Lecture Notes in Electrical Engineering 598

Bálint Németh   Editor

Proceedings of the
21st International
Symposium
on High Voltage
Engineering
Volume 1
Lecture Notes in Electrical Engineering

Volume 598

Series Editors
Leopoldo Angrisani, Department of Electrical and Information Technologies Engineering, University of Napoli
Federico II, Naples, Italy
Marco Arteaga, Departament de Control y Robótica, Universidad Nacional Autónoma de México, Coyoacán,
Mexico
Bijaya Ketan Panigrahi, Electrical Engineering, Indian Institute of Technology Delhi, New Delhi, Delhi, India
Samarjit Chakraborty, Fakultät für Elektrotechnik und Informationstechnik, TU München, Munich, Germany
Jiming Chen, Zhejiang University, Hangzhou, Zhejiang, China
Shanben Chen, Materials Science and Engineering, Shanghai Jiao Tong University, Shanghai, China
Tan Kay Chen, Department of Electrical and Computer Engineering, National University of Singapore,
Singapore, Singapore
Rüdiger Dillmann, Humanoids and Intelligent Systems Lab, Karlsruhe Institute for Technology, Karlsruhe,
Baden-Württemberg, Germany
Haibin Duan, Beijing University of Aeronautics and Astronautics, Beijing, China
Gianluigi Ferrari, Università di Parma, Parma, Italy
Manuel Ferre, Centre for Automation and Robotics CAR (UPM-CSIC), Universidad Politécnica de Madrid,
Madrid, Spain
Sandra Hirche, Department of Electrical Engineering and Information Science, Technische Universität
München, Munich, Germany
Faryar Jabbari, Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA,
USA
Limin Jia, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland
Alaa Khamis, German University in Egypt El Tagamoa El Khames, New Cairo City, Egypt
Torsten Kroeger, Stanford University, Stanford, CA, USA
Qilian Liang, Department of Electrical Engineering, University of Texas at Arlington, Arlington, TX, USA
Ferran Martin, Departament d’Enginyeria Electrònica, Universitat Autònoma de Barcelona, Bellaterra,
Barcelona, Spain
Tan Cher Ming, College of Engineering, Nanyang Technological University, Singapore, Singapore
Wolfgang Minker, Institute of Information Technology, University of Ulm, Ulm, Germany
Pradeep Misra, Department of Electrical Engineering, Wright State University, Dayton, OH, USA
Sebastian Möller, Quality and Usability Lab, TU Berlin, Berlin, Germany
Subhas Mukhopadhyay, School of Engineering & Advanced Technology, Massey University, Palmerston
North, Manawatu-Wanganui, New Zealand
Cun-Zheng Ning, Electrical Engineering, Arizona State University, Tempe, AZ, USA
Toyoaki Nishida, Graduate School of Informatics, Kyoto University, Kyoto, Japan
Federica Pascucci, Dipartimento di Ingegneria, Università degli Studi “Roma Tre”, Rome, Italy
Yong Qin, State Key Laboratory of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing, China
Gan Woon Seng, School of Electrical & Electronic Engineering, Nanyang Technological University,
Singapore, Singapore
Joachim Speidel, Institute of Telecommunications, Universität Stuttgart, Stuttgart, Baden-Württemberg,
Germany
Germano Veiga, Campus da FEUP, INESC Porto, Porto, Portugal
Haitao Wu, Academy of Opto-electronics, Chinese Academy of Sciences, Beijing, China
Junjie James Zhang, Charlotte, NC, USA
The book series Lecture Notes in Electrical Engineering (LNEE) publishes the latest developments
in Electrical Engineering - quickly, informally and in high quality. While original research
reported in proceedings and monographs has traditionally formed the core of LNEE, we also
encourage authors to submit books devoted to supporting student education and professional
training in the various fields and applications areas of electrical engineering. The series cover
classical and emerging topics concerning:
• Communication Engineering, Information Theory and Networks
• Electronics Engineering and Microelectronics
• Signal, Image and Speech Processing
• Wireless and Mobile Communication
• Circuits and Systems
• Energy Systems, Power Electronics and Electrical Machines
• Electro-optical Engineering
• Instrumentation Engineering
• Avionics Engineering
• Control Systems
• Internet-of-Things and Cybersecurity
• Biomedical Devices, MEMS and NEMS

For general information about this book series, comments or suggestions, please contact leontina.
dicecco@springer.com.
To submit a proposal or request further information, please contact the Publishing Editor in
your country:
China
Jasmine Dou, Associate Editor (jasmine.dou@springer.com)
India
Aninda Bose, Senior Editor (aninda.bose@springer.com)
Japan
Takeyuki Yonezawa, Editorial Director (takeyuki.yonezawa@springer.com)
South Korea
Smith (Ahram) Chae, Editor (smith.chae@springer.com)
Southeast Asia
Ramesh Nath Premnath, Editor (ramesh.premnath@springer.com)
USA, Canada:
Michael Luby, Senior Editor (michael.luby@springer.com)
All other Countries:
Leontina Di Cecco, Senior Editor (leontina.dicecco@springer.com)
** Indexing: The books of this series are submitted to ISI Proceedings, EI-Compendex,
SCOPUS, MetaPress, Web of Science and Springerlink **

More information about this series at http://www.springer.com/series/7818

Bálint Németh
Editor

Proceedings of the 21st

International Symposium
on High Voltage Engineering
Volume 1

123
Editor
Bálint Németh
High Voltage Laboratory
Budapest University of Technology
and Economics
Budapest, Hungary

ISSN 1876-1100 ISSN 1876-1119 (electronic)

Lecture Notes in Electrical Engineering
ISBN 978-3-030-31675-4 ISBN 978-3-030-31676-1 (eBook)
https://doi.org/10.1007/978-3-030-31676-1
© Springer Nature Switzerland AG 2020
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with regard
to jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

High voltage engineering is one of the most traditional fields of the electrical
engineering. It covers the application, the useful use and proper working of high
voltages and high fields and is extremely important for the reliable design, safe
manufacture and operation of electric devices, equipment and electric power
systems. At the 21st International Symposium of High Voltage Engineering (ISH),
the tradition meets the challenges of the twenty-first century.
The ISH series was founded by Professor Hans Prinz in 1972. The 21st edition
of the ISH conferences is organized by the High Voltage Engineering Group of
Budapest University of Technology and Economics.
The challenges of the twenty-first century such as the HVDC transmission, the
distribution generations and the smartening of the networks as well as the speed of
development and the improvement of the infrastructure and the adaptation of the
results in the everyday life are presented in high-quality papers at the 21st
International Symposium on High Voltage Engineering.
The topics covered include:
• Electromagnetic fields
• Transient voltages
• High voltage and high current testing techniques
• Advanced materials and insulation systems
• Monitoring and diagnostics
• High voltage systems and smart technologies
• HVDC technologies and applications
• Industrial applications of high voltage
• Live-line working
• Space charges
Special thanks go to the Steering Committee, the Program Committee, the
reviewers and last but not least the authors for their valuable contributions in
making the publication of these volumes possible.

v
Contents

Electromagnetic Fields
Partial Discharges of High Frequency Transformer for Space
Application in Near Vacuum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
M. T. Arentsen, C. L. Bak, F. F. da Silva, and S. Lorenzen
Optimization of Magnetic Shunts Towards Efficient and Economical
Power Transformers Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Ali Al-Abadi, Ahmed Gamil, and Franz Schatzl
Assessment of Thermal and Electric Field Characteristics
of HVDC Cable According to the Inner Filler Size of XLPE . . . . . . . . . 27
Ho-young Lee, Ik-Soo Kwon, Mansoor Asif, Chae-Kyun Jung,
Jae-Sang Hwang, Min-Ju Kim, and Bang-Wook Lee
Study on Conducted EMI Based on Wide-Band Model
for a –500 kV MMC-HVDC Station . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Weidong Zhang and Yuning Wei
Study on Simulation and Measurement of EMF in Transmission
Underground Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Jung Eui-hwan, Lee Soo-bong, Park Sang-taek, and Kim Jung-nyun
Further Developments of Metrological and Simulation-Based
Characterization of the Non-contact Measurement of Electrostatic
Charge by Means of Electric Field Meters . . . . . . . . . . . . . . . . . . . . . . . 55
C. Schierding, D. Möckel, M. Thedens, M. Beyer, and M. Kurrat
Calculation of Circulating Current Inside Power Transformers
in Non-symmetrical Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Bruno Jurišić, Tomislav Župan, and Leonardo Štrac
Influence of Different Turret-Tank Connection Types
on Transformer Turret Eddy Currents . . . . . . . . . . . . . . . . . . . . . . . . . 75
Tomislav Župan, Bruno Jurišić, and Franjo Kelemen

vii
viii Contents

Introducing an Inductive Loop Sensor as an Alternative to Record

the Phenomena of a Dense Plasma Focus of 400 J . . . . . . . . . . . . . . . . . 85
Luis Orellana, Jorge Ardila, Gonzalo Avaria, Benjamín Cevallos,
Cristian Pavez, Roger Schurch, and Leopoldo Soto
A New Approach in Calculation of Step Voltages for Complex
Grounding Systems by Analytical Considerations Only . . . . . . . . . . . . . 96
Martin Hannig, Ralph Brocke, and Volker Hinrichsen
Numerical Simulation on DC Breakdown of Polyimide Based
on Charge Transport and Molecular Chain Displacement . . . . . . . . . . . 108
Yuwei Li, Chenyu Yan, Daomin Min, Shengtao Li, Zhaoliang Xing,
Liangxian Zhang, and Chong Zhang
Thermal Instability Analysis of Station Class Surge Arresters
Based on Electrothermal Finite Element Simulation . . . . . . . . . . . . . . . 118
Yvonne Späck-Leigsnering, Maren Greta Ruppert, Erion Gjonaj,
Herbert De Gersem, and Volker Hinrichsen
The Comparison of Approaches to Power Frequency
Electromagnetic Field Hygienic Regulations . . . . . . . . . . . . . . . . . . . . . . 131
Nina Rubtsova, Sergey Perov, and Olga Belaya
Two Test Methods Comparison for Power Frequency Electric
Field Shielding Materials Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Sergey Perov, Olga Belaya, Tatyana Konshina, Elizaveta Tiutiunnik,
Balint Nemeth, Gabor Göcsei, and Valentin Faradzhev
Investigation of Electric and Magnetic Field in the Application
of Dynamic Line Rating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Levente Rácz, Dávid Szabó, and Gábor Göcsei

High Voltage Systems and Smart Technologies

Development and Characterization of a Transient Overvoltage
Recorder in Distribution Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
L. Perdomo, A. Alfonso, F. Santamaria, and F. J. Román
Condition Evaluation of High Voltage Transmission Line
in Thailand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Waraporn Luejai, Thanapong Suwanasri, and Cattareeya Suwanasri
Risk Assessment of Power Transformer in Thailand’s
Distribution Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
Thanapong Suwanasri, Suphon Kumpalavalee, and Cattareeya Suwanasri
Contents ix

Cost-Benefit for HV Transmission Line Renovation

and Replacement Based on Failure Probability
and Risk-Based Maintenance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Cattareeya Suwanasri, Thanapong Suwanasri, Waraporn Luejai,
and Surapol Saribut
Study on Charge and Discharge Phenomenon of Lithium Ion
Battery Under High Electric Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
S. Tsuji, Y. Fujita, H. Urushibata, A. Kono, M. Koyama, and R. Hanaoka
Online and On-site Partial Discharge Measurement of Long Length
Power Cables by Using Joints with Integrated PD Sensors . . . . . . . . . . 219
D. Passow, M. Beltle, S. Tenbohlen, J. Hohloch, and R. Grund
Development of Support Program for Managing Assets
by Considering Regular Maintenance Cost and Statistically
Expected Failure Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
Tsuguhiro Takahashi
SIAD-AERO: A New Methodology for the Inspection
of Energy Assets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Alexandre Dominice, Fernando Teixeira Abrahão,
Ricardo Augusto Tavares, and Alexandre Barreto
Thermal Aging of Photovoltaic Cables Based Cross-Linked
Polyolefin (XLPO) Insulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Ramy S. A. Afia, Ehtasham Mustafa, and Zoltán Ádám Tamus
Investigation of Complex Permittivity of XLPO Insulated
Photovoltaic DC Cables Due to Thermal Aging . . . . . . . . . . . . . . . . . . . 261
Ehtasham Mustafa, Ramy S. A. Afia, and Zoltán Ádám Tamus

Industrial Applications of High Voltage

Influenced Voltages on Pipelines by Overhead Lines Operated
at AC, DC and with Lightning Conditions . . . . . . . . . . . . . . . . . . . . . . . 273
T. Huecker and V. Schröder
The Variation of Electric Field on the Conductor Surface
Characterized by Space Charge Density . . . . . . . . . . . . . . . . . . . . . . . . . 282
Ruijin Liao, Hongbo Liu, Xuetong Zhao, Lulu Ren, and Yuandi Lin
Two-Dimensional Stray Loss Calculation of High
Voltage Transformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Joo HyunWoo, Bae ChaeYoon, Choi JongUng, and Kim YoungGeun
x Contents

Influence of Different Factors on the Interruption Characteristics

of Paralleled of High-Voltage SF6 Circuit Breakers with a Highly
Coupled Split Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
Haibo Su, Wenxiong Mo, Yong Wang, Lin Gan, Junxiang Liu, Ze Guo,
Le Gu, Lu Zhu, Li Chen, and Shengya Qiao
Investigation of High Altitude/Tropospheric Correction Factors
for Electric Aircraft Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
Athanasios C. Mermigkas, David Clark, and A. Manu Haddad
Evaluation of Wind Turbine Earthing System . . . . . . . . . . . . . . . . . . . . 316
S. A. Pastromas, I. A. Naxakis, S. I. Xerra, K. N. Koutras,
and E. C. Pyrgioti

Live-Line Working
Electromagnetic Interference Mitigation in a High Voltage
Inspection Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
S. B. Barnett, A. G. Swanson, T. Lorimer, and M. Brown
HCLS Special Hot Yoke for Live Line Maintenance . . . . . . . . . . . . . . . 342
Kristianto, Kurniawan Danu Diharja, and Iwan Safarudin
Evaluation of the Performance of Glass Insulator String
with Broken Units Under Artificial Pollution . . . . . . . . . . . . . . . . . . . . . 353
Frederico Tassi de Souza Silva and Ricardo Wesley Salles Garcia
Ten Years’ Experience of Live Work Implementation
in the Slovenian Electrical Energy Field . . . . . . . . . . . . . . . . . . . . . . . . . 362
A. Lovrenčič and V. Lovrenčić

Monitoring and Diagnostics

A Model Based on First Principles for the Simulation
of Partial Discharges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
A. Villa, L. Barbieri, and R. Malgesini
Contactless Thermal Online-Monitoring of Electrical Equipment
Under Load at High Voltage to Determine the Load Level
and Damage Avoidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
Thomas Gräf
Decomposition Characteristics of SF6 and Component Features
Extraction Under Negative DC Partial Discharge . . . . . . . . . . . . . . . . . 396
Mingxuan Zhang, Yulong Miao, Qiang Yao, Fuping Zeng, and Ju Tang
Contents xi

GIS Insulation State Evaluation Based on Multi-source

Information Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
Qiang Yao, Siying Wu, Yulong Miao, Ju Tang, Shiling Zhang,
and Fuping Zeng
Operational Aged Switchgear with the Age up to 50 Years –
Investigations, Testing, Results – Considerations for Design
and Operation of New Switchgear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
Thomas Gräf
Insulator Diagnostics Through a Dielectric E-Field Sensor . . . . . . . . . . 429
A. Villa, L. Barbieri, and R. Malgesini
Investigation on Performances and Functions of Power Emergency
Diesel Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
ZiChong Zhang, YuQuan Liu, WenXiong Mo, HongBin Wang,
ZhiGuo An, Han Liu, LinHuan Luo, and Zhong Wang
Condition Assessment of DC XLPE Cables Based on Combination
Weighting Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
Yufeng Zhu, Siyuan Wang, Cong Li, Yongpeng Xu, Gehao Sheng,
and Xiuchen Jiang
Identification of Hoarfrost Corona Losses Events Based
on Advanced Operational Data Analysis and Weather Modelling . . . . . 458
I. Gutman, Y. Solovyev, H. Ágústsson, and H. Mc Innes
Influence of Oil Temperature on Repeatability of Measurements
in Frequency Response Analysis of Power Transformers . . . . . . . . . . . . 472
Satoru Miyazaki, Yoshinobu Mizutani, Yoshihiro Wada,
and Chikara Hayashida
Detection of CF4 Gas Using a Nanomaterial-Based Gas Sensor
Fabricated by Dielectrophoresis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
Nisarut Phansiri, Daichi Maenosono, Takumi Furumoto, Hidefumi Sato,
Michihiko Nakano, and Junya Suehiro
Noise Reduction and Classification Method for Partial Discharge
Signal Detected by Transient Earth Voltage Sensor Using
Wavelet Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Hiroaki Cho, Yuuki Fujii, and Yusuke Nakamura
Comparison of Wavelet, Stationary Wavelet and Wavelet
Packet Methods for De-noising of Partial Discharge on Power
Cable System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
Arunjothi Rajendran, Thirumurthy, and K. P. Meena
Pattern Recognition of Partial Discharges on Power Cable Systems . . . 510
Arunjothi Rajendran, Thirumurthy, and K. P. Meena
xii Contents

Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV

GSU Transformers in Salto Grande Hydro Power Plant . . . . . . . . . . . . 521
E. Briosso, B. Gorgan, and W. Koltunowicz
Partial Discharge Phenomena in Power Capacitor Unit Insulation
Under Harmonic Resonance Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
Chongrag Boonseng, Rapeepornpat Boonseng, Nutthaphan Boonsaner,
Vijit Kinnares, Promsak Apiratikul, and Kunyanuth Kularbphettong
Partial Discharges Pulse Shape Analysis at AC and DC . . . . . . . . . . . . 549
Josef Pihera, Jaroslav Hornak, Ales Vobornik, Lukas Kupka,
Svatoslav Chladek, and Rainer Haller
Digital Filtering Methods for Interferences on Partial Discharges
Under DC Voltage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
Benedikt Hochbrückner, Martin Spiertz, Andreas Küchler,
Markus H. Zink, Thomas Steiner, Erik Winkelmann,
and Karsten Backhaus
Fiber-Optic Acoustic Sensors for Partial Discharge Detection
in Power Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 572
Chen Hao, Xu Yang, Qian Sen, and Su Lei
Development of PRPS Conversion Algorithm and Generator Using
PRPD Pattern of Underground Power Transmission Cable . . . . . . . . . . 582
Jae-Seop Lim, Gye-Hyun Joh, Jeon-Seon Lee, and Won Choi
Double-End Excitation of Transformer Winding Model
for Improved Frequency Response Analysis . . . . . . . . . . . . . . . . . . . . . . 594
Saurav Pramanik and Aravind Ganesh
Electrical Discharge Localization for Gas Insulated Line Based
on Distributed Acoustic Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 606
G. Ma, W. Qin, C. Shi, H. Zhou, Y. Li, and C. Li
Mineral Oil and Ester Based Oil/Paper Insulation Decaying
Assessment by FTIR Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 615
U. Mohan Rao, I. Fofana, R. Kartheek, K. M. L. Yapi, and T. Jaya
Experimental Research on Cumulative Deformation of Transformer
Winding Induced by Short-Circuit Current Impacts . . . . . . . . . . . . . . . 625
Fuqiang Ren, Shengchang Ji, and Lingyu Zhu
Construction of Novel Online Partial Discharge Monitoring System
in Model Power Apparatus Using Power Line Communication . . . . . . . 635
Hina Yoshikawa, Seiya Masuda, Yasutomo Kakimoto,
Nuraida Puspitasari, Yusuke Nakano, Masahiro Kozako, Masayuki Hikita,
Hidefumi Sato, Hideki Tagashira, and Kazuhito Ishida
Contents xiii

Identification of Partial Discharge Source in Power Apparatus

in Practical Substation Utilizing Artificial Neural Network . . . . . . . . . . 643
Yasutomo Kakimoto, Nuraida Puspitasari, Hina Yoshikawa,
Tadahiro Fujimura, Masahiro Kozako, Masayuki Hikita, Osamu Eda,
Yasuharu Shiina, Masanori Otsuka, and Hiroshi Kaneko
Partial Discharge Measurement and Its High Frequency
Characteristics in Cast Resin Transformer . . . . . . . . . . . . . . . . . . . . . . 653
Tadahiro Fujimura, Seiya Masuda, Masahiro Kozako, Masayuki Hikita,
Kazuo Iida, Tokihiro Umemura, Yusuke Nakamura, Tetsuo Nakamae,
Teruhiko Maeda, and Masakazu Higashiyama
Propagation Characteristics of Partial Discharge Radiated
Electromagnetic Waves for Insulation Diagnosis
of Oil-Filled Bushing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 662
Tadahiro Fujimura, Seiya Masuda, Masahiro Kozako, Masayuki Hikita,
Takashi Yamamoto, Shigemitsu Okabe, and Masaaki Nakahata
Diagnostics for Vacuum Degree of Vacuum Interrupter Based
on Partial Discharge Mechanisms Under Medium Vacuum Level . . . . . 673
Yusuke Nakano, Masahiro Kozako, Masayuki Hikita, Tsuyoshi Tanaka,
and Masato Kobayashi
Research on Energy Harvesting and Its Application in Thermal
Condition Monitoring of Three-Core Power Cables . . . . . . . . . . . . . . . . 684
T. X. Sun, X. F. Zeng, Y. X. Lu, and H. J. Li
Multidimensional Analysis of a Real Transformer Fleet Based
on the Evaluation of Oil Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697
Sebastian Schreiter, Holger Lohmeyer, and Peter Werle
Multidimensional Investigation of Transformer Oil Properties . . . . . . . . 705
Sebastian Schreiter, Holger Lohmeyer, and Peter Werle
Determining Temperature Dependence of Dielectric Response
by Temperature Normalization Method . . . . . . . . . . . . . . . . . . . . . . . . . 714
Yanfeng Gao, Wei Cai, Yi Lu, Chun Deng, Shun’an Gao, Wenxiang Xue,
Shuyuan Wang, Hui Wang, and Jifei Zhang
Terminal Configuration and Sensitivity Analysis of Tank Current
Measurement for FRA Diagnostics of Three-Phase Transformer
Winding Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724
V. S. B. Chaitanya Duvvury and Saurav Pramanik
Reliable Diagnostics on Rotating Machines Using FRA . . . . . . . . . . . . . 738
S. Uhrig, F. Öttl, N. Augeneder, and R. Hinterholzer
xiv Contents

Partial Discharge Characteristics of Palm Fatty Acid Ester

for Dissolved Gas Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 752
Yunwen He, Kiyoshi Wakimoto, Yang Xu, and Shengtao Li
Condition Assessment of Underground Cable System Using Health
Index and Conditional Multiplying Factor . . . . . . . . . . . . . . . . . . . . . . . 763
Tanachai Somsak, Thanapong Suwanasri, and Cattareeya Suwanasri
Comparison of a Thermally Accelerated Aged Oil-Paper Insulation
of an Open and Hermetically Sealed System . . . . . . . . . . . . . . . . . . . . . 777
T. Münster, T. Kinkeldey, P. Werle, K. Hämel, and J. Preusel
Investigation on the Accelerated Ageing Behaviour
of an Oil-Paper-Insulation Using Different Weight Ratios . . . . . . . . . . . 788
T. Münster, T. Kinkeldey, P. Werle, K. Hämel, and J. Preusel
Localization of Deviations in Cable Geometry Using FDA Fault
Localization Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 799
Y. Norouzi, C. Frohne, J. Rothfeld, and P. Werle
Evaluation of Numerical Indices for Objective Interpretation
of Frequency Response to Detect Mechanical Faults
in Power Transformers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 811
M. Tahir, S. Tenbohlen, and M. H. Samimi
Statistic Based Method for Post-processing Analysis in Lifetime
Investigations of Multi-factor Aged Winding Insulation . . . . . . . . . . . . . 825
A. Cimino, J. Horst, and F. Jenau
Electromagnetic Wave Radiation Due to Partial Discharges Inside
Power Transformers in the Frequency Domain . . . . . . . . . . . . . . . . . . . 838
M. Akbari Azirani, M. Ariannik, P. Werle, and A. Akbari
Simulation of the Mechanical Vibrations in a High
Voltage Transformer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 850
Chunhui Gu, Lin Gan, Wenxiong Mo, Yong Wang, Yi Rao, and Yu Qin
Non-destructive Testing Method for Composite Insulators Based
on Digital Shearography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 862
Lishuai Liu, Chenjun Guo, Liming Wang, and Hongwei Mei
Investigation on the Acid Removal Performance of Oil
Regeneration Sorbent Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 871
S. Y. Matharage, S. Liu, Q. Liu, and Z. D. Wang
Terahertz Time-Domain Spectroscopy Characterization of Aged
XLPE Cable Insulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 878
Yuanyuan Zhang, Ze Lian, Jianying Li, and Shengtao Li
Contents xv

Evaluation and Performance Benchmarking of Partial Discharge

Measurement Sensors Technologies Based on Non-conventional
PD Measurement System for Medium Voltage GIS Applications . . . . . . 887
David Jebamony, Manjunath Ramesh, Bastian Wölke, and Pascal Thyssen
Spectral Power Analysis of Partial Discharges Waveforms During
Electrical Tree Growth Under Different Excitation Frequencies . . . . . . 899
Roger Schurch, Osvaldo Munoz, and Jorge Ardila-Rey
Development of Testing Method for Static Electrification
Within Power Transformer Using Suspended Cellulose Fibers
in Insulating Oil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 908
Masanobu Yoshida, Soichi Moriguchi, Gaku Sato, Yoshinori Konishi,
and Masayuki Hasegawa
Development of a Digital Twin for the Determination
of Transmission Line Conductor Asset Health . . . . . . . . . . . . . . . . . . . . 917
J. C. Toth, J. Marmillo, and G. Biedenbach
Estimation of the Ageing Condition of Oil-Filled Transformers
Based on the Oil Parameters Using a Novel Fuzzy Logic Algorithm . . . 926
Tobias Kinkeldey, Tobias Münster, Peter Werle, Suwarno, Kai Hämel,
and Jörg Preusel
Optical Partial Discharge Measurement with Integrated Optical
Fibers as Sensing Element . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 937
Inna Kübler and Daniel Pepper
Condition Monitoring of High Voltage Press-Pack IGBT
with On-State Collector-Emitter Voltage . . . . . . . . . . . . . . . . . . . . . . . . 949
Cao Zhan, Lingyu Zhu, Chenshuo Liu, Liang Pan, Jiangyu Liu,
and Shengchang Ji
Influence on Formation of Methanol and Ethanol in Oil-Filled
Transformers Under Accelerated Thermal Aging . . . . . . . . . . . . . . . . . 958
K. Homeier, M. Imani, T. Münster, T. Kinkeldey, P. Werle, and D. Philipp
Applying Unsupervised Machine Learning Method on FRA Data
to Classify Winding Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 969
Xiaozhou Mao, Shuntao Ji, Zhongdong Wang, Paul Jarman,
Andrew Fieldsend-Roxborough, and Gordon Wilson
Interpreting First Anti-resonance of FRA Responses Through Low
Frequency Transformer Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 982
Bozhi Cheng, Peter Crossley, Zhongdong Wang, Paul Jarman,
Andrew Fieldsend-Roxborough, and Gordon Wilson
xvi Contents

Feasibility of Detecting Partial Discharge in Smoothing Reactor

Using UHF Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 993
Shusheng Zheng, Yangzhen Cui, and Bo Qi
Speciality and Applicability of Advanced Response Methods
in Contrast to the Traditional Dielectric Measurements
for Condition Assessment of Power Transformers . . . . . . . . . . . . . . . . . 1003
Richárd Cselkó, Máté Szirtes, and Gusztáv Csépes
Simulation of Void Discharges and Time Domain Examination
of Their Emitted Electromagnetic Signals . . . . . . . . . . . . . . . . . . . . . . . 1015
Máté Szirtes and Richárd Cselkó
A Novel Approach of Critical Span Analysis . . . . . . . . . . . . . . . . . . . . . 1025
Dávid Szabó, Levente Rácz, and Gábor Göcsei

Space Charges
Simple Method to Visualize Surface and Space Charges
by Specially Processed Colour Pigments . . . . . . . . . . . . . . . . . . . . . . . . . 1035
Thomas Huecker
Simulation of Partial Discharge Influenced by Space Charges
in Silicone Rubber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1044
Haozhe Cui, Zhaoliang Xing, Chong Zhang, Liangxian Zhang,
Daomin Min, and Shengtao Li
Calculation of Space Charge and Field Distributions
in a Cable Joint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1055
Yang Wu, Chuanhui Cheng, Rui Su, Kai Wu, and Lijun Wang
Study of Deep Dielectric Charging Characteristics and Suppression
Method Under Space Irradiation Environment . . . . . . . . . . . . . . . . . . . 1067
Xiaoping Wang, Shusai Zheng, Daomin Min, Shengtao Li, Xinbin Hou,
and Li Wang
Charge Traps Depended Space Charge Dynamics and Electrical
Breakdown Characteristics of Polymer Insulating Materials . . . . . . . . . 1077
Yuanwei Zhu, Peng Wei, Zichao Shen, Huize Cui, Yu Jing, Dongfan Li,
Zihao Wang, Dongri Xie, Guanghao Lu, and Shengtao Li
Effect of Deep Traps and Molecular Motion on Dc Breakdown
of Polyethylene Nanocomposites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1087
Rui Mi, Chenyu Yan, Zhaoliang Xing, Qingzhou Wu, Daomin Min,
and Shengtao Li
Modelling and Verification of Ion Currents Under Various
Air Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1097
J. Kortenbrede and F. Jenau
Contents xvii

Numerical Modelling and Influence of Defects on Space Charges

in Epoxy Resin Under HVDC Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . 1108
G. Ortiz, P. Mbolo Noah, S. Agnel, P. Notingher, C-T. Vu, and A. Girodet
Investigations on the Polarity Dependent Charge Carrier Injection
in Dielectric Liquids at High DC Voltage Stress . . . . . . . . . . . . . . . . . . 1121
Tobias Gabler, Karsten Backhaus, Ronny Fritsche,
and Steffen Großmann
Modelling Space Charge in HVDC Cable Insulation . . . . . . . . . . . . . . . 1132
Yunpeng Zhan, George Chen, Miao Hao, Zhiqiang Xu, Lu Pu,
Xuefeng Zhao, Haofei Sun, Sen Wang, Anxiang Guo, and Jian Liu
Study on Signal Correction Method for Measurement
of Space Charge Distribution of Multi-dielectrics Using
PEA Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1143
HyungGyu Kim, HyunCheol Jung, and Seung Hwangbo
Charge Accumulation on Slightly Conductive Barrier Systems
and Its Effect on Breakdown Voltage in an Air Insulated Rod
Plane Arrangement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1155
M. Schueller, A. Blaszczyk, F. Mauseth, H. K. Meyer, N. Stieger,
and J. Smajic

Transient Voltages
Current and Electric Field Characteristics of 35 Return Strokes
from Negative Lightning Measured at Peissenberg Tower Germany . . . 1169
Christian Paul, Fridolin H. Heidler, and Wolfgang Schulz
Improvement of Overhead Transmission Lines Lightning Protection
by Line Arresters with Separate Groundings and Shielding Wires
Fixed at Insulation Racks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1180
G. V. Podporkin
Study on Consistency of Failure Probability Characteristics
of Oil-Paper Insulation Under Different Impulse Voltages . . . . . . . . . . . 1192
Yu Zhang, Shijun Xie, Xiongwei Jiang, Lian Ye, Chenmeng Zhang,
Potao Sun, Zhou Mu, and Wenxia Sima
Insulation Withstand Testing on Surge Arresters and the Influence
of Voltage Grading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1207
Hans Sjöstedt and James Taylor
Characteristics of Failed Bypass Diodes for Photovoltaic Module
by Artificial and Natural Lightning . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1218
Toshiyuki Hamada, Kenta Nakamoto, Ikuo Nanno, Norio Ishikura,
Shinichiro Oke, and Masayuki Fujii
xviii Contents

Relationship Between Electron Activation Energy Absorption

and Impulse Breakdown Voltage in Polymers . . . . . . . . . . . . . . . . . . . . 1225
Xiongwei Jiang, Wenxia Sima, Potao Sun, and Qingjun Peng
Characterization of Electric Fields Produced by Preliminary
Breakdown Pulses Observed in Bogotá, Colombia . . . . . . . . . . . . . . . . . 1235
C. A. Granados, H. E. Rojas, C. A. Rivera, and F. J. Román
Duality-Based Potential Transformer Model Including
Black-Box Circuit for High-Frequency Transient Simulation . . . . . . . . . 1244
Daixiao Peng, Ming Yang, Wenxia Sima, Jinwei Chu, Zhicheng Xie,
and Yonglai Liu
Propagation of Lightning, Oscillating and Non-standard Impulse
Waveforms in Transformer Windings . . . . . . . . . . . . . . . . . . . . . . . . . . 1254
Marek Florkowski, Jakub Furgał, Maciej Kuniewski, and Piotr Pająk
Recent Developments of Field Grading for High Voltage Vacuum
Circuit Breakers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1265
B. Kühn, D. Gentsch, B. Weber, M. Hilbert, and M. Kurrat
Dimension Estimation of Transformer Windings Based
on Frequency Response Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . 1276
Ebrahim Rahimpour, Vahid Rashtchi, and Reza Aghmashehr
Estimation of Number of Model Units in Transformer Detailed
RCLM Model Based on Terminal Measurement in the Case
of Unavailable Design Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1285
Ebrahim Rahimpour, Vahid Rashtchi, and Reza Aghmashehr
Fractal-Based Approach for Modelling Electric Breakdown of Air
Gaps: An Application to a 75 cm Positive Rod-Plane Gap . . . . . . . . . . . 1295
A. I. Ioannidis, P. N. Mikropoulos, T. E. Tsovilis,
and N. Karanikiotis
Self Calibrating High Voltage Divider . . . . . . . . . . . . . . . . . . . . . . . . . . 1306
S. Passon, N. Rühmann, F. Schilling, J. Meisner, and M. Kurrat
High Voltage Topologies for Very Fast Transient Measurements . . . . . . 1316
A.-P. Elg, S. Passon, J. Meisner, and J. Hällström
Methods for Field Measurement of the Frequency-Dependent
Soil Electrical Properties: Evaluation of Electrode Arrangements
Through FEM Computations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1328
Z. G. Datsios, P. N. Mikropoulos, and E. T. Staikos
Effect of Quenching on Dielectric Properties of ZnO
Varistor Ceramics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1340
Kangning Wu, Yuwei Huang, and Jianying Li
Contents xix

Analysis of Lightning Overvoltage Base on Grounding Method

of Lightning Arresters in Wind Farm . . . . . . . . . . . . . . . . . . . . . . . . . . 1347
Jin-Hyuk Kim, Kyo-Ho Kim, and Jung-Wook Woo
Resonance Behaviour of Shielded High-Voltage Laboratories
and the Effect of Interference on Impulse Measurements . . . . . . . . . . . . 1356
T. C. Schlüterbusch, S. Passon, and J. Meisner
On the Use of UHF Sensors in the Detection and Characterization
of Pulsed Plasma Discharges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1367
Luis Orellana, Jorge Ardila, Gonzalo Avaria, Benjamín Cevallos,
Cristian Pavez, Roger Schurch, and Leopoldo Soto
Lightning Fault Rate of Power Distribution Line in Wind Farm
in Winter Lightning Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1377
Koji Michishita and Shigeru Yokoyama
The Criterion for Electromagnetic Compatibility Evaluation
in High Voltage Substations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1390
Tomaž Živic
Simplified Approach for Investigating Overvoltages in DC Cables
in a –320 kV Symmetrical Monopolar HVDC System . . . . . . . . . . . . . . 1401
T. Karmokar and O. Lennerhag
Experimental Investigation of the Overvoltage Steepness Effects
on Corona Inception Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 1413
Evanthia I. Bousiou, Pantelis N. Mikropoulos, and Vasileios N. Zagkanas
LPS and Grounding System for Light Rail Transit in Tropical
Area with High Lightning Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1423
Wisnu Adyatma S and Reynaldo Zoro
Verification of the Aging Impact of MOV Ceramics by Recording
Wide-Range V/I Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1437
S. Joerres, A. Claudi, and G. Finis
Experimental Analysis and Suppression for Very Fast Transient
Disturbance in Device Power Port During Switching Operation
in GIS Substations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1446
Siqiao Ge and Weidong Liu
Efficient Lightning Protection of a Gas Insulated Substation
Replacing an Air-Insulated Substation . . . . . . . . . . . . . . . . . . . . . . . . . . 1458
Constantin Balzer, Volker Hinrichsen, and Wolfgang Tausend
Lightning Performance of an Air Insulated High
Voltage Substation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1471
C. A. Christodoulou, V. Vita, and L. Ekonomou
xx Contents

Estimation of the Minimum Backflashover Current of 150

and 400 kV Overhead Transmission Lines Through ATP-EMTP
Simulations: Effect of the Lightning Stroke Location Along
Line Spans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1482
Z. G. Datsios, P. N. Mikropoulos, T. E. Tsovilis, V. T. Karakostas,
and S. P. Mavidou
Analysis of the Frequency Response of a Grounding System Using
the Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1491
Arthur Francisco Andrade, George R. S. Lira, Pedro Henrique F. Monteiro,
Helder A. Pereira, Edson G. Costa, and Rômulo O. Teixeira
Analysis of the Influence of Soil Stratification Models on Different
Grounding System Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1502
George R. S. Lira, Matheus R. Alves, André Igor Nóbrega,
Arthur Francisco Andrade, Edson G. Costa, and Rômulo O. Teixeira
Research on a New Type of Fault Current Limiter Based on High
Coupled Split Reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1513
Wenxiong Mo, Junxiang Liu, Yong Wang, Lin Gan, Haibo Su, Lu Zhu,
Shengya Qiao, Kaijian Wu, and Zhao Yuan
Dynamic Non-linearity of Impulse Digitisers and Its Effects
on Impulse Voltage Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1523
Wei Yan and Yi Li
Non-linear Dynamic Response of a Ground Rod in Resistive Soils . . . . 1533
R. R. Diaz, J. N. Silva, and A. Parellada
Lightning Performance of Medium Voltage Distribution Networks . . . . 1544
V. Vita, C. A. Christodoulou, and L. Ekonomou

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1555

Electromagnetic Fields
 Partial Discharges of High Frequency
Transformer for Space Application
in Near Vacuum

M. T. Arentsen1,3(&), C. L. Bak1, F. F. da Silva1, and S. Lorenzen2

1
Aalborg University, Pontoppidanstræde 111, 9220 Aalborg Ø, Denmark
mtarentsen@yahoo.com, {clb,ffs}@et.aau.dk
2
Flux A/S, Industrivangen 5, 4550 Asnæs, Denmark
sll@flux.dk
3
DEIF A/S, Frisenborgvej 33, 7800 Skive, Denmark

Abstract. This paper presents partial discharge measurements and a model-

based investigation for corona activity on a prototype of a high frequency
transformer from Flux A/S, designed to operate on a space travelling vessel.
Previous iterations of the prototypes failed external partial discharge tests, and
the environment that the transformer was designed to be installed and tested in,
was suspected to cause the partial discharge to onset. In normal operation on the
space vessel (and during the required design-tests), the transformer has its ter-
minals exposed to the surrounding atmosphere, whose pressure varies from
standard atmospheric pressure to near vacuum conditions during the launch of
the space vessel. The most critical condition for external partial dielectric
breakdown will be encountered under its intended operation, similar to the knee
point in the Paschen curve. The conditions at this point yields the largest stresses
on the dielectric, in this case the surrounding atmosphere of the transformer.
A predictive method for evaluation of corona activity onset is therefore inclu-
ded. This is done based on calculation of the effective ionization coefficient
along the critical electrical field line, a obtained by Finite Element Method
(FEM) models and is evaluated for a given set of pressure values to emulate the
intended operating conditions of the transformer.  a itself is a function of the
electrical field strength along the critical line, as well as the relative air density.
The FEM models of the transformer were therefore designed to calculate the
electrical field strength distribution around the transformer, and to locate the
critical field line. It was found that 2D models yielded a satisfactory accuracy for
the intention of the simulation, and that 3D models would only yield a slightly
improved accuracy for a substantially increased computational burden.
For the partial discharge measurements, the test voltage was 1 kV at 50 Hz,
and the geometry of the transformer and distances to objects in proximity to the
transformer will remain constant, leading to the electrical field strength distri-
bution also being constant. The only varying parameter which is affecting  a is
the variation of the atmospheric pressure. The corona activity onset condition is
evaluated by evaluating the integral value of a over its region that yields a net
positive ionization and compare this value with the criteria for the Townsend
mechanism. The simulated pressure range includes only values that are realiz-
able with the available equipment in the HV-laboratory of Aalborg University,
which is 1.0 to 0.2 bars of absolute pressure. This is done with the intention to

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 3–14, 2020.
https://doi.org/10.1007/978-3-030-31676-1_1
4 M. T. Arentsen et al.

compare the predictive method for corona onset with actual partial discharge
measurements. The PD measurements showed that no external partial discharge
activity was present for the given experimental conditions and pressure range,
and the new design of the transformer prototype was therefore improved.

Keywords: Corona
Space application
Partial discharge measurement
High

voltage measurement techniques
FEM simulation

1 Introduction

An experimental- and model-based investigating of external partial discharge activity

on a prototype of a high frequency transformer for space applications will be presented
in this paper. The transformer prototype is designed for an ion-motor on a space vessel
by Flux A/S, and testing of the first iterations of the prototype showed unsatisfactory
levels of PD-activity according to the governing standard ECSS (European Coopera-
tion of Space Standardization). The measurements were performed at various absolute
atmospheric pressure levels to emulate the intended operational environment of the
transformer during launch into space of the space vessel.
The primary goal was to investigate the prototype by experimental means, while a
method for model-based investigation was highly desired as well, since none was
available to Flux A/S in their design process. Both investigations were conducted and
are shown in this paper.
The model-based investigation was done by designing a Finite Element Method
(FEM) model of the transformer, in order to calculate the electrical field distribution as
well as locating the critical field line. Conventional corona onset evaluation was not
used, which usually is done by assessing the field strength at the critical point. Instead,
corona onset evaluation will be done based on calculation of the effective ionization
along the critical field line, which is a function of the applied voltage, the geometry of
the Device Under Test (DUT) and surrounding atmosphere and its relative density.
This modelling- and corona onset evaluation approach is described in more detail in the
State-of-the-Art section, and the intention of the model-based investigation is to serve
as a rough estimation tool for evaluating the expectancy of corona being present on the
test object for a given set of test conditions

2 PD Measurement Setup and Results

The PD-measurement test performed on the first iterations of the prototypes were
lacking useful details, since the equipment available to Flux A/S had no Phase-
Resolved-Partial-Discharge (PRPD) measurement capability – however, a corona glow
was observed during their test.
In accordance with IEC 60270 “High-Voltage Test Techniques – Partial Discharge
Measurements” [1], a PD-measurement setup with PRPD measurement capabilities
was designed to test the prototype for any PD activity, and the circuit can be seen in
Fig. 1.
 Partial Discharges of High Frequency Transformer for Space Application 5

Fig. 1. PD-measurement circuit, consisting of calibration Unit CAL 542, coupling quadripole
CPL 542, PD measurement unit MPD 600 and the ADC MCU 502 [2].

The DUT being the transformer prototype, is situated in the vacuum chamber, seen
in Fig. 2, which is used to vary the atmospheric pressure to emulate the ambient
conditions during assent of the space vessel. According to ECSS [3], the PD-
Measurement test must be performed in as close to intended operating conditions as
possible, rendering the atmospheric pressure variation a necessity.

Fig. 2. (a) Transformer prototype in suspension chamber. (b) Vacuum chamber, which
suspension chamber is inserted into.
6 M. T. Arentsen et al.

Conducting the PD-Measurements according to ECSS procedure and requirements,

conforming with IEC 60270, and repeating for pressure levels from 1.0 bar to 0.2 bars
of absolute pressure (Maximum capability of pressure vessel and pump) to emulate the
intended operating environment during assent yielded PD-Measurements seen in
Fig. 3.

Fig. 3. PD-measurement of transformer prototype at 0.2 bar over duration of 10 min. Apparent
activity hotspots at *−45° from zero-crossing.

Looking at Fig. 3, it seems like PD activity was present between the peak and zero-
crossing of the applied voltage waveform. It was discovered that this was caused by the
measurement equipment picking up noise levels of 0.8 pC phase shifted 180° of each
other, and wrongly getting corelated to a fixed spot of the applied voltage. This was
discovered by monitoring the PD-measurements while no voltage was applied, where
these two peaks drifted along the phase-axis of Fig. 3, until voltage was applied
gradually, and the peaks was wrongly corelated to these particular phase positions.
Figure 3 clearly indicates that no PD-activity was present during the 10-min
measuring period for the absolute air pressure of 0.2 bar and applied test voltage of
1 kV peak. In fact, no activity was present during any of the pressure levels tested,
from 1.0 to 0.2 Bars with decrements of 0.1 bar.
It should be noted that it was not possible to achieve absolute pressures lower than
0.2 bar, due to limitations of the available equipment. Conclusion of the measurement
must therefore be, that no PD-activity was present for any achievable pressure level,
and the transformer prototype therefore passes the ECSS requirements in all of the
possible test conditions, only limited by the actual pressure vessel and vacuum pump.
An out-of-scope experiment was also conducted, where the voltage was gradually
increased for the most critical pressure level, in order to experimentally determine the
 Partial Discharges of High Frequency Transformer for Space Application 7

PD onset voltage level. The PD onset was found and measured over the same 10-min
period as the other PD-measurements and had an onset voltage at a peak voltage of
2 kV, as seen in Fig. 4, and were internal PDs.

Fig. 4. PD onset at 2 kV for 0.2 bar.

Further increasing of the voltage was not done, since this would put the transformer
in the risk of being destroyed – eliminating future experiments on it completely. Under
a handful of the units are in existence, risking destroying one was not deemed worth it.

3 State-of-the-Art

This paper also demonstrates a strategy for evaluating corona onset based on calcu-
lation of the effective ionization coefficient along the critical electrical field line, in
contrary to evaluating for corona onset purely by maximum surface electrical field
strength at the critical point, adjusted to atmospheric conditions when deviating from
standard conditions. The effective ionization, a is given by:

a ¼ a  g ð1Þ

Where a is the ionization coefficient along the critical electrical field line, and η is
the attachment coefficient. It should be noted that Eq. (1) ceases to be valid when
a  η. The electrical field strength is simply too weak to provide a net positive
ionization from this point onwards.
8 M. T. Arentsen et al.

The ionization and attachment coefficient are calculated based on expressions

developed by Phillips et al. [4], and can be seen in Eqs. (2, 3):
(

a 3632
exp 168
Ed for 1:9\ Ed  45:6
¼
 ð2Þ
d 7385
exp 200:8
Ed for 45:6\ Ed  182:4
   2
g E E
¼ 9:9865  0:541
103 þ 1:118
108 ð3Þ
d d d

As can be seen in Eqs. (2, 3), The ionization and attachment coefficients are
functions of the electrical field strength E, as well as the relative air density d. These
expressions are deduced by experimental means and are empirical equations but are
also valid for non-radially emitting electrical fields [4].
The effective ionization coefficient, a, is then integrated along the critical field line,
seen in Eq. (4) [5, P. 343], until the point xc , where a  g ¼ 0, in order to determine
the number of charge carriers along the critical field line:
Z xc  d
a dx ¼ lnðNcr Þ ð4Þ
0

For corona onset evaluation based on Eq. 4, Townsend’s mechanism is the gov-
erning mechanism [5, P. 325], and taking the natural logarithm on the number of
charge carriers Ncr will equal 8 when the corona onset conditions are met.

4 FEM Model

The model-based investigation of the transformer prototype was be done by means of

FEM models. The electrical field distribution can be obtained by carefully mapping the
geometry of the DUT, applying a satisfactory fine mesh which has its densest areas
around the conducting surfaces and applying initial and boundary conditions. Mesh
density refinement was done in an iterative manner, where an initial mesh size was
applied and iteratively reduced between simulations until electrical field distribution
converged, meaning that additional downsizing of the mesh did not yield any mean-
ingful improvement of results compared to the increase in computational power.
Ideally the FEM models should be three dimensional to account for the asymme-
tries on the transformer prototype and test environment, but the influence of the extra
dimension was shown to be quite insignificant compared to the increase in the required
computational power needed in such case. Pairing this with a constraint on the max-
imum allowed number of notes, it was deemed that the 2D models yielded a satis-
factory accuracy, to be used as a rough predictive tool for corona onset evaluation in
the design process of the transformer.
Several 2D FEM models were developed based on various cross-sections of the
transformer, however, only the model yielding the largest electrical field strengths was
chosen to be analysed, since corona would first onset from this modelled area. The
cross-section yielding the largest electrical fields can be seen in Fig. 5:
 Partial Discharges of High Frequency Transformer for Space Application 9

Fig. 5. Red circle: screws investigated. Green line: vertical symmetry axis. Blue line: modelled
cross-section.

The red circle shows two screws which possess 1 kV and ground potential
respectively, since the right hand one is in galvanic contact with the grounded metal
case behind the polymer. The blue line represents the cross-section which is modelled
by FEM. The two brass-coloured terminals have a vertical symmetry line, which is
advantageous when modelling by FEM, since the use of such line in the program can
allow for more available notes.

4.1 Simulation Results of the FEM Model

The potential distribution, seen in Fig. 6 is obtained by simulating the FEM model of
the cross-section in Fig. 5.

Fig. 6. Potential distribution of the modelled cross-section.

10 M. T. Arentsen et al.

In Fig. 7, the electrical field distribution was obtained by a similar approach.

Fig. 7. Electrical field distribution and critical electrical field line of the modelled cross-section.

The critical field line originates from the point with the largest electrical field
strength and propagates in such a way that it intersects the contour lines of the electrical
field distribution perpendicularly and terminates on the external ground electrode
mounted on the inside on the vacuum chamber. The largest electrical field strength was
11.6 kV/cm. The electrical field strength along the critical field line was then obtained
and can be seen in Fig. 8.

Fig. 8. Electrical field strength along the critical electrical field line.
 Partial Discharges of High Frequency Transformer for Space Application 11

4.2 Calculation of the Ionization and Attachment Coefficient

Calculation of the ionization and attachment coefficient along the critical electrical field
line can then be done by Eqs. (2, 3). It should be noted that the electrical field dis-
tribution will remain constant as the geometry and applied voltage level is the same for
all simulations and conducted experimental tests. Therefore, the only parameter which
influences the ionization and attachment is the relative air density d. All pressure levels
used in the experimental PD-measurements were simulated for, but only the one
yielding the largest effective ionization was used to evaluate the corona onset criterion.
This level was an absolute pressure of 0.2 bar. The ionization and attachment coeffi-
cient along the critical line can be seen in Fig. 9. The effective ionization  a is only
defined until the intersection point, a  g ¼ 0. Therefore, the integral boundaries in
Eq. (4) is from zero to the point xc where a  g ¼ 0.

Fig. 9. Ionization and attachment coefficient along the critical electrical field line.

Evaluation of the integral in Eq. (4) on obtained data in Fig. 9 gives:

Z 2:907
105
ða  gÞdx ¼ lnðNcr Þ ¼ 0:0049 ð5Þ
0

From Eq. (5) it is evident that no corona should onset for any of the pressure levels
with an applied test voltage of 1 kV in the tested and modelled environment.
This simulation was repeated multiple times with varying voltage level for an
absolute pressure of 0.2 bar, to find the theoretical corona onset voltage level, and this
was found to be around an applied peak voltage of 3.5 kV.
12 M. T. Arentsen et al.

5 Evaluation of Corona Onset Criterion and Ionization

and Attachment Expressions

Experimental PD-measurements in Sect. 2 and FEM model results in Sect. 4.2 showed
that no corona onset was happening or was expected to happen.
Modelling a coaxial cylindrical arrangement in FEM, for which well documented
corona onset levels exists [5], would allow for a rough validation of validity of the
corona onset criteria as well as the ionization and attachment expressions. Arbitrarily,
two radii of 0.5 mm and 2 mm were chosen respectively for the inner and outer
cylinder to be modelled. Similarly, the relative air density was chosen to be d = 1.0.

Fig. 10. Corona onset field strength as a function of the product of relative air density and radius
of curvature [5].

Based on Fig. 10, the corona onset field strength was determined to be 73.18 kV.
This information was used to calculate the potential on the inner cylinder which yields
this electrical field strength for the given radii [5]:

V
Emax ¼ ð6Þ
r1
lnðr2 =r1 Þ

kV
V ¼ 73:18
0:05 cm
lnð0:2 cm=0:05 cmÞ ¼ 5072 V ð7Þ
cm
 Partial Discharges of High Frequency Transformer for Space Application 13

The coaxial cylindrical geometry was then modelled in the FEM program, and the
voltage of the inner cylinder was set to 5072 V. The exact same approach as in
Sect. 4.1 was then used to find the electrical field distribution as well as the critical field
line, which can be seen in Fig. 11:

Fig. 11. Electrical field distribution of the coaxial cylindrical geometry. r1 = 0.5 mm,
r2 = 2 mm, V = 5072 V, d = 1.0.

The same approach as in Sect. 4.2 was applied to this model in order to determine
the ionization and attachment coefficient along the critical field line, and this can be
seen in Fig. 12:

Fig. 12. Ionization and attachment coefficient along the critical field line for the coaxial
cylindrical geometry.
14 M. T. Arentsen et al.

Again, Eq. (4) was evaluated to check the corona onset criterion and yielded:

lnðNcr Þ ¼ 7:497 ð8Þ

The intention of the model-based investigation was to give a rough estimate on the
expectancy of corona being present for a given test case. It is concluded that the
demonstrated approach in Sect. 4 can be utilized, even though the criterion is only
close to being fulfilled with an integral value of 7.497 < 8. It should also be noted that
the demonstration in this chapter is for a radial geometry, which is not the case for the
transformer prototype.

6 Conclusions

The PD-Measurement setup was built to conform with the ECSS standard as well as the
IEC 60270. The PD-measurements of the DUT showed no activity was present for the
specified test cases in scope. The PD-activity onset voltage was found to be 2 kV at 0.2
bar, but of internal nature – and out of scope of this investigation. It can therefore be
concluded that the design of the investigated prototype was better than the previous
design with respect to PDs.
2D FEM models of the transformer were deemed to provide sufficient accuracy for
the intended use of the model-based investigation. The only model presented and
analysed in this paper was the one yielding the largest electrical field strengths. The
analysis of this showed that no corona was to be expected during any of the PD-
Measurements, which conforms with the initial experimental findings.
The work, results and methodology both in the model-based as well as the
experimental based investigation of the transformer prototype provided Flux A/S state-
of-the-art tools to perform similar investigations on future prototype and has influenced
their design of their own PD-measurement facilities.

References
1. IEC 60270: High-voltage test techniques – partial discharge measurements. (IEC 60270:2000)
(2001)
2. OMICRON Electronics GmbH (AT): MPD 600 user Manual. MPD600.AE.7 (2013)
3. European Cooperation For Space Standardization: High Voltage Engineering and Design
Handbook. ECSS-E-HB-20-05A, December 2012
4. Phillips, D.B., Olsen, R.G., Pedrow, P.D.: Corona onset as a design optimization criterion for
high voltage hardware. IEEE Trans. Dielectr. Electr. Insul. 7(6) (2000). ISSN 1070-9878
5. Kuffel, E., Zaengl, W.S., Kuffel, J.: High Voltage Engineering Fundamentals, 2nd edn.
(2000). ISBN 978-0-7506-3634-6
 Optimization of Magnetic Shunts Towards
Efficient and Economical Power
Transformers Design

Ali Al-Abadi(&), Ahmed Gamil, and Franz Schatzl

SGB Power Transformers (SGB-SMIT Group), Regensburg, Germany

ali.al-abadi@sgb-smit.group

Abstract. The focus of European and international markets and standards in

the last few years is to achieve highly efficient power transformer designs,
depending on the application sector (generation, transmission or distribution).
The efficiency is directly related to total losses and in turn the heat emission.
Losses in the power transformer are the summation of load and no-load losses.
The load losses, which generally dominate the total losses, consist of ohmic and
stray losses. However, reduction of ohmic losses is limited by the conductivity
and design of the winding conductors. The stray losses are generated inside and
outside the windings. Losses outside the windings (leads, tank and clamping
structure) can be reduced through control of the leakage flux and the trade-off
between different electromagnetic materials, such as magnetic shunts. In the
current study, we present intensive investigations and optimizations of the
application of magnetic shunts to control the leakage flux in power transformers
to achieve lower stray losses. Eventually, design rules for the tank and yoke
shunts are developed. The developed design tools are applied on case study
transformers with different ratings and short-circuit impedances. Results show
significant reduction in load losses. Economically, the newly developed design
rules can be used to reduce the emitted heat, which will provide the opportunity
to optimize the final power transformer design with a potential saving in
materials, and therefore the total cost, in addition to meet market requirements.

Keywords: Power transformers
Magnetic shunts
Stray losses

1 Introduction

Losses in power transformers dissipate energy in the form of heat and are therefore the
main source of high local temperatures and degradation of insulations that measure and
limit transformer life. The efficient power transformer corresponds to lower losses,
which are the summation of load and no-load losses components. The load losses in the
power transformer, which generally dominate the total losses, consist of ohmic and
stray losses. However, reduction of the ohmic losses is limited by material conductivity
of the winding conductors. The stray losses are generated inside and outside the
windings. Losses outside the windings (leads, tank and clamping structure) can be
reduced through control of the leakage flux and the trade-off between different elec-
tromagnetic materials, such as magnetic shunts.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 15–26, 2020.
https://doi.org/10.1007/978-3-030-31676-1_2
16 A. Al-Abadi et al.

The stray losses in the structural parts of a power transformer are generally esti-
mated by empirical equations during the design stage. The advantage of using
empirical equations is to save computational time. The challenge of the empirical
equations is that they need to cover a wide range of design variants and complex
geometries with acceptable deviation in comparison to the measurement during the
testing stage. However, the high speed of computational time and accuracy of
numerical tools (e.g. finite-element method FEM) today make it possible to simulate
complex 3-D geometries for electromagnetic fields and precisely calculate the stray
losses during the early design stage. 3-D magnetic field FEM simulations were used to
investigate the effect of yoke-beam distance to windings, flitch-plate material and
design, and modified edge stack design on the losses and the localized hot-spot [1].
Different means were investigated and suggested with the use of FEM tools to reduce
the stray losses in different components of the power transformer [2–4].
Selecting the proper material for the shields helps to reduce the stray losses and
local hot spots in transformer metal components. The hot spots can be mitigated using
aluminum shields on the tank wall [5]. Magnetic shunts as compared to aluminum
shields can effectively reduce eddy current losses and decrease the rise in temperature.
Different structures of magnetic shunts in power transformers are studied using FEM
analysis [6–9]. Analyzing the stray losses in a power transformer, especially in tank,
using 3-D FEM to simulate the tank losses with and without magnetic shunts showed
that the tank losses may be decreased by about 4.3 kW [10]. The use of wall-mounted
horizontally oriented magnetic shunts was proposed to reduce stray losses in power
transformers, hence providing a cost-effective method for magnetic shielding material
[11].
The material of the magnetic shields not only influences the losses in the shield,
tank, and clamping-structure, but also on the short-circuit impedance [12]. Horizontally
mounted yoke magnetic shunts were studied for their effectiveness on loss reduction
[13, 14]. A 3-D transient FEM and magnetic thermal coupling method was used to
optimize the structure of magnetic shunts for minimum eddy current loss and tem-
perature rise [15], whereas [16] presented a case study transformer with applied yoke
shunts to control the leakage flux. A comparative study of the use of grain-oriented
electrical steels (GOESs) and non–oriented electrical steels (NOESs) in magnetic
shunts of power transformers showed that both have the same effect in reducing the
stray losses, temperature, and costs in power transformers without adding any risk to
their operation [17].
In the current study intensive investigations and optimizations of power transformer
magnetic shunts to control the leakage flux for lower stray losses are performed.
Eventually, complete design rules of the tank and yoke shunts are developed. The
developed design tools are applied on case study transformers with different power
ratings and short-circuit impedances. Results show significant reduction in short-circuit
losses. Economically, the newly developed design rules can be used to reduce the
emitted heat which provide the opportunity to optimize the final power transformer
design with a potential saving in materials and therefore the total cost in addition to
fulfilling market requirements.
 Optimization of Magnetic Shunts 17

2 Stray Losses

The load loss in power transformers consists of ohmic resistance loss in windings and
stray losses, which constitute 10–40% of total load losses. Stray losses appear in all
metal parts exposed to magnetic leakage flux. In addition to ohmic loss, they are
therefore the reason for any potential local hot spots in transformer components. Stray
losses in power transformers comprise: stray losses in conductors (e.g. windings, leads
and connecting conductors), stray losses in steel (e.g. clamping structure and tank),
stray losses in non-magnetic materials (e.g. tank aluminum shields) and stray losses in
the first packages of the core.
Stray losses occur in conductors due to the stray alternated flux of the conductor
itself and the stray flux of surrounding conductors. Both fluxes induce eddy currents in
the conductor skin which in turn resists to flow within the conductor and cause the eddy
losses. If the CTC conductors are not suitably transposed, the circulating currents
within the parallel conductors will increase due to unequally induced voltages which
are induced by the varying leakage flux between parallel strands. Generally, it can be
accurately calculated because of the well-known geometry and electromagnetic prop-
erties of conductors and windings.
Magnetic flux is generated when current flows in the windings, leads and con-
necting conductors. The main part of the magnetic flux, which is in-core plane, enters
the core limbs and yokes (low reluctance parts) to close its paths. This represents the
low stray losses part because the transformer’s core is laminated, which minimizes the
induced eddy current from the alternated magnetic field. The other part of the magnetic
flux, which is out-of-core plane, has two paths to close. The first part tries to enter the
core through the clamping structure (flitch plates and press beams) and the second part
is attracted to the tank due to its high relative permeability. The high conductivity
materials such as magnetic steel, which is typically used for clamping structure parts
and tank, induce a large eddy current when exposed to external magnetic flux. The
eddy current induces an opposite magnetic field to the original leakage magnetic flux.
The eddy current then flows through the materials and generate stray losses.
Controlling the leakage flux is decisive for minimizing the stray losses and
therefore mitigating the local hot spots, which has a direct impact on transformer life
time. Furthermore, decreasing the stray losses will increase the efficiency and lead to a
cost-cutting effect through savings in coolers, oil, and materials.

3 Shielding

Shielding is a common technique used to control the stray flux outside the windings
and therefore to reduce the losses and mitigation of local over-heating spots. Different
types of shielding are available. The shielding type is then selected based on the
required function.
18 A. Al-Abadi et al.

3.1 Non-magnetic Shields

Non-magnetic shields (such as non-magnetic steel) have a high resistance to the stray
flux. Maximum effectiveness of the non-magnetic shields will be achieved only when
applying it on parts exposed to small stray flux (tank cover, lead exits to bushings, etc.).
If the stray flux source is strong (e.g. top and bottom regions of the main windings) it
cannot be completely prevented from entering the non-magnetic material, so therefore
the stray losses can significantly increase due to the higher penetration depth as
compared to conventional magnetic steel.

3.2 Active Magnetic Shields

When placing conductive material such as aluminum or copper shields within an
alternating magnetic field, it will induce eddy current which in turn induces the
magnetic field opposite the source field. The strength of the source field will be
exponentially attenuated with the depth from the incident point (skin effect). Increasing
the strength of the source magnetic field will increase the induced eddy currents in the
conductive material shields, whereas the shielding effect will remain the same. The
active magnetic shields are effective in reducing the stray losses of the protected
part. Nevertheless, their reactive nature to the source magnetic field will push it towards
other metal components, which in turn can generate stray losses. As a result, the total
stray losses can remain the same.

3.3 Passive Magnetic Shields

Passive magnetic shields (or magnetic shunts), such as silicon steel sheets, effectively
attract the magnetic stray flux. The lamination of these types of shunts will only induce
a very small amount of eddy current, and therefore radically decrease stray losses.
However, it needs to be carefully designed in order to avoid magnetic saturation, which
can lead to excessive hysteresis losses or skipping of the overflow magnetic flux to the
protected parts. The design must consider the shunts material, cross-section area,
length, orientation of the lamination in respect to the stray flux, and orientation of the
shunt itself, etc.
Magnetic shunts are typically used to control the stray flux towards the tank (tank
shunts) and press-beams (yoke shunts). However, yoke shunts are mainly designed to
collect the main windings flux which provide three-phase paths for flux compensation.
As a result, tank shunts material can be significantly reduced.
There are many practical constraints when designing yoke shunts. Properly
designed shunts reduce the stray losses. However, yoke shunts are normally placed on
top and under press-beams, where they obstruct the route of leads and oil ducts. For
most cases, the space for mounting shunts is limited due to voltage clearances. In
 Optimization of Magnetic Shunts 19

addition, adding yoke shunts corresponds to more complex mechanical design,

increases the production time and the material of the shunt itself which in turn increases
the price of the transformer. This can be higher than possible savings due to loss
reduction. Eventually, the cost effectiveness of adding yoke shunts is guaranteed only
when the shunts design is optimized.

4 Measurements

In order to develop and validate a magnetic flux control approach by means of tank and
yoke magnetic shunts, a plan for placing measurement coils is to set up for two units.
The measuring coils placed on the tank shunts are used for measuring the leakage flux
towards the tank, whereas the measuring coils placed on the yoke shunts are used to
measure the leakage flux towards the clamping structure (press-beams and flitch-plates)
as shown in Fig. 1.

Fig. 1. Designed measurement coils for measuring the flux density in tank shunts (left) and in
yoke shunts (right).

The measured magnetic flux densities in tank and yoke shunts as per the designed
measurement coils at different nominal currents IN (100%, 75% & 50%), tap positions
(nom, plus & min) and frequencies of 50/60 Hz is shown in Fig. 2.
20 A. Al-Abadi et al.

Fig. 2. Measured flux densities at different ratings, tap positions and frequencies 50/60 Hz in
(top) tank shunts and (bottom) yoke shunts

5 FEM Analysis

FEM simulations are performed with ANSYS-Maxwell© to validate the measured

fluxes in different shunts. Figure 3 shows the agreement between the measured and
FEM calculated flux densities. After gaining a high confidence level of the FEM
calculations (meshing fidelity and convenient specified boundary conditions) versus
measurements, a mass of FEM simulations is further performed to investigate the
parameters influencing the design of tank and yoke shunts.

Fig. 3. Measured versus FEM calculated flux densities in the tank shunts.
 Optimization of Magnetic Shunts 21

The influencing parameters on the design of yoke shunts started by investigating

the effect of the ampere turns (N.I) on the calculated flux density in the tank and yoke
shunts as well as the flux density in the main gap, Fig. 4. The relation to short-circuit
impedance (uk%) is found, Fig. 4. The importance of using the impedance is because it
has the information of the magnetic energy and directly indicates the leakage flux
towards the tank and the clamping structure.

Fig. 4. (Left) ampere-turns (N.I), and (right) short-circuit impedance (uk%) versus flux density
in the tank and yoke shunts and main gap.

The distances between the main gap to the tank and yoke shunts are important
factors, since the distance decides the leakage flux attracted by the shunts. The distance
ratio is investigated, and the non-linear exponents of the tank and yoke fluxes are
found, Fig. 5.

Fig. 5. Tank/yoke to winding distance ratio versus flux densities in tank and yoke shunts.

In order to optimize the tank shunts for minimum losses while limiting the tank size
(compact design), parameters influencing the tank shunts design are investigated. These
parameters are: the distance from winding to tank, the cross-section area of the tank
22 A. Al-Abadi et al.

Fig. 6. Parameters influencing tank and yoke shunts design.

shunts, rated power and impedance. Relations are shown in Fig. 6, respectively. All
exponents are found as per relation.
The investigated parameters for tank shunts are formulated based on their
weighting and exponents factors as follow:
Cross sections of tank and yoke shunts are,


Ayoke ¼ f SN ; uk ; dtank ; dyoke ; Byoke ; Atank ; ½a; ½m ð1Þ


Atank ¼ f SN ; uk ; dtank ; dyoke ; Btank ; Ayoke ; ½b; ½n ð2Þ

Where [a] & [m] are vectors of weighting and exponent factors of yoke shunts
parameters, respectively. [b] & [n] are vectors of weighting and exponent factors of
tank shunts parameters, respectively. Btank & Byoke are the desired magnetic flux
densities in the tank and yoke, respectively. dtank & dyoke are distances from winding to
tank and yoke, respectively.
The investigated parameters influencing the shunts design have been first driven by
a case study transformer of (31.5 MVA, 12.4 uk%) and then the developed formulas
based on the studied parameters are applied to a range of transformers with different
rated powers and short-circuit impedances. With the developed design formulas, it is
possible to design the tank and yoke shunts. The magnetic flux density in each shunt
can be adjusted for the design goal. This gives the designer the possibility to trade-off
 Optimization of Magnetic Shunts 23

between minimum losses design (efficient) or compact design (economical). Applying

the design formulas on a case study transformer (77 MVA, 27.7 uk%) shows a sig-
nificant reduction of the flux density in tank shunts when adding yoke shunts as
compared to flux density in tank shunts without adding yoke shunts, Fig. 7.

Fig. 7. Flux density in the tank shunts without adding yoke shunts (left) and with adding yoke
shunts (right).

The eddy current losses in the magnetic shunts are computed with FEM by using
the 3-D transient analysis [17, 18]:
"      #
kc dBx ðtÞ 2 dBy ðtÞ 2 dBz ðtÞ 2
Peddy ðtÞ ¼ þ þ ð3Þ
2p2 dt dt dt

Assume the magnetic flux density B(t) in each direction of the magnetic shunt as

BðtÞ ¼ Bm sinðxtÞ ð4Þ

Substituting Eq. (4) in Eq. (3), to calculate the average eddy current loss of the first
harmonic term,

x 2 kc  2 
Peddy ðtÞ ¼ Bmx þ B2
my þ B2
mz 1 þ ejxt ð5Þ
4p2
where kc is the eddy current loss coefficient. Equation (5) represents the average eddy
current loss in the first harmonic term. In 3-D calculation, the eddy current losses in
magnetic shunts are obtained by integrating Eq. (5) in the volume of the magnetic
shunts.
The stray losses in the structural parts, such as tank and clamping structure are
calculated by applying surface impedance boundary conditions [19], which are cal-
culated as,
24 A. Al-Abadi et al.

Fig. 8. Stray losses in the clamping structure when using only tank shunts (left) and when using
tank and yoke shunts (right).

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Z
pf lrcs l0
Ps ¼ Ht
Ht0 dS ð6Þ
4rcs
S

where Ht is the tangential magnetic field, Ht0 is its conjugate, and S is the surfaces of the
structural part.
Calculation of stray losses in the clamping structure shows significant reduction
when using tank and yoke shunts in comparison to a design with only tank shunts,
Fig. 8.

6 Validation on Case Study

The developed design formulas are applied to a series of units of two case study
transformers (Trafo A: 77 MVA, 27.7% uk & Trafo B: 31.5 MVA, 12.4% uk) to
optimize the design of tank and yoke shunts, Fig. 9.
The original design of Trafo A was only with tank shunts. The original units are 1
& 2 and meet the guarantee value of nominal rating. Applying optimization on tank and
yoke shunts on four units (units 3–6) with the currently developed formulas results in
obvious reduction in short-circuit losses (Psc) of the three operation cases (nominal,
plus & minus). This reduction is mainly due to the reduction of stray losses (Pstray). The
same case applies to Trafo B (units 1–8), which was originally designed with aluminum
shields on the tank. Reduction of the Psc is clear in units 9 & 10. Reduction in Psc
represents the efficient design. The economical design is represented in units 11 & 12
of Trafo B, where the design is optimized for maximum material savings while Psc still
meets the guarantee value.
 Optimization of Magnetic Shunts 25

Fig. 9. Total short circuit losses (Psc) and stray losses (Pstray) of case study transformers.

7 Conclusions

In the current study, design tools of the influencing parameters on tank and yoke shunts
designs are developed based on FEM simulations and validated with measurements.
The developed design tools have proven their feasibility through application on case
studies with different power ratings and short-circuit impedances. The tools allow the
designer to trade-off between efficient design via loss reduction, and therefore less
emitted heat, or economical design via optimization for maximum material savings,
and therefore the total cost-cutting, in addition to fulfilling market requirements.

References
1. Adalja, C.C., Jain, M.L.: Analysis of stray losses in power transformers by 3-D magnetic
field simulation. In: Fifteenth National Power Systems Conference (NPSC), IIT Bombay
(2008)
2. Dave, K.V., Kanani, S.M.: Use of FEM for reduction of transformer stray loss. Int. J. Sci.
Eng. Technol. Res. (IJSETR) 5(2), 509–514 (2016)
3. Kralj, L., Miljavec, D.: Stray losses in power transformer tank walls and construction parts.
In: The XIX International Conference on Electrical Machines - ICEM 2010, Rome, pp. 1–4
(2010). https://doi.org/10.1109/icelmach.2010.5607891
4. Krasl, M., Jiřičková, J., Vlk, R.: Stray losses in transformer tank. In: AMTEE 2009: Eighth
International Conference on Advanced Methods in the Theory of Electrical Engineering,
Czech Republic (2009)
26 A. Al-Abadi et al.

5. Magdaleno-Adame, S., Olivares-Galvan, J.C., Escarela, R., Raichenko, O., Kladas, A.G.:
Hot spots mitigation on tank wall of a power transformer using electromagnetic shields. In:
International Conference on Electrical Machines (ICEM) (2014)
6. Hernandez, C., Arjona, M.A., Sturgess, J.P.: Optimal placement of a wall-tank magnetic
shunt in a transformer using FE models and a stochastic-deterministic approach. In: 12th
Biennial IEEE Conference on Electromagnetic Field Computation (2006)
7. Chen, Y., Yang, J., Yu, H., Tang, R.: Study on eddy current losses and shielding measures in
large power transformers. IEEE Trans. Magn. 30(5), 3068–3071 (1994)
8. Song, Z., Wang, Y., Mou, S., Wu, Z., Zhu, Y., Xiang, B., Zhou, C.: Tank losses and
magnetic shunts in a three-phase power transformer. In: 2011 International Conference
Electrical Machines and Systems (ICEMS) (2011)
9. Song, Z., Wang, Y., Mou, S., Wu, Z., Zhu, Y., Xiang, B., Zhou, C.: The edge effects of
magnetic shunts for a transformer tank. In: 2011 International Conference on Electrical
Machines and Systems (ICEMS), August 2011
10. Wang, Y., Song, Z., Chi, J.: Effect of magnetic shunts on the eddy losses of a transformer
tank. Adv. Mater. Res. 482–484 (2012). https://doi.org/10.4028/www.scientific.net/AMR.
482-484.1126. ISSN 1662-8985
11. Moghaddami, M., Sarwat, A.I., de Leon, F.: Reduction of stray loss in power transformers
using horizontal magnetic wall shunts. IEEE Trans. Magn. (2016). https://doi.org/10.1109/
tmag.2016.2611479
12. Tsili, M.A., Souflaris, A.T., Paparigas, D.G.: Geometry optimization of magnetic shunts in
power transformers based on a particular hybrid finite-element boundary-element model and
sensitivity analysis. IEEE Trans. Magn. 41(5), 1776–1779 (2005)
13. Djurovic, M., Monson, J.: 3-dimensional computation of the effect of the horizontal
magnetic shunt on transformer leakage fields. IEEE Trans. Magn. 13(5), 1137–1139 (1977)
14. Djurovic, M., Monson, J.: Stray losses in the step of a transformer yoke with a horizontal
magnetic shunt. IEEE Trans. Power Appar. Syst. PAS-101(8), 2995–3000 (1982)
15. Yang, H., Zhang, Y.: Analysis on the magnetic shunt structure of large power transformer.
IET Gener. Transm. Distrib. 12(11), 2755–2761 (2018). https://doi.org/10.1049/iet-gtd.
2017.1490
16. Štrac, L.: Three-phase shunts for stray magnetic field. Procedia Eng. 202, 183–188 (2017)
17. Magdaleno-Adame, S., Kefalas, T.D., Fakhravar, A., Olivares-Galvan, J.C.: Comparative
study of grain oriented and non–oriented electrical steels in magnetic shunts of power
transformers. In: IEEE International Autumn Meeting on Power, Electronics and Computing
(ROPEC), Ixtapa, Mexico (2018)
18. Lin, D., Zhou, P., Fu, W.N., Badics, Z., Cendes, Z.J.: A dynamic core loss model for soft
ferromagnetic and power ferrite materials in transient finite element analysis. IEEE Trans.
Magn. 40(2), 1318–1321 (2004)
19. Olivares-Galvan, J.C., Magdaleno-Adame, S., Campero-Littlewood, E., Escarela-Perez, R.,
Georgilakis, P.S.: Techno-economic evaluation of reduction of low-voltage bushings
diameter in single-phase distribution transformers. Electr. Power Compon. Syst. 39(13),
1388–1402 (2011)
 Assessment of Thermal and Electric Field
Characteristics of HVDC Cable According
to the Inner Filler Size of XLPE

Ho-young Lee1 , Ik-Soo Kwon1 , Mansoor Asif1 ,

Chae-Kyun Jung2 , Jae-Sang Hwang2, Min-Ju Kim2,
and Bang-Wook Lee1(&)
1
Hanyang University, 55 Hanyangdeahak-ro, Sangnok-gu,
Ansan, Republic of Korea
bangwook@hanyang.ac.kr
2
KEPCO Research Institute, 105 Munji-ro, Yooseong-gu,
Daejeon, Republic of Korea

Abstract. Increasing the rating of the HVDC cable causes an inevitable

increase in the heat generated inside cable. Thus, it is necessary to improve the
thermal conductivity of the Cross-linked polyethylene (XLPE) insulation
material used in the extruded cable for effective thermal management. Also,
several methods have been pro-posed so far to increase the thermal conductivity
of XLPE. However, the influence of the filler addition on the thermal conduc-
tivity has been mainly studied experimentally so far. Therefore, numerical
analysis was performed to investigate the effect of filler size and its content on
thermal conductivity of XLPE. In the modelling of XLPE composite for
numerical analysis spherical Al2O3 particles with a radius of 25, 50, 100 and
200 nm were considered. Additionally, filler contents were considered with up
to 20 wt%. Numerical analysis of XLPE composite model showed that the
effective thermal conductivity increased with the smaller filler size. Also, as the
content of filler increased, the effective thermal conductivity increased. A cou-
pled electro-thermal model considering the previously derived thermal con-
ductivities was used to investigate the influence of the thermal conductivity of
XLPE on temperature and electric field distribution. From the simulation results
it can be concluded that, higher thermal conductivity of XLPE can reduce the
thermal and electrical stress of the cable. Therefore, it is considered that the
smaller the size of the filler added to the insulation or the larger the content of
the filler, the thermal and electrical stress in the cable can be reduced.

Keywords: XLPE
HVDC cable
Filler
Thermal conductivity
Numerical

analysis

1 Introduction

Due to the progress in development of inter country power grid connection, world-wide
interest in long distance power transmission is increasing [1]. HVDC cable has been
widely used for long distance power transmission. Since the beginning of the current

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 27–37, 2020.
https://doi.org/10.1007/978-3-030-31676-1_3
28 H. Lee et al.

century, the use of extruded cables using XLPE insulation has dramatically increased
[2]. Therefore, for reliable long-term operation of HVDC XLPE cable, efforts for the
improvement of XLPE insulation are being actively carried out.
The thermal properties as well as the electrical properties of XLPE should be
considered to improve its reliability [3]. If the heat generated in the cable is not
dissipated to the outside environment, the physical and chemical characteristics of
dielectric could be changed which resulting in accelerated ageing of the cable. In
addition, the rise of temperature gradient causes greater electric field inversion and
intensification near sheath side [4]. Therefore, it is important to increase the thermal
conductivity of the insulation material to maintain the desirable temperature and
temperature gradient inside the cable.
Fillers can be added to XLPE to increase the breakdown strength and thermal
conductivity of HVDC extruded cables. The addition of a small amount of filler does
not dramatically increase the thermal conductivity. However, the temperature and
electric field of a DC cable is significantly influenced by a slight change in thermal
conductivity. Thus, if the filler is added in the adequate amounts, desirable thermal as
well as dielectric characteristics can be achieved for the lifetime of cable [5]. Efforts to
improve thermal conductivity by adding fillers to polymeric materials have been
actively carried out. There are many experimental and analytical studies on the
improvement of thermal conductivity according to filler content [6]. However, the
numerical analysis considering both the size and the content of the added filler is
insufficient. Therefore, in this paper, numerical analysis was performed to derive the
effect of filler size and content on thermal conductivity.
Prior to the numerical analysis, modelling of three-dimensional XLPE composite
was performed considering fillers of various sizes and contents. Also, the coordinates
of each filler particle were randomly generated and placed to inside of XLPE. As a
result, the effective thermal conductivity of each three-dimensional model could be
derived. In addition, electric field analysis was performed to investigate the effect of
filler size and content added to XLPE on cable temperature and electric field. The
electric field analysis of the 500 kV HVDC XLPE cable at steady state was performed
by applying effective thermal conductivity the derived by numerical analysis. Also, the
heat generated by the current flowing in the cable conductor was considered. Finally,
the influence of increase in thermal conductivity according to the size and content of
filler added to XLPE on cable temperature and electric field distribution was derived.

2 Derivation of Effective Thermal Conductivity from Three-

Dimensional XLPE Composite

XLPE used in cable is manufactured by crosslinking by applying high temperature and

high pressure to LDPE added with antioxidant, cross linking agent, filler and so on.
Thus, three-dimensional modelling of XLPE composite with nanofiller was performed
to confirm the effective thermal conductivity variation according to the size and content
of the filler in XLPE. Coordinates of filler particles were randomly generated and
imported to AutoCAD 2019 software for three-dimensional geometry drawing. After
that, numerical analysis of the three-dimensional model was carried out to derive
 Assessment of Thermal and Electric Field Characteristics of HVDC 29

effective thermal conductivity. COMSOL Multiphysics 5.4 was used for numerical
analysis.

2.1 Modelling of Three-Dimensional XLPE Composite

The kinds of filler added to XLPE include Al2O3, ZnO, MgO, and so on. However, we
only used Al2O3 in order to derive the change of thermal conductivity according to the
size of the filler. The radius of the filler was 25, 50, 100, and 200 nm, and the shape of
the filler was set to be spherical. The content of filler was set to 0, 5, 10, 15 and 20 wt%.
Three-dimensional models of XLPE composite were produced by randomly arranging
spherical Al2O3 filler particles inside LDPE with different numbers of particles
according to filler size and content. The three-dimensional model of XLPE composite
with various filler contents shown in Fig. 1 contains filler particles with a radius of
25 nm. The same modelling approach was adopted while considering different filler
sizes. The detailed material properties are shown in Table 1.

Fig. 1. Three-dimensional modelling of XLPE composite for calculating effective thermal

conductivity - spherical filler with a radius of 25 nm.

Table 1. Properties of materials used in three-dimensional modelling.

Material LDEP Al2O3
Thermal conductivity in W/(m K) 0.33 42
3
Density in kg/m 950 3950
Heat capacity in J/(kg K) 1800 950

2.2 Set Up of Boundary Conditions

The effective thermal conductivity of a three-dimensional model of XLPE composite
can be calculated from the heat flux due to the temperature gradient of the top and
bottom surfaces of the regular hexahedron. The Boundary condition of three-
dimensional model for numerical analysis is set as shown in Fig. 2.
In order to simulate the internal temperature gradient and the heat flow in the
numerical analysis, a temperature gradient of 100 K was set by applying 273.15 K to
the top surface and 373.15 K to the bottom surface. In order to consider the heat flux
30 H. Lee et al.

generated only along Z-axis. The sides of the cube were set to adiabatic condition. The
effective thermal conductivity was calculated from Eq. (1) [7].
The effective thermal conductivity derived by numerical analysis is compared with
the approximation method proposed by Maxwell [8]. The Maxwell approximation is
given by Eq. (2) and can be used to derive the effective thermal conductivity of a
polymer composite containing randomly arranged fillers. However, the Maxwell
approximation method can calculate the change of the thermal conductivity according
to the content of the filler, but there is a limitation that the influence of the size of the
filler cannot be considered.

Fig. 2. Boundary condition of three-dimensional model for numerical analysis.

keff ¼ £
d=DT ð1Þ

where: keff = Effective thermal conductivity (W/m K)

£ = Heat flux density in the z-direction (W/m2)
d = thickness of the cube in z-direction (m)
DT = Temperature difference between the top and bottom surfaces (K).


2km þ kf  2 km  kf u
keff ¼ km
ð2Þ
2km þ kf þ 2 km  kf u

where: keff = Effective thermal conductivity (W/m K)

km = Thermal conductivity of polymeric materials (W/m K)
kf = Thermal conductivity of filler (W/m K)
u = Volume of added filler (m3).
 Assessment of Thermal and Electric Field Characteristics of HVDC 31

2.3 Effective Thermal Conductivity According to Filler Size and Content

Numerical analysis was performed on the three-dimensional model of XLPE compo-site
to obtain effective thermal conductivity. In order to compare with numerical analysis
results, the thermal conductivity according to the content of filler was calculated by
Maxwell approximation. The effective thermal conductivity according to filler size and
content is shown in Fig. 3, and detailed values are summarized in Table 2. As the filler
size decreased, the effective thermal conductivity of the XLPE composite increased.
Also, the effective thermal conductivity increased with the in-crease in filler content.
The effect of filler size on effective thermal conductivity in-creased with increasing filler
content. The effective thermal conductivity calculated by the Maxwell approximation
also increased with increasing filler content. Also, it showed higher values than the
effective thermal conductivity derived from the numerical analysis in all cases.

Fig. 3. Effective thermal conductivity according to filler size and amount.

Table 2. Result of effective thermal conductivity derivation.

Effective thermal conductivity in W/m K
Filler radius 5 wt% 10 wt% 15 wt% 20 wt%
200 nm 0.3437 0.3591 0.3744 0.3963
100 nm 0.3441 0.3613 0.3837 0.3986
50 nm 0.3449 0.3618 0.3840 0.4039
25 nm 0.3451 0.3627 0.3844 0.4133
Maxwell approximation 0.3467 0.3664 0.3897 0.4178
32 H. Lee et al.

3 Electric Field Analysis of HVDC XLPE Cable

The coupled electro-thermal analysis was performed to investigate the influence of the
thermal conductivity of the insulation on the cable internal temperature and electric
field distribution. The steady state analysis was performed on the 500 kV
HVDC XLPE cable geometry. In addition, coupled electro-thermal model reflects both
electrical and thermal characteristics. The temperature gradient inside the cable was
simulated by considering the cable heating due to the current flowing in the conductor.
In order to compare the temperature distribution according to the thermal conductivity
of the XLPE, the effective thermal conductivity calculated from the numerical analysis
of the three- dimensional model of XLPE composite was applied.

3.1 Electro-Thermal Coupled Modelling of 500 kV HVDC XLPE Cable

For 500 kV XLPE cable modelling, copper conductor and XLPE insulation layer were
considered, and the surrounding material was set to air [9]. The geometry of the cable is
shown in Fig. 4, the radius of the conductor is 30.2 mm, and the thickness of the
insulation layer is 27 mm. Air around the cable was set at 293.15 K to consider the
room temperature condition. The voltage of the cable conductor was set to 500 kV and
a high current of 2500 A was applied to heat up the XLPE cable to its maximum
allowable temperature i.e., 363.15 K.

Fig. 4. Cable geometry for electric field analysis.

The heat generated by the conductor can be calculated by Eq. (3) which relates to
the current and resistance of the conductor [10]. The value of the conductor resistance
must be calculated before calculating the heat generated by the conductor [11]. Since
the resistance of a conductor varies with temperature, it can be derived from Eq. (4).
The electrical conductivity of the insulation material varies depending on the
 Assessment of Thermal and Electric Field Characteristics of HVDC 33

temperature and electric field strength. Therefore, Eq. (5), which considers the tem-
perature and electric field dependence of the conductivity of the XLPE, is applied to the
analysis of the DC electric field of the electro thermal model. The thermal conductivity
of the XLPE was derived from the numerical analysis and Maxwell approximation. The
detailed physical properties used in the electric field analysis by coupled electro-
thermal model are shown in Table 3.

Table 3. Material properties of XLPE for electric field analysis.

3.2 Distribution of Temperature Inside Cable According to Thermal

Conductivity of XLPE
Numerical analysis of the three-dimensional model of XLPE composite revealed effec-
tive thermal conductivity to be in the range of 0.33 to 0.42 W/(m K) depending on the
particle size and content. Thus, the thermal conductivity increased from 0.33 to 0.42 W/
(m K) in steps of 0.03 W/(m K) was applied to electric field analysis. The distribution of
temperature inside the cable is shown in Fig. 5. Electro thermal coupled model can reflect
the conduction of heath generated due to joule heating to outside through the insulation
layer and the cable cooling by the outside air. In the case of XLPE having a thermal
conductivity of 0.33 W/(m K) without any filler content, the conductor side temperature
reached 90.97 °C. Whereas, the temperature at the sheath was 64.15 °C. Therefore, a
temperature gradient in the case without filler was 26.82 °C. In the case of XLPE having a
thermal conductivity of 0.36 W/(m K), the conductor side was heated to 88.49 °C and the
34 H. Lee et al.

sheath side to 64.01 °C. The temperature gradient between the conductor side and
the sheath side was 24.48 °C. In the case of XLPE with a thermal conductivity of 0.39 W/
(m K), the conductor side was heated to 86.39 °C and the sheath side to 63.88 °C, and a
temperature gradient of 22.51 °C occurred. When the heat conductivity of XLPE is
0.42 W/(m K), the conductor side was heated to 84.62 °C and the sheath side was heated
to 63.78 °C, and a temperature gradient of 20.84 °C occurred.
As the thermal conductivity of XLPE increases from 0.33 W/(m K) to 0.42 W/(m K),
the conductor side and sheath side temperature are lowered by 6.35 °C and 0.27 °C,
considering the same quantity of joules heating. The temperature gradient also decreased
by 5.98 °C. The higher the thermal conductivity of the XLPE, the faster the heat gen-
erated by the conductor is dissipated to the outside of the cable, there-by lowering the
temperature and temperature gradient. Therefore, the increase of thermal conductivity
through the addition of the inner filler of insulation material will prevent thermal aging of
XLPE, which will also aid the reliable long-term operation.

Fig. 5. Temperature distribution of cable due to thermal conductivity.

3.3 Electrical Conductivity and Electric Field Distribution of Cable

According to Thermal Conductivity of XLPE
The electric field distribution of the DC cable is determined by the electrical con-
ductivity of the insulation material. Electrical conductivity of insulation depends on
temperature and field strength. As mentioned above, according to the thermal con-
ductivity of the XLPE insulation material, the temperature distribution inside the cable
and the temperature gradient inside the insulation are different. Therefore, the electrical
 Assessment of Thermal and Electric Field Characteristics of HVDC 35

Fig. 6. Electrical conductivity distribution of cable according to thermal conductivity.

conductivity inside the insulation layer also showed a considerable gradient, and the
distribution of electrical conductivity inside the insulation due to the thermal con-
ductivity is shown in Fig. 6.
When the thermal conductivity of the XLPE was 0.33 W/(m K), the conductivities
of the conductor side and the sheath side were 1.52E−13 S/m and 2.06E−14 S/m,
respectively, that of conductor side being about 7.38 times higher. When the thermal
conductivity increased to 0.36 W/(m K), the conductivities of the conductor side and
sheath side were 1.22E−13 and 1.94E−14 S/m, respectively, and that of conductor side
being 6.29 times higher. When, the thermal conductivity of XLPE was 0.39 W/(m K),
the conductivities of the conductor side and the sheath side were 1.02E−13 and 1.85E
−14 S/m, respectively, and the conductor side was about 5.51 times higher. When the
thermal conductivity of XLPE was 0.42 W/(m K), the conductivities of the conductor
side and the sheath side were 8.72E−14 and 1.78E−14 S/m, respectively, and the
conductor side was 4.9 times higher. In all cases, there was a difference in electrical
conductivity between the conductor side and the sheath side. As a result, an electric
field inversion occurs resulting in intensification of electric fields on the sheath side. An
electric field inversion occurred in the cable due to the electrical conductivity gradient
between the conductor side and the sheath side. The cable electric field distribution
according to the thermal conductivity is shown in Fig. 7.
The thermal conductivity of the XLPE was 0.33 W/(m K), the electric field
intensity was 7.69 and 30.36 kV/mm on the conductor side and the sheath side,
respectively. When the thermal conductivity of the XLPE was 0.36 W/(m K) the
conductor side electric field was 8.64 kV/mm and the electric field intensity was
28.95 kV/mm at the sheath side. The maximum electric field intensity was 4.64%
lower than that when the thermal conductivity was 0.33 W/(m K). As the thermal
36 H. Lee et al.

conductivity of the XLPE increased to 0.39, the electric field intensity was 9.5 and
27.69 kV/mm on the conductor side and the sheath side, respectively. In addition, the
maximum field strength decreased by 8.79% as compared with the case where the
thermal conductivity was 0.33 W/(m K). The thermal conductivity of the XLPE
increases to 0.42, the electric field intensity was 10.26 and 26.79 kV/mm on the
conductor side and the sheath side, respectively. In addition, the maximum field
strength decreased by 11.76% as compared with the case where the thermal conduc-
tivity was 0.33 W/(m K).
When the thermal conductivity of the insulation was 0.33 W/(m K), the tempera-
ture gradient inside the cable was the highest. Also, the difference in conductivity
between the conductor side and the sheath side was the highest. Therefore, if the
thermal conductivity of the insulation material is poor, the electric field inversion inside
the cable will be intensified. Therefore, the electric field analysis results show that the
electrical stress generated in the cable can be reduced if the thermal conductivity of the
insulation material is outstanding.

Fig. 7. Electric field distribution of cable according to thermal conductivity.

4 Conclusions

To improve the long-term reliability of HVDC cables, a numerical study has been
conducted to increase the thermal properties of XLPE. The effect of the size of the filler
added to the XLPE on the thermal conductivity was derived through numerical anal-
ysis. Smaller the size of filler in XLPE, the higher was the thermal conductivity. And,
 Assessment of Thermal and Electric Field Characteristics of HVDC 37

as the filler content increased, the thermal conductivity changes with filler size also
increased.
Additionally, the electric field analysis using effective thermal conductivity
achieved from numerical analysis was also carried out. Higher the thermal conductivity
of the XLPE, faster the heat dissipation to the outside of the cable, resulting in a
decrease in temperature and temperature gradients. Therefore, better the thermal con-
ductivity of the XLPE, lesser will be the electric field reversal of the cable. Even if the
thermal conductivity of the XLPE increased slightly, the temperature of the cable and
the maximum electric field reduced significantly.
In subsequent studies, the consistency of the simulation should be verified by
measuring the thermal conductivity according to the size and content of the filler. In
addition, several methods will be devised for determining the optimum trade-off
between thermal conductivity and dielectric strength of XLPE by electrical experiment.

References
1. Feng, X., Misra, S., Xue, G., Yang, D.: Smart grid—the new and improved power grid: a
survey. IEEE Commun. Surv. Tutor. 14, 944–980 (2011)
2. Chen, G., Hao, M., Xu, Z., Vaughan, A., Cao, J., Wang, H.: Review of high voltage direct
current cables. CSEE J. Power Energy Syst. 1, 9–21 (2015)
3. Mecheri, Y., Bouazabia, S., Boubakeur, A., Lallouani, M.: Effect of thermal ageing on the
properties of XLPE as an insulating material for HV cables. In: International Electrical
Insulation Conference, Birmungham (2013)
4. Huang, Z.Y., Pilgrim, J.A., Lewin, P.L., Swingler, S.G.: Real-time electric field estimation
for HVDC cable dielectrics. In: 2015 IEEE Electrical Insulation Conference, Seattle (2015)
5. Li, Z., Du, B.: Polymeric insulation for high-voltage DC extruded cables: challenges and
development directions. IEEE Electr. Insul. Mag. 34, 30–43 (2018)
6. Lee, G.W., Park, M., Kim, J.K., Lee, J.I., Yoon, H.G.: Enhanced thermal conductivity of
polymer composites filled with hybrid filler. Compos. A Appl. Sci. Manuf. 37(5), 727–734
(2006)
7. Kiran, R.: Thermal conductivity of polymer composites filled with nanofillers (2016). https://
doi.org/10.20944/preprints201608.0216.v1
8. Maxwell, J.C.: A Treatise on Electricity and Magnetism. Oxford Classic Texts in the
Physical Sciences, vol. 1. Clarendon Press, Oxford (1873)
9. Yonemoto, N., Muneta, Y., Yamanouchi, H, Seo, S., Kumada, Y., Itoh, M., Kunimura, S.,
Nakamura, S., Fujii, Y., Ishii, T.: Construction of the world’s first long-distance 500 kV
XLPE cable line. Fujikura Technical Review (2003)
10. Zhan, Y., Chen, G., Hao, M.: Space charge modelling in HVDC extruded cable insulation.
IEEE Trans. Dielectr. Electr. Insul. 26, 43–50 (2019)
11. Mardiana, R.: Parameters affecting the ampacity of HVDC submarine power cables. In: 2011
2nd International Conference on Electric Power and Energy Conversion Systems (EPECS),
Sharjah (2011)
 Study on Conducted EMI Based on Wide-Band
Model for a –500 kV MMC-HVDC Station

Weidong Zhang(&) and Yuning Wei(&)

School of Electrical and Electronic Engineering,

North China Electric Power University, Beijing, China
zwd@ncepu.edu.cn, shuilalala@163.com

Abstract. In China, the HVDC technology has been widely applied for the
advantage of large capacity as well as high efficiency. The HVDC project
adopted modular multi-level converters (MMC) has attracted more and more
attention. Zhangbei MMC-HVDC power transmission project is the first four-
terminal ±500 kV MMC-HVDC transmission system in the world. The wide-
band frequency noise is generated during the periodic turn-on and turn-off
process of the IGBTs in converter valve. The noise will transmit along the
primary circuit in the high voltage switch yard and may interfere with the
secondary system, such as the power line communication (PLC) systems,
control systems, and so on. A wideband equivalent model of the primary
electrical system from the hall of the converter valve to the AC grid was built in
this paper. It was to simulate conducted electromagnetic interference
(EMI) along the primary high voltage circuit during the steady operation pro-
cess. Then, the response of the HVDC system influenced by equipment char-
acteristics was described. The Influence of the primary equipment along the
propagation route of the EMI was analyzed, and the effect of the PLC filter in
the broadband equivalent circuit was discussed. The wideband equivalent model
was based on the scheme of Zhangbei ±500 kV MMC-HVDC project and the
operating characteristics of the HVDC system. The wideband equivalent model
consists of the equivalent circuit model of the major equipment in HVDC
system, including converter transformers, reactors, AC filters, PLC filters, and
the wiring forms between them. The wideband equivalent model was deter-
mined according to measurements of equipment parameters and other HVDC
engineering applications considering high frequency characteristic. The excita-
tion source containing EMI derived from the IGBTs was obtained from the test
result of the converter valve, which were assembled to form the converter valve
used in the Zhangbei project.

Keywords: MMC-HVDC
EMI
PLC

1 Introduction

When the converter station operated normally, the voltage and current at both ends of
the sub-module will abruptly change during the process of the switching-on and
switching-off of the converter valve. These changes generate wide frequency band
disturbance in the circuit. The disturbance signal will affect the measurement and

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 38–47, 2020.
https://doi.org/10.1007/978-3-030-31676-1_4
 Study on Conducted EMI Based on Wide-Band Model 39

communication circuit of the secondary system through conduction, near-field coupling

and radiation. A computing model is built based on the planning of the Zhang-
bei ±500 kV MMC-HVDC Transmission Project and the operation characteristics of
the existing HVDC project.
The model ranges from the outlet of the valve hall of the converter valve to the AC
network. The measured waveform of sub-module ports of the same type of converter
valves applied in the project is introduced as the excitation conditions to obtain the
reaction characteristics and frequency components of it on the AC side of the converter
station. This paper analyses some frequency band noise in the power transmission
system, hoping to provide some reference for setting up communication facilities along
the line in the future, especially for Power Line Carrier communication (PLC) system
[1–3].

2 Converter Station Computation Model

According to the planning and design of MMC-HVDC Transmission Project, it is

expected that four converter stations will be constructed to form a four-terminal ring
network and their interconnection modes are shown in Fig. 1.

Kangbao Fengning
station station
205.1km
187.1km
49.6km

206.4km
Zhangbei B eijing
station station

Fig. 1. Diagram of the converter station of the Zhangbei ±500 kV MMC-HVDC project.

Taking one converter station as an example, this paper uses PSCAD platform to
build a wide-frequency circuit model of converter station equipment, including con-
verter transformer, reactor, lightning arrester, filter and AC line, etc., as in Fig. 2.
According to the measured output voltage of the sub-module port in converter valve,
the distribution of the conducted disturbance in the AC field of the converter station is
obtained by using time domain method, and the noise distribution in the PLC frequency
band is analyzed.

2.1 Converter Transformer

Converter transformer is an important equipment of converter station, which connects
converter valve and AC network. The operation characteristics of transformers and its
equipment parameters at low frequencies are taken into account in calculation,
40 W. Zhang and Y. Wei

Fig. 2. Wiring diagram of a converter station of Zhangbei ±500 kV MMC-HVDC project.

including leakage impedance R1, L1, R2 and L2 of primary and secondary sides,
representing magnetic circuit coupling, iron loss and core saturation characteristics
respectively. Its equivalent circuit model is shown in Figs. 3 and 4.

Fig. 3. The wideband model of a converter transformer.

Fig. 4. The equivalent circuit of the converter transformer.

The parasitic parameters of converter transformer at high frequency need to be

considered. The broadband model of converter transformer ensures the accuracy of the
calculation results in high frequency part [5]. There is no need to consider a more
accurate model (frequency dependent model) for the reason that PLC communication
systems usually work under the frequency range from 20 kHz to 500 kHz in China.
Parameters almost remain unchanged during this range. The capacitance effect of
windings at high frequencies is expressed by lumped parameter capacitance. The stray
capacitance of primary coil to ground is expressed by C1 and C10. The stray
 Study on Conducted EMI Based on Wide-Band Model 41

capacitance of secondary coil to ground is C2 and C20. The stray capacitance between
primary coil and secondary coil is C12. The capacitance between primary coil and
secondary coil is Cg1 and Cg2, respectively. The high frequency parameters come from
the experiments of the manufacturer.

2.2 Reactor Model

In most cases, the reactor model needs to consider the main function of the reactor and
the loss of equipment. The equivalent circuit is shown in Fig. 5. R1 is the equivalent
resistance of equipment loss; L1 is the inductance; C1 is the inter-turn stray capacitance
of reactor windings; Cac is the stray capacitance at AC network side; Cv is the stray
capacitance to the ground at exchange valve side. The equivalent lumped parameters
are in the Table 1 as follows.

Fig. 5. The equivalent circuit of the reactor.

Table 1. The wide-band parameters of the reactor

Wide band parameters Value
Inductance 0.15 H
Equivalent resistance 0.062 X
Inter-turn stray capacitance 0.3 lF
Stray capacitance at AC 0.15 lF
Stray capacitance to the ground 0.15 lF

2.3 PLC Filter

In this calculation, a hybrid C filter, which combines parallel mono-tuning and series
mono-tuning structures, is selected. The parameters of each element are shown in
Table 2 [6].
42 W. Zhang and Y. Wei

Table 2. The wide-band parameter of the C model of PLC filter

Wide band parameter Value
R1 900 X
C1 2.587 nF
L1 2 mH
R2 5X
C2 200 nF

2.4 Lightning Arrester

MMC-HDC system usually installs lightning arrester at DC line, converter neutral line
and converter valve outlet to protect equipment and lines from fault and lightning
overvoltage.
In this paper, the circuit model of arrester proposed by IEEE is selected. Its
parameters are determined by the structure and function of arrester. Its circuit model
includes non-linear resistance A0, A1 and ground stray capacitance C. Its equivalent
circuit is shown in Fig. 6.

Fig. 6. Diagram of the surge arrester model proposed by IEEE.

When the height of the arrester is d (m) and the number of columns of the arrester
shunt resistor is n, the lumped element parameters are selected as below [7, 8]:

L0 ¼ 15d=n ðlHÞ ð1Þ

R0 ¼ 65d=n ðXÞ ð2Þ

L1 ¼ 0:2d=n ðlHÞ ð3Þ

R1 ¼ 100d=n ðXÞ ð4Þ

C ¼ 100n=d ðpFÞ ð5Þ

 Study on Conducted EMI Based on Wide-Band Model 43

3 Equivalent Circuit Wideband Computation

3.1 Excitation Signal
In this paper, the excitation signal (Fig. 7) used in calculation is the waveform mea-
sured in equipment test. It can be considered that the proportion of each frequency
component is the same as the output voltage of the converter valve in actual operation.

Fig. 7. The test excitation waveform.

3.2 Computation Results

After that, the circuit from the AC side of the converter valve to the AC network side is
analyzed and calculated in several switching cycles of the converter valve sub-module.
Three measuring points are set in the main circuit to record the current and voltage at
the outlet of the converter transformer and the converter valve, and at the bus con-
necting the AC network. According to the data obtained, the electromagnetic envi-
ronment and conducted electromagnetic disturbance of AC system in converter station
are as followed.
The voltage and current waveforms at the outlet of the converter valve are shown in
Fig. 8. The waveform is close to the excitation signal, and the amplitude is basically
unchanged. Because of the waveform characteristics of the excitation source, there is
an overshoot phenomenon in the process at the switch on of the converter valve, which
forms a peak. The voltage at the outlet of the converter valve varies more than 2000 V
in less than 10 ls, and the high frequency component is very rich. The peak current
reaches 1200 A. It shows a current pulse with an amplitude exceeding 1000 A
appearing in less than 5 ls.
44 W. Zhang and Y. Wei

Fig. 8. Simulation results at the reactor (V1, I1) and converter transformer (V2, I2).

The output voltage of the converter valve is transformed by Fourier transform to

summarize its frequency domain characteristics. Its voltage spectrum is shown in
Fig. 9. It can be seen that it mainly distributes in the low frequency part, and the DC
component accounts for the vast majority. The high frequency component decreases
with the increase of frequency, and the voltage spectrum is concentrated at some
frequency points between 0 and 1 MHz.

Fig. 9. The spectrum density voltage on the reactor.

 Study on Conducted EMI Based on Wide-Band Model 45

Some representative frequency points is selected to record the disturbance level in

Table 3. The frequency components showed a gradual downward trend, and the
downward speed gradually slowed down, from 96 dBlV/Hz to 52 dBlV/Hz at 10–
1000 kHz, and then to below 52 dBlV/Hz after 1 MHz.

Table 3. Conducted EMI level of the converter.

Frequency point (kHz) Value (dBlV/Hz)
10 137
20 129
50 122
100 112
200 106
500 100
1000 93

4 Computation Result in a Period

In order to simulate the primary side waveform during the actual operation of the
project, according to the modulation strategy of the converter valve, the voltage
waveform of the sub-module port of the converter valve is used to fit the AC voltage
waveform with an amplitude of 250 kV, as shown in Fig. 10.

Fig. 10. The excitation signal (upper) and its stretch-out view.

The voltage and current waveforms at the outlet of the converter valve are shown in
Figs. 10 and 11. The time-domain voltage at the outlet of the converter valve before the
converter transformer is a step wave. The amplitude of a single step is 2 kV and the
amplitude is 250 kV. Due to the waveform characteristics of the excitation source,
there is a periodic jump phenomenon during the switch on of the converter valve. The
voltage in the transient process of a single step has a variation range of more than
2000 V in less than 10 ls, and the high frequency component is very rich. The peak
46 W. Zhang and Y. Wei

Fig. 11. The current at the reactor (upper) and its stretch-out view.

current is more than 4 kA. It oscillates rapidly in a short time, showing peaks one by
one. After taking a short time window, it can be found that the peak value of the current
at the outlet of the converter valve is 3 kA with a short pulse in 33 ls. It can be seen
that the wave characteristic analysis of the current can be simply replaced by this part
of the waveform in 33 ls to shorten the calculation time.

5 Conclusions

In this paper, the influence of conducted electromagnetic disturbance caused by switch

on and switch off of the converter valve in normal operation is investigated for the
MMC-HVDC system. The measured waveform of the sub-module ports is introduced
as the excitation conditions and used to fit the voltage waveform of the converter valve
outlet in actual operation. The conducted electromagnetic disturbance under actual
operation conditions is analyzed. The frequency spectrum analysis of the specific
frequency band noise is carried out to obtain the response characteristics of the
equipment on the AC side of the converter station in this transient process.
 Study on Conducted EMI Based on Wide-Band Model 47

The calculation results show that when the outlet voltage of the sub-module is the
excitation source, the maximum frequency domain component of the conducted dis-
turbance in the frequency band of 10–1000 kHz decreases from 96 dBlV/Hz to
52 dBlV/Hz. After adding PLC filter, it can restrain electromagnetic disturbance of
10–500 kHz in power line carrier frequency band.
The results of period wave computation show that the frequency domain compo-
nent of AC side voltage mainly distributes in the low frequency part, and the DC
component accounts for the vast majority.

Acknowledgement. This work is supported by The National Key Research and Development
Program of China under grant No. 2016YFB0900904.

References
1. Wu, F., Ma, Y., Mei, N., Zou, X.: Design of main connection scheme for Zhoushan flexible
multi-terminal HVDC transmission project. Power Syst. Technol. 38(10), 2651–2657 (2014)
2. Sun, H.: Research and application of wideband circuit modelling method of converter systems
in converter stations. North China Electric Power University (2009)
3. Li, H., Wu, W., He, Z.: Insulation coordination for the high-voltage bulk power transmission
VSC-HVDC systems. Power Syst. Technol. 40(6), 1908–1913 (2016)
4. State Power Economic Research Institute: Research on Insulation Coordination (2018)
5. Yu, Z., He, J., Zhang, B.: Time-domain simulation of conducted EMI caused by HVDC
valves in substations. Proc. CSEE 29(10), 17–23 (2009)
6. Xu, C.: Researches on the high orders harmonics and their filters in UHVDC transmission
systems. North China Electric Power University (2009)
7. Jones, R.A., Clifton, P.R., Grotz, G.: Modelling of metal oxide surge arresters. IEEE Trans.
Power Deliv. 7(1), 302–309 (1992)
8. Xu, W.: Over voltage analysis and protection of converter station in MMC-HVDC system.
North China Electric Power University (2014)
 Study on Simulation and Measurement
of EMF in Transmission Underground Cables

Jung Eui-hwan(&), Lee Soo-bong, Park Sang-taek,

and Kim Jung-nyun

LS Cable & System, 228 Suchul-daero, Gumi-si, Gyeongbuk, South Korea

sikemaro@lscns.com

Abstract. In this paper the finite element analysis method and the experiment
when mitigating the magnetic field generated by underground power cables are
analyzed. We installed a mock test line and measurement system to obtain the
magnetic field data. Based on simulations made with the finite elements method
software Maxwell 2D this paper presents mitigation offered by Aluminium plates.
A 154 kV Power cable is installed on mock test line in a phase loop shape.
And the current source is connected to control the conductor current. In addition,
we measured the phase difference between the conductor current and the sheath
circulating current using Ni-DAQ. The magnetic field around the power cable
measured by the EFA-300 device. Finally, a measurement system was installed
to enable synchronized data acquisition in real time using NI-lab view program.
In the mock line, the measured results obtained at a distance of 1M above the
ground. The measured EMF data were compared with the simulation data. As a
result, the difference between the magnetic value and the simulation value was
about 3%.

Keywords: EMF
Underground cables
Measurement vs. simulation

1 Introduction

Due to the recent improvement in living standards, environmental and health problems
have been socially highlighted, and people are frequently exposed to electromagnetic
fields generated from electric power facilities [1]. To face the high increasing electric
power demands in metropolitan cities the next decade, a greater electrical power
transmission network capacity is required. Several technical publications are describing
the ways to engineer power cable-circuits with low electromagnetic field emission at
the close vicinity of cables [2]. However, there is a lack of specific researches and
research data on electromagnetic disturbance and actual conditions. Especially, in case
of extremely low frequency electromagnetic field, exposure to the human body for a
long time has a negative effect on the human body [3].
On the other hand, many people have raised questions about the severity of the
effects of very low frequency electromagnetic fields associated with underground
transmission and distribution lines, but researchers and technicians do not place EMF
reduction management in the area of interest. However, while studies on the human
body are being conducted, only the conductor management technology for the

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 48–54, 2020.
https://doi.org/10.1007/978-3-030-31676-1_5
 Study on Simulation and Measurement of EMF 49

influence of the magnetic field in the cable has been studied as a practical alternative.
However, these techniques have many problems in practical applications due to the
limitation of high current control. In order to compensate the technical limitations of
the Conductor Management technology, basic researches on EMF mitigation tech-
nology such as shielding technology using conductor, magnetic materials, etc. [4–8].
These studies consider either measurement or simulation techniques, being the most
extended ones the international standard approach and 2-D finite element analysis
approach. Simulation line was installed. And the study compared measure EMF data to
simulation data.

2 Experiment and Simulation

2.1 Configuration of Mock Line

Figure 1 shows the schematic diagram of the mock line for Electro-magnetic field
measurement around the underground transmission cable. As shown in Fig. 2, a current
loop are connected to the cable conductor and lead sheath, respectively. All metal
materials have been removed to minimize the factors affecting the magnetic field
measurement.

Fig. 1. Configuration of mock line.

Fig. 2. Picture of underground power cable.

50 J. Eui-hwan et al.

2.2 EMF Measurement Equipment

The EFA-300 model with an accuracy within 0.1% was used in the measurement range
of 0–320G. Figure 3 and Table 1 show the detailed functions of measurement equip-
ment (H/W) and data analysis program (EFA-TS, S/W).

Fig. 3. EMF measurement equipment.

Table 1. Specifications of EMF measurement equipment

Spec.
Hardware Model EFA-300 (NARDA-STS)
Method of measurement Integrated coil system
Evaluated measurement axes 3-dimensional(isotropic) x, y, z
Frequency range 5 Hz–32 kHz
Measurement range 0–32 mT (0–320G)
Accuracy <±0.1%
Software Model EFA TS (NARDA-STS)
Function Recode, Remote control

2.3 Temperature Measurement

The temperature measurement using T/C (Thermo-couple) was performed to compare
measured data. T/C was installed at three points of cable conductor and ground. In
order to minimize the influence of outside temperature, between power cable and the
sensor connection were reinforced by soldering.
 Study on Simulation and Measurement of EMF 51

2.4 Monitoring System

The factors to be measured for the EMF influence around the underground transmission
cable are largely classified into conductor, sheath current, magnetic field, and tem-
perature. In particular, an integrated measurement system is capable of real-time
monitoring. It save data of magnetic field, sheath, conductor temperature by NI-DAQ,
Lab-view system, Respectively (Fig. 4).

Fig. 4. Integrated measurement equipment.

The conductor and sheath current can be measured using a Rogowskii coil type
current probe (CWT-LF) with an accuracy of 2% in the range of 0 to 3 kA. The current
magnitude and phase difference can be displayed using NI-Lab view Respectively. In
addition, J-Type TC is connected to conductor and sheath for temperature measure-
ment, and 16 channels are programmed to measure and store temperature in real time.
Finally, the magnetic field measurement values through the EFA-300 were configured
on the same PC using an RS-232 serial communication and synchronized with the
current and temperature measurements. Figure 5 shows the integrated monitoring
system main screen configured with Lab-view.

Fig. 5. Monitoring screen of integrated measurement equipment.

52 J. Eui-hwan et al.

The conductor and sheath current can be measured using a Rogowskii coil type
current probe (CWT-LF) with an accuracy of 2% in the range of 0 to 3 kA. The current
magnitude a 154 kV XLPE 2500SQ single-core cable (about 7 m) was placed as
shown in Fig. 2. A sheath loop cable was installed using 400 SQ cable. The conductor
currents supply 1000 A, sheath current supplied the calculated value. The conductors
and sheath currents were measured using a Rogowskii coil for the current magnitude
and phase difference. It were observed by an oscilloscope. EMF measurement was
installed horizontally at 1 m above the cable. EMF data was acquired at 0.9 m from
both sides of the cable center. In order to ensure the accuracy of the measured data, all
the metallic materials around the test line are removed.
EMF measurement equipment was used for EMF measurement. It was analyzed
through a magnetic field data analysis program. Measured magnetic field data is the
RMS value and 10 or more data are acquired at one point in consideration of the
measurement error of the magnetic field measuring device. It summarized as the sum of
the x-y-z data and the z-axis data and the rms value excluding the background noise. In
order to obtain the reliability of the data current and phase difference that have a large
influence on the magnetic field measurement.

2.5 EMF Simulation

Before the measurement of the magnetic field on the cable in the simulation line, the
cable modelling, magnetic field simulation were performed using the finite element
method. The model cable is a 154 kV XLPE 2500SQ single-core cable and the con-
ductor and sheath layers required for magnetic field analysis in the 2D plane. The cable
arrangement is equivalent to single phase in three phases. The magnetic field obser-
vation point is 1 m above the ground and 0.9 m on both sides.
The applied current and phase data based on the simulation results using EMTP-RV
Program. Also, eddy current effect induced on the sheath layer were considered.

2.6 Comparison of Simulation and Measurement

Figure 6 shows comparison measurement and simulation data from the 1 m position on
the cable. The magnetic field distribution curve shows the highest level with an
absolute value at the center of the cable, and it tends to decrease at both sides of the
cable. Figure 6(a), (b) show the difference in the magnitude of magnetic field.
Tables 2 and 3 show degree of errors (measurement vs simulation). Comparing the
analysis results with the measurement results, it can be seen that the values are almost
similar. And the analytical error according to results is about 1–3%.
 Study on Simulation and Measurement of EMF 53

(a) Case 1

(b) Case 2

Fig. 6. Magnetic field distribution vs Measurement distance

Table 2. Simulation vs Measurement (case 1)

Left-0.9 m Center Right-0.9 m
Simulation 0.846 1 0.846
Measurement 0.832 0.969 0.873
Error (%) 1.61 3.12 3.11

Table 3. Simulation vs Measurement (case 2)

Left-0.9 m Center Right-0.9 m
Simulation 0.854 1 0.854
Measurement 0.865 0.987 0.869
Error (%) 1.35 1.29 1.76
54 J. Eui-hwan et al.

3 Conclusions

Numerical simulations made with the finite elements software Maxwell 2D have
allowed to assess the mitigation effectiveness, thanks to simultaneous effect of con-
ductivity and high permeability. Simulation will require the ability to measure accu-
rately parameters like magnetic permeability. The reliability of the simulation
technique is secured by comparing/verifying the magnetic field values obtained from
actual measurement and simulation. It is confirm that the error rate is about 1–3%.
The harmful effects on the human health due to long-term exposure to EMF gen-
erated in the underground transmission line As the regulations are strengthened cen-
tering on developed countries, the influence of EMF around the cable is grasped and
the reduction rate and the allowable current reduction rate by various EMF abatement
methods are quantitatively It is very important to do. Therefore, in this study, a sim-
ulated test line was constructed in order to obtain the actual data for grasping the
shielding factor (SF) of the underground transmission cable EMF effect and reduction
method. We also implemented an integrated measurement program that enables mea-
surement at the same time to understand the characteristics due to the interaction of
temperature and magnetic field due to the application of the EMF reduction method of
the underground transmission cable. The simulated test line and the integrated mea-
surement program are expected to be useful in securing various measured data for the
influence of the EMF around the underground transmission cable and various reduction
techniques.

References
1. EPRI Report EL-7340-D: Single Conductor Transmission Cable Magnetic Fields PR 7898-
29, August 1991
2. EPRI Report TR-102003: Transmission Cable Magnetic Field Management PR 7898-37,
September 1993
3. Farang, A.S., Al-Sheri, A.M., Bakhashwain, J., Cheng, T.C., Zheng, G., Curci, V., Ghaemi,
M., Khajavi, M.: Magnetic field modeling and management for transmission underground
cables. In: Proceedings of the Ji-Cable’95, Paris, June, pp. 596–601 (1995)
4. Du, Y., Cheng, T.C., Farag, A.S.: Principles of power frequency magnetic field shielding with
flat sheets in a source of long conductors. IEEE Trans. Electromagn. Compat. 38(3), 450–459
(1996)
5. Habiballah, I.O., Farag, A.S., Dawoud, M.M., Firoz, A.: Underground cable magnetic field
simulation and management using new design configurations. J. Electr. Power Syst. Res. 45
(2), 141–148 (1998)
6. CIGRE TB 373: Mitigation Techniques of Power Frequency Magnetic Fields Originated from
Electric Power Systems, February 2009
7. CIGRE WG B1.23: Impact of Electromagnetic Fields on Current Rating and Cable Systems.
EMF-ELF 2011: D-O-01, March 2011
8. Brakelmann, H.: SC-IEC Cables - Single Core Cables with Integrated Electromagnetic
Shielding. JICABLE, B.1.3, pp. 394–398, June 2011
 Further Developments of Metrological
and Simulation-Based Characterization of the
Non-contact Measurement of Electrostatic
Charge by Means of Electric Field Meters

C. Schierding1(&), D. Möckel1, M. Thedens1, M. Beyer1,

and M. Kurrat2
1
Physikalisch-Technische Bundesanstalt, Bundesallee 100,
Brunswick, Germany
carola.schierding@ptb.de
2
Technische Universität Braunschweig, Schleinitzstr. 23, Brunswick, Germany

Abstract. Electrostatic charges and discharges on surfaces can result in safety-

relevant problems under certain conditions in different areas of application, from
electrical energy technology to explosion protection. In the field of electrical
energy technology, electrostatic charging is important for high-voltage direct
current transmission (HVDC). In this case, the stress of insulation systems
results in effects such as volume conductivity and surface charges. Furthermore,
electrostatic discharges in potentially explosive atmospheres can lead to ignition
of an explosive mixture and thus to explosion. Many electric and mechanical
devices have enclosures made of chargeable insulating materials and must be
evaluated for use in potentially explosive atmospheres.
The measurement of the electric field strength by means of electrical field
meters can be carried out without contact or disturbing the object under
investigation - the charge of the material - and without causing a discharge. In
order to establish the electrical field meter as a measuring instrument, precise
knowledge of the measurement method, the characterization of the influencing
parameters on the measurement and the traceability of the measured values to
national standards are required. For the metrological characterization of the non-
contact measurement of electrostatic charge, a test setup was developed which
represents the parameters to be investigated (e.g. real surface geometries, dis-
tance dependence). Furthermore, theoretical approaches for a simulation-based
characterization were considered. Preliminary results of the characterization for
idealized conditions (homogeneous electric field configurations) showed that the
measurement accuracy with regard to distance dependence and object size could
be determined. In addition, a cone detection method of the electric field meter
measuring head has been verified by simulation. As a next step, to the
simulation-based determination of the measuring angle a metrological method
for the determination is developed. Thereby, the measurement setup is refined
which is here considered together with the analytical evaluation. Furthermore, it
is necessary to quantify the simulation-based characterization for different types
of electrical field meter measuring devices. The influence of the diameter of the
electric field meter measuring head on the measuring angle and the resulting
detection cone is investigated.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 55–66, 2020.
https://doi.org/10.1007/978-3-030-31676-1_6
56 C. Schierding et al.

Keywords: Electric field meter
Field-simulation
Electrostatics

1 Introduction

In order to measure the amount of electrostatic charges on surfaces with indication of

the corresponding measuring accuracy, traceable measurement technology is required.
Established measuring instruments for the detection of electrostatic charges are the
handcoulombmeter and the shunt [1, 2]. A disadvantage of these methods is that they
influence the measurement object, when the surface charge is measured, by deter-
mining the electrostatic charge of the surface through a discharge. Furthermore, these
measurement methods cannot be used if an explosive atmosphere exists. This problem
is eliminated using an electric field meter to determine the surface charge. The electric
field meter is a non-contact and non-discharge measuring method for electrostatic
fields, which allows a derivation without direct influence on the measured object [3, 4].
In order to establish the electric field meter as a measuring instrument, it is necessary to
have a precise knowledge of the test method, the characterization of influencing
parameters and the traceability of the measured values to national standards. Therefore,
preliminary investigations to determine the influencing parameters for the electric field
meter measurement have already been carried out. On the basis of these results on
distance dependence and measured object size, using idealized arrangements (homo-
geneous electric field configurations), it has been determined that the measurement of
electrostatic charges depends on the size of the measuring surface and the distance
between electric field meter and measured object. Initial findings of the measurement
accuracy regarding distance dependencies in homogeneous E-field configurations are
achieved. Due to the increase of the distance, an increase of the measuring deviation
has been found when using the same measuring object [5]. Because of that, it was
necessary to get a better understanding of the measuring method of the electric field
meter in order to obtain high accuracy. For this purpose, a simulation model based on
the measurement setup of homogeneous electrode arrangements is developed. Thereby
a passive investigation method for the characterization of electric field meters was
developed. The simulative investigations show that the measuring head can be char-
acterized by a cone detection method. Here, the cone is formed by the measuring angle
a and the examined distance. As a result of a, and the resulting detection cone, the
measuring head detects not only the measured object but also the formed measuring
surface dcone [6]. This explains the increase of the measuring deviation with the
increase of the distance. Based on the preliminary measurement results of the distance
dependency studies [5], a measuring angle a could be determined by simulation for the
investigated electric field meter [6]. In addition to the simulation-based characterization
of a of an electric field meter, a metrological determination is necessary. For this a
metrological methodology is developed, described and verified in this paper, based on
measurement and simulation results. Furthermore, the influence of the diameter of the
electric field meter measuring head on the measuring angle and the resulting detection
cone is investigated by simulation.
 Further Developments of Metrological and Simulation-Based Characterization 57

2 Methodology of Metrological Characterization

For further metrological characterization of electric field meters, the measurement setup
[5] must be refined in order to be able to determine and verify the measuring angle,
which was previously only deduced by simulation [6]. The necessary adaption of the
measuring setup is described here, as well as the explanation for the developed method
for determining the measuring angle. A comparison between the derived measurement
angles by simulation and measurement is carried out for verification. In the following,
the principal function of the electric field meter (EMF58 eltex) used for the metro-
logical determination of the measuring angle is described.

2.1 Electric Field Meter

With an electric field meter, static charges as well as charging and discharging pro-
cesses can be displayed and measured. The measuring method of the electric field
meter is based on the principle of electrostatic induction [3, 4]. The electric field meter
consists of a measuring head and electronic measuring equipment. There are different
design of the system. The electric field meter used here works according to the most
common design [4, 7]. The design of the measuring head consists of a rotating
grounded chopper, which periodically covers the induction electrode, the measuring
surface A (t). The rotation of the chopper causes a proportional sinusoidal current on
the electric field meter, due to the electric field (E-field) [2–4]. With this measurement
method the electric field meter does not directly influence the electrostatic charge.
However, it is necessary to consider that the electric field meter changes the shape of
the E-field distribution. This is a result of the electric field meter and, in particular, the
edge of the grounded measuring head, which is not homogeneous. Because of this, the
E-field is distorted in the area of inhomogeneity (Fig. 1a) [4]. This kind of distortion
can be prevented by placing the measuring head into a grounded electrode. As a result
of the homogenization, the measuring head causes no significant influence on the
distribution of the E-field, see Fig. 1b.

a) b)

Fig. 1. (a) E-field distortion due to measuring head; (b) E-field homogenization by a grounded
electrode (according to [4])
58 C. Schierding et al.

2.2 Measurement Setup for Metrological Determination of Measuring

Angle
The developed measurement setup for the methodical investigation of the potential
influencing parameters on the measuring method of electric field meter [5] is refined in
this work. This allows, next to the determination of the measuring angle a, the
investigation of the interdependencies of the influencing parameters, distance between
the field meter and the measured object as well as the measured object size (Fig. 2a).
An idealized application of the electric field meter is used for the design of the mea-
suring setup for homogenization of the electric field distribution. In the setup, the
electric field meter measuring head is placed in a centered ground electrode (dge =
400 mm) and attached to an insulated support, as shown in (Fig. 2b). The test objects
are placed in parallel and opposite on a second insulating support. The objects are metal
profiles which are connected to an adjustable DC high voltage (HV) source (HV-DC
generator; 0 kV–40 kV). The DC voltage UHV is determined by a calibrated voltage
divider. The electric field meter (EMF58) has an external recorder output for recording
the measuring signal over a digital multimeter [5]. To determine a metrological, four
circular electrodes (HV-electrodes) with diameters of dHVe = 100 mm, 200 mm,
300 mm and 400 mm are used. With this parallel electrode arrangement for achieving a
homogeneous E-field, distance investigations (distances a = 20 mm–800 mm) are
carried out and used to derive the measuring angle by measurement and analysis.

a) b)

1: Electric field meter EMF58 6: Multimeter: recorder output EMF58

2: Ground elecrode dge = 400 mm 7: Multimeter: voltage UHV
3: HV-electrode dHVe = 400 mm 8: Grounding
4: Voltage divider 9: Measuring head dFM = 55 mm
5: HV-DC generator

Fig. 2. (a) Measurement setup for metrological characterization; (b) Electric field meter with
ground electrode [5]
 Further Developments of Metrological and Simulation-Based Characterization 59

2.3 Methodology of Determination of Measurement Angle

The methodology for the metrological determination of the measuring angle is based on
the principle of distance dependency investigations. Hereby the measured electric field
strength Emeasured is compared with the expected electric field strength Eh of the
homogeneous field for different distances a. For this, the ratio factor xm according to
Eq. (1) is calculated, with (2) [2].

xm ¼ Eh =Emeasured ð1Þ

Eh ¼ UHV =a ð2Þ

It is assumed that the detection cone of the measuring head only detects the mea-
suring object while xm * 1. Taking measurement deviations of the measurement results
into account, the threshold for the ratio factor xm is defined as xm = 1.11. When this
threshold value is reached or before it is exceeded, the assumption is that the measuring
area of the detection cone (dcone) at the thus determined distance is equal to the measured
object size (dHVe), as exemplary shown in Fig. 3. The definition of which threshold
value of the distance a is taken, depends on the ratio factor xm. When the ratio of
xm = 1.11 is reached for the last time over the distance a, with no subsequent decreases
and only a steady increase, then the threshold value for a is found. This means, when
xm > 1.11 is achieved, that the measuring head not only detects the measuring object.
After determining the distance a where xm  1.11 applies, the trigonometric calcula-
tion according to Eq. (3) determines the measuring angle a which defines the detection
cone. Here the value of bHVe is required, which is the difference between the radius of the
HV-electrode rHVe and the radius of the measuring head rFM (4).

a ¼ tan1 ðbHVe =aÞ ð3Þ

bHVe ¼ rHVe  rFM ð4Þ

This approach is carried out for four measurement objects with different diameters
dHVe (dHVe = 100 mm, 200 mm, 300 mm and 400 mm) in order to determine a defined
and reproducible measuring angle a within a homogenized measurement setup inde-
pendent of the measuring object size. When the metrological methodology is applied,
the electric field Emeasured (averaged over 10 measurements) is determined for the
distances from a = 20 mm to 200 mm in measuring intervals of Da = 10 mm, and for
a distance of 400 mm and of 800 mm (Fig. 3). Due to the increasing measuring
deviation found in preliminary investigations (dge = 400 mm, dHVe = 400 mm) [5], it
was decided to increase the distance intervals from a distance of a = 200 mm, because
it is assumed that the electric field meter detects not only the measuring object. Fur-
thermore, the maximum value of the DC voltage UHV applied to the HV-electrode is
determined by the measuring range of the investigated electric field meter based on the
minimum distance of a = 20 mm. This is done in order to be able to avoid possible
measuring errors of the measuring electronics because that would exceed the measuring
range. Also, this prevents discharges between the HV-electrode and the measuring head
of the electric field meter and thus damage to the measuring device.
60 C. Schierding et al.

Fig. 3. Principle of methodology for metrological determination of measuring angle a (showed

as example xm = 1.11 at a = 130 mm)

2.4 Metrological Results and Verification of Measurement Angle

To determine the measuring angle a of the electric field meter considered here, the
described methodology is used. For this purpose, the electric field meter is first
investigated in the measurement setup with the HV-electrode of the maximum diameter
dHVe = 400 mm for distance dependence, with the distance range a = 20 mm up
to 200 mm, 400 mm, 800 mm respectively with the appropriate intervals Da (Fig. 3).
Then, the HV-electrodes of different size are varied. A constant DC voltage of
UHV = 4 kV is applied to the HV-electrode due to the measuring range of the electric
field meter. Through the ratio factors xm (1) each threshold value and resulting dis-
tances for the HV-electrodes, where xm  1.11 applies, are determined. Finally, the
measurement angles are calculated and yield a measurement angle of a = 45° ± 3°. In
addition, the measurements are repeated with a DC voltage of UHV = 10 kV, which
required an increased minimal investigation distance of a = 40 mm. This is done to
determine whether the developed methodology is unaffected by the electric field or the
applied DC voltage. It is also used to determine whether the methodology is repro-
ducible. The results of the metrological investigations can be seen in Table 1. It shows
that for the HV electrodes dHVe = 400 mm, 300 mm and 200 mm identical measuring
angles could be determined for UHV = 4 kV and 10 kV. This was not possible for
the HV-electrode dHVe = 100 mm due to the minimum distance of a = 40 mm at
UHV = 10 kV. Nevertheless, it can be stated that the developed methodology is inde-
pendent of the measured field strength Emeasured and reproducible.
 Further Developments of Metrological and Simulation-Based Characterization 61

Table 1. Measurement results for measuring head diameter of dFM = 55 mm (marked rows:
threshold values for a, where xm  1.11 applies)

UHV
dHVe bHVe a
4 kV 10 kV α in°
in mm in mm in mm
xm xm
160 1.11 1.10 47
400 172.5 170 1.10 1.11 45
180 1.15 1.13 44
120 1.10 1.10 46
300 122.5 130 1.11 1.11 43
140 1.13 1.13 41
70 1.09 1.10 46
200 72.5 80 1.11 1.11 42
90 1.14 1.14 39
20 1.11 - 48
100 22.5 30 1.12 - 37
40 1.14 1.12 29

Now, by comparing the measuring angle determined by the metrological approach

with the previously by simulation determined measuring angle from [6] (Table 2), it
appears that these angles correspond. Yet, the measuring angle determined metrological
indicates a more accurate range of a = 45° ± 3°. Thus, an additional verification of the
methodology for the metrological determination of the measuring angle takes place and
can be used for the characterization of an electric field meter.

Table 2. Comparison of metrological and by simulation determined measuring angle a

(simulation results from [6])
Methodology Measuring angle a
Metrological 45° ± 3°
Simulation 45° ± 5°

3 Simulation-Based Characterization

In addition to the discussed methodology of the metrological characterization of the

measuring angle of an electric field meter, which was verified through the simulation
results from [6], further simulation-based characterizations are necessary to quantify
different types of electric field meters. For this purpose, the measuring angle of electric
field meters with different measuring head diameters are determined by simulation and
compared with each other. This is done to investigate the influence of the diameter of
the electric field meter measuring head on the measuring angle and the resulting
detection cone. The developed simulation model from [6] is used.
62 C. Schierding et al.

3.1 Simulation Model

The simulation model is created with COMSOL Multiphysics (Version 5.2). Thereby,
an electrostatic, 3D stationary model was implemented. The simulation model repre-
sents the measurement setup (Fig. 2a) without the measuring equipment (Fig. 2a, (4–
7)). The electric field meter is only represented by its measuring head, since the
grounded housing has no influence on the simulation model due to the homogenizing
grounded electrode. The measuring head is idealized in the simulation model by a
measuring surface (dms = 55 mm) of air (er = 1). In this way, the influence of the
measuring head material and the rotational speed of the chopper are neglected. The
measuring surface is centered ìn the ground electrode (dge = 400 mm, Rge = 4 mm)
made of aluminum. The measuring objects are plate electrodes (dHVe = 400 mm and
dHVe = 100 mm; RHVe = 4 mm) also made of aluminum, which are positioned parallel
to the ground electrode and centered around the measuring surface. The electrodes are
attached to supports (d = 100 mm, h = 0.5 m) made of insulation material (er = 3.56).
The resulting homogeneous electrode configuration is in air. The distance between the
homogenizing ground electrode with measuring surface and the plate electrode is
varied from a = 20 mm to a = 800 mm. The ground electrode is permanently set to
ground potential. A high-voltage (HV) potential is applied to the plate electrode,
whereby the height of the potential can be varied using a parameter sweep. The high-
voltage potentials depend on the electric field meter measuring range, the distance
between electric field meter and measured object and the resulting measuring distances,
which were used for previous investigations for the metrological characterization
according to [5]. The geometry of the simulation model is exemplary shown in Fig. 4,
with the HV-electrode dHVe = 400 mm.

Fig. 4. Simulation model of measurement setup for a homogeneous electrode configuration [6]

The geometry of the simulation model is extended with a detection cone. Therefore
the measuring surface is projected onto the high voltage electrode and the resulting
cylinder is extended through a parametrized measuring angle [6]. This results in a cone
with a diameter of dcone according to Eq. (5).
 Further Developments of Metrological and Simulation-Based Characterization 63

dcone ¼ 2
rFM þ 2
HZyl
tan a ð5Þ

Here, rFM represents the radius of the electric field meter measuring head and Hzyl
the height of the cylinder, which is equivalent to the distance a and together with the
measuring angle a builds the cone, which is displayed exemplary in Fig. 5. To
determine the electrical field Esim, the average over the volume of the resulting cone is
calculated.

Fig. 5. (a) Principle of detection cone (b) Simulation model for a = 400 mm and a = 45

3.2 Simulation Results

For investigation of the influence of the measuring head diameter on the measuring
angle a and the resulting detection cone, three different electric field meters are con-
sidered. The electric field meters, which are investigated here, have the following
measuring head diameters:
• no.1: dFM = 55 mm (EMF58; eltex [2, 5]),
• no.2: dFM = 20 mm (EM03; eltex [2, 5]),
• no.3: dFM = 100 mm (JCI131; Chilworth [2, 7]).
The applicability of the simulation model was validated with the measuring head
diameter of dFM = 55 mm (no.1) and by the measurement results, whereby the mea-
suring angle a = 45° ± 5° was determined [6]. In order to derive further measuring
angles, the diameter of the measuring surface dms that is equivalent to the measuring
head diameter was adjusted in the simulation model accordingly. In order to find the
correct measuring angle a of an electric field meter measuring head, a parameter sweep
of 5° to 90° in 5° steps is used, varying the angle during simulation with the experi-
mental parameters. Then the calculated electric field Esim is compared to the mea-
surement results of the electric field strength Emeasured using the ratio factor x, according
to Eq. (6).

x ¼ Esim =Emeasured ð6Þ

64 C. Schierding et al.

To decide whether a measuring angle a is applicable for representing the detection

cone of an electric field meter measuring head, the factor should be approximately
x = 1. Additionally, to verify the simulated measuring angle a, simulation results for
both HV-electrodes (dHVe = 400 mm and dHVe = 100 mm) are compared with each
other. During the simulations, different distances were used depending on the electric
field meter. Furthermore, different HV-potentials were applied to the used HV-
electrode, whereby the highest possible voltage for each distance, without causing a
discharge, was divided into five voltage potentials. For each measurement angle a, the
average and standard deviation of the ratio factor x were determined from these 5
simulation results of the E-field. Additionally, the average of the ratio factor over the
entire distance range was derived. If the ratio is approximately x * 1, the measurement
angle of the electric field meter is found. The results of no.2 are presented in Table 3 as
well as the results of no.3 in Table 4.

Table 3. Simulation results for no.2 with dFM = 20 mm (dHVe = 400 mm and dHVe = 100 mm)
Average x over distance 20 mm–200 mm
HV 4 kV–100 kV
Measuring object dHVe 400 mm 100 mm
Measuring angle Average of x 1.0 1.0
a = 50° Standard deviation s 0.43 0.46
Measuring angle Average of x 1.0 1.0
a = 55° Standard deviation s 0.42 0.42
Measuring angle Average of x 0.9 0.9
a = 60° Standard deviation s 0.41 0.39
Measuring angle Average of x 1.0 0.9
a = 65° Standard deviation s 0.41 0.51

It can be seen that the electric field meter no.2 (dFM = 20 mm) shows a measuring
angle of a = 57.5° ± 7.5°. This measuring angle is verified through different mea-
suring objects (dHVe). Thereby a wide measuring angle range with a ratio factor of
x * 1 results from the simulation, this is due to the high measuring deviations of the
electric field meter [5]. For electric field meter no.3 it was possible to determine a
measuring angle of a = 40° ± 5° using the simulation model. For this measuring angle
range, the average of the ratio factors of approximately x * 1 was found over the
entire distance range (a = 20 mm up to 800 mm).
 Further Developments of Metrological and Simulation-Based Characterization 65

Table 4. Simulation results for no.3 with dFM = 100 mm (dHVe = 400 mm and
dHVe = 100 mm)
Average over distance 20 mm–800 mm
HV 0.8 kV–100 kV
Measuring object dHVe 400 mm 100 mm
Measuring angle Average of x 1.1 1.1
a = 35° Standard deviation s 0.23 0.50
Measuring angle Average of x 1.0 1.0
a = 40° Standard deviation s 0.17 0.45
Measuring angle Average of x 0.9 0.9
a = 45° Standard deviation s 0.14 0.39

The comparison of the determined measuring angles of the electric field meters
(no.1–no.3) in Table 5 shows that the measuring angle a decreases with the increase of
the measuring head diameter from dFM = 55 mm to dFM = 100 mm. Equally, a
decrease of the measuring head to dFM = 20 mm increases the measuring angle. As a
result, the detection cone of no.2 forms a larger measuring surface diameter dcone than
the detection cone of no.1 at the same distance a. As an example, the measuring
diameters dcone = 454 mm and dcone = 647 mm result at a distance of a = 200 mm.
Thereby the electric field meter no.2 detects a larger measuring surface dcone, which
results in a higher measuring deviation depending on the object size dHVe.

Table 5. Comparison of measuring angles a of electric field meters (Results of no.1 from [6])
Measuring Measuring Measuring
head dFM angle a surface dcone (a = 200 mm)
no.1 55 mm 45° ± 5° 454 mm
no.2 20 mm 57.5° ± 7.5° 647 mm
no.3 100 mm 40° ± 5° 436 mm

4 Conclusions

This paper describes further developments of the metrological and simulation-based

characterization of electric field meters. Therefore, the developed methodology for the
metrological determination of the measuring angle a is described as well as the nec-
essary refinement of the already existing measurement setup for characterization.
Through the metrological methodology a measuring angle of a = 45° ± 3° could be
determined for the primarily investigated electric field meter, which defines the
resulting detection cone. Meanwhile, the methodology could be verified by further
measurements. Additionally, the measurement angle could be confirmed by compar-
ison to a previous simulation result with a = 45° ± 5°. Furthermore, the influence of
the diameter of the electric field meter measuring head on the measuring angle and the
resulting detection cone was investigated by simulations. Necessary adjustments to the
66 C. Schierding et al.

existing simulation model are explained as well. Due to measurement results from
preliminary investigations, the measurement angles of different electric field meters
were determined by simulation. Thus, with increasing diameter of the measuring head a
reduction of the measuring angle occurs. Consequently, with decreasing diameter the
measuring angle increases.

References
1. Langer, T., Beyer, M., Pidoll, U.: Messtechnische Charakterisierung elektrostatischer
Entladungen, Technisches Messen, pp. 516–524 (2008)
2. Küchler, A.: Hochspannungstechnik - Grundlagen - Technologie – Anwendungen, 3. neu
bearbeitete und erweiterte Auflage. Springer, Heidelberg (2017)
3. Lüttgens, G.: Expert Praxislexikon Statische Elektrizität. Expert Verlag (2000)
4. Lüttgens, G.: Statische Elektrizität begreifen - beherrschen – anwenden, 4. völlig neu
bearbeitete Auflage. Expert Verlag (2002)
5. Schierding, C., Möckel, D., Ladda, C., Beyer, M.: Messtechnische Charakterisierung der
berührungslosen Messung elektrostatischer Aufladungen mittels Feldmühlen, 13. Fachtagung
Anlagen-, Arbeits- und Umweltsicherheit, Köthen, Germany, November 2017
6. Schierding, C., Möckel, D., Runge, T., Meyer, T., Ladda, C., Beyer, M.: Metrological and
simulation-based characterization of the non-contact measurement of electrostatic charge by
means of electric field meters. In: 12th International Symposium on Hazards, Prevention and
Mitigation of Industrial Explosions, Kansas City, USA, August 2018
7. Chubb, J.N.: Two new designs of “Field Mill” type fieldmeters not requiring earthing of
rotating chopper. IEEE Trans. Ind. Appl. 26(6), 1178–1181 (1990)
 Calculation of Circulating Current
Inside Power Transformers
in Non-symmetrical Configurations

Bruno Jurišić1(&), Tomislav Župan1, and Leonardo Štrac2

1
Končar – Electrical Engineering Institute, Fallerovo šetalište 22,
10002 Zagreb, Croatia
bjurisic@koncar-institut.hr
2
Končar – Power Transformers Ltd., Josipa Mokrovića 12,
10090 Zagreb, Croatia

Abstract. In power transformers with split type two secondary windings, non-
symmetrical configurations are possible. The design of these transformers
considers two low voltage windings mounted one on top of the other and the
regulating and high voltage windings divided in two parallels. Due to the
specific design of these transformers, circulating currents can exist in parallels of
the high voltage and regulating windings. These currents can affect the load
losses inside the transformer as more parts of the winding are energized than in
the symmetrical configuration. Moreover, they may affect the design of the
winding itself since the amplitudes of these circulating currents can be signifi-
cant and therefore have to be taken into account during the design phase.
It is possible to calculate these currents using the finite element method
(FEM) software in order to check if their values exceed the design values and to
more accurately calculate the load losses of the transformers. In this paper, the
results of the 3D FEM calculation of such power transformers are shown and
compared with the measurement results in terms of circulating currents ampli-
tudes in high voltage and regulation windings parallel parts. Additionally, a
comparison between calculated and measured load losses is given.

Keywords: Power transformers
Circulating currents
Split type
Load losses

1 Introduction

Specific configuration of power transformers that considers two windings of same

phase and voltage level mounted axially with respect to each other is possible. For
instance, split-winding transformers have two low voltage windings that are usually
used when two generator units or distribution networks have to be connected and the
space and instrumentation in the substation is limited. Moreover, such design decreases
the values of the short circuit currents in the individual circuits [1]. High voltage and
regulation windings are in that case divided in two parallels allowing the circulating
currents to flow through the parallels. It is important to control these currents as their
values may be significant in the case of non-symmetrical configurations when only one
low voltage winding is loaded.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 67–74, 2020.
https://doi.org/10.1007/978-3-030-31676-1_7
68 B. Jurišić et al.

In Sect. 2 of this paper a finite element method (FEM) model for calculation of
circulating currents of a real split-winding power transformer that allows non-
symmetrical loading is shown. A method for measuring the circulating currents, used to
validate the model, is presented in Sect. 3. Then a comparison of measured with
calculated results as well as a discussion is given in Sect. 4. Finally, conclusions are
given in Sect. 5.

2 FEM Model of a Power Transformer for Calculating

the Circulating Currents

FEM modelling has been carried out in the commercial electromagnetic FEM analysis
software Infolytica© MagNet, using the time harmonics 3D solver [2]. As the circu-
lating currents are not significantly affected by the non-linearity of the transformer core
in the case of the load loss test, linear time harmonics solver was sufficient to address
this problem. Using FEM for the calculation allowed a straightforward way for cal-
culating the effect of the reaction field on the main leakage field, allowed taking into
account both radial and axial leakage flux components as well as the real 3D geometry
of the transformer [3]. 3D FEM solver was used since one of the aims of this research
was to see whether a difference exists in the circulating currents induced in the
windings of different phases. Using 2D axisymmetric FEM solver is also possible but
would only give results for a single phase. Similar approaches can be found in the
literature both for calculating the circulating current between parallel conductors inside
the windings [3–5] and between the windings [1, 6, 7].
A 240 MVA, 220/33 kV transformer unit used for the calculation consisted of two
low voltage windings mounted axially with respect to each other and high voltage and
regulation windings connected in parallel (Fig. 1). The transformer is a three-phase
network power transformer. Geometry of the transformer window is shown in Fig. 1.
The transformer under observation requires a single tap changer to operate.
Therefore, it has a parallel connection between each tap section of the regulation
winding allowing the current to flow even in the case when the tap sections are not
loaded. Regulation windings of the observed transformer unit are helical in one layer
with twelve tap sections and seven turns per tap section.
In the scope of the paper, main tap position has been observed. In this position only
seven turns (a single section) of each regulation winding are loaded. Other sections are
non-loaded, but the current is expected to flow through them due to the closed current
path of the parallels.
As the validation measurements for the circulating current were done without a
tank, only active part of the transformer has been modelled in FEM, as can be seen
from Fig. 2.
Model materials that are used in the FEM model are represented as linear as the
circulating currents are not significantly (as it is the case for the losses or magnetic
inductions) affected by the non-linearity in the load loss test.
 Calculation of Circulating Current Inside Power Transformers 69

LV2_1 LV2_2

LV2 HV2
RW2

HV

RW1
LV1 HV1

LV1_1 LV1_2 N

Fig. 1. Geometry of the winding of the power transformer with two split-type LV windings.

Fig. 2. A model of the observed transformer unit in 3D FEM analysis software Infolytica©
MagNet.
70 B. Jurišić et al.

A boundary of the calculation domain is set as flux tangential boundary at the

distance ten times larger than the largest dimension of the model. A mesh is set in such
way that its effect on the calculated induced currents and total magnetic energy results
can be neglected. Using the inbuilt mesh adaption tool, several simulations with dif-
ferent mesh sizes have been calculated in order to reach satisfying level of the mesh.
Every turn of the regulation winding is presented separately, and their connections
are connected in the same way as it is the case in the real transformer during the load
loss measurements. The current source is connected to the LV 2 winding while the
LV 1 winding is left open-circuited, in order to simulate non-symmetrical case. HV and
N terminals of the high voltage windings are short-circuited to ensure the correct
current path as during the measurements.
The results of the simulation are shown in Sect. 4.

3 Measurements of the Circulating Currents

in the Transformer Windings Connected in Parallel

The measurement configuration considered LV 2 winding short-circuited, LV 1

winding open-circuited and voltage source applied to the HV winding. Voltage source
generated a current of 20 A flowing through the HV winding. The currents were
measured using current clamps in two parallels of the HV winding and in each tap
section of the regulation winding. In total, 14 different current measuring positions
were used for the 240 MVA transformer unit. The measurements were done on one
phase and the transformer without the tank.
The results of the measurements are shown in Sect. 4.

4 Result Comparison for Circulating Current and Load

Losses

For the tap position under the observation, only Sect. 1 of the regulation winding is
loaded, while all the other tap sections of the regulation windings are non-loaded. Due
to the parallel connections of the winding, current loops exist between tap sections of
the regulation winding, allowing current to flow even though the sections that are not
loaded.
The measured and calculated currents are shown in Table 1, in the percentage of
the nominal currents.
Table 1 shows that a difference between the currents of different phases exists since
the transformer geometry is not perfectly symmetrical (side phases differ from the
middle phase). Difference also exists due to the numerical error since each section of
the regulation winding has only seven turns and a small fraction of voltage is induced
in the current loop with a limited resistance. Therefore, a small error in the calculation
of the induced voltage can significantly affect the values of the circulating currents.
 Calculation of Circulating Current Inside Power Transformers 71

Table 1. Comparison of measured and calculated circulating currents in the case of non-
symmetrical configuration of the split-winding transformer.
Phase FEM current per FEM average Measured Difference
phase (%) current (%) currents (%) (PP)
LV 1 A 0,00 0,00 0,00 0,00
B 0,00
C 0,00
LV 2 A 100,00 100,00 100,00 0,00
B 100,00
C 100,00
HV 1 A 1,79 1,86 1,20 −0,66
B 2,03
C 1,74
HV 2 A 101,70 101,78 101,50 −0,28
B 101,96
C 101,67
RW 1 - A 49,97 40,74 22,80 −17,94
section 1 B 49,97
C 22,27
RW 1 - A 37,23 34,84 31,80 −3,04
section 2 B 37,28
C 30,01
RW 1 - A 32,19 29,85 25,50 −4,35
section 3 B 32,11
C 25,25
RW 1 - A 37,80 36,30 35,30 −1,00
section 4 B 38,09
C 33,00
RW 1 - A 25,01 23,62 20,60 −3,02
section 5 B 24,50
C 21,35
RW 1 - A 47,40 42,30 39,10 −3,20
section 6 B 41,44
C 38,06
RW 1 - A 16,95 15,73 19,10 3,37
section 7 B 15,27
C 14,97
RW 1 - A 57,46 50,90 44,50 −6,40
section 8 B 48,97
C 46,27
RW 1 - A 16,19 13,36 18,50 5,14
section 9 B 13,44
C 10,46
(continued)
72 B. Jurišić et al.

Table 1. (continued)
Phase FEM current per FEM average Measured Difference
phase (%) current (%) currents (%) (PP)
RW 1 - A 68,26 62,13 51,60 −10,53
section 10 B 60,41
C 57,73
RW 1 - A 37,11 34,48 30,30 −4,18
section 11 B 36,69
C 29,63
RW 1 - A 37,11 60,48 59,00 −1,48
section 12 B 73,65
C 70,68

As the current measurements were done on only one phase of the transformer,
although all three phases were energized, the average calculated current is taken as a
reference for the comparison with the measured values. When compared with the
measurements, it can be seen that the calculated circulating current values differ in the
observed case less than 18% points for the regulation winding’s tap sections and less
than 1% point for the high voltage winding’s parallels. The difference between mea-
sured and calculated results is significantly lower in the case of the HV winding as the
number of turns is much higher for that winding than for the section of the regulation
winding.
Current density distribution in the transformer windings, for the non-symmetrical
configuration calculated using FEM is shown in Fig. 3.
Figure 3 shows that the current is limited in the lower parallel of the high voltage
winding. However, the current exists in the lower parallel of the regulation windings
due to current loops that are formed.
In addition, the comparison between measured and calculated load loss has been
shown. It is important to note that the circulating current in the transformer windings
can increase the load losses. The comparison is done for two transformer units of a
same design. Both calculated and measured values are shown in Table 2, for 75 °C.
For the results shown in Table 2, standard calculations from the transformer factory
were used. These calculations include ohmic and eddy current losses in the windings
and lead connections as well as the stray losses in the structural parts of the transformer.
Circulating currents affect all the mentioned segments of the load losses by magnifying
them. Therefore, taking these currents into account lowers the gap between the mea-
sured and the calculated values and puts the calculation results on the safe side.
 Calculation of Circulating Current Inside Power Transformers 73

Fig. 3. Current density distribution (in A/m2) in the split-winding transformer for the non-
symmetrical configuration.
74 B. Jurišić et al.

Table 2. Comparison of measured and calculated load losses in the case of non-symmetrical
configuration of the split-winding transformer.
Measured Calculating without circulating Calculated with circulating
[kW] currents [kW] currents [kW]
Unit 1 295,3 285,6 301,2
Unit 2 293,9

5 Conclusions

In the case of non-symmetrical configuration, significant circulating currents are

induced in the parallels of the split-type transformers during the load loss measure-
ments. These currents can affect the load losses and have to be taken into account
during the design stage of the power transformer.
In the paper, a 3D FEM calculation of the circulating currents in such transformer
has been shown. The calculated current values are compared with the measured ones.
The difference between measured and calculated results is small enough to consider the
calculations as applicable. Furthermore, measured and calculated losses during the load
loss test have been compared. By taking the circulating currents into account, the
accuracy of the load loss calculation for the non-symmetrical case is similar to the one
of the symmetrical cases. Therefore, it is important to consider the effect of the cir-
culating current when dealing with the power transformers that have a split-type
winding.

References
1. Kumbhar, G.B., Kulkarni, S.V.: Analysis of short-circuit performance of split-winding
transformer using coupled field-circuit approach. IEEE Trans. Power Deliv. 22(2), 936–943
(2007)
2. Mentor Graphics: MagNet (2017)
3. Koppikar, D.A., Kulkarni, S.V., Ghosh, G., Ainapure, S.M., Bhavsar, J.S.: Circulating-
current loss in transfromer windings. IEE Proc.-Sci. Measur. Technol. 145(4), 2–6 (1998)
4. Isaka, S., Tokumasu, T., Kondo, K.: Finite element analysis of eddy currents in transformer
parallel conductors. IEEE Trans. Power Appar. Syst. PAS-104(10), 2730–2737 (1985)
5. Pern, J., Yeh, S.: Calculating the current distribution in power transformer windings using
finite element analysis with circuit constraints. IEE Proc.-Sci. Measur. Technol. 142(3), 231–
236 (1995)
6. Fukumoto, H., Furukawa, T., Itoh, H., Ohchi, M.: Calculating leakage reactance of 9-winding
transformer using time-dependent 3D FEM analysis. In: IECON 2015-41st Annual
Conference of the IEEE Industrial Electronics Society, pp. 4459–4464 (2015)
7. Stuchl, P., Doležal, I., Zajic, A., Hruza, J., Weinberg, O.: Performance of transformers with
split winding under nonstandard operation conditions. CIGRE Bien Session Paris, France
(2000)
 Influence of Different Turret-Tank Connection
Types on Transformer Turret Eddy Currents

Tomislav Župan1(&), Bruno Jurišić1, and Franjo Kelemen2

1
Končar – Electrical Engineering Institute Inc., Fallerovo šetalište 22,
10002 Zagreb, Croatia
tzupan@koncar-institut.hr
2
Končar – Power Transformers Ltd., Josipa Mokrovića 12,
10090 Zagreb, Croatia

Abstract. All metal parts of the transformer tank are usually electrically con-
nected and grounded in order to prevent the possibly hazardous floating
potential from emerging. The magnetic field of transformer leads and bushings’
conductors induces eddy currents on nearby metal parts such as tank top and
bushing turrets. Since the above-mentioned electrical connections can form the
unwanted conducting path loops, the type of these connections can help in
restricting the eddy currents and consequently the losses in metal parts and the
heat generated in them. This is especially true for high-current power trans-
formers such as generator step-up units. This research presents calculated and
measured results of the influence of different bushing turret-tank connection
types on the induced eddy current values. The simulation results were obtained
using the finite element method based software. Comparison against the mea-
surements shows that these eddy currents can be adequately modelled and
predicted using calculations which can prove useful in the design phase of the
transformer manufacturing process. By carefully choosing the connection type
and the location of the shorting element, the eddy current conducting path loops
can be restricted and therefore the values of these currents can be limited as well.
This can, in turn, lead to lower local losses and reduced heating of the metal
parts, all of which can be easily shown using Rogowski coils for current
measurements and thermal imaging systems for thermography measurements,
respectively.

Keywords: Bushing
Eddy currents
Losses
Power transformer
Turret

1 Introduction

Power transformers’ tanks, since predominantly made from steel, are a source of
additional losses arising from induced eddy currents [1]. Due to a number of techno-
economical reasons, the tanks are made as small in volume as possible, regarding the
dielectric, electromagnetic and mechanical constraints. Since they are exposed to rel-
atively high magnetic fields arising from windings and leads placed within the tank, the
additional eddy current losses within them can become significant. These currents,
through ohmic loss in tank material, can cause significant local increase in temperature

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 75–84, 2020.
https://doi.org/10.1007/978-3-030-31676-1_8
76 T. Župan et al.

arising from the generated heat within [2–4]. Along with unwanted excess losses, this
can cause faster deterioration of paint, gaskets or bolts in the vicinity of these hotspots.
Therefore, a number of different techniques for reducing these losses are in use.
They mainly focus on shunting or shielding the stray magnetic flux within the tank, and
so the induced eddy currents and consequently losses are reduced. A number of papers
exist on these topics and therefore these approaches will not be a scope of the work
presented here [3–7].
One of the main areas where the problem of high magnetic field cannot be easily
mitigated is on the tank top, where connecting leads pass through bushing turrets
outside of the tank [6, 8]. On low voltage side, where high current leads are placed, this
problem is more pronounced. This is especially true for high-current power trans-
formers such as generator step-up units [5, 6]. In those areas, the typical solutions for
reducing losses range from stainless steel inserts to displaced tank top and three-phase
turrets outside of the tank.
All parts of the connected metallic equipment and fittings, that are considered a part
of the transformer, need to have an electrical connection with the tank and need to be
grounded. This applies to additional equipment, control/signal/regulation cabinets, oil
cooling ducts and radiators, bushing turrets, etc. All of these elements need to be
grounded in order to prevent a possibly hazardous floating potential from emerging.
A non-defined electric potential of a power transformer tank and adjacent elements can
lead to problems such as paint corrosion, sparking, partial discharges and can even be
harmful to nearby personnel. A special care is taken to prevent the forming of con-
ducting loops through interconnected equipment. These loops, if formed, can lead to
unwanted circulating currents [8, 9].
In some specific geometric structures and designs, however, conducting loops
cannot be avoided easily or even at all. This is mainly the case in different turret-tank
designs.
This paper presents the influence of different turret-tank connection locations and
types on the induced eddy currents within the turret. The following section briefly
presents the problem and the scope of the work with different options given. Section 3
gives the results obtained from numerical simulations and shows the verification
through measurements. Finally, conclusion of the presented work is given.

2 Turret-Tank Model

One of the most complicated tank designs when dealing with high-current leads’
magnetic field reduction is three-phase turret outside of the tank. If a three-phase
transformer has low voltage leads with high line current, it is very difficult to design the
tank with low eddy current losses. These three-phase leads, going out of the tank
through bushing turrets, generate high magnetic field in the turret area, and hence high
eddy current losses.
To mitigate this problem, without enlarging the whole tank which would in turn
negatively affect the overall transport volume and mass, delta connection is made
outside of the tank in a three-phase turret. In that case, both winding ends from each
phase pass through the tank top. Since the ends of the same winding carry the same
 Influence of Different Turret-Tank Connection Types on Transformer 77

current in different directions, the overall magnetic field is greatly reduced due to the
cancelling effect. Hence, the tank top should not have problems with large eddy current
losses and can be made without stainless steel inserts [8, 10, 11].
However, this approach imposes additional outer three-phase turret in which a delta
connection is finally made. To keep the steel and oil mass, together with the eddy
current losses to minimum, these turrets usually have to be made from stainless steel.
The computer model used for the simulation of eddy current losses in three-phase
turret is visible in Fig. 1. It consists of tank top, lower single-phase turrets, three-phase
turret or delta connection box and upper single-phase turrets through which line current
leads finally connect the low voltage side of the power transformer to the electrical
network. Jumpers on the upper turret are also visible in the figure, but jumpers on the
lower turret cannot be seen from this angle. For the sake of simulations presented here,
the low voltage side is shorted, and nominal current is imposed through each of the
phases.

Fig. 1. Computer model of tank top and turrets with outer delta connection box.

Both lower and upper mechanical contacts between single- and three-phase turrets
can electrically be either connected or not. Both approaches have their own difficulties
when manufacturing. Electrical connection can be easily obtained if both parts need to
be on the same electrical potential, simply through one of the flange’s bolts. However,
if a large continuous eddy current flows through a weak electrical connection, tem-
perature hotspots can occur. This is usually seen in local overheating of connecting
bolts that carry unsuitably high current during load loss test [4]. For that reason, even
seemingly simple electrical connection between turret flanges needs to be designed
properly if subjected to high magnetic fields.
78 T. Župan et al.

On the other hand, to ensure electrically insulated mechanical contact between

flanges, double gaskets with insulating barrier have to be usually used along with
electrically insulated washers with hollow cylinders for connecting bolts. Both
approaches can be used, depending on the overall design and transformer manufac-
turer’s decision.

3 Results

Since all the metal parts above the tank top are usually electrically connected using
jumpers, the main goal of this research is to investigate the influence of the position and
type of different turret-tank connections. For the sake of clarity, focus is given only on
upper single-phase turret jumpers. Both electrically conductive and insulated
mechanical contacts between single- and three-phase turrets have been used. All
simulations were done using Infolytica© MagNet 7.9 software.

3.1 Numerical Simulations

Three single-phase upper turrets were named X1–X3, looking from the high voltage
side, as visible in Fig. 1. Seven different jumper positions/types were used (positions
from the high voltage side):
1. Single copper jumper on the right side.
2. Single copper jumper on the left side.
3. Single steel jumper on the right side.
4. Single copper jumper: X1-right; X2-middle; X3-left side.
5. Dual jumpers on each turret – combining variants 1 and 2.
6. Same as 5, but with electrically insulated flanges.
7. Four jumpers on each turret.
For every case except case 6, the connecting flange between three- and upper
single-phase turret was electrically connected. The results of the effective current
flowing through each of the jumpers are given in Table 1, and the visualizations of
current density for each of the cases can be seen in Figs. 2, 3, 4, 5, 6, 7 and 8. In each
of the figures, the colour scale is the same in order to be able to visually compare the
results.
Since copper jumpers have electrically negligible resistance, it can be seen from the
results that they form a conductive pathway for induced eddy currents if placed in the
specific position. For instance, it can be seen in Fig. 2 that the current flowing through
jumper X1 is relatively low compared to jumpers X2 and X3. However, if the position
of the jumper changes by rotating it 180° around the turret, the current through the
same jumper greatly increases, as visible in Fig. 3.
The resistance of jumpers influences the current flowing through them, as visible in
Fig. 4, where copper jumpers are replaced with steel ones.
Figure 5 shows that, by carefully choosing the right position for the jumpers, the
jumper current can be greatly affected. This will, in turn, help in constraining the
overall eddy currents and hence the additional losses.
 Influence of Different Turret-Tank Connection Types on Transformer 79

Table 1. Simulated effective jumper current for each of the presented cases.
Case no. IRMS [A]
X3 X2 X1
1 (right) 1508.8 1632.9 239.3
2 (left) 391.4 1818.3 1861.2
3 (steel, right) 396.4 429.6 60.6
4 (combination) 387.5 221.2 236.5
5 (dual, connected) 390.1 1844.4 1901.1
1516.7 1628.8 263.6
6 (dual, insulated) 802.8 1879.6 1441.0
802.8 1879.6 1441.0
7 (four jumpers) 422.6 1542.9 1630.8
424.3 1539.9 1621.3
1639.3 1524.3 394.1
1663.2 1522.0 392.8

Fig. 2. Simulated current density distribution for case 1 (single copper jumper on the right side).

Cases 5 and 6, results of which are given in Figs. 6 and 7, show how the electrical
connection between flanges influences the jumper current. If there is an electrical
connection between the two turrets at the flanges, the current can flow through all three
pathways in each of the turrets and will distribute itself with regards to the induced
eddy currents and connection resistances (Fig. 6). On the other hand, if the flanges are
electrically insulated, then obviously the current through each of the two jumper pairs
on each turret will be the same, since eddy currents form a pathway through them.
Moreover, in the case where the flanges are electrically insulated and only a single
80 T. Župan et al.

Fig. 3. Simulated current density distribution for case 2 (single copper jumper on the left side).

Fig. 4. Simulated current density distribution for case 3 (single steel jumper on the right side).

jumper exists per turret, no current shall flow through the jumper as no current path will
be formed.
Figure 8 shows the results for case 7, where multiple current pathways exist. As
can be seen, the outer jumpers carry considerably lower amount of current. The results
 Influence of Different Turret-Tank Connection Types on Transformer 81

Fig. 5. Simulated current density distribution for case 4 (single copper jumper: X1-right; X2-
middle; X3-left side).

Fig. 6. Simulated current density distribution for case 5 (dual jumpers on each turret).

in Table 1 are presented in a left-right manner for cases 5 and 6, and clockwise starting
from lower left for case 7, when looking at the turrets as in given figures, from the high
voltage side.

3.2 Measurements
The model of the outside delta connection box, together with upper and lower turrets,
presented in previous section, was manufactured with the possibility of attaching
82 T. Župan et al.

Fig. 7. Simulated current density distribution for case 6 (dual jumpers on each turret with
electrically insulated flanges).

Fig. 8. Simulated current density distribution for case 7 (four jumpers on each turret).

jumpers on upper singe-phase turrets similar to the ones presented in case 7. Four
Rogowski coils were used, together with Dewetron DEWE-571-PQA, for measuring
the jumper currents waveshapes.
Figure 9 gives the measured values of jumper currents for upper single-phase turret
X2, together with their overall sum. The jumper positions are defined in the same way
as in the simulation case 7 explained in the previous section. Results show that eddy
currents mainly flow between left and right pairs of jumpers (J1/J2 and J3/J4 in phase,
J1/J3 and J2/J4 out of phase), which is in agreement with simulated results in case 7, as
 Influence of Different Turret-Tank Connection Types on Transformer 83

2
J1 J2 J3 J4 Sum

1.5

0.5
Current [kA]

0
0 5 10 15 20 25

-0.5

-1

-1.5

-2
Ɵme [ms]

Fig. 9. Measured current through four jumpers (J1–J4) on turret X2.

well as with similar cases 5 and 6. The calculated sum current is equivalent to the
current flowing through the electrically conductive flange connection between three-
and upper single-phase turret.
Measurement results are comparable with simulations and confirm the applicability
of finite element method based numerical models for eddy currents prediction and
optimal jumper positioning.

4 Conclusion

A brief insight, through several numerical cases, on the influence of conductive

pathways on overall induced eddy currents is given. Measurements confirm the plau-
sibility of used models.
All the results show that the positioning and material of the jumpers influences the
amount of current flowing through them. By carefully choosing the jumper position, its
cross section can be adequately designed and possible electrical current constraint on
flange bolts and hotspots can be mitigated. This can lead to greater durability of
surrounding parts and paint layers.
Results also confirm that, using detailed numerical models, eddy currents induced
in power transformer tank and surrounding metallic parts such as turrets can be ade-
quately calculated. This can lead to further understanding of local temperature hotspots
and overall additional losses in transformer tank and hence design optimization can be
made.
84 T. Župan et al.

References
1. Milagre, A.M., Da Luz, M.V.F., Cangane, G.M., Komar, A., Avelino, P.A.: 3D calculation
and modeling of eddy current losses in a large power transformer. In: Proceedings - 2012
20th International Conference on Electrical Machines, ICEM 2012, vol. i, pp. 2282–2286
(2012)
2. Zhang, Y., Yan, B., Cao, F., Xie, D., Zeng, L.: Analysis of eddy current loss and local
overheating in oil tank of a large transformer using 3-D FEM. In: 2011 International
Conference on Electrical Machines and Systems, ICEMS 2011, pp. 3–6 (2011)
3. Magdaleno-Adame, S., Olivares-Galvan, J.C., Escarela, R., Raichenko, O., Kladas, A.G.:
Hot spots mitigation on tank wall of a power transformer using electromagnetic shields. In:
Proceedings - 2014 International Conference on Electrical Machines, ICEM 2014, pp. 2235–
2238 (2014)
4. Olivares-Galvan, J.C., Magdaleno-Adame, S., Escarela-Perez, R., Ocon-Valdez, R.,
Georgilakis, P.S., Loizos, G.: Experimental validation of a new methodology to reduce
hot spots on the screws of power transformer tanks. In: Proceedings - 2012 20th International
Conference on Electrical Machines, ICEM 2012, pp. 2318–2322 (2012)
5. Yang, J.: New preventive measures against stray field of heavy current carrying conductors.
IEEE Trans. Magn. 9(Suppl. A), 271–1492 (1996)
6. Renyuan, T., Li, V., Lin, F., Lijian, T.: Resultant magnetic fields due to both windings and
heavy current leads in large power transformers. IEEE Trans. Magn. 32(3 PART 2), 1641–
1644 (1996)
7. Takahashi, N., Kitamura, T., Horii, M., Takehara, J.: Optimal design of tank shield model of
transformer. IEEE Trans. Magn. 36(4 I), 1089–1093 (2000)
8. Li, Y., Ho, S.L., Wang, N., Tang, R.Y.: Numerical analysis of eddy current field in the
ascending flange for the bushings and tank wall of a large transformer. In: 2008 Joint
International Conference on Power System Technology. POWERCON IEEE Power India
Conference, POWERCON 2008, vol. 44, no. 6, pp. 1522–1525 (2008)
9. Yan, X., Yu, X., Shen, M., Xie, D., Bai, B.: Research on calculating eddy-current losses in
power transformer tank walls using finite-element method combined with analytical method.
IEEE Trans. Magn. 52(3), 1–4 (2016)
10. Maximov, S., Olivares-Galvan, J.C., Magdaleno-Adame, S., Escarela-Perez, R., Campero-
Littlewood, E.: New analytical formulas for electromagnetic field and eddy current losses in
bushing regions of transformers. IEEE Trans. Magn. 51(4), 1–10 (2015)
11. Maximov, S., Escarela-Perez, R., Magdaleno-Adame, S., Olivares-Galvan, J.C., Campero-
Littlewood, E.: Calculation of nonlinear electromagnetic fields in the steel wall vicinity of
transformer bushings. IEEE Trans. Magn. 51(6), 1–10 (2015)
 Introducing an Inductive Loop Sensor as
an Alternative to Record the Phenomena
of a Dense Plasma Focus of 400 J

Luis Orellana1 , Jorge Ardila1(B) , Gonzalo Avaria2 , Benjamı́n Cevallos1 ,

Cristian Pavez2 , Roger Schurch1 , and Leopoldo Soto2
1
Universidad Técnica Federico Santa Marı́a, 8940000 Santiago, Chile
luis.orellanag@alumnos.usm.cl, jorge.ardila@usm.cl
2
Chilean Nuclear Energy Commission, Casilla 188-D, Santiago, Chile

Abstract. Pulsed plasma discharges are of interest for their capability

to produce pulsed high energy radiation and particle beams. One of them
is the dense plasma focus, a device that has renewed interest for physics
research and possible industrial applications. The plasma produced with
this device corresponds to the z-pinch phenomena and lasts between 1–
10 ns. Electrical quantities such as the voltage across the electrodes of the
vacuum chamber where the plasma dynamic takes place and the circuit
current are used for the diagnostic of these devices. The short duration
of the high density plasma that is produced implies a challenge when
choosing a proper sensor. Typically, Rogowski coils wrapped around a
conductor are used for measuring the time derivative of the circuit cur-
rent. The reproducibility of Rogowski coils and variation of its frequency
response due to movements can be an issue. An inductive loop sensor was
developed as an alternative to measure fast and low amplitude transients
of partial discharges. In this work, the use of the inductive loop sensor
is proposed as an alternative sensor for the dense plasma focus opera-
tion measurement. Simultaneous measurements of both the Rogowski coil
and inductive loop sensor were carried out on a low energy dense plasma
focus. A comparison between both sensors was made in time and time-
frequency domains. Wavelet transform was used in the time-frequency
analysis. Further studies regarding the pinch detection were made with
correlations between the signal values at this particular instant of the
device operation. The results indicated that similar results in the char-
acterization of the dense plasma focus operation can be obtained with
the inductive loop sensor. In terms of the frequency response at time
of pinch both sensors yielded similar results. A linear tendency between
the signal values at time of pinch was found. Future work will be carried
out to fully adapt the inductive loop sensor for its application in pulsed
power technology.

Keywords: Inductive loop sensor · Plasma focus · Fast current

measurement

c Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 85–95, 2020.
https://doi.org/10.1007/978-3-030-31676-1_9
86 L. Orellana et al.

1 Introduction
The dense plasma focus (DPF) is a pulsed power device that produces high
density plasma [1,2] using the z-pinch phenomena [3], which lasts in the order
of 1–10 ns, until it is disrupted by instabilities. This device has been used for
scientific investigations and also for possible industrial applications such as [2,4]:
a pulsed source of X-ray or neutrons, X ray imaging, to test fusion reactor wall
materials, nanosatellite propulsion, high energy radiation therapy, etc. The DPF
is a device composed of a vacuum chamber and an electrical circuit, the first is
where the plasma dynamics takes place, while the second drives the device. Inside
the vacuum chamber a coaxial electrode system is placed. The center electrode
corresponds to the anode and the outer one the cathode which is usually made
out with a certain number of bars. The anode is partially covered with an insula-
tor where the gas breakdown initially occurs and a plasma sheath is created. Gas
at a pressure in the order of 1–10 mbar fills the chamber. Typically hydrogen or
deuterium is used, although heavier gases and gas mixtures had also been used.
The device is electrically driven by a capacitor bank. The charging voltage of
the bank depends on its size, being possible to find miniature sized models to
greater sized devices. This range of voltages, from a few kV to even thousands
of kV, implies a circuit energy in the order of 0.1 J to MJ. The particular DPF
used in this work is a low energy device of around 297 J. In general, six stages
of the discharge can be identified [2,4]. (I) initial gas breakdown: this process
allows the formation of a plasma sheath on the surface of the insulator, (II)
axial rundown: due to Lorentz force produced by the circuit current, the plasma
sheath moves upwards toward the end of the anode, (III) radial rundown or com-
pression: on top of the anode the Lorentz force compresses the plasma sheath
into a column, (IV) pinch: the moment of maximum compression of the plasma
column, (V) axial plasma shock: after pinch disruption due to instabilities and
(VI) plasma jet: after the axial plasma shock advances away from the anode,
plasma jets are ejected. At the time of pinch, electron and positive ion beams
(also neutrons if deuterium is used) are produced along with electromagnetic
radiation in a broad spectrum, from radiofrequencies to X rays [1–4]. The most
simplified circuit model for DPF consists in an equivalent capacitor bank in
series with the equivalent stray inductance and circuit losses, connected to time
variant resistance and inductance that models the plasma. This model produces
an underdamped sinusoidal behaviour for its electrical quantities. The simplest
diagnostic of DPF is made with electrical signals of voltage measured across the
electrode of the vacuum chamber and the circuit current [1,2,4]. The Rogowski
coil is widely used for measuring the circuit current. This sensor has been par-
ticularly studied for pulsed power applications [5]. For the DPF operation, the
short duration of the pinch imposes bandwidth requirements on the inductive
sensors, so the development of this type of sensors remains an open field of study.
Other inductive architectures have been developed to be an alternative to the
Rogowski coil. In recent years, the inductive loop sensor (ILS), has been used
to measure the fast and low amplitude electrical transients associated to partial
discharges [6]. The aim of this work is to evaluate the ILS performance in the
 ILS Sensor for Plasma Focus 87

diagnostic of the DPF operation in comparison to the Rogowski coil. Simulta-

neous measurements of a low energy DPF were carried out. The signals were
analyzed in time and time-frequency domains. The latter was made with the
use of the wavelet transform. Particular details of the ILS behaviour when mea-
suring the evidence of the pinch event were further studied. The comparison of
the wavelet transforms from signals of both sensors for this particular time was
made. Finally, the correlation between the ILS and Rogowski using the signal
parameter that is used to quantify the pinch evidence was carried out in order
to verify a linear tendency from both signals.

2 Sensors and Experimental Setup

2.1 Inductive Sensors

As was mentioned above, the inductive measurement of the DPF operation is

one basic diagnostic which can be used to infer the plasma dynamic and its
properties. In this study, two inductive sensors are compared:

Fig. 1. Inductive sensors utilized: (a) Rogowski coil and (b) Inductive loop sensor.

– Rogowski coil : This coil is usually used in the measurement of fast electric
transients from pulsed power devices [5]. The Rogowski available in the lab-
oratory was wrapped around one of the return leads from the vacuum cham-
ber, see Fig. 1a. Little details of its construction or electrical parameters were
available. However, this sensor is adequate for detecting the dip (evidence
of the pinch) and serves for the circuit current calculation via integration of
its signal. This sensor is the reference to which the proposed inductive loop
sensor is compared to.
– Inductive loop sensor (ILS): This sensor was initially designed for measuring
the fast and low amplitude transients related to partial discharge phenomenon
[6]. This sensor is shown in Fig. 1b. The ILS consists in a copper strip line in
rectangular shape printed in a circuit board. Its main geometrical dimensions
are summarized in Table 1. No direct measurement of the frequency response
of the ILS was made in this study, although the response can be simulated.
88 L. Orellana et al.

Table 1. Geometrical dimensions of the ILS according to Fig. 1b

Geometrical dimension Length (mm)

Loop side l1 120.4
Loop side lw 10.4
Strip thickness 0.035
Circuit board width 0.35

As a reference, previous measurement of its frequency response indicated that

the ILS has a derivative behaviour until 31.83 MHz which can be considered to
be the cut off frequency for the di/dt measurement [7]. At higher frequencies,
a proportional behaviour and a resonance are also found [8]. Due to not being
shielded from EM noise, radiated components of the measurement or other
noise can be coupled into its signal.

2.2 Experimental Setup

Simultaneous measurements using the ILS and Rogowski were taken on a low
energy DPF PF-400 J [9]. The electrical parameters of this device are: 880 nF
capacitor bank, 38 nH circuit inductance, 26 kV charging voltage, 297 J of elec-
trostatic energy and approximately 330 ns to achieve the maximum current. The
operating conditions for this experiment were: a charging voltage of 26 kV for
the capacitor bank, and the vacuum chamber was filled with pure hydrogen at
9 mbar. The Rogowski was wrapped around a returning lead from the vacuum
chamber to one of the capacitors. Meanwhile, the ILS was placed at 2 mm from
another lead. The acquisition system used for the measurements was a NI-PXI
5162 oscilloscope embedded in a NI-PXIe 1082 chassis with a NI-PXIe-8115 con-
troller (dual core i5-2510E processor). The measurement characteristics of the
oscilloscope are: 2 channels, sampling frequency of 6.25 GS/s, 3 GHz of band-
width and 8 bit of vertical resolution. Attenuators of 20 dB and 30 dB with a
bandwidth of 0–18 GHz and 0–6 GHz, respectively, were used to keep the signals
in the safe range of 1 Vpp (peak to peak) for the equipment. The Rogowski coil
was connected to one channel, which was used for triggering the data acquisi-
tion; meanwhile the ILS was connected to the other. Both signals were recorded
with a time window of 1 µs. A high voltage trigger unit was used to control
the beginning of the DPF operation via a pressurized spark gap filled with pure
nitrogen at 0.5 bar. Thirty discharges from the DPF were made with a 14 s
interval between each one.

3 Results and Signals Comparison

The behaviour of the sensors are first compared in the time domain. The physical
phenomena that can be inferred from the Rogowski signal is used to qualitatively
 ILS Sensor for Plasma Focus 89

characterize the discharge in terms of the proposed ILS signal. Both the di/dt and
its numerical integration, i.e. the circuit current, are evaluated in this manner.
To further understand the sensor capability to detect particular events from the
discharge, the wavelet transform is used to analyze the signals. Time-frequency
domain approximation obtained via wavelet transform was preferred over the
fast Fourier transform (FFT), as different changes associated to the discharge
produce time variant frequency content signals, also known as non-stationary
signals. Particular details for the pinch inference are of most interest, as is in this
stage that the high density plasma is produced and particle beams and radiation
is emitted. Thus, any sensor used for DPF diagnostic requires to adequately
measure this phenomena. A final correlation between the sensors was carried out
using the dip value feature that represents the pinch detection in the inductive
signals.

3.1 Waveform Analysis in Time Domain and with Wavelet

Transform
Figure 2 shows an example of both Rogowski and ILS signals from a same dis-
charge. Some stages of the discharge mentioned in the introduction can be
inferred from the commonly used Rogowski signal. The identification of these
stages serves to characterize the discharge in terms of the ILS signal which is
fairly similar to the Rogowski one. The numerical integration of the inductive
signals yields the circuit current, see Fig. 3a for Rogowski signal and Fig. 3b
for the ILS signal. The diagnostic of the discharge based on the Rogowski coil
(Fig. 2a) is as follows: first, the initial spark gap conduction indicates that the
anode is now connected to the high voltage of the capacitor bank. This partic-
ular event is not clearly identified by the Rogowski, but since the ILS which

Fig. 2. Example of the Rogowski and Fig. 3. Current waveform obtained

ILS signals from a discharge: (a) from both sensors: (a) Rogowski and
Rogowski coil and (b) ILS. (b) ILS.
90 L. Orellana et al.

is more susceptible to EM radiation, a definitive change in the signal is seen.

The beginning of this event corresponds to the first blue vertical line in Fig. 2.
Then, the stages of the discharge follow. The initial gas breakdown, at the end
of stage I, can be interpreted to happen when the Rogowski signals starts to
increase due to the plasma sheath created allowing for the current to flow from
anode to cathode. Stage II of axial rundown can be inferred to happen next as
current keeps on increasing. The compression of the plasma sheath, stage III,
is identified when the Rogowski signal starts to decrease more sharply than in
stage II. See the dotted line in Fig. 2. The maximum compression of the plasma
is achieved at the pinch, stage IV highlighted with a red line in Fig. 2, which can
be seen as dip in the Rogowski signal. Also, due to pinch there is a slight current
change as can be seen in Fig. 3, this is often called the current interruption that
follows the pinch [2]. After stage IV, the pinch disruption occurs due to instabil-
ities. It is not possible to directly observe this disruption through the inductive
measurement. Later, it can be seen that the signal returns to an under-damped
sinusoidal behaviour which is expected in these devices. Note that high frequen-
cies components are excited again in both sensors. They are notorious in the
ILS signal, see Fig. 2b. The first time that these high frequencies components
appeared were during the initial breakdown, stage I, due to the gas breakdown
which is basically a switching action as reported by Bruzzone et al. [10]. Even
in the shielded Rogowski, high frequency oscillations are observed at the pinch
phase, because its disruption produces a switching action in a similar way as the
gas breakdown [10]. The interaction of stray capacitances and equivalent induc-
tance of the device had been identified as an explanation of these frequencies
[10,11]. The ILS further identifies those oscillations and even radiated compo-
nents that conforms the EM burst emitted from these devices. Stages V and VI
of the axial plasma shock and the plasma jet ejection cannot be observed with
the inductive measurement. Qualitatively, from Fig. 2 it can be said that the ILS
signal is similar to the Rogowski coil. Furthermore, the stages detected by the
Rogowski can also be detected by the ILS. An analysis of the frequency compo-
nents that appears as the discharge process evolves is shown next, based on the
wavelet transform of the sensors signals. This technique is adequate to identify
transients in noisy data [12], and previously was introduced by Piriaei et al. [13],
for analyzing the signal from a Mirnov coil applied to the DPF diagnostic. In
this work, the wavelet transform was computed using the Morlet wavelet and
was implemented with Matlab. R The wavelet transform produces a coefficient
matrix that indicates the time localization of a scaled wavelet, which in turn
can be interpreted as the time localization of an equivalent sinusoidal having a
particular frequency. Thus, a higher value of the coefficient indicates a strong
presence of the certain frequency. Each transform result obtained was normalized
by the maximum coefficient value of each transform. This allowed the interpre-
tation of the coefficients with respect to the maximum value. In the wavelet
transform results the reddish zones of the plot indicate a clear time-frequency
identification whereas the blue ones indicate no presence of the frequency at a
certain time during the signal.
 ILS Sensor for Plasma Focus 91

The results from the wavelet transform of the signals from Fig. 2 are shown
in Fig. 4a for the Rogowski signal and Fig. 4b for the ILS. Time of the pinch
is highlighted in the plot. Three zones of frequencies are of interest in both
signals, these are summarized in Table 2. The first frequency f0 corresponds to
the approximation of the main discharge frequency which can be estimated by
the capacitor bank and stray inductance values, see Eq. 1.
1
f0 ≈ √ = 870 kHz (1)
2π C0 L0

Fig. 4. Wavelet transform of the signals from Figure: (a) Rogowski and (b) ILS.

Table 2. Summary of the frequencies of interest from the wavelet results in Fig. 4.

Highlighted Rogowski (MHz) ILS (MHz)

frequencies
f0 ≤1 ≤1
f1 ≤26 ≤24
f2 [50 − 500]/135 [50 − 500]/120

Where C0 is the equivalent capacitor bank of 880 nF and L0 the equivalent

stray inductance of 38 nH. The estimation of 870 kHz is consistent with the
wavelet transform result, ≤1 MHz. Note that f0 is identified during the discharge,
92 L. Orellana et al.

at the lowest frequency value. Next, f1 is highlighted as the limit frequency which
is identified a Δ shape with light blue colour in Fig. 4 which indicates part of
the signal with abrupt changes. The events that are associated to these changes
are the gas breakdown and, most importantly, the pinch event. In the Rogowski
sensor this frequency is up to 27 MHz whereas in the ILS is up to 24 MHz. Finally,
a higher frequency band f2 appears because of pinch and lasts some ns later.
In the case of the ILS this transients lasts longer. In both wavelet results this
frequency band extends from 50 MHz to 500 MHz, where in the Rogowski the
frequencies around 120 MHz are more clearly identified than the others. In the
ILS, frequencies around 135 MHz marked with red colour, are more visible than
the others. In this representation it is evident that the ILS is more susceptible
to higher frequency components that can be part of the inductive measurement
or due to radiated components.

3.2 Dip Value Features of the ILS

As was mention before, the pinch evidence seen as a dip in the di/dt signal is
one of the most interesting features of the inductive measurement. It is univer-
sally agreed that a narrow dip with high value intensity is associated with the
emission of particle beams and high energy radiation [1,2]. This dip value is
even used to calculate the induced voltage at time of pinch along with a voltage
measurement. So, this inductive diagnostic serves not only as an indicator of the
plasma compression at pinch, but also to make other plasma dynamic estima-
tions. The proposed ILS needs to be able to detect this feature in a similar way
as the Rogowski coil. Basically, the dip can be thought as a kind of singularity,
narrow inverse peak, in the di/dt signal which otherwise is expected to follow
an underdamped sinusoidal behaviour. An adequate bandwidth of the sensor is
needed to properly measure this narrow peak to make reasonably estimations of
the voltage induced at time of pinch for example. A comparison of the frequency
content at time of pinch between the ILS and the Rogowski was carried out
using the wavelet transform results from Fig. 4. A plot of the wavelet coefficients
at time of pinch is shown in Fig. 5. The Rogowski result indicates that the dip
feature consists in particular frequencies, from the main discharge frequency to
higher ones centered around 100 MHz. Note that the ILS results follows a sim-
ilar behaviour. It is reasonable to say that both signals yield similar frequency
content at time of pinch. Variability of pinch is a common topic in DPF opera-
tion. Different pinch results detected by the dip values of the diagnostic signals
can be obtained due to the complex plasma dynamic inside the chamber [2]. To
evaluate the performance of the ILS in detecting this variability, a scatter plot
between the dip values of both sensors was analyzed. Figure 6 shows this result.
A linear tendency was found. Despite not being a perfect linear fit of the data,
the result indicates that pinch variability can also be inferred with the proposed
ILS signal.
 ILS Sensor for Plasma Focus 93

Fig. 5. Wavelet coefficients at time of Fig. 6. Dip value correlation between

pinch. the Rogowski and ILS.

4 Discussion and Conclusions

With the aim of searching better bandwidth sensors for the pulsed power devices
diagnostic, such as the plasma focus, the inductive loop sensor (ILS) was pro-
posed in this work. A comparison between a commonly used Rogowski coil and
the ILS was carried out. In the time domain, the discharge stages that can be
identified with the Rogowski could also be identified with the ILS, so the dis-
charge can also be characterized with this proposed sensor. The straightforward
integration of the ILS signal yields the circuit current as it is done with the
Rogowski signal without significant differences. Not direct measurement of the
bandwidth of both sensors was made, only a reference bandwidth measurement
[8] of the ILS was available. This measurement indicates that the ILS has a
derivative behaviour up to approximately 32 MHz.
Despite not being able to measure the frequency response of the sensors,
in this work the comparison between them was carried out using the discharge
phenomena of the DPF itself. The analysis of the frequency content was sufficed
to compare both signals from the discharge. The non-stationary signals from the
signals were analyzed with the wavelet transform. The stages of the discharge
that represent abrupt changes in circuit current were identified using this trans-
form. These stages corresponded to the initial gas breakdown and the pinch. In
both sensor the wavelets shown a Δ shaped zone indicated the occurrence of a
change in the signal. Moreover, by looking at the time domain signals and their
wavelet results it was evident higher frequencies were clearly identified in the
ILS signal. The ILS by not being shielded against electromagnetic radiation rep-
resents both an advantage and disadvantage. Higher frequencies from the DPF
operation are expected as indicated by Bruzzone et al. [10]. These components
94 L. Orellana et al.

were also found in the shielded Rogowski, so they are part of the measurement.
The ILS thus has the capability to measure directly these components. On the
other hand, electromagnetic noise from other parts of the DPF can coupled into
the ILS measurement distorting its signal. Strong electromagnetic (EM) burst is
also emitted from the DPF as was reported by Gerdin et al. [14] and Escalona
et al. [15]. The relationship between the EM burst and the inductive measure-
ment remains as an open field of investigation in these devices. The behaviour
of the signals at time of pinch was also studied using the wavelet transform
result. The dip value which is considered to be evidence of the pinch can be
interpreted as a narrow peak in the inductive signals. As was obtained in the
wavelet results for this particular time, both sensors yielded similar results. A
frequency band around 100 MHz was found to be present at time of pinch in
both sensors. This result gives hints about the adequate measurement of the
dip feature using inductive sensors. For the ILS, a shorter version could give
better results as the equivalent inductance would yield a higher bandwidth [6].
The influence of the ILS size along with the influence of the distance from the
DPF for the dip feature measurement will be considered in future work. The
variability of dip values obtained with the Rogowski coil was also obtained with
the ILS. The scatter plot using the dip value measured from each sensor showed
a linear tendency. In conclusion, the ILS performance was shown to be similar
to the Rogowski coil used for the diagnostic of DPF. The characterization of the
DPF operation can be carried out using the ILS and also similar frequency con-
tent between the reference Rogowski coil was obtained. Future work guidelines
regarding the measurement of higher frequency components using the ILS will
be considered to propose the ILS as a fully alternative to the Rogowski coil.

Acknowledgments. This work was supported by “Fondecyt Iniciación” 11160115

and “Dirección de Postgrados y Programas” of Federico Santa Maria Technical Uni-
versity under PIIC grant.

References
1. Mather, J.: In: Methods in Experimental Physics, pp. 187–249, vol. 9. Elsevier
(1971)
2. Bernard, A., Bruzzone, H., Choi, P., Chuaqui, H., Gribkov, V., Herrera, J., Hirano,
K., Krejci, A., Lee, S., Luo, C., et al.: J. Moscow Phys. Soc. 8, 93 (1998)
3. Haines, M.: Plasma Phys. Control. Fusion 53(9), 093001 (2011)
4. Soto, L.: Plasma Phys. Control. Fusion 47(5A), A361 (2005)
5. Pellinen, D.G., Di Capua, M.S., Sampayan, S.E., Gerbracht, H., Wang, M.: Rev.
Sci. Instrum. 51(11), 1535 (1980)
6. Robles, G., Martinez-Tarifa, J.M., Rojas-Moreno, M.V., Sanz-Feito, J.: IEEE
Trans. Instrum. Measur. 58(11), 3907 (2009)
7. Ardila-Rey, J., Barrueto, A., Zerene, A., Castro, B., Ulson, J., Mas’ud, A., Valdivia,
P.: Sensors 18(7), 2324 (2018)
8. Ardila-Rey, J., Rojas-Moreno, M., Martı́nez-Tarifa, J., Robles, G.: Sensors 14(2),
3408 (2014)
 ILS Sensor for Plasma Focus 95

9. Silva, P., Moreno, J., Soto, L., Birstein, L., Mayer, R.E., Kies, W.: Appl. Phys.
Lett. 83(16), 3269 (2003)
10. Bruzzone, H., Kelly, H., Moreno, C.: IEEE Trans. Plasma Sci. 18(4), 689 (1990)
11. Bruzzone, H., Acuña, H., Barbaglia, M., Milanese, M., Clausse, A.: J. Fusion
Energy 36(2–3), 87 (2017)
12. Lilly, J.M.: Proc. R. Soc. A: Math. Phys. Eng. Sci. 473(2200), 20160776 (2017)
13. Piriaei, D., Mahabadi, T., Javadi, S., Ghoranneviss, M.: Phys. Plasmas 24(8),
083508 (2017)
14. Gerdin, G., Tanis, M., Venneri, F.: Plasma Phys. Control. Fusion 28(3), 527 (1986)
15. Escalona, I., Avaria, G., Dı́az, M., Ardila-Rey, J., Moreno, J., Pavez, C., Soto, L.:
Energies 10(9), 1415 (2017)
 A New Approach in Calculation of Step
Voltages for Complex Grounding Systems
by Analytical Considerations Only

Martin Hannig1(B) , Ralph Brocke1 , and Volker Hinrichsen2

1
DEHN SE + Co KG, 92318 Neumarkt, Germany
martin.hannig@dehn.de
2
Technische Universität Darmstadt, 64283 Darmstadt, Germany

Abstract. Lightning strikes can have devastating effects to human

beings. Not only direct strikes, connecting leaders, induced or touch volt-
ages are dangerous, but also step voltages are an issue. Although touch
voltages are more critical in terms of mortality rate, in case of step volt-
ages the involved area on ground can be quite large. This is the reason
why in half of all cases, where lightning injures human beings, step volt-
ages are responsible. To overcome this threat, the grounding system plays
a major roll, when the lightning current should be diverted into ground.
An inadequate grounding system may not only lead to a high ground
potential rise, it may also lead to high step voltages in direct vicinity
to the structure to be protected. However, when a grounding system
is planned, step voltages are not considered in detail, because calcu-
lation of step voltages is quite time consuming. Some regulations are
given for placing down conductors, because the largest ground potential
rise exists at the point where the lightning current enters ground. Stan-
dards also give information on grounding rod arrangements or grounding
rings as well as some simple equations, from which an assembled low
frequency grounding resistance can be derived. Nowadays step voltages
are calculated using complex simulation tools. Hereby different effects
can be taken into account, like soil ionization, travelling waves, disper-
sion or even complex soil layers. However, most of the time these effects
are neglected, because when considering a worst case, soil ionization for
instance leads to smaller grounding resistances and in consequence to a
lower step- or touch voltage. In this work a new approach is presented, in
which an analytical consideration is used to describe a complex grounding
system that consists of electric conductors of certain radii. This method
has been implemented in a MATLAB algorithm and is able to calculate
step voltages on ground level very fast, in a couple of milliseconds. This
enables the user to plan a grounding system more efficiently, because
the user has a direct response on the step voltage, when a grounding ele-
ment is adjusted. Because the solution is based on analytical expressions,
the method can also give reference information to stationary simulation
tools. In this work an overview on the methodology is given, as well as
a verification and some examples showing the full potential of this new
methodology.
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 96–107, 2020.
https://doi.org/10.1007/978-3-030-31676-1_10
 New Approach in Calculation of Step Voltages 97

Keywords: Lightning · Grounding systems · Step voltages ·

Analytical consideration · Line current source

1 Introduction
Today different standards give information on safety limits regarding lightning
protection [1,2]. When it comes to grounding, the standards give information
on how a proper lightning grounding system has to look like in order to meet
acceptable grounding resistances and equipotentials in order to achieve accept-
able touch and step voltages. The standards also give some advice on how to
place down conductors to reduce the danger of touch and step voltages. How-
ever, there is no simple procedure to survey step voltages and the assembled
grounding resistance of a given arrangement. For this reason, often complex
simulation tools are applied [5]. This work describes the idea of calculating step
voltages on ground surface, based on complex grounding systems. In contrast to
other simplified methods, which are most commonly used in the standards [3,4].
The approach presented here is based on superimposed analytical expressions
of line current sources. The ground is represented as a homogenous isotropic
material with a given uniform soil resistivity. For this reason, the methodology
is only applicable to stationary problems and cannot take nonlinear or frequency
dependent effects into account.

2 Theory and Methodology

In order to calculate the step voltage on a plane ground surface the potential
distribution on this surface has to be known. The potential distribution can be
achieved, when the current density distribution in the ground is known, and
this can be calculated using analytical expressions of different current sources.
Basically, the whole procedure originates from the potential distribution of a
single line current source. Hereby the current leaves the line perpendicularly.
This source can be construed as a grounding rod of an individual diameter
and length. An analytical expression can be found that describes the potential
distribution. This expression can be easily formulated in cylindrical coordinates
and has to be transformed into Cartesian coordinates in order to rotate and
shift the expression, so that an individual position and orientation can be taken
into account. A mirrored copy of the source is also needed to take into account
the perfectly insulating ground plane. For each grounding element an analytical
expression is added. All expressions are superimposed and lead to the whole
potential distribution in uniform soil. One really demanding aspect is, that not
the analytical expression of the line is rotated and shifted, instead it is the
environment that is individually rotated and shifted, in order to stick to the
analytical solution in cylindrical coordinates. When the potential of the rods
itself is kept constant the unknown currents, flowing off each current source,
can be found, and hereby the whole current density distribution in ground can
98 M. Hannig et al.

be calculated [10]. This methodology is very fast, precise and relatively easy to
be implemented in a tool. Besides, the methodology can also easily derive the
potential and the electrical field strength distribution, not only on the surface,
but also in the whole environment.

2.1 Limitations in Case of Lightning

Stationary Assumption. The proposed methodology calculates potential dis-
tributions under the assumption of stationary currents. However, a lightning
return stroke current is some kind of impulse. In the standards first return
strokes (FRS) and subsequent return strokes (SRS) from negative and posi-
tive downward lightning are discriminated. The upper threshold for a lightning
peak current is 200 kA and is derived from a positive downward lightning, which
is exceeded in only 1% of all cases, according to the CIGRÉ peak current dis-
tribution. Such a value can lead to a high ground potential rise (GPR) and
therefore may lead to high step voltages. Fortunately, the positive FRS has a
very flat slope, and the equivalent frequency amounts to 25 kHz. What should
be kept in mind is that the size of the grounding system matters. Most of the
time grounding systems are planned for peak currents based on the lightning
protection levels. In Table 1 empirical rules are given for estimating the prop-
agation velocity in ground at a given soil resistivity as well as the concluding
wave length. The wave length of the first return stroke current in ground with a
soil resistivity of 1000 Ωm amounts to 632 m. A problem to be solved should be
smaller than about one tenth of this length. The lower the soil resistivity is, the
smaller can be the grounding systems to be depicted. However, step voltages are
less problematic in regions with low soil resistivity, because of a lower ground
potential rise.

Table 1. Dependency of penetration velocity in ground and its wave length according
to a specific soil resistivity and equivalent frequency based on first and subsequent
return strokes [9]

Parameter feq v in m µs−1 λ in m

√ √
FRS 25 kHz 0, 5 · ρe 20 · ρe
√ √
SRS 1 MHz 3.2 · ρe 3.2 · ρe

Soil Ionization. A known effect of grounding systems is that when the ground
potential rise (GPR) is high, the electrical field strengths in direct vicinity of the
grounding electrodes exceed the local electrical break-through field strength in
ground and in consequence bridge small regions by ionization. This effect leads
to an increase of the virtual grounding diameter and, therefore, lowers the overall
grounding impedance. In case of soil ionization, it is commonly accepted that
 New Approach in Calculation of Step Voltages 99

the ground potential rise as well as the step voltages are less critical compared
to cases without soil ionization. Therefore, a conservative approach is to neglect
this effect [5].

Different Soil Layers. The methodology presented here is not able to take
into account different soil layers. Although this might be possible an infinite
number or at least a large number of refraction planes would be needed, which
leads to a large amount of memory and computing power and therefore foils the
advantages of this methodology. Furthermore, real soil resistivity can locally vary
very much in dependency of day time, weather season and also the measured
depth. Therefore, it is advisable to use relatively large values in order to be as
conservative as possible.

2.2 Step Voltages

Step voltages on ground are basically potential differences between two feet
standing on ground in a distance of 1 m [11]. Usually the step voltage is evaluated
on a line starting at a point near a structure of interest towards distant ground.

Fig. 1. Exemplary differences between step voltage (black) and electrical field strength
(blue) originating from four circular grounding rings (simulated with COMSOL
MultiphysicsR
) with a Potential of 1 V

The intention is to find the highest step voltage, because if a person stands
on an equipotential level no step voltage will be effective. Therefore, the step
voltage is evaluated in a “‘worst case”’ approach. In order to plot the step voltage
on ground, always the maximum value at each point has to be found, and this
value has its maximum when someone stands perpendicularly to an equipotential
100 M. Hannig et al.

line on ground. It is obvious that this can easily be achieved by evaluating the
electrical field strength on ground.

E = −∇φ

where: E = Electrical field strength (V m−1 )

φ = Electrical potential (V)

Besides, the electrical field strength is very similar to the step voltage, when
normalized to the step length of 1 m. Because of 1, the gradient on a straight
line can be numerically calculated by 1 and is transformed to the step voltage
when Δx is set to 1 m.

f (x + Δx) − f (x)
f  (x) =
Δx
The step voltage differs from the electrical field strength in two ways. First,
the electrical field strength is the infinitesimal gradient on every point on ground
and leads to some kind of exaggeration, because these values can locally be quite
large. Second, the step voltage on a point is always the difference to a point 1 m
ahead, which leads to a shift of the step voltage towards the electrical field
strength by 0.5 m. Evaluating the electrical field strength as step voltage will
give more conservative results as can be seen in Fig. 1.

Calculation of Step Voltages. It should be noted that in all cases depicted

in this paper only unloaded step voltages are issued. The main interest of this
work is to calculate the step voltages on ground level. For this purpose, the
analytical expressions for the electric field strength and the potential on ground
level are needed. In order to calculate the correct step voltages, the field strength
of a point of interest is evaluated with regard to its orientation. The orientation
is used to calculate a second point in a distance of 1 m. On both points the
electrical potential is evaluated by the analytical solution in order to calculate
the difference between both potentials. This consideration is exemplarily shown
in a very coarse grid in Fig. 2. The arrangement consists of two vertical rods,
and the calculated potential distribution is shown. The blue arrows show the
orientation of the electrical field strength on ground, and the length of each
arrow is normalized to 1 m. The black dots are the coordinates of the potentials
that have to be evaluated in order to determine the correct step voltage towards
the red dots.
 New Approach in Calculation of Step Voltages 101

Fig. 2. Example of a potential distribution on ground level along with meshed evalua-
tion points for the step voltage. Red and black dots are interconnected by blue arrows,
indicating the maximum electrical field strength.

Because there is an analytic solution, the potential of the black dots can be
calculated exactly.

Fig. 3. Unloaded step voltage calculated by the new approach.

Finally, a color plot of the step voltages can be derived as shown in Fig. 3.
The step voltage colors are defined in a logarithmic scale. It should be noted
102 M. Hannig et al.

that, when dealing with step voltages, due to the finite resistance of the human
body the voltage difference between the two feet of the body standing on ground
will be different from the unloaded step voltages calculated here [12].

Loaded Step Voltages. When a human being stands on ground, the person’s
impedance is in parallel to the resistance of the ground between both feet. There-
fore, the resulting impedance is lower than in absence of the person. The person
and the ground form a current divider. Because the overall current remains the
same in both cases, the voltage drop along the stepping distance is reduced. The
issue has been discussed in recent publications, and in order to take this effect
into account, the step voltage can be reduced by a linear scaling factor based on
the actual soil resistivity [5,12]. Based on recent studies this scaling factor can
be calculated by 1.

Uu
= 0.0061555 · ρ + 1.0415
Ul

where: Uu = unloaded step voltage (V)

Ul = loaded step voltage (V)
ρ = specific resistance (Ωm)

Such scaling factor can easily be implemented. Using this procedure, a loaded
step voltage can be found, which should not be exceeded. This acceptable loaded
step voltage is still under discussion. Often a value of 25 kV is used [5].

3 Validation of the Methodology

The proposed new methodology is based on an analytical solution, originating

from thin lines, from which current is drained to ground. In order to validate
not only a single line but a complex arrangement, a building with a foundation
grounding as well as a ring earth electrode was evaluated. The building measures
54.1 m in length, 14.9 m in width, and the lowest elevation of the grounding
system is −1 m. All conductors have a diameter of 2 cm and are assumed to be
perfect electric conductors (PEC). The arrangement can be seen in Fig. 4. This
complex arrangement has been implemented in two commercially available field
simulation tools. At first, COMSOL Multiphysics R
is used, which comprises
a 3D finite element method (FEM). The simulation is somewhat simplified in
order to get a solution: The conductors are represented as straight lines, set to a
static potential. This avoids meshing the conductors by very small tetrahedrons.
Nevertheless, the whole surrounding area, along with infinite boundary elements,
has to be meshed, which is still quite ambitious. The computing time was about
394 s.
The second commercial simulation tool is CST Studio Suite. R Here, the elec-
trostatic solver uses a capacitance matrix and charges to calculate the solution.
 New Approach in Calculation of Step Voltages 103

Fig. 4. Normal component of the electrical field strength on ground level originating
from a complex grounding system made up of a foundation and a ring grounding system
(black), along with its mirror conductor (grey); computed with the new approach in
MATLAB.

The computing time was 578 s. These computing times are hard to compare,
because the degree of complexity is obviously quite different. Anyway, the ana-
lytical solution in MATLAB needs less than a second! In order to better compare
the accuracy of all three solutions, a line is drawn on ground evaluating the elec-
trostatic potential from this complex grounding system. The evolution of the
potential is shown in Fig. 5.

Fig. 5. Electrostatic potential on ground level calculated with COMSOL

Multiphysics
R
(blue, dashed line), CST Studio Suite R
(red, dotted line) and with
the new approach in MATLAB (orange, straight line) in 3D from a complex grounding
system, assuming a total current of 100 kA (LPL III) in ground with a soil resistivity
of 1000 Ωm.
104 M. Hannig et al.

It can be observed that the new methodology is very accurate in compar-

ison to sophisticated simulation tools. Deviations of the approach in compar-
ison to CST Studio Suite R
is within ±0.8% and in comparison to COMSOL

Multiphysics is within ±3.4% [7,8].
R

4 Opportunities of the New Approach

The new methodology can be applied to evaluate many different grounding rod
arrangements in a short period of time, with special regard to step voltages. It
can be implemented in the MATLAB graphical user interface (GUI) and can
be used to design a well-adapted grounding system, even in case of complex
structures.

4.1 Evolution of Step Voltages for Four Grounding Electrodes

The following example demonstrates the benefit of the new approach. Four
grounding electrodes are considered. Normally grounding electrodes are driven
vertically into the ground.

Fig. 6. Distribution of loaded step voltages for four grounding rods, inclined by 45◦ ,
with their critical areas where a loaded step voltage of 25 kV is exceeded.

It is assumed that they can be driven with a certain inclination angle in order
to optimize the loaded step voltage on ground level. In this example a total peak
current of 100 kA and a soil resistivity of 1000 Ωm has been assumed. To get
the loaded step voltages, all unloaded step voltages are divided by a factor of
7.2 [6,12]. The four grounding poles have a length of 9 m each and have fixed
positions of 5 m away from a virtual point of origin, where they enter ground. The
situation for an inclination angle of 45◦ can be seen in Fig. 6. Hereby, an overall
resulting grounding resistance of 34.1 Ω is achieved. Additionally, an area of
 New Approach in Calculation of Step Voltages 105

54 m2 , in which the critical loaded step voltage is exceeded, can be computed. In

order to find the arrangement with the smallest critical area, the whole procedure
is repeated for 1000 different inclination angles. The results can be found in
Fig. 7. The best setup in terms of step voltages can be reached (49.3 m2 ), when
an inclination angle of 31◦ is assumed, while the lowest assembled grounding
resistance, which amounts for 32.8 Ω, can be achieved, when an inclination angle
of 65◦ is assumed.

Fig. 7. Development of the critical area, in which a critical loaded step voltage of
25 kV is exceeded, in dependence of the inclination angle, together with the calculated
resulting overall grounding resistance.

4.2 Graphical User Interface in MATLAB

MATLAB offers the opportunity of creating an individual graphical user inter-
face in order to control a certain functionality. Because the new approach can cal-
culate even complex grounding systems very fast, the user is able to set grounding
rods in real time and see the effect on the step voltage in direct response. Hereby,
problems, e.g. originating from the metallic sheath of a water supply underneath
the entrance of a building, can be identified and mitigated by counter poises.
Figure 8 shows a screenshot of the GUI in MATLAB.
106 M. Hannig et al.

Fig. 8. Example of an arbitrary GUI in MATLAB, using the new approach in calcu-
lating the step voltages of complex grounding systems in real time, with several plots,
import and export functions.

5 Conclusion
In this work a new approach has been presented for calculating step voltages
from a given grounding system by purely analytical approaches. When only worst
cases shall be issued and only a survey of step voltage is needed, this method is
able to calculate even complex arrangements in very short periods of time (in the
order of only a few milliseconds, and by a factor of several hundred faster than
commercial simulation software, respectively), which makes this method capable
to be used in real time in order to design a grounding system not only by the
use of standardized estimations, but also to take a 3D grounding geometry into
account. There are only some few restrictions in case of complex soil, nonlinear
effects or dispersion issues related to large grounding systems. However, this is
a minor issue, because the diversity of parameters is quite large as well as their
variation. Within the known constraints the whole method has proven to be very
accurate, and it can even be used to validate complex simulation tools.

References
1. IEC 62305-3:2010: Protection against lightning – Part 3: Physical damage to struc-
tures and life hazard (2010)
2. IEC 61936-1:2010+AMD1:2014 Consolidated Version: Power installations exceed-
ing 1 kV a.c. – Part 1: Common rules (2014)
3. Sunde, E.D.: Earth Conduction Effects in Transmission Systems. Dover Publica-
tions Inc., New York (1949)
4. Liew, A.C., Darveniza, M.: Dynamic model of impulse characteristics of concen-
trated earths. IET Proc. Inst. Electr. Eng. 121, 123–135 (1974)
 New Approach in Calculation of Step Voltages 107

5. Suchanek, S., Hinrichsen, V., Brocke, R., Müller, K.P.: Investigations of earth
termination systems with respect to optimised step voltages. In: IEEE International
Conference on Lightning Protection (2012)
6. Suchanek, S., Hinrichsen, V., Gao, J., Munteanu, I., Brocke, R., Müller, K.P.:
Simulation of earth termination systems under consideration of step voltages. In:
XIV International Conference on Atmospheric Electricity (2011)
7. Comsol Multiphysics GmbH: Comsol Multiphysics Reference Manual (2018)
8. CST Computer Simulation Technology GmbH: CST STUDIO SUITE Help (2018)
9. Heidler, F., Stimper, K. Blitz und Blitzschutz. VDE Verlag (2009)
10. Hannig, M.: Calculation of the assembled grounding resistance from complex
grounding systems by using analytical considerations only. In: International Con-
ference on High Voltage Engineering (2018)
11. IEC 60050-195:1998/AMD1:2001: International Electrotechnical Vocabulary (IEV)
- Part 195: Earthing and protection against electric shock (1998)
12. Suchanek, S.: Untersuchungen an Blitzschutzerdungsanlagen unter besonderer
Berücksichtigung der Schrittspannung, Dissertation (2014)
 Numerical Simulation on DC Breakdown
of Polyimide Based on Charge Transport
and Molecular Chain Displacement

Yuwei Li1, Chenyu Yan1, Daomin Min1(&), Shengtao Li1,

Zhaoliang Xing2, Liangxian Zhang3, and Chong Zhang3
1
State Key Laboratory of Electrical Insulation for Power Equipment,
Xi’an Jiaotong University, 710049 Xi’an, China
forrestmin@mail.xjtu.edu.cn
2
Laboratory of Advanced Power Transmission Technology,
Global Energy Interconnection Research Institute Co., Ltd.,
102209 Beijing, China
3
XIAN XD Transformer Co., Ltd., 710077 Xi’an, China

Abstract. A DC breakdown model combining charge transport and molecular

chain displacement is utilized to simulate the thickness-dependent DC electrical
breakdown of polyimide and reveal the physical mechanism of DC breakdown.
The free volume existing in dielectric materials provide electrons with free path
to be accelerated and gain energy under the electric field. Molecular chains with
occupied deep traps can be displaced by Coulomb force under electric field,
furthermore, the displacement will enlarge the local free volume. The energy of
electron w is determined by the local electric field F and the length of free
volume kL, which can be expressed as w = eFkL. When the maximum energy of
electrons exceed the deep trap energy level, the local current and temperature
will rise in a surge, triggering breakdown eventually. The simulation results
reveal the dynamics of space charge and electric field inside polyimide material
before the DC electrical breakdown occurs. The breakdown strength Fb of
polyimide films obtained from the DC breakdown model decrease with an
increase in sample thickness d, which satisfies an inverse power law Fb = kd−n
with n = 0.30. A strong dependence can be found between breakdown field and
sample thickness when the influence from molecular chain displacement on free
volume is taken into consideration. The simulation results indicate that the DC
electrical breakdown may be the result of the interaction of space charge
accumulation effect and molecular chain displacement.

Keywords: DC breakdown model
Polyimide
Molecular chain

displacement
Free volume
Thickness

1 Introduction

It is known from the literature review that the electrical breakdown strength Fb of
dielectric materials are decreased with an increase in sample thickness d, following the
inverse power law Fb = kd−n [1–3]. The relationship above is only an empirical equation

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 108–117, 2020.
https://doi.org/10.1007/978-3-030-31676-1_11
 Numerical Simulation on DC Breakdown of Polyimide 109

obtained from experimental results without reasonable theories for interpretation. With
the fast-growing development of electrical power systems, thinner dielectric films with
higher breakdown strength are increasingly required. Therefore, in order to promote the
DC breakdown strength of dielectric materials more effectively, it is extremely necessary
to understand the physical mechanism of DC electrical breakdown in dielectric materials.
The formation and dynamic of space charges inside dielectric materials under
applied high voltage will determine the distribution of internal electric field and have a
strong influence on the electrical breakdown strength [4–6]. Matsui et al. have found
that the maximum electric field in LDPE before the insulation breakdown are almost
the same value of about 520 Vlm−1, which may be an intrinsic breakdown filed
existing in the polymeric materials [5]. Chen et al. proposed a model based on the
bipolar charge injection model and the formation of charge packet under high electric
filed to explain the thickness-dependent DC electrical breakdown in LDPE [7]. The
simulation results demonstrated that the electric breakdown was consequent upon the
charge dynamics. In addition, free volume exist in polymers, whose scale are very
small below the nm level, usually referring to the vacancy without molecular chains or
the unoccupied volume at the end of molecular chains [8, 9]. According to the free
volume breakdown theory proposed by Artbauer [10], free volume provide electrons
with free path to be accelerated and gain energy under the electric field. The molecular
chain with occupied deep traps will move a distance driven by the Coulomb force
under electric field, which will cause the free volume around enlarged. As a result, the
expansion of free volume enable electrons to gain higher energy. When the energy of
electrons are high enough to overcome the potential barrier of deep traps, the local
current and temperature will rise in a surge, which will trigger breakdown eventually.
In the previous work, the relationship between sample thickness and electrical break-
down field was simulated successfully in LDPE based on the effect of enlarged free
volume caused by molecular chain displacement [11].
Polyimide (PI) is a macromolecule polymeric material with excellent performances
in thermal, electrical, mechanical and radiation resistant properties and is widely used
for power equipment insulation [2, 12]. In this paper, a DC electrical breakdown model
combining space charge dynamics and molecular chain displacement effect is proposed
to simulate the thickness-dependent DC electrical breakdown of polyimide and
research the physical mechanism of DC electrical breakdown.

2 DC Breakdown Model

2.1 Charge Transport

Figure 1 shows the schematic diagram of bipolar charge transport model. In bipolar
charge transport model [7, 13, 14], electrons and holes are injected into the dielectric
material from cathode and anode respectively by Schottky thermionic emission. The
contact potential barriers exist between both metal electrodes and dielectric material,
expressed as Ein(e) and Ein(h) in eV. Electrons and holes will migrate toward anode and
cathode under electric filed after injected into the dielectric material, which may be
captured by the deep traps during migration. The accumulation phenomenon of space
110 Y. Li et al.

charges will form gradually at the interfaces near both metal electrodes with more and
more electrons and holes injected and trapped. The deep traps are only with a single
energy level, ET(e) and ET(h) for electrons and holes, respectively. Recombination occur
when electrons and holes encounter each other.

Fig. 1. Schematic diagram of bipolar charge transport model.

The injection currents densities at cathode and anode are determined by the contact
potential barriers between the material and its electrodes, electric fields at the interfaces,
and temperature [7, 13–16].

 "pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
EinðeÞ eFð0; tÞ=4pe0 er
jinðeÞ ð0; tÞ ¼ AT exp 
2
exp ð1Þ
kB T kB T

 "pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
EinðhÞ eFðd; tÞ=4pe0 er
jinðhÞ ðd; tÞ ¼ AT exp 
2
exp ð2Þ
kB T kB T

Here, jin(e) and jin(h) represent the injection current densities in Am−2 for electrons and
holes respectively; A is the Richardson constant (=1.20  106Am−2K−2); kB is the
Boltzmann constant; T is the temperature in K; e is the elementary charge; e0 is the
permittivity of free space; er is the relative permittivity of the insulating material; F(0, t)
and F(d, t) are electric fields in Vlm−1 at the interfaces of cathode and anode, respectively.
The space charges injected into the dielectric material will migrate with assistance
of shallow traps, and the conduction current densities of electrons and holes can be
expressed as:

jcðeÞ ðx; tÞ ¼ qfreeðeÞ ðx; tÞl0ðeÞ Fðx; tÞ ð3Þ

jcðhÞ ðx; tÞ ¼ qfreeðhÞ ðx; tÞl0ðhÞ Fðx; tÞ ð4Þ

 Numerical Simulation on DC Breakdown of Polyimide 111

Here, jc(e) and jc(h) are the conduction current densities for electrons and holes
respectively in Am−2; qfree(e) and qfree(h) are the density of free electrons and holes in
Cm−3; l0(e) and l0(h) are the mobilities for electrons and holes in m2V−1s−1, controlled
by shallow traps [7, 13–16].
The probability of charges trapped in deep traps Ptr(e,h) is proportional to the carrier
mobility and deep trap density NT(e,h) in m−3, but inversely proportional to the
dielectric constant [17]. which can be expressed as,

Ptrðe;hÞ ¼ eNTðe;hÞ l0 =e0 er ð5Þ

The detrapping probability Pde(e,h) of trapped charges released from deep traps can
be expressed as,

ETðe;hÞ
Pdeðe;hÞ ¼ vATE expð Þ ð6Þ
kB T

Here, ET(e,h) are the deep trap energy level for electrons and holes respectively in
eV, and vATE is the attempt-to-escape frequency in s−1.
The recombination coefficient between free electrons and holes Rel,hl and trap-
assisted recombination coefficient between free electrons and trapped holes Rel,ht, and
that between trapped electrons and free holes Ret,hl can be described as [18, 19]:
 
Rel;hl ¼ l0ðeÞ þ l0ðhÞ =e0 er ð7Þ

Rel;ht ¼ l0ðeÞ =e0 er ð8Þ

Ret;hl ¼ l0ðhÞ =e0 er ð9Þ

The unit of recombination coefficient is m3C−1s−1.

Charges in dielectric materials satisfy a set of charge continuity equations, which
can describe the trapping, detrapping, and recombination processes of electrons and
holes [7, 13–16].
 
@qfreeðeÞ ðx; tÞ @jcðeÞ ðx; tÞ qtrapðeÞ
þ ¼ PtrðeÞ qfreeðeÞ 1  þ
@t @x eNTðeÞ ð10Þ
PdeðeÞ qtrapðeÞ  Rel;hl qfreeðhÞ qfreeðeÞ  Rel;ht qfreeðeÞ qtrapðhÞ
 
@qtrapðeÞ ðx; tÞ qtrapðeÞ
¼ PtrðeÞ qfreeðeÞ 1  
@t eNTðeÞ ð11Þ
PdeðeÞ qtrapðeÞ  Ret;hl qtrapðeÞ qfreeðhÞ
 
@qfreeðhÞ ðx; tÞ @jcðhÞ ðx; tÞ qtrapðhÞ
þ ¼ PtrðhÞ qfreeðhÞ 1  þ
@t @x eNTðhÞ ð12Þ
PdeðhÞ qtrapðhÞ  Rel;hl qfreeðeÞ qfreeðhÞ  Ret;hl qtrapðeÞ qfreeðhÞ
112 Y. Li et al.

 
@qtrapðhÞ ðx; tÞ qtrapðhÞ
¼ PtrðhÞ qfreeðhÞ 1  
@t eNTðhÞ ð13Þ
PdeðhÞ qtrapðhÞ  Rel;ht qfreeðeÞ qtrapðhÞ

The subscripts l and t represent the mobile and the trapped charges, respectively.
The electric field can be calculated from the electric potential, F ¼ r/. The
electric potential distribution inside dielectric materials can be obtained by the Poisson
equation [7, 13–16].

@ 2 /ðx; tÞ qnet ðx; tÞ

¼ ð14Þ
@ x
2 2 e0 er

Here, qnet is the net charge density inside the dielectric material.
A boundary condition is required in order to solve the Poisson equation. In the
following simulation, the applied voltage is set as a positive ramp voltage with a rising
rate of 1 kVs−1. The electric potential at anode is equal to the value of applied voltage
and the potential at cathode is zero. Thus the boundary conditions can be expressed as
/(d, t) = Vappl(t) = kramptramp. Here, Vappl is the applied voltage in V, kramp is the rising
rate of the applied voltage in kVs−1, and tramp is the elapsed time after applying a DC
voltage.

2.2 Molecular Chain Displacement

The molecular chain with deep traps occupied by charges will move away from the
initial location driven by the Coulomb force under the electric field, as shown in Fig. 2.
The velocity equation for the motion can be expressed as: [11, 20]

dkD kD
¼ lmol F  ð15Þ
dt smol

Here, kD is the displacement of molecular chains in m; lmol is the mobility of

molecular chains in m2V−1s−1; and smol is the relaxation time of molecular chains in s.
The mobility of molecular chains lmol is determined by the carrier mobility controlled
by shallow trap and the probability of trapping and detrapping, namely lmol = l0Pde/
(Ptr + Pde). The relaxation time constant of molecular chains smol is equal to the
retention time of charges in deep traps, smol = s0exp(ET/kBT).
The displacement of molecular chains with deep traps occupied by charges under
electric field will enlarge the free volume existing in the polymer. Electrons will be
accelerated to obtain energy in the free volume under the influence of electric field. The
energy w of electrons depends on the local electric field F and the length kL of the free
volume, defined as w = kLeF. Therefore, the energy w will be increasing with the
increase of the length of free volume and the local electric filed. When the maximum
energy of electrons wmax = (kLeF)max exceed the deep traps energy level ET, insulation
breakdown will be initiated.
 Numerical Simulation on DC Breakdown of Polyimide 113

Fig. 2. Schematic diagram of the displacement of molecular chains with trapped charges.

In order to calculate the displacement of molecular chains, the distribution of

electric field inside the PI films need to be determined firstly. Then, we can obtain the
velocity and displacement of molecular chains with trapped charges by solving
Eq. (15). Since the initial length of free volume is negligible, it can be assumed that the
length of free volume is approximately equal to the displacement of molecular chains
with occupied deep traps. Hence the maximum energy of electrons can be expressed as
wmax = (kDeF)max. When the maximum energy of electrons is higher than the deep trap
level, a large number of electrons will jump across the deep trap barriers without
hindrance to form a large current, which will eventually lead to the insulation break-
down. Therefore, a DC electrical breakdown model combining the space charge effect
and molecular chain displacement is proposed.

2.3 Parameters
In the following simulation, the thickness of PI films ranged from 30 to 230 lm. The
temperature was set as 303 K. The energy of deep traps was extracted from the TSDC
experimental results. The density and energy of deep traps for electrons and holes were
assumed to be the same value, which were 6.25  1020 m−3 and 0.82 eV. The effective
contact barrier between the metal electrode and its dielectric material was 1.0 eV. The
carrier mobility of electrons and holes controlled by shallow traps in PI samples were
4.1  10−14 m2V−1s−1 and 2.0  10−14 m2V−1s−1, respectively, which were obtained
from the surface potential decay (SPD) test. The relative permittivity of PI was 3.4, and
the applied DC voltage increased at a rate of 1 kVs−1.

3 Simulation Results

Figure 3(a) and (b) indicate the distribution of space charge and internal electric field
inside a PI film with a thickness of 150 lm as a function of position at different times
calculated from the DC breakdown model. It can be seen from Fig. 3(a) that the charge
density due to space charge accumulation is small at the initial time, since the applied
voltage is low and the amount of charge injection from metal electrodes into the PI film
is little at the beginning. The accumulation of space charges increase gradually with the
114 Y. Li et al.

increase of applied voltage. Positive and negative charges accumulate near the anode
and cathode respectively, and then migrate toward cathode and anode under the electric
field. The local electric field is determined by the applied voltage and space charge
distribution. As shown in Fig. 3(b), at the beginning, the internal electric field has a
uniform distribution with no significant distortion. The space charge density increases
with the increasing applied voltage, leading to a more and more serious distortion of the
local electric field. For instance, at x = 15 lm near the cathode, the space charge density
q = −8.99 cm−3 at t = 10 s, q = −142.89 cm−3 at t = 30 s, and q = −251.78 cm−3 at
tb = 52.4 s. If a is defined as the ratio of the local maximum electric field to the applied
electric field, we can get a = 1.06 at t = 10 s, a = 1.29 at t = 30 s, and a = 1.33 at
tb = 52.4 s. The maximum value of internal electric field appears at the middle of the PI
film, which increases non-linearly with time.

1600 600
t1=10s (Vappl=10kV) t2=20s (Vappl=20kV)
t1=10s (Vappl=10kV)
t3=30s (Vappl=30kV) t4=40s (Vappl=40kV)
Space charge density (Cm -3)

1200 t2=20s (Vappl=20kV) 500

tb=52.4s (Vappl=52.4kV)
t3=30s (Vappl=30kV)
Electrical field (Vμm -1 )

800 t4=40s (Vappl=40kV) (a) 400 (b)

tb=52.4s (Vappl=52.4kV)
400
300
0
200
-400
100
-800
0
0 30 60 90 120 150 0 30 60 90 120 150
Cathode Position (μm) Anode Cathode Position (μm) Anode

Fig. 3. Numerical results of the space charge density (a) and internal electric field (b) as a
function of position at various times. tb corresponds to the moment of electrical breakdown, and
Vappl represents the applied voltage at different times.

Figure 4(a) demonstrates the displacement of molecular chains with occupied deep
traps inside the PI film with a thickness of 150 lm as a function of position at various
times. According to the equation of molecular chain displacement, molecular chains
with positive and negative charges will move toward cathode and anode respectively
under the internal electric field, and the velocity of molecular chain displacement is
proportional to the electrical field. Therefore, the curves of Fig. 3(b) and (a) have the
same trend. It can be seen that the maximum value of molecular chain displacement
kmax increases as the applied voltage increases. For example, kmax = 0.06 nm at
t = 15 s, kmax = 0.53 nm at t = 30 s, and kmax = 1.80 nm at t = 52.4 s. The maximum
value of molecular chain displacement also appears at the middle position of the PI
film, and the value of the displacement is relatively small near both metal electrodes.
Figure 4(b) shows the maximum energy of electrons obtained from the local
electric field. The molecular chain with occupied deep trap will move a distance driven
by Coulomb force under the electric field. The displacement of molecular chain could
 Numerical Simulation on DC Breakdown of Polyimide 115

cause the free volume around enlarged. In the initial state, since the length of free
volume inside PI is negligible at room temperature, it can be assumed that the length of
free volume kL is equal to the value of molecular chain displacement kD. Electron can
be accelerated to get a certain energy in free volume under electric field. The energy of
electron w obtained from electric field in free volume depends on the length of free
volume and the local electric field strength, which can be described as w = kDFe. When
the maximum energy of electrons wmax = (kDFe)max are higher than the deep trap
energy level ET, local current and temperature will rise in a surge, hence triggering
electric breakdown eventually. As shown in Fig. 4(b), breakdown occurs when the
maximum energy of electrons exceed the deep trap energy. The breakdown time tb of
PI films are increased non-linearly with the increase of films thickness d, so the
breakdown field Fb of PI films also increase with the thickness d due to Fb = kramptb/d.

t1=10s (Vappl=10kV) t2=20s (Vappl=20kV)

Molecular chain displacement (nm)

t3=30s (Vappl=30kV) t4=40s (Vappl=40kV) 0.8

Maximum energy of electron (eV)
2.0 30μm
tb=52.4s (Vappl=52.4kV) 70μm
(a) (b)
110μm
1.5 0.6
150μm
190μm
1.0 0.4 230μm

0.5 0.2

0.0
0.0
0 30 60 90 120 150 0 20 40 60 80 100
Position (μm) Time (s)

Fig. 4. Numerical results of molecular chain displacement as a function of position at various

times (a) and numerical results of the maximum energy of electrons as a function of time of
polyimide films with various thickness (b).

Figure 5 shows the DC breakdown strength of PI films as a function of thickness

calculated by the DC breakdown model in double logarithmic coordinates. It is obvious
that the DC breakdown strength of PI decreases with an increase in films thickness.
There is a linear relationship between electric breakdown strength Fb and thickness d in
the double logarithmic coordinates, which can be expressed as an inverse power
function Fb = kd−n with an inverse power index n = 0.30. A strong dependence can be
seen between breakdown field and sample thickness when considering the influence
from the molecular chain displacement on free volume.
116 Y. Li et al.

650

Electrical breakdown strength (Vμm-1)

600
550
500
450
n=0.30
400

350

300

101 102 103

Sample thickness (μm)

Fig. 5. Numerical results of the DC electrical breakdown field as a function of thickness of

polyimide films in the double logarithmic coordinates.

4 Conclusions

A DC electrical breakdown model modulated by charge transport and molecular chain

displacement are introduced. The molecular chain with trapped charges can move a
distance driven by the Coulomb force under local electric field, which will lead to the free
volume enlarged. The length of free volume is approximately equal to the displacement of
molecular chain. Electron can obtain a certain amount of energy by being accelerated in
free volume under local electric field, and the obtained energy is determined by the length
of free volume and the local electric field strength, namely w ¼ kL Fe. When the max-
imum energy of electrons are large enough to jump across the deep trap barriers, local
current and temperature will increase in a surge, eventually triggering electrical break-
down. The dynamics of space charge and the distribution of internal electric field before
insulation breakdown are calculated by the charge transport model.
The simulated results of electrical breakdown field Fb of PI films decrease with the
increase of sample thickness d, following an inverse power law with power index
n = 0.30. A strong dependence between electric breakdown field and sample thickness
can be seen from the simulation results when considering the influence from molecular
chain displacement on the free volume and taking the maximum electron energy
exceeding the deep trap barrier as the criteria for the occurrence of insulation break-
down. The thickness-dependent DC electrical breakdown may be the result of the
interaction of space charge accumulation effect and molecular chain displacement.
Further experiments are needed to verify the simulation results.

Acknowledgments. This work was supported by State Key Laboratory of Advanced Power
Transmission Technology (Grant No. GEIRI-SKL-2018-010), the National Basic Research
Program of China (grant No. 2015CB251003), and the National Natural Science Foundation of
China (grant No. 51507124).
 Numerical Simulation on DC Breakdown of Polyimide 117

References
1. Cooper, R., Rowson, C.H., Watson, D.B.: Intrinsic electric strength of polythene. Nature 197
(4868), 663–664 (1963)
2. Diaham, S., Zelmat, S., Locatelli, M.L.: Dielectric breakdown of polyimide films: area,
thickness and temperature dependence. IEEE TDEI 17(1), 18–27 (2010)
3. Kim, H.K., Shi, F.G.: Thickness dependent dielectric strength of a low-permittivity dielectric
film. IEEE TDEI 8(2), 248–252 (2001)
4. Laurent, C., Teyssedre, G., Le Roy, S., Baudoin, F.: Charge dynamics and its energetic
features in polymeric materials. IEEE TDEI 20(2), 357–381 (2013)
5. Matsui, K., Tanaka, Y., Takada, T., Fukao, T., Alison, J.M.: Space charge behavior in low
density polyethylene at pre-breakdown. IEEE TDEI 12(3), 406–415 (2005)
6. Zha, J.W., Dang, Z.M., Song, H.T., Yin, Y., Chen, G.: Dielectric properties and effect of
electrical aging on space charge accumulation in polyimide/TiO2 nanocomposite films.
J. Appl. Phys. 108, 094113 (2010)
7. Chen, G., Zhao, J., Li, S., Zhong, L.: Origin of thickness dependent dc electrical breakdown
in dielectrics. Appl. Phys. Lett. 100, 222904 (2012)
8. Fox, T.G., Flory, P.J.: The glass temperature and related properties of polystyrene. Influence
of molecular weight. J. Polym. Sci. 14(75), 315–319 (1954)
9. Fox, Jr., T.G., Flory, P.J.: Further studies on the melt viscosity of polyisobutylene. J. Phys.
Chem. B. 55(2), 221–234 (1951)
10. Artbauer, J.: Electric strength of polymers. J. Phys. D. 29(2), 446–456 (1996)
11. Min, D., Li, S., Ohki, Y.: Numerical simulation on molecular displacement and DC
breakdown of LDPE. IEEE TDEI 23(1), 507–516 (2016)
12. Dang, Z.M., Zhou, T., Yao, S.H., Yuan, J.K., Zha, J.W., Song, H.T.: Advanced calcium
copper titanate/polyimide functional hybrid films with high dielectric permittivity. Adv.
Mater. 21(21), 2077–2082 (2010)
13. Li, S., Min, D., Wang, W., Chen, G.: Modelling of dielectric breakdown through charge
dynamics for polymer nanocomposites. IEEE TDEI 23(6), 3476–3485 (2017)
14. Li, S., Zhu, Y., Min, D., Chen, G.: Space charge modulated electrical breakdown. Sci.
Rep. 6, 32588 (2016)
15. Hoang, A.T., Pallon, L., Liu, D., Serdyuk, Y.V., Gubanski, S.M., Gedde, U.W.: Charge
transport in LDPE nanocomposites Part I—experimental approach. Polymers 8, 87 (2016)
16. Le Roy, S., Segur, P., Teyssedre, G., Laurent, C.: Description of bipolar charge transport in
polyethylene using a fluid model with a constant mobility: model prediction. J. Phys. D. 37
(2), 298–305 (2004)
17. Kuik, M., Koster, L.J.A., Wetzelaer, G.A.H., Blom, P.W.M.: Trap-assisted recombination in
disordered organic semiconductors. Phys. Rev. Lett. 107, 25 (2011)
18. Shockley, W., Read, W.T.: Statistics of the recombinations of holes and electrons. Phys.
Rev. Lett. 87(5), 835–842 (1952)
19. Kuik, M., Koster, L.J.A., Dijkstra, A.G., Wetzelaer, G.A.H., Blom, P.W.M.: Non-radiative
recombination losses in polymer light-emitting diodes. Org. Electron. 13(6), 969–974 (2012)
20. Lowell, J.: Absorption and conduction currents in polymers: a unified model. J. Phys. D. 23
(2), 205–210 (1990)
 Thermal Instability Analysis of Station
Class Surge Arresters Based on
Electrothermal Finite Element Simulation

Yvonne Späck-Leigsnering1(B) , Maren Greta Ruppert1 , Erion Gjonaj1 ,

Herbert De Gersem1 , and Volker Hinrichsen2
1
Technische Universität (TU) Darmstadt,
Institut für Teilchenbeschleunigung und Elektromagnetische Felder (TEMF),
Schloßgartenstr. 8, 64289 Darmstadt, Germany
spaeck@temf.tu-darmstadt.de
2
TU Darmstadt, Fachgebiet Hochspannungstechnik,
Fraunhoferstr. 4, 64283 Darmstadt, Germany
http://www.temf.de/

Abstract. The phenomenon of thermal instability is an important prac-

tical problem of station class surge arresters. This effect may occur after
single or repetitive energy impulse injections to an arrester connected to
the power grid. Thermal stability of full scale arresters cannot be inves-
tigated in the laboratory. Finite element simulation provides detailed
insight to this complex electrothermally coupled problem. Therefore,
an ungraded arrester is modeled including all relevant electrothermal
physics. After the model is validated, various thermally stable and unsta-
ble scenarios are investigated. A criterion that assesses and predicts ther-
mal instability of an arrester is determined. Furthermore, selected model
parameters that increase the thermal stability limit are identified in order
to optimize future arrester designs.

Keywords: Surge arrester · Thermal instability · Electrothermal

simulation

1 Introduction

Surge arresters are crucial devices to protect electrical equipment in power sys-
tems from transient overvoltages. Their core consists of metal oxide resistors,
commonly zinc oxide (ZnO). These absorb the impulse energy due to a strongly
nonlinear, field- and temperature-dependent conductivity [6]. Thermal instabil-
ity occurs if an impulse exceeds the maximum permissible temperature while
the arrester is connected to the power grid. This leads to catastrophic failure
of the arrester. In the laboratory, thermal stability of station class arresters can
be assessed according to the IEC 60099-4 operating duty test only for small,
thermally equivalent arrester models [7,8].

c Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 118–130, 2020.
https://doi.org/10.1007/978-3-030-31676-1_12
 Thermal Instability Analysis 119

Electrothermal Finite Element (FE) simulation, is an alternative approach

to study arresters. In recent years, FE simulations of surge arresters have been
performed by several authors. The arrester’s internal axial operating voltage and
temperature distribution was investigated in [1,5,9]. Zheng et al. [15,16] studied
thermal stability of small arresters. However, no thermally unstable scenario was
considered. A specialized technique for the numerical solution of the nonlinear
and coupled electroquasistatic-thermal (EQST) problem of large station class
arresters was introduced in [11–13] and applied to study arresters subjected to
continuous and impulse operating conditions [3,4]. The simulations reproduced
laboratory experiments with very high accuracy.
In this work, we employ the simulation method of [12] to study thermal insta-
bility of an ungraded, full-scale station class surge arrester. Lightning current
impulses are simulated as instantaneous temperature rises.
During the cool-down period, the modeled arrester remains connected to the
grid. The transient cooling rate is, then, computed. Parameters that influence
the cooling behavior and may, in the worst case, lead to thermal instability, are
identified. This is the first step to define suitable parameters and goal functions
for optimizing the arrester design.

2 Arrester Model and Simulation

2.1 Setup
The investigated station class arrester (continuous operating voltage, Uc =
330 kV (rms)) is depicted in Fig. 1. It was experimentally investigated in the
High Voltage Laboratory of the Technische Universität Darmstadt (see [3]). The
4-segment arrester is 4.3 m high and consists of a central ZnO resistor column
which is inserted in a porcelain housing, as shown in Fig. 1 (b), (c). The arrester
is equipped with four distributed metallic heat sink elements and six temperature
measurement disks in each segment. The porcelain sheds are neglected. No field
grading system is used and the arrester model is, thus, two-dimensional (2D)
axisymmetric. Further dimensions and material characteristics are summarized
in Table 1. The ZnO resistors have a comparably small diameter of 47.8 mm. The
conductivity, σ, and permittivity, ε, of the ZnO resistors vary strongly with the
electric field, E, and the temperature, T . In this work, the electrical ZnO mate-
rial characteristics are provided by analytical, field- and temperature-dependent
functions (see Fig. 2). Their respective coefficients are obtained by fitting mea-
sured U-I-T-curves of the ZnO resistors used in the measurements [2]. The char-
acteristics are fitted in the leakage and early breakdown region (see [10]). The
conductivity reads,
 α 
|E|
σ(|E|, T ) = + a2 exp[(T − a3 )a4 ]. (1)
a1
The relative permittivity is given by,
ε(|E|, T ) = [exp(b1 |E|) + b2 ] exp[b3 T − b4 ], (2)
120 Y. Späck-Leigsnering et al.

where the coefficients are a1 = 4.00·105 Vm−1 , a2 = 1.00·10−7 Sm−1 , a3 = 290 K,

a4 = 0.02 K−1 , b1 = 2.61 · 10−5 mV−1 , b2 = 943, b3 = 2.90 · 10−3 K−1 and b4 =
0.99. The nonlinearity coefficient of the resistors, α, is the most important param-
eter as it defines the switching point of the arrester. In a first step of the inves-
tigation, α = 20 is adopted.

Fig. 1. (a) Investigated station class surge arrester [2]. (b) Detail of single segment ZnO
column with distributed heat sink elements and temperature measurement disks, [2].
(c) Schematic view of the arrester model (without grading system) with detailed view
of the ZnO column with centered heat sink element, air gap and porcelain housing.

2.2 Electrothermal Simulation

In order to simulate the capacitive-resistive behavior of an arrester, the transient

electroquasistatic (EQS) equation is solved,

∂t div (ε grad(φ)) + div (σ grad(φ)) = 0, (3)

where the electric potential is φ. The electric material characteristic of the ZnO
resistors, σ = σ(|E|, T ), ε = ε(|E|, T ) are strongly field and temperature depen-
dent, respectively. Heat transfer in surge arresters involves thermal conduction,
natural convection and thermal radiation. External heat transfer to the environ-
ment can be described by boundary conditions. Moreover, in [12] the authors
presented an approach to model the internal heat transfer in the arrester air
 Thermal Instability Analysis 121

Table 1. Model specifications.

Name Unit Value

Total arrester height mm 4216
Height of single segment, h mm 902
Height of single ZnO stack mm 250
ZnO stack radius, r1 mm 23.9
Housing, inner radius, r2 mm 68
Housing, outer radius mm 100
ZnO thermal conductivity W(mK)−1 21.5
3 −1
ZnO volumetric heat capacity J(cm K) 2.63
Porcelain relative permittivity – 6
Porcelain thermal conductivity W(mK)−1 1.6
Porcelain volumetric heat capacity J(cm3 K)−1 2.16
ZnO emissivity, εrad,1 – 0.96
Porcelain emissivity, εrad,2 – 0.85
Flange emissivity – 0.25
External convection W(m2 K)−1 3

gap. For this purpose, an effective material with nonlinear thermal conductiv-
ity, λ(T ), was introduced in the air gap (cf. [12] for a detailed description). A
complete thermal model is obtained by solving the transient heat conduction
equation,
∂t (cv T ) − div(λ(T ) grad(T )) = q̇, (4)
where q̇ is the power loss density computed in (3), λ and cv are the heat con-
ductivity and specific volumetric heat capacity, respectively.
The coupled system of Eqs. (3) and (4), is solved with the multirate time
integration technique of [12]. This approach exploits the extremely different time
constants. The electrical time constant is τel ≤ 100 µs in continuous operation.
It decreases to 1 µs after an overvoltage event. In contrast, the thermal time con-
stant is τel ≈ 1 min. The idea is that each subsystem is solved with its respective
time step. First, the EQS problem (3) is computed until a local electric station-
ary state is obtained due to the alternating current (ac) voltage excitation. The
power losses per ac period are obtained. Second, (4) updates the temperature
distribution. This procedure is repeated until the simulation is finished.

3 Results
3.1 Model Validation
For the numerical validation of the 3D FE arrester model, the simulated and the
measured steady state temperature distribution are compared. The measurement
122 Y. Späck-Leigsnering et al.

298 K
2,000

relative permittivity εr
333 K
388 K
423 K
463 K
1,500

1,000

0 0.5 1 1.5 2
electric field |E| in V m−1 ·105
(a)
10−2
298 K
conductivity σ in S m−1

−3 333 K
10
388 K
423 K
10−4
463 K

10−5

10−6

10−7
0 0.5 1 1.5 2
−1
electric field |E| in V m ·105
(b)

Fig. 2. (a) Relative permittivity and (b) electric conductivity of ZnO material.

data are taken from [3]. Two FE models are simulated. In model (A), the heat
sink elements are distributed, as shown in Fig. 1(b) (likewise to the experimental
setup of [3]). In model (B), all heat sink elements are combined and centered in
each segment, as shown in Fig. 1(c). The heat sinks are hollow, the wall thickness
is set to 8.5 mm. The arrester is energized with the ac voltage (50 Hz) of 330 kV.
The computational domain spans 8 m in radial- and 9 m in z-direction, respec-
tively. The floor is grounded. After approximately six hours, the arrester reaches
the EQST steady state in the experiment and the simulation, respectively. The
computation time is in the (low) range of hours for a problem size of tens of
thousands of FE nodes (Fig. 2).
Figure 3 shows the comparison of the measured and simulated steady-state
temperature distribution. An excellent agreement of the measurement and simu-
lation model (A) is obtained (relative error ≤5.8%). Model (B) with a simplified
 Thermal Instability Analysis 123

geometry of the heat sinks, yields an increased peak temperature, of 419 K. Thus,
even for this simplified model, the temperature deviation in the simulation is not
larger than 16 K. The mean temperature of both simulation models is 327 K. In
the following, model (B) is used to study thermal instability as it allows for
faster simulations and simpler variation of multiple parameters.

420
measurement
400 simulation, model (A)
simulation, model (B)
temperature T in K

380

360

340

320

300

280
0 0.5 1 1.5 2 2.5 3 3.5 4
height z in m

Fig. 3. Comparison between the steady-state temperature distribution obtained in

measurement and simulation.

3.2 Energy Injection

In the power transmission system, lightning strikes or switching occurs. The

impulse energy of these events immediately rises the temperature in the ZnO
resistors. In this work, instantaneous temperature impulses of the magnitude,
ΔT , are applied to the ZnO resistors to simulate the energy injection event.
The impulse injection event occurs while the arrester is connected to the grid.
Hence, the EQST steady-state temperature distribution Teqst = T (
, z, t = 0) is
assumed as initial condition.
The cooling of the ZnO resistors, ηZnO , along the arrester axis is studied.
Therefore, the cooling rate is defined as
∂T
η := ηZnO (z, t) = −(T − Teqst )−1 . (5)
∂t
A ZnO element is stable, if η(z, t) ≥ 0. If this holds for all ZnO elements for
t → ∞, the surge arrester is thermally stable.
124 Y. Späck-Leigsnering et al.

If the injected energy is above the arrester’s thermal stability limit, a thermal
runaway occurs, i.e. T (
, z, t → ∞) → ∞. Otherwise, the arrester cools down to
the previous steady state T (
, z, t → ∞) → Teqst (
, z, t = 0).

3.3 Thermal Instability Analysis

Figure 4 shows the cooling rate, η(t), in each segment for two scenarios, (a)
and (b). In Fig. 4(a), an impulse of ΔT = 250 K and in (b) ΔT = 270 K is
applied, respectively. These impulses are by approximately factor two above the
temperatures expected after injecting the nominal energy (cf. [6]) in order to
provoke thermal instability. The cooling rate is computed for three time instants,
at 2 min, 6 min, and 10 min after energy injection, respectively. In scenario (a),
the lower two segments initially heat up for the first 2 min and afterwards start
cooling down (η(t ≤ 2 min) ≤ 0). The two upper segments cool down from the
beginning. Thus, this scenario is thermally stable. In scenario (b), the lower
segments heat up continuously at all time instants (η(t) ≤ 0). The top segment
cools down initially. Finally, all segments heat up. Thus, scenario (b) is thermally
unstable.
Figure 5 shows the mean ZnO temperature, T ZnO (t), (a), and mean cool-
ing rate, η ZnO (t) (averaged over all segments), (b), for both scenarios. Thermal
instability can be detected already after one thermal time step based on the
value of η ZnO (t). This is an important observation, as the numerical simulation
of unstable scenarios is computationally expensive. The computation of each
thermal time step takes several hours (for a problem size of several tens of thou-
sands of nodes). This is because the working point of the ZnO resistors at very
high temperatures is shifted to the severely nonlinear region. In the following,
different parameters that affect the thermal stability limit of the arrester are
analyzed. Therefore, η ZnO is evaluated after a single thermal step.
How can the thermal stability limit of the arrester be increased? The heat
transfer to the environment can be improved. To investigate the effect of heat
transfer, we introduce model (C) which is equipped with solid heat sinks. Figure 6
shows that the thermal stability limit is slightly increased (5 K) by using solid
heat sinks compared to the hollow heat sinks of model (B).
A stronger effect can be observed by improving the heat transfer in the
arrester air gap. Based on [12,13] the heat transfer in the air gap is modeled
accurately by an equivalent thermal conductivity. It includes natural convection
and thermal radiation. The nonlinear thermal conductivity in the air gap is given
by,

λgap (T ) = λair + λconv + λrad (T ), (6)

where λair , is the conductive, λconv is the convective, and λrad is the radiative
heat transfer contribution in the air gap, respectively. Heat conduction and con-
vection are described by the mean Nusselt number, Nu, for a closed annular gap
 Thermal Instability Analysis 125

6 2 min
6 min

cooling rate η in h−1

10 min
4

bottom, lower mid, upper mid, top

arrester segment
(a)
4
2 min
2 6 min
cooling rate η in h−1

10 min
0

−2

−4

−6
bottom, lower mid, upper mid, top
arrester segment
(b)

Fig. 4. (a) Cooling rate in each segment for stable scenario (250 K). (b) Cooling rate
in each segment for unstable scenario (270 K).

(see [12,14]),
λair + λconv = Nuλair , with (7)
 2
c1 hδ
Nu = 3
h r1
 h 9/4 , (8)
r2 4 + c2 δ

where
c1 = 5.62 · 10−4 Gr(T1 − T2 ) Pr, (9)
854−1 ,
3/4
c2 = [Gr(T1 − T2 ) Pr] (10)
h is the height of the segment, δ is the air gap width, T1,2 are the temperatures
at the inner and outer air gap surface, Gr is the Grashof number, and Pr is
126 Y. Späck-Leigsnering et al.

mean temperature T ZnO in K

600 250 K
270 K

550

1 2 3 4 5 6 7 8 9
time t in min
(a)
mean cooling rate η ZnO in h−1

250 K
270 K
0

−2
2 3 4 5 6 7 8 9
time t in min
(b)

Fig. 5. Mean temperature (a) and mean cooling rate (b) the stable (250 K) unstable
scenario (270 K).

the Prandtl number. Radiative heat transfer in the air gap is described by (see
[12,13]),

4Cs T 3 r1 ln( rr21 )

λrad (T ) = , (11)
(ε−1
rad,1 +
r1
r2 · (ε−1
rad,2 − 1))

where Cs is the Stefan-Boltzmann constant and εrad,1,2 are the emissivities of

the inner and outer walls of the air gap, respectively.
In model (D), the arrester segment is divided in separate sub-segments of
h = 0.2 m. This enhances the convective transfer. Furthermore, the inner surfaces
of the air gap vessel are considered to be perfectly heat absorbing (εrad,1,2 = 1).
This increases the radiative heat transfer. Figure 7 shows, that these measures are
very effective. The thermal stability limit is shifted by 20 K. Both modifications
 Thermal Instability Analysis 127

mean cooling rate η ZnO in h−1

model (B)
2 model (C)

−2

−4
240 260 280
impulse temperature ΔT in K

Fig. 6. Cooling rate comparison of model (B) and (C) for different impulses.

may be realized easily. Non-conducting sheets could be inserted in the segments

to subdivide the air gap in separate compartments. A black coating could be
applied to the ZnO column and the inner surface of the housing, respectively.
mean cooling rate η ZnO in h−1

−2 model (B)
model (D)

240 260 280

impulse temperature ΔT in K

Fig. 7. Cooling rate comparison of model (B) and (D) for different impulses.

Moreover, the electric material nonlinearity influences the thermal stability

limit strongly. Figure 8 shows the cooling rate for different degrees of nonlinear-
ity, α, in the ZnO conductivity characteristic σ(α, |E|, T ) of (1). The stronger
the nonlinearity with respect to |E| in (1) is, the higher is the thermal stabil-
ity limit. The difference of the instability limit amounts to 70 K between the
lowest (α = 18) and highest (α = 22) degree of nonlinearity considered in the
investigation. The reason is that α determines the resistive current contribu-
tion during the ac voltage excitation after the impulse injection. In this mode
128 Y. Späck-Leigsnering et al.

mean cooling rate η ZnO in h−1

2

0
α = 18
α = 19
−2 α = 20
α = 21
−4 α = 22

160 180 200 220 240 260 280 300

impulse temperature ΔT in K

Fig. 8. Cooling rate comparison for increasing degrees of nonlinearity α of the ZnO
conductivity characteristic, σ(α).

of operation, the current is predominantly capacitive. An increased α further

reduces the resistive component. Thus, the losses generated by the ac voltage
are minimized. Figure 9, additionally, shows that the instability limit cannot
be increased indefinitely. A saturation of the critical temperature for high α is
observed. An optimization could, thus, aim for an α that balances optimal ac
operation, requirements regarding the impulse switching point and the thermal
stability limit, respectively.

300
stability limit ΔTintab

280

260

240

220

18 19 20 21 22
fitting parameter α

Fig. 9. Thermal stability limit over increasing degrees of nonlinearity α of the ZnO
conductivity characteristic, σ(α).
 Thermal Instability Analysis 129

4 Conclusion
Electrothermal Finite Element simulation is a powerful tool to investigate ther-
mal instability of full-scale station class surge arresters. In a first step, a valida-
tion of an ungraded Finite Element arrester model was presented. An excellent
agreement of measurement and simulation for ac operation was achieved. Ther-
mal instability following energy injections was identified after a single thermal
time step based on the mean cooling rate of the zinc oxide resistors. The air-
gap heat transfer and zinc oxide conductivity characteristic affected the thermal
stability limit. Improved radiation and natural convection in the arrester air
gap increase the thermal stability limit. Furthermore, increasing the degree of
nonlinearity with respect to the electric field rises the instability limit. Both
are promising parameters for a thermal stability optimization of future arrester
designs.

References
1. Clemens, M., Steinmetz, T., Weida, D., Hinrichsen, V.: Coupled thermal-
electroquasistatic 3D field simulation of high-voltage surge arrester structures.
In: Computational Electromagnetics Conference (CEM), Aachen, Germany, April
2006
2. Gießel, M.: Elektrothermisches Verhalten von Hochspannungs–Metalloxid–
Ableitern mit reduzierten Steuersystemen in Wechselspannungsnetzen. Disserta-
tion, Technische Universität Darmstadt (2018)
3. Gießel, M., Hinrichsen, V., Göhler, R., Späck-Leigsnering, Y., Gjonaj, E.,
De Gersem, H.: Einfluss unterschiedlicher steuerringkonfigurationen auf die ther-
mische stabilität von überspannungsableitern. In: VDE-Fachtagung Hochspan-
nungstechnik 2016, November 2016
4. Gießel, M., Hinrichsen, V., Göhler, R., Späck-Leigsnering, Y., Gjonaj, E.,
De Gersem, H.: Electro-thermally coupled finite-element simulations of high volt-
age station arresters with and without grading, November 2017
5. Göhler, R., Schubert, M., Weck, K., Hinrichsen, V., Tuczek, M.N., Clemens, M.,
Appel, R.: Special requirements on surge arrester design for UHV A.C. systems
above 800 kV system voltage. In: CIGRÉ Konferenz Report A3-104-2010, Paris,
August 2010
6. Hinrichsen, V.: Metal-Oxide Surge Arresters in High-Voltage Power Systems - Fun-
damentals, 3rd edn. Siemens AG, Munich (2011)
7. Hinrichsen, V., Gießel, M., Tuczek, M.: Thermal stability of HV and UHV arresters
with reduced grading systems. In: The 2015 INMR World Congress on Insulators,
Arresters, Bushings & Cable Accessories, Munich, October 2015
8. IEC 60099-4: Surge arresters – Part 4: Metal-oxide surge arresters without gaps
for a.c. systems. 3 edn. International Electrotechnical Commission (IEC), Geneva,
Switzerland, June 2014
9. Richter, B., Dellallibera, A., Greuter, F., Holzer, M., Ishizaki, Y., Kobayashi,
M., Rawi, I.M., Tuczek, M., Woodworth, J., Comber, M., Göhler, R., Hinrich-
sen, V., Ishibe, S., Johnnerfelt, B., Fan, L., Späck-Leigsnering, Y., Nakajima, M.,
Österlund, R.: MO surge arresters: metal oxide varistors and surge arresters for
emerging system conditions; Technical Broschure 696. Cigré Working Group A3.25
(2017)
130 Y. Späck-Leigsnering et al.

10. Späck-Leigsnering, Y., Gjonaj, E., De Gersem, H.: Electrothermal optimization of

field grading systems of station class surge arresters. IEEE J. Multiscale Multiphys.
Comput. Tech. 4, 29–36 (2019)
11. Späck-Leigsnering, Y., Gjonaj, E., De Gersem, H., Weiland, T., Gießel, M., Hinrich-
sen, V.: Thermal analysis of a station class arrester model in continuous operation
and under the operating duty test. In: 19th International Symposium on High
Voltage Engineering, August 2015
12. Späck-Leigsnering, Y., Gjonaj, E., De Gersem, H., Weiland, T., Gießel, M., Hin-
richsen, V.: Electroquasistatic-thermal modeling and simulation of station class
surge arresters. IEEE Trans. Magn. 52(3), 1–4 (2016)
13. Späck-Leigsnering, Y., Gjonaj, E., De Gersem, H., Weiland, T., Gießel, M., Hin-
richsen, V.: Investigation of thermal stability for a station class surge arrester.
IEEE J. Multiscale Multiphys. Comput. Tech. 1, 120–128 (2016)
14. VDI-Gesellschaft Verfahrenstechnik und Chemieingenieurwesen (ed.): VDI Heat
Atlas. Springer (2010)
15. Zheng, Z., Boggs, S.A.: Heat sink effects in thermal stability tests of ZnO arresters.
In: Conference on Electrical Insulation and Dielectric Phenomena, Annual Report,
pp. 473–477 October 2003
16. Zheng, Z., Boggs, S.A., Imai, T., Nishiwaki, S.: Computation of arrester thermal
stability. IEEE Trans. Power Deliv. 25(3), 1526–1529 (2010)
 The Comparison of Approaches
to Power Frequency Electromagnetic
Field Hygienic Regulations

Nina Rubtsova, Sergey Perov(&), and Olga Belaya

FSBSI “Izmerov Research Institute of Occupational Health”, 31, Prospect

Budennogo, Moscow 105275, Russia
sergey.perov@gmail.com

Abstract. Power frequency electromagnetic fields are the risk factor for occu-
pational and general public health. International and national electromagnetic
safety guidelines set the exposure limit values and have some significant differ-
ences. The goal of this paper is to compare Russian hygienic norms with ICNIRP
guidelines, EU Directive requirements for 50 Hz electromagnetic fields occupa-
tional and general public exposure, analyze and demonstrate the principal dif-
ferences. The protection from harmful human electromagnetic field effect is based
on principles: protection by time, protection by distance and protection by pro-
tective equipment. Various hygienic regulations use different approaches to the
human harmful effect definitions therefore there are distinctions of electromag-
netic fields permissible limit values in International and national electromagnetic
safety standards and guidelines. ICNIRP guidelines and Directive EU regard
safety limits only from short-term, acute effect. Chronic electromagnetic field
exposure harmful effect threshold is a basis for Russian hygienic norm. These
thresholds are defined as a results of biomedical complex researches. Protection by
time principle is realized in Russian hygienic norms, which are strong time
dependent for occupational exposure. It is the main distinctive characteristics from
International hygienic guidelines and is based on chronic exposure harmful human
health effect threshold definition and electromagnetic field cumulative effects
concept. Russian permissible limit value (25 kV/m) may be higher than ICNIRP
(10 kV/m) and Directive EU (20 kV/m) levels, but is limited by working time (no
more than 10 min per day). In Russia for general public permissible limit values
are graded according to possible exposure time per day.

Keywords: Power frequency electromagnetic field
Protection principles

Permissible limit value

1 Introduction

Human safety from electromagnetic field (EMF) exposure is actual issue due to
increasing environmental electromagnetic background for workers and general public.
The various common applications, such as electrical transport systems, industrial and
medical equipment, overhead and cable transmission lines, transformers and distribu-
tive substations, are the main sources of power frequency EMF.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 131–137, 2020.
https://doi.org/10.1007/978-3-030-31676-1_13
132 N. Rubtsova et al.

Power frequency EMF exposure is a risk factor for human health. Power frequency
magnetic field relates to extremely low frequency and is classified as possibly car-
cinogenic to humans (Group 2B) by International Agency for Research on Cancer [1].
Occupational and general public Russian hygienic norms are the basis of human
electromagnetic safety and must ensure the health of present and subsequent genera-
tions’ preservation. International and national electromagnetic safety guidelines set the
exposure limit values and have some significant differences. The aim of this paper is to
compare Russian hygienic norms [2–5] with ICNIRP guidelines [6], Directive EU [7,
8] requirements for 50 Hz EMF exposure, analyze and demonstrate the principal
differences.

2 Electromagnetic Field Human Protection Principles

The protection from harmful human EMF effect is based on 3 principles: protection by
time, protection by distance and protection by protective means (collective and
individual).
Protection by distance principle is maximal removal of the workplaces or non
occupational exposure places from EMF source and high intensity EMF exposure zone.
This principle for workers’ exposure condition is realized by means of staff authorized
restriction, automation, mechanization, remote control, manipulators, etc. In particular,
for general public protection, right-of-way (sanitary-protective zones) are organized.
Protection by time allows to increase staff exposure limit values higher than
hygienic norms for all working day. This protection principle is applied when it’s
necessary to work in higher EMF levels during short time period and there is no
opportunity to reduce it.
Collective and individual protective means use allows to reduce the high intensity
EMF at work places to permissible levels by shielding devices, clothes, etc.
Safe EMF exposure levels for humans should be in accordance with permissible
levels established as hygienic requirements. Hygienic norms are based on harmful EMF
effect threshold definition. Various hygienic regulations use different approaches to the
human adverse EMF effect definitions because of distinctions of EMF permissible level
values in international and national electromagnetic safety standards and guidelines.
ICNIRP guidelines and Directive EU regard safety limits from short-term, acute
EMF effect only, but long-term exposure are excluded from the their scope. In con-
tradiction to international EMF guidelines chronic (long-term) EMF exposure adverse
effect threshold is the basis of Russian hygienic norms. Its are defined as a results of
complex biomedical researches (hygienic, clinical, physiological, experimental and
epidemiological) [9].
International EMF safety standards set 2 parameters for exposure limit: basic
restrictions and reference levels. The internal electric field (mV/m), induced in bio-
logical body and tissues by power frequency electric or magnetic fields, is used as
ICNIRP basic restrictions and Directive EU exposure limit values (ELV) for power
frequency. Internal electric field limits are based directly on established health effects
and biological considerations as the relevant biophysical parameter to characterise the
nervous tissue excitation. But internal electric field quantities are impractical and
 The Comparison of Approaches to Power Frequency EMF 133

cannot be simply measured or calculated. Another quantities are introduced by inter-

national EMF safety guidelines for practical use. External electric field (kV/m) and
magnetic induction (mT, lT) level are set as ICNIRP reference levels and Directive EU
action levels (AL) for power frequency. These parameters are used for practical EMF
exposure assessment.
Furthermore, Directive EU [7] introduces two distinct thresholds for both the ELVs
and the ALs for occupational exposure conditions: low and high.
Principle protection by time is realized in Russian hygienic norms [2], which are in
dependency by time for occupational exposure. It is the main distinctive characteristic
from International hygienic guidelines and is based on EMF long-term exposure
adverse human health effect threshold definition and EMF cumulative effects concept.
The power frequency electric and magnetic field international reference levels, EU
Directive AL and Russian permissible levels are compared in the following sections.

3 Power Frequency Electric Field Permissible Levels

Permissible levels of power frequency electric field according to ICNIRP guidelines,

Directives EU and Russian hygienic norms for occupational and general public
exposure are presented in Table 1.

Table 1. 50 Hz electric field permissible levels.

Hygienic guideline Permissible level E, kV/m
Min Max
Occupational exposure
ICNIRP 2010 10
Directive 2013/35/EU 10 20
SanPiN 2.2.4.3359-16 5 25
General public exposure
ICNIRP 2010 5
Directive 1999/519/EC 5
SanPiN 2.1.2.2801-10, SN 2971-84 0.5 1–20*
*20 kV/m - Remote district

As mentioned above Directive EU [7] permits two external electric field thresholds
for occupational exposure, in particular, 10 and 20 kV/m as low and high AL
respectively. Low AL is equal ICNIRP reference level and are based on limiting the
internal electric field below the ELVs (1,1 V/m for health effects and 0,14 V/m for
sensory effects) and limiting spark discharges in the working environment [7]. Occu-
pational exposure levels below high AL (20 kV/m) means that the internal electric field
does not exceed the ELVs and annoying spark discharges are prevented, provided that
the special protection measures are taken [7].
134 N. Rubtsova et al.

As shown in Table 1, Russian norms permit 50 Hz electric field occupational

exposure levels in the range from 5 to 25 kV/m according to time dependency pre-
sented in Fig. 1.
Russian hygienic norms substantiation based on the long-term experimental studies
with the goal of adverse health effects threshold determination. This data (with hygienic
safety factor including) in collection with hygienic, epidemiological and peer-reviewed
publications information are substantiation of dose-dependent hygienic norms.
According to SanPiN 2.2.4.3359-16 [2] minimal (5 kV/m) permissible level of
electric field occupational exposure is permitted during all working day (8 h). The
maximum (25 kV/m) permissible level is permitted during short period up to 10 min
per day and other work time should be in general public condition. Under case of
electric field levels are from 5 kV/m to 20 kV/m permissible work time change. Power
frequency electric field occupational exposure hygienic norms are different according
to possible exposure time by equation T = 50/E−2 as principle of protection by time
(Fig. 1).

Fig. 1. 50 Hz electric field occupational exposure limits in dependence on work time.

If occupational exposure level is higher than 25 kV/m the worker should use
protective means, which ensure exposure level reduction below permissible levels.
Otherwise work in such conditions is not allowed.
It is possible to compare Russian and EU permissible levels by means of time
dependency shown in Fig. 1. For example, low AL (10 kV/m), permitted for all work
time per day in EU and only 3 h per day in Russia.
As presented in Table 1 50 Hz electric field permissible level for general public
exposure are equal in ICNIRP guidelines and Directives EU. At the same time they are
higher than Russian norms. In addition SanPiN 2.1.2.2801-10 [4] permits several
hygienic norms for general public in dependence on possible residence time and place.
Inside of residential, public and office buildings the power frequency electric field
 The Comparison of Approaches to Power Frequency EMF 135

exposure should not exceed the minimal permissible level (0.5 kV/m). The permissible
exposure level on a housing estate territory is 1 kV/m as well as at the border of 330–
1150 kV overhead transmission lines right-of-way (sanitary-protective zones).
The principal protection by time for general public is realized by taking into
account the different type of territory and possibility general public duration stay. In
addition to Table 1 data: permissible level in residential areas is 1 kV/m; in populated
areas outside of residential buildings permissible level is 5 kV/m; at intersections with
I-IV class roads - 10 kV/m and in unpopulated areas 15 kV/m.

4 Power Frequency Magnetic Field Permissible Levels

Permissible levels of power frequency magnetic field according to ICNIRP guidelines,

Directives EU and Russian hygienic norms for occupational and general public are
presented in Table 2.
Permissible levels for power frequency magnetic field occupational exposure are
differ for total (whole body) and local (limb) exposure.
Low AL (1 mT) is derived from the sensory effects ELV and high AL (6 mT) is
derived from the human health effects ELV for internal electric field related to electric
stimulation of peripheral and autonomous nerve tissues in head and trunk [7]. AL for
limbs exposure is derived from the health effects ELVs for internal electric field related
to electric stimulation of the tissues in limbs by taking into account that the magnetic
field is coupled more weakly to the limbs than to the whole body [7].

Table 2. 50 Hz magnetic induction permissible levels.

Hygienic guideline Permissible
level B, mT
Min Max
Occupational exposure
ICNIRP 2010 1
Directive 2013/35/EU
Whole body 1 6
Local to limb 18
SanPiN 2.2.4.3359-16
Whole body 0.1 2
Local to limb 1 8
General public exposure
ICNIRP 2010 0.2
Directive 1999/519/EC 0.1
HN 2.1.8/2.2.4.2262-07 0.005 0.02
136 N. Rubtsova et al.

Table 2 shows, that Russian norms of 50 Hz magnetic induction occupational

50 Hz exposure are differ for whole body exposure from 0.1 to 2 mT, for limbs
exposure from 1 to 8 mT and are significant less than Directive EU ALs. Russian
norms of magnetic induction exposure are dependence on work time per day and are
presented in Fig. 2 for whole body exposure as example.
According to SanPiN 2.2.4.3359-16 [2] whole body minimal (0.1 mT) permissible
level of magnetic induction occupational exposure is permitted for all working day
(8 h) and the maximal (2 mT) permissible level is permitted up to 1 h per day.
As shown in Fig. 2, low AL (1 mT), permitted for all work time per day in EU and
2 h per day only in Russia. High AL (6 mT) is not allowed for whole body occupa-
tional exposure in Russia, for limb exposure during 1.5 h per day only.
As presented in Table 2, 50 Hz magnetic field permissible level for general public
exposure are the highest in ICNIRP guidelines.
Russian norms permits different 50 Hz magnetic field levels for general public in
dependence on possible time of exposure. The minimal permissible limit value is 5 lT
in premises, children’s, preschool, school, general educational and medical institutions.

Fig. 2. 50 Hz magnetic induction occupational exposure (whole body) hygienic norms in

dependence on work time.

Hygienic norm is 10 lT for uninhabited premises of residential buildings, public

and office buildings, manned territories; 20 lT - for occupied district outside of zone of
housing estate, including overhead transmission lines and cable transmission line zone
above 1 kV; work of persons, professionally not connected with power objects
maintenance.
 The Comparison of Approaches to Power Frequency EMF 137

5 Conclusion

Comparison of current hygienic regulations of power frequency electric and magnetic

fields shows the differences between International and Russian electromagnetic safety
standards and norms. This is due to significant different principles in the basis hygienic
regulations. The main Russian permissible limit values are time dependent. This
approach provides possibility of work in more intensive EMF exposure, but during
short time period per work day. In some work cases the Russian hygienic norm
(25 kV/m) may be higher than ICNIRP (10 kV/m) and Directive EU (20 kV/m) levels
but are limited by working time (no more than 10 min per day). For general public in
Russia permissible limit values are graded to restrict inhabited places from high
intensity EMF source. In the case of prolonged exposure Russian norms are primary
stricter than others standards, because human protection from adverse cumulative EMF
effect is considered.
Hygienic Russian regulation is prognostic in terms of complicated 50 Hz electric
and magnetic filed human health effects possibility and evaluation of these exposure as
occupational and general public risk factor.

References
1. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, “Non-ionizing
radiation, Part 1: Static and extremely low-frequency (ELF) electric and magnetic fields”,
Lyon, vol. 80 (2002)
2. SanPiN 2.2.4.3359-16 Sanitary and epidemiological requirements for physical factors in the
workplace (2016). (in Russian)
3. HN 2.1.8/2.2.4.2262-07 Threshold permissible values of 50 Hz magnetic fields in residential
construction, living quarter and residential area (2007). (in Russian)
4. SanPiN 2.1.2.2801-10 Changes and addendums No. 1 to SanPiN 2.1.2.2645-10 Sanitary and
Epidemiological requirements to residential construction and living quarter residence
conditions, Moscow (2010). (in Russian)
5. SN 2971-84 Sanitary norms and rules of general public protection from electric field created
by dc power frequency overhead transmission lines (1984). (in Russian)
6. ICNIRP Guidelines for limiting exposure to time-varying electric and magnetic fields (1 Hz to
100 kHz). Health Phys. 99(6), 818–836 (2010)
7. Directive 2013/35/EU of 26 June 2013 on the minimum health and safety requirements
regarding the exposure of workers to the risks arising from physical agents (electromagnetic
fields) (2013)
8. Council Recommendation 1999/519/EC of 12 July 1999 on the limitation of exposure of the
general public to electromagnetic fields (0 Hz to 300 GHz) (1999)
9. Rubtsova, N., Paltsev, Yu., Perov, S., Bogacheva, E.: Dosing as intensity-time dependence
criterion in the EMF hygienic rating in Russia. Electromagn. Biol. Med. 37(1), 43–49 (2018)
 Two Test Methods Comparison for Power
Frequency Electric Field Shielding
Materials Evaluation

Sergey Perov1(&), Olga Belaya1, Tatyana Konshina1,

Elizaveta Tiutiunnik1, Balint Nemeth2, Gabor Göcsei2,
and Valentin Faradzhev3
1
FSBSI “Izmerov Research Institute of Occupational Health”,
31, pr. Budennogo, Moscow 105275, Russia
sergey.perov@gmail.com
2
Budapest University of Technology and Economics,
Mûegyetem rkp, 7-9, Budapest 1111, Hungary
3
Electrostatic Ltd., Alsó Zöldmáli u. 16, Budapest 1025, Hungary

Abstract. The personal protective equipment (conductive suit) should decrease

the power frequency electric field high levels up to occupational permissible
limit values for human safety. The personal protective equipment screening
property realized by conductive materials which has shielding coefficient. It is
relevant for practice to develop test method of conductive suit and materials
protection properties by direct power frequency electric field measurement. The
paper presents the comparison of results shielding materials testing by different
methods. There were compared data of direct power frequency electric field
measurement and standard test method according to IEC 60895. The suggested
shielding efficiency test method was based on electric field strength measure-
ments inside the metal holder covered by materials test sample placed in high
level power frequency electric field. 14 test samples of protective suit textile
with use two methods was compared. The results obtained by both methods
were analyzed by means of linear approximation function as a correlation
between suggested and standard test methods. This function defined minimal
shielding efficiency requirements for conductive material of power frequency
electric field personal protective equipment measured by suggested method. In
prospect, the new method will allow conductive suits manufacturers to predict
maximum possible power frequency electric field decrease for compliance with
permissible limit values at workplaces. Therefore, the suggested method of
shielding efficiency assessment is nearest to standardization in hygienic practice
and control at workplaces parameter.

Keywords: Power frequency electric field
Shielding efficiency
Test method

1 Introduction

Human health protection from the possible harmful electromagnetic field (EMF) effects
is one of the important issues of occupational safety. The workers in switchyards and in
the area of overhead transmission lines may be exposed by elevated permissible limit

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 138–144, 2020.
https://doi.org/10.1007/978-3-030-31676-1_14
 Two Test Methods Comparison for PF EF Shielding Materials Evaluation 139

values (PLV) power frequency EMF. Occupational safety is regulated by hygienic

standards. EU Directive [1] and ICNIRP Guidelines [2] set framework requirements for
workers power frequency (PF) electric field (EF) protection that are 10 kV/m in the
European Union. Russian PF EF hygienic standard [3] for all work day is 5 kV/m that
is stricter than in the European Union.
Modern personal protective equipment (PPE) is one of the possible way to workers
protection from PF EF high level exposure. Traditional PPE is the conductive suit,
which includes conductive clothes, facescreen, gloves and footwear. PPE screening
property bases on the principle of “Faraday cage” which realized by conductive
materials. PPE manufacture important step is its protective properties assessment as the
test of conductive material by shielding coefficient (К).
PF EF levels at work places are controlled and must be in compliance with hygienic
standards [1–3] for human health protection. Therefore, PF EF values decrease up to
level not higher than permissible should be important measure for worker protection. It
is known that PF EF elevated levels are human health risk factor as possible reason of
cardiovascular diseases [4].
PPE use goal is to decrease the PF EF levels up to PLV for worker safety ensuring.
Therefore, it is relevant to measure PF EF strength attenuation for shielding efficiency
assessment as conductive suit protection properties. PF EF measurements for PPE
shielding efficiency assessment approach has been used in some studies [5–10]. Live-
line conductive clothing properties with different types of face mesh were evaluated by
EF with use simulation and measurements [5]. PF EF strength measurements in lab-
oratory air capacitor shielding properties of conductive garments were tested [6]. There
were analyzed the possibility of PF EF decrease by Faraday cages in 400-kV substation
[7] and mesh shields under 400-kV overhead transmission lines [8] as well as PPE
jackets [9]. PF EF measurements at linemen face were used to compare PPE protective
properties of face screens [10].
PF EF evaluations were carried out for human body (face especially) [5, 7–10]
protection possibility in cases of the occupational limits exceeding [1, 2]. It is important
that PF EF values allow to direct estimate the protection properties by regarding
hygienic requirements. However, there is no standardized PPE test method based on
PF EF attenuation, and it is relevant for practice to develop it.
The background of PPE is the use of high quality protective materials of different
part of these means. Therefore, the first step of PPE shielding efficiency assessment
must be the testing of screening characteristic of shielding material. The shielding
properties of material define the potential protective efficiency of conductive suits made
from it.
The goal of study was the substantiation of conductive materials direct shielding
coefficient test method by means of electric field exposure evaluation.

2 Materials and Methods

Materials protective properties were evaluated by means of IEC standard test method
[11] and suggested direct test method.
140 S. Perov et al.

The suggested direct method is based on PF EF strength measurements inside the

metal holder covered by materials test sample placed in high level PF EF. The reference
PF EF strength is measured in the same case without material sample.
The suggested direct test method was realized by means of high voltage 50 Hz EF
test setup. The setup was open air capacitor, with plate sizes of 2.4  1.8 m and 0.7 m
distance between the plates. The high-voltage setup was the source of uniform PF EF in
the area of the holder location, materials and PF EF sensor. EF exposure levels were
from 7 to 60 kV/m.
PF EF levels were measured by a portable electric field analyzer EFA-300 (Narda
Safety Test Solution, Germany) with calibrated isotropic electric probe using a fiber
optic cable and EHP-50F with NBM-550 Controller and Display Unit (Narda Safety
Test Solution, Germany). The electric field probe was placed in the middle of the
holder and at 0.1 m high above the grounded plate. PF EF measurements were carried
out inside the holder without materials, and then with materials on the holder at the
same voltages at upper plate. The holder, materials as well as lower plate were
grounded.
The holder size was selected to ensure the required homogeneous PF EF at the
control points inside holder taking into account the measurement probe size, exposure
setup proportions and maximal possible conductive material sample size. The PF EF
distribution was estimated by simulation with various holder forms and dimensions.
Then the PF EF levels inside holder with selected geometry were measured without any
material sample.
Simulations were carried out using the software product SEMCAD X v.14.8
(«SPEAG AG», Switzerland) and high-performance computing systems NVIDIA Tesla
C2057 (NVIDIA Corporation, USA). The Finite-Difference Time-Domain method
(FDTD) was used for the calculation for the extremely low frequency (the equations
were transformed into frequency domain).
14 protective suit textile samples were tested by two methods: suggested and
standard. The samples size was matched with the holder size. Obtained results of
shielding efficiency assessment were analyzed and compared.
The shielding coefficient by proposed test method was estimated according to the
formula (1), where E1 is the results of PF EF measurements without materials and E2 –
with materials:

KE = 20IgðE1 /E2 Þ ð1Þ

where: KE = shielding coefficient, dB;

E1 = PF EF strength inside the holder without suit materials, kV/m;
E2 = PF EF strength inside the holder with suit materials, kV/m.
The material shielding coefficient KU was estimated by IEC standard test method
[11] and calculated by the formula (2):


KU = 20Ig Uref /U ð2Þ
 Two Test Methods Comparison for PF EF Shielding Materials Evaluation 141

where: KU = shielding coefficient, dB;

Uref = reference voltage between the line and earth connections without suit
materials, V;
U = voltage between the line and earth connections with suit materials, V.
The IEC method is based on IEC 60895:2004. The prescribed RMS voltage of the
inspection is 400 V, while the frequency of the sinusoidal waveform is 5 kHz. A well-
defined arrangement has to be set up to determine both reference and measured voltage
values. This high-frequency measurement is very sensitive to environmental noises,
properly shielded instruments and cables are essential.
The material electrical resistance was measured by microhmmeter CA4105 (Uman
plant “Megommetr”, Ukraine) and Digital multimeter APPA-97 (APPA Technology
Corporation, Taiwan. Measurement of KE, KU and material resistance were done 3
times. The experimental data are presented as mean ± SD by used Mathlab R2010a
(The MathWorks, USA) and Statistica 8.0 (StatSoft Inc., USA).

3 Results

There are presented the suggested method and comparison results for different mate-
rials tested by suggested method and standard test method according to IEC 60895-
2002 [11].
The developed models include setup (plane-parallel capacitor) with required PF EF
levels, shielding material and holder. Optimal holder shape and size (Fig. 1) selection
for developed method were chosen according to simulation data. The capacitor consists
of two metal plates, the size corresponds to the real test setup. The holder were placed
inside the capacitor and PF EF level was 40 kV/m in the work area, as shown in Fig. 1.

Fig. 1. PF EF value inside holder, PF EF sensor in the middle.

The chosen test holder was box shape with 0.9  0.9 m size and 0.2 m height. This
holder geometry was suitable for exposure setup dimensions, provided the required
field distribution and measurement probe placement inside holder.
142 S. Perov et al.

The test metal holder was made based on simulation results and used for next
experiments. Test sample set included conductive, low conductive and non-conductive
textiles. All samples shielding properties by both methods and its electric resistance
values measurement data are presented in Table 1. KU and KE are shielding coefficients
determined by IEC method and suggested method respectively.

Table 1. Results the measurement of materials resistance and shielding properties, mean ± SD.
Sample № R, Ohm КU, dB КE, dB
1 Non-conductive 9.93 ± 2.99 7.48 ± 1.23
2 Non-conductive 10.59 ± 2.60 7.85 ± 0.47
3 Non-conductive 2.94 ± 0.27 0.11 ± 0.09
4 0.48 ± 0.02 11.67 ± 0.47 61.83 ± 0.52
5 0.01 ± 0.00 41.78 ± 1.53 94.04 ± 1.88
6 0.06 ± 0.01 40.43 ± 0.28 88.30 ± 1.44
7 0.01 ± 0.00 15.86 ± 0.51 58.95 ± 0.17
8 0.37 ± 0.02 46.48 ± 0.23 98.96 ± 2.81
9 1.47 ± 0.09 30.43 ± 1.06 82.55 ± 0.57
10 0.30 ± 0.10 33.21 ± 0.91 84.28 ± 0.70
11 0.85 ± 0.12 65.84 ± 1.60 112.39 ± 7.14
12 6.04 ± 0.04 19.96 ± 0.36 69.76 ± 1.38
13 2.94 ± 0.42 63.13 ± 1.48 106.51 ± 6.90
14 0.93 ± 0.04 43.99 ± 0.42 97.06 ± 1.57

The Table 1 data show that test textile samples can be grouped in accordance with
their shielding coefficients. The non-conductive materials have the weakest shielding
efficiency (below 10 dB in KE and in KU). The material with low shielding efficiency
have 50–70 dB in KE and 11–20 dB in KU. The most of test samples are grouped by
shielding coefficient range from 80–100 dB in KE and 30–50 dB in KU. The highest
shielding efficiency is above 100 dB in KE and 60 dB in KU. Shielding efficiency of the
most conductive test samples are located above 50 dB in KE and 10 dB in KU.
Obtained data show sufficient discrepancies between the results received by dif-
ferent testing methods for the same sample. Possible correlation between results of
different methods analyses is presented in Fig. 2.
Scientific based criteria are necessary for any new methods main requirements
substantiation. The IEC requirement to material shielding efficiency is 40 dB. Linear
approximation function shown in Fig. 2 provides the simplest transition from KU and
KE values. The obtained line function is the rough approximation and suitable for
boundary shielding coefficient set for developed test method use. For example,
according to approximation the 40 dB by KU is equal to 80 dB by KE.
 Two Test Methods Comparison for PF EF Shielding Materials Evaluation 143

Fig. 2. KU and KE measurement and approximation results.

However, the data of some textile samples shielding properties are incomparable,
for example: KE above 80 dB and KU about 30 dB. Therefore, the approximation line
is not complete suitable for all test samples. Test samples group was very different by
textile resistance. Figure 2 show the appliance compatibility criteria for high conduc-
tive materials. It is possible that increase of test sample quantity will allow to find more
close approximation approach from KU to KE.
Our results conformed the possibility to apply suggested approach based on direct
textile evaluation of shielding efficiency by power frequency electric field attenuation
as adverse human health factor. This is important advantage to assess possible PF EF
level attenuation at workplace up to permissible levels or lower.
But, suggested method data are preliminary. Further research is needed for clari-
fication of both test methods uncertainty.

4 Conclusion

The data of investigation show the possibility protective suit materials efficiency
assessment by method differ from traditionally used by IEC. Suggested direct method
of PF EF assessment is nearest to standardization in hygienic practice and control at
workplaces parameter. The data analysis in comparison with standardized method has
defined minimum requirements for conductive materials shielding efficiency measured
by suggested method. In prospect, developed method will allow manufacturers PPE to
predict maximum possible PF EF attenuation for compliance with PLV at workplaces
ensuring from the beginning at the first step of new shielding material testing.
144 S. Perov et al.

References
1. Directive 2013/35/EU of 26 June 2013 on the minimum health and safety requirements
regarding the exposure of workers to the risks arising from physical agents (electromagnetic
fields) (2013)
2. ICNIRP Guidelines for limiting exposure to time-varying electric and magnetic fields (1 Hz
to 100 kHz). Health Phys. 99(6), 818–836 (2010)
3. SanPiN 2.2.4.3359-16 Sanitary-epidemiological requirements for physical factors in the
workplace (2016)
4. Rubtsova, N., Paltsev, Yu., Pokhodzey, L., Perov, S., Tokarskiy, A.: Main principles of
electromagnetic field occupational exposure risks management in Russia. Occup. Environ.
Med. 75, A420 (2018)
5. Göcsei, G., Berta, I.S., Németh, B.: Safety considerations regarding to the shielding of
electric fields during high voltage live-line maintenance. Acta Technica Jaurinensis 8(2),
153–164 (2015)
6. Neira, L., Pascual, H., Portillo, M., Pérez, F., Albanese, A., Fata, O., Franchini, R., Burna,
A., Stivanello, I.: A research on conductive clothing for life working. In: 11th International
Conference on Live Maintenance (ICOLIM), Budapest, Hungary (2014)
7. Pirkkalainen, H., Elovaara, J.A., Korpinen, L.: Decreasing the extremely low-frequency
electric field exposure with a Faraday cage during work tasks from a man hoist at a 400 kV
substation. Prog. Electromagn. Res. M 48, 55–66 (2016)
8. Pääkkönen, R., Korpinen, L., Tarao, H., Gobba, F.: Possibilities to decrease the electric field
exposure with a shield over worker under the 400 kV power lines. In: 2016 Progress in
Electromagnetic Research Symposium (PIERS), Shanghai (2016)
9. Korpinen, L., Pääkkönen, R.: Possibility to decreasing the 50 Hz electric field exposure with
different jackets. Bulgarian J. Public Health 7(2), 62–65 (2015)
10. Barbieri, L., De Maria, L., Chemelli, C., Gondola, M., Malgesini, R., Villa, A., De Donà, G.:
A comprehensive analysis of facial screens: sensitivity analysis and construction technolo-
gies. In: 12th International Conference on Live Maintenance (ICOLIM), Strasbourg, France
(2017)
11. IEC 60895-2002 Live working - Conductive clothing for use at nominal voltage up to
800 kV a.c. and ±600 kV d.c. (2002)
 Investigation of Electric and Magnetic Field
in the Application of Dynamic Line Rating

Levente Rácz(&), Dávid Szabó, and Gábor Göcsei

Budapest University of Technology and Economics, 18 Egry József Street,

Budapest 1111, Hungary
racz.levente@vet.bme.hu

Abstract. Dynamic Line Rating (hereinafter referred to as DLR) is an effective

grid management method for transmission system operators (TSOs). It provides
solution for the demand of the increasing transmission capacity and for cost-
effective network development at the same time. By the application of special
temperature sensors, weather stations and proper DLR models the ampacity of
the overhead lines (OHLs) could be increased significantly in the 95% of the
time. In the existing DLR models several aspects of the implementation are
taken into account and in addition to technological realization there are also
references which discuss the economic side of this method. Another important
question is that in case of DLR application what extent of a direct impact of the
increased current load has on the emerging magnetic field, and how affects the
electric field in the vicinity of the OHLs indirectly. Although this consequence
of DLR can be a risk to human life, it has not been investigated, yet. According
to this, researches on this issue can make the application of DLR safer, or
occasionally impose restrictions on the application of this method.

Keywords: Dynamic Line Rating
DLR
Overhead lines
Electric field

Magnetic field
Risk analysis

1 Introduction

For the power system operators is inevitable to utilize the existing grid with proper
safety and reliability level. In order to achieve this, each OHL has a maximum
allowable transfer capacity value based on the thermal equilibrium of the conductor.
The conventional way of calculating OHLs’ transmission capacity is the so-called
Static Line Rating (SLR). According to the SLR the ampacity of the conductor is
determined based on the worst-case scenario of the weather parameters’ combination
which increases security, however far from the optimum. With the spread of renewable
energy source based power generation units and the concept of the Integrated Elec-
tricity Market (IEM) increasing the ampacity of important OHLs is appreciated. One
cost-effective method is DLR, which in addition to reaching increased transfer capacity,
has many supplement benefits such as increased system stability, prevention of icing,
maintain the level of reliability etc. The essence of this method is to install sensors and
weather stations onto predetermined spans and observe the load and weather param-
eters in very frequent discrete periods such as in every 5–15 min. These data are sent to

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 145–153, 2020.
https://doi.org/10.1007/978-3-030-31676-1_15
146 L. Rácz et al.

a server wireless, and algorithms use them to calculate the ampacity value in real-time
and ahead. Based on the calculation the system operator makes a decision about the
transfer capacity value [1–3].
It is important to mention that there are cases where DLR cannot be configured on
the power line. Based on the current convention, for a particular OHL, the applicability
of DLR is determine based on which network element is the weak point of the system.
If the conductor is the weak point, which means that other parts such as current
transformers are able to carry more power without any malfunction than there is no
restriction for DLR methods [3].

2 Electric and Magnetic Field in the Vicinity of OHLs

The operation frequency of power transmission systems is 50 or 60 Hz, and in this

frequency range under normal operating conditions the effect of the electric and
magnetic field must be examined separately [4, 5].
Radiations below 100 kHz are called non-ionizing radiation, and the physiological
effects associated with these are investigated by WHO (World Health Organization)
and ICNIRP (International Commission on Non-Ionizing Radiation Protection) in
Europe. In case of non-ionizing radiation, the specified limit values may not be
exceeded even for a short time [4, 5].

2.1 Electric Field

The electric field around the OHLs is based on surface charge accumulation. As a result
of capacitive charge equalization an uncomfortable sensation may occur above a
10 kV/m field strength value. As a further consequence, low frequency electric fields
can cause direct nerve or muscle stimulation or even glare [4, 6].

Table 1. Reference levels for general public and occupational exposure to time-varying electric
field [4]
Exposure Frequency range E-field strengths [kV m−1]
General public 0.025–0.8 kHz 250/f
Occupational 0.025–0.82 kHz 500/f
Note: f as indicated in the frequency range column

According to Table 1, in case of 50 Hz the limit value is 5 kV/m for the general
public and 10 kV/m for occupational exposure.

2.2 Magnetic Field

At a frequency of 20 Hz, 5 mT is the lowest value where the magnetic field can cause
dazzle due to the stimulation of vision-related nerves. Low-frequency magnetic fields
can also affect sleep, mood and perception, but these mechanisms are not clarified, yet.
 Investigation of Electric and Magnetic Field in the Application of DLR 147

The relationship between the low-frequency magnetic field and cancer has long been a
subject of discussion, and many publications and studies have been conducted on this
subject. Although there are cases where a weak correlation can be detected, no study
has been carried out to dispel any doubts. On the other hand, IARC, the WHO Cancer
Research Organization categorized the low-frequency magnetic fields into the group
“2B - possibly carcinogenic to humans” [4, 7, 8].

Table 2. Reference levels for general public and occupational exposure to time-varying
magnetic field [4]
Exposure Frequency range B-field [µT]
General public 0.025–0.8 kHz 10/f
Occupational 0.025–0.82 kHz 50/f
Note: f as indicated in the frequency range
column

According to Table 2, in case of 50 Hz, the magnetic flux density limit is 200 µT
for the general public and 1000 µT for occupational exposure. However, research on
the effects of electric and magnetic fields is still ongoing, and the limit values may
change in the light of recent findings.

3 Effects of DLR on the Electric and Magnetic Field

The essence of DLR is to determine from time to time the transfer capacity of the OHLs
and in this way it increases the ampacity of the conductor in nearly 95% of the time.
The increase in current is closely related to the change in the magnetic field, which is
therefore worth examining when applying this method. DLR has no direct effect on the
electric field as the field strength does not change significantly in the vicinity of the
OHLs. However, with the increase of the sag the distance between the lower phase
conductor and the ground could be decreased which can cause problems in case of the
electric field. In the following some examples of potential threats are shown due to the
application of DLR.

3.1 Possible Dangers of Electric Field

All of the OHLs has a maximum temperature that cannot be exceeded in order to avoid
sag and clearance problems. The DLR models take note of this temperature limit, and
the algorithms are set not to exceed this value during the calculations. According to
this, electric field cannot cause problems while applying DLR. On the other hand, in a
real case this theoretical assumption is not always true. Increased sag can occur even if
the limit temperature is not exceeded according to the model, because every model has
some kind of error level. In addition, in the case of DLR forecast, the input parameters
148 L. Rácz et al.

are predicted weather parameters, which are also characterized by uncertainty. In

addition, the sensors and weather stations also has a measurement uncertainty. That is,
in an unfavorable case – even with the correct operation of the model – a difference of
up to 5–10° can occur between the discreetly measured wire temperature and the actual
hot spot depends on the length of the line and the applied sensors number.
Another important issue is the ageing of the conductor. With the aging of the
conductors, the sag of the power lines’ span may increase, which may be enhanced by
the application of DLR. In such a case, the DLR can cause an electric field problem, as
the permitted safety distance may decrease.
It also important to mention that there are special cases, when buildings, roads or
sports facilities are under the OHL. In these cases it is important that both calculations,
measurements and simulations have to be carried out to make sure that the general
public is not exposed to electric field hazards.
In case of occupational exposures there can be also critical points. Live working
(LW) on crossing OHLs which are under each other may be limited due to in case of
middle voltage LW there is no protection against the electric field. However, with the
change of high voltage OHLs’ sag, the circumstances could be changed.
Similar sag problems may occur in the last span when the conductors are directed
from the last high voltage tower to the transformer of the substation.

3.2 Possible Dangers of Magnetic Field

The magnetic field decreases reciprocal as a function of distance. For this reason, flux
density that is above the limit for general public may occur in the immediate vicinity of
the conductor. The temperature difference due to DLR models, forecasting systems,
and measurement error of sensors in very rare cases can lead to a situation where sag
growth can lead to a dangerous proximity of the magnetic field to the general public.
However, the situation is different during high voltage LW maintenance when the
trained person is working in the immediate vicinity of the conductor. In such cases,
further calculations and simulations are absolutely necessary in order to be able to
report with certainty that the use of DLR does not pose any additional health risk for
the worker.

4 Case Study

In order to find out the possible risk of electrical field and magnetic flux density case
studies were carried out for two independent high voltage power lines with different
locations. The simulations shown here could be a good basis for further calculations
and simulations for other locations with different geometric parameters.

4.1 110 kV Transmission Line

In the first case a 110 kV single circuit OHL located in Central Europe was investi-
gated. National laws regulated the clearance value to 7 m at this voltage level, but the
 Investigation of Electric and Magnetic Field in the Application of DLR 149

terrain conditions do not allow the keep this level at all the spans. One span with high
variation of ground level and elevation indicated to reduce the clearance limit to 5.5 m
in a short section. Therefore, the simulations were carried out for this span to find out
the possible dangers (Table 3).

Table 3. The main parameters of the 110 kV overhead-line

Parameter Value Parameter Value
Type of the conductor ACSR 240/40 mm2 Resistance at 25 °C 0.120 Ω/km
Maximum temperature 40 °C Resistance at 75 °C 0.144 Ω/km
Outer diameter 21.9 mm Average mass per unit lengths 0.987 kg/m
Outer strand diameter 3.45 mm Clearance 7 m or 5.5 m
Summer SLR 530 A Winter SLR 645 A

Figure 1 shows the cross-sectional electric field distribution under the line in case
of 5.5 m clearance from the ground. It can be seen, that the electric field strength
reaches 5 kV/m (marked with red) at 3 m height from the ground, which means that the
exposure was determined below the safety level for general public.

Fig. 1. Electrical field distribution under a 110 kV transmission line

As the major aim of a DLR system is to increase the transmission capacity of

OHLs, the magnetic flux density was examined as an effect of surplus current, which
could cause 40% higher value compared to the static rating of the line.
150 L. Rácz et al.

Fig. 2. Distribution of magnetic flux density under a 110 kV transmission line

Figure 2 shows that the increased load level in this special case does not increase
the exposure risk for general public.

4.2 220 kV Transmission Line

Secondly, a 220 kV double circuit OHL located in Southern Europe was inspected in
order to get a more comprehensive picture of the field strengths in general. National
regulations also determine the ground clearance level to 7 m, despite of the difference
nominal voltage level of the OHL. Nevertheless, the construction documentation of the
transmission line consists simulations for sag, where the lowest clearance value is
10.7 m (Table 4).

Table 4. The main parameters of the 220 kV overhead-line

Parameter Value Parameter Value
Type of the conductor ACSR 400/55 mm2 Resistance at 50 °C 0.0797 Ω/km
Maximum temperature 85°C Resistance at 85°C 0.0902 Ω/km
Outer diameter 27.7 mm Mass per unit lengths 1.458 kg/m
Outer strand diameter 3.08 mm Clearance 7m
Summer SLR 905 A Winter SLR 1066 A

A recent study showed that the conductor temperature variation can reach 10 °C in
case of a medium length line [9, 10]. Therefore, conductor sag simulation was carried
out, which showed that the clearance decreased by 15 cm as the effect of 10 °C
increase in conductor temperature. Thus, the field simulations were carried out for this
reduced clearance value. The electric field displayed in the simulation does not refer to
a single period just a snapshot for the chosen worst case scenario where the critical
distance is between the lower phases and the ground.
 Investigation of Electric and Magnetic Field in the Application of DLR 151

Fig. 3. Electrical field distribution under a 220 kV transmission line

Figure 3 shows that the electric field strength reaches 5 kV/m at about 6 m height
from the ground level. Therefore, the increased sag caused by DLR method does not
raise the risk level of electric field for public exposure.

Fig. 4. Distribution of magnetic flux density under a 220 kV transmission line

Figure 4 illustrates how the magnetic field distribution varying around the line,
when the possible increasing effect of sag due to conductor temperature uncertainty
was take into consideration, while the load was considered 40% higher than the static
rating of the line. It can be stated, that in his special case and geometry nor the
increased transmission capacity neither the increased sag raised the risk factor occur by
magnetic flux density.
152 L. Rácz et al.

4.3 Future Plans

Although, based on the results of the case studies presented, the risk factor in the
selected critical spans of these two OHLs is not increased by the current load increase
and the potential sag increase, further simulations and calculations are required to
provide a more accurate picture of the changes in risk factors, especially in case of
occupational exposures during LW.
It is important to notice, that the drawings showing the terrain conditions do not
always include the buildings, so they were not taken into account in the simulations.
However, there were spans during the visual inspection, where the safety clearance
seemed doubtful due to buildings, so in the future such layouts should be considered in
the simulations and calculations.

5 Conclusion

DLR is a cost-effective grid management method that can significantly increase the
transmission capacity of transmission lines by calculating the ampacity in real-time.
There are number of references to technical and economic issues related to this method,
as well as to the sensor application and forecasting methods, but none of them
undermine the impact of possible changes in electric and magnetic fields. This article
discusses the basics of the change of these low frequency fields as they can have a
significant impact on the applicability of the method.
This paper lists some cases where the magnetic field due to the effect of the current
increase and the electric field due to sag growth according to various model uncer-
tainties may cause a problem. In order to show the possible effects, simulations were
performed for two transmission lines, for both magnetic and electric field distribution in
the vicinity of the OHLs. According to the simulations, in the selected spans of the
OHLs nor the increased magnetic neither the electric field due to sag caused significant
risk increase. On the other hand, further simulations, calculations and local measure-
ments are required in order to state the safety of DLR method with higher certainty.

Acknowledgement. This work has been developed in the High Voltage Laboratory of Budapest
University of Technology and Economics within the boundaries of FLEXITRANSTORE project,
which is an international project. FLEXITRANSTORE (An Integrated Platform for Increased
FLEXIbility in smart TRANSmission grids with STORage Entities and large penetration of
Renewable Energy Sources) aims to contribute to the evolution towards a pan-European trans-
mission network with high flexibility and high interconnection levels.
 Investigation of Electric and Magnetic Field in the Application of DLR 153

References
1. McCall, J.C., Servatius, B.: Enhanced economic and operational advantages of next
generation dynamic line rating systems, Paris (2016)
2. Electric Power Research Institute: Evaluation of Instrumentation and Dynamic Thermal
Ratings for Overhead Lines (2013)
3. Rácz, L., Szabó, D., Németh, B., Göcsei, G.: Grid management technology for the
integration of renewable energy sources into the transmission system. In: 7th International
Conference on Renewable Energy Research and Applications, ICRERA 2018, Paris, France
(2018)
4. International Commission on Non-Ionizing Radiation Protection: ICNIRP guidelines for
limiting exposure to time-varying electric and magnetic fields (1 Hz–100 kHz). Health Phys.
99(6), 818–836 (2010)
5. WHO IARC: IARC monographs on the evaluation of carcinogenic risks to humans. Non-
Ionizing Radiation, Part 1: Static and Extremely Low-Frequency (ELF) Electric and
Magnetic Fields, vol. 80. IARCPress, Lyon, France (2002)
6. Reilly, J.: Applied Bioelectricity: From Electrical Stimulation to Electropathology. Springer,
New York (1998)
7. World Health Organization: Environmental Health Criteria 238 Extremely low frequency
(ELF) fields. WHO, Geneva, Switzerland (2007)
8. Xi, W., Stuchly, M.: High spatial resolution analysis of electric currents induced in men by
ELF magnetic fields. Appl. Comput. Electromagnet. Soc. 9, 127–134 (1994)
9. Rácz, L., Szabó, D., Göcsei, G., Németh, B.: Application of Monte Carlo methods in
probability based dynamic line rating models. In: 10th Advanced Doctoral Conference on
Computing, Electrical and Industrial Systems, DoCEIS 2019, Caparica, Portugal, pp. 11–
124. Springer (2019)
10. Szabó, D., Rácz, L., Göcsei, G., Németh, B.: A novel approach of critical span analysis. In:
International Symposium on High Voltage Engineering, ISH 2019, Budapest, Hungary
(2019)
High Voltage Systems and Smart
Technologies
 Development and Characterization
of a Transient Overvoltage Recorder
in Distribution Networks

L. Perdomo1,2(&), A. Alfonso1,2, F. Santamaria1, and F. J. Román2

1
Universidad Distrital Francisco José de Caldas, Cra. 7 # 40B-53,
Bogotá, Colombia
leperdomoo@correo.udistrital.edu.co
2
Universidad Nacional de Colombia, Ciudad Universitaria, Bogotá, Colombia

Abstract. Transient overvoltages in distribution networks can be produced by

direct and indirect lightning strikes, due to energization or de-energization of
distribution networks, reactive compensation banks, distribution transformers, or
transfer of transients from higher voltage levels through the capacitive coupling
of distribution transformers. Therefore, the characterization of this type of
transient overvoltages aims to adjust the design criteria of devices and equip-
ment in the electricity sector, such as the level of immunity to galvanic or
radiated coupling signals.
In order to characterize these transitory phenomena, a new technological
development is proposed, based on the appropriation of knowledge, which
allows the generation of alternative recording and measurement techniques that
help to propose solutions to problems in the equipment of the power system. In
this regard, this article describes the development, evaluation, and characteri-
zation of a prototype for a transient overvoltage recorder in distribution net-
works. Initially, the design criteria and their implementation in printed circuit
boards necessary to ensure the treatment and processing of the registered signals
are presented. Subsequently, the frequency response of the compensated voltage
divider, the time domain response of the galvanic protection, and the conversion
and processing stage of the transient signal are analyzed; the firmware structure
implemented in the processing unit, the prototype latency in offline mode, and
the storage in SD card are also evaluated. Finally, the results of the test for the
prototype under different overvoltages and operating conditions are presented.

Keywords: Printed circuit board
Transient overvoltage recorder

Electromagnetic compatibility

1 Introduction

Transient overvoltages in distribution networks resulting from the direct and indirect
impact of lightning discharges reduce the useful life of the interconnected devices [1–
6]. The costs associated with damage to distribution transformers resulting from
transient overvoltages due to lightning reach sums in the order of 302,665.70 USD per
year [7, 8].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 157–170, 2020.
https://doi.org/10.1007/978-3-030-31676-1_16
158 L. Perdomo et al.

For this reason, a prototype for a transient overvoltage recorder was developed to
characterize the dielectric stresses to which a device interconnected to the electrical
system is subjected due to transient disturbances. The main features of this prototype
are local storage of records, transfer of information from multiple communication
channels, modularity in development for future extensions, maintenance, and
improvements.
The development of this study is a contribution to research on electrical distur-
bances in distribution networks as well as the development of prototypes for their
characterization.

2 Methodology

The development of the prototype for transient overvoltage recorder, which includes
the selection of electronic components, was based on the characterization of over-
voltages in distribution networks resulting from direct and indirect impact of lightning
[9–11]. This article is based on the application of electromagnetic compatibility criteria
(EMC) [12] for the design of printed circuit boards (PCB) [13–22], with the aim of
preserving the signal in the different modules of the prototype: signal conditioning, data
processing, and communications.
The characterization of each of the constituent modules of the prototype for the
transient overvoltage recorder is described below. Finally, the integral evaluation of the
prototype is presented in the Results section. IEC standards [23–26] have been con-
sulted for prototype evaluation of transient overvoltages however, some test charac-
teristics have been modified in order to enforce the prototype sampling rate condition.

2.1 Compatibility Criteria for Printed Circuit Board Design

In order to ensure that the information recorded by the prototype and the subsequent
transfer through the cellular network to the server is preserved, the signal integrity
analysis is performed on specific parts of the printed circuit boards according to the
following methodology:
To analyze the electrical behavior of a microstrip, to develop the printed circuit
board by applying electromagnetic compatibility criteria (EMC), and to implement
IBIS models in the integrated technology.
Figure 1 shows the variation of the characteristic impedance (Zo) of the microstrip
with respect to the relation of trace width (w) and the substrate thickness (h). Four
variations in the value of the relative permittivity of the substrate were considered. The
minimum thickness of the boards and the substrate material to be used must be defined
with the PCB manufacturer. For instance, the substrate thickness for designed FR-4
composite material boards is 0.32 mm; their relative permittivity is 4.4. If it is nec-
essary to reduce the characteristic impedance of the trace while preserving the physical
dimensions of the substrate, the trace width must be increased; this is useful for trace
tuning.
 Development and Characterization of a Transient Overvoltage Recorder 159

Fig. 1. Determination of the characteristic impedance of the microstrip in relation to the

variation of the physical dimensions of the trace width and the substrate thickness.

2.2 Signal Conditioning Board

Figure 2 shows the final layout of the printed circuit board. In order to avoid electrical
disruptions between the traces carrying the transient overvoltage signals with high
voltage levels and the reference of the electronic device, the ground plane in this area
was eliminated.

Fig. 2. Printed circuit board (signal acquisition) Top: Altium Designer 3D visualization
software. Bottom: prototyping.

The proper processing of the transient signal was verified from tests performed on
the data conditioning board. These tests are: the frequency domain response of the
160 L. Perdomo et al.

compensated divider, the transient signal magnitude cut from TVS diodes, the signal
recording in the low-voltage branch of the divider, and the output of the second stage of
signal decoupling.
The signal conditioning board was equipped with a low-voltage compensated
divider with an ideal transformation ratio of 323.62. This printed circuit board was used
for the interconnection with an external compensated divider whose transformation
ratio ranged between 13.2 kV and 1 kV from the 1 kV BNC connector.
Figure 3 shows the frequency response of the signal conditioning board. The
transformation ratio of the compensated divider ranged between 333.64 and 329.18 for
frequencies between DC and 50 kHz. The maximum error reached for the transfor-
mation ratio of the compensated divider during measurements was 3.11% of the cal-
culated value.

Fig. 3. Transformer ratio of the compensated divider.

Figures 4 and 5 show the operation of the TVS diodes used as protection against
transient overvoltages, which may affect the electronic design implemented from the
low-voltage branch of the divider. The evaluation and the response in the time domain
of the TVS diodes were obtained from the application of test signals whose charac-
teristics closely exemplify the signals present in the power system during stable state
and when transient voltage signals occur.
Signals whose magnitude was affected due to the operation of the TVS diodes reach
a peak magnitude of 4.48 V (positive cycle) and 4.32 V (negative cycle) (see Fig. 4),
and 4.8 V (positive cycle) and 4.32 V (negative cycle) (see Fig. 5).
 Development and Characterization of a Transient Overvoltage Recorder 161

Fig. 4. Operation of the TVS diodes. Evaluation frequency: 41.98 Hz.

Fig. 5. Operation of the TVS diodes. Evaluation frequency: 41.98 Hz.

Figure 6 shows the signal record at the output of the conditioning board when a test
signal was applied to the low-voltage branch of the divider. The characteristics of the
test signal are 1.45 µs pulse width measured between times 13.27 µs and 14.72 µs, and
376 mV peak voltage; times between 10% and 90% in relation to the peak magnitude
of the signal are 90 ns and 230 ns, respectively.
162 L. Perdomo et al.

Fig. 6. Pulse register in the low-voltage branch of the divider and at the output of the
conditioning board.

2.3 Data Processing Board (Microcontroller Unit - MCU)

The development of the embedded system associated with the data processing board
had the ADC converters configured at an 8-bit and interleave mode; its sampling and
conversion time were 2.5 and 8.5 cycles, respectively. The clock frequency associated
with the ADC peripherals was 80 MHz through a PLL, which serves as a phase
follower for generating multiple frequencies and sampling rate of 7.2 MSPS.
The RAM has a maximum capacity of 65,500 records of 32 bits each when
interacting with the DMA. In addition, the RAM was configured to store up to 500
records, which correspond to a sampling time of about 50 µs. The DMA has the
characteristic of capturing information from the RAM for special records associated
with the ADC1-2 peripherals in word size (32 bits) and subsequent transfer to the
general RAM in byte size (8 bits). This methodology helps to free the CPU space of the
microcontroller from the process of information transfer to the general RAM; this
decreases the storage time since it does not have to read the lines of code in the CPU.
Figure 7 shows the data processing board (MCU). In the upper part, there is the
arrangement of the elements associated with the power supply unit (DC/DC converter,
frequency: 180 kHz, voltage supply range: 8–36 V current supply range: 0–5 A,
maximum power: 75 W), data processing unit (MCU) (512 kilobytes of internal
memory, 128 kilobytes of RAM, maximum CPU frequency: 80 MHz, voltage: 1.71–
3.6 V, 3 USART operating 115,000 BSPS, SD/MMC proprietary protocol for com-
munication with SD card), SD data storage unit configured at 1 MHz, wired data
transfer unit, and DC comparison reference.
 Development and Characterization of a Transient Overvoltage Recorder 163

Fig. 7. Data processing board (MCU). Top: Altium Designer 3D visualization software.
Bottom: prototype.

The transient signal acquired by the transient overvoltage recorder, whose sampling
rate was 7.2 MSPS, was compared to that recorded by the oscilloscope whose main
characteristics are: 100 MHz bandwidth and 1 GSPS sampling rate. Different signal
types were selected, varying the parameters of rise time, fall time, peak magnitude, and
ramp-up. With the rise and tail times selected during these tests, it was guaranteed that
the prototype would be capable of recording signals with slower rise and fall rates.
164 L. Perdomo et al.

Figure 8 shows a transient overvoltage whose technical characteristics are: rise

time: 250.9 ns, fall time: 1.1 µs, ramp-up: 9.13 V/µs, and peak voltage: 3.22 V. The
reproduction of the signal was adequate; however, the main problem in the digital-
ization of the signal was due to the sampling rate of 7.2 MSPS compared to that of the
oscilloscope, whose rate was 1 GSPS. The maximum error reached for the reproduction
of this signal was 9.2% in the voltage samples during the start of digitization.

Fig. 8. First record of transient overvoltage.

Figure 9 shows a transient overvoltage whose technical characteristics are: rise

time: 826.98 ns, tail time: 1.22 µs, ramp-up: 3.28 V/µs, and peak voltage: 3.12 V. The
maximum error reached was 4.8% in the voltage samples during the start of the signal
digitization. The discretization of the signal was adequate during the transient phe-
nomenon; however, the error was due to the resolution of the ADC, which is close to
12.9 mV per record.

Fig. 9. Second record of transient overvoltage.

 Development and Characterization of a Transient Overvoltage Recorder 165

Figure 10 shows a transient overvoltage whose technical characteristics are: rise

time: 181.47 ns, tail time: 1.06 µs, ramp-up: 14.42 V/µs and peak voltage: 2.88 V. The
maximum error reached was 5.2% in the voltage samples during the start of the signal
digitization.

Fig. 10. Third record of transient overvoltage.

2.4 Communications Board

The prototype for the transient overvoltage recorder was equipped with wireless
communication, which provides redundancy in communication channels for data
transfer. The communication module designed and implemented has additional benefits
such as the acquisition of georeferencing frames and time stamps, which are added to
the information on transient overvoltage for further data processing. The main technical
characteristics of this communication module whose design was based on the
EC20 QUECTEL communication module are: 1.4/3/5/10/15/20 MHz bandwidth;
3.3 V–4.3 V supply voltage; LCC encapsulation; control via AT, GNSS GPS/
GLONASS commands; supported protocols: TCP, UDP, PPP, FTP, HTTP, SMTP,
MMS, NTP, PING, DTMF, FILE, CMUX, and QMI. The power supply was the one
recommended by the manufacturer MIC29302WU; its technical characteristics are: 5 V
of power supply, 3.8 V regulated at 3 A. Figure 11 shows the wireless data transfer
board. The 3D design created on Altium Designer is on the left and the prototype is on
the right.
166 L. Perdomo et al.

Fig. 11. Data transfer board. Left: design on Altium Designer software. Right: prototype.

One of the timer modules of the microcontroller was configured to identify the
capture time of a transient overvoltage in offline mode. This was activated in counting
mode from the moment the overvoltage occurred until the storage in the SD card was
completed. For testing purposes, this value was sent via the serial port to a terminal.
The value did not exceed 500 ms.
 Development and Characterization of a Transient Overvoltage Recorder 167

3 Results

The evidence presented below shows the integral behavior of the prototype for the
recorder when faced with the incidence of transient signals controlled by voltage
generators. The tests have two stages: the first one aims to evaluate the storage in
offline mode of the device, and the second one aims to evaluate the information transfer
towards the server.

3.1 Storage of Information in Offline Mode

The results obtained are presented below. Figure 12 shows the records of transient
overvoltage signals obtained at different points of the prototype.

Fig. 12. Record of transient overvoltage. Test 1.

The incident signal in the high-voltage branch has the following technical char-
acteristics: rise rate: 422.7 mV/ns, and maximum voltage reached: 110 V. The records
of the transient signals were multiplied by the transformation ratio of the compensated
divider. The maximum error reached was close to 5.24%, between the signal recorded
in the low-voltage branch of the compensated divider and the record stored in the SD
card. The discretization of the incident signal was not adequately reproduced by the
MCU; this was the case of the transients present during the rise and fall of the signal,
due to the sampling rate and conversion range of the ADC unit.

3.2 Storage of Information on the Server

Figure 13 shows the record of a transient signal in the low-voltage branch of the
compensated divider and the record obtained in the server, from the wireless transfer via
GPRS. The technical characteristics of the transient signal are: rise rate: 98.27 V/µs, and
peak voltage: 98.98 V. The reproduction of the transient signal was adequate, because
168 L. Perdomo et al.

the rise rate was slower than those used for the characterization of the prototype. The
maximum error was 2.83%.

Fig. 13. Record of transient overvoltage. Test 2.

Evidence of the storage of the record was based on the interaction of the software
for visualizing transient overvoltage events, as shown in Fig. 14.

Fig. 14. Record of transient overvoltage on the database.

4 Conclusions

A prototype for the acquisition of the fundamental parameters of transient overvoltages

present in distribution networks was developed.
As a result of the laboratory tests, it was possible to guarantee the recording of
transient overvoltages in offline mode in less than 0.5 [s]. The time is determined from
 Development and Characterization of a Transient Overvoltage Recorder 169

the activation of the interruption of the overvoltage comparator, which initializes the
timer and ends with the storage of information in the SD card and file closing.
The digitization of transient signals was guaranteed with a ramp-up of 7.9 kV/µs,
for a transformer ratio of 333.64 (compensated divider of 1 kV/3.3 V). As an example,
with the previous tests it is possible to determine a signal of 5 kV/µs by multiplying the
ratio of the compensated divider and the ramp-up stored in the MCU board, which are
333.64 and 14.42 V/µs, respectively. This is because the ADC converter operates in
the range 0–3.3 V with an 8-bit resolution and the signal recovery is performed on the
written software but not on the embedded software.

Acknowledgments. This research is promoted by the National Fund for Funding of Science,
Technology and Innovation “Fondo Francisco José de Caldas” of the Administrative Department
of Science, Technology and Innovation - COLCIENCIAS (Contract: FP44842-321 2015).

References
1. Chowdhuri, P.: Electromagnetic Transients in Power Systems. Research Studies Press Ltd.,
New York (1996)
2. Greenwood, A.: Electrical Transients in Power Systems. Wiley, New York (1991)
3. Martínez, J.: Coordinación de aislamiento en redes eléctricas de alta tensión, Primera edn.
McGraw - Hill, Madrid (2008)
4. Martínez Velasco, J.A.: Power System Transients: Parameter Determination. CRC Press,
New York (2009)
5. Hileman, A.R.: Insulation Coordination for Power Systems. CRC Press, New York (1999)
6. Keri, A.J.F., Musa, Y.I., Halladay, J.A.: Insulation coordination for delta connected
transformers. IEEE Trans. Power Deliv. 9(2), 772–780 (1994)
7. Rojas Benitez, L.F.: Sistema electrónico de facturación de energía vía telefónica. Col Patent
G6F 17/60 (2001)
8. Roman Campos, F.: Análisis de fallas en transformadores causadas por la operación de
pararrayos ante sobretensiones externas. Ing. e Investig, pp. 34–46 (1987)
9. Perdomo, L., Alfonso, A., Santamaria, F., Roman, F.: Sensitivity study of induced transient
overvoltages on distribution networks. In: ISH 2017 - 20th International Symposium on
High Voltage Engineering (2017)
10. Perdomo, L., Santamaria, F., Roman, F.: Transient overvoltage in distribution networks in
rural areas. In: 2016 33rd International Conference on Lightning Protection (ICLP), pp. 1–6
(2016)
11. Alfonso, A., Perdomo, L., Santamaria, F., Gómez, C.: Transient surges analysis in low
voltage networks. Rev. Tecnura 18(Edición especial doctorado), 41–50 (2014)
12. Alfonso, A., Perdomo, L., Santamaria, F., Amortegui, F., Roman, F.: Integrity analysis of the
signal transmitted to a 4G LTE module. In: 2017 European Conference on Circuit Theory
and Design (ECCTD), pp. 1–4 (2017)
13. O’Hara, M.: EMC at Component and PCB Level. Newnes, Oxford (1998)
14. Montrose, M.I.: EMC and the Printed Circuit Board: Design, Theory, and Layout Made
Simple. Wiley, New York (2004)
15. Montrose, M.I.: Printed Circuit Board Design Techniques for EMC Compliance: A
Handbook for Designers, 2nd edn. Wiley, New York (2000)
16. Ott, H.W.: Electromagnetic Compatibility Engineering. Wiley, New Jersey (2011)
170 L. Perdomo et al.

17. Morgan, D.: A Handbook for EMC Testing and Measurement. The Institution of
Engineering and Technology, London (1994)
18. Ludwig, R., Bogdanov, G.: RF Circuit Design: Theory and Applications. Prentice Hall,
London, New York (2009)
19. Teppati, V., Ferrero, A., Sayed, M. (eds.): Modern RF and Microwave Measurement
Techniques. Cambridge University Press, Cambridge (2013)
20. Bowick, C.: RF Circuit Design. Newnes, Indianapolis (1982)
21. Gonschorek, K.-H., Vick, R.: Electromagnetic Compatibility for Device Design and System
Integration. Springer, Heidelberg (2009)
22. Redouté, J.-M., Steyaert, M.: EMC of Analog Integrated Circuits. Springer, Dordrecht
(2009)
23. International Electrotechnical Commission. IEC 61000-4-5: Testing and measurement
techniques. Surge immunity test. In: Electromagnetic Compatibility, p. 50 (2005)
24. International Electrotechnical Commission. IEC 61000-4-4 Testing and measurement
techniques Electrical fast transient/burst immunity test. In: Electromagnetic Compatibility,
Ginebra, p. 36 (2004)
25. International Electrotechnical Commission. IEC 61000-4-25: Testing and measurement
techniques HEMP immunity test methods for equipment and systems. In: Electromagnetic
Compatibility, Ginebra, p. 48 (2001)
26. IEC. Insulation co-ordination - Part 1: Definitions, principles and rules, IEC 60071-1, 8th
edn. International Electrotechnical Commission (2006)
 Condition Evaluation of High Voltage
Transmission Line in Thailand

Waraporn Luejai, Thanapong Suwanasri(&),

and Cattareeya Suwanasri

King Mongkut’s University of Technology North Bangkok,

1518 Pracharat 1 Rd., Wongsawang, Bangsue, Bangkok 10800, Thailand
thanapong.s.epe@tggs-bangkok.org

Abstract. This paper presents a novel approach for practical condition evalu-
ation of HV overhead transmission lines in Thailand. The transmission com-
ponents includes conductor, conductor accessories, insulator, steel structure,
foundation, lightning protection system, tower accessories and right of way.
Significant and scientific laboratory tests as well as criteria to evaluate the
condition are proposed such as special tests e.g. tensile strength, torsional
ductility, and loss of zinc test of conductor, thermography, visual inspection all
components, and etc. The weighting and scoring method is applied to calculate
the health index of transmission line. According to standards and experts’
experiences, the condition evaluation criteria are scored and ranged into six
levels from very good condition till end of life while importance weights of
components and testing methods are determined by applying Analytical Hier-
archy Process. Then, the condition in form of health index and maintenance task
are classified into three levels as normal condition with scheduled maintenance,
moderate condition with increasing of scheduled maintenance as well as med-
ium term planning for renovation or replacement and poor condition that needs
urgent replacement of defective equipment or replacement planning by new
transmission line. The on-line web-application was developed for a simple,
effective and real-time usage to all users. Twenty of 115, 230 and 500 kV
transmission lines with the practical testing data were evaluated. The results
show an effectiveness of the health index determining procedure to prioritize and
manage maintenance tasks accordant to the exact condition of transmission lines
to reduce maintenance cost and maintain better system reliability.

Keywords: Condition evaluation
Health index calculation
Weighting and

scoring method
Asset management
Transmission network maintenance

1 Introduction

Nowadays, the economic problem has forced electric utility to provide a better quality
of electricity under lower operating and maintenance cost and low capital investment.
Moreover, some HV overhead transmission lines (OHLs) in Thailand are aged and
gradually degraded, a deterioration, abnormal condition and failure in OHL’s compo-
nents such as broken conductor, flashover on insulator, corrosion of steel structure, etc.
keep increasing, which lead to significant amount of renovation/replacement work. The

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 171–183, 2020.
https://doi.org/10.1007/978-3-030-31676-1_17
172 W. Luejai et al.

aged HV overhead lines are very significant in power transmission system because
aging equipment may cause unexpected failure with huge interruption while decreasing
system operating performance and reliability [1, 2]. Then, the effective asset man-
agement and planning for condition evaluation and management is of prime importance
to gain maximum benefit, especially for the OHL in a competitive market [3]. In [4, 5],
the condition assessment of different components for OHL was introduced while
various testings, techniques and criteria were suggested. In [5], the Failure Mode Effect
Criticality Analysis (FMECA) approach was applied to determine the health index;
however, this method requires historical failure data, which any utility may encounter
difficulty in historical data collection. Ranking aging OHLs in replacement, refur-
bishment, and renovation task is challenged; then risk to failure and reliability were
evaluated in [6, 7] while the cost analysis was assessed the equipment to maintain
reliability [8]. However, those mentioned condition assessment techniques are mainly
visual inspection.
In this paper, the comprehensive procedure to calculate the condition overhead
OHL (OHL) is proposed. The significant and scientific laboratory test to determine the
aging of conductor as tensile test and torsional ductility together with visual inspection
are incorporated. The testing criteria to evaluate the condition are differentiated. The
weighting and scoring method is applied to calculate the condition of OHL in term of
health index. By choosing the worst score of individual component in overall OHL
spans instead of weight average technique, this novel approach is applied determine the
worst condition of OHL because the traditional weight average technique could hide
the worst score of any span in the same line route due to averaging with many good
condition spans. Finally, the maintenance planning could be effectively ranked
accordant to the health index.

2 HV Transmission Line Evaluation

2.1 Transmission Line Component, Sub-component and Its Testing

The HV OHL is classified into eight components consisting of conductor, conductor
accessories, insulator, steel structure, foundation, lighting protection system, tower
accessories, and right of way as given in Table 1. The sub-components of each
component are also given. The inspection and special tests for assessing the condition
of HV transmission sub-components in eight different groups are needed to evaluate
sub-components, whose degradation could be detected by different test methods as
shown in Table 2. The WSM are applied for evaluation while AHP method is used to
determine the weight of each criterion. The score is used to classify sub-components’
condition as divided into six levels as 0 (very good), 1 (good), 2 (satisfy), 3 (moderate),
4 (degradation) and 5 (reaching end of life) as shown in Table 3. Table 4 shows score
classification for loss of zinc of ACSR conductor. Detail of thermography inspection of
compression joint as sub-component of conductor accessories is shown in Table 5.
 Condition Evaluation of High Voltage Transmission Line in Thailand 173

Table 1. Components of OHL evaluation.

