Lecture Notes in Networks and Systems 602

Duy Cuong Nguyen · Ngoc Pi Vu ·

Banh Tien Long · Horst Puta ·
Kai-Uwe Sattler   Editors

Advances in
Engineering
Research and
Application
Proceedings of the International
Conference on Engineering Research
and Applications, ICERA 2022
Lecture Notes in Networks and Systems

Volume 602

Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland

Advisory Editors
Fernando Gomide, Department of Computer Engineering and Automation—DCA,
School of Electrical and Computer Engineering—FEEC, University of
Campinas—UNICAMP, São Paulo, Brazil
Okyay Kaynak, Department of Electrical and Electronic Engineering,
Bogazici University, Istanbul, Turkey
Derong Liu, Department of Electrical and Computer Engineering, University of
Illinois at Chicago, Chicago, USA
Institute of Automation, Chinese Academy of Sciences, Beijing, China
Witold Pedrycz, Department of Electrical and Computer Engineering, University of
Alberta, Alberta, Canada
Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland
Marios M. Polycarpou, Department of Electrical and Computer Engineering,
KIOS Research Center for Intelligent Systems and Networks, University of Cyprus,
Nicosia, Cyprus
Imre J. Rudas, Óbuda University, Budapest, Hungary
Jun Wang, Department of Computer Science, City University of Hong Kong,
Kowloon, Hong Kong
The series “Lecture Notes in Networks and Systems” publishes the latest
developments in Networks and Systems—quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core
of LNNS.
Volumes published in LNNS embrace all aspects and subfields of, as well as new
challenges in, Networks and Systems.
The series contains proceedings and edited volumes in systems and networks,
spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor
Networks, Control Systems, Energy Systems, Automotive Systems, Biological
Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems,
Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems,
Robotics, Social Systems, Economic Systems and other. Of particular value to
both the contributors and the readership are the short publication timeframe and
the world-wide distribution and exposure which enable both a wide and rapid
dissemination of research output.
The series covers the theory, applications, and perspectives on the state of the art
and future developments relevant to systems and networks, decision making, control,
complex processes and related areas, as embedded in the fields of interdisciplinary
and applied sciences, engineering, computer science, physics, economics, social, and
life sciences, as well as the paradigms and methodologies behind them.
Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.
For proposals from Asia please contact Aninda Bose (aninda.bose@springer.com).
Duy Cuong Nguyen · Ngoc Pi Vu ·
Banh Tien Long · Horst Puta · Kai-Uwe Sattler
Editors

Advances in Engineering
Research and Application
Proceedings of the International Conference
on Engineering Research and Applications,
ICERA 2022
Editors
Duy Cuong Nguyen Ngoc Pi Vu
Faculty of Electronic Engineering Faculty of Mechanical Engineering
Thai Nguyen University of Technology Thai Nguyen University of Technology
Thai Nguyen, Vietnam Thai Nguyen, Vietnam

Banh Tien Long Horst Puta

Vietnam Association for Science Editing Institute for Automation and Systems
Hanoi University of Science Engineering
and Technology Ilmenau University of Technology (IUT)
Hanoi, Vietnam Ilmenau, Germany

Kai-Uwe Sattler
Department of Computer Science
and Automation
Ilmenau University of Technology (IUT)
Ilmenau, Germany

ISSN 2367-3370 ISSN 2367-3389 (electronic)

Lecture Notes in Networks and Systems
ISBN 978-3-031-22199-6 ISBN 978-3-031-22200-9 (eBook)
https://doi.org/10.1007/978-3-031-22200-9

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2023, corrected publication 2023
This work is subject to copyright. All rights are solely and exclusively licensed 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

This book contains the proceedings of the 5th International Conference on Engi-
neering Research and Application 2022 (ICERA 2022) which was organized by
the Thai Nguyen University of Technology (TNUT), Vietnam, and cooperated with
Ilmenau University of Technology, Germany. The first conference started in 2018.
So far, the conference has attracted a lot of contributions from researchers of many
different universities around the world.
The aim of this conference is to bring together researchers from a broad range in the
field of research, theories, and practices in engineering research and applications. The
topics of this volume are Mechanical Engineering, Materials and Mechanics of Mate-
rials, Mechatronics and Micromechatronics, Automotive Engineering, Electrical and
Electronics Engineering, and Information and Communication Technology.
The up-to-date contributions reported in this book were carefully reviewed by
experts and also approved by editors for the last review. All 103 accepted papers
were presented and discussed at ICERA 2022, held in Thai Nguyen City, Vietnam,
on December 1–2, 2022. The total papers sent to this conference are 253 papers. As a
result of the two-stage review process, only 103 excellent contributions were selected
for presentation at the conference and publication in this book. The readers will find
here representative samples of the most modern techniques available nowadays for
the solution of challenging problems arising in engineering research and application.
We gratefully thank the authors for their contributions to the conference with very
interesting papers. We deeply thank the reviewers for their evaluations and recom-
mendations in order to improve the quality of the selected papers. We would like to
especially thank our Keynote speakers, Prof. Kai-Uwe Sattler (Ilmenau University of
Technology, Germany), Prof. S.A. Sherif (University of Florida, USA), Prof. Roger
A. Sauer (RWTH Aachen University, Germany/Gdansk University of Technology),
and Prof. Sunny Joseph Kalayathankal (Jyothi Engineering College, Cheruthuruthy,
India), for their valuable and inspiring contributions to scientists, researchers, and
listeners. We also thank the members of the Organizing Committee of ICERA2022
for their excellent technical and editorial support.

v
vi Preface

Last but not least, we would like to deeply thank Springer Publishers and its Editor
staff for helping us in the publication of the proceeding volume of this book.

Thai Nguyen, Vietnam Duy Cuong Nguyen

December 2022 Ngoc Pi Vu
Banh Tien Long
Horst Puta
Kai-Uwe Sattler
Keynote 1: In-Database Machine Learning for
the Masses

Kai-Uwe Sattler
Ilmenau University of Technology
kus@tu-ilmenau.de

Abstract Data scientists and analysts regularly need to process large amounts of
data. In recent years, the Python language has evolved to the de-facto standard for
these users and powerful frameworks have been developed supporting a wide range
of processing and analytical tasks. However, most of these frameworks extract data
to be analyzed from the database server to the client machine to process it there in
Python. Not only the large transfer costs but also the limited client memory impacts
the usability of this approach, as data might exceed typical client memory amounts.
On the other hand, database systems are no longer used only for storing structured
data and simple queries. Ongoing research aims at integrating ML models, e.g. neural
networks, and their use for prediction and classification.
In the talk, we present approaches to leverage the power of database systems to
support analysis tasks of data scientists.
First, we describe the mapping of Python functionalities provided in popular
frameworks such as Pandas to SQL queries. We further discuss approaches for inte-
grating inference steps of Machine learning tasks into database query processing.
Second, using the example of graph databases, we demonstrate how analytics on
in-database graphs can be accelerated by the seamless integration of GPU processing.

vii
Keynote 2: Advances in Decarbonizing the US
Economy and Net-Zero Energy and Latest
Research in Energy Efficiency
S. A. Sherif
University of Florida
sasherif@ufl.edu

Abstract The US Department of Energy is pursuing an ambitious agenda aiming

at decarbonizing the energy economy of the United States. This presentation will
discuss the latest advances in energy efficiency research and development and will
give a status report on the progress toward decarbonizing the US economy and the
progress made toward achieving net-zero energy. The DOE Earthshots Initiative will
also be discussed in the context of this ambitious agenda.

Biography
Dr. S. A. Sherif is a Professor of Mechanical and Aerospace Engineering at the
University of Florida. He is a Life Fellow of the American Society of Mechanical
Engineers (ASME); a Life Fellow of the American Society of Heating, Refriger-
ating, and Air-Conditioning Engineers (ASHRAE); a Fellow of the Royal Aero-
nautical Society; an Associate Fellow of the American Institute of Aeronautics and
Astronautics (AIAA); a Vice President of Commission B-2 on Refrigerating Equip-
ment of the International Institute of Refrigeration; a Member of the Advisory Board
of Directors of the International Association for Hydrogen Energy (IAHE); and a
Founding Member of the Board of Directors of the American Society of Thermal
and Fluids Engineers (ASTFE). Dr. Sherif currently serves as Editor-in-Chief of the
ASME Journal of Solar Energy Engineering. He also served as Editor-in-Chief of
the ASME Journal of Thermal Science and Engineering Applications (2014–2019).
He is on the editorial boards of 27 other archival journals. Dr. Sherif has more than
300 refereed papers, one book, 20 bound volumes, 24 book chapters, 200 technical
reports, and two US patents.

ix
Keynote 3: Isogeometric Finite Elements:
Motivation, Implementation, and Applications

Roger A. Sauer
Graduate School AICES, RWTH Aachen University/Civil Engineering, Gdansk
University of Technology
sauer@aices.rwth-aachen.de

Abstract Isogeometric finite elements are a new class of finite elements that have
been developed by combining computer-aided design (CAD) with finite element
analysis [1]. They use spline-based shape functions that have several advantages over
classical Lagrange-based finite element shape functions. They can be locally refined,
and they can be embedded straightforwardly into existing finite element software
using the Bezier extraction operator [2], although special attention has to be paid to
boundary conditions, especially for curved boundaries. The isogeometric paradigm
is particularly advantageous for solving high-order partial differential equations,
such as those of the Kirchhoff-Love shell theory [3]. This is illustrated by several
computational application examples from science and engineering.

References

1. Hughes, T. J. R., Cottrell, J. A., and Bazilevs, Y. (2005). Isogeometric analysis: CAD, finite
elements, NURBS, exact geometry and mesh refinement. Comput. Methods Appl. Mech.
Engrg., 194:4135–4195.
2. Borden, M. J., Scott, M. A., Evans, J. A., and Hughes, T. J. R. (2011). Isogeometric finite
element data structures based on Bezier extraction of NURBS. Int. J. Numer. Meth. Engng.,
87:15–47.
3. Duong, T. X., Roohbakhshan, F., and Sauer, R. A. (2017). A new rotation-free isogeometric
thin shell formulation and a corresponding continuity constraint for patch boundaries. Comput.
Methods Appl. Mech. Engrg., 316:43–83.

xi
Keynote 4: Fuzzy Modeling
and Decision-Making Applications
in Engineering Science
Sunny Joseph Kalayathankal
Principal (Professor & Dean of Research), Jyothi Engineering College,
Cheruthuruthy, India
sjkalayathankal@jecc.ac.in

Abstract The thought process involved in the act of decision-making is a complex

array of streaming possibilities in which a person selects or discards information made
available from diverse sources. In doing so, one is led by a meaningful analysis of
available information and optimal selection out of several apparently equi-efficient
decisions. Since Zadeh (1965) published the fuzzy set theory as an extension of classic
set theory, it has been widely used in many fields of application, such as pattern recog-
nition, data analysis, and system control. The unique characteristic of this theory, in
contrast to classic mathematics, is its operation on various membership functions
(MF) instead of the crisp real values of the variables. Molodtsov (1999) initiated the
concept of soft set theory as a new mathematical tool for dealing with uncertainties.
Pabitra Kumar Maji et al. (2001) introduced fuzzy soft set theory which also deals
with uncertainties. Out of the several higher order fuzzy sets, intuitionistic fuzzy sets
by Atanassov (1985) and Ordered intuitionistic fuzzy sets proposed by Kalayathanal
et al. (2010) have been found to be highly useful to deal with vagueness. Intuition-
istic fuzzy set is described by two functions: a membership function and a non-
membership function. We develop and apply similarity measures between ordered
intuitionistic fuzzy sets to multiple attribute decision-making (MADM) under a fuzzy
environment.
In 1965, Zadeh published the fuzzy set theory as an extension of classic set theory,
and it has been widely used in many fields of application, such as pattern recogni-
tion, data analysis, and system control. The unique characteristic of this theory, in
contrast to classic Mathematics, is its operation on various membership functions
(MF) instead of the crisp real values of the variables. Molodtsov (1999) initiated
the concept of soft set theory as a new mathematical tool for dealing with uncer-
tainties. Pabitra Kumar Maji et al. (2001) introduced fuzzy soft set theory which
also deals with uncertainties. Atanassov (1985) introduced intuitionistic fuzzy sets
which contain membership and non-membership functions. We develop and apply
similarity measures between various higher order fuzzy sets and their applications
under a fuzzy environment.

xiii
xiv Keynote 4: Fuzzy Modeling and Decision-Making Applications …

Biography
Prof. Dr. Sunny Joseph Kalayathankal received an M.Sc. degree from Kerala Univer-
sity, Kerala, India, in 1986; B.Ed. from Calicut University, Kerala, in 1987; M.Phil.
from Kerala University in 1993; Ph.D. (Mathematics) degree in 2010 from Kerala
University; MCA from Indira Gandhi National Open University, New Delhi, India,
in 2002; M.Tech. IT from Karnataka State Open University in 2013; and Ph.D. in
Computer Science under Bharathiar University, Coimbatore, India, in 2018. He is
currently working as a Principal (Professor & Dean of Research) at Jyothi Engi-
neering College Affiliated to APJ Abdul Kalam Technological University, Thrissur,
Kerala, India, and has 34 years and 8 months of teaching and 16 years of research
experience. He has published more than 91 papers in the areas of Fuzzy modeling
and decision-making, Graph theory, and Applied Mathematics. He has served as
Keynote and invited speaker at various National and International conferences. He
is the reviewer of the Iranian Journal of Fuzzy System, the International Journal of
Fuzzy system, and the Journal of Mathematical Modeling and Computer Simulation.
Contents

Cost Optimization Study for Two-Stage Helical Gearbox

with Second Stage Double Gear Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Nguyen Hong Linh, Tran Huu Danh, Bui Thanh Danh, Tran Minh Tan,
Nguyen Van Trang, and Luu Anh Tung
A Design and Implement of Fuzzy Controller
for Taking-off and Landing for Unmanned Aerial Vehicles . . . . . . . . . . . . 13
Dong L. T. Tran, Hai T. Do, Hoang T. Tran, Thang Hoang,
and Minh T. Nguyen
A Detailed Procedure of Squirrel-Cage Three-Phase Induction
Motor Parameter Estimation Using Polynomial Regression . . . . . . . . . . . . 23
Son T. Nguyen, Stefan Goetz, Tu M. Pham, Anh Hoang,
and Tuan V. Pham
A Hybrid Energy Sliding Mode Controller for the Rotary Inverted
Pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Van-Anh Nguyen, Duc-Binh Pham, Duy-Thi Pham, Ngoc-Tam Bui,
and Quy-Thinh Dao
A Neural Network-Based Fast Terminal Sliding Mode Controller
for Dual-Arm Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Hue Luu Thi, Van Trong Dang, Nhu Toan Nguyen, Duc Thinh Le,
and Tung Lam Nguyen
A Novel Approach for Determination of Main Dimensions
and Armature Reactions of Axial Flux Permanent Magnet
Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Thanh Nguyen Vu, Tan Nguyen Viet, and Vuong Dang Quoc
A Prototype of Lotus Fiber Extracting Machine . . . . . . . . . . . . . . . . . . . . . . 65
Ngoc-Kien Nguyen and Van-Tinh Nguyen

xv
xvi Contents

A Simulation Investigation of Dynamic Wheel Load of a Heavy

Truck with Hydro-Pneumatic Suspension System . . . . . . . . . . . . . . . . . . . . . 74
Le Xuan Long, Dang Viet Ha, Nguyen Van Tuan, Vu Thanh Niem,
and Vu Thi Hien
A Study of Kinematics, Dynamics of the Stair Climbing
Wheelchairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Nguyen Hoang Hiep, Le Hong Ky, Duong Tan Dat,
and Le Hong Ky Duyen
A Study of Polygonal Holes Machining Using Attachment . . . . . . . . . . . . . 93
Anh-Tuan Dang and Dinh-Ngoc Nguyen
Study on Multi-criteria Decision Making in CBN Grinding SKD11
Tool Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Bui Thanh Danh, Tran Huu Danh, Vu Van Khoa, Pham Duc Lam,
Nguyen Hong Linh, Nguyen Van Tung, and Bui Thanh Hien
Study on Multi-criteria Optimization in CBN Grinding SKD11
Tool Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Trieu Quy Huy, Vu Van Khoa, Bui Thanh Danh, Tran Huu Danh,
Hoang Xuan Tu, Vu Ngoc Pi, and Bui Thanh Hien
A Study on Optimization of Helical Worm Gearboxes Based
on Volume Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Bui Thanh Danh, Le Hoang Anh, Tran Huu Danh,
Tran Ngoc Huy Thinh, Trinh Kieu Tuan, Luong Viet Dung,
and Tran Thi Phuong Thao
A Study on Wearing and Tool Life of Shredding Machine Cutters
When Cutting Fabric Industrial Waste . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
Trong Hai Nguyen and Ngoc Tuyen Bui
A System Engineering Approach to Model and Implement Planar
Trajectory-Tracking Controllers for Autonomous Underwater
Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Ngo Van Hien, Hoang Cong Liem, and Ngo Van He
Active Disturbance Rejection Control of an Antagonistic Muscle . . . . . . . 150
Quy-Thinh Dao, Duc-Canh Nguyen, Minh-Chien Trinh,
Van-Vuong Dinh, Trong-Hieu Do, and Minh-Duc Duong
An Approach for Optimizing the Hedge-Algebras-Based
Controller and Application in Structural Vibration Control . . . . . . . . . . . 157
Hai-Le Bui
An Enhanced Hybrid Jaya Algorithm for Size Optimization
of Truss Structure Under Frequency Constraints . . . . . . . . . . . . . . . . . . . . . 166
Nga T. T. Nguyen, Sy Nguyen-Van, Thuy T. T. Diem,
Ngoc Nguyen-Dinh, Tien-Dat Hoang, and Luong Viet Dung
Contents xvii

An Evaluation of Some Specifications of Turbine Blades Made

by 3D Printing Technology and Processed on CNC Milling
Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Nguyen Hoang Hiep, Le Hong Ky, Pham Hoang Anh,
Le Hong Ky Duyen, and Tran Vinh Hung
An Optimal Cascade Reservoir Operation Based on Multi-objective
Water Cycle Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
Thi-Kien Dao, Trong-The Nguyen, The-Vinh Do, Trinh-Dong Nguyen,
and Vinh-Tiep Nguyen
Analysis of Insulation Characteristics of C2 F6 -N2 Mixtures . . . . . . . . . . . . 201
Pham Xuan Hien and Do Anh Tuan
Analysis of Surface Defects and Tool Wear in Edge Trimming
of CFRPs by Optical Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Dinh-Ngoc Nguyen and Anh-Tuan Dang
Anti-torque Controller Using Sliding Mode Linear Quadratic
Regulator Applied for Coaxial BLDC Motor . . . . . . . . . . . . . . . . . . . . . . . . . 217
Lam Cuong Quoc Thai, Van Trong Nguyen, Cong Thang Doan,
Van Tu Duong, Huy Hung Nguyen, and Tan Tien Nguyen
Application of EDAS Method for Best Dressing Mode for Internal
Cylindrical Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Trieu Quy Huy, Nguyen Hong Linh, Tra n Huu Danh,
Bui Thanh Danh, Do Thi Tam, Luu Anh Tung, and Le Xuan Hung
Application of MARCOS Method for Selecting the Best Schema
of Scissors Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
Trieu Quy Huy, Nguyen Ba Liem, Tran Quang Hau,
Dang Quoc Cuong, Tran Huu Danh, Nguyen Thi Thanh Nga,
Vu Ngoc Pi, and Nguyen Ngoc Thieu
Application of MCDM Method in Selection of Schema for Optimal
Design of Double Scissor Lift Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
Nguyen Hong Linh, Trieu Quy Huy, Tran Minh Tan, Tran Huu Danh,
Bui Thanh Danh, Nguyen Thi Thanh Nga, Vu Ngoc Pi,
and Nguyen Ngoc Thieu
Application of TOPSIS Method to Determine Best Alternative
in Wire-EDM 90CrSi Tool Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
Trieu Quy Huy, Le Hong Ky, Le Hoang Anh, Bui Thanh Danh,
Nguyen Manh Cuong, and Nguyen Thanh Tu
Application of WASPAS Method for Determining Best Process
Factors in External Grinding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Tran Huu Danh, Bui Thanh Danh, Trieu Quy Huy, Tran Minh Tan,
Nguyen Hong Linh, Tran Ngoc Giang, and Hoang Xuan Tu
xviii Contents

Auto-balancing Ballbot Systems: A Fractional-Order Sliding

Mode Based Radial-Basis Neural Network Approach . . . . . . . . . . . . . . . . . 270
Minh Duc Pham, Cong Minh Pham, Thu Giang Do,
Phuong Thao Dao, Trong Dang Van, Quoc Dong Hoang,
and Tung Lam Nguyen
Backstepping Sliding Mode Control Design for Active Suspension
Systems in Half-Car Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Viet Hung Pham, Duc Thinh Le, Nhu Toan Nguyen, Van Trong Dang,
Van-Anh Nguyen-Thi, Danh Huy Nguyen, and Tung Lam Nguyen
Balancing Composite Motion Optimization and Artificial Neural
Network for the Prediction of Critical Load of Concrete-Filled
Steel Tubes Under Axial Compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
Huan Thanh Duong, Tien-Thinh Le, Xuan Son Nguyen,
Minh Vuong Le, Hieu Chi Phan, Lu Minh Le, and Van Hai Nguyen
Building Power Demand Forecasting Using Machine Learning:
Application for an Office Building in Danang . . . . . . . . . . . . . . . . . . . . . . . . 297
Hoang-Anh Dang and Van-Dung Dao
Calculating Optimum Input Factors for Maximum Material
Removal Rate in Wire-EDM SKD11 Tool Steel . . . . . . . . . . . . . . . . . . . . . . . 305
Nguyen Hong Linh, Tran Huu Danh, Bui Thanh Danh,
Tran Minh Tam, Nguyen Anh Tuan, Trinh Kieu Tuan,
Nguyen Manh Cuong, and Nguyen Thi Thanh Nga
Calculating Optimum Main Design Factors of a Two-Stage Helical
Gearboxes for Minimum Gearbox Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
Do Quang Khai, Nguyen Hong Linh, Tran Huu Danh,
Tran Minh Tan, Nguyen Manh Cuong, Bui Thanh Hien, Vu Ngoc Pi,
and Nguyen Thi Quoc Dung
Deep ANFIS for Understanding Emotions in Movie Clips . . . . . . . . . . . . . 326
Tuan-Linh Nguyen
Determination of Best Input Factors for PMEDM 90CrSi Tool
Steel Using MABAC Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
Nguyen Hong Linh, Phan Dang Phong, Thangaraj Muthuramalingam,
Tran Minh Tan, Tran Huu Danh, Vu Ngoc Pi, Hoang Xuan Tu,
and Nguyen Van Tung
Determination of Optimum Main Design Parameters
of a Two-Stage Helical Gearbox for Minimum Gearbox
Cross-Section Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
Do Quang Khai, Tran Huu Danh, Bui Thanh Danh, Tran Minh Tan,
Nguyen Manh Cuong, Le Xuan Hung, Vu Ngoc Pi,
and Tran Thi Phuong Thao
Contents xix

Determination of the Location of the Photo- and Video-Fixation

System of Traffic Rules in the Context of Vietnam . . . . . . . . . . . . . . . . . . . . 354
Nguyen Xuan Hien, Vu Thi Van Anh,
and Krylov Grigory Alexandrovich
Determining Best Dressing Parameters for Internal Cylindrical
Grinding Using MABAC Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
Nguyen Hong Linh, Trieu Quy Huy, Tran Huu Danh,
Tran Ngoc Huy Thinh, Bui Thanh Danh, Le Xuan Hung,
Hoang Xuan Tu, and Do Thi Tam
Determining Best Input Parameters for CBN Grinding Al6106 T6
Using WASPAS Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Do Quang Khai, Tran Huu Danh, Bui Thanh Danh,
Nguyen Manh Cuong, Hoang Xuan Tu, and Nguyen Van Trang
Determining Partial Gear Ratios of a Two-Stage Helical Gearbox
with First Stage Double Gear Sets for Minimizing Total Gearbox
Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
Tran Huu Danh, Trieu Quy Huy, Bui Thanh Danh, Tran Minh Tan,
Nguyen Van Trang, and Luu Anh Tung
Determining the Best Alternative in Powder-Mixed Electrical
Discharge Machining 90CrSi Tool Steel Using the TOPSIS Method . . . . 389
Nguyen Anh Tuan, Trieu Quy Huy, Tran Huu Danh, Bui Thanh Danh,
Do Thi Tam, Nguyen Van Tung, and Le Xuan Hung
Develop an Advanced Driver’s Behaviors Detection System . . . . . . . . . . . . 395
Tung-Duong Pham, Hai-Hung Bui, Nhat-Linh Nguyen-Bui,
Dinh-Tien Nguyen, Trong-Phuoc Le, Chi-Quan Le,
Thanh-Tung Nguyen, and Hoang-Phuc Dam
Developing an Intelligent Waste Sorting with 6DOF Robotic Arm . . . . . . 402
Vo Thu Ha, Than Thi Thuong, and Vo Thanh Ha
Discrete Element Modeling Based Determination of Shear
Behavior in a Granular Medium Through Displacement Field . . . . . . . . . 412
Tien-Thinh Le, Thong Chung Nguyen, Xuan Son Nguyen,
Huan Thanh Duong, Lu Minh Le, Anh Tu Nguyen, and Van Hai Nguyen
Double-Loop PID Control with Parameter Optimization
for an Autonomous Electric Vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Quoc-Dong Hoang, Le Dinh Nghiem, Nguyen Duc Toan,
Pham Khanh Hung, Luu Quang Do, Dang Trung Hieu,
Tung Lam Nguyen, and Le Sy Dung
xx Contents

Effect of Input Factors on Material Removal Speed in Electrical

Discharge Machining SKD11 Tool Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
Nguyen Hong Linh, Bui Thanh Danh, Tran Minh Tam,
Tran Huu Danh, Nguyen Huu Quang, Nguyen Thanh Tu,
and Do Thi Tam
Effect of Main Blade Configuration on the Performance
of the Optimized Multicurve Savonius Wind Turbine . . . . . . . . . . . . . . . . . 433
Banh Duc Minh, Le Dinh Anh, Tran Cuong Hung, Doan Viet Ha,
Tran Cong Manh Hung, and Nguyen Thi Thu Phuong
Effect of Mass Reduction and Aerodynamic Drag Coefficient
on Fuel Consumption Reduction on Student Fuel-Saving Car . . . . . . . . . . 439
Nguyen Thanh Quang, Ngo Quang Tao, and Le Hong Quan
Effect of Composite Materials to Vehicle Body Vibration . . . . . . . . . . . . . . 448
Nguyen Thanh Quang
Effect of Printing Parameters on Characteristics of PCL Scaffold
Fabricated by Direct Powder Extrusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
Quang Duy Do, Du Khac Nguyen, Tung Van Nguyen,
Lan Xuan Phung, and Trung Kien Nguyen
Effect of Process Parameters on Material Removal Speed in CBN
Grinding Al6061 T6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 462
Bui Thanh Danh, Tran Huu Danh, Nguyen Hong Linh,
Nguyen Anh Tuan, Nguyen Van Tung, Hoang Xuan Tu,
and Nguyen Van Trang
Effects of Voltage Unbalance on Matrix Converter Induction
Motor Drive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
Nhu Y Do and Xuan Cuong Ngo
Electrocardiogram (ECG) Circuit Design and Using the Random
Forest to ECG Arrhythmia Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
Van Nam Pham and Hoai Linh Tran
Electromagnetic Torque and Force Analysis of Switched
Reluctance Motors with Different Rotor Pole Shapes . . . . . . . . . . . . . . . . . 495
Bui Duc Hung, Dinh Bui Minh, and Vuong Dang Quoc
Energy-Based Tracking Control with Vertical Vibration
Suspending for a 6-DOF Bridge Crane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
Quoc-Dong Hoang, Le Anh Tuan, Do Duc Luu, and Le Sy Dung
Enhancing Total Transfer Capability via Optimal Location
of Energy Storage Systems Using a Hybrid Improved Min-Cut
Algorithm and Genetic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512
Tung Linh Nguyen, Ngoc Sang Dinh, Viet Anh Truong,
Thanh Long Duong, Dao Huy Du, and Do Anh Tuan
Contents xxi

Evaluation of Model Order Reduction Algorithms for Unstable

High-Order System Applied to Large Power System . . . . . . . . . . . . . . . . . . 525
Thanh-Tung Nguyen, Huy-Du Dao, Ngoc-Kien Vu, Van-Ta Hoang,
and Thanh-Binh Nguyen Thi
Evaluation of Rollover Stability of Liquid Tank Truck in the Lane
Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
Xuan Ngoc Nguyen, Van Nhu Tran, Van Tan Vu, and Tien Phuc Dang
Factors Affecting Saving Energy Intention of Urban Residents
in Hanoi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
Tung Thanh Nguyen, Nhung Tang Cam, Kien Trung Duong,
and Tuan Anh Do
H∞ Optimal Tracking Control for Robot Manipulators with Input
Constraint and Disturbances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
N. D. Dien, N. T. Luy, and L. K. Lai
Hybrid Energy Storage Systems: A Brief Overview . . . . . . . . . . . . . . . . . . . 573
Nicola Campagna, Vincenzo Castiglia, Rosario Miceli, Fabio Viola,
Alessandro Busacca, and Minh T. Nguyen
Improving the Surface Characteristics of Mg AZ31B Alloy by RF
Magnetron Sputtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 580
Anup Malik, Harlal Singh Mali, and Arvind Kumar Singh
Influence of Dressing Parameters on Surface Roughness and Wheel
Life in Internal Grinding SKD11 Tool Steel . . . . . . . . . . . . . . . . . . . . . . . . . . 589
Tran Huu Danh, Nguyen Hong Linh, Bui Thanh Danh,
Nguyen Anh Tuan, Nguyen Thi Quoc Dung, Hoang Xuan Tu,
and Le Xuan Hung
Influence of Input Factors on Surface Roughness When Using
Wire-EDM SKD11 Tool Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
Tran Huu Danh, Nguyen Hong Linh, Bui Thanh Danh,
Tran Minh Tam, Nguyen Huu Quang, Tran Ngoc Giang,
Nguyen Manh Cuong, and Nguyen Thi Thanh Nga
Influence of Input Process Factors on Surface Roughness in CBN
Grinding Al6061 T6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607
Nguyen Hong Linh, Bui Thanh Danh, Tran Huu Danh,
Nguyen Anh Tuan, Nguyen Van Tung, Hoang Xuan Tu,
and Nguyen Van Trang
Influence of WEDM Process Parameters on Material Removal
Rate When Machining Circular Profile of Hardened SKD11 Steel . . . . . . 613
Nguyen Anh Tuan, Nguyen Hong Linh, Tran Huu Danh,
Nguyen Huu Luan, Bui Thanh Danh, Do Thi Tam,
Nguyen Thi Quoc Dung, and Nguyen Dinh Ngoc
xxii Contents

Iterative Learning Control for Completely Uncertain CSTR

with Matched Disturbance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623
Trung Thanh Cao, Nam Hoai Nguyen, and Phuoc Doan Nguyen
Load Sharing Control of Two Ac Motors with a Common Stiff
Shaft by Using an Adaptive Fuzzy Controller . . . . . . . . . . . . . . . . . . . . . . . . 633
Nguyen Trung Thanh and Dang Danh Hoang
Mathematical Model of the Workflow of a Modern Spark-Ignition
Engine Running on Ammonia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
Andrey Y. Dunin, Dung Van Nguyen, and Nguyen Khac Minh
Mixed Convection of a Non-Newtonian Fluid in a Square Enclosure
with Rotating Cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 662
Swaranjali Maurya, Ram Dayal, and Manish Kumar
Modeling and Dynamic Simulation of a Multi-slope Vibrating
Screen to Determine the Basic Working Parameters . . . . . . . . . . . . . . . . . . 671
Tan Nguyen Dang
Module Integrated Converters and Independent MPPT Technique . . . . . 685
Bui Van Hien, Truong Viet Anh, Nguyen Duc Minh,
Trinh Trong Chuong, Y. Do Nhu, and Trieu Viet Phuong
Multi-criteria Decision Making in Wire-EDM Process Using
MARCOS Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699
Trieu Quy Huy, Tran Huu Danh, Bui Thanh Danh,
Nguyen Van Cuong, Nguyen Manh Cuong, and Nguyen Thanh Tu
Multi-objective Optimization in Internal Grinding SKD11 Tool
Steel in Terms of Surface Roughness, Material Removal Rate
and Wheel Life . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 707
Tran Huu Danh, Nguyen Hong Linh, Bui Thanh Danh,
Nguyen Anh Tuan, Do Thi Tam, Hoang Xuan Tu, and Le Xuan Hung
Novel Energy Efficient Schemes for Wireless Sensor Networks
Utilizing Mobile Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 718
Minh T. Nguyen, Cuong V. Nguyen, Quynh C. Le, Binh D. Ha,
Hoang T. Tran, Vuong Q. Dang, Mien L. Trinh, and Fabio Viola
Numerical Analysis of Capacity Augmentation of Finned-Tube
Type Waste Heat Recovery Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 733
Amit Arora and P. M. V. Subbarao
Numerical Investigation into the Effects of Fuel Injection
Parameters and Piston Bowl Geometry on Exhaust Pollutants and
Performance of Diesel Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 739
Doan Nguyen Cong and Khanh Nguyen Duc
Contents xxiii

Obstacle Avoidance Algorithm for Autonomous Mobile Robots

in the Indoor Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 752
Nguyen Hong Thai, Trinh Thi Khanh Ly, Nguyen Thanh Long,
and Than Thi Thuong
Optimization of Central Pattern Generator-Based Locomotion
Controller for Fish Robot Using Deep Deterministic Policy Gradient . . . 764
Quoc Tuan Vu, Minh Huy Pham, Van Dong Nguyen, Van Tu Duong,
Huy Hung Nguyen, and Tan Tien Nguyen
Optimization of Main Design Parameters for a Two-Stage Helical
Gearbox Based on Gearbox Volume Function . . . . . . . . . . . . . . . . . . . . . . . . 771
Tran Huu Danh, Trieu Quy Huy, Bui Thanh Danh, Nguyen Van Trang,
and Le Xuan Hung
Pavement-Friendly Performance Analysis of an Agricultural
Truck with Two Types of Hydro-Pneumatic Suspension Struts . . . . . . . . . 780
Ta Tuan Hung, Le Xuan Long, Nguyen Van Tuan, Hoang Anh Tan,
and Vu The Truyen
Performance Analysis of a Mining Dump Truck Ride Comfort
with a Hydro-Pneumatic Suspension System under Different
Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 790
Dang Viet Ha, Le Van Quynh, and Le Xuan Long
Problematic Issues of Improving the Quality of Reception in VHF
SDR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 798
I. Yu. Malevich, Minh Tuan Nguyen, P. V. Zayats,
Thuong TK. Nguyen, and Quynh C. Le
Radial Basis Function Based Finite Element Method for Bending,
Vibration and Buckling Analysis of Laminated Composite
Mindlin-Reissner Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 806
D. Nguyen Kien, Xuefeng Chen, Xiaoying Zhuang, and Timon Rabczuk
Thermal Analysis of Drum Brake Mechanism for a Heavy Truck . . . . . . 823
Bui Van Cuong, Do Van Quan, Le Van Quynh, and Nguyen Thanh Cong
Research on Building the System of the Identification and Detection
Traffic Violations Vehicles at Vietnamese Intersections . . . . . . . . . . . . . . . . 831
Trinh Luong Mien and Vu Van Duy
Modified Skyhook Control for Semi-active Electric Vehicle
Suspension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 839
Le Van Quynh, Bui Van Cuong, Hoang Anh Tan, and Canh Chi Huan
Ride Comfort Evaluation for a Wheel Loader with Cab’s
Hydraulic Isolation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 846
Canh Chi Huan, Dang Viet Ha, Le Anh Vu, Tran Van Thoan,
and Le Van Quynh
xxiv Contents

Robust Hierarchical Sliding Mode Control for Planar

Flexible-Joint Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855
Nguyen Thi Thu Hien, Tran Xuan Minh, and Dang Danh Hoang
Selection of Input Parameters in External Cylindrical Grinding
Using EDAS Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 864
Tran Huu Danh, Bui Thanh Danh, Trieu Quy Huy, Tran Minh Tan,
Nguyen Hong Linh, Luu Anh Tung, and Hoang Xuan Tu
Stability of Microgrids: An Application of Virtual Synchronous
Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 873
Guido Ala, Nicola Campagna, Massimo Caruso, Vincenzo Castiglia,
Antonino Oscar Di Tommaso, Rosario Miceli, Claudio Nevoloso,
Giuseppe Schettino, Fabio Viola, and Minh Nguyen
Effects of Design Parameters of Cab’s Suspension System
on an Agricultural Tractor Ride Comfort . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
Bui Van Cuong, Canh Chi Huan, Le Van Quynh, and Doan Thanh Binh
Study the Impact of Cutting Edge Number Engaged on the Surface
Integrity During Machining Laminate Composites . . . . . . . . . . . . . . . . . . . 887
Dinh-Ngoc Nguyen and Anh-Tuan Dang
Studying Optimal Set of Input Parameters for CBN Grinding
Aluminum 6061T6 on CNC Milling Machine . . . . . . . . . . . . . . . . . . . . . . . . . 895
Bui Thanh Danh, Luu Anh Tung, Tran Huu Danh, Tran Minh Tan,
Trinh Kieu Tuan, Duong Van Phuong, and Nguyen Dinh Ngoc
Studying the Braking Efficiency of the Tractor Semi-trailer
on a Straight Road with Road Surface Profiles According to ISO
8608–2016 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903
Nguyen Thanh Tung and Luong Van Van
The Design of the Compound Non-circular Gear Train Has
an Internal Gears Pair with an Improved Cycloid Tooth Profile . . . . . . . . 910
Nguyen Thanh Trung, Phung Van Thom, and Nguyen Hong Thai
Effect of Main Design Factors on Two-Stage Helical Gearbox
Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 924
Tran Huu Danh, Nguyen Hong Linh, Bui Thanh Danh, Tran Minh Tan,
Nguyen Van Trang, Tran Thi Phuong Thao, and Nguyen Manh Cuong
The Experimental Research on Mechanical Properties
and Microstructural Characteristics of Narrow Gap Welding
Carbon Steel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 933
Minh Hung Ha, Trong Binh Ngo, and Duong Vu
Contents xxv

The Experimental Research on Plasma Welding of SUS 304

Stainless Steel Thin Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 942
Minh Hung Ha, Hai Tinh Do, Duong Vu, Duy Trung Dao,
and Thu Quy Le
The Orbit Tracking Adaptive Control for Omnidirectional Mobile
Robot Based on RBF Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 952
V. T. T. Linh, V. V. Thong, R. V. Hoa, V. D. Hung, T. D. Chuyen,
Tran Manh Hung, and Pham Ngoc Dien
Trends in the Development of Battery Thermal Management
Systems on Electric Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 968
S. Bazavluk, R. Kurmaev, K. E. Karpukhin, N. K. Minh, and N. K. Tuan
Wear Analysis and Microstructural Characterization of Al-MMC
Manufactured by Stir Casting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 977
Pankaj Kumar Gupta, Mayank Verma, and Umesh Vishwakarma
Correction to: Determination of Best Input Factors for PMEDM
90CrSi Tool Steel Using MABAC Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . C1
Nguyen Hong Linh, Phan Dang Phong, Thangaraj Muthuramalingam,
Tran Minh Tan, Tran Huu Danh, Vu Ngoc Pi, Hoang Xuan Tu,
and Nguyen Van Tung

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 985

 Cost Optimization Study for Two-Stage Helical
Gearbox with Second Stage Double Gear Sets

Nguyen Hong Linh1 , Tran Huu Danh2 , Bui Thanh Danh3 , Tran Minh Tan4 ,
Nguyen Van Trang5 , and Luu Anh Tung5(B)
1 Electric Power University, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

luuanhtung@tnut.edu.vn

Abstract. This paper presents the results of a cost optimization problem to deter-
mine the optimum partial gear ratios of a two-stage helical gearbox with the
second-stage double gear-sets. To accomplish the task, a simulation experiment
was designed and solved by computer programming and applying the Minitab R18
software and the Taguchi method. The influence of the main design parameters
as well as cost components on the optimal partial gear ratio of the first stage was
investigated. In addition, a regression formula to calculate the optimal partial ratio
has been introduced.

Keywords: Helical gearbox · Partial gear ratio · Optimum gearbox design ·

Gearbox cost

1 Introduction

A mechanical drive system (Fig. 1) is made up of a motor, a gearbox, and two couplings,
or a coupling and a V-belt or chain drive. The gearbox is the most important component,
as it reduces the speed and torque from the motor shaft to the working shaft (conveyor
drive shaft in Fig. 1). As a result, many researchers are pursuing the optimal design of
the mechanical drive system and the optimal design of the gearbox.
Up to now, there have been numerous studies on the optimal design of gearboxes.
Bernd-Robert Höhn et al. [1] discussed various methods for reducing power loss in
a gearbox. Abhishek Parmar et al. [2] proposed a novel multi-objective optimization
method for planetary gearboxes based on a discrete variant of the Non-Dominated Sorting
Genetic Algorithm. Yulong Lei et al. [3] conducted research on the vibration and noise
reduction of electric bus gearboxes using multi-objective optimization. The authors in
[4] used Hybrid Taguchi-Grey Relational Analysis to optimize process parameters for
thermal behavior of worm gearbox lubricant oil. Pan Hongxia et al. [5] proposed a study
on gearbox fault diagnosis using Particle Swarm Optimization.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 1–12, 2023.
https://doi.org/10.1007/978-3-031-22200-9_1
2 N. H. Linh et al.

The optimal partial gear ratio has received a lot of attention in optimal gearbox design.
This is due to the fact that it is the primary determinant of the gearbox’s size, mass, and
cost. Gearbox design optimization investigations have been reported out with a variety of
objective functions, including minimum gear mass [6, 7], minimum length of gearboxes
[7–11], minimum height of gearboxex [12, 13], or minimum gearbox cross-section area
[14–19]. Several cost optimization studies have recently been conducted to calculate the
optimal gear ratio to obtain the lowest gearbox cost [20, 21]. It is noteworthy that the
cost optimization problem in [21] has taken into account the cost of the bearings in the
gearbox. However, to date, no cost optimization study has been conducted to determine
the optimal gear ratio for two stage helical gearboxes with second stage double gear-sets,
taking bearing cost into account to date.

Fig. 1. Schema of a mechanical drive system: (1) Motor (2) Coupling (3) Gearbox (4) Chain drive
(5) Belt conveyor

In this paper, a cost optimization study to determine the optimal partial gear ratios
of a two-stage helical gearbox with double gear-sets in the second stage was introduced.
The effects of the main design parameters on the optimal gear ratio of the first stage u1
were evaluated. Furthermore, a regression model to find u1 was proposed.

2 Methodology
2.1 Calculation of Gearbox Cost
For a two-stage helical gearbox with second stage double gear-sets, the cost of the
gearbox Cgb can be determined by (Fig. 1):

Cgb = Cg + Cs + Cgh + Cb (1)

where, Cg , Cs , Cb , and Cgh are the cost of all the gears, shafts, bearings of the gearbox,
and the cost of gearbox housing. These cost elements are calculated by:

Cg = Cg,m · mg (2)
 Cost Optimization Study for Two-Stage Helical Gearbox 3

Fig. 2. Calculated schema

Cs = Cs,m · ms (3)

Cgh = Cgh,m · mgh (4)

n
Cb = C (5)
i=1 b,i

In which, Cg,m , Cs,m , Cgh,m are the cost per kilogram of gears, shafts, and gearbox
housing (USD/kg). Cb,i is the cost of a pair of bearings (USD/pair). mg , ms , and mgh
are the mass of the gears, shafts, and the mass of the gearbox housing (kg), respectively.
These mass elements are determined in the following subsections.

2.2 Calculation of Gear Mass

The mass of all gears of the gearbox can be calculated by:
mg = mg1 + mg2 (6)
where mg1 and mg2 are the mass of gears of stages 1 and 2, which are found by:
 
π · e1 · dw11
2 ·b
w1 π · e2 · dw21
2 ·b
w1
mg1 = ρg · + (7)
4 4
4 N. H. Linh et al.

 
π · e1 · dw21
2 ·b
w2 π · e2 · dw22
2 ·b
w2
mg2 = ρg · + (8)
4 4

where, ρg = 7.82 (kg/dm3 ) is the density of steel gear material; e1 and e2 are the volume
coefficients of stages 1 and 2, respectively [22]; e1 = 1 and e2 = 0.6; bw1 and bw2 are
the gear widths of stages 1 and 2.

bw1 = Xba1 · aw1 (9)

bw2 = Xba2 · aw2 (10)

In the above Equations, Xba1 and Xba2 are the wheel face width coefficients of stages
1 and 2. aw1 and aw2 are the center distances of stages 1 and 2 which are detemined by
[22]:

aw1 = ka · (u1 + 1) · 3 T11 · kH β /(AS21 · u1 · Xba ) (11)

aw2 = ka · (u2 + 1) · 3
T12 · kH β2 /(AS22 · u2 · Xba2 ) (12)

where, T11 and T12 are the torque on the pinions of stage 1 and stage 2:
 
T11 = Tout / ug · η2hg · ηb3 (13)
 
T12 = Tout / 2 · u2 · ηhg · ηbe
2
(14)

In which, Tout is the system output torque (N.mm); ηhg = 0.95÷0.97 is the efficiency
of a helical gear set [22]; ηb = 0.99÷0.995 is the efficiency of a rolling bearing pair
[22].

2.3 Calculation of Shaft Mass

The mass of all shafts of the gearbox can be calculated by:

ms = ms1 + ms2 + ms3 (15)

wherein, ms1 , ms2 and ms3 are the mass of shafts 1, 2 and 3 of the gearbox which are
found by:

ms1 = ρs · π · ds1
2
· ls1 /4 (16)

ms2 = ρs · π · ds2
2
· ls2 /4 (17)

ms1 = ρs · π · ds2
2
· ls2 /4 (18)
 Cost Optimization Study for Two-Stage Helical Gearbox 5

wherein, ds1 , ds2 , and ds3 are the diameters of shaft 1, 2, and 3 (mm); ls1 , ls2 , and ls3 are
the lengths of shaft 1, 2 and 3 of the gearbox (mm). These diameters and lengths can be
found by:

ds1 = [T11 /(0.2 · [τ ])]1/3 (19)

ds2 = [T12 /(0.2 · [τ ])]1/3 (20)

ds3 = [Tout /(0.2 · [τ ])]1/3 (21)

In the above Equations, [τ] = 15÷20 is the allowable shear stress (MPa) [22].

2.4 Calculation of Gearbox Housing Mass

The mass of the gearbox housing of the gearbox is determined by:

mgh = ρgh · Vgh (22)

In which, ρgh = 7.2 (kg/dm3 ) is the density of the gray cast iron (the gearbox housing
material); Vgh is the volume of the gearbox housing (m3 ) which is determined by:

Vgh = 2 · (Vb + VA1 + VA2 ) (23)

where, Vb , VA1 , VA2 are the volume of the bottom, the sides A1 , and A2 of the gearbox
housing, respectively. These volume elements are found by:

Vb = L · B · SG (24)

VA1 = L · (H − 2 · SG ) · SG (25)

VA2 = (H − 2 · SG ) · B1 · SG (26)

In the above Equations, L, H, B1 and SG can be determined by:

dw21 dw12
L = dw11 + + + dw22 + 20 (27)
2 2

H = max(dw21 , dw22 ) + 8.5 · SG (28)

B1 = 2 · bw1 + bw2 + 5 · SG (29)

B = B1 + 2 · SG = 2 · bw1 + bw2 + 7 · SG (30)

SG = 0.005 · L + 4.5 [23] (31)

6 N. H. Linh et al.

2.5 Calculation of Bearing Cost

The cost of all bearings of the gearbox (USD) can be found by [21]:

Cb,i = kcb · Cb,mi (32)

where, Cb,mi is the cost of a medium-sized deep groove ball bearing on the i-th shaft
which has the inner diameter of 5Ne,i ; Cb,mi can be found by [21]:

Cb = 0.2707 · Ne,i
2
− 1.2566 · Ne,i + 3.3056 (33)

In which, kcb is the bearing selling coefficient (kcb < 1).

2.6 Optimization Problem

The optimization problem can be described based on the prior analysis as follows:

Minimize Cgb (34)

With the following constraints:

1 ≤ u1 ≤ 9; 1 ≤ u2 ≤ 9 (35)

3 Experimental Work
To investigate the impact of the input factors on the partial gear ratio u1 for getting
minimum gearbox cost, a simulation experiment was conducted. Table 1 shows the
input parameters and their levels. Besides, the 1/8 fraction design was chosen to reduce
the number of experimental runs while still assessing the effect of the input factors.
Therefore, a number of 210–3 = 128 test runs were performed. Also, the Minitab R18
software was used for experimental design and the result analysis. The experimental
plan and the results of optimal gear ratio u1 are shown in Table 2.

4 Results and Discussions

The effect of the input factors on the optimum gear ratio of the first stage u1 is depicted
in Fig. 2. According to this graph, u1 is proportional to ug , Tout , AS1 , Cg , and inversely
proportional to AS2 , CS , and Kcb . Furthermore, ug is the most powerful factor on u1 ,
followed by AS2 „ Cg , Cs , Tout , AS1 , and Kcb . Moreover, Xba1 and Cgh have no effect
on u1 .
The Pareto chart can also be used to determine the influence of the input variables and
their interactions on u1 (Fig. 3). The length of the blue column shows the magnitude of
the variables’ impact. With a level of significance of 0.05, the factors with the magnitudes
greater than the red reference line have a major impact on u1 . Furthermore, A (ug ), F
(AS2 ), H (Cg ), B (Tout ), J (Cs ), E (AS1 ), K (kcb ), G (Cgh ), and the interactions AF, AE,
AH, AJ, AB, AK, BJ, BK, BH, and BE all significantly impact u1 .
 Cost Optimization Study for Two-Stage Helical Gearbox 7

Table 1. Input factors and their levels

No. Parameters Code Unit Level

Low High
1 Total gearbox ratio ug – 6 30
2 Output torque Tout Nm 5000 50000
3 Coefficient of the face width of the first gear stage Xba1 – 0.35 0.4
4 Coefficient of wheel face width of the second gear stage Xba2 – 0.35 0.4
5 Allowable contact stress of stage 1 AS1 MPa 350 420
6 Allowable contact stress of stage 2 AS2 MPa 350 420
7 Cost of gearbox housing Cgh USD/kg 1 5
8 Cost of a kilogram of gears Cg USD/kg 5 9
9 Cost of a kilogram of shafts Cs USD/kg 1.5 5
10 Bearing selling price coefficient kcb – 0.7 1

Table 2. Experimental plan and output results (u1 )

No. ug Tout Xba1 Xba2 AS1 AS2 Cgh Cg Cs kcb u1

1 6 5000 0.35 0.4 420 420 5 5 5 1 1.18
2 30 50000 0.4 0.35 350 350 5 9 5 1 5.88
3 6 50000 0.35 0.35 350 420 5 9 1.5 0.7 1.66
4 30 5000 0.35 0.35 350 420 5 9 5 1 4.89
5 6 50000 0.4 0.35 420 350 5 5 1.5 0.7 1.74
6 30 5000 0.4 0.35 420 420 1 9 5 0.7 5.55
…
127 30 50000 0.35 0.35 420 420 1 9 5 1 5.71
128 30 5000 0.35 0.4 420 350 5 9 5 1 5.9

The normal plot of normalized effects (Fig. 4) is used to determine the direction,
influence, and interaction of the input variables. On this diagram, the variables denoted
by red squares have a major impact on u1 . The variables that deviate from the standard
line produce a greater impact. The variables to the right of the standard line, such as
H, B, E, and G, as well as the interactions AE, AH, AB, BK, BE, FJ, and EH, pose a
positive effect on u1 . Conversely, the variables to the left of the standard line, such as F,
J, and K, as well as their interactions AF, AJ, AK, BH, and BJ, have a negative effect on
u1 (Fig. 5).
After removing the parameters that have little or no impact on u1 , a regression
equation for defining u1 was proposed using Minitab 18 software. Table 3 shows the
estimated coefficients of the input variables and their interactions for u1 . From the
8 N. H. Linh et al.

Fig. 3. The effect of input factors on u1

Fig. 4. Pareto chart of standardized effects for u1

Table, it was noted that the R-sq and R-sq (adj) of the regression model are very high
(99.93% and 99.91%), whichindicates that the equation is well-matched with the data.
The proposed model to determine u1 is in the form as follows:

u1 = 1.659 + 0.20448ug − 0.000008Tout − 0.001812AS1 − 0.001481AS2

+ 0.01336Cgh − 0.0413Cg + 0.0075Cs − 0.3906kcb + 0.000000ug ∗ Tout
+ 0.000229ug ∗ AS1 − 0.000321ug ∗ AS2 + 0.003984ug ∗ Cg − 0.003676ug ∗ Cs
 Cost Optimization Study for Two-Stage Helical Gearbox 9

Fig. 5. Normal distribution chart of effect of input factors and interactions for u1

− 0.02283ug ∗ kcb + 0.000000Tout ∗ AS1 − 0.000000Tout ∗ Cg

− 0.000001Tout ∗ Cs + 0.000008Tout ∗ kcb + 0.000199AS1 ∗ Cg
− 0.000253AS1 ∗ Cs + 0.000263AS2 ∗ Cs (36)

The proposed model is validated by calculating the difference between the trial and
u1 using the residual evaluation distribution plot (Fig. 6). This figure shows that the
errors (blue dots) are distributed very close to the normal distribution (solid lines), and
that these errors are located very close to the normal distribution. This is also evident in
the frequency graph, where most errors are close to zero. The other two graphs depict
the relationship between the error and the corresponding regression model values, which
are both randomly distributed. It demonstrates that u1 is not significantly affected by the
survey parameters. Furthermore, the adjustment coefficients R-sq and Rsq-adj are both
greater than 99.8% (Table 3). This demonstrates that the proposed model is appropriate
for using.

5 Conclusions
The purpose of this paper is to introduce a cost optimization study of a two-stage helical
gearbox with second-stage double gear-sets for determining the optimal partial gear
ratios u1 . To accomplish this, a simulation experiment with ten input parameters was
designed and carried out. The impact of the main design factors and cost components
on the the optimal gear ratio of the first stage u1 was studied. It was discovered that the
gearbox ratio ug has the greatest effect on u1 . The influence of other factors and their
interactions on u1 was also assessed. Furthermore, a regression model for estimating
u1 was proposed. The R-sq values are all greater than 99.8%, which confirms that the
proposed model is appropriate for use.
10 N. H. Linh et al.

Table 3. Estimated coefficient of input fcators and their interaction for u1

Coded coefficient
Term Effect Coef SE Coef T-Value P-Value VIF
Constant 3.55562 0.00559 635.92 0.000
ug 4.14125 2.07063 0.00559 370.33 0.000 1.00
Tout 0.31781 0.15891 0.00559 28.42 0.000 1.00
AS1 0.24719 0.12359 0.00559 22.10 0.000 1.00
AS2 −0.44844 −0.22422 0.00559 −40.10 0.000 1.00
Cgh 0.05344 0.02672 0.00559 4.78 0.000 1.00
Cg 0.38063 0.19031 0.00559 34.04 0.000 1.00
Cs −0.26312 −0.13156 0.00559 −23.53 0.000 1.00
kcb −0.17781 −0.08891 0.00559 −15.90 0.000 1.00
ug*Tout 0.13844 0.06922 0.00559 12.38 0.000 1.00
ug*AS1 0.19219 0.09609 0.00559 17.19 0.000 1.00
ug*AS2 −0.26969 −0.13484 0.00559 −24.12 0.000 1.00
ug*Cg 0.19125 0.09562 0.00559 17.10 0.000 1.00
ug*Cs −0.15437 −0.07719 0.00559 −13.80 0.000 1.00
ug*kcb −0.08219 −0.04109 0.00559 −7.35 0.000 1.00
Tout* AS1 0.03750 0.01875 0.00559 3.35 0.001 1.00
Tout*Cg −0.03844 −0.01922 0.00559 −3.44 0.001 1.00
Tout*Cs −0.05844 −0.02922 0.00559 −5.23 0.000 1.00
Tout* kcb 0.05125 0.02563 0.00559 4.58 0.000 1.00
AS1*Cg 0.02781 0.01391 0.00559 2.49 0.014 1.00
AS1*Cs −0.03094 −0.01547 0.00559 −2.77 0.007 1.00
AS2*Cs 0.03219 0.01609 0.00559 2.88 0.005 1.00
Model Summary
s R-sq R-sq(adj) R-sq(pred)
0.0632581 99.93% 99.91% 99.98%
 Cost Optimization Study for Two-Stage Helical Gearbox 11

Fig. 6. The distribution graphs evaluate the residuals

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Höhn, B.-R., Michaelis, K., Hinterstoißer, M.: Optimization of gearbox efficiency. Goriva i
maziva: časopis za tribologiju, tehniku podmazivanja i primjenu tekućih i plinovitih goriva i
inžinjerstvo izgaranja 48(4), 441–461 (2009)
2. Parmar, A., Ramkumar, P., Shankar, K.: Macro geometry multi-objective optimization of
planetary gearbox considering scuffing constraint. Mech. Mach. Theory 154, 104045 (2020)
3. Lei, Y., et al.: Research on vibration and noise reduction of electric bus gearbox based on
multi-objective optimization. Appl. Acoust. 158, 107037 (2020)
4. Chougale, A., et al.: Optimization of process parameters by hybrid Taguchi-Grey relational
analysis for thermal behaviours of lubricant oil of worm gearbox. Mater. Today: Proc. (2021)
5. Hongxia, P., Qingfeng, M., Xiuye, W.: Research on fault diagnosis of gearbox based on particle
swarm optimization algorithm. In: 2006 IEEE International Conference on Mechatronics.
IEEE (2006)
6. Romhil, I., Linke, H.: Gezielte Auslegung Von Zahnradgetriben mit Minimaler Masse auf der
Basis neuer Berechnungsverfahren. Konstruktion 44, 229–236 (1992)
7. Pi, V.N.: Optimum determination of partial transmission ratios of three-step helical gearboxes.
WSEAS Trans. Appl. Theor. Mech. 3(2), 43–52 (2008)
8. Cam, N.T.H., et al.: Calculating optimum gear ratios of mechanical drive systems using
two-stage helical gearbox with second-stage double gear sets and chain drive for minimum
gearbox length. In: Sattler, K.-U., Nguyen, D.C., Vu, N.P., Tien Long, B., Puta, H. (eds.)
ICERA 2019. LNNS, vol. 104, pp. 155–163. Springer, Cham (2020). https://doi.org/10.1007/
978-3-030-37497-6_18
12 N. H. Linh et al.

9. Hung, L.X., et al.: Calculation of optimum gear ratios of mechanical driven systems using
two-stage helical gearbox with first stage double gear sets and chain drive. In: International
Conference on Engineering Research and Applications. Springer (2019)
10. Van Cuong, N., Le Hong, K., Tran, T.H.: Splitting total gear ratio of two-stage helical reducer
with first-stage double gearsets for minimal reducer length. Int. J. Mech. Prod. Eng. Res. Dev.
(IJMPERD) 9(6), 595–608 (2019)
11. Cam, N.T.H., Pi, V.N., Hong, T.T.: A study on calculation of optimum gear ratios of a two-stage
helical gearbox with second stage double gear sets. a a 500, 1 (2019)
12. Tuan, N.K., et al.: Optimum calculation of partial transmission ratios of mechanical driven
systems using a V-belt and a three-step bevel helical gearbox. In: International Conference
on Engineering Research and Applications. Springer (2018)
13. Tuan, N.K., Pi, V.N.: Optimum calculation of partial transmission ratios of mechanically
driven systems using a V-belt and a helical gearbox with first-step double gear sets. Int. J.
Mech. Eng. Robot. Res. 8(2) (2019)
14. Tuan, N.K., et al.: Determining optimal gear ratios of a two-stage helical reducer for getting
minimal acreage of cross section. In: MATEC Web of Conferences. EDP Sciences (2018)
15. Pi, V.N., Tuan, N.K., Hung, L.X., Tung, L.A.: A new study on determination of optimum
gear ratios of a two-stage helical gearbox. In: Awang, M., Emamian, S.S., Yusof, F. (eds.)
Advances in Material Sciences and Engineering. LNME, pp. 107–115. Springer, Singapore
(2020). https://doi.org/10.1007/978-981-13-8297-0_14
16. Pi, V.N., Tuan, N.K., Hung, L.X., Dung, N.T.Q., Hien, B.T.: Determining optimum gear
ratios of a worm—helical gearbox for minimum acreage of the cross section. In: Awang,
M., Emamian, S.S., Yusof, F. (eds.) Advances in Material Sciences and Engineering. LNME,
pp. 89–96. Springer, Singapore (2020). https://doi.org/10.1007/978-981-13-8297-0_12
17. Thao, T.T.P., et al.: Determining optimum gear ratios of mechanical driven systems using
three stage bevel helical gearbox and chain drive. In: International Conference on Engineering
Research and Applications. Springer (2019)
18. Pi, V.N., et al.: Calculating optimum gear ratios of a two-stage helical reducer with first
stage double gear sets. In: IOP Conference Series: Materials Science and Engineering. IOP
Publishing (2019)
19. Pi, V.N., et al. :Determination of optimum gear ratios of a two-stage helical gearbox with
second stage double gear sets. In: IOP Conference Series: Materials Science and Engineering.
IOP Publishing (2019)
20. Vu, N.-P., et al.: The influence of main design parameters on the overall cost of a gearbox.
Appl. Sci. 10(7), 2365 (2020)
21. Quang, N.H., et al.: Optimizing the partial gear ratios of the two-stage worm gearbox for
minimizing total gearbox cost. East.-Eur. J. Enterp. Technol. 1(1), 115 (2022)
22. Chat, T., Van Uyen, L.: Design and Calculation of Mechanical Transmissions Systems, vol.
1. Educational Republishing House, Hanoi (2007)
23. Römhild, I., Linke, H.: Gezielte Auslegung Von Zahnradgetrieben mit minimaler Masse auf
der Basis neuer Berechnungsverfahren. Konstruktion (1981) 44(7–8), 229–236 (1992)
 A Design and Implement of Fuzzy Controller
for Taking-off and Landing for Unmanned
Aerial Vehicles

Dong L. T. Tran1 , Hai T. Do2(B) , Hoang T. Tran1 , Thang Hoang1 ,

and Minh T. Nguyen2
1 Center of Electrical Engineering, Duy Tan University, Da Nang 550000, Vietnam
2 Thai Nguyen University of Technology, Thai Nguyen 240000, Vietnam

dotrunghai@tnut.edu.vn

Abstract. Fuzzy controller applications have become increasingly common in

our lives, ranging from consumer products such as cameras, washing machines,
and cars to industrial products including robots, subway trains, medical instru-
ments, and home appliances. In our research, we apply and demonstrate the fuzzy
controller on Unmanned Aerial vehicles (UAVs) for takeoff and landing purposes.
Different from conventional controllers, the controller of fuzzy is not based on a
mathematical model of the system. It is a form of intelligent control by experi-
ment to adjust parameter control to be good for the system. By implement of UAV
system, we enable to determine the rule of fuzzy control. Addition, we will show
the performance of UAV by the algorithm of the fuzzy controller on LabVIEW.
The outcomes presented highlight positive and useful aspects.

Keywords: Unmanned Aerial Vehicle (UAV) · Fuzzy controller · Conventional

controllers · Takeoff · Landing

1 Introduction

Fuzzy logic applications have become more and more popular in our life. The applica-
tion ranges from consumer products such as camera surveillance, washing machines,
cars to industrial products including robots, underground trains, and medical instrumen-
tation [1–3]. Unlike the conventional controller, Fuzzy logic controller design is not
based on the mathematical model of system. Instead, it is an automatic controller that
controls an object in accordance with desire behavior. For a complex system whose math-
ematical model is difficult to define or undefined, the traditional controller is impotent.
Fortunately, fuzzy logic controllers are very useful in that case [4, 5].
Quadcopter is an aerial vehicle plat form which has become very popular among
researchers in the recent past due to the advantages it offers over conventional helicopters
[6, 7]. Quadcopter is very simple and unique, but it is inherently unstable from aerody-
namics point of view. In recent past researchers have proposed many control schemes
for quadcopter likely P, I, PID controllers [8–11]. Different from previous work, in this

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 13–22, 2023.
https://doi.org/10.1007/978-3-031-22200-9_2
14 D. L. T. Tran et al.

study, we present fuzzy logic controller for quadcopter to evaluation performance of

algorithm operation.
The rest of this paper is organized as follows. Section 2 provide a description of the
system with mathematical models that lead to be solved. Section 3 address the design of
input and output fuzzy controllers. Some results and discussion are provided in Sect. 4.
Finally, conclusions and future work are given in Sect. 5.

2 System Description
In the most common design for quadcopter, the two sets of rotors (1, 3) and (2, 4) rotate
in opposite directions as shown in Fig. 1. By changing the rotor speeds, l if t forces
can be changed and motion can be created in desired direction. Changing the speeds of
all four rotors generate vertical motion. Increasing or decreasing the speeds of rotor 2
and 4 conversely produces roll rotation coupled with lateral motion. Pitch rotation and
the corresponding lateral motion result from changing the speed of rotors 1 and 3. Yaw
rotation results from the difference in the counter torque between the pairs of rotors [12,
13].

Fig. 1. Quadcopter motion, the arrow width is proportional to propeller rotational speed

The inertial frame is defined with respect to the ground, with gravity pointing in the
negative Z-axis. (Fig. 2a) the body frame is defined by the orientation of the quadcopter,
with the rotor pointing in the positive Z-axis and the arm extensions pointing in the
positive/negative X and Y axes. (Fig. 2a) All metrics with respect to either the inertial
or body frame are implied in braces [12, 14–16].
Thrust is the force that is orthogonal to the propeller. It is generated by rotor spin at
a certain velocity (Eq. 1). Moreover, it accelerates the body in the direction of its force.
In all the equations henceforth ‘i’ refers to the particular rotor in Fig. 2b.
− 
→
 Tl  = ρAυi2 (1)

In Eq. 1, T is determined using the parameters v (velocity), (air-density) and A (cross-

sectional area) of the propeller. A real-time air-density reading is necessary because the
density of air is correlated to external environmental factors that are not always constant.
The CCW direction contributes negative net thrust {Z-axis Body frame} on the quad-
copter body, thereby enabling translational motion about the negative Z-axis {Inertial
 A Design and Implement of Fuzzy Controller 15

Fig. 2. a) Body and Inertial frame; b) hovering

frame}. All the rotors are spinning in the same direction and velocity [17]. Equation 2
gives the net thrust on the body as
4 2
net T = ρA [vi ] (2)
i=1

Equation 3 shows the net thrust from the four rotors. The arrows indicate the direction
of thrust in the Z-axis {Body frame}.
2 2 4 2
net T = ρA [vi ] − ρA [vi ] = 0 (3)
↑i=1 ↓i=3

From the law of conservation of energy, we obtain Eq. 4, which is the velocity
required for the quadcopter to climb from ho to any desired altitude of hf regardless of
its angular orientation.
 
v2 = 4g hf − ho (4)

Equation 5 is the total thrust that is generated by the rotors in order to keep the
quadcopter airborne at a certain desired altitude of hf given its total mass (m) and
angular orientation:
−   
→ ρA4g hf − ho + mg
T  = (5)
cos(θ )cos(ϕ)
The magnitude of the total thrust from Eq. 6 is used to obtain the magnitude of
the component vectors Tx, Ty, and Tz using Eqs. 6, 7 and 8. The vector component
magnitudes are with respect to the inertial frame.
  