Group Components Sub-components
1 Conductor Conductor
2 Cond. accessories Spacer, Damper, Joint, Dead end, PG clamp
3 Insulator Insulator, Fittings, Arrester
4 Steel Structure Structure, Anchor and Guy
5 Foundation Concrete foundation, Grillage foundation, Stub
6 Lightning protection OHGW/OPGW, OHGW fitting, Marker, Grounding
7 Tower accessories Danger sign, Tower number sign, Phase plate
8 Right of way Right of way

Table 2. Testing methods for sub-component of transmission tower and its span.
Group Sub-components Testing method
1 Conductor Visual inspection, Loss of zinc, Tensile strength, Torsional
ductility
2 Damper Visual inspection
Spacer Visual inspection, Special test
Dead end Resistance, Thermography
Joint Resistance
3 Insulator Visual inspection, Thermography
Fittings Visual inspection
Arrester Visual inspection
4 Structure Monopole visual inspection, Concrete pole visual inspection,
Steel lattice visual inspection
Anchor & Guy Visual inspection, Tensile strength
5 Concrete/Grillage Grillage visual inspection, Concrete visual inspection, Non-
foundation destructive test of concrete
Stub Visual inspection
6 OHGW/OPGW Visual inspection, Loss of zinc, Tensile strength
OHGW fittings Visual inspection
Marker Visual inspection
Grounding system Earth resistance, Visual inspection
7 Danger sign Visual inspection
Tower no. sign Visual inspection
Phase plate Visual inspection
8 Right of way Visual inspection, Elec. clearance
174 W. Luejai et al.

Table 3. Evaluation criteria for visual inspection of ASC/GA, ASC/AW, AAC conductor.
Score Corrosion of steel core - loss of zinc
0 New conductor, outer surface glossy
2 Outer surface of stand line rough, light gray color and not glossy
3 Change to black or red color from surface stand rust, Slag around stand and gap
between outer
4 Inflatable, Change to black or red color all conductor and aluminum stand surface
less than 25%
5 Inflatable and aluminum stand surface more than 25%

Table 4. Evaluation criteria for loss of tensile strength ACSR conductor.

Score Corrosion of steel core - loss of zinc
0 Remaining Tensile Strength > 100% Rated Tensile Strength (RTS)
1 95% RTS < Remaining Tensile Strength < 100% RTS
2 90% RTS < Remaining Tensile Strength < 95% RTS
3 85% RTS < Remaining Tensile Strength < 90% RTS
4 80% RTS < Remaining Tensile Strength < 85% RTS
5 Remaining Tensile Strength < 80% RTS, Loss of Aluminum Cross Sectional Area
15%

Table 5. Evaluation criteria for thermography of compression joint.

Score Severity Temp rise (°C)
1 Minor <10
2 Intermediate 10–35
3 Serious 35–75
4 Critical >75

2.2 The Analytical Hierarchy Process

For the condition evaluation, the Weighing and Scoring Method (WSM) and a tech-
nique of multi-criterion analysis have been applied to determine condition indices. The
Analytical Hierarchy Process (AHP) [9, 10] is applied to determine a percentage of
important weight of all criteria or sub-components. Then, condition evaluation of
components (%CI) and health index (%HI) of OHL are calculated.
Weighting assigned by using Analytical Hierarchy Process (AHP) is used to define
suitable weighting factor of each testing method and component. The AHP is a multi-
decision method for multi-persons in multi-criteria with involving comparison. As the
AHP is based on pair-wise comparison of individual criterion must be firstly paired
against other criteria in a hierarchical structure. Finally, the weighting factor of each
criteria can be calculated. The condition weights of eight criteria by expert 1 is given in
Table 6 as example, which the pair-wise matrix (the paired value is =1, >1, <1 means
 Condition Evaluation of High Voltage Transmission Line in Thailand 175

equal, more, lower importance) and sum value in each column from judgment matrix
form. Next, each element in the matrix by column is divided to generate the normalized
judgment matrix. Then divide the normalized column sum by number of criteria (8) to
generate weight matrix. The priority weight vector is obtained. The result shows that
condition weights for 8 criteria beginning with conductor, conductor accessories,
insulator, steel structure, foundation, tower accessories and right of way are 0.30, 0.15,
0.08, 0.11, 0.05, 0.26, 0.02 and 0.03, respectively as given in the last column in
Table 7.

Table 6. A pair-wise matrix of component by expert 1.

Criteria C1 C2 C3 C4 C5 C6 C7 C8
C1 1 7 5 3 5 1 7 9
C2 1/7 1 3 3 5 1/3 7 5
C3 1/5 1/5 1 1/3 3 1/5 5 3
C4 1/3 1/3 3 1 3 1/3 5 5
C5 1/5 1/5 1/3 1/3 1 1/5 3 3
C6 1 3 5 3 7 1 7 7
C7 1/7 1/7 1/5 1/5 1/3 1/7 1 1/3
C8 1/9 1/5 1/3 1/5 1/3 1/5 3 1
Sum 3.12 12.07 17.86 11.06 24.66 3.40 38 33.33

Table 7. Weights of judgment matrix for 8 component.

Criteria C1 C2 C3 C4 C5 C6 C7 C8
C1 0.32 0.58 0.28 0.27 0.20 0.30 0.18 0.27
C2 0.05 0.08 0.17 0.27 0.20 0.10 0.18 0.15
C3 0.06 0.03 0.06 0.03 0.12 0.06 0.13 0.09
C4 0.11 0.03 0.17 0.09 0.12 0.10 0.13 0.15
C5 0.06 0.02 0.02 0.03 0.04 0.04 0.08 0.09
C6 0.32 0.25 0.28 0.28 0.28 0.30 0.18 0.21
C7 0.05 0.01 0.01 0.02 0.01 0.04 0.03 0.01
C8 0.04 0.02 0.02 0.02 0.01 0.04 0.08 0.03
Sum 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Finally, check the consistency of judgment matrix using consistency analysis to

calculate maximum eigenvector, kmax as shown in (1).
176 W. Luejai et al.

2 3
2 3 1 (a W þ a W þ . . . þ a W )
kmax;c1 6 W1n
11 1n 12 2n 1n nn
7
6 kmax;c2 7 6 1 (a W1n þ a22 W2n þ . . . þ a2n Wnn ) 7
6 7 6 w2n 21 7
6 .. 7 ¼ 6 . .. 7 ð1Þ
4 . 5 6 . 7
4 . . 5
kmax;cn 1 (a W þ a W þ . . . þ a W )
1n n2 2n nn nn
W n1nn

The maximum eigenvectors (kmax) using the values in Tables 6 and 7 are calculated
by using Eq. (1). Then, kmax,C1 = 10.00, kmax,C2 = 9.26, kmax,C3 = 7.48, kmax,C4 = 8.95,
kmax,C5 = 9.69, kmax,C6 = 8.73, kmax,C7 = 9.19 and kmax,C8 = 8.99. The kmax is average
value of the eigenvectors is 9.03. The Consistency Index (CI = (kmax − n)/(n − 1)) and
Consistency Ratio (CR = CI/RI) are calculated with the value of Random Index (RI) is
1.41 when number of criteria is equal to 8.
For the judgment matrix more than 5  5, the CR must be less than 0.1. Conse-
quently, the obtained CR is accepted (CR = 0.09 < 0.1). With the AHP pair-wise
comparison matrixes by 10 expert’s, the average percentages of component weight for
every transmission tower are 25.31, 2.91, 10.48, 15.48, 19.06, 16.89, 3.10, 6.77,
respectively as given in %Wj in Table 11.

2.3 Health Index Calculation

After the components and sub-components of the OHL have been classified. Visual
inspection on tower and special tests for some equipment are performed in order to
evaluate the condition of each transmission tower. Once the evaluated condition of
components of each transmission tower of one span ahead is evaluated (component
level), the condition of each tower is then evaluated (tower level). Finally, the overall
condition of an OHL is assessed (OHL level) as given Fig. 1.

Fig. 1. OHL assessment hierarchy.

 Condition Evaluation of High Voltage Transmission Line in Thailand 177

The condition evaluation applies WSM approach, which composes of scores and
weights used to calculate component’s conditional index (%CIcomponent), the health
index of transmission tower (%HItower) and the overall health index of OHL (%HIline).
All indices are shown in form of percentage. Therefore, the condition of component
and OHL can be evaluated and shown the ranking of health index. Finally, the reno-
vation or replacement task and schedule of individual OHL can be performed. For
condition evaluation by applying the WSM method, the %CIcomponent of each com-
ponent is calculated and shown in Eq. (2).
P
M
(Si  Wsubcomponent;i )
i¼1
%CIcomponent ¼  100 ð2Þ
P
M
(Smax  Wsubcomponent;i )
i¼1

where Si and Smax are score and maximum score of each sub-component, %Wsub-
component,i is important weight of sub-component ith while M is number of sub-
components of each component.
From Eq. (2), %CIcomponent,j of each component is further used to calculate the
health index (%HItower) of transmission tower as shown in Eq. (3). The %CIcomponent,j
includes conductor (%CIC), conductor accessories (%CICA), insulator (%CII), steel
structure (%CISS), foundation (%CIF), lighting protection (%CILP), tower accessories
(%CITA), and right of way (%CIROW).
P
N
ð%CIcomponent;j  Wj )
j¼1
%HItower ¼  100 ð3Þ
P
N
ð%CIcomponent;j;max  Wj )
j¼1

where %Wj is important weight of component jth. The %CIcomponent,j,max is the maxi-
mum of each %CIj and N is a total number of component.
From Eq. (3), the worst-%CIcomponent (%CIWcomponent) of each component in OHL
is further used to calculate the health index of OHL (%HIline) as given in Eq. (3).

P
N
ð%CIWcomponent;j  Wj )
j¼1
%HIline ¼  100 ð4Þ
P
N
ð%CIcomponent;j;max  Wj )
j¼1

where %CIWcomponent,i is the worst-%CIcomponent,j whereas %Wj and N are similar to

Eq. (3).
178 W. Luejai et al.

Table 8. Health index of OHL condition and required action.

%HI Condition Requirements Color
band
0–21 Good Normal condition, Normal scheduled maintenance Green
22–54 Moderate Increasing scheduled maintenance, Medium term plan to Yellow
renovate/replace
55–100 Poor Needs replacement of defective equipment, Short term Red
replacement planning by new OHL

The health index is divided into three zones and displayed as three traffic light
colour bands as “green” for “normal condition” with normal scheduled maintenance,
“yellow” for “moderate condition” that needs following up the condition and increasing
the frequency of scheduled maintenance as well as medium term planning to
renovate/replace the OHL and “red” for “poor condition” that needs for urgent
replacement of defective equipment or short term renovation or replacement planning
by new OHL. The colour bands are practically easy to quickly understand the condition
of OHL component. The ranges of %HI are shown in Table 8.
The obtained %HI results are useful information to renovation or replacement
planning and maintenance scheduling management in order to extend a service life of
each line and maintain system reliability. Various maintenance experiences on OHL
components were described in [11–17] in order to better operating efficiency and
system reliability while the equipment lifetime could be prolonged.

3 Result and Discussion

The 230 kV A-B#1 OHL including 10 towers is assessed and given as example.
Step 1: Calculating conditional index of components (%CI). The scores (Si) and
weight (Wi) of conductor and conductor accessories after performing visual
inspection and special tests are given in Tables 9 and 10. By applying Eq. (2), %CIC
and %CICA are calculated and equal to 40% and 39%, respectively. Similarly for the
remaining 6 components, the %CIcomponent are calculated and shown in Table 11.

Table 9. Conditional index of conductor of tower#1.

Group Sub-Component Testing Si Wi %CICA
1 Conductor Visual inspection 3 25 40
Loss of zinc 1 25
Loss of tensile strength 2 25
Torsional ductility 2 25
Max Si = 5 100
 Condition Evaluation of High Voltage Transmission Line in Thailand 179

Table 10. Conditional index of conductor accessories of tower#1.

Group Sub-Component Testing Si Wi %CICA
2 Damper Visual inspection 2 20 39
Spacer Visual inspection 0 10
Special test 1 10
Dead end Resistance 2 10
Thermography 3 10
Joint Resistance 0 10
Thermography 1 10
PG clamp Thermography 4 20
Max Si = 5 100

Table 11. %CIcomponent,j and %HItower#1 of TL 230 kV A-B#1.

Tower Component %Wj %HItower#1 %HItower#1
Tower#1 Conductor 40 25.31 39.76
Conductor accessories 39 2.91
Insulator 30.8 10.48
Steel structure 60 15.48
Foundation 50 19.06
Lightning protection 23.53 16.89
Tower accessories 80 3.10
Right of way 0 6.77
Max %CI = 100 100

Step 2: Calculating health index of each tower (%HItower). %CIcomponent of eight

components as conductor (%CIC), conductor accessories (%CICA), insulator (%CII),
steel structure (%CISS), foundation (%CIF), lighting protection (%CILP), tower
accessories (%CITA), and right of way (%CIROW) are used to calculate conditional
index of tower no.1 (%HItower#1) of OHL 230 kV A-B#1. Similarly, by using the
same application, the health indices of all 10 towers are calculated. The result is
given in Table 12. By applying Eq. (3), %HItower#1 to %HItower#10 are calculated. %
HItower#1 is equal to 39.76 while others are shown in Table 12. Then, the worst-%
CIcomponent (%CIW) of each component as the maximum %CI of each component is
concluded in the last row and further used to calculate the health index of the OHL
(%HIline).
180 W. Luejai et al.

Table 12. %CICOMPONENT,J and %HITOWER of 10 towers of TL 230 kV A-B#1

Tower %CIcomponent,j %HItower
%CIC %CICA %CII %CISS %CIF %CILP %CITA %CIROW
#1 40 39 30.77 60 50 23.53 80 0 39.76
#2 12.63 80 30.77 60 50 18.82 80 0 41.76
#3 12.63 40 30.77 60 50 18.82 80 0 36.96
#4 12.63 40 30.77 60 50 9.41 80 0 36.39
#5 12.63 80 30.77 0 0 23.53 80 0 19.94
#6 12.63 40 24.62 60 40 18.82 80 100 38.66
#7 12.63 40 24.62 60 50 18.82 0 100 38.76
#8 12.63 40 24.62 60 40 18.82 0 100 36.26
#9 12.63 40 24.62 60 50 18.82 80 100 41.16
#10 12.63 40 24.62 60 50 9.41 80 100 40.59
%CIWj 40 80 30.77 60 50 23.53 80 100 41.76

Step 3: Calculating overall health index of %HIline of TL 230 kV A-B#1. Applying

Eq. (4), %HIline of TL 230 kV A-B#1 is equal to 47.72% as given in Table 13.

Table 13. %HILINE OF TL 230 kV A-B#1.

Line Component %CIWcomponent,j Wj %HILINE
230 kV A-B#1 Conductor 40 25.31 47.72
Conductor accessories 80 2.91
Insulator 30.77 10.48
Steel structure 60 15.48
Foundation 50 19.06
Lightning protection 23.53 16.89
Tower accessories 80 3.10
Right of way 100 6.77
Max %CIW = 100 100

According to the scoring method as presented in Sect. 2.1, the score is classified as
0 (very good), 1 (good), 2 (satisfy), 3 (moderate), 4 (degradation) and 5 (reaching end
of life), which means that the higher %HIline, the worst condition of the equipment. The
%HIline of TL 230 kV A-B#1 is located in yellow zone area. Then, the maintenance
requires a moderate care and increasing maintenance schedule with medium term
planning to renovate/replace a OHL.
 Condition Evaluation of High Voltage Transmission Line in Thailand 181

Table 14. Evaluated results on actual twenty OHLs in Thailand.

Line %HILINE Zone

230 kV Line#1 (A-B#1) 47.72% Moderate
115 kV Line#1 48.99% Moderate
115 kV Line#2 48.40% Moderate
115 kV Line#3 68.06% Poor
115 kV Line#4 61.09% Poor
115 kV Line#5 63.59% Poor
115 kV Line#6 48.39% Moderate
115 kV Line#7 50.97% Moderate
230 kV Line#2 2.13% Good
230 kV Line#3 49.54% Moderate
230 kV Line#4 27.97% Moderate
230 kV Line#5 44.74% Moderate
230 kV Line#6 15.53% Good
230 kV Line#7 19.13% Good
500 kV Line#1 27.04% Moderate
500 kV Line#2 31.84% Moderate
500 kV Line#3 40.66% Moderate
500 kV Line#4 42.34% Moderate
500 kV Line#5 16.96% Good
500 kV Line#6 20.10% Good

After obtaining this developed procedure, additional nineteen OHLs of 115, 230
and 500 kV with actual data in Thailand’s transmission network are selected for
evaluation. Those new OHL, moderate age OHLs and high aging OHLs was selected to
justify the determined health index ranges as normal, moderate and poor condition. For
the normal condition, scheduled maintenance is applied. For the moderate condition,
the increasing of scheduled maintenance as well as medium term planning for reno-
vation or replacement should be conducted. Lastly for the poor condition, the urgent
replacement of defective equipment or replacement planning by new OHL are needed.
The results are shown in Table 14.

4 Program Development

Database management system and information technology are used to develop the on-
line web application providing useful information for on-line recording testing results
and database, evaluating the whole transmission network, showing HV OHL details
182 W. Luejai et al.

and evaluation results as well as effective maintenance planning and maintaining

activities. The web-application program applies a freeware SQL Server program for on-
line recording database, Microsoft SQL Server Management Studio for creating tables
in database, Adobe Dreamweaver program for designing webpage using PHP language
connecting SQL database, and HTML language for designing webpage structure.
JavaScript language is used for creating dynamic updating content, control multimedia,
animation, and other features. The webpages show HV OHL condition on details,
evaluation results of components, circuits, lines and overall transmission network
information, as well as a requirement for HV OHL maintenance activities. The program
is real-timely, effectively, and user friendly to all users.

5 Conclusions

A novel procedure to evaluate the condition of OHL is proposed in this paper. The
important and necessary testing methods and visual inspection as well as criteria for
condition assessment of all transmission components are suggested. The weighting and
scoring method was applied to calculate health index of transmission components and
overall lines. The practical data from on-site tests, visual inspection and special tests of
twenty OHLs in 115, 230 and 500 kV Thailand’s transmission network was used.
Finally, the overall conditions of those OHLs were shown. The results show that three,
twelve, and five OHLs are in poor, moderate and good condition, respectively. The
transmission utility applies these results to effectively prioritize urgent lines with the
urgent maintenance tasks as well as successfully manage maintenance planning
according to the actual condition of OHLs in order to replace/renovate or propose an
investment for the new transmission line.

Acknowledgement. Authors would like to acknowledge the Department of Transmission

System Maintenance, Department of Research and Development, the Electricity Generating
Authority of Thailand for funding and all cooperation.

References
1. Cigre: Overview of Cigre publications on asset management topics. Cigre Technical report,
Electra no. 262, June 2012
2. Cigre: Ageing of the system impact on planning. Cigre Brochure 176, WG 37.27 (2000)
3. Correia, D.M., Brasil, D.O.C., Lima, D.C.: A proposal for the evaluation of transmission line
performance in a competitive market. Cigre Session 22-104 (2002)
4. Van Der Wal, A., Ross, A.A.H.J., Kema, T.D.C.: Condition assessment of overhead lines.
Cigre Session B2-204 (2004)
5. Ibrahim, A., Pharmatrisanti, A.: Transmission line assessment. In: 2012 IEEE International
Conference on Condition Monitoring and Diagnosis, Bali, Indonesia, 23–27 September 2012
(2012)
6. de Saboia Stephan, J.C., Ferreira Costa, C.: Uprate and upgrade of overhead transmission
lines methodologies and reliability. CIGRE Session B2-202 (2008)
 Condition Evaluation of High Voltage Transmission Line in Thailand 183

7. Brennan, G.F.: Refurbishment of existing overhead transmission lines. Cigre Session B2-203
(2004)
8. Cigre: International survey of component costs of overhead transmission lines. Cigre Electra
no. 137 (1991)
9. Tanaka, H., Tsukao, S., Yamashita, D.: Multiple criteria assessment of substation conditions
by pair-wise comparison of analytic hierarchy process. IEEE Trans. Power Deliv. 25(4),
3017–3023 (2010)
10. Saaty, T.L.: Relative measurement and its generalization in decision making: why pairwise
comparisons are central in mathematics for the measurement of intangible factors-the
analytic hierarchy/network process. Rev. R. Span. Acad. Sci. 102, 251–318 (2008)
11. Van Der Wal, A., Voncken, R.M.A.M.: Extending the service life of aged overhead line
towers. CIGRE Session B2-212 (2008)
12. Phillippe Grand: The life extension policy of overhead lines. Cigre Session B2-306 (2010)
13. Cigre SC 22 WG13: Working group preliminary report: management of existing overhead
OHLs. Cigre Session 22-107 (2000)
14. Chone, F., Gaudry, M., Parrotta, J., Gourit, L., Larripa, B.: Assessment of existing overhead
transmission lines and solutions for extending their residual lifetimes: Cigre Session 22-201
(2000)
15. Smirnov, A.A., Romanov, P.I.: Database for 330–750 kV transmission line insulation:
structure, data mining, use for the purpose of improvement of line operation. Cigre Session
15-101 (2002)
16. Garcia-Alamo, C.J., Palacios, J.: Structural analysis for transmission lattice steel tower in the
400 kV transmission lines EL Tablazo – Cuatricenteneario No. 1 and 2. Cigre Session B2-
302 (2010)
17. Cigre WG B2.03: Guide for the assessment of old cap and pin and long-rod transmission line
insulators made of porcelain or glass: what to check and when to replace? Cigre Technical
report, Electra no. 228, October 2006
 Risk Assessment of Power Transformer
in Thailand’s Distribution Grid

Thanapong Suwanasri(&), Suphon Kumpalavalee,

and Cattareeya Suwanasri

King Mongkut’s University of Technology North Bangkok,

Bangkok 10800, Thailand
thanapong.s.epe@tggs-bangkok.org

Abstract. Once power transformer has been installed in service, it is degraded

by thermal, electrical and mechanical stresses. To upkeep its condition and
prevent catastrophic consequences from its failure the preventive and condition-
based maintenances should be applied appropriately. The appropriate mainte-
nances are considered based on power transformer usage risk, which is evalu-
ated from present condition and importance of power transformer. To determine
the present condition, the health index of power transformer and its components
are at first determined by scoring the test results of visual inspection, electrical
and insulating oil tests. Then, the obtained scores and weights, which were
determined by applying the analytical hierarchy process to brainstorm the util-
ity’s expertise, were further calculated as component and overall health index of
power transformer. Moreover, the major concerns of power transformer usage,
such as loading percentage, load importance, redundancy, system fault level,
number of through fault per year, transformer age, operation and maintenance
satisfaction, are evaluated as importance index of power transformer by scoring
and weighting technique. Then, the health and importance indices are coherently
considered to access power transformer usage risk displayed on risk matrix. The
aforementioned procedures were developed in Microsoft Excel to access 67
transformer’s usage risks in Thailand’s distribution grid. The software was
developed to display the result as traffic light color as red, yellow and green for
high, moderate and low usage risk respectively. The obtained risk are used to
manage maintenance strategies and proper usage of power transformer
according to its condition and usage risk.

Keywords: Condition
Maintenance strategy
Power transformer
Usage risk

1 Introduction

The aim of this paper is to determine the usage risk of power transformer [1–4] fleet in
a distribution grid of a utility in Thailand as known as risk index [1–3]. Hence, the risk
index is evaluated by considering health index [5] and importance index [1–3] of power
transformer together and subsequently located the risk value in the risk matrix [1–3].
The health index of each power transformer consists of its major component [6]
conditions as active part, load tap changer, bushing, surge arrester, insulating oil
condition, cooling part, self-protection and indicating device etc. The mentioned

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 184–196, 2020.
https://doi.org/10.1007/978-3-030-31676-1_18
 Risk Assessment of Power Transformer in Thailand’s Distribution Grid 185

component conditions are determined as component health index by applying scoring

and weighting techniques. To determine the component condition, test methods and
results of each component are considered according to the recommendation in the
international standards [6–9] and maintenance experiences of a utility. The obtained
component conditions are further determined together by using scoring and weighting
technique to achieve the overall health index of power transformer. However, another
required parameter is importance index [1–3]. The importance index is evaluated by
considering several aspects [10–13] such as the percentage of supplying load, load
importance, redundancy, system fault level, number of through fault per year, age of
transformer, technology and satisfactory in operation and maintenance.
Eventually, the risk of power transformer usage is determined by considering the
obtained health index and importance index of power transformer. On the risk matrix,
the x-axis refers to the importance of power transformer [1] and y-axis refers to the
health index or power transformer condition [1] as shown in Fig. 1. The risk is mea-
sured by the distance normal to the 45° line from the x-axis for equal weighting
between condition and importance index.

Fig. 1. Located risk and risk measurement in the risk matrix

The risk index and the coordination location in the risk matrix is an useful infor-
mation, which helps utility to effectively manage the proper strategy of power trans-
former [10] utilization, such as relocation to reduce the stress due to high loading
percentage, refurbishment for old or damaged transformers, performing condition
based instead of preventive or corrective maintenance to upkeep the power transformer
condition and extend the useful life of power transformer [12, 13].

2 Weighting and Scoring Technique

In this section, the scoring and weighting technique used to evaluate the risk of power
transformer usage is described.
186 T. Suwanasri et al.

Weight technique is performed to prioritize the significance of power transformer’s

major components such as active part, insulation part, bushing, surge arrester, main-
tank, load tap changer, cooling system, protective device and monitoring device. In
addition, sub-components of each major component as well as test methods of indi-
vidual sub-components are weighted. Moreover, importance criteria such as load
property, outage interval effect, fault effect, aging effect, technology, load type, and
manufacturer satisfaction are also weighted. Hence, several components, sub-
components, test method and importance criteria are considered in the evaluation
procedure. Thus, the analytical hierarchy process (AHP) [14] as an effective judgement
tool using pair-wised comparison technique to prioritize the significance of those cri-
teria because the procedure involves different aspects of many experts.
Scoring technique is performed by considering the recommended value of each test
technique for pass/fail test or good/bad condition in the international standards of
power transformer tests. The recommended limit value of individual electrical testing is
used to define the maximum or minimum score in the scoring techniques in order to
represent the condition of power transformer component. Moreover, experiences
gained from power transformer testing in a Thailand’s utility has been combined
together for ranging the scores from 1–5, while 5 represents the failed condition and 1
represents the good condition.
Moreover, not only the electrical tests but also visual inspection and insulating oil
test have been considered and converted the obtained results to numerical value by
applying the similar scoring technique. The visual inspection and insulating oil test are
importance tests because both tests are the basis requirement and can be frequently and
regularly tested at the pre-determined interval. Any failed result is unacceptable. Failed
result, which can’t be fixed immediately, might lead to power transformer failure, this
transformer is classified as “Unhealthy power transformer”.

3 Power Transformer Risk Evaluation

3.1 Health Index Evaluation

The main purpose of power transformer assessment is to determine the present con-
dition and arrange the proper strategy of power transformer usage. Hence, the present
condition should be accurately evaluated. Nevertheless, power transformer condition
evaluation is complicated. Hence, the health index is initially developed by classifying
the major components into group due to its function. The classified components include
active part, insulator, bushing, surge arrester, main tank, tap changer, cooling and
protective device. The test results, which are used to evaluate the condition for each
power transformer component, are shown in Table 1.
 Risk Assessment of Power Transformer in Thailand’s Distribution Grid 187

Table 1. Test method for condition evaluation

Major component Sub – component Applied test method
Active part Magnetic core Insulation resistance
HV winding Turn ratio
Power factor
Capacitance
DC winding resistance
Exciting current
Single phase leakage impedance
Three phase short circuit impedance
Polarization index
LV winding Power factor
Capacitance
DC winding resistance
Polarization index
TV winding Power factor
Capacitance
DC winding resistance
Polarization index
Insulation part Paper Furan
Maintank oil Oil contamination
Power factor
DGA
OLTC oil Oil contamination
Bushing – Power factor
Capacitance
Surge arrester – Leakage current
Watt loss
Maintank – Visual inspection
Load tap changer – Visual inspection
Functional test
Cooling system Radiator Visual inspection
Fan Functional test
Pump
Protecting device Pressure relief Visual inspection
Conservator tank Functional test
Protective relay
Monitoring device Oil level indicator Visual inspection
Winding temperature meter Functional test
Top oil temperature meter
Protective relay
188 T. Suwanasri et al.

Table 2. Range score and standard limit

Score Range
Test Method
5 4 3 2 1
Turn Ratio :
>1 1-0.9 0.9-0.7 0.7-0.5 <0.5
Nameplate Error <1%
Power Factor :
>1 1-0.8 0.8-0.6 0.6-0.5 <0.5
Commissioning Error <1%
Capacitance :
>20 20-17 17-14 14-10 <10
Commissioning Error <20%
DC Winding Resistance :
>5 5-3.5 3.5-2.5 2.5-1 <1
Commissioning Error <5%
Exciting Current :
>10 10-8 8-6 6-5 <5
Commissioning Error <10%
1Ø Leakage Impedance :
>5 5-3.5 3.5-2.5 2.5-1 <1
Commissioning Error <3%
3Ø Short Circuit Impedance :
>3 3-2 2-1 1-0.5 <0.5
Nameplate Error <3%
Polarization Index :
>3 3-2 2-1 1-0.5 <0.5
R10min/R1min<1

Table 3. High voltage winding condition evaluation

Test Method Test Result Score Weight Result

Turn Ratio 0.37 0 5 0
Power Factor 0.157 0 4 0
Capacitance 0.37 0 3 0
DC Winding Resistance 2.66 3 5 15
Exciting Current 3.48 0 3 0
1Ø Leakage Impedance 1.455 3 3 9
3Ø Short Circuit Impedance 1.5079 3 4 12
Polarization Index 1.3056 0 3 0
Summation 30 36
High Voltage Winding Average Condition 24
High Voltage Winding Worst Case Condition 60

The evaluation criteria of HV winding is presented in Table 2. After determining

the score range, high voltage winding health index is further calculated by combining
the obtained score with its weight of test result. Weight of test results are determined by
AHP and brainstorming experts in utility according to the effectiveness of failure
finding, severity of detected failure, frequency of performing test etc. The condition
evaluation of high voltage winding is given in Table 3.
 Risk Assessment of Power Transformer in Thailand’s Distribution Grid 189

On average, the evaluated HV winding condition of power transformer is 36/150 or

24% representing as healthy condition. Despite, some test results and their scores are
shown as suspected condition. Hence to avoid the overlooking of hidden possible
defects by averaging technique, the worst case condition is developed to prior notice
and manifest/questionable poor condition from some testing as useful information to
perform proper maintenance action. In this case the worst case condition is from DC
winding resistance, single phase leakage impedance and 3 phase short-circuit impe-
dance as 3/5 or 60%.
Other sub-component conditions including magnetic core, low voltage winding and
tertiary winding are evaluated subsequently and further evaluated as active part con-
ditional health index as shown in Table 4 for averaging technique and Table 5 for
worst case condition technique. Finally, the overall power transformer conditional
health index is systematically evaluated as shown in Table 6.

Table 4. Active part condition evaluation by averaging technique

Sub - Component Evaluated Result Weight Result

Magnetic Core 0 30 0
High Voltage Winding 24 40 9.6
Low Voltage Winding 8 20 1.6
Tertiary Winding 5 10 0.5
Summation 100 11.7

Table 5. Active part condition evaluation by worst case condition technique

Sub - Component Evaluated result Weight Result

Magnetic Core 0 30 30
High Voltage Winding 60 40 24
Low Voltage Winding 80 20 16
Tertiary Winding 80 10 8
Summation 100 78
190 T. Suwanasri et al.

Table 6. Overall power transformer health index evaluation

Component Condition Evaluation Result

Major – Component % Full
Average Worst Average Worst
Active Part 11.7 78 20 2.34 15.6
Insulation Part 10.8 12 20 2.16 2.4
Bushing 32 63 15 4.8 9.45
Surge Arrester 62 80 10 6.2 8
Main tank 0 10.5 10 0 1.05
Load Tap Changer 0 11.4 10 0 1.14
Cooling System 0 0 5 0 0
Protecting Device 0 0 5 0 0
Monitoring Device 0 0 5 0 0
Summation 100 15.5 37.64

The obtained health index of each power transformer component are designed as a
graphic user interface in the MS-Excel program [10] as shown in Fig. 2 for immediate
recognition of individual major component condition.

Fig. 2. Power transformer condition

3.2 Importance Index Evaluation

Importance of power transformer in the electrical network is complicated and difficult
to identify. With the co-operation from Thailand’s utilities, power transformer’s duties,
major and minor concerns are classified as shown in Table 7. The first aspect is system
impact considering load type, percentage of power transformer loading and
 Risk Assessment of Power Transformer in Thailand’s Distribution Grid 191

redundancy. While the second aspect is probability of failure at the installed location
considering fault level, age, technology obsolescence and satisfactory in operation and
maintenance. Regarding to Table 7, power transformer importance which is located at
A1 substation in Thailand distribution grid is evaluated and shown in Table 8.

Table 7. Criteria for importance evaluation

Criteria Description Result Score
Load property %Load level compared to rated power of <20% 1
transformer 20–40% 2
40–60% 3
60–80% 4
>80% 5
%Amount of important feeder, which has been <50% 1
set under frequency relay 50% 3
>50% 5
Outage interval N-1 Criteria, Possibility for continuous Yes 0
effect shutdown No 5
Fault effect Fault level (kA) at main bus 22 kV <5 1
5–10 2
10–15 3
15–20 4
>20 5
Through fault (times/year) <1 1
1–5 2
5–10 3
10–15 4
>15 5
Aging effect Transformer age (years) <5 1
5–10 2
10–15 3
15–20 4
>20 5
Technology New technology, safety operating, Yes 0
environment friendly No 5
Load type Type and area consuming power form Countryside 1
transformer Downtown 3
Industry 5
Manufacturer Failure history, failure rate, spare part and Satisfaction 1
satisfaction team service Moderate 3
Unsatisfactory 5
192 T. Suwanasri et al.

Table 8. Importance evaluation result of power transformer at location TR1 in A1 substation

Criteria Description Result Score
Load property %Load level compared to rated power of >80% 5
transformer
%Amount of important feeder, which has been >50% 5
set under frequency relay
Outage interval N-1 Criteria, Possibility for continuous Yes 0
effect shutdown
Fault effect Fault level (kA) at main bus 22 kV >20 kA 5
Through fault (times/year) 5–10 3
Aging effect Transformer age (years) 23 5
Technology New technology, safety operating, No 5
environment friendly
Load type Type and area consuming power form Industry 5
transformer
Manufacturer Failure history, failure rate, spare part and Unsatisfactory 5
satisfaction team service

The obtained score is consequently used to evaluate the importance index by

applying score and weight technique. Similarly, AHP is applied once again to deter-
mine the weight of each importance index criteria. The importance index of position
TR1 at A1 substation is shown Table 9.

Table 9. Importance index evaluation

Criteria Result Score %Full Evaluation result
%Load level >80% 5 12 12
Important feeder >50% 5 8 8
N-1 Criteria Yes 0 12 0
Fault level (kA) >20 kA 5 8 8
Through fault 5–10 3 10 6
Age 23 5 5 5
Technology No 5 5 5
Load type Industry 5 20 20
Satisfaction No 5 20 20
Summary 84

The importance index of power transformer TR1, which is located at A1 substation,

is 84 classified as important transformer to the network.
 Risk Assessment of Power Transformer in Thailand’s Distribution Grid 193

3.3 Risk Evaluation

Risk index represents usage risk of power transformer. Based on the known condition
and importance of power transformer as mentioned previously, the usage risk can be
calculated by using Eq. 1. The obtained risk is further used to prioritize the mainte-
nance tasks and arrange the proper usage of power transformer. These proper and
effective usage can extend the power transformer life and reduce the excessive main-
tenance cost.
p
d ¼ ð%H:I: þ %I:M:I:Þ= 2 ð1Þ

Whereas
%H.I. = Percentage of Health Index
%I.M.I. = Percentage of Importance Index
d = Power Transformer Risk Index
The usage risk of TR1 is 70.357 and located in zone 6 of Fig. 3 representing
moderate condition and high importance.
Regarding to Table 10, TR1 has moderate usage risk and should be repaired the
defective surge arrester immediately due to its high importance to the network. Outage
of TR1 could lead to adverse impact to the system because of its high loading per-
centage and significant load. Even though having moderate condition, it is 23 years old
with obsolete technology and no more satisfactory in operation due to high failure rate
and difficulty in spare part finding. Hence with this obtained information, it should be
replaced by new.

Fig. 3. Location of TR1 on risk matrix

194 T. Suwanasri et al.

Table 10. Risk assessment and recommendation

Risk Index Risk Action

0 – 28 Very low Normal Maintenance
29 – 56 Low Normal Maintenance
57 – 84 Moderate Repair
85 – 112 High Refurbish, Relocate
113 - 142 Very high Replace

4 Risk Assessment and Maintenance Strategy

The power transformers in distribution network are evaluated by the aforementioned

procedure. 206 115/22 kV power transformers are completely evaluated and displayed
the risk assessment result in the risk matrix as shown in Fig. 4. Zone 1, 2, 4 and 5 have
low to moderate condition and low to moderate importance, which is acceptable risk.
Then power transformer in these zones can be normally maintained with normal usage.
The highest risk zone is zone 9, which requires immediate action to reduce the
importance and improve the poor condition of power transformer. Zone 7 and 8 are
interesting because they represent poor condition with low to moderate importance,
respectively. Hence power transformer within these zones requires repair or refur-
bishment. Zone 3 and 6 represent high importance with low to moderate condition,
respectively. Hence power transformer within these zones requires immediate reduction
of system impact by load reduction, reduction of forced outage or refurbishment.
According to the risk assessment result, 7 transformers are located in zone 2, 1
transformer in zone 4, 176 transformers in zone 5, these power transformer can be
normally maintained with normal usage. The moderate condition of power transformer
in zone 4 and 5 is mainly caused by poor condition of high voltage winding for 94%,
moderate condition of oil insulation in main tank and OLTC for 63.4% and poor
condition of arrester for 5% of the whole number of fleet.
There are 10 power transformers in zone 6 having high importance to the network
because most of them are carrying load more than 80% of their rated power and feeding
to industrial area. To reduce risk of these power transformers, load reduction and
redundancy improvement are required. Then, 12 transformers in zone 8 have poor
condition with moderate importance. Therefore, the relatively urgent action should be
performed to improve the condition. From 8 of 12 power transformers, the defective
parts are HV winding, LV winding, insulating oil in main tank and OLTC, which are in
poor condition. Moreover, over 700 °C thermal fault was detected by DGA test result
of these eight defective power transformers.
 Risk Assessment of Power Transformer in Thailand’s Distribution Grid 195

Fig. 4. Result of risk assessment of 206 power transformers

5 Conclusion

The risk of power transformer usage in distribution grid has been accessed by using the
risk matrix and scoring and weighting technique to calculate the health index and
importance index of power transformer fleet. The scoring has been performed to
classify the condition of power transformer components based on the recommendation
in the international standard and experience of maintenance crew of a utility. The
weight of each criterion comes from the brainstorming of experts from several divisions
working with power transformer by using the analytical hierarchy process with geo-
metric sum. Subsequently, this procedure has been applied to 206 power transformers
in power distribution grid. There are 184 power transformers or 89% are normal to
moderate condition and can be normally used and maintained. There are 10 transformer
locations, which has high importance to the grid, which requires load reduction or
redundancy improvement immediately. Lastly, 12 power transformers needs for thor-
ough inspection to determine the amount of repair or restoration works. This is useful
information for a utility to effectively manage the maintenance work and proper uti-
lization of power transformer fleet.

Acknowledgements. The authors would like to acknowledge the Research Division, Substation
Maintenance Division and Transformer Division of the Provincial Electricity Authority of
Thailand (PEA) for technical support and funding.

References
1. Muhr, M.: Asset and risk management of electrical power equipment. In: International
Symposium on Electrical Insulating Materials, Kitakyushu, Japan (2005)
2. Muhr, M.: Aging and degradation, their detection and monitoring & asset management. In:
International Symposium on Electrical Insulating Material, Mie, Japan (2008)
196 T. Suwanasri et al.

3. Muhr M., Sumereder, C.: Applying risk management for high voltage equipment. In: IEEE
International Conference on High Voltage Engineering and Application, China (2008)
4. Arshad, M., Islam, S., Khaliq, A.: Power transformer asset management. In: International
Conference on Power System Technology POWERCON, Singapore (2004)
5. Dominelli, N.: Equipment health rating of power transformers. In: IEEE International
Symposium on Electrical Insulation, USA (2004)
6. CIGRE WG A2.34. Guide-for-Transformer-Maintenance (2011)
7. IEEE 62-1995. Guide for Diagnostic Field Testing of Electric Power Apparatus (2005)
8. Gradnik, M.K.: Physical-chemical oil test monitoring and diagnostic of oil-filled
transformers. In: Proceeding of 14 International Conference on Dielectric Liquids, Graz,
Austria (2002)
9. IEC 60422-2013. Mineral Insulating Oils in Electrical Equipment - Supervision and
Maintenance Guidance (2013)
10. CIGRE WG B3.06. IT Strategies for Asset Management of Substations - General Principles
(2014)
11. CIGRE WG 37.27. Ageing of The System Impact on Planning (2000)
12. CIGRE WG A2.18. Life Management Techniques for Power Transformers (2003)
13. CIGRE WG A2.20. Guide on Economics of Transformer Management (2004)
14. Tanaka, H., Tsukao, S., Yamashita, D.: Multiple criteria assessment of substation conditions
by pair-wise comparison of analytic hierarchy process. IEEE Trans. Power Deliv. 25(4),
3017–3023 (2010)
 Cost-Benefit for HV Transmission Line
Renovation and Replacement Based on Failure
Probability and Risk-Based Maintenance

Cattareeya Suwanasri(&), Thanapong Suwanasri, Waraporn Luejai,

and Surapol Saribut

King Mongkut’s University of Technology North Bangkok,

1518 Pracharat 1 Rd., Wongsawang, Bangsue, Bangkok 10800, Thailand
cattareeya.s@eng.kmutnb.ac.th

Abstract. In this paper, the innovative method for cost-benefit analysis with
risk-based maintenance cost and strategies for renovation and replacement of
aged HV overhead transmission line (OHL) in Thailand is proposed. In the
analysis, a transmission usage risk involving with renovation and importance
index are concerned together with maintenance costs in order to calculate a risk-
based maintenance cost. This cost comprises cost of replaced component,
maintenance cost, reputation cost and cost of loss of selling electricity. The
renovation index is computed from practical scientific quantitative test results and
visual inspection while the importance index is computed from the criteria such as
percentage loading, system usage, voltage level, contingency, failure record, age,
human impact, pollution, public image of individual OHL; and finally those
indices are plotted in a risk matrix. The maintenance tasks are suggested for
different risk level; however, the risk-based maintenance cost is further compared
with the new investment cost to determine a breakeven point for replacing the old
line by the new one. Twenty OHLs in Thailand were analyzed in this paper. The
results show that five OHLs with high risk and the age is over than 35 years should
be planned for a new line while the others be proper maintained according to risk-
based maintenance strategy. This innovative method can effectively help trans-
mission utility making decision on transmission maintenance planning.

Keywords: Renovation index
Importance index
Cost-benefit analysis

Risk-based maintenance
Failure probability
Asset management

1 Introduction

Currently, the practical cost-benefit analysis in transmission network is a key factor to

minimize operating and maintenance cost while maximize reliability of system [1–3].
Some HV overhead transmission lines (OHLs) are aged and degraded causing unex-
pected failure such as broken conductor, flashover on insulator, collapse of steel
structure, etc. resulting to power interruption. Consequently, a condition assessment of
old transmission line is now prime concerned [4, 5]. The need to justify whether
renovation and replacement in OHL is widely interested [6, 7]. Different diagnostic
techniques and significant required data to assess the condition of OHL were

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 197–207, 2020.
https://doi.org/10.1007/978-3-030-31676-1_19
198 C. Suwanasri et al.

mentioned for different components [8–11]. However, the almost of methods and
condition assessment were based on visual inspection. The maintenance strategies
could be applied for different condition [12, 13].
Therefore, this paper proposes an innovative method for cost-benefit analysis based
on risk assessment for the practical OHL transmission lines in Thailand. The practical
classification of the transmission components and criteria according to utility experi-
ences are focused. The testing methods that includes scientific laboratory tests of each
OHL components, electrical tests and visual inspection for all components are pre-
sented. The methodologies showing state-of-the-art to calculate renovation and
importance index by applying weighting and scoring technique, analytical hierarchy
process (AHP) [14], risk-based maintenance cost, and finally cost-benefit analysis for
renovation or replacement planning of OHL are introduced. The OHL line, which the
risk-based maintenance cost is higher than the new investment cost, should be planned
for a replacement by a new OHL.

2 Risk-Based Maintenance

Risk matrix in Fig. 1 consists of %RI and %ImI of OHL line. %RI is calculated from
the results of practical scientific quantitative tests and visual inspection of transmission
components. %ImI is computed from different practical and significant criteria to
system reliability such as percentage loading, system usage as tie line, loop line or
radial line, voltage level, contingency analysis, failure record, age, human impact,
pollution, public image of individual OHL. By applying weighting and scoring method,
the risk in forms of renovation index (%RI) and importance index (%ImI) of individual
transmission line are calculated and plotted in risk matrix along y-axis and x-axis,
respectively. By means of both %RI and %ImI in the risk matrix, the risk-based
management on failure probability can be implied. The %RI and %ImI are divided into
3 levels as given in Tables 1 and 2. The colors of %RI are green, yellow and red
representing good, fair and poor condition while that of %ImI representing low,
moderate, and high importance to the system. For zoning %RI, the practical ranges are
justified according to test results reflecting %RI that need experience from experts to
get proper ranges, better effective and efficient maintenance management. The practical
ranges for %ImI are set due to the utility’s policy.

Fig. 1. Risk matrix.

 Cost-Benefit for HV Transmission Line Renovation and Replacement 199

Table 1. Maintenance requirements according to condition

%LRI Condition Requirements Color Code

100-55 Poor Replacement of defective equipment, construct a new OHL. Red
54-22 Fair Increasing of scheduled maintenance to follow condition
Yellow
and plan for new transmission line.
0-21 Good Normal maintenance, scheduled maintenance. Green

Table 2. Range of importance in horizontal matrix of the risk matrix

%ImI Importance Requirements Color Code

64-100 High Modify such high %ImI equipment or transmission line
such as reducing power transmission.
Add the transmission line to reduce the importance Red
Modify/correct or find out the methods to reduce the
fault/failure of transmission line.
36-63 Moderate Tracking the increase in power in line, improve the relia-
Yellow
bility of the transmission line.
0-35 Low Normal Operation Green

In Fig. 1, there are 9 clusters in risk matrix presented. The lowest risk (no. 1) has the
green color while the highest risk (no. 9) has the red color. The condition and importance
together with maintenance tasks for each cluster are detailed as follows. Cluster 1: Good
condition with low importance. There is minimal risk of use. It can be used for main-
tenance, repair or replacement when it is broken and check following normal conditions
or corrective maintenance (CM) with routine inspection. Cluster 2: Good condition with
moderate importance. Time-based maintenance (TBM) can be applied. Cluster 3: Good
condition with high importance. TBM can be applied together with condition-based
maintenance (CBM). Cluster 4: Moderate condition with low importance. Normal
maintenance or CM can be applied. Cluster 5: Moderate condition with moderate
importance. Preventive maintenance (PM) or regular maintenance or TBM can be
applied. Cluster 6: Moderate condition with high importance. CBM can be applied.
Cluster 7: Poor condition with importance. Repair or replace when fail (without
blackout). Cluster 8: Bad condition with moderate importance. Use the maintenance,
repair or rehabilitation or replacement on the economic suitability or budget or
replace/repair/refurbish by economic. Cluster 9: Bad condition with high importance.
This cluster has the highest risk and must be maintained repair or reconstruction
immediately. Construct a new OHL or replace/repair/refurbish immediately.

3 Principle of Evaluating Condition of Transmission Line

For condition assessment, the OHL is classified into eight components consisting of
conductor, conductor accessories, insulator, steel structure, foundation, lighting pro-
tection system, tower accessories, and right of way. The novel procedure for condition
assessment of OHL is shown in Fig. 2.
200 C. Suwanasri et al.

%RI

Fig. 2. Structure for condition assessment of OHL in transmission system.

The testing methods needed for condition evaluation for each groups are given in
Table 3. The Analytical Hierarchy Process (AHP) method [14] is applied to determine
importance weights of the components as given in Table 4 and of importance criteria
given in Table 6 for health index and important index calculation. %CI of each
component is calculated by using Eq. 1.

P
n
ðSi  %Wi Þ
i¼1
%CIc ¼ n   ð1Þ
P
Si;max  %Wi
i¼1

where %CIC is a percentage condition index of each component, Wi is importance

weight of each test method. For example, the conductor consists of only conductor,
then n is 1. %CIc for calculated according to four tests with equal %Wi of 25% is 80%.
%CIc of conductor together with %CI of other components are further used to
calculate overall condition or renovation index (%RIL) of OHL by using Eq. 2. The
calculating results are given in Table 5.
 Cost-Benefit for HV Transmission Line Renovation and Replacement 201

m 
P 
%CIj  %Wj
j¼1
%RIL ¼ m   ð2Þ
P
%CIj;max  %Wj
j¼1

where Wj is weight of each components and m is number of components.

Table 3. Testing methods for condition evaluation

Group Sub-components Testing method
1 Conductor Visual inspection, loss of zinc, tensile strength, torsional
ductility
2 Damper Visual inspection
Spacer Visual inspection, special tests
Dead end Resistance test, thermography
Joint Resistance test
3 Insulator Visual inspection, thermography
Fittings Visual inspection
Arrester Visual inspection
4 Structure Monopole, concrete pole, steel lattice visual inspection
Anchor and guy Visual inspection, tensile strength
5 Concrete/grillage Grillage visual inspection, concrete visual inspection, mon-
foundation destructive test of concrete
Stub Visual inspection
6 OHGW/OPGW Visual inspection, loss of zinc, tensile strength
OHGW fittings Visual inspection
Marker Visual inspection
Grounding system Earth resistance test, visual inspection
7 Danger sign Visual inspection
Tower sign Visual inspection
Phase plate Visual inspection
8 Right of way Visual inspection, electrical clearance checking

Table 4. Testing methods for conductor evaluation

Testing methods %Wi %Si %Wi*%Si %CIC
Visual inspection 25 4 100 80%
Loss of zinc 25 5 125
Loss of tensile strength 25 3 75
Torsional ductility 25 4 100
202 C. Suwanasri et al.

Table 5. Example on condition indices and theirs weight for overall condition assessment
Components Condition index %Wj %RIL
Conductor %CIC = 80% 25.31 84.39%
Cond. accessories %CICA = 75% 2.91
Insulator %CII = 90% 10.48
Steel structure %CITI = 95% 15.48
Foundation %CIF = 85% 19.06
Lightning protection %CILP = 75% 16.89
Tower accessories %CITA = 88% 3.10
Right of way %CIROW = 92% 6.77

4 Principle of Evaluating Importance of Transmission Line

The criteria for evaluating the importance of OHLs are as follows.

1. Percentage loading is considered from the maximum percentage of highest power
demand in a year compared to rated power (MVA) of each OHL. The higher
loading will cause the wider impact to customers, when the outage occurs.
2. System usage is an importance of using OHL in power system by considering that
OHL is used to connect power plant, used as tie line, rapid load shedding, radial line
or just normal line.
3. Voltage level is to consider power transmitted through OHL impacting to cus-
tomers. 500 kV OHLs should have more power transmitted than 115 kV OHL
resulting to more impact to customer in case of failure.
4. Contingency is considered by reliability of supply electricity. The lines without N-1
will cause system to be more affected than the line with N-1 because of redundancy.
5. Failure statistics of occurrence of malfunction/failure of OHLs means a frequency
or probability of problems in various causes by considering a number of the line
tripping from the electrical system.
6. Age of OHL is a reflection of the invisible deterioration and a chance of power
outage. OHL with aging and old design has specifications lowering than present
condition that may not be operated against high electrical and mechanical stresses.
7. Impact to customers considered by load or area that OHL is supplying power are
divided into 3 groups as industrial estate, big province/tourist attraction/important
business area, and general provinces.
8. Pollution and usage in the area near pollution zones such as salt vapor, chemicals,
rust corrosion, agriculture etc. are very important to cause OHL outage.
9. Impact on community as utility image is an interest issue because of limitations in
design and installation. For example, the OHL around Bangkok produces
electric/magnetic field interference to people as well as power outage causing power
usage interruption. All affects are rather impact to community than OHL connected
to the small province.
The importance criteria of each OHL are arranged in order of priority or “Scoring”
from very low (0: Very Low) to very high (4: Very High). The weight or “Weighting”
 Cost-Benefit for HV Transmission Line Renovation and Replacement 203

has been assigned to each criteria by considering the severity of the impact on the
electrical system. As summarized in Table 6, then scoring and weighting technique are
applied to calculate % ImI of each OHL as shown in Fig. 3.

P
p
ðSk  %Wk Þ
k¼1
%ImIL ¼ ð3Þ
P
p
Smax;k  %Wk
k¼1

where Sk is score of each criterion, Smax,k is maximum score, Wk is weight of each

criterion, and p is number of importance criteria.

Table 6. Importance criteria of HV OHLs to electrical system

Criteria Sk %Wk
(0) (1) (2) (3) (4)
%Loading 20–0 21–30 31–40 41–50 >50 14
System Usage No tie line Radial Tie line/rapid load 11
line shedding/generator
connected
Voltage level (kV)  115 230 500/300 kVDC 5
Contingency Non N-2 N-1 10
Analysis
Failure record 0 1–5 6–10 >10 6
(event/yr)
Age (yr) 10–0 20–11 25–21 30–26 >30 32
Public image Normal Industrial estate/big 7
province province tourist and
business area
Pollution (more Rice Plant Bird Coastal area/Industrial 5
than 20% of total field/agricultural burning/firing chit estate
line length) area
Human Normal line Compact line 10
impact/right of
way

Table 7. %ImI of TL 230 kV A-B#1

No. Important criteria Information Sk %Wk Sk*Wk Smax*Wk
1 Percentage loading (%) 48.5% 3 14 42 56
2 System usage Low 1 11 11 44
3 Voltage level (kV) High 3 5 15 20
4 Contingency analysis Very low 0 10 0 40
5 Failure record (event/yr) High 3 6 18 24
6 Age (yr) Very high 4 32 128 128
7 Human impact Very low 0 7 0 28
8 Pollution Moderate 3 5 15 20
9 Public image Low 1 10 10 40
204 C. Suwanasri et al.

5 Risk and Cost-Benefit Analysis

The measurement of the risk of OHLs can be achieved by measuring risk distance
(distance “d”) compared to the 45° line as shown in Fig. 3. The angle is defined by
giving importance weight between the condition and importance criteria of the OHL.
The angle of 45° means that both condition and importance criteria play equally
important role in transmission system of 50:50%. The distance “d” is calculated from
Eq. (3) shown below. Line “d” is perpendicular to a line doing 45° with both x-axis and
y-axis.

xþy
d ¼ pffiffiffi  100 ð4Þ
2

where x is %RIL and y is %ImI, these are plotted in the risk matrix as shown in Fig. 3.

Fig. 3. Risk matrix and distance “d” of 76.47%.

To answer the question, how long should this line be continually operated? Then,
the risk together with cost-benefit analysis is introduced in this paper. This is called
risk-based maintenance cost that includes cost of replaced components, maintenance
cost i.e. manpower and transportation, reputation cost and cost of loss of selling
electricity, which is used to compare with an investment of the new line.
The results of %RIL, %ImI, age and distance “d” of 20 lines are given in Table 8. In
Table 7, the 230 kV Line#8 is given as example. This line is a double circuit with
1  1272 MCM ACSR/GA conductor. Investment cost of the new line is 4.29
MTHB/km. The maintenance cost and required information of 230 kV line are given in
Table 9 including inflation rate, loss of sale, outage down time, electricity rate and loss
of reputation rate.
 Cost-Benefit for HV Transmission Line Renovation and Replacement 205

Table 10 shows the total risk-based maintenance cost of 20 OHLs. By multiplying

“d” of 76.47% with the replacement cost and maintenance cost of OHL 230 kV
Line#8, the results is called risk-maintenance costs that are equal to 2.88 and 0.40
MTHB, respectively. The loss of sale and reputation cost are of 1.88 and 0.12 MTHB,
respectively with the assumption that the line loss of 100 MW in 3 h with 3% inflation
rate. The overall risk-based maintenance cost is 2.88 MTHB.
This risk-based cost is further used compared with the investment of the new line of
4.29 MTHB. The risk-based maintenance cost is still lower than the investment cost of
new line. Consequently, this line should be continuously doing maintenance tasks as
suggested in Cluster 5 with PM or regular maintenance or TBM. Lastly, the results of all
lines in Table 10 show that the lines with high risk ages more than 30 years and the risk-
based maintenance cost is higher than new investment should be replaces by a new line.

Table 8. %ImI and %d of 20 lines

Line Age (yr) %RI (%FPline) %ImI %d
115 kV Line#1 37 48.99% 62.25% 78.67
115 kV Line#2 30 48.40% 52.00% 71.00
115 kV Line#3 28 68.06% 52.00% 84.91
115 kV Line#4 30 61.09% 52.00% 79.98
115 kV Line#5 30 63.53% 52.00% 81.70
115 kV Line#6 19 48.39% 49.25% 69.05
115 kV Line#7 38 53.33% 63.50% 82.62
230 kV Line#8 40 47.72% 59.25% 76.47
230 kV Line#9 12 2.13% 0% 1.51
230 kV Line#10 35 49.54% 59.75% 77.29
230 kV Line#11 18 27.97% 44.50% 51.25
230 kV Line#12 33 44.74% 59.75% 73.90
230 kV Line#13 24 15.53% 48.25% 45.11
230 kV Line#14 25 19.13% 48.25% 47.65
500 kV Line#15 18 27.04% 33.75% 42.99
500 kV Line#16 18 31.84% 29.25% 43.20
500 kV Line#17 23 40.66% 63.75% 73.84
500 kV Line#18 35 42.34% 70.50% 79.80
500 kV Line#19 12 23.36% 38.75% 43.93
500 kV Line#20 25 20.10% 44.25% 45.51

Table 9. Costs of 230 kV; practical experience in Thailand

Investment of new line (MTHB/km.) 4.29
230 kV double circuit, 1x1272 MCM ACSR/GA
Inflation rate (%) 3
MW loss of sale (MW) 100
Down time (hours) 3
Electricity Rate (THB/kWh) 2.513
Loss of reputation rate (THB/kWh) 0.5
206 C. Suwanasri et al.

Table 10. Cost-benefit analysis (MTHB/km) according to distance “d”

Age Line “d” Replaced Maintenance Loss of selling Reputation cost Risk-based
components’ cost cost (100 MW) (100 MW) maintenance cost
12 19 0.44 1.65 0.23 0.47 0.03 2.38
12 9 0.02 0.06 0.01 0.02 0.00 0.09
18 16 0.43 1.63 0.22 0.55 0.04 2.44
18 15 0.43 1.62 0.22 0.55 0.04 2.43
18 11 0.51 1.93 0.26 0.66 0.04 2.89
19 6 0.69 2.60 0.36 0.91 0.06 3.93
23 17 0.74 2.78 0.38 1.10 0.07 4.33
24 13 0.45 1.70 0.23 0.69 0.05 2.67
25 20 0.48 1.79 0.25 0.75 0.05 2.84
25 14 0.46 1.71 0.24 0.72 0.05 2.72
28 3 0.85 3.19 0.44 1.46 0.10 5.19
30 5 0.80 3.01 0.41 1.30 0.09 4.81
30 4 0.82 3.07 0.42 1.33 0.09 4.91
30 2 0.71 2.67 0.37 1.22 0.08 4.34
33 12 0.74 2.78 0.38 1.48 0.10 4.74
35 18 0.77 2.91 0.40 1.64 0.11 5.06
35 10 0.80 3.00 0.41 1.69 0.11 5.21
37 1 0.79 2.96 0.41 1.77 0.12 5.26
38 7 0.83 3.11 0.43 1.91 0.13 5.58
40 8 0.76 2.88 0.40 1.88 0.12 5.28

6 Conclusions

The innovative cost-benefit analysis based on risk assessment for OHLs in Thailand is
introduced. The practical diagnostic techniques and importance criteria for condition
and importance assessment are proposed. The weighting and scoring method and
procedure to determine risk-based maintenance cost are presented. Finally, the cost-
benefit analysis of individual OHL are analyzed. This proposed method and procedure
are the effective tools for transmission utility in order to compare the risk-based
maintenance cost with the new investment and make decision on continuous doing
maintenance tasks for the line or rebuilding the new OHL in scientific way.

Acknowledgement. Authors would like to acknowledge the Department of Research and

Development and Department of Transmission Network Maintenance, Electricity Generating
Authority of Thailand (EGAT) for technical support and funding.

References
1. CIGRE: Overview of CIGRE publications on asset management topics. CIGRE, Electra no.
262, June 2012
2. Neumann, C., Rusek, B., Balzer, G.: Optimization of maintenance, diagnostic technique and
design of substation equipment based on analysis of service experience. CIGRE B3-301 (2007)
 Cost-Benefit for HV Transmission Line Renovation and Replacement 207

3. Jeromin, I., Balzer, G., Backes, J., Huber, R.: Life cycle cost analysis of transmission and
distribution systems. In: 2009 IEEE Bucharest Power Tech Conference, Bucharest,
Romania, 28 June–2 July 2009 (2009)
4. CIGRE: Guide for the assessment of old cap and pin and long-rod transmission line
insulators made of porcelain or glass: what to check and when to replace? CIGRE, Electra
no. 63, October 2006
5. Wuller, A.I., Pharmatrisanti, A.: Transmission line assessment. In: IEEE International
Conference on Condition Monitoring and Diagnosis, Bali, Indonesia, 23–27 September 2012
(2012)
6. de Saboia Stephan, J.C., Ferreira Costa, C.: Uprate and upgrade of overhead transmission
lines methodologies and reliability. CIGRE B2-202 (2008)
7. Brennan, G.F.: Refurbishment of existing overhead transmission lines. CIGRE B2-203
(2004)
8. Chone, F., Gaudry, M., Parrotta, J., Gourit, L., Larripa, B.: Assessment of existing overhead
transmission lines and solutions for extending their residual lifetimes. CIGRE 22-201 (2000)
9. Van Der Wal, A., Ross, A.A.H.J.: Condition assessment of overhead lines. CIGRE B2-204
(2004)
10. Garcia-Alamo, C.J., Palacios, J.: Structural analysis for transmission lattice steel tower in the
400 kV transmission lines EI Tablazo - Cuatricenteneario No. 1 and 2. CIGRE B2-302
(2010)
11. Smirnov, A.A., Romanov, P.I.: Database for 330–750 kV transmission line insulation:
structure, data mining, use for the purpose of improvement of line operation. CIGRE 15-101
(2002)
12. Johngen, R.A., Gunski, E., Smit, J.J.: Statistical analysis of diagnostic-and life time data of
high voltage components. In: the 8th International Power Engineering Conference (IPEC
2007), Singapore, 3–6 December 2007 (2007)
13. Frimpong, G., Taylor, T.: Developing an effective condition based maintenance program for
substation equipment. In: Rural Electric Power Conference, Raleigh-Durham, NC, USA, 4–6
May 2003 (2003)
14. Saaty, T.L.: Relative measurement and its generalization in decision making: why pairwise
comparisons are central in mathematics for the measurement of intangible factors-the
analytic hierarchy/network process. Rev. R. Span. Acad. Sci. 102, 251–318 (2008)
 Study on Charge and Discharge Phenomenon
of Lithium Ion Battery Under High
Electric Field

S. Tsuji(&), Y. Fujita, H. Urushibata, A. Kono, M. Koyama,

and R. Hanaoka

Kanazawa Institute of Technology, 7-1 Ohgigaoka, Nonoichi,

Ishikawa 921-8501, Japan
b6800956@planet.kanazawa-it.ac.jp

Abstract. It aims to clarify the influence of high electric field on lithium ion
battery and to contribute to the establishment of appropriate use of it. In this
study, a lithium ion battery is installed in the vicinity of a spherical electrode
(diameter 100 mm) kept at a direct current ±10 kV and the presence or absence
of a voltage change of the lithium ion battery. (1) A gap (20 mm) is provided
between a needle electrode (tip radius of curvature 51 lm) attached to the
positive electrode of a lithium ion battery and a spherical electrode, and the
negative electrode is grounded. When applying +10 kV, the voltage rose by
about 15 mV in 300 s, and the lithium ion battery was charged. When −10 kV
was applied, the voltage dropped by about 10 mV in 300 s, and it was dis-
charged. (2) A gap (10 mm) is provided between the needle electrode attached
to the positive electrode of the lithium ion battery, and the spherical electrode,
and the needle electrode attached to the negative electrode, and the grounded
spherical electrode (diameter 100 mm). In this case, the lithium ion battery
voltage has changed. Therefore, Lithium ion battery was found to be affected by
high electric fields. It is presumed that the voltage of the battery changes because
corona discharge causes charge transfer in the gap and a chemical reaction of the
lithium ion battery occurs.

Keywords: Lithium ion battery
High electric field
Corona discharge

1 Introduction

Lithium ion batteries are widely used in portable devices because of high energy
density and power density per weight/volume. Furthermore, due to its installation in
electric vehicles and expansion of production volume, significant cost reduction is
expected [1]. Along with cost reduction, there is a possibility that lithium ion batteries,
which are originally low voltage devices, may be incorporated as electricity storage
devices in high voltage systems such as transmission and distribution [2]. However, the
influence when a lithium ion battery is installed in the vicinity of high-voltage
equipment has not been scientifically studied [3]. To clarify the influence of high
electric field on lithium ion battery and to contribute to the establishment of suitable
usage of lithium ion battery, we investigated voltage fluctuation phenomenon of

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 208–218, 2020.
https://doi.org/10.1007/978-3-030-31676-1_20
 Study on Charge and Discharge Phenomenon of Lithium Ion Battery 209

lithium ion battery in high electric field. In this experiment, the following two points
were examined in order to investigate the influence on the lithium ion battery in the
vicinity of the high voltage power supply. (1) Presence or absence of change in voltage
of lithium ion battery under the condition that spark discharge by high electric field
does not occur. (2) Relationship between voltage change of lithium ion battery and
polarity of corona discharge.

2 Experiments

2.1 Evaluation Method of Prototype Lithium Ion Battery

In this experiment, instead of a commercially available lithium-ion battery, we use a
self-made lithium-ion battery with a bipolar cell (Fig. 1). The reason for this is that
since it is smaller in capacity than commercially available, it is easy to detect the
voltage change and the risk is low for high voltage experiments. The structure of the
prototype lithium ion battery is shown in Fig. 2. Graphite containing lithium cobalt
oxide for the positive electrode and lithium for the negative electrode is used. A pro-
totype lithium ion battery is subjected to a charge and discharge test to evaluate its
performance. The discharge curve of the prototype lithium ion battery used in this
experiment is shown in Fig. 3. Depending on the discharge time of the battery, the
discharge capacity of the prototype lithium ion battery can be known. The current value
during discharging was set based on the C rate used for battery evaluation. C rate 1 C is
a current value that charges or discharges the theoretical capacity of the battery in 1 h.
In this experiment, 0.2 C (5 h discharge) generally used for specifying the battery
capacity is adopted [4]. Since the theoretical capacity of the prototype lithium ion
battery is 3.89 mAh, the current value is 0.778 mA. Constant current (CC) discharge
starts from the upper limit voltage 4.1 V (charging state 100%) of the lithium ion
battery. At this time, the voltage slowly decreases due to the internal resistance of the
battery and finally reaches the lower limit voltage of 2.7 V (charging state 0%) sud-
denly change. Thereafter, constant voltage (CV) discharge is started, and the current
value is reduced to 0. In this experiment, the discharge current was set to 5 h, but since
the discharge of the prototype lithium ion battery was completed in about 4.5 h, the
discharge capacity of the battery was 3.47 mAh. Since the measured capacity with
respect to the theoretical capacity is about 90% and the performance is close to that of
commercially available batteries, we decided to use this battery for the experiment [4].

10mm

Fig. 1. Appearance of bipolar cell

210 S. Tsuji et al.

Fig. 2. Internal structure of bipolar cell

Fig. 3. Discharge curve of prototype lithium ion battery

Fig. 4. Experimental circuit diagram (positive electrode side gap - negative electrode side
ground)
 Study on Charge and Discharge Phenomenon of Lithium Ion Battery 211

2.2 Voltage Measurement of Lithium Ion Battery (Positive Electrode Side

Gap - Negative Electrode Side Ground)
The experimental circuit diagram is shown in Fig. 4. Apply a DC high voltage
(±10 kV) to a spherical electrode with a diameter of 100 mm using a high voltage
power supply (Matsushika Precision, HAP - 10B10). Experiment is performed by
attaching a stainless steel needle electrode (tip curvature radius 51 lm: refer to Fig. 5)
to the positive electrode side of the lithium ion battery and changing the gap with the
spherical electrode to 10 mm, 15 mm, 20 mm. Connect the negative side of the lithium
ion battery directly to the ground and measure the voltage of the lithium ion battery at
that time with a digital multimeter. Figure 6 shows an experimental apparatus equipped
with a high voltage power supply and a lithium ion battery. For the order of experi-
ments, first charge the lithium ion battery to 2.8 V. Thereafter, the measurement with
the digital multimeter is started. The open circuit state is held in 0 to 300 s. Operate the
high voltage power supply at 300 to 600 s. Regarding the experimental environment,
measure the voltage of the lithium-ion battery with the temperature at 23 °C ± 0.5 and
the humidity at 50% or less.

μm

Fig. 5. SEM image of the tip of the needle electrode

Fig. 6. Experimental system (positive electrode side gap - negative electrode side ground)
212 S. Tsuji et al.

2.3 Voltage Measurement of Lithium Ion Battery (Positive Electrode

Gap - Negative Electrode Side Gap)
The experimental circuit diagram is shown in Fig. 7. Let the gap with the spherical
electrode connected to the high voltage power supply be 10 mm with the positive
electrode side of the lithium ion battery as the needle electrode. A similar needle
electrode is also connected to the negative electrode side of the lithium ion battery, and
the gap from the spherical electrode connected to the ground is set to 10 mm. The total
of both gaps is 20 mm. For the order of experiments, charge the lithium ion battery to
2.8 V, then start measurement with a digital multimeter. The open circuit state is held at
0 to 300 s. Operate the high voltage power supply at 300 to 600 s. However, directly
measuring high potential on a digital multimeter causes equipment failure. Therefore,
remove the digital multimeter while high voltage is applied, connect the digital mul-
timeter again to the lithium ion battery at 600 s, and measure the voltage. The
experimental environment is the same as in Sect. 2.2.

Fig. 7. Experimental circuit diagram (positive electrode side gap - negative electrode side gap)

3 Results and Discussion

3.1 Voltage Measurement Result of Lithium Ion Battery (Positive

Electrode Side Gap - Negative Electrode Side Ground)
The electric field becomes unequal, and it turns out that corona discharge is occurring
by becoming a high electric field. In the corona discharge, the state of occurrence of
discharge is different depending on the polarity of the voltage applied to the needle
electrode [5]. In this experiment, since a high voltage is applied to the spherical
electrode, a positive corona discharge is generated when a negative voltage is applied
relatively, and a negative corona discharge when a positive voltage is applied Become.
Figure 8 is a photograph of negative corona discharge when +10 kV is applied to the
 Study on Charge and Discharge Phenomenon of Lithium Ion Battery 213

spherical electrode. In this experiment, no apparent difference in corona discharge due

to the polarity of the applied voltage was observed. When a DC high voltage was
applied to the spherical electrode, it was confirmed that the tip of the needle electrode
shines as shown in Fig. 8.

20mm

Fig. 8. Photo of negative corona discharge (when +10 kV is applied)

First, the voltage change of the lithium ion battery when the polarity of the applied
voltage is changed when the gap is 20 mm is shown in Fig. 9. This is a graph when
±10 kV is applied. From 300 to 600 s, corona discharge occurred due to the high
electric field, which revealed that the voltage of the lithium ion battery changes with a
small amount. When +10 kV was applied, the voltage of the lithium ion battery was
increased by 14.5 mV, and the charging phenomenon of the battery occurred. The
current value flowing at this time is about 10 lA. When -10 kV was applied, the
voltage of the battery dropped 10.1 mV, and the discharge phenomenon of the battery
occurred. The current value flowing at this time is about 6 lA. It was found that the
voltage change slightly increased compared with the positive corona discharge for the
negative corona discharge. This may be due to the difference in reaction amount of
charge on the surface of the needle electrode. As shown in Fig. 10, when negative
corona discharge (+10 kV) is generated, neutral molecules are ionized at the tip of the
needle electrode. Electrons are attracted to the spherical electrode side and positive
charges are attracted to the needle electrode side [5]. In the case of negative corona
discharge, electrons are supplied to the needle electrode, so that recombination of
electrons and positive charges is frequently performed at the tip of the needle.
Therefore, at the time of negative corona discharge, since the flow of electrons becomes
active, it is expected that the current increases and the voltage change of the lithium ion
battery becomes large.
214 S. Tsuji et al.

Fig. 9. Voltage change of the lithium-ion battery due to difference in applied polarity
(positiveelectrode side gap - negative electrode side ground)

Fig. 10. Negative corona reaction model at the tip of the needle electrode (when +10 kV is
applied)

Next, the voltage change of the lithium-ion battery when the gap is changed to
10 mm, 15 mm, 20 mm with constant applied voltage is shown in Fig. 11. This is a
graph when +10 kV is applied. It was found that the voltage change becomes larger as
the gap becomes smaller [6]. Figure 12 shows the impedance of the gap. This impe-
dance was derived from the applied voltage and the circuit current. From Fig. 12, it can
be confirmed that the impedance becomes smaller as the gap becomes smaller.
Therefore, when the gap becomes smaller, the impedance becomes smaller and the
current flowing in the lithium ion battery becomes larger, so it is considered that the
voltage change of the battery increases. Also, when comparing the positive corona
discharge and the negative corona discharge, the impedance is small at the time of the
negative corona discharge, and it can be said that a larger current flows.
 Study on Charge and Discharge Phenomenon of Lithium Ion Battery 215

Fig. 11. Voltage change of lithium ion battery due to gap difference (when +10 kV is applied)

Fig. 12. Relation between gap and impedance

3.2 Voltage Measurement Result of Lithium Ion Battery (Positive

Electrode Side Gap - Negative Electrode Side Gap)
In Sect. 3.1, it was confirmed that the voltage of the battery varies due to the high
electric field with the negative electrode side of the lithium ion battery directly con-
nected to the ground. In this section, the voltage change of the battery when there is a
gap at both ends of the lithium ion battery was verified. Figure 13 shows the voltage
change of the lithium ion battery when ±10 kV is applied. Even when there was a gap
at both ends of the lithium ion battery, it was confirmed that the voltage of the lithium
ion battery was affected. The change in the voltage of the lithium-ion battery was about
5 mV when applying ±10 kV, respectively, and it was found that they are almost the
same value even if the polarity of the applied voltage is different. The current value
216 S. Tsuji et al.

flowed at this time is 3 lA. Since the sum of the gaps at both ends is 20 mm, it is
compared with the current value of the gap of 20 mm when the negative electrode side
of the lithium ion battery is directly connected to the ground. It was found that the
current value when there is a gap at both ends of the battery is smaller by about 3 to
7 lA. This is thought to be due to the fact that the place where corona discharge occurs
increases, electrons are released into the atmosphere, and the current has decreased [7].

Fig. 13. Voltage change of lithium ion battery (positive electrode side gap - negative electrode
side gap)

Consider the charging phenomenon when the voltage of the battery increases by
corona discharge when +10 kV is applied to the spherical electrode. As shown in
Fig. 14, negative corona discharge occurs on the positive electrode side of the lithium
ion battery. As a result, positive charges extract electrons inside the lithium ion battery
and recombine. As the electrons are withdrawn, lithium ions move from the positive
electrode to the negative electrode via the separator. At this time, positive corona
discharge occurs on the negative electrode side of the lithium ion battery, and ionized
electrons are supplied to the lithium ion battery. The electron and lithium ion are
bonded at the negative electrode. As this process continues, it is considered that
charging reaction occurs [8]. When −10 kV is applied to the spherical electrode, a
discharge reaction opposite to the charging reaction described above occurs. In the gap
on the high voltage electrode side, positive corona discharge occurs, and in the gap on
the ground electrode side, negative corona discharge occurs. Section 3.1 explained that
there is a difference in the magnitude of the current depending on the polarity of the
corona discharge. Corona discharge of different polarity occurred in two gaps when
there was a gap at both ends of the battery, so it is considered that the voltage change of
the lithium ion battery became almost the same value. From the above, it is believed
that charge transfer occurs due to corona discharge, which causes a chemical reaction
of the lithium ion battery.
 Study on Charge and Discharge Phenomenon of Lithium Ion Battery 217

Fig. 14. Charging phenomenon model diagram by corona discharge (when +10 kV is applied)

4 Conclusions

In this experiment, the following three points were clarified.

(1) The corona discharge at the tip of the needle electrode causes a chemical reaction
inside the battery, resulting in a charge and discharge reaction of the lithium ion
battery.
(2) As the negative corona discharge has a larger current value than the positive
corona discharge, the voltage change of the battery increases.
(3) When there is a gap at both ends of the lithium-ion battery, the battery voltage
changes due to the high electric field.
Under the conditions of this experiment, the voltage change of the lithium ion
battery is about 10 mV, the change is small, and the current flowing through the battery
is also very small. Therefore, it is considered that the effect of directly deteriorating the
lithium ion battery under high electric field is small. However, it is expected that even if
the battery is not connected in terms of electric circuit, leaving the lithium ion battery in
a high electric field for a long time will cause over charge and over discharge of the
battery.

Acknowledgment. This research was funded by JSPS Grants-in-Aid for Research and Devel-
opment 26289080, 18K04089.

References
1. Ishikawa, H., Umeda, M., Sone, Y., Ucida, I.: Comparison of apparent thermal deterioration
and charge/discharge activation engergies of commercial Lithium-ion secondary cell. Bunseki
Kagaku 62(11), 1029–1035 (2013)
2. Yoshino, A.: Lithium-ion Battery 15 Years and Future Technology, pp. 1–37 (2016)
3. Hata, T.: Effect of high electric field on charging level of power storage device with spherical
electrodes. In: 20th International Symposium on High Voltage Engineering (2017)
218 S. Tsuji et al.

4. Mukai, D., Kurahashi, T., Hashizaki, K., Kobayashi, K., Matsueda, N., Kogure, M.:
Development of high performance and large-sized Lithium-ion battery for energy storage and
industrial uses. Mitsubishi Heavy Industries Technical report, vol. 49, no. 1, pp. 8–9 (2012)
5. Hanaoka, R., Ishida, T.: High Voltage Pulse Power Engineering, pp. 79–86 (2014)
6. The Institute of Electrical Engineers of Japan. Discharge Handbook First Volume, pp. 139–
141 (1998)
7. Sekimoto, K., Takayama, M.: Study of ion formation and evolution in ambient corona
discharge. Earozoru Kenkyu 26(3), 203–213 (2011)
8. Saida, T., Watanabe, R.: Lithium-ion battery model that simulates performance deterioration
by discharge and charge. Proc. Automob. Eng. Assoc. 43(6), 1287–1288 (2012)
 Online and On-site Partial Discharge
Measurement of Long Length Power Cables
by Using Joints with Integrated PD Sensors

D. Passow1(&), M. Beltle1, S. Tenbohlen1, J. Hohloch2,

and R. Grund3
1
University of Stuttgart, Stuttgart, Germany
daniel.passow@ieh.uni-stuttgart.de
2
Pfisterer Switzerland AG, Altdorf, Switzerland
3
Pfisterer Kontaktsysteme GmbH, Winterbach, Germany

Abstract. This contribution investigates integrated capacitive Partial Discharge

(PD) sensors in joints. Firstly, the required sensor bandwidth, sensitivity and
selectivity are analyzed in context with long length power cable installations and
the unavoidable influence of phenomena like attenuation and dispersion on PD
pulse propagation along the cable. Secondly, the principle of an integrated
capacitive PD sensor is discussed. Furthermore, a model for the integrated
capacitive PD sensor is introduced based on material properties and geometry
considerations deduced from actual joints. Different parameters like the relative
permittivity and the specific resistance per meter are obtained in advance from
measurements within the targeted bandwidth. In the next step, an early prototype
of a joint with an integrated capacitive PD sensor is built and analyzed in a
laboratory setup. Therefore, the transfer function of the integrated capacitive PD
sensor is determined using S-Parameter measurements. Subsequently, the
modelling approach and simulation results are compared with the obtained
measurement data. Based on these findings, the previously developed sensor
model is verified. Additionally, the measurement sensitivity of the integrated PD
sensor is determined using calibration measurement with an IEC 60270 conform
calibrator where the integrated capacitive PD sensor is used as an external
coupling capacitor. Finally, the findings in this work will be summarized.

Keywords: Partial discharge
Cable
Joint

1 Introduction

There are different reasons, which can lead to an insulation breakdown in a power cable
system, e.g. transient overvoltages due to lightning or switching surges, mechanical
stresses, thermal and environmental stress or normal operational voltage in already
degraded insulating systems. In most cases prior to the actual failure partial discharges
(PDs) will occur as result of localized field stress or material degradation. Due to the
material properties of commonly used insulating materials like cross-linked poly-
ethylene (XLPE) or silicone rubber (SIR) PDs will permanently damage the insulating
system, starting an irreversible erosion process [1–3]. In order to find insulation defects

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 219–231, 2020.
https://doi.org/10.1007/978-3-030-31676-1_21
220 D. Passow et al.

before an insulation breakdown occurs on-site PD tests or PD monitoring systems

should be applied [4, 5]. With growing lengths of future power cable installations, e.g.
the planned HVDC cable connection Suedlink in Germany or several other long length
cable connection projects in Europe, conventional PD measurement systems will face
new challenges due to the high cable capacities and subsequent attenuation and dis-
persion during PD pulse propagation. It has been shown that measurement sensitivity
and localization performance in long length power cable PD measurement can be
improved using a two-sided PD measurement system [6]. This indicates that multipoint
PD measurement with additional PD sensors, e.g. installed in joints, will enable further
improvements in terms of measurement sensitivity and localization performance. To
obtain detailed localization information, a multipoint PD measurement system should
synchronize all measurement points. Different sources show that synchronization can
be achieved using GPS or injected synchronization pulses [7, 8]. In Sect. 2 the influ-
ence of different aspects like attenuation and dispersion on the PD pulse propagation
and the required frequency bandwidth for on-site and online PD measurement systems
will be discussed. In Sect. 3 an integrated capacitive PD sensor for joints will be
introduced and a model for this sensor will be proposed. The proposed model is then
verified through measurement data acquired from an early laboratory prototype of the
integrated capacitive PD sensor. Finally, Sect. 3 will be concluded by evaluating the
capability of the integrated capacitive PD sensor for online and on site PD
measurement.

2 Frequency Dependent Signal Propagation

2.1 Influence of Attenuation and Dispersion
Signal propagation in power cables is influenced by attenuation and dispersion [9–11].
Both phenomena lead to a frequency dependent signal propagation of the different
frequency components contained in a PD signal. Attenuation is caused through
dielectric losses in the insulating material and through the losses caused by the prop-
agation of the radial displacement current in the semiconducting layers [9, 10]. Dis-
persion describes the effect of variation in propagation velocity with frequency. In
direct comparison of both phenomena attenuation is more severe [11, 12]. In Fig. 1 the
attenuation behavior of a 110 kV XLPE-cable is shown.
Figure 1 shows a linear increase of the attenuation with frequency. In addition,
attenuation can be scaled with cable length, so it is obvious that in case of power cables
with lengths of several tenths of kilometres attenuation will become a significant factor.
In case of frequencies above a few tenths of MHz even just a several hundred meters of
cable can cause a severe pulse amplitude reduction [13].
 Online and On-site Partial Discharge Measurement of Long Length Power Cables 221

Fig. 1. Measured attenuation behavior of a 1 m piece of 110 kV XLPE-cable

2.2 Required Bandwidth for PD Measurement

By comparing the attenuation curve in Fig. 1 with the recommended range of PD
measurement frequencies in IEC 60270 [14] it is conclusive, that there is little to almost
none attenuation in the range from 30 kHz up to 1 MHz. Even though high frequencies
are attenuated severe PD measurement systems for on-site and online measurement
should be as broadband as possible in order to avoid background noises [4]. Consid-
ering the above-mentioned fact, that frequencies above a few tenths of MHz will result
in severe attenuation, the bandwidth of the integrated PD sensor is set to a bandwidth
from 30 kHz to 30 MHz.

3 Integrated Capacitive Sensor: Principle, Modelling

Approach, Sensor Validation

The integrated capacitive PD sensor needs to fit into the general structure of a joint
without changing the dielectric properties and electric field distribution inside the joint.
Figure 2 shows the general structure of a cross bonding joint with the screen inter-
ruption at the right side. At the edge between the insulating body and the screen
different kinds of semiconducting layers are included to smoothen the distribution of
the electric field stress. This type of junction will be used to integrate an additional
capacitive layer, forming the capacitive PD sensor.
222 D. Passow et al.

XLPE-insulation
screen insulating body (SIR)
field control layer screen interruption

connector stress cone

conductor conductive layers

Fig. 2. Structure of a joint with screen interruption

The detailed structure of the integrated capacitive PD sensor is shown in Fig. 3.

The additional capacitive layer consists of a copper electrode, which is applied on top
of the semiconducting layers and is surrounded with an exterior insulating material
forming a gap between the copper electrode and the metallic screen of the joint. This
gap separates the screen from the copper electrode, resulting in an additional capaci-
tance C2 between the electrode and the screen.

connector
conductive layer 2 exterior insulating material
copper electrode
screen
C2

C1

insulating body field control layer conductive layer 1

Fig. 3. Rotational symmetric cut through a joint with an integrated capacitive sensor

This kind of sensor application is possible because the semiconducting layers will
terminate the 50 Hz AC component of the electric field but are permeable for higher
frequencies. As a result the insulating system will remain intact at the operational
frequency forming the so called power frequency (PF) ground. For high frequency
(HF) components, e.g. PD pulses, the conductivity of the semiconducting layers is too
low to terminate the HF components of the electric field. In this case the ground
potential will be represented by the metallic screen of the joint resulting in capacitive
coupling across C2. Side note: Besides the use of capacitive coupling sensors, direc-
tional coupler sensors can also be applied to cables and joints [15, 16].
 Online and On-site Partial Discharge Measurement of Long Length Power Cables 223

3.1 Modelling Approach for the Integrated Capacitive Sensor

The basic equivalent circuit diagram in Fig. 4 can be used in order to model the
integrated capacitive PD sensor. In particular, the quality of the equivalent circuit
diagram representation depends on the material properties of the semiconducting lay-
ers, which are here represented as R1 and the relative permittivity
of the exterior
insulating material, influencing C2.

C1
IPD
U
R1 C2 Umeas

Fig. 4. Equivalent circuit diagram of the integrated capacitive sensor

To obtain an equivalent circuit model of sufficient quality it may be needed to

substitute the representation of the semiconducting layers R1 with the equivalent circuit
diagram for semiconducting materials proposed in [17, 18] and also model a frequency
dependent permittivity. In reality, the capacitive PD sensor is surrounded by the joint
geometry to the left and the right side of the sensor. Thus, not only sensor itself needs
to be modelled but also the joint. Figure 5 shows the equivalent circuit diagram of joint
and sensor.

Lleft Rleft L1 Lright Rright

C1

Gleft Cleft Gright Cright

R1 C2

joint and cable at the capacitive sensor, joint and cable at the
left-hand side from field area above field right-hand side from
control layer control layer field control layer

Fig. 5. Equivalent circuit diagram of joint and sensor

As other modelling approaches show both the transmission line representation of

the joint at the left hand and the right hand side to the sensor can get very complex and
is depending on frequency range and joint geometry [19, 20]. In this case, a more
complex modelling approach is neglected due to the relative low frequency range.
224 D. Passow et al.

Obtaining the Frequency Dependent Parameters of the Semiconducting Layers

and the Relative Permittivity
To model the joint and sensor according to the modelling approach presented in the
previous section C2 and R1 need to be determined. For this purpose, a Vector Network
Analyzer (VNA) and S-Parameter measurement are used. The relative permittivity of
the used exterior insulating material was determined in compliance with the mea-
surement setup proposed in IEC 62631 [21]. Figure 7 shows the relative permittivity
over frequency. It can be noted, that the relative permittivity slightly decreases with
frequency.

Fig. 6. Measurement of the relative permittivity exterior insulating material from 1 kHz to
30 MHz

Because, the applied semiconducting layers consist on the one hand of a semi-
conducting tape which is wrapped around the joint and on the other hand of a semi-
conductive paint, the conductivity of the semiconducting layers was measured using a
an assembled joint to include the effect of variation of the semiconducting layer
thickness.
From Fig. 7 it can be obtained, that the semiconducting layer 2 can be modelled by a
pure resistance as proposed in Fig. 4, the slight raise of the impedance at the end of the
frequency range is due to the inductance of the measurement setup. For semiconducting
layer 1, it can be stated, that a simple representation with a resistance won’t fit. To
determine the exact material properties of semiconducting layer 1, a different
 Online and On-site Partial Discharge Measurement of Long Length Power Cables 225

measurement setup will be needed, which will be part of future work. This contribution
focuses on the modelling of the capacitive sensor using semiconducting layer 2, which
is on top of semiconducting layer 1. Due to parallel application of both layers semi-
conducting layer 2 will dominate.

Fig. 7. Measurement of the impedance per meter of the semiconducting layers

3.2 Transfer Function Measurement of the Capacitive Sensor

and Validation of the Proposed Model
The measurement setup, which is used in this section, is shown in Fig. 8. The con-
ductor is replaced by an analogue solid aluminum cylinder replication. The capacitive
sensor has been applied as shown in Fig. 3. For the transfer function measurement the
high-ohmic low frequency (LF) ports of the VNA were used. Port T was applied to the
copper electrode as shown in Fig. 8. The obtained measurement data is compared with
results of a spice simulation using the equivalent circuit model proposed in Fig. 5.
Therefore, the resistance R1 of the semiconducting layer was calculated from the in
advance obtained conductivity measurement data. In order to model the capacitance C2
the frequency dependent permittivity from Fig. 6 was averaged over frequency.
226 D. Passow et al.

VNA

LF-Out T R

point of C2
measurement R1

C1

Rtermination

Fig. 8. Transfer function measurement setup

Fig. 9. Comparison between measurement and simulation for the transfer function behaviour of
the capacitive sensor

Figure 9 shows the comparison of the simulation result and the transfer function
measurement of the integrated capacitive PD sensor. It can be seen, that the transfer
function curves increases with approximately 20 dB/decade until the cut-off frequency
fc around 1 MHz showing a purely capacitive behavior. Besides small differences in
 Online and On-site Partial Discharge Measurement of Long Length Power Cables 227

amplitude, the simulation result resembles the measurement data. This justifies the
approach to obtain the actual transfer function of the capacitive sensor by simulation
and validates the proposed model.

3.3 Parameter Variation and Their Influence on the Transfer Function

of the Integrated Capacitive PD Sensor
The previous section showed that the equivalent circuit diagram in Fig. 4 can be used
to describe the behavior of the integrated capacitive PD sensor. From this observation
the transfer function of the integrated capacitive PD sensor can be calculated and
transformed leading to the cut-off frequency fc and the amplitude factor a.

fc ¼ 1=ð2pR1 ðC1 þ C2 ÞÞ ð1Þ

a ¼ C1 =ðC1 þ C2 Þ ð2Þ

The capacitance per meter of C1 can’t be changed because the distance from the
connector to the copper electrode of the sensor is fixed. Hence, either the resistance of
the semiconducting layers or the thickness of the exterior insulating material need to be
changed to influence the behaviour of the transfer function. Reducing the value of C2
will lead to a higher amplitude and will result in a slight increase of the cut-off
frequency fc. Due to geometric limitations, the value of C2 will not change in multiple
orders of magnitude. As a result the influence for the cut-off frequency fc can be
neglected. Whereas the influence on the amplitude factor a will be more severe,
because in the assembled setup C2  C1. Since the exterior insulating material consists
of a number of wrapped layers of tape, the value of C2 the can be changed easily to
verify the previous assumptions.
Figure 10 shows the measurement data for 5 different numbers of layers, starting
with a thickness of 4 layers increasing the number to 12. As it can be seen from Fig. 10
the measurement data confirms the previous made assumptions, that a change of the
capacitance C2 mainly influences the amplitude a of the transfer function. Whereas a
change of the capacitance C2 has only a minor effect on the cut-off frequency fc.
To set and change the cut-off frequency, e.g. set it to lower frequencies, the
resistance R1 of the semiconducting layer needs to be increased. This can be achieved
by applying a different kind of semiconducting layer to the joint or by removing the
semiconducting layer 2, which will leave only semiconducting layer 1 on the joint
surface. Figure 11 shows results of the transfer function measurement, comparing the
measurements with and without semiconducting layer 2. It is obvious, that removing
semiconducting layer 2 leads to a reduction of the cut-off frequency. As a result, the
integrated capacitive PD sensor offers a more broadband measurement capability.
228 D. Passow et al.

Fig. 10. Comparison of the transfer function behavior for different numbers of insulating layers

Fig. 11. Comparison of the transfer function behavior for different semiconducting layers
 Online and On-site Partial Discharge Measurement of Long Length Power Cables 229

3.4 Sensitivity Check of the Integrated Capacitive PD Sensor

As was shown in previous sections, the integrated capacitive sensor seems to be
appropriate for measuring PD. To demonstrate the PD measurement capability of the
sensor in a practical setup, a sensitivity check will be performed, using a setup similar
to Fig. 8. For the sensitivity check an IEC 60270 conform PD calibrator and an
Omicron MPD 500 were used to investigate the sensor sensitivity. The calibrator was
applied to one side of the solid aluminum conductor, whereas the integrated capacitive
PD sensor was used as coupling capacitance. For the sensitivity check a sensor setup
without semiconducting layer 2 was used. Two different center frequencies where
chosen, f1 = 500 kHz and f2 = 10 MHz. The bandwidth was set to Df = 800 kHz.
Figures 12 and 13 show the measurement results of the sensitivity check. For fre-
quencies in the IEC60270 measurement frequency range it’s possible to distinguish a
PD-level of 2 pC from the noise level (Fig. 12). For frequencies around 10 MHz the
amplitude of the transfer function increases Fig. 10, hence the signal to noise ratio
increases. This leads to distinguishable PD-level of 1 pC. Since in most cases insu-
lation failures happen in or near to joints, this type of PD sensor and the detection
sensitivity from 1 pC is well suited for on-site PD measurement or PD online
monitoring.

Fig. 12. Sensitivity check result at f1 = 500 kHz calibrated to 5 pC

230 D. Passow et al.

Fig. 13. Sensitivity check result at f2 = 10 MHz calibrated to 5 pC

4 Conclusion

The proposed modelling approach for the entire system consisting of sensor and joint
has been validated by measurement. The influencing parameters in terms of cut-off
frequency fc and bandwidth as well as sensor output amplitude have been identified.
Finally, the practical suitability of integrated capacitive sensor for PD measurement
was demonstrated by applying a sensitivity check to the sensor. The minimum
observable PD-level, which is distinguishable from noise is 1 pC. Besides this, the
sensitivity check showed the capability of the sensor to measure PD in the IEC 60270
frequency range and also at higher frequencies, offering a sensor bandwidth from
30 kHz up to 30 MHz. As a result, this allows a coarse identification of the PD-source
location by examining the measured frequency components.

References
1. Crine, J.: A molecular approach to electrical aging of XLPE Cables. In: Jicable (2007)
2. Harlin, A., Danikas, M.G., Hyvönen, P.: Polyoelfin insulation degradation in electrical field
below critical inception voltages. J. Electr. Eng. 56, 135 (2005)
3. IEEE: 400-2012 - IEEE guide for field testing and evaluation of the insulation of shielded
power cable systems rated 5 kV and above (2012)
4. Cigre Working Group B1.28. On-site partial discharge assessment of HV and EHV cable
systems. Cigre Technical Brochure 728 (2018)
5. Cigre Working Group B1.04. Maintenance for HV cables and accessories. Cigre Technical
Brochure 279 (2005)
 Online and On-site Partial Discharge Measurement of Long Length Power Cables 231

6. Wild, M., Tenbohlen, S., Gulski, E., Jongen, R.: Basic aspects of partial discharge on-site
testing of long length transmission power cables. IEEE Trans. Dielectr. Electr. Insul. 24,
1077–1087 (2017)
7. Mohamed, F.P., Siew, W.H., Soraghan, J.J., Strachan, S.M., McWilliam, J.: Partial
discharge location in power cables using a double ended method based on time triggering
with GPS. Trans. Dielectr. Electr. Insul. 20, 2212–2221 (2013)
8. Li, Y., Wouters, P.A.A.F., Wagenaars, P., van der Wielen, P.C.J.M., Steennis, E.F.:
Temperature dependent signal propagation velocity: possible indicator for mv cable dynamic
rating. IEEE Trans. Dielectr. Electr. Insul. 22, 665–672 (2015)
9. Boggs, S., Pathak, A., Walker, P.: Partial discharge. XXII. High frequency attenuation in
shielded solid dielectric power cable and implications thereof for PD location. IEEE Electr.
Insul. Mag. 12, 9–16 (1996)
10. Xu, C., Zhou, L., Zhou, J., Boggs, S.: High frequency properties of shielded power cable -
part 1: overview of mechanisms. IEEE Electr. Insul. Mag. 21, 24–28 (2005)
11. Shu, E., Boggs, S.: Effect of dispersion on PD pulse propagation in shielded power cable. In:
Conference on Electrical Insulation and Dielectric Phenomena, Quebec, QC, Canada (2008)
12. Shu, E., Boggs, S.: Dispersion and PD detection in shielded power cable. IEEE Electr. Insul.
Mag. 24, 25–29 (2008)
13. Oussalah, N., Zebboudj, Y., Boggs, S.: Partial discharge pulse propagation in shielded power
cable and implications for detection sensitivity. IEEE Electr. Insul. Mag. 23, 5–10 (2007)
14. IEC 60270. High-voltage testing - partial discharge measurement (2016)
15. Pommerenke, D., Strehl, T., Heinrich, R., Kalkner, W., Schmidt, F., Weissenberg, W.:
Discrimination between internal PD and other pulses using directional coupling sensors on
HV cable systems. IEEE Trans. Dielectr. Electr. Insul. 6, 814–824 (1999)
16. Craatz, P., Plath, R., Heinrich, R., Kalkner, W.: Sensitive on-site PD measurement and
location using directional coupler sensors in 110 kV prefabricated joints. In: Eleventh
International Symposium on High Voltage Engineering, London, UK (1999)
17. Heinrich, R., Bonisch, S., Pommerenke, D., Jobava, R., Kalkner, W.: Broadband
measurement of the conductivity and the permittivity of semiconducting materials in high
voltage XLPE cables. In: Eighth International Conference on Dielectric Materials,
Measurements and Applications, Edinburgh (2000)
18. Wang, X., Yu, D., Wang, H., Duan, S., Lao, Y., Wu, K.: Influence of the characteristics of
semiconducting materials in high frequency on electric-field distribution in cable accessory.
In: 12th International Conference on the Properties and Applications of Dielectric Materials
(ICPADM), Xi’an (2018)
19. Hou, Z., Li, H., Liu, B., Chen, X., Sun, Z., Ji, S.: A novel high frequency circuit model of
power cable joint. In: International Conference on Condition Monitoring and Diagnosis
(CMD), Xi’an (2016)
20. Li, Y., Wouters, P.A.A.F., Wagenaars, P., van der Wielen, P.C.J. M., Steenis, E.F.: Power
cable joint modeling high frequency. In: IEEE International Conference on Condition
Monitoring and Diagnosis, Bali (2012)
21. IEC 62631-2-1. Dielectric and resistive properties of solid insulating materials - Part 2-1:
relative permittivity and dissipation factor - Technical Frequencies (0,1 Hz - 10 MHz) - AC
Methods (2018)
 Development of Support Program
for Managing Assets by Considering
Regular Maintenance Cost and Statistically
Expected Failure Cost

Tsuguhiro Takahashi(&)

Central Research Institute of Electric Power Industry (CRIEPI),

2-6-1 Nagasaka, Yokosuka-shi 240-0196, Japan
shodai@criepi.denken.or.jp

Abstract. The asset management technique, which was developed in the field
of financial engineering, is expected to rationalize business operations. There
have been some previous investigations on introducing this technique into the
maintenance divisions of electric power transmission and distribution facilities
in Europe. Therefore CRIEPI (Central Research Institute of Electric Power
Industry) is examining its effective introduction into Japan in the last about 15
years by referring to these investigations. In this paper, the evaluation is in terms
of cost. Two types of cost, the maintenance cost in ordinary operation and the
failure-handling cost including a penalty cost of electric outage, are considered
in this investigation. The basic concept and a support program used to select
better maintenance scenarios for an oil-immersed power transformer are
described in this paper. The candidates are the successive operation of aged
equipment by repairing it whenever possible or by large-scale repair (overhaul)
until expected lifetime, and replacement by new, low-maintenance-cost equip-
ment immediately or at a certain age. The cumulative maintenance cost is cal-
culated for each scenario as the NPV (Net Present Value). In replacement
scenarios, the depreciation is considered for realism. At present, the maintenance
cost characteristic and the failure probability by age, which should be input to
the support program as parameters, are under investigation. The method and
difficulties are also described.

Keywords: Asset management
Managing assets
Maintenance cost
Failure

cost
Depreciation

1 Introduction

Recently, the electric power industry has been facing of difficult problems, such as the
strong social request for cost cutting and the unprecedented aging of facilities. Addi-
tionally, accountability to prove the validity of any investment will become much more
important owing to the “Electricity Systems Reform” next year in Japan. The asset
management technique, which has been developed in the field of financial engineering,
is expected to help solve these problems. There are some preceding investigations on
introducing this technique into the maintenance divisions of electric power
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 232–242, 2020.
https://doi.org/10.1007/978-3-030-31676-1_22
 Development of Support Program for Managing Assets 233

transmission and distribution facilities in Europe. For example, the progress of the
aging was pointed [1], the maintenance priority was examined by considering
importance and condition of each equipment [2], and the concept of the asset man-
agement for electric power equipment was discussed [3, 4]. Some concrete programs
for the asset management were also proposed ([5, 6], for example). Therefore, CRIEPI
is examining its effective introduction to Japan in the last about 15 years by referring to
these investigations [7]. It has been founded that actual maintenance-related data
should be collected, and a concrete evaluation should be carried out on the basis of
these data, because individual situations are generally different depending on the
equipment and sometimes on the utility company. The author has been collecting such
data by consulting with utility companies and other organizations, and developing
concrete support programs for maintenance strategies [7–9]. In such support programs,
possible maintenance scenarios should be defined and compared, and actual mainte-
nance data and possible maintenance strategies have to be examined. There are some
approaches to rationally managing assets. A relative evaluation by condition scoring
([5, 6, 10], for example,) is useful to a certain extent, but among different equipment or
different divisions in a company, a quantitative evaluation with a cost comparison is
necessary. Additionally, accountability to society comes to be very important recently.
A black-box evaluation cannot be accepted, and a simple and easy to understand
evaluation method is required. In this paper, the evaluation is carried out in terms of
cost with a very simple model. Two types of cost: the maintenance cost in ordinary
operation and the failure-handling cost including a penalty cost of electric outage, are
considered in this investigation. The former is a real cash flow and the latter is the
statistically expected cost. The statistically expected cost is not an actual cash flow, but
by considering it, the importance of the equipment and the influence when failure
occurs can be evaluated. The basic concept and a support program used to select better
maintenance scenarios for an oil-immersed power transformer are described.

2 Cost Evaluation Model for Oil-Immersed Transformer

Power transformers are one of the main facilities in substation division. Most of them
are oil-immersed type, whose reliability is sufficiently high for more than 30 years’
operation. On the other hand, there is some deterioration in the insulation paper,
pressboard and gaskets; therefore, appropriate repair and premeditated renewal should
be scheduled.

2.1 Maintenance Expenses in Ordinary Operation

In ordinary operation, there are some necessary maintenance expenses. The author
proposes their classification into the three kinds from the viewpoint of their relation to
aging.
(a) A repair cost that increases with age
(b) Expense for a periodic inspection
(c) Expense for a non-periodic inspection, such as a legal inspection
234 T. Takahashi

Expense (a) is assumed to be proportional to the age for simplicity, as in a past

examination [1].

2.2 Statistically Expected Failure Cost

The expense required for renewal with a failure is introduced in the subsequent section,
together with that for a premeditated renewal. The “failure cost” is defined as the
additional expenses excluding the installation cost of new equipment in this paper. The
“statistically expected failure cost”, which is the product of the “failure cost” and the
failure probability, is accumulated to LCC (Life Cycle Cost). This cost is not a real cash
flow but can be used to consider the importance of the equipment by the increasing
failure cost.
The failure cost should include the lost revenue from selling electricity, the
emergent recovery cost, a penalty resulting from service interruption, and so on. Its
estimation is generally difficult, and it is regarded as a parameter in this program.
The failure probability corresponding to the aging property is generally found by
statistical analysis (hazard analysis) based on actual accidents. It is generally fitted by a
Weibull distribution, whose shape parameter and scale parameter are specified in this
program. One point should be pointed here for these parameters obtained from failure
results. The present basic approach to maintenance is preventive and failure results
should be small. For example, it is difficult to estimate a change in the failure proba-
bility, which is expressed by Weibull parameters, when the maintenance budget is
reduced and maintenance activities decrease in the future.

2.3 Installation Cost of Renewed Equipment

The premeditated and accidental renewal cost is not spent in a single year, but by
instalments over several years, by considering depreciation. For example, for a trans-
former costing 100 million yen in Japan, the instalments are shown in Table 1. The
property tax should also be considered during these years.

2.4 Pending Problem of Individual Equipment

Some faults with various degrees of severity that need to be repaired may be found
during periodic inspections. In this program, the degree of severity is divided into four
ranks depending on the urgency as follows:
S rank : Immediate repair is required.
A rank : Repair should be made in a couple of years.
B rank : Repair can be postponed to the next periodic inspection or so-called
“overhaul” (the repair expenses are assumed to increase in proportion to the
number of postponed years).
C rank : Repair is made at next periodic inspection or overhaul only when there are
sufficient operation years until the planned renewal
 Development of Support Program for Managing Assets 235

Table 1. Depreciation and property tax for transformer

costing 100 million yen (depreciable life: 22 years,
depreciation rate: 0.091, guaranteed rate: 0.03182)
Age Terminal Depreciation Property
(years) value (Yen) charge (Yen) tax (Yen)
0 100,000,000
1 90,900,000 9,100,000 763,560
2 82,628,100 8,271,900 694,076
3 75,108,943 7,519,157 630,915
4 68,274,030 6,834,913 573,501
5 62,061,094 6,212,936 521,313
6 56,413,535 5,647,559 592,342
7 51,279,904 5,133,631 538,438
8 46,613,433 4,666,471 489,441
9 42,371,611 4,241,822 444,901
10 38,515,795 3,855,816 404,415
11 35,010,858 3,504,937 490,152
12 31,824,870 3,185,988 445,548
13 28,642,383 3,182,487 400,993
14 25,459,896 3,182,487 356,438
15 22,277,409 3,182,487 311,883
16 19,094,922 3,182,487 267,328
17 15,912,435 3,182,487 222,774
18 12,729,948 3,182,487 178,219
19 9,547,461 3,182,487 133,664
20 6,364,974 3,182,487 89,109
21 3,182,487 3,182,487 44,554
22 1 3,182,486 0

3 Maintenance and Renewal Planning Support Program

3.1 Maintenance Scenarios to Be Compared
Generally, when the asset management technique is utilized for maintenance and
renewal planning, the possible scenarios of maintenance should be examined before-
hand. LCC is evaluated for each scenario. Therefore, this scenario setting is important
in this support program.
To examine the average maintenance plan, time-based renewal scenarios (such as at
40, 50, and 60 years) with and without overhaul (OH) are considered. The OH is
assumed to rejuvenate a transformer with a certain cost. Its effect (rejuvenation years),
cost and timing are specified as parameters.
236 T. Takahashi

In addition, to examine an individual maintenance plan by considering each

pending problem, immediate renewal without repairing and continuous operation with
repairing are compared.

3.2 Change in Aging Distribution

In this program, when some aged transformers in service are removed as a result of
failure and premeditated renewal, the same number of new transformers is assumed to
be installed. That is, the total number of facilities does not change. This means that the
aging distribution changes over time.
For example, enough many new same transformers whose failure probability and
hazard rate are expressed as the Weibull function shown in Fig. 1 are assumed. The
total number of transformers is normalized to 1 hereafter, in the same manner as the
statistical consideration for a single transformer. The number of transformers with an
age of i years in the j-th year is expressed as n(i, j), and the hazard rate of transformer
with an age of i years is expressed as k(i). In the first year, n(i, 1) is expressed as


nð1; 1Þ ¼ 1
ð1Þ
nði; 1Þ ¼ 0 ði ¼ 2; 3;


Þ:

In the j-th year, the total number of failures N(j) is expressed as

X
imax
NðjÞ ¼ nði; jÞ
kðiÞ: ð2Þ
i¼1

N(j) new transformers are also installed in the j-th year. In the second year, n(i, 2) is
expressed as
8
< nð1; 2Þ ¼ nð1; 1Þ
kð1Þ
nð2; 2Þ ¼ nð1; 1Þ  nð1; 1Þ
kð1Þ ð3Þ
:
nði; 2Þ ¼ 0 ði ¼ 3; 4; . . .Þ:

That is, n(i, j) is expressed as

8
< iP
max
nði; jÞ ¼ nðk; j  1Þ
kðkÞ ði ¼ 1Þ
ð4Þ
: k¼1
nði; jÞ ¼ nði  1; j  1Þ  nði  1; j  1Þ
kði  1Þ ði  2Þ
 Development of Support Program for Managing Assets 237

0.1

failure probability / hazard

0.08
hazard
0.06

0.04
failure probability
0.02

0
0 20 40 60 80
age (years)

Fig. 1. Assumed failure probability (shape parameter: 4, scale parameter: 50 years)

The initial age distribution is shown in Fig. 2 and the distribution in the 10-th year
is shown in Fig. 3. The failure probability is small for relatively new transformers and
the age distribution seems to shift to the right with time, as shown in Fig. 3(a). Also, a
few transformers are installed every year, which appear in Fig. 3(b). The distribution in
the 60th year is shown in Fig. 4. Such changes in the age distribution should be
considered in the LCC calculation.

1.2
number of equipment(normalized)

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50 60
age (years)

Fig. 2. Initial age distribution

3.3 Developed Program

The support program for managing assets has been developed on Microsoft Excel by
considering the above items. Evaluation examples for average maintenance planning
and individual maintenance planning are shown in Figs. 5 and 6, respectively. The
parameters shown below are assumed for each case.
238 T. Takahashi

1.2

number of equipment(normalized)
1

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50 60
age (years)

(a) Normal vertical scale

0.01
1
≈ ≈
number of equipment(normalized)

0.008

0.006

0.004

0.002

0
0 10 20 30 40 50 60
age (years)

(b) Expanded vertical scale

Fig. 3. Age distribution in 10th year

0.12
number of equipment(normalized)

0.1

0.08

0.06

0.04

0.02

0
0 10 20 30 40 50 60
age (years)

Fig. 4. Age distribution in 60th year

 Development of Support Program for Managing Assets 239

＜Case A＞
Installation cost: 100 million yen
Interest rate: 0.05
Failure cost: 10 million yen
Periodic inspection: 1 million yen, every 6 years
Non-periodic inspection; 5 million yes, at 35th year
Overhaul: (for only current equipment)
at 30th year, cost: 10 million yen,
rejuvenation year: 20 years
Rate of increase in repair cost with age:
Current equipment: 0.0002
New equipment: 0.00002
Failure probability: (Weibull function)
Current equipment
Shape parameter: 2.5
Scale parameter: 150
New equipment
Shape parameter: 2.5
Scale parameter: 300
＜Case B＞
(Items shown below are changed from Case A)
Rate of increase in repair cost with age:
Current equipment: 0.001
Failure probability: (Weibull function)
Current equipment
Shape parameter: 4
Scale parameter: 50
＜Case C＞
(Items shown below are added to Case A)
Present age: 25 years
Pending problem:
Cost of rank S: 20 million yes
＜Case D＞
(Items shown below are added to case A)
Present age: 25 years
Pending problem:
Cost of rank S: 60 million yen
240 T. Takahashi

95

cumula ve cost NPV, million Yen)

90

85

80 renewal at 60 years
renewal at 50 years
renewal at 40 years
75 renewal at 60 years (with OH)
renewal at 50 years (with OH)
70 renewal at 40 years (with OH)
20 40 60 80 100
elapsed years from installa on years

(a) Case A: low repair and failure costs

115
110
cumula ve cost NPV, million Yen)

105
100
95
90
renewal at 60 years
85 renewal at 50 years
80 renewal at 40 years
renewal at 60 years (with OH)
75 renewal at 50 years (with OH)
70 renewal at 40 years (with OH)
20 40 60 80 100
elapsed years from installa on years

(b) Case B: high repair and failure costs

Fig. 5. Evaluation example of average maintenance planning for transformers

From Fig. 5(b), if the repair and failure costs are sufficiently high, early renewal
and OH show some cost merits. From Fig. 6(b), if the pending problem cost is suffi-
ciently high, immediate renewal shows some cost merits. This program provides a
simple evaluation environment by considering several conditions.
 Development of Support Program for Managing Assets 241

90
80

cumula ve cost NPV, million Yen

70
60
50
40
30
renewal at 60 years
20
immediate renewal
10
0
0 20 40 60 80 100
elapsed years from present years

(a) Case C: low pending problem cost

100
90
80
cumula ve cost NPV, million Yen

70
60
50
40
30 renewal at 60 years
20
immediate renewal
10
0
0 20 40 60 80 100
elapsed years from present years

(b) Case D: high pending problem cost

Fig. 6. Evaluation example of maintenance planning for individual service-aged transformer

4 Conclusion

As an asset management support tool for oil-immersed transformers, a Microsoft Excel-

based program by LCC evaluation has been developed. A simple maintenance cost
model is adopted, and the selected examination scenarios are time-based renewal at
several ages with and without OH, and immediate renewal. Average planning by
considering the characteristics of the maintenance cost with aging, and individual
planning by considering pending problems, are realized. Trial application to actual
maintenance scenes and the investigation of necessary parameters will be carried out in
future.

References
1. Cigre TB176. Ageing of the System – Impact on Planning (2000)
2. Balzer, G., et al.: Life cycle assessment of substations: a procedure for an optimized asset
management. Cigre paper 23-302 (2002)
242 T. Takahashi

3. Quak, B., Gulski, E., Wester, Ph.: Fundamental aspects of information processing and the
decision process to support Asset Management. In: Proceedings of ISH2003, no. 790 (2003)
4. Balzer, G., et al.: Strategies for optimising the use of substation assets. Cigre paper B3-101
(2004)
5. Frijmersum, G.J., et al.: Asset management decision support model for distribution power
cable. In: Proceedings of ISH2005, no. F-13 (2005)
6. Dominelli, N., et al.: Life extension and condition assessment. IEEE Power Energy Mag. 4
(3), 25–35 (2006)
7. Takahashi, T., et al.: Study of decision support programs for maintenance strategy of electric
power equipment -proposal of diagnosis database application and repairing cost evaluation
method -. CRIEPI report, no. H06014 (2007). (in Japanese)
8. Takahashi, T., Okamoto, T.: Development of asset management support tools for oil
immersed transformer. IEEE Trans. Dielectr. Electr. Insul. 23(3), 1643–1648 (2016)
9. Takahashi, T.: Utilization of hierarchical data for maintenance strategy planning of electric
power equipment. In: Proceedings of ISH2017, no. 461 (2017)
10. Cigre TB541. Asset Management Decision making using different Risk assessment
methodologies (2013)
 SIAD-AERO: A New Methodology
for the Inspection of Energy Assets

Alexandre Dominice1,2(&), Fernando Teixeira Abrahão2(&),

Ricardo Augusto Tavares2,3(&), and Alexandre Barreto3,4(&)
1
EDP Brazil, São Paulo, SP 04547, Brazil
alexandre.dominice@edpbr.com.br
2
ITA, Technological Institute of Aeronautics,
São José dos Campos, SP 12228, Brazil
abrahao@ita.br
3
Energias Energy Efficiency, São José dos Campos, SP 12228, Brazil
ricardo.tavares@energias.com.br
4
George Mason University, Fairfax, VA 22030, USA
adebarro@gmu.edu

Abstract. This paper aims to present the SIAD-AERO Project. The Project has
the objective to develop a semiautonomous system for energy assets inspection,
capable to deploy embedded sensors in an aerial platform, to capture and pro-
cess images in the visible, infrared and ultraviolet bands, detecting anomalies
automatically and presenting an optimal plan of action. A better quality of
service and operational efficiency were the premises guiding the project. The
initial requirements for the system have been included and described, as well as
a description of the designed system, which includes its four subsystems and its
development methodology using Feature Driven Development (FDD). The
achieved results are presented, discussing the paradigms broken during the
project’s development.

Keywords: Remotely piloted aircraft
Inspection
Energy assets

1 Introduction

The need for energy transmission and distribution system having a transport high
capability, coupled with a failure low probability due to problems in its physical
infrastructure has increased significantly in recent years due to the increase in con-
sumption. These results from the expansion of the quality of life’s families, the reg-
ulator’s requirement for product quality and the energy distribution service, among
other factors.
In the growing demand, strong investments in energy transmission capacity would
be required, which has not been happening in a strong way. One of the consequences of
this is that the transmission lines and distribution networks are operate in situations
their maximum capacities, implying the need for low failure rates of the system.
One of the ways to ensure a system with high operational reliability is the regular
predictive and preventive maintenance of the physical infrastructure of the transmission

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 243–252, 2020.
https://doi.org/10.1007/978-3-030-31676-1_23
244 A. Dominice et al.

lines and distribution networks, as well as the main assets of the electric power
companies. However, the transmission lines and networks have thousands of kilome-
ters that require visual inspection, which results in a significant increase in inspection
costs.
In this context, the most common way to inspect the conditions of the system, in
addition to the ground inspections, is the visual inspection by helicopters equipped with
camera systems designed for this purpose and teams of specialists in this subject.
However, the use of this type of solution implies high costs, which are charge per hour
flown, in addition to the risk of aeronautical accidents due to the flight profile required
by the inspection, which can easily make the air inspection activity unfeasible.
Already in the ground inspection, there are diverse difficulties for the teams of
electricians. These factors occur to locate of such teams, which remain below and far
from the points being inspected.
The fact of these lines is exposed to atmospheric conditions physically stressing
them, such as large variations in temperature, humidity, wind, rain, lightning, electric
discharge, among many other factors, increases the need for regular inspection of them
in shorter term intervals.
Considering the scenario presented, it is reasonable to conclude that the use of
remotely piloted aircraft (RPA) in the inspection of assets of the electric sector allows
the reduction of risk to the maintenance teams and the overall reduction of operating
costs.
This scenario is significant, with RPAs equipped with sensors and decision support
systems, providing for: reducing the time needed for decision-making, minimizing the
time required for conducting inspections, increasing the effectiveness and efficiency of
operations, as well as the traceability of all actions taken during the process.
In this way, the Project has been conceived focusing on five basic premises that
guide all the objectives to be achieved: (a) Increase the probability of identification of
anomalies in transmission networks, reducing system failures; (b) Reduce the possi-
bility of human error in the inspection, with the use of algorithms of automatic
detection of anomalies; (c) Reduce the risk to which maintenance teams are expose
during helicopter inspection when performing line inspection tasks; (d) Reduce the
inspection time of the lines; and (e) Reduce inspection costs for lines when compared
to the costs associated with inspections using helicopters.
Based on the project’s premises, the objective has been defined as being the
development of a semiautonomous energy asset inspection system capable of using
sensors embedded in an aerial platform (Remote Piloted Aircraft System - RPAS),
which is based on a fixed wing platform and multirotor platform, to capture (online)
and to process (off-line) images in the visible, infrared and ultraviolet bands, including
the detection of existing anomalies automatically and presenting an optimal plan of
action (maintenance and inspection) of assets (Substations - ETDS and ETCS, Medium
and High lines, wind farms and PCHS).
To allow that all actions carried out to achieve maximum efficiency operational in
financial and temporal terms, being auditable at all levels, as well as minimizing the
operator’s exposure to the various risks involved in the activity.
 SIAD-AERO: A New Methodology for the Inspection of Energy Assets 245

Within this scope, the Project was denominated with the following title:
AUTONOMOUS-COOPERATIVE SYSTEM OF PLANNING AND EXECUTION
OF ENERGY ASSETS (SIAD-AERO).
In addition to the general objective presented, the SIAD-AERO Project also has the
following specific objectives:
(a) Adapt a multirotor platform and a fixed wing platform for inspection of assets,
equipped with a payload adjustable to each different asset inspection operation;
(b) Develop an operational software dedicated to managing and follow the assets
inspections of the EDP Brazil Group based in a military Command and Control
(C2) structure;
(c) Semi-autonomously inspect the energy assets, ensuring a safe procedure and
complete coverage of the surface, by adapting the control and navigation system
to the assets under analysis, based on information contained in the bank of mis-
sions, and automatic obstacle detection, simplifying and minimizing the inter-
ventions of a specialized operator;
(d) Enable the capacity to develop inspection plans configurable by type of asset to be
inspected, as well as the insertion of an example of these plans (demo plan) for
each of the assets inspected (Substations - ETDS, ETCS, Medium and High lines,
wind farms and PCHs); and
(e) Provide an intelligent planner capability to generate an optimal inspection flight
plan, based on the preferences and necessities provided by the user, seeking the
greatest efficiency of one or more requirements (cost, time, probability of success
of imaging, etc.).
The SIAD-AERO Project was started on September 29, 2016, having 24 months to
be developed. Project was registered at and has the clearance from the Brazilian
Electric Energy Agency (ANEEL) to be developed. The project sponsorship comes
from the EDP Group and has been carried out by Energias Energetic Efficiency Ltd. It
is one of the few R&D Project been developed by a private company sponsored by a
Brazilian federal R&D regulation.

2 System Description

2.1 System’s Initial Requirements

The process of defining technical requirements is a process that transforms the
expectations of the sponsor or stakeholder into a complete definition of the problem to
be addressed, after some interactions, a set of requirements that must be put forward as
assertive or assertions that can be used to define a technical design solution.
In the context of the SIAD-AERO, the requirements were initially defined to meet
the company’s inspection issues focusing to deploy aerial platforms. First interactions
of the system development cycle were the goal to define the main requirements of
inspection system as itself, as following: (a) capability to plan and follow the
inspections; (b) capability to process the generated images using artificial intelligence
(IA) algorithms to detect the anomalies automatically; (c) to allow the company’s
246 A. Dominice et al.

technicians to analyze and validate the anomalies; (d) to generate the maintenance
orders once the anomalies are validated.
The second phase was the definition of RPA Fixed Wing requirements. In the
general way it needs to achieve theses performance: (a) Maximum range of 40 km;
(b) BVLOS operations (Beyond Visual Line of Sight); (c) Flight endurance of at least
5 h; (d) Maximum payload of 15 kg for takeoff; (e) Take-off and assisted landing;
(f) Safe return (fly-to-home); (g) Parachute system (emergency landing); (h) Semiau-
tonomous navigation; (i) and remote sense capabilities (EO, IR-LWIR, UV).
The second platform is the multirotor RPA, where requirements are: (a) Maximum
range of 6 km; (b) VLOS operations (visual line of sight); (c) Flight autonomy of at
least 1 h; (d) Maximum payload of 15 kg for takeoff; (e) Take-off and assisted landing;
(f) Safe return (fly-to-home); (g) Parachute system (emergency landing); (h) Semiau-
tonomous navigation; and (i) remote sense capabilities (EO, IR-LWIR, UV).

2.2 System Overview

According to the proposed objective and based on the initial requirements defined
during the design of the Project, the System was organized into a set of sub-systems,
which is illustrated by means of Fig. 1 and described below:
(a) Autonomous RPA (Multirotor & Fixed Wing): remote sensing sub-system
consisting of a fixed-wing and multirotor RPA, both equipped with an embedded
computer system that can perform semi-autonomous fly. It is composed of the
following subsystems: i. Flight Control; ii. Flight Plan; iii. Navigation; iv.
Emergency; v. Surveillance; vi. Command & Control; vii. Sensor & Actuator; and
viii. Radio.
(b) Planner: intelligent planning subsystem, capable of automatically creating opti-
mized and configured inspection plans for each type of asset to be inspected,
individual and/or cooperative (with more than one RPAS at the same time) and
which take into account user preferences. It is composed of the following sub-
systems: i. Flight Plan Manager; ii. Simulation; iii. Playback; iv. Mission Anal-
ysis; and v. Storage.
(c) Ground Control System (GCS): subsystem responsible for monitoring, co-
controlling and controlling the RPAS and their respective senses in real time,
through computer screens or augmented reality subsystems. From these stations it
is possible to obtain instantaneously, respecting the intrinsic limitations inherent
to the electromagnetic environment, data from the RPAS. It can be portable
(Portable C2) or mobile (Mobile C2 - shipped in a car). It is composed of the
following subsystems: i. Antenna Tracking; ii. Command & Control; iii. Image
Processor; iv. Radio; and v. Storage.
(d) Remote Sensing: subsystem responsible for processing, after the flight, the
images obtained by sensors embedded in the semi-autonomous RPA, detecting
the anomalies pre-defined by EDP and presenting them to the analysts for iden-
tification and confirmation. It consists of the following subsystems: i. Sensor; ii.
Image Acquisition; iii. Features Selection; iv. Features Classification; and v.
Storage.
 SIAD-AERO: A New Methodology for the Inspection of Energy Assets 247

EMERGENCY FLIGHT PLAN

Autonomous SUBSYSTEM SUBSYSTEM Planner
RPA USB FLIGHT PLAN
SIMULATION
MANAGER
SUBSYSTEM
SUBSYSTEM
FLIGHT SENSOR E
NAVIGATION CAMERA
CONTROL ACTUATOR
SUBSYSTEM
SUBSYSTEM SUBSYSTEM MISSION
PLAYBACK
ANALYSIS
SUBSYSTEM
SUBSYSTEM
SURVEILLANCE COMMAND &
CONTROL
SUBSYSTEM
SUBSYSTEM
STORAGE SUBSYSTEM

RADIO
SUBSYSTEM
NET Remote Sensing
DATA-LINK DATA-LINK
FEATURES FEATURES
RADIO CLASSIFICATION SELECTION
SUBSYSTEM SUBSYSTEM SUBSYSTEM

ANTENNA COMMAND & IMAGE

TRACKER CONTROL PROCESSOR
SUBSYSTEM SUBSYSTEM SUBSYSTEM IMAGE ACQUISITION SUBSYSTEM

Ground
Control STORAGE SUBSYSTEM STORAGE SUBSYSTEM
Station

Fig. 1. Schematic diagram of the system

3 Methodology and Deliverables

3.1 Methodology
The development of SIAD-AERO was completed in a period of 24 months, using
Development Driven Development (FDD methodology [1–4]. This methodology
consists of an agile method for system management and development, whose main
characteristic is to be objective and practical, although also to develop documentary
artifacts associated to the generation of useful deliverables for the user.
The FDD can be seen through two iterative and incremental macro processes:
(a) Conception and Planning: In this phase, the problem has been structured, and
the inspection requirements will be surveyed with clients, which will be based on
a series of case studies related to the operations of monitoring and protection of
transmission lines, distribution systems and assets. To perform this phase, the
Strategic Options Development and Analysis (SODA) method will be used [5, 6];
(b) Construction: As presented previously, it is this phase that develops artifacts of
interest to the user (deliverables), as well as all supporting documentation. This
phase can be divided into macro activities that can be described and presented as
structuring builds. That is, large sub-phases that characterize a large set of
deliverables that are subject to in-depth external audit. These sub-phases (builds)
are shown as follows.
a. Build 1 - Problem Structuring;
b. Build 2 - Initial Developments of Autopilot, Data Link Subsystem, Basic
Allocation Algorithm of Allocation and Algorithm of Detection of Anomalies;
c. Build 3 - Final Development of Autopilot, development of the Advanced
Allocation Algorithm and the complete Inspection Plan, development of the
248 A. Dominice et al.

Radiometric Methodology, with implementation of adjustments in the

Algorithm;
d. Build 4 - Application of Inspection Project, development of the Programming
Module of the recording profile of the sensors based on user preference and
inspection plan by mission profile;
e. Build 5 - Application of Operation: the final adjustments on the Autopilot for
navigation over long distances, in addition to tests and validation in ground
and in flight. The Multirotor and Fixed Wing flight envelope will be built.
Adjustments to real-time image transmission systems. Development of project
feedback and review module. IR and UV in-flight image recording tests;
f. Build 6 - Post-processing of data: Adjustment in the RPA flight envelope.
Development of the extended reality system and the mission simulator. Final
tests and validation of assets imaging;
g. Build 7- Interactive Database: Development of the interactive mission data-
base, aiming at the storage and subsequent criticism of the mission.
h. Build 8 - Certification of the RPAS (fixed and rotating wing): closing of the
certification process with ANAC (National Civil Aviation Agency), ANATEL
(National Telecommunications Agency) and DECEA (Department of Space
Control Aerial), so that the developed equipment can be operated in Brazil, in
addition to the final adjustments in the operating manual and the training of the
operating teams.

3.2 Project Deliverable

The deliverables of SIAD-AERO are (a) inspection system; (b) certification of oper-
ation; and (c) training for the operators’ teams. With regard to the inspection system, it
has already been presented. Within two years the entire hardware and software system
were finalized, as originally planned.
About certification, the development team was also responsible for delivering the
systems certified by Brazilian Civil Aviation Authority (ANAC) and Brazilian Air
Force Air Traffic Department (DECEA), ready to be operated. The certification process
is still being done.
Regarding the training for operator’s teams, a specific training was conducted for
piloting and sensors operation. In this case, the operators indicated the professionals to
be trained, who only started the training if they were fit according to specific regula-
tions of ANAC and DECEA.
Must be noted that deliverables cannot be analyzed individually. As presented at
the beginning of this article, the objective of the Project is developed an Operation
System, which implies the delivery of inspection tools, as well as the delivery of a new
inspection methodology, which can be defined as a new operational concept for the
Distributor.
The application of this new concept had some impacts that were already discussed
by EDP Brazil. The first major impact to be highlighted was the possibility of a change
in the way the inspections were carried out. With the delivery of the system, all
procedures have been reviewed and updated. With the system’s implementation, a new
function was created within the companies responsible for execution of the project,
 SIAD-AERO: A New Methodology for the Inspection of Energy Assets 249

which is the function of RPA operator for inspections. This function will be complex,
requiring both technical training and psychomotor skills (required for pilotage), there
was a need to define what type of vocational training would be required, the level of
training, among many already raised.
In addition to these, the greatest expected impact is the possibility of creating a new
business for the Sponsors that will be related to all the possibilities that this System will
be able to provide. In this respect, the developed product can be applied throughout the
national territory and, perhaps, international, requiring only operational adaptations.

3.3 Applicability and Relevance

This project was applied in all concession areas under the responsibility of EDP Brazil
and has the potential to reach the results of all three million final customers. The fact
that the effective inspection of the transmission and distribution lines guarantees the
efficiency of supply for its consumers, thus mitigating violations of the electricity
supply indicators (DEC and FEC, following the Brazilian Electric Power Authority -
ANEEL) and avoiding the payment of fines.
Regarding the verification of the results when the project finished, field validation
tests were conducted, based on a specific methodology called Systems Test and
Evaluation [7], which uses state-of-the-art applied statistics techniques such as
Experiments (Design and Analysis of Experiment - DOE) and multivariate regression,
among others [8–10].
The verification of the applicability of SIAD-AERO was demonstrated through the
tests carried out in laboratories, as well as through field tests.
With regard to the relevance of SIAD-AERO in scientific terms, it has been
observed that projects were found that focus only on the development of the air
platform or only on the specification of sensors used to identify anomalies in the
transmission system. In none of the cases was it possible to find a similar research line
that has a holistic and integrated view on the problem of automation of the process of
inspection of energy assets.
During the development of the project, two-master’s programs were realized for
members of the development team at the Technological Institute of Aeronautics (ITA).
All the works carried out had back-to-back themes for direct application in the project.
With regard to the diffusion of knowledge, it was planned to organize and to hold at
least one workshop to provide the exchange of experiences between the project team
and the professionals of the country’s energy distribution sector.
To technological relevance, attention for the fact that the development of the SIAD-
AERO project represented a concrete evolution insofar as the Brazilian market had an
innovative technological system available.
Commercially, the SIAD-AERO system is available and consists of a platform
capable of generating new business models associated with the inspection and man-
agement of energy assets, with secondary application in civil defense and public safety
actions. All the results obtained by the project team were submitted to the intellectual
property protection agencies, which may allow future royalties.
Finally, the project team promoted specialized training for the EDP team, training
pilots and maintainers of the aerial platform, as well as training in the maintenance and
250 A. Dominice et al.

resolution of technical problems for the computer system. All this made possible the
immediate use of the new system.
Analyzing the discussion about deliverables and related applicability’s, it can be
concluded that the Project described here has brought a great amount of innovations in
the Brazilian scenario, incorporated into the systemic treatment. This includes the use
of an inspection tool and subsystems to support planning, mission analysis and decision
support which also incorporates the distributor’s business through a System Operation
Concept.
This has led paradigms of project development approach to energy asset inspection
to be broken. That is, the focus shifted from the development of tools and turned to the
development of a system developed based on an operational concept, both developed
jointly and interactively.
In addition to the paradigm break, this project also continues the certification for
system operation. This process was initiated in parallel with the beginning of the
project, so that at the end of the project, there were no legal problems that could prevent
the operation of the system.
In this way, it is possible to affirm that the development of a complex development
project, such as the one presented, has a high probability of success when developers
and sponsors are involved in the development of concepts that guide the technical
design decisions.

4 Partial Results
4.1 Multisensory Pattern Recognition (MPR) Process
The MPR process has two phases. In the first one, the imagery analyses have been done
using computer vision algorithms (SVM - Support Vector Machine classifier) and the
Histogram of Oriented Gradients (HOG) was used as the feature representation.
However, this method was proven not effective.

Fig. 2. Faster R-CNN object detection result.

 SIAD-AERO: A New Methodology for the Inspection of Energy Assets 251

Thus, as deep learning approaches have proven their superiority in similar image
recognition/classification problem [11]. A Two-stage Faster R-CNN Tensorflow
architecture [12] is currently used. Two-stage architectures first categorize potential
objects in two classes: foreground or background. Then all foreground potential objects
are classified in more fine-grained classes: pylons, utility poles, isolators, transformers,
etc. This two-stage method is very slow compared to a one-stage approach, but also
produces the best accuracy. As every convolutional neural network (CNN), it takes an
image as input and processes it through several convolution kernels. Each kernel’s
output is a feature map- the first feature maps capture high-level features (such as a line
or a color). The further we go down in the network, the smaller the feature maps become
because of the pooling layers. While they are smaller, they also represent more fine-
grained information. The input image has three channels (red, blue, green), but every
subsequent feature map has dozens of channels. Each of them represents a different kind
of feature that it captured. Figure 2 presents the Faster R-CNN object detection result.

4.2 Business Impact Analysis

The expected operational impact through the implementation of SIAD-AERO can be
translated as the effective increase of EDP’s inspection capacity, in a concomitant way
with the significant reduction of inspection costs. Table 1 presents the comparison
between the inspection capacity of a team inserted in the process formerly used in EDP
and another team, already making use of the paradigm generated by SIAD-AERO.

Table 1. Operational impact

Inspection capacity (Km)
Urban visual Rural visual IR urban IR rural
inspection inspection inspection inspection
EDP current 5.670 km 2.514 km 10.584 km 2.514 km
capacity
SIAD-AERO 20.412 km 102.060 km 20.412 km 102.060 km
Percentage 360% 4.060% 193% 4.060%
gain

The results showed in the previous table gives a good idea how the system oper-
ation will impact in the EDP inspection planning and execution. The capacity will
increase and the time to perform the inspections will decrease. The use of human
resources and labor risk will decrease too. Therefore, the system has the possibility to
revolute the inspection paradigm in the Brazilian Utilities Industry.

5 Final Remarks

These his article discussed the presentation of the SIAD-AERO project, covering all its
technical and conceptual aspects, as well as the motivators and premises for carrying
out this project.
252 A. Dominice et al.

The initial requirements of the system to be designed were presented, emphasizing

that they are defined in an interactive way between the researchers and the technical
sectors of the stakeholders based on an operational concept developed together.
The interaction was necessary during the development of the project and allowed
the refinement and detailing of the technical requirements as well as the refinement of
the operational concept. These interactions increased the project’s likelihood of success
because they guided all technical decisions during project development to meet the
operational needs.
The main deliverable or product of the SIAD-AERO Project was the System
composed of four subsystems, with certification and operational training.

Acknowledgments. Authors and researchers would like to thank stakeholders for their trust and
partnership in the development of this Project, especially for ANEEL, EDP Brazil Group, ITA
and ENERGIAS.

References
1. France, R., Rumpe, B.: Model-driven development of complex software: a research
roadmap. In: 2007 Proceeding of Future of Software Engineering, FOSE 2007, pp. 37–54.
IEEE Computer Society, Washington, DC. https://doi.org/10.1109/fose.2007.14, ISBN 0-
7695-2829-5
2. Highsmith, J., Cockbur, A.: Agile software development: the business of innovation. IEEE
Comput. 34(9) (2001). https://doi.org/10.1109/2.947100, ISSN 0018-9162
3. Schmidt, D.C.: Model-driven engineering. IEEE Comput. 39(2), 25–31 (2006). ISSN 0018-
9162
4. Palmer, S.R., Felsing, M.: A Practical Guide to Feature-Driven Development. Pearson
Education (2001). ISBN 0130676152
5. Eden, C., Ackermann, F.: Group decision and negotiation in strategy making. Group Decis.
Negot. 10, 119–140 (2001)
6. Eden, C., Ackermann, F.: SODA: the principles. In: Rosenhead, J., Mingers, J. (eds.)
Rational analysis for a problematic world revisited: problem-structuring methods for
uncertainty and conflict, pp. 21–41. Wiley, Chichester (2001)
7. United States of America. Test and Evaluation Management Guide, 6th edn. Department of
Defense, Washington (2012). Accessed 15 Jan 2017
8. Sargent, R.G.: Verification and validation of simulation models. In: 2007 Proceedings of
Winter Simulation Conference, Syracuse, pp. 124–137. IEEE, Syracuse (2007)
9. Montgomery, D.C.: Design and Analysis of Experiment, 7th edn. Wiley, Hoboken (2008)
10. Kleijnen, J.P.C., et al.: A user’s guide to the brave new world of designing simulation
experiments: state-of-the-art review. In-forms: J. Comput., 263–289 (2005)
11. Huang, J., Rathod, V., Sun, C., Zhu, M., Korattikara, A., Fathi, A., Fischer, I., Wojna, Z.,
Song, Y., Guadarrama, S., Murphy, K.: Speed/accuracy trade-offs for modern convolutional
object detectors. In: IEEE CVPR, vol. 4, July 2017
12. Ren, S., He, K., Girshick, R., Sun, J.: Faster R-CNN: towards real-time object detection with
region proposal networks. In: Advances in Neural Information Processing Systems, pp. 91–
99 (2015)
 Thermal Aging of Photovoltaic Cables Based
Cross-Linked Polyolefin (XLPO) Insulation

Ramy S. A. Afia1,2 , Ehtasham Mustafa2,3,

and Zoltán Ádám Tamus2(&)
1
Department of Electrical Power and Machines Engineering,
Faculty of Engineering, Helwan University, Cairo, Egypt
ramysaad@h-eng.helwan.edu.eg
2
Department of Electric Power Engineering,
Faculty of Electrical Engineering and Informatics,
Budapest University of Technology and Economics, Budapest, Hungary
{mustafa.ehtasham,tamus.adam}@vet.bme.hu
3
Department of Electrical Engineering, Faculty of Engineering and Technology,
Gomal University, Dera Ismail Khan, Pakistan

Abstract. Photovoltaic and wind turbine energy systems have a pivotal role to
meet the growth in power demand. The power generated from the renewable
energy sources has main advantages of the absence of harmful emissions and
infinite availability. The integration of photovoltaic systems into the power grid
is expected to increase with the continuous reduction in the capital cost of these
systems. The photovoltaic systems are exposed to extreme ambient conditions
such as solar radiations, wind, humidity and thermal stresses. Cables in the
photovoltaic systems are of great importance since they provide the link
between the main components of the photovoltaic systems so, the reliability,
efficiency and lifetime of the photovoltaic systems affected greatly by the
functionality of these cables. This research presents a study of the insulation
condition of Cross-Linked Polyolefin insulation based photovoltaic cable sam-
ples exposed to elevated temperature of 120 °C for four cycles. The analysis
based on the measurement of the insulation capacitance, insulation resistance
and the dissipation factor or tangent delta after each cycle as a function of
frequency ranged from 100 Hz to 500 kHz. The results showed that the
capacitance had a slight change with increasing the ageing period while the
insulation resistance and the dissipation factor were more affected.

Keywords: Photovoltaic cables
Thermal aging
Capacitance
Insulation

resistance
Dissipation factor

1 Introduction

Electric power utilities face many challenges from the huge development towards an
extensive integration of different renewable energy sources into the power grid such as
photovoltaic (PV) energy systems [1]. Taking into consideration the PV source relia-
bility, availability, visibility and other factors such as the stability of the power grid, the
integration process should guarantee the mutual benefits between the two energy

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 253–260, 2020.
https://doi.org/10.1007/978-3-030-31676-1_24
254 R. S. A. Afia et al.

sources. A typical PV system comprises of solar panels which convert the incident of
solar radiations into DC power, inverter which provide the interface between the solar
panel and the utility network, batteries and charge controller as main components [2].
The design of the PV system components should be done accurately to obtain maxi-
mum output power and minimum losses in the system. The PV cables are used to
connect between the components of the PV system. PV cable management is of a great
importance aspect of the safety and lifetime of the PV system. This is primarily due to
the usage of exposed cables in the PV array since the PV system is installed outdoors
on rooftops and open fields where harsh ambient conditions such as solar radiation,
wind, humidity, and hot & cold temperatures stress exposed components as modules
and cables [3]. Theses harsh ambient conditions affect the insulation integrity of the
cables used in the PV systems where insulation degradation becomes more significant
affecting the overall PV system [2]. For this reason, the functionality of these cables
should be guaranteed over a long period since the estimated thermal lifetime of theses
cables is about 25 years [4–7]. In this research, samples of Cross-Linked Polyolefin
(XLPO) insulation-based PV cables were subjected to accelerated thermal aging to
investigate the effect of thermal stresses on the insulation integrity. The evaluation of
the insulation state carried out by measuring the insulation capacitance, insulation
resistance and the dissipation factor (tand) using a precision component analyzer with
output voltage of 5 V over a range of frequency from 100 Hz to 500 kHz.

2 Experimental Work
2.1 Specimens
The measurement of the insulation parameters was carried out on samples of low
voltage PV cable. The cable consists of three main parts, fine wire tin-plated copper
conductor with cross sectional area of 4 mm2, inner and outer insulation made of
XLPO as in Fig. 1. The cable technical specifications are given in Table 1.

Fig. 1. Cross section of the cable.

 Thermal Aging of PV Cables Based Cross-Linked Polyolefin (XLPO) Insulation 255

Table 1. Technical specifications of the cable.

Parameter Value
Rated voltage (AC) 1 kV
Inner insulation XLPO
Insulation thickness 1 mm
Outer (jacket) insulation XLPO
Jacket thickness 0.77 mm
Outer diameter 5.8 mm
Max. conductor temperature 120 oC

2.2 Accelerated Aging Procedure

Before starting the thermal aging, the samples have been preconditioned in an oxygen-
controlled oven to remove any moisture from the samples as shown in Fig. 2. The
preconditioning was for one day at temperature of 70 °C. After that, the samples were
kept at room temperature for one day. Before starting the thermal aging, a benchmark
measurement for the capacitance, insulation resistance and tand was carried out. The
accelerated thermal aging was at a temperature of 120 °C for different periods of 240,
480, 720 and 960 h. Between each period, the measurements were carried out to clearly
investigate the effect of different aging periods. For the measurement of the capaci-
tance, Insulation Resistance (IR) and tand the samples were covered with an aluminum
foil so, the measurement was between the conductor and the aluminum foil at room
temperature. All the measurements carried out in Faraday cage to eliminate any
interference during the measurement.

Fig. 2. Samples in the oven.

256 R. S. A. Afia et al.

3 Results and Discussion

Figure 3 depicts the capacitance, the insulation resistance (IR) and tand of unaged
samples over the whole frequency range (100 Hz to 500 kHz). The values of the
capacitance and IR decreased with increasing the frequency as shown in Fig. 3a and b
respectively while Fig. 3c shows that the values of tand increased moving to the higher
frequencies showing how the leakage current dominate in the insulation at higher
frequencies.

Fig. 3. The benchmark measurement (a) Capacitance (b) Insulation resistance (c) tand.

Figure 4 illustrates the capacitance after the thermal aging. The capacitance had the
same profile as the unaged case i.e., the capacitance values decreased with increasing
the frequency. Besides, after the first aging cycle (240 h), the capacitance decreased
slightly below the baseline measurement over the whole frequency range showing that
the dielectric insulation lost around 2% of its electrical capacitance [8]. While it
increased after the second cycle (480 h) showing that the insulation starts to retain its
original capacitance since the decrease in the capacitance became 1.2%. Moving to the
third and fourth cycles (720 and 960 h) respectively, the insulation recovered its
 Thermal Aging of PV Cables Based Cross-Linked Polyolefin (XLPO) Insulation 257

original capacitance since only 0.1% and 0.22% difference after the third and fourth
cycles respectively [9].

Fig. 4. Capacitance of aged and unaged samples.

The plot of the insulation resistance is shown in Fig. 5. Like the capacitance, the
insulation resistance had the same profile of the unaged case i.e., the higher values of
IR were shifted to the lower frequency range. Also as shown in Fig. 5, starting from
100 Hz towards 2 kHz, the IR values increased with increasing the aging period while
the IR values were less affected with the aging period increased for the frequency range
from 5 kHz to 500 kHz.
Basically, the dielectric dissipation factor is an indication for the losses in the
dielectric material. The active losses in the dielectric are conduction losses (Pc),
polarization losses (Pp) and ionization losses (Pi). The conduction losses depend on the
insulation resistance. The dissipation factor is dependent on the applied electric field,
frequency, insulation resistance of the dielectric material and temperature [8, 10]. The
dissipation factor is calculated as follows,

tan d ¼ IR =IC ¼ 1=ðx: C: RÞ ð1Þ

Where IR and IC are the leakage and charging current respectively and x is the
angular frequency. Figure 6 depicts the dissipation factor of the aged and unaged
samples. The tand values decreased with increasing the aging period starting from
100 Hz to 2 kHz and this can be explained by the increase of the insulation resistance
which leads to reduction in the leakage current (IR) for the same frequency range as
shown in Fig. 5. After the fourth cycle (960 h), the tand values increased with
increasing the frequency from 5 kHz to 500 kHz due to the decrease in the insulation
258 R. S. A. Afia et al.

Fig. 5. Insulation resistance of aged and unaged samples.

Fig. 6. tand of aged and unaged samples.

 Thermal Aging of PV Cables Based Cross-Linked Polyolefin (XLPO) Insulation 259

resistance after this cycle for the same frequency range showing more leakage current
passing through the insulation.

4 Conclusion

In this work, low voltage XLPO insulation based low voltage photovoltaic cable
samples were thermally aged at temperature of 120 °C for four aging cycles 240, 480,
720 and 960 h. The capacitance, insulation resistance and dissipation factor were
investigated to find out the integrity of the insulation after the thermal aging. The
measurements were carried out using precision components analyzer with frequency
range from 100 Hz to 500 kHz.
The results showed that, the capacitance decreased after the first thermal cycle and
it increased with increasing the aging period. The change in the capacitance after the
end of the fourth cycle was very slight, 0.22% of its original value before aging.
The values of the insulation resistance increased with aging period increased for
frequency range from 100 Hz to 2 kHz and it decreased moving to the high frequency
range showing more conduction losses in the insulation. Consequently, the profile of
the dissipation factor was correlated to the insulation resistance profile. For frequency
range from 100 Hz to 2 kHz, the dissipation factor values decreased with the aging
period increased while it increased moving to the high frequency range showing more
leakage current passed through the insulation.
Chemical investigations are needed to provide more explanations to the obtained
results.

References
1. Colli, A.: Failure mode and effect analysis for photovoltaic systems. Renew. Sustain. Energy
Rev. 50, 804–809 (2015)
2. Ekici, S., Kopru, M.A.: Investigation of PV system cable losses. Int. J. Renew. Energy Res.
7, 807–815 (2017)
3. Parise, G., Martirano, L., Parise, L.: Life monitoring tool of insulated cables in photovoltaic
installations. IEEE Trans. Ind. Appl. 50, 2156–2163 (2014)
4. Parise, G., Rubino, G.: A comparison of the ANSI/IEEE and the CENELEC/IEC approaches
to overload protection of insulated power cables. IEEE Trans. Ind. Appl. 33, 1362–1366
(1997)
5. Parise, G., Martirano, L.: Circuits operation control and overloads protection: power cables
equivalent ampacities. IEEE Trans. Ind. Appl. 36, 22–29 (2000)
6. Parise, G., Martirano, L.: Life loss of insulated power cables: a “count hours” device. In:
2001 IEEE Industrial and Commercial Power Systems Technical Conference. Conference
Record (Cat. No. 01CH37226), pp. 91–95. IEEE (2002)
7. Parise, G., Martirano, L.: System management strategy to monitor insulated power cables.
In: Conference Record of the 2004 IEEE Industry Applications Conference, 39th IAS
Annual Meeting, pp. 1755–1759 (2004)
260 R. S. A. Afia et al.

8. Kumru, C.F., Arikan, O., Kocatepe, C., Kalenderli, O.: Investigation of effect of temperature
variation in a high voltage cable on electrical insulation. In: 2013 International Symposium
on High Voltage Engineering, pp. 1104–1109 (2013)
9. Celina, M., Gillen, K.T., Wise, J., Clough, R.L.: Anomalous aging phenomena in a
crosslinked polyolefin cable insulation. In: Radiation Physics and Chemistry, pp. 613–626
(1996)
10. Asipuela, A., Mustafa, E., Afia, R.S.A., Adam, T.Z., Khan, M.Y.A.: Electrical condition
monitoring of low voltage nuclear power plant cables: tand and capacitance. In: 2018
International Conference on Power Generation Systems and Renewable Energy Technolo-
gies (PGSRET), pp. 1–4. IEEE (2019)
 Investigation of Complex Permittivity of XLPO
Insulated Photovoltaic DC Cables Due
to Thermal Aging

Ehtasham Mustafa1,2, Ramy S. A. Afia2,3,

and Zoltán Ádám Tamus2(&)
1
Department of Electrical Engineering, Gomal University,
Dera Ismail Khan, Pakistan
2
Department of Electric Power Engineering,
Budapest University of Technology and Economics, Budapest, Hungary
{mustafa.ehtasham,tamus.adam}@vet.bme.hu
3
Department of Electric Power and Machines Engineering,
Helwan University, Cairo, Egypt
ramysaad@h-eng.helwan.edu.eg

Abstract. To study the effect of thermal aging on the low voltage DC cables
used in photovoltaic systems, XLPO based cable samples were thermally aged
for four different cycles of 240, 480, 720 and 960 h at 120 °C. This aging time
was equivalent to 20, 40, 60 and 80 years of service period at 60 °C. The
dielectric properties, real and imaginary part of permittivity were investigated
for a frequency range of 20 Hz to 500 kHz using impedance analyzer. The
polarization and leakage current effects have been discussed due to the thermal
aging. The results show that the real part of permittivity increased with the aging
due to the creation of dipoles because of the morphological changes happening
inside the material. While the imaginary part of permittivity decreased at low
frequencies and increased at higher frequencies. Also, there has been shifting of
minimum value of imaginary part of permittivity to lower frequency, 500 Hz
from 2 kHz with aging, showing that the resistance of the material increased at
low frequency and the material becomes stiffer and hence an end-of-point to the
degradation of insulation material has reached.

Keywords: Thermal aging
Photovoltaic systems cables
Dielectric property

XLPO

1 Introduction

The depletion of the naturally occurring energy sources and the effect of the carbon
emission on the environment have made the use of renewable energy more important.
They have the advantages of being clean, green and reusable. Photovoltaic (PV) system
is one such renewable energy source out of other available sources, being considered
safe and environment-friendly. On the other side, the reported number of faults
occurring in the system has increased with an increase in its usage. The failure of the
cable insulation in general and of DC cables in particular in one such failure. Since, the

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 261–269, 2020.
https://doi.org/10.1007/978-3-030-31676-1_25
262 E. Mustafa et al.

DC cables are used in connecting the PV modules and inverters and are exposed to a
number of stresses, i.e., ultraviolet radiation, heat, humidity, mechanical, electrical and
thermal. As a result, the phenomenon of degradation, aging, in these cables is more as
compared to other components of the PV systems.
Since the insulation of the cables is made of polymer, so the thermal stress which
affects the structure of the polymer is considered to be more important than others. The
thermal stress could be due to the Ohmic current or environmental temperature. Cross-
linked polyolefin (XLPO) has been widely used polymer in PV insulation cables due to
number of advantages such as thermal stability, dielectric properties, and solvent
resistance. However, due to aging the chemical and physical properties of the XLPO
will change under thermal stress and may affect the performance of the cable. The
aging of insulation may result in leakage current which may increase with the passage
of time and result in electric sparks and hence system failure. This makes the study of
the real cause of aging of the insulation inevitable. A reliable condition monitoring
(CM) technique helps in better understanding the aging of insulation. The CM tech-
niques can be used to study either mechanical, chemical or electrical properties. The
first two properties are studied by the destructive methods, while in recent times some
non-destructive CM techniques have been reported to study the electrical properties of
low voltage cables [1–7]. But none of them have been used to study the behavior of LV
cables used in PV systems.
This paper has been aimed to study the dielectric properties; real ðe0 Þ and imaginary
00
ðe Þ part of permittivity of LV PV cables under the thermal stress. The method is useful
in understanding the production of polarization and conduction losses due to the
morphological changes happening inside the insulation due to the thermal stress. In this
work, XLPO based LV PV DC cables were thermally aged at 120 °C for four cycles
each of 10 days. The e0 and e00 were studied using impedance analyzer for a frequency
range of 20 Hz–500 kHz for each thermal cycle. The method was very helpful in
studying the effect of thermal stress on the polarization and conduction losses in the
insulation material, which makes the method capable of using it in PV systems as a
potential CM technique.

2 Experimental Work

2.1 XLPO Based LV PV DC Cable

The cross-sectional view of LV XLPO based PV DC cable used for the study is shown
in Fig. 1. XLPO has been used both as the primary insulation and the jacket of the
cable. The XLPO is a semi-crystalline material which has a complex crystalline melting
point [8]. Table 1 summarized different characteristics of the cable.
 Investigation of Complex Permittivity of XLPO Insulated PV DC Cables 263

Fig. 1. XLPO based DC cable under investigation

Table 1. PV cable characteristics parameters.

Parameters Values
Conductor Tin platted copper conductor
Conductor diameter 2.25 mm
Outer insulation/Jacket XLPO
Outer diameter 3.55 mm
Primary insulation XLPO
Rated conductor temperature −40 °C to 120 °C
Rated voltage DC: 1.5 kV/AC: 1.0 kV

2.2 Accelerated Aging

A group of five cable samples, each one meter long, were subjected to thermal stress at
120 °C for four different oven aging periods of 240, 480, 720 and 960 h. These were
equivalent to 20, 40, 60 and 80 years of service time at 60 °C, which were calculated
using the Arrhenius model. The jacket and the insulation were kept in contact with the
conductor to obtain the results as the cable is subjected to the field conditions. The
samples were removed from the oven after each thermal cycle for the dielectric
property analysis.

2.3 Measurement of Dielectric Properties

The measurement of dielectric properties was carried out using impedance analyzer for
a frequency range of 20 Hz to 500 kHz at an applied voltage of 5 Vrms. The impedance
of the cable was measured in R-C parallel configuration. The input signal terminal was
connected to the conductor and the output signal terminal to a wire braid positioned on
the outer surface of the cable. To reduce any pickup from the external noises, the cable
samples along with the electrodes setup were placed in the Faraday cage. The
264 E. Mustafa et al.

measurements were carried out at a temperature of 25 °C ± 2%. The real ðe0 Þ and
imaginary ðe00 Þ parts of the complex permittivity were calculated using the relationships
[9]:

dC
e0 ¼ ð1Þ
Aeo

d
e00 ¼ ð2Þ
AxReo

Where d and A are the thickness of the insulation and area of the electrode, respec-
tively. The permittivity of vacuum shown as eo and x is the angular frequency.

3 Experimental Results

The behavior of the real part ðe0 Þ and imaginary part ðe00 Þ of permittivity are shown in
Figs. 2 and 3 for a frequency range of 20 Hz to 500 kHz. The e0 increases as the
frequency is decreased irrespective of the aging. After the first thermal cycle, e0 has
decreased for all the frequencies. But an increase has been observed in e0 after the
second and third thermal cycles. After the fourth cycle of thermal stress, a decrease in e0
at 20, 40 and 100 Hz has been observed while at all other frequencies it has increased.

Fig. 2. Real part of permittivity (in a scale of 102) v/s frequency

For the e00 the graph has been plotted between 400 Hz and 500 kHz, as at low
frequencies below 400 Hz, the resistance of the jacket and insulation is higher than the
 Investigation of Complex Permittivity of XLPO Insulated PV DC Cables 265

range of impedance analyzer. The e00 have increased as the frequency is increased
regardless of the aging. The effect of the first thermal cycle on e00 has been a decrease in
its value. While after the second thermal cycle, an increase has been observed between
a frequency range of 400 Hz and 20 kHz. While at all other frequencies a decrease has
been detected. After the third thermal cycle, a decrease in the values of e00 has been
noted at 200 Hz and at the frequency range of 800 Hz and 5 kHz, while at all other
frequencies, an increase has been observed. At the end of the fourth thermal cycle
stress, there is an increase between 5 kHz and 500 kHz while a decrease has been
observed at 400 Hz and 2 kHz.

Fig. 3. Imaginary part of permittivity v/s frequency

For a better understanding of thermal stress effect on the jacket and insulation,
reference frequencies have been selected to study the changing behavior of e00 . Figure 4
shows the behavior of e00 against the equivalent aging years at a reference frequency of
400 Hz, 500 Hz and 10 kHz. From the plot, it has been observed that at 400 Hz and
500 Hz there is decrease in the values of e00 with aging while an increase has been
observed at a higher frequency, 10 kHz.
266 E. Mustafa et al.

Fig. 4. Behavior of imaginary part of permittivity at 400 Hz, 500 Hz, and 10 kHz

4 Discussion

Since the dipole orientation has a great contribution to the real part of permittivity as
the insulation material used in cables are polar in nature. The general profile of the real
permittivity with the change to frequency can be attributed as the increase in frequency
restricts the dipole orientation. The initial decrease in the e0 values after the first thermal
cycle can be due to the moisture evaporation as the water is naturally polar. But the
increase after each cycle, can be recognized due to the structural changes in the
insulation material which resulted in the creation of dipoles. While the decrease at low
frequencies 20, 40 and 100 Hz shows that at the low frequencies the dipoles con-
tributing to polarization have decreased due to more structural changes happening in
the material due to thermals stress.
The imaginary part of permittivity, e00 is related to the leakage current loss or the
conductivity of the material. The decrease of e00 after the first thermal cycle, indicates
the evaporation of the moisture, as a result, the charges contributing to conduction are
restricted. While an increase of e00 in the frequency range between 400 Hz and 20 kHz
after the second thermal cycle shows that due to the structural changes happening
inside the material, the charged particles are contributing to the leakage current. The
increase of e00 at higher frequencies after the third thermal cycle, 10 kHz to 500 kHz
shows that the charge carriers present at higher frequencies have been created since at
the second thermal cycle at this range of frequency e00 decreased. This trend shows that
the thermal stress has strongly effected the structure of the XLPO. This change has
been shown as at lower frequency range, 400 Hz to 5 kHz, where the e00 values have
decreased. This means that inside the material new bonds have been created and the
material resistance has increased resulting in the decrease of charge carriers. The fourth
thermal cycle has a more profound effect on the structure of XLPO, as more increase in
the value of e00 at higher frequency has been noted while at low frequency the decrease
 Investigation of Complex Permittivity of XLPO Insulated PV DC Cables 267

in the values is more. This shows that XLPO has become stiffer with aging, as a result
at low frequency there are less charge carriers.

Fig. 5. Variation of the imaginary part of permittivity with aging

The change of e00 at lower and higher frequencies has been studied by plotting the
graph of e00 versus equivalent aging time (years) for three frequencies i.e., 500 Hz,
10 kHz and 200 kHz, Fig. 5. As discussed earlier at a lower frequency, 500 Hz, the e00
has decreased with aging while at higher frequencies, 10 kHz and 200 kHz it has
increased. The structural changes happening in XLPO due to the thermal aging have
resulted in more conduction charges in the high frequency range.
The shifting of minimum values of e00 to lower frequency from 2 kHz to 500 Hz
with thermal aging is shown in Fig. 6. After shifting of the minimum value at 500 Hz
at first thermal cycle, it remains at this frequency for the next three thermal cycles,
which shows that the polar or ionic particles at low frequencies have decreased which
were contributing to the conduction. The decrease in the values of e00 at low frequencies
and shifting of the minimum value to lower frequency shows that the phenomenon
contributing to the alteration to the dielectric property of the insulation material have
reached to either conclusion or stabilization after 40-80 years.
It should be noted that during the fabrication process of insulation material different
inorganic compounds are added for different purposes such as anti-oxidants, plasti-
cizers, flame retardants, and dyes. These contribute to the structural changes happening
inside the material due to aging and hence may affect the conduction and polarization
processes. Consequently, for a more in-depth understanding of the effect of thermal
aging on the XLPO material a good chemical analysis (TGA, DSC, and OIT) will be
helpful.
268 E. Mustafa et al.

Fig. 6. Shifting of the imaginary part of permittivity to a lower frequency

5 Conclusion

The effect of thermal aging on the LV PV DC cables has been studied in this paper. The
insulation material used in the cable was XLPO, semi-crystalline. The dielectric
property, real and imaginary part of permittivity was used to study the effect of aging.
The overall effect of thermal stress on XLPO has been an increase in the polarization
effect, which is shown as an increase in e0 . While the e00 showed a wide variation in its
behavior with aging. At low frequencies the value of e00 decreased with aging while at
high frequencies it increased. This behavior suggests that morphological changes have
happened inside the XLPO due to aging which are shown in the formation of dipoles, e0
and conduction charge particles, e00 .
Besides, a shift in the minimum value of e00 from the higher frequency of 2 kHz
after first thermal cycle to 500 Hz and then being stable for next three cycles is an
indication that the material has become stiffer and an end-of-point of insulation
degradation has been reached. As a result, its resistance at low frequency has increased.
A reliable chemical investigation technique will be helpful in better understanding the
discussed dielectric behavior of XLPO.

References
1. Verardi, L., Fabiani, D., Montanari, G.C., Zak, P.: Electrical condition monitoring techniques
for low-voltage cables used in nuclear power plants. In: 2013 IEEE Electrical Insulation
Conference (EIC), pp. 504–508. IEEE, New York (2013)
2. Verardi, L., Fabiani, D., Montanari, G.C.: Electrical aging markers for EPR-based low-
voltage cable insulation wiring of nuclear power plants. Radiat. Phys. Chem. 94, 166–170
(2014)
 Investigation of Complex Permittivity of XLPO Insulated PV DC Cables 269

3. Fantoni, P.F.: Condition monitoring of electrical cables using line resonance analysis (LIRA).
In: 17th International Conference on Nuclear Engineering (ICONE), pp. 171–178. ASME
(2009)
4. Asipuela, A., Mustafa, E., Afia, R.S.A., Tamus, Á.Z., Khan, M.Y.A.: Electrical condition
monitoring of low voltage nuclear power plant cables: tand and capacitance. In: IEEE 4th
International Conference on Power Generation Systems and Renewable Energy Technologies
(PGSRET), pp. 1–4. IEEE, New York (2018)
5. Mustafa, E., Tamus, Á.Z., Afia, R.S.A.: Thermal degradation and condition monitoring of low
voltage power cables in nuclear power industry. IFIP Advances in Information &
Communication Technology, vol. 553, pp. 405–413 (2019)
6. Csányi, M.G., Tamus, Á.Z., Iváncsy, T.: Investigation of central frequency and central loss
factor values on dicotylphthalate (DOP) plasticized PVC films. In: 2018 IEEE International
Conference on Diagnostics in Electrical Engineering (Diagnostika), pp. 1–4. IEEE, New York
(2018)
7. Csányi, M.G., Tamus, Á.Z., Kordás, P.: Effect of enhancing distribution grid resilience on low
voltage cable aging. IFIP Advances in Information & Communication Technology, vol. 521,
pp. 300–307 (2018)
8. Celina, M.C., Gillen, K.T., Li, E.R.: Nuclear power plant cable materials - review of
qualification and currently available aging data for margin assessments in cable performance.
Sandia National Laboratories, Albuquerque, New Mexico (2013)
9. Barsoukov, E., Macdonald, J.R.: Impedance Spectroscopy Theory, Experiment, and
Applications, 2nd edn. Wiley, Hoboken (2010)
Industrial Applications of High Voltage
 Influenced Voltages on Pipelines
by Overhead Lines Operated at AC, DC
and with Lightning Conditions

T. Huecker1(&) and V. Schröder2

1
University of Applied Sciences – HTW-Berlin, Berlin, Germany
thomas.huecker@htw-berlin.de
2
Steffel KKS GmbH, Lachendorf, Heidelberg, Germany

Abstract. The Pipelines for oil or gas are often build for many kilometers in
parallel to high voltage overhead lines. Modern ones are well isolated by
polyethylene coating for corrosion protection reason. Especially if short-circuit
or lightning currents are present on the overhead lines the influenced voltage on
the pipelines can reach very high magnitudes. To prevent deadly over voltages
pipelines are earthed every few kilometer. As this grounding is often performed
by 90 m deep drillings or huge flat earth electrodes to be less than 2 X, it can be
rather costly. In this paper, a self-developed software is used to calculated
pipeline over voltages and to optimize the number of earthing points. As
nowadays AC overhead lines get sometimes upgrade to DC applications for
capacity reasons expectable changes in influenced pipeline voltages are simu-
lated and discussed for short-circuit and lightning conditions.

Keywords: Influenced voltages
Overhead lines
AC
DC
Lightning

1 Introduction

The demand for energy is continuously growing. Therefore, many countries are
investing in upgrading overhead transmission lines (OHTL) to higher transport
capacities or building new ones. Also the network of oil and gas pipelines (OGP) is
growing continuously. To use available space economically OHTLs and OGPs are
sharing the same energy routs for many kilometres. The not perfectly balanced mag-
netic field of a 3 phase AC transmission line is inducing an electromagnetic force into
buried pipelines and depending on insulation and grounding of the pipeline the
resulting electric potential on the pipeline can reach dangerous levels for humans.
European standards [1] allow a maximum AC r.m.s. voltage of 60 V against ground.
To stay within this limit a pipeline needs to be grounded multiple times along the
energy route. As this grounding is often performed by 90 m deep drillings or huge flat
earth electrodes to be less than 2 X, it can be rather costly. Modern OGPs are well
isolated by polyethylene coating for corrosion protection reason. This worsens the
pipeline grounding situation resulting in higher needs for earthing.
Even higher electric potentials are induced in a pipeline when the OHTL is affected
by short circuit currents ore lightning strikes. For such short time transients higher

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 273–281, 2020.
https://doi.org/10.1007/978-3-030-31676-1_26
274 T. Huecker and V. Schröder

limits apply in the standards. Unfortunately, the different organizations do not rec-
ommend same values (see Table 1).

Table 1. Limits for short term overvoltage’s on pipelines

Standard Limits for short term overvoltage’s
TE 30 [2] 2000 V
EN 50443 [1] 1500 V
TE 7 [3] 100 V

Also DC overhead lines can induce voltages on OGPs. With stationary currents the
induced potential by inductive coupling is just caused by the current ripple (<3%) and
is therefore neglect able. For buried pipelines, also the capacitive coupling does not
need to be considered. However, the situation changes if an earth fault or lightning
phenomena occurs on the DC line. Explicit standards for the calculation of the influ-
ence of DC OHTLs on pipelines do not exist today. Please note that the first paragraph
of a section or subsection is not indented.
Whenever the configuration of an energy route is changed, limits of the induced
pipeline voltages for all operation modes must be checked and possibly the grounding
of the pipeline needs to be improved. This is not only required when a new OHTL or
OGP is build but also if the current rating is increased. In this paper situation is
discussed for the upgrading of a 380 kV AC transmission line with an additional DC
system.

2 Magnetically Induced Potential on Buried Pipeline

The calculation principle for induced voltages on buried conductors is well described in
the ITU standards [4]. Commercial software exists (e.g. CDEGS) and works well for all
cases where the current frequency on the OHTL is 50 or 60 Hz. For lightning or
switching transients with high frequency components some researchers using this
software report extremely high induced voltages [11, 12], which cannot be easily
explained. Therefore, here a self-programmed software is used based on the ITU model
[4] to get results on a different way. The calculation method used is based on the
following equations (Eqs. 1–4 and Fig. 1). For fast transient currents on the OHTL the
skin effect component d (Eq. 4) limits the induced pipeline voltages to reasonable
values when the relevant frequency of the transient currents is considered.
The buried pipeline is modelled as a transmission line with longitudinal EMF
(voltage) sources (Fig. 2 and Table 2). Were in Uij is given by sum of the counter
electromotive-forces along a considered segment.
In this pipeline model, the complex line to ground resistance ZE needs to be
individually calculated for each segment depended on its pipeline geometry, insulation,
 Influenced Voltages on Pipelines by Overhead Lines Operated at AC, DC 275

soil resistance and potential grounding at that point. The loop current Ii can be obtained
by inversion of the system.


V pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
E0 ¼ 1; 4
A2 þ B2  A
BÞ
ð1Þ
kA
km

d ðy0  y2 Þ2 þ x22 ða  x2 Þ2 þ y22

A ¼ ln
ln þ 22
ln ð2Þ
y20 þ a2 2
ðy0  y1 Þ þ x21 ða  x1 Þ2 þ y21

d ðy0  y3 Þ2 þ x23 ða  x3 Þ2 þ y23

B ¼ ln
ln þ 22
ln ð3Þ
y20 þ a2 ðy0  y1 Þ2 þ x21 ða  x1 Þ2 þ y21

d ¼ 658
qf ð4Þ

where: E′ = Induced voltage per km and kA

f = Frequency of the TL current (Hz)
q = Soil’s Resistivity in Ohm (Xm)
d = Skin effect component

Fig. 1. Geometry for ITU model calculation

276 T. Huecker and V. Schröder

Fig. 2. Calculation model for the pipeline voltage (comp. [5])

Table 2. Calculation model for the pipeline voltage

I0 I1 I2 I3
UI0 ZL þ ZE þ ZE ZE
UI1 ZE ZL þ ZE þ ZE ZE
UI2 ZE ZL þ ZE þ ZE ZE
UI3 ZE ZL þ ZE þ ZE

3 Modelling of Lightning and DC Fault Currents

To simplify transient signals, they have been approximated by linear current ramps
with a constant di/dt. Figure 3 shows the approximation for a lightning current of
10 kA hitting a power line (worst case: no back flashover considered; class III lightning
protection by earth wire). Lightning strikes hitting the overhead grounding wire can
have much higher amplitudes but their current is lead to the ground at each pole [13].
Their resulting geometrical length with significant current is much lower than from
lightning currents on a life conductor. This leads to smaller induced potentials on the
OGP.
Figure 4 shows the current approximation for a DC line to earth fault close to a
converter station (worst case reported in [6]). Other events on HVDC lines like circuit
breaker operation have lower rise times and produce lower induced voltages [7–9].

Fig. 3. Simplified lightning current on OHTL

 Influenced Voltages on Pipelines by Overhead Lines Operated at AC, DC 277

Fig. 4. Simplified DC line to earth fault current

4 Energy Route Characteristics

To obtain realistic results the geometry of a real life scenario was chosen: An existing
380 kV OHTL that runs for 31 km in parallel to a DN1200 gas pipeline (Fig. 5).

Fig. 5. Distance of pipeline to OHTL

The transmission line is modelled based on standard line geometry of a lattice tower
(2 circuits in barrel shape). The rated thermal max. current is 3,4 kA and the short
circuit current to earth is 13,2 kA. The pipeline material is steel (q = 0,16 µX m;
lr = 200) with a 3 mm PE insulation (q = 1 MX m).

5 Results

The resulting pipeline potentials for the rated current of 3,4 kA is shown in Fig. 6.
Without grounding the pipeline voltage to earth would reach a magnitude of 120 V.
This is significant above the 60 V limit of EN 50443. With 2 optimized grounding
points (4 X at 2 km and 3,5 X at 25 km) the voltage stays within acceptable limits.
278 T. Huecker and V. Schröder

Fig. 6. Pipeline potential at 3,4 kA rated thermal current limit

If a short circuit to earth happens, the investigated OHTL can have a maximum
short circuit current of 13,2 kA which is hopefully cleared in less than 1 s. For such
short events, EN 50443 allows a maximum potential rise to 1500 V. Figure 7 shows
the calculation result for such an event. Again the maximum voltage is significant
above the limit, if the pipeline is not grounded. Optimized earthing points (0,5 X at
2 km, 0,5 X at 25 km and 20 X a km 31) will help to stay within limits. The grounding
requirement resulting from this lightning calculations are tougher than from the rated
thermal current case above. A calculation using this 3 earthing points can show that the
potential stays also below 60 V at 3,4 kA rated current.

Fig. 7. Pipeline potential at 13,2 kA short circuit current

Currents from lightning events to a life wires of the OHTL result in much faster
current rise times. Some researchers report results in the range of 100 to 500 kV for the
pipeline voltage [11, 12] as the induced voltage is proportional to the product of di/dt
and the mutual impedance between OHTL and OGP. But this mutual impedance is not
constant for different frequencies (comp. skin effect Eq. 4). A fact, which is not always
modelled correctly and leads to a significant result change. For the calculation in this
paper the lightning current is separated in its frequency domain components and
 Influenced Voltages on Pipelines by Overhead Lines Operated at AC, DC 279

represented by their median frequency. With this approach the results shown in Fig. 8
are achieved. Voltages stay far below the 1500 V limit. No additional pipeline
grounding is required for lightning strikes. This result fits much better to the observed
behaviour at existing pipelines. Over voltages of a few 100 kV on the pipelines would
result in a breakdown of the 3 mm PE insulation, which are rarely reported. Further
measurements are required to prove the results.

Fig. 8. Induced pipeline Potential from 10 kA lightning current hitting the OHTL without any
pipeline grounding

HVDC transmission lines are used to increase the power capacity especially over
long distances. For economic reasons, but also for the public acceptance, existing AC
lines are getting sometimes upgraded by DC systems. In such cases the impact on
pipelines nearby needs to be investigated. The induced pipeline voltage at steady state
operation with <3% ripple is very small. But in the case of a line to earth fault currents
can rise very fast and induce an overvoltage on the pipeline. Figure 9 shows the
resulting overvoltage of a DC line to earth fault. The resulting potential is much higher
than for the lightning event described above but stays with 420 V, a value significantly
below the 1500 V limit.

Fig. 9. Pipeline potential resulting from a DC line to earth fault without any pipeline grounding.
280 T. Huecker and V. Schröder

6 Conclusions

The voltages induced on buried pipelines by OHTLs can be calculated well based on
the ITU model in case of:
• 50/60 Hz rated currents
• 50/60 Hz short circuit currents
• Currents from lightning events
• Currents of DC line to earth failures
Also the self-written calculation program can optimize the number of required earth
points. This helps to save cost for new installations and to check situation for the power
upgrade of existing overhead transmission lines. The results presented here indicate
that the grounding requirements resulting from an AC short circuit event are most
challenging and will often limit the induced voltages at normal operation conditions to
satisfying values.
Induced voltages from lightning events or DC line to earth failures reported here are
significantly lower than published at other points [11, 12]. In most cases lightning and
DC induced voltages will stay even without grounding within allowed limits. This has
to be proven by further measurements.

References
1. BS EN 50443. Effects of electromagnetic interference on pipelines caused by high voltage A.
C. electric traction systems and/or high voltage A.C. power supply systems, 12th edn., 31
January 2012
2. TE 30: Technische Empfehlung Nr. 30, Maßnahmen bei Bau und Betrieb von Rohrleitungen
im Einflussbereich von Starkstromanlagen mit Nennspannungen über 1 kV, Technisches
Komitee für Beeinflussungsfragen, (VEÖ), Ausgabe April 1987
3. TE7: Maßnahmen beim Bau und Betrieb von Rohrleitungen im Einflussbereich von Hoch-
spannungs-Drehstromanlagen und Wechsel-strom-Bahnanlagen, SfB, Germany, February
2014
4. ITU-T Directives, Volume V: Inducing currents and voltages in power transmission and
distribution systems (2008)
5. Cercev, L., Janokov, V., Filiposki, V.: Induced voltages in metallic pipelines near power
transmission lines. In: Japanese-Bulgarian-Macedonian Joint Seminar on Applied Electro-
magnetics. IEEE Macedonian Section, pp. 108–115, 269 p. (2002). ISBN 9989-57-094-9
6. Yanushkevich, A., Scharrenberg, R., Kell, M., Smeets, R.: Switching phenomena of HVDC
circuit breaker in multi-terminal system. In: 11th IET International Conference on AC and
DC Power Transmission, Birmingham, 10–12 February 2015
7. Xu, Z., Xiao, H., Xiao, L., Zhang, Z.: DC fault analysis and clearance solutions of MMC-
HVDC systems. MDPI Energies 11, 941 (2018)
8. Callavik, M., Blomberg, A., Häfner, J., Jacobson, B.: The hybrid HVDC breaker. An
innovation breakthrough enabling reliable HVDC grids. ABB Grid Systems, Technical
Paper, November 2012
 Influenced Voltages on Pipelines by Overhead Lines Operated at AC, DC 281

9. Mokhberdoran, A., Carvalho, A., Leite, H., Silva, N.: A review on HVDC circuit breakers.
In: 3rd Renewable Power Generation Conference (RPG 2014), Naples, Italy, 24–25
September 2014
10. Baeckmann, W.V., Paul, H.-U., Feist, K.-H.: Contribution to the interference and conductor
acting as earth electrodes. CIGRE-Report 36-02 (1982)
11. AlShahri, A.S.A.: Assessment of induced potentials on pipelines from high voltage power
lines, thesis. The University of Auckland (2016)
12. Peabody, A.W., Verhiel, A.L.: The effects of high-voltage AC transmission lines on buried
pipelines. IEEE Trans. Ind. Gen. Appl. IGA-7(3), 395–402 (1971)
13. Liu, X., Wang, W., Yu, H.: Electromagnetic effect on underground pipeline of the lightning
strike 330 kV transmission lines. In: 2017 4th International Conference on Information
Science and Control Engineering (ICISCE) (2017)
 The Variation of Electric Field
on the Conductor Surface Characterized
by Space Charge Density

Ruijin Liao1, Hongbo Liu1(&), Xuetong Zhao1, Lulu Ren1,

and Yuandi Lin2
1
State Key Laboratory of Power Transmission Equipment and System Security
and New Technology, Chongqing University, Chongqing 400044, China
cqulhb@cqu.edu.cn
2
State Grid Jiangsu Electric Power Company Research Institute,
Nanjing 211100, Jiangsu, China

Abstract. The accuracy of surface electric field (SEF) is important for the
research of corona discharge phenomenon, both in numerical simulation and
actual application. However, it is widely considered that the SEF is invariable
after corona occurs (Kaptzov’s assumption). This theory makes corona study
more convenient, but slightly inconsistent with the actual situation. In this work,
SEF for the negative corona discharge in a coaxial cylinder’s model is inves-
tigated through the space charge density. Space charge density is measured by
sound pulse method, which is a non-contact measurement method. It can avoid
the introduction of interference in the study of SEF. By selecting the maximum
charge density (charge density around the conductor) closely related to the SEF
as the research objective, the variation of SEF intensity can be obtained. Results
show that the values of charge density increase with the rise of applied voltage
and the diameter of the conductor. Through the maximum charge density near
the wire, the variation formula of SEF can be obtained. It shows that SEF is not
a fixed value and not equal to the onset field intensity, and it has a decreasing
trend with the applied voltage rising after corona initiates.

Keywords: Sound pulse method
Space charge density
Surface electric field

1 Introduction

Corona discharge is a partial self-sustained discharge of gaseous medium in inhomo-

geneous electric field, and is the most common form of gas discharge [1]. Corona
phenomenon has a positive effect on numerous industrial applications and scientific
research such as aerospace, steel and biomedical applications [2–4]. However, problems
such as noise and corona loss caused by corona discharge not only affect people’s lives
but also cause huge energy waste. Therefore, corona discharge has caused widespread
concern among researchers all over the world. Surface electric field (SEF) has been
proved an effective factor to study corona, both in experiments and simulation. SEF can
be used to study corona loss and audible noise [5–7], besides, it also can be as boundary

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 282–291, 2020.
https://doi.org/10.1007/978-3-030-31676-1_27
 The Variation of Electric Field on the Conductor Surface 283

condition in the numerical simulation [8–10]. Therefore, the study of SEF intensity has a
positive effect on studying corona discharge.
At present, ultra-high voltage direct current (UHVDC) transmission lines have been
widely constructed in China [11]. Due to the increase of voltage grade, applied voltage
and wire radius varies greatly from previous projects, the effect of space charge is more
significant. This has created a growing interest in the study of charge distribution on dc
transmission lines. Changes in voltage grade and conductor size have been proved to be
important effect on the corona phenomenon. This means it also has affects the variation
of SEF intensity. Among the numerous indicators for evaluating corona discharge, SEF
intensity is a vital factor to study corona phenomenon.
In most researches, Kaptzov’s assumption (the electrical field strength remains
constant and equal to the onset value after the corona occurs) is widely adopted [12,
13]. However, this assumption is not consistent with the actual variation of SEF. In
order to solve this problem, many attempts to measure or calculate SEF had been
carried out. For measurements, Penney [14, 15] used the potential mapping methods
and projectile methods to study the SEF intensity. However, the results obtained by
above methods fluctuated greatly and difficult to obtain the exact value. Then, Waters
[16] used cylindrical electrostatic flux meter to measure the electric field intensity. He
made some progress in the law of SEF variation, but the conductor line used in this
system must be hollow and a rotor kept running in a high speed during the measure
process. This is not only inconsistent with the condition of actual conductor line but
also leads to a great distortion of electric field due to the introduction of the rotor.
In this work, space charge density is adopted as a tool to study the variation of SEF.
It has been proved an effective way to evaluate the intensity of corona discharge, shown
in reference [17]. In this paper, space charge density close to the conductor is extracted
to characteristic the SEF intensity. Based on the charge density, the maximum charge
density ratio is introduced to determine the variation of the SEF.

2 Numerical Simulation

Finite element method is set up to calculate charge density distribution. When applied
voltage in the wire exceeds inception voltage, corona occurs. Governing equations
about electric field and charge density are as followed:
q
r2 u ¼  ð1Þ
e0

r
j¼0 ð2Þ

j ¼ kqE ð3Þ

E ¼ ru ð4Þ

where u is the electric potential, q is charge density, e0 is the air permittivity in a

vacuum, j is the electric current density, k is ion mobility, and E is the electric field
intensity.
284 R. Liao et al.

Considering that the physical mechanism of corona discharge is highly compli-

cated, to simplify the nonlinear problem, the following assumptions were adopted.
(a) The thickness of the corona layer around the conductor is sufficiently small com-
pared with the radius of the conductor line, which can be neglected; (b) The mobility of
the ions is constant, and their diffusion effect is neglected.
The boundary conditions for finite element method can be divided into electric
potential conditions and electric field conditions. For the electric potential,
u = V (applied voltage) is specified on the surface of wire and u = 0 is specified on the
grounding end. For the electric field on the surface of the wire, surface electrical field is
equal to the onset field strength (Kaptzov’s assumption). In order to correct the criterion
of SEF strength, different electric field strengths are applied in the simulation to
compare with measured results. Through the comparison results, the variation of SEF
and the effect of Kaptzov’s assumption on the simulation can be obtained.

3 Experimental Principle and Facilities

3.1 The Experimental Principle
After a negative DC voltage is applied to the wire, the space charge appears in the air
around the wire. When the sound wave propagates in the air, the sound wave will
vibrate the air and space charge in its propagation area. The vibratory space charge can
produce an electric field (E-field) signal that can be received by an antenna. The
distribution of charge density can be obtained by analyzing the electric field signal.
After obtaining the distribution characteristics of space charge density, the maximum
space charge density close to the wire is selected to study the change of SEF.

3.2 The Space Charge Measurement Facilities

The electrode configuration is shown in the Fig. 1. A corona cage combined with
charge density measurement equipment is set up in the laboratory. The cylinder radius
is set as 15 cm, and the length of the corona cage is 100 cm. The conductor in the axis
of the cylinder is applied negative DC voltage through high voltage power supply, and
the cylinder is grounded. PC, control board, power board and DC voltage source are
combined together to drive ultrasonic transducer. The power board is energized by a
DC voltage source which can provide a maximum 600 V DC voltage output. Half-
bridge inverter circuit in power board can provide a square-wave voltage whose
parameters are controlled by control board. PC communicates with control board
through Bluetooth. Thus, all the parameters of square-wave voltage can be directly set
in PC. The frequency of square-wave voltage is 25 kHz. The conductor in this work is
aluminum and has a diameter of 0.15 cm. The ambient temperature ranged from 25 °C
to 30 °C, the air pressure was normal, and relative humidity ranged from 55% to 60%.
 The Variation of Electric Field on the Conductor Surface 285

Ultrasonic transducer

Support frame
(insulation)

Grounding end
High-voltage end
Wire

Fig. 1. The figure of electrode configuration and wire sample

4 Experimental Principle and Facilities

4.1 Charge Density Under Different Applied Voltages
The space charge density diagram under different applied voltages was obtained based
on the conductor (line radius is 0.15 cm). For this measurement, the space charge
density was measured at four different applied voltages. Therefore, the space charge
density distribution of the four applied voltages were obtained and shown in Fig. 2.
Two obvious characteristics can be observed from the charge distribution diagrams.

1.6
Simulation-34kV
C/m3)

1.4 Simulation-35kV
Simulation-36kV
-4

1.2 Simulation-37kV
Experiment-34kV
Space charge dsensity(

1.0 Experiment-35kV
Experiment-36kV
0.8 Experiment-37kV

0.6

0.4

0.2

0.0
0 2 4 6 8 10 12 14
Distance(cm)

Fig. 2. Distribution of space charge density (wire radius = 0.15 cm)

First, measured charge density distribution has the same trend and same order of
magnitudes as the simulated result. The simulated results are calculated by the finite
286 R. Liao et al.

element method. The distribution trend for different applied voltage is similar. Nev-
ertheless, there are still some deviations between experiment and simulation.
Second, the distribution of charge density in air was influenced by applied voltage
significantly. The space charge density increased with an increase of applied voltage.
Obviously, high voltage promoted the generation of space charge. The effect of applied
voltage on charge density distribution at the position closing to the grounding end is
not obvious. As shown in Fig. 2, when the measured position was close to the
grounding end, the curves of different voltages are almost consistent with each other.
However, at the position closing to the conductor surface, the effect of voltage on
charge distribution was significant.

4.2 Charge Density Under Different Conductor Radii

In the construction of transmission lines, different voltage levels require a corre-
sponding size of wire, which also has effect on the corona discharge. The charge
distribution under different radii was drawn in Fig. 3. Experimental results are
extracted form results of different applied voltage. These experiments in laboratory can
provide theoretical basis for actual transmission line.

Fig. 3. Space charge density distributions under different wire radii when the applied voltage is
−34 kV

The curves show a higher distribution of space charge density under a conductor line
with a radius of 1.5 mm, compared with the charge density under a conductor line with
the radius of 1.0 mm. According to the Peek’s formula, the thick wire has a smaller
electrical field. In addition, Liao [17] discussed the relationship between the charge
density and electrical field strength. The results of the research indicated that electrical
field strength can inhibit the increase of charge density. This explains why the conductor
with a radius of 2.0 mm produces greater charge density than that of a 1.0 mm con-
ductor. On the other hand, they are basically consistent in the grounding end. This
changing rule is same with the charge distribution under numerical simulation. On the
other hand, the measured maximum charge density under the radius of 1.0 mm has a gap
 The Variation of Electric Field on the Conductor Surface 287

with the corresponding simulation, whereas, it is very close to the simulated value under
the radius of 1.5 mm. This phenomenon is discussed in next part.

5 Discussion

In order to study the variation of SEF, the space charge density near the wire is
introduced because it has the highest charge density in this area. High charge density is
easy to test and the measuring deviation is smaller. This can reflect the changing of E-
field more accurate. In this part, maximum charge density ratio, n, was used to evaluate
the error between experiment and simulation, which is defined as:
qsmax
n¼ ð5Þ
qemax

where qs-max is the maximum charge density by simulation (10−4C/m3), qe-max is the
maximum charge density by experiment (10−4C/m3). The maximum charge density ratio
shows the difference in value between simulated results and measured ones. Maximum
charge ratio n close to 1 means that the error between experiment and simulation is small.
Comparing the maximum charge density of simulation and experiment (shown in
Fig. 4), it can be found that maximum charge density ratio decreases with the increasing
applied voltage. It means that the deviation between simulation and experiment depends
on some factors that affect the distribution of space charge, such as the applied voltage. In
the previous work [18], this phenomenon is attributed to the SEF intensity that is not
constant and should be lower than the onset electrical field strength. Some new numerical
simulations were carried out to analyze the SEF, but the results were not satisfactory
because of lacking the measured data. In this work, different applied voltages and wire
radii were introduced to correct the criterion of SEF.

0.86 1.0mm
1.5mm
Maximum charge density ratio m

2.0mm
0.84

0.82

0.80

0.78

0.76
30 32 34 36 38 40 42 44 46
Applied voltage (kV)

Fig. 4. The variation of maximum charge density ratio for different applied voltages and wire radii
288 R. Liao et al.

Table 1. Maximum charge density ratio of wires

with radius of 1.5 mm under different Eonset
Radius Applied SEF(Eonset) n
(mm) voltage
(kV)
1.5 −34 1.0000 0.854
0.9800 0.891
0.9600 0.930
0.9400 0.969
0.9200 0.999
0.9000 1.042
−35 1.0000 0.846
0.9800 0.871
0.9600 0.935
0.9400 0.947
0.9200 0.987
0.9100 1.000
0.9000 1.027
−36 1.0000 0.841
0.9800 0.877
0.9600 0.906
0.9400 0.943
0.9200 0.981
0.9050 0.999
0.9000 1.020
−37 1.0000 0.827
0.9800 0.864
0.9600 0.898
0.9400 0.934
0.9200 0.970
0.9000 1.000
0.8980 1.100
0.8800 1.109
−38 1.0000 0.823
0.9800 0.861
0.9600 0.894
0.9400 0.927
0.9200 0.954
0.9000 0.991
0.8950 1.000
0.8800 1.030
 The Variation of Electric Field on the Conductor Surface 289

53

Surface electrical field (kV/cm)

52
Calculated value
51 Fitted curve
Kaptzov's assumption
50

49

48

47

33 34 35 36 37 38
Applied voltage (kV)

Fig. 5. Comparison of SEF between Kaptzov’s assumption and calculation in this paper

It can be found that difference value between simulation and experiment increases
with the applied voltage, however, maximum charge density ratio decreases with the
applied voltage under the same wire radii. At the same applied voltage (−34 kV), the
difference value between measurement and simulation of thick wire is smaller than that
of thin wire. However, the maximum charge ratio n of thick wire is greater than that of
thin wire under the same situation. This further illustrates the impact of discharge
intensity on our work. This indicates that the difference between calculation and
measurement depends on measurement conditions, i.e. applied voltage and wire radius.
In this paper, different voltages and wire radii were introduced to correct the SEF. Take
a wire with radius of 1.5 mm as an example. The results are shown in Table 1. It is
obvious that the maximum charge density ratio n changing with the variation of the
SEF. Under the appropriate and low electrical field, the ratio n is equal to 1 roughly.
This means the experimental results is consistent with the simulated results when SEF
is lower than the onset electrical field. Similar change rule can be found when wire
radius is 1.0 and 2.0 mm. It indicates that the actual electrical field after corona occurs
is lower than the onset field. Whereas Kaptzov’s assumption says that SEF is constant
after corona occurs. This is the main reason why the simulated value is always smaller
than the measured values. According to the experiment and calculation, the fitted curve
of the SEF is shown in Fig. 5. The fitting formula of electric field strength varying with
applied voltage is as follows:
U
E ¼ 47:6 þ 2:4  1016
e 0:9 ð6Þ

where E is the surface electrical field strength, U is the applied voltage. The R-square
and Reduced Chi-Sqr are 0.994 and 0.025 respectively. The large R-square and small
Reduced Chi-Sqr show that the fitting curve is very close to the experimental results.
The fitted results show that actual SEF should be lower than the onset field strength.
290 R. Liao et al.

In this paper, applied voltage and wire radii are adopted to correct the calculation rule
of SEF, but corona discharge is influenced by more factors, such as relative humidity,
temperature, atmospheric pressure. So the full assumption correction and surface field
strength calculation should consider all the factors that affect it. Only in this way, the
correct formula can be applied to practical engineering. The relevant work is under
way.

6 Conclusion

In this work, a device for measuring the space charge density in a gas medium based on
the sound pulse method is established. Through this device, the distribution of space
charge density under different applied voltages and different sizes of wires is obtained.
On the basis of the known charge density distribution, the maximum charge density
close to the wire is selected to study the variation of the electric field strength on the
wire surface. The experimental results show that the surface electric field intensity is
not maintained at the value of the onset electric field and remains unchanged, but has a
decreasing tendency as the discharge intensity increases. This can correct the Kaptzov’s
assumption and improves the accuracy of the simulation calculation, meanwhile, it also
helps to further understand the corona discharge mechanism.

Acknowledgments. All the authors thank the National Basic Research Program of China (Grant
No. 2011CB209401), National Natural Science Foundation of China (Grant No. 51877016), the
National Key Research and Development Plan (2016YFB0900700) for the financial support
provided and Professor Kazimierz Adamiak for his suggestions to the paper revision.

References
1. Zhou, Z., Cui, X., Lu, T.: Measurements of space charge density under HVDC conductor
based on aspirator-type charge meter. High Voltage Eng. 42(9), 2962–2967 (2016)
2. Sun, H., Lu, B., Wang, M.: The role of photoionization in negative corona discharge: the
influences of temperature, humidity, and air pressure on a corona. Phys. Plasmas 24(10),
103506 (2017)
3. Modi, A., Koratkar, N., Lass, E.: Miniaturized gas ionization sensors using carbon
nanotubes. Nature 424(6945), 171 (2003)
4. Bussiahn, R., Brandenburg, R., Gerling, T.: The hairline plasma: an intermittent negative dc-
corona discharge at atmospheric pressure for plasma medical applications. Appl. Phys. Lett.
96(14), 143701 (2010)
5. Qiao, J., Zou, J., Li, B.: Calculation of the ionised field and the corona losses of high-voltage
direct current transmission lines using a finite-difference-based flux tracing method. IET
Gener. Transm. Distrib. 9(4), 348–357 (2015)
6. He, W., He, J., Wan, B.: Influence of altitude on radio interference level of AC power lines
based on corona cage. IET Sci. Measur. Technol. 9(7), 861–865 (2015)
7. Li, X., Cui, X., Lu, T.: The correlation between audible noise and corona current in time
domain caused by single positive corona source on the conductor. IEEE Trans. Dielectr.
Electr. Insul. 22(2), 1314–1320 (2015)
 The Variation of Electric Field on the Conductor Surface 291

8. Maruvada, P.S.: Electric field and ion current environment of HVdc transmission lines:
comparison of calculations and measurements. IEEE Trans. Power Deliv. 76(18), 4153–
4159 (1971)
9. Zhang, B., Mo, J., Yin, H.: Calculation of ion flow field around HVdc bipolar transmission
lines by method of characteristics. IEEE Trans. Magn. 51(3), 1–4 (2015)
10. Takuma, T., Ikeda, T., Kawamoto, T.: Calculation of ion flow fields of HVDC transmission
lines by the finite element method. IEEE Trans. Power Apparatus Syst. PAS-100(12), 4802–
4810 (1981)
11. Bian, X., Yu, D., Meng, X.: Corona-generated space charge effects on electric field
distribution for an indoor corona cage and a monopolar test line. IEEE Trans. Dielectr.
Electr. Insul. 18(5), 1767–1778 (2011)
12. Cladt, J.J., Gary, C.H.: Predetermination of corona losses under rain: experimental
interpreting and checking of a method to calculate corona losses. IEEE Trans. Power
Apparatus Syst. PAS-89(5), 853–860 (1970)
13. Zhou, X., Cui, X., Lu, T.: A time-efficient method for the simulation of ion flow field of the
AC-DC hybrid transmission lines. IEEE Trans. Magn. 48(2), 731–734 (2012)
14. Penney, G.W., Hewitt, J.G.: Some measurements of abnormal corona. Trans. Am. Inst.
Electr. Part I: Eng. Commun. Electron. 77(3), 319–327 (1958)
15. Penney, G.W., Matick, R.E.: Potentials in DC corona fields. Trans. Am. Inst. Electr. Part I:
Eng. Commun. Electron. 79(2), 91–99 (1960)
16. Waters, R.T.: A cylindrical electrostatic flux meter for corona studies. J. Phys. E: Sci.
Instrum. 5(5), 475 (1972)
17. Liao, R., Liu, K., Liu, H., Wu, F.: Measurement of space charge in negative corona on a
small corona cage. IEEE Trans. Dielectr. Electr. Insul. 23(4), 2344–2352 (2016)
18. Zheng, Y., He, J., Zhang, B.: Surface electric field for negative corona discharge in
atmospheric pressure air. IEEE Trans. Plasma Sci. 39(8), 1644–1652 (2011)
 Two-Dimensional Stray Loss Calculation
of High Voltage Transformer

Joo HyunWoo(&), Bae ChaeYoon, Choi JongUng,

and Kim YoungGeun

LS Industrial Systems, 68, Wolmyeong-ro #201, Heung deok-gu,

Cheongju-si, Chungcheongbuk-do, South Korea
hwjoo@lsis.com

Abstract. In this paper, a new approach to calculate stray loss in power trans-
former is presented. Three-dimensional Finite Element Method (FEM) is widely
used to calculate a local overheating in consequence of the loss. However, there
are limits to use three-dimensional (3D) FEM in the beginning of design process
due to its calculation time and the absence of 3D model. And it is difficult to apply
two-dimensional (2D) FEM or axi-symmetric method for electromagnetic cal-
culation of high voltage transformer because rectangular shaped tank wall and
cylindrical winding core should be solved simultaneously. In order to overcome
the difficulties of usage of 2D/3D FE calculation, 2D finite element method
combined with numerical interpolation is used for leakage flux on the tank wall.
And then stray loss is calculated by considering skin depth effect on tank walls.
Finally, the loss and calculation time obtained by employing the proposed method
are compared with 3D calculation results. From this result, it is proved that stray
loss method in this paper can reduce expensive design modifications before
testing.

Keywords: High voltage transformer
Stray loss
Finite Element Method

1 Introduction

With the increase of capacity of power transformer, stray loss caused by magnetic leakage
flux in the structure part of transformer constitutes a significant part of the total losses.
Due to uneven distribution of the stray losses, local overheating in various metal parts of a
high voltage transformer can be generated and be a root cause of a transformer failure.
Because the stray losses cannot be measured directly and there is no standardized
method of determining them, a transformer manufacturer mainly relies on their cal-
culation of the loss by using empirical equation from experiment [1]. Recently, many
researcher have investigated on the losses calculation by adopting 3D FE calculation.
However, 3D method cannot be applied without 3D model and can be time-consuming
process because of non-linearity and geometrically thin penetration depth [2, 3].
In this paper, stray loss on a transformer tank wall, which has more than 80% of
total stray losses, by using two-dimensional (2D) finite element method combined with
numerical interpolation is calculated. The results were analyzed and compared with 3D
calculation results.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 292–299, 2020.
https://doi.org/10.1007/978-3-030-31676-1_28
 Two-Dimensional Stray Loss Calculation of High Voltage Transformer 293

2 Surface Impedance

At the surface of good conductors, relationship between the tangential component of

the electric filed and the tangential component of magnetic field can be approximately
defined as Eq. (1) below [1].

ET ¼ ZðmÞ ½HT  n ð1Þ

Where ZðmÞ is the complex surface impedance of the metal material.

Under assumption that the most losses occur in the range of depth of penetration
and the magnetic field is constant in whole depth of penetration, then the surface loss
per unit area of the conductor can be calculated by Eq. (2);

W ¼ 0:5  RefZðmÞgjHTj2 ð2Þ

p
Where: RefZðmÞg ¼ ðxl=2rÞ
x = Angular frequency (rad)
l = permeability (H/m)
r = conductivity (S/m)

3 Stray Loss Calculation

From Eq. (2), the stray loss calculation can be concluded how to calculate the tan-
gential component of magnetic field at the tank wall, especially by applying 2D FE
method in this research. A power transformer generally consists of rectangular shaped
tank wall and cylindrical winding core as shown in Fig. 1. In order to solve the
tangential component of magnetic field by 2D FE method, cross section along Y-Z
axis, A-A’, was selected and the leakage flux distribution of a simulated model in the
cross section is shown in Fig. 2.

Fig. 1. Plan view of a high voltage transformer.

294 J. HyunWoo et al.

In the simulation, magnetomotive force of high voltage winding (N1I1) is set equal
to that of low voltage winding (N2I2) and impedance boundary condition is applied on
the surface (line in case of 2D). 2D FE analysis in this paper was carried out with the
aid of FEMM [4].

Fig. 2. Plan view of a high voltage transformer.

A difficulty in applying 2D FEM to stray loss calculation is to calculate a magnetic

leakage flux on shaped tank wall considering 3D geometry of a high voltage trans-
former. In order to overcome this difficulty, 2D FE method and numerical interpolation
skill is adopted in this paper. The magnetic leakage flux in the plan view along Z axis
as shown in Fig. 2 is the flux on A1 point in Fig. 3. Then, the distance between A1
point and B1 point can be set as a base distance (d1).

Fig. 3. Schematic view for loss calculation.

 Two-Dimensional Stray Loss Calculation of High Voltage Transformer 295

Suppose that a leakage flux on a certain point of tank wall, Ak in Fig. 3, has linear
relationship with that on the base point, A1, then the leakage flux on Ak point can be
simply calculated using Eq. (3).

Hk ¼ H1  ðd1 =dk Þ ð3Þ

Where dk is the minimum distance between High voltages three phase winding
(Bk) and tank wall (Ak), H1 is the leakage flux on A1 and Hk is the leakage flux on Ak.
The leakage flux distribution on A1 and Ak is shown in Fig. 4.

Fig. 4. Leakage flux distributions along Z axis.

Leakage flux distribution on the all sampling points in Fig. 3 can be obtained by
calculating the minimum distance between winding and tank wall and Eq. (4). Sam-
pling was done in every 10 mm on the tank wall and every 10 rad on the winding in
this research.
Here is the summery of the proposed method.
• Step1 Solve 2D FEM in order to get leakage flux distribution as shown in Fig. 4.
• Step2 Set the obtained field intensity as a base value.
• Step3 Calculate each minimum distance between Ak and Bk.
• Step4 Set the distance d1 (between A1 and B1) as base distance.
• Step5 Find the minimum distance on the Ak point to the winding point (Bk) and
calculate the distance from Ak to Bk (set this distance as dk)
• Step6 Calculate leakage field intensity on the point Ak by proportional equation,
Hk = H1  (d1/dk).
• Step7 Evaluate stray loss on the tank walls by integrating the stray loss result,
calculated by Eq. (2) on each sampling points.
296 J. HyunWoo et al.

4 Simulation Result

The load loss of a transformer consists of DC loss (I2R), eddy current loss of winding
and stray loss. DC loss can be calculated by Joule heating equation and eddy current loss
by 2D electromagnetic simulation. However, it is impossible to measure the stray loss
separately. In this study, 3 phase, 60 Hz, 350/390 MVA and 161/20 kV model is
selected and the calculation results were compared with 3D loss calculation results.
Tables 1 and 2 tabulate current input and number of turn in this simulation respectively.

Table 1. Current input

Windings Current [A]
Minimum tap Rated tap Maximum tap
1st low voltage 6500 6500 6500
2nd low voltage 6500 6500 6500
High voltage 1473.3 1398.7 1331.3
High voltage tap 1473.3 1398.7 1331.3

Table 2. Number of turn

Windings Number of turn [N]
Minimum tap Rated tap Maximum tap
1st low voltage −34 −34 −34
2nd low voltage −34 −34 −34
High voltage 300 300 300
High voltage tap 0 16 32

Figure 5 shows leakage flux distribution for each tap conditions tabulated in
Tables 1 and 2.

Fig. 5. Leakage flux distributions for each tap condition

 Two-Dimensional Stray Loss Calculation of High Voltage Transformer 297

The leakage flux has the highest value at the maximum tap condition and the hot
shot due to local temperature rise should be critical in the maximum condition.
However, the maximum flux position in case of the minimum tap condition is different
from the case of the maximum tap. So, the hot spot on the tank wall can be changed in
each tap conditions. It is important to design a magnetic shunt to prevent from local
heating on the tank wall. In order to consider the effect magnetic shunt, the shunt is
modelled as line in 2D model and its magnetic properties is set as an exactly same
magnetic characteristic as the main core in the simulation. Figure 6 shows the effect of
magnetic shunt on leakage flux at the rated tap condition.

Fig. 6. Effect of magnetic shunt on leakage flux distribution

By using the proposed method and the field distributions as shown in Figs. 5 and 6,
the stray losses are calculated and compared with 3D FE calculation results (Table 3).

Table 3. Simulation results

Minimum tap Rated tap Maximum tap
Without shunt 2D [kW] 32.5 40 53.4
3D [kW] 35.2 44.2 54.2
Error [%] 7.7 9.5 1.5
With shunt 2D [kW] 18.9 24.5 32.1
3D [kW] 17.7 21.1 27
Error [%] 6.7 16.1 18.9

The error between 2D and 3D simulation without considering magnetic shunt is up

to 10 [%] and 20 [%] for simulation with magnetic shunt. The error in case of magnetic
shunt can be increased because the slot, for reduction of circulating magnetic flux, on
the surface of magnetic shunt along X axis direction in Fig. 1 cannot be included in 2D
298 J. HyunWoo et al.

simulation and there is no effect of magnetic nonlinear effect of magnetic shunt in 2D

simulation. However, the proposed method can be useful because the calculation time
by 2D method can remarkably be reduced by approximately 1000 times compared with
3D calculation.

5 Software for 2D Stray Loss

A software for 2D stray loss calculation based on the proposed method is developed as
design tool. A designer can evaluate the stray loss in the early stage of design process
without 3D modelling. Figures 7 and 8 shows GUI interface for core geometry and
winding information respectively.

Fig. 7. GUI interface for core geometry

Fig. 8. GUI interface for winding information

 Two-Dimensional Stray Loss Calculation of High Voltage Transformer 299

6 Conclusions

The results in this study show that 2D FE analysis combined with numerical inter-
polation can be applied to a stray loss calculation of a power transformer without
determining 3D model in the beginning of a transformer design. The proposed method
was tested and verified by comparing example transformer. Finally, its validity and
availability was proved by comparing with 3D calculation results and time. This
method thus yield advantages of reduced calculation time and complexity in the
transformer design.

References
1. Holland, S.A., O’Connell, G.P., Haydock, L.: Calculating stray losses in power transformers
using surface impedance with finite elements. IEEE Trans. Magn. 28(2), 1355–1358 (1992)
2. Li, Y., Li, L., Jing, Y., Zhang, B.: 3D finite element analysis of the stray loss in power
transformer structure parts. Energy Power Eng. 5, 1089–1092 (2013)
3. Olivares-Galvan, J.C., Escarela-Perez, R., Georgilakis, P.S.: Reduction of stray losses in
Flange–Bolt regions of large power transformer tanks. IEEE Trans. Ind. Electron. 61(8)
(2014)
4. Meeker, D.: Finite Element Method Magnetic - Version 4.2
 Influence of Different Factors
on the Interruption Characteristics
of Paralleled of High-Voltage SF6 Circuit
Breakers with a Highly Coupled Split Reactor

Haibo Su1(&) , Wenxiong Mo1, Yong Wang1 , Lin Gan1,

Junxiang Liu1, Ze Guo2, Le Gu1 , Lu Zhu1, Li Chen2,
and Shengya Qiao1
1
Guangzhou Power Supply Co., Ltd.,
Guangzhou 510410, People’s Republic of China
shbfighter@qq.com
2
State Key Laboratory of Electrical Insulation and Power Equipment,
Xi’an Jiaotong University, Xi’an 710049, People’s Republic of China

Abstract. Paralleled high-voltage SF6 circuit breakers with a highly coupled

split reactor (HCSR) have an excellent interruption performance, because it is
expected that this structure could realize the function of automatic current-
sharing when the current flow through both branches and automatic current-
limiting when the current flow through only one branch. In this paper, a
mathematical model had been built for this interrupting device by coupling the
equivalent circuit equations into the magneto-hydro-dynamics (MHD) arc
model. The influence of different factors including different operation moments,
different filling pressures and different inductances are investigated. The
dielectric recovery strength after the interruption processes is also investigated.
The simulation results show that the paralleled high-voltage SF6 circuit breakers
with a highly coupled split reactor have excellent interruption performance. The
automatic current-sharing and current-limiting functions of the HCSR under
different interruption conditions all show excellent performance, and the 252 kV
paralleled SF6 circuit breakers with HCSR has successfully interrupted
252 kV/101 kA short circuit current in state laboratory of China recently.

Keywords: Paralleled SF6 circuit breakers
Highly coupled split reactor

MHD modeling
Dielectric recovery strength

1 Introduction

The parallel technology of the circuit breaker is an important way to solve the high
current breaking issue [1, 2]. The key problem of the parallel breaking technology of
the circuit breaker is how to realize the current sharing of the parallel branch during the
breaking process to ensure that the current of each branch does not exceed the capacity
of the circuit breaker. The paralleled SF6 circuit breakers with a highly coupled split
reactor (HCSR) could realize the function of automatic current-sharing when the

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 300–307, 2020.
https://doi.org/10.1007/978-3-030-31676-1_29
 Influence of Different Factors on the Interruption Characteristics 301

current flow through both branches and automatic current-limiting when the current
flow through only one branch. The HCSR shall prevent the short circuit current from
transferring sharply from one break to the other effectively without bringing in a large
inductance to the power grid.
In this paper, two 252 kV–63 kA circuit breakers are paralleled to interrupt an
85 kA short-circuit current. A mathematical model had been built for this interrupting
device by coupling the equivalent circuit equations into the magneto-hydro-dynamics
(MHD) arc model [3]. What’s more, the dielectric recovery process is evaluated by
comparing the electric field with the critical dielectric strength data predicted by
solving the two-term Boltzmann equation. Based on the simulation model, the inter-
ruption processes with different operation moments, different filling pressures and
different inductances are simulated.

2 Simulation Method

The simplified equivalent electrical circuit of the paralleled high-voltage SF6 circuit
breakers with HCSR was shown in Fig. 1. S was a 252 kV alternating voltage source.
Ls and Rs represented the equivalent line reactance (5.45 mH) and line resistance
(4 mX) respectively. The portion surrounded by the dotted line was the paralleled
circuit breakers with a HCSR, which was to be investigated in detail in this paper. The
split reactor of the HCSR has an inductance of 2.82 mH. The mutual inductance
coefficient of the HCSR is 0.98.

Ls
Rs
i

HCSR

i1 i2
+ + +
S us CB1 u1 u2 CB2
- - -

Fig. 1. A Simplified equivalent circuit of the paralleled circuit breakers with a HCSR

When there is a short-circuit situation, both CB1 and CB2 shall operate to interrupt
the short-circuit current. At first, the current flow through both CBs and this period is
called the current-sharing period because the HCSR shall make sure that CB1 and CB2
share the short-circuit current. After a current-zero point, the arc in one CB might
extinguish while that in the other CB may restrike. Hereafter, the current shall flow
302 H. Su et al.

through only one CB and the HCSR shall work as a current limiting reactor, the period
of which is named the current-limiting period.
During the current-sharing period, the short-circuit current flowed through both
branches. The equivalent electrical circuit equations during the current-sharing period
were given by:

di1 di2 di2 di1

L1 M þ u 1 ¼ L2 M þ u2 ð1Þ
dt dt dt dt
i1 þ i2 ¼ i ð2Þ

u1 ¼ f1 ði1 Þ ð3Þ

u2 ¼ f2 ði2 Þ ð4Þ

where L1 and L2 were the self-inductances of the two arms of the HCSR respectively,
M was the mutual inductance, i1 and i2 were the currents in the two branches, u1 and u2
were the arc voltages of the paralleled circuit breakers CB1 and CB2, i was the total
short-circuit current. Equations 3 and 4 represent the dependency of the arc voltages on
the arc currents in CB1 and CB2, which was obtained through the MHD arc model.
During the current-limiting period, the short-circuit current flowed through only
one branch, for example CB1. The equivalent electrical circuit equations during this
period were described as:

di1
ðL1 þ Ls Þ  þ Rs  i1 þ u1 ¼ us ð5Þ
dt
u1 ¼ f1 ði1 Þ ð6Þ

where us was the voltage of the voltage source. The short-circuit current was assumed
to flow through only the branch of CB1 reasonably.
The dynamic arc behaviors during the interruption process were simulated by
adopting a 2D MHD arc model which took account of the Ohmic heating, the elec-
tromagnetic effect (the Lorenz force), the nozzle ablation, the radiation loss and the
turbulence enhanced mass, momentum and energy transplant. Based on the arc MHD
simulation, the pressure and temperature distributions inside the arc quenching
chamber can be obtained during the dielectric recovery period and thus the dielectric
recovery process is evaluated. The evaluation method adopted in this paper is similar to
our previous work [4]. Using the calculated Ecr data, the distribution of the critical
electric-field strength in the arc chamber can be predicted. At the same time, according
to the applied recovery voltage and the configuration of the contacts and nozzle, the
distribution of the electric field Ea can be calculated too. By calculating the ratio of Ea
to Ecr, the dielectric breakdown probability of each point can be drawn. Ultimately, the
critical dielectric strength can be obtained as:
 Influence of Different Factors on the Interruption Characteristics 303

Ua
Ucr ¼ ð7Þ
ðEa =Ecr Þm

where the ðEa =Ecr Þm is the maximum value of the ratio of Ea to Ecr and Ua is the
recovery voltage applied.

3 Simulation Results
3.1 Influence of Operation Moment Differences
Figure 2 shows the typical simulation results of the paralleled circuit breakers with a
HCSR. In this simulation, the different operation moments is set as the cause of the
different current situation. The difference is set as 1 ms. The first half-wave is the
current-sharing period and the second half-wave is the current-limiting period. The
simulation results show that the current-sharing function of the HCSR is quite satis-
fying. The difference between the currents in the two branches is incomparable with the
short-circuit current. The current peak value in the current-limiting period is 78.7 kA,
corresponding to an RMS current of 55.7 kA. The HCSR is able to reduce the short-
circuit current to a value below the interruption limit of the CBs.

Fig. 2. Typical simulation results of the paralleled circuit breakers with a HCSR

Figure 3 shows the current difference in the paralleled branches in the current
sharing period with different operation moment differences. It can be seen that the
current difference keeps increasing during the current-sharing period. With the increase
of the operation moment difference, the current difference in the paralleled branches
increases. When the operation moment differences are set as 0.5 ms, 1 ms and 3 ms,
the maximums of the current difference are 35.9 A, 60.8 A and 240.2 A respectively.
304 H. Su et al.

Fig. 3. Current differences between the paralleled CBs with different operation moment
differences

In the current-limiting period, the current peaks with operation moment differences
of 0.5 ms, 1 ms and 3 ms are 78.70 kA, 78.71 kA and 78.78 kA, showing a satisfying
current-limiting function of the HCSR. After the current-limiting period, the dielectric
recovery strength are calculated and presented in Fig. 4. It can be seen that the
dielectric recovery strength is apparently higher than the applied recovery voltage.
Also, the dielectric recovery decreases with the increase of the operation moment
difference.

Fig. 4. Dielectric recovery strength of CB1 with different operation moments

3.2 Different Filling Pressures Different Inductances

In actual operation, there may be leak in the CBs. If one of the paralleled CB has a leak
and lower filling pressure, the interruption performance shall be worse. Thus, the filling
pressure of CB1 is set as 0.5 MPa to study the interruption performance under this
 Influence of Different Factors on the Interruption Characteristics 305

condition because CB1 extinguishes later than CB2. The calculated current difference
under this situation is shown in Fig. 5. It can be seen that the current difference is
smaller when there is a leak in CB1, showing a better current-sharing effect.

Fig. 5. Current difference between the paralleled CBs with different filling pressures

Figure 6 shows the dielectric recovery strength of CB1 with different filling pres-
sures. The dielectric recovery strength of CB1 is apparently lower when the filling
pressure decreases from 0.6 MPa to 0.5 MPa. However, the dielectric recovery
strength of CB1 with a filling pressure of 0.5 MPa is still higher than the applied
voltage during most of the time after current-zero. Thus, the chance of dielectric
recovery failure is very low under this condition.

Fig. 6. Dielectric recovery strength of CB1 with different filling pressures

306 H. Su et al.

3.3 Different Inductances

In The inductance of the split reactor is an importance parameter of the HCSR and shall
have importance influence on the interruption performance of the paralleled CBs with a
HCSR. Figure 7 shows the current differences between the paralleled CBs during the
current-sharing period with different inductances. Obviously, the increase of the
inductance shall reduce the current difference and improve the current-sharing effect.
The maximum of the current differences with the inductances of 2.82 mH and 4.2 mH
are 60.8 A and 41.1 A respectively.

Fig. 7. Current differences between the paralleled CBs with different filling inductances

Figure 8 shows the arc current in the paralleled CBs with an inductance of 4.2 mH
in the HCSR. The current peak of CB1 in the second half-wave is 67.5 kA, which is
smaller than that in the 2.82 mH case. Figure 9 shows the dielectric recovery strength
of CB1 with different inductances. The simulation results show no apparent difference.

Fig. 8. Arc current in the paralleled CBs with an inductance of 4.2 mH in the HCSR
 Influence of Different Factors on the Interruption Characteristics 307

Fig. 9. Dielectric recovery strength of CB1 with different inductances

4 Conclusions

In this paper, a simulation model is developed for paralleled CBs with a HCSR by
coupling the circuit equations with the MHD arc model. Two 252 kV-63 kA gas CBs
are paralleled to interrupt an 85 kA short-circuit current. The conclusions are drawn as
follows:
(1) The paralleled CBs with a HCSR show satisfying current-sharing and current-
limiting effects.
(2) The current difference between the paralleled CBs shall increase with the increase
of operation moment difference, and decrease with the increase of the inductance
as well as the decrease of filling pressure in CB1.
(3) The dielectric recovery strength calculation results show that the paralleled CBs
with a HCSR have excellent interruption performance.

References
1. Liu, Z.Y., Wang, J.M., Wang, Z.Y., Yuan, S.: Current transfer process in open operation
between parallel breaks of vacuum generator circuit breaker. In: Proceedings of ISDEIV
Conference, Xi’an, China, vol. 2, pp. 403–406 (2000)
2. Pertsev, A.A., Chistjakov, S.P., Rylskaya, L.A., Chistjakov, R.S.: Parallel connection of
several vacuum interrupters in a circuit-breaker pole. In: Proceedings of ISDEIV Conference,
Xi’an, China, vol. 2, pp. 407–410 (2000)
3. Guo, Z., Li, X.W., Su, H.B., et al.: Interrupting characteristics of paralleled SF6 circuit
breakers with a highly coupled split reactor. IEEE Trans. Compon. Packag. Manuf. Technol. 7
(5), 768–776 (2017)
4. Jiang, X., Li, X., Zhao, H., Jia, S.: Analysis of the dielectric breakdown characteristics for a
252-kV gas circuit breaker. IEEE Trans. Power Deliv. 28, 1592–1599 (2013)
 Investigation of High Altitude/Tropospheric
Correction Factors for Electric
Aircraft Applications

Athanasios C. Mermigkas(&), David Clark, and A. Manu Haddad

AHIVE Research Centre, Cardiff University,

The Parade, Cardiff CF24 3AA, Wales, UK
mermigkasa1@cardiff.ac.uk

Abstract. With rising fuel costs and CO2 emissions, the aviation industry is
moving rapidly toward increased electrification of aircraft, and power demand
for propulsion and safety critical systems necessitates a move to on-board dis-
tribution voltages in excess of 1 kV. The increased stress experienced by cable
insulation, connectors and other equipment, combined with extreme and
dynamic environmental conditions experienced in flight, presents a number of
technical challenges.
This research project proposes to quantify the effect of atmospheric conditions
on partial discharge thresholds in uprated aircraft electrical systems, and to
derive atmospheric correction factors appropriate to in-service operating con-
ditions to assist the aircraft electrical design engineer in the insulation coordi-
nation of modern aviation systems.
The development of a controlled atmospheric test facility for the precise
replication of in-flight conditions is outlined, and an initial visual assessment of
partial discharge activity in an ex-service wire harness at a range of pressures is
presented. We also present plans for the ongoing development of the facility and
test capabilities.

Keywords: Atmospheric correction factors
Corona
Partial discharge

Electric breakdown
Cable
Electric aircraft

1 Introduction

The civil aviation industry is pursuing a more-electric aircraft concept, with a view to
achieving higher system efficiency through electrification of traditional mechanical and
hydraulic aircraft subsystems [1, 2]. This shift is considered a stepping stone to real-
ising a new generation of hybrid and all-electric aircraft. Such aircraft will be equipped
with diverse and safety-critical sub-systems [3], based on high density power electronic
converters and operating at system voltages in excess of 1000 V. The increased
dielectric stress experienced by insulation, connectors and other equipment, combined
with extreme and dynamic flight conditions, presents several technical challenges for
the electrical design engineer.
The voltage uprating of on-board electrical power systems necessitates control of
electrostatic phenomena which will arise within the normal operating regime of the

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 308–315, 2020.
https://doi.org/10.1007/978-3-030-31676-1_30
 Investigation of High Altitude/Tropospheric Correction Factors 309

aircraft [4, 5], and there are a number of factors that must be taken into account when
assessing the performance of such systems, such as:
Increased use of switched dc power and its influence on insulation stress [6];
Increased system voltages [7–9]; increased frequency and severity of temporary over-
voltages; Changes in altitude and atmospheric conditions (temperature, pressure and
humidity) and appropriate altitude correction factors [10–12]; Icing and condensation;
Indirect effects induced by lightning strikes [13].
There is to date very little published work concerned with dynamic atmospheric
effects on ascent and descent, the influence of condensation and icing or the implica-
tions of short term system- and atmospherically-induced over-voltages. A fundamental
understanding of these parameters in medium-voltage systems is key to enabling
increased electrification of aircraft power distribution, and this places renewed
emphasis on insulation coordination and the mitigation of partial discharges as oper-
ational conditions shift toward the Paschen minimum at high altitudes [14].
The research project introduced in this paper proposes to quantify the effect of
atmospheric conditions on partial discharge thresholds in uprated electric aircraft, and
to derive atmospheric correction factors appropriate to in-service operating conditions.
From the published literature there are considerable differences between existing cor-
rection methods, with greater deviation between methods at lower air pressures [10].
This paper discusses the development of a controlled high-voltage test chamber for
the precise replication of in-flight environmental conditions and outlines the ongoing
development of test capabilities in this area. We also present results from the com-
missioning phase of the facility, comprising a visual assessment of partial discharge
activity in an ex-service equipment wire harness at a range of atmospheric pressures.

2 Simulated Altitude Test Facilities

2.1 System Overview

The test chamber, shown in Fig. 1, is the result of a major refurbishment of test
equipment previously employed in the investigation of pre-breakdown ionisation at
high gas pressures [15], and later work defining atmospheric correction factors for high
voltage test standards [11, 12].
With reference to the schematic shown in Fig. 2, the system comprises a stainless-
steel vacuum vessel (1) having internal volume *0.8 m3, clad in a cryogenic insula-
tion jacket (2). A high-voltage feedthrough (3) traverses an oil-filled bushing (4)
mounted in the roof of the vessel. Test objects are loaded and unloaded via the main
access port (5) and energised with reference to a ground plane (6), internally isolated
from the vessel wall.
Visibility of the test object is via a CaF2 viewport (7) and sensors integrated via a
reconfigurable feedthrough panel (8). Pressure control is by dry scroll pump (9) and
isolation/relief valve assembly (10, 11), connected to the vessel via a ceramic break.
Temperature control is achieved using a recirculating chiller (12) feeding a heat
exchange coil (13) and internal circulation fans (14).
310 A. C. Mermigkas et al.

Fig. 1. Simulated-altitude HV test chamber

Fig. 2. Component breakdown of the simulated altitude test chamber.

 Investigation of High Altitude/Tropospheric Correction Factors 311

Control and monitoring is by modular data-acquisition system (15), connected via

optical link (16) to the main control PC (17).
The developed facility permits high voltage testing at temperatures between −40
and +70 °C and at partial vacuum down to 50 mbar, conditions to which aircraft cables
are routinely subjected. Discussion of secondary systems for the investigation of
condensation and icing has been omitted as these were not employed in the work
presented here.

2.2 Condition Monitoring

Vessel conditions are monitored using a high performance pressure transducer, dew-
point transmitter and four K-type thermocouples, entering the vessel via feedthroughs
in the side port. Sensor power and signal acquisition is fully administered through a
National Instruments c-DAQ chassis, equipped with a range of Analogue, Digital I/O
and Thermocouple conditioning modules.

2.3 Control Signalling

Pump control and HV amplifier gate signalling is achieved using the modular DAQ,
employing Analogue Out and Solid-state relay modules, thereby eliminating any risk of
the user coming into direct electrical contact with equipment during a test.

3 Experimental Setup

In order to investigate the dielectric performance of aerospace cables under various

atmospheric conditions, an experimental set-up was devised, comprising HV waveform
generation, test vessel, data acquisition system and an image capturing device as
depicted in Fig. 3.
Low-voltage test waveforms were generated using an Agilent 33220A 20 MHz
function generator.

Fig. 3. Experimental setup for high voltage arbitrary waveform excitation under partial vacuum
conditions.
312 A. C. Mermigkas et al.

These were fed to the input of a TREK Model 30/20A High Voltage Power
Amplifier, having a gain of 3000 V/V, its output terminal in turn to the vessel HV
feedthrough. Inside the vessel a twisted pair of insulated aircraft equipment wire was
arranged, with one strand connected to the HV electrode and the other bonded to the
ground plane. The conductors used are ex-service, with a nominal insulation rating of
600 V, and were extracted from a cable harness of a decommissioned passenger aircraft
(115 V/400 Hz).
On the outside of the vessel, a Nikon D7200 camera was set up to capture long-
exposure images of the test object when subject to different HV excitation and pressure
conditions. A purpose-built 3D printed shroud was used to mount the camera’s lens
against the CaF2 window, ensuring that no light entered the vessel during test. Figure 4
shows the camera view of the test object, illuminated prior to sealing the vessel.
As a precursor to experimental validation of aerospace cable performance in future
more electric aircraft, preliminary experiments were conducted to observe corona dis-
charge around cables energised at a range of voltages and under reduced atmospheric
pressures. HV waveforms were selected to represent those likely to be found in modern
and future aircraft: 400 Hz AC, switched DC and steady DC with impressed ripple.

Fig. 4. Bright-field image of the test wire pair prior to sealing (Red - HV, Yellow - Ground)

4 Results

The experimental results are presented in Fig. 5 as long-exposure digital images,

comparing the visual corona of different energisation waveforms under identical
atmospheric conditions (partial vacuum). Photographs were taken at 30 s exposure,
with the camera ISO set to 2000 and an aperture of f5.6. With reference to Fig. 5,
column (a) shows the result of 400 Hz AC excitation at 3 kV rms. Column (b) is for
switched DC at ±3 kV, 750 V/µs slew rate and 50% duty cycle to achieve an
equivalent rms.
 Investigation of High Altitude/Tropospheric Correction Factors 313

Fig. 5. Long-exposure imaging of corona activity on aircraft equipment wires at different air
pressures; (a) AC 3 kVrms/400 Hz, (b) Switched DC ±3 kV/400 Hz. (ISO:2000; f/5.6;
Exposure: 30 s; T = 20 °C).

For the 400 Hz AC energisation, a plot of a function of the onset of visible corona
discharge on the surface of the insulated cables as a function of pressure is shown in
Fig. 6. Circles represent the presence of corona discharge, while the crosses its
absence. Experiments were conducted by applying a waveform of sufficient amplitude
to cause corona discharges, and reduced by regular intervals until corona was no longer
detected. The dashed line indicates a near linear relationship with air pressure, con-
sistent with [16, 17]. As is expected from literature, as the pressure increases, corona
threshold voltage also increases [18].
314 A. C. Mermigkas et al.

Fig. 6. Corona visibility threshold as function of pressure at 400 Hz AC excitation.

Steady DC excitation with impressed 1 kHz ripple at 2% was performed up to

7.5 kVdc at 100 mBar with no corona activity recorded at the stated camera sensitivity.

5 Conclusions

The present work was aimed at investigating the suitability of aerospace cables for
future more electric aircraft operating at elevated power densities. To these ends an
experimental system has been designed and constructed, in order to simulate flight
conditions at cruising altitude, and dynamic conditions experienced during take-off and
landing.
Test waveforms were selected to represent those likely to be found in modern and
future aircraft, namely 400 Hz AC and switched DC, and corona activity observed
using a camera having fixed acquisition settings.
Initial results were found to be in accordance with the literature; corona onset
occurring at lower voltages in lower pressures. Future work will be based around the
precise determination of partial discharge inception and extinction in aerospace elec-
trical systems, the influence of condensation and icing in these systems, and the
development of appropriate correction factors.

Acknowledgements. This work was completed with the support of the Engineering and
Physical Sciences Research Council (EPSRC) under grant ref. EP/R012881/1.
The authors would also like to thank Prof. Emer. R. T. Waters for his expertise and contri-
bution to the project.

References
1. Rosero, J.A., Ortega, J.A., Aldabas, E., Romeral, L.: Moving towards a more electric
aircraft. IEEE Aerosp. Electron. Syst. Mag. 22(3), 3–9 (2007)
2. Sarlioglu, B., Morris, C.T.: More electric aircraft: review, challenges, and opportunities for
commercial transport aircraft. IEEE Trans. Transp. Electrif. 1(1), 54–64 (2015)
 Investigation of High Altitude/Tropospheric Correction Factors 315

3. Boglietti, A., Cavagnino, A., Tenconi, A., Vaschetto, S.: The safety critical electric machines
and drives in the more electric aircraft: a survey. In: IECON Proceedings (Industrial
Electronics Conference), pp. 2587–2594 (2009)
4. Bilodeau, T.M., Dunbar, W.G., Sarjeant, W.J.: High-voltage and partial discharge systems.
IEEE Electr. Insul. Mag. 5(2), 34–43 (1989)
5. Dunbar, W.G.: AFAPL-TR-76-41: High voltage design guide for airborne equipment,
pp. 10–202 (1976)
6. Taghia, B., Billard, T., Carayon, J., Malec, D., Piquet, H., Belinger, A.: Investigations on
partial discharges risk in aeronautical rotation machine fed by HVDC 540 V DC network.
In: IEEE Electrical Insulation Conference (EIC), pp. 491–494 (2018)
7. Nya, B.H., Brombach, J., Schulz, D.: Benefits of higher voltage levels in aircraft electrical
power systems. In: Electrical Systems for Aircraft, Railway and Ship Propulsion, ESARS
(2012)
8. Cotton, I., Nelms, A., Husband, M.: Higher voltage aircraft power systems. IEEE
Aerosp. Electron. Syst. Mag. 23(2), 25–32 (2008)
9. Madonna, V., Giangrande, P., Galea, M.: Electrical power generation in aircraft: review,
challenges, and opportunities. IEEE Trans. Transp. Electrif. 4(3), 646–659 (2018)
10. Riba, J.-R., Larzelere, W., Rickmann, J.: Voltage correction factors for air-insulated
transmission lines operating in high-altitude regions to limit corona activity: a review.
Energies 11(7), 1908 (2018)
11. Ortega, P., Waters, R.T., Haddad, A., Hameed, R., Davies, A.J.: Impulse breakdown
voltages of air gaps: a new approach to atmospheric correction factors applicable to
international standards. IEEE Trans. Dielectr. Electr. Insul. 14(6), 1498–1508 (2007)
12. Davies, A.J., Hameed, R., Ortega, P., Waters, R.T.: Air breakdown at high altitude. In:
Proceedings of Lightning & Mountains 1994, p. 214 (1994)
13. Petrov, N.I., Haddad, A., Griffiths, H., Waters, R.T.: Lightning strikes to aircraft radome:
electric field shielding siumlation. In: Proceedings of the 17th International Conference on
Gas Discharges and Their Applications, pp. 513–516 (2008)
14. Alrumayan, F., Cotton, I., Nelms, A.: Partial discharge testing of aerospace electrical
systems. IEEE Trans. Aerosp. Electron. Syst. 46(2), 848–863 (2010)
15. Davies, W.E.V.J., Dutton, J., Harris, F.M.: An apparatus for the investigation of pre-
breakdown ionization in gases at high voltages, high gas pressures and large electrode
separations. J. Sci. Instrum. 43(7), 457–461 (1966)
16. Hu, Q., Shu, L., Jiang, X., Sun, C., Zhang, S., Shang, Y.: Effects of air pressure and humidity
on the corona onset voltage of bundle conductors. IET Gener. Transm. Distrib. 5(6), 621
(2011)
17. Abadie, C., et al.: Influence of pressure on partial discharge spectra. In: EIC 2016 (Electrical
Insulation Conference), pp. 507–510 (2016)
18. Kuffel, J., Zaengl, W.S., Kuffel, P.: High Voltage Engineering Fundamentals, 2nd edn.
Elsevier, New York (2000)
 Evaluation of Wind Turbine Earthing System

S. A. Pastromas1(&), I. A. Naxakis1, S. I. Xerra2, K. N. Koutras1,

and E. C. Pyrgioti1
1
Department of Electrical and Computer Engineering, University of Patras,
Rion-Patras, Greece
{pastromas,naxakis,e.pygioti}@ece.upatras.gr,
koskoutr@gmail.com
2
Department of Physics, University of Patras, Rion-Patras, Greece
soxerra@gmail.com

Abstract. The majority of the wind farms are constructed in high altitudes and
isolated areas which put in risk the wind turbines due to their exposure to
lightning activity and their high total height. Wind turbine earthing system is
part of its lightning protection design according to IEC61400-24 and shall
ensure a safe path for the lightning current hitting the wind turbine and a safe
working environment for the live beings. The earthing system contributes also in
earth fault current dissipation and is considered in the electrical studies. In the
current paper, the performance of a wind turbine earthing system of an existing
wind farm with medium soil resistivity is discussed, based on simulations in
ATP-EMTP which is suitable for steady-state and transient studies. Two cases
(Case A and Case B) were examined relatively to the lightning current, i.e. 5 kA
and 150 kA for the same concrete and soil resistivity, applied at the base and the
top of the tower, respectively. In Case A, two occasions were considered by
using or not an extra vertical rod placed at the base of the tower. In Case B, two
scenarios were assumed by considering or not the ionization effect. It was
noticed the reduction of the voltages and currents, when the ionization effect was
regarded. The results can be used to evaluate the earthing system performance
and the need for improvement to ensure a reliable operation of the whole asset
mitigating the operational risk for the equipment.

Keywords: Earthing
Lightning
Wind turbine
Safety

1 Introduction

European Community has established ambitious targets for renewable energy pene-
tration in the electrical grids. The recent agreement has set the goal of increasing the
penetration of renewables by 32% till 2030. The main contribution on this purpose
according to European commitment is expected to be covered by wind farms. Most of
wind energy projects are developed in areas with complex terrain and usually in
altitude higher than 500 m. In such conditions, wind turbines are put at risk, since they
are exposed to lightning activity usually ranging from medium to high. The soil of
these areas is also of high resistivity adding an extra factor that has to be taken into
account in a wind farm development. Wind turbine earthing system is a part of its

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 316–328, 2020.
https://doi.org/10.1007/978-3-030-31676-1_31
 Evaluation of Wind Turbine Earthing System 317

lightning protection design [1] and shall ensure a safe path for the lightning current
hitting the wind turbine and a safe working environment for the live beings. The
earthing system also contributes to earth fault current dissipation and is an important
part of the electrical studies.
In the current paper, the performance of a wind farm developed in the island of
Crete in Greece, which consists of seven wind turbines, is studied. The performance of
the earthing system is investigated regarding voltage and current dissipation in a wind
turbine operating with nominal characteristics considering soil conditions and maxi-
mum acceptable earth fault current. The study is based both on simulations with ATP-
EMTP software, which is suitable for steady-state and transient analysis, and on cal-
culations based on the requirements of the relevant standards [1–3]. The results
demonstrate how effective and safe the earthing system of the wind turbines is and
point out the possible need for extra measures. These results are of high importance
since the expected lifetime of the equipment of the relevant wind farms is about twenty
years (Fig. 1).

Fig. 1. Upward lightning strike at the top of wind turbine [4].

2 Wind Turbine Earthing System Evaluation

and Methodology

A wind turbine (WT) earthing system is an extended earthing system which usually
covers the area where a wind turbine is placed. The earthing system of a wind farm is
comprised by the interconnection of the earthing systems between wind turbines and
the substation. The wind turbines are interconnected and finally are connected in a
medium voltage (MV) switching station. The MV substation and wind farm earthing
systems are interconnected via buried earthing wire. For the purposes of the current
study, the earthing system of one wind turbine will be investigated.
318 S. A. Pastromas et al.

2.1 Wind Turbine Earthing System

The wind turbines of the wind farm have the same earthing system which consists of:
(i) the grounding embedded in the reinforcement, (ii) external earthing in the soil. All
earthing rings are interconnected via a double wire, while the same wire is used for the
interconnection of rest of wind turbines with the MV substation. The earthing system of
each wind turbine is shown in Fig. 2. The soil resistivity is 500 Xm. In the current
paper, different cases were studied with different lightning peak currents.

Fig. 2. Wind turbine earthing system under investigation.

2.2 Earthing System Modelling in ATP-EMTP

The wind turbine earthing modeling in ATP-EMTP is based on the circuit theory, using
the pi equivalent circuit [5–7], as it is demonstrated in Fig. 2. This theory has been
chosen taking advantage of its simplicity and its ability to simulate taking into account
the soil ionization [6, 7].

Fig. 3. Pi equivalent circuit used for wind turbine “earthing” modelling.

A grounding electrode buried horizontally in the ground under the impact of a

lightning strike can be represented by the network of distributed parameters of Fig. 3.
An elementary section of the conductor composes the serial resistance (ri) and
inductance (Li) as well as the conductivity (Gi) and capacitance (Ci) [6].
 Evaluation of Wind Turbine Earthing System 319

According to [8], the resistance and inductance remain unchanged relatively to the
soil ionization region. Therefore, the ionization affects only the elements of the parallel
branch. The formulas for the calculation of the per-unit length parameters are given and
clarified below:


l0 2li
Li  : ln  1 ð1Þ
2p a
2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi3
q 42h þ a li þ l2i þ a2 a l i þ l 2
þ 4h 2
2h
 1 þ ð Þ2 5
i
ri ¼ þ ln  1 þ ð Þ2 þ ln
2pli li a li 2h li

ð2Þ

2peli
ci ðai Þ ¼ pffiffiffiffiffiffiffiffiffiffi ð3Þ
ai li þ l2i þ a2i qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ai 2
ffi
li þ ln ai 1 þ ð li Þ

ci
Gi ¼ ð4Þ
e:q

where:
q = soil resistivity (Xm)
li = the total length of the electrode (m)
a = the radius of the electrode (m)
h = the burial depth of the electrode (m)
l0 = the magnetic permeability of vacuum, i.e. 4 p  10−7 Hm−1
e = dielectric constant of the soil (F/m)
ai = the equivalent radius of the ground electrode, when ionization is regarded (m)
In order the capacitance Ci to be calculated, the equivalent radius of the ground
electrode ai (when the soil ionization is included) is primarily calculated from the
following equation in accordance with Mousa criterion [9]:

q:Imi
ai ¼ ð5Þ
2pli E0

where:
Imi = the current in part i (A)
E0 = 300 kV/m, the ionization field (V/m)
The structure of wind tower is a cone, which is usually made of steel. Due to it is
not a standard cylinder, it is difficult to calculate its electrical parameters. The height of
the tower is generally in dozens or even more than one hundred meters, so the tower
320 S. A. Pastromas et al.

can be approximate as a hollow cylinder, when calculating its electrical parameters.

The inductance value of the tower can be calculated by Eq. (6) [7]:


l0 2H l
Lt ¼ : ln  1  ln c ð6Þ
2p rt l0

where:
H = the height of the tower (m)
rt = the average radius of the tower (m).
c = the parameter which is determined by the ratio of the inside radius and outside
radius.
The resistance of the tower can be calculated by Eq. (7).

H
Rt ¼ q ð7Þ
S
where:
q = the resistivity of the tower (Xm)
S = the conducting part of the tower cross section (m2)
The tower capacitance to the ground can be calculated by Eq. (8) [7]:

2peH
C0 ¼ ð8Þ
lnð2H=rt Þ

Table 1. WT tower technical data

Parameter Value Unit
Height 80 m
Tower diameter 3.5 m
Surface 0,438 m2
Resistivity 1,5*10−7 Xm
l/l0 100 –
Tower thickness 0,02 m

Table 1 depicts the wind turbine technical data. The wind turbine earthing system
model in the ATP-EMTP is depicted in Fig. 4. In the same figure, the tower of the wind
turbine is also depicted, which was modelled in the same software too.
 Evaluation of Wind Turbine Earthing System 321

Fig. 4. Wind turbine earthing system in ATP-EMTP.

3 Simulation Results

As it is presented in the previous section, the wind turbine earthing system consists of
several earthing rings, which are interconnected achieving a common earthing system.
This wind turbine earthing system is interconnected with the neighboring wind turbines
creating a wind farm earthing system which will be of low earthing impedance and
efficient for its purpose, i.e. the lightning current and earth fault current dissipation. In the
current section, the simulation results are presented, considering different peak currents.
It was also assumed a single wind turbine of the cretan wind farm. Before simulation
execution, the expected striking distance Rs was calculated [10–12] for the two values of
lightning peak current Ip1 = 5 kA, 10/350 ls and Ip2 = 150 kA, 10/350 ls:

RS ¼ 10  lp0:65 ð9Þ

where:
Rs = Striking distance (m)
Ip = Peak current (kA)
The striking distance for Ip1 = 5 kA is Rs1 = 28.5 m and for Ip2 = 150 kA is
Rs2 = 259.7 m. So, it is expected that only a lightning of 5 kA or less, has the
322 S. A. Pastromas et al.

probability to hit the base of tower and a lightning of 150 kA will hit the top or the
blades of wind turbine. The cases that have been simulated are: (a) foundation of
reinforced concrete with resistivity qcon = 50 Xm, surrounding soil of 500 Xm and
peak current Ip = 5 kA and (b) foundation of reinforced concrete with resistivity
qcon = 50 Xm, surrounding soil of 500 Xm and peak current Ip = 150 kA.

3.1 Lightning Strike Case a (50/500/5)

In the first case a lightning strike was simulated at the base of the wind turbine tower.
The peak current is 5 kA and resistivities are for the reinforced foundation concrete
q1 = 50 Xm and the surrounding soil q2 = 500 Xm. In Figs. 5, 6, 7 and 8, the voltages
that appear in the tower and the earthing system are depicted, while Figs. 9 and 10
demonstrate the currents in selected branches of high importance.

Fig. 5. Voltages in node A with and without the extra rod.

Fig. 6. Voltages in node B with and without the extra rod.

 Evaluation of Wind Turbine Earthing System 323

Fig. 7. Voltages in node C with and without the extra rod.

Fig. 8. Voltages in node D with and without the extra rod.

Fig. 9. Current in branch B-2 with and without the extra rod.
324 S. A. Pastromas et al.

Fig. 10. Current in branch D-E with and without the extra rod.

From the simulation results, it is obvious that in the outer ring, defined by branch
D-E, the current is very low, so there is no risk of occurrence of step and probably
touch voltages. As for the use of the extra rod, the figures above prove that its existence
causes overvoltage and current reduction in all the nodes, but not significantly. Last but
not least, the voltages appearing at the base of the tower could cause touch voltages,
although there is limited probability for the personnel to get in contact with the tower at
the time of the strike.

3.2 Lightning Strike Case B (50/500/150)

In this case, a lightning strike was simulated at the top of the wind turbine tower with
peak current of 150 kA and resistivities similar to the ones of the case A. In Figs. 11, 12,
13, 14 the voltages that appear in the tower and the earthing system are depicted, while
Figs. 15 and 16 demonstrate the currents in selected branches of high importance.

Fig. 11. Voltages in node A with and without considering the ionization effect.
 Evaluation of Wind Turbine Earthing System 325

Fig. 12. Voltages in node B with and without considering the ionization effect.

Fig. 13. Voltages in node C with and without considering the ionization effect.

Fig. 14. Voltages in node D with and without considering the ionization effect.
326 S. A. Pastromas et al.

Fig. 15. Current in branch B-2 with and without considering the ionization effect.

Fig. 16. Current in branch D-E with and without considering the ionization effect.

Simulation results show the presence of high voltages (MV) at the point of the
strike, which took place at the top of the tower. These voltages are reduced at the first
earthing ring and are even more reduced at the outer one, protecting the personnel from
step voltages. Furthermore, when the ionization effect is taken into account, according
to Figs. 11, 12, 13, 14, 15 and 16, the reduction of the maximum values of voltages and
currents is noticed.
On the other hand, the oscillations are reduced in a much slower rate in contrast to
the corresponding results without considering ionization, where after 0.20 ms they are
dampened notably. At the tower, the ionization effect does not affect the results, as the
range of all the measured parameters is approximately the same. However, the oscil-
lations are dampened without regarding the ionization effect in this occasion as well. As
 Evaluation of Wind Turbine Earthing System 327

regards to the currents, at the bottom of the tower, high current values appear at the
tower base and are lowered at the outer ring. Taking into account the ionization, these
values are even reduced relatively to each peak value respectively.

4 Conclusions

The current study is investigating the performance of the earthing system of an existing
wind farm implemented in Crete, Greece. Since all wind turbines consist of the same
earthing system, one of these simulated in order the earthing system performance to be
investigated. Simulation results have shown that for low peak current, i.e. 5 kA, the
earthing system is reliable in dissipating the lighting currents and emerging voltages.
In case of 150 kA, much higher voltages and currents are resulted, when ionization
is ignored in every node and branch of interest. As for the results obtained by con-
sidering the ionization effect, the corresponding voltages and currents are reduced,
apart from the node A referring to the tower, where the amplitudes are almost the same.
In terms of the oscillation dampening, when the ionization is considered, it is delayed.
The simulation results confirm that away from the tower base, at the outer earthing
ring, the voltages and currents are extremely low. The interconnection of the wind
turbine earthing systems contributes on having an improved performance while around
the tower base extra measures can be taken in order to mitigate the risk for the
equipment and personnel.

References
1. IEC 61400-24, Wind turbines Part 24: Lightning protection, edition 1.0 (2010)
2. ANSI/IEEE Std 80-2000, IEEE Guide for Safety in AC substation grounding (2000)
3. IEC62305-3, Protection against lightning Part 3: Physical damage to structures and life
hazard (2006)
4. https://www.flickr.com/photos/meteorio/45254236794/in/dateposted-public/. Accessed 24
Feb 2019
5. ANSI/IEEE Std 81.2-1983, IEEE Guide for measuring earth resistivity, ground impedance
and earth surface potentials of a ground system (1983)
6. Zeng, R., Gong, X., He, J., Zhang, B., Gao, Y.: Lightning impulse performances of
grounding grids for substations considering soil ionization. IEEE Trans. Power Deliv. 23(2),
667–675 (2008)
7. Zhang, T., Zhang, Y., Tan, X.: Study on overvoltage of signal line in wind turbine by
lightning strike, Shanghai (2014)
8. Yanqing, G., Rong, Z., Jinliang, H., Xidong, L.: Loss transmission-line model of grounding
electrodes considering soil ionization of lightning impulse. In: 2002 3rd International
Symposium on Electromagnetic Compatibility, 21–24 May 2002, pp. 294–298 (2002)
9. Mousa, M.A.: The soil ionization gradient associated with discharge of high currents into
concentrated electrodes. IEEE Trans. PWRD 9(3), 1669–1676 (1994)
10. Rakov, V.A.: Lightning discharge and fundamentals of lightning protection. J. Light. Res. 4
(Suppl. 1: M2), 3–11 (2012)
328 S. A. Pastromas et al.

11. Naxakis, I., Mihos, G., Pastromas, S., Pyrgioti, E.: Examining the operation of the grounding
system of a PV installation. In: Proceedings of the 2018 IEEE International Conference on
High Voltage Engineering and Application (ICHVE), Athens, Greece, 10–13 September
2018 (2018)
12. Pastromas, S.A., Maimaris, K., Stasinos, I.K., Naxakis, I.A., Pyrgioti, E.C.: Assessment of
wind turbine grounding system. In: Proceedings of the 2018 IEEE International Conference
on High Voltage Engineering and Ap-plication (ICHVE), Athens, Greece, 10–13 September
2018 (2018)
Live-Line Working
 Electromagnetic Interference Mitigation
in a High Voltage Inspection Robot

S. B. Barnett(&), A. G. Swanson, T. Lorimer, and M. Brown

University of KwaZulu-Natal, 238 Mazisi Kunene Road,

Durban 4041, South Africa
barnetts@ukzn.ac.za

Abstract. A challenge in developing any electronic system to be operated near

high voltage power lines, is achieving reliability in environments associated with
electromagnetic interference. In this paper, success was achieved in the charac-
terization and mitigation of the effects of the interference that an inspection robot
would experience during proximity dielectric breakdown, electromagnetic cou-
pling and electric arcing to a robot chassis. The source (electric arc) tested was
found to couple to communication lines with a magnitude disruptive to Universal
Serial Bus communications. Shielding was adequate to mitigate this interference,
but when arcing occurred to the chassis itself, it needed to be controlled to
maintain the integrity of the shield.

Keywords: Inspection
Robot
Shielding

1 Introduction

The inspection of high voltage power lines is an undertaking associated with a large
monetary expense and physical risk to the workers undertaking the task. The current
methods of inspection also have limitations with regard to accuracy. Often helicopters, or
ground personnel with optical equipment are employed to visually inspect the power lines
from a distance. It follows that the ability to traverse the conductor itself safely would
provide great advantage in the detailed inspection of the line; while negating limitations
imposed by the terrain, or the cost of commissioning aircraft. However, doing so safely is
inherently complicated if attempted directly by personnel. The field of robotics therefore
lends itself to this application, with the ability to displace personnel from danger; while
still providing some of the advantages of being directly on the line. But power lines are
known to emit high levels of electromagnetic interference (EMI), which hinders the
operation of electronics within their vicinity. This interference occurs with characteristics
relating to the voltages and currents associated with the line, as well the lines topology,
structural abnormalities and the power frequency associated with the line.

2 Considerations for Compatibility

For EMI to exist, there must be a source of emission, a coupling path and a system to
which interfering signals are coupled [1]. Research into the methods of EMI mitigation
highlights frequency as a pertinent consideration [2, 3]. It is shown that the magnitude
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 331–341, 2020.
https://doi.org/10.1007/978-3-030-31676-1_32
332 S. B. Barnett et al.

of interfering signals within a victim device is directly proportional to frequency [3],

and that the characteristics of both the signal and the coupled system impacts the extent
to which interference occurs.

2.1 EMI Bandwidth

The nature of the electromagnetic (EM) emissions produced at high voltage depends on
the occurrence and subsequently the nature of discharge [4]. It is suggested that the
frequency of the EMI that results from breakdown rarely exceeds 1.5 GHz [5, 6].
Further, that corona discharges produce emissions below 3 MHz, and that the emis-
sions from gap type discharges range in frequency up to 1 GHz [4]. A correlation is
also identified between the significance of interference, the magnitude of the source
voltage and weather conditions; with some power lines above 110 kV producing noise
from gap-type discharges up to 2.5 GHz in dry conditions [7], with higher altitudes
being said to increase frequency [8].
It is seen that literature cannot agree upon precise boundaries for the frequency of
emissions [4, 6]. This could be explained by the stochastic nature of dielectric
breakdown [9] and the factors of breakdown that affect the frequency of emissions;
such as the sustainability of an arc, the material of the electrodes, and the nature (or
frequency) of arc re-ignition [10]. It has also been concluded that when measuring
interference due to gas discharge, the dielectric strength of the medium affects the
frequency of the waveform, with higher frequencies correlating with higher dielectric
strengths [11]. It also affects the rise time of partial discharge (PD) pulses, and sub-
sequently the frequency range within which they interfere [13]. EMI is also seen to be
subject to the type of defect causing the PD [4, 6]. However, the waveform produced
by breakdown is like that of an impulse [9]. It follows therefore, that when considering
the frequency of the EM emissions that result from breakdown, that both the frequency
with which the waveform occurs, and the spectrum of the waveform itself must be
considered.

2.2 Cabling and Screening

Dependent on the frequency, magnitude and coupling of EMI, literature suggests
different methods of mitigation [2]. The use of twisted pair cables for communication is
suggested as a good means to mitigate the effects of induced voltages [2, 13]. However,
results have shown [13] limited effectiveness to frequencies within the megahertz
range, and implementation is suggested for interference that does not exceed 2 MHz;
with the number of turns per meter influencing the frequencies for which mitigation is
effective. For higher frequencies, coaxial cables are considered more effective [1].
However, weight, cost, flexibility and effectiveness are largely dependent on the type of
screening used to create the barrier against EM waves.
Enclosing a system with a solid metallic shielding structure is shown as an effective
method for the mitigation of EMI [2]. However, in the context of a power line
inspection robot, weight is a concern. An aluminum body offers inherent shielding
properties, but apertures, joints and conductors traversing within the enclosure can
reduce the shielding effectiveness [2, 13]. Areas of high resistance are also responsible
 Electromagnetic Interference Mitigation 333

for a large loss in shielding [14]. In practical applications it is suggested that this can
result from the use of welding joints [14] or untreated metals; which can form a
dielectric oxidation layer [13].

2.3 Interference Model

For the application of a power line inspection robot, it is assumed that the robot would
not be directly connected to earth, but would rather electrically float near, or at the
potential of the line; whereby the lowest point of the robot is at the lowest potential
imposed by its location in the electric field that results from the source. As such,
capacitive coupling would be the most pertinent consideration. In Fig. 1a the effects of
capacitive coupling can be seen for a cable partially enclosed by a shielding structure.
Figure 1b in contrast shows the effects of direct arcing to a shielding structure.

Fig. 1. Capacitive coupling to partially shielded line (a) and arcing to complete shield (b).

The noise voltage VN can be represented mathematically as shown in Eq. 1.

However, it is noted that the parallel output node set up by the resistance and capacitor
combination has a weighted influence on the noise voltage. The lower impedance of the
two will draw the most current and be subsequently more influential in defining the
voltage drop. This allows for the simplification of Eq. 1 into Eq. 2. It follows that the
developed potential is dominated by the exposed cable and its’ resistance to ground.


C12
jx C12 þ C2G þ C2S
VN ¼ V1 ð1Þ
jx þ RðC12 þ C12G þ C2S Þ

VN ¼ jxRC12 V1 ð2Þ

At higher voltages however, breakdown in the gas between the line and the robot
body may cause the ignition of an electrical arc. This results from the electric field
developed in the capacitor formed between the robot chassis and power line exceeding
that which the dielectric can withstand. To predict the noise voltage seen at the inner
conductor, the representation seen in Fig. 1b is used. From Fig. 1b, Eq. 3 can be
developed.
334 S. B. Barnett et al.

0 0
ZSG
Z2G
VN ¼ 0 0 0
V1 ð3Þ
Z1S þ ZSG
Z2G þ Z2G

If it is assumed that dielectric breakdown causes the impedance between the source
and shield to tend towards zero, then in a similar fashion to Fig. 1a, it is seen that the
noise voltage largely depends on the impedance of the shield to the ground plane.
Further, that a reduction in the impedance between the line and the shield tends the
noise voltage towards the source voltage. A decrease in shield impedance would
therefore lead to a decrease in the resultant interference voltage.

2.4 Shield Design

The effectiveness of a shield in a certain environment can be quantified through
defining its ability to attenuate an EM wave. It is defined as the ratio of the power of the
transmitted wave, as compared to the power of the incident wave. To a large extent,
attenuation is a combination of the losses due to the absorption and reflection of the
incident wave. Because it has been assumed that the source will be of a high voltage,
and low current (due to interference occurring from dielectric breakdown), shielding
effectiveness (SE) can be defined by Eq. 4 [2].
 
Et
SE ¼ 20 log ð4Þ
Ei

t
A ¼ 20
qffiffiffiffiffiffiffi
logðeÞ ð5Þ
2
xlr

Where:
T = Thickness (m)
x = Angular frequency (rad/s)
r = Conductivity (S)
µ = Permeability
A difference in characteristic impedance between two media will result in a
reflection loss for an EM wave attempting to traverse from one media to another. This
loss results from the wave being partially reflected and partially transmitted. The
electric field intensity of a wave traversing a boundary can therefore be defined by
Eq. 6.

2Z2
E1 ¼ E0 ð6Þ
Z1 þ Z2

where:
E1 = Electric field in second medium (V/m)
E0 = Electric field in first medium (V/m)
 Electromagnetic Interference Mitigation 335

Z1 = Characteristic impedance of first medium (Ω)

Z2 = Characteristic impedance of second medium (Ω)
Using Eq. 7 for characteristic impedance and applying Eq. 6 twice (to account for
the two boundaries of the shielding medium), the ratio of E1 and E0 can be substituted
into Eq. 4 to calculate the expected reflection loss.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jxl
Z0 ¼ ð7Þ
r þ jxe

where: e = Permittivity
In the theoretical investigation of the EMI resulting from high voltage power lines,
it was expected that the magnitude of interference should not exceed a rise of 53 dBm
[1, 7]. Using this result, and assuming that the combination of absorption loss and
reflection loss adequately describes shielding effectiveness, Table 1 can be used to
describe the necessary shield thickness for different metals.

Table 1. Required shield thickness.

Material Skin depth [µm] A [dB] R [dB] Thickness [µm]
Copper 65.97 – 108.17 –
Aluminium 84.47 – 106.03 –
Steel (430) 20.86 – 68.17 –
Carbon 1866.3 10.05 49.95 2200
Graphite 1409.3 5.17 54.83 838

2.5 Internal Communication Protocol

Whether EM waves interfere with communication, depends on the communication
protocol used. In this research the universal serial bus (USB) protocol was chosen due
to its’ prominent use in electronics; and in some devices used by current inspection
robots.
Each iteration of this protocol increases the available rates of data transfer. USB 1.0
supports low and full-speed data transfer rates of 1.5 Mb/s and 12 Mb/s respectively
[15]. USB 2.0 supports low, full and high-speed rates of 480 Mb/s. USB 3.0 supports
previous modes and SuperSpeed, with data rates of 5 Gb/s [16].
It is seen therefore, that knowing the iteration of USB used within a system is
important in determining the frequency of the data carrier waveform, and subsequently
the EMI that is of concern. From the expected emission frequencies, and the carrier
wave frequencies of the different USB iterations, Fig. 2a shows that low to high speed
modes are at greatest risk for EMI.
How tolerant a protocol is to induced noise is also important. Inherently, the use of
differential signaling reduces susceptibility. But USB devices will only tolerate certain
voltage ranges to represent logic levels, these fall in the millivolt range [16].
336 S. B. Barnett et al.

Fig. 2. Interference compared to USB mode (a) and cross section of a USB 3.0 cable (b).

In hardware, SuperSpeed is supported over two shielded twisted pairs in a USB 3.0
cable, one for forward traffic, and one for backward traffic. Slower modes utilize a
separate unshielded twisted pair for forward and backward traffic [16]. These are shown
in Fig. 2b.

3 Experimentation

An experiment was devised to quantify the effects of cable shielding, a shielded

enclosure and the twisting of cables operated near high voltage electric arcing. The test
was performed in the high voltage laboratory located at the University of KwaZulu-
Natal. This is a controlled environment and tests were performed near standard
conditions.
A challenge in performing such an investigation was ensuring that measurements
were taken without endangering those undertaking the experiment, nor the equipment
used. Initial testing was done with a Rohde & Schwarz FSH8 spectrum analyzer from
3 m, with a spark gap of 4.5 cm and a source voltage of 100 kV. The results of this
initial experimentation are seen in Fig. 3. However, more direct testing could not be
achieved using this method.
A system was therefore designed (seen in Fig. 4a) to electrically floated mea-
surement equipment near the potential of the source of interference, and to commu-
nicate measured data wirelessly a safe distance. This allowed the effects of arcing to a
robot (seen in Fig. 4b) to be safely investigated.
To maintain the relevance of the acquired data, the chassis of an existing high
voltage power line inspection robot was modified to enclose the equipment. This meant
that similar grade, and similarly aged metal (aluminum) was used, as well as similarly
dimensioned piping through which cables traverse.
 Electromagnetic Interference Mitigation 337

Fig. 3. Interference when arcing to a sphere.

Fig. 4. Shielded test system (a) and example of arcing to robot (b).

3.1 Direct Exposure to Interference

In this experiment, two cable types were laid at a distance of 1.5 m parallel to a spark
that resulted from a 100 kV source. The arc was set to bridge a 4.5 cm air gap to a
brass sphere. The cables tested were an unshielded and untwisted cable of 1 m length,
and a similar shielded and twisted cable referred to as a USB cable. A diagrammatic
representation of this is seen in Fig. 5.
338 S. B. Barnett et al.

Fig. 5. Radiated interference test.

Fig. 6. Time domain analysis of exposed cables at 196 µs (a) and 1.953 µs (b) resolution.

3.2 Cables Inside Shielded Enclosure

In this experiment, two cable types were placed inside the shielded enclosure shown in
Fig. 4a. The enclosure was then placed a distance of 1.5 m from the arc, with the cables
either parallel or perpendicular. This experiment was similar to that which is seen in
Fig. 5, however cables were fully enclosed within the shielded enclosure.

3.3 Arcing to the Enclosure

The untwisted and unshielded cable, as well as the shielded and twisted (USB) cable
were placed inside the shielding enclosure shown in Fig. 4a. The enclosure was then
placed 4.5 cm from the high voltage source. The voltage was then raised such that an
arc between the source and the enclosure formed. The voltage was raised incrementally
until the internal communications of the system failed, which occurred sporadically
between 70 kV and 100 kV. The signals present on the test cables were then noted.
 Electromagnetic Interference Mitigation 339

Fig. 7. Time domain analysis of enclosed cables at 196 µs resolution parallel to spark.

Fig. 8. Time domain analysis with arcing to enclosure with a 196 µs (a) and 1.953 ns
(b) resolution.

4 Discussion

From the results obtained through directly exposing the two cable types to EMI, it was
seen (in Fig. 6a and b) that the twisted and shielded USB cable significantly reduced
the interference that was coupled onto the test cable. When placed in the shielding
enclosure, it was seen that a negligible voltage was induced (Fig. 7) when arcing near
the enclosure, this implies that the shield was effective to radiated emissions. Changing
the orientation, the distance to the source from 1 m to 0.5 m, or the source voltage
magnitude from 100 kV to 60 kV, did not change the signals seen on the lines when
inside the enclosure. This result implies that the signal seen on the line in this test was
not in relation to the high voltage interference source near the enclosure, but rather was
inherent to the environment; which includes ambient EMI, and noise induced by the
measurement systems internal electronics and communication systems.
340 S. B. Barnett et al.

When arcing occurred to the enclosure (Fig. 8a), it was seen that the low frequency
interference, such as the 50 Hz power frequency, did not couple to the measurement
equipment. However, in Fig. 8b, waveforms similar to that of an impulse were seen to
be present. The difference in magnitude between the two cables tested was not as vast
as with experiments where arcing occurred near the exposed cables. This implies that
additional shielding within the enclosure did not significantly improve the overall
shielding effectiveness when the interference source was applied to the enclosure itself.
This shows that despite the applied shielding, the ground impedance was still seen to
develop a potential when exposed to impulse type waveforms.

5 Conclusion

It was shown in Eq. 2 that the EMI levels in a shielded system depends almost entirely
on the exposure of cables. When arcing occurs to the shield itself, it is the shields
ground impedance that builds a potential. It follows that the minimization of this would
reduce EMI susceptibility.
Forcing communication to occur at a frequency much higher than that of EMI (seen
in Fig. 2a) would also reduce its’ effects. However, so would opting for a more robust
protocol, or the use of optical communication.
It was found that a shielded enclosure reduced susceptibility to radiated EMI sig-
nificantly. However, when arcing occurred to the enclosure, a destructive potential still
developed. It was seen in Table 1 that the necessary shield thickness is relatively thin
and as such, the mechanical properties are most important. Electrically, the ground
impedance is most pertinent; as the impulse type waveform was seen to easily couple
and develop a potential. Apart from lowering this impedance, a multilayer shield could
be developed to direct the charge to ground, while still maintaining the integrity and
functionality of the shielding enclosure. This would prevent it from acting like a
conductor and developing a potential that is perceived on communications lines
electrically far from their ground reference.

Acknowledgements. The authors would like to thank Eskom TESP, THRIP and the NRF for
funding this research.

References
1. Weston, D.: Electromagnetic Compatibility Principles and Applications, pp. 2–5. Marcel
Dekker, New York (2001)
2. Ott, H.: Electromagnetic Compatibility Engineering, pp. 238–260. Wiley, New Jersey (2009)
3. Paul, C., Whites, K., Nasar, S.: Introduction to Electromagnetic Fields, pp. 511–577.
McGraw-Hill, Boston (1997)
4. Maruvada, S.: Corona in Transmission Systems, pp. 189–197. Crown Publications,
Johannesburg (2011)
5. Raja, K., Devaux, F., Lelaidier, S.: Recognition of discharge sources using UHF PD
signatures. IEEE Electr. Insul. Mag. 18(5), 8–14 (2002)
 Electromagnetic Interference Mitigation 341

6. Giussani, R., Cotton, I., Sloan, R.: Detection of corona with RF methods and spectra
analysis. In: IEEE International Symposium on Electrical Insulation, San Juan (2012)
7. Pakala, W.E., Chartier, V.: Radio noise measurements on overhead power lines from 2.4 to
800 kV. IEEE Trans. Power Appar. Syst. 90(3), 1155–1165 (1971)
8. Silva, J.M., Olsen, R.G.: Use of global positioning system (GPS) receivers under power-line
conductors. IEEE Trans. Power Deliv. 17(4), 938–940 (2002)
9. Kuffel, J., Kuffel, P.: High Voltage Engineering Fundamentals, pp. 287–364. Elsevier,
Amsterdam (2000)
10. Garzon, R.: High Voltage Circuit Breakers, pp. 1–26. Marcel Dekker, New York (2002)
11. Moore, P.J., Portuguese, I.E., Glover, I.A.: Radiometric location of partial discharge sources
on energized high-voltage plant. IEEE Trans. Power Deliv. 20, 2264–2272 (2005)
12. Kasten, D., Liu, X., Sebo, S., Grosjean, D.: Partial discharge measurements in air and argon
at low pressures with and without a dielectric barrier. IEEE Trans. Dielectr. Electr. Insul. 12
(5), 362–373 (2005)
13. Williams, T.: EMC for Systems and Installations. Newnes, Oxford (2000)
14. Chatterton, P., Houlden, M.: EMC Electromagnetic Theory to Practical Design, pp. 145–
149. Wiley, New York (1992)
15. Anderson, D., Dzatko, D.: Universal Serial Bus System Architecture, pp. 13–17. Addison-
Wesley Developers Press, Boston (2001)
16. Axelson, J., Complete, U.S.B.: The Developer’s Guide. Lakeview Research LLC, Chinook
(2009)
 HCLS Special Hot Yoke for Live
Line Maintenance

Kristianto1, Kurniawan Danu Diharja1,2(&), and Iwan Safarudin1

1
PT PLN (Persero), P3B Sumatera, Jakarta, Indonesia
{kristianto83,kurniawan.danu}@pln.co.id
2
STEI, Bandung Institute of Technology, Bandung, Indonesia

Abstract. At present, the demand for electricity on the island of Sumatra is

proliferating. This need encourages electric power providers to make efforts to
increase the capacity of electricity distribution. One of the most efficient ways to
improve the distribution power is by reconducting existing transmission lines.
Reconducting did by replacing ACSR conductors as HCLS conductors. This
High Capacity Low Sag (HCLS) conductor technology has different mainte-
nance treatments compared to ACSR, especially in tension isolator replacement
work. Initially, we had to carry out maintenance with an off-line condition, but
after we made these new, tested equipment, we were able to do the work without
causing the transmission line to off-line again. This paper will explain how
we do it.

Keywords: HCLS
Hot yoke
Live line maintenance

1 Introduction

The transmission line is the main link that connects power plants as a source of elec-
tricity to consumers who use electricity. This link can connect distances from tens to
hundreds of kilometers. Almost all transmission lines owned by the Perusahaan Listrik
Negara (PLN) are overhead transmission line using conductors of Aluminum Conductor
Steel Reinforced (ACSR) as primary conductors to distribute the electricity. ACSR
conductors that widely used for 150 kV conductors in Sumatra are ACSR 240 mm2.
The growing demand for electricity in Sumatra has made it necessary to increase
the capacity of transmission lines. One of the most efficient ways to increase the size of
electricity distribution is by reconducting. Reconducting has done by replacing the
ACSR conductor installed on the transmission line with the Aluminum Conductor
Composite Core (ACCC). ACCC conductors as a type of HCLS conductor that has a
higher capacity, lower losses and the ability to work at higher temperatures compared
to ACSR conductors of the same size are chosen [1–5]. Until September 2017, 30
transmission lines in P3B Sumatra have used HCLS conductors, and there is a pos-
sibility that the number of installed conductors will increase again. Size of HCLS
conductor that commonly used is ACCC Lisbon 310 mm2.
In the installation and maintenance of HCLS conductors have their difficulties.
Because the core conditions that cannot experience bending are higher than 300, this
conductor must be installed and maintained especially [2]. If we do not pay attention to

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 342–352, 2020.
https://doi.org/10.1007/978-3-030-31676-1_33
 HCLS Special Hot Yoke for Live Line Maintenance 343

good and correct installation and maintenance techniques, the conductor has great
potential to break up.
Transmission lines that use HCLS conductors are not transmission lines that are
free of maintenance. Some cases of insulator damage have occurred in this kind of
transmission line. The insulator types of HCLS conductors that experience damage
varies by the kind of suspension or tension. The isolator replacement work with live
line maintenance conditions previously only done on the kind of suspension insulator.
Live line work for a kind of tension insulator cannot be done because the pulling
equipment of the available conductor can only be used for ACSR conductor types. So
that for the replacement of tension-type insulators can only be done with off-line
maintenance, and this results in the unreliability of electricity distribution to consumers.

2 General Definition

2.1 HCLS Conductor

High Capacity Low Sag (HCLS) is a type of high-temperature low sag (HTLS) con-
ductor. A conductor technology that can deliver more electrical power, smaller power
losses, with resistance to work at high temperatures. In Indonesia there are two types of
HCLS conductors, the first is Aluminum Conductor Composite Core (ACCC) and
Aluminum Conductor Fiber Reinforced (ACFR). Both types of conductors use com-
posites as the core conductors.
The ACCC conductor started to develop by CTC Global in 2002 and commercially
deployed in 2005. The new conductor achieved its high capacity objective while
offering a substantial reduction in thermal sag compared to any other commercially
available conductor, due to its very low coefficient of thermal expansion. Due to the
composite core’s decreased weight compared to steel, an ACCC Drake size conductor,
for instance, could incorporate 28% more aluminum using compact trapezoidal shaped
strands, with a slight overall reduction in weight. The added aluminum content and
improved conductivity allow them to operate more efficiently compared to any other
available conductor of the same diameter and comparable weight. The ACCC con-
ductor achieves the highest ampacity at the coolest operating temperature compared to
the other high-temperature capable conductors [1].

ACSR ACCC

Fig. 1. The physical form of (from left to right) ACSR, ACCC, and ACFR.
344 Kristianto et al.

A key component of the ACFR is the unique stranded carbon fiber reinforced
plastic (see Fig. 1). Compared to ACSR, ACFR has about one-fifth the weight, one-
tenth the thermal expansion coefficient, and a higher tensile strength. It does not suffer
from magnetic losses. It is highly flexible so that it can be wound around a small drum
and is highly resistant to corrosion, unlike iron and steel. Last, its expected creeping
deformation level at high temperatures is similar to that of iron. The ACFR has lower
electrical resistance than the ACSR, since the trapezoidal shape has a larger contact
area than the round shape. The lightweight CFCC core allows an increased aluminum
content, at constant conductor weight. The transmission loss expected from this design
is 27% less than that of similarly sized ACSR because of the increased content of
aluminum, zero magnetic loss, and high electrical conductivity resulting from the use
of trapezoidal-shaped annealed aluminum. Since the CFCC core has a small thermal
expansion coefficient at temperatures as high as 175 °C, its thermal expansion is less
than that of ACSR. The transmission capacity expected from this design is more than
double that of similarly sized ACSR [6]. This type of conductor installed in one section
of 150 kV transmission line in Sumatera.
How to Thread HCLS. Several main points should be carefully understood before
using the HCLS conductor. The aluminum of HCLS is softer than conventional con-
ductor because it conducts electricity better, and all the conductor’s rated tensile
strength designed into the HCLS composite core after the knee point achieved, not
relying on the aluminum for rated tensile strength). All handling and tools which made
contact with the conductor must avoid gouging or scratching the aluminum surface,
which can create corona effects which can cause damage and failures at voltages over
138 kV.
The individual aluminum strands are trapezoid shaped and fit tightly together to
form each layer of strands. The strands can slide axially concerning one another, but
not as easily as conventional round strands. This characteristic makes the conductor
more susceptible to the condition known as the birdcage. While loosening or birdcage
conditions themselves are not a defect or failure, they can and should be avoided and
corrected.
The composite carbon core of HCLS is much stronger than conventional steel core
and is quite flexible, but not as flexible as equivalent steel core. Bending beyond a
minimum bending diameter will fracture the core instead of yielding like conventional
steel core. This one difference accounts for almost all installation issues and drives
nearly all of the elements of the installation and maintenance guidelines [2].

2.2 Live Line Work Conditions

Live line work in Indonesia must comply with five main aspects. Including available
rules and procedures/work instructions, available personnel who are competent to do
the type of work, available work and safety equipment that meets work requirements,
repaired installations have a protection system for live work, the latest one is weather
and supporting environmental conditions to implement the live work. If one of the five
conditions does not comply, the live work cannot be done. In developing new work
 HCLS Special Hot Yoke for Live Line Maintenance 345

Fig. 2. Offline maintenance work on HCLS conductors.

equipment, complete work instructions, personnel capabilities must be considered and

certainly ready before new equipment can be used.
In Indonesia, several kinds of methods can be used to do a 150 kV tension isolator
work. The type of method used depends on the type of construction to done. The most
common method used is using the strain pole-chain hoist method. This method uses a
series of isolator replacement conductor equipment such as Fig. 2. In ACSR conductors
that are widely used, come along is used to bite the conductor and pull it with a chain
hoist. Come along in ACSR conductor attached with a distance of 50–70 cm from the
tension press clamp. In HCLS conductor, you must attach a certain type of come-along
(has a minimum jaw length of 8–11 in.) and you must attach it 4–5 m from tension
press clamp (Fig. 3).

Fig. 3. Live line maintenance work on the tension-type tower of ACSR conductor.
346 Kristianto et al.

3 Problem Solving
3.1 Discussion with Cable Manufacturer
The first process that we did before making a special hot yoke was to invite HCLS
conductor manufacturers in Indonesia to discuss and to have better understand the
character details of the parts of the HCLS as well as to be able to plan design of
auxiliary equipment by HCLS conditions.
The live line maintenance team also conveyed the types of work methods possibly
do and what types of equipment were used. From this meeting, we conclude that for
insulator replacement with HCLS conductors:
• For insulator replacements in suspension type towers, work can still be done with
existing equipment and procedures
• The withdrawal must be carried out using a wire-grip with a minimum length of
8 in. and 11 in. because it avoids slippage and keeps the compressive power evenly
distributed on aluminum parts.
• Installation of wire grip at ACCC is done 4–5 m from the tip of the tension press
clamp to avoid bird caging or damage to the aluminum layer.
• There is an option from the PDKB procedure that can be used to replace tension
isolators using a pole-yoke set strain, but a suitable hot yoke is still needed to do this.

3.2 Creating Design and Manufacturing Process

After obtaining conclusions from the previous meeting, we proceed with the design of
pull equipment that can be used according to the requirements of the cable manufac-
turer. This design underwent several attempts and improvements so that the last design
appeared.
This design consists of three main parts, namely the Adapter, Sleeve, and Puller
section. The adapter functions to stabilize the yoke movement and maintain the dis-
tance of the yoke with the tension press clamp. The sleeve serves to clamp and press
the aluminum part of the conductor. The puller is the part that will press the sleeve and
hold the pull from the Puller that connected to the trunnion and strain pole (Fig. 4).

Fig. 4. Special hot yoke design.

 HCLS Special Hot Yoke for Live Line Maintenance 347

The material we have used is Durall Aluminum cast which has been certified, and
the manufacturing system that has done is machining and cold forging system. The
choice of this material considers the mechanical strength that must be held and also the
weight of the material. The estimated safe mechanical design load is 50 kN. The
dimensions of our design match the width of the existing ACSR yoke. We designed a
yoke to be able to fit with the trunion/series of pulling devices (Fig. 5).

Fig. 5. Discussion of yoke design with yoke manufacturer.

In October 2016, the special hot yoke was started to build. Initially, this aluminum
material was cast according to size requirements (see Fig. 6). Then the material is
reshaped with a machining and cold forging process until we get the desired shape. The
end of the process we got this special yoke (see Fig. 7).

Fig. 6. Aluminum cast material for the hot yoke

348 Kristianto et al.

Fig. 7. The shape of the yoke after reshaped and cold forging process

3.3 Testing Process

The special hot yoke is then tested with tensile strength testing to determine the tensile
resistance of special hot yoke before it can be used in the actual installation. The types
of tensile strength tests performed are breakdown test and working load test.
Breakdown test is done by pulling the yoke until the yoke breakdown. The test
value is expected to be more than the design breakdown capability of 200 kN. Sche-
matic testing is described in Fig. 8.

Fig. 8. Breakdown test schematic

The result of the breakdown test is the special yoke has a breakdown on the load of
205.3 kN (Fig. 9).
 HCLS Special Hot Yoke for Live Line Maintenance 349

Fig. 9. Breakdown test result. Before (top image) and after the test (bottom image).

Working load test is applied with a tensile load of 30 kN and 50 kN. From this test
will be seen the ability of special yoke to survive for 60 min and the value of the shift
that occurs in the conductor. A schematic working load test is described in Fig. 10.

Fig. 10. Breakdown test schematic

The result of the working load test is that the yoke can withstand in the 60-minute of
30 kN tensile test with a 4 mm shift (see the top image in Fig. 11) and the 60 min of
350 Kristianto et al.

50 kN tensile test for with a 10 mm shift (see the bottom image in Fig. 11). With this shift
value, there will be no problems arising in the HCLS aluminum section.
With the results of this test, we are sure to be able to use this equipment in the real
transmission line

Fig. 11. Shifting of the special yoke on the conductor with a working load test.

3.4 Creating Working Procedure Process

Before it can be used in the actual installation, we must first make the work instructions
for implementing the isolator replacement using this yoke. Because without work
instruction that has been approved, live line work cannot be done. This is to ensure that
each step of the work has been tested and is safe for each person who carries out the
work and equipment for the installation that is maintained.
The draft work instructions were prepared by the live line team of Transmission
Services Unit Medan and were reviewed together with several live line teams in P3B
Sumatra on April 18, 2017. Then on April 19, 2017, an offline trial was carried out at
one of tension-type tower 150 kV transmission line using HCLS conductor, and on the
next day, a trial was carried out again with a live line condition at the same location.
The results of both trials that have been done are well with safe, and the work
arrangement of the steps was agreed upon (Fig. 12).
 HCLS Special Hot Yoke for Live Line Maintenance 351

Fig. 12. Hot end yoke trial work with live line method on the HCLS conductor tension-type
insulator.

After it was decided that this draft work instruction feasible and fulfilling the rules
of safety for the personnel, equipment, systems, and environment, the draft work
instructions can be ratified or formalized by the P3B Sumatera live line committee and
national live line committee. After this work instruction has been approved by the live
line committee, the live line work has fulfilled the requirements to be able to do.

4 Application and Benefits

This special hot end yoke has advantages over using come along in consideration of:
• Replace the come along function as a conductor clamping device.
• Can be installed close to dead end compression.
• Eliminates bending, bird caging and is safe for conductors.
• Personnel can be comfortable and safe while working.
Financially, the benefits that obtained are the availability of electricity that can be
delivered to consumers because maintenance work is done without de-energized the
transmission line. Also, the reliability of the system conditions maintained by con-
ducting live line work.
The following Table 1 gives a list where this yoke implemented.
352 Kristianto et al.

Table 1. Locations of the special yoke implemented.

Date Line Condition
22-03-2017 Perbaungan – Tebing Tinggi T.27 Offline trial
19-04-2017 Perbaungan – Tebing Tinggi T.27 Offline trial
20-04-2017 Perbaungan – Tebing Tinggi T.27 Live line work
16-05-2017 Perbaungan – Tebing Tinggi T.27 Live line work
07-06-2017 Belambangan Umpu – Baturaja T.332 Live line work
09-08-2017 Kotabumi – Bukit Kemuning T.73 Live line work
10-08-2017 Kotabumi – Bukit Kemuning T.73 Live line work
30-08-2017 Belambangan Umpu – Baturaja T.210 Live line work
08-09-2017 Belambangan Umpu – Bukit Kemuning T.126 Live line work
27-02-2018 Sutami – Natar T.74 Live line work
20-03-2018 Balai Pungut – Duri T.67 Live line work
19-09-2018 Garuda Sakti – Balai Pungut T.169 Live line work
02-01-2019 Rantau Prapat – Padang Sidempuan T.140 (ACFR) Live line work
24-01-2019 Garuda Sakti – Balai Pungut T.169 Live line work

5 Conclusions

The special hot yoke that made for the work of replacing tension-type insulators in
HCLS conductors (ACCC or ACFR) with live line method proven work with safe and
well. This equipment made with a secure process, tested in the laboratory, tried out and
can be implemented in 150 kV Single HCLS conductor tension-type tower. For bundle
or more than one conductor, this method is not applicable.
Subsequent this yoke provides excellent benefits in maintaining the availability of
electric power distribution when transmission line maintenance activities are carried
out. For financially and system reliability, this yoke is very helpful to electric power
companies to be able to maintain their performance.

References
1. CTC Global Corporation: Engineering Transmission Lines with High Capacity Low
Sag ACCC Conductors, 1st edn. CTC Global Corporation, Irvine (2011)
2. CTC Global Corporation: ACCC Conductor Installation Guidelines. CTC Global Corpora-
tion, USA
3. Kumar, M., Rahangdale, R.: Comparative analysis of ACSR and HTLS conductor. Int.
J. Futur. Revolut. Comput. Sci. Commun. Eng. 4(5), 29–35 (2018)
4. Patel, H.B.: Re-conductoring scenario & payback calculations of ACSR moose and its
equivalent conductors for 400 kV transmission line (thermal uprating). Int. J. Adv. Eng. Res.
Dev. 2(5), 1283–1290 (2015)
5. EPRI: Demonstration of Advanced Conductors for Overhead Transmission Lines. Energy
Research and Development Division Final Project Report, CEC-500-2013-030. USA (2008)
6. Ohki, Y.: Development of a low sag carbon fiber reinforced aluminum conductor for
transmission lines [News from Japan]. IEEE Electr. Insul. Mag. 33(6), 54–57 (2017)
 Evaluation of the Performance
of Glass Insulator String with Broken Units
Under Artificial Pollution

Frederico Tassi de Souza Silva(&) and Ricardo Wesley Salles Garcia

CEPEL, Rio de Janeiro, Brazil

frederico@cepel.br

Abstract. Glass insulators are more susceptible to vandalism and the presence
of broken units can modify the supportability of the insulator string. It is nec-
essary to know the influence of broken insulators regarding to the quantity and
position in the string to verify the supportability of the remaining units. The aim
is to evaluate the electrical performance of glass insulator strings with broken
units in AC systems, under pollution conditions, in various configurations. The
artificial pollution method used will be the solid layer according to the recom-
mendations of IEC-60507, considering three levels of pollution for each con-
figuration. The SDD (salt deposit density) values used were: 0.03, 0.07 and
0.10 mg/cm2. In this work, 9 configurations were tested. The test was carried out
under steam fog, applying the non-standardized method named RFO (Rapid
Flashover), which has been used by some laboratories and proposed as an
alternative to the up and down method. The tests identified that the position of
the broken insulators in the string influences the breakdown voltage. The results
show that glass insulator strings with broken units in random and intermediate
positions may represent a more pronounced drop in the supportability when
compared to the configurations in which these units are located at the ends, the
one in which the broken units are close to the conductor presents the least loss of
supportability.

Keywords: Broken glass insulator
Artificial pollution
Rapid flashover
Salt

density deposit
Electrical supportability

1 Introduction

The reliability of electric power transmission systems is an increasingly important

factor for modern society, because the global economy requires energy with high
quality of supply and consumers are more aware of their rights, forcing service pro-
viders to develop mechanisms to minimize disruptions and ensure such characteristics,
suffering heavy fines if pre-set targets are not met. Therefore, being able to prevent
discontinuities in the supply of electricity is paramount in the current context of the
electric energy companies.
Keeping a transmission or distribution line in good working order requires a high
investment in both equipment and labor. The reliability required and the serious
consequences in the event of a disruption in the supply of electricity lead companies to

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 353–361, 2020.
https://doi.org/10.1007/978-3-030-31676-1_34
354 F. T. de Souza Silva and R. W. S. Garcia

seek increasingly efficient methods and equipment in order to avoid interruption as

much as possible and, if these occur, efficiently reduce the time and cost of repair [1].
The electrical insulation failures are the main causes of interruptions in the electric
power supply to the various consumers, so the performance of the power transmission
lines is directly related to the performance of their insulators [2].
There are studies that indicate the presence of residual electrical supportability even
in extreme conditions of all broken glass units in dry environments [3]. However, there
are no references in the literature to studies concerning the performance of insulator
strings with broken units under pollution and humidification conditions.
The goal of this work is to evaluate the electrical performance of glass insulator
strings with broken units for a.c. systems, which are subject to pollution and humid-
ification conditions.

2 Test Preparations

The evaluation was performed considering the number of broken units and their
position in the string, for three different levels of pollution - light, moderate and heavy -
using steam fog as a humidifier element, the reference being the SDD.

2.1 Object Under Test

The tests were performed on glass standard insulator strings (Figs. 1a and b). Their
characteristics are shown in Table 1. The broken units were shown in Fig. 1c. A string
with 10 insulators was considered as reference.

Table 1. Dimensional characteristics of the insulator.

Characteristics (units) Nominal values
Diameter (D), in mm 255
Height (H), in mm 146
Creepage distance, in mm 312
Top and bottom surface area, in cm2 633 and 939

Fig. 1. (a) Insulator profile (b) Intact insulator (c) Broken insulator
 Evaluation of the Performance of Glass Insulator String 355

2.2 Pollution Characteristics

The method used to apply the pollution was the solid layer. The composition of the
contaminant suspension was prepared using, for each liter of distilled water, 40 g of
kaolin and approximately 13.5 g/l of commercial purity salt for the level of light,
32.0 g/l for the moderate and 42 g/l for heavy pollution level.
The degree of pollution reference on the insulator has an established tolerance of
±15% and the kaolin used must have the characteristics of inert material as described
in IEC 60507: 2013 [4].
The application of the layer of pollution was made by dipping the insulator in the
suspension. The subsequent drying of the polluted insulator was performed at room
temperature. The application of the pollution layer was not performed on the insulators
with the broken dielectric. The deposit density of non-soluble material, NSDD, is
0.10 mg/cm2.

2.3 Selected Arrangements

The insulator strings were mounted in the test chamber in a vertical position. The test
was performed considering 9 (nine) distinct arrangements (Fig. 2), varying the number
and position of broken units, according to Table 2.

Fig. 2. Types of string configurations tested

Although the percentage of broken units may seem large, it is perfectly possible to
be found in vandalism situations.
356 F. T. de Souza Silva and R. W. S. Garcia

Table 2. Dimensional characteristics of the insulator.

Config. Nominal values Description and location of broken insulators
Intact Broken
#1 10 0 10 intact insulators
#2 8 0 8 intact insulators
#3 6 0 6 intact insulators
#4 8 2 8 intact and 2 broken in a row, from the conductor
#5 6 4 6 intact and 4 broken in a row, from the conductor
#6 6 4 6 intact and 4 broken in a row, in the middle of string
#7 6 4 6 intact and 4 broken randomly distributed
#8 6 4 6 intact and 4 broken in a row, from the ground
#9 8 2 8 intact and 2 broken in a row, from the ground

3 Test Procedures

The humidification of the insulators under test was done using a steam generator
producing a constant flow during the test of approximately 0.12 l/h/m3. A plastic tent
constructed at CEPEL was used to contain the fog (Fig. 3). The fog flow was not
pointed directly at the insulator string.
Due to the number of arrangements, and consequently the number of voltage
applications required to obtain the 50% probability of breakdown voltage, U50, fol-
lowing the IEC 60507: 2013 [4], would imply in a long period of tests, the method
known as rapid flashover (RFO) was adopted.
This method, initially proposed by Lambeth in 1988 [5], has been used by some
laboratories [6, 7] and was recently recommended by Cigré [8], aiming to reduce the
time and cost of the tests under pollution. However, the lack of specific criteria to
calculate the value of U50, from the obtained results, has limited its application.
In the case of this project, as the main objective is a comparative analysis, this
aspect does not imply compromised information. In the adopted RFO method, an initial
voltage is applied in the insulator string, during a period of 20 min, as soon as the fog
begins to enter the plastic tent.
After this time, the applied value is increased in levels of 10%, and maintained for a
period of 3 min, until the occurrence of flashover. From this situation, the voltage is
decreased or increased at levels of 5% of the initial value, if the result of the previous
application is flashover or withstand, respectively, and kept for periods of 3 min. The
total test duration is 100 min.
The initial voltage is chosen close to the discharge voltage according to the
arrangement.
 Evaluation of the Performance of Glass Insulator String 357

Fig. 3. Configuration for the steam fog test at CEPEL’s pollution chamber

4 Results and Analysis

The results are showed in a “U” curve format, the purpose of which is to obtain a
minimum value of the test flashover voltage caused by the reduction of the resistance of
the pollutant layer to the humidification, a characteristic process of solid layer pollution
tests. To obtain this curve, the mean values of a number of successive discharges are
calculated.
The graphs in Figs. 4, 5 and 6 show, for each of the 9 (nine) arrangements tested,
the behavior of the 7-discharge mean curve for the 3 pollution levels (0.03, 0.07 and
0.10 mg/cm2, respectively).
It is possible to note, as expected, that for each pollution level, arrangements with
all the intact insulators (1, 2 and 3) have discharge voltages directly proportional to the
number of insulators in the string. Moreover, by comparing arrangements with the
same number of insulators, for different levels of pollution, it is verified that the higher
the pollution the lower the observed flashover voltage.
By analyzing separately the groups of arrangements with 8 intact units (configu-
rations 2, 4 and 9) and 6 intact units (configurations 3, 5, 6, 7 and 8), for each of the
levels of pollution tested, it is noticed that there is a difference in voltage values,
358 F. T. de Souza Silva and R. W. S. Garcia

indicating that the position of the broken insulation in the string influences the flash-
over voltage. It is also verified that for the highest level of pollution the trend has
reversed between arrangements 4 and 9 and between 5 and 8, indicating the need to
evaluate later if this trend continues to levels even higher.

240
0.03 mg/cm²
1
220

4
200
9
Breakdown Voltage (kV)

180 2
5
3
160
8

140 6
7

120
Config. 1 Config. 2 Config. 3

100 Config. 4 Config. 5 Config. 6

Config. 7 Config. 8 Config. 9

80
0 20 40 60 80 100
Time (min)

Fig. 4. Discharge curve for nominal pollution with DDS of 0.03 mg/cm2

Table 3 presents the minimum values of the calculated means of 7 consecutive

discharges, related to the actual DDS values of each arrangement, for the different
levels of pollution considered.
The graphs in Fig. 7 show the points of Table 3, with curves for each arrangement,
calculated according to expression (1), which is representative of the flashover voltage
behavior according to the pollution level [9]:

U ¼ A xa ð1Þ

Where:
U = Voltage in Volt (V)
A = Configuration of the string
x = Level of pollution
a = Influence of pollution
 Evaluation of the Performance of Glass Insulator String 359

240
0.07 mg/cm²
220 Config. 1 Config. 2 Config. 3
Config. 4 Config. 5 Config. 6
200 Config. 7 Config. 8 Config. 9
Breakdown Voltage (kV)
1
180
4
160 9

5 2
140
8
120 6
3

100 7

Time (min)
80
0 20 40 60 80 100

Fig. 5. Discharge curve for nominal pollution with DDS of 0.07 mg/cm2

240
0.10 mg/cm²
220 Config. 1 Config. 2 Config. 3
Config. 4 Config. 5 Config. 6
Config. 7 Config. 8 Config. 9
200
Breakdown Voltage (kV)

1
180

160 9

4
140
2

120 8
5
3
100 6
7
80
0 20 40 60 80 100
Time (min)

Fig. 6. Discharge curve for nominal pollution with DDS of 0.10 mg/cm2
360 F. T. de Souza Silva and R. W. S. Garcia

Table 3. Description of arrangements.

2
Config. 0.03 mg/cm 0.07 mg/cm2 0.10 mg/cm2
SDD (mg/cm2) Um (kV) SDD (mg/cm2) Um (kV) SDD (mg/cm2) Um (kV)
#1 0.034 219.4 0.085 177.5 0.101 165.2
#2 0.032 173.9 0.053 143.0 0.098 137.8
#3 0.032 157.3 0.078 121.3 0.098 97.9
#4 0.032 193.4 0.053 169.0 0.101 145.5
#5 0.032 157.9 0.060 133.5 0.098 111.5
#6 0.035 141.6 0.060 131.7 0.094 97.8
#7 0.035 134.9 0.067 110.4 0.094 96.6
#8 0.035 156.0 0.067 131.7 0.089 105.5
#9 0.035 187.9 0.063 158.8 0.101 158.8

Breakdown
voltage Mean values of breakdown voltage
220 (kV)

200
y = 93.649x-0.253
R² = 0.9908
180

y = 42.255x-0.397 y = 104.46x-0.171
160
R² = 0.917 R² = 0.8615
y = 82.735x-0.245
140 y = 42.295x-0.373 R² = 0.9994
R² = 0.8579 -0.304 y = 84.244x-0.201
y = 55.692x
R² = 0.8337
120 R² = 0.9911

Config. 1 Config. 2 Config. 3 y = 43.014x-0.379

100
Config. 4 Config. 5 Config. 6 y = 43.615x-0.339 R² = 0.9242
Config. 7 Config. 8 Config. 9 R² = 0.9962 SDD (mg/cm²)
80
0.00 0.02 0.04 0.06 0.08 0.10 0.12

Fig. 7. Mean values of breakdown voltage for all configuration strings

5 Results and Analysis

In general, from the information presented in Fig. 7, it can be observed that the
exponents of the curves of the six-unit intact arrangements are, in magnitude, much
larger than the value of 0.22, indicated in the literature as characteristic of a.c. systems
[10]. The other results also show some variation, but smaller in dispersion. This
behavior should be analyzed by conducting further tests to see if the method applied
produces some influence or if some other effect not taken into account, such as the
humidification process, may cause the variations presented.
 Evaluation of the Performance of Glass Insulator String 361

Considering the configurations 4 and 9 which have 2 broken insulators in a row,

close to the conductor and the structure respectively, they are found between the curves
of the configurations 1 and 2, which do not have broken units. This shows that breaking
some units reduces the breakdown voltage, but the broken insulation contributes in
some way to the total supportability of the string.
Comparing them to each other, it can be seen that the strings that have the insu-
lators broken in sequence near the ends (arrangements 5 and 8) have higher voltages
than arrangement 3, repeating what happens to the arrangements of two broken insu-
lators. However, the results of arrangements 6 and 7 are below, indicating that these
configurations may represent greater reduction.
The fact that their curves fall below the six-unit arrangement curve (arrangement 3)
suggests the need for further investigation to verify if this behavior is also present in
insulator strings with more units.
Additionally, an assessment of the electrical supportability of a string with all units
broken under humidification will be important to identify how the arc behaves in the
string breakdown process under these conditions.

References
1. Rawat, A., Gorur, R.: Microstructure based evaluation of Field aged and new porcelain
suspension insulators. IEEE Trans. Dielectr. Electr. Insul. 16(1), 107–115 (2009)
2. Pressi, R.A., Resener, M., Ramos, M.: Economic evaluation of preventive inspections of
transmission circuits through the emission of corona. In: Congreso Internacional de
Distribuición Eléctrica, Buenos Aires, Argentina, 516 (2014)
3. George, J.M., Prat, S., Tartier, S., Lodi, Z.: Electrical characteristics and properties of a stub
(damaged toughened glass insulator). In: 18th International Symposium on High Voltage
Engineering, 25–30 August 2013, Seoul, Korea, paper OE1-03 (2013)
4. IEC 60507:2013 - Artificial pollution tests on high-voltage insulators to be used on A.C.
systems (2013)
5. Lambeth, P.J.: Variable-voltage application for insulator pollution tests. IEEE Trans. Power
Deliv. 3(4), 2103–2111 (1988)
6. Gutman, I.: Time-effective rapid test procedures for pollution testing of different types of
insulators. In: INMR Conference (2015)
7. Vlastós, A.E., Feiju, Y., Boubitsas, V.: Clean fog rapid procedure test of artificially and
naturally polluted HVDC porcelain barrel insulators. IEEE Trans. Power Deliv. 6(4), 1791–
1797 (1991)
8. Cigré Technical Brochure nº 691 - Pollution Test of Naturally and Artificially Contaminated
Insulators, July 2017
9. Jiang, X., et al.: Study on pollution flashover performance of short samples of composite
insulators intended for ± 800 kV UHV DC. IEEE Trans. Dielectr. Electr. Insul. 14(5),
1192–1200 (2007)
10. Pigini, A.: Optimizing insulation design for harsh service environments. INMR article, 9
September 2018
 Ten Years’ Experience of Live Work
Implementation in the Slovenian Electrical
Energy Field

A. Lovrenčič1(&) and V. Lovrenčić2

1
Faculty of Electrical Engineering, University of Ljubljana, Tržaška 25,
Ljubljana, Slovenia
Ancka.lovrencic@gmail.com
2
C&G d.o.o. Ljubljana, Riharjeva 38, Ljubljana, Slovenia

Abstract. We have been effectively carrying out live work (LW) in the
Slovenian electrical energy field for the last ten years. In January 2009 the first
permission to an electrical maintenance worker was issued in Krško Nuclear
Power Plant (NEK). We implemented a confirmed system for live work based
on the French method. Systematic manuals were prepared for the electrical
energy (transmission & distribution) and industrial companies. We successfully
trained more than one hundred line and maintenance workers and more than one
hundred coordinators from the electrical energy, industrial and service fields.
Systematic and independent audits of carrying out LW confirm that electricians
consistently use the prescribed and required personal protective equipment
(PPE) and tools for safe work. We have reached a notable level of efficiency and
effectiveness of carrying out LW. Our main concern in Slovenia is safety and
health at work. During last ten years there was not any dangerous situations and
no accidents while carrying out LW. Now in Slovenia, we carry out LW on low
voltage (LV) and we started on medium voltage (MV) but just cleaning of
substation. In Slovenia we still have many challenges ahead of us, mostly to
implement the LW method in additional working milieus and use it with the aim
to improve electrical energy quality indicators and reduce unnecessary discon-
nections during preventive planned maintenance works in utilities and process
industry. Despite more than 100 years’ experience, our neighbouring countries
and other countries worldwide continue with the development of LW methods
on all voltage levels, from LV to HV. One of important aim in Slovenia is to
carry out LW like a method as a method which supports the “zero accidents”
idea. Stakeholders need to promote the safe implementation of LW, so that the
safety of LW will be emphasised as much as the economic impacts.

Keywords: Live working
Live maintenance
Slovenian experience
Safety

at work
Zero accidents

1 Introduction

Slovenian power engineering community had to overcome many obstacles to achieve

the current level of LW development. When talking about ten years’ experience in LW
in the Slovenian electricity sector, one needs to know that this development was
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 362–372, 2020.
https://doi.org/10.1007/978-3-030-31676-1_35
 Ten Years’ Experience of Live Work Implementation 363

dependent on efficient and successful time synchronization of numerous stakeholders,

their expectations, interests and wishes [1].
Many stakeholders and interested parties are involved in the development and
implementation of LW in companies. Internal stakeholders are mainly managers and
employees (top management, heads of maintenance and health and safety departments,
maintenance workers, lineman and many others). External stakeholders are owners,
potential investors, electricity consumers, suppliers of LW tools, equipment and
technology, institutions (institutes, training centres), professional organizations (IEC,
CENELEC, SIST, CIGRE, CIRED, IEEE, LWA), competitive companies in the LW
market and a wider social community (Ministry of Labour, labour inspectors, trade
unions, international associations) [1].
Each of the mentioned stakeholders has its own point of view on LW based on its
specific interests. Stakeholders consider LW in terms of organization and implemen-
tation of maintenance, quality, health and safety at work, efficiency and effectiveness or
economic and social effects, as well as linemen training, upgrading management sys-
tems and ensuring customer satisfaction [1].
We have ten years of successful implementation of LW in Slovenian power
engineering field behind us. The first LW maintenance authorization was issued in
NEK in January 2009.

2 History of LW Development

The historical framework of LW development in Slovenia should be viewed in the

context of global activities, while also taking into account the circumstances of LW
development in Slovenia, including the period before the independence (joint devel-
opment steps in Yugoslavia) and after the independence in 1991.
The length limits of this paper make it impossible to comprehensively describe
Slovenian history of LW. A complete description of this subject is available in the
doctoral dissertation [1]. The truth of the matter is that much can be learned from the
global history of LW development on all voltage levels and used as support in real-life
implementation.

2.1 History of LW Development in the World

LW is a method of maintaining electrical installation that was originally established at
the beginning of the 20th century. The first use of the method was mentioned as early
as in 1913 in the USA. In spite of a hundred year tradition, live working has gained a
new development impetus in the recent period due to the restructuring of the sector of
transmission and distribution of electric power. It could be argued that by the opening
of the electric power market, old monopolies in the electric power transfer and dis-
tribution seemed to have vanished [1–3].
LW has a hundred years tradition in the world. The history of LW development is
represented by many authors [1–6]. Table 1 presents several historical data.
364 A. Lovrenčič and V. Lovrenčić

Table 1. History of LW development

Year Country
1913 The first implementation of LW was documented in USA
1920 Canada
1920–1930 Germany, Sweden, SSSR (Russia), Switzerland
1932 Australia
1933 (1975) Poland
1939 Great Britain
1945 Chile
1952 China
1962 Hungary
1963 France
1971 Argentina
19xx Brazil, Spain, Italy, Hungary, Ireland, Colombia, Peru, Ecuador, Uruguay,
Venezuela, Romania, Czech Republic, Slovakia, Portugal, Norway, Belgium,
New Zealand, India, etc.
20xx Croatia, Slovenia, Turkey

In the 1980s, LW expanded across almost all European countries. Activities also
started in Yugoslavia, in Slovenia and Croatia, the latter has taken upon itself to
construct the Yugoslav LW training centre (nowadays called HEP NOC), located in
Velika, Slavonia, near the city Požega. In 1987, Croatia was also the first Yugoslav
country to adopt a national regulation that allows LW [1].

2.2 History of LW Developments in Slovenia

LW Development in Slovenia 1961–1991
When studying the history of LW development in Slovenia, one has to consider the fact
that until 1991 Slovenia was part of Yugoslavia. It is therefore necessary to present an
overview of historical sources from this period. It should be noted that many records
from this period have been lost due to frequent reorganizations in power engineering
companies in Slovenia and Yugoslavia. Frequent relocations of archives and inac-
cessibility of private collections are a constraint in this study.
In Slovenia, the initiative for the introduction of LW appeared in 1961. The
Slovenian electricity community organized a LW consultation. Representatives of
power engineering companies from all the republics participated and the conclusions of
this consultation are still very topical today [1]:

• it is necessary to organize LW due to the demands of consumers (especially

industrial companies) for an uninterrupted power supply,
• when implementing LW, it is necessary to examine the work methods in order to
achieve the necessary safety,
• LW should first be introduced on LV and then gradually on higher voltages,
 Ten Years’ Experience of Live Work Implementation 365

• the applicable technical regulations do not allow LW and should be adapted to the
new requirements,
• a LW commission and four sub-commissions (for training, preparation of tools,
implementation on LV and MV and on HV) should be established.
In the period 1961–1978 there were no significant changes in this field. In
November 1979, the Yugoslav Council of electrical power engineers (YUKO CIGRE)
established a working group 22.06 “Live working” within the study committee 22 for
Overhead lines. In 1981, the Assembly of Yugoslav electricity community (JUGEL),
based in Belgrade, adopted the sectoral standard GSE – 40/81 (Rules and Measures for
Safety at Work in Electricity Facilities). The standard (JUGEL, 1981) was the first
Yugoslav document that mentioned the possibility of implementing LW on MV and on
HV network above 250 V.
During the construction of the 380 kV Yugoslav transmission network “Nikola
Tesla” at the end of the seventies, several expert visits were carried out abroad. The
delegations that travelled to Russia, France, USA, Hungary and elsewhere also
included Slovenian experts from transmission and distribution companies.
LW activities in Croatia were intense, especially in 1985/86, due to many years of
cooperation between the company Elektroslavonia, Osijek, and the distribution com-
pany DEDASZ, Pecs, from Hungary. Croatia took the initiative and in 1987 the first
republican regulation that permitted LW was approved. First Croatian instructors for
LW on LV were trained in Hungary in the spring of 1990 and in France, in
EDF SERECT, in the winter of the same year.
LW Development in Slovenia 1991–2009
Due to the disintegration of Yugoslavia and the beginning of war in Croatia, the
development of LW was on a standstill for a decade. The Croatian army and later
refugees from Bosnia and Herzegovina moved into the HEP NOC facilities in Slavonia,
leaving the buildings in a very poor condition.
The historical events in Croatia are very important for the Slovenian development of
LW, because the cross-border cooperation started after HEP d.d. started to carry out
LW. In 2000, HEP d.d. launched the renovation of the HEP NOC training centre for
LW. In 2003, HEP d.d. signed a contract with EDF SERECT for the transfer of LW
technology on LV and MV. HEP ODS, d.o.o. (distribution company) started with the
implementation of LW on LV in September 2005.
The turning point for the development of LW in Slovenia was the year 2006, when a
delegation of EG, IRSD, REI and C&G representatives visited HEP and encouraged a
cooperation. In 2007, HEP NOC and C&G signed a long-term cooperation agreement
with the aim of transferring technology and training. In the same year, manuals were
translated as the basis the training. At the same time, intensive communications
between C&G and NEK led to the first training of Slovenian experts in HEP NOC,
Croatia, in 2008.
LW Development in Slovenia 2009–2019
The NEK’s system requirements set high standards for the organization of the first
training. Before the training could begin, C&G (HEP NOC) had to fulfil all internal
366 A. Lovrenčič and V. Lovrenčić

requirements, which were prescribed by services responsible for maintenance, safety at

work, quality management and internal training.
After the validation and acceptance of all internal acts on training, implementation
and deployment of LW in LV in NEK, they started to implement the project. This is the
case, which is an exception in the global practice of LW, Slovenia started firstly in
industrial environment than in the distribution or transmission. As the first Slovenian
legal entity, NEK issued authorization for LW on LV in January 2009 “Fig. 1”.

Fig. 1. LW on LV in NEK [1]

Thanks to the promotion of LW implementation in industry, interest in distribution

and transmission has accelerated. Therefore, in 2011, LW were recorded on LV in TSO
(ELES) and DSO (EG, EL and EM), which spread throughout Slovenia.
An important step in the development of LW was made in 2013, a hundred years
after the first LW in the US, OHIO, when distribution (EG, EL and EM) began with
LW on MV with transformer station (TS) cleaning (Fig. 2).

Fig. 2. LW on MV (cleaning) [1]

 Ten Years’ Experience of Live Work Implementation 367

3 International Framework of LW

Experts in the field of LW clearly understand that without international connections in

Slovenia there is no successful development of LW as they cooperate with neigh-
bouring countries and global associations (LWA, CIGRE, IEEE ESMO, CITTES).

3.1 Training of LW in Slovenia

As already mentioned, the most important international cooperation is the connection
with Croatia, HEP NOC, a training centre in Velika, which, on the basis of the
C&G/HEP NOC “Fig. 3“ contract, conducts training at the Croatian location and on
operators’ facilities when Croatian instructors, together with C&G, complete practical
training in Slovenia.
As already mentioned, the most important international cooperation is the con-
nection with Croatia, HEP NOC, training centre in Velika. On the basis of the
C&G/HEP NOC “Fig. 3“ contract, conducts training at the Croatian location and on
operators facilities with Croatian instructors, together with C&G, complete practical
training in Slovenia.
Today, C&G/HEP NOC has already trained 145 lineman and 106 coordinators
for LW in more than 20 companies, of which 15 lineman and 10 LW coordinators for
MV - cleaning S/S.

Fig. 3. Training centre in HEP NOC, Croatia [1]

368 A. Lovrenčič and V. Lovrenčić

3.2 International Association

With the introduction of the LW, Slovenian experts recognized the need for cooper-
ation with international professional associations that are active in the LW field.
Since 2010 they have been actively participating in CIGRE, SC B2/B3 JWG27,
which published the Technical Brochure No 561 »Live Working – A Management
Perspective« in December 2013 [3].
Following the invitation from Croatian colleagues, Slovenia was accepted as an
observer in The Live Working Association (LWA) at International Conference On Live
Maintenance (ICOLIM) 2011 in Zagreb. This invitation opened the possibility of
international cooperation, which intensified after ICOLIM 2014 in Budapest, when the
collaboration between Hungarian (BME) and Slovenian experts (C&G) strengthened.
Thanks to this collaboration, the new CIGRE WG B2.64 “Inspection and Testing of
Equipment and Training for Live-Line Work on Overhead Lines” was activated in
2015 [7].
Today, Slovenian experts are actively collaborating with colleagues from Austria,
Canada, Czech Republic, Croatia, Italy, Hungary, Germany, Norway and Slovakia and
are also actively working with LWA, CIGRE, IEEE ESMO and CITTES.
SIST TC/DPN was established as a mirroring committee of IEC/EN TC 78 »Live
working« and is actively following international standardization. The committee
translated the EN 50110-1 [8] standard as a guidance for future LW work.

4 Surveys and Research

In the recent period, extensive study and research have been carried out [1]. Previously,
surveys and research were made in the Slovene and international environment [9]
among LW specialist.

4.1 Complex Research of the Impact of LW as Maintenance Method [1]

Organisations have been facing expectations and requirements of various stakeholders
when considering the requirements for quality, health and safety at work, sustainable
development and last but not least, the responsible conduct towards the society in
which an organisation operates. In such conditions, many organisations search for
innovative ways to meet the basic and specific requirements of different stakeholders,
such as society, regulators, employees as well as the expectations of their customers in
order to increase the satisfaction with a service.
Current aspects on quality management and development of contemporary main-
tenance concepts open versatile possibilities of integrating different maintenance
methods that lead an organisation to the efficient and effective management of main-
tenance processes. This recognition has contributed to the fact that the maintenance
process has become an integral element of a strategic consideration of numerous
stakeholders or interested parties in the organisation including employees, manage-
ment, owners, local communities and the state.
 Ten Years’ Experience of Live Work Implementation 369

Organisations invest large amount of funds and sources in innovative processes as

they introduce new approaches to be able to achieve the objectives as well as to
improve their performance. The efficient and effective management of business, pro-
duction or service processes and the maintenance processes is of vital importance for a
further successful development of organisations in many industries.
The introduction and implementation of maintenance of electrical installation using
the live working method can be classified among the innovative efforts of
organisations.
Although a hundred-year tradition of live working exists, live working represents a
new method of maintenance of electrical installation in the generation, transmission
and distribution of electric power as well as in the industry and different institutions
(e.g. hospitals) in Slovenia. The latter brings to the forefront a research problem that
has not been sufficiently addressed in prior literature.
Therefore, the main purpose of the study is to contribute to an in-depth research and
understanding of live working within the framework of the quality, health and safety at
work as well as the efficiency and effectiveness of maintenance processes of electrical
installations. However, studies investigating the interaction between the quality man-
agement, health and safety at work and maintenance process performance of electrical
installations by live working method are relatively scarce.
The theoretical framework presents the main area of the research. The underpinning
points of the theoretical framework are delineated by the resource-based theory
(RBT) and the stakeholder theory (ST). Using both theories the study has been built on
three main segments of the research: (1) quality management in the context of main-
tenance, (2) health and safety in maintenance, and (3) maintenance performance.

4.2 Study of Live Working Effects in Slovenia

Based on agreement with the Slovenian utilities; Elektro Gorenjska (EG), Elektro
Ljubljana (EL) and Elektro Maribor (EM), an outsourcing survey was carried out on
“the introduction and implementation of LW on LV in distribution companies” [9].
The aim of survey was to study how workers, coordinators and other contributors in
the process of deployment and implementation of the LW at LV in the Slovenian
utilities evaluate the performance of preventive maintenance processes.
The questionnaire contained 99 questions divided into six groups. The analysis
included 55 respondents (26 workers and 29 coordinators), out of which 15 are from
the EG, 12 from EL and 28 from EM.
Based on the data from the questionnaire, we made an empirical survey. In order to
empirically assess whether there are significant differences between means of the two
categories of respondents, we applied an independent t-test. The results of the t-tests show
that there is significant difference between mean values for the “Competencies for
implementation LW on LV” and “Efficiency and effectiveness in the implementation of
LW at LV” (t = 2.631, p = 0.011, t = 2.396, p = 0.026, respectively). It appears that
there are no significant differences regarding the “Safety and health (OSH) of LW at LV”
and “Quality management in the implementation of LW at LV” (p > 0.05) (Table 2).
370 A. Lovrenčič and V. Lovrenčić

Table 2. Summary of the results regarding the comparison between workers and coordinators
[9]
Dimension Mc M w t p
Competencies for LW 4.14 3.83 2.631 0.011
OSH at LW 4.75 4.66 0.851 0.408
Quality management at LW 4.58 4.20 2.006 0.066
Efficiency and effectiveness at LW 4.30 4.02 2.396 0.026
Mc – Mean value for coordinators, Mw – mean value for workers.

4.3 International Study of the Live Working Effects [1, 9]

International study of the live working performance was carried out among the LWA
members from Croatia, Czech Republic, France, Germany, Hungary, Ireland, Italy,
Poland, Portugal, Romania, and Slovenia and participants of ICOLIM. Later the survey
included LW specialists who are active in specialised working group of CIGRE [1, 9].
The international survey contained 83 questions divided into five groups. A total of
171 respondents from 36 countries participated in the survey. Most of the respondents
were from Slovenia (44 or 25.7%) and Croatia (27 or 15.8%). International LW experts
were in average represented with a size of 1 to 5. The above-average response was seen
in Hungary (10), Canada (9), Germany (8) and Poland (6).
We carried out a statistical analysis. Observing the four dimensions (“Quality
management at LW” - QM, “Health and safety at LW” - HS, “Reliability and Avail-
ability of LW” – R&A and “Performance of LW providers” – Providers Perf), we can
see that the highest mean value of Slovenian dataset corresponds to the “Reliability and
Availability of LW” (M = 4.52), while the lowest value corresponds to the “Perfor-
mance of LW providers” (M = 4.05). Given the international dataset our results
showed that the highest mean value corresponds to the “Health and safety at LW”
(M = 4.61), while the lowest value corresponds to the “Performance of LW providers”
(M = 4.17). Furthermore, t-test (Table 3) was used to examine whether there are sig-
nificant differences between international and Slovenian dataset concerning the LW
dimensions.

Table 3. T-test for equality of means between international and Slovenian dataset [1, 9]
Dimension t p MD 95% CI
Lower Upper
QM −1.235 0.219 −0.162 −0.421 0.097
HS 1.653 0.103 0.175 −0.037 0.388
R&A 0.428 0.669 0.047 −0.171 0.266
Providers Perf 0.769 0.443 0.118 −0.185 0.420
MD - Mean Difference, CI - Confidence Interval, QM –
Quality management during LW implementation, HS –
Health and safety at work during LW implementation,
R&A – Reliability and availability, PP – Performance of
providers.
 Ten Years’ Experience of Live Work Implementation 371

According to the results of the t-test, there are no significant differences between
two groups of respondents, namely international and Slovenian LW experts. Several
regression tests were performed in order to analyse the effects of deploying the quality
management as well as health and safety practices on efficiency and effectiveness of
LW implementation. It can be argued that both quality management and health and
safety positively influence reliability and availability (b = 0.487, p = 0.000, b = 0.710,
p = 0.000, respectively) as well as performance of LW providers (b = 0.339,
p = 0.000, b = 0.617, p = 0.000, respectively).
The aim of the research was to study the perception of LW experts on LW from the
Slovenian and international perspective. The study delivers empirical evidence telling
that there is no substantial discrepancy between Slovenian and international perspec-
tives regarding the LW implementation factors and LW results. The significant com-
parison between Slovenian and international dataset is shown in “Fig. 4”.

Fig. 4. The results of Slovenian and international survey [1, 9]

5 Conclusions

During the ten-year period of training and implementation of LW, a lot of experience
has been gained, which needs to be upgraded and complemented by ongoing devel-
opments of LW.
In the future, it is necessary to follow trends in Europe and the world, where the
LW method is established on all voltage levels. Slovenian electricity companies can
still be considered as not fully convinced of the need to implement LW both from a
security perspective and from an economic point of view.
In all distribution environments, it is necessary to increase the scope of LW to LV
and MV, the same applies to the rest of the electricity and industrial companies.
For LW it is crucial to train and acquaint a large number of lineman with LW, to
educate all workers actively performing work on the distribution network in order to
achieve a higher level of education and a broad spectrum as a basis for the LW
activities.
372 A. Lovrenčič and V. Lovrenčić

A logical step forward is the deployment and implementation of LW on the MV

network, not just cleaning TS, which is currently being implemented. Although this is a
technical and financial layout for a relatively small network, this step is feasible, but
most likely, if distribution companies establish a joint team that would perform LW on
MV for all distribution companies.
Many stakeholders in companies need an assessment of the effectiveness and
efficiency of the LW implementation at LV and MV, which is the task of active
stakeholders, both electricity and industrial companies. A complex research has already
been done but needs to be presented to interested stakeholders.
The aim in Slovenia is to implement LW method as a method that enables the
maintenance of electrical installations that supports the “zero accidents” idea [10].
The current debate addresses the protection of electricians from an electric arc.
A wider range of stakeholders need to promote the safe implementation of LW, so that
the safety of LW will be emphasized as much as the economic impacts.

References
1. Lovrenčić, V.: The impact of live working as a maintenance method for electrical
installations. PhD dissertation; University of Maribor, Faculty of Organizational Sciences,
Kranj (2018)
2. RTE: Live working. A cutting-edge technique. 50 years of French history, Paris (2014)
3. CIGRE: Live Work – A Management Perspective (Technical Brochure No 561). Joint
Working Group B2/B3.27, Pariz (2013)
4. Looms, J.S.T.: Live working on high-voltage lines. IEEPROC 128(2), 89–106 (1981)
5. Barg, I.G., Polevoj, S.V.: Remont vozdušnyh linij èlektroperedači pod naprjaženiem (1980).
http://forca.ru/knigi/arhivy/remontvlpodnapryazheniem.html
6. Žuravlev, B.: Paboty pod naprjaženiem bezopasnee rabot na otključennyh èlektrous-
tanovkah. NOVOSTI ÈLEKTROTEHNIKI, Žurnal №2 (50), 2008. ENTSOE, Dynamic
Line Rating for overhead lines – V6, CE TSOs current practice 2015
7. CIGRE: Inspection and Testing of Equipment and Training for Live-Line Work on
Overhead Lines, Draft, March 2019
8. CENELEC, EN 50110-1:2013 Operation of electrical installations - Part 1: General
requirements
9. Lovrenčić, V., Pantoš, M., Maletič, D., Maletič, M., Brezavšček, A., Gomišček, B.: Study on
the quality management and health and safety aspects on perceived live working
implementation dimensions. In: 12th International Conference on Live Maintenance,
ICOLIM 2017. LWA, Strasbourg (2017)
10. Lovrenčić, V., Gomišček, B.: Live working as an example of electrical installation
maintenance with the zero accidents philosophy. In: 11th International Conference on Live
Maintenance, ICOLIM 2014. LWA, Budapest (2014)
Monitoring and Diagnostics
 A Model Based on First Principles
for the Simulation of Partial Discharges

A. Villa(B) , L. Barbieri, and R. Malgesini

RSE, 20134 Milan, Italy

andrea.villa@rse-web.it

Abstract. Partial discharges represent one of the main mechanisms of

ageing of dielectrics especially when an alternated current is used. This
problem is particularly severe when partial discharges are associated with
degenerative phenomena such as the electrical treeing. In this work, we
present the latest development of our simulation codes, which are capable
of simulating the evolution of discharges in complex three-dimensional
geometries through parallel high-performance-computing technologies.
The code is capable of both predicting the evolution of some macroscopic
quantities, that can be measured directly, and estimating the progression
of the internal ageing. For instance, it is possible to simulate the creation
of chemically active species in the gas and their interactions with the
surfaces of the branches. Some examples of the results obtained in a set
of test cases will be discussed here.

Keywords: Partial discharges · Simulation · Parallel computing

1 Introduction
In this work we present the capabilities of a 3d simulation tool based on the
solution of a set of partial differential equations (PDEs) for the simulation of the
inception of electrical discharges in gases and of the interaction with insulating
surfaces. The code is both capable of dealing with external discharges such as
surface discharges and internal discharges where the gas regions are internal and
completely surrounded by an insulating matrix. This last configuration is typical
of the evolution of the electrical treeing [1].
In both cases the code has to take into account two regions: the first one
characterized by a mixture of gas species and the second one by a set of insulat-
ing materials. Moreover, since we want to compute the evolution of each single
discharge and the relative effects on insulating materials, we consider a fixed
geometry. The treeing usually includes an evolution of the shape of branches
but we assume that this last evolution is slow with respect to the one of a single
discharge.
Simulations have been widely used in this field. First circuital models [2]
have evolved into more complex ones. The numerical computation of the inter-
nal electric field has been a first step [3,4]. In recent years, also the evolution of
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 375–383, 2020.
https://doi.org/10.1007/978-3-030-31676-1_36
376 A. Villa et al.

the plasma, that fills the branches of electrical trees, has been taken into con-
sideration thus leading to more complete models based on the solution of a set
of PDEs [5–8].
Our approach can be regarded as a further evolution of the latter approach
and, in particular, we pursue two main objectives i.e.: the estimation of the
ageing insulating materials subjected to gas discharges and the reconstruction
of relevant quantities that can be directly measured by external gauges. For
instance, it is important to estimate the electric field in defects to check whether
it is compatible with the dielectric strength of the solid insulating material.
Moreover, the evolution of the internal pressure may affect the internal stress
state of the material thus leading to cracks and the progression of the treeing.
Finally, the generation and deposition of conductive species on the surface of
branches or on the surface of an insulator may cause more intense discharges
leading to the final failure of electrical components.
In the next pages, we will provide a brief introduction to the physical app-
roach which lays behind the simulation code and we will discuss the results of
some technically relevant test cases.

2 The Model

The model we have implemented can be regarded as an evolution of the one

described in [5] where the treatment of the complex chemical gas chemistry,
detailed in [9], has been added. The algorithm can handle relatively large time
steps and mesh spacing without incurring numerical instabilities, for more details
see [10–12]. Thus the numerical code can handle some complex, realistic geome-
tries substantially reducing the computational burden.
From a physical point of view, the model takes into account the evolution of
a reacting gas, by means of the Euler equations of gas-dynamics. A proper set
of chemical reactions describes the creation and annihilation of reactive species
and free charges. These latter ones are drifted by the electric field and their
motion is coupled by means of the electrostatic equation. Most of the charges
are created by impact ionization by fast electrons; however a few are emitted from
the surface of insulating materials by Schottky effect. These electrons provide
the inception of internal discharges. Some more seed electrons may be created
by photoionization which has been simulated using the approach detailed in [13].
The pressure field is computed in the gas region while the temperature field is
solved for both the gas and the insulating regions.
Moreover the above-described approach has been complemented by the sim-
ulation of the surface chemistry, which includes the interaction between a react-
ing gas and a polymeric surface. In principle many chemical databases can be
treated, however, here we consider the approach of [14] which takes into account
mainly the oxidation of a polyethylene surface. Some future evolutions will also
consider the emission of hydrocarbons in the gas and the formation of graphite
on the surface. Another novelty is the inclusion of the dynamics of charges, such
as electrons and holes, in insulating and semi-conductive regions, as described in
 Simulation PD 377

[15]. This feature is particularly important when dealing with applications where
a DC power supply is used although we anticipate that most of the applications
we are going to discuss deal with AC supplies.
The code we have developed is also capable of estimating some macroscopic
quantities that can be measured directly such as the induced currents on a proper
set of electrodes. This feature has been mainly used in the validation phase which
has been carried out in [5,9].

3 Surface Discharges Modeling

To show the performances of the code regarding external discharges we have

considered the simulation of a dielectric barrier discharge or, in other words,
the evolution of surface discharges. We have considered a cylindrical specimen
of polyethylene with a radius of 50 mm and 12.3 mm thick. The bottom of the
specimen is electrically grounded and above its top, at a distance of 2.2 mm,
a conical hollow electrode is placed. The smallest radius of the cone is 2 mm,
the largest one is 5 mm and the electrode is energized through an AC power
supply with a voltage of 9.08 kV and a frequency of 50 Hz. The same geometry
has been implemented in an experimental apparatus to gain an insight of the
chemical modifications due to surface discharges, see [14] for more details. The
gas has been modeled using the chemical database A described in [9], the initial
relative concentrations of gases are 79%, 20%, 1% and the pressure, temperature
conditions are considered standard: T = 298 K, P = 101 kPa. The simulation
covered a period of 80 ms. We will now discuss some relevant results concerning
the last discharge which is quite close to the simulation final time t = 80 ms.
In Fig. 1(a) we have depicted the concentration of electrons: as we notice,
the discharges are asymmetrical and involve only a minor part of the electrode.
This is due to the fact that the electrode is quite big when compared to those
most commonly used for tests in the literature, see for instance [16]. Due to its
dimensions the front tends to develop quickly some instabilities and to lose its
symmetry. Moreover, after a large number of discharges also the concentration
of charges on the insulating surface is non-symmetrical, see Fig. 1(b). As a con-
sequence, the electric field on the needle exhibit some more intense regions that
correspond to the regions where new discharges will more likely incept. Each
discharge affects a different region near the needle, injecting some energy in the
gas resting in this area. The energy is either consumed by endothermic chemical
reactions or converted into heat. As we can see from Fig. 2, the variations of
temperature are modest and mostly concentrated in the region just below the
electrode. It is still clearly visible a temperature gap of a few degrees between
the air and the gas that is due to initial conditions.
Most of the energy produced by discharges is employed for the creation
of chemically reactive species, such as atomic oxygen, as depicted in Fig. 3(a).
This atom and other radicals, such as O, play a major role in the oxidation of
polyethylene surfaces, in fact they abstract hydrogen atoms, previously attached
to the polymers chains. The void is quickly filled by other elements leading to
378 A. Villa et al.