1
|Tx | = −|T |2 cos(θ )2 1 − (6)
cos(θ )2
 
Ty  = |T |cos(θ )sin(ϕ) (7)

|Tz | = |T |cos(θ)cos(ϕ) (8)

A specific altitude can be maintained by total thrust from the rotors such that total
Tz {Inertial frame} in Eq. 8 is a constant.
16 D. L. T. Tran et al.

Vertical speed controller: The simplified transfer function that describes the vertical
position [17]:
−0.0424
Gw (s) = (9)
s(s + 10)

3 Design the Input and Output of Fuzzy Controller

3.1 Ultrasonic Sensor

Ultrasonic ranging module HC-SR04 provides 2 cm–400 cm non-contact measurement

function, the ranging accuracy can reach to 3 mm. The modules include ultrasonic
transmitters, receiver, and control circuit.

Fig. 3. a) Ultrasonic sensor operating; b) ultrasonic sensor position

According to experiments on our system, the distance information from the sensor is
unstable. Perhaps this is caused by oscillating of pendulum create an angle of deflection
like the left-top case in the above figures (Fig. 3).

3.2 Design Fuzzy and PID Controller for UAV

Design Fuzzy controller
The Table 1 and Table 2 show the parameter of input membership functions. Distance is
input variable to measure the distance from UAV to landscape. Another input variable
is speed which describe the velocity of UAV in z-axis direction. For output control, the
analog signal is selected to control the level of UAV. In Table 3, it presents the range
value of output membership functions. Figure 4a, 4b and Fig. 5 are used to visually
represent input and output membership functions.
Output membership functions
 A Design and Implement of Fuzzy Controller 17

Table 1. The input of variable distance (cm)

Membership function Shape Points

NVL Trapezoid −70; −70; −60; −50
NL Trapezoid −60; −50; −50; −40
NS Trapezoid −50; −30; −20; −20
PS Trapezoid 20; 20; 30; 50
PL Trapezoid 40; 50; 50; 60
PVL Trapezoid 50; 60; 70; 70

Table 2. The input of variable speed (cm/s)

Membership function Shape Points

VERY SMALL Trapezoid −8; −8; −6; −4
SMALL Trapezoid −6; −4; −4; −2
NEAR−ZERO Trapezoid −4; −2; −2; 0
ZERO Trapezoid −2; 0; 0; 2
NEAR+ZERO Trapezoid 0; 2; 2; 4
BIG Trapezoid 2; 4; 4; 6
VERY BIG Trapezoid 4; 6; 8; 8

Rules of fuzzy controller

See Table 4.
Input/output relationship
After finishing the designed rule of fuzzy controller, we got the relationship between
input and output variable of system in Fig. 6.
Design PID controller
The parameters of PID controller are designed based on the mathematical model of
UAV [18, 19]. The Fig. 7 shows the block diagram of algorithm PID controller. The
three parameters on the PID control are respectively proportional as, Gain (Kc = 0.012);
Integral time (Ti = 0.015); Derivative time (Td = 0.02).

4 Results and Discussion

After designing the controller, we conduct experiments to observe the controller prop-
erties. Figure 8 illustrates an experimental model of a UAV connected to a central con-
troller to execute control algorithms. The results of performance controller are indicated
in Fig. 9a and Fig. 9b (Table 5).
18 D. L. T. Tran et al.

Fig. 4. The input of membership functions for fuzzy controller: a) input variable of distance; b)
input variable of speed

Table 3. The output signal control motor (V)

Membership function Shape Points

Very small Trapezoid 1; 1; 1.5; 2
Small Trapezoid 1.5; 2; 2; 2.5
Big Trapezoid 2; 2.5; 2.5; 3
Very big Trapezoid 2.5; 3; 4; 4

Fig. 5. The output of membership functions for fuzzy controller

 A Design and Implement of Fuzzy Controller 19

Table 4. Rules of fuzzy controller

Distance Delta
Very Small Near-zero Zero Near + Big Very big
small zero
NVL Very Very Very small Small Small Small Very
small small small
NL Very Very Very small Very Very small Very Very
small small small small small
NS Small Small Small Small Small Small Very
small
PS Big Big Big Big Big Small Small
PL Very big Very big Big Big Very small Small Very
small
PVL Very big Very big Big Big Big Big Small

Fig. 6. Input/output relationship of membership functions for fuzzy controller

Fig. 7. The control model using PID controller

Discussion: As mentioned with the models, the design and also the results, we apply the
algorithm control to send signals to the remote controller by RF waves, thus ensuring
the sensitivity and response of the system. Application of fuzzy controller allows to
control objects with complex kinematics, difficult to model modeling. However, in the
control process, there is still noise type because the kinematic model of the quadcopter
20 D. L. T. Tran et al.

Fig. 8. The diagram of experiment controller for UAV

Fig. 9. a Diagram of the control process by PID method; a diagram of the control process by
Fuzzy control method

Table 5. Evaluation of PID and Fuzzy control on UAV

Type Rise time Evaluation

Fuzzy 0.05 or more Unstable
Fuzzy-PID 0.01 Stable

is quite complicated. The response time of system is small about 0.01 s with Fuzzy-PID
controller. We had proposed the method of using Fuzzy controller to improve control
quality for takeoff and landing of the UAVs.

5 Conclusions and Future Work

This paper aims to apply Fuzzy controller to support UAVs in taking-off and landing
purposes. In order to ensure the system’s sensitivity and reactivity, we use the algorithm
control to transmit signals through RF waves to the remote controller. The use of fuzzy
controllers enables the control of objects with complicated kinematics that are challeng-
ing to model. Mathematical models are provided to support the problems to be solved.
A design is completely addressed to support the problems of UAVs. The results show
promising points for both taking-off and landing. With a fuzzy controller, the system
responds in just 0.01 s on average. To enhance the control quality for an unmanned aerial
vehicle’s takeoff and landing, we had suggested utilizing a fuzzy controller.
 A Design and Implement of Fuzzy Controller 21

In our future work, following this approach, experimental and practical results should
be provided based on real UAVs that have enough capacities to programmed and coded.
In addition, some advanced communication technologies should be considered to support
the UAVs for better performance.

Acknowledgements. The author would like to thank Thai Nguyen University of Technology,
Viet Nam for the support.

References
1. Nguyen, M.T., Boveiri, H.R.: Energy-efficient sensing in robotic networks. Elsevier Mea-
surement 158 (2020)
2. Do, H.T., Truong, L.H., Nguyen, M.T., Chien, C.-F., Tran, H.T., Hua, H.T., Nguyen, C.V.,
Nguyen, H.T.T., Nguyen, N.T.T.: Energy-efficient unmanned aerial vehicle (UAV) surveil-
lance utilizing artificial intelligence (AI). Wirel. Commun. Mob. Comput. 2021, Article
ID 8615367, 11 (2021)
3. Nguyen, M.T., La, H.M., Teague, K.A.: Collaborative and compressed mobile sensing for data
collection in distributed robotic networks. IEEE Trans. Control Netw. Syst. 5(4), 1729–1740
(2018)
4. Mamdani, E.H., Assilian, S.: An experiment in linguistic synthesis with a fuzzy logic
controller. Int. J. Man. Mach. Stud. 7(1), 1–13 (1975). https://doi.org/10.1016/S0020-737
3(75)80002-2
5. Palm, R.: Scaling of fuzzy controllers using the cross-correlation. IEEE Trans. Fuzzy Syst.
3(1), 116–123 (1995). https://doi.org/10.1109/91.366559
6. Nguyen, M.T., Truong, L.H., Le. T.T.H.: Video surveillance processing algorithms utilizing
artificial intelligent (AI) for unmanned autonomous vehicles (UAVs). MethodsX 8, 101472
(2021)
7. Do, H.T., Hua, H.T., Nguyen, H.T.T., Nguyen, M.T., Tran, H.T.: Cooperative tracking
framework for multiple unmanned aerial vehicles (UAVs). In: International Conference on
Engineering Research and Applications, pp. 276–285. Springer, Cham (2021)
8. Satla, Z., Elajrami, M.: P, PI, PID controller designed for UAV quadrotors trajectory (2019)
9. Burggräf, P., Pérez Martínez, A.R., Roth, H., Wagner, J.: Quadrotors in factory applications:
design and implementation of the quadrotor’s P-PID cascade control system. SN Appl. Sci.
1(7), 1–17 (2019). https://doi.org/10.1007/s42452-019-0698-7
10. Tran, H.T., Tran, D.L.T., Nguyen, V.Q., Do, H.T., Nguyen, M.T.: A novel framework of
modelling, control, and simulation for autonomous quadrotor UAVs utilizing Arduino mega.
Wirel. Commun. Mob. Comput. 2022 (2022)
11. Noshahri, H., Kharrati, H.: PID controller design for unmanned aerial vehicle using genetic
algorithm. In: IEEE International Symposium on Industrial Electronics, pp. 213–217 (2014).
https://doi.org/10.1109/ISIE.2014.6864613
12. Mian, A.A., Daobo, W.: Nonlinear flight control strategy for an underactuated quadrotor
aerial robot. In: 2008 IEEE International Conference on Networking, Sensing and Control,
Apr. 2008, pp. 938–942 (2008). https://doi.org/10.1109/ICNSC.2008.4525351
13. Salih, A.L., Moghavvemi, M., Mohamed, H.A.F., Gaeid, K.S.: Flight PID controller design
for a UAV quadrotor. Sci. Res. Essays 5(23), 3660–3667 (2010)
14. Li, J., Li, Y.: Dynamic analysis and PID control for a quadrotor. In: 2011 IEEE International
Conference on Mechatronics and Automation, ICMA 2011, pp. 573–578 (2011). https://doi.
org/10.1109/ICMA.2011.5985724
22 D. L. T. Tran et al.

15. Lopez, J.C.: Quadcopter Simulation (2019)

16. Falconi, R., Melchiorri, C.: Dynamic Model and Control of an Over-actuated Quadrotor UAV,
vol. 45, no. 22. IFAC (2012). https://doi.org/10.3182/20120905-3-HR-2030.00031
17. Argentim, L.M., Rezende, W.C., Santos, P.E., Aguiar, R.A.: PID, LQR and LQR-PID on a
quadcopter platform. In: 2013 International Conference on Informatics, Electronics & Vision,
ICIEV 2013 (2013). https://doi.org/10.1109/ICIEV.2013.6572698
18. Tran, H.T., Vo, T.C., Tran, D.L.T., Nguyen, Q.N.A., Ha, D.M., Pham, Q.N., Le, T.Q., Nguyen,
T.K., Do, H.T., Nguyen, M.T.: Extended Kalman Filter (EKF) based localization algorithms
for mobile robots utilizing vision and odometry. In: 2022 IEEE 21st Mediterranean Elec-
trotechnical Conference (MELECON), pp. 91–96 (2022). https://doi.org/10.1109/MELECO
N53508.2022.9843066
19. Thanh, V.C., Quan, N.N.A., Le Thang Dong, T., Hoang, T.T., Nguyen, M.T.: Fusion of
inertial and magnetic sensors for autonomous vehicle navigation and freight in distinctive
environment. In: Nguyen, D.C., Vu, N.P., Long, B.T., Puta, H., Sattler, K.-U. (eds.) ICERA
2021. LNNS, vol. 366, pp. 431–439. Springer, Cham (2022). https://doi.org/10.1007/978-3-
030-92574-1_45
 A Detailed Procedure of Squirrel-Cage
Three-Phase Induction Motor Parameter
Estimation Using Polynomial Regression

Son T. Nguyen1(B) , Stefan Goetz2 , Tu M. Pham1 , Anh Hoang1 , and Tuan V. Pham3
1 Hanoi University of Science and Technology, Hanoi, Vietnam
son.nguyenthanh@hust.edu.vn
2 Technical University of Kaiserslautern, Kaiserslautern, Germany
3 Vinh University of Technology Education, Vinh, Vietnam

Abstract. Induction motors are the most frequently used electrical motor in high-
performance drive systems. The vector and sensorless speed control schemes of
induction motors usually require exact values of motor parameters to obtain the
robust control. This paper proposes an effective offline estimation of the equiva-
lent circuit parameters of squirrel-cage three-phase induction motors using time-
varying stator voltage and current with different load torques. The relationship
between the input impedance and the slip rate can be expressed as a fractional
polynomial. Based on conditions for obtaining the minimum of the objective
functions, the motor electrical parameters can be conveniently estimated. The
mechanical parameters of the motor can be derived by using determined electrical
parameters and motor dynamic simulation using MATLAB/Simulink. The main
advantage of this method is that it could facilitate us to gain the optimal values
of the motor parameters without the need of any initial values. In addition, the
effectiveness of the proposed method can be validated by a low-cost experimental
system.

Keywords: Induction motor · Equivalent circuit · Parameter estimation

1 Introduction
Three-phase induction motors are widely used in various industrial applications because
they are manufactured with low expense and do not require the frequent maintenance.
The performance characteristics of an induction motor are usually determined from
its equivalent circuit [1]. However, manufacturers do not provide the equivalent circuit
parameters except several catalog data such as rated voltage, current, power factor,
efficiency, torque, and rotor speed. Nowadays, vector control can be seen as the most
popular control method for high-performance induction motor drives. In addition, the
performance of sensorless vector control induction motor drives strongly depend on
the accuracy of the motor parameters. These motor parameters include stator resistance,
rotor resistance, core loss resistance, stator leakage inductance, rotor leakage inductance,
and magnetizing inductance [2].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 23–33, 2023.
https://doi.org/10.1007/978-3-031-22200-9_3
24 S. T. Nguyen et al.

The well-known method of squirrel-cage induction motor parameter estimation has

been based on no-load and locked rotor tests. However, the locked rotor test is difficult
to be performed as it requires the full control of the voltage with appropriate instruments
to mechanically secure the rotor in the locked condition. Several studies have focused on
the single-cage model parameter estimation by using steady-state, variable frequency,
and transient test data [3, 4]. In general, motor parameter estimation methods can be
classified into online estimation methods [5, 6] and offline estimation methods [7, 8]. In
some circumstances, online estimations must include power converters and additional
controllers and then the estimation method is embedded in the control algorithm. Com-
pared to online estimation methods, offline estimation methods require less devices or do
not use any controllers. But the disadvantage of offline estimation methods is that they do
not consider the change of the stator and rotor resistances according to the temperature
variation.
This paper describes a detailed procedure of an offline estimation of the equivalent
circuit parameters of a laboratory squirrel-cage three-phase induction motor [9]. The
time-varying voltage, current, and rotor speed are used to compute the equivalent circuit
parameters. The input impedance and the slip rate of the motor can be expressed as a
polynomial. Based on the condition of minimizing an objective function for the input
impedance, the equivalent circuit parameters can be conveniently found. The paper is
organized as follows. Section 2 presents the formation of a polynomial of the input
impedance depending on the slip rate and the polynomial regression for obtaining the
values of coefficients in the polynomial. Next, these coefficients are used to compute the
equivalent circuit parameters. Section 3 describes an experimental setup to validate the
effectiveness of the proposed method. Finally, Sect. 4 is the conclusion of this research.

2 Materials and Methods

2.1 Motor Input Impedance
To estimate the equivalent circuit parameters of an induction motor, a reasonable model
of the motor firstly needs to be formed. Figure 1 represents a steady-state per-phase
induction motor equivalent circuit. The equivalent circuit does not take the core loss
into account. Therefore, the circuit only has five parameters including: Rs (the stator
resistance), Rr (the rotor resistance referred to the stator side), Xm (the magnetizing
reactance), Xs (the stator reactance), and Xr (the rotor reactance referred to the stator
side).
Where s is the slip rate. According to Fig. 1, the input impedance can be computed
as follows:
⎡   ⎤
jXm jXr + Rsr
Z = R + jX = Rs + jXs + ⎣  ⎦ (1)
jXm + jXr + Rsr

where Z is the input impedance, R is the input resistance, and X is the input reactance.
From Eq. (1), the input resistance R can be derived as follows:
Xm2 Rr s
R = Rs + (2)
(Xm + Xr )2 s2 + R2r
 A Detailed Procedure of Squirrel-Cage Three-Phase Induction Motor 25

Fig. 1. Steady-state per-phase induction motor equivalent circuit without the core loss resistance.

From Eq. (1), the input reactance is also determined by:

Xm2 (Xm + Xr )s2

X = Xs + Xm − (3)
(Xm + Xr )2 s2 + R2r
Equation (2) can be shortened as follows:

a2 s2 + a1 s + a0
R= (4)
b2 s2 + 1
where
Xm2 (Xm + Xr )2 (Xm + Xr )2
a0 = Rs ; a1 = ; a2 = Rs ; b2 = (5)
Rr R2r R2r
Equation (3) can be shortened as follows:

a4 s2 + a3
X = (6)
b2 s2 + 1
where
(Xs + Xm )(Xm + Xr )2 − Xm2 (Xm + Xr )
a3 = Xs + Xm ; a4 = (7)
R2r
Based on Eqs. (4) and (6), the input impedance can be expressed as follows:
 
a0 + a1 s + a2 s2 + j a3 + a4 s2
Z= (8)
1 + b2 s2
Obviously, the input impedance Z is a fractional polynomial. At a specific rotor
speed, the slip rate s can be easily determined. The a0 , a1 , coefficients, a4 , and b2 can
be computed based on a polynomial regression mentioned in the next section.

2.2 Polynomial Regression

The coefficients in Eq. (8) can be determined by using a polynomial regression. By
minimizing a defined objective function, a set of results can be obtained with estimated
values close to actual values.
26 S. T. Nguyen et al.

During the operation of the induction motor, the time-varying values of voltage,
current, and rotor speed can be measured by using appropriate sensors. Based on these
acquired data, a series of the time-varying input resistance can be calculated as follows:
Vn
Zn = (9)
In
where Vn is the n-th measured motor voltage, In is the n-th measured motor current, Zn
the n-th calculated input impedance, and N is the number of measured points.
The phase difference α between the motor voltage vector V = [V1 , V2 , ..., VN ] and
the motor current vector I = [I1 , I2 , ..., IN ] is determined as follows:

cos−1 (V .I )
α= (10)
V I 
After the phase difference is computed, the n-the input resistance (Rn ) and the n-th
input reactance (Xn ) are determined as follows:

Rn = Zn cos(α) (11)

Xn = Zn sin(α) (12)

The difference between the n-th measured data value and the n-th polynomial value
can be defined as follows:
 
a0 + a1 s + a2 s2 + j a3 + a4 s2
En = Rn + jXn − (13)
1 + b2 s2
Re-arranging (13) gives:

  
  
Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 + j Xn 1 + b2 sn2 − a3 + a4 sn2
En = (14)
1 + b2 sn2

To obtain the minimization of polynomial (14), an objective function SE is firstly

defined as follows:
N
    2    2
SE = Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 + Xn 1 + b2 sn2 − a3 + a4 sn2
n=1
(15)

The minimum of the function SE can be obtained if all its partial derivatives are equal
to zero.
The partial derivative of SE with respect to b2 has the following form:

∂S   
N    
=2 Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 Rn sn2 +
∂b2
n=1
(16)
 
N     
+2 Xn 1 + b2 sn2 − a3 + a4 sn2 Xn sn2
n=1
 A Detailed Procedure of Squirrel-Cage Three-Phase Induction Motor 27

∂SE
When ∂b2 = 0, we have:

N  
    
Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 Rn sn2 +
n=1
N  
    
+ Xn 1 + b2 sn2 − a3 + a4 sn2 Xn sn2 = 0 (17)
n=1

Re-arranging (17) gives:


N          
−b2 R2n sn4 + Xn2 sn4 + a0 Rn sn2 + a1 Rn sn3 + a2 Rn sn4 + a3 Xn sn2 + a4 Xn sn4
n=1

N
= R2n sn2 + Xn2 sn2 (18)
n=1

The partial derivative of SE with respect to a0 has the following form:

∂SE   N   
=2 Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 (−1) (19)
∂a0
n=1

∂SE
When ∂a0 = 0, we have:


N   
n
−b2 Rn sn2 + a0 + a1 sn + a2 sn2 = Rn (20)
i=1 i=1

The partial derivative of SE with respect to a1 has the following form:

∂SE   N   
=2 Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 (−sn ) (21)
∂a1
n=1

∂SE
When ∂a1 = 0, we have:


N   
n
−b2 Rn sn3 + a0 sn + a1 sn2 + a2 sn3 = Rn sn (22)
i=1 i=1

The partial derivative of SE with respect to a2 has the following form:

∂SE   N    
=2 Rn 1 + b2 sn2 − a0 + a1 sn + a2 sn2 −sn2 (23)
∂a2
n=1

∂SE
When ∂a2 = 0, we have:


N   
n
−b2 Rn sn4 + a0 sn2 + a1 sn3 + a2 sn4 = Rn sn2 (24)
i=1 i=1
28 S. T. Nguyen et al.

The partial derivative of SE with respect to a3 has the following form:

∂SE   N   
=2 Xn 1 + b2 sn2 − a3 + a4 sn2 (−1) (25)
∂a3
n=1

∂SE
When ∂a3 = 0, we have:

N
    
n
−b2 Xn sn2 + a3 + a4 sn2 = Xn (26)
i=1 i=1

The partial derivative of SE with respect to a4 has the following form:

∂SE   N    
=2 Xn 1 + b2 sn2 − a3 + a4 sn2 −sn2 (27)
∂a4
n=1

∂SE
When ∂a4 = 0, we have:

N
    n
−b2 Xn sn4 + a3 sn2 + a4 sn4 = Xn sn2 (28)
i=1 i=1

From Eqs. (18), (20), (22), (24), (26) and (28), we can derive a matrix equation as
follows:

A.B = C (29)

where
⎡ ⎤
N 
 N  N  N  N  N 
− R2 s4 + X 2 s4 R s 2 R s 3 R s 4 X s 2 X s 4
⎢ n n n n n n n n n n n n n n ⎥
⎢ n=1 n=1 n=1 n=1 n=1 n=1 ⎥
⎢ N N N ⎥
⎢ ⎥
⎢ − Rn sn2 N sn sn2 0 0 ⎥
⎢ ⎥
⎢ n=1 n=1 n=1 ⎥
⎢ N N N N ⎥
⎢ − R s 3 s s 2 s 3 0 0 ⎥
⎢ n n n n n ⎥
⎢
A=⎢ n=1 n=1 n=1 n=1 ⎥
N N N N ⎥
⎢ − 4 2 3 4 ⎥
⎢ Rn sn sn sn sn 0 0 ⎥
⎢ n=1 n=1 n=1 n=1 ⎥
⎢   ⎥
⎢ N N ⎥
⎢ − Xn sn2 0 0 0 N sn2 ⎥
⎢ ⎥
⎢ n=1 n=1 ⎥
⎣ N N N ⎦
− Xn sn4 0 0 0 sn2 sn4
n=1 n=1 n=1
(30)

T
B = b2 a0 a1 a2 a3 a4 (31)

 N

T
 
N 
N 
N 
N 
N
C= R2n sn2 + Xn2 sn2 Rn Rn sn Rn sn2 Xn Xn sn2 (32)
n=1 i=1 i=1 i=1 i=1 i=1
 A Detailed Procedure of Squirrel-Cage Three-Phase Induction Motor 29

Vector B can be found as follows:

B = A−1 C (33)

The distribution of the reactance is expressed as a ratio as follows:

Xm + Xr
η= = 0.95 ÷ 1.05 (34)
Xm + Xs
As the vector B can be found using Eq. (33), the coefficients of the polynomial can
be used to find the parameters of the motor equivalent circuit as follows:

a32  
Rs = a0 ; Rr = ; Xm = a1 .Rr ; Xs = a3 − Xm ; Xr = b2 .R2r − Xm (35)
b2

2.3 Mechanical Parameter Estimation

Mechanical parameters of the induction motor consist of the inertia moment of the rotor
J and the friction factor D. These parameters can be determined if the output torque and
speed can be measured. Based on Newton’s equation, we have:

dω
T − Dω = J (36)
dt
In discrete-time, Eq. (36) can be expressed as follows:

Tn − Dωn = J (ωn − ωn−1 ) (37)

To estimate the inertia moment and the friction coefficient, an objective function
needs to be defined as follows:


N

2
ST = Tn − J (ωn − ωn−1 ) − Dωn (38)
n=1

When the objective function reaches the minimum value, the gradients of Eq. (38)
are equal to zero. Thus, J and D can be calculated as follows:
⎡ ⎤−1 ⎡ N ⎤

N 
N 
  (ωn − ωn−1 )2
ω (ω − ω ) T (ω − ω )
J ⎢ n=1 n n n−1 ⎥ ⎢ n=1 n n n−1 ⎥
=⎢
⎣
n=1 ⎥ ⎢
⎦ ⎣
⎥ (39)
⎦
D 
N N 2
−1 
N
ωn (ωn − ωn−1 ) ωn Tn ωn
n=1 n=0 n=1
30 S. T. Nguyen et al.

3 Experimental Validation
3.1 Electrical and Mechanical Parameter Identification
The experimental system is shown in Fig. 2. The system is formed by the following
elements:

• A 175 W Lab-Volt laboratory squirrel-cage three-phase induction motor

• A Lab-Volt three-phase AC source
• A Lab-Volt dynamometer working as an adjustable load torque
• Three 600 V/12 A Triacs used to switch the motor on/off
• Voltage and current transducers
• A National Instruments USB-6009 data acquisition device with the sampling rate of
10 k/second
• A PC/laptop

Table 1 shows the measured voltages, currents, and rotor speeds of 175 W three-phase
induction motor used to obtain the motor parameters. Table 2 indicates the estimated
parameters of the induction motor.

Fig. 2. The experimental system.

3.2 Comparison Between the Actual and Simulated Start-up Speeds

(A) Case 1: The load torque is 0.75 (N.m)

When the load torque is adjusted to be 0.75 (N.m), the actual and simulated start-up rotor
speed curves coincide each other as shown in Fig. 3. According to Fig. 4, the shapes of
the actual and simulated stator current waveforms are also identical. This consolidates
the high similarity between the real and simulated parameters of the motor.
 A Detailed Procedure of Squirrel-Cage Three-Phase Induction Motor 31

Table 1. Measured voltages, currents, and rotor speeds of the 175 W three-phase induction motor.

Load torque (N.m) Voltage Current Phase angle Speed

(V) (A) (Degree) (rpm)
0 220 0.317 70 1470
0.15 220 0.337 65 1456
0.3 220 0.361 59 1443
0.45 220 0.389 55 1435
0.6 220 0.419 50 1417
0.75 220 0.453 47 1404
0.9 220 0.491 44 1387
1.05 220 0.530 41 1370
1.2 220 0.572 39 1353
1.35 220 0.617 37 1331
1.5 220 0.665 36 1306

Table 2. Estimated parameters of the 175 W three-phase induction motor.

 
Rs () Rr () Ls (H) Lr (H) Lm (H) J kg.m2 D (N.m.s)

49.0034 45.9152 0.1135 0.1135 2.1141 0.0056 0.0021

Actual speed
1600
Simulated speed
1400

1200
Rotor speed (rpm)

1000

800

600

400

200

0
0 0.2 0.4 0.6 0.8 1
Time (s)

Fig. 3. The actual and simulated start-up rotor speeds (load torque is 0.75 N.m).

(B) Case 2: The load torque is 1.5 (N.m)

As the load torque is increased up to 1.5 (N.m), actual and simulated start-up rotor speed
curves coincide each other as depicted in Fig. 5. The shapes of actual and simulated stator
current waveforms are identical as shown in Fig. 6. This also means that the similarity
between the actual and simulated start-up rotor speeds is very high.
32 S. T. Nguyen et al.

Actual current
3
Simulated current

Current (A)
0

-1

-2

-3
0 0.2 0.4 0.6 0.8 1
Time (s)

Fig. 4. The actual and simulated start-up stator current waveforms (load torque is 0.75 N.m).

1600
Actual speed
1400 Simulated speed

1200
Rotor speed (rpm)

1000

800

600

400

200

0
0 0.2 0.4 0.6 0.8 1
Time (s)

Fig. 5. The actual and simulated start-up rotor speeds (load torque is 1.5 N.m).

Actual current
3
Simulated current

1
Current (A)

-1

-2

-3
0 0.2 0.4 0.6 0.8 1
Time (s)

Fig. 6. The actual and simulated start-up stator current waveforms (load torque is 1.5 N.m).

4 Conclusion
An effective offline parameter identification method of induction motors has been
deployed with a noticeably clear practical procedure. Especially, this method can be
developed without the use of expensive hardware. The high similarity of the actual and
simulated parameters such as the rotor speed and the stator current has proven the effec-
tiveness of the proposed method. This method can be also applied to various types of
induction motors. The future work of this study is to consider the fluence of other fac-
tors such as the temperature rise in stator and rotor windings on the motor parameter
estimation.
 A Detailed Procedure of Squirrel-Cage Three-Phase Induction Motor 33

Acknowledgments. The authors would like to acknowledge the valuable comments and sug-
gestions of Professor Rong-Ching Wu (Department of Electrical Engineering, I-Shou University,
Taiwan) for this research.
The authors would also like to acknowledge the financial support from School of Electrical and
Electronic Engineering, Hanoi University of Science and Technology (HUST) for this research.

References
1. Haque, M.H.: Determination of NEMA design induction motor parameters from manufacturer
data. IEEE Trans. Energy Convers. 23(4), 997–1004 (2008)
2. Zheng, J., Wang, Y., Qin, X., Zhang, X.: An offline parameter identification method of induction
motor. In: 2008 7th World Congress on Intelligent Control and Automation, pp. 8898–8901
(2008)
3. Monjo, L., Kojooyan-Jafari, H., Córcoles, F., Pedra, J.: Squirrel-cage induction motor parameter
estimation using a variable frequency test. IEEE Trans. Energy Convers. 30(2), 550–557 (2015)
4. Toliyat, H.A., Levi, E., Raina, M.: A review of RFO induction motor parameter estimation
techniques. IEEE Trans. Energy Convers. 18(2), 271–283 (2003)
5. Zamora, J.L., Garcia-Cerrada, A.: Online estimation of the stator parameters in an induction
motor using only voltage and current measurements. IEEE Trans. Ind. Appl. 36(3), 805–816
(2000)
6. Nayeem Hasan, S.M., Husain, I.: A Luenberger–sliding mode observer for online parameter
estimation and adaptation in high-performance induction motor drives. IEEE Trans. Ind. Appl.
45(2), 772–781 (2009)
7. Lee, K., Frank, S., Sen, P.K., Polese, L.G., Alahmad, M., Waters, C.: Estimation of induction
motor equivalent circuit parameters from nameplate data. In: 2012 North American Power
Symposium (NAPS) (2012)
8. Simba, K.G., Quilumba, F.L., Granda, N.V.: Parameter estimation of a three-phase induction
motor from direct starting stator transient measurements. In: 2020 IEEE ANDESCON (2020)
9. Wu, R.-C., Tseng, Y.-W., Chen, C.-Y.: Estimating parameters of the induction machine by the
polynomial regression. Appl. Sci. (2018)
 A Hybrid Energy Sliding Mode Controller
for the Rotary Inverted Pendulum

Van-Anh Nguyen1(B) , Duc-Binh Pham1 , Duy-Thi Pham1 , Ngoc-Tam Bui2 ,

and Quy-Thinh Dao1(B)
1 Hanoi University of Science and Technology, Hanoi, Vietnam
{anh.nguyenthivan1,thinh.daoquy}@hust.edu.vn
2 Shibaura Institute of Technology, Tokyo, Japan

Abstract. The rotary inverted pendulum (RIP) is a classical and underactuated

system with two degrees of freedom. In this paper, we concentrate on addressing
the swing up and stabilization of the rotary inverted pendulum, which are its two
fundamental issues. In the beginning, we proceed to model the system. The two
problems described previously are then solved by an energy sliding mode con-
troller (ESMC). The effectiveness of the controller has also been verified through
simulation results under the influence of external disturbance.

Keywords: Sliding mode control · Energy-based control · Rotary inverted

pendulum · Swing up control · Stabilization control

1 Introduction
Underactuated systems, which have control inputs less than their number of degrees
of freedom, are increasingly common. These systems often have rear components that
mechanically bond with the former so that when there is an impact on the former, the
latter will be impacted indirectly and move along. Controlling underactuated systems is
more difficult than full-actuated systems due to constraints and components that cannot
be controlled directly. One of the typical underactuated systems is a rotary inverted
pendulum (RIP). The RIP is composed of two connecting rods and a DC motor. The
second rod is attached to the end of the first rod, which is directly connected to the motor.
The two main control goals of the RIP are swing up and stabilization. The first is trying
to bring the pendulum to an upward vertical position, and the second focuses on main-
taining the pendulum in that position. Regarding the swing-up control algorithm, there
are several approaches such as sliding mode control (SMC) [1], Proportional Deriva-
tive (PD) controller [2], Energy-based controller [3], etc. The above control methods
can swing the pendulum up, but they also have limitations. While the PD controller
takes a relatively long time to bring the pendulum to the desired position, the sliding
mode controller can quickly get there. However, programming for the SMC is also more
complicated. Controlling swing up with energy is the most popular and widely used
method because of its simple structure and efficiency. To maintain the pendulum in the
equilibrium position, several stabilization controllers have been used such as sliding

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 34–41, 2023.
https://doi.org/10.1007/978-3-031-22200-9_4
 A Hybrid Energy Sliding Mode Controller 35

mode control [4], evolutionary algorithms [5, 6], adaptive control [7], and have shown
promising results.
The sliding mode control approach is widely used and successfully designed con-
trollers for nonlinear models. In this study, we further improve this method to swing up
and stabilize the inverted pendulum by including the energy controller. An energy-based
controller brings the pendulum to the upward position, and the sliding mode controller
stabilizes the pendulum and keeps it in an equilibrium position. The combination of
these two controllers can be called an Energy sliding mode controller (ESMC).
The following parts of this research are organized as follows: Sect. 2 is about the
mathematical model of the RIP. The control problems of the rotary inverted pendulum,
including swing up and stabilization, are investigated in Sect. 3. Section 4 provides the
simulation results, and some conclusions are discussed in Sect. 5.

2 System Modeling
The research is carried out based on the rotary inverted pendulum model as illustrated
in Fig. 1. The 3D model of the rotary inverted pendulum. A pendulum bar with a mass
of m (kg) and 2L in length is attached to the M (kg) swing arm. This swing arm has a
length of r (m). The rotation angles of the pendulum bar and the arm are α (deg) and θ
(deg) respectively.

Fig. 1. The 3D model of the rotary inverted pendulum.

Considering the Oxy coordinate system as in Fig. 1. The 3D model of the rotary
inverted pendulum. We have:

x = rθ − Lsin(α)
(1)
y = Lcos(α).
36 V.-A. Nguyen et al.

According to Newton’s 2nd law, we can obtain:


mr θ̈ + mLsin(α)α̇ 2 − mLcos(α)α̈ = fx .
(2)
−mLcos(α)α̇ 2 − mLsin(α)α̈ = fy − mg

where fx , fy are the total forces affect the pendulum in the x and y-direction.
Applying Euler’s equation to the rotation of the pendulum, we have:
1 2
mL α̈ = fx Lcos(α) + fy Lsinα (3)
3
Substituting (2) into (3), we get:
1 2
mL α̈ = [mr θ̈ + mLsin(α)α̇ 2 − mLcos(α)α̈]Lcos(α) + [−mLcos(α)α̇ 2 − mLsin(α)α̈]Lsin(α)
3
4
⇔ −mLrcos(α)θ̈ + mL2 α̈ − mgLsin(α) = 0 (4)
3

Moment of inertia of load at DC motor shaft:

km kg ηg km2 kg2 ηg
T= Vin − θ̇ − Jm kg2 ηg θ̈ . (5)
Rm Rm
where Jm is rotor inertia of DC motor, kg is the gearbox ratio, km is back electro-motive
force constant, Rm is armature resistance, Vin is armature voltage, ηg is the efficiency of
the gearbox.
Applying Euler’s equation to the rotation of the arm, we have:

Jeq θ̈ = T − Beq θ̇ − fy r (6)

in which Jeq is moment of inertia of the arm and the pendulum about the axis of rotation
at O , Beq is the friction coefficient of the pendulum arm.
Substituting (2) into (5), we get:

Jeq θ̈ = T − Beq θ̇ − [mr θ̈ + mLsin(α)α̇ 2 − mLcos(α)α̈]r

⇔ (Jeq + mr 2 )θ̈ − mLrcos(α)α̈ + mLrsin(α)α̇ 2 = T − Beq θ̇ (7)

From (4) and (7) we have the equation of motion of the pendulum:

−mLrcos(α)θ̈ + 43 mL2 α̈ − mgLsinα = 0
(8)
(Jeq + mr 2 )θ̈ − mLrcos(α)α̈ + mLrsin(α)α̇ 2 = T − Beq θ̇

Substituting (5) into (8), we obtain:

ηg kg2 km
2
km kg ηg
(Jeq + mr 2 + Jm kg2 ηg )θ̈ − mLrcos(α)α̈ + mLrsin(α)α̇ 2 + (Beq + )θ̇ = Vin (9)
Rm Rm

From (4) and (9), we get:


c1 θ̈ + c2 cos(α)α̇ + c2 sin(α)α̇ 2 + c5 θ̇ = c6 Vin
(10)
−c2 cos(α)θ̈ + c3 α̈ − c4 sinα = 0
 A Hybrid Energy Sliding Mode Controller 37

ηg kg2 km2
with c1 = Jeq + mr 2 + Jm kg2 ηg ; c2 = mLr; c3 = 43 mL2 ; c4 = mLg; c5 = Beq + Rm ;
km kg ηg
c6 = Rm .
 T
Putting the state variables as x = α θ α̇ θ̇ and representing the equations in the
form of:

ẋ = Ax + Bu (11)

with, ⎡ ⎤ ⎡ ⎤
0 0 1 0 0
⎢ 0 0 0 1 ⎥ ⎢ 0 ⎥
A=⎢
⎣
⎥ and B = ⎢
⎦ ⎣
⎥.
c1 c4 sinα
α 0 −c2 α̇cosαsinα −c2 c5 cosα
2 c2 c6 cosα ⎦
c2 c4 cosαsinα
α 0 −c2 c3 α̇sinα −c3 c5 c3 c6

3 Control Problem

To solve the swing up problem, we employ the energy control law given by K. J. Astrom
and K. Furuta [8]. Accordingly, a corresponding energy is supplied to bring the pendulum
to the upright position. We choose the pendulum’s energy in the standing position to be
zero. Then, to get the required energy, the control law is given as follows:

u = satξ g (k(E − E0 ))sgn(α̇cos(α)) (12)

in which u is the acceleration of the DC motor, ξ is a tuning parameter, k is a design

parameter, the function satξ g denotes a function which saturates at ξ g.
To convert between the shaft acceleration and the motor input voltage, we have the
following equation:
RMLu
Vin = + kg km θ̇ (13)
ηg kg km2

When the pendulum’s angle reaches the vicinity of the desired position, we switch
to using a stabilization controller to keep the pendulum balance (α = 0). We developed
a pendulum stabilization controller based on the sliding mode control approach. From
(10), the rotary inverted pendulum system can be presented by the following state-space
equations:
⎧ .
⎪
⎪ x1 = x2
⎪
⎪
⎨ x.2 = F1 (x) + G1 (x)
. (14)
⎪
⎪ x.3 = x4
⎪
⎪
⎩
x4 = F2 (x) + G2 (x)

where
c1 c4 sin(x1 ) − c2 cos(x1 )(c2 sin(x1 )x22 + c5 x4 ) c6 c2 cos(x1 )
F1 (x) = ; G1 (x) = ;
c1 c3 − c22 cos(x1 )2 c1 c3 − c22 cos(x1 )2
38 V.-A. Nguyen et al.

c3 F1 (x) − c4 sin(x1 ) c6 c3
F2 (x) = ; G2 (x) = (15)
c2 cos(x1 ) c1 c3 − c22 cos(x1 )2

It can be seen that the system in (14) is combined from two subsystems, which are
respectively related to α and θ. Two sliding surfaces can be chosen as follow:

s1 = x2 + λs1 x1
s2 = x4 + λs2 x3 (16)

The following Lyapunov function is considered to ensure the stability of the system:

V = |s1 | + λV |s2 | (17)

with λV is a positive scalar with the value between 0 and 1 set by users. The control
signal u can be chosen such that:
V
V̇ = −βsat( ) (18)

where β > 0 is a selected coefficient related to the reaching rate to zero of V . sat( 
V
)
is the saturation function to reduce the chattering phenomenon This function can be
determined as follows:
 V
V  if  < |V |
sat( ) = (19)
 sgn(V ) otherwise

From Eq. (19), we have:

V = |s1 | + λV |s2 |
= s1 sgn(s1 ) + λV s2 sgn(s2 ) (20)

Then, take the derivative of the Eq. (20):

. .
V̇ = s1 sgn(s1 ) + λV s2 sgn(s2 )
= (F1 + G1 u + λs1 x2 )sgn(s1 ) + λV (F2 + G2 u + λs2 x4 )sgn(s2 ) (21)

The control signal u can be inferred from Eqs. (18) and (21) as follows:

−βsat( 
V
) − (λs1 x2 + F1 (x))sgn(s1 ) − λV (λs2 x4 + F2 (x))sgn(s2 )
u= . (22)
G1 (x)sgn(s1 ) + λV G2 (x)sgn(s2 )

4 Simulation and Results

The parameters of the pendulum are selected for simulation as: g = 9.81 (m/s2 ), Jeq =
3.5812 × 10−3 (kgm2 ), L = 0.335 (m), m = M = 0.125 (kg), r = 0.125 (m) Beq =
0.04 (Nmsrad−1 ), kg = 1, Rm = 2.6 ( ), Jm = 3.87 × 10−7 (kgm2 ), km = 7.67 ×
10−3 (Vsrad−1 ), ηg = 0.9. The effectiveness of the applied controller is verified through
 A Hybrid Energy Sliding Mode Controller 39

Fig. 2. Simulation results of pendulum angle.

the pendulum angle in Fig. 2. It is evident that the pendulum quickly reaches the desired
position within one second and maintains it until there is an external force impact on the
pendulum.
Figures 3, 4, and 5 show the change in the variables θ, α̇, θ̇ over time. In particular,
the value θ in Fig. 3 changes much when the pendulum needs to swing up, and that value
is relatively stable when the pendulum is in equilibrium. Similarly, the pendulum’s
velocity varies most at 3 s in the presence of disturbance. Considering the value of θ̇ in
Fig. 5, the velocity of the arm changes continuously; however, the maximum value is
only about 1000 deg/s and is consistent with the limits that the actuator can achieve.

Fig. 3. Simulation results of arm angle.

40 V.-A. Nguyen et al.

Fig. 4. Simulation results of pendulum velocity.

Fig. 5. Simulation results of arm velocity

5 Conclusion
In this paper, the problem of bringing the pendulum up from the bottom and keeping
it in the equilibrium position has been solved using an ESMC controller, combining
an energy-based and sliding mode control. Overall, the controller performed well when
performing two tasks of swing up and stability. That has been verified through simulation
results. The system’s response in terms of time is relatively fast and has high stability.
The angular velocity of the arm has also been considered and has shown the feasibility
of applying the control method in practice.

References
1. Kurode, S., Chalanga, A., Bandyopadhyay, B.: Swing-up and stabilization of rotary inverted
pendulum using sliding modes. IFAC Proc. Vol. 44(1), 10685–10690 (2011)
2. Oltean, S.E.: Swing-up and stabilization of the rotational inverted pendulum using PD and
fuzzy-PD controllers. Procedia Technol. 12, 57–64 (2014)
 A Hybrid Energy Sliding Mode Controller 41

3. Qin H., Shao, W., Guo, L.: Research and verification on swing-up control algorithm of rotary
inverted pendulum. In: The 26th Chinese Control and Decision Conference (2014 CCDC),
pp. 4941–4945. IEEE (2014, May)
4. Rajan, A., Kumar, A.A., Kavitha, C.S.: Robust control methods for swing-up and stabilization
of a rotary inverted pendulum. In: 2016 International Conference on Emerging Technological
Trends (ICETT), pp. 1–6. IEEE (2016, October)
5. Hassanzadeh, I., Mobayen, S.: Controller design for rotary inverted pendulum system using
evolutionary algorithms. Math. Prob. Eng. 2011 (2011)
6. Rahimi, A., Raahemifar, K., Kumar, K., Alighanbari, H.: Controller design for rotary inverted
pendulum system using particle swarm optimization algorithm. In: 2013 26th IEEE Canadian
Conference on Electrical and Computer Engineering (CCECE), pp. 1–5. IEEE (2013, May)
7. Chen, Y.-F., Huang, A.-C.: Adaptive control of rotary inverted pendulum system with time-
varying uncertainties. Nonlinear Dyn. 76(1), 95–102 (2013). https://doi.org/10.1007/s11071-
013-1112-4
8. Åström, K.J., Furuta, K.: Swinging up a pendulum by energy control. Automatica 36(2),
287–295 (2000)
 A Neural Network-Based Fast Terminal Sliding
Mode Controller for Dual-Arm Robots

Hue Luu Thi1 , Van Trong Dang2 , Nhu Toan Nguyen2 , Duc Thinh Le2 ,
and Tung Lam Nguyen2(B)
1 Electric Power University, Hanoi, Vietnam
2 Hanoi University of Science and Technology, Hanoi, Vietnam

lam.nguyentung@hust.edu.vn

Abstract. The paper studies a robust adaptive fast terminal sliding mode con-
troller (FTSMC) based on a radial basis function neural network (RBF NN) for
a dual-arm robot manipulator system that coordinates the motion of the general
object. First, kinematics, a general dynamic model of the system consisting of
manipulators and the object, is inferred about the position and direction of the
object as the states of the derived model. Second, an FTSMC method is designed,
followed by the construction of two RBF systems: one approximates uncertain-
ties and external disturbances, and the other estimates the force applied to the
object according to the object’s uncertainty model. Next, the Lyapunov theory is
employed to demonstrate the stability of the closed-loop system and derive the
adaptation laws for RBF NN. Finally, simulation results of the dual-arm robot
system with three degrees of freedom (3-DOF) manipulators are presented to
illustrate the effectiveness of the proposed method.

Keywords: Dual-arm robotic · Fast terminal sliding mode controller (FTSMC) ·

Radial basis function (RBF) neural network

1 Introduction

Nowadays, dual-arm robot manipulator robots are increasingly being applied in many
different fields. Challenges in cooperative control include synchronizing position and
controlling the force applied to an object. If too much force is applied to the object may
deform its shape, or if the robot does not follow the predetermined path well, it can
endanger the operator due to sudden movements. Therefore, it is necessary to design
an efficient controller for this system to control position, orientation, and force with an
uncertain dynamic model.
The problem of control under the kinematic uncertainty of the system has been
solved by proposing the application of advanced control theories such as the sliding
mode control in [1, 2] designed a controller for the robot to improve trajectory tracking
quality. An adaptive controller for a dual-robot to hold an object with an unknown center
of mass and moment of inertia has been proposed in [3]. A hybrid adaptive controller has
been proposed in [4] to improve the efficiency of motion tracking and force adjustment

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 42–52, 2023.
https://doi.org/10.1007/978-3-031-22200-9_5
 A Neural Network-Based Fast Terminal Sliding 43

when a cooperative dual-arm robot grasps an unknown object. In [5], a non-chattering

robust sliding controller enhanced by a fuzzy logic unit to track the desired trajectory
with high accuracy and safely transport the object. In addition, terminal sliding mode
control (TSMC) is proposed by [6, 7] to improve the design of the sliding surface and
lead to the tracking errors converging to zero in a finite time. In [8, 9] the authors present
fast terminal sliding mode control (FTSMC) for nonlinear systems and demonstrate the
convergent feature of the controller superior to the conventional sliding mode control.
An adaptive hybrid position/force controller using fuzzy backstepping has been pro-
posed in [10], the fuzzy logic approximation properties are used to estimate the unknown
dynamics of the system. The controller ensures that and the tracking errors of both posi-
tion and force converge to the desired small value by selecting the appropriate design
parameters. In [11] proposed an adaptive neuron force/position controller, the controller
achieves the desired trajectories of the object and internal force tracking. A feedfor-
ward neural network is used to learn the unknown dynamics of the system with updated
law based on Lyapunov stability analysis. Some research for the combined dual-arm
robot system has solved two problems simultaneously: uncertain dynamic and uncer-
tain environment. The neuron impedance controller has been proposed in [12] with two
control loops, a position control loop and a force control loop to guarantee the desired
trajectory and cooperation force. Several other powerful approximation techniques are
based on differential equations or polynomials to estimate the uncertainty dynamic and
disturbances of a multi-manipulators system moving an object [13, 14]. The robust adap-
tive controller proposed in [15] to deal with the problem of uncertain base coordinates,
uncertainty model, and external disturbance, using the RBF neural network to estimate
all kinds of these uncertainties.
Inspired by the analysis of previous studies mentioned above, in this paper, we focus
on developing a generalized model for the dual-arm robot manipulator. Besides, an
adaptive fast terminal sliding mode control is based on two neural networks to deal with
nonlinear elements, uncertainties, and external disturbances. Finally, simulation results
and conclusions are made to verify and show the achievements of this paper.

2 Dynamics of the Dual-Arm Robot–Object System

A dual-robot system under this study is illustrated in Fig. 1. The system consists of two
manipulators arranged sort by duality, each arm is a robot with n degrees of freedom.
The coordinate frames are defined as follows:

{OXY}: The reference coordinate frame, located at the origin of the first robot arm;
{ov xv yv }: An object frame, located at the mass center of the object.

E i (i = 1 − 2) is the contact of the end-effector of the arm-robot ith with surface

of the object; z = [roTv O , φoTv O ]T denotes the position rov O = [x, y, z]T vector and
orientation φov O = [ψ, ϕ, θ ]T vector of the object frame concerning the reference
coordinate frame; zEi ov = [rETi ov , φETi ov ]T denotes the position and orientation vector of
the object frame concerning E i frame.
44 H. L. Thi et al.

Fig. 1. Model of a dual-arm robot holding an object

The position and orientation of the object is determined through the homogenous
transformation matrices of the corresponding coordinate frames and represented as
follows:

TOov = TOEi .TEoiv . (1)

From Eq. (1), the kinematic relationship between the object and the joint angles
of the manipulators are built. From there, the relationship between the object’s motion
velocity and the joint’s velocity is obtained [11]

ż = Ai q̇i ; z̈ = Ai q̈ + Ȧi q̇

That means

q̇ = A.ż; q̈ = A.z̈ + Bż, (2)

where A = [(A−1 −1 T T −1 −1 T −1 −1 T T
1 ) , (A2 ) ] ; B = [(−A1 Ȧ1 A1 ) , (−A2 Ȧ2 A2 ) ] .
T

The dynamic of the dual-arm can be formulated as in [10]

H (q) q̈ + C(q, q̇) q̇ + G(q) + τ d + JBT (q) F = τ , (3)

where q = [q T1 , qT2 ]T is the joint angle (2n × 1) vector; τ = [τ T1 , τ T2 ]T

denotes the torque (2n × 1) vector applied to the joints; τ d = [τ Td 1 , τ Td 2 ]T
denotes bounded unknown disturbances; F = [F1T , F2T ]T is the vector of
forces/moments on the object exerted by the manipulators at the end-effector; JB (q) =
blockdiag[J1 (q1 ), J2 (q2 )] represents the Jacobian analytical matrix of the manipulators;
H (q) = blockdiag[H1 (q1 ), H2 (q2 )], Hi (qi ) is the inertia matrix of the ith manipula-
tor C(q, q̇) = blockdiag[C1 (q1 , q̇1 ), C2 (q2 , q̇2 )] denotes the matrix of Coriolis and
centrifugal; G(q) = [G1T (q1 ), G2T (q2 )]T denotes the gravity vector.
The dynamic of the object also can be written in the following form [11]:

Hz z̈ + Cz (z, ż) ż + gz = Fz , (4)

 A Neural Network-Based Fast Terminal Sliding 45

where Hz , Cz , gz have the same meaning for the object as Eq. (3); Fz denotes the
force/moment vector applied at the center of the object by the two manipulators.
Sum of force/moment applied at the object is expressed through the force and moment
at the contact point generated by the manipulators at the end-effector: Fz = E F, where
E is the grasp matrix at the contact point, so:

F = E + Fz , (5)

where E + = E T (E E T )−1 is the pseudo-inverse of grasp-matrix E.

3 Fast Terminal Sliding Mode Controller Design

First of all, we define a reference velocity of the object by
m
żr = żd + γep + βepn , (6)

where γ = diag(γ1 , γ2 , γ3 , γ4 , γ5 , γ6 ),β = diag(β1 , β2 , β3 , β4 , β5 , β6 ),m, n (m > n)

are positive odd numbers, and 1 < mn < 2. The desired force is generated by an estimated
reference model of the object as follows

Fzd = Ĥz z̈r + Ĉz (z, żr )żr + ĝz (7)

The reference model of the object is estimated by using RBFNN

Hz z̈r + Cz (z, żr )żr + gz = W∗T

0 φ0 (x) + ε0 , (8)

where ε0 is the approximate error, so

T
Ĥz z̈r + Ĉz (z, żr )żr + ĝz = Ŵ0 φ0 (x)
 T
The tracking errors are defined ep = ep1 ep2 ep3 ep4 ep5 ep6 = zd − z.
The error between actual and reference velocities is given by.
m
s0 = ż − żr = −ėp − γep − βepn (9)

Consider that s0 = 0, we have:

m m
ėp + γep + βepn = 0 or ėpk + γk epk + βk epk
n
=0 (10)
1
in which k = 1 ÷ 6. Let uk = epk
n
then epk = ukn substitute this into (10) leads to

nukn−1 u̇k = −γk ukn − βk ukm (11)

Divide both sides by ukm into (11) leads to

duk n
nukn−m−1 = −γk ukm − βk (12)
dtk
46 H. L. Thi et al.

Integral of both sides into (12) leads to

 tk  0
un−m−1 duk
dtk = n k n (13)
0 epk (0) −γ u m − β
k k k

Thus,
n   1− m
 
tk = ln −γk epk n (0) + βk − ln βk (14)
(n − m)γk
The desired force applied to the object is updated based on the estimates of the
dynamic parameters of the object. The desired force at contact points should satisfy the
relation given in (5)
F d = E + Fzd (15)

3.1 Design Control Law

Define the reference velocity of the ith robot: q̇r .
The error between actual and reference velocities is given:
s = q̇ − q̇r = A s0 (16)
The dynamic of the dual-arm robot in Eq. (3), the matrices H (q), C(q), and vector
G(q) contain all the parameters of the manipulators such as length, mass, and moment
of inertia v.v. It is difficult to determine exact parameters due to errors in measurement,
and environment v.v. Therefore, it is assumed that the components in Eq. (3) are written:
H (q) = H0 (q) + H (q), C(q, q̇) = C0 (q, q̇) + C(q, q̇), G(q) = G0 (q) + G(q),
where H0 (q), C0 (q, q̇) and G0 (q) are nominal values and H (q), C(q, q̇) and G(q)
are the component uncertainty of the dynamics model. So The dynamic of the dual-arm
robot in Eq. (3) can be rewritten:
H (q) ṡ = τ − JBT (q) F − f0 (q, q̇r , q̈r ) − C(q, q̇) s − ( f (q, q̇, q̈) + τ d ), (17)
where f0 (q, q̇r , q̈r ) = H0 q̈r + C0 q̇r + G0 , f (q, q̇, q̈) = H q̈ + C q̇ + G
The dynamic uncertainty and disturbances components are estimated using RBFNN.
D = f (q, q̇, q̈) + τ d = W∗T φ(x) + ε, (18)

where ε is the error approximation, so D̂ is the estimation of the function D,

T
D̂ = Ŵ φ(x) (19)
So Eq. (17) as:
H (q) ṡ + C(q, q̇) s + W ∗T φ(x) + ε = τ − JBT (q) F − f0 (q, q̇r , q̈r ) (20)
Now, position control and force control without the measurement of force at contact
points is proposed:
τ = f0 (q, q̇r , q̈r ) + Ŵφ(x) + JBT F d − K1 (A s0 ) − K2 sgn(s) − K3 sgn(s0 ) (21)
 A Neural Network-Based Fast Terminal Sliding 47

3.2 Update Law for NN Weighs

Let an estimated error the dynamics of the object following is then obtained:
   
Hz z̈ + Cz ż + gz − Ĥz z̈r + Ĉz żr + ĝz = Fz − Fzd (22)

substitute Eqs. (8), (9) into (22) as:

T
Hz ṡ0 + Cz s0 + W̃0 φ0 (x) + ε0 = Fz − Fzd (23)

The system is considered stable according to the Lyapunov stability principle, the
candidate for the Lyapunov function can be selected as
1 T 1 1 1
V = s H s + tr(W̃ T Γ −1 W̃ ) + sT0 Hz s0 + tr(W̃0T Γ0−1 W̃0 ) (24)
2 2 2 2
The following lemmas as [11].

Lemma 1 AT JBT = E and AT JBT E + = I , where I is the identity matrix.

The time derivative (24) of the Lyapunov function becomes:

1 ˙ ) + sT H ṡ + 1 sT Ḣ s + tr(W̃ T Γ −1 W̃
V̇ = sT H ṡ + sT Ḣ s + tr(W̃ T Γ −1 W̃ ˙ )
0 z 0 z 0 0
2 2 0 0 0

Using Eqs. (20) and (23) and the matric Ḣ −2C and Ḣz −2Cz are skew-symmetric, so
sT (Ḣ − 2C)s = 0 and sT0 (Ḣz − 2Cz )s0 = 0. Using Eqs. (15), (16) and due to properties
of matrices sT0 AT = (A s0 )T , and using Lemma 1, the derivative of the Lyapunov is
determined as follows:


V̇ ≤ −(A s0 )T Ks A s0 − sT (K2 sgn(A s0 ) − ε) − (A s0 )T W̃ T φ(x)

˙ ) − sT sgn(s ) − ε ) − sT AT J T E + (F − F d )
+ tr(W̃ T Γ −1 W̃ 0 (K3 0 0 0 B z z
˙ )
+ sT0 (Fz − Fzd ) − sT0 W̃0T φ0 (x) + tr(W̃0T Γ0−1 W̃ (25)
0

According to the Lyapunov stability principle, the condition for stability of closed
dynamics is V̇ ≤ 0 then
˙ ) − (A s )T W̃ T φ(x) = 0
tr(W̃ T Γ −1 W̃ 0

and
˙ ) − sT W̃ T φ (x) = 0
tr(W̃0T Γ0−1 W̃ 0 0 0 0

So, the updated law for the weights of the neural network may have the form:
˙ = −Γ φ(x) (A s )T
Ŵ 0
˙
Ŵ0 = −Γ0 φ0 (x) sT0 (26)
48 H. L. Thi et al.

Now, the derivative of the Lyapunov function is as follows:

V̇ ≤ −(A s0 )T Ks A s0 − sT (K2 sgn(A s0 ) − ε) − s0T (K3 sgn(s0 ) − ε0 ) (27)

If K2 and K3 are selected, K2 > ε; K3 > ε0  then V̇ ≤ 0.

It is possible to prove that the dynamic system is stable under the control input (21)
combined with the updated law (26). According to the Lyapunov stability principle, the
system is stable.

4 Simulation Results
In this section, numerical simulation tests are carried out to illustrate the effectiveness
of the proposed method. The dual-arm robot is specified, with each arm being a 3-DOF
manipulator that works in 2D space to preserve the computational time. The desired
position and rotation angle of the object are set out with initial and terminal values as
follows:

x0 = 0.54 (m); y0 = 1.4(m); θ0 = 0 (rad)

xf = 1.2 (m); yf = 1.9 (m); θf = 0.349(rad)

To ensure smooth motion, the quintic polynomial curve is employed to construct

reference trajectories from the initial pose to the terminal pose. Choosing the execution
time Tf = 2.0592 (s), the desired path is given by:

xd = 0.54 + 0.756t 3 − 0.5508t 4 + 0.107t 5

yd = 1.4 + 0.5728t 3 − 0.4173t 4 + 0.0811t 5
θd = 0.3999t 3 − 0.2913t 4 + 0.0566t 5

The parameters for proposed controller are chosen as.

Ks = diag([100, 100, 100, 100, 100, 100]); γ = diag([300, 600, 600]); β =
diag([100, 100, 150]) For the RBF system, the number of neurons in hidden layer is
1000, the center and width of Gaussian function are linspace(−2, 2, 1000) and 30 respec-
tively, the adaptive gains are Γ = 200; Γ0 = 200. Since it is difficult to obtain accurate
system parameters, we assume the uncertainty accounts for 30% of total dynamic model.
Furthermore, the external noises acting on each arm are presented in Fig. 2. To clearly
show the advantages of the proposed method, the comparisons in terms of tracking
performance between the FTSMC controller and FTSMC integrated with the RBF neu-
ral network are implemented. From a qualitative perspective, Figs. 3 and 4 show the
object’s trajectory on the x-axis and y-axis, respectively. The rotation angle of the object
is presented in Fig. 5. The blue line, green line, and red dash line indicate the tracking
error of the NN-FTSMC, original FTSMC, and the reference path respectively. Although
the FTSMC method expresses robustness against disturbances, its performance is still
adversely impacted, leading to large errors in tracking trajectories, especially on the x
and y axis. On the other hand, with the support of the RBF approximation, the effects of
disturbances are compensated and adapted in the control signal. Therefore, the system
 A Neural Network-Based Fast Terminal Sliding 49

can ensure both robust and adaptive properties. The tracking errors of the NN-FTSMC
are much lesser than the original FTSMC. It can be seen more specific in quantitative
results using the root mean square error (RMSE) metrics on tracking errors in x, y-axis,
and rotation angle as in Table 1. As compared to the original FTSMC controller, the
tracking errors of NN-FTSMC are improved significantly, with 75.7% on the x-axis,
59.01% on the y-axis, and 68.62% in the rotation angle.

Fig. 2. External disturbances

Fig. 3. The trajectory of the object in x-axis

Fig. 4. The trajectory of the object in y-axis

The total forces and moments acting on the center of the object are also illustrated in
Fig. 6. It can be seen that the sum of the forces and moments applied to the object at the
balance position is zero, which means the object is held stable at the balance position.
The sign function is substituted by the saturation function to alleviate the chattering
phenomenon; hence, the control inputs are applicable in real systems.
50 H. L. Thi et al.

Fig. 5. The rotation angle of the object

Table 1. RMSE comparison

Tracking error NN-FTSMC FTSMC Improvement (%)

x 0.003 (m) 0.0122 (m) 75.7
y 0.0042 (m) 0.0102 (m) 59.01
θ 0.0177 (rad) 0.0564 (rad) 68.62

Fig. 6. Total forces and moments acting on the center of the object

5 Conclusion
This paper proposed a robust adaptive controller based on a neural network to control a
dual-arm robot manipulator manipulating a rigid object with uncertainties and external
disturbances. Since in practice, it is difficult to accurately determine the parameters of
the dynamic model of the system, so using two NN algorithms to estimate the model
uncertainty of the dual-arm robot, the external disturbance, and the applied force. The
stability of the closed-loop control system is proven stable according to the Lyapunov
stability theory. The online adaptive learning law is suggested to ensure a stable system by
using the Lyapunov principle. The simulation results on Matlab/Simulink have confirmed
 A Neural Network-Based Fast Terminal Sliding 51

the stability, robust, fast response, the trajectory of the object convergence with the
error to zero, and the guaranteeing object holding stable at the balanced position of the
proposed controller.

Acknowledgement. Van Trong Dang was funded by Vingroup JSC and supported by the Master,
PhD Scholarship Programme of Vingroup Innovation Foundation (VINIF), Institute of Big Data,
code VINIF.2021.ThS.26.
This research is funded by the Hanoi University of Science and Technology (HUST) under project
number T2022-PC- 003.

References
1. Galicki, M.: Finite-time trajectory tracking control in a task space of robotic manipulators.
Automatica 67, 165–170 (2016)
2. Corradini, M.L., Fossi, V., Giantomassi, A., Ippoliti, G., Longhi, S., Orlando, G.: Minimal
resource allocating networks for discrete time sliding mode control of robotic manipulators.
IEEE Trans. Industr. Inf. 8(4), 733–745 (2012)
3. Ren, Y., Chen, Z., Liu, Y., Gu, Y., Jin, M., Liu, H.: Adaptive hybrid position/force control of
dual-arm cooperative manipulators with uncertain dynamics and closed-chain kinematics. J.
Franklin Inst. 354(17), 7767–7793 (2017)
4. Jiao, C., Yu, L., Su, X., Wen, Y., Dai, X.: Adaptive hybrid impedance control for dual-arm
cooperative manipulation with object uncertainties. Automatica 140, 110232 (2022)
5. Hacioglu, Y., Arslan, Y.Z., Yagiz, N.: MIMO fuzzy sliding mode controlled dual arm robot
in load transportation. J. Franklin Inst. 348(8), 1886–1902 (2011)
6. Liang, X., Wang, H., Zhang, Y.: Adaptive nonsingular terminal sliding mode control for
rehabilitation robots. Comput. Electr. Eng. 99, 107718 (2022)
7. Wu, X., Huang, Y.: Adaptive fractional-order non-singular terminal sliding mode control
based on fuzzy wavelet neural networks for omnidirectional mobile robot manipulator. ISA
Trans. 121, 258–267 (2022)
8. Jouila, A., Nouri, K.: An adaptive robust nonsingular fast terminal sliding mode controller
based on wavelet neural network for a 2-DOF robotic arm. J. Franklin Inst. 357(18), 13259–
13282 (2020)
9. Wang, Y., Zhang, X., Zhang, X.: Nonsingular fast terminal sliding mode-based robust adaptive
structural reliable position-mooring control with uncertainty estimation. Ocean Eng. 253,
111329 (2022)
10. Baigzadehnoe, B., Rahmani, Z., Khosravi, A., Rezaie, B.: On position/force tracking control
problem of cooperative robot manipulators using adaptive fuzzy backstepping approach. SA
Trans. 70, 432–446 (2017)
11. Panwar, V., Kumar, N., Sukavanam, N., Borm, J.-H.: Adaptive neural controller for coopera-
tive multiple robot manipulator system manipulating a single rigid object. Appl. Soft Comput.
12(1), 216–227 (2012)
12. Zhai, A., Zhang, H., Wang, J., Lu, G., Li, J., Chen, S.: Adaptive neural synchronized
impedance control for cooperative manipulators processing under uncertain environments.
Robot. Comput.-Integr. Manuf. 75, 102291 (2022)
13. Izadbakhsh, A., Nikdel, N., Deylami, A.: Cooperative and robust object handling by multiple
manipulators based on the differential equation approximator. ISA Trans. (2021)
52 H. L. Thi et al.

14. Izadbakhsh, A., Nikdel, N.: Robust adaptive control of cooperative multiple manipulators
based on the Stancu-Chlodowsky universal approximator. Commun. Nonlinear Sci. Numer.
Simul. 111, 106471 (2022)
15. Zhai, A., Wang, J., Zhang, H., Lu, G., Li, H.: Adaptive robust synchronized control for
cooperative robotic manipulators with uncertain base coordinate system. ISA Trans. (2021)
 A Novel Approach for Determination of Main
Dimensions and Armature Reactions of Axial
Flux Permanent Magnet Generators

Thanh Nguyen Vu(B) , Tan Nguyen Viet, and Vuong Dang Quoc

Department of Electrical Engineering, School of Electronic and Electrical Engineering, Hanoi

University of Science and Technology, Hanoi, Vietnam
thanh.nguyenvu@hust.edu.vn

Abstract. Nowadays, axial flux permanent magnet generators have been widely
used in practice, such as: electric vehicles, renewable energy (e.g., wind energy,
submersible energy), because they have advantages of the high working efficiency,
high power density and low noise. In particular, its structure is simple in compari-
son with normal generators. Moreover, the structure of these generators is usually
a type of disc, with the larger diameter rotor. Thus, it is very suitable for electric
drives of the high torque and low speed. Otherwise, in order to increase the power
supply, it can be used the multi-disc making more stages together. In this research,
the main dimensions of the two generators have been successfully computed and
analytically designed and the influence of armature reactions of axial flux per-
manent magnet generators with different loads is proposed via a finite element
method.

Keywords: Axial flux permanent magnet generators · Disc-type generator ·

Wind generator · Submersible generator · Armature reactions · Finite element
method

1 Introduction
As we have known, axial flux permanent magnet (AFPM) machines have more advan-
tages than normal electric machines due to the high efficiency, high power density and
better heat-dissipation. With the development of material technology, the efficiency
and production costs of AFPM machines has been remarkably improved. Therefore
these machines have been widely used in practice, such as: robotic joints, hoist, electric
vehicles, aircraft propellers and other industrial equipments with narrow spaces. These
AFPM machines can be directly drived without any gearboxes [1–7]. In practice, there
are many AFPM machines designed with different structures as presented in Fig. 1, such
as single-sided structure, double-sided structure with double-stator single-rotor, double-
sided structure with double-rotor single-stator or multi-disc machines with multi-rotor
multi-stator [1, 4, 5]. For the lower power, it is used a slotless stator and permanent
magnets mounted on the rotor surface [1, 2, 5, 8].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 53–64, 2023.
https://doi.org/10.1007/978-3-031-22200-9_6
54 T. N. Vu et al.

Fig. 1. Basic structures of AFPM machines (a) single-sided machine, (b) double-sided machines
with double-rotor single-stator, (c) double-sided machines with double-stator single-rotor, (d)
multi-disc machines with multi-rotor multi-stator [4]

Many authors have recently studied and analysed different AFPM machines, that
is, nine models of the AFPM generator using vertical axis wind turbines with different
structures is presented in [2]. The slotted stators of these model are the rectangular slots
combined with three different permanent magnet shapes. The paper has also investigated
with the same load current for nine generators to estimate the variable voltage, power
and efficiency. However, the effect of armature reactions for different loads (inductive
load or capacitive load) has been mentioned by this paper yet. In [9], the double-sided
structure with double-stator single-rotor is directly connected to the wind turbine in the
wind generator system. By this way, the use of gearboxes is eliminated; this leads to
reduction of mechanical losses and weight, and increase of efficiency.
The first step for desinging AFPM generators is to determine the main dimensions
from the input parameters (power, voltage, speed, torque …). This is a very important
step to define the volume of the machines relating to directly to installation space. The
main dimensions include as pointed out in Fig. 2 pole pairs; inner and outer diameter of
stator and rotor; effective stack length of stator and rotor; axial length of rotor yoke and
stator yoke; axial length of permanent magnet.

Fig. 2. Main dimensions of the AFPM generators

The determination of the main dimensions has been recently presented by many
papers in [3, 10]. However, the generator’s working characteristics and the magnetic
 A Novel Approach for Determination of Main Dimensions 55

field distribution in the generator have been not investigated in detail. The size of the
general AFPM machine is determined from the general sizing equations of the radial
flux machine [10], but in the article does not specify how the equations will apply to
electric motors or generators. The main dimensions of the slotless stator AFPM motor
were also presented in [8], where the equations are applied to design the AFPM motor
so that the flux density flows in stator and rotor cores are below the saturation value
of the B-H magnetization curve of steel. The sizing equation is also used to determine
the size parameters of the double-sided AFPM in [7] and considered the end-winding
length, but does not mention in determinate the size of the magnet.
When the generator is directly coupled to a low power vertical axis wind turbine,
the single-sided AFPM generator has the advantage of compact design and high torque.
There are two new ideas proposed in this paper, that is: Firstly, the equations of deter-
mining the size of AFPM machines in general are developed to apply for the AFPM
generator in detail, with a slotted stator using non-oriented electrical steel, rotor sur-
face mounted permanent magnets. Secondly, the influence of armature reactions on the
AFPM generator is also investigated.

2 Determination of the Main Dimensions

The main dimensions of AFPMG generators are determined based on the parameters:
electric loading (Am ); the number of turns per phase on a stator (N 1 ); the armature phase
current of a stator winding (I a ); pole embrace (α i ); the magnetic flux density in the
air-gap is (Bmg ); winding factor (k w1 ); the ratio between the inner and outer diameter
(k d ). The average diameter of stator core is defined as [3, 5, 8, 10].
Do + Di
D= (1)
2
The ratio (k d ) between the inner (Di ) and outer (Do ) diameter (or inner (Ri ) and outer
(Ro ) radius) is expressed as
Di Ri
kd = = (2)
Do Ro
The specific electric loading is determined as the number
√ of turns per phase on a
stator multiplied by the peak value of the armature current 2Ia divided by the armature
circumference [5, 11].
√
4 2m1 Ia N1
Am = (3)
πDo (1 + kd )
where Am has a value from 10,000 A/m for low power AFPM machines, 40,000 A/m
for medium power AFPM machines [12].
The magnetic flux per pole pitch (φf ) is then defined as

Ro  
2π π
φf = Bavg rdr = Bavg Do2 1 − kd2 (4)
2p 8p
Ri
56 T. N. Vu et al.

It should be noted that if the waveform of the air-gap flux density is a sinusoidal
distribution, the average flux density Bavg can be computed [5].
π
p
1 2
Bavg = π Bmg sin(pα)d α = Bmg (5)
p −0 π
0

By combining Eqs. (4) and (5) with the induced voltage in the stator winding, it can
be calculated from the following equation:
√  
√ π 2
Ef = π 2fN1 kw1 φf = ns N1 kw1 Bmg Do2 1 − kd2 (6)
4
where ns is the shaft speed (rev/s).
The apparent electromagnetic power is:

Selm = m1 nEf Ia (7)

where n is the number of stators for a multi-disc AFPM generator.

By substituting Eqs. (3) and (6) into Eq. (7), one gets

π2  
Selm = n kw1 ns Bmg Am Do3 (1 + kd ) 1 − kd2 (8)
16
The ratio k D is then defined as
1  
kD = (1 + kd ) 1 − kd2 (9)
8
In addition, the term S elm can also be calculated via [5].
Pout
Selm = ε (10)
η cos ϕ
where ε is the induced voltage factor ( = Ef /V1 , with motor  < 1, with generator
 > 1), cosϕ is the power factor and η is the efficiency.
Based on the Eqs. (8), (9) and (10), the output power can be defined as
n 2 1
Pout = π kD kw1 ns Bmg Am Do3 η cos ϕ. (11)
2 ε
The outer diameter (Do ) is then computed

εPout
Do = 3 n 2 (12)
2 π kD kw1 ns Bmg Am η cos ϕ

It is worth noting that the formula to evaluate the outside diameter of the machine
does not need to know the value of the induced voltage generating the phase current, but
only the induced voltage in a stator winding E f .
 A Novel Approach for Determination of Main Dimensions 57

The axial length of the stator core (L cs ) is also defined via the Do , i.e.,

Bmg πDo (1 + kd )
Lcs = (13)
Bcs 4p
Thus, the axial length of the stator can be approximated by:

Ls = Lcs + 1.6Wcui (14)

where Wcui is the end winding from the stator slot in the radial direction. One can be
defined as

Di2 + 4A mD
Kfill Js − Di
Wcui = (15)
2
where Js is the stator current density and Kfill is a slot fill factor.
The axial length of the rotor is obtained as follows:

Lr = Lcr + hpm (16)

where the axial length of the rotor core (L cr ) is defined as

Bu πDo (1 + kd )
Lcr = (17)
Bcr 8p
where Bu is the ratio between air-gap flux density and the leakage factor kpm . The axial
length of permanent magnet (hpm ) is then calculated

μr Bmg
hpm = Bmg
(g + Wcui ) (18)
Br − kpm

where μr is the relative permeability, kpm is the leakage factor and Br is the remanence.
After almost the main dimensions of the machine are determined, it is necessary to
verify that the input parameters such as torque, power…. Are satisfied. Otherwise, the
electric loading (Am ) or the flux density in the air-gap (Bmg ) must be changed until the
required parameters are satisfied.

3 Application Problems
Based on the equations developed in the previous section, two generators with different
powers and speed are applied to define the main dimensions, namely:
The parameters of generator (G1 ) of 1.2 kVA are given as: frequency (f ) = 40 (Hz);
number of pole pairs (p) = 12; stator slots = 18; air-gap flux density (Bmg ) = 0.528 (T);
electric loading (Am ) = 23000 (kA/m); permanent magnet NdFeB30.
The parameters of generator (G2 ) of 0.5 kVA are given as: frequency (f ) = 300 (Hz);
number of pole pairs (p) = 6; stator slots = 18; air-gap flux density (Bmg ) = 0.5(T);
electric loading (Am ) = 28000 (kA/m); permanent magnet NdFeB30.
58 T. N. Vu et al.

The characteristics, properties and parameters of permanent magnets can reference

in [13]. The obtained results are given in Tables 1 and 2. It should be noted that the
diameter ratio (k d ) is an important parameter affecting directly to the design process of
the AFPM generator, because the copper loss depends on this factor. So as to achieve
a maximum power and torque, the optimal value of k d is respectively defined √1 and
3
0.63 [14].

Table 1. Main dimensions of G1 with the speed of 200 (rpm)

Parameters Values
Outer diameter (Do ) 398 mm
Diameter ratio (kd ) 0.6
Air-gap (g) 1.5 mm
Axial length of the stator (Ls ) 27.5 mm
Axial length of the rotor (Lr ) 14.2 mm
Axial length of permanent magnet (hpm ) 5.3 mm

Table 2. Main dimensions of G2 with the speed of 3000 (rpm)

Parameters Values
Outer diameter (Do ) 120 mm
Diameter ratio (k d ) 0.583
Air-gap (g) 1 mm
Axial length of the stator (Ls ) 20.8 mm
Axial length of the rotor (Lr ) 11.8 mm
Axial length of permanent magnet (hpm ) 5.6 mm

The models of two generators (G1 and G2) are shown in Figs. 3 and 4.

Fig. 3. Model of G1
 A Novel Approach for Determination of Main Dimensions 59

Fig. 4. Model of G2

4 Investigation of Influence of Armature Reactions

The main dimensions and 3D-model of two generator are presented in Sect. 3. In this
part, the effect of armature reactions with different loads of the two AFPM generators
is investigated by a finite element method (FEM) to check the main dimensions of the
generators being suitable with the design requirements or not.

4.1 No-Load Operation

For the no-load operation, the generator terminals are not connected to the load, and the
rotor is kept at rated speed. The electromotive force (EMF)/induced voltage waveforms
of the G1 and G2 are pointed out in Fig. 5. The RMS value of G1 is 57 (V) and 107 (V)
for G2 . The distributions of flux density and flux lines in the stator and rotor cores are
presented in Figs. 6 and 7. It can be seen that the flux density near the slots is higher
than other areas.

Fig. 5. Three-phase electromotive force of G1 (top) and G2 (bottom), with no-load operation

4.2 Rated Load Operation

In order to see the effect of armature reaction when the generator supplies power to the
load. The following types of loads are used: pure resistive load, pure inductive load, pure
60 T. N. Vu et al.

Fig. 6. Flux density of G1 (left) and G2 (right)

Fig. 7. Flux lines of G1 (left) and G2 (right)

capacitive load, lagging load with cosϕ of 0.8 and leading load with cosϕ of 0.8. For
different loads, the speed is kept at the rated value.
The air-gap flux density distribution of the G2 with different loads can be shown that
the effect of armature magnetic field on the rotor magnetic field (the magnetic field of
the permanent magnet) (Fig. 8). It should be noted that the waveforms of the flux density
in the air-gap are changed with the different loads. When comparing at the same spatial
location in the air-gap for 5 typical loads, the flux densities are given in Table 3 and
Fig. 8. From the obtained data, it is found that for the leading load and pure capacitive
load, the total magnetic field in the generator is added by the action of the magnetizing.
This is reflected in the stronger increased air-gap flux density, which is highest with the
pure capacitive load.

Table 3. Air-gap flux density value of G2

No. Electric loads B (T)

1 Pure capacitive load 1.03
2 Leading load: cosϕ = 0.8 0.98
3 Resistive load 0.93
4 Lagging load: cosϕ = 0.8 0.89
5 Pure inductive load 0.82

For lagging inductive and pure inductive load, the magnetic field in the generator
is strongly attenuated due to the demagnetizing armature reaction. This is reflected in
more decreased air-gap flux density, which is lowest in the case of pure inductive load
(Table 3).
 A Novel Approach for Determination of Main Dimensions 61

The change of air-gap flux density due to the armature reaction is directly influenced
on variable output voltages. Figures 9 (a) and 10 (a) show that for the pure capacitive
load, the current is leaded in phase of 90° with the voltage, the output voltage of G1 is
79.5 (V) and 126.3 (V) for the G2 . Figures 9 (b) and 10 (b) show that for leading load,
the current is leaded in phase with the voltage when cosϕ of 0.8; terminal voltage of
generator G1 is 55.2 (V) and that of generator G2 is 108.9 (V). Figures 9 (c) and 10 (c)
show that voltage and current are in phase with each other when the load is resistive,
terminal voltage of generator G1 is 42.1 (V) and that of generator G2 is 101 (V). Figures 9
(d) and 10 (d) show that for lagging load, the current is lagged in phase with the voltage
when cosϕ of 0.8; terminal voltage of generator G1 is 31.5 (V) and that of generator
G2 is 90.7 (V). Figures 9 (e) and 10 (e) show that for pure inductive load, the current
is lagged in phase of 90° with the voltage, terminal voltage of generator G1 is 23.7 (V)
and that of generator G2 is 89 (V).

Fig. 8. Air-gap flux density waveform of generator G2 : Density distribution (left); values at the
same spatial location (right)

Fig. 9. Waveform of voltage and current of generator G1 with. a) pure capacitive load. b) leading
load when cosϕ of 0.8. c) resistive load. d) Lagging load when cosϕ of 0.8. e) Pure inductive load

The voltage waveforms in five different load cases of the G1 and G2 are depicted in
Fig. 11. From the voltage waveforms and the RMS values presented in Tables 4 and 5,
62 T. N. Vu et al.

the peak value of output voltage gradually increases corresponding to the pure inductive
load, lagging load with cosϕ of 0.8; resistive load, leading load with cosϕ of 0.8 and the
pure capacitive load. This is shown that the armature reactions are very evident through
the change of the output voltage values. This change is very consistent with the change
in air-gap flux density.

Fig. 10. Waveform of voltage and current of generator G2 with. a) pure capacitive load. b) leading
load when cosϕ of 0.8. c) resistive load. d) Lagging load when cosϕ of 0.8. e) Pure inductive load

Table 4. RMS values with different loads of generator G1

No. Electric loads Z ( ) Idm

- (A) UMF (V)
1 Pure capacitive load 5.68 14.01 79.5
2 Leading load when cosϕ of 0.8 3.94 14.02 55.2
3 Resistive load 2.98 14.07 42.1
4 Lagging load when cosϕ of 0.8 2.25 13.98 31.5
5 Pure inductive load 1.68 14.05 23.7

5 Conclusion
This paper has proposed a general method to determine the main dimensions of AFPM
machines, thereby applying specific computation to single-sided AFPM generators with
cored stator and permanent magnets mounted on the rotor surface. The main dimensions
of the two generators G1 and G2 have been successfully computed and analytically
designed. Specially, the influence of armature reactions with the type of different prop-
erties/loads have been shown that a good agreement between the change of air-gap flux
densities and output voltages.
 A Novel Approach for Determination of Main Dimensions 63

Table 5. RMS values with different loads of generator G2

No. Electric loads Z ( ) Idm

- (A) UMF (V)
1 Pure capacitive load 31.5 4.01 126.3
2 Leading load when cosϕ of 0.8 26.95 4.04 108.9
3 Resistive load 25.44 3.97 101
4 Lagging load when cosϕ of 0.8 22.56 4.02 90.7
5 Pure inductive load 21.7 4.1 89

Fig. 11. Output voltage waveforms of G1 (left) and G2 (right)

In the future, it is possible to develop a general design process to determine the main
dimensions suitable for various topologies of AFPM generators.

References
1. Mahmoudi, A., Rahim, N.A., Hew, W.P.: Axial-flux permanent-magnet machine modeling,
design, simulation and analysis. Acad. J. 2525–2549 (2011)
2. Wirtayasa, K., Hsiao, C.-Y.: Performances comparison of axial-flux permanent magnet
generators for small-scale vertical-axis. Alexandria Eng. J. 61 (2022)
3. Laxminarayan, S.S., Singh, M., Saifee, A.H., Mittal, A.: Design, modeling and simulation of
variable speed axial flux permanent magnet wind generator. Sustain. Energy Technol. Assess.
19, 114–124 (2017)
4. Parviainen, A.: Design of axial-flux permanent-magnet low-speed machines and performance
comparison between radial-flux and axial-flux machines. Diss. Lappeenranta University of
Technology (2005)
5. Gieras, J.F., Wang, R.J., Kamper, M.J.: Axial Flux Permanent Magnet Brushless Machines,
2nd edn. Springer, New York (2008)
6. Cetin, E., Daldaban, F.: Analyzing distinctive rotor poles of the axial flux PM motors by using
3D-FEA in view of the magnetic equivalent circuit. Int. J. 20, 1421–1429 (2017)
7. Tak, B.-O., Ro, J.-S.: Analysis and design of an axial flux permanent magnet motor for
in-wheel system using a novel analytical method combined with a numerical method (2020)
8. Radmanesh, H.: Sizing equations of axial flux permanent magnet (AFPM) machine based on
an analytical method. J. Appl. Res. Electr. Eng. 1, 14–21 (2021)
9. Rossouw, F.G.: Analysis and design of axial flux permanent magnet wind generator system
for direct battery charging applications. South Africa (2009)
64 T. N. Vu et al.

10. Huang, S., Luo, J., Leonardi, F., Lipo, T.A.: A comparison of power density for axial flux
machines based on general purpose sizing equations. IEEE Trans. Energy Convers. 14, 185–
192 (1999)
11. Kim, S.-A., Li, J., Choi, D.-W., Cho, Y.-H.: Design and analysis of axial flux permanent
magnet generator for direct-driven wind turbines. Int. J. Power Syst. 2 (2017)
12. Otuoze, A.O., Ogidi, O.O., Mohammed, O.O., Jimoh, A.A., Olaiya, B.S.: Sizing of wind
powered axial flux permanent magnet alternator using analytical approach. Niger. J. Technol.
35, 919–925 (2016)
13. Gieras, J.F.: Permanent Magnet Motor Technology Design and Applications, 3rd edn. Taylor
and Francis Group, LLC (2010)
14. Daghigh, A., Javadi, H., Torkaman, H.: Optimal design of coreless axial flux permanent
magnet synchronous generator with reduced cost considering improved PM leakage flux
model. Electr. Power Compon. Syst. 1–15 (2017)
 A Prototype of Lotus Fiber Extracting Machine

Ngoc-Kien Nguyen and Van-Tinh Nguyen(B)

School of Mechanical Engineering, Hanoi University of Science and Technology, Hanoi,

Vietnam
tinh.nguyenvan@hust.edu.vn

Abstract. Lotus silk is one of the rarest fabrics in the world. It is produced
only in some Asian countries such as Cambodia, Myanmar, and, more recently,
Vietnam. Taking silk from thousands of lotus stems is a very sophisticated and
hard job. Extracting enough lotus silk for one scarf can take two months, and the
final product can cost ten times as much as regular silk. Thus, lotus silk-made
products are usually very expensive. With the aim of improving the productivity
while preserving the traditional craft, this paper introduced the first lotus fiber
extracting machine in the world.

Keywords: Lotus fiber · Extraction · Machine · Silk

1 Introduction

Lotus fiber is high strength and low stretch fiber that is used to weave lotus silk. These
fibers are fragile, but once woven, can be as durable as traditional silk [1, 2]. In recent
years, it has been picked up by international fashion brands searching for new luxury
fiber. Nowadays, the lotus fiber is only manually extracted by a few skilled craftspeople
across the world. The lotus threads need to be processed within 24 h while they are
still wet; otherwise, they will be broken. And so harvesting has to be done each day.
Extracting enough lotus silk for one scarf can take ten days, and the final product can
cost 10 times as much as regular silk. To promote the growth of the lotus skill in the
future, the automation of the lotus fiber extracting process is essential.
Review on the machines in the agriculture, many auto machines or systems are men-
tioned by different researchers such as W. Lili et al. developed the robot for harvesting
a tomato in greenhouse, this robot applied the laser navigation control system and the
proportional-integral-derivative (PID) algorithm [3]. X. Ni proposed a robot for picking
an apple using image recognition technique [4]. Likewise, Zhao De-An designed an
apple harvesting robot which consists of a manipulator, end-effector and image-based
vision servo control system [5]. Many researchers invented a plucking machine [6–8].
Observing the field of the fiber extracting machine, there are some banana fiber extract-
ing machines [9, 10]. James McCrae has patented the pineapple leaf fiber extracting
machine [11]. Extraction Device for Fiber in Plant Stems and Leaves is developed by
Zhangjiagang Shengtai fiber company [12].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 65–73, 2023.
https://doi.org/10.1007/978-3-031-22200-9_7
66 N.-K. Nguyen and V.-T. Nguyen

Realizing the potential of the lotus weaving craft, this study builds a lotus fiber
extracting machine to raise the productivity of fiber spinning that is the world’s first pro-
totype. The machine simulated the lotus fiber pulling and spinning process of craftsmen.
The lotus fiber extracting machine includes three main sections. The first section is to
take and hole lotus stalks, and cut them along. The second section is to take out lotus
fiber from the stalk. In the third section, fiber is spun into thread. All working stages are
operated automatically.

2 Overall Structure Design of the Auto Lotus Fiber Extracting

Machine
2.1 Overview of the Structure

The overall structural assembly model of the lotus fiber extracting machine designed in
this study is shown in Fig. 1. The mechanical design of the lotus fiber extracting machine
was performed using SolidWorks 2020 (Dassault Systems, SE). The framework of the
machine is mainly made of steel; meanwhile, some is made by 3D printing technology
that will reduce manufacturing costs [13]. Lotus fiber extracting process is controlled
sequentially through three stages such as workpiece feeding, fiber pulling and fiber
spinning as shown in Fig. 2. Input devices are sensors, limit switches and button while
output devices are motors, solenoid valves and air cylinders. S7-1200 PLC controller
(Siemens, Germany) is used for this machine.

Fig. 1. Overview of auto lotus fiber extracting machine.

Feeding Pulling Spinning Lotus thread

Fig. 2. Overview of lotus fiber extracting process.

 A Prototype of Lotus Fiber Extracting Machine 67

The auto lotus fiber extracting machine is modularly designed, this structure will be
convenient for assembling, arranging flexibly and repairing. The machine consists of
four main modules as shown in Fig. 1.
(1) Workpiece feeding module to provide lotus stalks.
(2) Fiber pulling module to take out lotus fiber from the stalks.
(3) Fiber spinning module to spin fiber into thread.
(4) Machine frame module to store all parts of the machine.

2.2 Workpiece Feeding Module

Fig. 3. Workpiece feeding module.

As shown in Fig. 3, this module consists of five main parts: ball screw, guide chute,
air cylinder, cone bushing, and 3-claw gripper. The lotus stalk (green) is cut into 400
segments and place on guide chute, it gradually moves to working zone by ball screw.
Cone bushing moves forward and acts on 3-claw gripper to hold the lotus stalk by air
cylinder as described in Fig. 4.

F
l

θ
A
β
C α/2
P
y x Fm Pin
O
a) b)
Fig. 4. The proposed 3-claw gripper: a) 3D design; b) Kinematic diagram with design parameters.
68 N.-K. Nguyen and V.-T. Nguyen

To evaluate the performance of the proposed gripper, the grasping index (GI) men-
tioned in [14] is applied as design criterion, and it expresses the relationship between
input and output forces as defined in Eq. (1).
F
GI = (1)
Pin
where Pin is the force generated by the cylinder on cone bushing, F the grasping force.
After some substitutions, the Grasping Index for the proposed gripper can be expressed
as Eq. (2).

sin(β) l1
GI = . (2)
3 cos α2 l2

where α is the angle of bushing cone, β is the contact angle between the first link and
the cone bushing in grasping process and is variant depending on the position of the first
link. l 1 and l2 are respectively the length of the first and second link of the gripper. As
described in Eq. (2), the GI depends on the ratio (l1 /l 2 ) of the length of the first link to the
second link, the contact angle between the first link and the cone bushing that depends
on the position of the first link and the configuration angle of the gripper. The parameter
of the link depends on the diameter of lotus stalk that ranges from 15 mm to 20 mm. We
selected the length of the first link, l1 = 20 mm and l 2 = 40 mm for the second one.
These geometrical parameters are considered on convenient manufacturing factor and
working ability of the gripper. As can be seen in Eq. (2), GI reaches maximum when
the contact angle between the first link and the cone surface, β = 90°. In this paper, the
angle of bushing cone, α is set to 20°, thus GI = 0.169.

2.3 Fiber Pulling Module

Fig. 5. Fiber pulling module.

 A Prototype of Lotus Fiber Extracting Machine 69

As shown in Fig. 5, this module consists of six main parts: sensor, knife, air cylinder,
2-claw gripper, cylinder, and ball screw. When sensor detects the lotus stalk at the position
of 20 mm away from the cutter, the knife will cut on the lotus stalk at the depth of 1 mm
by the air cylinder. While cutting process, the lotus stalk rotates on the horizontal axis.
After that, the gripper will move to the right by ball screw and grasp the lotus stalk by air
cylinder, then, the gripper brings 20 mm-lotus-stalk segments to the left to expose the
lotus fiber. Implementing the experiment on pulling the lotus fiber, we see that the lotus
fiber is broken when the pulling velocity, V is greater than 0.046 (m/s). Assuming that
the maximum speed of motor, nmax is 600 (rpm), pitch of worm is calculated by Eq. (3).
v 46.60
l≥ = = 4.6 (mm) (3)
nmax 600
We selected that pitch of worm of ball screw (2.6), l = 5 (mm), thus the speed of
motor is determined by Eq. (4).
v 46.60
n= = ≈ 550 (rpm) (4)
l 5

2.4 Fiber Spinning Module

Fig. 6. Fiber spinning module.

This module consists of five main parts: ball crew 1, ball crew 2, upper spinning
table, lower spinning table, eccentric pin, and rope drum as shown in Fig. 6. This module
simulates the lotus fiber spinning process of the craftsman. When upper spinning table
moves in horizonal and vertical direction (green arrow) by ball screw 1 and 2, lower
spinning table moves in horizontal direction (red arrow) by eccentric pin. After spinning,
70 N.-K. Nguyen and V.-T. Nguyen

lotus thread is collected to rope drum. The velocity of spinning table is defined by
measuring the velocity of craftsman’s hand, V = 0.083 (m/s). We choose pitch of worm
of ball screw (3.1) and (3.2), l = 10 (mm), the speed of motor, n = 500 (rpm).

3 Hardware Design
Table 1 lists the motors and cylinders used in the auto lotus fiber extracting machine.
Seven 57HS7630A4 stepper motors are used for lotus fiber extracting process and an
ES-3910 direct current motor (DC motor) is used to collect the lotus threads on rope
drum. The stepper motor specifications are shown in Table 2. Two SMC CQSB25-75DC
cylinders are used to grasp and cut the lotus stalk, a MHZ2-32D cylinder is to grasp left
head of the lotus stalk to expose the lotus fiber. Table 3 shows the cylinder specifications.

Table 1. List of motors and cylinders used in the machine.

Devices Function
1 Stepper motor 1 To push lotus stalk to working area
2 Stepper motor 2 Rotating lotus stalk while cutting process
3 Stepper motor 3 To move gripper to grasp lotus stalk
4 Stepper motor 4 Spinning motion of upper hand in vertical direction
5 Stepper motor 5 Spinning motion of upper hand in horizontal direction
6 Stepper motor 6 Spinning motion of lower hand
7 Stepper motor 7 Pulling and collection lotus thread
8 DC motor Distributing lotus thread on rope drum to avoid getting stuck
9 Cylinder 1 To operate grasping process of 3-claw gripper
10 Cylinder 2 To operate cutting process of knife
11 Cylinder 3 To implement grasping process of 2-claw gripper in lotus fiber pulling
operation

Table 2. 57HS7630A4 stepper motor specifications.

Step angle 1.8o

Step accuracy ± 5%
Temperature rise 80 °C Max
Ambient temperature range −20 o C to + 50 o C
Insulation resistance 100 M Min. 500V DC
Dielectric strength 500V AC 1 min
 A Prototype of Lotus Fiber Extracting Machine 71

Table 3. SMC CQSB25-75DC and MHZ2-32D cylinder specifications.

SMC CQSB25-75DC MHZ2-32D

Tube inner diameter 25 Number of fingers 2 pcs
Rod operation method Single Rods Dimensions when closed 26 mm
Stroke 75mm Gripping force (when 0.5 Mpa at 158 N
closing)
Operating temperature −10 o C to + 70 o C Dimensions when open 48 mm
Operating pressure 0.05 to 1 (MPa) Gripping Force (when 0.5 Mpa at 193 N
opening)

Fig. 7. Diagram of control system for lotus fiber extracting machine.

A diagram of control system is shown in Fig. 7. Control system consists of two PLC
S7-1200 controllers which are connected each other via Ethernet for information transfer.
HMI monitor performs the information of velocity, spinning number and working time.
Outputs are DC motor, solenoid valves and drivers of stepper motor. Electrical diagram
is shown in Fig. 8.

4 Result and Discussion

Fabrication of automatic lotus fiber extracting machine is shown in Fig. 9. Parameter of
the machine is 1300 mm (length) × 620 mm (width) × 800 mm (height), the weight of
70 kg. The lotus stalk has a diameter in range of (15–18) mm. The productivity of the
proposed machine is four times as much as the craftsman. The quality of lotus thread is
good and as similar as one extracted by craftsman. Working process of the machine is
expressed in Fig. 10.
72 N.-K. Nguyen and V.-T. Nguyen

Ethernet

PLC S7 1200

Driver

Fig. 8. Electrical component diagram.

Fig. 9. A prototype of control system for lotus fiber extracting machine: a) Completed prototype;
b) Workpiece supplying module; c) Fiber pulling module; d) Fiber spinning module.

Feeding Cutting Pulling Spinning

Fig. 10. Extracting process of auto lotus fiber extracting machine.
 A Prototype of Lotus Fiber Extracting Machine 73

To evaluate the effectiveness of this machine, we used a success rate (SR) of the
spinning as described in Eq. 5.
Ns
SR = (5)
Nt
where Nt is the number of the tests, we conducted 30 experiments; Ns is the number of
successful spinning. In the first experiment, the velocity of the spinning motor was set
to 500 rpm, the success rate was 83.33%. The failure rate of the lotus fiber breakage rate
were 6.67% and 10% for unsuccessful spinning.

5 Conclusions
This paper presented the first prototype of auto lotus fiber extracting machine in the
world. The mechanical and electrical components of the machine were introduced. The
design of this machine is modularized to easily assemble and repair. It consists of four
main modules: workpiece feeding module, fiber pulling module, fiber spinning module
and machine frame module where the principle of the spinning module simulates the
motion of craftsman in working. Its productivity is four times as much as craftsman
and the quality of lotus thread is good. The success rate in the working process of the
proposed machine was 83.33% and will be improved in the further future.

References
1. Pan, Y., Han, G., Mao, Z., Zhang, Y., Duan, H., Huang, J., Qu, L.: Structural characteristics
and physical properties of lotus fibers obtained from Nelumbo nucifera petioles. Carbohydrate
Polym. 85(1) (2011)
2. Chen, D.S., Gan, Y.J., Yuan, X.H.: Research on structure and properties of lotus fibers. Adv.
Mater. Res. 476–478, 1948–1954 (2012)
3. Lili, W., et al.: Development of a tomato harvesting robot used in greenhouse. Int. J. Agric. &
Biol. Eng. 10(4), 140–149 (2017)
4. Ni, X., Wang, X., Wang, S., Wang, S., Yao, Z., Ma, Y.: Structure design and image recognition
research of a picking device on the apple picking robot. IFAC-PapersOnLine 51(17), 489–494
(2018)
5. De-An, Z., Jidong, L., Wei, J., Ying, Z., Yu, C.: Design and control of an apple harvesting
robot. Biosyst. Eng. 110(2), 112–122 (2011)
6. Erada, J.C.: Traveling type tea plucking machine. JP Patent No. 2008301831 (2008)
7. Aokiha, J.A.: Tea leaf-picking machine. JP Patent No. 2011193754 (2011)
8. Hu, X.G.: Dc tea plucking machine. CN Patent No. 102668812 A (2012)
9. Veera ajay, C., Vignesh Ramamoorthy, K., Subash, V., Robinston, R., Ragashwar, M., Justus
Panicker, C.T.: Design and fabrication of manually operated banana fiber extracting machine
for agriculture applications. Mater. Today: Proc., Part 9 45, 8199–8202 (2021)
10. Poudel, S., Chapai, S., Subedi, R.K., Giri, T.R., Adhikari, S.: Design, fabrication and testing
of banana fibre extraction machine. J. Innov. Eng. Educ. 2(1) (2019)
11. James, M.: (US2722039): apparatus for obtaining fibers from plant leaves (1955)
12. Zhangjiagang Shengtai fiber Co. Ltd. (CN201018898): plant stem leaf fibre in device for
extracting extraction device for fiber in plant stems and leaves (2012)
13. Konarova, M., Jones, G., Rudolph, V.: Enabling compact GTL by 3D-printing of structured
catalysts. Results Eng. 6 (2020)
14. Ceccarelli, M.: Fundamentals of Mechanics of Robotic Manipulation, vol. 27. Springer
Science & Business Media, Berlin (2013)
 A Simulation Investigation of Dynamic Wheel
Load of a Heavy Truck with Hydro-Pneumatic
Suspension System

Le Xuan Long1(B) , Dang Viet Ha2 , Nguyen Van Tuan2 , Vu Thanh Niem2 ,
and Vu Thi Hien1
1 Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
lexuanlong@tnut.edu.vn
2 Vietnam Register, 12059, Hanoi, Vietnam

Abstract. The main objective of this paper is to investigate the effects of operat-
ing conditions of a mining dump truck with hydro-pneumatic suspension system
(HPSs) on road surface friendliness, a three-dimensional vehicle-road coupled
dynamic model of a mining dump truck with 11 degrees of freedom (DOF) is
set up to investigate. The dynamic load coefficient (DLC) is chosen as objective
function. The different operating conditions such as road surface roughness, vehi-
cle speed and vehicle body mass are investigated through DLC value. The study
results show that the vehicle operating conditions have a great influence on the
road friendliness. The study results are the theoretical basis to optimize the design
of the suspension system for off-road vehicles.

Keywords: Mining dump truck · Hydro-pneumatic suspension · Dynamic wheel

load · DLC

1 Introduction
An analytical model of the HPSs is proposed to formulate considering polytropic change
in the seal friction, gas state, and the gas-oil emulsion flows through valves and orifices
[1]. Artificial intelligence has been applied to simulate and predict the suspension sys-
tem performance for vehicles which would incorporate the artificial neural networks,
Mamdani fuzzy logic and a hybrid system, the hybrid neural fuzzy interference system,
for achieving the goal [2]. A HPSs model based on fractional order was built with the
advantage of fractional order in viscoelastic material modeling considering the mechan-
ics property of multiphase medium of HPSs [3]. The performance of the HPSs of vehicle
on the ride quality of road surfaces was proposed to compare to the air spring and rubber
suspension systems using a full dynamic model of a heavy vehicle [4]. A semi-active
HPSs was proposed to improve ride comfort performance, without compromising the
road holding and load carrying capabilities of the passive suspension through an active
damper used in parallel with a gas spring [5]. The effects of HPS parameters such as the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 74–83, 2023.
https://doi.org/10.1007/978-3-031-22200-9_8
 A Simulation Investigation of Dynamic Wheel Load 75

accumulator charge pressure, damping diameter, and the track tensioning pressure were
investigated using the vehicle multibody dynamics model to improve the ride safety
[6]. The performance of the HPSs of heavy truck on vehicle ride comfort was proposed
and compared to those of leaf, rubber, and air suspension systems [7–9]. A novel sus-
pension configuration with both hydraulically interconnected suspension and electronic
controlled air spring with different suspension configurations under different maneuvers
and suspension height modes was evaluated the handling stability and ride comfort of
vehicle [10]. The initial pressure and initial volume of accumulator, and the diameter
of damping orifice of HPSs were optimized to improve vehicle ride comfort by using
a method that combines with an optimization technique and a virtual and real proto-
type experiment integrated-Kriging model [11]. The novel method optimization with a
multi-objective optimization and a mathematical model of HPS with gas–oil emulation
was proposed to design HPSs for both driver comfort and rollover stability [12]. The
vibration damping performance of HPSs of heavy truck was studied by theoretical and
experiment [13]. A semi-active HPSs based on the electro-hydraulic proportional valve
control was developed with computer simulation to evaluate the effectiveness of HPSs
[14, 15]. The main object of this paper is to investigate the effects of operating condi-
tions of a mining dump truck with hydro-pneumatic suspension system (HPSs) on road
surface friendliness using a mining dump truck with 11 DOF. The different operating
conditions such as road surface roughness, vehicle speed and vehicle body mass are
investigated through DLC value.

2 Mathematical Model of HPS Strut

Structural schematic of hydro-pneumatic suspension (HPS) strut with two oil chambers
and one gas chamber is shown in Fig. 1 which consist of the main oil chamber (1), ring
oil chamber (2), the gas chamber (3), the oil pressures of the oil chambers (1), (2) and
gas chamber (3), p1 , p2 and pa , the effective areas of cylinder, rod and floating piston
and gas chamber, A1 , A2 , and A3 , the area of the orifices between oil chamber (1) and
chamber (2), A12 , the vertical displacements of vehicle axle, floating piston, and vehicle
body, za , zd and zb . Nonlinear dynamic model of HPS strut is shown in Fig. 2 which
consist of the stiffness and damping coefficients kh and ch , vehicle axle and vehicle body
masses ma and mb and vertical force of HPS strut, Fh .
From Fig. 2, the vertical dynamic force of HPSs is determined by

Fh = Fk + Fc (1)

where, Fk is the elastic force of the gas chamber and Fc is the damping force of the oil
chambers.
The elastic force of the gas chamber of HPSs is determined by

Fk = (pa − p0 )A3 (2)

The pressure in the gas chamber is defined as an adiabatic process following the laws
of thermodynamics

p0 V0n = pa Van (3)

76 L. X. Long et al.

where p0 and V0 are the initial absolute pressure and volume in the gas chamber, pa and
Va are the absolute pressure and volume in the gas chamber, n is the polytrophic rate (1
< n < 1.4).
 n
V0
pa = p0 , Va = V0 + A3 (zb − za ) (4)
Va

zb Vehicle body

Cylinder
1 p1 A1 A
12
Combined piston
zd p
2 2
Floating piston
3
pa A2
za A3

Vehicle axle

Fig. 1. Structural schematic of HPS strut.

Fh Vehicle body
zb
mb

kh ch

za
ma
Fh Vehicle axle

Fig. 2. Nonlinear dynamic model of HPS strut.

Combination of Eqs. (4) and (2), the elastic force of the gas chamber of HPSs is
achieved by
 
V0
Fk = p0 − 1 A3 (5)
V0 + A3 (zb − za )
The damping force of HPSs is determined by
 
md g
Fc = p1 (A1 − A3 ) − p2 (A1 − A2 ) − p0 − (A2 − A3 ) (6)
A3
 A Simulation Investigation of Dynamic Wheel Load 77

where A1 , A2 , and A3 are the area of cylinder, rod and floating piston, p1 and p2 are the
pressures in the oil chamber (1) and chamber (2).
The flow rate through the orifice is determined by:

(p2 − p1 )
Q = Cd A12 2 sign(żb − ża ) (7)
ρ

where, Cd is coefficient of discharge, A12 is the area of the orifices and ρ is the density
of oil.
On the basic of volume balance, the flow rate is inferred by

Q = (A1 − A2 )(żb − ża ) (8)

Combination of Eqs. (7) and (8), the relationship p1 and p2 is achieved by

 
ρ(A1 − A2 )2 żb − ża (żb − ża )
p2 = p1 + (9)
2Cd2 A212

The differential motion equation of floating piston can be determined by

(p1 − pa )A3 − md g = md z̈d (10)

The Eq. (10) can be rewritten by

md (z̈d + g)
p1 = pa − (11)
A3
Combination of Eqs. (11), (9) and (6), the damping force of HPSs is achieved by
 
ρ(A1 − A2 )3 żb − ża (żb − ża )
Fc = −
2Cd2 A212
 n 
V0 md z̈d
+ p0 (A2 − A3 ) −1 − (A2 − A3 ) (12)
V0 + A3 (zb − za ) A3

3 Three-Dimension Dynamic Model of a Mining Dump Truck

A three-dimensional vehicle-road coupled dynamic model of a mining dump truck with

11 DOF under random road surface roughness is set up to analyze the effectiveness of
HPSs in the dynamic tire load, as shown in Fig. 3 which consists of the tire stiffness and
damping coefficients kti and cti , the stiffness and damping coefficients of HPSs khi and
chi , the tire stiffness and damping coefficients of cab isolation system kci and cci , the
tire stiffness and damping coefficients of driver’s seat suspension system ks and cs , the
vertical displacements of front axle, rear axle, vehicle body, cabin and driver’s seat za1 ,
za2 , zb , zc and zs , the angular displacements of axles, vehicle and cab bodies θa1 , θa2 ,
ϕb , θb and ϕc , θc , the inertia moments of axle, vehicle and cab masses Iax1 , Iax2 , Ibx , Iby
78 L. X. Long et al.

l8 zs
zs
k zc
zc s cs l7
ϕc θc b4
kc3 Icx,Icy kc4
kc1 cc1 zb kc3 b5
cc3 cc3 cc4 zb
l5 l6 ϕb mb b3
l3 l4 θb Ibx,Iby
kh1 ch1 kh3 ch3 kh3 ch3 kh4 ch4
za1 za2 θa2 za2
ma1 ma2 Iax2
b2
kt1 ct1 kt3 ct3 kt3 ct3 kt4 ct4
q1 q2 q3 q4
l1 l2
b1

Fig. 3. Three-dimensional vehicle-road coupled dynamic model of a mining dump truck.

and Icx , Icy , the distances lj and bk and the road surface excitations qi (i = 1 ÷ 4, j = 1
÷ 8, k = 1 ÷ 5).
The multi-body system theory is used to separate the system into subsystems which
are linked by the force and moment equations. D’Alembert’s principle is used to set up
force and moment equations to describe vehicle dynamic subsystems. The equations of
motion can be written as.
The equations of motion for front axle of vehicle are written as follows


2 
2
ma1 z̈a1 = Fhi − Fti (13)
i=1 i=1


2
b1  b2
2
Iax1 θ̈a1 = (−1)i+1 Fti + (−1)i Fhi (14)
2 2
i=1 i=1

The equations of motion for rear axle of vehicle are written as follows


4 
4
ma2 z̈a2 = Fhi − Fti (15)
i=3 i=3


4
b1 
4
b2
Iax2 θ̈a2 = (−1) i+1
Fti + (−1)i Fhi (16)
2 2
i=3 i=3

The equations of motion for vehicle body are written as follows


4 
4
mb z̈b = Fci − Fhi (17)
i=1 i=4


2 
4 
2 
4
Iby ϕ̈b = Fhi l3 − Fhi l4 − Fci (l5 + l6 + l7 )− Fci l7 (18)
i=1 i=3 i=1 i=3


4
b2
Ibx θ̈b = (−1)i+1 Fhi − (Fc1 + Fc3 )(b3 + b4 ) − (Fc2 + Fc4 )b4 (19)
2
i=1
 A Simulation Investigation of Dynamic Wheel Load 79

The equations of motion for cab are written as follows


4
mc z̈c = Fci − Fs (20)
i=1


2 
4
Icy ϕ̈c = Fci l5 − Fci l6 + Fs l8 (21)
i=1 i=3


4
b3
Icx θ̈c = (−1)i+1 Fci + Fs b5 (22)
2
i=1

The equations of motion for seat vertical motion are written as follows

ms z̈s = −Fs (23)

where, Fti , Fhi , Fci and Fs are the vertical forces of tires, HPSs, cab’s isolation and driver’s
seat suspension system.
The vertical force of driver’s seat suspension system is determined as follows
 
Fs = ks (zs − zc − ϕc l8 − θc b5 ) + ks żs − żc − ϕ̇c l8 − θ̇c b5 (24)

The vertical forces of HPSs are determined by Eq. (12) and the remaining forces are
determined similarly to force of driver’s seat suspension system.
Road surface excitation: The filtering white noise method is used to describe the
time domain excitation of the road surface based on reference [20] and time domain
representation of the road surface can be given

q̇(t) + 2πf0 q(t) = 2πn0 Gq (n0 )vw(t) (25)

where, Gq (n0 ) is the road roughness coefficient which is defined for typical road classes
from A (very good) to H (very poor) according to ISO 8608(2016) [17], v = f/n is the
speed of vehicle from 10 m/s to 30 m/s, n is the road space frequency from 0.013 m−1
to 3.33 m−1 , and it can guarantee the temporal frequency of road surface f ranges from
0.33 Hz to 28.3 Hz which is the low excitation frequencies of road surface transmitted
to vehicle body; f0 is a minimal boundary frequency with a value of 0.0628 Hz; n0 is a
reference spatial frequency which is equal to 0.1 m; w(t) is a white noise signal.
Dynamic load coefficient: DLC is chosen as objective function which is defined by
a ratio of the root mean square of the vertical dynamic tire force over static load [16, 18,
19] as follows
FT ,RMS
DLC = (26)
Fs
where, FT,RMS , Fs are the root mean square of the vertical dynamic and the static tire
force.
80 L. X. Long et al.

4 Results and Discussion

The three-dimensional dynamic model of a mining dump truck with 11 DOF is simulated
by the MATLAB/Simulink software with a set of vehicle and HPSs parameters in the
reference [9]. The dynamic tire load at right rear axle of a mining dump truck when
vehicle moves on ISO class D road surface (poor condition) and ISO class E road
surface (very poor condition) at speed of 30 km/h and full load are shown in Fig. 4.

4 4
Dynamic tire force (N)

Dynamic tire force (N)

x 10 5 x 10
2

0 0

-2
-5
0 10 20 30 40 50 0 10 20 30 40 50
Time(s) Time(s)
a) ISO Class D a) ISO Class E

Fig. 4. The dynamic tire force at axle of vehicle under two different road surface conditions.

From the simulation results of Fig. 4 could be determined according to Eq. (27).
The DLC values rear axle under ISO class D and ISO class E road surface conditions
are respectively 0.176 and 0.297. The DLC value of right rear axle under ISO class E
road surface condition increase 40.68% in comparison with ISO class D road surface
condition. The DLC value increases quickly when vehicle moves on the bad road surface
conditions which makes the negative effects on the road surface. In order to investigate
dynamic wheel load of heavy vehicle with HPSs, the different operating conditions such
as road surface roughness, vehicle speed, body mass, discussions will be presented in
the next sections.
Road surface roughness: Five road surface conditions from ISO class A (very good)
to ISO class E (very poor) in ISO 8608 [17] are used when vehicle moves at the speed of
15km/h and 30km/h, full load. The DLC values at right rear axle of vehicle when vehicle
moves under the different road surface conditions are shown in Fig. 5. The results in
Fig. 5 show that the DLC values increase more quickly when vehicle move on the worse
road surface conditions. The DLC values at right rear axle at vehicle speed of 30 km/h
increase 32.28%, 32.05%, 32.11%, 30.84% and 31.28%, respectively in comparison with
vehicle speed of 15 km/h. The results also show that the DLC value at vehicle speed of
30 km/h increase the negative impacts on road surface much worse than at vehicle speed
of 15 km/h.
Vehicle speed conditions: The values of vehicle speed from 10 km/h to 60 km/h are
chosen to investigate dynamic wheel load of heavy vehicle with HPSs when vehicle move
moves on the ISO class D and ISO class E road surface conditions and full load. The
DLC values at right rear axle of vehicle with HPSs under two road surface conditions
at different vehicle speed conditions are shown in Fig. 6. The result of Fig. 6 show
that the DLC values at right rear axle under ISO class E condition increase 40.69%,
43.37%, 40.68%, 42.28%, 42.54% and 43.3%, respectively in comparison with ISO
class D condition when vehicle speed value increases. The DLC values increase quickly
 A Simulation Investigation of Dynamic Wheel Load 81

at vehicle speed conditions from 10 km/h to 30 km/h which increase very slowly at
vehicle speed conditions from 30 km/h to 60 km/h. It leads to the negative impact on
road surface of vehicle with HPSs at high speed conditions.
Vehicle body mass conditions: The vehicle body mass conditions including m =
[25, 50, 75, 100, 125, 150, 175 200]% mb are used to investigate dynamic wheel load of
heavy vehicle with HPSs under the ISO class D and ISO class E road surface conditions
at vehicle speed of 30 km/h. The DLC values of right rear axle of vehicle are shown
in Fig. 7. The results of Fig. 7 show that the DLC values at right rear axle of vehicle
decrease quickly when vehicle body mass value increases which lead to reduce the
negative impact on road surface. However, it has no benefit to the durability of the part
of vehicle and the safe movement of vehicle.

0.3
v=30km/h
0.2 v=15km/h
DLC

0.1

0
A B C D E
Different road surfacce conditions

Fig. 5. DLC values at right rear axle under the different road surface conditions.

0.4
ISO Class D
ISO Class E
DLC

0.2

0
10 20 30 40 50 60
Different speed conditions, v (km/h)

Fig. 6. DLC values at right rear axle under the different vehicle speed conditions.

0.4 ISO Class D

ISO Class E
DLC

0.2

0
25 50 75 100 125 150 175 200
Different vehicle body mass, mb(%)

Fig. 7. DLC values at right rear axle under the different vehicle body mass conditions.

5 Conclusions

In this paper, the operating conditions of a mining dump truck with hydro-pneumatic
suspension system (HPSs) are investigate its effect on surface friendliness using 11
82 L. X. Long et al.

DOF three-dimensional dynamic model. The main study conclusions drawn from the
investigation results are: (1) the DLC values increase more quickly when vehicle move on
the worse road surface conditions; (2) the DLC values at right rear axle under ISO class E
condition increase 40.69%, 43.37%, 40.68%, 42.28%, 42.54% and 43.3%, respectively
in comparison with ISO class D condition when vehicle speed value increases; and (3)
The DLC values at right rear axle of vehicle decrease quickly when vehicle body mass
value increases. The Study results are useful references for designers and manufacturers
in the field of off-road vehicle suspension systems.

Acknowledgment. The work described in this paper was supported by Thai Nguyen University
of Technology.

References
1. Yin, Y., et al.: Characterization of a hydro-pneumatic suspension strut with gas-oil emulsion.
Mech. Syst. Signal Process. 106, 319–333 (2018)
2. Ali, D., Frimpong, S.: Artificial intelligence models for predicting the performance of hydro-
pneumatic suspension struts in large capacity dump trucks. Int. J. Ind. Ergon. 67, 283–295
(2018)
3. Zhang, J., Chen, S., et al.: Research on modeling of hydropneumatic suspension based on
fractional order. Math. Probl. Eng. (2015)
4. Long, L.X., et al.: Performance analysis of the hydro-pneumatic suspension system of heavy
truck. IJMET 9, 1128–1139 (2019)
5. Solomon, U., Padmanabhan, C.: Semi-active hydro-gas suspension system for a tracked
vehicle. J. Terramechanics 48(4), pp. 225–239 (2011)
6. Han, S., Chao, Z., et al: Research on the effects of hydropneumatic parameters on tracked
vehicle ride safety based on cosimulation. Shock Vib. 2017, 1–10 (2017)
7. Long, L.X., Quynh, L.V., et al.: A comparison of ride performance of hydro-pneumatic sus-
pension system with those of rubber and leaf suspension systems. IOP Conf. Ser.: Mater. Sci.
Eng. 914 012037 (2020)
8. Long, L.X., Quynh, L.V., et al.: Ride performance evaluation of air and hydro-pneumatic
springs of suspension system. IJARET 12(1), 439–447 (2021)
9. Long, L.X.: Performance analysis of the hydro-pneumatic suspension system on vehicle ride
comfort. Science Project, Thai Nguyen University of Technology. Thai Nguyen, Viet Nam,
(2019)
10. Qi, H., Zhang, Y., et al. (2019) Improvement of both handling stability and ride comfort of
a vehicle via coupled hydraulically interconnected suspension and electronic controlled air
spring. Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile
Engineering, 095440701985653 (2019)
11. Gong, B., et al.: Ride comfort optimization of a multi-axle heavy motorized wheel dump
truck based on virtual and real prototype experiment integrated Kriging model. Adv. Mech.
Eng. 7(6), 168781401558425 (2015)
12. Kwon, K., Seo, M., Kim, et al.: Multi-objective optimization of hydro-pneumatic suspension
with gas–oil emulsion for heavy-duty vehicles. Veh. Syst. Dyn. 58, 1–20 (2019)
13. Xuefeng, S., Shengjie, J., et al.: Research on the vibration damping performance of hydro-
pneumatic suspension of mining dump truck. Vibroengineering PROCEDIA 20, 113–119
(2018)
 A Simulation Investigation of Dynamic Wheel Load 83

14. Yue, W.C., Li, S.C.: Investigation of semi-active hydro-pneumatic suspension for a heavy
vehicle based on electro-hydraulic proportional valve. World J. Eng. Technol. 5, 696–706
15. Sim, K., Lee, H., Yoon, J.W., et al.: Effectiveness evaluation of hydro-pneumatic and
semi-active cab suspension for the improvement of ride comfort of agricultural tractors. J.
Terrramech. 69, 23–32 (2017)
16. L. V. Quynh, N. V. Tuan, et al: Optimal design parameters of air suspension systems for
semi-trailer truck. Part 1: modeling and algorithm. Vibroengineering PROCEDIA, vol. 33,
pp. 72–77, (2020)
17. ISO 8608: Mechanical Vibration-Road Surface Profiles-Reporting of Measured Data.
International Organization for Standardization, London (2016)
18. Quynh, L.V., Tuan, N.V., et al.: Optimal design parameters of air suspension systems for
semi-trailer truck. Part 2: results and discussion. Vibroengineering PROCEDIA 33, 147–152
(2020)
19. Buhari, R., Rohani, M.Md., et al.: Dynamic load coefficient of tyre forces from truck axles.
Appl. Mech. Mater. 405(408), 1900–1911 (2013)
20. Long, L.X., Quynh, L.V., Cuong, B.V.: Study on the influence of bus suspension parameters
on ride comfort. Vibroengineering PROCEDIA 21, 77–82 (2018)
 A Study of Kinematics, Dynamics of the Stair
Climbing Wheelchairs

Nguyen Hoang Hiep1 , Le Hong Ky1(B) , Duong Tan Dat1 , and Le Hong Ky Duyen2
1 Vinh Long University of Technology Education, Vinh Long, Vietnam
kylh@vlute.edu.vn
2 International University—Vietnam National University, Ho Chi Minh City, Vietnam

Abstract. The objective of this article is to report the research results of designing
electric wheelchair that can climb stairs. The study focuses on innovating and
developing mechanical transmission assemblies of the stair climbing structure.
Observing kinetics, dynamics and simulating the wheelchair motion are done by a
professional software. Manufacturing and testing are implemented at Vinh Long
University of Technology Education.

Keywords: Dynamics · Electric-powered wheelchair · Kinetics · Stair-climbing

structure

1 Introduction
Wheelchairs for disabled people are increasingly being emphasized by manufacturers in
terms of design and function. Stair climbing wheelchairs for disabled people have helped
the elderly and disabled. Ghanshyam Das, Palkar Aman Manohar published research
on a seat backrest adjustment system that had been designed to adjust the centre of
gravity before climbing up and downstairs [1]. G. Quaglia, W. Franco has presented
a new version of a stair-climbing wheelchair [2]. With the introduction of a different
architecture and, more specifically, the proper design of a cam mechanism that manages
the seat orientation, the proposed objective has been achieved. New Horizon College of
Engineering, Bangalore, India has presented an electric wheel chair that can move on
structured and unstructured planes and go up and down staircase. It can also climb over
the obstacles with low jerk felt by the user [3]. André A. Jorge, Luis A. M. Riascos,
Paulo E. Miyagi presented a new type of wheelchair, the Hybrid Motorized Wheelchair
(HMW), which is under development. It has two locomotion systems, a conventional
wheel system and another system with tracks, which enables it to transpose obstacles,
such as stairs [4]. Tian-ci Jiang, Sheng-hao Yin, Eiichiro Tanaka, introduced a wheelchair
that is based on the three mechanisms with advantages of all of them. With linkage
mechanisms, the wheelchair can also adjust the location of the center of gravity (COG)
of the user for safer climbing [5]. Additionally, the linkage mechanism can also assist
the user to stand up. However, today’s popular wheelchairs are not able to help them
overcome obstacles or climb stairs.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 84–92, 2023.
https://doi.org/10.1007/978-3-031-22200-9_9
 A Study of Kinematics, Dynamics of the Stair Climbing Wheelchairs 85

Recently, a research team of PhD students from the Swiss Federal Institute of Tech-
nology and the University of Zurich has introduced a test version of the Scewo electric
wheelchair with 2 more rubber tracks to move on a variety of stairs [6, 7]. This wheelchair
is considered to be a revolutionary product for people in wheelchairs. Since 2015, the
research team of Vinh Long University of Technology Education has introduced several
versions of stair climbing electric wheelchairs that have been tested in the laboratory [8,
9].
Basically, the introduced stair climbing electric wheelchairs usually operate on 2
wheels like electric wheelchairs when running on ordinary surfaces. When climbing
stairs, the wheelchair will use 2 rubber tracks to help support to climb in the opposite
direction. According to the test results, the wheelchair will run at a steady speed for
climbing stairs of about 1 step/second.

2 Modeling the Stair Climbing Mechanism

According to [10, 11], when moving up the stairs, the static load of the mechanism is
formed due to the forces impeding the movement, including 3 main components: Friction
force Fr (including static friction and rolling friction), the friction force of the air (wind)
Fg, and the gravitational force of component P1 according to the formula:
P1 = m.g.sinα (1)
In which, m includes the mass of the person and the wheelchair, α is the angle
between the plane through the top of the stairs (Fig. 1) and the horizontal plane (α =
30o).
To determine the number of revolutions of the motor or the torque of the speed
reducer, it is necessary to first determine the load of the speed reducer or the load on the
2 rubber tracks.
The pair of pulleys is considered to be driven from the speed reducer to calculate as
Fig. 1.

Fig. 1. The components of the wheelchair motion resistance.

F t : tractive force
F r : friction force
Fr = m(C1 + C2 .v) (2)
86 N. H. Hiep et al.

C 1 , C 2 : coefficient of static friction and coefficient of rolling friction C 1 (m/s2 ); C 2

(m/s2 ) depends on tire and road surface.
F g is wind resistance force, but the wheelchair runs slowly which can be ignored.
The impact components of gravitational force due to uphill:

P1 = m.g.sinα (3)

Newton’s 2nd Law is applied to the wheelchair’s uphill motion

Ft = m · v + Fg + Fr + m · g · sin α (4)

Total load torque acting on the wheel axle T w includes the torque to accelerate the
wheel, the friction torque of the wheel axle and the torque required to create the tractive
force F t .

Tw = Jw · Nw + Tms + r · Ft (5)

In which:
Jw and r are the inertia and the active pulley radius.
N w is the number of revolutions of the tracks.
(4) is replaced by (5) with v = N w . r (air friction is ignored), the total pulley load is:
 
Tw = Jw + w.r 2 .N w + Tms + m.r(r + C1 + r.C2 ).NW + r.m.g.sinα (6)

Assuming that the weight and friction force between the outer surface of the rubber
tracks and the road surface are evenly distributed on the right (R, Right) and left (L, Left)
rubber tracks, then the load for the left rubber track (Tw1 L ) and right (Tw1 R ) respectively
are:
  
Tw1 R = 21  Jw + wr 2 Nw1 R + Tms1 R + 21 m.r(r + C1 + r.C2 ).Nw1 R + 21 r.m.g.sinα.
Tw1 L = 21 Jw + wr 2 Nw1 L + Tms1 L + 21 m.r(r + C1 + r.C2 ).Nw1 L + 21 r.m.g.sinα.
(7)

The friction torques can be simulated as viscous friction:

Tms1 R = d · Nw1 R (8)

Tms1 L = d · Nw1 L (9)

With d: coefficient of viscous friction, the system of differential Eqs. (7) is rewritten
as follows:
  
Tw1 R = 21  Jw + w.r 2 .N w1 R + d .Nw1 R + 21 m.r(r + C1 + r.C2 ).Nw1 R + 21 r.m.g.sinα
Tw1 L = 21 Jw + w.r 2 .N w1 L + d .Nw1 L + 21 m.r(r + C1 + r.C2 ).Nw1 L + 21 r.m.g.sinα
(10)

From the above formulas, by using the Simulink tool in Matlab [12, 13], a wheelchair
simulator has been built according to the principle as in Fig. 2, showing a detailed model
of the designed stair climbing mechanism.
 A Study of Kinematics, Dynamics of the Stair Climbing Wheelchairs 87

Fig. 2. The Simulink graph of wheelchair simulation.

In Fig. 2, the active torque of the motor axle is transmitted through the differential
reducer to the active pulley pair (P_L, P_R) of the rubber tracker belt transmission. When
the rubber tracker belt is supplied with torque, the transmission will operate, and the
wheelchair move up the stairs. When moving up the stairs, the rubber tracker is resisted
by other torques (viscous friction torque, left or right friction torque, gravitational force).
These torques acting on the rubber tracker may be uneven and cause the speed of the two
transmissions to be different, and to overcome this, the differential reducer will work and
divide the speed appropriately between the two trackers. This way, when the wheelchair
goes up the stairs, it will be safer.
With the calculated values: P1 = 2177,33 N, JW = 0.012 kg.m2 , r = 10,5 cm, d =
15,02 N.s/m, by Simulink, the response of the wheelchair with the active torque of the
engine can be seen. The response (torque, revolutions) of the engine and the Simulink
active pulleys can be observed on the Figs. 3, 4 and 5.

Fig. 3. Simulation results of engine moment.

Fig. 4. Simulation results of engine revolutions.

88 N. H. Hiep et al.

Fig. 5. Simulation results of left and right revolutions of active pulley.

3 Calculation of Kinematics, Dynamics of the Stair Climbing

Wheelchairs
Normally, when the wheelchair is running on a flat road (or uphill with an inclining angle
less than 150), the tracker is responsible for transmitting the motion from the engine to
the rear wheel axle, causing the wheelchair to move.
When meeting the stairs, the stair climbing mechanism will operate. In this case, the
transmission line also starts from the motor axle, through the worm-gear transmission to
the 2 semi-axles as the active pulley to perform rotational movement, through the rubber
tracker belt as the wheelchair to climb stairs as simulated in the Simulink diagram in
Fig. 3. In which, the rubber tracker belt has an inner working surface along with the
main pulleys which are synchronous belts, Fig. 6.
Thus, the mechanical structure of the stair climbing mechanism consists of three
main transmissions: the worm-gear transmission, the synchronous belt transmission and

Fig. 6. Model of mechanical structure of stair climbing mechanism of the electric wheelchair.

Fig. 7. Stair climbing mechanism speed reducer.

 A Study of Kinematics, Dynamics of the Stair Climbing Wheelchairs 89

the planetary gear system (Fig. 7). This is a problem for a multi-object dynamic system,
which can be separated into transmissions to perform calculations [14].

3.1 Worm-Gear Transmission

In the case of this study, the worm plays an active role (drive). Knowing that the worm
has a revolution of n1 (rpm), the power of P1 (kW), then the input and output torques are
determined according to ISO standards [16–18] as formulas 11, 12.
30000.P1
T1 = (Nm) (11)
π.n1
30000.P2
T2 = (Nm) (12)
π.n2

Table 1. Calculation results of the worm-gear transmission.

Parameters Symbol Worm (1) Worm gear (2)

Power P 5.000 (kW) 4.524 (kW)
Speed n 650.00 (rpm) 100 (rpm)
Torque T 73.456 (Nm) 432.046 (Nm)
Efficiency η 0.905 (ul)
Radial force Fr 2177.334 (N)
Tangential force Ft 5996.419 (N) 4923.601 (N)
Axial force Fa 4923.601 (N) 5996.419 (N)
Normal force Fn 8049.195 (N)
Circumferential speed v 1.361 (mps) 0.817 (mps)
Slide speed vk 1.332 (mps)

Here, assuming that a worm-gear transmission uses a 24V DC motor, with a power
of 5kW, the number of revolutions of 650 rpm. Equations (11) to (12) as well as software
application [15] are used to calculate the kinematics and dynamics of the transmission.
According to ISO for the worm-gear transmission formulas [16–18], the calculation
results are shown in Table 1.

3.2 The Synchronous Belt Transmission

The stair climbing mechanism using a rubber tracker belt with an inner working surface
with the main pulleys is a synchronous belt transmission (The Dual Synchronous Belt
transmission). Two active pulleys are mounted on the 2 semi-axles of the differential
mechanism and receive transmission directly from the worm rigidly mounted on the
90 N. H. Hiep et al.

planetary gear box case. Also, according to ISO for the synchronous belt transmission
formulas [16–18], the calculation results are shown in Table 2.
The worm-gear transmission has the power, torque and revolutions calculated as
shown in Table 1. In the case of a synchronous belt transmission, because the power is
divided among 2 semi-axles, each active pulley has a power of 2,762 (kW), torque of
216,005 (Nm) and the number of revolutions is still 100 (rpm). Here, only the parameters
for a synchronous belt transmission are calculated, making it easy to deduce the other
transmission. Calculation results of basic parameters are as shown in Table 2.
Using the software, it is also possible to determine the synchronous belt transmission
parameters. The results are as in Table 2.

Table 2. Calculation results of the synchronous belt transmission.