Fig. 1. Concentration of electrons (Fig. 1(a)) and surface electrons (Fig. 1(b)) during
the last discharge event of the simulation (units 1020 m−3 ).

Fig. 2. Temperature field during the last discharge event of the simulation (units K).

a sequence of reactions that may eventually cause the β-scission of polyethylene

chains. However in these conditions, and during this relatively short simulation
time, the most probable event after the abstraction of an H atom is the filling
of the void by an O2 molecule. In Fig. 3(b) we have depicted the density of sur-
face states occupied by oxygen molecules such as (−CH2 −) − OO. As soon as
the chemical databases that describe the surface reactions will be completed, it
will become possible to predict the formation of graphite and surface liquid and
solid compound as observed in [14] with electron microscope and spectrometer
analysis.

4 Medium Voltage Cable Modeling

We consider the geometry of a medium voltage cable (rating close to 20 kV r.m.s.)

with a radius of the internal conductive region of 8 mm, an external radius of the
insulating material of 14 mm and a semi-conductive layer spanning the interval
between 14 and 15 mm. The insulating region is assumed to be polyethylene with
a relative permittivity of 2.2 while the semi-conductive one has a value of 6. In
this case, we have placed, at a distance of 10 mm from the axis of the cable, a
spherical defect of 1 mm in diameter. The cable is energized with an AC power
supply of 30 kV (peak voltage) and 50 Hz. The sphere is filled having air with
 Simulation PD 379

Fig. 3. Concentration of atomic oxygen (Fig. 3(a)) and of surface sites of OO radicals
(Fig. 3(b)) during the last discharge event of the simulation (units 1020 m−3 ).

the same characteristics described in the previous section. The total simulation
time is 80 ms and a discharge has been generated for each semi-period.
The overall geometry and the electric field corresponding to the first discharge
is depicted in Fig. 4 while, in Fig. 5, we have shown the details of the evolution of
the streamer. The electrons emitted from the surface of the defect, by Schottky
emission, are accelerated by the electric field and, if conditions are favourable,
they generate an avalanche effect. We stress that no inception condition has to
be imposed, this is implicitly determined by the model itself and the swarm
parameters used. When enough positive charges are accumulated in the gas,
then a ionization front travels through the defect: a positive streamer has been
created. This reaches the opposite surface and it disappears after a short while.
The ionization front may also be guided by the surfaces of the defect creating
some more complex discharge structures [17].

Fig. 4. Two views of the geometry (Fig. 4(a)) and of the internal spherical defect
(Fig. 4(b)). The electric field corresponding to the first discharge is shown (units kV/m).

To show the capabilities of the code we have repeated our calculations sub-
stituting the sphere with a more involved geometry that loosely resembles the
380 A. Villa et al.

Fig. 5. The evolution of a streamer corresponding to the first discharge of the simula-
tion. Both the inception (Fig. 5(a)) and the extinction (Fig. 5(b)) of the streamer are
depicted. The concentration of electrons is shown (units 1020 m−3 ) in a section of the
sphere. The mesh on the surface of the sphere is outlined as well.

complexity of the branches of electrical trees. This has been obtained joining
three tubes: the biggest is 1 mm long and has a diameter of 0.2 mm, the others
are 0.5 and 0.2 mm long with diameters of 0.1 mm and 0.125 mm, respectively.
Contrary to the sphere case, this one is un-symmetrical since the upward-
travelling streamers, see Fig. 6, encounter a geometry that is different from the
one encountered by the downward-travelling streamers, see Fig. 7. Moreover sev-
eral different evolutions have been observed for the upward-travelling streamers:
some travel only through the shorter channel, while others involve both channels.

Fig. 6. The evolution of two, upward-directed, streamers: the first (Fig. 6(a)) involves
only the upper left branch while the second (Fig. 6(b)) involves both. The concentration
of electrons is shown (units 1020 m−3 ).

From the post-processing of our data it is possible to extract much useful

information such as the electric field as depicted in Fig. 8. In particular, it can
be checked whether the electric field remains below the dielectric strength of the
insulating material. If this threshold is crossed, one can expected the evolution of
the treeing to be quite fast as, in this case, each discharge would cause substantial
 Simulation PD 381

damages to the insulating bulk. Another useful quantity is the ageing of the
surfaces of the defect. In Fig. 9 we have depicted the concentration of (−CH2 −)−
OO radicals as we did in the previous section. This gives some information
regarding the regions that most likely will be eroded and could generate new
branches.

Fig. 7. The evolution of a, downward directed streamer. The concentration of electrons

is shown (units 1020 m−3 ).

Fig. 8. The electric field corresponding to an upward-directed streamer (units kV/m).

Fig. 9. Concentration of surface sites of OO radicals (units 1020 m−3 ) at the end of the
simulation period.
382 A. Villa et al.

5 Conclusions
We have demonstrated that the computational approach we have developed can
handle complex and realistic configurations, generating a comprehensive set of
data, which can shed new light on the ageing of polymeric materials. The code
has been designed from the scratch to include many effects such as space charge
accumulation in the insulating bulk and on surfaces of discontinuity between
different materials. This makes it possible, as soon as a proper set of character-
ization parameters of the plastic materials used is available, to extend readily
the analysis to HVDC cases.
The details and comprehensiveness of this approach come at a price since the
computational burden can be handled, in competitive time scales, only using high
performance parallel machines. For these reasons, some novel multi-scale tech-
niques will be developed to speed up the computations when long-time ageing
estimations are required.
Nevertheless the proposed approach has already demonstrated to be very
useful to develop new measurement systems for sensing partial discharges. In
particular, the computational approach is very attractive during the optimiza-
tion of the device, when several instrument configurations can be tested at the
same time without needing to physically construct them. Moreover, using this
approach, it is possible to get a strict control on the defect geometry. Also con-
sidering the fact that, on the contrary, the construction of a known defect in an
insulating system is, typically, a challenging task.

Acknowledgments. This work has been financed by the Research Found for the
Italian Electrical System under the Contract Agreement between RSE and the Ministry
of Economic Development. The authors wish to thank L. Barbareschi for her valuable
contribution and suggestions.

References
1. Wu, K., Chengwei, Z., Wu, G., Xie, H.: A new fractal model for describing time-
dependence of treeing growth. In: International Symposium on Electrical Insulating
Materials, pp. 157–160 (1995)
2. International Electrotechnical Committee: International standard IEC 60270.
Technical report, IEC (2015)
3. Rowland, S.M., Schurch, R., Pattouras, M., Li, Q.: Application of FEA to image-
based models of electrical trees with uniform conductivity. IEEE Trans. Dielectr.
Electr. Insul. 22(3), 1537–1546 (2015)
4. Enokizono, M., Tsutsumi, H.: Finite element analysis for discharge phenomenon.
IEEE Trans. Magn. 30(5), 2936–2939 (1994)
5. Villa, A., Barbieri, L., Gondola, M., Leon-Garzon, A.R., Malgesini, R.: A PDE-
based partial discharge simulator. J. Comput. Phys. 345, 687–705 (2017)
6. Lin, B., Zhuang, C., Cai, Z., Zeng, R., Bao, W.: An efficient and accurate MPI
based parallel simulator for streamer discharges in three dimensions. Technical
report, arXiv (2018)
 Simulation PD 383

7. Callender, G., Golosnoy, I.O., Rapisarda, P., Lewin, P.L.: Critical analysis of partial
discharge dynamics in air filled spherical voids. J. Phys. D Appl. Phys. 51(12),
125601 (2018)
8. Callender, G., Tanmaneeprasert, T., Lewin, P.L.: Simulating partial discharge
activity in a cylindrical void using a model of plasma dynamics. J. Phys. D Appl.
Phys. 52(5), 055206 (2019)
9. Villa, A., Barbieri, L., Gondola, M., Leon-Garzon, A.R., Malgesini, R.: Simulation
of the AC corona phenomenon with experimental validation. J. Phys. D Appl.
Phys. 50, 435201 (2017)
10. Villa, A., Barbieri, L., Gondola, M., Malgesini, R.: An asymptotic preserving
scheme for the streamer simulation. J. Comput. Phys. 242, 86–102 (2013)
11. Villa, A., Barbieri, L., Gondola, M., Leon, A., Malgesini, R.: An implicit three-
dimensional fractional step method for the simulation of the corona phenomenon.
Appl. Math. Comput. 311, 85–99 (2017)
12. Villa, A., Barbieri, L., Gondola, M., Leon-Garzon, A.R., Malgesini, R.: Stability of
the discretization of the electron avalanche phenomenon. J. Comput. Phys. 296,
369–381 (2015)
13. Ségur, P., Bourdon, A., Marode, E., Bessieres, D., Paillol, J.H.: The use of an
improved Eddington approximation to facilitate the calculation of photoionization
in streamer discharges. Plasma Sources Sci. Technol. 15(4), 648–660 (2006)
14. Leon, A.: Physico-chemical study and modelling of partial discharge phenomenon
in polymeric dielectric materials. Ph.D. thesis, Politecnico Di Milano (2017)
15. Serdyuk, Y., Gubanski, M.S.: Computer modeling of interaction of gas discharge
plasma with solid dielectric barriers. IEEE Trans. Dielectr. Electr. Insul. 12(4),
725–735 (2005)
16. Tran, T.N., Golosnoy, I.O., Lewin, P.L., Georghiou, G.E.: Numerical modelling of
negative discharges in air with experimental validation. J. Phys. D Appl. Phys.
44(1), 015203 (2011)
17. Babaeva, N.Y., Kushner, M.J.: Structure of positive streamers inside gaseous bub-
bles immersed in liquids. J. Phys. D Appl. Phys. 42(13), 132003 (2009)
 Contactless Thermal Online-Monitoring
of Electrical Equipment Under Load at High
Voltage to Determine the Load Level
and Damage Avoidance

Thomas Gräf(&)

Hochschule für Technik und Wirtschaft Berlin,

Berlin University of Applied Science for Engineering and Economics,
Wilhelminenhofstr. 75A, 12459 Berlin, Germany
thomas.graef@htw-berlin.de

Abstract. The meanwhile high age of the electric energy distribution in the
industrialized countries as well as the avoidance of new investments, the target
to reduce maintenance resources and to use the assets at the maximum effi-
ciency, accelerate the desire for the implementation of online monitoring
systems.
One of the most important physical dimensions for the judgement of the
condition of an electrical device is the temperature at the current path. Prob-
lematical, with the temperature monitoring is that the measuring places are often
not directly accessible or on high voltage level, so that measurements need to be
done contactless.
Today new electronic components with miniaturized dimensions and low
energy consumption allow very small design applications, so that the imple-
mentation inside high voltage equipment will be possible using of energy
harvesting.
The measurement of the temperature inside or at electrical equipment like
switchgear, HRC fuses, GIS- or GIL-systems, cabinets, cable connections, cable
joints or bushings allows the judgement of the load level and the estimation of
the thermal stress of the insulation or contacts. Additional functionality will be
generated when the ongoing current is measured together with the temperature.
The analysis of both parameters, allows the estimation of e.g. contact’s condi-
tion as well as the thermal condition of high voltage HRC fuses before these will
explode due to aging cased by overheating. The detection of thermal stress due
to short circuit current flow is possible too.

Keywords: Energy harvesting
Contactless
Temperature monitoring

1 Introduction

The measurement of temperatures at electrical equipment will be quiet difficult if the

equipment is in operation and additionally high voltage is present. Problematical,
however, with the temperature monitoring is that the measuring spots, which have to be
supervised, are often not directly accessible or on high voltage level, so that

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 384–395, 2020.
https://doi.org/10.1007/978-3-030-31676-1_37
 Contactless Thermal Online-Monitoring of Electrical Equipment 385

measurements need to be done contactless. The development of new electronic com-

ponents with low energy consumption and miniaturized dimensions allow very small
designs, so that the implementation inside high voltage equipment will be possible.
Problematic is the energy distribution for the electronic measuring system working at
high voltage level because wire connections are not suitable. Due to this fact it would
be advantageous to use the heat energy around the measuring places caused by the
current flow.

2 Energy Harvesting Sensor - Design of Measuring Sensor

Preliminary investigations [3–6] have shown that the high voltage- and the thermal-
requirements mainly influence the housing design of the temperature sensor. At low
voltage applications, the housing can be made out of plastic or ceramic with an outer
contour that is insulating. When used in the vicinity of high voltage, the outer contour
of the housing must be continuous in its course, which means there must be no “edges”
and be electrically conductive. Such a developed housing is shown in Fig. 1 and was
used successfully for prototype testing within a switchgear (Fig. 12). An additional
application is shown in Fig. 2. This figure shows a housing for curved applications like
cables, bushing or HRC fuses. Figure 3 shows details about the optimization of the
design of the housing contours to reduce the electrical field strength around the sensor
case. As an example the temperature monitoring for HRC fuses will be shown within
Sect. 3.2. Within this chapter the temperature distribution is shown while the HRC fuse
degenerates due to repetitive load cycles. Additional applications like cable temperature
monitoring, measurement of the temperature of the case of gas insulated line mea-
surement at contact locations, transformer bushing temperature monitoring, tempera-
ture monitoring of contacts at circuit breaker or similar applications are now possible to
measure with the presence of high voltage.

15

114 43

Fig. 1. Housing of the TOC measuring sensor (dimensions in mm)

2.1 Internal Design

The internal structure of the measuring sensor is simple and flexible and shown in
Fig. 3. In the first place, temperatures are recorded. These are relatively easy to
measure and provide a good indication of the workload and condition of a resource.
386 T. Gräf

With little extra effort, other physical quantities can be recorded too. The measuring
system is supplied with electricity generated with a thermogenerator - Peltier element -,
which converts the thermal energy from the heat output of e.g. a busbar into electrical
energy and stores it in a buffer. This makes it possible for a short time to provide a
relatively large amount of energy for the radio transmitter, so that the information can
be sent galvanically isolated to the next collecting node outside the high voltage area.
On average, the system requires significantly less than 1 mW on power for transmitting
data. For the housing material was deliberately no copper, but a special alloy used. The
selected material combination results in a significant temperature difference between
the housing surface and the busbar over a longer period of time and thus also on both
sides of the thermogenerator. This ensures that there is always enough electrical energy
available for the measuring system when there are temperature changes. If the tem-
perature does not change, no measured value is required. According to this principle,
the firmware of each measuring sensors is designed. If the temperature at the sensor
changes, the measuring sensor wakes up from the standby state and sends the current
measured value to the collecting node and returns afterwards to the standby state. This
saves energy. The measuring sensors work as pure analogue-digital-converter. For
future applications it is possible to combine the measured temperatures with e.g. the
relative humidity or vibrations measured by the same sensor. This allows additional
monitoring aspects for the safety and reliability of electrical equipment. The collecting
node receives by radio signal the encrypted directly converted ad value of the mea-
suring element. On the one hand, this increases interference immunity and at the same
time saves energy. On the other hand, the overall system remains flexible, as software-
related adjustments can be easily made in the collective node, which is integrated in the
IT-infrastructure of the plant operator. An intervention on the actual, built-in measuring
sensor is therefore not necessary (Fig. 4).

Fig. 2. Housing of the TOC-fuse-sensor for curved applications like fuse temperature
monitoring
 Contactless Thermal Online-Monitoring of Electrical Equipment 387

Standardized electrical field strength

Cuboid with sharp

edges
Cuboid with radial edges
Clothoid contours
Actual contours

Standardized length along outer contours

Fig. 3. Calculated electrical field strength as a function of the contours (ref. Fig. 1)

Fig. 4. Structure of temperature-measuring sensor with contactless data transmission

2.2 Data Handling Solutions

Different kind of solutions exists for the data receiving and handling. Figure 5 shows
the already existing solutions based on mainstream technologies. The first solution uses
a programmable logic controller (1. solution). With this kind of solution the interaction
and communication with industrial hard- and software is directly possible. There are no
customized and proprietary components necessary. The second solution uses a cloud
basis to collect the received data. The operator is able to define temperature limits and
messages that allows seeing the actual temperature and exceeded limits. As well it is
possible to switch off equipment in case of an emergency case. The information
is transferred by email or sms to all available predefined communication systems
(2. solution). Another solution is based on the idea, that it is dangerous to open
388 T. Gräf

switchgear while these are in operation. The switchgear is metal-enclosed and closed
while it is in operation, so a temperature measurement with thermocouple or with a
pyrometer is impossible. The opening of a door for measurement will not be accepted
due to the danger of arcing and the impact to the maintenance stuff and also the
temperature will immediately decrease (3. solution).
But only at that time it is possible to measure the temperature because a current
needs to flow generating a remarkable temperature. For evaluation of the existing
temperature (distribution) as a function of the current flow the actual temperature needs
to be measured and compared to former values. Additional information may be
gathered if nowadays data analytical software are used like modelling and deep
learning algorithms for evaluating the equipment.

Fig. 5. Different solutions for the data handling of monitoring the electrical equipment

3 Thermal Online Monitoring at HRC Fuses

The application deals with the use of a TOC fuse-sensor for monitoring the thermal
condition of medium voltage high rupture capability fuses (HRC fuses) [6] while these
are in operation. The fuses have inside a different number of parallel used fuse elements
to transfer the rated current. Because of the flowing current and thereby caused losses
the fuses heat up. There are a couple of aspects known that influence the lifetime of a
fuse. Several of these physical aspects lead to an unexpected damage of the ambient
area of the fuse due to an explosion. Some of the root causes for the degradation of
fuses are well known [7–10]. Further causes have become more apparent in the recent
past like
• variations in production quality,
• wrong engineered switchgear,
• mechanically induced vibrations,
• strongly pulsate currents.
 Contactless Thermal Online-Monitoring of Electrical Equipment 389

HRC fuses are used within compact medium voltage switchgear and these are often
integrated within the towers of wind energy converters or part of solar energy plants. In
the meantime it is known that strongly pulsation of currents together with the
mechanical expansion and contraction lead to a break of fuse elements inside the fuse.
The breakdown of a single fuse element increases the thermal losses of the fuse but
does not interrupt the current flow. The fuse is still able to transfer the nominal current
due to the fact, that there are a few fuse elements in parallel to transfer the rated current.
Figure 6 shows the principal combination of the number of broken fuse-elements and
the increasing thermal losses. Often the fuses are installed within a fuse tube, so that
there is no access given to measure the temperature by pyrometer or a thermocouple.
Supplementary problem arises because the developing additional heat caused by bro-
ken fuse elements cannot be forwarded due to the housing effect of the fuse tube. As a
result of this the sealing of the fuse tube gets damaged, the insulation gas disappears
and the risk of flashover with following short circuits currents dramatically rises.

Fig. 6. Thermal losses PV vs. broken fuse elements, Pr losses at rated current

3.1 Thermal Investigations

On the basis of the small design of the arrangement fuse, fuse tube and the thermal
situation inside, investigations were needed. The aim was the sufficient energy har-
vesting to power the TOC-fuse-sensor, to measure the temperature and to transfer the
measured temperature by radio frequency to the receiver which was located outside of
the high voltage area. Figure 7 shows six possible positions to place the TOC-sensor
for monitoring the increasing temperature of the HRC fuse elements. The fuse tube was
installed in a compact SF6 gas insulated compartment, so that the positions 1–6 were
accessible for fixing the TOC-fuse-sensor.
Additionally measurements were taken to determine the largest temperature gra-
dient for energy harvesting and for measuring in the the event that fuse elements will
break.
390 T. Gräf

Important for the function of the TOC-sensor was the availability of a temperature
gradient of at least 5 °C to gain enough thermal energy for the conversion to electrical
energy and to power the electronic circuits. The temperature was additionally measured
10 mm above the surface of the HRC fuse because of the thickness of the TOC-fuse-
sensor. The HRC fuse had the hottest area in the middle, to the ends the temperature
decreased due to thermal conduction of the accessories and the conductors. As well
there was a convective flow of heat inside the fuse tube that led to the measured
temperature distribution. At the bottom side of the fuse the temperatures was lower than
at the top side. Similarly the temperature decreased to the ends of the fuse.
As a result of the performed measurements the TOC-fuse-sensor was placed at the
position 5 shown in Fig. 7 corresponding to Fig. 8.

Fig. 7. Measuring points inside switchgear around HRC fuse

TOC

Fig. 8. TOC-fuse-sensor fixed at HRC fuse while inserted in GIS-switchgear

 Contactless Thermal Online-Monitoring of Electrical Equipment 391

For the testing of a TOC-fuse-sensor within an original SF6 medium voltage

switchgear with a rating of 24 kV a test setup consisting of a switchgear with HRC
fuses was erected and equipped with thermocouples.

3.2 Thermal HRC Fuse Monitoring

The number of fuse elements of the fuse at phase B was adapted for the simulation of a
degenerated HRC fuse. The original HRC fuse with a rating of 63 A had 4 fuse
elements inside whereas the fuse manipulated by the manufacturer for investigations of
the rate of rise of temperature and testing the function of a TOC-fuse-sensor had only
one fuse element implemented.
In general the positioning of HRC fuses inside fuse tubes led to increased thermal
losses while the fuses where burdened with 2/3 = 41 A of the nominal rated current of
the fuse. This reduced current represented the maximum allowed current otherwise the
temperature inside the fuse tube reaches more than 110 °C and led to a degradation
process to the fuse tube and to destroy the switchgear, because the SF6 tightness got
lost. The fuse tube consists of fibre reinforced cast resin. Figure 9 shows the result on
the temperature measurements while three fuses were in operation inside the SF6 gas
insulated medium voltage switchgear. The manipulated fuse together with the TOC-
fuse-sensor was installed at phase B.
There the temperature reaches 110 °C after 30 min of operation which was the
maximum tolerable temperature for the fuse tube. Phase A and phase C reached at that
time a maximum temperature level of round about 36 °C. The temperature decreased to
both ends of the fuse and reached the highest temperatures in the centre of the fuse. The
heat transfer led to the heating of the upper fuses within the fuse tube for phase A and
similar to fuse tube of phase B. There the lower fuse at phase C heated up the upper
fuse with its fuse tube at phase B so that the final temperature was higher than the fuse
below in phase C.

Fig. 9. Temperatures inside fuse tubes at different locations at three phases, point 15 with loose
thermocouple
392 T. Gräf

The dotted points represent measured temperatures by the TOC-fuse-sensor. The

temperature values were transmitted from inside the fuse tube to a receiver placed
outside the closed switchgear cabinet to receive the data. Figure 10 shows the unbal-
anced temperature distribution while a degenerated HRC fuse heats up.

Fig. 10. Thermal image of a HRC fuse with one broken fuse element at a heat-up phase [9]

4 Measurement at GIS/GIL-Systems

Another application deals with the use of the TOC-sensor for the monitoring of the
temperature of contacts mounted in GIS- and GIL-systems for ac or dc applications. In
these systems it is difficult to place any temperature measuring system inside because
of the high field strength and the danger of occurrence of partial discharges. Because of
this phenomenon the TOC-sensor can be mounted around the GIS- or GIL-
compartment, as well as inside. Inside the GIS- or GIL-compartment the thermal gas

Fig. 11. Thermal image of a GIS-tube with implemented contact-system

 Contactless Thermal Online-Monitoring of Electrical Equipment 393

flow will heat up the body of the system and lead to a temperature distribution at the
surface (Fig. 10). There it is possible to measure the temperature and the temperature
gradients. Together with the knowledge of the thermal model of the GIS- or GIL-
system (gas pressure, thermal conductivity of the used metal case, etc.) and the current,
it is possible to calculate the actual temperature and to estimate the condition of the
contact system. With this kind of solution it will be possible to supervise the contacts
without the need to open the gas filled system. For already existing GIS- or GIL-
systems it can be easily subsequently assembled and used for maintenance and mon-
itoring purpose. Figure 12(a) shows the principle arrangement of a TOC-sensor to a
GIS- or GIL-system, Fig. 12(b) shows the gas flow inside and outside the heated area
around the contact system, which is responsible for the heat transfer. So it will be
possible to measure the contact system temperature and start maintenance or load
management as a function of the current and temperature and time. Additionally it is
possible to detect deviations in the temperature profile due to arising contact problems.
This will be worth to notice, because the intended German high energy connections
from the north of Germany to the south might be erected as GIL-systems with several
thousand contact systems (Fig. 11).

Fig. 12. (a) Temperature distribution along contact system – model, (b) thermal caused gas flow
inside and outside GIL- or GIS-system case – cross-section
394 T. Gräf

5 Monitoring at Bushing Inside Switchgear

An additional application is shown in Fig. 13. So it is possible to measure the tem-

perature at critical points inside metal-enclosed switchgear [1, 2, 3, 5] for example the
bushings, screw joint or at the terminals of circuit breakers. Because of the high radio
frequency it is possible to receive the transmitted signals outside the metal enclosed
switchgear. There is no need to implement any antenna inside the switchgear, which
would lead to discussions about already existing type testing reports and emc
compatibility.

3
2

Fig. 13. Photo of the busbar within switchgear, screw joints and bushing

1 – TOC-sensor
2 – Reference-temperature at the screw joint of the bushing
3 – Thermocouple for the measurement of the ambient temperature

6 Conclusions

The contribution deals with thermal monitoring considering the ambient temperature
for prevention of upcoming damages. With the implementation of thermal monitoring
additionally short term overload management is possible. It is possible to generate a
reduction on invest by using the maximal load capacity under consideration of the
thermal status of electrical assets. Even it is possible to monitor the temperature while
short circuit current exists. Additionally support for maintenance purposes is possible
together with the reduction of the risk of opening of doors at encapsulated switchgear.
The measurement at unreachable measuring spots is shown. Connectivity to any
communication standard is established. Even a cloud based solution is possible for
managing the whole asset of an electrical energy distribution system. The TOC-sensors
need no batteries, so they are free of maintenance.

Acknowledgments. The IPH Institut “Prüffeld für elektrische Hochleistungsprüfungen” GmbH

Berlin – member of the CESI group - should be expressly thanked for the support of short circuit
current testing of the TOC-sensor system.
The NH-HH recycling Verein should be thanked for the support of the research project.
 Contactless Thermal Online-Monitoring of Electrical Equipment 395

References
1. IEC 62271-1:2007 + AMD1:2011 High-voltage switchgear – Part 1: Common specifications
2. IEC 62271-200:2011 High-voltage switchgear and controlgear – Part 200: AC metal-
enclosed switchgear and controlgear for rated voltages above above 1 kV and up to and
including 52 kV
3. Patent: Autarke Temperaturmessung DE 102013002206.1, January 2013
4. Haas, U.: Thermischer Schaltanlagenschutz durch Hochspannungssicherungen mit inte-
griertem Temperatur-Begrenzer unter Berücksichtigung der IEC 420:1990. Sonder-
druck SIBA Lünen (1990)
5. Gräf, T.: Berührungsloses Online-Monitoring an elektrischen Betriebsmitteln zur Bestim-
mung des Auslastungsgrades und Schadenvermeidung. ETG Fachtagung Diagnostik
elektrischer Betriebsmittel Berlin 25./26. November 2014, VDE Verlag Berlin (2014)
6. DIN EN 20282-1 VDE 0670-4 Hochspannungssicherungen – Teil 1: Strombegrenzende
Sicherungen (IEC 60282-1:2009) VDE Verlag GmbH, Berlin
7. Gräf, T.: Schadenvermeidung beim Einsatz von Hochspannungs-Hochleistungssicherungen
durch Online-Temperatur-Monitoring, Netzpraxis, Jg. 54, H. 7–8, 22–26 (2015). ISSN
1611-0412-D 7656 E
8. Cigré Session Paris 2016, France, Gräf, T.: Contactless thermal online-monitoring of
electrical equipment under load to determine the load level and damage avoidance. A3-204
(2016)
9. IEC 60282-1:2009 High-voltage fuses – Part 1: Current-limiting fuses
10. Gräf, T.: Patent Anordnung mit einer elektrischen Sicherungseinrichtung und einer an der
Sicherungseinrichtung angeordneten Messeinrichtung sowie Messeinrichtung. DE 10 2015
100 399.6
 Decomposition Characteristics
of SF6 and Component Features Extraction
Under Negative DC Partial Discharge

Mingxuan Zhang1 , Yulong Miao2 , Qiang Yao2 ,

Fuping Zeng1(&) , and Ju Tang1
1
Wuhan University, Wuhan, China
Fuping.zeng@whu.edu.cn
2
Electric Power Research Institute State Grid Chongqing
Electric Power Company, Chongqing, China

Abstract. To obtain the decomposition characteristics of SF6 under different

types of negative DC partial discharge, and then establish a fault diagnosis
method based on decomposition component analysis, the author extracts four
typical defects that may appear in SF6 gas-insulated electrical equipment, and
obtains the decomposition characteristics caused by the four kinds of defects
under negative DC. The differences of SF6 characteristic decomposition com-
ponent concentration and concentration ratio in the four cases are analysed, and
the results show that the SF6 gas will decompose under the negative DC-PD
caused by the four defects and generate five stable decomposed components that
are CF4, CO2, SO2F2, SOF2, and SO2. The concentration and concentration ratio
of five gas both change with its specific regulation. Spectral Clustering algorithm
is used to identify and diagnose the two feature quantities, which shows that the
two types of characteristic parameters can effectively reflect the PD type, and the
concentration ratio performs better.

Keywords: Negative DC PD
Decomposition component of SF6
Spectral

Clustering algorithm

1 Introduction

Pure SF6 has excellent insulation and arc extinguishing properties at normal temper-
ature and pressure, which has been widely used as insulation medium and arc extin-
guishing medium inside gas insulated equipment since 1960s [1]. Nowadays DC SF6
gas-insulated electrical equipment has attracted great attention at home and abroad due
to its technical advantages in improving system operation reliability and reducing
equipment footprint [2, 3]. However, in the design, manufacture, transportation,
installation, operation and maintenance of SF6 gas-insulated switchgear (GIS) equip-
ment, various insulation defects will inevitably occur inside, which can gradually
evolve in insulation faults, and brings potential threats to equipment and the entire
power grid [4]. Literature [4–6] shows that SF6 gas will decompose under PD fault, and
carry out a series of reactions to generate products such as SO2F2, SOF2, H2S, and HF,

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 396–405, 2020.
https://doi.org/10.1007/978-3-030-31676-1_38
 Decomposition Characteristics of SF6 and Component Features Extraction 397

CO2, SO2, CF4, SiF4. The literature [7–9] shows that the PD type and discharge energy
caused by different insulation defects are different, which leads to the difference in the
content, type and production rate of SF6 decomposition products. The literature [10,
11] analyzed the decomposition characteristics of SF6 under different AC PD and the
feasibility of fault diagnosis for GIS equipment. Therefore, it is possible to detect the
insulation state and fault diagnosis of SF6 GIS equipment by gas component analysis
(GCA), which can effectively avoid the interference of external electromagnetic and
noise [12, 13].
At present, the researches on the decomposition characteristics of SF6 at home and
abroad mainly focuses on the AC field, although the research on the DC field started
earlier, it has not been systematically studied. Van Brant used the needle plate model to
study the decomposition mechanism of SF6, and proposed a ‘regional decomposition
model’ [14, 15]. However, there is no research on the decomposition characteristics
and fault diagnosis methods of SF6 under different PD types inside DC GIS equipment.
Therefore, based on the SF6 DC PD decomposition platform and four typical insulation
defect models, the decomposition components of SF6 under PD caused by four dif-
ferent defects are obtained. Based on the decomposition components, three concen-
tration ratios: C(SOF2 + SO2)/C(SO2F2), C(SO2F2 + SOF2 + SO2)/C(CF4 + CO2), C
(CO2)/C (CF4) are constructed. Spectral Clustering algorithm is used to identify and
diagnose the two feature quantities: component concentration and concentration ratio,
and it is found that both can characterize the PD type, which lays a foundation to
evaluate the insulation state and fault diagnosis of DC SF6 GIS equipment using GCA
method.

2 Experiment

2.1 Experimental Wiring

The experimental wiring of SF6 decomposition under negative DC-PD is illustrated in
Fig. 1. A voltage regulator (T1: 0–380 V) and a testing transformer (T2: 50 kVA/
100 kV) provide the AC high voltage (AC-HV). The AC-HV is converted into DC
testing voltage by using a half-wave rectifier circuit, which comprises a HV silicon
stack (Ds: 100 kV/5 A) and a filter capacitor (Cf: 0.2 lF). Two protective resistors (R1:
20 kX, R2: 20 kX) are used to protect the system. A capacitive voltage divider (Cv) is
used to measure the value of AC output voltage of the transformer, and a resistive
voltage divider (Rv) is employed to measure the value of DC testing voltage applied
across the defect. A coupling capacitor (Ck: 500 pF/100 kV) is used to extract the pulse
voltage. A non-inductive detection impedance (Zm: 50 X) is used to send the pulse
current signal to the digital storage oscilloscope (DSO). The DSO is used to monitor
the PD magnitude. The structure of the gas chamber is shown in Fig. 2, and its volume
is 60 L. A Gas Chromatography-Mass Spectrometer (GC/MS) is used to measure the
sample gas components quantitatively. The GC/MS uses He as the carrier gas and
deploys the special capillary column CP-Sil5CB to separate different components.
398 M. Zhang et al.

R1 R2 Gas chamber
Ds
Ck DSO
CV Cf GC/MS
RV
1 00

80

60

Voltage
40

T1 T2 V V
20

650 7 00 750 800 8 50

Sample Point

Defect model Zm

Fig. 1. Experimental wiring of SF6 decomposition under negative DC-PD.

HV electrode HV electrode HV electrode HV electrode

Φ10 Diameter Φ90 Φ120 Φ120
100mm,
30 thickness
1.5mm
Radius of
10

10
10

curvature
0.3mm

2
20 Φ80
5 Φ80
25

Φ50
20

Copper

50
10 Φ4 Epoxy Gap
2 cuttings
25
10

Aluminium ball Epoxy

Φ120
Ground electrode Ground electrode Ground electrode Ground electrode

(a) Protrusion defect (b) Particle defect (c) Pollution defect (d) Gap defect

Fig. 2. Four kinds of insulation defect models (Unit: mm).

2.2 Insulation Defect Model

As shown in Fig. 2, typical defects in practical GIS include: protrusion defect, which is
usually manifested as the abnormal bulge on the HV conductor; particle defect, which
is generally shown as the metal powder that can move freely in the cavity; pollution
defect, which is formed by various pollution on the surface of the insulator; gap defect,
which is formed by peeling a gap between a HV conductor and disc insulator.
According to the characteristics of these defects, the four defect models are
designed in this study for experimental research, and the specific dimensions are shown
in Fig. 2. All electrodes shown in Fig. 2 are made of stainless steel, and all plate
electrodes have the same size: the thickness and diameter are 10 mm and 120 mm,
respectively. A needle-plate model is used to simulate the protrusion defect (Fig. 2a).
The model of the particle defect consists of a ball electrode (HV electrode), a bowl
electrode (ground electrode), and 20 aluminum balls (Fig. 2b). Some copper cuttings
are adhered on the surface of an epoxy cylinder to simulate the pollution defect
(Fig. 2c). An epoxy cylinder is inserted between the two plate electrodes to create the
gap defect model (Fig. 2d).
The experimental results show that the PD initial voltage of the system without the
defect is 82 kV. After the four insulation defect models are placed in sequence, the
corresponding initial discharge voltage PDIV and the experimental voltage Us were
obtained, where Us = 1.2 PDIV, no breakdown occurred at each experimental voltage.
 Decomposition Characteristics of SF6 and Component Features Extraction 399

In each set of experiments, the average discharge data per second Qsec obtained by the
pulse current method, the PD initial voltages of the system after placing the four
defects, and the corresponding experimental voltages are shown in Table 1.

Table 1. Experimental voltages under four kinds of defects.

Voltage Defect type
Protrusion Particle Pollution Gap
PDIV/kV 31 28 20 49
Us/kV 37.2 33.6 24 58.8
Qsec/pC 7160.4 3381.0 1317.6 523.8

2.3 Experimental Method

The purity of SF6 gas is 99.9995%. The laboratory temperature and relative humidity
are maintained at 20 °C and 50% respectively. The defect model is placed in the gas
chamber, which is cleaned three times with SF6 gas. Then, the chamber is filled with
0.3 MPa SF6. The concentration of H2O and O2 in the chamber satisfy the industrial
standard of DL/T 596-1996. The experimental voltage is raised gradually until the
oscilloscope can detect the PD signal. This voltage is the PD initial voltage (PDIV).
The decomposition experiment of SF6 is conducted for 96 h under the voltage of 1.2
PDIV. The SF6 decomposed components are collected every 12 h. GC/MS is used to
measure the component concentration. All the experiments are conducted under the
same conditions to ensure comparability.

Unit ppm

400

300

200

100

SOF2 Protrusion
SO 2F2 Particle
SO 2
Pollution
CO2
CF4 Gap

Fig. 3. Concentration of SF6 decomposed components at 96 h

3 Experiment Results

3.1 Concentration of SF6 Decomposed Components

Concentration of SF6 decomposed components at 96 h is shown in Fig. 3. Experi-
mental results show that the DC partial discharges caused by the four defects
decompose SF6 gas and generate five stable decomposed components, namely, CF4,
CO2, SO2F2, SOF2, and SO2. The relationship between the total amount of five stable
400 M. Zhang et al.

decomposition components is SOF2 > SO2F2 > CO2 > SO2 > CF4, the content of
SOF2 is much higher than the other four gases, which is the main decomposition
product of SF6 under DC negative PD. The relationship between the total amount of
SF6 decomposition of different defect types is: protrusion defect > particle
defect > pollution defect > gap defect.

2.5 Protrusion
20 Protrusion
Particle
Particle
2.0 Pollution

C(CO2) / ppm
Pollution
C(CF4) / ppm

15 Gap
Gap
1.5 0.3
0.2 10
1.0 0.1
0.0 5
0.5 0 24 48 72 96

0.0 0
0 12 24 36 48 60 72 84 96 0 12 24 36 48 60 72 84 96
t/h t/h
(a) CF4 (b) CO2

Fig. 4. Concentration of CF4, CO2 with discharge time under four kinds of defect

In order to analyze and compare the decomposition characteristics of SF6 under

different insulation defects, the variation curves of the decomposition components of
SF6 under four kinds of defects are shown in Figs. 4 and 5. As shown in Fig. 4a, the
CF4 concentration presents a ‘linearly saturated’ growth trend with time under the four
kinds of defect. On the amount of CF4 produced, the relationship is: protrusion
defect > particle defect > pollution defect > gap defect. At the beginning of the
experiment, the amount of CF4 produced is very low. As the PD intensity accumulated,
the production rate of CF4 gradually increased, and the amount of production begins to
saturate at 72 h.
CF4 is mainly formed by combining a C atom in metal or epoxy material with an F
atom from SF6. At the end of 96 h, the amount of CF4 produced under the four defects
was relatively low. On one hand, higher energy is required for the release of C atoms,
and they would also combine with O to form CO2. On the other hand, F atoms will react
with metal material to form metal fluorides, these factors restrict the production of CF4.
Among them, the amount of CF4 produced under particle defect is much larger than the
others, reaching 2.37 ppm. This is because of the small aluminum ball jumping up and
down in the bowl electrode during the test to precipitate C atoms, and the F atoms
produced by the decomposition of SF6 will further combine with C to form CF4.
As shown in Fig. 4b, the concentration of CO2 presents an approximately linear
growth trend with time under the gap defect. Under the other three defects, it presents a
‘linearly saturated’ growth trend. At 96 h, the concentration relationship of O2 under
the four defects is: protrusion defect > pollution defect > particle defect > gap defect.
CO2 is generated by the reaction of C atoms and O2. With PD time grows, the O2 in
the gas chamber is continuously consumed, resulting in the continuous decrease in the
rate of the formation of CO2, and finally comes to saturate. Compared with the other
three defects, the PD is extremely unstable under the gap defect, and the overall
discharge repetition rate is very low, resulting in the small amount of the
 Decomposition Characteristics of SF6 and Component Features Extraction 401

decomposition of SF6, so the consumption of O2 is also very small. On the other hand,
the epoxy contains a large amount of C in the gap defect, so the amount of CO2
generated increases with time, and no saturation is observed until 96 h.

6 Protrusion
75 Protrusion 300 Protrusion
Particle Particle
Particle
5 Pollution

C(SO2) / ppm
250
C(SO2F2) / ppm

60 Pollution Pollution

C(SOF2) / ppm
Gap Gap
Gap 4
45 200
6 30 3
4 150 20
30
2
2 100 10
15 0 0 1
0 24 48 72 96 50 0 24 48 72 96
0 0
0 12 24 36 48 60 72 84 96 0 0 12 24 36 48 60 72 84 96
0 12 24 36 48 60 72 84 96 t/h
t/h
t/h
(a) SO2F2 (b) SOF2 (c) SO2

Fig. 5. Concentration of SO2F2, SOF2, SO2 with discharge time under four kinds of defect

Under the four defects, the concentration of SO2F2, SOF2, and SO2 approximately
presents a “linearly saturated” growth trend with time (Fig. 5). The concentration
relationship of SO2F2, SOF2, and SO2 under the four defects is the same, namely,
protrusion defect > particle defect > pollution defect > gap defect. This is because the
S-containing products are all formed by the SFx, H2O, O2 and other impurities under
the action of discharge energy. In each group of experiments, the micro water and
oxygen content is strictly controlled, so the yield of the reaction product will eventually
reach saturation as the consumption of H2O and O2. Among them, SOF2 is mainly
derived from the hydrolysis reaction of SF4, and SO2F2 is mainly formed by the
reaction of SF2 and O2. Since SF2 is more difficult to generate than SF4, SOF2 is
generated more than SO2F2. SO2 is mainly produced by the hydrolysis reaction of
SOF2, so its concentration change law is basically consistent with that of SOF2.

3.2 Concentration Ratios of SF6 Decomposed Components

The ratio method can avoid the influence of volume effect on fault diagnosis results,
which has been widely used in power equipment fault diagnosis. C(SOF2 + SO2)/C
(SO2F2), C(CO2)/C(CF4), and C(SO2F2 + SOF2)/C(CF4 + CO2) are selected as char-
acteristic quantities to recognize the PD types [12, 13]. These three concentration ratios
can describe PD severity, the structure of insulation defects and the deterioration degree
of carbonaceous material (metal and organic insulation). Considering SOF2 is prone to
hydrolyze to produce SO2, C(SOF2 + SO2)/C(SO2F2), Ln(C(CO2)/C(CF4)), and
C(SO2F2 + SOF2 + SO2)/C(CF4 + CO2) are selected.
It can be seen from Fig. 6 that the concentration ratio C(SOF2 + SO2)/C(SO2F2)
under four kinds of defects is saturated at 36 h. The concentration ratio C(SOF2 + SO2)/
C(SO2F2) under the particle defect ranges from 5 to 12, and that of the other three
defects ranges from 2 to 5 while there is no intersection. This is because of the airflow
formed by the jumps of the free metal particles enables the SF2 and SF4 to be in full
contact with the F ions compared to the other three defects. SF2 has stronger reducibility
than SF4, which is easier to combine with F ions, resulting in the less generation of
402 M. Zhang et al.

SO2F2. Therefore, C(SOF2 + SO2)/C(SO2F2) under free particle defects is always larger
than the other three defects. Under the particle defect, pollution defect and gap defect, C
(CO2)/C(CF4) has a little fluctuation. The concentration ratio under the protrusion defect
gradually decreases with time and shows a saturation trend, but it is always larger than
the others. Among them, the concentration ratio under the protrusion defect varies from
100 to 600, while that of the other three defects is less than 100, and the concentration
ratio curves under the four defects do not intersect. This is because the generation
conditions of CF4 are harsh, and the generation amount of CF4 under four defects is very
small. Under the protrusion defect, limited by the amount of O2 and tip ablation pas-
sivation, the formation of CO2 gradually reaches saturation. Among the other three
defects, since the discharge energy is not such high and unstable, the total amount of
CO2 does not change much, so C(CO2)/C(CF4) fluctuates less.

C SOF2+SO2F2+SO2 C CF4+CO2
12 Protrusion 8 Protrusion
25 Protrusion
C SOF2+SO2 /C SO2F2

Particle Particle Particle

/C CF4

10 Pollution Pollution Pollution

6 20
Gap Gap Gap
8
15
2

6 4
10
4
2 5
Ln

2
0 0
0 0 12 24 36 48 60 72 84 96
0 12 24 36 48 60 72 84 96 0 12 24 36 48 60 72 84 96
t/h t/h
t/h
(c)C(SOF2+SO2F2+SO2)
(a) C(SOF2+SO2)/C(SO2F2) (b) C(CO2)/C(CF4)
/C(CF4+CO2)

Fig. 6. Concentration ratios of SF6 characteristic decomposition products

The ratio C(SOF2 + SO2F2 + SO2)/C(CF4 + CO2) under the protrusion defect
gradually increases with the discharge time and reach saturation at 36 h, and it is bigger
than that of the other three defects, within the range of 12–18. It have a small fluc-
tuation under the other three defects, and the range of variation under particle defect is
4–6 while its value of pollution defect and gap defect varies from 0 to 4, and there is an
intersection between the latter two defects. This is because the formation of the carbon-
containing compounds (CF4 + CO2) requires more energy than sulfur-containing
compounds (SOF2 + SO2F2 + SO2). With the consumption of H2O and O2, the for-
mation of sulfur-containing compounds is gradually saturated, and the concentration
ratio also tends to be stable. Under the gap defect and pollution defect, on the one hand,
there is lower discharge energy, on the other hand, the defect model of the epoxy resin
contains a large amount of C, and the discharge reaction is occurred at the gas-solid
interface, so the formation of sulfur-containing products and carbon-containing prod-
ucts is roughly balanced.
It can be seen from the above that the three sets of concentration ratio curves under
four kinds of defects all vary within the specific interval range, and generally reach
saturation at 36 h. Except the partial crossover of the C(SOF2 + SO2F2 + SO2)/C
(CF4 + CO2) curve under pollution defect and gap defect, the others are easy to dis-
tinguish. Therefore, the three concentration ratios can be identified as feature quantity
of the PD type identification.
 Decomposition Characteristics of SF6 and Component Features Extraction 403

4 Feature Quantity Validity Analysis

Considering the decomposition compound of SF6 and three concentration ratios can be
used as the feature quantity for PD identification, in order to compare the performance
of the two characteristic parameters, the Spectral Clustering (SC) algorithm is used to
classify the original data under these two kinds of feature quantities. Spectral clustering
is a graph-based clustering method. The basic idea is to regard the sample points to be
clustered as the vertices of the weighted undirected graph. The distance relationship
between each sample point is characterized by the weighted edge wij of the graph,
thereby transforming the clustering problem of the data set into the partitioning
problem of the graph. The clustering results obtained by different classification criteria
are quite different. The more common ones are Ratio Cut, Normalized Cut (N-cut),
Min-max Cut, and et al. In this paper, the N-cut method is adopted, which can ensure
that all points in the same sub graph have a high similarity and low similarity of points
in different sub graphs. The steps of the algorithm are as follows:
1. Enter the component concentration and concentration ratio raw data separately,
X = [x1, x2, …, x32]T(xi = [C(CF4), C(SO2), C(CO2), C(SO2F2), C(SO2F2)]), Y =
[y1, y2, …, y32]T(yi = [C(SOF2 + SO2)/C(SO2F2), C(CO2)/C(CF4), C(SO2F2 +
SOF2 + SO2)/C(CF4 + CO2)]), set the number of cluster groups k = 4;
2. Construct a weighted adjacency matrix W representing the sample data and a degree
matrix D (D is a diagonal matrix), where:

X
4
dii ¼ wij ð1Þ
j¼1

3. Perform a Laplacian transformation on the similar matrix W, calculate the unnor-

malized Laplace matrix L as formula (2) and according to the formula Lu ¼ kDu,
the first 4 eigenvectors u1, u2, u3, u4 of the matrix L are calculated, and the
eigenvector space is constructed.

L ¼ D1=2 WD1=2 ¼ I  D1=2 LD1=2 ð2Þ

4. Construct the matrix U = (u1, u2, u3, u4) with the column vectors u1, u2, u3, u4, and
cluster the matrix U by the k-means clustering algorithm to obtain the final four
clusters.
The clustering results are shown in Table 2. The clustering accuracy rate of the SF6
decomposition component concentration as the feature quantity is 71.875% while that
of concentration ratio can reach to 81.25%. Therefore, the concentration ratio with
higher accuracy of cluster analysis is more suitable as the feature quantity.
404 M. Zhang et al.

Table 2. Clustering result under two types of characteristic quantities

Characteristic quantity Sample number Defect type
Protrusion Particle Pollution Gap
/ Initial number 8 8 8 8
Component concentration Correct number 4 6 5 8
Accuracy (%) (4 + 6 + 5 + 8)/32 * 100% = 71.875%
Concentration Ratio Correct number 8 7 3 8
Accuracy (%) (8 + 6 + 8 + 7)/32 * 100% = 81.25%

5 Conclusion
1. The negative DC partial discharges caused by the four defects decompose the SF6
gas and generate five stable decomposed components, namely, CF4, CO2, SO2F2,
SOF2, and SO2. However, there are significant differences in the decomposition
characteristics of SF6 under various defects. Each component exhibits a specific
variation with the extension of discharge time. There is a correlation between SF6
decomposition components and component ratios and PD types, which can be used
for PD identification.
2. Extract and construct three concentration ratios C(SOF2 + SO2)/C(SO2F2),
C(SO2F2 + SOF2 + SO2)/C(CF4 + CO2) and C(CO2)/C(CF4), both of them are
saturated at 36 h, and has a good discrimination, which can be used as the feature
quantity for PD recognition.
3. Clustering the two types of feature quantities under the SC algorithm respectively.
The PD recognition rate under the characteristic component ratio is 81.25%, which
is higher than the 71.875% under the decomposition component. Therefore, the SF6
decomposition component concentration ratio is more suitable as the characteristic
quantity of PD recognition.

Acknowledgments. This work is partially supported by the by national natural science foun-
dation of China (51607127).

References
1. Zhang, X., Yao, Y., Tang, J., et al.: Actual and perspective of proximate of SF6 decomposed
products under partial discharge. High Volt. Eng. 34(4), 664–669 (2008)
2. Chu, F.Y.: SF6 decomposition in gas-insulated equipment. IEEE Trans Electr. Insul. 21(5),
693–725 (1986)
3. Ji, S., Zhong, L., Liu, K., et al.: Research status and development of SF6 decomposition
components analysis under discharge and its application. Proc. CSEE 35(9), 2318–2332
(2015)
4. Tang, J., Yang, D., Zeng, F., Zhang, X.: Research status of SF6 insulation equipment fault
diagnosis method and technology based on decomposed components analysis. Trans. China
Electrotech. Soc. 31(20), 41–54 (2016)
 Decomposition Characteristics of SF6 and Component Features Extraction 405

5. Tang, J., Liang, X., Yao, Q., et al.: Influence of oxygen and moisture on feature
concentration ratios of SF6 decomposition products under partial discharge. Proc. CSEE 32
(31), 78–84 (2012)
6. Liu, Y., Wu, L., Gong, Y.: Investigation on SF6 decomposition products in GIS and affecting
factors. Power Syst. Technol. 33(31), 58–61 (2009)
7. Tang, J., Li, T., Hu, Z., et al.: Analysis of SF6 gaseous decomposition components under two
kinds of PD defects. High Volt. Eng. 35(3), 487–492 (2009)
8. Tang, J., Chen, C., Liu, F., et al.: Detection of constituents from SF6 decomposition under
partial discharge and recognition of insulation defect coding. Power Syst. Technol. 35(1),
110–116 (2011)
9. Qi, B., Li, C., Luo, L., et al.: Experiment on the correlation between partial discharge and gas
decomposition products in GIS. High Volt. Eng. 36(8), 957–963 (2010)
10. Tang, J., Liu, F., Zhang, X., et al.: Partial discharge recognition through an analysis of SF6
decomposition products part 1: decomposition characteristics of SF6 under four different
partial discharges. IEEE Trans. Dielectr. Electr. Insul. 19(1), 29–36 (2012)
11. Tang, J., Liu, F., Meng, Q., et al.: Partial discharge recognition through an analysis of SF6
decomposition products part 2: feature extraction and decision tree-based pattern recogni-
tion. IEEE Trans. Dielectr. Electr. Insul. 19(1), 37–44 (2012)
12. Tang, J., Hu, Y., Yao, Q., et al.: Decomposition characteristics of SF6 under partial discharge
at different gas pressures. High Volt. Eng. 40(8), 2257–2263 (2014)
13. Qiu, Y., Tang, J., Fan, M., et al.: Photoacoustic spectroscopy detection of SOF2
characteristic component engendered in SF6 de-composition under partial discharge. High
Voltage Engineering 39(5), 1163–1169 (2013)
14. Van Brunt, R.J., Herron, J.T.: Plasma chemical model for decomposition of SF6 in a negative
glow corona discharge. Phys. Scripta 53(2), 9–29 (1994)
15. Van Brunt, R.J., Herron, J.T.: Fundamental processes of SF6 decomposition and oxidation in
glow and corona discharges. IEEE Trans. Electr. Insul. 25(1), 75–94 (1990)
 GIS Insulation State Evaluation Based
on Multi-source Information Fusion

Qiang Yao1 , Siying Wu2 , Yulong Miao1 , Ju Tang2 ,

Shiling Zhang1 , and Fuping Zeng2(&)
1
Electric Power Research Institute, Chongqing Power Company,
Chongqing, People’s Republic of China
2
Wuhan University, Wuhan, People’s Republic of China
Fuping.Zeng@whu.edu.com

Abstract. With the wide application of Gas Insulated Switchgear (GIS) in the
field of ultra-high voltage, the identification and evaluation of insulation state in
GIS equipment has been a hot topic in this field, Therefore, how to effectively
integrate the multi-source information representing GIS insulation state and
establish a scientific state evaluation model is a major problem to be solved. In
this paper, the insulation degradation process of typical insulation defect inside
GIS were simulated, and ultra-high frequency (UHF) and ultrasonic information
of defect in different degradation stages were obtained. At the same time, the
genetic algorithm is used to further characterize the PD severity assessment
feature, and the PD assessment index set based on UHF and ultrasonic infor-
mation was constructed. Based on this, the fuzzy hierarchy analysis was used as
the evaluation method framework, the factor analysis weighting method and
fuzzy evaluation function based on mutual information was used to obtain the
preliminary evaluation result. Finally, the preliminary evaluation result was used
to form a complete comprehensive evaluation model. The evaluation scheme
constructed in this paper can effectively complement the multi-source infor-
mation, thereby improving the reliability of the GIS insulation state evaluation
result.

Keywords: Gas insulated switchgear
Ultra high frequency
Ultrasonic
PD

severity assessment

1 Introduction

Gas-insulated switchgear (GIS) uses SF6, a gas with excellent insulation and arc-
extinguishing properties, as the main insulating medium, and seals components such as
circuit breaker, disconnector and transformers in a metal casing. Therefore, GIS has the
advantages of small footprint, less impact from external environment, and excellent
insulation performance. At the same time, the maintenance period of GIS equipment is
prolonged, which has improved its economic benefits to some extent.
The insulation fault of GIS equipment has been paid much attention as the main
factor of its equipment failure. At present, typical insulation defects in GIS are mainly
expressed in the form of partial discharge (PD). With the development of PD detection

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 406–416, 2020.
https://doi.org/10.1007/978-3-030-31676-1_39
 GIS Insulation State Evaluation Based on Multi-source Information Fusion 407

technology, acoustic, optical, and electrical detection methods can effectively capture
the physical and chemical change information caused by the generation and develop-
ment of PD. These multi-source PD information provide the basis for the construction
of GIS insulation state database. At present, the most commonly used PD detection
methods include ultrasonic, ultra-high frequency (UHF), chemical detection method,
etc. These methods can achieve online or offline monitoring without affecting the
operation of GIS equipment. Through the analysis of these PD information, the
identification of PD specific failure modes can be effectively realized [1–6].
As the field application of various condition monitoring means becomes more
mature, it is possible to obtain more diversified GIS internal insulation fault infor-
mation through joint monitoring, and multi-information joint monitoring is also an
established trend in the field of equipment monitoring in the future [7]. PD information
joint monitoring technology provides a rich information foundation for condition
evaluation. How to effectively use the collected multi-source information, mine data
related to the insulation condition of the equipment, establish a mathematical rela-
tionship between the useful information and the insulation condition of the equipment,
and complete the construction of the final evaluation scheme, these are the problems
that must be faced in the field of GIS insulation state evaluation.
This paper introduces the genetic algorithm to further comprehensively evolve the
evaluation index of unified optimal UHF PD source information for four typical
insulation defects, and combines the optimized evaluation index of ultrasonic fre-
quency domain information to ensure the scientific nature of the source information. In
the construction of GIS insulation state evaluation model, the fuzzy membership
function is introduced to shield the ambiguity between the insulation states, and the
factor analysis weighting method based on mutual information is proposed to realize
the weighting of the indicators of PD source information. The evaluation confidence of
each source information is used as the weight of the two types of PD source infor-
mation for the fusion evaluation of multi-source information, which enhances the
objective comprehensiveness of the evaluation model.

2 Experimental Platform and Method

This paper designs a joint experimental platform for PD UHF and ultrasonic infor-
mation detection in GIS, as shown in Fig. 1. The test chamber is filled with 0.4 MPa of
SF6 gas, and artificial defects are placed in the sealed chamber. PD signal acquisition
part is an UHF-ultrasonic data acquisition system based on a combination of various
detection methods. The PD signal is collected and stored by a DPO7104C oscilloscope
(analog bandwidth: 1 GHz, maximum sampling rate: 20 GS/s, memory depth: 48 M).
Main parameters of UHF sensor: bandwidth: 340 MHz–440 MHz, central fre-
quency: 390 MHz, measured band gain: 5.38 dB. Ultrasonic acquisition system [8] is
composed by resonant sensor D9241A (differential output mode, frequency bandwidth:
20–60 kHz, central resonant frequency: 30 kHz, frequency bandwidth: 20–180 kHz,
sensor sensitivity: about 82 dB) and preamplifier (operating bandwidth is 20 kHz–
1.2 MHz, with three-stage manual adjustment gain of 20 dB/40 dB/60 dB).
408 Q. Yao et al.

Rr

Test chamber
T2

T1

C1
UHF sensor
AC ~
C2 Pre-amplifier
Ultrasonic sensor

Artificial defects

Detecting impedance

Fig. 1. PD information joint collection test wiring schematic

Taking the typical metal protrusion defects in the field GIS equipment as an
example, an artificial physical model of insulation defects was designed. The structure
and parameters are shown in Fig. 2.
The defect model is pressurized by the step pressure method. The step pressure
method uses high voltage stress to accelerate the degradation of the defect to simulate
the PD process of insulation defect in GIS. Table 1 shows the voltage level settings of
UHF-ultrasonic signal acquisition during experiment.

Φ4

electrode
40

radius: 0.4mm
10

12

metal protrusion

Fig. 2. Artificial insulation defect models

Table 1. PD test voltage

Defects Applied voltage/kV
Protrusion defect 27.3!28.6!29.3!30.4!31.6!33.5!34.6!36.4!39.4

3 Construction of GIS State Evaluation Index System

3.1 Evaluation Index Based on UHF PD Information
In the construction process of the actual GIS insulation state evaluation model, it is
extremely cumbersome in the field implementation if a set of evaluation index system
is established for each insulation fault. In order to ensure the universality of the index
system and facilitate the construction of evaluation methods suitable for the practical
application of the project, it is necessary to establish a unified evaluation index system.
 GIS Insulation State Evaluation Based on Multi-source Information Fusion 409

The phase resolved partial discharge (PRPD) mode is a commonly used PD signal
analysis method, which analyzes the development process of PD by studying variation
distribution characteristic of discharge times n, discharge quantity q (replaced by pulse
amplitude u in this paper) with the distribution characteristics of power frequency
phase u. The mode can describe and distinguish the original characteristic information
of different PD severity levels, and has a widely effective in GIS condition monitoring
[9, 10]. Table 2 shows the original feature parameters description of the UHF PD in
this mode.

Table 2. The original feature parameters description of the UHF PD

Order Features Description
V1–V6 Sk+m, Sk−m, Skm, Sk+n , Sk−n , Skn Skewness of positive half cycle, negative half
cycle, and whole cycle
V7–V12 Ku+m, Ku−m, Kum, Ku+n , Ku−n , Kun Steepness of positive half cycle, negative half
cycle, and whole cycle
V13–V14 Qm, Qn Amplitude and discharge time ratio between
positive and negative half cycles
V15–V16 CCm, CCn Cross-correlation coefficient of positive and
negative half cycles

A genetic algorithm is used to search for an optimal set of evaluation parameters

[11]. Finally, {V1, V3, V4, V6, V11, V13, V14, V15, V16} are selected as the evaluation
index of UHF PD. It can be seen that the characteristic information characterize the
difference in the positive and negative half cycle discharge plays a leading role.

3.2 Evaluation Index Based on Ultrasonic PD Information

As shown in Table 3, the time domain characteristic parameters and frequency domain
characteristic parameters of the ultrasonic PD signal can represent the original PD
information.

Table 3. The original feature parameters description of the ultrasonic PD

Order Features Formula Order Features Formula
sffiffiffiffiffiffiffiffiffiffi P
V17 Root mean PN
2
V21 Skewness N

3
ðXi XÞ
Xi
square i¼1 i¼1
N ðN1ÞðN2Þsd 3

V18 Variance P
N V22 Maximum power max½Psdðf Þ; f 2 ð1; NÞ
1
2
ðXi  XÞ
N1
i¼1
spectrum
V19 Absolute P
N V23 Median P N
1
jXi j2 Psdðf Þ
integral mean N
i¼1
frequency f ¼1
2
V20 Kurtosis P
N
V24 Average power P
N

4
ðXi XÞ f
Psdðf Þ
i¼1 frequency f ¼1
ðN1Þsd 4 P N
Psdðf Þ
f ¼1
410 Q. Yao et al.

where V17–V21 is time domain characteristic parameters, and V21–V24 is frequency

domain characteristic parameters. These parameters are used as the original feature
quantities of the ultrasonic PD information.
Among the original eight characteristic parameters of four typical defects, the three
characteristic parameters in the frequency domain have obvious advantages for the
evaluation of the severity of the four typical insulation defects. Therefore, frequency
domain characteristics {V22: maximum power spectrum, V23: median frequency, V24:
average power frequency} can be selected as the evaluation index of the ultrasonic
information source.

4 GIS Insulation State Evaluation Model

4.1 GIS Equipment Insulation State Definition

At present, there are few studies on the insulation state evaluation of GIS equipment,
and there is no clear explanation on the detailed evaluation or the subdivision of the
insulation state. In this paper, the insulation state of the equipment is divided into three
states: Normal, Attention and Dangerous. The detailed definitions are as shown in
Table 4. The division of voltage levels in each state is obtained by a fuzzy c-means
clustering algorithm (FCM).

Table 4. GIS insulation state level

States Significance Voltage level/kV
Normal All monitoring data of the equipment are within the 27.3/28.6/29.3
(H1) normal range of the rated, and the long-term trend of the
monitoring data is not abnormal. This situation proves
that the equipment is in good operation condition, and the
operation and maintenance management of the equipment
can be formulated according to the original periodic
maintenance strategy
Attention The equipment status monitoring data is abnormal, and 30.4/31.6/33.5
(H2) there are unknown defects in the equipment. It is
necessary to further judge through long-term monitoring
data. This state needs to increase equipment vigilance,
increase equipment care input, and continuously pay
attention to equipment status
Dangerous The equipment status monitoring data is obviously 34.6/36.4/39.4
(H3) abnormal. There are defects in the equipment that have a
great influence on the insulation status of the equipment.
It is necessary to immediately maintain the equipment
 GIS Insulation State Evaluation Based on Multi-source Information Fusion 411

4.2 GIS Insulation State Evaluation Model

Based on the analysis in Sect. 2, the evaluation index of the insulation state includes
two aspects based on UHF and ultrasound, including 12 evaluation indexes. In the
synthetical evaluation of all indexes, it is necessary to allocate the weight of the two
types of information, the sum of the weight in the entire weight matrix is 1. In the
process of constructing the state evaluation model, the multi-layer evaluation is used as
the model structure, and the decision level information fusion is conducted on the
evaluation results.
The specific implementation process of decision information fusion is as follows:
1. Set the evaluation set X for the evaluation model to: X = {H1: normal; H2:
attention; H3: dangerous}.
2. Based on the multi-source PD information, the SF6 evaluation index set U is
constructed, with 9 UHF PD evaluation indexes and 3 ultrasonic PD evaluation
indexes.
3. Establish a fuzzy evaluation matrix R, which is used to evaluate the membership of
each evaluation object. On the basis of determining the defect type of the evaluation
sample, the corresponding membership degree boundary is selected to construct the
membership function, and then the membership degree matrix R of the evaluation
sample is obtained.
4. Calculate the weight of each evaluation index V = [y1, y2, …, ym], where m is the
number of indicators included in each source information. The factor analysis
method based on mutual information is used to calculate the weight of the evalu-
ation index of each source information.
5. Calculate the evaluation matrix B ¼ V  R ¼ [b1 , b2 ; . . .; bn  The final evaluation
result is the evaluation matrix, which is the probability of the sample belonging to
each state.
6. Decision fusion multi-source PD information. The evaluation confidence of each
source information is used as the weight of the two types of PD source information
for the fusion evaluation of multi-source information.

Evaluation Index Weight Determination. In the mutual information theory, the

mutual information I(x, y) represents the correlation between two random variables
x and y, and in the definition p(x), p(y), p(x, y) respectively represent the probability
density of variables. The analysis of the collected historical fault data can be used as the
theoretical experience of experts to analyze the advantages and disadvantages of these
parameters. Therefore, on the basis of data mining analysis, evaluation indicators can
be weighted. This method of objective analysis is factor analysis based on the mutual
factor theory.
In the process of constructing the method, the information correlation between the
evaluation index vi of evaluation parameter set V and the insulation state level H can be
P
measured by D ¼ jV1 j Iðvi ; HÞ , and the correlation between the evaluation index vi
vi 2V
P
and vj is measured by R ¼ jV1 j Iðvi ; vj Þ . D and R are used as datum quantities for the
vi 2V
evaluation index weight calculation.
412 Q. Yao et al.

The weight of the index is essentially the importance of the index to the final
evaluation result. Therefore, based on the essential meaning of the weight, and draw on
the correlation measure criterion of mutual information, we can mine the relevance
degree between each index and the severity assessment through the existing data
analysis, and overcome the subjectivity of the existing expert valuation method to the
greatest extent, making the final assessment conclusion more accurate.
The operator U3 is designed here to measure the information relevance of all
variables:

yT D yT Ry
max U3 ; U3 ¼  ð1Þ
y k kðk  1Þ

where the index vector y = [y1, y2, ……]T is a measure factor, the significance of the
formula is to weigh the two aspects of information. The optimization condition of the
quadratic function is:
X
s:t: yi ¼ k; yi 2 ½0; 1 ð2Þ
i

where k is the number of evaluation indexes. The y value of the indication vector
corresponding to each evaluation index is obtained, so that the information system
constructed by all evaluation indexes reaches the maximum relevant-minimum
redundancy. y is the weight factor we need.
Based on the index weights analysis method above, the weights of the two PD
source information are obtained by using the existing data, as shown in Table 5.

Table 5. The weights of the two PD source information

PD information Weights
UHF PD 0.1043, 0.066, 0.0926, 0.0755, 0.0958, 0.1304, 0.1258, 0.1332, 0.1764
Ultrasound PD 0.4112, 0.3642, 0.2246

Evaluation Index Membership Analysis. In the process of state evaluation of actual

transmission and transformation equipment, it is difficult to have absolute threshold
division between the defined evaluation states according to the actual operational state
of the equipment. Therefore, in the process of processing, fuzzy theory is introduced,
and the membership degree of each evaluation index is defined through the fuzzy
membership function, as far as possible to avoid the miscalculation of adjacent levels at
the cross boundary.
 
HðlÞ ¼ li;H1 ; li;H2 ; li;H3 ð3Þ

where li,Sm is the membership degree of the evaluation index vi belonging to the state Hm.
In this paper, evaluation indexes of four insulation defects are selected, and the
boundary determination of membership function needs to use the sample data of four
 GIS Insulation State Evaluation Based on Multi-source Information Fusion 413

kinds of defects. Therefore, this section selects the simpler triangle and trapezoidal
function as the fuzzy membership function of the evaluation index, as shown in Fig. 3.

µi , Hm
H1 H2 H3
1

0.5

0
a1 a2 a3
Evaluation index

Fig. 3. Membership function curve

Therefore, for a certain evaluation index vi, the possible degree of belonging to three
insulation states H1, H2, and H3 is shown in formula (4)–(6) respectively:
H1: Normal
8
< 1 Ui \a1
ri1 ¼ aa22U i
a 1 \U i \a2 ð4Þ
: a1
0 Ui [ a2

H2: Attention
8
>
> 0 Ui \a1
< Ui a1 a1 \Ui \a2
a2 a1
ri2 ¼ a3 Ui ð5Þ
>
> a2 \Ui \a3
: a3 a2
0 U i [ a3

H3: Dangerous
8
< 0 Ui \a2
a3 Ui
ri4 ¼ a2 \Ui \a3 ð6Þ
: a3 a2
1 Ui [ a3

where Ui represents the value of the i-th evaluation parameter in the evaluation index
v1-v18 defined in this section, and ri1, ri2, and ri3 respectively indicate the membership
degree of the i-th evaluation index to the three insulation state levels of normal,
attention, and dangerous.
Then, the fuzzy evaluation matrix R can be finally obtained:
2 3
r11 r12 r13
6 r21 r22 r23 7
6 7
R ¼ 6 .. .. .. 7 ð7Þ
4 . . . 5
r91 r92 r93
414 Q. Yao et al.

In order to obtain the membership function of each evaluation index, the key is to
determine the membership degree boundary of each evaluation index, that is, the value
of a1, a2, and a3 in formulas (4)–(6).
Therefore, the fuzzy C-means clustering (FCM) algorithm is introduced here to
mine the historical data of each evaluation index, and obtain the cluster center and
cluster radius of the indexes in the sample system constructed by the existing evalu-
ation indexes. a1, a2, and a3 is constructed by the class center and the cluster radius, as
shown in Table 6.

Table 6. State center values of features

Features a1 a2 a3 Features a1 a2 a3
v1 0.11 0.42 0.87 v7 0.12 0.42 0.77
v2 0.09 0.35 0.66 v8 0.25 0.52 0.89
v3 0.21 0.46 0.69 v9 0.15 0.42 0.7
v4 0.17 0.42 0.66 v10 0.20 0.29 0.61
v5 0.16 0.42 0.82 v11 0.19 0.30 0.46
v6 0.18 0.42 0.79 v12 0.15 0.33 0.44

5 Case Study

The metal protrusion defect was set under the external voltage of 28.2 kV in the
laboratory, and it was preliminarily judged that it was normal. UHF sensors and
ultrasonic sensors were used to collect a certain amount of PD pulse signals, and the
selected evaluation values were calculated. The UHF and ultrasonic data were nor-
malized to obtain the feature quantities: F1 = {0.26, 0.23, 0.18, 0.33, 0.01, 0.25, 0.33,
0.36, 0.28}; F2 = {0.23, 0.27, 0.18}.
According to the previous membership function, the membership degree matrices
of UHF and ultrasonic are obtained as follows:
2 3
0:77 0:23 0
6 0:46 0:54 07
6 7
6 1 0 07
6 7
6 0:52 0:48 07
6 7
R1 ¼ 6
6 1 0 077 ð8Þ
6 0:65 0:35 07
6 7
6 0:30 0:70 07
6 7
4 0:59 0:41 05
0:52 0:48 0
2 3
0:68 0:32 0
R2 ¼ 4 0:27 0:73 05 ð9Þ
0:83 0:17 0
 GIS Insulation State Evaluation Based on Multi-source Information Fusion 415

Based on the weights of the indicators shown in Table 5, the evaluation matrices
for UHF and ultrasound are calculated as:

B1 ¼ V1  R1 ¼ [ 0:63 0:37 0 ð10Þ

B2 ¼ V2  R2 ¼ [ 0:56 0:44 0 ð11Þ

The confidence level of each information source is based on the average evaluation
accuracy of the two types of PD source information in the evaluation of the PD
severity. Historical data indicates that the accuracy of UHF and ultrasound information
is 0.92 and 0.90, respectively, and the reliability of the two source information is
calculated to be 0.51 and 0.49, respectively. The credibility is used as the weight of the
two types of evaluation indicators for multi-source information fusion evaluation. The
comprehensive evaluation matrix after integration is:

B ¼ V  R ¼ [ 0:60 0:40 0 ð12Þ

It is judged to be in a normal state and conforms to the actual situation.

6 Conclusions

In the weight determination of the evaluation indexes, a weighting method based on

mutual information-based factor analysis is constructed. In the process of finally
integrating the evaluation results of multi-source information, the evaluation results of
the two types of PD source information are fused to realize the output of the evaluation
results of the insulation state.

Acknowledgments. This work is supported by the national natural science foundation of China
(51877157 and 51607127).

References
1. Yao, Y., Yue, Y., Huang, X.: Field application of GIS UHF/ultrasonic PD detection method.
High Volt. Eng. 34(2), 422–424 (2008)
2. Tang, J., Liu, F.: Partial discharge recognition through an analysis of SF6 decomposition
products part 2: feature extraction and decision tree-based pattern recognition. IEEE Trans.
Dielectr. Electr. Insul. 19(1), 37–44 (2012)
3. Suresh, S.D.R., Usa, S.: Cluster classification of partial discharges in oil-impregnated paper
insulation. Adv. Electr. Comput. Eng. 10(1), 90–93 (2010)
4. Chang, W.Y., Yang, H.T.: Application of fuzzy C-means clustering approach to partial
discharge pattern recognition of cast-resin current transformers. In: International Conference
on Properties & Applications of Dielectric Materials, pp. 372–375. IEEE (2007)
5. Swedan, A., El-Hag, A.H., Assaleh, K.: Acoustic detection of partial discharge using signal
processing and pattern recognition techniques. Insight 54(12), 667–672 (2012)
416 Q. Yao et al.

6. Karthikeyan, B., Gopal, S., Vimala, M.: Conception of complex probabilistic neural network
system for classification of partial discharge patterns using multifarious inputs. Expert Syst.
Appl. 29(4), 953–963 (2005)
7. Bernardo, J.T.: A methodology for hard/soft information fusion in the condition monitoring
of aircraft. In: Conference on Multisensor, Multisource Information Fusion - Architectures,
Algorithms, and Applications, vol. 8756. SPIE, Baltimore (2013)
8. Dong, Y.: Research on feature extraction and severity assessment of partial discharge in GIS.
M.S. thesis. Chongqing University, Chongqing (2015)
9. Mas’Ud, A.A., Stewart, B.G., Mcmeekin, S.G.: An investigative study into the sensitivity of
different partial discharge j-q-n pattern resolution sizes on statistical neural network pattern
classification. Measurement 92, 497–507 (2016)
10. Gao, W., Ding, D., Liu, W.: Research on the typical partial discharge using the UHF
detection method for GIS. IEEE Trans. Power Deliv. 26(4), 2621–2629 (2011)
11. Zhao, L.: Improved genetic simulated annealing algorithm based on BP neural networks and
application in recognition of GIS partial discharge. M.S. thesis. South China University of
Technology, Guangdong (2013)
 Operational Aged Switchgear with the Age
up to 50 Years – Investigations, Testing,
Results – Considerations for Design
and Operation of New Switchgear

Thomas Gräf(&)

Berlin University of Applied Science for Engineering and Economics,

Hochschule für Technik und Wirtschaft Berlin, Wilhelminenhofstr. 75A,
12459 Berlin, Germany
thomas.graef@htw-berlin.de

Abstract. The service life of electrical switchgear is designed for a service life
of 30 years. Meanwhile, the average age of switchgear is significantly higher
than 40 years. The main causes for this are a low susceptibility to failure as well
as the compulsion to reduce costs and optimize their use. Due to the high age of
the electrical infrastructure, the question then arises as to whether and how aging
effects can be determined after the end of the service life of switchgear that have
been operationally aged.

1 Introduction

Meanwhile, switchgear systems are often operated for more than 40 years. Although
the control technology will be renewed during this time, the primary technology will
remain in operation mostly unchanged [3]. Within this long service life, aging effects
must be expected. There are no individual physical loads, but combinations of tem-
perature changes with mechanical loads and radiation are conceivable. It is therefore
only logical that aging effects on switchgear must have taken place. As a rule, old
switchgear are no longer subject to testing, but scrapped. In particular, it should also be
noted that during the previous period of operation no tests corresponding to the type
tests have taken place. Therefore, the following investigations have found particular
interest in various operators who provided their old switchgear for investigations. In
order to be able to answer the question about ageing effects, operational aged
switchgear were investigated in cooperation with the IPH Institut “Prüffeld für elek-
trische Hoch-leistungsprüfungen” GmbH – CESI group.

2 Operational Aged Switchgear

Investigation results on the aging of operational aged electrical equipment can be found
in some sources, which are mostly limited to equipment with rated voltages 110 kV
[1, 2]. The effects caused by aging were often documented by the evaluation of asset

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 417–428, 2020.
https://doi.org/10.1007/978-3-030-31676-1_40
418 T. Gräf

management maintenance and maintenance records and assigned to specific assets [1,
2, 4]. Extensive scientific literature can be found to physical aging effects on individual
(model) arrangements. It should be noted, however, that the “conversion” of artificial
aging under more severe conditions to real aging is generally not possible [6, 7].
Therefore, the idea was developed to investigate operational aged switchgear in detail,
since here a multiple physical aging process took place under real environmental
conditions. Table 1 describes the investigated switchgear.

Table 1. Examined medium voltage switchgear

1. switchgear 2. switchgear
Manufacturer ESK Hoppe Berlin VEB Stark-strom
Anlagenbau
Type MSK31 CSI 10/250
Erfurt
Year of manufacture 1992 1967
Degree of protection IP 40 IP 30
Operating voltage 3/E/12 kV 3/10 kV
Rated peak withstand current 40 kA 38 kA
Rated short-time withstand 16 kA 1 s 15 kA 1 s
current
Rated short-duration power 28 kV 28 kV/45 kV
frequency withstand voltage
Rated impulse withstand voltage 75 kV 60 kV/85 kV
Standards DIN VDE 0101/DIN VDE 0670 TGL 26055 [11],
Teil 6 PEHLA-guideline 4 TGL 26469 [12]

The 1st switchgear was used within the supply infrastructure of a Berlin hospital.
The 2nd switchgear had been in operation since 1967 and last used in a waste
disposal company. The starting points for the investigations were questions such as:
• Which aging effects can be detected by simple optical inspection?
• Are the properties changing so that future operational safety could be endangered?
• Can aged, existing switchgear units meet newer requirements through taking retrofit
measures?
• Are there gaps in current standards that need to be closed by the lessons learned?
[10].

3 Investigations

3.1 Dismantling – First Optical Analyzes

The switchgear have been prior dismantling already optically examined by the HTW
stuff. For this purpose, the resistance values of the main current paths as well as the
 Operational Aged Switchgear with the Age up to 50 Years 419

torques of the screw connections were measured before disassembly. The transport and
reconstruction of the switchgear in the test institute were carried out in such a way that
only a few interventions on the aged switchgear were necessary to ensure the lowest
possible changes in the switchgear. Already the resistance measurement of the main
current paths showed during the dismantling already clear deviations per phase. This
was due to different torques of the screw connections in the current paths. It had been
determined screw points, which were only designated as “hand-tight”.

3.2 Optical Inspections at the Operational Aged Switchgear

in the Testing Institute
3.2.1 Optical Inspections at Switchgear MSK 31
At switchgear 1 concentrated dust deposits were found at the busbar bushings, although
the rest of the switchgear had no appreciable dust deposits. When removing the bus-
bars, copper patina was associated with the marking of the bushing (Fig. 1). The later
dielectric investigations revealed traces of partial discharges both on the busbar and on
the insulating body of the bushing (Fig. 2).

Fig. 1. Switchgear 1 – busbar with dust deposits and copper patina

3.2.2 Optical Inspection at Switchgear CSI

The screwing of the back panels was partially attacked by corrosion that washers rusted
away. Loose screws could be found on the drive rods on both switch-fuse combina-
tions. This is noteworthy because the feed-in and metering fields were maintained by
the grid operator and the outgoing feeder by the plant operator. On the doors of the
switchgear were signed stickers of the companies, which confirmed the timely
maintenance.
When replacing the switch-fuse combination in switchgear 2 in 1979, the spring
discs were installed the wrong way round, which led to a displacement of aluminium at
the busbar holes. In addition, copper and aluminium busbars were bolted directly
together. The ceramic supports showed massive dust deposits in the area of the busbar
as well as approaches to the formation of creeping marks.
420 T. Gräf

Fig. 2. View inside the bushing of the busbar with partial discharge traces

3.3 Dielectric Investigations

3.3.1 Determination of the Impulse Withstand Voltage at Switchgear
MSK 31
Switchgear 1 showed flashovers at a value of −74 kV at phase A and 68 kV at phase C
in field 1. The flashovers took place in the vicinity of the bushings. The reason for the
failure was the design, since the square busbars in conjunction with the bolts of the
busbar connection in the immediate vicinity of the bushings and the edges of the plates
in the area of the bushings were an unfavourable constellation for the electric field.
Further flashovers revealed weak spots caused by sharp edges of copper bars near the
circuit breaker and a door-ground lead laid too close to a phase conductor (Fig. 3).

3.3.2 Determination of the Impulse Withstand Voltage at Switchgear CSI

Switchgear 2 passed the withstand voltage test at 85 kV [11] across the isolators and
60 kV for the insulation between conductors, conductors and grounded housing. The
required impulse withstand voltage values according to DIN EN 62271-200 with
±75 kV were also maintained despite contamination. Decisive for achieving the
required impulse withstand voltage values were the large distances between the elec-
trode geometries.

Fig. 3. Switchgear 1 – flash-over at earthing strip of the door at −74 kV

 Operational Aged Switchgear with the Age up to 50 Years 421

3.3.3 Determination of the Short Time Alternating Voltage

Both switchgear systems were able to meet the required withstand voltage according to
the currently valid DIN EN 62271-200 [8].

3.3.4 Measurement of Partial Discharges

In the MSK31 switchgear, the bushings and the fused switch-disconnectors were
identified as partial discharge sources. The rectangular busbars had a slight air gap in
the bushings, initiated here partial discharges. The effect of the partial discharges,
which obviously has been occurred for some time, can be seen in Fig. 1 on the basis of
the discoloration of the copper and in the bushing through a white coating in Fig. 2.
The inception voltage was 15,6 kV, the exception voltage at 13,2 kV with apparent
charges of up to 4 nC measured in the ring feeder 1. The measured apparent charges
decreased significantly after all fused-switch-disconnectors were opened. The CSI
switchgear also showed partial discharges from an inception voltage of 10 kV without
connected current transformers. The inception voltage was determined to be 9,2 kV.
Apparent charges of up to 750 pF occurred in field 4 (transfer field) at phase C. One
reason for this was the strong surface contamination of the insulators (Fig. 4). Although
cleaning the insulators significantly reduced the apparent charge level, it did not reduce
the level of pd inception voltage. The post insulators were obviously not cleaned, as
they were accessible only by dismantling the switchgear. Pinched aluminum painted
busbars and sharp edges on the busbars were located as further partial discharge
sources.
The partial discharge measurements on both switchgear showed that undisturbed
operating voltage of Un/√3 = 10 kV/√3 did not require the occurrence of partial dis-
charges. Although the partial discharge inception voltage of the current transformer of
the 2nd switchgear with 6,8 kV was not significantly above the operating voltage
Un = 10 kV/√3 of the grid.

3.4 Short Time Current and Impulse Current, Making and Breaking
Current
In addition to the dielectric tests, tests have been carried out with short-time current and
short-circuit making and breaking tests with corresponding test sequences. Separate
tests of the switch-disconnector and fuse switch-disconnector were initially dispensed
with. The test currents were successively increased from 30% to 100% of the rated short-
circuit current, since the existing mechanical strength and thermal properties were not
known due to the aging effects that had occurred. The maximum short-circuit current
was the possible short-circuit current specified at the grid connection point by the grid
operator. It was important to note that the short circuit current values specified by the
grid operator at the connection point of the switchgear were changed over the decades to
larger values. While the thermally equivalent short-circuit current with 14 kA 1 s was
given up to 1990, this value increased from 1990 to 16 kA 1 s. A further increase was
specified starting in 2015 with a short-circuit current of 20 kA 1 s.
422 T. Gräf

Fig. 4. Polluted post insulator of the busbar

3.4.1 Short Time Withstand Current and Peak Withstand Current Tests
at the CSI Switchgear
The short time current tests carried out were increased in steps starting at 5 kA, 10 kA,
12,5 kA and a current flow time of 0,1 s up to the maximum rated short-circuit current
of the switchgear of 15 kA at 1 s. The SCI1-10/630/250-MAA oil-filled circuit-breaker
from the year of construction 1967 successfully passed all making and breaking tests in
accordance with DIN EN 62271-100 [9]. However, the circuit breaker blew off the oil
surplus by deleting the switching arc and the associated expansion of the oil in the
switching chamber and thus contaminated the switchgear. The test with a short circuit
current of Ik3p = 10,4 kA, 0,1 s current flow duration with an impulse current
amplitude of ip3pL3 = 20,4 kA led to the melting of the lower contact of the phase B
of the withdrawable unit. Figure 5 shows the molten contact of phase B. The cause was
a lack of contact force. This short-circuit current was significantly smaller than the
16 kA short-circuit current defined at the grid connection point and, with a current flow
duration of 0,1 s and was on the order of magnitude of the time of a protective
shutdown. If this damage had now occurred in the course of regular operation of the
switchgear, the contact resistance of the current path would have increased consider-
ably. With the flow of the operating current would have to be expected with a sig-
nificant increase in temperature at the contact point. How then could a determination of
the temperature rise at the contact take place? When would you notice the physical
change of the contact? What would have happened if a short-circuit current in the order
of magnitude of the rated data of the switchgear had occurred in reality?
Due to the damage occurring at the contact of the phase B of the withdrawable unit,
it was necessary to change the phase to be loaded at the highest. A further short-circuit
current test on phase C resulted in a maximum short-circuit current load of Ik3p =
14,8 kA at a current flow time of 1 s (nominal data of the switchgear) and a maximum
impulse current of ip3p = 39,4 kA destroying the upper contact of phase C at the
withdrawable unit. This was due to the high lateral tolerance in the guide rails, so that
the withdrawable unit could move transversely, as well as the non-functioning
mechanical locking of the bent locking rail.
 Operational Aged Switchgear with the Age up to 50 Years 423

Fig. 5. Melted phase B contact, short circuit current Ik3p 10 kA, 0,1 s, ip3p = 20,4 kA

3.4.2 Short-Time Withstand Current and Peak Withstand Current Tests

at the MSK31 Switchgear
This substation was also gradually loaded with increasing short-circuit currents. The
short-circuit current and the short-circuit duration were gradually increased starting
from 50% (Ik3p = 8 kA) and 0,1 s current flow time over 12 kA, 0,1 s and 12 kA, 1 s
up to the rated value of the switchgear in three stages.
Here, too, damages were already apparent before the rated values of the switchgear
were reached. As a result of the current forces, the busbar of phase B was shifted in the
passage (Fig. 6). It formed a small air gap between busbar and bushing.
At the circuit-breaker, the insulating body already broke at a short-circuit current
level of 16 kA (rated value) with a short-circuit duration of only 0,1 s, which was well
below the rated value of the current-carrying capacity of the switchgear (Fig. 7) [9].

Fig. 6. Shifted busbar phase B, Ik3p = 16.2 kA, 1 s, ip3pB2 = 23.9 kA

424 T. Gräf

Fig. 7. Broken insulation at the circuit breaker phase B, Ik3p = 16.2 kA, 1 s, ip3pB = 23.9 kA

3.5 Internal Arcing

The investigations into the behaviour of internal faults (arc fault test) were carried out
in favour of the possibility to make video recordings without indicators. Nevertheless,
none of the switchgear passed this test. In both switchgear doors or flaps opened. This
is particularly noteworthy since the MSK 31 switchgear was tested in accordance with
PHELA guideline no. 4 and accordingly classified as spurious-arc-proof. The opening
of the dampers and doors, however, is not due to aging effects, but the design and
execution of the locking of the doors (CSI) or the use of plastic locks (MSK 31) on the
circuit-breaker doors are the cause.
Figure 8 shows a picture take from the high speed video while the internal arcing
happens. The arcing was initiated at the upper contacts of the load switch. At the time
of 100 ms the door of the circuit breaker control opens due to the high pressure caused
by the internal arcing. Figure 9 shows the travelling of the arc inside the switchgear.
The initiation of the arcing was below the circuit breaker, but the arc sprang from the
end of the cable contacts backwards to the busbar and moved forward to the end of the
busbar til the field unit 4 (right side in Fig. 9).

Fig. 8. Arcing at t = 100 ms, opening of the circuit breaker control door, switchgear 1
 Operational Aged Switchgear with the Age up to 50 Years 425

In addition, it must generally be taken into account that a test with regard to internal
faults in switchgear has no statistical significance, since in the minimum only one test
may be sufficient to pass the type test requirements. Another aspect is the short-circuit
current carrying capacity required by the grid operator with 20 kA since 2015. The tests
carried out show that both switchgear systems could not withstand the loads with rated
data. In the end, would it not then be necessary to replace the aged inventory assets?

Fig. 9. Arcing at t = 100 ms, opening of flaps and travelling arc along busbar, switchgear 2

4 Evaluation of the Results

4.1 Effect of Maintenance and Servicing

All tested switchgear have received proper maintenance and service either from service
providers or from the grid operator. Deficiencies that were already present by the
design or emerged during the commissioning of the switchgear were also not recog-
nized or eliminated by regular maintenance.
In the investigated systems, the very different operating concepts and the switching
interlocks could only be recognized with some consideration. The forces to be applied
for switching operations were also very different. The CSI switchgear obviously had
such high forces applied at earlier times that the guide rail was bent and the mechanical
lock for the drawer stopped working. The electrical interlock, on the other hand, still
worked and released the circuit breaker. This results in the normative requirement to
provide a specification for the testing of mechanical interlocks and to verify that they
are working properly at cyclic intervals. Likewise, a visible instruction for the handling
of the switching and locking should be attached to the switchgear. For service providers
in the area of maintenance and servicing, who work with changing personnel, there is a
risk of mistreatment in handling a wide variety of operating concepts due to a lack of
experience and handling of various switchgear combinations.
426 T. Gräf

4.2 Addition of Standards

The current standards do not take into account the meanwhile high and increasing
service life of switchgear over a period of more than 30 years [5]. In particular, a
change in use or a conversion of the switchgear is to be expected over such periods,
especially in the industrial network area, which is considered to be customary and
therefore normatively not currently considered.
According to the current understanding of standards, type tests would no longer be
valid after a change to a tested switchgear, which would result in far-reaching legal
consequences and liability issues in the event of damages. This raises the question of
whether existing standards meet real-world situations and usage requirements.
A hazard analysis by the plant operator would have to take into account the fact that
the short-circuit currents change over time and, in view of the aging plastic insulating
components and the insufficient mechanical strength that would no longer be expected,
lead to replacement of the switchgear. A discussion about the costs would be pre-
programmed. In addition, the grid operator would be obliged to automatically inform
plant operators about changed physical limits.

4.3 Retrofit of Switchgear

The term retrofit is often understood in connection with switchgear as replacement of
the circuit breaker on the basis of no longer available spare parts [4]. The switching
performance tests with the service-aged circuit-breakers and the rated data of the
switchgear alone would not justify the replacement as part of a retrofit. Switching times
and capacity corresponded to the original rated data of the circuit breaker. Much more
meaningful would be the increasing short-circuit currents coming from the network.
Discussions currently exist regarding the risk of electric arcs in the operation of
electrical systems. The hazard analysis forces us here to act with regard to the
occurrence of toxic gases, bang, flash of light, etc., and an imminent loss of production
with high consequential costs. These hazards are currently not considered in the
electrotechnical standards. Currently, standardization committees are engaged in
activities for the introduction of standards for the equipment of switchgear with arc-
detecting and rapidly switching-off facilities. With the present test results and the
determined mechanical strengths of operational-aged switchgear, the meaningfulness of
the retrofitting or retrofitting with quick-disconnecting equipment must be negated.
Rather, the problem arises that broken plastic insulation material could lead to the
emergence of partial discharges with corresponding late effects, or simply the broken
plastic insulation material, due to the complex structure of the switchgear, are not
recognized. As a rule, after a retrofit measure, no short-term current tests are carried out
on existing systems that prove the mechanical and thermal stability of the systems.

5 Summary

The dielectric investigations showed that the required impulse withstand voltages were
not reached at the time of switchgear construction, during initial commissioning and
due to design conditions. Mounting shortages during the construction of the switchgear
 Operational Aged Switchgear with the Age up to 50 Years 427

on site are another cause. The occurrence of partial discharges occurred in the aged
switchgear from a voltage level above the nominal network voltage.
In the 50-year-old CSI switchgear, as far as the current transformers are disre-
garded, due to the materials used, the partial discharges cannot be classified as harmful
and do not lead to a degradation associated with the risk of a subsequent system failure.
In contrast, this is not the case with the much younger MSK 31 switchgear through the
use of plastic insulation material.
The short-term current tests caused damage to all switchgear. Particularly note-
worthy is the fact that these damages occurred at short-circuit current values lower than
the nominal short-circuit current defined at the grid connection point of the switchgear.
While the recent MSK 31 switchgear suffered damage due to mechanical overloading
of plastic insulating components due to aging and design defects, the damage to the old
CSI switchgear was due to design, poor service and aged movable contacts. The
damages that have occurred here would have been avoidable by attentive and complete
maintenance and inspection.
An aging of the circuit breakers in terms of switching capacity was not detectable.
In essence, this was due to the low number of operations during the life of the
switchgear and their use. The regular maintenance and servicing carried out according
to legal regulations does not lead to a reduction of defects or identified shortcomings.
Particularly in the case of retrofit measures on existing systems, it must be taken
into account that many tests that can be carried out in the laboratory at the installation
site of the switchgear can no longer be realized, especially testing for heating, impulse
voltage or mechanical strength under current loads. Thus, after a retrofit the state of the
switchgear is not clearly defined.

Acknowledgments. The IPH Institut “Prüffeld für elektrische Hochleistungsprüfungen” GmbH

Berlin – member of the CESI group - should be expressly thanked for the support and the
facilitation of the examinations of aged switchgear.

References
1. Balzer, G., Schorn, C.: Asset Management für Infrastrukturanlagen. Springer Vieweg, Berlin
(2014)
2. Smith, P., Balzer, G., et al.: Ageing of the system - impact on planning. Working Group 37-
27, Cigre report 176, December 2000
3. Windmöller, R.: Die wirtschaftliche Bedeutung der Versorgungsqualität. EW 97, H. 25, S.
17 (1998)
4. Asset-Management von Verteilnetzen. Technischer Bericht 299. FHG 2006 (2006)
5. DIN EN 62271-1 (VDE 0671-1) 2009-08, Hoch-spannungs-Schaltgeräte und –schaltanlagen
– Teil 1: Gemeinsame Bestimmungen
6. Küchler, A.: Hochspannungstechnik. Springer-Verlag, Heidelberg, 3. neu bearb. Aufl.
(2009)
7. Kaltenborn, U.: Die Eignung von Diagnoseverfahren zur Beurteilung der Frühphase der
Oberflächenalterung von Epoxidharz-Formstoffen bei simultaner elektrischer und klimatis-
cher Beanspruchung. Diss. D. 17, Darmstadt (2005)
428 T. Gräf

8. DIN EN 62271-200 (VDE 0671-200) 2012-08, Hochspannungs-Schaltgeräte und –

schaltanlagen – Teil 200: Metallgekapselte Wechselstrom-Schaltanlagen für Bemessungss-
pannungen über 1 kV bis einschließlich 52 kV
9. DIN EN 62271-100 (VDE 0671-100) 2009-08, Hochspannungs-Schaltgeräte und –
schaltanlagen – Teil 100: Wechselstrom-Leistungsschalter
10. Gräf, T.: Investigations at operational aged switchgear with the age up to 50 years. In: 25th
International Conference on Electricity Distribution, Cired, Madrid, 3–6 June 2019 (2019)
11. TGL-Norm 26469/03 Fabrikfertige Baueinheiten für Nennspannungen über 1 kV bis
72,5 kV
12. TGL 26055 Schaltanlagen CSIM12 für Wechselspannung
 Insulator Diagnostics Through
a Dielectric E-Field Sensor

A. Villa(B) , L. Barbieri, and R. Malgesini

RSE, 20134 Milan, Italy

andrea.villa@rse-web.it

Abstract. The on line diagnostic of high voltage insulators, used in

overhead lines, has proved a powerful tool to detect and prevent the
failure of such components. Among the many diagnostic techniques, E-
field sensors have been extensively used to assess the presence of surface
or internal defects. Electric field sensors are mostly based on capacitive
metallic gauges, which induce a significant electric field distortion and
may pose a threat to live line workers who perform the measurement.
In this work, we introduce a new class of electric field sensors that are
fully dielectric. The sensor is capable of measuring the modulus and
the direction of the electric field in a plane with a high rejection of the
components orthogonal to that plane in a very broad frequency interval.

Keywords: Insulator · E-field sensor

1 Introduction
In this work, we show some applications of an optical E-field sensor for the
diagnostics of high voltage line insulators and, more in general, for the diagnostics
of insulating materials.
As outlined in [1], composite insulators have represented a significant step
forward as they are lighter and they are more resistant in polluted environ-
ments. Unfortunately, premature failures of these components are more difficult
to be detected as they can develop surface and internal defects. Many diagnostic
systems have been developed, among them two main classes have emerged i.e.
optical instruments and E-field gauges. The first class takes into account both
infrared [2] and ultra-violet sensors [3] which are capable of detecting the heat
and the ionization radiation produced by surface discharges. Some devices are
also capable of combining these two information using a common optical unit
[4].
In spite of their versatility, optical sensors cannot detect internal defects
since internal discharges do not produce an external, significant, radiation. Only
if a major thermal effect is observed the fault can be isolated. On the con-
trary, though they are less versatile, E-field sensors [1,5] can also detect internal
defects. In fact, these kinds of sensors have to be placed near the insulator to be
monitored and they are usually handled by live line workers (LLWs).
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 429–436, 2020.
https://doi.org/10.1007/978-3-030-31676-1_41
430 A. Villa et al.

Most of the E-field gauges are based on capacitive effects and this means they
contain a large amount of metallic parts. This fact poses further difficulties to
LLWs since each metallic object must be treated with care and this aspect has
also some consequences on the quality of the measurement. In fact, large metallic
objects can distort significantly the electric field they are designed to measure:
in many cases the sensor has dimensions similar to those of the insulator sheds.
Therefore, a fully dielectric gauge provides a significant step forward in this
field, both from a diagnostic and safety viewpoint. Moreover, we aim to develop
a new measurement unit, which is capable of extracting as much information as
possible. Therefore, the sensor we are going to describe is capable of measuring
the modulus and direction of the electric field on a plane. This plane will be
properly placed to measure the component of the electric field aligned with the
axis of the insulator (axial component) and the radial component. Moreover, the
bandwidth of the unit is particularly broad as frequencies ranging from a few
Hz to tens of MHz can be acquired accurately. This makes it possible to identify
defects, that may be present, and to characterize the discharges that are taking
places. This provides some information regarding the severity of the fault.
The measurement system is based on optical technology and has already been
described in [6] therefore we will sum up here the relevant outline of the sensor
and provide some test cases on relevant electric components in a laboratory
environment.

2 An Outline of the Sensor

The measurement system is made of two parts: a measurement head, see

Fig. 1(a), and an electronics unit Fig. 1(b). The head is fully dielectric, mostly
made of polyethylene and some optical components. Its dimensions are very
small, it is 10 cm long, it has a diameter of less than 5 cm and its weight is
a few hectograms. The head holds an electro-optical crystal that changes its
characteristics depending on the direction and magnitude of an external applied
electric field. A laser beam is used to probe the crystal so that the electric field
can be deduced from the polarization state of the light exiting the crystal. To
be more precise, the laser beam is split into four beams by a set of Wollaston
prisms. As a consequence, the intensity of the four beams can be linked to the
characteristics of the external electric field. The laser source, in the electronics
unit, is connected, by a polarization maintaining fibre, to the head and the four
beams coming out from the sensor are brought to the electronics unit by four
identical mono-mode fibres. In other words, the head is connected to the data
acquisition device only by a bundle of fibres, which can be very long according
to the specific application.
The electronics unit contains the laser source and the analogic conversion
unit which converts the light fluctuations to properly amplified electric signals
that can be digitalized using some standard commercial units. The sensor is
capable of measuring the magnitude of the electric field and its direction but,
using this technique, it is not possible to discriminate its sense. Currently a first
 Dielectric E-Field 431

Fig. 1. The sensor head in a capacitor for testing purposes. The head is attached to a
plastic rod (a). The electronics unit containing the four analogic signal treatment units
and the laser source (b).

prototype has been assembled and tested as reported in [6], where are described
all the calibration and verification measurements that have been carried out. A
second, enhanced and more compact, prototype is, currently, under construction.

3 Some Test Cases

3.1 Streamer and Corona in a Capacitor

This test case has been performed with a relatively simple geometry to bet-
ter understand how to interpret the signals produced in regions where some
discharges are present. In particular, we have considered a capacitor with two
plates at a distance of 30 cm, the sensor head has been placed just in the middle
of the two plates. Then we have considered a thin straight wire, which protrudes
12 cm from the upper plate arriving just above the head. The wire has a diame-
ter of 1 mm, so it can easily generate electrical discharges such as corona effect
and treeing. The upper plate has been energized with an alternated waveform at
50 Hz with three different peak voltages at 11.35 kV, 13.6 kV and 15.9 kV. The
lower plate has been grounded. The electric fields measured by the head of the
sensor have been depicted in Fig. 2(a), while in Fig. 2(b) we have included just
the first period of the AC source. The first peak corresponds to the positive peak
while the second one to the negative one. At low voltage, i.e at 11.35 kV, the
inceptions of the discharges are not stable. Not all the semi-periods correspond
to a discharge event. On the contrary, higher voltage levels are associated to
more stable inceptions.
Now taking into account Fig. 2(b) we undeline that the electric field grows
till nearly 3 ms where a sudden change of the measured field takes place. This
corresponds to the development of a positive streamer originating from the tip
432 A. Villa et al.

of the wire and travelling in the direction of the head. As the streamer phase ter-
minates, the charges produced by the discharge are drifted towards the ground
plate and are captured by the surface of the head. This generates a rise of the
measured electric field characterized by a rate which is higher than the elec-
trostatic one observed in the interval 0–3 ms. The electric field keeps rising till
reaching its peak just after 5 ms and then it diminishes becoming null at 10 ms.
At this point the polarity changes and the measured electric field modulus rises
again till 13 ms where a new discharge phase starts. This phase, contrary to
the positive discharge, is characterized by a diffused glow discharge and we can
mainly observe the deposition of electrons on the surface of the sensor which
generates an enhancement of the electric field. This is due to the fact that nega-
tive streamers usually incept at higher voltages than those of positive ones. The
electric field peak is reached just after 15 ms and then it vanishes again at 20 ms.

Fig. 2. The evolution of the electric field measured by the sensor head with three
different voltages applied (a). A period of the evolution of the electric field measured
by the sensor head with three different voltages applied (b).

This test has outlined how these kinds of sensors are sensitive both to rapid
variations of the electric field and to the rather slow deposition of charges on
their surface. The rapid variations are associated mostly with streamers.

3.2 Insulator Without Defects

Now we pass to some more complicated and realistic geometries, in particular we
measure the radial and axial electric field in proximity of a high voltage insulator.
A mobile guide has been placed, see Fig. 3, aligned with the axis of the insulator,
at a distance of 15 cm. The head of the instrument has been placed on the guide
while the electronics used has been kept in a safe position far away from any
energized part. A 50 m bundle, five fibres as described above, connected the two
parts of the instrument. Therefore, in this way, the head is capable of measuring
the electric field at different positions along the insulator.
The insulator is a high voltage component, its insulating pars has a length
of 3.28 m, the maximum diameter of the sheds is 14 cm and a total 87 sheds are
present. The upper terminal has been grounded while the lower one has been
energized with an AC power supply at 50 Hz and 220 Kv r.m.s. The lower end
 Dielectric E-Field 433

has been attached to a single conductor and, finally, the insulator is brand new
without any defect and ageing.

Fig. 3. The high voltage insulator undergoing tests.

The actual geometry of the insulator has been reproduced using a commercial
simulation tool based on the finite element method (FEM). Since, in this case, no
corona effect or internal discharges take place, a direct and accurate comparison
between measurements and simulations is feasible.
In particular, we have compared the magnitude, see Fig. 3, and the angle, see
Fig. 4(a), of the electric field. A null angle corresponds to a purely radial electric
field.

Fig. 4. Comparison between the simulated and measured electric field modulus (a) and
direction (b).
434 A. Villa et al.

As we can see, the measured electric field modulus is very close to the theo-
retical expectations. The phase angle shows a maximum difference between the
measurements and the computations of 0.1 radiants which corresponds to a few
degrees. This test underlines the capability of the sensor to measure accurately
both the axial and radial electric fields also in complicated, realistic, geometries.

3.3 Insulator with Defects

We now test the sensor in a more relevant environment, i.e. we measure the
electric field near a faulty insulator. The insulator has the same geometrical
characteristics of the one described in the previous section, however this unit has
been used in a high voltage power line for years and has a clear carbonization
near the shed number 57. There is a hole which perforates the silicon cladding
and reaches the glass fibre core.
We have performed the same type of measurements as we did in the previous
section but, in this case, we outline the temporal evolution of the electric field to
better show the bandwidth of the sensor and the way it can be used to diagnose
the presence of defects. In Fig. 5(a) we have depicted the field measured near the
shed number 51: the evolution is nearly sinusoidal with no major distortion or
rapid variation of the field.
A marked difference can be seen repeating the same measurement at the
shed number 54, see Fig. 5(b). A distortion of the sine wave is present, due to
the deposition of surface charges. Moreover, much more evidently, some steep and
rapid variations of the signal, which are correlated to streamers and high power
discharges, can be observed. These discharges are even more evident measuring

Fig. 5. Electric field modulus near the shed no. 51 (a), 54 (b), 57 (c).
 Dielectric E-Field 435

the field at the shed number 57, which is the actual position of the defect, see
Fig. 5(c). Moreover, an even larger distortion of the sine wave is present, due to
a massive injection of charges in the air.

4 Conclusions
The tests have proved that this optic field sensor can be used successfully for the
diagnostic of overhead line insulators. Its dielectric characteristics make its use
quite straightforward even near high voltage sources. These tests have shown that
the system can be used to identify not only the position of the defects but also
to evaluate their severity. The defects are identified not only by comparing the
spatial electric field profile along the insulator but also interpreting the temporal
evolution of the measured electric field. Some first data interpretation methods,
based on the experience gained on simpler cases where corona discharges are
present, have been provided in this work. Moreover also new simulation tools
[7–9] may be used to better interpret the results.
The reduced dimensions of the head and the possibility of using a better
packaging for the electronics unit makes it possible to significantly reduce the
dimensions of the system. In fact, it is foreseeable that the sensor will be embed-
ded in automatic robotic systems. These robots can be used either remotely or
they can be fully automatic. This subsequent evolution would solve one of the
main issues related to the E-field sensors i.e. the need of using some very skilled
LLWs for insulator diagnosis.

Acknowledgments. This work has been financed by the Research Found for the
Italian Electrical System under the Contract Agreement between RSE and the Ministry
of Economic Development. The authors wish to thank L. Barbareschi for her valuable
contribution and suggestions.

References
1. Schmuck, F., Gutman, I., Mahatho, N., Perez, M., Phillips, A., Pigini, A., Pirovano,
G., Seifert, J., Shariati, M.R., Sklenicka, S., Vosloo, W., Wesley, R.: Assessment of
in-service composite insulators by using diagnostic tools. Technical report, Cigre
WG B2.21 (2013)
2. Gutman, I.: Application of helicopter-based IR technology for detecting internal
defects in composite insulators. INMR 17, 60–67 (2009)
3. de Nigris, M., Tavano, T., Zagliani, F., Redina, R.: Diagnostic methods of non-
ceramic insulators for HV lines. In: Cigre WG B2 (2000)
4. Jin, L., Ai, J., Tian, Z., Zhang, Y.: Detection of polluted insulators using the infor-
mation fusion of multispectral images. IEEE Trans. Dielectr. Electr. Insul. 24(6),
3530–3538 (2017)
5. Pigini, A., Pirovano, G., Rocchetti, G.: Live line work on HV lines equipped with
composite insulators: state of the art. In: INMR (2011)
6. Barbieri, L., Gondola, M., Potenza, M., Villa, A., Malgesini, R.: A sensor for vector
electric field measurements through a nonlinear anisotropic optical crystal. Rev. Sci.
Instrum. 88, 113114 (2017)
436 A. Villa et al.

7. Villa, A., Barbieri, L., Gondola, M.: Towards complex physically-based lightning
simulations. In: International Colloquium on Lightning and Power Systems (2016)
8. Villa, A., Barbieri, L., Gondola, M., Leon-Garzon, A.R., Malgesini, R.: Simulation
of the AC corona phenomenon with experimental validation. J. Phys. D Appl. Phys.
50, 435201 (2017)
9. Villa, A., Barbieri, L., Gondola, M., Leon-Garzon, A.R., Malgesini, R.: A PDE-based
partial discharge simulator. J. Comput. Phys. 345, 687–705 (2017)
 Investigation on Performances and Functions
of Power Emergency Diesel Generators

ZiChong Zhang1(&), YuQuan Liu1, WenXiong Mo1,

HongBin Wang1, ZhiGuo An1, Han Liu1, LinHuan Luo1,
and Zhong Wang2
1
Guangzhou Power Supply Co., Ltd., Guangzhou, People’s Republic of China
475165733@qq.com
2
School of Electrical Engineering and Information, Sichuan University,
Chengdu, People’s Republic of China

Abstract. In emergency repair of sudden disasters, the reliability and stability

of emergency power generators largely determine the quality and efficiency of
power supply restoration. According to China’s national standards and relevant
technical regulations, the performance and function of 39 power emergency
diesel generators in service in a power supply bureau are studied. The research
found that after 4–9 years of service, nearly 40% of the steady or transient
output voltage deviation was too large, nearly 70% of the transient frequency
deviation was too large, and 25% of the diesel power generators could not
operate continuously with rated load for a long time. The deterioration of the
performance of diesel power generators has seriously affected the quality and
reliability of emergency power supply, which brings hidden dangers to emer-
gency power supply. More attention should be paid to the detection and
maintenance of diesel power generators.

Keywords: Electric emergency repair
Emergency power generator
Diesel

power generator
Reliability

1 Introduction

As a national basic energy industry, electric power enterprises play an important role in
the national economy and people’s life. In recent years, extreme bad weather has
occurred frequently. In all previous rescue and power supply restoration processes,
problems such as insufficient rescue experience of emergency teams, poor quality of
emergency equipment, and lack of basic emergency knowledge and disposal skills have
been exposed. Guangzhou, as a national central city, is one of the city networks with
the highest power supply load density in the country, which requires extremely high
reliability and safety of power supply. Due to its wide coverage area and complex
climate and topography, the China Southern Power Grid is often affected by severe
weather disasters such as typhoons, rainstorms and thunderstorms, which have great
hidden trouble of power failure [1]. In order to ensure the residents’ electricity con-
sumption and social safety, the quality [2] and efficiency of emergency repair and
restoration are very important.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 437–448, 2020.
https://doi.org/10.1007/978-3-030-31676-1_42
438 Z. Zhang et al.

In the process of emergency rescue of sudden disasters, the reliability and stability
of emergency power generator have a great impact on the rapid recovery of electricity.
If quality problems occur in emergency power generator, it will bring great hidden
dangers and adverse effects to emergency repair work [3–6]. At present, there is no
national standard or industry standard for emergency power supply vehicles used for
emergency repair of power grids, and power grid companies also lack the detection
capability of emergency equipment products; Moreover, there are many manufacturers
of emergency power generator, mixed with good and evil people, with uneven product
quality and low threshold. Therefore, it is urgent to strengthen quality control.
According to China’s national standards and relevant technical regulations, this article
studies the performance and function of 39 power emergency diesel generators in
service in a power supply bureau. The research is expected to lay a foundation for the
formulation of technical standards, technical specifications and operation and mainte-
nance guidance of emergency power generator, improve the management level of
emergency power generator, and enhance the operability and reliability of the
equipment.

2 Sample

This article studies the operation performance of 39 power emergency diesel generators
in service in a power supply bureau. These power generators have been in use for 4 to 9
years with a rated output voltage of 230 V and a rated output power of 200, 400 or
500 kW. Before the test, no damage was found on the tested generator.

3 Reference Standards and Performance Parameters

This research work is carried out strictly in accordance with the requirements of
China’s national standard GB/T 20136-2006 “General Test Methods for Internal
Combustion Engine Power Stations”, industry standard YD/T 502-2007 “Diesel
Generator Set for Communication” and relevant technical agreements of a power
supply bureau [7, 8]. The performance judgment is based on the technical agreement.
When the technical agreement is lower than the national standard, industry standard or
enterprise commitment, the higher requirements in the national/industry standard or
enterprise commitment shall be taken as the performance judgment basis.
The performance and parameters studied include steady-state voltage deviation and
steady-state frequency deviation, transient voltage deviation and transient frequency
deviation, voltage and frequency recovery time, three-phase voltage imbalance, fre-
quency drop, phase voltage sine wave distortion rate, insulation resistance and other
non-electrical performance and parameters, as well as operation state parameter
monitoring function, overload protection function, emergency stop function, continu-
ous operation function, generator appearance and phase sequence arrangement.
 Investigation on Performances and Functions 439

4 Test Items
4.1 Steady-State Voltage and Frequency Deviation [9]
The research steps of this performance parameter are as follows: start and set the
generator to operate stably under rated conditions, and reduce the load to no load after a
period of time; Gradually load from no-load to 25%, 50%, 75%, 100% of rated load,
then reduce the load from 100% to no-load step by step according to this grade, and run
for a period of time under each grade of load; Test the three-phase voltage, frequency
and output power under various load levels. According to relevant technical require-
ments, the deviation between steady-state voltage and rated voltage shall be  ±0.5%,
and the deviation between steady-state frequency and rated frequency shall be
 ±0.5%.

Upper boundary 230*(1+0.5%) V

232 (a) 600

230 Lower boudary 230*(1-0.5%) V

500
228 100%

Output Power P/kW

Phase Voltage U/V

226 400
75%
224 75%
300
222
50% 50%
220 200

218 25% 25%

Phase A 100
216 Phase B
Phase C
214
0 100 200 300 400 500 600 700 800
Time t/s

50.4 (b) Upper boundary 50*(1+0.5%) Hz 600

50.2 500
100%
Output Power P/kW

50.0 400
Frequency f/Hz

75%
75%
49.8 300
50% 50%
49.6 Lower boundary 50*(1-0.5%) Hz 200
25% 25%
49.4 100

49.2 0
0 100 200 300 400 500 600 700 800
Time t/s

Fig. 1. Time variation characteristics of output voltage, frequency and power of 500 kW power
generator (steady state)
440 Z. Zhang et al.

Figure 1 shows the time variation characteristics (steady-state) of three-phase

voltage, frequency and output power under different load rates for a 500 kW diesel
generator under inspection. In order to visually judge whether the steady-state voltage
and steady-state frequency meet the requirements (the steady-state voltage deviation
should be less than or equal to 0.5%, and the steady-state frequency deviation should
be less than or equal to 0.5%), the upper and lower boundary lines of the steady-state
voltage and steady-state frequency are respectively made in the figure and the upper
and lower limit values are marked.
Of the 39 diesel generators tested, 15 failed in steady-state voltage deviation and
did not fail in steady-state frequency deviation. The absolute value of the maximum
steady-state voltage deviation of unqualified power generators is shown in Fig. 2. For
convenience of observation, the required upper limit boundary line of steady-state
voltage deviation is made in the figure. As can be seen from Fig. 2, the steady-state
voltage deviation of most unqualified diesel generators is concentrated between 1.00%
and 1.70%, except that the steady-state voltage deviation of one generator is as high as
2.5%. It can be seen that after a period of service, the steady-state output voltage
deviation of diesel power generators is more obvious, but the time stability of voltage
frequency is better.

3.00%

2.50%
Absolute value of steady-state

2.00%
voltage deviation

1.50%

1.00%

0.50%

0.00%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Unqualified diesel generator
Fig. 2. Absolute value of steady-state voltage deviation of unqualified diesel generator

4.2 Transient Voltage, Frequency Deviation and Voltage and Frequency

Recovery Time [10]
The research steps of this performance parameter are as follows: start and set the
generator to operate stably under rated conditions, and suddenly reduce to no load after
a period of time; Sudden load increase to rated load, after stable operation, sudden load
reduction to no load; Test the three-phase voltage, frequency and output power during
 Investigation on Performances and Functions 441

sudden load increase and sudden load decrease. According to relevant technical
requirements, transient voltage deviation should be  ±15%, transient frequency
deviation should be  ±7%, voltage recovery time should be  1 s, and frequency
recovery time should be  3 s.

232 500
(a)

228 400

Output Power P/kW

Phase Voltage U/V

224 ΔUC 300

ΔUB
ΔUA
220 200

Phase A
216 Phase B 100
Phase C

212 0
0 20 40 60 80 100
Time t/s

51.0 500
(b)

50.5
400
Output Power P/kW

50.0
Frequency f/Hz

300
49.5
Δf 200
49.0

100
48.5

48.0 0
0 20 40 60 80 100
Time t/s

Fig. 3. Time variation characteristics (transient) of output voltage, frequency and power of
500 kW diesel power generator during sudden load change

Figure 3 shows the time variation characteristics (transient) of three-phase voltage,

frequency and output power when a 500 kW diesel generator under inspection sud-
denly reduces its load. It can be seen from figure (a) that when the load is suddenly
reduced from the rated load to no load, the three-phase voltages all have “V” type
abrupt changes, and their maximum drop values are about 8 V, 6 V and 4 V respec-
tively, which are still within the allowable ±15% transient voltage deviation range. As
442 Z. Zhang et al.

can be seen from Fig. 3(b), when the load suddenly decreases from the rated load to no
load, the frequency also appears “V” type abrupt changes with the maximum decrease
value of about 1.5 Hz, which is still within the allowable ±7% transient frequency
deviation range. Similar studies show that when the load suddenly increases from no-
load to rated load, the maximum drop of three-phase voltage exceeds the allowable
range of ±15% transient voltage deviation. The maximum frequency drop is still
within the allowable ±7% transient frequency deviation.

30.00%
Absolute value of transient voltage

25.00%

20.00%
deviation

15.00%

10.00%

5.00%

0.00%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Unqualified diesel generator
Fig. 4. Absolute value of transient voltage deviation of unqualified diesel generator

30.00%
Absolute value of transient

25.00%
frequency deviation

20.00%

15.00%

10.00%

5.00%

0.00%
1 3 5 7 9 11 13 15 17 19 21 23 25 27
Unqualified diesel generator
Fig. 5. Absolute value of transient frequency deviation of unqualified diesel generator
 Investigation on Performances and Functions 443

Of the 39 diesel generators tested, 15 failed in transient voltage deviation and 27

failed in transient frequency deviation. The absolute value of transient voltage devia-
tion and the absolute value of transient frequency deviation of unqualified power
generators are shown in Figs. 4 and 5 respectively. In order to facilitate visual judg-
ment, the required upper boundary lines of transient voltage and frequency deviation
are respectively made in the figure. As can be seen from Fig. 4, the transient voltage
deviation of unqualified diesel power generators is distributed between 15.00% and
25.00%. As can be seen from Fig. 5, except that the transient frequency deviation of
two power generators exceeds 20%, the transient frequency deviation of most
unqualified diesel power generators is distributed between 7.00% and 20.00%. It can be
seen that after a period of service, the transient performance of diesel power generators
decreases obviously, and a large number of power generators have the problem of
excessive transient voltage and frequency deviation. 39 diesel generators tested, 15
failed in transient voltage deviation and 27 failed in transient frequency deviation. The
absolute value of transient voltage deviation and the absolute value of transient fre-
quency deviation of unqualified power generators are shown in Figs. 4 and 5 respec-
tively. In order to facilitate visual judgment, the required upper boundary lines of
transient voltage and frequency deviation are respectively made in the figure. As can be
seen from Fig. 4, the transient voltage deviation of unqualified diesel power generators
is distributed between 15.00% and 25.00%. As can be seen from Fig. 5, except that the
transient frequency deviation of two power generators exceeds 20%, the transient
frequency deviation of most unqualified diesel power generators is distributed between
7.00% and 20.00%. It can be seen that after a period of service, the transient perfor-
mance of diesel power generators decreases obviously, and a large number of power
generators have the problem of excessive transient voltage and frequency deviation.

4.3 Three-Phase Voltage Imbalance [11]

The research steps of this performance parameter are as follows: start and set the
research steps of this performance parameter are as follows: start and set the generator
to operate stably under rated conditions, reduce the load to no-load after a period of
time, and test the voltage imbalance degree when no-load occurs. According to the
relevant technical requirements, the voltage unbalance shall be  1% when no load is
applied. The generator to operate stably under rated conditions, reduce the load to no-
load after a period of time, and test the voltage imbalance degree when no-load occurs.
According to the relevant technical requirements, the voltage unbalance shall be  1%
when no load is applied.
Figure 6 shows the time variation characteristics of the three-phase voltage when a
tested 500 kW diesel generator vehicle is running stably at no load. The tester shows
that its three-phase voltage imbalance is 0.26%, which meets the requirement that the
voltage imbalance should be  1% at no load.
Of the 39 diesel generators tested, none had the problem that the voltage imbalance
was greater than 1% when no load was applied. It can be seen that the time stability of
the three-phase voltage unbalance degree is good when the power generator is in no-
load operation after a period of service.
444 Z. Zhang et al.

232

230
Phase Voltage U/V

228

226
Phase A
Phase B
224 Phase C

222
0 40 80 120 160 200 240
Time t/s

Fig. 6. Time variation characteristics of three-phase voltage for 500 kW diesel generator under
no-load stable operation.

4.4 Frequency Drop

The research steps of this performance parameter are as follows: start and set the
generator to operate stably under rated conditions, and reduce the load to no load after a
period of time; Record the voltage frequency at full load and no load. According to
relevant technical requirements, the frequency drop should be  3%.
Of the 39 diesel generators tested, only one generator has a frequency drop of
4.06%, while the rest meet the technical requirements. It can be seen that the time
stability of the frequency drop is good after the power generator has been in service for
a period of time.

4.5 Phase Voltage Sine Wave Distortion Rate [12]

The research steps of this performance parameter are as follows: the generator is in a
cold state or a hot state, and the generator is started and adjusted to operate stably under
the rated working condition; After a period of time, reduce the load to no load and
adjust the frequency to the rated value to test the three-phase voltage and its sine wave
distortion rate. According to relevant technical requirements, the sine wave distortion
rate of three-phase voltage shall be  3% when no load is applied and  10% when
full load is applied.
Figure 7 shows the time variation characteristics of the three-phase voltage and its
sine wave distortion rate when a 500 kW diesel generator under inspection is unloaded.
It can be seen from the figure that the sine wave distortion rate of the three-phase
voltage fluctuates up and down at 1.3 when no load is applied, and the distortion rate of
phase B is the smallest and that of phase C is the largest, but all meet the requirement
that the sine wave distortion rate of the voltage is  3% when no load is applied.
 Investigation on Performances and Functions 445

Of the 39 diesel power generators tested, the distortion rate of three-phase voltage
sine wave exceeds 10% when 4 power generators are fully loaded, and the other power
generators meet the technical requirements. It can be seen that the time stability of
phase voltage sine wave distortion rate is good after the power generator has been in
service for a period of time.

232 5

Voltage Sine Wave Distortion Rate Uthd

4
230
Phase Voltage U/V

3
228
Voltage sine wave distortion rate of Phase C 2
Voltage sine wave distortion rate of Phase A

226
Phase A 1
Voltage sine wave distortion rate of Phase B Phase B
Phase C
224 0
0 20 40 60 80 100
Time t/s

Fig. 7. Time variation characteristics of three-phase voltage and sine wave distortion rate of
500 kW diesel generator under no-load condition

4.6 Continuous Operation

The research steps of this performance parameter are as follows: start and set the
generator to run stably under the rated working condition, making it run continuously
for 60 min under full load; During the operation, the generator will not stop without
warning. After the operation, the generator unit will be checked for oil leakage, water
leakage and other abnormal phenomena. The diesel generator is considered to have
normal continuous operation capability.
Of the 39 diesel generators tested, 10 failed in continuous operation; One of them
can run continuously for 60 min, but after the operation, the unit was found to have oil
leakage; The other 9 power generators will stop when they are running for less than
60 min. The continuous operation time of most unqualified diesel generators is
between 20 min and 30 min, except that the continuous operation time of 2 generators
is less than 15 min. In addition, the detection found that the reason why most power
generators stop is that the unit cooling water temperature is too high, which leads to
alarm shutdown. It can be seen that after a period of service, the continuous operation
capability of some power generators cannot meet the technical requirements, and the
most likely reason is the reduction of heat dissipation capability.
446 Z. Zhang et al.

4.7 Other Parameters

Other performances and parameters of the electric emergency diesel generator that are
not directly related to the output performance are studied including insulation resis-
tance, overload protection, emergency stop function, appearance, monitoring function
and phase sequence arrangement.
The detection method of insulation resistance is to use insulation resistance tester to
study the insulation resistance of different ports of the generator set to ground, requiring
the insulation resistance of each port to be >0.3 MX. The 39 diesel power generators
tested all meet the technical requirements. It can be seen that the insulation perfor-
mance of different ports to the ground is stable after the power generators have been in
service for a period of time.
The overload protection detection method is to adjust the overload protection
setting value of the power generator to a lower value, and then gradually load the value
to observe whether the power generator is stopped due to overload alarm. The 39 diesel
generators tested all triggered overload protection normally. It can be seen that the
overload protection function remains good after the generators have been in service for
a period of time.
The detection method of the emergency stop function is to press the emergency
stop switch when the power generator is running to observe whether the power gen-
erator is stopped in an emergency. The emergency stop function of the 39 diesel power
generators tested is normal. It can be seen that the emergency stop function remains
good after the power generators have been in service for a period of time.
When testing the appearance of the generator, observe whether the car surface of
the generator is flat, whether the exposed metal is seriously corroded, whether lighting
devices are equipped and operating normally, whether the supporting cables are seri-
ously damaged and whether firefighting equipment is equipped. Among the 39 power
generators tested, 26 have different defects in appearance. The most common problems
are surface paint damage, failure of lighting device and expiration of extinguishing
agent in the matching fire extinguisher. It can be seen that after a period of service, the
appearance of the power generator deteriorates obviously.
The monitoring function inspection method is to observe the generator control
panel and check whether it has the function of monitoring generator set operation state
parameters such as generator voltage/current, frequency, active/reactive power, oper-
ation time, water temperature, oil pressure, etc. Of the 39 power generators tested, only
one power generator has a control panel failure, and the other power generators have
normal monitoring functions. It can be seen that the monitoring system functions well
after the power generator has been in service for a period of time.
When detecting the phase sequence arrangement, observe the three-phase sequence
arrangement of the connection terminals. The correct phase sequence arrangement shall
be from left to right or from top to bottom facing the connection terminals. The phase
sequence arrangement of 39 diesel generators tested is correct, which indicates that the
generators are designed according to the specifications.
 Investigation on Performances and Functions 447

5 Conclusions

According to China’s national standard GB/T 20136-2006 “General Test Methods for
Internal Combustion Engine Power Stations”, industry standard YD/T 502-2007
“Diesel Generator Set for Communication” and relevant technical regulations, this
article systematically studies various performance parameters of 39 electric emergency
diesel generators after serving for a certain period of time, and tests various items and
unqualified rates as shown in Table 1. The following conclusions are drawn from the
research:
1. After a certain period of service, the steady-state frequency deviation, voltage
imbalance, frequency drop, insulation resistance, overload protection, emergency
stop function and monitoring function have not occurred or only a small number of
power generators have failed, which indicates that the above performance and
function are stable.
2. After a certain period of service, nearly 40% of the power generators have excessive
steady-state or transient voltage deviation, and nearly 70% of the power generators
have excessive transient frequency deviation. The deterioration of output voltage and
frequency of power generators is obvious, which seriously affects the quality of
emergency power supply. The maintenance of diesel power generators should attach
great importance to these quality indicators and shorten the test cycle appropriately.
3. After a certain period of service, 25% of the power generators are not qualified for
continuous operation, and the operation time with rated load is less than 60 min.
This seriously affects the reliability of emergency power supply and brings hidden
dangers to emergency power supply. The maintenance of diesel power generators
should pay more attention to full load operation capability to ensure normal con-
tinuous operation capability.

Table 1. Items and failures rate of diesel generators

No. Test items Failure rate
1 Steady-state voltage deviation 38.46%
2 Steady-state frequency deviation 0.00%
3 Transient voltage deviation and voltage recovery time 38.46%
4 Transient frequency deviation and frequency recovery time 69.23%
5 Voltage imbalance 0.00%
6 Frequency drop 2.56%
7 Total harmonic distortion rate of voltage 10.26%
8 Insulation resistance 0.00%
9 Overload protection 0.00%
10 Emergency stop function 0.00%
11 Continuous operation 25.64%
12 Appearance check 66.67%
13 Monitoring function check 2.56%
14 Check phase sequence 0.00%
448 Z. Zhang et al.

Acknowledgments. This article is completed with the support of the science and technology
project “Research and Application of Key Technology of Life Cycle Based on High Reliability
for Emergency Equipment (GZHKJXM20170110)” of Guangzhou Power Supply Bureau Co.,
Ltd.

References
1. Wang, M., Ye, Q., Wang, D.: The development of natural disaster response capacity for
power system (in Chinese). Power Syst. Prot. Control. 36(13), 57–60 (2008)
2. Analysis of common problems in customer electrical equipment power quality (in Chinese).
China Electric Power Press (2013)
3. Zhang, M.: Emergency Management of Emergency Response to Power Supply (in Chinese).
Master, North China Electric Power University (2012)
4. Zhao, Z.: Application of integrated emergency warning system of electric power equipment
(in Chinese). Guangdong Electr. Power 22(1), 74–77 (2009)
5. Wang, Y.: Research on Management System and Capability Evaluation of Power System
Sudden Disaster Emergency Project (in Chinese). Master, North China Electric Power
University (2015)
6. Wang, Z., Sun, L., Lin, Z.: Optimal allocation of emergency power sources considering
uncertainties of loads and dispatching time periods (in Chinese). Autom. Electr. Power Syst.
42(2), 34–41 (2018)
7. GB/T 20136-2006: General Test Methods for Internal Combustion Engine Power Stations
(in Chinese). Lanzhou Power Vehicle Research Institute, China Electrical Equipment
Industry Association (2006)
8. YD/T 502-2007: Diesel Generator Set for Communication (in Chinese) (2007)
9. Feng, J., Wang, X., Zhang, C., Gao, C.: Active synchronization control of inverter based on
frequency recovery control (in Chinese). Adv. Technol. Electr. Eng. Energy 35(6), 7–12
(2016)
10. Chui, P.: Voltage frequency recovery and reactive power distribution based on consistency
algorithm (in Chinese). Master, Shenyang Agricultural University (2017)
11. An, J.: Three-phase voltage unbalance detection based on symmetrical component filter (in
Chinese). Electr. Autom. 33(6), 71–73 (2011)
12. Heusdens, R., Jensen, J., Korten, P.: Rate-distortion optimal high-resolution differential
quantization for sinusoidal coding of audio and speech. In: IEEE Workshop on Applications
of Signal Processing To Audio and Acoustics, pp. 243–246. IEEE, New Paltz (2005)
 Condition Assessment of DC XLPE Cables
Based on Combination Weighting Method

Yufeng Zhu1(&), Siyuan Wang2, Cong Li2, Yongpeng Xu1,

Gehao Sheng1, and Xiuchen Jiang1
1
Department of Electrical Engineering, Shanghai Jiao Tong University,
Shanghai, China
zhu-yufeng@sjtu.edu.cn
2
State Grid Jinan Electric Power Supply Company, Jinan, China

Abstract. Aiming at condition assessment for multiple characteristics, this

paper proposes the combination weighting method to make full use of infor-
mation detected by routine tests. Fuzzy analytical hierarchy process (FAHP) is a
modified subject weighting method and anti-entropy method is a modified
objective weighting method. Combination weighting method combines FAHP
and anti-entropy method to get the more reasonable results. First, the index
system of DC XLPE cables are established. Second, weight parameters of each
index are calculated by FAHP and anti-entropy method. Third, associating
weight parameters calculated above, weight parameters of combination
weighting method are deduced. Then, calculate the relative degree of deterio-
ration and degree of membership of each index. Finally, use weight vector
calculated by combination weighting method and degree of membership matrix
to get the condition membership. Compared to FAHP and anti-entropy method,
combination weighting method has more smooth distribution of membership
and can reflect the potential faults according to some sensitive indexes.

Keywords: Condition assessment
DC XLPE cables
Combination weighting

method
FAHP
Anti-entropy method

1 Introduction

DC cross-linked polyethylene (XLPE) cables have been more and more important since
the high voltage direct current (HVDC) program promoted. At present, large amount of
DC XLPE cables have come into service and contribute to modern power transmission,
especially in China. Therefore, diagnosis and condition assessment of DC XLPE cables
are vital to guarantee the economic and social security.
Recently, various studies mainly focus on partial discharge (PD) [1], space charge
[2] and nanocomposites [3] of DC XLPE cables. And the fault diagnosis usually
depends on PD signals based on k-nearest neighbor (KNN), artificial neural network
(ANN), support vector machine (SVM) and other intelligent algorithms [4].
Although PD is the one of the most significant indexes of potential faults, routine tests
such as insulation resistance and DC withstand voltage can also reflect the potential
problems [5]. Therefore, taking various indexes into consideration can get more

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 449–457, 2020.
https://doi.org/10.1007/978-3-030-31676-1_43
450 Y. Zhu et al.

information and evaluate more accuracy conclusion. Weighting method is one of the
popular methods considering various indexes, including subjective weighting method,
objective weighting method and combination weighting method [6, 7]. Subjective
weighting method mainly uses expertise to decide the weight parameters, but it cannot
use the various information embedded in data. And objective weighting method mainly
uses data discrepancy to determine the weight parameters, so the weight parameters
may change rapidly according to the different data set. Combination weighting method
associates subjective weighting method with objective weighting method and takes the
advantages of both two methods.
In this paper, combination weighting method is used to determine the weight of the
indexes reflecting the potential faults in DC XLPE cables. Then condition assessment is
represented depending on weight parameters given by combination weighting method.
Finally, different weighting methods are compared to verify the effectiveness of
combination weighting method.

2 Combination Weighting Method

Every index has different influence on condition assessment. Therefore, it is quite

important to determine the weight parameters of each index. The most popular
weighting methods include subjective weighting method and objective weighting
method. Subjective weighting method contains expert judgment method, analytic
hierarchy process (AHP) and empirical weighting method. Objective weighting method
contains rough set method, variation coefficient method and entropy method. Combi-
nation weighting method combine subjective weighting method and objective
weighting method together to get more reasonable weight parameters.

2.1 Fuzzy Analytical Hierarchy Process

Analytical hierarchy process (AHP) is one of the most popular subjective weighting
methods. It is a multi-objective decision-making method proposed by American
operational research expert Saaty [8] in 1970s. Judgement matrixes made by AHP need
to satisfy the consistency, and may be adjusted many times. Moreover, the consistency
of matrix has significant difference from the consistency of human thinking. Therefore,
fuzzy analytical hierarchy process (FAHP) is proposed to solve this consistency
problem using fuzzy consistent matrix.
First, we use 0.1–0.9 rating scale method to provide the priority
 judgement matrix
satisfied fuzzy complementarity. Suppose matrix A ¼ aij nn , if A satisfies
aij þ aji ¼ 1ði; j ¼ 1; 2;


; nÞ, then we say A is a fuzzy complementarity matrix.
Second, transfer the priority judgement matrix A to fuzzy consistent matrix


 Pn
R ¼ rij nn , where rij ¼ ri  rj =2n þ 0:5, ri ¼ aij ; i ¼ 1; 2;


; n. n is the num-
j¼1
ber of indexes.
 Condition Assessment of DC XLPE Cables 451

Finally, calculate the weight parameters of each index. Relational ranking method
has advantages of high resolution and obviously difference between each index.
Therefore, we use relational ranking method to calculate the weight parameters.

1 1 1 Xn
w1i ¼  þ rik ; i ¼ 1; 2;


; n ð1Þ
n 2a na k¼1

Where: w1i is the weight parameter of i-th index determined by FAHP. a ¼ ðn  1Þ=2,
a is adjustment parameter.

2.2 Anti-entropy Method

Entropy method is one of the common objective weighting method. However, it has
highly sensibility on diversity of indexes and may get 0 weight under extreme con-
dition. Therefore, we use anti-entropy method [9] to solve this problem and reflect the
difference between indexes.

First, get the evaluation index matrix C ¼ cij mn , whose each element is between
0 and 1. Where m is the number of data record.
Then, calculate the anti-entropy.

X
n

hi ¼  rij ln 1  rij ; i ¼ 1; 2;


; m ð2Þ
j¼1

P
n
Where: hi is the anti-entropy of i-th index. rij is calculated by rij ¼ cij = cij .
j¼1
Finally, determine the weight parameters by normalized.

X
m
w2i ¼ hi = hi ð3Þ
i¼1

Where: w2i is the weight parameter of i-th index calculated by anti-entropy method.

2.3 Combination Weighting Method

Depends on weight parameters determined by FAHP and anti-entropy method, we can
get the weight parameters of combination weighting method using multiplication
integration method.

X
m
wi ¼ w1i w2i = w1i w2i ð4Þ
i¼1

Where: wi, w1i and w2i are the weight parameters of i-th index determined by
combination weighting method, FAHP and anti-entropy method respectively.
452 Y. Zhu et al.

3 Condition Assessment of DC XLPE Cables

3.1 Index System
Indexes of DC XLPE cables can be divided into insulation system, outer sheath and
cable core according to existing standard [5]. And each index can be subdivided into
many sub-indexes. The index system of DC XLPE cables is shown below (Fig. 1).

Fig. 1. Index system of DC XLPE cables.

3.2 Membership Function

Membership function is the fundamental of the fuzzy evaluation matrix. The common
membership functions include ridge distribution function and triangular distribution
function. Among them, ridge distribution function has the correspondence between
non-linear feature and deterioration process in reality. Therefore, we use ridge distri-
bution function to describe the fuzzy relation between each state.
In this paper, we divide the state into four rank: good, normal, attention and serious,
represented by p1, p2, p3 and p4 respectively. And we simply use uniform distribution
to determine the peak points of each state, as 0.2, 0.4, 0.6, 0.8 respectively. The fuzzy
membership function is shown below (Fig. 2).
8
< 1 x  0:2
lp1 ð xÞ ¼ 0:5  0:5 sin p2 ðx  0:3Þ 0:2\x  0:4 ð5Þ
:
0 x [ 0:4
8
>
> 0 x  0:2
<
0:5  0:5 sin p2 ðx  0:3Þ 0:2\x  0:4
lp2 ð xÞ ¼ ð6Þ
> 0:5 þ 0:5 sin 2 ðx  0:5Þ 0:4\x  0:6
> p
:
0 x [ 0:6
 Condition Assessment of DC XLPE Cables 453

8
>
> 0 x  0:4
<
0:5  0:5 sin p2 ðx  0:5Þ 0:4\x  0:6
lp3 ð xÞ ¼ ð7Þ
>
> 0:5 þ 0:5 sin p2 ðx  0:7Þ 0:6\x  0:8
:
0 x [ 0:8
8
< 0 x  0:6
lp4 ð xÞ ¼ 0:5 þ 0:5 sin p2 ðx  0:7Þ 0:6\x  0:8 ð8Þ
:
1 x [ 0:8

Fig. 2. Fuzzy membership function.

3.3 Comprehensive Evaluation

After getting the weight vector W and fuzzy membership matrix L, we can use fuzzy
comprehensive evaluation formula Y ¼ W  L to get the degree of state membership
Y = [y1, y2, …y4]. Where:  is the fuzzy operator, we usually chose weighted mean fuzzy
operator. y1, y2, …y4 are the degree of membership that this DC XLPE cable belongs to p1,
p2, …, p4 state. Finally, we can conclude the condition state of this DC XLPE cable using
the maximum membership principle and realize the condition assessment.

4 Result and Discussion

4.1 Calculation of Weight Parameters

As to established index system, we use FAHP, anti-entropy method to calculate the
weight parameters respectively. And then use combination weighting method to
determine the final weight parameters.
454 Y. Zhu et al.

Hierarchical priority relation matrix established by FAHP can be divided into

target-index and index-sub-index two groups. Suppose hierarchical priority relation
matrix of target-index is F_S, hierarchical priority relation matrixes of index-sub-index
are S1_T, S2_T and S3_T.
2 3
0:5 0:8 0:8
F S ¼ 4 0:2 0:5 0:5 5 ð9Þ
0:2 0:5 0:5
2 3
0:5 0:7 0:7 0:8
6 0:3 0:5 0:5 0:7 7
S1 T ¼ 6
4 0:3
7 ð10Þ
0:5 0:5 0:6 5
0:2 0:3 0:4 0:5
 
0:5 0:6
S2 T ¼ ð11Þ
0:4 0:5

S3 T ¼ ½0:5 ð12Þ

Then we establish the fuzzy consistent matrix and use relational ranking method to
get the weight parameters of FAHP about each sub-index.
Anti-entropy method mines information entropy of different index using given data
set. Then determine the weight parameters of each sub-index using normalization.
Finally, we use combination weighting method to calculate the weight parameters
by associating weight parameters calculated by FAHP and anti-entropy method.
Weight parameters of each method is shown below (Table 1).

Table 1. Weighting parameters of each index

Index Sub-index FAHP Anti-entropy Combination weighting
Insulation system PD 0.262 0.169 0.298
Leakage current 0.183 0.097 0.119
DC withstand voltage 0.174 0.207 0.243
Insulation resistance 0.134 0.109 0.099
Outer sheath DC withstand voltage 0.095 0.207 0.133
Insulation resistance 0.078 0.105 0.055
Cable core DC resistance 0.075 0.106 0.053

4.2 Relative Degree of Deterioration

PD is the most sensitive indicator reflecting the potential faults of DC XLPE cables.
Therefore, we concentrate most on the sub-index of PD. The characteristics of PD
under DC are mainly represented by discharge amplitude q and time interval between
 Condition Assessment of DC XLPE Cables 455

two adjacent discharges Dt. in this paper, we use average discharge amplitude, max-
imum discharge amplitude, average time interval between two adjacent discharges and
discharge repetition rate of 300 continuous discharge sequence to represent the severe
condition of PD.
According to existing acquired data, use normalization method to get the relative
value of each characteristic about PD. The base value of each characteristic about PD is
shown below (Table 2).

Table 2. Base value of each characteristic in PD

Characteristic Average q Max q Average Dt Repetition rate
Base value 100 pC 1 nC 100 ms 1000 s−1

The evaluation of state indexes can be divided into benefit index and cost index.
Benefit index means the larger the value, the better the effect. However, larger value
means worse deterioration in the calculation of relative degree of deterioration, such as
average discharge amplitude, maximum discharge amplitude and discharge repetition
rate. On the contrary, the cost index means that smaller value represents worse dete-
rioration in calculation of relative degree of deterioration, such as average time interval
between two adjacent discharges. The benefit index formula g1(x) and cost index
formula g2(x) is presented below.
8
< 0 x\xmin
g1 ð x Þ ¼ xxmin
xmin  x  xmax ð13Þ
: xmax xmin
1 x [ xmax
8
< 1 x\xmin
xmax x
g2 ð x Þ ¼ xmin  x  xmax ð14Þ
: xmax xmin
0 x [ xmax

After calculation of relative degree of deterioration, we calculate average value of

four above results as the relative degree of deterioration of PD. Then we take it along
with 6 other sub-index weight parameters to assess the condition of DC XLPE cables.

4.3 Instance and Comparison

After one high voltage test, we get a data sequence of a DC XLPE cables. We calculate
the relative degree of deterioration of each sub-index as L = [0.7, 0.5, 0, 0.3, 0, 0.3,
0.1] according to Eqs. 13–14. We get the degree of membership by using FAHP, anti-
entropy method and combination weighting method respectively (Table 3).
456 Y. Zhu et al.

Table 3. Degree of membership calculated by 3 methods

Condition state Good Normal Attention Serious
FAHP 0.4498 0.2174 0.2021 0.1307
Anti-entropy method 0.6273 0.1553 0.1329 0.0845
Combination weighting method 0.4060 0.2364 0.2085 0.1490

As we can see, the relative degree of deterioration of PD sub-index is relatively

large. Therefore, it may have some defaults in this DC XLPE cable. Although 3
methods all conclude to the “good” state, the combination weighting method has the
highest degree of membership about “serious”, which means it reflects this cable may
have some potential defaults. In addition, compared to FAHP and anti-entropy method,
combination weighting method have most smooth distribution of degree of member-
ship. And it is consistent with human thinking that distribution of two adjacent state
should be not highly fluctuated. Thus, combination weighting method has more rea-
sonable result and can reflect the potential faults according to some sensitive sub-
indexes.

5 Conclusions

This paper proposes the combination weighting method using FAHP and anti-entropy
method. Compared to FAHP and anti-entropy method, combination weighting method
has more smooth distribution of membership and can reflect the potential faults
according to some sensitive indexes. The proposed method contributes more reason-
able evaluation to the condition assessment of DC XLPE cables.

Acknowledgments. This work was supported by the National Key R&D Program of China
(2016YFB0900700).

References
1. Gu, X., He, S., Xu, Y., Yan, Y., Hou, S., Fu, M.: Partial discharge detection on 320 kV VSC-
HVDC XLPE cable with artificial defects under DC voltage. IEEE Trans. Dielectr. Electr.
Insul. 25(3), 939–946 (2018)
2. Cao, Z., et al.: Space charge measurement system based on laser PWP method for the full-size
long HVDC cable. In: 2018 Condition Monitoring and Diagnosis (CMD), Perth, WA, pp. 1–4
(2018)
3. Lei, W., et al.: Are nano-composites really better DC insulators? A study using silica
nanoparticles in XLPE. IEEE Trans. Dielectr. Electr. Insul. 24(4), 2268–2270 (2017)
4. Xu, Y., Qian, Y., Yang, F., Li, Z., Sheng, G., Jiang, X.: DC cable feature extraction based on
the PD image in the non-subsampled contourlet transform domain. IEEE Trans. Dielectr.
Electr. Insul. 25(2), 533–540 (2018)
5. IEC 62895-2017: High voltage direct current (HVDC) power transmission. Cables with
extruded insulation and their accessories for rated voltages up to 320 kV for land applications.
Test methods and requirements (2017)
 Condition Assessment of DC XLPE Cables 457

6. Lu, C., Li, L., Wu, D.: Application of combination weighting method to weight calculation in
performance evaluation of ICT. In: 2015 IEEE 15th International Conference on Advanced
Learning Technologies, Hualien, pp. 258–259 (2015)
7. Qu, Q., Guo, X., Zhao, C., Li, Y.: Study on overseas electricity regulatory environment
assessment model based on combination weighting. In: 2018 37th Chinese Control
Conference (CCC), Wuhan, pp. 8667–8670 (2018)
8. Saaty, T.L.: The Analytic Hierarchy Process. McGraw-Hill, New York (1980)
9. Bengfort, B., Xirogiannopoulos, K., Keleher, P.: Anti-entropy bandits for geo-replicated
consistency. In: 2018 IEEE 38th International Conference on Distributed Computing Systems
(ICDCS), Vienna, pp. 1438–1441 (2018)
 Identification of Hoarfrost Corona Losses
Events Based on Advanced Operational Data
Analysis and Weather Modelling

I. Gutman1(&), Y. Solovyev2, H. Ágústsson3, and H. Mc Innes4

1
Independent Insulation Group, Ludvika, Sweden
igor@i2group.se
2
ABB, Ludvika, Sweden
3
Kjeller Vindteknikk, Kjeller, Norway
4
Norwegian Meteorological Institute, Oslo, Norway

Abstract. Corona losses on overhead line conductors due to hoarfrost lead to

the highest corona losses in comparison to other weather conditions initiating
corona (e.g. rain or wet snow). In this paper a promising methodology for
identification and calculation of corona losses induced by hoarfrost using
operational measurements, as well as a numerical weather model to predict
hoarfrost on overhead transmission lines, is proposed. The methodology com-
prises the practical simplified procedure including operational measurements of
the total line losses, subtraction of resistive losses, evaluation and subtracting of
a measuring error from operational data. The results are further verified by
application of advanced modelling using atmospheric data from a numerical
weather prediction model and tested for actual overhead lines. The proposed
hoarfrost model confirms its feasibility in identifying of observed corona loss
events due to hoarfrost.

Keywords: Corona losses
Hoar frost
Operational analysis
Modelling

1 Introduction

Hoarfrost events on the conductors of overhead lines do not occur as often as e.g. rain,
but they result in much higher corona losses and may exist longer than other weather-
related events. Hoarfrost on conductors is a reason for high power losses due to corona
discharge activity from the hoarfrost needles, see illustration of hoarfrost appearance in
Fig. 1. For example, the levels of corona losses measured at 400 kV line in Sweden can
be estimated as up to 84 kW/km [1].
As corona losses increase exponentially with voltage it may be profitable to
decrease the system voltage, apply different de-icing techniques (e.g. ice melting) or
even to switch off the lines affected by high corona losses [1]. However, before the
application of any countermeasures, prediction of hoarfrost events and reliable esti-
mation of corona losses based on analysis of operational data is needed. This is a
challenge due to the high demand on the precision of operational measuring equipment,
the calculation of the varying transmission losses and input meteorological data. This
challenge was the driving force for this investigation intended to establish a practical
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 458–471, 2020.
https://doi.org/10.1007/978-3-030-31676-1_44
 Identification of Hoarfrost Corona Losses Events 459

methodology applicable for identification of hoarfrost events and calculation of

respective hoarfrost corona losses.
Modelling of hoarfrost forced by atmospheric data given from a numerical weather
prediction model is proposed as an integral part of a forecasting system for identifi-
cation of hoarfrost events on transmission lines.

Fig. 1. Example of visual appearance of hoar-frost growth on the energized conductors (adopted
from [1]).

2 Identification of Corona Losses

2.1 General
Detailed description of the innovative methodology for the identification of corona
loses based on operation measurements and its verification is presented in [2]. In this
paper this methodology is summarized only briefly.

2.2 Available Measuring Data

Two overhead lines (OHL) of Norwegian TSO Statnett 300 kV and 420 kV were
selected for this study. The measuring data was collected during more than a 2-year
period from January 2015 to June 2017. For both OHL active power (MW), reactive
power (MVAr), operating bus bar voltage (kV) and load current (A) in rms values were
measured at both ends of lines with 1-min interval. These parameters are further
referred as operational data.
In Fig. 2 a simplified schematic diagram for measuring and processing of opera-
tional data is presented. Some disturbances with high amplitude peaks were identified
and after additional analysis were explained by not fully synchronized measurements at
line ends. The application of classical data processing techniques (filtering) smoothed
out the high-frequency fluctuations and the data could be presented in a more con-
venient way. This was achieved by using a 25-point moving average filter. In general,
460 I. Gutman et al.

the smoothing of the moving average filter decreases the amplitude of the random noise
due to uncertainties in measurements.

Fig. 2. Schematic diagram of measuring and processing of operational data.

2.3 Basic Principle for Calculation of Corona Losses

The basic principle is that corona losses are determined by subtracting resistive losses
from the total active losses calculated as the difference between the values of active
power P1 and P2 measured at both ends of an OHL. The actual total instantaneous
active power P at each end of a line is calculated by a data processing system as the
sum of the measured active powers in each three phases. Considering the presence of a
measuring error, the corona losses can be calculated as follows:
(
DP
 ¼ P1  P2  3
R
I2 ¼ Pk þ d qffiffiffiffiffiffi
2
ð1Þ
R ¼ R0
1 þ 0:004
Tamb  20 þ 8:3
j2
300
F
;

where 3
R
I22 – three-phase resistive losses, MW; R – line resistance, Ohm; I2 – load
current measured at one of the line ends, A; Pk – corona losses, MW; d – measuring
error consisting of systematic and stochastic (random) components; R0 –resistance of
conductor bundle at 20 °C, Ohm; Tamb – measured ambient temperature, °C; j – current
density calculated from operational measurements, A/mm2; F – cross-section area of
conductor bundle, mm2.
 Identification of Hoarfrost Corona Losses Events 461

2.4 Estimation of Measuring Errors

Estimation of measuring errors is an essential part of the proposed methodology. For
the subtracting of the measuring error d from (1) it is proposed to find the relation
between the measured and calculated corona losses for the so-called “reference peri-
ods” with fair weather. During such periods a conductor is dry and frost-free and, thus,
the corona losses are at their minimum. For this case corona losses can be calculated
using Tihodeev’s equation from [3], in kW/km:
 !
U0 0:5
U
B

Pk0 ðU Þ ¼ 3
A
U02
e
106 ; ð2Þ

where U – measured phase-to-earth voltage, kVrms; U0 – corona inception voltage

dependent on conductor configuration and OHL design, kVrms; B = 9.1; A – weather
dependent coefficient, A = 0.5 (for fair weather). Then for any identified reference day
with fair weather a measuring error is calculated as follows:

d ¼ ðP1  P2 Þ  3
R
I22  Pk0 ð3Þ

Measuring errors are always divided into two components: systematic error and
random error. Systematic errors are associated with measuring instruments or envi-
ronmental interference. The random error is an intrinsic feature of any measurement. In
our case it is caused by inherent fluctuations in the readings due to not fully syn-
chronized measurements, i.e. when the time difference between the measurements at
the line ends is not zero but differs by a few seconds. The measuring precision for field
measurements can be increased by removing random errors using the appropriate
statistical method. Details on the evaluation of measuring errors are in [2].
The random (stochastic) error can be characterized in statistical form by the stan-
dard deviation r as follows:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 XN  2
r¼
d  d ; ð4Þ
N 1

where N represents number of measurements per day, i.e. N = 1440. In our case the
standard deviation r is 2.46 MW.
The standard deviation r plays a central role in the methodology, because it is later
used to identify the time intervals when corona losses increase significantly most
probably due to the change of weather conditions.

2.5 Principles of Identification of Events with High Corona Losses

Weather conditions, which differ from fair the corona losses are calculated by sub-
tracting the systematic error:
462 I. Gutman et al.


Pk ¼ ðP1  P2 Þ  3
R
I22  d; ð5Þ

It should be noted, however, that the corona losses calculated by (5) are distributed
statistically, thus they still include the random component characterized by standard
deviation r. Based on that, the results can be presented as a “band”, i.e.
[Pk0  r; Pk0 þ r], which is illustrated in Fig. 3 using as an example data for April
2017.

Fig. 3. Identified periods with hoarfrost corona losses (red circled).

Using the values Pk0 and r calculated for the reference day with fair weather, the
level of losses ðPk [ Pk0 þ rÞ is 2.62 MW that approximately corresponds to corona
losses at wet snow conditions. In case of hoarfrost the calculated corona losses are
5.67 MW and that is visualized by the brown coloured curve in Fig. 3 together with the
closely placed green curve corresponding to the level of Pk0 þ 2
r ¼ 5:1MW.
It is clearly seen that the corona losses corresponding to the hoarfrost conditions are
approximately at the same level as calculated using Pk0 þ 2
r and this is valid for the
whole measuring period. This observation is considered as the most important finding
from the analysis and thus it is proposed to use this criterion (Pk > Pk0 þ 2
r) further
to identify the events with hoarfrost corona. Using the criterion as above the general
criterion will be as follows:

Pk [ Pk0 þ k
r; ð6Þ

where k is a coefficient related to the type of weather affecting the corona losses: the
proposal is to use k ¼ 1 for rain/wet snow conditions; and k ¼ 2 for hoarfrost
conditions.
The simplified flow-chart summarizing the described methodology is shown as a
step-by-step procedure in Fig. 4, adopted from [2].
 Identification of Hoarfrost Corona Losses Events 463

Fig. 4. Flow-chart diagram for the step-by-step application of the proposed [2].
464 I. Gutman et al.

2.6 Summary from Calculation of Hoarfrost Corona Losses Based

on Measurements
The results from calculations of corona losses based on 2-year measurements for
420 kV OHL are summarized in Table 1.
Actual corona losses identified at the most probable hoarfrost conditions during the
analyzed period of measurements vary between 6 MW and 12 MW providing an
intensity variation between 52 kW/km and 104 kW/km. These values are in good
agreement with STRI’s (Sweden) earlier measurements on an outdoor test span where
the maximum intensity about 100 kW/km was observed [1].

Table 1. Summary of results obtained from calculations of the corona losses related to hoar-
frost events on 420 kV OHL.
Voltage, kV Year Month Day(s) Maximum
level of
corona losses
MW kW/km
420 2015 Nov. 17, 29 8 70
Dec. 4, 27 11 96
2016 Jan. 8, 23 7 62
Feb. 7, 8 8 69
Nov. 4, 17, 18, 19, 20, 22 12 104
Dec. 10 6 52
2017 Feb. 1, 2, 3, 5, 6, 21, 22, 27, 28 8 70
Apr. 11, 12, 15,16, 23 7 60

3 Modelling of Hoarfrost on Overhead Transmission Lines

3.1 General
For the analysis and the prediction of significant hoarfrost events the proposed hoar-
frost model is a time-dependent numerical model which calculates the thickness,
growth rate and layer density of hoarfrost on a conductor.
In the model the deposition of water vapour into solid form on the conductor is
based on the detailed calculation of the temperature and energy balance at the surface
of the conductor at each time step. The model considers the removal of ice mass
through melting and sublimation.

3.2 Model Description

The core of the hoarfrost model is based on the following equations describing the
deposition and sublimation rate at the conductor as well as heat balance at the
conductor/ice surface:
 Identification of Hoarfrost Corona Losses Events 465


I ¼ c0:62
h
ð es  ea Þ
p
ca ; ð7Þ
qe þ qs þ qi ¼ qc þ qeff

where cp is the specific heat of air, pa the atmospheric pressure and h the convective
transfer coefficient; es and ea are the water vapor pressures over ice and in air
respectively; qe is the heat flux due to the release of latent heat of sublimation as ice is
deposited on the conductor, directly from water vapor; qc is the convective heat flux
both wind- and surface temperature difference driven; qeff is the radiative fluxes which
are most important when the sky is clear; qs is incoming solar radiation; qi is the
conductive heat flux due to joule heating.

3.3 Applicability of the Model

A simplified version of Makkonen’s model [4] is considered as an appropriate starting
point for an analysis of hoarfrost events which may lead to significant corona losses. To
simplify and speed up the running of the hoarfrost model, the current study employs an
approach where joule heating of the conductor is neglected. Free convection, i.e.
convective processes at the conductor surface during no wind is also neglected.

3.4 Input Data

Typically, the model uses meteorological data saved at specific weather stations.
However, observations from such weather stations are typically not representative for
larger regions during hoarfrost conditions associated with temperature inversions, weak
winds, local fog or partially clear skies. This is a case for OHL traversing complex
terrain where the observation sites are few and far apart, i.e. typical Norwegian case.
Consequently, input data has been retrieved from a state-of-the-art numerical weather
model. The input data includes a description of the cloud cover, relative humidity, wind
speed and air temperature, as well as material constants describing the specific heat
capacity of the conductor.

3.5 The Numerical Weather Prediction Model

Data from AROME, a numerical weather prediction (NWP) model has been used
instead as an input in the hoarfrost model. The AROME model is a non-hydrostatic,
convective scale NWP model that has been run at the Norwegian Meteorological
Institute (MET Norway) and the Swedish Meteorological and Hydrological institute
(SMHI) since 2013.
The model is run at a domain covering entire Scandinavia as well as adjacent areas
at a horizontal grid spacing of 2.5 km (Fig. 5).
466 I. Gutman et al.

Fig. 5. Domain for the MetCoOp system, run operationally by MET Norway, SMHI and FMI.

There are 65 levels available in the vertical direction, with the lowest layer at
12.5 m above surface and the uppermost at approximately 33 km. A 66-hour forecast
with 10 ensemble members is initialized four times a day, forced at the boundaries by
data from the global NWP model run at the European Centre for Medium Range
Weather Forecasts (ECMWF).

3.6 Data Provided by the Model

Hoarfrost accumulation was calculated on a diurnal basis at the positions corresponding
to every fourth OHL tower, i.e. 90 positions for the 420 kV OHL in total. The hourly
hoarfrost accumulation was then aggregated for all towers. Corona losses calculated
from operational data (red circles) and modelled hoarfrost accretion on 420 kV OHL
during February 2016/2017 are presented in Fig. 6.
Hoarfrost accretion includes only hoarfrost accumulation and not its removal, and
in this respect the figure can be interpreted as presenting a hoarfrost event as “poten-
tial”. Weather parameters obtained from the model for the same February 2017 are
shown in Figs. 7 and 8. Hoarfrost events identified from operational data are during the
periods of the 1st–6th, 21st–22nd and 27th–29th of February 2017, and the model pre-
dicted hoarfrost for the same periods, although with a very weak signal for the 21st–
22nd (Fig. 6). Between the 6th and 10th of February 2017 the model predicted intensive
hoarfrost accretion, but there were no corresponding corona events identified from
operational data. When considering the meteorological data from AROME (Figs. 7 and
8) there is a notable difference between the modelled hoarfrost events with and without
 Identification of Hoarfrost Corona Losses Events 467

corresponding corona loss. While the temperature during the 1–6 February 2017 is
between −5 and −2 °C, the temperatures during the period of 6th–10th of February are
between −15 °C and −10 °C, and during the other identified corona events the tem-
peratures are close to 0 °C. It can be speculated that higher temperatures may lead to a
change in the structure of hoarfrost needles (make them thinner and sharper), thus
forcing them to produce more corona discharges and higher losses.

Fig. 6. 420 kV OHL: observed corona losses (upper figure, red circles), modelled hoarfrost
accretion (lower figure) for February 2017; shaded areas reflect a hit rate.

The comparison of dates/times of identified hoarfrost corona events together with

modelled hoarfrost accretion and the meteorological input data from the AROME model
indicated that in identified corona events occurred when AROME predicted relatively
high temperatures. This indication is in line with previous studies from Sweden [1] when
468 I. Gutman et al.

during measurements on a 250 km long 400 kV OHL high corona loss at conductor
temperatures were observed in the intervals −4 °C to 2 °C and −20 °C to −15 °C and
air temperature from −3 °C to −5 °C and from −20 °C to −15 °C. A possible expla-
nation is the influence of air temperature on the development/changing of structure of ice
crystals.

Fig. 7. Modelled temperature and wind speed from grid points along 420 kV line for February
2017. Black line is median, while the shaded area shows 10 percentile to 90 percentile of the
values.
 Identification of Hoarfrost Corona Losses Events 469

Fig. 8. Modelled relative humidity and cloud cover from grid points along 420 kV line for
February 2017. Black line is median, while the shaded area shows 10 percentile to 90 percentile
of the values.

4 Discussion

In Table 2 the dates for hoarfrost events identified from analysis of operational data and
events obtained from simulated hoarfrost accumulation on 420 kV OHL are compared.
Even in rather simplified form, the hoarfrost model shows good ability in identi-
fying observed corona loss events due to hoarfrost. While the simplified hoarfrost
model fails to simulate hoarfrost for only 5 of the 36 identified hoarfrost corona events
(the average hit rate is 31/36 = 86%), there are many occasions when the model
predicts hoarfrost without corresponding events identified from operational data. A part
of the discrepancy between the corona events identified from operational data and
modelled hoarfrost events may be caused by some hoarfrost events that were not being
considered as actual high losses according to the proposed criterion (6). Such losses can
470 I. Gutman et al.

be attributed to wet snow events. The correlation between observed monthly corona
losses and monthly simulated hoarfrost accretion is generally confirmed on actual OHL
420 kV (and even on 300 kV OHL not presented in this paper). The total monthly
corona losses calculated from measurements include both wet snow and hoar-frost
corona, and the months with measured corona losses between 0.4 and 0.7 GWh and
near zero hoarfrost accretion are likely to have a high contribution of corona loss due to
wet snow or rain.
Another possible reason is that calculations of hoarfrost are quite sensitive to cloud
cover, relative humidity and temperature biases in the input data obtained from the
AROME model.
Also, the simplifications employed in the hoarfrost model, especially exclusion of
joule heating, may explain a large part of the apparent difference between prediction by
modeling and by using operational data. Joule heating may impact the energy balance
on the conductor surface and the accretion of hoarfrost [5]. Due to increasing of the
temperature on the conductor surface hoarfrost accretion may melt off or simply not
formed at slightly negative ambient temperatures. This could be a reason why the
simplified model over-predicts the hoarfrost accretion, especially at temperatures near
freezing point. Joule heating will also affect hoarfrost accretion at lower temperatures;
the air at the conductor will be heated by the sensible heat flux, contributing to keeping
the air at the conductor surface above the saturation point and requiring significant
radiative cooling to bring the air to saturation.

Table 2. Days with monitored hoarfrost corona and modelled hoarfrost accretion on 420 kV
OHL.
Month Days identified Days identified by Hit Days with modelled hoar-
from modelled hoarfrost rate frost accretion (no corona
operational data accretion losses from measurements)
Nov.’15 17, 29 17, 29 2/2 15, 16, 18, 20, 23
Dec.’15 4, 27 4, 27 2/2 2, 16, 28–31
Jan. ‘16 8, 23 8, 23 2/2 1–17, 19, 27, 29, 31
Feb. ‘16 7, 8 7, 8 2/2 1,2, 5, 6, 14, 17–21, 29
Mar.’16 NaN 1–9
Oct. ‘16 NaN 16, 21–25
Nov.’16 4, 17–20, 22 17–22 5/5 8, 9, 12
Dec.’16 10 10 1/1 20, 23
Jan. ‘17 NaN 1, 3, 4, 9, 10, 14–15, 28–30
Feb. ‘17 1, 2, 3, 5, 6, 21, 1–6, 22, 27, 28 8/9 7–10, 25
22, 27, 28
Mar.’17 NaN 3–8, 19
 Identification of Hoarfrost Corona Losses Events 471

5 Conclusion

A comparison between dates of events with corona losses identified from operational
data and from the simplified hoarfrost model indicated that these events can be pre-
dicted. Although a hit rate of 86% between modelling and measurements is promising,
the model predicts several occurrences of hoarfrost accretion without corresponding
losses according to operational measurements. This might be because the simplified
model was used, and important updates of this simplified hoarfrost model are needed
for its further development. A more comprehensive version of the weather modeling
would be needed for further implementation in an operational pilot forecasting system.

Acknowledgment. The authors wish to acknowledge the Nordic project Frontlines No.
245370 for the support of this work.

References
1. Sollerkvist, F.J., Maxwell, A., Rouden, K., Ohnstad, T.M.: Evaluation, verification and
operational supervision of corona losses in Sweden. IEEE Trans. Power Deliv. 22(2), 1210–
1217 (2007)
2. Solovyev, Y., Gutman, I., Byrkjedal, Ø., Thorsteinsson, B.: Practical methodology for
calculation of corona losses induced by hoarfrost using operational measurements. Cigré Sci.
Eng. 10, 77–87 (2018)
3. Tikhodeev, N.N.: Mitigation of corona losses on EHV overhead lines through voltage control.
In: Proceedings IEEE St. Petersburg Chapter, St. Petersburg, Russia, pp. 1–10 (2000)
4. Makkonen, L.: A model of hoarfrost formation. In: Proceedings of the 12th International
Workshop on Atmospheric Icing on Structures. IWAIS, Yokohama (2007)
5. Makkonen, L.: A model of hoarfrost formation on a cable. Cold Regions Sci. Technol. 85,
256–260 (2013)
 Influence of Oil Temperature on Repeatability
of Measurements in Frequency Response
Analysis of Power Transformers

Satoru Miyazaki1(&), Yoshinobu Mizutani1,2, Yoshihiro Wada2,

and Chikara Hayashida2
1
Central Research Institute of Electric Power Industry,
2-6-1, Nagasaka, Yokosuka-shi, Kanagawa, Japan
m-satoru@criepi.denken.or.jp
2
Kyuhen Co. Inc., 2-2-1, Hanamigahama, Fukutsu-shi, Fukuoka, Japan

Abstract. Power transformers are one of the key apparatus in electric power
networks. To ensure the reliability of power transformers, the diagnosis of
mechanical faults, such as the deformation and displacement of windings, is
important. Frequency response analysis (FRA) has attracted considerable
attention as a diagnostic method for mechanical faults of power-transformer
windings. In FRA, the frequency responses of power transformers, namely,
transfer functions (TFs), are measured in a wide frequency range. Faults are
detected by detecting changes in TFs compared with reference data measured
when the power transformers were operating normally. Therefore, it is important
to ensure the repeatability of measurements. In this study, the TFs of two power
transformers are measured shortly before and after a temperature-rise test and a
subsequent cooling stage. Changes in TFs due to a change in oil temperature are
found at frequencies higher than 400 kHz. Furthermore, the influence of the oil
temperature on the repeatability of the measurements in FRA is objectively
evaluated by employing a numerical index, which represents the identification
degree of two compared TFs. It is found that the influence of the oil temperature
is negligible when the average oil temperature is less than 50° C.

Keywords: Power transformer
Frequency response analysis
Repeatability

of measurements

1 Introduction

Power transformers are one of the key apparatus in electric power networks. To ensure the
high reliability of power transformers, the diagnosis of mechanical faults, such as the
deformation and displacement of windings, is important [1]. Even if the mechanical faults
of windings do not necessarily lead to immediate failure of a transformer, its ability to
withstand future mechanical and dielectric stresses may be greatly reduced [2, 3].
Frequency response analysis (FRA) [4–6] has attracted considerable attention as a
diagnostic method for mechanical faults of power-transformer windings. In FRA,
frequency responses, namely, transfer functions (TFs), are measured. A TF is defined
by the amplitude ratio and the phase difference between the voltages measured at two

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 472–480, 2020.
https://doi.org/10.1007/978-3-030-31676-1_45
 Influence of Oil Temperature on Repeatability of Measurements 473

terminals of the tested transformer over a range of frequencies when one of the ter-
minals is excited by a voltage source [4, 5]. TFs are determined by the electrical
parameters of power transformers, such as the inductances of windings, the capaci-
tances between windings, and the capacitances between windings and the ground.
These electrical parameters change as a result of mechanical and electrical faults, and
hence, the TFs change. FRA has the potential to detect these mechanical faults with
high sensitivity compared with other conventional diagnostic methods, such as short-
circuit impedance measurement [7].
Basically, in FRA, faults are identified by detecting changes in TFs compared with
reference data measured when the power transformers were operating normally [2].
Therefore, it is important to ensure the repeatability of measurements so that the
measured TFs are identical as long as a transformer does not have any faults.
For example, possible factors affecting the repeatability of measurements in FRA
are as follow.
• Arrangement of measuring cables
• Employing different connection schemes for measurement cables [8]
• Contact resistance between measuring cable and transformer terminal
• Apparatus connected to tested transformer terminal (e.g., GIS, cable, etc.)
• Temperatures of materials inside transformer (windings, insulation oil, insulation
papers, press boards, etc.)
The influence of the temperature on the repeatability of measurements is an
important issue. However, it is difficult to discuss the influences of temperatures of
windings, insulation papers, and press boards separately because they are difficult to
measure. The temperature of oil in power transformers is a good indicator of their
temperature and easy to measure. Under such circumstances, CRIEPI and Kyuhen Co.,
Inc., have studied the influence of the oil temperature on the repeatability of mea-
surements in FRA [9, 10]. In this paper, a summary of this study is given.

2 Experimental Method

Table 1 shows the specifications of the tested transformers, which are newly estab-
lished power transformers. Temperature-rise tests are conducted on the transformers as
factory acceptance tests. The TFs of transformer A are measured shortly before and

Table 1. Specification of tested transformers

Transformer A Transformer B
Rated voltage 66/6.9 kV 66/6.9 kV
Rated capacity 20 MVA 15 MVA
Vector group Yy0+d Yy0+d
Manufacture year 2013 2014
Measurement • Cooling stage • Cooling stage
• Just before and after temperature-rise test
474 S. Miyazaki et al.

(i) 20Hz~1MHz (ii) 100kHz~1MHz

(a) primary winding, W-V phase (open-circuit measurement)

(i) 20Hz~1MHz (ii) 100kHz~1MHz

(b) primary winding, W-V phase (short-circuit measurement)

(i) 20Hz~1MHz (ii) 100kHz~1MHz

(c) secondary winding, w-v phase (open-circuit measurement)

Fig. 1. Transfer functions measured shortly before and after temperature-rise test and evaluated
CCF*.
 Influence of Oil Temperature on Repeatability of Measurements 475

(i) 20Hz~1MHz (ii) 100kHz~1MHz

(d) secondary winding, w-v phase (short-circuit measurement)

Fig. 1. (continued)

after the temperature-rise test. In this measurement, the open-circuit and short-circuit
measurements of the high-voltage and low-voltage terminals specified in the IEC
standard [4] are employed. Furthermore, the TFs of transformers A and B are
repeatedly measured during the cooling stage after their temperature-rise test. In the
measurements during the cooling stage, only the TFs of the HV winding are measured
and the arrangement of the measuring cables is not changed. Therefore, the changes in
the TF are entirely due to the temperature rise of the oil.

3 Experimental Result

Figure 1 shows the TFs of transformer A measured shortly before and after the
temperature-rise test. Here, the oil temperature is defined as the average oil temperature
in the upper and lower parts of the tested transformer. The figures on the left and right
sides of Fig. 1 show the TFs in the frequency range from 20 Hz to 1 MHz and from
100 kHz to 1 MHz, respectively. CCF* displayed in Fig. 1 will be explained in
Sect. 4.2. The resonant frequencies higher than 400 kHz shift toward low-frequencies
with the temperature rise of oil.
Figure 2 shows the TFs of transformers A and B measured during after the cooling
stage of the temperature-rise test. The resonant frequencies higher than 400 kHz shift
toward lower frequencies when the average oil temperature increases.
476 S. Miyazaki et al.

(i) open-circuit measurement (ii) short-circuit measurement

(a) Transformer A

(i) open-circuit measurement (ii) short-circuit measurement

(b) Transformer B

Fig. 2. Transfer functions measured during cooling stage after temperature-rise test and evaluated
CCF*

4 Discussions
4.1 Mechanism of Changes in Transfer Functions Due to Difference
in Oil Temperature
When the oil temperature rises, the thermal expansion of windings, the variation of the
permittivity of oil, and the variation of the permittivity of oil-impregnated insulation
papers and press boards will occur. In this section, their influence on the TFs in FRA is
discussed.
1. Thermal expansion of windings
The thermal expansion coefficient of copper is 16.8  10−6/°C. When the winding
temperature rises by 50 °C, the thermal expansion ratio is 0.084%. This may result
in a 0.084% increase in winding radius. However, such a small change will hardly
affect the inductance or capacitance, and hence the TFs.
2. Variation of permittivity of oil
Figure 3 shows the variation of the permittivity of oil with its temperature [11]. The
permittivity is lower at higher temperatures. When the permittivity decreases, the
 Influence of Oil Temperature on Repeatability of Measurements 477

resonant frequencies increase. However, the shifts of the resonant frequencies in

Figs. 2 and 3 are in opposite directions. This indicates that the influence of the
variation of the permittivity of oil due to the temperature rise should be negligible in
FRA.
3. Variation of permittivity of oil-impregnated insulation papers and press boards
Figure 4 shows the variation of the permittivity of oil-impregnated press boards
[12]. Note that the numbers in Fig. 4 are the moisture contents of the papers relative
to the dry weight of the press boards in %. The permittivity is higher at higher
temperatures. When the permittivity increases, the resonant frequencies decrease.
This direction of the shift of the resonant frequencies agrees with the measurements.
Furthermore, the variation of the permittivity of the insulation papers and press
boards results in the variation of the inter-turn capacitances and the capacitances
between disks. These capacitances determine resonant frequencies in the high fre-
quency range [13]. This is also in agreement with the measured results.

Fig. 3. Variation of permittivity of insulation oil due to temperature [11].

Fig. 4. Variation of permittivity of oil- impregnated press board due to temperature [12].
478 S. Miyazaki et al.

From the above discussion, it can be concluded that the changes in TFs due to the
temperature rise of oil is due to the variation of the permittivity of oil-impregnated
insulation papers and press boards.

4.2 Influence of Oil Temperature on Diagnosis of Mechanical Faults

of Transformer Winding by FRA
Numerical indices [14] that evaluate the identification degree of two compared TFs are
useful for quantitative evaluation of the influence of the oil temperature on the
mechanical-fault diagnosis of transformer windings by FRA. For a quantitative dis-
cussion, a criterion for the diagnosis is necessary. Among various numerical indices, a
tentative criterion employing LCCF* to detect the mechanical faults of transformer
windings has been proposed [15, 16]. LCCF* is obtained as follows. Cross-correlation
factor (CCF) [14] is repeatedly calculated for a frequency window while shifting it by
one measurement frequency step, and a curve of CCF is drawn as a function of
frequency. A linear transformation is then performed using Eq. (1) to obtain CCF* for
the compared TFs in the same figure, where LCCF* is the minimum of CCF* over the
measured frequency band.

CCF  ¼ ð1  CCF Þ  100 ð1Þ

LCCF* has been successfully used to separate the measured TFs of normal
transformers and those of transformers whose windings are deformed or axially dis-
placed [15, 16]. On the basis of this discussion, a tentative criterion of LCCF* = −5 for
a window width of 1 octave has been proposed for detecting mechanical faults; an
LCCF* lower than −5 indicates that a transformer winding has a mechanical fault and
an LCCF* higher than −5 indicates that the transformer is normal. Here, LCCF* is zero
when two compared TFs are identical.
In Figs. 1 and 2, the calculated CCF* curves are also drawn. In the calculation of
LCCF*, the evaluated TFs are compared with reference data, which are the TFs
measured when the oil is at its lowest temperature. Here, for example, the calculated
CCF* for the frequency window from 10 kHz to 20 kHz is plotted at 10 kHz. In
Fig. 5, the calculated LCCF* is shown as a function of oil temperature. As discussed in
Sect. 4.1, the change in TFs due to the temperature rise of oil is due to the variation of
the permittivity of oil-impregnated insulation papers and press boards. The variation of
the permittivity of oil-impregnated press boards is nonlinear with the temperature as
shown in Fig. 4. Therefore, it is better to focus on the temperature itself rather than the
temperature difference. Note that the oil temperatures are 28 and 33 °C when the
reference data are measured, which are not particularly high considering the climate of
Japan. In the cooling stage after the temperature-rise test, LCCF* decreases with
increasing oil temperature. In Fig. 5, the calculated LCCF* is higher than −5 when the
oil temperature is lower than 50 °C. Therefore, it can be concluded that the influence of
the oil temperature is not so significant when the oil temperature is less than 50 °C.
From the above discussion, the measurement of TFs when the oil temperature is
less than 50 °C is recommended. For example, the oil temperature shortly after
 Influence of Oil Temperature on Repeatability of Measurements 479

stopping the operation of a power transformer may be higher than 50 °C. Therefore,
measurement after several hours, when oil and other materials are sufficiently cool, is
recommended.

Fig. 5. Variation of LCCF* due to average oil temperature.

5 Conclusion

In this study, the TFs of power transformers are measured shortly before and after
temperature-rise tests and during the subsequent cooling stage. The measured TFs
change with the temperature rise of oil. This is considered to be due to the variation of
the permittivity of oil-impregnated insulation papers and press boards. On the basis of a
discussion using a numerical index, the influence of the oil temperature on the
repeatability of measurements in FRA should be negligible when the average oil
temperature is less than 50 °C.

References
1. Tenbohlen, S., Coenen, S., Djamali, M., Muller, A., Samimi, M.H., Siegel, M.: Diagnostic
measurements for power transformers. Energies 9, 347 (2016)
2. Christian, J., Feser, K.: Procedures for detecting winding displacements in power
transformers by the transfer function method. IEEE Trans. Power Deliv. 19, 214–220 (2004)
3. McNutt, W.J., Johnson, W.M., Nelson, R.A.: Power transformer short-circuit strength—
requirements, design, and demonstration. IEEE Trans. Power Appar. Syst. PAS-89, 1955–
1969 (1970)
4. IEC60076-18: Measurement of frequency response, ED. 1, IEC Std. (2012)
5. IEEE guide for the application and interpretation of frequency response analysis for oil-
immersed transformers. IEEE Std. C57.149-2012, pp. 1-72 (2013)
480 S. Miyazaki et al.

6. Frequency Response Analysis on Winding Deformation of Power Transformers. The

Electric Power Industry Standard of People’s Republic of China, Std. DL/T911-204,
ICS27.100, F24, Document No. 15182-2005 (2005)
7. Miyazaki, S., Mizutani, Y., Suzuki, H., Ichikawa, M.: Identification of trouble of transformer
by frequency response analysis. IEEE J. Trans. PE 129, 517–525 (2009). (in Japanese)
8. Miyazaki, S., Mizutani, Y., Tahir, M., Tenbohlen, S.: Influence of employing different
measuring systems on measurement repeatability in frequency response analyses of power
transformers. IEEE Electr. Insul. Mag. 35(2), 27–33 (2019)
9. Miyazaki, S., Mizutani, Y., Wada, Y., Hayashida, C.: Influence of oil temperature of power
transformer on transfer function in frequency response analysis. In: Conference on Static
Apparatus of IEEE, Japan, SA-14-096, Tokyo (2014). (in Japanese)
10. Miyazaki, S., Mizutani, Y., Taguchi, A., Murakami, J., Tsuji, N., Takashima, M., Kato, O.:
Proposal of diagnosis criterion of abnormality of transformer winding by Frequency
Response Analysis (FRA). CRIEPI report H14010 (2015). (in Japanese)
11. The Japan Petroleum Institute: Electrical Insulation Oil Hand Book, Part 1, Chapter 3,
Section 2.6, pp. 89–911987
12. Ishii, T., Shirai, M.: Effect of moisture adsorption mechanism on dielectric properties of
pressboard. J. Inst. Electr. Eng. Jpn. 87(11), 2184–2193 (1967). (in Japanese)
13. Miyazaki, S., Mizutani, Y., Wada, Y., Hayashida C.: Abnormality diagnosis of transformer
winding by Frequency Response Analysis (FRA) using circuit model. In: Proceeding of
2012 International Conference on Condition Monitoring and Diagnosis, no. A-82 (2012)
14. Picher, P., Tenbohlen, S., Lachman, M., Scardazzi, A., Patel, P.: Current state of transformer
FRA interpretation. Procedia Eng. 202, 3–12 (2017)
15. Miyazaki, S., Mizutani, Y., Taguchi, A., Murakami, J., Tsuji, N., Takashima, M., Kato, O.:
Proposal of objective criterion in diagnosis of abnormalities of power-transformer winding
by Frequency Response Analysis. In: Proceeding of 2016 International Conference on
Condition Monitoring and Diagnosis, pp. 74–77 (2016)
16. Miyazaki, S., Mizutani, Y., Taguchi, A., Murakami, J., Tsuji, N., Takashima, M., Kato, O.:
Diagnosis criterion of abnormality of transformer winding by Frequency Response Analysis
(FRA) - detailed diagnosis of slight abnormality. Electr. Eng. Jpn. 201, 25–34 (2017)
 Detection of CF4 Gas Using
a Nanomaterial-Based Gas Sensor
Fabricated by Dielectrophoresis

Nisarut Phansiri1(&), Daichi Maenosono1, Takumi Furumoto1,

Hidefumi Sato2, Michihiko Nakano1, and Junya Suehiro1
1
Kyushu University, Fukuoka 819-0395, Japan
artto@hv.ees.kyushu-u.ac.jp
2
Kyushu Electric Power CO., INC., Fukuoka 819-0395, Japan

Abstract. Partial and arc discharge in gas insulated switchgear (GIS), such as a
switching component or circuit breaker, can lead decomposition gas generation
of sulphur hexafluoride (SF6), which is generally used as an electrical insulation
medium. The authors had demonstrated that SF6 decomposition products, such
as HF, SO2, SOF2, can be detected by using a carbon nanotube (CNT) gas
sensor, which was fabricated by dielectrophoresis. In this study, we selected
carbon tetrafluoride (CF4) as a target decomposition gas to be detected by
nanomaterial-based gas sensors. Because CF4 is hard to be removed by using a
molecular sieve or absorbent, the decomposition product accumulates in a long
time and can be used for GIS diagnosis. Three kinds of semiconducting nano-
material, carbon nanotube, SnO2 nanoparticle, ZnO nanowires, were integrated
on a microelectrode by dielectrophoresis respectively to fabricate a gas sensor. It
was found that SnO2 gas sensor showed the highest response to CF4 gas at 1%
concentration in SF6. The conductance of the SnO2 sensor gradually decreased
with elapsed time after exposure to the CF4 gas depending on the operating
temperature.

Keywords: Gas Insulated Switchgear
Decomposition of SF6

Nanomaterial-based gas sensor

1 Introduction

Gas Insulated Switchgear (GIS) is well-known and essential equipment for the power
distribution system. Generally, sulphur hexafluoride (SF6), is used for gas insulation
due to its excellent dielectric strength and non-toxic property. However, partial and arc
discharge in the GIS system, such as a circuit breaker or switching component, can lead
to the decomposition of SF6. When the SF6 is exposed to electrical discharge, it will
dissociate and recombine to produce decomposition products, such as HF, SO2, CO2,
CF4, SiF4, SOF2, SO2F2, S2F10 and SF4 [1]. Semiconductor gas sensors have been
studied for detection of SF6 decomposition. Nanomaterial- based gas sensor is a good
choice due to its advantage such as real-time and on-site analysis, low cost, and short
response time.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 481–488, 2020.
https://doi.org/10.1007/978-3-030-31676-1_46
482 N. Phansiri et al.

The authors have previously demonstrated a diagnosis method of GIS based on

detection of partial discharge generated decomposition products of SF6. The decom-
position products such as HF, SO2, and SOF2 were successfully detected by using a
carbon nanotube (CNT) gas sensor fabricated by dielectrophoresis (DEP) [2, 3].
Recently, the Interesting byproduct is carbon tetrafluoride (CF4), which is a
perfluorocarbon (PFCs). CF4 is found when the reaction involves solid or polymeric,
plastic material base on carbon, insulators located in the GIS system. Its properties
include toxicity and a long lifetime [4]. Removing CF4 is difficult by using molecular
sieve because its effective molecular diameter is nearly same size as SF6. The moni-
toring process of SF6 byproducts is essential requirement for GIS system.
In the present study, we explored the application of the DEP-fabrication gas sensor
to CF4 detection, aiming the development of GIS diagnosis based on the sampling
method, As CF4 sensing transducer, several different kinds of the nanomaterial, such as
carbon nanotube (CNT), zinc oxide (ZnO) nanowire and tin oxide (SnO2) nanoparticle
were tested. These nanomaterials were trapped onto a microelectrode by DEP in order
to fabricate a nanomaterial-based gas sensor. The operating temperature dependency of
these gas sensors were also investigated.

2 Materials and Methods

2.1 Nanomaterial-Based Gas Sensor

Several metal oxide materials are suitable for gas detection, which provide the sensor with
the reducing or oxidizing gas. Tin dioxide (SnO2) and Zinc oxide (ZnO) are the com-
monly used material-base gas sensor. They are n-type granular material whose electrical
conductivity is dependent on the density of pre-adsorbed oxygen ions on its surface.
Metal oxide semiconductor (MOS) have to deal with the problem of low response at the
room temperature. Normally, MOS has to be heated by an additional heater for operation
while carbon nanotube (CNT) gives high and fast responses. Therefore, SnO2, ZnO, and
CNT will be investigated for the ability of gas detection [5–7].

2.2 DEP Process for Fabrication of Nanomaterial Gas Sensor

Properties of nanomaterial used as the gas sensor transducer are shown in Table 1. The
SnO2 nano-particles were suspended with the three different concentrations to inves-
tigate effects of the number of nanoparticles trapped on the microelectrode and the
name list of sensors are shown in Table 2.
A schematic diagram of fabrication setup of nanomaterial-based gas sensors is shown
in Fig. 1. The microelectrode had 4 electrodes fingers and 3 straight gaps. The electrode
fingers had 50 µm in length and a minimum clearance of 5 µm between each electrode
finger, the microelectrode was patterned on a glass substrate by using phot lithography
technique. The microelectrode was surrounded by a silicone rubber spacer to form a
sealed chamber in which nanoparticles suspension was stored. The nanoparticle sus-
pension was continuously fed into the sealed chamber from a peristatic pump at a flow
rate 0.8 ml/min. The DEP trapping of nanoparticles to the microelectrode was performed
 Detection of CF4 Gas Using a Nanomaterial-Based Gas Sensor 483

with an ac voltage of 100 kHz frequency and 20 Vpeak-peak of amplitude. The ac voltage
and the conductance between the electrodes were simultaneously measured by an
impedance analyzer controlled by PC. After 50 min, the DEP process was stopped and
the suspension liquid was gently evaporated at room temperature (RT). Scanning electron
microscope (SEM) was used for observation of nanomaterial trapped on the
microelectrode.

Table 1. Nanomaterials used as the gas sensor transducer.

Material Diameter Length
SnO2 (particle) 15.9–136.7 nm –
ZnO (wire) 50 nm 300 nm
CNT (tube) 1 nm 1 µm

Table 2. Name list and detail of fabricated sensors.

Sensor Material Medium Concentration
A-1s SnO2 Ethanol 1 mg/ml
A-2s SnO2 Ethanol 5 mg/ml
A-3s SnO2 Ethanol 20 mg/ml
B-1s ZnO Ethanol 50 µg/ml
C-1s CNT Water 0.1 mg/ml

Fig. 1. Schematic diagram of nanomaterial-based gas sensor fabrication setup by positive DEP
process.

2.3 Experimental Apparatus for Gas Sensor Performance

Figure 2 shows the schematic diagram of the experimental setup for investigation of
the sensing response of the nanomaterial-based gas sensors. The sensor was placed in a
484 N. Phansiri et al.

Fig. 2. Schematic diagram of experimental setup for gas sensor performance evaluation.

chamber made of stainless steel and the working temperature was controlled by a
ceramic heater. The sensor was energized by ac voltage (1 Vpeak-peak and 100 kHz of
frequency) and the sensor conductance was measured by a lock- in amplifier. The gas
flow rate was controlled by mass flow control unit. CF4 gas was diluted by SF6 gas to
1% concentration. After steady base line was attained by initialize gas, exposing gas
sequences were included the initialize gas (SF6) for time t1 (300 s), exposing the target
gas for time t2 (600 s) and backed to initialize gas for time t3 (600 s). The sensing
response was shown as the normalized values of conductance of sensor.

3 Results and Discussion

3.1 DEP Process

Figure 3 shows SnO2 nanoparticles trapped on the microelectrodes. The conductance
of the nanomaterial-based (SnO2) gas sensor in DEP process is shown in Fig. 4.
The DEP fabrication method enables the control of nanoparticle trapped on the
microelectrode by impedance measurement.

Fig. 3. SEM image of SnO2 nanoparticles trapped on the microelectrodes.

 Detection of CF4 Gas Using a Nanomaterial-Based Gas Sensor 485

3.2 Effect of Transducer Material and the Sensor Operating

Temperature
The response (normalized response, DG) of the sensor is defined by the following
equation

Ga  G0
DG ¼ ð1Þ
G0

where is G0 the conductance in initialize gas and Ga is the conductance in the presence
of the target gas. Response time is expressed in term of time (tR) needed for 90% of full
response [8].
Figure 5 shows CF4 response comparison of three gas sensors, which were made of
different nanomaterials. The highest response was obtained with the A-2s sensor, which
was made of SnO2 nanoparticles. The B-1s sensor (ZnO nanowire) response was about
30% of A-2s, while the C-1s sensor (CNT) showed little response. Operating tem-
perature dependency of these sensors are shown in Fig. 6. The temperature dependency
of A-2s and B-1s demonstrated a similarity. Their responses were the highest at room
temperature and decreased with the temperature below 200 °C. Above 200 °C, the
response increased back with the operating temperature. C-1 s sensor response became
the highest at 200 °C.
Figure 7 shows temporal variation of A-2s sensor response to CF4 gas. As previ-
ously shown in Fig. 6, the peak value of the sensor response became the highest at
room temperature. On the other hand, the sensor response time tended to become faster
at elevated temperature. Table 3 compares response times (tR) and peak values (pv) at
room temperature and 300 °C. Figure 7 also demonstrates that the sensor conductance
showed a peak value at elevated temperatures, while monotonously increased with time
at room temperature. This result implied CF4 sensing mechanism of SnO2 nanoparticle-
based sensor depended on the temperature. Working principle for gas detection by
using metal oxide material is explained in the simple case for n-type semiconducting
metal oxide; in the atmosphere [9–12]. In this study, the authors would focus on the
experimental results and comparison of several sensors used for CF4 gas detection. The
sensing mechanism would be discussed in the future works.

Fig. 4. Temporal change of the microelectrode conductance during DEP process.

486 N. Phansiri et al.

Fig. 5. Response of sensors against CF4 at room temperature of A-2s, B-1s, and C-1s.

Fig. 6. Response of sensors against CF4 at various temperatures of A-2s, B-1s, and C-1s.

3.3 Effect of Nanoparticle Suspension Concentration

In DEP fabrication process of nanomaterial-based gas sensor, the amount of nano-
material (transducer) trapped on the microelectrode can be controlled by the concen-
tration of nanomaterial suspension. We investigated how the SnO2 nanoparticle amount
influences the sensor response and the results is shown in Fig. 8. It was found that the
sensor response depended on the SnO2 nanoparticles concentration and became the
highest response at 5 mg/ml concentration. It was probably due to enhanced
agglomeration of nanoparticles which reduced effective gas adsorption sites on their
surface.
 Detection of CF4 Gas Using a Nanomaterial-Based Gas Sensor 487

Fig. 7. Effect of the temperature on temporal variation of the A-2s sensor response.

Table 3. Effect of the temperature on the response times tR and peak values pv of the A-2s
sensor response.
Temperature tR (s) pv
RT 230 0.37
300 °C 74 0.28

Fig. 8. Effect of SnO2 nanoparticle suspension concentration on the CF4 gas response.

4 Conclusion

Aiming to an application for GIS diagnosis, semiconducting nanomaterial-based gas

sensors were fabricated by DEP. SnO2 nanoparticles, ZnO nanowires, and carbon
nanotubes were tested as the sensing transducer. It was found that the SnO2
488 N. Phansiri et al.

nanoparticles-based sensor showed the highest response at room temperature and the
fastest response time at 300 °C. In the future, further improvements in the response
value and response time are necessary for practical application.

References
1. Koch, D.: SF6 Properties, in MV and HV Switchgear, The Cahiers Techniques no 188,
collection is part of the Schneider Electric’s Collection Technique (2003). http://www.
schneider-electric.com
2. Ding, W., Hayashi, R., Ochi, K., Suehiro, J., Imasaka, K., Hara, M.: Analysis of PD-
generated SF6 decomposition gases adsorbed on carbon nanotubes. IEEE Trans. Dielectr.
Electr. Insul. 13(6), 1200–1207 (2006)
3. Martin, Y., Li, Z., Tsutsumi, T., Shou, R., Nakano, M., Suehiro, J.: Detection of SF6
decomposition products generated by DC corona discharge using a carbon nanotube gas
sensor. IEEE Trans. Dielectr. Electr. Insul. 19(2), 671–676 (2012)
4. Beroual, A., Haddad, A.: Recent advances in the quest for a new insulation gas with a low
impact on the environment to replace sulfur hexafluoride (SF6) gas in high-voltage power
network applications. Energies 10, 1216 (2017)
5. Liu, X., Cheng, S., Liu, H., Hu, S., Zhang, D., Ning, H.: Review a survey on gas sensing
technology. Sensors 12, 9635–9665 (2012)
6. Wang, C., Yin, L., Zhang, L., Xiang, D., Gao, R.: Review metal oxide gas sensors:
sensitivity and influencing factors. Sensors 10, 2088–2106 (2010)
7. Simon, I., Bârsan, N., Bauer, M., Weimar, U.: Micromachined metal oxide gas sensors:
opportunities to improve sensor performance. Sens. Actuators, B 73, 1–26 (2001)
8. Jaaniso, R., Kiang Tan, O.: Semiconductor Gas Sensor. Electronic and Optical Materials,
vol. 38. Woodhead Publishing, USA (2013)
9. Korotcenkov, G.: Review metal oxides for solid-state gas sensors: what determines our
choice? Mater. Sci. Eng., B 139, 1–23 (2007)
10. Barsan, N., Koziej, D., Weimar, U.: Metal oxide-based gas sensor research: how to? Sens.
Actuators, B 121, 18–35 (2007)
11. Bârsan, N., Weimar, U.: Understanding the fundamental principles of metal oxide based gas
sensors; the example of CO sensing with SnO2 sensors in the presence of humidity. J. Phys.:
Condens. Matter 15, R813-R839 (2003)
12. Henrich, V.E., Cox, P.A.: The Surface Science of Metal Oxides. Cambridge University
Press, Cambridge (2000)
 Noise Reduction and Classification Method
for Partial Discharge Signal Detected
by Transient Earth Voltage Sensor
Using Wavelet Decomposition

Hiroaki Cho(&), Yuuki Fujii, and Yusuke Nakamura

Toshiba Infrastructure Systems & Solutions Corporation,

1, Toshiba-Cho, Fuchu-Shi, Tokyo, Japan
hiroaki.cho@toshiba.co.jp

Abstract. Partial Discharge (PD) is one of the most harmful phenomenon in

high voltage electrical equipment. While measurement of PD is widely accepted
to examine whether insulation performance of the equipment is deteriorated or
not, external noise interference often lower then the detection performance.
Transient earth voltage (TEV) method has some advantages such as easy to
attach, simple structure, but the detected signal is also disturbed by noise. In this
article, a noise reduction method using wavelet decomposition to identify the
PD signals from noisy signal is presented. The proposed method showed good
performance for multiple PD classification and noise reduction.

Keywords: Partial Discharge
Transient Earth Voltage
Insulation diagnosis

1 Introduction

Switchgears are widely used in electric power transmission and distribution network
seriously, and they are one of the most important apparatus to maintain stable power
supply. Many kinds of dielectric materials such as SF6 gas, epoxy resins are used as the
insulation component of switchgears. Almost all these materials are seriously degraded
by partial discharge. Deterioration of insulation performance due to degradation of
dielectric materials induce partial discharges (PDs) which finally causes electrical
breakdown. Thus, PD detection is widely utilized for insulation diagnosis of
switchgears.
PD detection methods with UHF, High Frequency CT, Acoustic Emission and
Transient Earth Voltage (TEV) sensor are effective methods in on-line testing [1]. For
live-line insulation diagnostics, the TEV sensor has following advantages. It can
sensing without installation inside switchgears, no need to interruption, very easy
handling etc. The authors are reported to detection mechanisms of TEV sensor [2]. We
also reported that PD detection signal contains two kind of current which induced by
electromagnetic waves and discharge current [3, 4].
It is common problem for on-site PD measurement that electrical interference by
noise. Many researchers reported the noise reduction techniques to extract the PD
signal from noisy signal [5–7]. In order to find dominant frequency of a signal, Fast
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 489–497, 2020.
https://doi.org/10.1007/978-3-030-31676-1_47
490 H. Cho et al.

Fourier Transform (FFT) is commonly used. The disadvantage of FFT is lost the time
information of signals. In the PD analysis, time (phase) information is very important to
identify the cause PD. Therefore, in order to use the both time information and fre-
quency information for identification of PD, discrete wavelet transform which able to
analyse the frequency component of signals without loss of time information was
applied.
Discrete wavelet transform is known as a powerful technique for de-noising [8], but
it is a non-linear filtering method that makes filtered signal’s interpretation difficult.
Moreover, wavelet transform decompose and reconstruct the analysing signal based on
some thresholding rules which very complex to set appropriate values. In this article,
wavelet coefficient during wavelet decomposition process was used to find the PD
occurrence time, and FFT and linear Finite Impulse Response filer were used to find the
dominant frequency and extract the PD signals, respectively.

2 Experimental Method
2.1 PD Samples and Measurement System
Creeping discharge sample and artificial voltage transformer defect sample (VT sam-
ple) were used to measure the partial discharge signals. Among the switchgear com-
ponent, voltage transformer may have potential defect due to complex structure around
primary and secondary windings, therefore, partial model of voltage transformer were
used in this study. The PD may be generated in the defects in case of anomaly
situations such as overvoltage, overheating and long-time aging over expected lifespan.
Figure 1 shows the composition of a partial model of voltage transformer used in this
study. This sample is mainly composed with winding (0.8 mm diameter enameled
wire) and the insulating paper. The coils were molded by the silica-filled bisphenol A
type epoxy resin under vacuum. Testing voltage was applied between the second and
the third layers.
Figure 2 shows the PD measurement circuit used in this study. PD signals were
measured by two methods. The one is PD detection instrument (Nippon Keisokuki
Seizousho, CD-5), the other is a TEV sensor which was attached to the iron plate with
grounding. Detected signals were measured with oscilloscope (Tektronix, DPO 7254).
The sampling rate of oscilloscope was 100 MS/s.

2.2 TEV Sensor

Reeves reported the TEV sensor which can detect partial discharges occurred in the
metal-clad high voltage instruments by attaching the sensor to the exterior creeping of
the item [9]. Recently, PD detection method with TEV sensor is extensively studied by
many researchers [1, 3, 4]. Itose et al. were reported the detection mechanism of the
TEV sensor [3]. TEV sensor can detect both high frequency component (HF signal)
and relatively low frequency circuit current (LF signal) due to electromagnetic wave
and current to earth line by partial discharges respectively. The frequency characteristic
of LF signal depends on the circuit which PD derived current flows and also depends
 Noise Reduction and Classification Method for PD Signal Detected 491

on original PD signal frequency. It is difficult to precisely predict the frequency

characteristic of the circuit of diagnosed apparatus. Accordingly, careful signal pro-
cessing is important to identify the PD signals.

Fig. 1. Cross-sectional view of PD sample.

Fig. 2. Measurement setup

2.3 Signal Processing Method

Figure 3 shows the flow-chart of signal processing proposed in this study. In the TEV
method, dominant frequency of LF signal may have different values depend on the
measuring site (circuit). In order to get the high signal-to-noise ratio signals on the
different sites, it is important to set the appropriate filtering frequency during mea-
surement. From this point of view, digital filtering method was used to get good
performance for de-noising.
Wavelet transform is an attractive method to extract the PD signals from noisy
signals. Discrete Wavelet transform is mainly composed of three processes (decom-
position, thresholding and reconstruction). Among these processes, thresholding is
tentatively studied by many researchers, since this process is fairly complex to get the
appropriate analysis parameters [4–6]. In this study, only wavelet coefficient during
decomposition process was utilized for de-noising to LF signals.
492 H. Cho et al.

Fig. 3. Flowchart of proposed method.

First of all, PD-like pulse were enhanced according to the wavelet decomposition
process as depicted in Fig. 3. This time, 5-levels decomposition was conducted. The
“bior_3-1” wavelet was selected for mother wavelet as follow.
1. Get the wavelet coefficient of 5 PD pulses.
2. Calculate the enhanced rate defined as Eq. (1) of each mother wavelet

Enhanced rate ¼ Wavelet coefficient of PD pulse=Original PD pulse ð1Þ

As a result, “bior_3-1” which has the maximum enhanced rate, was selected.
Next to the decomposition process, the pulse having the highest signal level and over
the thresholding value was extracted from processed wavelet coefficient. Then, domi-
nant frequency of the extracted pulse was analysed by FFT analysis at each decom-
position level of the signal to use as the center frequency of following filtering process.
Finally, finite impulse response filter with narrow band pass frequency width was
applied to the original signal with extracted dominant frequency as center frequency.

3 Result and Discussion

Partial discharge signals of VT sample and creeping discharge sample measured by

TEV sensor was shown in Figs. 4 and 5, respectively. As shown in Figs. 4 and 5, PD
waveform of VT sample is fairly different to creeping discharges. Frequency analysis
was conducted to confirm the dominant frequency of these signals. Figure 6 shows the
 Noise Reduction and Classification Method for PD Signal Detected 493

result of FFT analysis of VT sample and creeping discharge signals. In Fig. 6, around
2 MHz and 5 MHz components are dominant. In addition to those frequencies,
creeping discharge have 10–100 MHz component. This difference suggest the
importance of frequency analysis to identify the type of discharges. According to this
result, decomposition of the signal finely on frequency domain is useful to get the high
S/N ratio and to distinguish the discharge mode. The reason of some detected common
frequency bands may related to the electrical resonance of detection circuit [3].
PD signals of VT sample and creeping in Fig. 7 were used to evaluate the proposed
method. The two signals were numerically added to make the contaminated signal as
depicted in Fig. 7(c). As far as we see in Fig. 7(c), two kinds of PD signals are
contaminated, and can’t identify each signal.
The contaminated signal shown in Fig. 7 is analyzed by proposed method, and the
result is shown in Fig. 8. It can be seen two kinds of discharge signals are effectively
distinguished. Note that, in Fig. 8 two frequency bands were manually selected from
automatically selected frequencies in this time. The result suggest the validity of
proposed method in on-site, because this method automatically selects the possible
frequencies without pre-information about PD frequency as mentioned above.

0.02
0.015
0.01
voltage [V]

0.005
0
-0.005
-0.01
-0.015 1usec
-0.02
time [sec]

Fig. 4. TEV signal of a VT sample discharge.

Fig. 5. TEV signal of a creeping discharge.

494 H. Cho et al.

Fig. 6. FFT result of discharge signals.

(a) VT sample discharge signals.

(b) Creeping discharge signals.

(c) Numerically superimposed signals of VT sample discharges and Creeping dis-

charges.

Fig. 7. PD signals used as the input signal.

 Noise Reduction and Classification Method for PD Signal Detected 495

Fig. 8. Distinguished signals using proposed method. Red-line: Creeping discharges Blue-line:
VT sample discharges

(a) Actual noise signal.

(b) PD signal of VT sample.

(c) Contaminated signal of VT sample and actual noise.

Fig. 9. Analyzed signals.

496 H. Cho et al.

Fig. 10. Verification result of proposed method.

The noise signals measured in actual site were used to evaluate the de-noising
performance. Figure 9 shows the PD signal of VT sample and the actual noise signal as
well as contaminated signal of them. As shown in Fig. 9(c), the PD signal was heavily
interfered by the noise signal. Figure 10 shows adopted result of the proposed method.
As shown in Fig. 10, the PD signal was able to identify after signal processing. Again,
the frequency that getting the good distinguished result was manually selected from
automatically selected frequencies. After de-noising, one can use well-known PRPD
pattern to identify the discharge mode. The verification of the validity of proposed
method in the actual sites is on-going.
Although noise frequency depends on the measurement site locations and detection
circuit, etc., partial discharge signals may have some dominant frequency bands. The
proposed method can find the PD-like pulse signals by applying wavelet decomposition
and extract the possible frequency to get high S/N signals for filtering. This system is
more preferable to get the high-S/N signal than conventional system with fixed filtering
frequency, because conventional system may lost the important frequency band. Fur-
thermore, by using the digital filtering, diagnostic hardware system can made compact
and cost-effective.

4 Conclusions

Partial discharge detection method using wavelet decomposition was developed and
evaluated. This method could discriminate two kind of another discharge signal and
successfully de-noised the contaminated signals both PD and actual noise. The veri-
fication of the validity of proposed method in the actual sites is on-going.

References
1. Fan, W., Guan, S., Fu, J., Li, L., Li, J., Wang, W., Yan, X.: Comparison study of partial
discharge detection methods for switchgears. In: 2016 International Conference on Condition
Monitoring and Diagnosis (CMD2016), pp. 318–322 (2016)
2. Fujii, Y., Cho, H., Nakamura, Y.: Highly sensitive partial discharge detection by TEV method
under severe noise conditions. In: 2017 4th International Conference on Electric Power
Equipment - Switching Technology (ICEPE-ST), pp. 783–786 (2017)
 Noise Reduction and Classification Method for PD Signal Detected 497

3. Itose, A., Kozako, M., Hikita, M.: Partial discharge detection and induced surface current
analysis using transient earth voltage method for high voltage equipment. In: 2016
International Conference on Condition Monitoring and Diagnosis (CMD2016), pp. 455–459
(2016)
4. Yoshizumi, H., Koga, T., Kozako, M., Hikita, M., Fujii, Y., Nakamura, Y., Cho, H.:
Grounding effect on transient earth voltage signal induced by partial discharge in metal box
model. In: 8th International Symposium on Electrical Insulating Materials, vol. 2, no. 19,
pp. 555–558 (2017)
5. Ray, P., Basuray, A., Maitra, A.K.: Optimum wavelet bases selection for wavelet based de-
noising in partial discharge measurement. In: 2013 IEEE Conference on Information &
Communication Technologies, pp. 1110–1113 (2013)
6. Chan, J.C., Ma, H., Saha, T.K., Ekanayake, C.: A novel level-based automatic wavelet
selection scheme for partial discharge measurement. In: 22nd Australasian Universities Power
Engineering Conference, pp. 164–169 (2012)
7. Morshuis, P., Montanari, G.C., Fornasari, L.: Partial discharge diagnostics - critical steps
towards on-line monitoring. In: IEEE Power Energy Society (PES) Transmission Distribution
(T&D) Conference and Exposition, pp. 1039–1043 (2014)
8. Zhang, H., Blackburn, T.R., Phung, B.T., Sen, D.: A novel wavelet transform technique for
on-line partial discharge measurements part 1: WT de-noising algorithm. IEEE Trans.
Dielectr. Electr. Insul. 14(1), 3–14 (2007)
9. Reeves, J.H.: Non-invasive detection, location and monitoring of partial discharges in
substation H.V. plant. In: IEEE Conference on Revitalising Transmission and Distribution
Systems, no. 273, pp. 82–86. IEE Conference Publication (1987)
 Comparison of Wavelet, Stationary Wavelet
and Wavelet Packet Methods for De-noising
of Partial Discharge on Power Cable System

Arunjothi Rajendran(&), Thirumurthy, and K. P. Meena

Central Power Research Institute, Bangalore, India

{arunjothi,thiru,meena}@cpri.in

Abstract. The failure of Power Cable systems is mainly attributed to Partial

Discharge activity in power Cable insulation and accessories such as terminations
and joints. Different types of sensors are available for online and offline moni-
toring of Partial discharge activity. However the main difficulty involved in the
Partial discharge measurement of Power Cable system at site is to eliminate the
interference from other external and internal high frequency noisy signals. In this
paper, de noising the PD signals by the wavelet transforms, stationary wavelet
and wavelet packets techniques are compared for its merits and demerits. The
method used for selection of mother wavelet is based on minimum entropy
criteria. The two types of mother wavelet debauchees and symlet are used. The
effectiveness of level dependent hard thresholding method in retrieving the
original partial discharge signals over the other conventional thresholding
methods is also analyzed. The laboratory measured Partial discharge signals are
mixed with white Gaussian noise and discrete spectral interference (DSI) signals
in such a way that the partial discharge signals are completely immersed in the
simulated noises. The partial discharge signals are measured for power cables
system with Improper Stress control tubing termination.

Keywords: Partial Discharge
Power Cable
Wavelet
Noise elimination

1 Introduction

Power Cables systems comprises of accessories like Joints and terminations. The
accessories of the power cables are the weakest links for the cable system. Especially
for medium voltage and EHV Cables, the failure of cable system due to poor design of
cable is very minimal compared to that of failure due to poor design of accessories.
Inspite of the higher stresses that occur in high voltage cables, if the accessories are
designed more carefully to handle these higher stresses the failure rate can be mini-
mized. Partial Discharge is one of the main cause for failures in cable accessories. For
service cables the partial discharge measurements are being carried out in online mode.
However these acquired PD signals are hampered with external and internal distur-
bances. The discrete Wavelet transform (DWT), Wavelet Packet & Stationary wavelet
methods are found to be a powerful signal processing tool to de-noise the PD signals.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 498–509, 2020.
https://doi.org/10.1007/978-3-030-31676-1_48
 Comparison of Wavelet, Stationary Wavelet and Wavelet Packet Methods 499

2 Denoising Techniques

The general de-noising method proceeds in three steps: (a) decomposing the noisy
signal, (b) thresholding the wavelet coefficients and (c) reconstructing the signal as
shown in Fig. 1.

Fig. 1. Wavelet-based de-noising procedure

2.1 Discrete Wavelet Transform

The principle of discrete wavelet transform is passing the signal in two additional filters
(low-pass filter and high-pass filter) to obtain the coefficients of discrete wave-let
transform. Approximation is all the coefficients obtained at the output of low pass filter
and contains the information of low frequencies. The detail is all the coefficients
obtained at the output of high pass filter and contains the information of high fre-
quencies. The decomposition process can be iterated, with successive approximations
being decomposed in turn, so that one signal is broken down into many lower-
resolution components. This is called the wavelet decomposition tree.

2.2 Wavelet Packet Transform

Here both the low-pass and the high-pass output undergo splitting at the subsequent
level. Therefore, WPT have the capability of partitioning the high-frequency bands to
yield better frequency resolution. Therefore, to achieve the best de-noising perfor-
mance, there is a need of choosing the best tree for representing a signal in wavelet
packet domain.

2.3 Stationary Wavelet Transform

Stationary Wavelet Transform (SWT), also known as un-decimated wavelet transform
or translation-invariant wavelet transform is a modification of the Discrete Wavelet
Transform that does not decimate coefficients at every transformation level.
The key point is that it gives a better approximation than the discrete wavelet
transform (DWT) since, it is redundant, linear and shift invariant [6]. These properties
provide the SWT to be realized using a recursive algorithm. Therefore, the SWT is very
useful algorithm for analyzing a linear system.
500 A. Rajendran et al.

The key point is that it gives a better approximation than the discrete wavelet
transform (DWT) since, it is redundant, linear and shift invariant. These properties
provide the SWT to be realized using a recursive algorithm. Therefore, the SWT is a
very useful algorithm for analyzing a linear system.
In both DWT and WPT, after filtration the coefficients will be down-sampled, that
prevents redundancy and allows using the same pair of filter in different levels. And so,
these transforms will suffer from the lack of shift invariance, which means that small
shifts in the input signal can cause major variations in the distribution of energy
between coefficients at different levels and may causes some error in reconstruction [1].
This problem is carried out by eliminating the down sampling steps after filtration
at each level in stationary wavelet transform (SWT). Stationary Wavelet Transform
(SWT) is similar to Discrete Wavelet Transform (DWT) but the only process of down-
sampling is suppressed. By eliminating down sampling, the number of coefficients at
each level is as long as original signal. Figure 2 shows decomposition of a signal by
SWT up two levels.

Fig. 2. SWT composition scheme

3 Background

In general the noise elimination techniques and thresholding techniques are applied on
the simulated PD signals of DOP and DEP signals mixed with white noise and Discrete
Spectral Interference (DSI) and the effectiveness of the noise elimination techniques
have been compared. However in this paper the simulated and as well as laboratory
measured PD signals of Power Cable systems are used to check the effectiveness of the
algorithms. Moreover the train of measured PD pulses is considered in order to check
the effectiveness of the proposed algorithms in retrieving all the pulses.
The noise levels considered in this paper is also very high and the Signal to Noise
Ratio (SNR) of the noisy signals are in the range of −25. The noises are added to the
laboratory measured signals in such a way the PD signals are completely immersed in
the noises. In this manner the worst case scenario of PD signals completely merged in
noises are considered. The prevailing denoising methods of Discrete wavelet,
 Comparison of Wavelet, Stationary Wavelet and Wavelet Packet Methods 501

Stationary wavelet and wavelet packet methods are applied to the noisy signals and the
effective and suitable noise elimination technique is arrived which reproduces the
original signal in a better way. In general the selection of mother wavelet in the noise
elimination is also done in arbitrary manner. In this paper a systematic approach of
selection of mother wavelet based on minimum entropy criteria is applied.
According to Madhu [2] the maximum energy based mother wavelet selection &
level dependent thresholding algorithm was very helpful to extract the simulated partial
discharge signals from noises and reduce distortion of de-noised signals. In the present
study the same approach is extended to laboratory PD signals and it was also found that
minimum entropy criteria mother wavelet selection is better in reproducing the PD
signals.
According to Kyprianou [3], a new algorithm using entropy based mother wavelet
selection & hard thresholding wavelet packet denoising with symlet family of wave-
lets successfully recovered the PD signal in most of the cases including field signal,
showing the potential of the wavelet packet algorithm to effectively denoise highly
corrupted simulated PD signals and as well as field measured PD signals. The similar
approach of entropy based de noising technique is applied with DWT, wavelet packet
and stationery wavelet methods for laboratory measured PD signals and the results are
compared. In this paper the suitability of the three techniques DWT, wavelet packet and
stationary wavelet for simulated PD signals and measured PD signals are also analyzed.

4 Methodology

In order to select the optimum mother wavelet, minimum entropy criteria is applied for
the wavelet family of Daubechies (db1 to db45) and symlet (sym1 to sym 41) and the
noise elimination was carried out with the selected optimal mother wavelet of Dau-
bechies and symlet. The optimal decomposition level selected for this study is given by
Eq. (1).

J ¼ fix ðlog2 ðN ÞÞ ð1Þ

J denotes the number of decomposition level and N is the signal length, and fix is to
get the nearest integer of the (log2(N)) toward zero. The decomposition level is selected
in such a way that the signal length at the highest decomposition level is not less than
the length of the wavelet filter being used. Hence, the maximum number of decom-
position levels denoted by Jmax, Eq. (2), is determined based on the length of both the
PD signal and the wavelet filter being used [4].



N
Jmax ¼ fix log2 1 ð2Þ
Nw

Where N is the length of the signal and Nw is the length of the decomposition filter
associated with the employed mother wavelet. In order to select the optimum mother
wavelet, entropy have been calculated from the wavelet family of Daubechies (db1 to
502 A. Rajendran et al.

db 45) and Symlet (sym1 to sym 41) separately. The entropy of wavelet co-efficient W
are defined for each level is given by Eq. (3) [5].
Xn
Entropy H ðsÞ ¼ i¼1
Pi log2 Pi ð3Þ

2
where Pi ¼ PjnW ðs;iÞj
i¼1
jW ðs;iÞj2
By calculating the values of entropy for the noisy PD signal the mother wavelet
with minimum entropy value among the wavelet family of db1 to 45 & sym 1 to 41 are
considered. Hence wavelet transform, Stationary Wavelet transform and wavelet packet
transform techniques are applied for the noisy signal with minimum entropy mother
wavelet in both the Daubechies and symlet cases. The thresholding rule adopted was
level dependent thresholding which is as given below in Eq. (4)
mj pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kj ¼ 2 log nj ð4Þ
0:6745

Where; kj indicates the threshold value at level j, mj is the median value of the
coefficients at level j, and n is the length of given signal under analysis [6, 7]. Then the
performance parameters as stated below are calculated with respect to the original for
all the three methods. Signal to Noise Ratio.

j f j2
SNR ¼ 10 log 10 ð5Þ
jf  r j2

Mean Square Error

1 Xn
MSE ¼ ½f ðiÞ  r ðiÞ2 ð6Þ
n i¼1

Peak Amplitude Distortion



 fmax  rmax 
PAD ¼ 100 
 ð7Þ
rmax

Cross Correlation Coefficient

P
ðf  f Þðr  r Þ
cc ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P P ð8Þ
ðf  f 2 Þ ðr  r 2 Þ

Where,
n – length of the Signal, f – The original PD signal, r –De-noised PD signal,
fmax – Maximum value of original PD signal, rmax– Maximum value of de-noised PD
signal f – Mean value of signal f, r – Mean value of signal r.
 Comparison of Wavelet, Stationary Wavelet and Wavelet Packet Methods 503

5 Results and Discussion

5.1 Simulated PD Signal
In order to compare the three types of transform techniques, initially the simulated PD
signal as shown below in Fig. 3 is considered.

Fig. 3. Simulated PD signal

The simulated PD signal is severely corrupted by mixing −15 db white noise and
DSI signal and the following noisy signal of SNR: −24.97 is simulated (Fig. 4).

Fig. 4. Simulated noisy signal of SNR: −24.97

Table 1. Performance parameters of noise elimination

Type of transform CC PAD SNR MSE
Wavelet db45 0.91 1.52 17.4 4.62e−4
Sym40 0.90 2.64 16.58 4.76e−4
Wavelet packet db45 0.85 24.6 10.61 7.01e−4
Sym40 0.88 23.3 13.29 5.75e−4
Stationary wavelet db45 0.72 60.6 2.82 1.17e−3
sym40 0.80 51.6 6.31 8.6e−4

Table 1 shows the effectiveness of noise elimination by calculating performance

parameters for minimum entropy Daubechies & Symlet mother wavelet with LD
thresholding.
504 A. Rajendran et al.

From Table 1, it is observed that all the three transform techniques with LD hard
thresholding are able to retrieve the PD signal with improved values of signal to noise
ratio. However the wavelet transform yields better results in terms of all the parameters
compared to wavelet packet and stationary wavelet transform. In case of stationary
wavelet transform the Peak Amplitude Distortion is more and in the range of 60
(Figs. 5 and 6).

Fig. 5. De-noised signal with wavelet db45

Fig. 6. De-noised signal with SW Sym 40

5.2 Laboratory Measured PD Signals

The partial discharge signals considered for this study are the laboratory measured
signals. The signal shown in Fig. 7 is measured during the failure of SCTM termination
of 33 kV XLPE Cable.

Fig. 7. PD signal of a 33 kV XLPE Cable

 Comparison of Wavelet, Stationary Wavelet and Wavelet Packet Methods 505

Along with the measured PD signal, the simulated −10 db of white Gaussian noise
and Discrete Spectrum Interference (DSI) signals are combined and the noisy signal as
shown in Fig. 8 is obtained.

Fig. 8. Noisy PD signal with (SNR: −24.05)

Since the other thresholding methods like Minimax, universal does not eliminate the
noise, the level dependent hard thresholding is applied throughout this paper. Table 2
shows the performance parameters obtained using minimum entropy Daubechies &
symlet mother wavelet with LD hard thresholding in eliminating the noises effectively.

Table 2. Performance parameters of noise elimination

Type of transform CC PAD SNR MSE
Wavelet db43 0.62 51.8 1.67 1.77e−20
Sym41 0.63 11.7 1.31 1.85e−20
Wavelet packet Db43 0.32 57.7 −5.38 3.23e−20
Sym41 0.50 5.5 0.32 2.93e−20
Stationary wavelet Db43 & sym41 0.06 390.7 0.04 7.15e−18

From Table 2, it is observed that the stationary wavelet transform with level
dependent thresholding is unable to retrieve the original signal. The wavelet transform &
wavelet packet transform with level dependent thresholding are able to retrieve the signal.

Fig. 9. Denoised signal with wavelet db43

506 A. Rajendran et al.

Fig. 10. Denoised signal with wavelet Sym41

Fig. 11. Denoised signal with wavelet Packet Sym41

In case of wavelet, the cross correlation, mean square error, signal to noise ratio are
better than that of wavelet packet. In case of peak amplitude distortion the mother
wavelet symlet is better in both wavelet and wavelet packet transform.
Figures 9 and 10 shows the de-noised signal with wavelet transform for db43 &
sym41 mother wavelet. Figure 11 shows the de-noised signal with sym41 wavelet
packet transform. In all the cases the higher amplitude pulse of the original PD signal is
retrieved, however the small PD signal are not retrievable.
In order to check the effectiveness of the wavelet transform the train of pulses of the
same PD signal of a 33 kV cable is considered as shown in Fig. 12. Figure 13 shows
the noisy signal with −10 db white Gaussian noise and DSI noise.

Fig. 12. Train of PD signals of a 33 kV XLPE Cable

 Comparison of Wavelet, Stationary Wavelet and Wavelet Packet Methods 507

Fig. 13. Noisy signal of SNR: −32.35

In order to check the effectiveness of the wavelet transform the train of pulses of the
same PD signal of a 33 kV cable is considered as shown in Fig. 12. Figure 13 shows
the noisy signal with −10 db white Gaussian noise and DSI noise.

Table 3. Performance parameters of noise elimination by wavelet transform

Wavelet transform CC PAD SNR MSE
db45 0.78 77.8 8.11 1.49e−20
Sym41 0.79 78.7 8.27 1.44e−20

Fig. 14. Denoised signal with wavelet Sym 41

From Table 3 it is evident that both symlet and Daubechies mother wavelet yields
almost same parameters. In this case the cross correlation coefficient is more (0.79) than
the case of single pulse (0.63). Figure 14 shows the de-noised PD signal with Sym 41.
It is observed that all the higher magnitude pulses are retrieved; however peak
amplitude distortion is high. Figure 15 shows the one more laboratory measured PD
signal of power cable system failure.
508 A. Rajendran et al.

Fig. 15. Measured PD signal of a defective power cable system

Figure 16 shows the de-noised signal with wavelet sym41 when the original signal is
highly corrupted with SNR of −25. The de-noised signal of both sym41 & db44 has the
cross correlation co efficient of 0.73 and the peak amplitude distortion is as high as 38%.

Fig. 16. Denoised signal with wavelet Sym 41

6 Analysis

In this study various noise elimination techniques and its effectiveness in removing the
noise is discussed. From this study the following observations are drawn.
For the simulated PD signal merged with noise all the three wavelet, wavelet packet
& stationary wavelet transform are able to retrieve the PD signal. However among the
three transforms, the wavelet transform with level dependent has low peak distortion,
high cross correlation co-efficient and high signal to noise ratio. Stationary wavelet
transform has the highest peak amplitude distortion in retrieving the original simulated
signal.
For the laboratory measured PD signal with one pulse merged with high noise the
stationary wavelet transform is not effective in retrieving the original signal. In this case
both wavelet and wavelet packet transform with level dependent thresholding is able to
reproduce the original PD signal with both Daubechies & Symlet mother wavelet.
However the cross correlation coefficient of wavelet transform is better than the
wavelet packet transform. The minimum entropy symlet wavelet yields better peak
amplitude distortion in both wavelet and wavelet transform techniques.
 Comparison of Wavelet, Stationary Wavelet and Wavelet Packet Methods 509

For the train of PD pulses the wavelet transform yields better results than that of
wavelet packet transform.

7 Conclusion

From this study it is inferred that minimum entropy mother wavelet selection yields
better results in elimination the noise. The level dependent hard thresholding works
effectively in retrieving the partial discharge signals. For the simulated PD signals, all
the three wavelet, wavelet packet & stationary wavelet transforms are able to retrieve
the signals. For real signal stationary wavelet is not able to reproduce the original PD
signal. The wavelet transform with Daubechies & symlet mother wavelet yield better
performance in retrieving the train of pulsed of partial discharges.

References
1. Mortazavi, S.H., Shahrtash, S.M.: Comparing denoising performance of DWT, WPT, SWT
and DT-CWT for partial discharge signals. In: 43rd International Universities Power
Engineering Conference, UPEC 2008 (2008)
2. Madhu, S., Amit, S., Shivam, G., Sumathi, S., Vidya, H.A.: Level dependent wavelet
selection for denoising of partial discharge signals simulated by DEP and DOP Models. In:
International Conference on Computer Science and Mechanical Engineering, 10th August
2014, Jaipur, India (2014)
3. Kyprianou, A., Lewin, P.L., Efthimiou, V., Stavrou, A., Georghiou, G.E.: Wavelet packet
denoising for online partial discharge detection in cables and its application to experimental
field results. Meas. Sci. Technol. 17, 2367–2379 (2006)
4. Hussein, R., Shaban, K.B., El-Hag, A.H.: Wavelet transform with histogram based threshold
estimation for online partial discharge signal denoising. IEEE Trans. Instrum. Meas. 64(12),
3601–3614 (2015)
5. Ray, P., Maitra, A.K., Basuray, A.: Entropy-based wavelet de-noising for partial discharge
measurement application. In: 2016 IEEE First International Conference on Control,
Measurement and Instrumentation (CMI), India (2016)
6. Jayakrishnan, M., Rao, B.N., Meena, K.P., Arunjothi, R., et al.: Optimum threshold estimator
for de-noising partial discharge signal using wavelet transform technique. In: IEEE
International Conference CATCON 2015 (2015)
7. Evagorouet, D., et al.: Evaluation of partial discharge denoising using the wavelet packets
transform as a preprocessing step for classification. In: CEIDP 2008 (2008)
 Pattern Recognition of Partial Discharges
on Power Cable Systems

Arunjothi Rajendran(&), Thirumurthy, and K. P. Meena

Central Power Research Institute, Bangalore, India

{arunjothi,thiru,meena}@cpri.in

Abstract. Partial discharge measurements are considered to be an important

diagnostic/condition monitoring tests on Power Cable System. Various off line
and online measurement systems have come up with inductive and capacitive
sensors. The high frequency partial discharge signals occur in power cable
system due to various defects such as voids/cavity in the power cable insulation,
defective termination, defective stress control materials and defective joints.
However, the partial discharge pattern differs for each type of defects. Hence to
identify the type of defect, proper analysis of pattern is required. The statistical
parameters such as mean, skewness, Kurtosis etc. with respect to the phase
angle, highest discharge magnitude etc. helps in extracting the feature infor-
mation of each pattern. PD-fingerprints such as Skewness (Sk) and kurtosis
which measures the degree of asymmetry of a distribution & sharpness of a
distribution, along with the average value of each half cycle (Mean) are esti-
mated using MATLAB programming for various partial discharge signals of
laboratory failed Power Cables and accessories. In this work, an attempt is made
to develop some finger prints for various defects on power cable systems using
the statistical parameters and PD pattern.

Keywords: Partial discharge
PD pattern
Power Cable
Statistical

parameters

1 Introduction

The reliability of power distribution and transmission system depends largely on the
insulation quality of electrical equipment. As per the failure statistics, majority of
failures in distribution sector is connected with power cable system especially the cable
accessories. The major failure mode of power cable system is associated with degra-
dation of insulation due to partial discharge activity. Even though standard specifies the
maximum discharge magnitude as the main criteria for acceptance, the exact location of
the defects with the magnitude decides the criticality. Hence the maximum discharge
magnitude shall be read in conjunction with the phase resolved pattern to check its
significance.
As per the reference [1], there are four usual types of PDs reported in the XLPE
insulation of high voltage underground cables (1) the first one is an internal PD in
insulation material with different types of cavities; (2) the corona discharge is a PD
produced by sharp edges; (3) electrical arborescence is generated by the combination of

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 510–520, 2020.
https://doi.org/10.1007/978-3-030-31676-1_49
 Pattern Recognition of Partial Discharges on Power Cable Systems 511

the corona discharge and superficial PD; (4) superficial PDs occur when there is effort
on the surface of the dielectric, and usually they are presented in cable ends [1]. In
many published articles the PD patterns of XLPE Cable insulation are classified
according to the type of defects of the insulation. In the laboratory newly manufactured
power cables along with different types of joints and terminations are being type tested
for its compliance to the required standard. Since the cables are manufactured at the
factory following stringent quality norms, the chances of defects from the cable
insulation is a rare case in the new cable compared to the cable accessories which are
prepared at site. As per reference [2], the PD patterns and their classification are
discussed for four typical created defects in the terminations and joint [2]. Hence the
power cables needs to be considered as a system along with the accessories for its PD
pattern classification and statistical parameter calculation. The intent of this paper is to
classify the partial discharge pattern of XLPE cable accessories of power cables by
means of statistical parameters.

2 Power Cable Systems

Power cable insulation, started with oil impregnated paper had graduated to Cross
linked Polyethylene, due to its very good thermal and electrical characteristics. How-
ever, extruded polymers like XLPE are more sensitive to PD activity. Preliminary PDs
expand to a weak spot very fast, depending upon the electric field stress. In the case of
PILC cables, defects such as localized voids is avoided to a large extent by the presence
of oil, where as in the case of extruded solid insulation such as XLPE, localized defects
will develop partial discharges on energization. With the improvement in manufac-
turing technology, chances of partial discharges with XLPE power cables due to poor
manufacturing process has reduced to minimum. In addition, the power cables after
production, is getting routine tested by conducting Partial discharge test and AC high
voltage test on drum length. However, the critical components of power cable system,
the termination and joints are mounted at site. The quality of these accessories depends
largely on the quality of accessory components, proper preparation of cable ends and
the skillset of the jointer. Common requirement for all cable accessories are adequate
conductor connection and the reinstatement of insulation. For all medium voltage
accessories, operating at or above 3.8/6.6 kV, electrical stress control is required at the
screen terminus points. In the case of joints, in addition to the screen terminus points,
radial stress at the connector position also has to be alleviated through proper stress
control.

3 Partial Discharge Measurement

Generally, partial discharges are classified in three main groups, namely internal dis-
charges, surface discharges and corona discharges. Internal discharges occur at
dielectric with a number of cavites within the insulation media which develops further
electrical trees as a result of cumulative PD activity. Surface discharge occurs along the
dielectric interface. The formation of a conductive path along the surface of the
512 A. Rajendran et al.

insulation results in tracking which further converts to electrical treeing and eventually
complete breakdown. External corona discharges occur at sharp edges at high voltage
side. Such corona discharges are considered to be less harmful to the insulation as
compared to the internal and surface discharges depending on the material.
Coming to power cable system, the uniform electrical stress distribution of a power
cable gets disturbed at the terminations and joints due to the interruption of screen
terminus, which in turn produce high tangential and radial stresses. These higher
stresses at the screen terminus are alleviated by providing proper stress control mate-
rials. Hence the main part of cable accessory is the stress control system, the effec-
tiveness of the same is evaluated by partial discharge test. In Medium voltage cable
accessories of heat shrinkable accessories, stress control is provided by high permit-
tivity materials at the screen terminus points. Improper stress control occurs due to
ineffective materials, improper application and improper dimensions.

4 Phase Resolved Pattern and Statistical Parameters

Primitive partial discharge detection was by visually observing the pattern in an

oscilloscope with elliptical time base. The advancement in electronic computation and
digital signal processing techniques has opened up new opportunities for exploiting the
pattern recognition techniques to efficiently utilize the vast stream of data to diagnose
potential defects in insulating media.
One of the undoubted advantages of a computer-aided measuring system is the
ability to process a large amount of information and to transform this information into
an understandable output [3]. The illustration of the partial discharge activity relative to
the 360° of an AC cycle, which is commonly known as Phase Resolved Partial Dis-
charge (PRPD) Pattern, allows for identifying the prominent root cause of partial
discharges, therefore appropriate corrective actions can be implemented. There are
many types of patterns that can be used for PD source identification. If the differences
between different patterns can be presented in terms of statistical parameters, identi-
fication of the type of defect from the observed PD pattern is much simpler. As each
defect has its own particular degradation mechanism, it is important to know the
correlation between discharge patterns and the kind of defect. Therefore, progress in
the recognition of internal discharge and their correlation with the kind of defect is
becoming increasingly important in the quality control in insulating systems [4]. In this
study, the partial discharge test results of certain MV cable accessories with defects
were analyzed from the patterns and their respective statistical parameters.
The important parameters to characterize PDs are phase angle u, PD charge
magnitude q and PD number of pulses n. PD distribution patterns are composed of
these three parameters. Statistical parameters are obtained for phase resolved pattern
(n − q). Due to the stochastic nature of partial discharges, the signals need to be
analyzed for a time span much longer than the duration of one voltage cycle. These
quantities can be analyzed as a function of time and as a function of the phase angle.
The quantities as function of time describe the changes of the basic quantities in the
course of time. The quantities as function of the phase angle represent the recurrence of
partial discharges related to their phase angle. The voltage cycle is divided into 4000
 Pattern Recognition of Partial Discharges on Power Cable Systems 513

phase windows and the main quantities such as average discharge magnitude, pulse
count & maximum discharge magnitude etc. are determined for each phase window. In
addition, the statistical operators such as skewness and kurtosis derived from mean and
standard deviation of the signals acquired per phase window, also provides some
additional data for evaluation of pattern. In all the cases, 400 voltage cycles of partial
discharge data are considered and the PD data of each voltage cycle has been divided
into 4000 bins and the statistical parameters are calculated.

5 Experimental Set up and Procedure

In this paper, the conventional partial discharge detection as per IEC 60270 is proposed
for carrying out defective power cable systems. The measurements were carried on
10 m length of sample by connecting the sample to a 200 kV PD free transformer and a
200 kV PD free coupling capacitor. After connecting the sample, calibration was
carried out by injecting a short duration current pulse of known charge magnitude
repetitively into one of the termination of cable sample and measuring the same in the
PD detector. Calibration is performed for the complete test configuration including the
cable system under test, coupling capacitor and the PD measuring system. In a similar
manner each cable system requires calibration as the capacitance of each system differs.
The test set up with high voltage source and coupling capacitor connected to the sample
is shown in Fig. 1.

Fig. 1. Test set up for partial discharge test

514 A. Rajendran et al.

The samples considered for measurement are as given in Table 1.

Table 1. Types of cable systems analyzed

Case study nos Type of the cable system
I Transition joint connecting PILC and XLPE 11 kV cables (defective)
II Heat shrink joint on 11 kV XLPE cable
III Heat shrinkable termination of 33 kV XLPE cable
IV Heat shrinkable termination of 11 kV XLPE cable

The partial discharge data were collected for 400 cycles of voltage and the data is
analyzed by splitting one cycle phase window in to 4000 bins. Then the discharge
patterns are analyzed along with phase resolved patterns and respective statistical
parameters such as average value, skewness and kurtosis in both positive and negative
voltage cycles were computed.
Skewness and Kurtosis are two numerical measures of shape which gives infor-
mation by checking its deviation from normality. Skewness is a measure of asymmetry
or degree of tilt of data with respect to normal distribution. For symmetric distribution,
SK = 0, If it is asymmetric towards left, SK > 0, If it is asymmetric towards right,
SK < 0. Kurtosis is an indication of sharpness of distribution. If the distribution has
same sharpness as a normal distribution, Ku = 0. If it is sharper than the normal then
Ku > 0. Else if it is flatter than the normal distribution, then Ku < 0. Ku = positive
means a sharp distribution and Ku = negative means a flat distribution. The skewness
and kurtosis are calculated as shown in Eqs. 1 and 2 respectively [5, 6].
Pn 3
i¼1 ðx
P i  lÞ f ðxi Þ
Skewness Sk ¼ n ð1Þ
r3 i¼1 f ðxi Þ
Pn 4
i¼1 ðx
P i  lÞ f ðxi Þ
Kurtosis Ku ¼  3:0 ð2Þ
r4 ni¼1 f ðxi Þ

where,
x = number of pulses n
f(x) = PD charge magnitude q,
l = average mean value of q,
r = variance of q

6 Experimental Results and Analysis

The partial discharges were measured at 1.73Uo after pre-stressing the cable with
accessories at 2Uo for 1 min, where Uo refers to the rated voltage of the cable. The
observed phase resolved partial discharge pattern (PRPD) for the four samples con-
sidered are given in Figs. 2, 3, 4 and 5.
 Pattern Recognition of Partial Discharges on Power Cable Systems 515

From Figs. 2, 3, 4 and 5, it is evident that the pattern shows a strong asymmetry in
the case II & IV, 11 kV heat Shrink Joint and 11 kV heat shrink Cable termination
suggesting a significant difference in the discharge behavior for positive and negative
discharge.
The statistical parameters computed for Partial Discharge data for the four case
studies for 400 V cycles of data are tabulated in Table 2. The values of Skewness and
Kurtosis are calculated separately for both positive half cycles (Phase angle 0 to 180°)
and negative half cycle (Phase angle 180 to 360).
In each window the average value of discharges observed is determined. The mean
pulse height distribution Hq which represents the average amplitude in each phase
window as a function of the phase angle. Hq is derived from the total discharge amount
in each phase window divided by the number of discharges in the same phase window
and Hq is calculated for both positive and negative half cycle of the applied Voltage.

Fig. 2. Case study – I - Observed PRPD pattern

Fig. 3. Case study – II - Observed PRPD pattern

Fig. 4. Case study – III - Observed PRPD pattern

516 A. Rajendran et al.

Fig. 5. Case study – VI - Observed PRPD pattern

Table 2. Calculated statistical parameters

Sl.no Type of failure Skewness (SK) Kurtosis (Ku)
+Ve cycle −Ve cycle +Ve cycle −Ve cycle
1 PILC XLPE transition joint 0.983 0.978 2.299 2.299
2 11 kV heat shrink joint 2.303 1.541 8.672 4.778
3 33 kV cable termination 2.615 3.014 13.15 12.76
4 11 kV heat shrink cable termination 3.007 2.855 13.45 15.39

Figures 6, 7, 8 and 9 shows the PD histogram i.e. the number of PD pulses with
respect to the Average PD magnitude.

Fig. 6. Case study – I - PD histogram for the cable system

Fig. 7. Case study – II - PD histogram for the cable system

 Pattern Recognition of Partial Discharges on Power Cable Systems 517

Fig. 8. Case study – III - PD histogram for the cable system

Fig. 9. Case study – VI - PD histogram for the cable system

From Figs. 6, 7, 8 and 9. It is observed that in all the four cases the numbers of PD
pulses are as much as 2500 to 3000 for the low value of average PD magnitude i.e. less
than 20 pC. Since all the cables are new cables mounted with new accessories and
subjected for type test, the average magnitude of partial discharges are comparatively
less.
Figures 10, 11, 12 and 13 shows the plot of average pulse magnitude (Hq) with
respect to phase angle. The terminations and joints are made partial discharge free by
providing proper stress control tubings. In the case of termination, stress control tubing
is provided to control the longitudinal stress experienced at the screen terminus points.
For joints, it experiences both tangential and radial stresses, tangential stress at the
screen terminus of the cable ends and radial stress between conductor to screen at the
ferrule region.

Fig. 10. Case study – I - Average PD pulse magnitude vs phase angle

518 A. Rajendran et al.

Fig. 11. Case study – II - Average PD pulse magnitude vs phase angle

Fig. 12. Case study – III - Average PD pulse magnitude vs phase angle

Fig. 13. Case study – VI - Average PD pulse magnitude vs phase angle

The PD pattern of the faulty transition joint (Case Study I) is carried out after
thermal short circuit test. The discharge pattern and the Hq plotis symmetrical at both
cycles and it starts from zero crossing, indicating surface discharges due to loose
contacts at high voltage or ground region. Hence it will experience surface discharges
and internal discharges. The skewness and kurtosis values of this case I are equal at
positive and negative cycles.
However, for 11 kV Heat shrinkable joint (Case study II), the discharges are higher
with positive half cycle indicting improper stress control at the screen terminus points.
Both skewness and kurtosis are higher at positive half cycles. The average pulse
magnitude (Hq) is also higher in positive half cycle than that of negative half cycle.
Regarding partial discharge pattern with respect to terminations, the starting point is
generally from a higher phase angle and the discharge pattern forms a curvy shape
which follows the sinusoidal voltage waveform indicating that the electric field causing
the PD follows the applied voltage. When the applied voltage increases, the PD
 Pattern Recognition of Partial Discharges on Power Cable Systems 519

magnitude also increases due to higher electric field. Approximately similar pattern in
both positive and negative cycle for 33 kV termination (Case study III) indicates the
symmetry of the defect. Since the product is a cable with stress control tubing, the
defect is with poor stress control tubing material of improper permittivity. In the case of
11 kV termination, (Case study IV) the discharge magnitude in the positive half cycle
is higher than the negative half cycle indicating the discharge zone is near to the ground
point. Improper shrinking of stress tubing or improper installation may be the reason
for these discharges.
Comparing the skewness and Kurtosis parameters, it is observed that these values
are lower for joints and higher for terminations.
For joints, the partial discharge signals starts from a much lower phase angle (0 to
20°) compared to terminations (above 40°). It is also observed that generally, Kurtosis
the sharpness of pattern also will be much higher with terminations than joints.

7 Conclusion

In this study, partial discharge measurements are carried out for four different types of
defective MV power cable system, as per IEC 60270. Phase resolved Discharge pattern
and statistical parameters such as Skewness & Kurtosis are calculated for all the four
types of cable systems. The following observations are derived from this study.
1. The cable accessories (joints & terminations) are the weakest link of cable systems
which lead to partial discharges.
2. If the partial discharge is surface discharge due to improper contacts, the discharge
pattern is symmetrical in both positive and negative cycle. The discharge also starts
from zero crossing itself.
3. The partial discharge patterns of terminations and joints are different.
4. In case of termination failure the partial discharge initiates after 40° phase angle in
the positive cycle and 220° phase angle in the negative cycle. The discharge
magnitudes are higher in case of terminations.
5. In case of Joint failure the PD initiates between 0 to 20° in positive cycle and 180 to
200° in the negative cycle.

References
1. Figueroa Godoy, F., Jimenez, J.T., Vacio, R.J., Yáñez Mendiola, J., Colin, J.Á.: Analysis of
Insulating material of XLPE cables considering innovative patterns of partial. Math. Prob.
Eng. 2017, 10 (2017). Article ID 2379418
2. Mazzetti, C., Mascioli, F.F., Baldini, F., Panella, M., Risica, R., Bartnikas, R.: Partial
discharge pattern recognition by neuro-fuzzy networks in heat-shrinkable joints and
terminations of XLPE insulated distribution cables. IEEE Trans. Power Deliv. 21(3), 1035–
1044 (2006)
3. Kothoke, P.M., Bhosale, N.R., Despande, A., Cheeran, A.N.: Analysis of partial discharge
using phase-resolved (n − q) statistical techniques. Int. J. Eng. Res. Appl. (IJERA) 3(3),
pp. 1317–1323 (2013). www.ijera.com, ISSN: 2248-9622
520 A. Rajendran et al.

4. Gulski, E., Kreuger, F.H.: Computer-aided recognition of discharge sources. IEEE Trans.
Electr. Insul. 2. Sahoo, N.C., Salama, M.M.A., Bartnikas, R.: Trends in partial discharge
pattern classification: a survey. IEEE Trans. Dielectr. Electr. Insul. 12(2) (2005)
5. Bhosale, N., Kothoke, P., Deshpande, A., Cheeran, A.N.: Analysis of partial discharge using
phase-resolved (F-Q) and (F-N) statistical techniques. Int. J. Eng. Res. Technol. (IJERT) 2(5)
(2013)
6. Nair, R.P., Nageshwar Rao, B., Sumangala, B.V.: Statistical analysis of surface discharges in
rotating machine stator insulation systems. In: ISH (2017)
 Partial Discharge Monitoring of Single-Phase
Shell-Type 500 KV GSU Transformers in Salto
Grande Hydro Power Plant

E. Briosso1, B. Gorgan2(&), and W. Koltunowicz2

1
CTM Salto Grande Power Plant, Salto, Uruguay
2
OMICRON Energy Solutions GmbH, Lorenzweg 5, 12099 Berlin, Germany
bogdan.gorgan@omicronenergy.com

Abstract. In this paper, the challenges in the condition assessment of bushing

and winding insulation in single-phase shell type 500 kV, 100 MVA GSU
transformers in Salto Grande Hydropower Plant are described. Since after some
years of service, dissolved gas analysis recorded elevated levels of hydrogen and
methane within several transformer units, measurements of partial discharge
(PD) activity were introduced. It was determined that PD measurements using
the conventional method at the bushing taps alone was not sensitive enough
because of high external noise, so PD measurements using the ultra-high fre-
quency method were additionally proposed for the new, recently commissioned
transformer units. The advantages provided by both methods were combined
into one PD monitoring system to better identify and localize the source of PD.
This modern PD monitoring system employs advanced techniques of multi-
channel PD source and noise separation. The modular design and user-friendly
management of the system are also described.

Keywords: Shell GSU transformers
UHF monitoring
Conventional PD

method

1 Introduction

The Salto Grande hydro power plant feeds 1890 MW into the 500 kV systems of
Argentina and Uruguay. The project is located on the Uruguay River along the
Argentina-Uruguay border (Fig. 1). It includes a 500 kV transmission system and an
unified control center for the power house and associated transmission system. Brown
Boveri delivered the complex equipment for the two original control centers, including
the control systems for the 500 kV networks, the generator bus ducts and circuit
breakers, 6.6 kV switchgear, distribution transformers, and other equipment.
Eight Generator Step-Up banks, for a total of 24 mono-phase generator step-up
transformers (GSU), 100 MVA each unit, raise the voltage level to the 500 kV system
voltage. The first transformer unit was commissioned in 1979 and the last in 1982. It is
operated by Comision Tecnica Mixta de Salto Grande. In the Salto Grande substation,
70 500 kV transformers and reactors are installed.
The first detections of partial discharges (PDs) were performed with dissolved gas
analysis (DGA) method in 1992 on several GSU units. High values of hydrogen and
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 521–533, 2020.
https://doi.org/10.1007/978-3-030-31676-1_50
522 E. Briosso et al.

Fig. 1. Salto grande hydro power plant.

methane were detected. The PDs, probably located at the static plate/shield resulted to
be non-destructive since the units are still in service. Since 2006, the PD measurements
using the conventional method on bushing measuring taps were periodically performed
with portable equipment [1].
The reduced insulation dimensions with lower safety coordination factors in the
new designed units of transformers and reactors imposed the necessity to apply modern
PD measurement and continuous monitoring systems.
The sensitivity of the on-site partial discharge (PD) measurements is strongly
limited by the high noise level. Therefore, the high electromagnetic interferences with
the values close to the maximum acceptable PD level in the HV equipment made on-
site application of conventional method very difficult. In this paper, the advanced
features based on synchronous multi-channel techniques for noise and PD signal
separation are presented, and the experience gathered from PD monitoring is shown.
The large part of insulation defects develops over the lifetime of a transformer. To
detect such changes at an early stage, information on the actual insulation condition is
needed [4]. This information can be derived from PD monitoring during the equipment
operation.
To increase the sensitivity of the PD measurements, it was decided to install ultra-
high frequency (UHF) PD sensors, on newly-installed transformers. The PD activity is
monitored based on the conventional method at the bushing measuring taps and the
unconventional UHF method by placing a sensor inside the transformer tank. These
methods have a complementary character.
The conventional-type measurements indicate the defects both in bushing and
winding insulation, while the UHF measurements are only sensitive to detect winding
problems [5, 6].
Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV GSU Transformers 523

The continuous PD monitoring started a year ago on three new GSU single-phase
units. In this paper, the first experience from the PD monitoring performed with such
system is described. Special attention is paid to the monitoring system calibration and
to the proper settings of the system parameters to obtain sensitive measurement results
in a noisy environment.

2 PD Monitoring System in Salto Grande

PD monitoring is nowadays a worldwide accepted method for the on-line condition-

based assessment of power transformers. Different solutions for sensors and acquisition
techniques have been recommended, applied and lately acknowledged by international
technical bodies [5, 7]. Therefore, there is a strong need for a versatile PD monitoring
system that can be applied to different types of PD sensors.
The advantage of having a fully digital system with synchronous measurements
between the channels and freely selectable measurement frequency is shown. The
settings can be easily remotely adjusted.

2.1 Hardware
The architecture of the monitoring system is shown in Fig. 2. At each transformer
bushing, one capacitive tap sensor is installed with a tap adapter to measure PD,
capacitance and dissipation factor. The sensor is equipped with multiple redundant
protections to guarantee safe operation. The ambient temperature and relative humidity
are also measured within the same tap sensor to compensate the measured values.
Two UHF antennas are inserted inside the oil drain valves, one located at the top
part and the other one at the bottom of the tank. The antennas are connected to two
UHF down converters and finally to the fourth channel of the acquisition units (two
acquisition units per transformer). The connection between the acquisition unit and the
server is done via two master control units. The monitoring server is installed in the
control room.
During the commissioning, calibration of the monitoring system was performed
and the level of alarms and measuring intervals was selected in the monitoring soft-
ware. The threshold values for the measured parameters were chosen based on the
recommendations in the standards and from experience.

2.2 Software
The monitoring software (Fig. 3) runs on a central computer (server). It can either be
directly accessed on-site or remotely using a trusted network via the convenient web-
based interface.
The software is a management tool for system configuration, continuous data
acquisition and the visualization of real-time and stored measurement data. The
monitoring system provides data from the acquisition unit in two modes: permanent
and periodic mode.
524 E. Briosso et al.

Fig. 2. Monitoring system architecture.

Fig. 3. Real-time data displayed in the graphical user interface.

Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV GSU Transformers 525

During the permanent mode, the data is acquired every second, compared with
threshold values and displayed in the graphical user interface in real time. When this
data is within normal margins, it will be colored in green. If the values exceed
thresholds for “warning” or “alert”, they are accordingly colored in yellow or red.
The periodic measurements are initiated in equidistant time spans, e.g. every hour.
The duration of the periodic measurement is normally one minute. During this time
span, all relevant scalar values (charge, repetition rate, voltage phase, time stamp) are
calculated and PRPD (Phase-Resolved PD) diagrams as well as the matching 3PARD
(3-Phase Amplitude Relation Diagram) are acquired. This data is saved for later post-
processing and trend visualization.

3 PD Methods and Their Advantages

3.1 Conventional PD Monitoring

The PD signal is synchronously detected at the measuring tap of the bushing via a
special coupling unit. It includes several levels of redundant over-voltage protections
and the measuring tap is automatically grounded when the communication with the
acquisition unit is interrupted. Trend and traditional PRPD patterns are recorded and
used for detailed data evaluation (Fig. 4) [8].
This helps to identify the PD defect type and gives its rough location. The PRPD
patterns are complex with overlapped signals from different PD sources.
In order to separate clusters of different PD sources from noise, a synchronous
multi-channel PD evaluation technique known as 3PARD is applied [8, 9].

Fig. 4. Three-phase PD trend with example of PRPD and 3PARD diagrams.

3.2 Unconventional UHF PD Monitoring

The UHF PD signal is measured in the frequency range from 0.1 to 2 GHz by antenna
type sensors (two for each transformer) installed inside the tank. In order to obtain more
information in the frequency sweep diagram (Fig. 5), two spectra of the signal are shown.
526 E. Briosso et al.

The upper spectrum is based on the maximum amplitude of the time domain signal
acquired at each value of the frequency during the sweep. The lower spectrum corre-
sponds to the minimum amplitude. PD activity is always visible on the upper spectrum,
while noise from external interferences (e.g. corona discharge, radio waves and GSM) is
visible on both spectra within different frequency bandwidths (e.g. Fig. 5) [5].

Fig. 5. UHF signal spectra [5].

While the transformer was de-energized, an off-line frequency sweep was per-
formed during the installation of the monitoring system. The off-line spectra give
information about the interference sources produced by other equipment in the sub-
station. Additional electromagnetic interferences can couple to the measurement system
at operating voltage [8].

4 Settings of Monitoring System Parameters

4.1 System Calibration

Conventional PD detection requires a calibration of the measuring system according to
[10]. To fulfill this requirement, PD pulses of a known charge are injected in the current
path of the bushing and measured at their taps.
The ratio between the known charge and the measured signal is the calibration
factor. Example of calibration factors at different center frequencies are presented in
Table 1 for Transformer R.

Table 1. Calibration factors for Transformer R. Injected charge 500 pC; measurements at
different center frequencies and bandwidth of 300 kHz.
Center frequency [MHz] 1 4 7
Calibration factor 13.0 6.7 1.9
Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV GSU Transformers 527

The measurement frequency can be freely adjusted within the signal FFT area (Fig. 6)
and tuned in the area with an acceptable noise level (e.g. center frequency of 4 MHz).
Different types of defects generate signals of different apparent charge magnitude
and PD pulse repetition rate. The decision about the criticality of measured PD signals
requires analyses of PD patterns.

Fig. 6. FFT spectrum for noise signals and measurement frequency tuned to 4 MHz.

Unconventional UHF PD measurements with wide or narrow bandwidths can be

applied. In wide bandwidth measurements, the magnitude of the UHF signal is mea-
sured in the frequency range from 0.1 GHz to 2 GHz. In this case, the noise can
influence the result of measurements. To overcome this, narrow band measurements are
performed in the selected frequency range.
When the transformer is out of service, the UHF signal spectra are acquired (Fig. 7)
and the external noise is identified. The center frequency and the bandwidth of the
measurements are selected in the “clean” area of the spectra.
The spectra are verified when the transformer is back in service (Fig. 8) and the
frequency range of the measurements can be adjusted, if necessary. However, the
spectra can change, depending on the intensity of different noise sources.

Fig. 7. UHF frequency sweep at the top UHF sensor; Transformer S is out of service.
528 E. Briosso et al.

Fig. 8. UHF frequency sweep at the top UHF sensor; Transformer S is in service.

4.2 Settings of Conventional PD System

Different PRPD diagrams were compared to find the best system settings for different
center frequencies at operating voltage. Results are presented in Table 2. Three types of
noise signals have been identified (Table 2, Figs. 9 and 10).

Transformer R Transformer S

Transformer T 3PARD

Fig. 9. PRPD diagrams for the three transformers and the corresponding 3PARD diagram for
fc = 4 MHz.
Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV GSU Transformers 529

Table 2. Conventional method, identification of different noise signals for Transformer S.

Noise
fc [MHz] Continuous [nC] Floating potential [nC] Surface discharge [nC]
1 4.6 Not visible Not visible
4 0.08 0.4 1.9
7 0.2 0.8 0.9

A center frequency of 4 MHz was a compromise. At this frequency, the continuous

noise and floating potential noise signals are low. Figure 9 presents the PRPD diagrams
corresponding to a measurement frequency of 4 MHz for the three transformers as well
as the related 3PARD diagram.
Figure 10 describes the use of the 3PARD tool to separate different PD sources and
noise signals. Surface discharge and floating potential were identified as noise. These
signals are coupling to all units.

Back transformation to PRPD of selected

3PARD cluster; Transformer S as most probable
source

Back transformation to PRPD of selected

3PARD
cluster; similar coupling to all transformers

Fig. 10. Use of 3PARD tool to separate different PD sources and noise.
530 E. Briosso et al.

4.3 Setting of UHF System

Figure 11 presents the frequency sweep signals as well as the PRPD patterns of the
center frequency (fc = 120 MHz and 250 MHz, Df = 70 MHz) for Transformer S,
which correspond to the UHF sensor at the top of the transformer.
The UHF PD measurement identifies the floating potential signal source, when
measuring with 120 MHz center frequency, but not the surface discharge signal
(Table 3). However, when adjusting the frequency to 250 MHz and 500 MHz, the
floating potential pattern is no longer visible. The measured signal is visible for the
measurement frequency of 120 MHz with all six UHF sensors installed at the top and
bottom part of the transformers.

fc = 120 MHz fc = 250 MHz

Δf = 70 MHz Δf = 70 MHz

Fig. 11. Example of Frequency sweep spectrum and PRPD diagrams for Transformer S with top
UHF sensor at different center frequencies.
Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV GSU Transformers 531

Figure 12 presents the PRPD patterns for a center frequency (fc = 250 MHz) for
two timestamps. If the noise signals are stronger, then the signals can be visible also at
a center frequency of 250 MHz.

fc = 250 MHz fc = 250 MHz

Δf = 70 MHz Δf = 70 MHz
January 10th, 2019 January 16th, 2019

Fig. 12. Example of PRPD diagrams for Transformer S, at different moments with top UHF
sensor.

5 Current Status of the Monitoring System

The monitoring system is in operation with the optimum frequency settings for the PD
measurements using conventional and UHF methods.
The detected signal is classified as noise and its value is monitored. The mea-
surement examples performed in February 2018 and in January 2019 are shown for
comparison in Fig. 13.
There is no internal defect activity detected by the PD measurements. The DGA
measurements performed in January 2019 confirm this statement (Table 4) [2, 3].

Table 3. Identification of different noise sources using the UHF method at different center
frequencies on Transformer S.
fc [MHz] Position of UHF sensor UHF signal [µV] Detected noise pattern
Df = 70 MHz
120 MHz Top 1300 Floating potential
Bottom 1700 Floating potential
250 MHz Top 7.3 No noise
Bottom 7.5 No noise
500 MHz Top 7.9 No noise
Bottom 7.5 No noise
532 E. Briosso et al.

Table 4. DGA results for Transformer S, January 2019.

Gas type Concentration [ppm]
Hydrogen (H2) 3.6
Methane (CH4) 16.6
Ethane (C2H6) 67.8
Carbon monoxide (CO) 84.4
Carbon dioxide (CO2) 1489

February 2018

January 2019

Fig. 13. Example of the PD measurements (PRPD and 3PARD) using the conventional method
performed on the three single-phase units.
Partial Discharge Monitoring of Single-Phase Shell-Type 500 KV GSU Transformers 533

6 Conclusions

The complementary character of the conventional (IEC 60270) and unconventional

(UHF) methods for PD data evaluation is confirmed. A proper selection of the mea-
surement frequency was crucial to identify different noise sources, while the multi-
channel and synchronous technique (3PARD) helped to separate them.
A proper setting of the monitoring system parameters allows users to tune the
system to the optimum PD signal-to-noise ratio range. A high sensitivity of the PD
measurements is obtained, so incipient internal defects can be detected.

References
1. Briosso, E.: Acciones realizadas en equipos con diagnostico de Descargas Parciales.
Experiencia de Salto Grande. In: IEEE EPIM (2018)
2. IEC 60599 Mineral oil-impregnated electrical equipment in service – Guide to the
interpretation of dissolved and free gases analysis
3. IEEE Std C57.104-1991 IEEE Guide for the Interpretation of Gases Generated in Oil-
Immersed Transformers
4. Badicu, L.V., Koltunowicz, W., Piccolo, A., McGuigan, A., Feely, C.: Monitoring of a
distribution transformer at winchelsea substation. In: Proceedings of TechCon Asia-Pacific
2014, pp. 259–268, 1–2 April 2014, Sydney, Australia (2014)
5. CIGRE Technical Brochure 662: Guidelines for Partial Discharge Detection Using
Conventional (IEC 60270) and Unconventional Methods. CIGRE Working Group D1.37,
August 2016
6. Badicu, L.V., Broniecki, U., Koltunowicz, W., Körber, S., Krüger, M., Voegel, E.:
Prevention of transformer failure through continuous monitoring. In: Paper 274, Proceedings
of the 19th International Symposium on High Voltage Engineering (ISH) in Pilsen (2015)
7. IEEE C57.113-2010: IEEE recommended practice for partial discharge measurements in
liquid-filled power transformers and shunt reactors
8. Koltunowicz, W., Badicu, L.V.: Challenges in monitoring of power transformers. In:
Proceedings of My Transfo, November 2016
9. Koltunowicz, W., Plath, R.: Synchronous multi-channel PD measurements. IEEE Trans.
Dielectr. Electr. Insul. 156, 1715–1723 (2008)
10. IEC 60270: High-voltage test techniques - Partial discharge measurements. International
Electrotechnical Commission, Edition 3.1 (2015)
 Partial Discharge Phenomena in Power
Capacitor Unit Insulation Under Harmonic
Resonance Effects

Chongrag Boonseng1(&) , Rapeepornpat Boonseng1(&) ,

Nutthaphan Boonsaner1(&) , Vijit Kinnares1(&),
Promsak Apiratikul2(&), and Kunyanuth Kularbphettong3(&)
1
Faculty of Engineering, King Mongkut’s Institute of Technology
Ladkrabang (KMITL), Ladkrabang, Bangkok 10520, Thailand
{chongrag.bo,59011122,kkvijit}@kmitl.ac.th,
nut_work8@hotmail.com
2
Faculty of Engineering, Rajamangala University of Technology,
Pathumthani 12130, Thailand
promsaka@gmail.com
3
Faculty of Science and Technology, Suan Sunandha Rajabhat University,
Bangkok 10300, Thailand
kunyanuth@ssru.ac.th

Abstract. The capacitor in the electrical system for power factor correction is
always damaged when under voltage and current waveforms are distorted. There
was a test found that the electrical insulation was damaged and destroyed. In
order to learn this damage mechanism it requires a non-destructive inspection
method. The partial discharge (PD) testing is a method that can detect deteri-
oration and a defect in the electrical insulating material that is not visible. The
purpose of this paper is to investigate the impact of harmonic resonance on
power capacitor unit insulation. High voltage capacitor units consist of parallel
and series connected elements. The rating of element of capacitor units applied
in the power system is 6.67 kVAR/Phase, 400 V. Series resonance can lead to
high voltage distortion levels at a low voltage side of transformer resulting in
high harmonic current in the element of capacitors. The inspection and element
of capacitor unit test are used to obtain power system resonance, a standard
operating procedure, testing and measurement between phase to ground using
the partial discharge model MPD 600 OMICRON, dielectric losses and
capacitance model CPC 100 OMICRON. Finally, the scanning electron
microscope (SEM) is used. The comparison of results between new element of
capacitor and element under harmonic resonance are performed. The results
convey that breakdown in insulating of capacitors under harmonic resonance
can be used for considering capacitor voltage rating. The failure of insulation an
increase in obtained and emphasized by photograph from SEM.

Keywords: Partial discharge
Breakdown of insulation
Harmonic resonance

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 534–548, 2020.
https://doi.org/10.1007/978-3-030-31676-1_51
 Partial Discharge Phenomena in Power Capacitor Unit Insulation 535

1 Introduction

Electrical breakdown related with the failure of solid insulating materials and used
inside capacitors in the electricity distribution grid, usually resulting in a short circuit or
a blown fuse. Determining whether power capacitors suffer from internal arcing or
dangerous levels of partial discharge (PD) is important because failure without warning
can result in damage to neighbouring equipment, customer dissatisfaction, disruption to
economic activity and the imposition of regulatory fines. PD, as defined by IEC 60270,
are localized dielectric discharge in a partial area of a dielectric insulation system under
high voltage field stress [1, 2]. Effective asset management of these capital assets is
therefore a key business objective of generation, transmission and distribution com-
panies. An imperative that is driving the research and development of advanced con-
dition monitoring systems and diagnostic tools PD occurs within a capacitor where the
electric field exceeds the dielectric strength of the insulation. Possible causes include
insulation damage due to over voltages and lightning strikes, incipient weakness due to
manufacturing defects, or deterioration caused by natural ageing processes.
Although PD may initially be quite small, it is by nature a damaging process that
causes chemical decomposition and erosion of materials. Left unchecked, the damaged
area can grow, eventually risking electrical breakdown.
This paper focuses on the series resonance causing high harmonic distortion which
may affect significantly typical element of power capacitor, such as low voltage self
healing capacitors used for reactive power compensation. Moreover, the effect of
distortion voltage and current generated by AC drives on electrical insulation of
capacitor is also investigated by taking the capacitors that fall under series resonance
for 60 days and checking by measuring capacitance, dielectric losses, PD and SEM.

2 General Background

In the electrical system, there is a phenomenon that destroys electrical insulation and
equipment such as overvoltage, over current, transient and lightning. Electrical insulation
that has been severely damaged by such phenomena may be deteriorated, but may begin
to deteriorate and may be damaged at a later time. Electrical, mechanical, thermal effects
and environment-specific stresses during operation cause aging in high-voltage insula-
tion. PD can occur in the resulting local defects, the corrosive effect of which damages the
insulation progressively [3]. Preventive maintenance on high-voltage insulation requires
efficient diagnostic methods. In addition to the various integral current measurement
(measurement of discharge and charging current, as well as dielectric loss factor and
capacitance) PD measurement also been performed increasingly over the last years.
These measurements enable the more exact analysis of defects in the insulation with
regard to cause and location of the damage. Large power converters are the main factor
responsible for voltage and current distortion in electrical networks. These harmonic
currents cause harmonic voltage drop along the line impedance distorting the voltage at
bus. Supply and bus voltage distortion can be magnified further by the parallel resonances
[4]. In additional, there is a rule that the harmonic current flows from outside of the
transformers. These can draw harmonic current through the transformer by the capacitor,
resulting in the same high voltage distortion, also known as a series resonance. Which can
536 C. Boonseng et al.

happen often. In industrial estates, there are many factories, if the control of harmonic is
not good, there may be a flow of the harmonic currents between the transformers in the
system, resulting in the capacitor deteriorate and being damaged.

3 Harmonic Resonance and Effects

3.1 System Configurations

In Fig. 1, there is the one line diagram of utility substation supplying 3 factories. On
transformers TR1-TR2, the loads consists partly of AC drives but on transformer TR3
there is normal ac motor loads only. Reactive power compensation was made using
automatically controlled capacitor banks without any reactor. The manufacturer of the
capacitor banks was informed that several capacitor units and fuse bases were thermally

22kV

TR1 TR2 TR3

4MVA 2000kVA 1600kVA
22kV/6.6kV 22kV/400V 22kV/400V

PM

AC AC
Drive Drive M M M
3 x 200kW Capacitor Banks
2.5 MW 650 kW AC Motor 8 x (2x20) kVAR

Fig. 1. One line diagram of power systems

Fig. 2. Measured %THDv and %THDi at PM

 Partial Discharge Phenomena in Power Capacitor Unit Insulation 537

damaged on the TR3. A harmonic measurement was carried out. The voltage and
current distortion value of PM (point of measurement) on the transformers TR3 are
shown in Fig. 2. According to phase A of the systems, when capacitor bank is switched
on the %THDv and %THDi is increased from 4.36% and 1.63% to 9.52 and 12.82%
respectively which show the existence of resonance in electric system.

3.2 Series Resonance

In case of series resonance, the voltage of the upstream network which is distorted the
series resonance circuit, formed by capacitance of the capacitor bank and the short
circuit inductance of supplying transformer, resulting in drawing high harmonic current
through the capacitors. Series resonance can lead to high voltage distortion levels at
low voltage side of transformer. For example of series resonance circuit referring to
Fig. 3. The 5th harmonic current in the supply is 8 A. Harmonic measurement shown
abnormal high voltage distortion of 9.52% at 400 V bus and the 5th harmonic current
in the supply was 96 A. In this voltage resonance case, amplification factor reached a
high value of approximately 12. The capacitor unit of 20 kVAR under resonance
condition has been PD tested.

5th Harmonic Current

22kV

TR1 TR2 TR3

4MVA 2000kVA 1600kVA
22kV/6.6kV 22kV/400V 22kV/400V

AC AC
Drive Drive M M M
3 x 200kW Capacitor Banks
2.5 MW 650 kW AC Motor 8 x (2x20) kVAR

Fig. 3. The series resonance of electrical system

4 Experimental Setup

4.1 Capacitance and Dielectric Losses

The CPC 100 is powerful testing devices, this makes testing with the CPC100 a time-
saving and cost-effective alternative for conventional testing methods. Schering bridge
is one of the very important ac bridges to measure capacitance and dielectric loss of
high voltage apparatus as shown in Fig. 4. The test objects are power capacitor 20
538 C. Boonseng et al.

Fig. 4. Capacitive and tand test circuit

Fig. 5. Capacitor under investigation and capacitance and tand test device

a) Without resonance condition b) With resonance condition

Fig. 6. Example of test result

kVAR 400 V which are a new capacitor (Capacitor I) and the capacitors under har-
monic resonance condition (Capacitor II). The Capacitor under investigated and
capacitance and tan d test devices are shown in Fig. 5. The results are shown in Fig. 6.
 Partial Discharge Phenomena in Power Capacitor Unit Insulation 539

4.2 Partial Discharge Measurement

PD measurement is a non-destructive insulation diagnostic method that can be per-
formed offline. Basic PD test circuit is shown in Fig. 7. The methods have proved to be
more appropriate for the detection of PD in transformer system, and they are, generally,
categorized as conventional and nonconventional test techniques [1, 5]. The test objects
are always connected to a high voltage source and a coupling capacitor. The experi-
mental setup for the PD measurement under laboratory conditions is shown in Fig. 7.

Fig. 7. PD test circuit

The measurements were realized on the basic of the standard. The MPD 600 partial
discharge analysis system is an acquisition and analysis toolkit for detecting, recording,
and analyzing partial discharge events in many applications. It is suited for laboratory
measurements of high-voltage systems, instrument and power transformers, and
rotating machines. It is controlled by the integrated mtronix software featuring real-
time visualization and analysis options of PD detection and system parameters.

4.3 Scanning Electron Microscope

SEM enables a clear observation of very small structures. The SEM is routinely used to
generate high-resolution images of shapes of objects and to show spatial variations in
chemical compositions. The SEM is also widely used to identify phases based on
qualitative chemical analysis and/or crystalline structure. The capacitor insulation is
composed of polypropylene (metalized) film, which insulation level is 4/15 kV.
The SEM introduced here utilizes an electron beam whose wavelength is shorter than
of light and therefore observing a structure of polypropylene to several µm in scale
become possible as shown in Fig. 8.
540 C. Boonseng et al.

Fig. 8. SEM model JSM series, JEOL

5 Results

5.1 Capacitance and Dielectric Losses

The quality of a capacitor is usually expressed in terms of capacitance and dielectric
loss angle defined as the angle by which determine the insulation dielectric power loss
by measuring the power angle between an applied ac voltage and the resultant current.
The lower the quality of the insulation condition, the more resistive it will appear and
the more power loss will be dissipated through it, and thus the dielectric loss angle will
be high. The comparison of results of the test between new capacitor and the capacitor
under harmonic resonance are shown in Table 1.

Table 1. Measurement results of the capacitor

Capacitor No.of test Test voltage (Volt) Capacitance (pF) tan d (%)
Capacitor I 1 504 11.114 26.3200
2 996 11.227 27.3391
Capacitor II 1 504 5.111 34.4310
2 997 5.808 31.5738
 Partial Discharge Phenomena in Power Capacitor Unit Insulation 541

5.2 Partial Discharge

OMICRON‘s MPD600 is a high-end measurement and analysis system for partial
discharges. The MPD 600 system consists of a measurement unit, a USB controller and
sophisticated analysis software. The modular plug-and-play system enables a number
of state of-the-art display and assessment features and achieves outstandingly high
measurement accuracy. The general requirements for PD measurement according to the
IEC standard 60270 consider the PD circuit, the measuring instruments, the calibration
and the measuring procedures [6].

Fig. 9. PD pattern of Capacitor I applied for phase A

Fig. 10. PD pattern of Capacitor I applied for phase B

542 C. Boonseng et al.

Fig. 11. PD pattern of Capacitor I applied for phase C

The experimental results of PD measurements of the capacitor I and II are shown in

Figs. 9, 10, 11, 12 13 and 14. Figures 9, 10 and 11 depict the PD patterns that were
measured under applied voltage around 2.4 kV, mostly PD not appear and with the
background noise appearing. Figures 12, 13 and 14 show the results started to show PD
patterns under applied voltage to 1.242 kV in phase C.

Fig. 12. PD pattern of Capacitor II applied for phase A

 Partial Discharge Phenomena in Power Capacitor Unit Insulation 543

Fig. 13. PD pattern of Capacitor II applied for phase B

Fig. 14. PD pattern of Capacitor II applied for phase C

5.3 Scanning Electron Microscope

We examined the polypropylene (metalized) film with the thermal field emission
scanning electron microscope. The SEM images are shown in Fig. 15. The localization
of damaged areas are shown in Fig. 15, d, e and f which the insulation is penetrated by
5th harmonic current from series resonance processing along 60 days. The positions of
insulation that are penetrated have many points resulting in a reduction in the capac-
itance of capacitors which are reduced and an increase in the dielectric loss. When we
tested capacitor for measuring PD in insulation, PD can occur in the resulting local
defect. The corrosive effect damages the insulation progressively.
544 C. Boonseng et al.

Fig. 15. SEM image of Capacitor I & II

6 PD Analysis and Recommendations

Figure 16 shows the results of PD measurement of the number of PD generated within

60 s. Phase C has PD generated more than other phases because the insulator in phase
C are destroyed more than occur that in phase A and phase B. The experimental result
is a data of capacitor II, while the capacitor I has not.
 Partial Discharge Phenomena in Power Capacitor Unit Insulation 545

10
Phase A Phase B Phase C
8

6
QIEC (pC)

0
0 10 20 30 40 50 60
Time(Sec)

Fig. 16. QIEC test results with time recording of capacitor II

Table 2 shows comparison between the results of PD measurement in capacitors I

and II. The capacitor I was applied with voltage of 2.2–2.4 kV. There is no PD because
of the insulations are perfected and not damaged. The capacitor II was installed in
harmonic resonance system for 60 days after that PD measurement is made. The
capacitor II has PD occurrence with applied voltage around 1.1–1.2 kV because of the
insulator due to deteriorated and damaged states of the insulator.

Table 2. Measurement results of the capacitor I, II

Capacitor Phase Duration QIEC QMax VRMS fv
(sec) (kv) (Hz)
Capacitor I A 66.9 570.42 fC 807.56 fC 2.39 50.0
(New units) B 73.7 600.41 fC 927.17 fC 2.37 50.0
C 67.4 824.24 fC 974.94 fC 2.36 50.0
Capacitor II A 65.4 1.34 pC 3.61 pC 1.25 50.0
(Units under B 68.9 2.10 pC 3.34 pC 1.20 50.0
resonance) C 75.6 1.84 pC 5.29 pC 1.10 50.1

6.1 Limitations of PD Measurement

6.1.1. The testing voltage level is approximately 2.5 kV. The supply transformer for
PD is 75 kV which is a big size. As a results non-sinusoidal waveform is achieved.
pffiffiffi pffiffiffi
6.1.2. The ratio between the value of peak voltage divided by 2 ( V= b 2) and the
RMS voltage (VRMS ) must be less than 5% as shown in Eq. (1), In Fig. 8 the ratio is
equal to 2.38%.
pffiffiffi
^
V 2
 5% ð1Þ
VRMS
546 C. Boonseng et al.

6.1.3. The QIEC value (the apparent charge as defined in IEC 60270) QIEC and
QIECMax obtained from the reading curve.
6.1.4. The aging deterioration and temperature effects may increase with this medium
voltage resonance case. Amplification factor reaches a high value. The amplification
factor is 12.

6.2 PD Behaviour
Initial discharge starts in phase A, where the PD level is not very high, about 1.8–2.4
pC and after increasing the applied voltage about 1.1–1.2 kV can maintain constant PD
levels duration of 1–20 s. Then the PD is as high as 3.8 pC and falls to about 1.8 pC. In
phase B, the PD occurs at a time of 20–24 s, with a value of approximately 1.7–3.6 pC.
After that, the PD does not occur until the end. The occurrence of PD in phase C starts
at 4 s, which is approximately 1.2 pC and disappears until 30 s, so it occurs as high as
5.2 pC and occurs throughout the last period. The development of highest PD occur-
rences at the time of 44–56 s of phase C has a continuous PD showing in Figs. 17, 18
and 19. From examining the result of the test of the electrical insulation in the
capacitor, there is a continuous PD occurrence. As it is possible, it can be concluded
that the electrical insulation has deteriorated and accelerated the formation of PD faster
and easier. For the result of 15 capacitors that under the same harmonic resonance
condition, it is found that the results of the examination for all capacitors with PD
appear all the same way.

Fig. 17. The capacitor II has PD level of 4.020 pC after applied voltage 1.134 kV for phase C
 Partial Discharge Phenomena in Power Capacitor Unit Insulation 547

Fig. 18. The capacitor II has PD level of 4.811 pC after applied voltage 1.321 kV for phase C

Fig. 19. The capacitor II has PD level of 5.251 pC after applied voltage 1.242 kV for phase C

7 Discussion

The influences of harmonics voltage distortion and series resonance at AC systems can
damage insulation systems, that the phenomena hasten the degradation of insulator and
discharge. Finally, the electrical insulation begins to become damaged. The electrical
insulation damage in the capacitor starts from the 5th harmonic current and the voltage
distortion which it causes a lot of heat. Due to the resonance phenomena in which the
heat of the insulation is high, it is the reason that causes the penetration between
electrical insulation in capacitor II as shown in Fig. 15d, e and f. When using the
sample insulation to check the PD, it is found that the initial PD is actually shown in
Figs. 17, 18 and 19. The relationship between the 5th harmonic current and the damage
of electrical insulation in the capacitor can be confirmed by the capacitance mea-
surement in the capacitor II. And likewise the 5th harmonic current and the series
resonances affect the waste and the deterioration of electrical insulation. Therefore, in
order to prolong operating electrical insulation life, the harmonic elimination and
wrestling still against resonance should be done urgently. The passive or active filter
can be used [7].
548 C. Boonseng et al.

8 Conclusions

Insulating material behaviour of polypropylene (metalized) film exposed to distortion

voltage results in electrical insulating being accelerated due to interactions between
applied voltage and electrical properties of dielectric. Effecting the electrical insulation
to be more than 70 °C and while the resonance current increased by 2–3 times.
Electrical insulation cannot be tolerated. The polypropylene (metalized) film is
destroyed in a short time. The nature of the destruction is a hole and a cavity. The
resulting electrical capacitance of the capacitor will be greatly reduced, If there is a high
resonance, some capacitors can be looked at the signs of damage from the outside. In
this monitoring system, resonance occurs frequently at a relatively high level. Damage
to capacitors is therefore very high. PD testing and measurement is another way to
evaluate and interpretation of electrical insulation damage.

References
1. IEC60270: High Voltage Test Techniques - Partial Discharge Measurements, 1st edn. British
Standards Institution (2000)
2. Treyer, P., Mraz, P., Hammer, U., Gonzalez., S.: Breaking the limit of power capacitor
resonance frequency with help PD pulse spectrum to check and setup PD measurement. In:
19th ISH 2015 (2015)
3. Siemens, A.G.: Generator Partial Discharge Measurement-Flyer (2007)
4. Fabiani, D., Montanari, G.C.: The effect of voltage distortion on ageing acceleration of
insulation systems under partial discharge activity. IEEE Electr. Insul. Mag. 17(3), 24–33
(2001)
5. IEC62478: High Voltage Test Techniques-Measurements of Partial Discharge by Electro-
magnetic and Acoustic Methods. British Standards Institution, 1st draft (2012)
6. IEC 60060-1: Ed.3.0 High-voltage test techniques- Part 1 (2010)
7. Boonseng, C.: Passive and active filter installation. Department of Electrical Engineering,
Faculty of Engineering, KMITL (2016)
 Partial Discharges Pulse Shape Analysis
at AC and DC

Josef Pihera(B) , Jaroslav Hornak, Ales Vobornik, Lukas Kupka,

Svatoslav Chladek, and Rainer Haller

Faculty of Electrical Engineering, Regional Innovation Centre for Electrical

Engineering, University of West Bohemia, Pilsen, Czech Republic
pihera@ket.zcu.cz

Abstract. Partial discharge (PD) signal analysis is a well-known tool

for monitoring the conditions of insulation systems of electrical machines
and equipment. Classical analyzes of PD are mostly based on the eval-
uation of the phase angle of the test voltage at which partial discharges
occur. At the present, the partial discharge detection is mostly based on
the measurement of charging currents that are needed to resume the volt-
age on the sample that was there before the partial discharge occurred.
Other electrical methods are, for example, the scanning of an electro-
magnetic field emitted during a partial discharge. Another developed
methodology in the field of measuring and analyzing partial discharges
is the monitoring of pulse parameters of the detected discharge. Exper-
iments suggest that measuring the rise and fall time of the pulse edges
and amplitude measurements can evaluate the partial discharge signal.
Especially this method could be very helpful at the moment when par-
tial discharges at DC are studied and evaluated. The disadvantage of
partial discharges at DC is the fact there is no phase and phase angle
position for PD patterns recognition. Thus, another evaluation method
development is necessary for PD recognition DC. There is presented in
the study an experiment where the PD pulse shapes, in the manner of
rise time and fall time, were investigated at laboratory PD models at
both voltage types AC and DC. The PD models simulated corona dis-
charges. Each individual pulse of a partial discharge can be detected
by a suitable scanning impedance and viewed with fast digital oscillo-
scopes. As has been said, there is a presumption that different discharge
arrangements or even variously degraded insulating materials have differ-
ent waveforms of the sensed pulses. The main monitored parameters of
individual pulses are their rise time and amplitude. Since the occurrence
of partial discharge is essentially a random phenomenon, statistical anal-
ysis of a significant amount of data is required. A view of the behaviour
of the partial discharge pulse in various cases of the discharge activity
brings different results in the observation of the pulse rise time and its
amplitude in the various observed setups. Monitoring the rise time and
pulse amplitude is one way of evaluating partial discharges, regardless of
their superimposed position on the sine wave of the supply voltage. Thus
this PD monitoring could be a part of PD evaluation at DC. These pulse
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 549–559, 2020.
https://doi.org/10.1007/978-3-030-31676-1_52
550 J. Pihera et al.

parameters together with other PD analysis at DC (as pulse sequence

analysis) could be used to estimate the partial discharge type as well as
to evaluate the deterioration of the insulation of electrical machines.

Keywords: Partial discharges · Pulse shape · Measuring impedance ·

Statistical analysis

1 Introduction

Detection and evaluation of partial discharges (PD) are important for quality
assurance and diagnosis of electrical insulation systems. With the increasing use
of DC voltages in electrical transmission and distribution systems the field of PD
under DC voltage stress needs to be deeply investigated, especially in the manner
of the field distribution in insulation systems used in DC voltage systems, and
the physical processes of the PD phenomena under DC voltage stress.
The well introduced diagnostic tools for PD measurement at AC cannot be
used easily under DC conditions, therefore, appropriate methods, procedures
and evaluation rules for PD measurement at DC should be developed.
It was investigated already for several years whether the introduced diagnos-
tic method of Pulse- Sequence- Analysis (PSA) [1,2] could be applied even at
DC conditions. But, there is still missing a key parameter of PD signal to have
the information about PD type. And because there is y no phase at DC thus
some other methods and parameters must be found.
The presented paper is based on the assumption that each individual pulse
of a partial discharge can be detected by a suitable measuring impedance and
viewed with fast digital oscilloscopes. A lot of studies (for example [3–5]) have
been performed for this purpose. The PD impulse shape may be affected due
to, e.g. insulation defect [6], selected test set-up [7], the presence of impurities
[8] or different connection of PDs detector [9]. It confirms a presumption that
different discharge arrangements or even variously degraded insulating materials
have different waveforms of the detected pulses.

2 Experiment
Specific details are provided in the following sections.
The PD measurement was performed using the basic arrangement - needle-
plane in the air with gap distance between tip and plane 10 mm. The tip radius
(needle) was ∼0.29 mm. The PMMA was used as the barrier between the tip
and plate (thickness 2 mm and relative permittivity r ∼ 3) (Fig. 1).
 Partial Discharges Pulse Shape Analysis at AC and DC 551

Fig. 1. PD test objects - corona (top), barrier corona (bottom)

The reference PD signal was measured using a standardized PD circuit

(Fig. 2) and measuring device PD Smart. Tables 1 and 2 show the test voltages
for corona arrangements testing. The used voltage was AC and DCnegative .

Fig. 2. Standardized PD test circuit - AC and DC H.V. source (≤200 kV AC, ≤135 kV
DC), Coupling capacitor, (C = 1000 pF ± 10%)

First, the main issue of the PD signal parameters is the fact the bandwidth
of measuring impedance (MI) determines the signal value of individual pulse
rise time (tr ). The different signal of corona streamer (6.87 kV) PD pulse using
different MI is shown in Fig. 3. As could be seen from the Fig. 3 there are different
rise times (tr ) of the signal when using different MI.
552 J. Pihera et al.

Table 1. Needle-plane-set

Setup Test voltage (kVpeak )

Corona AC (10 mm) - Trichel 6.36
Corona AC (10 mm) - Streamer 6.87
Corona DC (10 mm) 6.5

Table 2. Needle-plane-set with PMMA barrier

Setup Test voltage (kVpeak )

Corona PMMA AC (10 mm) - Trichel 6.5
Corona PMMA AC (10 mm) - Streamer 6.5
Corona PMMA DC (10 mm) 6.5

For the purpose of this study different MI were used to test the variation of
the signal. There were used the MI referred as PDX from PowerDiagnostix, PD
Smart measuring impedance designed and referred as LEMKE and finally, the
KET measuring impedance designed at UNI Pilsen and referred as KET MI.

Fig. 3. PD signal - different MI - full pulse shape (left), detail (right)

2.1 Hardware

There was tested several MI in the experiment. The results described further in
this study are those ones using the KET measuring impedance designed at the
UNI Pilsen. The impedance was build (Fig. 4) for the purpose of the study to
 Partial Discharges Pulse Shape Analysis at AC and DC 553

have as realistic PD impulse shape as possible. Due to this fact, the fall time of
the signal is not cut off.

Fig. 4. KET measuring impedance block diagram

The measuring impedance has been designed as a frequency crossover with

cut-off frequency at 100 kHz. Low-frequency output allows measurement of the
excitation signal with divider ratio 5000:1 or 50000:1 in the frequency range
from 10 Hz to 2 kHz. High-frequency output allows measurement of signals in
the range from 300 kHz to 140 MHz in tolerance 3 dB with basic attenuation of
6 dB (Fig. 5).

Fig. 5. KET measuring impedance frequency characteristic

554 J. Pihera et al.

Figures 6 and 7 show the KET MI output of Trichel impulse at 6.36 kVpeak
AC and the same voltage level at DC respectively. As mentioned above the total
pulse count for post-processing and statistical analysis were 1000 pulses.

Fig. 6. PD signal of Corona at AC (needle distance 10 mm) - Trichel impulse voltage

level 6.5 kVpeak

Fig. 7. PD signal of Corona at DC (needle distance 10 mm) - Trichel impulse voltage

level 6.5 kVpeak

2.2 Data Acquisition

The main monitored parameters of individual pulses are their rise time and
amplitude. And because the PD signal depends on the variety of conditions the
statistical analysis of a significant amount of data is required. For purpose of
this study, it was stored and analyzed 1000 impulses using the data acquisition
system and statistical technique described below.
For statistical analysis of measured data, a supporting application in Lab-
VIEW has been designed. The software controls a Tek MSO oscilloscope and
uses it as an input device for capturing the equivalent of PD signal. Sampled
data are transferred to the application for further processing. The storing of
measured data is done when the oscilloscope is re-triggered to catch the next
signal. The result of long term measurements is a file in CSV format and could
contain selected parameters of an acquired signal such as a timestamp of the
 Partial Discharges Pulse Shape Analysis at AC and DC 555

sample, rise time (tr ), fall time (tf ), amplitude, peak to peak values, impulse
area etc. (Fig. 8).

Fig. 8. LabView data acquisition front panel

2.3 Data Analysis

The stored data were consequently analysed using the statistical calculations
and test. Due to the structure of the measured data, scatterplots (xy graphs)
are the most useful tool for statistical analysis. These graphs include histograms
for displaying the probability of density.
At first, all measured data were cleaned from outliers and extreme values
before statistical analysis. These outlier values could negatively affect the results
of the statistical analysis and its interpretation. Tukey’s boxplot [10] is an effec-
tive tool for this purpose (see Fig. 9).

Fig. 9. Tukey’s boxplot [11]

Two histograms of important quantities (rise time and amplitude) were cre-
ated from the cleaned data set. A histogram is an accurate representation of the
distribution of numerical data. It is very useful because it is an estimate of the
probability distribution of a variable.
556 J. Pihera et al.

The presented graphs (Figs. 10, 11 and 12) are the scatterplots of rise time
and amplitude with histograms for both variables. It can be seen the dependence
of these variables.

3 Test Results
The test results obtained during this presented study are summarized in Tables 3
and 4. Table 3 presents data of corona and Table 4 the data of corona with the
barrier. The values in the tables are the mean value of statistical analysis. The
detailed data variation is shown in Figs. 10, 11 and 12.
The tests were provided at the same voltage level for ACpeak and DCnegative .
The inception voltage of Trichel and streamer impulses was measured first at
AC and consequently, these voltage levels were set for DC setup.

Table 3. Needle-plane-set experimental data

Setup Test voltage Rise time Amplitude

(kVpeak ) (ns) (mV)
Corona AC (10 mm) - Trichel 6.5 100 230
Corona AC (10 mm) - Streamer 6.87 80 1900
Corona DC (10 mm) 6.5 30 270
Corona DC (10 mm) 6.87 30 320

Table 4. Needle-plane-set with PMMA barrier experimental data

Setup Test voltage Rise time Amplitude

(kVpeak ) (ns) (mV)
Corona AC (10 mm) - Trichel 6.5 90 190
Corona AC (10 mm) - Streamer 6.5 85 1900
Corona DC (10 mm) 6.5 30 20–120

When looking at the data of corona (Table 3) there is seen the difference
between AC Trichel and streamer impulse. There is different inception voltage,
amplitude and risetime. This is not the same for DC, where the impulses at
different voltage levels have equal risetimes. The difference in signal amplitudes
is not as significant as in AC, compared the investigated voltage levels.
 Partial Discharges Pulse Shape Analysis at AC and DC 557

Fig. 10. Statistical data - corona at AC (needle distance 10 mm) - Trichel impulse -
voltage level 6.5 kVpeak (top), corona at DC (needle distance 10 mm) - Trichel impulse
- voltage level 6.5 kVpeak (bottom)

Fig. 11. Statistical data - corona at AC (needle distance 10 mm) - Streamer impulse
- voltage level 6.5 kVpeak (top), corona at DC (needle distance 10 mm) - Streamer
impulse - voltage level 6.87 kVpeak (bottom)

When analysing the results from Table 4 it can be seen several differences
compared to the corona without the barrier. First, the inception for Trichel and
streamer is at one voltage level. The AC risetimes are more or less equal. DC
risetime is significantly lower and equal to the AC voltage case.
The difference in amplitudes is evident; streamer impulse at AC has signifi-
cantly higher values than Trichel. The amplitude of the signal at DC varies from
20 to 120 mV (see detailed in Fig. 11). Here comes into effect the charging of the
barrier surface and additional electrical field forcing against the main field and
thus suppressing the PD magnitudes [12].
558 J. Pihera et al.

Fig. 12. Statistical data - corona PMMA at AC (needle distance 10 mm) - Trichel
impulse - voltage level 6.5 kVpeak (left), corona PMMA at AC (needle distance 10 mm)
- Streamer impulse - voltage level 6.5 kVpeak (center), corona PMMA at DC (needle
distance 10 mm) - voltage level 6.5 kVpeak (right)

4 Conclusion
The study shows the comparison of AC and DC pulses when using one particular
measuring impedance. There was shown the fact when using different measuring
impedances for one PD setup there exists different PD signal outputs. This fact is
due to the different design of measuring impedances from different manufactures.
Based on this fact it would be very difficult to compare the PD signal values,
especially risetimes, of different measuring impedances. But when looking for
the partial discharge type at DC, the additional parameter for advanced PSA
must be found.
This study shows a possible way of detecting several additional parameters
of PD signal and their statistical analysis.
Next study would be focused on the PD multiple sources setups such as
corona/surface PD and the clustering at AC and DC in multiple PD.

Acknowledgement. This research has been supported by the Ministry of Education,

Youth and Sports of the Czech Republic under the RICE New Technologies and Con-
cepts for Smart Industrial Systems, project No. LO1607 and by the Student Grant
Agency of the University of West Bohemia in Pilsen, grant No. SGS-2018-016 “Tech-
nology and Material Systems in Electrical Engineering”.
 Partial Discharges Pulse Shape Analysis at AC and DC 559

References
1. Fromm, U.: Interpretation of partial discharges at DC voltages. IEEE Trans.
Dielectr. Electr. Insul. 2, 761770 (1995). https://doi.org/10.1109/94.469972
2. Pirker, A., Schichler, U.: Partial discharges at DC voltage - measurement and
pattern recognition. In: 2016 International Conference on Condition Monitoring
and Diagnosis (CMD), pp. 287–290, Xian (2016). https://doi.org/10.1109/CMD.
2016.7757811
3. Altamimi, G., Illias, H.A., Mokhtar, N., Mokhlis, H., Bakar, A.H.A.: Corona dis-
charges under various types of electrodes. In: 2014 International Conference on
Power and Energy, (PECon), pp. 5–8, Kuching (2014). https://doi.org/10.1109/
PECON.2014.7062403
4. Kim, J., Kim, D., Nam, K., Choi, W., Lee, B., Koo, J.: Characteristics of partial
discharge by AC and DC. In: 2016 International Conference on Condition Moni-
toring and Diagnosis (CMD), pp. 489–492, Bali (2012). https://doi.org/10.1109/
CMD.2012.6416185
5. Klueter, T., Wulff, J., Jenau, F.: Measurement and statistical analysis of partial
discharges at DC voltage. In: 2013 48th International Universities’ Power Engineer-
ing Conference (UPEC), pp. 1–5, Dublin (2013). https://doi.org/10.1109/UPEC.
2013.6714912
6. Li, X., Wu, G., Zhang, X., Bian S.: Partial discharge pulse shape detection and
analysis under DC condition in typical defect models. In: 2007 Conference on
Electrical Insulation and Dielectric Phenomena (CEIDP), pp. 188–191, Vancouver
(2007). https://doi.org/10.1109/CEIDP.2007.4451593
7. Klueter, T., Wulff, J., Jenau, F., Wienold, D.: Evaluation of surface- and corona
discharges at DC voltage. In: 2013 13th International Conference on Environment
and Electrical Engineering (EEEIC), pp. 255–259, Wroclaw (2013). https://doi.
org/10.1109/EEEIC-2.2013.6737918
8. Li, X.G.Y., Zhang, Q., Li, J., Wang, T., Wang, Z., Gao, M., Ni, H.: Study on the
shape transformation of PD fibrous degradation products in oil-impregnated paper
insulation under DC voltage. In: 19th International Symposium on High Voltage
Engineering (ISH), Pilsen (2015)
9. Alvarez, F., Ortego, J., Garnacho, F., Sanchez-Uran, M.A.: A clustering technique
for partial discharge and noise sources identification in power cables by means
of waveform parameters. IEEE Trans. Dielectr. Electr. Insul. 23, 469481 (2016).
https://doi.org/10.1109/TDEI.2015.005037
10. Raschka. S: Dixons Q test for outlier identification a questionable practice. http://
www.grroups.com/blog/dixons-q-test-for-outlier-identification-a-questionable-
practice
11. McGill, R., Tukey, J.W., Larsen, W.A.: Variations of box plots. Am. Stat. 32, 1216
(1978). https://doi.org/10.2307/2683468
12. Akishev, Y.S., Demyanov, A.V., Karalanik, V.B., Monich, A.E., Trushkin, N.I.:
Comparison of the AC barrier corona with DC positive and negative coronas and
barrier discharge. Plasma Phys. Rep. 29, 8291 (2003). https://doi.org/10.1134/1.
1538505
 Digital Filtering Methods for Interferences
on Partial Discharges Under DC Voltage

Benedikt Hochbrückner1(&), Martin Spiertz1, Andreas Küchler1,

Markus H. Zink1, Thomas Steiner2, Erik Winkelmann2,
and Karsten Backhaus3
1
University of Applied Sciences Würzburg-Schweinfurt,
Ignaz-Schön-Strasse 11, 97421 Schweinfurt, Germany
benedikt.hochbrueckner@fhws.de
2
HIGHVOLT Prüftechnik GmbH, Marie-Curie-Strasse 10,
01139 Dresden, Germany
3
Technical University Dresden, 01069 Dresden, Germany

Abstract. In this contribution, a new separation method for partial discharge

signals and disturbances is introduced. In contrast to frequency rejection filter
techniques like low or high-pass filters as de-noising methods, a correlation filter
is used to separate the acquired signals into different signal classes. For that, a
test circuit for the measurement of partial discharge signals under dc voltage was
set up. Two test samples for the generation of corona and surface discharges as
typical partial discharge sources were used to obtain test signals for the pre-
sented classification algorithm. Besides partial discharge signals, additional
disturbances from the power supply, which was realized by the application of a
half-wave rectifier, were also measured. Spectral analyses of the measured
signals show the difficulties when using common frequency-rejection filter
methods since the relevant spectral contents of discharge signals and distur-
bances are located in the same frequency range. An application of these filter
methods for de-noising would influence the waveform of the partial discharge
signals and impede further analyses due to a limitation respectively a loss of
their relevant spectral contents. The new approach provides a separation method
for different signals without an influence on the pulse shape by using a corre-
lation filter method. Histogram analyses of the correlation coefficients show a
clear differentiation of the signal classes and present thresholds for the dis-
tinction between correlated and uncorrelated events. The consistence of each
single signal class is proven to ensure a high similarity of the events obtained
from the same test sample. The correlation between two signal classes lead to
coefficients below the determined threshold value, which indicates a comparison
of events from different classes. In a last experiment, the results from the his-
togram analysis were implemented in a classification algorithm to separate the
raw data into specific classes and presents a new approach for the grouping of
unknown signals in comparison to multi-channel processing tools like the 3
Center Frequency Relation Diagram (3CFRD).

Keywords: Partial discharges
DC
Filtering

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 560–571, 2020.
https://doi.org/10.1007/978-3-030-31676-1_53
 Digital Filtering Methods for Interferences on Partial Discharges 561

1 Introduction

In history, power transmission was built up as AC voltage systems to transport elec-

trical energy over long distances. This kind of energy transmission has the disadvan-
tage of high reactive power losses due to the inductive and capacitive effects of the
transmission line. A change to a DC voltage transmission system for the highest
voltage levels is able to overcome these problems and ensure a more efficient transport
of the electrical energy.
High voltage equipment like cables, transformers and bushings have to be tested
according to valid standards. The evaluation of the insulation system quality of high
voltage devices is mainly performed with a partial discharge test according to valid
standards like DIN EN 60270 [1]. While there is a reliable and practicable way for the
diagnosis of insulation defects by the use of the Phase Resolved Partial Discharge
Diagram (PRPD) for AC, such a method is still missing for the diagnostics on PD
under DC voltage [2]. In this case, single discharge impulses have to be analyzed in
time and frequency domain to differentiate various PD sources from disturbances. The
evaluation of PD measurements can be influenced by the presence of noise and
especially pulse-shaped disturbances, so an effective de-noising method has to be used
for the suppressing of unwanted signals as a first step in data processing [3]. Figure 1
shows measured partial discharge impulses from the test samples and an additional
disturbance signal.

Fig. 1. Measured discharge signals with disturbance (2500 samples, 10 µs).

Hence, many filter techniques can be applied [4] but some like mainly frequency-
rejection filters lead to a distortion of the waveform of partial discharges (PD). Due to a
high similarity between PD discharges and disturbances in frequency domain [3, 5, 6].
In order to avoid a loss of relevant signal information, new techniques like wavelet
transform filters [5] or the 3 Center Frequency Relation Diagram (3CFRD) [3, 7] were
562 B. Hochbrückner et al.

introduced. A further possibility for de-noising is described in [8] as a “nearest-

neighbor”-method by the use of heuristic signal features from the signal amplitude for
the classification of unknown events.
A new approach for the distinction of disturbance and PD signals as an efficient de-
noising method is introduced by using a combination of histogram analysis on cor-
relation coefficients and a matched filter algorithm as a closed loop de-noising system
[9]. In contrast to the usage of complex filter techniques like wavelet transform, the
focus of this contribution lies on the application of a matched filter and the determi-
nation of suitable thresholds for an effective signal differentiation.

2 Correlation Filter Methods

As an elementary operation in digital signal processing, the correlation indicates the

similarity between two signals [10]. The mathematical operation for the discrete cor-
relation is very similar to the discrete convolution, where rxy[k] represents the cross-
correlation function between two discrete signals x[n] and y[n] (1).
X1
rxy ½k ¼ n¼1
x½ny½n þ k ð1Þ

By using the Fourier transform, the rxy[k] can be determined by the inverse Fourier
transform of the so-called cross-energy density spectrum (2).

rxy ½k ¼ F 1 ðF  ðx½nÞ
F ðy½nÞÞ ð2Þ

To obtain applicable values for the similarity of two correlated events, a normal-
ization of rxy[k] on the signal energies Ex and Ey has to be performed. This leads to the
normalized cross-correlation coefficient uxy[k] (3).
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uxy ½k ¼ rxy ½k
ðEx Ey Þ1 ð3Þ
P
with Ex ¼ Nn¼0 x2 ½n
The detection of known signals in noisy environments with a correlation-based
approach is also introduced as a matched filter approach, for example in [11].

3 Test Setup and Procedure

The partial discharge measurements were performed by using a partial discharge test-
circuit referring to IEC 60270 within a shielded high voltage test-chamber. The power
supply for the test object was realized as a half-wave rectifier consisting of a trans-
former (Fig. 2, T) with an output voltage Û2,max = 140 kV, high voltage diodes
(D) and a capacitor CS = 10 nF for voltage smoothing. The resistor RP was used as a
 Digital Filtering Methods for Interferences on Partial Discharges 563

blocking impedance and reduced the influence on the measurements by PD originating

of the voltage source. The voltage measurement at the test samples (TS) was done with
a resistive voltage divider (VD) in parallel to the coupling capacitor CK. Discharge
signals of the test samples were directly recorded with a transient recorder (TR) by
using a 1 MX input impedance with a 50 X terminal resistance.

Fig. 2. Test circuit for measurements

The partial discharges were generated by test samples for two typical insulation
defects. A needle-plate arrangement for the generation of corona impulses represents
one signal source (test sample C). The distance between both electrodes is about
15 mm. A second test sample consisting of glass sheets between a flat and a cylindrical
electrode was used to generate surface discharges in order to extent the number of
signal classes for a test of the effectiveness (test sample S).
The measurements were divided into a pre-investigation for the determination of
the partial discharge inception voltage (PDIV) of each test sample and afterwards the
recording of the test signals. All measurements were performed with a negative polarity
of the test voltage due to the polarity effect on the test sample C. First step was a test on
PD activity of the environment within the test chamber. Therefore, the test voltage was
increased up to a level of 40 kV at which no PD events were observed.
Since an electrical breakdown of the test sample may damage the measuring
devices like the applied transient recorder, the PDIV for each test sample is determined.
For this pre-investigation, a commercial PD measuring system consisting of a mea-
suring impedance, a sensor and an evaluation software was used to determine the PDIV
of the test samples. First PD impulses from the test sample S were detected at a voltage
level of −5 kV. Since the occurrence of the PD events were not stable, the test voltage
was increased to a final level of −7.5 kV. This also led to a higher repetition rate of the
PD pulses. The test sample C showed an inception of PD signals at a voltage level of
−4 kV.
Due to the successful resistance of the test samples against the applied test voltage
level, the coupling device was removed and the transient recorder was connected
directly to the coupling capacitor with a coaxial cable. For each test sample, 200 events
564 B. Hochbrückner et al.

were recorded with a sampling frequency of fa = 250 MHz and a length of 2500
samples per trace which results a duration of 10 µs. The direct measurement of the PD
discharges avoided any suppression of relevant signal information by the band-pass
characteristic of the coupling device before an analog-digital conversion.

4 Test Results and Discussion

Analyses in time domain of the acquired data show significant differences of the single
events. Here, a first distinction into four signal classes is possible due to the waveform
of the single impulses:
• surface discharges (group S)
• corona discharges (group C)
• ground noise
• unknown disturbance (group D)
Each test sample shows specific discharge events with a significant pulse shape of
these discharge impulses. PD of the test sample C occur with a smaller time distance
between two consecutive events than the surface discharges. Furthermore, group C
signals do not show any oscillations in their waveform in contrast to group S signals.
Their traces have higher amplitudes for some events and show clear oscillations. It is
assumed that these oscillations were caused by reflections through a mismatch of wave
impedances of the signal transmission path. Both discharge signals show a high
reproducibility in pulse shape and a scatter of the signal amplitude.
Besides these two signal classes as types of different partial discharge sources, the
remaining signals are classified as ground noise and an unknown signal class, which
also occurs when no test sample is connected to the test circuit. This leads to a
classification as a disturbance that seems to originate in the applied voltage source.
Figure 3 shows one trace per signal class except ground noise, which is superimposed
to almost all traces.
An additional analysis of the spectra for the obtained signal templates show a band-
pass behavior up to 20 MHz for group S and C signals (3 dB-criteria) and up to
200 kHz for signals of group D (Fig. 4). Furthermore, events of group D reveal specific
peaks in their amplitude response within a frequency range from 10 MHz up to
70 MHz. These spectra show relevant information within a frequency range that also
refers to significant contents of the group S and C signals. Like mentioned before, an
application of common filter methods like high-pass or low-pass filtering for de-noising
would also affect the spectra of the measured PD signals which results in a distortion of
the discharge waveform. This would also influence further analyses of the waveform,
like the classification of PD sources by a comparison of impulse shape parameters,
since relevant spectral information is suppressed respectively lost [12].
 Digital Filtering Methods for Interferences on Partial Discharges 565

Fig. 3. Acquired signal templates with a length of 151 samples (604 ns).

Fig. 4. Spectra of signal templates.

566 B. Hochbrückner et al.

Due to the difficulties in de-noising with common linear filter methods, a new
approach based on the histogram analysis of cross-correlation coefficients is introduced.
This method works on cross-correlation within a signal class as a so-called intraclass
correlation and evaluates the similarity of events obtained from the same source with a
normalized correlation coefficient uxy[k]. A correlation of events between different
signal classes as interclass correlation is expected to generate lower values for
uxy[k] than a correlation within one signal group. This approach can be used to indicate
a similarity between signals and further the affinity to a signal class. To distinguish
signals into correlated and uncorrelated ones, a threshold has to be determined.
Therefore, a histogram analysis of the intraclass correlation for group S signals (uSiSj),
group C signals (uCiCj) and an interclass correlation between group S and group C (uSC)
was done. Figure 5 shows the distribution of the correlation coefficients of this analysis.

Fig. 5. Histogram of intraclass (uSiSj, uCiCj) and interclass (uSC) correlation of group S and D
signals.

The coefficients for intraclass correlated events show a Gaussian distribution above
u = 0.8, interclass correlated coefficients below. There, a clear distinction into corre-
lated and uncorrelated signals can be estimated empirically at this threshold. This
means a belonging to a signal class for u > 0.8 and no relation for values below
u = 0.8. For the correlation, the relevant events of all traces were manually rectangular
windowed over a length of 151 samples (see Fig. 3)
With the results of the interclass correlation between group S and group C dis-
charges, an intraclass correlation of the signals within group D shall prove the con-
sistence of the events from this signal class. This was done in the same manner with a
manual windowing on the disturbance signals. The results are visualized in Fig. 6.
The distribution for intraclass correlated group D events of the test sample S
(uDsiDsj), sample C (uDciDcj) shows an accumulation above 0.7 with some outliers
below. Moreover, the interclass correlation between group D signals of both test
 Digital Filtering Methods for Interferences on Partial Discharges 567

samples show a significant concentration above 0.8 with two peaks at 0.8 and 0.9
(Fig. 6, bottom). In this case, the disturbances are divided into two groups with a high
similarity to each other. For further evaluations, the disturbances are assumed, to be
well correlated and considered as one signal class.

Fig. 6. Histogram of intraclass correlation of group D signals.

Fig. 7. Interclass correlation between dicharge signals (S, C) and disturbance signals (D).
568 B. Hochbrückner et al.

A last investigation in the distribution uDC for an interclass correlation between

group D and C as well as uDC (group D and S) proves the non-similarity of the groups
S and C to group D (Fig. 7). All plots show a concentration for u below 0.6 with some
outliers above 0.65 (uSDc). The obtained traces show a clear non-similarity between the
discharge classes and the disturbance signals. By this, a second threshold value at 0.6 is
determined. After determining the threshold values for the distinction of different
signals into classes, these were applied to the classification. In this case, the full traces
with a length of 2500 samples were used as input signals for the matched filter to detect
the single events and group them through an evaluation of their correlation coefficients.
The signal flow chart of the overall classification algorithm is presented in Fig. 8.

Recording of templates for the signal clas-

ses S,C and D

Recording of the signal x[k]

Correlation Yes
Signal S detected
φxS> ν1

No

Correlation Yes
φxC> ν1 Signal C detected

No

Yes
Correlation Signal D detected
φxD> ν2

No

Unknown signal

Fig. 8. Flow chart of classification

Digital Filtering Methods for Interferences on Partial Discharges 569

Fig. 9. Classification of group S signals.

Fig. 10. Classification of group C signals.

570 B. Hochbrückner et al.

Figure 9 shows the results for a detection of surface discharges (group S) by the
correlation of the trace uS[k] with the templates of Fig. 3. The cross-correlation
function uUS for the group S template shows a peak with a value >0.8, so at this
position the similarity to a group S trace is quite high. Obviously, a group S discharge
is detected there. The trace of uUC shows the correlation to a group C template with
values <0.8 at the position of the surface discharge (no similarity to group C) as well as
no similarity to group D uUD < 0.6. Besides the surface discharge, a disturbance can be
seen on the top trace in time domain. The correlation analysis uUD classifies this as a
group D event with no similarity for both discharge templates and a high similarity for
a group D template. The proposed classifier applied to events from group C (Fig. 10)
results in a positive detection of group C discharges for uUC > 0.8. In this case, three
clear discharge impulses are detected. The cross-correlation with other templates shows
values below 0.8 (uUS) and 0.6 (uUD)

5 Conclusions and Outlook

A classification of PD impulses based on a correlation analysis is introduced. The

thresholds are determined by a histogram analysis of intra- and interclass correlations.
A distinction between disturbances and discharge signals as well as a distinction
between two discharge signals was successfully applied.
In further investigations, this approach will be tested on the measurement of two
defects at once. Also a limitation of the spectral content according to [1] shall deter-
mine the applicability of this method. Furthermore, a real-time mode will be established
to use this method in an automated measuring set up. At least, the application on one
test sample with several defects and disturbances is planned.

Acknowledgement. The presented investigations were performed within the research project
“AnalyTEG” (sign nr. 13FH200PX6) related to the funding program “Forschung an Fachhoch-
schulen” of the German Ministry of Education and Research.

References
1. DIN EN 60270:2000+A1:2016. Hochspannungs-Prüftechnik – Teilentladungsmessugen
(2016)
2. Küchler, A.: High Voltage Engineering: Fundamentals, Technology, Applications. Springer,
Heidelberg (2018)
3. Rethmeier, K., Obralic, A., Kraetge, A., Krüger, M., Kalkner, W., Plath, R.: Improved noise
suppression by real-time pulse waveform analysis of PD pulses and pulse-shaped disturbances.
In: 16th International Symposium on High Voltage Engineering, Johannesburg, South Africa,
August 2009
4. Sriram, S., Nitin, S., Prabhu, K.M.M., Bastiaans, M.J.: Signal denoising techniques for
partial discharge measurements. IEEE Trans. Dielectr. Electr. Insul. 12(6), 1182–1191
(2005)
 Digital Filtering Methods for Interferences on Partial Discharges 571

5. Zhang, H., Blackburn, T.R., Phung, B.T., Sen, D.: A novel wavelet transform technique for
on-line partial discharge measurements part 1: WT de-noising algorithm. IEEE Trans.
Dielectr. Insul. 14(1), 3–14 (2007)
6. Zhang, H., Blackburn, T.R., Phung, B.T., Sen, D.: A novel wavelet transform technique for
on-line partial discharge measurements part 2: on-site noise rejection application. IEEE
Trans. Dielectr. Electr. Insul. 14(1), 15–22 (2007)
7. Rethmeier, K., Kruger, M., Kraetge, A., Plath, R., Koltunowicz, W., Obralic, A., Kalkner,
W.: Experiences in on-site partial discharge measurements and prospects for PD monitoring.
In: 2008 International Conference on Condition Monitoring and Diagnosis, Beijing, China,
April 2008
8. Wenzel, D., Borsi, H., Gockenbach, E.: Unterdrückung impulsförmiger Störsignale bei der
Vor-Ort-Teilentladungsmessung mit Hilfe der Naheste-Nachbar-Klassifikation. Teilent-
ladungserfassung an elektrischen Isoliersystemen 56, 171–176 (1995)
9. Shim, I., Soraghan, J.J., Siew, W.H.: Digital signal processing applied to the detection of
partial discharge: an overview. IEEE Electr. Insul. Mag. 16(1), 6–12 (2000)
10. Deergha Rao, K., Swamy, M.N.S.: Digital Signal Processing: Theory and Practice. Springer,
Singapore (2018)
11. Oppenheim, A.V., Verghese, G.C.: Signals, Systems and Interference: Class Notes for
6.011: Introduction to Communication, Control and Signal Processing. Spring 2010.
Massachusetts Institute of Technology: MIT OpenCourseWare. License: Creative Commons
BY-NC-SA. https://ocw.mit.edu/
12. Vogt, T.: Teilentladungen bei Gleichspannung. Ph.D. thesis, TU Dortmund (2015)
 Fiber-Optic Acoustic Sensors for Partial
Discharge Detection in Power Cables

Chen Hao1 , Xu Yang1, Qian Sen2(&) , and Su Lei3

1
State Key Laboratory of Power Equipment and Electrical Insulation,
Xi’an Jiaotong University, Xi’an, China
2
Guangzhou Institute of Advanced Technology, Chinese Academy of Sciences,
Guangzhou, China
sen.qian@giat.ac.cn
3
School of Engineering and Materials Science,
Queen Mary University of London, London, UK

Abstract. Partial discharge acoustic detection is a powerful technique for

assessment of the insulation integrity of power cables. In contrast to conven-
tional method in which piezoelectric transducers are commonly used to detect
the partial discharge acoustic emission, fiber-optic sensors recently become
excellent candidates for partial discharge acoustic detection because of their
several benefits, such as immunity to electromagnetic interference and long
range remote sensing capabilities. In this paper, we demonstrate a fiber-optic
sensor system that can successfully detect the partial discharge acoustic emis-
sion from power cable joint defect. We first designed a fiber sensor coil that
worked as the acoustic sensing element, and built an interrogation unit based on
Sagnac interferometry to extract the detected acoustic signal from the fiber
sensor coil. Then, a power cable joint defected by stress cone dislocation was
used as the model to test the efficiency of our designed fiber-optic sensor system.
An electric detection unit that conformed to the IEC 60270 was also used as a
reference to monitor the partial discharge electric signals. The statistical analysis
on the phase resolved pulses detected by the fiber-optic sensors and the ones
detected by the electric sensors verify that the fiber-optic sensors can success-
fully detect the partial discharges induced within the stress cone dislocation of
the power cable joint. We expect that fiber-optic sensors to become an excellent
candidate for on-site partial discharge acoustic detection in the future.

Keywords: Fiber-optic sensors
Partial discharge
Acoustical detection

Power cables

1 Introduction

Optical fibers, after the demonstration that they are ideal media for light transmission
[1], are also expected to be excellent sensors. The benefits of optical fiber sensors
(OFS) include immunity to electromagnetic interference and galvanic isolation which
make OFS very attractive in high voltage applications where electromagnetic inter-
ference is common and the safety of human operators are under major concern.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 572–581, 2020.
https://doi.org/10.1007/978-3-030-31676-1_54
 Fiber-Optic Acoustic Sensors for Partial Discharge Detection in Power Cables 573

Partial discharge (PD) detection, as a commonly used tool for insulation evaluation,
is one of the scenarios where the applications of OFS is promising. OFS in different
forms have been put into practice for the PD measurement. Svoma et al. [2] proposed a
solution for integrated condition monitoring assets, which consisted of distributed
temperature strain sensing and partial discharge on-line monitoring along the subsea
power cable systems. PD detection was achieved by fitting HFCTs to the earth return
connections on the cable terminations, and high frequency signals were transmitted via
optical fiber cables back to the PD monitor located on the off-shore substation. In this
circumstance, optical fiber can only play a role in the data transmission with ultra-long
range. Tian et al. [3, 4] proposed an optical PD detection technique based on electro-
optic modulator and demonstrated its application in a 400 kV cable joint. This
approach provides a feasible PD monitoring technique for remote high voltage cable
systems. Zagari et al. [5] designed a fiber-optic ultrasonic pressure sensor for acoustic
detection of PD in transformers, in which the optical fiber was tightly wrapped around
the porcelain stem of a 66 kV current transformer, and an interferometry setup was
used to extract the ultrasonic signal which was related to PD level in the equipment.
Zhang et al. [6] presented another interferometer-based optical fiber sensing system
which designed a fiber sensor head with an elastomer cylinder to detect acoustic
emissions generated from PD in GIS. Habel et al. [7] embedded two types of OFS in a
HV silicone rubber insulation material: a fiber-optic acoustic sensor with Fabry-Perot
interferometer to measure acoustic waves in polymeric insulation generated by PD, and
a PMMA-based fluorescent optical fiber to detect the optical emission during ionization
process in the insulation material. Rohwetter et al. [8] developed an all-silicone elas-
tomer fluorescent optical fiber integrated into the stress cone elements of HV/EHV
cable accessories to enhance the optical PD detection sensitivity. They also introduced
a coherent Rayleigh time-domain reflectometry (C-OTDR) for PD distributed acoustic
sensing. The results proved that C-OTDR could be a promising solution for long range
sensing applications such as the submarine cable PD monitoring.
In this paper, a fiber-optic acoustic sensor was designed and fabricated, and it was
interrogated by a Sagnac interferometer to detect acoustic wave induced by partial
discharges. A 8.7/15 kV power cable joint was fabricated with stress cone dislocation
defect, which is commonly seen on-site. The partial discharges were induced within the
defect of the cable joint, and the fiber-optic sensor was deployed on the cable joint for
PD acoustic detection. An electric instrument, which followed the instruction specified
in IEC 60270, was also used as a reference.

2 Fiber-Optic Acoustic Sensor Design and Detection

The fiber interferometer that exhibit extreme sensitivity to acoustic wave was con-
sidered to detect the acoustic emission from partial discharges. It relies on demodu-
lating the phase change of the light wave that propagating through an optical fiber
under acoustic wave perturbation. According to [9], the phase / of the light wave after
going through a section of fiber with length l is / ¼ bl, where b is the optical wave
574 C. Hao et al.

propagation constant. Then perturbing the fiber under acoustic pressure wave caused
the phase at the output to shift by

db db
D/ ¼ bDl þ l Dn þ l DD ð1Þ
dn dD
in which, the first term represents the effect of fiber length change Dl due to strain,
while the second and third term respectively come about from the strain-optic effect
where by the strain changes the refractive index of the fiber Dn and the fiber mode
dispersion effect due to a change in fiber diameter DD produced by the strain.
The above mentioned effect can be utilized to develop a variety of fiber-optic sen-
sors. We first designed and fabricated a fiber coil sensor. The fiber coil sensor was a
compact circular donut, which was made by wrapping the single-mode fiber around a
mandrel. Figure 1 illustrates the fiber coil sensor geometry. A bend-insensitive single-
mode fiber with coating diameter of 250 lm was winded onto a cylindrical mandrel
backbone, using adhesive to glue the fibers into a donut shape. Once the fiber coil was
fabricated, the backbone mould was removed, and the inner hole of the coil was filled
with adhesive. It was then placed under dry air condition until the coil became solidified.

Fig. 1. The fiber coil sensor geometry modelling

The overall optical phase shift induced within the fiber coil with the geometric
parameters shown in Fig. 1 can be expressed as:

Zh=2 Zr2
1 jKz cos h
D/coil ¼ 2 pc ejXt e dz D/unit
2prJ0 ðKr sin hÞdr ð2Þ
d
h=2 r1

where D/coil is the overall optical phase shift, D/unit is the normalized optical phase
change induced in fiber of unit length by unit pressure [10, 11], pc is the averaged
 Fiber-Optic Acoustic Sensors for Partial Discharge Detection in Power Cables 575

incident acoustic pressure over the surface of fiber coil cross-sectional area, X is the
acoustical frequency of incident wave, K is the acoustical wave-vector of incident
wave, h is the incident angle of acoustic wave relative to the normal of fiber coil cross-
section, d is the diameter of single-mode optical fiber, and J0(Krsinh) is the zero-order
Bessel function.
Figure 2 shows the calculated acoustic frequency response of the fiber coil sensor
with optimized geometry designed according to Eq. (2) in the normal incident con-
dition: inner radius r1 = 10 mm, outer radius r2 = 15 mm, and thickness h = 10 mm.
The frequency response in Fig. 2 shows a monotonic decreasing response with the
acoustic frequency until it reaches the resonance frequency of 92 kHz, which relates to
geometric dimensions of the fiber coil.

Fig. 2. Calculated acoustic frequency response of an optical fiber coil sensor for the normal
incident acoustic wave (h = 0). The fiber coil sensor geometric parameters: inner radius
r1 = 10 mm, outer radius r2 = 15 mm, thickness h = 10 mm.

Sagnac interferometry can be used to retrieve the small optical phase shift induced
within the fiber coil sensor. The scheme of an all-fiber Sagnac interferometer with a
fiber coil sensor is shown in Fig. 3. Laser light emitted from a wideband laser diode is
injected into one of the input port of a 3  3 single-mode fiber coupler. On the other
side of the fiber coupler, the two output ports are connected to the both ends of a long
length single-mode fiber, which incorporates with the above designed fiber coil sensor
embedded at an arbitrary position (apart from the midpoint of the fiber length).
Through the 3  3 fiber coupler, the light power is split equally at each of the output
port and the light will transmit to a clockwise (CW) direction and a counter-clockwise
(CCW) direction simultaneously in the common path fiber loop. With the fiber sensor
placed asymmetry in the fiber, the light will experience a differential optical phase shift
576 C. Hao et al.

along the CW and CCW path when the fiber is under some locally perturbations, i.e.
acoustic wave perturbation, as expressed in Eq. (3),
  
z Lz
D/ ¼ /CW t   /CCW t  ð3Þ
v v

where /CW and /CCW are the optical phase shift due to the acoustic emission wave
perturbation along the CW and CCW path respectively, and D/ is the differential
phase. z is the fiber sensor position, L is the fiber length, and v is the velocity of light in
fiber core. After the light travelling in the two opposite direction along the fiber, they
will meet at the 3  3 fiber coupler again, then the interference effect happens. The
other two input port of the fiber coupler could direct the interference light outward and
feed into a balanced photodetector, in which a 180 degree differential phase is intro-
duced for the two fed light signals, thus SNR could be improved by a factor of 2 in the
balanced photodetector. The final output signal has an expression in Eq. (4):
pffiffiffi
2 3
Iout ¼ <E 2 sinð2D/Þ
9 pffiffi 0 ð4Þ
’ 4 9 3 <E02
D/ for D/  1

where < is the responsivity of the balanced photodetector in unit of A/W, E20 is
corresponding to the input light power, and Iout is the output current. A data acquisition
device was used to collect the electric signal transformed by the balanced photodetector
(BPD). The sinusoidal function relationship figures the characteristics of 3  3 fiber
coupler and balanced photodetector. In the scenario of PD induced acoustic pressure
wave that is weakly perturbing the fiber coil, the sinusoidal function approximately
degenerates to a linear relationship with a minute phase shift, as expressed in Eq. (4).

Fig. 3. The configuration of an all-fiber Sagnac interferometer. Acronyms: LD – laser diode, FC –

fiber coupler, FUT – fiber under test, FS – fiber sensor, AE – acoustic emissions, BPD – balanced
photodetector. DAQ – data acquisition, CW – clockwise wave, CCW – counter-clockwise wave.
 Fiber-Optic Acoustic Sensors for Partial Discharge Detection in Power Cables 577

3 Experiment and Results

As a proof-of-concept demonstration, the proposed fiber-optic sensor system was

deployed on a power able joint to evaluate its performance for partial discharge
acoustic detection, as shown in Fig. 4. We made a stress cone dislocation defect within
the 8.7/15 kV medium voltage heat shrink joint. The overall length of the cable joint
was 170 cm and the stress cone dislocation defect was made by separating the bottom
of stress cone component in the cable joint from the cable’s outer semiconductor with
approx. 20 mm spacing. When this part of cable-to-joint bonding area is under elec-
trical stress, it is much likely to initiate partial discharge. A portion of this discharge
energy is released in form of pressure waves as a primary source of acoustic emission.
Acoustic measurements of PD activity rely on the detection of the pressure wave
formed within insulation materials.
The sample cable was stressed with a high voltage source according to the basic test
circuit specified in IEC 60270, as shown in Fig. 4. A digital PD measuring instrument
(Doble Lemke GmbH) was used to collect electrical PD pulses through a measuring
impedance Z and to synchronize with the AC voltage phase. The optical fiber coil
sensor was attached to the outer surface of the cable joint, leaving about 12.5 cm away
from the stress cone dislocation defect. In addition, A set of armoured optical com-
munication fiber cable with 1000 m length was used to form a fiber loop and was
launched into the Sagnac interferometer.

Fig. 4. The experimental test circuit for PD measurement of cable joint by comparison of IEC
standard electrical method and optical fiber sensor based acoustic method. Acronyms: HV – high
voltage source, F – high voltage filter, Ck – coupling capacitor, Z – measuring impedance, PD –
partial discharge measuring system, GND – electric ground, FS – fiber sensor, SI – Sagnac
interferometer.

The digital PD system can record the electrical signal continuously, and the testing
time in the experiment was set to be 60 s. Meanwhile, the acoustical signal was
interrogated with Sagnac interferometer periodically in comparison. The inception
578 C. Hao et al.

voltage (PDIV) was 5 kV in our experiment. The apparent discharge level detected by
the electric sensor was 30–342 pC. However, no obvious acoustic signal can be
observed under this applied voltage for the Sagnac interferometer. When the applied
voltage was raised to 10 kV, both the electric sensor and the OFS can respectively
detect the electric signal and acoustic signal from the partial discharges. Figure 5(a)
shows the phase-resolved PD diagram reimaged from the electrical PD data. The
maximum apparent discharge increased to the level of 1800 pC. The PD repetition rate
was 972 counts/s in average. Meanwhile, acoustic PD pulses from optical fiber sensor
were observed within every AC cycling period of the 10 kV voltage level. Similarly as
in the electrical method, the acoustic PD pulses were also synchronized to the applied
AC voltage phase, therefore, an analogous phase-resolved acoustic PD diagram was
drawn by picking every acoustic burst peak value versus corresponding AC phase
angle, as shown in Fig. 5(b).The phase-resolved acoustic PD diagram detected by OFS
agreed well with that of the electric sensor, where PD primarily occurred during the
first quadrant and third quadrant. However, the repetition rate of acoustic PD detected
by OFS was only 59 counts/s in average that was much lower than that of the electric
sensor.

Fig. 5. The phase-resolved PD measurement result according to IEC 60270 electric method
(a) and the fiber-optic sensor based acoustic method (b), for the cable joint with stress cone
dislocation defect under a 10 kV applied voltage.
 Fiber-Optic Acoustic Sensors for Partial Discharge Detection in Power Cables 579

An instant slide of a single PD pulse waveform detected by both of the electric

sensor and the OFS is shown in Fig. 6. The electric single pulse only last for
nanoseconds, whereas the acoustic single pulse last for milliseconds. Nevertheless, it is
not always the case that an electric PD pulse must follow a corresponding acoustic PD
pulse. This may be the reason why the repetition rate of acoustic PD pulse is much
lower than that of electrical PD pulse.

Fig. 6. A simultaneously detected PD signal with electric method (upper) and fiber-optic sensor
based acoustic method (lower).

Figure 7 shows the FFT spectrum of the specific acoustic PD pulse shown in
Fig. 6. The main frequency band was in the range of 10 kHz to 100 kHz, with several
unique resonant peaks in 9 kHz, 24 kHz, 41 kHz and 91 kHz.
580 C. Hao et al.

Fig. 7. FFT spectrum of an instant acoustic PD pulse corresponding to that in the Fig. 6.

4 Conclusions

This research presented a fiber-optic acoustic sensor to detect the partial discharge
induced acoustic emission wave within the power cable joint. The acoustic frequency
response of the fiber coil sensor was firstly analyzed and then a geometrically opti-
mized fiber coil was fabricated based on the theoretical analysis, which showed a
considerable sensitivity in the range between 10 kHz and 100 kHz. The experiment
results showed that fiber-optic acoustic sensor could successfully detect the partial
discharges within the stress cone dislocation defect in power cable joint. The acoustic
signals detected by the fiber-optic sensor agreed well with the electric signals detected
by the standardized electric detection unit, although the sensitivity of the fiber-optic
acoustic detection was not as high as that of electric detection unit. We expect that
fiber-optic sensors to become an excellent candidate for partial discharge acoustic
detection in field applications.

Acknowledgements. This work was funded by the National Science Foundation of China
(Grant No. 51577149), the State Key Laboratory of Electrical Insulation and Power Equipment of
China (Grant No. EIPE14117), and the Guangdong Power Grid Co. Ltd. (Grant No.
GDKJXM20172510). This work was supported by the Engineering and Physical Sciences
Research Council [Grant Number EP/L022559/2].

References
1. Kao, K.C., Hockham, G.A.: Dielectric-fibre surface waveguides for optical frequencies.
Proc. Inst. Electr. Eng. 113, 1151–1158 (1966). https://doi.org/10.1049/piee.1966.0189
2. Svoma, R., Smith, C., Conway, C.: Integrated condition monitoring for subsea power cable
systems. In: 20th International Conference on Electricity Distribution. IET, Stevenage, UK,
Paper 1000 (2009)
 Fiber-Optic Acoustic Sensors for Partial Discharge Detection in Power Cables 581

3. Tian, Y., Lewin, P.L., Pommerenke, D., Wilkinson, J.S., Sutton, S.J.: Partial discharge on-
line monitoring for HV cable systems using electro-optic modulators. IEEE Trans. Dielectr.
Electr. Insul. 11, 861–869 (2004). https://doi.org/10.1109/TDEI.2004.1349792
4. Tian, Y., Lewin, P.L., Wilkinson, J.S., Schroeder, G., Sutton, S.J., Swingler, S.G.: An
improved optically based PD detection system for continuous on-line monitoring of HV
cables. IEEE Trans. Dielectr. Electr. Insul. 12, 1222–1234 (2005). https://doi.org/10.1109/
TDEI.2005.1561802
5. Zagari, A., Blackburn, T.: Acoustic detection of partial discharges using non-intrusive
optical fibre sensors. In: 6th International Conference on Conduction and Breakdown in
Solid Dielectrics, New York, USA. IEEE, pp. 573–576 (1998). https://doi.org/10.1109/icsd.
1998.709350
6. Zhang, T.Z., Pang, F.F., Liu, H.H., Cheng, J.J., Lv, L.B., Zhang, X.B., Chen, N., Wang, T.
Y.: A fiber-optic sensor for acoustic emission detection in high voltage cable system. Sensors
16, 2026 (2016). https://doi.org/10.3390/s16122026
7. Habel, W.R., Buchholz, U., Heidmann, G., Hoehse, M., Lothongkam, C.: Fibre-optic
sensors for early damage detection in plastic insulations of high-voltage facilities. In: XVII
International Symposium on High Voltage Engineering. VDE-Verlag, Berlin, pp. 1–6 (2011)
8. Rohwetter, P., Siebler, D., Eisermann, R., Lothonkam, C., Habel, W.: Progress in optical PD
detection for translucent and transparent HV cable accessories with improved fluorescent
optical fibres. In: 9th International Conference on Insulated Power Cables, Jicable,
Versailles, France, D5.4 (2015)
9. Bucaro, J.A., Lagakos, N., Cole, J.H., Giallorenzi, T.G.: Fiber optic acoustic transduction.
Phys. Acoust. XVI, 385–457 (1982). ISBN 0-12-477916-6
10. Hocker, G.B.: Fiber-optic sensing of pressure and temperature. Appl. Opt. 18, 1445–1448
(1979). https://doi.org/10.1364/AO.18.001445
11. Jarzynski, J., Hughes, R., Hickman, T.R., Bucaro, J.A.: Frequency response of interfero-
metric fiber-optic coil hydrophones. J. Acoust. Soc. Am. 69, 1799–1808 (1981). https://doi.
org/10.1121/1.385917
 Development of PRPS Conversion Algorithm
and Generator Using PRPD Pattern
of Underground Power Transmission Cable

Jae-Seop Lim1(&), Gye-Hyun Joh1, Jeon-Seon Lee2, and Won Choi2

1
Korea Electric Power Corporation, Daejeon-si, Korea
limjs@kepco.co.kr
2
SMND Co., Ltd., Anyang-si, Korea

Abstract. Widely used method for PD (partial discharge) measurement and

analysis of existing underground power transmission cables is to store the
detected signal pulses as PRPD (Phase Resolved Partial Discharge) data and to
analyze the PD defect factors by the pattern type. Although the PRPD pattern
data of various facility defects are acquired through the underground power
transmission cable diagnosis, due to the PRPD data storage structure, the pulse
generation information is destroyed over time, and reconversion into 3D PD
pattern (PRPS: Phase Resolved Pulse Sequence) is impossible. Also recently, as
a method for verifying the reliability of the power utility prevention diagnosis
system, a simulated partial discharge is generated for each facility defect, and
the diagnostic system verification that is capable of simultaneously measuring
pulses and PD patterns using various sensors is performed, and it is necessary to
develop a PD pulse generator based on simulated PRPS data in order to generate
the pulses based on PRPS pattern using the existing PRPD pattern, and input the
pulses into the system to perform the defect judgment performance test. To solve
this problem, this study developed and verified a PD pattern generator that can
generate pulses based on PRPS data, and an algorithm that converts existing
PRPD pattern data currently owned by Korea Electric Power Corporation
(KEPCO) to PRPS pattern data.

Keywords: PRPD
PRPS
3D PD pattern
PD measurement
Defect

Underground power transmission cables
Diagnosis system

1 Introduction

Failure of the extra-high voltage underground power transmission cable often gets
exacerbated into a large-scale problems, and as it incurs significant losses in terms of
time and finance for restoration, the prevention diagnosis in advance is needed. For this
reason, there have been various researches on the development of extra-high voltage
XLPE (Cross-Linked Polyethylene) power cable insulation diagnosis method world-
wide, and for XLPE power cable, insulation diagnosis based on the partial discharge
measurement is recognized as the most effective method. [1–3] For partial discharges

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 582–593, 2020.
https://doi.org/10.1007/978-3-030-31676-1_55
 Development of PRPS Conversion Algorithm and Generator 583

occurring at the on-site power cables, HFPD (High Frequency Partial Discharge
Measurement) has been proposed since the mid-1980 s, and it has been widely applied
for the completion test and diagnosis for deterioration of extra-high voltage XLPE
power cables mainly in the developed countries since the early 2000’s. [4–9] In Korea,
widely used method for PD (partial discharge) measurement and analysis of existing
underground power transmission cables is to store the detected signal pulses as PRPD
(Phase Resolved Partial Discharge) data and to analyze the PD defect factors by the
pattern type. Although the PRPD pattern data of various facility defects are acquired
through the underground power transmission cable diagnosis, due to the PRPD data
storage structure, the pulse generation information is destroyed over time, and recon-
version into 3D PD pattern (PRPS: Phase Resolved Pulse Sequence) is impossible.
Also recently, as a method for verifying the reliability of the power utility prevention
diagnosis system, a simulated partial discharge is generated for each facility defect, and
the diagnostic system verification that is capable of simultaneously measuring pulses
and PD patterns using various sensors is performed, and it is necessary to develop a PD
pulse generator based on simulated PRPS data in order to generate the pulses based on
PRPS pattern using the existing PRPD pattern, and input the pulses into the system to
perform the defect judgment performance test. To solve this problem, this study
developed and verified a PD pattern generator that can generate pulses based on PRPS
data, and an algorithm that converts existing PRPD pattern data to PRPS pattern data.

2 Analysis Algorithm

This study explains the PRPD and PRPS data structures, which are primarily used as
PD pattern analysis method, and describes the algorithm that was devised in the study
which converts PRPD Rawdata into PRPS Rawdata.

2.1 Data Structures of PRPS and PRPD

PRPS is the PRPD data plotted on the time axis. By using the PRPS method, it is
possible to determine whether the partial discharge signal is intermittent or persistent,
and therefore, it is possible to continuously analyze data information for a certain
period of time. While PRPD is valid for observing the U − q − n pattern, PRPS has the
advantage that the pattern can be observed with time information. PRPS data can be
expressed as cumulative data of a one-dimensional array of amplitude values with a
resolution of 256 sizes based on 1 cycle of 60 Hz, as shown in Fig. 1.
584 J.-S. Lim et al.

Fig. 1. PRPS data pattern

The PRPD data is the PD pattern in which the pulse frequency is accumulated in a
two-dimensional array composed of phase and amplitude coordinates for a one-
dimensional array of PRPS data for each cycle. As shown in Fig. 2, it is the amplitude
values for each cycle of PRPS data shown in PRPD format. Since the PRPD data is a
two-dimensional array data in which the frequency of occurrence is accumulated, there
is a disadvantage in that pulse generation information for each cycle is lost, but on the
other hand, the there is an advantage in that the storage capacity of Rawdata can be
smaller than that of PRPS data.
 Development of PRPS Conversion Algorithm and Generator 585

Fig. 2. PRPD data conversion and pattern

2.2 PRPS Conversion Algorithm

In order to calculate the valid number of pulses of the PRPD data, it is necessary to
calculate the base noise level and to replace the data value with ‘0’. The procedure to
remove unnecessary pulse data of the base noise is shown in Fig. 3.
586 J.-S. Lim et al.

Fig. 3. Process of PRPS conversion algorithm

In order to generate the PRPD pattern data in accordance with the output cycle
desired by an analyst, calculating the 1 cycle average pulses generation count is
required, and due to the asymmetric characteristics of the PD pattern, it is necessary to
calculate the number of the average pulse generations per cycle for each positive and
negative polarities as shown in Eqs. (1) and (2), respectively. Then the random phase
and pulse size values are extracted to compare whether the data has the identical values
with the PRPD data, and if the data has a valid value, the PRPS pattern is formed by
creating and accumulating to the PRPS 1D (1 cycle) data.
P127 P255
 j¼0 i¼0 PRPD½i½j
0  180 Phase average pulse generation count ¼ ð1Þ
0  180 Phase
 Development of PRPS Conversion Algorithm and Generator 587

P255 P255
 j¼128 i¼0 PRPD½i½j
181  359 Phase average pulse generation count ¼ ð2Þ
0  180 Phase
The PRPS pattern data generated by using the existing accumulated PRPD pattern
data can be applied to the developed PD pattern simulator, and generates the output
signal in real-time in the same manner as in the actual on-site, and the PRPS and the
PRPD patterns can be obtained through the measurement equipment as shown in Fig. 4.

Fig. 4. An example of PRPS conversion data utilization

3 Development of PD Simulator
3.1 PD Simulator H/W Configuration
Figure 5 shows the configuration of the PD Simulator H/W. There is a touch-type
TFT LCD screen where the user can check the PD pattern and configure the various
settings, and BNC input terminal which can be used for external sync and signal input.
The internal circuit configuration consists of a signal processing module that is capable
of storing and generating the PRPS Rawdata pattern, and a signal generating module
that is capable of adjusting the generated pulse information for the size gain, and also
generating the square wave pulses. In addition, there is an analog filter and a capacitor
588 J.-S. Lim et al.

module which can convert square wave pulse generated into partial discharge pulse
pattern as shown in Fig. 6. The Amplitude of the pulse can be calculated by the
amplitude of the generated pulse and the size of the capacitor. For example, a pulse
with an amplitude of 500 pC can be generated when 100 pF is applied to a 5 V square
wave pulse. The corresponding equation is Q = CV.

Fig. 5. PD simulator internal configuration

Fig. 6. Convertor square wave to PD pulse

Figure 7 illustrates the principle of pulse generation. Based on the internal power
phase, external sync, or phase information, the phase information of the pulse is
generated at the corresponding time interval from the previously stored PRPS data.
When the generated pulse is measured and analyzed by the partial discharge diagnostic
device, the PRPS and the PRPD patterns can be obtained as shown in Fig. 6.
 Development of PRPS Conversion Algorithm and Generator 589

Fig. 7. PD simulator pulse generator principle

The exterior design and explanation to the parts of the developed product as shown
in Fig. 8.

Fig. 8. PD simulator outline and parts explanation

590 J.-S. Lim et al.

Table 1 shows the specifications of the developed products. There can be up to 3

pulse generation channels, and the minimum rising time of the pulse can be up to 2 ns.

Table 1. PD simulator specifications

Item Contents
Output pC or mV
Unipolar output
PD waveform output
Output signal length 256 sample  Max 3600 cycle
Frequency band 0–50 MHz
Pulse Rising time: 2 ns
Rising/falling time Falling time: proportional to size
Channel 3 Channel BNC Type
LCD 7” TFT LCD
External sync. Power 220 V sync
Load current synchronization
User interface Compatible with PC
Ability to upload Excel CSV files

3.2 PD Simulator H/W Configuration

Figure 9 shows the S/W configuration. When the power is initially applied, the boot
screen is displayed, and after booting is completed, user can see total of 4 screen
modes. In the Master View screen, user can adjust the channel size and select the
source of the signal sync. In the Channel View screen, user can select the simulation
pattern to create, and also can check and print the PRPD and the PRPS of the existing
patterns. In the File View screen, user can select the pattern file to upload from the
external memory. In the Setting View screen, screen brightness, sound, and IP settings
are available.

3.3 PD Simulator Output Verification

Figure 10 shows the PD Simulator output verification configuration. The input data
were the PRPS data acquired with the setting of 3600 cycle for each corresponding
defect, which were stored and verification was conducted thereon. As the measurement
verification equipment, the previously developed Compact T7 equipment was utilized.
 Development of PRPS Conversion Algorithm and Generator 591

Fig. 9. S/W configuration

Fig. 10. PD simulator output verification configuration

592 J.-S. Lim et al.

Table 2 summarizes the verification results.

Table 2. PD simulator output verification result

PD simulator Measurement Defect type and measurement set-

Channel View result tings

Void Pattern
frequency: 20MHz
Cycle: 3600Cyc.

Void Pattern
Measuring frequency: 20MHz
Cycle: 3600Cyc.

Surface Pattern
frequency: 20MHz
Cycle: 3600Cyc.

Corona pattern
frequency: 20MHz
Cycle: 3600Cyc.

Corona pattern
frequency: 20MHz
Cycle: 3600Cyc.

4 Conclusions

As a result of development of the PD Simulator for underground transmission cable,

storage and implementation were carried out by utilizing the Rawdata for each defect,
and the actual discharge pulse, PRPS and PRPD could be obtained. In addition, the data
that is difficult to utilize due to the characteristics of PRPD data having no pulse
information of time was converted to PRPS data by applying the algorithm, then by
applying to the PD pattern generator which can generate the real-time signals, the
utility of the existing PRPD pattern data stored has increased, which made it possible to
verify the partial discharge pattern analysis function of the underground cable partial
discharge diagnostic system. Going forward, it is considered to be possible to utilize it
to verify the performance of the underground monitoring system and the portable
diagnostic equipment.
 Development of PRPS Conversion Algorithm and Generator 593

References
1. Research Report, Korea Electric Power Research Institute: Establishment of diagnosis method
of insulation deterioration of power transmission cable and prediction of life span
99Electricity-762 (2016)
2. Tanaka, T., Greenwood, A.: Advanced power cable technology. In: 9th International
Proceedings on Proceedings, pp. 1–2. CRC Press, Inc. Boca Raton (2010)
3. WG 21.05 on behalf of study committee 21: Diagnostic methods for high voltage cable
systems. CIGRE session 15/21/33-05 (1996)
4. Lee, S.K., Lee, C.Y., Baek, J.H., Kim, D.W., Kim, C.S.: Characteristics of high frequency
partial discharge for artificially defected extra high voltage accessories. In: IEEE Conference
on Electrical Insulation and Dielectric Phenomena, pp. 682–685 (2000)
5. Boggs, S., et al.: High frequency attenuation in shielded solid dielectric power cable and
implications thereof for PD location. IEEE Trans. Electr. Insul. 12 (1996)
6. Ahmed, N.H., Srinivas, N.N.: On-line partial discharge detection in cables. IEEE Trans.
Dielectr. Electr. Insul. 5(2), 181–188 (1998)
7. Boggs, S.A., et al.: Attenuating voltage surges in power cable by modifying the semi
conductive shields. In: IEEE International Symposium on Electrical Insulation, June 1992
8. Pultrum, E.: On-site testing of cable systems after laying, monitoring with HF partial
discharge detection. The Institution of Electrical Engineers. IEEE, Savoy Place, London
(1995)
9. Pommerenke, D.: Discrimination between internal PD and other pulses using directional
coupling sensors on HV cable systems. IEEE Trans. Dielectr. Electr. Insul. 6(6), 814–824
(1999)
 Double-End Excitation of Transformer
Winding Model for Improved Frequency
Response Analysis

Saurav Pramanik(&) and Aravind Ganesh

Indian Institute of Technology, Kharagpur 721302, India

saurav.pramanik@gmail.com,
aravindganesh.1205@gmail.com

Abstract. FRA is essentially a comparison-based measurement technique

which compares the response of a newly manufactured transformer winding
with the responses of the same transformer measured subsequently during
operation. Any change or deviation in subsequently measured frequency
responses compared to the healthy response of the winding determines the fault
in winding viz. mechanical damage or deformation, internal short-circuit,
insulation damage etc. If the healthy response is not available (which may occur
in practice for several reasons), the applicability of FRA method will be severely
affected. In absence of healthy response, FRA measurement alone can’t detect
the fault in transformer winding. This is how a limitation does exist for several
years in conventional practice of FRA method and no solution is proposed so far
in the literature to overcome this problem. To this effect, this paper proposes an
advanced measurement technique with double-end excitation of the winding to
identify the fault even in absence of healthy response. For feasibility studies of
the proposed technique, initially the simulation was performed for a four-section
ladder-network model followed by actual measurements on a practical winding
model. Results demonstrate that the proposed method is indeed very promising
to explore the possibility of identifying the fault in absence of healthy responses.

Keywords: Frequency response
Driving point admittance
Double-end

excitation
Ladder network model
Fault diagnostics

1 Introduction

It is needless to emphasize the importance of monitoring the health of a transformer

winding to prevent undesirable failures of transformer during operation. If such failures
occur due to winding damage or deformation, insulation damage and/or internal short-
circuits in the winding; it actually describes the breakdown of mechanical health of the
transformer. In a transformer, mechanical damage or deformation generally occurs as
consequences of several short-circuit forces appear across the winding [1, 2]. Further
the incidence of overvoltage and overcurrent transients also causes the insulation
damage in transformer winding. Such failures in transformer are often dealt with huge
economic-impact for utilities as well as for industries; it is not only about the cost of the
transformer, but also a huge penalty for unplanned outages caused by the undesirable

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 594–605, 2020.
https://doi.org/10.1007/978-3-030-31676-1_56
 Double-End Excitation of Transformer Winding Model 595

failure of transformers. Therefore, monitoring the health and necessary maintenances

are paramount even for a minor amount of fault in transformer winding [3]. Methods
developed so far for fault detection in transformer winding are the measurements of
short-circuit impedance, shunt-reactance, FRA/SFRA measurement technique etc. FRA
technique is by far the most sensitive method amongst all of them [2, 3]. FRA response
is measured by supplying a sinusoidal voltage of low magnitude (in range of 10–15 V)
across the transformer winding and plotting one of the system responses (i.e. spectrum
of output response divided by spectrum of input) as the frequency of applied voltage is
varied usually from few kHz to 1 MHz [1]. When a new transformer is manufactured,
FRA measurement is performed on the winding and is considered as the ‘reference’ or
‘fingerprint’ of the transformer. In conventional practice, this measurement is repeated
on the same winding at regular interval in future and is compared with the ‘reference’.
If deviations are observed, it interprets the fault in transformer winding [4–6].
Therefore, essentially having the ‘reference’ or ‘healthy’ magnitude response is
MUST for successful implementation of FRA method absence of which FRA method
is of no use. But, the healthy magnitude response of a transformer may not be always
available for several reasons in practice. Therefore, the question now arises how the
FRA method could be utilized for such a scenario. Literature suggests to have ‘finger-
print’ of a sister unit but this may not be always available in practice [7]. Furthermore,
this requires the dependency of other transformer as well as inherent difference between
two sister-units may also cause a problem. To this point, a new philosophy is described
in this paper for FRA measurement of transformer winding to identify the fault even in
absence of ‘healthy’ response.

2 Proposed Methodology

A new methodology based on measuring and comparing the responses of upper and
lower half of a single isolated transformer winding is proposed in this paper to identify
the fault in winding. In present work, winding is assumed to be uniformly wound for its
entire length and hence, it is to be considered that the upper and lower halves of the
winding are physically identical in terms of their length and parameter distribution.
Therefore, their individual magnitude responses (impedance or admittance) are also
likely to be identical, if the responses are separately possible to measure. If the fault
occurs in any half of the winding, two halves will remain no more identical and hence,
their magnitude responses would also undergo a change or shift from each other to
indicate the fault. This is the philosophy of the proposed method. For feasibility test as
well as for the sake of clarity in explaining the proposed method, only an isolated
winding model is considered for the present scope of work.
However, the challenge for implementing such an idea in practice is about how to
measure the responses separately for upper and lower halves of a given winding. One
possibility could be grounding the center point of the winding (i.e. the mid-point of the
winding length) and conduct the measurement of impedance or admittance from their
respective end-terminals by exciting one end at a time. Figure 1 schematically shows
such a possible connection diagram for an isolated winding.
596 S. Pramanik and A. Ganesh

Fig. 1. A schematic diagram for connection setup to measure the impedance/admittance

magnitude responses for each half of the coil.

In above Figure, grounding the middle point enables the individual excitation of
upper and lower halves of the winding by exciting the end terminals separately one
after the other. Since the middle point is grounded, the source connected to one end of
the winding is not likely to excite the other half of the winding since the current driven
by the source from one end continues to flow till the middle point (since it is grounded)
and ceases to flow in other half. This implies that the impedance seen by the source,
connected to one end of the winding, is actually half the impedance of the winding (i.e.
impedance of one half). Both the halves being physically identical; the responses
measured from either ends of the winding are supposed to be identical. But, the
practical issue for conducting such measurements in reality is that the middle point is
not obtainable outside for its grounding (only terminals are accessible). Therefore, an
alternative must be explored to impose a virtual ground at the centre (instead of
physically grounding it) using the terminals.
One feasible idea is about simultaneously exciting the end terminals by two voltage
sources. Sources must produce sinusoidal voltages of equal magnitude but opposite in
phase (i.e. opposite polarity) for all frequencies. This would enable the middle point of
a uniformly would transformer winding virtually grounded. Connection for such a
measurement is shown in Fig. 2.
For excitations in Fig. 2, impedance or admittance magnitude responses measured
from either ends of the winding should appear as identical as long as both the halves of
the winding remain identical. If fault occurs in any half of the winding, two halves are
to be seen no more identical (since the distributions of winding parameters are dif-
ferent) and hence, the middle point of the winding would remain no more a virtual
ground point. This will introduce a mismatch for both the responses and thus will also
 Double-End Excitation of Transformer Winding Model 597

Fig. 2. Connection diagram for double-end excitation of a coil/winding.

interpret the fault. This is the philosophy of the proposed method. A flowchart below
demonstrates the steps of implementation for the proposed method. Voltage and current
signals used in demonstration of the flowchart correspond to the voltage and currents
for the model in Fig. 2.
Flowchart: Steps for FRA measurement with double-end excitation
598 S. Pramanik and A. Ganesh

3 Simulation Results

To demonstrate the feasibility of the proposed method initially an arbitrary four-section

ladder-network model was considered for simulation (shown in Fig. 3). Such a coupled
ladder-network model (consisting of series and shunt capacitances, series inductances
with couplings) inherently captures the intricacies of winding behavior under excitation
for FRA measurement. For healthy condition, all the sections of the model are con-
sidered identical. Parameter values assigned for the model are obtained from literature
and are also listed in Table 1.

Fig. 3. Four-section ladder-network model used for simulation

Table 1. Model parameters

Parameter Value
R 2X
Ls 0.5 mH
Cs 0.5 nF
Cg 0.3 nF
M12 = M23 = M34 0.25 mH
M13 = M24 0.167 mH
M14 0.125 mH

Four-section model in Fig. 3 is only an example. Users may also perform the
exercise for any number of sections to test the feasibility of the proposed method.
A program was coded in MATLAB for simulation to perform the double-end excitation
of the model as well as to compute the admittance magnitude responses. For simula-
tion, following steps were considered:
1. Two source voltages of equal magnitude but of opposite polarity (i.e. out of phase)
were considered for simultaneous excitation of end terminals of the model (i.e.
double-end excitation).
2. State-space analysis was used to compute the admittance magnitude responses from
both the ends for both excitations (i.e. I1/E1 and I2/E2).
3. Node between second and third sections of the model appears as the virtual ground
and this was verified by computing the voltage at this node.
 Double-End Excitation of Transformer Winding Model 599

Computed responses are plotted in Fig. 4.

Fig. 4. Admittance magnitude responses computed for each half of the model for double-end
excitation (healthy condition)

In Fig. 4, it is clearly visible that the admittances from either ends of the model are
truly identical in terms of the following:
• Admittance magnitude for the entire range of frequency
• number of resonance peaks and
• the resonance frequencies
Input current expressions were also computed for lossless condition (R = 0, for
ease of computation) for unit step voltages and are included below. Both the current
expressions are found identical. This ensures that the impedances seen by the sources
from either ends of the model are identical and hence, also interprets the healthy state of
the model.

119 24 117612
I1 ¼ I2 ¼ þ 532 þ 3445
ð1Þ
260 s 221s2 þ 1272

Where: I1 = Input current for source-1.

I2 = Input current for source-2.
Next, a fault was introduced in the model by changing self-inductance and series
capacitance of the second section. Changed values are 0.47 mH, and 0.37 nF respec-
tively. Rest of the arrangements for the model was kept unchanged and the admittance
magnitude responses were computed again. New responses are plotted in Fig. 5.
600 S. Pramanik and A. Ganesh

Fig. 5. Admittance magnitude responses computed for a fault condition in the model with
double-end excitation

Following observations are enlisted below from Fig. 5:

1. Responses don’t match with each other and hence, they are no more identical
2. New resonance peaks also appeared in both the magnitude responses
Mismatch in two responses is exactly what is expected as per the demonstration for
proposed method in Sect. 2. Employing the philosophy of the method, mismatch
between the above two responses directly reveals the presence of fault in model. Fault
interpretation in this way does not require any comparison with the ‘healthy’ or ‘ref-
erence’ magnitude response and this is the significance of the proposed method.
Authors would also like to mention that simulations were performed for numerous
number of models with different set of model parameters (considering symmetry in
model parameters for healthy case) and always similar results were obtained. For
brevity, only one set of result is included in this paper.

4 Experimental Study

A homogeneous winding model represented by a lumped parameter ladder-network

(identical parameters for all sections) was considered for experimental validation of the
proposed method. Model is shown in Fig. 6. This is an eight-section model for which
series and ground capacitances are connected externally. Inductances are provided by
copper-wound coils on a cylindrical insulating former. Outside diameter of the former
is 200 mm and for self-inductance of each section, eighteen turns were used. Cs and Cg
values for each section are 0.68 nF and 0.24 nF respectively.
 Double-End Excitation of Transformer Winding Model 601

Fig. 6. Eight-section ladder network model for experimental measurement (side and front view)

For double-end excitation of the model, it was ensured that the sinusoidal excita-
tions from both ends of the model are of ‘equal magnitude’ but of ‘opposite polarity’
(i.e. 180° phase shift) for all excitation frequencies. This was ensured by the following:
1. Two signal generators capable of producing maximum 20 Vp-p sinusoids with a
frequency range of 0–80 MHz are connected in Master-Slave mode using coaxial-
cables and are synchronized by a common clock signal to trigger at the same instant
(using in-build features of signal generators).
2. Input voltage waveforms were acquired in an 8-bit digital oscilloscope and phase of
one source voltage was adjusted by 180° difference keeping the other source voltage
as the reference. 180° phase difference was maintained for all excitation frequen-
cies. This was also implemented using the features of signal generators.
3. Source currents were also measured in oscilloscope using two clamp-on current
probes with a sensitivity of 1 mA/mV, and bandwidth 450 Hz–60 MHz.
Measurement setup is shown in Fig. 7.
602 S. Pramanik and A. Ganesh

Fig. 7. Measurement setup using model coil for double-end excitation

Employing the sweep frequency method, peaks of all the voltage and current
waveforms were measured from the oscilloscope and thereafter were used to plot the
magnitude responses for I1/E1 and I2/E2 in Fig. 8.

Fig. 8. Measured admittance magnitude responses for upper and lower halves of the model for
double-end excitation (healthy condition).

It is seen that the admittance magnitude responses measured from either ends of the
model are closely matching with each other. Peak and trough frequencies and their
corresponding magnitudes are also found the same for both the responses. If the
responses are zoomed, a minor mismatch may be visible in magnitudes but could be
reliably ignored. This mismatch is actually due to inherent deviation in dimensions for
 Double-End Excitation of Transformer Winding Model 603

each section of the model since the coil-sections were manually built in work-
shop. However, the results found are very consistent with the simulation results and
thus also interpret the symmetry between upper and lower halves of the winding-
model. Hence, this may be considered as the healthy condition of the model.
For experimental validation of fault diagnostic feature, next a fault was introduced
in the model by changing Ls and Cs in third section and Cg at the node between second
and third sections. This simulates the mechanical fault in practice for actual winding
damage or deformation. Therefore, the two halves of the model are no more identical in
terms of parameter distribution. Self-inductance was also varied by short-circuiting two
adjacent turns in third section. Cs and Cg values were changed by simply changing the
corresponding external capacitors with new values. New values are 0.63 nF and 0.3 nF
for Cs and Cg respectively. With these changes in model, measurement was repeated
following the similar procedure described for healthy case. New responses are plotted
in Fig. 9.

Fig. 9. Measured admittance magnitude responses of the model for double-end excitation
(faulty condition).

Responses measured by source-1 and source-2 are also marked in the plot. Dif-
ference between two responses is clearly visible in the plot. In addition, new peaks are
also visible compared to the peaks in heathy response. These are the similar features
which were also observed for simulation results in Sect. 3. The reason why two
admittances seen by source-1 and source-2 are significantly different from each other
could be explained by physical asymmetry between two halves of the model. A simple
comparison between two simultaneous measurements directly reveals the presence of
fault in the model and thus validates the feasibility of the proposed method.
604 S. Pramanik and A. Ganesh

5 Remarks and Future Scope

• For an odd number of sections in the model, method is also verified by simulations.
In this case, the entire section in the middle is an equivalent to the virtual ground
point of the model.
• Two identical faults symmetrically located in either half of the model may produce
identical measurements for the model (or winding) but such fault scenario is very
unlikely to occur in practice for any winding/model.
• Validation of the proposed method for winding model in this paper may explore the
possibility of extending the method in future for its application to actual transformer
winding as well as for three-phase transformer.

6 Conclusion

Absence of healthy magnitude response reduces the scope of applicability for con-
ventional FRA method to detect the damage/deformations in transformer winding. To
identify the fault in absence of healthy magnitude response, this paper presented a
novel measurement technique by measuring (and comparing) the magnitude responses
of upper and lower halves of an isolated winding model. Magnitude responses for both
the halves were measured by simultaneously exciting the model from both ends by
equal and opposite sinusoids for all frequencies. Any change/deviation in two
responses, if observed, determines the fault in the model. Or else it would be recog-
nised as healthy condition. Before the proposed technique was validated for a practical
winding model, it was also verified for simulations on a four-section ladder network
model. For all simulations and practical measurements, results are always found very
consistent and thus it recognizes the feasibility and the potential of the proposed
method. This is the contribution and significance of the present work. In future, it
would be really interesting to extend the method for its application to actual trans-
former winding as well as for three-phase transformer.

Acknowledgments. Authors thank SRIC, IIT Kharagpur for granting “Institute Scheme of
Innovative research and Development” fund (ISIRD) to build the experimental setup in laboratory.

References
1. Bagheri, M., Naderi, M.S., Blackburn, T.: Frequency response analysis and short-circuit
impedance measurement in detection of winding deformation within power transformers.
IEEE Electr. Insul. Mag. 29(3), 33–40 (2013)
2. Abu-Siada, A., Hashemnia, N.: Understanding power transformer frequency response
analysis signatures. IEEE Electr. Insul. Mag. 29(3), 48–56 (2013)
3. Small, B.J., Abu-Siada, A.: A new method for analysing transformer condition using
frequency response analysis. In: 2011 IEEE Power and Energy Society General Meeting,
pp. 24–29 (July 2011)
 Double-End Excitation of Transformer Winding Model 605

4. Reykherdt, A.A., Davydov, V.: Case studies of factors influencing frequency response
analysis measurements and power transformer diagnostics. IEEE Electr. Insul. Mag. 27(1),
22–30 (2011)
5. Ludwikowski, K., Siodla, K.: Investigation of transformer model winding deformation using
sweep frequency response analysis. IEEE Trans. Dielectr. Electr. Insul. 19(6), 1957–1961
(2012)
6. Gomez-Luna, E., Mayor, G.A., Gonzalez-Garcia, C., Guerra, J.P.: Current status and future
trends in frequency-response analysis with a transformer in service. IEEE Trans. Power
Delivery 32(2), 587–598 (2014)
7. Al Murawwi, E., Mardiana, R.: Effects of terminal connections on sweep frequency response
analysis of transformers. IEEE Electr. Insul. Mag. 28(3), 8–13 (2012)
 Electrical Discharge Localization
for Gas Insulated Line Based
on Distributed Acoustic Sensing

G. Ma(&) , W. Qin , C. Shi , H. Zhou , Y. Li , and C. Li

State Key Laboratory of Alternate Electrical Power System with Renewable

Energy Sources, North China Electric Power University, Beijing, China
ncepumgm@gmail.com

Abstract. To overcome the high cost, low spatial resolution of the electrical
discharge localization systems for the Gas Insulated Line (GIL), a non-intrusive
distributed optical fiber sensing system based on Distributed Acoustic Sensing
(DAS) is proposed. Firstly, the principle of the proposed discharge localization
system is introduced. Then, efficiency of the proposed sensing system was
demonstrated based on pencil-lead break experiments. Furthermore, the local-
ization results of the experiments using the GIS spacer flashover discharge
indicate that the system meets the requirement of GIL discharge localization.

Keywords: Gas Insulated Line
Distributed Acoustic Sensing
Discharge

localization

1 Introduction

Gas insulated line (GIL) has the advantages of less harm to environment, smaller land
occupation and higher running stability [1, 2]. An ultra-high voltage GIL is under
construction in the Sutong tunnel which is 75 m under the Yangtze River. If there is a
discharge on an insulator or in the SF6 in the GIL, the insulator or the gases should be
changed because of the insulation reduction. Because the GIL is about 1 km long, a
monitoring system is needed to locate the discharge.
Currently, there are mainly two kinds of discharge localization system. (1) An
electrical measurement system. The sensor used in the system is similar as the transient
sensor used for very fast transient overvoltage (VFTO) detection [3]. The discharge
location is determined based on the time difference of the surge travel wave from the
location of the discharge to the two sensors located at different end of the GIL.
However, the transient voltage sensors which need to be installed inside the GIL tank
are not welcomed by the utilities. In addition, the rise time attenuations as the transient
travels along the GIL may also reduce the localization accuracy. (2) Electric acoustic
emission (AE) sensors. The acoustic emission generated by a discharge is monitored.
As the measurement range of the AE sensor is about ten meters. One hundred sensors
need to be mounted to the outside surface of the GIL tank to achieve discharge
localization for one kilometers length GIL.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 606–614, 2020.
https://doi.org/10.1007/978-3-030-31676-1_57
 Electrical Discharge Localization for GIL Based on DAS 607

In this paper, an optical distributed sensing method is proposed to provide a

potential method to solve the problem.

2 Principle of Phi-OTDR

In fiber optic transmissions, scattering is the loss of signal caused by the diffusion of a
light beam. But the back scattering can be used in optical sensing. Scattering in glass
arise from microscopic variations in the material density, from compositional fluctu-
ations and from structural inhomogeneties or defects occurring during fiber manufac-
ture. The back scattering consists of Rayleigh, Brillouin and Raman scattering.
Moreover, different distributed optical fiber sensors (DOFS) were developed based the
different scattering mechanisms [4–6].

Fig. 1. The scattering in the optical fiber

Rayleigh scattering was used in the paper. As for Rayleigh-based sensing, it typ-
ically requires lower averaging than Raman or Brillouin [7]. Therefore, it is suitable for
the discharge localization in GIL. Because DOFS allows for the continuous mea-
surement over long fiber distances, it provide a cost-effective solution for the GIL
discharge localization (Fig. 1).
A distributed optical fiber sensing system based on the phi-OTDR is proposed to
achieve the discharge localization, as shown in Fig. 3.
An ultra-narrow linewidth laser was used as the optical source. The linewidth of the
laser is 3 kHz which provide a better signal noise ratio for interferometer sensing. The
608 G. Ma et al.

wavelength of the laser is 1550.12 nm that can be transmitted in the single mode fiber.
An Acousto-optic modulator (AOM) was used to modulate the laser generated by the
ultra-narrow linewidth laser. The input of the AOM was an electric pulse which width
was 20 ns. Then the light in the fiber was amplified by an erbium-doped fiber amplifier
(EDFA). After that, the optical pulses are coupled into a sensing fiber.
As the pulse of light travels down the optical fiber, interactions within the fibre,
which result in light reflections known as the backscatter. When the pulse transmitted
forward in the sensing fiber, the reflected signals of different parts in of the pulse
overlaped and transmitted backward together. Because the coherence length of the laser
is much longer than the spatial interval of the pulse, the reflected optical signals will
interfere with each other (Fig. 2).

Fig. 2. Diagram of the discrete model of Rayleigh backscattering in optical fiber

The scattered lights reflected back into the optical circulator and are detected by a
photo detector.
Acoustic emission generated during a discharge in the GIL. As the optical fiber was
mounted on the outside surface of the GIL. There was an acoustic wave pressure forced
on the optical fiber. As we known, the phase of the light can be easily affected by the
stain of the optical fiber. a phase shift can be induced by the pressure. Then, the
interference light backed to the photo detector changed with the acoustic pressure
generated by the discharge in the GIL.

3 Methodology of Phi-OTDR

In our system, a Rayleigh scattering trace is a voltage curve converted by the photo
detector. It is generated with one pulse. The x-axis time point in the trace represent the
distance (Dist) z. The discharge generates ultrasonic vibration pressed on the optical
 Electrical Discharge Localization for GIL Based on DAS 609

fiber. Then, the voltage at the distance where the ultrasonic vibration pressed on the
optical fiber was changed, which is indicated in the Rayleigh scattering trace.
The pulse frequency was set to 5 kHz. The interval of the measured Rayleigh
Scattering Traces was 0.2 ms. The ultrasonic vibrations induced by a discharge can be
sensed by subtraction of a phi-OTDR trace from an earlier stored trace (Fig. 4).

Fig. 3. Sensing system of merged phi-OTDR and Michelson interference

Fig. 4. Measurement of ultrasonic vibrations induced by a discharge

610 G. Ma et al.

The key parameter in the phi-OTDR is the spatial resolution of the sensing system.
As mentioned before, the reflected optical signals interfered with each other in a pulse.
Thus, the theoretical spatial resolution Dz of the phi-OTDR depends on the width of the
light pulse [8]:

cTp
Dz ¼ ð1Þ
2gg

Where: c is the propagation speed of light in vacuum, Tp is the width of the light
pulse, and gg is the group refractive index.
Besides, the sampling rate of the DAQ used in the sensing system and the detector
bandwidth of the photodetector also adversely affects on the spatial resolution. The
theoretical spatial resolution of the proposed system is 2 m.

4 Performance Tests

Pencil-lead breaks are widely used as a reproducible source for test signals in acoustic
emission (AE) applications. Thus, a pencil-lead break experiment was carried out to
investigate the efficiency of the proposed sensing system [9].
The acoustic signal generated by the pencil-lead break lasted 20 ms to 200 ms.
A PicoScope 6000 was used to record the signals. The data were transmitted to a
computer and analyzed by a program to calculate the discharge position. The pencil-
lead break was first set at 115 m of the 1.1 km long sensing fiber. The recorded
Rayleigh scattering traces are shown in Fig. 5.
The position of the acoustic source was calculated by subtracting the Rayleigh
scattering traces. There is a pulse at 115 m, and the amplitude of the pulse is 2.1 mV
(shown in Fig. 6). It is the same located where the pencil-lead break occurred. The
pencil-lead break was then set at 1000 m of the 1.1 km long sensing fiber. The
recorded Rayleigh scattering traces are shown in Fig. 7.

Fig. 5. Rayleigh Scattering Traces measured when the pencil-lead break at 115 m
 Electrical Discharge Localization for GIL Based on DAS 611

Fig. 6. Measured pencil-lead break position by subtracting the detected Rayleigh scattering
traces (at 115 m)

Fig. 7. Rayleigh Scattering Traces measured when the pencil-lead break at 1000 m

The position of the acoustic source were calculated by subtracting the Rayleigh
scattering traces. There is a pulse at 1000 m, and the amplitude of the pulse is 1.8 mV
(shown in Fig. 8). It is the same location where the pencil-lead break occurred.

Fig. 8. Measured pencil-lead break position by subtracting the detected Rayleigh scattering
traces (at 1000 m)
612 G. Ma et al.

The experiments indicated that the proposed system can obtain the position of the
vibration source. The measurement range is over one kilometres.

5 Localization Experiment of Discharge

Localization experiments based on GIS spacer flashover discharges were carried out to
obtain the accuracy and range of the proposed system. A metal particle was arranged on
a spacer of a 110 kV GIS to simulate the defeat. The applied voltage of the experiment
was a standard 250/2500 ls switching impulse induced by an impulse generator. Ten
meters fibers in 1100 m sensing fiber were wrapped on the outer surface of the GIS
tank to measure the acoustic emission induced by the flashover discharges. Besides, the
other parts of the sensing fiber were used to simulate the fibers mounted to the outside
surface of a long-distance GIL tank. The setup of the experiment platform is shown in
Fig. 9.

Fig. 9. GIS discharge localization test platform

To verify the detection range of the distributed sensing system, the discharge
location was set approximately 975 to 990 m in the 1100 m sensing fiber. At first, the
impulse voltage generator was set to 10% below the estimated discharge voltage of the
defeat. Then, the amplitude of the impulse voltage was gradually increased until the
flashover discharges occurred on the space. The proposed phi-OTDR system was used
to measure the acoustic emission induced by the flashover discharges. Then, the cor-
responding position of the discharge can be sensed by subtraction of a phi-OTDR trace
from an earlier stored trace.
The measured Rayleigh scattering traces and the subtractions of the Rayleigh
scattering traces are shown in Fig. 10. The located discharge was 980 m to 987 m. The
difference between the set location and the measured location is only 5 m. The results of
the experiment demonstrated that the system can locate the discharge even the discharge
is nearly 1000 m away, and the location accuracy of the fault is smaller than 15 m.
 Electrical Discharge Localization for GIL Based on DAS 613

Usually, the length of a GIL section is 15 m. The requirement of the monitoring

system is to locate the GIL section a discharge happened. Thus, localization accuracy
of the proposed system meets the requirement. As the discharge current in the operation
GIL is much larger than that in the laboratory, the amplitude of the ultrasound induced
by the discharge is significantly higher than that in our experiment. As we successfully
located the small discharge in the laboratory, the sensitivity is very good for discharge
location in field.

Fig. 10. Localization result of discharge occurred at approximately 985 m in the 1100 m
sensing fiber

6 Conclusions

A distributed optical fiber sensing system based on the phi-OTDR was proposed for
GIL discharge localization. Efficiency of the proposed sensing system was demon-
strated based on pencil-lead break experiments. Furthermore, discharge localization
experiment results indicate that the sensing range of the system is over one kilometre
and the location accuracy is smaller than 15 m. Take the advantages of non-intrusive,
better spatial resolution, and lower cost, the proposed method provides a potential
method to locate the discharge in GIL.

Acknowledgments. This work was supported by National Key R&D Program of China (Grant
No. 2017YFB0903800).

References
1. Koch, H.: Gas Insulated Transmission Lines (GIL). Wiley, Hoboken (2012)
2. Lin, C., Li, Q., Li, C., Zhang, B., Liu, W., Yang, Y., Liu, F., Liu, X., Hu, J., He, J.:
Novel HVDC spacers by adaptively controlling surface charges – part iii: industrialization
prospects. IEEE Trans. Dielectr. Electr. Insul. 25(4), 1259–1266 (2018)
3. Ma, G.-M., Li, C.-R., Chen, W.-J., Sun, Z.-L., Guo, P.-H., Chen, M., Ding, W.-D., Li, Z.-B.:
A reliable wide-bandwidth VFTO sensor based on surface-mounted devices. IEEE Trans.
Power Delivery 28(3), 1839–1846 (2013)
614 G. Ma et al.

4. Wang, Z., Li, J., Fan, M., Zhang, L., Peng, F., Wu, H., et al.: Phase-sensitive optical time-
domain reflectometry with Brillouin amplification. Opt. Lett. 39(15), 4313–4316 (2014)
5. Bao, X., Zhou, D.-P., Baker, C., Chen, L.: Recent development in the distributed fiber optic
acoustic and ultrasonic detection. J. Lightwave Technol. 35(16), 3256–3267 (2017)
6. Ma, G., Qin, W., Du, Y., Zhou, H., Li, Y., Yan, C., Li, C.: A centimeter resolution GIS
insulator strain distribution measurement method based on OFDR. IEEE Sens. J. 19(8), 2962–
2969 (2018)
7. Pastor-Graells, J., Martins, H.F., Garcia-Ruiz, A., Martin-Lopez, S., Gonzalez-Herraez, M.:
Single-shot distributed temperature and strain tracking using direct detection phase-sensitive
OTDR with chirped pulses. Opt. Express 24(12), 13121–13133 (2016)
8. Juarez, J.C., Taylor, H.F.: Polarization discrimination in a phase-sensitive optical time-
domain reflectometer intrusion-sensor system. Opt. Lett. 30(24), 3284–3286 (2005)
9. Sause, M.G.R.: Investigation of pencil-lead breaks as acoustic emission sources. J. Acoust.
Emiss. 29, 184–196 (2011)
 Mineral Oil and Ester Based Oil/Paper
Insulation Decaying Assessment
by FTIR Measurements

U. Mohan Rao1, I. Fofana1(&), R. Kartheek2, K. M. L. Yapi1,

and T. Jaya1
1
Research Chair on the Aging of Power Network Infrastructure (ViAHT),
University of Quebec at Chicoutimi, Chicoutimi, QC G7H 2B1, Canada
ifofana@uqac.ca
2
University Center for Aluminum Research (CURAL),
University of Quebec at Chicoutimi, Chicoutimi, QC G7H 2B1, Canada

Abstract. Esters based dielectric fluids have been widely researched since
1990’s for application in high voltage insulation applications. Since then,
researchers are affirmative towards usage of ester based insulating fluids as a
replicated to mineral insulation oils. The operating properties and aging per-
formance of ester oils proved to be potential candidates for high voltage
applications. In view of high temperatures and longevity of insulation systems,
there also is a need to understand the chemical perspectives along with aging
behaviour of ester oils. Accelerated aging of oil/paper insulation associated with
mineral oil and synthetic ester with cellulose insulant has been experimentally
simulated as per ASTM D 1934 at 115 °C. Fourier Transform Infrared spec-
troscopy analysis of oils and cellulose papers is carried out at different aging
factors. The compositional changes in oils and cellulose kraft paper with aging
have been enumerated. The changes in the absorbance area for appropriate
functional groups have been also reported. It is found that, the chemical stability
of synthetic esters is superior to that of the mineral oil.

Keywords: Transformers
Insulation
Ester oils
FTIR spectroscopy

1 Introduction

Since years, mineral insulating oils are been successfully used in transformer tech-
nology for insulation and cooling purposes. However, with experience, engineers
realized the requirements of high dielectric and thermal performance of the insulating
fluids. Consequently, mineral oil (MO) is facing some serious critiques in terms of fire
point resources and biodegradability. Global research on alternative insulating fluids is
affirmative towards usage of ester based dielectric fluids as a replicate to mineral ones
[1]. Studies revealed that, synthetic ester (SE) is a suitable candidate for use in breather
transformers and natural ester (NE) performs well in sealed transformers [2–4].
Performance of insulating oil degrade with operating times and is required to
monitor periodically to ensure oil pristine conditions. Hence, there are several
parameters associated with oil that are to be maintained in proper limits [5]. Dielectric

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 615–624, 2020.
https://doi.org/10.1007/978-3-030-31676-1_58
616 U. Mohan Rao et al.

and physiochemical parameters of synthetic esters and natural esters have been widely
investigated and is established that, ester based fluids are superior to mineral oils [ref].
Insulating oil is used in conjunction with insulation paper and hence the compatibility
of new insulating oils with paper is to be ensured. Researchers have investigated the
performance of insulation paper in various insulating oils and reported the reduced
degradation of cellulose papers in new oils [6, 7].
The change in the quality of oil in transformer may be attributable to several
reasons including electrical stress, thermal stress, and oil/paper interface properties. It is
to be noticed that, these attributes involves in changing the chemical state of the
oil/paper insulation. This is because, aging of oil involves in formation of aging
products like acids, dissolved decay contents, dissolved gases, polar solvents and
sludge which hinders the performance of insulation system. Aging of paper is also
evident with production of some chemicals like furfurals, methanol, and acids. The
degradation aspects of oil/paper insulation is involved with several chemical changes.
Thus, there is a need to study the changes in chemical compositions that occur with
aging of oil/paper insulation. Fourier Transform Infrared spectroscopy (FTIR) analysis
of oil/paper insulation based on mineral oil and cellulose insulant have been reported in
[8, 9]. Also, chemical compositional changes for natural ester and cellulose based
insulation have been reported by researchers [10]. Comparative FTIR analysis of
mineral oil, natural ester, and synthetic ester is also reported by researchers [11].
In this paper, FTIR analysis of oil/paper insulation associated with mineral oil and
synthetic ester have been reported for different aging factors. FTIR analysis of mineral
oil, synthetic ester, and kraft paper (aged in MO and SE) have been reported. It is
noticed that, surface interaction of synthetic ester with cellulose kraft showed a sig-
nificant stability in its chemical structure as compared to that of the mineral oil.

2 Experimental

Initially, mineral oil, synthetic ester, and cellulose kraft paper were subjected to drying
in order to remove moisture. Dry oils and papers are transferred to aging cells and are
later subjected to thermal aging as per ASTM D 1934. Oil/paper ratio adopted is 10:1,
which is universally followed for oil/papers insulation studies for transformers. Ther-
mal aging followed 500, 1000, 1500, and 2000 h of test durations at 115 °C. After
every test duration, FTIR characterization is carried out for insulation oils. Simulta-
neously, cellulose kraft papers aged in mineral oil and synthetic ester are subjected to
degreasing in hexane fumes for one hour to remove the traces of oil absorbed by
cellulose fibres. Later, aged and degreased cellulose kraft papers are also subjected to
FTIR characterization. In order to establish the base line prior to aging, fresh mineral
oil, fresh synthetic ester and unused cellulose kraft paper are also subjected to FTIR.
The peaks obtained during FTIR characterizations are compared and analysed to
comment on the compositional changes with aging.
 Mineral Oil and Ester Based Oil/Paper Insulation Decaying Assessment 617

3 Results and Discussion

To understand the changes in chemical composition of oil/paper insulation with aging,

different aging factors have been simulated under laboratory conditions. The changes in
the surface functional groups are monitored by FTIR.

3.1 FTIR Spectrum of Mineral Oil

FTIR spectra of fresh and aged MO at various aging factors is presented in Fig. 1.

2000 h aged

1500 h aged
Absorbance (a.u)

1000 h aged

500 h aged
2922
2854

1458
1378

Non-aged

4000 3500 3000 2500 2000 1500 1000 500

Wavenumber (cm-1)

Fig. 1. FTIR spectra of fresh and aged MO

Broad O-H stretching vibration is observed in the fresh and aged samples. How-
ever, intensity profile of O-H peaks is noticed to be low. Figure 2(a) shows the high
frequency FTIR regions for peak positions of 2922 and 2854 cm−1. These peaks are
attributed to −CH2 vibrations accompanied by a small peak at 2954 cm−1 that is related
to −CH3 vibrations [9]. The polar type of hydrocarbon bonds are the internal chemical
signatures of MO. The degradation performance of MO is studied by observing the
peak area under CH bonds for different aging conditions as shown in Fig. 2(b). The
hydrocarbon group from the 2000 h MO showed decrease in the CH bond peak area
compared to the fresh MO.
618 U. Mohan Rao et al.

2000 h aged

2922
2954

2854
1500 h aged
Absorbance (a.u)

1000 h aged

500 h aged

Non-aged

3200 3100 3000 2900 2800 2700 2600

Wavenumber (cm-1)
(a)
100

80
Relative area (%)

60 C-H peak area

40

20

0
0 500 1000 1500 2000
Aging time (h)
(b)

Fig. 2. (a) Mineral oil peak positions at 2954, 2922 and 2854 cm−1 (b) the corresponding area
quantifications

FTIR spectrum of peak positions for 1458 and 1378 cm−1 are presented in Fig. 3.
The peak positions are attributed to −CH2 bending vibrations that result in the similar
trend as the main −CH2 and −CH3 vibrations.
 Mineral Oil and Ester Based Oil/Paper Insulation Decaying Assessment 619

1458

1378
2000 h aged

1500 h aged
Absorbance (a.u)

1000 h aged

500 h aged

Non- aged

1500 1400 1300

-1
Wavenumber (cm )

Fig. 3. Mineral oil peak positions of 1458 and 1378 cm−1.

3.2 FTIR Spectrum of Synthetic Ester

FTIR spectra of fresh and aged synthetic ester at 500, 1000, 1500 and 2000 h are
presented in Fig. 4.

2000 h aged

1500 h aged
Absorbance (a.u)

1000 h aged

500 h aged

Non- aged

4000 3500 3000 2500 2000 1500 1000 500

-1
Wavenumber (cm )

Fig. 4. FTIR spectra of SE at different aging conditions

620 U. Mohan Rao et al.

The chemical signature of synthetic ester resulting as CH2 and CH3 vibrations are
noticed at the peak positions of 2854 and 2922 cm−1 respectively. In addition, C = O
peaks positions are also observed in all the samples at 1750 cm−1. To understand the
influence of surface oxidation on synthetic ester, hydroxyl peak (OH) at 3200–
3500 cm−1 is quantified and the relative area is shown in Fig. 5. The relative increase
in the intensity of OH vibrations at higher aging times is related to surface oxidation of
its chemical structure (see Fig. 5(b)) [11]. Similarly, ester’s signature peak of carboxyl
group for aging factors and corresponding relative area is plotted in Fig. 6.

0.5 100

Fresh
OH peak 500 h
0.4 80 OH peak
1000 h
1500 h Relative area (%)
2000 h
0.3 60
Absorbance

0.2 40
Fresh

0.1 20

0.0 0
3200 3300 3400 3500 3600 3700 0 500 1000 1500 2000

Wavenumber (cm ) -1 Aging time (h)

(a) (b)

Fig. 5. (a) FTIR peak position at 3200–3500 cm−1 and its (b) corresponding peak area (SE)

1.0
100
Fresh
C=O peak 500 h
0.8 C=O peak
1000 h 80
1500 h
2000 h
Relative area (%)

0.6 60
Absorbance

0.4 40

0.2 20
Fresh

0.0 0
1600 1650 1700 1750 1800 1850 1900 0 500 1000 1500 2000
Aging time (h)
Wavenumber (cm-1)
(a) (b)

Fig. 6. (a) FTIR peak position at 1750 cm−1 and its (b) corresponding peak area (SE)
 Mineral Oil and Ester Based Oil/Paper Insulation Decaying Assessment 621

The relative area associated with C = O peaks is observed to increase with aging of
insulation. However, the percentage relative area at different aging factors at 1500 and
2000 h resulted in a small change as compared to fresh SE.

3.3 FTIR Spectrum of Cellulose Kraft Paper Aged in MO and SE

Further, to understand the properties of oil/paper interactions and their changes in the
chemical structure. Aging of MO and SE is carried out in presence of cellulose. Kraft
paper is degreased in hexane fumes to remove the traces of oil absorbed by cellulose
fibres. Degreasing is done in order to observe the chemical compositional changes that
occur in the insulation paper with aging of oil/paper insulation. The FTIR peaks of
degreased kraft paper aged in mineral oil at different aging factors is presented at Fig. 7.

1500 h

1000 h
Absorbance (a.u)

500 h

Mineral oil

Kraft paper

4000 3500 3000 2500 2000 1500 1000 500

Wavenumber (cm-1)

Fig. 7. FTIR spectra of cellulose kraft paper aged in Mineral oil

Fresh and dehydrated kraft paper showed the broad range of signature at the
hydroxyl group from 3500 to 3000 cm−1. This is the result of atmospheric OH inter-
action with the chemical structure of cellulose. Apart from this, the strong absorbance
peak at 2200 cm−1 is a result of C  C functional group from the cellulose structure. In
fingerprint region, below 1500 cm−1, very less information observed. In comparison to
the fresh kraft paper, kraft paper aged in MO shows a distinct feature at around
1700 cm−1. This is attributed to the C = O vibrations as a result of oil interaction with
cellulose during aging [10]. To understand the decomposition of C  C functional
group, aged samples of 2200 cm−1 is plotted as shown in Fig. 8.
622 U. Mohan Rao et al.

1.0

Fresh 2300-2000
500 h 60
0.8 1000 h
1500 h

Relative area (%)

0.6
Absorbance

40

0.4

20

0.2

0.0 0
1900 2000 2100 2200 2300 0 200 400 600 800 1000 1200 1400 1600

Wavenumber (cm-1) Aging time (h)

(a) (b)

Fig. 8. (a) FTIR peak position at 2200 cm−1 and (b) the corresponding relative peak area of
cellulose paper aged in MO

There are two distinct signatures observed at 2150 and 2200 cm−1 which is
ascribed to alkynes of C  C. The calculated relative peak area indicate a significant
decrease in quantity of C  C for aged samples. This correspond well with the
appearance of new peak of C = O at 1700 cm−1. FTIR spectra of degreased kraft paper
aged synthetic ester in illustrated in Fig. 9.

1500 h

1000 h
Absorbance (a.u)

500 h

Synthetic ester 7131

Kraft paper

4000 3500 3000 2500 2000 1500 1000 500

Wavenumber (cm-1)

Fig. 9. FTIR spectra of cellulose kraft paper aged in SE

 Mineral Oil and Ester Based Oil/Paper Insulation Decaying Assessment 623

Distinctive signature of cellulose C  C functional group is observed for kraft

paper aged in SE. However, the intensity of C  C vibrations decreased due to the
surface coverage of synthetic ester on the cellulose chemical structure. In addition, the
signatures of −CH2 and CH3 vibrations are clearly observed along with the cellulose
structure at 2854 and 2954 cm−1. The C = O signature at 1700 cm−1 on ester internal
chemical structure has a shift when it reacted with cellulose on aged samples. However,
the shift was observed at 1650 cm−1 along with the new peak appeared at 1550 cm−1.
These peaks were attributed to C = O and C = C respectively. Similarly, the functional
group of C  C vibrations are used for quantification analysis and the relative area
shows much lesser degradation than the mineral oil. The corresponding peak area and
quantification results are shown in Fig. 10.

1.0 60

Fresh 2300-1900
500 h
0.8 1000 h
1500 h
Absorbance

40
0.6
Absorbance

0.4
20

0.2

0.0 0
2000 2050 2100 2150 2200 2250 2300 0 200 400 600 800 1000 1200 1400 1600
-1
Wavenumber (cm ) Aging time (h)
(a) (b)

Fig. 10. (a) FTIR peak position at 2200 cm−1 and (b) Corresponding relative peak area of
cellulose paper aged in SE

4 Conclusion

FTIR spectra of fresh and aged samples of mineral oil and synthetic ester are studied at
different aging durations at 115 °C. The spectrum of mineral oil showed distinctive
peaks positions of its internal chemical structures at CH2 and CH3 vibrations. The
corresponding quantification analysis confirmed very small reduction in its peak area.
Similarly, synthetic ester displayed the important C = O vibrations. Chemical stability
of synthetic esters under different aging times confirmed the strong surface property of
synthetic ester by relative area quantification. Compared to the surface interaction of
mineral oil with Kraft paper under different aging conditions, synthetic ester showed
significant stability in its chemical structure. This could be due to its internal chemical
structure and the presence of C = O group on its structure.
624 U. Mohan Rao et al.

References
1. Fofana, I.: 50 years in the development of insulating liquids. IEEE Electr. Insul. Mag. 29(5),
13–25 (2013)
2. Fernandez, I., Ortiz, A., Delgado, F., Renedo, C., Perez, S.: Comparative evaluation of
alternative fluids for power transformers. Electr. Power Syst. Res. 98, 58–69 (2013)
3. Hamdi, A., Fofana, I., Mahi, D.: Stability of mineral oil and oil–ester mixtures under thermal
ageing and electrical discharges. IET Gener. Transm. Distrib. 11(9), 2384–2392 (2017)
4. Mohan Rao, U., Sood, Y.R., Jarial, R.K.: Oxidation stability enhancement of a blend of
mineral and synthetic ester oils. IEEE Electr. Insul. Mag. 32(2), 43–47 (2016)
5. ASTM D117-18: Standard Guide for Sampling, Test Methods, and Specifications for
Electrical Insulating Liquids. ASTM International, West Conshohocken (2018)
6. Abdelmalik, A.A.: Analysis of thermally aged insulation paper in a natural ester-based
dielectric fluid. IEEE Trans. Dielectr. Electr. Insul. 22(5), 2408–2414 (2015)
7. Tokunaga, J., Koide, H., Mohami, K., Hikosaka, T.: Comparative studies on the aging of
thermally upgraded paper insulation in palm fatty acid ester, mineral oil and natural ester.
IEEE Trans. Dielectr. Electr. Insul. 23(1), 258–265 (2016)
8. Fofana, I., Bouaicha, A., Farzaneh, M.: Characterization of aging transformer oil–pressboard
insulation using some modern diagnostic techniques. Eur. Trans. Electr. Power 21, 1110–
1127 (2011)
9. Munajad, A., Subroto, C., Suwarno, S.: Fourier transform infrared (FTIR) spectroscopy
analysis of transformer paper in mineral oil-paper composite insulation under accelerated
thermal aging. Energies (11), 1–12 (2018). 364
10. Bandara, K., Ekanayake, C., Saha, T.K., Annamalai, P.K.: Understanding the aging aspects
of natural ester based insulation liquid in power transformer. IEEE Trans. Dielectr. Electr.
Insul. 23(1), 246–257 (2016)
11. Mohan Rao, U., Pulluri, H., Kumar, N.G.: Performance analysis of transformer oil/paper
insulation with ester and mixed dielectric fluids. IEEE Trans. Dielectr. Electr. Insul. 25(5),
1853–1862 (2018)
 Experimental Research on Cumulative
Deformation of Transformer Winding Induced
by Short-Circuit Current Impacts

Fuqiang Ren(&), Shengchang Ji, and Lingyu Zhu

State Key Laboratory of Electrical Insulation and Power Equipment,

Xi’an Jiaotong University, Xi’an 710049, China
renfuqiang@stu.xjtu.edu.cn

Abstract. Power transformer is one of the most important apparatus to the

normal operation of power system. The main cause of winding deformation for
transformers is the enormous electromagnetic force induced by the short-circuit
current. Therefore, it is significant to study the winding deformation under this
condition. The sweep frequency impedance (SFI) technique, a combination of
frequency response analysis (FRA) and short-circuit impedance (SCI) method,
contains more reliable indices to detect winding deformation accurately. In this
paper, the cumulative effect of the winding deformation under repeated short-
circuit current impacts is experimentally studied based on a customized power
transformer. The SFI data of the transformer is acquired every time after the
impact, with which the condition of the transformer winding can thus be eval-
uated. The disassembly results of the transformer after the whole experiments
reveals that severe radial deformation and short-circuit fault within the winding
have occurred. The SFI data obtained after every impact indicates that the
winding condition just holds slight change with relatively small short-circuit
current value in the initial stage while deteriorates irreversibly with the increased
current value and test times. This finding can illustrate the cumulative tendency
of winding deformation for the actual in-service transformer to some degree. For
the actual transformer suffering many short-circuit current impacts in service, the
deformation for its winding may initially develop slowly but in latter stage can
abruptly collapse after many short-circuit current impacts, which suggests that
timely overhaul on older transformers is necessary.

Keywords: Correlation coefficient
Short-circuit current
Sweep frequency

impedance

1 Introduction

Power transformer is one of the most critical components in power system [1]. The
transformers in service are vulnerable to mechanical defects for various reasons, such
as transportation, installation, load shedding, insulation failing, and short circuits. The
mechanical changes, mostly resulted in by electromagnetic forces, typically manifest as
winding deformation, which will ultimately develop into transformer failures if not
diagnosed in time [2]. According to the relevant statistics, the rate of transformer

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 625–634, 2020.
https://doi.org/10.1007/978-3-030-31676-1_59
626 F. Ren et al.

winding deformation caused by short-circuit fault currents is high, therefore, it is

significant to study winding faults under short-circuit condition in order to keep its safe
running.
In general, power transformers may suffer multiple times of short-circuit current
impacts during their service life. However, a single impact will hardly lead to the
collapse of winding but can induce elastic or slight plastic deformation of winding.
Further, these tiny defects will accumulate into severe irreversible winding deformation
with the increased impact times and lead to the deterioration of the transformer.
Therefore, it is significant to study the cumulative tendency of the winding deforma-
tion. However, few paper can be found in this field due to the complicated simulation
or the rare chance for the repeated short-circuit impact tests on actual power trans-
formers. Wang [3] analyzed the cumulative deformation of transformer winding under
short-circuit fault with a magnetic–structural coupling model, considering the rela-
tionship between total plastic deformations and impacts of short-circuit current and the
electromagnetic force. In addition, rare report on experimental research of the cumu-
lative winding deformation can be found.
In this paper, a power transformer is designed to suffer multiple short-circuit current
impacts. Every time after the impact, the SFI trace of transformer is recorded, from
which two indices, namely the SCI percentage value of winding at the rated frequency
(Zk) and the correlation coefficient (CC) between different SFI traces, can be obtained.
Finally, with Zk and CC, the condition of transformer winding is evaluated and thus the
cumulative effect of the deformation is analyzed.

2 Experimental Setup

In this paper, a three-phase three-winding power transformer is custom-made to suffer

the impact test, with a rated capacity of 50 MVA, a rated voltage ratio of
110/35/10 kV, and a connection group of YNynd11, as shown in Fig. 1.

Fig. 1. The tested power transformer.

 Experimental Research on Cumulative Deformation of Transformer Winding 627

The short-circuit tests are conducted according to standard [4]. The duration of
impulse current is 0.25 s, and the short-circuit tests are conducted for 33 times. The
current is injected into high-voltage winding with the low-voltage winding opened and
the medium-voltage winding shorted.

I ¼ Isafe
70% ð1Þ

Current I calculated with Eq. (1) is set as the baseline current value injected into the
winding. During the tests, the ratio of actual current to baseline current I is use to
quantify the short current. The ratio values are listed in Table 1. The short current
injected into the winding is set to be gradually increased, from 75% to 105%. Then the
current is kept to be 95% until obvious winding deformation occurs. The SFI trace of
transformer after every test is acquired with the measurement equipment depicted in
Fig. 2.

Table 1. The current ratio in all 33 short-circuit impact rounds.

Test times Ratio Test times Ratio
1–3 70% 4–6 75%
7–9 80% 10–12 85%
13–15 90% 16–18, 25–33 95%
19–21 100% 22–24 105%

3 Theory of SFI Method

3.1 Measurement System
The layout of SFI measurement system is shown in Fig. 2. As can be seen, for the
measurement system, the driving voltage generated with signal source is amplified by
the power amplifier to enhance the signal-to-noise ratio and promote the anti-
interference ability of system. The SFI method combines the advantage of FRA and
SCI, with a typical measurement frequency range of 10 Hz–1 MHz. Therefore, Zk and
the SFI trace that resembles FRA curve can be obtained with a single SFI test. With
accurate frequency-dependent data and more diagnostic indices, SFI can be more
effective for winding deformation diagnosis [5].
As shown in Fig. 2, SFI measurement requires the secondary winding of trans-
former shorted, which is in accordance with the layout of short-circuit current impact
tests. Therefore, during the impact test, it is very convenient for the execution of SFI
measurement.

3.2 Calculation of Impedance

The amplified sinusoidal signal is injected into the driving point of winding, and its
value, U_ 1 , and the response voltage at neutral end, U_ 2 , are acquired by two sample
628 F. Ren et al.

resistors R1 and R2 respectively, whose values all equal 50 X. Therefore, the impedance
can be calculated as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  
U_ 1 ðjxÞ2 U_ 1 ðjxÞ

Zk ðjxÞ ¼ 50    2    cos ðhÞ þ 1 ð2Þ
U_ ðjxÞ 2 U_ ðjxÞ 2

where x is the angular frequency, and h is the difference of angular phase between
driving-point voltage (excitation voltage) and neutral-end voltage (response voltage).

Signal Generator
Transformer
Power U1
Amplifier R1
I1 I2

U2

R2 I3
Data
Acquisition
Device

Fig. 2. SFI measurement setup.

With the impedance at the rated frequency, Zk for the three-phase transformer can
be calculated by
pffiffiffi
3  Z  Ir
Zk ¼ ð3Þ
Ur

where Ir and Ur are respectively the rated current and voltage of the tested power
transformer.

3.3 Diagnostic Indices of SFI

Percentage SCI Value at Rated Frequency Zk

The rated Zk can be obtained with the data plate of the power transformer. Zk is related
to its magnetic flux leakage determined by the specific structure of the transformer
winding. Therefore, this parameter can be regarded as an index to evaluate the winding
 Experimental Research on Cumulative Deformation of Transformer Winding 629

condition since the change of winding condition will surely alter its structure and thus
induce the variation of Zk. According to the Chinese standard applicable to the
transformer with a Zk larger than 4% [6], the threshold of this index is defined in
Table 2. In this standard, the transformer with a rated capacity below 100 MVA and a
voltage grade below 220 kV is defined as the I-type transformer and that with a rated
capacity above 100 MVA and a voltage grade above 220 kV is the II-type.

Table 2. Threshold of Zk
Comparison mode Transformer type Threshold
Vertical comparison I ±2.0%
II ±1.6%
Horizontal comparison I ±2.5%
II ±2.0%

CC Between Two Curves

According to the Chinese standard [7], frequency range of 1 kHz–1 MHz is manually
divided into three parts, namely low frequency (LF) range of 1 kHz–100 kHz, middle
frequency (MF) range of 100 kHz–600 kHz, and high frequency (HF) range of 600
kHz–1 MHz. In general, the changes of SFI curve on resonance positions and
amplitudes at LF range indicate the variation of lumped inductance. Similarly, changes
in MF range usually mean the variation of distributed capacitances and inductances,
and changes in HF range signify the alteration of stray capacitances of winding [8]. The
calculation of CC can be found in [7]. The deformation levels and the corresponding
RXY thresholds at LF, MF, and HF ranges are illustrated in Table 3.

Table 3. Threshold of CC
Deformation level Threshold for Rxy
Severe RLF \0:6
Moderate 0:6  RLF \1:0 or RMF \0:6
Slight 1:0  RLF \2:0 or 0:6  RMF \1:0
Normal RLF  2:0; RMF  1:0; and RHF  0:6

4 Experimental Results

Before short-circuit impact experiments, SFI data for original transformer is acquired as
the reference fingerprint. Further, the SFI curves are collected every time after the
impact test, which are utilized to evaluate the condition of transformer winding, of
which several plots in the LF, MF, and HF ranges are exhibited in Fig. 3.
630 F. Ren et al.

600
Fingerprint The 21th test
The 5th test The 26th test
100
The 10th test The 32th test
The 15th test The 33th test
10

1
1 10 100
100
Impedance/kΩ

10

1
100 200 300 400 500 600
4

0.3
600 700 800 900 1000
Frequency(kHz)

Fig. 3. SFI curves in the LF, MF, and HF ranges.

As can be seen in Fig. 3, before the last (33rd) short-circuit current impact test, SFI
curves nearly keep consistent, which indicates that no obvious enormous variation on
winding condition exists after these tests. However, SFI curves do have some deviation
from each other but show no apparent law, which suggests that the transformer winding
should possess slight plastic deformation in this stage. However, as mentioned before,
no obvious irreversible deformation occurs.
However, after the last (33rd) short-circuit current impact test, the SFI has signif-
icant changes. In the frequency domain of 10 Hz–1 kHz, the SFI amplitude is sharply
dropt. While in 1 kHz–1 MHz, which is also the traditional frequency range of FRA,
SFI curve also holds descend or ascend trends of amplitude, the shift or disappearance
of resonance points, and the generation of new resonance peaks or troughs. This
phenomenon implies that severe winding deformation probably occurs after the last
impact abruptly. To further evaluate the evolution law of transformer winding condi-
tion under repeated short-circuit current impact, Zk and CC are calculated according to
the Subsect. 3.3 in this paper. The variation trend of Zk deviation to its original value
and the CC between the SFI plots of several tests and the fingerprint trace are separately
displayed in Fig. 4 and Table 4. The abbreviation “WF” in Table 4 refers to the ‘whole
frequency’, namely 1 kHz–1 MHz in this paper.
 Experimental Research on Cumulative Deformation of Transformer Winding 631

According to Chinese standard, the transformer in this paper belongs to the I-type.
Therefore, the threshold for vertical comparison of Zk is ±2%. In addition, in this
paper the value ±1% of vertical comparison of Zk is set as the alarm value, beyond
which the impact current is set to be constantly 95% to continue the test until apparent
severe winding deformation occurs. As can be seen in Fig. 4, the trend of Zk derivation
in the former 32 tests is magnified and exhibited in this figure as well. As can be seen,
the whole trend for Zk variation is rising with small fluctuation as impact time
increases. Zk variation exceeds the positive alarm deviation value after the 24th impact
test. Afterwards, the deviation value is kept above 1% and ascending. However, after
the 33th test, the winding collapses abruptly with a sharp drop of Zk deviation
(−21.85%), indicates severe irreversible plastic winding deformation has occurred.
As can be observed in Table 4, for the former 32 tests, with the test time increased,
CC in WF range between fingerprint and the SFI curve is dropt on the whole.
According to Tables 3 and 4, CC value in LF is between 1 and 2 and keeps dropping as
well, meaning that slight winding deformation occurs within the transformer in this
stage and the degree of winding deformation gradually gets higher with the increasing
impact times. However, after the 33rd impact test, the CC of LF is sharply down to
0.52, which implies that the winding deformation gets severe during this impact,
holding similar rule that Zk possesses.

0
Deviation to original Zk/%

-5
1.2

1.1

-10 1.0

0.9

0.8

-15 0.7

0.6
Deviation to original Zk/%
0.5
Positive threshold of Zk
-20 0.4
Negative threshold of Zk
0.3
0 5 10 15 20 25 30 35 Positive alarm value
Negative alarm value
-25
0 5 10 15 20 25 30 35
Test times
Fig. 4. Variation trend of Zk Versus impact times.
632 F. Ren et al.

Table 4. CC between SFI plots in several rounds

Test times CC of WF CC of LF CC of MF CC of HF
5th 1.83 1.82 1.71 2.51
10th 1.47 1.52 1.38 1.85
15th 1.28 1.42 1.66 3.75
20th 1.53 1.73 1.53 2.04
26th 1.47 1.57 1.45 1.44
32nd 0.84 1.16 1.54 1.57
33rd 0.87 0.52 1.01 1.54

As mentioned before, the impact current is gradually increased initially and then is
kept as a constant value when Zk deviation exceeds ±1%, to emulate the repeated
short-circuit current impacts that the actual in-service power transformer in the system
sustains as much as possible. From above discussion, both these two indices have
shown consistent evolution rules on winding cumulative deformation. At the initial
stage, under a shorter impact current, the winding might hold slight plastic deformation
and accumulate to severe deformation gradually with the increased short-circuit current
value and the impact times, which is evidenced by a Zk deviation from 0.42% to 1.16%
and a CC of LF from 1.82 to 1.16 in this stage. With the cumulative deformation, the
ability for winding to withstand short circuit is weakened. Then, in the last impact test,
the winding abruptly collapses with severe deformation, with a sharp drop of Zk
deviation and CC, separately from 1.16% to −21.85% and from 1.16 to 0.52. Hence,
for this newly-designed power transformer, the repeated short-circuit current impact
tests lead to slight winding deformation initially, which gradually gets severe with the
increased current value and the impact times. In the evolution progress, the ability for
transformer to withstand short circuit is weakened, meaning that the winding may
collapse at any time when next impact occurs, which has been evidenced in this paper.
In addition, in the experiments in this paper, the mechanical properties variation of the
metal and insulation materials for actual transformer in service is hardly possible to
emulate, which will even aggravate the ability to withstand the short circuit. Therefore,
in actual occasion where the in-service power transformer has been in operation for
years and has sustained several times of short-circuit current test, timely monitoring on
its condition is very necessary, since the winding may collapse with severe winding
faults during the next impact, as revealed in this paper.
For the last impact test, since the Zk deviates from the original value by −21.85%, it
is inferred that short-circuit fault has occurred inside the winding, which is proved by
the disassembly results of the transformer, as shown in Fig. 5, with obvious ablation
trace inside the winding. As can be seen, in addition to short-circuit fault, severe
winding radial deformation, induced by the enormous electromagnetic force, also
occurs in the middle of high voltage winding.
 Experimental Research on Cumulative Deformation of Transformer Winding 633

Fig. 5. Disassembly results of power transformer.

5 Conclusion

This paper studies the cumulative deformation of the transformer winding under short-
circuit current impact tests. After every impact test, the SFI curves are acquired and
analyzed. Two indices, namely Zk deviation and CC between the fingerprint and the
SFI trace after every test are utilized to evaluate the winding condition. It is found that
Zk gradually rises with the impact times increased, indicating the winding holds slight
plastic deformation that slowly gets severe at this stage, and at the same time the anti-
short circuit ability of the winding is weakened as well. Therefore, the winding col-
lapses abruptly with a sharp drop of Zk. Similar rules can be obtained as well for CC.
The disassembly results of transformer demonstrated that the winding impacted by
short-circuit current holds severe deformation of radial bulking and short-turns inside
the winding, which offers a validation of the above analysis. It is advised that timely
overhaul on older transformers, which may have been impacted for many times, is
necessary, since winding collapse may be induced merely by even one impact on these
transformers.

References
1. Pham, D., Pham, T., Borsi, H., Gockenbach, E., et al.: A new method for purposes of failure
diagnostics and FRA interpretation applicable to power transformers. IEEE Trans. Dielectr.
Electr. Insul. 20(6), 2026–2034 (2013)
2. Behjat, V., Vahedi, A., Setayeshmehr, A., et al.: Diagnosing shorted turns on the windings of
power transformers based upon online FRA using capacitive and inductive couplings. IEEE
Trans. Power Delivery 26(4), 2123–2133 (2011)
634 F. Ren et al.

3. Wang, S., Zhang, H., Wang, S., et al.: Cumulative deformation analysis for transformer
winding under short-circuit fault using magnetic-structural coupling model. IEEE Trans.
Appl. Supercond. 26(7), 0606605 (2016)
4. IEC Standard 60076-5: Power transformers-part 5: ability to withstand short circuit (2006)
5. Liu, Y., Ji, S.C., Yang, F., et al.: A study of the sweep frequency impedance method and its
application in the detection of internal winding short circuit faults in power transformers.
IEEE Trans. Dielectr. Electr. Insul. 22(4), 2046–2056 (2015)
6. China Standard DL/T 1093-2008: Guide for reactance method to detect and diagnose winding
deformation of power transformer (2008)
7. China Standard DL/T 911-2004: Frequency response analysis on winding deformation of
power transformer (2004)
8. IEC Standard 60076-18: Measurement of frequency response (2012)
 Construction of Novel Online Partial Discharge
Monitoring System in Model Power Apparatus
Using Power Line Communication

Hina Yoshikawa1(&) , Seiya Masuda1, Yasutomo Kakimoto1 ,

Nuraida Puspitasari1, Yusuke Nakano1 , Masahiro Kozako1 ,
Masayuki Hikita1 , Hidefumi Sato2, Hideki Tagashira2,
and Kazuhito Ishida2
1
Kyushu Institute of Technology, 1-1 Sensui-Cho,
Kitakyushu-Shi, Fukuoka, Japan
o108133h@mail.kyutech.jp
2
Kyushu Electric Power Co., Inc., Fukuoka, Japan

Abstract. In recent years, partial discharge (PD) measurement attracts attention

as insulation diagnosis of power equipment. A lower-cost and higher sensitive
detection method is required, but it has not been established yet. In this research,
an attempt is made to apply a ferrite core (FC), usually used for suppression of
high frequency noise in electric circuit, and power line communication (PLC),
known as a data communication on AC power transmission. By combining
those tools, we aim to develop the low cost and high sensitive continuous
monitoring technique. In the previous paper, the frequency characteristic of
impedance of FC was clarified and PD measurement using FC was performed.
This paper presents a novel concept of PD detection by PLC and FC, con-
structed test circuit and experimental verification. In general, PLC is used for
information communication by superimposing the data on distribution systems
with commercial frequency. Principle of the novel PD measurement using PLC
is as follows: PD measurements are made by injecting PD signal generated in an
artificial defect model into the constructed test circuit consisting of PLC and FC.
The inductance component of FC causes dominant frequency band that PD
signal originally possesses to shift to the frequency band allowable to transmit in
PLC between 2 MHz and 28 MHz. In addition, the impedance of the FC
amplifies the signal intensity detected from a test lead connected to the two
terminals in the ground line. In principle, PD detection can be realized by
identifying the PD signal as a noise signal in the communication frequency
region of PLC. Note that PD signal causes a communication error resulting in an
abnormal state in PLC. The novel PD detection method extracts PD signal as a
disturbance from the communication signal. An attempt is also made to measure
PD using the constructed monitoring system so as to verify the PD detection
principle.

Keywords: Power line communication (PLC)
Ferrite core (FC)
Partial

discharge (PD) detection

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 635–642, 2020.
https://doi.org/10.1007/978-3-030-31676-1_60
636 H. Yoshikawa et al.

1 Introduction

At present, partial discharge (PD) measurement attracts attention as insulation diag-

nosis of power equipment. Because PD does not occur continuously, online constant
monitoring is useful. Under the situation, high precision and reliable online PD
monitoring has not been established yet. Besides, inexpensive PD detection method is
required [1–6] depending on kind of equipment. In this research, we aim to construct a
novel PD monitoring system using inexpensive power line communication (PLC). In
the previous report [7], it was found that the signal detected with Ferrite core (FC) as
the sensor has a strong signal strength in PLC communication frequency 2 to 28 MHz
band. This paper reports on the PD monitoring system construction and PD detection
results using the principle that PLC generates errors due to disturbance arising from
PD.

HV Cc
60 Hz 1000pF PET film
50 μm PG Pulse calibrator
200μm

Power Equipment
(PD generation circuit)
Ferrite cores Disturbance (PD)
HPF PG Simulated pulse generator

Transmitting side A Receiving side

PLC signal
P
PC 2 PLC ATT PLC PC 1
B
Variable attenuators

Noise Filter Noise Filter Transformer AC 100V

AC100 V

Fig. 1. PD monitoring system using PLC and FC.

2 Construction of Monitoring System by PLC

2.1 PD Monitoring by PLC

Figure 1 shows a PD detection and monitoring circuit using PLC (I-ODATA, PLC-HP
240EA-S, 2 to 28 MHz). In this circuit, PD signal is regarded as disturbance to PLC
signal. This state is recognized by PLC as deterioration of the communication state,
thereby changing the modulation method. This change indicates that a PD signal has
been detected. Normally PLC signal is superimposed on the power supply line, but for
this research PLC signal is separated from the power supply signal. A noise filter (TSK
- Lambda, RSEN - 2010) is inserted in the power supply line (blue line) of PLC,
thereby cutting off PLC signal. On the other hand, the insertion of the noise filter
transformer (NF Corporation, NT - 500C/NT - 1000C) to the transmission line (green
 Construction of Novel Online Partial Discharge Monitoring System 637

line) of PLC signal cuts off the commercial power supply. This configuration allows
only the PLC communication signal to be attenuated by a variable attenuator: ATT
(KEYSIGHT, 8496 A, 8494 A, 1 to 121 dB), i.e. the PD signal is regarded as
disturbance.

2.2 PLC Characteristics on Disturbance Occurring

Note in Fig. 1 that PD signal (red line) enters point P (receptacle plug) and acts on PLC
signal (green line). In a normal PLC state, the communication is carried out by a
modulation method 32 PAM (pulse amplitude modulation) that makes the information
transmission speed become the most efficient. On the other hand, disturbance larger
than PLC signal prohibits the receiver side PLC from receiving it. Under the situation,
changing modulation scheme (for instance, 2 PAM) to the suitable state permits stable
communication to be achieved (change between 2 and 32 PAM). This confirmation is
performed by PC 1 so as to take the state of the communication path as a change of the
modulation method. Note that the initial state is set when the power of the PLC is
turned on. Decrease in PAM indicates a sign that PD signal has been detected.

2.3 PLC Signal Control and PD Signal Detection PLC

The ATT is inserted into the middle (green line) of the receiving/transmitting side PLC
as shown in Fig. 1. It attenuates PLC signal level to create a state susceptible to
disturbance. Figure 2 illustrates relationship between PLC signal and PD signal when
the ATT is used. It is seen in the figure that increasing the attenuation level decreases
the magnitude of the PLC signal (blue to green) so that PD signal can be detected. Note
that PLC signal can be set to communicate in the first and third quadrants or second and
fourth quadrants. In Fig. 2, communication states in the first and third quadrants are
shown.

Utility frequency

PD signal
Voltage

Time

PLC signal PLC signal

(Use ATT) (No use ATT)

Fig. 2. Schematic diagram of the relationship between ATT and PLC signals.
638 H. Yoshikawa et al.

3 Simulation Pulse Signal Detection by PLC

Figure 3 shows an example of PLC signal having a larger signal level at 1.9 to
15 MHz. The PD generating circuit in Fig. 1 was removed, and instead a pulse gen-
erator (NF Corporation, WF 1968) was connected (purple line). Pulse signals with a
given voltage (Vpp) and time width (tw) at 5 MHz were injected through high pass filter
(HPF, Mini-Circuits, ZFHP- 1 R 2 +, 1.2 MHz or more).

Fig. 3. Frequency of PLC transmission power level Pattern I.

Figure 4 shows an example of a change in PAM value when PD simulated pulse

with Vpp = 7 V and tw = 50 ns is injected. Results revealed that there were the cases
where PD detection became impossible depending on Vpp and tw. It was found from
this result that PLC has thresholds of Vpp and tw to recognize signal as disturbance.
Error(Pulse) First condition

2 4 8PAM 16PAM 32PAM

D2 2PAM 4PAM 8PAM

0 10 20 30 40 50 60 70

PAM(%)

Fig. 4. PAM value of Initial condition and injecting PD simulated signal with Vpp = 7 V and
tw = 50 ns.

Figure 5 shows the critical ATT attenuation rate as a function of thus determined
threshold voltage Vpp as recognized disturbance by PLC for PD detection for simulated
pulses with tw = 10 ns and different Vpp = 50 mV to 7 V. It can be seen from Fig. 5
that the critical attenuation rate of the ATT decreases with increasing Vpp.
 Construction of Novel Online Partial Discharge Monitoring System 639

120

100

Attenuator value [dB]

80

60

40

20

0
1 2 3 4 5 6 7

Voltage [V]

Fig. 5. The critical ATT attenuation rate as a function of voltage of PD simulated pulse as
recognized to be disturbance by PLC.

Next, an attempt was made to investigate the circuit at Point P with the charge
amount of PD injected in Fig. 1 against changes in the PAM value of PLC, i.e. the
influence of PD detection at PLC circuit will be examined. Using a pulse calibrator
(NIHON KEISOKUKI SEIZOSHO CO., LTD., NPG-2) (broken line), a constant
charge amount was injected to a needle-plate electrode with needle diameter 1 mm and
inter-electrode distance 200 lm, sandwiching a PET film (50 lm). An insulated wire
with five FCs (SEIWA, E04SR401938) was connected to the ground wire side. A test
lead was connected to the upper and lower terminals of the ground wire, and PD
waveform was recorded with an oscilloscope (Rhode & Schwartz, RTH 1004,
500 MHz, 2.5 GS/s).
Figure 6 shows waveforms detected with FCs for the charge amount q = 2000 pC
detected at points A and B. It is noticed from the figure that the pulse signal magnitude
attenuates after passing through point P.

8
FC:n=5(PG)
4
From HPF (Point A)
0
Signal intensity [V]

-4
8

4
From outlets (Point B)
0

-4
-200 0 200 400 600 800 1000
Time [ns]

Fig. 6. The Signal waveforms detected at Point A and B for injected charge q = 2000 pC.
640 H. Yoshikawa et al.

4 Detection of Real Partial Discharge Signal by PLC

AC voltage (KIKUSUI, TOS 5050 A) of 1.7 kV was applied to the needle-plate

electrode to generate PD. Waveform of PD pulse extracted by five FCs was recorded
with an oscilloscope (Tektronix, DPO 4054, 2.5 GS/s). Simultaneously, an attempt was
made to detect PD for the PLC transmission power level frequency pattern as drawn in
Figs. 3 or 7.

Fig. 7. Frequency of PLC transmission power level Pattern II.

Figures 8 and 9 show an example of PD signal waveform extracted with five FCs
and its wavelet transformation when PLC determines PD as disturbance. It should be
noticed that the detected waveform given in Fig. 8 exhibits Vpp = 9 V and tw = 40 ns,
which exceeds the threshold of PD detection by the PLC system. It is found from Fig. 9
that the signal strength is high at 5 MHz and 20 MHz or less. This result means that the
PLC system satisfies the condition for detecting the disturbance when five FCs are used
as a sensor.

10

8 FC : n=5
6
Signal intensity [V]

-2

-4

-200 0 200 400 600 800 1000 1200

Time [ns]

Fig. 8. One example of PD signal waveform detected with five FCs when partial discharge is
regarded as disturbance by the PLC system.
 Construction of Novel Online Partial Discharge Monitoring System 641

Figure 10 shows an example of PAM value change pattern when the PLC system
detects actual PD signal only for the PLC transmission power level pattern given in
Fig. 7. As can be seen in the figure, the PLC communication changes from 16 PAM to
2 PAM, meaning that the PD signal can be detected. Further investigation is underway
on the relationship between the PLC transmission power level pattern and the PD
detection sensitivity, that is, the PAM value change pattern.

100
Frequency [MHz]

10

9
0 1 2 3 4 5 6 7 8 9 10x10
Time [150ns/div]

Fig. 9. Wavelet transformation of the waveform given in Fig. 8.

First condition

16PAM
Error(PD)

D2 2PAM 4

0 5 10 15 20 25 30 35 40

PAM(%)

Fig. 10. Example of PAM value change pattern when the PLC system detects actual PD signal
for the PLC transmission power level pattern given in Fig. 7.

5 Conclusion

In this paper, we constructed a novel PD signal monitoring system that extracts PD

using FC and monitors PD by PLC. It was found that PLC has thresholds of Vpp and tw
to recognize signal as disturbance by injecting the pulse signal as a disturbance. Fur-
thermore, it can be concluded that PLC with transmission power level frequency
pattern II of high frequency side is able to detect disturbance on PLC signals due to PD
642 H. Yoshikawa et al.

signal obtained from five FCs. In consideration of the above, in addition to improving
PLC dedicated software, we will study the optimum values of the transmission signal
level, communication frequency band, phase and ATT of PLC suitable for PD detec-
tion for practical application. Moreover, by improving the optimum number of FC and
signal strength in the PLC circuit, we aim to construct a more accurate monitoring
system.

References
1. Liao, Y., Feng, B., Gu, X., Sun, T., Xu, Y., Zhang, Z.: Application of the online partial
discharge monitoring for the EHV XLPE cable system. In: International Conference on
Condition Monitoring and Diagnosis, pp. 913–916 (2016)
2. Blokhintsev, I., Kozusko, J., Oberer, B., Anzaldi, D.: Continuous and remote monitoring of
partial discharge in medium voltage switchgear. In: Electrical Insulation Conference, pp. 205–
208 (2017)
3. Siddiqui, B.A., Pakonen, P., Verho, P.: Novel inductive sensor solutions for on-line partial
discharge and power quality monitoring. Trans. Dielectr. Electr. Insul. 24, 209–216 (2017)
4. Rethmeier, K., Krüger, M., Kraetge, A., Plath, R., Koltunowicz, W.: Experiences in on-site
partial discharge measurements and prospects for PD monitoring. In: International Conference
on Condition Monitoring and Diagnosis, Beijing (2008)
5. Stone, G.C.: Partial discharge diagnostics and electrical equipment insulation condition
assessment. Trans. Dielectr. Electr. Insul. 12(5), 891–903 (2015)
6. Caprara, A., Cavallini, A., Garagnani, L., Guo, J.: A novel approach for continuous
monitoring of Partial Discharge phenomena on Medium Voltage equipments. In: Electrical
Insulation Conference, pp. 495–498 (2018)
7. Yoshikawa, H., Masuda, S., Nakano, Y., Kozako, M., Hikita, M., Mawatari, T., Matsumoto,
T.: Basic study on partial discharge detection with high sensitivity using ferrite core. In:
International Conference on Condition Monitoring and Diagnosis, P62 (2018)
 Identification of Partial Discharge Source
in Power Apparatus in Practical Substation
Utilizing Artificial Neural Network

Yasutomo Kakimoto1(&) , Nuraida Puspitasari1,

Hina Yoshikawa1 , Tadahiro Fujimura1 , Masahiro Kozako1 ,
Masayuki Hikita1 , Osamu Eda2, Yasuharu Shiina2,
Masanori Otsuka2, and Hiroshi Kaneko2
1
Kyushu Institute of Technology, 1-1 Sensui, Kitakyushu, Fukuoka, Japan
p108029y@mail.kyutech.jp
2
Tokyo Densetsu Service Co., Ltd., Tokyo, Japan

Abstract. In the previous report, two types of artificial neural network

(ANN) were constructed based on waveform data and phase-resolved partial
discharge (PRPD) patterns obtained from simulated defect samples in the lab-
oratory. In this report, the two types of ANN were used to discriminate between
PD and noise for signals acquired from a 72 kV tank type gas insulated
switchgear (GIS) installed at a substation. The signals obtained from the GIS
also include those caused by abnormalities due to floating electrodes that were
confirmed by internal inspection after the measurement. It was found that the
signals caused by the abnormality were mainly judged as noise and PD by
ANN_WP and ANN_PR, respectively. The disagreement by ANN_WP is
attributed to the waveform information change due to signal propagation and
attenuation, while ANN_PR provides better judgement owing to keeping the
phase information at which PD occur. As a result, ANN_PR is found to be more
effective means to judge the PD source in the field.

Keywords: Partial discharge
Artificial neural network
Waveform

parameter
Phase-resolve PD pattern
PD type classification

1 Introduction

In recent years, partial discharge (PD) phenomenon attracts attention as a precursor

phenomenon of dielectric breakdown in order to develop insulation deterioration
diagnosis technology with high reliability of power equipment. The purpose of this
research is to develop a diagnostic technology that can easily identify the partial
discharge source (PD source) inside the equipment by installing multiple sensors
outside the equipment and detecting PD signal. In previous reports [1, 2], PD signals of
four kinds of simulated electrodes and four types of artificial noise signals were
acquired using three kinds of sensors in the laboratory. Waveform parameters and

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 643–652, 2020.
https://doi.org/10.1007/978-3-030-31676-1_61
644 Y. Kakimoto et al.

phase resolved discharge (PRPD) pattern were extracted from the obtained signals. We
constructed artificial neural network (ANN) and conducted training and evaluation of
developed ANN. The trained ANN can classify the type of PD source inside the
equipment into (i) protrusion on high voltage conductor (POC), (ii) protrusion on
ground (POG), (iii) void defect (void), (iv) surface discharge defect (surface), and
distinguish from various noises. Several PD classification methods using various
algorithms have been reported [3–9, 10], but few cases apply to data acquired by actual
power equipment. From the viewpoint, the previous report [2] showed that the
developed ANN has been applied to data acquired by various sensors attached at a
cubicle type gas insulated switchgear C-GIS so as to provide successful identification
of PD signal sources [1, 2]. Following the previous report, this article presents an
attempt to identify PD sources from data acquired PD with the same PD sensors
attached on a 72 kV gas insulated switchgear GIS installed in a different substation
from that of the C-GIS.

2 ANN Training in the Laboratory

2.1 Construction and Training of ANN Waveform Parameters

(ANN_WP)
1000 signals with four electrodes simulating different PD sources, artificial noise and
background noise were obtained using three kinds of sensors (transient earth voltage
sensor TEV, surface current sensor SCS, and high frequency current transformer
HFCT) and a digital oscilloscope (HVPD Longshot (registered trademark), 200 MHz,
2.5 GS/s). From each PD signal and noise obtained, nine waveform parameters ((i)
PeakValue, (ii) Phase, (iii) Risetime, (iv) Falltime, (v) PulseWidth, (vi) PulseArea,
(vii) EventWidth, (viii) FreqOfMaxAmpl, (ix) Freq (1st Mom)) were extracted. These
nine parameters were used as training for ANN consisting of input layer using
waveform parameters (ANN_WP). ANN_WP was constructed with back propagation
(BP) algorithm using MATLAB program. Namely, ANN_WP consists of nine inputs
(9 waveform parameters), 8 outputs (4 simulated electrodes and 4 different types of
noise), and a hidden layer associated with inputs and outputs. Hyperbolic tangent
sigmoid function was used as an activation function by changing the number of neu-
rons in the hidden layer from 5 to 20. One thousand waveform signals caused by four
types of noise and four types of simulated electrodes were collected. Among the
number of signals, 700 data were used for training, 150 data for verification of
ANN_WP, and another 150 data were used for testing ANN_WP. Training ANN_WP
was done with changing the number of neurons from 5 to 20. As a result, it was found
that ANN_WP with 20 neurons showed the best performance with respect to training
ability and recognition rate.
 Identification of Partial Discharge Source in Power Apparatus 645

2.2 Construction and Training of ANN PRPD Pattern (ANN_PR)

To improve the reliability of Phase-Resolved Partial Discharge (PRPD) pattern, each
cycle of the PRPD pattern was divided into 60 phase angles with the width of each
phase angle window 6°. Three parameters are picked up for calculation in ANN using
PRPD pattern data (ANN_PR) as follows:
1. Number of pulses Hn (u) represents number of PD signals occuring in each phase
angle window 6°.
2. Maximum value Hmax (u) represents signal maximum value in each phase angle
window.
3. Mean value Havg (u) represents signal average value in each phase angle window.
Accordingly, 3  60 = 180 feature quantities were obtained from one cycle of
PRPD pattern.
Similar with ANN_WP, ANN_PR was constructed with back propagation
(BP) algorithm using MATLAB. It consists of 180 inputs, 8 outputs (representing 4
simulated electrodes and 4 types of noises) and 1 hidden layer with hidden neurons
varying from 5 to 20. 160 PRPD patterns were obtained for each simulated electrode
and artificial noise. 112 data were used for training, 24 data for verification and the
remaining 24 data for testing ANN_PR.

3 Partial Discharge Diagnosis at Substation

3.1 Measuring Method

Figure 1 shows layout of three TEV sensors (HVPD) placed on each side of the tank of
72 kV GIS installed in a substation. Those sensors were connected to an oscilloscope
(HVPD LongshotTM) via a coaxial cable, and 100 to 1500 pulses of detected signals
were recorded at a sampling rate of 100 MS/s. Note in Fig. 1 that star symbol indicates
the position where a floating electrode defect shown in Fig. 2 was found by internal
inspection of the GIS after the measurement. The photo shows carbonized trace with
black soot.

Fig. 1. Sensor mounting position for PD measurement in 72 kV tank type GIS.

646 Y. Kakimoto et al.

Fig. 2. Contamination point in the lower part of degradation point inside GIS.

Figure 3 shows signal intensity detected with the three TEV sensors. It is seen in
the figure that the signal acquired with TEV 3 exhibits the largest. Next, further PD
measurement was carried out by placing TEV 4 and 5 on the same tank surface with
TEV 3 as shown in Fig. 4, and at the same time, a high frequency CT (HFCT) was set
to the ground line.

Fig. 3. Comparison of Peak to peak values of three TEV sensors in the GIS in the substation.
 Identification of Partial Discharge Source in Power Apparatus 647

Fig. 4. TEV sensors and HFCT attached on the 72 kV GIS.

Waveform parameters and PRPD pattern were extracted from signals detected with
HFCT, TEV 4, and TEV 5 in the second measurement so as to be input to ANN, and
testing was carried out to identify the signal source in the 72 kV GIS type tank.

3.2 Results and Discussion

Figure 5 shows typical example of PRPD pattern of all the signals acquired with
HFCT, TEV4, and TEV5. Figures 6 and 7 show examples of waveforms acquired by
TEV 4 and TEV 5, respectively. It is found from these figures that waveforms of the
signals detected by the three sensors are similar to those obtained for the simulated
defect POC. It should be also noticed as shown in Fig. 5 that the signals detected by the
three sensors occur synchronously in phase. These results strongly suggest that the
signal detected should be judged to be PD.

Fig. 5. PRPD pattern of signals detected with the HFCT (orange), TEV 4 (pink) and TEV5
(blue) sensors.
648 Y. Kakimoto et al.

Fig. 6. Example of a waveform detected with TEV 4 in the order of ns in the GIS substation.

Fig. 7. Example of a waveform detected with TEV 5 in the order of ns in the GIS substation.

Fig. 8. Examples of POC waveforms measured in the laboratory as training data for ANN_WP.

3.2.1 Signal Waveform Detected by Various Sensors

(i) Waveform detected in the field compared with training data

Figure 8 shows two examples of waveforms obtained for the POC to use as
training data of ANN_WP. As a result, the waveforms given in Figs. 6 and 7
taken in the laboratory were classified as POC by ANN_WP. Note that PD
waveforms acquired in the GIS shown in Figs. 6 and 7 resembles to those for
POC shown in Fig. 8.
(ii) Segment data comparison in field and training data
Figure 9 shows a radar diagram of nine parameters of several signals detected by
TEV4 and TEV5 in the GIS in the field. The definition of each parameter is
 Identification of Partial Discharge Source in Power Apparatus 649

(i) PeakValue defined as peak to peak value, (ii) PhaseInCycle as pulse gener-
ation phase angle, (iii) Risetime as time elapsed for pulse instantaneous value
from 10% to 90% of the peak value, (iv) Falltime as the time interval taken for
the instantaneous signal with 90% to 10% of the peak value, (v) PulseWid as
time from 50% threshold of rising edge to 50% threshold of next falling edge,
(vi) PulseArea: area integration of signal waveform, (vii) EventWidth as time
interval of acquired signal, and (viii) FreqOfMaxAmpl as frequency giving
maximum amplitude, and (ix) Freq (1stMom) as the first frequency determined
by the method of moments. For comparison, Fig. 10 shows a radar diagram of
nine parameters of waveform obtained for POC. Comparing the results shown in
Figs. 9 and 10 indicates that PulseWid is 0 and large values of FreqofmaxAmpl
and PulseArea are likely to be classified as POC. The parameter of PeakValue,
EventWidth, Freq (1st Mom) do not have a certain tendency in training data, so
that it seems that those parameters do not really affect classification.

Fig. 9. Radar diagram of nine parameters of waveform detected with TEV 4 in the field.

Fig. 10. Radar diagram of nine parameters of waveform data for POC in the lab.
650 Y. Kakimoto et al.

(iii) Comparison of PD Source Identification for Data Obtained from Three Kinds of
Sensors by ANN_WP
Figure 11 shows determination results of PD source identification by ANN_WP
when all signals obtained by HFCT, TEV1 and TEV2 in the field are input. As
can be seen from the Figure, the PD source judgement probability of the input
data by TEV 4 and TEV 5 is 72.4% and 46.3%, respectively, while HFCT data
provides the external noise as the judgement probability 90.6%. As shown in
Figs. 6 and 7, waveform is determined to be PD by ANN_WP, while as in
Fig. 11, ANN_WP provides low judgement probability 7.0%, 72.4%, 46.3% as
PD from signal acquired with HFCT, TEV4, TEV5 in 72 kV GIS.

Fig. 11. Judgement result of PD source identification by ANN_WP in inputting all signals
acquired by HFCT, TEV1, and TEV2 in the field.

3.3 PRPD Pattern Acquired by Various Sensors

Figures 12 and 13 show normalized PRPD pattern of approximately 130 signals
detected with TEV 5 and void PRPD pattern used for training data. Note that the PRPD
pattern by TEV 5 in Fig. 12 was determined as Void by ANN_PR. In addition, the
PRPD pattern of the signal acquired by HFCT and TEV 4 was also determined as Void
by ANN_PR. The reasons for being judged to be Void are as follows: (i) the number of
discharges is increased in the first quadrant and the third quadrant, and (ii) the dis-
charge maximum value around 50° and 250° is about 1. Since the signals acquired by
any sensor are classified as Void, this implies that ANN_PR is not affected by sensor
placement. Therefore, it is considered that ANN_PR is better than ANN_WP in which
waveform information changes due to signal propagation and attenuation.
 Identification of Partial Discharge Source in Power Apparatus 651

Fig. 12. Normalized PRPD pattern of the signal acquired with TEV 5 in the field.

Fig. 13. Normalized PRPD pattern of void from training data in the lab.

4 Conclusion

In this report, the identification of the PD signal source was studied for the purpose of
applying ANN constructed to identify the PD signal obtained from various sensors of
72 kV GIS. The signals obtained from the GIS also include those caused by abnor-
malities due to floating electrodes that were confirmed by internal inspection after the
measurement. It was found that the signals caused by the abnormality were mainly
judged as noise and PD by ANN_WP and ANN_PR, respectively. The disagreement
652 Y. Kakimoto et al.

by ANN_WP is attributed to the waveform information changes due to signal propa-

gation and attenuation, while ANN_PR provides better judgement owing to keeping
the phase information at which PD occur. As a result, ANN_PR is found to be more
effective means to judge the PD source in the field.

References
1. Sukma, T.R., Khayam, U., Suwarno, Sugawara, R., Yoshikawa, H., Kozako, M., Hikita, M.,
Eda, O., Otsuka, M., Kaneko, H., Shiina, Y.: Classification of partial discharge sources using
waveform parameters and phase-resolved partial discharge pattern as input for the artificial
neural network. In: International Conference on Condition Monitoring and Diagnosis, no. P09
(2018)
2. Sukma, T.R., Khayam, U., Suwarno, Sugawara, R., Yoshikawa, H., Kozako, M., Hikita, M.,
Eda, O., Otsuka, M., Kaneko, H., Shiina, Y.: Determination of type of partial discharge in
cubicle-type gas insulated switchgear (C-GIS) using artificial neural network. In: International
Conference on Condition Monitoring and Diagnosis, p. 10 (2018)
3. Kozako, M., Sukma, T.R., Yoshikawa, H., Hikita, M., Eda, O., Otsuka, M., Shiina, Y.,
Kaneko, H.: Partial discharge type classification by artificial neural network using waveform
parameters and phase-resolved PD pattern. In: Symposium on Electrical and Electronic
Insulating Materials and Applications in Systems, no. A-4 (2018)
4. Mas’ud, A.A., Albarracin, R., Ardila-Rey, J.A., Muhammad-Sukki, F., Illias, H.A., Bani, N.
A., Munir, A.B.: Artificial neural network application for partial discharge recognition: survey
and future directions. MDPI 9(574), 1–18 (2016)
5. Gulski, E., Krivda, A.: Neural networks as a tool for recognition of partial discharges. IEEE
Trans. Electr. Insul. 28(6), 984–1001 (1993)
6. Suwarno: Diagnosis of High Voltage Equipment using Multi-Parameters, p. 20, Penerbi ITB,
Bandung (2010)
7. Satihs, L., Zaengi, W.S.: Artificial neural networks for recognition of 3-d partial discharge
patterns. IEEE Trans. Dielectr. Electr. Insul. 1(2), 265–275 (1994)
8. Candela, R., Mirelli, G., Schifani, R.: PD recognition by means of statistical and fractal
parameters and a neural network. IEEE Trans. Dielectr. Electr. Insul. 7(1), 87–94 (2000)
9. Mazroua, A.A., Bartnikas, R., Salama, M.M.A.: Neural network system using the multi-layer
perceptron technique for the recognition of PD pulse shapes due to cavities and electrical
trees. IEEE Trans. Power Delivery 10(1) 92–96 (1995)
 Partial Discharge Measurement
and Its High Frequency Characteristics
in Cast Resin Transformer

Tadahiro Fujimura1(&) , Seiya Masuda1, Masahiro Kozako1 ,

Masayuki Hikita1 , Kazuo Iida2, Tokihiro Umemura2,
Yusuke Nakamura3, Tetsuo Nakamae4, Teruhiko Maeda4,
and Masakazu Higashiyama4
1
Kyushu Institute of Technology, 1-1 Sensui-cho,
Kitakyushu-shi, Fukuoka, Japan
o108101t@mail.kyutech.jp
2
Mie University, 1577 Kurima-machiya-cho, Tsu, Mie, Japan
3
Toshiba Infrastructure Systems & Solutions Co.,
1 Toshiba-cho, Fuchu-shi, Tokyo, Japan
4
Toshiba Industrial Products and Systems Co.,
2121 Nao, Asahi-cho, Mie-gun, Mie, Japan

Abstract. Cast resin transformer (CRTr) adopting the solid insulation method
is superior in flame resistance as compared with the transformer adopting other
insulation method. In addition, since it can be made compact and lightweight, it
is used in various industries. However, internal defects such as voids and
delamination are likely to occur in the manufacturing process. Partial discharge
(PD) generated by such defects causes dielectric breakdown and ultimately leads
to equipment failure. So, early detection of defects by PD detection is desired.
Therefore, in this research, to establish an optimal detection method of PD for
CRTr is aimed. PD signal has components in various frequency bands, that
depend on test system and equipment. To understand the frequency component
of PD signal in advance makes it possible to distinguish it from noise and lead to
highly sensitive PD detection. From the above, in this paper, the influence of test
system and sample on frequency characteristics of PD current and electro-
magnetic wave (EMW) is studied for CRTr and artificial void sample. It was
found that PD EMW dominantly has a frequency component depending on test
system and sample as well as PD current in addition to the high frequency
radiated from the defect. Furthermore, since it was found that PD signal dom-
inantly has a component of the tens of MHz band dependent on CRTr itself,
elucidation of the cause was addressed.

Keywords: Partial discharge
Electromagnetic wave
Loop sensor
HFCT

Cast resin transformer

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 653–661, 2020.
https://doi.org/10.1007/978-3-030-31676-1_62
654 T. Fujimura et al.

1 Introduction

Cast resin transformer (CRTr) is widely used in distribution lines. Defects are likely to
occur inside the CRTr, and these defects cause power failure due to dielectric break-
down. Therefore, it is important to detect defects in CRTr at an early stage, but no
diagnostic technique is yet established. In previous reports, diagnostic techniques using
partial discharge (PD) measurement of CRTr have been studied to improve the optimal
defect detection method [1–4]. PD signal consists of various frequency components
from MHz band to GHz band [5]. Diagnostics are also being studied in various fre-
quency bands even in power equipment having windings such as transformers [6–8]. In
these reports, PD is detected with high sensitivity by current transformer (CT) and
electromagnetic wave (EMW) sensor having characteristics in tens of MHz band. It is
also known that the measurement of PD current is affected by test circuit [9, 10]. From
the above viewpoints, this paper deals with investigation on the influence of test system
and sample on the frequency characteristics of PD current and EMW for CRTr as well
as a void defect sample is examined. The influence of a kind of CRTr on the dominant
frequency components of PD signal is also reported.

2 Experimental

A 6.6 kV class transformer with void as defect was used as in the previous paper [4];
void was found to be near the base of the U terminal and between the primary coil and
the secondary coil as the main insulation. Figure 1 shows the internal structure and two
connection types of the 6.6 kV class CRTr. Phase to phase and power distribution
connections simulate the wiring of the shipping test, and the wiring actually used,
respectively. Although the voltage distribution varies, a large electric field is applied to
the defect part in any of the connection methods. Experiments were also carried out
using 33 kV class CRTr, as shown in Fig. 2, with defects in the main insulation
part. The connection type is phase to phase.

(a) Phase to phase connection (b) Power distribution connection

Fig. 1. Connection type of 6.6 kV class CRTr.

 Partial Discharge Measurement and Its High Frequency Characteristics in CRT 655

(a) Appearance (b) Internal structure

Fig. 2. 33 kV class CRTr.

(a) Appearance (b) Model drawing

Fig. 3. Artificial void sample.

Figure 3 shows photo and illustration of artificial void sample (AVS) simulating
internal defects of CRTr. The diameter of the void is 2a = 0.65 mm, and the distance
between the electrodes is D = 3 mm. Voids were molded with epoxy resin together
with electrodes.
Figure 4 shows PD measurement system and sensor arrangement. In order to
clarify the influence of the sample on the PD signal, the wiring length was made as
short as possible. The applied voltage was set to 8 kV and 5 kV for 6.6 kV class CRTr

Fig. 4. PD measurement system and sensor arrangement for 6.6 kV class CRTr and AVS.
656 T. Fujimura et al.

and AVS, respectively. Moreover, PD was generated by X-ray irradiation for both
samples [1]. A self-made loop sensor (LS, diameter:10 mm, 1.5 GHz band) was used
to detect EMW directly emitted from the PD source with high sensitivity in the high
frequency band in the hundreds of MHz [11]. PD signal was measured with an
oscilloscope (Tektronix, DPO 7254, 2.5 GHz, 10 GS/s) using CT (ETS-LINDGREN,
94430-1, 1 MHz - 1 GHz) and LSs. PD EMW was measured simultaneously using two
LSs placed at d = 5 cm and 50 cm. By triggering CT, PD EMW and noise were
distinguished. The amount of charge was about 10–80 pC in any sample.

3 Results and Discussion

3.1 Frequency Characteristics of PD Signal in Artificial Void Sample

(l = 150 Mm)
Figure 5 shows an example of PD signals in AVS simultaneously acquired by CT and
two LSs. It can be seen from the figure that the rising part of PD emitted EMW
dominates the high frequency signal and thereafter the low frequency signal is prop-
agated. Figure 6 shows the average value (the number of measurements: N = 3) of the
frequency spectrum of acquired PD signal. It can be seen from the figure that both CT
and LS have large components around 80 MHz. The LS detected signal also has a large
frequency component of 200 MHz or more.
It was found from the above, the PD signal has a large component at the resonance
frequency determined from the constants of test circuit and sample. From the attenu-
ation characteristics, it was also revealed that PD EMW emitted from the defect is high
frequency.

Fig. 5. Detection waveforms of PD current and EMW in AVS (l = 150 mm).

 Partial Discharge Measurement and Its High Frequency Characteristics in CRT 657

Fig. 6. Frequency spectrum of PD current and EMW in AVS (l = 150 mm).

3.2 Frequency Characteristics of PD Signal in 6.6 KV Class Cast Resin

Transformer (l = 150 Mm)
Figure 7 shows an example of acquired PD signal of 6.6 kV class CRTr with phase to
phase connection. As in case of AVS, the rising edge of PD EMW detected in CRTr
mainly consists of high frequency, and then followed by the low frequency signal.
Figure 8 shows the average value (N = 3) of the frequency spectrum of detected PD
signal. It is evident from the figure that the frequency characteristics of PD signal
detected with CT substantially agree with that by LS. On the other hand, the largest
frequency component in CRTr is the 30 MHz band, which is different from that for
AVS. This is thought to arise from the different inductance and capacitance of the
sample [9]. In addition, frequency components existing in 90 MHz and 150 MHz
bands are considered to be the third and fifth harmonics of the signal in 30 MHz band.
It was found from the above results that the following two EMWs are detected
irrespective of sensor characteristics; these are the high frequency signal radiated
directly from the defect and the low frequency signal caused by test system generated
from PD current., in precise defects location by EMW necessitates to remove noise and
low frequency signal radiated from the ground line and to pay attention to high fre-
quency signals of 200 MHz or more.

Fig. 7. Detection waveform of PD current and EMW in 6.6 kV class CRTr (l = 150 mm).
658 T. Fujimura et al.

Fig. 8. Frequency spectrum of PD current and EMW in 6.6 kV class CRTr (l = 150 mm).

3.3 Change in Frequency Component of PD Signal with Respect to Test

Condition
Figure 9 shows the average value of the frequency spectrum of PD signal (N = 3)
acquired for different wiring length of the circuit (l = 5000 mm) in AVS and 6.6 kV
class CRTr, and for power distribution connection of CRTr. It can be seen in the figure
that increasing the wiring length causes the frequency with the highest intensity of AVS
to shift from 78 MHz to 64 MHz to the low frequency; i.e. the peak appears at 20 MHz
also in CRTr. The results are explained by the resonance frequency calculated from the
lumped circuit constant based on the loop antenna theory [9]. On the other hand, note
that the CRTr has a fixed frequency component mainly in the 30 MHz band which is
not affected by wiring length and connection type.

Fig. 9. Frequency spectrum of CT signal for change of test condition.

3.4 Signal Propagation Characteristics of Class Cast Resin Transformer

Studies have been conducted to clarify the characteristics of signals propagating inside
the transformer using a network analyzer (NA) [12, 13]. Experimental results showed
 Partial Discharge Measurement and Its High Frequency Characteristics in CRT 659

that PD signal of 6.6 kV class CRTr has a dominant component in 30 MHz band. To
elucidate the cause, insertion loss IL (S21) was measured using NA (Keysight Tech-
nologies, N9913A). Port 1 and 2 are connected to the primary and secondary side,
respectively. A coupler was used so as not to be influenced by the ground. IL is defined
by the following formula.

IL ¼ 20 log 10 ðPout = Pin Þ ð1Þ

where: Pin is the intensity of the signal input to the primary side of the CRTr and Pout
is the intensity of the signal transmitted to the secondary side.
Figure 10 shows the comparison of the frequency spectrum of CT signal (Fig. 7)
and IL in the 6.6 kV class CRTr. It can be seen from the figure that IL has the
maximum value in 30 MHz band as well as PD signal. Figure 11 shows the com-
parison result between the frequency spectrum of the CT signal and the IL in the 33 kV
class CRTr. It can be seen also from this figure, the peaks of the PD signal and IL
match at 27 MHz. PD signal has also the dominant component of about 10 MHz. It
seems that is caused by the test system including the test transformer used to generate
high voltage.

Fig. 10. Comparison of frequency spectrum of CT signal and IL in 6.6 kV class CRTr.

Fig. 11. Comparison of frequency spectrum of CT signal and IL in 33 kV class CRTr.

660 T. Fujimura et al.

From the above, it is suggested that the component of the tens of MHz band
dominantly possessing the PD signal depends on the sample and is determined by IL.
This result makes it possible to identify the device generating the PD by understanding
the frequency at which the IL of the device shows the maximum in advance.

4 Conclusion

In this paper, PD EMW were roughly divided into two types due to difference in
frequency characteristics. One is the low frequency signal in the tens of MHz band
radiated from PD current affected by the impedance of sample itself and test circuit.
The other is high frequency signal in the hundreds of MHz band directly radiated from
defect. In addition, it was found that CRTr has unique frequency characteristics that are
not affected by the wiring length of circuit and connection type. It coincides with the
frequency at which the IL becomes maximum. In the future, further investigations of
the signal and examination in the equivalent circuit model will be attempted to apply to
diagnostic technique.

References
1. Inatomi, A., Kozako, M., Hikita, M., Nakamura, Y., Iida, K., Umemura, T., Maeda, T.,
Higashiyama, M.: Effect of X-ray irradiation on partial discharge inception and extinction
characteristics in cast resin transformer. In: Electrical Insulation Conference, vol. 24, no.
4 (2015)
2. Matsuo, K., Inatomi, A., Kozako, M., Hikita, M., Iida, K., Umemura, T., Nakamura, Y.,
Maeda, T., Higashiyama, M., Nakamae, T.: Impulse and subsequent AC partial discharge
properties in cast resin transformer. In: International Conference on Dielectrics, no.
85797 (2016)
3. Maruyama, H., Kozako, M., Hikita, M., Iida, K., Umemura, T., Nakamura, Y., Maeda, T.,
Higashiyama, M., Nakamae, T.: AC partial discharge characteristics and accumulation
charge after lightning impulse in cast resin transformer. In: International Symposium on
High Voltage Engineering, no. 416 (2017)
4. Fujimura, T., Masuda, S., Nakano, Y., Kozako, M., Hikita, M., Iida, K., Umemura, T.,
Nakamura, Y., Maeda, T., Higashiyama, M.: Partial discharge location in cast resin
transformer by multiple small loop sensors. In: International Conference on Condition
Monitoring and Diagnosis, p. 64 (2018)
5. Stone, G.C., Sedding, H.G., Chan, C., Wendel, C.: Comparison of low frequency and high
frequency PD measurements on rotating machine stator windings. In: Electrical Insulation
Conference, vol. 082-S11, pp. 349–352 (2018)
6. Chen, M.-K., Chen, J.-M., Cheng, C.-Y.: Partial discharge detection in 11.4 kV cast resin
power transformer. Trans. Dielectr. Electr. Insul. 23(4), 2223–2231 (2016)
7. Zheng, Z., Han, Y., Sun, M.: Improved design of grounding transformer to facilitate partial
discharge online test. In: International Symposium on High Voltage Engineering, no.
407 (2017)
8. Caprara, A., Cavallini, A., Garagnani, L., Guo, J.: A novel approach for continuous
monitoring of partial discharge phenomena on medium voltage equipments. In: Electrical
Insulation Conference, vol. 115-S19, pp. 495–498 (2018)
 Partial Discharge Measurement and Its High Frequency Characteristics in CRT 661

9. Yoshizumi, H., Nakano, Y., Kozako, M., Hikita, M., Fujii, Y., Nakamura, Y., Cho, H.:
Consideration of generation mechanism of low frequency component of partial discharge
signal detected by transient earth voltage. In: International Conference on Condition
Monitoring and Diagnosis, p. 63 (2018)
10. Heredia, L.C.C., Mor, A.R.: Effects of RLC parameters of a measuring circuit on the
frequency spectrum of partial discharges. In: Electrical Insulation Conference, vol. S7-3
(2018)
11. Mitsudome, H., Akinaga, Y., Matsuo, K., Kozako, M., Hikita, M., Ikeda, Y., Taniguchi, K.,
Nakamura, Y., Okamoto, K.: Basic study on partial discharge location in power module. In:
The First International Conference on Dielectrics, vol. 05a-1, pp. 451–454 (2016)
12. Tran-Anh, T., Auriol, P., Tran-Quoc, T.: High frequency power transformer modeling for
power line communication applications. In: Power Systems Conference and Exposition,
pp. 1069–1074 (2006)
13. Spahiu, M., Partal, H.P.: High frequency modeling and impedance matching of power
transformers for PLC applications. In: Innovative Smart Grid Technologies (2014)
 Propagation Characteristics of Partial
Discharge Radiated Electromagnetic Waves
for Insulation Diagnosis of Oil-Filled Bushing

Tadahiro Fujimura1(&) , Seiya Masuda1, Masahiro Kozako1 ,

Masayuki Hikita1 , Takashi Yamamoto2, Shigemitsu Okabe2,
and Masaaki Nakahata3
1
Kyushu Institute of Technology,
1-1 Sensui, Tobata, Kitakyushu, Fukuoka, Japan
o108101t@mail.kyutech.jp
2
Tokyo Electric Power Company Holdings, Inc., 4-1 Egasaki, Tsurumi,
Yokohama, Kanagawa, Japan
3
TEPCO Power Grid, Inc., 1-5-3, Uchisaiwaimachi, Chiyodaku, Tokyo, Japan

Abstract. Currently, as a failure event due to deterioration of the oil-filled

transformer, the existence of defects due to oil-in bushing has been confirmed.
On the other hand, insulation diagnostic techniques for oil-filled bushings have
not yet been established. The purpose of this research is to develop reliable
insulation diagnostic technique for the oil filled bushing by measuring partial
discharge (PD). Focusing on electromagnetic wave (EMW) radiated by partial
discharges generated in oil-filled bushings, we have studied its propagation
characteristics using finite-difference time-domain (FDTD) analysis software.
PD source can be located from the frequency characteristics of EMW leaking to
the outside of the condenser type oil bushing. Simulation shows that PD source
located at the top of the bushing emits EMW leaking to the outside the bushing
as if the bushing acts as a coaxial cylindrical waveguide consisting of center
conductor, capacitor foil and porcelain insulator. On the other hand, PD source
at the bottom of the bushing causes the resonance frequency to appear as a
coaxial cylindrical waveguide inside the bushing, whereas the resonant fre-
quency appears strongly in the rectangular cavity resonance arising from the
tank size. The above results suggest that PD signal and external noise can be
distinguished by estimating the frequency maximum of the EMW leaking
strongly to the outside.

Keywords: Partial discharge
Insulation diagnosis
Oil-filled bushing

Condenser-type bushing
Electromagnetic wave propagation
Finite-difference
time-domain

1 Introduction

It is very important to grasp the signs of failure of oil-filled transformers to prevent

equipment failure beforehand. Partial discharge measurement such as ultrasonic
detection has been proposed as one of diagnostic methods in transformer, while
external diagnostic technology for oil-filled bushing has not been established yet [1–5].
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 662–672, 2020.
https://doi.org/10.1007/978-3-030-31676-1_63
 Propagation Characteristics of Partial Discharge Radiated Electromagnetic Waves 663

In this research, we aim to establish insulation diagnosis of oil-filled bushing

without stopping transformer operation. Radiated electromagnetic wave (EMW) signal
by partial discharge (PD) generated in the bushing is detected by an EMW sensor
placed outside the bushing. We investigate methods for insulation diagnosis which
enables us to identify noise or PD from the signal detected by the EMW sensor.
Furthermore, to identify whether PD occurs inside the bushing or from the transformer,
it is important to understand the EMW propagation characteristics inside the trans-
former as well as the bushing. From the above viewpoints, we have been studying the
propagation characteristics of PD radiated EMW generated within the bushing using
finite difference time domain (FDTD) analysis.
This report deals with investigation on PD emitted EMW propagation for two kinds
of models simulating a bushing; (1) simple model; small tank size, consisting of a
single capacitor foil, and (2) practical model; tank size closer to actual unit, simulating
four layers of capacitor foil. The simplified model is used to elucidate numerically the
mechanism of PD radiated EMW propagation in oil-in bushing. On the other hand, the
practical model provides a diagnostic method by identifying frequency components
with relatively larger magnitude propagating to the outside of the bushing with the
same propagation principle revealed by the simplified model. Results of examining
each analysis model will be also discussed.

2 Analysis Method

Figure 1(a) and (b) show three-dimensional structure diagram and the sectional view of
the simplified model, respectively. This simplified model simulates a capacitor foil
inside a bushing with a cubic type tank of 500 mm side. Note that this model simulates
a capacitor foil such as carbon or metal foil applied on the surface of oil impregnated
paper in practical equipment. The surface of the metal cylinder simulates the capacitor
foil assumed as a perfect electric conductor (PEC) which has a conductivity of infinity,
and its thickness set to 0.
Figure 2(a) and (b) show three-dimensional structure diagram and the cross-
sectional view of the practical bushing model, respectively. This figure provides the
basis of actual equipment with the cubic type tank of 2000 mm side and the thickness
20 mm, and the PEC surface. The bushing has a cylindrical structure with an outer
diameter of 125 mm and a length of 1620 mm. This bushing is covered with a ceramic
having relative permittivity er = 6.5 and thickness of 8 mm. The center conductor as
PEC has a diameter of 40 mm and a length of 500 mm. A disk as PEC with a thickness
of 8 mm and a diameter of 200 mm was placed on the upper end of the bushing. This
disk simulates the bottom surface of a tank with er = 2 for supplying insulating oil to
the bushing [6]. The practical model consists of the capacitor foil with four layers.
664 T. Fujimura et al.

(a) 3D model (b) Sectional view

Fig. 1. Simple model of condenser type oil-filled bushing.

(a) 3D model (b) Sectional view

Fig. 2. Basic analytical model of practical condenser type oil-filled bushing.

An excitation source located in the lower part of the bushing simulates insulation
deterioration due to moisture intrusion into the bushing to emit EMW. Here, the
excitation source was placed between the first and second layer of the capacitor foil
under the bushing.
PD radiated EMW propagation was analyzed by FDTD software (MAGNA /TDM
Ver 8.3, Itochu Techno-Solutions Co., Ltd.). A Gaussian pulse current with a crest
value of 1 A and a resonance frequency of 1 GHz was injected into the excitation
source in the X direction of one mesh. The wall of the analysis area was taken as
perfect matched layer (PML). In addition, the electric field component in the X
direction was extracted by Check Point (CP) placed inside and outside of the bushing.
 Propagation Characteristics of Partial Discharge Radiated Electromagnetic Waves 665

3 Elucidation of the Principle of Electromagnetic Wave

Propagation in Simple Model

Figure 3 shows calculated electric field distribution of PD emitted EMW signal in the
simple model at time t = 4.3 ns after the excitation. It can be seen in the figure that
EWM propagates between the center conductor and the capacitor foil ①, between the
capacitor foil and the porcelain insulator ②, and inside the tank ③. Let Cc - Mf
represent between the center conductor and the Metal (capacitor) foil, Mf - Ins between
the metal foils and the insulation tube, and Cc - Ins between the center conductor and
the insulation tube. Figure 4 shows the frequency spectra of the electric field intensity
waveform of PD radiated EMW at each CP in the simple model. As can be seen in the
figure, maximum of the signal intensity at CP 3 appears around 648 MHz.

Fig. 3. Calculated electric field distribution of PD emitted EMW signal in the simple model at
time t = 4.3 ns after the excitation (simple model).

Fig. 4. Frequency spectrum of electric field intensity waveform at various check points in the
simple model.
666 T. Fujimura et al.

When the bushing part (Cc - Mf, Mf - Ins, Cc - Ins) is assumed to be a coaxial
cylindrical waveguide, the resonance frequency is expressed by Eq. (1).
c0 m
f ðTEmn Þ ¼ pffiffi ðn ¼ 1Þ ð1Þ
pða þ bÞ e

where a and b are radii between the center conductor and the capacitor foil, and the
insulation tube, respectively, c0 is speed of light 3.0  108 m/s, and e is the dielectric
constant of insulating oil with er = 2.2, and m and n are integers ≧0. Equation (1)
provides the resonance frequency in TE11 mode as fCc - Mf11 = 920 MHz for Cc - Mf,
fCc - Ins11 = 537 MHz for Cc - Ins and fMf - Ins11 = 429 MHz for Mf - Ins. It should be
noticed in Fig. 4 that fCc - Mf11, fCc - It11, and fMf - It11 calculated by Eq. (1) agree with
the frequencies exhibiting the maxima of the signal intensity. On the contrary, the
resonance frequency by Eq. (1) does not match the frequency around 648 MHz outside
the bushing. This indicates that the EMW does not receive much resonance at the
bushing which is considered as a coaxial waveguide. The resonance frequency when
the tank is assumed to be a rectangular cavity resonator is expressed by Eq. (2).
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c0 m2  n 2  p 2
fmnp ¼ pffiffiffiffi þ þ ð2Þ
er 2l 2l 2l

Calculation with the tank size l = 500 mm from Eq. (2) provides f013 = 661 MHz
for m = 0, n = 1, and p = 3 mode. This calculated value roughly agrees with 648 MHz
which dominantly propagates to the outside of the bushing. It is suggested that the
resonant frequency due to the tank size strongly propagates in the EMW that leaks
strongly to the outside of the bushing.
The simplified model consists of a condenser type oil-filled bushing including one
capacitor foil with the small size tank. Numerically, the propagation principle of the PD
radiated EWM was clarified. In the simplified model, it was shown that the EMW
detected outside was more likely to be affected by the tank than the bushing.

4 Results and Discussion in Practical Model

4.1 Electromagnetic Field Propagation Outside the Bushing

Figure 5 shows frequency spectrum of EMW electric field component at CP (1000 mm
from the center) arranged outside the tank with bushing, the bushing unit only, and the
tank unit only. It is evident from Fig. 5 that frequency components around 700 to
800 MHz are prominently leaked to the outside regardless of presence or absence of
tank (dotted area in Fig. 5). Note that the basic model as named Tank with bushing in
the figure shows the same frequency spectrum as the Bushing unit only. Contrary to
this result, in the simplified model in the previous section, the resonance frequency due
to the tank size was strongly observed outside the bushing. The difference is that the
practical model has larger tank length of 2000 mm than the simplified model.
 Propagation Characteristics of Partial Discharge Radiated Electromagnetic Waves 667

Therefore, the resonance frequency due to the tank size does not propagate strongly, so
that it can be interpreted as being not observable externally.
Figure 6 shows calculated electric field distribution of PD radiated EMW in the
practical equipment model at t = 10 ns. It can be seen in the figure that the radiated
EMW travels between the capacitor foils and at the same time propagates inside the
tank. Calculated result provides the electric field strength between inside and outside of
the bushing 2.5 times larger than that inside the tank. The result indicates that the
resonance frequency dependent on the tank size is not observed outside at the practical
model.

Fig. 5. Frequency characteristics of electric field component under each condition.

Fig. 6. Calculated electric field distribution of PD radiated EMW in the practical equipment
model at t = 10 ns (practical model).
668 T. Fujimura et al.

4.2 Electromagnetic Field Propagation Inside the Bushing

Figure 7 illustrates more detailed bushing model with the tank simulating only the
opening part and CP placed between each capacitor foil.
Figure 8 shows calculated electric field intensity E and the frequency fmax providing
maximum intensity at CP4 as a function of bushing length L. As can be seen in Fig. 8,
fmax is almost independent of L, whereas E decreases with increase in L. The decrease
in E with L is explained in terms of EMW attenuation caused by increase in the
distance from the check point to the opening part of the tank at which PD source is
located. This result suggests that dominant frequency component leaking to the outside
of the bushing is independent of L.

Fig. 7. Bushing size (L = 3220 mm, W = 250 mm).

Fig. 8. Relation among bushing length L and electric field intensity at peak frequency.
 Propagation Characteristics of Partial Discharge Radiated Electromagnetic Waves 669

Figure 9 shows frequency fmax giving maximum intensity calculated by Eq. (1) as a
function of bushing size, i.e. the bushing diameter W. Triangular symbols in the figure
represent result analyzed by FDTD for W = 125 mm (Wa) and 250 mm (Wb) with
changing the size between the capacitor foils according to the difference in W. Note that
the first and second layer of the capacitor foils having excitation sources adjacent to
each other forms coaxial cylindrical waveguide. It is confirmed that the resonance
frequency fr obtained from Eq. (1) agrees with fmax irrespective of L. It means that the
two layers of capacitor foils behave as a coaxial cylindrical waveguide with its reso-
nance frequency observed outside the bushing.
Figure 10 shows frequency spectrum of the electric field intensity waveform cal-
culated at each CP. Equation (1) provides fr as f1-2 = 379 MHz between the first and
second layer, f2-3 = 280 MHz between the second and third layer, and f3-4 = 222 MHz
between the third and fourth layer. It follows that each fr agrees with the frequencies
fmax giving maxima as shown in Fig. 10. Note that the maximam intensity appears at
fmax = f1-2 at CP4, which is the most likely to leak to the outside of the bushing, and the
same fmax also appears at other CPs. The result indicates that EMW propagating from
the upper part between the first and second layer continues to propagate between the
other capacitor foils, and then to the outside of the bushing.
As a consequence, it was found that the difference in size between the capacitor
foils allows us to estimate specific resonance frequencies. This estimation will provide
effective means for distinguishing the specific frequencies produced by PD radiated
EMW propagation in the bushing from that of external noises.
In the next section, based on the above results, we propose a novel diagnosis
method from outside a bushing in the condenser type oil-filled one.

Fig. 9. Analysis values and estimated values for each capacitor foil inter-size.
670 T. Fujimura et al.

Fig. 10. Frequency spectrum of electric field intensity waveform at various CPs.

4.3 Proposal of a Novel Diagnostic Method in Oil-Filled Bushing

Let us compare PD signal generated inside the bushing with external noise. Figure 11
(a) and (b) show calculated waveform at CP for an excitation source placed, as shown
in Fig. 2, at the lower part and the outside of the bushing, respectively. It is obvious
from Fig. 11(a) that the waveform is characterized as high frequency vibration arising
from EMW propagation inside the bushing. On the other hand, the waveform in Fig. 11
(b) exhibits even less vibration due to the signal propagating as a spherical wave.
Based on the above consideration, a novel diagnostic method using PD EMW
detection is proposed in Fig. 12. First, EMW signal is measured with an external
antenna. Second, compare a frequency giving dominant intensity of the signal with the
resonance frequency propagating as the coaxial cylindrical waveguide between the
capacitor foils. In case where no relation between the two specific frequencies, judge it
as external noise. In case where the dominant frequency in the detected EMW agrees
with the resonance frequencies provided by a given condenser layer foils, judge as high

(a) Under bushing (b) Outside of bushing

Fig. 11. Calculated waveform at CP for an excitation source placed, as shown in Fig. 2 at the
lower part and the outside of the bushing, respectively.
 Propagation Characteristics of Partial Discharge Radiated Electromagnetic Waves 671

possibility of PD located in the lower part of the bushing. If the specific frequencies do
not agree with calculated resonance frequencies, one can judge as PD generated in the
transformer tank.

Fig. 12. Proposal of a novel diagnostic method for condenser type oil-filled bushing.

5 Conclusion

This paper presented investigation on propagation of electromagnetic wave caused by

simulated PD source in the condenser type bushing. It was found that the bushing can
be regarded as a coaxial cylinder wave guide through which electromagnetic wave
propagates with the resonance frequencies determined by the size between the capacitor
foils. As a result, a novel diagnostic method is proposed using the specific frequency
characteristics of PD emitted electromagnetic wave detected from the outside of the
bushing.

References
1. Hikita, M., Ohtsuka, S., Teshima, T., Okabe, S., Kaneko, S.: Electromagnetic (EM) wave
characteristics in GIS and measuring the EM wave leakage at the spacer aperture for partial
discharge diagnosis. IEEE Trans. Dielectr. Electr. Insul. 14, 453–460 (2007)
2. Sriyono, Kim, Y.-J., Khayam, U., Harjo, S., Hikita, M.: Characteristics of external loop
sensor located near bushing on partial discharge induced electromagnetic wave measurement.
Int. J. Electr. Eng. Inf. 5(1), 320–323 (2013)
3. Okabe, S., Ueta, G., Hama, H., Ito, T., Hikita, M., Okubo, H.: New aspects of UHF PD
diagnostics on gas-insulated systems. IEEE Trans. Dielectr. Electr. Insul. 21, 2245–2258
(2014)
4. Badicu, L.V., Broniecki, U., Koltunowicz, W., Subocz, J., Zenker, M., Mrozik, A.: Detection
of bushing insulation defects by diagnostic monitoring. In: Electrical Insulation Conference,
International Conference on Condition Monitoring and Diagnosis, vol. A2-4 (2016)
672 T. Fujimura et al.

5. Wang, H., Cheng, J., Ding, D., Liu, W., He, L.: A feasibility study on bushing tap UHF
sensing method for partial discharge in transformer. In: International Conference on Condition
Monitoring and Diagnosis, no. 17 (2018)
6. Liao, R., Hao, J., Chen, G., Ma, Z., Yang, L.: A comparative study of physicochemical,
dielectric and thermal properties of pressboard insulation impregnated with natural ester and
mineral oil. IEEE Trans. Dielectr. Electr. Insul. 18(5), 1626–1637 (2011)
 Diagnostics for Vacuum Degree of Vacuum
Interrupter Based on Partial Discharge
Mechanisms Under Medium Vacuum Level

Yusuke Nakano1(&) , Masahiro Kozako1 , Masayuki Hikita1 ,

Tsuyoshi Tanaka2 , and Masato Kobayashi2
1
Kyushu Institute of Technology, 1-1 Sensui-cho, Tobata-ku,
Kitakyushu 804-8550, Japan
yusuke_nakano.3513@live.com
2
1450, Nishihara, Tarazaki, Hitachinaka 312-0003, Japan

Abstract. A loss of vacuum in vacuum Interrupter (VI) can lead to a crucial

failure in vacuum circuit breaker (VCB). The purpose of present research is to
develop the continuous monitoring and diagnostic technique for degradation of
vacuum degree in VIs. The loss of vacuum originates from a natural slow
leakage, corrosion at flange face, mechanical damages at bellows or flanges, etc.
during long-term operations. The partial discharge (PD) phenomena, occurring
in the VI under lower vacuum level from 0.1 Pa to 100 Pa, have been focused
on, and its characteristics and mechanisms have been discussed. A glass VI with
close contact state was used as a specimen, whose vacuum degree could arbi-
trarily be kept by connecting vacuum pumping systems. PDs inside the VI could
be detected with a detecting resistor connecting to a plane-round electrode
located 50 mm from a vapor shield of VI. Applied voltage and electrical
potential of the floating vapor shield were also measured with a voltage divider,
a capacitive voltage divider directly connecting to the vapor shield of VI.
Furthermore, to determine the PD location in the VI, framing images light
emission of PD were measured by digital camera with image intensifier, whose
exposure time and its timing were controlled by gate pulse from a pulse gen-
erator. The present study found out the characteristics of first PD, which can be
measured in degradation process of VI. The pressure range corresponds to the
Paschen left region. The differences in PD polarities were clarified through
above measurement, and temporal change of charge potential of vapor shield
was calculated. From the above results and discussion, the applicability of PD
detecting for vacuum degree deterioration to operating VCBs in the field and its
method including sensing and signal processing were proposed.

Keywords: Vacuum circuit breaker
Vacuum degree deterioration

Diagnosis
Partial discharge

1 Introduction

Vacuum circuit breakers (VCBs) have been widely operated in distribution system for
more than half a century. Their reliability and performance records of the vacuum
switching technology are outstanding in the medium voltage range up to 52 kV [1].
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 673–683, 2020.
https://doi.org/10.1007/978-3-030-31676-1_64
674 Y. Nakano et al.

Nowadays, user interest is focusing on the monitoring tools to estimate the residual
lifetime of medium voltage VCB, installed in the field in excess of 30 years [2, 3].
Generally, the VCB provides very long electrical life due to low arcing voltage, arc
rotation and low erosion of the contacts. Furthermore, the VCB designed for medium
voltage applications has low contact strokes in the range of 10 mm and low moving
energy in the kg-range, which enables them to operate more than 10,000 operations.
In rare cases where VCB fails, it is normally related to a loss of vacuum in vacuum
interrupter (VI). In the unlikely event of vacuum leakage, its origin can be divided into
four causes [4]:
1. During manufacturing process (but detectable) with leakage rates of >10−2 Pa/s
2. Mechanical damages at bellows or flanges with leakage rates of >10−2 Pa/s
3. Corrosion at flange face with leakage rates of >10−3 Pa/s
4. Flashover outside and through the insulators: leakage rates of >10−1 Pa/s. This
would lead to loss of vacuum in a day
For the early detection of loss of vacuum and its continuous monitoring during
operation, the present study focuses on the partial discharge (PD) phenomena inside VI,
occurring with the increase in the pressure inside the VI. This is because the loss of
vacuum can lead to the decrease in discharge voltage according to Paschen’s law [5]. In
the previous research, PD characteristics in the medium vacuum region from 1 Pa to
1000 Pa order (generally, “medium vacuum” is defined as 0.1 Pa to 100 Pa), whose
vacuum range corresponds to the left region and vicinity of Paschen minimum, was
discussed [6].
The present study found out the characteristics of first PD, which can be measured
in degradation process of VI through the measurement of PD current, shield potential,
which is newly measured with the capacitive divider directly connected to the vapor
shield, and light emission image of PD. The differences in PD polarities are clarified
through above measurement, and temporal change of charge potential of vapor shield is
calculated. From the above results and discussion, the applicability of PD detecting for
vacuum degree deterioration to operating VCBs in the field and its method including
sensing and signal processing are proposed.

2 Experimental Setup

2.1 Experimental Circuit with Vacuum System

Figure 1 shows the overview of glass VI used. Figure 2 shows the constructed
experimental setup with glass VI, connected with vacuum systems. A vacuum pressure
in the VI can arbitrarily be kept through an insulating envelope and several gate valves,
which is measured with a Pirani gauge and an ionization gauge. Main contact of VI was
closed as a normal operation. PDs inside the VI were generated by applying AC
voltage (60 Hz) through a transformer, which was measured with a voltage divider.
 Diagnostics for Vacuum Degree of Vacuum Interrupter 675

Fixed Vapor Movable

electrode shield electrode

Metal
Glass bellows
insulator

Fig. 1. Overview of glass VI, whose contact is closed.

Vacuum Glass VI (closed

system contact) Vapor shield

HV
Applied PD Shield
source
voltage current potential
(60 Hz)

Fig. 2. Experimental setup and measurement system.

2.2 PD Current Measurement

PD current was measured with a detecting resistor RD of 1 kΩ connected to a plane
electrode, which was located 50 mm below the VI.

2.3 Shield Potential Measurement

The shield potential of vapor shield, which is normally a floating potential, is deter-
mined by the capacitance CC-S between conductor and vapor shield, and parasitic
capacitance CS-G between vapor shield and surrounding grounded electrode. Capaci-
tances of insulator should also be considered, but here it is assumed to already be
included in CC-S.
Figure 3 shows the equivalent capacitive circuit of VI. In the present study, the
vapor shield potential was controlled and measured by connecting the additional
capacitances C1 and C2 of 25.4 pF and 12.1 nF as a capacitive divider. From the
capacitive ratio, the shield potential was kept to 47.6% of applied voltage without
discharges. Here, a conventional HV probe including resistance component cannot be
used for the measurement of shield potential, because its charge accumulated in the
vapor shield by PDs would be released through the resistance with time constant less
than 1 ms.
676 Y. Nakano et al.

Applied voltage Va

CC-S
(31.5 pF)

CS-G C1
(6.5 pF) (25.4 pF)

Shield
PD current
potential
RD C2
(1 kΩ) (12.1 nF)

Fig. 3. Equivalent capacitive circuit and measurement system.

2.4 Measurement of Light Emission of Partial Discharge

To determine the location of a single PD pulse inside the VI, the light emission images
of PDs were measured by a still camera with an image intensifier (I.I.). Exposure time
and its timing of light emission image were arbitrarily controlled by gate pulse signal
by pulse generator.

3 Experimental Results and Discussion

3.1 PD Current Waveform and PD Light Emission
Figure 4 shows measured waveforms of applied voltage, shield potential and PD
current under vacuum pressure of 3.7 Pa, and applied voltage of 13.2 kVrms. In this
case, the first PD occurred with negative polarity. Before the first negative PD, the
shield potential fluctuated only by the capacitive division. After the subsequent PDs,
the shield potential was changed with PD current pulses. Figure 5 shows the light
emission image of PD inside the VI with exposure time of 16.6 ms of I.I. with gate
pulse control, which corresponds to the band of Fig. 4. The intense light emission on
glass surface inside the VI can be seen. Therefore, measurable PD in this VI would be
surface discharge along the insulator surface.
 Diagnostics for Vacuum Degree of Vacuum Interrupter 677

Exposure of I.I.
Applied voltage (Fig. 5)
25
20
15
10
5
0
-5
-10
-15
-20
-25
Shield potential

25
20
15
10
5
0
-5
-10
-15
-20
-25

Fig. 4. Applied voltage, shield potential and PD current waveforms under pressure of 3.7 Pa
inside the VI.

Fig. 5. Light emission image of PD inside the VI of 1 cycle AC voltage through image
intensifier (I.I.) with gate pulse control.

3.2 Characteristics of Positive and Negative PD Pulses

Figure 6(a) and (b) show the shield potential and PD current waveforms of positive and
negative PDs. These waveforms in Fig. 6 were measured with higher time resolution
than those in Fig. 4. The difference in rise of PD current pulse can be seen. In both
polarities, as the result of pulsed PD, the potential difference between conductor and
vapor shield was polarity reversed. The vapor shield was completely charged when PD
678 Y. Nakano et al.

current reached its peak. In particular, in negative PD, gradual increase in PD current
and shield potential can be seen. Further measurement with higher resolution in vertical
axis, precursor current before PD pulse can also be seen as shown in Fig. 6(c).
25 25
20 20
15 15
10 10 Applied voltage
5 5
0 Shield potential 0
-5 -5
-10 -10
-15 Applied voltage -15 Shield potential
-20 -20
-25 -25

25 25
20 20
15 15
10 10
5 5
0 0
-5 -5
-10 -10
-15 -15
-20 -20
-25 -25

(a) positive PD (b) negative PD

10
Shield potential
5

0
Applied voltage
-5

-10

0.5

-0.5

-1.0

-1.5

-2.0

(c) negative PD with more sensitive measurement

Fig. 6. Applied voltage, shield potential and PD current waveforms for positive and negative
polarities.
 Diagnostics for Vacuum Degree of Vacuum Interrupter 679

Figure 7(a) and (b) also show the light emission images of positive PD pulse and
negative PD pulse, captured in the whole time range shown in Fig. 6(a) and (b). In both
polarities, PDs occur along insulator surface. In Fig. 7(a), intense light emission can be
seen below the shield edge, which extends toward high voltage shield ring on left-side.
On the contrary, in Fig. 7(b), light emissions concentrate on insulator surface and near
shield ring. From the polarity of PD current and light emission image, PD mechanism
and its model would be discussed in next section.

(a) positive PD (b) negative PD

Fig. 7. Light emission images of PD inside the VI for positive and negative polarities.

4 Temporal Change of Shield Potential of Vapor Shield

by Partial Discharge

As is mentioned above, PD can occur between conductor and vapor shield. In this
section, firstly, the relationship between PD and vapor shield potential would be dis-
cussed. Then, PD mechanism inside the VI along insulator surface is also discussed.

4.1 Temporal Change of Shield Potential of Vapor Shield

Figure 8 shows the temporal change of shield potential of vapor shield by partial
discharge with the applied voltage of 15.4 kVrms, the internal pressure of 3.7 Pa. The
first PD occurred at positive polarity. The displacement shield potential without any
discharge, determined by capacitive division, is also shown in Fig. 8. The charge
potential of vapor shield was calculated by subtracting displacement component from
measured shield potential.
The vapor shield is charged up to about 10 kV by PD current in both positive and
negative polarities. It corresponds to the PD charge amount of several tens nC. The
attenuation of charge potential after PD pulse can be explained by the resistance
component e.g. insulator surface and residual plasma in space generated by discharges
in contact with the vapor shield. From the estimated time constant, the total resistance
would be several MΩ * several tens MΩ. In attenuation region, no discharges could
occur, since there are no light emissions captured by digital camera. However, the
discharge current of attenuation cannot be measured because the current value may be
quite small.
680 Y. Nakano et al.

Unlike the attenuation of charge potential, the time region in which the amount of
charge potential increase can be seen in a few cycles in Fig. 8. Thereby, the small
current flows between conductor and vapor shield before and after PD current.

Applied voltage
25
20
15
10
5
0
-5
-10
-15
-20
-25
Shield potential
25
20
15
10
5
0
-5
-10
-15
-20
-25

25
20
15
10
5
0
-5
-10 Charge potential
-15
= Shield potential − Displacement
-20
-25

Fig. 8. Temporal change of charge potential by PD current in VI.

4.2 PD Current and Change of Charge Potential of Vapor Shield

Figure 9 shows the relationship between PD current and change of shield potential due
to PD pulse. As shown in Fig. 8, same procedures for the calculation of charge
potential were applied for 10 sample waveforms. PD inception voltage was different in
every voltage application from 12 kVrms to 17 kVrms.
 Diagnostics for Vacuum Degree of Vacuum Interrupter 681

Positive PDs have linear characteristic between two parameters. Compared to

positive PDs, negative PDs have larger amount, whose dispersion can be explained by
the difference in PD occurrence phase and electron multiplication process.

25

20

15

10

5 positive PD
negative PD

0
0 5 10 15 20 25

Fig. 9. Relationship between PD current and change of shield potential due to PD.

4.3 PD Mechanism Inside the VI Along Insulator Surface

From above results, the PD mechanism inside the VI for each polarity would be
discussed. Figure 10(a) and (b) shows the model of PD along insulator surface inside
the VI.
For the positive PD, it can begin with the electron emission from the vapor shield.
Since the charge accumulated in the shield is released, the shield potential decreases
with the electron emission from the vapor shield, which results in good linearity
between PD current and change of shield potential as shown in Fig. 9. For the negative
PD, it can begin with the electron emission from the conductor with applied voltage of
negative polarity. The discharge propagates along the insulator surface toward the
vapor shield with the electron multiplication because of the electrons remaining on the
insulator surface due to positive PDs, which results in the measured PD current and its
dispersion as shown in Fig. 9. The current in this process can be measured as the
gradual increase in PD current. In both polarities, the small current after PD pulse was
measured, which could be explained by the movement of remained ions in the space
and insulator surface.
682 Y. Nakano et al.

− Electron

HV electrode Shield HV electrode Shield

− −

− − −− −− − − −− −−−
Glass insulator Glass insulator

(a) positive PD (b) negative PD

Fig. 10. Schematic model of positive and negative PD along insulator surface in VI.

5 Diagnostics of Vacuum Degradation by Detecting Partial

Discharge

From above results and discussion through PD measurement with vacuum degradation
inside the VI, a few of diagnostic techniques would be suggested. As mentioned in
previous sections, the current and the change of shield potential can be measured, when
PD occurs inside the VI. Compared to other PDs or noise in pC or tens pC order, PD
charge inside the VI is enough large to detect its PD signal. As measured in the present
paper, it is possible to detect the PD current by using plane sensor connecting detecting
resistor. The sensor with a coupling capacitor would also be applicable to measure the
superimposed charge on shield potential.

6 Conclusion

In the present study, for the development of degradation of vacuum degree inside the
VI, PD phenomena occurring in the VI with the increase in the internal pressure were
focused on. As a new measurement, the shield potential was measured directly con-
nected with the capacitive divider. From the shield potential measurement and light
emission images of PD inside the VI, the PD mechanisms for each polarity were
discussed. Furthermore, a few of diagnostic techniques of degradation of vacuum
degree inside the VI were suggested.

References
1. Falkingham, L.T., Reeves, R.: Vacuum life assessment of a sample of long service vacuum
interrupters. In: 20th CIRED, paper 0705 (2009)
2. Falkingham, L.T.: A study of vacuum levels in a sample of long service vacuum interrupters.
In: 25th ISDEIV, pp. 181–184 (2012)
3. Smeets, R., et al.: The impact of the application of vacuum switchgear at transmission
voltages, CIGRE WG A3.27, A3-14(SC) 02 IWD (2014)
 Diagnostics for Vacuum Degree of Vacuum Interrupter 683

4. Renz, R., Gentsch, D., Slade, P., Fink, H., Schlaug, M.: Vacuum interrupters – sealed for life.
In: 19th CIRED, Paper 0156 (2007)
5. Schellkens, H.: Continuous vacuum monitoring in vacuum circuit breakers. In: 26th ISDEIV,
pp. 465–468 (2014)
6. Nakano, Y., Kozako, M., Hikita, M., Tanaka, T., Kobayashi, M.: Discharge characteristics
and mechanisms under medium vacuum region in vacuum interrupter. In: IWHV 2018, ED-
18-092/SP-18-045/HV-18-127 (2018)
 Research on Energy Harvesting
and Its Application in Thermal Condition
Monitoring of Three-Core Power Cables

T. X. Sun1(&), X. F. Zeng2, Y. X. Lu3, and H. J. Li3

1
Zhuhai Power Supply Bureau, Guangdong Power Grid Corporation, No.296,
Cuixiang Road, Xiangzhou District, Zhuhai, Guangdong Province, China
1912843157@qq.com
2
State Grid Changsha Power Supply Company, No. 443, Baisha Road,
Tianxin District, Changsha, Hunan, China
3
Xi’an Jiaotong University,
No. 28, Xianning West Road, Beilin District, Xi’an, Shaanxi, China

Abstract. It is of great significance to monitor the thermal condition of three-

core power cables in order to ensure the secure and stable operation of power
system. Nevertheless, the power supply of thermal condition monitoring system
terminal of three-core cables remains to be solved. A three-core power cables
energy harvesting method was proposed by analyzing the magnetic field dis-
tribution around the three-core cable surface. Then a three-core cable energy
harvesting and temperature measuring device was designed and produced. The
three-core cable harvesting model was built by Ansys Maxwell, and the impact
of different factors on the performance of energy harvesting was analyzed. The
cable core temperature in steady state and transient state was calculated using
finite element method based on the load current obtained by induced voltage of
energy harvesting coils and the cable surface temperature measured by tem-
perature measuring device. Results obtained by simulation were very close to
the measured ones, which indicated the thermal condition monitoring system of
three-core cables based on energy harvesting method was effective.

Keywords: Energy harvesting
Three-core cable
On-line monitoring

Temperature field
Finite element method

1 Introduction

Cross-linked polyethylene (XLPE) cables are increasingly used in distribution net-

works due to their excellent insulation, heat resistance and mechanical strength. When
the cable runs for a long time under the condition of high voltage and large current, its
temperature will rise. And the temperature of the conductor of the cable core can
basically reflect its running condition. If the continuous heating of the core cannot be
timely discovered and dealt with, it may lead to the occurrence of extreme accidents
such as fire and even explosion, which will greatly affect the safe operation of the
power system and cause very large economic losses and serious social impact [1].
Therefore, many scholars have carried out a large number of studies on XLPE cable

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 684–696, 2020.
https://doi.org/10.1007/978-3-030-31676-1_65
Research on Energy Harvesting and Its Application in Thermal Condition Monitoring 685

core temperature detection, developed a lot of detection equipment, which has also
been applied in the power grid [2, 3]. However, most of the cable temperature mea-
suring and testing equipment is installed on the cable in a distributed way, and the
middle part of the cable is a fully enclosed structure, which cannot be connected with
an invasive energy collecting device. Therefore, the power supply problem of the
temperature measuring device has become a major problem limiting the temperature
measuring device of XLPE cable.
In view of the power supply problem of XLPE cable temperature detection system,
the main solutions are as follows: 1. The distributed optical fiber temperature mea-
suring device is adopted to avoid the power supply problem, but its cost is high. And
the 10 kV XLPE distribution cable is mostly directly buried, so it is difficult to install
the optical fiber temperature measuring device on the installed XLPE cable. 2. Lithium
battery power supply, which is the most widely used method at present, can solve the
power supply problem of short-term temperature measuring device, but with the
increase of service life, the battery needs to be replaced regularly, which brings great
inconvenience to site staff. 3. Non-invasive energy collecting equipment is adopted to
collect electricity on cables, which is not difficult to install and can provide permanent
power supply, and can be used as an effective power supply mode for temperature
measuring system.
The research content of this paper belongs to the third power supply mode men-
tioned above. At present, literature [4] has carried out the research on the non-invasive
power supply mode of XLPE single-core cable, and the energy collection device
obtained can effectively power the temperature measuring system. However, there is no
effective way to extract energy for three-core XLPE cable. To solve this problem, this
paper proposes an effective energy collection method for three-core cable.
The research content of this paper is as follows: the second section theoretically
analyses the magnetic field distribution characteristics around the three-core XLPE
cable, proposes the three-split coils around the three-core cable to obtain energy. And
the finite element analysis software Ansys Maxwell [5] is used to calculate the voltage
distribution of the three-split coil [6, 7]. The third section describes in detail the design
of the device of the energy collecting coil. The fourth section describes the validity
verification experiment of the device under laboratory conditions.

2 Principle of Three-Core Cable Energy Harvesting

2.1 Theoretical Analysis of Energy Harvesting of Three-Split Core Coils

The three-core cable three-phase current vector sum is zero, so it is impossible to get
the induced voltage via the traditional circular core coils [4]. According to the char-
acteristics of magnetic field distribution, a specially designed core coil made of a
circular core divided into three parts was proposed, as shown in Fig. 2. Three-split core
coil were marked as A, B, and C, which output induced voltage eA(t), eB(t) and eC(t)
respectively. The number of turns per coil is N. According to Faraday electromagnetic
induction principle, the induced voltage of coil A, eA(t) could be expressed as (Fig. 1):
686 T. X. Sun et al.

dUA
eA ðtÞ ¼ N ð1Þ
dt
Where: ФA = magnetic flux through the coil A in Wb.

(a) t=0s (b) t=0.0017s

(c) t=0.005s (d) t=0.0083s

Fig. 1. Magnetic field value around the three-core cable at different times

Fig. 2. Three-split coils around the three-core cable

The magnetic flux through coil A is the surface integral of the magnetic flux density
through the section of coil A. However, the magnetic field around the three-core cable
with armor is unevenly distributed. So, the average magnetic flux density cannot be
used to calculate the magnetic flux according to the traditional complete Rogowski coil
calculation scheme. ФA could be calculated as follows:
Z
UA ¼ B
d~
~ S ð2Þ
S

The relationship between the magnetic flux density B(r, h) and the magnetic field
strength H(r, h) of point P(r, h) could be expressed as:
Research on Energy Harvesting and Its Application in Thermal Condition Monitoring 687

Bðr; hÞ ¼ lr l0
H ðr; hÞ ð3Þ

Where, µr represents the relative permeability of coil A. µ0 represents the vacuum

permeability. Based on (2) and (3), the output voltage of coil A eA(t) could be
expressed as:
R
d ~
d~
lr l0
H S
eA ð t Þ ¼ N S
dt ð4Þ
Rb
d lr l0 h
H
dq
¼N a
dt

where: a = the inner radius of coil A in Meter(m)

b = the outer radius of coil A in meter(m)
H’ = the component of the magnetic field strength in the tangential direction of
point P(r, h) in Tesla(T), which was set because the direction of the magnetic field
of some point outside the three-core cable surface is not parallel to the tangential
direction.

2.2 Simulation of Energy Harvesting of Three Partial Core Coils

The 3D model of a YJLV22-10 kV-3 * 50 three-core cable and three-split energy
harvesting coils installed outside the cable surface were built in Ansys Maxwell [5], as
shown in Fig. 3.

Fig. 3. The 3D model of the three-core cable and energy harvesting coils

In the analysis, a 50 Hz three-phase current of 80 A was adopted. The simulation

start time is 0 s, the termination time is 0.02 s, and the simulation step is 0.001 s. The
radius of the conductor is 4 mm. The insulation thickness is 3.9 mm. The semicon-
ductor shield is 0.8 mm. The armour thickness is 0.5 and the diameter of the cable is
53 mm. The core and the armour of the cable are copper. The insulation material of the
cable is polyethylene. For the energy harvesting coils, the material of the core is ferrite.
The rest of the space is vacuum. The Length based mode is used to split the model, the
688 T. X. Sun et al.

max size is selected as 5 mm to ensure the validity of the simulation. It is observed that
when 80 A three-phase current was adopted, an induced three-phase AC voltage of
about 400 mV is obtained on the energy harvesting coils, as Fig. 4 showed.

Fig. 4. Simulation waveform of energy harvesting coils

Factors affecting the performance of the three-core cable three-splitting energy

harvesting coils may be the cable current, the air gap between the three split cores, the
position of the three-split core relative to the three-phase conductor of the cable, etc. all
these factors can be analyzed by Ansys Maxwell [5].
Cable current varied linearly from 20 A to 160 A in steps of 20 A in the 3D model.
Figure 5 showed that the coil output voltage was proportional to the cable current.

Fig. 5. Coil voltage as a function of cable current

The curvature of the core is parameterized in the 3D model. The greater the cur-
vature, the smaller the air gap between the cores. The core arc curve changed linearly
from 40° to 119° in steps of 10°, Fig. 6 showed the relationship between the output
voltage of a coil and the core arc. Therefore, theoretically, in the case where the
magnetic cores are not in contact, the smaller the air gap between the split cores, the
larger the output voltage of the coil.
Research on Energy Harvesting and Its Application in Thermal Condition Monitoring 689

The three-split core coils can be arbitrarily rotated on the cable surface with the
cable center as the axis, so the three-split core has different positions relative to the
cable three-phase conductor. The position of the three-split core is described by the
angle between the line connecting the center of the core and the center of the cable and
the line connecting the center of the conductor and the center of the cable, allowing it to
vary linearly from 0° to 120° in steps of 10°. Figure 7 is a plot of coil output voltage as
a function of core rotation angle. It is observed that when the center of the core, the
center of the cable conductor, and the geometric center of the cable are collinear, the
output voltage of the coil is maximized.

Fig. 6. Coil voltage as a function of core radius

3 Development of Device

Energy harvesting coils composed of three-split cores were designed to be mounted

outside the cable surface based on the principle of three-core cable energy harvesting,
which can output three-phase AC voltage. The three-phase AC is converted to DC by a
three-phase bridge full-controlled rectifier circuit, and then boosted by a DC/DC
module. Then the power management module performs voltage regulation to provide a
5 V operating voltage for the single-chip control system.
On the other hand, the single-chip control system also controls the DC/DC module.
At the same time, the voltage monitoring module monitors the voltage amplitude of the
output of the three-phase rectifier bridge, and feeds back to the single-chip control
system to determine whether to use the energy harvesting coils output voltage to
directly supply energy or use the alternate energy storage battery to power the sub-
sequent circuit. The system block diagram of energy harvesting and temperature
measuring device for three-core are presented in Fig. 8 [8, 9].
690 T. X. Sun et al.

Fig. 7. Coil voltage as a function of core rotation angle

Fig. 8. Block diagram of energy harvesting and temperature measuring device

3.1 Design of the Energy Harvesting Coil

Under low-frequency working conditions, Mn-Zn ferrite has the characteristics of high
magnetic permeability and high magnetic flux density, and the loss is low. Therefore,
Mn-Zn ferrite is selected as the material of the magnetic core. Its relative magnetic
permeability is 2500, and the maximum magnetic induction is 0.4T [10]. The structure
of the energy harvesting coil is demonstrated in Fig. 9, and its structural parameters are
shown in Table 1.

Fig. 9. Energy harvesting coil structure

Research on Energy Harvesting and Its Application in Thermal Condition Monitoring 691

Table 1. Parameters of coil

Parameter Value Parameter Value
h1 in ° 28 h2 in ° 32
h2 in ° 4 h1 in ° 116
ri in mm 33 Ro in mm 45
Core height in mm 25 Coil number in N 240

3.2 Rectifier and Boost Circuit

A typical bridge three-phase rectifier circuit was used to convert the induced three-
phase voltage outputted on the three-split coil into direct current.
In order to avoid the situation that cable current is too small, the coil induced
voltage amplitude is lower than the forward voltage drop of the conventional silicon or
germanium diode, resulting in the rectifier circuit not working, a kind of special diode
with very low forward voltage drop made by Texas Instruments was chosen. Its typical
forward voltage drop at 8 A is typically only 26 mV, meeting the requirements of this
device.
The DC power output from the three-phase rectifier bridge is converted into a stable
5 V DC power using a typical boost circuit and a voltage regulator chip for use by the
single-chip control circuit.
In the initial state, the energy harvesting device itself has no energy supply, and the
circuit cannot work normally. Therefore, in the initial state, the boost circuit is powered
by the fully charged rechargeable lithium battery. The combination circuit of lithium
battery and boost circuit is shown in Fig. 10. The output voltage of the boost circuit is
detected by a resistor divider. When it is detected that the voltage value coupled from
the cable body is too low or zero, the MOSFET connected to the lithium battery is
turned on, and the lithium battery replaces the boost circuit to power the latter circuit.
The MOSFET uses a P-channel chip component with a turn-on voltage of −1 V, while
the lithium battery has a rated output voltage of 9 V and a capacity of 800 mAh.

Fig. 10. Voltage monitoring circuit

692 T. X. Sun et al.

3.3 Control Unit

The main function of the control unit is to control the PWM of the switch in the boost
circuit and control the temperature measurement unit. The low-power, high-
performance AT89C51RC chip with programmable and erasable memory and its
peripheral circuits are used.

3.4 Temperature Measuring Unit

The HDC1000YPAR temperature measurement chip with extremely low power con-
sumption and high precision is used to measure the temperature of the cable surface.
The temperature test range is −40 °C–125 °C, the accuracy is ±0.2 °C, and the power
supply voltage is 3 V–5 V.

4 Measurement and Calculation

4.1 Measurement of the Load Current
According to theoretical analysis and simulation analysis, the load current of the three-
core cable is proportional to the induced voltage of the three-split core coil. Therefore,
the magnitude of the cable load current can be obtained by measuring the magnitude of
the induced voltage output of the three-split core energy harvesting coil installed
outside the cable surface. The proportional coefficient k between the load current value
I/A and the induced voltage value U/mV is 0.21 based on the simulation result above,
which could be expressed as:

I ¼ 0:21U ð5Þ

4.2 Measurement of the Cable Surface Temperature

The three-core distribution cable buried directly in the ground has a relatively stable
surrounding environment. The temperature of the three-core cable surface is evenly
distributed and changes uniformly with the change of the load current. Therefore, the
cable surface temperature value of the three-core cable can be measured at any point on
the surface of the three-core cable, which can be shown as Fig. 11:

Fig. 11. Measurement of the cable surface temperature

Research on Energy Harvesting and Its Application in Thermal Condition Monitoring 693

4.3 Calculation of Cable Core Temperature

The internal temperature field of 10 kV three-core cable was modelled and simulated
using ANSYS finite element simulation software.
In steady state, the conductor loss value of the cable (calculated by cable current,
conductor resistivity and geometric parameters) and the skin temperature was input into
the model built by ANSYS, and the temperature field distribution can be obtained. The
conductor loss was calculated as 1.45  104 W/m3 when the current was 50 A. The
cable surface temperature was set as 37 °C, and the steady-state temperature distri-
bution of the three-core cable was shown in Fig. 12. It can be seen that the core
conductor temperature is calculated to be about 65.7 °C.

Fig. 12. Temperature field distribution of three-core cable section under steady state

In transient state, the step load current waveform shown in Fig. 13 was input into
the model, and the external temperature is set to be 28.4 °C. The core temperature
change curve can be obtained as shown in Fig. 14.

Fig. 13. Cable current as a function of Fig. 14. Transient core temperature as a function
time of time
694 T. X. Sun et al.

5 Experimental Verification

An experimental platform for thermal condition monitoring of 10 kV three-core XLPE

power cable was built, as shown in Fig. 15. The current in the cable conductor can be
regulated by the three-phase voltage regulator. Information of the magnitude of the
cable current was obtained by the current transformer.

Fig. 15. Experimental platform for thermal condition monitoring of 10 kV three-core XLPE
power cable

5.1 Measurement of the Load Current

Steps in 5 A, the cable current was gradually increased from 10 A to 65 A. The
relationship between the output voltage of the coils and the load current was shown in
Fig. 15. It was observed that the cable load current was proportional to the output
voltage of the energy-harvesting coils, and the ratio was about 0.20, which was very
close to the simulation result, 0.21.

5.2 Calculation of Core Temperature

At steady state, the room temperature was first measured to be 27.2 °C. And the cable
core temperature and surface temperature was measured when the cable current was
20 A, 30 A, 40 A and 50 A respectively. Then the core temperature was calculated by
ANSYS. The results were shown in Table 2. It was observed that the error was within
3%.
In the transient state, the room temperature is measured to be 28.4 °C, and the load
current of the step was adjusted according to the step current waveform in Fig. 14. The
core temperature measurement value and the calculated value are extracted every
10 min, and the relative error was calculated, which was within 5% (Table 3).
Research on Energy Harvesting and Its Application in Thermal Condition Monitoring 695

Table 2. Results at steady state

Current/A Core temperature Core temperature Error
measured/°C calculated/°C
30A 42.6 43.8 2.82%
40A 51.6 52.7 2.13%
50A 62.8 64.2 2.23%
60A 76.4 78.5 2.74%

Table 3. Results at transient state

Time/min Core temperature Core temperature Error
measured/°C calculated/°C
10 30.0 30.5 1.67%
20 30.7 31.4 2.28%
30 32.3 33.8 4.64%
40 33.8 34.9 3.25%
50 36.6 38.4 4.90%
60 39.3 40.2 2.29%
70 35.6 35.1 −1.40%
80 32.6 32.1 −1.53%
90 29.6 30.2 −2.03%
100 28.4 29.0 2.11%
110 30.0 30.5 1.67%
120 30.7 31.4 2.28%

6 Conclusions

In this study, a three-core power cables energy harvesting method is proposed and a
three-core cable energy harvesting and temperature measuring device is designed to
monitor the thermal condition of the three-core power cables. The output voltage of
three-split coils was transformed into a stable 5 V DC. The temperature measuring
device can measure the cable surface temperature. The load current can be obtained by
the output voltage of the energy-harvesting coils. Then the core temperature of the
cable can be calculated using ANSYS. Errors in calculation of the core temperature
may not exceed 3% at steady state and 5% at transient state.

Acknowledgments. This work has been initiated and supported by the project of
GDKJXM20185371.
696 T. X. Sun et al.

References
1. Zheng, Y.L., Wang, N., Li, H.J., et al.: Methods and development of current carrying
capacity calculation for power cables. In: Power Distribution, vol. 29, no. 3, pp. 26–31
(2010). IEC 60287-1-1 Electric cables-calculation of the current rating, part 1: current rating
equations (100% load factor) and calculation of losses, section 1: general (2006)
2. Liang, Y.C., Wang, Q.L., Yan, C.H., et al.: Temperature field and ampacity calculation of
cable buried in local conduit using 3D finite element method. High Voltage Eng. 37(12),
2911–2917 (2011)
3. Liang, Y.C., Zhao, Y.C.J., Yan, C.H.: Numerical calculation of temperature field and
ampacity of power cables in channel. High Voltage Eng. 38(11), 3048–3053 (2012)
4. Yuan, Y., Dong, J., Gan, J., et al.: A three-core power cable online monitoring system based
on phase current sensing. In: IEEE International Instrumentation and Measurement
Technology Conference, pp. 1–6. IEEE, Turin (2017)
5. ANSYS. https://www.ansys.com/
6. Xing, Y., Li, H.J.: Calculation and restoration of underground cable ratings based on a
combined technique. Electr. Eng. 14(10), 1 (2013)
7. Zhao, X., Keutel, T., Baldauf, M., et al.: Energy harvesting for overhead power line
monitoring. In: International Multi-Conference on Systems, Signals and Devices, pp. 1–5.
IEEE, Chemnitz (2012)
8. Li, H.Y., Shang, Y.W., Wu, C.C., et al.: Research on a electromagnetic induction fed power
supply from high voltage cable. Power Electron. 46(10), 83–86 (2012)
9. Schalkwyk, J., Hancke, G.P.: Energy harvesting for wireless sensors from electromagnetic
fields around overhead power lines. In: IEEE International Symposium on Industrial
Electronics, pp. 1128–1135 (2012)
10. Liu, J.G., Fu, X.M.: Development of Mn-Zn ferrite in magnetic properties. J. Magn. Mater.
Devices 36(2), 113–119 (2005)
 Multidimensional Analysis of a Real
Transformer Fleet Based on the Evaluation
of Oil Properties

Sebastian Schreiter1,2(&), Holger Lohmeyer3, and Peter Werle4

1
Faculty of Electrical Engineering and Information Technology,
Department of Electric Power Systems, HTWK Leipzig,
Leipzig University of Applied Science, Leipzig, Germany
sebastian.schreiter@htwk-leipzig.de
2
FTZ Leipzig e.V., Research and Transfer Centre at the HTWK Leipzig,
Leipzig, Germany
3
ABB AG, Halle/Saale, Germany
4
Institute of Electric Power Systems, Division of High Voltage Engineering
and Asset Management, Schering-Institute, Leibniz University Hannover,
Hanover, Germany

Abstract. Transformers belong to the most important and most elementary

elements in the power system. Because an unplanned outage of a transformer
can lead to interrupted power supply or the stop of production, the related costs
can be much higher than the costs of the transformer itself. Therefore it is of
elementary interest for operators of transformers to know the condition of the
transformers, identify risky units and to avoid unplanned outages. Therefore
different methods of condition assessment were developed. The most important
groups of those methods are electric and dielectric measurements of the trans-
former and the different analyzing techniques on the insulation liquid (oil tests).
The different oil tests can be separated into two main groups of parameters:
Firstly the parameters of the standard oil test (SOT). Secondly, failures at the
active part can be indicated with the help of the results of the dissolved gas
analysis (DGA). In this contribution both groups of parameters - the results of
the standard oil test as well as the results of a dissolved gas analysis - were
analyzed in a multidimensional investigation to observe if dependencies
between both groups of parameters can be found. Therefore it is investigated if
typical increased gas contents can be found depending on oil properties or if
both groups of parameters can influence each other. These investigations were
based on a huge number of oil test results of real transformers of different
voltage levels, applications, age, and manufacturers.

Keywords: Power transformer
Dissolved gas analysis oil test

1 Introduction

Transformers are among the elementary components at the nodes of the electrical
energy supply. Therefore, the assessment of their condition is for the operator of
equally elementary interest, because a failure of the transformer leads by the outage of
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 697–704, 2020.
https://doi.org/10.1007/978-3-030-31676-1_66
698 S. Schreiter et al.

the supplied systems and the resulting energy supply or loss of production to further,
sometimes significantly higher costs. Those costs can significantly exceed the initial
cost of the transformer, which is why it is of particular interest to transformer operators
to detect the condition of the transformer, early oncoming faults and to be able to make
a well based decision in case of a fault. For this purpose, different methods of condition
assessment and fault diagnosis are available. These methods can be separated into two
main groups: Electrical measurements and oil diagnostics. Electrical measurements
were carried out using the bushings of the transformer and can help to assess differ-
entiated diagnostic. To do such measurements it is necessary to take the transformer out
of service [1].
Oil samples on the other hand can be taken during operation and are therefore much
easier to handle. Within the oil diagnostic there are two main applications: Firstly the
investigation of the oil properties (standard oil test, SOT) and secondly the dissolved
gas analysis (DGA). In order to examining the oil properties, the physical and chemical
properties of the insulating liquid are examined to investigate how far the oil can still
withstand the thermal and dielectric stresses occurring during operation [2]. The second
application, the dissolved gas-in-oil analysis can be used to find indications of faults in
the active part of the transformer [3].
The evaluation of the SOT is easy: The measurement results were compared to
limiting values, named in the standard IEC 60422 [2]. In order to interpret the results of
a DGA two steps are necessary: In a first step a distinction is made between typical and
atypical gas concentrations and rates of increase. Only in case of atypical gas con-
centrations or increasing rates in a second step the nature of the reason for these
abnormalities - the type of failure - is identified [3].
This method is therefore based on typical gas concentrations. The standard gives
examples of typical behaviour but in the end, the typical concentrations must be defined
for each group of transformer or even each individual transformer and might depend
not only of the type of the transformer, but also its operation as well as its ageing status.
In this work, both the insulating oil properties and the results of dissolved gas-in-oil
analysis were included in a multi-dimensional analysis in order to identify possible
connections between these two parameter groups. In particular, it will be investigated
whether typical gas concentrations or typically increased gas concentrations occur
depending on the oil properties. The basis of this investigation is a large group of
results from oil samples, in which the results of a gas-in-oil analysis as well as a
simultaneous investigation of the insulating oil properties were combined. These values
were achieved from a large transformer population with different voltage levels, power
classes and applications over many years.

2 Background
2.1 The Investigated Transformer Fleet
For this study the results of the measurements of dissolved gas-in-oil analyses as well
as results of standard oil test of real transformers were involved. Figure 1 shows as an
example the age structure of the underlying fleet. It can be seen that more than half of
 Multidimensional Analysis of a Real Transformer Fleet 699

the transformers are 30 years or older and the largest age cohort is 40 to 50 years old.
This means that the oil properties included in this study can be considered as typical
aged during operation. Therefore a representative behaviour of the database can be
assumed.

Fig. 1. Age of the transformer fleet.

2.2 Multidimensional Evaluation of Big Data Sets by Cluster Analysis

One mathematical method to find groups of similar data objects (or clusters) in a
multidimensional data set is the cluster analysis. To perform a cluster analysis, two
basic methods can be differentiated: Hierarchical and partitioning methods: The hier-
archical methods compute the multi-dimensional similarity between the individual
objects and combine then a few but similar objects into a group. Then these groups
were combined to bigger groups and so on until all objects are combined into a single
cluster (agglomerative methods). A other possibility for hierarchical methods are
divisive methods: here, initially, all objects are in one big cluster and this is succes-
sively divided into groups depending on their multi-dimensional similarity with fewer
objects until at the end each object represents a cluster [4].
By contrast, partitioning methods require in advance a number of target groups or
cluster to be defined by the user. Those methods first iteratively determine the areas, in
which the respective properties should lie, and then divides the objects depending on
their multi-dimensional similarity into these areas. The process is repeated until the
different clusters are as little as possible similar [5].
The similarity of the objects can be determined for both methods according to
various mathematical methods. Prior to this, however, a standardization of the data is
necessary so that all parameters can be considered uniformly. Also for the calculation
of the similarity or the distance of two objects there are different calculation methods.
One of these will be explained using the example of the squared Euclidean distance,
700 S. Schreiter et al.

which is mathematically described in Eq. (1) according to: In a two-dimensional view,

two parameters of a set of data can be normalized to the respective maximum value and
displayed case-wise in an X-Y diagram. The Euclidean distance between two values in
this diagram is equal to the sum of the squares of the two path lengths that must be
travelled in X and Y direction to get from one case to the next. In a multi-dimensional
consideration, this distance must be summed up not only in 2 but in N directions if N is
the number of dimensions.

X
N
Deukl ðx; yÞ ¼ ð yi  xi Þ 2 ð1Þ
i¼1

Where:
Deukl Euclidean distance
x first case
y second case
xi i-th parameter of case 1
yi i-th parameter of case 2
i control variable
N Number of cases

Fig. 2. Graphical interpretation of a 2-dimensional cluster analysis: cases with similar

parameters are summarized to groups or clusters.

The formation of clusters is illustrated in Fig. 2 in a simplified, two-dimensional

representation: The objects (each represented by a cross) are entered here in the two-
dimensional space. The parameters are called m and n. Based on their 2-dimensional
similarity the cases were grouped into different areas or clusters.
The Euclidean distance is also represented in Fig. 2: The distance between two
cases (blue crosses) is shown with green arrows. Both distanced must be geometrical
added to get the Euclidean distance.
 Multidimensional Analysis of a Real Transformer Fleet 701

In multi-dimensional space, more parameters are included in the formation of the

clusters, and the clusters are formed analogously by calculating proximity in multidi-
mensional space.

3 Results

3.1 Consolidation of the Data

In order to compare only suitable data with each other, the data were initially con-
solidated: only samples from the main oil compartment of the transformer were con-
sidered (oil compartments such as cable terminations, dome or load switch vessel were
excluded). In addition, samples made from hermetic-type transformers and samples
derived from insulating fluids other than mineral oil were excluded. In addition, only
one sample was included per year and transformer.

3.2 Procedure of the Cluster Analysis

In a previous study it was investigated whether there are typical gas concentrations
depending on the oil properties. It was found that the hierarchical methods cannot be
used successfully due to the large amount of data and parameters [6, 7].
Subsequently, a partitioning cluster analysis was used to investigate how many
target clusters can be usefully assigned. It was found that the cluster analysis for more
than 3 target clusters does not provide meaningful results, because otherwise clusters
with only a few or even only individual cases are formed, which however map extrema.
It also turned out that the failure gases alone form such a large variety of parameters
that uniform clusters would be difficult to form. Therefore, in this paper, the results of
the gas-in-oil analysis were first summarized into one parameter, in which the sum of
all failure gases was formed, in which the following gases are contained:
• Hydrogen
• Methane
• Ethan
• Ethene
• Acetylene
• Propane
• Propene
In the following, this summary is called the sum of the failure gases or gas sum.
The atmospheric gases nitrogen and oxygen were not included, as they are mainly
dissolved by contact with air in the oil. Carbon dioxide and monoxide were not used,
because these gases have relatively large proportions even in typical gas-in-oil analyses
compared to the considered gases. In addition, the following oil properties were
included:
• Breakdown voltage (BV),
• Colour,
• Dielectric Dissipation Factor (DDF),
702 S. Schreiter et al.

• Acidity
Since the water content is comparable only at oil temperatures during sampling
above 20 °C [2], all samples with lower sampling temperatures need to be excluded,
which would led to a significant (and undesirable) case reduction. Therefore, this
parameter was not included in the consideration at all. In addition, the interfacial
tension was not included to reduce the degree of complexity, resulting in clearer cluster
results and the interfacial tension according to IEC 60422 is only mentioned as a
complementary measurement.
In order to make the different parameters, which are heterogeneous in the numerical
values, more comparable, all measured values were first scaled to typical values, so that
measured values, which were normalized to the numerical value one represent typical
(good) behavior. Larger numerical values always indicate worse behavior, numerical
values below one better behavior. During normalization, the mean value of the samples
considered to be typical for the gas concentrations were assumed to be typical for the
insulating oil values and was kept to normative suggestions. Thus, the color number
became 3, the breakdown voltage to 60 kV (limit between good and fair for category A
equipment), the neutralization number to 0.1 mg KOH/g (limit between good and fair
for category A equipment) and the dielectric dissipation factor to 100  10−3 (limit
between good and fair for Category A equipment).

3.3 Results of Cluster Analysis

The results of the partitioning cluster analysis with the search for three target clusters
are shown in the following diagrams and are discussed below. As mentioned before,
values around the value one represent the nominal case, while values below symbolize
better properties. Values above one indicate worse cases. All six diagrams are scaled
for clarity, that the ordinate always contains the range 0 to 4, good properties are
therefore symbolized by representation in the lower quarter (coloured light green)
(sometimes the light green area is completely obscured), objects in the upper quarter
(light red area) are, stand for worse properties.
In the first cluster, which can be seen in Fig. 3, the majority of the samples are
represented. As it can be seen from the upper diagram, the normalized gas sums are
always less than one and thus classified as typical. Also, the insulating oil properties
(represented here by color, neutralization number and breakdown voltage) have values
well below or around one and only in exceptional cases beyond.
In the second cluster (Fig. 4), the diagrams show predominantly typical gas sums,
but here, in particular, the breakdown voltage is significantly decreased, while the
dielectric dissipation factor has predominantly moderate values. Thus, this cluster can
be considered as the cluster of typical gas sums but abnormal oil properties.
Figure 5 shows the cases of the third cluster. This summarizes the cases, in which
both the gas sums and the oil properties are particularly high (the values are up to 3.5
times the mean value). In particular the dielectric dissipation factor can be estimated
with values up to more than four times higher than the normal value (Figs. 6, 7 and 8).
 Multidimensional Analysis of a Real Transformer Fleet 703

Fig. 3. Cluster 1: no abnormalities in gases. Fig. 4. Cluster 1: no abnormalities in SOT.

Fig. 5. Cluster 2: abnormalities in gases. Fig. 6. Cluster 2: abnormalities in SOT.

Fig. 7. Cluster 3: no abnormalities in gases. Fig. 8. Cluster 3: abnormalities in SOT.

704 S. Schreiter et al.

4 Conclusions

With the help of the cluster analysis, relationships between different parameters or
measured variables can be examined on the basis of larger data sets. In this contri-
bution, a procedure for the formation of clusters including both the results of dissolved
gas-in-oil analysis and insulating oil properties was presented. With this approach,
three target clusters were found that summarized three sets of results: in the first and
largest group, there are cases with typical results, both in terms of gas sums and
insulating oil properties. The second cluster contains cases with typical gas sums, but
impaired insulating oil properties, in particular reduced breakdown voltages. In the
third cluster the cases are summarized, which contain both increased gas sums as well
as conspicuous insulating oil properties.
With the help of the cluster analysis it could be shown that correlations between the
gas concentrations and the insulating characteristic values can be observed. Thus, the
gas concentrations from the gas-in-oil analysis can provide indications of the insulation
characteristics. In addition, the cluster analysis shows the possibility of dividing the
data of a fleet into the three areas good (green), medium (yellow) and bad (red), so that
this procedure can be used in the for asset management strategies.

References
1. CIGRE Working Group A2.34: Guide for Transformer Maintenance, CIGRE Technical
Brochure 445, February 2011
2. IEC 60422 ed. 4.0: Mineral insulating oils in electrical equipment Supervision and
maintenance guidance, January 2013
3. IEC 60599 ed. 3.0: Mineral oil-impregnated electrical equipment in service - guide to the
interpretation of dissolved and free gases analysis, September 2015
4. Werle, P.: Hierarchical cluster analysis of broadband measured partial discharges as part of a
modular structured monitoring system for transformers, Bd. 1999, S. v5-29-v5-29 (1999)
5. Bynen, E.J.: Cluster Analysis: Survey and Evaluation of Techniques. Springer, Netherlands
(1973)
6. Schreiter, S.: Individualisierte Bewertungskenngrößen für die öldiagnostische Untersuchung
von Leistungstransformatoren, Masterarbeit, Fernuniversität in Hagen, Hagen (2014)
7. Schreiter, S., Lohmeyer, H., Werle, P.: Mehrdimensionale Betrachtung ölanalytischer
Ergebnisse am Beispiel einer großen, realen Transformatorenflotte, VDE-Fachtagung
Hochspannungstechnik, Berlin, Deutschland (2018)
 Multidimensional Investigation of Transformer
Oil Properties

Sebastian Schreiter1,2(&), Holger Lohmeyer3, and Peter Werle4

1
Faculty of Electrical Engineering and Information Technology,
Department of Electrical Power Systems, HTWK Leipzig,
Leipzig University of Applied Science, Leipzig, Germany
sebastian.schreiter@htwk-leipzig.de
2
FTZ Leipzig e.V., Research and Transfer Centre at the HTWK Leipzig,
Leipzig, Germany
3
ABB AG, Halle/Saale, Germany
4
Institute of Electric Power Systems, Division of High Voltage Engineering
and Asset Management, Schering-Institute, Leibniz University Hannover,
Hanover, Germany

Abstract. The investigation of the oil properties is one of the most basic, but
also one of the most powerful methods of condition assessment on power
transformers. With the standard oil test (SOT) it can be tested if the oil is still
able to withstand the stresses occurring during operation. In order to do a
standard oil test different parameters are measured. With some of these
parameters for example the breakdown voltage mainly the capability to with-
stand dielectric stresses can be investigated. Other parameters such as acidity or
dielectric dissipation factor can help to assess the aging of the insulating liquid.
Because some parameters indicate the same properties (such as ageing state) it
could be possible that there are pairs or groups pf parameters, which give
redundant results. With other words: it could be possible that some parameters
indicate each other. In this case it would be possible to reduce the number of
parameters, which were measured during the SOT. To investigate if such
behaviour can be observed in real oil properties of real transformers, the dif-
ferent parameters of oil properties of a huge database were analysed with cor-
relation analysis. Additionally it is shown how often the different oil properties
can be classified as good, fair or poor according to the applicable standard IEC
60422. Using these information operators of power transformer are able to
evaluate the results of the SOT of their own transformers not only according to
the standard, but also according to the performance of other real transformers.

Keywords: Power transformer
Standard oil test
Correlation

1 Introduction

The investigation of the properties of the insulation liquid – standard oil test SOT - is
one of the most basic and also most approved methods of condition assessment on
power transformers.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 705–713, 2020.
https://doi.org/10.1007/978-3-030-31676-1_67
706 S. Schreiter et al.

With the help of the SOT it is examined if the insulating oil still is able to withstand
stresses, which occur during operation [1]. Therefore this method provides a valuable
contribution to avoid unplanned outages. Furthermore the SOT is as well as the dis-
solved gas analysis (DGA) [2] one of the advantageous methods. The handling of oil test
is very easy: The transformer can stay in operation to take an oil sample. The oil sample
is than analysed in a laboratory were a DGA as well as an SOT can be done. While the
DGA can indicate e.g. electrical or thermal failures in the active part of the transformer
the SOT shows the condition of the insulating system of the transformer [3].
Because the oil is involved in a continuous ageing process the condition must be
checked in regular intervals to ensure that the oil is still able to withstand the stresses
during operation. Therefore different chemical and physical test were done such as
measurement of the breakdown voltage, moisture content or dissipation factor. Some of
these parameters examine the remaining ability to withstand the dielectric stresses
during operation some other parameters can indicate ageing of the oil, e.g. due to
thermal stresses [2].
Because some of the parameters tend to the same properties (such as ageing) it
could be possible that the results correlate which each other and therefore give
redundant results. In a previous study, the relationships between the different gas
concentrations were already examined with the help of the correlation method [4].
In order to investigate whether such behaviour occurs in real transformers, corre-
lations between the different insulating oil properties for several groups of typical
transformers were determined on the basis of real measured values of oil samples from
a transformer fleet with the help of the correlation method addition, this transformer
fleet was used to investigate how often the measured parameters of the insulating oil -
individually or as a whole - correspond to the categories mentioned in the standard IEC
60422: good, fair and poor.

2 Background

2.1 Interpretation of the Results

The standard IEC 60422 specifies a series of tests, which can or must be carried out to
investigate the residual quality of the insulating oil. There are those that are used to
answer the question of how far the insulating oil can still withstand the dielectric
stresses. In this case the breakdown voltage is the most important parameter. Other
parameters are used to indicate the ageing of the oil. Such parameters are the dielectric
dissipation factor, the acidity or the interfacial tension. The standard names different
ageing mechanisms as the cause of changes in these parameters [1].
In order to assess the condition of the oil, the standard proposes limiting values,
which are depending on the nominal voltage of the equipment and its application.
Based on these limit values, the measured results can then be sorted into the ranges
good, fair and poor [1].

2.2 Investigated Transformers Fleet

For this study the results of the standard oil test of many thousand transformers of
different voltage and power levels as well as different applications were involved.
 Multidimensional Investigation of Transformer Oil Properties 707

The age structure of the transformer fleet is shown in Fig. 1. It can be seen that the
largest age cohort is 40 to 50 years old. More than half of the transformers are 30 years
or older. This means that the oil properties included in this study can be considered as
typical aged during operation. Therefore a representative behaviour of the database can
be assumed.

Fig. 1. Age of the transformer fleet.

2.3 Correlation for 2-Dimensional Applications

In order to investigate whether different insulating oil parameters are related to each
other, the correlation is appropriate: With the help of the correlation, the lues of two
different parameters can be analysed over a large data set. To do so this two parameters
were considered as a function of each other. The aim is to get a representative value,
which describes how strong theses parameters are connected to each other. The cor-
relation is done according to Eq. (1).
PN
i¼1 ðxi  xm Þ  ðyi  ym Þ
1
r ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
N
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN ð1Þ
N 2 2
1
N i¼1 ð x i  x m Þ  1
N i¼1 ðyi  ym Þ

Where:
r Pearson correlation coefficient
xi single value of parameter x
yi single value of parameter y
xm Mean of all values of parameter x
ym Mean of all values of parameter y
i control variable
N Number of cases
708 S. Schreiter et al.

If all available data of this both parameters were included in this calculation, a single
value is the results. This value, the correlation coefficient, can have values between
minus one and one. Here zero represents a pair of parameters, which shows no corre-
lation (no relationship between these two parameters). A correlation coefficient of one
represents an ideal positive relationship (with the rise of one parameter increases to the
same extent the other parameter). Minus one stands for an ideal negative relationship
(with the rise of one parameter, the other parameter decreases to the same extent).
Values between one (or minus one) and zero show a relation, which is not ideal [5].
Figure 2 illustrated the understanding of the correlation coefficient: In both dia-
grams, the two values of a set of two parameters are shown as a function of each other.
Each pair of values can be considered as a case.
In the upper diagram there are pairs of values (symbolized by a cross), which have a
high value both in parameter 1 and in parameter 2, and those which have a medium or
even small value in both parameters. This means with the knowledge of one parameter
the behaviour of the other can be assumed or predicted. The correlation coefficient is
therefore high; e.g. it will have a value close to one. If all the stars would be straight on
the green line, the relationship would be ideal and the correlation coefficient would be
equal to one.
In the lower diagram, the cases are scattered across the entire diagram - one
parameter can therefore not predict the other. The correlation coefficient of this dataset
is zero or very close to zero.
A (positive) correlation between two insulating oil properties would therefore mean
that the knowledge of one property can predict or at least approximate the other
property.

Fig. 2. Visualization of the principle of correlation: upper diagram: good correlation (high
correlation coefficient); bottom: low correlation (low correlation coefficient).

It is important to understand that with the help of the correlation no statement can
be made about the cause of the relation. In case of the oil properties a high correlation
coefficient would not mean that the reason of the increase of one parameter is the
increase of the other one. A high correlation coefficient means that both are rising in the
same extant due to a common reason. In case of the oil properties, e.g. the ageing of the
oil could be a reason for such behavior.
 Multidimensional Investigation of Transformer Oil Properties 709

3 Results
3.1 Consolidation of the Data
In order to use only representative data a consolidation of the used data was necessary:
First of all, only measurement results of oil samples, taken from the main tank were
used. Separated parts of the tank such as cable sealing boxes, domes or the diverter
switch compartment were not involved. Furthermore, only samples taken from open
breathing transformers were included to avoid mixing the results from hermetic
transformers with the results of open breathing ones. Finally only results from trans-
formers filled with mineral oil were taken into account. To ensure that all transformers
are represented equally per year and transformer only one oil sample was included in
this study. The age of the transformer at sampling date was calculated by the knowl-
edge of the year of manufacturing and the sampling date.

3.2 Correlation
Initially the correlation coefficient of each pair of the following oil properties was
calculated:
• Breakdown voltage (BV), Color, Dielectric dissipation factor (DDF), Water content
(at 20 °C), Acidity, Interfacial tension (IFT), Age
The results of this analysis are concluded in Table 1 (left). The correlation coef-
ficients describe the relation between the properties at the top of the column respec-
tively the begin of the line. For a better understanding the results were colored marked
depending on the value (depending on the strength of the correlation):
• r < 0,35: no correlation (no colour)
• r ± 0,35 to ±0,5 weak correlation (yellow)
• r ± 0,5 to ±0,7 middle correlation (red)
• r > ± 0,7 strong correlation (purple)
Correlation coefficients lower than 0,35 were not shown beacause of their low
impact. Because each pair of parameters would appear twice and due to the fact that the
correlation coefficient of a parameter which itself is exact one the fields in the upper
part of the table are empty.
It can be seen in Table 1 (left) that especially the breakdown voltage does not
correlate with any other parameter. So, it would not be possible to predict the break-
down voltage by another parameter.
A strong correlation can be observed between interfacial tension and colour and a
middle one between interfacial tension and acidity. Weak correlations can be found
between water content and breakdown voltage, dielectric dissipation factor, acidity and
interfacial tension.
The age of the transformer at sampling date show only a weak correlation to the
interfacial tension.
710 S. Schreiter et al.

Table 1. Correlation coefficients of all included pairs of parameters, left: all cases, right:
included only pairs of parameters if they are considered as fair or poor

3.3 Correlation of Fair or Poor Results

In the previously discussed investigation all samples were included. To ensure that
such low correlations as they can be observed are not due the possibility that the
majority of good results influence the results inadequately, a second investigation was
done. Therefore only results, which can be categorized as fair or poor according to the
standard were included. Then, the correlation coefficients were calculated again for
each pair of oil properties. The results of this second run are shown in Table 1 (right).
The correlation coefficients are marked in the same way as it was done in Table 1
(left). The table must therefore be read row by row: In the second row, for example the
correlation coefficient between the breakdown voltage and the other oil parameters is
shown. In order to calculate theses values only cases with fair or poor values of the
breakdown voltage were included. The correlation coefficients show how good the
other parameters can predict a decreased breakdown voltage. In the next row the
correlation coefficients in case of fair or poor values of the dissipation factor to the other
oil parameters was calculated and so on.
Due to this method, each pair of parameters is named twice: The cases, which were
included in the correlation between breakdown voltage and dissipation factor are dif-
ferent in case of fair or poor breakdown voltage than in case of fair or poor results of the
dissipation factor.
The table can therefore be red in such a way that high correlation coefficients mean
that the parameter at the top of the column indicates the parameter at the beginning of
the row.
Since the age of the equipment at sampling date cannot be bad, but should be taken
into account, the last column shows the relationship between the increased insulating
oil parameters and the age of the transformer. This results in an asymmetry in the table.
These results confirm the results of the first investigation: Table 1 (right) shows that
the breakdown voltage has no correlation to any other oil property neither to the water
content. The interfacial tension shows a middle correlation to the colour and the acidity.
A weak correlation can be found between acidity and interfacial tension. The water
content is increasing with other oil parameters but is not indicated by other parameters.
One explanation for this outcome is that the breakdown voltage is not depending on
the absolute water content but on the relative moisture. However, the relative moisture
is depending on many parameters like the temperature but also the ageing of the liquid,
thus no clear correlation to only one parameter can be seen Furthermore, the ageing
 Multidimensional Investigation of Transformer Oil Properties 711

parameters are indicating not the same ageing processes but different ones and therefore
do not depend on each other.
Finally it could be shown that no oil parameter is depending on the age of the
transformer e.g. an old transformer does not necessarily show bad ageing properties in
the oil.

3.4 Classification of the Oil Properties Depending on the Category

of the Transformer
At the end, it should be presented informatively how often the examined samples and
thus the entire fleet fulfil the requirements of the standard or not. This is to give
operators of power transformers the possibility to reflect the results of their own
transformers not only to the standard but also to a typical fleet of transformers. For this
purpose, each measured value was classified according to IEC 60422 based on the
equipment characteristics in the areas good, fair or poor and afterwards the distribution
for each equipment category was determined.

Fig. 3. Graphical illustration of the distribution of the insulating oil properties for all cases.

Figure 3 illustrate the distribution of the investigated cases into the ranges good,
fair and poor according to the standard independent from the equipment category. It is
shown that, apart from the interfacial tension, only a few percent of the cases must be
classified as poor, in all equipment categories.
712 S. Schreiter et al.

Furthermore, only a few cases were considered as fair, the results of the oil test for
most of the properties and equipment in more than 90% of cases can be classified as
good. Only the interfacial tension must be classified as fair or poor in about 50% of the
cases.
However, the results could also be interpreted that approximately 10% of the
transformers of the investigated fleet have critical values, which lead to the question of
the overall reliability of the whole system, which contains all these transformers.

4 Conclusions

Based on the correlation analysis, it was investigated whether the insulating oil
properties indicate each other. For this purpose, the correlation coefficients between the
different insulating oil parameters were determined. This was done twice: In the first
calculation all samples were included. In the second run only cases, where at least one
parameter was considered as fair or poor according to the standard were used.
Both investigations show that in particular the breakdown voltage is not indicated
by any single insulating oil parameter. But even between the other insulating oil
parameters, there are only minor relations. A strong correlation for example was found
between the interfacial tension and colour. Furthermore a middle correlation was found
between colour and acidity. The water content finally shows only weak correlations to
other oil properties.
The standard describes that the different oil properties indicate different ageing
processes and different stresses. Therefore the different parameters will react inde-
pendently, because in each single case not all ageing mechanism will act with the same
intensity.
The discoursed results therefore show, that the different oil properties do not predict
each other. Therefore it is necessary to measure all oil parameters and take them into
account for a condition assessment in order to ensure a comprehensive evaluation of the
oil.
Finally, the results of many thousand SOT were evaluated to show the overall
condition of the representative transformer. The relative distribution of parameters in
the 3 categories good, fair, poor for each category of transformer according to the
standard IEC 60422 results in the following: The insulation oil fulfils in more than 90%
of the cases the requirements according to the standard and can be considered as good.
On the other hand, approximately 10% of the cases show values, which must be
considered as fair or poor. Such transformers may be subject to particular restrictions
or may already be in a dangerous condition.
With this representation, operators of power transformers may be able to use these
exemplary results not only to classify the results of their own resources according to the
limit values of the standard, but also to assess them in the context of other transformers,
i.e. the fleet considered here.
 Multidimensional Investigation of Transformer Oil Properties 713

References
1. IEC 60422 ed. 4.0: Mineral insulating oils in electrical equipment Supervision and
maintenance guidance, January 2013
2. IEC 60599 ed. 3.0: Mineral oil-impregnated electrical equipment in service - guide to the
interpretation of dissolved and free gases analysis, September 2015
3. IEC 60475 ed. 2.0: Method of sampling insulating liquids, October 2011
4. Schreiter, S.: Individualisierte Bewertungskenngrößen für die öldiagnostische Untersuchung
von Leistungstransformatoren, Masterarbeit, Fernuniversität in Hagen, Hagen (2014)
5. Ross, S.M.: Statistik für Ingenieure und Naturwissenschaftler/Sheldon M. Ross. Aus dem
Amerikan. übers. von Carsten Heinisch, 3. Aufl., Spektrum, Akad. Verl. Elsevier, München
(2006)
 Determining Temperature Dependence
of Dielectric Response by Temperature
Normalization Method

Yanfeng Gao1(&), Wei Cai1, Yi Lu1, Chun Deng1, Shun’an Gao1,

Wenxiang Xue2, Shuyuan Wang2, Hui Wang2, and Jifei Zhang2
1
State Grid Jibei Electric Power Co., Ltd., Research Institute,
North China Electric Power Research Institute Co., Ltd., Beijing 100045, China
gaoyfer@126.com
2
State Grid Jibei Electric Power Co., Ltd., Beijing 100045, China

Abstract. Temperature is a significant external influence of dielectric response

which almost always influence the characteristic relaxation frequency. These
different effects can be most easily be revealed by the temperature normalization
method of plotting the data from different temperatures. This method consists in
shifting the dielectric response measured at different temperature into coinci-
dence, by which the ‘master curve’ is determined. The physical significance of
temperature normalization method is to describe the relationship between the
temperature and spectral shape function of dielectric response (the spectral shape
function describes the relationship between dimensionless frequency and
dimensionless susceptibility, which directly relates to the micro-structure of
dielectric). In order to obtain the correct master curve, the temperature nor-
malization method should be carried out in log-log plot by shifting the real and
imaginary components of dielectric response simultaneously. Logarithmic axes
are used because translations along them are the same as changing the frequency
and amplitude by a multiplicative factor, i.e. the frequency and amplitude
become scaled. The locus of the reference point in the translation process should
also be recorded in detail, because it defines the relationship between inverse
characteristic (relaxation) frequency and inverse amplitude as a function of
temperature, which becomes a hidden variable. In this paper, the master curves
of dielectric response of HTV silicone rubber is constructed to explain and to
demonstrate the relevant issues in the application of the temperature normal-
ization method in the analysis of dielectric response.

Keywords: Dielectric response
Temperature
Normalization
HTV silicone

rubber

1 Introduction

Dielectric response is the response of dielectric to the external electric field and can
provide an effective method to investigate the microstructure and charge motion of
dielectric materials [1]. In an alternating electric field, the dielectric response is a
function of frequency and has been widely investigated in the frequency domain [1, 2].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 714–723, 2020.
https://doi.org/10.1007/978-3-030-31676-1_68
 Determining Temperature Dependence of Dielectric Response 715

In the frequency domain, the information regarding the dielectric response is contained
in the frequency-dependent susceptibility which is defined as the Fourier transform of
the response function. The frequency-dependent susceptibility is a complex function
whose real and imaginary components obey the Kramers-Kronig relation [3, 4]. Fur-
thermore, the susceptibility can be characterized and expressed in detail in terms of the
product of the amplitude and the spectral shape function. There are many models
proposed to describe the spectral shape function of susceptibility, such as Debye model
[5], Cole-Cole model [6], Davidson-Cole model [7], Havriliak-Negami model [8],
Dissado-Hill model [9, 10], etc. Among these models, the Dissado-Hill model is the
most sophisticated and describes dielectric response with parameters having definite
physical meaning [11], in comparison with other models, the interaction between
relaxing entities and the micro structure in dielectric can be quantitatively characterized
by using the Dissado-Hill model [12].
Temperature is a significant influence affecting the dielectric response [13], by
investigating the dielectric response at different temperatures, the temperature depen-
dence of relaxing entity (such as the thermal activation energy of given dielectric
response) can be obtained. In general, with increasing of temperature, the dielectric
response shifts along the direction of high frequency [1, 2]. It is noteworthy that a large
number of experimental data [14, 15] indicate that in the temperature range in which
the micro structure of dielectric do not alter significantly, the shape of spectra function
of susceptibility remains the same. Thus, it is possible to normalize the measured
dielectric response data at different temperatures by shifting the data laterally into
coincidence obtaining a mutually overlapping curve, i.e. the ‘master curve’ [1]. The
method of shifting the data measured at different temperatures into coincidence is
known as the temperature normalization method. Compared to the data obtained at a
given temperature, the master curve can extend the frequency range and increase the
data density which can eliminate the accidental error in the test and improve the
accuracy of the results. More importantly, the master curve can be employed to
investigate the influence of temperature on the micro structure of dielectric. In addition
to temperature, other external factors such as pressure [16] and humidity [17] also have
influence of dielectric response, analogously, the influence of these factors on dielectric
response can also be analyzed by using normalization method [18, 19]. In Jonscher’s
famous academic masterpieces on dielectric response, ‘Dielectric relaxation in solids’
[1] and ‘Universal relaxation Law’ [1, 2], the temperature normalization method has
been used in depth to study the temperature influence on the dielectric response of
various dielectrics, a lot of applications regarding temperature normalization method
can be seen there, these classical applications remind the researchers that in the
implementation process of temperature normalization method, some key points should
be concerned and identified clearly, they are, (1) the physical meaning of temperature
normalization, (2) the object to be processed in temperature normalization, (3) the
mathematical basis of temperature normalization, (4) the locus of reference point, and
(5) whether it is inevitable to obtain a coincidence master curve or not.
In this paper, the detailed method of temperature normalization and above key
points are described and analyzed. Taking the dielectric response of HTV silicone
rubber [20] as examples this paper gives a detailed explanation of the application of
temperature normalization in the analysis of dielectric response.
716 Y. Gao et al.

2 Master Curve and Temperature Normalization Method

2.1 The Spectra Function of Susceptibility
As introduced in Introduction, the information of the dielectric response is contained in
the frequency-dependent susceptibility, in general, the susceptibility vðxÞ can be
expressed in terms of the product of the amplitude vð0Þ and the spectral shape function
Fðx=xx Þ,

vðxÞ ¼ vð0ÞFðx=xx Þ ð1Þ

The specific expression of shape function Fðx=xx Þ varies from model to model,
the detailed expression of shape function can be found in the original literatures for
these models [5–10], the description of dielectric response model is not the main
purpose of the present paper, for brevity, the mathematic formula of these dielectric
response model are not given here.
It must be stressed that in the spectral shape function Fðx=xx Þ, the frequency x is
normalized by the characteristic frequency xx (its means that the frequency x is made
dimensionless by dividing by the characteristic frequency xx ). Similarly, the suscep-
tibility vðxÞ can also be made dimensionless by dividing by the amplitude vð0Þ, i.e. the
spectral shape function Fðx=xx Þ is the dimensionless susceptibility. In summary, the
spectral shape function Fðx=xx Þ describes the relationship between the dimensionless
frequency and dimensionless susceptibility.

2.2 Discussion of the Situation Where a Single Master Curve Can Be

Obtained
A large number of experimental data [1, 2, 14, 15] indicate that in the temperature
range where no significant change of structure of dielectric occurs, the spectral shape
function Fðx=xx Þ is independent of temperature, the temperature influence on the
dielectric response is mainly reflected in two aspects, i.e. the influence of temperature
on characteristic frequency xx and on the amplitude vð0Þ. Based on this phenomena,
the Eq. (1) can be re-written as a function of temperature T,

vðx; TÞ ¼ vð0; TÞFðx=xx ðTÞÞ ð2Þ

Where vð0; TÞ denotes the relationship between temperature and amplitude and
xx ðTÞ the relationship between temperature and characteristic frequency.
In the temperature range where no structure change of dielectric occurs, the
dielectric response measured at different temperatures can be normalized, the specific
method is described as follows: the normalization should be carried out in the log-log
plot; the measured dielectric response at different temperatures are shifted in parallel
along the frequency axis (horizontal translation) and the amplitude axis (vertical
translation) to form a coincidence and mutually overlapping curve, i.e. the ‘master
curve’. Compared to the data obtained at a given temperature, the master curve can
 Determining Temperature Dependence of Dielectric Response 717

extend the frequency range and increase the data density which can eliminate the
accidental error in the test and improve the accuracy of the results.
During the shifting process, the locus of reference point should be recorded in
detail. The information of the locus of reference point reflects the influence of tem-
perature on the amplitude and characteristic frequency of the dielectric response.
There are four key points that should be stressed in the temperature normalization
method:
1. According to Eq. (2), the temperature normalization object is susceptibility vðxÞ
rather than capacitance CðxÞ or permittivity eðxÞ. This is due to the fact that the
real part of capacitance C 0 ðxÞ contain the unavoidable C1 and the imaginary part of
capacitance may contain a contribution of conductance G, either of which may have
a very different temperature dependence from the susceptibility vðxÞ. In most
instances, it can be found that the premise of obtaining of a single master curve is to
analyse a single relaxation process, however, it should be noted that sometimes
there exists a measured dielectric response consisting of different relaxation pro-
cesses for which a single master curve can be obtained [19, 21]. This means that
these relaxation processes have the same temperature dependence, and are therefore
probably connected physically.
2. According to Eq. (2), the susceptibility is a product of amplitude vð0; TÞ and spectra
shape function Fðx=xx ðTÞÞ, thus, the normalization method only can be carried out
in the log-log plot in which the parallel translation is equivalent to the multiplication
or division operation. The normalization cannot be operated in the linear or semi-
log plot.
3. The normalization of susceptibility should be conducted for both real and imaginary
part of susceptibility with the same shift parameters. According to the Kramers-
Kronig relation [3, 4], the real and imaginary components of susceptibility reflect
the same physical process, and these two parts can be transformed to each other by
the Hilbert transform. The temperature impact on the dielectric is therefore reflected
in both the real and imaginary components of the susceptibility. If the normalization
of susceptibility is only conducted for a single component part of susceptibility, the
degree of freedom of judgment to determine whether or not the coincidence occurs
during the shifting process is too large, which could lead to the shift parameters
obtained from shifting the real part (imaginary part) of susceptibility not being
suitable for shifting to coincidence of the imaginary part (real part) of susceptibility.
Some mistaken analyses could then be carried out.
4. During the shifting process, the detailed information of the locus of reference point
should be recorded. As mentioned above, the information of the locus of reference
point contains the temperature influence on dielectric response. In summary, the
purpose of temperature normalization is to obtain two kinds of data, one is the
master curve, and the other one is the locus of reference point, both are
indispensable.
718 Y. Gao et al.

2.3 Discussion of the Situation Where a Single Master Curve Cannot Be

Obtained

1. In the temperature range where the change of dielectric structure occurs.

It is well known that the dielectric response has a directly relationship with its physical
structure [22]. If the change of temperature leads to the varying of micro structure of
dielectric, then the spectra shape function Fðx=xx Þ will also change with the varying
of temperature, which means the shape of susceptibility at different temperatures is not
maintained the same, so the parallel translation cannot lead to a coincidence and
mutually overlapping curve, in other words, a single master curve cannot be obtained in
this condition.
2 The measured dielectric response consists of multi different temperature depen-
dence relaxation processes
According to Eq. (2), it can be found that the premise of obtaining of a single master
curve is to analyse a single relaxation process. If the measured dielectric response
consists of a number of different temperature dependence relaxation processes, these
relaxation process may have different micro structure and charge motion properties, the
temperature influence on these relaxation process may not the same, thus, a single
master curve cannot be obtained in this condition.
In the dielectric response consisting of many different temperature dependence
relaxation processes, in certain frequency range the dielectric response may be domi-
nated by a specific relaxation process. In this frequency range, the temperature nor-
malization method can be applied and a partial master curve can be obtained. In this
situation, the locus of reference point reflects the temperature influence on this dom-
inant relaxation process.
In the situation where a single master curve cannot be obtained, the influence of
temperature on the dielectric response can be studied by using the equivalent circuit
method, the detailed procedure for this analysis is to obtain the specific parameters of
the equivalent circuit describing the measured dielectric at given temperature firstly,
then, the temperature dependent of these parameter can be analyzed which reflect the
influence of temperature on the dielectric response.

3 The Application of Temperature Normalization

3.1 Experimental
The dielectric materials used in this research is HTV silicone rubber, the raw rubber used
is polydimethylsiloxane (PDMS), the fillers are ATH filler and nano silica filler. The
mass ratio of PDMS, ATH and nano silica was 100:100:40. The detailed prepare pro-
cedure of this kind of HTV silicone rubber was given in Literature [20]. The samples
were cut in the form of circular slabs with a thickness of 2 mm. The dielectric response
was measured over the frequency range from 10−4 to 104 Hz. The experimental tem-
peratures were set to 20 °C, 40 °C, 60 °C, and 80 °C, the diameter of the measured
electrode was also 44 mm, the magnitude of the applied ac voltage was set to 2 V (rms).
 Determining Temperature Dependence of Dielectric Response 719

3.2 Results
The measured complex capacitance of HTV silicone rubber at 20 °C, 40 °C, 60 °C and
80 °C are given in log-log representation in Fig. 1, the individual pairs of complex
capacitance are displaced vertically for clarity by three decades. In the imaginary part
of complex capacitance, a substantial dielectric dispersion can be clearly identified
which moves in the direction of high frequency with the increase of temperature,
however, the real part of the complex capacitance in the high frequency region is
dominated by the infinite frequency capacitance C1 which can be seen to be tem-
perature independent, i.e. the dielectric dispersion in the real part of complex capaci-
tance is obscured by the infinity frequency capacitance, thus, the temperature
normalization method can not be executed on the data of complex capacitance directly.
In order to investigate the influence of temperature on the dielectric performance of
HTV silicone rubber, the susceptibility should first be obtained.

Fig. 1. Complex capacitance measurement results of HTV silicone rubber under different
temperatures

Based on the geometric dimension of the samples, the susceptibility of the HTV
silicone rubber can be calculated in terms of

C 0 ðxÞ  iC 00 ðxÞ  C1
vðxÞ ¼ v0 ðxÞ  iv00 ðxÞ ¼ ð3Þ
e0
S=d
720 Y. Gao et al.

Where e0 denotes the permittivity of free space whose values equals to 8.854 pF
m−1,
S=d is the geometrical sample factor.
Figure 2 shows the susceptibility results of HTV silicone rubber at different tem-
peratures, from which it can be found that in all the experiments the principal feature of
the dielectric response is the quasi-dc process for which the parallel real and imaginary
asymptotic behavior is clearly observed [10, 23]. At the lowest frequencies and highest
temperature, there is an indication of a decrease in the frequency exponent of the loss
which is evidence for the presence of a series diffusion barrier [24]. The points ‘A’ in
Fig. 2 is used as a reference point, the horizontal translation of the moving trace
describes the temperature dependence of the characteristic frequency of susceptibility
and the vertical translation gives that of the amplitude. The 80 °C is taken as the
reference temperature and the susceptibilities at 20 °C, 40 °C, 60 °C were shifted to the
susceptibility at 80 °C with the real and imaginary parts of susceptibility being shifted
simultaneously. The master curve and the locus of reference point are shown in Fig. 3.
In the frequency region indicated as region II in Fig. 3, the susceptibility data from
different temperatures come into coincidence and a perfect master curve is obtained.
The master curve in frequency region II shows the principle feature of quasi-dc process,
furthermore, it can be concluded that the micro structure responsible for the quasi-dc
process does not change in the temperature range form 20 °C to 80 °C, this is the
physical basic for the formation of a single master curve in region II. In the frequency
region III indicated in Fig. 3, the susceptibility data from different temperatures can not
come into coincidence by the same frequency-amplitude shifts as that for the quasi-dc
process, this means that the relaxation process in this frequency region has a different
physical origin as well as temperature dependence.

Fig. 2. Complex susceptibility results of HTV silicone rubber under different temperatures
 Determining Temperature Dependence of Dielectric Response 721

In the region II marked in Fig. 3, a partial master curve is obtained, thus, the locus
of reference point reflects the temperature influence on this dominant relaxation pro-
cess. It should be noted that in addition to a shift in the characteristic frequency of the
quasi-dc process the locus of the trace point in Fig. 3 shows that there is a substantial
increase in its amplitude with increasing temperature. This means that the concentration
of moving ions and their separation at a given frequency increase with increasing
temperature.

Fig. 3. The master curve of HTV silicone rubber dielectric response

Fig. 4. The temperature dependences of the shift frequency and amplitude applied in
temperature normalization method

Arrhenius relationship is employed to analyze both the relationship between the Ln

(shift frequency) and the reciprocal of the temperature and the relationship between the
Ln(shift amplitude) and the reciprocal of the temperature,
722 Y. Gao et al.

Wv
shift vðTÞ ¼ shift v0
exp ð4Þ
kT
Where shift vðTÞ is amplitude range for shifting in the temperature normalization,
shift v0 is a constant, Wv is the thermal activation energy of the amplitude of sus-
ceptibility. The Arrhenius plots in Fig. 4 show that both the shift frequency and shift
amplitude obey the thermally activated process and the activation energies of the quasi-
dc characteristic frequency and amplitude determined from the slope of the straight
lines are 0.27 eV and 0.24 eV respectively.

4 Conclusion
1. The physical meaning of temperature normalization is to describe the relationship
between temperature and spectral shape function. The spectral shape function
describes the relationship between the dimensionless frequency and dimensionless
susceptibility.
2. The temperature normalization should be conducted in the log-log plot and the real
and imaginary parts of dielectric response should be shifted simultaneously, the
locus of reference point should also be recorded in detail.
3. To obtain a single master curve is not an inevitable result. In the condition of the
dielectric response consisting of multiple relaxation processes with different tem-
perature dependences and in the temperature range where the structure of the
dielectric changes, a single master curve cannot be obtained.
4. A single master curve of dielectric response of HTV silicone rubber cannot be
obtained in the frequency range of 10−4 Hz to 104 Hz and in the temperature range
of 20 °C–80 °C. In the frequency range where the quasi-dc process dominates the
dielectric response, a partial master curve can be obtained, which indicates that the
shift frequency and shift amplitude of quasi-dc process both obey the thermally
activated process and the related thermal activation energies are 0.27 eV and
0.24 eV respectively.

Acknowledgments. Y. G. thanks Prof. L. A. Dissado and Prof. Xidong Liang for their brilliant
theoretical guidance in the analysis of the experimental results.

References
1. Jonscher, A.K.: Dielectric Relaxation in Solids. Chelsea Dielectric press (1983)
2. Jonscher, A.K.: Universal Relaxation Law. Chelsea Dielectrics Press (1996)
3. Kramers, H.A.: La diffusion de la lumiere par les atomes (1927)
4. Kronig, R.D.L.: On the theory of dispersion of x-rays. JOSA 12, 547–557 (1926)
5. Debye, P.: Some results of kinetic theory of isolators. Preliminary announcement.
Physikalische Zeitschrift 13, 97–100 (1912)
6. Cole, K.S., Cole, R.H.: Dispersion and absorption in dielectrics I. Alternating current
characteristics. J. Chem. Phys. 9, 341–351 (1941)
 Determining Temperature Dependence of Dielectric Response 723

7. Davidson, D.W., Cole, R.H.: Dielectric relaxation in glycerol, propylene glycol, and n-
propanol. J. Chem. Phys. 19, 1484–1490 (1951)
8. Havriliak, S., Negami, S.: A complex plane representation of dielectric and mechanical
relaxation processes in some polymers. Polymer 8, 161–210 (1967)
9. Dissado, L.A., Hill, R.M.: A cluster approach to the structure of imperfect materials and their
relaxation spectroscopy. In: Proceedings of the Royal Society of London. A. Mathematical
and Physical Sciences, vol. 390, pp. 131–180 (1983)
10. Dissado, L.A., Hill, R.M.: Anomalous low-frequency dispersion. Near direct current
conductivity in disordered low-dimensional materials. J. Chem. Soc. Faraday Trans. 2: Mol.
Chem. Phys. 80, 291–319 (1984)
11. Das-Gupta, D.K., Scarpa, P.C.N.: Modeling of dielectric relaxation spectra of polymers in
the condensed phase. IEEE Electr. Insul. Mag. 15, 23–32 (1999)
12. Dissado, L.A., Hill, R.M.: Non-exponential decay in dielectrics and dynamics of correlated
systems. Nature 279, 685–689 (1979)
13. Hill, R.M., Dissado, L.A.: The temperature dependence of relaxation processes. J. Phys. C:
Solid State Phys. 15, 5171–5193 (1982)
14. Ishida, Y.: Dielectric relaxation of high polymers in the solid state. J. Polym. Sci. Part A - 2:
Polym Phys. 7, 1835–1861 (1969)
15. Hill, R.M., Jonscher, A.K.: The dielectric behaviour of condensed matter and its many-body
interpretation. Contemp. Phys. 24, 75–110 (1983)
16. Williams, G., Watts, D.C., Nottin, J.P.: Dielectric relaxation of acrylonitrile-butadiene
copolymers as a function of frequency, temperature and applied pressure. J. Chem. Soc.
Faraday Transa. 2: Mol. Chem. Phys. 68, 16–24 (1972)
17. Bano, N., Jonscher, A.K.: Dielectric properties of humid mica surfaces. J. Mater. Sci. 27,
1672–1682 (1992)
18. Chaudhry, M.A., Jonscher, A.K.: The dielectric properties of mica paper in variable
temperature and humidity. J. Mater. Sci. 20, 3581–3589 (1985)
19. Chalashkanov, N., Dodd, S., Dissado, L., Fothergill, J.: Re-examination of the dielectric
spectra of epoxy resins: bulk charge transport and interfacial polarization peaks. IEEE Trans.
Dielectr. Electr. Insul. 21, 1330–1341 (2014). 01 January 2014
20. Gao, Y., Liang, X., Dissado, L.A., Dodd, S.J., Chalashkanov, N.M.: Dielectric response of
filled high temperature vulcanized silicone rubber. IEEE Trans. Dielectr. Electr. Insul. 23,
3683–3695 (2016)
21. Abdelmalik, A.A., Dodd, S.J., Dissado, L.A., Chalashkanov, N.M., Fothergill, J.C.: Charge
transport in thermally aged paper impregnated with natural ester oil. IEEE Trans. Dielectr.
Electr. Insul. 21, 2318–2328 (2014)
22. Dissado, L.A., Nigmatullin, R.R., Hill, R.M.: Dynamical processes in condensed matter. In:
Advances in chemical physics, vol. 63. Wiley, New York (1985)
23. Jonscher, A.K.: Low-frequency dispersion in carrier-dominated dielectric. Philos. Mag.
B 38, 587–601 (1978)
24. Hill, R., Pickup, C.: Barrier effects in dispersive media. J. Mater. Sci. 20, 4431–4444 (1985)
 Terminal Configuration and Sensitivity
Analysis of Tank Current Measurement
for FRA Diagnostics of Three-Phase
Transformer Winding Model

V. S. B. Chaitanya Duvvury(&) and Saurav Pramanik

Indian Institute of Technology, Kharagpur 721302, India

dvsbchaitanya259@gmail.com, saurav.pramanik@gmail.com

Abstract. This paper presents an in-depth analysis of tank current measure-

ment for fault detection in a three-phase transformer winding model. In the first
part of analysis, a unique fault detection sensitivity of tank current is presented
for a three-phase winding model followed by determining the best terminal
configuration to improve the sensitivity further. As a consequence of fault in
three-phase winding model, it is found that tank current measurement always
exhibits new additional resonance peaks. This feature is very unique for tank
current response compared to the features in conventional FRA measurements.
Further, an attempt is also made to quantify the fault-detection sensitivity of tank
current by defining two indices. One index (named as “S-index”) is defined by
the number of new extra resonance peaks in tank current magnitude response as
a consequence of fault. Another is the well-known “root mean square deviation”
between the two responses before and after the fault. Higher the index values,
more is the sensitivity for tank current to fault detection. Using these indices,
this paper also determines that the fault detection sensitivity is more for tank
current if the non-tested phase terminals are kept shorted and grounded. Sim-
ulation results and experimental measurements on a practical three-phase
winding model demonstrate the facts too. For practical viability, authors would
also like to extend the present work for actual transformers in near future.

Keywords: Frequency response
Tank current
Transformer windings

Resonance peaks

1 Introduction

Enhancing the fault detection sensitivity for winding damage or deformation in three-
phase transformer (power or distribution-class) has been always a key motivating factor
to the innovation of new measurement techniques for frequency response analysis
(FRA). As far as the sensitivity of fault detection is concerned, several efforts have been
reported in the literature to identify different terminal connections for FRA and their
sensitivity to the deformation in transformer winding [1]. Transformers when are in
service are often prone to winding related faults [2], and hence, FRA sensitivity in this
regard is very crucial to identify the preliminary damage before it results in a

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 724–737, 2020.
https://doi.org/10.1007/978-3-030-31676-1_69
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 725

catastrophic failure. The conventional FRA measurements [3, 4] (e.g. input admittance,
transfer admittance, transferred voltage) identifies the fault by recognizing only the
“changes” or “deviation” in subsequently measured frequency responses as compared
to the “healthy” response. But, recently a new measurement of tank current response
was introduced for the first time for FRA diagnostics of an isolated transformer
winding in [5]. Tank current response is actually defined by the current through the
ground capacitances of the windings (IT = RICg = IL − IN) and its contributions are
distributed throughout the length of the winding. Tank current magnitude response
exhibits additional new resonance peaks to the occurrence of fault in single isolated
winding. Further, it also experiences the change/shift in magnitude and position of
existing peaks and troughs. Appearance of new resonance peaks is the unique feature of
tank current compared to the conventional FRA measurements (for other voltages or
currents) which only experience the deviation but fail to exhibit any such new reso-
nance peaks. This is how the tank current measurement appears to be an alternative for
improved fault-detection sensitivity in FRA method. Following on the same line of
investigation, this paper presents a feasibility test of tank current method for its
application to a three-phase winding model and also proposes how the sensitivity of
tank current measurement could be enhanced further to indicate the fault. Objective of
the present work is detailed in Sect. 2.

2 Objective

Analysis of tank current measurement for three-phase winding is not straight-forward

as the non-tested windings influence the measurement on tested winding. Therefore,
first an equivalent three-phase model of transformer (i.e. a mutually coupled three-
phase ladder-network model shown in Fig. 1) was considered for simulation to com-
pute the tank current magnitude response and thereafter the responses for different
cases are carefully studied in steps for the following objectives:
• Is the FRA data for tank current response applicable for fault detection in three-
phase winding model?
• Do the new resonance peaks appear in tank current magnitude response for a fault in
all the phase windings?
• If yes, then study how the fault-detection sensitivity of tank current response could
be enhanced further using different terminal configurations.
Ladder-network model in Fig. 1 (consisting of series and shunt capacitances, series
inductance with couplings, interphase inductance and capacitances) inherently captures
the intricacies of winding behaviour under excitation for FRA measurement. For
healthy case, the ladder-networks for every phase are considered identical and they are
also coupled to each other through mutual inductance and capacitances in a symmet-
rical manner (details are given in Sect. 3). The definition of tank current for three phase
winding model remains the same and hence is computed by subtracting the neutral
current (IN) from line-current i.e. input-excitation current (IT = IL − IN). All these
currents are also shown by arrows in Fig. 1.
726 V. S. B. C. Duvvury and S. Pramanik

Fig. 1. A delta-connected three-phase ladder network model.

For the present study, only a delta-connected three-phase winding model is con-
sidered for two possible terminal configurations (Configuration-I: non-tested winding
terminals are shorted and grounded, Configuration-II: only the remote end of the tested
winding is grounded but the non-tested winding terminals are not shorted, see Fig. 1
for the connections). Tank current magnitude responses corresponding to each terminal
configuration of the model are computed for both healthy and faulty cases. These
responses are then compared for investigation to answer the questions above. It is to be
noted that for computing tank current magnitude responses, ladder-network in the
middle phase is always considered for excitation since it shares an identical coupling
with two outer phases. Objective of the primary investigation is to examine whether
new resonance peaks appear in tank current magnitude response or not when a fault
condition is occurred in the model. If the new resonance peaks are observable in tank
current magnitude response, the fault-detection sensitivity later could be analysed in the
following way: “higher the number of new resonance peaks, higher is the sensitivity for
tank current response to detect the fault”. This is further used to identify the best
terminal configuration for improved fault detection sensitivity of tank current. Further,
the features obtained above are also verified by the ‘conventional sensitivity mea-
surement technique’ i.e. by measuring the total deviation of the faulty responses from
the healthy response (i.e. root mean square deviation). However, since the present work
is only a preliminary work, only the faults in HV windings are considered and hence,
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 727

the ladder-networks for LV windings are not included in Fig. 1. Furthermore, the LV
windings in a transformer, if shorted and grounded; their influence on the measurement
for HV windings is also found very negligible in [6, 7].

3 Simulation Results

3.1 Results for Healthy Model

As demonstrated above, to compute the tank current magnitude responses (i.e., the ratio
of tank current spectrum to the spectrum of input voltage) a mutually coupled three-
phase delta-connected ladder-network model is considered for three HV windings.

Table 1. Three-phase winding model parameters

Inductances (mH)
Individual phase Mutual phase inductances
Inductances Adjacent phases Outer phases
LA1-A1 3/5 LA1-B1 1/4 LA1-C1 1/5
LA1-A2 1/3 LA1-B2 1/6 LA1-C2 1/7
LA1-A3 1/5 LA1-B3 1/8 LA1-C3 1/9
LA1-A4 1/7 LA1-B4 1/10 LA1-C4 1/11
LA1-A5 1/9 LA1-B5 1/12 LA1-C5 1/13
Capacitances (nF) and Resistance (Ω)
Individual phase Inter-winding capacitances
Cs 0.3 Two adjacent phases Outer phases
Cg 0.5 Cw 0.4 Cw neglected
R 0.4
* All are per section values
** A, B and C are three phases in Fig. 1 from left to right

For each phase, ladder network comprises of five sections. Parameter values for the
model are obtained from the literature and are also listed in Table 1. Details for
symmetry of the model are discussed above in Sect. 2. Tank current magnitude
responses for the aforesaid two terminal configurations of the model are computed
using MATLAB and are plotted in Figs. 2 and 3. From Figs. 2 and 3, it is clearly
visible that total four and seven number of resonance peaks are observable in tank
current magnitude response for terminal configuration-I and –II respectively. To
facilitate an easy counting, peaks are also labelled by the numbers in both the figures.
Later, for measuring the fault-detection sensitivity of tank current, faults were intro-
duced in the model for both the terminal configurations and the corresponding tank
current magnitude responses were computed again. Details are discussed below in
Sect. 3.2.
728 V. S. B. C. Duvvury and S. Pramanik

Fig. 2. Computed tank current magnitude response of a three-phase winding model (healthy
case, terminal configuration-I).

Fig. 3. Computed tank current magnitude response of a three-phase winding model (healthy
case, terminal configuration-II).

3.2 Results for Faulty Model

Here, the objective is to see how the resonance behavior of tank current is influenced
when a fault is introduced in the model. For this purpose, model parameters for Cs, Cg
and L in the first section of left-outer-phase were changed (changed values are
0.285 nF, 0.575 nF and 0.57 mH respectively) to simulate the mechanical fault for
winding damage/deformation and the corresponding tank current magnitude responses
were computed for both the terminal configurations. If any change is observed, the next
task is to examine which terminal configuration affords better sensitivity for tank
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 729

current measurement to the occurrence of fault. Computed new responses are plotted in
Figs. 4 and 5. Figures 4 and 5 are for the tank current magnitude responses for identical
faults in the model but with two different terminal configurations. From Figs. 2 and 4, it
is observed that seven number of new resonance peaks appear in tank current mag-
nitude response for terminal configuration-I whereas for terminal configuration-II, this
number is five (compare the responses in Figs. 3 and 5).

Fig. 4. Computed tank current magnitude response for a three-phase winding model with fault
in first section of left outer phase, terminal configuration-I.

Fig. 5. Computed tank current magnitude response for a three-phase winding model with fault
in first section of left outer phase, terminal configuration-II.
730 V. S. B. C. Duvvury and S. Pramanik

Therefore it is observable that the resonance behavior of tank current for isolated
winding which was observed earlier in [5] seems also to be a typical feature for three-
phase winding model. Thus the fault in a three-phase model can be identified simply by
observing new resonance peaks in tank current magnitude response measured subse-
quently after the occurrence of fault. Next, for analysing the sensitivity afforded by tank
current response for fault detection, an index ‘S’ is defined by the number of how many
extra new resonance peaks appear in tank current magnitude response. Certainly,
higher the number of S-index, improved is the fault-detection sensitivity for tank
current. Therefore, the sensitivity index ‘S’ for terminal configuration-I is found seven
whereas the same is five for terminal configuration-II. Thus it may be interpreted that
amongst two possible terminal configurations, fault detection sensitivity is higher for
tank current, for terminal configuration-I (at least for the present set of results) i.e. the
non-tested windings are kept shorted and grounded. It is also important to mention that
the sensitivity measurement of this kind for fault detection is never reported in the
literature for any conventional FRA measurements. The reason is the conventional
FRA measurements do not exhibit new resonance peaks as a consequence of fault in the
model (or winding). However, for tank current responses any non-zero value for S-
index will interpret the fault in the model. This is the unique aspect of tank current
measurement for fault detection. Further, the fault detection sensitivity of tank current
response with the above two terminal configurations are also verified by computing the
deviation between two magnitude responses for both healthy and faulty cases. This
deviation is quantified by another index “Root Mean Square Deviation” (RMSD). This
deviation is actually the conventional practice of measuring the sensitivity for FRA
data. It is defined as follows:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN 2
i¼1 ðY ðiÞ  X ðiÞÞ
RMSD ¼ ð1Þ
N

Where, ‘Y’ and ‘X’ are the tank current magnitude data for faulty and healthy traces
and ‘N’ is number of samples. Higher the RMSD value better is the sensitivity for fault
detection. Now, if the responses in Figs. 4 and 2 are compared for terminal
configuration-I, RMSD value is obtained as 64.86. For similar fault in the model, if the
responses in Figs. 3 and 5 are compared for terminal configuration- II, RMSD value is
obtained as 30.34. Therefore, it could be inferred that the terminal configuration-I is
more sensitive for tank current to detect the fault. This is also similar to what was
conferred for sensitivity measurement using S-index. Therefore, the conclusion is that
the tank current afforded better sensitivity in fault-detection if the non-tested winding
terminals are kept shorted and grounded. However, to verify the consistency of fault
detection sensitivity for the above two terminal configurations; another similar fault
was introduced in the third section of middle phase. Corresponding tank current
magnitude responses were computed again for both the above terminal configurations
(and these are the only possible terminal configurations for middle-phase excitation)
and the responses are plotted in Figs. 6 and 7. Here, for both the terminal configura-
tions, three new resonance peaks are observable. This is obtained by comparing the
healthy and faulty responses in Figs. 2 and 6 and the responses in Figs. 3 and 7.
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 731

Fig. 6. Computed tank current magnitude response for a three-phase winding model with fault
in third section of middle phase, terminal configuration-I.

Fig. 7. Computed tank current magnitude response for a three-phase winding model with fault
in third section of middle phase, terminal configuration-II.

Important finding is that the new resonance peaks always appeared for both the
terminal configurations and hence, the S-index is found a non-zero value to indicate the
fault. Later, the RMSD values are also computed for healthy and faulty responses for
both the terminal configurations and the values are obtained as follows: 67.31 for
terminal configuration I and 36.96 for terminal configuration –II. This concludes that
the overall fault detection sensitivity is always better for tank current with terminal
configuration-I. Summary of the results for the faults in middle and left-outer phase for
both the terminal configurations are listed in Table 2.
732 V. S. B. C. Duvvury and S. Pramanik

Table 2. Sensitivity comparison for simulated faults

Fault in # RMSD S-index
Left-outer phase Terminal config. -I 64.86 7
Terminal config. -II 30.34 5
Middle phase Terminal config. -I 67.31 3
Terminal config. -II 36.96 3

The following inferences are possible to draw from the results in Table 2:
• FRA data for tank current response seems promising for fault detection in three-
phase model
• Fault can be identified simply by observing a non-zero S-index for measured tank
current magnitude responses.
• It is also important to notice that that the sensitivity of tank current response for
fault-detection in three-phase model varies with terminal configurations.
• Simulations results in Table 2 infer that the terminal configuration-I is more sen-
sitive to fault for tank current (i.e. if the non-tested windings are kept shorted and
grounded).
• A lateral observation is that the terminal configuration-I appears better suited for
improved sensitivity if the fault occurs in non-tested winding compared to faults in
tested winding (i.e. middle phase).
The above inferences are also verified by computing the simulations for several
faults in different sections of three-phase model (and also with different set of model
parameters). For every case, the consistency is always found for tank current sensitivity
to the fault but the results are not included in the paper for brevity. However, an
experimental validation was performed later by conducting measurements on a prac-
tical three-phase winding model.

4 Experimental Measurements

To experimentally verify the sensitivity of tank current measurement for fault-detection

in three-phase model, a practical three-phase winding model was built in laboratory and
is shown in Fig. 8. Each phase of the model comprises of eight sections and all sections
are built in identical manner to ensure the uniformity of the model. Inductances are
provided by single layer copper turns on three identical insulating formers. Diameter of
each former is *200 mm (outer diameter) and 28 turns were used for inductance of
each section of the model. Cs, Cg, and inter-winding capacitances (Cw) of the model
are provided by externally connecting the lumped capacitors. Further, the insulating
formers were placed with equal gap distances to ensure the symmetry in mutual
inductances for two adjacent phases. Externally connected lumped capacitors for Cs,
Cg and Cw are 0.54 nF, 0.63 nF and 0.475 nF respectively.
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 733

Fig. 8. A three-phase winding model in delta-connection for tank-current measurement.

Measurement was done using the following setup:

• A signal generator capable of producing 20 Vp-p sinusoid with freq. range of 0–
100 MHz
• A 2.5GS/s, 8-bit Digital storage oscilloscope
• A clamp-on current probe with a sensitivity of 1 mA/mV, bandwidth: 0–100 MHz
• 50 Ω coaxial cables
Tank current of the model was measured sequentially for both the terminal con-
figurations. Middle phase of the model was excited by a 20 VP-P sinusoids and the
measurement was performed using sweep frequency method. For each frequency,
signals for line-current (IL), neutral-current (IN) and input-voltage (Ein) were acquired
on the oscilloscope using measurement cables but the tank current signal was obtained
by subtracting the neutral current from the line-current using “Math” operation in
Oscilloscope. From the measured data, tank current magnitude responses are plotted in
Figs. 9 and 10 for both the terminal configurations. From Figs. 9 and 10, it is clearly
visible that total ten and thirteen number of resonance peaks appear in tank current
magnitude response for terminal configuration-I and –II respectively (low-magnitude
peaks are zoomed and identified). Later, to examine the fault detection sensitivity of
tank current for both the terminal configurations one fault was introduced in the
second-section of left outer phase of the model by shorting two adjacent turns and
changing Cs and Cg to 0.359 nF and 0.949 nF respectively. After introducing the fault,
the measured responses are plotted in Figs. 11 and 12. For both the terminal config-
urations, measured responses in Figs. 11 and 12 are compared with their corresponding
healthy responses in Figs. 9 and 10.
734 V. S. B. C. Duvvury and S. Pramanik

Fig. 9. Measured tank current magnitude response of a three-phase winding model (healthy
case, terminal configuration-I).

Fig. 10. Measured tank current magnitude response of a three-phase winding model (healthy
case, terminal configuration-II).

Fig. 11. Measured tank current magnitude response of a three-phase winding model (faulty
case- I, terminal configuration-I).
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 735

Fig. 12. Measured tank current magnitude response of a three-phase winding model (faulty
case-I, terminal configuration-II).

“S-index” values for sensitivity measurement (i.e. new number of resonance peaks)
are obtained as six and five for terminal configuration-I and -II. Further, root mean
square deviations are also tabulated in Table 3. It is observed that similar to simulation
results in Sect. 3, terminal configuration-I is observed more sensitive to fault-detection
for tank current. It is also to be noted that the feature of tank current resonance
behaviour i.e. the appearance of new resonance peaks for fault condition is also
observed for a practical three-phase winding model. Later measurements are also
repeated for a similar fault in the fourth section of middle phase and the measured
responses are plotted in Figs. 13 and 14 for terminal configurations I and II respec-
tively. “S-index” values are obtained as six and two for terminal configuration-I and II.
Corresponding RMSD values are also included in Table 3.

Fig. 13. Measured tank current magnitude response of a three-phase winding model (faulty
case-II, terminal configuration-I).
736 V. S. B. C. Duvvury and S. Pramanik

Fig. 14. Measured tank current magnitude response of a three-phase winding model (faulty
case-II, terminal configuration-II).

Table 3. Sensitivity comparison for faults in practical model

Fault in # RMSD S-index
Section 2, left outerTerminal configuration-I 7.8215 6
Terminal configuration-II 4.8917 5
Section 4, mid-phase Terminal configuration-I 9.6891 6
Terminal configuration_II 2.3388 2

So, for a similar fault in middle phase, terminal configuration-I is again observed as
more sensitive to fault detection. Results are summarized in Table 3. Findings are
similar to what was observed for simulation results for three-phase model.

5 Conclusions

This paper contributed towards identifying unique fault detection sensitivity of tank
current for a practical three-phase winding model. Further, it also determines the best
terminal configuration for enhanced measurement sensitivity of tank current for fault
detection. Best terminal configuration is determined based on the appearance of more
number of resonance peaks in tank current magnitude response for fault condition and
also by the “root mean square deviation” between the responses before and after the
fault. Findings from simulation results are supported by the actual measurements on a
practical three-phase winding model. From the results, it is also observed that sensi-
tivity is more for a fault in non-tested outer-phase instead of a fault in excitation phase.
Sensitivity of tank current measurement reported in this paper also needs to be rein-
forced for several other fault conditions in practice. For practical viability, authors
would also like to extend this preliminary work for actual transformers in near future.
In the end, authors believe that the tank current method is indeed a new potential for
FRA diagnostics to draw more attention of utilities and manufacturers in near future.
 Terminal Configuration and Sensitivity Analysis of Tank Current Measurement 737

Acknowledgments. Authors thank SRIC, IIT Kharagpur for granting ‘ISIRD’ fund to build
experimental laboratory setup.

References
1. Samimi, M.H., Tenbohlen, S., Akmal, A.A.S., Mohseni, H.: Effect of different connection
schemes, terminating resistors and measurement impedances on the sensitivity of the FRA
method. IEEE Trans. Power Delivery 32(4), 1713–1720 (2017)
2. Jayasinghe, J.A.S.B., Wang, Z.d., Jarman, P.N., Darwin, A.W.: Winding movement in power
transformers: a comparison of FRA measurement connection methods. IEEE Trans. Dielectr.
Electr. Insul. 13(6), 1342–1349 (2006)
3. Picher, P., Lapworth, J., Noonan, T., Christian, J.: Mechanical condition assessment of
transformer windings using frequency response analysis, CIGRE WG A2.26, Technical
Brochure 342 (2008)
4. Tenbohlen, S., Jagers, J., Gebauer, J., Muller, P., Lapworth, J., Yukiyasu, S., Desai, B.,
McIntosh, A., Bastos, G., Bo, L.: Transformer reliability survey: interim report. Electra, 46–
49 (2012)
5. Pramanik, S.: Resonance behavior and sensitivity to detect mechanical change in transformer
winding: shunt current versus neutral current. IEEE Trans. Power Delivery 30(5), 2276–2283
(2015)
6. Li, C.X., Zhu, T.Y., Xia, Q., Yao, C.G., Zhao, Z.Y.: Influence of untested winding in FRA
test for winding diagnosis. IET Gener. Transm. Distrib. 12(8), 1704–1711 (2018)
7. Pramanik, S., Duvvury, V.S.B.C., Sahoo, S.: Tank current measurement of three-phase
transformer: its resonance behavior and sensitivity to detect mechanical faults, available in
early access. IEEE Transactions on Power Delivery. (https://doi.org/10.1109/tpwrd.2019.
2914249)
 Reliable Diagnostics on Rotating Machines
Using FRA

S. Uhrig1(&), F. Öttl2, N. Augeneder3, and R. Hinterholzer3

1
Munich University of Applied Science, Munich, Germany
stephanie.uhrig@hm.edu
2
OMICRON electronics GmbH, Klaus, Austria
3
voestalpine Stahl GmbH, Linz, Austria

Abstract. The frequency response analysis (FRA) is a known highly sensitive

method to detect mechanical and electrical defects in power transformers.
Investigations already showed, that this technique is applicable to detect broken
windings or shorted turns in rotor and stator of rotating machines. Here, the
applicability of FRA to other defects was investigated, like broken rods of
damping windings in synchronous machines or in a squirrel cage of induction
motors, which often can be detected only by disassembling the machine. The
reason lies in the design: damping windings as well as the squirrel cages have no
accessible outer terminal, which can be used for electrical measurements. As the
FRA method is capable to measure also capacitive or inductive couplings, it
might be a possible low effort alternative. This investigation focuses on the
reliability of failure detection. For a clear conclusion, numerous tests were
performed on rotating machines of different types. It was found, that the fre-
quency response can be measured on assembled rotating machines highly
reproducible up to 1 MHz in different setups. Main factors to be considered are a
good connection technique and a clear description of the measurement setup for
later tests. A comparative analysis is only possible for the same phase config-
uration as well as the same rotation angle of the rotor. Similar to power trans-
formers a time-based comparison is preferred. Different phases have not exactly
the same frequency responses. For the investigation described in this contri-
bution, broken rods in a squirrel cage of a 55 kVA induction motor and in the
damping winding of a 37 kVA synchronous machine were reproduced. It was
found, that the FRA method is sensitive enough to identify even single defects.

Keywords: FRA
Synchronous motor
Induction motor
Damping winding

1 Introduction

Rotating machines in service are continuously exposed to mechanical forces, which can
cause different defects. Besides mechanical components like shaft or bearing, also
windings can be affected in terms of shorted turns in rotor or stator or broken rods of
damping windings in synchronous machines or in a squirrel cage of induction motors.
Especially the broken rods in damping windings are difficult to detect, as they are
completely insulated from the machine windings and therefore as well from outer
accessible terminals. So there is no possibility for a direct contact, which can be used

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 738–751, 2020.
https://doi.org/10.1007/978-3-030-31676-1_70
 Reliable Diagnostics on Rotating Machines Using FRA 739

for measurement. Like broken rods in squirrel cages of induction motors, these defects
can often only be detected after disassembling.
It was already shown, that Frequency Response Analysis (FRA) is suitable for
detecting shorted turns in rotating machines [1–4]. This was also confirmed by sim-
ulation [5]. As it is also possible to measure capacitive and inductive couplings by
means of FRA, it was investigated whether it is possible to reliably detect broken rods
of damping windings in synchronous machines or in a squirrel cage of induction
motors. The aim was the diagnosis of these defects without disassembling the machine.
Therefore, only connections in the terminal box are used for measurements.

2 Frequency Response Analysis

The Frequency Response Analysis

(FRA) is an approved method to
assess the condition of a power
transformer [6]. It bases upon the
fact, that components inside an
asset, like electric windings, mag-
netic parts as well as the tank/
housing form a complex electrical
network (Fig. 1), having a unique
frequency response, similar to a
finger print. Ruptures, short-circuits
and even geometrical deformations
Fig. 1. Representation of the electrical circuit of an will change the electrical network
asset, here the example of a power transformer and subsequently the frequency
response. The FRA uses this fact:
On a possibly defect asset, the frequency response is measured and compared to a
reference, measured in healthy state. By analysing the deviations, the defect can be
confirmed or refuted.
The FRA uses a sinusoidal voltage source U1 of a known amplitude and phase in a
wide frequency sweep. The voltage source is applied to an accessible terminal and the
response signal U2 is measured from a second terminal. The frequency response is
generated by using the amplitude ratio and the phase shift between both signals.

3 Measurement Objects and Setups

3.1 Squirrel Cage Motor

A 55 kVA squirrel cage motor was used for the investigation (Fig. 2). The connection
schema of the stator phases were changed for different measurements. A reference
position was marked at the shaft of the rotor, to retrace the influence of the rotor position.
Furthermore, some rods of the squirrel cage were sawed to simulate a broken rod (Fig. 3).
Initially measurements with a single rod rupture were performed, followed by
measurements with two rods ruptured. Here, a second rod was sawed in a distance of
740 S. Uhrig et al.

six rods (Fig. 3). To identify clear tendencies, two more rods between both mentioned
rods were sawed, reproducing a squirrel cage with four broken rods. For final mea-
surements, the rods were brazed similar to a typical repair. All measurements were
performed on a completely assembled machine.

Fig. 2. Investigated induction motor with squirrel cage, assembled

Fig. 3. Squirrel cage with marked reference position (blue) and two sawed rods (red)

3.2 Synchronous Motor with Damping Winding

The second measurement object was a 37 kVA synchronous motor, having six ter-
minals of the stator winding and two of the rotor winding accessible for measurements.
To ensure the reproducibility, the rotor reference position was defined and marked at
the brushes (marked white in Fig. 4).
 Reliable Diagnostics on Rotating Machines Using FRA 741

The damping winding consisted of 36 rods. Five of these rods were manipulated to
simulate a rupture. These five rods, named A to E, were situated at 0°, two at 180°, at
225° and at 270° compared to the reference position mark (Fig. 5).

Fig. 4. Rotor with marked reference position (white) and screwed bolt in damping winding (red)

The ruptures were reproduced by drill holes with threads. A screwed bolt is used for
a highly conductive contact, making the damping winding complete, a missing bolt is
simulating a rod rupture. Using this configuration, it was possible to investigate dif-
ferent failure combinations. The frequency response, measured with all screwed bolts
was used as reference for later analysis.

Fig. 5. Position and name of manipulated rods

742 S. Uhrig et al.

4 Reproducibility and Influencing Factors

The FRA method is based on comparisons of the measured frequency response to a

reference curve, preferably from a previous measurement in a known healthy condition,
from a sister- unit or from another phase of the same unit. In any case the changes of
the frequency response must be clear and distinct. Therefore, influencing factors must
be known, eliminated or at least minimized to ensure a reliable result of the analysis.
Thus, systematically investigations were performed, to identify influencing factors and
their impact on the frequency response of rotating machines.
Fundamental influencing factors are known from power transformers [6]. Fur-
thermore, rotating machines can alter their rotor position. Taking these into account,
following aspects were systematically investigated:
• Connection technique
• Measurement voltage
• Rotor position
• Connection schema of the stator winding (Y/D/single phases)
Results of this contribution are given only for the induction motor from 3.1, which
have the same qualitative behaviour as measured for the synchronous motor. The only
exception is the influence of the rotor position, which is shown for both machines.
To distinguish clearly between different influences, every parameter is altered
separately, whereas all other parameters are kept constant. If not given differently, the
rotor was in reference position (0°) and the measurement voltage was 10 VPP.

4.1 Connection Technique and Measurement Voltage

The measurements were conducted on the induction motor with a single rod rupture by
using different connection clamps like crocodile clips or screw clamps, but with regards
of the known aspects for a high reproducibility for power transformers [6]. This meant
e.g. ensuring a low contact resistance, using short cable connections and a proper
ground connection using aluminium braids. After initial measurements all connections
were removed and renewed several months later to indicate the reproducibility.
Measurements were performed, having the stator in Y – connection, D – connection
or leaving all terminals unconnected, which is later called “single phase”. All terminals,
not used for measurement were left floating. Figure 6 shows comparisons of a mea-
surements in Y – connection (blue and light blue) with a time gap of several months,
the same for D – connection (green and light green) and for single phase measurements
(red and orange). As it can be seen, the measurement is highly reproducible up to about
1 MHz. Deviations are visible only in the very high frequency range. Therefore, it is
assumed that the connection technique is suitable for rotating machines.
All measurements were performed with a measurement voltage of 1 VPP as well as
with 10 VPP. A direct comparison showed no deviations between the measured fre-
quency response.
 Reliable Diagnostics on Rotating Machines Using FRA 743

Fig. 6. Comparison of two independent measurements with a single rod rupture

4.2 Phase-to-Phase Comparisons

If no reference curves from previous measurements or sister units are available, a
phase-to-phase comparison might be used. Here, it is essential to distinguish between
deviations caused by defects or originating from a slightly different design of the other
phase. As shown in Figs. 7, 8 and 9, the frequency responses of all phases have a
similar characteristic shape. This is visible for each connection schema of the stator
winding. However, slight differences can be seen above 1 kHz.

Fig. 7. Comparison of phase U, V and W measured as single phase, single rod rupture
744 S. Uhrig et al.

Larger defects within the direct electrical circuit, like broken rods or shorted turns
of windings, would influence the characteristic shape of the frequency response and
might be recognizable in a phase-to-phase comparison. The defects investigated here,
are based only on capacitive and inductive couplings. Therefore, the influence on the
frequency response is much smaller and can not be clearly identified in a phase-to-
phase comparison. In practice, an assessment based purely on a phase-to-phase com-
parison should be avoided for this type of defect in rotating machines.

Fig. 8. Comparison of phase U, V and W measured in Y – connection, single rod rupture

Fig. 9. Comparison of phase U, V and W measured in D – connection, single rod rupture

 Reliable Diagnostics on Rotating Machines Using FRA 745

4.3 Rotor Position of the Squirrel Cage

Defects within the squirrel cage lead to asymmetries of the elecromagnetic coupling
along the circumference. It was assumed, that the influence of the rotor position on the
frequency response would be larger for a higher asymetrie, i.e. more rods were broken.
Therefore, the results of a four rod rupture is discussed below.

Fig. 10. Measurement of the single phase U with four rods ruptured at different rotor positions
(red: 0°, blue: 60°, green: 90°)

For all connection schemas slight deviations are recognizable in a frequency range
of 1 kHz to 300 kHz, which have the largest impact on the first extremum of the
magnitude (Fig. 10). Thus, the influence of the rotor position is existent, but small.

4.4 Rotor Position of the Synchronous Motor

Only inductive measurements can be performed for the induction motor, having both,
the source and response connected to the stator. For the synchronous motor also a
capacitive measurement between rotor and stator windings is possible. Measurement
results show (Sect. 5), that a defect in the damper windings, situated between rotor and
stator, has a much stronger influence on the capacitive measurement compared to the
inductive approach. Although the capacitive measurements are very promising to detect
broken damper bars, these are affected quite strongly by the rotor position. Measure-
ments with the same defect performed on different rotor position showed a significant
influence on the result (Fig. 11). The deviations are already visible at low frequencies
up to 5 kHz and change the characteristic shape of the frequency response. This means,
in practice measurements have to be performed always in the same rotor position,
otherwise they can not be used for comparison.
746 S. Uhrig et al.

Fig. 11. Rotor position influence for capacitive measurements at the synchronous motor

5 Broken Rods in the Squirrel Cage of an Induction Motor

The influence of a single, double and quadruple rod rupture as well as the repaired
squirrel cage was investigated. The measurements were performed in all three con-
nection schemas each (Y/D/single phases). For all connection schemas the frequency
response showed deviations mainly in the range between 1 kHz and 300 kHz. Below
and above this range, the curves for different defects show no significant change. For
further discussion Figs. 12, 13 and 14 show the mentioned frequency range between
1 kHz and 300 kHz, with respect to the different connection schema.

Fig. 12. Different defects at single phase U (magnified extract)

 Reliable Diagnostics on Rotating Machines Using FRA 747

It can be seen, that the deviation is increasing with increasing defect for Y – and
D – connection. The influence itself is small, but it was possible to reproduce them. As
they are in the same order as the influences of rotor position and different used phases
discussed in Sect. 4, a reliable assessment is difficult.

Fig. 13. Different defects at phase U in Y - connection (magnified extract)

Fig. 14. Different defects at phase U in D - connection (magnified extract)

748 S. Uhrig et al.

6 Broken Rods in the Damping Winding of a Synchronous

Motor

For the synchronous motor the terminals of all stator windings as well as the terminals
of the rotor winding is accessible. Therefore, different setups for a frequency response
measurement are possible:
• source and response ! stator
• source and response ! rotor
• source ! rotor; response ! stator
As the damper winding is situated between rotor and stator, the third is assumed to
be the most sensitive setup. The measurement voltage is applied to the clamps of the
shorted rotor winding, where also the reference measurement is connected. The
response is measured at the stator winding in Y – connection, as this is the most
common connection schema in practice. Using this setup, even single broken rods
within the damper windings can be clearly and reproducible detected. The deviations
caused by the broken rods are visible in two ranges of 1 kHz to 10 kHz and 200 kHz to
2 MHz (Fig. 15).

Fig. 15. Capacitive measurements at reference position of rotor (0°) for different defects and
combinations of defects
 Reliable Diagnostics on Rotating Machines Using FRA 749

Fig. 16. Extract 1 of results from capacitive measurements at reference position of rotor (0°) for
different defects and combinations of defects

Fig. 17. Extract 2 of results from capacitive measurements at reference position of rotor (0°) for
different defects and combinations of defects
750 S. Uhrig et al.

The assumption, that more defects will cause larger deviations is only conditionally
correct, as shown in Figs. 16 and 17. Defect A, D and E are situated within an angle of
90°. Here, the deviation is increasing with increasing number of defects as it can be
seen in Fig. 16. However, defect B and C are rotated compared to A by an angle of
nearly 180°. This double defect causes the largest deviation compared to the reference,
but in the opposite direction as caused by A, D and E. The combination of all five
defects causes a response again closer to the reference. It is assumed, that the influence
off al five defects are superimposing and thus mitigating the overall deviation.

7 Conclusion

It was investigated, whether defects in rotating machines which are not directly within
the electrical measurement circuit can be reliably detected using FRA. Broken rods in
the squirrel cage of an induction motor and in the damper winding of an synchronous
motor were simulated and measured.
To achieve reproducible and meaningful results, following aspects are essential for
a frequency response measurement:
• Use of suitable connection technique
• Rotor position must be the same as for the reference measurement
• Connection schema of the stator winding (Y/D/single phase) must be the same as
for the reference measurement
• Measurement direction must be kept the same as for the reference measurement
• Avoid phase-to-phase comparison
It was found, that broken rods in the squirrel cage of an induction motor cause
small, but reproducible changes of the frequency response measured from the stator
windings. The deviations are mainly visible between 1 kHz and 300 kHz.
Broken rods in the damper winding of a synchronous motor cause clear deviations
in a frequency response measurement between rotor and stator. The strongest influence
is visible in two different frequency ranges. It is assumed, that the deviation is larger the
more the defect is unsymmetrical.
Further measurements are necessary, to confirm whether the results are qualitatively
applicable to other machines.

References
1. Platero, C.A., Blázques, F., Blánquez, F.R., Rebollo, E., Battle, B., Fernandez-Sanchez, E.:
Application of Sweep Frequency Response Analysis (SFRA) for inter-turn detection of in
medium-voltage coils manufacturing. In: CIGRÉ Session 2012, Paris, France, 26–30 August
2012, paper A1-205 (2012)
2. Aranda Carmona, A.J.: Diagnostic method of electrical rotors by applying the sweep
frequency response analyzer (SFRA). In: CIGRÉ Session 2012, Paris, France, 26–30 August
2012, paper A1-106 (2012)
 Reliable Diagnostics on Rotating Machines Using FRA 751

3. Venikar, T.G., Ballal, M.S., Umre, B.S.: Application of sweep frequency response analysis
for the detection of winding faults in induction motor. In: 42nd Annual Conference of the
IEEE Industrial Electronics Society, IECON 2016, Florence, Italy, 23–26 October 2016,
pp. 1458–1463 (2016)
4. Blánquez, F.R., Platero, C.A., Rebollo, E., Blázquez, F.: Evaluation of the applicability of
FRA for inter-turn fault detection in stator windings. In: 2013 9th IEEE International
Symposium on Diagnostics for Electric Machines, Power Electronics and Drives
(SDEMPED), Valencia, Spain, 27–30 August 2013, pp. 177–182 (2013)
5. Florkowski, M., Furgał, J.: Modelling of winding failures identification using the frequency
response analysis (FRA) method. Electr. Power Syst. Res. 79, 1069–1075 (2009)
6. Cigré WG A2/26: Mechanical condition assessment of trans-former windings using
Frequency Response Analysis (FRA), Brochure 342, Paris (2008)
 Partial Discharge Characteristics of Palm Fatty
Acid Ester for Dissolved Gas Analysis

Yunwen He1, Kiyoshi Wakimoto2, Yang Xu1(&), and Shengtao Li1

1
Xi’an Jiaotong University, Xi’an, People’s Republic of China
xuyang@xjtu.edu.cn
2
Meidensha Corporation, Shizuoka, Japan

Abstract. Currently, natural ester insulating oil is widely used in transformers

because of its fire safety and environmental sustainability. Natural ester insu-
lating oil has high viscosity and poor oxidation stability compared with mineral
oil. Therefore, the modified ester palm fatty acid ester (PFAE) has been
developed, which has a monoester structure. PFAE has a lower viscosity and
greater oxidation stability than mineral oil. In this paper, the characteristics of
partial discharge lasted for 120 h were compared between PFAE and mineral oil
using needle-plane and creepage discharge electrodes. The dissolved gas pro-
duction characteristics for different discharge durations were analysed. The test
results showed that the discharge was large in the PFAE compared with mineral
oil for needle-plane electrode. And for creepage, discharge was comparable in
two oils. The main gases produced under the discharge were H2 and CO for both
oils, and C2H6 was not produced significantly in PFAE. The amounts of gen-
erated hydrocarbon gases in PFAE were lower than that in mineral oil for similar
discharge energies. The basic gas ratios method and the Duval pentagon showed
that the fault type in both oils were PD and D1.

Keywords: Insulating oil
Partial discharge
Dissolved gas analysis
PFAE

oil

1 Introduction

At present, natural ester is gradually being investigated and applied in transformers due
to its fire safety and environmental sustainability. Compared with mineral oil, the
relative permittivity of natural ester is closer to that of oil-impregnated insulating paper,
so the concentration of the electric field on the oil side can be suppressed [1]. However,
the commonly used soybean-based and rapeseed-based natural esters have high vis-
cosity and poor oxidation stability. Palm fatty acid ester (PFAE) was developed by the
Japanese Lion Corporation, Japan AE Power Systems Corporation, Hitachi, Fuji
Electric, and Meidensha jointly in 2006 [2]. PFAE is a monoester that is produced
through the reaction of palm oil and alcohol. On the one hand, PFAE has a high
electrical insulation performance, good thermal stability, biodegradability and a higher
relative permittivity compared to mineral oil. On the other hand, PFAE a lower vis-
cosity and greater oxidation stability than mineral oil [3].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 752–762, 2020.
https://doi.org/10.1007/978-3-030-31676-1_71
 Partial Discharge Characteristics of Palm Fatty Acid Ester 753

Dissolved gas analysis is an important technique for evaluating the insulating state
of power transformers. An understanding of the degradation of PFAE under electrical
stress is still limited. There are studies on PFAE using a needle-plane [4] or a spherical
metal particle electrode [5] that compared the partial discharge characteristics of PFAE
with those of mineral oil. Analyses of the gas production of PFAE under partial
discharge and breakdown have also been conducted, but the results were not consistent
for different researchers [6, 7]. According to the gas production results of PFAE in a
local thermal fault, PFAE without C-C double bond is more similar to mineral oil than
natural ester with C-C double bond under thermal stress [8, 9].
In this paper, the 120-hour partial discharge characteristics of PFAE using needle-
plane and creepage electrodes were studied and compared with the corresponding
characteristics of mineral oil. Then, the gas production characteristics of PFAE under
partial discharge and whether the commonly used DGA fault type diagnose methods
are applicable to PFAE were discussed.

2 Experimental Descriptions

2.1 Test Materials

PFAE from Lion Specialty Chemicals was used in this study, namely, PASTELL NEO,
and mineral oil was taken for a comparison. A pressboard with a 1.5 mm thickness was
used as the solid dielectric between electrodes.

2.2 Sample Processing

The oil was first filtered by Whatman 42 filter paper. Then, the oil was heated at 80 °C
in vacuum (below 100 Pa) for 48 h to remove gases and moisture. After the pre-
treatment, the relative humidities were controlled to less than 20% of the saturation
levels.
The pressboard was first dried for 24 h at 105 °C under vacuum and then
impregnated with the oil sample under vacuum below 100 Pa at 85 °C for 24 h.

2.3 Test Setup

The structure of the electrical ageing chamber is shown in Fig. 1. The cylinder test cell
is made of Perspex and has a capacity of 9 L. There are two pipes that connect to the
bottom of the chamber: one long pipe is used for vacuuming, and the other is used for
sampling and oiling. The electrode system is set at the centre of the test chamber shown
in Fig. 2. The electrode systems included the needle-plane and creepage discharge
electrodes. Medical needles are selected as the needle electrode in the two electrode
systems. The asymmetric half-taper-shaped needle tip can ensure close contact with the
pressboard surface. The tip radius is controlled to between 4 and 5 lm from the front
view. The plane electrode used is 30 mm in diameter in Fig. 2(a) and 20 mm in Fig. 2
(b). The gap distance is fixed at 40 mm.
754 Y. He et al.

Fig. 1. The electrical ageing chamber

R<5μm R<5μm
40mm
40mm

30mm 20mm
a) Needle-plane b) Creepage

Fig. 2. Electrode systems

The experimental setup is shown in Fig. 3. A resistor was connected between the
test cell and the supply transformer. An over-current relay at the low voltage side was
set to trip the power supply. A commercial Doble LDS-6 PD detector was connected in
series with a 150 pF coupling capacitor to detect the PDs.
To make the amount of gas generated by the insulating oil more obvious, 120 h
was selected as the test duration. Additionally, 30 kV and 37 kV were selected as the
test voltages for the needle-plane and creeping electrode systems, respectively. Partial
discharge data were recorded every 12 h, and the oil was taken every 24 h for the DGA
test.
 Partial Discharge Characteristics of Palm Fatty Acid Ester 755

T1 T2 R

Ck

AC Zx

Zpd

LDS-6 PD
system

Fig. 3. Test setup for the PD measurements

3 Results and Discussion

3.1 Partial Discharge Inception Voltage
The partial discharge inception voltage (PDIV) is identified as the applied voltage at
which repetitive partial discharges higher than twice the maximum allowed background
PD noise were first observed for the test object. Ten tests were repeated for each case to
obtain the scale parameters of a Weibull distribution as the results. The PDIVs for
needle-plane and creepage discharge in the two liquids are listed in Table 1.

Table 1. PDIV of different electrode systems

Electrode system PFAE/kV Mineral oil/kV
Needle-plane 26.6 26.4
Creepage 26.0 24.4

The results indicate that the presence of the pressboard surface does not change the
PDIV compared to an open gap, as the PDIVs of each oil were comparable. In addition,
the results show that discharges of the two liquids behave similarly in the early stage of
development.

3.2 Partial Discharge

Needle-Plane Electrode System. The PRPD pattern of the needle-plane electrode

system using two oils at the beginning of the test is shown in Fig. 4. The maximum PD
magnitude was 2162pC for PFAE and 907pC for mineral oil, so the discharge was
large in PFAE. The PRPD patterns for PFAE and mineral oil of the needle-plane
electrode system were similar.
Breakdown occurred in electrode using PFAE at 102 h after the test. Therefore, PD
data of this electrode only has the data of first 96 h. Figure 5 shows the maximum PD
756 Y. He et al.

a) PFAE b) Mineral oil

Fig. 4. PRPD pattern of the needle-plane electrode system with 30 kV

amplitudes and pulse repetition rate during the test. Maximum PD amplitudes for two
oils had a stable trend, and the pulse repetition rate decreased during the test. Maxi-
mum PD amplitude and pulse repetition rate in PFAE were higher. And the more
intense discharge in PFAE was consistent with the breakdown occurred in it. These
inferior PD properties of PFAE than mineral oil seem to be similar tendency of nature
ester with triester structure [10, 11].

2500 PFAE
mineral oil 103 PFAE
2000 mineral oil

1500 PFAE broke down at

102h after the test 102
Qmax/pC

n/min-1

1000 PFAE broke down

at 102h after the test
1
500 10

0
0 24 48 72 96 120 100
t/h 0 24 48 72 96 120
t/h

a) Maximum PD amplitudes b) Pulse repetition rate

Fig. 5. PD properties of the needle-plane electrode system at different times

Creepage Electrode System. The PRPD pattern of the creepage discharge electrode
system using two oils at the beginning of the test is shown in Fig. 6. The PRPD
patterns for PFAE and mineral oil were similar. The PRPD pattern shows that all the
PDs occurred in both positive and negative half cycles. The presence of the pressboard
promotes negative discharge. However, compared with the discharge of the needle-
plane, the PD amplitude decreased and pulse repetition rate increased. The maximum
PD amplitude was 27pC in PFAE and 52pC in mineral oil.
Figure 7 shows the maximum PD amplitudes and pulse repetition rate during the
tests. Although the maximum discharge in PFAE and mineral oil was different at the
beginning of the test, the maximum discharge in mineral oil decreased, which was close
to that of PFAE. In the later stage of the tests, despite the maximum discharge of the
 Partial Discharge Characteristics of Palm Fatty Acid Ester 757

a) PFAE b)Mineral oil

Fig. 6. PRPD pattern of the creepage electrode system with 37 kV

two oils fluctuating, the trend of the variation was basically stable. The discharge
repetition rates were comparable in the two oils, but the PFAE discharge repetition rate
fluctuated more significantly. Compared with the test results of the needle-plane
electrode, the discharge repetition rate in the creepage electrode was significantly
increased. This is because the placement of the paperboard parallel to the electric field
promotes the development of discharge.

60 PFAE PFAE
105
mineral oil mineral oil
50

40 104
Qmax/pC

n/min-1

30
103
20
102
10

0 101
0 24 48 72 96 120 0 24 48 72 96 120
t/h t/h

a) Maximum PD amplitudes b) Pulse repetition rate

Fig. 7. PD properties of the creepage electrode system at different times

3.3 DGA
Generated Combustible Gases. The results of DGA during the tests are shown in
Fig. 8.
The main gases produced during the tests were H2 and CO in both oils. These gases
are the typical gases of partial discharge. Unlike other natural ester, such as FR3, which
produces a significant amount of C2H6 under PD, the dominant hydrocarbon gas was
CH4 in PFAE. For needle-plane electrode system, the gas produced in two oils were
comparable. And for creepage electrode, mineral oil produced more gas than PFAE.
758 Y. He et al.

Dissolved Gas Concentration/µL·L-1 24h 24h

Dissolved Gas Concentration/µL·L-1

12 48h 12 48h
72h 72h
10 96h 10 96h
120h
8 8

6 6

4 4

2 2

0 0
H2 CH4 C2H6 C2H4 C2H2 CO H2 CH4 C2H6 C2H4 C2H2 CO
a) PFAE (needle-plane) b) Mineral oil (needle-plane)

Dissolved Gas Concentration/µL·L-1

Dissolved Gas Concentration/µL·L-1

16 24h 16 24h
48h 48h
14 14
72h 72h
12 12 96h
96h
10 120h 10 120h
8 8
6 6
4 4
2 2
0 0
H2 CH4 C2H6 C2H4 C2H2 CO H2 CH4 C2H6 C2H4 C2H2 CO

c) PFAE (creepage) d) Mineral oil (creepage)

Fig. 8. Dissolved gas concentrations at different times

The discharge power can be calculated from the sum total energy of all the PDs that
occurred in a test using Eqs. (1) and (2).

X
T
Pi ¼ uj qj =T ð1Þ
j

X
t
W¼ Pi
ti ð2Þ
i

where:
Pj = PD power during the PD record time T(W)
qj = Apparent charge of a PD pulse (C)
uj = Instantaneous voltage applied when the PD pulse occurs (V)
W = Total discharge energy of the test (J)
ti = Time between two PD records
Figure 9 shows the concentrations of the total dissolved combustible gases (TDCG)
with total discharge energy for PFAE and mineral oil for various durations. For the
same type of liquid, the overall trend is that the TDCG increases with increasing total
 Partial Discharge Characteristics of Palm Fatty Acid Ester 759

discharge energy. In addition, for the same discharge energy, mineral oil produced
more TDCG than PFAE.

25 30
PFAE PFAE
mineral oil 25 mineral oil
20
20

TDCG/µL·L-1
TDCG/µL·L-1

15
15
10 10

5 5

0
0
0 2 4 6 8 10 12 14 16 18
0 1 2 3 4 5 6 7 Discharge energy/J
Discharge energy/J

a) Needle-plane b) Creepage

Fig. 9. The correlation between TDCG and discharge energy

DGA Diagnosis. The methods for fault type diagnosis in ester insulating oil are the
Duval triangle [12] and Duval pentagon [13]. These two methods depend on the type of
ester. However, they do not provide a criterion for PFAE. According to Abdul Rajab’s
study, it can be concluded that monoesters without C-C double bond exhibit gas
production characteristics that are similar to mineral oil. Therefore, the Duval pentagon
for mineral oil and the IEC basic gas ratios are used to analyse the DGA results in the
tests.
The Duval pentagon method uses five gases (H2, CH4, C2H6, C2H4, and C2H2) to
diagnose the fault type. The relative percentage of each gas is plotted on its corre-
sponding gas axis, providing five different points that form an irregular polygon. The
centre (“centroid”) of the irregular polygon represents the DGA results of this example
in the pentagon configuration.
Figure 10 shows the diagnosis results of the fault types using the Duval pentagon.
For the needle-plane electrode system, the DGA results of PFAE fall in PD and then
fall in low-energy discharge (D1) areas. For mineral oil, the DGA results first fall in
stray gassing (S) area and then fall in PD. The last two-day results fall in D1. For
creepage discharge, the results of PFAE fall in D1 close to PD, and the results of
mineral oil fall in PD.
The IEC basic gas ratios interprets the DGA results by utilizing three basic gas
ratios, C2H2/C2H4, CH4/H2 and C2H4/C2H6.
Table 2 shows the IEC basic gas ratios diagnostic results for all the oil samples. The
results were compared to the results using the Duval pentagon.
760 Y. He et al.

Mineral oil 24h PFAE 24h

Mineral oil 48h PFAE 48h
Mineral oil 72h PFAE 72h
Mineral oil 96h PFAE 96h
Mineral oil 120h
a) Needle-plane

Mineral oil 24h PFAE 24h

Mineral oil 48h PFAE 48h
Mineral oil 72h PFAE 72h
Mineral oil 96h PFAE 96h
Mineral oil 120h PFAE 120h
b) Creepage

Fig. 10. Diagnosis of PD faults using the Duval pentagon for the oil samples

The IEC basic gas ratios method could not identify the fault type in all situations,
especially for the results of the first few days of needle-plane electrode system.
However, for the identified results, the fault type results were consistent with the Duval
pentagon results.
 Partial Discharge Characteristics of Palm Fatty Acid Ester 761

Table 2. Diagnosis results for the oil samples

Electrode system Liquid Time/hour Fault type
Duval pentagon IEC ratio methods
Needle-plane PFAE 24 PD /
48 D1 /
72 D1 /
96 D1 /
Mineral oil 24 S /
48 PD /
72 PD /
96 D1 /
120 D1 D1
Creepage PFAE 24 D1 D1
48 D1 D1
72 D1 D1
96 D1 D1
120 D1 D1
Mineral oil 24 S /
48 PD /
72 PD PD
96 PD PD
120 PD PD

4 Conclusions

PD measurements were performed with PFAE and mineral oil using needle-plane and
creepage electrode systems in this research. Furthermore, DGA was carried out for the
oil samples exhibiting PD. Some conclusions can be made as follows:
(1) The PDIVs of the two electrodes using two oils are comparable.
(2) For the needle-plane electrode, discharge in PFAE is larger. For creepage, dis-
charge is comparable in the two oils. The presence of the pressboard caused the
maximum PD amplitudes to decrease, while the pulse repetition rate increased.
(3) The main gases produced during the tests, H2, CO and CH4, are similar for PFAE
and mineral oil. Mineral oil produced more TDCG than PFAE for the same
discharge energy.
(4) The diagnostic results of the Duval pentagon and the IEC basic gas ratios method
are PD and D1 for mineral and PFAE, while IEC basic gas ratios method cannot
identify the fault type for all the cases.
762 Y. He et al.

References
1. Ruijin, L., et al.: A comparative study of physicochemical, dielectric and thermal properties
of pressboard insulation impregnated with natural ester and mineral oil. IEEE Trans.
Dielectr. Electr. Insul. 18(5), 1626–1637 (2011)
2. Ohki, Y.: News from Japan. IEEE Electr. Insul. Mag. 27(3), 55–57 (2011)
3. Takaaki, K., et al.: Analyses of electro-chemical characteristics of Palm Fatty Acid Esters as
insulating oil. In: 2008 IEEE International Conference on Dielectric Liquids, pp. 1–4 (2008)
4. Khayam, U., Suwarno, A.S., et al.: Partial discharge characteristics and dissolved gas
analysis of vegetable oil. In: Proceedings of 2014 International Symposium on Electrical
Insulating Materials, pp. 330–333 (2014)
5. Azmi, K., Zuhairi, A., Ishak, D., et al.: Partial discharge characteristics of spherical metal
particle in mineral oil and PFAE under AC voltage. In: 2017 IEEE Region 10 Conference on
TENCON 2017, pp. 2588–2593 (2017)
6. Ab Ghani, S.S., Muhamad, N.A.: Hydrocarbon gases dissolved in palm-based insulation oil
after experienced breakdown. In: 2015 IEEE 11th International Conference on the Properties
and Applications of Dielectric Materials (ICPADM), pp. 460–463 (2015)
7. Susilo, A., Muslim, J., Hikita, M., et al.: Comparative study of partial discharge
characteristics and dissolved gas analysis on palm-based oil as insulating material. In: The
2nd IEEE Conference on Power Engineering and Renewable Energy (ICPERE), pp. 232–
236 (2014)
8. Rajab, A., Tsuchie, M., Kozako, M., et al.: Low thermal fault gases of various natural
monoesters and comparison with mineral oil. IEEE Trans. Dielectr. Electr. Insul. 23(6),
3421–3428 (2017)
9. Rajab, A., Tsuchie, M., Kozako, M., et al.: Comparison of low thermal fault gases of various
fatty acid mono esters. In: Properties and Applications of Dielectric Materials, pp. 64–67
(2015)
10. Eberhardt, R., Muhr, H.M., Lick, W., et al.: Partial discharge behaviour of an alternative
insulating liquid compared to mineral oil. In: Power Modulator and High Voltage
Conference, pp. 1–4 (2010)
11. Wang, X.: Partial discharge behaviours and breakdown mechanisms of ester transformer
liquids under AC stress (2011)
12. Duval, M.: The duval triangle for load tap changers, non-mineral oils and low temperature
faults in transformers. IEEE Electr. Insul. Mag. 24(6), 22–29 (2008)
13. Duval, M., Lamarre, L.: The new Duval Pentagons available for DGA diagnosis in
transformers filled with mineral and ester oils. In: IEEE Electrical Insulation Conference,
pp. 279–281 (2017)
 Condition Assessment of Underground Cable
System Using Health Index and Conditional
Multiplying Factor

Tanachai Somsak, Thanapong Suwanasri(&),

and Cattareeya Suwanasri

King Mongkut’s University of Technology North Bangkok,

1518 Pracharat 1 Rd., Wongsawang, Bangsue, Bangkok 10800, Thailand
thanapong.s.epe@tggs-bangkok.org

Abstract. This paper presents condition assessment of underground cable

system for an independent power producer in Thailand. The underground cable
is very important in both underground transmission and distribution system.
Nowadays, the underground cable system is widely utilized to improve system
reliability and aesthetic view. When cables have been used for a period of time,
they become deteriorated and finally damaged because of electrical and
mechanical stresses, and etc. If the failures were not recognized and repaired in
time, the catastrophic failure could probably occur. Furthermore, this affects to
customer’s reliability and trustworthy. Therefore, the condition assessment and
health index calculation are needed to effectively maintain the system. The
underground cable system consists of power cable, termination, joint, manhole,
and duct bank. Normally, visual inspection and special testing with those
components are required for assessment. Without power interruption, on-line
testing consists of visual inspection, infrared thermography inspection and
partial discharge measurement. The off-line testing, which is performed during
maintenance, consists of polarization index measurement, dissipation factor
measurement, leakage current measurement and very low frequency withstand
test as required. These testing methods are considered together to evaluate the
actual condition of all components by using weighting and scoring method.
A conditional factor, considering service year, load current, failure frequency,
laying, and environment, is proposed to provide more accurate result in term of
practical usage condition. The web-application software is applied to setup
systematic database for technical data and test results. It has been developed for
processing condition assessment and facilitating maintenance activities.

Keywords: Conditional factor
Electrical test
Health index
Scoring and

weighting
Underground cable system

1 Introduction

Underground cable plays an important role in transmission and distribution system in

industrial area. It is used to transmit electricity from generation to customer. It keeps
increasing in number of installation to improve reliability and security of power supply.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 763–776, 2020.
https://doi.org/10.1007/978-3-030-31676-1_72
764 T. Somsak et al.

When cables have been used for a period of time, they become deteriorated and
damaged because of electrical, thermal, ambient and sometimes mechanical stresses.
The condition assessment and health index calculation are needed to effectively
maintain the system. However, the condition evaluation of underground cable system is
difficult and complex because of number of components, routing, and required testing
methods. Thus, this paper proposed state of the art for condition evaluation of
underground cable system applying weighting and scoring method as well as database
setup and data management by using web-application tool for real time evaluation, easy
access data management, and maintenance planning for underground cable system
according to its actual condition.

2 Data Structure

In Fig. 1, to evaluate condition of cable system, it needs to classify the major com-
ponents into five groups as cable, joint, termination, manhole, and duct bank based on
basic structure.

Generation Customer A
Substation Substation
Manhole Termination

MH 1 MH 2 MH 3
FD_1 FD_1
Cable
FD_2

Duct bank Customer B

Substation

FD_2
Joint
MH 4 MH 5

Fig. 1. An underground cable system basic structure.

2.1 Technical Data

The important technical data of five major components as cable, joint, termination,
manhole and duct bank can be described as follows. Firstly, the cable’s technical data
consists of feeder name, rated voltage, rated current, manufacturer, model type,
installation date, conductor size, number of joints, number of manhole, and cable
length of the entire route. Secondly, the joint’s technical data consists of joint name,
parent name, bonding type, manufacturer, model type, installation date and location.
Next, the termination’s technical data consists of termination name, installation type,
rated voltage, feeder name, location, manufacturer, model type, and installation date
[1]. Fourthly, the manhole’s technical data consists of manhole name, feeders in
manhole, location, manufacturer and model type. Finally, the duct bank’s technical data
consists of duct bank name, starting manhole, destination manhole, as well as numbers
of total duct, used duct, damage duct, and available duct.
 Condition Assessment of Underground Cable System 765

2.2 Measurement and Inspection Data

Regularly, there are two types of the underground cable system testing and inspection.
One is the on-line testing consisting Infrared thermography inspection, partial dis-
charge measurement and visual inspection. Another one is off-line testing composing of
VLF withstand testing, leakage current measurement, dissipation factor measurement,
polarization index measurement and partial discharge measurement. However, not all
of measurements can be applied to detect the abnormal condition of all major com-
ponents. For underground cable, the effective inspections are VLF withstand test,
leakage current measurement, dissipation factor measurement, polarization index
measurement, partial discharge measurement and visual inspection. For joint and ter-
mination, the effective inspections are VLF withstand test, dissipation factor mea-
surement, polarization index measurement, partial discharge measurement and visual
inspection [2]. For manhole and duct bank, there is an effective inspection just visual
inspection.

2.3 Environment and Operation Data

For getting the actual condition of the system, the environment and operation data are
significantly concerned. The data presents not only the situation of cable system, but
also a risk of power system. For the underground cable system, the factors affecting to
the system are operating temperature, failure frequency, age, percentage of load,
operating voltage, humidity, and cable laying [3, 4]. The underground cable system
with a few year in service, a less percentage of load current, a low failure frequency
with good installation practice and installed in cable trench should provide the higher
condition multiplying factor refereeing to better actual condition, or vice versa.

2.4 Web Application Development

In this work, the dynamic programing via web-application is the prime importance for
software tool development. It is used for a centralized database setup, data manage-
ment, graphic user interface, retrieving data for condition evaluation and result pre-
sentation. Database server is an important element to communicate with client or
worker for complying the request command, connecting to database, performing
operation according to a written script command to compute a result and sending the
result back to client. Then PHP program and JavaScript language are chosen to pro-
gram a website for condition assessment of underground cable system with the MySQL
as database program because of the open source and ability for self-learning of the
program developer. The developed software tool can be used via web-browser of
personal computer, tablet and smart phone. Then the client can access the database by
using internet from anywhere and anytime.
766 T. Somsak et al.

3 Testing Methods and Criteria

To provide the actual condition of the underground cable system, the system has to
recognize and collect data of defects by many types of methods. After that, it is
necessary to identify the criteria for each test method as well as environment and
operating condition in order to calculate the overall health index of cable system.

3.1 VLF Withstand Test

VLF withstand test or AC withstand test is the most elementary of all electrical tests on
cable insulation to identify that the system is pass or fail. It applies nominal voltage or
higher for a prescribed period of time to forecast a pre-breakdown phenomena of aged
underground cable, an erosion process associated with partial discharges that can
produce a breakdown of a weak insulation. Following the IEEE Std 400.2-2013 [3], the
typical withstand testing voltages range for a new cable is approximately, U0 to 2U0.
U0 is an operating voltage. However, to test the aged cable, the test might not
appropriate to inject much higher voltage than nominal voltage level. So, this work
would like to use a test voltage and time as shown in Table 1. The field VLF withstand
test applies frequencies between 0.01–1 Hz. It is applicable to all types of cable sys-
tems. For a more advanced diagnostic result, partial discharge and dielectric response
can be measured.

Table 1. Voltage and time of VLF withstand test

Nominal voltage U0/U (kV) Test voltage (kV) Time (min)
66.4/115 1.2 U0 = 99.6 60

3.2 Leakage Current Measurement

For leakage current measurement, it can be monitored and recorded during a VLF
withstand test. It is an additional test installed together with AC withstand test for
checking the insulation loss. The result of this method presents a degradation of cable
insulation. The extremely degraded insulation should provide a high value of leakage
current [4, 5]. Thus, this method can help to identify the level of cable insulation
degradation as other methods.

3.3 Dissipation Factor

The dissipation factor measurement (tan delta) is one of analytical method for under-
ground cable testing to determine a quality of cable insulation. Typically, the tan delta
needs to shut down power system and energize an external AC source to measure it [1,
2]. However, it would be easy if there is the AC withstand testing during cable system
maintenance because it can be measured when the AC voltage is applied [6].
 Condition Assessment of Underground Cable System 767

3.4 Partial Discharge

Partial discharge (PD) event is another way to investigate cable insulation degradation
defect. Normally, the PD events are always attended with emission of light, sound,
heat, and electromagnetic pulse. PD measurement, as shown in Fig. 2, can detect not
only the amplitude of PD event, but also can indicate discharge type and location in the
cable system. Although some system cannot find out any PD event, it does not mean
the system has no PD because such defect can initiate PD event in a case of only
temporary AC overvoltage or PD inception voltage is more than normal operating
voltage. Therefore, a combination of AC withstand test and PD measurement is nec-
essary to analyze the condition of cable insulation [5–8].

Fig. 2. On-site PD measurement.

3.5 Polarization Index

Polarization measurement is performed off-line by applying a low DC voltage to the
termination of cable system under the test for a duration. The charging current during
voltage application is called polarization current. Under measurement, the polarization
current decreases more slowly versus time representing a resistive current due to a
resistance of insulation system. Polarization index can be measured and analyzed in a
time domain to evaluate overall degradation of cable system in field by calculating the
ratio of measured insulation resistance at 10 min divided by the measured value at
1 min. The result of the measurement presents the water trees in extruded insulation by
the decrease of polarization index. Thus, this can be used to indicate the level of
insulation degradation in cable system [5, 7].

3.6 Visual Inspection

Visual inspection means something that can be detected by human sense such as
seeing, touching, reading, smelling and hearing at underground cable system compo-
nents. It is an organized examination or evaluation. However, the activities may
involve visualization, measurement, testing, and etc. applied to certain components.
768 T. Somsak et al.

Visual inspection items and criteria for underground cable system for the evaluation are
given in Table 3. Figure 3 shows an activity during manhole and cable visual
inspection.

Fig. 3. Manhole and cable visual inspection.

3.7 Infrared Thermography

Infrared thermography is equipment or method, which detects infrared energy emitted
from object, converts it to temperature, and displays image of temperature distribution
as shown in Fig. 4.

Fig. 4. Cable IR thermography inspection.

3.8 Testing Criteria and Scoring

After all of testing and inspection have been done, the tested data are completely
collected. They are necessary to change all tested results into scores of all criteria for
health index calculation of underground cable system. The criteria and their scores are
proposed according to standards [1–3] as given in Tables 2 and 3. Table 4 represents
criteria and their scores for condition factor (k) calculation used for more accuracy in
term of practical usage condition.
 Condition Assessment of Underground Cable System 769

Table 2. Electrical test criteria and scores

Measurement Score (Si)
Good (5) Moderate (3) Poor (0)
AC withstand Pass – Fail
Leakage current <20 nA 0.02–1 µA >1 µA
Dissipation factor <50 50–80 >80
Partial discharge Corona Surface Internal
Polarization index >2 2–1 <1
Infrared thermography No hot spot >2 °C rise >5 °C rise

Table 3. Visual inspection criteria and scores

Inspected item Score (Si)
Good (5) Moderate (3) Poor (0)
Manhole cover Normal Stained Disappear
Concrete racking poles Normal Stained Rusted
Steel structure poles Normal Stained Rusted
Hanger for supporting cable Normal Stained Rusted
Pillow insulator supporting cable Normal Cracked Broken
Cable rack Normal Stained Rusted
Cable tray Normal Stained Rusted
Cable jacket Normal – Abnormal
Spices joint Normal – Abnormal
Cable termination Normal Stained Bloated
Grounding system Normal High resistance Abnormal
Manhole situation Clean – Dirty

Table 4. Operating condition criteria and scores for condition factor (k)
Operating condition Score (Si)
Good (5) Moderate (3) Dangerous (0)
Operating temperature <35 °C 35–60 °C >60 °C
Failure frequency (in 5 years) 0 1 >1
Age in service <10 years 10–20 >20 years
Load percentage <50% 50–80% >80%
Operating voltage LV level MV level HV level
Humidity Dry Humid Flooded
Cable laying In trench – Direct buried
770 T. Somsak et al.

4 Health Index Calculation

The condition assessment procedure for underground cable system is shown in Fig. 5.
The health index of each major component (%HIcomponent) can be calculated from the
mentioned test methods by using the Weighting and Scoring Method (WSM). After
that, the health index of each major component and its assigned weight are used to
calculate the health index of underground cable system (%HIsystem) [9]. Finally, the
health index of underground cable system will be multiplied by the conditional factor
(k), which is observed during the preventive maintenance in order to obtain the final
overall health index (%HIoverall).

AC Withstand Testing
Leakage Current
Dissipation Factor Cable %HIcable
Partial Discharge
Polarization Index

AC Withstand Testing
Partial Discharge %HIjoint
Joint
Infrared Thermography %HIcable system
Visual Inspection

AC Withstand Testing
Partial Discharge
Termination %HItermination
Infrared Thermography
Visual Inspection
Multiplying

Visual Inspection Manhole %HImanhole

%HIoverall
Visual Inspection Duct Bank %HIduct bank

Operating Temperature
Failure Frequency
Age in Service Environment
Load Percentage &Operating Conditional Factor
Operating Voltage Condition
Humidity
Cable Laying

Fig. 5. Health index calculation diagram.

 Condition Assessment of Underground Cable System 771

The WSM in Eq. (1) together with scores (Si) and assigned weights (Wi) of testing
criteria are used to calculate the health index of major components of cable, joint,
terminator, manhole and duct bank as %HIcable, %HIjoint, %HItermination, %HImanhole
and %HIductbank, respectively.
PN
ð Si  W i Þ
%HIcomponent ¼ PN i¼1
   100 ð1Þ
i¼1 Smax;i  Wi

where Si is a score according to individual test result of each component, Smax,I is a

maximum score of individual test result, Wi is a weight of test method ith starting from
1 to 5 referring to the lowest to highest importance of testing as given in Tables 5 and
6, and N is a number of all testing methods of the component.
Then, the component health indices are used to calculate the system health index
(%HIsystem) of underground cable by using Eq. (2).
PM  
j¼1 %HIcomponent;j  %Wj
%HIsystem ¼ ð2Þ
100
where %HIcomponent,j is a calculated component health index jth, Wj is an important
weight of component jth as given in Table 9, and M is a number of all component
health indices.
Finally, the health index of underground cable system will be multiplied with a
condition factor (k) to obtain the overall health index (%HIoverall) as given in Eq. (3)
that shows an actual condition of cable system depending on practical usage condition
such as temperature, failure frequency, age, load percentage, operating voltage,
humidity, and cable laying as given in Table 9.

%HIoverall ¼ %HIsystem  k ð3Þ

where %HIsystem is the calculated cable system health index, and k is a conditional
factor of the cable system.
To simply present actual condition of underground cable system, the %HIoverall is
classified into three levels and displayed as traffic light colors as green, yellow and red
representing good, moderate and poor condition, respectively. The color bands are
practically easy to understand for the actual condition. Moreover, this work gives a
suggestion to improve the condition of underground cable system.

5 Result and Discussion

This section shows a procedure for %HIoverall calculation. Feeder FD_1 in Fig. 1 is
selected as an example. The procedure includes component health index calculation,
cable system health index calculation, conditional factor calculation, and finally,
overall health index calculation.
772 T. Somsak et al.

Step 1: Calculate the component health index of major components as cable, joint,
termination, manhole and duct bank. Using scoring technique, the all test results in
underground cable system must be compared with the values in Table 2. From Table 6,
scores together of important weights of all testing are used to calculate underground
cable health index using WSM in Eq. (1). The calculation is expressed below. Then,
the %HIcable is equal to 57.14%. Similarly, %HI of other components in FD_1 are also
calculated. The results are shown in Table 7.

%HIcable ¼ ðð50 
 5Þ þ ð3  3Þ þ ð3  4Þ þ ð5  4Þ þ ð3  3Þ þ ð5  2Þ
5Þ þ ð5  3Þ þ ð5  4Þ þ ð5  4Þ þ ð5  3Þ þ ð5  2Þ  100
%HIcable ¼ 57:14%

Step 2: From step 1, the worst health indices of all components in Table 7 are selected
to calculate a cable system health index.
Step 3: Calculate a underground cable system health index (%HIsystem) by using Eq. (2)
with the selected worst %HIcomponent in Table 8. By applying Eq. (2), the underground
cable system health index of FD_1 based on the weight percentage (%Wj) is calculated
and equal to 61.59%.

%HIsystem ¼ ð57:14  27Þ þ ð59:09  30Þ þ ð66:36

100
 30Þ þ ð69:09  8Þ þ ð60  5Þ

%HIsystem ¼ 61:59%

Step 4: To calculate a conditional k-factor, Eq. (1) is used together with condition data
in Table 9 converted into their scores. The k-factor needs to normalize into a maximum
value of 1. Then, k-factor is calculated and equal to 0.7364.

k ¼ ð5  3Þ þ ð3  5Þ5þ ðð33 þ55Þ þþ ð55þ 44Þþþ1ð0þ31þÞ þ2Þð3  3Þ þ ð5  2Þ  100

100
k ¼ 0:7364

Table 5. Underground cable health index of FD_1

Component Measurement Score (Si) Weight (Wi) %HIcable
Cable AC withstand 0 5 57.14
Leakage current 3 3
Dissipation factor 3 4
Partial discharge 5 4
Polarization index 3 3
Visual inspection 5 2
 Condition Assessment of Underground Cable System 773

Table 6. The worst %HIcomponent in FD_1

Component Testing method Si Wi %HIcomponent
Cable AC withstand 0 5 57.14
Leakage current 3 3
Dissipation factor 3 4
Partial discharge 5 4
Polarization index 3 3
Visual inspection 5 2
Joint AC withstand 0 5 59.09
Dissipation factor 3 3
Partial discharge 5 4
Polarization index 3 3
Infrared thermography 3 4
Visual inspection 5 3
Termination AC withstand 0 5 66.36
Dissipation factor 3 3
Partial discharge 5 4
Polarization index 3 3
Infrared thermography 5 4
Visual inspection 5 3
Manhole Visual inspection 3.14 5 69.09
Duct bank Visual inspection 3 5 60

Table 7. All component health indices of FD_1

Section Cable Joint Termination Manhole Duct bank
Gen.–MH.1 57.14 – 66.36 – –
MH.1–MH.2 57.14 66.36 – 96.36 60
MH.2–MH.3 57.14 59.09 – 69.09 100
MH.3–Cus.A 57.14 81.81 75.45 100 86.67
Worst %HIcomponent 57.14 59.09 66.36 69.09 60

Table 8. %HIsystem of FD_1

Component %HIcomponent %Wj %HIsystem
%HIcable 57.14 27 61.59
%HIjoint 59.09 30
%HItermination 66.36 30
%HImanhole 69.09 8
%HIduct bank 60 5
774 T. Somsak et al.

Table 9. Conditional factor of FD_1

Operating condition Condition Si Wi k-factor
Operating temperature 27 °C 5 3 Conditional
Failure frequency 1 3 5 k-factor = 0.7364
Age in service 15 years 3 5
Load percentage 50% 5 4
Operating voltage 115 kV 0 1
Humidity High-humidity 3 3
Cable laying In trench 5 2

Step 5: Calculate the %HIoverall of underground cable by multiplying the calculated

cable system health index (%HIsystem) with the condition k-factor in Table 10. Finally,
the %HIoverall of FD_1 is equal to 45.35%.

%HIoverall ¼ 61:59  0:7364 ¼ 45:35%

This value is compared with the ranges of %HIoverall Table 11. The result shows
that the system has poor condition. Thus a very close investigation and maintenance are
required while planning and acting for repairing or replacement are also needed. The
developed software shows the result of overall health index via the web-application as
presented in Fig. 6.

Table 10. Overall health index (%HIoverall) of FD_1

Variable Value %HIoverall
%HIsystem 61.59 45.35
Conditional factor 0.7364

Table 11. The ranges of overall health index of underground cable system
%HIoverall Condition Action Color
code
81–100 Good Normal maintenance Green
61–80 Moderate Frequent rounds of investigation/maintenance, special Yellow
tests needed
0–60 Poor Very close investigation and maintenance while Red
planning and acting for repairing/replacement

Fig. 6. Overall health index display.

 Condition Assessment of Underground Cable System 775

Three cases are evaluated. The results show poor conditions. The first case pre-
sented as calculating example encounters AC withstand failure with %HIoverall of FD_1
equal to 45.35%. The second case encounters surface discharge at its termination in
destination substation side resulting to the bloated termination jacket. The %HIoverall of
FD_2 is equal to 57.12%. Lastly, the internal partial discharge was found in the third
system, which the operating condition is heavy. The %HIoverall of FD_3 is equal to
48.06%. After solving all problems, the %HIoverall of them become good condition with
normal maintenance.

6 Conclusions

Condition assessment using WSM is proposed to determine health index of under-

ground cable system consisting of cable, joint, termination, manhole, and duct bank.
The calculation starts with component health index calculation, cable system health
index calculation, conditional k-factor calculation, and lastly, overall health index
calculation. Scoring is obtained from the test results while weighting is defined from
importance of test methods including VLF withstand test, leakage current measure-
ment, dissipation factor, polarization index, partial discharge measurement, infrared
thermography, and visual inspection. Finally, the underground system conditions as
good, moderate and bad are differentiated from three ranges of overall health index and
shown in form of traffic lights as green, yellow and red, respectively. The maintenance
strategies for all conditions are mentioned. A web-application is developed as an
effective tool for user-friendly usage during on-site inspection and maintenance, col-
lecting important data, calculating all steps, showing results, and efficiently manage
maintenance task by ranking the conditions of equipment with less time consuming in
condition evaluation for underground cable system.

References
1. IEEE Std 1511.1-2010: IEEE Guide for Investigating and Analyzing Shielded Power Cable
Failures on Systems Rated 5 kV Through 46 kV (2010)
2. IEEE Std 400-2012: IEEE Guide for Field Testing and Evaluation of the Insulation of
Shielded Power Cable Systems Rated 5 kV and Above (2012)
3. IEEE Std 400.2-2013: IEEE Guide for Field Testing of Shielded Power Cable Systems Using
Very Low Frequency (VLF) (2013)
4. Cherukupalli, S., Buchholz, V., Colwell, M., Crine, J., Keefe, R.J.: Condition assessment of
distribution PILC cables from electrical, chemical, and dielectric measurements. IEEE Electr.
Insul. Mag. 22(4), 6–12 (2004)
5. Kurihara, T., Sato, T., Homma, K.: Insulation characteristics of dry-cured and extruded three-
layer 6.6 kV removed XLPE cables with additional accelerated water-tree degradation. In:
2017 International Symposium on Electrical Insulating Materials, pp. 110–113 (2017)
6. Smit, J.J., Gulski, E.: Advanced condition assessment of high voltage power cables. In: 2005
International Symposium on Electrical Insulating Materials, Kitakyushu, Japan, pp. 869–872
(2005)
776 T. Somsak et al.

7. McDermid, W., Bromley, J.C.: Condition assessment of service aged XLPE distribution
cable. In: Electrical Insulation Conference and Electrical Manufacturing and Coil Winding
Conference, Ohio, USA, pp. 349–352 (2001)
8. Montanari, G.C., Cavallini, A., Puletti, F.: A new approach to partial discharge testing of HV
cable systems. IEEE Electr. Insul. Mag. 22, 14–23 (2006)
9. Naderian, A., Cress, S., Piercy, R., Wang, F., Service, J.: An approach to determine the health
index of power transformers. In: The 2008 IEEE International Symposium on Electrical
Insulation, Vancouver, BC, Canada, pp. 192–196 (2008)
 Comparison of a Thermally Accelerated Aged
Oil-Paper Insulation of an Open
and Hermetically Sealed System

T. Münster1(&), T. Kinkeldey1, P. Werle1, K. Hämel2, and J. Preusel2

1
Division of High Voltage Engineering and Asset Management,
Institute of Electrical Power Systems, Schering-Institute,
Leibniz Universität Hannover, Callinstr. 25A, 30167 Hannover, Germany
muenster@ifes.uni-hannover.de
2
GRIDINSPECT GmbH, Mühlhof 3a, 36325 Feldatal, Germany

Abstract. This study investigates differences in the ageing behaviour of an oil-

paper insulation system in an open and hermetically closed system concerning
relevant ageing markers of three different insulating liquids. In this context, the
correlation of the ageing markers measured in the liquid and the current condition
of the paper insulation is investigated. Accordingly, paper samples are impreg-
nated with a non-inhibited and two different inhibited insulating oils having
different oil components as well as different inhibitor contents, which are cur-
rently used in field. Although different papers from the past have shown inves-
tigations on ageing of oil-paper systems [1, 2] usually oil was used, which is not
longer available today and furthermore no simultaneous comparative study was
carried out in the past, as it is presented in this paper. The ageing for both systems
takes place over a period of 15 weeks at the temperature of 130 °C. These results
are also supposed to be used to determine the current paper condition using the
values of ageing markers of the oil without the furan content. The investigation
shows that the type of used system and the oil types influence the ageing process
of the insulation system and consequently the production of respective param-
eters. Moreover, there are differences depending on the system, which must be
taken into account in the indirect state analysis of the insulating paper, since the
ageing process in the closed system is slower than in the open system due to the
lack of atmospheric oxygen and protection against ambient humidity.

Keywords: Ageing
Transformer
Oil
Paper
Degree of polymerization

Oxidation
Temperature
Insulation degradation
Insulation life

1 Introduction

Even nowadays, the oil-paper insulation is still one of the most important insulation
systems for high-voltage units, such as power transformers, which can be built as a free
breathing or hermetically closed system. However, the insulation is subjected to con-
tinuous ageing due to thermal, electrical, mechanical and chemical influences [2]. In fact,
the rate of ageing depends on many factors, e.g. the location and load of the transformer.
As the paper insulation ages, it becomes brittle and fragile over time resulting in reduced

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 777–787, 2020.
https://doi.org/10.1007/978-3-030-31676-1_73
778 T. Münster et al.

mechanical strength. This increases the probability of failure of the transformer and
therefore the condition of the paper insulation is a limiting factor of the lifetime [3].
Currently, the measurement of the DP-value is the used direct method for evaluating the
current paper condition. For this, a paper sample is required, which can only be obtained
from a transformer under very high costs and efforts [4]. However, samples of the
insulating liquid are relatively easy to obtain, hence they are often used to evaluate the
condition of the insulation system. For this purpose, various ageing markers of the
insulation liquid are measured to achieve an indirect assessment of the paper condition.
Furthermore, in order to clarify the influence of the different systems, this study focuses
on the differences in ageing behaviour and its effects on the measured ageing markers.

2 Oil-Paper Samples

For both investigated systems Grade K Insulation Paper is used, which is impregnated
with three different insulating liquids. In order to study the influence of the type of
insulating liquid, three different types of today used conventional oils are investigated.
Oil A is a mineral oil based uninhibited electrical insulating oil, while oil B and oil C
are two different inhibited oils with different compositions, inhibitor contents and
manufacturing processes. For the study the paper is cut into 10 cm long strips, which
can be seen in Fig. 1. Every paper strip has a width of 25 mm and a nominal thickness
of 0.1 mm.

Fig. 1. Untreated Kraft paper with a length of 10 cm

For the preparation of comparable samples with a same initial state, the samples are
prepared according to the same method. Firstly, the samples are dried in a stainless
steel vessel in a vacuum chamber, by a heat-vacuum-process, at a pressure of 0.1 mbar
and a temperature of 80 °C. The vessel has separated compartments with 160 g
untreated insulation paper in each compartment. The drying time is about 48 h. After
the drying process, dried and degassed insulating oil is added into the chamber under
vacuum, so that the paper samples are covered with the insulating liquid. Subsequently,
a waiting time of 10 min passed by, then the vacuum oven is ventilated. In the initial
state, the oil has a residual moisture content under 5 ppm determined at 20 °C by the
Karl-Fischer-method. The impregnation process takes place at 80 °C for further 48 h.
After the impregnation process was completed, the paper from every compartment is
 Comparison of a Thermally Accelerated Aged Oil-Paper Insulation 779

added into a chemically inert conical flask, which is filled up additionally with 1600 g
of the insulating liquid used for the impregnation, hence a paper-oil ratio of 1:10 is
reached. Each test sample also has a copper tube as catalyst, with a weight of 66 g, a
length of 10 cm and a wall thickness of 0.1 mm. For every insulating liquid and
system, 10 samples are prepared, so that a total quantity of 30 samples is reached for
every system and 60 in total.

3 Implementation of Ageing

In order to simulate the real ageing time of a transformer, the oil-paper insulation is
aged thermally accelerated. For the ageing process, the open and hermetically closed
flasks, as shown in Fig. 2 are aged at a temperature of 130 °C in a convection oven to
guarantee a homogenous temperature at every point, over a period of 15 weeks. The
open system is an unsealed Erlenmeyer flask, which allows permanent contact with the
ambient air. In order to simulate the hermetically sealed system, the flasks are sealed
with silicone rubber plugs, which means that the insulation is no longer in direct
contact with the ambient air and only a small gas cushion remains, which does not
influence the results.

Fig. 2. Different systems for the ageing process

Table 1. Examination times of the samples

Removal Ageing Name of the sample Name of the sample
time time in h (open system) (closed system)
Start 0 A, B, C 0 AG, BG, CG 0
Week 1 168 A, B, C 1 AG, BG, CG 1
Week 2 336 A, B, C 2 AG, BG, CG 2
Week 3 504 A, B, C 3 AG, BG, CG 3
Week 4 672 A, B, C 4 AG, BG, CG 4
Week 6 1008 A, B, C 6 AG, BG, CG 6
Week 8 1344 A, B, C 8 AG, BG, CG 8
Week 10 1680 A, B, C 10 AG, BG, CG 10
Week 12 2016 A, B, C 12 AG, BG, CG 12
Week 15 2520 A, B, C 15 AG, BG, CG 15
780 T. Münster et al.

With regard to the expectation that larger changes will occur at the beginning of
ageing rather than at the end, the examination times in Table 1 are chosen [5, 6]. For
the evaluation, one flask from each insulating liquid and system is taken out at the same
removal time.

4 Results

In Fig. 3 the colour change of the three aged insulating liquids can be seen. The
number on the samples represents the ageing time of the samples in weeks. It is evident
that the oils in the open system have a stronger colour change, except oil B. Also oil C
for the closed system appears to reach earlier a colour change compared to the open
system. Here the colour turns yellow after 3–4 weeks and for the open system it seems
to reach the yellow colour after 8 weeks. The stronger change for oil A with the open
system is caused by the stronger influence of oxidation processes due to the permanent
contact with the ambient air [7, 8]. The colour change for oil B and C is more delayed
because of the inhibitor, which decreased the influence of oxidation processes on the
ageing behaviour.

A0 A15 AG0 AG15

B0 B15 BG0 BG15

C0 C15 CG0 CG15

Fig. 3. Comparison of the aged oil samples (open system left/closed system right)

Figure 4 shows the colour change of the paper samples. It shows that a stronger
colour change can be observed with the open system, due to the aforementioned reason,
for all the paper samples.
Since the paper impregnated with oil A contained very dark samples, the paper was
divided into two groups for both systems. The first group consisted of light stripes,
which make up the bulk, and dark stripes, which were present in small quantities. It can
also be seen here that the paper strips for the open system are significantly darker.
 Comparison of a Thermally Accelerated Aged Oil-Paper Insulation 781

A0 A15 AG0 AG15

A0D A15D AG0D AG15D

B0 B15 BG0 BG15

C0 C15 CG0 CG15

Fig. 4. Comparison of the aged paper samples (open system left/closed system right)

Some selected values of the DP of the paper insulation during the degradation for
the two different systems, which are measured according to IEC 60450 in an external
accredited laboratory, can be seen in Fig. 5. The degree of polymerization (DP) is the
average number of glucose units in each chain [9]. The values for 0 h, 1344 h (8
weeks) and 2520 h (15 weeks) are shown, thus values can be compared at the
beginning, middle and end of the ageing study. For oil A, the values for the light paper
are displayed, as this made up the majority of the paper insulation. For the dark paper
samples, slightly reduced values could be determined, whereby the exact gradients can
be found in the papers [5] and [6] as well as the other DP-values.

Fig. 5. Comparison of the DP-value of the open system and closed system
782 T. Münster et al.

Very interesting differences can be observed here. It can also be observed that the
paper ages faster in the open system than in the closed system, which is consistent with
previous investigations [10–13]. But at the end of the investigation this does not apply
to the paper, which is impregnated with oil A. The light-coloured paper has a DP-value
of 389 in the open system for oil A after 15 weeks and a value of 397 for the closed
system, so the end ageing process was similar. On the other hand, the inhibitor
influences the ageing of the closed system, which can be clearly seen in the inhibited
oils B and C. The final value of the DP-value after 15 weeks for the paper with oil B is
237 for the open system and 509 for the closed system. For oil C, the values are 226
and 364. Thus, an influence of the inhibitor on the aging behaviour of the paper
samples can be seen in various systems. Due to the lack of oxygen in the closed system,
it can therefore be assumed that oxidation processes are inhibited in this system and
that the ageing rate is lower. It also appears that the colour of the oil is not directly
related to the DP-value of the paper (see Fig. 3), while the colour of the paper seems to
be related to the DP-value.
For a better overview Fig. 6 shows the compensation curves of the interfacial
tension (IFT), measured according to ASTM D971-12.

Fig. 6. Comparison of the IFT of the open system and closed system

First, it should also be noted here that the open system shows a greater decrease of
IFT values because of the greater influence of oxidation processes. For oil A and B the
value decreases below 20 mN/m and for oil C under 30 mN/m. With the closed system
the end values for oil A and B are between 30–35 mN/m and for oil C 40 mN/m. In
both systems a quite good correlation with the DP-value can be seen, whereby the
correlation with the closed system seems to be rather linear, certainly the correlation for
the open system for oil B and C is to be regarded more as an exponential function. It
becomes clear that when ageing markers are used to predict the current condition of the
paper insulation or the DP-value, a distinction must be made between a free-breathing
system and a closed system, since different values are recognizable for the IFT for the
 Comparison of a Thermally Accelerated Aged Oil-Paper Insulation 783

same DP-values. The initial conditions are the same for both systems and the values are
between 40 and 50 mN/m.
The trends of the acidity in the insulating liquid are shown in Fig. 7. Compensation
curves are also shown here for a better comparability.

Fig. 7. Comparison of the acid content of the open system and closed system

As it can be clearly seen, a higher acid content is achieved during ageing in the
open system. The values for oil A and B, at 0.3 mg KOH/g at the end of the ageing are
more than twice as high as those for the closed system, whereby a value of about
0.15 mg KOH/g is achieved at the end. It is interesting to note that the final value for
oil C in both systems is slightly above 0.1 mg KOH/g, although this value is reached
earlier in the closed system, since a value of 0.1 mg KOH/g is reached here with a DP-
value of slightly below 500. In the open system, this value is reached between a DP-
value of 300 - 200. These results can also be related to the paper ageing. In [12] it was
shown that different acids in oil, which are produced during the ageing, have a negative
influence on paper ageing and they will accelerate the degradation. As with the IFT,
this ageing marker has a good correlation with the DP-value, as the correlation is
relatively linear. For oil C this correlation seems to be exponential for both systems,
whereas for oil B in the open system the correlation is more exponential. However, also
in using this ageing marker for conclusions about the current state of the paper insu-
lation, it must be taken into account which system is involved. Thus, different acid
contents are also present here with the same values of the DP-value. Obviously the
permanent contact with the ambient air in the open system causes a greater influence by
oxidation processes, resulting in the formation of a larger amount of acids dissolved in
both the oil and the paper [9].
Figure 8 shows the course for the gases CO and CO2 as well as the ratios for CO2/
CO and O2/N2 measured with DGA. As it can be seen from the graphs, significantly
higher gas concentrations of CO and CO2 have been measured in the closed system,
which is similar in scope to [10].
784 T. Münster et al.

Fig. 8. Course of CO, CO2, CO2/CO and O2/N2 for the open system and closed system
 Comparison of a Thermally Accelerated Aged Oil-Paper Insulation 785

Fig. 8. (continued)

For the gas content of CO2, there are almost tenfold concentration values compared
to the open system. The values for CO are also nearly twice as high. It is also inter-
esting to note that in both systems a maximum is reached for CO2 during the ageing
period. With the closed system, however, this maximum is achieved at a later time.
According to [14], the CO2/CO ratio for normal paper ageing is between 3 and 10. The
investigation showed that values below 3 were only achieved for the open system,
whereas values above 10 were partially achieved for the closed system. Therefore, a
clear difference is visible here, the values must be considered with regard to the system,
which is used. According to [14], where a ratio of O2/N2 below 0.3 gives information
about a greater influence of oxidation processes, this threshold seems to be equal for
both systems.

Fig. 9. Comparison of the relative permittivity for the open system and closed system at
# = 25 °C
786 T. Münster et al.

The course of the relative permittivity er at a temperature of # = 25 °C can be seen

in Fig. 9. While the value for er remains approximately constant over the entire range of
the DP-value with the closed system, the value changes with the open system and a
correlation is recognizable for oil A and B, since a continuous increase is observable,
while oil C changes also here not over the DP-value. Further investigations have shown
that the gradient is independent of the water content in the open system. For this
purpose, all samples were subsequently brought to the same moisture content in a
climatic chamber at 40 °C for two weeks at a relative humidity of 40% and measured
again. The gradient of the curve was retained.
Figure 10 shows the course of the viscosity η for the temperature of # = 10 °C
over the DP-value for the different insulating liquids and systems. It can also be seen
that the open system shows a stronger change in values. For oil C, no change in
viscosity over ageing time and thus over the DP-value can be observed in both systems.
This parameter shows differences according to the used system even for the same DP-
value. Thus, for example, at a DP-value of 400 oil A has a viscosity value of about
29 cP for the closed system and a value of about 32 cP for the open system.

Fig. 10. Comparison of the viscosity for the open system and closed system at # = 10 °C

5 Conclusions

It was shown that different transformer systems and types of today used insulating
liquids and their composition cause different ageing behaviours and different visual
appearance of the oil and paper. Thus, it could be shown that for indirect analysis of the
insulating paper conditions the system (closed or free breathing), used in each case,
must be also considered. The different systems showed different values for the
investigated parameters such as acid content, interfacial tension etc. for the same DP-
values of the paper.
It has also been shown that the type of insulating fluid, which is used, has an
influence on the ageing behaviour. Thus, the inhibited liquids have better performance
 Comparison of a Thermally Accelerated Aged Oil-Paper Insulation 787

in the investigated parameters. It is also interesting to note that after an ageing period of
15 weeks, similar values were achieved for oil A in both the open and closed systems.
In addition, oil C has the best ageing properties in the oil parameters, although the DP-
value of the paper is lowest in both systems.

Acknowledgments. The authors would like to express their gratitude to GRIDINSPECT GmbH
and AiF/ZIM for the financial support as well as Weidmann Electrical Technology AG for the
support with insulation materials and Analysen Service GmbH for the analysis.

References
1. Lundgaard, L.E., Hanse, W., Linhjell, D., Painter, T.J.: Aging of oil-impregnated paper in
power transformers. IEEE Trans. Power Deliv. 19, 230–239 (2004)
2. Bakar, N.A., Abu-Siada, A., Islam, S.: A review on chemical diagnosis techniques for
transformer paper insulation degradation. In: Australasian Universities Power Engineering
Conference, AUPEC 2013, Hobart, TAS, Australia, 29 September–3 October 2013
3. Abu-Elanien, A.E.B., Salama, M.M.A.: Survey on the transformer condition monitoring. In:
Large Engineering Systems Conference, pp. 187–191 (2007)
4. Teymouri, A., Vahidi, B.: CO2/CO concentration ratio: a complementary method for degree
of polymerization of power transformer paper insulation. IEEE Electr. Insul. Mag. 33, 24–30
(2017)
5. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel, J.: Investigation on the
accelerated ageing behaviour of oil-paper-insulation using different insulating oils. VDE
Fachtagung, Hochspannungstechnik, Berlin, Germany (2018)
6. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel, J.: Investigation on the ageing
parameters of a thermally accelerated aged paper-oil-insulation in a hermetically sealed
system. In: Condition Monitoring and Diagnosis, CMD, Perth, Australia (2018)
7. Liapis, I., Danikas, M.G.: A study of parameters affecting the ageing of transformer oil in
distribution transformers. In: IEEE International Conference on Dielectric Liquids,
Trondheim, Norway (2011)
8. Technical Brochures, Reference: 738: Ageing of Liquid Impregnated Cellulose for Power
Transformers, CIGRE, August 2018
9. IEC 60450: Measurement of the average viscometric degree of polymerization of new and
aged cellulosic electrically insulating materials (2004)
10. Kachler, A.J., Höhlein, I.: Aging of cellulose at transformer service temperatures. Part 1:
influence of type of oil and air on the degree of polymerization of pressboard, dissolved
gases, and furanic compounds in oil. IEEE Electr. Insul. Mag. 21, 15–21 (2005)
11. Shroff, D.H., Stannet, A.W.: A review of paper aging in power transformers. Gener. Transm.
Distrib. IEE Proc. C 132, 312–319 (1985)
12. Lelekakis, N., Martin, D., Wijaya, J.: Ageing rate of paper insulation used in power
transformers part 2: oil/paper system with medium and high oxygen concentration. IEEE
Trans. Dielectr. Electr. Insul. 19, 2009–2018 (2012)
13. Lundgaard, L.E., Hansen, W., Ingebrigtsen, S., Linhjell, D., Dahlund, M.: Aging of kraft
paper by acid catalyzed hydrolysis. In: ICDL: 2005 IEEE International Conference on
Dielectric Liquids, New York, 2005
14. IEC 60599: Mineral oil-filled electrical equipment in service – Guidance on the
interpretation of dissolved and free gases analysis (2015)
 Investigation on the Accelerated Ageing
Behaviour of an Oil-Paper-Insulation
Using Different Weight Ratios

T. Münster1(&), T. Kinkeldey1, P. Werle1, K. Hämel2, and J. Preusel2

1
Institute of Electrical Power Systems, Division of High Voltage Engineering
and Asset Management, Schering-Institute, Leibniz Universität Hannover,
Callinstr. 25A, 30167 Hannover, Germany
muenster@ifes.uni-hannover.de
2
GRIDINSPECT GmbH, Mühlhof 3a, 36325 Feldatal, Germany

Abstract. This research presents a comparative study of the ageing process of

an oil-paper-insulation using Kraft paper. The main goal of this research is the
comparison of the ageing behaviour of the insulation with different weight ratios
and the influence on pertinent ageing markers, which are used to estimate the
remaining lifetime of an oil-paper insulated transformer. Therefore, an investi-
gation has been conducted to measure different parameters of a non-inhibited
insulating oil during a long-term thermal stress over 15 weeks at a temperature
of 130 °C to examine the influence of the ratio on the formation of these
markers. The higher temperature compared to the normal operation temperature
of a transformer is used to accelerate the thermal ageing of the insulation
material in order to simulate the lifetime of a transformer. In addition to the
ageing of pure paper and pure insulating oil, the paper-oil weight ratios 1:5, 1:10
and 1:20 will be investigated. The aim is to study how the respective insulating
medium ages individually and to determine the ageing behaviour of the insu-
lating materials when used together as a mixed dielectric, as it can be found in
real transformers. Therefore, a typically used uninhibited liquid based on min-
eral oil is used as insulating liquid. Ageing markers such as water content,
acidity as well as DGA analyses and other parameters are investigated. Fur-
thermore, the DP-value of the paper samples is also measured. For this, an open
system is reproduced, so that a permanent contact with the ambient air is given
to simulate a free breathing transformer. The main aim of this study is to
investigate the influence of the various weight ratios and how it influences the
possibility of indirectly determining the condition of the insulating paper. For
this purpose, ageing markers are to be investigated. The obtained results showed
that the weight ratio has no influence on the ageing rate of paper but on some
ageing markers.

Keywords: Ageing
Transformer
Oil
Paper
Degree of polymerization

Oxidation
Temperature
Insulation degradation
Insulation life
Ratio

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 788–798, 2020.
https://doi.org/10.1007/978-3-030-31676-1_74
 Investigation on the Accelerated Ageing Behaviour of an Oil-Paper-Insulation 789

1 Introduction

Even today the oil-paper-insulation is still one of the most important insulation systems
for high-voltage components like transformers or reactors. This kind of dielectric
medium undergoes permanent stresses during operation in the form of thermal, elec-
trical, mechanical and chemical stresses [1]. Accordingly, the insulation is exposed to a
slow permanent ageing process, which makes it more vulnerable to different failures.
However, the ageing time of every transformer is different and depends on several
factors, e.g. the location or the amount of load. Therefore, the service age of these
assets should not be simply seen as a function of the years of operation, but rather
influenced by the insulation end-of-life [2]. In general, the pyrolysis has a degrading
effect on the cellulose of the paper resulting in a decrease of the mechanical strength
and consequently, its insulation quality. Currently, the method of measuring the degree
of polymerization (DP-value) is the mostly used direct technique to determine the
condition of the paper insulation. Usually, a DP-value in a range below 200 is con-
sidered as end-of-life criteria for a transformer, which can be seen as a value based on
experience. At this value, the paper is fragile and brittle even with reduced mechanical
stresses [3]. However, for this measurement a paper sample is needed, for which the
transformer has to be de-energized and opened with consecutive very high costs [4].
Therefore, in order to reach an optimal lifetime of the transformer and to give rec-
ommendations about their future life expectation, it is necessary to observe the
degradation of the cellulose during the operation time. Usually, for the determination of
the condition of the paper insulation, oil samples are taken to measure different ageing
markers, because oil samples can be easily taken.