Parameters Symbol Values

Power P 2.262 kW
Torque T 216.005 N m
Speed n 100.000 rpm
Efficiency η 0.956 ul
Tension factor k1 1.300 ul
Belt Speed V 0.851 mps
Belt flex frequency fb 1.243 Hz
Effective pull Fp 2657.543 N
Centrifugal force Fc 0.841 N
Belt installation tension Ft 551.690 N
Maximum tension in belt span F tmax 883.882 N

3.3 Planetary Gear Transmission

The speed reducer uses a planetary gear system (Fig. 8) instead of a straight-line adjust-
ment mechanism when climbing stairs. The planetary gear system is responsible for
adjusting the number of revolutions of the 2 active pulley axles of the rubber tracks to
balance the friction force between the tracks and the corners of stair steps to help the
wheelchair climb stairs safely. In terms of transmission, a pair of planetary gears and
active bevel gears correspond to the Bevel Gear transmission. Also according to ISO for
the planetary gear transmission formulas [16–18], the calculation results are shown in
Table 3.
Because 2 bevel gears mounted on a semi-axis use the same pair of planetary gears
and 2 planetary gears have the same number of revolutions and torques, here only one
pair (a planetary gear and an active bevel gear) is counted and then the other side is
inferred. Since the power is divided among the two semi-axles, each active pulley has a
power of 2,762 (kW).
 A Study of Kinematics, Dynamics of the Stair Climbing Wheelchairs 91

Equations [16–18] as well as the application of software to calculate the parameters

of the planetary gear system are used to obtain the same results as Table 3.

Table 3. Calculation results of planetary gear transmission system.

Parameters Symbol Gear 1 Gear 2

Power P 2.262 kW 2.217 kW
Speed n 100.00 rpm 70.18 rpm
Torque T 216.005 N m 301.651 N m
Efficiency η 0.980 ul
Tangential force Ft 1890.540 N
Normal force Fn 2011.871 N
Circumferential speed v 1.196 mps
Resonance speed nE1 m

4 Conclusion

When climbing the stairs, wheelchair components from the engine to the pulley were
analyzed and modeled. Based on that, a wheelchair simulator was built on the Mat-
lab’s Simulink tool. The simulation results show that this model can be used to under-
stand the response of wheelchair components when running in different conditions.
From the wheelchair simulator with initial parameters, kinematics and dynamics of the
wheelchair’s transmissions were calculated. The wheelchair simulator is also used to
assist in the design of controls for the inputs to automate the stair climbing wheelchair.
The stair climbing transmission cluster model of the electric wheelchair is designed as a
removable cluster that can be easily disassembled with the chassis. With the simulation
tool, virtual testing shows that the transmission cluster works well on flat roads, rough
roads or uphill with an inclining angle of less than 15° and works very well when going
up stairs. Different controllers can be tested on this model to reduce design time as well
as avoid possible risks on the real model.

References
1. Das, G., Manohar, P.A.: Design of advanced multi desire wheel-chair. Int. J. Innov. Eng.
Technol. (IJIET) 7(4) (2016)
2. Quaglia, G., Franco, W., Nisi, M.: Kinematic analysis of an electric stair-climbing wheelchair.
Ing. Univ. 21(1), 27–48 (2017)
3. Kumark, V., Nayakam, A., Garamoti, A.: Design of stair climbing wheelchair. Int. Res. J.
Eng. Technol. (IRJET) 07(05) (2020)
92 N. H. Hiep et al.

4. Jorge, A.A., Riascos, L.A.M., Miyagi, P.E.: Modelling and Controlling of a Hybrid Motorized
Wheelchair on Flat and Inclined Surfaces. In: 2019 6th International Conference on Control,
Decision and Information Technologies (CoDIT’19). Senior Member, IEEE, Paris, France
(2019)
5. Jiang, T.-C., Yin, S.-H., Tanaka, E.: Wheelchair able to assist the elderly to move on stairs
and stand up. In: 2019 58th Annual Conference of the Society of Instrument and Control
Engineers of Japan (SICE) September 10–13, Hiroshima, Japan (2019)
6. Information on the revolutionary wheelchair rising to a challenge. BBC Future
7. Information on The wheelchair of the future—made in Switzerland. House of Switzerland
8. Ky, L.H.: Research on design and manufacture of the stair climbing electric wheelchairs.
Vietnam J. Mech., Hanoi, no. 3 (2017)
9. Ky, L.H.: Survey of kinematics and dynamics of the stair climbing electric wheelchairs.
Vietnam J. Mech., Hanoi, no 3 (2017)
10. Chat, T., Uyen, L.V.: Calculation and Design of Mechanical Drive Systems, vol. 1, 2. Vietnam
Education Publishing House, Hanoi (2009)
11. Hiep, N.T., Lam, N.V.: Design of Machine Parts. Vietnam Education Publishing House, Hanoi
(1999)
12. Information on http://edu.levitas.net/Tutorials/Matlab (2018)
13. Information on https://www.mathworks.com/products/simulink-online.html
14. Shabana, A.A.: Dynamic of Multybody Systems. Wiley Publication, Hoboken (2011)
15. Information on https://www.autodesk.com/products/inventor (2020)
16. Künne, B.: Köhler/Rögnitz, Maschinenteile 1. Springer, Berlin (2007)
17. Hütter, D.I.: Des Ingenieuvs Taschenbuch. Springer, Berlin (2007)
18. Dubbel, H., et al. (eds.): Taschenbuch für den Maschinenbau. Springer, Heidelberg (1914).
https://doi.org/10.1007/978-3-642-64925-7
 A Study of Polygonal Holes Machining Using
Attachment

Anh-Tuan Dang1 and Dinh-Ngoc Nguyen2(B)

1 Department of Mechanical Engineering, National Kaohsiung University of Science and
Technology Kaohsiung, Kaohsiung, Taiwan
2 Faculty of International Training, Thai Nguyen University of Technology, Thai Nguyen,
Vietnam
ngocnd@tnut.edu.vn

Abstract. Despite the applicability in machining processes, current polygonal

holes are manufactured by stamping but not by cutting. This will raise the required
components’ price and influence the machining procedure for mechanical prod-
ucts. This paper proposes a concept method for machining parts with polygonal
cross-section holes by cutting. Based on the motion of gear transmission in con-
structing Epitrochoid curves, a unique mechanism to control the position of straight
groove drills bit to machines polygons’ holes is presented. Numerical calculations
have been built to allow the detection of optimized machining parameters. Results
from the study show that polygons holes machined from the new method can be
applied in machining multiple types of polygonal without replacing cutting bits.

Keywords: Cutting · Epitrochoid curve · Polygonal hole · Attachment · Gear

transmission

1 Introduction
For producing non-circular sockets, metal forming is the traditional process to be used.
Due to this machining limitation, polygonal socket structures are not popularly applied
for manufacturing. Polygonal drilling originated from Watt’s idea in 1916 when rotating
a Reuleaux triangle within a square [1]. This method uses special jigs to guide drill
movements in holders to form square holes. Although providing practical values, this
device is not using a mathematically precise solution to the square-drill problem, which
leads to the polygonal’s sides not being straight and the movement path of the drill not
circle, which limits the machinability of the device (Fig. 1).
Based on Watt’s model, anonymous research modified the shape of holders and
placed the cutting bit at the center of the non-Reuleaux arc to produce perfect square
shapes. This method can create an exact polygonal with desirable properties of cutting
bits was then modeled by Bryant and Sangwin, who then introduced a physical device
using Oldham coupling [2].
Along with the studies on attachment design for square hole cutters, some researchers
work on extending the machining characteristics to increase the number of faces in

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 93–98, 2023.
https://doi.org/10.1007/978-3-031-22200-9_10
94 A.-T. Dang and D.-N. Nguyen

Fig. 1. Design of Watt using Reuleaux triangle for cutting square hole [1].

polygon sockets. Specifically, to make a hexagonal hole, Cox [3] recommended using
a pentagonal drill with the drill bits cutting all edges simultaneously. The research also
states that drilling an n-sides polygonal hole is required to manufacture a corresponding
n-1 side drill bit. However, studies of tool paths also point out that holes different in
sides and dimensions need specific attachment and drill bit, making it difficult for mass
production.
The paper presents a method to machine holes with polygonal cross-section by
drilling using a single cutter based on hypotrochoid curve construction. By changing
some parts in the attachment, the proposed model can be used for cutting polygonal
differences in dimension shape while ensuring the tolerance and flatness of sides.

2 Theoretical Method

An Epitrochoid is a roulette traced by point M attached to a circle c2 of radius R2 rolling

around a fixed circle c1 of radius R1 , with the given distance L1 = R1 + R2 and L2
= O2 M. Since the circles are circumscribed, movement of the two circles are related,
meaning when circle c1 is fixed and O1 O2 rotates with an angle θ1 , circle c2 will rotate
with a corresponding angle θ2 , and the positive ion for point M can be presented as
(Fig. 2):
→ −
− →
OM = L1 + L2 (1)

Or

OM = L1 .ejθ1 + L2 .ejθ2 (2)

Project into x- and y- axes of the world coordinate system:


xM = L1 .cosθ1 + L2 . cos θ2
(3)
yM = L1 .sinθ1 + L2 . sin θ2

Surveying different values of L1 , L2 , and the radii ratio between the circles shows
that when L2 reaches a large enough distance and ratio i = -α2 /α1 is integer, the resulting
Epitrochoid curve will have a polygonal shape (see Fig. 3).
 A Study of Polygonal Holes Machining Using Attachment 95

y c2

c1 O2

θ2
L2
M
L1

θ1
x
O1

Fig. 2. An example of the Epitrochoid curve [4].

Fig. 3. The constructed curves with different value of i and L2 .

Equation (3) can be rewritten as:

⎧    
⎪ α α
⎪
⎨ x = L1 .cosα + L2 . cos − n − 1 = L1 .cosα + L2 . cos n − 1
    (4)
⎪
⎪ α α
⎩ y = L1 sinα + L2 . sin − = L1 sinα − L2 . sin
n−1 n−1
where α is the rotation angle of L1 , and n is the side of polygon.
To calculate the appropriate data for constructing n-side polygon,
the
trajectory of
2π(n−1)
point M is analyzed, with the surveyed angle α in the range of 0; n .
Flatness of polygonal sides is then determined by the concavity e by following
equation:
Dmax − Dmin
e= (5)
L
π(n−2)
with D is distance from point M (x, y) to the corresponding line y = tan 2n x:


 π(n−2) 
tan 2n x − y
D = 
2 (6)
tan π(n−2)
2n +1

and L is the range between peaks in the surveyed curve.

During adjusting of parameters, there is a circumstance of L1 where the edge of the
polygonal profile is convex at the middle of the survey curve (Dmin is not in the middle
96 A.-T. Dang and D.-N. Nguyen

but the starting and ending points of the curve). Respectively for this case, the concavity
e of polygon’s sides can not determinated. Detail of these cases are presented in Fig. 4:

10 10
9
Dmax

Distance D (mm)
7.5 7.5
6

5
Dmin 5
3
L
2.5 2.5
0

y (mm)
y (mm)

0 3 6 9 12 15
0 Length (mm) 0

-2.5 9 -2.5
Dmax
Distance D (mm)
-5 6 -5
Compared_line Compared_line
-7.5 3 Dmin -7.5
Polygonal locus Polygonal locus
Survey path Survey path
-10 0 -10
-7.5 -5 -2.5 0 2.5 5 7.5 10 12.5 0 3 6 9 12 15 -7.5 -5 -2.5 0 2.5 5 7.5 10 12.5
x (mm) Length (mm) x (mm)

Fig. 4. Extracting a part of the polygonal locus for flatness analyzing.

3 The Constructing Model

We can see from the concept model that if only circles c1 and c2 are used, determining
the transmission ratio between links L1 and L2 is difficult. Not only due to the limitation
of the structure but also the selection of suitable parts to meet the transmission ratio
from them. In addition, because the hole to be machined is small, the attachment layout
is also bounded by the cutting space and the shape of the cutting tool.
Thus, in the new setup, we use a planetary gear set including four gears to control the
movement of the cutting tool and a D-bit cutter mounted in the holder with adjustable
distances L1 and L2 to change the movement of the cutting profile, as illustrated in Fig. 5.

Fig. 5. Schematic diagram of the mechanism.

 A Study of Polygonal Holes Machining Using Attachment 97

The cutting velocity also extracted for analysys, using Eq. (2):

VM = Vx2 + Vy2 (7)

With:
⎧  
⎪ ω1 α
⎪
⎨ x
V = −L ω
1 1 sin α − L2 sin
n−1 n−1
  (8)
⎪
⎪ ω α
⎩ Vy = L1 ω1 cosα − L2 1 cos
n−1 n−1
This value reaches the maximum at the middle of the sides and minimum at the
corners of the polygon, as illustrated in Fig. 6.

Fig. 6. Velocity diagram of point M in the polygonal path.

Several sample dimension sets are selected to analyze the influence of initial
parameters on the characteristic of the constructed polygon, as presented in Table 1.

Table 1. Different input and properties of the obtained polygons.

n L1 L2 Dmin L e Vmax //ω Vmin //ω

3 2 8 6 - n/a 6.00 2.00
2.2 8 5.8 10.00 0.07 6.20 1.80
2.6 8 5.4 15.34 0.25 6.60 1.40
4 0.9 10 9.1 - n/a 4.23 2.43
1.2 10 8.8 6.28 0.14 4.53 2.13
1.8 10 8.2 14.29 0.03 5.13 1.53
5 0.7 12 11.3 - n/a 3.70 2.30
0.9 12 11.1 7.85 0.09 3.90 2.10
1.2 12 10.8 11.97 0.03 4.20 1.80
n/a: Not available.
98 A.-T. Dang and D.-N. Nguyen

4 Conclusion
The proposed model proves we can use attachment in drilling for machining polygons
holes with initial requirements. The input parameters (L1 , L2 , i) are investigated to
evaluate their influence on the profile of the generated polygon (geometry variations,
flatness tolerance, convexity). Some of the comments obtained from the study include:

1. Changing the intermediate gears in the transmission (changing the gear ratio i) allows
the generation of polygons with different numbers of edges, making manufacturing
easy and overcoming the shortcomings of previous research, which requires precise
multiple groove drill bit.
2. Adjusting the dimensions of L1 and L2 can change the shape of drilled holes,
increasing the flatness of the edge or reducing the radius at the corners.
3. Mechanical parts with polygonal structures, such as socket heads, toys, and even
non-metallic parts, can be produced using this method instead of pressing or forging,
allowing improved efficiency and flexibility of the manufacturing process.

Acknowledgments. A special thanks to the Thai Nguyen University of Technology for funding
this research.

References
1. WATTS H. J.: Drill or boring member. US patent Appication US1241176A (1916)
2. Bryant, J., Sangwin, C.: How Round Is Your Circle? Princeton University Press, Princeton, NJ
(2008)
3. Cox, B., Wagon, S.: Drilling for Polygons. Am. Math. Mon. 119(4), 300–312 (2012)
4. Tsiotras, P., Reyes Castro, L.I.: The artistic geometry of consensus protocols. Springer, In
Controls and Art (2014)
 Study on Multi-criteria Decision Making in CBN
Grinding SKD11 Tool Steel

Bui Thanh Danh1 , Tran Huu Danh2 , Vu Van Khoa3 , Pham Duc Lam4 ,
Nguyen Hong Linh5 , Nguyen Van Tung6 , and Bui Thanh Hien6(B)
1 University of Transport and Communications, Ho Chi Minh, Vietnam
2 Vinh Long University of Technology Education, Vinh Long, Vietnam
3 National Research Institute of Mechanical Engineering, Durgapur, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh, Vietnam
5 Electric Power University, Ho Chi Minh, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

buithanhhienkc@tnut.edu.vn

Abstract. This paper presents the results of a multi-criteria decision-making

(MCDM) study when CBN grinding SKD11 tool steel. In this study, the MOORA
method was applied to solve the MCDM problem and the Entropy method was
used to determine the weight of criteria. Two criteria including the surface rough-
ness (RS) and the material removal speed (MRS) were selected for the problem.
In addition, an experiment with design L18 (61 + 33 ) using Taguchi method was
conducted. Four input parameters including the depth of dressing cut aed (mm),
the wheel speed Rpm (rpm), the feed rate Fe (mm/min), and the wheel diameter d
(mm) were investigated. According to the study’s findings, the best experimental
setup was proposed.

Keywords: CBN Grinding · MCDM · MOORA method · Surface roughness ·

Material removal speed · SKD11

1 Introduction

In mechanical machining processes, selecting the best or the most reasonable solution
among many given options is a very common and complex problem. Especially this
problem needs to be solved when the criteria set out in conflict. The requirement for
increasing the material removal rate (MMR) leads to the need to increase the feed rate
and increase the depth of cut. As a result, the surface roughness (SR) will increase. In
other words, the requirements or criteria for SR and MMR are contradictory. To solve
the above problem, the multi-criteria decision making (MCDM) problem is often used.
There have been a number of studies to date on the use of MCDM methods to
determine the best solution in mechanical processing in general and the grinding process
in particular. In [1] used the MARCOS method to MCDM for the surface grinding
of 90CrSi steel by CBN grinding wheel. In this study, 5 criteria including SR, the
vibration of the grinding wheel shaft in the three directions Ax, Ay, and Az and material

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 99–105, 2023.
https://doi.org/10.1007/978-3-031-22200-9_11
100 B. T. Danh et al.

removal yield (Q) were selected for MCDM problem to get the best option for the
Input process parameters. In [2] conducted a study on grinding wheel abrasive material
selection using fuzzy TOPSIS method. In this work, eight grinding wheel abrasives were
evaluated based on seven criteria. In [3], the results of an MCDM problem in the dressing
process for internal grinding were presented. Six input parameters were investigated in
this study: Fine dressing depth and time, coarse dressing depth and time, non-feeding
dressing, and dressing feed rate. Furthermore, four MCDM methods, including TOPSIS,
MARCOS, EAMR, and MAIRCA, were used for the MCDM to obtain the minimum SR
and maximum MRR at the same time. In [4], a study on MCDM when external cylindrical
grinding of 65G steel was presented. Two MCDM methods consist of WASPAS and
PIV methods were used for the investigation. The results of an MCDM study on surface
grinding SKD11 tool steel were reported in [5]. The MCDM problem was solved using
MOORA and COPRAS techniques in this study. The MARCOS method was applied to
solve MCDM problem in turning, milling, and gridning process [6]. The comparison of
MAIRCA, EAMR, MARCOS and TOPSIS methods when MCDM in turning process
was done in [7]. In [8] compared the MARCOS, TOPSIS, and MAIRCA methods when
MCDM for PMEDM process. In [9] presented the application of the TOPSIS and PIV
methods in solving MCDM problems when hard turning.
The results of an MCDM study in CBN grinding on CNC milling machines are
presented in this paper. In the study, SR and MRS were chosen as criteria. The MOORA
method was used as the MCDM method, and the Entropy method was used to find
the creation weights. In addition, four input parameters were taken into account: aed ,
Rpm , Fe , and d. The MCDM problem was solved, and the best experimental setup was
proposed.

2 Methodology

2.1 Method for Multi-criteria Decision Making

In this work, the MOORA method was used to solve the MCDM problem.
The following steps must be taken in order to use this method [10]:
Step 1: Making the initial decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X =⎢
⎢ r1 · · · rij · · · rin ⎥
⎥ (1)
⎢ . . . . . ⎥
⎣ .. . . .. . . .. ⎦
rm1 · · · rmj · · · rmn

where, m is the number of the alternative; n is the criteria number.

Step 2: Calculate the normalized matrix by:
rij
nij = (2)
maxrij
 Study on Multi-criteria Decision Making 101

minrij
nij = (3)
rij
Equation (2) is used for surface roughness criterion, and Eq. (3) is used for material
removal speed criation. 
Step 3: Create a normalized decision making matrix Xij m×n :
yij
Xij =
 (4)
m 2
i=1 yij

Step 4: Compute the decision-making matrices after normalizing the weights by:

Wij = wj × yij (5)

Step 5: Calculate Pi and Ri by:

1 
Pi = Wij (6)
|B|
j∈B
1 
Ri = Wij (7)
|NB|
j∈NB

In which B and NB represent the number of the criterion the larger the better and
the number of the criterion the smaller the better, respectively.
Step 6: Calculating Qi by:

Qi = Pi − Ri (8)

Step 7: Rank the alternatives using the principle that the alternative with the highest
Qi is the best.

2.2 Method for Calculation of the Weight of Criteria

In this work, the Entropy method was used to calculate the weights of the criteria. The
steps outlined below can be used to put this method into action [11].
Step 1: Determine the indicator’s normalized values by the following Equation:
xij
pij = m (9)
m+ 2
i=1 xij

Step 2: Estimate the Entropy for each indicator by:

m    m   m 
mej = − pij × ln pij − 1 − pij × ln 1 − pij (10)
i=1 i=1 i=1

Step 3: Find the weight of each indicator:

1 − mej
wj = m  (11)
j=1 1 − mej
102 B. T. Danh et al.

3 Experimental Setup
An experiment was carried out to determine the best experimental setup for obtaining the
lowest SR and highest MRS when grinding with a CBN wheel on a CNC milling machine.
Figure 1 shows the experimental setup, which includes a CNC milling machine (Model
M-V50C-Japan), a CBN grinding wheel, CBN wheel dresser equipment (V-TDM-2
type), SKD11 workpiece samples, and a roughness meter (SURFTEST SV-3100). The
input parameters and their levels are shown in Table 1. In addition, the Taguchi method
was used with design L18 (61 + 33 ). As a result, the investigation required 18 test runs.
Table 2 displays the experimental matrix as well as the RS and MRS values.

Table 1. Input parameters and their levels.

No Input factors Level

1 2 3 4 5 6
1 Depth of dressing cut (mm) 0.005 0.01 0.015 0.02 0.025 0.03
2 Revolution of spindle speed (rpm) 4000 4500 5000 - - -
3 Feed rate (mm/min) 2000 2500 3000 - - -
4 Wheel diameter (mm) 100 125 150 - - -

Fig. 1. Experimental setup [12] a Dressing setup; b Grinding setup; c Grinding schema

4 Determining the Best Alternative When CBN Grinding SKD11

Tool Steel

This section discribes finding the best experimental setup when CBN grinding SKD11
tool steel using the MOORA method for MCDM problem and the Entropy method for
calculation the criteria weights.
 Study on Multi-criteria Decision Making 103

Table 2. Experimental plan and output results.

No Input parameters Output results

aed Rpm Fe d Ra (µm) MRS (g/h)
1 0.005 4000 2000 100 0.5640 2.1933
2 0.005 4500 2500 125 0.5140 21.4465
3 0.005 5000 3000 150 0.2660 0.9161
4 0.01 4000 2000 125 0.3950 3.6771
5 0.01 4500 2500 150 0.3613 1.2421
6 0.01 5000 3000 100 0.3993 5.7262
7 0.015 4000 2500 100 0.3997 5.9976
8 0.015 4500 3000 125 0.2820 3.2727
9 0.015 5000 2000 150 0.2037 3.3251
10 0.02 4000 3000 150 0.7403 3.7553
11 0.02 4500 2000 100 0.5997 4.7941
12 0.02 5000 2500 125 0.3277 20.9084
13 0.025 4000 2500 150 0.6490 3.4458
14 0.025 4500 3000 100 0.7457 7.8089
15 0.025 5000 2000 125 0.2737 5.0633
16 0.03 4000 3000 125 0.8177 4.0054
17 0.03 4500 2000 150 0.4487 3.6447
18 0.03 5000 2500 100 0.6483 6.5775

4.1 Calculating the Weights for the Criteria

The weights of criteria are determined using the Entropy method as follows (see
Sect. 2.2): The normalized values pij are calculated using Eq. (9). Determined the Entropy
value for each indicator mej using Eq. (10). Finally, using Eq. (11), compute the wj weight
of the criteria. The Ra and MRS weights were found as 0.5485 and 0.4515, respectively.

4.2 Finding the Best Experimental Setup Using MOORA Method

The steps for MCDM using the MOORA method are as described in Sect. 2.1. Using
Eqs. (2) or (3) to compute the normalized matrix. The normalized decision making matrix
is then calculated using Eq. (4). After that, calculate the decision-making matrices using
Eq. (5). Afterward, using Eq. (6), compute Pi and Ri (7). Then, using the formula,
determine Qi (8). Finally, order the alternatives so that the one with the lowest Qi is the
best. Table 3 displays the results of some parameters as well as the alternative rankings.
From Table 3, option 12 is the best alternative. This is because it has the highest value
of Qi (Qi = 0.189). Therefore, the optimal solution includes the following parameters:
104 B. T. Danh et al.

Table 3. Several calculated results and ranking of alternatives.

Trial Input Data Xij Wij Ri Pi Qi Rank

Ra MRR Ra MRR Ra MRR
1 0.564 2.193 0.278 0.064 0.152 0.029 0.152 0.029 -0.123 15
2 0.514 21.446 0.253 0.630 0.139 0.284 0.139 0.284 0.145 2
3 0.266 0.916 0.131 0.027 0.072 0.012 0.072 0.012 -0.060 9
4 0.395 3.677 0.195 0.108 0.107 0.049 0.107 0.049 -0.058 8
5 0.361 1.242 0.178 0.036 0.098 0.016 0.098 0.016 -0.081 11
6 0.399 5.726 0.197 0.168 0.108 0.076 0.108 0.076 -0.032 6
7 0.400 5.998 0.197 0.176 0.108 0.080 0.108 0.080 -0.029 5
8 0.282 3.273 0.139 0.096 0.076 0.043 0.076 0.043 -0.033 7
9 0.204 3.325 0.100 0.098 0.055 0.044 0.055 0.044 -0.011 4
10 0.740 3.755 0.365 0.110 0.200 0.050 0.200 0.050 -0.150 17
11 0.600 4.794 0.296 0.141 0.162 0.064 0.162 0.064 -0.099 14
12 0.328 20.908 0.162 0.614 0.089 0.277 0.089 0.277 0.189 1
13 0.649 3.446 0.320 0.101 0.175 0.046 0.175 0.046 -0.130 16
14 0.746 7.809 0.368 0.229 0.202 0.104 0.202 0.104 -0.098 13
15 0.274 5.063 0.135 0.149 0.074 0.067 0.074 0.067 -0.007 3
16 0.818 4.005 0.403 0.118 0.221 0.053 0.221 0.053 -0.168 18
17 0.449 3.645 0.221 0.107 0.121 0.048 0.121 0.048 -0.073 10
18 0.648 6.578 0.320 0.193 0.175 0.087 0.175 0.087 -0.088 12

The dressing cut depth aed = 0.02 (mm); the wheel revolution Rpm = 5000 (rpm); the
feed rate Fe = 2500 (mm/min.); and the wheel diameter d = 125 (mm).

5 Conclusions

In this paper the MOORA method was applied to find the best experimental setup when
CBN grinding SKD11 tool steel on CNC machines to get the minimum SR and maximum
MRS at the same time. Besides, the weights of the criteria were determined using the
Entropy method. From the results of the study, it was found that the alternative 12 is the
best solution as it has the highest value of Qi . Experimental run No. 12 demonstrated the
best performance feature among the 18 trials, with maximum value of Qi (Qi = 0.189).
The best experimental setup was proposed as follows: aed = 0.02 (mm); Rpm = 5000
(rpm); Fe = 2500 (mm/min); and d = 125 (mm).

Acknowledgment. Thai Nguyen University of Technology provided support for this research.
 Study on Multi-criteria Decision Making 105

References
1. Thinh, H.X., Trung, D.D.: A research on application of the measurement of alternatives and
ranking according to compromise solution method for multi-criteria decision making in the
grinding process. EUREKA: Physics and Engineering, (2), p. 101–110, (2022)
2. Maity, S.R., Chakraborty, S.: Grinding wheel abrasive material selection using fuzzy TOPSIS
method. Mater. Manuf. Processes 28(4), 408–417 (2013)
3. Nguyen, H.-Q., et al.: A Comparative Study on Multi-Criteria Decision-Making in Dressing
Process for Internal Grinding. Machines 10(5), 303 (2022)
4. Trung, D.D., The combination of Taguchi–Entropy–WASPAS–PIV methods for multi-criteria
decision making when external cylindrical grinding of 65G steel. J. Mach. Eng., 21, (2021)
5. Nguyen, N.-T., Trung, D.D.: Combination of Taguchi method, Moora and Copras techniques
in multi-objective optimization of surface grinding process. J. Appl. Eng. Sci. 19(2), 390–398
(2021)
6. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev., 9 (2022)
7. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: A comparative study. Adv. Prod. Eng. &
Manag. 16(4), 443–456 (2021)
8. Nguyen, H.-Q., et al.: Multi-Criteria Decision Making in the PMEDM Process by Using
MARCOS, TOPSIS, and MAIRCA Methods. Appl. Sci. 12(8), 3720 (2022)
9. Trung, D.D., Application of TOPSIS and PIV methods for multi-criteria decision making in
hard turning process. Journal of Machine Engineering, 21, (2021)
10. Alinezhad, A.: New methods and applications in multiple attribute decision making (MADM).
2019: Springer
11. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Vietnam. J. Agric. Sci. 17(4), 32–2331 (2019)
12. Vu, N.-P., et al.: Optimization of grinding parameters for minimum grinding time when
grinding tablet punches by CBN wheel on CNC milling machine. Appl. Sci. 9(5), 957 (2019)
 Study on Multi-criteria Optimization in CBN
Grinding SKD11 Tool Steel

Trieu Quy Huy1 , Vu Van Khoa2 , Bui Thanh Danh3 , Tran Huu Danh4 ,
Hoang Xuan Tu5 , Vu Ngoc Pi5 , and Bui Thanh Hien5(B)
1 University of Economics—Technology for Industries, Hanoi, Vietnam
2 National Research Institute of Mechanical Engineering, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Vinh Long University of Technology Education, Vinh Long, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

buithanhhienkc@tnut.edu.vn

Abstract. A multi-objective optimization study for CBN grinding SKD11 tool

steel on a CNC milling machine is presented in this paper. In this work, two objec-
tive functions consisting of minimizing surface roughness (SR) and maximizing
material removal speed (MRS) were considered. Four input factors dressing cut
depth aed (mm), wheel speed Rpm (rpm), feed rate Fe (mm/min), and wheel diam-
eter dw (mm) were investigated. The effects of the variables on the multi-objective
function were assessed. Furthermore, an optimal set of input process parameters
was suggested.

Keywords: Grinding · CBN grinding · Multi-objective optimization · Surface

roughness · Material removal speed · SKD11

1 Introduction
As a cutting tool a CBN wheel made of cubic boron nitride is commonly used in CBN
grinding. CBN is one of the hardest on the market, only second to diamonds. The main
advantage of the material is its high abrasion resistance and thermal conductivity, which
allows it to retain its sharp cutting edges. As a result, CBN grinding is becoming more
popular in mechanical processing.
Thus far this year, various investigations on CBN grinding have been conducted. [1]
proposed a model of the thermal aspects of grinding that forecasts the critical removal rate
at which workpiece burn occurs. Specific grinding energy, grain thermal conductivity,
and wear flat area were all investigated. In [2] a new grinding method for screw rotors that
employs a cubic boron nitride (CBN) grinding wheel was described. Small electroplated
CBN grinding wheels were first used to grind screw rotors in this work. The authors in
[3] introduced a new water-based grinding fluid formulation capable of meeting both
the performance and environmental requirements for CBN grinding. In [4] examined
the grinding performance of AISI 4340 steel in cylindrical grinding using two types of
vitrified-bonded CBN grinding wheels, taking into account the results of the parameters

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 106–115, 2023.
https://doi.org/10.1007/978-3-031-22200-9_12
 Study on Multi-criteria Optimization in CBN 107

surface roughness, roundness error, micro hardness, acoustic emission, and average
power. In [5] an experimental study was conducted to assess the effect of CBN crystal
characteristics and plating thickness on the performance of electroplated CBN wheels. A
study on optimizing grinding factors to get the minimum grinding time in CBN grinding
tablet punches on a CNC milling machine was reported in [6]. The researchers in [7]
introduced an investigation study on the wear of CBN wheels and the influence of the
grinding process on the wheel topography and the grinding behavior. The effects of
compressed cold air on spindle shaft materials (SCM4 and SCM21) when cylindrically
ground with CBN wheels was presented in [8]. The effect of vitrified bond structure on
the radial wear of CBN wheels was investigated in [9, [10] introduced the application of
several types of cutting fluid in plunge cylindrical grinding with conventional and CBN
grinding wheels. The determination of optimal input parameters was also performed for
external grinding [11, 12], surface grinding [13, 14] and internal grinding [15].
In practice, CBN wheels are commonly used in cylindrical external grinding, internal
grinding, surface grinding or on specialized grinding machines. In this study, CBN
grinding process was applied on a CNC milling machine. The processing SKD11 tool
steel was investigated in a multi-target optimization work with two criteria including
SR and MRS. The influence of the input parameters on the multi-objective function was
evaluated. Additionally, the optimal input process factors were recommended.

2 Experimental Work

Fig. 1. Experimental setup.

An experiment was conducted to solve the multi-objective optimization problem

in order to obtain the lowest SR and highest MRS. Figure 1 depicts the experimental
setup, which includes a CNC milling machine (Model M-V50C—Japan), CBN grinding
wheels, SKD11 tool steel samples, and a roughness meter (SURFTEST SV-3100). The
input parameters and their levels are displayed in Table 1. Besides, the Taguchi method
with design L18 (61 + 33 ) was also employed. As a result, 18 test runs were required for
108 T. Q. Huy et al.

the investigation. The experimental matrix, as well as the output results (RS and MRS)
are shown in Table 2.

Table 1. Input parameters and their levels.

No Input factors Level

1 2 3 4 5 6
1 Depth of dressing cut (mm) 0.005 0.01 0.015 0.02 0.025 0.03
2 Revolution of spindle speed (rpm) 4000 4500 5000 – – –
3 Feed rate (mm/min) 2000 2500 3000 – – –
4 Wheel diameter (mm) 100 125 150 – – –

Table 2. Experimental matrix and output results.

No Input factors Output results

aed Rpm Fe d Ra (μm) MRS (g/h)
1 0.005 4000 2000 100 0.5640 2.1933
2 0.005 4500 2500 125 0.5140 21.4465
3 0.005 5000 3000 150 0.2660 0.9161
4 0.01 4000 2000 125 0.3950 3.6771
5 0.01 4500 2500 150 0.3613 1.2421
6 0.01 5000 3000 100 0.3993 5.7262
…
17 0.03 4500 2000 150 0.4487 3.6447
18 0.03 5000 2500 100 0.6483 6.5775

3 Multi-objective Optimization

For multi-objective optimization, the gray relational analysis method combined with
Taguchi was utilized. In this problem, two objectives including minimum surface rough-
ness (Ra) and maximum material removal rate (MRS) have been investigated. The S/N
index of Ra and MRS is calculated according to the following formulas:
For the minimum SR:

1 2
n
S/N = −10log 10 ( yi ) (1)
n
i=1
 Study on Multi-criteria Optimization in CBN 109

For the maximum MRS:

1 1
n
SN = −10log 10 ( ) (2)
n yi2
i=1

The calculated values of S/N ratios of the output results are described in Table 3.

Table 3. Calculated values of S/N of output results.

No Input parameters Ra (μm) MRS (g/h)

aed Rpm Fe d Mean S/N Mean S/N
1 0.005 4000 2000 100 0.5640 4.9734 2.1933 6.7907
2 0.005 4500 2500 125 0.5140 5.7805 21.4465 26.6271
3 0.005 5000 3000 150 0.2660 11.5008 0.9161 –0.7620
4 0.01 4000 2000 125 0.3950 8.0678 3.6771 11.3096
5 0.01 4500 2500 150 0.3613 8.8408 1.2421 1.8825
6 0.01 5000 3000 100 0.3993 7.9724 5.7262 15.1569
7 0.015 4000 2500 100 0.3997 7.9660 5.9976 15.5578
8 0.015 4500 3000 125 0.2820 10.9920 3.2727 10.2976
9 0.015 5000 2000 150 0.2037 13.8202 3.3251 10.4356
10 0.02 4000 3000 150 0.7403 2.6111 3.7553 11.4885
11 0.02 4500 2000 100 0.5997 4.4417 4.7941 13.6114
12 0.02 5000 2500 125 0.3277 9.6911 20.9084 26.4064
13 0.025 4000 2500 150 0.6490 3.7546 3.4458 10.7444
14 0.025 4500 3000 100 0.7457 2.5490 7.8089 17.8513
15 0.025 5000 2000 125 0.2737 11.2538 5.0633 14.0867
16 0.03 4000 3000 125 0.8177 1.7484 4.0054 12.0500
17 0.03 4500 2000 150 0.4487 6.9614 3.6447 11.2316
18 0.03 5000 2500 100 0.6483 3.7637 6.5775 16.3610

To analyze the gray relationships based on S/N values, it is necessary to convert them
to comparison series with dimensionless quantities or to normalize the data.
To examine gray relationships from the S/N values, they must be converted to values
which have dimensionless quantities or the data must be normalized. In this case, S/N
values are normalized by the value of Zij (0 ≤ Zij ≤ 1) which is determined by the
following Equation:
SN ij − min(SN ij , j = 1, 2, ..k)
Zij =   (3)
max SN ij , j = 1, 2, ..n − min(SN ij , j = 1, 2, ..n)
In which n is the total of test runs; yi the observed data.
110 T. Q. Huy et al.

The grey relation coefficient γ(k) can be found by:

min + ζ max
γ (k) = (4)
j (k) + ζ max

where, j(k) is the deviation of reference data; j (k) = Z0 (k)−Zj (k) and Z0 (k) is the
reference data or best data. min = minminZ0 (k) − Zj (k) minimum value of j (k).
∀j∈i ∀k
max = maxmaxZ0 (k) − Zj (k) maximum value of j (k). ζ is the distinguishing or
∀j∈i ∀k
identification coefficient; ζ = 0 ÷ 1 (In this case ζ = 0.5).
The gray relational grade is obtained by the following Equation:

1
m
γj = γij (5)
k
i=1

wherein, k is the number of objectives.

The calculated results of the gray relation values yi and mean gray relation value yi
of the experiments are shown in Table 4.

4 Results and Analysis

Theoretically, the higher the average gray relation value is, the better the harmony
between output parameters is. As a result, the multi-objective problem is considered as
a single problem, with the mean gray relation value as the output. The Taguchi method
is employed to assess the impact of technological parameters on the mean gray relation
value. The S/N value is calculated using formula 2 with the goal of increasing the mean
gray relation value as much as possible.
The influence of the input parameters on the mean gray relation value is analyzed by
ANOVA method and the results are shown in Table 5. From Table 5, it is recognized that
Rpm has the greatest influence on yi (35.75%); followed by the effects of d (21.09%),
aed (20.62%), and Fe (7.59%).
The influence order of the parameters on the gray relation value through ANOVA
analysis is described in Table 6. According to this result, the influence order of the input
parameters is Rpm , aed , d, and Fe , respectively.
Figure 2 expresses the influence of the input parameters on yi . From this Figure, it
can be seen that when aed increases from level 1 to level 2, yi decreases slightly. When
aed goes up to level 3, yi spikes. However, increasing aed to levels 4, 5 and 6 will cause
yi decrease. Rpm increases from level 1 to level 3, yi increases sharply. Rpm is also the
parameter that has the greatest influence on yi . The effect of Fe is as follows: yi expands
when Fe increases from level 1 to level 2. However, when Fe rises to level 3, yi declines.
The effect of d is similar to Fe: When d increases from level 1 to level 2, yi increases;
but when it increases from level 2 to level 3, yi drops.
When determining the set of optimal input factors, it is required to consider the effect
of the input parameters on the S/N ratio. As previously discussed, the set of parameters
with the highest yi value for each input factor is the most reasonable. As previously
stated, the greater the y value is, the larger the value of S/N will be. This implies that
 Study on Multi-criteria Optimization in CBN 111

Table 4. Calculated values of S/N, Zi, i (k), yi, and yi .

No S/N Zi i (k) Gray relation yi

value yi
Ra MRS Ra MRS
Reference value Ra MRS Ra MRS
1.000 1.000
1 4.9734 6.7907 0.267 0.276 0.733 0.724 0.406 0.408 0.407
2 5.7805 26.6271 0.334 1.000 0.666 0.000 0.429 1.000 0.714
3 11.5008 –0.7620 0.808 0.000 0.192 1.000 0.722 0.333 0.528
4 8.0678 11.3096 0.523 0.441 0.477 0.559 0.512 0.472 0.492
5 8.8408 1.8825 0.588 0.097 0.412 0.903 0.548 0.356 0.452
6 7.9724 15.1569 0.516 0.581 0.484 0.419 0.508 0.544 0.526
7 7.9660 15.5578 0.515 0.596 0.485 0.404 0.508 0.553 0.530
8 10.9920 10.2976 0.766 0.404 0.234 0.596 0.681 0.456 0.569
9 13.8202 10.4356 1.000 0.409 0.000 0.591 1.000 0.458 0.729
10 2.6111 11.4885 0.071 0.447 0.929 0.553 0.350 0.475 0.412
11 4.4417 13.6114 0.223 0.525 0.777 0.475 0.392 0.513 0.452
12 9.6911 26.4064 0.658 0.992 0.342 0.008 0.594 0.984 0.789
13 3.7546 10.7444 0.166 0.420 0.834 0.580 0.375 0.463 0.419
14 2.5490 17.8513 0.066 0.680 0.934 0.320 0.349 0.609 0.479
15 11.2538 14.0867 0.787 0.542 0.213 0.458 0.702 0.522 0.612
16 1.7484 12.0500 0.000 0.468 1.000 0.532 0.333 0.484 0.409
17 6.9614 11.2316 0.432 0.438 0.568 0.562 0.468 0.471 0.469
18 3.7637 16.3610 0.167 0.625 0.833 0.375 0.375 0.572 0.473

the noise effect has been reduced. Figure 3 depicts the effect of the input parameters on
the S/N values. This diagram also shows the optimum set of the input parameters for the
multi-objective function (Table 5).
The Minitab 19 can be used to predict the results of the objective functions using
the optimal set of parameters (Table 6). In this case, the predictive results are Ra≈1.273
(μm) and MRS≈12.9831 (g/h) (Table 7).
Evaluation of proposed model
The residual plots (Fig. 3) were used to evaluate the suitability of the experimental
model. The Figure shows that the experimental dots (the blue points) on the histogram
are very close to the normal distribution line (the red solid line), which indicates that the
deviation is very narrow. The error frequency Histogram graph reveals that the majority
of errors fall between -0.06 and 0.06. The experimental errors are distributed at random
112 T. Q. Huy et al.

Table 5. Effect of input factors on the gray relation values yi .

Analysis of variance for means

Source DF SeqSS AdjSS Adj MS F P C(%)
aed 5 0.04693 0.047 0.009 1.66 0.277 20.62
Rpm 2 0.08135 0.081 0041 7.18 0026 35.75
Fe 2 0.01728 0.017 0.009 1.52 0.292 7.59
d 2 0.048 0.048 0.024 4.23 0.071 21.09
Residual error 6 0.03401 0.034 0.006 14.94
Total 17 0.22758
Model summary
s R-Sq R-Sq(adj)
0.0753 85,05% 57.65%

Table 6. Influence order of input process parameters on yi .

Response table for means

Level aed Rpm Fe d
1 0.5498 0.4449 0.5269 0.4780
2 0.4901 0.5226 0.5630 0.5974
3 0.6093 0.6095 0.4871 0.5017
4 0,5512
5 0.5033
6 0,4505
Delta 0.1588 0.1646 0.0759 0.1195
Rank 2 4 3
Average of grey relational coefficient: Q.S26

across the other two graphs. This means that the built model is greatly influenced by the
input parameters chosen, but the order of experiments is unaffected (Fig. 5).
The Anderson-Darling technique was used to test the suitability of the experimental
model (Fig. 4). The model results show that the experimental points (blue dots) are all
within a 95% confidence interval defined by two limit lines. Furthermore, the P value of
0.09 is greater than 0.05, indicating that the empirical model used is appropriate.

5 Conclusions
This paper discussed the results of a study on the effect of process parameters on mate-
rial removal speed in CBN grinding SKD11 tool steel. Two objective functions were
 Study on Multi-criteria Optimization in CBN 113

Fig. 2. Effect of input factors on yi .

Fig. 3. Effect of input factors on S/N of y.

Table 7. Optimum input facors.

No Input factors Code Unit Level Optimum value

1 Depth of dressing cut aed mm 3 0.015
2 Rpm Rpm rpm 3 5000
3 Feed rate Fe mm/min 2 2500
4 Wheel diameter d mm 2 125

considered in this work: Minimizing surface roughness (SR) and maximizing material
removal speed (MRS). The dressing cut depth aed (mm), the wheel speed Rpm (rpm),
114 T. Q. Huy et al.

Fig. 4. Probability residual plots for S/N ratios.

Fig. 5. Probability plot of the fit of proposed model

the feed rate Fe (mm/min), and the wheel diameter dw were all investigated (mm). The
effects of these variables on the multi-objective function were explored. In addition, the
optimal input process parameters were proposed as follows: aed = 0.015 (mm); Rpm =
5000 (rpm); Fe = 2500 (mm/min.); and d = 125 (mm).

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

 Study on Multi-criteria Optimization in CBN 115

References
1. Lavine, A., Malkin, S., Jen, T.: Thermal aspects of grinding with CBN wheels. CIRP Ann.
38(1), 557–560 (1989)
2. Wei, J., Zhang, G.: A precision grinding method for screw rotors using CBN grinding wheel.
Int. J. Adv. Manuf. Technol. 48(5), 495–503 (2010)
3. Oliveira, J., Alves, S.M.: Development of environmentally friendly fluid for CBN grinding.
CIRP Ann. 55(1), 343–346 (2006)
4. Lopes, J.C., et al.: Effect of CBN grain friability in hardened steel plunge grinding. Int. J.
Adv. Manuf. Technol. 103(1–4), 1567–1577 (2019). https://doi.org/10.1007/s00170-019-036
54-w
5. Upadhyaya, R., Fiecoat, J.: Factors affecting grinding performance with electroplated CBN
wheels. CIRP Ann. 56(1), 339–342 (2007)
6. Vu, N.-P., et al.: Optimization of grinding parameters for minimum grinding time when
grinding tablet punches by CBN wheel on CNC milling machine. Appl. Sci. 9(5), 957 (2019)
7. Shi, Z. and S. Malkin, Wear of electroplated CBN grinding wheels. (2006)
8. Choi, H., Lee, S., Jeong, H.: A comparison of the cooling effects of compressed cold air and
coolant for cylindrical grinding with a CBN wheel. J. Mater. Process. Technol. 111(1–3),
265–268 (2001)
9. Herman, D., Krzos, J.: Influence of vitrified bond structure on radial wear of cBN grinding
wheels. J. Mater. Process. Technol. 209(14), 5377–5386 (2009)
10. Monici, R.D., et al.: Analysis of the different forms of application and types of cutting
fluid used in plunge cylindrical grinding using conventional and superabrasive CBN grinding
wheels. Int. J. Mach. Tools Manuf 46(2), 122–131 (2006)
11. Trung, D.: Influence of cutting parameters on surface roughness in grinding of 65G steel.
Tribol. Ind. 43(1), 167 (2021)
12. Le, H.-A., et al.: Determining the best dressing parameters for external cylindrical grinding
using MABAC method. Appl. Sci. 12(16), 8287 (2022)
13. Nguyen, N.-T.: Combination of Taguchi method, Moora and Copras techniques in multi-
objective optimization of surface grinding process. J. Appl. Eng. Sci. 19(2), 390–398 (2021)
14. Do Duc, T., et al., A research on multi-objective optimization of the grinding process using
segmented grinding wheel by Taguchi-Dear method. EUREKA: Physics and Engineering,
(1), p. 67–77, (2021)
15. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
 A Study on Optimization of Helical Worm
Gearboxes Based on Volume Function

Bui Thanh Danh1 , Le Hoang Anh2 , Tran Huu Danh2 , Tran Ngoc Huy Thinh3 ,
Trinh Kieu Tuan4 , Luong Viet Dung5 , and Tran Thi Phuong Thao5(B)
1 University of Transport and Communications, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
3 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
4 University of Economics - Technology for Industries, Hanoi, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

tranphuongthao@tnut.edu.vn

Abstract. This paper aims to determine the optimal gear ratios of a helical worm
gearbox in order to achieve the smallest gearbox volume. To solve this problem,
a simulation experiment is designed and carried out. In the experiment, four main
input factors including the total gearbox ratio, the coefficient of wheel face of stage
2, the allowable contact stress of stage 2, and the output torque were investigated.
The effect of input factors on optimal gear ratios was evaluated. A regression
formula for determining the optimal gear ratio has also been proposed.

Keywords: Helical worm gearbox · Gear ratio · Optimum gearbox design ·

Volume function

1 Introduction
The gearbox is a kind of common transmission device which is widely applied to trans-
mission systems. However, due to the number of mixed variables, the intricate design
and the gear geometry make the design process a complex task to handle, so it always
requires designing improvements. Consequently, how to optimize the transmission gear-
box becomes more and more important. So far, there have been many studies on the
optimal gearbox design [1–8]. The oldest method is the graph method [9, 10], whereby
the gear ratios are found based on the graph of the relationship between the component
ratios and the total gearbox ratio. Another method was proposed by Milou et al. [11],
accordingly, gear ratios are calculated based on the practical data, e.g. the data given by
the gearbox manufacturing factories. Modelling method is widely utilized to calculate
the optimum gear ratios. In this method, various objective functions are established such
as the minimum mass of gear [12], the minimum area of gearbox cross-section [13], the
minimum gearbox length [14], and the minimum total mass of the gearbox [15]. There
are also some other researching methods such as the one with optimization algorithm,
or finite element analysis method to individually optimize each part of the gearbox [16,
17].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 116–127, 2023.
https://doi.org/10.1007/978-3-031-22200-9_13
 A Study on Optimization of Helical Worm 117

The results of previous studies showed that the optimal gear ratios directly impact
the design of the optimal gearbox. However, up to now, there has been no research
on calculating the optimal gear ratios with volume objective function. Therefore, this
article presents a method to study volume optimization of the helical worm gearbox, with
the objective function is to optimize the transmission ratio u2. Accordingly, the main
design factors were chosen to investigate their dependences on the output response. The
Design of Experiment technique combination with Minitab®19 software is used to carry
out the experiment simulation tests. Simultaneously, a useful regression model will be
proposed based on the experiment data to evaluate the influence of input factors and
their interaction on the objective functions.

2 Methodology
2.1 Gearbox Volume Analysis
The volume of the gearbox Vgb is calculated by (Fig. 1):

L
A T1 A-A
A1 A1
S1 S2
SG

bW2
1.5SG

lS1

dS3
L1
1.5SG

10 10
22
dW bW1
H

1.2dS3
dW11
aW2

dS1

1.2dS1
aW1
H1

10 10
dW21
dW12

S3
dS2

SG SG
h

SG
5SG

1.5SG
A 1.5SG B
lS2
SG

B1 lS3

Fig. 1. Schema for determination of volume gearbox

Vgb = L · H · B + L1 · H · B (1)

where, L, H, L1 , and B1 can be estimated by:

L = L + 2SG (2)

L = dw22 + 20 (3)

SG = 0.005 · L + 4.5 [18] (4)

118 B. T. Danh et al.

H = dw22 /2 + aw2 + max(dw21 /2; dw12 /2) + 8.5 · SG (5)

L1 = bw1 + 4 · SG (6)

B = max(dw21 ; dw12 ) + 20 + 2 · SG (7)

In the above equations, dw11 , dw21 , dw21 , and dw22 are the pitch diameters of the
pinion and the gear of the worm and the helical gear sets. These diameters will be
determined in the bellow sub-sections.

2.2 Calculating Center Distance and Pitch Diameters of Helical Gear Drive

For the helical gear drive, the center distance aw1 is calculated by [19]:
  
aw1 = ka · (u1 + 1) · 3 T11 · kH β / [σH 1 ]2 · u1 · Xba (8)

where, kH β is the contacting load ratio for pitting resistance; for the helical gear set
kH β = 1.05 1.27 [19]. Therefore, it can be chosen kH β = 1.16; [σH 2 ] is the allowable
contact stress of the first stage (MPa); ka is the material coefficient; As the gear material
is steel, ka = 43 [19]; Xba is the coefficient of wheel face width of the helical gear set;
T11 is the torque on the pinion (N mm) which is determined by:
 
T11 = Tout / ug · ηw · ηhg · ηb3 (9)

In which, Tout is the output torque of the system

 (N mm); ηhg is the transmission
efficiency of the helical gear unit (ηhg = 0.96 0.98 [19]); ηb is the transmission effi-
ciency of a rolling bearingpair (ηh = 0.99/0.995 [19]); ηw is the transmission efficiency
of a worm set (ηw = 0.7 0.82). Choosing ηhg = 0.97; ηw = 0.76;, ηb = 0.992; and
substituting them into (9) gives:

T11 = 1.3896 · T out /ug (10)

2.3 Calculating Center Distance and Pitch Diameters of Worm Drive

The worm gear center distance aw2 (mm) is calculated by the following equation [20]:
 1/3
aw2 = Ka · KHV · KH β · T22 /[σH 2 ]2 (11)

In which, Ka is a coefficient; Ka = 610 [20]; KHV is the internal dynamic coefficient;

[19] and we can get KHV = 1.2; KH β is the load concentration coefficient; [20] and we
can get KH β = 1.2; T22 is the wheel torque (Nmm); T22 = Tout .
In the Eq. (11), [σH 2 ] is the allowable contact stress of the worm unit (N/mm2 );
[σH 2 ] is influenced by the wheel material. As the wheel material is tinless bronze or
 A Study on Optimization of Helical Worm 119

soft grey iron, [σH 2 ] can be determined by the following regression equation which was
found from the data in [19] (with the determination coefficient R2 = 0.9906):
[σH 2 ] = 5.0515 · vsl2 − 49.742 · vsl + 189.9 (12)
wherein, νsl is the slip velocity which is found by [19]:
 1/3
vsl = 0.0088 · P1 · u · n2w (13)

where, P1 is the power on the worm shaft; P1 can be found by:

 
P1 = T12 · nw / 9.66 · 106 (14)

In which, nw is the rotational speed of the worm; T12 is the worm torque (N mm):
 
T12 = Tout / uw · ηw · ηb2 (15)

Choosing ηw = 0.76, ηb = 0.992 as in the Sect. (2.2) and substituting them into (15)
gives:
T12 = 1.3371 · Tout /uw (16)
If the wheel material is tin bronze, [σH 2 ] is calculated by [19]:
[σH 2 ] = KHL · vsl · [σH 0 ] (17)
where, [σH 0 ] is the allowable contact stress when the stress change cycle is 107 :
  
[σH 0 ] = 0.7 0.9 · σt (18)
In which, σt is the tensile stress (N/mm2 ); the value of σt depends on the slip velocity
νsl: σt = 260 if νsl = 5/8; σt = 230 if νsl = 8/12 and σt = 285 if νsl = 8/25 [19].
KHL is the service life ratio which is determined as [19]:
 1/8
KHL = 107 /NHE (19)

where, NHE is the equivalent loading cycle number for the wheel teeth:
NHE = 60 · nwl · t (20)
In which, t is t the service lifetime (h); nwl is the rotational speed of the wheel
(rpm).
From above analysis, the optimization problem is defined as:
MinimizeCgb (21)
With the following constraints
1 ≤ u1 ≤ 9

8 ≤ u2 ≤ 80 (22)

aw1 + dw11 /2 + 2.5·S G < aw2

120 B. T. Danh et al.

3 Experimental Setup
A simulation experiment was conducted to investigate the influence of experimental
factors on the objective functions. The purpose of the investigation is to determine what
the main factors affecting the objective function or the survey process are; to evaluate
the influence degree of the parameters; to evaluate the influence of interaction between
parameters, and to propose a simple model describing the input-output relation. All input
parameters are chosen to investigate their impacts on the output response, the partial
optimum gear ratio u2. The input parameters and their levels are shown in Table 1. The
Minitab® 19 software is selected to develop the experimental plan and data analysis using
the 2-level full factorial design. In this work, 16 tests for the simulation experiment were
used. The output responses are presented in Table 2.

Table 1. Input factors

Factor Code Unit Low High

Total gearbox ratio ug – 60 110
Coefficient of wheel face width of stage 2 Xba – 0.2 0.25
Allowable contact stress of stage 2 AS 2 MPa 350 420
Output torque Tout Nm 8000 10000

4 Results Discussion
4.1 The Influence of Input Parameters and Their Interactions
The main influencing factors, the effects on transmission ratio u2 response, can be deter-
mined qualitatively through Main Effects Plot as shown in Fig. 2. In this graph, the effect
of each experimental variable is plotted independently on the same chart.
In Fig. 2, the graph of the effects of the variables is plotted in independent cells.
Considering the graph showing the effect of ug , we observe that when ug changes from
60 to 110, the target function u2 changes from 30.5934 to 55.2516. The slope of this
graph is (55.2516 − 30.5934)/2 = 12.3291. Similarly, the slope of u2 when Xba , AS2 ,
and Tout changes respectively: (42.9927 − 42.8523)/2 = 0.0702; (43.1118 − 42.7269)/2
= 0.19245, and (56.6667 − 29.1784)/2 = 13.74415. From a qualitative comparison, it
shows that the slope of the Tout graph is the largest, next is the graph of ug , AS2 , and
finally X ba .
Another way to evaluate the influence of input factors on u2 is to use the normal
plot of the standardized effects (Fig. 3) or the Pareto chart of the standardized effects
(Fig. 4). In Fig. 3, the graph is plotted for the objective function with significance level
α = 0.05. The most significant factors are represented in squares, including A (ug ), C
(AS 2 ), D (T out ) and quadratic interaction: AD (ug *T out ), CD (AS 2 *T out ). Accordingly,
variable D has the strongest significance, followed by A, AD, C, and finally CD.
 A Study on Optimization of Helical Worm 121

Table 2. Experimental plans and output responses

RunOrder ug Xba AS2 Tout u2

1 110 0.2 350 1000 37.801
2 60 0.2 350 1000 21.583
3 110 0.25 420 1000 36.424
4 110 0.2 350 8000 73.333
5 110 0.25 350 8000 73.333
6 60 0.25 350 8000 40.000
7 60 0.25 420 8000 40.000
8 60 0.2 350 8000 40.000
9 60 0.25 420 1000 20.833
10 60 0.2 420 8000 40.000
11 110 0.25 350 1000 37.543
12 110 0.2 420 8000 73.333
13 60 0.2 420 1000 20.979
14 110 0.2 420 1000 36.913
15 110 0.25 420 8000 73.333
16 60 0.25 350 1000 21.352

Fig. 2. Main effects plot of factors for the response of u2

In Fig. 4, Minitab R19 uses the significance level α to draw the limit line (Value is 2.6
on the horizontal axis) of the area that rejects the inversion hypothesis. The influential
values (standardized), are represented as horizontal bars. Factors that are to the right of
the limit line have a significant effect, and to the left of the limit line, have a weak effect.
The graph shows the order of influence of factors and interactions from high to low: D,
A, AD, C, and finally CD. In other ways, the interactions between the input parameters
on the response (u2 ) can be shown in Fig. 5. From this graph, it is noted that only the
above-mentioned interactions, AD (ug *T out ), CD (AS 2 *T out ), are significant to u2 .
122 B. T. Danh et al.

Fig. 3. Normal plot for u2

Fig. 4. Pareto chart of the standardized effects for u2

4.2 Determining Regression Model for Calculating u2

The regression model at the two-factor interaction level is proposed as shown in Table 3,
in which systems b0 , bx , bxx are listed by Minitab R19 in the Coef column; the names of
the elements, including experimental variables and interactions, are listed in the Term
column. After the factors and interactions are removed with weak or no influence on u2 ,
the results as shown in Table 4 are received.
The constant values in the equation are denoted as constants in the Term column of
Table 4. Values for the coefficients are listed in the column “Coef”. Column T represents
the value in the t-value distribution of the quantity to be considered; Column P lists the
value of the probability p-value when testing the statistical hypothesis for the possibility
that the coefficients are zero. A p-value is greater than the significance level α signals
that the existence of the corresponding coefficient is not statistically significant. In other
 A Study on Optimization of Helical Worm 123

Fig. 5. The interactions between input parameters on the response for u2

words, when p > α, it is possible to trust (1 − α)% to get that coefficient to zero. That
is, the influence of the respective component is negligible on the objective function.
From the p-value column, we have conclusions about the components that can affect
the objective function. First of all, the p-values of the parameters and their interactions
are both smaller than the significance level α = 0.05. This proves that the parameters
and their interactions have a strong influence on u2 . The last row of the table shows the
regression model evaluation parameters. The parameters R-Square and R-Square (adj)
are approximately 100% and 99.99%, proving that the model is well-matched with the
data. The relation equation between u2 and the main influence parameters can be written:

u2 = 7.212 + 0.27009 · ug − 0.01277 · AS 2 − 0.000901 · Tout + 0.00005 · ug · Tout

+ 0.00005 · ug (23)

4.3 Analysis of Variance—ANOVA

To quantitatively conclude the impact of each parameter and its interactions in responses,
Analysis of Variance is necessary. Table 5 reveals the Analysis of Variance in the case
of response, where the insignificant impacts are ignored. We can see that the main effect
A, C, D, AD and CD have very small values (p < 0.001), showing that the main effects
are statistically significant. In other words, the main influences and interactional effects
are both significant on the target function.

4.4 Validation of Proposed Model

The estimation of errors resulting from the difference between experiments and the
model of u2 is qualitatively described in Fig. 6. The Normal Probability Plot (Fig. 6.a)
shows the residuals of the distribution very close to the normal distribution. The Versus
Order graph (Fig. 6.b) shows the points of random distribution which does not follow
124 B. T. Danh et al.

Table 3. Information of the regression model

Table 4. Coded coefficients and model summary

any rules. It is found from Table 2 that the value of u2 is not affected by factors with other
rules except for survey parameters. Therefore, the relationship between u2 and survey
parameters represented in formula (23) has a pretty high reliability.
 A Study on Optimization of Helical Worm 125

Table 5. Analysis of variance for u2

5 Conclusions
The results of an optimization problem to find the optimum gear ratios of a helical
worm gearbox in order to get the smallest gearbox volume are presented in this paper. A
simulation experiment with full factorial design was designed and carried out to solve
this task. The effect of the main designing factors on optimal gear ratios of the gearbox
was also investigated. Furthermore, a regression model for calculating the best gear
ratios has been proposed. The proposed model was also found to be very consistent with
the experimental data.
126 B. T. Danh et al.

Fig. 6. Normal probability (a) and versus order plot (b) for u2

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Vu, N.P.: A new study on the optimal prediction of partial transmission ratios of three-step
helical gearboxes with second-step double gear-sets. Appl. Theory Mech. 2(11), 229–238
(2007)
2. Pi, V.N.: A study on optimal determination of partial transmission ratios of helical gearboxes
with second-step double gear-sets. Int. J. Mech. Mechatronics Eng. 2(1), 26–29 (2008)
3. Pi, V.N., Tuan, N.K.: Optimum determination of partial transmission ratios of three-step
helical gearboxes for getting minimum cross section dimension. J. Environ. Sci. Eng. A 5(11)
(2016). https://doi.org/10.17265/2162-5298/2016.11.004
4. Cam, N.T.H., Pi, V.N., Hong, T.T., Ky, L.H., Tung, L.A.: A study on calculation of optimum
gear ratios of a two-stage helical gearbox with second stage double gear sets. Int. J. Mech.
Prod. Eng. Res. Dev. 9(2), 613–620 (2018). https://doi.org/10.24247/ijmperdapr201961
 A Study on Optimization of Helical Worm 127

5. Pi, V.N., Tuan, N.K.: Determining optimum partial transmission ratios of mechanical driven
systems using a chain drive and a two-step bevel helical gearbox. Int. J. Mech. Eng. Robot.
Res. 8(5), 708–712 (2019). https://doi.org/10.18178/ijmerr.8.5.708-712
6. Pi, V.N., Thi Hong, T., Thi Phuong Thao, T., Khac Tuan, N., Hung, L.X., Anh Tung, L.:
Calculating optimum gear ratios of a two-stage helical reducer with first stage double gear
sets. In: IOP Conference Series: Materials Science and Engineering, vol. 542, no. 1. https://
doi.org/10.1088/1757-899X/542/1/012017
7. Thao, T.T.P., et al.: Determining optimum gear ratios of mechanical driven systems using
three stage bevel helical gearbox and chain drive. In: Sattler, K.-U., Nguyen, D.C., Vu, N.P.,
Tien Long, B., Puta, H. (eds.) ICERA 2019. LNNS, vol. 104, pp. 249–261. Springer, Cham
(2020). https://doi.org/10.1007/978-3-030-37497-6_29
8. Hung, L.X., et al.: Calculation of optimum gear ratios of mechanical driven systems using
two-stage helical gearbox with first stage double gear sets and chain drive. In: Sattler, K.-
U., Nguyen, D.C., Vu, N.P., Tien Long, B., Puta, H. (eds.) ICERA 2019. LNNS, vol. 104,
pp. 170–178. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-37497-6_20
9. Kudreavtev, V.N., Gierzaves, I.A., Glukharev, E.G.: Design and calculus of gearboxes (in
Russian) (1971)
10. Chat, T., Uyen, L.V.: Calculation of Mechanical Driven Systems. Education Publisher, Ha
noi (1996)
11. Milou, G., Dobre, G., Visa, F., Vitila, H.: Optimal design of two step gear units, regarding the
main parameters. VDI Berichte 1230, 227–244 (1996)
12. Phan, Q.T. Vu, N.P., Nguyen, D.B., Vu, Q.D.: Optimal calculation of total transmission ratio
of three-step helical gearboxes for minimum mass of gears. J. Sci. Technol. Tech. Eng. Univ.
170–178 (2006)
13. Pi, V.N.: A new method for optimal calculation of total transmission ratio of two step helical
gearboxes. In: The National conference on Engineering Mechanics (2001)
14. Pi, V.N.: A new study on optimal calculation of partial transmission ratios of two-step helical
gearboxes. Comput. Eng. 2(1), 162–165 (2008)
15. Vu, N.P.: Optimal determination of partial transmission ratios of three-step helical gearboxs
with second double gears (2008)
16. Xu, X., Chen, Z., Liu, Y., Dong, P., Liu, Y.: Enumerative optimization procedure for the gear
train optimization problem of a two-speed dedicated electric transmission. Energies 10(9)
(2017). https://doi.org/10.3390/en10091362
17. Liang, M., Hu, J., Li, S., Chen, Z.: Topology optimization of transmission gearbox under
multiple working loads. Adv. Mech. Eng. 10(11) (2018)
18. Romhild, I., Linke, H.: Gezielte Auslegung Von Zahnradgetrieben mit minimaler Masse auf
der Basis neuer Berechnungsverfahren. Konstr. 44, 229–236 (1992)
19. Chat, T., Van Uyen, L.: Design and calculation of Mechanical Transmissions Systems.
Educational Republishing House (2007)
20. Karl-Heinrich Grote, E.K.A.: Springer Handbook of Mechanical Engineering, vol. 47, no.
01. Springer (2009)
 A Study on Wearing and Tool Life of Shredding
Machine Cutters When Cutting Fabric
Industrial Waste

Trong Hai Nguyen and Ngoc Tuyen Bui(B)

Hanoi University of Science and Technology, Hanoi, Vietnam

tuyen.buingoc@hust.edu.vn

Abstract. Fabric waste treatment of the textile industry is a matter of current

concern. Fabric shredding machines are used to shred fabrics for fuel combustion
or recycling. In the process of working, the cutter is gradually worn, affecting
the working ability of the machine. When the tool wears to the limit value cor-
responding to one cycle of tool life, the machine must stop working to change
the tool. The fabric shredding efficiency and productivity of these machines are
highly dependent on the tool life of the cutters. This paper presents a study on
tool wear and tool life of cutters of fabric shredding machines. The influence of
factors such as cutting speed, cleanliness or soiling of the input fabric on tool life
is investigated for two types of tool materials as SKD11 steel and YG9C sintered
tungsten carbide. From this, an optimal model to achieve the maximal productivity
in a tool life cycle has been developed. Through optimization problem solving,
optimal cutting speeds for specific cases have been determined.