2 Oil-Paper Samples

For the investigation samples with different weight ratios are prepared. For this purpose
Kraft paper and an uninhibited insulating liquid based on mineral oil is used. The paper
used for this study is Grade K Insulation Paper as seen in Fig. 1.

Fig. 1. Untreated Grade K Insulation Paper

790 T. Münster et al.

In order to obtain comparable samples for the ageing process, the paper samples and
the insulating oil are prepared. First, the insulating paper with a thickness of 0.1 mm and
a width of 25 mm is cut into strips with a length of 10 cm. The paper strips are then
placed in a stainless steel container, which is divided into individual segments. The
paper is then put into the individual segments for drying and degassing. The stainless
steel container is placed in a heatable vacuum oven for drying and impregnation. At a
temperature of 80 °C and a pressure of 0.1 mbar, the paper is finally dried and degassed
over a period of 48 h in a heat-vacuum process. In the meantime, the insulating oil is
dried and degassed, which is also achieved under heat and vacuum, by means of an oil
treatment system. After the drying process of the paper, the dried insulating oil, which
after processing has a moisture content of less than 5 ppm, is put into the vacuum oven
under vacuum. The moisture content is determined at a temperature of 20 °C using Karl
Fischer titration. After the oil filling and a 10 min waiting time the vacuum oven is
ventilated with dried air. Erlenmeyer flasks made of borosilicate glass, which are
chemically inert and therefore do not influence the formation of ageing markers, are
used for the accelerated ageing process. To investigate the ageing behaviour of pure oil
without the presence of paper, 10 flasks are filled with 1600 g insulating oil each. In
order to study the behaviour of different paper to oil weight ratios, 10 flasks for each
ratio are also filled with 1600 g oil and insulating paper is added according to the weight
ratio. Thus, for the paper to oil weight ratio 1:5 320 g paper is added to the flasks. For
the ratio 1:10 160 g is added and for the ratio 1:20 80 g of the Kraft paper is added.
Additionally, to investigate the ageing behaviour of pure paper without oil, 160 g of
Kraft paper is aged in an open borosilicate glass vessel. To simulate the lifetime of a real
transformer, the samples are thermally accelerated aged by means of increased tem-
perature. The prepared samples are placed in a convection oven to ensure uniform
temperature distribution. The ageing is finally carried out at a temperature of 130 °C.
During ageing, the vessels are kept open to simulate a free breathing transformer. The
samples with different weight ratios are taken out at the same point in time to ensure
comparability, which can be seen in Table 1.

Table 1. Sample examination times

Removal time Ageing time in h Name of the sample
Start 0 A, B, C, D, R 0
Week 2 336 A, B, C, D, R 1
Week 4 672 A, B, C, D, R 2
Week 6 1008 A, B, C, D, R 3
Week 8 1344 A, B, C, D, R 4
Week 10 1680 A, B, C, D, R 5
Week 12 2016 A, B, C, D, R 6
Week 14 2352 A, B, C, D, R 7

The designation “A” denotes samples, which have been aged as pure oil without
paper. “B” corresponds to the weight ratio paper to oil of 1:5, “C” to the ratio 1:10 and
“D” to the ratio 1:20. The designation “R” refers to the pure paper samples, which have
been aged not impregnated under air without oil.
 Investigation on the Accelerated Ageing Behaviour of an Oil-Paper-Insulation 791

3 Results

In this section, the results of the ageing study will be presented and discussed. In Fig. 2
the colour change of the aged oil samples with the different paper contents can be seen
at the various ageing points in time. The strongest colour change can be observed for
samples A, followed by samples C and D, since the samples appear slightly brighter
after 2 weeks of ageing. For the B samples with the paper to oil weight ratio 1:5 the
colour change is delayed, which can be observed especially with the sample B1
compared to A, C, D 1. The colour in this case is yellow brown compared to a brown
colour of the other samples. Due to the largest amount of paper presents at a ratio of
1:5, the ageing of the oil is delayed compared to the other samples, because the paper
binds most of the water content from the insulating oil [5]. The paper can absorb much
more moisture than the insulating oil. In this respect, a smaller amount of water is
dissolved in the insulating oil (see Fig. 5), which means that ageing by hydrolysis for
samples B takes place to a lesser extent compared to the other samples.

Fig. 2. Colour change of the aged oil samples

Figure 3 shows the colour change of the different paper samples over the ageing
time of 14 weeks. The paper samples aged under oil are divided into two groups. The
paper is divided into light and dark strips. The light paper strips made up the bulk of the
paper insulation.
Where they were overlaid by other strips of paper, they have lighter areas. Dark
discoloration of the paper over the ageing period can be observed in all samples. It is
interesting to note that the samples of C darken faster at the beginning. The dark paper
sample for C1 seems to be darker than for B1 and D1. From the fourth week of ageing
onwards, the sample for D2 appears to be darker, with the sample for B2 appearing
slightly lighter than C2 and D2. During further ageing the papers show similar colours.
The bright paper samples also do not seem to differ optically over the entire ageing
period. Also the paper strips, which are aged under air, become darker with the time. In
general, they appear somewhat lighter than the oil-impregnated samples, which is also
reinforced by the lack of oil in the paper.
792 T. Münster et al.

Fig. 3. Colour change of the aged paper samples

The course of the DP-value measured according to IEC 60450 is shown in Fig. 4.
The figure shows the DP-value of the light paper samples because they make up the
bulk. The measurements are done in an external accredited laboratory. As it can be
seen, there is a big difference between the ageing of paper samples that have been aged
together with oil and the paper samples that have been aged directly in air. The final
value for pure paper after 14 weeks has a DP-value of about 200, while it is about 400
for papers impregnated with oil. Obviously the surrounding oil has a protective effect
on the paper insulation. The ageing rate of the impregnated papers is slower. Since the
samples are prepared in the same vessel and come from the same paper roll, they have
the same DP-value at the beginning.

Fig. 4. Course of the DP-value of the different paper to oil ratios.

 Investigation on the Accelerated Ageing Behaviour of an Oil-Paper-Insulation 793

It is also interesting that the final value of the impregnated samples is the same. The
ageing time shows that there is almost no difference between the ratio 1:5 and 1:20.
Over time, the DP values for the ratio 1:5 are only slightly smaller. A larger difference
can be observed for the ratio 1:10. Thus, the DP-values during ageing are above those
of the other ratios, e.g. after 4 weeks with about 100 DP. The DP-value for the different
weight ratios starts at the same value, because all paper samples were prepared in the
same container and the average DP-value of all samples was determined, because this
measuring point was sent to the laboratory as one sample. For pure paper, this value is
slightly below the initial value, since the samples come from another paper roll on the
one hand and on the other hand [2] gives information that measuring new paper can be
difficult, because it is more problematic to dissolve for the viscosimetric measurement.
Figure 5 shows the trend of the moisture content of the oil during the ageing
process. The graph shows that the higher the paper content, the lower the moisture
content of the insulating oil, which can be explained by the hygroscopic properties of
the paper [6]. Nevertheless, the difference between the ratios seems small compared to
pure oil, where the moisture content is about twice as large. However, this is not
directly reflected in the ageing of the paper, as the ratio of 1:10 has the highest DP-
values during ageing. It cannot therefore be directly assumed that the wetter the oil is,
the faster the paper ages.

Fig. 5. Course of the moisture content of the different paper to oil ratios

In the papers [7] and [8] it has already been shown that the interfacial tension
(IFT) and the acid number correlate very well with the DP-value and thus allow an
indirect state analysis of the paper. Therefore, these ageing parameters will be con-
sidered in the following and the influence of the weight ratios on these markers will be
investigated. For a better overview, these ageing parameters are represented by
exponential compensation curves.
The course of the IFT of the different weight ratios can be seen in Fig. 6. As it can
be seen in the graph, there is a decreasing trend of the IFT over the ageing period. The
interfacial tension values decrease from about 40 mN/m at the beginning of the
794 T. Münster et al.

investigation to a value between 20 and 15 mN/m after 14 weeks of ageing. All curves
show an exponential progression. It is also interesting that the values of pure oil lie
between the values of the different ratios.

Fig. 6. Course of the IFT of the different paper to oil ratios

Figure 7 shows the course of the acid value over the ageing time of the different
ratios. As it can be seen, from week 10 the more paper is inside the flask, the higher the
acid content is. The acid content for the ratio 1:5 after 14 weeks is about 0.45 mgKOH/g
and for the pure oil the value is less than 0.35 mgKOH/g. However, until week 4, the
behaviour is the opposite. In this case, the less paper there is, the more acid can be
detected in the oil. It is generally known that paper can absorb acids, whereby paper
mainly absorbs low-molecular acids [9, 10]. It seems that till week 4 the paper absorbs
some of the acid and bind it in the paper. The acidity was determined by means of the
potentiometric titration method according to IEC 62021-1, whereby with this method it
is only possible to measure the total acidity in the oil. Larger amounts of acid generally
have a negative influence on the ageing behaviour of the paper insulation [11].

Fig. 7. Course of the acid content of the different paper to oil ratios
 Investigation on the Accelerated Ageing Behaviour of an Oil-Paper-Insulation 795

Figure 8 shows the progressions for CO, CO2 and the ratios CO2/CO and O2/N2,
for the different paper to oil weight ratios and the pure oil. It can be observed, that the
less paper there is inside the flask, the less CO2 and CO is present in the oil in dissolved
form. This is consistent with the results from [2, 4, 12–14] that a large part of the
formation of CO2 and CO is due to the ageing of insulating paper.
For pure oil, the maximum for CO2 is about 2000 ppm after 14 weeks of ageing. At
a ratio of 1:20 a maximum of 2500 ppm is observed after 4 weeks. For the ratio 1:10
this maximum is reached after 8 weeks with 3000 ppm and for the ratio 1:5 with
6000 ppm after 2 weeks. It can also be said that the more paper there is, the more
pronounced is the maximum. It is also interesting that the concentration of CO2 after
reaching the maximum for all weight ratios drops to approximately 2000 ppm after
14 weeks, which corresponds to the final value of the ageing of the pure oil. With the
ageing of pure oil and the ratio 1:20, the maximum CO concentration is reached after 6
weeks with a value of about 1000 ppm, whereby the value at 1:20 is slightly higher.
After that, both systems decrease to 500 ppm after 14 weeks of ageing. The weight
ratio of 1:10 also shows a maximum after 6 weeks with a value of more than
1000 ppm. In this case the value also drops slightly. However, the value rises again, so
that the value after 14 weeks is again at 1000 ppm. At a ratio of 1:5, the maximum CO
concentration is reached after 8 weeks with a value of about 2000 ppm. At this ratio,
the values for CO already exceed 1000 ppm after 2 weeks and thus exceed the con-
centrations of the other ratios due to the higher paper content. With regard to the CO2/
CO ratio, the values for the various weight ratios are predominantly in the range
between 2 and 3. According to IEC 60599 the ratio of CO2/CO should lie between 3
and 10 for normal paper ageing [15]. In general, a fluctuating progression of the ratio
can be observed, with a value below 3 being reached after only 2 weeks for the weight
ratios 1:5, 1:10 and 1:20. Also for pure oil the value is below 3 after 2 weeks, whereby
the ageing took place without the participation of paper insulation. However, Höhlein-
Atanasova et al. [16] also pointed out that a CO2/CO ratio of less than 3 can exist
without irregularities as long as no other fault gases such as H2 are formed.
According to IEC 60599 a value of O2/N2 below 0.3 gives information about an
excessive consumption of oxygen. The values for O2/N2 are below 0.3 for all weight
ratios already after 2 weeks, which means that increased oxidation processes can be
assumed, which are comparable in all ratios.
796 T. Münster et al.

Fig. 8. Course of CO, CO2, CO2/CO and O2/N2 for the different paper to oil ratios
 Investigation on the Accelerated Ageing Behaviour of an Oil-Paper-Insulation 797

4 Conclusions

The investigations have shown that the different paper to oil weight ratios have an
influence on the formation of some ageing markers. This should be taken into account
in the indirect condition assessment of insulating paper when performing an analysis
using ageing markers determined by insulating oil. Thus, this could be important for
the asset management of different transformers. This study had shown that the paper
ages more slowly under oil than under air. It had also been shown that the paper had the
highest DP-value for the paper to oil weight ratio of 1:10, although the moisture content
in the oil was least at the ratio 1:5. However, there is no clear overall difference in the
DP-value between the ratios, so that the ageing rate can be regarded as independent of
the weight ratio. IFT, on the other hand, showed also no clear trend. The acid mea-
surement showed that the more paper the system contains, the more acid is produced.
At the beginning of aging, on the other hand, more acid is detectable if less paper is
present, since the acid is not dissolved in the paper. The determination of the DGA
analysis also showed that more CO and CO2 are dissolved in the oil when larger
quantities of paper are used. These results can also be important for the development of
algorithms for the condition evaluation of insulating paper and thus the determination
of the further lifetime of a transformer, with the use of different ageing markers.

Acknowledgments. The authors would like to express their gratitude to GRIDINSPECT GmbH
and AiF/ZIM for the financial support as well as Weidmann Electrical Technology AG for the
support with insulation materials and Analysen Service GmbH for the analysis.

References
1. Bakar, N.A., Abu-Siada, A., Islam, S.: A review on chemical diagnosis techniques for
transformer paper insulation degradation. In: Australasian Universities Power Engineering
Conference, AUPEC 2013, Hobart, TAS, Australia, 29 September–3 October 2013
2. Technical Brochures, Reference: 738: Ageing of liquid impregnated cellulose for power
transformers, Cigre, August 2018
3. Lundgaard, L.E., Hanse, W., Linhjell, D., Painter, T.J.: Aging of oil-impregnated paper in
power transformers. IEEE Trans. Power Delivery 19, 230–239 (2004)
4. Teymouri, A., Vahidi, B.: CO2/CO concentration ratio: a complementary method for degree
of polymerization of power transformer paper insulation. IEEE Electr. Insul. Mag. 33, 24–30
(2017)
5. Du, Y., Zahn, M., Lesieutre, B.C., Mamishev, A.V., Lindgren, S.R.: Moisture equilibrium in
transformer paper-oil systems. IEEE Trans. Dielectr. Electr. Insul. (1999)
6. Jeffries, R.: The sorption of water by cellulose and eight other textile polymers. J. Text. Inst.
Trans. 51, 339–374 (1960)
7. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel, J.: Investigation on the
accelerated ageing behaviour of oil-paper-insulation using different insulating oils. In: VDE
Fachtagung, Hochspannungstechnik, Berlin, Germany (2018)
8. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel, J.: Investigation on the ageing
parameters of a thermally accelerated aged paper-oil-insulation in a hermetically sealed
system. In: Condition Monitoring and Diagnosis, CMD, Perth, Australia (2018)
798 T. Münster et al.

9. Ingebrigtsen, S., Dahlund, M., Hansen, W., Linhjell, D., Lundgaard, L.E.: Solubility of
carboxylic acids in paper (Kraft)-oil insulation systems. In: Annual Report Conference on
Electrical Insulation and Dielectric Phenomena, CEIDP, pp. 505–514 (2004)
10. Lelekakis, N., Wijaya, J., Martin, D., Susa, D.: The effect of acid accumulation in power-
transformer oil on the aging rate of paper insulation. IEEE Electr. Insul. Mag. 30, 19–26
(2014)
11. Lundgaard, L.E., Hansen, W., Ingebrigtsen, S., Linhjell, D., Dahlund, M.: Aging of Kraft
paper by acid catalyzed hydrolysis. In: ICDL: 2005 IEEE International Conference on
Dielectric Liquids, New York (2005)
12. Kachler, A.J., Höhlein, I.: Aging of cellulose at transformer service temperatures. Part 1:
influence of type of oil and air on the degree of polymerization of pressboard, dissolved
gases, and furanic compounds in oil. IEEE Electr. Insul. Mag. 21, 15–21 (2005)
13. Emsley, A.M., Stevens, G.C.: Review of chemical indicators of degradation of cellulosic
electrical paper insulation in oil-filled transformers. IEE Proc.-Sci. Meas. Technol. 141, 324–
334 (1994)
14. Duval, M., Pablo, A.D., Atanasova-Hoehlein, I., Grisaru, M.: Significance and detection of
very low degree of polymerization of paper in transformers. IEEE Electr. Insul. Mag. 33, 31–
38 (2017)
15. IEC 60599: Mineral oil-filled electrical equipment in service – guidance on the interpretation
of dissolved and free gases analysis (2015)
16. Höhlein-Atanasova, I., Frotscher, R.: Carbon oxides in the interpretation of dissolved gas
analysis in transformers and tap changers. IEEE Electr. Insul. Mag. 26, 22–26 (2010)
 Localization of Deviations in Cable Geometry
Using FDA Fault Localization Method

Y. Norouzi1(&), C. Frohne1, J. Rothfeld1, and P. Werle2

1
Nexans Deutschland GmbH, Kabelkamp 20, 30179 Hannover, Germany
Younes.Norouzi@Nexans.com
2
Institute of Electric Power Systems, Division of High Voltage Engineering
and Asset Management, Schering-Institute, Leibniz Universität Hannover,
Callinstr. 25A, 30167 Hannover, Germany

Abstract. Frequency domain analysis (FDA) is a possible method to detect,

classify, and localize the faults in cable systems. This method extracts the fre-
quency response of the cable by applying a sweep voltage. Principle of this
method is based on transmission line theory and could detect changes in the
core, conducting screen, semi-conductor layers, and especially changes in the
insulation bulk. Changes in the insulation of a cable alters its electrical
parameters, which leads to a deviation in the frequency response of the damaged
cable compared to the response of a healthy cable system. It is common to
consider faults in cables just as a resistive fault, but in reality, most of the faults
in cables cause a deviation in geometry of the cable at the fault position. This
can be interpreted as a change in the capacitance, inductance, and conductance
of the cable. These variations in the cable parameters must be taken into account
to localize the faults precisely. In this contribution, the sensitivity of the FDA
method to the abnormalities and deviations in the cable geometry is demon-
strated. Several measurements on a long MV cable were performed to examine
the sensitivity of the FDA method in real cases. The used MV cable is damaged
and a small part of its geometry is slightly deviated, which could not be detected
using common fault localization methods, e.g., time domain reflectometry
(TDR). The FDA method however, was applied and could successfully detect
and localize the change in the cable geometry.

Keywords: Cable
Fault localization
Frequency Domain Analysis (FDA)

1 Introduction

To increase the availability of extruded cable systems, faults should be detected,

localized and as quickly as possible repaired. The first step is to classify the faults, thus
the most suitable localization method for this kind of fault can be chosen. Already
many fault localization methods for AC and some for DC cable systems are used. To
date, various methods have been proposed to localize faults in the cable systems.
Methods which use signals reflectometry in time or frequency domain or in both time
and frequency domain simultaneously [1, 2] are preferred by cable producers and
transmission system operators; because these methods apply low voltage signals to
localize faults in the cable systems. Some examples of the methods that operate in
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 799–810, 2020.
https://doi.org/10.1007/978-3-030-31676-1_75
800 Y. Norouzi et al.

frequency domain are frequency domain analysis (FDA) [3, 6], frequency domain
reflectometry (FDR) [7, 8], and line impedance resonance analysis (LIRA) [9]. FDA
method, which is used in this work can be distinguished from other methods, e.g. FDR
and FRA. FDA measures in frequency domain and transfer the results mathematically
by Inverse Fast Fourier Transformation (IFFT) in time domain and analysis them in
time domain. But FDA analysis the results in both frequency domain and after
transformation in time domain. Analysis in frequency domain is based on the position
of resonance points, phase of the input impedance and shape of curves, which change
in case of faults in cable. Moreover, FDA does not only measure just the input
impedance, but also analysis the admittance and the gain of the injected sweep voltage.
To cite an instance, if the quadrupole model of a transmission line is imagined and S-
parameters are analyzed to localize the faults in cable systems, FDA uses reflection
coefficient (s11) and transmission coefficient (s12), whereas FDR and LIRA use just s11.

2 Frequency Domain Analysis

A method to classify and to localize the faults in both HVAC and HVDC cable systems
with a large measuring range from some meters up to some hundred kilometers is the
FDA method [3, 6]. This method works on the principle of line theory and measures
the complex input impedance of the cable as a function of the frequency. This input
impedance depends on the material properties, the dimensions and the length of the
cable. The frequency response of the injected sweep voltage is dependent on cable
parameters (R′, L′, C′ and G′), cable length and the frequency of the input sweep
voltage, where:
R′ = Resistance in Ohm per Meter (X/m)
G′ = Conductance in Siemens per Meter (S/m)
L′ = Inductance in Henry per Meter (H/m)
C′ = Capacitance in Farad per Meter (F/m)
If these parameters change in the cable, for example because of aging or damages,
the position, amplitude and phase of the frequency-dependent complex impedance
change. As a result, damages in the conductor, as well as in the insulation and the outer
sheath could be identified. The FDA can also be used for cable condition diagnostics.
For this purpose, an FDA is usually carried out after cable laying as finger print, which
can be used for further measurements as a reference.
However, applicability of FDA methods, especially for the long and ultra-long
HVDC cable systems, is still an area with limited experiences in the field and needs to
be more investigated and improved for practical applications. The proposed fault
localization method in this contribution differs from the time domain reflectometry in
the sensitivity to abnormalities in the cable and in the possibility of providing a
diagnostic even before a fault occurs.
 Localization of Deviations in Cable Geometry 801

2.1 Complex Input Impedance

A cable can be described as a symmetrical, linear quadrupole system as shown in
Fig. 1.

Fig. 1. Input impedance of a line as a function of wave impedance, propagation coefficient and
cable length.

The input impedance as a function of the output impedance ðZ 2 Þ and characteristic

impedance ðZ w Þ according to Fig. 1 is:

Z 2 þ Z W tanhðclÞ
Z1 ¼ ZW ð1Þ
Z W þ Z 2 tanhðclÞ

2.2 Resonance Frequencies

Resonance behaviour of a cable is in the following described by a lossless model of a
cable. For a lossless cable, the attenuation measure is a = 0, and the calculation of
open-circuit impedance ZOC is obtained by assuming an infinite termination impe-
dance, which results in the following expression.

1
Z OC ¼ lim Z 1 ¼ Z W ¼ Z W cothðclÞ ð2Þ
Z 2 !1 tanhðclÞ

1 cosðblÞ
Z OC ¼ jZ W ¼ jZ W ð3Þ
tanhðblÞ sinðblÞ

Consequently, the following condition applies to determine the resonance fre-

quencies of the open-circuit impedance.

2p k
bl ¼ l ¼ np ! l ¼ n ; n ¼ 1; 2; 3; . . .
k 2 ð4Þ
n
fn ¼ pffiffiffiffiffiffiffiffiffi for n ¼ 1; 2; 3; . . .
2l L0 C0
802 Y. Norouzi et al.

3 Inverse Fast Fourier Transformation

There is a possibility to transform the results of the frequency domain into the time
domain to facilitate a direct evaluation of the measurement results and localize the
faults or deviations in the insulation system of the cable and obtain a signal image
equivalent to the TDR. Discrete Fourier Transform (DFT) is the discrete version of the
Fourier Transform (FT) that transforms a signal (or discrete sequence) from the time
domain representation to its equivalent in the frequency domain; whereas, Fast Fourier
Transform (FFT) is any efficient algorithm for calculating the DFT.

X
N 1
DFT: Xn ¼ xk :eðj2p nk=NÞ
K¼0
ð5Þ
1XN 1
IDFT: Xn ¼ xk :eðj2p nk=NÞ
N K¼0

where:
Xn/k = Spectral field with k/n numbers
xk/n = Input signal of value k/n in the TD
k = input value
n = Length of the input signal in the TD
N = Length of the output signal in the FD

FFT is a fast and efficient algorithm to calculate the spectral components [10]. For
that purpose, the FFT needs a number of input values (N) equal to a power of two [10].
If the number of input values is not a power of two, then the FFT can’t be used.
MATLAB solves this problem in its function (FFT) with a zero-padding at the end of
the original signal.

4 Test Object

A 11.277 m/33 kV three phase medium voltage cable with XLPE insulation and
copper conductor, as shown in Fig. 2, was used as the test object. The length of the
cable was achieved by adding 5 cables together with 4 joints (J1–J4).

Fig. 2. Test object: a 11277 m 33 kV medium voltage cable with XLPE insulation and copper
conductor.
 Localization of Deviations in Cable Geometry 803

4.1 First Step: Before Removing the Outer Sheath

In the first measurements, the third phase L3 of the 11.2 km long MV cable has a fault
in its jacket (outer sheath) without any damages in the insulation or semi-conductor
system. This is schematically shown in Fig. 3. The performed measurements exhibit no
difference between the FDA results of the three phases of the cable as shown in Fig. 4,
because the FDA measurements were done between conductor and outer metal screen.
Therefore, just the changes in conductor, metal screen or in all cable layers between
them are expected to be detected.

Fig. 3. Schematic structure of a three-core XLPE cable with a fault at 2906 m in the outer
sheath of L3.

Fig. 4. FDA measurement up to 1 MHz (shown up to 100 kHz) of all three phases from
wagon 1.

To convert the signal into the time domain, the Inverse Fast Fourier Transform
(IFFT) algorithm is used without any further filtering of the signal. The time axis is
scaled over the largest signal, and the cable length is calculated using the propagation
velocity. Therefore, if the ratio between the power of an incident electromagnetic wave
and that of its reflection in a very wide frequency range is measured and afterwards an
804 Y. Norouzi et al.

IFFT is performed, the location, where the complex impedance varies in the cable can
be determined as shown in Fig. 5. This Figure illustrates the transformed signal of the
frequency response of the input impedance as a function of the cable length, which is
normalized to the peak of the reflection from the cable end. The distance can be
calculated by multiplying the time steps resulting from the IFFT with the propagation
velocity of an electromagnetic wave inside the cable.

Fig. 5. IFFT from the FDA measurement up to 1 MHz of all three phases from wagon 1.

4.2 Second Step: After Removing the Outer Sheath and Peeling the Outer
Semiconductor; the Metal Screen Is Cut at the Fault Position
In the next step the outer sheath at the fault position is removed. Figure 6 depicts the
cable under investigation and its schematic drawing, respectively. Moreover, the metal
screen is cut at the fault position and the outer semi-conductor has been peeled to
determine how deep the damage is.

Fig. 6. Fault position of the 11.2 km cable and its schematic structure.
 Localization of Deviations in Cable Geometry 805

The FDA measurements have been done again on three phases. In this case the
metal screen of L3 is cut at the fault position at 2906 m. Therefore, according to (4) it is
expected to have less resonances for L3 than for L1 and L2 in the same frequency
range. In this case the ratio of number of resonance frequencies of L1 or L2 to L3 is:
11277 m
2906 m  3:88. It means the number of resonance points of L3 are almost 25% of the
number of those of L1 or L2. FDA measurements in terms of amplitude and phase of
the input impedance of three phases of the test object are shown in Fig. 7.

Fig. 7. FDA measurements up to 20 MHz (shown up to 1 MHz) from waggon 1. L3 with

opened outer sheath and cut metal screen.
806 Y. Norouzi et al.

In Fig. 8, the IFFT results of amplitude, phase and complex (amplitude and phase)
form of FDA measurements are depicted. The position of cut metal screen is clearly
visible in all three pictures, even if not the full information (amplitude and phase) is
used. In the case of IFFT from amplitude and complex signal some other small
reflection can be observed, which are the 4 factory joints J1–J4, which are also visible
by this method, but not by TDR. It means that in the case of significant changes in the
geometry of the cable, e.g. cut metal screen, all three sorts of IFFT could be used. But
in case of smaller changes there is a need for further investigations.

a) b)

C)

Fig. 8. FDA measurements up to 20 MHz. IFFT from amplitude (a), phase (b) and complex
(c) of input impedance from Waggon 1. L3 with removed outer sheath, peeled semi-conductor
and cut metal screen.
 Localization of Deviations in Cable Geometry 807

4.3 Third Step: After Removing the Outer Sheath and Peeling the Outer
Semiconductor: The Metal Screen Is Restored
In the next step the metal screen of L3 is restored as schematically shown in Fig. 9. The
surface of the peeled semi-conductor is 2 mm2.

Fig. 9. Schematic structure of the restored fault position of L3 of the 11.2 km cable.

The FDA measurements of three phases have been performed up to 20 MHz and in
Fig. 10 the results only up to 100 kHz are shown in order to make the resonance points
more visible. Figure 10 makes it clear, that there is almost no difference between FDA
measurements in all three phases. One reason is the low frequency resolution of the
measurements, because of the large measured frequency range. The reason that a large
frequency range is selected is to have a higher spatial resolution of IFFT results. The
spatial resolution is explainable with the maximum measured frequency and the phase
velocity at this frequency. The resolution is calculated as follows:

maxðvÞ
DL ¼ ð6Þ
2:maxðf Þ

where: DL = Spatial resolution in Meter (m)

V ¼ imag
2:p:f
ðcÞ
calculated phase velocity.

Fig. 10. FDA measurements up to 20 MHz. Amplitude (left) and phase (right) of the complex
input impedance from Waggon 1. L3 with removed outer sheath, peeled semi-conductor and
restored metal screen.
808 Y. Norouzi et al.

In the next step, the results of the FDA measurements are transferred to time
domain. If the cable had no other damages or no change in the geometry beside the
reconstruction of the metal screen, there would be no visible reflection at the con-
nection position in the IFFT. But in this case there is a small deviation in the outer
semi-conductor, as shown in Fig. 9, and also an almost negligible deviation in the cable
metal screen after the restoration. Figures 11, 12 and 13 show the IFFT results in
amplitude, phase and complex signal of the input impedance of the test object. The
fault position can be detected from all three IFFTs, although they have different
amplitudes of reflection. If just the IFFT from amplitude or from the phase is con-
sidered, some information of the signal will be lost. The only signal, which contains all
the information is the IFFT of the complex signal. But in some cases, there is a special
interest of detecting a particular fault or deviation in the cable geometry. In such a case
the desired reflection may cause a smaller reflection than the other reflections.
Therefore, in all cases it is recommended to investigate all the three mentioned IFFTs to
have a better understanding of the faults locations.

Fig. 11. FDA measurements up to 20 MHz. IFFT from amplitude of input impedance from
Waggon 1. L3 with restored metal screen.

Fig. 12. FDA measurements up to 20 MHz. IFFT from phase of input impedance from Waggon
1. L3 with restored metal screen.
 Localization of Deviations in Cable Geometry 809

Fig. 13. FDA measurements up to 20 MHz. IFFT from complex signal of input impedance from
Waggon 1. L3 with restored metal screen.

5 Conclusion and Future Works

This contribution shows the high sensitivity of the FDA method to small faults and
deviation in cable geometry and in cable parameters. The deviations and faults could be
detected and localized precisely.
Moreover, the results of this work show that IFFT represented in amplitude, phase
or complex part of input impedance are different in some cases. Therefore, for every
fault scenario all these three forms of IFFT should be performed and compared to each
other. For a better interpretation of the IFFT results following steps will be taken in
future works:
– Define the best frequency range for measuring different fault scenarios
– Using denoising algorithm for a clear localization of abnormalities or changes in
FDA
– Using time of arrival algorithm to improve the preciseness and accuracy of the
localization method.

Acknowledgements. The authors gratefully acknowledge funding by the (BMBF) within the
Kopernikus Project ENSURE ‘New ENergy grid StructURes for the German Energiewende’.

References
1. Kwon, G.Y., Lee, C.K., Lee, G.S., Lee, Y.H., Chang, S.J., Jung, C.K., Kang, J.W., Shin, Y.
J.: Offline fault localization technique on HVDC submarine cable via time-frequency domain
reflectometry. IEEE Trans. Power Deliv. 32(3), 1626–1635 (2017)
2. Shin, Y.J., Choe, T.S., Hong, C.Y., Song, E.S., Yook, J.G., Park, J.B.: Application of time-
frequency domain reflectometry for detection and localization of a fault on a coaxial cable.
IEEE Trans. Instrum. Measur. 54(6), 2493–2500 (2005)
3. Norouzi, Y., Frohne, C., Gauler, V., Aganbegovic, M., Werle, P., Stagge, H.: Identification
and classification of faults in DC cable systems. In: ISH Conference 2017, Buenos Aires
(2017)
810 Y. Norouzi et al.

4. Norouzi, Y., Frohne, C., Gauler, V., Werle, P., Seifi, S.: Feasibility of FDA method for the
detection and classification of faults in extruded HVDC cables. In: IEEE Conference-ICD
2018, Budapest, Hungary (2018)
5. Norouzi, Y., Frohne, C., Werle, P.: High frequency model of cables for frequency domain
analysis used for fault localization. In: IEEE-ICHVE Conference 2018, Athens, Greece
(2018)
6. Norouzi, Y., Braun, S., Frohne, C., Seifi, S., Werle, P.: Effect of Cable Joints on Frequency
Domain Analysis. In: IEEE-CEIDP CONFERENCE 2018, Cancun, Mexico (2018)
7. Ohki, Y., Yamada, T., Hirai, N.: Precise location of the excessive temperature points in
polymer insulated cables. IEEE Trans. Dielectr. Electr. Insul. 20(6), 2099–2106 (2013)
8. Ohki, Y., Hirai, N.: Effects of the structure and insulation material of a cable on the ability of
a location method by FDR. IEEE Trans. Dielectr. Electr. Insul. 23(1), 77–84 (2016)
9. Fantoni, P.F.: Line resonance analysis system. US Patent 7,966,137 (2011)
10. Kammeyer, K., Kroschel, K.: Digitale Signalverarbeitung. 7th edn. Vieweg + Teubner,
Wiesbaden, Germany (2009). ISBN 978-3-8348-0610-9
 Evaluation of Numerical Indices for Objective
Interpretation of Frequency Response to Detect
Mechanical Faults in Power Transformers

M. Tahir1(&), S. Tenbohlen1, and M. H. Samimi2

1
Power Transmission and High Voltage Technology (IEH),
University of Stuttgart, Stuttgart, Germany
mehran.tahir@ieh.uni-stuttgart.de
2
High Voltage Institute, ECE Department, University of Tehran, Tehran, Iran

Abstract. For assessing the mechanical integrity of power transformers using

the well-known frequency response analysis (FRA) method, proper interpreta-
tion of FRA results is crucial. In spite of many research efforts in developing a
systematic interpretation methodology, there is still no universally accepted and
definitive algorithm for interpretation of FRA results. The purpose of interpre-
tation of the FRA results is to detect the extent and type of a mechanical fault
followed by a judgement about the condition of the transformer, whether it
needs a repair or it can continue its normal operation. In literature, numerical
indices are one of the mathematical algorithms, which have been proposed for
the interpretation, where the amount of an index extracted from two FRA sig-
natures can show the degree of a mechanical fault. To ease the FRA based
condition monitoring, this paper evaluates different numerical indices, present-
ing a number of new indices and collecting them on a single platform, as an
interpretation tool for FRA results. To discuss the sensitivity of the indices
against different faults, axial displacement (AD) and radial deformation (RD) as
two common transformer-winding faults, with various extents, were practically
applied to the windings of a 1 MVA distribution transformer. For each case,
numerical indices are evaluated in different frequency bands to provide a deeper
understanding to their characteristics. To discuss the sensitivity of different
connection schemes, indices are evaluated and compared for four connection
schemes; end-to-end open circuit (EE-OC), end-to-end short circuit (EE-SC),
capacitive inter-winding (CIW) and inductive inter-winding (IIW) are recom-
mended by IEC standard. The results show the advantage of the proposed
indices over the others in the FRA interpretation, however further studies from
the field are needed to settle the new indices. The research can be beneficial for
the standardization process of the numerical indices.

Keywords: Frequency response analysis (FRA)
Power transformers

Transformer diagnosis
Mechanical faults
Numerical indices

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 811–824, 2020.
https://doi.org/10.1007/978-3-030-31676-1_76
812 M. Tahir et al.

1 Introduction

The key role of power transformers in electric power networks has brought up different
condition monitoring and diagnostic methods to ensure their healthy operation.
Nowadays, mechanical faults are one of the major reasons behind transformer failures.
Among these mechanical faults, radial deformation (Hoop Buckling), axial winding
elongation (Telescoping) and short circuit faults are the frequent mechanical faults of
power transformers. Though the minor mechanical faults do not necessarily lead to an
immediate transformer failure, however, its capability to withstand future mechanical
and electrical stresses strongly reduced [1].
Transfer function assessment using the well-known Frequency response analysis
(FRA) is a widely accepted technique for diagnosing transformer winding mechanical
faults and several other types of faults that are caused during manufacture, trans-
portation, installation and/or service life [2]. Previous contributions in the field of FRA
has resulted the standardization of its measurement procedures as described in the IEC
and IEE standards [3, 4]. However, the interpretation of FRA results is still limited to
the analysis of the experts in the field since there is no universally accepted and
definitive methodology available to assess the mechanical condition of a transformer
based on the FRA results. As FRA is a comparative diagnostic method which needs a
reference measurement for comparison. Time based, construction based and type based
comparison can be performed. Many researchers have suggested model experiments
and simulation techniques, simulating different faults as an approach for interpretation
of FRA results [5]. These simulation techniques have a drawback that high accuracy is
required to reproduce the FRA measurements [6].
Other FRA interpretation methodologies are based on assessment of FRA signa-
tures by means of numerical indices, which quantifies the differences in the FRA traces
[7]. The final target is a decision about the condition of a transformer based on the
quantified values of these indices. There are numerous indices proposed in the literature
[2, 8, 9]. However, there is no comprehensive comparison study to assess the sensi-
tivity of these numerical indices against different faults in different frequency bands.
Consequently, if it is intended to use a numerical index for interpretation of FRA
signatures, it is not known which one is appropriate. Therefore, the definition of a
standard numerical index is the need of the hour. Additionally, the standardization of
the FRA interpretation process through numerical indices demands a comparison
between these numerical indices followed by selecting the appropriate indices. This
contribution is an effort in this direction, which evaluates different numerical indices
based on their characteristics, presenting a number of new indices and collecting them
on a single platform, to announce the most appropriate indices as the standard ones, for
objective interpretation of FRA signatures.
In this paper, axial displacement (AD) and radial deformation (RD) as two common
transformer-winding faults, with various extents, were practically applied to the
windings of a 1 MVA distribution transformer. For each case, numerical indices are
evaluated in different frequency bands to provide a deeper understanding to their
characteristics. The results of sensitivities of different connection schemes against two
faults are also reported in this contribution.
 Evaluation of Numerical Indices for Objective Interpretation 813

2 Experimental Setup

The transformer used in this study is a single-phase transformer with HV and LV

windings. The windings correspond to a transformer with a rating of about 1 MVA.
The HV winding is a continuous disk winding (height = 865 mm) with 660 turns in 60
disks and the LV winding is a helical winding (height = 865 mm) with 24 turns and 12
parallel conductors in each turn. Two hollow, copper cylinders are employed outside
and inside of the windings to model the tank and the core, respectively. As in this case,
the FRA traces are measured above 10 kHz, and due to small skin depth in this
frequency range, the core can be replaced with the hollow metallic cylinder [5].
To evaluate and compare the characteristics of numerical indices, two case studies
are considered in the framework of this research. Specific details of each case is
provided in the following sections.

(a) Axial displacement (AD) (b) Radial deformation (RD)

Fig. 1. Experimental setup and different mechanical faults

2.1 Case 1: Axial Displacement

In case 1, axial displacement is introduced on the HV winding, with various levels of
AD being implemented, by inserting a number of spacers at the bottom of the winding
814 M. Tahir et al.

as shown in Fig. 1(a). Five steps of AD are implemented in the HV winding in upward
direction, each step of 5 mm. Therefore, the maximum displacement is about 3% of the
height of the HV winding.

2.2 Case 2: Radial Deformation

In case 2, radial deformation is simulated by deforming one section of the LV winding
through the winding height. LV winding is deformed in seven steps; in each step, the
radius of that section is decreased by 2.5 mm. Figure 1(b) shows the deformed LV
winding as described in [9].
Four connection schemes, specified in IEC standard [3], are employed including
OC, SC, CIW and IIW as shown in Fig. 2.

Fig. 2. FRA measurement connection schemes [3] (a) Open circuit (OC) (b) Short circuit
(SC) (c) Capacitive inter-winding (CIW) (d) Inductive inter-winding

3 Numerical Indices

Interpretation of FRA results using numerical indices is based on the fact that the
numerical indices quantify the differences between reference and present FRA traces.
In literature, researchers have proposed different indices. In the following, some
commonly used and some newly proposed indices are presented. The names of the
indices along with the references are given in Table 1. The equations of these indices
are also summarized. In the following equations, X and Y are the magnitude vectors of
reference and present FRA trace respectively, X(i) and Y(i) are the ith elements of these
vectors, N is the number of data points, f is the frequency vector, and uX and uY are the
phase vectors of the reference and present FRA trace:
 Evaluation of Numerical Indices for Objective Interpretation 815

Table 1. Definition of numerical indices.

Abbreviation Definition Reference
CCF Cross correlation Factor [10]
CC Correlation coefficient [8, 13]
Rxy Relative factor [11]
CD Complex distance [12, 13]
ED Ecludean Distance [12, 13]
SD Standard deviation [12, 13]
SSE Sum squared error [8, 13]
SSD Stochastic spectrum deviation [12, 13]
RMSE Root mean square error [12, 13]
ASLE Absolute sum of logarithmic error [8, 13]
SDA Standard difference area [2, 13]
LSE least squared error [12]

PN
ðXðiÞ  XÞðYðiÞ   YÞ 
q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
CCF ¼ P i¼1 ð1Þ
P
N  2 N ½YðiÞ  Y 2
i¼1 ½XðiÞ  X  i¼1

PN
i¼1 XðiÞYðiÞ
CC ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN ð2Þ
2 PN 2
i¼1 ½XðiÞ i¼1 ½YðiÞ


10 1  q\1010
Rxy ¼
 log10 ð1  qÞ otherwise
PN P ð3Þ

i¼1 ðY ðiÞ:X ðiÞÞ
 Y  ðiÞ ¼ YðiÞ  N1 Ni¼1 YðiÞ
q ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN PN ; PN

 ðiÞÞ2 X ðiÞ ¼ XðiÞ  N
1
i¼1 ð Y  ðiÞÞ2
i¼1 ð X i¼1 XðiÞ

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
" #ffi
u
uXN ðXðiÞ cos u ðiÞ  YðiÞ cos u ðiÞÞ 2
CD ¼ t i¼1
X Y
ð4Þ
þ ðXðiÞsin uX ðiÞ  YðiÞsin uY ðiÞÞ2
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
XN
ED ¼ i¼1
ðYðiÞ  XðiÞÞ2 ð5Þ

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN 2
i¼1 ðYðiÞ  XðiÞÞ
SD ¼ ð6Þ
N1
PN
ðYðiÞ  XðiÞÞ2
SSE ¼ i¼1
ð7Þ
N
816 M. Tahir et al.

 
100 XN YðiÞ  XðiÞ
SSD ¼ i¼1 
ð8Þ
N XðiÞ 
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!2ffi
u
u 1 XN jYðiÞj  jXðiÞj
RMSE ¼ t PN ð9Þ
i¼1 jXðiÞj
N i¼1 1
N

PN
j20 log10 YðiÞ  20 log10 XðiÞj
ASLE ¼ i¼1
ð10Þ
N
R
jYðf Þ  Xðf Þjdf
SDA ¼ R ð11Þ
jXðf Þjdf
PN

ðXðiÞ  XÞðYðiÞ 
 YÞ
LSE ¼ i¼1
PN ð12Þ
 2
i¼1 ½XðiÞ  X 

4 Interpretation Experiences with Numerical Indices

There are some cases, which reported unsuccessful performance of some numerical
indices in FRA interpretation, namely CC, Rxy, SSE, and SSD [8, 9, 13, 14]. It is
noteworthy to recall that previous contributions have employed different frequency
ranges for different indices. Thereby, it is difficult to distinguish the attributes of each
frequency range, as the data of each contribution is also different. In order to perform
an investigation, to standardize the indices, all indices should be calculated from dif-
ferent frequency ranges of one database and, correspondingly, their characteristics
should be assessed. Moreover, criteria for abnormality of transformer had been pro-
posed in [10] and [11] as summarized in Tables 2 and 3. However, ref. [14] reports
cases where the transformers have obvious deformations, but the Chinese standard fails
to show deformations, indicating a “normal winding”. Similarly, ref. [15] reported two
cases where CCF criterion could not detect any fault. Such inconsistencies show that
these criteria need revision. By considering the above-mentioned observations, it can
be concluded that improving the interpretation of FRA needs more complementary
numerical indicators.

Table 2. Criteria for FRA diagnosis proposed in [10].

Category Limits for the CCF
Investigate <0.95
Marginal 0.96–0.97
Good 0.98–1.00
 Evaluation of Numerical Indices for Objective Interpretation 817

Table 3. Criteria for FRA diagnosis proposed in [11].

Category Limits for the relative factor (Rxy)
Severe (RLF < 0.6)
Obvious (1.0 > RLF  0.6) or (RMF < 0.6)
Slight (2.0 > RLF  1.0) or (0.6  RMF < 1.0)
Normal (RLF  2.0), (RMF  1.0) and (RHF  0.6)
LF: 1–100 kHz, MF: 100–600 kHz, HF: 600–1000 kHz

5 Proposed Numerical Indices

5.1 Lin’s Concordance Coefficient (LCC)

CC and CCF can only measure deviations from the best-fit line, which is precision while
these indices fail to detect agreement between two variables. It implies that if Y ¼ cX or
Y ¼ X þ c, in which c is a constant, these indices show no deviation in the traces
although the traces are completely different. To overcome these problems, this study
proposes a desirable reproducibility index, to be called a Lin’s Concordance Coefficient
(LCC), which evaluates precision as-well-as agreement between two variables by
measuring variation through the concordance line [16]. LCC is defined using (13)

LCC ¼ ðYXÞ2S
2
XY
þ S2 þ S2
¼ qCb
Y X

P
N
SXY ¼ N 1
ðXðiÞ  XÞðYðiÞYÞ
i¼1
P
N P
N
S2Y ¼ N1 ðYðiÞYÞ2 ; Y ¼ N1 YðiÞ
i¼1 i¼1
ð13Þ
P
N
2 P
N
S2X ¼ N1
ðXðiÞXÞ ; X¼N 1
XðiÞ
i¼1 i¼1

Cb ¼ ½ðv þ 1
v þ u2 Þ=21 ;
ðYXÞ
v ¼ SSXY = Scale shift, u¼p ffiffiffiffiffiffiffi = Location shift
S S X Y

Here, 0 < Cb  1 is a bias correction factor which is a measure of accuracy and q

is the Pearson correlation factor measures precision of two measurements. This implies
that LCC can detect frequency deviations as well as the magnitude deviations between
two FRA traces.

5.2 Sum of Error (SE)

SE is derived from SSE, which calculates square of the distance between two traces.
Due to square term, SSE shows a parabolic behavior toward the mechanical changes
that makes it a less linear index. In contrast, SE calculates the linear distance between
two traces. SE is defined using (14)
818 M. Tahir et al.

PN
jYðiÞ  XðiÞj
SE ¼ i¼1
ð14Þ
N

6 Measurement Results and Index Evaluation

The measured TFs for AD and RD faults are shown in Figs. 3 and 4 respectively. As
can be seen in the measured FRA traces for AD, the impact of the AD on FRA traces is
noticeable in the medium (100 kHz–600 kHz) and high frequency (600 kHz–1 MHz)

(a) (b)

(c) (d)

Fig. 3. FRA measurements for different levels of axial displacement fault (a) Open circuit
(OC) (b) Short circuit (SC) (c) Capacitive inter-winding (CIW) (d) Inductive inter-winding (IIW)

(a) (b)

(c) (d)

Fig. 4. FRA measurements for different levels of radial deformation fault [9] (a) Open circuit
(OC) (b) Short circuit (SC) (c) Capacitive inter-winding (CIW) (d) Inductive inter-winding (IIW)
 Evaluation of Numerical Indices for Objective Interpretation 819

range. Results also show that AD fault increases the resonance frequencies in the
medium and high frequency range with only a slight change in magnitude in the CIW
connection scheme. The change in magnitude is more pronounced in IIW connection
scheme. While, in case of RD, the impact of the RD on FRA traces is obvious in the
medium (100 kHz–600 kHz) and high frequency (600 kHz–1 MHz) range and con-
trary to the effect of AD faults on FRA traces, RD faults shift the resonance frequencies
to the left in the medium and high frequency range with a slight change in magnitude.

6.1 Index Evaluation

Firstly, all indices are calculated for different steps of mechanical faults and for dif-
ferent connection schemes in both cases. Afterwards, the calculated indices are plotted
against fault steps as shown in Fig. 5. From such plots, three characteristics of the
indices can be extracted: monotonicity, linearity and the sensitivity.
A good numerical index must exhibit monotonic behaviour against increasing
amounts of the mechanical fault since more severe faults increase the deviations in the
FRA traces and, consequently, an index should possess higher values for larger
deviations. Conversely, if an index is non-monotonic, it cannot be used to define the
extent of a mechanical fault. Typical behaviour of some monotonic and non-monotonic
indices is shown in Fig. 5. After evaluating each index for two cases, it was found that
most of the indices including CCF, CC, LCC, SD, ED, RMSE, ALSE, SE, and SSE are
purely monotonic.
Linearity of the indices is essential for quantitative diagnosis and it is crucial to
define thresholds to the indices as a diagnostic criterion. To quantify the linearity of the
indices, a regression analysis is performed which is a statistical tool, to define, how
much two variables are linearly coupled. The fault steps and the amounts of the indices
are taken as input variables. The output of the regression analysis is the coefficient of
determination that ranges from 0 to 1. Its value equal to 1 indicates a complete linear
relation. The results of linearity check are presented in Fig. 6. The coefficient of
determination reported here is the average of both cases. The result shows that ASLE,
LCC, CCF, ED, SD, RMSE, SDA, and SE exhibit good linear behaviour. While
indices CC, LSE, SSE, and SSD are weaker in this respect.
The indices including ASLE, LCC, CC CCF, ED, SD, RMSE, SDA and SE show
monotonic and linear behaviour. However, ED and RMSE are not properly normalized
which means their values depend on sample numbers in the data, the more sample
numbers there are, the higher are the amounts of the indices. Likewise, ASLE, cal-
culates the logarithmic sum while FRA data which is already in dB. Hence, only six
indices: CCF, CC, LCC, SD, SDA and SE are further used for sensitivity evaluation.
Tables 4 and 5 summarize the calculated indices against 1% of AD and 7th fault
step of RD, respectively. Here the indices are evaluated for two frequency sub-bands;
MF: 20 kHz – 400 kHz and HF: 400 kHz – 1 MHz, and the results of all connection
schemes are also compared. It can be seen from the results that CC could not detect any
deviation against 1% AD fault and 7th level of RD fault. SDA possess quite low
sensitivity in both cases. SE shows a good sensitivity to detect AD fault especially in
CIW and IIW connection schemes. Based on the CCF criterion proposed in Table 2
[10] and looking at the values of CCF in Tables 4 and 5, it can be seen that all the
820 M. Tahir et al.

Fig. 5. Monotonic and non-monotonic behavior of indices against different extents of radial
deformation fault in short circuit measurement

Fig. 6. Coefficient of determination showing the linearity behavior of the indices

values of CCF lie in good transformer range. Thus, CCF could not detect 1% AD fault
and 7th level of RD fault with this criterion. Similarly, the values of Rxy lies in normal
winding range for both cases. While LCC, which is an improved form of the CCF,
shows much better sensitivity and detects both faults based on the criteria proposed in
Table 2. It can be seen that LCC have detected 1% AD fault in both CIW and IIW
measurements. While in case of RD, it shows marginal values in three measurements:
OC, SC and IIW. It is noteworthy that both criteria, summarized in Tables 2 and 3, are
based on CCF. Hence, both criteria can be revised by using the proposed LCC, which
proved to be more sensitive compared to CCF.
 Evaluation of Numerical Indices for Objective Interpretation 821

From Tables 4 and 5, it can be seen that different indices hold different sensitivities
in different connection schemes depending upon their definition. Hence, sensitivity
analysis of different connection schemes should not be based on a single numerical
index rather it should consider the effect of all the selected indices. To achieve this, the
feature normalization (FN) method is implemented [17] to normalize the selected four
indices: CCF, LCC, SD, and SE. Finally, the average normalized index (ANI) is cal-
culated for each connection scheme. In this way, the effect of all four indices is
considered for sensitivity comparison of connection schemes. The average normalized
index (ANI) is calculated using (15) and (16).

Table 4. Calculated indices for 10 mm (1%) AD

Scheme OC SC CIW IIW

MF HF MF HF MF HF MF HF
CC(a) 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
SDA(a) 0.002 0.004 0.003 0.004 0.014 0.032 0.018 0.024
SD(a) 0.100 0.218 0.214 0.225 0.637 0.955 0.802 0.987
CCF(a) 0.999 0.998 0.999 0.997 0.997 0.985 0.984 0.980
Rxy(b) 3.522 2.698 3.154 2.537 2.408 2.537 1.703 1.756
LCC(a) 0.999 0.997 0.998 0.996 0.996 0.955 0.970 0.922
SE(a) 0.066 0.177 0.157 0.188 0.493 1.115 0.510 1.071

(a) MF: 20 kHz – 400 kHz, HF: 400 kHz – 1 MHz,

(b) MF: 100 kHz – 600 kHz, HF: 600 kHz – 1 MHz

Table 5. Calculated indices for step 7 of RD

Scheme OC SC CIW IIW

MF HF MF HF MF HF MF HF
CC(a) 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
SDA(a) 0.017 0.013 0.017 0.013 0.015 0.015 0.012 0.008
SD(a) 0.765 0.659 0.739 0.651 0.766 0.739 0.609 0.409
CCF(a) 0.989 0.982 0.989 0.982 0.997 0.993 0.981 0.980
Rxy(b) 1.790 1.787 1.821 1.772 2.431 1.995 1.675 1.585
LCC(a) 0.988 0.978 0.988 0.977 0.996 0.991 0.979 0.974
SE(a) 0.523 0.534 0.48 0.529 0.641 0.621 0.353 0.316

(a) MF: 20 kHz – 400 kHz, HF: 400 kHz – 1 MHz,

(b) MF: 100 kHz – 600 kHz, HF: 600 kHz – 1 MHz
822 M. Tahir et al.

The larger values of CCF, CC, and LCC indicate small deviation
 
Index  Indexmin
FN ¼ ð15Þ
Indexmax  Indexmin
1X
ANIcs ¼ FN index ð16Þ
5

Figure 7 shows the sensitivity comparison of different connection schemes for 1%

AD and 7th step of RD faults. For AD: IIW and CIW measurements show highest
sensitivity. While in case of RD: CIW, OC and SC measurements show good sensi-
tivity. It is noteworthy to point out that derived sensitivity also depends on the defi-
nition of the used indices. As some indices are less prone to magnitude changes in the
FRA traces. According to the principle of TF measurement, both frequency and
amplitude can change with the change of the impedance of the transformer. Therefore,
sensitivity of indices to detect both frequency and amplitude shifts is an important
characteristic that should also be considered.
In summary, LCC and SE have detected both faults: AD and RD. However, CCF,
CC, and Rxy cannot detect it when they are evaluated in certain frequency sub-bands.
TFs of CIW measurement possess good sensitivity to detect both AD and RD faults
among the studied connection schemes.

Fig. 7. Sensitivity comparison of different connection schemes for AD and RD faults through
four numerical indices: CCF, LCC, SD and SE
 Evaluation of Numerical Indices for Objective Interpretation 823

7 Conclusion

This paper collects and evaluates all the candidate numerical indices on a single
platform to provide a deeper understanding of their characteristics for the assessment of
transformer FRA signatures. The study extended the previous research works by
presenting some new numerical indices. The behavior of the indices is characterized by
monotonicity, linearity and sensitivity. The results of two case studies are considered.
The results provided in this contribution leads to the conclusion that the proposed
indices can satisfactorily detect minor winding faults. The sensitivity of different
connection schemes against different faults is also analyzed and it was found that CIW
measurements hold good sensitivity to detect investigated mechanical faults. The
results show the advantage of the proposed indices LCC and SE over the others for
FRA interpretation. However, further studies are required, particularly with large power
transformers to make a general conclusion.

References
1. Tenbohlen, S., Coenen, S., Djamali, M., Muller, A., Samimi, M.H., Siegel, M.: Diagnostic
measurements for power transformers. Energies 9, 347 (2016)
2. Rahimpour, E., Jabbari, M., Tenbohlen, S.: Mathematical comparison methods to assess
transfer functions of transformers to detect different types of mechanical faults. IEEE Trans.
Power Delivery 25(4), 2544–2555 (2010)
3. IEC60076-18: Measurement of frequency response, ED. 1, IEC Std, March 2012
4. IEEE guide for the application and interpretation of frequency response analysis for oil-
immersed transformers, IEEE Std. C57.149-2012, pp. 1–72, March 2013
5. Tenbohlen, S., Tahir, M., Rahimpour, E., Poulin, B., Miyazaki, S.: A new approach for high
frequency modelling of disc windings. In: Cigré 2018, Paris, paper no. A2-214 (2018)
6. Picher, P., Tenbohlen, S., Lachman, M., Scardazzi, A., Patel, P.: Current state of transformer
FRA interpretation. Proc. Eng. 202, 3–12 (2017)
7. Heindl, M., Tenbohlen, S., Kraetge, A., Krüger, M., Velásquez, J.: Algorithmic determi-
nation of pole-zero representations of power transformers’ transfer functions for interpre-
tation of FRA data. In: 16th International Symposium on High Voltage Engineering,
Johannesburg, Paper D-26 (2009)
8. Kim, J.-W., Park, B., Jeong, S.C., Kim, S.W., Park, P.: Fault diagnosis of a power
transformer using an improved frequency-response analysis. IEEE Trans. Power Deliv. 20
(1), 169–178 (2005)
9. Samimi, M.H., Tenbohlen, S., Akmal, A.A.S., Mohseni, H.: Effect of different connection
schemes, terminating resistors and measurement impedances on the sensitivity of the FRA
method. IEEE Trans. Power Deliv. 32(4), 1713–1720 (2017)
10. Kennedy, M., McGrail, A.J., Lapworth, J.A.: Using cross-correlation coefficients to analyse
transformer sweep frequency response analysis (SFRA) traces. In: 2007 IEEE Power
Engineering Society Conference and Exposition in Africa, pp. 1–6 (2007)
11. Frequency response analysis on winding deformation of power transformers. The Electric
Power Industry Standard of People’s Republic of China, Std. DL/T911-2004, ICS27.100,
F24, Document No. 15182-2005
824 M. Tahir et al.

12. Pourhossein, K., Gharehpetian, G., Rahimpour, E.: A probabilistic feature to determine type
and extent of winding mechanical defects in power transformers. Electr. Power Syst. Res.
82, 1–10 (2012)
13. Samimi,M.H., Tenbohlen, S.: The numerical indices proposed for the interpretation of the
FRA results: a review. In: VDE High Voltage Technology 2016, ETG-Symposium, Berlin,
Deutschland (2016)
14. Gonzales Arispe, J., Mombello, E.: Detection of failures within transformers by FRA using
multiresolution decomposition. IEEE Trans. Power Deliv. 29(3), 1127–1137 (2014)
15. Miyazaki, S., et al.: Proposal of objective criterion in diagnosis of abnormalities of power-
transformer winding by Frequency Response Analysis. In: 2016 International Conference on
Condition Monitoring and Diagnosis (CMD), Xi’an, pp. 74–77 (2016)
16. Lawrence, I., Lin, K.: A concordance correlation coefficient to evaluate reproducibility.
Biometrics 45, 255 (1989)
17. Zhang, H., Lin, H., Li, Y.: Impacts of feature normalization on optical and SAR data fusion
for land use/land cover classification. IEEE Geosci. Rem. Sens. 12, 1061–1065 (2015)
 Statistic Based Method for Post-processing
Analysis in Lifetime Investigations
of Multi-factor Aged Winding Insulation

A. Cimino1(&), J. Horst2, and F. Jenau1

1
Institute of High Voltage Engineering, TU Dortmund University,
Friedrich-Woehler-Weg 4, 44227 Dortmund, Germany
alessandro.cimino@tu-dortmund.de
2
Bielefeld University of Applied Science,
Interaktion 1, 33619 Bielefeld, Germany

Abstract. To evaluate the condition of electrical insulation systems the

knowledge of predominant aging mechanisms is of particular importance. As
experiments are very expensive and require a lot of time, and the experimental
setup only permits a limited number of test samples at the same time, it is
essential to analyze the already available lifetime data with the use of statistical
methods within post-processing.
Rotating machines are important components in power generation. There is a
high level of demand for machine reliability and availability. Negative influ-
ences on machine reliability and life expectancy might be thermal, electrical and
mechanical stress. Integration of decentralized renewable energy sources lead to
a fluctuating load profile of power engineering equipment, such as generators.
This increasingly stresses the electrical insulation system in form of mechanical
and thermomechanical forces. Regarding mechanical aging the condition of
winding insulation of large rotating machines has to be investigated more
detailed.
In this context, experimental investigations with a focus on mechanical ageing
are necessary to understand the complex insulation system in order to prevent
failures. Previously, initial evaluations using Weibull distribution have been
used. It has to be noted that only a small sample size is available, which limits a
statistical significant evaluation. For this reason, different statistic methods are
used to optimize the empirical data post-processing for assessing statistical
accuracy and to improve the quality of predictions for small sample size. Using
multiple regression and bootstrapping, present empirical data are analyzed.

Keywords: Multi-factor ageing
Lifetime investigation
Bootstrapping

1 Introduction

In the last few years, the use of renewable energies has steadily increased. This results
in a fluctuating load profile [1]. Power generation equipment, such as generators,
operated in the past with a constant continuous base load are subjected to an
increasingly time-dependent load, which is reflected in increased switching cycles,

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 825–837, 2020.
https://doi.org/10.1007/978-3-030-31676-1_77
826 A. Cimino et al.

whereby generators are switched off or reactivated into operation at shorter intervals
[1]. The faultless performance of the generator is of fundamental relevance for the
reliability of power supplies. The fault-critical component of generators is the stator
winding. The most common cause of stator failure is the ageing of the electrical
insulation system of generator winding bars. Up to one-third faults refer to the insu-
lation system [2]. The exact knowledge of the ageing process and possible estimation
of the residual life of the insulation system is therefore particularly required. In this
context, electrical, thermal and mechanical ageing are of major importance. In opera-
tion mode, the individual ageing factors act as simultaneous multi-stress. The three
mentioned individual factors interact in complex dependencies and influence each
other, which lead to synergy effects. The ageing process under simultaneous electrical,
thermal and mechanical stress is rarely completely known. The verification of the
theoretical model requires experimental investigations on generator winding bars.
These are considered to be cost- and especially time-intensive [1–6].
Previously, numerous series of accelerated measurements have been conducted, but
only limited lifetime data is available, which restricts the statistical assessment. For this
reason, this paper presents a statistically based method for the evaluation of investi-
gations using small sample size.

2 Statistical Fundamentals

Essential fundamentals for evaluating existing lifetime data using statistical methods
are presented in the following. Statistics is the combination of certain methods for the
analysis of empirical data [7].
• Linear regression: functional relationship between dependent variable and one or
more independent variables
• Nonlinear regression: nonlinear combination of the model parameters and depends
on one or more independent variables
• Multiple linear regression: several independent variables/functions of independent
variables
A sample (e.g. lifetime data) of size n is given. The used statistical model (e.g.
linear regression) requires estimates of the regression coefficients in order to minimize
the sum of the squares of the residuals. The usual procedure uses the requirement of
Gaussian distribution of residues. Due to the limited number of lifetime data in the
present study, the hypothesis of Gaussian distribution of residuals is not valid [7].
A method that does not require the Gaussian distribution of residuals and a minimum
number of data is the bootstrapping. Confidence intervals for the requested regression
parameters are calculated using the bootstrapping method.

2.1 Bootstrapping
Bootstrapping is a statistic method, which is based on random sampling with
replacement. Repeated statistics are calculated on the basis of only one sample.
Bootstrap methods are used typically if the theoretical distribution is not known. It is a
 Statistic Based Method for Post-processing Analysis 827

process which starts from a simple system and activates a more complex one. In this
context, it “pulls itself up by its bootstraps”, which means that it improves itself by its
own efforts. Bootstrapping enables accurate assessments (e.g. confidence intervals) to
sample estimates. Using the bootstrap distribution, parameters of a distribution are
estimated. This involves a special resampling technique, which offers several
advantages:
• Bootstrap resampling does not require any precondition for the distribution of
random variables, such as normal distribution (Gaussian).
• This method is applicable for multiple estimates without checking and developing
new distribution models and their properties.
• The sample size is not required to be as large as to have implications from the
central limit theorem.
The maximum number of possible distinct bootstrap samples is specified as Eq. (1),
which uses the binomial coefficient (left side of equation: ð2n  1Þ over n) [8].
 
2n  1 ð2n  1Þ!
¼ ð1Þ
n n!ðn  1Þ!

Figure 1 shows the influence of the sample size. For Example, the bootstrap dis-
tribution consists of 4537567650 samples from an existing data set of n ¼ 18.

Fig. 1. Distinct bootstrap samples as a function of distinct value n showing influence of sample
size
828 A. Cimino et al.

3 Experimental Investigations and Lifetime Database

3.1 Multi-factor Ageing Test Setup
In order to enable experimental investigations, a special test setup is used for defined
simultaneous multi-factor ageing of the electrical insulation system during electrical,
thermal and mechanical stress. This test setup was designed and built in [9] and is
continuously used for further multi-factor investigations [4, 10–13]. The test system
consists of three main components:
• electro-dynamic exciter
• circulating air oven (max. temperature #max ¼ 210  C)
• single-phase transformer with high voltage bushing (max. voltage Vmax ¼ 30 kV)
Figure 2 shows the schematic structure of the used test setup for simultaneous
multi-factor ageing and Fig. 3 presents the inside view of the oven with mechanical
fixture and two mounted test bars.

circulating air oven tested stator bars

mechanical
motion
high
voltage
bushing electro-
dynamic
Transformer exciter

Fig. 2. Schematic overview of the used test setup for multi-factor ageing of generator stator bars

Fig. 3. Multi-factor ageing test setup with two fixed generator stator bars
 Statistic Based Method for Post-processing Analysis 829

3.2 Data Set of Life Cycle Data from Empirical Investigations

The generation of lifetime data is particularly demanding with respect to time. Table 1
shows an overview of the research work performed in the last few years concerning
mechanical influenced lifetime investigations. As part of this work, the data are ana-
lyzed and evaluated statistically.

Table 1. Data set of multi-factor accelerated ageing tests considering variation of bending stress
while the temperature is set to 100 °C
# Voltage in kV Bending stress in N/mm2 Lifetime in h
1 26.4 39.82 66.05
2 26.4 39.82 624.82
3 26.4 51.19 33.17
4 26.4 51.19 12.37
5 26.4 62.57 3.60
6 26.4 62.57 4.83
7 26.4 73.95 0.28
8 26.4 73.95 0.93
9 26.4 85.32 0.37
10 26.4 85.32 0.03
11 22 39.68 52.45
12 22 39.68 161.18
13 22 48.49 53.18
14 22 48.49 14.60
15 22 85.32 3.13
16 22 85.32 5.97
17 22 39.82 586.93
18 22 39.82 583.25

4 Statistic Based Method for Life Cycle Analysis

and Application on Lifetime Data

The generated lifetime data is identified on the basis of empirical results (Sect. 3).
Subsequently, these data are interpreted by statistical model using (multiple) regression
and bootstrapping. The general statistic based method is shown in Fig. 4.
830 A. Cimino et al.

STATISTIC BASED METHOD

Regression linear
acquisition of empirical data

nonlinear

statistic influence of
electrical voltage

categorical
variable
Multiple
regression

parameter
bootstrapping
estimation

Fig. 4. Statistic based method for lifetime analysis

In a first step, the method “linear regression” checks whether there is a linear
correlation between lifetime L and mechanical bending stress r. The hypothesis in
Eq. (2) is given by:

L ¼ b0 þ b1
r ð2Þ

where:
L = lifetime (in min)
bj = regression coefficients, j ¼ 1; 2
r = bending stress (in N/mm2)

Equation (2) results in the regression parameter:

^ ¼ 31611 and b
b ^ ¼ 414:2
0 1

with a coefficient of determination of R2 ¼ 32%, which is quite small and requires

further investigations of the linear correlation (Fig. 5).
 Statistic Based Method for Post-processing Analysis 831

Fig. 5. Linear regression: Lifetime as a function of bending stress; R2 ¼ 32%

As the correlation between lifetime L and bending stress r is only slightly linear, a
nonlinear correlation is investigated in the following. By using the natural logarithm of
lifetime, Eq. (3) results:

lnðLÞ ¼ b0 þ b1
r ð3Þ

Equation (3) is transformed mathematically to (4):

L ¼ eb0 þ b1
r ð4Þ

The result of regression using Eq. (3) is plotted in Fig. 6. The coefficient of
determination is R2 ¼ 76%. Thus, an improvement by factor of 2.375 compared to the
linear regression is evident.
832 A. Cimino et al.

Fig. 6. Regression: natural logarithm of lifetime as a function of bending stress; R2 ¼ 76%

According to Table 1, besides the variation of the mechanical bending stress,

different electrical voltages are available, which differs by 20%. The tested generator
stator bars are designed for a rated voltage of 22 kV. Data are available for rated
voltage and a voltage 20% above. Although this influence is usually negligible, this is
investigated statistically.
The dependent variable lifetime L is analyzed using the independent metric variable
bending stress r and the independent categorical variable electrical voltage V. The
model is defined in Eq. (5):

0; for V ¼ 22:0 kV
lnðLÞ ¼ b0 þ b1
r þ b2
ð5Þ
1; for V ¼ 26:4 kV

which results in regression coefficients:

^ ¼ 14:65; b
b ^ ¼ 0:12 and b
^ ¼ 1:48
0 1 2

and is visualized in Fig. 7.

 Statistic Based Method for Post-processing Analysis 833

Fig. 7. Natural logarithm of lifetime as a function of bending stress considering categorical

variable of electrical voltage

Based on the already gained results, the natural logarithm of lifetime is used for
further multiple linear regression analysis. The model is defined as (6):

lnðLÞ ¼ b0 þ b1
r þ b2
V ð6Þ

By the use of the exponential function Eq. (6) results in Eq. (7):

L ¼ eb0 þ b1
r þ b2
V ð7Þ

Considering the model (6) as well as (7) and the multiple regression, the parameter
estimation is given by the regression coefficients:

^ ¼ 22:03; b
b ^ ¼ 0:12 and b
^ ¼ 0:34:
0 1 2

The coefficient of determination is calculated to R2 ¼ 83%. Figure 8 shows the

lifetime surface as a function of bending stress and electrical voltage.
834 A. Cimino et al.

Fig. 8. Lifetime as a function of bending stress and electrical voltage; R2 ¼ 83%

To evaluate the parameters for bending stress and electrical voltage bootstrapping
method is used. The bootstrap procedure uses the following algorithm with start value
i ¼ 0:
1. A new sample i is created from the existing data set by resampling (with return).
2. The regression coefficients b^ðiÞ ; b
^ðiÞ ; b
^ðiÞ of the new sample i (from step 1) are
0 1 2
calculated.
3. i ¼ i þ 1 for i\10000 and return to step 1.
The bootstrap algorithm distributes the estimated parameters. Figures 9 and 10
show the bootstrapping distribution as well as the 95% bootstrap confidence interval.
^ . Furthermore,
The bootstrap distribution is centred around the estimated value b j
the lower and upper bound of the 95% bootstrap confidence interval is given, which is
an estimate of the 95% confidence interval of the unknown true distribution. Therefore,
the regression coefficients are specified with 95% probability.
The confidence intervals contain the true parameters with a probability of 95%.
Using Eqs. (6) and (7) it is now possible to estimate lifetime, which depends on the
variables deflection and voltage (Fig. 8).
 Statistic Based Method for Post-processing Analysis 835

^ ¼ 0:12 for bending stress

Fig. 9. Bootstrapping distribution of parameter b 1

^ ¼ 0:34 for electrical voltage

Fig. 10. Bootstrapping distribution of parameter b2

The benefits of bootstrapping method are summarized as follows:

• For an estimate of the 95% confidence interval, only small datasets are necessary to
enable a reliable estimation.
• The standard methods for calculating confidence intervals require specifications for
the investigated variables, such as independently normal distribution. Bootstrapping
works without these additional requirements.
836 A. Cimino et al.

5 Conclusion and Outlook

The present paper evaluates empirical lifetime data using statistical methods. The linear
regression provides a coefficient of determination of 32%. The combination of the
natural logarithm and a regression results in a coefficient of determination of 76%. The
use of multiple regression including independent variable bending stress and electrical
voltage returns a coefficient of determination of 83% for the dependent variable life-
time. In addition, the variation of the parameters is investigated using bootstrapping
method. The presented statistical method for the evaluation of lifetime data allows an
estimation of lifetime depending on the variables deflection and voltage without
knowing the exact distribution of residuals, which is in particular a challenge for data
with small sample size.
In additional work the influence of bending stress and electrical voltage is con-
sidered and extended by the influence of temperature, which corresponds to a model in


the fourth dimension R4 . Furthermore, a model has to be defined which takes into
account the interaction of the individual influencing parameters.

References
1. Fölting, C.: Physikalisch basierte Nachbildung des Alterungsprozesses bei simultaner
elektrischer, thermischer und mechanischer Belastung, Ph.D. dissertation, TU Dortmund
University, Germany (2017)
2. Tavner, P., Ran, L., Penman, J., Sedding, H.: Condition of rotating electrical machines. The
Institution of Engineering and Technology (IET) (2008). https://doi.org/10.1049/PBPO056E
3. Stone, G.C., Culbert, I., Boulter, E.A.: Electrical Insulation for Rotating Machines. IEEE
Press Series on Power Engineering (2004)
4. Staubach, C., Steins, H., Cimino, A., Jenau, F.: Analysis of locally resolved dissipation
factor measurements on the insulation system of mechanically aged generator bars, ETG-
VDE Hochspannungstechnik, Berlin, Germany, pp. 568–573 (2016)
5. Fölting, C., Jenau, F.: Modellbasierte Betrachtung des mechanisch initiierten elektrischen
Bäumchenwachstums auf Basis zeitraffender Lebensdaueruntersuchungen an originalen
Generatorwicklungsstäben bei simultaner elektrischer, thermischer und mechanischer
Belastung. (engl. “Theoretical considerations on the mechanically initiated electrical treeing
based on accelerated multi-factor aging test on generator stator bars with original
dimensions”), ETG-Diagnostik elektrischer Betriebsmittel, Berlin, Germany, pp. 1–5 (2014)
6. Fölting, C., Jenau, F.: Implementation and analysis of multi-factor aging tests on the winding
insulation of original sized generator bars used in large rotating machines. In: International
Conference on High Voltage Engineering and Application, Poznan, Poland, pp. 1–4 (2014)
7. Hedderich, J., Sachs, L.: Angewandte Statistik. Springer, Heidelberg (2018). https://doi.org/
10.1007/978-3-662-56657-2
8. Efron, B., Tibshirani, R.J.: An introduction to the Bootstrap. Chapman & Hall, London/New
York (1993)
9. Fölting, C., Jenau, F.: Design und Konstruktion eines Versuchsaufbaus für Alterungsver-
suche an originalen Generatorstäben bei simultaner Belastung mit mehrfachen Einfluss-
größen“ (engl. “Design and Construction of a test setup for multi-factor aging of real
generator stator bars”), ETG-Diagnostik elektrischer Betriebsmittel, Fulda, Germany, pp. 1–
5 (2012)
 Statistic Based Method for Post-processing Analysis 837

10. Cimino, A., Staubach, C., Jenau, F.: Analysis of accelerated multi-factor aging tests on the
insulation system of generator stator bars used in large rotating machines. In: 2018 IEEE
Electrical Insulation Conference (EIC), San Antonio, TX, pp. 110–113 (2018)
11. Cimino, A., Staubach, C., Jenau, F.: Analysis of localized dissipation factor measurements
on the insulation system of mechanically aged generator stator bars. In: 2016 IEEE
International Conference on Dielectrics (ICD), Montpellier, France, pp. 674–677 (2016)
12. Cimino, A., Staubach, C., Jenau, F.: Ageing behaviour of the insulation system used in
rotating machines. In: 2017 INSUCON - 13th International Electrical Insulation Conference
(INSUCON), Birmingham, pp. 1–6 (2017)
13. Cimino, A., Staubach, C., Jenau, F.: Analysing mechanical aging behaviour of the electrical
insulation system used in rotating machines. In: International Symposium on High Voltage
Engineering (ISH), Buenos Aires, Argentina, pp. 1–6 (2017)
 Electromagnetic Wave Radiation Due
to Partial Discharges Inside Power
Transformers in the Frequency Domain

M. Akbari Azirani1(&) , M. Ariannik1, P. Werle1, and A. Akbari2

1
Institute of Electric Power Systems, Division of High Voltage Engineering and
Asset Management, Schering-Institute, Leibniz Universität Hannover,
Hannover, Germany
azirani@ifes.uni-hannover.de
2
Department of Electrical Engineering, K. N. Toosi University of Technology,
Tehran, Iran

Abstract. Reliable operation of power systems depends highly on condition of

power transformers, hence making their condition monitoring top priority.
Partial discharge (PD) is among the most common and deteriorative faults in a
power transformer. PDs cause progressive deterioration of the insulation system
of the transformers, and may lead to catastrophic failures, if no countermeasure
is taken.
PD measurements in power transformers are among well-known tools for the
evaluation of the insulation system’s condition. For this purpose, the conven-
tional method described in IEC 60270 is applied; however, the ultrahigh fre-
quency (UHF) PD measurement technique, which is based on the radiation of
electromagnetic (EM) waves due to a PD fault in very high frequency (VHF)
and UHF range, has advantages, e.g., higher robustness against external noises,
possibility of PD localization as well as its applicability for both offline and
online PD measurements.
In this contribution, UHF PD pulses are studied from an EM stand point.
A specially designed PD fault model is placed inside a transformer tank model.
The EM waves propagated due to PD are captured via UHF probes connected to
a high sampling digital storage oscilloscope and transformed into the frequency
domain using discrete Fourier transform (DFT). A Histogram of the frequency
components is presented for different UHF probe positions. The results are
verified by recording phase resolved PD patterns using both the conventional,
and the UHF measurement method. Finally, a practice-oriented analysis of
suitable and unsuitable center frequencies and bandwidths for UHF measure-
ments is presented.

Keywords: UHF measurement technique
PD diagnosis
Power

transformers
Monitoring system

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 838–849, 2020.
https://doi.org/10.1007/978-3-030-31676-1_78
Electromagnetic Wave Radiation Due to Partial Discharges Inside Power Transformers 839

1 Introduction

In order to increase the reliability of the transformers and minimize operating costs,
online condition monitoring of transformers is recommended. The condition of the
transformer insulation system is in focus when the transformer is to be monitored [1].
Since partial discharges (PD) occur before the onset of a severe damage, PD
monitoring and evaluation can be used to provide early warning of imminent failure.
Advantages such as robustness against external disturbances, the possibility of online
monitoring and localization of the PD make the UHF method seem particularly
attractive [2].
To characterize the type of PD, one of the most commonly used methods is Phase
Resolved Partial Discharge (PRPD) pattern recognition analysis. It uses both the
amplitude and phase of the voltage cycle of each individual PD pulse to obtain a
pattern. By using the PRPD method, the PD data can be interpreted. The UHF method
can also be used to record such patterns.
The UHF sensor is the main part of a UHF measuring system. Despite different
versions of the sensors for use in power transformers [3], UHF sensors for standard oil
valves are particularly widespread.
In this contribution, all parameters that influence the sensitivity and calibration of
this method are discussed in detail and analyzed by performing laboratory measure-
ments. Furthermore, the correlation between the electrical PD measurement according
to IEC 60270 and the UHF measurement method is investigated. Moreover, the
robustness of the UHF measurement method against external disturbances is assessed
by means of measurements on a transformer [4].

2 PD Pulses in Frequency Domain

In this section, the characteristics of PD signals are investigated using an internal PD

fault model for generation of PD pulses.

2.1 Measurement Setup

For the investigations conducted in this contribution, a transformer tank model with six
oil drain valves is utilized. This tank is equipped with two OMICRON UVS610 valve
sensors, which are conical monopole antennas [5]. The transformer tank model,
depicted in Fig. 1, has a length, width, and height of 1236 mm, 856 mm, and
1596 mm, respectively. High-voltage is introduced into the interior of the tank via its
123 kV bushing. The UHF sensors are inserted through the lowest and highest oil drain
valves with an insertion depth of 6 cm.
840 M. A. Azirani et al.

Fig. 1. Schematic illustration of the transformer tank model.

An internal PD fault model is utilized to generate PD pulses. The model consists of

two electrodes with Rogowski profiles encapsulated with epoxide resin. The solid
insulation material incorporates a centered fine drilled hole with a diameter of 1 mm up
to the midpoint of the electrodes. The PD fault model used in this contribution is
visualized schematically in Fig. 2.
Electromagnetic Wave Radiation Due to Partial Discharges Inside Power Transformers 841

Fig. 2. Schematic view of the internal PD fault model.

The particular feature of this fault model, which makes it suitable for the mea-
surements conducted in this contribution, is that it generates a pseudo identical phase
resolved partial discharge (PRPD) pattern with similar PD intensity. The PRPD pattern
at a voltage slightly above 20 kV is shown in Fig. 3.

Fig. 3. Electrical PRPD pattern of the utilized PD fault model.

The patterns in both positive and negative half cycles are nearly identical with PD
starting at the zero-crossing and continuing to the extrema of the voltage, clearly
indicating a void or cavity in the solid insulation [6].
842 M. A. Azirani et al.

The test circuit used for capturing PD pulses is shown in Fig. 4. The UHF probes
are connected to a 10 GS/s digital storage oscilloscope with a maximum bandwidth of
3.5 GHz. Simultaneously, the PD intensity is monitored using a PD measurement
circuit according to IEC 60270.

Fig. 4. Overview of the test setup for PD measurements.

The introduced PD fault model, which is depicted in Fig. 2, is placed at the center
of the tank model. The upper electrode is connected to a metal rod enclosed by the
bushing and hence, placed on high-voltage, while the lower electrode is grounded.

2.2 Frequency Analysis

In order to analyze the electromagnetic waves propagated by the PD fault model, 100
data sets are captured and converted into the frequency domain using discrete Fourier
transform (DFT) [7]. The resulting signals in the frequency domain are added together
and normalized. The resulting signal is demonstrated in Fig. 5.
Electromagnetic Wave Radiation Due to Partial Discharges Inside Power Transformers 843

1
Probe 1
Probe 2
0.8
Normalized Amplitude

0.6

0.4

0.2

0.2 0.4 0.6 0.8 1

Frequency [GHz]

Fig. 5. Captured PD pulses of the internal discharge model in frequency domain with two UHF
probes.

The results, presented in Fig. 5, can be interpreted by dividing the frequency range
to four categories as listed in the following:
1. Below 160 MHz
The tank of a transformer is a metallic enclosure and acts as a rectangular cavity.
Each cavity has a dominant mode, which is a function of its size, form, and the medium
with which it is filled. Considering the transformer tank model in this contribution, the
dominant frequency equals 160 MHz. Theoretically, a signal with a frequency below
this value will be attenuated swiftly [8]. The dominant mode of a rectangular
waveguide depends on its size and the medium, in which the EM waves propagate. The
frequency components observed in the frequency range below 160 MHz result from
the bushing conductor that protrudes into the tank. In this case, the PD fault model acts
as the excitation source of the bushing conductor, which in this context can be con-
sidered as a monopole antenna. Due to the relative small size of the tank, the signals in
this frequency region are still picked up by the probes, before getting damped. This
region is not suitable for UHF measurements.
844 M. A. Azirani et al.

2. Between 160 MHz and 300 MHz

The frequency components in this category are due to the electromagnetic waves
generated by the PD fault model. The maximum is observed at a frequency around
238 MHz for both probes. It is noteworthy that the frequency of the component with
the maximum amplitude depends on many factors, such as the size of the transformer,
the position of the probe, the antenna characteristics of the probe, the insulating
medium, etc.; however, this range is recommended for performing UHF measurements.
3. Between 400 MHz und 700 MHz
The frequency components in this frequency range are mainly the reflections of the
previous category. The shortest path between the PD fault model and probe 1 is the
direct uninterrupted path. This is rarely the case in the practice, since the active part of
the transformer usually disturbs the direct path. By considering the walls of the tank,
further traveling paths of the UHF signals can be calculated. For performing UHF
measurements, this region is also suitable. The highest peak is observed at 468 MHz.
4. Above 700 MHz
There are barely any frequency components observed in this frequency range. Since
the PD fault generates signals in the lower frequency bands, only reflections could be
observable in the higher frequency bands; however, since by each reflection, the signals
lose a fraction of their energy, they are attenuated. Hence, this frequency range is not
suitable for the UHF measurements.
Based on the frequency analysis of the UHF pulses in this section, certain regions
were recommended for performing UHF measurements. In order to investigate the
eligibility of the selected frequency regions, UHF PRPD measurements are conducted,
and the results are presented in the following section.

3 UHF PRPD Pattern

A PRPD pattern is obtained by depicting the intensity of the PD pulses measured by the
conventional method in regard to the phase angle of the applied voltage. Similarly, the
amplitude of the signals, received via a UHF probe, can be displayed with respect to the
phase of the power frequency of the AC cycle. The PRPD patterns attained by electrical
PD measurements enable the operators to differentiate various types of PD [6]. During
on-site PD measurements, the background noise level sometimes does not allow
obtaining an electric PRPD pattern. The UHF PRPD pattern is applicable in such cases
to determine the PD type [9], since the UHF measurement technique offers a higher
robustness against external disturbances [2].
UHF PRPD measurements are conducted to verify the results of the investigation in
the previous section.
Electromagnetic Wave Radiation Due to Partial Discharges Inside Power Transformers 845

3.1 Measurement Setup

The test circuit for these measurements is shown in Fig. 6.

Fig. 6. Overview of the test setup for UHF PRPD measurements.

The established electrical PD measurement according to IEC 60270 is applied. The

detection frequency is set from 100 kHz to 400 kHz. Furthermore, probes are inserted
through the same oil drain valves as the previous section for capturing UHF pulses.
Using the bandwidth converter OMICRON UHF620, the PD measurement range is
extended to 100 MHz to 2 GHz. Hereby, the measurement bandwidth can be set to
70 MHz for a mediumband measurement or to 1.9 GHz for a broadband measurement.
The introduced internal PD fault model is placed in the center of the tank, similar to the
previous section.
For the results presented in this section, the top row shows the conventional
electrical PRPD pattern, while row 2 and 3 show the UHF PRPD pattern for probe 1
and probe 2, respectively, as defined in Fig. 1. Each pattern shows the recorded PD
pulses for a duration of 90 s.
846 M. A. Azirani et al.

3.2 Broadband Measurement

The result of the broadband UHF PRPD measurement is shown in Fig. 7.

Fig. 7. PRPD patterns of a broadband UHF measurement.

The resulting PRPD pattern for a broadband UHF measurement in Fig. 7 indicates
that while a certain correlation between the electrical PRPD pattern and the UHF PRPD
patterns in terms of the phase angle of the PD pulses is observed, the intensity of the
UHF pulses do not conform to the electrical PD pulses.
In case the bandwidth is set to 1.9 GHz, although UHF PD pulses are captured,
their amplitude is incorrect. The reason is that the amplifier settings of the measurement
system cannot be adjusted properly, either leading to overdriving or insufficient sen-
sitivity. Although a broadband UHF measurement can be applied for the detection of
PD, it is not suitable for PRPD measurements.

3.3 Mediumband Measurement

For a Mediumband UHF measurement, the bandwidth is set to 70 MHz. The quality of
the UHF PRPD pattern depends highly on the chosen center frequency. In the fol-
lowing, different cases are presented to demonstrate the importance of the chosen
center frequency.
Suitable Frequency Ranges
Based on the results presented in Sect. 2.2, categories 2 and 3 were recommended for
UHF measurements. For these frequency regions, the highest peak is set as the center
frequency for the UHF PRPD measurements. Hence, the corresponding center fre-
quencies for categories 2 and 3 are 238 MHz, and 468 MHz, respectively. The results
are shown in Figs. 8 and 9.
Electromagnetic Wave Radiation Due to Partial Discharges Inside Power Transformers 847

Fig. 8. PRPD patterns of a mediumband UHF measurement with a center frequency of

238 MHz.

Fig. 9. PRPD patterns of a mediumband UHF measurement with a center frequency of

468 MHz.

The UHF PRPD patterns correspond in both cases with the electrical pattern.
The UHF PRPD patterns obtained at the center frequency of 238 MHz, the number of
UHF pulses is less than the measurement with a center frequency of 468 MHz. This
agrees with the frequency spectrum of the UHF pulses as depicted in Fig. 5. Since the
number of the peaks in category 3 is noticeably higher than category 2, the number of
captured UHF PD pulses reflects this fact.
Unsuitable Frequency Ranges
As examples of unsuitable frequency ranges for conducting UHF measurements, two
center frequencies are investigated. According to the presented frequency spectrum, the
848 M. A. Azirani et al.

region between category 2 and category 3 as well as the region defined as category 4 do
not include any peaks. For illustration purposes, the center frequencies 350 MHz and
800 MHz are selected. The resulting PRPD patterns are depicted in Figs. 10 and 11.

Fig. 10. PRPD patterns of a mediumband UHF measurement with a center frequency of
350 MHz.

Fig. 11. PRPD patterns of a mediumband UHF measurement with a center frequency of
800 MHz.

When the chosen frequency is unsuitable, no UHF pulses are captured via the
probes. This might lead to a misinterpretation of the result and the device under test
could inaccurately be declared as PD-free. In order to avoid such cases, the frequency
Electromagnetic Wave Radiation Due to Partial Discharges Inside Power Transformers 849

range for performing UHF measurements needs to be chosen carefully by considering

the specific test object and selecting an appropriate center frequency and bandwidth.

4 Conclusion

In this contribution, captured UHF pulses of an internal PD fault model were converted
into the frequency domain using DFT. The resulting frequency components were
analyzed and suitable frequency ranges for conducting UHF measurements were rec-
ommended. The selected ranges were verified by performing UHF PRPD measure-
ments in the transformer tank model. The difference between suitable and unsuitable
frequency ranges was demonstrated.

Acknowledgements. The authors would like to express their sincere gratitude towards Mr.
Abdullah Malik Ibrahim for his cooperation in this research.

References
1. Chakravorti, S., Dey, D., Chatterjee, B.: Recent Trends in the Condition Monitoring of
Transformers. Theory, Implementation and Analysis. Power Systems. Springer, London
(2013)
2. Akbari, A., Werle, P., Akbari Azirani, M., Mirzaei, H.R.: Challenges in calibration of the
measurement of partial discharges at ultrahigh frequencies in power transformers. IEEE
Electr. Insul. Mag. 32, 27–34 (2016)
3. Sinaga, H.H.: Detection identification and localization of partial discharges in power
transformers using UHF techniques. Ph.D. thesis, The University of New South Wales
Australia (2012)
4. Akbari Azirani, M., Werle, P., Akbari, A., Jahangir, H., Szczechowski, J.: An investigation on
the relation between PRPD patterns acquired by conventional and UHF nonconventional PD
measuring technique for power transformers. In: International Symposium on High Voltage
(2017)
5. OMICRON electronics GmbH: MPD 600. www.omicronenergy.com/en/products/mpd-600/.
Accessed 15 May 2019
6. Fuhr, J., Aschwanden, T.: Identification and localization of PD-sources in power-transformers
and power-generators. IEEE Trans. Dielect. Electr. Insul. 24, 17–30 (2017)
7. Oppenheim, A.V., Willsky, A.S., Hamid Nawab, S.: Signals and Systems. Always Learning,
2nd edn. Pearson Education, Harlow (2013)
8. Schelkunoff, S.A.: On representation of electromagnetic fields in cavities in terms of natural
modes of oscillation. J. Appl. Phys. 26, 1231–1234 (1955)
9. Akbari Azirani, M., Werle, P., Akbari, A., Jahangir, H., Szczechowski, J.M. (eds.): An
investigation on PRPD patterns generated by the UHF measurement technique for power
transformers. In: IEEE Iranian Conference on Electrical Engineering (2017)
 Simulation of the Mechanical Vibrations
in a High Voltage Transformer

Chunhui Gu(&), Lin Gan, Wenxiong Mo,

Yong Wang, Yi Rao, and Yu Qin

Guangzhou Power Supply Co., Ltd., China Southern Power Grid,

Guangzhou, China
goochunhui@qq.com

Abstract. Winding fault is one of the components that cause more problems in
the transformer. The main faults of transformer winding are irreversible changes
in the mechanical structure of winding under the action of electromagnetic force
or mechanical force. Common faults include loosening, warping, bulging and
dislocation of winding. The live detection technology of transformer winding
deformation based on noise and vibration is of great significance. In this paper,
the mechanism and characteristics of transformer vibration are studied firstly.
The vibration of transformer is caused by the vibration of its main body (core,
winding, etc.) and the vibration of cooling device. The core vibration mainly
comes from magnetostriction of silicon steel sheet, which is expressed by
magnetostriction ratio. Through mathematical deduction, it is found that the
vibration acceleration is based on two times of the power supply frequency, and
the amplitude is proportional to the square of the voltage. Winding vibration is
closely related to structural stiffness and dynamic characteristics. When the
transformer winding coil flows through the load current, due to the existence of
leakage magnetic field, dynamic electromagnetic force is generated between
windings, wire cakes and turns, which causes winding vibration. In the study of
the mechanism and characteristics of transformer vibration, the axial structure of
winding coil is simplified to MASS-SPRING-DAMP model. The upper and
lower pressure plates are considered as rigid bodies and fixed constraints. Each
layer of wire cake is equivalent to a mass module, and the insulation pad is
equivalent to an elastic element. The distributions of radial and axial electro-
magnetic force and acceleration are obtained by dynamic simulation analysis.

Keywords: Transformer
Simulation
Mechanical vibration

1 Introduction

The normal operation of power transformer plays a decisive role in the reliable
transmission and flexible distribution of electricity in the power grid. When the power
transformer is in sudden accident during operation, it may cause casualties and large-
scale power outage [1]. One of the main faults of transformer is transformer winding
deformation, which shall cause cumulative effect, inter-turn short circuit and insulation
damage [2]. The vibration of transformer, which is closely related to the internal
mechanical state, contains abundant state information. The on-line testing method of

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 850–861, 2020.
https://doi.org/10.1007/978-3-030-31676-1_79
 Simulation of the Mechanical Vibrations in a High Voltage Transformer 851

winding deformation based on noise and vibration has strong anti-interference, and
does not affect the internal structure of transformer. Meanwhile, the testing system can
be realized on-line detection, with the merits of small size and portability. Therefore,
it is of great importance to carry out research on the on-line detection technology of
transformer winding deformation based on noise and vibration.

2 Research on Mechanism and Characteristic

of Transformer Vibration

Transformer vibration is caused by the vibration of its main body (core, winding, etc.)
and cooling device. The deformed transformer winding will cause the change of
leakage magnetic field, geometric shape and natural mode frequency, and furtherly
effect the vibration signal transmitted to the surface of fuel tank [3, 4]. So the defor-
mation of transformer winding can be detected by vibration signal.

2.1 Mechanism of Winding Vibration

Because of the leakage magnetic field in the transformer, the winding coil with
alternating current during normal operation will produce axial vibration due to the
electromagnetic force. In the research of the axial dynamics of windings, the axial
structure of pie coils is often simplified to a mass-spring mathematical model as shown
in Fig. 1. In this mathematical model, the upper and lower pressure boards of windings
are considered as rigid bodies with fixed positions and no displacement; each layer of
wire cake can be equivalent to a mass module; and the insulation pads between the
winding wire cake can be considered as elastic elements.

T xT=0

K0 C0

m1 F1
x1

K1 C1

m2 F2
x2

mn-1 Fn-1
xn-1

Kn-1 Cn-1

mn Fn
xn

Kn Cn

B xB=0

Fig. 1. Mass-spring mathematical model of pie winding

852 C. Gu et al.

In Fig. 1, mi is the mass of layer i, Fi is the electromagnetic force on layer i, xi is

the displacement of layer i, Ki is the elastic coefficient and Ci is the damping coeffi-
cient. The displacement, velocity and acceleration of the winding are calculated and
analyzed by establishing and solving the differential equations of motion.
Assuming that the power transformer operates at a load current of
I ¼ Im
cosðwt þ w0 Þ, among them: Im – load current amplitude; w0 – initial phase.
Then the electromagnetic force acting on the winding can be calculated by the
following formula:


1 1
F ¼ pIm2 þ cosð2xt þ 2u0 Þ ð2  1Þ
2 2

d2 x dx
M þ C þ Kx ¼ F þ Mg ð2  2Þ
dt2 dt

The fourth-order Runge-Kutta method is used to solve the differential equations.

Simplified winding vibration acceleration can be obtained.

d2 x
a¼
dt2 ð2  3Þ
¼  x2a Ae2M sinðxa t þ aÞ  pIm2 D sinð2xt þ 2u0 þ bÞ
Ct

According to formula (2-3), the vibration acceleration of transformer winding

consists of a steady state and a attenuation with a frequency twice the power supply
frequency. Therefore, the actual steady-state operation of the power transformer
vibration signal only contains the steady-state component, that is, the transformer
winding vibration signal is twice the power frequency (domestic 100 Hz) as the basic
frequency of vibration.

2.2 Radial Vibration Mechanism of Winding

The winding is circumferentially divided by cushion blocks to form a line turn with two
ends supported by cushion blocks and braces and suspended in the middle. The radial
vibration of line turns under periodic force is regarded as in-plane bending vibration of
n  w. Among them, r is the average radius of line turns, n is
arc arch with arc length 2pp
the number of cushions and w is the width of cushion blocks, as shown in Fig. 2.
 Simulation of the Mechanical Vibrations in a High Voltage Transformer 853

Fig. 2. Schematic diagram of winding and turn structure

2π r
−w
n
pr
y
R
x
L

Fig. 3. In-plane bending vibration model of two-hinged arch

According to the plane deflection equilibrium equation of the arch and the inertia
force of Darumbell, the vibration equation of two-hinged circular arch with equal cross-
section under the condition of radial force density and simple harmonics can be
obtained (Fig. 3).

 
  
1 @3 @w @ 2 v 1 @ @w @ 2 v @ @2w @pr
EI þ 2 þ 4 EI þ 2 þ q0 S 2 þ ps 
R @h
4 3 @h @h R @h @h @h @h @t @h
¼0
ð2  4Þ

Boundary conditions:


0 EI d 3 w dw
w ¼ 0; w ¼ v ¼ 0; M ¼  2 þ ð2  5Þ
R dh3 dh

The periodic vibration response of line turns can be obtained by modal superpo-
sition method. If the deflection of the line turn under static load is approximated as the
vibration response of the line turn under harmonic load, then the maximum deflection y
of the line turn center can be obtained as follows:
854 C. Gu et al.

h q  i
yðx; tÞ ¼ x4 þ 2Lx3  L2 x2 sinðxtÞ ð2  6Þ
24EI
Considering the distribution characteristics of electromagnetic force in transformer
windings, the radial vibration acceleration amplitude of the line cake on the same axis
decreases gradually from the middle to the end of the winding, and the vibration
amplitude in the same arc arch decreases gradually from the center of the arch to the
cushion block. The frequency of the vibration acceleration and the electromagnetic
force is the same as 100 Hz.

2.3 Vibration Mechanism of Iron Core

The core vibration of power transformer is mainly caused by the phenomenon that the
shape of the silicon steel sheet of the transformer core changes due to excitation
(magnetostriction phenomenon) and the electromagnetic force caused by magnetic
leakage at the joint of the core. Due to the progress of modern transformer manufac-
turing technology, the stacking method of core is optimized and improved. The
vibration signal of core caused by magnetic leakage between stacks is small and almost
negligible, while the energy of magnetostriction phenomenon is large and difficult to
control. Therefore, it is considered that the vibration of transformer core is basically
determined by the magnetostriction degree of silicon steel sheet.
The core vibration is mainly determined by the magnetostriction of silicon steel
sheet. In general, the magnetostriction is solved by the following formula (2-7):
Formula (2-7): e – Magnetostriction ratio, DL – the variation of the length of silicon
steel sheet during excitation, L – the length of silicon steel sheet in the original state.
The vibration acceleration of core silicon steel sheet caused by magnetostriction is
as follows:

d 2 DL 2Les Us2
ac ¼ ¼ cos 2xt ð2  7Þ
dt 2
ðNcore S
Bs Þ2

Formula (2-7) shows that the value of vibration acceleration is proportional to the
square of voltage. If the power supply frequency of the transformer is 50 Hz, the main
body vibration of the transformer core silicon steel sheet is 100 Hz, which is twice the
power supply frequency.

3 Simulation and Analysis of Vibration Model of 110 kV

Transformer Winding

3.1 Vibration Model of 110 KV Transformer Winding

The solid three-dimensional model of 110 kV power transformer with three-phase
double windings is established by using the three-dimensional visualization entity
simulation software Inventor. As shown in Fig. 4, the geometric size information is
shown in Table 1. In the process of modeling, the transformer structure is appropriately
 Simulation of the Mechanical Vibrations in a High Voltage Transformer 855

simplified, the transformer core is simplified by using square pillars, and the winding
turns of the same cake are equivalent to continuous entities. There is no electromag-
netic wire connection between the cakes. The winding coil is divided into several
sections from top to bottom, and the windings between adjacent sections are composed
of pads between cakes. Transformers with multi-turn coils in the same section are
modeled as integral coils and four insulating pads are equivalent to one with the same
height. Among them, 16 groups of insulating pads are evenly distributed in the high
voltage windings to form the inter-section oil channel.

Fig. 4. Three-dimensional model of 110 kV transformer

Table 1. Geometric dimensions of 110 kV power transformers

Parameter Value
Type of Transformer SF9/25000
Cooling mode Self cooling
Impedance voltage/% 10.74
Core window height/mm 1230
Low Voltage Winding Internal Diameter/mm 276
Turn Number of Low Voltage Winding 141
High Voltage Winding Internal Diameter/mm 371.5
Turn Number of High Voltage Winding 938
Low Voltage Winding Height/mm 1025
Young’s modulus of cushion block/MPa 300
Core radius/mm 260
Outer Diameter of Low Voltage Winding 335
Number of Cakes in Low Voltage Winding 74
External Diameter/mm of High Voltage Winding 457
Cake Number of High Voltage Winding 80
High Voltage Winding Height/mm 999
Poisson’s ratio of cushion block 0.33
856 C. Gu et al.

ANSYS MAXWELL is used to calculate the leakage magnetic field and simulate
the structure dynamics of the intermediate phase windings in three-phase windings.
According to the structure and magnetic field distribution characteristics of transformer,
the following assumptions are made for calculating the leakage magnetic field of
transformer windings:
(1) Consider the ampere-turn balance in the winding, that is to say, the number of
turns at different wire cakes of the winding is the same;
(2) Simplify the structure of oil passage of winding and consider that the size of oil
passage is the same.
(3) The influence of cushion block and other accessories on the calculation of
magnetic field is neglected, and 1/2 model is selected according to the symmet-
rical structure of winding to calculate the eddy current field.
According to the material property settings of the 10 kV power transformer in the
previous section, the simplified calculation model of the leakage magnetic field of the
winding and the electric force of the wire cake is shown in Fig. 5.

Fig. 5. Leakage magnetic field calculation model

In the simulation process, the condition of rated current should be taken into
account, so the effective values of the coils of high voltage winding and low voltage
winding are 119.29 A and 793.65 A respectively, and the sinusoidal excitation current
with initial phase difference of 180° is applied; the symmetrical boundary conditions
are added to the left side of the solution region, and the impedance boundary conditions
are adopted in the other three directions to simulate the restraint effect of the box on the
magnetic field. Fifty Hz, the meshing condition of electromagnetic field analysis is
selected, and the model is subdivided by adaptive meshing method. The results are
shown in Fig. 6. The meshing results meet the accuracy requirements of leakage
magnetic field simulation calculation.
 Simulation of the Mechanical Vibrations in a High Voltage Transformer 857

Fig. 6. Partitioning result

3.2 Calculation Results of Magnetic Field

The calculation results of magnetic field in windings and cores are shown in Fig. 7. The
magnetic field intensity in cores is about 1.6T, and the middle guard of high and low
voltage windings is in the high magnetic field intensity region.

Fig. 7. Magnetic field calculation results

3.3 Axial Electrodynamic Force Distribution of High Voltage Winding

According to the left-hand law, the direction of electromagnetic force on the winding
can be judged. The axial force on the transformer winding produces a tendency of
compression from both ends to the middle. The radial force makes the high-voltage
winding expand outward and away from the core pillar.
The distribution of electromagnetic force vector on the upper half of transformer
high voltage winding is shown in Fig. 8. It can be seen from the figure that, consistent
with the theoretical analysis, the axial force makes the high-voltage winding com-
pressed. The maximum value of the axial force mainly appears at the end of the
winding, and the axial electromagnetic force in the middle of the winding is small. The
axial electric force at the part of the wire cake is opposite to the end direction, which
makes the part of the winding compressed towards the end of the winding, which is
slightly different from the results of the 10 kV two-dimensional calculation.
858 C. Gu et al.

Fig. 8. Electromagnetic force distribution of high voltage winding

3.4 Electrodynamic Force Distribution of Low Voltage Winding

According to the left-handed rule, it is easy to judge the direction of the low-voltage
winding of the transformer. The axial force makes the winding compressed, and the
amplitude force makes the winding close to the core column and away from the high-
voltage winding. At the same time, due to the small amplitude size of low-voltage
windings, and more concentrated in the high magnetic field intensity area between high
and low-voltage windings, the axial electrodynamic force of low-voltage windings is
also greater than that of high-voltage windings.
The distribution of electromagnetic force vector on the upper half of transformer
low-voltage winding is shown in Figs. 9 and 8. It can be seen from the figure that,
consistent with the theoretical analysis, the axial force makes the low-voltage winding
compressed. The maximum of the axial force mainly appears at the end of the winding,
and the axial electromagnetic force in the middle of the winding is smaller. The low-
voltage windings are compressed in the axial direction.

Fig. 9. Electromagnetic force distribution of low voltage winding

 Simulation of the Mechanical Vibrations in a High Voltage Transformer 859

4 Dynamic Analysis of Winding Structure

4.1 Modeling
The 110 kV high voltage winding model is established by using the simplified cal-
culation method of the model and the property of material setting in the previous
section. In the process of establishing the model, according to the simulation of the
axial vibration of the winding in the previous section and the simulation results of the
axial electrodynamic force in this section, it is found that the winding coil cake located
in the middle of the winding axis receives less axial force and is difficult to vibrate.
Therefore, the winding model can be properly divided along the middle symmetric
plane, and only one-second model can be retained, as shown in Fig. 10.

Fig. 10. High voltage winding 1/2 model

4.2 Boundary Conditions Setting

According to the structure of transformer winding, axial force and appropriate sim-
plification principle, boundary conditions are applied to the dynamic model of trans-
former winding axial vibration. Fixed constraints are applied to the top and bottom of
1/2 model, displacement constraints are applied to other pies and cushions, and only
axial degrees of freedom are retained. The axial force of each coil cake is applied to
each coil cake to solve the problem.

4.3 Result Analysis

The displacement cloud of the 1/2 model of high voltage winding is shown in Fig. 11.
It can be seen from the figure that the maximum displacement of the 1/2 model is about
1/3 of the model, that is to say, for full-size transformer windings, the maximum
displacement of vibration is about 1/6 of the top and bottom of the winding. At this
860 C. Gu et al.

Fig. 11. 110 kV winding displacement nephogram

point, the axial vibration acceleration of the winding wire cake will reach the maximum
value.
For the 1/2 model, the maximum displacement under the action of 100 Hz elec-
trodynamic force is about 8.3  10−5 mm, which is about 0.03 m/s2 if converted into
acceleration.
Through the dynamic simulation of 110 kV transformer winding, it is found that
the vibration acceleration will be affected by the winding preload, oil temperature and
operation life, and show the same influence law. In the axial vibration model of
transformer windings, the value of Young’s modulus is closely related to its vibration
amplitude, and the Young’s modulus of windings is affected by many non-fault factors.
The Young’s modulus increases with the increase of pre-tightening force, while the
Young’s model decreases with the increase of temperature and aging time.

5 Conclusions

In this paper, the mechanism and characteristics of transformer vibration are studied,
including the vibration mechanism of winding and core. The vibration model of
110 kV transformer winding is simulated and analyzed. The vibration model of 110 kV
transformer winding is established. The distribution of transformer magnetic field and
the electrodynamic force of high and low voltage windings are calculated, and the
dynamic analysis of winding structure is carried out. The conclusions are as follows:
(1) The fundamental frequency of transformer winding vibration signal is 2 times of
the power supply frequency (domestic 100 Hz). In the radial vibration of wind-
ings, the frequency of vibration acceleration and electromagnetic force is 100 Hz.
(2) The value of vibration acceleration is proportional to the square of voltage. The
main body vibration of transformer core silicon steel sheet is twice the frequency
of power supply (100 Hz) as the fundamental frequency.
 Simulation of the Mechanical Vibrations in a High Voltage Transformer 861

(3) The magnetic field intensity in the core is about 1.6T, and the middle guard of
high and low voltage windings is in the high magnetic field intensity region.
(4) The axial force makes the high-voltage winding compressed. The maximum of the
axial force mainly appears at the end of the winding, and the axial electromagnetic
force in the middle of the winding is small.
(5) For full-scale transformer windings, the maximum vibration displacement is
located at 1/6 of the winding distance from the top and bottom.

References
1. Weiser, B., Pfutzner, H., Anger, J.: Relevance of magnetostriction and forces for the
generation of audible noise of transformer cores. IEEE Trans. Magn. 36(5), 3759–3777
(2000)
2. Zheng, J., Wang, J.D., Guo, J., et al.: Vibration analysis of power transformer cores.
J. Electronic Meas. Instrum. 24(8), 763–768 (2010)
3. Yu, X.B., Jiang, Z.X., Han, T.: Noise of transformer and its reduction. Noise Vibr. Control 21
(5), 35–38 (2001)
4. Zhu, Y.Y., Ji, S.C., Zhang, F., et al.: Vibration mechanism and influence factors in power
transformers. J. Xi’an Jiao Tong Univ. 49(6), 115–125 (2015)
 Non-destructive Testing Method for Composite
Insulators Based on Digital Shearography

Lishuai Liu1(&), Chenjun Guo2, Liming Wang1, and Hongwei Mei1

1
Graduate School at Shenzhen, Tsinghua University, Shenzhen 518055, China
lls17@mails.tsinghua.edu.cn
2
Electric Power Research Institute, Yunnan Power Grid Co., Ltd.,
Kunming 650217, China

Abstract. Due to the special characteristics of materials, structures and manu-

facturing processes, composite insulators are prone to generate internal defects
during the production process and on-site operations. In this paper, an internal
defect detection method for composite insulators based on digital shearography is
proposed to eliminate faulty insulators in production and during operation. This
method diagnoses the defects of composite insulators and the severity by mea-
suring surface abnormal deformation caused by internal defects. Compared with
other non-destructive testing techniques, it has the characteristics of high speed,
high accuracy, non-contact, and full-field imaging. Insulator samples with man-
made defects were examined, and the defect information was quickly and intu-
itively detected by the interference fringe pattern and phase profile obtained. The
feasibility of digital shearography for the detection of internal defects in com-
posite insulators was verified by artificial defects detection experiments.

Keywords: Composite insulator
Nondestructive testing
Digital

shearography

1 Introduction

Composite insulators are widely used in power transmission projects, especially UHV
transmission lines, because of their high specific strength, high anti-pollution capa-
bility, ease of operation and maintenance, and difficulty in damage during trans-
portation and installation. However, due to the special characteristics of materials,
structures and manufacturing processes, composite insulators are prone to generate
internal defects during the production process and on-site operations. In operation, the
composite insulator is under the action of electric field, acid rain, salt spray, mechanical
load, ice coating, wind pressure, and sudden temperature change. The internal defects
will further expand, resulting in insulator breakdown, flashover, and internal discharge
ablation channels, aging of silicone rubber, reduction of mechanical strength, and even
string breakage of insulators [1]. Therefore, it is urgent to study a fast, accurate and
effective internal defect detection method for composite insulators. It can be used in the
production process and operation of composite insulators to detect hidden faults as
early as possible, and to prevent the insulators failure and ensure the safety of the
power grid.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 862–870, 2020.
https://doi.org/10.1007/978-3-030-31676-1_80
 Non-destructive Testing Method for Composite Insulators 863

At present, the detection methods of composite insulator defects that have been
formed at home and abroad include ultrasonic testing, electric field measurement, infrared
temperature measurement, steep wave testing, improved water diffusion testing, and so
on. However, most of these methods are complex, less sensitive, time-consuming, and
vulnerable to environmental conditions, and have not been applied in large quantities. In
recent years, scholars in related fields have proposed many new detection techniques for
the detection of composite insulator defects such as X-ray, ultrasonic phased array,
photoelectric field sensor, microwave, ultrasonic guided wave, terahertz time-domain
spectroscopy, etc. [2–5]. These techniques have achieved a series of positive results, and
have certain implications for the detection of composite insulator defects.
Digital shearography has the characteristics of high speed, high accuracy, non-
contact, and full-field imaging. It has wide applications in the field of stress mea-
surement, vibration analysis, non-destructive testing, etc. It is increasingly concerned
and valued by academics and industry. Digital shearography has been proved to be
particularly suitable for defect detection of composite materials. It has been applied to
aerospace and defense industries, such as aviation tire testing, aerospace honeycomb
structure inspection, rocket pillar coating quality inspection, turbine blade inspection,
etc. [6–8]. The application of digital shearography in the detection of power equipment
defects has not yet been reported. In this paper, the feasibility of digital shearography to
detect the internal defects of composite insulators is studied by artificial defects
detection experiments.

2 Theoretical Background

The principle of digital shearography is to measure the surface displacement of an

object under different loading conditions. The detected object is deformed to a certain
degree by pressure change, heating, etc. If there is a defect inside the object, the stress
concentration caused by the defect makes the deformation of the part inconsistent with
the deformation of the normal area, and the abnormal deformation of this defect
position will change. In order to measure the optical path difference of the optical path,
it is displayed by means of imaging. Detection principle and detection system diagram
shown in Fig. 1. A laser light source is used to irradiate the surface of the detected
object and is imaged by an imaging system composed of a CCD camera and a dis-
location imaging device. The misalignment imaging device used can divide the inci-
dent light into two light beams with a certain angle, and record the two misaligned
images on the CCD camera through the optical lens, that is, a point on the object can be
photosensitive in the CCD. The image on the surface is two points; in turn, the
misalignment device brings the two non-parallel beams scattered at two different points
on the surface of the object into approximately collinear lines and interfere with each
other, that is, each point on the image corresponds to two points on the object. The
distance between the two pairs of points and the plane distance along the misalignment
is defined as the amount of misalignment. Since the laser is highly coherent light and
the object to be detected usually has an optically rough surface, the two misaligned
images on the imaging plane interfere with each other to form a random speckle
pattern. After the object is deformed by pressure or heat, the speckle pattern will also
864 L. Liu et al.

change. In the detection, the two speckle images before and after the deformation are
digitized by the image acquisition system, and the interference fringe pattern is
obtained by computer operation. The level of the interference fringe reflects the
deformation displacement partial derivative of the detected object in the misalignment
direction.
The intensity distribution I of the shearing speckle image can be expressed as
follows:

I ¼ I0 ð1 þ c cos /Þ ð1Þ

Where, I0 is an object image (can be regarded as a DC component), c is a mod-

ulation amplitude, and / is a random phase angle. Formula (1) illustrates that the object
image is modulated into a random speckle field.
When an object is deformed by an external force or heated, the optical path dif-
ference between the two interference points on the imaging surface changes, so the
phase difference between these two points becomes the relative phase change of the
two points before and after the deformation / þ D, and D the deformation of the two
points is characterized. Relative displacement information caused. The deformed
speckle field can be expressed as:
0
I ¼ I0 ½1 þ c cosð/ þ DÞ ð2Þ

Fig. 1. Principle and experimental setup of digital shearography.

The two speckle patterns before and after the deformation are subtracted to obtain
the following expression:
    
 0  D D 
  
Id ¼ I  I ¼¼ 2I0 c sin / þ sin ð3Þ
2 2 
 Non-destructive Testing Method for Composite Insulators 865

The subtracted fringe pattern characterizes the distribution of relative phase

changes. When D ¼ 2np (n is an integer), dark fringes appear; when D ¼ ð2n þ 1Þp,
light fringes appear. The fringe order reflects the relative displacement before and after
the deformation of the interference point. The larger the relative displacement caused
by the defect, the more obvious the fringe pattern is.

3 Experiments and Results

3.1 Experimental Setup
The main components of the experimental device include two thermally excited light
sources, a laser array, an optical imaging system composed of a high-accuracy CCD
camera and misalignment device, and an image computing and processing system.
Thermally excited light sources are used to heat the test specimens for deformation.
The maximum power of each thermal excitation source is 1 kW. In this experiment, the
helium-neon laser was used to measure the surface deformation of the object. The laser
wavelength was 632.8 nm and the laser power was 120 mW. During the detection
experiment, the CCD camera collects the speckle images before and after the deformed
specimens are subjected to thermal excitation, and is transmitted to a computer for
image processing to obtain interference fringe patterns and phase maps, from which the
defects can be identified and analyzed. The heat load time was 6 s in the experiment.
Composite insulators are mainly composed of silicon rubber sheds and fiberglass-
epoxy mandrels. The mandrels and sheaths respectively assume the external insulation
properties and mechanical strength of the insulators. The mandrels, sheaths, and
mandrels of the composite insulators are protected. All three areas of the interface may
produce defects during production and operation.

3.2 Detection of Voids in the Core

Silicone rubber and glass fiber epoxy resin bonded plate models were prepared and
drilled on the epoxy resin in advance to simulate void defects in the core. The defective
specimen model and related parameters are shown in Fig. 2. The standard test speci-
men is a square plate with a side length of 200 mm and a thickness of 10 mm. The rear
side is processed with four rows of 20 flat bottom holes with different depths and
diameters. The depth of flat bottom holes in the same row is the same, and the depth is
2 mm, 4 mm, 6 mm, and 8 mm in sequence. The diameter of each flat bottom hole in
the same row is 20 mm, 10 mm, 4 mm, 2 mm, and 1 mm in descending order. Two
epoxy resin plates with the above-mentioned void defects were bonded to silicon
rubber pieces with the thickness of 3 mm and 5 mm respectively, and digital
shearography detection was performed on the surface of the silicone rubber. The
resulting speckle interference pattern and phase map are shown in the Fig. 3. After the
test piece is deformed by heat during the inspection process, relative displacement will
866 L. Liu et al.

occur between the surface of the defect position and the surface of the non-defective
position, and the size is related to the material properties, defect parameters and thermal
excitation.

Fig. 2. The EP specimen model with flat bottom hole defects.

(a) Fringe pattern (b) Phase map

Fig. 3. Detection results of void defect samples with the silicone rubber thickness of 3 mm.
 Non-destructive Testing Method for Composite Insulators 867

(a) Fringe pattern (b) Phase map

Fig. 4. Detection results of void defect samples with the silicone rubber thickness of 5 mm.

(a) Schematic of cracks (b) Phase map

Fig. 5. Detection results of crack defect samples with the silicone rubber thickness of 2 mm.

It can be seen from Fig. 3 that the defect position shows pairs of stripe patterns, and
the size of the stripe pattern is approximately the size of the defect. A clear 10 mm
diameter defect can be detected under the 3 mm silicone rubber layer, and the detection
effect of the defect with the diameter of 4 mm is not obvious. A clear 10 mm diameter
defect can be detected under the 5 mm silicone rubber layers, as shown in Fig. 4. For
the same type of specimen, the sensitivity of digital shearography detection is closely
related to the size and depth of the defect.
868 L. Liu et al.

3.3 Detection of Cracks in the Core

Figure 5(a) shows a sample of the adhesive plate with three cracks in the EP section.
The size of the EP board is 100 mm  100 mm  10 mm. The SIR board has the
thicknesses of 2 mm. Each crack is 80 mm long and 2 mm deep. The cracks are
marked as CR1, CR2 and CR3, respectively, and their widths are 2 mm, 1 mm and
0.5 mm, respectively.
As shown in Fig. 5(b), all three cracks below 2 mm thick silicone rubber plate can
be clearly recognized in the generated phase map.

3.4 Detection of Interfacial Disbond

The glass fiber epoxy resin flat plate and silicone rubber sheets are bonded with a
coupling agent, and a certain area of the adhesive surface is left uncoated with a
coupling agent to simulate the interfacial disbond of the composite insulator between
the core and the housing. The length of the epoxy resin flat plate is 200 mm, width is
100 mm, thickness is 10 mm, and the thickness of the silicone rubber layer is 3 mm.
There are 3 disbond defects at the interface of the defective model. Digital shearog-
raphy detection was performed on the surface of the silicone rubber. The interference
fringe pattern, wrapped phase map and unwrapped phase map obtained by the detection
are shown in Fig. 6. It can be seen from Fig. 6 that the digital shearography detection
can clearly identify the disbond defects set under the 3 mm silicone rubber layers, and
the fringe pattern clearly shows the presence of defects.
The disbond defects at the interface between the core and the sheath is a major
safety hazard for composite insulators and an important cause of breakdown and
fracture accidents of composite insulators. The traditional nondestructive testing
techniques such as ultrasonics, radiation, etc., whose principle is to detect the inho-
mogeneity of the internal structure of the object, could detect the disbond defect of the
composite insulator mainly by identifying the obvious air gap at the interface, and they
are difficult to detect that the disbond interface is still stuck. In contrast, digital
shearography demonstrates a clear detection advantage for disbond type defects.
Since the digital shearography detects the surface displacement information caused
by the stress concentration of the defect, the structural and mechanical properties of the
detected object could be both revealed, and thus digital shearography can more directly
reflect the damage of the defect to the strength of the insulator structure. It is a feature
that other non-destructive testing technologies do not have. For the disbond defect of
composite insulator that the interface is still stuck, no obvious air gap has been formed
in the interface region. It is difficult to identify such defects by ultrasonic, X-ray, and
other detection techniques that rely on the heterogeneity of the defect in the object
being detected. Although insulator with this kind of disbond defect is not significantly
different from the normal product in structure, the interface adhesiveness has been
greatly reduced, and the mechanical properties of the product have changed. At this
time, digital shearography detection can be performed to produce an abnormal defor-
mation at the defect position, and such disbond defects can be detected.
 Non-destructive Testing Method for Composite Insulators 869

(a) Fringe pattern

(b) Wrapped phase map

(c) Unwrapped phase map

Fig. 6. Detection results of interfacial disbond with the silicone rubber thickness of 3 mm.

4 Conclusion
(1) The digital shearography can detect the typical defects of composite insulator,
such as voids, cracks and interfacial disbond. Compared with other non-
destructive testing techniques, it has the characteristics of high speed, non-contact,
and full field measurement. Hence, it has great application prospects.
870 L. Liu et al.

(2) Digital shearography detects the surface displacement caused by the stress con-
centration of the defect, and it also reveals the information of its mechanical
properties while characterizing the structural properties of the detected object.
Hence, it can more directly reflect the damage of the defect to the strength of the
insulator structure. The digital shearography has obvious advantages in the
detection of interfacial disbond defects without obvious changes in the structural
characteristics.
(3) The detection accuracy of digital shearography is closely related to the mechanical
properties of the material, the size and depth of defects, and the loading of the
excitation. The quantitative identification of composite insulator defects using
digital shearography remains to be further studied.

References
1. Zhang, F., Song, L., Li, R., et al.: Evaluation method of defect and fracture reason for
composite insulator. High Voltage Eng. 38(11), 3093–3100 (2012)
2. Xie, C., He, Z., Ling, Y., et al.: Using ultrasonic phased array to inspect the internal defects of
composite insulators. Proc. CSEE 32(1), 63–68 (2012)
3. Volat, C., Jabbari, M., Farzaneh, M., et al.: New method for in live-line detection of small
defects in composite insulator based on electro-optic E-field sensor. IEEE Trans. Dielectr.
Electr. Insul. 20(1), 194–201 (2013)
4. Deng, H., He, Z., Chen, L., et al.: Ultrasonic guided wave-based detection of composite
insulator debonding. IEEE Trans. Dielectr. Electr. Insul. 24(6), 3586–3593 (2017)
5. Cheng, L., Wang, L., Mei, H., et al.: Research of non-destructive methods to test defects
hidden within composite insulators based on THz time-domain spectroscopy technology.
IEEE Trans. Dielectr. Electr. Insul. 23(4), 2126–2133 (2016)
6. ASTM International: ASTM E2581-14 Standard practice for shearography of polymer matrix
composites and sandwich core materials in aerospace applications. ASTM International, West
Conshohocken (2014)
7. Angelis, G., Dati, E., Bernabei, M., et al: Development on aerospace composite structures
investigation using thermography and shearography in comparison to traditional NDT
methods. In: 2015 IEEE Metrology for Aerospace, Benevento, Italy (2015)
8. Huang, Y., Ho, H.: Shearography: an optical measurement technique and applications. Mater.
Sci. Eng., R 49(3), 61–87 (2005)
 Investigation on the Acid Removal
Performance of Oil Regeneration
Sorbent Materials

S. Y. Matharage , S. Liu, Q. Liu(&) , and Z. D. Wang

The University of Manchester, Manchester M13 9PL, UK

qiang.liu@manchester.ac.uk

Abstract. Transformers being one of the most expensive equipment in the

networks require proper utilisation, which includes regular maintenance and
application of life-extension techniques in feasible situations. Ageing process of
the mineral oil and cellulose paper insulation system creates by-products such as
moisture, acids and sludge, which further accelerate the degradation of the
transformer insulation. Therefore, the removal of these by-products through oil
regeneration could potentially prolong the lifetime of transformers. Typically,
oil regeneration process includes percolation of oil through an adsorbent system
followed by a filtration and a degasification process. Various sorbent types such
as fuller’s earth, alumina, molecular sieves, silica-kaolin-sand mixture are used
for the regeneration process. This paper investigates the acid removal perfor-
mance of a modified alumina based sorbent, against two conventional sorbents
i.e. fuller’s earth and alumina. It was found that the overall performance of
modified alumina was higher than the conventional sorbent materials. Further-
more, all the three sorbent materials showed higher absorption capacity for low
molecular weight acids than the high molecular weight acids.

Keywords: Transformer
Oil regeneration
Fuller’s earth
Alumina

1 Introduction

Mineral insulating oil refined from crude oil sources are still the predominant insulating
liquids used in power transformers. These oils consist of mainly paraffinic and naph-
thenic hydrocarbons. Due to various electrical, mechanical, thermal and chemical
stresses the liquid insulation degrades mainly through oxidation, producing various by-
products such as alcohols, aldehydes, ketones, acids and water. Ageing of paper occurs
through acid catalysed hydrolysis which produces water and low molecular weight
acids. These accumulated by–products could further accelerate ageing and degradation
of the oil and paper insulation system. By removing the ageing by-products, the rate of
degradation of the oil/paper insulation of processed transformers could be reduced and
thereby their lives extended appreciably [1].
Oil regeneration/reclamation has been a topic of interest, since early 20th century
when mineral oil was first used as an insulating liquid for transformers [2–4]. In the
past, lack of mineral oil resources in some countries and the much faster degradation of

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 871–877, 2020.
https://doi.org/10.1007/978-3-030-31676-1_81
872 S. Y. Matharage et al.

available oil leading to sludge formation in transformers were the main driving forces
for oil regeneration [4]. Both solvent extraction and percolation techniques were tried
to regenerate the oils by removing the polar compounds and sludge formed during the
ageing processes [2, 3, 5]. However, oil regeneration was not widely practiced due to
extra costs involved with the process and the potential worry of regenerated oil being
oxidised faster than new oils. Invention of the hindered phenolic inhibitor, 2,6-Di-tert-
butyl-p-cresol (DBPC) in mid-20th century helped to manufacture mineral oil with
higher oxidation stability than the uninhibited counterpart [6]. Furthermore, it was
found that DBPC would not only improve the oxidation stability of new oil but also the
regenerated oil which resulted in the concept of multiple life extensions for mineral
oils.
Modern days, oil regeneration is performed not only to improve oil quality but also
with the attempt to improve the quality of solid insulation and hence as life extension
technique for the whole transformer [7–9]. Both online and offline techniques are used
for oil regeneration. Typical oil regeneration units are either a combination or indi-
vidual units with an oil filtration unit, moisture removal unit and an adsorbent based
cleaning unit. Conventional sorbent materials used in regeneration systems include
fuller’s earth, alumina, molecular sieves and Silica-Kaolin-Sand mixture. This paper
compares the performance of a modified alumina based sorbent material against con-
ventional sorbent materials fuller’s earth and alumina in terms of their acid absorption
capacity.

2 Experimental Description

Performance of the modified alumina based sorbent with regard to acid removal from
oil was compared against two commonly used sorbent materials fuller’s earth and
alumina. Experiments were conducted with an inhibited mineral oil spiked with dif-
ferent types of acids and an in-service aged uninhibited mineral oil. Samples prepared
by mixing sorbent materials with oils were first stirred on a magnetic stirrer for different
durations allowing the sorbent to absorb acids from the oil. At the end of the stirring
period oil was separated from the sorbent by filtering through a 0.5 µm nylon mem-
brane. Acidity of the initial and filtered oil samples was measured by potentiometric
titration according to BS EN 62021-1 [10]. Performance of the sorbents was compared
in terms of their absorption capacity and the absorption speed.

2.1 Selection of Oil to Sorbent Ratio

A preliminary study was first conducted to understand the relative absorption capacity
of each sorbent material. Typically, the size of filters used in oil regeneration units is
identified based on their volume. The study was conducted by mixing 2 ml of sorbent
material with 25 ml of an in–service aged uninhibited mineral oil. It can be seen from
Fig. 1 that for the same oil-to-sorbent volume ratio, the modified alumina absorbs more
than four times the amount of acids absorbed by the fuller’s earth and alumina.
Therefore, it was decided to use four times more fuller’s earth and alumina than
modified alumina for the tests as this would result in significant acid extraction from oil
 Investigation on the Acid Removal Performance 873

by fuller’s earth and alumina and a much clearer difference between the initial and the
final acidity values.

1.2

1.0

Acidity in oil (mg KOH/g oil)

0.8

0.6

0.4

0.2 2ml Fullers earth and 25 ml oil

2ml Modified alumina and 25 ml oil
2 ml Alumina and 25 ml oil
0.0
0 10 20 30 40 50 60 70 80
Duration (Hours)

Fig. 1. Variation of acidity in oil mixed with different sorbent materials

When comparing the performance of different sorbent materials, volume mea-

surement of solids can be less repeatable due to the differences in the size and the shape
of the solid particles. Furthermore, it was found that a unit volume of the sorbent
materials studied in this work are in a similar weight range (0.51–0.57 g/ml). There-
fore, weight of the sorbents was used for the following experiments. Table 1 shows the
amount of oil and sorbent materials used for the experiments. Sorbent to oil ratio for
fuller’s earth and alumina was 1:5 while 1:20 for modified alumina.

Table 1. Amount of oil and sorbent materials

Amount of oil 20 g
Amount of sorbent material Fullers earth 4g
Alumina 4g
Modified alumina 1 g

2.2 Details of the Oil Samples

Three different types of oil samples including two spiked with acids and one in–service
aged oil sample were used for the experiments. Formic acid and stearic acid were used
to represent low molecular weight acids (LMA) and high molecular weight acids
(HMA) respectively. Table 2 shows the acidity of the three oil samples prior to the
tests.
874 S. Y. Matharage et al.

Table 2. Acidity in the oil samples

Sample type Acidity in oil
Inhibited mineral oil spiked with formic acid 0.65 mg KOH/g oil
Inhibited mineral oil spiked with stearic acid 0.46 mg KOH/g oil
In-service aged uninhibited oil 0.99 mg KOH/g oil

2.3 Parameters Used to Compare the Sorbent Materials

Performance of a sorbent material can be indicated through acidity measurement in oil.
However, it is not the best way to compare the performance of sorbent materials as the
acidity in oil depends on the oil to sorbent ratio and the initial acidity in oil. Hence, the
acidity measurements were converted into absorption capacity as shown in Eq. 1,

Woil
AC ¼ ½Acidini  Acidend   ð1Þ
Wsorb

Where, AC is the absorption capacity in mg KOH/g sorbent, Acidini and Acidend are
initial and final acidity in oil measured in mg KOH/g oil, and Woil and Wsorb are weight
of oil and sorbent measured in g.
Figure 2 shows the acidity in oil and the relevant absorption capacity calculated for
an experiment conducted with alumina and oil mixture. Two sets of samples were used
for the experiment with sorbent to oil ratios of 1:20 and 1:5. Even with the same
sorbent material, final acidity in oil was 0.07 mg KOH/g oil and 0.77 mg KOH/g oil
for 1:5 and 1:20 ratios, respectively. On the other hand, the absorption capacity cal-
culated for both instances were similar at about 4.6 mg KOH/g sorbent. This result
shows that the absorption capacity is a better indicator than acidity to compare the
performance of oil regeneration sorbent materials. Therefore, absorption capacity was
used to compare the sorbents used in this study.

1.0 5
Absorption capacity (mg KOH/g absorbent)

0.9

0.8 4
Acidity in oil (mg KOH/g oil)

0.7

0.6 Acidity 3
1 g sorbent to 20 g oil
0.5
4 g sorbent to 20 g oil
0.4 Absorption capacity 2
1 g sorbent to 20 g oil
0.3 4 g sorbent to 20 g oil
0.2 1

0.1

0.0 0
0 10 20 30 40 50 60 70 80
Duration (Hours)

Fig. 2. Variation of acidity in oil and the absorption capacity of alumina for different sorbent to
oil ratios.
 Investigation on the Acid Removal Performance 875

3 Results and Discussions

3.1 Performance of Sorbents in New Mineral Oil Spiked with Formic
and Stearic Acids
Figure 3 shows the formic acid absorption capacity against stirring duration for the
three sorbent materials. It can be seen that all three sorbent types have reached their full
removal capacity within the first three hours. The fast acid removal could have occurred
due to lack of contaminant in oil other than formic acid. However, it can be seen that
the formic acid absorption capacity for modified alumina is again about four times that
of conventional fuller’s earth and alumina sorbents.
Figure 4 shows the stearic acid absorption capacity of the three sorbent types.
Similar to formic acid, all three sorbent materials reached their capacity within three
hours. Modified alumina had the highest adsorption capacity followed by alumina, and
then fuller’s earth.

10
12
Absorption capacity (mg KOH/g absorbent)

Absorption capacity (mg KOH/g absorbent)

Fuller's earth
Modified alumina Fuller's earth
8 Modified alumina
10 Alumina
Alumina

8 6

6
4
4

2
2

0 0
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Duration (Hours) Duration (Hours)

Fig. 3. Formic acid absorption capacity of the Fig. 4. Stearic acid absorption capacity of the
sorbent materials three sorbent materials

Table 3 shows the absorption capacity values for the three sorbent materials. It can
be seen that all the sorbent materials have higher absorption capacity for formic acid
than the stearic acid. This is advantageous as LMA such as formic acid contributes for
paper ageing [11]. For both acid types modified alumina has the highest absorption
capacity followed by alumina and fullers earth.
876 S. Y. Matharage et al.

Table 3. Final absorption capacity for the three investigated sorbent materials
Acid type Sorbent type Absorption capacity
mg KOH/g sorbent)
Formic Fuller’s earth 3.1
Alumina 3.2
Modified alumina 12.6
Stearic Fuller’s earth 0.9
Alumina 2.3
Modified alumina 9.2

3.2 Performance of Sorbents in In-service Aged Mineral Oil

Figure 5 shows the results of the tests carried out with in-service aged transformer oil
samples. Similar to the tests with new oil, the highest and lowest absorption capacity
values for aged oil were obtained for modified alumina and fuller’s earth, respectively.
The final absorption capacity for fuller’s earth, alumina and modified alumina were 1.3,
4.6 and 16.8 mg KOH/g absorbent, respectively.
It can be seen that the absorption capacity values obtained for the in-service aged
oil are generally higher than for acid spiked new oils. This could be due to an increase
in the amount of dissociated acids compared to the new oil. In-service aged oil has a
higher moisture content than the degassed and dried new oil. This would produce more
H+ ions through acid dissociation than in the new oil. These small H+ ions are easily
adsorbed onto the pores available in the sorbent material resulting in an apparent
increased absorption capacity.

18 Fuller's earth
Absorption capacity (mg KOH/g absorbent)

Modified alumina
16 Alumnia

14

12

10

0
0 10 20 30 40 50 60 70 80
Duration (Hours)

Fig. 5. Absorption capacity of the three sorbent materials in in-service aged mineral oil

The order of absorption capacity values of the sorbent materials was similar to that
obtained for new oil where modified alumina has the highest capacity followed by
alumina and fuller’s earth. In addition to the absorption capacity, the rate of acid
removal was also different between the new oil spiked with acids and the in-service
 Investigation on the Acid Removal Performance 877

aged oil. For new oil it only took about three hours to reach the maximum capacity
while in aged oil it took more than 20 h to reach the maximum capacity. The com-
peting nature between acids and the ageing by-products such as alcohols, aldehydes,
ketones and water could have also resulted in absorption of some acids getting delayed.

4 Conclusions

Acid removal performance of a modified alumina based oil regeneration sorbent was
compared with two other conventional sorbent materials; fuller’s earth and alumina.
Investigations were conducted with new mineral oil spiked with acids and in-service
aged oil sample.
Absorption capacity is defined and used to compare the performance among different
sorbent materials. Absorption capacity of all three sorbent materials showed a depen-
dency on the type of acids with a higher absorption capacity for low molecular weight
acids than high molecular weight acids. Out of three sorbent materials, modified alumina
based sorbent had the highest absorption capacity followed by alumina and fuller’s earth.
Similar to the absorption capacity, modified alumina had a higher initial absorption
speed than fuller’s earth and alumina sorbents for the in–service aged oil. The acid
removal rate from the in-service aged oil was lower than the new oil for all the three
sorbent types. This could have occurred due to the competing nature between acids and
other ageing by-products such as alcohols, aldehydes, ketones and water in the oil.

References
1. Venkatasubramanian, R., Liu, Q., Wang, Z.D., Marshal, P.: Assessment of regenerated oil
through accelerated thermal ageing experiments. In: 2015 50th International Universities
Power Engineering Conference (UPEC), pp. 1–4 (2015)
2. Friese, R.: Process of regenerating oil used for electrical purposes. US Patent 1,103,499
(1914)
3. Cox, D.C.: Process of reclaiming oil. US Patent 1,752,238 (1930)
4. Ambrose, H.A.: In chemical progress on insulating oil. Digest Lit. Dielectr. 3, 12–16 (1938)
5. Housley, J.E.: Reconditioning of insulating oils by activated alumina. Electr. Eng. 58(4),
172–178 (1939)
6. Doble, F.C.: A new concept of insulating oil life characteristics. Trans. Am. Inst. Electr. Eng.
Part III: Power Apparatus Syst. 71(1), 532–538 (1952)
7. Koestinger, P., Bruaroy, T. M.: Drying of power transformers in the field, applying the LFH-
technology in combination with oil reclamation. In: CIGRE Paris Session, Paper No A2-106,
Paris, France (2006)
8. Rasor, R.T., Moleski, H.L., Lubbeck, H.U.: Moisture in transformers and online dryer
performance. IEEE PES T&D 2010, 1–5 (2010)
9. Sisic, E.: Transformer insulation regeneration and drying in online mode. Transformer Mag.
5(2), 116–122 (2018)
10. BS EN 62021-1: Insulating liquids - Determination of acidity - Part 1: Automatic
potentiometric titration (2003)
11. Azis, N., Liu, Q., Wang, Z.D.: Ageing assessment of transformer paper insulation through
post mortem analysis. IEEE Trans. Dielectr. Electr. Insul. 21(2), 845–853 (2014)
 Terahertz Time-Domain Spectroscopy
Characterization of Aged XLPE
Cable Insulation

Yuanyuan Zhang, Ze Lian, Jianying Li, and Shengtao Li(&)

State Key Laboratory of Electrical Insulation and Power Equipment,

Xi’an Jiaotong University, Xi’an 710049, China
sli@mail.xjtu.edu.cn

Abstract. Cross-linked polyethylene (XLPE) has been widely used in high

voltage power cables as insulation material due to its superior mechanical and
electrical performance. Thermal aging causes irreversible chemical and mor-
phological change that may strongly reduce the dielectric strength of XLPE,
producing a limitation to the effective service life of the power cables. There-
fore, it is worthwhile to monitor the aging condition of XLPE insulation so that
the safety of the power system can be ensured. Terahertz technology has
attracted much attention because many molecular structure information is
involved in terahertz frequency. In this paper, terahertz time-domain spec-
troscopy (THz-TDS) was employed to study the effect of thermal aging on
XLPE power cable insulation. XLPE sheets were aged at 130 °C for different
periods in an air forced oven. The pristine and aged samples were studied by
differential scanning calorimeter (DSC) and the crystallinity were determined.
The extent of oxidation of the insulations were assessed by Fourier transform
infrared (FTIR) spectroscopy. Our experimental results show that the index of
refraction and relative dielectric constant in the frequency range from 0.5–2 THz
increase with aging time. The aged samples show an increase in oxidation
products and a decrease in the crystallinity, which indicates the molecular chain
scission and oxidation. This result suggests that THz-TDS has potential to
obtain information about the cable insulation condition.

Keywords: XLPE cable insulation
Thermal aging
Condition assessment

Terahertz time-domain spectroscopy

1 Introduction

XLPE is an important material and extensively used as cable insulation due to its
superior mechanical and electrical performance. The aging and degradation of insu-
lation can cause damage and shorten the life of XLPE power cable [1]. Insulation
condition assessment plays a vital role in the power system.
Various methods have been proposed for XLPE power cable insulation condition
assessment. Tensile test is commonly performed to assess the condition of the insu-
lation, but it is a destructive method and consumes great amount of samples. This
motivates the use of other test methods that use only small amounts of sample, such as

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 878–886, 2020.
https://doi.org/10.1007/978-3-030-31676-1_82
 Terahertz Time-Domain Spectroscopy Characterization 879

measurement of crystallinity, gel fraction, oxidation induction time and oxidation

products [2]. The aging effects on physical and chemical properties were well presented
through these methods. For non-destructive methods, dielectric spectroscopy at audio
frequency has a proven potential for condition assessment of XLPE power cable
insulation [3].
In recent two decades, terahertz science and technology developed dramatically
because of the emergence of ultrafast lasers and semiconductors technologies [4].
Terahertz wave generally refers to the electromagnetic radiation with the frequency
from 0.1 to 10 THz. THz-TDS is a promising technology to acquire dielectric spec-
troscopy of polymer at terahertz frequency and it is also used for aging condition
assessment. Li et al. [5] applied THz-TDS to investigate the deterioration of different
transformer oils. They found that THz-TDS is a useful technique in characterizing the
condition of dielectric system. Yan et al. [6] reported that the dielectric permittivity of
fresh 10 kV XLPE power cable from 0.7 to 2.5 THz are larger than that of thermally
aged XLPE. The results of Komatsu et al. [7] showed that the real and imaginary part
of dielectric permittivity for XLPE nuclear power cable from 0.6 to 3.6 THz increased
as radiation dose increased. Since THz region is a relative new area, the relation
between the structure and dielectric response at terahertz frequency of aged XLPE still
needs further study.
In this paper, dielectric constant in terahertz frequency of XLPE before and after
thermal aging were investigated by THz-TDS. Furthermore, DSC and FTIR were
employed to characterize the underlying structure changes of XLPE.

2 Experimental

2.1 Sample and Thermal Aging

The samples in this work were insulation of a new 110 kV XLPE cable supplied by a
cable manufacturer. XLPE sheets were obtained by cutting the cable along axial
direction with a lathe. Thermal aging was conducted in an aging oven at 130 °C for
1200 h.

2.2 THz-TDS Measurement

The spectra of aged and pristine XLPE samples were measured using THz-TDS
spectrometer (DHC Co. Ltd, China). The test temperature was 23 °C and relative
humidity was kept below 3% by dry nitrogen to exclude vaporous water. The time-
domain signal with and without XLPE samples were recorded by computer. The
frequency-domain signal including amplitude and phase is obtained through fast
Fourier transform (FFT) to time-domain signal. The optical parameters of XLPE
samples are calculated through formulas (1) and (2) [8]
c
nðxÞ ¼ 1  uðxÞ ; ð1Þ
xd
880 Y. Zhang et al.

4nðxÞ c
jðxÞ ¼ lnf 2
g : ð2Þ
AðxÞ½nðxÞ þ 1 xd

where:
n(x) = Index of refraction
u(x) = Ratio between phase of sample signal and reference signal
j(x) = Extinction coefficient
A(x) = Difference between amplitude of sample signal and reference signal

The relation between relative dielectric constant and n and j is shown in formula (3)

e ¼ n 2  j2 ð3Þ

where: e = Relative dielectric constant

2.3 DSC Measurement

We analysed the crystallinity with differential scanning calorimeter (DSC822e,
METTLER, Switzerland) under nitrogen environment. The heat flow curves were
acquired by heating approximate 5 mg sample from 30–150 °C with a scanning rate of
10 °C/min. The crystallinity X was calculated through Eq. (4) [9]

DHm
X¼  100% ð4Þ
DH0

where:
Hm = Melting enthalpy of sample (J/g)
H0 = Melting enthalpy of completely crystallized XLPE, the value is 287.3 J/g

2.4 FTIR Measurement

FTIR is a useful tool to characterize the chemical composite and structure of polymers
because of the specific molecular vibrations. The oxidation products of XLPE before
and after aging were measured with a FTIR spectrometer (Nicolet iN10, America). The
carbonyl index is used to quantify the degree of oxidation, defined as the ratio of the
absorption peak area at 1720 cm−1 (carbonyl band) and that at 2010 cm−1, which is not
sensitive to oxidation [10].

3 Results and Discussion

3.1 Effect of Aging on the Crystallinity

Thermal aging may influence the distribution of molecular chains of XLPE, leading to
the variation in amorphous and crystallized region. Figure 1 is DSC spectra of different
 Terahertz Time-Domain Spectroscopy Characterization 881

XLPE samples, showing that a melting peak moves to lower temperature when aging
time becomes longer.

Fig. 1. DSC curves of different XLPE samples.

According to Fig. 1, the variation of crystallinity of XLPE samples with respect to

aging time is shown in Fig. 2. It is found that the crystallinity becomes smaller
gradually when aging time increases. The reduction of crystallinity indicates that
degradation occurs in the material caused by thermo-oxidation reaction and chain
scission. Due to long exposure to high temperature, the crystalline parts of XLPE
become amorphous which makes oxygen diffusion easier.

Fig. 2. The crystallinity of XLPE samples with aging time.

882 Y. Zhang et al.

3.2 Effect of Aging on Chemical Structure

Oxidative reaction is one mechanism of structure and property deterioration for XLPE.
Figure 3 shows the IR absorption in a wavenumber range from 500 to 3500 cm−1 for
unaged and aged XLPE samples. As can be seen, the absorption peak of carbonyl at
1720 cm−1 increases with aging time. The inset in Fig. 3 magnifies the absorption band
from 1700–1800 cm−1. According to the oxidation mechanism of XLPE, oxygen
would combine with radicals to produce carbonyl group such as ketone, ester and
carboxylic acid [11].

Fig. 3. The FTIR spectra of different XLPE samples.

In this regard, the degree of oxidation can be reflected by the carbonyl absorbance.
Figure 4 illustrates the carbonyl index of XLPE samples as a function of aging time,
defined as the ratio of the absorption peak area at 1720 cm−1and that at 2010 cm−1
[10]. The carbonyl index shows a great increment with aging time. The carbonyl index
of unaged XLPE is around 0.4, after 1200 h exposure it increases to 1.0, which means
that XLPE samples experienced thermo-oxidation. This result is consistent with the
drop of crystallinity.
 Terahertz Time-Domain Spectroscopy Characterization 883

Fig. 4. The carbonyl index of XLPE samples with aging time.

3.3 Effect of Aging on Dielectric Permittivity at Terahertz Frequency

Terahertz spectroscopy is a novel method to obtain the dielectric response of polymers.
Figures 5 and 6 are time-domain and frequency-domain spectra of reference and dif-
ferent XLPE samples at terahertz region. Here, the reference signal is nitrogen. As can
be seen from two figures, the amplitudes of XLPE samples are smaller than that of
reference signal. This phenomenon reveals that terahertz radiation is absorbed by
XLPE samples. Besides, there is a time delay between samples and reference signal
because XLPE has larger refractive index than nitrogen.

Fig. 5. Terahertz time-domain spectra of reference and different XLPE samples.

884 Y. Zhang et al.

Fig. 6. Terahertz frequency-domain spectra of reference and different XLPE samples.

The index of refraction of XLPE samples from 0.5 to 2.0 THz is shown in Fig. 7,
obtained through analyses of the amplitude and phase of THz-TDS measurement
results. As shown in Fig. 7, the index of refraction of XLPE samples increases when
the aging time increases. Furthermore, the relative dielectric constant of XLPE were
also calculated, showing same tendency during aging period as seen in Fig. 8.
The dielectric constant spectra of polymers often contain important information
about the molecular vibration. The most prominent vibration modes in THz frequency
range are skeletal vibrations, such as oscillations of the backbone, hydrogen bonds, and
orientation polarization [12].

Fig. 7. The index of refraction of XLPE samples from 0.5–2.0 THz.

 Terahertz Time-Domain Spectroscopy Characterization 885

Fig. 8. The relative dielectric constant of XLPE samples from 0.5–2.0 THz.

When polyethylene was oxidized, optically inactive skeletal vibrations in poly-

ethylene would become active [7]. The increment of index of refraction and dielectric
constant in terahertz region may be attributed to the the accumulation of oxidation
products when aging time increases.

4 Conclusions

In this study, the effects of thermal aging on XLPE were studied based on DSC, FTIR
and THz-TDS. The following conclusions were drawn: When the XLPE cable insu-
lation sheets were aged at 130 °C, the molecular chain was broken and oxidized. As a
result, the refractive index and relative dielectric constant at THz region increase with
aging time increasing. Besides, the decrease of crystallinity and the increase of oxi-
dation products prove the aging process of XLPE. It is promising to use THz-TDS to
assess the condition of XLPE cable insulation.

References
1. Li, W., Li, J., Yin, G., Li, S., Zhao, J., Ouyang, B., Ohki, Y.: Frequency dependence of
breakdown performance of XLPE with different artificial defects. IEEE TDEI 4(19), 1351–
1359 (2012)
2. Boukezzi, L., Boubakeur, A., Lallouani, M.: Effect of artificial thermal aging on the
crystallinity of XLPE insulation cables: X-ray study. In: Conference on Electrical Insulation
& Dielectric Phenomena, pp. 65–68. IEEE, Canada (2007)
3. Linde, E., Verardi, L., Fabiani, D., Gedde, U.W.: Dielectric spectroscopy as a condition
monitoring technique for cable insulation based on crosslinked polyethylene. Polym. Test.
44, 135–142 (2015)
886 Y. Zhang et al.

4. Ferguson, B., Zhang, X.: Materials for terahertz science and technology. Nat. Mater. 1(1),
26–33 (2002)
5. Li, M., Tong, M., Flecher, J., Dong, Z.: A novel approach to investigate the deterioration of
insulation of oils in power transformers with terahertz time-domain spectroscopy.
IEEE TDEI 2(24), 930–938 (2017)
6. Yan, Z., Shi, W., Hou, L., Xu, M., Yang, L., Dong, C., Li, S.: Investigation of aging effects
in cross-linked polyethylene insulated cable using terahertz waves. Mater. Res. Express 1(4),
015304 (2017)
7. Komatsu, M., Ohki, Y., Mizuno, M., Fukunaga, K.: Terahertz spectroscopic observation of
cross-linked polyethylene aged simultaneously by heat and gamma-rays. In: 10th
International Conference on the Properties and Applications of Dielectric Materials. IEEE,
India (2012)
8. Withayachumnankul, W., Naftaly, M.: Fundamentals of measurement in terahertz time-
domain spectroscopy. J. Infrared Millim. Terahertz 8(35), 610–637 (2014)
9. Liu, X., Yu, Q., Liu, M., Li, Y., Zhong, L., Fu, M., Hou, S.: DC electrical breakdown
dependence on the radial position of specimens within HVDc XLPE cable insulation.
IEEE TDEI 3(24), 1476–1484 (2017)
10. Xu, Y., Luo, P., Xu, M., Sun, T.: Investigation on insulation material morphological
structure of 110 and 220 kV XLPE retired cables for reusing. IEEE TDEI 4(21), 1687–1696
(2014)
11. Gulmine, J., Akcelrud, L.: FTIR characterization of aged XLPE. Polym. Test. 7(25), 932–
942 (2006)
12. Pearce, E.: Polymers: chemistry and physics of modern materials. J. Polym. Sci. Pol. Chem.
8(30), 1777 (1992)
 Evaluation and Performance Benchmarking
of Partial Discharge Measurement Sensors
Technologies Based on Non-conventional
PD Measurement System for Medium Voltage
GIS Applications

David Jebamony1(&), Manjunath Ramesh1(&), Bastian Wölke2(&),

and Pascal Thyssen2(&)
1
Nuventura GmbH, 12681 Berlin, Germany
{dj,mr}@nuventura.com
2
Westnetz GmbH, 44139 Dortmund, Germany
{bastian.woelke,pascal.thyssen}@westnetz.de

Abstract. Continuous monitoring solutions exist for medium voltage

(MV) switchgear which enable real-time monitoring of switchgear health and
provide operators with benefits such as lower maintenance costs and reduced
risk of failures. They are predominantly used in air insulated switchgear (AIS) or
externally in gas insulated switchgear (GIS). However, monitoring solutions
inside the critical GIS core vessel are not common due to environmental con-
cerns around SF6 emissions and technical difficulties. Nuventura has developed
a MV switchgear that uses dry air as an insulating medium which enables the
use of these solutions. An additional advantage is that sensor solutions that can
be integrated within the GIS core vessel at manufacturing. Partial Discharge
(PD) provides valuable information about the condition of the insulation of
high-voltage (HV) electrical equipment such as switchgear and is a key indicator
of the asset health. Recent advances in non-conventional measuring techniques
have created opportunities to monitor switchgear health in real-time and without
interruption. Commercially available methods, such as ultra-high frequency
(UHF) range measurement of electromagnetic radiation, are low cost and pas-
sive and can be integrated into switchgear at the time of manufacture. They are
expected to last for long periods of time without interruption. In the current
evaluation, commercially available UHF sensor solutions are benchmarked to
identify a technically suitable, economically viable early warning detection
system for use inside the MV GIS vessel. Other non-conventional sensing
techniques such as Transient Earth Voltage (TEV), High Frequency Current
Transformer (HFCT) and acoustic methods are also tested for comparison. The
conventional method, described in IEC 60270, is used as a reference.

Keywords: Non-conventional
Partial discharge
UHF method
Gas

insulated switchgear
Sensor

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 887–898, 2020.
https://doi.org/10.1007/978-3-030-31676-1_83
888 D. Jebamony et al.

1 Introduction

Distribution grid operators constantly look for ways to improve service reliability,
reduce maintenance costs and increase the lifetime of critical assets such as switchgear.
Through advancements in sensor technology, economically viable solutions already
exist that can measure and monitor key indicators of asset health such as partial
discharge. Increasing partial discharge is an indicator of failing insulation in switchgear
and is responsible for over 50% of failure cases [1] in GIS. Online monitoring and
condition-based assessment is expected to help operators realize benefits such as lower
maintenance costs, increased asset lifetime and reduced risk of failures. These solutions
are becoming more readily available and are already used in air insulated switchgear
(AIS) as well as in gas insulated switchgear (GIS) in places such as cable compartments
or externally on the switchgear enclosures. There are challenges in monitoring the
insulation within the GIS core compartment. This is mainly due to the use of SF6
within GIS and technical challenges in sensor installation.
Replacing SF6 by dry compressed air offers numerous advantages. One of the most
important advantages is accessibility to the GIS vessel for maintenance and service. No
SF6 evacuation process is involved and leakage of dry air will not lead to adverse
environment effects. In addition, a dry air solution is easy to maintain and expensive
gas handling at the manufacturing process or in field use is eliminated. Additionally,
installation and replacement or service of additional technologies such as sensors
becomes possible.

2 Non-conventional PD Detection Methods

2.1 Advantages
Detection of PD by electrical methods can be broadly categorized into the conventional
method based on the IEC 60270 standard and those that are implemented using non-
conventional methods based on the IEC 62478 standard.
The conventional method of partial discharge (PD) detection is very accurate and
measures electrical pulses caused by PD in the frequency range below 1 MHz as per
the IEC 60270 standard. This is suitable for controlled laboratory environments but is
susceptible to high electrical noise in live environments [2]. It is also intrusive,
requiring outages to the device under test for the measuring period, and comparatively
costly making it unsuitable for continuous online monitoring.
In order to avoid the issues associated with high electrical noise on site, alternative
non-conventional methods which operate in higher frequency ranges are required.
Commercially available solutions exist that are economically viable and non-intrusive
allowing for continuous monitoring in real life conditions without outages to the
equipment. These can be classified as internal (invasive) or external (non-invasive)
sensors, according to whether they are mounted inside or outside the equipment.
The UHF method is a non-conventional method that includes passive sensors which
can be installed inside the metallic chambers of GIS compartments such as the core
vessel. The absence of active electronics and moving components in the sensors means
 Evaluation and Performance Benchmarking 889

that they can be expected to operate for long periods of time without maintenance. As
they are classified as invasive, they can also be mounted inside the core GIS vessel
during manufacture. The advantages of this are the high sensitivity to discharges due to
inner electrical resonance, low inherent losses and high immunity to electrical noise
interference. Additionally, it is expected that any insulation defect detected from a
sensor inside the switchgear confines the detection to be inside this particular
switchgear (including cable terminations connected to it).

2.2 UHF Method in MV GIS

The UHF detection technique is based on electromagnetic sensors with a high sensi-
tivity within the 300 MHz–3000 MHz bandwidth. Within this range, partial discharge
phenomena with short pulse duration are known to exhibit high spectral energy and are
thus easier to detect and analyze [3]. The distribution of freed energy in an occurrence
of a PD event is a non-directed radiation pattern, caused by the sudden changes in
velocity of charge. These transient electromagnetic waves can be picked up with
suitable antennas.
Within a metalclad environment such as the gas-filled tank of a GIS, electromag-
netic radiation is reflected on the wall surfaces and bounced around within the
enclosure. Since the typical PD event is characterized as a pulse, it is beneficial to
measure at a high frequency, where the typical GIS tank dimensions are larger than the
wavelengths of the propagating signal. For the usual tanks being of dimension below
1 m  1 m  1 m, this means that frequencies above 300 MHz are preferable. This
coincides with the comparatively common corona type discharge creating high-gain
noise mostly within a lower frequency band [4].
Dissipation is comparatively slow within this environment, which usually contains
mostly other conductors and only few dielectric components to absorb radiated energy.
Natural dampening can be as low as 2 dB/m [5]. Additionally, the noise level at the
UHF bandwidth within a closed GIS compartment is low, since common environ-
mental noise is mostly blocked out by the tank acting as a Faraday cage. The lack of
electronics or transmission devices inside the enclosure means any naturally occurring
emission is closer to the low frequency of the power transmission.

3 Benchmarking of Solutions
3.1 Testing Criteria
The purpose of these tests is to provide a technical evaluation of the UHF detection and
measurement functionality. There are however other factors that are important in
determining the technical suitability of the systems that are not covered, such as
identifying if additional discharges are introduced by the sensors or lifecycle testing to
understand long term performance and degradation of the sensors over time. Also, the
testing does not consider other factors such as cost of the systems, ease of use and
deployment of the systems which are would be important from an operator perspective.
890 D. Jebamony et al.

The main criteria is to identify which of the systems are able to detect different
types of discharge and act as an early warning detection system. This would allow
operators to use existing tools and processes to further diagnose the exact location of
the fault within the switchgear. As the UHF methods are used within the GIS gas
vessel, it is assumed that the discharge can be localized to be within the GIS.
The sensitivity of the systems is also important as it is crucial to know if they would
detect defects early enough to provide time for an operator to respond within their
planned maintenance cycles. Identification of the specific type of discharge is useful
but not mandatory for an early warning detection system and thus identifying the
specific discharge or differentiation between multiple types of simultaneous discharge
is not considered compulsory.
Setup tests were successfully carried out to determine if the sensor signal could be
extracted from within the GIS vessel via a hermetic feedthrough without the loss of
pressure in the tank, as this is an important factor for use within a pressurised GIS.

3.2 Test Design and Methodology

The test setup involves a demonstration version of a GIS tank as the device under test.
This simplifies handling inside a laboratory and reduces the costs and the complexity of
the tests. A high voltage source feeds a capacitive divider formed by a coupling
capacitor and an impedance, which provides a sinusoidal signal of 50 Hz. The variable
voltage source allows variation of the applied voltage magnitude. A PD free cable
connection ensures voltage connection into the device under the test (DUT). Blocking
impedances are used in the circuit to avoid high currents in case a discharge occurs due
to intense PD activity. The lab setup is shown in Fig. 1 with sensors as shown in
Table 2.

Fig. 1. Overview of test setup showing sensor positioning

 Evaluation and Performance Benchmarking 891

The used GIS tank consists of a steel compartment, with a steel lid providing an
access to the compartment. When the lid is closed, the compartment can be pressurized
up to 0.5 bar pressure. Pressurized dry air is used as the insulation medium in the GIS
tank. The lid consists of a holding arrangement for the antennas as well as hermetically
sealed feedthroughs for the SMA interface for the signal reception. The testing of all
sensors are run in parallel and data is logged into the respective systems for each of the
tests (Fig. 2).

Fig. 2. Picture of test setup showing all elements

A PD activity can be established in a controlled manner by applying high voltage

across two points and varying the distance between these points. As the points get
closer, a discharge is induced due to the increased electric field strength. The GIS core
vessel is evacuated to remove any moisture so that testing could be carried out with dry
air at atmospheric pressure (1.0 bar abs).

3.3 Test Methods

Different test configurations were used to induce different types of discharge as shown
in Table 1 and Fig. 3.

Table 1. Overview of test configurations

Test method Test configuration Expected primary discharge type
#1 Wire to plate Corona discharge
#2 Rod to acrylic sheet Surface discharge
#3 Metallic particles Particle discharge
892 D. Jebamony et al.

In test method #1, discharge was created by passing a high voltage between a 2 mm
wire with a sharpened tip and a metallic plate to simulate a defect. This is expected to
create corona discharges due to the ionisation of the air at the tip of the wire. Irregu-
larities on the conductors or damaged insulators are examples where this situation
could occur within a GIS tank.
Test method #2 uses a 3 mm thick transparent acrylic plexiglass sheet fixed firmly
between a metallic rod at high voltage and a metal disc at earth potential. This is
expected to create surface discharges on the acrylic sheet similar to what could occur
on dirty, contaminated or aged insulator surfaces.

Fig. 3. Test configurations for methods 1, 2 and 3 respectively

Test method #3 uses small metallic particles inside the high voltage field to sim-
ulate the defect. This is expected to create particle discharges that can be created similar
to what can occur from wear and tear of metallic parts, such as the disconnector or bolts
within the GIS vessel, rubbing against each other during normal operation.

3.4 Test Systems

The Omicron MPD600, which uses the conventional method described in IEC 60270,
is used as a reference system. Three non-conventional UHF methods based on IEC
62478 are assessed. Each has a passive sensor located inside the GIS gas vessel which
is connected via a hermetic feedthrough in the access to its respective monitoring unit
or collector where noise cancellation, signal processing methods and other algorithms
are used to detect the presence of discharges.
System-A uses a low profile (43 mm height) passive sensor to sense the amount of
charge (Quhf), which is a non-physical quantity and uses a non-linear scale. It also
autonomously identifies surface discharge (including corona) and partial (internal)
discharges as well as the noise floor.
 Evaluation and Performance Benchmarking 893

System-B uses a cylindrical pipe shaped sensor which couples the variation of
transient electromagnetic field caused by discharges. The data is measured in mV and
pulses per second and a phase resolved partial discharge (PRPD) pattern is also pro-
vided. System-B also provides a Transient Earth Voltage (TEV) sensor as well as an
UHF Transient Magnetic (UHF TM) sensor. The TEV sensor is installed externally on
the GIS vessel while the UHF TM sensor is coupled to the HV cable.
System-C does not utilise the UHF method but is used for comparison to other non-
conventional methods. It has TEV, Ultrasonic (Contact Method) and HFCT sensors
and provides measurement using a decibel scale.

Table 2. Overview of sensors tested

Sensor ID Technique Location Frequency range
Reference IEC 60270 External cable sensor <1 MHz
A1 UHF Internal to GIS vessel 300 MHz to 1500 MHz
B1 UHF Internal to GIS vessel 150 MHz to 1200 MHz
B2 UHF TM External cable sensor 30 MHz to 900 MHz
B3 TEV External on GIS vessel 200 MHz to 800 MHz
C1 TEV External on GIS vessel 10 MHz to 500 MHz
C2 HFCT External cable sensor 1.5 MHz to 20 MHz
C3 Ultrasonic Contact External on GIS vessel 30 kHz to 88 kHz

As per Table 2, the UHF sensors are located inside the GIS vessel. The UHF TM
and HFCT sensors are located externally on the HV cable. The TEV and Ultrasonic
Contact sensor is located externally on the GIS vessel enclosure.

4 Results

4.1 Method #1 – Corona Discharge

The test was started at 0.5 kV at which point some background noise was detected.
Inception was reached at 11 kV and further measurements were taken at 15 kV, 17 kV
and 18 kV. At each of these steps the voltage was maintained for five minutes. Finally,
the voltage was reduced to 0.5 kV where only some background noise was detected.
On the reference system, a pattern indicating corona discharges was observed. As
the voltage level increased, a higher the pulse activity in amplitude and intensity
(pulses per second) is observed (Fig. 4).
894 D. Jebamony et al.

Fig. 4. Omicron reference system measurements at (a) 11 kV, (b) 15 kV, (c) 17 kV and
(d) 18 kV

UHF Sensor A1 (System-A) detects low amounts of partial discharge at 0.5 kV and
11 kV. The surface discharge (SD) increased in step with the voltage applied from
15 kV to 18 kV. The “Total SD” includes surface discharge and corona discharges and
there is no differentiation made between the two in this system. The time intervals and
the corresponding applied voltages are shown in Fig. 5 below.

Fig. 5. Discharge measured by System-A

In System-B, UHF Sensor B1 and TEV Sensor B3 do not detect any discharges.
The external UHF cable Sensor B2 detects a steady increase in discharges as the
voltage is increased from 15 kV. At 17 kV and 18 kV, a noticeable increase in the
pulses per second is also detected. The PRPD pattern (Fig. 6) is consistent with what is
shown in the reference system.
 Evaluation and Performance Benchmarking 895

Fig. 6. Discharge measured by sensor B2

In System-C, the TEV sensor does not detect any discharge, however the Ultrasonic
Contact and HFCT sensors detect an increase in discharge from 15 kV and increase in
step with the voltage rises (Fig. 7).

Fig. 7. Discharge measured by System-C

4.2 Method #2 – Surface Discharge

The test was started at 0.5 kV at which point some background noise was detected.
Inception was reached at 3.1 kV and further measurements were taken at 3.5 kV
(*1000 pC), 4 kV, 4.5 kV, 5 kV and 7.5 kV. Finally, the voltage was reduced to
0.5 kV where only some background noise was detected.
On the reference system, a pattern indicating surface discharges was observed with
formations in both half cycles as well as partial discharges. As the voltage level
increased, a higher pulse activity in amplitude and intensity (pulses per second) is
observed for the discharges (Fig. 8).
896 D. Jebamony et al.

Fig. 8. Reference system measurements at (a) 3.1 kV, (b) 3.5 kV, (c) 5 kV and (d) 7.5 kV

UHF Sensor A1 (System-A) detects PD at inception voltage (3.1 kV) and

increasing amounts of both SD and PD are seen from 3.5 kV onwards. At 7.5 kV, the
amount of SD increases significantly. Both PD and SD detected amounts are not seen
after the voltage is decreased to 0.5 kV at the end of the testing.
In System-B, the UHF Sensors (B1 & B2) detect discharges at 3.1 kV (inception)
and the TEV sensor (B3) detects discharge at 3.5 kV. All three sensors show a clear
increase in discharge levels as the voltage applied is increased.
In System-C, the TEV and HFCT sensors detect discharge at 3.5 kV and the
Ultrasonic Contact sensor detects strong discharge at 7.5 kV.

4.3 Method #3 – Particle Discharge

Inception was reached at 3 kV with 50 pC of discharge. The voltage applied is
increased at regular intervals to 7 kV (*50 pC), 11 kV (*70 pC), 15 kV (*300 pC)
and 17 kV (*550 pC) where the voltage is held constant for a period of time before
decreasing it to 0.5 kV. On the reference system, a pattern was observed with cloud
formations in the negative half cycle and other discharges developing at a higher level.
UHF Sensor A1 (System-A) detects weak amounts of SD at inception voltage
(3 kV) and very high amounts of SD from 7 kV onwards however this reduces slowly
over the time period even as the voltage applied increases up to 17 kV. The amount of
SD detected decreases steadily until a stable level is reached at 17 kV. This discharge
was mainly noted in the 600 MHz frequency band.
In System-B, the internal UHF sensor (B1) detects weak discharge from 17 kV and
the external UHF cable sensor (B2) detects increased discharge at 15 kV and 17 kV.
In System-C, the TEV and HFCT sensors do not detect discharge and the Ultra-
sonic Contact sensor detects low amounts of discharge at 15 kV.
 Evaluation and Performance Benchmarking 897

5 Discussion

Table 3 summarizes the results of the test methods. It must be noted that for method 2,
where the table shows strong detection, it was only tested with high amounts of
discharge and the actual sensitivity of the sensors could not be discerned with accuracy.

Table 3. Summary of testing

UHF sensor A1 was able to detect all the different methods of discharge tested with
an accuracy of less than 300 pC.
UHF sensor B2 was able to detect all methods tested with varying sensitivity and
are also able to act as an early warning detector. Other offline methods (e.g. portable or
other handheld devices) and processes could be used additionally to provide detailed
diagnoses such as identification of the type of discharge and localization of the
defective part(s).
UHF sensor B1 was not able to detect discharges in method 1. However, it was able
to provide a PRPD pattern which could be useful in helping operators further identify
the type of occurring discharge(s).
Other non-invasive methods can be used additionally by operators to supplement
the internal UHF sensors and the combined results can be used to make further
diagnoses remotely.

6 Conclusion and Outlook

In summary, the results indicate that UHF sensors within a GIS vessel are able to detect
different types of discharge and in some cases clearly detect small amounts of dis-
charge. This can be supplemented with other types of non-conventional measurement
methods that are non-invasive and can be installed by the operator at a later stage.
Existing processes and methods, such as portable diagnosis devices, will still be
important in performing detailed diagnoses.
Future work is required to determine the quality and long-term suitability of the
sensors within GIS vessels. Lifecycle testing will be important to understand how the
sensors perform over long periods of time and if there is degradation of the sensors. It
898 D. Jebamony et al.

needs to be tested if there will be any decomposition of the sensor housing that could
impact the system, for example by causing surface deterioration of the insulators,
leading to surface discharges.

Acknowledgments. The project has been the joint effort of Nuventura and Westnetz
GmbH/innogy SE. Nuventura acknowledges and deeply appreciates the effort and interest from
Westnetz/innogy in driving the innovation into the forefront in SF6-free GIS solutions with
integrated monitoring.

References
1. Zeng, F., Tang, J., Zhang, X., Zhou, S., Pan, C.: Typical Internal Defects of Gas-Insulated
Switchgear and Partial Discharge Characteristics. IntechOpen (2018)
2. Álvarez Gómez, F., Albarracín-Sánchez, R., Garnacho Vecino, F., Granizo Arrabé, R.:
Diagnosis of insulation condition of MV switchgears by application of different partial
discharge measuring methods and sensors. Sensors 18, 720 (2018)
3. Judd, M.D., Yang, L., Hunter, I.B.B.: Partial discharge monitoring of power transformers
using UHF sensors. Part I: sensors and signal interpretation. IEEE Electr. Insul. Mag. 21(2),
5–14 (2005)
4. Bartnikas, R., McMahon, E.J.: Engineering Dielectrics. Volume L—Corona Measurement
and Interpretation. ASTM Publication No. STP 669 (1979)
5. Tenbohlen, S., Denissov, D., Hoek, S.M., Markalous, S.M.: Partial discharge measurement in
the ultra high frequency (UHF) range. IEEE Trans. Dielectr. Electr. Insul. 15(6), 1544–1552
(2008)
 Spectral Power Analysis of Partial Discharges
Waveforms During Electrical Tree Growth
Under Different Excitation Frequencies

Roger Schurch1(&), Osvaldo Munoz1, and Jorge Ardila-Rey2

1
Universidad Técnica Federico Santa María, Av. España 1680,
Valparaíso, Chile
roger.schurch@usm.cl
2
Universidad Técnica Federico Santa María, Av. Vicuña Mackenna 3939,
Santiago, Chile

Abstract. Electrical trees are the main mechanism of failure in solid polymeric
insulation. Their growth is associated to partial discharge (PD) activity. Mea-
suring and analyzing PDs during tree growth is crucial for the insulation con-
dition assessment and thus, the risk of failure of power equipment. Here, PD
pulse waveforms are analyzed using power ratios maps. The aim of the study is
to evaluate the ability of power ratio maps to determine the state of progression
of electrical trees. Electrical trees were grown in epoxy resin samples at different
frequencies, while PDs were measured using very-high frequency (VHF) and
ultra-high frequency (UHF) techniques. The progression of the tree was
observed using an optical camera and it was correlated with PD data. The results
showed that power ratio maps obtained from VHF and UHF signals yield similar
patterns. Especially in VHF, it was observed that the centroids of the clusters
associated to each stage of tree-growth moved with tree-progression. PDs at the
beginning of tree-growth were located at the top-right of power maps, while in
more advanced stages, they were located more to the bottom-center of the maps.
This means that the PD pulse waveform had higher frequency content in the
beginning than in the end of tree growth. In base of this initial results, it is
concluded that power ratio maps are sensitive to the state of growth of electrical
trees, and thus, can be used for assessing the risk of failure of insulation sub-
jected to electrical treeing degradation mechanism.

Keywords: Partial discharges
Electrical trees
Power ratio maps

1 Introduction

Electrical treeing is a mechanism of degradation in solid polymeric insulation that leads

to the failure of power equipment. It is known that electrical trees grow by the erosion
effect of partial discharges (PDs) [1], which are localized electric breakdown of micro-
volumes in dielectrics. Thus, measuring and analyzing PD activity is a method of
assessing the insulation condition of electrical power equipment. However, an
important challenge in PD analysis of electrical trees is that they are interconnected
networks which as they grow, the PD pattern evolves having a dynamic and complex

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 899–907, 2020.
https://doi.org/10.1007/978-3-030-31676-1_84
900 R. Schurch et al.

behavior [2]. In addition to this complexity, the issue of interpreting PDs under dif-
ferent frequencies of the applied voltage is a relevant aspect in insulation condition
assessment. On the one hand, high frequencies (higher than industrial 50 or 60 Hz) are
present nowadays in the form of harmonic components in power networks, mainly due
to the use of power converters and non-linear loads [3]. On the other hand, very low
frequency (VLF, lower than 1 Hz) is currently used for diagnostic testing of power
cables [4], and it has been suggested that new methods of PD analysis are required for
VLF testing [5].
PDs from electrical trees have been analyzed using various techniques: ‘phase-
resolved’ PD (PRPD) analysis, ‘pulse sequence’ analysis (PSA) and the analysis of the
PD pulse waveform [6–9], amongst others. A number of studies have analyzed the
growth of electrical trees at frequencies other than industrial frequency [10–13],
however, most of them were not focused on PD analysis. More connected to this
present work, reference [8] has studied electrical treeing under harmonic voltages
exploiting ultra-high frequency (UHF) techniques for measuring PD activity. Related to
treeing analysis, they characterized the UHF signal using power spectral analysis to
classify the shape of the electrical tree formed (branch, bush or fibrillar tree-type).
Conventionally, spectral power analysis, and specifically power ratios maps have been
used to classify PD sources (i.e. types of PDs: internal, superficial or corona) and to
separate them from noise [14, 15]. The novelty of this present paper is that we used
power ratios maps to analyses the progression of electrical trees under different fre-
quencies. Thus, the purpose of this paper is to evaluate the ability of power ratio maps
to determine the state of progression of electrical trees. Ultimately, this research seeks
to contribute in condition monitoring of power equipment evaluating the risk of failure
in insulation subjected to electrical trees.

2 Methodology

2.1 Experimental

Experimental Setup. Samples were made of epoxy resin (Mepox – 1685/L, a DGBA
epoxy system) and using the conventional needle-to-plane geometry with a gap of
*2 mm between the needle tip and the bottom-plane of the sample. Electrical trees
were grown and monitored in a test facility specially assembled for treeing experi-
ments. The electrical test circuit was the balanced circuit arrangement according to the
IEC 60270 standard [16], as shown in Fig. 1. The power source was a high voltage
amplifier Trek model 20/20C-HS, its input signal was provided by a signal generator
(Rigol DG-1022, ‘SG’ in Fig. 1). With this setup, sinusoidal voltages up to 14 kV
(*20 kV peak) and frequencies from 0.1 Hz to 450 Hz were generated for the tree
growth experiments. The samples were fed through a limiting resistance (Rx) to protect
the instruments in the event of breakdown. The sample Ca and the dummy sample (PD
free) Ck were placed into a transparent oil container to prevent unwanted surface
discharges. The signals from the treeing and dummy samples were subtracted in the
subtracting circuit SC, which output fed a commercial PD system that allowed
 Spectral Power Analysis of Partial Discharges Waveforms 901

continuous capturing of PD. Simultaneously, PD were captured in both very-high

frequency (VHF) and UHF modes for PD waveform analysis. This is detailed in
Sect. 2.1.3. The electrical tree growth was also monitored by an optical camera that
was taking pictures during the growth to correlate the PD measurements with the
progression of the tree.

Fig. 1. Circuit for electrical tree growth and monitoring.

Electrical Tree Initiation. To initiate an electrical tree in the samples, an incremental

scheme of three minutes steps of voltage and frequency application was devised. The
starting step was 10 kV 50 Hz, then the frequency or the voltage was increased in
alternating mode, until the voltage limit was reached and only the frequency could be
increased: 10 kV 150 Hz, 12 kV 150 Hz, 12 kV 300 Hz, 14 kV 300 Hz, 14 kV
500 Hz, 14 kV 600 Hz, 14 kV 700 Hz and 14 kV 800 Hz. At any point of the initi-
ation procedure, as soon as an electrical tree was optically detected, the voltage was
turned off. If no tree was initiated, the sample was discarded as not suitable for treeing
experiments. Following this procedure, the tree initiation process was experimentally
separated with the tree growth process. The initiation process was not analyzed, since is
out of the scope of this study.
Electrical Tree Growth and PD Measurements. A number of experiments of tree
growth at different voltages and frequencies were carried out. Eight samples were
selected to be analyzed here; details of them are presented in Table 1. One sample per
frequency of analysis was selected; at the ‘low frequencies’, the voltage was set to
either 14 or 16 kV, otherwise the experiment was too long or even imperceptible
growth was observed, especially in the case of 0.1 Hz.
Tree growth was monitored with an optical camera and three PD recording modes
simultaneously: conventional (wide band) of the electrical pulses, VHF and UHF
modes. The conventional measurement is not analyzed here (i.e. PRPD plots), it was
used to monitor the tree growth and PD behavior, and to correlate with the tree growth
time. For the growth at 0.1 Hz, there is no recording in VHF mode. VHF and UHF
measurements were integrated using a PXI high-speed acquisition system. The VHF
sensor was a commercial High Frequency Current Transformer (HFCT), while the
UHF sensor was an antenna (monopole of 5 cm). The sampling rate was set to 4 GS/s.
As is shown in Fig. 1. The HFCT was placed in the earth-side-wire of the sample and
the antenna 26 cm from the sample, at the same height. The distance was slightly
adjusted at the beginning of each experiment so the amplitude of the measured
902 R. Schurch et al.

Table 1. Samples and characteristics.

Sample Time to BD # PDs - Stage duration [h:min:s]
Stage 1 Stage 2 Stage 3
0.1 Hz-16 kV 24 h 22 min 9500 - 8:3:0 9500 - 8:5:0 9400 - 8:2:0
1 Hz-14 kV 2 h 39 min 541 - 0:4:0 2630 - 0:54:0 45105 - 1:15:0
10 Hz-14 kV 1 h 18 min 22806 - 0:26:0 22806 - 0:26:0 22806 - 0:26:10
50 Hz-12 kV 23 min 15 s 4584 - 0:6:10 4584 - 0:6:10 4332 - 0:6:10
150 Hz-12 kV 21 min 2 s 3210 - 0:2:32 3210 - 0:2:32 3210 - 0:2:32
250 Hz-12 kV 8 min 47 s 2537 - 0:2:30 2537 - 0:2:35 2537 - 0:2:35
350 Hz-12 kV 8 min 30 s 2009 - 0:2:20 2009 - 0:2:20 2009 - 0:2:20
450 Hz-12 kV 8 min 30 s 490 - 0:1:30 490 - 0:1:35 66 - 0:0:30

discharges will not exceed the limit of the channel. For the VHF measurement, a filter
of 60 MHz and an attenuator of 20 dB were used to protect the acquisition system.
The PXI acquisition system was composed by a NI PXIe-1071 chassis, a PXIe-5185
card and a NI PXIe-8135 controller. The system has two channels, sampling frequency
of 12.5 GS/s, 3 GHz of bandwidth and 8 bit of vertical resolution, the input limit is 1
Vpp for each channel (50 Ω).

2.2 PD Analysis – Power Ratio Maps

Power ratios maps have been used to classify PD sources and to separate them from
noise [14, 15]. Two parameters to characterize PD pulse waveform are calculated: the
power ratio for low frequencies (PRL) and the power ratio for high frequencies (PRH);
and the point (PRLi, PRHi) of the PD number “i” is plotted in a 2D graph called ‘PR
map’. The power ratios are calculated as follows [15]:
Pf2L 2
f j sð f Þj
PRL ¼ P1L
ft 2

100 ð1Þ
0 j sð f Þj

Pf2H 2
f jsð f Þj
PRH ¼ P1H
ft 2

100 ð2Þ
0 jsð f Þj

where s(f) is the spectral magnitude distribution of the pulse waveform and f1L, f2L, f1H,
f2H and ft are the limits of the frequency intervals (or frequency ranges or bands) and
are configurable by the user.
The PRL and PRH were calculated for each PD measured by both VHF and UHF
modes, thus the behavior of the PR maps could be compared for both modes. The
frequency bands for PRL and PRH were selected according to where the FFT of the
signal had more variability. For doing this, the variance was calculated for the entire
frequency range of 0 to 2 GHz, and the zones of larger variance were selected. The
results, which were used for all cases, are shown in Table 2:
 Spectral Power Analysis of Partial Discharges Waveforms 903

Table 2. Frequency limits of the intervals used for the calculation of power ratios.
Mode f1L f2L f1H f2H ft
VHF [MHz] 2 8 32 38 50
UHF [MHz] 20 50 500 1000 1000

To evaluate how sensitive are the position of the PDs in the PR maps with respect
to the progression of the electrical tree, the entire tree growth (i.e. all the PDs of each
sample) was divided in three stages. For each stage, the centroid of the cluster of
discharges in the PR map was determined, in order to obtain a clearer view of their
development/movement with the tree growth. This development was correlated to the
progression of the tree, which was determined through the images of the tree obtained
from the optical camera; the results are presented in the next section. In addition, the
slope and the dispersion of the data (PDs) of the clusters were calculated. The slope of
the clusters was calculated as the slope of the linear regression of the best fit of the
discharges of the cluster (or stage). The dispersion of the discharges of the clusters was
calculated as the standard deviation of the discharges for both PRL and PRH axes.

3 Results and Analysis

The time-to-breakdown of the samples is shown in Table 1. As expected [10], trees

grow faster as the frequency is increased, and this relation is not linear. However, note
that time-to-breakdown of samples aged at 250, 350 and 450 Hz is almost the same.
The structure and shape of electrical trees grown at different frequencies, prior to
breakdown is shown in Fig. 2. As it is known, trees grow in a more filamentary and
branchy way at lower frequencies, which is observed here.

Fig. 2. Electrical trees prior to breakdown.

The following analysis presented in this section is composed of: PR maps, the maps
of the centroids of the clusters or stages, and the slope and dispersion of the clusters of
the stages. Due to the amount of graphical information, only one case is fully-presented
here as an example; the case selected is 150 Hz. Figure 3 shows the PR maps from PDs
904 R. Schurch et al.

from the tree grown at 150 Hz, the top-row of graphs is from UHF and the bottom-row
from VHF measurements. Stage 1 (blue) is on the left, Stage 2 (red) in the center and
Stage 3 (black) on the right. It can be seen how as the tree grew, the clusters of the PR
maps moved. This is especially noticeable in UHF measurement, where in Stage 1 the
cluster is mainly located in top-left position of the map and ending in Stage 3 located
bottom-centered position of the map. Note that these positions relate with the frequency
intervals selected for PRL and PRH (Table 2). This indicates that in this case, PR maps
are sensitive to the progression of the tree. To analyses the clusters in more detail,
Fig. 4 presents the centroid of the clusters (left), their slope (center) and the dispersion
in both PRL and PRH axes (right), for both UHF (top row) and VHF (bottom row)
measurements. In this method of presentation, it is clearer the observation of the
evolution of the centroids of the clusters of each stage.

Fig. 3. PR maps of PDs from electrical tree growth at 150 Hz. Top row: UHF, bottom row:
VHF.

In general, similar results were obtained for VHF and UHF measurements. The
position of the centroids of the clusters, either from VHF or UHF, is shown in Fig. 5.
The criterion of showing centroids from VHF or UHF was selecting the map that
yielded a clearer evolution or progression of the centroids with electrical tree devel-
opment. It can be observed that in most cases there was a movement of the centroids
from the top-left of the map towards the bottom-right. This indicates that in the
beginning of tree-growth, the PD pulse waveform has higher frequency content than in
a more advance stage of growth.
The analysis of the clusters of each stage is presented in Fig. 6. The slope (left),
dispersion in PRL axis (center) and dispersion in PRH axis (right) is shown for both
UHF (top row) and VHF (bottom row) measurements. The idea behind of presenting the
results in this way is that the variation/evolution of the parameters along the stages of
growth can be easily evaluated for each excitation frequency. Especially in UHF
measurements, in most cases, for each frequency, the slope of the clusters did not greatly
change along the stages of growth. In contrast, the dispersion of the discharges in each
cluster of the stages showed greater changes for each frequency of voltage applied.
 Spectral Power Analysis of Partial Discharges Waveforms 905

Fig. 4. Analysis of clusters of PR maps - tree growth at 150 Hz.

Fig. 5. PR maps of the centroids of clusters of each stage.

Fig. 6. Analysis of clusters: slope (left), dispersion in PRL axis (center) and dispersion in PRH
axis (right). Top-row graphs for UHF and bottom-row graphs for VHF measurements.
906 R. Schurch et al.

4 Discussion and Conclusions

This paper has exploited power ratios maps to analyze the progression of electrical
trees under different frequencies. Power ratio maps are used to characterize the pulse
waveform of partial discharges, and here, they were evaluated in their ability of
determining the state of progression of electrical trees.
In general, PR maps obtained from VHF and UHF signals yielded similar pattern,
this points out that any of them can be used for the analysis and can draw similar
results. PR maps were analyzed through the slope and data dispersion of the clusters of
each stage of growth. It can be observed that in most cases, there was a movement of
the centroids from the top-left of the map towards the bottom-right. Note that these
positions relate with the frequency intervals selected for PRL and PRH (Table 2). This
movement shows that in the beginning of tree-growth, the PD pulse waveform has
higher frequency content than in a more advance stage of growth. This result indicates
that PR maps are sensitive to the growth of electrical trees, and therefore, they can be
used to infer the state of progression of trees. This is an important result, because it
means that PR maps could be used for condition assessment of insulation of power
equipment subjected to electrical trees. More research is needed to generalize this
initial, but promising results.

Acknowledgments. RS acknowledges funding support from Chilean Research Council

(CONICYT) under FONDECYT project Nº 11181177. JA acknowledges support from FON-
DECYT project Nº 11160115. The authors also thank Universidad Técnica Federico Santa María
for the grant PI_L_18_19.

References
1. Dissado, L.A., Fothergill, J.C.: Electrical Degradation and Breakdown in Polymers. Peter
Peregrinus Ltd., London (1992)
2. Schurch, R., Donoso, P., Aguirre, P., Cardenas, O., Zuniga, M., Rowland, S.M.: Electrical
tree growth and partial discharges analyzed by fractal and correlation dimensions. In: IEEE
Conference on Electrical Insulation and Dielectric Phenomenon (CEIDP), pp. 785–788
(2017)
3. Mohan, N., Undeland, T.M.: Power Electronics: Converters, Applications, and Design.
Wiley, Hoboken (2007)
4. IEEE 400.2-2013: IEEE Guide for Field Testing of Shielded Power Cable Systems Using
Very Low Frequency (VLF) (less than 1 Hz) (2013)
5. Dao, T., Phung, B.T., Blackburn, T., Nguyen, H.V.P.: A comparative study of partial
discharges under power and very low frequency voltage excitation. In: IEEE Conference on
Electrical Insulation and Dielectric Phenomenon (CEIDP), pp. 164–167 (2014)
6. Vogelsang, R., Fruth, B., Farr, T., Fröhlich, K.: Detection of electrical tree propagation by
partial discharge measurements. Eur. Trans. Electr. Power 15, 271–284 (2005)
7. Chalashkanov, N.M., Dodd, S.J., Dissado, L.A., Fothergill, J.C.: Pulse sequence analysis on
PD data from electrical trees in flexible epoxy resins. In: IEEE Conference on Electrical
Insulation and Dielectric Phenomenon (CEIDP), pp. 776–779 (2011)
 Spectral Power Analysis of Partial Discharges Waveforms 907

8. Sarathi, R., Oza, K.H., Pavan Kumar, C.L.G., Tanaka, T.: Electrical treeing in XLPE cable
insulation under harmonic AC voltages. IEEE Trans. Dielectr. Electr. Insul. 22(6), 3177–
3185 (2015)
9. Schurch, R., Orellana, L., Donoso, P., Ardila-rey, J., Montana, J.: Pulse waveform, phase-
resolved and pulse sequence analysis of partial discharges during electrical tree growth in
epoxy resin. In: International Symposium on High Voltage Engineering (ISH) (2017)
10. Chen, G., Tham, C.: Electrical treeing characteristics in XLPE power cable insulation in
frequency range between 20 and 500 Hz. IEEE Trans. Dielectr. Electr. Insul. 16(1), 179–188
(2009)
11. Bao, M., Yin, X., He, J.: Structure characteristics of electrical treeing in XLPE insulation
under high frequencies. Phys. B Condens. Matter 406(14), 2885–2890 (2011)
12. Gao, Y., Deng, Y.D., Du, B.X., Li, S.W., Wang, N.: Electrical treeing behavior in XLPE
under kHz-AC voltage. In: IEEE International Conference on Dielectrics, vol. 2, pp. 724–
727 (2016)
13. Ildstad, E., Fauskanger, K., Hølto, J.: Electrical treeing from needle implants in XLPE during
very low frequency (VLF) voltage testing. In: IEEE International Conference on Solid
Dielectrics (ICSD), pp. 800–803 (2013)
14. Ardila-Rey, J., Martínez-Tarifa, J., Robles, G., Rojas-Moreno, M.: Partial discharge and
noise separation by means of spectral-power clustering techniques. IEEE Trans. Dielectr.
Electr. Insul. 20(4), 1436–1443 (2013)
15. Martínez-Tarifa, J.M., Ardila-Rey, J.A., Robles, G.: Automatic selection of frequency bands
for the power ratios separation technique in partial discharge measurements: part I,
fundamentals and noise rejection in simple test objects. IEEE Trans. Dielectr. Electr. Insul.
22(4), 2284–2292 (2015)
16. IEC-60270:2000: High-voltage test techniques - Partial discharge measurements (2000)
 Development of Testing Method
for Static Electrification Within Power
Transformer Using Suspended Cellulose Fibers
in Insulating Oil

Masanobu Yoshida1(&), Soichi Moriguchi1, Gaku Sato2,

Yoshinori Konishi2, and Masayuki Hasegawa2
1
Chubu Electric Power Company, 20-1 Kitasekiyama Ohdaka-cho Midori-ku,
Nagoya, Japan
Yoshida.Masanobu@chuden.co.jp
2
YUKA Industries Co., Ltd., 2-3-17 Chidori Oota-ku, Tokyo, Japan

Abstract. This paper describes a new testing method that has been developed
to estimate the quantity of electrification in a power transformer. In addition, the
influence that the degradation of both a press board (PB; consists of cellulose
fibers) and insulating oil has on static electrification can be evaluated by using
the new testing method. In this testing method, cellulose fibers that are sus-
pended in insulating oil in a transformer were used. These fibers might come off
from the PB and the surface of insulating paper in a transformer. It was con-
sidered these cellulose fibers were useful samples for estimating the electrifi-
cation level of a PB in an actual power transformer. The insulating oil was taken
from a running power transformer, and cellulose fibers in the oil were filtered
with a metal filter. This filtering process was isolated from the air. These cel-
lulose fibers and insulating oil (oil with cellulose removed) were testing samples
for measuring the quantity of electrification in the static electrification process.
Static electrification occurs when insulating oil flows over cellulose fibers on a
metal filter. The volume of electrification in the static electrification process can
be known by measuring the electric current on a conductor connected between
the metal filter and the ground. Experimental results showed that the volume of
electrification on a degraded PB was larger than that of a new PB. Besides, there
was also a tendency for the volume of electrification on degraded insulation oil
(large value for electrostatic charging tendency) to be larger than that of new
insulation oil.

Keywords: Power transformer
Static electrification
Cellulose fiber

1 Introduction

An electrical breakdown may occur when static electrification phenomena occur in an

oil-immersed transformer. In conventional practice, the electrostatic charging tendency
(ECT) value of insulating oil is measured in order to evaluate the ease of generating a
charge on insulating oil. Various methods are used to measure the ECT value [1–6]. In
recent years, the accumulation charge density measuring method has also been
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 908–916, 2020.
https://doi.org/10.1007/978-3-030-31676-1_85
 Development of Testing Method for Static Electrification 909

developed and is widely used in Japan [7]. It was found that a degraded solid insulator
(e.g. PB, insulating paper) has a tendency to generate a lot of charge caused by static
electrification phenomena. Thus, it is necessary to estimate the magnification of charge
generation that depends on the degradation level of solid insulators. In the accumu-
lation charge density measuring method, degraded insulating oil and a degraded PB
(that was extracted from an actual power transformer) are needed to construct a sample
electrode for the static electrification test. Since it is usually difficult to pick out a
degraded PB from a running transformer, a new PB is used for the sample electrode; in
addition, a correction value corresponding to the transformer’s age is multiplied by the
new PB’s value. However, the degradation mechanism of a PB is not simple. Many
physical factors affect it (e.g. the past thermal history of the transformer, humidity in
the PB, and volume of oxygen in the transformer). Accordingly, development of a
direct method for evaluating the electrification level on a degraded PB is required
(without multiplying by a calibration factor).

2 Experimental Setup and Measuring Method

In this chapter, the experimental setup for measuring the quantity of charge generation
caused by the static electrification process is described.

2.1 Experimental Setup

Continuous ECT measurement apparatus was constructed (see Fig. 1) [8]. This appa-
ratus can measure the electrostatic charging tendency (ECT) value and the dielectric
loss tangent (tan d) of insulating oil under an arbitrary temperature condition. Sample
cellulose fibers were set on the metal filter in the electrification cell in Fig. 1.

Fig. 1. Experimental setup

910 M. Yoshida et al.

The electrification cell was set on the oil circulation pipe. Here, the metal filter was
insulated electrically from the oil circulation metal pipe and it was grounded through a
pico-ammeter. Insulating oil was circulated in the metal pipe. The oil flow velocity was
controlled by an oil pump. Static electrification occurred when insulating oil flowed
into the electrification cell. The volume of electrification on the static electrification
process can be obtained from the value of the electric current shown on the pico-
ammeter.

2.2 Structure of the Electrification Cell

The electrification cell consists of a filter holder and metal filter (see Fig. 2). The metal
filter has two roles. First, it keeps in place the cellulose fibers that are the source of
static electrification. Also, it functions as an electrode that detects the electric current
caused by static electrification. The filter pore size is 50 lm which is smaller than a
cellulose fiber’s width. The metal filter with an arbitrary amount of cellulose fibers is
clamped by a metal filter holder. This filter holder was insulated electrically from the
oil circulation metal pipe. The filter holder is compatible with an electrification cell on a
conventional mini-static electrostatic charging tendency tester [5, 6].

Cellulose fiber

Pore size φ50μm

(a) Metal filter

(b) Filter holder

Fig. 2. Metal filter and filter holder.

 Development of Testing Method for Static Electrification 911

2.3 Measuring Method

Filter holder was installed into continuous ECT measurement apparatus. The streaming
current was detected by a pico-ammeter when insulating oil flowed into the filter holder
(with cellulose fibers). The streaming current intensity is related to the intensity of the
static electrification.

3 Basic Study of the Streaming Current Measurement Using

Cellulose Fibers

This chapter will describe an investigation into the influence that the quantity of the
cellulose fibers, oil flow rate, and oil temperature have on the streaming current
intensity. In this examination, a new PB’s cellulose fibers and forcibly deteriorated
insulating oil (ECT value on mini-static electrostatic charging tendency tester = 250
pC/mL) were used.
The relationship between the streaming current intensity and the quantity of cel-
lulose fibers is shown in Fig. 3. This figure reveals a trend in which the streaming
current increased with an increase in the amount of cellulose fibres. It was also found
that a significant amount of streaming current was observed with an exceedingly small
cellulose fiber volume of 0.1 mg. In the case where the weight of the cellulose fibers
exceeded 1 mg, the cellulose fibers accumulated on top of each other on the metal
filter. The trend of a relationship between the streaming current and amount of cellulose
fibers changes in this condition.

Fig. 3. Relationship between weight of cellulose fibers and streaming current (1.2 mL/s, 23 °C).

Figure 4 shows the relationship between streaming current intensity and oil flow
velocity/oil temperature. A new PB’s cellulose fibers and degraded oil were used as
samples. The insulating oil was forcibly degraded. In particular, insulating oil with a
912 M. Yoshida et al.

copper catalyst (surface area 44.8 cm2/100 ml oil) was heated up to 105 °C degrees for
23 days. (ECT value on mini-static electrostatic charging tendency tester after heat-
ing = 140 pC/mL). There was a trend in which the streaming current increased with an
increase in the oil flow velocity, and the oil temperature also influenced the streaming
current intensity.

Fig. 4. Influence of flow rate and temperature (weight of cellulose fiber: 0.3 mg)

4 Streaming Current Caused by Insulator Other Than

Cellulose Fiber of PB

Many kinds of solid insulators other than natural cellulose fiber (PB and kraft insulating
paper) have been used in a power transformer, such as manila-hemp, amine-added
cellulose fiber, synthetic fiber (e.g. Rayon), chemical fiber (e.g. PET), and varnish.
There is a possibility that these insulators contaminate the cellulose fibers on the metal
filter when filtering the insulating oil. This chapter describes a measurement of the
static electrification intensity on various insulators that are used in a power transformer.
The streaming current value is normalized by the surface area of the insulator (mm2)
because a static electrification phenomenon occurs on the surface of solids. The value
of streaming current per unit surface area on the insulator was defined as the generated
charge density (the unit is pC/mL/mm2).
Figure 5 shows the ratio of generated charge density for various insulators com-
pared with cellulose fiber. Error bars show the minimum and maximum value of the
ratio of generated charge density for various test conditions (oil temperature and oil
flow rate). The solid points on Fig. 5 show the representative value that assumed the
condition of oil temperature and oil flow rate in the transformer (50 °C degrees,
4.2 mL/s). A high amount of static electrification was observed on PET. In contrast,
varnish has a small value. In other insulators, the generated charge density was at the
same level as cellulose fiber.
 Development of Testing Method for Static Electrification 913

It is thought that these contaminants have a very small influence on the measure-
ment of generated charge density, because the amount of contaminants was much
smaller than the cellulose fiber volume in insulating oil in the actual case.

(Cellulose fiber)

Fig. 5. Generated charge density of contaminant substance.

5 Streaming Current Measurement Simulated Actual Power

Transformer

In this chapter, the influence that degradation of a cellulose fiber has on streaming
current is discussed.

5.1 Sample Preparation of Degraded Cellulose Fibers

Cellulose fibers weighing 0.3 mg (picked off from a new PB) and 50 ml of new
insulating oil were sealed together in a vial. These samples were heated in a thermo-
static chamber at 105 °C for between 3 and 12 months. The degraded cellulose fibers
after heating processing were filtered with a metal filter. (The degraded cellulose fibers
were set on the metal filter, as a result).

5.2 Streaming Current on the Degraded Cellulose Fibers

The streaming current was measured under the test conditions of an oil temperature of
50 °C and oil flow rate of 1.2 mL/s by using continuous ECT measurement apparatus
(Fig. 1). Two degradation levels of insulating oil (non-degraded oil: ECT value on
mini-static electrostatic charging tendency tester = 8 pC/mL, and degraded oil: ECT
value = 140 pC/mL), and four degradation levels of cellulose fibers (heating period: 0,
3, 6, 12 months) were used as samples.
914 M. Yoshida et al.

Figure 6 shows the relationship between the generated charge density and degra-
dation level of insulating oil/cellulose fibers. There was a trend in which the generated
charge density increased as the degradation level of cellulose fibers became higher.
This result shows that static electrification phenomena occur remarkably on degraded
cellulose fiber. This is in agreement with another paper’s results.

Fig. 6. Generated charge density under different degraded oil and cellulose fiber conditions.

6 Streaming Current Measurement Using Cellulose Fibers

in Actual Power Transformer

This chapter describes the measurement of streaming current on cellulose fibers in an

actual power transformer and comparison with the value for a new PB’s cellulose
fibers. The specifications of the transformer and insulating oil are shown in Tables 1
and 2, respectively.
Twenty liters of insulating oil was acquired from an active transformer. The cel-
lulose fibers suspended in the insulating oil were filtered with a metal filter. It is known
that several hundreds of cellulose fibers are suspended in the oil (per 10 L) in a
transformer [9]. Generated charge density was measured by using these cellulose fibers
on the metal filter and insulating oil on actual power transformers.
Figure 7 shows the relationship between the ECT value of insulating oil (acquired
by using mini-static electrostatic charging tendency tester [5, 6]) and generated charge
density (acquired by continuous ECT measurement apparatus). It became clear that
compared with the case of using a new PB’s cellulose fibers, a higher generated density
value was shown when cellulose fibers in an aged (degraded) power transformer were
used. There is also tendency in which a high generated density value was observed with
a high ECT value of insulating oil.
 Development of Testing Method for Static Electrification 915

Table 1. Specification of transformer

Transformer A B C
Capacity [MVA] 100 1,000 1,000
Voltage class [kV] 500/275 500/275 500/275
Operating years 14 31 24

Table 2. Specification of insulation oil

Transformer A B C
ECT [pC/mL] 237.8 1344.5 89.5
Accumulation charge density [nC/cm2] [7] −11.0 −5.0 −7.6
Soluting copper [mg/kg] <0.01 <0.01 0.1
Sulfur content [%] 0.22 0.21 0.06
Sulfide [ppm] 220 215 61
Sulfoxide [ppm] 5 2 0

Fig. 7. Relationship between generated charge density and ECT value of insulating oil in power
transformer (50 °C, 1.2 mL/s).

7 Conclusions

A new method for diagnosing static electrification phenomena was developed. This
unique method uses cellulose fibers that are suspended in insulation oil in an actual
power transformer. The ease of static electrification was defined as generated charge
density (pC/mL/mm2). In a combination of degraded cellulose fibers and degraded
insulation oil, a high generated charge density value was observed.
The static electrification intensity on various insulators (except cellulose fibers) that
are used in a power transformer was also measured. These insulators could contaminate
insulating oil in an actual transformer and could have a bad influence on the new testing
916 M. Yoshida et al.

method. However, it was thought that these contaminants have a very small influence
because the volume of contaminants was much smaller than the cellulose fiber volume
in an actual case.

References
1. Morin II, A.J., Zahn, M., Melcher, J.R.: Fluid electrification measurements of transformer
pressboard/oil insulation in a couette charger. IEEE Trans. Electr. Insul. 26(5), 870–901
(1991)
2. Lee, M.J., Nelson, J.K.: Flow-induced electrification and partial discharge measurements in
transformer duct structures. IEEE Trans. Electr. Insul. 26(4), 739–748 (1991)
3. Oommen, T.V., Lindgren, S.R.: Streaming electrification study of transformer insulation
system using a paper tube model. IEEE Trans. Power Deliv. 5(2), 972–983 (1990)
4. Kedzia, J.: Investigation of transformer oil electrification in a spinning disk system. IEEE
Trans. Electr. Insul. 24(1), 59–65 (1989)
5. Oommen, T.V., Petrie, E.M.: Electrostatic charging tendency of transformer oils. IEEE Trans.
Power Apparatus Syst. 103(7), 1923–1931 (1984)
6. CIGRE JWG 12/15: Static Electrification in Power Transformers, GIGRE Technical Brochure
170 (2000)
7. Electric Technology Research Association: The guideline for refurbishment of electric power
transformers. Electr. Technol. Res. 1(65) (2009)
8. Sato, G.: Study of evaluation method of electrostatic charge tendency for transformers oil by
continuous circulating measurement. In: IEEJ PES 2010, no. 272 (2010)
9. Yoshida, M.: Diagnosis method of aging degradation for power transformer using cellulose
fibers in insulating oil. CIGRE SC A2 & C4 JOINT COLLOQUIUM (2013)
 Development of a Digital Twin
for the Determination of Transmission Line
Conductor Asset Health

J. C. Toth1(&), J. Marmillo2, and G. Biedenbach2

1
Enginomix Consulting Inc., 6355 Salish Dr., Vancouver, BC, Canada
janos.toth@enginomix.com
2
LineVision Inc., 444 Somerville Ave., Somerville, MA 02143, USA

Abstract. Electrical transmission lines are an essential part of a functional

economy delivering electricity to residential, commercial and industrial cus-
tomers. One important component of the transmission line is the conductor and
transmission line asset managers are responsible for decisions on how to
maintain and, when necessary, replace conductors currently in service. The
decision-making process requires reliable long-term and up-to-date information
on the state and condition of the conductors. Currently, most asset managers
lack such information. LineVision is a company that provides non-contact
transmission line monitoring of conductor assets. Originally developed for
Dynamic Line Rating (DLR), the LineVision systems collect information that is
invaluable for utility asset managers. An Asset Health Module (AHM) has been
developed which creates an asset digital twin, integrating both historical and
real-time data from the target transmission line in order to compute its baseline
Asset Health. These inputs include high-temperature cycling, extreme sag and
blowout forces, occurrences of galloping and their intensity, vibration and
precipitation loads from icing, and other events the conductor experiences
during its lifetime which negatively impact asset health. Without monitoring
equipment many of these events go unrecorded and even unnoticed unless they
cause catastrophic failures which can be dangerous, cause forced service
interruptions, and take months or years to repair. The AHM integrates engi-
neering specifications of the design, information collected after a line is built,
and real-time field measurements obtained via a field-based monitoring system.
Based on field measurements, an extensive proprietary model determines the
operating conditions and the asset health of the transmission line conductor. The
output of the system is a real-time, continuous assessment of a transmission
line’s operational condition and an estimate of the conductor’s remaining life
based on the historical and collected data, industry standards, and proprietary
conductor analysis. The AHM can assist asset managers and system operators to
prioritize the criticality of multiple monitored sections. The AHM improves line
reliability, overall system reliability and can lead to significant cost savings by
supporting asset management decisions, such as postponed conductor replace-
ment which is the costliest conductor asset decision. The details of the AHM and
the approaches it uses along with the benefits for transmission line operators will
be discussed in the paper.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 917–925, 2020.
https://doi.org/10.1007/978-3-030-31676-1_86
918 J. C. Toth et al.

Keywords: Dynamic Line Rating
DLR
Remote sensing
Transmission grid

safety
Transmission grid reliability improvement
Asset management

Conductor asset health model
Transmission line monitoring

1 Introduction

Delivering reliable electricity to residential, commercial and industrial customers is an

essential part of well-functioning economies. A large portion of this electricity is
delivered by the high voltage transmission grid, and the electrical overhead transmis-
sion line conductors. The conductors represent one of the most significant cost com-
ponents of the entire line. Maintaining and making operational and planning decisions
about the conductors are an important part of overall transmission line asset manage-
ment. The decision-making process requires well supported up-to-date information
about the conductor’s current condition. Unfortunately, most transmission line asset
managers do not have access to actual asset condition data for their work.
When the condition of a conductor is assessed, the results are mainly based on
visual inspections carried out over large time intervals which can be up to several years.
However, there is now available a system, called LineVision, that monitors the con-
ductor’s motion as well as other physical and electrical properties and conveys that
information to decision makers in near real-time that can provide significant reliability,
monetary and safety benefits. Moreover, this equipment is easy to install and relocate as
needed. It Is not attached to the conductor either.
LineVision’s equipment was originally developed to monitor transmission line
Dynamic Line Rating (DLR) by monitoring the position of the conductor with a non-
attached electromagnetic field (EMF) sensor. Further developments have been imple-
mented and now the LineVision V3 system is able to precisely monitor conductor sag,
horizontal motion, conductor temperature, galloping, ice build-up, as well as detecting
anomalies using a combination of sensor types. The collected information can be
further analyzed, and an Asset Health Model of the transmission line conductor has
been developed. The model works by creating a “digital twin” of the span which is
relatable to the overall stringing section.

2 The Model

The model uses a mathematical representation of the conductor based on the field
measurements called the “digital twin” (DT). This model incorporates the various
conductor support, sag, temperature, position, and motion for the specific span. The
model parameters can accommodate the condition changes of the conductor and forms
the basis of further evaluation of conductor status.
The available mechanical properties and catenary position of the conductor are used
to create a reference state that is corrected to a specific temperature. The reference state
is then compared to the conductor’s behavior in real-time. A variety of factors can be
observed which negatively impact the conductor asset health.
 Development of a Digital Twin 919

The major components that can be derived from the measurements are: Thermal
aging, creep, precipitation overload, sag change, and galloping. There is development
to monitor aeolian vibration, and natural frequency changes of the conductor. The
information and results computed by the Asset Health Model can be sent to the
transmission operation control center, transmission line asset management department,
and/or field operations.

3 Principles of the LineVision V3 System’s Operation

The LineVision V3 system is a remote, non-conductor-contact transmission line

monitoring system that is typically installed on the transmission line’s supporting
monopole or lattice structure. The sensors continuously monitor multiple aspects of the
lines, both physically in space and electrically. The equipment uses a combination of
patented EMF and optical sensors to measure the electrical properties of the line and
the physical position of the conductors. This data is then used to calculate asset health
parameters and dynamic and forecasted line ratings which are delivered to the trans-
mission line owner via a secure data feed or private cloud interface.

4 Areas of the Model

The developed Asset Health Model covers Thermal aging, Historical sag, Precipitation
overload, Sag change, Galloping, and Aeolian Vibration.

4.1 Thermal Aging

One of the most important variables which must be detected by a real time monitoring
system is the temperature of the conductor. This is derived from the measured sag and
the line calibration equation [1] of the monitored span as related to that span’s specific
geometry. Transmission lines are designed for a certain maximum operational tem-
perature. Operating the line at a high temperature above 90 C can degrade the strength
of the conductor and can lead to rapid deterioration of the conductor’s health. There are
several well-established thermal degradation models that provide an indication of the
impact of high operational temperature events and their duration on the remaining
strength of the conductor. The most widely accepted and used model was developed
and published by Harvey [2, 3]. IEEE standard 1293 provides detailed information [4].
Other models are also available published by Morgan [5, 6], Jakl et al. [7] and Bhuiyan
et al. [8] (Fig. 1).
920 J. C. Toth et al.

Fig. 1. Catenary change after installation where T-max is the maximum conductor temperature
with conductor temperature, ice loads, wind loads and time after installation where T-max is the
maximum conductor temperature [9].

From the sensor measurements, including the data sets that measure temperature
cycle levels and durations (conductor exposure), the strength reduction can be esti-
mated. More accurate models can be developed for the conductor using detailed
temperature strength reduction in laboratory environments. Using the digital twin
model, the remaining strength of the conductor is compared to the line’s acceptable
limit. This acceptable limit must be defined by the transmission line’s design
parameters.
When the conductor’s historical records are evaluated over time, the annualized
degradation rate can be determined. Assuming a constant rate of degradation and
conductor loss of strength, the conductor’s remaining useful life can be estimated. The
“digital twin” of the conductor can be used for checking design, schedule replacement
and making business justification on decisions regarding the conductors.

4.2 Historical Sag

For as long as monitoring equipment has been installed, it is able to provide historical
records of the individual conductor sag and horizontal motion. These sag measure-
ments and details along with the calculated deviation from the originally designed state
and temperature corrections can be used for computing the long-term elongation of the
conductor (creep). These permanent plastic elongations can then be evaluated and
compared to acceptable limits.
 Development of a Digital Twin 921

Historical records of the sag and the real time measured sag can be compared to the
engineering designed acceptable criteria and the remaining life as a percentage of the
initial design can be determined. When sag monitoring is carried out from the instal-
lation date of the conductor, the various conductor elongations can be directly com-
puted (Fig. 2).

Fig. 2. Types of elongation in conductor [10].

4.3 Precipitation Overload

Over the lifespan of a conductor, it may be experience abnormal mechanical loading
caused by various forms of precipitation such as snow loading, ice formation, or rime.
In cloud icing, or freezing rains, ice build-up from falling snow can be one of the
highest mechanical and structural loads that a transmission line experience.
LineVision equipment continuously monitors the conductor sag. When ice or snow
build up occurs, it is usually progressive and gradual. The sensors can indicate that on-
conductor precipitation build up is in progress and send an alert can to the transmission
control centre. This provides a valuable indicator to system operators that an event
which can compromise system reliability is occurring. The alert may trigger inter-
vention from the control centre to implement de-icing protocols, or system operating
orders to mitigate the precipitation event impact on the transmission line. For example,
system operators may implement switching protocols to load the icing line with
additional current, so that it heats up and the ice is melted off the conductor, preventing
it from experiencing plastic deformation from the additional ice loading.
The digital twin model of the span can be used as a reference for evaluating how
much ice or snow formation has occurred on the conductor at the current situation.
From the sag and the original position of the conductor the unit weight of the pre-
cipitation accumulated on the conductor can be computed using catenary formulas. The
input items are the physical and designed properties of the conductor and the sag
922 J. C. Toth et al.

position of the conductor. The output is the additional load form the precipitation acting
on the line.
These sag measurements will also indicate any additional, operational and safety
related issues from the increased sag under precipitation load. These can include
potential flash over or conductor sagging excessively at certain critical locations where
these excessive sags cannot be tolerated. Examples are long spans crossing roads,
railways and waterways. Additionally, long crossings in commercial, industrial and
residential areas where clearances need to be kept for safety reasons are of most
concern.
Based on the measurements during a snow or icing event, the transmission line
asset management group as well as the transmission line design group can gain
invaluable data on the local load conditions of the line. Moreover, the duration of the
ice build-up and the rate of loading and unloading is also monitored in near real time.
This is also very valuable for both system operation and asset management point of
view.
After the icing or snow event is over, the sag of the conductor is again measured
and taken in account as a part of the AHM system’s continuous operation. By ana-
lyzing the sag measurements after the precipitation loading event and comparing them
to the digital twin before the event, the plastic deformation of the conductor can be
computed. The sag condition after the icing event can be also evaluated and decisions
can be made if remedial actions are needed on the line or not due to excessive sag
(Fig. 3).

Fig. 3. Clear example of excessive sag from icing on a transmission line [11].
 Development of a Digital Twin 923

4.4 Sag Change

LineVision equipment monitors in real time the conductor’s position and relates this to
its sag and horizontal motion. If a sudden change is detected which may be an indi-
cation of damage to either the conductor support (structure, cross arm, insulator,
hardware etc.) or to the conductor itself, the asset health model can trigger an anomaly
detection alert. For example, a tree falling on the line, a large branch of a tree blown
into the line, a piece of tarp or other item blown by the wind into the conductor. Other
security concerns can be detected. If one conductor were to suddenly start sagging
lower due to a foreign intrusion, this change in position can be immediately detected.
These changes can flag potential problems with the line’s operating condition and
trigger an alert to a utility that a site inspection is required to determine the severity of
the issue. This type of transmission line monitoring equipment increases reliability and
safety of the line.

4.5 Galloping
Conductor galloping is the high-amplitude, low-frequency oscillation of overhead
power lines due to wind. The movement of the wires occurs most commonly in the
vertical plane, although horizontal or rotational motion is also possible.
Galloping can have significant influence on the conductor remaining life. Signifi-
cantly larger axial forces caused by the conductor’s motion and possible phase to phase
touching causing flashover burns on the conductor surface can happen.
LineVision equipment is able to monitor the conductor motion in near real time
which can provide operators with information that can indicate when dangerous con-
ductor galloping is occurring and provide the conductor’s minimum and maximum
amplitude of galloping.
Once the conductor position measurement has been recorded, the conductor’s
modal frequency and amplitude determined. From the conductor motion’s frequency
and amplitude, structural and mechanical overload forces of the conductor and structure
can be computed. The formulas are available in Ma et al. [12]. After the galloping event
has occurred, the catenary position of the conductor is identified and compared to pre-
galloping position of the Digital Twin. Any deviations are then identified which are
evidence of possible plastic deformation or damage to the conductor support.
Galloping can cause permanent damage to bundled conductors creating a twisted
permanent position of the bundle after the galloping is over. This twisted position also
can be detected and flagged for further attention. Field services crews can be alerted
and dispatched to site addressing potential damage to the line from galloping.
After evaluating the potentially damaging events, asset managers are able to
evaluate and justify required actions for the line’s improvement such as redesign,
phase-to-phase insulators or additional dampers.
Based on historical records of the conductor’s position along with damage suffered
from galloping as determined by the minimum and maximum amplitudes of the gal-
loping, estimates can be generated for the remaining life of the conductor (Fig. 4).
924 J. C. Toth et al.

Fig. 4. Mathematical model of a galloping conductor [12].

4.6 Aeolian Vibration

Aeolian vibration is characterized as high frequency, low amplitude conductor motion.
Caused by vortices formed via non-turbulent air flow over a conductor. Generally
produced by wind velocities below 25 kph (15 mph) [13].
Approximate frequency range for aeolian vibration is 3 to 150 Hz indicating that
the amplitude of the conductor motion is smaller than the diameter of the conductor.
Aeolian vibration can cause metal fatigue in the conductor and such damage can
develop within 3 months to 20 years of a conductor exposed to aeolian vibration.
Monitoring aeolian vibration and detecting changes could indicate that something
happened to the span.
For example, dampers moved or damaged. Further benefit of monitoring aeolian
vibration is that the potentially implemented vibration mitigating measures perfor-
mance is monitored and can be evaluated. Possible indicating either satisfactory per-
formance or he need for further intervention.

5 Conclusions

A transmission line conductor Asset Health Module (AHM) has been constructed
which creates an asset digital twin, integrating both historical and real-time data from
the target transmission line in order to compute its baseline Asset Health from Line-
Vision system data. These inputs include high-temperature cycling, extreme sag and
blowout forces, occurrences of galloping and their intensity, vibration and precipitation
loads from icing, and other events the conductor experiences during its lifetime.
Based on the various derived models, such as thermal aging, historical sag, sag
under precipitation, and galloping, and aeolian vibration, the plastic deformation and
other permanent deformities that the conductor has experienced can be determined.
 Development of a Digital Twin 925

These deformations can be compared to the allowable and tolerable limits as

defined by transmission engineers and a historical rate of wear and tear can established.
The remaining life of the conductor can then be estimated, and data from the AHM
can be used by asset managers to improve system reliability and inform decisions on
conductor maintenance and replacement.

Acknowledgements. Authors would like to acknowledge the support and feedback from other
LineVision members, Hudson Gilbert CEO, Nathan Pinney CTO.

References
1. CIGRE: Guide for Application of Direct Real-Time Monitoring Systems. Technical
Brochure 498 (2012)
2. Harvey, J.R.: Effect of elevated temperature operation on the strength of aluminum
conductors. IEEE PES Winter Meeting Paper T 72-189-4 (1971)
3. Harvey, J.R.: Effect of elevated temperature operation on the strength of aluminum
conductors. IEEE Trans. PAS-91(5), 1769–1772 (1972)
4. IEEE Guide for Determining the Effects of High Temperature Operation on Conductors,
Connectors and Accessories. IEEE Standard 1283-2004 (2004)
5. Morgan, V.T.: The loss of tensile strength of hard-drawn conductors by annealing in service.
IEEE Trans. Power Apparatus Syst. PAS-98(3), 700–709 (1979)
6. Morgan, V.T.: Effect of elevated temperature operation on the tensile strength of overhead
conductors. IEEE Trans. Power Deliv. 11(1), 345–352 (1996)
7. Jakl, F., Jakl, A.: Effect of elevated temperatures on mechanical properties of overhead
conductors under steady state and short-circuit conditions. IEEE Trans. Power Deliv. 15(1),
242–246 (2000)
8. Bhuiyan, M.M.I., Musilek, P., Heckenbergerova, J., Koval, D.: Evaluating aging charac-
teristics of electric power transmission lines. In: CCECE 2010, Calgary, AB, pp. 1–4 (2010).
https://doi.org/10.1109/ccece.2010.5575137
9. CIGRE: Sag-tension calculation methods for overhead lines, Technical Brochure 324 (2016)
10. CIGRE, Paris: Conductors for the uprating of overhead lines, Technical Brochure 244, April
2004
11. Our Experience of Ice on Powerlines, Statnett, Icebox, Presentation, Oslo, Norway, 3
October (2018)
12. Ma, G.-M., Li, Y.-B., Mao, N.-Q., Shi, C., Li, C.-R., Zhang, B.: A fiber bragg grating-based
dynamic tension detection system for overhead transmission line galloping. Sensors 18(2),
365 (2018)
13. Transmission Line Reference Book, Wind-Induced Conductor Motion, EPRI Project 792.
The Orange Book (2005)
 Estimation of the Ageing Condition of Oil-Filled
Transformers Based on the Oil Parameters
Using a Novel Fuzzy Logic Algorithm

Tobias Kinkeldey1(&), Tobias Münster1, Peter Werle1, Suwarno2,

Kai Hämel2, and Jörg Preusel3
1
Institute of Electric Power Systems, Division of High Voltage Engineering and
Asset Management, Schering-Institute, Leibniz Universität Hannover,
Hannover, Germany
tobias.kinkeldey@ifes.uni-hannover.de
2
School of Electrical Engineering and Informatics,
Institute of Technology, Bandung, Indonesia
3
GRIDINSPECT GmbH, Feldatal, Germany

Abstract. The condition of the transformer insulation determines the remaining

life of the transformer. Over the transformer service life, both liquid and solid
insulation undergoes a continuous aging process under electrical, chemical,
mechanical and thermal stresses. The insulating liquid of a transformer can be
reconditioned or replaced; however, this is not the case for the cellulose insu-
lation. Therefore, the condition of the paper insulation is the major factor for
determining the aging status of a transformer. To assess the paper condition, the
common method is to measure the degree of polymerization (DP) of the paper
insulation as a significant parameter. This method is destructive as it requires a
sample of paper from inside of the transformer. Therefore, it could not be
applied for transformers in operation. There are several approaches to approx-
imate the DP value without direct measurements of paper samples.
This research presents an improved method based on a fuzzy logic system for
the estimation of the DP based on dissolved gases and chemical parameters of
the liquid insulation. The algorithm is developed to create the rules for the use of
fuzzy sets based on the information gain extracted by the entropy of the labo-
ratory measurement data. This algorithm employs entropy to examine the
sample homogeneity. Entropy is a measure of information theory that can
determine the dataset characteristics concerning impurity and homogeneity. The
algorithm uses fuzzy sets of oil parameters like Acidity, Interfacial Tension
(IFT), Carbon Dioxide (CO2) and Carbon Monoxide (CO) and the breakdown
voltage (BDV) for determination of the DP value.

Keywords: Fuzzy logic algorithm
DP value
Paper ageing

1 Introduction

High voltage equipment such as power transformer, circuit breaker, surge arrester, etc.
are highly important components in power systems. Among them, power transformers
are usually quite reliable, and have an average life expectancy of approx. 30 years [1].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 926–936, 2020.
https://doi.org/10.1007/978-3-030-31676-1_87
 Estimation of the Ageing Condition of Oil-Filled Transformers 927

To stay competitive in a changing market, the utility companies expose the assets to
higher electrical and mechanical stresses in long-term use for profit optimization. Thus
increase the probability of failures, as several studies have shown [2, 3]. This makes
reliable diagnostic and maintenance strategies necessary for the asset manager, because
of the fact that a power transformer failure is normally accompanied with high
replacement costs and long outage times. This includes not only the repair or
replacement costs for the asset, but also the consequential costs, such as contractual
penalties for undelivered energy as well as loss of revenues [4].
The failure risk of a transformer is depending on the DP value as the average length
of the cellulose chains. While a new transformer has a DP value of 1200 and higher, a
DP value of 200 is considered as the end of life of this asset. Indirectly, the estimation
of the DP value could be also done using a with the furan content in the oil. But there is
the uncertainty that the DP value is not correctly calculated, especially in case of a local
overheating, where a local area of the insulation result in a rise of the furan content.
However, the measured furan content is therefore not an adequate indicator of the
average DP value of the whole insulation material. Therefore, additional indirect
methods are needed in order to obtain more reliable results regarding the DP of the
whole insulation system.
In this contribution, a method is presented, which includes several oil and gas
parameters in order to make a statement on the overall condition of the paper insula-
tion. Accordingly, a special fuzzy logic algorithm is developed to calculate the average
DP value.

2 Initial Situation

In a previous publication, a method was presented in which fuzzy logic should be used
to calculate the expected DP value [5]. It should be mentioned that this method has
shown some optimization potentials regarding the depth of the state tree and the initial
states. To enhance the accuracy of the calculation method, 20 initial data sets, obtained
by an accelerated thermal aging of a transformer model with uninhibited insulation oil
were used to establish the evaluation algorithm [6–8].
The paper samples of 160 g impregnated with 1600 g of insulating liquid were
aged with a copper rod of 66 g as the catalyst at 130 °C over a period of up to 15
weeks. Within the scope of this study, the measurements of the oil characteristics were
performed according to IEC-60422-2013 [9]. The concentration of fault gases were
determined by means of gas chromatography in accordance to IEC 60567 [10]. The DP
value of the aged paper sample was measured according to the IEC-60450-2008 [11].
The investigation revealed that parameters well-correlated to the DP value are the
Acidity, IFT, CO and CO2 concentration, the Water content and the Breakdown
Voltage (BDV) [6–8]. A statistical analysis using Pearson correlation found the highest
degree of correlation with respect to the DP value for these parameters.
928 T. Kinkeldey et al.

3 Decision Tree

Based on the measurement results a state tree was developed using an ID3 algorithm
(Iterative Dichotomiser 3). The state tree attributes the changes in the physico-chemical
parameters to the DP value. This gives the fuzzy algorithm described in this work a set
of rules and defines possible conditions. The ID3 uses the entropy and the information
gain to define a decision tree.
To create a decision tree with ID3 algorithm, two types of variables, the predictors
and the target value, are required. The predictors are independent variables, e.g. gas
concentrations like CO and CO2 or oil properties like Acidity, IFT and BDV, whereas
the target is the product of the predictors, in this case the DP value.
The ID3 algorithm uses the entropy to calculate the sample homogeneity. For
samples which are homogeneous, the entropy value is zero. To calculate the entropy of
the DP, Eq. (1) is employed. Hereby, E(DP) is the Shannon entropy and pi the
probability of the DP condition [12, 13].
Xm
EðDPÞ ¼ ði¼1Þ
pi
log2 ðpi Þ ð1Þ

In this study, the DP values are classified in four groups with corresponding
probability, as shown in Table 1.

Table 1. DP condition and corresponding probability values.

DP condition Occurrences Probability
New 2 0.1
Good 6 0.3
Average 8 0.4
Aged 4 0.2

The information gain is defined as the difference in the entropy of the predictors and
the target, as shown in Eq. (2) [13, 14].
X j Sv j
GainðDP; AÞ ¼ EntropyðDPÞ  EntropyðSv Þ ð2Þ
v
j Sj

Where A is the particular attribute, |Sv| is the number of elements in the considered
class of the attribute, |S| stands for the total number of elements in the considered class
of the attribute and v for all the possible values of the attribute. The calculation of the
information gain for the DP value must be done for all input variables.
The predictor that has the highest information gain will be set as the root in the
decision tree. The next step is similar to the previous step with a new target and new
predictors, where the new target is the class of the attribute with the highest information
gain. This procedure is repeated until the predictors are processed. The calculation
procedure for the ID3 is shown in Fig. 1.
 Estimation of the Ageing Condition of Oil-Filled Transformers 929

Fig. 1. Calculation procedure of the ID3 algorithm.

As shown in Fig. 2, a four-layer decision tree is the result from the ID3 calculation,
based on the input variables.

Fig. 2. 4-level-decision tree.

930 T. Kinkeldey et al.

The decision tree is shown from top to bottom. The top is called the root. The
distribution from top to bottom involves dividing the data homogeneously into subsets
containing instances of the same value.
Table 2 shows the corresponding defined rules.

Table 2. Linguistic fuzzy rules of the decision tree.

No. Linguistic rule
1 If Acidity is fair and BDV is poor, then DP is aged 1
2 If Acidity is fair and BDV is fair, then DP is aged 2
3 If Acidity is fair and BDV is good and CO2 is in condition 1, then DP is good 2
… …
13 If Acidity is good and CO is in condition 4 and water is poor and CO2 is in
condition 1, then DP is good 2
14 If Acidity is good and CO is in condition 4 and water is poor and CO2 is not in
condition 1, then DP is average

The elements marked in blue are conditions calculated directly from the ID3
algorithm. The orange-marked elements are additionally introduced conditions in order
to increase the accuracy and to supplement missing information in the state tree due to
missing input data.
For example, “con 2” was not presented in the input data, when the classes were
calculated for CO values. This case was additionally introduced to get a defied state for
the output. The state “good 2” represents a good state, which indicates a higher DP
value than the state “good 1”. Analogously, the state “aged 2” represents a higher DP
value than “aged 1”. In the parameter “BDV”, the state “aged 1” was introduced in
order to map a very poor state of the insulating oil.

4 Fuzzy Inference System (FIS)

Fuzzy logic is a method that transforms crisp values of physical independent param-
eters to linguistic values to draw a conclusion from uncertain variables enabling a
decision making [14, 15]. Therefore, a scheme of fuzzy sets, for CO, CO2, Acidity, IFT
and BDV were defined and the membership of the input data to the fuzzy sets will give
a membership level to the fuzzy set of the output: the DP value. The calculation for the
output level of membership is controlled by a fuzzy inference system, which means
logical operations like minimum implication and maximum aggregation. The structure
of the presented Fuzzy Inference System (FIS) method is shown in Fig. 3.
 Estimation of the Ageing Condition of Oil-Filled Transformers 931

Fig. 3. Structure of the FIS calculation method.

The process of fuzzy inference is governed by logical if-then rules. These rules are
created by using the former described decision tree [15, 16]. In Fig. 4, the application
of the min implication for the following rule is shown:
If “Acidity” is “good”, and “CO” is in “con 4” and “CO2” is in “con 1” and
“Water” is “poor”, then “DP” is in con “good 1”.
That means, in the first step, the Algorithm focuses on the comparison between
Acidity and DP, and estimates a degree of membership between this input parameter
and the DP as output parameter according to the rule assigned class. In the second step,
the CO is considered as input parameter and the degree of membership determining the
DP is estimated by the degree of fulfillment of the input parameter. It is represented by
the blue lines in Fig. 4. That process repeats until all the input parameters are used to
get a degree of membership of the output parameter i.e., DP. Finally the minimum
value (min inference) of the Degree of membership is chosen to estimate the output of
degree of membership for this rule.

Fig. 4. Application of the min implication.

932 T. Kinkeldey et al.

As it can be seen the most effective parameter in this case is the CO content, so it
limits the area of the degree of membership in the output set.
This procedure must be repeated for all the rules for the given input values. The
results are output quantities with individual degrees of membership. These output
quantities are superimposed using the maximum aggregation to a total output quantity,
as the Fig. 5 illustrates.

Fig. 5. Application of the maximum aggregation.

Fig. 6. Defuzzification step.

 Estimation of the Ageing Condition of Oil-Filled Transformers 933

The Defuzzification is done to get the output value as a natural value by mathe-
matical calculation of the DP value using the center of gravity (CoG) method, see
Fig. 6.
The influencing factors are the degree of the membership and the condition classes
determined by the rules. As it can be seen the final DP value is influenced by all the
output quantities involved. That means, if results with worse classes are determined,
then the center of gravity of the output quantity shifts towards lower DP values.

5 Results

In order to evaluate the functionality of the FIS model, the algorithm developed by
laboratory data was examined based on the transformer data of real cases, in which the
required physico-chemical and gas characteristic data were known.
In the first study, a fleet of transformers of a utility company as asset owner was
considered. The DP values were calculated by measurement of the concentration of
Furfural (2FAL) in the insulating oil and the evaluation using the Chendong equation,
as shown in Eq. 3 [17].

log10 ð2FALÞ  1:51

DP ¼ ð3Þ
0:0035

Table 3 shows the results of the DP calculation using the FIS algorithm for 69
power utility transformers.
It can be seen that the FIS algorithm determines the correct condition class for the
DP for the overall possible condition states and the DP value determined using the oil
and gas characteristic data is in the range determined by the Chendong method.

Table 3. FIS results for transformer fleet in service.

No. Calculation FIS
DP Condition DP Condition
1 861 Good 885 Good
2 778 Good 616 Average
3 519 Average 906 Good
4 1112 New 1016 New
5 981 Good 989 Good
6 337 Aged 306 Aged
7 392 Aged 301 Aged
8 193 Aged 325 Aged












66 694 Good 800 Good
67 1256 New 755 Good
68 275 Aged 570 Average
69 269 Aged 394 Aged
934 T. Kinkeldey et al.

Furthermore, the study shows that most of the considered assets were classified
correctly. Of the 69 units considered, 38 were classified correctly, as shown in Fig. 7.
This indicates a significant improvement compared to the previous result of 31 cor-
rectly classified units out of the total 69 transformers.

Fig. 7. Accuracy of the FIS on a transformer fleet in service.

Besides in the case of not-correctly classified transformers, none of them catego-

rized as “new”, which not the case in the reality. This eliminated critical false indexing.
Furthermore, in this study, the difference between the determined condition classes
by the FIS model and Chendong method were usually one class, as shown in Fig. 8.

Fig. 8. FIS model results for the not correctly classified transformers.
 Estimation of the Ageing Condition of Oil-Filled Transformers 935

Only at three assets is still a class difference of more than one. However, this seems
less critical, because they are already aged units and the algorithm indicates the higher
aged state, which should lead to a more intense inspection of the asset.
But it needs to be emphasized that the DP values calculated by Chendong was used
as reference and hence was considered as the true values. However, judgment about the
correctness of the results could not be done.
Therefore, in the second study, validity of the FIS algorithm to measure data of
transformers was checked.
Here power transformers as the assets are not diagnosed as faulty; however, they
were taken out of the service and subjected to a post-mortem analysis. The DP values
were determined by a direct analysis of the paper and no furan analysis was carried out.
The results are listed in Table 4.

Table 4. FIS results of post mortem analysis of the transformers.

No. Acidity IFT CO CO2 BDV Water Given FIS
mgKOH/g mN/m ppm ppm kV ppm DP
1 0.07 25 233 2409 73 9 705 878
2 0.05 23 153 3183 77 10 765 910
3 0 33 42 1013 70 4 1192 1015
4 0.04 28 39 662 75 10 910 936

It can be seen that the algorithm correctly classifies all 4 considered units based on
the input data.
Deviation regarding those results with direct measurement and those estimated by
the FIS algorithm is ranging from a few DP up to a DP value of 170.

6 Conclusion

This paper presents a method for the calculation of the degree of polarization of
transformer papers without furan measurement, based on oil and gas parameters. In this
case, an algorithm is developed with the basis of defined input fuzzy sets using a fuzzy
inference system, which means logical operations like minimum implication and
maximum aggregation. The algorithm generates an output set with appropriate degree
of membership. The calculation process is controlled by rules that are specified from a
decision tree. With the input data, it was possible to generate a 4-stage state tree that
takes input data IFT, Acidity, CO, CO2 and BDV into account.
The functionality of the algorithm has been proven for a wide range of paper aging
condition. The applicability of the developed algorithm is by conducting to case
studies, in which the data of transformers are provided by a German transformer service
provider, and transformer fleets of an Indonesian network operator.
The results show that this model can be used to predict the value of DP accurately.
To improve the reliability of the results, this method could be combined with Furan
analysis.
936 T. Kinkeldey et al.

Acknowledgments. The authors would like to express their graduate to GRIDINSPECT GmbH
and AiF/ZiM for the financial support as well as Weidmann Electrical Technology AG for the
support with insulation materials and Analysen Service GmbH Leipzig for the analysis. Fur-
thermore, the authors would like to thank the company ABB for the provision of comparative
data.

References
1. Oommen, T., Cellulose, A.P.: Insulation in oil filled power transformer: part II-maintaining
insulation integrity and life. IEEE Electr. Insul. Mag. 22(2), 5–14 (2006)
2. CIGRE 227 - Life Management Techniques for Power Transformer. CIGRE (2003)
3. CIGRE: Tutorial of CIGRE WG A2.37, Transformer Reliability Survey, Technical Brochure
642
4. Balzer, G., Schorn, Chr.: Asset Management für Infrastrukturanlagen - Energie und Wasser.
Springer (2010). ISBN 978-3-642-05391-7
5. Kinkeldey, T., Münster, T., Werle, P., Nasution, E., Suwarno, S., Hämel, K., Preusel, J.:
Estimation of the degree of polymerization (DP) of oil-paper insulated transformers using a
novel fuzzy logic algorithm. In: VDE Hochspannungstechnik (ETG) 2018, Berlin, Germany
(2018)
6. Kinkeldey, T., Werle, P., Münster, T.: Investigation on aging markers of thermally
accelerated aged oil-impregnated papers. In: ISH in Buenos Aires, Argentina (2017)
7. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel J.: Investigation on the accelerated
ageing behavior of oil-paper-insulation using different insulation oils. In: VDE-
Hochspannungstechnik (ETG) 2018, Berlin, Germany (2018)
8. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel, J.: Investigation on ageing
parameters of a thermally accelerated aged paper-oil-insulation in a hermetically sealed
system. In: CMD 2018, Perth, Australia (2018)
9. IEC Std 60422: Mineral Insulation Oil in Electrical Equipment: IEC (2013)
10. IEEE Std C57.104: IEEE Guide for the Interpretation of Gasses Generated in Oil-Immersed
Transformers. IEEE, New York (2009)
11. DIN EN 60450: Messung des durchschnittlichen viskosimetrischen poly-merisationsgrades
von neuen und gealterten cellulosehaltigen Elektroisolierstoffen (2008)
12. Han, J., Kamber, M., Pei, J.: Data Mining: Concepts and Techniques (2012). ISBN 978-0-
12-381479-1
13. Runkler, T.A.: Data Mining (2010). ISBN 978-3-8348-0858-5
14. Chowdary, M.L., Singh, A., Bansal, R., Jarial, R.K.: A fuzzy logic approach to analyze
change in dissolved decay content in correlation with density, IFT and acidity of transformer
oil. IEEE Xplore (2005)
15. Zadeh, L.A.: Fuzzy Logic, University of California, Barkeley. IEEE Xplore (1988)
16. Zadeh, L.A.: Fuzzy sets, Department of Electrical Engineering and Electronics Research
Laboratory, University of California, Berkeley (1965)
17. Ortiz Fernández, F., Fernández Diego, C., Santisteban Díaz, A., Delgado San Román, F.,
Ortiz Fernández, A.: Estimating the age of power transformers using the concentration of
furans in dielectric oil. In: International Conference on Renewable Energies and Power
Quality (ICREPQ 2016) (2016)
 Optical Partial Discharge Measurement
with Integrated Optical Fibers
as Sensing Element

Inna Kübler and Daniel Pepper(&)

Beuth University of Applied Sciences Berlin, Luxemburger Straße 10,

13353 Berlin, Germany
Daniel.pepper@beuth-hochschule.de

Abstract. A reliable assessment of the current condition and preferably the

remaining lifetime of an electrical equipment of high voltage cable systems is
playing an important part in achieving a sustainable development and upgrading
of the power grid. Partial Discharge (PD) measurements and diagnosis are
considered as a valid tool for evaluating the condition especially of cable sys-
tems. The recently discussed optical measurement of PD in high voltage cable
accessories using integrated fiber optical sensors has significant advantages over
the electrical PD measurement according to IEC 60270 since it is completely
insensitive to any electromagnetic interference. Non-electrical methods are an
interesting alternative, especially for on-site testing. This article describes the
on-going work on the integration of special fluorescent optical fibers made of
silicone elastomer (E POF) into a test specimen with a tip-plate configuration
and the optical detection of PD using these optical sensors. The used insulating
material is a transparent silicone elastomer whose optical properties have also
been investigated. Measurements are taken in a darkened cabin on a PD mea-
suring circuit according to IEC 60270 using an optical converter (APD).
Beforehand, the adaptation of the sensors and the APD was examined to confirm
the functionality of the system.

Keywords: Optical PD detection
Optical converter
Monitoring

1 Introduction

The safety and reliability of energy networks are important parameters that must be
maintained at the lowest possible cost. In this sense monitoring of the equipment can
contribute to a more efficient use of the service lifetime of equipment. The partial
discharge measurement (PD measurement) has proven to be a valid method to detect
damage in the insulation system. For equipment which are either to be supervised by
using monitoring systems or tested on-site, severe external electromagnetic interference
makes a sensitive PD measurement according to IEC 60270 hardly possible. However,
if the PDs are detected optically, the electromagnetic interference no longer affects the
sensitivity of the measurement. The necessary prerequisites, conditions and procedural
methods are being developed within the scope of the joint research project “Monalisa”
by the participating partners Federal Institute for Materials Research (BAM), Beuth

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 937–948, 2020.
https://doi.org/10.1007/978-3-030-31676-1_88
938 I. Kübler and D. Pepper

University of Applied Sciences Berlin (BeuthHS), University of Applied Sciences

(HTW), Technical University of Berlin (TUB) together with the industrial partners
50 Hz Transmission GmbH (50 Hz) and IPH Institute “Prüffeld für elektrische
Hochleistungstechnik” GmbH.
One major requirement for the detection of the optical signals of a PD is the
transparency of the insulating material. Furthermore, sensors must be integrated which
can detect the optical signals of the PD on the one hand and on the other hand do not
adversely affect the dielectric properties of the insulating system. These requirements
are fulfilled by transparent silicone elastomer in which optical fibers of the same
material are embedded as sensors. Thus insulating material and sensors are equally
elastic, which is essential for the installation of the accessories of energy cables. The
optical signals are processed by converters, which provide electrical impulses at their
output which directly correspond to the optical input and thus enable connection to
conventional PD measurement technology. Due to the low light intensities of the
optical PD signals, only photon counters and avalanche photodiodes can currently be
used as optical-electrical converters. The optical fibers are doped with a fluorescent dye
in order to shift the spectra of the optical PD and adjust it to the spectral sensitivity of
the converter [1]. Within this publication the operating environment and the sensitivity
of the sensors are examined using exemplary PD sources. First, the optical transmission
and detection properties of the transparent silicone elastomer and the E–POF are
examined. Furthermore, in order to determine the sensitivity, the optical PD signals of
an internal defect of a tip-plate configuration are measured using different sensor
arrangements and compared to the results of a conventional electrical PD-measurement
according to IEC 60270.

2 Functional Validation of the Optical PD Measuring System

When a PD occurs very small amounts of light are emitted. Capturing these signals
requires sensitive optical sensors and a converter, which translates the optical signals in
an adequate way for the subsequent evaluation technique. In the present work,
fluorescent optical fibers made of silicone elastomer (E–POF) were used as sensors
together with an optical-electrical converter (APD) based on an avalanche photodiode.
These two components are briefly described below. First, however, the optical spec-
trum of an arcing in air is measured and presented.

2.1 Optical Spectrum of an Arcing in Air

Using a tip-sphere configuration with a distance of 5 mm and the spectrometer LR2–T
from Lasertack the spectrum of the emitted light was recorded. For this purpose, the
optical fiber Thorlabs FT 030-blue was mounted on the sphere electrode in a way that it
is aligned with the discharge channel, as shown in Fig. 1. The voltage was increased to
an extent that stable discharges between the electrodes were observed. The discharge
current was limited by a resistance of 10 MX. The material of both electrodes is
stainless steel.
 Optical Partial Discharge Measurement with Integrated Optical Fibers 939

Fig. 1. Setup of a discharge channel in air

Figure 2 shows the recorded spectrum of the arcing in air. It contains mainly high
energy components in the visible blue and UV range. Assuming that inner PDs in
insulating materials have a similar spectrum suitable sensors should have a high sen-
sitivity in this range.

Fig. 2. Optical spectrum of a discharge in air

2.2 Optical Converter

The converter (APD) used converts optical signals into proportional electrical impul-
ses. The advantage over the use of photon counters is that the information of the light
intensity is taken into account during conversion. It is included in the conversion in
such a way that the subsequent PD-measuring systems can interpret this as a charge [2].
This means that defects can not only be recognized as such, but can also be identified in
their nature, for example via the phase-resolved patterns. One drawback is that the
940 I. Kübler and D. Pepper

sensitivity of the APD is lower than that of photon counters. Due to the spectral
sensitivity of the APD, which is in the visible range of higher wavelengths, signals with
wavelengths below 500 nm can hardly be detected. For this reason, a frequency-shift of
the light emitted by the PDs in a spectral range detectable by the APD is necessary.

2.3 Frequency Shift due to Fluorescence

With regard to the use of optical PD-measurement in cable accessories, a fluorescent
fiber made of silicone elastomer – a so-called E–POF (elastomeric polymer optical
fiber) – has been developed [1]. It is intended to enable non-reactive integration into
the insulating material. The main advantage is that the E–POF is flexible enough to
withstand elongation on the cable accessory without damage. As a result, the proven
assembly technique of sliding the accessories onto the cable ends can be maintained.
The E–POF has a diameter of around 2 mm with a few micrometers thick cladding
layer. The light is transmitted along the fiber length by total reflection at the transition
from core to cladding. In addition, incident light can be coupled into the fiber laterally
and forwarded therein. The fluorescence causes photons in the core of the fiber to be
multiplied, thus amplifying the input signal. This also leads to a shift of the input
spectrum according to the doped dye. This shift of the light spectrum is essential for the
optical PD-detection, since the spectra of the optical signals can be adapted to the
receiving devices. In this study E–POFs were used, which are doped to represent a
received signal in the green spectral range. The maximum of the emission spectrum is
about 520 nm. The absorption spectrum ranges according to the manufacturer from
360 nm to 480 nm with a maximum at 425 nm and is thus optimally suitable for the
detection of PDs. Using a broadband light source and a spectrometer the spectral
sensitivity as well as the frequency shift of the used E–POF could be detected. For this
purpose, the fiber was centered and fixed laterally at the output of the light source and
the spectrum was measured at the other end of the fiber as shown in Fig. 3. Light in the
visible spectrum was selected at the light source superimposed with the switchable
ultraviolet (UV)-source as input signal. The power of the different signals in the
spectral range of 200–1000 nm was measured with the spectrometer. Table 1 lists the
power for the input signals and the output signals generated by the E–POF. The input
signals were measured beforehand using the optical fiber Thorlabs FT 030-blue, which
was connected directly between the source and the spectrometer. The UV range causes
a significant increase in the power of the output signal of the E–POF. Although the
power of the input signal only increases by a factor of 1.46 if the UV component is
added, the power of the shifted signal is higher by a factor of 6.29. Figures 4 and 5
show the spectra of the input signals as well as the frequency-shifted output spectrum
of the E–POF.
 Optical Partial Discharge Measurement with Integrated Optical Fibers 941

Fig. 3. Setup for measuring frequency shift

Table 1. Power density of input and output of the E–POF

Power density [µW/cm2]
Spectral range Vis UV Vis + UV Vis + UV/UV
Input 1,95  10 8,23  10 2,85  105 1,46
5 4

Output 1,46 7,8 9,19 6,29

Fig. 4. Optical Spectrum of the light source

Fig. 5. Optical spectrum at the output of the E–POF

942 I. Kübler and D. Pepper

2.4 Attenuation of the E-POF

The attenuation of the sensor fibers is determined by a non-destructive side-
illumination fluorescence method [3]. A setup as shown in Fig. 3 is used. The light
source is moved along the length of the E–POF as the power density of the output
signal is measured. A 175 cm long E–POF is used and the measurement is performed
twice using each end as output. Figure 6 shows the obtained signals for different
distances. Using an exponential fitting of the data the overall attenuation can be cal-
culated as shown in Fig. 7 and is around 2,1 dB/m. The deviation of both measure-
ments can be explained by a possibly inhomogeneous distribution of the doped dye as
well as a varying diameter of the fibers. This appears plausible since the E–POFs are
prototypes and manufactured manually (Table 2).

Fig. 6. Intensity of the output signal according to wavelength and distance from light source

2.5
Power Density [μW/cm2]

UV+Vis 1
y = 2.01e-0.41x
2 UV+Vis 2
R² = 0.96

1.5

1
y = 2.02e-0.57x
R² = 0.92
0.5
0 0.5 1 1.5 2
Distance [m]

Fig. 7. Exponential fitting of the data

 Optical Partial Discharge Measurement with Integrated Optical Fibers 943

Table 2. Calculation of the attenuation of the E–POFs

Measurement Fitting Coefficient of determination Attenuation [dB/m]
1 2 * e−0,41x 0,96 1,8
2 2 * e−0,57x 0,92 2,5

2.5 Optical Properties of the Transparent Silicone Elastomer

The optical PD-measurement within an insulating material requires that the light sig-
nals can propagate through the material. For this a transparent insulating material is
necessary, which is not common nowadays [4]. In this work, a highly transparent,
addition-curing silicone elastomer (type Lumisil 7601) from Wacker Chemie AG was
used, which is designed for optical applications and therefore has excellent optical
properties. To determine the spectral attenuation of the silicone elastomer a test piece
with a thickness of 33 mm was produced. This fitted exactly into the cuvette holder
enclosed with the spectrometer, which was connected to the light source and the
spectrometer with commercially available optical fibers, see Fig. 8 inset.
The transmission spectrum was acquired so that the optical losses of the insulating
material are known. The transmission spectrum was measured using a broad band light
source and a spectrometer.

Fig. 8. Transmission spectrum of the silicone elastomer (red) and of air (blue), the darker lines
represent the moving average of the measuring data; measuring setup (inset)

That way the measurement of the silicone object was compared with the trans-
mission in air. Figure 8 shows the results of the measurement normalized to the
transmission in air. The transmission of the silicone elastomer proceeds very smoothly
from a wavelength of about 600 nm to lower frequencies. Below 600 nm and espe-
cially in the blue and UV range, the spectrum shows a frequency dependent attenua-
tion. The maximal value of the attenuation is about 30%.
944 I. Kübler and D. Pepper

3 Sensitivity of the Optical Measuring System

In order to determine the sensitivity of the E–POF in connection with the APD, the
optical PD measurement is compared with an electrical measurement according to
IEC 60270 taking place simultaneously. In order to enforce partial discharges within a
silicone volume a tip-plate arrangement is used representing a typical and severe defect
in the insulating material.
The measurements are carried out with different arrangements of the E–POF. First
an optimal arrangement is chosen in order to detect the maximum sensitivity. Further a
more realistic layout is measured using a spirally wound E–POF within the silicone
elastomer. The different arrangements are shown in Fig. 9 and the dimensions are given
in Table 3.

Fig. 9. E–POF arrangement inside the test specimen

Table 3. Dimensions of the test specimens with different sensor arrangements

Arrangement d[mm] D[mm] r[µm] S[mm] E-POF[mm]
(a) 18 30 1 20 70
(b) 18 45 1 20 500

3.1 Production of the Test Specimen

All test specimens are produced in the in-house laboratory. Stainless Steel needles are
used as high voltage electrodes. In order to ensure little variance within the group of
specimens the radius of the tip of each needle is measured with a microscope at three
shaft rotating angles. In doing so unemployable needles because of e. g. too large tip
radius, flattened or unsymmetrical tips are excluded from the experiment. For each
specimen a new needle is used. A brass plate is used as ground electrode. The two
component silicone rubber Lumisil LR7601/50 from Wacker Chemie AG is used to
fabricate the specimens. The two components of the raw silicone are homogeneously
mixed together, then the air which entered during this process is extracted by applying
low pressure (100 mbar) in a vacuum chamber followed by the vulcanization.
 Optical Partial Discharge Measurement with Integrated Optical Fibers 945

3.2 Optical Measurement of an Internal Defect

Figure 10 shows the PD measuring circuit according to IEC 60270. The detected
optical signals are fed through a conventional clear optical fiber into the APD
converter. In the optical channel the filter settings of the PD-measuring system are:
Df = 160 kHz and fCenter = 160 kHz. The converted light signals are processed by the
PD measuring system as electrical impulses and presented in the software as electrical
charge with the unit pC. However, it should be noted that the optical channel cannot be
calibrated. Therefore the optical signals are labeled pC* within this study. The mea-
surement is started at 2 kV and increased by 1 kV every minute until PDs set in. The
noise level for both the electrical and optical channel was 1,4 pC/pC*.

Fig. 10. PD measuring circuit with electrical and optical channel

3.3 Results of the Optical and Electrical Measurement

Figure 11 shows specimen P-a1 where the needle tip is brought as close as possible to
the E–POF. In this case the highest sensitivity is expected. The measurement showed,
that PDs with around 3 pC* could be detected optically at a PD inception voltage of
9 kV while they were hardly visible in the electrical channel (2 pC). Increasing the
voltage further results in higher electrical and optical PDs. While the electrical dis-
charges rise noticeably to 31 pC when increasing the voltage level to 13 kV, the optical
discharges increase only to 12 pC*. At higher voltage stress the discharges increase
steadily, but it becomes clear, that the measured values of the optical channel remain
smaller than the electrical discharge values. At 17 kV electrical discharges with 100 pC
are detected, while they are at about 50 pC* in the optical measurement. Thus the
factor between the optical and the electrical channel is not constant within the mea-
surement but depends on the state of degradation of the insulating material (Table 4).
946 I. Kübler and D. Pepper

Fig. 11. Specimen P-a1 showing the optical PD signals in immediate proximity of the sensor E–
POF (please refer to Fig. 9)

Table 4. Measured electrical and optical signals and the corresponding optical PD noise level
(COPDNL) of the noise level of 1,4 pC* of Specimen P-a1
Voltage level [kV] Electrical [pC] Optical [pC*] Factor (opt/el) COPDNL [pC]
9 2 3,5 1,75 0,8
13 31 12 0,4 3,6
14 58 30 0,5 2,7
15 56 31 0,6 2,5
16 80 50 0,6 2,2
17 100 47 0,5 3

In specimen P-a2 the needle tip was placed around 1 mm away from the E–POF.
Additionally, the needle was decentered as shown in Fig. 12. In this case the optical
signals had to propagate through the insulating material before they were detected by
the sensor. The results in Table 5 show a significantly reduced sensitivity for this
measurement.

Fig. 12. Specimen P-a2 needle decentered and at a distance of 1 mm of the E–POF

Table 5. Measured electrical and optical signals and the COPDNL of the noise level of 1,4 pC*
of Specimen P-a2
Voltage level [kV] Electrical [pC] Optical [pC*] Factor (opt/el) COPDNL [pC]
13 10 <1,4 – –
14 30 2 0,07 21
15 50 4 0,08 17,5
16 80 10 0,13 11,2
17 100 10 0,1 14
 Optical Partial Discharge Measurement with Integrated Optical Fibers 947

For arrangement (b) the E–POF is spirally wound at a distance of about 2,5 cm
around the needle tip and embedded in one step to avoid interphase layers. For this
purpose, Teflon screws are inserted into the walls of the molds so the silicone fiber is
held in place during manufacturing. About 50 cm of an E–POF are embedded inside
the silicone volume. The optical and electrical PDs set in at 22 kV with 93 pC
respectively 2,8 pC*. This results in a factor of 0,03 and thus a corresponding optical
PD noise level of 46 pC. The corresponding optical PD noise level is calculated by
multiplying the measured optical noise level (1,4 pC*) with the inverse of the calcu-
lated factor of the optical and the electrical signals.
Increasing the voltage stress leads to a growing electrical tree and thus higher
electrical PD charges. Figure 13 shows the electrical and optical discharges at a con-
stant voltage stress of 30 kV. It can be seen, that the optical signals do not develop in
the same way. While the optical and the electrical signals behave similarly in the
beginning, the optical emissions do not increase along with the electrical discharges.
Instead they seem to remain almost constant at a level of 50 pC*. At a maximum the
factor between optical and electrical channel is 0,3 corresponding to an optical PD
noise level of 5 pC. Please note that the total values of the electrical and the optical
channel cannot be compared to each other because the unit pC* is not calibrated, as
mentioned above.

Fig. 13. Electrical (blue) and optical (red) discharges as a function of the voltage stress duration
at 30 kV

4 Conclusions

It has been shown that the spectral sensitivity of the E–POF is well suited for the
detection of the optical signals of PDs. In addition, the detected signals are shifted in
the frequency spectrum in order to adapt them to the receiving avalanche photodiode of
the APD.
948 I. Kübler and D. Pepper

A very high sensitivity can be reached for an optimal arrangement where the needle
tip is placed very close to the E–POF. However the sensitivity decreases significantly
as soon as the position of the E–POF departs from the optimal arrangement. These
findings are valid for very small PD defects like electrical trees that are just igniting at
electrodes with a radius of 1 lm up to 3 lm. In the case of a realistic arrangement with
a spirally wound E–POF the electrical and optical signals do not behave in the same
way, possibly depending on the length of the growing electrical tree.
However, it is also conceivable that the charge of the PD and the associated light
emission is not constant but changes with the degree of degradation of the insulating
material. It seems possible that the optical emissions and the charge of PDs are not
developing in the same way. This has to be investigated in more detail. It is also
possible that the optical spectrum of the discharges behaves differently within a cavity
of the insulating material than it does in air.

5 Outlook

Investigations with integrated E–POFs of different lengths and different arrangements

have to follow in order to determine the influence of the sensor position. Also the
relationship between the light emissions and the electrical discharges at PD inside the
insulating material needs to be clarified.

Acknowledgments. The authors thank the German federal ministry of education and sciences
for funding.

References
1. Siebler, D., Hoberg, M., Rohwetter, P.: A new generation of all-elastomeric fluorescent POF
for sensor applications. In: The 24th International Conference on Plastic Optical Fibers,
Nuremberg, Germany, 22–24 September 2015
2. Kübler, I., Pepper, D.: A novel converter for optical PD measurement with output as electrical
charge proportional to light intensity. In: The 20th International Symposium on High Voltage
Engineering, Buenos Aires, Argentina, 27 August–01 September 2017
3. Kruhlak, R.J., Kuzyk, M.G.: Side-illumination fluorescence spectroscopy. II. Applications to
squaraine-dye-doped polymer optical fibers. J. Opt. Soc. Am. B 16(10), 1756–1767 (1999)
4. Lothongkam, C., Rohwetter, P., Habel, W., Gockenbach, E.: Dielectric strength behavior and
mechanical properties of transparent silicone rubbers for HV cable. In: The 7th International
Symposium on Electrical Insulating Materials, Niigata City, Japan, 1–5 June 2014
 Condition Monitoring of High Voltage
Press-Pack IGBT with On-State
Collector-Emitter Voltage

Cao Zhan(&), Lingyu Zhu, Chenshuo Liu, Liang Pan, Jiangyu Liu,
and Shengchang Ji

Xi’an Jiaotong University, Xi’an 710049, China

zhancao@stu.xjtu.edu.cn

Abstract. Insulated gate bipolar transistor (IGBT) is the critical component in

Module multilevel converter (MMC) submodule and is aged under operating
condition. As the voltage rating of MMC-HVDC power transmission continues
raising, high voltage press-pack IGBT (PPI) will be used widely. Condition
monitoring of high voltage PPI is very important to maintain the reliable
operation of MMC-HVDC. The purpose of this paper is to use on-state
collector-emitter voltage during power cycling experiment as degradation
indicator of high voltage PPI. On-state collector-emitter voltage is largely
influenced by collector current and junction temperature. The impact of collector
current can be ignored in this experiment because the current remains 350A.
Junction temperature is very difficult to be directly measured, but gate internal
resistance can indirectly reflect the junction temperature. Gate internal resistance
can be calculated by peak gate current obtained together with on-state collector-
emitter voltage during power cycling experiment. The on-state collector-emitter
was measured under different junction temperature when the PPIs are under the
same degradation degree. The results show that there is a linear relationship
between on-state collector-emitter voltage and junction temperature. The linear
relationship is used to strip the influence of junction temperature on on-state
collector-emitter voltage obtained during power cycling experiment. After
stripping the influence of junction temperature, the results show that the on-state
collector-emitter voltages of test PPI samples all increase with the degradation
continuing. On-state collector-emitter voltage can be utilized to be degradation
indicator of high voltage PPIs.

Keywords: Power cycling
High voltage PPI
On-state collector-emitter

voltage
Gate internal resistance
Junction temperature

1 Introduction

Modular multilevel converter-high voltage direct current (MMC-HVDC) is very useful

in distributed generation grid-connection, the interconnection of regional grids, the
multi-terminal direct current (MTDC) transmission and so on [1, 2]. However, with the
increasing of operation period, the submodules of MMC are gradually deteriorating.
IGBT is the key component in MMC submodule. As the voltage rating of MMC-

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 949–957, 2020.
https://doi.org/10.1007/978-3-030-31676-1_89
950 C. Zhan et al.

HVDC power transmission continues raising, high voltage press-pack IGBT (PPI) are
widely used due to its endurance to high power and high voltage, double side cooling
as well as short-circuit failure mode [3, 4]. It is important to monitor and assess the
condition of IGBT to ensure the healthy operation of MMC.
Since degradation has a considerable impact on the electrical characteristics of
IGBTs, many references focus on the technology utilizing electrical parameters to
realize the condition monitoring of power module IGBTs [5]. On-state voltage VCE,ON
is widely used as the indicator of package-related bond wire failure. In [6], Smet et al.
performed five different high temperature power cycling experiments on nineteen
power module IGBTs to study the aging mechanism and condition monitor method of
IGBTs. It is found that VCE,ON is a good indicator of bond wire aging since it shows
various increasing trend under different power cycling experiment. In [7], Ui-Min Choi
et al. studied the junction temperature estimation method using VCE,ON at high current.
The result shows that comparing the estimation junction temperature and VCE,ON, the
dramatically increase of VCE,ON is not due to the solder joint fatigue but due to the
bond-wire fatigue in this test condition. Actually, VCE,ON is not only affected by IGBT
degradation, but also has an important relationship with junction temperature. So it is
necessary to separate the effect of junction temperature when condition monitor method
based on VCE,ON is utilized [8]. Most of the studies only focus on the technology of
monitoring wire-bonded IGBT, while few research is conducted on the condition
monitoring technology of high voltage PPI.
In this paper, power cycling experiment is performed to simulate accelerated aging
on PPIs. During power cycling experiment, on-state collector-emitter voltage and gate
current are both measured during the experiment. X-ray computed tomography (CT) and
scanning electron microscope (SEM) experiment are then performed on an obviously
failed PPI to study the physical failure mechanism caused by the deterioration.

2 Power Cycling Experiment

Power cycling test is utilized here to realize accelerated degradation by a thermal

cycling load platform shown as Fig. 1. This platform mainly consist of the device
under test (DUT), the clamping pressure device which can adjust the pressure on each
DUT, the drive circuit which provides gate voltage and the water cooled plate which is
connected with water cooler.
Constant heating current flowing from collector to emitter produces thermal power
loss. Once the junction temperature reaches upper limited value, drive circuit of the
control gate will output −5 V turn off signal to block PPI. Meanwhile, double side
cooling system tunes on to cool PPI. Then, drive circuit of the control gate output
+15 V turn on signal to switch PPI. Thermal stress on the device under test (DUT) will
eventually cause degradation.
There are totally six DUTs in power cycling experiment to suffer over 100 k cycles
until one of them is obviously failed. Actually, the aging degree of these DUTs are
quite different due to the difference of temperature swing between the water plates
above and below the DUT. From the top to the bottom of the DUTs placed in the
platform are named #1 to #6 in this paper. #2 PPI is failed even judging from the
 Condition Monitoring of High Voltage Press-Pack IGBT 951

appearance of the DUT since the collector package is swelling, shown in Fig. 2. #5 PPI
is judged as abnormal because the voltage between collector and emitter is much lower.

Fig. 1. Power cycling platform

Fig. 2. Failed PPIs

The stack model of IGBT is shown as Fig. 3. From the collector to emitter, is top
molybdenum plate, chip plate, bottom molybdenum plate and silver plate respectively.
Molybdenum plates that include upper and lower molybdenum plate has high hardness
property, so the double side of chip is balanced to prevent chip from damaging by
uneven pressure.

Top molybdenum plate

Chip plate
Bottom molybdenum plate
Silver plate

Fig. 3. Stack model

952 C. Zhan et al.

3 On-State Collector-Emitter Voltage of PPI

The value of on-state collector emitter voltage VCE and gate current Ig in the aging
process of the PPIs are recorded every 1000 cycles. During the data processing of VCE,
the voltage waveform during the turn-on process from 1 s to 18 s is shown as Fig. 4,
the curve is fitted by cubic spline curve method. VCE shows an upward trend since the
change of junction temperature and mechanical pressure during the heating phase
affects the value of the turn on resistance of PPI. When the current is constant, the
voltage drop between collector and emitter increases eventually. During the data
processing, the average temperature within two second of VCE in different aging
degrees is selected for comparison.

Fig. 4. Waveform of VCE

The junction temperature and the current both influence on-state collector emitter
voltage. During the experiment, the current maintains at 350 A, so the influence of the
current can be ignored. However, the temperature of the cooling water has a variation
range of about 5 °C, so the case temperature of PPI fluctuates continuously during the
aging process, which causes the fluctuation of the junction temperature.
In order to strip the influence the junction temperature on the VCE, it is necessary to
obtain the junction temperature during each power cycle. At present, the accurate
measurement of the junction temperature of PPI is the critical problem of the research.
This paper uses the indirect measurement method to obtain the junction temperature
during the power cycle, which is also known as temperature based sensitive electrical
parameter method. It is pointed out in the literature [9] that there is a positive corre-
lation between the junction temperature and the gate resistance of IGBT and the
internal resistance of the gate can be calculated by the peak current of the gate during
each cycle. Though measuring the gate current during the dynamic operation of IGBT,
the junction temperature can be obtained. Based on the power cycling test platform, the
on-state collector emitter voltage and the gate peak current are obtained under the same
current and clamping pressure. Figure 5 shows the linear relationship between the on
state collector emitter voltage and the internal resistance.
 Condition Monitoring of High Voltage Press-Pack IGBT 953

Fig. 5. The relationship between VCE and gate internal resistance.

The equation of the fitted curve shown in Fig. 5 is as follow:

VCE ¼ 0:0395  R þ 2:6439 ð1Þ

Where: VCE = Voltage in Volt (V), R = Resistance in Ohm (X).

Fig. 6. The relationship between VCE and aging degree of #2 PPI.

The correlation coefficient is −0.9842 which shows a highly correlation. The linear
relationship between the VCE influenced by junction temperature and the junction
temperature is utilized to normalize the result of VCE influenced by both degradation
and junction temperature. In this way, the relationship between VCE and the degrada-
tion of PPI is obtained. From Fig. 6, it is apparent that after striping the influence of
junction temperature, VCE shows an increased trend during 100 K cycling, but it still
shows fluctuation with the change of aging degree. One reason is that mechanical
pressure on PPI has important influence on VCE and mechanical pressure changes
largely with the temperature distribution inside PPI. Another reason is that obtained
954 C. Zhan et al.

normalized formula used to strip the influence of junction temperature has certain error
since the data points utilized for the fitted curve are not enough. Further experiment is
necessary to improve this method. From the present result, VCE measured in power
cycling experiment has a variation range of less than 0.2 V and a rate of change is
7.7%, which shows that VCE can be used as an indicator for condition monitoring of
PPI.

4 Physical Failure Mechanism

Physical failure mechanism is of significance to promote condition monitoring tech-

nique. In order to detect the structural abnormality caused by thermal degradation of #2
PPI before the disassembly, 450 kV X-ray computed tomography (CT) imaging
equipment is used to detect the structural change of every stack inside PPI package.

(1) Failed PPI (2) Healthy PPI

Fig. 7. Longitudinal tomography results.

Figure 7 shows the longitudinal tomography results of #2 PPI. There are two stacks
separating from the baseplate totally. It indicates that this IGBT suffers large thermal
and mechanical stress during power cycling test and eventually caused clamping
pressure lose. It’s illustrated in literature [10] the temperature distribution on chip face
is uneven: the inner surface temperature is higher than outer surface. Under the effect of
uneven thermal expansion, the stack is expanding in lateral direction and bending apart
from each other. The separated area of the chips is partial at heating phase, but once
heating over and power cycle turns into cooling phase, the separated area of chip
becomes large since chip shrinks faster than other parts. These analyses can explain the
longitudinal tomography result reasonably.
 Condition Monitoring of High Voltage Press-Pack IGBT 955

Fig. 8. Disassembly result of failed PPI.

After X-ray CT experiment, the Failed PPI is disassembled, as shown in Fig. 8. It

shows that the chip plate has totally ablated during power cycling. Since molybdenum
plate is very close to chip, there is a certain ablative impurity on the surface. What’s
more, the insulation encapsulation of the stack is carbonized apparently due to high
temperature caused by overheating. Poly-ether-ether-ketone (PEEK) is the material of
insulation encapsulation of the stack which is carbonized due to high temperature in
this case.
In order to study the failure mechanism of DUT after power cycling furthermore,
scanning electron microscope (SEM) is employed to investigate the surface charac-
teristics of chip plate, the result shown in Fig. 9. Obvious cracks appear on the surface
of chip. As mentioned before, the pressure is one of the most important factors causing
failure of chip for PPI. Coefficient of thermal expansions (CTE) of different materials
lead to micro friction on the surface of the chip plate. When junction temperature
increases largely due to power cycling, the mechanical stress caused by thermal
expansion becomes large enough to cause crack.
The above tests prove that the main failure mode of PPI under power cycle test is
the uneven pressure distribution caused by excessive local junction temperature, which
results in local deformation of PPI. On one hand, the change of the structure lead to the
increase of contact resistance and thus lead to the increase of VCE. On the other hand,
the increase of contact resistance can lead to large loss, which causes ablation and leads
to the failure of the device.
956 C. Zhan et al.

Fig. 9. The scanning result of chip surface.

5 Conclusion

In this paper, a power cycling platform for high voltage PPI is built. Combined with the
positive correlation between the gate internal resistance and junction temperature, the
influence of junction temperature on the VCE is stripped and the evolution relationship
between the VCE and aging degree is obtained. It is found that the VCE shows an upward
trend during the process of thermal degree, which can be utilized in monitoring the
condition of PPI in MMC. The disassembly test and SEM test can explain the
increasing reason of VCE.
In order to realize the condition monitoring method of high voltage press-pack
IGBT with on state collector voltage, it is necessary to study the accurate measurement
method of junction temperature and mechanical pressure. What’s more, the finite
element model of PPI should be built to study the physical failure mechanism of the
device under thermal degradation furthermore.

References
1. Deng, E., Zhao, Z., Lin, Z., Han, R., Huang, Y.: Influence of temperature on the pressure
distribution within press pack IGBTs. IEEE Trans. Power Electr. 33(7), 6048–6059 (2018)
2. Shengqi, Z., Luowei, Z., Pengju, S.: Monitoring potential defects in an IGBT module based
on dynamic changes of the gate current. IEEE Trans. Power Electron. 28(3), 1479–1487
(2013)
3. Yang, S., Xiang, D., Bryant, A., Mawby, P., Ran, L., Tavner, P.: Condition monitoring for
device reliability in power electronic converters: a review. IEEE Trans. Power Electron. 25
(11), 2734–2752 (2010)
4. Gunturi, S., Schneider, D.: On the operation of a press pack IGBT module under short circuit
conditions. IEEE Trans. Adv. Packag. 29(3), 433–440 (2006)
5. Liu, J., Zhang, G., Chen, Q., Qi, L., Geng, Y., Wang, J.: In situ condition monitoring of
IGBTs based on the miller plateau duration. IEEE Trans. Power Electron. 34(1), 769–782
(2019)
6. Smet, V., et al.: Ageing and Failure Modes of IGBT Modules in High-Temperature Power
Cycling. IEEE Trans. Industr. Electron. 58(10), 4931–4941 (2011)
 Condition Monitoring of High Voltage Press-Pack IGBT 957

7. Oh, H., Han, B., McCluskey, P., Han, C., Youn, B.D.: Physics-of-failure, condition
monitoring, and prognostics of insulated gate bipolar transistor modules: a review. IEEE
Trans. Power Electron. 30(5), 2413–2426 (2015)
8. Singh, A., Anurag, A., Anand, S.: Evaluation of Vce at inflection point for monitoring bond
wire degradation in discrete packaged IGBTs. IEEE Trans. Power Electron. 32(4), 2481–
2484 (2017)
9. Baker, N., Munknielsen, S., Iannuzzo, F., Iannuzzo, F.: IGBT junction temperature
measurement via peak gate current. IEEE Trans. Power Electron. 13(5), 3784–3793 (2016)
10. Poller, T., Basler, T., Hernes, M., D’Arco, S., Lutz, J.: Mechanical analysis of press-pack
IGBTs. Microelectron. Reliab. 52(10), 2397–2402 (2012)
 Influence on Formation of Methanol
and Ethanol in Oil-Filled Transformers Under
Accelerated Thermal Aging

K. Homeier1(&), M. Imani1, T. Münster1, T. Kinkeldey1, P. Werle1,

and D. Philipp2
1
Division of High Voltage Engineering and Asset Management,
Leibniz Universität Hannover, Institute of Electric Power Systems,
Schering-Institute, Hannover, Germany
homeier@ifes.uni-hannover.de
2
Analysen-Service GmbH, Environmental and Oil Laboratory,
Leipzig, Germany

Abstract. This contribution investigates the suitability of short-chain alcohols

for estimating the condition of a cellulose insulation in power transformers. To
model an accelerated thermal aging of a transformer, strips of insulation paper
impregnated with different insulation liquids were kept in a glass vessel, filled
with the dielectric liquid and catalyzer inclusions, in a heating chamber. After oil
sampling at predefined intervals, the concentrations of furanic compounds, the
degree of polymerization (DP) value and the concentrations of methanol and
ethanol in the insulation liquid were measured. Three different types of insu-
lation liquids were utilized, such as the well-known uninhibited and inhibited
mineral oil as well as a new inhibited insulation liquid with different oil com-
ponents, inhibitor content and manufacturing process. These insulation liquids
were investigated to obtain their various aging behavior regarding the generation
of 2FAL and short-chain alcohols by thermal accelerated stress.

Keywords: Methanol
Ethanol
Thermal aging
Cellulose deterioration
DP

value

1 Introduction

Insulation paper contains about 90% of cellulose, 6–7% of hemicellulose, and 3–4% of
lignin [1]. Cellulose is built out of molecules of glucose, which are ring structured and
are bonded via oxygen molecule under elimination of water by polycondensation. The
number of glucose rings is defined by degree of polymerization (DP) value. Due to an
increase in temperature, increasing moisture, presence of water and mechanical stress
the long chains of cellulose molecule are broken by three different aging mechanisms:
pyrolysis (thermal), oxidation and hydrolysis (chemical decomposition). At very high
temperatures the aging of cellulose is mainly determined by the pyrolysis. By
decreasing of the DP value from an initial value of about 1200 to 200, the mechanical
property of insulation paper reduces. For example short circuit faults in transformers
locally lead to these very high temperatures. In addition to breaking up the chains of

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 958–968, 2020.
https://doi.org/10.1007/978-3-030-31676-1_90
 Influence on Formation of Methanol and Ethanol 959

glucose molecules, different aging products are generated, e.g. carbonmonoxide, car-
bondioxide, water, furanic compounds and various acids, such as alcohols, ketones,
aldehydes. This aging products serve as catalyst for the continuous paper degradation
and partially diffuse in the surrounding insulation liquid [2–5].
To determine the lifespan of a transformer and to detect faults in an early stage,
aging indicators in the insulation liquids are required, which allow conclusions to be
drawn about the condition of the cellulose. For a long time, 2FAL were utilized for
determination of DP value of the insulation paper. However, various studies have been
shown that the concentration of 2FAL cannot easily be inferred from the DP by a
formula. For example, the concentration of 2FAL are influenced by the type of insu-
lation liquid and the presence of inhibitors. Furthermore, the construction of trans-
former, such as free-breathing or hermetically sealed transformers, leads to different
amount of 2FAL in the insulation liquids. In case of free-breathing transformers, the
2FAL can evaporate in the ambient air. Therefore, in this construction type the amount
of 2FAL is lower than in sealed transformers. Only the fact of a high amount of FAL
can signify the degradation of cellulose paper and can indicate a thermal fault of the
insulation paper in transformers [2–7]. Thus, methanol was recommended as a new
aging indicator of deterioration of the insulation paper [8]. There have been a number
of studies regarding to this topic [8–13].
Thus, the influence of the type of insulation liquid and the presence of inhibitors on
the generation and also degradation of short-chain alcohols, such as methanol and
ethanol, as well as of the furanic compound 2FAL for comparison were examined.
Moreover, the DP value of the insulation paper were measured. In this contribution, the
generation of this aging products in a hermetically sealed transformer were investigated
under accelerated thermal stress.

2 Experiment Description

To model the insulation system of an oil-filled power transformer, scaled glass vessels
filled with different insulation liquids combined with strips of insulation paper were
utilized. Three types of new insulation liquids were utilized: an uninhibited mineral oil
(oil A), an inhibited mineral oil (oil B) and an inhibited insulation liquid with different
oil components, inhibitor content and manufacturing process (oil C). The insulation
paper strips were dried in a vacuum chamber at a pressure of 0.1 mbar and a tem-
perature of 80 °C for 24 h. The dried samples were impregnated with the respective
insulation liquid, which was dried and degassed, for a duration time of 36 h. After
impregnation the test vessels were prepared with the impregnated cellulose paper, with
the dried and degassed insulation liquid and with a tube made out of copper as catalyst,
which are contained in a real transformer. The cellulose arrangement consists of 160 g
of 100 mm long cellulose strips with a thickness of 0.1 mm and a width of 25 mm in
combination with a 66 g copper tube, shown in Fig. 1. To obtain a weight ratio of 10:1,
1600 g insulation liquid were added.
960 K. Homeier et al.

Fig. 1. Scaled test vessel with paper strips and copper tube.

Reproducing a hermetically sealed transformer, the glass vessels were tightly

closed with a silicone rubber. Thus, the model of insulation system were kept without
contact to the atmosphere. Furthermore, one test vessel was used per sampling point, to
prevent the contact with air during the sampling and to maintain the weight ratio
between insulation paper and oil constant at 1:10 over the whole aging time.
These test vessels were aged under thermal accelerated stress at 130 °C in a heating
chamber for a duration time up to 2520 h (15 weeks) [14]. After sampling several
chemical properties, such as the concentration of 2FAL, methanol, ethanol and the DP
value as well as acid number, were measured.

3 Results and Discussion

Firstly the DP value of the insulation paper were measured based on the test standard
IEC 60450. In Fig. 2 the DP value over the aging time is depicted.

Fig. 2. DP value over the aging time.

 Influence on Formation of Methanol and Ethanol 961

3.1 Furanic Compounds

In a previous investigation the influence of the cellulose arrangement, once in wrapped
paper on the copper tube and once cut in paper strips, on the formation of furanic
compounds was examined [15]. The result of this investigation was that with a higher
contact surface between the insulation liquid and the insulation paper a higher amount
of furanic compounds in the insulation liquid was determined. This fact can be
explained with a higher generation of furanic compounds and a greater diffusion of the
furanic compounds out of the insulation paper in the insulation liquid. At the surface of
contact, various aging products can accelerate or initiate aging mechanisms of the other
insulating medium. Furthermore, the diffusion of chemical substances between the
insulation paper and insulation liquid needs a surface contact to determine a distri-
bution equilibrium. Considering the wrapped paper, the surface of the inner paper
layers has no direct contact to the surrounding oil. Thus, the generated 2FAL could be
retained in the wrapped paper and the amount of 2FAL in insulation liquid is much
lower to the striped papers. The previous investigation shows the possibility of diffu-
sion of 2FAL from paper into oil and the other way around. For this a defined amount
of 2FAL was added in an insulation liquid, which was kept in combination with
insulation paper in sealed vessels at 80 °C for two weeks. The result was a reduced
concentration of 2FAL [15].
In this contribution the concentration of 2FAL were measured with the measuring
method using liquid chromatography with a high performance liquid chromatograph
(HPLC) according to the IEC 61198 test standard. The concentration of 2FAL over the
aging time and DP value by accelerated thermal stress is shown in Figs. 3 and 4.

Fig. 3. Concentration of 2FAL over the aging time.

962 K. Homeier et al.

Fig. 4. Concentration of 2FAL over the DP value.

As expected, oil B and oil C, both filled with inhibitors, have got a lower amount of
2FAL over the whole aging time than oil A. Inhibitors slow down the aging process by
its own consumption. The amount of 2FAL in the insulation liquids increases generally
over the aging time and with decreasing of DP value. However, in addition to this
increase a decreasing trend is sometimes recognizable for oil A and oil C. This
decreasing trend might be explained by the abovementioned diffusion, by degradation
and by evaporation. A similar behavior of the concentration of 2FAL were determined
in the previous investigation using a model for a free-breathing transformer [15]. These
results also show that any interpretation, based on the measured concentration of 2FAL,
should be considered carefully. High concentrations of 2FAL could indicate a pro-
gressive thermal aging of the cellulose paper, but a direct calculation into an exact DP
value is not possible without consideration of other aging indicators.

3.2 Methanol
Figures 5 and 6 represent the different generation and degradation of concentration of
methanol in the various liquids over the aging time as well as over the DP value. The
concentration of methanol were measured with gas chromatography with head space
method.

Fig. 5. Concentration of methanol over the aging time.

 Influence on Formation of Methanol and Ethanol 963

Fig. 6. Amount of methanol over DP value.

There is a similar behavior of the three insulation liquids visible over the aging
time. All show decreasing trends in addition to the expected increasing trend, while the
maxima and minima of the concentration of methanol in the insulation liquids are at
similar aging times. In approximation a rather increasing behavior of the amount of
methanol over the aging time can be observed. The highest concentration of methanol
occurs in the inhibited oil C at about 2000 h. Due to the inhibitor, which interfere the
aging mechanism, a higher amount of methanol is recognizable in the uninhibited
mineral oil A than in the inhibited mineral oil B. Furthermore, plotted over the DP value
an up and down amount of methanol is also visible. At a specific DP value the
methanol concentration begins to increase significantly. For oil C, the rise in con-
centration can be seen at the lowest DP value, but with the highest gradient.
In order to recognize a relation between the concentration of furan and methanol,
the next three figures Figs. 7, 8 and 9 show both chemical aging products plotted in the
same graph for each oil.

Fig. 7. Concentration of methanol and 2FAL with the thermal accelerated uninhibited mineral
oil oil A.
964 K. Homeier et al.

Fig. 8. Concentration of methanol and 2FAL with the thermal accelerated inhibited mineral oil
oil B

Fig. 9. Concentration of methanol and 2FAL with the thermal accelerated inhibited insulation
liquid oil C.

Using oil A and oil C approximately the same trend is recognizable between the
amount of 2FAL and methanol. Only in case of the inhibited mineral oil there is no
relation visible. The reduction in the measured methanol concentration may be traced in
the possible diffusion of methanol from the insulation liquid into the insulation paper.
Since the test vessels are located in a closed system, evaporation should probably be
eliminated. Besides, because of the high temperature of 130 °C, degradation of those
compounds is also possible.

3.3 Ethanol
Figures 10 and 11 show the different curves of concentration of ethanol in the various
insulation liquids over the aging time as well as over the DP value. Similar to methanol,
the concentration of ethanol were determined with a gas chromatography with head
space method.
 Influence on Formation of Methanol and Ethanol 965

Fig. 10. Concentration of ethanol over the aging time.

Fig. 11. Concentration of ethanol over the DP value.

A similar behavior of the insulation liquids can be observed over the aging time and
also over the DP value. Approximately, both mineral oils show the same trend, shifted
with a small difference. As expected, the mineral oil without inhibitor possesses a
higher amount of ethanol than the inhibited mineral oil. However, the inhibited
oil C shows a constant behavior over the aging time with the lowest concentration of
ethanol. Similar to methanol an up and down concentration of ethanol is visible over
DP value. The amount of ethanol also begins to increase significantly at a specific DP
value, which is the lowest value using oil C.
In order to detect a relation between the concentration of furan and ethanol, the
following three figures Figs. 12, 13 and 14 show these chemical aging products for
each oil. Approximately there is the same trend between the concentration of ethanol
and 2FAL for all insulation liquids. Similar to the amount of 2FAL, the concentration
of ethanol possess an approximately rising trend over the aging time.
966 K. Homeier et al.

Fig. 12. Concentration of ethanol and 2FAL with the thermal accelerated uninhibited mineral
oil oil A.

Fig. 13. Concentration of ethanol and 2FAL with the thermal accelerated inhibited mineral oil
oil B.

Fig. 14. Concentration of ethanol and 2FAL with the thermal accelerated inhibited liquid oil C.

The reduction of ethanol concentration in sections can be explained comparatively

like methanol. Ethanol may be traced in the possible diffusion from the insulation liquid
into the insulation paper. Furthermore, an evaporation could probably be excluded by
the closed system model. Besides, a degradation of those compounds is also possible.
 Influence on Formation of Methanol and Ethanol 967

3.4 Acid Value

Matharage et al. investigated the formation and degradation of methanol in a pure
mineral oil without insulation paper under an accelerated thermal aging at 130 °C in a
closed system. Also without paper, a reduction in the methanol concentration can be
detected. The authors explained this fact by a possible esterification of the mineral oil,
if the acid content exceeds a threshold value. By this esterification the methanol could
be consumed [9].
Thus, in this contribution the acid number of the mineral oils were also measured
according to the IEC 62021-1, shown in Fig. 15 [14].

Fig. 15. Acid number of the different insulation liquids over the aging time.

Approximately a continuous increasing trend is recognizable. For methanol, the

important decreasing point of all insulation liquids is situated in a duration time
between about 900 h and 1300 h. In this duration time the acid number also possess for
all insulation liquids a decrease with a low gradient. However, after this duration time
the acid number continues to rise, while the amount of methanol also increases.

4 Conclusions

In this contribution, the influence of type of insulation liquid and the presence of
inhibitor were investigated under an accelerated thermal stress in a closed system. Both
parameters affect the formation of these aging products.
Various aging mechanisms superpose at this high temperature of 130 °C. Due to
the high initial concentration of long-chain glucose molecules of cellulose, and also of
hemicellulose and lignin, a lot of aging products, such as furanic compounds and short-
chain alcohols are generated. Although hemicellulose and lignin are contained to a
lower extent in insulation paper than cellulose, they are more susceptible to aging
processes. After consumption of the initial products, the curve should slowly be sat-
urated. However, various generated aging products, such as water and acids, catalyze
these aging mechanisms, so that the curve undergoes an additional increase.
968 K. Homeier et al.

Against this increasing trend, the possible diffusion of 2FAL, methanol and ethanol
from the insulation liquid back into the insulation paper as well as their decomposition
in possible secondary reactions counteract a decreasing trend. The diffusion of the
short-chain alcohols and 2FAL from the oil into the insulation paper and vice versa
requires time, so that the molecules can be distributed homogeneously depending on
the solubility of the paper for the respective aging product.
Thus, a determination of the DP value, based on the measured 2FAL or short-chain
alcohols, like methanol and ethanol, should be carefully treated. There are various
influences of the generation and also degradation of these aging products.

Acknowledgments. The authors would like to express their gratitude towards Weidmann
Electrical Technology AG and GRID INSPECT GmbH for their support of this research.

References
1. Emsley, A.M., Xiao, X., Heywood, R.J., Ali, M.: Degradation of cellulosic insulation in
power transformers. Part 2: formation of furan products in insulating oil. IEE Proc. Sci.
Measur. Technol. 147(3), 110–114 (2000)
2. Fofana, I.: 50 years in the development of insulating liquids. IEEE Electr. Insul. Mag. 29(5),
13–25 (2013)
3. Höhlein, I., Kachler, A.J.: Aging of cellulose at transformer service temperature. Part 2:
influence of moisture and temperature on degree of polymerization and formation of furanic
compounds in free-breathing systems. IEEE Electr. Insul. Mag. 21(5), 20–24 (2005)
4. Lundgaard, L.E., Hansen, W., Linhjell, D., Painter, T.J.: Aging of oil impregnated paper in
power transformers. IEEE Trans. Power Delivery 19(1), 230–239 (2004)
5. Unsworth, J., Mitchell, F.: Degradation of electrical insulating paper monitored with high
performance liquid chromatography. IEEE Trans. Electr. Insul. 25(4), 737–746 (1990)
6. De Pablo, A., Pahlavanpour, B.: Furanic compounds analysis: A tool for predictive
maintenance of oil-filled electrical equipment. Electra 175, 9–18 (1997)
7. Shroff, D.H., Stannett, A.W.: A review of paper aging in power transformers. IEE Proc.
C Gener. Transm. Distrib. 132(6), 312–319 (1985)
8. Ryadi, M., Jalbert, J.: Methanol marker for the thermal performance qualification of power
transformers. In: IEC 2013 (2013)
9. Matharage, S.Y., Liu, Q., Wang, Z.D., Wilson, G., Dyer, P., Mavrommatis, P.: Generation of
methanol and ethanol from inhibited mineral oil. In: INSUCON 2017 (2017)
10. Liland, K., et al.: Comparison between synthetic ester and mineral oil impregnated
transformer paper aging markers. In: ICDL 2017 (2017)
11. Bare, C., Merritt, S.: Methanol as an aeging marker for in service transformers. IEEE Trans.
Ind. Appl. 54(6), 6598–6602 (2018)
12. Jalbert, J., Gilbert, R., Denos, Y., Gervais, P.: Methanol: a novel approach to power
transformer asset management. IEEE Trans. Power Delivery 23(2), 514–520 (2012)
13. Matharage, S.Y., Liu, Q., Wang, Z.D., Mavrommatis, P., Wilson, G., Jarman, P.: Ageing
assessment of transformer paper insulation through detection of methanol in oil. In:
ICPADM 2015 (2015)
14. Münster, T., Kinkeldey, T., Werle, P., Hämel, K., Preusel, J.: Comparison of a thermally
accelerated aged oil-paper insulation of an open and hermetically sealed system. In: ISH
2019 (2019)
15. Imani, M.T., Homeier, K., Werle, P., Dräger, G.: How far are furan compounds reliable
indicators for thermal aging of oil impregnated cellulose insulation? In: CEIDP 2018 (2018)
 Applying Unsupervised Machine Learning
Method on FRA Data to Classify
Winding Types

Xiaozhou Mao1 , Shuntao Ji1, Zhongdong Wang1(&) ,

Paul Jarman2, Andrew Fieldsend-Roxborough2, and Gordon Wilson2
1
The University of Manchester, Manchester M13 9PL, UK
zhongdong.wang@manchester.ac.uk
2
National Grid Company, Warwick CV34 6DA, UK

Abstract. Over the years utilities have accumulated a large number of mea-
sured FRA data whilst the transformers’ design information such as winding
types may or may not necessarily be known. Different winding types own
different equivalent electrical parameters, i.e. capacitance and inductance. For
instance, the interleaved winding has higher series capacitance whilst the plain
disc winding has lower series capacitance. As a result, unalike features are
caused at specific frequency ranges of FRA. Consequently it is possible to
correlate FRA characteristics with known design features. Hierarchical clus-
tering is an unsupervised machine learning algorithm that groups similar objects
together. In this paper, using the National Grid FRA database as an example,
winding types are identified by Hierarchical Clustering method through
grouping similar FRA data. In addition, a pre-processing technique called
Dynamic Time Warping (DTW) is used to scale frequencies with the same FRA
features before applying Hierarchical Clustering, and this has been proved to be
the most suitable unsupervised machine learning methods to classify winding
types. National Grid has been retiring transformers, and each transformer retired
would go through forensic examination and knowledge acquired can then be
used for asset management. Same faults may occur to same winding types and
result in similar distortions of FRA features. With the technique employed in
this paper, in combination with expertise knowledge and forensic information
accumulated, the utility will be able to develop a strategy to manage similar type
of transformers and achieve effective asset management.

Keywords: Power transformers
FRA
Transformer windings

Classification
Machine learning

1 Introduction

Windings are the most important parts in a transformer; and the power and voltage
rating of a transformer influence the choice of transformer windings’ types signifi-
cantly. The higher the BIL level, the higher winding series capacitance it will require to
control the non-linear distribution of lightning voltage in a winding. In the UK power
transformers are mainly of core type and use disc-type windings. Plain Disc, Inter-
leaved Disc and Intershielded Disc windings are some widely used winding types.
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 969–981, 2020.
https://doi.org/10.1007/978-3-030-31676-1_91
970 X. Mao et al.

FRA is considered to be one of the most effective ways to detect the mechanical
integrity of transformer windings, particularly the winding movement and deformation,
caused by a short circuit fault in the system or mal-handling during the transportation and
installation [1, 2]. A low voltage sinewave signal at a frequency is injected at one end and
a signal is received at the other end of winding. And this process repeats over a large range
of frequency. FRA trace consists of the magnitude ratio and phase difference between the
two signals. Different frequency ranges of FRA trace are influenced by different trans-
former parts. In this paper, a group of 400/275/13 kV autotransformers are studied. For
this category of transformers, transformer core, inter-winding influence, properties of
winding-under-test and measurement setup are believed to dominate the bandwidths of
0–2 kHz, 2–20 kHz, 20–1000 kHz and >1000 kHz respectively [3, 4].
For windings of same type, similar winding mechanical faults may occur because
of their similar physical structure. In the UK forensic examination are routinely carried
out on retired transformers, and knowledge can then be acquired for asset management.
Therefore, any technique which helps to identify the unknown winding types of
transformers is desirable when considering the fact that a significant number of
transformers’ design information is unknown to the utilities. In this paper, Hierarchical
Clustering as an unsupervised machine learning method, is applied on the FRA traces
of 400/275/13 kV autotransformers. It groups similar traces together according to the
distances to each other for identifying the winding construction types.

2 Winding Types and Unsupervised Machine Learning

Methods

2.1 FRA Traces of Different Winding Types

White box models of equivalent electrical circuit have been built to understand FRA
traces [3, 5]. Different winding types own different electrical parameters, i.e., induc-
tance and capacitance. Resonance and anti-resonance are caused by the conjunct effect
of inductance and capacitance. They normally appear as local maximum and minimum
value on FRA traces, and can be described by complex zeros and poles mathematically
[6]. The variations in series capacitance, ground capacitance and inductance of different
winding types will lead to the difference of their FRA characteristics.
Generally, Plain Disc and Intershielded Disc winding types own low winding series
capacitance, whilst Interleaved Disc and Multiple Layer winding types have high
winding series capacitance. Plain Disc windings and Intershielded Disc windings have
been applied for a wide range of voltage levels (normally for a voltage level  132 kV
but sometimes 275 kV). Interleaved Disc windings are suitable for higher BIL level
(normally for a voltage level  275 kV). For historical reasons some manufacturers are
used to Multiple Layer windings’ design rather than disc-type windings, hence Multiple
Layer windings are used for all high and low voltage windings in a transformer. It is also
applicable for voltages from 33 kV right to 400 kV, or even up to 800 kV. All of the four
winding types mentioned above can be used for both series and common windings of the
400/275/13 kV transformers studied. As for the tertiary winding, single helical winding
type is customarily used, which has the lowest winding series capacitance.
 Applying Unsupervised Machine Learning Method on FRA Data 971

In the National Grid’s FRA database, 400 evenly distributed data points are
measured for the following frequency region: 5 Hz to 2 kHz, 50 Hz to 20 kHz, 500 Hz
to 200 kHz, 5 kHz to 2 MHz and 25 kHz to 10 MHz. The above mentioned five
winding types are shown in Fig. 1. Each trace in Fig. 1 starts from 5 Hz and ends at
1 MHz.

Intershielded Disc
Plain Disc
Multiple Layer
Interleaved Disc
Single Helical

Fig. 1. Typical FRA traces of winding types

It can be seen that Single Helical winding has a higher magnitude roughly from
0 Hz to 100 kHz. For the frequency range of 20 kHz to 1000 kHz, which is believe to
be dominated by winding properties, Multiple Layer winding’s FRA magnitude trace
rises about 20 dB per decade with obvious oscillations. With a higher series capaci-
tance, Interleaved Disc winding has the same rising speed but the tracer is smoother.
Camel humps usually appear on Plain Disc winding’s FRA traces. The FRA charac-
teristic of Intershielded Disc winding is not unique; it looks like Plain Disc winding
type. For both winding types, the smooth rising trend would occur after camel humps
feature [7]. Actually, the boundary points of bandwidth, 2 kHz, 20 kHz, and
1000 kHz, are empirical. The concerned FRA characteristic in one bandwidth can
move up or down to another bandwidth when the voltage and power ratings of winding
or winding type changes. For clarity, to avoid the influence of measurement setup, the
measured FRA data up to 200 kHz, are used to test the unsupervised machine learning
methods in this study.

2.2 Hierarchical Clustering

Hierarchical clustering method is an unsupervised machine learning algorithm which
groups similar objects [8]. Different from supervised machine learning method, it does
not require any input data with known classification [7]. For a given group of obser-
vations, each observation is initially treated as a separate cluster, and the algorithm
repeatedly identifies the closest two clusters and merges them, until all the clusters are
972 X. Mao et al.

merged into one cluster. As shown in Table 1, six dataset, A–F, the distances between
every two dataset are listed in Table 1(a). Hierarchical Clustering algorithm finds the
smallest distance in this distance matrix, which is 10 between E and F in Table 1(a),
thus E and F are clustered together first. Then, the smaller distance between A to E and
A to F is adopted as the distance from cluster A to the new cluster {E, F} which is 77.
This is called Single Linkage. The new distance matrix is recalculated in Table 1(b),
where the smallest distance is 16 between A and B. The algorithm keeps merging
clusters until only one cluster exists. Dendrogram in Fig. 2 shows the merging process.

Table 1. Distance matrix example

Fig. 2. Dendrogram of datasets A–F in Table 1

The clustering result can be decided either by final cluster number or cut off
distance. When the final cluster number is set as 2, then two clusters {A, B} and {C, D,
E, F} are divided. When the cut off distance is set at the height of the dash line in
Fig. 2, clustering result remains the same.
In fact, the distance between two observations reflects the similarity between them.
Euclidean Distance, Standardized Euclidean Distance and Correlation Distance are
some most commonly used distance types. Euclidean Distance is the root of accu-
mulated squared data deviation, and Standardized Euclidean Distance is the Euclidean
Distance of observations after standardization [7]. Correlation distance uses a real
number between 0 and 1 to reflect the resemblance between two observations. Apart
from the Single Linkage, Complete Linkage, Weighted Linkage can also be used.
Complete Linkage is opposite to Single Linkage. It uses the longest distance between
observations as the new distance, whilst Single Linkage uses the nearest one. The
Single Linkage may cause loose linkage, and the Complete Linkage may cause com-
pact linkage. Weighted Linkage can be viewed as a compromise between the above
 Applying Unsupervised Machine Learning Method on FRA Data 973

two linkages. It uses the average of two distances between a specific cluster and the two
observations to be merged, as the new distance. In the given example in Table 1, using
weighted linkage, the distance of newly merged cluster {E, F} to cluster A is (77 + 79)/
2 = 78.

3 Application of Hierarchical Clustering

All together 28 measured FRA traces from 400/275/13 kV autotransformers are studied
using Hierarchical Clustering, and their winding type information are listed in Table 2
and FRA traces are plotted in logarithmic scale in Fig. 3.

Table 2. Basic information of FRA traces

Winding Trace Trace

Type Number Label
Intershielded Disc 4 [1, 2, 3, 4]

Plain Disc 6 [5, 6, 7, 8, 9,10]

Multiple Layer 4 [11,12,13,14]

Interleaved Disc 6 [15,16,17,18,19,20]

Single Helical 8 [21,22,23,24,25,26,27,28]

(a)

Winding Trace Trace

Type Number Label
Series Winding 9 [5,6,7,12,13,15,16,17,18]

Common Winding 11 [1,2,3,4,8,9,10,13,14,19,20]

Tertiary 8 [21,22,23,24,25,26,27,28]
(b)

For each trace, there are 1120 data points for the frequency range from 5 Hz to
200 kHz. Clustering uses Euclidean Distance and Weighted Linkage. The Euclidean
Distance between two FRA traces is calculated as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n¼m
d¼ R ðXai  Xbi Þ2 ð1Þ
i¼1

where Xai and Xbi are the magnitudes at frequency point i on FRA traces a and b, in the
unit of dB.
974 X. Mao et al.

The clustering Dendrogram is shown in Fig. 4. Suppose the final cluster number is
set as 2, the left cluster contains 8 traces of Single Helical Winding type, and the right
cluster contains traces from the rest winding types. For the left hand side cluster, 8
traces {21–28} from Single Helical winding are grouped due to their high magnitude at
most frequencies in the whole range. Traces {21–24} are clustered together with
distances as tiny as 3 since they are from the same transformer. So are traces {25–27}
with distances around 100. Trace 28 is from a third transformer, and its distance to the
other Single Helical Traces from other transformers is about several hundreds.

Fig. 3. FRA traces of 400/275/13 kV transformers

Fig. 4. 5 Hz to 200 kHz clustering dendrogram using Euclidean distance weighted linkage
 Applying Unsupervised Machine Learning Method on FRA Data 975

For the right hand side cluster, all common windings {1–4, 8–10, 13, 14, 19, 20}
are cluster together in small group. Series windings {5–7} are grouped with common
windings afterwards. Finally the rest common windings traces {11, 12, 15–18} are
merged. This result is not desired, since it does not distinguish winding types, but
perhaps the differences in magnitude.
As known the frequency range 5 Hz to 200 kHz can be split into three bandwidths
and the relation among the distances exists as:

d 2 ¼ d12 þ d22 þ d32 ð2Þ

where d, d1, d2, d3 are the Euclidean distances of 1120 points from 5 Hz to 200 kHz,
400 points from 5 Hz to 2 kHz, 360 points from 2 kHz to 20 kHz and 360 points from
20 kHz to 200 kHz. It should be noted that the distance d of two FRA traces is not only
affected by the distance at each measured frequency but also the amount of measured
frequency points.
It is worth to study the distance in each bandwidth to understand their individual
contribution to the final clustering classification results.

3.1 Bandwidth Dominated by Core

Using FRA data in the 5 Hz–2 kHz bandwidth which is dominated core, the dendro-
gram is shown in Fig. 5. Set the final cluster number as 3 and a clear clustering result
can be obtained, i.e., the tertiary windings on the left, common windings in the middle
and series windings cluster on the right. Clearly d1 are sensitive to voltage levels and
good at classifying T, C and S windings. The classification result is reasonable as FRA
is dominated by the core in this bandwidth and the core inductance is proportional to
the voltage level squared.

Fig. 5. 5 Hz to 2 kHz clustering dendrogram using Euclidean distance weighted linkage

976 X. Mao et al.

3.2 Winding Interaction Controlled Bandwidth

The winding interaction dominated frequency bandwidth is from 2 kHz to 20 kHz, due
to its complexity, the clustering results using FRA data at this bandwidth only, are
messy and thus discussion on d2 is ignored in this paper.

3.3 Winding Controlled Bandwidth

For winding controlled bandwidth 20 Hz to 200 kHz, the clustering results for the
275 kV and 400 kV windings are shown in Fig. 6. Two clusters are clearly formed.
Multiple Layer and Interleaved Disc windings are clustered together whilst Plain Disc
and Intershielded Disc windings are gathered as the other cluster. Multiple Layer
windings and Interleaved Disc windings are both the type of windings with high series
capacitance, and Plain Disc windings and Intershielded Disc windings are the type of
windings with low series capacitance.

Fig. 6. 20 kHz to 200 kHz clustering dendrogram using Euclidean distance weighted linkage

Discussions on d3 are detailed here. For the left hand side cluster {11–20}, traces
{14, 15, 16, 19, 20} are assembled as one small group, and traces {11, 12, 13, 17, 18}
are assembled as another small group. Interleaved Disc traces {15, 16} are merged with
a distance as tiny as 4, and Interleaved Disc traces {19, 20} are merged with a distance
of 10. The distance between small cluster {15, 16} and cluster {19, 20} rises to 68,
mainly due to the large distances between Multiple Layer trace 14 and Interleaved Disc
traces {19, 20}, which are 104 and 108. Because of the same reason, the distance
between cluster {14, 15, 16, 19, 20} and cluster {11, 12, 13, 17, 18} rises to 150,
largely influenced by the most unalike traces members in the cluster. The final merging
 Applying Unsupervised Machine Learning Method on FRA Data 977

distance of the 20 trance is 356. Actually the merging distance is not only influenced by
the largest distance (which is 503 between Plain Disc winding 5 and Interleaved Disc
winding 17), but also every distance between every two traces. Details of the distances
for the right hand side cluster {1–10}, will be ignored here.
In fact, from 20 kHz to 200 kHz, the difference in winding series capacitance
causes the magnitude difference and leads to this clustering result.

4 Application of DTW Technique

Considering the shifting/scaling of winding structure’s feature frequencies caused by

sizing including rated voltage and power, a technique called dynamic time warping
(DTW) can be helpful to pre-process the data for better identifying the same FRA
characteristics at different frequencies, and then allow the processed data to be applied
with hierarchical clustering.

4.1 Pre-processing by DTW

In time series analysis, DTW is one of the algorithms for measuring similarity between
two temporal sequences, which may vary in speed [9]. It can be used to identify same
characteristics before and after scaling. DTW horizontally scales two traces to align
them to each other. Among all the possible paths, it chooses the one which has smallest
moving distance. Horizontal points will be inserted to compensate the part where the
height is not aligned. Generally the greater the difference between the two traces, the
more insertion points there are. Indeed, any data that can be turned into a linear
sequence can be analyzed with DTW.
In Fig. 7, comparison of two Plain Disc winding FRA traces, from 5 Hz to
200 kHz, before and after DTW applied is illustrated. The horizontal axis is data point
for both Fig. 7(a) and 7(b), instead of frequency, for the convenience of comparison
after DTW is applied. The original data points’ quantity increases from 1120 to 1802
after DTW is applied. The distance between the two traces changes from 406 to 137
after DTW is used to match the resonances and anti-resonances. Therefore, DTW is
able to enhance the similarity between two traces.
Another example is shown in Fig. 8 when DTW is applied on two windings from
difference winding types, i.e., Plain Disc winding and Interleaved Disc Winding. The
quantity of data points rises from 1120 to 1565. The distance between the two traces is
reduced from 383 to 233. Though the similarity between them is enhanced by DTW,
there still exists a significant difference between the distance of FRA traces of same
winding type (137) and the distance of FRA traces of difference winding type (233).
978 X. Mao et al.

Fig. 7. DTW applied on two windings of same type (plain disc windings)
 Applying Unsupervised Machine Learning Method on FRA Data 979

Fig. 8. DTW applied on two different windings (plain disc and interleaved disc windings)

4.2 Clustering DTW Processed Data

The Euclidean Distance should be calculated after the traces being processed by DTW.
When DTW is applied on every two FRA traces in Table 1 to calculate their Euclidean
Distance, a better clustering result than the former attempts can be produced as shown
in Fig. 9.
980 X. Mao et al.

Fig. 9. 5 Hz to 200 kHz clustering dendrogram using Euclidean distance weighted linkage

The overall distance is reduced as compared with Fig. 4, with the final distance
between the last two clusters decreasing from about 900 to about 550. It can be seen
that all traces are divided into three distinctive groups; one group contains all the 8
Single Helical windings and the others are separated into two groups, one with low
capacitance (Plain Disc winding and Intershielded Disc winding) and the other with
high capacitance (Multiple Layer winding and Interleaved Disc winding).
Once a method which can give a clear clustering result is developed, winding type
recognition could be achieved by clustering of mixed known and known windings. The
locations of known windings indicate the corresponding classification of unknown
windings.

5 Conclusions

In this paper, unsupervised machine learning method Hierarchical Clustering is used to

classify 28 FRA traces of 400/275/13 kV autotransformers, with the aim to identify
winding types. 8 Single Helical Windings can be easily picked out due to their high
magnitude in most frequencies, especially at the low bandwidth of 5 Hz to 100 kHz.
For the frequency range <20 kHz, a larger amount of data points is used as com-
pared with the high bandwidth from 20 kHz to 200 kHz, this means more weight is
given to the low bandwidth, and thus the difference between Single Helical windings
and other windings are emphasised.
With DTW technique applied to pre-process the FRA data, the same FRA features
at different frequencies can be scaled or shifted, this pre-processing technique signif-
icantly increases the chance to correctly identify the winding types, three clustering
groups can be identified as Single Helical winding, winding with low series capacitance
(Plain Disc winding and Intershielded Disc winding) and windings with high series
capacitance (Multiple Layer winding and Interleaved Disc winding).
 Applying Unsupervised Machine Learning Method on FRA Data 981

References
1. Sofian, D.M., Wang, Z.D., Jayasinghe, S.B.: Frequency response analysis in diagnosing
transformer winding movements – fundamental understandings. In: 39th International
Universities Power Engineering Conference, 2004. UPEC 2004, Bristol, UK, vol. 1, pp. 138–
142 (2004)
2. IEC-International Electrotechnical Commission: Power transformers - Part 18: measurement
of frequency response. IEC 60076-18 (2012)
3. Wang, Z.D., Li, J., Sofian, D.M.: Interpretation of transformer FRA responses—Part I:
influence of winding structure. IEEE Trans. Power Delivery 24(2), 703–710 (2009)
4. Sofian, D.M., Wang, Z.D., Li, J.: Interpretation of transformer FRA responses—Part II:
influence of transformer structure. IEEE Trans. Power Delivery 25(4), 2582–2589 (2010)
5. Ang, S.P., Li, J., Wang, Z., Jarman, P.: FRA low frequency characteristic study using duality
transformer core modeling. In: 2008 International Conference on Condition Monitoring and
Diagnosis, Beijing, pp. 889–893 (2008)
6. Mao, X., Wang, Z., Wang, Z., Jarman, P.: Accurate estimating algorithm of transfer function
for transformer FRA diagnosis. In: 2018 IEEE Power Engineering Society General Meeting,
Portland, OR (2018)
7. Mao, X., Wang, Z.D., Jarman, P., Roxborough, A.: Winding type recognition through
supervised machine learning using frequency response analysis (FRA) data. In: 2019
ICEMPE, Guangzhou, China (2019)
8. Ward, J.H.: Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc. 58
(301), 236–244 (1963)
9. Sakoe, H., Chiba, S.: Dynamic programming algorithm optimization for spoken word
recognition. IEEE Trans. Acoust. Speech Signal Process. 26(1), 43–49 (1978)
 Interpreting First Anti-resonance
of FRA Responses Through Low Frequency
Transformer Modelling

Bozhi Cheng1, Peter Crossley1, Zhongdong Wang1(&), Paul Jarman2,

Andrew Fieldsend-Roxborough2, and Gordon Wilson2
1
The University of Manchester, Manchester M13 9PL, UK
zhongdong.wang@manchester.ac.uk
2
National Grid Company, Warwick CV34 6DA, UK

Abstract. FRA is an effective detection method for mechanical deformation of

transformer windings; however the challenge of applying the FRA technique
lies in the correct interpretation of FRA measurement results. Modelling and
simulation can help understand the FRA characteristics and in this respect much
research work are needed especially to develop the modelling capability for
three-phase three-winding auto-transformers under the condition of single phase
excitation such as the FRA measurement set-up. This paper is a sister paper to
[7] which extends the discussion on how to interpret low frequency FRA fea-
tures especially the first anti-resonance. Through our previous publication, it has
become a common knowledge nowadays that the 1st anti-resonance of FRA is
produced by the coupling between the core inductance and the equivalent
winding capacitance, though a quantitative analysis is carried out in this paper to
prescribe an analytical equation first time showing the linear composition of the
equivalent winding capacitance made from winding series capacitances and
inter-winding capacitances. It is necessary to understand that although shifts of
resonance (and/or anti-resonance) frequencies are regarded as the key indicators
of winding deformation, the frequency shift, Df, of the resonance (and/or anti-
resonance) before (f1) and after (f2) the deformation, is actually not the best
indicative parameter as Df cannot be related to the change of electrical param-
eters in the model, instead the ratio, f2/f1 has a well-defined quantitative linkage
with the winding inductances and capacitances. It is therefore recommended to
use the frequency ratio, f2/f1, of the resonance (and/or anti-resonance) before (f1)
and after (f2) the deformation for FRA interpretation.

Keywords: FRA
Interpretation
Transformer
Modelling

1 Introduction

The failure of transformers can result in safety issues and economic losses.
Although FRA is widely used to detect mechanical winding deformation, the success of
this technique lies in the correct interpretation of the FRA plot and up to now there is
no IEC/IEEE standard for FRA interpretation although a CIGRE WG A2.53 working
group is currently working towards producing a technical brochure on this respect.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 982–992, 2020.
https://doi.org/10.1007/978-3-030-31676-1_92
 Interpreting First Anti-resonance of FRA Responses 983

Most researchers investigate the FRA characteristics in the high frequency region
through experiments and/or simulation due to the FRA characteristics in this region is
more winding structure related [1–6], nevertheless, the low frequency FRA feature, in
the frequencies generally up to 2 kHz, associated with global parameters of a trans-
former, is the “grey” area where not much research attention has been paid. A mea-
surement based approach is proposed in [7] proposed to establish a low frequency
transformer model for FRA simulation studies, and the simulation results are validated
through the comparison with measurement results. In [7] a 400/275/13 kV auto-
transformer is used as an example to build a low frequency first order transformer
model and sensitivity studies are conducted on the influencing electrical factors on the
low frequency anti-resonances and resonances. This paper is a sister paper to [7] which
continue to study and discuss how to interpret the first anti-resonance of FRA
responses.

2 Simulation Model

The transformer being modelled is a three-limb core type 1000 MVA 400/275/13 kV
auto-transformer. There are 3 windings per phase, i.e. series (S), common (C) and
tertiary (T) windings. Figure 1 shows the equivalent circuit which involves both
inductive and capacitive elements, where the model is a first order model where each
winding is represented by one unit of global parameters, inductance and capacitance.
All the representations of the symbols in the circuit can be found in Table 1, the
parameters except core leg and yoke inductances are calculated based on winding
geometry, while the core leg and yoke inductances are calculated based on FRA
measurement results which will be discussed in Sect. 3.1.
In [7] it is concluded that the model can produce reasonable simulation results up to
a few kHz. The anti-resonances in the low frequency region and a typical double-peak
feature of series and common winding FRA responses can be simulated, and sensitivity
studies are implemented to identify the influence of electrical components on the FRA
responses. The results show that the anti-resonance of FRA response in the low fre-
quency region is dominated by the core inductance, the double-peak feature is influ-
enced by winding series capacitance and air-core inductances, and the interwinding
capacitances can affect both the first anti-resonance and the double-peak feature sen-
sitively. While the FRA characteristics in the high frequency region cannot be reflected
due to the over-simplicity of the first order model [7].
984 B. Cheng et al.

CIWCT

CGS CGS
A3
CIWCS CSS LS
a2
CIWCS CSC LC
CGT 3A2
Na
CGC CGC LT CST
CGT

CIWCT
CIWCT

CGS CGS
B3
CIWCS CSS LS
b2
CIWCS CSC LC
CGT
Nb 3B2
CGC CGC CST
LT
CGT

CIWCT

CIWCT

CGS CGS
C3
CIWCS CSS LS
c2
CIWCS CSC LC
Nc CGT 3C2
CGC CGC LT CST
CGT

CIWCT

Fig. 1. Equivalent circuit of 400/275/13 kV auto-transformer

Table 1. Electrical parameters in the model

Parameters Values
Ground capacitance of series winding CGS 241.08 pF
Series capacitance of series winding CSS 1603.86 pF
Ground capacitance of common winding CGC 38.5 pF
Series capacitance of common winding CSC 30.21 pF
Interwinding capacitance between common and series windings CIWCS 1495.57 pF
Ground capacitance of tertiary winding CGT 2820.05 pF
Series capacitance of tertiary winding CST 5.67 pF
Interwinding capacitance between common and tertiary windings CIWCT 772.45 pF
Air-core inductance of series winding LS 159.2 mH
Air-core inductance of common winding LC 83.34 mH
Air-core inductance of tertiary winding LT 438.51 mH
Core leg inductance LLEG 13.43 H
Core yoke inductance LYOKE 12.61 H
N1 : N2 125:275
pffiffiffi
N1 : N3 125= 3 : 13
 Interpreting First Anti-resonance of FRA Responses 985

3 Interpretation of First Anti-resonance

3.1 Core Inductance Calculation Based on FRA
In [7], a group of special winding terminal tests are performed to obtain the electrical
parameters for the transformer model. Some of the parameters identified by the special
terminal tests may be found in the factory test reports and capacitance/tangent delta test
report; measurements of core leg and yoke inductances are special, and they can be
identified also by an alternative approach, demonstrated in this section. Based on three
phases FRA measurement data at low frequencies, core leg and yoke inductances can
be calculated and no extra terminal tests are required. In the core dominated low
frequency region of FRA response, the magnitude decreases as 20 dB per decade, and
the equivalent circuit of FRA end-to-end measurement can be regarded as a purely
inductance with a 50Ω in series connection, shown in Fig. 2.

Fig. 2. Equivalent circuit of FRA measurement in the low frequency region

The ratio of terminal voltages can be calculated by Eq. (1).

 
Vout 
A ¼ 20log10   ð1Þ
Vin 

Where A is the FRA magnitude data (dB), Vin and Vout are the voltages measured at the
input and output terminals of a winding. The equivalent core inductances can be
calculated from A and B phase FRA measurement data as they are different combi-
nations of core leg and yoke inductances, given as:

LeA ¼ ðLLEG ==LYOKE þ LYOKE Þ==ðLLEG ==LYOKE Þ ð2Þ

LeB ¼ ðLLEG ==LYOKE þ LLEG ==LYOKE Þ==LLEG ð3Þ

Where LeA is equivalent core inductance calculated based on A phase FRA mea-
surement data, equivalent core inductance of C phase has the same expression to A
phase, LeB is equivalent core inductance calculated based on B phase FRA measure-
ment data.
986 B. Cheng et al.

LeA and LeB are calculated based on FRA measurement data between 5 Hz and
60 Hz as higher than 60 Hz the capacitive elements start to influence the FRA
response, resulting in the FRA magnitude decreases more than 20 dB per decade. The
mean values of equivalent core inductances seen from A and B phase windings are
calculated, Table 2 gives the inductance calculation results from A and B phase
winding FRA measurement data. It should be mentioned that LLEG and LYOKE of three
windings (Series, Common and Tertiary) follows exactly the relationship of the ratio of
turn number squared.

Table 2. Equivalent core inductance, LLEG and LYOKE seen at three winding terminals
Parameters Series Common Tertiary
Mean LeA 4.85 H 23.56 H 158 mH
Mean LeB 6.61 H 31.97 H 215 mH
LLEG 13.43 H 64.82 H 438 mH
LYOKE 12.61 H 61.50 H 411 mH

Although the core and air core inductances plugged into the model are the values
seen from series winding side, with the involvement of ideal transformers, the model is
capable to perform FRA simulation on either common or tertiary winding terminals
without changing the values of any inductive elements.
The model has been validated in [7] to produce reasonable low frequency FRA
characteristics, as compared with measurement results. Next section will be based on
the model and conduct sensitivity simulation studies to investigate the frequency of the
first anti-resonance and its influencing factors.

3.2 Composition of Equivalent Winding Capacitance Ce Seen at Winding

Terminal
Based on the knowledge of how the first anti-resonance of FRA response is produced
[7], the frequency of the first anti-resonance can be expressed by Eq. (4).

1
f ¼ pffiffiffiffiffiffiffiffi ð4Þ
2p LCe

Where f is the frequency of the first anti-resonance, L is the equivalent core inductance
combined by LLEG and LYOKE , Ce is the equivalent winding capacitance seen as a
global parameter at the winding terminal, which is believed to be composed of all
geometric winding capacitances.
To quantify the influence of winding capacitances on the first anti-resonance, the
relationship between Ce and winding capacitances is required. In this section, sensi-
tivity studies on the contribution from individual winding capacitance to Ce is
implemented. Three windings of B phase are tested individually, LLEG and LYOKE are
not changed, the value of each individual winding capacitance is changed from 10%,
 Interpreting First Anti-resonance of FRA Responses 987

100%, 200%, … to 600% of the original value. To calculate the value of Ce in each
sensitivity study, the equivalent core inductances of B phase series, common and
tertiary windings are used according to Table 2, then the first anti-resonance frequency
is identified from the FRA response and Eq. (4) is used to calculate Ce . The baseline of
Ce seen from series, common and tertiary windings, are equal to 25556 pF, 4034 pF
and 826990 pF respectively.
After the sensitivity studies, the relationship between Ce at the first anti-resonance
and the winding capacitances for B phase windings are shown in Figs. 3, 4 and 5
respectively. It is found that no matter which winding is tested, the winding series
capacitances and interwinding capacitances always have a linear relationship with Ce ,
and CIWCS contributes most to Ce among all the capacitive elements.

CSS
1.0x105
Equivalent Capacitance (pF)

CSC
8.0x10 4 CST
CIWCT
6.0x10 4 CIWCS

4.0x104

2.0x104

0.0
0 1 2 3 4 5 6
Proportion (100%)

Fig. 3. Ce for series winding

1.8x104 CSS
Equivalent Capacitance (pF)

CSC
1.5x104
CST
1.2x10 4 CIWCT
CIWCS
9.0x103

6.0x103

3.0x103

0.0
0 1 2 3 4 5 6
Proportion (100%)

Fig. 4. Ce for common winding

988 B. Cheng et al.

3.0x106 CSS

Equivalent Capacitance (pF)

CSC
2.5x106 CST
CIWCT
2.0x106 CIWCS

1.5x106

1.0x106

5.0x105

0 1 2 3 4 5 6
Proportion (100%)

Fig. 5. Ce for tertiary winding

The influence of CIWCS can even transfer to the FRA response of tertiary winding,
consequently explains the result that CIWCS has the most sensitive influence on the 1st
anti-resonance of B phase winding FRA responses [7]. For series and tertiary windings,
CIWCT takes the secondary status. For common winding, CSS ranks only second to
CIWCS , on the other hand, the series capacitances of common and tertiary windings
have minor contribution to Ce because they are too small when compare to other
capacitive elements.
Through the linear regression method, the expressions of equivalent capacitance Ce
for B phase windings are calculated and shown as Eqs. (5), (6) and (7) respectively, the
unit for the variables in three equations is pF, CST is neglected in the equations since it
is so small that even 600% CST can hardly shift the first anti-resonance in any sensi-
tivity study case (frequency resolution is 1 Hz), in addition, for the constants (1715,
584 and 207590 pF) in the equations, it is believed that they represent the effect of
winding ground capacitances.

Ces ¼ 1:59CSS þ 7:46CSC þ 3:11CIWCT þ 4:64CIWCS þ 1715 ð5Þ

Cec ¼ 0:3CSS þ 1:38CSC þ 0:18CIWCT þ 0:88CIWCS þ 584 ð6Þ

Cet ¼ 47CSS þ 233CSC þ 103:6CIWCT þ 136CIWCS þ 207590 ð7Þ

With the relationship developed between Ce and other geometric winding capaci-
tances, it is now feasible to demonstrate that the frequency of the first anti-resonance
can be represented by electrical parameters in the model.

3.3 Discussion on Deformation Indicative Parameter Which Can Link

with Change of Electrical Parameter in Equivalent Circuit
The essence of FRA interpretation is to quantify the relationship among winding
geometry, electrical parameters of the model and the FRA plot, or in another word, the
 Interpreting First Anti-resonance of FRA Responses 989

change of winding geometric dimension, electrical parameters and the shift of reso-
nance or anti-resonance frequencies of the FRA response.
Generally, the frequency shifting ðDf Þ was used to describe and reflect the degree
of deformation, however in this section, a proposal is made, based on this study of the
first anti-resonance, to use instead the ratio expression of frequency before (f1) and after
(f2) the deformation to build a direct linkage with the change of electrical parameters in
the equivalent circuit.
Based on (4), it is possible to link the shift of frequency of the first anti-resonance
with the change of electrical parameters of the equivalent circuit as expressed by
Eq. (8).
pffiffiffiffiffiffiffiffiffiffiffiffi
f2 L1 Ce1
¼ pffiffiffiffiffiffiffiffiffiffiffiffi ð8Þ
f1 L2 Ce2

Where f 1 represents the frequency of baseline, f 2 represents the frequency after

deformation and its associated change of electrical parameters,L1 , Ce1 and L2 , Ce2
represent the electrical parameters before and after the deformation.

3.3.1 Demonstration Case 1: Core Inductance

To verify the effectiveness of expression (5) and the influence of core inductance on the
first anti-resonance, the sensitivity study results on the core inductance are extracted
from [7].

Table 3. Frequency ratio and Df

(a) Series winding (S) results

Case Fre (Hz) Δf
Baseline 386 Hz - -
50% of and 533 Hz 1.381 147 Hz
200% of 276 Hz 0.715 -110 Hz

(b) Common winding (C) results

Case Fre (Hz) Δf
Baseline 424 Hz -
50% of and 588 Hz 1.387 164 Hz
200% of and 318 Hz 0.750 -106 Hz

(c) Tertiary winding (T) results

Case Fre (Hz) Δf
Baseline 367 Hz -
50% of and 508 Hz 1.384 141 Hz
200% of and 277 Hz 0.755 -90 Hz
990 B. Cheng et al.

In reality, the core inductance can be changed due to winding short circuit fault or
during the transformer short circuit test. In this sensitivity study case, LLEG and LYOKE
are changed to 50% lower and then 200% higher. Following (5), when LLEG and LYOKE
are reduced by 50%, the ratio f 2 =f 1 should be equal to 1.414 and when LLEG and LYOKE
are increased by 200%, the ratio f 2 =f 1 should be equal to 0.707. The comparison on the
effectiveness of Df and the ratio, f 2 =f 1 to be related to the change of core inductance is
presented in Table 3.
According to the calculation results shown in Table 3, it can be concluded that
Eq. (5) can successfully link the change of core inductance to the frequency ratio of
f 2 =f 1 . As f 2 =f 1 based on the simulated FRA plots are very close to the calculated value
of 1.414 and 0.707, although slight deviations are found due to that the frequency
resolution (1 Hz) is not small enough. On the other hand, using Df cannot give an
intuitive impression that how the core inductance is changed.