Keywords: Fabric shredding · Cutter · Tool life · Cutting speed · SKD11 steel ·
YG9C sintered tungsten carbide

1 Introduction
Environmental pollution has been increasing gradually and causing serious impact on
living organisms including human-being. The problem of reducing waste to the envi-
ronment, including industrial waste in general and fabric waste in particular is increas-
ingly interested. Scraps, outdated clothing, etc. are cut into small pieces by shredding
machines, then recycled to make pillow cores, blankets, stuffed animals, etc. or used
as combustion fuel for cement plants, thermal power plants. There are many types of
fabric waste shredding machines with different types of principles (single-shaft, twin-
shaft, etc.), with different capacity. Single-shaft shredding machines are widely used in
industry.
Figure 1 is single-shaft shredding machine Libra 2000C. Two fixed rows of cutters
are fixed on both sides of the drum (Fig. 2). When the drum rotating, the cutter will
sweep the fabric waste into the gap between the rotating cutters and the fixed cutters that
performs the cutting process. The drum is driven by an electric motor with a capacity of
160 kw. Basic technical specifications of shredding machine Libra 2000C are presented
in Table 1.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 128–137, 2023.
https://doi.org/10.1007/978-3-031-22200-9_14
 A Study on Wearing and Tool Life of Shredding 129

Fig. 1. Shredding machine Libra 2000C.

Fig. 2. Cutter drum and fixed cutter: 1-Fix cutter, 2-Rotating cutters, 3-Drum.

Table 1. Basic technical specifications of shredding machine Libra 2000C.

No Specifications Values
1 Dimension (D × R × C)
2 Length of drum 2000 (mm)
3 Diameter drum 560 (mm)
4 Rated capacity/power 160 kw
5 Speed 750 (rpm)
6 Speed range drum: nmin − nmax 60–260 (rpm)
7 Capacity max 8 Ton/h
8 Product size 25–80 (mm)

The cutters are usually made of alloy steel or carbide. During the work, the cutting
edges are worn due to the mechanical abrasion mechanism, also due to the heating
mechanism. Industrial fabric shredders often work continuously for a long time, so the
abrasion takes place quickly. When the worn cutting edges reach the limit, it is necessary
to stop the machine to replace cutters. The cutter replacing activities affect the production
plan as well as productivity. Therefore, a research on the rule of wearing cutter edges
and optimizing the speed of movement of drum is necessary.
Studies of cutter design and wearing mechanism are presented thoroughly in spe-
cialized books. There has been a number of studies on shredding machines. I. M. Sanjay
Kumar, et al., studied the machine structure and cutting process to cut agricultural waste
130 T. H. Nguyen and N. T. Bui

Fig. 3. Detailed drawing of fixed cutter.

Fig. 4. Detailed drawings of rotating cutter.

[1]. Sudhakara Reddy, et al., developed a machine to shred plastic waste [2]. Chang, F.,
Joseph, Y. K. and Gu-Ming, Z. patents studies on paper shredding cutters and machines
[3–5]. Fauzia Siddiqui, et al., presented the design of machine to shred paper waste [6].
However, there is no study so far that investigates the cutter wearing of fabric shredding
machines.

2 Wearing and Tool Life of Cutters

Figures 3 and 4 are detailed drawings of rotating and fixed cutters. These two types
of cutters are made by SKD11 steel material or YG9C carbide. SKD11 steel is a JIS
Standard tool alloy steel, and is heat treated to a hardness of 60 HRC. YG9C carbide
is manufactured according to China Standard with hardness 70 HRC. During working,
rotating cutters are often worn out faster than fixed cutters. Wear occurs mainly on the
race face of the rotary blade over which the chip flows. Experiments show that when the
cutter blade is worn to the length limit on the race face [lt] = 1.1 mm (Fig. 5) then we
have to stop the machine, replace the cutter, and end the tool life cycle (Table 2).
 A Study on Wearing and Tool Life of Shredding 131

Fig. 5. The wear length on the race face.

Speed of wearing of the cutter depends on some of the following basic factors:
The first is the material of the cutter. It is found that the wearing depends a lot on the
hardness of the blade material, due to the higher the hardness of the material, the slower
the wearing speed. YG9C carbide will wear more slowly than SKD11 steel material.
The second is condition of clean or dirty input waste material. When the input waste
mixed with soil, sand, etc. is supplied to the machine, the cutter will wear out faster. The
third is the cutting speed of the cutter. The greater the cutting speed, the faster the cutter
wearing. The cutting speed of the cutter is calculated according to the Eq. (1):

V = π Dn/1000 (m/ min) (1)

where, drum diameter D = 560 mm; n is rotating speed of the drum, adjusted of about
60–260 rpm. Cutter wearing over time during production process with dirty fabric is
presented in Table 3. From this data table, the cutter wear of SKD11 cutter over time is
built according to Eq. 2 (Fig. 6) and YG9C cutter according to Eq. 3 (Fig. 7).

ltSKD11 = 0.217 + 0.000754t + 0.000264t2 − 0.000001t3 (2)

ltYG9C = 0.02332 + 0.000522t + 0.00004t2 (3)

where, ltSKD11 and ltYG9C are the wears of cutters made of SKD11 and YG9C,
respectively; t is the working time of the cutter.
From the measurement of cutter blade wear according to the working time corre-
sponding to each cutting speed, the relationship of tool life depending on cutting speed
is determined (Table 4).
For metal cutting tools, the relationship between tool life and cutting speed is an
exponential function: V.Tm = const [7]. The relationship between cutting speed and
tool life for fabric cutters is predicted to be similar. Using interpolation method with
supports of Minitab software [8], the relationship between tool life and cutting speed for
shredding cutters is determined by Eqs. (4), (5), (6), (7) and charts in Figs. 8, 9,10 and
11.
132 T. H. Nguyen and N. T. Bui

Table 2. Relationship of tool wear over cutting time.

Cutting time (h) Wear length lt (mm) Cutting time (h)

SKD11
8 0.2 8
16 0.3 16
32 0.4 32
48 0.6 48
64 0.9 64
80 1.1 80
140 – 140
200 – 200
280 – 280

Fig. 6. Tool wear variation with time for SKD11 blade.

Fig. 7. Tool wear variation with time for YG9C blade.

Case 1: Cuter blade made of SKD11 and dirty fabric waste.

T = 8996.99 V−0.799431 (4)

 A Study on Wearing and Tool Life of Shredding 133

Table 3. Tool life depends on cutting speed.

No Cutting speed V Tool life for SKD11: T(h) Tool life for YG9C: T(h)
(mph) Clean fabric Dirty fabric Clean fabric Dirty fabric
waste waste waste waste
1 150 440 163 3200 1340
2 200 290 131 2050 920
3 250 225 110 1470 700
4 300 175 95 1100 550
5 350 130 80 800 425
6 400 118 74 721 382

Case 2: Cuter blade made of SKD11 and clean fabric waste.

T = 308738.8. V−1.31151 (5)

Case 4: Cuter blade made of YG9C and dirty fabric waste.

T = 907087 V−1.3006 (6)

Case 3: Cuter blade made of YG9C and clean fabric waste.

T = 7786370.V−1.5558 (7)

Fig. 8. Plot of tool life versus cutting speed for SKD11 blade, dirty fabric waste.

3 Machining Regime Optimization

3.1 Machining Capacity
Machining capacity will increase as the cutting speed increases, i.e. when the drum rota-
tion speed increases. However, the dependence of machining capacity on cutting speed
134 T. H. Nguyen and N. T. Bui

Fig. 9. Plot of tool life versus cutting speed for SKD11 blade, clean fabric waste.

Fig. 10. Plot of tool life versus cutting speed for YG9C blade, dirty fabric waste.

Fig. 11. Plot of tool life versus cutting speed for YG9C blade, clean fabric waste.

is nonlinear. With drum design as detailed drawings, average statistics of machining

capacity depending on cutting speed are presented in Table 4. This relationship does not
depend on the tool material.
Using the interpolation method, we have the relationship of machining capacity and
cutting speed according to Eq. (8) and the graph in Fig. 12:

P = 0.58 + 0.04109V − 0.000046V2 (8)

 A Study on Wearing and Tool Life of Shredding 135

Table 4. Machining capacity depend cutting speed.

Speed V (m/p) Capacity P (ton/h)

100 4.2
150 5.8
200 6.9
250 8.0
300 8.8

Fig. 12. Capacity variation with cutting speed.

3.2 Optimization According to the Objective Function of the Maximum

Productivity
In order to achieve the maximum productivity output before the tool has to be stopped,
the tool working time must be maximum and the machining capacity must be maximum.
Therefore, we have the following objective function:

Q = t · P → Max (9)

where, Q is machining productivity (ton)

The constraints domain here includes: the conditions on the speed domain of the
rotating-drum must be within the allowable capacity of the machine (10); working time
must be less than the tool life of the cutter blade (11); cutter blade wear should be less
than the limit wear value (12).
1000V
n min ≤ n = ≤ n max (10)
πD
With nmin = 60 (rpm); nmax = 260 (rpm)

t < T = f(V) (11)

136 T. H. Nguyen and N. T. Bui

With f(V) corresponding to cases according to the Eqs. (4), (5), (6), (7)

lt <[lt] (12)

With [lt] = 1.1 mm.

Use SOLVE tool in Excell software to solve the above optimization problem, we
have specific results for the following cases:
Case 1—Cuter blade made of SKD11 and dirty fabric waste. Optimal cutting speed
is 374.9 (mph), corresponding with optimal rotating speed of 213 (rpm). And max
productivity is 750 ton.
Case 2—Cuter blade made of SKD11 and clean fabric waste. Optimal cutting speed
is 446,6 (mph), that corresponding with optimal rotating speed of 254 (rpm). And max
productivity is 768,5 ton.
Case 3—Cuter blade made of YG9C blade, dirty fabric waste. Optimal cutting
speed is 108 (mph), corresponding with optimal rotating speed of 61 (rpm). And max
productivity is 9194 ton.
Case 3—Cuter blade made of YG9C blade, clean fabric waste. Optimal cutting
speed is 200 (mph), corresponding with optimal rotating speed of 114 (rpm). And max
productivity is 14252.5 ton.

4 Conclusions
This paper presents a study on the wear and tool life of shredding machine cutters
Through statistical collection of tool wear data corresponding to the cases of changing
the state of the input material (clean or dirty), changing the type of tool material (SKD11
and YG9C), changing the cutting speed by changing the rotational speed of the cutter,
the influence of these factors on the tool wear as well as the tool life has been determined.
The results show that when increasing the cutting speed, the tool life decreases rapidly
with an exponential relationship. The influence of cutting speed on tool life also depends
on the tool material and input material state. Carbide has higher hardness, better wear
resistance than tool alloy steel. Therefore, the tool life of cutting tools made of carbide
is higher. The input dirty cloth waste contains a lot of soil, sand, …. Also makes the tool
life decrease quickly. The study also evaluates the dependence of machine productivity
on cutting speed as well as the dependence of tool wear on cutting time through building
empirical regression functions. As the cutting speed is increased, the cutting capacity
increases accordingly. Tool wear increases with cutting time. On the basis of the empirical
regression functions built, an optimal problem model according to the objective function
of the maximum output in a tool life cycle has also been built. From this, the optimal
rotational speeds and maximum machining productivity for one cycle of the tool life are
determined corresponding to specific cases of the type of tool material and the state of
the input fabric.

References
1. Sanjay Kumar, I.M., Hemanth Kumar, T.R.: Design and development of agricultural waste
shredder machine. IJISET Int. J. Innov. Sci. Eng. Technol. 2(10) (2015)
 A Study on Wearing and Tool Life of Shredding 137

2. Reddy, S., Raju, T.: Design and development of mini plastic shredder machine. In: IOP Con-
ference Series: Materials Science and Engineering, vol. 455, 2nd International Conference on
Advancements in Aeromechanical Materials for Manufacturing, 13–14 July 2018, Telangana,
India
3. Chang, F.: Blade assembly for paper shredder. US Patent 6089482, BO2C 18/06, BO2C 18/18
(2000)
4. Joseph, Y.K.: Paper shredding device. US Patent 6390397 B1 (2002)
5. Gu-Ming, Z.: Blade of paper shredder. US Patent 2008/0040934A1 (2008)
6. Siddiqui, F., Patil, H., Raut, S., Wadake, O., Tandel, S.: Design and fabrication of paper shredder
machine. Int. J. Sci. Eng. Res. 8(3) (2017)
7. Stephenson, D.A., Agapiou, J.S.: Metal Cutting Theory and Practice. CRC Taylor and Francis
Group (2016)
8. Van Du, N., Binh, N.D.: Experimental Planning in Engineering. Scientific and technical
publishing house, Hanoi, Vietnam (2011)
 A System Engineering Approach to Model
and Implement Planar Trajectory-Tracking
Controllers for Autonomous Underwater
Vehicles

Ngo Van Hien(B) , Hoang Cong Liem, and Ngo Van He

Hanoi University of Science and Technology, Hanoi, Vietnam

hien.ngovan@hust.edu.vn

Abstract. Following the Model-Driven Architecture (MDA) approach, we have

modeled and implemented planar trajectory planning and tracking controllers for
Autonomous Underwater Vehicles (AUVs). This model covers all steps including
the requirements management, analysis, design, and implementation to conve-
niently realize controllers for most standard AUV platforms. It also allows the
designed elements to be customizable and re-usable in the development of new
control applications for various AUVs. The paper describes step-by-step the devel-
opment lifecycle of planar trajectory-tracking controllers for AUVs. Based on this
proposed model, a horizontal planar trajectory-tracking controller of a low-cost
turtle-shaped AUV was developed and taken on trial trips with good feasibility.

Keywords: Autonomous Underwater Vehicles (AUV) · AUV control ·

Model-Based Systems Engineering (MBSE) design · Model-Driven Architecture
(MDA) · Real-Time UML/MARTE

1 Introduction
The study of oceans needs underwater vehicles such as AUVs with concrete aims to
enhance the effectiveness of civil society in economic as well as in other naval facilities,
e.g., the biological discovery of ocean resources, disaster and tsunami warnings, self-
operated underwater military means, etc. [1, 2].
In the construction of AUVs, the problem of designing an autonomous trajectory-
tracking controller is really challenging because the control design is tightly connected
with the AUV dynamic models. This controller can be considered as a model, which
combines discrete and continuous behaviors, is called the Hybrid Dynamic System
(HDS), and it can be modeled by Hybrid Automata (HA). Actually, there are many AUV
control applications, which have used different soft computing techniques to optimally
solve the control of AUV dynamics [3, 4]. Besides, the immersion in an industrial
control context makes that the development engineers must take into account costs and
existing standards for analyzing, designing, and implementing effectively the system.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 138–149, 2023.
https://doi.org/10.1007/978-3-031-22200-9_15
 A System Engineering Approach to Model and Implement Planar 139

The customization and reusability are thus factors to be associated with the production
of a new application, in order to reduce its costs, resources, and time development [5].
In the software engineering field, the Object Management Group (OMG) has stan-
dardized the Model-Driven Architecture (MDA), which has been started with the well-
known and long-established idea to separate the specification of system operations from
the details of the way that system uses the capabilities of its platform. Besides, the Uni-
fied Modeling Language (UML) [6] has been also standardized by OMG which was
an industry standard to visualize, specify, construct, and document the artifacts of a
software-intensive system. However, the UML lacks the constructs for modeling time
and duration constraints of the industrial control system. For these reasons, the MDA’s
features can be specialized by using the real-time UML/MARTE (Modeling and Anal-
ysis of Real-Time and Embedded systems) version [7], in order to model in detail the
analysis and design artifacts for real-time and embedded control systems, e.g. the AUV
controller.
Starting from the above considerations, we have developed an object-oriented anal-
ysis and design model, which is mainly based on the specialization of MDA’s features,
in combination with the real-time UML/MARTE, in order to conveniently implement a
planer trajectory-tracking controller for AUVs. In the current model, the AUV dynamics
are also adapted for control requirements that are then combined with the specialization
of MDA features composed of the Computation Independent Model (CIM), Platform
Independent Model (PIM) and Platform Specific Model (PSM); the control system then
permits an AUV to track a horizontal planar reference path in the Cartesian space.
Here, the CIM includes the use-case model specialized closely with an implemented
function block diagram to precisely capture the requirement analysis of control for an
AUV controller; the PIM is built on the identified CIM by specifying HA and real-time
UML/MARTE to intensively design the real-time control capsules with their ports, and
protocols for AUV controllers. The detailed PIM elements are then converted into PSMs
of open-source platforms such as OpenModelica [8] and Arduino [9], in order to quickly
simulate, realize and deploy the AUV controller. This proposed model is finally applied
to produce a new horizontal trajectory-tracking controller for a low-cost turtle-shaped
AUV.

2 AUV Dynamics and Control Structure

2.1 AUV Dynamic Model for Control
According to SNAME [10], the six motion components of an underwater vehicle are
defined as surge, sway, heave, roll, pitch, and yaw that are shown in Table 1.
From the large field of guidance, navigation and control of underwater vehicles [11],
the six Degree of freedom (DoF) dynamic model of AUVs can be written in equation
system (1).

η̇ = J (η)ν
(1)
M ν̇ + C(ν)ν + D(ν)ν + g(η) = τ (v, u)

where: η = [η1 T ,η2 T ]T includes the position η1 = [x,y,z]T (NED: North, East and
Down) and the orientation η2 = [φ,θ ,ψ]T (Euler RPY: Roll, Pitch and Yaw angles);
140 N. Van Hien et al.

Table 1. SNAME notations for underwater vehicles.

Degree of freedom Motions Force and moment Linear and angular Position and Euler
velocity angles
1 Surge X u x
2 Sway Y v y
3 Heave Z w z
4 Roll K p φ
5 Pitch M q θ
6 Yaw N r ψ

ν = [v1 T ,v2 T ]T is composed the linear v1 = [u,v,w]T and the angular v2 = [p,q,r]T
velocities; M = M RB + M A is a mass matrix, which denotes the 6 × 6 system inertia
matrix containing M RB —the generalized constant inertia matrix, and M A —the added
mass inertia matrix; C(ν) = C RB (ν) + C A (ν) is the 6 × 6 Coriolis and centripetal forces
matrix including added mass; Linear and nonlinear hydrodynamic damping are contained
within the 6 × 6 matrix D(ν) = D + Dn (ν), D contains the linear damping terms, and
Dn (ν) contains the nonlinear damping terms; g(η) is the 6 × 1 vector of gravitational
and buoyancy effects; τ (v, u) is the vector of resultant force and moment acting on the
underwater vehicle, and u is the control inputs, e.g. the rotational speed of the motors
related to the generated thrusts, the driving angles sent to the needed servo-motor for
sail planes and rudder.

2.2 General Control Structure for an AUV

Within the autonomy architecture of AUVs are three main systems as shown in Fig. 1.
These are: the guidance system, which is responsible for generating the trajectory for
the vehicle to follow; the navigation system, which produces an estimation of the current
state of the vehicle; and the control system, which calculates and applies the appropriate
forces to manoeuvre the vehicle [11].

Fig. 1. A general control structure for an AUV.

All three of these systems have their own individual tasks to complete, yet must
also work cooperatively in order to reliably allow an AUV to complete its objectives.
This allows the AUV to track a desired trajectory even in the presence of unknown
environmental disturbances. As previously mentioned, the AUV controller must take into
 A System Engineering Approach to Model and Implement Planar 141

account models with discrete events and continuous behaviors that can be considered as
an HDS of which global behaviors can be modeled by HA.
The horizontal motion of an underwater vehicle is often described by the motion
in surge, sway, and yaw. Therefore, we choose (x, y, ψ)T , (u, v, r)T and the Jacobian
matrix (J) typed 3 × 3 for obtaining the three dimension (3D) dynamic model issued
from the general model (1). In this work, we are interested in developing the planar
trajectory-tracking controller for AUVs, so we concentrate on this model to gather the
control algorithms with concrete guidance, e.g., the Line-of-Sight guidance [12].

3 Specializing the Model-Driven Architecture to Develop

the AUV’S Planar Trajectory-Tracking Controller

As previously stated, the CIM, PIM and PSM components of MDA combined with
the real-time UML/MARTE were chosen to model in detail the analysis and design
artifacts for real-time and embedded control systems, e.g., the AUV’s planar trajectory-
tracking controller. The real-time UML/MARTE includes the ‘capsules, ports, protocols,
connectors’ concepts that can be adapted by specializing a set of control capsules in
precise behaviors and structures of the developed AUV controller.

3.1 Defining the Computation Independent Model

To capture the general requirements in the object-oriented systems engineering model,

we present here a meta-model, which consists of the abstract classes by using UML
stereotypes and class diagrams to commonly describe functional components of AUV
(Fig. 2). In UML models, a stereotype is a model element that is an extensibility mecha-
nism, which we can use to identify the purpose of the model element to which we apply
it. For example, the Navigator stereotype can be applied to the abstract Navigation
Systems class in order to indicate that it is a navigation function for the AUV.

Fig. 2. A meta-model of an AUV controller.

The guidance system’s class is responsible for giving the desired trajectory for AUV
to follow. This responsibility is completed by taking the desired waypoints defined
142 N. Van Hien et al.

pre-mission with the possible inclusion of external environmental disturbances; then it

generates a path for this AUV to follow in order to reach each successive waypoint. The
navigation system’s class addresses the task of determining the current state, i.e., the
velocity, position and attitudes) of AUV. It has to provide the best estimate of the current
state of AUV, regardless of what sensor information is available. The controller’s class
is responsible for providing the corrective signals and events to enable AUV to follow a
desired path. This is achieved by receiving the desired state of AUV from the guidance
system’s class, and the current state of the vehicle from the navigation system’s class. The
controller’s class then calculates and applies a correcting force, through the use of the
various actuators on AUV, to minimize the difference between desired and current states.
The actuator’s class is used to present the forces applied to AUV. There are some types
of actuators used in AUV such as propellers, rudders, sailplanes, and tunnel thrusters.
The main two steps to construct the CIM of the AUV’s planar trajectory-tracking
controller are the followings:

– Identifying complex behaviors of the developed AUV by using the use case model.
The use case represents a set of functions or behaviors being provided by a developed
system to actors with its relationships of inclusion, extension and generalization [7].
From the above 3D dynamic model, LOS guidance [12] and general control structure
of AUVs, we present here the main use case model of the AUV planar trajectory-
tracking controller (Fig. 3a). Here, MDS is the Measurement and Display System;
MES is the Marine Environment System including disturbances such as the wind,
waves, ocean currents. In this step, it is necessary to provide industrial conditions,
e.g., the maximum swing angles of rudders, velocity, diving depth and other safe
operational modes of the developed AUV, in order to ensure its operational safety. We
find that the “Track a desired trajectory” use case is oriented towards control modes;
its complex behaviors must be specified by using sequence diagrams, e.g., Fig. 3b in
the real-time UML/MARTE’s convention.

Fig. 3. Use-case model (a) for the AUV’s planar trajectory-tracking controller, and an example
of trajectory-tracking scenarios (b) of the “Track a desired trajectory” use case.
 A System Engineering Approach to Model and Implement Planar 143

– Defining the implemented functional diagram, which permits us to model transforma-

tional activities of this system with events coming from outside; because the real-time
UML/MARTE lacks the constructs for modeling internal continuous behaviors for
each state on the state machine diagram. This is based on the considered dynamic
model of AUV, its industrial constraint conditions, and the concrete behaviors of the
main use case model with LOS guidance.

3.2 Building the Platform Specific Model

In the PIM of an AUV, HA is specialized to describe mathematical behaviors, i.e., the

dynamic model of AUV including terms such as Situations, Continuous State Variables,
Event, Transition, Global Continuous Behavior and Invariants. A HA of the AUV’s
planar trajectory-tracking controller is defined by data form (2). Here, Q is a set of
running cases of the AUV, qo ∈ Q is the starting situation,  X is the continuous state-
space of continuous elements, xco ∈ X is the initial value, is a set of external
 events,
A is a set of transitions between running cases corresponding to events σ ∈ , Inv is
an application tool, which is used to check xc ∈ inv(q), and F is the continuous global
model issued from the 3D dynamic model.

H AUV = (Q, X, å, A, Inv, F, qo , xco ), (2)

We find that direct transformations of CIM to the implementation environment must

be supplemented by the visual artifacts to carry out an industrial AUV and its re-use
in the new application development phase. For example, the above-identified CIM is
not well adapted to visualize, and model interconnection types between control objects
or sub-systems. In the detailed design phase of AUV, we transform the identified CIM
into PIM, which is based on the use case approach and the real-time UML/MARTE
concepts. We have developed the five main control capsules of PIM, which take part in the
HA realization of the AUV’s planar trajectory-tracking controller being developed: the
continuous part’s capsule, discrete part’s capsule, internal interface’s capsule, external
interface’s capsule and Instantaneous Global Continuous Behavior (IGCB’s capsule).
Figure 4 shows the capsule pattern of control by using the real-time UML/MARTE’s
structure diagram.
Here, the discrete part’s capsule consists of situations Q and transitions A in HA of
the AUV’s planar trajectory-tracking controller; the continuous part’s capsule contains
the continuous state-space X; the IGCB’s capsule implements concrete global continu-
ous behaviors as f ∈ F, where f is directly derived from the 3D dynamic model and the
implemented functional block diagram; the external interface’s capsule is an intermedi-
ary, which receives/sends events/signals between the AUV’s planar trajectory-tracking
controller and the MES/MDS; the internal interface’s capsule permits the Inv tool to
generate internal events in the HA evolution. The detailed specification of this capsule
pattern and the realization hypotheses of the HA evolution, which permit the invariant
Inv and guard control Guard can generate internal events for the AUV controller, can
be found in the author’s previous report [13].
The reusability is very important for developing general industrial control systems,
because it allows mitigating the development time, resources and cost. The various
144 N. Van Hien et al.

reusable views in the development phase are considered as follows: The reusable view is
based on the virtual mechanism of objects, classes or class hierarchies; the other re-use
view can be based on design components. Actually, we can find software tools, e.g., IBM
Rational Rose RealTime and IBM Rational Software Architect RealTime [14], which can
be used to support the re-use of designed control components. They permit us to make the
reverse-engineering of applications for tracking requirements of the developed systems.
The specializations, which permit the capsule pattern of a developed AUV controller to
be customizable and re-usable in new control applications for various AUVs, could be
seen in the author’s reports [13].

Fig. 4. Real-time capsule pattern for control of an AUV.

3.3 Constructing the Platform Specific Model

To realize industrial control systems such as the AUV’s planar trajectory-tracking con-
troller, the PSM is firstly implemented to the simulation model transformed from the
above-identified PIM. It is important to perform simulation models instead of carrying
out experiments on real systems because of expensive and dangerous experiments, inves-
tigated systems not yet exist, the incompatible time scale of the dynamics of the system
with the experimenter, inaccessible variables, etc. The simulation results also permit us
to theoretically evaluate the control performance and functionalities, and to easily opti-
mize control design elements of this system before we decide to realize and deploy it.
The ‘sub-system’ paradigms, which are supported by software tools such as LabView-VI,
MatLab-Simulink, OpenModelica are used to perform the control simulation model of
AUV; because they are easily adapted from the object-oriented design elements of PIM.
Then PIM, which is modified with the control elements optimized in the PSM of the
simulation model, is adapted to obtain the new updated PIM for realization models of
AUV that is called PIM+ . Finally, this PIM+ is converted into new PSMs by using differ-
ent specific platforms, which support object-oriented programming languages such as
 A System Engineering Approach to Model and Implement Planar 145

C++, Java, or Ada, in order to completely realize the AUV’s planar trajectory-tracking
controller with compatible microcontrollers.

4 Application

Based on the above-proposed model, a new planar trajectory-tracking controller, which

permits low-cost turtle-shaped AUV (T-AUV) to reach and follow the predetermined
trajectory, was deployed and taken on trial trips. The main operational parameters of this
T-AUV are summarized in Table 2.

Table 2. Main operational parameters of the T-AUV.

Parameter Value
Size (L × H × W) (1.26 × 0.40 × 0.61) m
Net dry weight 21.20 kg
Autonomous duration 20 min
2 × Li-Po battery 22.2 V, 20000 mAh
Ultimate capacity 324 W
Maximum submerging/rising speed 0.30 m/s
Maximum horizontal transferring speed 1.80 m/s
Maximum radius of action 450 m

In this case study, an implemented functional block diagram was defined to model the
continuous dynamics of the T-AUV controller, because it is difficult to use the real-time
UML/MARTE, which lacks the constructs for depicting complex internal continuous
evolution models combined with control methods, e.g., the Integral Backstepping (IB)
or Proportional-Integral-Derivative (PID) techniques. Thus, an implemented functional
block diagram (Fig. 5) was proposed to gather the internal continuous behaviors of the
T-AUV controller.
In Fig. 5, Desired trajectory and depth are events, which are used for providing,
respectively, the desired position and depth to the control blocks of positions and depths.
Ω di , i = 1, 4 are desired rotational speeds, which are applied to the four actuators
installed on the AUV, e.g., the blocks of motor control for the propellers and rudders.
ΣT and τ φ,θ,Ψ are the overall output forces and moments acting on the actuators of
the AUV. The IB expansion combined with the Control Lyapunov Functions (CLF) for
AUV controllers is well-used in many AUV control applications. In the case study, this
is also applied to the functional blocks of Depth Control, Position Control, and Attitude
Control (Fig. 5), which participate in the continuous evolution of the AUV controller.
The PID regulators are also used for the functional block of Motor Control. In this study,
we did not focus on the decomposition of these control techniques for the AUV because
they have been clearly developed in many AUV applications [11].
146 N. Van Hien et al.

Fig. 5. Functional block diagram for implementing the continuous evolution of the T-AUV.

The ATMEGA32-U2 and STM32 Cortex-M4 microcontrollers [9] were installed on

the main board, in combination with installed sensors of the Inertial Measurement Unit
(IMU): MPU6000 with a working frequency of 100 Hz [15], and Global Positioning
System (GPS): Ublox Neo 6M with working frequency 10 Hz [16], and the control pro-
gram was installed with Arduino’s IDE based on C++ language. The T-AUV installation
for trial trips is shown in Fig. 6a. The experimental scenarios of the T-AUV controller
were based on various desired courses with different desired course angles, and various
desired shape-based planar trajectories and mean transferring speeds.

Fig. 6. Installation and trial trips of the planar trajectory-tracking controller (a), the T-AUV
reached and followed a desired triangle-shaped trajectory (b).

Some of the main planar course-tracking test results are shown in Table 3. Figure 6b
indicates the horizontal planar trajectory-tracking controller that permits the T-AUV to
reach and follow the desired triangle-shaped trajectory. These values are carried out for
the three paths generated by Way-Points (WPs): WP0-WP1, WP1-WP2 and WP2-WP0.
 A System Engineering Approach to Model and Implement Planar 147

Table 3. Test scenarios and obtained results for the planar course-tracking of the T-AUV.

No Desired course angle (deg) Mean transferring speed Duration for the stabilized
(m/s) course (s)
1 010 1.0 6.30
2 010 1.5 5.60
3 020 1.0 6.50
4 020 1.5 5.90
5 030 1.0 7.60
6 030 1.5 6.50

Based on the comparison between the experimental data of these trial tests and the
obtained experimental results from the literature [17], in which the PID regulators were
only used for implementation, the planar trajectory-tracking controller of this T-AUV
was improved in terms of the stabilized course duration and trajectory error, which
decrease by about 6.0% s and 1.5%, respectively.

5 Conclusion and Future Work

The paper presented a hybrid control approach to conveniently deploy controllers for
AUVs. This model is based on the MDA specialization, in combination with the real-
time UML/MARTE, and HA to closely realize planar trajectory planning and tracking
controllers for the AUVs. The dynamic models of AUV were firstly adapted for control
and combined with the MDA’s features, such as the CIM, PIM, and PSM. In the CIM,
the use case model was defined with continuous behaviors such as the implemented
functional block diagram to capture the control requirements. The PIM was established,
in combination with the HA, and built up by the real-time capsule pattern. The designed
PIM was then converted into PSM through the round-trip engineering of the object-
oriented codes to form an AUV controller with suitable microcontrollers. Based on the
proposed model, a new planar trajectory-tracking controller of a low-cost turtle-shaped
AUV was implemented and tested out on the ATMEGA U2 and STM32-Cortex-M4
microcontrollers. Following the above-specialized MDA concepts, in combination with
the real-time UML/MARTE, engineering developers may be more capable of managing
the system complexity by using visual artifacts. However, as the real-time UML/MARTE
version lacks the constructs for modeling complex internal continuous behaviors com-
bined with control methods of developed systems, an implemented functional block
diagram must be then supplemented in the CIM to depict internal continuous behaviors
for the developed control system.
Actually, we just applied the above-proposed model to the low-cost AUV’s planar
trajectory-tracking controller, and intend to implement it in the new control applications
for autonomous coordinated AUV. In the following studies, we will extend it more
effective, in order to create cooperative controllers to balance the search and target
response in a coordinated team of multiple AUVs, which can be used for performing
148 N. Van Hien et al.

quickly missions in a wide range of actions in order to improve the efficiency of patrol
and monitor for the marine environment.

Acknowledgment. This research was supported by the Vietnam National Foundation for Science
and Technology Development (NAFOSTED) under grant number 107.03–2019.302. The authors
would like to warmly express their thanks for the support.

References
1. AUVAC: Autonomous Undersea Vehicles Applications Center. https://auvac.org/. Accessed
July 2022
2. Kang, S., Yu, J., Zhang, J., Jin, Q.: Development of multibody marine robots: a review. IEEE
Access ISSN 2169–3536(8), 21178–21195 (2020). https://doi.org/10.1109/ACCESS.2020.
2969517
3. Han, M., Lyu, Z., Qiu, T., Xu, M.: A review on intelligence dehazing and color restoration for
underwater images. IEEE Trans. Syst. Man Cybern. Syst. ISSN 2168–2216(50), 1820–1832
(2020). https://doi.org/10.1109/TSMC.2017.2788902
4. Liu, Z., Liu, Q., Xu, W., Wang, L., Zhou, Z.: Robot learning towards smart robotic manufac-
turing: a review. ID 102360. Robot. Comput. Integr. Manuf. 77, 21 (2022). Elsevier. ISSN
0736-5845. https://doi.org/10.1016/j.rcim.2022.102360
5. Soriano, T., Gies, V., Pham, A.H., Van Hien, N.: Mechatronics Iterative Design for Robots
Multi-agent Integration. In: Barkallah, M., Choley, J.-Y., Louati, J., Ayadi, O., Chaari, F., Had-
dar, M. (eds.) MECHATRONICS 2019. LNME, pp. 57–66. Springer, Cham (2020). https://
doi.org/10.1007/978-3-030-46729-6_6
6. OMG: Documents Associated With Unified Modeling Language™ (UML® version 2.5.1):
OMG formal/17–12–05. OMG (2017). http://www.omg.org/spec/UML/
7. OMG: UML profile for MARTE: modeling and analysis of real-time embedded systems
version 1.2. OMG ptc/18-07-05. OMG (2019). http://www.omg.org/spec/MARTE/
8. OpenModelica: OpenModelica. OpenModelica software, version 1.18. https://www.openmo
delica.org/. Accessed May 2022
9. Arduino: Open-source electronics prototyping platform for hardware and software. http://
www.arduino.cc/. Accessed April 2022
10. SNAME: Nomenclature for Treating the Motion of a Submerged Body through a Fluid.
SNAME, New York 18, N. Y., USA (1950)
11. Fossen, T.I.: Handbook of Marine Craft Hydrodynamics and Motion Control, 2nd edn. Wiley,
United Kingdom (2021)
12. Lekkas, A.M., Fossen, T.I.: Integral LOS path following for curved paths based on a
monotone cubic Hermite spline parametrization. IEEE Trans. Control Syst. Technol. ISSN
1063–6536(22), 2287–2301 (2014). https://doi.org/10.1109/TCST.2014.2306774
13. Hien, N.V., He, N.V., Diem, P.G.: A model-driven implementation to realize controllers for
Autonomous Underwater Vehicles. Appl. Ocean Res. Elsevier ISSN 0141–1187(78), 307–319
(2018). https://doi.org/10.1016/j.apor.2018.06.020
14. IBM: IBM rational’s methodology, software, online documentation and training kits. https://
www.ibm.com/academic/home. Accessed April 2022
15. InvenSense: Sensor system on chip. http://www.invensense.com/. Accessed March 2022
16. u-blox: Product selector. https://www.u-blox.com/en/product-search. Accessed May 2022
 A System Engineering Approach to Model and Implement Planar 149

17. Hien, N.V., et al., Diem P.G.: Research on, design and manufacture control systems with the
integration of object-oriented technology (MDA & Real-Time UML) and navigation units
(INS/GPS) for autonomous underwater vehicles. Research project report, funded by State of
Vietnam, KC03.TN05/11–15, Hanoi University of Science and Technology, Hanoi, Vietnam
(2013)
 Active Disturbance Rejection Control
of an Antagonistic Muscle

Quy-Thinh Dao1(B) , Duc-Canh Nguyen1 , Minh-Chien Trinh1 , Van-Vuong Dinh1,2 ,

Trong-Hieu Do1(B) , and Minh-Duc Duong1
1 Hanoi University of Science and Technology, Hanoi 11615, Vietnam
{thinh.daoquy,hieu.dotrong}@hust.edu.vn
2 Hanoi Vocational College of High Technology, Hanoi 10000, Vietnam

Abstract. Recently, pneumatic artificial muscle (PAM) has been widely used in
rehabilitation systems. These kinds of soft actuators are often used in an antago-
nistic configuration which allows conversion of longitudinal direction to rotating
movements. This paper focuses on modeling and tracking control of a PAMs’
antagonistic configuration. The mathematical model of the antagonistic actuator
is built based on the physical structure of a single PAM first. After that, an active
disturbance rejection control method is employed for trajectory tracking purposes.
Finally, the mathematical model and proposed control strategy are verified by
experiments with and without a load.

Keywords: Pneumatic artificial muscle · ADRC · Antagonistic configuration ·

Rehabilitation robot · Extended state observer

1 Introduction
Pneumatic artificial muscle (PAM) is one of the most promising actuators for the human-
interaction system, such as rehabilitation devices, medical robots, etc. PAMs present
many outstanding advantages, such as being incredibly lightweight, having a high ratio
of weight and volume to power, and inherent safety. However, PAM-based systems
face disadvantages such as uncertainty and significant delay due to the compressed
air. In previous studies, some nonlinear models, for example, Bouc-Wen [1], Prandlt-
Ishlinskii [2], Maxwell-Slip [3], or Preisach models [4] were employed to describe the
characteristic of PAMs. However, the procedures to identify the model parameters of
these methods are too complicated. To obtain a more straightforward way to model the
behavior of PAMs, linear mathematical models have been applied in the literature [5–9].
In these researches, the uncertainty and nonlinearity in the characteristic of PAMs are
observed by an extended state observer (ESO) and compensated by active disturbance
rejection controllers (ADRCs). These ADRCs are a promising control algorithm for the
PAM systems, which can achieve a good accuracy when tracking a human-gait pattern
trajectory.
This paper develops an ADRC control approach for two opposing PAMs connected
via a pulley. From the relationship between the supplying pressure of a single muscle

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 150–156, 2023.
https://doi.org/10.1007/978-3-031-22200-9_16
 Active Disturbance Rejection Control 151

and its length, the extension force of the PAM is obtained. After that, the dynamic
behavior of the antagonistic actuator is built. Finally, an ADRC is designed for trajectory
tracking purposes. All mathematical models of PAMs and control algorithm are verified
by experiments with and without a load.

2 System Description

Figure 1 illustrates the experimental platform used in this research. The system includes
a pair of muscles connected via a pulley. The pulley is rigidly connected to a potentiome-
ter, so the rotation angle can be measured. The length of each PAMs can be changed
by regulating two proportional electric control valves ITV1030-series from SMC com-
pany. The used PAM has one inch of diameter, and the length is fifteen inches. The
PAM can contract about 30% from its complete deflation state. The proposed ADRC is
implemented by LabVIEW software and Myrio-1900 hardware.

Fig. 1. The experimental system (a) and a single structural PAM (b).

3 Active Disturbance Rejection Controller Design

3.1 System Modeling

To obtain the mathematical model of the PAM-based actuator, we consider the kinetics
of a single structural artificial muscle as shown in Fig. 1b. In the report by Chou [10],
PAM acts similar to a spring with a hardness KS and we have:

F = KS L = KP P(L − Lmin ) (1)

where:
6πD02 PL
K= = KP P (2)
4b2
152 Q.-T. Dao et al.

With F is the tension force of PAM; L and Lmin are the length of PAM and its
theoretically possible minimum value, respectively; b is the length of a thread; D0 is the
diameter of PAM when the angle between a braided thread and the cylinder long axis is
zero; P is the pressure inside the PAM.
Equation (1) describes the relationship between the force generated by a single PAM
(F) and its pressure (P). The pressures inside each PAM of antagonistic configuration
are determined as follows:

PP = P0 + P
(3)
PA = P0 − P

in which P0 is the initial pressure for determining the actuator’s angle at startup, PA and
PP are the anterior and posterior muscle pressures. P is the control variable that leads
to changes in the actuator angle. We can obtain the length of each muscle of the actuator
as:

LA,P = L0 ± Rθ (4)

Otherwise, we can describe the dynamic motion of the entire system as:

T = J θ̈ (t) + Bθ̇ (t) + Kθ (t) (5)

where T is the torque vector generated by pair of muscles, J is the pulley’s moment
inertia. B and K represent the damping and stiffness coefficient of the actuator.
From Eqs. (1), (3), (4), and (5) we have:

2KP R(L0 − Lmin )P(t) = J θ̈ (t) + Bθ̇ (t) + (K − 2KP R2 P0 )θ (t) (6)

where y(t) = θ (t) is the measured output, u(t) = P is the control variable, Eq. (6)
becomes:

Kc u(t) = J ÿ(t) + Bẏ(t) + (K − 2KP R2 P0 )y(t) (7)

where: Kc = 2KP R(L0 − Lmin ).

Thus, Eq. (7) describes the dynamic behavior of PAMs in the antagonistic
configuration.

3.2 Controller Design

From Eq. (7), we can obtain:
 
B K − 2KP R2 P0 Kc
ÿ(t) = − ẏ(t) − y(t) + u(t) = f (t) + b0 u(t) (8)
J J J
where f (t) represents the unknown components of the system. In this paper, an observer
with an extended state called an extended state observer (ESO), is implemented to
observe f (t). With such estimation, the effect of f (t) on the output y(t) can be com-
pensated as a disturbance rejection problem. This idea is the main point of the ADRC
concept [12]. The constitution of the proposed control diagram is presented in Fig. 2.
 Active Disturbance Rejection Control 153

Fig. 2. ADRC controller structure.

Expressing Eq. (8) as state space with x1 (t) = y(t), x2 (t) = ẏ(t), x3 (t) = f (t),
 T  T
x(t) = x1 x2 x3 and ẋ(t) = ẋ1 ẋ2 ẋ3 we have:
ẋ(t) = Ax(t) + Bu(t) + C f˙ (t) (9)
where:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
010 0 0
A = ⎣ 0 0 1 ⎦; B = ⎣ b0 ⎦; C = ⎣ 0 ⎦ (10)
000 0 1
Though, we cannot measure the unknown disturbance f (t). The estimated value x̂3
of f (t) can be obtained from the control variable and measured output as follows:
˙ = Ax̂ + Bu + L(y − x̂1 )
x̂(t) (11)
 T
where L = l1 l2 l3 with l1 , l2 and l3 the parameters of the observer
ensuring x̂1 , x̂2 and x̂3 close to the value of y, ẏ and f . Consequently, We can obtain
the control signal of ADRC as:
u0 (t) − fˆ (t)
u(t) = (12)
b0
In which:
 
u0 (t) = KP r(t) − y(t) − KD ẏ(t) (13)
with r(t) the desired trajectory to track. Assume that the observer works well, x̂1 (t), x̂2 (t),
and x̂3 (t) can well approximate y(t), ẏ(t), and f (t), respectively. Substitute Eq. (12) into
the Eq. (8), we have:
 
ÿ(t) ≈ u0 (t) = KP r(t) − x̂1 (t) − KD x̂2 (t) (14)
According to the works of Herbs [12], the ADRC controller parameters can be tuned
via Tsettle as:
−6
KP = (sCL )2 ; KD = −2sCL ; sCL ≈ (15)
Tsettle
154 Q.-T. Dao et al.

4 Experimental Results
Tracking gait pattern trajectories is an essential requirement for any rehabilitation robot.
In order to test the applicability of artificial muscles to rehabilitation robots, human
footstep trajectories are used as reference signals of the system. The gait patterns are
the modified hip and knee trajectories collected from Winter experimental data [11].
In experiments, the well-tuned parameters of the ADRC controller are obtained with
Tsettle = 11 s. Figure 3 illustrates the experimental results when the system follows a gait
pattern trajectory without a load (dash-dot green line) and with 3 kg of load (dash red
line). The frequency of the cycling trajectory is 0.2 Hz. We can see that the controller can
achieve good performance with a root mean square tracking error (RMSTE) less than
0.5°. When a load is added to the system, the control performance is slightly decreased
to 1.1° of RMSTE. A similar conclusion is obtained when the system tracks the 0.3 Hz
gait pattern. To examine the robustness of the controller with external disturbance, we
conduct the second scenario of the experiment in which a noise is fed into the system
for a time 0.5 s. The noise has the same amplitude as the control variable but in reverse.
As shown in Fig. 4, the measured trajectory quickly returns to track reference signals
in about 1 s, and the maximum tracking error does not exceed 3°. It is acceptable for
the rehabilitation system. Table 1 provides the system’s RMSTE values under different
experimental scenarios.

Fig. 3. Experimental results under different conditions (no load and 3 kg of load) when tracking
the 0.2 Hz gait pattern: hip joint (a), and knee joint (b).

5 Conclusion
This paper presents an active disturbance rejection controller of PAM in an antagonistic
configuration. From the positive performance in reference tracking, we show that the
control quality when using the ADRC method is very promising. The trajectories take
the shape of the human gait pattern with reasonable control quality. That suggests the
application of artificial muscles and ADRC approach in robot systems interacting with
 Active Disturbance Rejection Control 155

Fig. 4. The system’s responses to external disturbance when tracking (a) 0.2 Hz and (b) 0.3 Hz
gait pattern trajectories.

Table 1. Experimental results.

Reference signal’s frequency RMSTE (°)

No load 3 kg of load Noise
0.2Hz 0.50 1.10 2.20
0.3Hz 0.60 2.08 3.40

humans, such as rehabilitation robots and medical robots. However, the orbit frequency
is only 0.3 Hz, which is still slow. This frequency is equivalent to a pedestrian speed of
about 1.5 km/h and needs further research to improve control performance.

References
1. Yabe, C., Okuyama, A.: Modeling of pneumatic artificial rubber muscle using Bouc-Wen
model. Proc. Conf. Kanto Branch 25, 19G03 (2019). https://doi.org/10.1299/jsmekanto.2019.
25.19G03
2. Xie, S., Mei, J., Liu, H., Wang, Y.: Hysteresis modeling and trajectory tracking control of the
pneumatic muscle actuator using modified Prandtl–Ishlinskii model. Mech. Mach. Theory
120, 213–224 (2018). ISSN 0094–114X
3. Vo-Minh, T., Tjahjowidodo, T., Ramon, H., Van Brussel, H.: A new approach to modeling
hysteresis in a pneumatic artificial muscle using the Maxwell-slip model. IEEE/ASME Trans.
Mechatron. 16(1), 177–186 (2011)
4. Zhao, X., Tan, Y.: Neural network based identification of Preisach- type hysteresis in piezo-
electric actuator using hysteretic operator. Sens. Actuators A: Phys. 126(2), 306–311 (2006).
ISSN 0924–4247
5. Yuan, Y., Yu, Y., Guo, L.: Nonlinear active disturbance rejection control for the pneumatic
muscle actuators with discrete-time measurements. IEEE Trans. Ind. Electron. 66, 2044–2053
(2019)
6. Zhao, L., Liu, X., Wang, T.: Trajectory tracking control for double-joint manipulator systems
driven by pneumatic artificial muscles based on a nonlinear extended state observer. Mech.
Syst. Signal Process. 122, 307–320 (2019). ISSN 0888–3270
156 Q.-T. Dao et al.

7. Yuan, Y., Yu, Y., Wang, Z., Guo, L.: A Sampled-data approach to nonlinear ESO-based active
disturbance rejection control for pneumatic muscle actuator systems with actuator saturations.
IEEE Trans. Ind. Electron. 66, 4608–4617 (2019)
8. Zhao, L., Cheng, H., Xia, Y., Liu, B.: Angle tracking adaptive backstepping control for a
mechanism of pneumatic muscle actuators via an AESO. IEEE Trans. Ind. Electron. 66(6),
4566–4576 (2019)
9. Yang, H., Yu, Y., Zhang, J.: Angle tracking of a pneumatic muscle ac- tuator mechanism
under varying load conditions. Control Eng. Pract. 61, 1–10 (2017). ISSN 0967–0661
10. Chou, C.-P., Hannaford, B.: Measurement and modeling of McKibben pneumatic artifi-
cial muscles. IEEE Trans. Robot. Autom. 12(1), 90–102 (1996). https://doi.org/10.1109/70.
481753
11. Winter, D.A.: Biomechanics and motor control of human movement, 3rd edn. University of
Waterloo Press, Waterloo, ON, Canada (1990)
12. Herbs, G.: A simulative study on active disturbance rejection control as a control tool for
practitioners. In: Siemens AG (ed) Clemens Winkler-Strabe, vol 3, Germany (2013)
 An Approach for Optimizing
the Hedge-Algebras-Based Controller
and Application in Structural Vibration Control

Hai-Le Bui1,2(B)
1 School of Mechanical Engineering, Hanoi University of Science and Technology, No. 1 Dai
Co Viet Street, Hanoi, Vietnam
le.buihai@hust.edu.vn
2 Intelligent Mechatronics and AI Research Group, School of Mechanical Engineering, Hanoi

University of Science and Technology, No. 1 Dai Co Viet Street, Hanoi, Vietnam

Abstract. Since introduced in 1990, the Hedge-algebras theory has been effec-
tively applied to many theoretical and practical problems in different fields. This
paper presents an approach for optimizing the controller based on the Hedge-
algebras theory. In which the design variables include the parameters for building
the Normalization and De-Normalization diagrams and the reference intervals
of the state and control variables. This approach is applied to a building model
subjected to earthquake loads. The objective functions are to minimize the sys-
tem’s maximum relative displacement and absolute acceleration. The simulation
results show that the optimal parameters make the controllers more efficient than
controllers in published studies. In addition, the approach in the paper allows
determining the controller parameters suitable for the investigated model instead
of calculating them by a trial and error method. The approach in this study can be
developed to design optimal controllers based on the Hedge-algebras theory for
different mechanical models.

Keywords: Hedge-algebras-based controller · Optimization · Vibration control

1 Introduction
The Hedge-algebras (HA) theory has been proposed since 1990 [1, 2]. The fuzzy measure
of linguistic values in HA is represented by real numbers instead of using fuzzy sets as
in traditional fuzzy set theory. This representation always ensures the natural semantic
order of linguistic values. Hence, the calculation steps in HA are simple and explicit.
HA theory has been applied in the control field since 2008 [3]. The HA-based con-
troller (HAC) includes function blocks: Normalization (to transfer state variables from
the real domain to the semantic domain), HA – Rule Base (to represent the control
rule base), HA – Inference (to demonstrate the inference mechanism when determining
control variables in the semantic domain), and De-Normalization (to transfer control
variables from the semantic domain to the real one) [4]. The applications of HA in
control focus on the simple inverted pendulum model [3], the damped-elastic-jointed

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 157–165, 2023.
https://doi.org/10.1007/978-3-031-22200-9_17
158 H.-L. Bui

inverted pendulum subjected to a follower force [5], the inverted pendulum-cart system
[6], the beam model subjected to bending vibration [7], the active suspension model
[8], the induction generator system [9], and especially building structures subjected to
earthquake loads [4, 7, 10–15]. The above publications have demonstrated many advan-
tages of HA theory over fuzzy set theory in applications of the control field, such as
simplicity in construction, clarity in implementation, ease in optimization, efficiency
in operation, and especially fast in computation. However, the above publications have
not mentioned the tuning of Normalization and De-Normalization steps to improve the
control efficiency of HACs.
Therefore, in this study, the problem of optimizing the parameters in the Normaliza-
tion and De-Normalization diagrams and the reference intervals of the state and control
variables is performed. Numerical simulations are carried out on a building subjected to
earthquake loads.

2 Investigated Model

Considering a 3-story building model subjected to seismic loads ẍ0 with structural param-
eters given as shown in Fig. 1a. The control force u is generated from the Active Bracing
System (ABS) with the maximum control force umax = 700 N [16–18].
The system’s state differential equations are as follows:
⎡ ⎤⎧ ⎫ ⎡ ⎤⎧ ⎫
m1 0 0 ⎨ ẍ1 ⎬ c1 + c2 −c2 0 ⎨ ẋ1 ⎬
⎣ 0 m2 0 ⎦ ẍ2 + ⎣ −c2 c2 + c3 −c3 ⎦ ẋ2
⎩ ⎭ ⎩ ⎭
0 0 m3 ẍ3 0 −c3 −c3 ẋ3
⎡ ⎤⎧ ⎫ ⎧ ⎫
k1 + k2 −k2 0 ⎨ x1 ⎬ ⎨ u − m1 ẍ0 ⎬
+ ⎣ −k2 k2 + k3 −k3 ⎦ x2 = −m2 ẍ0 (1)
⎩ ⎭ ⎩ ⎭
0 −k3 −k3 x3 −m3 ẍ0

The control force must satisfy the following conditions: |u| ≤ umax . In numerical
simulations, the peak acceleration of the seismic loads is scaled to 0.112 g.
For the model in Fig. 1a, the criteria that need to be minimized include the system’s
maximum relative displacement and absolute acceleration to ensure safety for structures
and humans.
Let CD and CA be the ratio as below, in which di , di∗ , ai , ai∗ are the maximum
relative displacement and absolute acceleration of the system between the cases of the
controlled and uncontrolled system, respectively.

max|di |
i
CD =  
maxd ∗ 
i
i
(2)
max|ai |
i
CA =  
maxai∗ 
i
 An Approach for Optimizing the Hedge-Algebras-Based Controller 159

Fig. 1. The 3-story building model (a) and system’s control diagram (b).

3 Control Design

Theorems in the theory of HA allow determining the fuzzy measures of linguistic values
in terms of real numbers in the interval [0, 1] [4, 11]. These measures are also called
linguistic values’ quantitative semantic mapping (SQM). Considering a HA structure
consisting of primary terms (Negative, Positive) = (N, P), hedges (Little, Very) = (L,
V), and neutral element W. Several typical linguistic values with their SQM value are
shown in Fig. 2. The system’s control diagram is plotted in Fig. 1b, in which z1 = x 1
and z2 = ẋ1 .
Assume that the reference range and linguistic values of the variables are given as:

z1 : [−a, a], (N, LN, W, LP, P); z2 : [−b, b], (LN, W, LP) (3)

u : [−c, c], (VN, N, LN, W, LP, P, VP) (4)

The operating principle of HAC is shown in Fig. 3. Role of function blocks Nor-
malization of state variables (Fig. 3a and b), HA – Rule Base (Fig. 3c), HA – Inference
(Fig. 3d), and De-Normalization of control variables (Fig. 3e) have been presented in
Sect. 1.
It is noted that the tuning problem to find the best Normalization and De-
Normalization schemes has not been considered in the publications on controlling
mechanical models using HA. In the present work, these diagrams are divided into
lines as shown in Fig. 3a, b, and e. Assuming z1 is in the interval i and i + 1, the formula
to convert z1 from the real domain to the semantic one (z1s ) is as follows:

Fig. 2. SQM value of typical linguistic values.

160 H.-L. Bui

Fig. 3. Hedge-algebras-based controller.

z1s = α1 z1 + β1
z1s (i + 1) − z1s (i) z1s (i + 1) − z1s (i) (5)
α1 = ; β1 = −z1 (i) + z1s (i)
z1 (i + 1) − z1 (i) z1 (i + 1) − z1 (i)
Similarly, if z2 is in the range j and j + 1, the formula to convert z2 from the real
domain to the semantic one (z2s ) is as follows:
z2s = α2 z2 + β2
z2s (i + 1) − z2s (i) z2s (i + 1) − z2s (i) (6)
α2 = ; β2 = −z2 (i) + z2s (i)
z2 (i + 1) − z2 (i) z2 (i + 1) − z2 (i)
A typical rule base consisting of 15 control rules is shown in Fig. 3c. Combined with
the SQM values of the linguistic values, as shown in Fig. 2, the geometric representation
of this rule base is plotted in Fig. 3d. With the selection of HAC’s parameters as above,
the rule surface in Fig. 3d is plane, and hence the representation of the control variable
in the semantic domain us is determined as follows:

us = z1s + z2s − 0.5 (7)

If us is in the interval k and k + 1, the formula for converting the control variable
from the semantic domain to the real one is as follows:
u = αu us + βu
u(k + 1) − u(k) u(k + 1) − u(k) (8)
αu = ; βu = −us (k) + u(k)
us (k + 1) − us (k) us (k + 1) − us (k)
Next, the HAC optimization problem is presented. The objective functions include:

CD → min; CA → min (9)

 An Approach for Optimizing the Hedge-Algebras-Based Controller 161

Design variables include the height of the intermediate points in the Normalization
and De-Normalization diagrams in Fig. 3a, b, and e, and the reference ranges of the state
and control variables are as below:

z1s (i), i = 1 ÷ 4; z2s (j), j = 1 ÷ 4; u(k), k = 1 ÷ 4; a; b; c (10)

Hence, the optimization problem consists of 15 design variables. Training data dur-
ing optimization using the acceleration of the 1940 El Centro earthquake. The optimal
controls based on the goals C D → min and C A → min are denoted HAd and HAa,
respectively. The optimization tool in this study is Balancing composite motion opti-
mization (BCMO), a recently published population-based optimization algorithm [19,
20].

4 Numerical Simulation
First, the results of the optimization parameters are shown in Figs. 4, 5 and Table 1. The
criteria C D and C A in the case of the El Centro earthquake are compared in Table 2,
in which LQR, MBBC, SSMC, and RC, are controllers of linear quadratic regulators,
modified bang–bang, saturated sliding mode, and robustness, respectively. The maxi-
mum relative displacements and absolute accelerations of the floors are presented in
Fig. 6. The time responses of the system are plotted in Fig. 7.

Fig. 4. Optimal Normalization (a and b) and De-Normalization (s) diagrams - HAd.

Fig. 5. Optimal Normalization (a and b) and De-Normalization (s) diagrams - HAa.

HAd and HAa are optimized based on data from the El Centro earthquake. In addition,
data from the 1994 Northridge, 1999 Duzce, and 1952 Kern County earthquakes are
also used for subsequent simulations to validate the effectiveness of HAd and HAa.
Simulation results of these earthquakes are presented in Fig. 8.
For the El Centro earthquake, it can be seen from Figs. 4, 5 and Table 1 that the
parameters of HACs need to be determined by an optimization problem instead of trial
162 H.-L. Bui

Table 1. Optimal values of reference ranges of variables.

Controller a, m b, m/s c, N Controller a, m b, m/s c, N

HAd 0.01541 0.00379 19747.6 HAa 0.02841 0.06543 15081.4

Table 2. Comparison of criteria C D and C A .

LQR [16] MBBC [16] SSMC [16] RC [16] RC [17] HAd HAa
CD 0.657 0.381 0.388 0.396 0.41 0.373 0.379
CA 0.584 0.548 0.560 0.543 0.53 0.487 0.399

Fig. 6. Peak relative displacement and absolute acceleration of floors – El Centro.

and error steps. As shown in Table 2, HAd and HAa give much higher control efficiency
than the controllers in [16, 17] for both criteria C D and C A with the same maximum
control force umax = 700 N. The control efficiency of HAa is higher than that of HAd,
which is also illustrated in Figs. 6, 7. In the time response of the control force in Fig. 7c,
the change of the control force of HAa is softer than that of HAd.
The simulation results in Fig. 8 show that HAd and HAa both have high control
efficiency for the remaining earthquakes. In addition, HAa gives better or equal control
efficiency than HAd when considering both criteria C D and C A , as listed in Table 3.
Hence, it can be concluded that HAa is the most suitable controller for the investigated
model in this study.
 An Approach for Optimizing the Hedge-Algebras-Based Controller 163

Fig. 7. Time responses of parameters – El Centro.

5 Conclusion

This study presents the optimization problem of Normalization and De-Normalization

diagrams, and variables’ reference intervals of the controller based on Hedge-algebras
theory. Numerical simulations are performed for the building model subjected to different
seismic loads. The research results allow determining the appropriate parameters for the
controlled system by the optimization problem instead of trial and error steps. The
optimal controllers have high control efficiency compared to controllers in published
studies. The approach in this paper can be applied to more complex systems or different
mechanical models.
164 H.-L. Bui

Fig. 8. Peak relative displacement and absolute acceleration of floors.

Table 3. Criteria C D and C A in validation cases

Earthquake Northridge Duzce Kern County

HAd HAa HAd HAa HAd HAa
CD 0.549 0.539 0.580 0.654 0.622 0.668
CA 0.736 0.566 0.713 0.638 0.723 0.676
Mean values 0.643 0.553 0.647 0.646 0.673 0.672

Acknowledgments. This research is funded by Hanoi University of Science and Technology

(HUST) under project number T2021-SAHEP-009.

References
1. Ho, N.C., Wechler, W.: Hedge algebras: an algebraic approach to structure of sets of linguistic
truth values. Fuzzy Sets Syst. 35(3), 281–293 (1990)
2. Ho, N.C., Wechler, W.: Extended hedge algebras and their application to fuzzy logic. Fuzzy
Sets Syst. 52(3), 259–281 (1992)
3. Ho, N.C., Lan, V.N.: Optimal hedge-algebras-based controller: Design and application. Fuzzy
Sets Syst. 159(8), 968–989 (2008)
 An Approach for Optimizing the Hedge-Algebras-Based Controller 165

4. Bui, H.-L., Nguyen, C.-H., Vu, N.-L., Nguyen, C.-H.: General design method of hedge-
algebras-based fuzzy controllers and an application for structural active control. Appl. Intell.
43(2), 251–275 (2015). https://doi.org/10.1007/s10489-014-0638-6
5. Bui, H.-L., Tran, D.-T., Vu, N.-L.: Optimal fuzzy control of an inverted pendulum. J. Vib.
Control 18(14), 2097–2110 (2012)
6. Bui, H.-L., Pham, M.-N., Nguyen, T.-T.-H.: Swing-up control of an inverted pendulum cart
system using the approach of Hedge-algebras theory. Soft Comput. 26(10), 4613–4627 (2022)
7. Bui, H.-L., Tran, Q.-C.: A new approach for tuning control rule based on hedge algebras
theory and application in structural vibration control. J. Vib. Control 27(23–24), 2686–2700
(2021)
8. Bui, H.-L.: Multi-objective optimization of hedge-algebra-based controllers for quarter-car
active suspension models. In: IFToMM Asian Conference on Mechanism and Machine
Science, pp. 876–885. Springer (2021)
9. Vukadinović, D., Bašić, M., Nguyen, C.H., Vu, N.L., Nguyen, T.D.: Hedge-algebra-based
voltage controller for a self-excited induction generator. Control. Eng. Pract. 30, 78–90 (2014)
10. Anh, N.D., Bui, H.L., Vu, N.L., Tran, D.T.: Application of hedge algebra-based fuzzy con-
troller to active control of a structure against earthquake. Struct. Control. Health Monit. 20(4),
483–495 (2013)
11. Bui, H.-L., Le, T.-A., Bui, V.-B.: Explicit formula of hedge-algebras-based fuzzy controller
and applications in structural vibration control. Appl. Soft Comput. 60, 150–166 (2017)
12. Bui, H.-L., Nguyen, C.-H., Bui, V.-B., Le, K.-N., Tran, H.-Q.: Vibration control of uncertain
structures with actuator saturation using hedge-algebras-based fuzzy controller. J. Vib. Control
23(12), 1984–2002 (2017)
13. Bui, V.-B., Tran, Q.-C., Bui, H.-L.: Multi-objective optimal design of fuzzy controller for
structural vibration control using Hedge-algebras approach. Artif. Intell. Rev. 50(4), 569–595
(2017). https://doi.org/10.1007/s10462-017-9549-3
14. Tran, D.-T., Bui, V.-B., Le, T.-A., Bui, H.-L.: Vibration control of a structure using sliding-
mode hedge-algebras-based controller. Soft. Comput. 23(6), 2047–2059 (2017). https://doi.
org/10.1007/s00500-017-2919-6
15. Nguyen, T.-D., Bui, H.-L.: General optimization procedure of the Hedge-algebras controller
for controlling dynamic systems. Artif. Intell. Rev. 1–36 (2022). https://doi.org/10.1007/s10
462-022-10242-0
16. Lim, C., Park, Y., Moon, S.: Robust saturation controller for linear time-invariant system with
structured real parameter uncertainties. J. Sound Vib. 294(1–2), 1–14 (2006)
17. Du, H., Zhang, N., Naghdy, F.: Actuator saturation control of uncertain structures with input
time delay. J. Sound Vib. 330(18–19), 4399–4412 (2011)
18. Mac, T.-T., Bui, H.-L.: LQR control design in vibration control of a benchmark building
structure subjected to seismic load. In: Regional Conference in Mechanical Manufacturing
Engineering, pp. 771–780. Springer (2022)
19. Le-Duc, T., Nguyen, Q.-H., Nguyen-Xuan, H.: Balancing composite motion optimization.
Inf. Sci. 520, 250–270 (2020)
20. Bui, H.-L., Tran, N.-A.: Multi-objective optimal design of TMDs for increasing critical flutter
wind speed of bridges. J. Wind Eng. Ind. Aerodyn. 225, 104992 (2022)
 An Enhanced Hybrid Jaya Algorithm for Size
Optimization of Truss Structure Under
Frequency Constraints

Nga T. T. Nguyen1 , Sy Nguyen-Van1 , Thuy T. T. Diem2 , Ngoc Nguyen-Dinh1 ,

Tien-Dat Hoang1 , and Luong Viet Dung1(B)
1 Thai Nguyen University of Technology, Thai Nguyen, Vietnam
{nguyennga,luongvietdung}@tnut.edu.vn
2 School of Foreign Languages, Thai Nguyen University, Thai Nguyen, Vietnam

Abstract. This paper proposed an enhanced hybrid Jaya algorithm, called AEHJ.
The proposed AEHJ is a new improvisation of the Jaya algorithm (Jaya) and the
differential evolution algorithm (DE) with two modifications. Firstly, the local
search is improved by using DE/best/1, DE/best/2, and Jaya operators. Secondly,
an elitist selection approach is used for choosing the best solution for the next
population. For validating the feasibility of AEHJ, the well-known benchmark
example of size optimization for a 10-bar truss is performed.

Keywords: Jaya algorithm · An enhanced hybrid Jaya algorithm (AEHJ) · Size

optimization · Truss optimization · Frequency constraints

1 Introduction

Over past decades, meta-heuristic algorithms have been received much interest from
many scholars from all over the world because they randomly search optimal solu-
tions in a given search space and no sensitivity analyses is required [1–4]. Generally,
meta-heuristic algorithms are classified into three main groups as follows: evolutionary
algorithms (EAs), swarm intelligence algorithms (SI) and physics-based algorithms [5].
Concretely, evolutionary algorithms are population-based algorithms. EAs are based
on natural evolution in biological populations. Some famous evolutionary algorithms can
be listed as follows: Genetic Algorithm (GA) [6], Differential Evolution (DE) [7], and
Genetic Programming and etc.
Regarding swarm intelligence algorithms (SI), SI mimics the social behavior of
self-organized systems, natural or artificial which are ant colonies, bee colonies and bird
flocking and etc. The most well-known swarm intelligence algorithms are Particle Swarm
Optimization (PSO) [8], Symbiotic Organism Search (SOS) [9], Whale Optimization
Algorithm (WOA) [5], Grey Wolf Optimizer (GWO) [4], and etc.
Physics-based methods simulate the physical rules. For instance, the most popu-
lar physics-based algorithms are Simulated Annealing (SA) [10], Gravitational Search

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 166–176, 2023.
https://doi.org/10.1007/978-3-031-22200-9_18
 An Enhanced Hybrid Jaya Algorithm for Size Optimization 167

Algorithm (GSA) [11], Central Force Optimization [12] and Space Gravitational
Optimization [13] and etc.
Among the above-mentioned meta-heuristic algorithms, another noticeable meta-
heuristic algorithm is the Jaya algorithm [14]. The performances of Jaya are also validated
in many engineering problems [15–18]. However, the performances of classical Jaya
need improvisation for reducing computational cost in finding an optimal solution.
From the above-mentioned analyses, the No Free Lunch (NFL) theorem should be
mentioned here [19]. It states that no algorithm can solve all problems or all algorithms
have similar performances for solving all optimization problems. Thus No Free Lunch
theorem reveals the motivation for proposing new meta-heuristic optimizations in this
study. In addition to our best knowledge, hybridization meta-heuristic optimizations
between DE and Jaya are very limited in works [1, 17]. Thus, this study aims to propose an
enhanced hybrid Jaya algorithm (AEHJ). The proposed AEHJ could obtain both the good
accuracy of optimal solutions and fast convergence speed. For enhancing good global
searchability, the mutation operator DE/rand/1 and DE/rand/2 are used. For better local
searchability, DE/best/1, DE/best/1, and Jaya operators are used. For switching between
the global and local searches, the mean and the best values of the objective function are
used. Finally, elitist selection is used for choosing the best individual between the last
and current populations.

2 Definitions of the Problems

This section statements the foundation of objective functions and constraints for size
optimization of truss structures. In this study, objective functions and constraints are
truss weight and frequency constraints, respectively. The design variables are truss sizes.
Thus, the objective functions and constraints are defined as follows:

t
Minimize : Truss weight(A) = Li ρi Ai (1)
i=1
⎧
⎨ ωk − ωL ≥ 0,
min
⎪
Constraints : ωp − ωH
max
≤ 0, (2)
⎪
⎩ min
Ai ≤ Ai ≤ Amax
i

where Truss weight (A) presents the truss weight; L i presents the length of the ith truss
element, ρi is the material density of the ith truss element; A = [A1 ,…, Ai ,…, At ] is the
vector of cross-sectional areas; t is the total number of truss elements; n is the number
of nodal coordinates; ωk and ωp signify the kth and pth natural frequencies of the truss
structure, respectively; ωLmin and ωHmax are the lower and higher bounds of ω and ω ,
k p
min max
respectively; Ai and Ai stand for the lower and higher bounds of Ai , respectively.
It shows that Eq. (1) is a constrained one, thus its transformation to unconstrained form
is required. As reported in [1, 3], the penalty function method is applied to tackle this
issue. Hence, the objective function and constraint are combined and are rewritten as
follows:
Minimize : fobjective (A) = (1 + α1 υ)α2 Truss weight(A), (3)
168 N. T. T. Nguyen et al.


m
 
With: υ = max 0, gz (A) , (4)
z=1

where υ is the sum of violated constraints; gs (A) and m are the zth frequency constraint
and the number of constraints, respectively; the parameters α 1 and α 2 are dependened on
the exploration and exploitation rates of the search space, respectively. In this research,
α 1 is set to be 1, and α 2 is equal to 1.5 at the first population and constantly increased
to 6.

3 Optimization Algorithms

3.1 Differential Evolution

Differential evolution is firstly introduced by Storn and Price [20]. The classical DE has
four main stages including initialization, mutation, crossover, and selection which are
given as follows:

• Initialization

Firstly, NP individuals are randomly generated in a given search space. Each ith
individual (i = 1, 2,…, NP) is a vector of D design variables and given as follows:
g=0

Xi,j = Lj + randi,j [0, 1] Hj − Lj , j = 1, 2, ..., D, (5)

where L, H are the lower and upper boundaries, respectively; NP is the number of
individuals in the population, D is the number of design variables, and rand[0; 1] is a
random value within the range of 0 and 1; g is the gth population. The ith individual
vector is given as follows.
 
g g g g g
Xi = Xi,1 , Xi,2 , . . . , Xi,j , . . . , Xi,D , (6)

• Mutation

g
To enhance the diversity of the population, a mutant vector Vi is generated by
using the following mutation schemes are “DE/rand/1”, “DE/rand/2”, “DE/best/1” and
“DE/best/2”. These four DE/operators are briefly summarized as follows:

g g g g
rand/1 : Vi = XR1 + F XR2 − XR3 , (7)

g g g g
best/1 : Vi = Xbest + F XR1 − XR2 , (8)
 
g g g g g g
rand/2 : Vi = XR1 + F XR2 − XR3 + F XR4 − XR5 , (9)
 An Enhanced Hybrid Jaya Algorithm for Size Optimization 169

 
g g g g g g
best/2 : Vi = Xbest + F XR1 − XR2 + F XR3 − XR4 , (10)

where R1 ; R2 ; R3 ; R4 and R5 are random differential numbers chosen from [1, 2, 3,…,
g
NP]; F is a random number between 0 and 1, and Xbest is the best individual vector in
the current population.

• Crossover

g
To get good optimal solutions,
  the crossover stage is used. The ith trial vector Ui =
g g g g g
Ui,1 , Ui,2 , ..., Ui,j , ..., Ui,D is generated by mixing the target vector Xi and the mutant
g
vector Vi as follows:
 g
g
Vi,j , if j = Rand or randi,j [0, 1] ≤ Cr,
Ui,j = g (11)
Xi,j , otherwise,

where Rand is a random number within the range of [1; NP], and the crossover value
Cr is randomly chosen in [0.7, 1].

• Selection

g
For comparing the trial individual of Ui and the target individual of, the selection stage
is given as follows:
 g g
g

g+1 Ui , if f Ui ≤ f Xi ,
Ui = g (12)
Xi , otherwise,

3.2 Jaya Algorithm

Jaya algorithm is first introduced in [14]. Mathematically, Jaya has four main stages:
initialization, reproduction, and selection. While the initialization and selection stages
are similar to those of DE, the reproduction of Jaya can be expressed as follows:
   
g g g  g  g  g 
Vi = XR1 + rand [0, 1] Xbest − XR2  − rand [0, 1] Xworst − XR3  (13)

where R1 ; R2 ; and R3 are differential numbers chosen from [1, 2, 3,…, NP]; rand [0,1]
is a random number between 0 and 1.

3.3 An Enhanced Hybrid Jaya Algorithm

This section presents an enhanced hybrid Jaya algorithm (AEHJ) which is a hybridization
between Jaya and DE. The AEHJ has two modifications. Firstly, for enhancing the global
search, DE/rand/1 and DE/rand/2 are used. In addition, to achieve a faster convergence
speed, DE/best/1, DE/best/2 and Jaya operator in Eq. (13) are performed.
170 N. T. T. Nguyen et al.

Table 1. Data for size optimization of truss problems with multiple frequency constraints.

Example Modulus of Density of truss Bounds of Frequency

elasticity, members, cross-sectional constraints (Hz)
E(N/m2 ) ρ(kg/m3 ) areas, A(m2 )
10 bar-truss 6.98 × 1010 2770 0.645 × 10–4 ≤ A ω1 ≥ 7, ω2 ≥ 15,
ω3 ≥ 20

To switch between these operators, three values of σi are introduced which are σ1 ,
σ2 and σ3 . The tuning of these parameters are introduced in Sect. 4.
To automatically shift from global search to local search, AEHJ uses an adaptive
parameter, δ, which is successfully applied in many other works [1, 3]. The parameter δ
is defined as follows:

δ = |fmean /fbest − 1|, (14)

where f mean and f best denote the mean and best values of the objective function in the past
population, respectively. It should be noted that δ are reduced during the optimization
process. In AEHJ, the global search is performed first in which DE/rand/1 and DE/rand/2
are used while the values of δ is smaller than 10–4 . In contrast, the DE/best/1, DE/best/2
and Jaya operator are utilized.
Secondly, an elitist selection approach [3, 21, 22] is applied to choose the best
individuals for the next iteration. Hence, the flowchart of the proposed AEHJ algorithm
is shown in Fig. 1.

4 Numerical Example

In this section, size optimization of 10 bar truss example under multiple frequency con-
straints are performed for validating the accuracy and feasibility of AEHJ. The optimal
cross-sectional areas obtained by AEHJ are compared with the results of other methods
available in the literature. For fairing comparison, all algorithms have the same number
of population and generations which are 20 and 1000, respectively. For the stopping
condition of the optimization process, the value of δ is less than or equal to 10–6 or the
number of FE analyses is equal to 20000. Data in terms of material properties, lower and
higher bounds of size, and frequency constraints are provided in Table. 1. In addition,
performances of AEHJ, DE, and Jaya in 10 differential runs are compared in terms of
best weight, the number of FE analyses, average weight, worst weight, and standard
deviation of the weight.

4.1 Effects of Mutant Vector, σ1 , σ2 and σ3

This section presents the influence of mutant vector F, σ1 , σ2 , and σ3 are investigated for
a 10-bar planar truss example is shown in Fig. 2. It should be noted that a lumped mass
of 454 kg is added at all free nodes.
 An Enhanced Hybrid Jaya Algorithm for Size Optimization 171

Fig. 1. The AEHJ flowchart.

Figures 3, 4, 5 show the effect of mutant vector F, σ1 , σ2 , and σ3 on the number of

FEAs analyses and the best weights. It shows that when the values of mutant vector F,
σ1 , σ2 and σ3 are [0.4–1], 0.3, 0.4 and 0.3 respectively, the AEHJ has a good balancing
between the exploration and exploitation abilities. Thus, the values of mutant vector F,
σ1 , σ2 and σ3 of [0.4–1], 0.3, 0.4 and 0.3 are recommended for AEHJ.
172 N. T. T. Nguyen et al.

Fig. 2. The 10-bar planar truss.

Fig. 3. Effect of σi on truss weight.

4.2 Comparison with Other Methods

Table 2 compares the optimal solutions gotten by the AEHJ and other algorithms. The
optimal weights of the AEHJ, and CDJ are the best and the AEHJ solutions are better
than those of other algorithms (537.98 kg for the PSO, 542.75 for the GA, 524.4602
kg for DE, and 524.5072 for Jaya). Regarding the number of FEAs analyses, the AEHJ
is the best which needs 4020 while only DE and Jaya require 14920 and 10660 FE
analyses, respectively. Table 3 shows the first eight optimal natural frequencies gotten
by the AEHJ and other algorithms. It shows that no frequency constraint is violated.
Also, the convergence speed of the DE, Jaya, and AEHJ are compared as shown in
Fig. 6. It shows that AEHJ has the fastest decreasing convergence rate but it still has
good population diversity for obtaining the good optimal solution. Overall, the suggested
algorithm is the best one in terms of the good balance between the solution accuracy
and the convergence rate.
 An Enhanced Hybrid Jaya Algorithm for Size Optimization 173

Fig. 4. Effect of σi on the number of FEAs analyses.

Fig. 5. Effect of the mutant vector F on truss weight and number of FEAs analyses.

5 Conclusion
This paper successfully proposed an enhanced hybrid Jaya, AEHJ, for size optimization
of truss structures with multiple frequency constraints. AEHJ has a good balancing
174 N. T. T. Nguyen et al.

Table 2. Optimal results of the 10-bar planar truss from other algorithms.

Design variables PSO GA CDJ This current study

Ai (cm2 ) [23] [24] [1] DE Jaya AEHJ
1 37.7120 42.2340 35.0741 35.1311 35.4966 35.1901
2 9.9590 18.5550 14.7192 14.7738 14.8088 14.6857
3 40.2650 38.8510 35.2368 35.0130 34.9476 35.0885
4 16.7880 11.2220 14.6819 14.7680 14.5111 14.7486
5 11.5760 4.7830 0.6453 0.6450 0.6450 0.6451
6 3.9550 4.4510 4.5598 4.5553 4.5622 4.5585
7 25.3080 21.0490 23.7564 23.6386 24.1423 23.6539
8 21.6130 20.9490 23.6777 23.6449 23.3838 23.7448
9 11.5760 10.2570 12.3676 12.5391 12.0836 12.3374
10 11.1860 14.3420 12.4229 12.4253 12.5912 12.4892
Weight (kg) 537.9800 542.7500 524.4551 524.4602 524.5072 524.4565
Number of FE analyses – – 5160 14920 10660 4020
Worst weight (kg) – – 530.6843 524.8842 541.2042 530.9719
Average weight (kg) 540.8900 552.4470 525.1077 524.5181 529.7029 528.9004
Standard deviation 6.8400 4.8640 1.9596 0.1304 5.0528 2.9968

Table 3. The first eight optimal natural frequencies of the 10-bar truss obtained by different
algorithms.

Frequency no PSO [23] GA [24] CDJ [1] This current study

DE Jaya AEHJ
1 7.0000 7.0080 7.0000 7.0000 7.0000 7.0000
2 17.7860 18.1480 16.1937 16.1865 16.1970 16.1921
3 20.0000 20.0000 20.0000 20.0001 20.0000 20.0000
4 20.0630 20.5080 20.0003 20.0004 20.0002 20.0000
5 27.7760 27.7970 28.5499 28.5723 28.4524 28.5505
6 30.9390 31.2810 28.9458 29.0261 28.9739 28.9664
7 47.2970 48.3040 48.5791 48.5819 48.5787 48.5811
8 52.2860 53.3060 51.0674 51.1025 51.0562 51.0759

between global search and the local sear for obtaining good accuracy of the optimal
solution and the convergence speed, respectively. In AEHJ, four new parameters of F,
σ1 , σ2 , and σ3 are added and are also investigated for choosing of proper values.
 An Enhanced Hybrid Jaya Algorithm for Size Optimization 175

The 10 bar truss example is tested for AEHJ, it shows that the proposed method is the
best in terms of both good optimal solution and fast convergence speed. In general, AEHJ
is very promising to apply in other engineering optimization such as path optimization,
mechanism synthesis, plates, shell and etc.

Fig. 6. The weight convergence of the 10-bar truss by using the DE, Jaya and AEHJ.

Acknowledgment. This research was supported by Thai Nguyen University of Technology.

References
1. Nguyen-Van, S., Nguyen, T.T.N., Nguyen-Dinh, N., Lieu, Q.X.: Truss optimization under
frequency constraints by using a combined differential evolution and jaya algorithm (2021).
https://doi.org/10.1007/978-3-030-64719-3_95
2. Nguyen-Van, S., Nguyen-Dinh, N., Duong, P.T.M., Hung, N.Q., Nguyen, T.T.N.: The dimen-
sional synthesis of the four-bar mechanism with a symbiotic organisms search algorithm. In:
Advances in Engineering Research and Application. pp. 780–791. Springer, Cham (2021).
https://doi.org/10.1007/978-3-030-64719-3_85
3. Lieu, Q.X., Do, D.T.T., Lee, J.: An adaptive hybrid evolutionary firefly algorithm for shape
and size optimization of truss structures with frequency constraints. Comput. Struct. 195,
99–112 (2018)
176 N. T. T. Nguyen et al.

4. Mirjalili, S., Mirjalili, S.M., Lewis, A.: Grey wolf optimizer. Adv. Eng. Softw. 69, 46–61
(2014). https://doi.org/10.1016/j.advengsoft.2013.12.007
5. Mirjalili, S., Lewis, A.: The whale optimization algorithm. Adv. Eng. Softw. 95, 51–67 (2016).
https://doi.org/10.1016/j.advengsoft.2016.01.008
6. Holland, J.H.: Genetic algorithms and adaptation. In: Adaptive Control of Ill-Defined Systems.
pp. 317–333. Springer US (1984). https://doi.org/10.1007/978-1-4684-8941-5_21
7. Storn, R., Price, K.: Differential evolution: a simple and efficient adaptive scheme for global
optimization over continuous spaces. J. Glob. Optim. 23, 341–359 (1997)
8. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of ICNN’95 - Interna-
tional Conference on Neural Networks, pp. 1942–1948. IEEE. https://doi.org/10.1109/ICNN.
1995.488968
9. Cheng, M.Y., Prayogo, D.: Symbiotic organisms search: a new metaheuristic optimization
algorithm. Comput. Struct. 139, 98–112 (2014)
10. van Laarhoven, P.J.M., Aarts, E.H.L.: Simulated annealing. In: Simulated Annealing: Theory
and Applications, pp. 7–15. Springer Netherlands, Dordrecht (1987). https://doi.org/10.1007/
978-94-015-7744-1_2
11. Rashedi, E., Nezamabadi-pour, H., Saryazdi, S.: GSA: a gravitational search algorithm. Inf.
Sci. (Ny) 179, 2232–2248 (2009). https://doi.org/10.1016/j.ins.2009.03.004
12. Formato, R.A.: Central force optimization: A new nature inspired computational framework
for multidimensional search and optimization. Stud. Comput. Intell. 129, 221–238 (2008).
https://doi.org/10.1007/978-3-540-78987-1_21
13. Hsiao, Y.T., Chuang, C.L., Jiang, J.A., Chien, C.C.: A novel optimization algorithm: space
gravitational optimization. In: Conference Proceedings - IEEE International Conference on
Systems, Man and Cybernetics, pp. 2323–2328 (2005). https://doi.org/10.1109/icsmc.2005.
1571495
14. Venkata Rao, R.: Jaya: A simple and new optimization algorithm for solving constrained and
unconstrained optimization problems. Int. J. Ind. Eng. Comput. 7, 19–34 (2016)
15. Frans, R., Arfiadi, Y.: Sizing, shape, and topology optimizations of roof trusses using hybrid
genetic algorithms. In: Procedia Engineering, pp. 185–195. Elsevier Ltd (2014)
16. Sonmez, M.: Performance comparison of metaheuristic algorithms for the optimal design of
space trusses. Arab. J. Sci. Eng. 43(10), 5265–5281 (2018). https://doi.org/10.1007/s13369-
018-3080-y
17. Luu, T.V., Nguyen, N.S.: Parameters extraction of solar cells using modified JAYA algorithm.
Optik (Stuttg). 203, 164034 (2020)
18. Alomoush, A., Alsewari, A.A., Alamri, H.S., Zamli, K.Z.: Solving 0/1 knapsack problem
using hybrid HS and jaya algorithms. Adv. Sci. Lett. 24, 7486–7489 (2018)
19. Wolpert, D.H., Macready, W.G.: No free lunch theorems for optimization. IEEE Trans. Evol.
Comput. 1, 67–82 (1997). https://doi.org/10.1109/4235.585893
20. Storn, R., Price, K.: Differential evolution–a simple and efficient heuristic for global
optimization over continuous spaces. J. Glob. Optim. 11, 341–359 (1997)
21. Padhye, N., Bhardawaj, P., Deb, K.: Improving differential evolution through a unified
approach. J. Glob. Optim. 55, 771–799 (2013)
22. Do, D.T.T., Lee, J.: A modified symbiotic organisms search (mSOS) algorithm for optimiza-
tion of pin-jointed structures. Appl. Soft Comput. J. 61, 683–699 (2017)
23. Gomes, H.M.: Truss optimization with dynamic constraints using a particle swarm algorithm.
Expert Syst. Appl. 38, 957–968 (2011)
24. Lingyun, W., Mei, Z., Guangming, W., Guang, M.: Truss optimization on shape and sizing
with frequency constraints based on genetic algorithm. Comput. Mech. 35, 361–368 (2005)
 An Evaluation of Some Specifications of Turbine
Blades Made by 3D Printing Technology
and Processed on CNC Milling Machines

Nguyen Hoang Hiep1 , Le Hong Ky1(B) , Pham Hoang Anh1 , Le Hong Ky Duyen2 ,
and Tran Vinh Hung3,4
1 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
kylh@vlute.edu.vn
2 International University, National University HCM City, Dı̃ An, Vietnam
3 Faculty of Mechanical Engineering and Mechatronics, PHENIKAA University, Hanoi 12116,
Vietnam
4 A&A Green Phoenix Group JSC, PHENIKAA Research and Technology Institute (PRATI),

No. 167 Hoang NganCau Giay, Trung Hoa, Hanoi 11313, Vietnam

Abstract. This paper presents the research results on manufacturing turbine

blades in turbocharger machined on Haas VF2 module TRT 160 5-axis CNC
milling machine and by 3D printing technology on Metal 3D printer HBD-280
series with the same material AlSi10Mg. The research sample is the compressed
turbine blades in the HX40W turbocharger mechanism. CAD data files for machin-
ing programming and 3D printing of turbine blades are scanned from an ATOS
Core 80 scanner with GOM and Geomagic Design X softwares. Some specifica-
tions of the two manufacturing methods are also presented. Research results show
that the error of average diameter and surface roughness of 3D printing technology
are larger than those ones of machining on a CNC milling machine while mechan-
ical properties are lower. This shortcoming can be overcome by heat treatment to
improve particle organization and mechanical properties.

Keywords: 3D printing · CNC milling · Microstructure · Rapid prototyping ·

Surface roughness · Turbine blade

1 Introduction
Turbocharger for engines is widely used not only in the field of motor vehicles but
also in the marine and aviation fields. The turbine blade of the turbocharger is a part
with a complex contour, freeform surface without regularity, and there are no studies on
geometric modeling, so it cannot be designed with a common software [1–6]. Previously,
the manufacture of turbine blades was mainly by casting in metal molds [7–9]. Currently,
it is possible to design turbine blades by reverse engineering from the original model
through specialized scanning equipment, and after processing and editing with software,
the desired design file will be obtained [10, 11]. Studies [12–16] show that complex
surfaces such as turbine blades can be machined on 5-axis CNC Milling machines.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 177–187, 2023.
https://doi.org/10.1007/978-3-031-22200-9_19
178 N. H. Hiep et al.

Authors Nguyen Chi Thong and Le Trung Hau have published research results on the
influence of technological parameters on the accuracy of diameter and surface roughness
of compressed turbine blades in the turbocharger mechanism machined on HASS VF2
module TRT 160 5-axis CNC Milling machines [17]. In general, although it is possible
to process turbine blades on 5-axis CNC Milling machines, in terms of economic and
technical criteria, it has not yet replaced the casting method in the traditional metal
mold. Today, the technology of Additive Manufacturing (AM) or 3D Printing using
developed metal materials is considered a breakthrough in the manufacturing technology
industry. AM/3D printing is not only applied to rapid prototyping, but this technology is
also applied in many fields such as the automotive industry, aerospace and biomedical
engineering to manufacture metal parts for direct use [18]. From the design file, complex
structural parts such as turbine blades can be manufatured by AM technology through
the following basic steps: printing, cleaning, and sintering [19]. Selecting which method
to manufacture turbine blades is an issue that requires scientific research and evaluation.

2 Research Sample and Measuring Device

2.1 Specifications of Turbine Blades

The turbine blade model selected in this study is a compressed blade part in the HX40W
turbocharger mechanism (Fig. 1a).

a) b)
Fig. 1. Designing turbine blades from samples

2.1.1 Materials
The Model Niton material analyzer is used to determine the material for making the
compressed turbine blades in the turbocharger HX40W, which is an aluminum alloy
with the basic element composition of aluminum (98.4%), copper (1.17%), iron (0.26%)
and other elements.
 An Evaluation of Some Specifications of Turbine Blades Made 179

2.1.2 Geometric Parameters

Coordinate Measuring Machine (CMM) is used to determine the average value of diam-
eters Ø1 and Ø2 at 3 measurements and the average value of large blades as shown in
Fig. 1. The measurement values and the average diameter of the HX40W compressed
turbine blade model are summarized as shown in Table 1.

Table 1. Diameter measurement results of turbine blade model

Measurement position Measurement results (mm)

1st time 2nd time 3rd time ØAvS
Ø1 59.9999 59.9998 60.0001 59.9999
Ø2 84.5002 84.5003 84.4998 84.5001

2.1.3 Surface Roughness

The study used the Mitutoyo SJ-210 roughness meter to determine the surface roughness
of 3 large blades (Fig. 2). The measurement values and average value of surface roughness
of the HX40W compressed turbine blade model are summarized as shown in Table 2.

Table 2. Surface roughness measurement results of turbine blades

1st time 2nd time 3rd time Overall mean

Ra Rz Ra Rz Ra Rz RaAvCNC RzAvCNC
2.655 15.453 2.447 15.277 2.734 13.116 2.612 14.615

2.2 Turbine Blade Design

The study used ATOS Core 80 scanner and GOM Inspect software to generate point
cloud data from the sample HX40W compressed turbine blades (Fig. 1a). After the
point cloud data is processed with Geomagic Design X software, a CAD data file can
be created for machining and manufacturing turbine blades as Fig. 1b [20].

3 Machining Turbine Blades on CNC Milling Machines

On the basis of the analysis of the sample turbine blade material composition, the study
uses the processing material of aluminum alloy A6061 with the chemical composition
as shown in Table 3. This material has good anti-corrosion properties, low melting
temperature and hardness. The molded ingots are heated to 5400 °C before pressing,
and then uniformly annealed for 5 h.
180 N. H. Hiep et al.

Table 3. Chemical composition of aluminum A6061

Elements Si Fe Cu Mn Mg Cr Zn Ti Al
% 0.4–0.8 0.7 0.15–0.4 0.15 0.8–1.2 0.04–0.35 0.25 0.15 Remaining

Turbine blades are machined through 4 operations:

Operation 1: Turning the positioning part and clamping for the next operations,
drilling holes smoothly. HAAS TL2 lathe machine.
Operation 2: Milling and breaking the cylindrical workpiece to create the designed
workpiece shape and widen the hole after drilling. HAAS VF3 Milling machine.
Operation 3: Finely milling the Turbine blades on Haas VF2 model TRT 160 5-axis
CNC Milling machine with cutting speed V = 60 (m/min), feedrate S = 0.24 (mm/rev)
and cutting depth t = 0.35 (mm), Fig. 2, [17]. Programming and simulation of machining
Turbine blades with Matercam software, machining on Haas VF2 model TRT 160 5-
axis CNC Milling machine corresponding to finishing work. Some important factors to
determine: machining workpiece, feedrate strategy, order of machining strategies, cutting
mode, cutting tool when machining. In this operation, simultaneous 5-axis feedrate paths
are used throughout the programming process. Three steps are performed, including:
finishing of turbine blades, finishing of cylindrical surfaces and machining of rounded
corners. The program Gcode is transferred by POSTPROCESSOR to the CNC machine
for processing.
Operation 4: Cutting off the positioning and clamping is done on a universal lathe.

Fig. 2. Machining turbine blades

4 Manufacture of Turbine Blades by 3D Printing Technology

The study used AlSi10Mg as the material for making turbine blades (Table 4). It is
a typical casting alloy with good casting properties and is commonly used for casting
parts with thin walls and complex geometrical shapes. AlSi10Mg is a material with great
 An Evaluation of Some Specifications of Turbine Blades Made 181

strength and hardness, so it is suitable for parts subjected to high-speed working loads.
AlSi10Mg components are ideal for applications that require a combination of good
thermal properties and low weight such as turbine blades.

Table 4. Chemical composition of aluminum AlSi10Mg

Elements Si Fe Cu Mn Mg Cr Zn Ti Al
% 9.0–11.0 ≤0.25 0.15–0.4 ≤0.1 0.25–0.45 ≤0.05 ≤0.1 ≤0.15 Remaining

Simulation of the 3D printing process is carried out on Simufact software with the
same technology mode as when manufactured on the printer, in which layer thickness
= 0.02 mm, Scan width = 0, Velocity = 1.0 m/s, Power = 500 W, Efficiency = 0.5
(Fig. 3b). Before simulation, the meshing of survey objects is carried out with geometrical
parameters as shown in Fig. 3a. The results of running simulations on any layer show
that the model has very little distortion, and the areas of the distortion blade profile are
larger than the rest (Fig. 4a). Effective stress is unevenly distributed in the part, and the
body and base of the part have a greater effective stress than the blade profile (Fig. 4b).

Fig. 3. Simulation mode setting

Turbine blades are manufactured by 3D printing technology using Metal 3D printer

HBD-280 series [21] controlled by Voxeldance additive software (Fig. 5b). The forming
chamber is initially filled with an inert gas to minimize oxidation of the AlSi10Mg and
then heated to the optimum forming temperature. A thin layer of metal powder is spread
on the printing bed and a high-powered laser scans the cross-section of the part, fusing
the metal particles together and creating the next layer. When the scanning is complete,
the bed moves down by one layer thickness and the coating spreads another thin layer
of metal powder as shown in Fig. 5a. The process is repeated until the entire part is
completed. After printing, cleaning and separating the part from the printing bed, the
turbine blade product is obtained as Fig. 5c. Since metal 3D printing supports are shaped
with the same material as the printed part and are always required to minimize warping
182 N. H. Hiep et al.

Fig. 4. Simulation results of the 3D printing process

and distortion that can result from high processing temperatures. When the barrel cools
to room temperature, the excess powder is removed and the product is heat treated while
still attached to the bed, reducing any residual stress.

Fig. 5. Turbine blade 3D printing process

5 Results and Discussion

After machining the turbine blades on a CNC milling machine (Fig. 2) and manufacturing
it by 3D printing technology (Fig. 5), similar to the sample part, a CMM coordinate
measuring device and SJ-210 Mitutoyo roughness meter (Fig. 6b, c) are used to measure
diameters (Ø1, Ø2) and surface roughness. The microstructure of the sample turbine
blade as well as the product processed on a CNC milling machine were analyzed and
3D printed using an Optika metallographic microscope at three different positions on
the large blade (Fig. 6a).

5.1 Microstructure
5.1.1 With the Original Model
See Fig. 7.
The image of the microstructure of the model turbine blade in the turbocharger
HX40W made by casting technology in all 3 positions shows coarse, large and branched
 An Evaluation of Some Specifications of Turbine Blades Made 183

a) b) c)
Fig. 6. Measuring some specifications of 3D printing product of turbine blades

Fig. 7. Image of the microstructure of the original model turbine blades in positions a) Close to
the edge of the blade 20x-T; b) Middle of the blade 20x-T; c) Close to the center of the shaft 20x-T

particles. There are many impurities in the model, with the pores distributed among the
branches, and the mechanical properties are not uniform in all directions. It is possible
that the turbine blade model uses molten casting technology, combining pressure for
processing. The analysis shows that the model may have undergone heat treatment
(aging-T6) for machining, and that the model is susceptible to fatigue damage.

5.1.2 With the Model Machined on CNC

The microstructure image of the model turbine blade machined on CNC shows that the
model has phase 2, containing alloying elements in the organization, and the element is
dissolved in solid solution α in the face-centered cubic lattice. The analysis shows that
the model may have undergone a heat treatment (aging-T6) (Fig. 8).

5.1.3 With the Model Made by 3D Printing Technology

See Fig. 9.
The model turbine blade is made by 3D printing technology using concentrated
heat energy to melt the crystalline powder metal and cover each layer of material. The
microstructure image shows that the model has metal droplets that flow and merge to
create good adhesion between the layers. On the same layer, there is the phenomenon
184 N. H. Hiep et al.

Fig. 8. Image of the microstructure of the turbine blades machined on CNC in positions a) Close
to the edge of the blade 20x-T; b) Middle of the blade 20x-T; c) Close to the center of the shaft
20x-T

Fig. 9. Image of the microstructure of the turbine blades made by 3D printing technology a) Close
to the edge of the blade 20x-T; b) Middle of the blade 20x-T; c) Close to the center of the shaft
20x-T

of not covering, local shrinkage and cavitation, and the phenomenon of shrinkage and
porosity. In general, the model turbine blade made by 3D printing technology has low
mechanical properties. This shortcoming can be overcome by heat treatment to improve
particle organization and mechanical properties.

5.2 Surface Roughness

The results of surface roughness measurement (Ra, Rz) together with the average values
(RzAvCNC, RaAvCNC) and (RzAv3D, RaAv3D) are shown in Tables 5 and 6. The graph
shows the average roughness of 3 times measured on the same surface, the same blade
of the turbine blade processed on a CNC milling machine (RzAvCNC), 3D printing
(RaAv3D) compared to the sample as shown in Fig. 10.
Tables 5 and 6 and Fig. 10 show that the Ra surface roughness of the 3D printing
technology on the Metal 3D printer HBD-280 series is larger than that one of the model
as well as when it is processed on the Haas VF2 module TRT 160 5-axis CNC milling
machine. It is completely similar with the measurement value Rz. This difference can
be completely overcome when machining by sandblasting or using a metal 3D printer
with higher machining accuracy.
 An Evaluation of Some Specifications of Turbine Blades Made 185

Table 5. Surface roughness measurement results of turbine blades processed on CNC milling
machines

1st time 2nd time 3rd time Overall mean

Ra Rz Ra Rz Ra Rz RaAvCNC RzAvCNC
1.408 7.420 1.636 6.534 1.656 6.079 1.567 6.678

Table 6. Surface roughness measurement results of turbine blades manufactured by 3D printing

technology

1st time 2nd time 3rd time Overall mean

Ra Rz Ra Rz Ra Rz RaAv3D RzAv3D
3.826 17.570 3.239 17.208 3.577 19.660 3.547 18.146

Fig. 10. Turbine blade surface roughness is processed on a CNC milling machine, 3D printing
compared to the original model

5.3 Diameter Ø1, Ø2

Measurement results of diameters (Ø1, Ø2) on 3 long blades with average values
(Ø1AvCNC , Ø2AvCNC ) and (Ø1Av3D , Ø2Av3D ) corresponding to 2 machining methods as
shown in Tables 7 and 8.
Table 9 shows that the average diameter size error (Ø1Av3D , Ø2Av3D ) of the 3D
printing technology is larger than that one when processed on a CNC milling machine
(Ø1AvCNC , Ø2Av3D ). This error can be overcome when using a metal 3D printer with
higher machining accuracy as well as in calculating CAD design files that note the
shrinkage of the part after printing and sintering at high temperature.
186 N. H. Hiep et al.

Table 7. Measurement results of large blade diameter of turbine blades processed by CNC milling
machine

Measurement position Measurement results (mm)

1st time 2nd time 3rd time ØAvCNC
Ø1 59.981 60.002 59.992 59.9918
Ø2 84.501 84.492 84.592 84.5280

Table 8. Measurement results of large blade diameter of turbine blades manufactured by 3D

printing technology

Measurement position Measurement results (mm)

1st time 2nd time 3rd time ØAv3D
Ø1 60.0324 60.0050 60.0039 60.0138
Ø2 84.3787 84.4656 84.2545 84.3663

Table 9. Average diameter error of the 2 machining methods compared with the model

Measurement position Measurement results (mm)

ØAvCNC Error ØAv3D Error
Ø1 59.9918 −0.0081 60.0138 0.0139
Ø2 84.5280 0.0279 84.3663 −0.1338

6 Conclusion

From the research sample of a compressed turbine blade in the HX40W turbocharger
mechanism, the ATOS Core 80 scanner with GOM and Geomagic Design X softwares
were used to create CAD data files for machining programming and 3D printing of tur-
bine blades. Haas VF2 module TRT 160 5-axis CNC milling machine and 3D printing
on Metal 3D printer HBD-280 series were used to manufacture turbine blades. Research
results show that the error of average diameter and surface roughness of 3D printing
technology is larger than that of CNC milling machine, in which the surface roughness
when processed on a 5-axis CNC milling machine is smaller than that of the sample.
Among the models studied, the CNC machined model has the best mechanical proper-
ties. Although metal 3D printing technology is not expected to replace most traditional
manufacturing methods, its outstanding features are seen as a breakthrough in the manu-
facturing industry, especially for parts with thin walls and complex geometries, subjected
to high-speed working loads, requiring a combination of good thermal properties and
low weight, such as turbine blades. The limited technical characteristics of 3D printing
technology compared to other manufacturing methods can be improved and enhanced by
 An Evaluation of Some Specifications of Turbine Blades Made 187

different solutions, and heat treatment to improve mechanical properties is one of them.
The study of the influence of the technology mode on the quality of the parts will yield
an optimal set of technologies for mass fabrication of turbine blades using 3D printing
technology.

References
1. Wu, C.: Arbitrary surface flank milling of fan, compressor, and impeller blades. J. Eng. Gas
Turbines Power 117, 534–540 (1995)
2. Tsay, D., Chen, H., Her, M.: A study on five flank machining of centrifugal compressor
impellers. J. Eng. Gas Turbines Power 124, 177–181 (2002)
3. Harik, R., Gong, H., Bernard, A.: 5-axis flank milling: a state-of-the-art review. Comput.
Aided Des. 45, 796–808 (2012)
4. Zhou, Y., Schulze, J., Schaffler, S.: Dual spherical spline and its application in tool path planing
for 5-axis flank milling. In: Vision, C. (ed.) Imaging and Computer Graphics, pp. 19–32.
Springer, Theory and Applications (2013)
5. Wu, C.: Arbitrary surface flank milling-flank sam in the design and manufacturing of jet
engine fan and compressor airfoils. In: Proceedings of ASME Turbo Ex-po, Copenhagen,
Denmark (2012)
6. Chen, S.-L., Wang, W.-T.: Computer aided manufacturing technologies for centrifugal
compressor impellers. J. Mater. Process. Technol. 115(3), 284–293 (2001)
7. Takeuchi, Y., Watanable, T.: Generation of 5-axis control collision-free tool path and post-
processing for NC data. CIRP Ann 41, 539–542 (1992)
8. Tsay, D.M., Chen, H.C., Her, M.J.: A Study on five flank machining of centrifugal compressor
impellers. J. Eng. Gas Turbines Power, 177–181 (1992)
9. Bi, Q., Chen, H., Zou, X., Zhu, L., Ding, H.: Five-axis flank milling for design and manufacture
of turbocharger compressor impeller. In: Proceedings of ASME Turbo Expo 2014: Turbine
Technical Conference and Exposition, pp. 1–11 (2014)
10. https://www.gom.com/3d-software/gom-system-software/atos-professional.html
11. https://aie.com.vn/geomagic-design-x/
12. Stute, G., Storr, A., Sielaff, W.: NC programming of ruled surfaces for five-axis machining.
CIRP Ann. 28(1), 267–271 (1979)
13. Rehsteiner, F.: Collision-free five-axis milling of twisted ruled surfaces. CIRP Ann. 42(1),
457–546 (1993)
14. Wu, C.Y.: Arbitrary surface flank milling of fan, compressor, and impeller blades, ASME. J.
Eng, Gas Turbines Power 117, 534–539 (1995)
15. Tsay, D.M., Her, M.J.: Identification of undercutting and machining of twisted ruled surfaces
ASME. J. Manuf. Sci. Eng. 123(4), 731–738 (2001)
16. Tsay, D.M., Yan, W.F., Ho, H.C.: Generation of five-axis cutter paths for turbomachinery
components. ASME J. Eng. Gas Turbines Power 123, 50–56 (2001)
17. Thong, N.C., Hau, L.T.: The effects of technological parameters on the accuracy and surface
roughness of turbine blades when machining on CNC milling machines. In: Proceedings of the
International Conference on Advanced Mechanical Engineering Automation and Sustainable
Development (AMAS), pp. 285–291 (2021)
18. Wohlers, T.: Wohlers report: additive manufacturing and 3D printing state of the industry. In:
Annual Worldwide Progress Report, Wohlers Assoc (2015)
19. https://markforged.com/3d-printers/metal-x
20. https://www.3dsystems.com/applications/design-and-engineering
21. https://www.directindustry.com/prod/hbd-metal-3d-printer/product-216125-2240469.html
 An Optimal Cascade Reservoir Operation Based
on Multi-objective Water Cycle Algorithm

Thi-Kien Dao1 , Trong-The Nguyen2,3 , The-Vinh Do4(B) , Trinh-Dong Nguyen2,3 ,

and Vinh-Tiep Nguyen2,3
1 School of Computer Science and Mathematics, Fujian University of Technology,
Fujian 350118, China
2 University of Information Technology, VNU-HCM, Ho Chi Minh City, Vietnam
3 Vietnam National University, Ho Chi Minh City 700000, Vietnam
4 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

thevinh8880@tnut.edu.vn

Abstract. Optimal cascade reservoir scheduling is a complex problem related to

the broad interests of society, economy, and environment. This study proposes
a solution for dispatching cascade reservoir operation optimization based on the
multi-objective water cycle algorithm (MWCA). Search strategies for confluence,
diversion, seepage, evaporation, and rainfall are established in the MWCA by
simulating the natural water cycle process. A relative gravity mechanism under
multi-objective is constructed to achieve an adequate search for optimal solutions.
In the simulation section, the calculation results of the proposed approach are
compared with the other methods in the literature, e.g., particle swarm optimization
(MOPSO) and the genetic algorithm (NSGA-II). Compared results show that
MWCA is superior to other algorithms in calculation diversity and an effective
solution to the multi-objective optimal scheduling problem of cascade reservoir
groups.

Keywords: Multi-objective water cycle algorithm · Relative gravity

mechanism · Cascade reservoir group · Multi-objective optimal scheduling ·
Optimization

1 Introduction

Thanks to the nation having a lot of water resources that is possible having the potential
to get abundant electrical power from hydropower plants [1]. The main challenge of
optimal scheduling of cascade reservoirs is thoroughly considering the needs of irri-
gation, power production, flood control, and water supply, which are related to the
overall interests of society, economics, and the environment [2]. However, cascading
reservoirs have hydrological issues [3]. Economics, electricity, and hydraulic power are
all closely related and exhibit multi-objective, nonlinear, numerous variable, and pow-
erful constraint features [4]. Directly solving traditional optimization techniques like
linear, nonlinear, and dynamic programming is challenging [5]. Global optimality and

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 188–200, 2023.
https://doi.org/10.1007/978-3-031-22200-9_20
 An Optimal Cascade Reservoir Operation Based on Multi-objective 189

dimensionality of calamity are not guaranteed [6]. Convergence issues also affect intelli-
gent population optimization techniques such as the genetic algorithm (GA) [7], particle
swarm optimization (PSO) [8], spider swarm algorithm (SSA) [9], and cuckoo algorithm
(CA) [10] that may cause early convergence and slow speed. Therefore, introducing a
new optimization technique in the best scheduling of cascade reservoir groups to strike
a compromise between convergence performance and computation speed remains a hot
and challenging area of academic research.
The water cycle algorithm (WCA) [11] is a recently intelligent population optimiza-
tion algorithm that simulates the process of the water cycle and water flow in nature and
uses the relative gravity mechanism to make water droplets search for targets in the solu-
tion space. The optimal solution of the function! Simultaneously construct confluence,
diversion, seepage, evaporation, and rainfall optimization strategies are listed below:

• The confluence strategy makes multiple adjacent water droplets converge into a single
water droplet, and the quality of the water droplets after the confluence becomes more
extensive, which can effectively improve the search speed so that the water droplets
can quickly approach the optimal solution.
• The diversion strategy can prevent premature convergence caused by the excessive
mass of a single water droplet particle. When the water droplet particle is larger than
the limit value, the diversion is divided into multiple tiny water droplets to continue
the search;
• The seepage strategy judges that when the relative gravity of the water droplet particles
is relatively weak, it splits into multiple tiny water droplets for a local search around
the current position, which significantly improves the algorithm’s optimization ability;
• The evaporative rainfall strategy makes the relative gravity zero, that is, iterative stag-
nation. The intelligent water droplet particles evaporated and left the current posi-
tion, then randomly rained into the solution space, making them re-participate in the
search, effectively solving the problem of intelligent droplet particles and falling into
the optimal local solution.

The optimization process of the WCA can work well over the iteration mechanism
based on the search above strategies [12]. The WCA independently follows the current
search situation, changes the quantity and quality of water droplet populations, allocates
search resources intelligently, and causes the algorithm to adapt to the existing comput-
ing requirements. As a result, WCA, which has high resilience and robust search ability,
really recognizes the population’s intelligence. It has applications in numerous fields and
has recently emerged as a promising new component of the intelligent population algo-
rithm [13]. However, this technique has not yet been used for multi-objective reservoir
group scheduling.
This paper studies the multi-objective water cycle algorithm (MWCA) based on
the Pareto front solution. The gravity iteration mechanism under the multi-objective
optimization problem is constructed by updating location equations for the optimization
problem. The multi-objective optimal scheduling of cascade reservoirs with MWCA
provides better performance in finding optimal results.
190 T.-K. Dao et al.

2 Cascade Reservoir Optimal Scheduling Model

Cascade reservoirs are needed for both power generating and water supply operations,
such as shipping, agriculture, and industry, so both requirements should be considered
while designing a scheduling strategy [2]. Under the condition of meticulous considera-
tion of various constraints and the water inflow in each time interval in a given dispatch
period, the model is to maximize the power generation as much as possible to improve
the efficiency of the power supply and, at the same time, maximize the volume of the
water supply to enhance the efficiency of the water supply [3].
The highest amount of power should be produced to optimize output given the limited
natural streams, add additional demand to the grid, enhance power production during
the dispatch period, and lower thermal power output, energy conservation, and emission
reduction [2, 3]. The first objective function is expressed as follows.
   H
max E(T ) = Tt=1 N p=1 βp Qp,t Hp,t t
G
, (1)
s.t.0 ≤ QpG,min ≤ Qp,t
G ≤ Q G,max
p

In the formula, E(T ) is the total power generation of the reservoir group in the total
dispatching time period T , kW ∗ h; T is the total dispatching time period; N H is the
number of reservoirs; βp is the output coefficient of the reservoir p; Qp,t
G is the power
3
generation flow of reservoir p in period t, m /s; H p,t is the power generation head of
reservoir p in period t, m; t is a delta of the number of hours in period t, h.
The second objective function is the largest of the water supply. The water supply
includes water for municipal, industrial, and agricultural irrigation. Among these, the
industrial and municipal water supplies are drawn from reservoir areas, while the farm
water supply is primarily derived from water diversion between rivers. Enhance the
reservoir and watershed groups’ overall integration. The water supply volume improves
the dispatch period water supply effectiveness and full water energy use of cascade
reservoir groups.
⎧ T NN
⎨ max W (T ) =
⎪ (QA + Qp,t
C )
 t=1 p=1 p,t t
A,min
0 ≤ Qp,t ≤ Qp,t A , (2)
⎪
⎩ s.t. C,min
0 ≤ Qp,t ≤ Qp,t
C

In the formula, W (T ) is the water supply volume of the reservoir group in the total
dispatching time period T, m3 ; Qp,t A is the average water supply flow of agricultural

irrigation in the upper reaches of the reservoir p in the period t, m3 /s; Qp,t
C is reservoir p

with average industrial and municipal water flow during t period, m3 /s.
Constraints of the objective functions are dealt with as follows.

(1) Water balance constraints. Cascade reservoirs are closely related to each other.
The outflow of upstream hydropower stations, the amount of abandoned water, the
amount of incoming water in the interval, the amount of water supply for agricultural
irrigation, the power generation flow of the hydropower station at this level, the
amount of municipal water supply, and the amount of discarded water are all related
 An Optimal Cascade Reservoir Operation Based on Multi-objective 191

to this. The storage capacity of the hydropower station and the power generation
of the downstream hydropower stations have a great influence, and the relational
expression is:

Vp,t+1 = Vp,t + (Rp,t + Qp−1

G
+ Sp−1,t − Qp,t
A
− Qp,t
C
− Qp,t
G
− Sp,t )t , (3)
In the formula, Vp,t and Vp,t+1 represent the water storage capacity of reservoir p
at the beginning and end of period t, respectively, m3 ; Rp,t is the inflow of reservoir
G
p in the interval of t period, m3 /s; Qp−1,t and Qp,tG are the power generation flow of

reservoir p-1 and reservoir p in t period, m3 /s; S p-1,t and S p,t represent the abandoned
water of reservoir p-1 and reservoir p in t period flow, m3 /s.
(2) Constraints of agricultural water supply flow as follows.
A,min
0 ≤ Qp,t ≤ Qp,t
A
, (4)
A,min
In the formula, Qp,t is the minimum agricultural water supply flow in the upper
reaches of reservoir p in the period t, m3 /s.
(3) Industrial and municipal water supply flow constraints. The constraints are
expressed as follows.
C,min
0 ≤ Qp,t ≤ Qp,t
C
, (5)
C,min
In the formula, Qp,t is the minimum industrial and municipal water supply flow
of reservoir p at time t, m3 /s.
(4) Constraints on water storage are given as follows.

0 ≤ Vpmin ≤ Vp,t ≤ Vpmax , (6)

where Vpmin and Vpmax are the minimum and maximum water storage capacity of
reservoir p, respectively, m3 .
(5) Power generation flow constraints as the constraints are presented as follows.

0 ≤ QpG,min ≤ Qp,t
G
≤ QpG,max , (7)

In the formula, QpG,min and QpG,max are the minimum and maximum power
generation flow of reservoir p, respectively, m3 /s.
(6) Power generation output constraints are presented as:

0 ≤ Ppmin ≤ Pp,t ≤ Ppmax , (8)

In the formula, Ppmax and Ppmin are the upper and lower limits of the output of
reservoir p, respectively, in kW.

3 Multi-objective Water Cycle Algorithm

The WCA algorithm is inspired by the natural water cycle and water flow phenomenon
through the mechanisms and strategies [11]: gravity optimization, convergence strategy,
diversion strategy, seepage strategy, and evaporative rainfall strategy.
192 T.-K. Dao et al.

3.1 The Mechanisms and Strategies

The gravity optimization mechanism is used to suppose the optimal solution problem is
minf(x), l <= x <= u where x is an D-dimensional vector, and the upper and lower
bounds are l and u, respectively. WCA first calculates the function value corresponding
to the water droplet population. It is known from physical principles that the water flow
moves from a higher altitude to a lower altitude under the influence of gravity [11]. In
general, the optimal value in an optimization problem is unknown before calculation,
so WCA defines a water droplet t with a high altitude. It always flows toward the water
droplets with low altitudes, that is, moving towards the current better position. There
are multiple low-altitude water droplets that are simultaneously pulled by their relative
gravity, and the resultant force to construct relative gravity is as follows.


B

 xk − xk
j i
Fik = mi f xi − f xjk
k
, (9)
j=1
|xj − xik |
k

In the formula, B is the set of water droplets that is better than water droplet i; mi is
the mass of water droplet i, and the initial mass is m0 ; k is the current iteration number;
f (x) is the objective function value corresponding to water droplet i. A new position of
the water droplet is introduced according to the following iterative equation, namely.

xik+1 = xik + λg Fik (10)

where λg is a random number, λg ∈ (0,1).

The iterative process simulates the convergence strategy by calculating the Euclidean
distance and the relative altitude difference between the water droplets. The water
droplets’ relative position is judged in the solution space. Then the confluence operation
is performed on the water droplet groups close to each other. The distance calculation
formula and relative altitude difference are expressed as follows.

 D

dij = 
k k − x k )2 ,
(xi,h j,h (11)
h=1
  
fd = f (xi ) − f xj , (12)

where d ij is the Euclidean distance between droplets i and j; D is the dimension; f d is

the relative altitude. If the two water droplets satisfy dijk <d 0 and f d < f d0 , it is judged
that they are close to each other. The confluence strategy gathers all the water droplets
that are close to each other into a new water droplet optimal value. i.e.


NM
mm = mi , (13)
i=1

xm = xi,best ∀i ∈ NM , (14)

where mi is the mass of water droplets participating in the confluence; N M is the number
of water droplets participating in the confluence; mm is the mass of the new water droplets
 An Optimal Cascade Reservoir Operation Based on Multi-objective 193

after the confluence; xi,best is the optimal droplet position participating in the confluence;
x m is the position of the new water droplets after the confluence.
The diversion strategy is to avoid the problem that the mass of water droplets may
be too large, the population size may decrease sharply, and the confluence strategy may
cause premature convergence; it is set that when the mass of the water droplet exceeds
a certain multiple of the initial group of the water droplet, half of its mass is shunted
to form a new water droplet, and given the random relative gravity of the new droplet
particles [11].

ms = 0.5mi , (15)

Fs = Fi + σsp , (16)

where mi is the mass of the split water droplet; ms is the mass of the new water droplet
after the split; Fi is the relative gravity of the split water droplet; Fs is the relative
gravity of the new water droplet after the split; σsp is an N-dimensional random vector.
The diversion strategy is the balance of the confluence strategy, which plays a role in
preventing premature convergence caused by excessive mass, and at the same time, the
random distribution of new water droplets after diversion also improves the population
diversity to a certain extent.
The seepage strategy simulated water infiltration to the surface and constructed the
seepage strategy of WCA. A random search is performed in the vicinity of the currently
searched optimal solution to find possible optimal values. The process is described
sequentially, evaluating each dimension’s optimal global solution. One-dimensional
small-scale random perturbation of the number is presented as.:
new
xg,best = xg,best
old
+ σin , (17)
new is the new position of the current optimal water droplet after
In the formula, xg,best
old
seepage; xg,best is the original position of the current optimal water droplet; sigin is
new ) is better than
a-dimensional random vector. If the fitness of the new solution f(xg,best
old ), then Replace the global optimal solution before percolation with the new
f(xg,best
solution. Through the percolation strategy, WCA can randomly search for a better solu-
tion next to the current global optimal solution to replace the current solution, which
can effectively improve the local fine search ability of the algorithm and improve the
convergence accuracy.
The evaporative rainfall strategy is formulated as below. In the iterative process of
WCA, the current optimal solution x g,best is constantly replaced, and the water droplets
in x g,best have the highest relative altitude. The evaporative rainfall causes a waste of
search resources. The rain into random positions in the solution space and the iteration
is re-launched.

NRF = mg,best /m0 , (18)

xRF,i = l + λRF (u − l)∀i ∈ NRF , (19)

194 T.-K. Dao et al.

In the formula, mg,best is the water droplet mass of the current optimal solution; N RF
is the number of water droplets produced after rainfall; λRF is a random number, λRF ∈
(0,1); x RF,i is the position of each water droplet after rainfall.

3.2 Multi-objective Water Cycle Algorithm (MWCA)

The problem supposes the multi-objective optimization with f 1 (x), f 2 (x), …, f m (x) are
member objective functions [14] can be expressed as follows.
 
minf (x) = f1 (x), f2 (x), . . . , fm (x) , (20)

In the formula, M is the number of objective functions; f1(x), f2(x), …, fm(x) are
multiple objective functions. For such multi-objective optimization problems, there is
no global optimum; therefore, there is no unique minimum altitude point for WCA.
The Pareto archived evolution scheme is used and some improvements are made, and
a multi-objective imitation water cycle algorithm [15]. The non-inferior solution set
(corresponding to the points on the frontier of the Pareto) is considered to be the lowest
points of the terrain. By calculating the density information of the water droplets in the
non-inferior solution, the water droplets have a greater tendency to move towards the
area with a smaller density value. The improved multi-objective gravity optimization
mechanism can be expressed as follows:


Nr
1
 xk − xfront,j
Fik = mi [f xik − ffront,j ] ik , (21)
ρfront,j xi − xfront,j 
j=1

where N P is the number of water droplets located in the non-inferior solution set A;
pfront,j is the density information of the water droplet j in the non-inferior solution set;
x front,j is the position of the water droplet j in the non-inferior solution set; x front,j is the
water droplet j The corresponding Pareto frontier point.
Slow search is caused by being guided by many non-inferior solutions simultane-
ously. For the multi-objective optimization problem with constraints, the penalty func-
tion method is adopted to deal with it; that is, by adding the penalty function of the
constraints to each objective function, the optimization problem under constraints can
be transformed into unconstrained optimization.
Specific steps are as follows:
Step 1. Initialize the MWCA droplet population, including the initial iteration number
k = 1; the maximum number of iterations K max, the population number N , the initial
position of the droplet group {x 1 ,x 2 , …, x N }, the initial fitness {f(x 1 ), f(x 2 ), …, f(x N )}
and the initial mass m0 , generate the non-inferior solution set A of the current population
according to Pareto archived evolution.
Step 2 According to the quality and location information of water droplets, implement
the strategies of confluence, diversion, seepage, evaporation, and rainfall, and optimize
the quality of water droplets and the size of the water droplet population.
Step 3 Calculate the relative gravity F i of each water droplet under multi-objective
conditions by Eqs. (21) and (10) is used to update the position to obtain xik+1 .
 An Optimal Cascade Reservoir Operation Based on Multi-objective 195

Step 4 Calculate the multi-objective function fitness f(xik+1 ) of each water droplet in
the new population.
Step 5 Update the non-inferior solution set solution A of the water droplet group, if
the number of water droplets stored in A reaches the upper limit of A, remove the water
droplet with the highest density value in A according to the density information, and
store the new water droplet.
Step 6 Repeat steps (2)–(5) until the maximum number of iterations is reached. The
position of the water drop in the non-inferior solution set A is the optimal solution of the
multi-objective optimization problem, and its fitness for optimizing the values, listing or
plotting the corresponding Pareto boundaries, the decision maker can choose the related
solution with different expectations.

4 Optimal Cascade Reservoir Scheduling with MWCA

Five main power stations are called Gi (i = 01, 02, …, 05) in a cascade reservoir group
in a particular basin, e.g., the G01 is the power station that is the leading power station,
and G05 is the power station last one, that is the terminal power station. The scheduling
period for water intake is one year, in which January to March is the dry season, June
and August are the high water period, and other months are the flat water period. The
installed capacity, rated storage capacity, normal water level, and other parameters of
each reservoir and power station are shown in Table 1.

Table 1. Cascade hydropower station parameters

Hydropower Adjustment Installed Water Dead Normal Maximum

station characteristics capacity storage water water power
(G) (MW) capacity level level generation
(106 m3 ) (m) (m) flow (106
m3 )
G01 Seasons of 1100 100.6 432 383 1100
adjustment
G02 Runoff 810 0.32 136 145 800
regulation
G03 Runoff 1220 2.75 126 121 1200
regulation
G04 Years of 3500 275 128 125 3400
adjustment
G05 Seasonal 2400 30 105 102 2300
adjustment

4.1 Comparative Analysis

The obtained results of the proposed MWCA are compared with the other methods in
the literature, e.g., Multive objective Particles swarm optimization (MOPSO) [16] and
196 T.-K. Dao et al.

Non-dominated Sorting Genetic Algorithm (NSGA-II) [15]. The MOPSO [16] method
and the NSGA-II [15] algorithm are utilized for comparison analysis, and the above
research scenarios are optimized for scheduling calculations to confirm the efficacy
and superiority of the MWCA algorithm. Analysis of MWCA dispatching results since
the non-inferior solutions at the Pareto boundary are all practical solutions to multi-
objective optimization problems, the choice of non-inferior solutions depends on the
decision maker’s dispatch strategy preference; some hope to increase power generation,
while others hope to increase water supply.
Tables 2 and 3 show the obtained optimization scheduling results of the proposed
MWCA for average outputs of power generation and water supply. Analysis of MWCA
dispatching results since the non-domination at the Pareto boundary are all effective solu-
tions to multi-objective optimization problems, the choice of non-domination depends
on the decision maker’s dispatch strategy preference, some hope to increase power
generation, while others hope to increase water supply.

Table 2. The obtained optimization scheduling results for average annual outputs (MW h)

Month G01 G02 G03 G04 G05

1 4.89E+02 1.75E+02 3.02E+02 1.13E+03 6.63E+02
2 4.98E+02 1.55E+02 2.64E+02 9.83E+02 6.76E+02
3 5.11E+02 1.99E+02 3.13E+02 1.32E+03 6.63E+02
4 6.03E+02 2.12E+02 3.37E+02 1.68E+03 8.29E+02
5 6.27E+02 3.44E+02 3.68E+02 1.98E+03 9.04E+02
6 7.26E+02 4.28E+02 5.14E+02 2.29E+03 1.16E+03
7 8.38E+02 5.30E+02 6.31E+02 2.76E+03 1.14E+03
8 8.97E+02 5.22E+02 6.70E+02 2.97E+03 1.37E+03
9 8.24E+02 4.14E+02 4.60E+02 2.09E+03 1.02E+03
10 7.24E+02 3.05E+02 4.31E+02 1.77E+03 8.61E+02
11 6.30E+02 1.94E+02 3.50E+02 1.49E+03 7.76E+02
12 5.22E+02 2.12E+02 3.92E+02 1.32E+03 7.64E+02

Table 2 shows a specific concession made a compromise by the hydroelectric plant’s

production for the obtained optimization scheduling results for average annual outputs.
Table 3 displays the dispatched water supply with unit million cube meters (m3 ) of
optimization scheduling results of the MWCA approach. The hydroelectric plant’s output
is reduced during the dry season by maintaining a minimum water supply and adhering
to dead water level restrictions.
Figure 1 depicts the Pareto solution bounds for three methods with a non-inferior
solution in the objective function space. The three algorithms’ strong convergence sta-
bility is seen in Fig. 1, where the Pareto border distribution of the produced non-inferior
solutions is broader and more uniform. Especially in the region with an average annual
power generation greater than 0.14 · 106 kW h, MWCA has searched a large number of
 An Optimal Cascade Reservoir Operation Based on Multi-objective 197

Table 3. Dispatched water supply with unit million cube meters (m3 ) of optimization scheduling
results

Month G01 G02 G03 G04 G05

1 2.7E+00 2.2E−02 2.2E−01 4.1E+00 8.2E−01
2 2.7E+00 2.1E−02 2.1E−01 4.1E+00 8.2E−01
3 2.7E+00 2.3E−02 2.2E−01 4.1E+00 8.2E−01
4 2.7E+00 2.4E−02 2.4E−01 4.3E+00 8.2E−01
5 2.8E+00 3.1E−02 2.4E−01 4.3E+00 8.3E−01
6 3.2E+00 4.0E−02 2.8E−01 4.8E+00 8.3E−01
7 3.8E+00 5.2E−02 3.0E−01 4.8E+00 8.3E−01
8 4.1E+00 5.3E−02 3.1E−01 4.9E+00 8.3E−01
9 3.9E+00 5.1E−02 2.6E−01 4.6E+00 8.3E−01
10 3.3E+00 4.0E−02 2.4E−01 4.4E+00 8.2E−01
11 3.0E+00 3.5E−02 2.3E−01 4.4E+00 8.2E−01
12 2.7E+00 2.8E−02 2.3E−01 4.4E+00 8.2E−01

Fig. 1. Pareto solution bounds of the MWCA, NSGA-II, and MOPSO methods with a non-inferior
solution in the objective function space

non-inferior sets, while MOPSO and NSGA-II have only obtained a few solutions in the
region. In addition, the overall Pareto boundary is also better than the two algorithms.
The dispatch strategy of the MWCA non-inferior solution can provide an average annual
water supply of 0.295 billion, while MPSO and NSGA-II can only provide about 0.250
billion (Table 4).
The strategy effectively improves the diversity of the water droplet population, and
also enhances the local optimization performance of MWCA, enabling it to obtain more
non-inferior solutions. Table 2 compares the obtained dispatch results of the MWCA
with the MOPSO and NSGA-II, respectively, for the average annual optimization water
198 T.-K. Dao et al.

Table 4. A comparison of the obtained dispatch results of the MWCA with the MOPSO and
NSGA-II for the average annual optimization water supply and power generation ratio, respectively

Typical Algorithm Minimum Maximum Minimum Maximum Number of Computation

seasons power power water water non-inferior time
(year) generation generation supply supply solutions (s)
(106 kW (106 kW h) (106 m3 ) (106 m3 )
h)
Rainy MWCA 55.72 365.16 112.31 119.64 84 11.07
season MOPSO 80.47 358.41 112.82 118.06 51 9.82
NSGA-II 72.19 313.15 113.47 118.32 48 11.50
Normal MWCA 44.58 293.73 89.85 95.71 67.20 8.86
season MOPSO 64.38 286.73 90.26 95.25 40.80 7.86
NSGA-II 57.75 250.52 90.78 95.46 38.40 9.20
Dry MWCA 39.00 257.01 78.62 83.75 58.80 7.75
season MOPSO 56.33 250.89 78.97 83.34 35.70 6.87
NSGA-II 50.53 219.21 79.43 83.52 33.60 8.05

supply and power generation ratio. It can be seen from Table 2 that the dispatch strategy
obtained by MWCA can get the highest middle yearly power generation of 336.16
million kW h, which is 358.42 more than the 313.15 million kW h of MOPSO and
31.315 million kW h of NSGA-II respectively. At the same time, the maximum annual
water supply obtained by MWCA is 119.64 million m3 , which is 118.06 million m3 and
118.32 million m3 more than that of MOPSO and NSGA-II in the rainy season.
In addition, the average annual minimum annual power generation ratio of the non-
inferior solutions obtained by MWCA is smaller than the NSGAII and MOPSO. This
indicates that the non-inferior solutions obtained by MWCA have a more comprehensive
range and more robust search performance. The operations of the MWCA can get 84
non-inferior solutions, while MOPSO and NSGA-II only obtain 51 and 48 non-inferior
solutions for the rainy season, respectively. In computation time, the MWCA, MPSO,
and NSGA-II average 11.07s, 9.82s, and 11.50s for the rainy season. Although MWCA
is not the least time-consuming, with a similar time, MWCA obtains more non-inferior
solutions than the other two algorithms, reflecting the algorithm’s superiority in iterative
speed and performance.
Figure 2 depicts the optimal scheduling of cascade reservoirs of the MWCA com-
pared to the NSGA-II and MOPSO schemes. The graph of the scheduling approach of
obtained outputs demonstrates the logic and justification of using MWCA to schedule
cascade reservoir groups optimally. In addition, it also reflects the water replenishment
scheduling function of ’storing abundant and replenishing dryness.’ Provide enough
water supply to meet the needs of industry and people’s livelihood when the dry season
comes. It can be seen that the suggested MWCA provides the outstanding performance
of the MOPSO and NSGA-II algorithms.
 An Optimal Cascade Reservoir Operation Based on Multi-objective 199

Fig. 2. The optimal scheduling of cascade reservoirs of the MWCA compared to the NSGA-II
and MPSO schemes

5 Conclusion

This paper suggested a solution for optimal cascade reservoir operation scheduling of
cascade reservoirs based on the multi-objective water cycle algorithm (MWCA). Multi-
objective optimal scheduling is mathematically modeled with the two most significant
objective functions, e.g., water supply and power generation, to solve the cascade reser-
voir scheduling problem. The multi-objective imitation water cycle algorithm (MWCA)
is designed via processing efficiently boost population diversity, search efficiency, con-
vergence, avoid local optimum, and avoidance of unnecessarily iterative processes. The
calculation results in the simulation of the proposed MWCA are compared with the other
methods in the literature, e.g., particle swarm optimization (MOPSO) and the genetic
algorithm (NSGA-II). Compared results show that MWCA offers a practical resolution
to the multi-objective optimal scheduling problem of cascade reservoir groups and has
promising research and technical application prospects.