3.3.2 Demonstration Case 2: Winding Capacitance

To verify the expressions of Ce for different B phase windings, the sensitivity study
results of B phase series winding in [7] are extracted, and new sensitivity studies of B
phase common and tertiary windings are conducted.
In the case of capacitive elements sensitivity studies, no inductive elements are
changed, each individual winding capacitance is doubled, the ratios of frequencies
(f 2 =f 1 ) at the first anti-resonance are calculated
pffiffiffiffiffiffiffi based on FRA simulation results, and
pffiffiffiffiffiffiffi
ratios of the square root of Ce Ce1 = Ce2 are calculated by Eqs. (5), (6) and (7). The
pffiffiffiffiffiffiffi pffiffiffiffiffiffiffi
frequency of 1st anti-resonance, the frequency ratio, the ratio of Ce1 = Ce2 and the
frequency shift Df are calculated and presented in Table 4.
Using the expressions of Ce derived in Sect. 3.2, doubled individual geometric
winding capacitance may only bring a small percentage change to Ce . The frequency
ratios (f 2 =f 1 ) canbe extracted from FRA plots, which are very similar to the ratio of the
pffiffiffiffiffiffiffi pffiffiffiffiffiffiffi
square root of Ce Ce1 = Ce2 before and after the change of capacitance for all cases.
The sensitive influence of CIWCS on Ce also provide quantitative explanation for the
results in [7]. On the other hand, Df cannot be related to the change of the electrical
parameters.
With the works presented in Sect. 3, an interpretation procedure towards the fre-
quency shift of first anti-resonance can then be summarized as follows:
(1) Calculating the equivalent core inductances of baseline FRA ðL1 Þ and diagnostic
FRA ðL2 Þ based on low frequency FRA data, by comparing the calculation
results, the variation of core inductance can be quantified.
pffiffiffiffiffi pffiffiffiffiffi
(2) Calculating the ratio of frequencies ðf 2 =f 1 Þ and L1 = L2 to indicate if winding
capacitances are also changed.
(3) In the case of winding capacitance being modified, the Eqs. (5), (6) and (7) can
then be used to further discuss the variation of capacitive elements.
 Interpreting First Anti-resonance of FRA Responses 991

Table 4. Frequency ratio f 2 =f 1 and Df caused by geometric winding capacitances

(a) Series winding results

Case Frequency (Hz) (pF) Δf (Hz)

Baseline 386 25556 - -

2 times 370 28022 0.959 0.953 -16
2 times 385 25793 0.997 0.996 -1
2 times 352 30875 0.912 0.917 -34
2 times 312 39392 0.808 0.805 -74

(b) Common winding results

Case Frequency (Hz) (pF) Δf (Hz)

Baseline 424 4034 - -

2 times 403 4505 0.951 0.945 -21
2 times 423 4076 0.998 0.995 -1
2 times 411 4323 0.969 0.967 -13
2 times 331 6664 0.781 0.777 -93

(c) Tertiary winding results

Case Frequency (Hz) (nF) Δf (Hz)

Baseline 367 826.99 - -

2 times 356 886.5 0.970 0.959 -11
2 times 367 834.03 1 0.996 0
2 times 338 985.65 0.921 0.918 -29
2 times 302 1234.6 0.823 0.824 -65

4 Conclusions

In this paper, a first order transformer model is established to discuss the FRA char-
acteristics in the low frequency region. Special attention is paid to the first anti-
resonance of FRA response, with the knowledge that the first anti-resonance is created
by the resonant coupling between core inductance and equivalent winding capacitance.
A quantitative analysis is carried out and for the first time a linear composition of the
equivalent winding capacitance made from winding series capacitances and inter-
winding capacitances is revealed, consequently, the linkage of the first anti-resonance
and the electrical components of the equivalent circuit, is proposed.
In the respect of diagnosis, frequency shifts of the resonance and/or anti-resonance,
are generally regarded as the key indicators of winding deformation. Under a much more
severe situation, creation of new or disappearance of old resonances can also occur. In
this paper, it is suggested that the frequency shift Df is not able to intuitively reflect the
change of electrical parameters; alternatively the frequency ratio ðf 2 =f 1 Þ can better
represent the quantitative change of electrical parameters in the equivalent circuit.
992 B. Cheng et al.

References
1. Samimi, M.H., Tenbohlen, S., Akmal, A.A.S., Mohseni, H.: Dismissing uncertainties in the
FRA interpretation. IEEE Trans. Power Delivery 33(4), 2041–2043 (2018)
2. Wang, Z.D., Li, J., Sofian, D.M.: Interpretation of transformer FRA responses - Part I:
influence of winding structure. IEEE Trans. Power Delivery 24(2), 703–710 (2009)
3. Sofian, D.M., Wang, Z.D., Li, J.: Interpretation of transformer FRA responses - Part II:
influence of transformer structure. IEEE Trans. Power Delivery 25(4), 2582–2589 (2010)
4. Hashemnia, N., Abu-Siada, A., Islam, S.: Improved power transformer winding fault
detection using FRA diagnostics - Part 1: axial displacement simulation. IEEE Trans. Dielectr.
Electr. Insul. 22(1), 556–563 (2015)
5. Hashemnia, N., Abu-Siada, A., Islam, S.: Improved power transformer winding fault
detection using FRA diagnostics - Part 2: radial deformation simulation. IEEE Trans. Dielectr.
Electr. Insul. 22(1), 564–570 (2015)
6. Pham, D.A.K., Pham, T.M.T., Borsi, H., Gockenbach, E.: A new method for purposes of
failure diagnostics and FRA interpretation applicable to power transformers. IEEE Trans.
Dielectr. Electr. Insul. 20(6), 2026–2034 (2013)
7. Cheng, B., Crossley, P., Wang, Z.D., Jarman, P., Roxborough, A.: Interpretation of FRA
results through low frequency transformer modelling. In: IEEE 2nd International Conference
on Electrical Materials and Power Equipment, Guangzhou, China (April 2019)
 Feasibility of Detecting Partial Discharge
in Smoothing Reactor Using UHF Method

Shusheng Zheng1,2(&), Yangzhen Cui1,2, and Bo Qi1,2

1
State Key Laboratory of Alternate Electrical Power System with Renewable
Energy Sources, North China Electric Power University, Beijing, China
zss4@ncepu.edu.cn
2
Beijing Key Laboratory of High Voltage and EMC,
North China Electric Power University, Beijing, China

Abstract. Smoothing reactors is of great importance in ultra-high-voltage

converter substation, so Partial Discharge (PD) detection is now suggested to
find out the insulation defects. However, until now, there is no report on this
work. To explore the feasibility of detecting PDs in smoothing reactor by using
Ultra-High-Frequency (UHF) method, three works have been done in this paper.
Firstly, the PD tests have been carried out to collect the UHF signals radiated
from the reactor. Secondly, polarization of electric field and magnetic field
components of electromagnetic wave is analyzed based on waveguide theory.
Thirdly, simulation work has been done to present the UHF signal propagation.
It is shown that UHF signals can radiate out from smoothing reactor. There are
two propagation paths, one is diffraction between every two adjacent winding
discs, and the other is leakage in cooling channel between two coils. The electric
field polarization directions of the first electro-magnetic wave are difference,
which are decided by the propagation paths. These conclusions is benefit for
promoting the UHF technology for detecting and locating PDs in smoothing
reactors.

Keywords: Smoothing reactor
Partial Discharge
Ultra-High-Frequency

1 Introduction

Dry-type smoothing reactor is one of the main devices in ultra-high-voltage converter

substation, which is used to suppress the harmonics current and the fault current [1]. In
the transient process of sudden failure in power system, the high transient overvoltage
may appear on the reactor. With the increase of operating frequency, the transient
overvoltage on the reactor will produce a detrimental effect on turn-to-turn insulation,
which is likely to cause the turn-to-turn short circuit and faults on the reactor. In recent
years, reactor burnouts have occurred in a majority of areas. According to statistics,
turn-to-turn insulation defect is the main reason for equipment failure of the smoothing
reactor [2].
Currently high frequency oscillation test is widely used to detect the turn-to-turn
insulation of the reactor. The high frequency oscillation test for the reactor is recom-
mended or required in the relevant international standards. Partial-Discharge (PD) test

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 993–1002, 2020.
https://doi.org/10.1007/978-3-030-31676-1_93
994 S. Zheng et al.

is also a widely used insulation detection method [3–5]. The whole reactor lie in high
voltage area, so only Ultra-High-Frequency (UHF) method can be used for detecting
PD in reactor in operation. But the turn-to-turn clearance is no more than 1 mm,
PD UHF signals may be shielded.
To explore the feasibility of UHF method for detecting PDs in smoothing reactors,
the PD tests are carried out on a 10 kV reactor, in which the UHF signal waveforms are
collected and observed. To reveal the UHF electro-magnetic wave propagation, some
simulations have been done by using FDTD method, and the electromagnetic wave
features are analyzed.

2 PD Tests

The PD tests are carried out on a 10 kV dry-type reactor. A PD model is inserted in the
cooling channel between two adjacent coils. The UHF antennas are set on two typical
positions to collect PD signals come out from different path.

2.1 Dry-Type Reactor and PD Model

The 10 kV reactor used in the PD test is shown in Fig. 1. The thickness of the coil is
about 1.5 cm. the thickness of the polyimide film around the coil conductor is about
0.25 mm, that means the turn-to-turn distance of two adjacent conductors is only
0.5 mm. The width of the cooling channels is about 1.9 cm. The out diameter and the
inner diameter of the reactor is 108.0 cm and 84.0 cm. The height of the reactor is
50.5 cm.
PD model is set in one of the cooling channels. The PD model is made up of high
voltage line, floating mental needle, and grounded line and reactor. As shown in Fig. 2.
The distance between floating mental needle and high voltage line is about 3 mm. The
distance between floating mental needle and grounded line is 15 mm. When the high
voltage on the line reach up to 5 kV, PD signals appear.

High voltage
PD model line

1.9 Floating
84.0 reactor mental needle
1.5
108.0
Grounded line

reactor

Fig. 1. The 10 kV reactor used in the PD test. Fig. 2. The PD model set in the reactor.
 Feasibility of Detecting Partial Discharge in Smoothing Reactor 995

2.2 UHF Antennas and Detection Points

Bow tie UHF antenna is chosen as PD sensor for coupling UHF electro-magnetic
waves. The largest detection gain will be achieved when the polarization direction of
antenna is consistent with that of electro-magnetic waves. The designed frequency band
of antenna is 1 GHz–5 GHz.
The antennas are arranged at two points, one is beside the reactor, and the other is at
the bottom of the reactor. The arrangements and the distance between PD model and
detection points are shown in Fig. 3.

0.97m
Point1 PD

Reactor

1.18m

Point 2

Fig. 3. The arrangements of the UHF antennas.

At each detection points, two antennas are set according to different polarization
direction. By comparing the amplitudes and arrival times of UHF waveforms collected
by different antennas, the UHF propagation features can be obtained.
UHF signal amplifiers are used to magnified UHF signals. Their frequency band is
250 MHz–1.5 GHz, and gain is about 30 dB. Oscilloscope is used to collect UHF
signals, the frequency band is 0–4 GHz, and the sampling rate is 20 GS/s.

2.3 UHF Signals and Comparative Analysis

(1) Comparison of UHF signals coupled by different polarization direction

antennas
Signals coupled by antennas at point 1:
Two antennas are set on point 1, one’s polarization direction is along the vertical
direction, and the other’s polarization direction is along the horizontal direction.
The antennas’ polarization directions are shown in Fig. 4.
The UHF signals coupled by the two polarization direction antennas at point 1 are
shown in Fig. 5. The arrival times of the two signals are all −7.2 ns. The
amplitudes of the two whole waveforms are about 10 mV. However, the ampli-
tudes of the first waveform are different. The amplitude of vertical polarization
direction antenna is larger than the one of horizontal direction antenna. The
former is 3.2 mV, and the latter is 2.4 mV.
996 S. Zheng et al.

Signals coupled by antennas at point 2:

Two antennas are set on point 2, one polarization direction is along the radial
direction, and the other polarization direction is along the normal direction. The
antennas’ polarization directions are shown in Fig. 6.

Fig. 4. The antennas set on point 1 and their polarization directions.

10 -7.2ns -3.2mV
Amplitude(mV)

-10

-20 -10 0 10 20
Time (ns)

(a) By the vertical polarization direction antenna.

10
-7.2ns -2.4mV
Amplitude(mV)

-10

-20 -10 0 10 20
Time (ns)

(b) By the horizontal polarization direction antenna.

Fig. 5. UHF signals coupled by the two polarization direction antennas at point 1.
 Feasibility of Detecting Partial Discharge in Smoothing Reactor 997

Fig. 6. The antennas set on point 2 and their polarization directions.

-7.4ns 3.7mV
10
Amplitude(mV)

-10

-20 -10 0 10 20
Time (ns)

(a) By the radial polarization direction antenna.

10
Amplitude(mV)

-6.6ns 4.9mV
-10

-20 -10 0 10 20
Time (ns)

(b) By the normal polarization direction antenna.

Fig. 7. UHF signals coupled by the two polarization direction antennas at point 2.

The UHF signals coupled by the two polarization direction antennas at point 2 are
shown in Fig. 7. The amplitudes of the two whole waveforms are about 15 mV.
However, the arrival times of two signals is different. It seems that the radial
polarization antenna get the UHF signals earlier. The difference of arrival times is
0.8 ns.
998 S. Zheng et al.

(2) Comparison of UHF signals coupled by antennas arranged at different

positions
As shown Fig. 1, two antennas are set near to the reactor for coupling UHF
signals. At point 1, the polarization direction of the UHF antenna is vertical
direction. At point 2, the polarization direction of the UHF antenna is radial
direction. With these polarization directions, the earlier arrival times and largest
first waveform amplitudes can be obtained at the two points.
Figure 8 is the waveforms got from the two points. The amplitudes of the whole
waveforms are all about 10 mV. However, the amplitudes of first waveforms are
different. The amplitude at point 1 is 2.4 times (3.4 mV/1.4 mV) of point 2. The
arrival time difference is 1.2 ns ((−9.0)−(−10.2)), that is larger than the theoretical
value (0.7 ns = (1.18−0.97)m/(0.3 m/ns)).

10 -10.2ns 3.4mV
Amplitude(mV)

-5

-10
-20 -10 0 10 20
Time (ns)

(a) UHF signals got in point 1.

15
-9.0ns .4mV
Amplitude(mV)

10
5
0
-5
-10
-20 -10 0 10 20
Time (ns)

(b) UHF signals got in point 2.

Fig. 8. UHF signals coupled by the two antennas set at point 1 and 2.

3 UHF Signal Propagation

3.1 Electro-Magnetic Wave Polarization
From the comparison of the amplitudes of the first UHF waveforms got from point 1,
the electric field components that is perpendicular to the crack between two mental
wires is always larger than the other components. That is consistent with planar
waveguide theory.
 Feasibility of Detecting Partial Discharge in Smoothing Reactor 999

The electric lines of power in the planar waveguide and those propagating out of
the planar waveguide are shown in Fig. 9. The directions of electric field vector are
always perpendicular to the crack of the two mental planar.

(a) Front view. (b) Side view.

Mental Planar E H

Fig. 9. The electric line of force in planar waveguide.

From the comparison of the amplitudes of the first UHF waveforms got from point
2, the electric field components that is along radial direction of the cylindrical reactor.
That is consistent with coaxial line waveguide theory.
The electric lines of power in the coaxial line waveguide and those propagating out
of the planar waveguide are shown in Fig. 10. The directions of electric field vector are
always along the radial direction of the coaxial line.
On the other hands, when the electro-magnetic waves go out of the waveguide, they
will overlap or mix with each other in every observation point, so, the polarization
direction will be complex.

(a) TEM. (b) TE11.

Coaxial Conductor E H

Fig. 10. The electric line of force in the coaxial line waveguide.
1000 S. Zheng et al.

3.2 Propagation Simulations

Simulation Method and Model
The software named XFDTD is used to simulate the electro-magnetic wave propaga-
tion. The 3-D reactor model is shown in Fig. 11. There are four coils in the model.
Every spiral coil is made from one conductor, whose cross section is square. The length
of the square side is 10 mm. The square conductor is packed by polyimide film, the
thickness of which is 0.25 mm. It means that the width of the crack between every two
adjacent conductor lines is 0.5 mm. The relative dielectric constant er is 4.0. An
insulation cylinder is set inside each coil, the thickness of which is 2.0 mm, and the
dielectric constant er is also 4.0. The width of the cooling channels between inner side
of the insulation cylinder and outside of the coils are 19 mm.

Fig. 11. The electric line of force in the coaxial line waveguide.

A current element is set as a PD source, which is arranged at the middle height of the
model, and between the two outside coils. The waveform of the current is Guess pulse
shape, the amplitude of which is 1A, the width of pulse is 1 ns. The simulation space is
divided into several cuboids. The dimension of every cuboids is 1  1  0.5 mm3.
Propagation and Arrival Times
The propagation of electro-magnetic wave is presented by several graphs of electric
field value in the inspection plane at different time, which are shown at Fig. 12.
Judging from Fig. 12(b), the electro-magnetic wave can pass through the polyimide
film filled in cracks of the coils, and go out in the space. The electro-magnetic wave can
also go out from the top and bottom of the reactor, as shown in Fig. 12(c). Spending
same time, the electro-magnetic can go longer distance in the side path than in the end
path, which is shown in Fig. 12(d), (e). That is to say, the arrival time will be a little
later if the observation point is under the reactor. This phenomenon is caused by the
reflection and polarization conversion of electro-magnetic wave. This phenomenon
confirms the arrival time difference between point 1 and point 2 shown in Fig. 8.
 Feasibility of Detecting Partial Discharge in Smoothing Reactor 1001

(a) t=0.15ns (b) t=0.74ns (c) t=1.34ns

(d) t=1.93ns (e) t=2.67ns (f) t=3.71ns

Fig. 12. The 3-D reactor model for simulation.

4 Conclusions

In the PD test, UHF electro-magnetic wave can go out of the reactor and be coupled by
UHF antennas. The propagate path will affect the polarization directors, especially for
the first wave of the UHF signal. For the reflection and polarization conversion, the
arrival time of the first wave will be a little later when observation point is under the
reactor.
These conclusions are good for improving UHF detection sensitivity and location
precision. On the one hand, antenna’s polarization direction should be along the electric
component of the electromagnetic waves to get higher sensitivity. On the other hand,
electro-magnetic waves going out from the similar path should be coupled for
improving PD location precision.

References
1. State Grid Corporation of DC Engineering Network Alliance: Pro-Curement Specifications of
Equipment of ±1100 kV DC Transmission Project from Zhundong Region: Technical
Specification of Dry-Type Smoothing Reactor. State Grid Corporation of DC Engineering
Network Alliance, Beijing (2014)
2. Zhang, M., Guojing, W., Yuehua, Z., Bin, H., Jingliang, X., Sun, W.: Insulation design and
test analysis of ±1100 kV UHV dry-type smoothing reactor. High Voltage Eng. 41(5), 1760–
1766 (2015)
1002 S. Zheng et al.

3. CIGRE WG D1.33: Guidelines for Unconventional Partial Discharge Measurements, France

(2010)
4. Judd, M.D., Yang, L., Hunter, I.B.B.: Partial discharge monitoring for power transformers
using UHF sensors Part 2: field experience. IEEE Insul. Mag. 21(2), 5–13 (2005)
5. Zheng, S., Li, C., Tang, Z., Chang, W., He, M.: Location of PDs inside transformer windings
using UHF methods. IEEE Trans. Dielectr. Electr. Insul. 21(1), 386–393 (2014)
 Speciality and Applicability of Advanced
Response Methods in Contrast
to the Traditional Dielectric Measurements for
Condition Assessment of Power Transformers

Richárd Cselkó(&), Máté Szirtes, and Gusztáv Csépes

Budapest University of Technology and Economics,

Egry J. u. 18, Budapest 1111, Hungary
cselko.richard@vet.bme.hu

Abstract. For oil-paper insulated transformers, an accurate assessment of

condition (moistening and ageing), evaluation of the service life and determi-
nation of the highest permissible operation temperature are complex tasks. The
inhomogeneous distribution of moisture, ageing by-products and temperature
make the assessment even more difficult. The existing evaluation standards are
based on traditional long-time established tests which characterize this com-
plicated and inhomogeneous insulation with a single parameter. In order to carry
out efficient condition assessment, extend transformer lifetime and increase load
capacity, it is necessary to apply powerful and reliable diagnostic tools. The
advanced dielectric methods (RVM, FDS, and PDC) characterize the insulation
with a spectrum, consequently they are able to provide more precise condition
assessment. Despite the fact that these methods are particularly useful for
transformers, an overall guide or standard that describe their application is not
available yet. This paper presents the applicability and advantages of the
advanced techniques compared to the traditional dielectric measurements.

Keywords: Condition assessment
Transformer diagnostics
RVM
FDS

PDC
Response methods

1 Introduction

The life of an oil/paper transformer is mostly influenced by the condition of solid

insulation (cellulose). Insulation degradation is more significant at higher temperatures
and moisture in the presence of acidity by-products. For these reasons, transformer life
management gained an ever-increasing interest over the past decade, due to both
economic and technical reasons. The condition of oil can be determined by sampling
and analyzing using laboratory chemical-physical tests. In contrast, the determination
of cellulose condition is not too easy because the sampling is quite expensive and
difficult. Moisture and ageing by-products in oil-paper insulation cause three dangerous
effects: they decrease the dielectric strength, accelerate cellulose aging and may gen-
erate bubbles in the oil at operating temperatures [10, 11].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1003–1014, 2020.
https://doi.org/10.1007/978-3-030-31676-1_94
1004 R. Cselkó et al.

Many classical methods (absorption factor, loss tangent, etc.) have been used for
many decades as testing of solid oil impregnated paper insulations. There are several
guides and standards that describe transformer diagnostics using classical methods.
These traditional methods using a single parameter, are not enough to characterize the
non-uniform insulation of transformers. More advanced tools, however, are able to
provide information on multiple failure modes (moistening, ageing, bubble generation,
etc.) and any indication of the type of action needed (refurbishment, oil regeneration,
replacement, etc.) but standards or guides are missing for these techniques [1–5]. At the
Budapest University of Technology and Economics (BME), in the middle of 1970’s,
among other things, the following parameters have been investigated during a large
research program [1–4]: return voltage measurements (RVM), tgd and capacitance C in
the dependency of (low) frequency (presently called “frequency domain spectroscopy -
FDS), and charge/discharge current (measurements of polarization and depolarization
currents - PDC). These methods are equivalents and all the three compared methods
reflect the same fundamental polarization and conduction phenomena in transformer
insulation. All data, which were measured with these methods, can be converted into
each other [1–5, 7–9, 12, 13].
The interpretation of data is considerably sophisticated, this is why this paper tries
to present the special features and the applicability of these advanced condition
assessment techniques. Without going into details here, we would like to clarify some
anomalies and show a possible interpretation of measured data with return voltage
technique (RVM). As it is well known, many asset managers currently use some kind
of transformer assessment indices (e.g. Health Index). That is why it should quantify
the advanced techniques, where possible.

2 Some Oil-Paper Insulation Fundamentals

The insulation system of power transformers consists of oil and paper/pressboard. The
oil is providing both the electrical insulation and the means for transferring the thermal
losses to the cooling system. The life of a transformer is mostly influenced by the
condition of its solid insulation. Knowing the literature, it can be said that there are
numerous misunderstandings, so we would like to emphasize the most important
fundamentals.

2.1 Some Words About the Oil Reclamation

Knowledge about the actual moistening and ageing in the cellulose provides essential
information about the condition of the insulation. Fortunately, there are several
maintenance actions that can be performed on transformer oil, e.g. the oil reclamation
[11]. The oil reclamation is a very powerful process, but we must call the attention to
the followings. Although oil reclamation is able to remove ageing by-products from the
oil, these products are partitioning between oil and paper, in the same way as water is
partitioned between oil and paper, the greater part is within the solid insulation. The
main goal of the “oil reclamation” is to remove water and acids from the “cellulose
insulation”. By the standard, before, during and after the reclamation process, the
 Speciality and Applicability of Advanced Response Methods 1005

laboratory oil test is recommended. If the oil indicates good condition but the cellulose
contains high amount of ageing by-products, when reaching equilibrium, the oil
reabsorbs the moisture and acids. It can be said, that the oil test alone is not enough to
check the efficiency of reclamation processes. Before and after the reclamation process,
RVM is a very powerful tool for checking the efficiency of oil reclamation [11].

3 Some Dielectric Fundamentals

The two key dielectric phenomena in the insulation of transformers are conduction, and
polarization, which are characterized by the conductivity and the spectrum of polar-
ization. Both of these processes strongly depend on the physical structure of the oil-
paper insulation. The polarization is a resultant of several elementary processes of
different intensities and relaxation times [1–5, 7–9]. If there is any deterioration caused
change in the physical insulation structure, that is directly reflected in the changing
values of the insulation dielectric parameters, as well.

3.1 Briefly About the Interfacial Polarization

Dielectric diagnostic methods measure the interfacial polarization effect, which origi-
nates from cellulose and the interfaces between cellulose and oil. Important remark:
interfacial polarization can also develop purely in the cellulose but in the oil the
interfacial polarization is insignificant. If the solid paper insulation is measured without
oil (“air impregnated cellulose” or oil free condition), a dominant interfacial polar-
ization can be detected. The oil is characterized by the conductivity, but the oil con-
ductivity determined by the condition of cellulose. If we would like to use the
conductivity of oil for diagnostics purpose, we can realize that it has only one fix value,
this alone is not able to characterize tons of inhomogeneous insulation.
For condition assessment of cellulose, we have to investigate interfacial polariza-
tion. The polarization is characterized by the time constant and related intensity
(Fig. 1). The time constant range of interfacial polarization spectrum in the impreg-
nated cellulose extends from 103 s up to 10 þ 4 s and it can be investigated by low
frequency AC (FDS) or DC (RVM, PDC) methods. Ageing and moistening processes
cause chemical change of certain molecule groups. The elementary polarizations,
connected to this group, show a characteristic alteration in its time constants and its
intensity. Consequently, each kind of the deterioration processes has its own special
effect to the polarization and thereby characteristically alters the range of long time
constants, too [1–3, 5, 9]. It is important to note that during the low frequency mea-
surements of composite oil-paper insulation, the results are mostly connected to the
solid impregnated paper.
It is known that the water is distributed between oil and paper in such way, that the
water content of the oil strongly depends on the water content of the paper. It is
important to note that these parameters are not independent, because law governs the
partition of water, acids, etc. During application of the “spectrum of polarization”, we
1006 R. Cselkó et al.

must interpret the spectrum, by which the polarization is characterized, meaning that
the time constant values and the relative magnitudes must be determined. There are
some misunderstandings that must be clarified regarding these parameters [1–3, 5, 9].

3.2 Feature of the Dielectric Loss

When applying the RVM methods the dielectric loss plays a significant role in inter-
pretation. The measured dielectric loss at a specific frequency can be interpreted as the
sum of elementary losses resulting from discrete processes. The behavior of the
dielectric loss value (tgd) strongly depends on the frequency. The frequency domain
distribution of the dielectric loss and the relative permittivity (er) has characteristic
frequencies caused by specific polarization mechanisms, as Fig. 1 illustrates.

Fig. 1. The frequency dependence of dielectric loss and relative permittivity

The dielectric loss values are usually concentrated around the characteristic fre-
quency of a polarization mechanism, showing a peak value at the exact proximity of
this frequency. The alterations of the different time-constant-ranges of the polarization
spectrum represent different types of moistening and ageing processes or an inhomo-
geneous condition. The long-time constant range of the polarization spectrum can give
a lot of valuable information about the deteriorating processes and the momentary
condition of insulation [1]. The intensity of ageing and moistening is “distributed”,
with other words they have inhomogeneous distribution, and therefore, the resulting
polarization spectrum is also distributed. The dielectric loss factor is much higher for
the aged and moist paper than for the new one. Figure 2 shows that the initiation
temperature of bubble formation is lower for moist and aged oil-paper insulation [6].
 Speciality and Applicability of Advanced Response Methods 1007

3.3 Loadability and Bubbling Effect

When wet paper is immersed and heated in oil, as the temperature rises, the water starts
to evaporate, leading to the formation of bubbles. As the gas bubbles have a much
lower dielectric strength compared to oil, partial discharge and breakdowns are likely to
occur in electrically stressed parts of the insulation [6].

Fig. 2. Initiation temperature of the bubble effect

For this reason, solid insulation must be analyzed on a regular basis to verify that its
moistening and ageing is within safe limits. When the transformer is new and dry, this
condition permits operation at higher temperatures without risk of the formation of
water vapor bubbles. IEC 60354 “Loading Guide for oil immersed transformers”
concedes a hot spot temperature for a short-term overload of transformers. In this case
the operator has to assess ageing and moistening condition and evaluate the risk of
bubble formation. Bubble formation occurs when the temperature of the insulation
reaches a limit. Figure 2 shows that the initiation temperature of the bubble effect is
dependent on moistening and ageing processes. It is apparent that the initiation tem-
perature of the bubble effect is lower for the aged paper than for the new one. It was
1008 R. Cselkó et al.

found that dielectric loss increases the local temperature of the insulation. This means
that it is essential to know the tgd dependence of the basic parameters of oil-paper
insulation [1, 2, 6]. Investigating the dielectric loss, it can be seen that the time constant
values strongly depend on temperature, water and ageing by-products and they are
shifting close to the power frequency as the condition gets worse. Near the time
constant of the network frequency, the value of dielectric loss steeply grows, showing a
strong correlation with temperature, moisture and ageing by-products (see Figs. 4 and
5). This leads to rising local temperatures in the insulation that can lead to the initiation
of the bubble effect if a critical temperature is reached [6]. Surveying the basic con-
nections between moistening, ageing and the bubble effect initiation temperature, we
are able to clarify the limitation of overload operation. That is why the RVM technique
could provide efficient management for moistening, ageing and overloading, by
checking the efficiency of oil reclamation and estimate life expectation.

4 Features of RVM Technique

As up to now, there are no internationally standardized procedures for spectrum

methods. Analyses are done by experts but run the risk of being judged with subjec-
tivity. Thus, the need for an algorithm enabling quantitative determination of polar-
ization spectra (measured by RVM) within a certain frequency range is obvious. This
article tries to present an algorithm to assess the condition of oil-paper insulation with
RVM. The result of this algorithm is verified using database obtained during the
mentioned Hungarian research work. The efficiency and sensitivity of this algorithm
and interpretation of measured data is of great importance compared to the conven-
tional methods.

4.1 The Main Feature of Spectrum Methods

Using the RVM method correctly, it is able to estimate the probability of failure and
give acting works to correct transformers in poor condition. One insufficient problem is
the interpretation of spectrum methods and understanding the meaning of the tem-
perature or moisture content when we know that the distributions of these parameters
are inhomogeneous. While the conventional methods provide a single parameter of
insulation, the spectrum method provides a “spectrum” with lots of information,
therefore it is necessary to have an advanced interpretation, which helps determining
the detailed condition of the insulation. We also have to provide information on the rate
of degradation and the optimal preventive or corrective actions.

4.2 Measurement of RVM Curves

The fundamental behavior determined by the dielectrics cellulose and oil is charac-
terized in the equivalent circuit by the numerical values of model RC series and parallel
elements (Fig. 3) [1–3, 5, 7–9]. Regarding the RVM curve, we are able to record which
polarization at which time constants are active. Two parameters of the return voltage
are measured during RVM insulation examination, namely the maximum voltage value
 Speciality and Applicability of Advanced Response Methods 1009

(Vr), which is proportional to the polarizability (a), and the initial slope of the signal
(Sr), which correlates with the polarization conductivity (b). An important remark: we
can use both RVM curves: maximum voltage and initial slope curve but the maximum
voltage curves much clear and spectacular solution. First data of initial slope is in direct
proportion to tgd measured in this time constant range. The RVM value measured at
20/10 ms, is in direct proportion to loss factor measured at 50 Hz. RVM, also called
“quasi” polarization spectrum methods, can be used as a diagnostic method to detect
dielectric defects in oil-paper insulation.

Fig. 3. Equivalent insulation circuit and RVM curve

4.3 The Main Goal of the RVM Technique

The goal is to provide efficient assessment about moistening and ageing phenomena,
the operation limits, and the life expectation. Similarly, information must be provided
about the rate of degradation and the optimal preventive or corrective actions. By
knowing the fundamentals of advanced dielectric methods, it is obvious that these
techniques are able to provide the required diagnostics. One of the possible interpre-
tations of RVM curve is if we make a comparison between the measured RVM curves
and the reference curves in the Hungarian database. We also have to be in possession of
the possible reference curve and an evaluating algorithm. The moisture and ageing
analysis in this approach is based on the comparison of onsite measurements to a
modelled dielectric response. This data pool consists of dielectric responses of new and
various aged pressboard samples measured at various temperatures and moisture
contents. Therefore, a reliable moistening and ageing analysis can be done even for
moist and aged oil-paper-insulations. The other possible interpretation of spectrum
methods with the RVM technique is based on the bubble formation which is influenced
by dielectric losses.
1010 R. Cselkó et al.

4.4 Further RVM Fundamentals

The three methods are equivalent [5, 7–9] and it is possible to convert data into each
other. Comparing the three spectrum methods it can be stated that the RVM technique
provides the most valuable solution [4]. To analyze the data, time constants and
intensities must be calculated. It can be seen that RVM curves directly indicate the time
constants. Measuring techniques and presentation of data are different but RVM pro-
vide such curves that the results are “visible to the naked eye” and the interpretation is
the most convenient.
For skeptical experts it can be repeated that a software technique is available that
converts polarization data between the three spectrum methods. The Hungarian data
pool [2] shows a strong correlation between the curves of initial slope and the tgd
curves in frequency range. With other words, the initial slope curves getting from the
RVM measurements can be used as tgd in frequency range.
In order to carry out an efficient condition assessment we have to use the polar-
ization spectrum methods combined with physicochemical oil analysis. Considering
the ageing by-products, we can say that its influence is similar to the moistening
phenomenon, but the temperature dependence differs from it (lower). In this case we
can use the reference curves from Hungarian research work (data pool).

4.5 Specialities of Advanced Interpretation

For the correct interpretation of “polarization” we have to apply “polarization assess-
ment” and use the whole spectrum with several time constants. Sometimes it can be
seen that some experts use only a single time constant in spite of the fact that the given
spectrum is in possession of more than one time constant. Moreover, it is surprising
that the RVM is accused of “simplistic interpretation”. We must make it absolutely
clear that to correctly interpret the RVM data, we have to use the whole spectrum and
we have to use the database getting from the Hungarian research work.

4.6 Possible Interpretation of the RVM Data

As mentioned before, the time constant values strongly depend on temperature, water
and ageing by-products. If the condition is good, the time constants are over 1000 s, if
the condition is bad, they are below 1 s at room temperature. With deterioration of
insulation, the time constants are shifting to the power frequency. The network fre-
quency is 50 Hz and its time constant is Tm = 1/x  3.3 ms (Figs. 4 and 5). In our
case, for the calculation of the dielectric loss we have to take the network frequency
into consideration. For this reason, the value of the dielectric loss has to be investigated
in the frequency range of Fig. 4. Near the power frequency time constant (Tm), the
dielectric loss shows a significant growth (Fig. 5). If the material is in good condition,
it will have time constants far above Tm, in the order of 1000 s. Figure 4 shows the
fundamental polarization loss distribution as the function of the time constant. The
figure shows that the basic processes of time constants ranging from Tl to Tu (0,1/x –
10/x) are causing the resulting loss [1, 6].
 Speciality and Applicability of Advanced Response Methods 1011

Fig. 4. Time constant range around 50 Hz

Initiation temperature of the bubble effect can be seen in Fig. 2. Using the RVM
curve we are able to determine the value of the lowest time constant and we can
calculate that the lowest time constant where is from 3.3 ms. Knowing the value of
lowest time constant at a given temperature, the risk of bubbling can be estimated
(Figs. 4 and 5). As it can be seen in the Fig. 5, if the time constant at operating
temperature is nearer than two orders to the Tm, after a certain time, in consequence of
the bubble generation, the breakdown is bound to happen. If we are able to operate the
transformer at lower temperature, the bubble evolution will not occur. With other
words, although the transformer has not too good condition, it can be operated at lower
temperature. With other words, we have flexibility in operation the transformer and
considering the bubble generation the risk of failure is low or still acceptable.

Fig. 5. RVM spectrum after breakdown [8]

Regarding transformer reliability, it is possible to determine the safe operation

limits of the equipment, namely the limits of moisture, temperature and by-products of
ageing. These limits, at which the transformer can operate safely, can be determined
using the curves of the bubbling-effect and the RVM method. For further clarification,
Fig. 5 shows the results of a transformer that has broken down at 60 °C temperature
during service. The curve of Fig. 5 was measured at room temperature (21 °C) after the
1012 R. Cselkó et al.

breakdown. In the paper, at 21 °C the lower dominant time constant of 0.3 s refers to
around 5% of “equivalent moisture content”. Using the reference curves or the
physicochemical test of oil we are to diagnose the origin of deterioration and decide
about the repairing action. The temperature dependency of these time constants show
that at higher temperatures they are shifting to lower values, for example in case of the
results of Fig. 5, at the temperature of 60 °C the lowest time constant value moved
down to 3 ms.
Regarding Figs. 4 and 5, we can see that the losses usually have a peak value
around a characteristic frequency of a specific type of polarization.
To avoid premature ageing of the cellulose insulation, the equipment’s water and
acid content must be kept at a low level. If the transformer has a significant water and
acid content, it is dangerous to operate the unit at rated power, drying and oil regen-
eration may retard ageing of the solid insulation [10]. Until the oil regeneration action
we have to decrease the temperature (the load) where the operation of the given
transformer is reliable. In order to estimate the temperature where the operation is
reliable, we have to determine the whole RVM spectrum and each time constants.
Knowing the lowest time constants (namely the real ageing and moistening condition)
we will be able to estimate the permissible temperature where the operation is reliable.
As it was mentioned that the distribution of temperature, moisture and ageing content is
not homogeneous, really it is a difficult case to speak a fix temperature, at most about
average value. For us the most important is to know the lowest time constants, which
can cause local temperature rise, consequently bubbling formation, and in the long run,
breakdown. With other words, we have to know the acceptable limits of moistening,
ageing and the bubbling temperature, then determining the limit temperature at which
the operation is safe [1–4, 6–9] or which repairing activities had to carry out in the long
run.

4.7 Summarizing of Applicability of RVM

Based on the RVM data pool and considering the bubble effect, it is possible to assess
the highest operation temperature of an oil-paper transformer where the moisture and
ageing by-products in the oil-paper insulation surpass a given limit. Using the RVM
technique it can be indicating that damages can be occurred during the transformer
operating for long periods of time at given loading condition. But if we decrease the
temperature the reliability is acceptable and later it is possible to switch off the unit to
improve the condition of insulation (e.g. oil regeneration).
Now it is demonstrated that the interpretation of spectrum methods (including the
RVM) is relatively easy. Otherwise the interpretation of data of spectrum methods is
more complicated than it was experienced in the classical methods. If the time con-
stants are around 3 ms, or key polarization processes occur between the time constants
of 3 ms and 300 ms, that means that the equipment is in a dangerous condition phase.
These polarization mechanisms can highly contribute to the 50 Hz dielectric loss
values of the insulation. Due to the steeply growing dielectric loss values, the local
temperature also starts to rise, which can cause the formation of the bubble effect.
By examining Fig. 5, we can presume that regarding the insulation breakdown, an
important role can be given to the increased levels of dielectric loss, and also the
 Speciality and Applicability of Advanced Response Methods 1013

created bubbles and the forming partial discharges. With all this in mind, a general rule
can be stated: for the sake of security and reliability, at the highest possible temperature
of operation the smallest time constant values need to be above 300 ms.

5 Conclusions

The standards are missing that is why a greater part of the experts are not able easily to
take full advantage of an advanced method. One of the reasons for that has been the
fact that the results of such methods cannot be interpreted confidently because the exact
relationship between the measured dielectric parameters and the fundamental dielectric
processes of insulations is not yet clearly understood. It is obvious that the conventional
diagnostics with a single parameter are not able to provide the required condition
assessment. Nevertheless, the dielectric spectroscopy tests characterize the insulation
with a spectrum, which is why they are able to provide more precise assessment of the
condition of the insulation. In this paper the focus is directed on the most important
aspects of mechanism of cellulose ageing and dielectric measurement (missing link),
with special regard to return (recovery) voltage measurement technique.

References
1. Nemeth, B., Csépes, G., Cselko, R.: Applicability of the dielectric response methods for
diagnostics of power transformers by application of return voltage measurement. In:
IEEE DEIS ICD BP 2018, Dialogue session 3/43 (2018)
2. Csépes, G., et al.: Dielectric response methods for diagnostics of power transformers –
Hungarian research work in the mid-1970 s. In: Paper D1-242, CIGRE SC-D1–Colloquium
in Hungary Budapest (2009)
3. Bognár, A., et al.: Diagnostic tests of high voltage oil-paper insulating systems (in particular
transformers insulation) using DC dielectrometrics. In: Proceedings of the 1990 CIGRE
Conference, Paris, France, 15/33-08 (1990)
4. Bognár, A., et al.: Comparing various methods for the dielectric diagnostics of oli-paper
insulation systems in the range of low-frequencies or long time constants. In: 8th ISH,
Yokohama, Japan (1993)
5. CIGRE Technical Brochure 254: Dielectric response methods for power transformers. In: TF
D1.01.09 (conv. S. Gubanski), Paris (2004)
6. Przybylek, P., et al.: Bubble effect in transformer insulation caused by dielectric losses. In:
15th ISH, Ljubljana, Slovenia (2007)
7. Qin, S., et al.: Study on return voltage measurement of oil-paper insulation testing
technology in transformers. In: 19th IEEE International Conference on Dielectric Liquids
(ICDL), Manchester (2017)
8. Shi-ling, Z.: The simulation analysis of transformer recovery voltage by field and circuit
method based on PSO algorithm. In: 12th IEEE International Conference on the Properties
and Applications of Dielectric Materials, Xi’an (2018)
9. Saha, T.K.: Deriving an equivalent circuit of transformers insulation for understanding the
dielectric response measurements. IEEE Trans. Power Delivery 20(1), 149–157 (2005)
10. CIGRE TB 349: Moisture equilibrium and moisture migration within transformer insulation
systems. WG A2.30, Paris (2008)
1014 R. Cselkó et al.

11. CIGRE TB 323: Ageing of cellulose in mineral-oil insulated transformers. TF D1.01.10,

Paris (2007)
12. Szirmai, Á., Tamus, Z.Á.: Modelling of dielectric processes in oil-paper insulation for
replacement of return voltage measurement. In: 2016 Conference on Diagnostics in
Electrical Engineering (Diagnostika), pp. 1–4. IEEE (September 2016)
13. Tamus, Z.Á., Csábi, D., Csányi, G.M.: Characterization of dielectric materials by the
extension of voltage response method. J. Phys. Conf. Ser. 646(1), 012043 (2015)
 Simulation of Void Discharges
and Time Domain Examination
of Their Emitted Electromagnetic Signals

Máté Szirtes(&) and Richárd Cselkó

Group of High Voltage Technology and Equipment,

Budapest University of Technology and Economics, Budapest, Hungary
mate.szirtes@gmail.com

Abstract. The goal of this paper is to present the time domain behavior of the
emitted electromagnetic signals of void discharges and to show the design
process of the measured partial discharge (PD) sources. After the introduction of
the used physical models and equations, the finite element calculations used to
create a modified empirical ignition criterion for the discharge and to determine
the optimal dimensions of the samples are presented. Using these sizes, two void
discharge sources were created by including different diameter air bubbles in a
thin layer of epoxy resin between two electrodes. After this, the emitted elec-
tromagnetic signals of the sources were measured using a noise-free GTEM cell
and a digital sampling oscilloscope. From the measured signals, PRPD patterns
have been created to examine the activity of the emitted void discharge signals
as the function of the phase angle of the AC testing voltage. This was followed
by the examination of the effect of bubble size on the PRPD characteristics. The
results of the emitted measurements were consistent with the conducted mea-
surement results of the literature and the partial discharge physics based model
of the air-filled cavity.

Keywords: Insulation diagnostics
Partial discharge
Void discharge
Finite

element simulation
GTEM cell
PRPD pattern

1 Introduction

Partial discharge measurements have an important role in the diagnostics of high

voltage equipment. Later occurring malfunctions can be prevented by successfully
detecting, finding and correcting the faulty part in which the discharge is present. It is
possible to determine the spatial position of the discharge by measuring and correctly
processing its emitted electromagnetic signals. To increase the effectiveness of these
promising methods, further examination of the emitted signals is needed. For proper
examination, sensitive measuring instruments may be used, which require limited
testing voltages compared to the general tens of kilovolts. While decreasing the volt-
age, the size of the measured discharge source must also be reduced. This requires
accurate dimensioning and precise design for the source to work properly. For this,
proper understanding of void discharge physics and thorough simulations are needed.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1015–1024, 2020.
https://doi.org/10.1007/978-3-030-31676-1_95
1016 M. Szirtes and R. Cselkó

2 Design Process of the Discharge Sources

2.1 Basics of Void Discharge Physics
Void discharges may occur in small, gas-filled cavities of solid insulating materials or
in gas bubbles of liquid insulating materials. The simplest experimental arrangement of
the void discharge is a spherical air bubble in the middle of a solid insulating material
placed in a homogeneous electric field. The relative permittivity and dielectric strength
of solid insulating materials are several times larger than the relative permittivity and
dielectric strength of air. In case of DC or 50 Hz AC testing voltages, the electric field
distribution of the model is given as the solution of an electrostatic problem. In this
case, the strength of the external homogeneous electric field is multiplied inside the
bubble as Eq. (1) shows [1, 2]. According to this model, the electric field inside a
spherical air bubble is approximated to be homogeneous and independent of the cavity
size. It should be noted that this approximation is only correct if the diameter of the
cavity is not close to the thickness of the insulating material. If this criterion is not
satisfied, the curvature of the void wall starts to distort the otherwise homogeneous
electric field inside the air bubble.

3
er;solid
Evoid;air ¼
Eext ð1Þ
1 þ 2
er;solid

Eext ¼U=d

Where:
– Evoid; air = Homogeneous electric field strength inside air bubble [kVrms/cm]
– er; solid = Relative permittivity of solid dielectric material [ ]
– Eext = Homogeneous electric field strength inside the solid dielectric [kVrms/cm]
– U = AC testing voltage between electrodes [kVrms]
– d = Electrode distance [cm]
Due to the higher electric field and the lower dielectric strength of air inside the
bubble, the ignition of a void discharge occurs. It must be taken into account that while
the electric field strength inside the void is approximately independent of its size, the
discharge ignition field strength increases sharply with the reduction of bubble size.
This ignition field strength value can be calculated from the empirical expression of
Eq. (2) which assumes normal atmospheric pressure and room temperature inside the
air bubble [2–4].
 
8:6
Eignit; void ¼ 24:2
1 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð2Þ
D
100

Where:
– Eignit; void = Ignition electric field strength of void discharge [kV/cm]
– D = Diameter of spherical air bubble [mm]
 Simulation of Void Discharges and Time Domain Examination 1017

By connecting 50 Hz AC test voltage to the sample, the field strength begins to

increase in the cavity proportionally to the voltage, until the ignition value is reached.
When this happens, a streamer discharge occurs if a required initiatory electron is
available. This discharge transfers a packet of charge carriers to the opposing wall of
the bubble, creating an electric field component opposing the external one, as Fig. 1
illustrates. This component decreases the resultant field to a residual value and stops the
current pulse of the streamer. The next pulse occurs when the ignition field strength
value is reached again. This results in repeating current pulses around or after the zero
crossing of the testing voltage as Fig. 2 illustrates. It must be noted that the time scale
of the illustrated PD activity is not representative, because the goal of this illustration is
to give a physical understanding of void discharges.

Fig. 1. Electric field development inside a gas cavity during void discharge [2]

Fig. 2. Voltage and current illustration of gas cavity in the presence of void discharge [1]

2.2 Simulations and Creation of the Samples

With the help of finite element calculations, it became possible to determine the most
suitable dimensions of the discharge sources to work properly. For these calculations,
the main geometry, the used materials and the testing voltage had to be chosen first. For
the solid insulating material, epoxy resin was chosen, with a measured relative per-
mittivity of er; solid ¼ 3:545. The two electrodes were rounded edge copper cylinders
with a diameter of 20 mm and an edge radius of 2.5 mm. The plain bases of the
electrodes must be sinking in the resin to avoid unwanted surface discharges along the
edges. The chosen 50 Hz AC testing voltage was U ¼ 7 kVrms , which was represented
1018 M. Szirtes and R. Cselkó

in the electrostatic simulations as 7 kVDC to give us the RMS values of the alternating
electric field distribution. The static finite element simulations were made in 3D using
COMSOL. During the 3D tetrahedral meshing of the model, the minimum and max-
imum element sizes were 0.1 lm and 50 lm.

Fig. 3. Electric field distribution of bubble model

Fig. 4. Electric field strength along the Z axis with normalized void size and varying D/d ratio
(d = 1 mm)

Throughout the simulations, the varying values were the electrode distance (d) and
the bubble diameter – electrode distance ratio (D/d). The goal of the simulations was to
find the optimal combination of these values. Figure 3 shows a 2D slice of the 3D
electric field distribution of the simulated model in one of the observed arrangements,
while Fig. 4 presents the field strength values along the Z axis, between the two
electrodes and inside the bubble, in case of different D=d ratios, compared to the
 Simulation of Void Discharges and Time Domain Examination 1019

theoretical homogeneous external electric field and void electric field values of Eq. (1),
showing the limits of the homogeneous approximation of this equation.
The curves of Fig. 4 show that the theoretical homogeneous void field strength
ðEvoid Þ of Eq. (1) is only accurate if the D=d ratio is under 50%. By further increasing
the relative void diameter, the electric field in the cavity becomes inhomogeneous and
its magnitude values also decrease. To determine the optimal thickness of the epoxy
layer and the size of air bubbles inside, this distortion in the void electric field had to be
taken into consideration. Accordingly, the empirically modified ignition criterion of
inequality (3) has been created by combining the simulation results of Fig. 4 with
Eqs. (1) and (2). The inequality uses a fitted third-degree polynomial curve of the
maximum electric field strength values inside the void, as the function of the D=d ratio.
If inequality (3) is satisfied, the maximum electric field strength inside the bubble is
theoretically enough to start a void discharge as soon as a free initiatory electron
becomes available.
"  3  2   !#   
U 3er D D D 8:6


0:705
1:123
þ 0:290
þ 9:988 = 24:2
1 þ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi [ 1
d 1 þ 2er d d d D
100
ð3Þ

Where:
– U = AC testing voltage between electrodes [kVrms]
– d = Electrode distance [mm]
– D = Diameter of spherical air bubble [mm]
– er = Relative permittivity of solid dielectric material [ ]

Fig. 5. Maximum electric field strength inside air bubble, relative to the ignition value (function
of d and D/d)
1020 M. Szirtes and R. Cselkó

To determine the optimal d and D values, the left side of inequality (3) was
calculated for different electrode distances as Fig. 5 presents. This figure shows that an
epoxy layer with a thickness lower than 2 mm should be used between the electrodes to
meet the ignition criterion. With the available tools, the preparation of thinner epoxy
layers would have caused serious difficulties, so the thickness was chosen to be
d ¼ 1:5 mm. In case of this electrode distance, the ignition criterion is satisfied for D=d
ratios higher than 20%, according to the results of Fig. 5.
Figure 6 shows the 3D simulation model and the final arrangement of the created
void discharge sources. For the measurements, two samples were created with different
bubble sizes inside the epoxy layer between the electrodes. The sample Void#1 had air
bubbles with diameters in the range of 0.1–0.7 mm, while the bubbles of Void#2 had
diameters of 0.4–1.2 mm.

Fig. 6. The simulation model and the final arrangement of the void discharge sources

3 Examination of the Created Void Discharges

3.1 Methodology
The most suitable equipment for direct examination of the emitted electromagnetic
signals is a GTEM cell (Gigahertz Transverse ElectroMagnetic cell). It is an electro-
magnetically shielded chamber with the ability to forward the vertically polarized
transverse components of the electromagnetic waves generated inside of it, to the
coaxial cable connected to its output. The cell works as the 50 Ω termination of this
coaxial cable and prevents the development of internal reflections. The applied
equipment can be used in the frequency band of 9 kHz–5 GHz, which covers the
possible VHF and UHF frequencies of the partial discharge signals. The vertically
positioned sources and the 7 kVrms test transformer were placed on an insulative stand
inside the cell for the time of the measurements. The measured voltage signals were
then recorded with a digital sampling oscilloscope with 4 GSa/s sampling rate and
500 MHz analog bandwidth. Figure 7 shows the simplified schematic of the presented
measuring arrangement.
 Simulation of Void Discharges and Time Domain Examination 1021

For the examination of the time domain behavior of void discharges, phase-
resolved partial discharge (PRPD) patterns have been created for the sources Void#1
and Void#2. These patterns show the partial discharge activity as a function of the
phase angle of the 50 Hz testing voltage. Typically, these diagrams are used in the
conducted PD measurements during conventional diagnostic examinations, so the PD
intensity is commonly given as the apparent charge on the vertical axis. This charge
value is obtained by the integration of the conducted discharge current pulses. Because
this is not possible in case of emitted signal measurements, the figures presented below
show the peak values of the measured voltage pulses. To create the PRPD diagrams,
the local maxima of the measured voltage signals were recorded for 2 min in a 20 ms
wide time window in the test voltage phase range of 0°–360°. Compared to conven-
tional PRPD figures, a positive aspect of this solution is that the polarity of the voltage
pulses is also displayed, so that the nature of the positive and negative discharges can
be examined more qualitatively. The main information that can be extracted from these
figures is the phase angle, intensity and repetition rate of the voltage pulses. From these
data, with the use of different transformations or pattern-recognition algorithms, it is
possible to identify and separate the signals of multiple concurrent discharges [5–7].
Similarly to the PRPD diagrams, these techniques are typically used for conductively
measured PD signals, thus the examination of the emitted electromagnetic signals can
open new development directions in the future.

Fig. 7. The simplified measuring arrangement of time domain examinations

3.2 Results and Evaluation

Before the examination of the discharge sources, the noise-free operation of the
measuring system had to be verified. In doing so, special attention had to be paid to the
possible partial discharges caused by the transformer. The PRPD pattern shown by
Fig. 8 was created by energizing the test transformer without a connected discharge
source. Only a negligible number of low amplitude pulses were measured during the
2 min test cycle.
Figures 9 and 10 show the PRPD patterns of the sources Void#1 and Void#2. The
effect of the symmetrical arrangement can be observed in both figures, as similar
discharge behavior occurs in both half-periods of the sine wave. By examining the half-
periods, the effect of the increment in field strength can be observed, as the amplitudes
1022 M. Szirtes and R. Cselkó

of the discharge pulses increase along the positive ascending and negative descending
sections. Also, when the sine wave reaches its peak, the pulses cease shortly as
expected according to the explanation of Figs. 1 and 2.

Fig. 8. The PRPD pattern of the unloaded measuring system

By comparing the two figures, it can be seen that the discharge pulses of Void#2
occur at a higher repetition rate than in case of Void#1. The reason for this is the
difference between the bubble sizes of the sources, because the bubble diameters of
source#2 are somewhat larger. Knowing the bubble sizes, based on the inequality (3)
and the curves of Fig. 5, a larger Evoid =Eignition ratio is obtained for Void#2, which may
contribute to more frequent discharges.
Because of the small size of the air bubbles and their closed arrangement, the
initiatory electrons required for the initiation of the streamer discharge are less likely to
develop in a unit of time. This is significantly influenced by the quality of the bubble’s
inner surface and the dielectric material, which can affect the characteristics of the
resulting PRPD pattern. For example, in case of Void#1 and Void#2, the discharges
only develop after the zero crossing of the sine wave, which is consistent with some
literature examples, but there are also results where the pulses start before the zero
crossing similarly to the ideal pulses of Fig. 2 [5, 6, 8].
The likelihood of the formation of an initiatory electron increases with bubble size
as the number of ionizable particles is getting higher with increased volume. This may
also explain why the larger bubbles in Void#2 show more frequent discharges, because
if the initiatory electron is more likely to develop, more will occur over a given period
of time, resulting in more frequent discharges if the needed electric field strength is
given.
 Simulation of Void Discharges and Time Domain Examination 1023

Fig. 9. The PRPD pattern of the Void#1 PD source

Fig. 10. The PRPD pattern of the Void#2 PD source

4 Conclusions

The presented results show the success of the design process, which used finite element
simulations combined with partial discharge physics. The PRPD patterns of the emitted
electromagnetic void discharge signals are similar to the ones constructed from con-
ductive measurements, which can be found in the literature. The phase angle, repetition
rate and amplitude of the discharge pulses can be extracted from these patterns for
further processing. With the help of this information, characteristic differences can be
found between the different types of partial discharges. Knowing the type of the
1024 M. Szirtes and R. Cselkó

detected PD can help us narrowing down the list of possible faulty parts in a high
voltage equipment. This can speed up the procedure of finding and fixing these faults,
and in special cases, this identification could be enough to determine the specific
location of the discharge, replacing the more complex positioning methods. Knowing
the characteristic differences between the different types of PDs can highly support the
operation of location approximation methods in case of multiple concurrent discharges.

Acknowledgment. The presented research was supported by the ÚNKP-18-1 New National
Excellence Program of the Ministry of Human Capacities of Hungary.

References
1. Kuffel, E., Zaengl, W.S., Kuffel, J.: High Voltage Engineering: Fundamentals. Second edn.
pp. 381–385 (2000)
2. Gutfleisch, F., Niemeyer, L.: Measurement and simulation of PD in epoxy voids. IEEE Trans.
Dielectr. Electr. Insul. 2(5), 729–743 (1995). https://doi.org/10.1109/94.469970
3. Illias, H.A., Tunio, M.A., Mokhlis, H., Chen, G., Bakar, A.H.A.: Determination of partial
discharge time lag in void using physical model approach. IEEE Trans. Dielectr. Electr. Insul.
22(1), 463–471 (2015). https://doi.org/10.1109/TDEI.2014.004618
4. Boggs, S.A.: Partial discharge. III. Cavity-induced PD in solid dielectrics. IEEE Electr. Insul.
Mag. 6(6), 11–16 (1990). https://doi.org/10.1109/57.63094
5. Kraetge, A., Hoek, S., Rethmeier, K., Krüger, M., Winter, P.: Advanced noise suppression
during PD measurements by real-time pulse-waveform analysis of PD pulses and pulse-
shaped disturbances. In: 2010 IEEE International Symposium on Electrical Insulation, San
Diego, CA, pp. 1–6 (2010). https://doi.org/10.1109/elinsl.2010.5549723
6. Boonpoke, S., Marungsri, B.: Pattern recognition of partial discharge by using simplified
fuzzy ARTMAP. Int. J. Electr. Comput. Energ. Electron. Commun. Eng. 4(5), 789–796
(2010)
7. Janani, H., Kordi, B.: Towards automated statistical partial discharge source classification
using pattern recognition techniques. High Voltage 3(3), 162–169 (2018). https://doi.org/10.
1049/hve.2018.5048
8. Herath, H.M.M.G.T., Kumara, J.R.S.S., Fernando, M.A.R.M., Bandara, K.M.K.S., Serina, I.:
Comparison of supervised machine learning techniques for PD classification in generator
insulation. In: 2017 IEEE International Conference on Industrial and Information Systems
(ICIIS), Peradeniya, pp. 1–6 (2017)
 A Novel Approach of Critical Span Analysis

Dávid Szabó(&), Levente Rácz, and Gábor Göcsei

Budapest University of Technology and Economics, Budapest, Hungary

szabo.david@vet.bme.hu

Abstract. The maintenance of operational safety and reliability are the major
challenges during the operation of electrical network both in distribution and
transmission system. New technical trends like the spreading of distributed
generation or the integration of renewable energy sources are also raising the
extend of these challenges, while the adequacy to the market trends and con-
sumer satisfaction also should be satisfied.
Dynamic line rating (DLR) technology offers a cost-efficient solution for the
optimal utilization of the transmission system through the monitoring of the
power lines. Therefore, not only the transmission capacity of the overhead lines
can be increased, but system operators receive information about the actual state
of the line. Although, there are different algorithms for ampacity calculation and
conductor temperature tracking, nevertheless the building conception of the data
providing sensors are less discussed in the international literature. The aim of
this paper is to demonstrate the problems with critical span analyzation and
review the possible solution for the optimal sensor placement to DLR system.

Keywords: Dynamic line rating
DLR
Critical span
Analysis

Identification
System planning
Hot spot

1 Introduction

According to traditional approach a static line rating was determined for transmission
lines, which limits the network-capable power in a conservative way. Although, this
method is the cheapest solution for transmission capacity management, because does
not require any sensor, however the proper utilization of the overhead lines untapped,
while the operational safety is also critical in about 5% of the time according to the
concept of this calculation mechanism.
Contrarily, dynamic line rating (hereinafter refereed as DLR) is a state-of-the-art
technology to adjust line rating to the rapidly changing environment. The method based
on the thermal equilibrium of the conductors, where the heating and cooling effect of
the environment keeps heat-balance with the Joule-heat generated by the current flow
through the phase wires. Therefore, the real-time conductor temperature can be cal-
culated based on the environmental factors measured by weather stations installed on
transmission line towers and the ampacity of the line can be determined according to
the difference between the actual and maximal allowed conductor temperature.
Further advantage of DLR process is the predetermined line rating calculated with
weather forecast, which can be used for generation schedule planning. Furthermore, ice
prediction module and anti-icing unit can also be implemented into the system
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1025–1031, 2020.
https://doi.org/10.1007/978-3-030-31676-1_96
1026 D. Szabó et al.

according to the weather parameters used for calculation. Therefore, a complex grid
management technology can be realized based on DLR method, which offers an
extensive monitoring system of overhead lines supplemented with the prediction of
thermal overloads and ice formation on phase wires [1–3].
On the other hand, the establishment of DLR system requires different monitoring
equipment in the vicinity of the power line. Real-time conductor temperature calcu-
lation demands the measurement of environmental factors such as ambient temperature,
solar radiation, relative humidity, precipitation intensity, wind speed and direction;
which can be solved by the placement of weather stations on the towers, as near to the
phase conductors as possible. Moreover, line monitoring sensors installed on phase
wires makes the calculation safety factor higher, as well as can be used for clearance
measurement or ice detection. It is hard to achieve harmony between the number and
location of the used monitoring equipment, the cost of the realization of DLR infras-
tructure and the safety factor of the system.

2 Critical Span Analysis Methods

To make a complex grid management system – which not only includes DLR system,
but available to predict icing events – it is important to get a comprehensive picture
from the whole line corridor. The data acquired by measuring equipment installed on
the line should represent the actual status of the whole line, where the measurements
results can only deviate in the direction of security relative to the conditions of the
whole line.

2.1 Basis of Critical Span Identification Processes

During DLR determination it is important to find the local hot spots of the transmission
line, as the thermal overloads will be occurred firstly in these places. Annealing con-
ductor aging is an adverse effect, the effect of which conductors lose their tensile
strength. This phenomenon accumulates over the time as the effect of local thermal
overloads, which reduces the lifetime of the conductors, thereby increasing the pos-
sibility of outages [4, 5]. Furthermore, the investigation of clearances is also an
important aspect of critical span analysis to find out where will be the legal limitations
firstly violated in case of raised conductor temperature [6]. International critical span
analyzation method offers different algorithms to find out the local hotspots or the
highest sagging spans, as well as the solution which considers both criteria [7, 8].
On the other hand, not only local hot spots can reduce the operational safety of
power lines, but in case of winter extreme conditions ice formation will occur on the
lowest temperature sections. Thus, the identification of these areas as important as the
determination of hot spots in the point of view of operational safety.

2.2 Disadvantages of Existing Methods

Several studies showed that wind is the most influencing parameter on the thermal
behavior of overhead line conductors, while the spatial and temporal changing of wind
 A Novel Approach of Critical Span Analysis 1027

speed and direction is also significant [1–3]. Notwithstanding the existing critical span
analyzation methods are able to find the highest sagging spans with great efficiency and
accuracy, but the variation of local hot spots place cannot be managed properly with the
algorithms according to the afore-mentioned reasons [7, 8].
Case Studies
Based on field data several case studies presented here to demonstrate the problems
with existing critical span analyzing methods. Firstly, a 220 kV transmission line
equipped with 6 weather stations was investigated. Weather stations placed equidis-
tantly with good approximation. The length of the line is almost 30 km. Thus, the
whole line covered with weather stations about 5 km spatial resolution. Weather and
load data are available from 17 months with 5 min time resolution.
Wind speed measurements as the most influencing factor illustrated in Fig. 1.

Fig. 1. Wind deviation in speed and measurement place

Figure 1 shows how the wind speed is varying along the line. The difference
between the minimal and maximal wind speed measured in the same time illustrated in
the left side, where the median value is 2.7 m/s and the 25th percentile is 2 m/s. It
means that the difference in wind speed between the maximal and minimal measured
places not less than 2 m/s in the 75% of the time. Field data showed that minimal wind
speed measurement place is quasi evenly distributed as it shown it right side of Fig. 1.
So, the place, where the minimal wind speed measurable evenly changing in the time
along the line corridor.
Other environmental parameters are also varying in a wide interval along the line,
such as wind direction which is stochastic in case of low wind speeds, or the solar
radiation as the effect of cloud cover variation. These effects combined with the mostly
slight load values means that environmental factors have large effect on the conductor
temperature distribution along the line. To illustrate this problem, the above-mentioned
line was divided into 21 sections based on the direction changing in the line path at
angle towers. Then the conductor temperature was calculated for all sections, which
means that nearly 2 million data point was analyzed.
1028 D. Szabó et al.

Fig. 2. Hot spot analysis

Figure 2 shows the result of hot spot analysis. The left side shows that the maximal
temperature along the line may differ by more than 10 °C from the average conductor
temperature, however the median of the difference is 5.7 °C and the 75th percentile is
8 °C. The right side of Fig. 2 shows how each section share of the time between each
other in case of the hottest section. There is only one section (section ID4), which is
lower than 1% of the time is the critical one. So, there is hard to mark critical spans,
because almost of the spans are critical in a significant time slot.
Critical Spans in Case of Icing
Anti-icing techniques are intended to prevent ice accretion on phase conductors, which
can be achieved by the increased heat dissipation of the wires. Therefore, critical spans
are the spans with the lowest temperature in case of winter time, when ice formation is
possible on the power line structures. The same analysis of critical span was carried out
for the chilliest sections as in case of the hottest spans.

Fig. 3. Chill spot analysis

Figure 3 shows the result of the analyzation of the chill sections. The median
difference between the average conductor temperature and the chill points is 4.6 °C,
while the 75th percentile of the temperature difference is 6.8 °C. In this case, there is
only one section (section ID 11) which less than 1% of the time critical, however the
first section is critical in 20% of the time. In this case, the critical points also cannot be
determined obviously, as all of the sections critical more than 1% of the time except the
11th section.
 A Novel Approach of Critical Span Analysis 1029

Field Measurements
Line monitoring sensors measured data from 6 months are also available from another
line thanks to the Flexitranstore European Union project. Four line monitoring sensors
installed on a 110 kV transmission line located in Central Europe, which able to
measure the conductor temperature, inclination and load values. The four sensors
placed on a short segment of the line, the outsides located within 800 m to each other.
In this case the sensor placement realized according to historical icing events.

Fig. 4. Conductor temperature deviation measured by line monitoring sensors

Figure 4 shows the deviation on the conductor difference according to the mea-
surement results of line monitoring sensors. As the sensors placed close to each other,
the difference between the places of the minimal and maximal temperature gives the
most relevant information. Accordingly, in 25% of the time, this temperature difference
exceeds 2.1 °C, which is significant considering that the sensors placed within 800 m.

3 Possible Solutions for Sensor Placement

Simulations was carried out to find out what extent can be the conductor temperature
change along the line. Firstly, a 30 km long line was investigated, where the weather
stations placed approximately equidistant. The changing of wind speed in wide range
indicates that the conductor temperature can also varying in extensive dimension. On
the other hand, field measurements revealed that the phase wires temperature alters not
negligible in case of sensors installed close to each other.
The results indicate that the existing critical span analyzation methods cannot
satisfy the proper operational safety level desired by transmission system operators.
The conductor temperature can vary in a wide range depending on the length of the
transmission line, while the share of the time between the sections changes approxi-
mately evenly.
1030 D. Szabó et al.

The required operational safety level both in case of DLR and ice prevention
system imply to equip transmission lines with more measuring equipment than in
current practice. Although, the line monitoring sensors used nowadays able to measure
multiple attributes of the line, these costs are significant during the building of a
complex line monitoring system. Nevertheless, the implementation of reliable system
requires only the accurate measurement of conductor temperature. For this purpose, a
low-cost conductor temperature measuring sensor under development in the High
Voltage Laboratory of Budapest University of Technology and Economics. Thus, the
realization of reliable DLR and ice prevention system can be fulfilled without signif-
icant increase of the system implementation costs.

4 Conclusion

Several pilot projects originated in order to develop grid management technologies for
transmission lines. Integration of DLR and ice prevention system offer a cost-efficient
solution for the increase of operational safety. Although, the development of the cal-
culation methods has a great emphasis, there are different challenges, which should be
handled.
The optimal placement of line monitoring sensors is essential to the realization of
line monitoring systems. The disadvantages of the existing models illustrated with
different case studies and field measurements. It is showed that the conductor tem-
perature varies in wide interval along the line path, while the location of the extreme
values are changes approximately evenly. These results indicate the improvement of
existing methods to offer not only a cost-efficient technology for system operators, but
realize the appropriate operational safety, too. Therefore, a low-cost conductor tem-
perature measuring sensor under development in the High Voltage Laboratory of
Budapest University of Technology and Economics.

Acknowledgements. This work has been developed in the High Voltage Laboratory of
Budapest University of Technology and Economics within the boundaries of FLEXITRAN-
STORE project, which is an international project. FLEXITRANSTORE (An Integrated Platform
for Increased FLEXIbility in smart TRANSmission grids with STORage Entities and large
penetration of Renewable Energy Sources) aims to contribute to the evolution towards a pan-
European transmission network with high flexibility and high interconnection levels.
 A Novel Approach of Critical Span Analysis 1031

References
1. Rácz, L., Szabó, D., Göcsei, G., Németh, B.: Grid management technology for the integration
of renewable energy sources into the transmission system. In: 2018 7th International
Conference on Renewable Energy Research and Applications (ICRERA), Paris, pp. 612–617
(2018)
2. Halász, B.G., Németh, B., Rácz, L., Szabó, D., Göcsei, G.: Monitoring of actual thermal
condition of high voltage overhead lines. In: Camarinha-Matos, L., Adu-Kankam, K.,
Julashokri, M. (eds.) Technological Innovation for Resilient Systems. DoCEIS 2018. IFIP
Advances in Information and Communication Technology, vol. 521. Springer, Cham (2018)
3. Németh, B., Göcsei, G., Szabó, D., Rácz, L.: Comparison of physical and analytical methods
for dynamic line rating (DLR) calculations. In: CIGRE-IEC, 2019 Conference on EHV and
UHV (AC & DC), April 23–26, Hakodate, Japan (2019)
4. Bhuiyan, M.M.I., Musilek, P., Heckenbergerova, J., Koval, D.: Evaluating thermal aging
characteristics of electric power transmission lines. In: CCECE 2010, Calgary, AB, pp. 1–4
(2010)
5. Musilek, P., Heckenbergerova, J., Bhuiyan, M.M.I.: Spatial analysis of thermal aging of
overhead transmission conductors. IEEE Trans. Power Delivery 27(3), 1196–1204 (2012)
6. Lovrenčić, V., Gabrovšek, M., Kovač, M., Gubeljak, N., Šojat, Z., Klobas, Z.: The
contribution of conductor temperature and sag monitoring to increased ampacities of overhead
lines (OHLs). Periodica Polytechn. Electr. Eng. (2015). https://doi.org/10.3311/PPee.8585
7. Teh, J.: Analysis of dynamic thermal rating system of transmission lines. Ph. D. thesis. The
University of Manchester (2016)
8. Matus, M., et al.: Identification of critical spans for monitoring systems in dynamic thermal
rating. IEEE Trans. Power Delivery 27(2), 1002–1009 (2012)
Space Charges
 Simple Method to Visualize Surface
and Space Charges by Specially Processed
Colour Pigments

Thomas Huecker(&)

University of Applied Sciences – HTW Berlin, Berlin, Germany

tomas.huecker@htw-berlin.de

Abstract. HVDC applications are getting globally more and more important
for the energy transport in the high voltage networks. To design apparatus
properly it is important to understand the space and surface charges that are
accumulating under DC stress in certain areas. Existing methods to measure or
visualize such charge accumulations like “Pressure Wave Propagation Method”,
“Thermal Step Method”, “Pulsed Electro Acoustic Method” or “Electro Static
Volt Meters” show often limitations applied to full size apparatus. Sometimes
the resolution or specimen thickness is limited or the application can be rather
time consuming or complex. A simple method to visualize surface charges is
known as “Lichtenberg’s dust figures”. It is known since hundreds of years but
rarely practiced as the best working chemicals (yellow sulphur and red lead) are
not health save. In this paper a fresh view is given to charge visualization by
colour pigments. It is shown that different pigments attach to positive or neg-
ative charges. The application of this method is easy and takes just a view
seconds. E.g. tree type structures can be visualized with a resolution of less than
1 mm. Hover the value of the charge magnitude cannot be obtained. The surface
charge behaviour on different materials like Epoxy, PE-HD, Silicone is shown.
In addition, the possibility of space charge visualization on specimen cross
sections is discussed.

Keywords: HVDC
Surface charges
Space charges
Visualization

1 Introduction

The understanding of space and surface charge accumulation gets more and more
important for the design of HVDC equipment [1]. Existing measurement methods like
“Pressure Wave Propagation Method”, “Thermal Step Method”, “Pulsed Electro
Acoustic Method” or “Electro Static Volt Meters” are often time consuming to apply or
show limitations in resolution or specimen size [2–14]. Bud with all mentioned
methods the charge magnitude and distribution can be obtained. Since 1777 it is well
known that charged areas attracts dust [15]. This effect is e.g. utilized in electrostatic
filters. In a similar way colour pigments can be used to indicate charged areas on
insulator surfaces. Laser printers are based on this principle. Traditionally red lead and
yellow sulphur pigments have been used to make surface charges visible. This sub-
stance as well as laser toner cannot be recommended from safety point of view as

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1035–1043, 2020.
https://doi.org/10.1007/978-3-030-31676-1_97
1036 T. Huecker

inhaling has health unpleasant effects. Which save colour pigments work well for HV
apparatus and which types of surface charges or space charges can be visualized well is
going to be investigated here.

2 Colour Pigments

Colour Pigments are in use since centuries for painting applications. Today’s choice of
pigments is rather large and chemically well-defined but not all pigments are attracted
in the same manor to electrically charged surfaces.
Conductive ones are not getting attracted at all. Pigments that are none polar needs
to be pre-charged before they get attracted. However pre-charging is difficult to handle.
But a third group: polar pigments, were positive and negative charge centre are dis-
placed, work quiet well. They adhere to charged surfaces as they are getting close to
them. A preselection of pigments can be made by a charged glass or polyester bar if
rubbed with the right cloth and brought close to the pigments. With this simple method,
12 pigments were preselected and then further investigated in a second step: The
surface of a polyethylene specimen was covered by an earthed grid metal and then
charged by a corona source (Fig. 1).

Fig. 1. Charging of a PE HD insulator, 60 kV, 1 h, positive or negative polarity.

Afterwards the PE surface was sprinkled with the different pigments and the results
were compared. Some of the pigments were obviously doing better than others and
some were more attracted by negative and some by positive charged surface areas
(Fig. 2).
 Simple Method to Visualize Surface and Space Charges 1037

Fig. 2. Polarity effect on Prussian Blue pigments: the colour pigment adheres mainly to the
areas with positive charge.

During the tests it could be clearly recognized that pigment size and their lump
formation tendency influences the results strong. Well results could be obtained with
Pigments that are micronized and dried. Laser toner pigments work quiet well but they
are not health save, therefore they have been excluded from further tests. Table 1
provides an overview on the results. Prussian Blue (PB) is in use since hundreds of
years without reported health issues. As PB is delivering the best resolution the further
focus of the investigation is on this pigment only (−0,9 mm see Fig. 3).
Table 1. Overview of the results obtained by different colour pigments.
Colour pigment Adhering strength Adhering to charge polarity
1 Prussian blue Good Positive
2 Laser toner Good Positive
3 Iron oxides yellow Ok Negative
6 Suns yellow Ok Positive
4 Chrome titanium yellow Marginal ?
5 Pepper Marginal ?
7 Sky blue Marginal ?
8 Silna nature Marginal ?
9 Power pink Marginal ?
10 Russet Marginal ?
11 Bohemian green earth Marginal ?
12 Colourless salt Marginal ?

Fig. 3. Tree structure obtained by a charged PE specimen coloured by Prussian blue.

1038 T. Huecker

3 Surface Charges

While section “2 Colour Pigments” was about the selection of the right pigments, the
focus is now on visualization of surface charges on different materials: Under inves-
tigation are: PE, PTFE, Epoxy, Silicon, Glass and Ceramics.
All the materials were charged by a sharp needle that is right on top of the specimen
without any air gap at 20 kV with both polarities for 30 min. Immediately after the
charging Prussian blue was sprinkled on the specimens without any discharging. On
glass and ceramics, the results were negative. The pigments were not adhering and no
pattern could be observed. On the other materials, surface charges are well detectable
by colour pigment PB (Fig. 4).

Fig. 4. Surface charges on different materials visualized by colour pigment Prussian blue.
Specimen size is 10  10 cm
 Simple Method to Visualize Surface and Space Charges 1039

Fig. 5. Lichtenberg’s dust pattern. Left side by a positive charge, right by negative charges [15].

The obtained pigment distribution shows fine structured patterns that change for
different materials and by polarity.
Positive charges result in tree like structures on all materials while negative ones
result more in circular structures. This is in line what Lichtenberg has reported [15]
(Fig. 5). For comparison Fig. 6 show results by Zhang [9] obtained on a similar set up
using an electrostatic voltmeter, a Kelvin probe, a scan mechanic and an optimized
algorithm to calculate the space charge density from the electrical potential. On Epoxy
surface similar ring type structures can be seen. Due to the size of the Kelvin Probe just
limited resolution can be obtained.

Fig. 6. Surface charge accumulation on a cone-type insulator after application of −20 kV dc

voltage for 6 h in air [9] (Epoxy 10 cm diameter).

4 Space Charges Close to Insulator Surface

In a next test the ability of the pigment to visualize space charges that are close to the
insulator surface is investigated. Several insulator specimens were charged by a sharp
needle 20 cm above the surface at 40 kV for 120 min. Than the surface charges were
eliminated by a stroke with an earthed cupper brush. The colour pigment was applied to
the surface after different waiting times. Figure 7 show the results for the PE specimens
charged with negative voltage.
Obviously not all charges could be eliminated with the brush. A characteristic brush
pattern remains (upward stroke direction). It can be seen that after approximately 20 s
1040 T. Huecker

Fig. 7. Recovery time dependent pattern of Fig. 8. Recovery time dependent pattern of
negative charged PE surfaces after surface positive charged PE (comp. Fig. 7)
charges have been removed by an earthed
cupper brush [16].

charges inside the insulator have been diffused close enough to the surface to be seen
through the brush pattern. This effect gets stronger over time. After 120 s. the diffusion
process reaches a charge equilibrium and the space charge patterns get lighter.
Charging the surfaces with positive voltage delivers a similar result (Fig. 8). Just
that PB adheres better to positive charges and delivering therefore a higher contrast
indicating first space charges already after 10 s.

5 Deep Space Charges

To investigate space charges deep inside the insulator material, two sharp needles were
implemented in a PE specimen. 30 kV have been applied for 21 h. Figure 9 shows the
test setup. After the 21 h voltage stress the specimen has been cut in 4 pcs. and surfaces
have been polished. Probably this preparation influences the space charge distribution.
Nevertheless, an attempt was performed to prove their existence. The colour pigment
has been applied to the 4 surfaces along the cut. Figure 10 shows the result.
 Simple Method to Visualize Surface and Space Charges 1041

Fig. 9. Test setup for deep space charge investigation. PE specimen under 21 h 30 kV stress.

Fig. 10. (a) No clear pattern can be seen along the cut surfaces. (b) All pigments can be easily
removed by blowing them of, indicating no clear adhering to space of surface charges [16].
1042 T. Huecker

6 Conclusions

Polar colour pigments are easy and well usable for surface charge visualisation.
Prussian blue works quiet well if micronized and dried. Laser toner provides also good
results but is not health save. Surface charges can be well visualized on polyethylene,
PTFE, epoxy and silicon specimens but not on glass and ceramics. Hover no infor-
mation on charge magnitude is obtained by this method. Space charges that are close to
insulator surfaces can be detected after the time they need to diffuse to the surface. But
resulting colour patterns distinguish soon and are not that easy to interpret.
It was tried to visualize space charges deep inside an insulator on specimen cross
sections without success. Further investigations are required.

7 Acknowledgments

This work has been funded by HTW Berlin. Specimens have been provided by Lapp
insulators, Siemens AG and others.

References
1. IEEE STd 1732: IEEE recommended practice for space charge measurements on high
voltage direct current extruded cables for rated voltages up to 550 kV. In: IEEE Dielectrics
and Electrical Insulation society (2017)
2. Küchler, A.: Ausbildung Elektrischer Felder Bei Gleichspannungsbelastung. RCC-
Tagungsbericht, Berlin (2015)
3. Krause, J.: Raumladungen Ladungsträger-Beweglichkeit und Dielektrische Vorgänge in
Polymeren Werkstoffen. RCC-Tagungsbericht, Berlin (2015)
4. Fuchs, K., Berger, F.: Modellierung von Raumla-Dungen auf die Elektrische Feldstärke in
VPE-Kabeln unter Hochspannungsgleichstrombean-spruchung. RCC-Tagungsbericht, Ber-
lin (2015)
5. Gockenbach, E.: Prüfmethoden und Diagnoseverfahren für den Einsatz von Polymeren
Isolierstoffen in Kabeln der Hochspannungsgleichstromübertragung. RCC-Tagungsbericht,
Berlin (2015)
6. Zhao, N., et al.: Effect of O2/F2/N2 on space charge behaviour of low density polyethylene.
In: Proceedings of 19th ISH 2015, Pilsen, CZ (2015)
7. Herbst, H.: Industrial Organic Pigments: Production, Properties, Applications. Wiley,
Weinheim (2004)
8. Kumada, A., Okabe, S.: Charge distribution measurement on a truncated cone spacer under
dc voltage. IEEE Trans. Dielectr. Electr. Insul. 11(6), 929–938 (2004)
9. Su, G.Q., Wang, Y.B., Guo, B.H., Song, B.P., Mu, H.B., Zhang, G.J.: Experimental
investigation of surface charge accumulation behaviours on PTFE insulator under DC and
impulse voltage in vacuum. IEEE Trans. Dielectr. Electr. Insul. 24(6), 3347–3356 (2017)
10. Takabayashi, K., Nakane, R., Okubo, H.: HVDC partial discharge mechanisms and flashover
characteristics with charging activities on solid insulators in air. In: IEEE Conference on
Electrical Insulation and Dielectric Phenomenon (CEIDP), Fort Worth, TX, USA (2017)
11. Schueller, M., Straumann, U., Franck, C.: Role of ion sources for spacer charging in SF6 gas
insulated HVDC systems. IEEE Trans. Dielectr. Electr. Insul. 21(1), 352–359 (2014)
 Simple Method to Visualize Surface and Space Charges 1043

12. Zhang, B., Gao, W., Qi, Z., Wang, Q., Zhang, G.: Inversion algorithm to calculate charge
density on solid dielectric surface based on surface potential measurement. IEEE Trans.
Instrum. Meas. 66(12), 3316–3326 (2017)
13. Winter, A., Kindersberger, J.: Stationary resistive field distribution along epoxy resin
insulators in air under DC voltage. IEEE Trans. Dielectr. Electr. Insul. 19(5), 1732–1739
(2012)
14. Zhang, B., Qi, Z., Zhang, G.: Thermal gradient effects on surface charge of HVDC spacer in
gas insulated system. In: IEEE Conference on Electrical Insulation and Dielectric
Phenomena (CEIDP), pp. 703–706 (2016)
15. Lichtenberg, G.C.: De Nova Methodo Naturam Ac Motum Fluidi Electrici Investigandi
(Göttinger Novi Commentarii, Göttingen (1777)
16. Wolf, T.: Nachweis von Raum und Oberflächenladungen in Geläufigen Isolierwerkstoffen
mit Farbpigmenten. Masterarbeit HTW, Berlin (2017)
17. Sam, Y.L., Lewin, P.L.: Dynamic measurement of surface charge, dielectric materials,
measurements and applications. In: Eighth International Conference on (IEE Conf. Publ.
No. 473) (2000)
18. Fouracre, R.A., Santos, E.: Surface discharge propagation: the influence of surface charge,
power modulator symposium. In: Conference Record of the 2006 Twenty-Seventh
International (2006)
 Simulation of Partial Discharge Influenced
by Space Charges in Silicone Rubber

Haozhe Cui1, Zhaoliang Xing2, Chong Zhang2, Liangxian Zhang3,

Daomin Min1(&), and Shengtao Li1
1
State Key Laboratory of Electrical Insulation and Power Equipment,
Xi’an Jiaotong University, Xi’an 710049, China
forrestmin@xjtu.edu.cn
2
State Key Laboratory of Advanced Power Transmission Technology,
Global Energy Interconnection Research Institute Co., Ltd.,
Beijing 102209, China
3
Xi’an XD Transformer Co., Ltd., Xi’an 710077, China

Abstract. As the fast development of modern city and industry, energy con-
sumption shows an increasing tendency in recent years. More power cables of
high-voltage and extra-high-voltage class are widely used to meet the demand of
sustainable growth of grid capacity. However, in power cable system, statistics
show that above all failures in recent years, the faults of cable accessories can
make up more than 70%. One of the main reasons for the high failure rate is the
performance degradation of insulating materials, which is strongly influenced by
the partial discharge occurrence. Hence, as an effective way to evaluate the
performance of insulating dielectrics, modelling of partial discharge is a sig-
nificant issue to cable accessories. The paper aims to give a simplified model of
partial discharge based on space charge transport along the silicone rubber
which is adopted as the main insulating material of cable accessories containing
one spherical air void, considering partial discharge happens when gas were
ionized by charge collision under AC applied voltage. The model contains the
physical processes of charge injection and extraction at the interface between
electrode and dielectric material, space charge transport along the material,
charge exchange at the interface between dielectrics and the spherical void.
Additionally, the influence of gas conductivity under various temperature and
electric field is also concerned. The parameters of applied voltage and the size of
void are analysed in this paper. The simulation results indicate that the range of
discharge phase concentrates on the first and third quadrant of one AC cycle
where the voltage amplitude ascends. With the increase of applied voltage and
the size of void, the magnitude of charge also rises. Higher applied voltage and
larger void size contribute to the accumulation of space charges in the void and
accelerate gas ionization process. Therefore, the probability of partial discharge
in the interior of silicone rubber increases as well. The model which links the
partial discharge phenomena with space charge transport along the material and
the interface between solid and gas provides the other efficient way to reflect the
condition of partial discharges in insulating material.

Keywords: Partial discharge
Space charges
Silicone rubber
Gas

ionization
Interface

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1044–1054, 2020.
https://doi.org/10.1007/978-3-030-31676-1_98
 Simulation of Partial Discharge Influenced by Space Charges 1045

1 Introduction

Partial discharge is a discharge phenomenon caused by electrical breakdown at local

area in an insulating material. Generally, the insulating part of power equipment
includes multi materials, such as gas-solid insulation and liquid-solid insulation [1, 2],
where electric field distribution is also uneven due to their diverse dielectric properties.
As for the insulating part with single material, during manufacturing process or long-
time operation time, voids and other impurities can be brought or generated, which also
contribute to the high electric field in certain area in the interior or on the surface of
dielectrics. Once the electric field reaches breakdown strength, local electrical break-
down regarded as partial discharge phenomenon will occur in the interior of materials
at the same time. Although partial discharge will not lead to bulk breakdown imme-
diately, the performance degradation of insulating materials caused by continuous
partial discharge activity can take adverse effect on the stability of power equipment
and even result in the failure of the whole insulation system [3, 4]. Consequently,
partial discharge phenomenon and its effect to the insulating material has been a
significant subject for many researchers.
Silicone rubber, with high elasticity, wide service temperature range, and excellent
electrical properties, is adopted as the main insulation material in power cable system
[3, 4]. However, due to complex structure of the insulation, harsh operation environ-
ment, defects easily introduced in and so on, severe partial discharge also happens in
silicone rubber, which accelerates the material erosion and leads to the rising failure
rate of power cable accessories. Thus, as partial discharge is considered as one of the
main reasons for the failure of insulating material, this paper proposes to establish the
typical model of partial discharge and make a further investigation on the process of
charge transport and charge exchange in the silicone rubber with defects.

2 Theory

The type of partial discharge can be generally varied in three categories, including
discharge in solids or the interfaces between solids and gases, discharges at the surface
of gas insulator and discharges in liquids or the interfaces between solids and liquids
[5]. In this paper, only the first category will be discussed. The simplified physical
model as Fig. 1 shows is based on the solid dielectric material with one spherical void.
Partial discharge activity happens when gas in voids generates collision ionization and
space charges exchange at the interfaces between dielectrics and voids under certain
applied voltage. As Fig. 1 demonstrates, j1, j2, jv are the current density caused by the
migration of charges directionally before charge injection into the void, the current
density after charge extraction from the void and the current density in the void,
respectively. Similarly, E1, E2, Ev are the electric field under the influence of applied
voltage in three regions and d1, d2, dv are the thickness of three regions, respectively.
1046 H. Cui et al.

Anode Dielectric Materials Cathode

Void

I II III

j1 jv j2
E1 Ev E2
d1 dv d2

Fig. 1. Physical model of partial discharge in dielectric material with one spherical void.

2.1 Charge Injection and Extraction at the Interfaces Between

the Electrode and Dielectrics
Charge injection at the interfaces between electrodes and dielectrics obeys the Schottky
thermal emission law which can be described as [6]
8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
>
> EinðhÞ  eF=4pe0 er
>
> AT exp 
2
; F0
>
< kB T
jin ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi! ð1Þ
>
> E  eF=4pe e
>
> in ð e Þ 0 r
: AT exp  ; F\0
2
>
kB T

where: A = Richardson constant = 1.2  106 Am−2K−2

T = Sample temperature in K; F = Electric field in V/m;
Ein(h), Ein(e) = Injection barrier for holes and electrons respectively in eV;
e = Elementary charge in C = 1.6  10−19 C;
e0 = Vacuum permittivity = 8.85  10−12 F/m;
er = Relative permittivity of the sample;
kB = Boltzmann constant = 1.38  10−23 J/K;
Charge extraction at the interfaces between the electrode and dielectrics obeys the
Ohmic law which can be described as
(
re F; F  0
jex ¼ ð2Þ
rh F; F\0

where: re = Conductivity of electrons in S/m; rh = Conductivity of holes in S/m;

2.2 Charge Injection and Extraction at the Interfaces Between Dielectrics

and Voids
While collision ionization happens, positive and negative ions will be generated in
voids. In this paper, only negative ions will be discussed. Positive and negative ions
generated in voids has a high mobility rate and migrate along the electric field to the
 Simulation of Partial Discharge Influenced by Space Charges 1047

interfaces between dielectrics and voids in a very short time. Then, under the influence
of electric field, ions will be adsorbed by the interfaces. When adsorption time is long
enough, the process of desorption can be neglected within one or several AC cycles.
Consequently, ions at the interfaces will exchange with the charges in the dielectric
material. While charge exchange happens between positive ions and hole traps, charges
migrate into the hole traps and positive ions get electrons. The positive ions which have
recombined with electrons change into neutral molecule and get back into the voids.
Meanwhile, the number of electrons in conduction band also decreases. Similarly,
charge exchange also happens between negative ions and electron traps and negative
charges transfer to electron traps. Negative ions get recombination with holes in
valence band. The current density which injects from dielectrics by the thermal
emission into voids obeys the following equation [7],
rffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
kB T Esg  e3 F ðd1 ; tÞ=4pe0
Ce ðd1 ; tÞ ¼ nfree exp  ð3Þ
m kB T
rffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!
kB T Esg  e3 F ðd1 þ d2 ; tÞ=4pe0
Ce ðd1 þ d2 ; tÞ ¼ nfree exp  ð4Þ
m kB T

where: nfree = Density of free charges in conduction band in dielectrics in m−3;

m = Mass of electron in kg; Esg = Interface barrier in V/m;

2.3 Charge Transport in Solid Dielectrics

Charges migrate directionally under the influence of electric field. Conduction current
of electrons and holes in solid dielectrics are described by the equation [8, 9]:

jcðeÞ ðx; tÞ ¼ qfreeðeÞ ðx; tÞl0ðeÞ F ðx; tÞ ð5Þ

jcðhÞ ðx; tÞ ¼ qfreeðhÞ ðx; tÞl0ðhÞ F ðx; tÞ ð6Þ

While continuity equation follows the equation:

@qfreeðeÞ ðx; tÞ=@t þ @qtrapðeÞ ðx; tÞ=@t þ @jcðeÞ ðx; tÞ=@x ¼ Se ð7Þ

@qfreeðhÞ ðx; tÞ=@t þ @qtrapðhÞ ðx; tÞ=@t þ @jcðhÞ ðx; tÞ=@x ¼ Sh ð8Þ

where: jc(e), jc(h) = Conduction current density of electrons and holes in A/m2.
qfree(e), qfree(h) = Density of free electrons and holes in C/m3.
qtrap(e), qtrap(h) = Density of trapped electrons and holes in C/m3.
l0(e), l0(h) = carrier mobility of free electrons and free holes in m2/(Vs).
1048 H. Cui et al.

2.4 Gas Conductivity Dependent on Temperature and Electric Field

Gas conductivity can be described by the following equation with various electric field
and temperature. Under low electric field, gas conductivity obeys the ohmic law.
Conductivity in medium electric field is a constant, while high-field conductivity shows
exponential relationship with collision ionization coefficient avoid and void length dvoid.
8
>
> rvohmic ; 0\F\1V=m
<
jsat =F ¼ rvohmic =F;  1  F\1  106 V=m
rv ðT; F Þ ¼ ð9Þ
>
> C exp  F avoid dvoid
; 1  10  F\3  106 V=m
6
: 10
10 S=m; F [ 3  106 V=m

where: rvohmic = Conductivity which obeys the ohmic law in S/m;

jsat = Saturation current density in A/m2; C = The scaling factor.

2.5 Electric Field in Solid Dielectrics

The calculation of electric field in solid dielectrics includes three parts: the electric field
dependent on applied voltage without the influence of space charges or interface
charges, the electric field tuned by space charges and the electric field tuned by
interface charges. Firstly, the electric displacement vectors are equal everywhere in
solid dielectrics without space charges and interface charges. The equation of electric
field is given below:

Dsolid1 ¼ Dvoid ¼ Dsolid2

ð10Þ
ersolid1 e0 Fsolid1 ¼ ervoid e0 Fvoid ¼ ersolid2 e0 Fsolid2

where: Dsolid1, Dvoid, Dsolid2 = Electric displacement factor of Region I, II, III in C/m2.
ersolid1, ervoid, ersolid2 = Relative dielectric constants of Region I, II, III.
Considering the integral of the electric field to the position is equal to the applied
voltage, the applied voltage V can be described as

Fsolid1 dsolid1 þ Fvoid dvoid þ Fsolid2 dsolid2 ¼ V ð11Þ

Electric fields in Region I, II, III can be calculated by the following equations:

V
Fsolid1 ¼ ð12Þ
dsolid1 þ ðersolid1 =ervoid Þdvoid þ ðersolid1 =ersolid2 Þdsolid2

V
Fvoid ¼ ð13Þ
ðervoid =ersolid1 Þdsolid1 þ dvoid þ ðervoid =ersolid2 Þdsolid2

V
Fsolid2 ¼ ð14Þ
ðersolid2 =ersolid1 Þdsolid1 þ ðersolid2 =ervoid Þdvoid þ dsolid2

Electric field influenced by space charges qsc can be calculated by Poisson equa-
tion, which can be described as [8, 9]
 Simulation of Partial Discharge Influenced by Space Charges 1049

@/2 =@x2 ¼ qSC =e0 er ð15Þ

The relation between electric field influenced by space charges and potential inside
the sample is described by

F ¼ r/ ð16Þ

As for the electric field tuned by interface charges, while gas ionization happens,
charges at the left and right interface between dielectrics and gas are named as qsgL and
qsgR, respectively. Due to the potential between two electrodes equaling to zero, the
following equation can be obtained that

qsgL  qsgR ð17Þ

To solve out the induced charges at both two opposite electrodes influenced by qsgL
and qsgR, it can be obtained that the distribution of electric field formed by interface
charges. Then, considering the influence of electric field formed by applied voltage,
space charges and interface charges, electric field distribution can be calculated.

2.6 Parameters for Charge Transport in Silicone Rubber with One

Spherical Void
The thickness of silicone rubber sample is set as 2 mm. The dielectric constant of
silicone rubber is 3.5 measured by broadband dielectric spectrometer at 50 Hz at room
temperature. The energy of deep hole and electron traps is set as 0.92 eV and trap
density is 6.2  1020 m−3 for both electrons and holes [10]. In addition, the injection
barriers at anode and cathode and at the interfaces between silicone rubber and the void
for electrons and holes are assumed to be same as the deep trap energy. The carrier
mobility of silicone rubber controlled by shallow traps is 7  10−14 m2 =ðV
sÞ [10].
For the gas parameters in the void, the pressure of gas is set as one standard atmo-
spheric pressure and the gas temperature is 300 K, while the threshold of electric field
for partial discharge is 30 kV/mm as the breakdown strength of air.

3 Simulation Results and Discussion

3.1 Simulation of Partial Discharge Dependent on AC Voltage Amplitude

As the amplitude of applied AC voltage ascends in one cycle, electric field near the
spherical void in silicone rubber also rises. When electric field reaches the threshold of
discharge in the void, partial discharge activity occurs in the sample. Figure 2
demonstrates the simulation results of partial discharge in silicone rubber under various
amplitude of AC voltage of 5 kV and 15 kV with the frequency of 50 Hz. All partial
discharge activities in a period time of 0.4 s are contained in simulation results. It can
be observed that most partial discharge activities concentrate on the first and the third
quadrant, while minority can be observed at the second and the fourth quadrant in one
1050 H. Cui et al.

AC cycle. Meanwhile, as the increment of voltage amplitude, the magnitude of partial

discharge also increases. For instance, under the voltage amplitude of 5 kV, the
maximum partial discharge capacity is only around 3 pC, which means the existence of
slightly partial discharge in silicone rubber. While the voltage amplitude rises to 15 kV,
the maximum partial discharge capacity reaches the value of around 50 pC indicating
that severer partial discharge happens in the interior of silicone rubber. In addition, as
the threshold of gas breakdown strength is the same, higher amplitude of AC voltage
with the same frequency means less time is needed to reach the gas breakdown
strength. Consequently, the phase where partial discharge mainly occurs shows the left-
shifted tendency as the AC voltage ascends.

4 5 60 15
(a) 5kV (b) 15kV
10

Voltage Amplitude(kV)
Voltage Amplitude(kV)

3
Partial Discharge(pC)
Partial Discharge(pC)

40 5

2 0 0

20 -5
1
-10

0 -5 0 -15
0 90 180 270 360 0 90 180 270 360
Phase(°) Phase(°)

Fig. 2. Simulation results of partial discharge in silicone rubber with various AC voltage.

Figure 3 demonstrates the simulation results of conduction current density in the

void and the space charge density at the interfaces between the void and the silicone
rubber in one AC cycle. While the electric field at the void reaches the breakdown
strength of the gas, conduction path is formed and conduction current can also be
observed in the void. The results indicate that conduction current which reflects the
intensity of discharge in the void is proportional to the amplitude of applied AC
voltage. The large conduction current could cause severe thermal effect to the sample
itself, which is one of the main reasons that contributes to the accelerating degradation
of the insulating material. As for the charge density at the interfaces between the
silicone rubber and the void, as the increment of applied AC voltage, more mobile
charges caused by severer discharge in the void can be captured by electron and hole
traps at the interfaces before the electric field reaches the breakdown strength of gas.
As partial discharge activity happens at relatively low applied voltage, the distor-
tion of electric field caused by the accumulation of space charges near the area between
the electrode and the silicone rubber sample can be neglected. Consequently, the
electric field at a certain place except the region of the void in the interior of the silicone
rubber is chosen as the electric field of silicone rubber sample. Figure 4 shows the
variation of electric field at the interface between the silicone rubber and the void and in
the silicone rubber respectively in one cycle under various AC voltage amplitude. It can
be found that the electric field at the interface is much higher than that in the other place
of silicone rubber. For example, the maximum electric field under the AC voltage of
5 kV reaches around 1.10  107 V/m, while in the other place of silicone rubber, the
 Simulation of Partial Discharge Influenced by Space Charges 1051

Conduction Current Density (x10 A/m ) 1.5 8 10

Conduction Current Density (x10 A/m )

2

2
Space Charge Density at Interfaces
(a) 5kV (b) 15kV

Space Charge Density at Interfaces

2
6
1.0

-3
-4

4 5
1
0.5
2

(x10 C/m )
3

(x10 C/m )
3
0.0 0 0 0

-5

-4
-2
-0.5 -1
-4 -5
-1.0
-6 -2
Left Side Left Side
-1.5 -8 -10
200 205 210 215 220 200 205 210 215 220
Time (ms) Time (ms)

Fig. 3. Simulation results of conduction current density and interface charge density under
various AC voltage.

maximum electric field is only around 3.52  106 V/m. Charges which accumulates at
the interfaces distort the electric field near the area between the void and the silicone
rubber. In addition, combined with Fig. 2, it can be observed that when partial dis-
charge happens, electric field also distorts in a small range for charge recombination
and the direction of electric field changes quickly to the opposite due to both the
injection and extraction of charges at the interfaces and the migration of free charges
under the influence of the electric field.

7
15kV-Silicone Rubber
2x10 5kV-Silicone Rubber
Electric Field (V/m)

7
1x10

7
-1x10

7 15kV-Interface
-2x10
5kV-Interface
200 205 210 215 220
Time (ms)

Fig. 4. Simulation results of electric field at the interface under various AC voltage.

3.2 Simulation of Partial Discharge Dependent on Void Thickness

Figure 5 demonstrates the simulation results of partial discharge with various thickness
of the void of 1 lm and 2 lm under the applied voltage of 10 kV. Results indicate that
the void thickness have a marked impact on the partial discharge. The discharge
magnitude rises as enlarging the void size. With the thickness of 1 lm, the maximum
value of partial discharge is around 7.5 pC, while the thickness of the void is set as
twice larger, severe partial discharge activity can be observed and the maximum partial
discharge rises to nearly 30 pC. To figure out the variation of the magnitude of partial
1052 H. Cui et al.

discharge, charge density at the interfaces between the void and the silicone rubber in
one AC cycle with the same applied voltage is analysed as Fig. 6 demonstrates. As the
increment of the thickness of the void, the surface area of the void where happens
partial discharge activity also increases. More space charges exchange at the interfaces,
which contributes to the increment of the density of interface charges and further the
magnitude of partial discharge.

10 10 30 10
(a) 1μm (b) 2μm

Voltage Amplitude(kV)
8

Voltage Amplitude(kV)
Partial Discharge(pC)
Partial Discharge(pC)

5 5
20
6
0 0
4
10
-5 -5
2

0 -10 0 -10
0 90 180 270 360 0 90 180 270 360
Phase(°) Phase(°)

Fig. 5. Simulation results of partial discharge in silicone rubber with various void thickness.

Figure 7 indicates the variation of electric field dependent on the void thickness
under the applied voltage of 10 kV. In one AC cycle, the electric field at the interface
compared to that in the silicone rubber is seriously distorted by the partial discharge
activity in the void. Considering that more space charges migrate to the surface of the
void and exchange with charges in the silicone rubber by the partial discharge activity
as the void size increases,the internal electric field formed by space charges in the void
which direction is opposite to the applied electric field also rises. Consequently, as the
increment of void thickness, it can be noticed that the electric field slightly decreases at
the first and third quadrant. The maximum electric field with the void thickness of 1 lm
is about 1.89  107 V/m while the maximum electric field reduces to 1.74  107 V/m
when the thickness of the void is 2 lm. Furthermore, compared to the variation of ac
voltage amplitude, the variation of thickness of the void contributes less to distortion of
electric field.

-4
4.0x10 Left Side
Interface Charge Denisty (C/m )

1μm
3

2μm
-4
2.0x10

0.0

-4
-2.0x10

-4
-4.0x10
200 205 210 215 220
Time (ms)

Fig. 6. Simulation results of interface charge density dependent on the void thickenss at the interface.
 Simulation of Partial Discharge Influenced by Space Charges 1053

7
2x10
1μm
2μm
7
1x10

Electric Field (V/m)

0

7
-1x10

7
-2x10
200 205 210 215 220
Time (ms)

Fig. 7. Simulation results of electric field at the interface in silicone rubber with various void
thickenss.

4 Conclusion

We investigated the partial discharge activity based on a model of space charge

transport along silicone rubber with one spherical void. The process of charge
exchange and electric field distribution at the interfaces between silicone rubber and
void were analyzed. As the increment of applied voltage and void thickness, the
process of gas ionization is accelerated and more space charges migrate to the inter-
faces and accumulate under the applied electric field. Consequently, the magnitude of
partial discharge rises significantly. The model can effectively simulate the condition of
partial discharge in dielectrics and it should also be noticed that the present work is
only based on one void existing in the sample, while the partial discharge activity in the
dielectrics is the interaction of multi voids in fact. Considering that charge transport
along the multi voids is complex, much work is needed to do in the future to discuss the
partial discharge activity in the dielectrics with multi voids.

Acknowledgement. This work was supported by State Key Laboratory of Advanced Power
Transmission Technology (Grant No. GEIRI-SKL-2018-010), the National Basic Research
Program of China (Grant No. 2015CB251003), and the National Natural Science Foundation of
China (Grant No. 51507124).

References
1. Shigemitsu, O., Genyo, U., Haruhisa, W., Hitoshi, O.: Partial discharge-induced degradation
characteristics of insulating structure constituting oil-immersed power transformers.
IEEE TDEI 17(5), 1649–1656 (2010)
2. Mengchen, H., Xutao, H., Qiuyang, L., Shengfu, W., Yao, C., Junhao, L.: Research on
partial discharge characteristics of protrusion defect in GIS under oscillating lightning
impulse voltage. In: 2016 International Conference on Condition Monitoring and Diagnosis,
pp. 920–923, Xi’an (2016)
3. Yuanxiang, Z., Yunxiao, Z., Ling, Z., Dawei, G., Xu, Z., Mingyuan, W.: Electrical tree
initiation of silicone rubber after thermal aging. IEEE TDEI 23(2), 748–756 (2016)
1054 H. Cui et al.

4. Boxue, D., Zongle, M., Yu, G., Tao, H.: Effect of ambient temperature on electrical treeing
characteristics in silicone rubber. IEEE TDEI 18(2), 401–407 (2011)
5. Lutz, N.: A generalized approach to partial discharge modelling. IEEE TDEI 2(4), 510–528
(1995)
6. Campbell, S., George, G.M.: Charge injection and recombination at the metal–organic
interface. Chem. Phys. Lett. 299(2), 115–199 (1999)
7. Igor, P., Sergey, P., Severine, L.R., Leanne, C.P.: Simulation of breakdown in small
confined volumes inside dielectrics for electrical ageing and diagnostics. In: 2010 Annual
Report Conference on Electrical Insulation and Dielectric Phenomena, pp. 1–4, Indiana
(2010)
8. Daomin, M., Shengtao, L., Yoshimichi, O.: Numerical simulation on molecular displace-
ment and DC breakdown of LDPE. IEEE TDEI 23(1), 507–516 (2016)
9. Daomin, M., Yuwei, L., Chenyu, Y., Dongri, X., Shengtao, L., Qingzhou, W.: Thickness
dependent DC electrical breakdown of polyimide modulated by charge transport and
molecular displacement. Polymers 10(9), 1012–1029 (2018)
10. Daomin, M., Chenyu, Y., Yin, H., Shengtao, L., Yoshimichi, O.: Dielectric and carrier
transport properties of silicone rubber degraded by gamma irradiation. Polymers 9(10), 533–
547 (2017)
 Calculation of Space Charge and Field
Distributions in a Cable Joint

Yang Wu , Chuanhui Cheng , Rui Su , Kai Wu(&) ,

and Lijun Wang

State Key Laboratory of Electrical Insulation and Power Equipment,

Xi’an Jiaotong University, Xi’an 710049, China
wukai@mail.xjtu.edu.cn

Abstract. At present, high-voltage direct current (HVDC) transmission tech-

nology is more and more widely used in long-distance transmission because of
its low line cost, low loss and easy control. But under high-voltage DC electric
field, the accumulation of space charge will occur in the DC insulation structure,
leading to the distortion of electric field, and thus influencing the aging and
failure of the insulation. Due to the limitation of space charge measurement
methods, we can only obtain the space charge distribution of simple structures
(e.g. films or coaxial cable samples), and the distribution of space charge in
complex insulation structure cannot be directly measured. Therefore, it is nec-
essary to simulate the distribution of space charge in complex insulation
structure using simulation software based on the existing space charge transport
theory. In this paper, based on COMSOL Multiphysics software, the space
charge and field distribution in a 3-dimensional structure can be calculated. The
model used in this paper is the bipolar charge transport model, to simulate the
distribution of space charge and electric field in various insulation structures
under the action of high voltage DC electric field. This paper calculates the
space charge and electric field distribution in a cable joint under temperature
gradient. Comparing with traditional way which only the conductivity was taken
into consideration, this method can accurately reflect the electric field distortion
caused by the difference of carrier injection, extraction and migration.

Keywords: Space charge
Simulation
Cable joint

1 Introduction

With the development of high-voltage direct current (HVDC) transmission technology,

crosslinked polyethylene (XLPE) cables have been widely used due to their good
electrical and mechanical properties. Under the action of HVDC, the accumulation of
space charge will occur inside the XLPE. The existence of these space charges will
make the distortion of electric field distribution in XLPE and thus affect the breakdown
performance of the cables [1, 2].
In the cable joint, there are two materials, XLPE and EPR (ethylene propylene
rubber). As the relative permittivity, mobility, electrical conductivity, injection and

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1055–1066, 2020.
https://doi.org/10.1007/978-3-030-31676-1_99
1056 Y. Wu et al.

extraction characteristics of the two materials are different, it is easy to accumulate

space charge at the interface of XLPE and EPR [3].
In order to study the distribution of space charge and electric field in cable joints,
this paper built a 3D numerical simulation model.

2 Simulation Theory

The accumulation of space charge involves many physical processes, including carrier
injection, extraction, transport, trapping, recombination, ionization, etc. These physical
processes interact and restrict with each other. On the one hand, they are affected by the
electric field; and on the other hand, they also affect the distribution of space charge,
leading to the electric field distortion. In order to simulate the distribution of space charge,
some processes must be reasonably neglected, using appropriate simulation models.

2.1 Research Status

In the past decades, a series of researches for the numerical simulation of space charge
in polymers have been carried out. Alison and Hill first proposed the use of charge
transport model to calculate and studied the space charge accumulation characteristics
of polymer dielectric materials in 1994 [4]. They used Schottky charge injection,
barrier charge extraction and single barrier hopping model. Roy developed a numerical
model for explaining charge transport in polyethylene including the effect of trapping
and recombination using a fluid model with a constant mobility [5]. Kai Wu system-
atically discussed the influencing mechanism of temperature gradient to the space
charge distribution in polyethylene using the bipolar charge transportation model based
on the blocked extraction of injected charge [6]. In this paper, we will use bipolar
charge transportation model to calculate the space charge and electric field distribution
in a cable joint under temperature gradient.

2.2 Bipolar Charge Transportation Model

From a macro perspective, the transport of charge in an insulating material can be
regarded as a typical fluid process. As shown in Fig. 1, electrons are injected from the
cathode and extracted at the anode; holes are injected from the anode and extracted at
the cathode. Inside the medium, electrons and holes move under the action of the
electric field. During this process, carrier trapping, detrapping and recombination occur.
The following equations describe the behaviour of charge carries:

@nðx;tÞ
jðx; tÞ ¼ lðx; tÞ nðx; tÞ Eðx; tÞ  Df @x
@nðx;tÞ @f ðx;tÞ
@t þ @x ¼ Sðx; tÞ ð1Þ
@Eðx;tÞ
@x ¼ qalle0ðx;tÞ
er
 Calculation of Space Charge and Field Distributions in a Cable Joint 1057

Fig. 1. Bipolar charge transportation model.

The first one is the transport equation, j is the flux of carries, l is the mobility of
carries, n is the density of species, E is the electric field, and Df is the diffusion
coefficient. The second is the continuity equation, S is the source term that includes
trapping and recombination. The third is the Poisson’s equation, qall is the charge
density of all carries and e is the permittivity.
The process of charge injection follows the Schottky law:

 rffiffiffiffiffiffiffi!
ew ei e eEc
jei ¼ ATc2 exp  exp
kTc kTc 4pe

 rffiffiffiffiffiffiffi! ð2Þ
ewhi e eEa
jhi ¼ ATa exp 
2
exp
kTa kTa 4pe

where jei and jhi are the fluxes of electrons and holes at the cathode and anode, wei and
whi are the Schottky injection barriers for holes and electrons at the boundary, Tc and Ta
are the temperature of cathode and anode, Ec and Ea are the electric field, A is the
Richardson constant.
For the extraction process of carriers, the method of setting the extraction rate
artificially is used to describe the carrier extraction limitation. And can be described as
following equations:

jeo ¼ Ce le ne jEa j
ð3Þ
jho ¼ Ch lh nh jEc j

where jeo and jho are fluxes of extraction, Ce and Ch are the extraction rates of electrons
and holes.
Moreover, in the actual operation of the cable, because of the current heating, the
temperature of core rises, a temperature gradient exists between the inner and outer
sides of the cable, and it affects the distribution of space charge. The temperature
distribution in the cable joint needs to be considered in the simulation.
1058 Y. Wu et al.

In this report, COMSOL Multiphysics, which is a general-purpose software plat-

form based on advanced numerical methods for modeling and simulating physics
problems, is adopted to make the simulation.

3 Simulation Models

This paper selects the integral prefabricated cable joint as the research object, use EPR
as the reinforced insulating material, and tightly combines with the cable body through
the elasticity of the preform itself.
The model cross-section of the integral prefabricated cable joint is shown in Fig. 2.
It mainly includes the stress cone, the cable main insulation material, the reinforced
insulation material and the high voltage shield structure.

Fig. 2. The cable joint model. 1-core, 2-XLPE, 3-crimping tube, 4-high voltage shield, 5-EPR,
6-stress cone.

In this model, the material of stress cone and the high voltage shield structure are
semiconductive material, which can be seen as ground connection. And the crimping
tube is made of metal. In addition, as the model is symmetrical, it can be simplified as
shown in Fig. 3.

Fig. 3. Simplified cable joint model.

 Calculation of Space Charge and Field Distributions in a Cable Joint 1059

A summary of the parameters of the model is presented in Table 1. In the model,

250 kV DC voltage is applied to the core and the temperature is set. The outside is
grounded and the temperature is room temperature (293 K).

Table 1. Parameters
Parameters XLPE EPR
er 2.25 2.75
Barrier height of injection wei 1.3 [eV] 1.25 [eV]
whi 1.3 [eV] 1.25 [eV]
Df 1e − 12 [m^2/s] 1e − 12 [m^2/s]
Recombination coefficients 4e − 3 [m^3/(C
s)] 4e − 3 [m^3/(C
s)]
Extraction rate 0.1 0.1

At the same time the mobility is in the form of the following equation, and is fitted
according to the relationship between mobility and electric field and temperature.


A C
l¼ exp B  E þ ð4Þ
E T

4 Result and Discussion

Using the above model and parameters, this paper carries out simulation calculations.

4.1 Model Verification

In order to verify the validity of the simulation model, this paper first establishes a
simple one-dimensional model to simulate the space charge based on the bipolar charge
transport model.
The length of the one-dimensional model is 0.4 mm. A voltage of 4 kV is applied
to the anode and the cathode is grounded. The results of this simulation are shown in
Fig. 4.
As shown in Fig. 4(a), the electrons accumulate at the anode, and the holes
accumulate at the cathode. The accumulation of heteropolar charge makes the electric
field distorted near the electrode, as shown in Fig. 4(b). Obviously, these results are
consistent with the experimental results, indicating the feasibility of the tool of
COMSOL.
1060 Y. Wu et al.

Fig. 4. The simulation result of one-dimensional model.

4.2 Simulation of Cable Joint

Using the simplified cable joint model, based on the bipolar charge transport model,
this paper simulate the space charge and electric field distribution in the cable joint, as
shown in Fig. 5. In this model, the temperature of the inside of the cable joint is 363 K,
which means the temperature gradient is 70 K.
From Fig. 5(b), it can be seen that the electric field near the root of the stress cone
and the end of the high voltage shield are severely distorted. For more careful obser-
vation, the results near these two parts are zoomed as shown in Fig. 6. In these two
parts, space charge accumulates and causes distortion of the electric field, which
become the weak points of the cable joint.
 Calculation of Space Charge and Field Distributions in a Cable Joint 1061

(a)Space charge (b)Electric field

Fig. 5. Space charge and electric field distributions in the cable joint.

Fig. 6. Space charge and electric field distributions near the root of the stress cone and the end
of the high voltage shield.

The electric field strength at the end of the high voltage shield is the largest, and the
electric field strength at the junction of the high voltage shield and EPR (b–c–d in
Fig. 3) is extracted, as shown in Fig. 7. The maximum electric field strength reaches
14.8 kV/mm, which is far exceeding the average field strength, 3.3 kV/mm.
1062 Y. Wu et al.

Fig. 7. The electric field strength at the junction of the high voltage shield and the EPR.

Figure 8 shows the electric field strength near the root of the stress cone (g–a–h in
Fig. 3). It can be seen that the maximum electric field strength appears at the point a in
Fig. 3, at where the three parts, XLPE, EPR and stress cone meeting.

Fig. 8. The electric field strength near the root of stress cone.

4.3 Interface Charge

And at the interface of the two materials, it is easy to accumulate space charge. In order
to observe the interface charge, this paper create a cut line (e–f in the Fig. 3) passing
through XLPE and EPR, the space charge at this line is shown in Fig. 9.
 Calculation of Space Charge and Field Distributions in a Cable Joint 1063

From Fig. 9, it can be seen that at the interface between XLPE and EPR, space
charge accumulation occurs. Because the mobilities of the two materials are different,
the charges transferred to here cannot be completely removed, resulting in the accu-
mulation of charges at the interface.

Fig. 9. Space charge at the cut line.

Along the interface between XLPE and EPR (a–b in Fig. 3), the charge accumu-
lated at the root of the stress cone is the most, as shown in Fig. 10. Here, the insulation
structure is complicated, and the charge injection, extraction and interface effects work
together, leading to the large charge accumulation.

Fig. 10. Space charge accumulation at the interface.

1064 Y. Wu et al.

4.4 Effect of Temperature Gradient

It can be seen from Eqs. 2 and 4 that the change of temperature affects the process of
carrier injection and migration, and thus influences the space charge distribution.
Figure 11 shows the temperature distribution in the cable joint at a temperature gra-
dient (TG) of 70 K. Taking the high-voltage shielding end where has the highest
electric field strength as an example, the influence of temperature gradient on space
charge distribution and electric field strength can be observed. As shown in Fig. 12, the
electric field strength increases as the temperature gradient between the inner and outer
sides of the cable joint increases. And the highest electric field strength at TG = 70 K is
about 14.82 kV/mm, while that at TG = 0 K is about 7.88 kV/mm.

Fig. 11. Temperature distribution (TG = 70 K).

Fig. 12. The electric field strength at the junction of the end of high voltage shield and the EPR
at different temperature gradient.
 Calculation of Space Charge and Field Distributions in a Cable Joint 1065

5 Conclusions

In this paper, the simulation model of cable joint is constructed based on bipolar charge
transport model and blocked charge extraction using COMSOL Multiphysics software.
Comparing with traditional way in which only the conductivity was taken into con-
sideration, this method can accurately reflect the electric field distortion caused by the
effects of carrier injection, extraction and migration.
According to the simulation results, space charge is easily accumulated at the root
of the stress cone and the end of the high voltage shield. The maximum electric field
strength in the cable joint appears at the end of the high voltage shield, which is much
higher than the average electric field strength. Moreover, the presence of a temperature
gradient can make the electric field distortion in the cable joint more serious.

Acknowledgements. Authors wish to thank the support of the National Key Research and
Development Program of China (Grant No. 2016YFB0900703), the National Key Research and
Development Plan of China (Grant No. 2016YFB0900804) and the State Grid Corporation of
Science and Technology Project of China (Grant No. 522722160001).

References
1. Fabiani, D., Montanari, G.C., Laurent, C., et al.: Polymeric HVDC cable design and space
charge accumulation. Part 1: insulation/semicon interface. IEEE Electr. Insul. Mag. 6(6), 11–
19 (2007)
2. Delpino, S., Fabiani, D., Montanari, G.C., et al.: Polymeric HVDC cable design and space
charge accumulation. Part 2: insulation interfaces. IEEE Electr. Insul. Mag. 1(1), 14–24
(2008)
3. Lan, L., Jiandong, W., Wang, Y., Yin, Y.: Space charge property at the interface in low
density polyethylene/ethylene propylene rubber double-layered insulation. Proc. CSEE 35
(5), 1266–1272 (2015)
4. Alison, J.M., Hill, R.M.: A model for bipolar charge transport, trapping and recombination
in degassed crosslinked polyethylene. Phys. D Appl. Phys. 27, 1291–1299 (1994)
5. Roy, S.L., Boufayed, F., Teyssedre, G., et al.: Computer simulation of space charge
distribution in an XLPE-EPR sandwich. In: 2005 Annual Report Conference on Electrical
Insulation and Dielectric Phenomena, Nashville, TN, United states, pp. 661–664 (2005)
6. Wang, X., Lv, Z.P., Wu, K., et al.: Study of the factors that suppress space charge
accumulation in LDPE nanocomposites. IEEE Trans. Dielectr. Electr. Insul. 21(4), 1670–
1679 (2014)
7. Lv, Z.P., Wang, X., Wu, K., et al.: Simulation of space charge distribution in polyethylene
under a temperature gradient. In: CEIDP 2014, Iowa, America, October 18–22, pp. 879–882
(2014)
8. Le Roy, S., Segur, P., Teyssedre, G., Laurent, C.: Description of bipolar charge transport in
polyethylene using a fluid model with a constant mobility: model prediction. J. Phys.
D Appl. Phys. 37, 298–305 (2004)
9. Le Roy, S., Teyssedre, G., Laurent, C.: Modelling space charge in a cable geometry. IEEE
Trans. Dielectr. Electr. Insul. 23(4), 2361–2367 (2016)
10. Terashima, K., Sukuki, H., Hara, M., et al.: Research and development of ± 250 kV
DC XLPE cables. 7–16 (1998)
1066 Y. Wu et al.

11. Wang, X., Zhu, Y., Wang, C., et al.: Effect of space charge on electric field distribution at the
insulating interface between DC cable and accessory. High Voltage Eng. 41(8), 2681–2688
(2015)
12. Kai, W., Cheng, C.: Interface charge between insulating materials. IEEE Trans. Dielectr.
Electr. Insul. 24(4), 2633–2642 (2017)
 Study of Deep Dielectric Charging
Characteristics and Suppression Method
Under Space Irradiation Environment

Xiaoping Wang1, Shusai Zheng1, Daomin Min1, Shengtao Li1(&),

Xinbin Hou2, and Li Wang2
1
State Key Laboratory of Electrical Insulation and Power Equipment,
Xi’an Jiaotong University, Xi’an 710049, China
sli@mail.xjtu.edu.cn
2
Qian Xuesen Laboratory of Space Technology, Beijing 100094, China

Abstract. With the rapid increasing interest on the space exploration, the
reliability of the spacecraft becomes a very important problem. The space solar
power station (SSPS) is inevitably exposed to space plasma, energetic particles
radiation, extreme temperature, cosmic rays, etc. Energetic electrons can pene-
trate through the aluminum shield and deposit in the deep-layer of insulating
materials, leading to partial accumulation of space charges and high electric
field. Electrostatic discharge (ESD) occurs when the maximum electric field of
insulating materials exceeds a certain threshold, resulting in deterioration of the
insulating material and even the failure of the entire electronic equipment. Deep-
layer dielectric charging has been the key scientific issue for developing high-
voltage and high-power spacecraft technology. In this paper, a physical model is
established to simulate the deep-layer charging characteristics of ethylene-tetra-
fluoro-ethylene (ETFE) under FLUMIC spectrum electron irradiation, based on
the processes of carriers’ transport and deposition of charge and energy. Two
operating conditions, i.e. typical GEO condition and extreme GEO condition
with varied flux enhancement, are studied. In addition, the possibility of sup-
pressing the deep dielectric charging properties of ETFE by the addition of
nano-boroncarbide (nano-B4C) is also investigated. The calculation results
show that the maximum electric field in the deep-layer of ETFE rapidly reaches
108 V/m under extreme GEO space environment. Electrostatic discharge is
easily to take place as it exceeds the breakdown threshold. It is found that the
time spent to reach the maximum value of potential and electric field is less than
one minute under extreme space radiation environment. Furthermore, the
addition of nano-B4C can suppress the deep dielectric charging properties of
ETFE to a large extent by introducing more shallow traps. This provides a
potential approach on suppressing the deep dielectric charge accumulation.

Keywords: Deep charging characteristics
Space irradiation environment

Nanocomposite dielectric
Suppression method

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1067–1076, 2020.
https://doi.org/10.1007/978-3-030-31676-1_100
1068 X. Wang et al.

1 Introduction

As energy harvesting in space is not influenced by season, day and night, as well as
weather, space solar power station (SSPS) is a prospective way to resolve the problems
of global energy shortage. Reliability of SSPS is one of the key problems due to the
extreme and complex operating condition in geosynchronous orbit (GEO) [1, 2]. It is
inevitably exposed to the flow of energetic electrons [3]. The electrons are accumulated
in the insulating materials, inducing deep dielectric charging and building up an
internal electric field. Electrostatic discharge (ESD) may occur when it exceeds the
threshold field, resulting in damage of insulation system [4]. Deep-layer dielectric
charging has been the key scientific issue for developing high-voltage and high-power
spacecraft technology.
Studies on deep dielectric charging have been gradually increased since the 1990s
[5]. There are two main theoretical models have been proposed to study charge
transport properties of insulating materials under the energetic electrons radiation, i.e.
radiation induced conductivity model (RIC model) and charge generation-
recombination model (GR model) [6]. For theoretical analysis, numerical simulations
were employed to calculate the potentials both on surface and internal of the dielectric
materials [7]. For experimental research, extensive studies focused on the measurement
of radiation-induced conductivity, surface potential, secondary electron emission
coefficient, discharging pulse energy, breakdown threshold and other parameters under
simulated space radiation environment [8]. In our group previous work, the depen-
dences of deep dielectric charging on insulating material properties and thickness were
discussed [9].
Ethylene-tetra-fluoro-ethylene (ETFE), with excellent resistance to extreme tem-
perature, radiation resistance, has been widely used in special cables especially in the
field of aerospace insulation. In this paper, we focus on the deep dielectric charging
properties of ETFE and its composites under two kinds of radiation conditions, i.e.
typical space environment and extreme space environment. Besides, a potential
approach on suppressing the deep dielectric charge accumulation is discussed.

2 Deep Dielectric Charging Model Under Energetic Electron

Radiation
2.1 Space Electron Radiation Environment
FLUMIC model was adopted to describe the electron radiation environment in
geosynchronous orbit (GEO). The model was established by Rodgers et al., based on
spacecraft data of GOES/SEM and STRV-1b/REM [10, 11]. Owing to the complete
description of seasonal and annual variations in energetic electron flux, FLUMIC
model is suitable for spacecraft charging risk assessment and spacecraft design.
Energetic electrons in the outer radiation belt continually exhibit a series of discrete
enhancement events. The electron flux of energy from hundreds of keV to several MeV
will jump instantaneously about two or three orders of magnitude in a few days and last
for about 10 days, which is known as relativistic electron enhancement [12]. In this
 Study of Deep Dielectric Charging Characteristics 1069

paper, we take the electron flux enhancement as 100 and 1000 times magnitudes of the
value of FLUMIC standard spectrum for convenience. Figure 1 depicts FLUMIC
spectrum under typical and extreme space environment.

14
10
12
10
10
10
-1

8
10
m s sr
-2 -1

6
10
Flux

10
4 FLUMIC at GEO
FLUMIC at GEO*100
10
2
FLUMIC at GEO*1000
0
10
0.1 1 10
Energy(MeV)

Fig. 1. The FLUMIC model value at GEO environment.

2.2 Space Electron Radiation Environment

The charge deposition model in dielectric is established by analysing the charge
injection process and conduction process, as shown in Fig. 2.

(a) Dielectric Electrode (b) Dielectric Electrode

σdark+σRIC σdark
Non-
radiative Non-
zone radiative
Radiative
Incident zone Incident Charge deposition zone
electron Induced electron
Penetration range Penetration range
L charge L

Fig. 2. Schematic of charge injection, conduction and deposition in the material under electron
beam radiation: (a) interaction between energetic electrons and materials; (b) charge deposition
process in the medium.

Figure 2a depicts a schematic diagram of the interaction between energetic electron

and material. As a result of physical mechanism of elastic scattering or inelastic
scattering, incident electrons penetrate into materials and their energy will gradually
transfer into target atoms. The electron energy loss reaches the maximum at a certain
depth, and then the electron energy is gradually reduced to zero. Those electrons are
most likely to be captured by traps with the material, forming charge deposition. In
addition, electrons conduct toward the material under the action of the inherent
1070 X. Wang et al.

conductivity (rdark) and the radiation induced conductivity (rRIC). When deposit rate is
larger than charge leakage rate, the space charge is deposited continuously in the
material. The Weber semi-empirical formula [13] was used to obtain the penetration
depth of energetic electrons in the material:
 
aEP b
R¼ 1 ð1Þ
q 1 þ cEP

where:
R = Maximum penetration depth (m)
EP = Incident electron energy (MeV)
q = Density of the material (kg
m−3)
a, b, c = Constant (a = 5.5 kg m−2 MeV−1, b = 0.9841, c = 3 MeV−1)
The charge deposition rate is related to the penetration depth of the incident
electron and the density of the medium [13]:

@Q(x; t) x3 x
¼ 14:42Jp 4 exp(  3:605( )4 ) ð2Þ
@t R R
where:
∂Q(x, t)/∂t = Deposited electron density (C m−3 s−1)
JP = Incident electron current density (A m−2)
x = Penetration depth (m)
The energy deposition on the unit distance is related to the incident electron energy
and dielectric density [13]:

dE Ep x
= 1.544 exp(  2.2(  0.7)2 ) ð3Þ
dx R R
where: dE/dx = Energy deposition in the unit distance (MeV m−1)

2.3 The Physical Process of Charge Transport

The inherent charge conduction process of insulating materials is described by hopping
conductivity equation [13].

n0 qkT ET qkT Fi
rdark ¼ n0 ql ¼ t0 exp(  )sinh( ) ð4Þ
3Fi kB T 2kB T

where:
n0 = Concentration of ion (1/m3)
q = Charge of an ion (C)
l = Mobility (m2/V/s)
m0 = Attempt to escape frequency (s−1)
 Study of Deep Dielectric Charging Characteristics 1071

ET = Barrier height (eV)

Fi = Electric field (V/m)
kB = Boltzmann constant
T = Temperature (K)
kT = Hopping distance (m)
Radiation induced conductivity [13] is given by equations:

 
DðT Þ
rRIC D ¼ kRIC ðT Þ
D ð5Þ

DðT Þ ¼ ð1 þ T=TRIC Þ1 ð6Þ

DðTÞ
kRIC ðT Þ = kRIC0  kRIC1 ½T=TRIC 22DðT Þ
3
ð7Þ

where:
D = Dose rate
D(T) = Temperature index coefficient, usually between 0.5–1.0
TRIC = A constant related to temperature
kRIC0 = A constant independent with temperature
kRIC1 = Related to temperature
Charge transport process satisfy the current conduction equation, the charge con-
tinuity equation and Poisson’s equation [13].

j(x; t) ¼ rtot EF(x; t) ð8Þ

@qðx; tÞ @jðx; tÞ @Qðx; tÞ

þ ¼ ð9Þ
@t @x @t
@EFðx; tÞ qðx; tÞ
¼ ð10Þ
@x e0 er

rtot ¼ rdark þ rRIC ð11Þ

where:
j(x,t) = Conduction current density (A/m2)
rtot = Total conductivity including dark conductivity and RIC (S/m)
EF(x,t) = Electric field (V/m)
q(x,t) = Space charge density (C/m3)

2.4 Definition of Parameters in Simulation

A summary of the parameters of simulation is presented in Table 1. We measured the
density of ETFE and 1wt% B4C/ETFE. We calculated activation energy and hopping
distance according to the results of the SPD experiment by the method given in the
reference [13].
1072 X. Wang et al.

Table 1. Definition of parameters in simulation.

Parameter Symbol ETFE 1wt% B4C/ETFE
Temperature in K T 300 300
Electron radiation environment / FLUMIC spectrum FLUMIC spectrum
Density in kg∙m−3 q 1603.03 1711.57
Activation energy in eV EA 0.92 0.68
Hopping distance in m kT 1.44  10−9 1.50  10−9
RIC coefficient in S∙m−1 kRIC0 3.11  10−12 5.25  10−12
RIC coefficient related to temperature kRIC1 4.6  10−5 4.6  10−5
RIC constant in K TRIC 745.3 745.3

3 Results and Discussion

3.1 The Effect of Electron Flux Enhancement on the Charging Properties
of ETFE
Under typical space environment, the total charge deposition, the maximum electric
field amplitude and the maximum potential amplitude of ETFE irradiated by electrons
after 30 min, 60 min, and 90 min are shown in Table 2.

Table 2. Internal charging characteristics of ETFE irradiated by electrons after 30 min, 60 min
and 90 min under typical space environment.
Radiation Total charge Maximum electric field Maximum potential
time accumulation (C/m3) amplitude (V/m) amplitude (V)
30 min −2.49  105 4.65  107 3748
60 min −3.37  105 5.45  107 4138
90 min −4.31  105 5.59  107 4186

Figure 3 depicts the maximum electric field and potential of ETFE versus time
under typical space environment. After 90-minute radiation, the potential and the
electric field in ETFE has reached the maximum value. The electric field is still in the
107 V/m order of magnitude. Discharge threshold electric field depends on material
properties, general material takes 106 V/m–108 V/m [11]. For excellent insulating
materials like ETFE, the threshold is 107 V/m or higher. Therefore, under typical space
environment, the discharge risk is small.
 Study of Deep Dielectric Charging Characteristics 1073

7
6x10 0
7
5x10
-1000

Electric field (V/m)

4x10
7 Electric field (V/m)

Potential (V)
Potential (V)
7 -2000
3x10
7
2x10 -3000
7
1x10 -4000
0
-5000
0 20 40 60 80 100
Time (min)

Fig. 3. The maximum electric field and potential of ETFE versus time under typical space
environment.

When the flux is enhanced by 1000 times, the maximum potential magnitude of
ETFE reaches 23952 V after 10 s radiation, as shown in Fig. 4. The potential reaches a
balance state at 27860 V within 1 min. The corresponding maximum electric field
reaches 3.80  108 V/m, which exceeds the discharge threshold and will lead to
extremely high discharge risk.

8
4.0x10 3
-5.0x10
8
3.5x10
4
8 -1.0x10
3.0x10
Electric field (V/m)
Electric field (V/m)

Potential (V)

8 Potential (V)
2.5x10 4
-1.5x10
8
2.0x10
4
8
-2.0x10
1.5x10
8 4
1.0x10 -2.5x10
7
5.0x10
4
-3.0x10
0 50 100 150 200 250 300
Time (s)

Fig. 4. The maximum electric field and potential of ETFE versus time under extreme space
environment.

3.2 The Effect of Adding Nano-B4C on the Charging Properties of ETFE

To take the 1000 times electron flux enhancement of FLUMIC spectrum and radiation
time of 300 s as an example, this paper discussed the influence of adding nano-B4C on
the charging properties of ethylene-tetra-fluoro ethylene (ETFE). Total charge accu-
mulation of 1wt% B4C/ETFE reached −2.48  106 C/m3 after 300 s’ radiation. The
1074 X. Wang et al.

maximum electric field strength reached 2.25  108 V/m. The maximum electric field
appeared near the grounding electrode. The maximum potential amplitude was
16564 V. Most of the charge deposited near the radiated surface, though part of the
charge driven by the electric field mitigated toward the grounding electrode in the
electric field, as shown in Fig. 5.

Fig. 5. The distribution of charge deposition, electric field and potential: (a) Charge deposition
distribution; (b) Electric field distribution; (c) Potential distribution

Table 3 shows comparisons of charging properties under extreme radiation envi-

ronment between neat ETFE and 1wt% nano-B4C/ETFE. The total charge accumula-
tion in neat ETFE is more than that in 1wt% nano-B4C/ETFE after the same radiation
condition. Moreover, the maximum electric field amplitude and the maximum potential
amplitude in neat ETFE are also higher than that in 1wt% nano-B4C/ETFE.
Dangling bonds exist on the surface of nano-B4C, which provide the hopping site
for electrons. Therefore, the value of activation energy decreases and the hopping
distance increases by introducing nano-B4C. Besides, As a result of introducing dan-
gling bonds, the electron affinity is reduced and more shallow traps are generated in
B4C/ETFE. The hopping conductivity of carriers in B4C/ETFE is increased and the
space charge accumulation is suppressed. Consequently, the addition of nano-B4C is
beneficial to suppress the charge accumulation, reduce the maximum electric field and
improve the anti-internal charging properties of ETFE.

Table 3. Internal charging characteristics comparisons between neat ETFE and 1wt% nano-
B4C/ETFE.
Parameters Neat ETFE 1wt% B4C/ETFE
Radiation time 300 s 300 s
Radiation condition 1000 times electron flux 1000 times electron flux
enhancement enhancement
Total charge −4.05  106 C/m3 −2.48  106 C/m3
accumulation
The maximum 3.80  108 V/m 2.25  108 V/m
electric field
The maximum potential 27860 V 16564 V
0111famplitude
 Study of Deep Dielectric Charging Characteristics 1075

4 Conclusion

In this study, the deep charging characteristics of ethylene-tetra-fluoro-ethylene and its

composites under electron radiation were investigated. The charge transport model
under FLUMIC spectrum electron irradiation was established. With this model, we
studied the effect of electron flux enhancement and the addition of nano-B4C on the
charging properties. The potential and electric field in ETFE reach the maximum value
after about 90-minute typical space environment radiation. Under typical space envi-
ronment, the discharge risk is small. The time spent to reach the maximum value of
potential and electric field is less under extreme space radiation environment, which is
about 1 min. Besides, the extreme condition would lead to high discharge risk. The
addition of nano-B4C can suppress the deep dielectric charging properties of ETFE by
introducing more shallow traps. This would be a potential approach on suppressing the
deep dielectric charge accumulation.

Acknowledgement. This work was supported by the National Natural Science Foundation of
China (NSFC) under Project with No. 11575140.

References
1. Jaffe, P., Hodkin, J., Harrington, F., Person, C., Nurnberger, M., Nguyen, B., LaCava, S.,
Scheiman, D., Stewart, G., Han, A., Hettwer, E., Rhoades, D.: Sandwich module prototype
progress for space solar power. Acta Astronautica 94, 662–671 (2014)
2. Lior, N.: Power from space. Energy Convers. Manag. 42(15–17), 1769–1805 (2001)
3. Laurent, C., Teyssedre, G., Le Roy, S., Baudoin, F.: Charge dynamics and its energetic
features in polymeric materials. IEEE Trans. Dielectr. Electr. Insul. 20(2), 357–381 (2013)
4. Min, D.M., Cho, M., Khan, A.R., Li, S.T.: Surface and volume charge transport properties of
polyimide revealed by surface potential decay with genetic algorithm. IEEE Trans. Dielectr.
Electr. Insul. 19(2), 600–608 (2012)
5. Garrett, H.B., Whittlesey, A.C.: Spacecraft charging, an Update. IEEE Trans. Plasma Sci. 28
(6), 2017–2028 (2000)
6. Sessler, G.M., Figueiredo, M.T., Leal Ferreira, G.F.: Models of charge transport in electron-
beam irradiated insulators. IEEE Trans. Dielectr. Electr. Insul. 11(2), 192–202 (2004)
7. Jun, I., Garrett, H.B., Kim, W., Minow, J.I.: Review of an internal charging code, NUMIT.
IEEE Trans. Plasma Sci. 36(5), 2467–2472 (2008)
8. Le Roy, S., Baudoin, F., Griseri, V., Laurent, C., Teyssèdre, G.: Charge transport modeling
in electron-beam irradiated dielectrics a model for polyethylene. J. Phys. D Appl. Phys. 43,
315402, 1–10 (2010)
9. Li, G., Li, S., Min, D., et al.: Deep dielectric charging characteristics of ring structure
irradiated by energetic electrons. IEEE Trans. Dielectr. Electr. Insul. 22(4), 2349–2357
(2015)
10. Wrenn, G.L., Rodgers, D.J.: Modeling the outer belt enhancements of penetrating electrons
introduction. J. Spacecr. Rockets 37(3), 408–415 (2000)
11. Rodgers, D.J.: The FLUMIC electron environment model. In: 8th Spacecraft Charging
Technology Conference (2003)
1076 X. Wang et al.

12. Lai, S.T.: Fundamentals of spacecraft charging: spacecraft interactions with space plasma,
1st edn. Princeton University Press, Princeton (2011)
13. Pan, S., Min, D., Wang, X., et al.: Effect of electron irradiation and operating voltage on the
deep dielectric charging characteristics of polyimide. IEEE Trans. Nucl. Sci. (99), 1–1
(2018)
 Charge Traps Depended Space Charge
Dynamics and Electrical Breakdown
Characteristics of Polymer
Insulating Materials

Yuanwei Zhu1,2, Peng Wei1,2, Zichao Shen1,2, Huize Cui2, Yu Jing3,

Dongfan Li1,2, Zihao Wang1, Dongri Xie2, Guanghao Lu1,2(&),
and Shengtao Li2(&)
1
Frontier Institute of Science and Technology, Xi’an Jiaotong University,
Xi’an 710054, China
2
State Key Laboratory of Electrical Insulation and Power Equipment,
Xi’an Jiaotong University, Xi’an 710049, China
{guanghao.lu,sli}@xjtu.edu.cn
3
States Grid Shaanxi Economic Research Institute, Xi’an 710065, China

Abstract. Electrical breakdown of insulating materials unavoidably occurs in

power equipment, which endangers the safe operation of power systems. Recent
investigations suggest both shallow and deep traps contribute to improved
breakdown performance of insulating materials. This contradiction implies that
no uniform conclusion is formed at present stage, and thus suggesting the
influence of charge trap on electrical breakdown and its related space charge
dynamics are still to be further investigated. Here in this paper, electrical
breakdown experiments were firstly conducted on three typical polymer insu-
lating materials (low density polyethylene, polypropylene and polyimide). Then,
charge trap depth and trap density of the three polymers were characterized by
isothermal surface potential decay experiments, and the obtained information of
charge traps were linked to breakdown results. After that, space charge
dynamics in electrical breakdown processes of low density polyethylene with
variations of charge trap depth were systemically simulated based on an
improved bipolar charge transport model. It is found that the electrical break-
down strength of insulating materials are strongly related to trap depth of space
charges, as deep charge traps corresponds to enhanced electrical breakdown
strengths. The profile of homo space charges under DC stress is controlled by
charge trap depth, and space charge induced electric field distortion dominates
the DC breakdown performance of insulating materials. A turning point of DC
breakdown strength is found around 0.58 eV of trap depth, as both shallower
traps and deeper traps lead to increased electrical breakdown strengths.

Keywords: Breakdown
Space charge
Trap depth
Polymer

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1077–1086, 2020.
https://doi.org/10.1007/978-3-030-31676-1_101
1078 Y. Zhu et al.

1 Introduction

Electrical breakdown of insulating materials occurs in key power equipment such as

transformers, gas insulated switchgears (GIS), power cables, which greatly endangers
the safety operation of power systems. At present stage, it is widely accepted that space
charges induced by high electrical stresses modulate breakdown performance of
insulating materials [1, 2]. Based on this acknowledgement, the breakdown perfor-
mance of insulating materials influenced by space charge injection and accumulation
has been initially studied and some of the widely recognized conclusion were achieved
[3, 4].
More specifically, the origin of thickness dependent DC breakdown of polymer
insulating materials is found to be controlled by homo charges injected during the
voltage ramping procedure in 2012 [3]. After that, the related issue has extend to the
situations of AC electrical stresses, and some widely accepted conclusions can be
drawn through both of experiments and numerical simulations, suggesting that space
charges accumulated near the interface between electrodes and the material dominate
AC breakdown of insulating materials [1, 4].
Some further investigations suggest that the procedures of space charge migration
and accumulation are controlled by the depth and amount of charge traps formed at
physical and chemical defects in insulating materials [5–7]. Theoretically, based on the
investigations in last decade, shallow traps suppress the non-uniform electric field
distortion, resulting in improved electrical breakdown performance [5]. Deep trap can
firmly capture free charges in materials, which reduces free charges involved in electric
conduction, leading to enhanced electrical breakdown voltages [6]. Summarily, recent
investigations suggest both of shallow and deep traps contribute to improved break-
down performance of insulating materials. This contradiction implies that no uniform
conclusion is formed with the influence of charge trap depth on electrical breakdown
and its related space charge dynamics of insulating materials at present stage.
Based on the present inadequate understanding, a combined investigation of
experiment and simulation on charge traps depended electrical breakdown performance
of polymeric insulating materials is carried out in this paper. Experimentally, the
electrical breakdown characteristics of three typical polymer insulating materials (Low
density polyethylene (LDPE), polypropylene (PP), and polyimide (PI)) are tested. The
charge trap depth and density are characterized by isothermal surface potential decay
(ISPD) experiment, and the obtained information of charge traps are linked to break-
down results. Based on an improved bipolar charge transport (BCT) model, the space
charge dynamics under ramping voltages in electrical breakdown processes of LDPE
are simulated. The space charge induced electric field distortion is described and is
comparatively analyzed with experimental breakdown results.
 Charge Traps Depended Space Charge Dynamics and Electrical Breakdown 1079

2 Setups
2.1 Electrical Breakdown
DC and AC electrical breakdown experiments of PP, PI and LDPE are carried out with
brass sphere-plate electrode. In experiments, the polymer films are clamped by the
electrode, and are kept in Karamay 25# transformer oil. High voltage is stress to the
upper sphere, while the lower plate electrode is grounded.
For each sample, the experiment was conducted for 15 times to ensure the reliable
data. The voltage ramping rate is 1 kV/mm, and the AC breakdown strength is
recorded in peak values.

2.2 Isothermal Surface Potential Decay

ISPD experiments were carried out to obtain the charge trap depth and trap density of
each polymer material. The ISPD testing system includes a mesh grid electrode placed
3 cm above the sample, and a needle electrode placed 7 cm above the mesh grid.
A schematic of the testing system can be found in our earlier investigation [1]. In
experiments, the charging voltage at the needle electrode is set to 10 kV, the bias
voltage applied to the mesh grid is 5 kV, and the charging duration is 300 s. After
charging, the polymer film is immediately moved under a testing probe for a 10000 s
surface potential test.

2.3 Bipolar Charge Transport Model

The space charge dynamics in polymer materials during DC breakdown process are
numerically simulated based on an improved bipolar charge transport model, as shown in
Fig. 1. One dimensional coordinate, x is set up, as x = 0 corresponds to the interface of
cathode and the sample, and x = L corresponds to the interface of anode and the sample.

Fig. 1. A schematic diagram of bipolar charge transport model under DC voltage.

As polymeric materials are commonly composed of amorphous regions and cross-

connected crystalline regions, the free charges injected into the material must jump
over a potential barrier to migrate between two localized areas. This kind of charge
1080 Y. Zhu et al.

migration obeys the law of charge hopping model in electric conduction mechanism of
dielectric materials, which is applied in the simulation program.
The simulation of charge transport of insulating materials is governed by a set of
self-consistent equations, as:
(a) Charge injection,

1=2 !
/in  q3e EF 4pe0 er
jin ¼ AT exp 
2
ð1Þ
kB T

(b) Charge transport,

jc ðx; tÞ ¼ Qðx; tÞlEF ðx; tÞ ð2Þ

(c) Charge advection-reaction,

@Qðx; tÞ @jc ðx; tÞ

þ ¼ Sðx; tÞ ð3Þ
@t @x
(d) Poisson’s equation,

@/2 ðx; tÞ Qðx; tÞ

¼ ð4Þ
@x2 e0 er

(e) Charge dynamics,

Sðx; tÞ ¼ RQe ðx; tÞQh ðx; tÞ ð5Þ

where jin is Schottky emission current in A/m2; A is the Richardson constant; T is the
absolute temperature in K; /in is the Schottky injection barrier in eV; EF is the electric
field in V/m; kB is the Boltzmann constant; qe is the elementary charge in C; e0 is the
vacuum permittivity in F/m; and er is the dielectric constant of the material (Table 1).

Table 1. The values of the parameters

Parameters Unit Value
Voltage ramping rate V/s 1000
Electron hopping distance m 2.6  10−9
Hole hopping distance m 1.4  10−9
Electron trap depth eV 0.40–0.75
Hole trap depth eV 0.45–0.80
Temperature K 300
Sample thickness m 1.0  10−4
Spatially discrete grid 200
Calculating time step s 1  10−6
 Charge Traps Depended Space Charge Dynamics and Electrical Breakdown 1081

3 Experimental Results
3.1 Electrical Breakdown of LDPE, PP, and PI
DC and AC breakdown tests were conducted with thickness of 100 um LDPE, PP and
PI, the details of the test are described in Sect. 2.1 and the results are shown in Fig. 2. It
is noticed that clear differences in breakdown voltages exist. The average DC break-
down strength of PI reaches 415.78 kV/mm, which is the highest among the three
insulating films, followed by 371.45 kV/mm of PP and 347.16 kV/mm of LDPE.
The AC breakdown strengths of the films are nearly 30% of their DC breakdown
strengths, and the trend of AC breakdown strengths are in consistence with that under
DC stresses. The mechanism of this phenomenon is explained in earlier investigation as
a result of differences of space charge accumulation profiles under DC and AC stresses
[1].

Fig. 2. DC and AC breakdown strengths of LDPE, PP and PI.

3.2 Characterization of Charge Traps of LDPE, PP, and PI

In order to analyze the charge trap depth and trap density of the three insulating films,
ISPD experiments were conducted. The details of the experiments are described in
Sect. 2.2, and the results are shown in Fig. 3(a). It can observed that the initial surface
potential of PI is the highest, suggesting good space charge storage characteristics (in
other words, higher charge trap density). More importantly, the decay of PI and PP are
obviously slower than LDPE, which indicate deeper charge traps in the polymer films.
To quantitatively analyze the charge trap characteristics of the three insulating
polymers, an improved Simmons model is applied [1], in which,

ET ¼ kB T lnðmATE tÞ ð6Þ
1082 Y. Zhu et al.

e0 er d/s ðtÞ
Qs ðtÞ ¼ t ð7Þ
qe L dt

where ET is trap depth in eV; mATE is the attempt-to-escape frequency; Qs(t) is trap
density in m−3; /s is the surface potential in V.
The variations of electron trap density with changes of trap depth is shown in
Fig. 3(b). It can be observed that the trap depth of the three films are all in range of
0.68–0.93 eV. The average trap depth of PI is the largest, while LDPE is the smallest.
These results are consistent with the electrical breakdown results in Fig. 2, indicating
strong connection of breakdown strength with charge trap depth in polymer insulating
materials. Furthermore, the amount of charge traps inside the material can be obtained
from the integral of the ISPD curve, however, this doesn’t seem to greatly affect the
breakdown performance, as the differences of trap densities are less than an order of
magnitude among the three materials.

(a) surface potential decay curves (b) electron trap density and trap depth

Fig. 3. ISPD results of LDPE, PP and PI.

4 Discussions

It is clearly demonstrated that differences in breakdown strengths exist among LDPE,

PP and PI films, and the polymer insulating material with deeper charge traps corre-
sponds to higher electrical breakdown strength under both DC and AC voltages.
Therefore, it reasonable to consider that the formation of deep charge traps contributes
to better electrical insulating performance of polymer materials.
In order to obtain a more in-depth understanding of charge traps depended electrical
breakdown of insulating materials, to avoid the interference of other factors, numerical
simulations are carried out with the same dielectric constant, temperature, and appli-
cation method of breakdown voltage on LDPE. The space charge dynamics and their
induced electric field distortions are systemically investigated with the variation of
charge trap depth (0.4–0.8 eV), as shown in Fig. 4.
 Charge Traps Depended Space Charge Dynamics and Electrical Breakdown 1083

(a) space charge profiles with 0.48 eV (b) space charge profiles with 0.63 eV
electron traps electron traps

(c) electric field distribution with 0.48 eV (d) electric field distribution with
electron traps 0.63 eV electron traps

Fig. 4. Space charge profiles of LDPE with variations of charge trap depth in DC breakdown

It can be observed that the space charge accumulation profiles vary greatly with the
changes of charge trap depth. In Fig. 4(a), for electron trap of 0.48 eV (and hole trap of
0.53 eV), charges can easily migrate to the other electrode under the applied DC
breakdown stress. The migration distance of electrons is much longer compared to
holes, owing to their relatively larger mobility. A clear interface can be found at about
20 lm near the anode during the whole voltage ramping procedure. Therefore, charge
recombination occurs during electrical breakdown under the situation of 0.48 eV
electron traps. Both of electron and hole density at every position inside the sample
raise as a result of ramped applied voltage. At the moment of pre-breakdown, the
electron density reaches −78.3 C/m3 in positions of 0–6  10−5m, and hole density
reaches 519.6 C/m3 at the position near anode.
In Fig. 4(b), the space charge profile under the situation of electron trap of 0.63 eV
(and hole trap of 0.68 eV) shows clear difference compared to Fig. 4(a). The space
charge migration is greatly suppressed owning to the relatively deeper charge traps.
Both of electrons and holes accumulate near the electrode of their charge injection.
Electron migration can be clearly observed with the increase of application duration of
1084 Y. Zhu et al.

the ramped voltage. At 31 s of pre-breakdown, electrons migrate less than 30 lm,

while holes only migrate for 4 lm. As the charge accumulation position is much closer
to the electrode, the maximum space charge density becomes larger when compare to
Fig. 4(a), as electron density reaches –360.2 C/m3, while hole density reaches
4585.3 C/m3. For the positions deeply inside the sample, no space charge accumulation
is achieved during the whole breakdown process, which is obviously different with the
situation of electron trap depth of 0.48 eV.
Space charges accumulation can tremendously distort the electric field distribution
inside the insulating materials. The electric field distributions correspond to Fig. 4(a)
and (b) are solved based on Poisson’s equation. The dotted lines in Fig. 4(c) and
(d) show real-time applied electric field. For ET-e = 0.48 eV, no obvious electric field
distortion is observed during the early 10 s in breakdown process. As the applied
electric field ramped to larger value, observable electric field distortion is demonstrated.
As homo charges lead to a unidirectional change electric field, the maximum electric
field distortion occurs at the interface between electrons and holes. At pre-breakdown
moment of 25 s, the maximum electric field inside the sample reached 3.5108 V/m,
which is 140% of its applied electric field of 2.5108 V/m.
For the situation of ET-e = 0.63 eV in Fig. 4(d), the profile of electric field dis-
tortion is obviously different with ET-e = 0.48 eV, as the value of electric field deeply
inside the sample keeps constant, while the electric field distortion mainly occurs near
the electrodes where most of the space charges accumulate. As the space charge density
near cathode is relatively larger compared to ET-e = 0.48 eV, space charge induced
electric field distortion can be observed from earlier moment (5 s) during the whole
breakdown process. However, the maximum electric field distortion is comparatively
smaller at the pre-breakdown moment, as it is only reached 113% of the applied electric
field (3.1108 V/m).

Fig. 5. The variations of DC breakdown voltages with change of electron trap depth.
 Charge Traps Depended Space Charge Dynamics and Electrical Breakdown 1085

Electric field distortion is one of the dominate factors that influence DC breakdown
of insulating materials. As the results above discuss the space charge accumulations
and their induced electric field distortions, the influence of charge trap depth on DC
breakdown voltages of LDPE is generally demonstrated based on numerical simula-
tions, as shown in Fig. 5.
In Fig. 5, a tuning point of breakdown voltage is observed near 0.58 eV of electron
traps, as both of shallower traps and deeper traps lead to the elevated electrical
breakdown voltages of LDPE. Due to the limitation of simulation program, the shal-
lowest electron trap depth is simulated to 0.40 eV, as shallower traps could lead to
oscillating of the simulation program. When comparing the results of Figs. 4 and 5, it is
concluded that trap depth dominate space charge accumulation, and space charge
induced electric field distortion during DC breakdown, which ultimately modulates the
breakdown performance, as larger maximum electric field distortion lead to easier
breakdown process, resulting in smaller breakdown voltages.

5 Conclusions

Generally, the scientific fundamental issue of charge trap depth depended electrical
breakdown characteristics of insulating material is analyzed and discussed in this
investigation through a combination of experiments and numerical simulations.
The breakdown experimental results of PP, LDPE and PI films show strong rela-
tionship of electrical breakdown voltages and charge trap depth, as deeper charge traps
(*0.75 eV) lead to enhanced electrical breakdown strength. Numerically, it is found
that space charge induced electric field distortion greatly affect the DC breakdown
performance of insulating polymers, as larger electric field distortion can tremendously
decrease the breakdown voltage. Moreover, a turning point of breakdown voltage is
found around 0.58 eV of trap depth, as both shallower traps and deeper traps lead to
increased electrical breakdown strengths.
The results could promote better understanding on charge traps related electrical
insulation properties of polymeric materials, and provide a potential possibility to
improve insulation characteristics of polymer materials by turning their charge trap
characteristics at the stage of material design.

Acknowledgments. The authors thank the financial support of China Postdoctoral Science
Foundation (Grant No. 2018M643648), the State Key Laboratory of Electrical Insulation and
Power Equipment (Grant No. EIPE 19308), the Fundamental Research Funds for the Central
Universities, and the National Science Foundation of China (NSFC) under projects with Nos.
51473132 and 21574103.
1086 Y. Zhu et al.

References
1. Li, S., Zhu, Y., Min, D., Chen, G.: Space charge modulated electrical breakdown. Sci. Rep. 6,
32588 (2016)
2. Zhu, Y., Li, S., Min, D., Li, S., Cui, H., Chen, G.: Space charge modulated electrical
breakdown of oil impregnated paper subjected to AC-DC combined voltages. Energies 11,
1547 (2018)
3. Chen, G., Zhao, J., Li, S., Zhong, L.: Origin of thickness dependent DC electrical breakdown
in dielectrics. Appl. Phys. Lett. 100, 222904 (2012)
4. Chen, G., Fu, M., Liu, X., Zhong, L.: AC aging and space-charge characteristics in low-
density polyethylene polymeric insulation. J. Appl. Phys. 97, 083713 (2005)
5. Li, S., Min, D., Wang, W., Chen, G.: Linking traps to dielectric breakdown through charge
dynamics for polymer nanocomposites. IEEE Trans. Dielectr. Electr. Insul. 5(23), 2777–2785
(2015)
6. Zheng, F., Dong, J., Zhang, Y., An, Z., Lei, Q.: Reduction of space charge breakdown in e-
beam irradiated nano/polymethyl methacrylate composites. Appl. Phys. Lett. 97, 012901-1
(2013)
7. Zhu, Y., Li, S., Min, D.: Origin of dielectric processes in aged oil impregnated paper. IEEE
Trans. Dielectr. Electr. Insul. 24, 1625–1635 (2017)
 Effect of Deep Traps and Molecular
Motion on Dc Breakdown
of Polyethylene Nanocomposites

Rui Mi1, Chenyu Yan1, Zhaoliang Xing2, Qingzhou Wu3,

Daomin Min1(&), and Shengtao Li1
1
State Key Laboratory of Electrical Insulation and Power Equipment,
Xi’an Jiaotong University, Xi’an 710049, China
forrestmin@xjtu.edu.cn
2
State Key Laboratory of Advanced Power Transmission Technology,
Global Energy Interconnection Research Institute Co., Ltd.,
Beijing 102209, China
3
Institute of Fluid Physics, China Academy of Engineering Physics,
Mianyang 621900, China

Abstract. Low-density polyethylene (LDPE) has been widely used as an

insulating material in high-voltage direct-current power cables. In this research,
we investigate how to improve the electrical breakdown strength of LDPE by
nanodoping method and the mechanism of improvement. MgO particles with an
average diameter of 50 nm are mixed with LDPE to fabricate nanocomposites
by using a toque rheometer. Five kinds of nanocomposite samples are fabricated
with nanofiller loadings of 0.25 wt%, 0.5 wt%, 1 wt%, 2 wt%, 3 wt% and pure
LDPE is made as the contrast. Then the nanocomposites are pressed into sheet
samples about 100 lm by plate vulcanizing machine. The images observed by
scanning electron microscope show nanoparticles are dispersed uniformly in
LDPE matrix. X-ray diffraction is used to measure the bonding effect between
nanoparticles and polymer matrix as well as the morphology of nanocomposites.
The trap parameters such as trap levels are characterized by thermally stimulated
depolarization current. The dc breakdown experiments indicate that the dc
breakdown strength increases firstly and then decreases with an increase in
nanofiller loading. The dc breakdown strength is enhanced by incorporating
nano MgO and reaches the maximum value 377.06 kV/mm at around 0.5 wt%,
which is 17.61% higher than the breakdown field of pure LDPE. The influences
of bonding effect, morphology, and trap properties on dc electrical breakdown
strength of LDPE nanocomposites are analyzed. It is found that incorporating a
small amount of MgO nanoparticles into LDPE matrix enhance the bonding
effect between nanoparticles and polymer matrix and establish isolated interfa-
cial regions around nanoparticles. Then, deep traps are formed in the interfacial
regions and molecular chains with occupied deep charges are difficult to move
under electric force. Consequently, the dc electrical breakdown performance is
improved. At higher nanofiller loadings, bonding effect is weakened and
interfacial regions are overlapped so that carriers can migrate more easily and
the dc electrical breakdown field is reduced.

Keywords: Charge transport
DC electrical breakdown
Molecular

displacement
Low-density polyethylene
Nanocomposite dielectrics

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1087–1096, 2020.
https://doi.org/10.1007/978-3-030-31676-1_102
1088 R. Mi et al.

1 Introduction

High-voltage direct-current (HVDC) power transmission systems increasingly occupy

the power transmission market owing to its advantages of low electrical loss, stability
and less expensive [1–3]. Low-density polyethylene is one of the widely used basic
electrical insulating material for HVDC power cables [4, 5] as for its’ great electrical
insulation characteristics [6, 7]. Studying the electrical breakdown strength of this kind
of insulating material has a great influence on its’ application.
Doping nanofillers is an useful method to improve the dc breakdown strength of
LDPE insulating materials [8–10], which enhances the electrical breakdown strength
firstly and then declines it with the increasing of nanofiller loadings [11, 12]. In
addition, nanofillers doping can also affect the crystalline properties, dc conductivity
properties and trap distributions of nanocomposites [13–16]. Xu et al. observed that the
crystallinity of the LPDE/MgO nanocomposites initially increases and then decreases
with the increasing of the loading. Nilsson et al. found that the dc conductivity reduces
firstly and then increases when the nanofiller loading increases. In [15], LDPE/Al2O3
nanocomposites was investigated and the author obtained that the deep trap energy
level has an increase and then lower down with increase of nanoparticle loading. The
simulation to study the relation of molecular displacement with dc breakdown of LDPE
and investigations about the effects of the thickness and temperature on dc breakdown
[17, 18] have already emerged. However, it is still unclear how deep traps and crys-
talline structure affect dc breakdown strength. To further explore the problem, in the
work, the LDPE/MgO nanocomposites were prepared to investigate the relation of
electrical breakdown strength, crystallinity, crystallite size, trap distributions and dc
conductivity.

2 Experimental

2.1 Materials
LDPE (FT6230) having melt flow rate of 2.0 g/10 min at 190 °C/2.16 kg and density
of 0.922 g/cm3 was used as the polymeric matrix. Milky-white spherical LDPE has a
low crystallinity and a melt temperature about 111 °C. MgO nanoparticles modified by
c-(2,3-expoxypropoxy) propyltrimetho-xysilane having purity of 99.9%, average size
of 50 nm and specific surface area of 50 m2/g were selected as the doping nanofillers.

2.2 Sample Preparation

Two-step melt blending method was carried out to make MgO/LDPE nanocomposites.
The main details are as follows. LDPE and MgO nanoparticles were dried in a vacuum
oven at 80 °C and 10 °C, respectively. Then nanoparticles were mixed with LDPE matrix
in the Torque Rheometer (RM-200C). The processing conditions were set as following:
experimental temperatures in 1, 2 and 3 districts: 110 °C, 100 °C and 110 °C, motor
speed: 100 rpm and mixing time: 30 min. In order to get a better dispersion effect of
nanoparticles, the mixing processes were divided into two steps. First, prepare nanofiller
 Effect of Deep Traps and Molecular Motion on Dc Breakdown 1089

loading of 20 wt% nanocomposites. And then, samples with loadings of 0.25 wt%,
0.5 wt%,1 wt%, 2 wt% and 3 wt% were obtained by using pure LDPE matrix to dilute
the loading of 20 wt% prepared in the first step and the neat LDPE was made as the
contrast.
Then films about 100 lm in thickness were prepared by Plate Vulcanizing machine
(YT-LH-20B). The experiments were performed the following steps. First, block
samples were preheating for 25 min without pressure. Then, to avoid bubbles in the
sheet samples during the pressing process, a step-by-step boosting method is adopted,
successively pressed at 5 Mpa, 10 Mpa and 15 Mpa for 5 min each pressure. Finally,
shut down the power and let the sample cooling down naturally. In this way, two sizes
of specimens were made, round sheets about 50 mm in diameter and square ones with
each length of 8 cm in shape. Based on the reason of eliminating the thermal history,
specimens for direct current (dc) breakdown experiments were placed into a vacuum
oven at 95 °C for 30 min, and then cooled down to 30 °C in a vacuum environment.
Then other films were put into an oven at 60 °C for 24 h.

2.3 Characterization
The scanning electron microscope (SEM) experiment was utilized to examine the
dispersion of MgO nanoparticles inside the LDPE matrix. The films were frozen in
liquid nitrogen, broken and then sputtered the fractured sections with gold which would
be observed under the potential of 5 kV or 15 kV.
X-ray diffraction (XRD, D8 ADVANCE A25) was applied to measure the crys-
tallinity and crystallite size of samples. The measured angle (2h) ranged from 10° to
50° with the scanning rate of 0.02°/s. After all measurements finished, the
DIFFRAC EVA software was used to analyze the data.
The conductivity is calculated out through the volume resistivity experiment.
Volume resistivity of dielectric nanocomposites were tested by the Model 65 High
Resistivity Measurement. The measurements were performed under ac voltage about
500 V at room temperature. The square sheet samples with a side length of 8 cm were
applied in volume resistivity tests. Each sample was measured for 11 times and 3 times
of them should be discarded to decrease the error.
The trap parameters were measured with thermally stimulated depolarization
(TSDC, Concept 90) in a temperature range from −100 °C to 90 °C. The test condi-
tions are following: polarization voltage: 250 V, polarization temperature: 70 °C,
polarization time: 20 min, and short-circuiting time: 3 min. The film samples should be
sputtered gold both sides firstly before the test.
The dc electrical breakdown strengths of samples were tested through two spherical
copper electrodes at room temperature. The measurement was performed at the dc ramp
rate of 500 V/s till the sample failed. Each kind of specimen was tested at least 20 times
in this experiment.
1090 R. Mi et al.

3 Results
3.1 Dispersion of Nanofillers in Polyethylene
The dispersion of MgO nanoparticles inside the film samples has been observed by
SEM, as shown in Fig. 1. The MgO nanoparticles can be clearly seen in the samples
and its number increases with the filler loading increasing. Generally speaking,
nanoparticles disperse well in LDPE matrix. Of course, it can be observed that the size
of most nanoparticles in samples is less than or just about 100 nm.

Fig. 1. SEM images of six kinds of MgO/LDPE nanocomposites: (a) 0 wt%, (b) 0.25 wt%,
(c) 0.5 wt%, (d) 1 wt%, (e) 2 wt%, (f) 3 wt%.

3.2 Crystalline Properties

XRD measurement was performed in characterizing the crystalline properties of
samples, as shown in Fig. 2. It is clearly seen that the crystallinity firstly increases and
then begins decreasing at loading of 0.5 wt% with the increase of nanofiller loading,
and the crystallinity reaches the maximum value at 0.5 wt% in XRD spectra. The
crystallinity nearly keeps unchanged above the loading of 0.5 wt% (40.2%). It can be
easily concluded that adding a small amount of nano MgO can increase the crystallinity
of the LDPE dielectric nanocomposite. Figure 2(b) shows the crystallite sizes of
samples which are calculated by following formula [19].

0:89k
D¼ ð1Þ
x cos h
where D is the crystallite size, k represents the wavelength of X-rays (k = 0.15406 nm),
x denotes the full width at half maximum (FWHM), and h indicates the Bragg angle.
 Effect of Deep Traps and Molecular Motion on Dc Breakdown 1091

41
20.50 Crystallite size
(a) Crystallinity (b)
40

Crystallite size (nm)

Crystallinity (%)

20.25
39

38 20.00

37
19.75
36

0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Nanofiller loading (wt%) Nanofiller loading (wt%)

Fig. 2. Results measured by XRD. (a) Calculated crystallinity of different nanofiller loadings of
LDPE/MgO from XRD. (b) Calculated crystallite size.

From the crystallite size curve, it is easily can be find that the crystallite size of samples
decreases firstly and then increases with the nanofiller loading increasing and has the
minimum value at 0.5 wt%. The change trend of crystallite size is just against with that
of crystallinity.

3.3 DC Conductivity
Figure 3 shows the dc conductivity of film samples with different nanofiller loadings. It
can be easily observed that the conductivity firstly decreases until at 0.5 wt% and then
increases with the increasing of nanofiller loading. And the dc conductivity of samples
reaches the minimum value about 5.49  10−15Sm−1.

-14 Conductivity
3.0x10
Conductivity(Sm )
-1

-14
2.0x10

-14
1.0x10

0.0 0.5 1.0 1.5 2.0 2.5 3.0

Nanofiller loading(%)

Fig. 3. Conductivity as a function of nanofiller loading.

3.4 TSDC Measurement

Figure 4 shows the TSDC results. It is easy to look out two obvious relaxation peaks
(known as a and c) from the TSC curves, and the temperatures of the a peaks are
shown in Table 1. The location of another relaxation peak is between the a and c
1092 R. Mi et al.

-9
0wt% 0.25wt%

Current density (Am )

1.2x10

-2
0.5wt% 1wt%
2wt% 3wt%
-10
9.0x10

-10
6.0x10

-10
3.0x10

0.0
-100 -50 0 50
ºC

Fig. 4. Results of TSDC measurement.

peaks, named as the b peak. The incorporation of nano-magnesia has an impact on TSC
curves. The a peak reveals the deep traps distribution. The trap parameters can be
obtained by fitting TSC results to the following equation [20]:
 Z T1   
Eb 1 Eb
jðTÞ ¼ B exp   exp  dT ð2Þ
kB T as0 T0 kB T

where j(T) is the TSDC current density in Am−2, B is a constant in Am2, Eb is the
activation energy of relaxation process in eV, a is the heating rate in Ks−1, s0 is the
relaxation time constant in s, kB is the Boltzmann constant, T1 is the temperature of
sample after heating in K, T0 is the initial temperature of sample at the beginning of
heating in K, and T is the temperature in K.
The parameters of trap energy of samples are shown in Table 1. The trap energy
levels decrease with an increasing in nanofiller loading.

3.5 DC Breakdown
In this work, two parameter Weibull distribution is applied to analyze the dc break-
down strength and is described as follows [21]:
  u 
E
PE ¼ 1  exp  ð3Þ
Ec

i  0:44
PE ði; nÞ ¼ ð4Þ
n þ 0:25

where E is the dc breakdown field value arranged from small to large in kVmm−1, PE
indicates the accumulative probability of dc breakdown strength at E, Ec represents the
characteristic electrical breakdown strength with accumulative probability of 63.2%, u
is shape distribution parameter, and n is the testing time for the same sample.
 Effect of Deep Traps and Molecular Motion on Dc Breakdown 1093

Table 1. The temperatures of a peaks and its calculated trap energy levels
Samples(wt%) Temperature(oC) Etrp(eV)
0 81 1.24
0.25 84 1.33
0.5 83 1.4
1 83.5 1.13
2 63 0.53
3 64 0.45

The Weibull cumulative probability of dc breakdown strength of the LDPE/MgO

nanocomposites is shown in Fig. 5. In addition, the fitting curves choose the trans-
formation of log[−ln(1 − P)] as the ordinate and the breakdown strength as the
abscissa. And Fig. 6 shows the dc characteristic breakdown field of LDPE/MgO
dielectric nanocomposites.
Webull parameter, log(-ln(1-P))

0.5

0.0
P=63.2%
-0.5 0wt%
0.25wt%
-1.0 0.5wt%
1wt%
-1.5 2wt%
3wt%
-2.0
250 300 350 400
-1
Electric field (kVmm )

Fig. 5. Weibull cumulative probability of dc breakdown strength of the LDPE/MgO nanocom-

posites with various nanofiller loadings.

The dc breakdown strength increases firstly and then decreases with an increase in
nanofiller loading. Slight nano-magnesia doping enhances the dc breakdown strength
of LDPE/MgO nanocomposites apparently, which reaches the maximum value
377.06 kVmm−1 (17.61% higher than the breakdown field of pure LDPE which is
320.59 kVmm−1) at around 0.5 wt%.

4 Discussion

Figure 7 shows the relationship between characteristic breakdown strength, crys-

tallinity and crystallite size. It is easy to know that both dc breakdown strength and
crystallinity increase at low nanofiller loading and then decrease when the loading
exceeds 0.5 wt% from Fig. 7(a). They get the maximum value at the particle loading of
0.5 wt%. Figure 7(b) indicates that the dc breakdown strength declines with the
1094 R. Mi et al.

Characteristic breakdown strength (kVmm )

-1
375

360

345

330 Characteristic breakdown strength

315
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Nanofiller loading (wt%)

Fig. 6. DC characteristic breakdown strengths in LDPE/MgO nanocomposites.

crystallite size increasing. The correlation between dc conductivity and dc breakdown

with trap energy level is shown in Fig. 8. As the increasing of trap energy levels, dc
conductivity decreases and dc breakdown strength increases just opposite to dc
conductivity.

Fig. 7. Relationship between characteristic breakdown strength, crystallinity and crystallite size,
(a) characteristic breakdown strength and crystallinity; (b) characteristic breakdown strength and
crystalline size.

The crystalline properties and traps are related to the molecular structure of the
LDPE/MgO nanocomposite. And the influences reflect on the conductivity and the
breakdown strength. The hypothesis can be proposed that adding a small amount of
MgO nanoparticles into LDPE matrix can enhance the bonding effect between
nanoparticles and polymer matrix and establish isolated interfacial regions around
nanoparticles. Then, deep traps are formed in the interfacial regions and molecular
chains with occupied deep charges are difficult to move under electric force. Conse-
quently, the dc electrical breakdown performance is improved. At higher nanofiller
loadings, bonding effect is weakened and interfacial regions are overlapped so that
carriers can migrate more easily and the dc electrical breakdown field is reduced.
 Effect of Deep Traps and Molecular Motion on Dc Breakdown 1095

Characteristic breakdown strength (kVmm )

-14
3.0x10 375

Conductivity(Sm )
-1
360
-14
2.0x10
345

-14
1.0x10 330

315
0.3 0.6 0.9 1.2 1.5
Trap energy levels (eV)

-1
Fig. 8. DC conductivity and dc breakdown strength as a function of trap energy levels.

5 Conclusion

In this work, the LDPE/MgO nanocomposites of nanofiller loadings of 0 wt%, 0.25 wt%,
0.5 wt%, 1 wt%, 2 wt% and 3 wt% were prepared. Then SEM, XRD, TSC, dc break-
down, and dc conductivity experiments were processed to test the properties. And it can
be concluded as follows:
1. With the increasing of nano MgO loadings increasing, the crystallinity increases
and crystallite size decreases initially and then begins to decrease and increase at the
loading of 0.5 wt%, respectively. Nanofillers has an impact on the morphology
which is reflected in deep traps, and further embodied in the dc conductivity and dc
breakdown strength.
2. A slight incorporation of nano MgO can enhance the dc breakdown strength about
17.61% compared with pure LDPE matrix which happens at the loading of 0.5 wt%.
As the increasing of trap energy levels, dc conductivity decreases and dc breakdown
strength increases.
In conclusion, the deep traps and molecular motion have effects on the dc break-
down properties of LDPE/MgO nanocomposites.

Acknowledgments. This work was supported by the State Key Laboratory of Advanced Power
Transmission Technology (Grant No. GEIRI-SKL-2018-010), the National Basic Research
Program of China (grant No. 2015CB251003), and the National Natural Science Foundation of
China (grant No. 51507124).

References
1. Jin, L., Du, B.X., Hang, X.: Suppressing interface charge between LDPE and EPDM for
HVDC cable accessory insulation. IEEE Trans. Dielectr. Electr. Insul. 24(3), 1331–1339
(2017)
2. Liu, D., Hoang, A.T., Pourrahimi, A.M., Pallon, L.K.H., Nisson, F., Gubanski, S.M.,
Olsson, R.T., Hedenqvist, M.S., Gedde, U.W.: Influence of nanoparticle surface coating on
electrical conductivity of LDPE/Al2O3 nanocomposites for HVDC cable insulations. IEEE
Trans. Dielectr. Electr. Insul. 24(3), 1396–1404 (2017)
1096 R. Mi et al.

3. Lu, T., Feng, H., Zhao, Z., Cui, X.: Analysis of the electric field and ion current density
under ultra high-voltage direct-current transmission lines based on finite element method.
IEEE Trans. Magn. 43(4), 1221–1224 (2007)
4. Chahal, J.S., Reddy, C.C.: Space charge dynamics in LDPE. In: International Conference on
Condition Assessment Techniques in Electrical Systems. Bangalore, India (2016)
5. Wang, S., Zha, J., Wu, Y., Ren, L., Dang, Z.: Preparation, microstructure and properties of
polyethylene/alumina nanocomposites for HVDC insulation. IEEE Trans. Dielectr. Electr.
Insul. 22(6), 3350–3356 (2016)
6. Chen, X., Wu, K., Wang, X., Cheng, Y., Tu, D., Qin, K.: Modified low density polyethylene
by nano-fills as insulating material of DC cable(I). High Volt. Eng. 38(10), 2691–2697
(2012)
7. Khalil, M.S., Jervase, J.A.: Development of polymeric insulating materials for HVDC cables
using additives: evidence from a multitude of experiments using different techniques. In:
Conference Record of the IEEE International Symposium on Electrical Insulation, CA, USA
(2000)
8. Murakami, Y., Imazawa, S., Kurimoto, M., Nagao, M., Inoue, Y., Reddy, C.C., Murata, Y.:
DC breakdown characteristic on LDPE/MgO nanocomposite influenced by DC prestress. In:
2011 Annual Report Conference on Electrical Insulation and Dielectric Phenomena, Cancun,
Mexico (2011)
9. Murakami, Y., Nemoto, M., Okuzumi, S., Masuda, S., Nagao, M.: DC conduction and
electrical breakdown of MgO/LDPE nanocomposite. IEEE Trans. Dielectr. Electr. Insul. 15
(1), 33–39 (2008)
10. Zhao, H., Peng, S.Y., Yang, J.M.: Improving the dispersity and DC breakdown of
MgO/LDPE nanocomposite by adding EVA. Adv. Mater. Res. 833(833), 339–342 (2014)
11. Li, Z., Du, B., Han, C., Xu, H.: Trap modulated charge carrier transport in
polyethylene/graphene nanocomposites. Sci Rep.-UK 7(1), 4015 (2017)
12. Wang, Y., Wang, C., Chen, W., Xiao, K.: Effect of stretching on electrical properties of
LDPE/MgO nanocomposites. IEEE Trans. Dielectr. Electr. Insul. 23(3), 1713–1722 (2016)
13. Xu, M., Zhao, H., Ji, C., Yang, J., Zhang, W.: Preparation of MgO/LDPE nanocomposites
and its space charge property. High Volt. Eng. 38(3), 684–690 (2012)
14. Nilsson, F., Karlsson, M., Pallon, L., Giacinti, M., Olsson, R.T., Venturi, D., Gedde, U.W.,
Hedenqvist, M.S.: Influence of water uptake on the electrical DC-conductivity of insulating
LDPE/MgO nanocomposites. Compos. Sci. Technol. 152, 11–19 (2017)
15. Zhao, N., Li, S., Wang, X., Li, G.: Effects of LDPE/nanofilled LDPE interface on space
charge formation. In: IEEE International Conference on Solid Dielectrics (2013)
16. Min, D., Li, S., Ohki, Y.: Numerical simulation on molecular displacement and DC
breakdown of LDPE. IEEE Trans. Dielectr. Electr. Insul. 23(1), 507–516 (2016)
17. Min, D., Li, S., Ohki, Y.: Analysis on the thickness and temperature dependent DC
breakdown of low density polyethylene. In: 2015 IEEE 11th International Conference on the
Properties and Applications of Dielectric Materials, (2015)
18. Wang, W., Min, D., Li, S.: Understanding the conduction and breakdown properties of
polyethylene nanodielectrics: effect of deep traps. IEEE Trans. Dielectr. Electr. Insul. 23(1),
564–572 (2016)
19. Uvarov, V., Popov, I.: An estimation of correctness of XRD results obtained at analysis of
materials with bimodal crystallite size distribution. CrystEngComm 17(43), 8300–8306
(2015)
20. Turnhout, J.V.: Thermally stimulated discharge of electrets. Polym. J. 2(2), 173–191 (1970)
21. Dissado, L.A., Fothergill, J.C.: Electrical degradation and breakdown in polymers. In: IEE
Materials and Devices Series (1992)
 Modelling and Verification of Ion Currents
Under Various Air Pressure

J. Kortenbrede(&) and F. Jenau

Institute of High Voltage Engineering, TU Dortmund University,

Dortmund, Germany
Jens.Kortenbrede@tu-dortmund.de

Abstract. The main purpose of this paper is to evaluate the results of multiple
simulation methods and verify them with results of experimental measurement.
Therefore, this paper presents three different models, a partial differential
equation system, an analytical calculation and a model based on plasma physics.
With these models, it is possible to calculate ion currents as a function of voltage
and air pressure. The simulated data are compared with each other and evaluated
with the focus of accuracy. Furthermore, the simulation results are verified by
experimental data. The coaxial experimental setup with the possibility to variate
voltage and air pressure is presented. This setup is used to create the mea-
surement data for the model verification.

Keywords: Ion current
Atmospheric modeling
Air pressure
Current

measurement

1 Introduction

The possibility to detect and predict the behaviour of ion current is important for the
coordination of high voltage direct current (HVDC) overhead transmission lines. Close
to overhead lines, there is an electromagnetic field as well as an ionic current. On the
one hand, there are some thresholds for ion current under HVDC overhead transmis-
sion lines and on the other hand is it important to predict the coupling between different
transmission systems. The level of ion current intensity depends on the ion generation
rate and the properties of the transmission path. Both are influenced by atmospheric
variations such as temperature and air pressure. Therefore, it is necessary to investigate
the behaviour of ion currents in dependence of air pressure. Because of this, a coaxial
measurement setup and different calculation models are presented and compared to
each other. Finally, the accuracy of the different models in accordance to the measuring
data is presented and discussed.

2 Experimental Measurement Setup

The used experimental setup is presented in [1]. This setup consist of three main parts: a
high voltage DC source (model: PNChp 100000 − 6 ump by Heinzinger), a coaxial
experimental arrangement and a current measurement system. The coaxial experimental

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1097–1107, 2020.
https://doi.org/10.1007/978-3-030-31676-1_103
1098 J. Kortenbrede and F. Jenau

setup is based on a part of a Gas Insulated Switchgear (GIS). A schematic view is

depicted in Fig. 1. In the centre of the GIS is the strained wire, which is fixed at the two
electrodes on both sides. The geometrical parameter of this setup are the radius of the
wire rwire ¼ 0; 5 mm, the radius of the outer electrode ra ¼ 115 mm and the length
l ¼ 600 mm. With help of a resistive shunt Rsh ¼ 1 kX current is measured. A data
acquisition card detects with 200 Sample/s the voltage over the shunt for a time of 10 s.
The maximum uncertainty of current measurement is about 4 µA caused by the addition
of standard deviation and capacitive current, calculated as in [2].

epoxy epoxy
resin cylindrical conductor resin

electrode for HV rwire

Fig. 1. Schematic view of the coaxial experimental setup [1]

The GIS is drawn to vacuum and filled with dried and compressed air up to the
desired pressure. Therefore, the relative humidity is set to zero in the calculation
models.

3 Ion Current Modelling

In this chapter, the three different methods are presented to calculate ion current with
variation of voltage and air pressure. The first model combines various differential
equations, the second is based on an analytical approach and the third one works with
plasma physics.

3.1 Space Charge Model

The space charge model taking into account the self-consistent transport of ions
without considering any possible collisions. The ionisation zone is neglected and the
wire surface is the source of negative ions. The radius of the ionisation zone is similar
to the radius of the wire and is much smaller than the distance between the two
 Modelling and Verification of Ion Currents Under Various Air Pressure 1099

electrodes. Generally the corona current is defined with the charge carrier density like
in the following formula:
Z
e   !
I¼ np vp  nn vn  ne ve
E0 dV ð1Þ
U
V

where:
e = elementary charge
U = voltage
np ; nn and ne = density of positive, negative ions and electrons
vp ; vn and ve = velocity of positive, negative ions and electrons
~
E 0 = Laplace field
In case of a constant negative corona current the electrons are getting accelerated to
the ionisation zone and the positive ions are moving towards the cathode where they
get absorbed. In conclusion of attachment the number of negative ions increases and
the number of electrons is decreasing. On basis of this, corona current results from the
charge density of negative ions [3]. Therefore, ions of the same polarity as the wire are
drifting towards the ground electrode where they get neutralized [4]. With this pre-
liminary considerations the current caused by corona is called ion current. With this
assumption and the assumption of an infinite long wire, the process of corona can be
described by the Poisson Equation (2) and the continuity Equation (3). These equations
lead to the combined differential equation system in a two-dimensional space. In
comparison to the plasma based model (Sect. 3.3) this model is neglecting all dis-
persions with exception of the ion-ion recombination.
q
D/ ¼  ð2Þ
e0 er

where:
/ = electrical potential
q = space charge density
e0 er = permittivity

@q
þ r
~
J¼0 ð3Þ
@t
where:
~
J = current density
t = time
1100 J. Kortenbrede and F. Jenau

The current density ~

J is subdivided in three parts:

~
J ¼ leff q~
E þ qvair  Drq ð4Þ

where:
leff = effective ion mobility
~
E = electric field
vair = wind velocity
D = diffusivity
The first part is the drift term and describes the charge transport as a result of the
electric field. The second part specifies the convection and the last term describes
possible processes of diffusion following the gradient of charge density [5]. Equa-
tion (3) as well as Eq. (4) could be derived of the Boltzmann equation with the method
of moments. When Eq. (4) is used in Eq. (3), the following equation is obtained:

@q  
þ r lq~
E þ rðqvair Þ þ rðDrqÞ ¼ 0 ð5Þ
@t

If the space charge density is dependent on the position and the velocity of wind is
independent on the position Eq. (5) becomes to Eq. (6) caused of the product rule for
vector analysis.

@q  
þ l qr
~
E þ~
E
rq þ vair rq þ DDq ¼ 0 ð6Þ
@t

With the help of the Gaussian law

q
r~
E¼ ; ð7Þ
e0 er

and the potential function for the electrostatic field under the assumption of time slow
variable magnetic fields

~
E ¼ r/; ð8Þ

results with er  1
 2 
@q q
þl  r/
rq þ~
vair rq þ DDq ¼ 0 ð9Þ
@t e0

Equation (9) describes the temporal development of space charge and forms the
partial differential equation system with Eq. (2), which has to be solved with suitable
boundary conditions in order to calculate the unipolar ion current with Eq. (4) for any
electrode configuration [6].
 Modelling and Verification of Ion Currents Under Various Air Pressure 1101

For the simulation within this paper the drift term is used. The time dependent term
is neglected because the stationary state is in focus. The term for convection is not used
in fact of the closed coaxial arrangement. Furthermore possible diffusion processes are
not performed. Therefore Eq. (9) is simplified to

q2
0 ¼ r/
rq  : ð10Þ
e0

Boundary conditions: To define boundary conditions the space charge density is

separated into two parts, a constant part q0 and a space dependent part dq:

q ¼ q0 þ dq ð11Þ

As a result of this distribution the following two equations are solved:

q0 þ dq
D/ ¼  ð12Þ
e0

ðq0 þ dqÞ2
0 ¼ r/
rdq  ð13Þ
e0

One boundary condition is that the emitted ions by the electrode surface are con-
stant and space independent, which refers to dq ¼ 0 at r ¼ ri .
Two further boundary conditions are:

/jAnode ¼ 0 ð14Þ

~
n
r/ ¼ EC ð15Þ

where:
~
n = unit vector in normal direction
Ec = corona onset gradient

Input parameter: The corona onset gradient EC is calculated with the empirical
formula [7, 8]:
 
K
Ec ¼ mE0 d 1 þ pffiffiffiffiffiffi ð16Þ
dri

where:
m = surface irregularity factor
ri = conductor radius
E0 and K = empirical constants
d = relative air density factor
1102 J. Kortenbrede and F. Jenau

with

273 þ T0 p
d¼
ð17Þ
273 þ T p0

where:
T = temperature of ambient air
p = pressure of ambient air
T0 ¼ 20  C
p0 ¼ 1013 hPa
In this paper the values of Whitehead for negative direct current are used for the
empirical constants ðE0 ¼ 31 kV=cm and K ¼ 0:308Þ [7, 8]. For the effective ion
mobility two different models are used, the model of Zhang [9] and Liu [10]. The
resulting onset gradients and effective mobilities are summarised in Table 1.

Table 1 Used corona onset gradient and effective mobilities

p in kPa Ec;Whitehead in kV l ðcmÞ2 ðcmÞ2
cm eff ;Zhang in Vs leff ;Liu in Vs
100 73.0 2.493 3.675
120 83.2 2.078 3.552

3.2 Analytical Ion Current Model

For simple electrode configurations such as coaxial arrangement, it is possible to
calculate the ion current for different voltages by analytical expressions. With con-
siderations of simplifications in Sect. 3.1 the Eqs. (2) and (9) can be solved analyticaly
[11]. For the steady-state and with neglecting the convection and diffusion part the
electric field in radial-direction is described by:
  1
E c ri ri2 ri2 2
E¼ 1 2f þ 2f ð18Þ
r ra ra

where f is the normalized current and is defined with

 
I ra 2
f¼ ð19Þ
2pleff e0 Ec ri

where:
Ec = corona onset gradient
ri =ra = inner /outer radius
I = current
leff = effective ion mobility
 Modelling and Verification of Ion Currents Under Various Air Pressure 1103

If the voltage U\Uc the normalized current f ¼ 0, so that the electric field is
equivalent to the Laplace field. With the assumption that

ri2
f1 ð20Þ
ra2

and the integration of Eq. (18) from ri to ra the relation between used voltage and
normalized current can be written as following
  pffiffiffiffiffiffiffiffiffiffi  
U  Uc ra 2
ln ¼ 1 þ f  1 þ ln pffiffiffiffiffiffiffiffiffiffi ð21Þ
Uc ri 1þ 1þf

To calculate an ion current with variation of voltage, geometry (inner and outer
radius), corona onset field strength and effective ion mobility are necessary to know.
Therefore the parameters in Table 1 are used. The resulting ion current for the used
voltage is calculated with Eq. (19).

3.3 Plasma Model

The plasma model works with generation and transport of charged particles. Therefore,
it is necessary to calculate the transport coefficients for the simulated gas (here air). The
coefficients for transport are calculated using a Boltzmann solver, which needs the
collision process of each used species and the atmospheric parameter like temperature.
In this paper are the species oxygen and nitrogen used which are taken from the Plasma
Exchange Project LXcat. The implemented reactions are listed in Table 2. Further-
more, it is implemented that the ions are neutralised when they reach the surface of an
electrode. Moreover, secondary emission is implemented for the cathode. Photoion-
ization is neglected. Further details are presented in [2].

Table 2 Implemented chemical reactions

Reaction Type
e þ A ! p þ 2e Ionization
eþA ! n Attachment
e þ 2A ! n þ A Attachment
eþp ! A Reaction
n þ p ! 2A Reaction

4 Analysis and Discussion

This chapter presents the results of the three different calculation models and com-
parison with measurement values. For the analytical model and the space charge model
are two different effective ion mobilities used in each case. Therefore are two curves for
1104 J. Kortenbrede and F. Jenau

the analytical and space charge model presented in each plot. Furthermore, the results
of the plasma simulation and the measuring points are presented.
The general behaviour of ion current over voltage is simulated with all models like
shown in Fig. 2. Furthermore, it is depicted that the gradient of current over voltage
depends on the effective ion mobility for the space charge model and the analytical
model. This effect is illustrated by the blue arrow in Fig. 2 and is justified in Eqs. (4)
and (19). In these equations the effective mobility is a factor which influences the ion
current linear. If the same effective mobility is used for both models, the resulting ion
current is different. The space charge model delivers smaller ion currents in comparison
with the analytical model.

100 kPa

Fig. 2. Ion current over voltage for different models with variation of effective ion mobility and
comparison with measurement data by 100 kPa air pressure

By comparison of Fig. 2 for 100 kPa and Fig. 3 for 120 kPa air pressure ion
current decreases with increasing air pressure. The reason for this is on the one hand the
lower effective mobility and on the other hand the higher onset gradient like shown in
Table 1. The measurement values are in the range of the simulated values for both air
pressures. For a better comparison of simulated data and measured data, relative
deviation in relation to the measured values for each voltage is calculated and depicted
in Figs. 4 and 5.
The deviation for each model over voltage for 100 kPa (Fig. 4) and 120 kPa
(Fig. 5) air pressure are shown. Using the analytical model deviation at 100 kPa air
pressure is in the range of 38%–76% with effective mobility of Liu in contrast to 6%–
20% deviation for mobility of Zhang. The deviation for lLiu and 120 kPa pressure is in
the range of 79%–118% and between 5% and 27% for lZhang : Therefore, the accuracy
 Modelling and Verification of Ion Currents Under Various Air Pressure 1105

120 kPa

Fig. 3. Ion current over voltage for different models with variation of effective ion mobility and
comparison with measurement data by 120 kPa air pressure

100 kPa a b

Fig. 4. Relative deviation of different models over voltage at 100 kPa air pressure
1106 J. Kortenbrede and F. Jenau

120 kPa

Fig. 5. Relative deviation of different models over voltage at 120 kPa air pressure

of the analytical model is higher with use of mobility calculated by the model of Zhang
at both pressures.
Using space charge model deviation is in the range of 32%–88% for lLiu and in the
range of 10%–23% for lZhang at 120 kPa air pressure, so the accuracy is higher by
using lZhang . In case of 100 kPa deviation is dependent on voltage level by comparing
mobility of Zhang and Liu. The two voltage sections are highlighted in Fig. 4 with the
letters a and b. In section a (26 kV–39 kV) deviation for lZhang is smaller (10%–18%)
than for lLiu (20%–63%). For voltage section b (40 kV up to 50 kV), deviation (3%–
18%) using lLiu is more exact than using lZhang (20%–30%).
When using plasma model for 120 kPa deviation decreases from 158% at 26 kV to
23% at 50 kV and for 100 kPa air pressure from 98% at 26 kV to 8% at 50 kV.
Therefore, accuracy increases with increasing voltage for plasma model.

5 Conclusions

This paper presents three kinds of ion current calculation in comparison with mea-
surement data. The general behaviour of ion current over voltage is rebuild with all
three presented models. Furthermore the deviation of the analytical and space charge
model divers with the chosen effective mobility. These two models deliver the best fit
for different mobilities and voltages. Therefore, it is necessary to select different
effective mobility’s for several models. The plasma model delivers good results for
higher voltages. Further research aspire a better performance near onset gradient for the
plasma model. Moreover, a better prediction of effective mobility for the different
models is an important aim.
 Modelling and Verification of Ion Currents Under Various Air Pressure 1107

Acknowledgments.
Many thanks to student Philipp Huber for his support by the
implementation of the models.
The authors would like to thank the federal ministry for economic
affairs and energy of Germany for the support of this work as a part
of the E2HGÜ project (FKZ 03ET7514).

References
1. Kortenbrede, J., Cimino, A., Jenau, F.: Ion current studies under various air pressure using a
coaxial experimental setup. In: 2017 INSUCON - 13th International Electrical Insulation
Conference (INSUCON), Birmingham, United Kingdom (2017)
2. Kortenbrede,J., Cimino, A., Jenau, F.: Evaluation of ion current modelling under variation of
air pressure based on measurement data. In: 2018 IEEE 2nd International Conference on
Dielectrics (ICD), Budapest, 1–5 July 2018 (2018)
3. Sato, N.: Discharge current induced by the motion of charged particles. J. Phys. D Appl.
Phys. 13(1), L3–L6 (1980)
4. Adamiak, K., Atten, P.: Simulation of corona discharge in point–plane configuration.
J. Electrostat. 61(2), 85–98 (2004)
5. Abdel-Salam, M.M., Al-Hamouz, Z.M.: A new finite-element analysis of an ionized field in
coaxial cylindrical geometry. J. Phys. D Appl. Phys. 25(10), 1551–1555 (1992)
6. Potrymai, E.: Time Dependent Modelling and Simulation of the Corona Discharge in
Electrostatic Precipitators. Linnaeus University (2014)
7. Maruvada, P.S., Bisnath, S.: Corona in transmission systems: Theory, design and
performance. Crown Publications, Johannesburg (2011)
8. Whitehead, J.B.: The high voltage corona in air (1911)
9. Zhang, B., He, J., Ji, Y.: Dependence of the average mobility of ions in air with pressure and
humidity. IEEE Trans. Dielect. Electr. Insul. 24(2), 923–929 (2017)
10. Liu, Y., Huang, S., Zhu, L.: Influence of humidity and air pressure on the ion mobility based
on drift tube method. CSEE Power Energy Syst. 1(3), 37–41 (2015)
11. Townsend, J.S.: XI. The potentials required to maintain currents between coaxial cylinders.
Lond. Edinb. Dublin Philos. Mag. J. Sci. 28(163), 83–90 (1914)
 Numerical Modelling and Influence of Defects
on Space Charges in Epoxy Resin
Under HVDC Stresses

G. Ortiz1(&), P. Mbolo Noah2, S. Agnel2, P. Notingher2, C-T. Vu1,

and A. Girodet1
1
SuperGrid Institute, 23 Rue Cyprian, 69628 Villeurbanne, France
guillermo.ortizhernandez@supergrid-institute.com
2
Institut D’Electronique et Des Systèmes, Université de Montpellier,
860 Rue de St Priest, 34090 Montpellier, France

Abstract. Space charge accumulation on High Voltage Direct Current Gas

Insulated Substations can produce electrical field reinforcements in the insula-
tion that need to be taken into account in the equipment design. The TSM
(Thermal Step Method) is one of the experimental techniques allowing to
determine space charge distributions in insulating materials. However localized
defects (i.e. microvoids, delaminations etc.) cannot usually be detected by this
technique. A new numerical approach to study the influence of structural defects
on Thermal Step Method currents is proposed. The method is based on a Finite
Element numerical simulation allowing to simultaneously solve electrical and
thermal equations. The effect of three different defects were studied. It results
that ring defects, with diameters smaller than 0.4 mm, produce less than 10% of
change on TSM current signals. This confirms the difficulty to detect small
defects by this method. It was also observed that delaminations can produce
variations in signal as high as 70%, and even generate signals of opposing sign
from the case without defect.

Keywords: Space charge
Epoxy resin
HVDC
Thermal step method

1 Introduction

Nowadays, thanks to the development of HVDC link and meshed HVDC network, Gas
Insulated Substations (GIS) and Gas Insulated Lines (GIL) will be one of key com-
ponents in the development of these networks. They consist on a high-voltage con-
ductor maintained by insulating spacers, the whole being placed in a metal enclosure
filled with a dielectric gas, generally SF6. An important problem to deal with when an
insulating material is subjected to a high DC field and a high temperature is the
injection and accumulation of space charges. These charges can produce a local
increase of the electric field [1, 2], which can be critical for GIS/GIL in case of
superimposed voltage or polarity reversal. It is therefore important to take into account
space charges in the equipment design.
The TSM (Thermal Step Method) is one of the experimental techniques allowing to
determine space charge distributions in insulating materials. Developed mainly for
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1108–1120, 2020.
https://doi.org/10.1007/978-3-030-31676-1_104
 Numerical Modelling and Influence of Defects on Space Charges in Epoxy Resin 1109

HVDC cables [3], this technique could be also applied for GIS and GIL spacers in
order to assess the modification of the electrical field created by space charges accu-
mulated at the interface of the epoxy resin and the metallic insert or in the bulk
material. Previous studies have shown that TSM can be applied to cylindrical epoxy
resin samples to study the evolution on space charges over time [4].
However, the determination of space charge distributions from data provided by
stimuli charge measurement methods, like the TSM, requires to suppose that the
electrical state of the material is homogeneous in a plane parallel to electrodes (for flat
samples) or in a cylindrical skin of given radius (in an axisymmetric configuration).
Thus, it is usually difficult to take into account the effect of localized defects (i.e.
microvoids, delaminations etc.) on acquired signals, both because they do not fulfil the
former hypothesis and because their contribution to the overall signal may be too small.
In the present work, we present a numerical model developed to evaluate the
influence of these localized defects on space charge distributions and to determine the
effect of their presence on TS current signals.

2 Theoretical Basis

2.1 Thermal Step Method

When a DC voltage is applied on a dielectric material, space charges can be generated.
The mechanism describing space charge evolution is very complex and can involve
accumulation around material’s electrical inhomogeneities, charge trapping, charge
generation by partial discharges or charge injection/extraction at electrodes interfaces.
TSM is a non-destructive technique developed to study the space charge distri-
butions on dielectric material. It consists on applying an abrupt thermal step on one of
the sample’s surface. This generates a thermal wave that diffuses throughout the sample
and interacts with the cumulated space charges. If the sample is short circuited, the
temperature variation produces a measurable displacement current named TS current
(ITS).
Equation 1 gives the analytical expression describing the time evolution of the
current ITS in cylindrical coordinates [5]. In the expression of the TS (Thermal Step)
equation, C stands for the electrical capacitance of the sample, a for a constant directly
depending on the variation of the capacitance with the temperature, E for the electric
field, T for the temperature, r for the radial distance coordinate and t for the time. As
previously stated, this equation supposes that sample is homogeneous in the cylindrical
plane at a given r coordinate.

rZ2 @T ðr; tÞ
ITS ðtÞ ¼ aC E ðr Þ dr ð1Þ
r1 @t
1110 G. Ortiz et al.

2.2 Numerical Modelling

The implementation of the proposed TSM calculation method, based on numerical
modelling, consists on simultaneously solving, electrical and thermal equations. This
was done by Finite Element calculations using COMSOL Multiphysics software.
Thermal expression is given by Eq. 2, where D is the thermal diffusivity, S the
source term, q the density and Cp the heat capacity.

@T S
 Dr2 T ¼ ð2Þ
@t qCp

Since DC constraints impose to take into account the material’s electrical con-
ductivity, electrical modelling consists on solving Eqs. 3 to 5, where J is the current
density, r is the conductivity, e the permittivity and q the electrical charge.

~ @e~
E
J ¼ r~
Eþ ð3Þ
@t

re~
E¼q ð4Þ

@q
r~
J¼ ð5Þ
@t

3 Numerical Procedure for TS Signal Simulation

3.1 Geometry and Studied Cases
For this study, 40 mm long and 3 mm thick cylindrical alumina filled epoxy resin
samples (as shown on Fig. 1) were modelled in 2D axisymmetric geometry using Finite
Element software COMSOL Multiphysics. The geometry was defined based on sam-
ples of a previous work [4]. Even though 2D axisymmetric geometry was chosen to
reduce as much as possible calculation time, the proposed method is completely
compatible to a 3D modelling.
During manufacturing process of a spacer, delamination may occur at the
conductor/resin interfaces and gas bubbles can form in the bulk. For this study, defects
were modelled as geometrical free space filled with air. Since the model was developed
in 2D axisymmetric geometry, all defects geometries were of toroidal shape following
the revolution axis (ring defects). Three different defects were studied (see Fig. 2):
• Air ring-type defect in the epoxy resin volume (Fig. 2a);
• Air delamination-type defect at the grounded electrode/epoxy resin interface
(Fig. 2b);
• Air delamination-type defect at the inner high voltage electrode/epoxy resin inter-
face (Fig. 2c).
Both defects at the electrodes interfaces are denoted as delaminations.
 Numerical Modelling and Influence of Defects on Space Charges in Epoxy Resin 1111

Fig. 1. Geometry used for space charge simulations representing a 40 mm long and 3 mm thick
hollow cylinder.

Fig. 2. 3D model representation (in red) of three studied defects: (a) air ring defect, (b) outer
(ground) delamination and (c) inner (HV) delamination.

3.2 Simulation Parameters

Epoxy resin electrical properties were implemented in the software through the use of
electrical conductivity and permittivity models. The modelling of conductivity as a
function of electric field and temperature model r(E,T) was based on experimental
measurements [6]. Figure 3 shows I(V) curves obtained by applying the conductivity
model, corresponding to conduction current as a function of applied voltage.
Relative permittivity dependence on temperature was modelled as a linear varia-
tion. For the studied material, the used value was 5.6 at room temperature, and the
variation in temperature of er was determined by using TSM calibration as in Sect. (4.1)
presented hereafter.
For all simulations, the air defect’s electrical properties were a constant conduc-
tivity of 10−12 S/m and a constant relative permittivity of 1.
Table 1 presents the values of materials thermal properties used in the simulations.
Epoxy resin properties were determined by experimental measurement presented in a
previous study [7]. Air properties come from COMSOL standard database.

3.3 Field, Space Charges and TS Current Simulation Procedure

Usually, in order to reveal the influence of the applied constraints on the accumulation
of space charge in the insulation, an electro-thermal stress is applied on the sample.
1112 G. Ortiz et al.

Fig. 3. I(V) curves for different temperatures in logarithmic scale calculated by using the r(E,
T) model in a reference geometry.

Table 1. Materials thermal properties, k stands for the thermal conductivity, Cp for the heat
capacity, q for the density and D for the thermal diffusivity.
k (W/m/K) Cp (J/kg/K) q (kg/m3) D (m2/s)
Epoxy resin 0.8 2400 2200 0.15  10−6
Air 0.026 1010 1.3 20  10−6

After that, in voltage off conditions, an abrupt thermal variation is applied to one of
sample’s surface to measure the TS current generated by the stimulated space charges.
For electro-thermal stress, a high DC voltage is applied on internal surface of the
hollow cylinder and the outer surface is grounded. During this step, sample’s tem-
perature is fixed to a stress temperature Tstress.
The same steps were followed for the numerical model. By simultaneously solving
electrical and thermal equations presented on Sect. 2.2, time evolution of electric field
throughout the sample can be calculated. The simulation duration on this phase was
chosen so that electric field is stabilized.
The change on electric field distribution E(t) compared to Laplace field ELaplace
(field at t = 0) is defined as the residual field Eres. Thus Eres is calculated with Eq. 6.

Eres ¼ EðtÞ  ELaplace ð6Þ

Finally, by applying Gauss law (Eq. 4) to residual field, an equivalent space charge
distribution is calculated.
The generation of the thermal step consists of applying a thermal variation on the
internal surface (inner electrode), while the temperature of the outer surface was set to
room temperature.
The final step of the method consists on the calculation of the resulting TS current.
This is done by integrating the current density through the external surface of the
sample (Eq. 7).
 Numerical Modelling and Influence of Defects on Space Charges in Epoxy Resin 1113

ZZ
J
d~
~ S ð7Þ
Outer
electrode

As mentioned above, the TS current depends on the variation in temperature of

sample capacitance (see Eq. 1). In a real TSM measurement, mechanical and electrical
variations may contribute to capacitance variation. However, in epoxy resin, the typical
coefficient of thermal expansion is small (10−6 to 10−5 K−1) compared to electrical
variations (10−4 to 10−3 K−1). Therefore, in this study, only the electrical contribution
to the a constant was taken into account. If this method needed to be applied to other
materials where both contributions were comparable, it would be necessary to include
mechanical modules in combination to thermal and electrical ones.

4 Results

In this section, results of numerical Thermal Step currents calculation method are
presented. First, a calibration step, based on comparison with experimental measure-
ments, was used to determine the experimental thermal step applied to epoxy resin
sample and its a constant. In a second part, the results of the calculation method applied
to a sample without defect is analyzed. Finally, the effect of different air defects (rings
and delamination) are compared to the reference case without defects.

4.1 Calibration Step

In practice, typical calibration procedures consist on applying a relatively low DC
electric field to the sample [5]. The applied electric field should be low enough so that
space charge level stays as low as possible and TS signal remains measurable.
Knowing the field distribution throughout the sample (Laplace field for cylindrical
sample), it is possible, by simulating the TS signal, to determine the experimental
applied thermal stimulus on the inner electrode and the a constant of the sample.
Figure 4(a) shows measured experimental TS signals obtained by applying a 3 kV
voltage (1 kV/mm in the insulation close to the inner electrode) and a thermal step of
−30 °C, with a liquid at −5 °C on the inner electrode.
Figure 4(b) shows the thermal step obtained by fitting simulated curves to exper-
imental ones on Fig. 4(a). The fitting curves were simulated following the numerical
procedure presented in Sect. 3. In addition to the thermal stimulus, the permittivity
variation with temperature was determined (de/dT = 7  10−4). The obtained tem-
perature profile and the a coefficient were used for all the studied cases presented in the
following sections.
To validate the proposed numerical procedure allowing to simulate TS current
signals from a given electric field distribution of 1 kV/mm on the inner electrode, two
different calculation methods were compared (Fig. 5).
1114 G. Ortiz et al.

Fig. 4. (a) TS calibration signal obtained by imposing a ±3 kV voltage (|Estress| = 1 kV/mm) on

sample’s internal surface. (b) Thermal stimulus determined from calibration.

Fig. 5. Comparison of analytical determination of TS current using Eq. 1 and numerical

resolution by integrating the current density over the outer electrode (Eq. 7).

The current density integration method (Eq. 7) and the application of the analytical
TS equation (Eq. 1) method give the same result. The use of the analytical equation is
only possible if the sample doesn’t contain any significant structural defects, such as for
example delamination. That is the reason why the current density integration method
was chosen for this study of the influence of defects on TS current signals.
 Numerical Modelling and Influence of Defects on Space Charges in Epoxy Resin 1115

4.2 Space Charge and TS Current Without Defects

To simplify notations, the stress field Estress is defined as the field at mid-gap on a
coaxial geometry, such as the one studied in this paper. The stress field is calculated
from the applied high voltage value (HV) using Eq. 8, where Rin, Rout and Rm-g are,
respectively, the inner, outer and mid gap radii.

HV 1
Estress ¼   ð8Þ
Rout Rmg
ln Rin

Figure 6 shows the evolution of the electric field, the residual field (see Eq. 6) and
the equivalent space charges with time along an r coordinate cut line (z coordi-
nate = 0). All curves were obtained following electro-thermal stress procedure
described in Sect. 3.3. A stress field Estress = 10 kV/mm and a stress temperature
Tstress = 100 °C was applied to the numerical sample.

Fig. 6. Cutline along r coordinate (a), and electric field (b), residual field (c) and space charges
(d), along the cutline, obtained for different times during the thermoelectric stress period (with
Estress = 10 kV/mm and Tstress = 100 °C).

The use of the conductivity model as a function of electric field r(E), on a sample
where the electric field distribution is non-homogenous (due to cylindrical geometry),
results on a conductivity gradient through the sample. This gradient on electrical
properties produces variations of equivalent space charges over time. Of course, space
1116 G. Ortiz et al.

charges distributions determined following this procedure are not representative of

those that can be obtained in real samples since trapping, partial discharges or
injection/extraction mechanisms are not taken into account. Still, these equivalent space
charges allow to study the effect of conductivity inhomogeneities.
The case without defect was used as a reference point for comparison to different
defects cases studied hereafter. Figure 7 shows the calculated TS signals from residual
fields and space charges presented on Fig. 6. It is clear that the signal is increased by
increasing the temperature or the stress field (both cases being equivalent to an increase
of conductivity as shown on Fig. 3). In the range of electric field and temperatures
studied, the temperature has a heavier impact on TS signal and thus on space charges
generation.

Fig. 7. TS currents obtained on a geometry without defects for different temperatures with a
stress field of 10 kV/mm (a) and for different stress fields at 100 °C (b).

In next sections, the effect of different defects are compared to the reference case
without defects. For all studied cases, the stress temperature was fixed at 100 °C, and
the stress field at 10 kV/mm.

4.3 Space Charge and TS Current with Ring-Type Defect

First studied defect was an air ring of 0.1 to 1 mm diameter located at mid-gap. Figure 8
(a) shows a 2D-axisymmetric representation of the calculated equivalent space charge,
after the stress phase (10 kV/mm at 100 °C during 24 h). Figure 8(b) shows the influence
of the defect diameter on the equivalent space charges. One can observe that the smaller
the defect diameter, the higher the maximum value of the obtained space charge.
Resulting TS signal for defects at mid-gap of different diameters are shown on
Figure 9. The signal change is very small, less than 10% for ring defect of diameter
lower than 0.4 mm. This result shows that, even if the peak value of the space charge is
higher for small defects, the influence on TS current signals seems to depend more on
the total volume occupied by the defect. This confirms the difficulty to take into
account the influence of localized space charge accumulation, generated by this kind of
defect, with this measurement technique.
 Numerical Modelling and Influence of Defects on Space Charges in Epoxy Resin 1117

Fig. 8. (a) Equivalent space charge of a 0.5 mm diameter air ring-type defect. (b) Equivalent
space charge profile along the shown r coordinate cutline for different defects diameters.

Fig. 9. TS currents from numerical sample with ring defect of different diameters.

One can also observe on Fig. 9 than when the defect’s diameter increases, the
signal decreases. For a diameter of 1 mm, the decrease of signal is of about 70% with
respect to reference signal without defect. Moreover, the signal shape changes since a
plateau is clearly observed.
Figure 10 shows TS signals obtained when the position of the defect varies along
the r-coordinate. Figure 10(a) gives the results for a defect of 0.1 mm in diameter.
Almost no change on the TS signals is observed for this defect, even when it is close to
inner electrode where the resolution of the method is the highest.
1118 G. Ortiz et al.

Fig. 10. TS currents obtained with a numerical sample including a ring defect of 0.1 mm (a) and
0.5 mm (b) in diameter at different positions along r coordinate (z = 0).

Fig. 11. TS current signals obtained in a geometry including a delamination-type defect on inner
HV electrode/epoxy resin interface (a) and outer grounded electrode/epoxy resin interface (b).
The defects are 0.1 mm thick (along r coordinate) and have different lengths Ldef (along
z coordinate).

In comparison, Fig. 10(b) gives the results for a defect of 0.5 mm in diameter.
When the defect is closest to inner electrode, the signal is increased by roughly 30%,
while close to the grounded electrode, the signal decreases by about 20%.
 Numerical Modelling and Influence of Defects on Space Charges in Epoxy Resin 1119

4.4 TS Current with Delamination-Type Defect

Figure 11 shows the TS signals for a 0.1 mm thick delamination-type defect on inner
(HV) electrode and on outer (grounded) electrode of different lengths Ldef. When
delamination is present on HV electrode (Fig. 11a) the TS signal increases, with
respect to the case without defect, when the length increases (up to a value close to
1 mm), then it starts to decrease when the length increases. In parallel, a progressive
sign inversion of the signal can be observed.
When the delamination is present on the grounded electrode (Fig. 11b), the TS
signal decreases as the length increases. Moreover, when the delamination length
exceeds 1 mm in z-axis, the signal sign changes and its magnitude keeps increasing
with length. For total delamination, the signal is of opposing sign with respect to the
case without defect, and its magnitude is roughly 6 times higher than the reference case.

5 Conclusions

In this paper, a new numerical approach to simulate Thermal Step Method currents
from given space charge distribution is proposed. The method is based on a Finite
Element numerical simulation using COMSOL software. It consists of simultaneously
solving electrical and thermal equations.
A calibration step, using experimental measurements, allowed to determine the
experimental thermal step variation applied to the samples and the variation of the
capacitance with the temperature of the epoxy resin. This parameters were used for all
the subsequent studied cases.
The case without defects was the first tested case for the proposed method. The use
of a field dependent conductivity model and the non-homogenous field of the cylin-
drical geometry produce a conductivity gradient, and thus, allows to determine an
equivalent space charge distribution. The TS current simulated from this space charge
distribution was set as reference to study the effect of the presence of localized defects.
Three different defects were studied: air ring-type defects in the sample bulk, air
delamination-type defects on the outer (grounded) electrode and on the inner (High
Voltage) electrode. The defects were modelled from constant electrical and thermal
properties. It came out that ring-type defects with diameter smaller than 0.4 mm pro-
duce less than 10% of change in TS signals. This confirms the difficulty to detect small
defects by this method. It was also observed that delamination-type defects, and more
precisely their dimension and localization in the insulation bulk, can produce variations
in signal as high as 70%, and even generate signals of opposing sign with respect to the
case without defect. It will be interesting to validate this results by comparing with
experimental measurements on samples with controlled defects.
Even if the proposed model was developed for a 2D axisymmetric geometry, it can
be easily extend to 3D geometries allowing the study of other defect geometries as
sphere voids. Moreover, the method can be used to study other dielectric materials by
implementing their electrical and thermal properties.
1120 G. Ortiz et al.

Acknowledgements. This work was supported by a grant of the French National Research
Agency (ANR) as part of the “Investissements d’Avenir” Program (ANE-ITE- 002-01).

References
1. Le Gressus, C., Blaise, G.: Breakdown phenomena related to trapping/detrapping processes in
wide band gap insulators. IEEE Trans. Electr. Insul. 27, 472–481 (1992)
2. Dissado, L.A., et al.: The role of trapped space charge in the electrical aging of insulating
materials. IEEE Trans. Dielectr. Electr. Insul. 4, 496–506 (1997)
3. Agnel, S., et al.: Study of space charge dynamics directly on power cables using the thermal
step method. In: IEEE CEIDP Annual Report (1999)
4. Noah, P.S.M., et al.: Development of a space charge measurement method applied to
HVDC GIS spacers. In: 2016 IEEE Conference on Electrical Insulation and Dielectric
Phenomena (CEIDP) (2016)
5. Notingher, P., et al.: Thermal step method for space charge measurements under applied dc
field. IEEE Trans. Dielectr. Electr. Insul. 8, 985–994 (2001)
6. Vu, C.-T., et al.: Simulation methodology for DC GIS/GIL. In: Conference MATPOST
(2015)
7. Desmars, L., et al.: High voltage electrical properties of epoxy / h-BN microcomposites. In:
2018 IEEE 2nd International Conference on Dielectrics (ICD) (2018)
 Investigations on the Polarity Dependent
Charge Carrier Injection in Dielectric
Liquids at High DC Voltage Stress

Tobias Gabler1(B) , Karsten Backhaus1 , Ronny Fritsche2 ,

and Steffen Großmann1
1
Institute of Electrical Power Systems and High Voltage Engineering,
Technische Universität Dresden, Dresden, Germany
tobias.gabler@tu-dresden.de
2
Energy Management Division Transformers, SIEMENS AG, Nuremberg, Germany

Abstract. A reliable operation of HVDC converter transformers requires

an optimized dielectric design of the oil-paper insulation system. There-
fore, the dielectric behaviour of the oil-paper-insulation at high DC volt-
age stress needs to be investigated. In previous investigations, a charge
carrier-based approach is utilised to explain occurring effects like the non-
linear, field strength dependent electric conductivity of mineral oil and to
model the dielectric behaviour of mineral oil at high DC voltage stress.
This model considers different species of charge carriers, their drift and
their effect on the electric field distribution. These physical processes
can be modelled using the Poisson-Nernst-Planck equation system. Both
species of oil-intrinsic charge carriers and charge carriers injected from
electrodes play a significant role in the nonlinear conductivity of mineral
oil and affect the dielectric stress of oil-paper-insulated arrangements. Fur-
thermore, a polarity effect in the electric conductivity of mineral oil was
observed, which is a result of the different injection rates of electrons and
holes from bare metal electrodes. The aim of this study is to examine the
nonlinear conductivity of mineral insulating oil and to apply the results
on the charge carrier-based approach in order to investigate the injection
of charge carriers and their influence on the dielectric stress. Current mea-
surements are carried out at an inhomogeneous electrode arrangement
at DC voltage stress at both polarites in order to investigate the electric
conductivity of an aged and unaged naphthenic and a paraffinic mineral
oil. Furthermore, the pure paraffin n-Dodecane was used to investigate
the influence of the different components of mineral oil on its dielectric
behaviour. The results show differences in the polarity dependent injec-
tion rates of the investigated liquids and between the aged and the unaged
mineral oil. The different components of the used oils as well as ageing
products affect the injection of charge carriers and thus the measured con-
ductivity. Therefore, it is assumed that the different ingredients of mineral
oils affect the generation of charge carriers and thus the dielectric stress
of an oil-paper-insulated arrangement at DC voltage stress.

Keywords: Mineral insulating oil · Nonlinear electric conductivity ·

Charge carrier injection · HVDC
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1121–1131, 2020.
https://doi.org/10.1007/978-3-030-31676-1_105
1122 T. Gabler et al.

1 Introduction
The increasing share of decentralized generated electric power and the required
transmission of electrical energy over long distances increase the demand on
HVDC transmission systems [1,2]. In order to ensure a reliable operation the
dielectric design of the equipment needs to be optimized. For this reason, the
dielectric behaviour of the insulation materials at high DC voltage stresses needs
to be investigated.
The dielectric properties of an insulation material at DC voltage stress are
commonly described by assuming an ohmic electric conductivity κ, which can
be used in field calculations — e. g. with FEM-calculations. In case of the oil-
paper insulation system of HVDC converter transformers previous investigations
showed a nonlinear electric conductivity of mineral oil dependent on the electrical
field strength [3–5]. The nonlinear conductivity is a result of the mutual influence
of charge carrier accumulation, injection and their influence on the electric field
distribution at the electrode–fluid interface, which can be described according
to a charge carrier-based approach [3,4,6].
Determining the field dependent conductivity of an insulating liquid at DC
voltage stress holds many challenges. According to [7] it is recommended to
measure the electric conductivity of dielectric liquids by using a homogeneous
or a coaxial, weak inhomogeneous electrode arrangement. However, experiments
showed that the effect of each charge carrier species on the field dependent
injection processes leads to a not reproducible measurement of the electric con-
ductivity at such arrangements [6,8]. Furthermore, a polarity effect caused by
the polarity dependent injection of charge carriers occurs. Therefore, it is rec-
ommended to use an inhomogeneous, coaxial electrode arrangement in order to
determine the polarity dependent injection in dielectric liquids [6].
Mineral oils used in HVDC transformers are mixtures of several organic
ingredients — e. g. naphthenes, paraffins, aromatic compounds and inhibitors in
order to obtain specific properties, dependent on their application. The dielectric
behaviour of these ingredients also influence the dielectric properties of mineral
oils at DC voltage stress. In this investigation, measurements were performed
with different aged and unaged mineral oils as well as a pure substance in order
to investigate the influence of the different composition of mineral oils and their
ageing on the injection of charge carriers. Comparing the results with the known
processes according to the charge carrier-based approach, the aim of this inves-
tigation is to improve the understanding of the influence of the ingredients of
minerals oils on their nonlinear conductivity.

2 Charge Carrier-Based Modelling

2.1 Physical Fundamentals
Previous investigations presented a charge carrier-based approach of modelling
the nonlinear behaviour of dielectric liquids [3,4]. The nonlinear electric conduc-
tivity in dielectric liquids is a result of
 Polarity Dependent Charge Carrier Injection in Dielectric Liquids 1123

• The generation of homo charges and hetero charges,

• Their movement due to electrostatic forces and diffusion processes and
• Their recombination with other charge carriers or at bare electrodes.
According to this modelling, there are different species of charge carriers of each
polarity. Oil intrinsic charge carriers (hetero charges), which are formed by disso-
ciation processes — e. g. of ageing products, admixed aromatic compounds in the
mineral oil, etc. [3,9,10], hardly recombine and thus accumulate at oppositely
charged electrodes due to their micelle structure. Homo charges (electrons e− or
holes h) are injected from the electrodes dependent on the local electrical field
strength and recombine at oppositely charged electrodes or with other injected
charge carries dependent on the local charge carrier density [3,4,11].
Hence, the drift of charge carriers and the field dependent charge carrier
injection causes a measurable current through an arrangement at DC voltage
stress. In stationary conditions the current density J in the arrangement and
thus the stationary electric conductivity κ is primary a result of the injection of
charge carriers of both polarities, dependent on the electrical field strength [6]:
J = Γ e ≈ Ee (zh μh nh + ze- μe- ne- ) (1)
— where J is the current density in stationary conditions, Γ is the injection rate,
e is the elementary charge, E is the electrical field strength, z is the valence
number, μ is the mobility and n is the concentration of charge carriers of a
species.
Furthermore, the injection of homo charges and thus the stationary conduc-
tivity κ is polarity dependent [6]. However, accumulated hetero charges at the
electrodes as well as injected homo charges affect the injection of homo charges
due to their influence on the local electrical field strength at the electrodes
(Fig. 1) – known as the Malter-effect [6,12].

w/o SC
x w/ SC

x
HV GND HV GND

hetero charges homo charges

Fig. 1. Schematically comparison of the electrical field strength distribution at high

positive DC voltage stress in stationary conditions with and without the influence of
space charges (SC)

Therefore, it is recommended to measure the electric conductivity and thus

the injection rates at a coaxial, inhomogeneous arrangement with a small inner
1124 T. Gabler et al.

electrode [6]. A small radius of curvature of the electrode leads to a low influence
of hetero charges on the electric field at the injecting electrode compared to a
homogeneous arrangement as recommended in [7]. Due to the different properties
of the ingredients of mineral oil and the varied amount of charge carriers, it is
assumed that different dielectric liquids show different injection rates of homo
charges.

2.2 Numerical Modelling

The Poisson-Nernst-Planck equation system (PNP) can be used to model the

influence of each charge carrier species i on the dielectric behaviour [3,4]. As
a part of the equation system, the Poisson equation (2) describes the effect of
each charge carrier species on the electric field. Furthermore, the Nernst-Planck
equation (3) describes the time dependent change in the concentration of each
species by their drift due to the electric field and the density dependent diffusion
as well as the injection and recombination of charge carriers.
e 
− Δϕ = zi ni (2)
εr ε0
∂ni 
= zi μi ni Δϕ + Di Δni + rij (3)
∂t
— where ϕ is the electric potential, ε is the permittivity, t is the time, D is the
Diffusion coefficient of a species, and r is the source term for the recombination
and ionization of each species i with species j.
These equations illustrate, that the concentration of each charge carrier
species influences the electric field distribution and thus the dielectric stress
within an insulation system and vice versa — e. g. due to the field dependent
injection of charge carriers. These mutual influences cause the nonlinear con-
ductivity of mineral oil, especially caused by the effect of its ingredients due to
the formation of hetero charges dependent on the composition and the ageing
conditions of the used oil.

3 Experimental Setup
3.1 High Voltage Test Circuit

The test circuit consists of a one-phase, one-pulse rectifier circuit with an addi-
tional damping resistance Rd which provides DC voltages up to UDC = ±130kV
(Fig. 2). A resistive voltage divider measures the voltage. The smoothing capaci-
tor C of the rectifier circuit, the damping resistor Rd and the capacitance of the
electrode arrangement (EA) form a low-pass filter in order to reduce the influ-
ence of the ripple and the caused displacement current through the electrode
arrangement of the DC voltage on the current measurements.
 Polarity Dependent Charge Carrier Injection in Dielectric Liquids 1125

One-phase rectifier circuit with damping Test

resistor and ohmic voltage divider vessel

D Rd
R1
C EA
R2 V A

Fig. 2. Schematic diagram of high voltage test circuit with test vessel and current
measurement system

3.2 Current Measurement System

The current measurement system is connected directly to the electrode arrange-

ment and consists of a transimpedance amplifier. Low pass filters and amplifiers
are used in order to measure the direct current through the arrangement without
a disturbing influence of displacement currents caused by stray capacitances of
the test setup (Fig. 3). The measured current and DC voltage of the arrangement
is visualized and stored simultaneously via Bluetooth connection.

VDC

VDC
t
RM
I
I
-
+

v v t
fc = 5 Hz fc = 5 Hz
Storage
Transimpedance amplifier with filters
unit

Fig. 3. Schematic diagram of current measurement system based on transimpedance

amplifiers with filters and storage unit

3.3 Test Setup and Electrode Arrangement

In the experiments, four different dielectric liquids were used: an unaged and an
aged naphthenic mineral oil, an unaged isoparaffinic mineral oil and the pure
paraffin n-Dodecane. The liquids were used in order to show differences in the
injection rates at DC voltage stress caused by the different composition of the
1126 T. Gabler et al.

used mineral oils compared to the pure substance without any additives. The
measurements were performed at room temperature without a special treatment
of the used liquids.
A coaxial wire-cylinder arrangement with a geometrical field efficiency fac-
tor of η = 0.14 is used as electrode arrangement (Fig. 4) and is located in a
vacuum desiccator used as test vessel. Previous investigations showed, that an
inhomogeneous, coaxial arrangement is more suitable in order to measure the
polarity dependent conductivity and injections rates of a dielectric fluid [6,8]
compared to a homogeneous arrangement as recommended in [7]. The grounded
outer cylindrical electrode was used as measurement electrode and was painted
on the outer surface in order to reduce the influence of ‘parasitic currents’ on
the measurements. Furthermore, an additional grounded metallic cage was used
as a guard electrode. The electrode arrangement was sealed with a barrier made
of PTFE (Polytetrafluorethylene) to avoid an influence of the surrounding vol-
ume on the current measurements. Additional charge carriers would influence
the measurements due to the current caused by their charge and their influence
on the electrical field distribution [6,13].

Guard
electrode
High voltage
electrode
Sealing
barrier

A Measurement A
electrode

Fig. 4. Schematic diagram of the used coaxial electrode arrangement

The experiments were performed at both polarities of the DC voltage in order

to investigate the polarity dependent behaviour of the liquids. The arrangement
had to be conditioned electrically before each measurement in order to achieve
reproducible and comparable measurements [6]. For this purpose, the experi-
ments were started at the maximum voltage step of about 75% of the expected
breakdown voltage of the arrangement until a stationary current is achieved.
After conditioning, the voltage was decreased and increased stepwise for several
times at each measurement. It is necessary to achieve stationary conditions and
the equilibrium in the generation and recombination of charge carriers in order
to evaluate the results according to the injection of homo charges.
 Polarity Dependent Charge Carrier Injection in Dielectric Liquids 1127

4 Experimental Results
4.1 Measurement of Electric Conductivity

The electric conductivity κ of each dielectric liquid is determined by the mea-

sured current Istat through the electrode arrangement in stationary conditions,
the voltage V at each voltage step and the geometry of the arrangement:
 
Istat rGND
κ= ln (4)
V 2πl rHV

— where l is the effective length of the measurement electrode, rGND is the

radius of the grounded measurement electrode and rHV is the radius of the high
voltage electrode.
The calculated conductivities of the dielectric liquids show the expected
polarity effect [6] with differences of several orders of magnitude (Fig. 5). The
electric conductivity determined at negative DC voltage stress is much higher
than the conductivity measured at positive DC voltage, especially at higher field
strengths.

0 1 2 3 4 5 6 0 1 2 3 4 5 6
-8 -8

-9 -9
lg κ in S/m

lg κ in S/m

-10 -10

-11 -11

-12 -12

-13 -13
Emean in kV/mm Emean in kV/mm
Naphthenic Naphthenic aged Naphthenic Naphthenic aged
Isoparaffinic n-Dodecane Isoparaffinic n-Dodecane
(a) Positive DC voltage (b) Negative DC voltage

Fig. 5. Calculated electric conductivity κ of the used liquids dependent on the mean
electrical field strength Emean in the electrode arrangement

The results also show a minimum of the measured electric conductivity

between (1. . . 1.5) kV/mm at positive polarity and a minimum at negative polar-
ity at lower field strength of less than 1 kV/mm for all liquids, which was
also found in [6]. Furthermore, the results show a higher conductivity of the
naphthenic oils at positive polarity for low field strengths compared to the
isoparaffinic oil or the pure paraffin. At negative polarity, the conductivity of
the isoparaffinic oil is lower at low field strengths but strongly increases between
1128 T. Gabler et al.

Emean ≈ (1 . . . 2) kV/mm. The paraffin n-Dodecane shows a similar behaviour

of the conductivity at negative polarity but with a smaller increase and a lower
conductivity compared to the mineral oils. Nevertheless, all of the investigated
mineral oils show a similar conductivity at high field strengths for each polarity
under DC voltage stress.

4.2 Calculated Injection Rates

The observed polarity effect in the determined electric conductivity of the used
liquids results from the polarity dependent injection of charge carriers because of
the much higher electrical field strength at the high voltage electrode as shown
in [6] (Fig. 1). According to (1), the current through the inhomogeneous arrange-
ment in stationary conditions is mainly caused by the injection of charge carriers
of a species i dependent on the polarity of the high voltage electrode — i. e. the
injection of electrons at negative polarity and the injection of holes at positive
polarity. Hence, the injection rate Γi for each species i is calculated by the sta-
tionary current Istat and the electrode area of the injecting high voltage electrode
AHV :
Jstat Istat
Γi = = (5)
e eAHV
The results show the expected field dependent injection of charge carriers
with higher injection rates of negative than positive homo charges dependent on
the analytic field strength EHV at the injecting high voltage electrode (Fig. 6).
The results also show a similar increase of the injection of positive homo charges
(holes) for all of the used liquids in the first place, but they also show higher
rates in the naphthenic oils compared to the paraffinic liquids (Fig. 6a).

20 20

19 19
lg ( in 1/m2s  AHV)

lg ( in 1/m2s  AHV)

18 18

17 17

16 16

15 15

14 14
0 10 20 30 40 0 10 20 30 40
EHV in kV/mm EHV in kV/mm
Naphthenic Naphthenic aged Naphthenic Naphthenic aged
Isoparaffinic n-Dodecane Isoparaffinic n-Dodecane
(a) Positive DC voltage (b) Negative DC voltage

Fig. 6. Comparison of the calculated injection rates Γ of the used dielectric liquids
dependent on the geometrical electrical field strength EHV at the injecting high voltage
electrode
 Polarity Dependent Charge Carrier Injection in Dielectric Liquids 1129

The comparison of the injection rates of negative homo charges (electrons)

shows also a similar behaviour dependent on the electrical field strength, but
also differences of the used liquids are observed (Fig. 6b). The increase in the
injection rates of the aged and unaged naphthenic oil is quite similar. In contrast,
the injection rates of the isoparaffinic oil strongly increases at low field strengths
with a decreasing slope at high field strengths. The injection rates of the pure
paraffinic liquid strongly increases with increasing field strength as well, but at
lower rates compared to the used mineral oils.

4.3 Discussion
The different injection rates of the used mineral oils compared to the pure paraf-
fin n-Dodecane demonstrate the influence of the various ingredients of mineral
oils on the electric field distribution and thus on the injection of charge carriers
in dielectric liquids and their electric conductivity. As shown in [6], additional
hetero charges within the arrangement affect the injection of charge carriers due
to their influence on the electric field at the injecting high voltage electrode
(Fig. 1) — especially at low field strengths. Accordingly, inhibitors and aromatic
compounds in the mineral oils may form additional hetero charges due to their
weak double bonds [4], which leads to higher injection rates of the mineral oils
compared to the pure paraffin.
In particular, this influence is apparent at negative DC voltage stress
(Fig. 6b). The influence of additional hetero charges on the local electric field
and thus on the injection decreases at higher field strengths at the injecting
electrode due to the increasing amount of homo charges [4,12]. Hence, the injec-
tion rates in the used mineral oils converge at high field strengths to an almost
direct dependency on the analytic electrical field strength at the injection elec-
trode. This effect is also visible in the rates in the pure paraffin, which converge
to the rates in the mineral oils. However, the increase of the injection rate of the
used oils starts decreasing at lower field strengths compared to the pure paraffin
because of the effect of hetero charges on the injection is stronger in the mineral
oils. Especially for the aged naphthenic oil a huge influence of the additional
hetero charges due to the high amount of ageing products occurs, which leads to
a strong effect on the local electric field and thus to higher injection rates at low
electric field strengths. The differences in the measured injection rates between
the naphthenic oils and the isoparaffinic oil at low geometrical field strength
may result from the different composition and the different share of inhibitors
or aromatic compounds of the used oils.
The determined injection rates at positive DC voltage show a similar
behaviour of all liquids. However, both naphthenic oils reveal higher injection
rates compared to the paraffinic liquids. The similar increase in the injection
rates of positive homo charges (holes) of the used liquids lead to the assumption
that the different composition of the used liquids leads to a comparable influ-
ence on the injection at positive polarity. Nevertheless, the components of the
mineral oils affect the injection of positive homo charges differently, which might
be the reason of the different injection rates of the used liquids. Comparable to
the injection of negative homo charges (electrons), the influence of the different
1130 T. Gabler et al.

components become negligible at higher field strengths, which leads to the con-
vergence of the positive injection rates of all liquids at higher field strengths. In
contrast to the injection of electrons, it appears that the convergence happens at
much higher field strengths due to the lower injection rates of holes. Similar to
the results shown in [4,6,13] an ascent of the liquids at the liquid–air interface
of the bare high voltage connection was observed during the experiments, which
was much stronger at the mineral oils compared to the pure paraffin. This effect
also illustrates that there is a higher amount of hetero charges in the mineral
oils compared to the pure substance, which affects the injection of homo charges
and thus their dielectric behaviour.
The determined injection rates can be applied to calculations according to
the charge carrier-based approach in order to determine the dielectric stress of
an oil-paper-insulated arrangement at DC voltage stress. Nevertheless, it has to
be considered that the injection rates are determined according to the analytic
field strength at the injecting electrode. The ‘true’ injection rates are dependent
on the local field strength at the oil–electrode interface, which is affected by each
charge carrier species (cf. Fig. 1). In order to determine the ‘true’ injection rates,
the local field strength at the injecting electrode need to be determined as well.

5 Summary and Conclusions

The presented results of the experimental investigations on the used naphthenic

and isoparaffinic mineral oils as well as the pure paraffin n-Dodecane confirm
the nonlinear and polarity dependent conductivity behaviour of dielectric liq-
uids due to the field dependent injection of homo charges (electrons and holes).
The comparison of the measured conductivity and the calculated injection rates
illustrates the strong influence of the injection on the conductivity properties
of dielectric liquids at DC voltage stress. Furthermore, the differences in the
injection rates of the used liquids demonstrate that the different ingredients of
the mineral oils strongly affect the injection and thus the conductivity of the
dielectric liquids. The reason in this behaviour might be the generation of hetero
charges from the ingredients of mineral oil. These charges affect the local electric
field at the electrodes and thus the injection of homo charges (cf. Fig. 1).
As presented in [8], the different injection rates affect the distribution of
the electrical field strength of an oil-insulated arrangement. Hence, different
injection rates of various dielectric liquids may lead to an influence on the field
distribution of oil-paper insulated arrangements as well. These effects may lead
to an impact on the dielectric stress of insulation systems with different dielectric
liquids. Accordingly, the increased injection rates of aged mineral oil compared
to unaged oil as well as the higher amount of generated hetero charges may be
another reason for an increased dielectric stress of aged insulation systems at
DC voltage stress.
Although the effect of the different ingredients on the dielectric behaviour
and the injection of homo charges cannot be explained exactly due to their
unknown composition and the electric behaviour of the ingredients, the results
show that differences in the injection of charge carriers and thus in the dielectric
 Polarity Dependent Charge Carrier Injection in Dielectric Liquids 1131

behaviour exist under DC voltage stress. Hence, the behaviour of the different
components need to be investigated with regard to their ability to generate hetero
charges, their quantity and their effect on the injection rates, especially at low
field strengths. Furthermore, it is assumed that the injection in dielectric liquids
shows a similar behaviour at electrode arrangements with a very small radius
of curvature of the injecting electrode due to the negligible influence of hetero
charges. With this knowledge, the injection of charge carriers can be evaluated
concerning the ‘true’ field strength at the oil–electrode interface.

Acknowledgement. The authors gratefully thank SIEMENS AG, Division Trans-

formers in Nuremberg, Germany for funding the research cooperation over the past
years.

References
1. Friends of the Supergrid, Roadmap to the Supergrid Technologies. www.friendsoft
hesupergrid.eu/wp-content/uploads/2013/07/Supergrid-Technological-Roadmap-
2016-FINAL1.pdf. Accessed on 30 Nov 2018
2. Arrillaga, J.: High voltage direct current transmission, 2nd edn. Institute of Elec-
trical Engineering, London (1998)
3. Backhaus, K., Speck, J., Hering, M., Großmann, S., Fritsche, R.: Nonlinear dielec-
tric behaviour of insulating oil under HVDC stress as a result of ion drift. In:
International Conference on High Voltage Engineering and Application (ICHVE),
pp. 1–4 (2014)
4. Backhaus, K.: Das dielektrische Verhalten der Öl-Papier-Isolierung bei Belastung
mit hoher Gleichspannung. Dissertation. Technische Universität Dresden, Dresden
(2016)
5. Liebschner, M.: Interaktion von Ölspalten und fester Isolation in HVDC-
Barrierensystemen. Dissertation, Technischen Universität IImenau, Düsseldorf
(2009)
6. Backhaus, K., Speck, J., Gabler, T., Großmann, S., Fritsche, R.: The nonlinear
conductivity properties of mineral insulating oil and their effect on conductivity
measurements. In: VDE-Hochspannungstechnik, Berlin, pp. 596–601 (2016)
7. CIGRE JWG A2/D1.41, HVDC Transformer Insulation: Oil Conductutivity, TB
646, Paris (2016)
8. Gabler, T., Backhaus, K., Götz, T., Großmann, S., Fritsche, R.: Effect of the non-
linear electric conductivity of mineral in-sulating oil on the dielectric strength at
high DC voltage stress. In: 2018 IEEE Conference on Electrical Insulation and
Dielectric Phenomena (CEIDP), Cancun, Mexico, pp. 54–57 (2018)
9. Onsager, L.: Deviations from Ohm’s law in weak electrolytes. J. Chem. Phys. 2(9),
599 (1934)
10. Wien, M.: Über den Spannungseffekt der elektrolytischen Leitfähigkeit in sehr
starken Feldern. Ann. Phys. 393(3), 400–416 (1929)
11. Felici, N.: High-field conduction in dielectric liquids revisited. IEEE Trans. Elect.
Insul. EI–20(2), 233–238 (1985)
12. Malter, L.: Thin film field emission. Phys. Rev. 50(1), 48–58 (1936)
13. Gabler, T., Backhaus, K., Speck, J., Großmann, S., Fritsche, R.: Investigations on
the charge carrier based breakdown behaviour of oil-paper-insulations at high DC
voltage stress. In: VDE-Hochspannungstechnik, Berlin, pp. 732-737 (2016)
 Modelling Space Charge in HVDC
Cable Insulation

Yunpeng Zhan1(&) , George Chen1, Miao Hao1 ,

Zhiqiang Xu1 , Lu Pu2 , Xuefeng Zhao2 , Haofei Sun2 ,
Sen Wang2 , Anxiang Guo2 , and Jian Liu2
1
University of Southampton, Southampton, UK
yz10n16@soton.ac.uk
2
State Grid Shaanxi Electric Power Research Institute, Xi’an, China

Abstract. The design of high-voltage direct-current extruded cable is one of

the most challenging issues in the cable industry, as the electric field distribution
across the insulation can be strongly affected by the presence of space charge,
which can subsequently affect its long-term reliability and life expectancy. In
this study, the bipolar charge transport model was utilized to calculate space
charge and field distribution in a polymeric cable insulation, and the result was
compared with the one obtained by the conductivity model which is commonly
used in the cable industry. It is shown that the simulation results of the bipolar
charge transport model are more comparable with the previous experimental
work, and the shortcomings of the conductivity model are presented. At last, the
feasibility and potential issues of the new method are discussed for further
development.

Keywords: HVDC cable
Field distribution
Bipolar charge transport model

1 Introduction

In recent decades, the market for high-voltage direct-current (HVDC) transmission

systems is booming, as this technology contributes large power delivery over long
distance and the integration of renewable electricity sources [1]. Due to its excellent
electrical and thermal properties, extruded cable becomes the key component for
HVDC transmission system [2]. However, the design of HVDC extruded cable requires
special considerations, as it is reported that the electric field distribution across the
insulation thickness can be distorted by the presence of space charge accumulation [3].
Indeed, space charge can easily accumulate in the polymeric insulation under DC field,
due to charge injection at electrodes and to dissociation of by-products and impurities
within the insulation bulk. If the space charge density becomes sufficiently high, the
local electric field strength may exceed the breakdown strength of the dielectric,
leading to insulation ageing and even failure [1].
With the assumptions of the conductivity/resistivity of the insulation depending on
temperature and field, some methods of calculating field distribution in DC cable have
been developed [4], and this kind of methods is serving as an important reference for
DC cable design. Nevertheless, these models cannot predict charge generation and
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1132–1142, 2020.
https://doi.org/10.1007/978-3-030-31676-1_106
 Modelling Space Charge in HVDC Cable Insulation 1133

transport, and the space charge is only derived from the conductivity gradient, which
could limit its performance [5]. Since the work proposed by Alison et al. on charge
transport modelling in polyethylene based material in 1994 [6], the bipolar charge
transport model has been applied to simulate charge dynamics by many researchers,
achieving a good fit when compared with the experimental results, but few attempts
have been made to investigate the space charge behaviour in cable geometry [7–9].
By using the COMSOL Multiphysics software, we compare the simulation results
obtained using the above two models in a polyethylene-insulated cable, considering a
temperature gradient across the cable insulation. Differences have been shown between
the conductivity model and the charge transport model, as the charge injection and
transport are not considered in the macroscopic model. Besides, some suggestions for
the further improvement of the bipolar charge transport model are given.

2 Descriptions of the Two Methods

From a macro perspective, a weak current could be formed in polyethylene-based

materials under DC field, and this current may not be uniform due to the non-
homogeneity of the insulating material. On the basis of the current continuity equation,
a dielectric sample where a DC current of density j is flowing, and where a divergence
takes place between the incoming and outgoing charge flow, will acquire a space
charge density q.

! @q
r
j ¼ ð1Þ
@t

By combing the other three essential equations i.e.: Gauss law, Ohm’s law and the
electrostatic electric field,
 !
r
e0 er E ¼ q ð2Þ

! ! !
j ¼ r E ¼ lnq E ð3Þ
!
E ¼ rV ð4Þ

where E is the electric field strength, r is the conductivity, e0 is the vacuum permit-
tivity, er is the relative permittivity, µ is the mobility of charge carriers and V is the
electrical potential, space charge density can be described in the non-homogeneous
weakly conductive material as:

e0 er @q ! e e 
0 r
q¼ þrE
r ð5Þ
r @t r
1134 Y. Zhan et al.

2.1 The Conductivity Model

By assuming the volume conductivity of the insulation depends on temperature and
field, the conductivity model was developed to anticipate the field distribution in loaded
cable systems [4]. From Eq. (5), space charge would accumulate if the ratio
permittivity/conductivity varies with position. Indeed, the permittivity of the insulation
could be treated as a constant within the working range of temperatures and fields,
however, the conductivity is generally considered to be dependent on both temperature
and field. The conductivity expression of Arrhenius’ law has been widely used to
describe such relationship in the synthetic insulation material [10, 11]:
 
Ea sinhðBjEjÞ
rðE; TÞ ¼ A
exp ð6Þ
kB T jE j

where kB is the Boltzmann’s constant, E is the electric field, T is the temperature.

Constants A, B and Ea are obtained by fitting the conductivity equation with the
conduction current measurements on the planar samples of the material under a range
of temperatures and fields. In this model, the conductivity gradient formed across the
insulation drives the charge dynamics, therefore the field distribution can be obtained
straightforward.

2.2 The Bipolar Charge Transport Model

Differently from the conductivity model, the bipolar charge transport model features the
bipolar injection of electronic charge carriers from the electrodes when the DC field
exceeds the threshold. As shown in Fig. 1, four species are considered in the bipolar
charge transport model. Under the force of the electric field, these injected positive and

Fig. 1. Four kinds of charge carriers in bipolar charge transport model. Si, Bi and Di are
recombination, trapping and de-trapping coefficients respectively. ni is the charge density.
Indexes e and h refer to electrons and holes; l and t refer to mobile and trapped charge carriers
[7].
 Modelling Space Charge in HVDC Cable Insulation 1135

negative charge carriers shift to the reversed polarity electrodes. For the big band-gap
material, polyethylene, the localized energy states which originated from physical or
chemical defects can trap charge carriers, and these captured charge carriers can also
escape from these traps. Additionally, these charge carriers with opposite polarity have
the possibility to recombine, resulting in electroluminescence. A hopping type of
mobility, which is field and temperature dependent, is used to describe the conduction
of charge carriers [8]. For sake of simplification, space charge generation only depends
on the injection from semi-conductive electrodes, and the ionization process is not
taken into account. The boundary condition has already been reported in the previous
work [9].
It should be noted that in the bipolar charge transport model, the source terms si,
which are defined to describe the local variations of density of given specie, are
introduced for solving the current continuity equation. The source terms encompass the
trapping, de-trapping and recombination processes, for mobile electrons, seµ can be
presented as:
 
@nel net
sel ¼ ¼ S1 nht nel  S3 nhl n  Be nel 1  þ De net ð7Þ
@t noet

3 Simulated Results

As shown in Fig. 2, a medium voltage cross-linked polyethylene (XLPE) cable of

4.5 mm insulation thickness, is used for simulation by the both models. The average
electric field is around 20 kV/mm, and a 20 °C temperature gradient is applied along
the radial direction. The electrical and thermal setting are following the previous
experimental work [11].

Fig. 2. Schematic representation of the medium voltage cable sample.

1136 Y. Zhan et al.

Table 1 gives the parameters which used in the macroscopic model, and they were
also from literature [11].

Table 1. Symbols used for the macroscopic model [11].

Symbol Value Unit
A 1*1014 A/m2
B 2*10−7 V−1m
Ea 1.48 eV

Figure 3 presents the time-dependent charge distribution simulated by the con-

ductivity model. It should be claimed first that the charge accumulation in such model
could be contributed by all types of charge carriers, including ions, holes and electrons.
With the positive voltage applying at the inner electrode, only one-polarity (positive)
charge can be observed in the insulation. The accumulation of charge is believed to
originated from the nonuniform current density along the radius. At first, due to the
presence of temperature gradient and the Laplacian field, a very large conductivity
gradient could be formed along the insulation, therefore charges accumulates very
quickly at this stage, especially at the inner part. As the steady field distribution is
resistive under DC stress, the outer part of the insulation shall bear more electric stress
due to a lower conductivity, as shown in Fig. 4. The field variation could decrease the
conductivity gradient and slow the charge accumulation. At last, the steady state is
reached till 20000 s.

Fig. 3. The distribution of charge density simulated by the conductivity model.

 Modelling Space Charge in HVDC Cable Insulation 1137

Fig. 4. Field distribution in the cable insulation predicted by the conductivity model.

For bipolar charge transport model, in order to simulate the transport process of
holes and electrons respectively, the module of “Transport of Diluted Species (TDS)”
has been selected in the COMSOL, a finite element method software, to calculate the
density of each specie migrating in the electric field. The parameters used in this model
are listed in Table 2 [8].

Table 2. Parameters applied in the

bipolar charge transport model [8].
Parameter Value Unit
Injection barrier heights
Wei for electrons 1.27 eV
Whi for holes 1.16 eV
Trapping coefficients
Be for electrons 0.05 s−1
Bh for holes 0.05 s−1
Trap depths
wµe for electrons 0.71 eV
wµh for holes 0.65 eV
Trap densities
noet for electrons 100 C/m3
noht for holes 100 C/m3
De-trapping barrier height
Wtre for electrons 0.96 eV
Wtrh for holes 0.99 eV
1138 Y. Zhan et al.

As shown in Fig. 5, the space charge distribution is very different from the result of
the conductivity model, because different processes dominate the charge accumulation
in the bipolar charge transport model together. At the beginning, the charge injection is
accountable for the charge accumulation at the interface. Compared with electrons,
more holes are injected into the bulk, this is not only because the higher temperature at

Fig. 5. Simulated space charge distribution within the insulation bulk evolving with time.

Fig. 6. Computed field distribution within the insulation by the bipolar charge transport model.
 Modelling Space Charge in HVDC Cable Insulation 1139

the inner side but also the assumption of lower injection barrier for holes. The large
amount of homo-charges decreases the field at the anode severely, as shown in Fig. 6,
resulting in a nearly zero field. For longer polarization time, the trapping/de-trapping
processes seem to have a bigger impact on the charge distribution. The charge deep
penetration and these trapped charges affect the field distribution greatly, making the
maximal electric field transfers from inner part to the outer gradually.

4 Comparisons of the Results

Both models are able to anticipate the field inversion in the cable with the presence of
temperature gradient, but there are many differences between their results and the
experimental observation. First, the space charge distribution performed by the con-
ductivity model is very different from the experimental data shown in Fig. 7, as the
conductivity model can only provide an overall contribution from all types of charge
carriers, while it lacks specific transport information of each kind of charge carriers.

Fig. 7. Space charge profiles obtained by PEA techniques with the presence of a 20 °C
temperature gradient [11].

Meanwhile, the last electric field distribution predicted by the macroscopic model at
the outer side is 25 kV/mm, which is fairly underestimated compared with the results
shown in Fig. 8. In contrast, the charge movement simulated by the bipolar charge
transport model is very consistent with the experimental observation, it is believed that
the well descriptions on charge injection and transport processes are very useful.
However, the large amount of accumulated charge at the vicinity of electrodes can
reduce the field strength there badly, resulting in a surprisingly low electric field at the
beginning of the simulation. The maximum electric field at the outer electrode pre-
dicted by the bipolar charge transport model is about 36 kV/mm, which is also very
closed to the experimental data.
1140 Y. Zhan et al.

Fig. 8. Field distribution in the cable insulation (applied voltage 90 kV). [11].

Figure 9 shows the total charge amount inside the bulk of the conductivity model, it
can be observed that the charge amount increased gradually in the first 10000 s, and the
final total charge amount was maintained at about 5.2  10−4 C/m2. The charge
amount of holes and electrons is shown in Fig. 10 separately. Differently from the
prediction of the conductivity model, the amount of holes was increased very fast in the
first 100 s, then it continued slightly increasing until 30000 s, as the nearly zero field at
the anode prevent further injection of holes. The amount of electrons increased more
gently, so the total net charge amount first increased, then decreased and maintained at
about 5.1  10−4 C/m2. It should be noted that the hetero-charge build-up is not
considered in the bipolar charge transport model, which may be accountable for the
difference.

Fig. 9. The total accumulated charge amount calculated by the macroscopic model.
 Modelling Space Charge in HVDC Cable Insulation 1141

Fig. 10. The charge amount for holes and electrons separately during the polarization simulated
by the bipolar charge transport model.

Another drawback of the conductivity model is found that the parameters of the
conductivity equation are not dependent on position. On the one hand, the surface
effect could affect the conductivity at the interface [12], making it differ from the one
insider the bulk. On the other hand, the impurity concentration could be nonuniform
along the radius, which also affects the inhomogeneity of the cable insulation.
Due to the well descriptions of charge injection and transport in the bipolar charge
transport model, the simulation is more complicated and can cost much more time than
the conductivity model. In spite of that, both the charge movement and the field
variation seem to be more consistent when compared with the experimental data.
However, it still need to be further developed before it can be applied in practical
applications. The parameters that related to charge generation and transport mecha-
nisms cannot be obtained by independent experiments straightforward. The estimation
of parameters based on experimental data still need to be developed. Additionally, the
ionization mechanism should be considered in the charge transport model in order to
expand its application.

5 Conclusions

Field estimation in HVDC cable insulation has been made by the macroscopic model and
the bipolar charge transport model. With the presence of a temperature gradient, the stress
inversion has been observed by both models. Compared with the experimental obser-
vations, the results of the bipolar charge transport model were more consistent, even with
parameters that are not optimized for a XLPE material. The well descriptions on the
charge injection at the interface and the complicated transportation processes were
1142 Y. Zhan et al.

believed as the main reasons for the better performance of the bipolar charge transport
model. It is suggested that the ionization mechanism requires to be considered in the new
approach for the further development, and the parameterization for XLPE is still needed.

Acknowledgments. We would like to acknowledge the financial support from the technology
project of State Grid Shaanxi Electric Power Company [5226KY16001G] and [2018GY-001].

References
1. Mazzanti, G., Marzinotto, M.: Extruded Cables for High-voltage Direct-current Transmis-
sion: Advances in Research and Development. Wiley-IEEE Press, Hoboken (2013). Power
Engineering Series
2. Ghorbani, H., Jeroense, M., Olsson, C.-O., Saltzer, M.: HVDC cable systems—highlighting
extruded technology. IEEE Trans. Power Deliv. 29(1), 414–421 (2014)
3. Chen, G., Hao, M., Xu, Z., Vaughan, A., Cao, J., Wang, H.: Review of high voltage direct
current cables. 10 Csee J. Power Energy Syst. 1(2), 9–21 (2015)
4. Morshuis, P.H.F., Bodega, R., Fabiani, D., Montanari, G.C., Dissado, L.A., Smit, J.J.:
Calculation and measurement of space charge in MV-size extruded cables systems under
load conditions. In: IEEE International Conference in Solid Dielectrics, ICSD 2007,
pp. 502–505 (2007)
5. Vu, T.T.N., Teyssedre, G., Vissouvanadin, B., Le Roy, S., Laurent, C., Mammeri, M.,
Denizet, I.: Field distribution in polymeric MV-HVDC model cable under temperature
gradient: simulation and space charge measurements. Eur. J. Electr. Eng. 17, 307–325 (2014)
6. Alison, J.M., Hill, R.M.: A model for bipolar charge transport, trapping and recombination
in degassed crosslinked polyethene. J. Phys. D Appl. Phys. 27, 1291–1299 (1994)
7. Le Roy, S., Teyssedre, G., Laurent, C., Montanari, G.C., Palmieri, F.: Description of charge
transport in polyethylene using a fluid model with a constant mobility: fitting model and
experiments. J. Phys. D Appl. Phys. 39(7), 1427 (2006)
8. Le Roy, S., Teyssedre, G., Laurent, C.: Modelling space charge in a cable geometry. IEEE
Trans. Dielectr. Electr. Insul. 23(4), 2361–2367 (2016)
9. Zhan, Y., Chen, G., Hao, M.: Space charge modelling in HVDC extruded cable insulation.
IEEE Trans. Dielectr. Electr. Insul. 26(1), 43–50 (2019)
10. Boggs, S., Damon, D.H., Hjerrild, J., Holboll, J.T., Henriksen, M.: Effect of insulation
properties on the field grading of solid dielectric DC cable. IEEE Trans. power Del. 16(4),
456–461 (2001)
11. Bodega, R.: Space charge accumulation in polymeric high voltage DC cable systems. PhD
dissertation, Delft University of Technology (2006)
12. Boggs, S.: A rational consideration of space charge. IEEE Electr. Insul. Mag. 20(4), 22–27
(2004)
 Study on Signal Correction Method
for Measurement of Space Charge Distribution
of Multi-dielectrics Using PEA Method

HyungGyu Kim(&), HyunCheol Jung, and Seung Hwangbo

Honam University, Gwangju, Republic of Korea

khg8532@gmail.com

Abstract. With the advancement of power transmission technology and

increase of long-distance transmission between countries, high voltage direct
current (HVDC) transmissions have attracted a lot of interest. However, there is
a problem with multi-dielectrics, arsing from the structure of the cable con-
nection boxes for long-distance transmission. The main issues are the temper-
ature gradient in the cable insulator under DC electric stress and measurement
waveforms distortion due to non-uniformity of the dielectric density and mis-
match of acoustic characteristics of adjacent dielectrics. Therefore, the signal
calibration process is essential to obtain accurate measurements and reliable
signals. In this study, the pulsed electro-acoustic method is used to measure
multi-dielectric space charge distribution signals and a new signal processing
algorithm is developed. The effects of space charges in the material are analysed
and the reliability of the measured signal is significantly improved.

Keywords: Space charge
Multi-dielectrics
PEA method
Signal

processing
Deconvolution

1 Introduction

In recent year, as the demand for large-scale power transmission and DC cables has
increased due to the connection between inter-country grid and offshore wind power
generation around the world, high voltage direct current (HVDC) cables have been
extensively studied. However, as for long-distance transmission, such as for power grid
and offshore wind power generators, a DC high voltage is continuously applied to the
internal insulator of the cable, there is a high probability of insulation breakdown due to
space charge. The space charge can cause electric insulation inside the cables, leading
to insulation breakdown by injecting, creating or moving electric charge inside the
insulator [1]. The cable connection box consists of several dielectrics and due to the
presence of interfaces between adjacent dielectrics, distortion of the electric field
occurs, and various problems such as electrical insulation occur. Attempts to solve such
problems have been made using the pulsed electro-acoustic (PEA) method, which,
being non-destructive and noise immune, has been used by many researchers. How-
ever, the space charge distribution signal measured using PEA method contains dis-
tortion. This includes ringing and noise due to reflections and refractions at interfaces.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1143–1154, 2020.
https://doi.org/10.1007/978-3-030-31676-1_107
1144 H. Kim et al.

In Furthermore, it is difficult to evaluate the performance of the insulator because the

influence of attenuation and dispersion is directly proportional to the insulator’s
thickness [2]. Therefore, it is necessary to correct the distorted signal. In this study, we
use the deconvolution method whereby the original signal is restored using the mea-
sured signal and the system response function, this method is an algorithm widely used
to obtain the reliability of the signal. We developed a program using signal processing
algorithms and LabVIEW S/W, and successfully corrected the measurement signal.

2 PEA Method

The PEA method was first proposed in 1985 by Takata in Japan and Cooke in the
United States as a non-destructive measurement method to directly measure the space
charge distribution in solid dielectrics. It is a technique to directly or indirectly measure
the space charge inside the insulator. It has excellent reproducibility and is used as a
representative technique in this study because the system configuration is relatively
simple. In this method, a high-voltage short pulse is directly applied to the dielectric to
generate a pressure wave. The space charge formed inside the dielectric is displaced by
the high voltage short pulse and the electrostatic stress, and the wave propagating in the
surroundings is acoustically detected by a piezoelectric element installed in the
detection unit, thereby measuring the space charge distribution [1].

Fig. 1. Principle of the PEA method

Figure 1 shows the principle of space charge distribution measurement using the
PEA method. Multi-dielectric samples with space charge density qðxÞ are placed
between the ground electrode (x = 0) and upper electrode (x = d). When the high
voltage VDC is applied, charges accumulate on the surface of the two electrodes, which
are semiconducting layers. When the high voltage short pulse ep ðtÞ is applied to the
upper electrode through the coupling capacitor Cc , the charges are locally displaced by
the Lorentz force due to the electrostatic stress. The pressure wave generated by the
space charge in the surface of the electrode and inside the sample travels to the
 Study on Signal Correction Method for Measurement of Space Charge Distribution 1145

piezoelectric device (PVDF) and is converted into the voltage signal, so that the space
charge distribution can be measured. The sensor signal from the space charge is linearly
proportional to the amount of charge, but attenuation and dispersion from the circuit
components and the material distort the signal. In the case of a plane specimen of multi-
dielectric, the electrostatic force generated per unit volume F by the external electric
field E can be defined as follows.
e0 2
F¼ E rer þ qE ð1Þ
2

Here, e0 is the vacuum permittivity, er is the dielectric constant of the insulator, and
q is the space charge density. The first term on the right-hand side of Eq. (1) represents
the electrostatic force generated by the discontinuous dielectric constant and the second
term represents the electrostatic force generated by the space charge q in the insulator.
This indicates that although there is no space charge, electric pulses are generated at the
multi-dielectric interfaces. The effects of pressure wave propagation and reflection in
multi-dielectric are omitted in this study because they have been extensively explained
in previous works.

Fig. 2. Schematic representation of the PEA cell for space charge measurements in multi-
dielectrics

Figure 2 shows a schematic diagram of the measurement-cell used to measure the

space charge distribution of XLPE-LSR and LSR-XLPE multi-insulator. The amplitude
of the high voltage short pulse generator ranges from 0 to 5 kV, the period is 10 ns, and
a 50 X feed-through is used for impedance matching of the input signal, thereby
reducing the distortion of the reflected signal. PVDF was used as a piezoelectric
element to convert the pressure wave signal into an electrical signal.
1146 H. Kim et al.

3 Signal Processing
3.1 Signal Processing System Construction
The purpose of this study was to improve the accuracy of space charge distribution
measurements, by correcting the distortion of the signals measured in multi-dielectrics
using the PEA method. Various procedures were attempted including the improvement
of the acoustic impedance matching in the detector, but there is a limit to the distortions
that can be removed. Finally we developed a signal correction program in LabVIEW
S/W. Figure 3 shows a flow chart of the signal correction program, which allows the
user to set parameters such as the thickness of the insulating material and the applied
voltage. In addition, the corrected data and graphs can be saved in Excel files in real
time.

Fig. 3. Signal processing flowchart

 Study on Signal Correction Method for Measurement of Space Charge Distribution 1147

3.2 Deconvolution Processing

Figures 4 and 5 show the space charge distribution signals measured with the PEA
system developed in this study for XLPE 500 lm − LSR 360 lm and LSR
340 lm − XLPE 500 lm, respectively. Input voltages are applied sequentially in 2 kV
increments from 2 to 10 kV. The experimental results confirm the linear relationship
between the applied voltage and the space charge distribution signal, as well as the
presence of distortions, such as noise and signal attenuation. In the case of general
noise, it is possible to compensate it by filtering. However, to correct the distortion of
the signal waveform due to the transfer function, a specific equation is necessary, and
deconvolution method was used in this study. Deconvolution is an algorithm-based
processing method used to separate convolved signals and is widely used for signal and

Fig. 4. Space charge distribution signal from a XLPE 500 lm − LSR 360 lm measured by
PEA method

Fig. 5. Space charge distribution signal from a LSR 340 lm − XLPE 500 lm measured by
PEA method
1148 H. Kim et al.

image processing. On the other hand, to measure the space charge distribution signal,
the signal is obtained through a broadband amplifier and cannot be measured precisely
because the amplifier distorts the signal as a high pass filter [2]. According to the
convolution theorem, the impulse response of the system can be obtained to calculate
the input and output relationship of the corresponding circuit. In other words, assuming
that the amplifier frequency band is sufficient and xðtÞ is an ideal space charge signal,
the measured signal is produced by resolution with the response function hðtÞ of the
system. Thus, if the measured signal and hðtÞ are known, the original ideal signal can
be obtained through deconvolution.

Fig. 6. Deconvolution process for signal correction

Figure 6 shows the deconvolution algorithm used in this study, where yðtÞ is the
signal measured through the oscilloscope, x'ðtÞ represents the ideal output signal and
hðtÞ represents the system response function. To perform the deconvolution hðtÞ must
be known. In this study, hðtÞ is regarded as a space charge distribution signal on the
ground electrode of the measurement signal yðtÞ. In other words, an ideal delta voltage
is applied to generate a delta pressure wave through interaction with the space charge
on the ground electrode, which is considered to be distorted by propagation. However,
it is assumed that there is no attenuation and dispersion at this time. Here, y'ðtÞ is a
signal simulating an input waveform and is generated by a Gaussian function. The
Gaussian function also acts as a filter. In Fig. 6, yðtÞ and hðtÞ are transformed into the
frequency domain by a fast Fourier transform (FFT) and the Gaussian function is then
applied. x'ðtÞ can be obtained by transforming it into the time domain through an
inverse FFT (IFFT) as shown in Eq. (4)
 Study on Signal Correction Method for Measurement of Space Charge Distribution 1149

yðtÞ ¼ hðtÞ  xðtÞ ð2Þ

Yðf Þ ¼ Hðf ÞXðf Þ ð3Þ


Yð f Þ 0
xðtÞ ¼ F 1 Y ðfÞ ð4Þ
Hð f Þ

Fig. 7. Space charge distribution signal from a XLPE 500 lm − LSR 360 lm after deconvo-
lution processing

Fig. 8. Space charge distribution signal from a LSR 340 lm − XLPE 500 lm after deconvo-
lution processing
1150 H. Kim et al.

Figures 7 and 8 show the space charge distribution signal on the compensated top
and bottom electrodes after deconvolution of the signals in Figs. 4 and 5, respectively,
where the reflected waves and noise in the measured signal are corrected. However, it
can be seen that the interfaces between adjacent dielectrics of different thickness
produce distortion. This leads to the next calibration step.

3.3 Calibration
Further corrections are needed to convert the measured signal from units of time and
voltage to units of insulation thickness and charge density. In this paper, the unit
compensation is performed using the space charge formed on the lower electrode.

re ¼ e0 er Ee ð5Þ

V
Ee ¼ ð6Þ
d

Here, V is applied voltage, d is sample thickness. The correction factor Zcal for the
conversion to charge density can be obtained from the lower electrode signal pðxÞ as
follows:
R x2
x1 pð xÞdx
Zcal ¼ ð7Þ
re

The space charge distribution unit can be converted from mV to C=m3 by the cor-
rection coefficient thus obtained. The signal after calibration is shown in Figs. 9 and 10.

Fig. 9. Space charge distribution signal from a XLPE 500 lm − LSR 360 lm after signal
processing
 Study on Signal Correction Method for Measurement of Space Charge Distribution 1151

Fig. 10. Space charge distribution signal from a LSR 340 lm − XLPE 500 lm after signal
processing

The unit and thickness conversions are performed through the calibration, but it can
be confirmed that the thickness correction was unsuccessful. This is attributed to the
difference in the insulating layers causing the acoustic characteristic mismatch.

Fig. 11. Space charge distribution signal using PEA method

1152 H. Kim et al.

Fig. 12. Space charge distribution signal from a XLPE 500 lm − LSR 360 lm after calibration

Therefore, we have developed an algorithm to show the exact thickness at heteroge-

neous interfaces by correcting the acoustic property mismatch.
Figure 11 shows the signal after the deconvolution process measured from the short
pulse of 2 kV voltage, 10 ns period, and 2 to 10 kV HVDC. The data were decom-
posed base on the red line and the calibration was performed considering the respective
acoustic velocities. Figures 9 and 10 show the space charge distribution signal, con-
sidering the acoustic characteristics of an insulator, ignoring the fact that it is a multi-
dielectric. Figures 12 and 13 show the signal after calibration.
 Study on Signal Correction Method for Measurement of Space Charge Distribution 1153

Fig. 13. Space charge distribution signal from a LSR 340 lm − XLPE 500 lm after calibration

It can be seen that the thickness of each insulator as well as the corresponding
electric field distribution and potential, is corrected in Figs. 12 and 13. Additionally, it
was confirmed that the upper signal considering the original acoustic characteristics
was corrected.

4 Conclusions

In this study, the space charge distribution signal was detected using the PEA method
by applying a short pulse of 2 kV, 10 ns and HVDC of 2 to 10 kV successively to
XLPE 500 lm − LSR 360 lm and LSR 340 lm − XLPE 500 lm samples. The
correction was performed using a specific type of algorithm. Furthermore, when
compensating the space charge distribution signal in multi-dielectrics, the thickness and
distortion signals were corrected by calibrating the acoustic characteristics of each
sample. At present, we are studying attenuation of multi-dielectrics space charge
accumulation and signal compensation for temperature signals. If attenuation is com-
pensated, more reliable signal correction is feasible.
1154 H. Kim et al.

ACKNOWLEDGMENTS. This work was supported by the Korea Institute of Energy Tech-
nology Evaluation and Planning (KETEP) and the Ministry of Trade, Industry & Energy
(MOTIE) of the Republic of Korea (No. 20171210201080).

References
1. Hwangbo, S., Kwun, Y.H., Jeon, S.I., Han, M.K.: Direct correlation between space charge
and conduction characteristics of LDPE at various temperature. Jpn. J. Appl. Phys. 37, 4419–
4427 (1998)
2. Tsekmes, I.A., et al.: Space charge accumulation in polymeric DC mini-cable. In: 2013 IEEE
International Conference on Solid Dielectrics (ICSD), pp. 452–455. IEEE (2013)
3. Kim, H.G., Jung, H.C., No, H.G., Hwangbo, S.: A study on the space charge distribution
signal correction of multi-dielectrics using PEA method. In: Korean Institute of Electrical
Engineers Conference, pp. 19–20 (2018)
4. Bodega, R., Morshuis, P.H.F., Smit, J.J.: Space charge measurements on multi-dielectrics by
mean of the pulsed electroacoustic method. IEEE Trans. Dielectr. Eletr. Insul. 13(2), 272–281
(2006)
5. Le Roy, S.: Computer simulation of space charge distribution in an XLPE-EPR sandwich.
IEEE (2005)
6. Bodega, R.: Space charge accumulation in polymeric high voltage dc cable system. Delft
University of Technology (2006)
7. Lan, L., Yin, Y., Wu, J.: Recovery of space charge waveform in multi-layer dielectrics by
means of the pulsed electro-acoustic method. Proc. CSEE 36(2), 570–576 (2016)
 Charge Accumulation on Slightly Conductive
Barrier Systems and Its Effect on Breakdown
Voltage in an Air Insulated Rod Plane
Arrangement

M. Schueller1(&), A. Blaszczyk2, F. Mauseth3, H. K. Meyer3,

N. Stieger1, and J. Smajic1
1
University of Applied Sciences Rapperswil, Rapperswil-Jona, Switzerland
mschuell@hsr.ch
2
ABB Ltd., Baden-Dättwil, Switzerland
3
NTNU – Norwegian University of Science and Technology,
Trondheim, Norway

Abstract. The barrier effect has been known since the 1930’s. It is of great
importance in the field of high voltage engineering as the breakdown voltage of
an electrode arrangement can be increased significantly when dielectric barriers
are applied. Nevertheless, an exact and general valid physical model explaining
the barrier effect in gaseous, liquid and solid insulation systems is still not
known. For gaseous insulation systems, the Marx-Roser model is widely
accepted for explaining the barrier effect. The Marx-Roser model explains the
barrier effect by a redistribution of the electric field in the gap. This redistri-
bution is due to the space-charge field in front the tip but also due to the surface
charge field formed on the barrier. This leads to a higher breakdown voltage of
the system. In a recent publication of the authors, it was shown that the surface
resistance of the barrier has a huge impact of the breakdown performance of the
barrier arrangement. The breakdown voltage of the system decreased when the
surface resistance of the barrier was decreased over a value of around 107 Ω.
To investigate this effect, the surface potential due to accumulated surface
charges on the barrier surface was measured for three different values of barrier
surface resistance. It was shown in the experiments that at lower values of
surface resistance, less surface charges are accumulated at the barrier surface.
This leads to a decrease of the field reduction effect between high voltage
electrode and barrier surface and might result in lower breakdown values of the
system.

Keywords: Barrier
Conductivity
Breakdown voltage
Surface charge

Space charge
Charge accumulation

1 Introduction

The effect of barrier systems on breakdown performance of electrode systems has been
known for a long time. First reports about the barrier effect have been published by
Marx and Roser when they studied air gap discharges in the 1930’s using AC, DC and

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1155–1165, 2020.
https://doi.org/10.1007/978-3-030-31676-1_108
1156 M. Schueller et al.

lightning impulse voltages (LI) [1, 2]. These early publications explain the barrier effect
in gases with the redistribution of the electric field in the gap [1–3].
In gaseous insulation systems, the model from Marx and Roser is still used today
by many authors to explain the barrier effect. It has also been used as explanation for
insulation systems consisting of liquid and solid dielectrics [4, 5]. At present, there are
various other hypotheses trying to explain the barrier effect in insulation systems.
Inhomogeneous polarization or electro-physical characteristics of the main and barrier
materials are suggested in literature [6].
According to [6], the Marx-Roser model can only explain the barrier effect in rather
short gas gaps. The theory does not apply for longer gas gaps, liquid dielectrics or solid
arrangements. This is explained by the drift velocity and free path length of the charge
carriers being too slow or small to form a significant surface charge layer on the barrier
within the by the Marx-Roser model required time frame [6]. However, in a recent
publication [7] it was shown that the saturation charge levels on dielectric barriers in air
can be reached in a few tens of nanoseconds of exposure to positive streamer channels
[7]. So according to [7] the Marx-Roser model can also be applied in longer gas gaps
like presented in this study.
According to [6], an exact and generally valid physical model explaining the barrier
effect is still not known. In a recent publication [8], the influence of the surface
conductivity of the barrier on the withstand voltage was investigated. Following the
qualitative explanation provided by the Marx-Roser model the value of the surface
conductivity of the barrier has a significant influence on the withstand voltage of the
whole system.
In [8] and [9], it was shown that the withstand voltage of a single barrier system
decreases when the conductivity of the barrier surface increases. A theory to explain
this measured effect was proposed in [9]. According to this theory, when a certain
conductivity of the barrier is exceeded, the charge at the barrier is distributed faster at
the surface due to the smaller resistance and the barrier behaves as a floating electrode.
A dielectric surface can only be charged up to the saturation level [10]. At satu-
ration the amount of accumulated surface charges is so large that the normal component
of the electric field is compensated and the surface charging stops. This saturation
charge level is strongly dependent on the surface conductivity of a dielectric barrier and
according to [1] and [2] may influence the breakdown performance of the system. This
effect has to be investigated in further detail.
To investigate this theory the surface charge on the barrier surface has to be
measured. In particular, the goal of experiments is to confirm if the accumulated surface
charge is reduced for decreased surface resistances, which may finally result in lower
field reduction at the electrode and a lower breakdown voltage.

2 Theoretical Background

In Fig. 1 results from [8] are presented. In this publication it was shown that with
decreasing surface resistance the breakdown voltage of the single barrier system
decreased as well.
These results can be explained with two different approaches.
 Charge Accumulation on Slightly Conductive Barrier Systems 1157

The Marx-Roser model explains the barrier effect with the creation of space charges
due to ionization near the electrode tip and accumulation of surface charge on the
barrier. In case of positive tip, fast electrons will move towards the positive tip. The
remaining slower positive space charges drift in direction of the field and accumulate at
the barrier surface. This might result in a reduction of the electric field and conse-
quently in an increase of the breakdown voltage [1, 2, 6]. Because less surface charge is
trapped at the barrier surface with the decreasing surface resistance less field due to the
surface charge is accumulated at the barrier surface. As a result, the overall electric field
in the gap between barrier and high voltage electrode gets higher with decreasing
surface resistance.

Fig. 1. 50% withstand voltage at positive lightning impulse (LI) stress depending on the surface
resistance of the barrier. The dotted lines represent the breakdown voltage of the 80 mm gap
without barrier. The solid line is for positive LI, data taken from [8]. Note: The experiment
shown here has been performed for the same electrode configuration as described in Sect. 3, but
with a barrier size of 525  700 mm.

Another approach to explain the results from [8] can be derived from a model
introduced in [10]. This model is based on streamer propagation stability field in
combination with a saturation charge model of the dielectric surface.
In general, the value of the accumulated surface charge on a dielectric barrier
depends on the discharge behavior around the electrode and of the surface conductivity
of the barrier. Both factors can have significant impact and make the problem very
complex. This complexity can be avoided when only the most extreme case of surface
charging is considered. The most extreme case is charging till saturation with a very
high (or infinite) surface resistance. In this saturation case, the amount of the surface
charge is so large that the normal component of the electric field pointing from the
barrier out in the air, EnAir, is zero. As a result no field line ends on the surface
anymore. So further charge accumulation at the barrier is not possible and the charging
1158 M. Schueller et al.

of the barrier surface stops. This model has been introduced in [10] and experimentally
confirmed in [11] and [12].
According to the stability field model [10] the lowest voltage in kV that enables a
breakdown of an air gap larger than 40 mm with inhomogeneous electric field can be
expressed as [10]:

Uws ¼ U0 þ d Est ð1Þ

where Uws is the withstand voltage, d is the distance between electrodes in mm. Est is
the internal field strength in kV/mm along the positive streamer behind its head, and
has the same value as the required external field for stable streamer propagation (sta-
bility field). A voltage of U0  20–30 kV is equivalent to the potential of the streamer
head needed to generate a breakdown. The value of Est is in the range of 0,54 kV/mm
for positive impulse [10].
Equation (1) can be applied not only for straight gaps between electrodes but also
for arrangements where the streamer propagates parallel to dielectric surfaces or has to
bypass dielectric barriers like it is present in this study [10]. In such cases, the distance
d in (1) represents the clearance between electrodes (the closest connection in air
between the rod tip and the grounded plate).
With this model it is possible to roughly calculate the high values of breakdown
voltage for high surface resistance cases in Fig. 1.

3 Methods

3.1 Experimental Setup

To measure the effect of accumulated surface charges on a barrier, a vertical single
barrier arrangement between a high voltage rod electrode and a grounded plate elec-
trode was used The dimensions used were similar as in [8], see Fig. 2. The grounded
plate electrode is made of copper and measures (height  width) 1100  1000 mm.
The barrier material and dimensions were a little different as used in [8]. As barrier, a
5 mm PMMA (“Plexi glass ®”) plate with dimensions 400  600 mm was used. The
high voltage rod electrode is made of alumina and has a radius of 7 mm. It is 260 mm
long and the tip is rounded with a radius of 3.5 mm. The tip of the high voltage
electrode was positioned in the middle of the barrier at a z position of 200 mm and a x
position of 300 mm. A 200 mm toroid was used as well to guard the connection of the
high voltage electrode.
As gap distance of 80 mm is chosen between the tip of the high voltage electrode
and the ground electrode. The barrier is positioned 20 mm in front of the high voltage
electrode.
The experiments presented in this paper were conducted with positive lightning
impulse voltage. An eight stage 800 kV, 40 kJ Marx generator was used to create
1.2/50 µs lightning impulses (LI). The earth current was monitored using a Pearson
6585 current monitor to ensure that there was corona inception at the high voltage
electrode during the lightning impulse.
 Charge Accumulation on Slightly Conductive Barrier Systems 1159

80mm

400mm

20mm
1100mm

Highvoltage electrode
with torroid

Barrier

Ground electrode

Fig. 2. Dimensions of experimental set up seen from the side with high voltage electrode, 5 mm
PMMA barrier and grounded copper plate.

To measure the surface potential of the PMMA barrier, a DC-Stable Electrostatic

Voltmeter 341B with a 3450 sensor from Trek was used. This is a high precision
electrostatic voltmeter using the principle of a vibrating Kelvin probe. Its principle is to
zero the electric field between itself and the measured surface which was placed 2 mm
away from the sensor by adjusting its potential [13, 14]. The sensor does not measure
surface charge in Coulomb. It measures the result that is due to the surface charges, the
electric potential in Volt caused by the charges at the surface. The sensor is mounted on
a robot stage which moves the sensor along the barrier in x and z direction to measure
the surface potential on the whole surface. After the LI impulse, the barrier and ground
plane were moved along a rail system and positioned 2 mm in front of the probe
tip. The whole surface potential on the barrier was then scanned using the robot stage
resulting in 2D maps of the surface potential distribution of the whole barrier.
For the experiments, the surface resistance of the barrier had to be varied. This was
done using a conductive graphite spray. During the measurement series, more and more
graphite spray was applied to the surface of the barrier decreasing the surface resis-
tance. The surface resistance of the barrier was measured with a Keithley source meter
2410 before every series of measurement. The source meter can provide up to
UDC = 1.1 kV and measures a minimal current of 1 pA. Thus, it was possible to
measure a surface resistance up to a value of R = 1015 Ω

3.2 Experimental Procedure

The measurements here were based on the saturation charge assumption [10], as
explained in Sect. 2. This was necessary as a complete breakdown of the barrier system
leads to an unreproducible part- or full discharge of the barrier as the conductive
plasma channel formed by the breakdown is a low resistance path to ground. Thus, a
1160 M. Schueller et al.

positive LI voltage of 80 kV was applied to the high voltage electrode and no

breakdown of the system occurred. The applied 80 kV was sufficient to ignite corona
discharges at the tip electrode and charge the barrier surface.

Fig. 3. Picture of the set up with surface charge measurement in the front and high voltage
electrode connected to the Marx Generator in the back.

After the charging of the barrier surface by corona discharges by the high voltage
electrode, the whole barrier and ground plate were moved on a rail system to the
surface potential measurement with the Trek electrostatic voltmeter. The experimental
setup can be seen in Fig. 3. The distance between the place where the high voltage LI
was applied and the place where the surface potential was measured was necessary to
protect the surface potential measurement system in case of a occurring flashover.

4 Results

In Fig. 4 a line plot of the surface potential, which is due to the accumulated surface
charges is shown for measurements and a simulation. The plot is shown in the middle
of the barrier at a z position of 200 mm over the whole 600 mm length of the barrier in
x direction.
The peak of the measured surface potential is located around the x position
300 mm.
The influence of the surface resistance can be seen in this plot. The maximal
accumulated surface potential decreases with decreasing surface resistance. 80 kV was
applied at the HV electrode in every measurement.
At the shown z position 200 mm the maximal accumulated surface potential of 14
920 V was measured when the barrier had its maximal surface resistance of 2
1012 Ω.
 Charge Accumulation on Slightly Conductive Barrier Systems 1161

This was the case without any graphite spray applied to the PMMA barrier surface. At a
surface resistance of 3
1011 Ω the maximal surface potential was 6210 V. At the lowest
measured surface resistance of 5
106 Ω, the accumulated surface potential was 60 V.
The last result might be affected by noise due to limited precision of the probe. Also the
60 V amplitude is not visible due to re scaling of the axis.

Fig. 4. Surface potential due to accumulated surface charges at the barrier along the middle of
the barrier. Shown are measurements with solid lines and a simulation with a dashed line.

The simulation was performed according to [10] based on a purely capacitive

model (with infinite surface resistance that approximately corresponds to the case
without graphite spray). The simulated saturated surface potential shows a maximal
value of 32 kV at the position of the high voltage electrode.
In Figs. 5, 6 and 7, the measurements results are shown along the whole surface of
the barrier for all three measured surface resistances of 2
1012, 3
1011 and 5
106 Ω in
form of color plots.

Fig. 5. Measured surface potential on whole barrier with a barrier surface resistance of
2 1012 Ω.
1162 M. Schueller et al.

Fig. 6. Measured surface potential on whole barrier with a barrier surface resistance of 3 1011 Ω

Fig. 7. Measured surface potential on whole barrier with a barrier surface resistance of 5 106 Ω.

For the measurement with the highest surface resistance, shown in Fig. 5, the
maximum surface potential is in the middle of the barrier directly under the high
voltage electrode and decays to zero till the edges of the barrier. The decay of the
surface potential is continuous from the middle till the edges.
For the measurement with 3
1011 Ω surface resistance this not the case anymore.
The surface potential still shows a maximum at the middle of the barrier where the rod
shaped high voltage electrode was located, but the decay to zero is much more abrupt
than the measurement with 2
1012 Ω showed.
At a surface resistance of 5
106 Ω the surface potential corresponds to the residual
charge remaining on the surface without clearly localized maximum. The pattern in
Fig. 7 indicates that the charging activity dominates in the vertical direction where the
distance to the edge of the barrier is shorter.
 Charge Accumulation on Slightly Conductive Barrier Systems 1163

5 Discussion

A significant influence of the surface resistance on the accumulated surface potential

has been clearly demonstrated in this work. At the highest value of surface resistance
the highest value of accumulated surface potential was measured. Following the Marx-
Roser model this would reduce the electric field between high voltage electrode and
barrier the most. Thus the highest breakdown voltage would be expected. This is
clearly the case like it can be seen in Fig. 1.
In Fig. 8, the results of 2D FEM simulations of the field distribution along the gap
are shown. In this simulation model the measured surface potential for a certain surface
resistance (shown in Fig. 4) was put as a boundary condition on the barrier surface. The
electric field strength in the 20 mm gap between barrier surface and high voltage
electrode is shown for the three measured surface resistances and the corresponding
surface potential. It can be clearly seen that at the case with the highest surface
resistance and thus the highest accumulated surface potential the electric field in the
gap is the smallest due to field reduction. So at 5
106 Ω surface resistance the electric
field strength at the high voltage electrode is increased from 7.7 kV/mm at 2
1012 Ω to
11.04 kV/mm, an increase of 43%, which is in both cases above the critical field of
2.6 kV/mm and above the inception level. So less surface charge on the barrier may
lead to an earlier breakdown due to less field reduction. However the field strength at
the totally insulating case (with the maximal surface potential) is above 2.6 kV/mm. In
this arrangement suppression of streamer inception due to the accumulated surface
potential cannot be confirmed. But still this effect cannot be totally excluded as
explanation for the behavior shown in [8].
In the measurements presented in this publication only 80 kV was applied at every
experiment. But in [8] much higher voltages up to 190 kV were applied. The saturation
charge at a barrier surface depends strongly on the applied voltage at the electrode. So
it possible that at 190 kV applied the surface potential of the barrier is much larger than
shown in these measurements at 80 kV and inception suppression might have hap-
pened at high surface resistance values due to a large potential accumulation and field
reduction. This has to be investigated in further experiments as it is questionable if a
much higher accumulated surface potential due to the higher voltage will be in such a
range to relax this much higher applied voltage.
In [9] and [8] it was also assumed that in addition to the possible effect of field
reduction the barrier with low surface resistance can act as a floating electrode at a
potential close to the high voltage electrode. This is due to a rather small voltage drop
along a streamer bridging the air gap between the rod tip and the barrier [11]. Based on
the stability field for the positive streamer, it was estimated that the potential of the
floating barrier is just 20 mm*0,5 kV/mm = 10 kV lower than the applied voltage
during the positive impulse [11].
Due to the high conductivity of the barrier this high potential will be transferred to
the barrier edge, where field enhancement occurs which creates many inception points
around the large and sharp barrier circumference which leads to the lower withstand
voltage. This might also be a valid explanation for the effect but this has to be
investigated in further detail to draw a conclusion which of the two effects is the main
1164 M. Schueller et al.

responsible for the decrease of the breakdown voltage at barrier systems with increased
surface conductivity.
According to the simulation procedure proposed in [12] the maximum surface
potential of around 32 kV has been calculated for the case with no graphite spray
applied at the barrier at 2
1012 Ω, see Fig. 4 This is around twice the value the
measurements showed. This difference could be due to charge decay. A rough esti-
mation including the surface resistance and surface capacitances of the setup has shown
that the time constant for the decay is below 1 min. This is about the time that is needed
to get the barrier ready for surface potential measurement after LI was applied. The
scan of the whole surface of the barrier also takes about 6 min.

Fig. 8. Electric field strength distribution along the gap between the barrier (length = 0) and the
tip of high voltage rod (length = 20 mm). Note: these curves have been calculated for different
surface resistances by assuming the corresponding measured surface potential as a boundary
condition along the barrier surface.

Still the calculations did show a qualitative agreement with the measurements. The
simulations showed field reduction at higher surface resistance due to the higher
amount of accumulated surface potential at the barrier but inception suppression is not
possible according to the calculations.
These effects have to be investigated and measured in much more detail to fully
understand the dependence of the withstand voltage of the barrier system depending on
the surface resistance on the barrier which is shown in Fig. 1.

6 Conclusions

According to the for gaseous insulation systems widely accepted Marx-Roser model
the surface charges at the surface of the barrier have a huge impact on the breakdown
voltage of the barrier system. In [9] it was shown that the breakdown voltage of a
barrier arrangement decreases severely when the surface resistance of the barrier is
decreased as well.
 Charge Accumulation on Slightly Conductive Barrier Systems 1165

To investigate this effect, the surface potential due to accumulated surface charges
on the barrier surface was measured for three different values of barrier surface
resistance. It was shown in the experiments that at lower values of surface resistance,
less surface charges are accumulated at the barrier surface. This leads to a decrease of
the field reduction effect between high voltage electrode and barrier surface and might
result in lower breakdown values of the system.
From the experiments presented in this publication we can conclude that less surface
charge accumulation due to increased surface conductivity takes place on the barrier
surface. This seems to have influence on the breakdown behavior of the barrier system
but this has to be investigated and quantified in further detail in future experiments.

References
1. Marx, E.: Der Durchschlag der Luft im unhomogenen elektrischen Felde bei verschiedenen
Spannungsarten. Electrotech. Z. H33, 1161–1165 (1930)
2. Roser, H.: Schirme zur Erhöhung der Durchschlagspannung in Luft. Electrotech. Z. H17,
411–412 (1932)
3. Voloschenko, N.F.: About the barrier effect mechanism. Russ. J. Electricity 3, 21–26 (1946)
4. Yoshino, K.: Electrical conduction and dielectric breakdown in liquid dielectrics. IEEE
Trans. Dielect. Electr. Insul. 21 (1986)
5. Leontev, Yu.N., Torbin, N.M.: Influence of the barrier position on the breakdown voltage of
solids. Trans. USSR Univ. Electr. Eng. (12), 34–37 (1961)
6. Lebedev, S.M., Gefle, O.S., Pokholkov, Y.P.: The barrier effect in dielectrics the role of
interfaces in the breakdown of inhomogeneous dielectrics. IEEE Trans. Dielectr. Electr.
Insul. 12(3), 537–555 (2011)
7. Meyer, H.K., Mauseth, F., Marskar, R., Pedersen, A., Blaszcyk, A.: Streamer and surface
charge dynamics in non-uniform air gaps with a dielectric barrier. IEEE Trans. Dielectr.
electr. Insul. 26(4), 1163–1171 (2019)
8. Schueller, M., Bucher, M.K., Franz, T., Smajic, J.: Effect of slightly conductive barrier
systems on the breakdown voltage of an air insulated rod plane arrangement shown in
experiment and simulation. In: International Symposium on High Voltage Engineering, ISH
2017, Argentina (2017)
9. Schueller, M., Blaszczyk, A., Krivda, A., Smajic, J.: Influence of the surface conductivity of
a single glass barrier on the breakdown voltage in an air insulated rod plane arrangement. In:
IEEE Conference Electrical Insulation Dielectrical Phenomena, Toronto (2016)
10. Pedersen, A., Blaszczyk, A.: An engineering approach to computational prediction of
breakdown in air with surface charging effects. IEEE Trans. Dielectr. Electr. Insul. 24(5),
2775–2783 (2017)
11. Pedersen, A., Christen, T., Blaszczyk, A., Böhme, H.: Streamer inception and propagation
models for designing air insulated power devices. In: IEEE Conference Electrical Insulation
Dielectrical Phenomena (CEIDP) (2009)
12. Meyer, H.K., Blaszczyky, A., Schueller, M., Mauseth, F., Pedersen, A.: Surface charging of
dielectric barriers in short rod-plane air gaps – experiments and simulations. In: IEEE
Conference on High Voltage Engineering and Application, ICHVE, Greece (2018)
13. Zismann, W.A.: A new method of measuring contact potential differences in metals. Rev.
Sci. Instrum. 3, 367–368 (1932)
14. Schueller, M., Gremaud, R., Baur, M., Hermann, L.G., Smajic, J.: Kelvin probe for surface
potential measurements on epoxy insulators for HVDC applications. In: IEEE Conference on
High Voltage Engineering and Application, ICHVE, Greece (2018)
Transient Voltages
 Current and Electric Field Characteristics
of 35 Return Strokes from Negative Lightning
Measured at Peissenberg Tower Germany

Christian Paul1(&) , Fridolin H. Heidler1 , and Wolfgang Schulz2

1
University of the Federal Armed Forces Munich, Chair for High Voltage
Technology and Lightning Research (EIT 7.1), Neubiberg, Germany
christian.paul@unibw.de
2
OVE Service GmbH, ALDIS, Vienna, Austria

Abstract. In this paper, we analyse the currents and electric fields of 35 neg-
ative return strokes, which have been measured since 2012 at Peissenberg
Tower, Germany. 27 were pure return strokes and 8 were return strokes with
superimposed symmetrical M-components. 2 out of this 8 were first return
strokes. The measured peak currents ranged from 3.1 kA to 40.8 kA, the
arithmetic mean value (AM) was 12.3 kA. Further we estimated the 10%-to-
90% rise time, which ranged from 1.0 µs to 7.4 µs, the AM was 1.9 µs. The
transferred charge varied from 0.1 C to 10.6 C, the AM was 1.0 C. The radiated
electric field was measured in a distance of about 180 m to the tower. The
electric field exhibits a first field change due to the descending leader. For the
description of this first field change we introduced ΔE1. The values of ΔE1
varied from 0.8 kV/m to 10 kV/m, the AM was 2.8 kV/m. The first field change
is immediately followed by a second field change of opposite polarity. We
introduced ΔE2 to describe this field change, which is caused by the return
stroke process. The values of ΔE2 varied from 1 kV/m to 14.2 kV/m, the AM
was 3.5 kV/m. All analysed return strokes were detected by the lightning
location system (LLS) EUCLID. The peak current inferred by EUCLID varied
between 3.9 kA and 53.0 kA, the AM was 15.0 kA. 10 out of 35 detected return
strokes were misclassified as cloud-to-cloud discharge.

Keywords: Lightning
Return stroke
Electric field
Peissenberg tower

Lightning location (LLS)
EUCLID

1 Introduction

A classical downward lightning contains one or more return strokes. Each of

these return strokes can be followed by a continuing current (CC), typical in the
range of several hundred of amperes [1–5]. The continuing current can be superim-
posed by various impulsive currents, mostly symmetrical current pulses, called M-
components [6]. M-components were first mentioned in the literature by Malan et al. in
the thirties of the last century. They found out that there is a temporary increase in
luminosity of the lightning channel after a return stroke [7]. Further, Kitagawa et al.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1169–1179, 2020.
https://doi.org/10.1007/978-3-030-31676-1_109
1170 C. Paul et al.

observed different brighter components, superimposed on the illumination caused by

the CC [8].
On the contrary, high rising structures with heights of more than 100 m are typi-
cally struck by upward lightning. Because of the building height, the electric field
strength at the top of the building is high enough to initiate a lightning strike.
Therefore, the upward lightning is triggered by the building itself [9–15]. The begin-
ning of a classical upward lightning is characterized by a slow-varying initial contin-
uous current (ICC), due to the leader moving upward from the top of the building. The
initial continuing current has a typical magnitude between some tens to some thousands
of amperes, the duration varies from some tens up to some hundreds of milliseconds
[16, 17]. Similar to the CC, the ICC can be superimposed by various impulsive currents
[6].
The electric field of the return strokes exhibits at close distance typically a V-shape
with a first field change, which is followed by a second field change of opposite
polarity. The first field change is caused by the approach of the downward leader. The
second field change is caused by the charge transfer to ground due to the return stroke
process [18]. The bottom of the V-shape is time-correlated with the onset of the current,
which indicates the transition from the leader stage to the return stroke stage [19].
Rakov et al. reported that return strokes with larger peak currents generate larger
changes in the electric field [20]. In addition it can be assumed that the second field
change is commonly higher than the first field change [21].
The record of the radiated electric and/or magnetic field with different stations
builds up the basis of a classical lightning location system (LLS). The LLS derives the
striking point of the return stroke generally by the use of two different techniques. On
the one hand, the striking point can be calculated from the time of arrival of the field at
different recording stations (TOA-method), and on the other hand, the direction of the
incident magnetic field at the different stations can be used to determine the striking
point (MDF-method). Modern LLSs make use of a combination of both methods
(IMPACT-method) [22]. Furthermore the LLS estimates a peak current (called inferred
peak current) for each detected return stroke, out of the recorded electric or magnetic
field [23, 24]. In Europe the detection of lightning activity is organized in EUCLID
(European Cooperation for Lightning Detection) since 2001. As of January 2018
EUCLID employs in total 165 sensors containing 34 analog sensors (IMPACT
ES/ESP) and 131 digital sensors (LS700X). The performance of the European LLS can
be evaluated by using the measurements at high buildings as ground truth data. There
are various projects like the Saentis Tower in Switzerland and the Gaisberg Tower in
Austria, which were used as references [25, 26]. In addition to that, different studies
showed, that there is a local field enhancement due to the presence of the high building,
which influences the inferred peak current of the LLS. This effect is called “tower
enhancement effect” and depends on the height of the structure as well as on the rise
time of the lightning current [27–30].
In this paper, we analyse 35 return strokes which were measured at Peissenberg
Tower and use them as a reference for the performance evaluation of the
 Current and Electric Field Characteristics of 35 Return Strokes 1171

EUCLID LLS. The main emphasis of this analysis is the quality of the inferred peak
current of the LLS determined from the recorded magnetic field.

2 Experiment and Data Overview

The mountain “Hoher Peissenberg”, about 940 m above mean sea level, is located
about 60 km west of Munich. On this mountain, there is located an about 150 m high
television broadcasting tower, called Peissenberg Tower. We instrumented the top of
this tower with a current probe and a di/dt-probe for measurement of the lightning
current and its time-derivative. In addition to that, we installed a field measuring station
in a distance of about 180 m to the tower. With this station we have measured the
radiated electric and magnetic field and their time derivatives. The tower measurement
system as well as the field measurement system is GPS time synchronized. The time
synchronization accuracy between the current and field record can be assumed as better
than 0.1 µs. The lightning current and the electric field is recorded with a measuring
device (NI PXI 5122) with a resolution of 14 bit with a sample rate of 100 MS/s (for
more details see [13, 14]). The current records were filtered numerically with a
350 kHz low pass filter (2nd order butterworth). For each return stroke, we determined
the peak current (Ip), the 10%-to-90% rise time (t10-90%) and the transferred charge (Q).
The current duration is characterized by the full width at half maximum (FWHM). An
impulsive lightning current is classified as return stroke as soon as its 10%-to-90% rise
time is smaller than 2 µs and its peak current Ip is greater than 2 kA, because the
smallest peak current of a return stroke that can exist in nature is estimated to be 2 kA
[31].
Table 1 gives an overview of the analysed return strokes, measured between Jan-
uary 2012 and August 2017. The peak current (Ip) ranged from 3.1 kA to 40.8 kA.
The AM was 12.3 kA (GM: 10.1 kA). The 10%-to-90% rise time (t10–90%) ranged
from 1.0 µs to 7.4 µs. The AM was 1.9 µs (GM: 1.7 µs). The current duration
(FWHM) varied between 6.3 µs and 79 µs. The AM was 39.5 µs (GM: 33.5 µs). The
transferred charge (Q) ranged from 0.1 C up to 10.6 C. The AM was 1.0 C (GM:
0.6 C).

Table 1. Overall values for the analyzed 35 return stroke currents. (*The minimum rise time
may be increased due to the numerical filtering with 350 kHz. AM: arithmetic mean value GM:
geometric mean value)
Ip [kA] t10–90% [µs] FWHM [µs] Q [C]
Min. 3.1 1.0* 6.3 0.1
Max. 40.8 7.4 79.0 10.6
AM 12.3 1.9 39.5 1.0
GM 10.1 1.7 33.5 0.6
1172 C. Paul et al.

0.9

70 140 210 t (μs) 280

(kV/m)

- 0.9
HW
∆E1 10% to 90% ∆E2
E(t)

- 1.8
rise time
- 2.7 ∙ ∆E1

- 3.6

Fig. 1. Typical waveform of a radiated electric field showing the used field parameters.

Figure 1 shows a typical waveform of an electric field pulse caused by a return

stroke. The waveform exhibits a first field change, which is described by ΔE1, followed
by a second field change of opposite polarity. The second field change is described by
ΔE2. The fast rise of the electric field is characterized by the 10%-to-90% rise time
(t10–90%,E) shown in Fig. 1. The duration of the electric field pulse is characterized by
the half width (HW).

Table 2. Overall values for the analyzed 35 return stroke electric field records.
ΔE1 [kV/m] ΔE2 [kV/m] HW [µs] t10–90%,E [µs]
Min. 0.8 1.0 2.3 0.4
Max. 10.1 14.2 142.3 5.7
AM 2.8 3.5 45.6 1.0
GM 2.3 2.8 35.6 0.9

Table 2 shows the overall values of the electric field records. The first field change
(ΔE1) varied between 0.8 kV/m up to 10.1 kV/m. The AM was 2.8 kV/m (GM:
2.3 kV/m). The second field change (ΔE2) ranged between 1.0 kV/m and 14.2 kV/m.
The AM was 3.5 kV/m (GM: 2.8 kV/m). The duration of the electric field pulse varied
from 2.3 µs up to 142.3 µs. The AM was 45.6 µs (GM: 35.6 µs). The 10%-to-90% rise
time ranged from 0.4 µs up to 5.7 µs. The AM was 1.0 µs (GM: 0.9 µs).
Figure 2 shows the cumulative frequency distribution (probability) of ΔE1. The
values fit the logarithmic normal distribution sufficiently. 95% of all data points had a
value of 0.8 kV/m or higher. The highest value of ΔE1 was 10.1 kV/m.
Figure 3 shows the cumulative frequency distribution (probability) of ΔE2. Similar
to the values of ΔE1, the values of ΔE2 fit the logarithmic normal distribution suffi-
ciently. 95% of all data points had a value of 1.0 kV/m or higher. The highest value of
ΔE2 was 14.2 kV/m.
 Current and Electric Field Characteristics of 35 Return Strokes 1173

99.9
99.7
99
98
95

Cumulative frequency [%]

90

(Probability) 75

50

25

10
5
2
1
0.3
0.1
0.1 1 10
∆E1 [kV/m]

Fig. 2. Cumulative frequency distribution of ΔE1 of the 35 analyzed electric field records.

99.9
99.7
99
98
95
Cumulative frequency [%]

90

75
(Probability)

50

25

10
5
2
1
0.3
0.1
0.1 1 10
∆E2 [kV/m]

Fig. 3. Cumulative frequency distribution of ΔE2 of the 35 analyzed electric field records.

3 Electric Field of Return Strokes

Figure 4 shows the characteristic waveform of the current (a) and the time-
synchronized electric field (b) of a return stroke (B343). For all 35 negative return
strokes, the electric field shows this characteristic “V-shape”, which is based on two
field changes of opposite polarity. The first one is due to the descending leader, this
phase is called leader mode. The second one with opposite polarity is related to the
1174 C. Paul et al.

beginning of the current flow in the lightning channel, when the return stroke front
develops upwards, in opposite direction of the preceding descending leader. This phase
is called return stroke mode. The beginning of the current flow was always time-
correlated with the bottom of the “V-shape” of the electric field. Thus, the bottom of the
“V-shape” marks exactly the time, when the transition from the leader mode to the
return stroke mode occurs.

100 400 1000

0
700 t (μs)

-4

(kA)
-8

i(t)
a) - 12

- 16

100 400 1000 - 20

0
700 t (μs)
(kV/m)

-2

b)
E(t)

-4

-6

Fig. 4. Characteristic current waveform (a) and corresponding time-synchronized electric field
(b) of a return stroke (B343), measured on 20th January 2012.

The comparison of the lightning current and the electric field revealed that the rise
of the electric field is much faster compared to the current rise. The 10%-to-90% rise
time of the electric field t10–90%,E is almost half of the 10%–90% rise time of the current
t10–90%. On contrary, the duration of the current pulse (FWHM) is nearly the same as of
the duration of the electric field pulse (FWHM), with the very small deviation of 6% for
the GM.
For the analysis of the first (ΔE1) and the second (ΔE2) field change we introduced
a field ratio factor f according to Eq. (1):

DE2 =DE1 ¼ f ð1Þ

The AM as well as the GM of the field ratio factor f is 1.2. This means, that the
second field change (ΔE2) is typically 20% higher compared to the first field change
(ΔE1) (see Table 2).
Figure 5 shows the correlation between the electric field changes ΔE1 and ΔE2 and
the measured peak current Ip. 33 out of 35 were subsequent return strokes (SRS) and 2
 Current and Electric Field Characteristics of 35 Return Strokes 1175

out of 35 were first return strokes (FRS). The field changes follow approximately the
linear regression according to Eqs. (2a) and (2b). The correlation coefficient is 0.96 for
the correlation between ΔE1 and Ip and 0.94 for the correlation between ΔE2 and Ip.

0:23
Ip ½kA ¼ DE1 ½kV=m ð2aÞ

0:3
Ip ½kA ¼ DE2 ½kV=m ð2bÞ

16

14

12

∆E2(Ip) = 0.3 ∙ Ip
10
∆E [kV/m]

6 ∆E1(Ip) = 0.23 ∙ Ip

4 SRS ∆E1
FRS ∆E1
SRS ∆E2
2 FRS ∆E2

0
0 10 20 30 40 50
Ip [kA]

Fig. 5. Correlation between the electric field changes ΔE1 and ΔE2 and the current peak value Ip
for the analyzed 35 return strokes.

4 Performance of the LLS EUCLID

All 35 analyzed return strokes were detected by the LLS EUCLID. For each return stroke,
the LLS inferred a peak current Ip,LLS from the recorded field. The values of the inferred
peak current varied between 3.9 kA and 53.0 kA. The AM was 15.0 kA (GM: 12.5 kA).
Figure 6 shows the correlation between the measured peak current Ip at Peissenberg
Tower and the inferred peak current Ip,LLS by the LLS EUCLID. The correlation
coefficient is 0.92. Due to the good correlation, the data points are quite close to the
regression line according to Eq. 3:

1:2
Ip ½kA ¼ Ip;LLS ½kA ð3Þ

Due to Eq. (3), the inferred peak current of the LLS is overestimated by about 20%.
The measured peak current Ip has a GM of 10.1 kA and the inferred peak current Ip,LLS
has a GM of 12.5 kA. It is likely that the difference is due to “tower enhancement effect”.
1176 C. Paul et al.

60

50

Ip,LLS = 1.2 ∙ Ip
40
Ip,LLS [kA]

30

20

10 SRS
FRS

0
0 10 20 30 40 50
Ip [kA]

Fig. 6. Correlation between the measured absolute peak current Ip and the inferred absolute
peak current Ip,LLS by the LLS.

3500,0

3000,0

2500,0
SRS
FRS
Location error [m]

2000,0

1500,0

1000,0

500,0

0,0
0 10 20 30 40 50
Number of sensors

Fig. 7. Correlation between the number of sensors and the location error between the striking
point inferred by the LLS and the Peissenberg Tower.

Figure 7 shows the correlation between the number of sensors, which contributed
to the localization of the return strokes, and the location error of the striking point
(inferred by EUCLID). It can be seen, that at minimum 4 sensors are necessary to
 Current and Electric Field Characteristics of 35 Return Strokes 1177

achieve a location error less than 500 m (from the Peissenberg Tower). There is no
significant difference between the location error of first (FRS) and subsequent
(SRS) return strokes.
Figure 8 shows the correlation between the number of sensors, which contributed
to the localization of the return strokes, and the relative current deviation ΔI. The
current deviation ΔI is given by the difference between the peak current Ip measured at
the Peissenberg Tower and the peak current Ip,LLS inferred by EUCLID. The correlation
coefficient is 0.1, which means that there is almost no significant correlation between
these two parameters. Thus it can be assumed, that the relative deviation between
measured and inferred peak current does not depend on the number of sensors, which
contribute to the detection of the lightning event.

60

50

SRS
40 FRS
∆I [%]

30

20

10

0
0 10 20 30 40 50
Number of sensors

Fig. 8. Correlation between the number of sensors and the relative deviation ΔI between
measured (Ip) and inferred (Ip,LLS) peak current.

5 Conclusion

In this paper we give an overview of the current and electric field waveforms which
were caused by a return stroke. The electric field waveform of a return stroke showed a
characteristically “V-shape”, where the bottom of the “V” is time- correlated to the
beginning of the current flow. Further we figured out, that on average the first field
change ΔE1 due to the descending leader is 20% smaller compared to the second field
change ΔE2 due to the return stroke process. The LLS EUCLID detected all return
strokes to the Peissenberg Tower. The electric field was used (by EUCLID) to localize
the strike point and to evaluate the peak current. EUCLID overestimated the peak
current of return strokes by about 20%. This value was independent from the number of
1178 C. Paul et al.

sensors which contributed to the localization. At minimum four sensors were required
to localize the strike point with an error less than 500 m.

Acknowledgement. We thank EUCLID for providing the LLS data for this study.

References
1. Berger, K.: Novel observations on lightning discharges: results of research on Mount San
Salvatore. J. Franklin Inst. 283(6), 478–525 (1967)
2. Rakov, V.A., Uman, M.A.: Lightning: Physics and Effects. Cambridge University Press,
New York (2003)
3. Shindo, T., Uman, M.A.: Continuing current in negative cloud-to-ground lightning.
J. Geophys. Res. 94(D4), 5189–5198 (1989)
4. Rakov, V.A., Uman, M.A.: Long continuing current in negative lightning ground flashes.
J. Geophys. Res. 95(D5), 5455–5470 (1990)
5. Cooray, V.: Lightning electromagnetics. The Institution of Engineering and Technology,
London (2012)
6. Paul, C., Heidler, F.H.: Properties of three types of M-components and ICC-pulses from
currents of negative upward lightning measured at the Peissenberg Tower. IEEE Trans. EMC
59(60), 1825–1832 (2018). https://doi.org/10.1109/temc.2018.2802720
7. Malan, D.J., Collens, H.: Progressive lightning III - the fine structure of return lightning
stroke. Proc. Roy. Soc. London Ser. A Math. Phys. Sci. 162(909), 175–203 (1937)
8. Kitagawa, N., Brook, M., Workman, E.J.: Continuing currents in cloud-to-ground lightning
discharges. J. Geophys. Res. 67(2), 637–647 (1962)
9. Wang, D., Takagi, N., Watanabe, T., Sakurano, H., Hashimoto, M.: Observed characteristics
of upward leaders that are initiated from a windmill and its lightning protection tower.
Geophys. Res. Lett. 35, L02803 (2008). https://doi.org/10.1029/2007gl032136
10. Wang, D., Takagi, N.: Characteristics of winter lightning that occurred on a windmill and its
lightning protection tower in Japan. IEEJ Trans. Power Energy 132(6), 568–572 (2012).
https://doi.org/10.1541/ieejpes.132.568
11. Miki, M., Rakov, V.A., Shindo, T., Diendorfer, G., Mair, M., Heidler, F., Zischank, W.,
Uman, M.A., Thottappillil, R., Wang, D.: Initial stage in lightning from tall objects and in
rocket-triggered lightning. J. Geophys. Res. 110(D2) (2005)
12. Miki, M., Miki, T., Wada, A., Asakawa, A., Asuka, Y., Honjo, N.: Observations of lightning
flashes to wind turbines. In: 30th International Conference on Lightning Protection ICLP,
Cagliari, Italy (2010)
13. Heidler, F., Manhardt, M., Stimper, K.: The slow-varying electric field of negative upward
lightning initiated by the Peissenberg Tower, Germany. IEEE Trans. EMC 55(2), 353–361
(2013)
14. Heidler, F., Manhardt, M., Stimper, K.: Characteristics of upward positive lightning initiated
from the Peissenberg Tower, Germany. IEEE Trans. EMC 57(1), 102–111 (2015)
15. Romero, C., Paolone, M., Rubinstein, M., Rachidi, F., Pavello, D., Giri, D.: A statistical
analysis on the risetime of lightning current pulses in negative upward flashes measured at
the Saentis tower. In: Proceedings of the 31st Conference on Lightning Protection (ICLP),
Vienna, Austria, paper 114, September 2012
16. Flache, D., Rakov, V.A., Heidler, F., Zischank, W., Thottappillil, R.: Initial-stage pulses in
upward lightning: leader/return stroke versus M-component mode of charge transfer to
ground. Geophys. Res. Lett. 35, L13812 (2008). https://doi.org/10.1029/2008gl034148
 Current and Electric Field Characteristics of 35 Return Strokes 1179

17. Wang, D., Rakov, V.A., Uman, M.A., Fernandez, M.J., Rambo, K.J., Schnetzer, G.H.,
Fisher, R.J.: Characterization of the initial stage of negative rocket-triggered lightning.
J. Geophys. Res. 104(D4), 4213–4222 (1999)
18. Heidler, F.H., Paul, C.: Some return stroke characteristics of negative lightning flashes
recorded at the Peissenberg Tower. IEEE Trans. EMC 59(5), 1490–1497 (2017). https://doi.
org/10.1109/temc.2017.2688587
19. Rubenstein, M., Rachidi, F., Uman, M.A., Thottappillil, R., Rakov, V.A., Nucci, C.A.:
Characterization of vertical electric fields 500 m and 30 m from triggered lightning.
J. Geophys. Res. 100(D5), 8863–8872 (1995)
20. Rakov, V.A., Kodali, V., Crawford, D.E., Schoene, J., Uman, M.A., Rambo, K.J., Schnetzer,
G.H.: Close electric field signatures of dart leader/return stroke sequences in rocket-triggered
lightning showing fields. J. Geophys. Res. 110(D07205) (2005). https://doi.org/10.1029/
2004jd005417
21. Mosaddeghi, A., Shoory, A., Rachidi, F., Diendorfer, G., Pichler, H., Pavanello, D.,
Rubinstein, M., Zweiacker, P., Nyffeler, M.: Lightning electromagnetic fields at very close
distances associated with lightning strikes to the Gaisberg tower. J. Geophys. Res. 115
(D17101) (2010). https://doi.org/10.1029/2009jd013754
22. Cummins, K.L., Murphy, M.J., Bardo, E.A., Pifer, A.E.: A combined TOA/MDF technology
upgrade of the U.S. National Lightning Detection Network. J. Geophys. Res. Atmos. 103
(D8), 9035–9044 (1998). https://doi.org/10.1029/98jd00153
23. Cummins, K.L., Krider, E.P., Malone, M.D.: The U.S. National Lightning Detection
Network and applications of cloud-to-ground lightning by electric power utilities. IEEE
Trans. on EMC 40(4), 465–480 (1998). https://doi.org/10.1109/15.736207
24. Cummins, K.L., Burnett, R.O., Hiscox, W.L., Pifer, A.E.: Line reliability and fault analysis
using the National Lightning Detection Network. In: Precise Measurements in Power
Conference, Arlington, VA, USA (1993)
25. Azadifar, M., Rachidi, F., Rubinstein, M., Paolone, M., Diendorfer, G., Pichler, H., Schulz,
W., Pavanello, D., Romero, C.: Evaluation of the performance characteristics of the
European Lightning Detection Network EUCLID in the alps region for upward negative
flashes using direct measurements at the instrumented Säntis Tower. J. Geophys. Res.
Atmos. 121, 595–606 (2016)
26. Diendorfer, G., Pichler, H., Schulz, W.: LLS detection of upward initiated lightning flashes.
In: Asia-Pacific International Conference on Lightning (APL), Nagoya (2015)
27. Baba, Y., Rakov, V.A.: Lightning electromagnetic environment in the presence of a tall
grounded strike object. J. Geophys. Res. 110(D9) (2005). https://doi.org/10.1029/
2004jd005505
28. Bermudez, J.L., Rachidi, F., Rubinstein, M., Janischewski, W., Shostac, V., Pavanello, D.,
Chang, J.S., Hussein, A.M., Nucci, C.A., Paolone, M.: Far field-current relationship for
lightning return strokes to elevated strike objects. IEEE Trans. EMC 47(1), 146–159 (2005)
29. Rachidi, F., Janischewsky, W., Hussein, A.M., Nucci, C.A., Guerrieri, S., Kordi, B., Chang,
J.S.: Current and electromagnetic field associated with lightning-return strokes to tall towers.
IEEE Trans. EMC 43(3), 356–367 (2001)
30. Heidler, F., Schulz, W.: Lightning current measurements compared to data from the
lightning location system BLIDS. In: CIGRE International Colloquium on Lightning and
Power Systems, Bologna, Italy (2016)
31. Cooray, V.: On the upper and lower limit of peak current in first return strokes of lightning
flashes. In: Proceedings 30th International Conference on Lightning Protection (ICLP),
paper 1291, Cagliari, Italy, September 2010
 Improvement of Overhead Transmission Lines
Lightning Protection by Line Arresters
with Separate Groundings and Shielding Wires
Fixed at Insulation Racks

G. V. Podporkin(&)

Streamer, 27 Gzhatskaya, St. Petersburg, Russia

georgij.podporkin@streamer.ru

Abstract. This paper sets out a proposal aimed at increasing efficiency of

overhead lines lightning protection by using lightning arresters only on one
phase (e.g. A) and stand-alone grounding. If it is not metal-connected to the pole
it is connected to the pole via earth. Due to the current flow in the earth around
the grounding there is distribution of potential, which decreases with offset from
the grounding. Conducting pole located at a distance from the grounding gets a
potential. Thus, voltage applied to phase B insulator decreases and the proba-
bility of its flashover is reduced. However, there is a possibility of phase A in-
sulator flashover. By application grounding at optimal distance from the pole
lightning protection level can be twice greater than at traditional pole grounding
with the same value of grounding resistance. Efficiency of application of insu-
lating racks for fixing shielding wires connected to standing alone groundings is
also shown. Use of standing alone groundings for connection of lightning
arresters on the power lines, as well as use of insulation racks for shielding wire
with grounding, located at an optimal distance from the poles, appear to be
effective technical solutions deserving further elaboration with participation of
scientific, design and operational organizations.

Keywords: Overhead lines
Lightning protection
Grounding

1 Introduction

In spite of multi-year efforts made by scientific and engineering organizations the issue
of ensuring the required overhead lines (OHL) lightning protection remains urgent
especially for regions with high soil resistivity, for example, in solid rocks or ever-
frozen grounds. It makes impossible to ensure a required low value of grounding
resistance, so the protection by shielding wire reveals to be inefficient due to the fact
that at lightning surge current flowing through high tower footing resistance its
potential increases considerably and the back flashover occurs.
For OHL protection application of lightning arresters (LA) of different types are
being worked out. Use of LA at OHL is expensive measure as for protection efficiency
LA should be installed in parallel to each line insulator, i.e. 3 pieces per pole. It is less

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1180–1191, 2020.
https://doi.org/10.1007/978-3-030-31676-1_110
 Improvement of Overhead Transmission Lines Lightning Protection 1181

efficient, but economically more justified, to install LA only on one or two phases of
the line, but there is a requirement to provide a low value of grounding resistance.
Australian specialists suggested and developed lightning protection system with
insulation rack connected to separate groundings [1]. Commercial application of the
system was successful. Author of the present paper was interested in this technology
and suggested to use stand alone groundings for arresters as well [2]. Rather recently it
was a publication in Internet about successful application of insulated shielding wire
and standalone grounding in Indonesia. The authors stated that this system is about two
time less expensive than traditional one [3]. Therefore it is interesting to study influence
of different factors (such as position of grounding, type of soil etc.) on effectiveness of
separate groundings and find optimal solution.
This paper sets out a proposal aimed at increasing efficiency of OHL lightninged
protection by using LA only on one phase and separate groundings. The level of
lightning protection may be doubled due to the use of standalone ground electrode at
optimal distance from the tower as compared to the case of using the electrode galvanic
connected to tower body. Efficiency of application of insulating racks for shielding
wires connected to separate groundings is also shown.

2 Standalone Grounding

2.1 Principle of Operation

Figure 1 shows the scheme of OHL lightning-surge protection by means of arrester P
connected to standalone grounding electrode.
When lightning strikes top phase A the lightning-surge current flows through
arrester and further on through electrode Rg resistance into ground. Voltages applied
respectively to insulators of phases A and B are respectively equal to:

Uins:A ¼ UA  Up ð1Þ

Uins:B ¼ UB  Up ð2Þ

where: UA, UB and Up – potentials of phase A and B conductors and tower pole
respectively.
The potential of phase A is calculated upon equivalent circuit shown on Fig. 2,
where zch – lightning channel surge resistance; z – surge resistance of line conductor,
x – inductance of lead connecting arrester to ground electrode; Ra – arrester resistance;
Im – maximum value of lightning current (further – “lightning current”). The lead
connecting arrester to ground is substituted by inductance x and ground electrode - by
resistance Rg.
Neglecting on a first approximation the inductance of lead, once the arrester is
operated (Ra = 0), the voltage on phase A conductor may be defined upon expression:
1182 G. V. Podporkin

Up

UA
Ca

UB
C A B

1
LA

Ug 2 Up

Cg

Fig. 1. OHL lightning protection diagram with stand-alone grounding: 1 – pole; 2 – stand-alone
grounding; ca – coupling factor between conductors of phases A and B by air; cg – coupling factor
between grounding and pole by ground; Up – pole potential; Ug – potential of grounded
electrode.

zch

e
z/2 Ra
Im*zch

Rg

Fig. 2. Equivalent circuit to calculate the potential of phase A at its shielding failure.
 Improvement of Overhead Transmission Lines Lightning Protection 1183

Im zch ð2z ==Rg Þ

UA ¼ ¼ Im zeqv ð3Þ
zch þ ð2z ==Rg Þ

z ð z ==R Þ
where: zeqv ¼ zchchþ2ðz==Rg g Þ - equivalent resistance; //- parallel addition sign of resistances.
2
Due to current spreading into ground there is a potential distribution created around
ground electrode, which decreases when moving from it. The conductive tower pole
being spaced some distance apart from ground electrode acquires the potential Up,
which may be determined upon expression:

Up ¼ cg Ug  cg UA ð4Þ

where: Ug – ground electrode potential; UA  Ug – phase A potential; cg – ground

coupling factor between ground electrode and pole (see Fig. 1).
The voltage applied to phase A insulator may be determined from (1) and (4):


Uins:A ¼ UA  Up ¼ UA  cg UA ¼ UA 1  cg : ð5Þ

The phase B potential can be defined through that one of phase A by means of air
coupling factor ca between conductors (see Fig. 1):

UB ¼ ca UA : ð6Þ

So, the voltage on phase B insulator may be defined from (2) and (6) respectively:


Uins:B ¼ UB  Up ¼ ca UA  cg UA ¼ UA ca  cg : ð7Þ

From (7) one may see that at ground coupling factor between tower and ground
electrode equal to air coupling factor between conductors ca = cg the voltage on phase
B insulator becomes equal to 0. At that, the voltage on phase A insulator is calculated
upon (5):


Uins:A ¼ UA 1  cg ¼ UA ð1  ca Þ: ð8Þ

It is reasonable to determine conditions under which voltages on phase A and

B insulators are equal. When making (5) equal to (7) we have:



UA 1  cg ¼ UA ca  cg ð9Þ

from where:

cg ¼ ð1 þ ca Þ=2 ð10Þ

and from (5) and taking into account (9) and (10) we may have the expression for
voltages on phase A and B insulators:
1184 G. V. Podporkin

Uins:A ¼ Uins:B ¼ UA ð1  ð1 þ ca Þ=2Þ ¼ UA ð1  ca Þ=2: ð11Þ

Comparing (8) and (11) we can see that for the latter the voltage on phase A in-
sulator is evenly twice as little. In other words, if the ground coupling factor is taken
from (10), i.e. in a way that the voltage on insulators of two phases A and B is equal, it
becomes evenly twice as little as that one on phase A insulator in case when ca = cg,
that ensures the absence of voltage on phase B.
Figure 3 gives relationships between voltages on insulators of phases A and B as a
function of ground coupling factor cg, they are calculated upon (5) and (7). The results
are in relative units (with relation to UA).
As we can see on Fig. 3 at cg = 0 (that corresponds to a great distance between
ground electrode and tower) the voltage on phase A insulator is maximum and is equal
to 1 in relative units, i.e. it is equal to phase A voltage. At cg increase the voltage Uins.A
is linearly decreasing and at cg = 1 (that corresponds to traditional galvanic connection
of arrester to grounded tower) it is equal to zero.

1
Voltage on insulator’s phases A and B, p.u.

0.8
Uins.A
|Uins.B|
0.6

0.4

0.2

0
0.2 0.4 0.6 0.8 1
-0.2 сg.opt
-0.4 сa = сg
Uins.B
-0.6

-0.8
Ground coupling factor, сg

Fig. 3. Relationships between voltages on phase A and B insulators as a function of ground

coupling factor at ca = 0.4.

At cg increase the voltage on phase B insulator changes in the following way. First,
UИЗ.B is decreased, at cg = ca it is equal to 0, while at its further increase it changes its
polarity and is increased again in magnitude. At cg = 1 Uins.B reaches the value of
(1 − ca) = 1 − 0.4 = 0.6.
From Fig. 3 it is shown that minimum values of voltage on insulators can be got
simultaneously at cg = cg.opt, to be defined upon (10).
It is interesting to compare the OHL lightning-surge protection at traditional mode
of tower grounding (cg = 1) and at tower grounding through stand-alone ground
electrode when the ground coupling factor is optimal cg = cg.opt.
 Improvement of Overhead Transmission Lines Lightning Protection 1185

Let us assume that if an arrester has been discharged and no one insulator has been
flashovered then the lightning protection was successful. At traditional mode of tower
grounding (cg = 1) the lightning-surge withstanding capability is defined by phase
B insulator flashover. The voltage of phase B insulator with account made for (7) and
(3) shall be calculated by the following expression:

Uins:B ¼ UA ðca  1Þ ¼ Im zeqv :ðca  1Þ ð12Þ

or for absolute voltage values:

jUins:B j ¼ Im zeqv :ð1  ca Þ: ð13Þ

ð1Þ
The level of lightning protection Im shall be determined at 50% discharge test
voltage of insulator U50%:

U50%
Imð1Þ ¼ : ð14Þ
zeqv ð1  ca Þ

In case when the tower is grounded through a stand-alone ground electrode and the
ground coupling factor is optimal cg = cg.opt the voltage on insulators of both phases
A and B is the same and shall be defined upon (11). Bearing in mind (7) the level of
ð2Þ
lightning protection in such a case Im may be defined by expression:

U50% U50%
Imð2Þ ¼ ¼2 ¼ 2Imð1Þ : ð15Þ
zeqv ð1  ca Þ=2 zeqv ð1  ca Þ

ð2Þ ð1Þ
By comparing (14) and (15) one can see that Im ¼ 2Im does not depend on U50%
and z. Thus, the level of OHL lightning-surge protection having a stand-alone ground
electrode at optimal ground coupling factor is twice as much as at traditional tower
grounding with the same value of grounding resistance. Besides, this relationship does
not depend on insulators breakdown voltage and on the value of equivalent resistance
and hence, on ground electrode resistance Rg.
One more explanation may be given to clarify the proposed solution. In case of
traditional galvanic connection of ground electrode to tower all “reverse overvoltage” is
found to be applied to one insulator of phase B, while in case of a standalone ground
electrode with optimal ground coupling factor the same “reverse voltage” is equally
distributed between insulators of phases A and B. Thus, the level of OHL lightning
protection is doubled.
The OHL tripping rate is proportional to the probability of lightning stroke current
occurrence equal or exceeding the level of OHL lightning protection Im. Such a
probability may be estimated upon expression [4]:

PIm ¼ e0:04Im : ð16Þ

1186 G. V. Podporkin

Hence, the ratio of OHL tripping rate with traditional tower grounding n(1) to that
one of line with standalone ground electrode n(2) may be expressed in such a way:
ð1Þ ð1Þ
e0:04Im e0:04Im ð1Þ
k¼ ð2Þ
¼ ð1Þ
¼ e0:04Im : ð17Þ
e 0:04Im e 0:042Im

Figure 4 exhibits the dependence of the ratio k of tripping rate of line with tradi-
tional tower grounding n(1) to the rate of line tripping with standalone ground electrode
n(2). As it is presented in (17) and on Fig. 4 k greatly depends on the lightning
ð1Þ ð1Þ
protection level of studied line Im . The more is Im , the more efficient becomes the use
of standalone ground electrodes.

60

50

40
Efficiency, k

30

20

10

0
0 20 40 60 80 100
Lightning protection level Im, kA
Fig. 4. Dependence of the efficiency of the use of standalone ground electrodes k on the
lightning protection level Im.

In particular, for 35 kV OHL 50% breakdown voltage of insulator is approximately

400 kV. At tower grounding resistance Rg = 10 X and coupling factor ca = 0.4 the
lightning protection level calculated upon (15) shall make:

U50% 400
Imð1Þ ¼  ¼ 67 kA: ð18Þ
zeqv ð1  ca Þ 10ð1  0:4Þ

Hence, the efficiency k, defined as per (17) shall be equal to:

 Improvement of Overhead Transmission Lines Lightning Protection 1187

ð1Þ
k ¼ e0:04Im ¼ e0:0467 ¼ 14:6  15:

In this context due to the use of standalone ground electrode the tripping rate of
35 kV OHL may be approximately reduced by 15 times as compared to traditional
mode of tower grounding.

2.2 Location of Ground Electrode in Relation to Tower

For the sake of simplicity, let us consider the case when the conductor length l is equal
to the length of buried part of pole. The grounding resistance of electrode Rg = R1 and
mutual resistance between ground electrode and tower R12 can be calculated by
expressions [5, 6]:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
q 2l r 2 ffi r
sp sp
R1 ¼ Arsh  þ1þ ð19Þ
2pl rsp 2l 2l
0 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
s 1
2 ffi
q @ 2l D D
R12 ¼ Arsh  þ1þ A ð20Þ
2pl D 2l 2l

where:
q – specific soil resistivity;
D – distance between ground electrode and pole axes;
rsp – ground electrode radius with account made for spark formation in ground
Hence, the ground coupling factor cg may be determined by expression:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

D 2
D
Arsh 2l 2l þ 1 þ 2l
D
R12
cg ¼ ¼ q
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð21Þ
R3 rsp 2 r
Arsh r2lsp  2l þ 1 þ 2lsp

With due account made for spark formation in ground the electrode radius is deter-
mined upon expression [6]:

Im q
rsp ¼ ð22Þ
2pEbr l

where: Ebr – soil breakdown strength.

From (22), one can see that rsp depends on lightning stroke current Im and on soil
breakdown strength Ebr. According to soil type and Im value of Ebr is within 6–
12 kV/m. For example, at lightning stroke current Im = 40 kA and Ebr = 6 kV/m (that
corresponds to sandy soil at q = 500 X m) at l = 3 m the radius of ground electrode is
equal to:
1188 G. V. Podporkin

Im q 40 kA
500 X
m
rsp ¼ ¼  1:8  2 m;
2pEbr l 2
3:14
600 kV/m
3 m

and in case of clay soil q = 50 X m and Ebr = 12 kV/m

Im q 40 kA
50 X
m
rsp ¼ ¼  0:09  0:1 m;
2pEbr l 2
3:14
1200 kV/m
3 m

i.e. is 20 times less.

From (21) it is seen that the ground coupling factor depends on rsp, D, and l. As rsp
greatly depends on soil type, D and l should be chosen according to each type of soil.
Figure 5 gives correlation between ground coupling factor and distance between axes
of ground electrode and pole D for different rsp calculated upon (21) at l = 3 m. The
optimal value of ground coupling factor for the above studied case makes cg.opt = 0.7
(see Fig. 3). In this case as it is shown on Fig. 5 the distance between pole and ground
electrode axes may change within 0.5 to 3.5 m with respect to rsp, i.e. in relation to soil
type. The higher is the conductivity, the less is the radius of ground electrode with
account made for spark formation in ground (22) rsp and the closer to tower should be
located the electrode (see Fig. 5).

1
Ground coupling factor, сg

rsp=2 m
0.8
rsp=1 m
0.6

0.4 rsp=0,5 m

0.2 rsp=0,1 m

0
0 1 2 3 4
Distance D, m
Fig. 5. Correlation between ground coupling factor and distance between axes of ground
electrode and pole D for different values of rsp at l = 3 m.
 Improvement of Overhead Transmission Lines Lightning Protection 1189

3 Uw

Up Ca

UA

C
UB
A
B

Ug 2 Up

Cg

Fig. 6. Diagram of OHL lightning protection using shielding wire installed on insulation rack
and connected to a stand-alone grounding: 1 – pole; 2 – stand-alone grounding; 3 – insulation
rack; ca – coupling factor between shielding wire and phase B by air.

3 Insulating Lightning-Resistant Rack

The use of insulating racks to suspend the overhead protection wire which is to be
grounded through a stand-alone ground electrode (see Fig. 6) [1, 3] makes it possible to
improve substantially the OHL lightning-surge protection. In this case the rack should
be made of insulating material, for example, of wood, insulating concrete etc.
At lightning strike to the shielding wire the lightning surge current flows on steel
lead into ground via resistor Rg. The voltage applied to insulating rack Ur is equal to:

Ur ¼ Uw  Up ð23Þ

where: Uw – voltage on shielding wire.

Neglecting the voltage drop in lead one may suppose that the wire voltage is
approximately equal to the voltage on grounding electrode Uw  Ug. Then (23) will
take the following form:
1190 G. V. Podporkin


Ur ¼ Uw  Up  Uw  cg Uw ¼ 1  cg Uw : ð24Þ

The voltage on strings of insulators Uins is approximately equal to:


Uins  Up  Ucond ¼ Up  ca Uw ¼ cg Uw  ca Uw ¼ Uw cg  ca ð25Þ

where: Ucond – phase conductor potential; ca – wire-phase conductor coupling factor

averaged for three phases.
At cg = ca the voltage on insulator strings is Uins  0. At that, the voltage applied
to the insulating lightning resistant rack shall be defined according to (24).
In general, if we exclude economic factors, the use of insulating rack may ensure a
rather high line lightning protection level without optimizing the location of ground
electrode. For example, the ground electrode may be located far from tower. Then the
ground coupling factor will be close to zero cg = 0 and according to (24) at lightning
strike all wire voltage will be applied to insulating lightning resistant rack Ur = Uw. To
prevent the insulating rack flashover its length should be sufficiently large. In case the
location of ground electrode is optimal the cg = ca voltage applied to insulating
lightning resistant rack will be considerably less. For instance, at cg = ca = 0.3 the
voltage applied to insulating rack will make Ur = (1 − 0.3)Uw = 0.7Uw, so the rack
length may be substantially reduced.

4 Conclusions
1. The level of lightning protection may be doubled due to the use of standalone
ground electrode as compared to the case of using the electrode galvanic connected
to tower body.
2. Such duplication of lightning protection level does not depend on insulator dis-
charge voltage and on the value of grounding resistance.
3. The higher is the level of OHL lightning protection the higher will be the efficiency
of the use of standalone ground electrodes, hence, this measure may be considered
as practical for 35 kV OHL and above.
4. Due to the use of standalone ground electrodes the rate of 35 kV OHL tripping may
be reduced approximately by an order of magnitude as compared to the traditional
mode of tower grounding.
5. The optimal location of ground electrode in relation to tower depends on soil
conductivity. The higher is the conductivity, the closer to tower should be located
the ground electrode.
6. For single grounding rods as long as 3 m the rod-tower distance is within 0.5 to
3.5 m depending on the type of soil.
7. The use of standalone ground electrodes for connecting lightning arresters on OHL
and the use of insulating lightning resistant racks with wire grounding through
standalone ground electrodes located at optimal distance from towers are considered
as efficient engineering solutions worth to be studied further on with the involve-
ment of scientific, engineering and operating organizations.
 Improvement of Overhead Transmission Lines Lightning Protection 1191

References
1. Grant, I., Shears, D.: Design of a lightning – resistant overhead transmission line. In: The
Institute of Engineers, Australia, Electrical Engineering Transactions, pp. 131–137 (1967)
2. Podporkin, G.: Overhead power line. Patent of Russian Federation №2312441 (2007)
3. ITB Alumnus Successfully Built Isolated Ground Shield Wire System for PLN Towers.
https://stei.itb.ac.id/en/blog/2017/10/11/tesladaya/
4. Kostenko, M., et al.: Lightning physics and OHL lightning protection. Leningrad Polytechnic
Institute (1982). (in Russian)
5. Neiman, L.R., Demirchian, K.: Foundations of Electric Engineering. Textbook for Higher
Education, vol. 2. Energoizdat, Leningrad (1981). (in Russian)
6. Kostenko, M., et al.: Grounding in HV networks and overvoltage protection devices.
Leningrad Polytechnic Institute (1983). (in Russian)
7. High-Voltage Engineering: Text book for students of electric and electric power institutes.
Under general editorship of D. Razevig, Energia (1976). (in Russian)
 Study on Consistency of Failure Probability
Characteristics of Oil-Paper Insulation Under
Different Impulse Voltages

Yu Zhang1, Shijun Xie1, Xiongwei Jiang2, Lian Ye2(&),

Chenmeng Zhang1, Potao Sun2, Zhou Mu1, and Wenxia Sima2
1
State Grid Sichuan Electric Power Research Institute, Chengdu 610000, China
2
State Key Laboratory of Power Transmission Equipment and System Security
and New Technology, Chongqing University, Chongqing 400044, China
yelian@cqu.edu.cn

Abstract. Due to the different overvoltage waveforms and the different dis-
tribution of the waveform parameters, the failure characteristics of insulating
medium under these overvoltage waveforms are significantly different. How-
ever, the failure probability characteristics are closely related to the risk
assessment of insulation failure of power system. Meanwhile, the failure
probability characteristics of insulating medium under different overvoltage
waveforms are not fully understood. Therefore, it is necessary to carry out
research on the difference of failure probability characteristics of insulating
material under different impulse voltages. In this paper, a double exponential
impulse wave (DEIW) and bipolar oscillatory attenuated impulse wave
(BOAIW) test platform was set up. Then, we obtained the failure probability of
oil-immersed insulation paper (OIP) for transformer under the DEIW and the
BOAIW. The differences and commonalities of the failure probability charac-
teristics of OIP under different waveforms were further analyzed. Finally, we
realized the prediction of the failure probability characteristics of oil-paper
insulation under different impulse voltages.

Keywords: Oil-paper insulation
Different impulse voltages
Failure

probability characteristics
Prediction

1 Introduction

The impulse voltage actually invading the transformer has an oscillatory attenuated
characteristic [1–3], so the waveform parameter characteristics are significantly dif-
ferent from the waveforms specified by the IEC. At present, a great deal of research has
been carried out on the on-line monitoring of the power system overvoltage and the
breakdown characteristics of insulating medium under the impulse voltage [4]. And it
is found that there are significant differences in the waveform characteristics of the
overvoltage in different nodes of the power system. Due to the different overvoltage
waveforms and the different distribution of the waveform parameters, the breakdown
characteristics of insulating medium under these overvoltage waveforms are signifi-
cantly different [5, 6].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1192–1206, 2020.
https://doi.org/10.1007/978-3-030-31676-1_111
 Study on Consistency of Failure Probability Characteristics 1193

The risk assessment of insulation failure of power system is the guarantee of safe
and stable operation of power system. However, the failure probability characteristics
are closely related to the risk assessment of insulation failure of power system [7].
Meanwhile, the failure probability characteristics of insulating medium under different
overvoltage waveforms are not fully understood [8].
In view of the above problems, the researchers have carried out preliminary study.
Some of researchers from North China Electric Power University had studied the
failure probability characteristics of transformer oil under DEIW with different
waveform parameters [9], and the influence of waveform parameters on the failure
probability characteristics of transformer oil was preliminarily obtained. Experts from
Institute of Electrical Power Engineering, Lodz University of Technology had studied
the failure probability characteristics of synthetic fiberboard impregnated with different
liquid dielectrics [10, 11]. The voltage strength of insulating paperboard was evaluated
by three-parameter Weibull distribution. And experts from Kumamoto University had
studied the failure probability characteristics of high-pressure carbon dioxide [12]. The
breakdown voltage and breakdown probability of high-voltage carbon dioxide under
the impulse voltage were measured in the case of different electrode structures, and
Weibull statistical analysis was also carried out.
At present, the research mainly focuses on measuring the failure probability
characteristics of different insulating media, and then comparing and analyzing the
electrical strength of insulating media. However, the research on the difference of
failure probability characteristics of the same insulating media under different wave-
forms is very scarce. Therefore, it is necessary to carry out research on the difference of
failure probability characteristics of insulating material under different waveform
impulse voltages. Further, it can provide a research basis for the risk assessment of
insulation failure of power system.
In this work, a test platform of the DEIW and BOAIW was set up. And we obtained
the failure probability of the oil-impregnated paper (OIP) under the DEIW and the
BOAIW with various waveform parameters. Based on the single-variable fitting of the
failure probability data, the failure probability characteristics of OIP was obtained. The
differences and commonalities of the failure probability characteristics of OIP under
different waveforms were further analysed.

2 Experimental Setup

2.1 Test Platform Setup

For the simulation of the slightly uneven electric field observed in winding insulation in
transformers, a column-column electrode structure was used for the test. The structure
is shown in Fig. 1(a). The parameters of the electrodes and the insulation papers are
shown in Table 1.
1194 Y. Zhang et al.

Table 1. Parameters of the electrode and insulating medium.

Types Diameter/mm Thickness/mm Surface roughness/lm
Brass electrode 25 25 0.8
Cellulose insulation paper 80 0.3 –

(a) Schematic of electrode structure.

Impulse generator system

a2 Rf R
D r
a'1 a2
Module of oscillation
attenuation wave
Module of double
exponential wave

C0
Rt C L
a'1 g2
g3 1 2 g3
g3
1
g2 9 8 10
a1
T
a1
Trigger 7
2 6
unit
5
g1 4
3
g1
Test tank
Charging
transformer
Measurement
and control
system Control system Oscilloscope

(b) Experimental platform for impulse voltage.

Fig. 1. Schematic of experimental equipment.

 Study on Consistency of Failure Probability Characteristics 1195

Then, an experimental platform of the DEIW (negative impulse voltages) and

BOAIW (negative impulse voltages) was constructed, as shown in Fig. 1(b). And the
detailed description of some of the devices and structures in Fig. 1(b) is shown in
Table 2.

Table 2. Introduction to the symbols of the experimental platform.

Symbols Description Symbols Description Symbols Description
1 Barometer 7 Insulation oil r Protection
resistance
2 Observation 8 Filling port Rf Wave front
resistance
3 Insulation 9 High-voltage side Rt Wave tail
bracket resistance
4 Drain port 10 High-voltage R Damping
divider resistance
5 Sample D High-voltage C Wave front
silicon stack capacitor
6 electrode C0 Charging capacitor L Series
inductance

The DEIW was generated by the MARX circuit, whereas the BOAIW was generated
by the RLC oscillation circuit. These two impulse voltage waveforms can be switched
by selecting either the test circuit 1 or 2 in Fig. 1(b). By changing the impedance
parameters of the test circuit, we can adjust the wave front time, wave tail time, oscil-
lation frequency, and attenuation constant, respectively. The high-voltage signal gen-
erated during the test was measured by a high-voltage divider and an oscilloscope.

2.2 Sample Pretreatment

All samples were pretreated before the test. We filtered the #25 (Freezing point below
−25 °C) naphthene-base transformer mineral oil to ensure that the transformer oil
meets the relevant technical requirements. The filtered transformer oil and insulation
paper were dried for 48 h at 60 °C and 50 Pa air pressure. The dried insulation papers
were placed into the hot oil, and the mixture was moved into a vacuum drying oven for
48 h (temperature of 60 °C, pressure of 50 Pa). After the transformer oil and insulation
paper were cooled to room temperature, they were stored in sealed cans until use. All
the samples met the technical requirements of the IEC Standard 60641-3-2 [13] and
ASTM Standard D3487-16e1 [14].

2.3 Waveform Parameter Definition and the U50% Measurement Method

The waveform parameters of the BOAIW (wave front time T1, oscillation frequency
f and attenuation constant a) are defined in Fig. 2, where Um is the maximum voltage
amplitude of the BOAIW, aUm is the first positive peak voltage value, bUm is the
1196 Y. Zhang et al.

aUm
bUm α=a-b
Voltage amplitude
T1

C
0

A
0.3Um

0.9Um B 1/f
Um Waveform of bipolar oscillatory
attenuated impulse voltage
Time

Fig. 2. Definitions of waveform parameters.

second positive peak voltage value, and a and b are constants. The attenuation constant
a characterizes the ratio of the difference between aUm and bUm to Um, which is the
difference between a and b.
The failure probability of OIP was obtained through the Multiple-level tests. The
test procedure is given in the IEC Standard 60060-1-2010 [15]. The effective number of
insulation tests under the action of each pair of waveform parameters is 20. The ratio of
the number of insulation breakdown to the effective number of insulation tests gives the
failure probability of OIP under the voltage of each level.

3 Experimental Results and Analysis

We obtained the failure probability of the oil-impregnated paper (OIP) under the DEIW
and the BOAIW with various waveform parameters. Based on the single-variable
fitting of the failure probability data, the failure probability characteristics of OIP was
obtained. The differences and commonalities of the failure probability characteristics of
OIP under different waveforms were further analyzed.

3.1 Failure Probability Characteristics of OIP Under Different

Waveform Impulse Voltages
The Multi-level tests of 0.3 mm OIP was carried out under the DEIW and the BOAIW
with different waveform parameters. The corresponding breakdown probability under
different voltages was measured and calculated. Then, the Gauss accumulation formula,
Weibull accumulation formula and Logistic accumulation formula were used to fit the
experimental data obtained under all impulse voltages. The fitting function expressions
and their goodness of fit are shown in Table 3. It can be seen that the Logistic
cumulative formula has the best goodness of fit, exceeding 0.98.
 Study on Consistency of Failure Probability Characteristics 1197

Table 3. Comparison of goodness of fit.

Distribution types Cumulative formulae Adj. R-Square
h  i
Gauss FðxÞ ¼ 2 1 þ erf rpffiffi2
1 xx c 0.88–0.95
k
Weibull FðxÞ ¼ 1  ekðxxc Þ 0.82–0.95
Logistic FðxÞ ¼ 1  1 þ eðxx
1
c Þ=d
0.98–1

Therefore, this paper uses Logistic accumulation formula to fit the experimental

data, and obtains the failure probability characteristics of OIP under the DEIW and the
BOAIW with different waveform parameters. The experimental results and their fitting
curves are shown in Fig. 3.

1.1
1.0
0.9
0.8
Failure Probability

0.7
0.6
0.5 Tf =1.2μs Tt=50μs
Tf =6.4μs Tt=59.2μs
0.4 Tf =10μs Tt=63.2μs
Tf =20μs Tt=77.6μs
0.3 Tf =2μs Tt=180μs
0.2 Tf =1.6μs Tt=408μs
Tf =250μs Tt=2760μs
0.1 Tf =320μs Tt=5040μs
Tf =240μs Tt=1700μs
0.0 Tf =112μs Tt=376μs
-0.1
30 32 34 36 38 40 42 44 46
Voltage/kV
(a) Fitting curves under the DEIW.
1.1
1.0
0.9
0.8
Failure Probability

0.7
0.6
0.5 T1=40μs f=142Hz α=0.37
T1=30μs f=357Hz α=0.301
0.4 T1=30μs f=746Hz α=0.3
T1=0μs f=847Hz α=0.345
0.3 T1=0μs f=417Hz α=0.152
0.2 T1=0μs f=167Hz α=0.367
T1=0μs f=3.97kHz α=0.311
0.1 T1=0μs f=7.94kHz α=0.341
T1=0μs f=15.1kHz α=0.319
0.0 T1=0μs f=26kHz α=0.293
-0.1
24 26 28 30 32 34 36 38 40 42 44 46 48
Voltage/kV
(b) Fitting curves under the AOBIW.

Fig. 3. Failure probability curves of the OIP.

1198 Y. Zhang et al.

3.2 Dispersion of the Failure Probability Characteristics

The Logistic distribution is used to fit the experimental results. The cumulative dis-
tribution function used for fitting is:

1
FðxÞ ¼ 1  ð1Þ
1 þ eðxxc Þ=d

Where x is a variable corresponding to the breakdown voltage, and xc and d are the
fitting parameters. The fitting parameter xc represents the U50% of the OIP. The fitting
parameter d is positively correlated with the dispersion of the failure probability
characteristics, that is, the larger d is, the greater the dispersion of failure probability
characteristics is.
The density function corresponding to the above cumulative distribution Formula
(1) is:

eðxxc Þ=d
f ðxÞ ¼ ð2Þ
dð1 þ eðxxc Þ=d Þ2

The relationship between the Formulae (1) and (2) is shown in Fig. 4. It can be seen
that the xc obtained by the fitting is U50% of the OIP under the corresponding wave-
form, and the Logistic cumulative distribution curve is centrosymmetric about the point
(xc, 0.5).

Logistic CDF
Logistic PDF
1
Failure Probability

0.73 F(x)

0.5

0.27
f (x)
d d
0
U0% xc/U50% U100%
Voltage/kV

Fig. 4. Diagram of the Logistic distribution.

Through the fitting analysis, the estimated values of xc and d corresponding to the
failure probability characteristics under the impulse voltage with different waveform
parameters were obtained, as shown in Table 4. And the relative deviation of all fitted
estimates is less than 5%.
 Study on Consistency of Failure Probability Characteristics 1199

Table 4. Parameters estimation of failure probability curves under impulse voltage with
different waveform parameters.
DEIW AOBIW
Wave Wave Parameter Parameter Wave Oscillation Attenuation Parameter
front tail time xc d front frequency constant xc
time Tt /µs time f/kHz a
Tf /µs T1/µs
1.2 50 38.11498 1.12398 40 0.142 0.37 39.01976
1.6 408 35.34879 1.13195 0 0.167 0.367 30.85901
2 180 36.85901 1.05472 30 0.357 0.301 40.14178
6.4 59.2 39.09340 1.32416 0 0.417 0.152 29.85353
10 63.2 39.86907 1.17828 30 0.746 0.3 40.74319
20 77.6 40.16337 1.24140 0 0.847 0.345 29.27733
112 376 37.96749 1.11347 0 3.97 0.331 31.02720
240 1700 36.40000 1.00342 0 7.94 0.341 33.12769
250 2760 38.40000 1.16377 0 15.1 0.319 32.82023
320 5040 34.31671 0.86123 0 26 0.293 38.40000

Since the value of xc represents U50%, the variation of the fitting parameter d with
the increase of U50% is obtained, as shown in Fig. 5. It can be seen that the fitting
parameter d of the failure probability characteristics under the action of DEIW and
BOAIW is positively correlated with U50%, and the correlation coefficients are 0.792
and 0.404, respectively. That is, the larger the U50% of OIP is, the larger the value of the
fitting parameter d is. Since the dispersion of the failure probability characteristics is
positively correlated with the fitting parameter d, the dispersion of the failure proba-
bility characteristics of OIP increases with the increase of U50%.

2.2
2.0 DEIW
BOA IW
1.8
1.6
1.4
1.2
d

1.0
0.8
0.6
0.4
0.2
0.0
28 30 32 34 36 38 40 42
xc (U50%)

Fig. 5. Relationship between parameter d and U50%.

1200 Y. Zhang et al.

4 Consistency of Relative Dispersion Coefficient

4.1 Definition of Relative Dispersion Coefficient (d)
In order to characterize the dispersion of the failure probability characteristics of oil-
paper insulation under impulse voltage, a relative dispersion coefficient d was defined
as the ratio of standard deviation r to the U50%, namely:
r
d¼  ð100%Þ ð3Þ
U50%

Where,

pffiffiffiffiffiffiffi Z þ1
r¼ DX ¼ ðx  EXÞ2
f ðxÞdx
1 ð4Þ
p
¼ pffiffiffi d  1:8138d
3
Z þ1
EX ¼ x
f ðxÞdx ¼ xc ð5Þ
1

U50% ¼ xc ð6Þ

By simplifying Formula (3), the relative dispersion coefficient d of the failure proba-
bility characteristic can be obtained as follows:

1:8138d
d¼  ð100%Þ ð7Þ
xc

Table 5. Calculation results of relative dispersion coefficients.

DEIW AOBIW
(xc, d) d (xc, d) d
(38.11498, 1.12398) 0.05349 (39.01976, 1.11054) 0.05162
(35.34879, 1.13195) 0.05808 (30.85901, 1.05472) 0.06199
(36.85901, 1.05472) 0.05190 (40.14178, 1.27228) 0.05749
(39.09340, 1.32416) 0.06144 (29.85353, 1.21650) 0.07391
(39.86907, 1.17828) 0.05360 (40.74319, 1.37198) 0.06108
(40.16337, 1.24140) 0.05606 (29.27733, 0.81968) 0.05078
(37.96749, 1.11347) 0.05319 (31.02720, 0.56804) 0.03321
(36.40000, 1.00342) 0.05000 (33.12769, 1.05056) 0.05752
(38.40000, 1.16377) 0.05497 (32.82023, 1.19717) 0.06616
(34.31671, 0.86123) 0.04552 (38.40000, 0.92868) 0.04387
 Study on Consistency of Failure Probability Characteristics 1201

The relative dispersion coefficients under the DEIW and BOAIW with different
waveform parameters were obtained by Formula (7), as shown in Table 5. It is further
known that the corresponding relationship between the U50% and the relative dispersion
coefficient d of the failure probability characteristics of oil-paper insulation.

4.2 Normality Test of Relative Dispersion Coefficient d

In order to verify the normal randomness of the relative dispersion coefficients in
Table 5, the relative dispersion coefficients need to be tested for normality. Figures 6
and 7 are the schematic diagrams of normality test for relative dispersion coefficients
under the DEIW and BOAIW, respectively.

Fig. 6. Normality test of d under double exponential waves.

(b)

Fig. 7. Normality test of d under oscillation waves.

In this paper, Anderson-Darling, Ryan-Joiner and Kolmogorov-Smirnov test meth-

ods are used to test the normal distribution characteristics of the relative dispersion
coefficients of failure probability characteristics under the DEIW and BOAIW. The test
results of the three normality test methods are shown in Table 6. The smaller the AD
1202 Y. Zhang et al.

value of Anderson-Darling test method, the better the fitting degree of normal distribution
to the data, that is, the better the normal distribution characteristics of the data. The
correlation coefficient RJ value of Ryan-Joiner test method is closer to 1, which indicates
that the data fits the normal distribution better. The smaller the KS value of Kolmogorov-
Smirnov test method, the better the normal distribution characteristics of the data.
As can be seen from the data in Table 6, the results of three test methods indicate
that the relative dispersion coefficients follow a normal distribution. Meanwhile, each
test index of the relative dispersion coefficients under the DEIW and BOAIW is
approximately equal.

Table 6. Results of the three normality test methods.

Test methods Parameters Results
DEIW AOBIW
Anderson-Darling AD 0.215 0.200
P-value 0.791 0.836
Ryan-Joiner RJ 0.981 0.983
P-value >0.100 >0.100
Kolmogorov-Smirnov KS 0.142 0.159
P-value >0.150 >0.150

4.3 Consistency of Relative Dispersion Coefficient d

The above-mentioned failure probability characteristics were obtained in the case of
Multiple-level test with effective experiment times of 20. The effective number of
experiments is closely related to the accuracy of the breakdown probability of oil-paper
insulation, so the relationship between the relative dispersion coefficient d and the
effective test times under DEIW (Tf = 1.2 ls, Tt = 50 ls) and BOAIW (T1 = 0 ls,
f = 847 Hz, a = 0.345) was studied.

0.064
Tf =1.2μs Tt=50μs
0.062
T1=0μs f=847Hz α=0.345
0.060

0.058
δ

0.056 0.05500

0.054 0.05491

0.052

0.050

10 20 30 40 50
times

Fig. 8. Relationship between d and effective test times.

 Study on Consistency of Failure Probability Characteristics 1203

The effective test times involved in the test were 10, 20, 30, 40 and 50 times. As
shown in Fig. 8, as the number of effective tests increases, the relative dispersion
coefficient d gradually tends to a constant. Moreover, the relative deviation between the
two constants of the relative dispersion coefficients under the two waveforms is only
1.64%. That is, the two constants of the relative dispersion coefficients under the two
waveforms are approximately equal and both approach 0.055.
Meanwhile, based on the corresponding relationship between xc and the relative
dispersion coefficient d in Table 5, the relationship between the relative dispersion
coefficient d and U50% was obtained, as shown in Fig. 9.

0.08

DEIW
0.07 AOBIW

0.06 Mean 0.05576

0.05 Mean 0.05383

δ

0.04

0.03

28 30 32 34 36 38 40 42
U50% / kV

Fig. 9. Relationship between d and U50%.

Based on the characteristics that the relative dispersion coefficient has a normal
distribution and Fig. 9, the relative dispersion coefficient d of the failure probability
characteristics of oil-paper insulation under the DEIW is randomly distributed around
the mean value of 0.05383 (5.383%, the percentage represents the proportion of r to
U50%), and the standard deviation of normal distribution is 0.004348 (0.435%). Sim-
ilarly, the relative dispersion coefficient d under the BOAIW is randomly distributed
around the mean value of 0.05576 (5.576%), and the standard deviation of normal
distribution is 0.01156 (1.156%). Meanwhile, the two relative deviations between the
two average values of the relative dispersion coefficients and 0.055 are only 2.13% and
1.38%. Therefore, the relative dispersion coefficients of the failure probability char-
acteristics under the DEIW and BOAIW are approximately equal, that is, the dispersion
of the failure probability characteristics under these two waveforms is consistent.
1204 Y. Zhang et al.

5 Prediction of Failure Probability Characteristics

In this section, the relative dispersion coefficient d of the failure probability charac-
teristics under different waveform impulse voltages is 0.055. Based on the consistency
of the relative dispersion coefficient d under different waveforms, the entire failure
probability characteristics can be predicted by measuring the single breakdown prob-
ability (0 < F(u) < 1) of the oil-paper insulation under the impulse voltage with
amplitude u.
Equations (8) and (9) can be derived from Eqs. (1) and (3) to obtain the parameter
^xc and d^ of the failure probability characteristics curve:
u
^xc ¼ pffiffiffi  ð8Þ
1 þ ð 3d pÞ
lnðFðuÞ=ð1  FðuÞÞÞ
u
d^ ¼  pffiffiffi ð9Þ
p ð 3dÞ þ lnðFðuÞ=ð1  FðuÞÞÞ

Where, ^xc and d^ are the positional parameter and shape parameter of the failure
probability characteristics curve, respectively, and F(u) is the breakdown probability of
the oil-paper insulation under the impulse voltage with amplitude u.

1.1
Tf =10μs Tt=75.5μs
1.0 T1=0μs f=22.1kHz α=0.405
1
0.9 Fitted curves
1'
0.8 Predicted curves
Breakdown probability

0.7
0.6
0.5
0.4 2'
2
0.3
0.2
0.1
0.0
-0.1
30 32 34 36 38 40 42 44 46
Voltage / kV

Fig. 10. Accuracy verification of consistency prediction curves.

The breakdown probabilities F(u = 39.4 kV) and F(u = 36 kV) of oil-paper
insulation under the DEIW (Tf = 10 ls, Tt = 75.5 ls) with amplitude 39.4 kV and
BOAIW (T1 = 0 ls), f = 22.1 kHz, a = 0.405) with amplitude 36 kV were measured
(effective test times were 50 times), respectively. The values of parameters ^xc and d^
 Study on Consistency of Failure Probability Characteristics 1205

were obtained from Eqs. (8) and (9). Furthermore, the consistent prediction curves of
the failure probability characteristics of oil-paper insulation under these two waveforms
were depicted, as indicated by dashed lines in Fig. 10. Meanwhile, the failure proba-
bility characteristics of oil-paper insulation under these two impulse voltages were
measured (the effective test times under each voltage level were 20 times). The Logistic
distribution was used to fit the data to obtain the two corresponding failure probability
characteristics curves, as shown by the solid line in Fig. 10.
As can be seen from Fig. 10, the predicted curves and the fitted curves almost
coincide. From the parameter values in Table 7, it can be seen that the relative devi-
ation of the position parameter between the predicted curve and the fitted curve is less
than 2.16%, and the relative deviation of the shape parameter is less than 5.92%. This
further indicates quantitatively that the difference between the predicted curves and the
fitted curves is very small.

Table 7. Comparison of parameters between consistency prediction

curves and fitted curves.
Predicted Test Breakdown Calculated Calculated
curves voltage probability F value ^xc value d^
u/kV (u)
1 36 0.5 36 1.09163
2 39.4 0.5 39.4 1.19469
Fitted curves Estimated Estimated
value xc value d
1′ 36.07016 1.12671
2′ 39.31482 1.26544

6 Conclusions

This paper focuses on the consistency of the failure probability characteristics of oil-
paper insulation under different impulse voltages. The research results are summarized
as follows:
The failure probability characteristics of oil-paper insulation under DEIW and
BOAIW obey Logistic distribution, and the goodness of fit exceeds 0.98. And the
greater the U50% of the oil-impregnated insulating paper is, the greater the dispersion of
the failure probability characteristics is. The relative dispersion coefficients of the
failure probability characteristics of oil-paper insulation under DEIW and BOAIW are
approximately equal, that is, the dispersion of the failure probability characteristics
under these two waveforms is consistent. When the relative dispersion coefficient d of
the failure probability characteristic of oil-paper insulation under the DEIW or the
BOAIW is known, the entire failure probability characteristics can be predicted by
measuring the single breakdown probability (0 < F(u) < 1) of the oil-paper insulation
under the DEIW or the BOAIW with amplitude u.
1206 Y. Zhang et al.

Acknowledgments. This work is supported by the National Natural Science Foundation of

China (51707023), National Natural Science Foundation of China (51837002). We are grateful
for their support.

References
1. Du, L., Dai, B., Sima, W.X., Chen, W., Jing, L.: Identifying the overvoltage in distribution
networks based on support vector machine. In: 2008 International Conference on High
Voltage Engineering and Application (ICHVE), Chongqing, China. IEEE (2008)
2. Chen, X., Yang, H., Li, H., Liu, Y., Zhang, L.: Overvoltage online monitoring for 35 kV
substation and its waveforms analysis. Electr. Eng. 4, 115–118 (2016)
3. Sima, W.X., Lan, X., Yang, Q., Yuan, T.: Statistical analysis on measured lightning
overvoltage surges in a 110 kV air-insulated substation. IET Sci. Meas. Technol. 9(1), 28–36
(2015)
4. Sima, W.X., Wu, J.Y., Sun, P.T., Yang, M., Hua, J.F.: Breakdown characteristics of oil-
impregnated paper and influential factors for damped alternating oscillation waveforms.
IEEE Trans. Dielectr. Electr. Insul. 23(4), 2079–2087 (2016)
5. Krithika, G., Usa, S.: Volt-time characteristics of OIP under non-standard impulses. In: 1st
International Conference on Condition Assessment Techniques in Electrical Systems
(CATCON), Kolkata, India. IEEE (2013)
6. Zhang, L., Liu, Z., Han, X., Li, J.: Breakdown characteristics of protrusion on HV conductor
in GIS under oscillating and standard lightning impulse voltages. In: 2016 International
Conference on Condition Monitoring and Diagnosis (CMD), Xi’an, China. IEEE (2016)
7. Wang, T.L., Li, Y., Wang, Z., Ni, Hl., Wen, T., Guo, C., Gao, M., Qin, Y.F., Zhang, Q.G.:
Turn-to-turn insulation breakdown characteristics under non-standard lightning impulse
voltages. In: 11th International Conference on the Properties and Applications of Dielectric
Materials (ICPADM), Sydney, Australia. IEEE (2015)
8. Zhang, Q., You, H., Guo, C., Qin, Y., Ma, J., Wen, T.: Experimental research of dispersion
of SF6 discharge breakdown voltage under lighting impulse. High Volt. Eng. 42(11), 3415–
3420 (2016)
9. Zhao, T., Yang, J., Wang, Z., Liu, Y., Lv, F.: Volt-time characteristics of transformer oil
under impulse voltage with different waveforms. In: 2016 IEEE International Conference on
High Voltage Engineering and Application (ICHVE), Chengdu, China. IEEE (2016)
10. Rozga, P.: Using the three-parameter Weibull distribution in assessment of threshold
strength of pressboard impregnated by different liquid dielectrics. IET Sci. Meas. Technol.
10(6), 665–670 (2016)
11. Rozga, P., Stanek, M.: Assessment of electrical strength of liquid immersed pressboard using
three-parameter Weibull distribution. In: 2016 International Conference on Condition
Monitoring and Diagnosis (CMD), Xi’an, China. IEEE (2016)
12. Kiyan, T., Ihara, T., Kameda, S., Furusato, T., Hara, M., Akiyama, H.: Weibull statistical
analysis of pulsed breakdown voltages in high-pressure carbon dioxide including
supercritical phase. IEEE Trans. Plasma Sci. 39(8), 1729–1735 (2011)
13. IEC 60641-3-2: Pressboard and press-paper for electrical purposes - Part 3: Specification for
individual materials - Sheet 2: Requirements for press-paper (2008)
14. ASTM D3487-16e1: Standard Specification for Mineral Insulating Oil Used in Electrical
Apparatus (2016)
15. IEC 60060-1-2010: High voltage test techniques - Part I: General definitions and test
requirements (2010)
 Insulation Withstand Testing on Surge
Arresters and the Influence of Voltage Grading

Hans Sjöstedt(&) and James Taylor

ABB AB, High Voltage Products, Ludvika, Sweden

hans.sjostedt@se.abb.com

Abstract. In order to reduce the risk of insulation failure to an economically

and operationally acceptable level, the insulation withstand and clearances for
substation equipment is selected with regard to anticipated overvoltages; taking
into account the expected protective characteristics of nearby surge arresters in
the station. Insulation coordination standards further recognize that, since the
primary function of a surge arrester is to limit, and thereby render harmless,
overvoltages to which the protected equipment is exposed, the arrester’s own
insulation is obviously the best protected of all. In contrast to other HV appa-
ratus, the insulation level for surge arresters therefore does not need to fulfil a
standardized insulation class since the arrester will effectively protect its own
insulation against overvoltages.
An intensive review of the surge arrester standard IEC 60099-4 resulted in the
release of Edition 3.0 during 2014. Amongst other matters, this introduced a new
procedure for verifying the dielectric withstand of the external insulation on an
arrester. Equations were defined which consider the relationship between dry
arcing distances and withstand voltages and thereafter the need to perform a
specific insulation withstand test or not. For arresters intended for use on systems
of Us > 245 kV, lightning and switching impulse voltage tests on complete
arrester assemblies based on the arrester’s protective level can be especially
defining. During the tests the nonlinear metal-oxide (MO) resistors are replaced by
linear resistors, capacitors or high gradient MO resistors in order to achieve the
supposed voltage distribution along the arrester during impulses. The MO resistor
distribution within a multi-unit arrester as well as the grading ring configuration
used may impact the withstand voltage of the arrester, and arresters with similar
appearance or with the same arcing distance can potentially perform differently.
This paper discusses the requirements for insulation withstand on surge
arrester housings and evaluates the role voltage grading plays on arrester per-
formance, both during dielectric testing and normal operation.

Keywords: Surge arrester
Insulation withstand voltage
Insulation

coordination

1 Introduction

Insulation coordination is the procedure for the selection of the most economical level
of insulation for electrical equipment which permits fault-free operation during both
normal and abnormal operating conditions. In order to define the desired insulation

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1207–1217, 2020.
https://doi.org/10.1007/978-3-030-31676-1_112
1208 H. Sjöstedt and J. Taylor

withstand levels for a particular electrical apparatus in a specific situation, it is nec-

essary to establish an acceptable risk of failure based on technical and economic
analysis and service experience. The risk defines the probability of insulation failure,
and the failure rate is then expressed in terms of the expected average frequency of
failures of the insulation resulting from events causing overvoltage stresses. Statistical
risk of an external flashover less than or equal to 0.1% per year is generally considered
acceptable [1].
Surge arresters provide the first line of defence against overvoltages for which other
apparatus within an electrical system are not inherently designed to withstand, i.e.
lightning and switching surge impulses beyond a defined magnitude. The insulation
withstand and clearances for station equipment is hence selected with regard to
anticipated overvoltages, taking into account the expected protective characteristics of
nearby surge arresters in the station. Thus, arresters play a vital role in minimizing
insulation breakdown on other equipment, but even their own design should take into
account the same principles of defining the overvoltage stresses which will actually be
applied to them in order to choose their most economical level of insulation.
Within the station, the surge arrester design and placement has to be such that it will
not increase the total risk for insulation failure. The insulation withstand of the surge
arrester itself - including across individual units, between flanges and grading rings, etc.
- thus has to be coordinated with its own protective characteristics. This paper discusses
the necessary requirements in this regard, with reference to the testing demands of the
surge arrester standard IEC 60099-4, Edition 3.0 [2] and the role voltage grading plays
on arrester performance, both during dielectric testing and normal operation.

2 Surge Arrester Insulation Withstand

As stated in clause 8.2 in the IEC surge arrester standard [2] for insulation withstand
tests on the arrester housing: “The voltage withstand tests demonstrate the voltage
withstand capability of the external insulation of the arrester housing”. This test is then
related to the external physical dimensions and not the internal components. In actual
service, through limiting the overvoltages subjected to other equipment, the arrester
protects its own insulation the best of all. Distance effects need not be considered since
the voltage across an arrester can never be higher than that determined by the arrester’s
protective characteristics. Only the need for an additional statistical safety margin and
correction for installation altitude can technically justify requiring a higher external
insulation strength. This is reflected in the standard’s subclauses 8.2.6–8.2.8 [2], where
the insulation requirements for arresters at altitudes up to 1000 m.a.s.l. are based on the
arrester’s protection levels with a reasonable safety margin added. In contrast to other
HV apparatus, the insulation level for surge arresters therefore does not need to fulfil a
standardized insulation class. Such reasoning is also followed through in a number of
other relevant IEC standards, as summarized below.
 Insulation Withstand Testing on Surge Arresters 1209

– IEC 60099-5: Clause 5.2.2.2 - The standard insulation levels of the insulation
coordination standards are not relevant to surge arresters and, in general, only the
requirements of IEC 60099-4 [2] should be applied [3].
– IEC 60071-1: Clause 5.5 - Required withstand voltages of the insulating housing on
surge arresters are based on the protective levels and therefore standardized with-
stand voltages shall not be selected [4].
– IEC 60071-1: Annex A - Lower clearance distances are permitted when a) separate
impulse testing is specified for the equipment and b) the overvoltages are antici-
pated to be lower than the standardized withstand voltages [4]. Both of these criteria
clearly inherently relate to surge arresters.
The ability for a surge arrester to perform with significantly reduced insulation with-
stand level has been trialed and verified in practice [5]. The 420 kV surge arrester in
question was approximately 2/3rd the height of that of a conventional design and passed
the applicable and valid type tests as well as a 3.5 year long site trial without issue or
flashover. Notwithstanding possible local regulations regarding safety clearances, only
the need for additional creepage distance because of adverse pollution conditions
and/or altitude correction at a given site location would require a taller arrester [6].

3 Dielectric Testing According to IEC Surge Arrester

Standard
3.1 General
Historically, the IEC surge arrester standard [7] only required dielectric testing to be
performed on the longest individual unit of a particular arrester type together with, if
not one and the same, the unit housing having the highest specific voltage stress. A new
method for external insulation impulse withstand verification was introduced in the
new edition [2], whereby arresters intended to be used on system voltages above
245 kV require determination of the insulation withstand of the complete arrester
assembly, including grading rings. This needs to be verified by test or calculation. The
purpose of the insulation withstand test is to verify the voltage withstand capability of
the external insulation of the arrester housing. During the withstand testing with
lightning and switching impulses the metal-oxide (MO) resistors are replaced by
resistors, capacitors or higher resistance MO resistors in order to obtain approximately
the same voltage distribution as the actual arrester would have during those type of
events.

3.1.1 Lightning Withstand Impulse Test

The arrester is subjected to a standard lightning impulse voltage dry test according to
IEC 60060-1 [8], with a test voltage of at least 1.3 times the residual voltage of the
arrester. The 1.3 factor includes a 15% coordination factor to account for discharge
currents higher than the nominal current as well as the statistical nature of the withstand
1210 H. Sjöstedt and J. Taylor

voltage of the insulation, and 13% margin to account for variation in air pressure from
sea level up to normal service altitudes of maximum 1000 m.
To verify the lightning withstand voltage of the housing, fifteen consecutive
impulses at the test voltage shall be applied for both polarities. For each polarity, a
maximum number of two external flashovers, and no internal, are allowed in order to
pass the test. The test is only required if the dry arcing distances in m are smaller than
the test voltage in kV divided by 500 kV/m.

3.1.2 Switching Withstand Impulse Test

The arrester is subjected to a standard switching impulse voltage wet test according to
[8], with a test voltage according to Eq. (1).

USIWV ¼ 1:1  em1000=8150  USIPL ð1Þ

where:
USIWV = Switching impulse withstand voltage in kilovolt (kV)
USIPL = Switching impulse protective level in kilovolt (kV)
m = 1 for arresters intended for use on system voltages equal to or less than
800 kV, m is taken from Fig. 9 [1] for arresters intended for use on system voltages
higher than 800 kV
To verify the switching withstand voltage of the housing, fifteen consecutive impulses
at the test voltage shall be applied for both polarities. For each polarity, a maximum
number of two external flashovers, and no internal, are allowed in order to pass the test.
The test is only required if the dry arcing distances in m are smaller than the calculated
value according to Eq. (2).
 
d ¼ 2:2  eðUSIWV =1069Þ  1 ; ð2Þ

where:
d = distance in meter (m)
USIWV = Switching impulse withstand voltage in kilovolt (kV), as calculated
according to Eq. (1)
Equation (2) is derived from equation F.3 in [1].

U50 ¼ k  1080  lnð0:46  d  1Þ; ð3Þ

where:
U50 = 50% switching impulse flashover voltage in kilovolt (kV)
k = is the gap factor and is assumed to be equal to 1.1. Two standard deviations of
0.05 each are taken into account to achieve the withstand voltage.
 Insulation Withstand Testing on Surge Arresters 1211

4 Influences of Voltage Grading

Ideally the voltage potential across the housing would increase linearly with the height
of the arrester. But, due to stray capacitances from the arrester to ground [9–13] and to
other phases, the potential is instead non-linearly distributed along the arrester length.
As a consequence, there is higher dielectric stress on the MO resistors in the upper part
of the arrester. If not compensated for, this may lead to degradation of the MO resistors
[14] and/or increased temperature which in turn may reduce the energy and temporary
overvoltage withstand of the arrester. Traditionally, adjustment to the voltage grading
of the arrester has been obtained by using external grading rings, and in some cases
internal grading capacitors.
To ensure stability at applied continuous voltage, the following methods may be
used (individually or in combination) to design an arrester with reasonable linear
voltage distribution and also fulfill the insulation withstand voltage requirements:
• Grading ring diameter: generally more linear voltage distribution with larger radius
• Grading ring down-hang: generally more linear voltage distribution with greater
suspension
• MO resistor distribution: generally more linear voltage distribution with more MO
resistors in the upper part of the arrester
Some arrester designs are configured with multiple grading rings suspended at
different distances from the top of the arrester. This is often done to equivalently reduce
the voltage stress in the upper part of the arrester when a large down-hang would
otherwise be required. Added benefits include a reduced electric field and increased
mechanical stability of the grading ring assembly during seismic activity.
For higher system voltages, the height of the arrester results in the need to use very
large grading rings suspended significantly. At the same time, the necessary distance to
ground to uphold the required switching or lightning surge withstand voltages sets a
limitation on the possible down-hang of the grading rings. This in turn results in an
arrester which may need to be much taller than required to simply house the necessary
number of MO resistors.

5 Impulse Withstand Test Experience

Since it is not practical to type test all arrester configurations, the expectation from the
new method is that surge arrester manufacturers will perform insulation withstand
testing on a number of representative arrester housing-configurations in order to define
the generic coordinating characteristic for insulation withstand of individual families of
surge arresters.
Definition of the lightning impulse performance is straightforward since the char-
acteristic is linear. As per IEC, the defined withstand is 500 kV/m and performed
testing has shown this to be a valid conservative representation that can be used for
evaluation, see Fig. 1.
Definition of the switching impulse performance is somewhat more complicated
since the characteristic is non-linear. Figure 2 shows the relationship between arcing
1212 H. Sjöstedt and J. Taylor

Fig. 1. Relationship between lightning impulse withstand voltage and arcing distance and test
values on different types of arresters.

distance and impulse voltage for switching surge impulse voltage based on theoretical
equations as well as actual test results on various types of arresters. The IEC curve
based on Eq. (2) is derived with consideration to the switching impulse withstand
voltage of an air gap having dependence not only on the gap length but also on the gap
geometry. Gap factors for a variety of gap geometries with values ranging 1.03–1.66
are discussed in [1]. Arguably the closest gap geometries to a surge arrester are con-
ductor - crossarm and conductor - window, for which ranges given are 1.36–1.58 and
1.22–1.32, respectively. In [2] a gap factor of K = 1.1 was chosen for the sake of
conservatism until more testing had been performed. Subsequent testing has indeed
shown that, especially at higher system voltages with large grading ring geometries,
Eq. (2) is extremely conservative and arresters can actually withstand a higher voltage
for a given arcing distance.
With more actual testing performed, confidence was gained to propose a more
pragmatic approach [15] by replacing the exponent 1069 with 1118 in Eq. (2). Com-
pared with either of these theoretical characteristics, the actual withstand of tested
arresters are higher than required, i.e. provided a given arrester has an arcing distance at
or below the value from the characteristic, then the testing made has verified that it is
capable of withstanding the defined USIWV.
 Insulation Withstand Testing on Surge Arresters 1213

Fig. 2. Relationship between switching impulse withstand voltage and arcing distance and test
values on different types of arresters.

5.1 Effect of Design Configuration

Depending on the system voltage as well as the arrester rated voltage, the significance
of the type of impulse withstand test required will differ. For multi-unit arresters,
lightning impulse withstand has most influence on system voltages up to 362 kV
compared to power frequency and switching withstand voltages. Due to the nonlinear
relationship between withstand voltage and arcing distance, switching surges will
influence the arrester design more on system voltages above 362 kV than lightning
impulse withstand and the specifics of the design can play a role as well.
Notwithstanding that an arrester assembly having a certain arcing distance has been
shown capable of withstanding a given switching impulse voltage, depending on the
arrester design the value may differ significantly if the arrester design was subsequently
altered, such as MO resistor distribution or grading ring configuration.
Two 360 kV rated arresters with the same MO resistor distribution but of different
station classes and with different grading rings, as shown in Fig. 3, were tested with
respect to switching impulse withstand voltage. Their partial arcing distances were
similar, although Arrester B had a slightly longer total arcing distance. However, their
withstand voltages were significantly different as shown in Table 1, with Arrester A
coping with a higher applied voltage.
It is necessary to not only consider the complete arcing distance, but also the
relative voltage distribution across the respective individual partial arcing distances as
1214 H. Sjöstedt and J. Taylor

Fig. 3. Surge arresters with same MO resistor distribution with different grading rings and class.

well. With consideration to the arcing distances and Eq. (2), it was judged necessary to
test both arresters to verify their respective switching surge insulation withstand. As
noted in Table 1, each arrester upheld the required USIWV value according to Eq. (1),
and therefore both fulfill the requirements stated in [2]. However, the difference in the
results highlights that, for switching impulses, it may not be sufficient to focus solely
on distance across the housing length as the sole criteria without also considering the
voltage grading as well as the grading ring design and configuration.
Furthermore, during lightning impulse testing, it was also observed that the MO
resistor distribution within a multi-unit arrester will affect the withstand voltage of the
arrester. When the applied voltage between grading ring and intermediate flanges is too
high, a partial flashover may occur rather than a direct flashover to ground.
Nonetheless, the testing performed indicates that the 500 kV/m value stated in [2] is
sufficiently conservative to be used for evaluation of an arrester with grading ring(s).
Based on test experience, due to the different relationship between arcing distance
and the respective impulse voltage, the arrester design needs to consider varying
aspects depending on the system voltages. Below system voltages of 362 kV, lightning
impulses risk flashover either across partial or the complete arcing distance, whereas
 Insulation Withstand Testing on Surge Arresters 1215

switching impulses typically flashover across the complete distance. For higher system
voltages, the dimensioning parameters of the arrester tend to be lightning impulse
withstand of partial arcing distances and switching impulse withstand of the complete
arcing distance.

Table 1. SIWV test results of the arresters in Fig. 3. Arrester A with

a USIPL of 714 kV and Arrester B with a USIPL of 728 kV.
Arrester Req’d Test voltage Impulses Number of
USIWV (kV) (corrected, kV) flashovers
A 887 982 15 pos 1
A 887 982 15 neg 0
B 905 905 15 pos 2
B 905 905 15 neg 0

6 Causes of Concern from Overdimensioning

As discussed herein, an arrester must have sufficient insulation withstand across its own
housing. However, users may not realize that there are notable drawbacks to defining a
higher value than that determined by coordination with the arrester’s protective
characteristic.
A taller than necessary surge arrester may lead to less effective protection for steep-
fronted surges for which the inductance of the arrester itself becomes more significant.
Hence, the stipulation of high external insulation withstand values, e.g. equal to that for
the protected equipment, may thus be disadvantageous for the protected equipment and
should not be done unnecessarily.
For a given design, the permissible applied mechanical loading is proportional to
the height of the arrester. Notwithstanding that mechanical forces on surge arresters
should always be minimized wherever possible, the fact remains that, for a given
applied force, a taller arrester will be loaded with more relative bending moment at its
base compared to a shorter housing.
A taller arrester of a given design and rating will likely require improved voltage
grading in order to keep it thermally stable in service. This is usually easily able to be
accommodated by appropriate redistribution of the MO resistors along with the use of
external grading rings of larger diameter and/or greater down-hang. However, it may
not be possible to place an arrester with larger grading rings in a given location if the
necessary clearances have not been considered in the station layout. This is of particular
concern in retrofit installations, where previous generation arresters were probably not
designed with the same criteria in mind. Furthermore, a larger ring down-hang may
mean the flashover distances are made worse, not better, requiring that an even taller
housing be used that may need even greater voltage grading, which in turn spirals out
to an impossibility of fulfilling all design criteria.
1216 H. Sjöstedt and J. Taylor

Finally, a taller arrester of a particular design and rating will surely cost more than a
shorter one with the same electrical properties. When the additional size is not tech-
nically justified for the design, the extra cost becomes an unnecessary added burden to
the station project finances, i.e. the user is paying for something which is not actually
needed.

7 Conclusion

Users may not fully grasp that surge arresters are unique amongst station equipment, in
so much that the impulse voltage across it will always be inherently limited by its own
protective characteristic. Thus, arrester housings are the best protected of all and
consequently do not need to uphold the same standardized insulation withstand values
as for the other electrical equipment in the same station. Instead the required surge
arrester impulse withstand voltage is directly proportional to its protective levels.
Specifying a higher than necessary value is unwarranted with respect to mitigation of
acceptable flashover risk, and conversely can instead result in significant technical and
financial disadvantages. Hence, if not required by safety regulations, the necessary
arcing distance across the surge arrester should be related to the protective level of the
arrester or verifying withstand voltage tests made by the manufacturer.
Introduction of procedures into the surge arrester standard IEC 60099-4, Edition 3.0
[2] during 2014 for verifying the dielectric withstand of the external insulation on
complete arrester assemblies, by calculations or testing, was a positive step towards
ensuring that the design as a whole is secure on transmission systems with system
voltages above 245 kV.
The stipulated relationship between arcing distance and withstand voltage
according to [2], for both lightning and switching withstand voltages, is conservatively
given, with the switching impulse requirements anticipated to be adjusted in the future
as more test results come forth. However, due to the influence of the grading ring(s)
and the MO resistor distribution within the arrester on the withstand voltage, care must
be taken when evaluating the overall design of a surge arrester with respect to both
continuous operating voltage and impulse withstand voltage.

References
1. IEC 60071-2, Edition 4.0, 2018-03, Insulation co-ordination - Part 2: Application guidelines
2. IEC 60099-4, Edition 3.0, 2014-06, Surge arresters - Part 4: Metal-oxide surge arresters
without gaps for a.c. systems
3. IEC 60099-5, Edition 3.0, 2018-01, Surge arresters - Part 5: Selection and application
recommendations
4. IEC 60071-1, Edition 8.1, 2011-03, Insulation co-ordination - Part 1: Definitions, principles
and rules
5. Stenström, L., Taylor, J., Westerlund, H.: An optimal surge arrester for EHV air insulated
stations utilizing high-gradient MO resistors. In: IEEE Electrical Insulation Conference
(EIC), Ottawa (2013)
 Insulation Withstand Testing on Surge Arresters 1217

6. Martin, M., Taylor, J., Popic, M.: Optimisation of air-insulated substations by consideration
of surge arrester protective characteristic. HRO CIGRÉ Session, Cavtat (2013)
7. IEC 60099-4, Edition 2.2, 2009, Surge arresters - Part 4: Metal-oxide surge arresters without
gaps for a.c. systems
8. IEC 60060-1, Edition 3.0, 2010-09, High-voltage test techniques – Part 1: General
definitions and test requirements
9. Csendes, Z.J., Hamann, J.R.: Surge arrester voltage distribution analysis by the finite
element method. IEEE Trans. Power App. Syst. PAS-100(4), 1806–1813 (1981)
10. Oyama, M., Ohshima, I., Honda, M., Yamashita, M., Kojima, S.: Analytical and
experimental approach to the voltage distribution on gapless zinc-oxide surge arresters.
IEEE Trans. Power App. Syst. PAS-100(11), 4621–4627 (1981)
11. Valsalal, P., Usa, S., Udayakumar, K.: Effect on stray capacitance on surge arrester
performance, In: Lecture Notes in Engineering and Computer Science, vol. 2178, no. 1,
pp. 541–544 (2009)
12. Haddad, A., Naylor, P.: Finite-element computation of capacitance networks in multiple
electrode systems: application to ZnO surge arresters. IEE Proc. Sci. Meas. Technol. 145(4),
129–135 (1998)
13. Sjöstedt, H., Stenström, L., Pusch, D., Ostrowski, J.: Voltage grading design of UHV surge
arrester using 3D transient capacitive-resistive field simulations. In: International Conference
on High Voltage Engineering and Application (ICHVE), New Orleans, pp. 32–35 (2010)
14. He, J.L., Hu, J., Gu, S.Q., Zhang, B., Zeng, R.: Analysis and improvement of potential
distribution of 1000-kV ultra-high voltage metal-oxide arrester. IEEE Trans. Power Del. 24
(3) (2009)
15. CIGRÉ TB 696: MO Surge Arresters – Metal Oxide Resistors and Surge Arresters for
Emerging System Conditions. CIGRE WG A3.25 (2017)
 Characteristics of Failed Bypass Diodes
for Photovoltaic Module by Artificial
and Natural Lightning

Toshiyuki Hamada1(&), Kenta Nakamoto1, Ikuo Nanno1,

Norio Ishikura2, Shinichiro Oke3, and Masayuki Fujii4
1
National Institute of Technology, Ube College, 14-1, Tokiwadai, Ube,
Yamaguchi 755-8555, Japan
hamada@ube-k.ac.jp
2
National Institute of Technology, Yonago College, 4448, Hikonacho, Yonago,
Tottori 683-8502, Japan
3
National Institute of Technology, Tsuyama College, 624-1, Numa, Tsuyama,
Okayama 708-0824, Japan
4
National Institute of Technology, Oshima College, 1091-1, Oazakomatsu,
Suooshima, Yamaguchi 742-2104, Japan

Abstract. In recent years, bypass diodes (BPDs) in photovoltaic (PV) modules

have failed because of lightning-induced surges, and burnout of BPDs has also
occurred. In this study, we investigate the characteristics of BPDs in PV
modules, with the most likely cause of failure being lightning-induced surges;
conduct an artificial lightning test of BPDs; and compare the failure charac-
teristics with those BPDs that failed due to natural lightning. As a result, we
confirmed that the schottky barrier diodes (SBDs) exposed to lightning-induced
surges show short-mode failures where the fault resistance decreases inversely
proportionally to the increase in the lightning surge amplitude. We also con-
firmed that the BPD in the PV module damaged by natural lightning fails in
short mode or open mode. When the BPD fails in short mode, the circulating
current generated in the bypass circuit increases the risk of heat generation and
ignition.

Keywords: Lightning-induced surge
Bypass diode
Schottky-barrier diode

Photovoltaic module
Artificial lightning
Natural lightning

1 Introduction

1.1 A Subsection Sample

Since the fixed-purchase system of renewable energy was introduced in Japan, PV
systems (PVSs) have become increasingly popular. Accordingly, accidents due to the
failure of PVSs have become very common. A typical PVS failure is the failure of
bypass diodes (BPDs) in a PV module due to an lightning-induced surge [1]. Bypass
diodes (BPDs) are known to fail owing to lightning-induced surges when lightning
strikes in the vicinity of their PVS, and the failure of BPDs due to lightning-induced

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1218–1224, 2020.
https://doi.org/10.1007/978-3-030-31676-1_113
 Characteristics of Failed Bypass Diodes for Photovoltaic Module 1219

surges has been experimentally verified by Haeberlin et al. [2–6]. We have previously
pointed out that if a BPD fails in short-mode due to lightning, a risk of overheating and
burnout of short-circuited BPDs is introduced by (1) the circulating current generated
in the bypass circuit consisting of PV cell and BPD; (2) the bypass current through
failed BPDs, originating from other cluster heads; and (3) the reverse current from a
other sound string appearing when a short-circuit failure occurs in the blocking circuit
at the same time [7, 8]. In particular, the risk of ignition and burnout of a PVS is high
if the short-circuit failure of a BPD occurs at a fault resistance of approximately
1–10 X [9].
In this report, BPDs in solar modules that failed after a natural lightning surge were
retrieved and their electrical properties were characterized. Furthermore, the electrical
characteristics of BPDs failed by artificial lightning surge are comparatively examined.

2 Experimental Method
2.1 Measurement of the Characteristics of Bypass Diodes in Solar
Modules Damaged by a Natural Lightning Surge
In 2017, at a 2-MW photovoltaic power plant operated in Kyushu Japan, an accidental
burnout occurred because of a lightning strike near a BPD (Schottky barrier type diode)
in a PV module. In most cases, a lightning strike mark, typically due to a direct
lightning strike, was not observed in the failed PVS, suggesting that lightning-induced-
surge damage is a very likely possibility. In this instance, on-site open-circuit voltage
measurements were performed on the damaged and removed PV module; thereby, if
the open-circuit voltage was abnormally low, the BPD of the PV module was recovered
for measurement of the electrical characteristics of the BPD. Notably, if the BPD fails
in the short mode, the output voltage of the PV module decreases if the bypass circuit
that comprises a PV cell and a BPD is short-circuited. The specifications of the PV
modules used in this report were as follows: 38 V, short-circuit current: 9.00 A,
Output: 261 W.

2.2 Artificial Lightning Tests of Bypass Diodes

Figure 1 shows a schematic of the lightning-induced-surge test configuration. In the
lightning-induced-surge test based on the test standard IEC6100-4-5 [10], the dielectric
breakdown test of schottky barrier diodes (SBDs) was conducted for the BPDs of PV
modules using the combination of an open-circuit voltage waveform 1.2/50 ls and a
short-circuit current waveform 8/20 ls. For the dielectric breakdown test, we used
SBDs with a repetitive peak reverse voltage VPRV of 45 V and an average rectified
current I0 of 30 A for the PV module, whose failure by lightning surge was confirmed
in the field investigation for PV power generation systems by our research group. After
the test, the electrical characteristics of the failed BPDs were measured and evaluated
1220 T. Hamada et al.

with an current–voltage (I–V) curve tracer. Afterwards, the failure characteristics of

BPDs due to lightning-induced surges were evaluated through the comparison of the
electrical characteristics of the BPDs that failed in the artificial lightning test and those
that failed from the surge induced from natural lightning.

Fig. 1. Schematic of the lightning-induced-surge test setup.

3 Experimental Results
3.1 Characteristics of BPDs in the PV Module that Failed Because
of a Natural-Lightning Surge
Figure 2 shows the I–V characteristics of the SBDs in the PV module that failed
because of natural lightning. Incidentally, the PV module that suffered lightning
damage had three bypass circuits, each comprising serially connected PV cells and a
BPD. Therefore, three BPDs are installed in each PV module, and this report describes
the characteristics of BPDs in the two PV modules (A and B) that were damaged by a
lightning surge. Figure 2 confirms that in PV module A, short-mode failures occurred
in two of the three BPDs. In PV module B, all three BPDs failed; however, two of these
failed in short mode while the remaining one failed in open mode. Note that when a
BPD fails in short mode, a circulating current flows in the bypass circuit and the output
voltage decreases. Therefore, short-circuited BPDs have a greater risk than open-
circuited BPDs of heat generation and burnout due to the circulating current. Also, in
module B, we confirmed that a BPDs exhibited open-mode failure.
 Characteristics of Failed Bypass Diodes for Photovoltaic Module 1221

(a) Module A (b) Module B

Fig. 2. Current–voltage characteristics of SBDs that failed because of natural lightning.

3.2 Characteristics of BPDs that Failed by Artificial Lightning-Induced

Surges
Figure 3 shows the I–V characteristics of SBDs to which lightning-induced surges
were applied in the forward direction. The test setting value was evaluated by calcu-
lating the charge of the lightning surge from the BPD current. The figure confirms that
the SBD is short-circuited if the charge of the lightning surge is >12.0 mA s. Evalu-
ation of the failures confirmed that the resistance decreases in the reverse direction at
voltages below VPRV (where no current normally flows) and in the forward direction
below the VF of the diode. However, because the forward characteristics of diodes were
also confirmed in the voltage region above VF, we considered that in the inner part of
an SBD, the Schottky barrier layer is partly short-circuited because of the heat gen-
eration caused by the lightning surge current. As the charge of the lightning surge
increases, the short-circuit failure area inside the SBD due to lightning surge also
expands. We considered that the slope of the I–V curve increases and that the failure
resistance decreases.
Figure 4 shows the I–V characteristics of SBDs in which a lightning-induced surge
was applied to the SBD in the reverse direction. The I–V curves confirm that the SBD
fails in the short mode when the charge of the lightning surge is greater than
0.502 mA s. The fault resistance is also found to decrease as the charge of lightning
surge increases, similar to the results of the forward-direction test. Therefore, we
clarified that in the lightning-induced-surge test, SBDs fail in short mode.
1222 T. Hamada et al.

Fig. 3. Current–voltage characteristics of SBDs after the lightning-induced-surge test (in the
forward direction).

Fig. 4. Current–voltage characteristics of SBDs after the lightning-induced-surge test (in the
reverse direction).

4 Discussion

By measuring the electrical characteristics of BPDs in the PV power generation module

recovered from the solar PV power plant damaged by natural lightning, we confirmed
that BPD failure can occur in short mode or in open mode. However, we also con-
firmed that all BPD failures in the artificial test with lightning-induced surges occurred
in short mode. Two explanations are offered for this observation. One type of BPD
failure in the open mode can originate from lightning-induced surges with a much
larger amount of charge than that used in the lightning-induced surges test in this case.
The second type of open-mode BPD failure can originate from overheating due to the
power generation current passing through the BPD that has failed in short mode.
 Characteristics of Failed Bypass Diodes for Photovoltaic Module 1223

Figure 5 shows the external appearance of an SBD (B-2) that failed in open mode
and an SBD (B-1) that failed in short mode as shown in Fig. 4. As shown in the figure,
the SBD that failed in short mode has an appearance similar to that of a sound SBD;
however, in the SBD that failed in open mode, the surrounding filler is melted because
of heat generation and we also confirm the presence of cracks in the SBD due to
expansion by heating. This observation suggests that damage is caused not only by
instantaneous heat generation due to lightning surges but also by heat generation over a
prolonged period of time. The heating mechanism of the failed BPD is as follows. In
the SBD used in the BPD that is short-circuited due to lightning, if the BPD is short-
circuited in the bypass circuit made of BPD and a PV cell, the generated current
circulates and heats up the BPD. Alternatively, the reverse current flows into the short-
circuited BPD because of the difference in the output voltage compared to that of a
sound string, causing heating of the BPD and resulting in an open failure.
In the PV cell power plant exposed to lightning, which was the subject of this
investigation, in addition to burnout damage and breakdown of BPDs in the PV
module, numerous strings were observed in which the blocking diodes (BLDs)
simultaneously failed in short mode. Therefore, because the string where the output
voltage decreases because of a short-circuit BPD failure causes a failure of the BLD
(the purpose of which was to prevent the reverse current flow), overheating of the
short-circuited BPD due to the reverse current in a sound string is established; an open
failure may have occurred for this reason. Note that our research group has demon-
strated, experimentally and by simulation, the burnout mechanism of a BPD [7, 8, 11].

(a) Sound SBD (b) B_2 (open failure) (c) B_1 (short failure)

Fig. 5. Current–voltage characteristics of SBDs after the lightning-induced-surge test (in the
reverse direction).

5 Conclusions

Failure mechanisms of SBDs used as the BPDs of PV modules were clarified in cases
of failures caused by lightning-induced surges. As a result, we confirmed that the SBDs
exposed to lightning-induced surges show short-mode failures where the fault resis-
tance decreases inversely proportionally to the increase in the lightning surge
1224 T. Hamada et al.

amplitude. We also confirmed that the BPD in the PV module damaged by natural
lightning fails in short mode or open mode. When the BPD fails in short mode, the
circulating current generated in the bypass circuit increases the risk of heat generation
and ignition. In addition, open failure mechanisms can include the effects of lightning
surges with a charge quantity exceeding 63.0 mA s, which is used in this test, or heat
generation by circulation current arising after the short-circuit failure.

Acknowledgements. A part of this study was supported by the New Energy and Industrial
Technology Development Organization (NEDO) and Power Academy. The author (One of the
authors (T. Hamada)) was supported through the Electric Technology Research Foundation of
Chugoku.

References
1. Takeya, K.: Photovoltaic System Case Study Guide; Lightning Protection and Design
Method, Design Technology Series, pp. 94–99 (2015)
2. Haeberlin, H.: Interference voltages induced by magnetic fields of simulated lightning
currents in photovoltaic modules and arrays. In: Proceedings of 17th EU-PVSEC, Munich,
pp. 2343–2346 (2001)
3. Haeberlin, H.: Damages at bypass diodes by induced voltages and currents in PV modules
caused by nearby lightning currents. In: Proceedings of 22th EU-PVSEC, Miran (2007)
4. Haeberlin, H., Kaempfer, M.: Measurement of damages at bypass diodes by induced
voltages and currents in PV modules caused by nearby lightning currents with standard
waveform. In: Proceedings of 23th EU-PVSEC, Valencia (2008)
5. Dhere, N.G., et al.: The reliability of bypass diodes in PV modules. In: Proceedings of SPIE
Solar Energy + Technology, San Diego, CA (2013)
6. Jiang, T., Grzybowski, S.: Electrical degradation of photovoltaic modules caused by
lightning induced voltage. In: Proceedings of IEEE EIC, Philadelphia, PA (2014)
7. Oke, S., et al.: Elucidation of overheating and firing processes of PV module which was
damaged by induced lightning. In: Proceedings of JSES/JWEA Joint Conference, vol. 79,
pp. 285–288 (2016)
8. Oke, S., Sakai, H., Tottori, H., Nanno, I., Hamada, T., Fujii, M., Ishikura, N.: Characteristics
and risks of broken bypass diode with induced lightning. In: Proceedings of Grand
Renewable Energy 2018 International Conference and Exhibition (GRE 2018), Yokohama,
p. a91154 (2018)
9. Oke, S., et al.: I-V characteristics of broken bypas diode on PV module. In: Proceedings of
35th EU-PVSEC, Brussel, pp. 1996–2000 (2018)
10. Japanese Industrial Standards: Electromagnetic Compatibility (EMC) – Part 4–5: Testing
and Measurement Techniques – Surge Immunity Test. Japanese Industrial Standards
Committee (2011)
11. Ishikura, N., et al.: Simulation analysis of really occurred accident caused by short circuit
failure of blocking diode and bypass circuit in the photovoltaics system. In: Proceedings of
IEEE Conference, ICRERA 2018, Paris, pp. 533–536 (2018)
 Relationship Between Electron Activation
Energy Absorption and Impulse Breakdown
Voltage in Polymers

Xiongwei Jiang1(&) , Wenxia Sima1 , Potao Sun1 ,

and Qingjun Peng2
1
State Key Laboratory of Power Transmission Equipment and System Security
and New Technology, Chongqing University, Chongqing, China
jiangxiongweicqu@163.com
2
Electric Power Research Institute of Yunnan Power Grid Corporation,
Kunming, China

Abstract. Impulse breakdown is a common phenomenon in power equipment

and electronic devices. To reveal impulse breakdown mechanism, a novel model
of activation energy absorption for trapped electrons is proposed based on the
mechanical vibration theory, considering the effect of trapped charges de-
trapping process on electrical breakdown. In the model, the frequency-
dependent dielectric properties, the frequency dependent absorption character-
istic and the frequency spectrum of the impulse voltage are considered. To
verify its reliability, the breakdown characteristics of polyethylene terephthalate
film (PET) under impulse voltages are tested. In the experiment, the influences
of different wave front/tail time and oscillation frequency on breakdown volt-
ages are obtained. The mechanism analysis of breakdown phenomenon based on
the proposed model is consistent with the experimental results. It is indicated
that the proposed model is suitable for explaining the phenomenon of break-
down dispersion and the breakdown voltage varies with waveform parameters.
This work provides a reference for the optimal design of insulation materials
with enhanced impulse breakdown voltage.

Keywords: Impulse breakdown
Activation energy absorption
Dispersion

Waveform parameters

1 Introduction

Polyethylene terephthalate (PET), polyethylene (PE) are the most popular insulating
materials for ultra-high voltage (UHV) electrical equipment [1, 2]. As the internal
insulation of electrical equipment, they are frequently affected by lightning and
switching transient overvoltage during decades-long operation, then surface flashover
and irreversible bulk breakdown phenomenon happen frequently [3]. It was also found
that the transient overvoltage that invading the electrical equipment sometimes has the
characteristics of damped alternating oscillation, defined as damped alternating oscil-
lation waveform (DAOW).

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1225–1234, 2020.
https://doi.org/10.1007/978-3-030-31676-1_114
1226 X. Jiang et al.

From 1922, to reveal the breakdown mechanism underlying the DC breakdown

process, several classic theories from the perspective of electronic [4], thermal
breakdown [5], and electromechanical breakdown [6] and local deterioration [7] have
been put forward. With the improvement understanding on the role of space charges
during electrical breakdown process, some investigations were conducted in this area
[8–10]. Extensive researches indicated that the accumulation of trapped charges pos-
sibly leads to two major consequences, (i) internal electric field distortion, initiating a
sharp increase in leakage current and (ii) charge transport process destroys molecular
chains or creates local defects. Despite much work has been done on DC breakdown
phenomena of polymers, the insulating performance of materials under impulse volt-
ages remains unclear, because of the complexity of this physical phenomenon and the
limited technology. Therefore, understanding the failure mechanism under transient
overvoltage is important to the stable operation of electrical equipment and the design
of materials.
With regard to electrical breakdown, it has been investigated that [9, 10] when the
electric strength in the materials exceeds the critical de-trapping field, it causes a rapid
de-trapping phenomenon happens, the local excess of internal polarization energy (5–
10 eV) then is released to the dielectric, resulting in the loss of insulating performance.
This theory is in well agreement with some electrical breakdown phenomena. How-
ever, in spite of the efforts, a clear analysis on the following interesting breakdown
phenomenon is highly required: the impulse breakdown strength dispersion, and the
breakdown voltage varies with waveform parameters, such as wave front/tail time and
oscillation frequency. Researchers are used to measuring the dielectric properties (er
and tand) of materials, but the underlying quantitative relationship between dielectric
properties and activation energy absorption characteristics is neglected.
This paper deals with breakdown phenomena in polymers under impulse voltages.
In particular, we focused on the de-trapping process of trapped electrons, which is
closely associated with the irreversible breakdown. Combined with the measured
dielectric characteristics and breakdown strength results, a quantitative model of acti-
vation energy absorption applicable to insulating polymers is proposed. We discussed
the differences in activation energy obtained from different double-exponential voltages
and DAOWs. In short, the effects of waveform parameters on breakdown strength are
perfectly explained.

2 Materials and Methods

2.1 Preparation of Materials

The material involved in this paper is PET film. The thickness of the film is
100 ± 5 lm. PET particles in molten state were hot-pressed with 15.0 MPa at 120 °C
for 30 min using the flat vulcanizing press. To ensure the reliability of samples, the
surfaces of the sample were scrubbed with acetone and were dried in a vacuum oven at
50 °C for 48 h. Some samples used for dielectric spectroscopy were evaporated with
Au electrodes with a diameter of 20 mm on both sides.
 Relationship Between Electron Activation Energy Absorption 1227

2.2 Characterization
In the dielectric spectroscopy test, the measurement of dielectric relaxation spec-
troscopy was performed at the temperature of 293 K, the measuring dielectric spec-
troscopy was conducted from 10−2 Hz to 106 Hz at an AC driving voltage of 1 V,
using a Novocontrol broadband dielectric spectrometer.
The whole impulse breakdown procedure was carried out at 293 K in a Plexiglas
test cell, as shown in Fig. 1. Figure 2 exhibits the specific waveforms and parameters
of double-exponential impulse voltage and damped alternating oscillation waveforms.

High Voltage Divider

Impulse Generator
System
Test Cell

D Rf

C0 Rt

g2
Protective
Resistance

Control System Oscilloscope

Fig. 1. Sketch of double-exponential impulse voltage and damped alternating oscillation

waveforms (DAOWs) breakdown test setup.

1.0 1.0
Damped Alternating Oscillation Waveform

0.8
Wave-front Wave-tail 0.5 1/f
Time Time
Amplitude

Amplitude

0.6
0.0
0.4 Double-exponential Impulse Voltage

-0.5
0.2 Wave-front
Time
(a) (b)
0.0 -1.0
0.0 5.0x10-4 1.0x10-3 1.5x10-3 2.0x10-3 2.5x10-3 0.0 2.0x10-4 4.0x10-4 6.0x10-4
Time (s) Time (s)

Fig. 2. Oscillograms of (a) Double-exponential impulse voltage and (b) Damped alternating
oscillation waveforms (DAOWs).
1228 X. Jiang et al.

3 Results and Discussion

3.1 Dielectric Spectroscopy Results
As important parameters of the activation energy absorption model, the frequency-
dependent dielectric characteristics of single-layered PET film are measured at 293 K,
which are illustrated in Fig. 3.

3.25 10-1
Relative Dielectric Constant
Dielectric Loss
Relative Dielectric Constant εr

3.20

Dielectric Loss Tanδ

10-2
3.15

3.10
10-3

3.05

3.00 10-4
10-2 10-1 100 101 102 103 104 105 106
Frequency (Hz)

Fig. 3. Frequency dependence of permittivity and loss tangent of single-layer PET film.

At low frequencies, the permittivity remains at a high level. As the frequency

increases, the dielectric constant decreases obviously, which is caused by different
polarization types depending on the frequency range [11, 12]. At low frequencies, the
enhanced dielectric permittivities mainly come from space charge polarization (inter-
facial polarization). Meanwhile, all the free polarizable functional groups in the PET
chains at low frequency have enough time to orient themselves. As the frequency
increases, due to the orientation time of dipole is usually longer than space charge
polarization, the orientational polarization is weakened, so the permittivity at high
frequency decreases obviously. Besides, the dielectric relaxation spectroscopy of PET
exhibits two dielectric loss peaks, which is broader than the Debye curve and indicates
the existence of more than one relaxation process. The first peak (at 10−2 Hz) is related
to the interfacial polarization, which is expected to occur at the lower frequency range.
The second peak (at about 106 Hz) can be attributed to the dipolar relaxation, which is
usually observed in the range of 1 Hz to 100 MHz [13].

3.2 Impulse Breakdown Results

In subsequent sections, the insulating performance of PET film under different impulse
voltages was investigated. The breakdown results were processed within the framework
of the Weibull statistical method, which reflects the breakdown voltage of the sample at
specific breakdown probability (Ft).
 Relationship Between Electron Activation Energy Absorption 1229

1.0 1.0
160 Hz, a=100 6.8 kHz, a=6500
0.5 750 Hz, a=100 0.5 6.8 kHz, a=4500
750 Hz, a=400 6.8 kHz, a=1700
0.0 0.0

ln(ln(1/(1-Ft)))
ln(ln(1/(1-Ft)))

-0.5 -0.5

-1.0 -1.0

-1.5 -1.5

-2.0 -2.0

-2.5 -2.5
(a) (b)
-3.0 -3.0
2.70 2.75 2.80 2.85 2.65 2.70 2.75 2.80 2.85
lnU lnU

Fig. 4. Relationship between breakdown voltage and failure probability of PET film under
different DAOWs. Breakdown occurs before the first peak-voltage. (a) Oscillation frequency is
less than 750 Hz; (b) Oscillation frequency is 6.8 kHz with diverse damping coefficient a.

Figure 4 indicates the relationship between breakdown voltage and failure proba-
bility of PET film under different DAOWs. DAOW with diverse oscillation frequency
f and damping coefficient a were used to study the effect of waveform parameters on
breakdown voltages. Both Figs. 4a and b show that breakdown occurs before the first
peak-voltage time. When oscillation frequency is less than 750 Hz, the breakdown
voltages at 100 Hz and 750 Hz are about 15.8 kV (Ft = 63.2%), thus, the frequency
has little effect on breakdown voltage. Nevertheless, a slight enhancement in break-
down strength (16.6 kV, Ft = 63.2%) is observed when coefficient a increases. From
Fig. 4b, as coefficient a increases, intriguingly, the breakdown voltage at 6.8 kHz also
increases (increases from 15.5 kV to 16.7 kV, Ft = 63.2%).
In addition, the breakdown data of PET under double-exponential impulse voltages
(lightning and switching impulse) was summarized in Table 1. It can be concluded that
the effect of wave-front time and wave-tail time on breakdown strength is exactly
opposite. The increase in wave-front time is beneficial to the improvement of break-
down strength, but the increase in wave-tail time has little influence on impulse
insulating performance. Detailed analysis will be carried out in subsequent sections.

Table 1. Breakdown voltages at Ft = 63.2% failure probability of PET films under different
double-exponential impulse waveforms.
Double-exponential impulse Waveform front/tail time Breakdown voltages
Lightning impulse 1.2/50 ls 16.4 kV
1.2/100 ls 16.3 kV
6/50 ls 17.0 kV
Switching impulse 250/2500 ls 17.1 kV
160/2500 ls 16.5 kV
1230 X. Jiang et al.

4 Mechanism Analysis
4.1 Activation Energy Absorption Characteristic
To further investigate the above interesting breakdown phenomenon, an improved model
of activation energy absorption in insulating polymers is proposed based on the Harmonic
oscillator model. Our basic assumption is that electrical breakdown occurs when acti-
vation energy absorbed by trapped electrons plus the reduction in trap energy DE is larger
than the initial trap level ET. That is, DE + WA> ET. Subsequently, based on the well-
known theory of Harmonic oscillator model, the vibration differential equation of the
system which contains electron and trap center is established, which is expressed as:

m x00 þ cx0 þ kx ¼ eE0 cos x1 t ð1Þ

where m* is the variable mass of a trapped electron, as a function of applied voltage

[14]. x00 is the acceleration of forced vibration electron. c = 2 m*nx0, n is damping
coefficient, according to the vibration theory, it is set to be 0.1. x0 is intrinsic fre-
quency, which is set equal to 109 rad/s. −kx is the elastic force, x is the distance from
trap center. e is elementary charge (e = 1.6  10−19 C), E0 is the electric field
amplitude. x1 is the driving frequency (rad/s) determined by the interceptive voltage
(waveform recorded by oscilloscope during breakdown).
As for the energy spectrum of impulse voltage, firstly, the mathematical expression
of impulse voltage was introduced, and the equivalent average electric field distribution
in the sample under double-exponential impulse voltage is described as:

EðtÞ ¼ k
Ep ðelt  eht Þ ð2Þ

where k is the correction parameter of amplitude, Ep is the equivalent average electric

field amplitude, l controls the wave-tail time, and h is related to the wave-front time,
respectively. In case of DAOWs, the mathematical expression of DAOW is expressed as
n o
EðtÞ ¼ Ep qt
h1 ðtÞ þ h2 ðtÞ
cos½x1 ðt  t1 Þeaðtt1 Þ ð3Þ

where h1 ðtÞ ¼ hðtÞ  hðt  t1 Þ, h2 ðtÞ ¼ hðt  t1 Þ  hðt  t2 Þ, h(t) is jump function, q is
wave-front steepness (V/s), t1 is wave-front time (s), t2 is the actual continuous time
(breakdown time), a is damping coefficient. Notably, the final energy distribution of
interceptive voltage is determined by the values of t1 and t2. The Continuous Fourier
Transform (CFT) process of different impulse voltages was dealt with Matlab program.
In the forced vibration system, in a cycle T, the total energy WP of a trapped
electron obtained from the external field is calculated by
Z T
WP ¼ EðtÞ
P0 dt ð4Þ
0

where P0 is the electric dipole moment P ¼ e
xðtÞ partial differentiation of t. In our

first attempt, it is thought that not all energy is converted into activation energy
 Relationship Between Electron Activation Energy Absorption 1231

(thermal energy) of trapped electron, so the dielectric loss is introduced into the model.
Considering the impulse voltage contains ample equivalent-frequencies, and based on
the Plancherel’s formula of Continuous Fourier Transform (CFT), the total activation
energy WA is shown in the integral form:
Z þ1
jEðix1 Þj2 e2 nx1 x0
WA ¼ 

tan dðx1 Þ dx1 ð5Þ
0 m ðx0  x1 Þ2 þ ð2nx1 x0 Þ2
2 2

where jEðix1 Þj2 is the energy spectrum of waveform E(t). E(t) represents the inter-
ceptive impulse voltage.
In addition, to reveal the breakdown mechanism underlying the DC breakdown
process, a percolation model has been put forward [9, 10], which is mainly described as
the presence of an equivalent average electric field E can modify the initial trap level,
and it is reduced by

DE ¼ bpf E00:5 ¼ ðe3 =peÞ0:5 E00:5 ð6Þ

where bpf is the Poole-Frenkel coefficient and e is the dielectric permittivity of test
sample. Then, the modified trap level EMT is expressed as:

EMT ¼ ET  DE ð7Þ

where ET is the initial trap level. It can be concluded from Eqs. (5) and (6), the
variation of breakdown voltage is closely related to the frequency-dependent er and
tand. When the electric field is high enough (WA > EMT), the electrons in traps are able
to escape from local sites (traps) by thermally assisted hopping process. Electrons
accelerate during the transfer process between the trap sites, finally leads to the elec-
trical breakdown.

4.2 Application of the Proposed Model

To investigate the feasibility of the model, the calculated activation energy WA and the
reduction in trap energy DE under different impulse voltages were summarized in
Tables 2, 3, respectively. Results in Table 2 is related to the DAOWs, and Table 3 is
associated with the double-exponential impulse voltage. In Tables 2 and 3, the wave-
front time t1 and the actual continuous time t2 are set corresponding to the 63.2%
breakdown voltage point. To set the critical breakdown criterion, the deep trap level ET
of PET is generally in the range of 0.82–0.92 eV according to our previous study [1].
For ease of analysis, the critical breakdown criterion (initial trap level) is set to 0.87 eV
in Eq. (7). Although in this case, the calculated breakdown strength would be slightly
different, it does not affect the analysis of breakdown dispersion and the effect wave-
form parameters on breakdown voltage.
From Table 2, the breakdown voltages vary with f and coefficient a are well
explained in Fig. 4. Meanwhile, the calculated results in Table 3 successfully reveal the
breakdown mechanism underlying the impulse voltages with different wave-front time
and wave-tail time.
1232 X. Jiang et al.

Table 2. WA and DE as a function of DAOW parameters. DE + WA= 0.87 eV. The sample is
PET. The calculated critical breakdown electric field Ec corresponds to a breakdown probability
of 63.5%.
Ec(V/m) t1(ls) t2(ls) f(Hz) a WA(eV) ΔE(eV)
1.78  108 6.8 6.0 160 100 0.2912 0.5794
1.78  108 6.7 6.0 750 100 0.2949 0.5759
1.81  108 6.7 6.3 750 400 0.3046 0.5659
1.73  108 6.4 5.2 6800 6500 0.2758 0.5958
1.71  108 6.2 4.8 6800 4500 0.2641 0.6070
1.68  108 5.5 4.0 6800 1700 0.2498 0.6214

Table 3. WA and DE as a function of double-exponential impulse voltages. DE + WA= 0.87 eV.

The sample is PET. The calculated critical breakdown electric field Ec corresponds to a
breakdown probability of 63.5%.
Voltage type Ec (V/m) Front t1 (ls) Tail (ls) WA (eV) ΔE (eV)
Lightning impulse 1.52  108 1.2 50 0.3419 0.5311
1.58  108 5 50 0.3302 0.5406
1.51  108 1.2 75 0.3414 0.5293
1.62  108 7 75 0.3279 0.5456
Switching impulse 1.91  108 250 2500 0.2895 0.5822
1.87  108 160 2500 0.3024 0.5699

In addition, the dispersion of impulse breakdown voltages under impulse voltage is

also a valuable phenomenon that needs to be concerned about. It should be pointed out
that the irreversible breakdown may occurred at 100 to 250 ls under standard witching
impulse voltage [15]. Such a breakdown dispersion problem also occurs in lightning
voltages and DAOWs.

Table 4. WA and DE as a function of breakdown time t2 under standard witching impulse

voltage. DE + WA= 0.87 eV.
E0 (V/m) Breakdown time t2 (ls) WA (eV) ΔE (eV) WA + ΔE (eV)
1.85  108 100 0.2894 0.5803 0.8697
1.86  108 120 0.2896 0.5811 0.8707
1.90  108 160 0.2830 0.5882 0.8712
1.91  108 180 0.2816 0.5887 0.8703
1.93  108 200 0.2816 0.5906 0.8722

The calculated results shown in Table 4 indicated that, although the critical
breakdown strength is different, the final WA+ ΔE is the same. It is demonstrated that
the electric strength is not the decisive factor causing the trapped charges to escape
 Relationship Between Electron Activation Energy Absorption 1233

from the traps, so the energy of impulse voltage cannot be simply calculated without
considering the frequency characteristics of electronic vibration. Therefore, the
breakdown can occur under different electric fields, but the corresponding breakdown
time t2 is different. For ease of analysis, the activation energy absorption characteristic
under the field of 1  108 V/m under standard switching impulse at different break-
down time t2 was investigated, the results are shown in Fig. 5. It can be seen from
Fig. 5, as the breakdown time t2 decreases, the activation energy distribution frequency
is more extensive. The actual activation energy calculated by Eq. (5) indicates that the
amount of actual activation energy increases as the breakdown time decreases. Besides,
the reduced trap level calculated by Eq. (6) also increases as the breakdown time
decreases. Therefore, the impulse breakdown can occur at different t2, leads to the
dispersive breakdown voltage. To sum up, the validity and accuracy of the proposed
model are confirmed.

1.8x10-6
t2=100 μs
-6 t2=120 μs
1.5x10
Activation Energy WA (eV)

t2=160 μs
-6 t2=180 μs
1.2x10
t2=200 μs
-7
9.0x10

6.0x10-7

3.0x10-7

0.0
102 103 104 105 106
w1 (Rad/s)

Fig. 5. The actual activation energy absorption characteristics under the field of 1  108 V/m
under standard switching impulse at different breakdown time t2.

5 Conclusions

In the present work, the impulse breakdown characteristics of polyethylene tereph-

thalate film were investigated. A quantitative model of activation energy absorption for
trapped electrons was proposed to reveal the breakdown characteristics. According to
the numerical calculation, the forced vibration characteristics of trapped electrons vary
with the applied frequencies reveal that the frequency-dependent energy absorption
characteristics should be taken into account in the activation energy absorption process
rather than simply calculating the original energy of impulse voltage. Considering the
influence of Poole-Frenkel effect and activation energy absorption characteristic,
improved dielectric constants and lower dielectric loss are required for the design of
materials with excellent insulating performance at high frequency. The practicability of
1234 X. Jiang et al.

the proposed model is demonstrated by the breakdown results. Our investigation

provides a quantitative method for analyzing the breakdown strength and its dispersion
of insulating materials under impulse voltages.

Acknowledgments. This work was financially supported by the National Basic Research
Program of China (973 Program) (2015CB251003), the China Scholarship Council (CSC) Fund,
the National Key Basic Research Program of China (2017YFB0902702), the State Key Pro-
gram of National Natural Science of China (51837002). We are grateful for their support.

References
1. Jiang, X., Sima, W., Peng, Q., Yang, M., Sun, T.: Effect of thermal ageing on space charge
characteristics in double-layered polyester film. IEEE Trans. Dielectr. Electr. Insul. 23(5),
3156–3164 (2016)
2. Yin, K., Zhou, Z., Schuele, D.E., Wolak, M., Zhu, L.: Effects of Interphase modification and
biaxial orientation on dielectric properties of poly (ethylene terephthalate)/poly (vinylidene
fluoride-co-hexafluoropropylene) multilayer films. ACS Appl. Mater. Interfaces 8(21),
13555 (2016)
3. Du, N., Guan, Y., Zhang, J., Niu, J., Yao, S.: Phenomena and mechanism analysis on
overvoltages caused by 40.5-kV vacuum circuit breakers switching off shunt reactors. IEEE
Trans. Power Deliv. 26(4), 2102–2110 (2011)
4. Fröhlich, H., Simpson, J.H.: Intrinsic dielectric breakdown in solids. Adv. Electr. Electr.
Phys. 2(1), 185–217 (1950)
5. Wagner, K.W.: The physical nature of the electrical breakdown of solid dielectrics. Trans.
Am. Inst. Electr. Eng. 41(12), 1034–1044 (1922)
6. Zebouchi, N., Malec, D.: Combination of thermal and electromechanical breakdown
mechanisms to analyze the dielectric breakdown in polyethylene terephthalate. J. Appl.
Phys. 83(11), 6190–6192 (1998)
7. Nagata, T., Shimizu, N.: Deterioration in aged XLPE cables. In: IEEE Conference on
Electrical Insulation and Dielectric Phenomena CEIDP: 579 (1992)
8. Li, S., Zhu, Y., Min, D., Chen, G.: Space charge modulated electrical breakdown. Sci.
Rep. 6, 32588 (2016)
9. Blaise, G.J.: The space charge physics and the breakdown process. Appl. Phys. Lett. 77(7),
2916–2927 (1995)
10. Wu, K., Dissado, L.A.: Percolation model for electrical breakdown in insulating polymers.
Appl. Phys. Lett. 85(9), 4454–4456 (2004)
11. Deligöz, H., Özgümüş, S., Yalçınyuva, T., Yıldırım, S., Deǧer, D., Ulutaş, K.: A novel
cross-linked polyimide film: synthesis and dielectric properties. Polymer 46(11), 3720–3729
(2005)
12. Elimat, Z.M., Hamideen, M.S., Schulte, K.I.: Dielectric properties of epoxy/short carbon
fiber composites. J. Mater. Sci. 45(19), 5196–5203 (2010)
13. Zhu, L.: Exploring strategies for high dielectric constant and low loss polymer dielectrics.
J. Phys. Chem. Lett. 5(21), 3677–3687 (2014)
14. Jadidian, J., Zahn, M.: Charge transport analysis in two-phase composite dielectric systems.
IEEE Trans. Plas. Sci. 41(9), 2464–2474 (2013)
15. Sima, W., Jiang, X., Peng, Q., Sun, T.: Improved model of activation energy absorption for
different electrical breakdowns in semi-crystalline insulating polymers. J. Phys. D Appl.
Phys. 51, 215301 (2018)
 Characterization of Electric Fields Produced
by Preliminary Breakdown Pulses Observed
in Bogotá, Colombia

C. A. Granados1(&), H. E. Rojas2, C. A. Rivera1, and F. J. Román1

1
Universidad Nacional de Colombia, Bogotá Campus, Bogotá, Colombia
cagranadosv@unal.edu.co
2
Universidad Distrital Francisco José de Caldas, Kra. 7 # 40-53,
Bogotá, Colombia
herojasc@udistrital.edu.co

Abstract. In this paper, the characteristics of electric fields due to preliminary

breakdown pulses (PBP) that precede the first return strokes (FRS) are pre-
sented. These pulses exist due to the electrical activity formed inside the clouds
prior to the occurrence of the stepped leader. The electric field signatures were
recorded in Bogotá, Colombia with an altitude of 2550 m. For the measurement
campaign, a broadband parallel-plate antenna system with a 500 ms full
observation window and a pre-trigger of 250 ms were used. The measurements
were carried out during two rainy seasons recording 180 waveforms. From
recorded data, it was found that 37.5% of negative cloud-to-ground (CG) flashes
exhibited PBP. With respect to the PBP trains a duration between 0.4 ms and
22.3 ms was observed. For the same temporal parameter, the arithmetic mean
(AM) and the geometric mean (GM) were 4.7 ms and 3.1 ms, respectively. On
the other hand, the GM of the time interval between the largest peak of the PBP
train and the peak value of the FRS was 10.1 ms, while its AM was 27.9 ms. In
addition, the GM of the ratio between the maximum value of the PBP train and
its respective FRS was 0.51, with changes between 0.12 (minimum) and 2.1
(maximum). Finally, the number of individual pulses per train between 4 and 39.

Keywords: Electric field
First return stroke
Individual pulses
Negative

cloud-to-ground flashes
Preliminary breakdown pulses
Total pulse train
duration

1 Introduction

Cloud-to-ground (CG) lightning flashes are transient discharges of high current that are
produced between clouds and the surface of the planet. According to the direction in
which the charges move and their polarity, these electrical discharges can be classified
in downward negative lightning, upward negative lightning, downward positive
lightning and upward positive lightning.
It is considered that upward flashes are less common than downward ones. The
upward lightning discharges are related with the presence of tall structures or objects

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1235–1243, 2020.
https://doi.org/10.1007/978-3-030-31676-1_115
1236 C. A. Granados et al.

with moderate height located on top of mountains. According to this, the negative
downward lightning flashes represent more than 90% of global CG discharges, while
10% or less are positive lightnings [1]. Due to this behavior and its high percentage of
occurrence, this work focuses on the study and characterization of negative CG flashes.
Several parameters of CG lightnings (electric charge transfer, current, frequency
spectrum) used for engineering applications can be obtained from the measurements of
electromagnetic fields radiated during the flash occurrence [1, 2]. The knowledge of the
lightning phenomena and its parameters is important to develop lightning discharge
models, design lightning protection systems and determine electromagnetic compati-
bility requirements for electric and electronic devices, among others.
The electric field waveform produced by a CG lightning flash sometimes reveals the
presence of a pulse train composed by individual discharges with a duration of some
microseconds, which precedes the first return stroke (FRS). This pulse train is known as
the preliminary breakdown pulses (PBP) and its features have been analyzed in different
regions of the world [3, 4]. The PBP is produced inside the cloud by a vertical discharge
between the negative charge packet and the lower positive charge packet [2, 5]. In other
words, the preliminary breakdown leads to the initiation of the downward moving
negative stepped leader. For this reason, PBP can provide information about the first
event that produces the electrical breakdown inside the clouds [6, 7].
Studies about the PBP processes in different geographical conditions can provide a
better understanding of the lightning process [2, 7, 8]. However, there are few studies
conducted in tropical regions (Sri Lanka, Malaysia and Indonesia). This work presents
the characteristics of the PBP observed in Bogotá, Colombia (tropical region located
above 2500 m of altitude) using signatures of the electric field radiated by negative CG
lightning flashes. This study is the first research carried out on this subject in the
country using more than 40 electric field signatures.

2 Instrumentation

The measuring system used in this work is composed by a 30 cm diameter parallel-

plate antenna, which is connected to an electronic circuit based on the buffer BUF602
by means a coaxial cable RG58U (0.6 m). The signals from the electronic circuit were
fed into a digital oscilloscope using a 12 m longitude RG-58/U coaxial cable, properly
terminated to avoid reflections. The general scheme used to measure lightning-
generated electric fields and the electronic circuit are shown in Figs. 1 and 2, respec-
tively. The elements that compose the electronic circuit are R1 ¼ 50 X, C ¼ 270 pF,
R ¼ 100 MX, Rb ¼ 100 X, Cb ¼ 0:1 lF; Cv ¼ 15 pF  230 pF, RO ¼ 48 X.
 Characterization of Electric Fields Produced by PBP 1237

Fig. 1. Scheme of the electric field measuring system. (A) parallel-plate antenna; (B) short
coaxial cable; (C) electronic circuit; (D) long coaxial cable; (E) digital oscilloscope.

Fig. 2. Electronic circuit used in the electric field measuring system [9].

The electronic circuit was installed inside a shielded cabinet and it was fed by a
12 V-DC battery bank. The bandwidth of the measuring system is 11 MHz. The
oscilloscope was configured with a 500 ms full observation window and a sampling
time of 100 ns. In addition, in order to acquire the signals before and after the trigger
transient pulse, a 250 ms pre-trigger was adjusted. The measuring system was located
on the roof of the building B5 of the Camilo Torres complex at the Universidad
Nacional de Colombia Campus. The approximate height between the ground plane and
the roof of the building is 15 m. The location of the antenna and the electronic circuit is
shown in Fig. 3.
1238 C. A. Granados et al.

Fig. 3. Location of the antenna and the electronic circuit on the roof of the building.

3 Methodology

The electric field signatures of PBP trains that precede the FRS were recorded in two
rainy seasons: August–September 2016 and October–December 2017. The pulses
trains were produced by negative CG lightning flashes in Bogotá, which is located on
the central region of Colombia (latitude: 4.641°N, longitude: 74.091°W and 2550 m
above sea level). From the measurement campaigns, 180 negative CG flashes were
recorded and, after to examine the electric field data, 68 signatures (37.5%) exhibited
PBP train. The remaining waveforms were not included in this study.
Before to characterize the PBP pulses, the signal-to-noise ratio (SNR) and the noise
content of the electric field signatures were analyzed. In this way, for the 180 wave-
forms the average of the SNR was 7.9 dB with a standard deviation of 2.6 dB. In
addition, the ratio between the maximum value of the noise and the peak value of the
first return stroke (Noise/FRS ratio) has an arithmetic mean (AM) of 0.09. These results
show that the noise content in the recorded signals does not affect significantly the
identification of the signal parameters.
The methodology applied to characterize the PBP trains is similar to that used by
Baharudin et al. [5], Wu et al. [10], and Johari et al. [11]. Thus, the criteria used for
analyze the waveforms of the PBP trains are defined as follows:
• Only were included pulses with amplitude larger than twice the average noise level.
• The individual pulses considered inside a PBP train where those separated by less
than 2 ms between them.
• Only pulse trains with at least three individual pulses were analyzed.
The pulses that compose a PBP train are mainly bipolar (usually from the beginning
up to the middle of the train). However, at the end of the train the pulses can be
unipolar [3, 12]. Figure 4 shows an example of the electric field signature produced by
 Characterization of Electric Fields Produced by PBP 1239

a negative CG lightning flash recorded in Bogotá. In this figure, it is possible to identify

the PBP train the FRS and two subsequent return strokes (SRS).

Fig. 4. Complete electric field signature produced by a negative CG flash recorded in Bogotá
(Colombia). Data obtained in November 16th 2017 at 18:08:06

On the other hand, Fig. 5 shows the PBP train and the FRS of the CG flash
presented in Fig. 4. In this way, each PBP frequently starts with one or two small
pulses, followed by some largest pulses, and it ends with pulses of decreasing
amplitudes [10]. In addition, it can be seen that the pulses are clearly bipolar between
−2.1 ms and −1 ms, while between −1 ms and −0.7 ms the pulses are unipolar until
the end of the pulse train.
In order to describe the characteristics of each PBP train, four parameters were
selected: (a) the pulse train duration (time between the first and the last detectable
pulse); (b) the time interval between the largest peak of the PBP train and the maximum
value of the FRS (PBP-FRS); (c) the ratio between the peak value of the PBP train and
its FRS (PBP/FRS); (d) the number of individual pulses in each train [5, 12]. Figure 6
shows the parameters analyzed for the each PBP. For the signature presented in Fig. 5,
the pulse train duration is 1.25 ms, the PBP-FRS separation is 1.59 ms, the PBP/FRS
ratio is 0.72 and the number of pulses was 16.

Fig. 5. Signature of the PBP train and the FRS for the negative CG flash shown in Fig. 4
1240 C. A. Granados et al.

Fig. 6. Parameters identified in the signature of the PBP train and the FRS

4 Results and Discussion

The characterization of the 68 PBP trains were performed using statistics parameters
such as geometric mean (GM), arithmetic mean (AM), minimum value (Min) and
maximum value (Max). Table 1 summarizes the statistical results for each parameter
included in this study. In addition, to analyze the behavior of the parameters mentioned
in Sect. 3, statistical results are presented between Figs. 7 and 10 using histograms.
Figures 7 and 8 show the results for the pulse time duration and the distribution of
PBP–FRS separation, respectively. It can be seen in Fig. 7 that 46 of 68 signatures
(67.6%) present a PBP duration less or equal to 4 ms, while the remaining waveforms
(12) exhibit a duration between 4 and 23 ms approximately. For this parameter, the
GM is 3.1 ms and the AM is 4.7 ms, with individual values from 0.41 ms up to
22.3 ms (Min and Max values).
With respect to Fig. 8, it is possible to observe that major part of the signatures
(42–61.8%) present a PBP-FRS separation between 0 and 20 ms, while 8 CG flashes
(11.7%) exhibited a time interval between the PBP train and its FRS greater than
100 ms. For this parameter, the AM and the GM are 27.9 ms and 10.1 ms, respec-
tively. The minimum value of PBP-FRS was 0.26 ms, while the maximum was
170.76 ms.

Table 1. Summary of the statistical results obtained from the characterization of PBP trains.
Statistical Parameter of the PBP
Total train duration (ms) PBP-FRS (ms) PBP/FRS Pulses per train
Min 0.41 0.26 0.12 4
Max 22.3 170.76 2.08 39
GM 3.09 10.1 0.51 11.4
AM 4.74 27.9 0.61 13.8
 Characterization of Electric Fields Produced by PBP 1241

50
GM: 3.1 ms

Number of flashes
40
AM: 4.7 ms
30 Min: 0.41 ms
Max: 22.3 ms
20

10

0
0--4 4--8 8--12 12--16 >16
Total pulse train duration (ms)

Fig. 7. Distribution of the PBP train duration

40 GM: 10.1 ms
Number of flashes

AM: 27.9 ms
30 Min: 0.26 ms
Max: 170.76 ms
20

10

0
0--20 20--40 40--60 60--80 >80
PBPóFRS separation (ms)

Fig. 8. Distribution of the PBP-FRS separation

The statistical distribution of the PBP/FRS ratio for the CG flashes observed in
Bogotá is depicted in Fig. 9. The results show that 58 (85.3%) signatures present a ratio
less than 1. In addition, it was found that 10 CG flashes (14.7%) presented at least one
bipolar pulse with a magnitude that exceeds the peak value of the FRS. The AM and
the GM of the PBP/FRS ratio were 0.61 and 0.51, respectively. In addition, for the
measurements analyzed, this parameter varies from 0.12 (Min) up to 2.08 (Max).
The histogram shown in Fig. 10 provides the distribution of the number of indi-
vidual pulses per train. It is possible to observe that 48 negative CG flashes (70.6%)
present up to 16 pulses per train, while only 4 signatures (5.8%) exhibit more than 32
pulses. In addition, the GM for the number of pulses is 11.4 and its AM is 13.8. The
minimum and the maximum values for this parameter were 4 and 39, respectively.

25 GM: 0.51
Number of flashes

AM: 0.61
20 Min: 0.12
15 Max: 2.08

10
5
0
0--0.25 0.25--0.5 0.5--0.75 0.75--1 1--1.25 >1.25
PBP/RS

Fig. 9. Distribution of the PBP/FRS ratio

1242 C. A. Granados et al.

25

Number of flashes
GM: 11.38
20 AM: 13.79
Min: 4
15 Max: 39
10

0
0--8 8--16 16--24 24--32 >32
Number of pulses

Fig. 10. Distribution of individual pulses per train

5 Conclusions

In this paper, the relation between the electric field waveform of preliminary break-
down pulse (PBP) and the first return stroke (FRS), produced by negative CG lightning
flashes in Bogotá, Colombia, has been analyzed. During the examination of the electric
field data, it was observed that some factors as the pre-trigger time, the noise level, and
relatively small PBP amplitudes could affect the adequate identification of the trains.
Waveform characteristics of the PBP are similar between signatures. The train is
composed by pulses that are mainly bipolar, generally starts with one or two small
pulses, followed by some largest pulses, and it ends with few pulses of decreasing
amplitudes (in this phase are usually unipolar). Considering the characteristics
described above it is possible differentiate the PBP signatures from those produced by
stepped leaders. However, in some cases it was complicated determine the point where
the PBP ends and where the step leader.
The statistic results obtained in this work show that PBP-FRS time interval varies
from 0.26 ms to 170.76 ms, while the AM and GM of the PBP/FRS ratio are 0.61 and
0.51, respectively. In addition, the total pulses train duration varies from 0.4 ms up to
22.3 ms. Despite these results, additional studies are needed in order to analyze (in a
complete way) the characteristics of PBP trains in Bogotá and other regions of
Colombia. These studies must include other parameters of the waveform, such as the
individual pulse duration and the inter-pulse duration.
This work is one of the first research conducted about PBP in Colombia. In order to
make a comparison between the results presented here and other works conducted
around the world, it is necessary to carry out additional measurement campaigns. From
a new set of data could be possible find a seasonal dependence of the PBP parameters
for lightnings occurred in mountainous regions.

Acknowledgments. H. E. Rojas express his gratitude to Universidad Distrital Francisco José de

Caldas for its support through the doctoral commission with contract code N° 0002-2016. In
addition, the authors would like to thank Prof. Mahendra Fernando PhD from University of
Colombo, Sri Lanka for his advice in the preliminar analysis of the measurements obtained in
Colombia during 2016.
 Characterization of Electric Fields Produced by PBP 1243

References
1. Rakov, V.A.: Lightning phenomenology and parameters important for lightning protection.
In: IX International Symposium on Lightning Protection SIPDA, November 2017, pp. 1–24
(2007)
2. Salimi, B., Mehranzamir, K., Abdul-Malek, Z.: Statistical analysis of lightning electric field
measured under equatorial region condition. Procedia Technol. 11, 525–531 (2013)
3. Arshad, S.N., et al.: Characterization preliminary breakdown in the measured lightning
electric fields. In: Proceedings of 2014 IEEE 8th International Power Engineering and
Optimization Conference, PEOCO 2014, March 2014, pp. 435–438 (2014)
4. Hazmi, A., Emeraldi, P., Hamid, M.I., Takagi, N.: Some characteristics of multiple stroke
negative cloud to ground lightning flashes in Padang. Int. J. Electr. Eng. Inform. 8, 438–450
(2016)
5. Baharudin, Z.A., Ahmad, N.A., Fernando, M., Cooray, V., Mäkelä, J.S.: Comparative study
on preliminary breakdown pulse trains observed in Johor, Malaysia and Florida, USA.
Atmos. Res. 117, 111–121 (2012)
6. Baharudin, Z.A., Fernando, M., Ahmad, N.A., Mäkelä, J.S., Rahman, M., Cooray, V.:
Electric field changes generated by the preliminary breakdown for the negative cloud-to-
ground lightning flashes in Malaysia and Sweden. J. Atmos. Solar-Terr. Phys. 84–85, 15–24
(2012)
7. Qie, X., Yu, Y.: Characteristics of cloud-to-ground lightning in Chinese Inland Plateau.
J. Meteorol. Soc. Japan 80, 745–754 (2002)
8. Arshad, S.N.M., Kadir, M.Z.A., Izadi, M.: Characterization of measured lightning electric
fields observed in Malaysia. In: 2014 International Conference on Lightning Protection,
October 2014, pp. 1058–1063 (2014)
9. Rojas, H.E., Rivera, C.A., Chaves, J., Cortés, C.A., Román, F.J., Fernando, M.: New circuit
for the measurement of lightning generated electric fields. In: 2017 International Symposium
on Lightning Protection, XIV SIPDA 2017, October 2017, pp. 188–194 (2017)
10. Wu, T., et al.: Preliminary breakdown pulses of cloud-to-ground lightning in winter
thunderstorms in Japan. J. Atmos. Solar-Terr. Phys. 102, 91–98 (2013)
11. Johari, D., Cooray, V., Rahman, M., Hettiarachchi, P., Ismail, M.M.: Characteristics of
preliminary breakdown pulses in positive ground flashes during summer thunderstorms in
Sweden. Atmosphere (Basel) 7(3), 1–18 (2016)
12. Wooi, C.L., Abdul-Malek, Z., Ahmad, N.A., Mokhtari, M.: Characteristic of preliminary
breakdown preceding negative return stroke in Malaysia. In: 2015 IEEE Conference on
Energy Conversion, CENCON 2015, pp. 112–115 (2016)
 Duality-Based Potential Transformer
Model Including Black-Box Circuit
for High-Frequency Transient Simulation

Daixiao Peng1(&), Ming Yang1, Wenxia Sima1, Jinwei Chu2,

Zhicheng Xie2, and Yonglai Liu1
1
State Key Laboratory of Power Transmission Equipment and System Security
and New Technology, Chongqing University, Chongqing 400044, China
pengdaixiao@cqu.edu.cn
2
Maintenance and Test Center of Extra High Voltage Power Transmission
Company of China Southern Power Grid, Guangzhou 510700, China

Abstract. The precise transformer modeling is of significant importance for the

analysis and simulation of transient overvoltage. This paper proposes an
improved hybrid transformer model combining nonlinear and frequency-
dependent effects and the parameters determination methods. The duality-
based transformer model is employed to describe the low-frequency character-
istics of the hybrid model by the direct application of the principle of duality.
The model parameters are obtained considering the variations of the transformer
electromagnetic behaviours under various operating conditions. The black-box
model is utilized to describe the high-frequency characteristics of the hybrid
model. The black-box modeling methodology is based on the transfer function,
which is derived from the two-port test data at the transformer terminals. The
model parameters are approximated by the vector-fitting method, which is used
in this paper to fit the transfer function by rational functions consisting of real as
well as complex conjugate poles and residues. The hybrid model is developed
by the merging of the dual transformer model and the black-box circuit model
by the filters. The hybrid model is validated with experimental data of the 10-kV
potential transformer on the inrush current, the harmonic voltage, and the
lightning excitation, respectively. There is a good agreement between the
measurement results and the simulation results of the voltage on the secondary
side. The hybrid model proposed in this paper is meaningful for the overvoltage
analysis in power systems.

Keywords: Duality
High-frequency
Black-box model
Vector-fitting

1 Introduction

Transformer is one of the crucial equipment in power systems. Potential transformer is

the widespread voltage monitored device in power system, and it is also the
transformer-type equipment, which is composed of core and winding. The precise
transformer modeling is of significant importance for the analysis and simulation of
transient overvoltage. But the parameters determination of transformer model is very

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1244–1253, 2020.
https://doi.org/10.1007/978-3-030-31676-1_116
 Duality-Based Potential Transformer Model 1245

difficult due to the complex features of the transformer core and windings, such as
nonlinearity and frequency dependence. Their high frequency behaviours, which are
influenced by their configuration and installation topology, are mainly characterized by
several resonance points due to the inductive and capacitive effects from the windings,
tank and core [1]. Therefore, this paper intends to present a hybrid transformer model
which can describe both the nonlinearity and wide band characteristics.
The high-frequency transformer models are needed in transient overvoltage studies
involving transformer network interactions as well as transfer of overvoltage between
windings [2, 3]. High-frequency models are routinely developed by the manufacturers
starting from detailed design information and used for predicting internal winding
overvoltage during the lightning impulse test [4]. These white-box models are usually
unavailable to external parties and they often lack in accuracy for use in general
network studies. When the model is intended for network studies only, a more practical
approach is to characterize the transformer terminal behaviours using frequency sweep
measurements followed by model extraction via curve fitting [5–8]. The latter black-
box model often gives highly accurate results when used in studies of high-frequency
transients although the information about internal voltages is lost.
It is known that the nonlinear-characteristics of the steel results in a shift in fre-
quency and magnitude of the impedance frequency responses up to several tens of
kilohertz. The low-frequency transient characteristics of transformers, such as inrush
current and ferroresonance, are deeply influenced by the nonlinearity of iron core. As a
remedy, it was proposed in [5] to merge the wideband data with those generated by a
standard 50/60 Hz model prior to model extraction via curve fitting. However, this
alternative model is still limited since it is not capable of taking core nonlinear effects
[6] into account.
This paper introduces an alternative way of combining transformer nonlinear and
frequency-dependent effects which permits including the transformer’s high-frequency
resonances. The duality-based transformer model, which is the most popular model due
to its clear physical meanings and high accuracy, is employed to describe the low-
frequency part of the hybrid model. The black-box model, approximated by rational
functions, is utilized to describe the high-frequency characteristics of the hybrid model.
For the convenience of incorporating the model into ATP-EMTP, the circuit synthesis
technique is further applied based on the approximations.

2 Hybrid Modeling Procedure

The dual reversible transformer model proposed in Fig. 1 is first briefly introduced. The
model is established by the direct application of the principle of duality. The dual
electrical equivalent circuit is shown in Fig. 1. The magnetizing inductance is divided
into two parts: leg plus half the yoke, and the yoke plus the returning leg.
1246 D. Peng et al.

Rs1 Ls Rs2

N1:N N:N2

L1 Rm1 Rm2 L2

Fig. 1. Dual electrical equivalent circuit of the single-phase transformer.

The model parameters are obtained considering the variations of the transformer
electromagnetic behaviours under various operating conditions. Rs1 and Rs2 are the
winding resistances. Ls is the leakage inductance between the windings. The nonlinear
inductances L1 and L2 and the shunt resistances R1 and R2 represent the magnetizing
branches of the primary and secondary sides, respectively. When the transformer iron
core is unsaturated or slightly saturated, the leakage inductance is very small compared
with the magnetizing inductance. Therefore, the magnetizing curves obtained by terminal
measurements can be equally distributed to the two magnetizing branches (see Fig. 1)
introducing negligible errors. When the iron core goes into deep saturation, its relative
permeability gradually decreases and tends to 1. The saturated iron can be considered the
same as the air, and it will affect the distribution pattern and magnitude of the magnetic
flux in the air. Therefore, the deep saturation inductance should be accurately measured.
However, the parameters of the model are obtained in the power frequency; the
application of the dual reversible transformer model is in the low-frequency electro-
magnetic transient. The transmission characteristics of transformer in high-frequency
cannot be accurately described with the dual model. Therefore, the dual transformer
model is modified in this paper by the method proposed in [8]. The modified transformer
model can not only simulate the low-frequency electromagnetic transient process, but
also describe the transmission characteristics in high frequency. The schematic diagram
of the modified model is shown in Fig. 2. The hybrid model is comprised by the dual
transformer model paralleled with the wide-band circuit model. In order to ensure the
smooth conversion between low frequency and high frequency, the low-pass filter is
connected to the dual model, and the high-pass filter is in series with the wide-band
model. By selecting the appropriate cut-off frequencies, the hybrid transformer model
could perform the smooth transition between low frequency and high frequency.

CH RH Black-box Model

Ls
RL Rs1 N1:N N:N2 Rs2

L1 Rm1 Rm2 L2
CL

Fig. 2. The schematic diagram of the hybrid model.

 Duality-Based Potential Transformer Model 1247

The wide-band model in this paper is represented by the black-box model, which is
developed by the two-port test data. The black-box model does not require detailed
information about the transformer and its material properties, and it is derived from the
two-port test data at the transformer terminals. The model parameters are estimated
using an optimization method, minimizing the error between the measured and the
calculated data. The black-box modeling methodology is based on the transfer function,
which is defined by the recorded voltage ratios at the transformer open-circuited ter-
minals. The transfer function can be obtained from the two-port S parameters, which
are measured by the network analyser. The transfer function presents different values as
frequency increases. The vector-fitting method is one of the rational approximation
methods. It is used in this paper to fit the transfer function by rational functions
consisting of real as well as complex conjugate poles and residues. The black-box
model is then implemented in the EMTP software by the synthesization method.

3 Parameters Determination and Implementation in EMTP

The hybrid transformer model proposed in this paper will be implemented on an

inductive 10-kV potential transformer (PT). The capacity of the PT is 15 VA. In order
to develop the hybrid model of PT in EMTP, the determination methods of model
parameters are introduced in this section.
Firstly, the parameters of the dual transformer model are obtained by the open-
circuit test, short-circuit test, and deep saturation test. The leakage inductance LS and
winding resistance RS1 and RS2 are obtained by the short-circuit test. The nominal
magnetizing curve and magnetizing resistance Rm are obtained by the open-circuit test.
The deep saturation part of the magnetizing curve is obtained by the deep saturation
inductance test. The deep saturation inductances L1-sat and L2-sat are determined by a
modified AC plus DC laboratory tests shown in Fig. 3. A fundamental voltage with DC
bias is connected to one transformer winding, and the other winding is open-circuited.
The DC bias excites the transformer into deep saturation, and the small AC voltage is
used for testing the inductance. Then, L1-sat and L2-sat are obtained by the allocation
method proposed in [9]. Based on the above method, the parameters of the dual
transformer model are shown in Table 1.

Table 1. The parameters of the dual transformer model.

RS1 RS2 LS Rm1 Rm2 L1-sat L2-sat
2250 X 0.43 X 1.96 mH 2475.4 X 2475.4 X 7.7 mH 3 mH
1248 D. Peng et al.

IAC+DC + Rdc
DC
UAC+DC

AC
-

Fig. 3. The AC + DC test circuit for deep saturation inductance.

Secondly, the circuit parameters of the black-box model are determined by the
synthesization method. The KEYSIGHT 5061B network analyser has been used to
measure the S parameters of the PT from 5 Hz to 100 kHz. Based on the relationship of
S parameters and the transfer function, the voltage transfer function can be represented
by

U 2 ðsÞ 2S12
H ðsÞ ¼ ¼ ð1Þ
U1 ðsÞ ð1 þ S11 Þð1  S22 Þ þ S12 S21

The frequency response of the voltage transfer function is shown in Fig. 4. It indicates
that the transfer function varies with frequency. The VF method [1] is adopted here to
fit the frequency-dependent voltage transfer function by means of rational function
using real poles as well as complex conjugate poles, and the function passivity is
enforced by means of the technique described in [6]. The VF method approximates the
frequency response with rational functions, expressed in the form of a sum of partial
fractions as in

XN
cn
f ðsÞ  þd ð2Þ
n¼1
s  pn

The poles pn and residues cn are either real quantities or complex conjugate pairs
while d is real. The VF method solves the Eq. (2) sequentially as a linear problem by
pole identification and residue identification. The approximation is achieved by
replacing a set of starting poles with an improved set of poles via a scaling procedure.
The high number of resonance peaks of the measured curve can be well fitted by the
complex starting poles.
The fitting result of the voltage transfer function is also shown in Fig. 4. The fitting
order is 8, including 2 real pole and 3 pairs of conjugate poles. The values of the poles
and the corresponding residues are shown in Table 2. The fitting result in Fig. 4 shows
that the VF method can fit the frequency-dependent voltage transfer function well.
 Duality-Based Potential Transformer Model 1249

Fig. 4. The frequency response of the voltage transfer function and the comparison of the fitting
result.

Based on the fitting expression of the voltage transfer function, the technique
proposed by [2] is adopted here to synthesize the circuit for the expression. The each
term of the expression can be represented by the specific circuit. The black-box circuit
model is comprised of the circuits of each term by series and parallel connection. The
constant term d can be synthesized with the ideal transformer with ratio d. For a pair of
real pole −pn (pn > 0) and residue cn, the synthesized circuit is shown in Fig. 5(a). The
parameters of the circuit can be obtained as follows

L1n ¼ 1=pn
cn ð3Þ
H1n ¼
pn

For the pairs of conjugate poles −prn ± j pin (prn > 0) and the residue −crn ± j cin, the
synthesized circuit is shown in Fig. 5(b). The parameters of the circuit can be obtained
as follows

2prn
C2n ¼
p2rn þ p2in
1
L2n ¼
2prn ð4Þ
crn prn  cin pin
H2n ¼ 2
p2rn þ p2in
crn
H3n ¼
prn

Table 2. The fitting poles and residues of the voltage transfer function.
p 0.0068 1.1*106 8.7*103 ± 2.6*105i 1.13*104 ± 3.9*105i 3.5*104 ± 5.1*105i
c −0.037 −4.7*104 −16 ± 178i −4.7 ± 36.5i −571 ± 2.4*103i
1250 D. Peng et al.

L1n 1:H1n

U1(s) R=1Ω U2(s)

(a)
L2n
1:H2n

C2n

U1(s) U2(s)

R=1Ω

1:H3n
(b)

Fig. 5. The circuit synthesization of the VF rational function.

After acquiring all the parameters of the black-box circuit model, the hybrid
transformer model can be developed in ATP-EMTP as shown in Fig. 6. The filter
parameters were chosen as fLP = 1 kHz and fHP = 1 kHz, by the method proposed in
[8]. Due to too many components, the black-box circuit model has been compressed as
a component named black-box model (BBM) in ATP-EMTP.

Fig. 6. The simulated diagram of the hybrid transformer model in ATP-EMTP.

4 Method Validation

The proposed model is validated with experimental data of the 10-kV potential
transformer on various conditions. The inrush current, the harmonic voltage and the
lightning excitation are applied on the primary side of the potential transformer for the
validation.
In order to demonstrate the simulation ability of the hybrid model in low frequency,
the simulation result of the inrush currents and the comparison result with the measured
inrush currents are presented in this section. The rated voltage is applied on the primary
side of the PT at the zero crossing of the voltage. As shown in Fig. 7, the simulated
 Duality-Based Potential Transformer Model 1251

inrush currents are very close to the measurement. It shows a 6.4% difference between
the simulated and experimental inrush currents (peak). The comparison result validates
the effectiveness of the hybrid transformer model in low-frequency studies.

Measured Inrush Current

30
Simulated Inrush Current
25
Current I [A]

20

15

10

0
0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 0.055 0.06 0.065
Time T [s]

Fig. 7. Comparison between the inrush currents of measurement with simulation.

In order to demonstrate the simulation ability of the hybrid model in high fre-
quency, the harmonic voltage with 80 kHz is applied on the primary side of the PT.
The amplitude of the sinusoidal voltage is 10 kV. As shown in Fig. 8, the secondary
voltages are obtained from the PT simulation studies and also the laboratory test. The
comparison result validates the effectiveness of the hybrid transformer model in high-
frequency studies. The ratios of the simulation and measured results are 0.0571 and
0.0578, respectively. It illustrates that the ratio changes with the frequency.

600
Measured Secondary Voltage
Simulated Secondary Voltage
400

200
Voltage U [V]

-200

-400

-600
0 1 2 3 4 5 6 7 8
Time T [s] x 10
-5

Fig. 8. Comparison between the harmonic voltages of measurement with simulation.

The simulation results of the lightning excitation and the comparison results with
the measurement are also presented to validate the hybrid transformer model in
1252 D. Peng et al.

transient simulations. The waveform of the lightning excitation is the standard lightning
waveform with 1.2/50 ls wavelength. The crest value is 10 kV (Because of the limit of
experiment condition). The test is also valid because the transformer exhibits linearity
due to the high frequency. As shown in Fig. 9, the simulated voltage of the secondary
side is very close to the measurement, which validates the effectiveness of the hybrid
transformer model for the studies of high-frequency transients.

200
Measured Secondary Voltage
150 Simulated Secondary Voltage
Voltage U [V]

100

50

-50
0 1 2 3 4
Time T [s] -4
x 10

Fig. 9. Comparison between the lightning excitation voltages of measurement with simulation.

In conclusion, there is a good agreement between the measurement results and the
simulation results on each condition; thus, the validity of the methodology is proved.

5 Conclusions

A hybrid transformer model combining nonlinearity and wide band characteristics is

proposed in this paper. The duality-based transformer model and black-box model are
employed to describe the low-frequency and high-frequency parts, respectively. The
filter is connected to the hybrid model to ensure the smooth conversion. The filter
parameters were chosen as fLP = 1 kHz and fHP = 1 kHz in this paper. The VF method
is utilized to fit the frequency-dependent voltage transfer function and shows good
coincidence. The hybrid model is implemented in ATP-EMTP by the circuit synthesis
technique. The hybrid transformer model is validated with experimental data of a 10-
kV potential transformer on the inrush current, the harmonic voltage, and the lightning
excitation, respectively. The results show that the proposed model can accurately
describe both the low-frequency and high-frequency characteristics of the transformer.

Acknowledgments. This work is supported by the National Key Research and Development
Program of China (2017YFB0902701), National Natural Science Foundation of China (51837002,
and 51807018), and Guangdong Power Grid Corporation Science and Technology Project
(GDKJXM20162522).
 Duality-Based Potential Transformer Model 1253

References
1. Gustavsen, B.: Wide band modeling of power transformers. IEEE Trans. Power Deliv. 19(1),
414–422 (2016)
2. Zhang, Z., Lu, F., Liang, G.: A high-frequency circuit model of a potential transformer for the
very fast transient simulation in GIS. IEEE Trans. Power Deliv. 23(4), 1995–1999 (2008)
3. Liu, X., Cui, X., Qi, L.: Calculation of lightning-induced overvoltages on overhead lines
based on DEPACT macromodel using circuit simulation software. IEEE Trans. Electromagn.
Compat. 54(4), 837–849 (2012)
4. Chrysochos, A.I., Nousdilis, A.I., Papadopoulos, T.A., Papagiannis, G.K.: A wide band
black-box model of power transformers in ATP/MODELS. In: 49th International Universities
Power Engineering Conference (UPEC), Cluj-Napoca, Romania (2014)
5. Gustavsen, B., Semlyen, A.: A filtering approach for merging transformer high-frequency
models with 50/60-Hz low-frequency models. IEEE Trans. Power Deliv. 30(3), 1420–1428
(2015)
6. Gustavsen, B., Semlyen, A.: Enforcing passivity for admittance matrices approximated by
rational functions. IEEE Trans. Power Syst. 16(1), 97–104 (2001)
7. Gustavsen, B., Silva, H.M.J.D.: Inclusion of rational models in an electromagnetic transients
program: Y-Parameters, Z-Parameters, S-Parameters, transfer functions. IEEE Trans. Power
Deliv. 28(2), 1164–1174 (2013)
8. Gustavsen, B.: Wideband transformer modeling including core nonlinear effects. IEEE Trans.
Power Deliv. 31(1), 219–227 (2015)
9. Jazebi, S., De León, F., Farazmand, A., Deswal, D.: Dual reversible transformer model for the
calculation of low-frequency transients. IEEE Trans. Power Deliv. 28(4), 2509–2517 (2013)
 Propagation of Lightning, Oscillating
and Non-standard Impulse Waveforms
in Transformer Windings

Marek Florkowski1, Jakub Furgał2, Maciej Kuniewski2(&),

and Piotr Pająk2
1
ABB Corporate Research, ul. Starowiślna 13A, 31-038 Kraków, Poland
2
Department of Electrical Engineering and Electrical Power,
AGH University of Science and Technology, al. Mickiewicza 30,
30-059 Kraków, Poland
maciej.kuniewski@agh.edu.pl

Abstract. The transformers are subjected to overvoltages caused by the

lightning hit, grid switching phenomena (especially in Gas Insulated Substa-
tions) and selected failure states. However, not only such extreme conditions are
causing additional stress on electrical insulation. During normal operation
application of voltage source converters in power systems or in industrial net-
works subjects electrical insulation to permanent stress resulting from fast,
repetitive switching pulses. In most cases, the pulse width modulation
(PWM) technique is used. Due to an operation of different sources of electrical
energy and propagation phenomena in electrical power systems transformers are
subjected to overvoltages representing various waveforms.
Overvoltages generated in electrical networks have a form of aperiodic waves
with high and low du/dt slope, periodical transient waveforms and aperiodic
surges contained transient damped periodical components. Overvoltage of a dif-
ferent characteristic with amplitudes lower than the protection level of the surge
arresters, act on the transformer terminals practically without amplitude damping
and waveform shape change. The response of a transformer to the overvoltages is
related to the waveform of the acting voltage and the transient phenomena in the
transformers. Therefore an analysis of internal overvoltages occurring in trans-
formers during the influence of overvoltages of different wave shapes is important
for developments aimed at increasing the reliability of power transformers.
Comparison of transformer winding responses to lightning, oscillating and
non-standard impulse waveforms will be presented in the paper. Measurements
of overvoltages generated in the transformer windings during the action of
different voltage waveforms, including superposition of impulse and high fre-
quency oscillating, modulated content, are the basis of comparisons.

Keywords: Transformer windings
Impulse testing
Non standard waveform

1 Introduction

During the technical and economic optimization performed at the design of trans-
formers, a reduction of electrical strength may be determined to be acceptable relative
to other design considerations. A detailed analysis of the comparison of overvoltages
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1254–1264, 2020.
https://doi.org/10.1007/978-3-030-31676-1_117
 Propagation of Lightning, Oscillating and Non-standard Impulse Waveforms 1255

generated during typical operation conditions as well as during voltage tests with use of
voltage impulses have to be made at the first stage.
The transformers in operation conditions are subjected to overvoltages caused by
the lightning hit, grid switching phenomena (especially in Gas Insulated Substations)
and selected failure states. However, not only such extreme conditions are causing
additional stress on electrical insulation. During normal operation application of
voltage source converters in power systems or in industrial networks subjects electrical
insulation to permanent stress resulting from fast, repetitive switching pulses. In most
cases, the pulse width modulation (PWM) technique is used. Due to an operation of
different sources of electrical energy and propagation phenomena in electrical power
systems transformers are subjected to overvoltages representing various waveforms.
Overvoltages generated in power systems impact on insulation systems of windings
of power transformers despite of overvoltage protection systems. Residual voltages of
surge arresters are about three times greater than rated voltage of protected trans-
formers. During service of surge arresters, when external overvoltages exceed values of
its residual voltages, voltage wave between transformer clamps and earth can be
approximated by a rectangular shape [1–4]. If values of overvoltages are smaller than
designed protection level maximal values of overvoltages are not changed. Overvolt-
ages with small maximal values reach transformers almost without amplitude distortion
and are sources of overvoltages with large maximal values generated inside windings
due to effect of resonance phenomena [4–8].
The response of a transformer to the overvoltages is related to the waveform of the
acting voltage and the transient phenomena in the transformers. The transformer
windings may be represented as a distributed RLC network with frequency dependent
parameters. In such a structure, waveforms and maximum values of internal over-
voltages in certain locations can be different than the excitation waveform of external
transient voltages [6, 7, 9–11].
Therefore a comparative analysis of internal overvoltages occurring in transformers
during lightning impulse as well as during typical operating conditions are important
for development and design to increase the reliability of power transformers.
In general experimental and theoretical investigations of transient phenomena in
transformers concern first of all on an analysis of propagation of overvoltage waves
only in one windings phase of transformers. Such investigations do not provide full
impact analysis of overvoltages on whole internal insulation systems in transformers
during action external waveforms with different shapes [7, 12–14].
Transformer winding responses to standard lightning impulses and operational
overvoltages are presented. Investigations were conducted with the all windings of
transformer being subjected to voltage impulse tests, as well as transient voltages with
waveforms simulating overvoltages generated in electrical power systems. Operational
overvoltages were modelled by use of rectangular voltage strokes, sinusoidal voltage
with frequency of wide range and transient voltages with waveforms of and impulse
voltage including of oscillating component with different frequencies. Surges with
rectangular waveforms imitate transient voltages occurring on the surge arresters
protecting of the transformer. Sinusoidal voltage in broad frequency range and impulse
voltages including oscillating component with different frequencies simulate oscillating
transient components of external overvoltages occurring in power systems during
operation conditions.
1256 M. Florkowski et al.

2 Measurement Stand and Experimental Object

The waveforms of overvoltages generated in electrical power systems were modelled as:
– oscillating sinusoidal voltage,
– lightning voltage impulse,
– lightning voltage impulse with sinusoidal component of different frequency,
– rectangular voltage stroke with different steepness du/dt.
Sinusoidal voltages with varying frequency show the frequency response of
windings to overvoltages of differing frequencies, which could potentially be generated
in electrical power systems. Sinusoidal voltages with frequency varied in the range of
100 Hz–2 MHz were used in the investigations. Application of lightning voltage
impulse has been simulating overvoltages generated in transformers during lightning
tests of insulation systems.

Fig. 1. Measurement setup for recording of internal overvoltages in transformer windings: a –

test setup for recording of time dependences of overvoltages, b – test setup for recording of
frequency characteristics of overvoltages; x – distance between measurement point and the earth
(x = 0), l – height of the winding

Investigations performed by use of standard lightning voltage impulse, with super-

imposed sinusoidal component of different frequencies, simulates transients in power
systems, including oscillating components at the terminals of the transformers generated
as a results of propagation phenomena in electrical networks or during repeated restrikes
at circuit breakers. Voltage strokes with rectangular shape reflect the waveforms of
transient voltages at transformer terminals during operation of surge arresters [15–17].
 Propagation of Lightning, Oscillating and Non-standard Impulse Waveforms 1257

The measurement stand used for recording the waveforms of the overvoltages
generated during the action of standard lightning voltage impulse, standard lightning
voltage impulse with sinusoidal component of different frequency, rectangular voltage
stroke with different steepness du/dt (Fig. 1a) consists of a function generator Tektronix
type AFG310, a digital oscilloscope type 784D and a computer. Schematics of the
measurement stands used in investigation generated in transformer windings during
action of oscillating sinusoidal voltage is presented in Fig. 1b. The function generator
and the oscilloscope were connected to the host computer by GPIB-USB interface. The
dedicated software for measurements of frequency characteristics of overvoltages has
been implemented in LabView™ from National Instruments.
A 20 kVA 15/0.4 kV transformer was used as the experimental object. The basic
electrical and design parameters of the transformer are presented in Table 1.

Table 1. Rated and construction parameters of the experimental transformer

Sn, kVA 20
UnHV, kV 15
UnLV, kV 0.4
Number of phases 3
Uz, % 4.2
DPFe, kW 0.114
DPCu, kW 0.525
I0, % 2.8
Type of coils uc* rc**
Number of turns in the coil 810 650
Number of coils 4 4
Height of the winding l, mm 280 250
Internal diameter do, mm 157 157
External diameter di, mm 205 205
Width of the coil h, mm 25 25
*
uc - usual coils, **rc - reinforced coils
(insulation systems of coils have greater
electrical strength)

The investigations included the measurements of waveforms of overvoltages to

ground at selected points x/l in the HV- and LV windings during the excitation from
voltage impulses between external terminals of winding for phases B and C. Mea-
surements were taken via additional taps in the windings. Overvoltages were recorded
between terminals of HV- and LV windings every phases (x/l = 1) and earth and
between internal point with coordinate x/l = 0.5 in windings and earth.
Overvoltages in the HV windings are expressed using the coefficient ki HV over-
voltage calculated from the formula:
ui HV x=l
ki HV ¼ ð1Þ
uHV1:0max
1258 M. Florkowski et al.

where: ui LV x/l(t) - time dependence of the internal overvoltage between point with
coordinate x/l in the HV winding and earth, uHV 1.0 max - maximal value of external
voltage impulse between clamps of phases B and C.
Overvoltages generated in the LV windings are expressed using the coefficient kt LV
calculated by use of the formula:

ut LV x=l ðtÞ
kt LV ¼ # ð2Þ
uHV1:0max

where: ut LV x/l(t) - time dependence of transferred overvoltage between the point x/l in
the LV winding and earth, # - ratio of the transformer.
The measurements have been carried out using voltage stimulus with maximal
value 20 Vpp. The ferromagnetic core has practically no influence on the transient
phenomena in transformer windings for frequencies higher than several dozen of
kilohertz. For these frequencies transformer windings have practically a linear character
and the maximum values of voltage stimuli practically have no influence on the time
dependences and maximal values of the internal overvoltages in the windings [18].

3 Lightning Overvoltages in Windings of Transformer

Time waveforms of overvoltages in the HV and LV windings of the transformer

measured during the lightning stroke are presented in Fig. 2.

Fig. 2. Waveforms of overvoltages generated during action the full lightning impulse in selected
points, with coordinates (x/l = 1 and x/l = 0.5) (Fig. 1), of: (a) HV windings, (b) LV windings
 Propagation of Lightning, Oscillating and Non-standard Impulse Waveforms 1259

As a results propagation phenomena inside transformer, the waveforms of over-

voltages inside a winding have a different shape to the waveforms generated by the
lightning voltage stroke at the of the transformer terminals. The overvoltages contain
oscillations components. From investigations it is possible to observe that for example
the overvoltage coefficient kiHV of voltage U0.5C (Fig. 1) in the HV winding of phase C
is 0.75 but for voltage UA at terminal of phase A reach about 0.5.
On the basis of results presented in Fig. 2b we can see that for example value of
coefficient kiLV for voltage Ua is equal 0.7 and for voltage U0.5c reach about 2.4.

4 Response of Transformer to Rectangular Voltage Stroke

with Different Steepness du/dt

Waveforms of overvoltages generated in HV windings of the transformer subjected to

rectangular voltage strokes with different shape du/dt are presented in Fig. 3.

Fig. 3. Waveforms of selected overvoltages in HV windings during action of rectangular

voltage impulse with different time rise: (a) 10 ns, (b) 100 ns

On the basis of analysis it is seen that overvoltages generated in HV windings

during action of rectangular voltage impulse with different risetime in range from 10 ns
up to 100 ns have the same waveforms. Also maximal values were practically not
changed. It results from low self frequencies of HV windings (Fig. 5).
1260 M. Florkowski et al.

Fig. 4. Waveforms of overvoltages in LV windings during action of rectangular voltage impulse

with different time rise: (a) 10 ns, (b) 100 ns

However overvoltages generated inside the LV windings during action of rectan-

gular voltage strokes strongly depend on risetime (Fig. 7). From investigations it is
possible to observe that, for example, the overvoltage coefficient kiLV for voltage U0.5c
(Fig. 1) in the LV winding of phase “c” for risetime of 10 ns is about 3.3 but for
risetime 100 ns is about 1.6. Overvoltage coefficient kiLV of transient voltage Ua at the
terminal of phase “a” for rise time 10 ns reach about 1.95 but for rise time of 100 ns is
equal only to 0.7 (Fig. 4).

5 Propagation of Oscillating Waveforms in Transformer

Windings

For the purposes of determining the response of a transformer winding to the oscillating
components of overvoltages generated in electrical power systems, the frequency
dependencies ux/l/u0 = g(f) of overvoltages inside the windings were measured. The
results of investigations are presented in Fig. 5. The character of this reaction depends
on the values of overvoltages affecting the insulation systems of transformers. On the
basis of the frequency characteristic one can observe that sinusoidal voltages at certain
frequencies are strongly amplified inside the windings.
Investigations confirm that overvoltages containing transient waveforms with an
oscillatory character can be strongly amplified inside transformers due to resonance
phenomena, particularly if the frequencies of these waveforms are equal to the self
frequencies of the transformers.
 Propagation of Lightning, Oscillating and Non-standard Impulse Waveforms 1261

Fig. 5. Frequency dependencies of overvoltages between terminals and point in half of

windings and earth: (a) HV windings, (b) LV windings

On the basis of the frequency dependencies of overvoltages in HV- and LV wind-

ings of the transformer it can be observed that oscillating sinusoidal voltages at selected
frequencies propagated in transformer are strongly amplified inside the windings.
Results of presented investigations confirm that overvoltages generated in transformers
during propagation of oscillating waveforms can have large values for a few frequen-
cies. In HV windings oscillating overvoltages reach maximum for frequencies
0.65 kHz, 13.1 kHz, 36.7 kHz and 51 kHz and in LV windings for frequencies:
0.64 kHz, 8.7 kHz, 107 kHz and 0.85 MHz. It is very important to note that if resonant
frequencies of the transformer is larger then the probability of multiplication of internal
overvoltages in transformers is greater. It was also noticet that oscillating overvoltages
achieve maximal values in windings of all phases practicaly for the same frequencies.

6 Propagation of Lightning Voltage Impulses with Sinusoidal

Component of Different Frequency

In order to analyze the internal overvoltages generated in the winding as a result of

resonance phenomenon at impulse voltage with superimposed oscillating component,
the special stimulus has been designed. It contained the aperiodic part as a lightning
voltage surge but also a superimposed oscillating component.
The response of the HV windings to periodic voltage surge containing oscillating
components with resonance frequencies (selected for example from frequency
1262 M. Florkowski et al.

Fig. 6. Response of the HV windings to lightning impulse containing oscillating component

with frequency of: (a) 8.3 kHz, (b) 13.1 kHz (Fig. 5)

Fig. 7. Response of the LV windings to lightning impulse containing oscillating component

with frequency of: (a) 107 kHz, (b) 0.85 MHz (Fig. 5)
 Propagation of Lightning, Oscillating and Non-standard Impulse Waveforms 1263

characteristics presented in Fig. 5) of 8.3 kHz and 13.1 kHz are illustrated in Fig. 6.
One may see that the oscillating components related to winding resonance frequencies
are amplified inside the windings. Overvoltages in HV windings have greater values
than those generated during action of lightning impulse.
Time relations of overvoltages generated in LV windings as a response to an
aperiodic voltage surge containing oscillating component with resonance frequencies
(selected from Fig. 5) of 107 kHz and 0.85 MHz are presented in Fig. 7. From results
of investigations it is possible to see that oscillating overvoltages in LV windings are
also strongly amplified. For example for frequency 0.85 MHz the overvoltage coeffi-
cient for voltage U0.5c (Fig. 1) equal to 3.4 but for voltage Ua is about 1.5.

7 Conclusion

Investigations of overvoltage propagation presented in this paper were performed on all

windings of transformer. Measurements were obtained for the windings of transformer
being subjected to voltage impulse tests, as well as transient voltages with waveforms
simulating overvoltages generated in electrical power systems. Results have shown that
the transient phenomena in windings occurring during action of external voltage
impulses change waveforms and maximal values of overvoltages propagated inside
transformer. The overvoltage exposure of selected parts of transformer insulation
systems, in response to selected voltage surges appearing during transformer operation,
is higher than overvoltage exposures occurring during standard lightning voltage
impulse. Overvoltages inside transformer windings during service are dependent on
shapes of voltage waveforms generated in power systems. Overvoltages reach maximal
values for surges with an oscillating transient component with the frequency equal to
the transformer resonant frequency. Results of investigations reveal that the transformer
has usually few resonant frequencies.

References
1. Rocha, C.A.O.: Electrical transient interaction between transformers and the power system.
In: CIGRE-Brazil Joint Working Group JWG - A2/C4-03, pap. C-104, pp. 1–6 (2008)
2. Joint Working Group A2/C4.39-CIGRE: Electrical transient interaction between transform-
ers and the power system (part 1: Expertise, part 2: Case studies). CIGRE (2014)
3. Massaro, U.R., Antunes, R.: Electrical transient interaction between transformers and power
system - Brazilian experience. In: International Conference on Power Systems Transients
(IPST 2009), Kyoto, Japan, 3–6 June 2019, pp. 1–6 (2019)
4. Furgał, J.: Analysis of overvoltages in windings of power transformers protected by use of
metal oxide surge arresters. Euro. Trans. Electr. Power Eng. 19, 400–410 (2009)
5. Lapworth, J.A., Wilson, A.: Transformer internal overvoltages caused by remote energi-
sation. In: IEEE PES Power Africa 2007 Conference and Exposition, Johannesburg, South
Africa, 16–20 July 2007, pp. 1–6 (2007)
6. Hori, M., Nishioka, M., Ikeda, Y., Naguchi, K., Kajimura, K., Motoyama, H., Kawamura,
T.: Internal winding failure due to resonance overvoltages in distribution transformer caused
by winter lightning. IEEE Trans. Power Deliv. 21(3), 1600–1606 (2006)
1264 M. Florkowski et al.

7. Gustavsen, B., Brede, A.P., Tande, J.O.: Multivariate analysis of transformer resonant
overvoltages in power stations. IEEE Trans. Power Deliv. 26(4), 2563–2572 (2011)
8. Popov, M., van der Sluis, L., Smeets, R.P.P.: Evaluation of surge-transferred overvoltages in
distribution transformers. Electr. Power Syst. Res. 78(3), 441–449 (2008)
9. Borghetti, C., Morched, A., Napolitano, F., Nucci, C.A., Paolone, M.: Lightning-induced
overvoltages transferred through distribution power transformers. IEEE Trans. Power Deliv.
24(1), 360–372 (2009)
10. Florkowski, M., Furgał, J., Pająk, P.: Analysis of fast transient voltage distributions in
transformer windings under different insulation conditions. IEEE Trans. Diel. Electr. Insul.
19(6), 1991–1998 (2012)
11. Holdyk, B., Gustavsen, B.: External and internal overvoltages in a 100 MVA transformer
during high-frequency transients. In: International Conference on Power Systems Transients
(IPST 2015), Cavtat, Croatia, 15–18 June 2015, pp. 1–6 (2015)
12. Fujita, S., Hosokawa, N., Shibuya, Y.: Experimental investigations of high frequency
oscillation in transformer windings. IEEE Trans. Power Deliv. 13(4), 1201–1207 (1998)
13. Popov, M., Sluis, L., Smeets, R.P.P., Roldan, J.L.: Analysis of very fast transients in layer-
type transformer windings. IEEE Trans. Power Deliv. 22(1), 238–247 (2007)
14. Florkowski, M., Furgał, J., Kuniewski, M.: Propagation of overvoltages in distribution
transformers with silicon steel and amorphous cores. IET Gener. Transm. Distrib. 9(16),
2736–2742 (2015)
15. PN-EN 60099-5 Surge arresters. Recommendation for selection and application
16. IEEE Working Group 3.4.11: Modeling of metal oxide surge arresters. IEEE Trans. Power
Deliv. 7(1), 302–309 (1992)
17. EN 60071-2 Insulation coordination. Part. 2
18. Wilcox, D.J., Conlon, M., Hurley, W.G.: Calculation of self and mutual impedances for coils
in ferromagnetic cores. IEE Proc. 135(7), 470–476 (1988)
 Recent Developments of Field Grading
for High Voltage Vacuum Circuit Breakers

B. Kühn1(&), D. Gentsch2, B. Weber1, M. Hilbert1, and M. Kurrat1

1
Institute for High Voltage Technology and Electrical Power Systems,
Technische Universitaet Braunschweig, Schleinitzstraße 23,
38106 Brunswick, Germany
b.kuehn@tu-braunschweig.de
2
ABB AG, Calor Emag Mittelspannungsprodukte,
Oberhausener Str. 33, 40472 Ratingen, Germany

Abstract. In the medium-voltage range, circuit-breakers with vacuum as

interruption medium have advantages such as environmental friendliness and
lower maintenance costs compared to sulphur hexafluoride (SF6). Current
political and industrial developments suggest an effort to expand the field of
application of vacuum circuit breakers (VCB) up to high voltage levels. With a
rising operating voltage level, the influence of parasitic capacitances on floating
components, such as metal vapour shields, result in an increased absolute
voltage drift of these components. This reduces the dielectric strength of the
VCB, especially in case of lightning impulse stress. To ensure the dielectric
strength of the high voltage VCBs, larger insulation distances are required. An
approach with pure (linear) upscaling of all components and the gap distance is
not sufficient from the technical as well as economical point of view.
This paper presents a novel approach to stabilize floating potentials in VCBs
based on a combination of shielding from external influences and grading
floating potentials by internal capacitive coupling mechanisms. This is achieved
through an additional shield arrangement in an outer surrounding vacuum
chamber. In this chamber, shields on defined potentials are coupled with the
dielectric displacement to floating shields and thus stabilize those potentials.
Following this approach, the coupling between the shields is simulated with
COMSOL as well as proven over empirical studies. To determine the dielectric
strength of vacuum and thus to derive the required distances between the
shields, empirical investigations are considered. A prototype of the shield
arrangement for a double breaking VCB is developed and tested. Finally,
comparative measurements of the double breaking VCB arrangement with and
without external disturbances are presented and show the functionality of the
combined shielding and grading. The main goal of this work is an approach to
stabilize floating potentials in high voltage VCB and thus ensure dielectric
insulation under lightning impulse stress.

Keywords: Shielding
Vacuum double break
Field grading

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1265–1275, 2020.
https://doi.org/10.1007/978-3-030-31676-1_118
1266 B. Kühn et al.

1 Introduction

The vacuum circuit breaker (VCB) is established as a circuit-breaker in medium

voltage levels up to 52 kV. In comparison to other interruption approaches, like sulfur
hexafluoride, vacuum interrupters are nearly maintenance free in operation and feature
a high environmental sustainability. Recent political and industrial developments show
tendencies to enlarge the operational level of VCB to high voltage. Next to changed
requirements during the arcing phase, the dielectric requirements for VCB change with
a rising operational voltage as well. One challenge is to ensure the dielectric strength of
the VCB in open contact position. Most of the single break VCB have components like
metal vapor shields on floating potential, Fig. 1 (left). In case of double break high
voltage VCB arrangements, the whole middle contact part is floating, Fig. 1 (right).
The potential of those parts is mainly defined through couplings over the dielectric
displacement of internal and external parts. In a worst case, the influence of external
couplings result in rising electrical fields and cause a breakdown of the VCB. In case of
double break applications, the usage of external grading capacitors to stabilize floating
potentials is a common and well known approach [1–3]. In case of high voltage single
break VCB applications, first methods to stabilize the floating shield potential are
presented in [4, 5].

Single Break Double Break

C4
C4 C2
C2 C1

C3 CE3
C5
C1

C5
C3
CE CE2
C3 C1
C2 CE1
C5
C4

C1 capacitance of the contact gap

C2 - C5 internal capacitances couple to the floating components
CE, CE1- CE3 parasitic external capacitances couple to the floating components

Fig. 1. Capacitive coupling in single- and double break vacuum circuit breakers in open contact
position
 Recent Developments of Field Grading for High Voltage VCB 1267

Most solutions come with additional external components, such as grading

capacitors or diodes, which are not suitable for industrial solutions. In this paper a new
approach to stabilize floating potentials in case of single and double break is presented.

2 Stabilized Floating Potentials in Circuit Breakers

For a better understanding of the floating potentials the equivalent circuit diagrams of
the VCB based on [6] were considered. In case of single break the voltage distribution
UC4⁄UC5 is influenced by the parasitic capacitance CE according to Eq. (1).

Equivalent circuit diagram of single break

VCB VCB with grading capacitors
C2 C4 UC4
C2 C4 Cg
UC4
UC1
C1 UC5 CE UC1
C1 CE
C5 UC5
C3 C5 Cg
C3

| for >>
for >0
Typical values: C1 ≈ 3.2 pF; C2 = C3 ≈ 3.5 pF; C4 = C5 ≈ 1.3 pF; CE ≈ tens
of pF; Cg ≈ hundreds of pF up to nF

Fig. 2. Equivalent circuit diagram of a single break VCB with and without additional grading
capacitors

To stabilize the floating potentials it is common to use grading capacitors. Those

grading capacitors work well according to Fig. 2, when the grading capacitance Cg is
significantly larger than the influences of CE. This is applicable for single and double
break approaches. Several publications (based on double break VCB) show that
grading capacitances in the range of 150 pF lead to a voltage distribution of 40% to
60%. A capacitance of up to 1200 pF is necessary for a field grading with a ratio of
49% to 51% [1–3]. According to Fig. 2 the second possibility is to reduce the influence
of the parasitic component CE due to a field shielding.

3 New Shield Design for Vacuum High Voltage Circuit

Breakers

In the field grading system, two new approaches are combined. The first is to design
grading capacitors in an additional outer chamber (with vacuum-, liquid- or solid
insulation) according to Fig. 3 additional shields are arranged in the outer chamber.
1268 B. Kühn et al.

Additional separate outer chamber

Stabilized floaƟng
shield

grading capacitors

Fig. 3. Approach of additional grading capacitors in a separate outer chamber with vacuum-,
liquid- or solid-insulation

The dielectric displacement between the shields should be used to design grading
capacitors. Therefore the maximal permitted electrical field needs to be defined as well
as the value of required grading capacitance to stabilize the floating potential. As model
a shield design for a 72.5 kV VCB is taken. The shield design should be as simple as
possible and function for single- and double break applications in the same way.

3.1 Dielectric Strength of the New Shield Design

The shield design is based on dielectric properties. To define upper limits for the
electrical field strength between the shields, one of the most dielectric stressing test
according to the international standard [7] was taken into account.
Average breakdown field strength in
[kV/mm]

Gap length in [mm]

Fig. 4. Dielectric strength of vacuum in dependence of gap length from [8, 9]

Previous works investigated the dielectric strength of vacuum during lightning

impulse [8, 9]. The dielectric strength is dependent on several factors like gap length
 Recent Developments of Field Grading for High Voltage VCB 1269

and material or surface roughness. The electrical field design in this work is based on
the results shown in Fig. 4.

3.2 Achievable Capacitance with Vacuum Grading Capacitors

To design the vacuum grading capacitors, the operation voltage level of the VCB and
the dielectric strength of vacuum in dependence of the distance between the shields
were considered as maximum permitted electrical field stress between the shields. The
capacitance of the grading capacitors is mainly restricted by geometrical dimensions of
the VCB, which define the areas to be coupled. In a simplification the VCB could be
expressed as a cylinder capacitor (Fig. 5). Considering these factors, simple geomet-
rical calculations show that a coupling between the shields would reach a max. possible
grading capacitance Cg in the range of 40 pF.

Fig. 5. Calculation of max. possible shield length according to the dimensions of a VCB

The floating metal vapour shield of a VCB in a gas insulated switchgear (GIS) is
coupled to ground with approx. CE  20 pF. Considering Eq. (2), and assuming for
C35 = C24  5 pF a max. ratio of 59% to 41% is possible. To reach a uniform
distribution the influence of CE needs to be reduced.

3.3 Development of a Combined Field Shielding and Field Grading

To reduce the influence of the parasitic capacitances CE, two additional shields
(Fig. 6A and B) are designed in the outer vacuum chamber G. The small gap between
A and B reduces the influence from CE on the floating metal vapour shield D. During a
lightning impulse voltage (LIV) stress, shield A is supposed to be on high voltage
(HV) potential and shield B is supposed to be on ground potential. The floating metal
vapour shield D is coupled to the outer shields (A and B) over the dielectric dis-
placement and determine the grading capacitance Cg. The inner switching vacuum
chamber H and the outer chamber G are two separate chambers. In the outer chamber G
vacuum-, liquid- or solid-insulating can be applied. The advantage of the grading will
be similar.
1270 B. Kühn et al.

A) Outer shield (on HV)

E) Internal ceramics

D) Internal metal
vapour shield (floating)

G) Separated outer
chamber

C) Grading
capacitances

F) Outer ceramics

H) Internal vacuum
chamber

E) Internal ceramics

B) Outer shield Surrounding

(on ground) grounded
tube (GIS)

Fig. 6. Design of a shield arrangement with a combined shielding/grading in a separate outer

chamber

3.4 Simulation of the Capacitances and the Electrical Field Distribution

During LIV Application
Simulatory approaches of high voltage components with floating parts, based on [10],
are used for the coupling capacitances between the shields with the software COMSOL
(electrostatics and electric currents physic, time invariant and time dependent). The
geometry is surrounded by a zero charge line. In the simulation the geometry from
Fig. 6 is implemented. The outer shield (A) is on high voltage potential and the
opposite shield (B) on ground potential. The potential of the internal metal vapour
shield (D) is defined through the capacitive coupling. The simulated grading capaci-
tance with vacuum as dielectric (er = 1) Cg is 74 pF which is proven by measurements
with an RLC-Meter (80 pF measured). In a first simulation the voltage distribution
without (Fig. 6 - left) and with (Fig. 6 - right) disturbances was calculated. Therefore, a
grounded tube surrounded the system in the second case. In both cases the potential of
the floating part (D) held 50% of the stressing voltage. Figure 7 shows the results of
additional simulations with and without the shield arrangement in case of a disturbance
over a grounded tube. In case of a disturbed system with the shield arrangement the
floating potential is constant on 50% of the stressing voltage (Fig. 7 right). In case the
same stress without the shield arrangement, the floating potential of the internal metal
vapour shield drifts to 19% of the stressing voltage (Fig. 7 left). In a worst case
scenario this changed voltage distribution could lead to high internal electrical fields
and cause internal breakdowns. Figure 7 also shows the electrical field distribution in
both cases.
 Recent Developments of Field Grading for High Voltage VCB 1271

325 kV = 100 % 325 kV = 100 %

20 kV/mm

450 mm
162 kV ≈ 50 %
62 kV ≈ 19 % (floating)
(floating)

100 mm
130 mm

Ground potential Ground potential

Fig. 7. Simulation results of the potential- and the electrical field distribution in case of a
disturbed system with and without shielding

The reason of the stabilized potential is the combination between the field shielding
and the strong coupling between the shields with a max. electrical field of 20 kV/mm.
Due to a distance of 10 mm between the shields, the electrical field stress is, according
to the literature not high enough to trigger internal flashovers (Fig. 4) [8, 9]. The result
of this simulation shows that the vacuum grading capacitors are optimized as a com-
promise of achievable capacitance and dielectric strength between the shields.

4 Practical Investigations

For practical investigations of the combination between field shielding and field
grading, a prototype of the shield arrangement is build. Aim of the investigations is to
show the voltage stabilisation and the dielectric strength of the system. The shields are
made from aluminium and positioned by insulating plastic rods. The whole test setup is
surrounded by a plastic tube in a diameter of 400 mm.

4.1 Potential of the Floating Middle Shield

For first low voltage tests (results shown in Fig. 9) air is chosen due to the comparable
permittivity er and the higher flexibility in handling.
Figure 8 shows the test setup with a special probe to measure the potential of the
floating middle shield based on capacitive coupling [11]. The probe couples over a
capacitance of Cp = 6.2 pF to the floating middle shield. Thus this probe enables a
nearly reactionless measurement of the floating potential. As shown in the circuit
diagram (Fig. 8), the capacitance of the probe was used to construct a capacitive
1272 B. Kühn et al.

CP=6.2 pF
b
UHV
c
Probe

22000 pF UProbe

D1 R1 = 1 Meg CB RD = 416 DUT

Zaengl

UHV CP=6.2 pF

CS = 6000 p RE = 9500 Cb = 1200 p C = 22000 pF

UProbe
D1 diode; R1, RD and RE resistors; CB crow bar; Cs and Cb capacitances;
DUT device under test

Fig. 8. Model of the shield arrangement to measure floating middle potential; Equivalent circuit
diagram of the supply circuit

voltage divider with a dividing factor of 3548. To supply the test LIV, a one stage
impulse voltage generator was used. For the measurements the upper shield (a) is
connected to high volt (lightning impulse, positive and negative up to 30 kV) and the
lower shield (c) is connected to ground potential. The whole test setup is positioned in a
test field with a distance of more than 1.5 m to all grounded parts. The results are
shown in Fig. 9 (Reference). As described before, the upper shield (a) shows a
potential of the LIV (measured with Zaengl-divider). The floating middle shield
(b) reaches almost 50% of the LIV. The ratio (b/a) shows a symmetric voltage dis-
tribution during the whole impulse duration. To consider the influence of external
disturbances, the test setup is surrounded by grounded aluminum foil with a radius of
200 mm and the LIV test is repeated (Fig. 6 disturbed). The results show almost no
detectable difference between the disturbed and undisturbed case.
 Recent Developments of Field Grading for High Voltage VCB 1273

+
35 50
Upper shield potential a a

Percental potential deviation from 50%

30
b
25
25 Deviation
Voltage in [kV] 20
0
15

10
-25
5
Middle shield potential b
0 -50
0 20 40 60 80 100
Time in [μs]

Fig. 9. LIV-test in air ! Measurement of voltage deviation with and without external
disturbances

4.2 Dielectric Stress of the System

It is well known that additional shields in vacuum interrupter influence the dielectric
strength of the whole system [12]. Therefore the ability of the test setup to withstand
the maximum dielectric stress of 325 kV LIV should be shown. In this test the insu-
lation medium in the outer chamber is exchanged from air to oil due to the comparable
dielectric strength of oil and vacuum. The system is tested for both polarities with 20
times of voltage applications of a 325 kV (1.2/50 µs) LIV according to the
regulations/standards.

4.3 Application for High Voltage Vacuum Interrupters

To apply the shield arrangement on high voltage vacuum interrupters, two concepts are
presented in Fig. 10. The grading and shielding electrodes are positioned in the
described separate outer vacuum chamber. These electrodes are coupled to the middle
electrode, which is vacuum-tightly brazed to the inner ceramics and thus build the inner
vacuum chamber. In this chamber the single- or double break contacts operate. In case
of double break it is possible to divide both contact gaps with a middle-flange. Inde-
pendent of a single- or double-break application the symmetric voltage distribution
over the shields is ensured.
1274 B. Kühn et al.

Single break Double break

Grading and
shielding electrodes

Metal vapor and

middle electrode

Outer ceramics Inner ceramics

Fig. 10. Concepts for industrial high voltage single- and double break VCB

5 Conclusions and Outlook

In this paper a new approach to handle floating shield potentials for high voltage
vacuum circuit breaker is presented. This approach is adaptable for single or double
break applications. The main advantage is a combination between a field shielding and
field grading through additional shields in a separate outer vacuum chamber. Due to
long term industrial experience, it is possible to reach almost perfect vacuum conditions
in the outer chamber without any influence of metal vapour during switching opera-
tions. In addition to that, the approach can easily be integrated in the production
process of vacuum circuit breakers. To design the internal electrical field distributions
FEM-simulations are used. To define the maximum permissible electrical field between
the shields, former research was considered. Based on the simulation results a proto-
type is built to prove the functionality of the shield arrangement. In different tests with
and without external disturbances it could be proven that the floating middle shield’s
potential is nearly independent of external influences. The withstand capability of the
max. permissible field stress of the system (for 72.5 kV operating voltage) is proven
with a 325 kV lightning impulse test. Finally, several approaches to adapt the shield
arrangement on high voltage vacuum circuit breakers are proposed. In the next step, the
investigations according to the max. achievable dielectric strength between the shields
in the outer chamber should be intensified. Therefore, an industrial prototype should be
manufactured for further dielectric tests.

References
1. Horn, A., Lindmayer, M.: Investigations on the series connection of two switching gaps in
one tube in vacuum. IEEE Trans. Plasma Sci. 33(5), 1594–1599 (2005)
2. Fugel, T., Koenig, D.: Influence of grading capacitors on the breaking performance of a 24-
kV vacuum breaker series design. IEEE Trans. Dielectr. Electr. Insul. 10(4), 569–575 (2003)
3. Betz, T., Konig, D.: Influence of grading capacitors on the breaking capacity of two vacuum
interrupters in series. IEEE Trans. Dielectr. Electr. Insul. 6(4), 405–409 (1999)
 Recent Developments of Field Grading for High Voltage VCB 1275

4. Nakano, Y., Surges, B., Hinrichsen, V.: Increasing the internal field strength of vacuum
interrupters with vapor shield potential control. IEEE Trans. Power Deliv. 33, 3155–3161
(2018)
5. Kato, K., Kaneko, S., Okabe, S., Okubo, H.: Optimization technique for electrical insulation
design of vacuum interrupters. IEEE Trans. Dielect. Electr. Insul. 15(5), 1456–1463 (2008)
6. Kühn, B., Kurrat, M., Gentsch, D.: Field grading method of a double breaking vacuum
chamber arrangement. In: VDE Hochspannungstage, Berlin, October 2016
7. High-voltage switchgear and controlgear - Part 100: Alternating-current circuit-breakers,
IEC 62271-100
8. Schümann, U., Budde, M., Kurrat, M.: Influence of shield capacity on the breakdown
voltage of vacuum tubes: discharges and electrical insulation in vacuum. IEEE Trans.
Dielectr. Electr. Insul. 2004, 618–621 (2004)
9. Sato, S., Koyama, K.: Relationship between electrode surface roughness and impulse
breakdown voltage in vacuum gap of Cu and Cu-Cr electrodes. IEEE Trans. Dielectr. Electr.
Insul. 10, 576–582 (2003)
10. Kühn, B., Kurrat, M., Hilbert, M., Gentsch, D.: Capacitance simulation method for HV-VCB
with floating potential shield design based on FEM. In: 20th International Symposium on
High Voltages, Buenos Aires, Argentina, August 2017
11. Kühn, B., Gentsch, D., Hilbert, M., Weber, B., Kurrat, M.: Novel field grading shield design
for double breaking vacuum chambers under lightning impulse stress. In: International
Symposium on Discharges and Electrical Insulation in Vacuum (ISDEIV), Greifswald,
Germany (2018)
12. Kahl, B.: Zur elektrischen Festigkeit von Dreielektroden-Anordnungen im Hochvakuum.
German Ph.D. thesis, VDI-Verlag, Düsseldorf (1994)
 Dimension Estimation of Transformer
Windings Based on Frequency Response
Measurement

Ebrahim Rahimpour1(&), Vahid Rashtchi2, and Reza Aghmashehr3

1
ABB AG, R&D Department, Bad Honnef, Germany
ebrahim.rahimpour@de.abb.com
2
Department of Electrical Engineering, University of Zanjan, Zanjan, Iran
3
IGMC, National Grid Dispatching Deputy, Tehran, Iran

Abstract. Frequency response of a transformer winding is a powerful tool to

determine the parameters of fast transient models in case that the internal
structure and design documents are not available. One method to determine the
model parameters is to estimate the geometric dimensions of the winding and
calculating the parameters of the model based on the estimated dimensions and
existing analytical formulas. In the current paper, dimension estimation of a
transformer winding to estimate the parameters of a detailed RCLM model for
fast transient analysis is converted to an optimization problem. The resulting
model is called transient gray box model. The gray box model of the trans-
formers’ windings is used for analysis of internal behavior of a winding when a
transformer’s factory documentation is not accessible. Particle swarm opti-
mization algorithm is used to minimize a proper cost function to achieve the best
overall fitting between measured and calculated frequency response of the
winding in a range of 10 kHz to 1 MHz. Several tests are conducted on a non-
interleaved double disk winding to evaluate the precision of the model. In
addition to the terminal frequency response measurement, internal frequency
response of the winding are measured. Comparison between simulation results
and both measured internal and external voltage waveforms, confirm that
determining the RCLM model parameters based on estimation of winding
dimension is a reliable and applicable method to study the transformer winding
behavior.

Keywords: Transformer winding
Frequency response
Dimension

estimation

1 Introduction

Frequency response (FR) of a transformer winding is a practical tool for parameter

determination of windings’ models for fast transient analysis and condition monitoring
of a transformer [1, 2]. Parameters of fast transient transformers’ models are estimated
either by analytical methods [3] or by measurements techniques [4]. Each of them has
their own advantages. Determination of transformers’ models’ parameters based on
analytical equations are precise and can be delivered to the customer when the

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1276–1284, 2020.
https://doi.org/10.1007/978-3-030-31676-1_119
 Dimension Estimation of Transformer Windings Based on FR Measurement 1277

apparatus is sold. However, this method is not applicable when factory documents are
not available [5]. In this condition the parameters of the models are identified by
measurements [6, 7].
In each frequency range, a special model is used to study the behavior of the
transformer. This paper deals with fast transient modeling of transformers windings.
The most important models to study fast transient phenomena in transformers are
detailed RCLM model [8] and multi transmission line model [9]. RCLM model is
investigated in this paper due to the following reasons [10]:
– The RCLM model is valid for a wide range of about 10 kHz up to MHz
– The RCLM model is able to present the physical features of transformers windings
– The model can be applied easily to any windings and also can be solved easily in
both frequency and time domain
In the current work, the winding dimensions estimation problem is defined as an
optimizing problem of a multivariable function. The objective function includes all the
necessary parameters for winding geometry. Particle swarm optimization algorithm
(PSO) is used to optimize the objective function based on the frequency response of the
winding.
A transformer winding containing 44 non-interleaved disks has been tested to
verify the simulation results.

2 Detailed Model of the Winding

There are different types of models to study and analyze the high frequency behavior of
the transformers windings. Among the existing models the detailed RCLM model
based on self and mutual inductances is suitable for calculating the transfer functions
and internal voltages and currents of the winding. The detailed model of the test
winding is shown in Fig. 1. The parameters of the detailed model are described below:
Li : Self-inductance of ith model unit
Mi;j : Mutual inductance between ith model unit and other units
Cei : Capacitance between ith model unit and ground (aluminium cylinder)
Ki : Resultant series capacitance of ith model unit
Rsi : Series resistance to represent winding losses
RKi : Parallel resistance due to dielectric losses of resultant series resistance
Gei : Parallel resistance due to capacitance between ith model unit and ground
(aluminium cylinder)

Each unit can contain one, two or more disks and turns. The model precision is
affected by the number of model units. The number of model units is considered as a
model parameter and can be determined based on the apparent resonances in the
frequency response of the winding.
1278 E. Rahimpour et al.

M i ,j

i −1 i i +1
R ssii L i ,i,i
R KKii

ZE
Ki
C eeii G eeii

Model Unit

Fig. 1. Detailed RCLM model of the winding

Both time domain and frequency domain can be used to calculate the nodal volt-
ages and current branches. In this study frequency domain is utilized to solve the model
due to considering the frequency dependence of the model elements and easy
implementation.

3 Test Bench

To evaluate the applicability of the dimension estimation method, a single transformer

winding is investigated. The winding contains 44 non-interleaved disks and each disk
consists of seven turns. The winding has been tested outside the transformer tank and
iron core. An aluminum cylinder is used to model the earth potential of the iron core.
The bottom terminal and aluminum cylinder are grounded. For internal frequency
response validation, two extra terminal are extracted at 17th and 25th disks. An impulse
wave with 310 V amplitude is applied at top terminal of the winding and the recording
data window is 163.8 µs. The voltage and current waves are acquired in time domain
and the waves are converted to frequency domain by mathematical transformation like
Fourier transform to calculate the transfer functions. The test winding plan and
experimental setup is shown in Fig. 2a and b respectively.
 Dimension Estimation of Transformer Windings Based on FR Measurement 1279

1. 1234567
258.5
4
2. 7654321 269.5
Conductor cross section
3. 1234567
paper thickness
288 =0.25
Press 2
board 515
Aluminum
43. 1234567 cylinder for 7.3 CU 7.8
modeling the core
44. 7654321

2.5

(a) (b)

Fig. 2. (a) The winding plan, (b) Experimental setup

4 Frequency Response Method for Geometry Determination

Measured frequency response at transformers terminal is a unique characteristic of each

transformer winding. Besides, the frequency response of transformers windings is
dependent to the windings’ geometry and internal structure of transformers. Thus, the
influence of the geometric position and the physical characteristic of the windings in
the frequency response of transformers make it possible to estimate the windings’
dimension and consequently calculating the parameters of the detailed RCLM model
using the estimated geometry. Hence the current approach tries to match the calculated
transfer function of the model to the measured one to estimate the geometry of the
winding. In this regard, the necessary geometric parameters and objective functions
must be defined properly. In addition a reliable optimization algorithm is needed to find
the optimum solution.

4.1 Unknown Parameters

A set of basic parameters are needed to calculate the electrical elements of the RCLM
model based on analytical formulae. These parameters are listed in Table 1 and include
the conductor’s dimension, dielectric property of the insulation and internal and radius
of winding and core.
1280 E. Rahimpour et al.

Table 1. Basic parameters of the winding

Description Notation
Width of conductor aw
Height of conductor bw
Inner radius of winding ri
Radius of core rc
Paper thickness of one winding Pw
Pressboard thickness between core and winding Pbw
Axial distance between two disks d
Dielectric loss factor tan d
Turns per disk TPD

4.2 Geometry Estimation and Objective Function

Defining a suitable objective function is a very important task to achieve the best
parameters. In fact, model accuracy depends to the precision of the parameters. In this
step, the winding geometry is determined by minimizing a proper objective function.
The objective function which takes into account square relative error and correlation
between calculated and measured transfer functions gives satisfactory results. Thus the
objective function is defined as:
0 11
P
K

K  2 ð Þ
 
B ðfi ÞTFsim ðfi Þ C
P TFexp ðfi ÞTFsim ðfi Þ
TFexp
OF ¼ þ bB r
@ P ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
i¼1
þ 1C
A
TFexp ðfi Þ K P K
i¼1
ðTFexp i Þ
 ðf Þ
2
ðTFexp ðf Þ
i Þ
2
ð1Þ
i¼1 i¼1

P
K
TF  ðfi Þ ¼ jTFðfi Þj  K1 jTFðfi Þj
i¼1

Where:
TFexp = Measured transfer function
TFsim = Simulated transfer function
b = Balancing coefficient
fi = Spectral frequency
k = Number of samples
The measured transfer function is the input impedance of the transformer winding
and it is calculated by dividing the applied voltage at top terminal of the winding to the
input current in frequency domain:
 
Vinput ðj xÞ
TFexp ðj xÞ ¼ Zin ðj xÞ ¼   ð2Þ
Iinput ðj xÞ
 Dimension Estimation of Transformer Windings Based on FR Measurement 1281

Where:
Vinput = Applied voltage
Iinput = Input current of the winding
TFexp ðj xÞ = Measured transfer function
Zin ðj xÞ = Input Impedance
x = Spectral frequency.

4.3 Optimization Algorithm

The present geometry estimation problem can be solved by classic mathematical
methods or by artificial intelligence algorithms. In the current work, intelligent algo-
rithms are preferred to solve the optimization problem. Among many existing algo-
rithm, PSO is implemented to find the optimum solution. Since PSO is a well-known
algorithm, a brief summery is presented here:
– Setting the algorithm parameters
– Generating the random positions and velocities of particles
– Initializing the local and global best positions
– Evaluation of the objective function for each particle
– Updating velocities and positions
– Updating the local and global best positions.

5 Result Evaluation

In this chapter the identified geometry are presented and the calculated transfer func-
tions based on the estimated geometry are compared with the measured transfer
functions.

5.1 Estimated Dimensions

Winding dimensions are identified by several run of the PSO algorithm and the best
results are given in Table 2. It must be mentioned that the number of model units in
RCLM model is determined according to the existing resonances in the terminal fre-
quency response of the winding. Because it has been assumed that there is no infor-
mation of the internal structure of the winding.
As it is seen from Table 2, differences between the identified dimension and real
dimension are unavoidable. Besides, some dimensions have large differences which are
not of concern. The reason is that the number of model units is assumed twice the
number of terminal frequency response. Thus, it is normal to obtain such results. In the
other words, the main goal is to obtain the precise frequency response, not the design
parameters.
1282 E. Rahimpour et al.

Table 2. Estimated parameters (dimensions are in mm)

Parameter Estimated Design data
aw 1.86 2
bw 24.8 7.3
ri 411 270
rc 391 260
pw 0.37 0.25
pbw 7.86 10
d 11 4
tan d  103 1.56 1
TPD 6 7

35
Measured TF
30 Calculated TF

25
TF (V/mA)

20

15

10

0
0 200 400 600 800 1000
Frequency (KHz)

Fig. 3. Comparison of calculated and measured transfer functions

5.2 Terminal Frequency Response

By knowing the winding dimensions and analytical formulae, the electrical elements of
the RCLM model can be calculated. Consequently, any transfer functions can be
computed easily. In Fig. 3 measured frequency response at top terminal is compared
with the calculated transfer function.
Figure 3 confirms that all of the resonance points except the first resonance of the
winding are appeared in the frequency response of the model precisely. Also the
amplitudes of the transfer function at resonance points have acceptable accuracy in
comparison with the measured transfer function.

5.3 Internal Frequency Response

Internal voltage computation helps the engineers to analyze the condition of the
transformer winding after a failure. In addition to terminal frequency response, the
model must be able to simulate the internal frequency response of the winding pre-
cisely. In this section, the internal behavior of the winding is investigated.
 Dimension Estimation of Transformer Windings Based on FR Measurement 1283

In order to verify the internal frequency response, voltage transfer functions of 17th
and 25th disks are measured and compared with the simulated transfer functions. The
measured voltage transfer functions are computed by the following equations:

jVDisk 17 ðj xÞj
TFV 17 ðj xÞ ¼   ð3Þ
Vinput ðj xÞ

jVDisk 25 ðj xÞj
TFV 25 ðj xÞ ¼   ð4Þ
Vinput ðj xÞ

Where:
Vinput = Applied voltage
VDisk 17 = Voltage of disk 17
VDisk 25 = Voltage of disk 25

6 7
Measured TF Measured TF
Calculated TF 6 Calculated TF
5
5
TF (VDisk 25 / Vin)
TF (VDisk 17 / Vin)

4
4
3
3

2 2

1
1
0
0 200 400 600 800 1000
0
0 200 400 600 800 1000 Frequency (KHz)
Frequency (kHz)
(a) (b)

Fig. 4. Comparison between measured and simulated internal voltages: (a) 17th disk voltage,
(b) 25th disk voltage

Figure 4a and b compare the simulated and measured internal voltage transfer
functions. As it is seen the amplitude and frequency of resonance points are simulated
precisely and so that the resulting frequency responses from simulation are fitted to the
measured transfer functions.

6 Conclusion

Winding geometry estimation to obtain RCLM model of a transformer winding is

investigated in this paper. The geometry determination is defined as an optimization
problem. By dimension estimation of the transformer winding a few parameters is
needed to construct the RCLM model.
1284 E. Rahimpour et al.

PSO is utilized to seek the optimum solution of the problem. The presented RCLM
model based on geometry estimation satisfies both terminal and internal frequency
response. Thus, the method is applicable to calculate internal voltages to analyse fast
transient events in a transformer winding.

References
1. Bagheri, M.: Transformer frequency response: a new technique to analyze and distinguish
the low-frequency band in the frequency response analysis spectrum. IEEE TEIM 34(5), 39–
49 (2018)
2. Jahan, M.S.: Locating power transformer fault based on sweep frequency response
measurement by a novel multistage approach. IET SMT 12(8), 949–957 (2018)
3. Eslamian, M.: New methods for computation of inductance matrix of transformer windings
for very fast transient studies. IEEE TPD 27(4), 2326–2333 (2012)
4. Sahnoune, M.: BA to construction of equivalent circuit of a transformer winding from
frequency response measurement. IET EPA 12(5), 728–736 (2018)
5. Mitchell, S.D.: Initial parameter estimates and constraints to support gray box modeling of
power transformers. IEEE TPD 28(4), 2411–2418 (2013)
6. Chanane, A.: Investigation of the transformer winding high-frequency parameters identi-
fication using particle swarm optimization method. IET EPA 10(9), 923–931 (2016)
7. Rashtchi, V.: Parameter identification of transformer detailed model based on chaos
optimization algorithm. IET EPA 5(2), 238–246 (2011)
8. Blanken, P.G.: A lumped winding model for use in transformer models for circuit
simulation. IEEE TPE 16(3), 445–460 (2001)
9. Hassan Hosseini, S.M.: New high frequency multi-conductor transmission line detailed
model of transformer winding for PD study. IEEE TDEI 24(1), 316–323 (2017)
10. Rahimpour, E.: Transfer function method to diagnose axial displacement and radial
deformation of transformer windings. IEEE TPD 18(2), 493–505 (2003)
 Estimation of Number of Model Units
in Transformer Detailed RCLM Model Based
on Terminal Measurement in the Case
of Unavailable Design Data

Ebrahim Rahimpour1(&), Vahid Rashtchi2, and Reza Aghmashehr3

1
ABB AG, R&D Department, Bad Honnef, Germany
ebrahim.rahimpour@de.abb.com
2
Department of Electrical Engineering, University of Zanjan, Zanjan, Iran
3
IGMC, National Grid Dispatching Deputy, Tehran, Iran

Abstract. Providing a precise transformer model for fast transient analysis

phenomena requires detailed design information and geometry. However due to
restrictions on the preservation of design documents in factories or their
destruction over time, the design documentation of transformers are not always
available. On the other hand, a valid detailed model to study high frequency
events is a necessary need in the absence of information. Detailed RCLM (R:
Resistance, C: Capacitance, L: Self-inductance, M: Mutual inductance) model is
a reliable and applicable model for calculation of internal voltage of trans-
formers’ winding when they are subjected to fast transient phenomena. In case
of insufficient knowledge of the windings topology, the number of model units
and parameters of each model unit in RCLM model cannot be determined
properly and consequently internal voltage calculation would not be calculated
accurately. This work presents a new method to identify the optimum value for
number of model units in detailed RCLM model based on measured transfer
function at transformers’ windings terminals. Two cases are investigated in this
study. In case one, the frequency dependency effect of the winding is discarded
while in the second case, the series resistance of the winding is considered as a
frequency dependent variable. Covariance matrix adaptation evolutionary
strategy is utilized to estimate the optimum value for the model unit numbers.
The methodology is verified by comparing the simulated and measured internal
voltages on a real transformer winding.

Keywords: Model units
Winding geometry
Frequency response

1 Introduction

Modelling of power system equipment is an important task for engineers to simulate

and analyse the behaviour of the system in different operation conditions. Among them,
transformers are of high importance due to critical role in power transmission and high
price [1, 2]. Switching in a power transmission system is a daily routine maneuver and
produces impulse voltages in a power network. Thus, knowing the status of trans-
formers when faced to fast transient high voltage wave shapes is necessary [3].
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1285–1294, 2020.
https://doi.org/10.1007/978-3-030-31676-1_120
1286 E. Rahimpour et al.

Li ,1 Li , n
Li ,i −1 Li ,i +1
Top Terminal

1 i −1 i i +1 n
Rs1 L1,1 R s (i −1)L(i −1),(i −1) R si Li ,i R s (i +1) L(i +1),(i +1) R sn Ln ,n
ZE

RK 1 R K (i −1) R Ki R K (i +1) R Kn

K1 K (i −1) Ki K (i +1)
Kn
C e1 Ge 1 C e (i −1) Ge (i −1) C ei Gei C e (i +1) Ge (i +1) C en Gen

Fig. 1. High frequency detailed model with n sections

Physical based transformers’ models for fast transient analysis are classified into
two groups: lumped parameters [4] or distributed parameter models [5]. Both models
are considered as white box or internal models [6]. Although these models present the
internal voltage distribution of transformers’ windings precisely, but they need accurate
geometry of the internal structure of transformers [7].
In case of no knowledge of internal geometry, the physical models could be
extracted by terminal tests, nameplate data and external dimension of the apparatus [8].
Frequency response of a transformer winding is a well-known tool to identify the
parameters of a fast transient model because it could provide internal characteristics of
the winding at winding terminals [9, 10].
Computing the internal voltage of transformers requires precise determination of
model parameters. In this study, parameters of detailed RCLM model are estimated in
case of unavailable design data and geometry. The model which is a simplified model
and its parameters are determined by experiments is called gray box model. One of the
most important parameters in gray box RLCM model is the number of model units
which it estimated by the help of artificial intelligence method. Also, the frequency
dependencies of loss parameters are taken into account and the model validity is
checked by experimental results.

2 High Frequency Model Based on Lumped Parameters

Figure 1 shows RCLM model of a typical winding. The model consists of n sections.
The number of sections is a modelling parameter and must be determined precisely to
calculate the nodal voltages accurately. In fact, the precision of the model in describing
the internal behavior of the windings depends upon the number of RCLM model
sections and parameters’ accuracy of each section.
The model parameters are explained in detail in [11]. The RCLM model can be
solved either in time domain or in frequency domain. Since the frequency dependent
behavior of the winding’s loss is considered in this investigation, the frequency domain
approach is preferred.
 Estimation of Number of Model Units in Transformer Detailed RCLM Model 1287

1. 1234567
258.5
4
2. 7654321 269.5
Conductor cross section
3. 1234567
paper thickness
288 =0.25
Press 2
board 515
Aluminum
43. 1234567 cylinder for 7.3 CU 7.8
modeling the core
44. 7654321

2.5
Fig. 2. Test winding geometry and dimension

3 Test Winding Description

A winding with 44 non-interleaved disks are prepared to evaluate the simulation

results. Each disk contains 7 turns. The measurements are performed in the case of the
winding is removed from iron core and transformer tank. To model the earth potential
of the iron core, an aluminum shield is place inside the winding. Both the aluminum
shield and bottom terminal of the winding are grounded at time of measuring the
frequency response of the winding. More information of the winding layout and
geometry is presented in Fig. 2.

4 Estimating the Number of Model Units

Gray box models are used when the design data of a transformer winding is not
available and there is necessary need to model the winding. In this case the parameters
of the model must be estimated by special techniques. Gray box models usually are
used when a transformer failure is occurred and engineers or researchers try to find the
defect factor. Accurate computation of internal voltages helps them to find the voltage
stress inside the winding. Hence estimating the number of model units in the RCLM
model and its parameters is a solution to calculate of the internal disks voltages. This
chapter describes the method of number of model units’ estimation in case of the
design data is not available. The methodology deals with estimation of transformers’
windings geometry and dimension and then using the analytical formulae to calculate
1288 E. Rahimpour et al.

the electrical elements of the RCLM model. Two cases are investigated in the current
essay. In the first case, the winding loss is considered as a constant value while in the
second case, the frequency dependency of the winding is considered.

4.1 Case 1
The electrical elements of RCLM model of a transformer winding are calculated using
the geometry design of the winding. In this regard several geometry parameters are
needed to be estimated all of the elements of the RCLM model. These parameters not
only include dimensions of the model, but also contain the characteristic of insulation
system and structure of the model. The most important parameter which clarifies the
precision of internal voltage computation is the number of model units. Estimation of
number of model units in RCLM model is investigated in this section. The unknown
geometric parameters of a transformer winding to present RCLM model considering a
constant value for winding losses are shown in Fig. 3a. The description of the unknown
parameters in Fig. 3 is as follows:

aw and bw : Width and height of conductor; d: Axial distance between two disks;
ri : Inner radius of winding; tan d: Dielectric loss factor;
rc : Radius of core; Rs : Series resistance of the winding;
Pw and Pbw : Paper and pressboard thickness; N: Number of model units;
TPD: Number of turns in each disk;

According to the abovementioned parameters by identifying 11 parameters, the

detailed RCLM model elements can be calculated.

4.2 Case 2
The only difference of this case with the case 1 in Sect. 4.1 is the way of modeling the
series resistance of the winding. In practice the winding loss is not a constant value and
the value of series resistance in RCLM model changes with the variation of frequency
due to skin and proximity effects. Hence in this chapter the series resistance of the
winding is considered as a frequency variable parameter. In this condition, the series
resistance of the winding is calculated based on the estimated geometry and existing
analytically formulae instead of determining individually as presented in case 1.
Therefore the unknown parameters in case 2 would be as shown in Fig. 3b.

aw bw ri rc Pw Pbw d tan δ Rs N TPD aw bw ri rc Pw Pbw d tan δ N TPD

(a) (b)

Fig. 3. Unknown parameters: (a) case 1, (b) case 2

As it is shown in Fig. 3b, one of the unknown parameters has been reduced. Thus
the number of unknown parameters decreases to 10 parameters. The advantage of
 Estimation of Number of Model Units in Transformer Detailed RCLM Model 1289

determining of model parameters based on geometry estimation is that a few number of

unknown parameters are need to be identified.
In the following chapters, the objective function and the optimization algorithm to
estimate the introduced parameters in case 1 and case 2 are explained.

5 Objective Function and Optimization Approach

To identify the optimum value of the unknown geometric parameters both classic
gradient approaches and artificial intelligence methods can be used. Due to high effi-
ciency and easy implementation artificial intelligence method is utilized in this paper.
The parameter determination problem is converted to an optimization problem.
Hence, a proper fitness function and appropriate optimization algorithm is required to
reach the best solution. The fitness function is constructed based on relative square
error and correlation factor and is defined in the following:

FF ¼ FFa þ bFFb

ZinðexpÞ  ZinðsimÞ 2
ð1Þ
FFa ¼

;
FFb ¼ corr ZinðexpÞ ; ZinðsimÞ
ZinðexpÞ

where:
ZinðexpÞ = Measured input impedance;
ZinðsimÞ = Simulated input impedance
corr = Correlation factor between simulated and measured input impedance
b = Balancing coefficient
The input impedance versus frequency which is considered as the winding transfer
function is calculated by dividing the input voltage to the input current of the winding.
Measurements are performed in time domain and then transformed to frequency
domain.
There are various artificial intelligence methods such as genetic algorithm (GA),
particle swarm optimization (PSO), simulated annealing (SA), imperialist competitive
algorithm (ICA), covariance matrix adaptation evolutionary strategy (CMA-ES) to
optimize the fitness function. Approximately they could give same results and they
differ only in run time and way of codes’ implementation. In this work, CMA-ES is
used to minimize the introduced fitness function for both case 1 and case 2. The
algorithm is a repetitive algorithm and contains several operators. The flowchart of the
optimization algorithm is shown in Fig. 4.
The CMA_ES algorithm seeks for the best solution in a search space in which the
unknown parameters are bounded to a minimum and maximum value. The existing
constrains are determined empirically by the knowledge of the user. Since the geometry
design is not accessible, experimental knowledge of the user would affect the con-
vergence speed of the algorithm.
1290 E. Rahimpour et al.

Start
Selection:
Selecting the best search
Initializing: point for recombination
1)Setting the algorithm parameters
2)Producing a population

Recombination:
Forming new search points
Evaluation of the objective in the population
function for each search
points
Step Size Control:
1) Updating the conjugate
evolution path
Termination No 2) Updating the step size
criterion
Satisfied?
Covariance Matrix Adaptation
Yes 1) Updating Evolution Path
2) Updating Covariance
Matrix
End

Fig. 4. Flowchart of CMA-ES

6 Analysis of Results

In this chapter, the identified geometry and the validity of model are evaluated. The
model validation is performed by comparing both input-output and internal frequency
response the winding. Also the results are analysed by numerical comparison of
winding responses based on suitable indexes.

6.1 Identified Parameters

The identified geometric parameters for both cases are presented in Table 1. As it is
seen in Table 1, there is large difference between some parameters of the winding. The
reason is that all the parameters have been estimated simultaneously and the number of
turns of each disk and number of model units are considered as unknown parameters
and they are identified in the optimization process.
It is worth mentioning that the winding parameters would be identified close to the
real dimensions if the number of model units and turns of each disk are not unknown.
The series resistance of the winding in both cases is shown in Fig. 5. The estimated
resistance in case 2 varies with frequency changes. Since the optimization algorithm
tries to fit the frequency response in the entire frequency range, the constant series
resistance lies between the minimum and maximum values of identified resistance in
case 2.
 Estimation of Number of Model Units in Transformer Detailed RCLM Model 1291

Table 1. The Winding geometry determined through case 1 and case 2 (dimensions are given in
mm).
Parameter Case 1 Case 2 Design data
aw 1.86 2.2 2
bw 24.8 16.4 7.3
ri 411 304 270
rc 391 289 260
pw 0.37 0.18 0.25
pbw 7.86 13.8 10
d 11 3.65 4
tan d  103 1.56 1.1 1
N 16 18 22
TPD 6 6 7

18
Case 1
16
Case 2
14

12
10
Rs

8
6
4

2
0
0 200 400 600 800 1000
Frequency (KHz)

Fig. 5. Identified series resistance through case 1 and case 2

6.2 Evaluation of Frequency Response of the Winding

After estimating the dimensions of the winding, the electrical parameters of the RCLM
model can be calculated by the analytical formulae. Consequently the RCLM model
can be solved by simple electric circuit laws. Therefore, the input impedance transfer
function of the winding can be computed simply. The simulated input impedance
transfer function through case 1 and case 2 is compared with the measured one in
Fig. 6.
In addition to input-output frequency response validation, input response of the
winding is verified by measuring the internal voltage transfer function of disk 17 and
25. The voltage transfer functions are calculated by dividing voltage of 17th and 25th
disks to the applied voltage at top of the winding. Calculated internal voltage transfer
functions for disk 17 and 25 are compared with measured transfer functions in Fig. 7a
and b respectively.
1292 E. Rahimpour et al.

35
Measured TF
30 Calculated TF by Case 1
Calculated TF by Case 2
25

TF (V/mA) 20

15

10

0
0 200 400 600 800 1000
Frequency (KHz)

Fig. 6. Comparison of calculated and measured transfer functions

6 7
Measured TF Measured TF
Calculated TF by Case 1 6 Calculated TF by Case 1
5
Calculated TF by Case 2 Calculated TF by Case 2
5
TF (VDisk 25 / Vin)
TF (VDisk 17 / Vin)

4
4
3
3

2 2

1 1

0
0 0 200 400 600 800 1000
0 200 400 600 800 1000 Frequency (KHz)
Frequency (kHz)
(a) (b)

Fig. 7. Voltage transfer function: (a) Disk 17, (b) Disk 25

Figures 6 and 7 confirm that the simulated RCLM model based on identified
geometry and design data can describe both terminal and internal frequency response of
the winding precisely. To compare the simulation and measurement with more details,
the resonance points of the transfer function are analyzed with more consideration. In
this regard, the frequency and amplitude deviation of the resonance points for input
impedance and internal voltage transfer functions are calculated using the following
indexes:

Dfi fsim;i  fexp;i

IFDi ¼  100% ¼  100% ð2Þ
fexp;i fexp;i

DAi Asim;i  Aexp;i

IADi ¼  100% ¼  100% ð3Þ
Aexp;i Aexp;i
 Estimation of Number of Model Units in Transformer Detailed RCLM Model 1293

where:
IFD = Index of frequency deviation
IAD = Index of amplitude deviation
f = Frequency of resonance points
A = Amplitude of resonance points
IFD and IAD of input impedance and voltage transfer functions are given in
Tables 2, 3 and 4. The results indicate that the amplitude and frequency of resonance
points in both cases are calculated precisely. Also, the indexes confirm that considering
the frequency dependent resistance of the winding in case 2, improves the amplitudes
of resonance points significantly when compared with case 1.

Table 2. IFD and IAD for input impedance transfer function in case 1 and case 2
IFD (%) Case 1 Case 2 IAD (%) Case 1 Case 2
IFD1 70.77 67.49 IAD1 −45.14 −2.76
IFD2 0 0 IAD2 37.06 22.94
IFD3 0 0 IAD3 32.57 16.57
IFD4 0 1.96 IAD4 14.39 17.62
IFD5 0.76 −0.76 IAD5 −8.04 −9.5
IFD6 −1.26 −3.15 IAD6 −23.74 −15.14

Table 3. IFD and IAD for voltage transfer function of disk 17 in case 1 and case 2
IFD (%) Case 1 Case 2 IAD (%) Case 1 Case 2
IFD1 −3.22 −3.22 IAD1 −12.44 2.59
IFD2 −1.54 −1.54 IAD2 −22.02 −10.40
IFD3 0 0 IAD3 −2.68 3.08
IFD4 1.61 0.81 IAD4 −37.51 −13.88
IFD5 1.33 0 IAD5 −52.01 −38.30

Table 4. IFD and IAD for voltage transfer function of disk 25 in case 1 and case 2
IFD (%) Case 1 Case 2 IAD (%) Case 1 Case 2
IFD1 0 0 IAD1 −1.08 −0.76
IFD2 0 −1.56 IAD2 28.36 6.72
IFD3 1.02 0 IAD3 −25.36 −10.14
IFD4 0.81 0.81 IAD4 −4.96 −9.92
IFD5 0 −0.66 IAD5 −52.27 −21.74

7 Analysis of Results

In this paper two cases are investigated to estimate the number of model units of
transformers windings’ RCLM model based on terminal measurements. The method is
applicable when the design data and geometry are not available. The outcome of the
current work is listed below:
1294 E. Rahimpour et al.

– The number of model units of a transformer winding in RCLM is identified using

terminal frequency response of the winding.
– Artificial intelligence methods are able to find the optimum value of design
parameters of a winding for calculating the RCLM model parameters.
– The RCLM model based on identified parameters is able to simulate the internal
response of the winding.
– A few parameters is needed to be identified to calculate the electrical elements of
the winding’s RCLM model.

References
1. Popov, M.: Modeling, simulation and measurement of fast transients in transformer windings
with consideration of frequency-dependent losses. IET EPA 1(1), 29–35 (2007)
2. Eslamian, M.: New equivalent circuit of transformer winding for the calculation of resonance
transients considering frequency-dependent losses. IEEE TPD 30(4), 1743–1751 (2015)
3. Smajic, J.: Simulation and measurement of lightning-impulse voltage distribution over
transformer windings. IEEE TM 50(2), 553–556 (2014)
4. Rahimpour, E.: Applying artificial optimization methods for transformer model reduction of
lumped parameter models. Elsevier EPSR 84(1), 100–108 (2012)
5. Liang, G.: Modeling of transformer windings under very fast transient overvoltages.
IEEE TEC 84(4), 621–627 (2006)
6. López, Z.L.: Modelling of transformer windings for fast transient studies: experimental
validation and performance comparison. IEEE TPD 32(4), 1852–1860 (2017)
7. Kazemin, R.: Estimation of design parameters of single-phase distribution transformers from
terminal measurements. IEEE TPD 32(4), 2031–2039 (2017)
8. Mossad, M.I.: Transformer parameters estimation from nameplate data using evolutionary
programming techniques. IEEE TPD 29(5), 2118–2123 (2014)
9. Aghmasheh, R.: Gray box modeling of power transformer windings for transient studies.
IEEE TPD 32(5), 2350–2359 (2017)
10. Aghmasheh, R.: Gray box modeling of power transformer windings based on design
geometry and particle swarm optimization algorithm. IEEE TPD 33(5), 2384–2393 (2018)
11. Rahimpour, E.: Experimental and theoretical investigation of disk space variation in real
high-voltage windings using transfer function method. IET EPA 4(6), 451–461 (2010)
 Fractal-Based Approach for Modelling Electric
Breakdown of Air Gaps: An Application
to a 75 cm Positive Rod-Plane Gap

A. I. Ioannidis(&) , P. N. Mikropoulos , T. E. Tsovilis ,

and N. Karanikiotis

High Voltage Laboratory, School of Electrical and Computer Engineering,

Faculty of Engineering, Aristotle University of Thessaloniki,
Building D, Egnatia Str., 54124 Thessaloniki, Greece
ialexios@ece.auth.gr

Abstract. A stochastic model for the dynamic simulation of the spatial growth
of electrical discharges as well as for assessing breakdown of air gaps is pre-
sented. The fractal-based model visualizes the stochastic behavior of discharge
development, in terms of channel branching and tortuosity, and yields a sta-
tistical distribution for the dielectric strength of the gap as affected by discharge
evolution. The main algorithm of the proposed simulation model, developed in
MATLAB environment, is applicable in both two (2D) and three (3D) dimen-
sions and considers as influencing parameters on discharge evolution: the critical
field required for discharge evolution, the voltage drop along the discharge
channel, the propagation parameter “η”, commonly employed in fractal studies,
to account for the dependence of propagation probability on the electric field, as
well as the size of the mesh. The electric potential in the gap is calculated at each
step of discharge evolution by using the finite difference method (FDM) with a
successive over relaxation (SOR) technique. An application of the model is
made to obtain the breakdown probability distribution of a 75 cm positive rod-
plane air gap. Simulations are conducted adopting the “multiple-level tests”
method according to IEC 60060-1, so as to compute the 50% breakdown voltage
and standard deviation. Simulation results are evaluated and discussed with
reference to the experimentally obtained breakdown probability distribution; a
very good agreement is shown to exist. Values of the model key parameters are
discussed on the basis of 2D and 3D simulations.

Keywords: Electrical discharges
Electric breakdown
Fractal-based model

1 Introduction

Electrical discharges in gaseous, solid and liquid dielectrics are strongly characterized
by features of branching and tortuosity, both in prebreakdown and breakdown phe-
nomena. Since 1982, when Mandelbrot [1], firstly introduced the fractal geometry
many researchers have employed the concept of fractals and their main characteristic of
self-similarity, in order to simulate the stochastic behavior of electrical discharge
evolution [2–5].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1295–1305, 2020.
https://doi.org/10.1007/978-3-030-31676-1_121
1296 A. I. Ioannidis et al.

Various stochastic models were introduced to simulate the dynamic growth of

electrical discharges and to describe these complex patterns taking into account their
fractal characteristics. They were implemented to a number of applications such as
simulation of surface discharges evolution (Lichtenberg figures) [2, 6], breakdown and
pre-breakdown phenomena in dielectrics [3, 7–9] as well as lightning protection [10–
12]. The major asset of all these models is their ability to combine both the deter-
ministic and statistical features of electrical discharges so as to consider the stochastic
nature of dielectric breakdown and derive more accurate results on discharge evolution.
In this study a stochastic model for the simulation of the spatial growth of electrical
discharges as well as for obtaining breakdown probability distribution of air gaps is
presented. The discharge evolution is simulated through steps of development, by
calculating at each step the electric potential in space. The fractal-based model visu-
alizes the stochastic behavior of discharge development in terms of channel branching
and tortuosity and yields a statistical distribution for the dielectric strength of the gap as
affected by the boundary conditions. An application of the fractal-based model is made
for a 75 cm positive rod-plane air gap. Simulation results are evaluated and discussed
with reference to the experimentally obtained breakdown probability distribution under
positive lightning impulse voltages. Values of key parameters of the model are dis-
cussed based on physical interpretations of simulation results considering both the 2D
and 3D approaches.

2 Model Description
2.1 Main Algorithm
The main algorithm of the proposed stochastic model has been developed in the
MATLAB environment. The space adopted in the simulations is discretized in both two
and three dimensions using an appropriate mesh size. The electrical discharge emanates
from the high voltage electrode, and progresses towards the grounded electrode.
Downward discharge evolves in steps, and at each step, which is equal to mesh size, the
electric potential, V, is calculated at the whole space. Firstly, the average potential
gradient between all nodes P that are already part of the discharge channel and their
surrounding points P′ are calculated from the equation:




VP  VP0

EPP0 ¼


ð1Þ
DPP0

where VP and VP′ are the electric potentials of points P and P′, respectively and DPP′ is
the distance between P and P′ (Fig. 1).
Possible candidate points of discharge progression are only those that fulfil the
following propagation criterion:

EPP0  Ecr ð2Þ

where Ecr is a critical electric field value required for discharge evolution.
 Fractal-Based Approach for Modelling Electric Breakdown of Air Gaps 1297

a)
b)

Fig. 1. Full dot (•) denotes a discharge node, P, and empty dots (◦) the surrounding points, P′, of
the node; (a) 2D and (b) 3D.

For example, considering that P is a point of the discharge then up to 8 surrounding

points in 2D (Fig. 1a) and up to 26 surrounding points in 3D (Fig. 1b), including
diagonal ones, are possible candidate points, M. Thus, the total candidate points at each
step are M < 8N (Fig. 2, 2D) or < 26N (3D), where N is the number of the discharge
path nodes.
The propagation probability p(P ! P′) of a surrounding point P′ to become part of
the electrical discharge channel is calculated by [12]:
8
> ðEPP0 Þg
<P N PMi ; for EPP0  Ecr
0
pðP!P Þ
¼ ðEPP0 Þg ð3Þ
>
: i¼1 j¼1
0; for EPP0  Ecr

where η is a propagation parameter [2]. This parameter accounts for the dependence of
propagation probability on the electric field. Lower values of η yield more intense
branching and tortuosity for the discharge channel, enhancing the stochastic behavior
in discharge growth.
The denominator in Eq. (3) refers to the summation of all possible candidate dis-
charge points Mi from all discharge nodes N and as a result normalizes the propagation
probability distribution [13]. A number uniformly distributed in the interval (0, 1) is
randomly obtained and by using the cumulative distribution function of p(P ! P′)
(Eq. 3) the next discharge channel point P′ is chosen [14]. The model belongs to single
element category; only one candidate point is selected per step. Then, the electric
potential V of the newly added to the discharge channel point is updated using the
formula:

VP ¼ V0  s
Ech ð4Þ

where V0 is the electric potential at the point of origin of the discharge, Ech is an
average electric field accounting for the voltage drop along the discharge channel, and
s is the length of the discharge along its path, between the point of origin and the newly
added discharge point.
1298 A. I. Ioannidis et al.

Fig. 2. Discharge points for a 2D simulation scenario. Full dots (•) denote the discharge nodes,
empty dots (◦) the candidate points and crosses () points that do not fulfil the propagation
criterion.

This procedure continues iteratively until either no point satisfies the propagation
criterion (Eq. 2) (withstand case) or when the electrical discharge channel reaches the
grounded electrode (breakdown case).

2.2 Electric Potential Calculation

Electric potential calculation is the most demanding part in the stochastic model,
leading to huge computational cost. In the present model, the finite difference method
(FDM) is used in both two and three dimensions in order to solve Laplace’s equation
and calculate the electric potential V at every point in the simulation area.

r2 V ¼ 0 ð5Þ

From the theory of partial differential equations (PDE’s), in a computational

domain discretized using equal grid spacings in all directions and neglecting higher
order terms in Taylor’s expansion, Laplace’s Eq. (5) is transformed into the below
discretized form in 2 dimensions

1
Vi; j ¼
ðVi þ 1; j þ Vi1; j þ Vi; j þ 1 þ Vi; j1 Þ; ð6Þ
4
which is known as the “5-point approximation’’ and to the “7-point approximation’’ in
3 dimensions respectively [15].

ðVi þ 1; j;k þ Vi1; j;k þ Vi; j þ 1;k þ Vi; j1;k þ Vi; j;k þ 1 þ Vi; j;k1 Þ
Vi; j;k ¼ ð7Þ
6
 Fractal-Based Approach for Modelling Electric Breakdown of Air Gaps 1299

In order to solve the above equations, a successive-over relaxation (SOR) technique

is used, which is found to significantly reduce the calculation time and reach faster to
convergence than other iterative methods [15, 16].

Vi;n jþ 1 ¼ ð1  xÞ
Vi;n j þ x
Vi; j ð8Þ

where

1
Vi; j ¼ nþ1

ðVinþ 1; j þ Vi1; nþ1
j þ Vi; j þ 1 þ Vi; j1 Þ
n
ð9Þ
4

and Vi;n jþ 1 denotes the (n + 1)th estimate of the potential solution [16]. Equations (8)
and (9) can be derived for 3 dimensions after appropriate adjustments. The over-
relaxation parameter, x, takes values between 1 and 2 and the optimal value speeds up
the simulation process [15, 16]. In this study the values of 1.9 and 1.6 [17] were used
for 2D and 3D simulations, respectively.
The electric potential is calculated by iteratively solving Eq. (8) until the relative
error in the potential for all points, between two successive iterations n and n + 1 of the
SOR algorithm, is smaller than 1%. Two types of boundary conditions were applied.
Dirichlet boundary conditions (V = ct.) were imposed on objects with fixed potential
(electrodes), as well as on discharge nodes. Neumann type boundary conditions
(dV/dx = 0), were chosen for the lateral boundaries defining the simulation area. When
a new point is added to the discharge channel, its potential is updated according to
Eq. (4) and then, with the new boundary conditions imposed, the potential in the whole
space is recalculated. Consequently, the modification of the potential at every new
discharge point affects the whole simulation and confers on it a dynamic attribute.

3 Model Application

An application of the fractal-based model is presented for the case of a rod-plane air
gap 75 cm in length. Under positive applied voltages, breakdown probability curves
were obtained by performing 2D and 3D simulations. The key parameter values shown
in Table 1 were employed in simulations, yielding satisfactory agreement with
experimental data and reasonable simulation time. For the average electric field
accounting for the voltage drop along the discharge channel, Ech, a value of 4.5 kV/cm
has been used, in accordance with the threshold positive streamer propagation field
[18]. The mesh size of 1 cm was considered adequate given the length of the gap and
computing effort; it must be noted that Ecr and η may vary depending on mesh size.
1300 A. I. Ioannidis et al.

Table 1. Key parameter values

Parameter 2D 3D
Ech (kV/cm) 4.5 4.5
Ecr (kV/cm) 11.5 21.5
η 1 15
Mesh size (cm) 1 1

3.1 Simulation Results

Figure 3 shows the discharge evolution for the case of 400 kV applied voltage at the
rod; it depicts both breakdown (Fig. 3a) and withstand (Fig. 3b) cases by visualizing
the stochastic behavior of discharge development in terms of channel branching and
tortuosity. It should be noted that as the discharge evolution is simulated dynamically,
the electric potential is recalculated at each step taking into account the updated value
of electric potential of the newly added point, thus the discharge path is computed
under the interaction of all discharge branches.

a)

b)

Fig. 3. Discharge growth under 400 kV applied voltage; 75 cm rod-plane gap, values in cm
(a) breakdown, (b) withstand.
 Fractal-Based Approach for Modelling Electric Breakdown of Air Gaps 1301

12
3D
2D
10

Electric potential (kV)

8

0
0 0.2 0.4 0.6 0.8 1
Simulation steps / total steps (p.u.)

Fig. 4. Variation of the electric potential at 1 cm above the grounded plane at gap axis as the
downward discharge propagates; 2D (93 steps) and 3D (1040 steps). 400 kV applied voltage
causing breakdown.

As an illustrative example, Fig. 4 shows the increase of the electric potential at

1 cm above the grounded plane at gap axis as the downward discharge develops
(simulation steps); it is evident that 2D simulations, although significantly faster, are
associated with overestimated values of electric potential in the gap.
Figure 5 shows the variation of the electric potential at 1 cm above the grounded
plane as the downward discharge propagates; evidently, as compared to the withstand
case, breakdown is associated with a sharp increase in the electric potential at the later
stages of discharge evolution.

6
breakdown
withstand
5
Electric potential (kV)

0
0 200 400 600 800 1000
Simulation steps

Fig. 5. Variation of the electric potential at 1 cm above the grounded plane at gap axis as the
downward discharge propagates. Comparison between a breakdown and a withstand case.
400 kV applied voltage, 3D simulations.
1302 A. I. Ioannidis et al.

100%
2D
90%
3D
80%
Experimental data
70%

60%
p (%)

50%

40%

30%

20%

10%

0%
360 370 380 390 400 410 420 430 440
Up (kV)

Fig. 6. Breakdown probability distributions of the 75 cm rod-plane air gap under positive
applied voltages; fitting curves, were drawn according to Normal distribution.

In order to obtain the breakdown probability distribution of the investigated gap,

2D and 3D simulations were performed by adopting the “multiple-level tests” method
according to IEC [19]. 100 simulations (voltage applications) per voltage level were
run at gradually increasing voltage levels. Figure 6 shows simulation results together
with the corresponding breakdown probability distribution obtained experimentally
[20]. The probability distributions, were found to be well approximated with the
Normal distribution; thus, from each distribution the 50% breakdown voltage, U50, and
the associated standard deviation, r, were computed. These values, together with those
obtained experimentally, are listed in Table 2.

Table 2. U50 and r of 75 cm positive rod-plane gap

Parameter 2D 3D Experimental
U50 (kV) 395.2 395.7 395.9
r (%) 1.5 2.3 2.5
Total 1h 52 h 2h
time (100 voltage appl. per (100 voltage appl. per (20 voltage appl. per
voltage level) voltage level) voltage level)
Intel Core i7-8700, 3.2 GHz, 32 GB RAM

As evident from Fig. 6 and Table 2, there is an excellent agreement between

simulation and experimental results. It must be noted that the U50 values of Table 2
compare well with the U50 value of 397.5 kV obtained using the empirical expression
for lightning impulse breakdown voltage [21]

U50 ¼ 530
d ð10Þ

 Fractal-Based Approach for Modelling Electric Breakdown of Air Gaps 1303

and they are also higher (*13%) than that of the DC breakdown voltage of the
investigated gap [21].

3.2 Discussion
Rod-plane air gaps are axis-symmetrical configurations allowing for electrostatic field
and electrical potential computation in 2 dimensions. However, where in simulations
the growth of electrical discharges is of concern, the symmetry of the gap does not
apply. Actually, the more intense the branching and tortuosity of the discharge channel
the more asymmetric becomes the electric field distribution between the electrodes.
Thus, 3D simulations are more appropriate for calculating accurately the electric field
and potential distributions.
Nevertheless, because of the significantly lower computational cost related to 2D
simulations (Table 2), the latter may also be employed in engineering applications by
considering a planar approach; however, this requires appropriate adjustment of key
parameters values in simulations for compensating the overestimation of the electric
potential within the gap (Fig. 4) as well as the fact that discharge growth is limited in 2
dimensions. Actually, using the same mesh size (1 cm) and average gradient of positive
discharge (Ech = 4.5 kV/cm), to obtain a satisfactory agreement between simulation
and experimental results, lower values for the critical electric field for discharge evo-
lution, Ecr in Eq. (2), and propagation parameter, η in Eq. (3), had to be employed in
2D than 3D simulations. It is important to note that 3D simulation results, as compared
to 2D ones, (U50, r, tortuosity, branching) are more sensitive to η variations, due to the
significantly larger number of candidate points for discharge evolution.
Future work necessitates the integration of time to the fractal-based model, so as to
take into account the temporal variation of discharge evolution under voltages both
stable and variable in time. Also, a sensitivity analysis on the dependence of the model
key parameters on mesh size, gap length and electrode configuration is currently
undertaken; this, together with relevant experimental results, may allow for the physical
interpretation of the model parameters and simulation results.

4 Conclusions

A stochastic model for the simulation of electrical discharge spatial evolution and for
determining the dielectric strength of non-uniform air gaps has been introduced. Dis-
charge evolution is simulated dynamically, by recalculating the electric potential at
each growth step; thus, the path of the discharge is computed as the latter evolves under
the interaction of all discharge branches. The fractal-based model has been applied to
obtain the breakdown probability distribution of a 75 cm rod-plane air gap under
positive lightning impulse voltages. By selecting appropriate values for the model key
parameters, the stochastic behavior of discharge development in terms of channel
branching and tortuosity has been visualized. The computed 50% breakdown voltage
and standard deviation r are in excellent agreement with experimental results. Accurate
simulation of the dynamic growth of electrical discharges in air gaps requires problem
solution in 3 dimensions. 2D simulations may yield acceptable results on breakdown
1304 A. I. Ioannidis et al.

probability distribution of air gaps after appropriately adjusting key model parameters,
such as the critical electric field required for discharge evolution and propagation
parameter, both affecting branching and tortuosity of the discharge channel.

Acknowledgments. Results presented in this work have been produced using the AUTH
Computing Cluster. Authors would like to acknowledge the support provided by IT AUTH and
especially Ms. Alexandra Charalampidou, throughout the progress of this research work.

References
1. Mandelbrot, B.: The Fractal Geometry of Nature. W. H. Freeman and Co., New York (1982)
2. Niemeyer, L., Pietronero, L., Wiesmann, H.J.: Fractal dimension of dielectric breakdown.
Phys. Rev. Lett. 52(12), 1033–1036 (1984)
3. Wiesmann, H.J., Zeller, H.R.: A fractal model of dielectric breakdown and prebreakdown in
solid dielectrics. J. Appl. Phys. 60(5), 1770–1773 (1986)
4. Barclay, A.L., Sweeney, P.J., Dissado, L.A., Stevens, G.C.: Stochastic modelling of
electrical treeing: fractal and statistical characteristics. J. Phys. D Appl. Phys. 23(12), 1536–
1545 (1990)
5. Kudo, K.: Fractal analysis of electrical trees. IEEE TDEI 5(5), 713–727 (1998)
6. Femia, N., Niemeyer, L., Tucci, V.: Fractal characteristics of electrical discharges:
experiments and simulation. J. Phys. D Appl. Phys. 26(4), 619–627 (1993)
7. Noskov, M., Karpov, D., Lopatin, V., Pleshkov, O.: The simulation of the discharge
channels propagation in liquids. In: 12th International Conference on Conduction and
Breakdown in Dielectric Liquids, Roma, Italy (1996)
8. Gao, Y., He, J., Zou, J., Zeng, R., Liang, X.: Fractal simulation of soil breakdown under
lightning current. J. Electrostat. 61(3–4), 197–207 (2004)
9. Charalambakos, V.P., Agoris, D.P., Pyrgioti, E.C., Kupershtokh, A.L.: Simulation of
streamer propagation under positive impulse voltage. In: 13th International Symposium on
High Voltage Engineering, Delft, The Netherlands (2003)
10. Petrov, N.I., Petrova, G.N., D’Alessandro, F.: Quantification of the probability of lightning
strikes to structures using a fractal approach. IEEE TDEI 10(4), 641–654 (2003)
11. Li, J., Yang, Q., Sima, W., Sun, C., Yuan, T., Zahn, M.: A new estimation model of the
lightning shielding performance of transmission lines using a fractal approach. IEEE TDEI
18(5), 1712–1723 (2011)
12. He, J., Zhang, X., Zeng, R., Tu, Y.: Protection zone estimation for high lightning rods by a
fractal approach. In: 16th International Symposium on High Voltage Engineering, Cape
Town, South Africa (2009)
13. Perera, M.D.N., Sonnadara, D.U.J.: Fractal nature of simulated lightning channels. Sri
Lankan J. Phys. 13(2), 9–25 (2012)
14. Riousset, J.A., Pasko, V.P., Krehbiel, P.R., Thomas, R.J., Rison, W.: Three-dimensional fractal
modeling of intracloud lightning discharge in a New Mexico thunderstorm and comparison
with lightning mapping observation. J. Geophys. Res. 112 (2007). Article no. D15203
15. LeVeque, R.J.: Finite Difference Methods for Ordinary and Partial Differential Equations.
Society for Industrial and Applied Mathematics, Philadelphia (2007)
16. Li, J., Chen, Y.: Computational Partial Differential Equations Using MATLAB. CRC Press,
New York (2008)
17. Sanudo, J., Gomez, J.B., Castano, F., Pacheco, A.F.: Fractal dimension of lightning
discharge. Nonlinear Process. Geophys. 2, 101–106 (1995)
 Fractal-Based Approach for Modelling Electric Breakdown of Air Gaps 1305

18. Mikropoulos, P.N., Stassinopoulos, C.A., Sarigiannidou, B.C.: Positive streamer propaga-
tion and breakdown in air: the influence of humidity. IEEE TDEI 15(2), 416–425 (2008)
19. IEC 60060-1. Ed.3.0: High-voltage test techniques-part 1 (2010)
20. Mikropoulos, P.N., Tsovilis, T.E.: Experimental investigation of the Franklin rod protection
zone. In: 15th International Symposium on High Voltage Engineering, Ljubljana, Slovenia
(2007)
21. Rizk, F.A., Trinh, G.N.: High Voltage Engineering, p. 228. CRC Press, Boca Raton (2014)
 Self Calibrating High Voltage Divider

S. Passon1(&), N. Rühmann1, F. Schilling1, J. Meisner1,

and M. Kurrat2
1
PTB, Bundesallee 100, 38116 Brunswick, Germany
stephan.passon@ptb.de
2
TU-Braunschweig, Elenia, Schleinitzstraße 23, 38106 Brunswick, Germany

Abstract. High voltage calibrations on metrology level require traceability

down to primary national reference standards. These standards are based on the
Josephson effect, hence the maximum voltage is limited to less than 10 V. To
transfer these reference voltages to several hundreds of kilovolts different ref-
erence dividers are required. Several intermediate steps are needed to utilize the
optimal measuring range of the utilized dividers and voltmeters. The drawback
of using several dividers is the increase of measurement uncertainty within each
intermediate step. Additionally, the time required for traditional traceable cali-
bration of the high voltage reference dividers can easily exceed several weeks.
Since the PTB is one of the world’s leading metrology institutes it has started
the development of a new high voltage divider design. This new development
includes a high voltage arm, very similar to a standard high voltage divider.
However, it is possible to connect all high voltage resistors in parallel. In this
parallel arrangement the scale factor can be adjusted and compared to the pri-
mary reference standard. The adjustment is implemented by including a second
precision divider that is only used at low voltages below 50 V. With the aid of a
Wheatstone bridge arrangement these two voltage dividers are compared to each
other. The adjustment is carried out with the aid of two potentiometers. One of
the potentiometers is placed in the second divider to create a precisely known
scale factor. The bridge is zeroed by a potentiometer in the high voltage arm.
Thus, the known scale factor of the second divider is transmitted onto the high
voltage divider. After finishing the adjustment, the parallel arrangement is
changed back to series for the high voltage mode. Thereby, the scale factor is set
to a calculable precise value. The divider that was built as a first prototype was
compared to the existing reference divider and the results are shown within this
paper.

Keywords: High voltage divider
Reference divider
Hamon principle
Self

calibrating divider

1 Introduction

Measurements for high voltage direct currents (HVDC) are required for a wide variety
of applications. These may be high voltage sources for X-ray generation or electrostatic
coating processes, which require the least precise measurements. The accounting for
energy transmission over HVDC transmission lines needs a lot higher precision levels
with uncertainties below 0.1%. The by far most precise measurements are required for
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1306–1315, 2020.
https://doi.org/10.1007/978-3-030-31676-1_122
 Self Calibrating High Voltage Divider 1307

research of nuclear and particle physics. The required uncertainties are several orders of
magnitude lower, down to few parts per million [ppm] (1∙10−6). Some examples are
located at the GSI, CERN or the KATRIN experiment where the lowest uncertainty
with 3 ppm (k = 2) is needed [1]. These precision levels are equal to national
metrology level. PTB’s uncertainty for voltages below 100 kV is 2 ppm (k = 2). So, to
increase the precision in high voltage DC measurements new calibration capabilities for
traceable methods have to be established.
Standard high voltage dividers are equipped with a high voltage resistor and a low
voltage resistor. The scale factor is the ratio between the low voltage resistor and the
series resistance of the high- and low voltage arms. To reduce the self-heating the
measuring current should be restricted to about 0.1 mA at the rated voltage. However,
this leads to resistances in the giga-ohm range for voltages above 100 kV. To deter-
mine the scale factor a simple measurement of the resistances could be carried out.
However, the best uncertainties for resistance certified measurement capabilities
(CMC’s) above 1 GX are 5 ppm (k = 2). To overcome this problem the Hamon
principle has been implemented within the constructed 100 kV divider.
The high voltage measurement system presented in this paper will set the foun-
dation for further high voltage divider construction projects at the PTB beginning in
2020. This method will allow the direct scale factor calibration down to primary
reference standards without intermediate steps.

2 Basic Principle

This measuring instrument is equipped with two voltage dividers. The first one is used
only for the alignment of the main high voltage divider and is therefore named support
divider within this paper. Its nominal scale factor is 2:1. The main high voltage divider
has a high voltage arm with the addon of spring-loaded contacts on every high voltage
resistor. These allow the parallel arrangement of all high voltage resistors by the aid of
two gold-plated stainless-steel bars. When these are inserted in the divider for the
adjustment the high and low voltage arm also form a divider with the scale factor of
2:1. Both dividers are arranged in a Wheatstone bridge and the bridge voltage is
measured by a zero voltmeter. One of the high voltage resistors is equipped with an
adjustable resistor to equilibrate the bridge. The support divider is calibrated by
exchanging the high and low voltage resistors by S4 with each other and adjusting them
to achieve no change in the bridge voltage. This setup is shown in the following
illustration:
1308 S. Passon et al.

Fig. 1. Voltage divider setup

For the adjustment several steps have to be carried out in the following order:
(1) The parallel arrangement has to be prepared first. In Fig. 1 the switches S1 and S2
establish this connection, which is done by the two gold plated bars.
(2) The support divider is also connected by the aid of S3, which in turn is also done
by one of the two bars.
(3) Then a voltage U1 of 20 V is applied to both dividers.
(4) The switch S4 is switched back and forth and the potentiometer P2 is adjusted. If
the bridge voltage does not change after interchanging the high and low voltage
resistors of the support divider both resistors are equal in resistance and the scale
factor of the support divider is exactly 2:1.
(5) Afterwards the actual high voltage divider is compared to the support divider and
adjusted accordingly by P1 to reduce the bridge voltage to zero. If this is guar-
anteed the high voltage parallel resistance and low voltage resistance are also
equal.
(6) As the final step the voltage U1 and the gold-plated bars are removed
Now the divider is adjusted. In the case of this divider with 99 high voltage
resistors every high voltage resistor R1–R99 has 99 times the average resistance of the
low voltage arm RLV. In fact, the spread of resistance within the high voltage arm is not
negligible. However, the mathematics behind this configuration will be discussed later.
When the parallel arrangement is changed back to the normal series operation the total
scale factor M with n high voltage resistors will be:
 Self Calibrating High Voltage Divider 1309

M ¼ n2 þ 1 ð1Þ

The mathematical proof is based on the Hamon principle and the fact that the high
voltage resistors R1–R99 need be valued within a multiple factor m of the low voltage
resistor RLV

R1 ¼ R2 ¼ Ri ¼ R ¼ m
RLV ð2Þ

These resistances are actually not perfect but include a tolerance that might have an
influence on the scale factor after the adjustment procedure. Therefore, every high
voltage resistor RHV and the low voltage resistor RLV gets a tolerance d.
 
dRLV
RLV;T ¼ RLV þ dRLV ¼ RLV
1 þ ð3Þ
RLV
 
dRLV
RLV
1 þ ¼ R
ð1 þ aLV Þ ð4Þ
RLV
 
dRHV;i
RHV;i;T ¼ RHV þ dRHV;i ¼ RHV
1 þ ð5Þ
RHV
 
dRHV;i

RHV
1 þ ¼ m
R
1 þ aHV;i ð6Þ
RHV

The coefficient a describes the relative deviation of the resistance to the nominal
value R. With this approach the parallel connection, the adjustment and afterwards the
back to series arrangement was calculated. As a result, the scale factor is dependent on
the mean square of this relative deviation. This means that with low tolerances of the
used resistors (0.01%) this influence on the scale factor is negligible. Similar equipment
is available on the market, however only voltages for up to 1000 V like the Fluke752A.
This technique was implemented within a divider for voltages up to 100 kV with
99 high voltage resistors. Its design is shown in the following section.

3 Divider Design

To test the performance of such a divider a setup of 99 high voltage resistors with
10 MX each has been chosen. The low voltage arm needs to have 1/99 the resistance of
one high voltage resistor, which amounts to 101010 X. All the high voltage resistors
are thin film resistors with a voltage rating of 1000 V each. They are held in place with
a 3D printed support structure made of PLA (polylactic acid). Around them is a small
gap of 0.5 mm thickness through which air is forced in a closed-loop system. The air is
temperature controlled to 30 °C with a maximum of 0.5 °C deviation. This guarantees
a constant temperature distribution along the entire divider. Additionally, every resistor
is attached to a gold coated SMD connector. These connectors and the resistor’s leads
1310 S. Passon et al.

do not have a direct contact to the printed PLA but are held in place by CNC milled
PTFE (Polytetraflourethylene) ribbons (Fig. 2).

Fig. 2. 3D printed support structure with PTFE ribbons

This guarantees a high insulation resistance and therefore hinders the flow of
parasitic currents [2]. Two aluminum toroids on top and bottom of the divider establish
a homogenous electrical field and reduce corona discharges which would otherwise
also establish parasitic currents. These currents have a direct influence on the linearity
of the divider and may lead to a voltage dependent scale factor.
In the base of the divider the temperature control as well as the zero-volt meter are
included. The setup of the entire divider is illustrated in the following (Fig. 3):
The parallel connection bars are made of stainless steel and have been galvanized
with a 20 µm thick layer of gold. This generates a lower contact resistance as well as a
more reliable connection over time due to the lack of an oxide layer.
 Self Calibrating High Voltage Divider 1311

Fig. 3. Finished 100 kV Voltage divider

4 Measurements

To verify the function principle and design of the newly developed divider it was
compared to the primary reference divider MT100 for high DC voltages at the PTB [2].
These measurements include the stability, linearity, offset voltage, purposefully mis-
adjustment and at least the repeatability of the divider.

4.1 Stability
The stability test shows several different properties of voltage dividers and especially
their resistors. Thereby the self-heating due to the measuring current and thermal mass
define the amount and time constant of scale factor drift. Within this divider the
temperature induced scale factor drift was counteracted by the controlled heater. Any
temperature increase due to measuring current will be compensated by a reduced power
in the heater. Therefore, the thermal mass of the divider does not have an influence on
1312 S. Passon et al.

the scale factor’s drift time constant. Only the resistors and their internal structure
define this drift.
The stability measurements have been performed at voltages of 50 kV and 100 kV
for a time period until a stable value has been achieved. The change in the scale factor
(against the start value) is illustrated against the elapsed time within the following
(Fig. 4):

Fig. 4. Stability of the divider - (each horizontal line is equal to a relative change of 2
10−6)

In order to exclude the influence of the reference divider on these measurements the
tested divider was at first disconnected from the power supply and the reference
divider. The reference and the power supply stayed at the desired voltage for several
hours, so their thermal induced drift remained stable. Then with the high voltage still
on, the high and low voltage connection to this new divider was established. Imme-
diately afterwards the change of the scale factor can be measured without the short-term
stability of the power supply and reference dividers. However, it should be noted that
the low voltage connection to the divider should be established subsequent to the high
voltage connection. Otherwise transient over-voltages might influence or damage the
voltmeters.

4.2 Linearity
The linearity test was also performed against the MT100 since it is well known for its
non-voltage dependent behaviour [3]. For this test the voltage was increased in five
equidistant voltage steps up to the nominal voltage of 100 kV. Since the adjustment of
the scale factor was done at low voltage the trend should intersect the calculated scale
factor on the ordinate. A voltage dependent scale factor is expected since the utilized
resistors are thin film resistors which have a voltage coefficient far higher than tradi-
tional wire wound resistors [4]. The results of several linearity measurements are
illustrated within the next figure:
 Self Calibrating High Voltage Divider 1313

Fig. 5. Linearity of the divider - (each horizontal line is equal to a relative change of 2
10−6)

The value of the scale factor was always obtained when a stable value was reached.
The voltage coefficient that can be extracted from this Fig. 5 is 6.2
10−11/V which is
mainly caused by the high voltage resistors of the Type USFS370. However, this is
smaller than the datasheet specifications of up to 5
10−8/V [5]. The fact that the linear
interpolation does not intersect the calculated scale factor of 9802:1 indicates that there
is still a systematical deviation which needs to be identified.

4.3 Offset Measurements

The measurement of the offset of a voltage divider can exhibit weak features of crucial
components. These are mainly located in the low voltage arm and primarily influence
the measurement of small voltages and induce a polarity as well as voltage dependent
effect.
Offset voltages are especially caused by thermoelectric voltages which are in turn
evoked by the interconnection of various metals with different temperatures. This leads
to the visibility of the temperature control mode in the offset voltage measurement. To
overcome this problem a better temperature control was implemented as well as better
contact materials such as pure copper or copper alloys instead of steel and stainless
steel. Especially oxidation of copper contacts needs to be prevented since copper oxide
has a thermoelectric voltage more than thousand times higher than pure copper [6].
Also, LEMO connectors were used instead of N-Type connectors to reduce the thermal
mass with the drawback of higher interference levels. All these efforts reduced the
offset voltage to less than 2 µV.
1314 S. Passon et al.

4.4 Adjustment Verification

To see the result of the adjustment of the divider a intentional mis-adjustment was
performed. It is required to show calculated dependencies of the scale factor on the
bridge voltage. Therefore, the divider was tuned to generate bridge voltages ranging
from −45 µV to + 200 µV. The difference of the scale factor according to the mis-
adjustment is calculated and simultaneously measured against the MT100. The results
are shown in the following (Fig. 6):

Fig. 6. Mis-adjustment of the divider

It is evident, that the measured values coincide well within the calculated ones. This
in turn proves that the calculations according to the model are correct.

4.5 Repeatability
In order to verify the principle of the divider further on it was misaligned purposely and
adjusted back. This procedure was carried out ten times to be sure to achieve a constant
performance of the divider. The mean of the scale factor was determined to 9801.95
with a relative standard deviation of the mean equalling 1.51
10−8.

5 Uncertainty Considerations

The uncertainty contributions have been evaluated according to the performance

measurements and calculations based on the dividers principle on the Hamon method
and in comparison, to the existing reference voltage divider. All the uncertainty con-
tributions except for the voltage dependency have been included. The voltage
 Self Calibrating High Voltage Divider 1315

behaviour was excluded since it is a contribution by the resistor technology and not the
principle itself. In the end the expanded relative measurement uncertainty (k = 2) of
this divider is 3.4
10−6.

6 Conclusion

The principle of the proposed divider has been presented within this paper. Future
research must be placed on the linearity on high voltage dividers and the resistor
technology to further improve this technique and reduce the uncertainty of the system.
Also, for higher voltages bigger dividers are necessary and therefore other approaches
for the parallel resistor arrangement are required.
Identification of certain equipment does not imply recommendation by the authors,
nor does it imply that the equipment is necessarily the best available for the purpose.

References
1. Bauer, S., Berendes, R., Hochschulz, F., Ortjohann, H., Rosendahl, S., Thümmler, T.,
Schmidt, M., Weinheimer, C.: Next generation KATRIN high precision voltage divider for
voltages up to 65 kV. J. Instrum. 8(10), 10026 (2013). https://doi.org/10.1088/1748-0221/8/
10/p10026
2. Keithley: Precision DC Current, Voltage, and Resistance Measurements, 7th edn
3. Marx, R.: 100 kV DC voltage standard divider of the shielded type, pp. 417–418 (2000).
https://doi.org/10.1109/cpem.2000.851053
4. Passon, S., Gitin, I., Meisner, J.: Investigating the properties of precision resistors for the
application in high voltage DC dividers, pp. 1–6 (2016). https://doi.org/10.1109/amps.2016.
7602872
5. Caddock.: Type USFS high voltage, ultra-stable, low TC selected resistor sets. http://www.
caddock.com/online_catalog/Mrktg_Lit/TypeUSFS.pdf
6. Kidd, M.L.: Watch out for those thermoelectric voltages!, June 2012
 High Voltage Topologies for Very Fast
Transient Measurements

A.-P. Elg1(&), S. Passon1, J. Meisner1, and J. Hällström2

1
Physikalisch-Technische Bundesanstalt,
Bundesallee 100, 38116 Brunswick, Germany
alf-peter.elg@ptb.de
2
Centre for Metrology MIKES, VTT Technical Research Centre
of Finland Ltd., Espoo, Finland

Abstract. High voltage equipment is subjected to several types of electrical

stress during operation, why testing by transient wave shapes is routinely per-
formed. Measurement systems are being developed by National Metrology
Institutes (NMI) for traceable calibration of sensors used in Gas Insulated
Switchgear (GIS) where front times below 10 ns are common. We refer to front
times below this as Very Fast Transients (VFT).
This paper presents advantages and disadvantages of various divider designs
to be used for VFT measurement systems. Many of these are a heritage from
lightning impulse measurement systems. Several divider designs are studied to
investigate the contribution on uncertainties in both amplitude and time
parameters in a measurement system. Four resistive designs based on ceramic
bulk resistors, bifilar wire winding, a resistive liquid and thick film SMD
resistors have been characterized and compared with a small damped capacitive
divider.
Step response measurements have been used to characterize several dividers
in different geometrical configurations, looking at the transient response out of a
wave propagation point of view with analysis of loop sizes, field distributions
and proximity effects. The findings are that and minimizing skin effects and
inductance in the dividers are essential for a fast response and settling.
Convolution of the step responses with ideal reference curves showed that
more development is needed to eliminate oscillations in the step response.
Further improvement of the geometries and impedance matching is needed to
achieve acceptable errors for VFT with front times T1 below 10 ns.

Keywords: Fast transients
High voltage impulse
High voltage divider

1 Introduction

High voltage equipment is subjected to several types of electrical stress during oper-
ation. Factory tests are defined to ensure that the equipment will perform satisfactorily
in service. Testing with standard lighting impulses [10] and conforming to IEC 60071,
using impulses exceeding the nominal voltage in the grid is well proven and used
method with established traceable calibration services.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1316–1327, 2020.
https://doi.org/10.1007/978-3-030-31676-1_123
 High Voltage Topologies for Very Fast Transient Measurements 1317

Measurement of transmitted overvoltage test for instrument transformers with

0.5 µs front time at 1 kV according to IEC 61869, currently lacking traceable cali-
bration services for HV has recently been developed [16]. Puncture testing of insulator
discs according to IEC 61211 with 0.2 µs front time has recently been developed
providing traceability of peak voltage measurement [11]. The need to correctly measure
faster impulses, i.e. transients, during testing is pushing metrology to even faster rise
times.
For measurement of transients in Gas Insulated Switchgear (GIS), measurement
systems with rise times less than 10 ns are needed. Development of traceability for
transients with rise times even below 10 ns and voltages up to 100 kV is ongoing at
national metrology institutes [1], using transient recorders with latest technology with
higher dynamic range and fast enough settling time [9]. Another major obstacle for
traceable measurement at high voltage is the design of a divider with a fast response,
which involves attention not only to the physical size of the voltage divider but also the
topology of the measurement circuit.

2 Divider Types

In this work we focus on different divider designs, and their uncertainty contribution in
amplitude and time parameter measurements. New divider designs have similarities to
dividers used for LI measurements, but in most cases quite different approaches are
used to improve the response time and minimize the overshoot at the front.
We have studied seven different dividers. Two dividers typically used in LI mea-
surements are wire-wound resistive dividers, as shown in Fig. 1, and mixed resistive
capacitive divider shown in Fig. 2. The newly developed dividers are: a divider for
puncture testing using ceramic resistive components shown in Fig. 3, a wire-wound
resistive divider shown in Fig. 5, a salt solution resistive divider depicted in Fig. 6, a
divider with Surface Mounted Resistive devices (SMD) in Fig. 4, and a damped
capacitive divider in Fig. 7.

2.1 Lightning Impulse Dividers

A commercial resistive divider for LI measurements [12] is shown in Fig. 1. The shape
of the corona ring has proven to be important for linearizing the field distribution along
the stack, crucial for achieving a good step response. This type of divider is designed
for LI measurements up to c.a. 500–700 kV [13].
1318 A.-P. Elg et al.

Fig. 1. Wire wound resistive divider

A damped-capacitive divider is shown in Fig. 2 [2] and is an optimum damped RC

divider of Zaengl design [14]. The damped-capacitive divider is mainly designed and
used for ultra-high voltage measurements. Results show an experimental response time
TN of 5.8 ns, a partial response time Ta of 13 ns and a settling time Ts of 190 ns [2].
Due to the size of the measuring circuit, parasitic inductances in combination with the
capacitance give oscillations, which is evident from the 190 ns settling time.
The designs of the LI dividers are well proven for measurements of wave fronts
down to 0.5 µs, for testing of HV equipment. However, the dimensions of these are
probably too large to achieve fast enough step-response for shorter front times needed
for VTF measurements (1–10 ns front time) where the dimensions of the high voltage
circuit become important.

Fig. 2. Optimum damped RC Mixed divider.

 High Voltage Topologies for Very Fast Transient Measurements 1319

2.2 Fast Transient Dividers

For fast transient measurement the signal propagation path needs to be shortened by
reducing the size of the HV circuit, which also lead to lower inductance. This is
achieved by building smaller dividers paying attention to the reduction of the flashover
voltage or confine the HV circuit in a protective gas like in GIS. We limit this study to
designs for applications in air, but an example of GIS application is described in [4].
Ceramic Disc Resistive Divider. This divider was originally designed for puncture
testing with 200 ns rise time in a previous study [3].

Fig. 3. Divider with ceramic bulk resistors in a 200 kV configuration. The step generator can be
seen attached to the right on the coaxial cable.

Here, compared with the original design, the number of ceramic bulk disk resistors
stacked in the high voltage arm has been reduced. The maximum voltage of the
modified divider in Fig. 3 is 200 kV instead of 600 kV, and the HV arm resistance is
1600 Ω. The disks are 25.4 mm high, have 50 mm outer diameter and a 20 mm
diameter through hole. The low voltage arm has a 4.7 Ω resistance, giving a scale factor
of 680.
In this divider we focus on minimizing mutual inductance by using a coaxial
symmetry in both the HV and LV arm. The design is based on the work of M Aro [5].
The ceramic discs have no measurable inductance, the skin depth is large e.g. 180 mm
@ 100 MHz giving a high effective relative conducting cross section of 96%. How-
ever, the scale factor of the divider is voltage dependent. A modified version of this
divider design has also been used in an ongoing project for calibration of VFT sensors
in GIS [4].
Thick Film SMD Divider. The coaxial design of the ceramic divider in Fig. 3
eliminates much of the inductance. One way to further minimize the skin effect is to use
film resistors with preferably thin film. A divider was built with SMD in a coaxial
arrangement as shown in Fig. 4.
1320 A.-P. Elg et al.

Fig. 4. SMD thick film resistors mounted in a coaxial arrangement designed for measurements
up to 200 kV with pulse widths up to 500 ns.

In selection of SMD resistor values, there is a trade-off in optimizing the response

time, which is hampered either by stray capacitance for large values, or by stray
inductance for lower values. Also, the maximum sustainable pulse energy and/or field
strength limits the resistor selection.
An optimum is found in the 100 Ω to 300 Ω range. The HV arm was built using ten
resistors, each 100 Ω, soldered parallel in a circle, stacking 50 such units to obtain a
HV resistance of 500 Ω. The low voltage arm comprises of ten 10 Ω resistors in
parallel in a coaxial arrangement, thus obtaining a scale factor of 501.
Wire Wound Divider. An alternative to discrete resistors is to use a bifilar winding of
resistive wire to reduce the mutual inductance. This has been used in the design of the
compact divider in Fig. 5. The divider is built using a bifilar winding of 140 µm thick
Manganin wire forming a high voltage resistance of 5 kΩ. The wire thickness brings
skin effect to a low value, e.g. with an effective relative conducting cross section of
56% at 100 MHz.

Fig. 5. Wire wound resistive divider designed for measurements up to 200 kV and time-to-half
values up to c.a. 60 ls
 High Voltage Topologies for Very Fast Transient Measurements 1321

For large dividers as the ones in Figs. 1 and 2, the influence of capacitive coupling
to surroundings needs to be taken care of. This can be done either by grading the
winding or capacitance in the HV arm or by using a toroid as shown in e.g. Figure 1.
The toroid in Fig. 1 makes the field along the HV stack more linear and reduces
capacitive coupling to the surroundings. The effect of the toroid ring for this divider is
discussed in the step response test section.
Salt Solution Divider. The high dielectric constant of water, together with added salt
to reduce the resistivity, make the salt solution (CuSO4) divider a candidate for
measurement of fast transients. However, the conclusion in earlier work was that the
divider scale factor was dependent on temperature, and it was not stable [6]. The design
shown in Fig. 6 use a copper sulphate solution to adjust the HV resistance value.

Fig. 6. Resistive salt solution divider designed for measurements up to 200 kV with time-to-half
values up to c.a. 60 µs.

The fluid is contained in a 2.5 mm thick layer between two PMMA tubes to min-
imize the inductance and skin effect. With the saturated CuSO4 solution a high voltage
resistance of 1.7 kΩ is achieved. The skin depth is e.g. 28 mm at 100 MHz and using a
thin layer design concentric geometry the effective relative conducting cross section is
97%. In Fig. 6 the divider is shown with toroid and a guard ring.
Damped Capacitive Divider. A miniaturized version of the damped capacitive
divider in Fig. 2 is the last candidate for VFT measurements. The divider is shown in
Fig. 7.
1322 A.-P. Elg et al.

Fig. 7. Damped capacitive divider for 100 kV and time-to-half values up to c.a. 60 µs.

3 Step Response

3.1 Step Generator and Digitizer

Traditionally a mercury wetted relay has been used to study the response on dividers
for LI. Relay design producing steps down to 400 ps fall times has been reported [7].
Figure 8 shows the step response of the resistive LI divider of Fig. 1 measured with a
75 Ω cable and termination. The rise time is 8 ns which is a bit too slow for VFT
measurements.

Fig. 8. Step response of SMR500 LI divider. The rise time is 8 ns.

In VFT measurements we use 50 Ω termination to increase the bandwidth and

avoid reflections at the recorder end of the measuring cable. In the low ns time domain,
 High Voltage Topologies for Very Fast Transient Measurements 1323

the reflections resolved by the propagation time in the cable will clearly be observed
and have an impact on measurement uncertainties.
A special avalanche technique is used to shorten the rise time of the step [8] below
100 ps. The developed step generator utilizes a strip-line technique to match with 50 Ω
cable impedance. A reference step response from the digitizer used here is shown in
Fig. 9.

Fig. 9. Step response of the PXIe-5164 digitizer with a rise time of 1.45 ns and 4.6 ns settling.

However, since most of these dividers are mixed RC dividers and cannot be loaded
with 50 Ω, the typical input impedance of a recorder used with those is 1 MΩ. This
often leads to a reduced bandwidth.

3.2 Step Response Setup

The step generator [8] shown in Fig. 10 is attached as close as possible to the HV
circuit of the divider as shown in Fig. 3.

Fig. 10. Avalanche step generator [4]. A d.c. voltage is applied from the right, the output is on
the left.

3.3 Divider Step Responses

In Figs. 11, 12, 13, 14 and 15 the step responses are shown for the dividers shown in
Figs. 3, 4, 5, 6 and 7. A fast response is important for a VFT divider; however,
1324 A.-P. Elg et al.

overshoot and oscillations lead to a slow settling time, and higher measurement
uncertainty [10].
The ceramic disk divider in Fig. 3 has a 4.7 ns rise time as shown in Fig. 11. The
200 kV version in Fig. 3 is more compact and has better step response than larger
versions [3], indicating the importance of a compact structure for VFT applications.
The bifilar wire wound divider in Fig. 5 performs best with the corona ring lin-
earizing the field but causing more capacitive coupling from the HV arm to earth. This
causes the faster oscillations observed on the front of the step response shown in Fig. 12.

Fig. 11. Step response of the ceramic resistive divider of Fig. 3. The rise time is 4.7 ns at 50 Ω
in a 200 kV configuration without corona sphere and coaxial guard tube.

Fig. 12. Step response of the wire wound divider of Fig. 5. The rise time is 9.5 ns for the green
case and 18 ns for the red and blue case.

The response time of the wire wound resistive divider in Fig. 5 is a little faster than
for the larger commercial resistive divider in Fig. 1, but oscillations and overshoot are
large.
The salt solution divider has a faster response without corona and/or guard ring as
shown in Fig. 13. Here, we reach 6.4 ns rise time with much less overshoot than for the
wire wound and the ceramic disk divider.
 High Voltage Topologies for Very Fast Transient Measurements 1325

Fig. 13. Step response of the salt solution divider of Fig. 6. The rise time is 8.4 ns with corona
ring and guard and 6.4 ns without corona ring and/or guard.

When we look at the step response of the SMD resistive divider in Fig. 14, the rise
time of 6.2 ns is the same as for the CuSO4 liquid divider without corona ring. The
overshoot is the lowest of all dividers, which makes the geometry and components of
this divider one of the better for VFT measurements. The settling time is the best
among the resistive dividers and is discussed more in Sect. 4.

Fig. 14. Step response of the SMD divider of Fig. 4. The rise time is 6.2 ns.

The step response and settling of the damped-capacitive divider of Fig. 7 is shown
in Fig. 15 in the next section.

4 Settling Time and Error Estimate

The settling time of the dividers gives a first hint of the shortest front time that could be
used without contributing too much to Up and T1 uncertainty [10].
The resistive dividers we have presented respond quickly but have oscillations after
a step, which makes it difficult to use them for short front times below 10 ns. The
1326 A.-P. Elg et al.

Fig. 15. Step response and the step integral, with a 6.3 ns rise time, and 7.5 ns settling time for
the damped-capacitive divider.

measured settling times are 60 ns for the wire wound divider, 20 ns for the salt solution
divider, 17 ns for the SMD divider and 22 ns for the ceramic PIKA divider.
The damped-capacitive divider has a rise time almost as fast as the resistive dividers
(6.3 ns), but it is settling much faster (7.5 ns shown in Fig. 15). The errors introduced
by the dividers can be determined by convolving the step responses with ideal test data
curves [15]. Results are shown in Table 1 for the two fastest dividers.

Table 1. Errors of damped-capacitive (Zaengl) and SMD dividers.

Ref. curve Up error [%] T1 error [%] Up error [%] T1 error [%]
Zaengl Zaengl SMD SMD
2/50 ns +0.5 +280 +7.4 +119
4/50 ns +2.4 +103 +4.4 +17.4
5/50 ns +2.5 +69.2 +1.9 −0.0
7/50 ns +1.3 +30.0 +0.9 −9.7
10/50 ns −0.14 +1.6 −0.1 +0.1

5 Conclusions

The advantages and disadvantages of various divider designs for use in VFT mea-
surements have been compared. The divider design has impact on uncertainties in both
impulse amplitude and time parameters measurement. A low inductance in a divider is
essential for a fast response. However, skin effect also must be considered. The wire
wound divider with a slower step response and large overshoot must be tuned further to
be qualified for VFT measurements.
The fastest rise time is obtained using the ceramic disk divider, followed by the
liquid divider and the SMD. The liquid divider does not have a stable scale factor and it
must be calibrated before and after use. The ceramic disk divider has an excellent
energy withstand, but its drawback is a voltage dependence, although it is repeatable
and can be corrected for. The SMD divider design has the lowest overshoot among the
 High Voltage Topologies for Very Fast Transient Measurements 1327

resistive dividers, it could have a small voltage dependence, and it has potential to
become fast enough for VFT measurements. The small damped capacitive divider has a
fast response but most important of all, also a fast settling time.
A first test using convolution of the step responses with ideal reference curves
showed that more development is needed to eliminate oscillations in the step response.
Further improvement of the test setup geometries and impedance matching is needed to
achieve low uncertainty for measurement of VFT with front time T1 below 10 ns.

Acknowledgments. The work presented here has received support from the EMPIR program
co-financed by the Participating States and from the European Union’s Horizon 2020 research
and innovation program.

References
1. European EMPIR program funded project 15NRM02 UHV. http://www.vtt.fi/sites/UHV
2. Meisner, J., Passon, S., Schierding, C., Hilbert, M., Kurrat, M.: PTB’s new standard impulse
voltage divider for traceable calibrations up to 1 MV. In: 20th ISH, Buenos Aires, Argentina,
Paper 415, August 2017
3. Hällström, J., Kazmi, S., Havunen, J., Pykälä, M.L.: Design and performance of a fast
divider for puncture testing. In: 20th ISH, Buenos Aires, Argentina, Paper 470 (2017)
4. Kamlichi, A., Garnacho, F., Hällström, J., Elg, A.-P.: Calibration setup for traceable
measurement of very fast transients (to be published in ISH 2019)
5. Aro, M.: Feasibility of impulse voltage puncture test on insulators in air. Ph.D. thesis, Acta
Polytechnica Scandinavica, El91 (1998)
6. Aro, M., Punkka, K., Huhdanmäki, J.: Fast divider for steep front impulse voltage tests. In:
5th ISH, Braunschweig, Germany, Paper 73.01 (1987)
7. Bergman, A., Elg, A.-P., Hallström, J.: Evaluation of step response of transient recorders for
lightning impulse. In: 20th ISH, Buenos Aires, Argentina, ISH 2017, August 2017
8. Passon, S., Schilling, F., Meisner, J., Schlüterbusch, T.C., Elg, A.-P. Kurrat, M.: Avalanched
based step generator (to be published at ISH 2019)
9. Elg, A.-P.: Qualifying a transient recorder for traceable measurements of very fast transients.
In: Conference on Precision Electromagnetic Measurements CPEM, Paris, France, July 2018
10. IEC 60060-2. Ed.3.0: High-voltage test techniques-part 2 (2010)
11. Hällström, J., Havunen, J., Li, Y., Yan, W., da Silva, M., Filho, O., Vinícius, M., Lopes, B.,
Laiho, M.: Comparison of Measuring Systems for Puncture Test According to IEC 61211 (to
be published at ISH 2019)
12. HighVolt SMR500 Divider
13. Spiegelberger, J.: Probleme der Dimensionierung und Eichung von ohmchen Stossspan-
nungteilern. Dissertation, TU Dresden (1966)
14. Zaengl, W.: Das messen hoher, rasch verändlicher Stossspannungen. Dissertation, TH
München (1964)
15. Glinka, M., Schon, K.: Numerical convolution technique for qualifying HV impulse
dividers. In: Proceedings of 10th ISH, Montreal, Canada, vol. 4, pp. 71–74, August 1997
16. Elg, A.-P., Garnacho, F., Garcia, T., Rovira, J., Hällström, J., Nieminen, T.: Traceable
measurement of transmitted overvoltages in instrument transformers (to be published in ISH
2019)
 Methods for Field Measurement
of the Frequency-Dependent Soil Electrical
Properties: Evaluation of Electrode
Arrangements Through FEM Computations

Z. G. Datsios(&) , P. N. Mikropoulos , and E. T. Staikos

High Voltage Laboratory, School of Electrical and Computer Engineering,

Faculty of Engineering, Aristotle University of Thessaloniki,
541 24 Thessaloniki, Greece
zdatsios@auth.gr

Abstract. The low radio-frequency electrical properties of soil (up to 3–

5 MHz) are important for lightning transient and grounding system studies.
Measurements of soil electrical properties are most commonly performed on
undisturbed or reconstituted soil samples even though it is difficult to use these
results for the representation of undisturbed ground in field conditions. This
work presents an evaluation of several electrode arrangements for field mea-
surement (in situ) of the frequency-dependent soil electrical properties at low
radio-frequencies through FEM computations. The developed model is validated
against results obtained theoretically and using grounding analysis software.
A parametric analysis is performed by varying electrode dimensions, separation
distances between electrodes and the excitation frequency (up to 100 kHz);
several uniform as well as two-layer soil models were investigated. Results are
compared and discussed based on deviations from the assumed hemispherical
symmetry of the potential distribution in the ground around the injecting elec-
trode taking into account the effects of excitation frequency and soil model
parameters. The most appropriate arrangement for field measurements com-
prises a hemispherical and a ring electrode for current injection as well as two
potential probes; for practical applications, the ring current-return electrode can
be replaced by four ground rods in order to minimize the time and effort required
for the setup of the electrodes. A preliminary investigation is conducted to assess
the applicability of the measured apparent electrical properties of non-uniform
soils to the computation of the impedance of ground electrodes. It is shown that
using the apparent soil electrical properties as determined for an arbitrary dis-
tance between potential probes, may result in significant errors in the computed
impedance of ground electrodes.

Keywords: Electrical properties
FEM
Field measurement methods

Frequency dependence
Soil

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1328–1339, 2020.
https://doi.org/10.1007/978-3-030-31676-1_124
 Methods for Field Measurement of the Frequency-Dependent Soil 1329

1 Introduction

The electrical properties of soil at low radio-frequencies are important for many
engineering applications, including lightning transient studies and grounding system
design and analysis. For such applications, soil can be characterized using the relative
permittivity, er, and the effective (equivalent) conductivity, reff, that is, the sum of the
DC and AC conductivity. er and reff are frequency-dependent owing to polarization
phenomena in soil. Actually, at low radio-frequencies er decreases with increasing
frequency due to the distributed relaxation of the interfacial polarization mechanism
and reff increases as a result of the associated polarization losses.
For frequencies up to 3–5 MHz related to lightning phenomena, measurement of
the frequency-dependent soil electrical properties is typically performed on undisturbed
or reconstituted soil samples using two-, three- or four-terminal cells [1–4]. In this way,
the actual electrical properties are obtained for the soil under test. However, due to the
non-uniformity of soil, it is difficult to use these results for characterizing the elec-
tromagnetic behavior of undisturbed ground in field conditions, thus also for deter-
mining an accurate soil model. This task requires field measurements to be conducted,
analogous to those used for determining the apparent DC (or power frequency AC) soil
resistivity, q. For DC soil resistivity measurement, several field methods and electrode
arrangements have been proposed in literature, as also adopted by IEEE standards [5,
6]. However, for field measurement of the frequency-dependent soil electrical prop-
erties only Visacro and Alipio [7] have proposed a method.
This work presents an evaluation of several electrode arrangements for field
measurement of the frequency-dependent soil electrical properties. This is accom-
plished through FEM computations using COMSOL Multiphysics software [8, 9]; the
latter is commonly used in literature for the analysis of grounding systems [10–13].
A parametric analysis is performed considering different electrode dimensions and
separation distances between electrodes, as well as several uniform and two-layer soil
models. Results are compared and discussed considering the effects of excitation fre-
quency and soil model parameters. It is shown that the most appropriate electrode
arrangement for field measurements comprises a hemispherical and a ring electrode; for
practical applications, the latter can be replaced by four equipotential ground rods.

2 FEM Modeling Using COMSOL

2.1 General Settings

The model employed in FEM computations was developed using the Electric Circuit
and Electric Currents interfaces of the AC/DC module [9] of the COMSOL Multi-
physics software [8]. The former interface was utilized to excite the evaluated electrode
arrangements in the frequency domain through an External I vs. U node. The electrode
arrangements were designed with the aid of the Electric Currents interface. The ground
surrounding the electrodes consists of two concentric hemispheres. The subdomain
between the hemispheres is characterized as an Infinite Element Domain; hence, the
1330 Z. G. Datsios et al.

ground is considered as semi-infinite and the total dimensions of the model do not
affect computation results.
The potential of the outer hemisphere surface was set to 0 V to simulate the remote
earth. The Electric Insulation boundary condition was used at the ground-air interface:
n
J ¼ 0, where n is the normal unit vector and J is the current density vector. The
investigated electrodes were excited using a Terminal node by coupling the Electric
Circuit and Electric Currents interfaces. Models were created in 3D and/or 2D
axisymmetric geometries (where possible) so as to minimize simulation time and
memory usage. The extremely fine physics-controlled mesh was used in all cases.
Several uniform and two-layer soil models were employed in computations; the
relative permittivity, er, and resistivity, q, values are given in Table 1. The electrical
properties of soil were considered as frequency-independent since computations are
performed in the frequency domain and each (er, q) set is employed for all investigated
frequencies. The selected er values of the uniform soil models (Table 1) decrease with
increasing q values as expected for actual soils [1–4]. Extremely high er values were
deliberately chosen so as to consider cases with relatively high electric displacement
currents. The two-layer soil models of Table 1 (soil cases 5 and 6) were implemented
in COMSOL by modeling the upper layer as a disk with a diameter almost equal to the
inner hemispherical subdomain and depth equal to that of the soil model (Table 1).

Table 1. Soil cases considered in this work.

Uniform soil models
Soil case q (Xm) er
1 10 10000
2 100 1000
3 1000 100
4 10000 10
Two-layer soil models
qupper (Xm) qlower (Xm) er Depth, h (m)
5 10 100 0.3, 0.5
10000
6 100 10 1.0, 2.0

Initially, validation was performed for DC and AC excitation by simulating ground

electrodes and a guarded two-electrode laboratory arrangement, respectively. The
frequency values used in this work were lower than 100 kHz to fulfil the criteria for the
quasi-static approximation at the highest frequency, taking into account the largest
dimensions of the investigated electrode arrangements. The AC/DC module of
COMSOL, which employs quasi-static analysis, is appropriate for problems with
electrical size less than 0.1 [9].
 Methods for Field Measurement of the Frequency-Dependent Soil 1331

2.2 Validation with DC Excitation

In order to validate the model under DC excitation, several hemispherical electrodes
and ground rods were simulated in uniform soils (q = 10 − 104 Xm, er = 104) con-
sidering both 3D and 2D axisymmetric geometries and using a Stationary Study. Each
electrode was excited by DC current and the obtained ground resistance was compared
with that calculated using well-known expressions [3] or computed with the XGSLab
software [14]. Also, results of 3D and 2D axisymmetric geometries were compared. For
all investigated cases, the obtained errors are less than 0.05%; typical results are shown
in Fig. 1.

(a) y=x

D=1 m

R3D ( )
(b) y=x

Fig. 1. Ground resistance of a hemispherical electrode: (a) Calculation and 3D simulation and
(b) 3D and 2D axisymmetric simulation; parameter: soil resistivity.

2.3 Validation with AC Excitation

Validation in the frequency domain was performed by simulating a guarded two-
electrode laboratory arrangement used for soil electrical properties measurements in the
frequency range from 42 Hz up to 1 MHz (Fig. 2 of [4]). The arrangement was sim-
ulated in 3D geometry using a Frequency Domain Study with uniform soil samples
between the electrodes. An AC voltage with variable frequency was applied on the
unguarded electrode with the other electrodes grounded. The computed admittance
between the guarded and unguarded electrodes was compared with the value calculated
for a lossy capacitor [4]; the obtained errors are less than 0.5%.
1332 Z. G. Datsios et al.

3 Visacro and Alipio Method

For field measurement of soil electrical properties in the frequency range up to 3–

5 MHz the only available method has been proposed by Visacro and Alipio [7]
(Fig. 2). A low-voltage impulse generator is used to inject a current between a
hemispherical electrode and four auxiliary rods. The developed potential difference
between the hemisphere and a point on the ground surface is measured and the
admittance in the frequency domain, Y*, is obtained. The electrical properties of soil are
calculated using (1), which is derived assuming hemispherically symmetric current and
potential distributions in the ground:
 
2p qð1xÞ þ jxe0 er ðxÞ
Y  ðxÞ ¼ ð1Þ
1
r1  r12

In (1) Y* is the admittance between r1 (m) and r2 (m) (Fig. 2), x (rad/s) is the angular
frequency, er is the relative permittivity of soil, e0 is the permittivity of free space
(8.854
10−12 F/m) and q (Xm) is the electrical resistivity of soil.
As evident from Fig. 2, the current returns to the impulse generator through the
auxiliary rods, which are not symmetrically positioned with respect to the hemisphere.
Even though the rods are placed at a considerable distance from the hemisphere, the
assumption of a hemispherically symmetric current distribution has to be investigated;
from a relevant study on impulse ground impedance measurement it can be deduced
that such a distribution may not always apply [15]. In addition, and most importantly,
this method uses a three-terminal arrangement; therefore, measurement errors due to
contact resistance and electrode polarization effects are inherent, as the hemisphere is
used for both current injection and voltage measurements.

Impulse
Generator

Α V
r1 = 0.23 m r1
r2 = 0.92 m r2 Auxiliary Rods
0.7 m

30 m

Fig. 2. Electrode arrangement of the Visacro and Alipio [7] method (not according to scale).
 Methods for Field Measurement of the Frequency-Dependent Soil 1333

4 Design and Evaluation of Electrode Arrangements

Several electrode arrangements were designed in this work in order to achieve sym-
metric current and potential distributions in the ground and to eliminate measurement
errors due to contact resistance and electrode polarization effects. A parametric analysis
was conducted to evaluate arrangements using a hemispherical electrode for current
injection in the ground and either a ring electrode or four equipotential ground rods as
return electrodes (Fig. 3). The voltage arising between two points on the ground sur-
face due to the injected current is measured using two test probes. It is noteworthy that
arrangements employing a ring electrode for the current return path have been used in
ground impedance measurements in [15, 16].

Dh Dh
r V r V

r1 r1
r2 r2
A A

(a) Dr (b)

Fig. 3. Investigated electrode arrangements for field measurements of the frequency-dependent

soil electrical properties (top view). Current-return electrode: (a) Ring and (b) Four ground rods.

4.1 Uniform Soil Models

The evaluated electrode arrangements of Fig. 3 were simulated in uniform soils using
3D or 2D axisymmetric geometry. Different ring burial depths, hr, hemisphere, Dh,
ring, Dr, and conductor, Dc, diameters were investigated. It was found that the current
and potential distribution symmetry in the ground is predominately affected by Dr;
thus, the latter was considered as an influencing parameter in what follows. In simu-
lations, the ring was placed at the ground surface taking into account the applicability
of the arrangement; the hemispherical electrode and conductor diameters were selected
0.5 m and 6 mm, respectively.
1334 Z. G. Datsios et al.

Figure 4 shows the variation of the relative difference between the maximum and
minimum potential on a hemispherical surface in the ground with its radius, r, for
several values of Dr (Fig. 3a); this variation is a measure of the asymmetry in current
and potential distributions in the ground. The potential can be considered as sym-
metrically distributed only close to the hemispherical electrode; potential difference
values lower than 5% were obtained for r < *0.18Dr. Based on the results of Fig. 4, a
ring diameter of 60 m is adopted resulting in enhanced symmetry in the potential
distribution. Figure 4 also includes the results referring to the Visacro and Alipio
arrangement (Fig. 2). It is evident that the potential distribution is symmetric for
considerably shorter distances from the center of hemispherical electrode as compared
with the arrangement of Fig. 3a for the corresponding case of Dr = 60 m. It is
important that the results of Fig. 4 are not affected by the electrical properties of the
uniform soil model (Table 1) and the excitation frequency.
In order to evaluate the applicability of (1) to the adopted electrode arrangement
(Dr = 60 m), the computed admittances between several points on the ground surface
(radii r1 and r2 from the center of the hemispherical electrode) were compared with
those calculated by using (1). Figure 5 shows the variation of the absolute relative error
of the computed admittance amplitude, |DY/Y|, with r1 and r2; these results are not
influenced by the excitation frequency and the uniform soil model (Table 1). As evi-
dent from Fig. 5, |DY/Y| is constantly lower than 1%; for the phase angle of the
admittance, errors are lower than 0.1%. Hence, (1) yields accurate results even when
the potential difference is measured at a considerable distance from the hemispherical
electrode.

Umin

Dr = 8 m 16 m
Umax
30 m 60 m

[7]

Fig. 4. Absolute relative difference, |DU/Umax|, between the minimum and maximum potential
on a hemispherical surface in the ground as a function of its radius, r, for the electrode
arrangements of Figs. 2 and 3a with Dr as a parameter.
 Methods for Field Measurement of the Frequency-Dependent Soil 1335

0.8

0.7

0.6 r1 = 5 m

0.5
|∆Y/Y| (%)

2m

0.4

0.3

0.2
m
0.1 0.25

0
0 3 6 9 12 15
r2 (m)

Fig. 5. Absolute relative error of the computed admittance amplitude, |DY/Y|, as a function of
the radii r1 and r2; arrangement of Fig. 3a (Dr = 60 m).

Figure 6 shows the corresponding relative error values, |DY/Y|, for the Visacro and
Alipio arrangement. Despite the fact that in their method r1 is equal to the hemi-
spherical electrode radius (Fig. 2), in this work the positions of both r1 and r2 were
varied along the axis connecting the hemispherical electrode with the auxiliary rods.
Apparently, the error values of the computed admittance amplitude are almost 1 order
of magnitude higher than those of Fig. 5; for the admittance phase angle, errors were
found comparable between the two arrangements. For the Visacro and Alipio
arrangement, (1) can be used with confidence when both potential probes are in close
proximity to the hemispherical electrode, where symmetry is acceptable (Fig. 4).
Alternatively, at least one potential probe should be placed at a very short distance from
the hemispherical electrode where the potential rise on the ground surface attains its
highest values, thus, dominating voltage measurements.

10
9
8 r1 = 5 m
7
|∆Y/Y| (%)

6 2m
5
4
3
2
1 0.23 m
0
0 3 6 9 12 15
r2 (m)

Fig. 6. Absolute relative error of the computed admittance amplitude, |DY/Y|, as a function of
the distances r1 and r2; Visacro and Alipio arrangement (Fig. 2).
1336 Z. G. Datsios et al.

In light of the above, the electrode arrangement of Fig. 3a has several advantages
with respect to that shown in Fig. 2. Specifically, by employing two potential probes
the effects of contact resistance and electrode polarization are eliminated. Also, due to
enhanced symmetry of the current and potential distributions, the potential probes can
be placed at greater distances from the hemispherical electrode (Figs. 5 and 6).
A simplified electrode arrangement has also been evaluated so as to facilitate
electrode setup in the field. The ring electrode of Fig. 3a is replaced by four equipo-
tential ground rods symmetrically placed around the hemispherical electrode (Fig. 3b)
at a distance of 30 m. In order to assess the effectiveness of this simplified arrangement,
the computed admittances were compared with those referring to the arrangement of
Fig. 3a for Dr = 60 m; typical results are shown in Fig. 7. For all cases it was found
that deviations are lower than *0.5% for both the amplitude and phase angle of the
admittance when the potential probes are placed along an axis that connects the
hemispherical electrode with a rod (Fig. 3b).

0.5
0.25 m 5m

0.4
|∆Y/Y| (%)

0.3

0.2

0.1

0
0 2 4 6 8 10
r2 (m)

Fig. 7. Absolute relative difference of the computed admittance amplitude, |DY/Y|, obtained by
the arrangements of Fig. 3a and b (Dr = 60 m) as a function of r1 and r2.

4.2 Two-Layer Soil Models

Two-layer soil models were also evaluated in order to better simulate field conditions.
Figure 8 shows the potential distribution around the hemispherical electrode for the
arrangement of Fig. 3a (Dr = 60 m) and soil case 6 (Table 1). The asymmetry is
evident even in close proximity to the electrode due to the different electrical resistivity
values of the two soil layers. The potential distribution in the ground is affected
considerably by the excitation frequency and the upper layer depth. Actually, symmetry
is enhanced with increasing excitation frequency and depth of the upper layer.
 Methods for Field Measurement of the Frequency-Dependent Soil 1337

0.3 m
0.5 m

1m

2m

Fig. 8. Potential distribution in the ground around the hemispherical electrode for the
arrangement of Fig. 3a (Dr = 60 m); =100 kHz, two-layer soils of case 6 (Table 1).

The values of soil electrical properties as determined using the computed admit-
tance in (1) were found to vary with excitation frequency and potential probe positions
(r1 and r2) due to the non-uniformity of the two-layer soil models. In fact, the measured
electrical properties of non-uniform soils are apparent quantities corresponding to a soil
volume up to a certain depth; the latter depends on excitation frequency and electrode
arrangement.
A preliminary investigation was conducted to assess the applicability of the
apparent soil electrical properties obtained using the arrangement of Fig. 3a to the
computation of the impedance of a ground electrode. A ground rod (length: 2 m,
diameter: 0.02 m) was simulated using the two-layer soils of case 5 (Table 1) and the
uniform soils with the corresponding apparent electrical properties; the frequency
spectra of the latter were determined from the arrangement of Fig. 3a (Dr = 60 m) for
(r1, r2) = (0.25 m, 0.9 m). Figure 9 shows the variation of relative error values of the
computed ground rod admittance with excitation frequency. Obviously, the apparent
electrical properties do not satisfactorily describe the two-layer soils. This was found to
apply also for the apparent electrical properties determined using the Visacro and
Alipio arrangement.
1338 Z. G. Datsios et al.

250
h = 0.3 m
0.5 m
200

|∆Y/Y| (%) 150 1.0 m

100
2.0 m
50

(a)
0
80
h = 2.0 m
1.0 m
70 0.5 m
0.3 m
60
|∆θ/θ| (%)

50

40

30
(b)
20
0.01 0.1 1 10 100
Frequency, f (kHz)

Fig. 9. Frequency spectra of the absolute relative error of the computed ground rod admittance
in two-layer (case 5, Table 1) and equivalent uniform soils using apparent electrical properties:
(a) amplitude, |DY/Y|, and (b) phase angle, |Dh/h|; parameter: upper layer depth.

Consequently, the use of the apparent electrical properties of non-uniform soils

determined for an arbitrary distance between potential probes may result in significant
errors in grounding system simulations. Further work is needed to develop appropriate
methods for measurement results interpretation, so as to derive soil models represen-
tative of non-uniform field conditions.

5 Conclusions

Several electrode arrangements for field measurement of the frequency-dependent soil

electrical properties have been evaluated through FEM computations. The developed
FEM model has been validated against results obtained using theoretical expressions
and grounding analysis software. A parametric analysis has been performed by varying
electrode dimensions, separation distances between electrodes and the excitation fre-
quency (up to 100 kHz); several uniform as well as two-layer soil models were
employed in computations.
The most appropriate electrode arrangement for field measurements of the
frequency-dependent soil electrical properties comprises a hemispherical and a ring
electrode (diameters: 0.5 and 60 m, respectively) used for current injection; the ring
electrode can be replaced by four equipotential ground rods to facilitate electrode
setup. The voltage arising between two points on the ground surface due to the injected
 Methods for Field Measurement of the Frequency-Dependent Soil 1339

current is measured using two test probes. Therefore, measurement errors due to
contact resistance and electrode polarization effects are eliminated.
The frequency-dependent electrical properties determined from field measurements
performed on non-uniform soils are apparent values, varying with potential probe
positions. Further work is needed to develop appropriate methods for measurement
results interpretation, so as to derive soil models representative of non-uniform field
conditions. In addition, investigation is required for frequencies higher than 100 kHz,
considering in computations wave propagation effects.

References
1. Scott, J.H., Carroll, R.D., Cunningham, D.R.: Dielectric constant and electrical conductivity
measurements of moist rock: a new laboratory method. J. Geophys. Res. 72(20), 5101–5115
(1967)
2. Snowden, D.P., Morris Jr., G.C., van Lint, V.A.J.: Measurement of the dielectric constant of
soil. IEEE Trans. Nucl. Sci. NS 32(6), 4312–4314 (1985)
3. He, J., Zeng, R., Zhang, B.: Methodology and Technology for Power System Grounding.
Wiley, Singapore (2013)
4. Datsios, Z.G., Mikropoulos, P.N.: Characterization of the frequency dependence of the
electrical properties of sandy soil with variable grain size and water content. IEEE Trans.
Dielectr. Electr. Insul. 26, 904–912 (2019)
5. IEEE Guide for Measuring Earth Resistivity, Ground Impedance, and Earth Surface
Potentials of a Grounding System. IEEE Std 81-2012 (2012)
6. IEEE Guide for Safety in AC Substation Grounding. IEEE Std 80-2013 (2015)
7. Visacro, S., Alipio, R.: Frequency dependence of soil parameters: experimental results,
predicting formula and influence on the lightning response of grounding electrodes. IEEE
Trans. Power Del. 27(4), 927–935 (2012)
8. COMSOL Multiphysics User’s Guide. COMSOL 4.3a (2012)
9. AC/DC Module User’s Guide. COMSOL 4.3a (2012)
10. Katsanou, V.N., Papagiannis, G.K.: Substation grounding system resistance calculations
using a FEM approach. In: Proceedings of the IEEE Bucharest PowerTech Conference,
Bucharest, Romania (2009)
11. Akbari, M., Sheshyekani, K., Alemi, M.R.: The effect of frequency dependence of soil
electrical parameters on the lightning performance of grounding systems. IEEE Trans.
Electromagn. Compat. 55(4), 739–746 (2013)
12. Akbari, M., Sheshyekani, K., Pirayesh, A., Rachidi, F., Paolone, M., Borghetti, A., Nucci, C.
A.: Evaluation of lightning electromagnetic fields and their induced voltages on overhead
lines considering the frequency dependence of soil electrical parameters. IEEE Trans.
Electromagn. Compat. 55(6), 1210–1219 (2013)
13. Trifunovic, J., Kostic, M.: Quick calculation of the grounding resistance of a typical 110 kV
transmission line tower grounding system. Electr. Pow. Syst. Res. 131, 178–186 (2016)
14. XGSLab User’s Guide, SINT Ingegneria, Release 8.3.1 (2018)
15. Zhang, B., He, J., Wang, S., Zhang, J., Li, Z., Li, W.: Discussion of measurement methods of
impulse grounding impedance. In: Proceedings of the 34th ICLP, Rzeszow, Poland, Paper
No. 152 (2018)
16. Hamzehbahmani, H., Haddad, A., Griffiths, H., Harid, N., Guo, D.: Application of an
analysis technique to characterise impulse response of grounding systems. In: Proceedings of
the 33rd ICLP, Estoril, Portugal, Paper No. 252 (2016)
 Effect of Quenching on Dielectric Properties
of ZnO Varistor Ceramics

Kangning Wu(&) , Yuwei Huang , and Jianying Li

State Key Laboratory of Electrical Insulation and Power Equipment,

Xi’an Jiaotong University, Xi’an, China
wkn.xjtu@stu.xjtu.edu.cn

Abstract. ZnO varistor ceramics quenched under different temperatures were

investigated from the aspect of dielectric responses. Nonlinear coefficient a and
electrical breakdown field E1mA were significantly different when quenching
temperature was around 600 °C. From dielectric responses under low temper-
atures, it was found that densities of both zinc interstitials and oxygen vacancies
increased at first and then decreased when quenching temperature decreased
from 1200 °C to room temperature. Noticeably, the quenching temperature
where zinc interstitial density reaches its peak was lower than that for oxygen
vacancy, suggesting that oxygen vacancies were more sensitive to the ambient
temperature. For dielectric responses under high temperatures, DC conductance
was so intense that relaxations with long relaxation times were covered and not
easily characterized via traditional dielectric spectra. Therefore, an improved
dielectric spectroscopy free of DC conductance was employed. In samples
quenched below 600 °C, two distinct relaxations originated from interfacial
polarization and interface states were found. The interface polarization only
appears when quenching temperature is below 600 °C, accompanied by notably
improved resistance in small-current region. With quenching temperature
increased, a new low frequency dielectric relaxation appeared, whose origin still
needs further investigation.

Keywords: ZnO
Varistor
Dielectric
Schottky barrier

1 Introduction

ZnO varistor ceramics have been widely employed in overvoltage protection because
of their excellent nonlinear current-voltage characteristics and surge current absorption
capability. It is acknowledged that double Schottky barriers (DSB) at grain boundaries
are responsible for the electrical nonlinearity. Negatively charged interface states
originated from absorbed oxygen, zinc vacancy, etc. are formed at grain boundaries [1].
Electrons in ZnO grains nearby to the surface are exhausted. Depletion layers are thus
formed, which are mainly composed of intrinsic point defects (zinc interstitial and
oxygen vacancy) [2]. Either the short-term electrical properties or long-term degra-
dation is inevitably modulated by behaviors of interface states and intrinsic point
defects in ZnO varistor ceramics.
Numbers of investigations were conducted on those defects in ZnO varistor
ceramics. For those intrinsic point defects in depletion layers, they were widely

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1340–1346, 2020.
https://doi.org/10.1007/978-3-030-31676-1_125
 Effect of Quenching on Dielectric Properties of ZnO 1341

characterized by frequency domain dielectric spectroscopy (FDS) [3], deep level

transient spectroscopy (DLTS) [4] and thermally stimulated depolarization current
(TSDC) [5]. Zni˙ was found as a shallow donor, which located at *0.05 eV below the
conduction band. On the contrary, Zni¨ and VO˙ were deep donors at *0.23 eV and
*0.33 eV below the conduction band, respectively [6]. Unfortunately, although
decisive role of interfacial relaxation on conduction behavior of ZnO varistor ceramics
was proposed recently [7], those interfacial relaxation responses are still not clear [8].
As a kind of precise electronic ceramics, electric properties of ZnO varistors depend
largely on sintering processes. Development of interfacial microstructure during
cooling of ZnO varistor ceramics were investigated by Olsson et al. Evolution of Bi-
rich interfacial microstructure was detailly discussed [9]. However, the evolution of the
barrier structure, especially those extrinsic defects, were seldom reported.
In our previous work, multiscale trapping behaviors at grain boundary in ZnO
varistor ceramics were characterized as dielectric relaxations by employing an opti-
mized dielectric spectroscopy [10–12]. In this paper, ZnO samples were quenched from
various temperatures during sintering process to reserve the barrier structure. Dielectric
relaxations induced by multiscale electron trapping behaviors of defects at grain
boundary were characterized, which show consistence with development of electrical
nonlinearity during sintering.

2 Experimental

ZnO ceramic samples of a commercial formula (Bi2O3, Co, Sb, Ni, Mn and Al) were
prepared by a semi-solution method. After ball milling in a polyamides bottle for 12 h,
the mixture was dried, sieved and calcined at 600 °C for 3 h. The calcined powders
were mixed with 1 wt% polyvinyl alcohol (PVA) solution as a binder and pressed into
green discs of 12 mm in diameter and 2 mm in thickness. Finally, the discs were
sintered in air at 1200 °C for 2 h with the heating rate of 150 °C/h and cooling rate of
100 °C/h. Quench process was conducted by quickly taking the samples from the
furnace to a stainless block in air. As shown in Fig. 1, samples were quenched in air
from 1200, 1000, 800, 600 and 400 °C to room temperature in 3 min (labelled as Q-
1200, Q-1000, Q-800, Q-600, Q-400 and As-prepared), respectively.

Q-1200
1200
Q-1000
Temperature (oC)

Q-800
800
Q-600

Q-400
400

As-prepared

0 3 6 9 12 15 18
Time (h)

Fig. 1. Sintering process and quenching temperatures for the investigated samples.
1342 K. Wu et al.

Current-voltage characteristics were measured using a DC power source

(WJ10001D, China) and a Multi-function digital meter (HP34401A, America) at room
temperature. Dielectric properties were measured by a broadband dielectric measure-
ment system (Novocontrol, Concept 80, Germany) in −90 °C–220 °C and 10−1 Hz–
107 Hz.

3 Results and Discussion

3.1 Electrical Properties
Current-voltage (I-V) characteristics, which are of great importance for varistors, are
plotted in Fig. 2. Electrical breakdown field E1mA (electric field at 1 mA/cm2) and a
(nonlinear coefficient) were calculated and shown in the inset of Fig. 2. Nonlinearity
(a > 10) was found in all the samples indicating the formation of Schottky barriers at
ZnO-ZnO grain boundaries. However, remarkable discrepancies were further detected
during cooling period, with 600 °C as a critical temperature. On one hand, as plotted in
inset of Fig. 2, remarkable increase of a from *20 to *40 is observed when the
quench temperature decreased below 600 °C. On the other hand, E1mA reached its
maximum under quench temperature of 600 °C. Samples quenched below 600 °C
exhibited abruptly improved electrical properties, which was accordant with other
reported results. Therefore, it is reasonable to deduce that well-developed Schot-
tky barrier and grain boundary resistance were mainly formed at temperature nearby
600 °C.

600
U (V)

320
40
300

400 30
As prepared
E1mA(V/mm)

280
20 260 Q-400
10 240 Q-600
220 Q-800
0
0 400 800 1200 Q-1000
Quench temperature (oC)
Q-1200
200
0.0 0.5 1.0
I (mA)

Fig. 2. I-V curves of ZnO ceramic samples quenched at different temperatures. Insets are
nonlinear coefficient a and breakdown field E1mA.

3.2 Intrinsic Point Defects

FDS is an effective way of revealing dielectric relaxation processes correlated with
those defects at grain boundaries. As shown in Fig. 3(a), two relaxation peaks A and B
were observed shifting towards higher frequency range with increasing temperatures in
e″ spectroscopy, indicating thermally activated processes. Therefore, their activation
 Effect of Quenching on Dielectric Properties of ZnO 1343

energies were calculated via Arrhenius equation. As a result, 0.23 eV and 0.33 eV
were acquired from peak frequencies under difference temperatures for relaxations A
and B, respectively. These two relaxations were reported independent on fabrication
processes, annealing, doping, etc., indicating intrinsic natures. It is widely proposed
that they are correlated with intrinsic point defects zinc interstitials (Zni¨) and oxygen
vacancies (VO˙), respectively. In consequence, development of oxygen vacancies and
zinc interstitials can be clearly characterized via e″ spectroscopy.

Fig. 3. Frequency dependence of e″ from −100 to −60 °C (a) and simulation of e″ spectroscopy
of as prepared sample based on Cole-Cole equation under −110 °C (b).

Generally, dependence of e″ on angular frequency follows Cole-Cole equation:

Xn ki ðxsi Þ1ai cosðpai =2Þ

e00 ðxÞ ¼ k0 xa0 þ ð1Þ
i¼1
1 þ 2ðxsi Þ1ai sinðxsi =2Þ þ ðxsi Þ2ð1ai Þ

where k0 is magnitude of DC conductivity while ki (i = 1, 2, 3 …) is magnitude of

permittivity from the ith relaxation (Δei). si is relaxation time and ai is depression angle.
e″ curves of all the samples at −110 °C were fitted according to Eq. (1). Calculated
parameters are listed in Table 1, and the fitting curves, taking as-prepared sample as an
example, are plotted in Fig. 3(b).

Table 1. Calculated parameters of relaxations A and B.

Samples Relax. A Relax. B
k1 a1 K2 a2
Q-1200 32.0 0.12 65.5 0.32
Q-1000 46.0 0.12 44.0 0.24
Q-800 57.0 0.12 51.0 0.30
Q-600 33.0 0.12 33.0 0.20
Q-400 37.5 0.12 37.5 0.20
As prepared 39.5 0.12 30.0 0.20
1344 K. Wu et al.

It is clear that densities of both Zni˙˙ and VO˙ increased firstly and then declined
during cooling process. Notably, VO˙ was found more sensitive than Zni˙˙ and the
tendency of density variation of Zni˙˙ followed the variation of VO˙. Maximum den-
sities of Zni˙˙ and VO˙ appear at about 1100 °C and 850 °C, respectively. Oxygen was
lost from the lattice at high temperatures so that oxygen vacancies were gradually
accumulated. Zinc atoms were consequently dragged into interstitial due to lattice
distortion. As temperature decreased, combination of oxygen became dominant leading
to reduction of both Zni˙˙ and VO˙. When the temperature decreased below 500 °C,
densities of both zinc interstitials and oxygen vacancies tended to become constant.

3.3 Extrinsic Defects

Besides of intrinsic point defects, extrinsic defects at grain boundary regions are also
important on modulating carrier transport in ZnO varistors. Unfortunately, relaxation
times for those defects are commonly so long that the corresponding dielectric
responses can be only observed in low frequency range where DC conduction is
intense. Component of DC conduction might cover dielectric in traditional dielectric
spectra so that further investigations on them are hard to be carried out. In our previous
work, an improved (∂e′/∂lnx)/e′ spectroscopy was proposed, which was able to
eliminate effect of DC conduction and unveil the electron trapping behavior of surface
states in ZnO varistor ceramics. In this paper, the frequency dependent (∂e′/∂lnx)/e′
was employed and the spectroscopy of ZnO varistor ceramics quenched at 1200 °C and
as-prepared under 170 °C–280 °C are exemplified in Fig. 4(a) and (b), respectively.

0.8 Relax. D (b)

Relax. E
(a) 0.2
Relax. C
Relax. D
0.6

0.4 Relax. C
0.1

0.2 170 oC 190 oC

150 oC 170 oC
210 oC 240 oC
260 oC 280 oC 190 oC 210 oC
0.0 0 0.0 0
10 101 102 103 104 105 10 101 102 103 104 105
f (Hz) f (Hz)

Fig. 4. Frequency dependence of (∂e′/∂lnx)/e′ in Q-1200 (a) and as-prepared (b) ZnO samples
under high temperatures.

In the as-prepared sample, there are two distinct relaxation peaks (Relaxation C and
D) at frequency range of 100–102 Hz and 102–104 Hz, respectively. Both relaxations
shifted to higher frequencies under higher temperatures, indicating thermally activated
processes. Their activation energies were calculated based on Arrhenius equations, as
shown in Table 2. Notably, relaxation C was undetectable in those samples quench at
high temperatures (>800 °C). It was distinctively found when the sample was cooled to
 Effect of Quenching on Dielectric Properties of ZnO 1345

*600 °C. The appearance of relaxation C was accomplished by the notably improved
resistance in small-current region. It is reasonable to deduce that relaxation C is closely
correlated with overall electrical properties. In our previous work, relaxation C is found
originating from interfacial polarization between intergranular phase and depletion
layers [6, 8].
When the quench temperature is higher than 600 °C, relaxation C was unable to be
clearly characterized. Peak of relaxation D became wider, as well. In addition, another
relaxation (denoted as Relaxation E) appeared at low frequency range (<102 Hz). The
activation energies of relaxation E under different quench temperatures were also
included in Table 2. It is clear that activation energies of relaxation D in those samples
quenched higher than 600 °C were much lower than those quenched at 600 °C. In our
previous report, it was proposed to be originated from interface states [8]. Electrons
were captured and emitted from interface states under applied AC voltage, resulting in
a dielectric relaxation with long relaxation time. Relaxation E was firstly found in ZnO
varistor ceramics. Its origin still needs further investigation.

Table 2. Activation energies of high-temperature dielectric relaxations of samples.

Samples EC/eV ED/eV EE/eV
Q-1200 – 1.12 1.52
Q-1000 – 0.89 1.01
Q-800 – 0.89 1.08
Q-600 0.60 1.44 –
Q-400 0.62 1.05 –
As prepared 0.63 1.08 –

4 Conclusions

Great discrepancies on nonlinearity of ZnO varistor ceramics were found when

quenching temperature was around 600 °C. Densities of both zinc interstitials and
oxygen vacancies increased at first and then decrease when quenching temperature
decreased from 1200 °C to room temperature. In addition, an improved dielectric
spectroscopy free of dc conduction is employed for high temperature relaxation
analysis. In samples quenched below 600 °C, two distinct relaxations originated from
interfacial polarization and interface states were found. The interface polarization only
appears when quenching temperature is below 600 °C. With quenching temperature
increases, a new low frequency dielectric relaxation appears.

Acknowledgments. This work is supported by the National Key Research and Development
Program of China (No. 2018YFB0905802), the State Key Program of National Basic Research
(973 Program) of China (No. 2015CB251003).
1346 K. Wu et al.

References
1. Stucki, F., Greuter, F.: Key role of oxygen at zinc oxide varistor grain boundaries. Appl.
Phys. Lett. 57, 446–448 (1990)
2. Cordaro, J.F., Shim, Y., May, J.E.: Bulk electron traps in zinc oxide varistors. J. Appl. Phys.
60, 4186–4190 (1986)
3. Cheng, P., Li, S., Zhang, L., Li, J.: Characterization of intrinsic donor defects in ZnO
ceramics by dielectric spectroscopy. Appl. Phys. Lett. 93, 988 (2008)
4. Ohbuchi, Y., Kawahara, T., Okamoto, Y., Morimoto, J.: Characterization of interface states
in degraded ZnO varistors. Jpn. J. Appl. Phys. 41, 190–196 (2002)
5. Eda, K., Matsuoka, M.: Thermally stimulated current in nonohmic ZnO ceramics. Jpn.
J. Appl. Phys. 16, 195–196 (1977)
6. Zhao, X., Li, J., Li, H., Li, S.: Intrinsic and extrinsic defect relaxation behavior of ZnO
ceramics. J. Appl. Phys. 111, 124106 (2012)
7. Yang, X., Hu, J., He, J.: Effect of interfacial charge relaxation on conducting behavior of
ZnO varistors under time varying electric fields. Appl. Phys. Lett. 110, 485–646 (2017)
8. Wu, K., Huang, Y., Xin, L., Li, J., Li, S., Liu, W.: Understanding of DC degradation of ZnO
varistor ceramics from the aspect of high-temperature relaxation. In: 12th International
Conference on the Properties and Applications of Dielectric Materials, Xi’an, China, May
2018
9. Olsson, E., Dunlop, G.L., Osterlund, R.: Development of interfacial microstructure during
cooling of a ZnO varistor material. J. Appl. Phys. 66, 5072–5077 (1989)
10. Wu, K., Huang, Y., Li, J., Li, S.: Space charge polarization modulated instability of low
frequency permittivity in CaCu3Ti4O12 ceramics. Appl. Phys. Lett. 111, 042902 (2017)
11. Wu, K., Huang, Y., Hou, L., Zhuang, T., Li, J., Li, S.: Effects of dc bias on dielectric
relaxations in CaCu3Ti4O12 ceramics. J. Mater. Sci. -Mater. Electron. 1–7 (2017)
12. Wu, K., Huang, Y., Xing, Z., Zhang, C., Hu, X., Guo, P., Zhang, J., Li, J.: Understanding the
validity of impedance and modulus spectroscopy on exploring electrical heterogeneity in
dielectric ceramics. J. Appl. Phys. 125, 084103 (2019)
 Analysis of Lightning Overvoltage Base
on Grounding Method of Lightning
Arresters in Wind Farm

Jin-Hyuk Kim1 , Kyo-Ho Kim1(&) , and Jung-Wook Woo2

1
Hankyong National University, Anseong, Republic of Korea
asas3212@naver.com, kyuho@hknu.ac.kr
2
Korea Electric Power Research Institute, Daejeon, Republic of Korea

Abstract. This study simulates lightning strikes and a wind farm using the
simulation tool EMTP-RV and analyses the effects of lightning overvoltage. It
compares the lightning overvoltage for two grounding methods of lightning
arresters-common ground and connected ground. For the simulations, the
ground impedance of the lightning arrester is selected according to the earth
resistivity and lightning frequency. The study also analyses the resulting mag-
nitude of the overvoltage when corona is the lightning strikes. Thereby, it
suggests a method of reducing the lightning overvoltage through various sim-
ulations, depicting various conditions such as the change of ground impedance
and the grounding method of the lightning arrester.

Keywords: Lightning
Ground impedance
Lightning arrester
Wind farm

1 Introduction

It is possible to generate overvoltage and overcurrent in the inner and outer power grid
of the wind power generation farm. Lightning strikes can cause damage to electronic
equipment used in wind power generation, cause breakage of blades, and other
equipment-related problems. Therefore, it is important to reduce the impact of lightning
strikes because lightning can lead to overvoltage in the grid of the wind turbine and
affect power quality. So, many countries actively conduct studies on the damage caused
by lightning strikes to wind power generation and the countermeasures to prevent these
damages. Holger Hecler studied the challenging operating conditions of the offshore
wind turbines and the requirement for the design of surge protection devices in these
operating conditions. They also describe the design principles and operational behavior
of this new arrester for wind turbines. Karbalaye Zadeh analyzed the metal oxide surge
arresters in IEEE Model, Pinceti Model, and Popov model and proposed a model which
is adequate for a sudden the rise of surge currents. Mungkung studied the lightning
wave in a transmission line at 115 kV voltage level in Thailand, using an ATP/EMTP
program to create the model of the transmission line and lightning surge [1–4]. This
paper analyses the effects of overvoltage resulting from lightning when the ground
impedance is changed. The grounding method of the lightning arrester is also taken into
account.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1347–1355, 2020.
https://doi.org/10.1007/978-3-030-31676-1_126
1348 J.-H. Kim et al.

2 System Configuration

Specific details are provided in the following sections.

2.1 Lightning Model

The lightning model is based on Isurge Source provided by EMTP-RV. This lightning
model has duration of 0 s to 0.5 ms and the maximum lightning current is 15 kA. The
equation of the lightning model is shown in (1) and the waveform can be adjusted by
changing the numerical values of A and B in the equation. The lightning waveform is
shown in Fig. 1 and the lightning constant of the model is shown in Table 1.

f ðtÞ ¼ Ilight
ðeAt  eBt Þ ð1Þ

Fig. 1. 15 kA lightning

Table 1. Isurge surge function

Constant Value
Amplitude [A] 25000
A [1/s] −9000
B [1/s] −60000
T start [ms] 0
T stop [ms] 0.5

2.2 Ground Impedance

Figure 2 shows the frequency spectra of standard surge voltage and current waveforms
[5]. Since the lightning surge voltage and current are distributed over a large frequency
band, the ground impedance must be considered.
 Analysis of Lightning Overvoltage Base on Grounding Method 1349

Fig. 2. Frequency due to surge voltage and current

Figure 3 shows the ground impedance according to the frequency of the lightning
surge when the earth resistivity is 1000 X m [5]. When the frequency of the lightning
surge is less than 10 kHz, it has a constant value. For the higher frequencies grounding
impedance increases due to the inductive effect [6–8].

Fig. 3. Ground impedance according to frequency

2.3 Grounding Method

Japanese Industrial Standard JIS A-4201 accepts a connecting ground method because
the current affects device frame when lightning strikes on a lightning rod and a steel
tower. In this case, it is difficult to reduce the individual ground impedances (Fig. 4).

Fig. 4. Common and connecting ground

1350 J.-H. Kim et al.

2.4 Corona Model

Corona has a significant effect on overvoltage and wave propagation. The currently
available method for corona modelling relies heavily on the experimental testing of
transmission lines for extracting the charge-voltage characteristic. The electric field
intensity at the onset of corona can be predicted from the Peek’s formula [9]. The
corona model is developed in a three-phase corona device using the Suliciu model [10].
This corona model is provided by EMTP-RV. The corona current occurs at the first
0.05 ms, and increases gradually to about 6.08 kA, 4.86 kA, and 3.65 kA (Fig. 5).

Fig. 5. Corona current waveform

3 Simulation Results

The simulation condition is that of the lightning striking on the blades of the wind
turbine. The ground impedance is about 60 X, 100 X, and 110 X and the lightning
surge frequencies are 100 kHz, 1 MHz, and 10 MHz, respectively, when the earth
resistivity is 1000 X m. The lightning overvoltage considering the corona effect and
the grounding method is compared.

3.1 Lightning Overvoltage with Respect to Ground Impedance

in Common Grounding
When the common grounding method is used for the lightning arrester, the maximum
value of the lightning overvoltage with respect to the ground impedance is compared.
When the ground impedance is 60 X, 100 X, and 110 X, the maximum value of the
lightning overvoltage is about 308.551 kV, 403.651 kV, and 421.874 kV, respectively.
Simulation results show that the higher the ground impedance, the greater the lightning
overvoltage (Figs. 6 and 7).
 Analysis of Lightning Overvoltage Base on Grounding Method 1351

Fig. 6. In the common grounding method, when the lightning arrestor ground impedance is
60 X

Fig. 7. In the common grounding method, when the lightning arrestor ground impedance is
110 X

3.2 Lightning Overvoltage with Respect to Ground Impedance

in Connecting Grounding
When the lightning arrester is connected using the connecting grounding method, the
maximum value of the lightning overvoltage with respect to the ground impedance is
compared. When the ground impedance is 60 X, 100 X, and 110 X, the maximum
value of the lightning overvoltage is about 246.834 kV, 322.142 kV, and 349.465 kV,
respectively. In the case of common ground, the higher the ground impedance, the
larger the lightning overvoltage. However, the maximum value of the lightning over-
voltage is reduced by about 17%–20%, when compared to the common grounding
method (Figs. 8 and 9).
1352 J.-H. Kim et al.

Fig. 8. In the connecting grounding method, when the lightning arrestor ground impedance is
60 X

Fig. 9. In the connecting grounding method, when the lightning arrestor ground impedance is
110 X

3.3 Lightning Overvoltage with Respect to Ground Impedance

in Common Grounding Method and Considering the Corona Effect
In this case, the simulation conditions are the same as in Sect. 3.1, and the corona effect
is also considered. The maximum lightning overvoltage is measured when the corona
effect is assumed to occur in a power line leading to a wind power generator and a
substation. When the ground impedance is 60 X, 100 X, and 110 X, the maximum
value of the lightning overvoltage is about 577.074 kV, 763.461 kV, and 799.177 kV,
respectively. The maximum overvoltage increased about 46–47% for case considering
the corona effect when compared with results given in Sect. 3.1 (Figs. 10 and 11).
 Analysis of Lightning Overvoltage Base on Grounding Method 1353

Fig. 10. In the common grounding method, when the lightning arrestor ground impedance is
60 X with corona

Fig. 11. In the common grounding method, when the lightning arrestor ground impedance is
110 X with corona

3.4 Lightning Overvoltage with Respect to Ground Impedance

in Connecting Grounding Method and Considering the Corona Effect
For this analysis, the grounding method of the lightning arrester is connecting
grounding, and the maximum value of the lightning overvoltage with respect to the
ground impedance is measured considering the corona effect. When the ground
impedance is 60 X, 100 X, and 110 X the maximum value of the lightning overvoltage
is about 445.08 kV, 614.202 kV, and 653.355 kV respectively. The maximum over-
voltage increases by about 44–46% when considering the corona effect in connecting
grounding method when compared to Sect. 3.2. When compared with Sect. 3.3, the
overvoltage was reduced from a minimum of 22.32% to a maximum of 29.66% when
the common grounding method is changed to the connecting grounding method
(Figs. 12 and 13).
1354 J.-H. Kim et al.

Fig. 12. In the connecting grounding method, when the lightning arrestor ground impedance is
60 X with corona

Fig. 13. In the connecting grounding method, when the lightning arrestor ground impedance is
110 X with corona

4 Conclusions

This paper analyses the effects of overvoltage caused by lightning strikes when the
ground impedance was changed. The grounding method of the lightning arrester and
the corona effect were also considered.
When the grounding method of the lightning arrester increases the ground impe-
dance of the lightning arrester both in the common ground and the connecting ground,
the lightning overvoltage also increases. When the corona effect is considered, a rise of
about 44%–47% is observed in the overvoltage as opposed to when only lightning
strikes are considered.
When using the connecting ground method, the maximum value of the lightning
overvoltage is reduced by about 20% compared to that of the common ground method.
The use of the connecting ground method resulted in a lesser rise in the overvoltage
due to the corona effect and a total overvoltage reduction of up to about 29% when
compared to the common ground method.
Therefore, it is necessary to keep the ground impedance as low as possible in order
to reduce the influence of the overvoltage, and it is effective to reduce the overvoltage
by using the grounding system of the lightning arrester.
 Analysis of Lightning Overvoltage Base on Grounding Method 1355

Acknowledgments. This research was supported by Korea Electric Power Corporation (Grant
Number: R18XA06-59).

References
1. Hecler, H., Wosgien, J., Wetter, M.: New lightning current arrester design for onshore and
offshore wind turbines. In: International Symposium on Lightning Protection (2011)
2. Sorensen, T., Jensen, F.V., Raben, N., Lykkegaard, J., Saxov, J.: Lightning protection for
offshore wind turbines. In: International Conference on Electricity Distribution (2001)
3. Karbalaye Zadeh, M., Abniki, H., Shayegani Akmal, A.A.: The modeling of metal-oxide
surge arrester applied to improve surge protection. In: International Conference on Power
Electronics and Intelligent Transportation System (2009)
4. Mungkung, N., Wongcharoen, S., Tanitteerapan, T., Saejao, C., Arunyasot, D.: Analysis of
lightning surge condition effect on surge arrester in electrical power system by using
ATP/EMTP program. Int. J. Electr. Comput. Energ. Electron. Commun. Eng. (2007)
5. Cho, S.-C., Lee, B.-H.: A method of computing the frequency-dependent ground impedance
of horizontally-buried wires. Trans. Korean Inst. Electr. Eng. 65(5), 745–752 (2016)
6. Choi, J.H., Lee, B.H.: Frequency-dependent grounding impedance of the counterpoise based
on the dispersed currents. J. Electr. Eng. Technol. 7(4), 589–595 (2012)
7. Verma, R., Mukhedkar, D.: Fundamental considerations and impulse impedance of
groundings. IEEE Trans. PAS 100(3), 1023–1030 (1981)
8. IEEE Std. 80-1986, IEEE Guide for Safety in AC Substation Grounding, pp. 277–284. IEEE
Inc. (1986)
9. Maruvada, P.S., Menemenlis, H., Malewski, R.: Corona characteristics of conductor bundles
under impulse voltages. IEEE Trans. Power Appar. Syst. 96, 102–115 (1977)
10. Suliciu, M.M., Suliciu, I.: A rate type constitutive equation for the description of the corona
effect. IEEE Trans. PAS-100(8), 3681–3685 (1981)
 Resonance Behaviour of Shielded High-Voltage
Laboratories and the Effect of Interference
on Impulse Measurements

T. C. Schlüterbusch(&), S. Passon, and J. Meisner

Physikalisch-Technische Bundesanstalt Braunschweig,

Bundesallee 100, 38116 Brunswick, Germany
tim.schlueterbusch@ptb.de

Abstract. As the national metrology institute of Germany, the Physikalisch-

Technische Bundesanstalt (PTB) is dedicated to ensure the quality of metrology
in Germany by performing its tasks. The working group 2.32 for High-voltage
metrology developed a transient recorder with a corresponding software. In the
Process of testing the new measuring system there were superimposed oscilla-
tions observed. In this paper we want to investigate these observations. More-
over, the influence of the oscillations on the analysis of the measurement data is
not investigated so far. The recorded curves in the laboratory are analysed
according to de IEC 60060-1:2010 in which context the measurement data is
treated by software-based filters and algorithms that perform a nonlinear
regression. If the input data is biased too much by interference it can be possible
that the measurement and the resulting values are negatively affected.
The impulse generator with its spark gaps is assumed to be the source of a
high energetic transient electric field. Like many other laboratories the PTB’s
high-voltage laboratory is fully shielded with copper foil on the inside of every
wall, the floor and the ceiling. The architecture of this laboratory room forms an
almost perfectly rectangular box-shaped cavity with electric conductive walls.
This formation is capable of behaving like a hollow waveguide conductor for
certain frequencies. Thus, it is possible that the resonance frequency of the
laboratory room is excited by the impulse generator.

Keywords: Measurement
Lightning
Impulse
Interference
Resonance

1 Introduction

In the high-voltage laboratory of the PTB we already observed oscillations while

performing measurements with lightning (LI) or switching Impulses (SI) in the past. In
the most cases these interferences where not visible due to a low sample-rate or
resolution of our previously used digital-recorder (digitizer). We became aware of the
oscillations in the process of developing a new measurement system with better
specifications. In the first tests several oscillations in the frequency range of about
20 MHz to 35 MHz were clearly visible superimposed to the double exponential
lightning impulse (LI).

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1356–1366, 2020.
https://doi.org/10.1007/978-3-030-31676-1_127
 Resonance Behaviour of Shielded High-Voltage Laboratories 1357

2 Source of the Interferences

We assume the impulse generator with its spark gaps to be the source of the inter-
ferences. After the ignition of the spark gaps the fully shielded high-voltage laboratory
acts as a cuboid-shaped resonator. The resonance frequency of an electro-magnetic
wave in a cuboid cavern is given in the literature as Eq. (1) [1].
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
c mp2 np2 lp2
fres; m; n; l ¼ þ þ ; n; m; l 2 N [ f0g ð1Þ
2p a b h

Where a set of n, m and l indicate the mode of the higher harmonics. These factors can
only have values of positive natural numbers or zero. The lowest frequency is deter-
mined by the longest dimension of the laboratory test room.
When calculating this equation with the dimensions of the laboratory of height
h = 12−13 m, length a = 20 m, and width b = 11 m the following frequencies in
Table 1 are in theory propagable frequencies (excitable modes) in the resonator. In
practical terms of course, these values will not be discretely visible in a measurement
due to obstacles in the laboratory and other imperfect circumstances.

Table 1. Examples of excitable modes in a resonator with the dimensions of PTBs high voltage
laboratory
m 1 2 1 2
n 0 1 2 2
l 0 0 0 0
Frequency 7.49 20.3 28.3 31.1
[MHz]

3 Examination of the Resonance Behavior in a Fully Shielded

High-Voltage Laboratory

In this section the results of the frequency response measurement in the high voltage
laboratory are presented. The purpose of this measurement is to examine the dominant
frequencies or the frequency range which is propagated in the laboratory room.

3.1 Test Setup

To determine the frequency dependent behaviour of the laboratory an EM-Test gen-
erator was used to generate an alternating electro-magnetic field of specific frequencies.
As an antenna a simple copper conductor was connected to the positive output of the
generator while the negative output was connected to the earth potential of the labo-
ratory shielding. To prevent the influence of the antennas length the test was repeated
with several different antenna lengths from 1 m up to 10 m. With the antenna physi-
cally placed near the impulse generator we try to accomplish circumstances which are
1358 T. C. Schlüterbusch et al.

possibly close to the situation when testing and measuring LI under normal conditions.
A high bandwidth digital oscilloscope which is used as a spectrum-analyzer with its
FFT-analysis is connected to the reference divider of the PTB for LI and SI. This way
the Impulse divider is used as antenna for the spectrum-analyzer. The Cable between
the oscilloscope and the reference divider is the dedicated measurement cable which is
always used in combination with the reference divider including its termination
resistor. The presence of the termination resistor should not have an influence on the
interferences due to their supposed nature as common mode signal. The following
Fig. 1 is a sketch of the setup in the high voltage laboratory as a top view. All four
walls, the ceiling and a layer beneath the floor tiles of the laboratory are covered with
copper sheets. All doors and windows are equipped with sealings aimed towards the
insulation of electromagnetic fields. This extensive shielding of the high voltage test
area was applied with the intention to protect other nearby laboratories on the property
of the institute from the possible interferences caused by the high voltages generated in
the laboratory.

Fig. 1. Laboratory setup for the EM-frequency-response-test

3.2 Test Method and Results

The EM-generator is capable of generating a frequency sweep from 1 MHz to
100 MHz. The amplitude of the output is set to its maximum of 30 V for this device.
The frequency is increased step-wise in steps of 100 kHz with a test time of 1.5 s and a
pause of 0.5 s between two frequencies. The spectrum analyzer saves the maximum
amplitude for every frequency component measured.
In Fig. 2 a spectrum is shown for the antenna length of 2 m blank copper and
another 2 m laboratory wire. This spectrum was chosen because in this test all fre-
quencies of the peak values are also present in the other results. Therefore, this can be
taken for the best representation of the frequency behavior of the laboratory. Clearly
visible are peaks in the range of 15 MHz to 35 MHz. Note that the amplitude in Fig. 2
does not allow a statement about the electric or magnetic field strength. The amplitude
is a combined result of the transmission behaviour of the antenna, the propagation
characteristics in the room and the as receiving antenna used reference voltage divider.
 Resonance Behaviour of Shielded High-Voltage Laboratories 1359

The values in µV are the peak voltages for the respective frequency the used oscil-
loscope was able to measure during the test.

Fig. 2. Measured frequency response of the laboratory

As shown in Fig. 1 there is not only equipment present in the laboratory which is
actively needed for the measurement of voltage impulses but also equipment for other
types of measurement. This equipment is present due to storage purposes, is not
moveable due to its size or the need of fixed installation. For this reason, one can expect
that the propagation of electromagnetic waves will be more or less - depending on the
material of the objects - obstructed by these objects in the laboratory. This will have an
influence on the imperfect behaviour of the laboratory as a resonator.

4 Examination of the Influence on the Analysis

of the Impulses

The standard IEC 60060-1:2010 [2] for the analysis of lightning impulses (LI) proposes
an algorithm, which was developed for the correct consideration of superimposed
oscillation lower than 500 kHz in the result parameters. This was done to correctly take
in account stress for insulation material caused by oscillations in this frequency range.
The method is based on a nonlinear regression, using the Levenberg-Marquard-Method
and a digital filter. The nonlinear regression generates parameters for the mathematical
description of a non-distorted equivalent LI. The mathematical description follows the
function in Eq. (2) [2]
 ðtt Þ ðttd Þ

d
uð t Þ ¼ U e s1
e s2
ð2Þ

This way the equivalent impulse should be a possibly good approximation of the
mathematical function based on all recorded samples of the impulse signal. The
equivalent curve is then subtracted from the original curve. The residual signal rep-
resents the isolated noise and superimposed interferences. On this residual signal the
1360 T. C. Schlüterbusch et al.

digital lowpass filter, which is described in the standard IEC 60060-1:2010, is applied.
The resulting filtered signal is then added to the equivalent curve. This leads to a so-
called test curve on which the impulse parameters like the peak value Up and the
risetime T1 are determined.
The fundamental problem with the observed oscillations in the MHz-range is that
although the filter eliminates the visible oscillations for the evaluation part of the
algorithm, a negative influence on the nonlinear regression can be possible. The
Levenberg-Marquardt-Method tries to find the optimum trend through the collection of
samples which are supplied to the algorithm. If the sampled signal is highly distorted
by the observed oscillations there is a risk of a biasing influence by the interferences.
Especially if the predominant interference consists of a multitude of frequency com-
ponents that can form complex signal shapes.

4.1 Method of Examination

To investigate the influence of the oscillations we executed the following operations in
a specifically developed software.
• Isolation of the Interference from a recorded Impulse Measurement
• Superposition of the interferences to a waveform of the “TDG” Test Data Gener-
ator, which is defined in the standard IEC 61083-1:2012 [3]
• Examination of the parameters in a measurement software for Lightning Impulses
which has got implemented the standard evaluation of the parameters.
With these operations we can manipulate an impulse with well-known resulting
parameters. We can then compare the evaluation results of the manipulated standard
curve with the parameters that result with a manipulated curve.

4.2 Results of the Examination

The following Figs. 3, 4, 5, 6, 7, 8, 9, 10 and 11 show the signals that have been used for
the examination. Figure 3 is a mathematically generated impulse curve “LI-A1” used and
defined by the standard IEC 61083-1:2012 [3] with given reference parameters.

Fig. 3. IEC 61083-1:2012 [3] TDG standard impulse “LI-A1”

 Resonance Behaviour of Shielded High-Voltage Laboratories 1361

Fig. 4. Measured impulse signal 1 with interferences

Fig. 5. Interference 1 isolated from measured impulse signal 1 in Fehler! Verweisquelle

konnte nicht gefunden werden.

Fig. 6. Frequency domain of the interference 1 in Fehler! Verweisquelle konnte nicht

gefunden werden.
1362 T. C. Schlüterbusch et al.

Fig. 7. LI-A1 (Fehler! Verweisquelle konnte nicht gefunden werden.) with superimposed
interference 1 (Fehler! Verweisquelle konnte nicht gefunden werden.)

Figures 4, 5, 6 and 7 illustrate the examination process for one measured signal
(marked with 1). For a more comprehensive examination all steps were repeated for
another measured impulse signal (marked with 2) with a slower T1 parameter illustrated
by Figs. 8, 9, 10 and 11.
The used original signals Figs. 4 and 8 are intentionally chosen measurements with
a worst-case scenario of interferences, without any measures taken to reduce the dis-
turbances. Figures 6 and 10 are the representations of the isolated interreference Sig-
nals (Figs. 5 and 9) in the frequency domain. Based on Figs. 6 and 10 the observation
can be made that there are indeed some approximate matches in the frequency com-
ponents of the recorded interferences, the frequency response in Fig. 2 and the
determined frequencies in Table 1. For this reason, the theory on the origin of the
interferences seems plausible.

Fig. 8. Measured impulse signal 2 with interferences

 Resonance Behaviour of Shielded High-Voltage Laboratories 1363

Fig. 9. Interference 2 isolated from measured impulse signal 2 in Fehler! Verweisquelle

konnte nicht gefunden werden.

Fig. 10. Frequency domain of the interference 2 in Fehler! Verweisquelle konnte nicht
gefunden werden.

The following Table 2 shall give an overview on the resulting parameters of the
superimposed signals. The percentage values printed in the lines 3 to 6 of Table 2
represent the deviation from the evaluated results of the standard curve LI-A1 in line 2.
The added interference in line 3 is extracted from a measurement with a slightly faster
T1 Parameter (see Fig. 4), meaning the bigger amplitudes of the interferences are
shifted more towards the beginning of the impulse. The examined Signal for this line is
shown in Fig. 7. The superimposed interference in line 4 origins from a measured
impulse with a longer T1 parameter (see Fig. 8) causing the interferences to be slightly
delayed. The examined signal for this line is shown in Fig. 11. Especially this signal
shows a severe deviation. Lines 5 and 6 contain the result for the same configuration as
in line 4 but with the interference signal multiplied with factors 0, 5 and 0, 1 before
superposition to the base curve LI-A1. Lines 4 to 6 show an approximate linear
coherence in terms of deviation.
1364 T. C. Schlüterbusch et al.

Fig. 11. LI-A1 (Fehler! Verweisquelle konnte nicht gefunden werden.) with superimposed
interference 2 (Fehler! Verweisquelle konnte nicht gefunden werden.)

Table 2. Evaluated parameters of the investigated impulses

Up [kV] T1 [µs] T2 [µs] b’[%]
1: LI-A1 ref. values 1049.6 0.84 60.16 0
2: LI-A1 software results 1049.6 0.83986 60.158 0.01
3: LI-A1 w. interference 1 (Fehler! Verweisquelle 1061.1 0.84934 58.824 4.47
konnte nicht gefunden werden.) (1.09%) (1.13%) (−2.22%)
4: LI-A1 w. interference 2 (Fehler! Verweisquelle 1057.1 0.90028 59.236 1.59
konnte nicht gefunden werden.) (0.71%) (7.19%) (−1.53%)
5: LI-A1 w. interference 2 * 50% 1053.4 0.86636 59.697 0.69
(0.36%) (3.16%) (−0.77%)
6: LI-A1 w. interference 2 * 10% 1050.4 0.84370 60.062 0.12
(0.07%) (0.46%) (−0.16%)

5 Countermeasures and Improvements

During the examination of the problem we found 2 possible, yet simple and practical
methods to reduce or suppress the interferences to a negligible amount.

5.1 Application of Ferrite Beads on the Measurement Cable

Ferrite beads are a widely used solution to so-called common mode interferences.
These interferences induce an electrical current which is flowing on both conductors of
a cable in the same direction (e.g. [4]). The cable is in this case a coaxial cable. The
induced common mode interference cannot be conducted directly to ground potential,
because many digitizers for the purpose of impulse measurements in the high voltage
metrology are galvanic insulated. The market offers ferrite beads in all necessary sizes
and also as a clip-on mounting type for supplementary installation on cables. We
applied various numbers of ferrite beads on the measurement cable, leading to our
 Resonance Behaviour of Shielded High-Voltage Laboratories 1365

impulse voltage divider. A smaller number of ferrite beads already showed sufficient
results. When changing the amount of ferrite beads we observed that any number
higher than 5 does not bring any additional visible benefit. On the other hand, we did
not detect any negative influence of this countermeasure in general or the amount of
ferrite beads in particular.

5.2 The Use of an Appropriate Combination of Digitizer Input Range

and Divider Ratios
When measuring high voltages, the accuracy can be increased by reducing the Signal-
to-Noise-Ratio (SNR). In this use-case we can reduce the amount of interferences by
choosing an adapted setup according to the approximately expected voltage. To
achieve this adjustment there are 3 parameters concerning the components of a typical
measurement system that can be varied with more or less effort. An essential measuring
system for high-voltage impulse measurements typically consists of a high-voltage
impulse voltage divider, a dedicated cable, a signal attenuator and a digital recorder or
digitizer [5].
• Input range of the digitizer: An input signal with a higher level is less vulnerable to
induced interferences.
• Ratio of the attenuator on the digitizer input: In practical terms we recommend
several attenuators with different, fixed ratios to match the ratio of the high voltage
divider and the input range of the digitizer. The digitizers input range shall be used
to at least 50% of its maximum.
• Ratio of the high voltage Impulse divider: This ratio is often fixed and a second
divider is not economic for most laboratories but would also serve this
countermeasure.
The aim is to have a possibly high voltage level on the connection lines between the
components of the measurement system. This way the induced interferences are smaller
respectively to the signal in interest.

6 Conclusion

Our Investigations are leading to the following conclusions

• The observed oscillations in the MHz-range do have an influence on the evaluated
Parameters of an impulse measurement. Up to several percent of deviation are
possible
• The Levenberg-Marquard-Method does not seem to be robust against all interfering
signals of this kind.
• The complete phenomena seem to affect the accuracy more, if the intensity of the
oscillations have a maximum during the rise time of the impulse.
• Not only the Signal-to-Noise-Ratio (SNR) but also the delay and the configuration
of frequency components of the interference seems to have an impact on the caused
deviations (see Table 2, line 3 vs. 4)
1366 T. C. Schlüterbusch et al.

In our laboratory we apply the two described countermeasures with sufficient

success so that we can assure the quality of measurements and accuracy on an adequate
level for a national metrology institute. Especially the application of ferrite beads seems
to be an easy and effective way to enhance the quality of the measurement.

Acknowledgements. The authors would like to gratefully thank the German Federal Ministry of
Economic Affairs and Energy for the financial support of the project “Universal Power Switch”
FKZ: 0324017C.

References
1. Pozar, D.M.: Microwave Engineering, 4th edn. Wiley, Hoboken (2012)
2. IEC 60060-1:2010: High-voltage test techniques–part 1 general definitions and test
requirements (2010)
3. IEC 61083-2:2010: Instruments and software used for measurements in high-voltage and
high-current tests–part 2 requirements for software for tests with impulse voltages and
currents (2012)
4. Franz, J.: EMV–Störungssicherer Aufbau elektronischer Schaltungen. Springer Fachmedien,
Wiesbaden (2013)
5. Schon, K.: High Impulse Voltage and Current Measurement Techniques. Springer
International Publishing Switzerland, Basel (2013)
 On the Use of UHF Sensors
in the Detection and Characterization
of Pulsed Plasma Discharges

Luis Orellana1 , Jorge Ardila1(B) , Gonzalo Avaria2 , Benjamı́n Cevallos1 ,

Cristian Pavez2 , Roger Schurch1 , and Leopoldo Soto2
1
Universidad Técnica Federico Santa Marı́a, 8940000 Santiago, Chile
luis.orellanag@alumnos.usm.cl, jorge.ardila@usm.cl
2
Chilean Nuclear Energy Commission, Casilla 188-D, Santiago, Chile

Abstract. Pulsed plasma discharges are of interest for their capability

to produce pulsed high energy radiation and particle beams. One of them
is the dense plasma focus, a device that has renewed interest for physics
research and possible industrial applications. In this work, the evaluation
of UHF antennas is carried out as an alternative remote diagnostic of this
device. The measurements were taken from a low energy dense plasma
focus device PF-400J operating at 26 kV of charging voltage and using
hydrogen at a pressure of 9.5 mbar. The device operation was simultane-
ously measured with an inductive sensor, an inductive sensor loop, and
an antenna. Two antennas were used for evaluating which one of them
could be used as a complementary diagnostic of the device operation: a
monopole and a Vivaldi antenna. Both antennas were tuned in the UHF
range and were previously designed for measuring partial discharge phe-
nomena. The characterization of the device operation with the antennas
was carried out comparing the time domain signals of the antennas and
the inductive sensor, whose signals features had been known beforehand.
Then, for quantifying both types of measurements, signal parameters of
the inductive sensor and the antennas were correlated. Peak value and
signal energy were determined from the antennas signal and they were
correlated with the dip value feature of the inductive sensor which is an
important indicator of the plasma dynamic of the discharge. The results
indicate that antennas can be used for the detection of abrupt changes
in the circuit current of the device. In particular, the plasma maximum
compression, known as pinch, could be inferred from the antenna signal
as a transient that started when the dip feature was detected with an
inductive measurement. The correlation between antenna signals param-
eter and the dip value of the inductive sensor showed high data disper-
sion, although an increasing trend of the antennas signal parameter with
dip values was found. This work, as an initial experimental survey, con-
cluded that UHF measurement techniques can be applied to the remote
diagnostic of pulsed plasma discharges. The understanding of the emit-
ting mechanism of the device and proper design of UHF sensors will be
considered in future work.

Keywords: UHF measurement · Vivaldi antenna · Plasma focus

c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1367–1376, 2020.
https://doi.org/10.1007/978-3-030-31676-1_128
1368 L. Orellana et al.