Acknowledgments. This study was supported by the Thai Nguyen University of Technology.

References
1. Cosslett, T.L., Cosslett, P.D.: Water Resources and Food Security in the Vietnam Mekong
Delta. Springer (2014)
2. Nguyen, T.-T., Wang, H.-J., Dao, T.-K., Pan, J.-S., Liu, J.-H., Weng, S.-W.: An improved slime
mold algorithm and its application for optimal operation of cascade hydropower stations. IEEE
Access. 8, 1 (2020)
3. Dao, T.-K., Chang, K.-C., Chu, K.-C., Nguyen, T.-T.-T., Ngo, T.-G., Nguyen, T.-T.: A solution
for cascade hydropower station reservoirs optimal operation based on improved bat algorithm
BT. In: Advanced Presented at the Machine Learning Technologies and Applications (2021)
4. Tsai, C.F., Dao, T.K., Pan, T.S., Nguyen, T.T., Chang, J.F.: Parallel bat algorithm applied to
the economic load dispatch problem. J. Internet Technol. 17, 761–769 (2016). https://doi.org/
10.6138/JIT.2016.17.4.20141014c
200 T.-K. Dao et al.

5. Nguyen, T.-T., Wang, M.-J., Pan, J.-S., Dao, T., Ngo, T.-G.: A load economic dispatch based on
ion motion optimization algorithm BT. In: Presented at the Advances in Intelligent Information
Hiding and Multimedia Signal Processing (2020)
6. Nguyen, T.T., Ngo, T.G., Dao, T.K., Nguyen, T.T.T.: Microgrid operations planning based on
improving the flying sparrow search algorithm. Symmetry 14 (2022)
7. Nguyen, T.-T., Shieh, C.-S., Horng, M.-F., Dao, T.-K.: A genetic algorithm with self-
configuration chromosome for the optimization of wireless sensor networks
8. Shi, Y., Eberhart, R.C.: Empirical study of particle swarm optimization. In: Proceedings of the
1999 Congress on Evolutionary Computation, 1999. CEC 99, vol. 3, pp. 1945–1950 (1999)
9. Cuevas, E., Cienfuegos, M.: A new algorithm inspired in the behavior of the social-spider for
constrained optimization. Expert Syst. Appl. 41, 412–425 (2014)
10. Yang, X.-S., Deb, S.: Cuckoo search via Lévy flights. In: 2009 World Congress on Nature &
Biologically Inspired Computing (NaBIC), pp. 210–214. IEEE (2009)
11. Eskandar, H., Sadollah, A., Bahreininejad, A., Hamdi, M.: Water cycle algorithm—a
novel metaheuristic optimization method for solving constrained engineering optimization
problems. Comput. Struct. 110, 151–166 (2012)
12. Nasir, M., Sadollah, A., Choi, Y.H., Kim, J.H.: A comprehensive review on water cycle
algorithm and its applications. Neural Comput. Appl. 32(23), 17433–17488 (2020). https://
doi.org/10.1007/s00521-020-05112-1
13. Nguyen, T.T., Pan, J.S., Dao, T.K.: An improved flower pollination algorithm for optimizing
layouts of nodes in wireless sensor network. IEEE Access. 7, 75985–75998 (2019). https://
doi.org/10.1109/ACCESS.2019.2921721
14. Dao, T.-K., Nguyen, T.-T., Nguyen, V.-T., Nguyen, T.-D.: A hybridized flower pollination
algorithm and its application on microgrid operations planning. Appl. Sci. 12, 6487 (2022)
15. Deb, K., Pratap, A., Agarwal, S., Meyarivan, T.: A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6, 182–197 (2002)
16. Coello Coello, C. A., Reyes-Sierra, M.: Multi-objective particle swarm optimizers: a survey
of the state-of-the-art. Int. J. Comput. Intell. Res. 2, 287–308 (2006). https://doi.org/10.5019/
j.ijcir.2006.68
 Analysis of Insulation Characteristics
of C2 F6 -N2 Mixtures

Pham Xuan Hien1(B) and Do Anh Tuan2,3

1 Department of Mechanical Engineering, University of Transport and Communications, Ha
Noi, Vietnam
hienpx@utc.edu.vn
2 National Center for Technological Progress, Ministry of Science & Technology, Ha Noi,
Vietnam
tuanda@acit.com.vn, datuan@most.gov.vn
3 A Chau Industrial Technology Joint Stock Company, Ha Noi, Vietnam

Abstract. This paper studied the insulation characteristics of C2 F6 -N2 mixture

based on the electron transport coefficients. The electron transport coefficients in
C2 F6 -N2 mixture were calculated using Monte Carlo method and the electron
collision cross section set for C2 F6 and N2 molecules. The present results, espe-
cially the calculated limiting field strength in C2 F6 -N2 mixtures, will be helpful
for the purpose of searching for suitable candidate for replacement of SF6 gas.

Keywords: C2 F6 · C2 F6 -N2 mixture · Insulation characteristics · SF6

1 Introduction
Perfluoroethane (C2 F6 ) plays a vital role in etching plasma processing [1–3], PECVD
chamber cleaning applications [4]. It is also widely used as a gaseous dielectric in
discharge switches and gas insulator in high voltage equipment [5–7] because of its
strongly electron negative character. The global warm potential (GWP) of C2 F6 is 9,200
[8] whereas that of SF6 is 23,900 [9]. Therefore, C2 F6 is expected to be an alternative to
substitute SF6 . Since the late 1970s, the research of SF6 replacements has been focused
on the mixture of SF6 or new eco-friendly gases with buffer gases. The buffer gases are
usually natural gases, such as N2 , CO2 , and dry air. In order to evaluate the potential to
replace SF6 gas, the insulation characteristics of gases or gaseous mixtures should be
analyzed. For these purposes, efficient numerical methods such as Boltzmann equation
analysis and Monte Carlo technique are widely used [10–12]. In this study, the insu-
lation characteristics of C2 F6 -N2 mixture were analyzed using Monte Carlo method in
comparison with the others.

2 Analysis
The insulation characteristics of gas mixtures are usually evaluated based on their dielec-
tric breakdown properties, such as density-normalized effective ionization coefficients,

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 201–207, 2023.
https://doi.org/10.1007/978-3-031-22200-9_21
202 P. X. Hien and D. A. Tuan

and critical electric field strength. Monte Carlo method and Boltzmann equation meth-
ods are common theoretical approaches to obtain these properties. Both methods require
the electron collision cross section set for using gas. The reliability of calculated results
depends on the accuracy of input data. Therefore, in this study, the electron collision
cross-section sets were chosen from [13] for the C2 F6 molecule and from [14] for the N2
molecule. The electron collision cross sections set for C2 F6 molecule, which includes the
momentum-transfer cross section, three vibrational-excitation cross sections, the disso-
ciative attachment cross section, two electronic excitation cross sections, the dissociation
cross section and the ionization cross section. This set has been obtained by using the
multi-term approximation of the Boltzmann equation. The multi-term approximation of
the Boltzmann equation is more appropriate than the two-term approximation to. How-
ever, the density-normalized effective ionization coefficients could not be applied for
electron collision cross section set, which includes fairly large inelastic cross sections
calculated, although they are important coefficients to evaluate the dielectric breakdown
properties. Therefore, in this study, the Monte Carlo method [15] was applied to calculate
the electron transport coefficients for C2 F6 -N2 mixtures. The electron transport coeffi-
cients, which include the electron drift velocity W, the density-normalized longitudinal
diffusion coefficient NDL , the density-normalized effective ionization coefficients (α −
η)/N. The Monte Carlo method has been presented in more detail by Tagashira et al.
[15] and Santos et al. [16], and the flow chart also shown in the Fig. 1. For each E/N, we
take a sample of 2000 electrons with initial energy 1i = 0 in TRIES-N2 mixture with
total pressure p = 1 Torr and temperature T = 300 K.
The electron collision cross section set for C2 F6 molecule from [13] has been slightly
modified before being used to calculate the electron transport coefficient for C2 F6 -N2
mixtures. In this study, the measured values of the electron drift velocity W and the
density-normalized longitudinal diffusion coefficient NDL in pure C2 F6 from [13] were
also presented. These parameters in the range of 0.1–100 Td at 0.1–500 Torr were
measured using the double shutter drift tube [13].

3 Results and Discussion

The variances of the electron transport coefficients in pure C2 F6 , pure N2 , and their
mixtures for various concentration mixtures, were shown in Figs. 2, 3 and 4.
The electron drift velocities, W, in the C2 F6 -N2 mixtures as functions of E/N, were
shown in Fig. 2. In this figure, the calculated and measured values of W for the pure
C2 F6 and the pure N2 , were also included. It is clear that the calculated values of W are
in good agreement with experiments for the pure C2 F6 [13] and the pure N2 [14]. At the
same E/N, the values of W in the pure C2 F6 are larger than those in the pure N2 , and the
mixtures W stand between them in the almost range of E/N < 100 Td. Figure 3 shows the
density-normalized longitudinal diffusion coefficient, NDL , in the C2 F6 -N2 mixtures.
The trend of variation of the NDL and W is similar. Figure 4 shows the relationship
between the density-normalized effective ionization coefficients, (α − η)/N, for the pure
of C2 F6 , N2 , and their mixtures against E/N. The (α − η)/N in the C2 F6 -N2 mixtures
are suggested to be between those of the pure gases over the entire range of E/N. The
shallow minimums appear on all the curves around 130 Td. For a given E/N value,
Analysis of Insulation Characteristics of C2 F6 -N2 Mixtures 203

Fig. 1. Monte Carlo flow chart

204 P. X. Hien and D. A. Tuan

Fig. 2. The electron drift velocity, W, in the C2 F6 -N2 mixtures

Fig. 3. The density-normalized longitudinal diffusion coefficient, NDL , in the C2 F6 -N2 mixtures

the density-normalized effective ionization coefficients become smaller with increasing

amounts of C2 F6 in the mixtures. The limiting field strength values (E/N)lim for C2 F6 -
N2 mixtures are derived from the variation curves of (α − η)/N, as shown in Fig. 5.
The values of (E/N)lim for SF6 -N2 mixture from [11] were also included in Fig. 5. It is
clear that the (E/N)lim decreases with percent N2 content. For the same N2 gas content,
 Analysis of Insulation Characteristics of C2 F6 -N2 Mixtures 205

Fig. 4. The density-normalized effective ionization coefficients

Fig. 5. Limiting field strength (E/N)lim for C2F6 and SF6 mixed with N2

SF6 -N2 mixtures have larger the (E/N)lim . However, as discussed above, the calculated
values (E/N)lim for C2 F6 -N2 mixtures are still useful for insulation applications.
206 P. X. Hien and D. A. Tuan

4 Conclusions
In this paper, the electron collision cross section set for C2 F6 molecule has been slightly
modified before using to calculate the electron transport coefficients in C2 F6 -N2 mix-
tures. The calculated electron transport coefficients in the pure C2 F6 are in agreement
with those in experiment. The electron transport coefficients, which include the elec-
tron drift velocity, the density-normalized longitudinal diffusion coefficient, the density-
normalized effective ionization coefficients in C2 F6 -N2 mixtures, were firstly calculated
and analyzed in the range of 0.1–100 Td using the Monte Carlo method. The values of
limiting field strength (E/N)lim for C2 F6 -N2 mixtures were also presented in comparison
with those for SF6 -N2 mixtures. In view of the lack of data on the insulation character-
istics of C2 F6 -N2 mixtures, the present theoretical results will be helpful for the purpose
of searching for suitable candidates for replacement of SF6 gas.

References
1. Park, H.H., et al.: Characterization and removal of silicon surface residue resulting from
CHF3 /C2 F6 reactive ion etching. J. Appl. Phys. 76, 4596–4602 (1994)
2. Miller, P.A., Greenberg, K.E.: Period-doubling bifurcation in a plasma reactor. Appl. Phys.
Lett. 60, 2859–2861 (1992)
3. Chae, H., Sawin, H.H.: Plasma kinetic study of silicon-dioxide removal with fluorocompounds
in a plasma-enhanced chemical vapor deposition chamber. J. Korean Phys. Soc. 51, 978–983
(2007)
4. Christophorou, L.G., Hunter, S.R., Carter, J.G., Mathis, R.A.: Gases for possible use in diffuse-
discharge switches. Appl. Phys. Lett. 41, 147–149 (1982)
5. Pirgov, P., Stefanov, B.: Elastic and inelastic e-C2 F6 and e-C3 F8 cross sections from swarm
data. J. Phys. B: At., Mol. Opt. Phys. 23, 2879 (1990)
6. Schaefer, G., Schoenbach, K.H., Kristiansen, M., Strickland, B.E., Korzekwa, R.A., Hutche-
son, G.Z.: Influence of the circuit impedance on an electron beam controlled diffuse discharge
with a negative differential conductivity. Appl. Phys. Lett. 48, 1776–1778 (1986)
7. Hunter, S.R., Carter, J.G., Christophorou, L.G.: Electron transport studies of gas mixtures for
use in e-beam controlled diffuse discharge switches. J. Appl. Phys. 58, 3001–3015 (1985)
8. Christophorou, L.G., Olthoff, J.K.: Electron interactions with C3 F8 . J. Phys. Chem. Ref. Data
27, 889–913 (1998)
9. Zhong, L., et al.: Compositions, thermodynamic properties, and transport coefficients of high-
temperature C5 F10 O mixed with CO2 and O2 as substitutes for SF6 to reduce global warming
potential. AIP Adv. 7, 075003 (2017)
10. Wu, B.T., Xiao, D.M., Liu, Z.S., Zhang, L.C., Liu, X.L.: Analysis of insulation characteristics
of c-C4 F8 and N2 gas mixtures by the Monte Carlo method. J. Phys. D Appl. Phys. 39, 4204
(2006)
11. Zhong, L., Wang, J., Wang, X., Rong, M.: Comparison of dielectric breakdown properties for
different carbon-fluoride insulating gases as SF6 alternatives. AIP Adv. 8, 085122 (2018)
12. Tuan, D.A.: Analysis of insulating characteristics of Cl2 -He mixture gases in gas discharges.
J. Electr. Eng. Technol. 10, 1734–1737 (2015)
13. Hien, P.X., Tuan, D.A., Jeon, B.H.: Determination of the electron collision cross section set
for the C2 F6 molecule. J. Korean Phys. Soc. 69, 1755–1761 (2016)
14. Nakamura, Y., Tokyo Denki Univ., Tokyo, Japan. Private Commun. (2010)
 Analysis of Insulation Characteristics of C2 F6 -N2 Mixtures 207

15. Tagashira, H., Sakai, Y., Sakamoto, S.: The development of electron avalanches in argon at
high E/N values. II. Boltzmann equation analysis. J. Phys. D: Appl. Phys. 10, 1051 (1977)
16. Santos, F.P., Dias, T.H.V.T., Rachinhas, P.J.B.M., Stauffer, A.D., Conde, C.A.N.: The elec-
troluminescence of Xe-Ne gas mixtures: a Monte Carlo simulation study. IEEE Trans. Nucl.
Sci. 45, 176–185 (1998)
 Analysis of Surface Defects and Tool Wear
in Edge Trimming of CFRPs by Optical Method

Dinh-Ngoc Nguyen and Anh-Tuan Dang(B)

Faculty of International Training, Thai Nguyen University of Technology, Thai Nguyenv,

Vietnam
anhtuanck@tnut.edu.vn

Abstract. Machining of CFRP composites are usually accompanied by appear-

ing defects induced in the machined surface. These kinds of damage are crucially
evaluated in both qualitative and quantitative aspects. There have been few tech-
niques which can meet two requirements so far. In this study, a noncontact optical
method will be utilized to consider both quantitative and qualitative assessment
of machining quality. Multidirectional CFRP plates are edge trimmed without
lubricant. The variation of surface damage levels observed by optical method is
analyzed. Moreover, areal roughness parameter, Sa, is chosen as an indicator for
quantitatively evaluating machined surface. The analysis of tool wear mechanism
is also conducted to show the advantages of the optical method for the machined
fibrous composite surfaces.

Keywords: CFRP · Optical method · Keyence · Surface defect · Tool wear

1 Introduction
Carbon fiber reinforced plastics composites have been widely utilized in industries like
aerospace, automobile, sport equipment, etc. for decades. This is attributed to the advan-
tages compared with traditional materials such as high strength-to-weight ratio, lighter
weight, high corrosion resistance. In order to satisfy the required shapes of appliances,
composite materials are typically fabricated close to the final shape or near-net-shape
[1–3]. Hence, after demolding the secondary machining operations are normally needed
to remove the excess materials. This is to meet the final dimensions of working com-
posite structures. During machining process, defects induced are frequently observed in
the wall or edge of machined surfaces. There defects are in the forms of delamination,
matrix-fiber cracking, matrix smearing, fiber breaking, or fiber debonding [4–7]. There
have been researchers documenting that the mentioned damage has harmful impact on the
working ability of composite structures [3, 8–11]. Therefore, characterizing machining
damage has been necessary [7, 12–14]. Scanning electron microscope (SEM) techniques
have been widely adopted for qualitatively assessment [1, 15]. In terms of quantification
of machining defects, roughness criterion, firstly applied to metallic material machining,
has been applied in industry [12, 13, 16]. Ramulu et al. [1] showed that the combination
between average surface roughness, Ra , and root mean surface roughness, Rq , should not

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 208–216, 2023.
https://doi.org/10.1007/978-3-031-22200-9_22
 Analysis of Surface Defects and Tool Wear in Edge Trimming 209

be selected as an good parameter for describing machining quality of fibrous composite

materials. Inversely, the parameters of Ry and Ten-point height of surface roughness, Rz ,
are better descriptors of machined surface. J. Sheikh-Ahmad et al. [13] when correlating
the tensile strength of CFRP composite specimens to machining quality concluded that
roughness parameter of Rz is better correlated defects to tensile strength than rough-
ness parameter of Ra . However, several researchers have shown that surface roughness
is not good indicator for describing the machining damage variation. This means that
the variation of surface roughness cannot support for explaining machining damage
alteration [2, 3, 10, 17]. Surface roughness can be measured in machined surface based
on profile (Ra ) or areal roughness parameter (Sa ) in which fiber direction has strong
effect on the measured values of Ra [6, 15]. Regarding the effect of machining damage
on residual strength of composite structures, Haddad [3] shows that when compressive
strength increasing with decreasing surface roughness. It can be said that this result is
logical. However, no relationship between compressive strength and surface roughness
are observed in each subgroup of tested specimens. Eriksen [18] investigated the effect
of the surface roughness of three machined short-fiber-reinforced thermoplastics on the
strength properties given by Charpy impact, monotonic bending and flexural fatigue.
It is shown that there are not relations between strength of surface quality. Keyence
has been successfully applied to characterize machining quality of metallic materials
by researchers [19–21]. However, there have been few studies dealing with using this
technology in evaluating and analyzing machining quality of composite materials.
This study aims to analyze surface damage and tool wear during edge trimming of
multidirectional CFRP composite materials. A full experimental design including three
feed speed and two spindle speed will be selected for doing the tests. Four helix flute end
mills made of carbide are utilized to generate the diffident defect levels. Defects resulting
from machining test will be quantitatively and qualitatively evaluated by optical method.
Furthermore, tool wear and wear mechanism is also considered.

2 Experimental Setup

Tested specimens are fabricated by laminating P2352 prepregs according to stack-

ing sequence of [90°/90°/-45°/0°/45°/90°/-45°/90°/45°/90°]s. The stacked plates have
dimensions of 300 mm × 150 mm × 5.2 mm in which the cutting direction is along
150 mm length. The fabrication of and properties of CFRPs in this study can be referred
in [9, 22]. In order to generate the different damage levels appearing in machined surface,
three degrees of feed speed, Vf are combined with two levels of cutting speed, Vc . There
result totally six machining conditions. Radial depth of cut is 2 mm for all machining
tests. The detail of machining condition is listed in Table 1. Machining is taken along
150 mm direction of CFRP plates. Mazak vertical center smart 530C (Yamazaki Mazak
Corporation, Japan) with the maximum spindle speed of 12000 rpm is utilized to con-
duct the trimming tests. The devices are setup as shown in Fig. 1. Six new carbide helix
end mill with four flutes and 8 mm diameter corresponding to six cutting conditions
are chosen for doing the tests. Tool geometry is referred in Fig. 2. Twenty paths will
be carried out for each machining condition. For this reason, total 3 m long of cutting
distance is created for one combination between feed speed and spindle speed. The setup
210 D.-N. Nguyen and A.-T. Dang

of evaluation of microstructures of machined surfaces can be detailed in [9, 22]. The

surface metrology of machined surface is captured by using an optical 3D measure-
ment device, KEYENCE VHX-6000 Digital Microscope (Fig. 3). This device allows
the acquisition of areal surface roughness, as well as the surface topography. The areal
surface roughness is measured in the fix area of 5.0 mm × 5.0 mm. Microstructure of
machined surfaces is investigated by scanning electron microscope (SEM) referenced
under “JEOL-JSM 5410 LV” (Fig. 3).

Fig. 1. Experimental setup showing devices used during trimming tests.

8 mm

Fig. 2. Carbide milling cutter with four helix cutting edges.

3 Results and Discussions

Figure 4 presents the evolution of areal surface roughness, Sa , as a function of machining
condition and machining length. It is observed that at small machining length (0.3 m)
influence of machining parameters on Sa is clear, e.g. an increase in feed speed and a
decrease in cutting speed lead to augment areal surface roughness. When feed speed
increases from 500 mm/min to 1500 mm/min, Sa varies from 1.923 µm to 3.13 µm
(increase by 63%), and 1.68 µm to 2.83 µm (increase by 69%) for cutting speed of
150 m/min and 250 m/min respectively. The impacts of process parameters in this case
can be directly explained by the items of theoretical chip thickness which was introduced
by [23]. It means that machining with high cutting speed and low feed speed generates
 Analysis of Surface Defects and Tool Wear in Edge Trimming 211

Fig. 3. Optical device (Keyence) for evaluating machined surfaces.

Table 1. Machining conditions

Cutting speed (m/min) 150 250

Feed speed (mm/min) 500 1000 1500 500 1000 1500
Feed per tooth (mm/t) 0.0208 0.0417 0.0625 0.0125 0.025 0.0375
Radial depth of cut (mm) 2 2 2 2 2 2

smaller theoretical chip thickness as presented in Fig. 5. This makes machining easily
happening. The results as previously analyzed are agreement with those documented by
researchers [9, 22].

Fig. 4. Evolution of process parameters and machining length as a function of areal surface
roughness.
212 D.-N. Nguyen and A.-T. Dang

Fig. 5. Theoretical chip thickness as a function of machining parameters.

The evolution of Sa as shown in Fig. 4 can be correlated to the damage level variation
in machined surface optically measured by Keyence device shown in Fig. 6. There is
no damage observed in case of using 500 mm/min, this result is completely reflected in
the value of Sa as presented in Fig. 5. Machining damage obtained by this machining
condition exhibiting in Fig. 7. The machining damage level in two remains images
shows similarly damage degree where almost defects appearing at −45° layer positions
in forms of craters and fiber pullout. These values are also well described in Sa revolution
in Fig. 5.

(a) (b) (c)

No damage

Craters Craters

1000 µm 1000 µm 1000 µm

Fig. 6. Optical images showing machining damage level at machining length of 0.3 m when
using cutting speed of 150 m/min and feed speed of (a) 500 mm/min, (b) 1000 mm/min, and (c)
1500 mm/min.

Regarding the Sa values of machined surface at machining length of 3.0 m, it is

seen that when machining with feed speed of 500 areal surface roughness exhibits the
highest value compared with two remain feed speed. In particular, at this feed speed the
Sa resulting from cutting speed of 250 m/min is superior to that given by cutting speed
of 150 m/min corresponding to 5.33 µm and 4.12 µm respectively. However, it can be
 Analysis of Surface Defects and Tool Wear in Edge Trimming 213

No damage

Fig. 7. Topography of machined surface at machining length of 0.3 m when using cutting speed
of 150 m/min and feed speed of 500 mm/min.

seen that Sa obtained by other cutting conditions are not different compared to those
at short machining length as exhibited in Fig. 4. The dominance of Sa at feed speed of
500 mm/min can be similarly explained as those shown in [24]. This result reveals that
longer machining time and wider contacting area between cutting edges and workpiece
make cutting tools quickly worn. Consequently, worn tools extremely generate machined
surfaces.

(a) (b) (c)

Fiber pullout Fiber pullout

Craters

Craters
Craters

1000 µm 1000 µm 1000 µm

Fig. 8. Optical images showing machining damage level at machining length of 3.0 m when
using cutting speed of 150 m/min and feed speed of (a) 500 mm/min, (b) 1000 mm/min, and (c)
1500 mm/min.

The damage level shown by optical images in Fig. 8 well described variation of
Sa in Fig. 4. Indeed, Fig. 8.a corresponding to using feed speed of 500 mm/min and
cutting speed of 150 m/min shows the more irregular surface where machining damage
in shapes of craters are observed across machined surface. Also the damage degree in
case of using cutting speed of 250 m/min is higher than those generated by other feed
speed. Moreover, there is no different in machining damage level in case machining
length of 3.0 m and that of 0.3 m.
214 D.-N. Nguyen and A.-T. Dang

Micro break of
Sharp cutting cutting edge
edge

(a) Abrasion
(b) mechanism

Fig. 9. Optical images showing cutting edge cartography at machining length of 3.0 m when
using cutting speed of 150 m/min and feed speed of (a) 1500 mm/min.

Figure 9 presents the cutting edge cartography in two case, namely new and worn cut-
ting edges. It is observed that the radius of the new cutting edge shows the small radius,
round 6 µm for carbide cutting tools. Regarding longer cutting length of 3.0 m, it is visu-
alized the mechanism of wear showing abrasion in some parts of cutting edge. Addition-
ally, micro fragments along cutting edge is also observed. The detail of using Keyence
device in analyzing tool wear of machining composite materials will be presented in
another study. The mentioned results exhibit the ability of using optical technique to
replace or compare to those given by traditional technique such as SEM, non-confocal
laser.

4 Conclusions
Analysis of surface defects and tool wear resulting from edge trimming of CFRP com-
posites are studied in this article. Machining damage is quantitatively and qualitatively
characterized by optical method. Moreover, tool wear and mechanisms are also ana-
lyzed. The results show that when machining process reaches small cutting length, cut-
ting speed and feed speed, directly influencing on the theoretical chip thickness, purely
impact on the machining damage levels and areal surface roughness. However, at longer
machining length tool wear have dominant effect on surface quality compared with pro-
cess parameters. Cartography of cutting edges measured by Keyence device shows the
wear mechanism. This including abrasion in some parts along cutting edges and micro
fragments are observed. Qualitative and quantitative analysis of machining quality and
cartography cutting edge by Keyence device based on laser principle show the ability of
this technique for machining composite materials.

Acknowledgments. The authors wish to thank Thai Nguyen University of Technology for
supporting this work.

References
1. Ramulu, M.: Machining and surface integrity of fibre-reinforced plastic (1997)
 Analysis of Surface Defects and Tool Wear in Edge Trimming 215

2. Ghidossi, P., Mansori, M., Pierron, F.: Influence of specimen preparation by machining on
the failure of polymer matrix off-axis tensile coupons. Compos. Sci. Technol. 66(11–12),
1857–1872 (2006)
3. Haddad, M., et al.: Study of trimming damages of CFRP structures in function of the machin-
ing processes and their impact on the mechanical behavior. Compos. B Eng. 57, 136–143
(2014)
4. Hintze, W., Hartmann, D., Schütte, C.: Occurrence and propagation of delamination during
the machining of carbon fibre reinforced plastics (CFRPs)—an experimental study. Compos.
Sci. Technol. 71(15), 1719–1726 (2011)
5. Azmi, A.I.: Chip formation studies in machining fibre reinforced polymer composites (2013)
6. Haddad, M., Redouane Zitoune, F.E., Castanié, B.: Machinability and surface quality during
high speed trimming of multi directional CFRP. Int. J. Mach. Mach. Mater. (2013)
7. Arola, D., Ramulu, M.: An examination of the effects from surface texture on the strength of
fiber reinforced plastics. J. Compos. Mater. 33(2), 102–123 (2016)
8. Nguyen-Dinh, N., Bouvet, C., Zitoune, R.: Influence of machining damage generated during
trimming of CFRP composite on the compressive strength. J. Compos. Mater. 54(11), 1413–
1430 (2019)
9. Ngoc, N.D., Hue, N.T.: A study on qualitative and quantitative characterization of machin-
ing quality of aerospace composite structures. In: Advances in Engineering Research and
Application, pp. 94–101 (2020)
10. Ngoc, N.D., et al.: The characterization of machined damage of CFRP composite: comparison
of 2D and 3D surface roughness performance. In: Advances in Engineering Research and
Application, pp. 771–779 (2020)
11. Nguyen-Dinh, N., et al.: New tool for reduction of harmful particulate dispersion and to
improve machining quality when trimming carbon/epoxy composites. Compos. Part A: Appl.
Sci. Manuf. 131 (2020)
12. Palanikumar, K., Karunamoorthy, L., Karthikeyan, R.: Assessment of factors influencing
surface roughness on the machining of glass fiber-reinforced polymer composites. Mater.
Des. 27(10), 862–871 (2006)
13. Shahid, A.H., Sheikh-Ahmad, J.: Effect of edge trimming on failure stress of carbon fiber
polymer composites_Jamal Sheikh-Ahmad_Hay dung bai nay (2013)
14. Ghafarizadeh, S., Lebrun, G., Chatelain, J.-F.: Experimental investigation of the cutting tem-
perature and surface quality during milling of unidirectional carbon fiber reinforced plastic.
J. Compos. Mater. 50(8), 1059–1071 (2015)
15. Ramulu, M., Wern, C.W., Garbini, J.L.: Effect of fibre direction on surface roughness
measurements of machined graphite epoxy composite. Compos. Manuf. 4 (1993)
16. Duboust, N., et al.: An optical method for measuring surface roughness of machined carbon
fibre-reinforced plastic composites. J. Compos. Mater. 51(3), 289–302 (2016)
17. Hejjaji, A., et al.: Surface and machining induced damage characterization of abrasive water
jet milled carbon/epoxy composite specimens and their impact on tensile behavior. Wear
376–377, 1356–1364 (2017)
18. Eriksen, E.: The influence of surface roughness on the mechanical strength properties of
machined short-fibre-reinforced thermoplastics_Else Eriksen. Compos. Sci. Technol. (2000)
19. Kaynak, Y., Kitay, O.: Porosity, surface quality, microhardness and microstructure of selective
laser melted 316L stainless steel resulting from finish machining. J. Manuf. Mater. Process.
2(2) (2018)
20. Li, J., et al.: Machining precision controlling method for plane surface assisted by SPDT.
Compos. Adv. Mater. 30 (2021)
21. Skrzyniarz, M., Nowakowski, L., Blasiak, S.: Geometry, structure and surface quality of a
maraging steel milling cutter printed by direct metal laser melting. Materials (Basel) 15(3)
(2022)
216 D.-N. Nguyen and A.-T. Dang

22. Ngoc, N.D., Truong, P.V., Hue, N.T.: Effects of machining configurations and process param-
eters on the machining damage generated during milling CFRP structures. In: Advances in
Engineering Research and Application, pp. 400–406 (2022)
23. Janardhan, P., Sheikh-Ahmad, J., Cheraghi, H.: Edge trimming of CFRP with diamond
interlocking tools (2006)
24. Nguyen-Dinh, N., et al.: Surface integrity while trimming of composite structures: X-ray
tomography analysis. Compos. Struct. 210, 735–746 (2019)
 Anti-torque Controller Using Sliding Mode
Linear Quadratic Regulator Applied for Coaxial
BLDC Motor

Lam Cuong Quoc Thai1,3 , Van Trong Nguyen1,3 , Cong Thang Doan1,3 ,
Van Tu Duong1,2,3 , Huy Hung Nguyen1,4 , and Tan Tien Nguyen1,2,3(B)
1 National Key Laboratory of Digital Control and System Engineering (DCSELab), Ho Chi
Minh City University of Technology (HCMUT), 268 Ly Thuong Kiet Street, District 10, Ho Chi
Minh City, Vietnam
nttien@hcmut.edu.vn
2 Faculty of Mechanical Engineering, Ho Chi Minh City University of Technology (HCMUT),
268 Ly Thuong Kiet, District 10, Ho Chi Minh City, Vietnam
3 Vietnam National University Ho Chi Minh City, Linh Trung Ward, Thu Duc District, Ho Chi
Minh City, Vietnam
4 Faculty of Electronics and Telecommunication, Saigon University, Ho Chi Minh City, Vietnam

Abstract. The goal of this paper is to investigate a controller that enables not
only to reject the anti-torque but also to minimize the energy consumption of
a coaxial BLDC motor propelling for underwater vehicles. First, a structure of
the coaxial BLDC motor with two counter-rotating rotors is introduced. Then,
dynamic mathematical modeling of the BLDC motor is expressed in order to
realize the proposed controller, a free chattering sliding mode control approach
is toward the speed equal to two rotors in the opposite direction. To optimize the
energy, a linear quadratic regulator (LQR) based control law combined with the
sliding mode controller is proposed. Finally, the proposed controller is evaluated
through simulation results.

Keywords: Coaxial BLDC motor · LQR sliding mode control · Torque balance ·
Energy consumption optimization

1 Introduction
Autonomous underwater vehicles (AUVs) appeared for the first time at the university of
Washington as early as 1957 by Stan Murphy, Bob Francois. So far, AUVs are applied
in many fields such as commerce, research, military, etc., owing to the ability to perform
many functions underwater. In AUV design, the propulsion for propelling the UAV has
been utilized by various equipment/approaches. However, the selection of the propulsion
is still a challenge for researchers. In fact, most of the existing studies are using a single-
propeller motor to power AUVs that may cause contra-torque affecting to motor and AUV
body. Hence, the coaxial motor is a well-suited solution to replace the conventional motor
and is essential to help AUV in terms of good performance and improved efficiency. To

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 217–226, 2023.
https://doi.org/10.1007/978-3-031-22200-9_23
218 L. C. Q. Thai et al.

achieve that, it requires a controller so that the anti-torque induced by the two rotors are
canceled each other out.
To deal with anti-torque rejection, [1] has presented a mathematical model of BLDC
motor and PWM converter, in which the speed of the main/auxiliary rotor is controlled by
separated two close-loop PI controllers. A torque balance controller is employed for the
auxiliary rotor to reject the anti-torques of two rotors with a speed difference of two rotors.
The simulation results indicate that different responses with variable coefficient choices.
Huang et al. [2] proposed a PI controller to control the speed of rotor one and a sliding
mode voltage compensation controller to balance the anti-torque by using the perceived
electrical current of dual BLDC motors. The effectiveness of this control strategy has
been verified through the simulation result analysis. However, this control strategy cannot
meet the requirement for eliminating the anti-torque of a coaxial BLDC motor because
the torque coefficients and the masses of two rotors are unequal. Huang and Wang [3]
proposed an adaptive variable structure sliding control algorithm with torque balance
control strategy based on the voltage compensation, which is demonstrated to be more
effective than PI controller in aspect of response speed, steady-state. Huang and Wang
[4] presented an equivalent switching variable structure sliding mode control which is
used with the strong capacity of resisting disturbance and fast response rate; however, it
causes steady-state vibration.
To address this issue, this paper concentrates on the anti-torque control problem of
a coaxial BLDC motor propelling for the AUVs. In Sect. 2, a coaxial BLDC motor is
introduced, and its mathematical model is given. And an anti-torque controller using
LQR sliding mode algorithm is developed for controlling the speed of two rotors to
ensure robustness against the system parameters change in Sect. 3. The reliability and
performance of the proposed controller are verified through the simulation results in
Sect. 4. Finally, some main points of this research are summarized in Sect. 5.

2 Coaxial BLDC Motor

2.1 System Description

The proposed motor structure consists of two external rotors and one common stator,
which is split into two separate parts and independently excited to drive the two rotors.
Thus, we consider a coaxial motor structure consisting of two conventional BLDC motors
shown in Fig. 1, with the outer shaft BLDC motor called the primary rotor and the inner
shaft motor called the secondary rotor. The configuration of two rotors, such as dimension
parameters of yoke/slot/tooth, winding scheme, steel core shape, etc., is equivalent to
the total power designed for the coaxial BLDC motor as Pds . Hence, each BLDC motor
is required to reach a haft of the total power. The electrical power for each BLDC motor
is determined by [5, 6].
Pds
P= (1)
2
where
Pds : design power of coaxial BLDC motor.
 Anti-torque Controller Using Sliding Mode Linear Quadratic Regulator 219

Rotor 2 Rotor 1
Stator 2 Stator 1

Stator 2 Stator 1
Rotor 2 Rotor 1

Fig. 1. Principle diagram of coaxial BLDC motor

P: design power of BLDC motor.

Two main components of the rotor are magnets and a yoke (back iron). The permanent
magnets are fixed to the yoke rotor, which is assembled to the shaft via a flange, and
the ball bearing is mounted on the shaft, which is nested in the steel core of the stator
to enable rotation. The stator is composed of a steel core which is formed by many
electrical engineering steel laminations. They are insulated from each other and shape
have many teeth, which are used to wrap wire. Two stators are associated by four bolts
which are mounted through a steel core and fixed to the outer motor frame. To select the
conductor cross-section, the maximum current can be calculated based on the standard
evaluation system of logarithmic wire. Then, the result is compared with the amperage
in the BLDC motor when reaching the maximum power.

2.2 Mathematical Model

Assumptions about coaxial BLDC motor [7–9]:

• The motor is not saturated.

• The resistance and inductance of each phase are the same.
• Hysteresis, eddy current losses, and load torque are eliminated.
• All semiconductor switches are ideal.

The voltage of each phase is determined based on Kirchhoff law:

⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
Va R00 ia L−M 0 0 ia ea
d
⎣ Vb ⎦ = ⎣ 0 R 0 ⎦⎣ ib ⎦ + ⎣ 0 L − M 0 ⎦⎣ ib ⎦ + ⎣ eb ⎦ (2)
dt
Vc 00R ic 0 0 L−M ic ec

where
M : one phase mutual inductance (H).
R: one phase armature resistance ().
L: one phase armature self-inductance (H).
ia,b,c : each phase current (A).
ea,b,c : each phase back-emf (V) (Fig. 2).
The back EMF voltage is directly proportional to the speed of the rotor and is
calculated by the following equation:
 
ea,b,c = Ke ωm F θea,b,c (3)
220 L. C. Q. Thai et al.

Fig. 2. Equivalent circuit diagram of BLDC motor driven by two-level inverter

where
ωm : angular speed of the rotor (rad s−1 ).
Ke : back EMF coefficient.
θea,b,c : electrical angle of the rotor (rad) (θea = θe ; θeb = θe −2π/3; θec = θe +2π/3).
 
F θea,b,c : each phase back EMF references the function of rotor position and is
calculated as follows:
⎧
⎪
⎪ 1, 0 ≤ θea,b,c < 2π
⎪   3
⎨ 1 − π θea,b,c − 3 , 2π 3 ≤ θea,b,c < π
6 2π
 
F θea,b,c = (4)
−1, π ≤ θea,b,c < 3
5π
⎪
⎪
⎪
⎩ −1 + 6 θ
ea,b,c − 3 , 3 ≤ θea,b,c < 2π
5π 5π
π

Based on the law of conservation of energy, the relation between the angular speed
of the rotor and the electromagnetic torque is presented as the following equation:

d ωm
J + Dωm = Te − Tl (5)
dt
where
J : the inertia of the rotor and coupled shaft (kg m2 ).
D: friction coefficient (Nms rad−1 ).
Tl : load moment of the rotor (N m).
Te : the electromagnetic torque depends on the back-emf or Q current and is calculated
by the following equation:

Te = (ea ia + eb ib + ec ic )/ωm = kt iq (6)

Since this control algorithm is based on the FOC control technique, the mathematical
model of the BLDC motor needs to be converted to the D-Q coordinate by using the
Park transform as the following equation:
⎡ ⎤  ⎡ ⎤⎡ ⎤
d cos(θe ) cos(θe − 2π/3) cos(θe + 2π/3) a
⎣q⎦ = 2 ⎣ −sin(θe ) −sin(θe − 2π/3) −sin(θe + 2π/3) ⎦⎣ b ⎦ (7)
3 √ √ √
0 1/2 1/2 1/2 c

where a, b, c, d , q: may be current or voltage

 Anti-torque Controller Using Sliding Mode Linear Quadratic Regulator 221

Combining Eqs. (2), (5), (6), and (7), the mathematical model of the three-phase
BLDC motor in the D-Q coordinate is found as follows:
⎧  
⎨ iq = uq −  Riq − ωm Lid − ωm kt /L
id = ud − Rid + ωm Liq /L (8)
⎩
ω̇m = kJt iq − DJ ωm − TJL

In the FOC control technique, the D current is controlled so that it is equal to zero
and supposing that load torque equals zero. The dynamic equation is found with angular
velocity output:

JL RJ + LD RD + kt2
ω̈m + ω̇m + ωm = uq (9)
kt kt kt
Therefore, the state-space model of the coaxial BLDC motor is rewritten in the
following equations:
⎧
⎪
⎪ ẋ1 = x2
⎨ = A1 X1 + B1 uq1
ẋ2 = a1 x1 + b1 x2 + c1 uq1
(10)
⎪
⎪ ẋ3 = x4
⎩ = A2 X2 + B2 uq2
ẋ4 = a2 x3 + b2 x4 + c2 uq2

where
RD +k 2
a1 = − J11 L t1 ; b1 = − RJ1J+LD
1L
1
; c1 = kt1
J1 L ;
RD2 +kt2
2
a2 = − ; b2 = − RJ2J+LD 2
; c2 = Jk2t2L ;
 
J2 L  2 L    
0 1 0 1 0 0
A1 = ; A2 = ; B1 = ; B2 = ;
a b a b c1 c2
 11  2 2
x x
X1 = 1 ; X2 = 3 .
x2 x4
x1 = ωm1 −ωm1ref is the error between current velocity and desired velocity of motor
1.
x2 = ω̇m1 is the acceleration of motor 1.
x3 = ωm2 −ωm2ref is the error between current velocity and desired velocity of motor
2.
x4 = ω̇m2 is the acceleration of motor 2.

3 Anti-torque Controller
The control objectives for the coaxial motor are both to ensure the required output speed
of the primary motor and ensure the total torque acting on the whole system is zero.
Instead of feedback torque affecting the system to control, a speed tracking controller
is designed as an anti-torque controller. Based on characteristics such as simple design
and robustness with system parameter changes, the sliding mode algorithm is chosen
to design for the anti-torque controller of the secondary motor. A PI speed controller is
chosen to control the speed of the primary rotor and as shown in Fig. 3.
222 L. C. Q. Thai et al.

Fig. 3. Block diagram controller of coaxial BLDC motor

The energy requirement that coaxial BLDC motor consumes is minimized, while
control specifications are still guarantee. The objective function [10] is proposed as
follows:
∞
1
J = xT Qx + uqT Ruq dt (11)
2
0
 
q1 0
where Q = is the weighting function of states (positive semi-definite matrix).
0 q2
R = r is the weighting function of control variables (positive definite matrix).
Hamilton function is stated by the following equation:
1 T  
H (α, u, x, t) = x Qx + udT Ruq + λ Ax + Buq (12)
2
Based on initial condition x(0) = [00] and final condition x(∞) = [xd (∞)0], Riccati
equation are solved to satisfy the necessary condition for optimal control, and results
are expressed as follows:

uq∗ = −R−1 BT SX2

= −KX2 (13)
 
where K = R−1 BT S = k1 k2 is acquired by choosing the appropriate weighting
matrices Q, R and S is a symmetric, positive definite matrix which is the solution of the
algebraic Riccati equation [11, 12]:

AT S + SA + Q − SBR−1 BT S = 0 (14)

Choosing sliding surface [13–15]:

t
s = CX2 − C (A2 − B2 K)X2 dt (15)
0
 
where C = c1 c2 are positive constant matrices
Differentiating (15):
 
ṡ = C A2 X2 + B2 uq2 − C(A2 − B2 K)X2 (16)
 Anti-torque Controller Using Sliding Mode Linear Quadratic Regulator 223

For x3 to converge towards 0, choosing conventional sliding mode control law as follows:

ṡ = −k.sign(s) (17)

Substitute (17) inside (16):

 
−k.sign(s) = C A2 X2 + B2 uq2 − C(A2 − B2 K)X2 (18)

The control signal of the motor is expressed as follows:

uq2 = (CB2 )−1 (−k.sign(s) − CB2 KX2 ) (19)

Proof. Consider a Lyapunov function:

V = 0.5s2 (20)

Differentiating V with respect to time yields:

   
V̇ = sṡ = s C A2 X2 + B2 uq2 − C(A2 − B2 K)X2 (21)

Substitute (17) inside (21):

 
V̇ = s C A2 X2 + B2 (CB2 )−1 (−k.sign(s) − CB2 KX2 ) − C(A2 − B2 K)X2
= −k.|s| (22)

where k is a positive constant, V̇ < 0∀s, so the close loop control system is stable.
Differentiating V̇ with respect to time yields:

V̈ = k 2 |s|2 (23)

Each term in Eq. (23) is bounded. Hence, V̈ is bounded and V̇ is uniformly continu-
ous, which means the error dynamics can reach s = 0 infinite time and angular velocity
error converges to zero x3 = 0. This completes the proof.
In order to reduce the chattering phenomenon, the following function can be used to
replace the sign function.
s
sign(s) = , ε is positive constant (24)
|s| + ε

4 Simulation

A simulation is performed to verify the stability and responsiveness of the controller.

The parameters of the system, as well as the controller, are presented in Table 1. The
reference speed of the secondary rotor tracks the speed of the primary rotor, which has
the reference speed as (25), and the load torque of the secondary rotor change from 0 to
0.01 (Nm) at t = 4 (s).
224 L. C. Q. Thai et al.

Table 1. Parameters of coaxial BLDC motor and controllers.

 
Coaxial BLDC motor J1 = 0.00252 kg.m2 ; J2 = 0.00241 kg.m2 ; R1 = R2 = 9.6();
L1 = L2 = 0.0043(H ); B1 = B2 = 0.1; kt1 = kt2 = 0.00496
Controller PI kP = 0.005; ki = 0.2
 
1 0  
LQR-SMC Q = ; R = 1; C = 0 −1/c2 ; ε = 10; k = 50
0 10−2


1200 t < 2
ω1ref = (25)
1200 + 300sin(2πt) t ≥ 2

In Fig. 4, at the transient stage, the maximum error speed tracking is nearly 120
(rpm), about 10% of the reference value at t = 0.2 (s) when the speed of two rotors
reaches a peak value. The main reason is that not only does the secondary rotor always
respond one step later than the primary rotor, but also speed change is large during initial
startup acceleration.
At the time t = 4, increasing the load torque and the desired speed has a sinusoidal
shape. The result of tracking the speed on the rising side responds lower in comparison
to the falling side.
With the same coefficient controller, the energy consumption levels are expressed
in Table 2. Compared with conventional SMC controllers, LQR-SMC has lower
consumption, but this difference is quite small, about 1.6%.

5 Conclusion

A sliding mode controller with an additional integrated sliding surface for optimization
by the LQR method is used to control the speed tracking with the moment of inertia of
the two rotors being almost equal, so the anti-torque problem has been resolved. And
energy consumption levels with the variation of the R matrix are also analyzed and give
the opposite result compared to the conventional LQR controller as well as conventional
sliding mode control. Thus, anti-torque control with energy consumption optimization
of coaxial BLDC rotor applied for autonomous underwater vehicles has been unraveled
in this paper.
 Anti-torque Controller Using Sliding Mode Linear Quadratic Regulator 225

Fig. 4. Speed response and error of two rotor

Table 2. Energy consumption of controllers

Controllers Energy consumption

Conventional sliding mode 115.4
LQR sliding mode 113.6

Acknowledgments. This research is funded by Vietnam National University Ho Chi Minh City
(VNU-HCM) under grant number TX2022-20b-01. We acknowledge the support of time and
facilities from National Key Laboratory of Digital Control and System Engineering (DCSELab),
Ho Chi Minh City University of Technology (HCMUT), VNU-HCM for this study.
226 L. C. Q. Thai et al.

References
1. Yuan, Z., Wang, H., Yang, J., Gu, G.: Research on torque balance of permanent-magnet
brushless DC motor. Proc. Int. Conf. Electr. Mach. Syst. ICEMS 2007 1, 786–788 (2007).
https://doi.org/10.1109/ICEMS.2007.4412190
2. Huang, Y., Wang, H., Li, W.: Research of torque balance control of dual-rotor motor based on
MATLAB. 2014 17th Int. Conf. Electr. Mach. Syst. ICEMS 2014 1219–1221 (2014). https://
doi.org/10.1109/ICEMS.2014.7013657
3. Huang, Y., Wang, H.: Torque balance of double motors drive system based on adaptive
VSSMC. ICEMS 2018–2018 21st Int. Conf. Electr. Mach. Syst. 1464–1469 (2018). https://
doi.org/10.23919/ICEMS.2018.8549337
4. Huang, Y., Wang, H.: Torque balance of dual-motor driven system based on equivalent switch-
ing variable structure sliding mode control. 23rd Int. Conf. Electr. Mach. Syst. ICEMS 2020
2059–2062 (2020). https://doi.org/10.23919/ICEMS50442.2020.9291195
5. Hanselman, D.C.: Brushless Motor Design (1994)
6. Hanselman, D.: Brushless permanent magnet motor design (2003)
7. Optimization, A.: Modeling and simulation methods for brushless DC motor drives. Abolfazl
Halvaei Niasar of Science & Technology Department of Electrical Isfahan University of
Technology Department of Electrical & Computer Iran University of Science & Technology
Department, pp. 1–6 (2005)
8. Tashakori, A., Ektesabi, M., Hosseinzadeh, N.: Modeling of BLDC motor with ideal back-
EMF for automotive applications. Proc. World Congr. Eng. 2011, WCE 2011. 2, 1504–1508
(2011)
9. Tibor, B., Fedak, V., Ďurovský, F.: Modeling and simulation of the BLDC motor in MATLAB
GUI. Proc.—ISIE 2011 2011 IEEE Int. Symp. Ind. Electron. 1403–1407 (2011). https://doi.
org/10.1109/ISIE.2011.5984365
10. Control, O., Edition, T.: Optimal Control, 3rd edn by Lewis, F.L., Vrabie, D.L., Syrmos, V.L.
(auth.) (z-lib.org).pdf. (2012)
11. Rafee Nekoo, S.: Nonlinear closed loop optimal control: a modified state-dependent Riccati
equation. ISA Trans. 52, 285–290 (2013). https://doi.org/10.1016/j.isatra.2012.10.005
12. Do, T.D., Choi, H.H., Jung, J.W.: SDRE-based near optimal control system design for PM
synchronous motor. IEEE Trans. Ind. Electron. 59, 4063–4074 (2012). https://doi.org/10.
1109/TIE.2011.2174540
13. Yu, G.R., Tseng, M.H., Lin, Y.K.: Optimal positioning control of a DC servo motor using
sliding mode. Proc. IEEE Int. Conf. Control Appl. 1, 272–277 (2004). https://doi.org/10.1109/
cca.2004.1387223
14. Shyu, K.K., Lai, C.K., Tsai, Y.W., Yang, D.I.: A newly robust controller design for the position
control of permanent-magnet synchronous motor. IEEE Trans. Ind. Electron. 49, 558–565
(2002). https://doi.org/10.1109/TIE.2002.1005380
15. Shyu, K.K., Shieh, H.J.: A new switching surface sliding-mode speed control for induction
motor drive systems. IEEE Trans. Power Electron. 11, 660–667 (1996). https://doi.org/10.
1109/63.506132
 Application of EDAS Method for Best Dressing
Mode for Internal Cylindrical Grinding

Trieu Quy Huy1 , Nguyen Hong Linh2 , Tra n Huu Danh3 , Bui Thanh Danh4 ,
Do Thi Tam5 , Luu Anh Tung5 , and Le Xuan Hung5(B)
1 University of Economics—Technology for Industries, Hanoi, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 Vinh Long University of Technology Education, Vinh Long, Vietnam
4 University of Transport and Communications, Hanoi, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

lexuanhung@tnut.edu.vn

Abstract. The results of a study on determining the best input process parame-
ters for dressing process in internal grinding SKD11 tool steel to achieve minimal
surface roughness (RS) and maximum wheel life (T) all at the same time are pre-
sented in this paper. To accomplish this, the EDAS (Evaluation based on Distance
from Average Solution) method was applied to solve the multi-criteria decision-
making (MCDM) problem. Besides, the Entropy method was used to calculate
the weights of the criteria. In addition, six input parameters were tested: coarse
dressing depth, coarse dressing passes, fine dressing depth, fine dressing passes,
non-feeding dressing, and dressing feed rate. The Taguchi method and an L16
orthogonal array were used to design this experiment. The best dressing mode for
internal cylindrical grinding has been proposed based on the findings of the study.

Keywords: Internal cylindrical grinding · MCDM · EDAS method · Entropy

method · Surface roughness · Wheel life · SKD11

1 Introduction

MCDM is a common technique for analyzing multiple options and selecting the best
one. It is widely used in various fields such as economics [1, 2], medicine [3, 4], military
[5, 6], engineering [7–12], etc. Many scientists are interested in applying the MCDM
technique in mechanical processing. This is because the machining process is frequently
required to meet a variety of criteria, such as maximum material removal rate (MMR),
minimal surface roughness (SR), and maximum tool wear rate.
So far, there have been many studies on the use of MCDM in machining processes.
Turning, milling, electric discharge machining, and other mechanical machining pro-
cesses have all been studied. Ch. Divya et al. [13] conducted a detailed literature review
of the application of various MCDM methods used for optimizing turning process param-
eters in the area. Prameet Vats et al. [14] presented a study that used hybrid AHP and
WASPAS methods to determine the best process parameters for turning AISI 1040

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 227–233, 2023.
https://doi.org/10.1007/978-3-031-22200-9_24
228 T. Q. Huy et al.

steel. Halil alşkan et al. [15] presented the findings of a study on material selection for
tool holders working under hard milling conditions using various multi criteria deci-
sion making methods. The TOPSIS (Technique for Order Performance by Similarity to
Ideal Solution), VIKOR (Viekriterijumsko Kompromisno Rangiranje), PROMETHEE
II (EXPROM2) (Preference Ranking Organization Method for Enrichment Evaluation),
and AHP (Analytic Hierarchy Process) were used in this study. V. C. Nguyen et al.
[16] presented an MCDM study using the TOPSIS method to determine the optimum
input factors in finishing the milling of Ti-6Al-4V titanium alloy under MQL conditions.
Recently, the results of the MCDM problem to find the best setup of dressing process
for internal grinding are presented in [17]. [18] presented the findings of an MCDM
study performed on cylindrically shaped parts in 90CrSi tool steel using powder-mixed
electrical discharge machining (PMEDM).
This paper presents an investigation into the use of MCDM in the dressing process for
internal cylindrical grinding. Two criteria were chosen for the investigation in this work:
Surface roughness RS and wheel life Tw . Furthermore, the EDAS method was chosen
to solve the MCDM problem. In addition, the Entropy method was used to calculate the
weights of the criteria. As a result, the EDAS method for determining the best dressing
mode for internal cylindrical grinding has been successfully applied.

2 Methodology
2.1 Method for Multi-criteria Decision Making
In this work, the EDAS method was used to solve the MCDM problem. The following
steps must be taken in order to use this method [19]:
Step 1: Making the initial decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X=⎢ r · · · r · · · r ⎥
in ⎥ (1)
⎢ 1 ij
⎢ . . . . . ⎥
⎣ .. . . .. . . .. ⎦
rm1 · · · rmj · · · rmn m×n

In which, m is the option number; n is the criteria number.

Step 2: Finding the average solution of each criterion by the following Equation:
m
rij
AV j = i=1 (2)
m
Step 3: Determining the positive and negative distances from average solution by:
For the criterion RS:

max 0, rij − AV j
PDAij = (3)
AV j

max 0, AV j − rij
NDAij = (4)
AV j
 Application of EDAS Method for Best Dressing 229

For criterion Tw :

max 0, AV j − rij
PDAij = (5)
AV j

max 0, rij − AV j
NDAij = (6)
AV j
Step 4: Calculating the weighted PDA and NDA by:

n
SP i = PDAij · wj (7)
j=1


n
SN i = NDAij · wj (8)
j=1

Step 5: Determining weighted normalized PDA and NDA by the following Equation:
SP i
NSP i = (9)
max(SP i )
i
SN i
NSN i = (10)
max(SN i )
i

Step 6: Calculating appraisal score for each option by:

AS i = (NSP i + NSN i )/2 (11)

Step 7: Rank the option using the rule that the option with the highest ASi is the best.

2.2 Method for Detemining Criterion Weight

In this work, the weights of the criteria were calculated using the Entropy method. This
method can be implemented using the steps outlined below [20]:
Step 1: Calculating the indicator’s normalized values by the following Equation:
xij
pij = m (12)
m+ 2
i=1 xij

Step 2: Determining the Entropy for each indicator by:

m 
 m   m 
mej = − pij × ln pij − 1 − pij × ln 1 − pij (13)
i=1 i=1 i=1

Step 3: Calculating the weight of each indicator by the following Equation:

1 − mej
w j = m
(14)
j=1 1 − mej
230 T. Q. Huy et al.

3 Experimental Work
To solve the multi-objective optimization problem, an experiment was carried out.
Besides, the Taguchi method and the L16 orthogonal array were used to design this
experiment (44 ×22 ). Table 1 shows the input factors and their levels. Figure 1 depicts
the experimental setup, and Tables 2 show the input dressing factors and their levels. The
experiment was conducted in the following order: The dressing procedure was conducted
with the experimental matrix as shown in Table 2. The grinding wheel was used to grind
test samples at the following grinding regime after dressing process: wheel speed of
12000 (rpm); workpiece speed of 150 (rpm); radial wheel speed of 0.0025 (mm/stroke);
and axial feed speed of 1 (mm/min.). After the experiments were completed, the SR (in
this case, Ra (m)) was measured and WL (min.) was calculated. The output results are
shown in the two right columns of Table 2.

Fig. 1. Setup of experiment.

4 Determining the Best Input Dressing Parameters

This section explains how to find the best input dressing parameters for internal grind-
ing SKD11 tool steel by solving the MCDM problem with the EDAS method and
determining criteria weights with the Entropy method.

4.1 Determining the Criation Weights

The weights of criteria are calculated as follows using the Entropy method (see Sect. 2.2):
The normalized values pij are calculated using Eq. (12); The Entropy value for each
indicator mej was calculated using Eq. (13). At last, using Eq. (14), determine the
weight of the criteria wj . Ra and Tw weights were calculated to be 0.5129 and 0.4871,
respectively.
 Application of EDAS Method for Best Dressing 231

Table 1. Input dressing factors.

No Input Factors Levels

1 2 3 4
1 Coarse dressing depth (mm) 0.025 0.03 - -
2 Coarse dressing time (times) 1 2 3 4
3 Fine dressing depth (mm) 0.005 0.01 0.015 0.02
4 Fine dressing time (times) 0 1 2 3
5 Non-feeding dressing (times) 0 1 2 3
6 Dressing feed rate (mm/min) 1 1.2 - -

Table 2. Experimental matrix and output results.

No ar nr af nf n0 Sd Ra (µm) Tw (min.)
1 0.025 1 0.005 0 0 1 0.3652 0.9186
2 0.03 1 0.01 1 1 1 0.214 12.733
3 0.025 1 0.015 2 2 1.2 0.195 12.697
4 0.03 1 0.02 3 3 1.2 0.242 11.423
5 0.03 2 0.005 1 2 1.2 0.185 11.803
6 0.025 2 0.01 0 3 1.2 0.248 12.430
7 0.03 2 0.015 3 0 1 0.252 11.147
8 0.025 2 0.02 2 1 1 0.217 12.077
9 0.03 3 0.005 2 3 1 0.306 12.547
10 0.025 3 0.01 3 2 1 0.324 12.077
11 0.03 3 0.015 0 1 1.2 0.341 12.027
12 0.025 3 0.02 1 0 1.2 0.354 12.503
13 0.025 4 0.005 3 1 1.2 0.318 12.177
14 0.03 4 0.01 2 0 1.2 0.313 10.620
15 0.025 4 0.015 1 3 1 0.326 13.153
16 0.03 4 0.02 0 2 1 0.363 12.230

4.2 Determining the Best Input Dressing Parameters

The EDAS method is applied to the MCDM problem as follows (see Sect. 2.1): Using
Eq. (2), compute the mean solution for each criterion. Following that, using Eqs. (3)
and (4) for surface roughness and (5) and (6) for wheel life, the positive and negative
distances from the average solution are calculated. The weighted PDA and NDA are
computed using Eqs. (7) and (8). Equations (9) and (10) are then used to calculate the
weighted normalized PDA and NDA. The appraisal score ASi for each option is then
232 T. Q. Huy et al.

calculated using the formula (11). Finally, arrange the options so that the option with
the highest ASi is the best. Table 3 displays a variety of calculated outcomes as well as
option rankings.

Table 3. Several calculated results and ranking of options.

Trial SPi SNi NSPi NSNi ASi Rank

1 0.0000 0.5920 0.0000 0.0000 0.0000 16
2 0.1849 0.0000 0.8508 1.0000 0.9254 3
3 0.2174 0.0000 1.0000 1.0000 1.0000 1
4 0.0786 0.0000 0.3618 1.0000 0.6809 6
5 0.1969 0.0000 0.9056 1.0000 0.9528 2
6 0.1108 0.0000 0.5098 1.0000 0.7549 5
7 0.0595 0.0112 0.2739 0.9810 0.6275 7
8 0.1515 0.0000 0.6970 1.0000 0.8485 4
9 0.0485 0.0383 0.2232 0.9353 0.5792 9
10 0.0285 0.0698 0.1309 0.8821 0.5065 11
11 0.0263 0.0998 0.1211 0.8313 0.4762 13
12 0.0467 0.1241 0.2147 0.7903 0.5025 12
13 0.0327 0.0590 0.1506 0.9003 0.5254 10
14 0.0000 0.0832 0.0000 0.8595 0.4297 15
15 0.0744 0.0734 0.3424 0.8760 0.6092 8
16 0.0350 0.1409 0.1610 0.7621 0.4615 14

From Table 3, it can be found that option 3 is the best option. This is because it
has the highest ASi value (ASi = 1.0000). As a consequence, the best dressing mode
contains the following input factors: ar = 0.025 (mm); nr = 1 (times); af = 0.015 (mm);
nf = 2 (times); n0 = 2 (times); Sd = 1.2 (m/min).

5 Conclusions
This paper presents the results of applying the EDAS method to solve the MCDM
problem and the Entropy method to calculate the criation weights in internal grinding
SKD11 tool steel. To fulfill this, the Taguchi method was used to design and analysis an
experiment with an L16 orthogonal array (44 ×22 ). The findings of the study lead to the
following conclusions:
- The EDAS method is being used for the first time to solve the MCDM problem in
internal grinding.
- The best input dressing parameters to achieve the minimum surface roughness and
the maximum wheel life are as follows: ar = 0.025 (mm); nr = 1 (times); af = 0.015
(mm); nf = 2 (times); n0 = 2 (times); Sd = 1.2 (m/min).
 Application of EDAS Method for Best Dressing 233

Acknowledgment. Thai Nguyen University of Technology provided support for this research.

References
1. Zavadskas, E.K., Turskis, Z.: Multiple criteria decision making (MCDM) methods in
economics: an overview. Technol. Econ. Dev. Econ. 17(2), 397–427 (2011)
2. Balezentis, T., et al.: Optimizing crop mix with respect to economic and environmental
constraints: An integrated MCDM approach. Sci. Total Environ. 705, 135896 (2020)
3. Ejegwa, P.A.: Generalized triparametric correlation coefficient for Pythagorean fuzzy sets
with application to MCDM problems. Granul. Comput. 6(3), 557–566 (2020). https://doi.
org/10.1007/s41066-020-00215-5
4. Wen, Z., et al.: Cold chain logistics management of medicine with an integrated multi-criteria
decision-making method. Int. J. Environ. Res. Public Health 16(23), 4843 (2019)
5. YILMAZ, B.Ö., H. TOZAN, and M.A. KARADAYI, Multi-Criteria decision making
(MCDM) applications in military healthcare field. J. Health Syst. Policies, 2020. 2(2):
p. 149–181
6. Sánchez-Lozano, J., Rodríguez, O.N.: Application of Fuzzy Reference Ideal Method (FRIM)
to the military advanced training aircraft selection. Appl. Soft Comput. 88, 106061 (2020)
7. Kumar, A., et al.: A review of multi criteria decision making (MCDM) towards sustainable
renewable energy development. Renew. Sustain. Energy Rev. 69, 596–609 (2017)
8. Hung, K.-C., et al.: A decision support system for engineering design based on an enhanced
fuzzy MCDM approach. Expert Syst. Appl. 37(1), 202–213 (2010)
9. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev., 9 (2022)
10. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: A comparative study. Adv. Prod. Eng. &
Manag. 16(4), 443–456 (2021)
11. Do, D.T., A combination method for multi-criteria decision making problem in turning
process. Manuf. Rev., 8 (2021)
12. Trung, D.D.: Application of TOPSIS and PIV methods for multi-criteria decision making in
hard turning process. J. Mach. Eng., 21 2021
13. Divya, C., Raju, L.S., Singaravel, B.: Application of MCDM methods for process parameter
optimization in turning process—a review. Recent trends in mechanical engineering, p. 199–
207 (2021)
14. Vats, P., et al.: Optimization of machining parameters in turning of AISI 1040 steel using
hybrid MCDM technique. Materials Today: Proceedings 50, 1758–1765 (2022)
15. Çalışkan, H., et al.: Material selection for the tool holder working under hard milling con-
ditions using different multi criteria decision making methods. Mater. Des. 45, 473–479
(2013)
16. Nguyen, V., Nguyen, T., Tien, D.: Cutting parameter optimization in finishing milling of
Ti-6Al-4V titanium alloy under MQL condition using TOPSIS and ANOVA analysis. Eng.,
Technol. & Appl. Sci. Res. 11(1), 6775–6780 (2021)
17. Nguyen, H.-Q., et al.: A Comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
18. Nguyen, H.-Q., et al.: Multi-Criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
19. Alinezhad, A., New methods and applications in multiple attribute decision making (MADM).
Springer (2019)
20. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Vietnam. J. Agric. Sci. 17(4), 32–2331 (2019)
 Application of MARCOS Method for Selecting
the Best Schema of Scissors Mechanism

Trieu Quy Huy1 , Nguyen Ba Liem2 , Tran Quang Hau2 , Dang Quoc Cuong3 ,
Tran Huu Danh4 , Nguyen Thi Thanh Nga2 , Vu Ngoc Pi2 , and Nguyen Ngoc Thieu2(B)
1 University of Economics - Technology for Industries, Hanoi, Vietnam
2 Thai Nguyen University of Technology, Thai Nguyen, Vietnam
ngthieu@tnut.edu.vn
3 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
4 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam

Abstract. A lift table is a very common piece of equipment in industry and pro-
duction. This device uses a scissors mechanism to lift goods or people. There are
numerous schemas (or diagrams) for the scissors mechanism. With the same set of
input parameters, each schema will produce a different set of output responses. As
a result, to achieve optimum design of a lift table, it is critical to determine the best
scheme. This paper presents the findings of a study that used multi-criteria decision
making (MCDM) technique to determine the best scheme for designing a lift table
to meet two criteria at the same time: minimum lifting time and maximum lifting
force. To accomplish this, nine different diagrams of the scissor lifting mecha-
nism were analyzed using the Measurement Alternatives and Ranking according
to COmpromise Solution (MARCOS) method for solving the MCDM problem
and the Method based on the Removal Effects of Criteria (MEREC) for the calcu-
lation of the weights of criteria. Finally, the best lift table design scheme has been
proposed.