1 Introduction
In recent years, the dense plasma focus (DPF) [1] devices have been widely used
in various scientific investigations for possible industrial applications such as
[2,3]: pulsed source of X-ray or neutrons, X ray imaging, to test fusion reactor
wall materials, space propulsion, high energy radiation therapy, etc. DPF corre-
sponds to a pulsed plasma discharge based on the Z-pinch phenomena [4], i.e.
a plasma compression due to its own circulating current. The plasma dynam-
ics in DPF devices takes place inside a vacuum chamber which is driven by an
electrical circuit [1–3]. A set of coaxial electrodes is placed inside the vacuum
chamber, where the central electrode corresponds to the anode. The anode is
partially covered with an insulator where the gas breakdown initially occurs and
the plasma sheath is created. For the DPF operation, first the air inside the
chamber is extracted and then, the chamber is filled with a working gas. Hydro-
gen or other gases, such as deuterium or heavier ones, are put at a low pressure
in the order of 1–10 mbar. A capacitor bank is charged at tens of kV with a
high voltage DC source, depending on the device size, and then the discharge is
initiated via a spark gap. Different energy/size DPF can be constructed follow-
ing scaling laws [5]. Electrically, the DPF operation resembles the behaviour of
an underdamped RLC circuit [1–3]. In general, six stages of the discharge can
be identified [2,3]: (I) initial gas breakdown; this process allows the formation
of a plasma sheath on the surface of the insulator, (II) axial rundown; due to
Lorentz force produced by the circuit current, the plasma sheath moves upwards
toward the end of the anode, (III) radial rundown or compression; on top of
the anode the Lorentz force compresses the plasma sheath into a column, (IV)
pinch; the moment of maximum compression of the plasma column, (V) axial
plasma shock; after pinch disruption due to instabilities and (VI) plasma jet,
after the axial plasma shock advances away from the anode, plasma jets are
ejected. At the time of pinch, electron and positive ion beams (also neutrons
if deuterium is used) are produced along with electromagnetic radiation in a
broad spectrum, from radio frequencies to X rays [1–3]. The electrical diagnostic
of the discharge is commonly made using inductive sensors, like Rogowski coils,
and voltage dividers at the discharge electrodes [1–3]. Related to the physical
phenomena of the discharge, high frequency components in the electrical signals
can be detected. These frequencies are higher than the main discharge frequency
determined by the capacitor bank and the equivalent inductance of the device
[6]. The high frequencies can be explained in terms of the electrical parameters
of the DPF [6,7]. It was suggested that some of the high frequencies are excited
when the circuit current undertakes abrupt changes, like in the initial breakdown
and pinch stages [7]. For the DPF, these abrupt current changes produce pulsed
radio frequency bursts that can be measured outside the chamber with antennas.
Previously, this electromagnetic burst had been identified and had been found
to be related to the discharge process [8,9]. Thus, complementary information
about the discharge can be measured at distance using antennas. This works eval-
uates the use of ultra-high frequency (UHF) tuned antennas as a method for the
remote diagnostic of the operation of these pulsed power devices. Simultaneous
 UHF Sensors for Plasma Focus Discharges 1369

measurements of the DPF operation were carried out with an inductive sensor,
which served as a reference measurement, and an antenna. Two different anten-
nas tuned in the UHF range were considered for this experiment: a monopole
antenna and a Vivaldi antenna. These antennas were evaluated in their ability
to detect the pinch of the DPF operation. For quantifying the antenna perfor-
mance in terms of the inductive sensor, correlations between signal parameters
from the antenna and the dip value from the inductive sensor were made.

2 Methodology

An example of the electrical diagnostic is shown in Fig. 1, where dotted lines show
the associated stages of the discharge. One of the main features of this diagnostic
is the detection of the pinch stage. This is seen as a dip in the inductive sensor,
see the dotted line for the stage IV in Fig. 1a, and as a voltage surge in the voltage
divider signal, in Fig. 1b. The evaluation of the antennas as a complementary
diagnostic to the one shown in Fig. 1, was carried out comparing the antenna
signal and the inductive sensor measurement, evaluating the signal parameter
correlations.

Fig. 1. Electrical diagnostic of a Fig. 2. Sensors used in the measure-

DPF discharge: (a) Inductive sensor, ments (from left to right): inductive
Rogowski coil, and (b) voltage divider. loop sensor (ILS), monopole antenna
and Vivaldi antenna.

2.1 Sensor Characteristics

The sensors used in this work are shown in Fig. 2: an inductive loop sensor (ILS)
[10], a monopole antenna [11] and a Vivaldi antenna [12]. For the antennas,
their S11 parameter, or reflection loss coefficient, was measured with a vecto-
rial network analyzer (VNA Master MS2035B). This parameter is an important
characteristic of the efficiency of the antennas because it indicates the frequency
for which the antennas are tuned. The details of each sensors are as follows:
1370 L. Orellana et al.

1. Inductive sensor loop (ILS): This inductive sensor corresponds to one rectan-
gular loop made in a circuit plate, as shown in Fig. 2. This sensor measures
signals similarly as a Rogowski coil. It was developed for measuring fast vary-
ing current pulses from partial discharges and its constructional details are
found in [10].
2. Monopole antenna (22 cm): A copper wire with 1.6 mm diameter was attached
to a BNC straight bulkhead socket (50 Ω). The length from the top of the
conductor to the socket was 22 cm. Additionally, a 1.6 mm of thickness and
10 cm of diameter aluminium circular plate was placed in the socket as a
ground plane. The blue colour line in Fig. 3 shows the measured S11 param-
eter of this monopole antenna. This type of antenna has an omnidirectional
radiation pattern with respect to the conductor axis. The first three reso-
nances were located at 410 MHz, 1 GHz and 1.6 GHz. This antenna, one of
the simplest that can be designed and built, was previously used for detecting
radiofrequency bursts from partial discharges [11].
3. Vivaldi antenna: This antenna corresponds to an exponential slot line (waveg-
uide) embedded in a substrate, where a micro-strip transmission line is used
to obtain the signal out to a SMA connector [12], see Fig. 2 for its shape.
The Vivaldi antenna was made from a circuit board with dimensions 8.5 cm
× 11.3 cm. In contrast to the monopole antenna, the Vivaldi antenna has a
directional radiation pattern in the direction of the slot line, so the influ-
ence of reflected waves or other sources are minimized. The green colour line
in Fig. 3 shows the measured S11 parameter. The Vivaldi had a resonance
around 500 MHz and a wideband adaptation in the range of 1.25–2.2 GHz.
This antenna has been also used for measuring partial discharges in insulation
systems [12].

Fig. 3. Measured S11 parameter for the antennas.

2.2 Experimental Setup

The DPF device under study was a low energy plasma focus PF-400J [13]. The
parameters of this DPF are: 880 nF capacitor bank, 38 nH circuit inductance, 26
 UHF Sensors for Plasma Focus Discharges 1371

kV charging voltage, ≈ 297 J of electrostatic energy and approximately 330 ns to

maximum current. The operational conditions for this experiment were set at 26
kV for charging voltage and the vacuum chamber was filled with pure hydrogen
at a pressure of 9.5 mbar. The layout indicating the position of the sensors with
respect to the DPF is shown in Fig. 4. The best alternative for placing the ILS
was at 2 mm away from one of the return leads that connects the capacitor bank.
A fixed distance of 40 cm was chosen for the antennas. The acquisition system
used for the measurements was a PXI system which consisted in: NI-PXIe 1082
chassis, NI-PXI 5162 acquisition card and a NI-PXIe-8115 controller with a
dual core i5–2510E processor. The measurement characteristics of the system
are: 2 channels, sampling frequency of 6.25 GS/s, 3 GHz of bandwidth and 8 bit
of vertical resolution. Attenuators of 20 dB and 30 dB with bandwidth of 0–18
GHz and 0–6 GHz, respectively, were used for the safety of the acquisition system
which allowed a maximum of 1 Vpp for the 50 Ω channels. As indicated by Fig. 4,
the ILS was connected to one channel, meanwhile an antenna was connected to
the other. The measurement campaign started with the simultaneous recording
of the ILS and monopole and then repeated for the ILS and Vivaldi. Recall that
the ILS is the reference measurement to the antenna signals. Each signal was
recorded using a time window of 1 µs. A trigger unit was used to control the
discharge of the spark gap which was filled with nitrogen at a pressure of 0.5
bar. The automatic control of the discharge allowed the operation of the DPF
on regular time intervals. A total of 200 discharges, with a 14 s interval between
each one, were recorded for each ILS-antenna data set.

Fig. 4. Measurement layout. The Fig. 5. Example of the ILS measure-

returning leads which serve as refer- ment of the DPF operation.
ence are highlighted in red.

3 Measurement Results and Data Processing

An example of the ILS signal from the measurement results is shown in Fig. 5. As
discussed above, one of the discharge features identified in the ILS is the pinch
seen as a dip in its signal. Due to the plasma dynamic, specifically its compression
phase until pinch, the dip can be interpreted as a rise of impedance of the plasma
[1–3]. The dip value, see Fig. 5, can serve as an indicator of the quality of the
1372 L. Orellana et al.

discharge, because it is related to the induced voltage that accelerate charge

carriers during pinch [3], and thus, is responsible of the radiation and particle
beam emissions. Dip value was the signal parameter used in the correlations
with antenna signal parameters because it can be used to determine whether
the pinch is detected by the antennas or not.

3.1 Waveform Analysis

The measured signals from the ILS-antenna data set are shown in Fig. 6 for the
ILS-monopole comparison, and in Fig. 7 for the ILS-Vivaldi. The initial spark
gap conduction and gas breakdown inside the chamber are highlighted in blue
colour. The abrupt rise of the circuit di/dt can be related to the gas break-
down, where the current starts to flow from the anode to cathode in the newly
formed plasma sheath along the surface of the insulator. Both antennas detected
a transient at this time. The combined effect of the switching action from the
spark gap and gas breakdown can be associated to the excitation of the high
frequencies seen in the ILS signal and the first electromagnetic burst detected
with the antennas. The time of pinch is highlighted in red in Figs. 6 and 7. A
new transient was detected by the antennas at this time. High frequency oscil-
lations were also excited again, as seen in the inductive sensor signal. For the
antennas, the oscillations were found to be highly intense slightly after pinch.
Although a dip feature cannot be seen in the antennas, as it is a unique feature
of the inductive measurement, the pinch can be detected indirectly by looking
at these high frequencies oscillations in the antennas. The transients associated

Fig. 6. Example of the waveforms mea- Fig. 7. Example of the waveforms mea-
sured from the DPF operation. (Top) ILS sured from the DPF operation. (Top)
signal and (bottom) monopole signal. ILS signal and (bottom) Vivaldi signal.
 UHF Sensors for Plasma Focus Discharges 1373

with pinch were seen in all discharges where also the ILS signal detected a dip.
The characterization of the DPF operation with antennas showed that the stages
that involves abrupt current changes, such as the initial spark gap/gas break-
down and pinch, can be detected. Especially, the latter can be associated with
the plasma dynamic, i.e. pinch phenomena. Different transient intensity from
the antennas signals due to the pinch were obtained. In the next subsection, it is
further investigated if this transient was somehow correlated with the dip values
of the ILS signal. Note that different dip values can also be obtained using the
same experiment conditions due to the plasma dynamics [3].

3.2 Signal Parameter Correlation

Due to the importance of the pinch in the DPF discharge process, only the tran-
sient that can be associated with the pinch was considered for the correlations.
This transient corresponds from the time of pinch (red line in Figs. 6 and 7)
until the end of the recording time. The rest of the signal is despised since it
only corresponds to the initial phenomena. Two parameters of the antenna sig-
nal, peak value Vp and energy En , were analyzed in the correlations. These two
parameters were proposed to observe see if one type of correlation yielded better
results than the other. Peak value was selected from the highest oscillation of the
transient in absolute value, meanwhile signal energy was calculated as indicated
by Eq. 1:
n
En = vi2 (1)
i=1

where En is the energy of a signal containing n samples and vi is the sample

value. Note that signal energy uses information of the whole transient related
to pinch, in contrast to the peak value which is achieved in one instant during
the transient. The results of the peak value correlations are shown in Fig. 8, and
those with signal energy in Fig. 9. No clear correlations were found using scatter
plots, i.e. parameter to parameter comparison, but instead clearer results were
obtained using boxplot. Dip values were grouped, and the antenna parameter
were distributed in those dip value ranges. Median values of each distribution
are marked with a circle with a dot in its centre. Only the median value of the
antenna parameter distribution was found to follow an increasing trend with
dip values. Polynomial fits were adjusted for these trends and visually they
showed good agreement with the median values. Note that with the monopole
antenna the median values followed a linear increase whereas with the Vivaldi
they followed a cubic polynomial. Either using peak value or signal energy, only
the median value trend was identified.
1374 L. Orellana et al.

Linear fit for median values Linear fit for median values

0.3
3.5

0.25 3

Monopole antenna
Signal Energy (a.u.)
Peak value (a.u.)

2.5
Monopole antenna

0.2
2
0.15
1.5

0.1 1

0.5
0.05

[0.09-0.10]

[0.12-0.13]

[0.13-0.14]

[0.14-0.15]

[0.15-0.16]
[0.10-0.11]

[0.11-0.12]
[0.09-0.10]

[0.12-0.13]

[0.13-0.14]

[0.14-0.15]

[0.15-0.16]
[0.10-0.11]

[0.11-0.12]

Dip value (a.u.) Dip value (a.u.)

Cubic fit for median values Cubic fit for median values
1.2
0.14
1
0.12

Signal Energy (a.u.)

Vivaldi antenna
Peak value (a.u.)
Vivaldi antenna

0.8
0.1

0.08 0.6

0.06 0.4

0.04
0.2
0.02
0

[0.12-0.13]

[0.13-0.14]

[0.14-0.15]

[0.15-0.16]
[0.10-0.11]

[0.11-0.12]
[0.12-0.13]

[0.13-0.14]

[0.14-0.15]

[0.15-0.16]
[0.10-0.11]

[0.11-0.12]

Dip value (a.u.) Dip value (a.u.)

Fig. 8. Correlation results between dip Fig. 9. Correlation results between dip
value of the ILS and peak value of the value of the ILS and signal energy of
antennas. The median value trend is the antennas. The median value trend
highlighted in red. is highlighted in red.

4 Discussion and Conclusions

In this work, the characterization of a dense plasma focus discharge using UHF
antennas was presented. Two antennas were tested for assessing which one could
be used as a remote diagnostic of the discharge. First, the characterization of
the discharge was carried out in terms of the simultaneous measurement with
an inductive sensor whose features were previously known. Two transients were
detected with the antennas, and they can be associated with abrupt current
changes. The phenomena that can be associated with the transients were the
initial spark gap conduction/gas breakdown and the maximum compression,
the pinch, of the plasma generated by the discharge. The former was due to a
switching action, either the initial spark gap conduction or the gas breakdown,
whereas the latter is related to the plasma dynamic. The pinch is of paramount
importance for this type of pulsed plasma discharge, so a remote measurement of
this phenomena with antennas could provide complementary information about
the discharge. The transient measured due to pinch it is believed to be according
to previous predictions about the high frequency oscillations related to the dense
plasma focus operation [7]. The pulsed electromagnetic burst emitted from the
 UHF Sensors for Plasma Focus Discharges 1375

high frequency oscillations during the discharge were correlated to the dip value
in the signal measured by an inductive sensor which is used commonly used as
a pinch detector. For the antennas, only the transient associated with the pinch
was considered. Peak value and signal energy of the antennas signals showed
an increasing trend with the dip value, although no clear correlation, parameter
to parameter, was found. An increasing trend with the dip value was expected
based on [3]. The median values of the antennas parameters clearly exhibited this
increasing trend, but the rest of the data showed high dispersion. For instance,
note that some high peak values are not necessarily related to high dip values. It
was not clear which antenna obtained better correlation results. In both cases,
only the trend was identified and the median values were successfully adjusted
with a polynomial fit. The same statement goes for the use of peak values or
signal energy correlations. The detection of transients related to the pinch and
the median value trend found, encourages further investigation of the application
of UHF sensors for these devices. More work is required studying the behaviour
of the dense plasma focus as an emitting antenna, the influence of distance in
the correlation and thus the design of proper antennas for this particular device.
In conclusion, the dense plasma focus operation can be remotely characterized
with antennas, and more importantly, the pinch can be detected. The correlation
of signal parameters of the electromagnetic burst measured with antennas and
the inductive measurement, showed a particular increasing trend, but with high
data dispersion. The relationship between the inductive measurement and the
radiated one needs further investigation for remote diagnostic of plasma focus
devices.

Acknowledgments. This work was supported by Fondecyt Iniciación 11160115 and

“Dirección de Postgrados y Programas” of Federico Santa Maria Technical University
under PIIC grant.

References
1. Mather, J.: In: Methods in Experimental Physics, vol. 9, pp. 187–249. Elsevier
(1971)
2. Soto, L.: Plasma Phys. Controlled Fusion 47(5A), A361 (2005)
3. Bernard, A., Bruzzone, H., Choi, P., Chuaqui, H., Gribkov, V., Herrera, J., Hirano,
K., Krejci, A., Lee, S., Luo, C., et al.: J.Moscow Phys. Soc. 8, 93 (1998)
4. Haines, M.: Plasma Phys. Controlled Fusion 53(9), 093001 (2011)
5. Soto, L., Pavez, C., Tarifeno, A., Moreno, J., Veloso, F.: Plasma Sources Sci. Tech-
nol. 19(5), 055017 (2010)
6. Bruzzone, H., Acuña, H., Barbaglia, M., Milanese, M., Clausse, A.: J. Fusion
Energy 36(2–3), 87 (2017)
7. Bruzzone, H., Kelly, H., Moreno, C.: IEEE Trans. Plasma Sci. 18(4), 689 (1990)
8. Gerdin, G., Tanis, M., Venneri, F.: Plasma Phys. Controlled Fusion 28(3), 527
(1986)
9. Escalona, I., Avaria, G., Dı́az, M., Ardila-Rey, J., Moreno, J., Pavez, C., Soto, L.:
Energies 10(9), 1415 (2017)
1376 L. Orellana et al.

10. Ardila-Rey, J., Barrueto, A., Zerene, A., Castro, B., Ulson, J., Mas’ud, A., Valdivia,
P.: Sensors 18(7), 2324 (2018)
11. Robles, G., Martı́nez-Tarifa, J., Rojas-Moreno, M., Albarracı́n, R., Ardila-Rey, J.:
2012 IEEE International Instrumentation and Measurement Technology Confer-
ence Proceedings, pp. 1496–1499. IEEE (2012)
12. Robles, G., Albarracı́n, R., Vázquez-Roy, J., Rajo-Iglesias, E., Martı́nez-Tarifa, J.,
Rojas-Moreno, M., Sánchez-Fernández, M., Ardila-Rey, J.: In: 2013 IEEE Interna-
tional Conference on Solid Dielectrics (ICSD), pp. 302–305. IEEE (2013)
13. Silva, P., Moreno, J., Soto, L., Birstein, L., Mayer, R.E., Kies, W.: Appl. Phys.
Lett. 83(16), 3269 (2003)
 Lightning Fault Rate of Power Distribution
Line in Wind Farm in Winter Lightning Area

Koji Michishita(&) and Shigeru Yokoyama

Shizuoka University, Hamamatsu, Japan

michishita.koji@shizuoka.ac.jp

Abstract. In Japan, the total number of the faults on a 6.6 kV power distri-
bution line has decreased due to installation of ZnO (zinc oxide) surge arresters
at the interval of less than 200 m. However, the proportion of faults caused by
lightning still keeps a high value, 10–15%. The damage of arresters occurs
frequently at the coast of the Sea of Japan where thunderstorms in winter are
notable. It is shown that the surge arresters are damaged when the absorption
energy exceeds the rated value. One of the characteristics of winter lightning is
high energy resulting in the arrester damage. In this paper, the authors consider
damages of surge arrester on a distribution line connected to the wind turbine
through the EMTP (Electro-Magnetic Transient Program). The authors show the
probability of surge arrester damage by the lightning flash to the turbine and
discuss the lightning protection design against surge arrester damage.

Keywords: Lightning
Distribution line
Fault rate

1 Introduction

In Japan, the total number of the faults on a 6.6 kV power distribution line has
decreased due to installation of ZnO (zinc oxide) surge arresters at the interval of less
than 200 m. However, the proportion of faults caused by lightning still keeps a high
value 10–15%. The damage of arresters occurs frequently at the coast of the Sea of
Japan where thunderstorms in winter are notable. It is shown that the surge arresters are
damaged when the absorption energy exceeds the rated value. One of the character-
istics of winter lightning is high energy resulting in the arrester damage. Therefore, the
lightning flash with the high energy draws much attention in Japan recently. The winter
lightning has another characteristic that the flash is initiated by upward propagating
leaders [1] and the initial continuous current (ICC) flows sometimes with pulses.
The ICC has low amplitude but large charge transfer [2].
The authors have carried out the measurement of lightning current at the instru-
mented wind turbines at two locations at the winter lightning area on the coast of the
Sea of Japan [3]. It is confirmed that the charge transfer associated with the ICC
initiated by upward propagating leaders is high compared with that of the return strokes
initiated by downward propagating leaders [4] observed at the summer lightning area.
Due to the high energy and/or the high current, severe damage of the distribution
line has been reported. In this paper, the authors consider damages of surge arrester on
a distribution line connected to the wind turbine through the EMTP (Electro-Magnetic

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1377–1389, 2020.
https://doi.org/10.1007/978-3-030-31676-1_129
1378 K. Michishita and S. Yokoyama

Transient Program) [5]. The authors show the probability of surge arrester damage due
to lightning discharge to the wind turbine and discuss the lightning protection design
against surge arrester damage.

2 Method of Analysis

2.1 Model Line

Figure 1 shows an example of a line subject to analysis and Table 1 shows the
parameters used in the analysis. A wind turbine is connected with one end of a dis-
tribution line through a service line and a pole-mounted transformer. A 960 m-long
model line is composed of horizontally arranged three phase wires and an OHGW
(Overhead ground wire). Another end of a line is terminated with a resistor network so
that no reflection of travelling waves can occur. Horizontal lines are modeled by the
J. Marti set up program in the EMTP [6].

Fig. 1. Configuration of line subject to analysis.

The surge arrester used for analysis is ZnO (zinc oxide) surge arrester with a gap
and the element has the V-I characteristics shown in Fig. 2. In the analysis, the pole-
mounted transformer isn’t modeled and only surge arresters connected in its vicinity
are modeled. This is because most of the lightning current flows into the ground
through the common ground at the pole, the OHGW and phase conductors through
surge arresters [7].
 Lightning Fault Rate of Power Distribution Line in Wind Farm 1379

Table 1. Parameter used in analysis.

Medium-voltage phase wires Line type ACSR-OC of 32 mm2
Sparkover voltage 120 kV
OHGW Line type Hard-drawn copper stranded
conductor of 22 mm2
Grounding interval 40 m
Grounding resistance 60 X
Surge arresters with gap Grounding interval 80 m
Grounding resistance 30 X
Concrete pole Surge impedance 300 X
Propagation speed 3  108 m/s
Span length 40 m
Wind turbine Surge impedance 200 X
Propagation speed 3  108 m/s
Length 100 m
Ground resistivity 100 X m
Lightning path impedance 1 kX

Fig. 2. V-I characteristics of surge arrester.

An OC (Outdoor Cross-linked polyethylene insulated wire) of 22 mm2 is modeled

as an overhead service line for simplicity and it is confirmed that the calculated result is
little influenced by use of the simplified model. The SPDs (Surge Protective Device)
are usually installed for the protection of the electrical equipment in the turbine,
however, they are not taken into account because of their low absorption energy. The
grounding resistance of the wind turbine is set 10 X. The transient current and voltage
is calculated by the EMTP.
Figure 3 shows the configuration of the OHGWs. The A-OHGWs are additional
OHGWs. The E-OHGW (extended overhead ground wire) [8] is a wire, installed from
the concrete pole from one end of the line and connected to the OHGW, extending to
the direction different from the distribution line. Grounding resistance of the E-OHGW
is set 60 X. When the number of the E-OHGWs is two or three, the configuration of the
wires are the same as that of the OHGWs shown in Figs. 3(a) and (b)
1380 K. Michishita and S. Yokoyama

(a) A-OHGW (b) 2 A-OHGWs (c) E-OHGW

Fig. 3. Various OHGW configuration.

2.2 Calculation of Probability of Surge Arrester Damage

The authors estimate the critical peak current for each time-to-half value of lightning
stroke current waveforms resulting in arrester damage through the numerical simulation
by the EMTP. The probability of arrester damage is calculated by (1) [7].
Z (Z )
þ1 þ1  
Pd ¼ f Ip dIp gðTt ÞdTt ð1Þ
0 y ð Tt Þ

Pd : probability of arrester damage

yðTtÞ: minimum current resulting in arrester damage
f Ip : probability density function of peak value of lightning current Ip
gðTt Þ: probability density function of time-to-half value of lightning current
waveform Ttp .
The current waveform is assumed triangular and parameters of lightning current
waveforms are shown in Table 2 [9]. It is assumed that the peak value and the time-to-
half value of lightning current waveforms are in logarithmic normal distribution
functions and are independent of each other.

Table 2. Parameter of winter lightning in [9]

Parameter 50% value 5% value
Peak value [kA] 24 83
Time-to-half value [ms] 0.089 2.21

3 Influence of Sparkover of Insulators

Figure 4 (a) shows a waveform of insulator voltage at the second pole from the end of
the line close to the wind turbine when lightning stroke current with peak of 50 kA
(2/100 ls, time-to-crest value: 2 ls, time-to-half value: 100 ls) hits the turbine and the
 Lightning Fault Rate of Power Distribution Line in Wind Farm 1381

grounding resistance of wind turbine is 10 X. The insulator voltage waveforms at all

phases are almost the same and the insulator voltage is quite lower than the sparkover
voltage, 120 kV, in spite of the high current amplitude of 50 kA. In the analysis,
sparkover does not occur.
In Fig. 4(a), the polarity of the waveform is initially negative and then becomes
positive due to the sparkover of the gap of surge arresters. At about 10 ls, the polarity of
the voltage waveform becomes negative again due to the increase of the current flowing
into the ground from the OHGW. Time-to-crest value of lightning stroke current is 2 ls,
but time-to-crest value of incoming lightning current to the distribution line is much
longer. As time passes, the impedance of the distribution line seen from the lightning
striking point decreases. Consequently, the incoming lightning current becomes large,
therefore, time-to-crest value of the incoming lightning current waveform is long.

Fig. 4. Waveforms of insulator voltage and incoming lightning current

4 Effect of OHGW

The effect of the conventional OHGW, installed above the phase wires, is classified as
follows: (1) shielding of the power line from direct strokes, (2) mitigation of the voltage
of the power line by the electromagnetic coupling and (3) carrying the lightning current
to the remote ground.

4.1 Current Distribution

Figure 5 shows the distribution of the current for the line configuration in Fig. 1 (the
line is consisted of 3 phase wires and an OHGW and no E-OHGW or A-OHEW is
installed) when the lightning strikes the top of the wind turbine. The lightning current
waveform has the peak of 60 kA, the front duration of 2 ls and the total duration of
1 ms. In the followings, the current waveforms under this condition are shown if not
otherwise mentioned. In the initial stage, the current distribution is determined by the
impedance of the distributed constant circuit. After the arrival of the reflected current
waves from the discontinuity points of the impedance, the distribution of the current is
influenced by the reflected waves from the discontinuity points. In the time range of
1382 K. Michishita and S. Yokoyama

less than 2 ls, corresponding to the wave front of the lightning current, most of the
current flows into the tower ground of the wind turbine of 10 X. The current flowing
into the surge arresters gradually increases after the flashover of the arrester gap due to
the potential rise of the pole by the current following into the pole ground. As the
current on the OHGW increases by the reflection of the current waves at the multiple
grounding points and becomes dominant, the arrester current gradually decreases after
about 0.1 ms from the initiation, and is almost zero at about 0.45 ms.

Fig. 5. Current waveforms at various point.

4.2 Installation of A-OHGW

Figure 6 shows the current waveforms through surge arresters and the OHGWs
including the A-OHGW at the pole closest to the wind turbine influenced by the A-
OHGW length.

Fig. 6. Influence of length of A-OHGW on arrester current waveform.

 Lightning Fault Rate of Power Distribution Line in Wind Farm 1383

With the increase of the A-OHGW length, the current through the OHGWs
increases while the current through the arresters decreases. The degree of the decrease
of the current on the surge arresters by expanding A-OHGW from 200 m to 400 m is
less than that by the installation of the 200 m-long A-OHGW to the line without the A-
OHGW.
Figure 7 shows the current waveforms through surge arresters and the OHGWs at
the pole closest to the wind turbine influenced by the number of OHGWs. The length of
the additional OHGW is 200 m. With the increase of the number of the OHGWs, the
current through the OHGWs increases while the current through the surge arresters
decreases. The degree of the decrease of the current on the surge arresters by adding the
A-OHGWs from 1 to 2 is less than that by the installation of the A-OHGW.

Fig. 7. Influence of number of A-OHGWs on arrester current waveform.

4.3 Installation of E-OHGW

Figure 8 shows the current waveforms through surge arresters and the OHGWs
including the E-OHGW at the pole closest to the wind turbine influenced by the length
of the E-OHGW. With the increase of the E-OHGW length, the current through the
OHGWs increases while the current through the surge arresters decreases. The degree
of the decrease of the current on the surge arresters by expanding the length of the E-
OHGW is similar to the case of the OHGWs in Fig. 6.
By installing the E-OHGW or the A-OHGW, the current flowing into the OHGWs
including the E-OHGW increases while the arrester current decreases. In this way the
absorption energy of surge arresters in positive relation with the arrester current
decreases.
Figure 9 shows the current waveforms through the OHGW when the E-OHGW or
the A-OHGW of 200 m in length is installed. Although the total current waveforms for
the line with the E-OHGW and the A-OHGW is similar as in Figs. 6 and 8, the current
on the OHGW with the E-OHGW is significantly high compared with the current on
the OHGW with the A-OHGW. In the case of the A-OHGW is installed, the current on
1384 K. Michishita and S. Yokoyama

the OHGW is almost the same with the current on the A-OHGW. When the E-OHGW
is installed, the current on the E-OHGW is higher than the current on the OHGW in the
time range of 0.32 ls and less, and the current on the OHGW becomes higher after
0.32 ls. Due to the high coupling of the wires in the cases of installation of the A-
OHGS, the decrease of the absorption energy of the surge arresters by the A-OHGW is
a little bit more than that by the E-OHGW.

Fig. 8. Influence of length of second OHGW on arrester current waveform.

Fig. 9. Current waveforms through OHGW, A-OHGW and E-OHGW at pole closest to wind
turbine.

5 Damage Ratio of Arresters

5.1 Threshold Line

Figure 10 shows an example of a result of the threshold line for the line configuration
in Fig. 1.
 Lightning Fault Rate of Power Distribution Line in Wind Farm 1385

Figure 11 shows the influence of the duration of the current waveforms, 1 ms or

10 ms, on the current waveforms flowing through three surge arresters and the OHGWs
including the 200 m-long E-OHGW. The current duration of the OHGW current is
almost the same with that of the assumed lightning current waveform, however, the
duration of the arrester current waveforms is influenced less significantly. This is
because most of the current flows through the OHGW as the time goes. Therefore, the
threshold of the current peak for the time-to-half value of longer than 1 ms in Fig. 10 is
little influenced by the duration of the lightning current waveforms.

1000
Peak value[kA]

Damage side
100

Safe side
10
10 100 1000 10000
Time-to-half value [μs]
Fig. 10. Damage threshold of surge arrester.

Fig. 11. Influence of duration of the lightning current waveform on arrester current waveform.

5.2 Damage Ratio

Effect of 200 m-long A-OHGW or E-OHGW. Figure 12 shows the relative damage
ratio of a surge arrester normalized by the ratio of estimated for the line configuration
1386 K. Michishita and S. Yokoyama

of in Fig. 1 (8.5% of all strokes to the turbine) when the 200 m-long E-OHGW or A-
OHGW is installed.
When the E-OHGW or the A-OHGW is installed at the end of the line far from the
wind turbine, the damage ratio of a surge arrester is little influenced.
When the E-OHGW or an additional OHGW is installed at the end close to the wind
turbine for 200 m, the damage ratio decreases by about 40%. Similarity of the degree
of the decrease of the ratio by the installation of the E-OHGW and the A-OHGW can
be expected by comparison of the arrester current waveforms in Figs. 6 and 8.
Effect of A-OHGW or E-OHGW of 400 m in Total Length. Figure 13 shows the
damage ratio of a surge arrester normalized by the ratio of estimated for the line
configuration of in Fig. 1 (8.5% of all strokes to the turbine) when the E-OHGW and/or
the A-OHGW are installed for the total length of 400 m.

Fig. 12. Relative arrester damage ratio influenced by 200 m-long A- or E-OHGW.

The damage rate of the arresters for the line with the 400 m-long E-OHGW is
almost the same for the line with the 400 m-long A-OHGW or the two 200 m-long A-
OHGWs. This can be expected from the similarity of the arrester current waveforms in
Figs. 7 and 8.
Figure 14 shows the arrester current waveform as well as the OHGW current when
the two 200 m-long E-OHGWs are installed commonly. Because E-OHGW plays a
role of carrying the lightning to the remote ground, the current on the E-OHGWs is
little influenced by doubling the wires on the same pole.
Figure 15 shows the current waveforms through the arresters and the OHGWs when
two 200 m-long E-OHGWs are installed on the separate poles. Different from the cases
of the two 200 m-long E-OHGWs are installed on the same pole, the current on two
200 m-long E-OHGWs increases, therefore, the current through the arresters decreases
significantly. This result in the decease of the damage rate in Fig. 13.
 Lightning Fault Rate of Power Distribution Line in Wind Farm 1387

Fig. 13. Relative arrester damage ratio influenced by A- and/or E-OHGWs of 400 m in total
length

Fig. 14. Current waveforms through surge arresters and OHGWs with two 200 m-long E-
OHGWs installed on same poles.

Figure 16 shows the current waveforms through the arresters and the OHGWs when
the 200 m-long E-OHGW and the 200 m-long A-OHGW are installed. The E-OHGW
current is almost the same as that of the E-OHGW current on the line with only one
200 m-long E-OHGW. The A-OHGW current is lower than that on the line with only
one 200 m-long A-OHGW, however, the total current through the OHGWs increases
compared with other cases in Fig. 13. This result in the lowest damage rate of the surge
arresters in Fig. 13.
1388 K. Michishita and S. Yokoyama

Fig. 15. Current waveforms through surge arresters and OHGWs with two 200 m-long E-
OHGWs installed separate poles.

Fig. 16. Sum of current waveforms flowing through surge arresters at pole closest to wind
turbine influenced by 200 m-long A- and E-OHGWs.

6 Conclusion

The authors conducted a study on damage of the surge arrester by winter lightning
based on EMTP simulation. The following results are obtained.
1. The probability of surge arrester damage due to the incoming lightning current from
the wind turbine is estimated for the line with additional overhead wires, installed
below the phase wires, and/or extended overhead ground wires. Both additional and
extended overhead ground wires are effective to reduce the arrester damage.
 Lightning Fault Rate of Power Distribution Line in Wind Farm 1389

2. When the OHGW is installed for 400 m in total length, the lowest rate is obtained
when the extended OHGW and the additional OHGW of equal length is installed.

References
1. Miyake, K., Suzuki, T., Takashima, M., Takuma, M., Tada, T.: Winter lightning on Japan Sea
coast-lightning striking frequency to tall structures. IEEE Trans. Power Deliv. 5, 1370–1376
(1990)
2. Miki, M., Miki, T., Wada, A., Asakawa, A., Asuka, Y., Honjo, N.: Characteristics of winter
lightning flashes to wind turbines in the coastal area of the Sea of Japan - Observation results
of lightning for wind turbines at Nikaho Kougen Wind Farm from 2005 to 2008. CRIEPI
report, H09005, June 2010. (in Japanese)
3. Furukawa, M., Michishita, K., Honjo, N., Yokoyama, S.: A study of characteristics of winter
lightning from measurements of current waveform on wind turbine. In: 27th Annual
Conference on P&E Society IEEJ (2015). (in Japanese)
4. Berger, K., Anderson, R.B., Kroninger, H.: Parameters of lightning flashes. Electra 41, 23–37
(1976)
5. K.U. Leuven EMTP Center: Alternative transients program (ATP) rule book.
Canadian/American EMTP User Group (1987)
6. Marti, J.R.: Accurate modelling of frequency-dependent transmission line in electromagnetic
transient simulations. IEEE Trans. Power Appar. Syst. PAS-101(1), 147–157 (1982)
7. Nakada, K., Yokota, T., Yokoyama, S., Asakawa, A., Kawabata, T.: Distribution arrester
outages caused by lightning backflow current flowing from customer’s facility into power
distribution lines. IEEJ Trans. Power Energy 117(10), 1382–1388 (1997). (in Japanese)
8. Yokoyama, S., Sugimoto, H., Wada, M., Koide, K., Kosuge, T., Nakada, K., Urata, T.:
Lightning protection of power distribution lines located in mountainous areas. CRIEPI report,
T64 (2001). (in Japanese)
9. Miyake, K., Szuki, T., Shinjou, K.: Characteristics of winter lightning current on Japan Sea
coast. IEEE Trans. Power Deliv. 7(3), 1450–1456 (1992)
 The Criterion for Electromagnetic
Compatibility Evaluation in High
Voltage Substations

Tomaž Živic(&)

Milan Vidmar Electric Power Research Institute, Hajdrihova 2,

1000 Ljubljana, Slovenia
tomaz.zivic@eimv.si

Abstract. Currently, there is no international standardized method for assess-

ing the provision of electromagnetic compatibility in substations. For more than
25 years a test method for measuring transient overvoltages in secondary circuits
of instrument transformers is in use in the Slovenian electric power system. As a
source of disturbances, this method uses the switching of the high voltage
disconnector. Measurements are performed mainly in secondary circuits of
current and voltage instrument transformers. After comparing the peak measured
values to limit values conclusions are made whether the electromagnetic com-
patibility is achieved or not. It has been found that a different analysis procedure
is needed, which is more flexible and provides more information. The solution
was found in statistical approach to the analysis. A retrospective statistical
analysis of oscilloscope recordings from 30 air-insulated high voltage substa-
tions was made. First, absolute values of recordings were calculated. Then they
were tested for normality of distribution and transformed, since the distribution
does not follow the normal one. On transformed data descriptive statistics
(arithmetic mean and standard deviation) were calculated, which, in addition to
the type test value of immunity for secondary equipment and two factors,
comprise a new criterion. Two examples are presented that illustrate the use of
the newly proposed criterion in assessing electromagnetic compatibility.

Keywords: Electromagnetic compatibility
High voltage substation

Transient overvoltage
Statistical analysis

1 Introduction

Electronic equipment in the control, protection and metering process has been present
in substations for a long time. This ancillary or secondary equipment is functionally
and electrically connected with primary, i.e. high voltage equipment. Secondary
equipment needs an appropriate electromagnetic environment for its proper functioning
and must have sufficient immunity to electromagnetic interference. Testing the
immunity of secondary equipment that is built into the substation is well covered. The
generic standard IEC 61000-6-5 [1] and individual product standards, e.g. IEC 60255
(especially Part 26), set the requirements and define the tests. However, there is cur-
rently no standard procedure or standardised limit values for testing the level of

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1390–1400, 2020.
https://doi.org/10.1007/978-3-030-31676-1_130
 The Criterion for Electromagnetic Compatibility Evaluation 1391

disturbance that occurs in the substation [2]. Transient overvoltages are conductive
disturbances occurring at the transition conditions of the substation [3]. The influence
on the transfer of a disturbance from high voltage part of a substation to secondary
circuits and secondary equipment is defined by [4]:
• geometric relationship between high voltage and secondary circuits of the
substation,
• the type and version of high-voltage devices (e.g. inductive voltage instrument
transformers),
• the type of cables (with or without screen) and their paths,
• the type of material for earthing and bonding,
• other mitigation measures (e.g. bonding).
In addition to these influences, the quality of the installation work plays an
important role. And all together forms a system that is unique for each substation. This
uniqueness affects the amplitude and frequency of the transient overvoltages. There-
fore, the following applies for testing in the substation:
• a source of interference is inherent to an individual substation,
• a criterion for assessing electromagnetic compatibility is based on the immunity of
installed secondary equipment.
Such an approach ensures the repeatability of testing – within a particular sub-
station it is a foundation for a defined measurement process and provides information
on electromagnetic environment in the substation and on the effectiveness of the
mitigation measures. On this basis we can conclude on the electromagnetic compati-
bility provision of the particular installation or of a whole substation.

2 Test Method

Measurements of transient overvoltages in secondary circuits as described below are

carried out for more than 25 years in Slovenian substations. The common mode
(asymmetric) transient overvoltages (i.e. damped oscillatory wave) generated in the
secondary circuit are the consequence of a switching with a high-voltage disconnector.
These overvoltages are measured at the terminals of the input port of the secondary
equipment, e.g. in the relay house. Measurement takes place during the switching of the
disconnectors of all three phases, which is part of normal operation. Switching and
measurements take place in the same high voltage bay, in the no-load state (i.e. the
circuit breakers are turned open). This applies to feeder and bus coupler bay (Fig. 1).
The overvoltages are systematically measured in secondary circuits of voltage and
current instrument transformers. These circuits transmit information about the condi-
tion of the high voltage part of the substation to the control, protection and measure-
ment equipment and are therefore important for the reliable operation of the substation.
The power line that is connected to the feeder bay is energised during the
1392 T. Živic

measurements. By closing or opening the feeder disconnectors (Q9), a voltage and

current wave travels to instrument transformers (T1 and T5). A similar occurrence
emerges, at the closing or opening selector disconnectors Q1 or Q2 in bus coupler bay.
Therefore, the corresponding system of busbars (G1 or G2) must be energised.

Fig. 1. Schematic of feeder bay (above) and bus coupler bay (below).

The bay is energised during the switching of the disconnector and is not discharged
after that. The measurements are therefore carried out with a trapped charge, which is
the same as at the automatic reclosing after the fault. For each measurement in a single
circuit, several consecutive measurements are made at the opening and closing of
disconnectors. This type of testing in a functioning substation is mentioned in a Cigré
publication [2] as one of the options for testing the actual state of electromagnetic
compatibility or as a procedure for controlling the effectiveness of the mitigation
measures – according to the implementation of the earthing, the bonding, the instal-
lation of secondary cables, etc. Therefore, such a test requires secondary equipment to
be connected in the circuit.
 The Criterion for Electromagnetic Compatibility Evaluation 1393

3 Criterion
3.1 The Previous One
The previous criterion for assessing the provision of electromagnetic compatibility
based on the measured transient overvoltages in secondary circuits was determined [5]
by the formula:
uTPV
ud  ð1Þ
fu
fn
fs

where:
– ud is the permissible peak value of the measured overvoltage in Volts (V),
– uTPV is the type test value (TPV) for the immunity of the secondary device in Volts
(V),
– fu is the ratio of the maximum operating voltage of the substation and voltage at
which the measurement is performed (=1,1),
– fn is the ratio between the flashover peak voltage and the line peak voltage (=2,9),
– fs is safety factor (=1,5).
At the time of the design of this criterion, transient overvoltage measurements were
intended only at the closing of the disconnector. It is also partly because in the past
there was no suitable measuring equipment that could cover the entire occurrence of
transient overvoltages, but it was possible to measure only the overvoltages at the first
breakthrough between the poles of the disconnector. With modern equipment, the
whole phenomenon can be measured. For the type test value of the damped oscillatory
wave of 2500 V, as set by the standard [1] for signal circuits running from the high
voltage switchyard to the room with secondary equipment, the peak value of the
transient asymmetric overvoltage according to the Eq. (1) shall be ud  522 V. In
practice, there has often been a question of what to do in the case when ud is more than
522 V? It turns out that the criterion is very rigid. Therefore, a different option was
sought. As part of the research for the master thesis [6], the measurement results from
30 air insulated substations of voltage levels of 110 kV, 220 kV and 400 kV were
processed. A retrospective statistical analysis was performed on the stored data of the
measured asymmetric transient overvoltages. Based on this analysis, a new criterion to
verify the electromagnetic compatibility was designed.

3.2 Data Preparation

As part of the processing of the measurement results, recordings for a certain circuit
from the individual measurements in a particular substation were merged and absolute
values were calculated. The data thus obtained represent one sample. The obtained
samples have several thousand elements (overvoltage amplitudes). For each sample,
noise and part of the recording before or after the transient overvoltage are present. The
noise and/or the part of the recording without overvoltages would distort the analysis.
Therefore, in the analysis, only those elements of the sample with amplitude greater
1394 T. Živic

than the noise were taken into account, on a case-by-case basis (Fig. 2). The noise in
the secondary circuit of the voltage transformer is the value of the steady line voltage.
The actual values were up to approximately 96 V because the network typically
operates at slightly higher voltages than the nominal one. Therefore, the noise level was
determined for each sample separately. The situation is different in the circuit of the
current transformer. No current is flowing through the current transformer (because the
circuit breaker in the bay is switched open), but noise is present in the recording due to
other influences. The noise amplitude was between 4 V and 24 V. Also, for these
samples, the noise level was determined for each one separately.

Fig. 2. Noise and overvoltage part of the recording of the voltage transformer secondary circuit.

3.3 Normality Testing and Data Transformation

Testing of the samples showed that asymmetry coefficients (skewness) are mostly
between 1 and 9, which means that the samples have a longer tail on the right (towards
higher values). The coefficients of flattening (kurtosis) have values in most cases
greater than 1 and up to about 33, indicating a very narrow form, and some are
 The Criterion for Electromagnetic Compatibility Evaluation 1395

flattened. Testing has shown that patterns generally do not have a normal distribution.
Since the statistical analysis based on the normal distribution is more powerful than
nonparametric, it is therefore reasonable to transform the samples that their deviation
from the normal distribution is as small as possible. This is accomplished with the
function of the decimal logarithm on the values in the samples:

X ¼ logðabsðY ÞÞ ð2Þ

where:
– Y is the set of overvoltage amplitudes in a sample in Volts (V),
– X is transformed set (no unit).
An example of the normality test for raw and transformed data is in Figs. 3 and 4.
In the case of transformed data, better matching between the data graph (blue circles)
and the normal distribution line (red line) is noticeable. For both graphs, the parameters
were derived from data, where only the values above the noise level were taken into
account (condition: v1 > 16).

Fig. 3. Graph of raw measured i.e. abs(Y) values probability (blue circles) against Normal
probability (red line).

After the sample distribution test was done, a data transformation was made and
descriptive statistics were calculated on transformed data (particularly the arithmetic
mean and standard deviation among the others).
1396 T. Živic

Fig. 4. Graph of transformed i.e. X values probability (blue circles) against Normal probability
(red line).

3.4 The New Criterion

The statistical analysis showed that some samples had a lower average value than
others but a higher variance or standard deviation. Based on this fact, the idea is that the
standard deviation is that part of the criterion, which will give the answer on how to
evaluate the measured transient overvoltages.
The concept of ensured electromagnetic compatibility means that there is a suffi-
ciently large margin of electromagnetic compatibility. The margin of electromagnetic
compatibility is calculated as the ratio between the transmitting level and the level of
immunity. In practice, the ratio of minimum 6 dB [7] has been established for the
electromagnetic compatibility margin, which as a voltage ratio is a factor of 2.
The normal distribution is symmetrical, but when considering the electromagnetic
compatibility assurance, only the positive side of the distribution where there are large
values is interesting, that is, the positive value of the standard deviation. Therefore, in
the statistical approach, the final value is calculated by inverse transformation of the
sum of the arithmetic mean and the multiple of the standard deviation. The result is
compared to compatibility margin:

UTV
10ðx þ ks
sÞ  ð3Þ
km
 The Criterion for Electromagnetic Compatibility Evaluation 1397

where:
– x is the arithmetic mean of transformed values,
– ks is the multiple for standard deviation of transformed values,
– s is the standard deviation of transformed values,
– UTV is type test value in Volts (V),
– km is the factor of electromagnetic compatibility margin (usually is equal 2).
In addition, the criterion may be tightened if the factor km is more than 2. In this
way, more stringent requirements may be considered, for example, for secondary
circuits, which are feeding the equipment which must meet the performance criterion A
requirement [1]. With the factor ks, the influence of the variance of the measurement
results can be optionally taken into account (ks = 2 corresponds to approximately
95,5%, ks = 3 to 99,7%, ks = 3,5 to 99,95% and ks = 4 to 99,99% of all overvoltages).
Factors ks and km shall be determined according to the desired reliability of equipment
operation.

4 Examples

The examples in Figs. 5 and 6 show merged recordings of overvoltages. The boundary
according to the previous criterion (“Limit 522 V”) is also written in the graph,
according to which it would be estimated that the electromagnetic compatibility in the
secondary system is not ensured because the overvoltages exceed this value. However,
the question is if such a measurement result really means that the electromagnetic
compatibility is not ensured, in case the type test value is 2500 V.
It is obvious that the deviation from 522 V in neither case is massive. It follows
from both cases that the assessment cannot be based solely on exceeding the permis-
sible value. A more accurate criterion should be taken into account. The following two
examples illustrate the use of the criterion based on statistics.

4.1 Example 1
The recordings for Example 1 were obtained in a relay house when measuring in the
secondary circuit of the current transformer of the transmission line bay. They were
processed according to the statistical analysis procedure. The results are descriptive
statistics in Table 1.

Table 1. The descriptive statistics of transformed values for the Example 1

Valid N Mean ðxÞ Std. Dev. (s)
4699 1,740510 0,476072
1398 T. Živic

Fig. 5. The statistical and EMC margin level of Example 1.

The value of Eq. (3) at km = 2 and ks = 3 is 1475 V (Fig. 5, “3sigma”). For

UTV = 2500 V the criterion is not fulfilled, because 1475 V > 2500/2 V = 1250 V,
although the maximum overvoltage is 800 V, which is less than 1250 V. In this case,
the estimate is consistent with the rating that would be obtained on the basis of the old
criterion, that electromagnetic compatibility is not guaranteed. The reason is a large
variance, although the arithmetic mean value itself is low (55 V).

4.2 Example 2
The recordings for Example 2 were obtained when measuring in a relay house in the
secondary circuit of the voltage transformer in the transmission line bay. They were
processed according to the statistical analysis procedure. The results are descriptive
statistics in Table 2.

Table 2. Descriptive statistics of transformed values for Example 2

Valid N Mean ðxÞ Std. Dev. (s)
4219 2,202842 0,210494
 The Criterion for Electromagnetic Compatibility Evaluation 1399

Fig. 6. The statistical and EMC margin level of Example 2.

The value of Eq. (3) at km = 2 and ks = 3 is 683 V (Fig. 6, “3sigma”). With

UTV = 2500 V, the criterion is fulfilled because 683 V < 2500/2 V = 1250 V. The
estimate thus obtained is more favourable than the rating it would have been obtained
on the basis of the old criterion, although the highest measured overvoltage was 712 V.
The reason is smaller variance, although the average value (160 V) is greater than in
Example 1. The result is still favourable, even when the factor ks = 4 and the calculated
value is 1109 V, which is less than 1250 V.

5 Conclusions

The new criterion based on a statistical analysis of the overvoltage measurements is an

improvement compared to the old criterion. By taking the standard deviation into
account, it is possible to optionally tighten or temper the criterion, making it adjustable
according to the given specifications of the secondary equipment and according to the
required properties of the electromagnetic environment.
The presented criterion may be, at least, one of the criteria to deal with evaluation
of electromagnetic compatibility in substations. The test procedure and statistical
analysis answer several questions:
• testing is carried out with an actual source of interference (high voltage
disconnector),
• the test is repeatable (within a particular substation),
1400 T. Živic

• the measurement method is defined,

• the result of the analysis provides information on the effectiveness of mitigation
measures and about the electromagnetic compatibility in a particular installation and
therefore provides a basis for evaluation of the EMC provision in a whole
substation,
• the test procedure and the analysis provide a basis for standardisation.
To standardise the measurement method and the analysis process, the following
should apply:
• accept the idea of uniqueness of particular substation, therefore measurement results
and the analysis result are valid only for that particular substation,
• specify the minimum specifications of the measuring equipment (e.g. frequency
bandwidth, sampling rate, memory depth of an oscilloscope, frequency bandwidth
of voltage probes),
• determine the minimum number of switching operations with the disconnector,
• define a multiple of the standard deviation in different electromagnetic
environments,
• define the margin of electromagnetic compatibility for various types of secondary
equipment.

References
1. Electromagnetic Compatibility (EMC) – Part 6-5: Generic standards – Immunity for
equipment used in power station and substation environment. IEC 61000-6-5. Ed.1.0 (2015)
2. EMC Within Power Plants and Substations. Publication 535. CIGRE Working Group C4.208,
pp. 242–245 (2013)
3. Ogorelec, A., Orel, B., Poberaj, S., Verč, E., Žunko, P., Kurnik, A., Podlipnik, J., Bergant, J.,
Lah, P., Gantar, J.: Transient overvoltages in secondary circuits of high-voltage substations -
Phase 1 (Tranzientne prenapetosti v sekundarnih vodih elektroenergetskih postrojev visokih
napetosti – 1. faza). Faculty of Electrical Engineering, Department of Power Systems and
Devices, Ljubljana (1977)
4. Kosec, D.: Overvoltages in wiring of relays protection (Prenaponi u ožičenju relejne zaščtite).
University of Zagreb, Faculty of Electrical Engineering, Zagreb (1978)
5. Ogorelec, A., Orel, B., Popović, M., Slivnik, T., Žunko, P., Turk, I., Kosec, D., Korbar, M.,
Kurnik, A., Kukovič, I., Zakonjšek, J., Gantar, J.: Transient overvoltages in secondary circuits
of high-voltage substations - Phase 3 (Tranzientne prenapetosti v sekundarnih vodih
elektroenergetskih postrojev visokih napetosti – 3. faza). Faculty of Electrical Engineering,
Department of Power Systems and Devices, Ljubljana (1980)
6. Živic, T.: Expected values of transient overvoltages on secondary circuits in electric power
substations at switching manoeuvres with disconnectors (Pričakovane vrednosti prehodnih
prenapetosti v sekundarnih tokokrogih elektroenergetskih postrojev pri stikalnih preklopih z
ločilniki). Master thesis. University of Ljubljana, Faculty of Electrical Engineering, Ljubljana
(2016)
7. MIL-STD-464: Electromagnetic environmental effects requirements for systems, pp. 4–5
(1997)
 Simplified Approach for Investigating
Overvoltages in DC Cables in a –320 kV
Symmetrical Monopolar HVDC System

T. Karmokar1(&) and O. Lennerhag2

1
NKT HV Cables AB, Rombvägen 4, 371 65 Lyckeby, Sweden
tanumay.karmokar@nkt.com
2
Independent Insulation Group, Storgatan 5, 771 30 Ludvika, Sweden
oscar@i2group.se

Abstract. Transient analysis of DC cable systems has gained momentum,

especially due to pressing demands on the availability and reliability of offshore
interconnectors. Such investigations form a vital part of the overall insulation
coordination studies, which are typically performed by utilities and transmission
system operators. Detailed converter models are typically limited to the
respective manufacturers, and from a cable manufactory perspective, the con-
verters are usually considered as a black box component. This makes it chal-
lenging for cable manufacturers to obtain such models for performing studies on
their DC cables, mainly aimed at the converter/cable interface. In this paper,
reduced converter models are utilized for performing DC-side fault studies.
Detailed cable models have been implemented, considering the properties of
every cable layer. The focus is on pole-to-ground faults for a symmetric
monopolar half bridge VSC based modular multilevel system. The simulation
results are analyzed and recreated in the form of a special overvoltage test
performed on a full-scale DC cable system, i.e. a cable, along with joints and
terminations. Based on the presented study results, reduced converter models
can be utilized as an alternative to detailed models within the cable industry. In
this manner, it is possible to achieve a first-hand impression of the profile of
transient stresses that the DC cables may experience. This can provide an
advantage regarding developmental activities, especially in designing special
and non-standard high voltage tests.

Keywords: DC cable systems
Overvoltage
Symmetrical monopolar

1 Introduction

The motivation for increasing the share of renewable electricity generation has caused a
paradigm shift in the power sector to develop smarter and more flexible transmission
grids. The large-scale realization of offshore windfarms, as well as the need to inter-
connect countries around the North Sea has led to an interest in offshore grids [1]. High
Voltage Direct Current (HVDC) transmission has become a key technology in enabling
the transition into a more sustainable power system, e.g. by increasing the share of
renewable electricity generation from remotely located sources, often offshore or near

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1401–1412, 2020.
https://doi.org/10.1007/978-3-030-31676-1_131
1402 T. Karmokar and O. Lennerhag

the sea. Advancements in extruded cross-linked polyethylene (XLPE) HVDC cables

and Voltage Source Converter (VSC) technology have made the realization of cost
effective and state-of-the-art HVDC interconnections practically feasible [2].
Understanding the performance of cable systems after they have been installed in
an HVDC station is of great importance to cable manufacturers. A better understanding
regarding the stresses on DC cables makes it possible to perform tests that can better
reflect service conditions. Stresses on DC cable systems can be assessed using existing
test methods [3], which are incorporated in standard qualification programs. In general,
any test method must sufficiently address critical parts of the cable system such that
they shall reflect the stresses occurring in a real situation.
In order to assess the transient performance of DC cable systems through simu-
lations, suitable models of the converters are a necessity. Detailed converter models,
including the complete characteristics of the converter system, are typically restricted to
manufacturers themselves. From a cable manufactory perspective, the converters are
usually considered as a black box component.
In this paper, simplified converter models are implemented for performing DC-side
fault studies. The primary focus is on pole-to-ground faults for a symmetric monopolar
half-bridge Modular Multi-Level Converter (MMC). Section 2 details a simplified
converter model suitable for DC-side fault studies. Section 3 describes the developed
model used for the study and Sect. 4 presents results from the simulations. Section 5
compares the simulation results with results from an overvoltage test performed on a
full-scale cable system and, finally, Sect. 6 presents the conclusions from the study and
discusses the representative nature of non-standard impulses for overvoltage tests.

2 A Simplified Converter Model for DC-Side Fault Studies

Figure 1 shows the topology of a half-bridge MMC, and the layout of the sub-modules.

Fig. 1. Topology of a half-bridge MMC, including the layout of a sub-module.

 Simplified Approach for Investigating Overvoltages in DC Cables 1403

Depending on the study, there are several approaches to the modelling of MMC-
based HVDC ranging from a full representation, where each sub-module is included in
detail [4], to reduced models, with equivalent circuits representing the sub-modules [5].
All these models have different considerations regarding accuracy, computational
burden and modelling effort required.
In order to select a suitable converter model for studying DC-side faults, it is
important to understand the behaviour of the half-bridge MMC during the fault. DC-
side faults can be divided into three different stages [5]:
i. Discharge of the DC-side capacitors
ii. AC transient infeed
iii. AC steady-state infeed.
During stage (i), the inserted submodules will discharge, which contributes to a
steep increase in the fault current. The AC-side infeed during this stage is limited.
During stage (ii), the IGBTs will be blocked and the current will flow through the free-
wheeling diodes. This stage is characterized by a combination of AC and DC transient
components. Finally, during stage (iii) the DC-side fault current is fed by the AC
system through the diodes. The three stages are illustrated in Fig. 2.

Fig. 2. DC-side fault current stages.

Based on these three stages, [5] proposes a simplified converter model suitable for
DC-side faults which has been verified through simulations by comparing the results
against detailed converter models. The simplified converter model is presented in
Fig. 3.
The main principle of the model can be explained as follows: at the instant of the
fault, SB1 is closed and SB2 is open. This represents stage (i). At the IGBT turn-off
(i.e. blocking) instant, SB1 opens and SB2 is closed. This represents stages (ii) and (iii).
Stages (i) and (ii) are separated by the blocking of IGBTs. Finally, the AC-side
breakers are opened to clear the fault.
1404 T. Karmokar and O. Lennerhag

Fig. 3. The simplified converter model.

3 Modelling and Study Cases

3.1 Overview
The system under investigation is a point-to-point symmetrical monopolar half-bridge
MMC as shown in Fig. 4. The main parameters of the AC-system are given in Table 1,
based on [6]. A model of the system has been implemented in PSCAD™/EMTDC™
version 4.6.3.

Fig. 4. Overview of the model. The right side (not shown) is identical to the left side.

Table 1. AC-side parameters.

Vac Zac R/X Sn uk V1/V2
(kV) (X) (MVA) (pu) (kV)
400 4 0.1 1000 0.18 400/333
 Simplified Approach for Investigating Overvoltages in DC Cables 1405

3.2 HVDC Cable Model

The cables used in the study are 2010 mm2 Aluminum XLPE-insulated DC land
cables. The rated voltage is 320 kV.
For modelling purposes, the cables are buried at a depth of 1.5 m, with a distance of
0.5 m between them. A soil resistivity of 100 Ωm is used.
The cables are divided into 5 km long segments, and the cable screens are con-
nected together and grounded at each joint. At the converters, a grounding resistance of
0.1 Ω is used, and for the other joints a value of 10 Ω is used. The cables are modelled
in PSCAD using frequency-dependent models [7].

3.3 Converter Model

The converters are modelled using the simplified representation described in Sect. 2,
Fig. 3. This model has been validated for an asymmetrical monopolar configuration [5]
but the behavior of the converter should be the same in a symmetrical monopolar
configuration. Table 2 presents the parameters used for the converter model, based on
[5, 6].

Table 2. Converter parameters

Vdc Csm N Ceq,u, Ceq,l Leq Req
(kV) (mF) (µF) (mH) (X)
±320 10 400 50 50 2

3.4 Surge Arresters

Surge arresters are located at the cable terminals. The data used for the surge arrester
characteristics is shown in Fig. 5, based on [6]. A base voltage of 345 kV is used.

Fig. 5. Surge arrester characteristics, adapted from [6].

1406 T. Karmokar and O. Lennerhag

3.5 Study Cases

The following study cases are considered:
i. Pole-to-ground fault at the terminals of converter 1
ii. Cable fault, 100 km from converter 1 (cable midpoint), at a joint
iii. Cable fault, 25 km from converter 1, at a joint.
A fault impedance of 1 Ω is used. A blocking time of 200 µs is used for the
simulations. Depending on the fault location, a delay is added to the blocking time due
to the time it takes for the fault to be visible at the converter terminals. The AC-side
breakers are opened 40 ms after the fault is detected.

4 Simulation Results

4.1 Pole-to-Ground Fault at Converter 1

Figure 6 shows the calculated voltages following a pole-to-ground fault at the terminals
of converter 1.

Fig. 6. Negative (upper) and positive (lower) pole voltages following a pole-to-ground fault at
the terminals of converter 1. Note the difference in scale.
 Simplified Approach for Investigating Overvoltages in DC Cables 1407

Following the fault inception, the negative (non-faulty) pole voltage increases to
1.60 pu with a time-to-peak of 6–7 ms. Oscillations due to reflections can be seen in the
positive (faulty) pole voltage at the cable midpoint and at the terminals of converter 2.

4.2 Cable Fault 100 km from Converter 1

Figure 7 shows the calculated voltages following a cable fault at the midpoint of the
cable.
The negative (non-faulty) pole voltages are similar to those obtained for the pole-
to-ground fault, with a maximum value of 1.63 pu with a time-to-peak of 5–6 ms. For
the positive (faulty) pole voltages, oscillations due to reflections can be seen at con-
verter 1 and 2.

Fig. 7. Negative (upper) and positive (lower) pole voltages following a pole-to-ground fault at
the terminals of converter 1. Note the difference in scale

4.3 Cable Fault 25 km from Converter 1

Figure 8 shows the calculated voltages following a cable fault 25 km from the ter-
minals of converter 1.
1408 T. Karmokar and O. Lennerhag

Fig. 8. Negative (upper) and positive (lower) pole voltages following a cable fault 25 km from
the terminals of converter 1. Note the difference in scale

The negative (non-faulty) pole voltages are similar to the other studied cases, with a
maximum value of 1.59 pu with a time-to-peak of 6–7 ms. For the positive (faulty)
pole, a cable fault close to the converter terminals leads to oscillations with a higher
frequency at the converter terminals, compared to faults that are further away. The
reason for this is that the distance to the fault is shorter, which consequently means that
the travelling time for the reflections is shorter.

5 High Voltage Tests

Existing standard high voltage tests for DC cable systems [3] are aimed at emulating
requirements which are specific to the operating conditions of point-to-point HVDC
systems. Considering the simulation results presented in Sect. 4, as seen in Figs. 6, 7
and 8, the Temporary Overvoltage (TOV) experienced by the non-faulty pole due to a
pole-to-ground fault can be seen as a double exponential varying wave shape. Such a
wave shape can be mathematically represented by a double exponential waveform.
Figure 9 shows a double exponential waveform starting at instant t0, which could
correspond to the fault instant. The maximum peak value achieved is Umax.
 Simplified Approach for Investigating Overvoltages in DC Cables 1409

Fig. 9. Double exponential waveform representing an overvoltage. The figure has been adapted
from [8].

This represents an aperiodic transient voltage in the form of an impulse waveform

as per [9] having an exponential rise to peak ðs1 Þ and a similar exponential fall to its
half value ðs2 Þ. The time constants for the two exponential waveforms are different and
are related as per Eq. (1), where V0 is the charging voltage of the impulse capacitors via
the spark gaps and g is the efficiency of the impulse generator.
s2  t t

V ðt Þ ¼ V 0 g   es2  es1 ð1Þ
s 2  s1

From a testing perspective, the time constants depend on the correlation between
the test circuit parameters and their interaction with the cable system as a test object.
Thus, by adjusting the resistance (R) and capacitance (C) of the test setup, desired front
(and respectively tail) times of the exponential waveform can be generated. The limits
are set by the R and C components of the test circuit.

5.1 Impulse Voltage Generator

An equivalent RC circuit representation can be used to explain the generation of
impulse voltages. An impulse capacitor (Cs) is charged via a charging resistor up to a
high enough DC breakdown voltage, which would then lie within the trigger range of
the sphere gap. An impulse voltage (Vi) is generated by the connected network com-
ponents. Subsequently, the load capacitor is charged via the front resistor (Rd) which
forms the front of the impulse voltage waveform. The impulse capacitor (Cs) is dis-
charged via the tail resistor (Re) and forms the tail of the impulse voltage waveform.
The superposition of both processes generates a double exponential waveform, which
could be used for testing purposes within laboratory conditions.
While being charged all capacitors are connected in parallel and the switching gap
is insulating. The charging time is given by the ladder network of impulse capacitors,
charging resistors, the number of stages and the charging unit. The discharging of
impulse capacitors occurs in series following an ignition of the switching gap of first
stage and then experiencing a voltage cascading effect on subsequent gaps in higher
stages (see Fig. 10).
1410 T. Karmokar and O. Lennerhag

Fig. 10. Schematic representation of discharging of an impulse generator on test object (left)
and simulation of the discharging behaviour in OrCAD PSpice® A/D simulation (right).

5.2 Long Impulse Composite Voltage Test Results

Figure 11 indicates the results of performing the long impulse test on a 320 kV
DC XLPE cable system as a test object. The long impulse with a time-to-peak of

Fig. 11. Test laboratory output waveform for same positive polarity (upper) and same negative
polarity (lower) of a long impulse TOV test.
 Simplified Approach for Investigating Overvoltages in DC Cables 1411

7000 µs and time-to-half of 25000 µs was superimposed on the rated DC voltage of

320 kV for both polarities, thus, executing a composite voltage test as defined in [9].
The accessories consisted of terminations and two prefabricated joints. The conductor
temperature was maintained to at least 70 °C during the test while maintaining a
minimum temperature gradient across the XLPE insulation according to [3]. For testing
purposes one hundred such long impulses were applied, and the cable system suc-
cessfully withstood and passed the test.

6 Conclusion

This paper investigates the use of a simplified model of a half-bridge MMC for ana-
lyzing overvoltage caused by DC-side faults. The simulation results indicate that the
TOV experienced by the non-faulty pole due to a pole-to-ground fault can be seen as an
exponentially varying wave shape.
An attempt has been made to experimentally generate a TOV wave shape, which
correlates to the simulations. The design of the experimental setup was executed in the
high voltage testing laboratory of NKT HV Cables AB by retrofitting conventional test
components to generate a much longer than standard impulse wave shape. Thus, it has
been proven that a substantially long tail time as compared to current test standards can
indeed be reproduced for testing purposes, as it has been demonstrated in this paper,
without investigating the upper limit on the maximum duration of the tail time.
The suitability of existing standard tests regarding their representative nature and
the actual impact of TOVs imposed on cables and accessories in a point-to-point
HVDC system is currently being investigated in CIGRE Joint Working Group
(JWG) B4/B1/C4.73 [10].

References
1. EU Horizon 2020 (H2020): Progress on Meshed HVDC Offshore Transmission Networks.
https://www.promotion-offshore.net/about_promotion/the_project/. Accessed 15 May 2019
2. Lundberg, P., Gustafsson, A., Jeroense, M.: Recent advancements in HVDC VSC systems –
HVDC and power electronics technology and development. In: Cigré Symposium, Lund,
Sweden (2015)
3. Cigré Technical Brochure 496: Recommendations for testing DC extruded cable systems for
power transmission at a rated voltage up to 500 kV. Working Group B1.32, April 2012
4. Cigré WG B4.57 TB 604: Guide for the Development of Models for HVDC Converters in a
HVDC Grid (2014)
5. Leterme, W., Beerten, J., Van Hertem, D.: Equivalent circuit for half-bridge MMC DC fault
current contribution. In: International Energy Conference (ENERGYCON), Leuven,
Belgium (2016)
6. Ajaei, F., Iravani, R.: Cable surge arrester operation due to transient overvoltages under DC-
Side faults in the MMC–HVDC link. IEEE Trans. Power Deliv. 31(3), 1213–1222 (2016)
1412 T. Karmokar and O. Lennerhag

7. Ametani, A., Ohno, T., Nagaoka, N.: Cable System Transients - Theory, Modelling and
Simulation, 1st edn. Wiley, Singapore (2015)
8. Kuechler, A.: High Voltage Engineering Fundamentals, Technology, Applications, 4th edn.
Springer, Heidelberg (2018)
9. IEC 60060-1: High-voltage test techniques – Part 1: General definitions and test
requirements, 3rd edn. (2010)
10. Cigré JWG B4/B1/C4.73: Surge and extended overvoltage testing of HVDC cable systems
 Experimental Investigation of the Overvoltage
Steepness Effects on Corona Inception
Characteristics

Evanthia I. Bousiou1(&) , Pantelis N. Mikropoulos1 ,

and Vasileios N. Zagkanas2
1
High Voltage Laboratory, School of Electrical and Computer Engineering,
Faculty of Engineering, Aristotle University of Thessaloniki,
54124 Thessaloniki, Greece
ebousiou@ece.auth.gr
2
Fieldscale P.C., 54250 Thessaloniki, Greece

Abstract. The overvoltage surges propagating along overhead line conductors

are gradually distorted and attenuated because of impulse corona discharge. The
extent of these effects is dependent mainly on the associated corona energy
losses. The inception characteristics of impulse corona are most commonly
assessed based on corona inception under steady or slowly varying electric
fields. This, however, may result in inaccurate assessment of the impulse corona
effects, since impulse corona characteristics are greatly affected by the over-
voltage steepness. This study presents an experimental investigation on corona
inception characteristics under lightning impulse voltages in the coaxial con-
figuration. The steepness and polarity of the applied impulse voltage are con-
sidered as influencing parameters. The corona inception voltage and the total
charge injected in the gap increase markedly as the overvoltage steepness
increases. This effect is more pronounced for impulse voltages with shorter
wavefronts and for positive than negative corona. There is a critical overvoltage
steepness, depending on impulse voltage wavefront, above which the increase of
corona inception voltage, as well as of the total corona charge injected in the
gap, with overvoltage steepness is less pronounced; this is associated with the
transition from the glow to streamer mode of corona inception. At transition
between corona modes, the total charge injected in the gap decreases with
increasing wavefront duration and is markedly lower for negative than positive
corona. In an effort to model corona losses for studies on attenuation and dis-
tortion of lightning overvoltages, a procedure is introduced that considers the
corona inception U-t characteristic and a working relationship for the estimation
of the total corona charge injected in the gap.

Keywords: Electrical discharges
Impulse corona
Overhead lines

Steepness
Surges

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1413–1422, 2020.
https://doi.org/10.1007/978-3-030-31676-1_132
1414 E. I. Bousiou et al.

1 Introduction

As regards to the overvoltage protection and insulation coordination of power systems,

surge corona discharge attenuates and distorts the overvoltages propagating on over-
head power lines and substations [1–4]. In transient studies, neglecting corona effects
on surges propagating along line conductors and incoming to the connected substations
may result in over-designed insulation systems and means for overvoltage protection.
Also, knowledge on the characteristics of surge corona is essential for the better
understanding of insulation flashover mechanism, thus for improving the design of
insulation systems and reliability of power systems.
The surge corona effects on overhead power lines related to energy losses are most
commonly assessed based on the inception characteristics of corona under steady or
slowly varying electric fields. However, such an approach may result in inaccurate
assessment of the corona effects. It is well established that the salient characteristics of
impulse corona discharge, namely inception voltage and charge, depend on the
steepness of the overvoltages arising on conductors [5–8].
This study presents an experimental investigation on impulse corona characteristics
under lightning impulse voltages of both polarities. Impulse corona characteristics,
namely inception time, voltage, as well as the associated charge injected in the coaxial
cylindrical electrode arrangement were measured at several applied voltage levels.
Thus, corona inception voltage-time and charge-voltage curves were obtained. The
effects of overvoltage steepness and polarity on corona inception characteristics are
assessed and discussed. A procedure for modelling corona losses is introduced, which
considers the corona inception U-t characteristic and a working relationship for the
estimation of the total corona charge injected in the gap.

2 Experimental Arrangement

The effects of overvoltage steepness on corona inception characteristics were experi-

mentally investigated by using a coaxial cylindrical electrode arrangement, comprising
an inner conductor with radius r0 = 0.028 cm and an outer cylinder with radius
R = 29.5 cm (Fig. 1). Lightning impulse voltages of both polarities with front time, T1,
0.4, 1, 1.3 and 2.4 ls and a time to half-value, T2, *50 ls were applied to the inner
conductor. They were produced by an impulse voltage generator 140 kV, 245 J and
were being monitored by using a capacitive divider. The outer cylinder was grounded
along 1 m of its length through a 9.41 nF capacitor, allowing for the charge injected in
the gap to be measured; two shielding sections 50 cm in length were used to eliminate
edge effects (Fig. 1).
Corona inception probability distributions were obtained by applying fifty impulse
shots per voltage level, for several levels gradually increasing in amplitude. The gap
was further overstressed, to investigate the effects of the amplitude of the applied
impulse voltage on corona inception characteristics. Thus, the average value and the
associated standard deviation of the corona inception time, t1, instantaneous voltage, U1
and total corona charge, q, over twenty impulse shots per several voltage levels were
estimated.
 Experimental Investigation of the Overvoltage Steepness Effects 1415

L2 = 50 cm L1 = 100 cm L2 = 50 cm

R = 29.5 cm
2r0

Impulse
voltage
generator
9.41 nF
OSC

Fig. 1. Schematic diagram of the experimental arrangement.

600
70
Qa (t)
60 500
qa (t)
50 400

Q (nC/m)
40
U (kV)

Qb (t) 300
Ut,a = 35.0 kV
30 U1,a qb (t)
Ut,b = 22.8 kV 200
20 U1,b
10 100
t1,a t1,b
0 0
0 1 2 3 4
t (μs)

Fig. 2. Simultaneous measurements of applied impulse voltage (1/48 ls) and charge in the gap;
q(t): corona charge, t1 and U1: corona inception time and voltage, respectively.

Experiments were performed in ambient air; absolute humidity and relative air
density varied naturally between 11.5–14.5 g/m3 and 0.992–1.012 p.u., respectively.

3 Experimental Results

3.1 Corona Inception Voltage

The corona inception probability distributions were found to be well approximated by
the normal distribution (Fig. 3). Thus, the values of the, applied voltage causing 50%
corona inception probability, Uin, and of the associated standard deviation, rin, were
computed (Table 1). Uin increases as the wavefront duration of the applied impulse
1416 E. I. Bousiou et al.

voltage decreases and is higher for positive than negative corona. As shown in Table 1,
with the exception of the impulse voltage with the longest front time (2.4 ls), Uin, is
higher than the dc corona inception voltage; the latter were estimated for both polarities
according to [9, 10]. Table 1 also lists the limiting values of corona inception voltage,
Uth, for both polarities. These values correspond to applied impulse voltages causing
2.5% inception probability and were obtained for the longest front time (2.4 ls). Uth
can be considered as the threshold values of corona inception voltage for the investi-
gated cases.
Figure 4 shows the average values of the instantaneous corona inception voltage,
U1, normalized with respect to the test impulse voltage, Ut, as a function of the latter.
At threshold inception conditions, corona initiates at voltages around the peak value of
the test impulse voltage. As the applied voltage increases in amplitude, corona initiates
sooner during the wavefront at voltages markedly lower than the peak of the test
impulse voltage; this is more pronounced for longer wavefronts and for negative than
positive corona. Also, at a fixed test impulse voltage, U1 is higher for shorter wave-
fronts and for positive than negative corona.
By assuming a linearly rising wavefront between the 10% and the 90% of the peak
value of the test impulse voltage, Fig. 5 shows the variation of corona inception
voltage, U1, with the overvoltage steepness, dU/dt, both magnitudes normalized with

1.0 (a) 1 (b)

0.9 0.9
Corona inception probability

Corona inception probability

0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3 0.4/48 μs
0.4/48 μs
0.2 1/48 μs 0.2 1/48 μs
1.3/50 μs 1.3/50 μs
0.1 0.1
2.4/50 μs 2.4/50 μs
0.0 0
10 15 20 25 30 10 15 20 25 30
Ut (kV) Ut (kV)

Fig. 3. Positive (a) and negative (b) corona inception probability distributions; fitting curves
were drawn according to normal distribution.

Table 1. Corona inception voltages

Impulse voltage Positive corona Negative corona
T1/T2 Uin (kV) rin (%) Uin (kV) rin (%)
0.4/48 ls 21.3 10.74 19.0 2.78
1.0/48 ls 18.9 8.09 18.6 2.22
1.3/50 ls 18.2 8.38 17.7 1.33
2.4/50 ls 16.7 1.64 15.8 1.36
Uth (kV) 16.2 15.4
Udc (kV) 16.9 17.3
 Experimental Investigation of the Overvoltage Steepness Effects 1417

1.0 1.0
0.4/48 μs 0.4/48 μs
1/48 μs 1/48 μs
0.9 1.3/50 μs 0.9 1.3/50 μs
2.4/50 μs 2.4/50 μs
0.8 0.8

U1/Ut (p.u.)
U1/Ut (p.u.)

0.7 0.7

0.6 0.6

0.5 0.5

(a) (b)
0.4 0.4
15 20 25 30 35 40 45 50 55 60 15 20 25 30 35 40 45 50 55 60
Ut (kV) Ut (kV)

Fig. 4. Variation of (a) positive and (b) negative normalized corona inception voltage, U1/Ut,
with the test impulse voltage, Ut; U1: average corona inception voltage per voltage level; vertical
bars denote r, curves denote average trends.

1.8 1.8
0.4/48 μs 0.4/48 μs
1.7 1/48 μs 1/48 μs
1.7
1.3/50 μs 1.3/50 μs
1.6 2.4/50 μs 1.6 2.4/50 μs
U1/Uth (p.u.)

U1/Uth (p.u.)

1.5 1.5

1.4 1.4

1.3 1.3

1.2 1.2

1.1 1.1
(b)
(a) 1.0
1.0
1.0 1.5 2.0 2.5 3.0 3.5 1.0 1.5 2.0 2.5 3.0 3.5
(dU/dt)/(dU/dt)th (p.u.) (dU/dt)/(dU/dt)th (p.u.)

Fig. 5. Variation of (a) positive and (b) negative corona inception voltage, U1/Uth, with
overvoltage steepness (dU/dt)/(dU/dt)th; U1: average inception voltage, Uth: threshold inception
voltage; vertical bars denote r, curves denote average trends.

their values corresponding to threshold corona inception conditions. It is obvious, that

as dU/dt increases due to higher applied voltages, U1 increases markedly reaching
values significantly higher than Uth (up to *1.8 times), especially for shorter wave-
fronts and for positive corona. As can be deduced from Fig. 5, for each impulse voltage
wavefront there is a critical overvoltage steepness, above which the rate of increase of
U1 with dU/dt is slower.
Figure 6 shows the corona U1−t1 inception characteristics for both impulse voltage
polarities. These characteristics can be well approximated by the following expression:
 
A
U1 ¼ Uth
1 þ ð1Þ
t1

where Uth is the threshold corona inception voltage corresponding to applied impulse
voltages causing 2.5% inception probability (Table 1) and A is a constant taking values
1418 E. I. Bousiou et al.

31
(+) 0.4/48 μs (-) 0.4/48 μs
29 (+) 1/48 μs (-) 1/48 μs
(+) 1.3/50 μs (-) 1.3/50 μs
27 (+) 2.4/50 μs (-) 2.4/50 μs
expression (1)
25
U1(kV)

23

21

19

17

15
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
t1 (μs)

Fig. 6. Corona U1−t1 inception characteristics; vertical bars denote r, fitting curves were drawn
according to (1).

0.21 and 0.18 for positive and negative corona, respectively. It must be noted that the
values of both Uth and A are specific to the investigated experimental configuration.

3.2 Corona Charge

Figure 7 shows the total corona charge injected in the gap, q, for the full duration of the
applied impulse voltages as a function of the overvoltage steepness, dU/dt. q increases
noticeably with increasing dU/dt. In an analogous way to the corona inception voltage
(Fig. 5), for each of the applied impulse voltages, a critical overvoltage steepness can
be determined, above which the rate of increase of q with dU/dt is slower. This change
in the dependence of q (Fig. 7), as well as of U1 (Fig. 5), on dU/dt is attributed to the
different modes of corona inception [11]. At threshold inception conditions and under
relatively low values of dU/dt, impulse corona initiates in the glow mode. For dU/dt
higher than the critical value the inception of a streamer corona is established (Fig. 7).
The total charge injected in the gap and the associated streamer corona inception
voltage, designated qst and U1st respectively, corresponding to the critical dU/dt
(Fig. 7), are shown in Fig. 8. These values are considered as characteristics of the
transition between the glow and streamer corona inception modes in the present
experimental configuration. At transition between corona modes, the total charge
injected in the gap, qst, decreases with increasing wavefront duration, tf, and is
markedly lower for negative than positive corona (Fig. 8a). These results are related to
the analogous variation of the streamer corona inception voltage, since qst and Ust are
linearly related (Fig. 8b). The strong polarity effect on qst certainly suggests a greater
growth of positive streamer corona, especially when considering that deviations in
 Experimental Investigation of the Overvoltage Steepness Effects 1419

streamer inception voltage between positive and negative corona are less than 10%
(Fig. 8b).

700 350
0.4/48 μs 0.4/48 μs
1/48 μs 1/48 μs
600 300
1.3/50 μs 1.3/50 μs
2.4/50 μs 2.4/50 μs
500 250

400 200

q (nC/m)
q (nC/m)

300 150

200 100

100 50
(a) (b)
0 0
0 20 40 60 80 100 0 20 40 60 80 100
dU/dt (kV/μs) dU/dt (kV/μs)

Fig. 7. Variation of (a) positive and (b) negative total corona charge injected in the gap, q, with
the wavefront steepness, dU/dt; curves show average trends, arrows denote the transition between
glow and streamer corona inception modes.

200 200
(+) 0.4/48 μs (-) 0.4/48 μs (+) 0.4/48 μs (-) 0.4/48 μs
(+) 1/48 μs (-) 1/48 μs (+) 1/48 μs (-) 1/48 μs
160 (+) 1.3/50 μs (-) 1.3/50 μs 160 (+) 1.3/50 μs (-) 1.3/50 μs
(+) 2.4/50 μs (-) 2.4/50 μs (+) 2.4/50 μs (-) 2.4/50 μs
qst (nC/m)
qst (nC/m)

120 120

80 80

40 40

(a) (b)
0 0
0.0 1.0 2.0 3.0 4.0 16 18 20 22 24
tf (μs) U1st (kV)

Fig. 8. Variation of the corona charge, qst, at streamer inception with (a) wavefront duration, tf,
and (b) streamer corona inception voltage, U1st.

Figure 9 shows the corona charge normalized with respect to the corona charge at
streamer inception, q/qst, as a function of the overvoltage stressing the gap, Ut/U1. The
following expression may well describe the relationship between charge and
overvoltage:

q Ut
¼a
b ð2Þ
qst U1

where a and b are constants that depend on the mode of corona inception (Table 2). In
the present configuration the streamer corona is established for overvoltage ratios Ut/U1,
higher than *1.1.
1420 E. I. Bousiou et al.

Table 2. Constants to be used in (2)

Glow corona Streamer corona
a 10 4.0
b 10 3.4

5
(+) 0.4/48 μs (-) 0.4/48 μs
(+) 1/48 μs (-) 1/48 μs
4 (+) 1.3/50 μs (-) 1.3/50 μs
(+) 2.4/50 μs (-) 2.4/50 μs
q/qst (p.u.)

3 Streamer corona
Glow corona
2

0
0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
Ut/U1 (p.u.)

Fig. 9. Total corona charge injected in the gap, q, normalized with respect to that corresponding
to streamer inception, qst, as a function of the overvoltage Ut/U1; Ut: test impulse voltage, U1:
corona inception voltage; lines were drawn according to (2).

3.3 Estimation of Q-U Curves

Overvoltage surges as they propagate along overhead line conductors are gradually
distorted and attenuated by impulse corona. The extent of these effects can be quan-
tified taking into account the associated energy losses [3]. The latter are represented by
the enclosed area of the charge-voltage (Q-U) curves following a hysteresis loop. The
Q-U curves can be described through empirical charge-voltage relationships, the
parameters of which most of the times are experimentally determined [12–16].
Figure 10 shows examples of measured and estimated Q-U curves. In this figure,
the measured Q-U curves have been derived by eliminating time from the charge and
voltage oscillograms of Fig. 2. For estimating Q-U curves, first the charge in the gap
increases linearly with applied voltage, with a slope equal to the geometrical capaci-
tance of the gap until the corona inception voltage, U1, is reached (point A). U1 is
estimated using the empirical expression (1) together with the double-exponential
function of the impulse voltage. Then, the charge in the gap is assumed to increase
linearly reaching its maximum value at the peak of the test impulse voltage (point B).
The maximum charge injected in the gap is estimated from expression (2), using U1
and the value of the charge at streamer inception, qst, corresponding to the wavefront
duration of the applied impulse voltage (Fig. 8a). Thereafter, during the wavetail of the
impulse voltage, the charge in the gap is assumed to decrease linearly considering the
geometrical capacitance of the gap.
 Experimental Investigation of the Overvoltage Steepness Effects 1421

As evident in Fig. 10, the estimated dynamic corona capacitance (instantaneous dq/
dU) deviates from the measured one which does not increase linearly; however, this is
not expected to affect appreciably the estimated energy losses associated with corona
discharge. To generalize the above procedure for estimating corona energy losses,
further experimental work is needed, by considering, in addition to wavefront duration
and overvoltage steepness, the conductor geometry and atmospheric conditions effects.

600

500 B

400
Q (nC)

300
(a)
200
(b)
100 measured
A
estimated
0
0 5 10 15 20 25 30 35 40
U (kV)

Fig. 10. Measured and estimated Q-U curves, under positive 1/48 ls with amplitude 35 kV
(trace a) and 22.8 kV (trace b).

4 Conclusions

Corona inception characteristics under lightning impulse voltages have been experi-
mentally investigated in the coaxial cylindrical electrode arrangement. The wavefront
steepness, amplitude and polarity of the applied impulse voltage were considered as
influencing parameters.
The corona inception characteristics, namely corona inception voltage and total
charge injected in the gap, increase markedly as the overvoltage steepness increases.
This effect is more pronounced for impulse voltages with shorter wavefronts and for
positive than negative corona.
There is a noticeable change of the rate of increase of the corona inception char-
acteristics with overvoltage steepness, associated with the transition between the glow
and streamer modes of corona inception. At transition between corona modes, the total
charge injected in the gap decreases with increasing wavefront duration and is mark-
edly lower for negative than positive corona. This polarity effect on streamer corona
charge certainly suggests a greater growth of positive streamer corona, given also that
deviations in streamer inception voltage between polarities is less than 10%.
In an effort to model corona losses for studies on attenuation and distortion of
lightning overvoltages, a procedure has been introduced that considers the corona
inception U-t characteristic and a working relationship for the estimation of the total
1422 E. I. Bousiou et al.

corona charge injected in the gap. Modelling of corona losses considering in addition to
the impulse voltage waveshape, the conductor geometry and the atmospheric condi-
tions, is subject of further work in progress.

References
1. Wagner, C.F., Lloyd, B.L.: Effects of corona on traveling waves. Trans. AIEE 74(3), 858–
872 (1955)
2. Gary, G., Cristescu, D., Dragan, G.: Distortion and attenuation of traveling waves caused by
transient corona. CIGRE Study Committee 33 (1989)
3. EPRI: AC Transmission Line Reference Book – 200 kV and Adove. 3rd edn. EPRI, Palo
Alto (2005)
4. Thang, T.H., Baba, Y., Nagaoka, N., Ametani, A., Itamoto, N., Rakov, V.: FDTD simulation
of direct lightning strike to a phase conductor: influence of corona on transient voltages at
the tower. EPSR 123, 128–136 (2015)
5. Maruvada, P.S., Menemenlis, H., Malewski, R.: Corona characteristics of conductor bundles
under impulse voltages. IEEE Trans. PAS 96(1), 102–115 (1977)
6. Gary, C., Dragan, G., Lungu, I.: Several physical characteristics of impulse corona discharge
around conductors. Part 2: onset voltage and delay time of corona discharge. IEE Proc.-A
136(4), 197–201 (1991)
7. Harid, N., Waters, R.T.: Statistical study of impulse corona inception parameters on line
conductors. IEE Proc. A Sci. Measur. Technol. 138(3), 161–168 (1991)
8. Mikropoulos, P.N., Zagkanas, V.N.: Impulse corona inception in the coaxial cylindrical
electrode arrangement in air: effects of the steepness of the applied voltage. In: 18th
International Symposium on High Voltage Engineering, Seoul, South Korea (2013)
9. Mikropoulos, P.N., Zagkanas, V.N.: Threshold inception conditions for positive DC corona
in the coaxial cylinder electrode arrangement under variable atmospheric conditions. IEEE
Trans. DEIS 22(1), 278–286 (2015)
10. Mikropoulos, P.N., Zagkanas, V.N.: Negative DC corona inception in coaxial cylinders
under variable atmospheric conditions: a comparison with positive corona. IEEE Trans.
DEIS 23(3), 1322–1330 (2016)
11. Bousiou, E.I., Mikropoulos, P.N.: Experimental investigation on corona charge-voltage
characteristics in the coaxial configuration under lighting impulse voltages. In: IEEE
International Conference on High Voltage Engineering and Application, Athens, Greece
(2018)
12. Mihailescu-Suliciu, M., Suliciu, I.: A rate type constitutive equation for the description of the
corona effect. IEEE Trans. PAS 100(3), 3681–3685 (1981)
13. Gary, C., Timotin, A., Cristescu, D.: Prediction of surge propagation influenced by corona
and skin effect. Phys. Sci. Measur. Instrum. Manag. Educ. – Rev. IEE Proc.-A 130(5), 264–
272 (1983)
14. Inoue, A.: Propagation analysis of overvoltage surges with corona based upon charge versus
voltage curve. IEEE Trans. PAS 104(3), 655–662 (1985)
15. Yang, P., Chen, S., He, J.: Lightning impulse corona characteristic of 1000-kV UHV
transmission lines and its influences on lightning overvoltage analysis results. IEEE Trans.
Power Deliv. 28(4), 2518–2525 (2013)
16. He, J., Zhang, X., Yang, P., Chen, S., Zeng, R.: Attenuation and deformation characteristics
of lightning impulse corona traveling along bundled transmission lines. EPSR 118, 29–36
(2015)
 LPS and Grounding System for Light Rail
Transit in Tropical Area with High
Lightning Density

Wisnu Adyatma S(&) and Reynaldo Zoro

Bandung Institute of Technology, Bandung, Indonesia

wisnoeas.united@gmail.com, inanzoro@gmail.com

Abstract. Palembang Light Rail Transit structure is higher than surrounding

object, so it is definitely an easy target for lightning strike. So far, the early
streamer emission air terminals only protect stations and important objects in
viaduct such as signal and telecommunication poles, and most of the track will
be an easy target for lightning strike. If it is hit by a direct strike, there will be
very high voltage elevation because the grounding is installed on every 5 piers,
in fact it can even reach 12 MV. Installation of grounding on each pier will
lower the voltage elevation by 57%. Besides the threat of direct strikes, the
voltage can arise from BTS tower and 70 kV transmission wire induction when
that objects are struck by lightning. To overcome the threat due to direct strikes
and induction, we can install Overhead Ground Wire above the track. Foun-
dation grounding systems significantly reduce the grounding resistance, thus it is
the best choice for structure like Palembang Light Rail Transit.

Keywords: Lightning
Light Rail Transit
Easy target
Protection

Grounding

1 Introduction

Light Rail Transit (LRT) is a mass transportation with right of way as well as a train.
LRT track can be either on ground, underground, or even elevated. In Palembang Light
Rail Transit case, 23-km LRT track is located 9–19 m above ground level, so it is
definitely an easy target for lightning strike. Thus, we need to analyze the existing
lightning protection system for Palembang Light Rail Transit.

2 MRT Train in Singapore Crippled by Lightning Strike

There are 2 types of MRT station in Singapore: elevated and underground station.
Unfortunately, trains on surface lines are occasionally struck by lightning. A lightning
strike on November 20, 2017, has stopped Singapore MRT train and injured the driver.
The incident caused a delay in service between Tanah Merah and Paya Lebar. Pas-
sengers were directed to the station, and then the train was withdrawn for further
investigation [1].

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1423–1436, 2020.
https://doi.org/10.1007/978-3-030-31676-1_133
1424 W. Adyatma S and R. Zoro

Another incident was on May 11, 2016, when an SMRT train on the North-South
Line was crippled by lightning. It disrupted services from Yishun to Yio Chu Kang
stations [1]. Such accidents may also occur in Palembang.

3 Lightning Parameters

These parameters would be useful in destruction effect analysis of lightning strikes [2].

3.1 Peak Current

Peak current (i) determines the resistive drop voltage.

U ¼ Rimp
i ð1Þ

where: U = resistive drop voltage (kV)

Rimp = surge impedance (X)
i = peak current (kA).

3.2 The Steepness of Current Rise

The steepness of current rise (di/dt) determines inductive drop voltage.

U ¼ L
di=dt ð2Þ

where: U = inductive drop voltage (kV)

L = inductance (lH)
di/dt = the steepness of current rise (kA/µs).

3.3 The Current Charge

The current charge (Q) is a measure of lightning current energy as a lightning jumping
current to a metal. In the tropics, lightning impulses occur longer tail. This can damage
the material physically (can melt the material).
Z
Q ¼ i dt ð3Þ

where: Q = current charge (kA . s)

i = peak current (kA).
 LPS and Grounding System for LRT in Tropical Area 1425

3.4 Impulse Force

Impulse Force (E), is the basis of the mechanical effect and the heat of the electrical
impulse on the resistor. Impulse force can destroy object.
Z
E ¼ i2 dt ð4Þ

where: E = impulse force (kA2 . s)

i = peak current (kA).

4 Tropical Lightning Characteristics

Indonesia is an archipelagic country with tropical climate. In this area cumulonimbus

cloud (cb cloud) is easily formed and then producing lightning. Lightning parameters
needed for the protection system is already measured and published since 1995 in
Lightning Measurement Station at Mt. Tangkuban Perahu in West Java operated by
ITB Lightning Research Center. Tropical lightning current characteristic which are
used for this research are peak current lightning characteristic (i) and lightning current
steepness (di/dt) characteristic is given in Table 1.

Table 1. Tropical lightning characteristic [3].

Lightning characteristic Negative polarity Positive polarity
Peak current (i) Maximum 335 kA 392 kA
50% Probability 40 kA 18 kA
Average 41 kA 30 kA
Steepness (di/dt) Maximum 119 kA/µs 120 kA/µs
50% Probability 30 kA/µs 20 kA/µs
Striking density 4.1–12.4 1.5–3.8
(Strike/km2/year)
Total striking density 7.9–15.5
(Strike/km2/year)

The 50% probability value for peak current and steepness from negative polarity
tropical lightning characteristic are applied to this research. Besides the lightning
measurement at Mt. Tangkuban Perahu, lightning detection system is also used either
to plan the suitable lightning protection system, analyze faults due to lightning strike,
even predict and give early warning about lightning strike. With this system, we can
easily get ground flash density data around Palembang, which can reflect the activity of
lightning in those area (Fig. 1).
1426 W. Adyatma S and R. Zoro

Fig. 1. Ground flash density around Palembang.

5 Rolling Sphere Method

During the sixties, the Edison Electric Institute (EEI) set up a research project with a
team of the Institute of Research and Technology Illinois headed by Whitehead. The
project consisted of a theoretical analysis and practical measures over a period of nine
years on the electric power networks lines. A statistical analysis of the performance of
air lines was the objective of the project. In order to improve the protection of electrical
networks, it was necessary to know two parameters: the striking distance and the
current that a phase can carry without any failure. Whitehead and Amstrong developed
the model of Wagner, by using the experimental results of Paris. They deduced the
relation [4]:

rs ¼ 6:7 I00;8 ð5Þ

 LPS and Grounding System for LRT in Tropical Area 1427

where: I0 = forecast lightning current (kA)

rs = striking distance (m).
Based on these observations of power supply networks, the critical lightning cur-
rent that a line can carry without damage I0 was determined. In 1973, Whitehead and
Love deduced the following expression [4]:
 
rs ¼ 2I þ 30 1  eI=6:25 ð6Þ

which was approximated by Gilman and Whitehead to the famous relationship cur-
rently used by all international standards [4]:

rs ¼ 10 I 0:65 ð7Þ

where: I = lightning current (kA)

rs = striking distance (m).
This model was named Electrogeometric Model (EGM). It is based on the principle
that the lightning strike position is determined only by the striking distance [4].
The Rolling Sphere Method is the direct application of the EGM for 3D geometries.
It is commonly used technique recommended in the International Standard IEC 62305-3
[5].
An imaginary sphere, of radius equal to the assumed striking distance, is rolled over
the structure. All surface contact points are likely striking points and deemed to require
protection, whilst the unaffected volumes are not struck for lightning currents equal to
or greater than I and are considered protected. A dashed line indicates the trajectory of
the center of the rolling sphere. It corresponds to the different positions of the down-
ward negative leader tip likely to strike the structure [4].
The RSM is relatively simple to use. However, this model disregards the physical
basis of the upward leader inception and propagation process and the importance of the
field intensification caused by the structure geometry or by object being over [4].

6 Coverage Area and Voltage Elevation

Palembang Light Rail Transit protection against direct lightning strikes is given by
installation of ESE air terminal on the top of lightning rod with N2XSY (single
shielded) cable in viaduct and Ericore (double shielded) cable as the down conductor at
stations. In addition, it has lightning counter from Prosurge and magnetic tape measure
from Lightning Research Center installed in the lightning event counter box (LEC
box). 3 lightning rods is installed on the roof of each station, while for the viaduct, it is
only mounted on the location of important objects such as a signal or telecommuni-
cation poles (Fig. 3).
1428 W. Adyatma S and R. Zoro

Fig. 2. Coverage area of lightning rods.

Fig. 3. 70 kV transmission lines crossing over LRT track.

Areas enclosed by circles are the coverage area of lightning rods, while the line that
across the track line is 70 kV transmission line. As we can see in Fig. 2, most of the
track will be an easy target for lightning strike because it is outside the coverage area of
lightning rods (Fig. 4).
 LPS and Grounding System for LRT in Tropical Area 1429

Fig. 4. Traction substation simplified diagram.

If the rolling stock (train) is hit by a direct strike, there will be an electromagnetic
wave travelling across nearby medium. It can be the rolling stock, third rail, rail,
handrail, or another object near the rolling stock. Firstly, if this travelling wave flow
inside the power system of rolling stock, then motor and the other equipment can be
damaged due to overvoltage, unless there are proper internal lightning protection
systems. Secondly, if this travelling wave flow on the third rail toward DC switchgear,
it should be cut by surge arrester in the outgoing panel of DC switchgear. Thirdly, if
this travelling wave flow on the rail toward negative panel, the negative panel will pass
it to the rectifier, and it should be cut by surge arrester on the DC output side of
rectifier. Lastly, if this travelling wave flow into handrail or another object, it may
produce touch voltage, step voltage, or side flash that hazardous for human near that
object, so for safety reasons, every metal object in viaduct should be grounded with a
proper grounding to make equipotential condition.
Unfortunately, when the piers were built, there were no planning to make
foundation-type grounding, so the handrail of viaduct is grounded every 5 piers using
vertical earthing rod with aluminum tape as its down conductor. The worst case of
direct strike is when the rolling stock is hit exactly in the middle of 2 grounded-pier.
For example, the height of pier-A and pier-B are 13 m and 15 m, the distance between
them is 120 m. The tropical lightning characteristic i = 40 kA, di/dt = 30 kA/µs, and
another parameter that we have set by R0 = 5 X, L = 1 µH/m are applied for this
research (Fig. 5).

Fig. 5. Voltage elevation example.

1430 W. Adyatma S and R. Zoro

If it is hit by a direct strike, there will be overvoltage V at viaduct.

V ¼ I : R0 þ L : di=dt ð8Þ

where: V = overvoltage on viaduct (kV)

I = lightning current (kA)
R0 = grounding resistance (X)
L = inductance of down conductor (µH)
di/dt = lightning current steepness (kA/µs).
So, the overvoltage V for the example above is

V ¼ i : R0 þ L : di=dt ¼ 40 kA : 5 X þ 1 lH=m : ðð13 m þ 15 m þ 120 mÞ=2Þ : 30 kA=ls

¼ 2:42 MV

With all piers measurement and this method, Palembang Light Rail Transit voltage
elevation profile due to direct strike can be obtained (Fig. 6).

Fig. 6. Palembang Light Rail Transit voltage elevation profile.

In the existing conditions (5-pier-grounding), the voltage elevation will be in the

range of 1.5 MV to 12 MV and the average value is 2.5 MV, whereas if there is
grounding on each pier, the voltage elevation will be in the range of 785 kV to 1.7 MV
and the average is 1 MV. If we install the grounding on each pier, we can reduce the
voltage elevation by 57%, even reaching 86% at the Ampera Bridge. So, grounding on
each pier is better than the existing condition.
Unfortunately, grounding on each pier is still dangerous. According to International
Standard IEC 62305-3, even the distance between down conductor for Class IV
Lightning Protection System is 20 m, and yet the distance between 2 piers is 30 m
(Table 2).
 LPS and Grounding System for LRT in Tropical Area 1431

Table 2. Typical preferred values of the distance between down-conductors according to the
class of LPS [5].
Class of LPS Typical distances m
I 10
II 10
III 15
IV 20

To improve the reliability of the lightning protection system, it is necessary to

install Overhead Ground Wire (OHGW) in the viaduct area. By using OHGW, the risk
of direct stroke to rolling stock or humans will decrease, due to the higher OHGW
structure, so that OHGW tends to be struck first [6].

7 Induced Voltage

In Indonesia, Jakarta and surrounding area, disturbances on the electric railway system
caused by lightning strike has made a lot of damage on the equipments and caused
shutdown on this transportation system. Most of disturbance are caused by induced
overvoltage coming from 1.5 kV DC overhead contact lines (OHCL) and nearby
telecommunication tower hit by lightning that damaged the railway signaling system
[7]. It proves that the disturbance is not only coming from the direct strike, but also
from indirect strike (Fig. 7).
There are 25 BTS towers as high as 70 m and 1 BTS tower as high as 100 m
around the Palembang LRT line. In addition to the BTS tower, there were also three
70 kV transmission lines that passed above the viaduct. This can be a source of
disturbance in Palembang LRT if these objects are struck by lightning strikes.
Induced voltage that appear in the LRT track can be calculated by simplified Rusck
formula [6].
0 1
Z 0 I0 h B
B1 þ v C
Vmax ¼ @ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiC
A ð9Þ
y p ffiffi
ffi
2v 0 1  2vv0

rffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 l0 1 4p : 107 H=m
Z0 ¼ ¼  30 X ð10Þ
4p e0 4p 109
36p F=m

v ¼ 1:2  108 m=s ð11Þ

where: Z0 = surge impedance (X)

I0 = lightning current (kA)
1432 W. Adyatma S and R. Zoro

Fig. 7. Electromagnetic induction due to direct strike toward BTS tower.

h = viaduct height (m)

y = strike distance from viaduct (m)
v = return stroke speed (m/s)
v0 = light speed in vacuum (m/s).
Assume that a 70 m BTS tower is hit by a direct strike with an I0 = 40 kA. If the
height of viaduct h = 12 m and distance from BTS tower to viaduct = 37 m, the
induced voltage is
0 1
Z 0 I0 h B
B1 þ v C
Vmax ¼ @ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiC
A
y pffiffiffi
2v 0 1  2vv0
 LPS and Grounding System for LRT in Tropical Area 1433

0 1
30 X : 40 kA : 12 m B 1:2 : 10 m=s8
C
Vmax ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi @1 þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi A
1:2 : 108 m=s
2
ð37 mÞ þ ð70 m  12 mÞ 2
3 : 10 m=s
8 2  3 : 108 m=s
¼ 275:5 kV

With the method above we can calculate induce voltage from both BTS towers and
70 kV transmission lines (Fig. 8).
Not only add the coverage area, Overhead Ground Wire can be used to reduce
voltage induction from another object [6]. So, OHGW is the best solution to protect
Palembang Light Rail Transit from lightning strike threat.

Fig. 8. Induced voltage from BTS towers and 70 kV transmission lines.

8 Grounding System Improvement

In the Palembang LRT existing condition, viaduct using multiple rod type grounding.
According to ANSI/IEEE Std 80-1986, grounding resistance for 3-rods-grounding can
be determined using formula [8]
 
1 q l
R ¼ loge 2:42
ð12Þ
3 2pl a=2

where: R = grounding resistance (X)

q = soil resistivity (Xm)
l = length of rod
a = radius of rod.
Assume that groundings are not connected galvanically with one another, with
q = 60 Xm, l = 3 m, and a = 0,008 m, then the ground resistance will be
1434 W. Adyatma S and R. Zoro

 
1 q l
R¼ log 2:4

2
3 2pl e a=2
 
1 60 Xm 3m
R¼ loge 2:42

3 2p : 3 m 0;008 m=2

R ¼ 7:72 X

Besides the existing grounding type above, the use of foundation grounding sys-
tems (FGS) has increased in the past decades. Foundation grounding systems (FGS) are
grounding systems comprising round steel rods (of diameter 10–16 mm) forming two
or three-dimensional grids. These are laid in the concrete foundations of buildings,
substations, transformer stations, overhead power line poles, etc. [9].
The skeleton elements of an FGS are welded together, or zinc-coated steel strips are
used to galvanically connect the already existing concrete reinforcements. Considering
the amount of reinforcement steel in the concrete, the influence of the strips on the
electrical characteristics of an FGS may be neglected [9].
The main advantage of an FGS is that it is located inside the mass of concrete
which protects it against corrosion, making its life practically unlimited. The concrete
also maintains humidity, which ensures a good electrical contact between the
grounding system and the surrounding soil over large areas. This results in a signifi-
cantly lower grounding resistance than for grounding conductors which are laid directly
in the ground. The shape and size of an FGS enables high intensity ground fault
currents to flow into surrounding soil. Relatively deeply situated in soil and with a
better contact between electrodes and the soil, an FGS contributes to equalizing the
surface potential, and thus to substantially lower the step and touch voltages [9].
Assume that both the concrete block and the surrounding soil are homogeneous, of
resistivities qc, and qg, respectively. The upper block surface is flush with the surface
of the ground and is in contact with the air.
With a simple FGS resistance formula and Table 3 below, [9]
qg q
Rf ¼ 0:45 pffiffiffi fð c Þ ð13Þ
S qg

S ¼ a:b ð14Þ

 
Table 3. Value of the Function f qc =qg [10]
qc/qg 0,2 0,5 0,8 1 2 5 8
f(qc/qg) 0,68 0,76 0,85 0,89 1,02 1,44 1,85

where: Rf = grounding resistance (X)

qg = soil resistivity (Xm)
 LPS and Grounding System for LRT in Tropical Area 1435

qc = concrete resistivity (Xm)

a = concrete width (m)
b = concrete length (m).
the grounding resistance for FGS can be approximated. It should be noted that formula
above was tested using the following assumption: radius of the rods forming the grid
(R) = 0.008 m, grid length (L1) = b − 1, and grid width (L2) = a − 1. For qc/qg > 1,
the following relation may be used: [10].
!
q qc
f c ¼ 0;14 þ 0;74 ð15Þ
qg qg

Assume that LRT pier dimensions were 1.5 m  1.5 m  10 m with

R = 0.008 m, L1 = b − 1, and L2 = a − 1. The soil resistivity is 60 Xm and the
concrete is 48 Xm. With given data and formula above, the foundation grounding
systems resistance is
qg q
Rf ¼ 0:45 pffiffiffi f ð c Þ
S qg
 
60 Xm 48 Xm
Rf ¼ 0:45 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi f ¼ 15:3 X
1:5 m  1:5 m 60 Xm

If all of Palembang LRT structures, piers and another track structures, is connected
galvanically, we can use a calculation method based on the principles of superposition
and reciprocity [11]. Applying this method to complex grounding systems created from
approximately 840 piers, the formula will be
 
1 1 1 1
¼g þ þ...þ ð16Þ
Req R1 R2 R840

where: Req = equivalent grounding resistance (X)

ƞ = coefficient reflecting the mutual effect among all foundation grounding.
Assume that all piers have grounding resistance value = 15.3 X, and g ¼ 0:6. The
equivalent grounding resistance is
 
1 1 1 1
¼g þ þ...þ
Req R1 R2 R840
 
1 1 1 1
¼ 0:6 þ þ...þ
Req 15:3 X 15:3 X 15:3 X

Req ¼ 0:03 X
1436 W. Adyatma S and R. Zoro

This results in significantly lower grounding resistance than both the existing
grounding resistance calculation (7.72 X) and measurement, which varies from 0.1 X
to 3 X. So, the foundation grounding systems (FGS) is definitely the best choice for
structure like Palembang Light Rail Transit.

9 Conclusions

5-pier-grounding installation won’t be enough to protect Palembang Light Rail Transit

from lightning direct strike. Besides that, tall objects like BTS towers and 70 kV
transmission lines could be another source of disturbance (voltage induction). In order
to improve it, installation of overhead ground wire (OHGW) can solve both of voltage
elevation and voltage induction problems. Furthermore, for the future structure
development, foundation grounding system should be considered as well.

Acknowledgments. The authors wish to acknowledge PT. Anaya Berkah Cemerlang for the
opportunity given to carry out the research concerning improvement of the protection and
grounding system in Palembang Light Rail Transit.

References
1. MRT lightning incident: Did you know Singapore is known as the ‘lightning capital’?,
https://www.straitstimes.com/singapore/mrt-train-on-east-west-line-struck-by-lightning-did-
you-know-singapore-is-also-known-as. Accessed 10 Jan 2019
2. Zoro, R.: Proteksi Sistem Tenaga: Proteksi Terhadap Tegangan Lebih pada Sistem Tegangan
Listrik. ITB, Bandung (2002)
3. Zoro, R.: Lightning characteristic and weather condition in the tropics measured in Mnt.
Tangkuban Perahu. Ph.D. dissertation, Post Graduate Faculty of Institut Teknologi Bandung.
Bandung, Indonesia (1999)
4. Alt-Amar, S., Berger, G.: A modified version of the rolling sphere method. IEEE Trans.
Dielectr. Electr. Insul. 16(3), 718–725 (2009). https://doi.org/10.1109/tdei.2009.5128511
5. IEC 62305-3. Protection Against Lightning - Part 3: Physical Damage to Structures and Life
Hazard. IEC, Geneva (2006)
6. Denov B.: Sistem proteksi petir pada light rail transit. ITB, Bandung (2018)
7. Zoro, R., Pakki, R.R., Komar, R.: Lightning protection for electric railway in Indonesia
telecommunication and signalling system. In: 2017 International Conference on High
Voltage Engineering and Power System, pp. 476–478. IEEE, Piscataway (2017). https://doi.
org/10.1109/ichveps.2017.8225893
8. ANSI/IEEE Std 80. IEEE Guide for Safety in AC Substation Grounding. IEEE, Piscataway
(1986)
9. Kostic, M.B., Popovic, B.D., Jovanovic, M.S.: Numerical analysis of a class of foundation
grounding systems. IEE Proc. C 137, 123–128 (1990). https://doi.org/10.1049/ip-c.1990.0017
10. Kostic, M.B.: Analysis of foundation grounding systems with external loops and rods. IEE
Proc. C 140(2), 73–76 (1993). https://doi.org/10.1049/ip-c.1993.0010
11. Kostic, M.B., Jovanovic, M.S.: A new approach of determining grounding resistance of one
class of complex grounding systems. Publ. Faculty Electr. Eng. Univ. Belgrade 120, 1–11
(1987). Ser. Pow. Belgrade
 Verification of the Aging Impact
of MOV Ceramics by Recording Wide-Range
V/I Characteristics

S. Joerres1(&), A. Claudi1, and G. Finis2

1
University of Kassel, Kassel, Germany
joerres@uni-kassel.de
2
Phoenix Contact GmbH & Co. KG, Blomberg, Germany

Abstract. Nowadays, varistors are widely used because of their energy

absorption capability (EAC) with concomitant economy and efficiency. Surges
and overvoltage transients unnoticeably degrade Metal Oxide Varistors
(MOV) over time. A newly developed condition monitoring system is based on
a time-resolved measurement of the discharge current. In order to be able to
compare and interpret acquired data sets from the monitoring system with rel-
evant aging models, extensive investigations of the measurable influences are
required. Therefore, an experimental setup has been developed for determining
different aging characteristics and process influences. A precise continuous
curve recording over nine current decades from the ohmic leakage current area
to the limiting resistance region up to several thousand Amperes allows studying
aging characteristics of differently stressed MOVs. This method combines
common spot-measurements from the curves like mA-Point, leakage current and
the nonlinearity exponent alpha. In this paper, a fully automated test bench is
presented. The performance of the system and its advantages are compared to
manually operated tests and different test equipment. First experimental results
show that single point measurements, depending on the type of stress, are
insufficient to detect degradation characteristics. Especially in the transition area
from the ohmic linear range to the non-linear range, irregularities terms of
degradation are ascertained. These studies are the basis for building up relevant
aging models depending on the process measurement variables. This results in
more accurate predictions for the state of health of varistor ceramics by using
recorded data sets as the time-resolved measurement of the discharge current.
Finally, it helps to understand the change of characterization by different
impacts. In addition, the system is scalable for the study of layered surge
arresters and thus applicable to all network levels.

Keywords: Metal oxide varistors
Degradation
State of health

1 Introduction

Surge arresters are important components in all electrical power systems in terms of
reliability and safety. Due to concomitant economy and efficiency, varistor-constructed
surge protective devices are widely used. Class II surge protective devices according to

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1437–1445, 2020.
https://doi.org/10.1007/978-3-030-31676-1_134
1438 S. Joerres et al.

IEC 61643-331 [1] are based on metal oxide varistor (MOV) technology. Varistors are
voltage dependant highly non-linear variable resistors that become low-ohmic in case
of surge events. Thus, the current associated with a surge voltage is discharged and the
voltage is limited to its protective level. Depending on the different stress levels and
repetition rates of surge transients, varistors age during their lifetime. Irreversible
formation of conductive semiconductor boundary layers between micro-varistors
(Fig. 1) change the electrical characteristics. This can cause unwanted and unexpected
equipment failures without further protection. To avoid such failures, condition mon-
itoring and the integration to Industrial Internet of Things (IIoT) can be used to increase
the system availability and permit condition-oriented and predictive maintenance.
A newly developed condition monitoring system measures the discharged current and
digitizes the surge current. The recording offers big data processing and an adaption of
relevant aging models. For building up degradation models depending on the measured
values provided by the monitoring system, the knowledge of alterations due to different
stress impacts are of elementary importance.

Zinc oxide
Intergranular boundary
Microvaristor

Fig. 1. Structure of a ZnO-based varistor with irreversible conductive paths according to [2]

The V/I characteristic curve is one of the most important parameter for an inves-
tigation of the varistor degradation. Therefore, a test bench has been developed which
measures the effect on the characteristics after different stress conditions. A powerful
ramp-controlled direct current source and a combination wave generator (CWG)
(V = 1.2/50 µs, I = 8/20 µs) were combined in a test bench to investigate the influ-
ences in the leakage current range (µA) up to the impulse current range (kA). Com-
bined, this enables V/I curve recording over nine decades and provides precisely
comparable data after different stress conditions or production processes like:
• Surge Stress
• Temperature, Humidity
• Temporary Overvoltage (TOV)
• Soldering/Thermal impact
 Verification of the Aging Impact of MOV Ceramics 1439

2 Motivation and Requirements

State of the art is the recording of the V/I characteristics of metal oxide varistors by
spot measurements of different points of interest as displayed in Fig. 2. The double
logarithmic reference representation allows the analysis of the current over many
decades and forms the typical V/I varistor characteristic. In general, the extent of
degradation often refers to the decrease of the varistor reference voltage measured at
1 mA. A common way is to measure the resulting varistor voltage VV direct current
controlled. In this measurement, the current should be maintained at a steady value
regardless of the load impedance. The source and measurement with constant current
should have a duration between 10 and 100 ms. During different stress tests, the
maximum permissible deviation for the varistor voltage VV should be in a range
of ± 10%. The maximum continuous voltage is defined as VDC and its leakage current
is an indicator for the state of health of varistor ceramics. The leakage current should be
less than the manufacturer’s specified value at VDC after the nominal discharge current
test [1]. Contemporarily handheld measuring equipment is a typical and precise
application for point determination in the leakage current region up to 1 mA after
normative tests.

IClamp
VV (1 A)
VV (1 mA)
Varistor Voltage in V

α≈40
VDC

Varistor Current in A

Fig. 2. Example for a V/I characteristic with relevant parameters according to [3]

Manufacturers often specify their products by measuring single points and inter-
polating the characteristic in the different areas of ohmic leakage current, limiting
resistance and high current with three different gradients a. They can be determined
according to Eq. (1) from the pairs of values for 1…10 mA or 1…1000 mA of the V/I
characteristic.

logðI2 Þ  logðI1 Þ
a¼ ð1Þ
logðV2 Þ  logðV1 Þ

For testing the protection level in a typical range of 300 A, a surge generator is
necessary. The protection level is the voltage drop across the varistor for surge
1440 S. Joerres et al.

currents >1 mA. The V/I characteristics show the maximum protection level as a
function of surge currents with an 8/20 µs waveform. [3]
The degradation of varistors by partial area spot measurements has been studied
over many years including the changes in electrical parameters as well as the
microstructure of the MOV. Nevertheless, for researching characteristic changes as a
function of stress tests or production impact, the knowledge of further data points are of
elementary importance. Statistics on the degradation of varistors and the need for a
large number of samples are required in the tests to eliminate the impact of randomness
of the results. Moreover, the need of possible aging capabilities such as surge aging or
aging due to temporary overvoltage under different ambient conditions are demanded.
The motivation for building up a fully automated test-bench for V/I recording with a
continuous characteristic up to several Amperes and a point measurement up to several
kilo Amperes has been taken after inexplicable failures in spite of passing the criterion
in Fig. 3. According to the IEC 61643-331 [1], this two spot measurements are a part of
passing criterion during tests or production impacts. Further research results in char-
acteristic changes around normative defined spots. To measure test batches in a sta-
tistically correct fashion extended test equipment and test algorithms are required as
described in the following section.

Fig. 3. Test conditions for varistor qualification. Extract from the norm IEC 61643-331 [1]

3 Implementation of the Test Bench

The requirements for the varistor characteristic test-bench are divided into following
tasks:
• recording at least from VDC to IClamp for varistors specified up to VV = 1000 V at
1 mA
• bipolar measurement
• high measuring process rate for recording high numbers of varistor samples
• storage of the measured data in a standard data type with summary test report
• flexible parametrization of the test sequence
• operator-friendly handling without required training period
• handling big amounts of data with prescribed evaluation methods
Figure 4 shows the overview system diagram of the newly developed fully auto-
mated test-bench for metal oxide varistors with its key-components. All devices are
 Verification of the Aging Impact of MOV Ceramics 1441

controlled with Matlab via TCPIP. This has the advantage that a single program acts as
control unit, data storage unit and evaluation unit with an integrated visualization. The
process of the program follows a specific algorithm. For starting a test procedure only
the electrical ratings such as VDC and IClamp, depending on the disc type, have to be
defined. Typical tests comprise a test series of several ceramics. Therefore, the program
automatically increments the initial device under test (DUT) number during a loop. The
first measuring process is realized by a powerful ramp controlled direct current source
with an adjustable dV/dt. The electric contactor 1 is closed and locked against the
second contactor, also controllable over TCPIP. The voltage ramp up to 1500 V with a
maximum current of 2 A is measured simultaneous and digitized into data storage and
processing. To achieve a high measurement accuracy of the current measurement
during the voltage ramp of several hundred volt per second, a shunt based meter with
an automatic measuring range switch is used, controlled by the program. This allows a
continuous curve recording and a simultaneous trigger of all components in the leakage
current area to the lower limiting resistance area without an affected energy impact.
After the change of state of the electric contactors 1 and 2, the curve is extended by
normative single surge currents by a combination wave generator with the curve shape
of 8/20 µs up to several kA in case of a short circuit.

A Oszilloscope Electric
Electric Probe
DC - Source Contactor 1 Pearson Contactor 2
Current 1 2
V Monitore
Hybrid
DUT Generator
Network Probe

Electric Electric
Contactor 1 Contactor 2

Ethernet
to DIO

Fig. 4. System diagram with key-components of the varistor characteristic test-bench

The voltage is measured by a passive probe and the current by a galvanically

isolated transformer. With four single impulses and increasing amplitudes it is possible
to interpolated the curve with a sufficiently accuracy in the limiting resistance area. The
interpolation between two V/I points is given by the equation:
logðI1i ÞlogðI2 Þ
þ logðV2 Þ
V1i ¼ e a ð2Þ

Therefore, the nonlinearity exponent a is determined between two surge records

and the current I1i is linearly interpolated into defined numbers of i-points. The
assembled curve is shown in Fig. 5. The points of interest are extracted from the data
and arbitrarily expandable. All results are compared and evaluated to special spot
measurement varistor test-equipment from a DIN EN ISO/IEC 17025 accredited lab-
oratory [4].
1442 S. Joerres et al.

Fig. 5. Result of characteristic curve measurement

In contrast to the standard IEC 61643-331 [1], a certain test value range such as
1 mA is not constant over a few milliseconds for a ramp-controlled source. This
requires a comparison to standard-compliant varistor test equipment consisting of the
determination of the varistor voltage and the leakage current. The varistor voltage is in
a direct comparison of a laboratory measuring instrument with measured values
between 540 … 590 V with an average deviation of 2.7 V with a corresponding
standard deviation 1r of 1.02 V (Fig. 6).

7
Average
Standard deviation σ1+
6 Standard deviation σ1-
Difference Varistor Voltage / V

Deviation
5

0
0 5 10 15 20 25 30 35
DUT

Fig. 6. Comparing measurement deviation

 Verification of the Aging Impact of MOV Ceramics 1443

In the case of leakage current measurement, the systematic deviation is substan-

tially higher, whereas the standard deviation is in a lower scatter range (Table 1). The
absolute measured values are in the range of 4 … 60 lA. The explanation for this is the
strong dependence of the measurement time on the leakage current absorption during
the measurement. A normative time specification does not exist. Despite the systematic
deviation, characteristic changes can be reliably measured.

Table 1. V/I characteristic parameter with comparison to varistor test equipment of DIN EN
ISO/IEC 17025 accredited laboratory
Parameter Systematic deviation average Std. dev.
VV @ 1 mA/V 2.73 1.02
IDC @ 450 V/µA 16.6 0.51

The investigation of the process stability with multiple measurements of the same
DUT is within the range of the measuring device tolerances and has an average
deviation of >±3r. This results in comparable results of deviations due to damage
despite a systematic error.

4 Performance and Results

The overall test procedure takes 58 s including recording and creating a test-report as
PDF, all test equipment raw-data as CSV, a log file as TXT and a summary of all points
of interest and additional information. The largest part is needed for charging the surge
generator voltage. With these defined test procedures it is possible to record the
characteristic of hundreds of varistor ceramics for research and building up statistically
correct data sets. One of the first results and explanation for inexplicable failures in
spite of passing the criterion of the varistor voltage at 1 mA is shown in Fig. 7. Five
D40K350 DUTs were recorded, which stands for: Disk type metalized, 40 = rated
diameter in mm, K = standard tolerance of VV at 1 mA with ±10% and a maximum
AC operating voltage of 350 V. The nominal current In is 20 kA with an 8/20 µs
waveform and the test procedure is performed 8 times with 25  In/2 and a repetition
rate of 60 s. Every step is measured and finally plotted with the initial and last test
procedure. It is shown that the point of no change move in the direction from 0.1…
10 mA during surge stress tests with current impulses under normative test environ-
ment conditions. Related recorded characteristics are published in [5]. This first test
already shows that it is important to continuously record the varistor curve rather than
performing a spot-point measurement.
1444 S. Joerres et al.

Fig. 7. Characteristics of D40K350 varistors in leakage current area without significant mA-
point change despite degradation

5 Conclusion

The presented fully automated test system for the recording of varistor V/I charac-
teristics combine two typical measuring methods. The leakage current area up to the
lower limiting resistance region is implemented by a powerful ramp-controlled direct
current source. A combination wave surge generator implements the upper limiting
resistance region including the recording of the clamping voltage. Combined, this
recorded data sets result in a nine decade current characteristic from µA to kA with its
resulting limiting voltage up to several kV. Due to the optimized test time and fully
automated evaluation, mass-tests of varistors are possible. This allows reliable evalu-
ations for degradation mechanisms through various aging tests. The accuracy of the
system is compared to customary varistor test equipment and confirmed in its accuracy.
First results show the advantages of a continuous curve recording over common spot
measurements. The informative value of such spot measurements for different stress
mechanisms has to be doubted as a single degradation indicator. Continuous recording
certainly helps to detect degradations which cause unexpected failures. Thus, the
change of degradation can be measured more precisely with the presented test-bench.
The acquired data sets help to understand the change of degradation by different
impacts and enables further research with statistically relevant numbers of varistors.
Due to the stacked varistor ceramic design in high-voltage surge arresters, the system
can be used and expanded for all network levels.
 Verification of the Aging Impact of MOV Ceramics 1445

References
1. IEC 61643–331 (2017-12): Components for low-voltage surge protective devices – Part 331:
Performance requirements and test methods for metal oxide varistors (MOV)
2. Finis, G., Schäfer, F., Reinbold, C., Claudi, A., Köhler, G.: Fully automated test system for
the thermal stability test of surge protective devices according to IEC 61643–1. In: 30th
International Conference on Lightning Protection (ICLP), Cagliari, Italy (2010)
3. Epcos, A.G.: Data Book SIOV Metal Oxide Varistors (2018)
4. Finis, G., Wetter, M., Durth, R., Depping, C.: Laboratory for the qualification-testing of SPDs
combining highest power performance parameters with unique fine adjustment possibilities.
In: International Conference on Lightning Protection (ICLP), Shanghai, China (2014)
5. Khanmiri, D.T., Ball, R., Mosesian, J., Lehman, B.: Degradation of low voltage metal oxide
varistors in power supplies. In: IEEE Applied Power Electronics, Northeastern University
Boston, USA (2016)
 Experimental Analysis and Suppression
for Very Fast Transient Disturbance
in Device Power Port During Switching
Operation in GIS Substations

Siqiao Ge and Weidong Liu(&)

State Key Laboratory of Power System, Department of Electrical Engineering,

Tsinghua University, Beijing 100084, China
gesq15@mails.tsinghua.edu.cn,
lwd-dea@mail.tsinghua.edu.cn

Abstract. Transient electromagnetic disturbance induced by disconnector and

circuit breaker operations in gas insulated switchgear (GIS) has a serious impact
on electronic equipment in substation secondary system. Typical forms of
transient electromagnetic disturbance are space electromagnetic radiation and
transient enclosure voltage (TEV) in GIS. Transient electromagnetic disturbance
can be easily coupled to the internal of electronic devices in many ways and one
of the coupling paths is power ports of electronic devices. This paper highlights
the coupling mechanism and characteristics of electromagnetic disturbance to
electronic equipment through its power port. Discharge in a SF6 gap is used to
simulate the complex electromagnetic process generated by switching opera-
tions in GIS, and then the coupled electromagnetic disturbance in power supply
ports of common field operation devices is measured in order to obtain the
coupling mechanism and characteristics of electromagnetic disturbance during
switching operations. Discharge experiment results show that the main fre-
quency components of electromagnetic disturbance in power supply ports are
1.2 MHz and 8.2 MHz, and meanwhile there are also 20–60 MHz and 120–
140 MHz frequency components. The relationship between power port coupled
disturbance and TEV on shielding box in test system is nearly linear. Further-
more, a deeply suppression isolation transformer is designed to decrease elec-
tromagnetic disturbance based on frequency characteristics of electromagnetic
disturbance, and the structure and installation method of the transformer are
given. Shielding performance tests are carried out to the transformer through
frequency sweep and discharge experiment. Results show the shielding effec-
tiveness of the transformer is up to −60 dB at the frequency components lower
than 20 MHz and −40 dB below 68 MHz, which demonstrates that the sup-
pression structure can protect the power port of secondary equipment in
substation.

Keywords: Switching operation
Space electromagnetic radiation
Transient

enclosure voltage (TEV)
Transient electromagnetic disturbance
Substation
secondary system

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1446–1457, 2020.
https://doi.org/10.1007/978-3-030-31676-1_135
 Experimental Analysis and Suppression for Very Fast Transient Disturbance 1447

1 Introduction

Disconnector and circuit breaker operations in gas insulated switchgear (GIS) may
induce very fast transient electromagnetic disturbance in power system substations.
During switching operations, spark discharges caused by breakdowns between switch
contacts continuously radiate high-frequency electromagnetic waves into space near
GIS, and meanwhile the traveling voltages wave induced by discharge pulses propagate
along GIS enclosure, and then refract and reflect when encountering the grounding
leads, forming TEV waveforms [1]. For example, the electric field strength of elec-
tromagnetic radiation can reach more than tens of kilovolts per meter and the TEV
magnitude can reach more than tens of kilovolts at ultra-high-voltage level, which may
incur serious impact on electronic devices in secondary systems [2, 3]. This is
becoming a critical factor for electromagnetic compatibility design in substations since
more and more secondary devices have been installed in primary equipment enclosures
[4].
The traditional power supply of electronic devices in primary equipment enclosures
is provided by substation which is grounded at far end. When very fast electromagnetic
disturbance occurs, the power line will receive space electromagnetic radiation, and
moreover the TEV will be transmitted through the power line, leading external dis-
turbance into electronic devices [5]. This is the main cause of internal interference in
electronic devices of GIS secondary systems, and may also cause secondary system
failure and power supply accident in substation.
Many studies have achieved progress in characteristics and suppression of the very
fast transient electromagnetic disturbance including space electromagnetic radiation
and TEV from experimental perspective [6–8]. Characteristic results, summarized by
CIGRE engaging in 72.5 kV–245 kV GIS substations, show that the magnitudes of
electric and magnetic field strength near GIS are 3 kV/m–10 kV/m and 30 A/m–
100 A/m, respectively, and the main frequency components distribute from 200 kHz to
3 MHz. The magnitudes of TEV reach 0.1–0.25 p.u., with the main frequency com-
ponents reaching 5–10 MHz [9, 10]. These studies are based on the characteristics of
disturbance, however we cannot specify the characteristics of disturbance after it passes
through the power port. As for suppression of disturbance, the current protection of
electronic devices includes using filters to absorb high frequency transient voltage
components, and using surge protectors to absorb voltage traveling wave on signal
transmission line of secondary equipment [11]. Most of the suppression methods are
based on cutting off the conducting path of electromagnetic disturbance on the power
line, however, electromagnetic disturbance of switching operations has high frequency
components. Thus, very fast transient disturbance could also couple to the internal of
electronic devices through radiation. The suppression effect of the devices on high
frequency components can only reach 30%–40% of that on low frequency components.
This paper aims to develop the internal coupling mechanism and characteristics of
electromagnetic disturbance through the power port of the electronic devices. This
paper measures the disturbance coupling characteristics in power supply ports of
electronic devices by simulating switching operations, and then a shielding protection
transformer for power supply is designed according to the interference characteristics.
1448 S. Ge and W. Liu

This research would help substations resolve electronic devices electromagnetic

compatibility problem.
The outline of this paper is as follows: Sect. 2 explains the practical power supply
of electronic devices in substation secondary systems, and analyzes coupling path of
very fast electromagnetic disturbance during switching operations, and a self-
developed power supply disturbance test circuit and measurement system is
designed. The experimental results and the characteristics of power supply disturbance
are laid out in Sect. 3. A deeply suppression isolation transformer is introduced to
decrease the level of power supply disturbance, and shielding performance tests is
carried out to verify its feasibility in Sect. 4. Section 5 concludes this paper.

2 Coupling of Electromagnetic Disturbance

and Measurement System

2.1 Common-Mode Interference Model of Electromagnetic Disturbance

to Electronic Devices
Very fast transient disturbances in substations can be coupled to the power supply ports
of electronic devices through space electromagnetic radiation and TEV conduction,
respectively, as illustrated in Fig. 1. Electronic devices on primary equipment enclo-
sures are protected by shielding, and the power supply is connected with isolation
transformer and filter. During switching operations, although the electronic device itself
is protected by the shield, the space electromagnetic wave can still radiate into the
shield body through the lead-out line of power supply. At the same time, due to the
existence of parasitic capacitance in isolation transformer and filter, the high frequency
component of TEV can still be transmitted to the shield by the power line. Therefore,
the radiation and conduction disturbance together form common mode interference on
power line.

Fig. 1. Schematic of substations common-mode interference model on power supply.

 Experimental Analysis and Suppression for Very Fast Transient Disturbance 1449

2.2 Test Circuit of Electromagnetic Coupling Characteristics

In order to obtain the coupling characteristics of very fast transient electromagnetic
disturbance on power port during switching operations, a self-developed test system
built in laboratory is shown in Fig. 2. The self-developed test system consists of
switching operation simulation, TEV measurement system and disturbance coupling
characteristics testing system.

Fig. 2. Simulation test system of disturbance coupling to power port.

The source of the test circuit includes a voltage regulator, a test transformer, a
current limiting resistor (1 MX), a high voltage diode, a SF6 discharge gas gap and a
capacitor (1 nF). The test transformer is connected to SF6 discharge gap, shielding box
and discharge inductance coil through rectifier diode and rectifier capacitance. When
SF6 gas gap breaks down as the voltage of the capacitor exceeds, high frequency
electromagnetic wave and transient enclosure voltage will be generated to simulate the
actual switching operation process.
Transformer with a shielding layer between its two coils is commonly used form of
electronic devices power supply. Connect two terminals of primary coil to ground, as
well as the shielding layer of transformer, and then measure the output of secondary
coil, which is the disturbance coupling to the power port under the common mode
interference, with data acquisition unit (oscilloscope, optical fibres and remote com-
puter) in shielding box (connected to an insulation platform). The measuring terminal
of the HV probe is connected to the ground wire, while the other side of the measuring
terminal is connected to the oscilloscope in the shielding box covered with shielding
enclosure. The outer shell of the HV probe is mounted on the shielding box, which
means the reference potential of the HV probe is shielding box potential. When the SF6
gap breaks down, the HV probe measures the voltage between the ground wire and the
shielding box, which reflects the level of transient enclosure voltage (TEV) on
shielding box. The frequency range of the measurement system is from quasi-dc to
250 MHz, the maximum sampling rate is 2.5 GS s-1, and the maximum measurable
voltages for transformer output and TEV are 200 V and 2 kV, respectively.
1450 S. Ge and W. Liu

3 Electronic Devices Power Port Disturbance Coupling

Characteristics

Experiments are carried out under different lengths of SF6 gas gap in order to change
the voltage level of TEV and power port disturbance coupling output. Three different
voltage level of measured results are recorded, and the typical measured waveforms are
shown in this section, upon which the statistical characteristics and the relationship
between TEV and power port coupled disturbance are then extracted and analyzed.

3.1 Experimental Waveforms of TEV and Power Port Coupled

Disturbance
The measured waveform of physical quantities tested in the experiment are shown in
Figs. 3 and 4, respectively.

Fig. 3. Experimental waveform of power port coupled disturbance.

Fig. 4. Experimental waveform of Shielding box TEV.

 Experimental Analysis and Suppression for Very Fast Transient Disturbance 1451

For each physical quantity, its overall waveform appears to be an oscillating

attenuation wave, and the main frequency components of the wave are determined by
the charging capacitor and discharge inductor (in Fig. 2) L-C oscillation circuit.
Moreover, higher frequency components can be found at the initial stage of each
measured voltage waveform, and these components form as a result of space electro-
magnetic radiation and the reflection and refraction of travelling voltage waves along
the shielding enclosure. As can be seen from Figs. 3 and 4, the oscillation start time of
disturbance coupling waveform is consistent with the one in TEV waveform, which
indicates that the disturbance in power port is caused by very fast transient disturbance
during gas gap discharge, and the maximum voltage in power ports reaches 150 V,
which will affect the electronic devices with this kind of power supply.
Different level of TEV and power port coupled disturbance can be obtained by
varying the lengths of discharge gas gap, and the amplitudes of the measuring voltage
are shown in Table 1. It can be seen that the relationship between power port coupled
disturbance and TEV is nearly linear, which indicates that severe interference will lead
into power port if the necessary suppression methods are not adopted when TEV level
is really high.

Table 1 Amplitude of TEV and power port disturbance in different voltage level.
Number TEV in kV Power port disturbance in V
1 3.5 54.1
2 7.54 135.6
3 14.5 195

3.2 Statistical Characteristics of Power Port Coupled Disturbance

The frequency spectrum of power port coupled disturbance single pulse in Fig. 3 is
carried out using the fast Fourier transform, shown in Fig. 5.

Fig. 5. Frequency spectrum of power port coupled disturbance.

1452 S. Ge and W. Liu

As drawn in Fig. 5, the disturbance coupling in power port contains abundant

frequency components. These components can mainly be classified into three cate-
gories. The first category consists of low-frequency components, usually around
1 MHz or lower. These components are determined by the source parameters and inner
structures of the testing circuit. The second one consists of high-frequency components,
usually from several MHz to above 10 MHz, and these components constitute the main
threat of the TEV. The last one consists of ultra-high-frequency components, which are
usually higher than dozens of MHz. These components are caused by reflections and
refractions of traveling voltage wave and the space electromagnetic radiation.

3.3 Comparison of Experimental Model Results with UHV GIS Results

A full scale test circuit with the practical 1100 kV GIS was established at UHV AC test
base in Wuhan [12]. The TEV statistical results show that the peak of TEV can be
achieved, with more than 16 kV at the end of GIS source side busbar, which shows that
the actual TEV level is the same with that in the experimental level. The main fre-
quency components of TEV in UHV GIS in Wuhan include 1.1 MHz caused by GIS
structure arrangement, and 5.9 MHz, 9.5 MHz, 20.9 MHz induced by the transient
voltage wave which are mostly multiple reflections and refractions in GIS, the
43.3 MHz and above caused by impedance changes outside GIS. Frequency distri-
bution results of laboratory tests in Fig. 5 are consistent with that in Wuhan UHV GIS
tests, which indicates the feasibility of laboratory model.

3.4 Relationship Analysis of Power Port Coupled Disturbance and TEV

in Frequency Domain
In theory, TEV and space electromagnetic radiation should give rise to power port
coupled disturbance together, as a result, the frequency domain distribution of TEV can
be used to observe the degree of coupling in power port, qualitatively.

40

20

0
Attenuation (dB)

-20

-40

-60

-80

-100
105 106 107 108
Frequency (Hz)

Fig. 6. Frequency spectrum of power port coupled disturbance.

 Experimental Analysis and Suppression for Very Fast Transient Disturbance 1453

The system coupling transfer function can be obtained, with the input of TEV and
the output of transformer coupling voltage. The transfer function is shown in Fig. 6. As
shown in Fig. 6, the traditional power supply mode of electronic devices can suppress
electromagnetic disturbance below 10 MHz, however, as the frequency of disturbance
increases, the suppression effect deteriorate gradually, and even when the frequency
reaches 20 MHz, the power supply amplifies the external disturbance, which indicates
that the traditional power supply mode of electronic devices will aggravate the coupling
of disturbance, hence it is essential to propose additional suppression method.

4 Deeply Suppression Isolation Transformer for Protection

of Electronic Devices Power Port

As it shown in Sect. 3, power port coupled disturbance includes abundant frequency

components above 1 MHz, and it will increase with the rise of TEV. Therefore, it is
necessary to develop an efficient suppression method to protect the power port. Since
the main forms of interference in power port are space radiation and TEV conducted by
parasitic capacitance of power isolation equipment, the main idea of shielding sup-
pression is to adopt an isolation device, which consists two parts. These two parts are
placed inside and outside of the shielding box respectively. Then the two parts are
perfectly connected to the shielding box, which ensures that the inside part is not
disturbed by external radiation, and at the same time, the parasitic capacitance between
the two parts is decreased by perfectly grounding. Thus, a deeply suppression isolation
transformer is designed based on the mentioned idea.

4.1 Structure of Deeply Suppression Isolation Transformer

The structure and installation of deeply suppression isolation transformer is shown in
Fig. 7.

Fig. 7. Structure and installation of deeply suppression isolation transformer.

1454 S. Ge and W. Liu

Two winding coils of the transformer are placed inside and outside of the shielding
box separately, and a shielding layer is installed on the shielding box between two
windings. A terminal of the secondary coil is connected to the shielding lay, thus the
parasitic capacitance between two windings can be deeply reduced through the
installation method. It is necessary to make a slit in the shielding layer between two
windings, which will not only ensure that the shielding layer can weaken the influence
of external high frequency electromagnetic wave, but also enables the transformer to
work normally without short circuit. Use the secondary winding coil connected with a
filter inside the shielding box as the power supply for a practical electronic device.

4.2 Measuring Results of Deeply Suppression Isolation Transformer

Replace the traditional power supply in Fig. 2 with the deeply suppression isolation
transformer, and the shielding properties of the transformer can be obtained. Figure 8
shows the output of the isolation transformer in the shielding box, and Fig. 9 shows the
ground potential near the shielding box.

Fig. 8. Waveform of deeply suppression isolation transformer secondary output.

Fig. 9. Waveform of ground potential near the shielding box.

 Experimental Analysis and Suppression for Very Fast Transient Disturbance 1455

From Fig. 8, we can see that the magnitude of transformer output voltage is 20 V,
which is much less than that in Fig. 3, and at the same time, the trend of ground
potential in Fig. 9 is consistent with Fig. 4 when the SF6 gap breaks down, which
illustrates the effectiveness of suppression transformer in electronic devices electro-
magnetic compatibility during practical switching operations.

4.3 Shielding Properties and Effectiveness of Deeply Suppression

Isolation Transformer
Figure 10 presents the frequency spectrum of deeply suppression isolation transformer
secondary output. Compared with Fig. 5, it can be noted that the frequency compo-
nents of transformer output below 1 MHz is still very small, while the higher frequency
components between 1 MHz and 68 MHz is weakened by its shielding property, and
the shielding effectiveness reach −40 dB. As for frequency components above
100 MHz, the suppression effect is less than that at lower frequency components,
which is due to the slit on shielding layer of transformer, however, the shielding
effectiveness is up to −20 dB and still sufficient for electromagnetic disturbance pro-
tection in substations. Therefore, isolation transformer installed before the power
supply of the electronic device in substations can suppress the electromagnetic dis-
turbance caused by switching operations.

Fig. 10. Frequency spectrum of deeply suppression isolation transformer secondary output.

The two-port transfer function of this deeply suppression isolation transformer can
be obtained based on ground potential and transformer secondary output in frequency
domain.
As shown in Fig. 11, deeply suppression isolation transformer can restrain external
disturbance in full frequency band of the test system. For frequency components of
disturbance below 20 MHz, the suppression effectiveness can reach −60 dB, and for
frequency components lower than 68 MHz the suppression effectiveness is up to
−40 dB, and for frequency components lower than 200 MHz it can still reach −20 dB.
1456 S. Ge and W. Liu

As the fact that the frequency components of switching operations electromagnetic

disturbance in substations are mainly below 200 MHz, the deeply suppression isolation
transformer can basically shield the disturbance coupling to the power supply port of
electronic devices.

-20
Attenuation (dB)

-40

-60

-80

-100

-120
105 106 107 108
Frequency (Hz)

Fig. 11. Transfer function of deeply suppression isolation transformer.

5 Conclusions

In this paper, the harm of electromagnetic disturbance caused by switching operation in

GIS substation to electronic devices in substation is analyzed and the necessity to
protect the power supply of electronic devices is pointed out.
Based on the simulation test system, the power port coupling disturbance waveform
of the electronic device during switching operation is measured and its coupling
characteristics are analyzed.
The structure and installation method of a deeply suppression isolation transformer
are introduced, and its disturbance coupling characteristics are tested. The results show
that the isolation transformer can effectively restrain the high frequency components
between 10 MHz and 100 MHz of disturbance and help the electromagnetic compat-
ibility of substation electronic devices.

References
1. Shu, Y., Chen, W., Li, Z.: Recent progress in investigations on very fast transient
overvoltage in gas insulated switch-gear. IEEE Trans. Power Deliv. 28(1), 458–466 (2013)
2. Cai, Y., Guan, Y., Liu, W.: Statistical characteristics of transient enclosure voltage in ultra-
high-voltage gas-insulated switchgear. Plasma Sci. Technol. 19(6), 74–81 (2017)
3. Cai, Y., Guan, Y., Liu, W.: Experimental study on the relationship between transient
enclosure voltage and very fast transient overvoltage. In: 19th International Symposium on
High Voltage Engineering, Pilsen, Czech Republic (2015)
 Experimental Analysis and Suppression for Very Fast Transient Disturbance 1457

4. Zhang, B., Cui, X., Zhao, Z.: Numerical analysis of the influence between large grounding
grids and two-end grounded cables by the moment method coupled with circuit equations.
IEEE Trans. Power Deliv. 20(2), 731–737 (2005)
5. Narimatsu, S., Yamaguchi, K., Nakano, S.: Interrupting performance of capacitive current by
disconnecting switch for gas insulated switchgear. IEEE Trans. Power Apparatus Syst. PAS
100(6), 2726–2732 (1981)
6. Cai, Y., Guan, Y., Liu, W.: Study of transient enclosure voltage coupling to secondary
cables in a gas-insulated substation. IEEE Trans. Power Deliv. 33(2), 761–768 (2018)
7. Ari, N., Blumer, W.: Transient electromagnetic fields due to switching operations in electric
power systems. IEEE Trans. Electromagn. Compat. 29(3), 233–237 (1987)
8. Thomas, D., Wiggins, C.: Prediction of electromagnetic field and current transients in power
transmission and distribution systems. IEEE Trans. Power Deliv. 4(1), 744–755 (1989)
9. Kostin, M., Matveyev, M., Zhivodernikov, S.: Some results of EMC investigations in
Russian substations. In: CIGRE, Paris, France (2002)
10. Imposimato, C., Hoeffelman, J., Eriksson, A.: EMI characterization of HVAC substations -
updated data. In: CIGRE, Paris, France (2002)
11. Ma, G., Li, C., Quan, J.: Measurement of VFTO on transformer entrance with transformer
bushing sensor. Proc. CSEE 30(33), 122–128 (2010)
12. Hu, R., Cui, X., Zhang, W.: Transient enclosure voltage measurement system of UHV GIS
and TEV statistical characterization. In: International Symposium on Electromagnetic
Compatibility, EMC Europe (2012)
 Efficient Lightning Protection
of a Gas Insulated Substation Replacing
an Air-Insulated Substation

Constantin Balzer1(&), Volker Hinrichsen1,2, and Wolfgang Tausend2

1
TU Darmstadt High Voltage Laboratories,
Fraunhoferstr 4, 64283 Darmstadt, Germany
balzer@hst.tu-darmstadt.de
2
Transnet BW, Osloer Straße 15-17, 70173 Stuttgart, Germany

Abstract. This contribution highlights a procedure that aims at assessing the

different possibilities to improve the lightning performance of an overhead line,
connected to a substation that will be transformed from air- to gas-insulated.
Therefore, the probability of a shielding failure was first calculated for each span
width, using the simplified electro-geometrical model. Then, the overvoltages
arising from these shielding failures were calculated using transient simulations.
Special attention was paid to the implementation of the dynamic flashover
characteristics of the insulators as well as corona-damping effects, as they will
significantly influence the outcome of the calculations. It was found that both, a
reinforced ground wire protection as well as the installation of line surge
arresters at the first tower can drastically reduce the probability of overvoltages
within the GIS. If the first option is chosen, an efficient length can be chosen
with the help of the calculated probability density functions, sorted after the
point of origin of the lighting strike.

Keywords: Statistical insolation coordination
Lightning protection

1 Introduction

In comparison to air-insulated substations (AIS), gas-insulated substations (GIS) re-

quire an enhanced overvoltage protection with respect to incoming travelling waves
originating from shielding failures on overhead lines. This can be achieved by a
reinforced ground wire protection of overhead lines in the vicinity of the substation, as
it is often best practice for TSOs so far.
However, if an AIS is converted into a GIS, an enhancement of the shielding
performance of the lines would entail considerable mechanical reinforcements of the
corresponding line towers. As these measures are both costly and – due to the nec-
essary legislative approval procedure – significantly time consuming, there is a high
interest in alternative solutions.
Against this background, two questions arise: Firstly, the efficiency of the sup-
plementary ground wire protection should be examined as a function of the covered
distance in order to find a reasonable (both economically and technically sufficient)
length. And secondly – due to the aforementioned difficulties in changing the tower
© Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1458–1470, 2020.
https://doi.org/10.1007/978-3-030-31676-1_136
 Efficient Lightning Protection of a Gas Insulated Substation 1459

geometry – it is the question whether the substation can be sufficiently protected by

omitting the double ground wire protection, rather than installing supplementary
arresters on the line instead.
Consequently, a statistical insulation coordination procedure according to the
standard IEC 60071-1 [1] and IEC 60071-2 [2] is performed in order to evaluate the
risk of overvoltages inside a GIS with and without the elaborate lightning protection on
the connected overhead lines. For this purpose, the probability of a shielding failure is
first calculated, using the tower geometry and the simplified electro-geometric model
that is proposed by Cigré [3]. In a second step, detailed transient calculations are
performed in order to determine the resulting overvoltage caused by a shielding failure.
The combination of the probability of a shielding failure with its resulting over-
voltage finally gives the desired overvoltage probability density function (also denoted
as “lightning performance”). As this is done separately for each tower, it allows the
correlation of the possible overvoltage with the location of the shielding failure. This
makes it possible to determine an efficient length of the supplementary ground wire
protection. By regarding the resulting cumulative probability functions, the impact of a
double ground wire protection or – alternatively – supplemental arresters on the line
can easily be evaluated.

2 Studied Configurations

Three different overhead lines (two of them already connected to a GIS) with Um =
420 kV have been studied: One line with a double ground wire protection all along its
length (in the following referred to as configuration “A”), and one with double ground
wire protection on its first three towers (denoted configuration “B”). The lightning
performances of these two lines are compared to the corresponding performance of one
line connected to a station that is presently air-insulated but will be converted into a
gas-insulated station (configuration “C”). At the moment, this line is protected only by
a double ground wire on the first span length between the station and the first tower.
The gas-insulated systems are connected directly at the station entrance together
with an air insulated surge arrester. Worst-case assumptions regarding the expected
overvoltages have been formulated, assuming an open end inside the GIS without any
branching as this leads to a reflection factor of 1 of the incoming surge at the end of the
GIS. For configurations “A” and “B” a length of 40 m between the beginning of the
GIS and its open end has been assumed. As the station of configuration “C” is quite
ample, the maximum length between the beginning and the open end of the GIS was
supposed to be 100 m. The overvoltages are evaluated at the open end of the outer,
upper phase of the towers, as it is most exposed to possible shielding failures.
With respect to possible improvements of the lightning performance of configu-
ration “C”, two measures have been included into the study: The extension of the
double ground wire protection to the second tower as well as the installation of line
surge arresters on the first tower after the station. The installation of gas-insulated
arresters was not taken into consideration.
1460 C. Balzer et al.

3 General Approach

As mentioned before, the calculation of the lightning performance necessitates two

main calculations: The determination of the probability of shielding failures and the
evaluation of the overvoltages from shielding failures. Whereas the latter are found
with the help of transient simulations, the shielding failure probability is derived with
the help of the simplified electro-geometric model. In the following, both procedures
are presented in detail.

3.1 Probability of a Shielding Failure

In order to calculate the probability of direct lightning strikes, the simplified electro-
geometric model provides a tangible, robust and well approved method, which is based
on two assumptions:
– A downward leader approaches earth strictly vertically
– If the distance between the downward leader and an object falls below a specified
value – called the striking distance – the lightning will strike into this object.
In general, it may be assumed that the striking distance r is a function of the
lightning peak current IF with:

r ¼ A
IFb ð1Þ

Following the recommendation of Cigré [3], the empirical relation according to

Brown and Whitehead was used, which distinguishes between the striking distance rc
to a phase and rg to ground:

rc ¼ 7:1
IF0:75 ð2Þ

rg ¼ 6:4
IF0:75 ð3Þ

Figure 1 illustrates the application of the model for one specific value of IF: The
simplified geometry consists of one phase wire at the height h1 (in red) and one ground
wire at the height h2 (in black). First, circles with the radius of rc around the phase and
ground wires are drawn, together with the horizontal line in a distance of rg to ground.
From the intersections of the circles and the line parallel to ground, the lateral distances
Dg and Dc for the phase under consideration can be determined. The distances Dg and
Dc then mark the area from which a downward leader produces a strike to ground or a
shielding failure, respectively.
By repeating the aforementioned procedure for different current values, the lateral
distance Dc as a function of the lightning peak current can be derived. As an example,
Fig. 2 shows the result for the outer, upper phase at the first tower of configuration C as
a solid line. As the phase is protected by only one ground wire, it can be stated that the
resulting distances are very high, with a maximum shielding failure current (the point
of intersection of the function with the abscissa) of Is,max = 62 kA. Installing a double
 Efficient Lightning Protection of a Gas Insulated Substation 1461

Fig. 1. Example of the simplified electro-geometrical model

Fig. 2. Lateral distances Dc as a function of the lightning peak current for the outer, upper phase
at the first tower in configuration C. Black: One single ground wire; Red: Double ground wire
protection

ground wire would result in a considerable reduction of the lateral distances as well as
the maximum shielding failure current, as it can be conceded from the dotted line in
Fig. 2.
In order to transform the values of Dc(IF) into the sought probabilities, the lateral
distances have to be multiplied by the length of the line in order to obtain the surfaces
Ac(IF). With respect to the changing tower geometries, the evaluation was not done in
three dimensions (see [5]), as this would increase the complexity of the calculations and
lead to less transparency of the results. Hence, Ac(IF) was calculated for each tower,
with the length corresponding to half the span width before and after the tower under
1462 C. Balzer et al.

consideration. Also, the line sag was neglected, resulting in an approximation on the
safe side, as the shielding failure current decreases with the height of the phase wire.
Afterwards, the flash density of a strike with the corresponding amplitude has to be
determined. This is done with the help of the overall ground flash density Ng and the
value of the log-normal distribution of current amplitudes [3]. In this case, only first
strokes had been considered, so that the probability density function can be expressed
as:

1 lnðIF =M Þ
eðZ =2Þ ; Z ¼
2
f ðIF Þ ¼ pffiffiffiffiffiffi ð4Þ
2p
b
IF b

The values of the Median M and the lognormal standard deviation b can be taken
from Table 1, and the value of the ground flash density was set to:

3
Ng ¼ ð5Þ
a
km2
which is a rather conservative assumption for middle Europe.

Table 1. Parameters for the lognormal distribution in Eq. (4).

IF < 20 kA IF > 20 kA
M 61 kA 33.3 kA
b 1.33 0.61

Consequently, the density function of the shielding failure currents Is at tower

m can be calculated with:

  Lsp; m1 þ Lsp; m þ 1

f Is;m ¼ Ng
f ðIF Þ
Dc;m ðIF Þ
ð6Þ
2
where Lsp,m-1 and Lsp,m+1 denote the span lengths to both sides of the tower.

3.2 Determination of Overvoltages

In order to assess the impact of a shielding failure, calculations of the resulting
overvoltage at the end of the GIS have been performed with the help of the transient
analysis program ATP/EMTP. The lightning strike itself was modelled as a current
source, following a concave curve proposed by Cigré [3]. The time to half value was
considered constant at th = 77.5 µs, the maximum steepness Sm and the front duration
td,30 were set as the correlated median to the peak current IF following the relations in
Table 2.
 Efficient Lightning Protection of a Gas Insulated Substation 1463

Table 2. Correlation between the maximum steepness Sm and the time to crest of the lightning
current td,30 with the amplitude IF
IF < 20 kA IF > 20 kA
Sm in ls 12:00
IF 6:50
IF0:376
kA 0:171

td;30 in ls 1:77
IF0:188 0:91
IF0:411

Lightning strikes of peak currents from 3 kA to 100 kA have been modelled at the
first seven towers on the line of each configuration. The maximum overvoltage in each
case at the open end of the GIS was then used to create the relation between the peak
current of the lightning strike (at tower m) and the maximum overvoltage, which allows
converting the probability distribution of shielding failure currents in Eq. (6) to a
probability distribution of overvoltages:
    
f Is;m ! f Umax;m Is;m ð7Þ

In excess of the existing models of overhead lines, surge arresters and transmission
line towers [4], two models have been developed to implement the dynamic flashover
characteristics of the line insulators as well as the effect of corona damping. The
importance of these two phenomena with respect to the resulting simulation output will
be highlighted in the following two sections.
Implementation of a Dynamic Flashover Model
As high amplitudes of the lighting current IF are correlated with a high steepness, the
flashover characteristic of the line insulator is no longer sufficiently described by its
statistical withstand voltage U10. This is due to the voltage-time area law, which yields
considerably higher flashover voltages for front times in the range of microseconds.
Therefore, a leader propagation model has been implemented based on the rec-
ommendations in [3]. It assumes that the velocity vL of the leader is proportional to the
difference of the resulting field strength within the air gap and a constant field strength
E0. Hence:
 
d‘L uð t Þ
vL ¼ ¼ k
uðtÞ
 E0 ð8Þ
dt d  ‘L

where:
vL: Velocity of the leader in m
s−1
‘L : Length of the leader in m
u(t): Voltage across the insulator in kV
d: Air gap of the insulator in m
k: Proportionality factor in m2
V−2
s−1
E0: Constant field strength in kV
m−1
1464 C. Balzer et al.

Equation (8) is evaluated if

uðtÞ  d
E0 ð9Þ

and the length ‘L updated in every time step of the simulation. Flashover occurs if:

‘L  d ð10Þ

For the two parameters k and E0, the values for negative voltage stresses on long
rod insulators were chosen [3]:
m
k ¼ 106 ð11Þ
V2
s
kV
E0 ¼ 670 ð12Þ
m

Figure 3 shows the simulated voltages on a phase at the point of impact (in the
vicinity of an insulator with an air gap of d = 2.95 m) for peak current amplitudes from
3 kA to 20 kA. Because Eq. (9) amounts to:

d
E0 ¼ 1977 kV ð13Þ

no flashovers will occur for peaks voltages below 1.9 MV, which is quite a conser-
vative assumption.

Fig. 3. Simulated voltages on the examined phase at the point of impact of lightning strikes with
the implemented dynamic flashover characteristic of the line insulator. Current amplitudes are
varied from 3 kA to 20 kA
 Efficient Lightning Protection of a Gas Insulated Substation 1465

Implementation of Corona Damping

In addition to the flashover characteristics of the insulators, the damping of voltage
surges due to corona effects is the second important phenomenon that needs to be
dynamically modelled within the simulation.
In order to do so, it is helpful to first recall the physical effects of corona: In general,
it leads to the formation of space charges around the conductor, so that the capacitive
current in the phase can be formulated as

duðtÞ dqc ðtÞ

iðtÞ ¼ Cgeo
þ ð14Þ
dt dt
where:
Cgeo : Geometric capacitance per unit length in F
m−1
u: Line voltage in kV
qc : Space charge as a function of time per unit length in As
m−1
As the space charge depends on the voltage itself, the chain rule of derivation can
be applied, leading to:

dqc ðtÞ dQc ðuðtÞÞ duðtÞ duðtÞ

¼
¼ Cdyn ðU Þ
ð15Þ
dt duðtÞ dt dt

with:
Qc : Space charge as a function of line voltage in A
s
Cdyn ðUÞ: Dynamic corona capacitance at voltage U
Now Eq. (14) can be reformulated as:

  duðtÞ
iðtÞ ¼ Cgeo þ Cdyn ðU Þ
ð16Þ
dt

Hence, implementing corona phenomena into the simulation will result in mod-
elling the supplementary voltage dependant capacitance Cdyn. As the interference
between the different phases can be omitted, Cdyn is branched directly between the
affected phase and ground. Moreover, it was implemented as a lumped element every
300 m, which proved to be a sufficient approximation with respect to shorter lengths.
The dynamic corona capacitance is often confounded with the expression of the
corona capacitance. The latter being defined as the static relation between the space
charge and the line voltage:

Q c ðU Þ
Ccor ¼ ð17Þ
U
1466 C. Balzer et al.

one may write:

dQc ðuðtÞÞ dðCcor ðU Þ
uðtÞÞ

¼
duðtÞ duðtÞ
ð18Þ
dCcor ðU Þ
¼ Ccor ðU Þ þ uðtÞ

dU
so that the relation between the corona capacitance and the dynamic capacitance is:

dCcor ðU Þ
Cdyn ðU Þ ¼ Ccor ðU Þ þ uðtÞ
ð19Þ
dU

According to [3], the corona capacitance can be assumed as linearly dependent on

the voltage, so that:

Cc ¼ Ci þ K
ðU  Ui Þ ð20Þ

with:
Ci : Capacitance per unit length at the inception of corona in A
s
m−1
K: Corona constant in pF
kV−1
m−1
Ui : Corona inception voltage in kV
Inserting Eq. (20) into (19) leads to:

Cdyn ¼ Ci  K
Ui þ 2
K
U ð21Þ

With respect to the initial capacitance, [3] proposes a value of

pF
Ci ¼ 3 ð22Þ
m
and – dealing with quad conductor bundles – for the corona constant:

pF
K ¼ 2:9
103 ð23Þ
m
kV

The corona inception voltage is determined with the help of the relation between
the corona inception field strength and the conductor radius:
 
  0:301 kV
~ 
Ei ¼ 2460 1 þ ð24Þ
rcon m

with:
rcon : Radius of a subconductor in m
Equation (24) is then inserted into the relation for the surface field strength of
bundle conductors:
 Efficient Lightning Protection of a Gas Insulated Substation 1467

  lnðd=rB Þ
Ui ¼ n
~
Ei 
 ð25Þ
1
rcon þ n1
2
rT

where:
n: Number of conductors in a bundle
d: Distance between the phases in m
rT : Pitch circle radius of the conductor bundle in m
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rB : Equivalent radius, rB ¼ n n
rcon
rTn1
Figure 4 illustrates the modelled effect of corona damping after 900 m of length on
two overvoltage surges, following a direct lightning strike on a phase: one full wave as
well as one chopped wave as a consequence of an insulator flashover.

Fig. 4. Simulated voltages on the examined phase at the point of impact of lightning strikes as
well as after 900 m length with the implemented dynamic flashover characteristic and corona
damping model

With regard to the alteration of the original voltage surge, it is obvious that the
implementation of corona effects reduces the calculated overvoltages due to two
effects:
– For non-chopped voltage surges, the steepness is largely reduced, thus increasing
the protection distance of the arresters
– For chopped voltage surges, the reduction in steepness leads to a considerable
reduction in amplitude
1468 C. Balzer et al.

4 Results

Evaluation of Eq. (7) for each tower and the representation of the outcome in the form
of a histogram yields the discrete probability distribution density as it is shown in
Fig. 5 for configuration C. The tower number is increasing with the distance from the
station, tower number one being the first tower after the station.

Fig. 5. Calculated probability function of overvoltages at the open end of the GIS in
configuration C. The numbers indicate the tower next to the point of strike. Tower numbers count
from the substation (i.e. number one = tower next to the station).

The representation in Fig. 5 reveals two aspects: First, it basically illustrates the
impact of corona damping (and that its implementation in a transient simulation is
mandatory), as the distribution of overvoltages for each single tower shifts towards
voltages that are lower the larger the distance to the station. Based on this effect, it
secondly allows answering the question of an “ideal” length of a possible double
ground wire protection: By fixing a “target maximum overvoltage”, the tower, up to
which a reinforced ground wire system is necessary, can readily be deduced.
For the purpose of comparing the lightning performance of the three different
configurations, it is useful to build the cumulative function of the distributions shown in
Fig. 5. This yields the curves in Fig. 6.
As it was expected, configuration A has the best lightning performance, with a risk
of only 10−3 per cent and year in exceeding an overvoltage of 800 kV per year,
whereas configuration C has the same risk for overvoltages up to 1.3 MV. The
restricted ground wire protection of configuration B echos nicely in the cumulative
 Efficient Lightning Protection of a Gas Insulated Substation 1469

Fig. 6. Calculated cumulative probability function of overvoltages at the open end of the GIS
for the three configurations A, B and C as well as for configuration C with line surge arresters on
the first tower (“C1”) and configuration C with a double ground wire protection between the first
two towers (“C2”).

probability function: Regarding high overvoltages (whose origin is a strike in the

vicinity of the station), it resembles configuration A, but the lower the overvoltage (i.e.
the longer the distance to the station, where configuration B is protected only by a
single ground wire) it approaches the risk of configuration C.
Moreover, the impact of a double ground wire protection on the line span between
the first and the second tower as well as the installation of line surge arresters is
included in Fig. 6. Although the double ground wire protection strongly reduces the
possibility of overvoltages higher than 1.1 MV, it is not totally eliminating the risk, as
the shielding angle in the assumed tower geometry is not as low as for configuration A
or B. However, the installation of line surge arresters can approach the lightning
performance of configuration C close to the one of configuration A or B without
extensive constructional change of the towers.
At last, it should be noted that the calculated probabilities in Fig. 6 are only valid
with respect to the time that the mentioned “worst case” configuration of the GIS lasts
(i.e. an open end without a connected line in parallel). If at all, this may only be the case
for a restricted time during the year. Strictly spoken, the functions in Fig. 6 should be
weighted by the duration of the worst case condition, leading to a further reduction in
overvoltage probabilities for the station in total.
1470 C. Balzer et al.

5 Conclusion

The proposed procedure is a useful tool in assessing the improvement of lightning

protection for gas-insulated substations. Based on the resulting cumulative probability
functions in Fig. 6, it can be seen that both, installation of line surge arresters at the first
tower next to the station or reinforcing ground wire protection, can strongly improve
the lightning performance of a formerly air-insulated substation. If the latter option is
chosen, an effective length of the reinforcement can readily be deduced by the prob-
ability density functions such as the one in Fig. 5, as the magnitude of the overvoltages
decreases with distance between the station and the point of a possible shielding failure.

References
1. IEC 60071-1 Edition 8.0 (2006-01) + Amendment 1 (2009-09): Insulation co-ordination –
Part 1: Definitions, principles and rules
2. IEC 60071-2 Edition 4.0 (2016-03): Insulation co-ordination – Part 2: Application guide
3. CIGRE WG 33.01: Guide to procedures for estimating the lightning performance of
transmission lines, CIGRE Technical Brochure No. 63, Paris (1991)
4. IEC TR 60071-4: Insulation co-ordination – Part 4: Computational guide to insulation co-
ordination and modelling of electrical networks
5. Horváth, T.: The protected space proved to be an undefined term. In: 2012 International
Conference on Lightning Protection (ICPL), Vienna, Austria, 2–7 September 2012 (2012)
 Lightning Performance of an Air Insulated
High Voltage Substation

C. A. Christodoulou, V. Vita, and L. Ekonomou(&)

Department of Electrical and Electronic Engineering Educators,

A.S.PE.T.E. - School of Pedagogical and Technological Education,
N. Heraklion, 14121 Athens, Greece
leekonomou@aspete.gr

Abstract. Lightning strokes in power systems is an issue of high importance

due to the developed overvoltages that may cause serious damages to equipment
and power supply interruptions, imposing an additional cost to the power utility
because of the undelivered energy. The current work deals with the lightning
performance of air-insulated high voltage substations. The study of the conse-
quences due to an atmospheric surge and the efficient design of the protection
system against external ovevoltages are discussed. In the current work, different
overvoltage protection schemes for an air insulated 150/20 kV substation are
examined, considering the electrical characteristics and the placement of the
surge arresters, the number of the connected overhead high voltage lines, the
lightning hit position and the tower footing resistance, in an effort to propose an
optimal protection configuration.

Keywords: High voltage substations
Lightning performance
Surge arresters

1 Introduction

Atmospheric surges can cause several damages to a high voltage/medium voltage

substation, stressing the insulation of the various components of the substation. If the
arising overvoltage gets greater than the insulation level of the equipment a breakdown
happens, leading to sever damages and unscheduled disruption of the power supply.
Power transformers, circuit breakers, switching devices, voltage and current mea-
surement transformers, high voltage and medium voltage cables (if any) are the basic
components of a typical high voltage/medium voltage substation that are affected by
potential lightning strokes and their damage results in abnormal operation of the power
system and dangers for human beings. Even if the substation is fully protected against
direct hits, in the case that a lightning flash strikes on one of the connected high voltage
lines, then a shielding failure or a backflashover may happen; the harshness of the
consequences relies on the position of the overhead protective ground wires, the peak
current of the lightning flash and the value of the grounding resistance (both of the
substation and the towers). An insulation fault generates a travelling wave that is
directed to the substation and stresses the insulation of the equipment of the substation,
resulting in thermal and electromechanical effects.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1471–1481, 2020.
https://doi.org/10.1007/978-3-030-31676-1_137
1472 C. A. Christodoulou et al.

The above analysis highlights the necessity for the efficient protection of a high
voltage substation against lightning overvoltates to mitigate outages due to atmospheric
phenomena and restrict the repairing or/and replacement costs for the damaged
equipment. Note that the stochastic nature of the lightning flashes and several techno-
economic parameters of a substation must be considered during the design of the
lightning protection system. Moreover, the lightning hit position, the configuration of
the external lightning protection system, the grounding system and the basic insulation
level are parameters that determine the efficacy of the lightning protection system. The
adoption of higher insulation level and the installation of ground wires and masts are
the most widely practices to protect a substation against lightning overvoltages. Added
to these, the installation of surge arresters, especially in regions with high soil resis-
tivity can enhance the lightning performance of the high voltage substations [1, 2].
In the current work, different overvoltage protection schemes for an air insulated
high voltage (150/20 kV) substation will be analysed, considering the placement of the
surge arresters, the number of the connected overhead lines and the grounding resis-
tance of the towers, in an effort to propose an optimal protection configuration.

2 Protection of High Voltage Substations Against Direct

and Incoming Lightning Overvoltages

High-voltage substations consist a critical link in the power transmission chain between
the power stations and the customers. Considering the insulation of the substation,
three basic configurations are possible [3]:
• Air-Insulated Substations (AIS): they are used wherever space restrictions and
environmental limitations are not strict. The exposure to environmental effects is the
main drawback of the AIS.
• Gas-Insulated Indoor or Outdoor Switchgear (GIS): their dimensions and their
compact design render them ideal in the case that overhead transmission lines are
not appropriate (e.g. urban and/or industrial areas).
• Mixed Technology: they include both Air-Insulated and Gas-Insulated components.
The severity of the consequences of a lightning hit to a high voltage/medium
voltage substation adds pressure to power utilities to adopt the essential protection
measures, in an attempt to mitigate equipment damages and power supply interruption.
The design of a lightning protection system obeys to national and international stan-
dards and follows the following appropriate methodologies:
• fixed angles method: uses vertical angles to adjust the number, the position, and the
height of the ground wires or the masts (Fig. 1) [4, 5].
 Lightning Performance of an Air Insulated High Voltage Substation 1473

Fig. 1. The fixed angles method [4, 5].

• empirical curves method: it is based on field and laboratory outcome to determine

lightning protection system configuration [4, 5].
• electrogeometrical (EGM) method: The EGM model is based on the striking dis-
tance that is affected by the lightning hit position and peak current of the lightning
current. Figure 2 depicts the implementation of the EGM method for a typical high
voltage substation, by applying the rolling sphere method (the same striking dis-
tances to the ground, a ground wire or a mast are assumed). The parts of the
equipment that remain below the curved surface of the imaginary sphere (the radius
of this sphere S is equal to the striking distance), that rolls over the surface of the
substation, are protected, in contrast with the parts that touch the imaginary sphere
[4, 5].

Fig. 2. Principle of the rolling sphere [4, 5].

1474 C. A. Christodoulou et al.

Nevertheless, the development of lightning surges cannot be totally avoided when a

high voltage/medium voltage substation is connected with overhead high voltage lines.
A lightning can hit either the phase (direct stroke) or the shield wire (backflashover) of
an overhead transmission line, resulting in overvoltages of great magnitude that exceed
the insulation level of the line; in this case, a voltage surge is developed that propagates
along the phase conductor to the substation and may damage the equipment of the
substation. To avoid the repercussions of a potential lightning stroke, the peak of the
developed overvoltages should be restricted in a way that the overvoltage can be below
the insulation strength of the equipment [6].
For the efficient protection of the substation the accomplishment of low grounding
resistances is an issue of priority. In general, main scope of the grounding system is to
keep the developed touch and step voltages in and around the yard of the substation
within the safety limits [7]. A well designed grounding system is requirement for the
safety of the equipment as well as the people working in the substation, since it diverts
the surge currents to ground, avoiding to exceed the dielectric strength of the substation
devices; at the same time, grounding system provides protection to personnel against
the repercussions of electric shock under fault conditions [5].
Except from the above, the installation of arresters at various positions of the
substation definitely contributes to the enhancement of the lightning performance of the
substation. Surge arresters play a key role in substations for restricting external and
internal overvoltages, ensuring that no insulation breakdown will occur. The appro-
priate placement of the arresters plays fundamental role, since voltage surges behave as
travelling waves [8, 9]. Nowadays, metal oxide (ZnO) gapless surge arresters, instead
of gapped SiC ones, are used that absorb negligible current in the normal operation of
the network and very high current (kA range) when faced with lightning surges [10,
11].

3 System Configuration and Simulation Models

Figure 3 depicts the AIS under examination. The power transformer, the busbars, the
circuit breakers, the disconnect switches, the ground switches, the current and voltage
transformers are the main components of a typical 150/20 kV AIS that have to be
protected against incoming surges.
 Lightning Performance of an Air Insulated High Voltage Substation 1475

incoming
transmission lines

high voltage bus

circuit breaker
disconnector swich
power transformer
40/50MVA
150/20kV

medium voltage bus

Fig. 3. AIS under examination.

Figure 4 depicts a typical 150 kV transmission line, consisting of ACSR conduc-

tors and glass or ceramic insulators. Table 1 contains the electrical characteristics of the
transmission lines as well as the medium voltage distribution lines that depart from the
medium voltage busbars.
The ground wires, installed according to the EGM, intercept the majority of the
lightning flashes. The towers are modelled according to equation [12]:
 
r 
1
Zr ¼ 60 ln cot tan1 ð1Þ
2 h

where r and h are the base radius and the height of the tower in m.
1476 C. A. Christodoulou et al.

Fig. 4. Configuration of a typical 150 kV tower (A: −4.8 m, 23.75 m, B: 0 m, 21.14 m, C:

−7 m, 21.14 m, D: 4.8 m, 23.75 m, E: −4.8 m, 23.75 m).

Table 1. Electrical characteristics of high voltage and medium voltage lines

High voltage transmission line Medium voltage distribution line
Type of conductor ACSR ACSR
Cross-section 636 MCM 3  95 mm2
R (X/km) 0.097 0.245
X (X/km) 0.422 0.334
C (X/km) 8.7 –
Zo = Ro + jXo (X/km) 0.497 + j2.349 0.363 + j1.556
Co (X/km) 4.1 –

In the case that the arising overvoltage across the insulators becomes greater than
their insulation level, then a flashover happens. The flashover strength VFO (in kV) is
calculated by the equation [13, 14]:
 
710
VFO ¼ 400 þ 0:75
D ð2Þ
t

where D is the length of the insulator string in m, and t is the elapsed time after
lightning stroke in ls.
 Lightning Performance of an Air Insulated High Voltage Substation 1477

For the representation of surge arresters the IEEE model is used [15]. Table 2
contains the electrical characteristics of the arresters that are used in the Hellenic
system.

Table 2. Electrical characteristics of the used metal oxide surge arresters.

Continuous operating voltage (COV) 114 kV rms
Rated voltage (Ur) 144 kV rms
Nominal discharge current (8/20 ls) 10 kA peak
Maximum residual voltage At 5 kA 330
at lighting impulse (8/20 ls) At 10 kA (lightning impulse 350
protective level)
At 20 kA 390
Energy capability Line discharge class 3
Energy withstand 8.5 kJ/Kv (Ur)
High current impulse withstand 100 kA peak
Short circuit withstand capability 31 kA rms

The grounding resistance of the substation is considered equal to 1 X, since the

tower footing resistance varies from 1 X to 25 X. The grounding resistance R(I) is
represented taking into consideration the soil ionization, according to (3) and (4) [16]:

Ro
RðI Þ ¼ qffiffiffiffiffiffiffiffiffiffiffiffi ð3Þ
1 þ IIg

1 Eo :q
Ig ¼ : ð4Þ
2p R2o

where Ro is the low current grounding resistance, Ig is the critical current for the
initiation of sufficient soil ionization and q is the soil resistivity.

4 Results and Discussion

Firstly, the effect of the number of the incoming transmission lines and the lightning hit
position on the lightning performance of the substation is examined. For the config-
uration of Fig. 3, the developed overvoltages at the entrance of the substation are
calculated, by using ATP-EMTP software, considering 2−4 incoming high voltage
lines and various positions of the lightning hit (100 kA for strokes to tower). The tower
footing resistance is equal to 5 X. Figure 5 depicts the estimated overvoltages at the
high voltage bus of the substation under study in function with the lightning hit
position and the number of the incoming lines.
1478 C. A. Christodoulou et al.

Fig. 5. Developed overvoltages on the high voltage busbars in function with the lightning hit
position and the number of the connected transmission lines.

The magnitude of the developed overvoltages decreases as more transmission lines

are connected with the substation, considering that the incoming (from one of the lines)
surge faces the low surge impedance of the parallel combination of the other lines. In
addition, the obtained outcomes indicate that the number of the connected high voltage
lines does not significantly influence the expected peak of the arising surges in the case
of lightning hits very close to the substation. However, for lightning events far away
from the substation, more lines contribute to the reduction of the incoming surges.
The second step of the current analysis includes a sensitivity analysis that examines
the impact of the placement of the surge arresters and the tower footing resistance (Rg)
on the lightning performance of the substation. A lightning flash strikes the shielding
wire of the 150 kV line between the second and the third tower. Three different
protection schemes are examined (two connected transmission lines are considered):
(a) installation of arresters at the high voltage side of the 150/20 kV transformer
(b) installation of arresters at the high voltage side of the 150/20 kV transformer and
the high voltage bus
(c) installation of arresters at the high voltage side of the 150/20 kV, the high voltage
bus and the three first towers of the connected lines.
The developed overvoltage due to backflashover of the incoming lines is measured
at the 150 kV side of the power transformer (Fig. 6) and at the high voltage bus
(Fig. 7).
 Lightning Performance of an Air Insulated High Voltage Substation 1479

Fig. 6. Developed overvoltages at the 150 kV side of the power transformer.

In the case that the installation under study is not protected by surge arresters,
lightning overvoltages on substation equipment get greater than the insulation level.
The placement of surge arresters at strategical position of the substation reduces the
magnitude of the developed surges below the critical value that results in insulation
breakdown. It is noteworthy that the implemented arresters, even only at the 150 kV

Fig. 7. Developed overvoltages at the high voltage bus.

1480 C. A. Christodoulou et al.

side of the transformer, improve the lightning performance of the substation. In pro-
tection schemes (b) and (c) the arising overvoltages are further reduced compared to
protection scheme (a), providing higher level of protection. As far as the tower footing
resistance is concerned, the magnitude of the developed overvoltages increases with
tower grounding resistance; this is more intense in the case that no arresters are
installed.

5 Conclusions

Lightning protection of 150/20 kV substations continue to be a prominent research

field for the operation of power systems in an effort to mitigate the ramifications
deriving from them. Note that the level of electrical power reliability and quality has
been closely linked to the adequate protection of the 150/20 kV substations against
lightning surges. In response to these challenges, the current work deals with the
analysis of the lightning performance of a 150/20 kV AIS. The role of the number of
the connected overhead 150 kV lines and the position of the lightning stroke on the
magnitude of the arising voltage surge are discussed. Moreover, different protection
schemes are examined, concerning the placement of the arresters and the grounding
resistance of the incoming transmission lines. The obtained results reveal the central
importance of the appropriate design of the lightning protection system of the sub-
stations and can be a practical tool for engineers and power utilities for the enhance-
ment of the lightning performance of already existed substations or for the more
effective design of new ones.

Acknowledgment. The authors acknowledge financial support for the dissemination of this
work from the Special Account for Research of ASPETE through the funding program
“Strengthening ASPETE’s research’’.

References
1. Christodoulou, C.A., Vita, V., Maris, T.I.: Lightning protection of distribution substations by
using metal oxide gapless surge arresters connected in parallel. Int. J. Power Energy Res. 1
(1), 1–7 (2017)
2. Christodoulou, C.A., Vita, V., Ekonomou, L.: Studies for the more effective protection of
MV/LV substations against lightning overvoltages. Int. J. Circ. Electron. 2, 11–15 (2017)
3. Siemens, Energy Sector, Power Engineering Guide: Answers For Energy, 5th edn. (2008)
4. IEEE Std. 998-1996: IEEE guide for direct lightning stroke shielding of substations (1996)
5. McDonald, J.: Electric Power Substations Engineering, 3rd edn. CRC Press, Boca Raton
(2006)
6. Hamza, A.H., Abdel-Gawad, N.M.K.: Investigation of overvoltages in a gas insulated
substation (GIS) caused by a lightning stroke. Energy Convers. Manag. 35(12), 1107–1116
(1994)
7. Kostica, V.I., Raicevic, N.B.: An alternative approach for touch and step voltages
measurement in high-voltage substations. Electric Power Syst. Res. 130, 59–66 (2016)
 Lightning Performance of an Air Insulated High Voltage Substation 1481

8. Christodoulou, C.A., Vita, V., Voglitsis, D., Milushev, G., Ekonomou, L.: A heuristic
method for the reduction of the outage rate of high-voltage substations due to atmospheric
overvoltages. Appl. Sci. 8, 273 (2018)
9. Trainba, M., Christodoulou, C.A., Vita, V., Ekonomou, L.: Lightning overvoltage and
protection of power substations. WSEAS Trans. Power Syst. 12, 107–114 (2017)
10. Kuczek, T., Chmielewski, T., Abdel-Latif, A.: Modeling insulation in high-voltage
substations. ABB Rev. 2, 38–43 (2017)
11. Hosseini, S.A., Mirzaie, M., Barforoshi, T.: Impact of surge arrester number and placement
on reliability and lightning overvoltage level in high voltage substations. Electr. Power
Energy Syst. 65, 146–158 (2015)
12. Chisholm, W.A., Chow, Y.L., Srivastava, K.D.: Travel time of transmission towers. IEEE
Trans. Power Apparatus Syst. PAS- 104(10), 2922–2928 (1985)
13. Darveniza, M., Popolansky, F., Whitehead, E.R.: Lightning protection of UHV lines. Electra
41, 39–69 (1975)
14. IEEE Working Group: Estimating lightning performance of transmission lines II–Updates to
analytical models. IEEE Trans. Power Deliv. 8(3), 1254–1267 (1993)
15. IEEE WG 3.4.11: Modeling of metal-oxide surge arresters. IEEE Trans. Power Deliv. 7(1),
302–309 (1992)
16. CIGRE Working Group 33.01: Guide to Procedures for Estimating the Lightning
Performance of Transmission Lines. Technical Bulletin 63, CIGRE, Paris, France (1991)
 Estimation of the Minimum Backflashover
Current of 150 and 400 kV Overhead
Transmission Lines Through ATP-EMTP
Simulations: Effect of the Lightning Stroke
Location Along Line Spans

Z. G. Datsios(&) , P. N. Mikropoulos , T. E. Tsovilis ,

V. T. Karakostas, and S. P. Mavidou

High Voltage Laboratory, School of Electrical and Computer Engineering,

Faculty of Engineering, Aristotle University of Thessaloniki,
541 24 Thessaloniki, Greece
zdatsios@auth.gr

Abstract. The lightning performance of overhead transmission lines is greatly

affected by backflashover, that is, flashover of line insulation due to fast front
overvoltages caused by lightning strokes terminating to towers or shield wires
along line spans. One of the parameters affecting the estimated backflashover
rate of overhead transmission lines is the minimum lightning current causing
backflashover. In estimating this critical current, the lightning stroke location
along the span is usually disregarded. Actually, the backflashover rate of
overhead transmission lines is most commonly estimated based on the critical
current corresponding to lightning strokes to transmission towers. The fact that
lightning strokes terminate to shield wires along the span is considered indi-
rectly by using a multiplier coefficient (span factor) of 0.6; however, this
coefficient is expected to depend on the several influencing factors. In this study
the minimum backflashover current of 150 and 400 kV double-circuit overhead
lines of the Hellenic transmission system is estimated through ATP-EMTP
simulations considering the lightning stroke location along the span. It is shown
that the lightning overvoltages stressing line insulators decrease with increasing
distance of the lightning stroke location from the tower, taking the lowest values
at midspan; this behavior is due to the lower current flowing through the tower
for midspan strokes. The minimum backflashover current increases with
increasing distance from the tower; this is less pronounced for relatively high
ground resistance values. The ratio of the total backflashover rate of the over-
head lines to the backflashover rate for strokes to tower only (span factor) is
found to vary between approximately 0.5 and 0.9, depending on tower ground
resistance and transmission line geometry and insulation level. Thus, the span
factor may take values significantly higher than the commonly used value of 0.6;
this may lead to underestimating backflashover rate, especially for overhead
lines with relatively high ground resistance values.

Keywords: ATP-EMTP
Backflashover
Backflashover rate
Lightning

Overhead transmission lines

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1482–1490, 2020.
https://doi.org/10.1007/978-3-030-31676-1_138
 Estimation of the Minimum Backflashover Current of 150 and 400 kV 1483

1 Introduction

Backflashover of overhead transmission lines is defined as flashover of line insulation

due to fast front overvoltages caused by lightning strokes terminating to towers or
shield wires along line spans. One of the parameters affecting the estimated back-
flashover rate, BFR, of overhead transmission lines is the minimum lightning current
causing backflashover, IBF. This critical current depends on:
– overhead line geometrical and electrical characteristics
– tower geometry
– tower grounding system geometry
– soil electrical properties
– lightning current waveshape
– phase angle of the power frequency voltage at the instant of the lightning strike
– location of the lightning stroke along the span.
These influencing factors can be taken into account in the estimation of IBF through
analytical calculations using simplified expressions [1–3] or through computer simu-
lations [4, 5]. The latter allow for the detailed modeling of the transmission line
components and of the complex phenomena involved in backflashover [6–10]. How-
ever, in estimating IBF the lightning stroke location along the span is commonly dis-
regarded. According to CIGRE [2], the BFR of overhead transmission lines is estimated
based on the critical current corresponding to lightning strokes to transmission towers.
In the same technical brochure [2], the fact that lightning strokes terminate to shield
wires also along the span is considered indirectly by using a conservative multiplier
coefficient (span factor) of 0.6. However, this coefficient is expected to depend on the
influencing factors listed above.
In this study the minimum backflashover current of 150 and 400 kV double-circuit
overhead lines of the Hellenic transmission system is estimated through ATP-EMTP
[11, 12] simulations considering the lightning stroke location along the span. It is
shown that IBF increases considerably with increasing distance of stroke location from
the tower; this is due to reduced lightning overvoltages stressing line insulation. The
ratio of the total BFR to that due to strokes to the tower only is obtained; this ratio
depends on tower ground resistance as well as transmission line geometry and insu-
lation level.

2 Transmission Line Modeling

Geometry and characteristics of the investigated 150 and 400 kV double-circuit

overhead lines of the Hellenic transmission system are shown in Fig. 1. These typical
transmission lines have been employed in lightning overvoltage simulations before in
[13–20]. Details on modeling of line components have been given in [19]. Briefly, ten
spans were considered in simulations, terminated by long sections so as to avoid
reflections. Each span was modeled with a JMarti [21] frequency-dependent model.
The phase angle of the power frequency voltage (cosinus function) at the lightning
stroke instant was varied from 0o to 330o in 30o increments.
1484 Z. G. Datsios et al.

150 kV 400 kV 5.6 m

Insulator length (m):
1.86 3.62
BIL (kV):
750 1425
Span length (m): 300 350 8.6 m
Shielding angle at tower (deg):31.6 19.2
Phase configuration: Superbundle 5.8 m
(A: upper phase)
3.2 m
8.5 m

5.2 m
9.4 m

3.9 m
5.2 m 8m

3.9 m
40 cm 6.4 m
4.1 m

20 m 20 m

5.8 m 8.4 m
150 kV 400 kV

Fig. 1. Geometry and characteristics of the 150 and 400 kV double-circuit overhead lines of the
Hellenic transmission system.

Four lossless frequency-independent distributed parameter lines connected in series

were used for tower modeling. The surge impedance of the double-circuit towers was
calculated by approximating them as cones. The Sargent and Darveniza [22] and the
Chisholm et al. [23] models were adopted for lightning strokes terminating respectively
to the tower (vertical current injection) and to the shield wire within the span (hori-
zontal current injection). Surge propagation velocity along towers was taken equal to
85% of the speed of light [24].
In order to obtain conservative simulation results, the concentrated tower grounding
systems were represented by their low current and low frequency ground resistance, R0,
which was considered as a parameter; soil resistivity values, q, used for line parameter
calculation were varied accordingly [19, 20]. Insulator flashover was predicted using
the CIGRE [2] leader development model with parameters for negative flashover of cap
and pin insulators. This model was implemented in ATP-EMTP with the aid of a type-
94 circuit component [17] considering the predischarge current flowing in the gap
during the leader development phase.
Several lightning stroke locations were considered in simulations by dividing the
span in 25 m segments. The CIGRE [2] current source was adopted for lightning
current representation (negative polarity). A 400 X resistance in parallel to the source
was used to simulate the effects of the return-stroke channel [15, 25, 26]. The adopted
current waveshape parameters correspond to median values: the time to half value was
77.5 ls, whereas the front time and maximum steepness were functions of the current
peak [2].
 Estimation of the Minimum Backflashover Current of 150 and 400 kV 1485

3 Results and Discussion

3.1 Computed Overvoltages Across Line Insulation
Figure 2 shows computed overvoltages (withstand cases) arising across the insulators
of the 150 and 400 kV overhead lines due to lightning strokes to the tower and midspan
at the time instant of the positive peak of the power frequency voltage of each phase
conductor. These maximum overvoltages were obtained for power frequency ground
resistance values of 25 X and 100 X. It can be seen that the overvoltages caused by
lightning strokes to midspan are considerably lower due to the smaller current flowing
through the tower (red lines in Fig. 2). Therefore, the minimum lightning current
causing backflashover of line insulation is expected to increase with increasing distance
from the tower reaching its maximum value at midspan.

R0 = 25 R0 = 25 400 kV
I = 65 kA Lower
Lower I = 150 kA phase
phase
Middle
phase Upper
phase Middle
phase

Upper
phase

Current
150 kV Current

R0 = 100 150 kV R0 = 100 400 kV

I = 25 kA Lower
Lower I = 60 kA phase
phase

Middle
Middle phase
phase

Upper
phase Upper
phase

Current Current

Fig. 2. Normalized overvoltages across the insulators of the 150 and 400 kV overhead lines
together with the current flowing through the tower (red lines). Lightning strokes to the tower and
midspan; withstand cases.

3.2 Minimum Backflashover Current

Figure 3 shows the variation of the minimum backflashover current, IBF, with the phase
angle of the power frequency voltage for several lightning stroke locations along the
span of the 150 and 400 kV overhead lines. From this figure it can be seen that the
phase angle determines the phase insulator where flashover occurs. For the investigated
double-circuit lines, the lower insulator is more prone to flashover due to the reduced
coupling of the lower phase conductor with the shield wire resulting in higher over-
voltages (Fig. 2). Consequently, the lowest (threshold) value of IBF is obtained for a
phase angle of 240o corresponding to the positive peak of the power frequency voltage
of the lower phase conductor.
From Fig. 3 it can also be observed that IBF varies considerably with the lightning
stroke location along the span. This variation is better demonstrated in Fig. 4 for the
1486 Z. G. Datsios et al.

Upper phase Middle phase Lower phase Upper

flashover flashover flashover
50

45
Critical current, IBF (kA)

40

35

30

25
Tower 25 m
20
50 m 75 m
15 100 m 125 m
150 kV 150 m
10
110

100
Critical current, IBF (kA)

90

80

70

60
Tower 25 m
50
50 m 75 m
40 100 m 125 m
400 kV 175 m
30
0 30 60 90 120 150 180 210 240 270 300 330
Phase angle (deg)

Fig. 3. Variation of the minimum backflashover current, IBF, with the phase angle of the power
frequency voltage (cosinus function) of the 150 and 400 kV overhead lines; R0 = 100 X.

threshold critical current IBFthr (lower insulator, phase angle: 240o) with R0 as a
parameter. Evidently, IBF increases with the distance of the stroke location from the
tower. This becomes less pronounced for higher R0 values, as can also be seen from
Fig. 5, which presents the variation of IBFthr with R0 for the cases of lightning stroke to
the tower and midspan. It is important that for strokes to the tower backflashover could
not be obtained for R0 < *6 X and 17 X for the 150 and 400 kV lines, respectively.
For midspan strokes, the corresponding R0 values are *13 X and 33 X.

200 200
150 kV 400 kV
175 175
Critical current, IBFthr (kA)

150 150

125 125

100 100

75 75

50 50

25 25

0 0
0 25 50 75 100 125 150 0 25 50 75 100 125 150 175
Distance (m) Distance (m)

Fig. 4. Variation of the threshold critical current, IBFthr, of the 150 and 400 kV overhead lines
with the distance of the lightning stroke location (along the span) from the tower; parameter:
ground resistance, R0.
 Estimation of the Minimum Backflashover Current of 150 and 400 kV 1487

200
Tower
Midspan
175

Threshold critical current, IBFthr (kA)

150

125
400 kV
100

75

50

25 150 kV

0
0 25 50 75 100 125 150 175 200
Power frequency ground resistance, R0 ( )

Fig. 5. Variation of the threshold critical current, IBFthr, of the 150 and 400 kV overhead lines
with ground resistance, R0, for lightning strokes to the tower and midspan.

3.3 Backflashover Rate

According to CIGRE [2], the backflashover rate, BFR, of an overhead transmission line
is given as (backflashovers/100 km/yr):
Z 1
BFR ¼ 0:6Ns f ðI ÞdI ð1Þ
IBF

where NS (strikes/100 km/yr) is the annual number of lightning strokes to shield wires,
IBF is the minimum backflashover current due to strokes to the tower and f(I) is the
probability density function of the lightning crest current distribution given by (2). In
the latter Ī and rln are the median value and the standard deviation of the natural
logarithm of the lightning crest current distribution taking values of 30.1 kA and 0.76
[27], respectively. The span factor of 0.6 has been introduced in (1) to account for the
effect of lightning strokes along the span to the BFR of the line [2].
" #
ðlnI  lnI Þ
2
1
f ðI Þ ¼ pffiffiffiffiffiffi exp  ð2Þ
2prln I 2r2ln

NS in (1) can be estimated as:

 
Ns ¼ 0:1Ng 2Req þ b ð3Þ

where Ng (strikes/km2/yr) is the ground flash density, b (m) is the separation distance
between the shield wires and Req (m) is the equivalent interception radius of the shield
wire:
1488 Z. G. Datsios et al.

Z 1
Req ¼ RðI; hÞf ðI ÞdI ð4Þ
0

where R(I, h) is the interception radius and h the average height of the shield wire. In
this study, the electrogeometric model of lightning attachment proposed by IEEE Std
1243 [28] was adopted for the calculation of Req.
For estimating the total BFR of the evaluated lines using the simplified approach
that is based on the IBF obtained for lightning strokes to the tower only together with
the span factor of 0.6, using (1)–(4), the BFR corresponding to each phase angle of the
power frequency voltage is calculated. The average of all these instantaneous BFR
values is the total BFR (simplified) of the line.
This procedure is also followed for all lightning stroke positions along the span of
the line and the total BFR is calculated as the average of the BFR values at all positions
in order to consider the effects of lightning strokes along the span. In this case, the span
factor of 0.6 in (1) is disregarded.
Figure 6 depicts the backflashover rate, BFR, of the evaluated 150 and 400 kV
overhead transmission lines as a function of R0 for lightning strokes to the tower (black
solid lines), midspan (black dashed lines) as well as the total BFR of the lines (red solid
lines), that is, the average of the BFR values along the span. BFR increases with a
progressively lower rate tending to upper limits at relatively high R0 values. Figure 6
also includes the total BFR estimated according to the simplified approach that is based
on the IBF obtained for lightning strokes to the tower together with the span factor of
0.6 (blue dotted lines). This BFR deviates from the total, depending on ground resis-
tance. Actually, the span factor, that is, the ratio of the total BFR to the BFR at tower,
was found to vary in the range from 0.47–0.91 and 0.55–0.85 for the 150 and 400 kV
overhead lines, respectively, depending on R0. Hence, using a span factor of 0.6, as
proposed by CIGRE [2], may lead to underestimating BFR, especially for lines with
relatively high R0 values. A thorough investigation is necessary to quantify the effects
of geometry and characteristics of typical overhead lines on the span factor; this calls
for further work.

100 10
150 kV 400 kV
BFR (Backflashovers/100 km/yr)

10

0.1

0.1
Tower
Midspan
Total
Total (simplified)
0.01 0.01
0 25 50 75 100 125 150 175 200 0 25 50 75 100 125 150 175 200
Power frequency ground resistance, R0 Power frequency ground resistance, R0

Fig. 6. Variation of the BFR of the 150 and 400 kV overhead lines with ground resistance, R0;
Ng = 1 strike/km2/yr.
 Estimation of the Minimum Backflashover Current of 150 and 400 kV 1489

4 Conclusions

The effect of lightning stroke location along an overhead line span on the minimum
backflashover current and backflashover rate of the line has been investigated. Two
typical 150 and 400 kV double-circuit overhead transmission lines were simulated
using ATP-EMTP software by considering the power frequency tower ground resis-
tance and the phase angle of the power frequency voltage as influencing parameters.
The lightning overvoltages stressing line insulators decrease with increasing dis-
tance of the lightning stroke location from the tower, taking the lowest values at
midspan; this behavior is due to the lower current flowing through the tower for
midspan strokes. The minimum backflashover current increases with increasing dis-
tance from tower; this is less pronounced for relatively high ground resistance values.
The ratio of the total backflashover rate of the overhead lines to the backflashover
rate for strokes to tower only (span factor) has been found to vary between approxi-
mately 0.5 and 0.9, depending on tower ground resistance and transmission line
geometry and insulation level. Thus, the span factor may take values significantly
higher than that of 0.6 proposed by CIGRE; this may lead to underestimating BFR,
especially for lines with relatively high R0 values.

References
1. Eriksson, A.J., Weck, K.H.: Simplified procedures for determining representative substation
impinging lightning overvoltages. In: CIGRE, paper no 33–16 (1988)
2. CIGRE Working Group 33.01: Guide to procedures for estimating the lightning performance
of transmission lines. In: Technical Brochure. 63 (1991)
3. Hileman, A.R.: Insulation Coordination for Power Systems. CRC Press, Boca Raton (1999)
4. IEEE Fast Front Transients Task Force: Modeling guidelines for fast front transients. IEEE
Trans. Power Del. 11(1), 493–506 (1996)
5. Ametani, A., Nagaoka, N., Baba, Y., Ohno, T.: Power System Transients: Theory and
Applications. CRC Press, Boca Raton (2014)
6. Banjanin, M.S., Savić, M.S.: Some aspects of overhead transmission lines lightning
performance estimation in engineering practice. Int. Trans. Electr. Energ. Syst. 26(1), 79–93
(2016)
7. Shariatinasab, R., Gholinezhad, J., Sheshyekani, K., Alemi, M.R.: The effect of wide band
modeling of tower-footing grounding system on the lightning performance of transmission
lines: a probabilistic evaluation. Electr. Pow. Syst. Res. 141, 1–10 (2016)
8. Araújo, A.R.J., Seixas, C.M., Kurokawa, S., Kordi, B.: Overvoltage analysis of transmission
towers considering the influence of tower-footing impedance. In: Proceedings of the
XIV SIPDA, Natal, Brazil (2017)
9. Rodrigues, A.R., Guimarães, G.C., Chaves, M.L.R., Boaventura, W.C., Caixeta, D.A.,
Tamashiro, M.A.: Lightning performance of transmission lines based upon real return-stroke
current waveforms and statistical variation of characteristic parameters. Electr. Pow. Syst.
Res. 153, 46–59 (2017)
10. Gholinezhad, J., Shariatinasab, R.: Time domain modeling of tower-footing grounding
system based on impedance matrix. IEEE Trans. Power Del. (2018). https://doi.org/10.1109/
tpwrd.2018.2881696
11. Canadian-American EMTP Users Group: ATP Rule Book (1997)
1490 Z. G. Datsios et al.

12. Dommel, H. W.: Electro-Magnetic Transients Program (EMTP) Theory Book. Bonneville
Power Administration (B.P.A.) (1986)
13. Datsios, Z.G., Mikropoulos, P.N., Tsovilis, T.E.: Estimation of the minimum shielding
failure flashover current for first and subsequent lightning strokes to overhead transmission
lines. Electr. Pow. Syst. Res. 113, 141–150 (2014)
14. Datsios, Z.G., Mikropoulos, P.N.: Modeling of lightning impulse behavior of long air gaps
and insulators including predischarge current: implications on insulation coordination of
overhead transmission lines and substations. Electr. Pow. Syst. Res. 139, 37–46 (2016)
15. Datsios, Z. G., Mikropoulos, P. N., Tsovilis, T. E.: Effects of lightning channel equivalent
impedance on lightning performance of overhead transmission lines. IEEE Trans.
Electromagn. Compat. (2019). https://doi.org/10.1109/temc.2019.2900420
16. Datsios, Z. G., Mikropoulos, P. N., Tsovilis, T. E.: Impulse resistance of concentrated tower
grounding systems simulated by an ATPDraw object. In: Proceedings of the IPST, Delft,
The Netherlands, paper no. 39 (2011)
17. Datsios, Z. G., Mikropoulos, P. N.: Implementation of leader development models in ATP-
EMTP using a type-94 circuit component. In: Proceedings of the 32nd ICLP, Shanghai,
China, 735–741, paper no. 270 (2014)
18. Datsios, Z. G., Mikropoulos, P. N.: Effect of tower modelling on the minimum backflashover
current of overhead transmission lines. In: Proceedings of the 19th ISH, Pilsen, Czech
Republic, paper no. 26 (2015)
19. Datsios, Z. G., Mikropoulos, P. N., Tsovilis, T. E., Angelakidou, S. I.: Estimation of the
minimum backflashover current of overhead lines of the hellenic transmission system
through ATP-EMTP simulations. In: Proceedings of the ICLPS, CIGRE SC C4, Bologna,
Italy, paper no. 32 (2016)
20. Datsios, Z. G., Mikropoulos, P. N., Tsovilis, T. E., Angelakidou, S. I.: Effect of concentrated
tower grounding system modeling on the minimum backflashover current and BFR of 150
and 400 kV overhead transmission lines. In: Proceedings of the ICHVE, Athens, Greece,
paper no. O-TE1–2 (2018)
21. Marti, J.R.: Accurate modelling of frequency-dependent transmission lines in electromag-
netic transient simulations. IEEE Trans. Power App. Syst. PAS-101(1), 147–157 (1982)
22. Sargent, M.A., Darveniza, M.: Tower surge impedance. IEEE Trans. Power App. Syst. PAS-
88(5), 680–687 (1969)
23. Chisholm, W.A., Chow, Y.L., Srivastava, K.D.: Travel time of transmission towers. IEEE
Trans. Power App. Syst. PAS-104(10), 2922–2928 (1985)
24. IEEE Working Group on Lightning Performance of Transmission Lines: A Simplified
Method for Estimating Lightning Performance of Transmission Lines. IEEE Trans. Power
App. Syst. 104(4), pp. 919–932 (1985)
25. Ametani, A., Kawamura, T.: A method of a lightning surge analysis recommended in japan
using EMTP. IEEE Trans. Power Del. 20(2), 867–875 (2005)
26. Bewley, L.V.: Traveling Waves on Transmission Systems, 2nd edn. Dover Publications,
New York (1963)
27. Lightning and Insulator Subcommittee of the T&D Committee: Parameters of lightning
strokes: a review. IEEE Trans. Power Del. 20(1), 346–358 (2005)
28. IEEE guide for improving the lightning performance of transmission lines. In: IEEE Std,
1243–1997 (1997)
 Analysis of the Frequency Response
of a Grounding System Using
the Finite Element Method

Arthur Francisco Andrade1 , George R. S. Lira1(B) ,

Pedro Henrique F. Monteiro1 , Helder A. Pereira1 , Edson G. Costa1 ,
and Rômulo O. Teixeira2
1
Departament of Electrical Engineering (DEE),
Center of Electrical Engineering and Computer Science (CEEI),
Federal University of Campina Grande (UFCG), Campina Grande, Paraı́ba, Brazil
{arthur.andrade,pedro.monteiro}@ee.ufcg.edu.br,
{george,helder.pereira,edson}@dee.ufcg.edu.br
2
Light SA, Rio de Janeiro, Brazil
romulo.teixeira@light.com.br

Abstract. This paper analyzed the frequency response of a simple

grounding system, consisting of a rod, through computational simula-
tions. A finite-element-based software for solving the electromagnetic
problem governed by differential equations was used. In the simulations,
the grounding system was submitted to currents with different frequen-
cies with a spectrum from 60 Hz to 1 MHz. From the relation between the
frequency and the calculated impedance, transfer functions that model
the behavior of the grounding system were obtained. For the estimation
of the function parameters, methods of optimization and curve-fitting
methods were used. In addition, since both the resistivity and the elec-
tric permittivity attributed to the soil should be considered the simulated
response, a sensitivity analysis was performed with different resistivity
and permittivity values. Obtaining analytical expressions for the descrip-
tion of the ground response provides a tool for the evaluation of the
grounding response to signals in the frequency or time domain. Consid-
ering the results obtained in the simulations, it is possible to evaluate
how the parameters resistivity and electric permittivity of the soil influ-
ence the analytical expression that characterizes a grounding system.
Thus, it is planned to compare in a future work the results obtained in
the simulation with field experiments.

Keywords: Computational simulation · Finite element method ·

Frequence response · Grounding system

1 Introduction
Grounding systems are essential for the safety of people and equipment in an
electrical system. In most technical studies of design and evaluation of grounding
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1491–1501, 2020.
https://doi.org/10.1007/978-3-030-31676-1_139
1492 A. F. Andrade et al.

systems, the analysis is performed only considering the low frequency resistance
parameter. However, when subjected to stresses with higher frequency compo-
nents, such as current impulses from atmospheric discharges or switching, both
the resistive, capacitive and inductive characteristics of the grounding can exert
an influence on the response, since reactive effects have a direct influence on the
actual impedance during a transient current input. So, a correct representation
must account for the frequency response [1].
Thus, some studies have been devoted to the analysis of grounding frequency
response of grounding [2–4]. Hasan et al. [2] performed tests with the objec-
tive of comparing experimental data with simulations performed to analyze the
harmonic impedance of a grounding system. It was observed that, for a given fre-
quency range, there is a good agreement between the experimental data and the
simulation data. Additional simulations were carried out with the objective of
evaluating the impact of resistivity and permittivity in the frequency response of
the ground. It was found that the resistivity has a direct impact on the impedance
behavior, while the influence of the permittivity was negligible. Choi et al. [3]
carried out measurements to analyze the frequency response of a grounding loop.
A variable-frequency current signal was applied, while the potentials in the loop
were measured and the harmonic impedance was obtained from the processing of
these data. Thus, transfer functions were obtained that described the behavior
of the grounding mesh considered and enabled the analysis of transients in the
time domain. In Velilla et al. [4] the frequency response of a grounding system
with horizontal electrodes was obtained. For this, a methodology that described
the grounding as an equivalent circuit was adopted. From the equivalent circuit
it is was obtained transfer function to compare its response with the frequency
response of the grounding system. The performance of the developed methodol-
ogy was evaluated by calculating the error between the transfer function obtained
and the frequency response data of the grounding system.
However, the previous works are based on experimental data from a pre-
existing grounding, whereas at the design stage the only available data are: the
specified mesh geometry and a soil model. In this context, this study has the
objective of proposing and analyzing tools for the representation of grounding
systems, in order to estimate the frequency response of the system even during
the design stage. Thus, this paper proposes an analysis of the frequency response
of a ground rod system using computational tools to analyze the influence of
resistivity and electric permittivity on the ground behavior and to obtain transfer
functions that describe the grounding from geometry and soil parameters.
In order to do so, the finite element method (FEM) was employed. Based
on the obtained data, a curve fitting process was carried out for the purpose of
analyzing the data of the simulated grounding system, thus enabling the obtain-
ment of transfer functions that can be applied for evaluation of performance in
the frequency domain and calculation of transients in the time domain. This
paper is organized as follows: Sect. 2 describes the theoretical basis required for
grounding systems and parameters used in the simulations. In Sect. 3, the mate-
rials and methods used to carry out the simulations are presented. In Sect. 4,
 Analysis of the Frequency Response of a Grounding System 1493

the results are described, analyzed and discussed. Finally, Sect. 5 presents the
conclusions.

2 Theoretical Foundation
In this section, the soil parameters considered in the simulations of grounding
systems (Subsect. 2.1) are described, as well as the numerical methods used for
the simulations (Subsect. 2.2).

2.1 Soil Parameters

In the context of electrical grounding systems, the resistivity (ρ) and the elec-
tric permittivity () are considered the properties of the soil with determinant
influence on its operation, since the relative permeability of the soil is commonly
considered equal to 1 [1]. The resistivity depends, among other factors, on the
nature of the soil and the climatic conditions. The permittivity corresponds to
the polarization characteristic of a material in response to an applied electric
field [5].

2.2 Finite Element Method

According to Sadiku [6], the finite element method is a powerful and versatile tool
for solving problems with complex geometry and inhomogeneous environments
that are governed by differential equations. The FEM application can be divided
into four stages: (1) discretization of the domain studied in a finite number of
subregions; (2) selection of interpolation functions; (3) formulation of the system
of equations and (4) resolution of the obtained system of equations.

3 Materials and Methods

In order to analyze a grounding system constituted by a rod, computational sim-
ulations were performed. Subsections 3.1 to 3.2 describe the procedures adopted,
from the definition of the grounding system geometry to the analysis of the data
obtained with the simulation.

3.1 Grounding System Simulation

In order to characterize the frequency response and to evaluate the impact of
soil resistivity on the grounding impedance, a grounding system consisting of a
rod was simulated. For that, the software COMSOL Multiphysics  R was used
as simulation tool. The development of the problem can be divided into the
following steps: (1) definition of the physics used in the simulation; (2) definition
of the grounding system geometry; (3) characterization of the materials that
compose the system and (4) definition of the resistivity and application of a
current input signal with a given frequency.
1494 A. F. Andrade et al.

Step 1. To simulate the grounding system, the physics “magnetic and elec-
tric fields” was used to apply the Eqs. (1) to (5) to model the electromagnetic
problem [7].
− −
→ →
∇ · J = 0, (1)
− −
→ → − →
∇×H = J, (2)
−
→ → −
− →
B = ∇ × A, (3)
−
→ →
− ∂− →
E = − ∇V − A, (4)
∂t
and
−
→ →
− ∂− →
J = σ E + D, (5)
∂t
→
− →
−
in which E represents the electric field, D the electric displacement field, V the
→
− →
−
electric scalar potential, σ the conductivity, H the magnetic field, B the mag-
→
− →
−
netic flux density, A the magnetic vector potential and J the current density.

Step 2. Due to the axial symmetry of the problem, an axisymmetric geometry

was used to simulate the grounding system as shown in Fig. 1.

Step 3. As shown in Fig. 1, the element indicated by 1 represents the single-rod

grounding that is 2.4 m long and 16 mm in diameter. The element indicated by
2 represents an air layer with a permittivity and permeability of 1 and elec-
tric conductivity equal to 10−8 S/m. The element indicated by 3 represents a
homogeneous soil with variable electric resistivity, depending on the case study
considered (Scenario 1 to 3), as described on Table 1. The magnetic permeability
was considered as equal to 1 and the permittivity electric variable. The element
indicated by 4 represents an infinite element, a computational resource used to
emulate domains with unlimited extension, such as a grounding system. At the
outer edge of the infinite element, the electric scalar potential was considered as
null, specified as boundary condition (reference earth).

Step 4. After steps 1 to 3, simulations were performed in which the parameters:

resistivity, relative permittivity of the soil as well as the frequency of the input
signal were varied. For the different relative permittivity values considered in the
simulations (4, 10 or 20) no significant differences were observed in the obtained
results. The resistivity values analyzed are shown in Table 1, as well as the type
of soil considered.
For each resistivity value considered (100, 1000 or 10000 Ωm), the frequency
was varied from 60 Hz to 1 MHz. Thus, the impedance of the grounding system
was obtained as a function of frequency for each of the resistivity values analyzed.
The relationship between potential and applied current, relative to the point of
current application, provides the harmonic impedance of the grounding system.
 Analysis of the Frequency Response of a Grounding System 1495

Fig. 1. Axisymmetric geometry used to simulate the grounding system.

Table 1. Simulation scenarios considered in relation to resistivity and soil type.

Scenario Resistivity [Ωm] Type of soil [8]

1 100 Moist soil
2 1000 Dry soil
3 10000 Bedrock

3.2 Data Processing and Retrieval of Transfer Functions

In order to characterize the frequency response of the grounding system and

to obtain transfer functions that represent the behavior of the obtained data,
a non-linear least squares method was used to fit curves [9]. This technique
aims to minimize the quadratic error between the data points obtained from
the simulation and the function to be estimated. For this, the method should
minimize the following expression
N
 2
min | Zest,n − Zsim,n | , (6)
n=1

in which Zest,n represents the estimated impedance values from the obtained
transfer function and Zsim,n represents the impedance values obtained with the
simulation performed. Initially, only the impedance module was fitted. Based on
these estimates, the obtained model was used as the initial guess for an optimiza-
tion method, which adjusted both the module and the impedance phase, thus
characterizing the frequency response of the grounding system. The MATLAB  R
platform was used to implement the curve fitting algorithm. For each scenario
considered, we performed lower order functions that described with acceptable
accuracy the behavior of the data obtained with the simulations.
1496 A. F. Andrade et al.

4 Results and Discussions

The simulation data were obtained considering different values of resistiv-
ity’ (100, 1000 and 10000 Ωm) in order to evaluate the impact of the parameter
on the frequency response behavior of the grounding system. Subsections 4.1 to
4.3 describe the results according to each scenario analyzed (Scenarios 1 to 3).
In addition, the behavior of the potential at the soil surface is discussed due to
the resistivity variations.

4.1 Obtained Transfer Functions

In order to characterize the frequency response of the grounding system based
on data obtained with simulation, an optimization method was used for adjust
transfer functions to simulated impedance data. The obtained transfer functions
describe the behavior of the simulated for each resistivity value considered. Equa-
tions (7) (Scenario 1) and (8) (Scenarios 2 and 3) are the models adopted in this
paper. In Table 2, the parameters calculated for each evaluated scenario after fit
procedure are shown.

s2 + Cs + D
Z(s) = (7)
As + B
and
A
Z(s) = . (8)
s+B
For each one of the obtained functions, the deviation in impedance magnitude
and phase angle between them and the simulated scenarios are calculated. For
that, two metrics were used to quantify the error, the mean square of the (α)
and the maximum value of the error (β).

Table 2. Coefficients estimated from the curve fit for each simulated scenario.

Parameter Scenario 1 Scenario 2 Scenario 3

A 1.0 × 105 4123.00 4.3 × 106
5
B 2.7 × 10 10.68 1.1
C 4.0 × 107 – –
7
D 9.0 × 10 – –

4.2 Frequency Response Simulations

In order to verify the adequacy of the obtained model for each data set, ampli-
tude and phase angle plots of the harmonic impedance of the simulated ground
system, were elaborated as shown in Fig. 2. Figure 2 describes the behavior of
the frequency response of the grounding system taking into account the resis-
tivity value of: (a) ρ = 100 Ωm (Scenario 1); (b) ρ = 1000 Ωm (Scenario 2)
 Analysis of the Frequency Response of a Grounding System 1497

and (c) ρ = 10000 Ωm (Scenario 3). The points (in blue color) represent the
data obtained in the simulations and the solid line (in red color) describes the
behavior of the transfer functions estimated from the simulated data.
Observing Fig. 2a to Fig. 2c, we can see the influence of the resistivity on
the harmonic impedance behavior of the grounding system. As expected, the
impedance had a value proportional to the soil resistivity at low frequencies.
However, in Fig. 2a, it is verified that, with the increase of the frequency, there is
an increase in the value of the harmonic impedance which evidences an inductive
characteristic of the ground. For Fig. 2b and Fig. 2c, the gauge with 1000 Ωm and
10000 Ωm, respectively, is an increase in the value of the grounding capacity with
the frequency of the frequency, implying a capacitive characteristic. It was also
found that the obtained functions described by Eqs. (7) to (8) agree with the data
obtained from the simulations. This can be seen in Fig. 2 and in the associated
error values and described in Table 3.

Table 3. Quadratic and maximum error values associated to the module and the phase
of each transfer function obtained with the respective resistivity value.

Scenario Module Phase

α (%) β (%) α (%) β (%)
−4
1 0.02 0.06 8.75 × 10 1.10 × 10−3
−4
2 0.60 1.90 2.42 × 10 7.20 × 10−4
3 2.30 6.80 4.30 × 10−4 1.20 × 10−3

4.3 Surface Potentials

Figure 3 describes the behavior of the electric potential as a function of the

distance considering the resistivity value of: (a) ρ = 100 Ωm (Scenario 1);
(b) ρ = 1000 Ωm (Scenario 2) and (c) ρ = 10000 Ωm (Scenario 3). According
to Fig. 3, it can be inferred that the resistivity has a direct influence on the
surface potentials of the grounding system, being the potential value proportional
to the resistivity value considered. Another pertinent observation is that with
increasing resistivity, the high frequency component of the electric potential is
attenuated.
Table 4 shows the relation of the touch potential for the different scenarios.
Note that the touch potential for Scenario 1 does not change considerably as
frequency increases, which is not the case in Scenarios 2 and 3, where an increase
in the frequency from 40 kHz to 1 MHz yields a considerable change in touch
potential.
1498 A. F. Andrade et al.

Magnitude of Z(f) [ ]
Grouding Impedance - Scenario 1
39

38.5
Scenario 1
38
101 102 103 104 105 106
Frequency [Hz]
10
Phase of Z(f) [º]

0
Scenario 1
-5
101 102 103 104 105 106
Frequency [Hz]

ρ = 100 Ω 1
Magnitude of Z(f) [ ]

Grouding Impedance - Scenario 2

400

350
Scenario 2
300
101 102 103 104 105 106
Frequency [Hz]
0
Phase of Z(f) [º]

-10

-20
Scenario 2
-30
101 102 103 104 105 106
Frequency [Hz]

ρ = 1000 Ω 2
Magnitude of Z(f) [ ]

Grouding Impedance - Scenario 3

4000

2000
Scenario 3
0
101 102 103 104 105 106
Frequency [Hz]
0
Phase of Z(f) [º]

-50
Scenario 3
-100
101 102 103 104 105 106
Frequency [Hz]

ρ = 10000 Ω 3

Fig. 2. Frequency response of the grounding system taking into account the resistiv-
ity value of: (a) ρ = 100 Ωm (Scenario 1); (b) ρ = 1000 Ωm (Scenario 2) and
(c) ρ = 10000 Ωm (Scenario 3).
 Analysis of the Frequency Response of a Grounding System 1499

Scenario 1
40
60 Hz
40 kHz
1 MHz

Electric potential (V)

30

20

10

0
0 1 2 3 4 5
Distance (m)
ρ = 100 Ω 1
Scenario 2
400
60 Hz
40 kHz
1 MHz
Electric potential (V)

300

200

100

0
0 1 2 3 4 5
Distance (m)
ρ = 1000 Ω 2
Scenario 3
4000
60 Hz
40 kHz
1 MHz
Electric potential (V)

3000

2000

1000

0
0 1 2 3 4 5
Distance (m)
ρ = 10000 Ω 3

Fig. 3. Earth surface potential as a function of the distance from the rod considering
the resistivity value of: (a) ρ = 100 Ωm (Scenario 1); (b) ρ = 1000 Ωm (Scenario 2)
and (c) ρ = 10000 Ωm (Scenario 3).
1500 A. F. Andrade et al.

Table 4. Ratio of touch potential for the different scenarios considered.

Frequency Touch potential (V )

Scenario 1 Scenario 2 Scenario 3
60 Hz 29.6 296.2 2963.4
40 kHz 29.6 296.1 2870.1
1 MHz 29.5 246.1 138.1

5 Conclusions
This paper proposed an analysis of the frequency response of a single-rod ground-
ing system. Computational tools were used to analyze the influence of resistivity
and electric permittivity on the grounding frequency response. Transfer func-
tions that describe the grounding impedance were obtained from geometry and
soil parameters. The transfer functions modeled with acceptable accuracy the
data obtained with the simulations. The maximum error obtained was 6.8% in
Scenario 3, in which the resistivity considered is 10000 Ωm. By means of the
proposed methodology, it is possible, based on a geometric model of a grounding
system and the adjacent soil, to estimate the behavior of the ground system
during transients, as well as to estimate the electric potential at the point of
injection in the time domain. It was found that the resistivity altered both the
impedance and distribution of surface potentials of the grounding system. In
addition, the analyses indicated that the permittivity exerted a reduced influ-
ence on the response of the grounding system, corroborating the analyses of
Hasan et al. [2]. In view of the observations, it is planned in a future work to
compare the results obtained in the simulations with experimental data obtained
in the field.

Acknowledgments. To light SA (Brazilian power utility) by support of this research

in the scope of ANEEL (Brazilian regulation agency) R&D program. To the High
Voltage Laboratory of Federal University of Campina Grande by provide the necessary
infrastructure.

References
1. Visacro, S.: A comprehensive approach to the grounding response to lightning cur-
rents. IEEE Trans. Power Delivery 22(1), 381–386 (2007)
2. Hasan, H.: Characterization of horizontal earth electrodes: variable frequency and
impulse responses. In: 50th International Universities Power Engineering Conference
(UPEC). IEEE, Stoke on Trent (2015)
3. Choi, J.: Direct measurement of frequency domain impedance characteristics of
grounding system. In: International Conference on Power System Technology,
pp.2218–2221. IEEE, Kunming (2002)
4. Velilla, E.: Using genetic algorithm and the simplex method to obtain equivalent
circuits of the grounding systems. In: IEEE/PES Transmission and Distribution
Conference and Exposition: Latin America, pp.1–5. IEEE, Bogota (2008)
 Analysis of the Frequency Response of a Grounding System 1501

5. Castro, M.S.: Sistema de aquisição de dados para diagnóstico de um sistema de

aterramento submetido a impulsos de corrente elétrica. Dissertation for a Mas-
ter’s degree in Electrical Engineering at the Federal University of Campina Grande,
Campina Grande (2012)
6. Sadiku, M.N.O.: Numerical Techniques in Electromagnetics, 2nd edn. CRC Press,
New York (2000)
7. Comsol, A.B.: Comsol Multiphysics: AC/DC Module User’s Guide. COMSOL, New
York (2013)
8. IEEE Standards Association: IEEE Guide for Safety in AC Substation Grounding.
IEEE Std 80TM . IEEE Power Engineering Society, New York (2013)
9. Mathworks: Solve nonlinear curve-fitting problems in least-squares sense. https://
www.mathworks.com/help/optim/ug/lsqcurvefit.html. Accessed 11 Feb 2016
 Analysis of the Influence of Soil
Stratification Models on Different
Grounding System Configurations

George R. S. Lira1(B) , Matheus R. Alves1 , André Igor Nóbrega1 ,

Arthur Francisco Andrade1 , Edson G. Costa1 , and Rômulo O. Teixeira2
1
Federal University of Campina Grande, Campina Grande, Brazil
george@dee.ufcg.edu.br
2
Light SA (Brazilian Power Utility), Rio de Janeiro, Brazil

Abstract. Grounding has great importance in power electrical systems.

The main parameter to evaluate its performance is the impedance, which
can be estimated in the design stage using the conductors geometry and a
soil stratification model based on measures of soil resistivity. In this con-
text, this work performed a comparative analysis of one-layer and two-
layer stratification models. The frequency response associated to each
of the soil models considered was calculated for two different grounding
systems and used to obtain the grounding response to a 8/20-µs current
impulse. Differences of up to 44% were obtained for impedance at low
frequency, and it was verified that the homogeneous soil model under-
estimates grounding impedance. That may lead to a worse-to-expected
performance during short-circuit or lightning. It can be concluded that
the modeling of the soil with the homogeneous model is discouraged
when designing a grounding system for substations.

Keywords: Grounding systems · Soil stratification · Finite element

method

1 Introduction

The importance of grounding to ensure the safety of electrical power systems is

well known. Although there are already standardized techniques and procedures
for the design and evaluation of ground structures [1,2], models that allow a
more reliable representation of the working conditions of grounding grids still
need to be studied and improved.
The purpose of a grounding grid is to provide a low impedance path for the
current flow to the ground [3], both in the case of currents from faults or from
lightning or switching.
The grounding performance is directly affected by the characteristics of the
soil in which the system is inserted. Thus, grounding systems project needs
consider the impact of the stratification models used to represent the soil in the
c Springer Nature Switzerland AG 2020
B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1502–1512, 2020.
https://doi.org/10.1007/978-3-030-31676-1_140
 Influence of Soil Stratification Models on Grounding Systems 1503

design parameters of grounding systems. According to [4], using the one-layer

stratification model, by means of the apparent resistivity method, can make
unfeasible the design of grounding grids, either from a financial or a technical
point of view. In this sense, according to [5], two-layer soil modeling generally
represents a good approximation in most cases.
Within this context, the objective of this paper is to compare one-layer and
two-layers soil stratification models, which are the most commonly models used
for soil stratification. To this end, two different ground configurations were con-
sidered. The changes in the value of ground impedance and ground potential rise
during a simulated current impulse were analyzed for the different soil models
considered. The novelty of this work consists of to simultaneously consider the
effects of soil representation and geometry on the harmonic impedance and on
the amplitude of transient overvoltages during a simulation study.
The analyses were performed using the finite element method for solving
differential electromagnetic equations. This paper is organized as follows: this
section presents an introduction to the theme, as well as the motivations of
this work, Sect. 2 describes the methods applied and the material used for the
development of the work, Sect. 3 presents the results obtained and Sect. 4, the
conclusions.

2 Material and Methods

The first step to perform the comparative analysis between soil stratification
models and their impact on ground impedance was the measurement of soil resis-
tivity, in order to obtain experimental data for the study. From the data obtained,
a commercial software was used to estimate the soil stratification models. Thus,
it was possible to obtain the soil resistivity value for the two situations, the first
referring to the assumption of a homogeneous soil, and the second for a two-layer
soil. Finally, with this data set, simulations to determine the ground impedance
were performed considering one-rod and two-rod groundings and one and two-
layer soil. The simulations were performed using the COMSOL Multiphysics R
software, which applies the finite element method (FEM) for the electromag-
netic field calculation. After the simulations, the results were compared with
each other and with analytical expressions obtained from the literature for the
calculation of ground resistance.

2.1 Field Measurements

The soil used for the study is located on a adjacent yard to the High Voltage
Laboratory of the Federal University of Campina Grande, in the northeast of
Brazil. In this way, to obtain experimental data on soil resistivity, the Wenner
method was used. This method consists in bury four aligned rods equally spaced
at a distance (a) and a depth (b), being the two end rods intended for current
injection (I), and the two central rods for the potential measurement (P1 and
P2 ) [6,7]. The measurement scheme is shown in Fig. 1.
1504 G. R. S. Lira et al.

Fig. 1. Wenner’s method for the measurement of soil resistivity

In the Wenner method, the resistivity for a given spacing is estimated from
the distance a, the injected current I and the potential difference V measured
between the central electrodes according to [6]:

4 × π × VI
ρ=     , (1)
1 − √a2a+b2 + √a22a
+4b 2

where V is the potential difference measured between the two central rods (V)
and I is the current injected into the outer rods (A).
If the condition b ≤ a/10 is satisfied, then (1) approaches (2):

2×π×a×V
ρ= (2)
I
Three parallel lines (A, B and C) along the yard were defined for measurements.
The distance values between the rods were 1 m, 2 m and 4 m. In Table 1, Ra , Rb
and Rc represents the resistance values obtained for each of the lines.

Table 1. Measurement table obtained by the Wenner method

Distance (m) Ra (Ω) Rb (Ω) Rc (Ω)

1 4.99 3.17 3.22
2 2.60 1.92 1.74
4 1.58 1.44 1.38

With the assumption of a homogeneous soil model, the obtained value for the
resistivity using the modeling method employed was 30.52 Ωm. For the stratified
soil, the resistivity and depth values of each layer can be seen in Table 2.

Table 2. Resistivity table for stratified soil

Layer Resistivity(Ωm) Depth (m)

1 26.48 4.46
2 421.40 ∞
 Influence of Soil Stratification Models on Grounding Systems 1505

2.2 Computational Simulations

For the simulations, a program that solves Maxwell’s equations was used. The
modeling of the problem requires the values of electrical permittivity, magnetic
permeability and electrical conductivity of all materials, which must be deter-
mined. At this stage of the research, a frequency-independent model was applied
to the soil resistivity, although the effective resistivity in the simulation has a
frequency-dependent characteristic due to magnetic induction effects, which was
considered. The values used in the analysis can be seen in Table 3.
In addition, boundary conditions and input parameters must be set. In
this case, the electric current injected into the grounding system was taken as
input parameter and the output variable obtained was the potential around the
grounding. The materials present in the simulation were soil, metal rod and air.
The soil was modeled from the data obtained by commercial software TecAt
Plus 5.2,
R shown in Table 2. The rod was considered to be composed of steel,
with a layer of copper coating, buried length of 2.4 m and a radius of 8 mm. This
rod configuration was adopted in all simulations.
Then, with the aim of analyzing another ground configuration from the same
scenario, grounding with 2 rods was simulated. The connection between the two
rods was carried out by means of a cable with the same material, using the same
properties of the metal conductor exposed in Table 3. The Fig. 2 represents this
configuration for the two models of soil stratification analyzed.

Table 3. Electrical parameters of simulated materials [9]

Material ρ (Ωm)  μ
Uniform soil 30.52 10 1
Air 0 1 1
Rod 1.81 × 10−8 106 100
Non-uniform soil layer 1 26.48 10 1
Non-uniform soil layer 2 421.4 10 1

For all analyzed cases, the frequency response of the systems was evaluated
by insertion of a current of amplitude 10 amps. The signal frequency was varied
from 1 Hz to 10 MHz in order to analyze the impedance behavior of the simulated
system over a large range of frequencies.
The evaluation parameters analyzed were: impedance of the grounding sys-
tem and distribution of surface potentials.

2.3 Evaluation of Simulated Parameters

In order to evaluate the results obtained with analytical expressions, the param-
eter analyzed was the ground resistance. The values obtained for the two ground
1506 G. R. S. Lira et al.

(a) homogeneous soil model (b) stratified soil model

Fig. 2. Representations of the two-rod grounding system

systems inserted in the two different soil models were compared with each
other and with the expressions. The objective was to determine the discrep-
ancy between the low frequency resistance values obtained through simulation
with those predicted by the technical literature.
According to [8], a rod embedded in homogeneous soil has its electrical resis-
tance given by:  
ρa 4L
R1rod = ln , (3)
2πL d
Where ρa is the apparent resistivity of soil (Ω . m) (value obtained to represent
a soil with several layers by only one layer), L is the length of rod (m) and d is
the diameter of the circle equivalent to the cross-sectional area of the rod (m).
For a two-rod grounding system, the ground resistance can be calculated
by the expression of resistances of the individual rods in parallel. However, the
individual resistance of each rod undergoes a change due to the insertion of new
rods [8]. Thus, when inserting a new grounding rod into a system that once had
only one, the individual resistance of each of the rods will no longer be given by
Eq. (3), but by expressions (4) and (5) [8]:


n
Rh = Rhh + Rhm , (4)
m=1m=h

Where Rh is the resistance presented by the rod, h, inserted in the set considering
the interference of other rods (Ω), Rhh is the individual resistance of each rod,
without the presence of others (given by Eq. (1)) (Ω), n is the number of rods in
parallel, Rhm is the increase in resistance at rod, h, due to mutual interference
of rod, h, given by the expression (5).
 
2
ρa (bhm + L) − e2hm
Rhm = ln 2 2 , (5)
4πL ehm − (bhm − L)
 Influence of Soil Stratification Models on Grounding Systems 1507

where ehm is the spacing between rod, h, and rod, m, (m), L is the length of the
rod (m), bhm is the distance between the lower end of rod, h, and upper tip of
rod, m, (m).

2.4 Obtaining Time-Domain Voltage

The waveform of the transient voltage during the injection of a current impulse
is also a relevant parameter for the evaluation of the performance of groundings
[3]. Thus, in order to analyze the difference between the potentials obtained at
the point of application of the current, it was considered the injection of an
impulsive current modeled as an exponential double function, according to the
expression (6):

i(t) = 94.12 × e−0.1322t − e−0.1361t (6)

In (6), i represents the current, in kA, and t is the time in μs. The parameters
of (6) were calculated using an optimization method to produce a pulse with
amplitude 1 kA and waveform 8/20 µs.
The Fourier transform of the modeled current according to (6) was calculated.
Then, using expression (7), the voltage in the frequency domain was determined
as:
V (ω) = Z(ω) · I(ω) (7)
Then, the voltage in the time domain was calculated using (8):
 ∞
1
v(t) = V (ω) · ejωt dω (8)
2π −∞

With the set of values obtained for the impedance of each of the grounding
systems for the established frequency band (considering the one and two-layer
model), it was possible to obtain the vector Z(ω). Applying the Fourier transform
in the signal of expression (6) and applying Ohm’s law in the frequency domain,
was obtained the signal V (ω), which represents the potential at the point of
application of the current in the frequency domain. Finally, using the inverse
Fourier transform, the signal v(t) was obtained for both soil stratification models.

3 Results Obtained

From the results of the simulations, the behavior of the ground systems sim-
ulated with the two models of soil stratification was analyzed. The results are
presented and discussed in this section. The results for the system with a single
rod is presented in Subsect. 3.1 and for the system with two rods is shown in
Subsect. 3.2.
1508 G. R. S. Lira et al.

3.1 System with a Single Ground Rod

In Fig. 3, the ground impedance values obtained for the single-rod grounding
by variating the frequency of the applied signal are presented for the frequency
range of 1 Hz to 100 kHz. Figure 4 shows the grounding harmonic impedance for
the frequency range between 100 kHz and 10 MHz. The impedance phase as a
function of frequency is shown in Fig. 5.

Fig. 3. Harmonic impedance absolute value obtained for the single-rod grounding
system

In situations where a low frequency signal is applied (range between 1 Hz

and 100 kHz), different impedance values were obtained for each of the adopted
stratification models. Using the two-layer soil, the ground impedance absolute
value was 8.3% higher than that the value achieved by using the homogeneous
soil model - a situation found in the 60 Hz frequency. In addition, for the fre-
quency range between 100 kHz and 10 MHz, the values obtained for the ground
impedance are practically coincident. Also, peaks in impedance values at high
frequencies are observed. These have as probable cause the resonance of the
system reached in these frequencies. With respect to the comparison of the low
frequency resistance with the expression (3), it can be observed that the resis-
tance obtained with the two-layer model best approximates the values of ground
resistance predicted by Eq. (1). In the table below the values of the ground loop
at the 60 Hz frequency are compared with the theoretical value (Table 4).
Following the procedure presented in Sect. 2.4, the transient voltages pre-
sented in Fig. 6 were obtained.
It can be observed that the potentials obtained considering the two-layer soil
model were slightly higher than when the homogeneous soil was adopted. In spite
of this, it is also seen that both impulses have the same shape. The voltage peaks
 Influence of Soil Stratification Models on Grounding Systems 1509

Fig. 4. Ground impedance absolute value for the single-rod grounding system at the
higher frequencies range

Fig. 5. Ground impedance phase for the single-rod grounding system at the higher
frequencies range

Table 4. Resistance values for the single-rod grounding system

Model Resistance (Ω) Percent error in relation

to the theoretical value
Homogeneous model 12.08 6.70
Soil with two layers 13.05 0.80
Theoretical value 12.95 -
1510 G. R. S. Lira et al.

Fig. 6. Potentials at the point of application of the impulsive current, for the models
of the homogeneous soil and the stratified soil (system with 1 rod)

obtained with the homogeneous and two layer soil were, respectively, 12.88 kV
and 13.42 kV, a percentage difference of 4.47%.

3.2 System with Two Ground Rods

In a similar way to the previous case, the behavior of grounding with two rods was
evaluated when considering the stratification models of the soil in one and two
layers. When evaluating the influence of the frequency on the system impedance,
it was found that, for the case where the adopted soil model was the two-layer
stratified, an increase in the value of about 78% occurred, thus demonstrating
that the two layers it is more difficult for the current to disperse in the soil. The
situation is illustrated in Fig. 7, which shows the simulated frequency response
for the two-rod system.
Table 5 presents the resistance values of the grounding at the frequency of
60 Hz in relation to the homogeneous soil, stratified soil in two layers and the
theoretical value calculated through (3–5).

Table 5. Resistance values for the two-rod grounding system

Model Resistance(Ω) Percent error in relation

to the theoretical value
Homogeneous model 5.71 22.47
Soil with two layers 10.14 37.67
Theoretical value 7.36 -
 Influence of Soil Stratification Models on Grounding Systems 1511

Fig. 7. Ground impedance in a two-rod system

Following the procedure presented in Sect. 2.4, the time domain voltage of
Fig. 8 was obtained.

Fig. 8. Potentials at the application point of the impulsive current, for the
homogeneous-soil and two-layers soil models, in a two-rod grounding

In the case of the two-rod grounding system, a greater difference was verified
between the use of the different models of soil stratification. The two-layer model
presented a time domain voltage peak value 42.2% higher than the single layer
model.

4 Conclusions
This paper analyzed the influence of soil stratification models on the resis-
tance, impedance and transient response of grounding systems. Computational
1512 G. R. S. Lira et al.

simulations of the frequency response of two grounding systems were performed.

The percentage difference between the impedance values obtained for the differ-
ent soil models had a variation large enough to impact the grounding design.
Differences up to 43.7% were obtained for the impedance at low frequency.
In the case studies carried out, it was verified that the homogeneous soil model
underestimates grounding impedance. By using simpler soil stratification models,
smaller ground impedances can be obtained, which will not faithfully reproduce
the system response. These simplifications may lead to worse-to-expected per-
formance against lightning or faults, when touch and step potential beyond the
limits stipulated in the design step may occur on the soil surface. According to
the results, it can be concluded that it is essential to design a grounding sys-
tem, mainly within substations, to avoid the representation of the soil by only a
homogeneous layer.

Acknowledgments. To light SA (Brazilian power utility) by support this research

in the scope of ANEEL (Brazilian regulation agency) R&D program. To the High
Voltage Laboratory of Federal University of Campina Grande by provide the necessary
infrastructure.

References
1. IEEE Std 81-2012: Guide for measuring earth resistivity, ground impedance, and
earth surface potentials of a grounding system (2012)
2. EG-0 Power System Earthing Guide, Part 1: Management Principles, ENA DOC
025-2010, May 2010
3. IEEE Std 80-2000: Guide for safety in AC substation grounding (2000)
4. Nassereddine, M., Rizk, J., Nasserddine, G.: Soil resistivity data computations; sin-
gle and two - layer soil resistivity structure and its implication on earthing design.
World Acad. Sci. Eng. Technol. Int. J. Electr. Comput. Eng. 7(1), 878–883 (2013)
5. Choi, J.-H., Lee, B.-H.: An analysis on the frequency-dependent grounding
impedance based on the ground current dissipation of counterpoises in the two-
layered soils. J. Electrostat. 70(2), 184–191 (2012)
6. Wenner, F.: A method of measuring resistivity. Natl. Bur. Stan. Sci. Pap. Rep. No.
258 12(3), 469–482 (1916)
7. NBR 7117: Earth resistivity measurement and soil stratification (2012)
8. Dwight, H.B.: Calculations of Resistance to Ground. Massachusetts Institute of
Technology, Cambridge (1936)
9. Comsol, A.B.: COMSOL multiphysics: material library (2018)
 Research on a New Type of Fault Current
Limiter Based on High Coupled Split Reactor

Wenxiong Mo1, Junxiang Liu1(&), Yong Wang1, Lin Gan1,

Haibo Su1, Lu Zhu1, Shengya Qiao1, Kaijian Wu2, and Zhao Yuan2
1
Guangzhou Power Supply Co., Ltd., Guangzhou, China
jxliu_2006@126.com
2
State Key Laboratory of Advanced Electromagnetic
Engineering and Technology, Wuhan, China

Abstract. With the rapid development of the power system, the short circuit
current level has increased sharply. In order to ensure security, stability and
reliable operation of power system, the fault current limiter (FCL) is one of the
effective measures to limit short-circuit current. The use of high coupled split
reactor (HCSR) to achieve limit and interruption of large current is a new
technique emerged in recent years. The FCL based on HCSR has many
advantages compared with other types of FCL, but it puts big pressure on the
insulation design of HCSR and affects the recovery voltage of fast mechanical
switch. The paper proposes an improved FCL topology, which retains the
advantages of HCSR and on the other hand reducing the insulation requirements
for HCSR. Focusing on the FCL based on HCSR, a simulation model is
established to analyse the effect of current limiting and over-voltage under
different working conditions. The effect of current limiting of FCL is in line with
the expected results, and setting capacitors in parallel with FCL is an effective
method to reduce the rate of rise of recovery voltage (RRRV) of main switch.

Keywords: Fault current limiter
HCSR
Recovery voltage

1 Introduction

With the continuous expansion of the scale of power grids, the magnitude of short-
circuit current has increase sharply. If no measures are taken, the short-circuit current in
high load density networks even exceeds the interrupting capabilities of the existing
circuit breakers. For example, the bus bar short-circuit current of some 500 kV sub-
stations in Guangdong power grid has exceeded 80 kA [1], while rated short-circuit
current of existing circuit breakers is 63 kA. In order to solve above problems, it is
necessary to develop circuit breakers with larger interrupting capacity, but it is extre-
mely difficult to manufacture circuit breakers with large capacity due to the limitation
of materials, processing technology, etc. However, certain measures could be taken to
limit the short-circuit current, and the fault current limiter (FCL) is one of the effective
solution to achieve it.
The measures to limit current in system-level are forced to sacrifice the transmis-
sion capacity and power supply reliability, while the FCL is a more competitive

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1513–1522, 2020.
https://doi.org/10.1007/978-3-030-31676-1_141
1514 W. Mo et al.

solution to control the short circuit current [2–4]. The FCL exhibits low impedance in
normal operation condition of power grid, and has no effect on the system. When a
fault occurs, the FCL changes to a high impedance and limits short circuit current.
There are many types of FCL, each one has its own characteristics and advantages.
Among them, the FCL based on high coupled split reactor (HCSR) is a relatively
economical solution and has many advantages compared with other types of FCL.
This paper analyses the current transfer and current limiting process of the FCL
with HCSR, and points out the problems exist in topology. In order to solve the
problems, an improved FCL topology is proposed. Meanwhile the current limiting
effect and over-voltage of a 500 kV FCL based on HCSR are analysed through sim-
ulation [5–8], which can provide reference for the selection of HCSR and fast
mechanical switch.

2 Working Principle of FCL Based on HCSR

and Improvement

HCSR is a kind of split reactor which has two reverse coupled windings. When the
current pass through two coupled windings of the HCSR simultaneously, HCSR shows
the small leakage inductance outside and has little effect on power system. When only
one winding of HCSR is connected to the grid, it is presented as a large inductance that
could limit the fault current. The current sharing and limiting characteristic of HCSR
can be applied to develop circuit breakers with larger interrupting capacity. The par-
alleled circuit breakers with HCSR are shown in Fig. 1, which composed of HCSR and
two circuit breakers. Each circuit breaker is connected in series with a winding of
HCSR respectively, thus forming two paralleled branches. In the normal operation
condition, currents flowing through two branches are identical, and the magnetic field
of HCSR’s two windings have opposite direction and almost identical value, which
means the leakage reactance of HCSR is really small. If one of the circuit breakers
doesn’t quenching the arc while the other does at a current zero point, HCSR will
changes roles as a current limiting reactor, which could limit the fault current and make
sure that the circuit breaker could interrupt the limited fault current successfully.

HCSR
CB2
CB1

Fig. 1. Schematic of paralleled circuit breakers.

The paralleled circuit breakers with HCSR have been applied in the occasion which
high current interruption is in need. The 126 kV/5000 A – 80 kA paralleled circuit
breakers have been connected with electric power net in Wuhan, China as shown in
Fig. 2, and the 252 kV/3150 A – 101 kA paralleled circuit breakers has successfully
 Research on a New Type of Fault Current Limiter 1515

interrupted 252 kV/101 kA short circuit current in the laboratory, and it is connected
with electric power net in Guangzhou, China, as shown in Fig. 3.

Fig. 2. 126 kV/5000 A – 80 kA paralleled circuit breakers prototype.

Fig. 3. Experimental diagram of 252 kV/3150 A – 101 kA paralleled circuit breakers.

In order to limit the short-circuit current more quickly and minimize the impact on
the equipment in 500 kV power system, the FCL based on HCSR is proposed as shown
in Fig. 4. The FCL composed of HCSR and a fast mechanical switch. In the normal
operation condition, the current is proportionally distributed between two branches and
the FCL only exhibits a small leakage inductance to the system. Once fault occurs, the
fast mechanical switch disconnects at first current zero point, thus limiting the fault
current and make sure the main circuit breaker in the system interrupts the limited fault
current successfully.
1516 W. Mo et al.

Fig. 4. Schematic of FCL based on HCSR.

The advantage of above topology is that the fast mechanical switch does not need to
interrupt the full capacity short-circuit current to achieve current transferring. When the
inductance of HCSR two windings is equal, the fast switch only needs to interrupt half
of the total short-circuit current. However, there is an insulation problem with the
topology. Since the two windings of HCSR is tightly coupled, although the fast
mechanical switch is turned off, the induced voltage of the disconnected winding exists,
which puts big pressure on the insulation design of HCSR, and affects the recovery
voltage of fast mechanical switch. When HCSR is applied to high voltage or ultra-high
voltage power system, the insulation requirement of HCSR is very high, which makes
it difficult to design HCSR as two windings of HCSR need to be highly coupled.
In order to solve the problems above, an improved FCL topology is proposed as
shown in Fig. 5. The improved FCL consists of a HCSR and two fast mechanical
switches. When the fault occurs, the fast mechanical switches (CB1 and CB2) are
disconnected at the same time, and the fault current is limited by a winding of HCSR.

C HCSR D
CB1 CB2
A B

Fig. 5. Schematic of the improved FCL based on HCSR.

In order to reduce the insulation requirements for HCSR, the way of inlet and outlet
of HCSR two-arm windings is adjusted. As shown in Fig. 6, when CB1 and CB2 are
disconnected, suppose that the potential at point A is U and the potential at point B is 0,
the voltage that needs to be tolerated between encapsulations of HCSR is approxi-
mately equal to 0. Compared with the original topology of FCL, the insulation
requirement of HCSR has dropped from 2U to 0.
 Research on a New Type of Fault Current Limiter 1517

A D
0 U CB2
0
0
0
0 0
0
0

B 0 C

CB1

Fig. 6. Diagram of potential distribution in the improved FCL.

The improved FCL topology retains the advantages of HCSR, eliminating the need
to transfer full-capacity fault currents, and on the other hand reducing the insulation
requirements for HCSR. However, for the fast mechanical switches and main circuit
breaker in the system, the insertion of the HCSR’s inductance may have an impact on
the interrupting process and the transient recovery voltage (TRV) of circuit breakers.
Focusing on the improved FCL applied in 500 kV power system, a simulation model is
established in PSCAD-EMTDC to analyse the current limiting effect and the over-
voltage distribution under different working conditions, and the parameter design
method of each component is proposed.

3 Simulations and Results

The simulation is based on PSCAD-EMTDC, and the ideal circuit breaker model in
PSCAD is adopted to simulate the fast mechanical switches and main circuit breakers.
Therefore, the TRV obtained by the simulation is only related to the parameters of
system components. In order to simulate the operating conditions when FCL connected
to the 500 kV system, firstly the system equivalent circuit model is established. In this
paper, the system side model is established based on the actual installation location of
the FCL prototype, and some lines, transformers, circuit breakers, bus bars and loads
are retained. The equivalent circuit diagram is shown in Fig. 7. The FCL is located in
the outgoing line of substation B, and connected to substation A through 66 km lines.
The FCL prototype has 8 fast mechanical switches as shown in Fig. 8.

A 66km Fault B
A1 B1
FCL
Line
A2 B2

Fig. 7. Schematic of the FCL in the grid.

1518 W. Mo et al.

CB1 CB2 CB3 CB4 CB5 CB6 CB7 CB8

Fig. 8. Schematic of the FCL prototype.

Suppose that the short circuit current is 90 kA, the inductance of each winding in
HCSR is 8 mH, and the coupling coefficient of HCSR is set to 0.97. The voltage-
sharing capacitance across each fast switch is 2000 pF.

3.1 Simulation Results Without FCL

Suppose that the FCL is not connected to power system, and a three-phase short circuit
to ground occurs as shown in Fig. 7. The main switch B1 is set to operate at t = 0.13 s,
and circuit breaker in B-phase is the first to complete the interrupting process. The
effective value of short circuit current is shown in Table 1. The TRV and the rate of rise
of recovery voltage (RRRV) of main switch B1 in B-phase are shown in Table 2.

Table 1. Simulation results of short circuit current.

Fault types Current at Current flows
fault point (kA) through B1 (kA)
Three-phase grounding fault 106.1 90.5
Single phase grounding fault 84.1 73.6
Inter-phase short circuit 96.6 81.7
Two phases grounding fault 99.4 83.4

Table 2. Simulation results of TRV (main switch B1).

Fault types Peak of TRV (kV) RRRV (kV/ls)
Three-phase grounding fault 815.5 0.82
Single phase grounding fault 698.8 0.69
Inter-phase short circuit 649.0 0.83
Two phases grounding fault 747.1 0.75

As shown in Tables 1 and 2, when the FCL is not put into the system, the effective
value of three-phase grounding short-circuit current flowing through the installation
position of FCL is 90.5 kA. The TRV peak of main switch B1 is 815.5 kV, and RRRV
is 0.82 kV/ls.
 Research on a New Type of Fault Current Limiter 1519

3.2 Simulation Results with FCL

When the FCL is connected to power system as shown in Fig. 7, suppose that a three-
phase short circuit to ground occurs. The FCL is set to operate at t = 0.11 s, and the
single winding of HCSR is put into the system. The FCL in B-phase is the first to
complete the current transfer and limiting process. The effective value of limited short
circuit current is shown in Table 3. The TRV and RRRV of main switch B1 in B-phase
are shown in Table 4, while the results of fast mechanical switches are shown in
Table 5.
Simulation results show that the improved FCL limits the fault current from
90.5 kA to 51.6 kA when a three-phase short circuit to ground occurs, which is in line
with the expected results. When fault occurs, the fast switches need to break 45 kA
short-circuit current to complete current transfer process, and the recovery voltage
waveforms of fast switches are basically the same. As shown in Table 5, the breaking
capacity requirement for each fast switch is successfully reduced. Comparing the
simulation results with Tables 2 and 4, it could be found that when FCL is connected to
power system, the RRRV of main switch B1 sharply rises, which is far beyond the
value in the standard (2 kV/ls).

Table 3. Simulation results of limited current.

Fault types Current at Current flows
fault point (kA) through B1 (kA)
Three-phase grounding fault 67.9 51.6
Single phase grounding fault 56.3 45.6
Inter-phase short circuit 59.8 44.9
Two phases grounding fault 63.1 48.3

Table 4. Simulation results of TRV (main switch B1).

Fault types Peak of TRV (kV) RRRV (kV/ls)
Three-phase grounding fault 803.7 14.3
Single phase grounding fault 705.8 12.8
Inter-phase short circuit 636.5 9.4
Two phases grounding fault 753.0 13.8

Table 5. Simulation results of TRV (fast switches).

Fault types Peak of TRV (kV) RRRV (kV/ls)
Three-phase grounding fault 83.1 0.13
Single phase grounding fault 69.4 0.11
Inter-phase short circuit 61.7 0.11
Two phases grounding fault 75.4 0.12
1520 W. Mo et al.

During the current transfer and limiting process, when the fast switches interrupt
current in one winding of HCSR, HCSR is connected to the grid in the form of current-
limiting inductance and makes up a high-frequency oscillation circuit with the stray
capacitance [9, 10]. High frequency oscillations are superimposed on the TRV
waveform as shown in Fig. 9, therefore affects recovery voltage of the main switch
which is open later. Even though HCSR limits the fault current, the actual interrupting
process of the main switch in power system may be more difficult. Hence, restrictions
are needed to reduce the RRRV.

(a) Three-phase grounding fault

(b)Inter-phase short circuit

Fig. 9. TRV waveform of Main switch B1.

3.3 Solutions to Reduce RRRV of Main Switch

From the analysis above, it can be seen that in order to reduce RRRV, the equivalent
frequency of the oscillation loop formed by HCSR and stray capacitance needs to be
changed. Therefore, setting capacitors in parallel with FCL is an effective method to
reduce RRRV of main switch. The effect of installing capacitors is shown in Table 6.
When the FCL is connected in parallel with 600 nF capacitors end to end, the RRRV of
main switch is reduced to 2 kV/ls, but the peak of TRV rises slightly.
 Research on a New Type of Fault Current Limiter 1521

Table 6. Simulation results of TRV (with capacitors).

Capacitors Peak of TRV (kV) RRRV (kV/ls)
500nF 845.03 1.97
600nF 835.8 1.78

The impact of HCSR on RRRV of main switch is analysed below through simu-
lation. It should be noted that the data of RRRV mentioned below may differ from
reality, as the stray capacitance of some devices (transformers, VT, HCSR, and so on)
are not accurate enough, but this problem needs to be taken into account when
designing the FCL based on HCSR.

4 Conclusions

According to the improved FCL based on HCSR, the equivalent circuit model is
established in PSCAD and the current transfer and limiting process is analysed through
simulation. It can be concluded that:
(1) The improved FCL based on HCSR retains the advantages of HCSR, eliminating
the need to transfer full-capacity fault currents, and on the other hand reducing the
insulation requirements for HCSR.
(2) The effect of current limiting of FCL is in line with the expected results.
Although HCSR limits the fault current, the actual interrupting process of the
main switch in power system may be more difficult as the RRRV rises.
(3) Setting capacitors in parallel with FCL is an effective method to reduce RRRV of
main switch, but the specific value of capacitors needs to be calculated and
verified with accurate system parameters.

Acknowledgments. This Project was supported by National Key R&D Program of China
(2018YFB0904300), and the Science and Technology Project of China Southern Power Grid
Co., Ltd. (GZHKJXM20180055).

References
1. Yuan, Z., Yin, X., Pan, Y., He, J.: Current dividing process of paralleled circuit breakers
with high coupled split reactor. High Voltage Eng. 38(8), 2008–2014 (2012)
2. Yin, T., Chen, X., Du, Y.: Design of large-capacity high-voltage breakers using tightly-
coupled reactor. High Voltage Eng. 36(6), 1577–1582 (2010)
3. Chen, X., Yin, T., Pan, Y.: High voltage and large capacity vacuum circuit breakers with
series-parallel structure. High Voltage Eng. 37(12), 3157–3163 (2011)
4. Du, Y., Chen, X., Liu, F., et al.: Design of a novel parallel circuit breakers. In: International
Conference on Electrical and Control Engineering, pp. 3558–3560. IEEE Computer Society,
Wuhan (2010)
1522 W. Mo et al.

5. Hongshun, L., Qingmin, L., Jie, L.: Impact of inductive FCL on the RRRV of circuit
breakers. Trans. China Electrotechnical Soc. 22(12), 84–91 (2007)
6. Calixte, E., Yokomizu, Y., Shimizu, H., et al.: Reduction of rating required for circuit
breakers by employing series-connected fault current limiters. IEE Proc.-Gener. Transm.
Distrib. 151(1), 36–42 (2004)
7. Szewczyk, M., Kulas, S.: Investigation on circuit breaker influence on transient recovery
voltage. In: 42nd International Universities Power Engineering Conference, pp. 1036–1043.
IEEE, Brighton (2007)
8. Calixte, E., Yokomizu, Y., Shimizu, H., et al.: Theoretical expression of rate of rise of
recovery voltage across a circuit breaker connected with fault current limiter. Electr. Power
Syst. Res. 75(1), 1–8 (2005)
9. Li, J., Li, F., Han, J.: Effects of 500 kV bus sectionalizing reactor on transient recovery
voltage and countermeasures. High Voltage Apparatus 50(10), 63–69 (2014)
10. Geng, S., Zhang, J., Zhang, M.: Influence of serial-resonant FCL on the characteristics of
circuit breakers during short-circuit faults. Power Syst. Prot. Control 37(18), 32–37 (2009)
 Dynamic Non-linearity of Impulse
Digitisers and Its Effects on Impulse
Voltage Measurement

Wei Yan(&) and Yi Li

National Measurement Institute Australia,

36 Bradfield Road, West Lindfield, NSW, Australia
wei.yan@measurement.gov.au

Abstract. Dynamic non-linearity of analogue-to-digital converters (ADCs)

used in digitisers may affect the accuracy of impulse measurements. IEC 61083-
1:2001 defines the method for determining dynamic differential non-linearity
(DNL) of impulse digitisers and sets the acceptance limit for the DNL of
impulse digitisers. However, the errors of impulse parameters that result from a
DNL within such a limit are yet to be studied. Here the DNL of several com-
mercial digitisers has been determined using triangular and sinusoidal test
voltages. The results show that the difference in dynamic DNL values obtained
with the two types of test voltages was negligible compared with the acceptance
limit set in IEC 61083 1:2001. A new parameter, the dynamic integral non-
linearity (INL) was calculated from the dynamic DNL of the digitiser. Then, the
measured values of code bin width (CBW) of the digitisers were derived from
their DNL and used for the digitisation of ideal lightning impulse waveforms.
The lightning impulse parameters of the ideally digitised waveforms and those
of the waveforms digitised with the measured CBW were compared. The results
showed that the parameter errors caused by the dynamic non-linearity of the
digitisers were more affected by the shape of their INL curves, rather than the
DNL. The differences in the errors of front time (T1) and time to half-value (T2)
caused by the two types of test voltage for obtaining the CBW were within 0.9%
and 0.6%, respectively. The method used in this study may be used for esti-
mating the uncertainties of lightning impulse measurement caused by dynamic
non-linearity of the digitiser.

Keywords: Digitisers
Dynamic non-linearity
Impulse voltage measurement

1 Introduction

Digital recorders, or digitisers, are indispensable components of modern impulse

voltage measurement systems. The analogue-to-digital converter (ADC) systems of
digitisers inherently exhibit static and dynamic non-linearity, causing distortion of the
recorded waveforms and resulting in errors in the waveform parameters. For impulse
measurement, more attention should be paid to dynamic non-linearity as the rate of
voltage change for impulses is high. IEC 61083-1:2001 [1] has defined the acceptance
limit of dynamic differential non-linearity (DNL) for approved impulse digitisers.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1523–1532, 2020.
https://doi.org/10.1007/978-3-030-31676-1_142
1524 W. Yan and Y. Li

However, the errors of impulse parameters caused by DNL within this limit have yet to
be studied.
Triangular and sinusoidal test voltages are commonly used for characterising the
dynamic non-linearity of ADCs [1–3]. In this study, triangular and sinusoidal test
voltages were used to first obtain the code occurrence histograms of three commercial
digitisers. Then, the DNL, dynamic integral non-linearity (INL), and code bin widths
(CBWs) of the digitisers were determined from their code occurrence histograms.
Finally, the CBWs of the digitisers were used to simulate digitisation of the ideal
impulse waveforms. The differences of the parameters between waveforms digitised
with ideal CBWs and measured CBWs were calculated. The results show that the
differences of DNL caused by different types of test voltage was negligible compared
with the acceptance limit set in [1]. Moreover, the accuracy of the measured parameters
was more likely to be affected by INL as INL reflects the cumulative non-linearity of
the digitiser. The differences in DNL caused by different types of test voltages were
determined. The influence of DNL and INL on the accuracies of the measured
parameters is discussed.

2 Experiment

2.1 Instrumentation and Test Settings

The 100 mV per division (mV/div) voltage scales of a Tektronix TDS 540C, a Tek-
tronix TDS 784D, and a Keysight MSOS 804A digitiser were tested. All three digi-
tisers have an 8-bit voltage resolution. Channel input impedances were set to 1 MΩ.
An analogue function generator was used for generating triangular and sinusoidal test
voltages for obtaining the code occurrence histograms of the digitisers. The frequency
of triangular test voltage was 2.48 MHz, which was calculated to meet the requirement
for digitisers used for measuring lightning impulse voltages with nominal front time of
0.84 µs according to [1]. The same frequency was used for sinusoidal test voltage. The
amplitude of the test voltages was set so that the peak-to-peak voltage of each
waveform was greater than 95% of the full scale deflection (f.s.d.) of the digitiser.

2.2 DNL Determination with Triangular and Sinusoidal Test Voltages

From the histograms obtained with triangular test voltages, DNL was determined from
Eq. 1 [1].

n½ k   n0
d ½k  ¼ ð1Þ
n0

Where d ½k  is DNL of code k, n½k is the recorded occurrence of code k, and n0 is

the average occurrence count in the range of code under examination.
 Dynamic Non-linearity of Impulse Digitisers and Its Effects 1525

With the histograms obtained with sinusoidal test voltages, DNL was determined as
the difference between the actual code occurrence and the theoretical occurrence. The
theoretical occurrence was calculated from Eq. 2 [3].



Bðk2N1 Þ Bðk12N1 Þ
L
p½k ¼ L
p1
arcsin  arcsin
A
2N A
2N ð2Þ
k 2 ½0; 2N1 

where L is the number of recorded points, p½k  is the probability of the occurrence of
code k, B is the f.s.d., A is the amplitude of the sinusoidal waveform, and N is the
number of bits of the digitiser. DNL was then calculated from Eq. 3.

n½k  L
p½k 
d ½k  ¼ ð3Þ
L
p½k 

where d ½k is the DNL of code k and n½k  is the recorded occurrence of code k.
For each type of test voltage, the occurrence histograms were generated for 40,000
periods of the test voltage, which produced more than 8 million sample points, so that
the influence of noise and the statistical uncertainty on the measured DNL could be
minimised.
The calculation of DNL defined in the present IEC standard [1] using triangular test
voltage removes the effect of the offset. The sum of d ½k for all codes is therefore
always zero due to this definition. However, the sum of d ½k may not be zero for the
DNL determined with the sinusoidal test voltage using Eq. (3). This is due to the
calculation of the DNL with the occurrence histogram recorded using sinusoidal test
voltage, where the DNL can be affected by the inaccuracy of the measured amplitude
(A in Eq. 2), and any offset in the source voltage and the digitiser. The offset of the
DNL determined with sinusoidal test voltage was removed using Eq. 4.

0
X
N 1
d ½k ¼ d½k  d½i=N ð4Þ
i¼0

Where d ½k is the DNL of code k, d 0 ½k is the offset-free DNL of code k, and N is the
total number of codes under investigation.
To ensure the DNL derived from the histograms reflected the behaviour of the
digitiser, instead of originating from the function generator, histograms were recorded
with a 12-bit function generator. The results show that the differences in the DNL
caused by different function generators were negligible compared to the average
magnitude of the DNL. Therefore, the effect of the function generator’s non-linearity
on DNL determination was neglected in this study.

2.3 Determination of Dynamic Integral Non-linearity and Code Bin

Widths
As dynamic INL for impulse digitisers is defined in relevant standards, we adopted the
definition of static INL in [1] and used Eq. 5 to determine dynamic INL from DNL.
1526 W. Yan and Y. Li

X
k
S½ k  ¼ d ½i=ðk2  k1 Þ  100% ð5Þ
i¼0

where: S½k  is the INL of code k in percentage of f.s.d.; d ½i is the DNL of code i; k1 and
k2 are the lowest and highest codes, respectively, of the code range being measured.
The experimental values of CBW were calculated from measured d ½k  using Eq. 6.

w½k ¼ w0
ð1 þ d ½kÞ ð6Þ

where w½k is the CBW of code k, w0 is the ideal CBW and d ½k  is the DNL of code k.

2.4 Waveform Digitisation

The measured CBWs of the digitisers were used to simulate the digitisation of ideal
lightning impulse waveforms on a computer. The ideal waveforms were also digitised
with linear CBWs. The peak voltage (Ut), front time (T1), and time to half-value (T2) of
the digitised waveforms were then calculated according to IEC 60060-1:2010 [5] and
the parameters were compared. The process of the simulated digitisation and the result
comparison is shown in Fig. 1. Three peak voltage values for the ideal lightning
impulse were chosen, which corresponded to 58%, 78%, and 97% of the f.s.d. Peak
voltages less than 50% of the f.s.d. are impractical and should not be used for 8-bit
digitisers as the effective voltage resolution would be too low for impulse voltage
measurement.

Fig. 1. The process of simulated digitisation and result comparison.

 Dynamic Non-linearity of Impulse Digitisers and Its Effects 1527

3 Results and Discussion

3.1 DNL Determined with Triangular and Sinusoidal Test Voltages
Figure 2 shows column charts of the DNL values determined with triangular and
sinusoidal test voltages. The code range under examination was from about 30 to 220.
As the amplitudes of triangular and sinusoidal test voltages were not exactly the same
due to the accuracy of the function generator, the range under examination with the two
types of voltage can be slightly different. It can be seen that, in most situations, the
local peaks of DNL appear at the same codes for both types of test voltage. The
maximum absolute values of DNL (Dm) and their occurrence code with different test
voltages for each digitiser are shown in Table 1. The range of Dm measured for the
three digitisers and two test waveforms is 0.32 ± 0.02. By comparison, Dm shall be
less than 0.8 for approved digitisers for impulse voltage measurement, as defined in [1].
The results indicate that the tested digitisers meet this requirement, and the difference in
Dm caused by the type of test voltage used for DNL determination is negligible.
Furthermore, Dm occurs at the same code independently of the test voltage for the
TDS 540C and TDS 784D, whereas it occurs at different codes for the MSOS 804A.
This is because there are multiple negative local peaks of DNL with similar values (e.g.
at codes 174, 195, and 223) for this digitiser as evident in Fig. 2(c). A slight difference
of DNL measurement could result in a different occurrence code.

3.2 INL Determined with Triangular and Sinusoidal Test Voltages

The INL values calculated from the DNL of the digitisers, corresponding to triangular
and sinusoidal test voltages, are shown in Fig. 3. The INL curves obtained with
sinusoidal test voltages have similar shapes to those obtained with triangular test
voltages but with a positive offset. This offset may result from the inequality of the
amplitudes and offsets used for constructing the ideal code occurrence histograms and
those of the actual sinusoidal test voltages applied in the test. Although an offset
removal process has been implemented in determining the DNL (Eq. 4), the INL values
still show a residual offset. As a result, the maximum absolute values of INL (Im)
determined with the two types of test voltage show relatively large difference, and the
maximum Im does not always occur at the same code on the INL curves with different
test voltages, as shown in Table 2.

3.3 Simulated Digitisation of Ideal Lightning Impulse Waveforms

Although the difference in Dm among different digitisers was less than 0.04, inde-
pendently of the test waveform (Table 1), the errors in voltage and time parameters of
the waveforms digitised with their CBWs were large and inconsistent. Different errors
in the parameters were obtained, depending on the peak voltage of the digitised impulse
waveforms and the test voltage used for obtaining the CBWs (through DNL as in
Eq. 6).
In the simulated digitisation process, the zero levels of the ideal lightning impulse
waveforms were fixed at code 0, and the Ut of the ideal lightning impulse waveforms
1528 W. Yan and Y. Li

Fig. 2. Differential non-linearity of (a) TDS 540C, (b) TDS 784D, and (c) MSOS 804A
determined with triangular and sinusoidal test voltages.
 Dynamic Non-linearity of Impulse Digitisers and Its Effects 1529

Table 1. The maximum absolute values of DNL (Dm) and their occurrence codes
Triangular Sinusoidal
test voltage test voltage
Digitiser Dm Code Dm Code
TDS 540C 0.34 226 0.34 226
TDS 784D 0.30 32 0.32 32
MSOS 804A 0.31 195 0.31 174
Dm: maximum absolute value of DNL

(58%, 78%, and 97% of the f.s.d.) appeared around codes 148, 199, and 247 respec-
tively. The errors of the digitised waveform parameters are shown in Tables 3, 4 and 5.
The results show that Ut is mostly influenced by the value of INL at the code where the
peak voltage appears. Larger variations in Ut error are found at low values of Ut. This is
because the corresponding INL curves, determined from triangular and sinusoidal test
voltages, deviate more significantly at mid and low code ranges, as can be seen in
Fig. 3. The Ut errors become zero at high code ranges because the INL is zero at such
code values, e.g. INL curves at code 247 are zero for both test voltages and all
digitisers, so it is apparent that the Ut errors are zero. However, it has to be noted that
zero INL in the high code range is caused by the DNL calculation procedure defined in
[1]. It may not reflect the real INL of the digitisers. More work needs to be done to
reveal the influence of INL on Ut error in the high code range. It should also be noted
that the filter used in calculating the impulse parameters [5] would significantly reduce
the effects of dynamic DNL on the errors of the impulse parameters.
The errors of time parameters are more sensitive to the integral voltage non-
linearity [6]. The impact of INL on T1 and T2 depends on the INL at the codes
corresponding to the instants when the voltage reaches 30% and 90% of Ut on the
rising slope and 50% of Ut on the falling slope, according to their definitions [5].
Therefore, the difference of T1 and T2 errors between different digitised waveforms
results from the shape of the corresponding INL curves.
The results show that the maximum T1 error is 2% and the maximum difference of
T1 error between the waveforms digitised with different CBWs is 0.9%. Both cases
occurred at Ut of 58% of f.s.d. with the MSOS 804A digitiser. On the other hand, the
maximum T2 error is 1% and the maximum difference of T2 error between the
waveforms digitised with different CBWs is 0.6%. The difference in measured CBWs
(hence INL shapes) is considered to be mainly due to the deviations of the actual test
voltage waveforms (triangular or sinusoidal) from their ideal counterparts. The other
cause could be the difference in the frequency components contained in the two types
of test voltage waveforms, leading to different digitiser responses.
1530 W. Yan and Y. Li

Fig. 3. Integral non-linearity of (a) TDS 540C, (b) TDS 784D, and (c) MSOS 804A determined
with triangular and sinusoidal test voltages.
 Dynamic Non-linearity of Impulse Digitisers and Its Effects 1531

Table 2. The maximum absolute values of INL (Im) and their occurrence codes
Triangular test Sinusoidal test
voltage voltage
Digitiser Im (% of f.s.d.) Code Im (% of f.s.d.) Code
TDS 540C 0.26 211 0.40 223
TDS 784D 0.41 37 0.35 111
MSOS 804A 0.40 150 0.59 150
Im: maximum absolute value of INL

Table 3. Errors caused by non-linear code bin widths of TDS 540C

Ut Test voltage for Ut Difference T1 Difference T2 Difference
(% of f.s.d) determining error of Ut error error of T1 error error of T2 error
CBWs (%) (%) (%) (%) (%) (%)
58 Triangular −0.1 0.3 0.1 0.5 −0.4 0.1
Sinusoidal −0.4 −0.4 −0.3
78 Triangular 0.3 0.3 0.5 0.5 −1.0 0.1
Sinusoidal 0.0 0.0 −0.9
97 Triangular 0.0 0.0 −0.3 0.2 −0.4 0.3
Sinusoidal 0.0 −0.1 −0.7
Error of the highest amplitude −0.4 0.5 −1.0

Table 4. Errors caused by non-linear code bin widths of TDS 784D

Ut Test voltage for Ut Difference T1 Difference T2 Difference
(% of f.s.d) determining Error of Ut error error of T1 error Error of T2 error
CBWs (%) (%) (%) (%) (%) (%)
58 Triangular −0.2 0.2 −0.3 0.1 0.5 0.3
Sinusoidal −0.4 −0.4 0.2
78 Triangular 0.0 0.0 0.1 0.3 0.2 0.6
Sinusoidal 0.0 0.4 −0.4
97 Triangular 0.0 0.0 0.2 0.2 −0.1 0.5
Sinusoidal 0.0 0.0 −0.6
Error of the highest amplitude −0.4 −0.4 −0.6
1532 W. Yan and Y. Li

Table 5. Errors caused by non-linear code bin widths of MSOS 804A

Ut Test voltage for Ut Difference T1 Difference T2 Difference
(% of f.s.d) determining error of Ut error error of T1 error Error of T2 error
CBWs (%) (%) (%) (%) (%) (%)
58 Triangular −0.1 0.0 1.1 0.9 −1.0 0.0
Sinusoidal −0.1 2.0 −1.0
78 Triangular 0.0 0.0 0.0 0.8 −0.2 0.5
Sinusoidal 0.0 0.8 −0.7
97 Triangular 0.0 0.0 0.0 0.2 −0.3 0.4
Sinusoidal 0.0 0.2 −0.7
Error of the highest amplitude −0.1 2.0 −1.0

4 Conclusion

It can be concluded that the type of test voltage, i.e. triangular or sinusoidal, does not
significantly affect the measured DNL values and their maxima Dm for the digitisers we
tested. Also, the values of Dm determined with either test voltage were much lower than
the acceptance limit set in IEC 61083-1:2001 [1]. However, the errors in the parameters
of the lightning impulse waveforms digitised with CBWs, derived with different types
of test voltage, can be quite different. These errors are more dependent on the peak
voltage of the lightning impulse and the shape of the INL curve of the digitiser. The
errors in the waveform parameters resulting from the non-linearity of the digitisers may
be used as components of uncertainty in lightning impulse measurement. It is also
possible to use dynamic INL values to make corrections for the measured lightning
impulse parameters if it can be demonstrated that the effect of the distortion of the test
voltage source is negligible. Further work is needed in this area.

References
1. IEC 61083-1.Ed. 2.0: Instruments and software used for measurement in high-voltage impulse
tests-Part 1 (2001)
2. Liu, F., Chen, J.: Development of a calibration system of digital recorders for impulse tests.
In: Conference on Precision Electromagnetic Measurements Digest, pp. 250–251. IEEE
(2008)
3. Texas Instruments: Application Report: Dynamic Tests for A/D Converter Performance
(2015). http://www.ti.com/lit/an/sbaa002a/sbaa002a.pdf
4. McComb, T.R., Malewski, R.A.: Digital waveform recorders (digitizers). Electra 171, 79–99
(1997)
5. IEC 60060-1.Ed.3.0: High-voltage test techniques-Part 1 (2010)
6. Li, Y.: Estimation of time parameter uncertainties due to voltage uncertainties in lightning
impulse voltage measurement systems. In: International Symposium on High Voltage
Engineering, Hannover, Germany (2011)
 Non-linear Dynamic Response
of a Ground Rod in Resistive Soils

R. R. Diaz1,2(&), J. N. Silva1, and A. Parellada1

1
National University of Tucuman, Av. Independencia 1800,
San Miguel de Tucumán, Argentina
rdiaz@herrera.unt.edu.ar
2
CONICET, G. Cruz 2290, Buenos Aires, Argentina

Abstract. The increasing use of intelligent electronic devices (IEDs) in sub-

stations, for supervisory control and data acquisition in power systems, makes
critical the understanding of transients in grounding electrodes under lightning
surges. Thus, acquiring greater knowledge of the dynamic response of
grounding systems becomes strategic. The flow of high currents in earthed
electrodes can produce high electric fields and the consequent ionization of the
soil. The non-linear conduction mechanism in the soil provokes a time-
dependent reduction of the impulse resistance. This reduction may be very
pronounced for several tens of microseconds and the effect can be advantageous
for the insulation coordination with soils of high resistivity, eventually limiting
the harmful inductive behaviour depending on frequency.
Previous experimental work and computer simulation in a laboratory scale
model, with a single soil resistivity value, demonstrated the importance of the
energy injected for the ionization of the soil around the earthed electrode.
A simplified physical model of the discharge progression in soils was proposed
in order to explain the relationship between energy and transient grounding
resistance. The model included waveform, peak value and rise time of the
injected current. Extending the proposed method, this contribution presents and
analyzes new experimental results obtained for several values of soil resistivity,
between 290 and 3800 Xm. Finally, from the experimental results obtained with
this approach, a simple and practical procedure is proposed to estimate transient
resistance reduction in actual vertical ground rods buried in soils having a
different resistivity.

Keywords: Soil ionization
Ground rod
Transient resistance

1 Introduction

The function of grounding systems is the protection of equipment and humans against
over-voltages. Ground rods are conventional parts of systems widely used to facilitate
the evacuation of lightning stroke currents, particularly in high resistive soils. The
growing use of IEDs for supervisory control and data acquisition requires mastering the
dynamic response of grounding systems in substations, and this paper is aimed at
providing some tools to design ground rods that take impulse response into account.

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1533–1543, 2020.
https://doi.org/10.1007/978-3-030-31676-1_143
1534 R. R. Diaz et al.

Stroke lightning currents in grounding systems can produce high electric fields and
ionization of the soil around earthed electrodes. Soil ionization provokes transient
reduction of impulse resistance in grounding systems. Such a reduction may be very
pronounced for several tens of microseconds. Nevertheless, this effect can be advan-
tageous for insulation coordination with highly resistive soils, reducing the inductive
effect dependent on the frequency that increases impedance for fast-rising current
impulses. Consequently, in order to optimize the design of grounding systems, it is
necessary to predict the transient behaviour of ground electrodes, particularly rods,
subjected to impulse currents associated with lightning strokes.
The non-linear resistance of earthed electrodes associated with soil ionization has
received several denominations. A wide range of terms is in use in the literature, such
as “transient resistance”, “transient impedance”, “dynamic impedance”, “impulse
resistance”, “impulse impedance”, etc. In this paper we will preferably use “transient
resistance”, as defined in Eq. (2).
Dynamic models predicting the non-linear surge-current characteristics of con-
centrated earth electrodes have been studied for many years [1–3]. The determination
of the ionization gradient of soils is critical for the non-linear response. Several articles
[4–7] have been published concerning the analysis of soil ionization, the effect of
impulse polarity, soil conductivity and electrode geometry.
CIGRE recommends a current-dependent grounding resistance model applicable to
concentrated electrodes [8]. Many different approaches of soil ionization have been
proposed, e.g. a current-dependent resistance model based on arc equations similar to
arc interruptions in circuit breakers [9]; a simulation model based on the finite-element
method [10]; and a hybrid frequency-time domain methodology applied to coupled
systems [11]. The dynamic performance of earthed electrodes due to soil ionization and
frequency-dependent behaviour has been analysed in [12]. The transmission line theory
was applied to study the ground impedance of buried horizontal wires submitted to
lightning or switching transients [13]. A detailed analysis of the inductive-capacitive
characteristics of rod electrodes, including the analytical expression of the critical
length, was proposed in [14].
In [15] and [16] the authors presented the laboratory results using a scaled model of
a grounding rod. The measurements show a correlation between the injected energy
and minimum transient resistance of the grounding system. From the experiments, the
energy appeared to be the most significant electrical parameter conditioning transient
grounding resistance, even over the current peak value.
The ionization process in soils is more complex than in gases and the parameters
associated with the inception and progression of discharges were previously studied [4–
6]. Field measurements of grounding transient resistances are not easily obtained under
lightning strokes and the experimental study of the ionization process in soils is difficult
due to the limited accessibility to the discharge parameters. The lack of experimental
data showing the effects of impulse currents parameters (wave shape, peak value, rise
time, etc.) on the transient resistance in actual grounding systems, can be overcome by
measurements in laboratory scale systems with well controlled conditions.
A physical approach based on the analysis of experimental results obtained in the
laboratory using a cylindrical test cell with a vertical rod submitted to current impulses
with several amplitudes and front-tail times was presented in [17] for a single value of
 Non-linear Dynamic Response of a Ground Rod in Resistive Soils 1535

soil resistivity. The model computes the space charge distribution around the electrode
and is based on the energy balance between the injected energy and the energy needed
to heat the soil and develop the ionized channels. In this paper the approach was
applied to several values of soil resistivity, between 290 and 3800 Xm, in order to
generalize the model and allow a fast estimation of the reduction of transient resistance
in any buried rod.

2 Experimental Set-Up

The tests were made in a laboratory scale grounding system designed to reproduce the
current density and electric field distributions in an actual grounded rod under lightning
stroke currents. The length of the earth electrode was shorter than the critical one [14]
for all tested soils. A detailed description of the cell is available in [15] and [17].
Figure 1 is a schematic diagram of the experimental cell.
The current impulses were produced with a Marx generator of 35 kJ working as a
current source. Currents were measured in the external grounded electrode with a 1 Hz
to 20 MHz bandwidth current monitor. Voltages between rod and ground were mea-
sured with a resistive divider and the waves were recorded with a digital storage
oscilloscope of 8 bits, 1 GS/s, 100 MHz.
The scaled grounding system was designed to represent similar electrical conditions
as a rod buried in the soil. The electric field E around the rod electrode was calculated
by the well-known equation:

E ¼ J :q ð1Þ

Rod Ф r

h
Sand

PVC tank

Floor

Fig. 1. Laboratory-scaled grounding system. r = 63 cm; Ф = 0.6 cm, h = 100 cm; a = 40 cm;
b = 60 cm.
1536 R. R. Diaz et al.

2.1 Transient Resistance

Six series of tests were performed with several double exponential impulses each, in
order to measure traces of voltage and current, for different resistivity soils. With the
measured traces, the minimum resistance values and their corresponding energies were
calculated for several current waveforms. Then, the ionization current was estimated
from these values.
Transient resistance is calculated as:

RðtÞ ¼ U ðtÞ=I ðtÞ ð2Þ

Figure 2 shows the voltage, current and transient resistance traces measured for a
non-linear conduction regime.

Fig. 2. Example of transient resistance, current and voltage as a function of time.

3 Results
3.1 Measured Minimal Transient Resistance
The evaluation of the ionization current Ic, corresponding to transient resistance tran-
sition due to the absence of ionization R = R0, is shown in Fig. 3 for each soil con-
dition, and the corresponding equations are presented in the figure caption. Table 1
summarizes the values of Ic, R0 and energy. The value of R0 matches the low frequency
(50/60 Hz) resistance of the grounding system.
 Non-linear Dynamic Response of a Ground Rod in Resistive Soils 1537

Fig. 3. Rm/R0 [pu] as a function of Imax [A] for the values of soil resistivity.
289 Xm : Rm =R0 ¼ 0:20 lnðImax Þ þ 1:86
323 Xm : Rm =R0 ¼ 0:17 lnðImax Þ þ 1:71
400 Xm : Rm =R0 ¼ 0:23 lnðImax Þ þ 1:70
1530 Xm : Rm =R0 ¼ 0:23 lnðImax Þ þ 1:59
1690 Xm : Rm =R0 ¼ 0:22 lnðImax Þ þ 1:52
3790 Xm : Rm =R0 ¼ 0:27 lnðImax Þ þ 1:55

Table 1. Ionization from test and measurements

q [Xm] R0 [X] Wc [J] Ic [A] Current impulse [ls]
3787 2840 0.5 7.7 4/20
1693 1270 2.5 10.6 15/36
1533 1150 3.4 13 15/36
400 300 10 21 8/40
323 242 15.8 65 15/36
289 217 32.4 73 15/36

3.2 Injected Energy at the Resistance Minimum Value

The energy injected until the time tm of minimum resistance value is calculated by
integration of the power:

Ztm
W ðt m Þ ¼ PðtÞ dt ¼ Wm ð3Þ
0
1538 R. R. Diaz et al.

Since the reduction of transient resistance depends on soil ionization and electric
field, a relation between the current and the lower value of transient resistance should
be expected. In a previous work [16], a good correlation was obtained between the
energy W(tm) and the minimum resistance Rm measured for several current impulses.
There seemed to be a good correlation between the current crest value Imax and the
resistance Rm only for similar current waveforms.
In [17] the total injected energy W(t) in the grounding system showed that it can be
associated with two electrical phenomena: the formation of the discharge channel by
accumulation of space charges with energy Wd(t) and the heating of sand by the loss of
energy Ws(t) in the resistive soil. In that case the energy balance equation is:

W ðtÞ ¼ Wd ðtÞ þ WsðtÞ ð4Þ

The ionization phenomenon appears when the electric field around the rod reaches
the dielectric strength gradient Ec [5] corresponding to the critical current Ic starting soil
ionization.
The tm value depends mainly on the Imax crest value, on the front and tail times of
the current impulses. Thus, tm increases when they do. In the present tests, tm measured
values were between 7 and 40 ls.
Figure 4 shows the Rm/R0 ratio calculated as a function of Wm/Wc, where Wm
represents the injected energy that produces the minimum of resistance Rm and Wc is
the lowest value of energy needed to ionize the soil.
Although the value correlation of resistance and energy is satisfactory, from the
point of view of engineering, the calculation of Wm and Wc is not easy. Assuming loss
of accuracy and physical background, a simple procedure is here proposed to estimate
the reduction of resistance from technically available parameters such as peak current,
soil resistivity and ground electrode geometry.

Fig. 4. Rm/R0 as a function of Wm/Wc.

 Non-linear Dynamic Response of a Ground Rod in Resistive Soils 1539

4 Discussion

In order to calculate Ic independently of the present measurements, another way is

proposed by applying Eqs. (1) and (7) and considering the Ec values obtained from
Fig. 5. For resistivity lower than 700 Xm, Ec may be considered constant and equal to
8.4 kV/cm; for resistivity greater than 30 kXm, Ec is about 18 kV/cm, while for
resistivity within the range of 700 Xm and 30 kXm, Ec may be estimated with Eq. (5):

Ec ¼ 2:5 ln q  8:5 ½Xm; kV=cm ð5Þ

Considering the buried surface of the rod:

S ¼ p:U:h ð6Þ

In the particular case of the laboratory cell the surface of the rod is S = 188.5 cm2.
The values of the critical density of current are calculated as:

Jc ¼ Ic =S ð7Þ

Fig. 5. Dielectric strength as a function of soil resistivity [5].

Table 2 shows the Ic values calculated by both methods. The agreement is satis-
factory, especially for high resistivity, and it can justify the reduction of the current
values to ionization conditions, as shown in Fig. 6.
1540 R. R. Diaz et al.

Table 2. Critical current Ic obtained from the cell measurements and the dielectric strength
gradient.
q [Xm] 3787 1693 1533 400 323 289
Ic [A] from Table 1 7.7 10.6 13 21 65 73
Ic [A] from Fig. 5 6 11 12 39 49 55

Fig. 6. Rm/R0 as a function of Imax/Ic.

By fitting the values of Fig. 6, a simple logarithmic formula can be obtained to

calculate the reduction of transient resistance depending on the current ratio:

Rm =R0 ¼ 0:23 lnðImax =Ic Þ þ 1 ð8Þ

Equation (8) shows that, with an approximation better than ±15%, it is possible to
estimate the reduction of transient resistances for different buried rods whose geometric
dimensions are known. In principle, the found relationship could be extended to soils of
greater resistivity, based on the experimental values of the dielectric strength of Fig. 5.
This simple method to calculate the minimum value for any grounding rod can be
used for impulse currents with front times between 4 and 15 ls, and tail times between
20 and 40 ls; as indicated in the flow chart of Fig. 7. R0 could be measured at low ac
frequency or calculated from the rod size and soil resistivity [19].
 Non-linear Dynamic Response of a Ground Rod in Resistive Soils 1541

Fig. 7. Procedure for calculating Rm

The radial electric field distribution around the inner electrode is comparable to the
radial field distribution of a 24 mm diameter long vertical ground rod with zero
potential 30 m away. From this comparison an electric field equivalence can be
assumed between the scaled grounding system and an actual long rod. Figure 8
illustrates the radial electric field distribution around the rods, calculated assuming
uniform soil resistivity and a current flow analysis [18]. A negligible difference can be
observed after 20 mm, showing that once the ionization started, the discharge develops
in a similar electric field, which should result in a similar non-linear resistance. The
field values on the surface of the rods are not the same because the rod diameters are
different, and cause changes in the ionization current.
It must be taken into account that our laboratory rod does not have a tip in the
ground, while the actual rods have a buried end that increases the electric field at the
rod tip, which initiates ionization with a lower current value. This effect could partially
compensate the difference of electric fields on the surface of the rods.
1542 R. R. Diaz et al.

However, with the aforementioned caution, when calculating Ic from q, S and Ec,
the last one obtained from Fig. 5, the value of the ionization current becomes inde-
pendent of the grounding laboratory cell, and Eq. (7) could be generalised to any rod.

Fig. 8. Radial electric field around the cell rod (radius Ф/2 = 3 mm, blue) and around a typical
long rod (Ф/2 = 6 mm, red) buried in uniform soil with a null potential at 30 m.

5 Conclusions

This work has achieved an engineering method to estimate the reduction of the tran-
sient resistance by soil ionization in grounding rods subjected to lightning surge cur-
rents. The study was carried out with impulse waveforms not very different from the
standard impulse 8/20 ls, with uniform soils and resistivity comprised between 290
and 3800 Xm.
The main finding is an equation to evaluate transient resistance reduction. The input
data are simple technical specifications, such as soil resistivity, expected peak of the
current impulse and rod size.
Further work is needed to confront the Rm values obtained in the laboratory with
those of the actual ground rods buried into the soil and, in addition, to extend the results
to soils with resistivity values outside the studied range.

Acknowledgments. The authors acknowledge the financial support of CONICET (the National
Scientific and Technical Research Council of Argentina).
 Non-linear Dynamic Response of a Ground Rod in Resistive Soils 1543

References
1. Bellaschi, P.L., Armington, R.E., Snowden, A.E.: Impulse and 60-cycle characteristics of
driven grounds - part 11. AIEE Trans. 61, 349–363 (1942)
2. Liew, A.C., Darveniza, M.: Dynamic model of impulse characteristics of concentrated
earths. Proc. IEE 121, 123–135 (1974)
3. Oettle, E.E.: A new general estimation curve for predicting the impulse impedance of
concentrated earth electrodes. IEEE Trans. Power Deliv. 3, 2020–2029 (1988)
4. Mousa, A.M.: The soil ionization gradient associated with discharge of high currents into
concentrated electrodes. IEEE Trans. Power Deliv. 9, 1669–1677 (1994)
5. Espel, P., Diaz, R., Bonamy, A., Silva, J.: Electrical parameters associated with discharges in
resistive soils. IEEE Trans. Power Deliv. 19, 1174–1182 (2004)
6. Nor, N., Haddad, A., Griffiths, H.: Determination of threshold electric field Ec of soil under
high impulse currents. IEEE Trans. Power Deliv. 20, 2108–2113 (2005)
7. Nor, N., Ramli, A.: Soil characteristics of wet sand under different impulse polarity and earth
electrode’s dimensions. IEEE Trans. Dielectr. Electr. Insul. 15, 910–914 (2008)
8. Cigre WG 33.01: Guide to procedures for estimating the lightning performance of
transmission lines. Tech. Broch. Nr. 63, Paris (1991)
9. Coelho, V., et al.: Soil resistivity behaviour under high current density values using the
energy balance method. In: Proceedings of the International Symposium on Lightning
Protection, XIII SIPDA, pp. 129–136, Brazil (2015)
10. Habjanic, A., Trlep, M.: The simulation of the soil ionization phenomenon around the
grounding system by the finite element method. IEEE Trans. Magn. 42, 867–970 (2006)
11. Salari, J.C., Portela, C.: Grounding systems modeling including soil ionization. IEEE Trans.
Power Deliv. 23, 1939–1945 (2008)
12. Sekioka, S.: Frequency and time-dependent grounding resistance model for lightning surge
analysis. IEEE Trans. Electromagn. Compat. 61(2), 419–425 (2019)
13. Theethayi, N., Thottappillil, R., Paolone, M., Nucci, C., Rachidi, F.: External impedance and
admittance of buried horizontal wires for transient studies. IEEE Trans. Dielectr. Electr.
Insul. 14, 751–761 (2007)
14. Bourg, S., Sacepe, B., Debu, T.: Deep earth electrodes in highly resistive ground. In: IEEE
International Symposium on Electromagnetic Compatibility, pp. 584–589, USA (1995)
15. Diaz, R., Silva, J.: Experimental results on non-linear behaviour of grounding systems
submitted to impulse currents. In: Proceedings of the International Symposium on Lightning
Protection, VIII SIPDA, pp. 551–555, Brazil (2005)
16. Diaz, R., Silva, J.: Energy injection effect on transient grounding resistance. In: Proceedings
of the International Symposium on High-Voltage Engineering, pp. 1763–1768, Johannes-
burg (2009)
17. Diaz, R., Silva, J.: Space charge and soil ionization: an electro-kinetic approach. IEEE Trans.
Dielectr. Electr. Insul. 18(6), 2032–2039 (2011)
18. DC Current Flow Package. https://www.quickfield.com/cflow.htm. Downloaded 11 Dec
2018
19. IEEE Std. 80: Guide for Safety in AC Substation Grounding, USA (2015)
 Lightning Performance of Medium Voltage
Distribution Networks

V. Vita, C. A. Christodoulou, and L. Ekonomou(&)

Department of Electrical and Electronic Engineering Educators,

A.S.PE.T.E. - School of Pedagogical and Technological Education,
N. Heraklion, 14121 Athens, Greece
leekonomou@aspete.gr

Abstract. The need for uninterruptable electrical power supply of good quality
emerges lightning protection of distribution networks an issue of high impor-
tance for power utilities. The adequate protection of distribution substations and
lines ensures the reliable operation of the network and prevents damages of the
equipment. The current work addresses the lightning performance of complex
medium voltage networks, considering the impact of the installation position of
the protective devices on the lightning failure rate. In details, the mean time
between failures for each transformer of the network under study is estimated for
both direct and indirect lightning overvoltages, applying an appropriate
methodology. The simulation results reveal that the appropriate placement of the
protective devices (surge arresters or/and arcing horns) at critical positions of the
network improves its lightning performance. The outcomes of the current work
are useful for power utilities to design in a more efficient way the lightning
protection configuration of their networks.

Keywords: Distribution networks
Lightning performance
Surge arresters

1 Introduction

The need for reduction of the momentary or permanent interruptions of the power
supply emerges the efficient lightning protection of distribution networks an issue of
priority for the power utilities. Added to these, the proliferation of sensitive loads, the
integration of renewable energy sources, the development of smart grids, the demand
by customers for power supply of good quality, the ever-increasing complexity of the
network are factors that necessitate the protection of the distribution network against
lightning phenomena.
To this context, the efficient protection of the distribution lines and substations is
requirement in order to ensure the normal operation of the network and avoid mal-
functions and interruption of the power supply. Overhead medium voltage lines and
medium voltage/low voltage substations are susceptible to lightning flashes, due to
their installation in wide-open areas. A lightning stroke on line or a transformer stresses
the insulation of the equipment and results in severe damages of the various devices
that can lead to momentary or permanent outages. The replacement or/and repair cost
of the damaged equipment, the cost of the non-distributed energy, the downgrade of the

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1544–1554, 2020.
https://doi.org/10.1007/978-3-030-31676-1_144
 Lightning Performance of Medium Voltage Distribution Networks 1545

power quality indices, the reputation of the power utility are parameters that render
urgent the adequate protection of the distribution network against internal and external
overvoltages [1–3].
The installation of overhead ground wires and the placement of surge arresters are
the basic protective means against lightning surges. The implementation of ground
wires is not recommended for distribution lines, considering the low insulation level
and the grounding resistance values. So, the installation of surge arresters at critical
positions of the network is the main solution for the enhancement of the lightning
performance of a distribution system. The installation of surge arresters at every pole or
at every transformer definitely improves the lightning performance of the line, elimi-
nating the expected lightning failure rate [4, 5]. Nevertheless, the cost of this invest-
ment would be not affordable. For this reason in the current work surge arresters are
installed at the medium voltage side of the distribution transformers, examining the
impact of various factors on the lightning performance of a typical medium voltage
network, in an effort to determine effective protective schemes that reduce the annual
lightning failure rate and ensure the uninterruptable operation of the network.

2 Lightning Performance of Distribution Networks

Both overhead and underground power lines of a distribution network are susceptible
to lightning events that may result in interruptions of the power supply and permanent
damages of the equipment. Lightning protection for the transmission lines and distri-
bution lines share the same principles. The inherently high insulation levels for
transmission lines render them less vulnerable to lightning damages compared to
distribution lines, and hence they suffer fewer outages than the less-well insulated
distribution lines, although the repercussions of a transmission line failure can be much
more costly [6].
The number of strikes to an overhead power line per kilometer of line per year, Nkm,
can be approximately estimated by using the concept of an “equivalent collective area”
on the ground in which the lightning would strike if the line were not there. Assuming
that all flashes that would hit the ground within a horizontal distance equal to the
double of the line height on each side of the center of the line will strike the line, the
equivalent collective area (equal to 4
h
l) is designed in Fig. 1 [6, 7]. For a length of
line l, that area is 4
h
l, where h is the line height. The number of strikes to an
overhead power line per kilometer of line per year, Nkm, is:

Nkm ¼ 4
h
Ng km1
years1 ð1Þ

where Ng is the ground flash density in ground flashes per square kilometer per year
(km−2 years−1).
1546 V. Vita et al.

downward
downward outside shaded
leader above area
shaded area
upward leader
upward leader from ground
from phase
conductor

h
4h l

Fig. 1. Lightning flashes striking an overhead power line (the equivalent collective area) [6].

The protection of overhead medium voltage lines against atmospheric overvoltages

can may be accomplished by the following methods:
• adoption of higher insulation levels
• installation of overhead ground wires
• installation of surge arresters at critical positions of the network.
The increase of the insulation level of the equipment is an expensive and techni-
cally difficult practice. On the other hand, the installation of ground wires is efficient for
transmission systems, but it is not efficient in the same way for distribution networks.
Indeed, an overhead ground wire could reduce the number of flashovers due to nearby
lightning flashes, but to be efficient against direct strikes it should be grounded at every
pole with very low ground resistance values to avoid back flashover phenomena; these
requirements consist major constraints to the effectiveness of ground wires against
direct strokes. The installation of ground shield wires may provide effective protection
only if sufficient withstand voltage between local ground and the phase conductors is
provided or low ground resistance values of the poles are achieved [7].
Considering the above, the installation of surge arresters at critical positions of the
network is the only practical method that can definitely improve the lightning per-
formance of a distribution system. The modern metal oxide (ZnO) gapless surge
arresters are installed between phase conductor and earth and clamp the developed
external and internal overvoltages to values below the insulation level of the equip-
ment, preventing flashovers and consequent insulation breakdown. Surge arresters are
usually installed at the primary of the distribution substations to limit the incoming
overvoltages below the Basic Insulation Level of the system [6, 7].
The appropriate placement of the arresters plays a key role on the efficient pro-
tection that they provide; the lightning surges behave as travelling waves, reflecting or
 Lightning Performance of Medium Voltage Distribution Networks 1547

refracting where the surge impedance changes. Considering a distribution transformer,

due to its behaviour as an unterminated end and the consequent reflections, a double
voltage magnitude will arise. To mitigate the reduplication of the voltage surge, surge
arresters must be installed near the equipment to be protected. Furthermore, the length
of the connection conductors and the grounding resistance are factors to be taken into
account. Figure 2 presents a simple configuration of a line protected by a gapless surge
arrester, where the developed travelling waves are shown.

medium voltage line

u
u’
arrester

transformer
Fig. 2. Distribution transformer protected against external overvoltages by surge arresters.

It is worth mentioning that surge arresters suffer due to the stress of the discharge
current through them. Indeed, if the absorbed by the arresters energy exceeds their
nominal energy absorption capability (see energy class of the arresters), then they
cannot cool back-down to their normal operating temperature and, consequently, they
fail and they cannot provide efficient protection to the substations [8, 9]. For this
reason, the selection of the electrical characteristics of the arresters is significant to
avoid the above repercussions.

3 Calculation Methodology

Atmospheric overvoltages are classified into overvoltages due to direct lightning and
induced (indirect) overvoltages. Taking into account the low insulation level of a
distribution line, a direct lightning stroke to a phase conductor may cause insulation
flashover, resulting in the development of two equal voltage surges that travel to both
directions. Note that more than 90% of the lightning currents have a magnitude that
exceeds 10 kA, so the arising voltage surges are expected to be greater than 2000 kV
(much higher than the impulse withstand voltage of the lines). On the other hand, a
lightning hit on the structures close to a distribution line or on the ground may induce
on the conductors of the line overvoltages greater than the insulation level [7].
A related point to consider is that the overhead ground wires have practically no
impact on the lightning performance of a medium voltage line in the case of a direct
1548 V. Vita et al.

lightning hit, since the rise of the ground potential due to the lightning current flow
through the pole ground impedance causes in general voltage differences between the
ground lead and the phase conductors greater than the critical flashover voltage (CFO)
of the line. Note that the CFO is the crest value of the standard lightning impulse wave
that results in flashover through the surrounding medium on 50% of the applications.
Assuming a Gaussian distribution of flashover data, any specific probability of with-
stand can be estimated from the CFO value and the standard deviation [7]. Grounding
at every pole, the accomplishment of low grounding resistances and an appropriate
CFO are prerequisites for the efficient protection that the ground wires provide [1–3].
In contrast, a ground wire can contribute to the reduction of the arising induced
overvoltages due to nearby lightning hits, considering its coupling with the phase
conductors. The applied electrogeometrical model for the installation position of the
overhead ground wires, the grounding resistance, the soil resistivity, the magnitude, the
rise time and the time to half-width of the lightning current are crucial factors for the
efficiency of the protection measures [10, 11]. The appropriate placement of surge
arresters is the most effective measure to protect critical equipment that is permanently
damaged in the case of insulation breakdown.
In the case of a single straight line, the lightning performance is generally expressed
in terms of number of events/year/km; nevertheless, in the case of a complex topology
of the network the mean time between failures (MTBF) is preferred, given by the
following equation [10, 12]:
ntotal
MTBF ¼ ð2Þ
n
A
Ng

where n is the number of lightning events that result in flashover, ntotal is the total
number of simulated lightning events, A is the striking area under study in km2 and Ng
is the ground flash density in flashes/km2/year.
Considering the Eq. (2), a Monte Carlo procedure can be applied for the random
generation of the parameters of each lightning event (the peak current, the rise time and
the location). As far as the lightning current is concerned, it follows the Cigre log-
normal probability distribution for negative first strokes, ignoring subsequent negative
flashes and positive strokes [1, 10]. A direct lightning hit is modelled as a current
source connected to the pole nearest the randomly-generated stroke location coordi-
nates [10, 12]. Note that direct lightning strikes are almost always expected to result in
insulation flashovers, even if overhead ground wires are installed. The implementation
of surge arresters at critical positions can limit the expected failure rate. As far con-
cerning induced lightning surges, the coupling model between the conductors of the
overhead line and the lightning electromagnetic pulse takes into consideration the
lighting current characteristics, the return stroke model and the lightning electromag-
netic pulse propagation above a lossy ground [10, 12]. The induced lightning events are
computed according to the electrogeometrical model presented in [1, 13].
 Lightning Performance of Medium Voltage Distribution Networks 1549

4 System Under Examination

Figure 3 depicts the configuration of the network under study, consisting of 22

20/0.4 kV distribution transformers. Table 1 presents the nominal apparent power and
the grounding resistance of the distribution transformers. Figure 4 depicts the geometry
of a typical overhead medium voltage line, consisting of three ACSR conductors
horizontally placed. The rated voltage of the lines is 20 kV (phase-to-phase) and the
withstand voltage of the distribution transformers in 125 kV. Tables 2 and 3 contain
the electrical characteristics of the lines and the arresters, respectively. It is supposed
that surge arresters are connected directly to the tank of the transformer with short
length conductors in order to ensure a more efficient protection against the incoming
surges, compared to the case that the arresters and the transformer are separately
grounded and long length connection conductors are used. The shielding factors Sf is
regarded equal to 0.91 that corresponds to the case of shielding provided by objects
having the same height [12]. The ground conductivity is 0.01 ℧/m.

0.7 km
1
2.3 km
2
6
0.4 km
7
0.4 km

5.9 km

8.1 km

1.7 km
3.2 km

8
0.6 km
3 9

4 5 10
13
0.8 km

0.3 km
2 km

14
11 3.5 km
15
6.1 km
18
12
fuse cutout (type A)
fuse cutout (type B) 16 17
20/0.4kV transformer 5.9 km
22
circuit breaker
1.2 km
0.5 km

disconnect switch
recloser
20 21
19
Fig. 3. Topology of the network under examination.
1550 V. Vita et al.

phase 1.5m 1.15m

insulator
wooden
wooden crossarm
pole 10m

Fig. 4. Unshielded wooden pole of a typical 20 kV distribution line.

Table 1. Characteristics of the substations.

Substation kVA Grounding resistance Substation kVA Grounding resistance
1 630 0.9 X 12 630 1.2 X
2 160 1.2 X 13 250 2.9 X
3 160 1.4 X 14 250 1.4 X
4 160 1.8 X 15 160 1.1 X
5 100 1.7 X 16 160 1.8 X
6 100 2.5 X 17 100 2.5 X
7 100 2.8 X 18 100 2.9 X
8 160 3.1 X 19 250 2.3 X
9 100 2.4 X 20 160 1.2 X
10 100 2.7 X 21 100 1.8 X
11 250 1.8 X 22 100 1.2 X

Table 2. Electrical characteristics of the distribution medium voltage overhead lines.

Cross-section R (X/km) X (X/km) Ro (X/km) Xo (X/km) Imax (A)
3  16 mm2 1.268 0.422 1.416 1.620 136
3  35 mm2 0.576 0.397 0.724 1.595 224
3  95 mm2 0.215 0.334 0363 1.556 448
 Lightning Performance of Medium Voltage Distribution Networks 1551

Table 3. Electrical characteristics of the surge arresters.

Rated voltage (Ur) 24 kV
Continuous operating voltage (Uc) 19.3 kV
Nominal discharge current 10 kA (peak), 8/20 ls
Residual voltage 8/20 ls 5 kA 61 kV
10 kA 67 kV
Energy capability Line discharge 2
class
Energy 3.7 kJ/kV (Ur)
withstand

5 Results and Discussion

The current work examines the lightning performance of the network configuration of
Fig. 3, regarding the following cases:
(1) no surge arresters are installed
(2) installation of surge arresters at all the distribution transformers
(3) installation of surge arresters at the distribution transformers with grounding
resistance greater than 2 X.
For each examined case the MTBF is estimated, according to the procedures
described in Sect. 3. Figure 5 depicts the obtained outputs for each substation for case
1, 2 and 3, respectively. The installation of surge arresters at every substation (case 2)
enhances the lightning performance of the network under study; indeed, the MTBF is

Fig. 5. MTBF of the substations.

1552 V. Vita et al.

greater compared with the case that no surge arresters are placed. Similar results are
also extracted for case 3, indicating that the strategical placement of the protective
devices can provide effective protection in a moderate cost.
The calculation of the MTBF is repeated, considering that the first pole of each
distribution substation is equipped with arcing horns (Fig. 6). The MTBF is recalcu-
lated for cases 1–3 and the obtained results are presented in Fig. 7.

medium voltage line

arcing horn

12.5 cm arrester

transformer
Fig. 6. Protection of a distribution transformer by surge arresters and arcing horns.

Fig. 7. MTBF of the substations in the case that the first pole of each substation is equipped
with arcing horns.
 Lightning Performance of Medium Voltage Distribution Networks 1553

The installation of arcing horns is technically practical and financially affordable

solution to upgrade the lightning performance of the network. Indeed, the implemen-
tation of arcing horns at critical positions is common practice for power utilities to
reduce the annual failure rate due to lightning phenomena. Nevertheless, main draw-
back of arcing horns is that in the case of a flashover an earth fault is produced,
resulting in a circuit outage until the fault is cleared by circuit breaker operation.

6 Conclusions

This paper conducts a lightning performance analysis for a distribution network,

consisting of overhead medium voltage lines and outdoor medium voltage/low voltage
transformers, considering both direct and induced overvoltages. Various cases for the
installation position of surge arresters and horning arcs are examined and the MTBF is
evaluated for each case. The calculated results reveal that the appropriate placement of
the arresters at critical positions reduces the annual failure rate due to lightning surges.
The outcomes of the current work are useful for power utilities to design in a more
efficient way the configuration of the lightning protection of their networks. Future
work includes the examination of the role of the ground conductivity, the environ-
mental shielding factor Sf as well as the need for installation of surge arresters at the
medium voltage lines.

Acknowledgment. The authors acknowledge financial support for the dissemination of this
work from the Special Account for Research of ASPETE through the funding program
“Strengthening ASPETE’s research”.

References
1. IEEE guide for improving the lightning performance of electric power overhead distribution
lines. In: IEEE Std. 1410 (2010)
2. Piantini, A.: Lightning protection of overhead power distribution lines. In: Proceedings
International Conference on Lightning Protection (29th ICLP), pp. 1–29, Uppsala, Sweden
(2008)
3. Piantini, A.: Analysis of the effectiveness of shield wires in mitigating lightning-induced
voltages on power distribution lines. Electr. Power Syst. Res. 159, 9–16 (2018)
4. Christodoulou, C.A., Vita, V., Ekonomou, L.: Studies for the more effective protection of
MV/LV substations against lightning overvoltages. Int. J. Circuits Electron. 2, 11–15 (2017)
5. Christodoulou, C.A., Vita, V., Maris, T.I.: Lightning protection of distribution substations by
using metal oxide gapless surge arresters connected in parallel. Int. J. Power Energy Res. 1
(1), 1–7 (2017)
6. Uman, M.: The Art and Science of Lightning Protection. Cambridge University Press,
Cambridge (2008)
7. Cooray, V.: Lightning Protection. IET Power and Energy Series, vol. 58. Institution of
Engineering and Technology, Stevenage (2010)
8. IEEE Std. C62.22–1991: IEEE guide for the application of metal-oxide surge arresters for
alternating-current systems
1554 V. Vita et al.

9. IEEE Std. 1410™-2004 (Revision of IEEE Std 1410–1997) 1410TM IEEE guide for
improving the lightning performance of electric power overhead distribution lines
10. Borghetti, A., Napolitano, F., Nucci, C.A., Tossani, F.: Response of distribution networks to
direct and indirect lightning: influence of surge arresters location, flashover occurrence and
environmental shielding. Electr. Power Syst. Res. 153, 73–81 (2017)
11. Vita, V., Maris, T.I.: Sensitivity analyses of parameters that affect the lightning performance
of distribution networks with distributed generation. J. Multidiscip. Eng. Sci. Stud. 2(8),
774–781 (2016)
12. Napolitano, F., Tossani, F., Borghetti, A., Nucci, C.A., Martinez, M.L.B., Lopes, G.P., Dos
Santos, G.J.G., Fagundes, D.R.: Lightning performance of a real distribution network with
focus on transformer protection. Electr. Power Syst. Res. 139, 60–67 (2016)
13. Bullich-Massague, E., Sumper, A., Villafafila-Robles, R., Rull-Duran, J.: Optimization of
surge arresters location in overhead distribution networks. IEEE Trans. Power Deliv. 30(2),
674–683 (2015)
Author Index

A Boonsaner, Nutthaphan, 534

Abrahão, Fernando Teixeira, 243 Boonseng, Chongrag, 534
Adyatma S, Wisnu, 1423 Boonseng, Rapeepornpat, 534
Afia, Ramy S. A., 253, 261 Bousiou, Evanthia I., 1413
Aghmashehr, Reza, 1276, 1285 Briosso, E., 521
Agnel, S., 1108 Brocke, Ralph, 96
Ágústsson, H., 458 Brown, M., 331
Akbari, A., 838
Al-Abadi, Ali, 15 C
Alfonso, A., 157 Cai, Wei, 714
Alves, Matheus R., 1502 Cevallos, Benjamín, 85, 1367
An, ZhiGuo, 437 ChaeYoon, Bae, 292
Andrade, Arthur Francisco, 1491, 1502 Chen, George, 1132
Apiratikul, Promsak, 534 Chen, Li, 300
Ardila, Jorge, 85, 1367 Cheng, Bozhi, 982
Ardila-Rey, Jorge, 899 Cheng, Chuanhui, 1055
Arentsen, M. T., 3 Chladek, Svatoslav, 549
Ariannik, M., 838 Cho, Hiroaki, 489
Asif, Mansoor, 27 Choi, Won, 582
Augeneder, N., 738 Christodoulou, C. A., 1471, 1544
Avaria, Gonzalo, 85, 1367 Chu, Jinwei, 1244
Azirani, M. Akbari, 838 Cimino, A., 825
Clark, David, 308
B Claudi, A., 1437
Backhaus, Karsten, 560, 1121 Costa, Edson G., 1491, 1502
Bak, C. L., 3 Crossley, Peter, 982
Balzer, Constantin, 1458 Cselkó, Richárd, 1003, 1015
Barbieri, L., 375, 429 Csépes, Gusztáv, 1003
Barnett, S. B., 331 Cui, Haozhe, 1044
Barreto, Alexandre, 243 Cui, Huize, 1077
Belaya, Olga, 131, 138 Cui, Yangzhen, 993
Beltle, M., 219
Beyer, M., 55 D
Biedenbach, G., 917 da Silva, F. F., 3
Blaszczyk, A., 1155 Datsios, Z. G., 1328, 1482

© Springer Nature Switzerland AG 2020

B. Németh (Ed.): ISH 2019, LNEE 598, pp. 1555–1560, 2020.
https://doi.org/10.1007/978-3-030-31676-1
1556 Author Index

De Gersem, Herbert, 118 Guo, Ze, 300

de Souza Silva, Frederico Tassi, 353 Gutman, I., 458
Deng, Chun, 714
Diaz, R. R., 1533 H
Diharja, Kurniawan Danu, 342 Haller, Rainer, 549
Dominice, Alexandre, 243 Hällström, J., 1316
Duvvury, V. S. B. Chaitanya, 724 Hamada, Toshiyuki, 1218
Hämel, Kai, 777, 788, 926
E Hanaoka, R., 208
Eda, Osamu, 643 Hannig, Martin, 96
Ekonomou, L., 1471, 1544 Hao, Chen, 572
Elg, A.-P., 1316 Hao, Miao, 1132
Eui-hwan, Jung, 48 Hasegawa, Masayuki, 908
Hayashida, Chikara, 472
F He, Yunwen, 752
Faradzhev, Valentin, 138 Heidler, Fridolin H., 1169
Fieldsend-Roxborough, Andrew, 969, 982 Higashiyama, Masakazu, 653
Finis, G., 1437 Hikita, Masayuki, 635, 643, 653, 662, 673
Florkowski, Marek, 1254 Hilbert, M., 1265
Fofana, I., 615 Hinrichsen, Volker, 96, 118, 1458
Fritsche, Ronny, 1121 Hinterholzer, R., 738
Frohne, C., 799 Hochbrückner, Benedikt, 560
Fujii, Masayuki, 1218 Hohloch, J., 219
Fujii, Yuuki, 489 Homeier, K., 958
Fujimura, Tadahiro, 643, 653, 662 Hornak, Jaroslav, 549
Fujita, Y., 208 Horst, J., 825
Furgał, Jakub, 1254 Hou, Xinbin, 1067
Furumoto, Takumi, 481 Huang, Yuwei, 1340
Huecker, Thomas, 273, 1035
G Hwang, Jae-Sang, 27
Gabler, Tobias, 1121 Hwangbo, Seung, 1143
Gamil, Ahmed, 15 HyunWoo, Joo, 292
Gan, Lin, 300, 850, 1513
Ganesh, Aravind, 594 I
Gao, Shun’an, 714 Iida, Kazuo, 653
Gao, Yanfeng, 714 Imani, M., 958
Garcia, Ricardo Wesley Salles, 353 Ioannidis, A. I., 1295
Ge, Siqiao, 1446 Ishida, Kazuhito, 635
Gentsch, D., 1265 Ishikura, Norio, 1218
Girodet, A., 1108
Gjonaj, Erion, 118 J
Göcsei, Gábor, 138, 145, 1025 Jarman, Paul, 969, 982
Gorgan, B., 521 Jaya, T., 615
Gräf, Thomas, 384, 417 Jebamony, David, 887
Granados, C. A., 1235 Jenau, F., 825, 1097
Großmann, Steffen, 1121 Ji, Shengchang, 625, 949
Grund, R., 219 Ji, Shuntao, 969
Gu, Chunhui, 850 Jiang, Xiongwei, 1192, 1225
Gu, Le, 300 Jiang, Xiuchen, 449
Guo, Anxiang, 1132 Jing, Yu, 1077
Guo, Chenjun, 862 Joerres, S., 1437
Author Index 1557

Joh, Gye-Hyun, 582 Li, Yuwei, 108

JongUng, Choi, 292 Lian, Ze, 878
Jung, Chae-Kyun, 27 Liao, Ruijin, 282
Jung, HyunCheol, 1143 Lim, Jae-Seop, 582
Jung-nyun, Kim, 48 Lin, Yuandi, 282
Jurišić, Bruno, 67, 75 Lira, George R. S., 1491, 1502
Liu, Chenshuo, 949
K Liu, Han, 437
Kakimoto, Yasutomo, 635, 643 Liu, Hongbo, 282
Kaneko, Hiroshi, 643 Liu, Jian, 1132
Karakostas, V. T., 1482 Liu, Jiangyu, 949
Karanikiotis, N., 1295 Liu, Junxiang, 300, 1513
Karmokar, T., 1401 Liu, Lishuai, 862
Kartheek, R., 615 Liu, Q., 871
Kelemen, Franjo, 75 Liu, S., 871
Kim, HyungGyu, 1143 Liu, Weidong, 1446
Kim, Jin-Hyuk, 1347 Liu, Yonglai, 1244
Kim, Kyo-Ho, 1347 Liu, YuQuan, 437
Kim, Min-Ju, 27 Lohmeyer, Holger, 697, 705
Kinkeldey, Tobias, 777, 788, 926, 958 Lorenzen, S., 3
Kinnares, Vijit, 534 Lorimer, T., 331
Kobayashi, Masato, 673 Lovrenčič, A., 362
Koltunowicz, W., 521 Lovrenčić, V., 362
Konishi, Yoshinori, 908 Lu, Guanghao, 1077
Kono, A., 208 Lu, Y. X., 684
Konshina, Tatyana, 138 Lu, Yi, 714
Kortenbrede, J., 1097 Luejai, Waraporn, 171, 197
Koutras, K. N., 316 Luo, LinHuan, 437
Koyama, M., 208
Kozako, Masahiro, 635, 643, 653, 662, 673 M
Kristianto, 342 Ma, G., 606
Kübler, Inna, 937 Maeda, Teruhiko, 653
Küchler, Andreas, 560 Maenosono, Daichi, 481
Kühn, B., 1265 Malgesini, R., 375, 429
Kularbphettong, Kunyanuth, 534 Manu Haddad, A., 308
Kumpalavalee, Suphon, 184 Mao, Xiaozhou, 969
Kuniewski, Maciej, 1254 Marmillo, J., 917
Kupka, Lukas, 549 Masuda, Seiya, 635, 653, 662
Kurrat, M., 55, 1265, 1306 Matharage, S. Y., 871
Kwon, Ik-Soo, 27 Mauseth, F., 1155
Mavidou, S. P., 1482
L Mbolo Noah, P., 1108
Lee, Bang-Wook, 27 Mc Innes, H., 458
Lee, Ho-young, 27 Meena, K. P., 498, 510
Lee, Jeon-Seon, 582 Mei, Hongwei, 862
Lei, Su, 572 Meisner, J., 1306, 1316, 1356
Lennerhag, O., 1401 Mermigkas, Athanasios C., 308
Li, Cong, 449, 606 Meyer, H. K., 1155
Li, Dongfan, 1077 Mi, Rui, 1087
Li, H. J., 684 Miao, Yulong, 396, 406
Li, Jianying, 878, 1340 Michishita, Koji, 1377
Li, Shengtao, 108, 752, 878, 1044, 1067, 1077, Mikropoulos, Pantelis N., 1295, 1328, 1413,
1087 1482
Li, Yi, 606, 1523 Min, Daomin, 108, 1044, 1067, 1087
1558 Author Index

Miyazaki, Satoru, 472 Philipp, D., 958

Mizutani, Yoshinobu, 472 Pihera, Josef, 549
Mo, Wenxiong, 300, 850, 1513 Podporkin, G. V., 1180
Mo, WenXiong, 437 Pramanik, Saurav, 594, 724
Möckel, D., 55 Preusel, Jörg, 777, 788, 926
Mohan Rao, U., 615 Pu, Lu, 1132
Monteiro, Pedro Henrique F., 1491 Puspitasari, Nuraida, 635, 643
Moriguchi, Soichi, 908 Pyrgioti, E. C., 316
Mu, Zhou, 1192
Munoz, Osvaldo, 899
Q
Münster, Tobias, 777, 788, 926, 958
Qi, Bo, 993
Mustafa, Ehtasham, 253, 261
Qiao, Shengya, 300, 1513
Qin, W., 606
N
Qin, Yu, 850
Nakahata, Masaaki, 662
Nakamae, Tetsuo, 653
Nakamoto, Kenta, 1218 R
Nakamura, Yusuke, 489, 653 Rácz, Levente, 145, 1025
Nakano, Michihiko, 481 Rahimpour, Ebrahim, 1276, 1285
Nakano, Yusuke, 635, 673 Rajendran, Arunjothi, 498, 510
Nanno, Ikuo, 1218 Ramesh, Manjunath, 887
Naxakis, I. A., 316 Rao, Yi, 850
Nemeth, Balint, 138 Rashtchi, Vahid, 1276, 1285
Nóbrega, André Igor, 1502 Ren, Fuqiang, 625
Norouzi, Y., 799 Ren, Lulu, 282
Notingher, P., 1108 Rivera, C. A., 1235
Rojas, H. E., 1235
O Román, F. J., 157, 1235
Okabe, Shigemitsu, 662 Rothfeld, J., 799
Oke, Shinichiro, 1218 Rubtsova, Nina, 131
Orellana, Luis, 85, 1367 Rühmann, N., 1306
Ortiz, G., 1108 Ruppert, Maren Greta, 118
Otsuka, Masanori, 643
Öttl, F., 738 S
Safarudin, Iwan, 342
P Samimi, M. H., 811
Pająk, Piotr, 1254 Sang-taek, Park, 48
Pan, Liang, 949 Santamaria, F., 157
Parellada, A., 1533 Saribut, Surapol, 197
Passon, S., 1306, 1316, 1356 Sato, Gaku, 908
Passow, D., 219 Sato, Hidefumi, 481, 635
Pastromas, S. A., 316 Schatzl, Franz, 15
Paul, Christian, 1169 Schierding, C., 55
Pavez, Cristian, 85, 1367 Schilling, F., 1306
Peng, Daixiao, 1244 Schlüterbusch, T. C., 1356
Peng, Qingjun, 1225 Schreiter, Sebastian, 697, 705
Pepper, Daniel, 937 Schröder, V., 273
Perdomo, L., 157 Schueller, M., 1155
Pereira, Helder A., 1491 Schulz, Wolfgang, 1169
Perov, Sergey, 131, 138 Schurch, Roger, 85, 899, 1367
Phansiri, Nisarut, 481 Sen, Qian, 572
Author Index 1559

Shen, Zichao, 1077 U

Sheng, Gehao, 449 Uhrig, S., 738
Shi, C., 606 Umemura, Tokihiro, 653
Shiina, Yasuharu, 643 Urushibata, H., 208
Silva, J. N., 1533
V
Sima, Wenxia, 1192, 1225, 1244
Villa, A., 375, 429
Sjöstedt, Hans, 1207
Vita, V., 1471, 1544
Smajic, J., 1155
Vobornik, Ales, 549
Solovyev, Y., 458
Vu, C-T., 1108
Somsak, Tanachai, 763
Soo-bong, Lee, 48 W
Soto, Leopoldo, 85, 1367 Wada, Yoshihiro, 472
Späck-Leigsnering, Yvonne, 118 Wakimoto, Kiyoshi, 752
Spiertz, Martin, 560 Wang, HongBin, 437
Staikos, E. T., 1328 Wang, Hui, 714
Steiner, Thomas, 560 Wang, Li, 1067
Stieger, N., 1155 Wang, Lijun, 1055
Štrac, Leonardo, 67 Wang, Liming, 862
Su, Haibo, 300, 1513 Wang, Sen, 1132
Su, Rui, 1055 Wang, Shuyuan, 714
Suehiro, Junya, 481 Wang, Siyuan, 449
Sun, Haofei, 1132 Wang, Xiaoping, 1067
Sun, Potao, 1192, 1225 Wang, Yong, 300, 850, 1513
Sun, T. X., 684 Wang, Z. D., 871
Suwanasri, Cattareeya, 171, 184, 197, 763 Wang, Zhong, 437
Suwanasri, Thanapong, 171, 184, 197, 763 Wang, Zhongdong, 969, 982
Suwarno, 926 Wang, Zihao, 1077
Swanson, A. G., 331 Weber, B., 1265
Szabó, Dávid, 145, 1025 Wei, Peng, 1077
Szirtes, Máté, 1003, 1015 Wei, Yuning, 38
Werle, Peter, 697, 705, 777, 788, 799, 838,
T 926, 958
Tagashira, Hideki, 635 Wilson, Gordon, 969, 982
Tahir, M., 811 Winkelmann, Erik, 560
Takahashi, Tsuguhiro, 232 Wölke, Bastian, 887
Tamus, Zoltán Ádám, 253, 261 Woo, Jung-Wook, 1347
Tanaka, Tsuyoshi, 673 Wu, Kai, 1055
Tang, Ju, 396, 406 Wu, Kaijian, 1513
Tausend, Wolfgang, 1458 Wu, Kangning, 1340
Tavares, Ricardo Augusto, 243 Wu, Qingzhou, 1087
Taylor, James, 1207 Wu, Siying, 406
Teixeira, Rômulo O., 1491, 1502 Wu, Yang, 1055
Tenbohlen, S., 219, 811
Thedens, M., 55 X
Thirumurthy, 498, 510 Xerra, S. I., 316
Thyssen, Pascal, 887 Xie, Dongri, 1077
Tiutiunnik, Elizaveta, 138 Xie, Shijun, 1192
Toth, J. C., 917 Xie, Zhicheng, 1244
Tsovilis, T. E., 1295, 1482 Xing, Zhaoliang, 108, 1044, 1087
Tsuji, S., 208 Xu, Yang, 752
1560 Author Index

Xu, Yongpeng, 449 Zhan, Yunpeng, 1132

Xu, Zhiqiang, 1132 Zhang, Chenmeng, 1192
Xue, Wenxiang, 714 Zhang, Chong, 108, 1044
Zhang, Jifei, 714
Y Zhang, Liangxian, 108, 1044
Yamamoto, Takashi, 662 Zhang, Mingxuan, 396
Yan, Chenyu, 108, 1087 Zhang, Shiling, 406
Yan, Wei, 1523 Zhang, Weidong, 38
Yang, Ming, 1244 Zhang, Yu, 1192
Yang, Xu, 572 Zhang, Yuanyuan, 878
Yao, Qiang, 396, 406 Zhang, ZiChong, 437
Yapi, K. M. L., 615 Zhao, Xuefeng, 1132
Ye, Lian, 1192 Zhao, Xuetong, 282
Yokoyama, Shigeru, 1377 Zheng, Shusai, 1067
Yoshida, Masanobu, 908 Zheng, Shusheng, 993
Yoshikawa, Hina, 635, 643 Zhou, H., 606
YoungGeun, Kim, 292 Zhu, Lingyu, 625, 949
Yuan, Zhao, 1513 Zhu, Lu, 300, 1513
Zhu, Yuanwei, 1077
Z Zhu, Yufeng, 449
Zagkanas, Vasileios N., 1413 Zink, Markus H., 560
Zeng, Fuping, 396, 406 Živic, Tomaž, 1390
Zeng, X. F., 684 Zoro, Reynaldo, 1423
Zhan, Cao, 949 Župan, Tomislav, 67, 75