Keywords: Lift table · Optimal design · MCDM · MARCOS · MEREC

1 Introduction

A lift table (Fig. 1) is a device that raises and lowers goods and/or people using a scissors
mechanism [1]. This type of device is typically used to move large, heavy loads over
short distances or to adjust their height to meet the needs of an application and the user.
In practice, lift tables are commonly used in logistics warehouses, manufacturing plants,
workshops, warehouses, factories, garages, and so on. As a result, scientists have become
interested in designing and optimizing scissors mechanism or lift tables.
There have been numerous studies on scissor lifts so far. G. Arunkumar et al. [2] con-
ducted research on hydraulic scissor lift design, analysis, and topological optimization.
Topology Optimization and FEA analysis techniques were used in this study, and as a
result, the weight of the lift table was reduced by 38%. Tian Hongyu and Zhang Ziyi [3]
conducted research on the design of a scissors lift platform using the 3D software Pro/E.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 234–243, 2023.
https://doi.org/10.1007/978-3-031-22200-9_25
 Application of MARCOS Method for Selecting the Best Schema 235

The platform in their study is designed to fold away doors to save space for convenient
storage. Furthermore, it can be raised to a height of 8 m. M. T. Islam et al. [4] used
a bond graph modeling framework to implement general multibody system dynamics
on a scissor lift mechanism. The mechanism is modeled and simulated in this work in
order to evaluate several application-specific requirements such as dynamics, position
accuracy, and so on. Tao Liu and Jian Sun [5] demonstrated a three-dimensional model
of a scissor lifting mechanism. The kinematical and kinetic simulation analysis in this
work was performed using MATLAB/Simulink. In addition, the mechanism’s design
was optimized in Pro/Mechanical based on the results of the simulation analysis. Ren
G. Dong et al. [6] conducted a study to better understand the major mechanisms and
factors influencing scissor lift stability. In this study, both experimental and modeling
approaches were used. A lumped-parameter model of the scissor lift was developed based
on the dynamic characteristics identified in the experimental results. Lukasz Stawinski
et al. [7] introduced a new drive system for scissor lift. The new system uses a frequency
inverter and a simple controller for a hydrostatic system with the ability to control the
velocity of a hydrostatic receiver such as a hydrostatic cyreceive. Wei Zhang et al. [8]
investigated the static stability of six types of scissor lifts with one hydraulic actuator
input force, with the input on the lines of the scissor lifts’ nodes.

Fig. 1. Lift table [9].

According to the above analysis, while there have been many studies on the scissor
lifting mechanism, there has been no research on selecting the most reasonable or best
schema for this type of mechanism. In this study, the problem of determining the best
lifting table scheme for designing was solved with the help of MCDM. In practice, the
application of MCDM technique to select the best option has been carried out in many
mechanical processing processes. For example, solved in milling [10], turning [10–12],
grinding [10, 13], electrical discharge machining (EDM) [14], and powder-mixed EDM
[15]. In this work, to address this issue, nine different lift table schemas were identified.
236 T. Q. Huy et al.

Besides, two criteria for selecting the mechanism diagram are chosen, including the
minimum lifting time and maximum lifting force. Furthermore, the MARCOS method
was used to solve MCDM problem and MEREC method was used to calculate of the
weights of criteria. Finally, the best schema for scissor lifting mechanism for optimum
design of lift tables was proposed.

2 Methodology

2.1 Method for Multi-criteria Decision Making

In this study, the MARCOS method is used to solve the MCDM problem. This method
is performed in the following order [10]:
Step 1: Using the following equation to generate the initial matrix:
⎡ ⎤
x11 · · · x1n
⎢ x21 · · · x2n ⎥
⎢ ⎥
X =⎢ . . ⎥ (1)
⎣ .. · · · .. ⎦
xmn · · · xmn

Step 2: Adding the ideal (AI) and anti-ideal solution (AAI) to the initial decision-
making matrix to create an extended initial matrix.
⎡ ⎤
AAI xaa1 · · · xaan
A1 ⎢⎢ x11 · · · x1n ⎥
⎥
⎢ ⎥
A2 ⎢ x21 · · · x2n ⎥
X = . ⎢ . . . ⎥
⎢ (2)
.. ⎢ .. .. .. ⎥ ⎥
⎢ ⎥
Am ⎣ xm1 · · · xmn ⎦
AI xai1 · · · xain

wherein, AAI = min(xij ) and AI = max(xij ) for criterion Wm ; AAI = max(xij ) and
AI = min(xij ) for criterion T.
Step 3: Normalizing
 the extended initial matrix (X). The elements of normalized
matrix N = nij m×n are found by

uij = xAI /xij (3)

uij = xij /xAI (4)

Equation (3) is used for criterion T and Eq. (4) is used criterion Wm .
 for
Step 4: Creating a weighted normalized matrix C = cij m×n by:

cij = uij · wj (5)

where, wj is the weight coefficient of the criterion j.

 Application of MARCOS Method for Selecting the Best Schema 237

Step 5: Determining the utility degree of alternatives Ki− and Ki+ by:

Ki− = Si /SAAI (6)

Ki+ = Si /SAI (7)

With

m
Si = cij (8)
i=1

Step 6: Calculating the utility function of alternatives f(Ki) by:

Ki+ + Ki−
f (Ki ) = (9)
1−f (Ki+ ) 1−f (Ki− )
1+ f (Ki+ )
+ f (Ki− )

In which, f(K i − ) and f(K i + ) are the utility functions associated with the anti-ideal
and ideal solutions. These functions are established by:
 
f Ki− = Ki+ /(Ki+ + Kii ) (10)

 
Ki+ = Ki− /(Ki+ + Kii ) (11)

Step 7: Calculating the alternative with the highest value of the utility function by
ranking the alternatives based on the final value of the utility functions f(Ki).

2.2 Method for Calculation of the Weights of Criteria

MEREC method is applied to find the weights of the criteria in this study. The MEREC
method can be used by the following steps [11]:
Step 1: Establishing the initial matrix by Eq. (1).
Step 2: Calculating the normalized values by:
minxij
hij = (12)
xij
xij
hij = (13)
maxxij

In which, Eq. (11) is used for criterion tl and Eq. (12) is used for criterion Wm .
Step 3: Finding the alternative performance Si by:
⎡ ⎛ ⎞⎤
1
  
Si = ln⎣1 + ⎝ ln hij ⎠⎦ (14)
n
j
238 T. Q. Huy et al.


Step 4: Calculating the performance of ith alternative Sij by:
⎡ ⎛ ⎞⎤
 1
  
Sij = Ln⎣1 + ⎝ ln hij ⎠⎦ (15)
n
k,k=j

Step 5: Finding the removal effect of the jth criterion Ej by:

  

Ej = Sij − Si  (16)
i

Step 6: Calculating the criterion weight by:

Ej
wj =  (17)
k Ek

2.3 Method for Determine Lift Table Schema

When designing a lift table, the problem is that with the same input parameters it is
necessary to find a diagram of the lift table so that the designed lift table has the minimum
lifting time and the maximum lifting force of the table. The input parameters, in this
case, include: The main dimensions of the lift table (AC and BD); the maximum lift
height hm, the piston force Fp, and the piston speed Vp. In fact, there are many different
options for lift table schema. Figure 2 shows nine different schemas for the design of a
lift table. Each schema will give different output responses.
To determine the best option for the lift table, the calculation process is presented
as follows: Calculate the output parameters for nine options of the lift table. After that,
apply the MARCOS method to determine the best solution - the one that simultaneously
satisfies the two criteria including the minimum lifting time and the maximum lifting
force.

3 Calculation of the Output Responses of the Lift Table

Figure 3.a shows the lift table solution 2 (as in Fig. 2). The input data are the force on
the piston Fp = 2000 (N), the velocity of the piston Vp = 10 (mm/s), the maximum
lifting height hm = 866 (mm), and the length of link AC and BD is L = 1000 (mm).
The output responses are the maximum lifting force Wm (N), and the time of lift tl (s).
From Fig. 3.a and Fig. 3.b the force balance equation is given as follows:

FA = FB = FC = FD = W /2 (18)

From Fig. 3.d it is found that FGx = 0; FGy = 0. Besides, from Fig. 3.c: = MA 0.
Therefore, the following Equation can be found:

Ly Lx
MA = −FC · Lx + FGx · cosβ · + FGx · sinβ · =0 (19)
2 2
 Application of MARCOS Method for Selecting the Best Schema 239

With Lx = L · cosα, Ly = L · sinα and FC = Lx · W /2.

It was found from (18) that:

W = FP · sin(α + β)/cosα (20)

Setting LAE = a, LAF = b, LEF = c.

From AFE:

c2 = a2 + b2 − 2 · a · b · cosα (21)

And:

Fig. 2. Different schemas for lift table design.


c= a2 + b2 − 2 · a · b · cosα (22)

We get:

H = cαmax − cαmin (23)

From AFE have:

b2 = a2 + c2 − 2 · a · c · cosβ (24)
240 T. Q. Huy et al.

Fig. 3. Schema for calculation of output responses of a lift table.

From (22):

a2 + c2 − b2
β = arccos( ) (25)
2·a·c
The stroke of piston, denoted as H, is computed by:

H = cmax − cmin (26)

where cmax and cmin are the maximum and minimum positions of a piston, respectively.
They are determined from Eq. (20).
The lifting time tl can be calculated by:

tl = H/VP (27)

With the remaining schema in Fig. 2, the calculation of the output responses is
similar to the above. Calculated for ten schemas using the above-mentioned input data,
the results are shown in Table 1.

4 Determining the Best Schema for Scissors Mechanism

This section presents the results when the MACOS method is used to solve MCDM
problem, and the MEREC method is used for calculation of the weights of the criteria.
 Application of MARCOS Method for Selecting the Best Schema 241

Table 1. Output results of different solutions of a lift table.

Solution tl (s) Wm (N)

1 36.27 348.62
2 45.55 2000.00
3 27.26 320.62
4 64.13 1015.47
5 9.39 69.98
6 11.28 186.82
7 10.90 167.88
8 49.62 174.98
9 10.70 770.07

4.1 Determining the Criteria Weights

To calculate the weights for the criteria using the MEREC method, perform the follow-
ing steps (see Sect. 2.2): Calculate the normalized values hj using Eqs. (11) and (12);

determine Si and Sij using Eqs. (13) and (14); calculate the criterion removal effect using
Eq. (15); and find the weight of the criteria wj using Eq. (16). Furthermore, tl and Wm
were discovered to have weights of 0.6142 and 0.3858, respectively.

4.2 Determining the Best Schema of Scissors Mechanism by Using MARCOS

Method
The application of the MARCOS method to multi-objective decision-making is carried
out according to the steps outlined in Sect. 2.1. First, the ideal solution (AI) and the
anti-ideal solution (AAI) are determined by formula (2). The obtained values of Ra and
MRR are 9.3913 (s) and 2000 (N) with AI, and 64.1269 (s) and 69.9802 (N) with AAI,
respectively. Then, the normalized values uij are determined according to formulas (3)
and (4). Also, the normalized value taking the weight cij into account is found by formula
(5). In
 addition, coefficients Ki− and Ki+ are determined by Eqs. (6) and (7). The values
−  +  
of f Ki and f Ki are determined by Eqs. (9) and (10). It was found that f Ki−
 +
= 0.0626 and f Ki = 0.9374. Finally, the values of f (Ki ) are calculated by formula
(8). Table 2 shows the calculated results of several parameters and the ranking of the
solutions.
Table 2 shows several calculated results and the ranking of solutions of scissors
mechanism when using MARCOS method with the calculation of the weight of criteria
by the MEREC method. Besides, Fig. 4 describes the relation between the solutions and
the values of f(Ki). From Table 2 and Fig. 4 it was reported that the best schema for
scissors mechanism is solution 9 because it has the maximum value of f(Ki) (0.00034).
From Table 1, it is easy to see that this solution has a lift time of 10.7 (s) and a maximum
lift force of 770.07 (N).
242 T. Q. Huy et al.

Table 2. Several calculated results and ranking of solutions.

Solution K− K+ f(K−) f(K+) f(Ki) Rank

1 0.0017 0.0001 0.0626 0.9374 0.00011 8
2 0.0038 0.0003 0.0626 0.9374 0.00025 5
3 0.0020 0.0001 0.0626 0.9374 0.00014 7
4 0.0021 0.0001 0.0626 0.9374 0.00014 6
5 0.0047 0.0003 0.0626 0.9374 0.00031 2
6 0.0041 0.0003 0.0626 0.9374 0.00027 4
7 0.0042 0.0003 0.0626 0.9374 0.00028 3
8 0.0011 0.0001 0.0626 0.9374 0.00007 9
9 0.0051 0.0003 0.0626 0.9374 0.00034 1

Fig. 4. Relationship between the solution and the value of f(Ki ).

5 Conclusions
This paper presents the results of a study to determine the best schema for scissors
mechanism for lift tables to obtain both the minimum lifting time and the maximum
lifting force simultaneously. In this study, the MARCOS method was used to solve the
MCDM problem and the MEREC method was also used to determine the weight of
criteria. The study findings lead to the following conclusions:

– This was the first time an MCDM method (the MARCOS method) was used to deter-
mine the best schema for scissors mechanism for lift tables to meet two competing
criteria at the same time: minimum lifting time and maximum lifting force.
– Solution 9 is the best schema for scissors mechanism because it has the highest value
of f(Ki) (0.00034);
 Application of MARCOS Method for Selecting the Best Schema 243

– The best schema for scissors mechanism (solution 9) can give a lifting time of 10.7
(s) and a lifting force of 770.07 (N).

Acknowledgments. This work was supported by Thai Nguyen University of Technology.

References
1. WikipediA: https://en.wikipedia.org/wiki/Lift_table. Accessed 20 May 2022
2. Arunkumar, G., Kartheeshwaran, R., Siva, J.: Investigation on design, analysis and topological
optimization of hydraulic scissor lift. J. Phys. Conf. Ser. IOP Publishing (2021)
3. Hongyu, T., Ziyi, Z.: Design and simulation based on Pro/E for a hydraulic lift platform in
scissors type. Proc. Eng. 16, 772–781 (2011)
4. Islam, M.T., et al.: Dynamic analysis of scissor lift mechanism through bond graph modeling.
In: 2014 IEEE/ASME International Conference on Advanced Intelligent Mechatronics. IEEE
(2014)
5. Liu, T., Sun, J.: Simulative calculation and optimal design of scissor lifting mechanism. In:
2009 Chinese Control and Decision Conference. IEEE (2009)
6. Dong, R.G., et al.: An investigation on the dynamic stability of scissor lift. Open J. Saf. Sci.
Technol. 2(01), 8 (2012)
7. Stawinski, L., Kosucki, A., Morawiec, A., Sikora, M.: A new approach for control the velocity
of the hydrostatic system for scissor lift with fixed displacement pump. Arch. Civil Mech.
Eng. 19(4), 1104–1115 (2019). https://doi.org/10.1016/j.acme.2019.06.001
8. Zhang, W., et al.: A study on the static stability of scissor lift. Open Mech. Eng. J. 9(1) (2015)
9. ULINE: Electric lift table. https://www.uline.com/Product/Detail/H-3935/Lift-Tables/Ele
ctric-Lift-Table-2000-lb-48-x-36. Accessed 20 May 2022
10. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev. 9 (2022)
11. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: a comparative study. Adv. Prod. Eng.
Manag. 16(4), 443–456 (2021)
12. Do, D.T.: A combination method for multi-criteria decision making problem in turning
process. Manuf. Rev. 8 (2021)
13. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
14. Huu Phan, N.. Muthuramalingam, T.: Multi criteria decision making of vibration assisted
EDM process parameters on machining silicon steel using Taguchi-DEAR methodology.
Silicon 13(6), 1879–1885 (2021)
15. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
16. Stević, Ž, et al.: Sustainable supplier selection in healthcare industries using a new MCDM
method: measurement of alternatives and ranking according to COmpromise solution
(MARCOS). Comput. Ind. Eng. 140, 106231 (2020)
17. Keshavarz-Ghorabaee, M.: Assessment of distribution center locations using a multi-expert
subjective–objective decision-making approach. Sci. Rep. 11(19461), 2021 (2021). https://
doi.org/10.1038/s41598-021-98698-y
 Application of MCDM Method in Selection
of Schema for Optimal Design of Double Scissor
Lift Tables

Nguyen Hong Linh1 , Trieu Quy Huy2 , Tran Minh Tan3 , Tran Huu Danh4 ,
Bui Thanh Danh5 , Nguyen Thi Thanh Nga6 , Vu Ngoc Pi6 ,
and Nguyen Ngoc Thieu6(B)
1 Electric Power University, Hanoi, Vietnam
2 University of Economics—Technology for Industries, Hanoi, Vietnam
3 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
4 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
5 University of Transport and Communications, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

ngthieu@tnut.edu.vn

Abstract. A scissor lift table is a valuable piece of industrial equipment. This

device is used to lift objects or people. The double lift tables were used to achieve
the extra reach for this equipment. There are numerous schemas for double lift
tables in practice. Each schema will produce different values of output results
when given the same set of input factors. As a result, determining the best scheme
is critical when designing an optimal double lift table. The purpose of this paper is
to present a study on multi-criteria decision making (MCDM) to determine the best
scheme for designing a double lift table that meets two criteria at simultaneously:
minimum lift time and maximum lift force. To gain this, nine different double
lift table schemas were analyzed using the Multi-objective Optimization Ratio
Analysis (MOORA) method for solving the MCDM problem and the Entropy
method for calculating criteria weights. Finally, the best design scheme for a
double lift table has been proposed.

Keywords: Double lift table · Optimal design · MCDM · MOORA · Entropy

method

1 Introduction

A scissor lift table is a device used to carry large, heavy objects over short distances or to
adjust their height to meet the needs of an application and the user. A double scissor lift
table (Fig. 1) can be understood as a mechanism that consists of two scissor lift tables
stacked on top of each other to increase the scissor lift table’s lifting capacity. Lift tables
and double scissor lift tables are commonly used in logistics warehouses, warehouses,
factories, garages, etc. As a result, many scientists are interested in the optimal design
of the scissor lift table as well as the double scissor lift table.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 244–253, 2023.
https://doi.org/10.1007/978-3-031-22200-9_26
 Application of MCDM Method in Selection of Schema 245

The design and analysis of a hydraulic scissor lift are presented in [1]. This work
is primarily concerned with the forces acting on the equipment as it is extended and
contracted. A study on the static stability of scissor lifts has been conducted [2]. This
paper investigates the static stability of six different types of scissor lifts with a single
hydraulic actuator input force applied to the scissor lifts’ node lines. Research on the
stress analysis of the scissor lift table using the finite element method was presented
in [3]. The study estimates and explains the methodology for structural analysis used
in the presented case study. The results of the FEM analysis serve as the foundation
for developing methods and guidelines for optimizing the current design in order to
obtain optimal parameters for the lifting table’s weight, stability, capacity, mobility, and
layout. The determination of appropriate dimensions in designing hydraulic scissor lifts
was proposed in [4]. The strength of the structural elements containing the frame was
investigated [5]. The device used in this work has a maximum load of 300 kg. Research
to achieve the parametric design of the various components and parametric assembly
of scissor lift tables was described [6]. In the study, the size of the model structure was
optimized using the finite element method. The actuator forces have been evaluated
under the rated loading of the scissor lift [7]. The mobile scissor lift in this work was
designed to be hydraulically powered to a height of 8 m carrying a maximum load of
1000 kg.

Fig. 1. A double scissor lift table [8]

From the above analysis, while many studies have been conducted on this field, no
research has been found on selecting the best schema for double lift tables. In this paper,
the MCDM method was used to solve the problem of determining the best double lifting
table scheme. In practice, using MCDM methods to determine the best solution for a
mechanical machining process is very popular. This issue has been addressed in milling
[9, 10], turning [9, 11, 12], grinding [9, 13], electrical discharge machining (EDM) [14],
and powder-mixed EDM [15]. In this work, to apply the MCDM technique, nine different
double lift table schemas were identified and two criteria including the minimum lifting
246 N. H. Linh et al.

time, and maximum lifting force are chosen for the investigation. Furthermore, the
MOORA method was used to solve the MCDM problem, and the Entropy method was
used to calculate the criterion weights. From the results of the work, the best double
scissor lift table schema was proposed.

2 Methodology
2.1 Method for Multi-criteria Decision Making
The MOORA method was used to solve the MCDM problem in this work. Using this
method, the following steps must be taken [16].
Step 1: Creating the initial decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X =⎢ ⎢ 1 r · · · r ij · · · r ⎥
in ⎥ (1)
⎢ . . . . . ⎥
⎣ .. . . .. . . .. ⎦
rm1 · · · rmj · · · rmn
In which m is the number of alternatives and n is the number of criteria.
Step 2: Finding the normalized matrix by the following Equation:
rij
nij = (2)
maxrij
minrij
nij = (3)
rij

Noted that Eq. (2) is used for lift time criation, and Eq. (3) is for
lift force criation.
Step 3: Making a normalized decision making matrix Xij m×n :
yij
Xij =
 (4)
m 2
i=1 yij

Step 4: Calculating the decision-making matrices after normalizing the weights by:

Wij = wj × yij (5)

Step 5: Determining Pi and Ri by:

1 
Pi = Wij (6)
|B| j∈B

1 
Ri = Wij (7)
|NB| j∈NB

In which B and NB are the criation number as large as better and criation number as
small as better, respectively.
 Application of MCDM Method in Selection of Schema 247

Step 6: Determining Qi by the following Equation:

Qi = Pi − Ri (8)

Step 7: Rank the options according to the principle that the option with the highest
Qi is the best.

2.2 Method for Criation Weight Determination

The Entropy method was used to calculate the weights of the criteria in this work. This
method can be implemented using the steps outlined below [17].
Step 1: Calculating the indicator’s normalized values by:
xij
pij = m (9)
m+ 2
i=1 xij

Step 2: Finding the Entropy for each indicator by:

m    m   m 
mej = − pij × ln pij − 1 − pij × ln 1 − pij (10)
i=1 i=1 i=1

Step 3: Determining the weight of each indicator by the following Equation:

1 − mej
wj = m . (11)
j=1 1 − mej

2.3 Method for Determing Double Scissor Lift Table Schema

The problem with designing a double lift table is that it is necessary to find a lift table
schema with the same input parameters so that the designed table has the shortest lift time
and the greatest lift force. In this case, the input parameters are the two main dimensions
of the double lift table (AC and BD); the maximum lift height hm, the piston force Fp ,
and the piston speed Vp . In fact, there are numerous double lift table schema options.
Figure 2 depicts nine different designs for a double lift table. Each schema will produce
different values of output factors (the lift time and the lift force).
The process for determining the best option for the double lift table is as follows:
Determine the output parameters for the double lift table’s nine options. After that, use
the MOORA method to find the best solution - that meets both the minimum lift time
and the maximum lift force criteria.

3 Calculation of the Output Results of Double Scissor Lift Tables

Figure 3a describes the double lift table solution 3 (as in Fig. 2.2). The input factors are
the force on the piston Fp = 2000 (N), the velocity of the piston Vp = 10 (mm/s), the
maximum lifting height hm = 1732 (mm), and the length of links AC, BD, DK, and
CI is L = 1000 (mm). The output factors are the maximum lift force Wm (N), and the
248 N. H. Linh et al.

time of lift tl (s). From Fig. 3a and b the Equation for the force balance can be given as
follows:
Lx = L · cosα (12)

Ly = L · sinα (13)

FPx = FP · cosβ (14)

FPy = Fp · sinβ (15)

W
FD = FI = FK = (16)
2
Setting LAC = LBD = LDK = LCI = L; LIE = a; LIF = b; LEF = c
It can be found from Fig. 3c that FGx = 0; FGy = 0.
From Fig. 3d have:
 − →
mC Fi = 0 (17)
And
FI .Lx − FP · sinβ(L − IF) · cosα − FP · cosβ(L − IF) · sinα = 0 (18)
Submitting FI = W
2 ;Lx = L · cosα; IF = 3·L
4 , we get:
FP · sin(α + β)
W= (19)
2 · cosα
From  IEF:
c2 = a2 + b2 − 2 · a · b · cos(90 − α) (20)
Or we have:

c= a2 + b2 − 2 · a · b · cos(90 − α) (21)
We get
 = cαmax − cαmin (22)
From  IEF have:
a2 = b2 + c2 − 2 · b · c · cos(β + α) (23)
 2 
b + c2 − a2
→ β = arccos −α (24)
2·b·c
The lift time tl can be determined by:

tl = (25)
VP
In which  is stroke of piston.
The calculation of the values of the output factors is similar in the remaining schemas
in Fig. 2. Table 1 shows the results of nine schemas calculated using the above-mentioned
input data.
 Application of MCDM Method in Selection of Schema 249

Fig. 2. Schemas for double lift table design

4 Determining the Best Schema for Double Scissor Lift Tables

This section explains how to use the MOORA method to solve the MCDM problem and
the Entropy method to calculate the creation weights to find the best schema for double
lift tables.

4.1 Determining the Creation Weights

The weights of criteria are calculated using the Entropy method (see Sect. 2.2) as follows:
Eq. (9) is used to compute the normalized values pij . Next, Eq. (10) was used to calculate
the Entropy value for each indicator mej . Lastly, using Eq. (11), compute the criteria’s
wj weight. Ra and MRS weights were determined to be 0.4993 and 0.5007, respectively.
250 N. H. Linh et al.

Fig. 3. Schema for determining output factors

Table 1. Output results of different schemas

Solution tl (s) Wm (N)

1 17.893 269.924
2 15.097 69.866
3 66.278 911.029
4 39.934 156.367
5 26.623 104.244
6 938.700 38.137
7 26.623 104.244
8 81.743 414.174
9 40.530 714.099

4.2 Determining the Best Schema for Double Scissor Lift Tables by Using
MOORA Method
The steps for MCDM using the MOORA method (see Sect. 2.1) are as follows: Compute
the normalized matrix using Eqs. (2) or (3). Equation (4) is then used to compute the
normalized decision-making matrix. After which, using Eq. (5) compute the decision-
making matrices. Whereupon, using Eqs. (6) and (7) compute Pi and Ri . Then, using
the Formula (8) calculate Qi . Finally, arrange the alternatives in such a way that the one
with the lowest Qi is the best.
 Application of MCDM Method in Selection of Schema 251

Table 2. Several calculated results and ranking alternatives

Input Data Xij Wij

Trial. Ri Pi Qi Rank
tl Wm tl Wm tl Wm
1 17.893 269.924 0.019 0.211 0.009 0.106 0.009 0.106 0.096 4
2 15.097 69.866 0.016 0.055 0.008 0.027 0.008 0.027 0.019 8
3 66.278 911.029 0.070 0.712 0.035 0.357 0.035 0.357 0.322 1
4 39.934 156.367 0.042 0.122 0.021 0.061 0.021 0.061 0.040 5
5 26.623 104.244 0.028 0.081 0.014 0.041 0.014 0.041 0.027 6
6 938.700 38.137 0.991 0.030 0.495 0.015 0.495 0.015 -0.480 9
7 26.623 104.244 0.028 0.081 0.014 0.041 0.014 0.041 0.027 7
8 81.743 414.174 0.086 0.324 0.043 0.162 0.043 0.162 0.119 3
9 40.530 714.099 0.043 0.558 0.021 0.279 0.021 0.279 0.258 2

Table 2 shows several calculated results and the ranking of solutions the double
scissor lift table when using MOORA method to solve MCDM problem and the Entropy
method to calculate the weights of criteria. Moreover, Fig. 4 shows the relation between
the solutions and the values of Qi . From Table 2 and Fig. 4 it was found that the best
schema for the double lift tables is solution 3. This is because this solution has the
maximum value of Qi (0.322). From Table 1 it is noted that this solution has a lift time
of 66.28 (s) and a maximum lift force of 911.03 (N).

Fig. 4. Relation between the values of Qi and solutions

252 N. H. Linh et al.

5 Conclusions
The results of a study to determine the best schema for double scissors lift tables to gain
both the minimum lift time and the maximum lift force simultaneously are presented in
this paper. The MOORA method was used to solve the MCDM problem in this work, and
the Entropy method was also used to determine the weight of the criteria. The findings
of the study lead to the following conclusions:

– This was the first time an MCDM method (the MOORA method) was used to deter-
mine the best schema for double scissor lift tables while meeting two conflicting
criteria: minimum lift time and maximum lift force;
– Solution 3 is the best schema for double scissor lift tables because it has the highest
value of Qi (0.322);
– Solution 3 can provide a lift time of 66.28 (s) and a maximum lift force of 911.03 (N).

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Kumar, M.K., et al.: Design & analysis of hydraulic scissor lift. Int. Res. J. Eng. Technol.
(IRJET) 3(6), 1647–1653 (2016)
2. Zhang, W., et al.: A study on the static stability of scissor lift. Open Mech. Eng. J. 9(1) (2015)
3. Rasovic, N., Vučina, A., Obad, M.: Stress analysis of lifting table using finite element method.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2019)
4. Dang, A.-T., Nguyen, D.-N., Nguyen, D.-H.: A study of scissor lifts using parameter design.
In: Sattler, K.-U., Nguyen, D.C., Vu, N.P., Long, B.T., Puta, H. (eds.) ICERA 2020. LNNS,
vol. 178, pp. 75–85. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-64719-3_10
5. Ganesan, G.: Strength analysis of scissor lift developed for household applications. Kauno
technologijos universitetas (2017)
6. Jiang, W.C., et al.: The parametric design of scissors lifting platform based on solidworks and
optimization analysis. In: Applied Mechanics and Materials. Trans Tech Publication (2015)
7. Paramasivam, V., et al.: Analytical investigation of hydraulic scissor lift for modular industrial
plants in ethiopia. Materials Today: Proceedings 46, 7596–7601 (2021)
8. https://www.astrolift.co.nz/product/131-Scissor-Lift-Table-Double-H-Electric. Accessed 20
June 2022
9. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev. 9 (2022)
10. Çalışkan, H., et al.: Material selection for the tool holder working under hard milling con-
ditions using different multi criteria decision making methods. Mater. Des. 45, 473–479
(2013)
11. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: a comparative study. Adv. Prod. Eng.
Manag. 16(4), 443–456 (2021)
12. Do, D.T.: A combination method for multi-criteria decision making problem in turning
process. Manuf. Rev. 8 (2021)
 Application of MCDM Method in Selection of Schema 253

13. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing

process for internal grinding. Machines 10(5), 303 (2022)
14. Huu Phan, N., Muthuramalingam, T.: Multi criteria decision making of vibration assisted
EDM process parameters on machining silicon steel using Taguchi-DEAR methodology.
Silicon 13(6), 1879–1885 (2021)
15. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
16. Alinezhad, A.: New Methods and Applications in Multiple Attribute Decision Making
(MADM). Springer (2019)
17. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Viet. J. Agricult. Sci. 17(4), 32–2331 (2019)
 Application of TOPSIS Method to Determine
Best Alternative in Wire-EDM 90CrSi Tool Steel

Trieu Quy Huy1 , Le Hong Ky2 , Le Hoang Anh2 , Bui Thanh Danh3 ,
Nguyen Manh Cuong4 , and Nguyen Thanh Tu4(B)
1 University of Economics - Technology for Industries, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nguyenthanhtucnvl@tnut.edu.vn

Abstract. The goal of this research is to find the best process input parameters
in wire-electrical discharge machining (wire-EDM) when processing 90CrSi tool
steel. The TOPSIS method was used as a tool in this study to solve the MCDM
problem. Furthermore, the MEREC method is employed to compute the weights
on the output parameters of the experiment result. According to the results of this
work, experiment No. 19 is the best experimental setup, with the following input
parameters: the cutting voltage VM = 9 (V), the pulse on time Ton = 12 (s), the
pulse off time Toff = 18 (s), the servo voltage SV = 35 (V), the wire feed WF =
8 (mm/min), and the feed speed SPD = 4.5 (mm/min).

Keywords: Wire EDM · MCDM · TOPSIS method · Surface roughness ·

Cutting speed · 90CrSi

1 Introduction

Wire-EDM is an unconventional machining process used to cut difficult-to-machine

materials and machine parts with narrow slots or small-radius inside corners. As a result,
numerous studies on the wire-EDM process and determining the optimal inputs have
been conducted so far.
In the machining process as well as wire-EDM, the output responses frequently
require different criteria, which commonly contradict each other. For example, increasing
material removal rate (MMR) will result in an increase in the depth of cut and feed rate,
as well as an increase in the surface roughness. As a result, determining the best option
for a machining process or a wire-EDM process is extremely difficult and necessitates
the use of a supporting tool, of which Multi-Criteria Decision Making (MCDM) is an
excellent choice.
In practice, the MCDM technique has been used to find the best input process param-
eters in different machining processes such as turning [1–3], milling [3, 4], grinding [3,
5], EDM [6], etc. Besides, there have been some studies on the use of MCDM in the wire-
EDM process so far. Majumder and Maity [7] predicted and optimized surface roughness

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 254–261, 2023.
https://doi.org/10.1007/978-3-031-22200-9_27
 Application of TOPSIS Method to Determine Best Alternative 255

and micro-hadnerss using a general regression neural network and Fuzzy logic combined
with multi-objective optimization on the basis of ratio analysis (MOORA). In this study,
five critical machining parameters were used as process input factors: pulse-on time,
discharge current, wire feed, wire tension, and flushing pressure. Sreeraj et al. [8] con-
ducted research on optimizing process parameters to improve machining performance
by combining different Multi-criterion Decision Making methods such as MOORA and
Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) with Prin-
cipal Component Analysis (PCA). Jaiswal and colleagues [9] presented a study on the
multi-response optimization of the wire-EDM process when machining D3 die steel.
Four process parameters were investigated in this work: pulse-on-time, pulse-off-time,
servo voltage, and wire tension. The MOORA approach was used to optimize pro-
cess parameters to achieve higher cutting speed and lower surface roughness. Patel and
Maniya [10] used the Operational competitiveness rating analysis (OCRA) method to
conduct an optimization study on wire-EDM to obtain optimum process parameters when
machining Aluminium metal matrix material Al+SiC. When using the Multi-Objective
Optimization Ratio Analysis (MOORA) method, the optimum input factors for wire-
EDM Inconel 718 were discovered in [11]. Four input parameters were investigated in
this study: pulse-on time, pulse-off time, pulse current, and servo voltage.
According to the above analysis, there have been a number of studies on the appli-
cation of MCDM methods to determine the best experimental setup when wire-EDM
with different materials. However, no research has been done with 90CrSi tool steels.
The results of using the TOPSIS method to determine the best set of input parameters
when wire-EDM 90CrSi tool steel are presented in this paper. The TOPSIS method was
chosen because it is the most widely used MCDM method in research on mechanical
machining processes. It has been used in turning [1], milling, internal grinding [5], EDM
[12, 13], PMEDM [14], and other processes. Furthermore, the MEREC method is used
to compute the criterion weights. In addition, five input parameters were investigated:
VM, Ton , Toff , SV, WF, and SPD. As a result, the best experimental setup solution was
discovered.

2 Methodology
2.1 Method for MCDM

In this study, the TOPSIS method was used to solve MCDM problem. The following
steps must be taken in order to use this method [15]:
Step 1: Forming initial decision-making matrix:
⎡ ⎤
x11 · · · x1n
⎢ x21 · · · x2n ⎥
⎢ ⎥
X =⎢ . . ⎥ (1)
⎣ .. · · · .. ⎦
xmn · · · xmn

where, m is the alternative number; n is the criteria number; x mn is the value of the
criterion n in the alternative m.
256 T. Q. Huy et al.

Step 2: Using the following equation to calculate the normalized values k ij :

xij
kij =  (2)
m 2
i=1 xij

Step 3: Calculating the weighted normalized decision matrix by:

lij = wj × kij (3)

Step 4: Determining the best alternative A+ and the worst s alternative A− by:


A+ = l1+ , l2+ , . . . , lj+ , . . . , ln+ (4)


A− = l1− , l2− , . . . , lj− , . . . , ln− (5)

wherein, lj+ and lj− are the best and worst values of the j criterion (j = 1,2,…, n).
Step 5: Calculating the values of better options Di+ and worse options Di− by the
following equations:

n  2
Di+ = lij − lj+ (6)
j=1
  2
n
Di− = lij − lj− (7)
j=1

In (6) and (7) i = 1, 2, …, m.

Step 6: Calculating closeness coefficient values Ri of each alternative by:

Di−
Ri = (8)
Di− + Di+
where, i = 1, 2, …, m; 0≤ Ri ≤ 1
Step 7: Ranking the order of alternatives by maximizing the value of R.

2.2 Method for Calculation of the Weight of Criteria

In this work, the MEREC method was used to find the weights of the criteria. The
MEREC method can be applied through the following steps [16]:

Step 1: Creating the initial matrix as in the step 1 of the TOPSIS method.
Step 2: Calculating the normalized matrix elements by:

– If the criterion j is as bigger as better:

minxij
hij = (9)
xij
 Application of TOPSIS Method to Determine Best Alternative 257

– If the criterion j is as smaller as better:

xij
hij = (10)
maxxij

Step 3: Determining the performance of the alternatives S i by:

⎡ ⎛ ⎞⎤
1   
Si = ln⎣1 + ⎝ ln hij ⎠⎦ (11)
n
j

Step 4: Finding the performance of ith alternative Sij concerning the removal of the
jth criterion by:
⎡ ⎛ ⎞⎤
 1   
Sij = Ln⎣1 + ⎝ ln hij ⎠⎦ (12)
n
k,k=j

Step 5: Determining the removal effect of the jth criterion Ej by the following
equation:
  

Ej = Sij − Si  (13)
i

Step 6: Calculating the weight for the criteria by:

Ej
wj = . (14)
k Ek

3 Experimental Setup
An experiment was carried out to determine the best solution that meets two criteria at
the same time: minimum surface roughness SR and maximum cutting speed CS when
wire-EDM 90CrSi steel. Six input parameters were carefully chosen for this experiment
(Table 1). In addition, a 2-levels 1/2 factorial experimental design was chosen. As a
result, 26–1 = 32 experimental tests were carried out. The experimental setup included
the following pieces of equipment: Fanuc Robocut −1 iA EDM machine (Fig. 1); brass
wire with a diameter of 0.25 (mm) (Taiwan); workpiece material: 90CrSi; samples with
a cross section of 22 × 22 (mm2 ); Dielectric fluid: Deionised water; surface roughness
tester: Mitutoyo 178–923-2A, SJ-201 (Japan).
The surface roughness of the workpieces was measured and the cutting speed was
calculated after conducting the experiment. Table 2 shows the various levels of input
factors as well as the output response results (Ra and CS).

4 Determining the Best Alternative in Wire-EDM 90CrSi Tool Steel

This section discusses MCDM when the TOPSIS method is used and the criteria weights
are calculated using the MEREC method.
258 T. Q. Huy et al.

Fig. 1. Wire-cut electrical discharge machine

Table 1. Input parameters and their levels

Parameter Code Unit Low High

Cutting voltage VM V 3 9
Pulse on time Ton µs 8 12
Pulse off time Toff µs 13 18
Server voltage SV V 25 35
Wire feed WF mm/minute 8 12
Feed speed SPD mm/minute 2.5 4.5

4.1 Calculating the Weights for the Criteria

The MEREC method can be used to calculate the weights for the criteria by performing
the following steps (see Sect. 2.2): calculating the normalized values hij using Eqs. (9)

and (10); determining Si and Sij using Eqs. (11) and (12); calculating the criterion
removal effect using Eq. (13), and determining the weight of the criteria wj using
Eq. (14). Furthermore, Ra and CS were discovered to have weights of 0.7098 and 0.2902,
respectively.

4.2 Determining the Best Alternative by Using TOPSIS Method

Section 2.2 describes how to use the TOPSIS method to solve the MCDM problem. As
a result, Eq. (2) is used to calculate the normalized values of kij, and Eq. (3) is used to
calculate the normalized weighted values of lij (3). Equations (4) and (5) determine the
A+ and A− values of Ra and CS (5). Ra and CS are equal to 0.1042 and 0.0755 for A+,
respectively, and 0.1441 and 0.0331 for A−. Furthermore, the Di+ and Di− values have
 Application of TOPSIS Method to Determine Best Alternative 259

Table 2. Experimental plan and output results

RunOrder VM Ton Toff SV WF SPD Ra (µm) CS (mm/s)

1 9 8 18 35 12 4.5 2.949 1.883
2 9 12 13 25 8 4.5 3.770 2.204
3 9 12 13 25 12 5.5 3.692 2.212
4 3 12 13 25 12 4.5 3.899 1.849
5 3 8 13 25 8 4.5 3.491 1.250
6 3 12 13 25 8 5.5 3.978 1.732
…
18 9 8 13 35 8 4.5 3.156 1.772
19 9 12 18 35 8 4.5 3.536 2.851
20 3 12 18 35 8 5.5 3.749 1.921
…
31 9 12 13 35 12 4.5 3.580 2.132
32 9 12 18 25 8 5.5 3.564 2.353

been calculated using Formulas (6) and (7). Finally, Eq. (8) has been used to compute
the ratio Ri. Table 3 shows the results of calculating several parameters and ranking the
alternatives using the TOSIS method.
Option 19 is the best option, according to Table 3 because it has the maximum
closeness coefficient value (R19 = 0.6589). As a result, the best solution has the following
parameters: VM = 9 (V); Ton = 12 (µs); Toff = 18 (µs); SV = 35 (V); WF = 8
(mm/min.); SPD = 4.5 (mm/min.).

5 Conclusions
In this paper the TOPSIS method was applied to find the best input factors of the wire-
EDM process when machining 90CrSi tool steel. From the results of the study, it is
recommended that using alternative 19 can achieve the lowest surface roughness and
highest cutting speed at the same time. Experiment No. 19 demonstrated the best per-
formance feature among the 32 trials, with the maximum closeness coefficient value of
0.6589. TOPSIS has identified the optimal input process parameters as VM = 9 (V);
Ton = 12 (s); Toff = 18 (s); SV = 35 (V); WF = 8 (mm/min.); SPD = 4.5 (mm/min.).
260 T. Q. Huy et al.

Table 3. Calculated results and ranking of alternatives

Trial kij Lij Di+ Di− Ri Rank

Ra CS Ra CS
1 0.1501 0.1719 0.1066 0.0499 0.0257 0.0411 0.6150 14
2 0.1919 0.2012 0.1362 0.0584 0.0363 0.0265 0.4218 8
3 0.1879 0.2019 0.1334 0.0586 0.0337 0.0277 0.4506 7
4 0.1985 0.1688 0.1409 0.0490 0.0453 0.0162 0.2635 21
5 0.1777 0.1141 0.1261 0.0331 0.0477 0.0180 0.2739 32
6 0.2025 0.1581 0.1437 0.0459 0.0494 0.0128 0.2055 24
…
18 0.1606 0.1617 0.1140 0.0469 0.0302 0.0331 0.5230 19
19 0.1800 0.2602 0.1278 0.0755 0.0235 0.0455 0.6589 1
20 0.1909 0.1754 0.1355 0.0509 0.0398 0.0198 0.3321 18
…
31 0.1822 0.1947 0.1294 0.0565 0.0315 0.0277 0.4673 9
32 0.1814 0.2148 0.1288 0.0623 0.0279 0.0330 0.5421 4

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: a comparative study. Adv. Prod. Eng.
Manag. 16(4), 443–456 (2021)
2. Do, D.T.: A combination method for multi-criteria decision making problem in turning
process. Manuf. Rev. 8 (2021)
3. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev. 9 (2022)
4. Singh, A., et al.: TOPSIS based selection of optimal end milling process parameters. In: AIP
Conference Proceedings. AIP Publishing LLC (2020)
5. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
6. Huu Phan, N., Muthuramalingam, T.: Multi criteria decision making of vibration assisted
EDM process parameters on machining silicon steel using Taguchi-DEAR methodology.
Silicon 13(6), 1879–1885 (2021)
7. Majumder, H., Maity, K.: Prediction and optimization of surface roughness and micro-
hardness using GRNN and MOORA-fuzzy-a MCDM approach for nitinol in WEDM.
Measurement 118, 1–13 (2018)
8. Sreeraj, P., et al.: Application of MCDM based hybrid optimization of WEDM process
parameters. Mater. Today: Proc. 50, 1186–1192 (2022)
 Application of TOPSIS Method to Determine Best Alternative 261

9. Jaiswal, A., et al.: Multi response optimization of wire EDM process parameters. In: IOP
Conference Series: Materials Science and Engineering. IOP Publishing (2018)
10. Patel, J.D., Maniya, K.D.: Optimization of WEDM process parameters for aluminium metal
matrix material Al+ SiC Using MCDM Methods. In: Advances in Manufacturing Processes,
pp. 59–70. Springer (2021)
11. Bavche, A.L., Valekar, M., Padaseti, B.K.: Application of MOORA to optimize WEDM
process parameters: a multi-criteria decision making approach. In: ICRRM 2019–System
Reliability, Quality Control, Safety, Maintenance and Management: Applications to Civil,
Mechanical and Chemical Engineering, p. 73 (2019)
12. Huo, J., et al.: Influence of process factors on surface measures on electrical discharge
machined stainless steel using TOPSIS. Mater. Res. Exp. 6(8), 086507 (2019)
13. Huu Phan, N., Muthuramalingam, T.: Multi-criteria decision-making of vibration-aided
machining for high silicon-carbon tool steel with Taguchi–topsis approach. Silicon 13(8),
2771–2783 (2021)
14. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
15. Hwang, C.-L., Lai, Y.-J., Liu, T.-Y.: A new approach for multiple objective decision making.
Comput. Oper. Res. 20(8), 889–899 (1993)
16. Keshavarz-Ghorabaee, M.: Assessment of distribution center locations using a multi-expert
subjective–objective decision-making approach. Sci. Rep. 11(1), 1–19 (2021)
 Application of WASPAS Method
for Determining Best Process Factors
in External Grinding

Tran Huu Danh1 , Bui Thanh Danh2 , Trieu Quy Huy3 , Tran Minh Tan4 ,
Nguyen Hong Linh5 , Tran Ngoc Giang6 , and Hoang Xuan Tu6(B)
1 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
2 University of Transport and Communications, Hanoi, Vietnam
3 University of Economics - Technology for Industries, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Electric Power University, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

hxt_dhktcn@tnut.edu.vn

Abstract. This paper describes the results of a study on multi-criteria decision-

making (MCDM) in cylindrical external grinding SKD11 tool steel to identify the
best input process parameters for dressing process in external grinding in order to
achieve minimal roundness (R) and maximum wheel life (T) at the same time. To
accomplish this, the MCDM problem was solved using the WASPAS (Weighted
Aggregates Sum Product Assessment) method, and the weights of criteria were
calculated using the Entropy method. In addition, an experiment was conducted
with six input process factors: fine dressing depth, fine dressing passes, coarse
dressing depth, coarse dressing passes, non-feeding dressing, and dressing feed
rate. The Taguchi method and an L16 orthogonal array were also used to design
the experiment. The experimental output results were measured and fed into the
MCDM problem, including the roundness and wheel life. The best dressing mode
for external cylindrical grinding has been proposed based on the findings of this
study.

Keywords: External grinding · MCDM · WASPAS method · Entropy method ·

Roundness · Wheel life · SKD11 tool steel

1 Introduction

Grinding is the most common type of machining in mechanical processing. It is very

useful for finishing and semi-finishing. In practice, the grinding wheel is continuously
worn while working, lowering the quality and productivity of the process. Therefore,
the dressing process is required to maintain the original specification of the grinding
wheel. The dressing process has an effect on the grinding wheel’s profile, topography,
and wear behavior. It has a significant impact on the efficiency and accuracy of the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 262–269, 2023.
https://doi.org/10.1007/978-3-031-22200-9_28
 Application of WASPAS Method for Determining Best Process Factors 263

grinding process. As a result, studying the dressing process as well as determining the
best grinding mode is critical.
Up to now, there have been many studies on the dressing process. The effect of
kinematical dressing parameters on the wheel topography has been evaluated in [1]. The
effect of dressing factors on grinding performance has been evaluated in [2]. The opti-
mum input dressing parameters have been determined for cylindrical external grinding
[3–5], surface grinding [6–8], and internal grinding [9, 10]. In addition, the optimization
problems for this process have been conducted to get the minimum surface roughness
[3, 5, 6, 10], the minimum flatness [8], the maximum material removal rate [9, 11],
maximum wheel life [4, 12, 13], or multi-objective function [12]. Studies to determine
the optimal input parameters for the dressing process have been conducted when grind-
ing 90CrSi [3, 6, 12], SKD11 steel [4, 7, 9–11, 14], aluminium composite [15], etc.
The results of a study on the selection of the best dressing factors in CNC cylindrical
grinding En-31 steel using the TOPSIS (Technique for Order of Preference by Similar-
ity to Ideal Solution) method were presented in [16]. A study on multi-criteria decision
making (MCDM) for dressing process in internal grinding was recently conducted [17].
In practice, MCDM methods are frequently used to find the best solution for a mechan-
ical machining process. For example in milling [18, 19], turning [18, 20, 21], grinding
[17, 18], electrical discharge machining (EDM) [22], and powder-mixed EDM [23]. In
this study, four MCDM methods were used to obtain the minimum SR and maximum
MRR simultaneously: TOPSIS, MARCOS (Measurement of Alternatives and Ranking
according to COmpromise Solution), EAMR (Evaluation by an Area-based Method of
Ranking), and MAIRCA (Multi-Attributive Ideal-Real Comparative Analysis).
It can be recognized from the above analysis, while there have been various optimiza-
tion study or MCDM study to find the optimum or the best input dressing parameters, no
MCDM studies have been conducted to determine the best dressing mode for external
cylindrical grinding when processing SKD11 tool steel.
The purpose of this paper is to present a study on MCDM for the dressing process
in external grinding. Two criteria, including the roundness R and the wheel life T, were
considered in this work. Furthermore, the WASPAS method was chosen to solve the
MCDM problem and the Entropy method was used to calculate the weights of criteria.
As a result, the WASPAS method has been successfully applied in external cylindrical
grinding to determine the best dressing mode.

2 Methodology

2.1 Method for Multi-criteria Decision Making

In this work, the WASPAS method was used to solve the MCDM problem. The following
steps are needed to use this method [24].
264 T. H. Danh et al.

Step 1: Making the first decision-making matrix:

⎡ ⎤
r11 ··· r1j · · · r1n
⎢ .. .. .. . . .. ⎥
⎢ . . . . . ⎥
⎢ ⎥
X=⎢
⎢ r1 ··· rij · · · rin ⎥
⎥ (1)
⎢ . .. .. . . .. ⎥
⎣ .. . . . . ⎦
rm1 · · · rmj · · · rmn m × n

wherein, m is the number of the alternative; n is the criteria number.

Step 2: Calculating the normalized values of the decision matrix by:
rij
rij∗ = (2)
maxrij
i
minrij
rij∗ =
i
(3)
rij

Equation (2) is used for R and (3) is used for T.

Step 3: Calculating the additive relative importance in the weighted normalized data
of each alternative:

n
Ai = rij∗ · wj (4)
j=1

In which, wj is the weight of criterion j; i = 1 ÷ m.

Step 4: Determining the multiplicative relative importance in the weighted normal-
ized data of each alternative:
n

wj
Mi = rij∗ (5)
j=1

Step 5: Calculating the joint generalizad criterion Qi by:

Qi = λ · Ai + (1 − λ)·Mi (6)

Step 6: Rank the alternatives by sorting Qi in a descending order and the highest
amount has the highest rank.

2.2 Method for Determination of Criterion Weights

The Entropy method was used to calculate the weights of the criteria in this work.
Their computation is as follows [25].
Step 1: Calculating the criation’s normalized values by the following Equation:
xij
pij = m (7)
m+ 2
i=1 xij
 Application of WASPAS Method for Determining Best Process Factors 265

Step 2: Finding the Entropy for each criation by:

m  
m 
m 
mej = − pij × ln pij − 1 − pij × ln 1 − pij (8)
i=1 i=1 i=1

Step 3: Computing the weight of each criation:

1 − mej
wj = m  . (9)
j=1 1 − mej

3 Experimental Work

To solve the MCDM problem, an experiment was carried out. Figure 1 illustrates the
external grinding experimental setup, including the following equipment: the grinding
machine CONDO-Hi-45 HTS (Japan); the grinding wheel: Ct80MV1-G 400 × 40 × 203
(Vietnam); the dresser: 3908-0088C type 2 (Russia); surface roughness tester: Mitutoyo
178-923-2A, SJ-201 (Japan). Table 1 describes the dressing process’s input parameters
and their level. Besides, the experiment was designed using the Taguchi method and the
L16 orthogonal array design (44 × 22 ). Table 2 shows the experimental plan.
The experiment is carried out in the following steps: First, the wheel dressing test
is carried out. After dressing, the samples are processed with the grinding wheel in the
following grinding mode: The workpiece is made of SKD11 tool steel; the total depth
of cut is 0.05 (mm); the grinding wheel speed is 29.3 (m/s); and the longitudinal feed
rate is 1.8 (m/min.). The values of R and T are determined as follows: Perform the
external grinding process for each wheel dressing mode using the above grinding mode.
After each test run, the grinding time and the roundness of the sample are measured.
This procedure is repeated indefinitely until the roundness exceeds the allowable value
(0.015 µm in this case). At this point, the total grinding time of all samples equals the
wheel life T. Table 2 displays the values of responses (the roundness R and wheel life
T) in the two left columns.

4 Determining the Best Experimental Setup

This section outlines how to determine the best experimental setup for external grinding
SKD11 tool steel using the WASPAS method for the MCDM problem and the Entropy
method to estimate criterion weights.

4.1 Calculating the Criation Weights

The criterion weights are calculated as follows using the Entropy method (see Sect. 2.2):
The normalized values pij are calculated using Eq. (7). The Entropy value for each criation
mej is calculated by Eq. (8). Then, the weight of the criteria wj is calculated by Eq. (9).
The weights of R and T were found as 0.3501 and 0.6499, respectively.
266 T. H. Danh et al.

Fig. 1. Experimental setup

Table 1. Input dressing factors

No Input factors Unit Levels

1 2 3 4
1 Coarse dressing depth ar mm 0.03 0.04 – –
2 Coarse dressing passes nr times 1 2 3 4
3 Fine dressing depth af mm 0.005 0.01 0.015 0.02
4 Fine dressing passes nf times 0 1 2 3
5 Non-feeding dressing n0 times 0 1 2 3
6 Dressing feed rate Sd m/min 1 1.5 – –

4.2 Determining the Best Experimental Setup

In this work, the WASPAS method is applied to solve the MCDM problem by following
steps (see Sect. 2.1): Calculate the normalized values of the decision matrix rij* by
Eq. (2) for the roundness and (3) for the wheel life. Then, compute the additive relative
importance in the weighted normalized data of each alternative using Eq. (4). After
that, determine multiplicative relative importance in the weighted normalized data using
Formula (5). The the joint generalizad criterion Qi is then found using Eq. (6). Finally,
sort Qi in a descending order to rank the alternatives. Several calculated results and the
rank of the alternative are presented in Table 3.
From Table 3 it was found that the best option is option 3. This is because it has
the highest Qi value (Qi = 0.5087). Therefore, the best experimental setup for dressing
 Application of WASPAS Method for Determining Best Process Factors 267

Table 2. Experimental plans and responses

Exp. no. Input parameters Responses

af nf n0 nr ar Sd R (µm) T (min)
1 0.005 0 0 1 0.03 1 0.0127 9.29
2 0.005 1 1 2 0.03 1.5 0.0087 8.98
3 0.005 2 2 3 0.04 1 0.0037 16.48
4 0.005 3 3 4 0.04 1.5 0.0050 8.52
5 0.01 0 1 3 0.04 1.5 0.0090 9.86
6 0.01 1 0 4 0.04 1 0.0090 9.55
7 0.01 2 3 1 0.03 1.5 0.0023 9.04
8 0.01 3 2 2 0.03 1 0.0043 9.37
9 0.015 0 2 4 0.03 1.5 0.0093 12.81
10 0.015 1 3 3 0.03 1 0.0057 10.82
11 0.015 2 0 2 0.04 1.5 0.0080 9.62
12 0.015 3 1 1 0.04 1 0.0050 9.19
13 0.02 0 3 2 0.04 1 0.0107 9.00
14 0.02 1 2 1 0.04 1.5 0.0110 10.32
15 0.02 2 1 4 0.03 1 0.0050 11.21
16 0.02 3 0 3 0.03 1.5 0.0080 8.22

process has the following input factors: ar = 0.04 (mm); nr = 3 (times); af = 0.005
(mm); nf = 2 (times); n0 = 2 (times); Sd = 1 (m/min.).

5 Conclusions

This paper presents a study on MCDM in dressing process for external grinding SKD11
tool steel using the WASPAS method for solving the MCDM problem and the Entropy
method for finding the criterion weights. To achieve this, the Taguchi method and L16
orthogonal array (44 × 22) were chosen for an experiment. From the results of the study,
several following conclusions were given:

– For the first time, the WASPAS method was applied to solve the MCDM problem in
external grinding SKD11 tool steel.
– The best setup for the input parameters of the dressing process to get the minimum
roundness and maximum wheel life was proposed: ar = 0.04 (mm); nr = 3 (times);
af = 0.005 (mm); nf = 2 (times); n0 = 2 (times); Sd = 1 (m/min.).
268 T. H. Danh et al.

Table 3. Several calculated results and rankings of alternatives

Trial rij * Ai Mi Qi Rank

R T
1 0.1842 0.5634 0.4306 0.0236 0.2271 16
2 0.2692 0.5448 0.4483 0.0334 0.2408 13
3 0.6364 1.0000 0.8727 0.1448 0.5087 1
4 0.4667 0.5167 0.4992 0.0549 0.2770 8
5 0.2593 0.5982 0.4795 0.0353 0.2574 10
6 0.2593 0.5794 0.4673 0.0342 0.2507 12
7 1.0000 0.5486 0.7067 0.1248 0.4157 2
8 0.5385 0.5685 0.5580 0.0696 0.3138 6
9 0.2500 0.7774 0.5927 0.0442 0.3185 4
10 0.4118 0.6562 0.5706 0.0615 0.3161 5
11 0.2917 0.5836 0.4814 0.0387 0.2601 9
12 0.4667 0.5575 0.5257 0.0592 0.2925 7
13 0.2188 0.5460 0.4314 0.0272 0.2293 15
14 0.2121 0.6261 0.4812 0.0302 0.2557 11
15 0.4667 0.6803 0.6055 0.0722 0.3389 3
16 0.2917 0.4987 0.4262 0.0331 0.2297 14

Acknowledgment. Thai Nguyen University of Technology provided support for this work.

References
1. Klocke, F., Linke, B.: Mechanisms in the generation of grinding wheel topography by dressing.
Prod. Eng. Res. Devel. 2(2), 157–163 (2008)
2. Buttery, T., et al.: Some effects of dressing on grinding performance. Wear 55(2), 195–219
(1979)
3. Tu, H.X., et al.: Influence of dressing parameters on surface roughness of workpiece for grind-
ing hardened 9XC tool steel. In: IOP Conference Series: Materials Science and Engineering.
IOP Publishing (2019)
4. Le, H.A., et al.: Impacts of dressing conditions on wheel lifetime when external grinding of
SKD11 steel. In: Nguyen, D.C., Vu, N.P., Long, B.T., Puta, H., Sattler, K.-U. (eds.) ICERA
2021. LNNS, vol. 366, pp. 479–487. Springer, Cham (2022). https://doi.org/10.1007/978-3-
030-92574-1_50
5. Giang, T.N., et al.: Studying the influence of dressing parameters on the surface roughness
when conducting the external grinding of SKD11 steel. In: Solid State Phenomena. Trans
Tech Publication (2021)
6. Trung, D.D., et al.: Calculating effects of dressing parameters on surface roughness in surface
grinding. In: International Conference on Engineering Research and Applications. Springer
(2019)
 Application of WASPAS Method for Determining Best Process Factors 269

7. Hong, T.T., et al.: Optimization of dressing parameters in surface grinding SKD11 tool
steel by using Taguchi method. In: International Conference on Engineering Research and
Applications. Springer (2020)
8. Tran, T.H., et al.: Optimizing dressing conditions for minimum flatness tolerance when
grinding SKD11 tool steel. In: Materials Science Forum. Trans Tech Publication (2021)
9. Nguyen, A.T., et al.: Optimization of dressing parameters for minimum surface roughness
and maximum material removal rate in internal grinding of SKD11 tool steel. In: E3S Web
of Conferences. EDP Sciences (2021)
10. Hung, L.X., et al.: Determining the optimum set of dressing parameters satisfying minimum
surface roughness when conducting the internal grinding of hardened SKD11 steel. In: Solid
State Phenomena. Trans Tech Publication (2021)
11. Hong, T.T., et al.: Multi response optimization of dressing conditions for surface grinding
SKD11 steel by HaiDuong grinding wheel using grey relational analysis in Taguchi method.
In: International Conference on Engineering Research and Applications. Springer (2020)
12. Hong, T.T., et al.: Multi-criteria optimization of dressing parameters for surface grinding
90CrSi tool steel using Taguchi method and grey relational analysis. In: Materials Science
Forum. Trans Tech Publication (2020)
13. Tran, T.H., et al.: Improvement of wheel life by optimization of dressing parameters in surface
grinding of SKD11 steel. In: Materials Science Forum. Trans Tech Publication (2021)
14. Nguyen, T.T., et al.: Influence of dressing parameters on roundness tolerance in cylindrical
external grinding SKD11 tool steel. In: Materials Science Forum. Trans Tech Publication
(2021)
15. Shanawaz, A.M., et al.: Grinding of aluminium silicon carbide metal matrix composite mate-
rials by electrolytic in-process dressing grinding. Int. J. Adv. Manuf. Technol. 57(1), 143–150
(2011)
16. Patil, S.S., Bhalerao, Y.J.: Selection of levels of dressing process parameters by using TOPSIS
technique for surface roughness of en-31 work piece in CNC cylindrical grinding machine.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2017)
17. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
18. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev. 9 (2022)
19. Çalışkan, H., et al.: Material selection for the tool holder working under hard milling con-
ditions using different multi criteria decision making methods. Mater. Des. 45, 473–479
(2013)
20. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: a comparative study. Adv. Prod. Eng.
Manag. 16(4), 443–456 (2021)
21. Do, D.T.: A combination method for multi-criteria decision making problem in turning
process. Manuf. Rev. 8 (2021)
22. Huu Phan, N., Muthuramalingam, T.: Multi criteria decision making of vibration assisted
EDM process parameters on machining silicon steel using Taguchi-DEAR methodology.
Silicon 13(6), 1879–1885 (2021)
23. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
24. Alinezhad, A.: New Methods and Applications in Multiple Attribute Decision Making
(MADM). Springer (2019)
25. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Viet. J. Agricult. Sci. 17(4), 32–2331 (2019)
 Auto-balancing Ballbot Systems:
A Fractional-Order Sliding Mode Based
Radial-Basis Neural Network Approach

Minh Duc Pham1 , Cong Minh Pham1 , Thu Giang Do1 , Phuong Thao Dao1 ,
Trong Dang Van1 , Quoc Dong Hoang2 , and Tung Lam Nguyen1(B)
1 Ha Noi University of Science and Technology, Dai Co Viet 01, Hanoi, Vietnam
lam.nguyentung@hust.edu.vn
2 Vietnam Marinetime University, Hai Phong, Vietnam

Abstract. This research presents an adaptive hierarchical sliding mode control

approach based on fractional order derivatives and radial basis function neural
networks for an uncertain 2D Ballbot system. Instead of using fixed integer order
derivatives as usual, fractional order derivatives are used because of the possibility
of adjustment. To estimate the unknown dynamic model, a radial basis function
neural network is used. The control law is constructed from a candidate of the
Lyapunov function in order to assure the stability of sliding surfaces, and an
updated law for neural network weight matrices is derived from it. Simulation
findings show that the suggested control technique is effective, resulting in a
lower swing and more precise location tracking as anticipated.

Keywords: Radial basis neural network · Sliding mode control ·

Fractional-order derivative · Ballbot

1 Introduction
Ballbot is a form of mobile robot that uses the inverse pendulum concept. The ballbot
has the enormous advantage of allowing the robot to move in all directions while also
lowering its size since there is only one point of contact with the ground. As a result,
even under restricted working environments, the robot can perform well.
The first ballbot was built in 2005 [1], which had the height, width, and weight of
a person and a high center of gravity that balanced dynamically on a single spherical
wheel. Ball IP, a balanced ballbot on a spherical wheel, was released in 2009 [2] as
another balanced ballbot variant. A transmission mechanism on the robot comprised
three omnidirectional wheels with stepper motors in contact with a spherical wheel,
allowing for friction control and reduction. The ballbot in this paper is for the purpose
of transporting objects, so we chose to design the controller based on the 2D model of
Ball IP because it has a compact size, low center of gravity, and is easily designed to
adapt to a variable mass placed on the body.
From the control perspective, the ballbot is a robot with an under-actuated system
dynamic model, in which the control signals are less than the outputs. Because the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 270–280, 2023.
https://doi.org/10.1007/978-3-031-22200-9_29
 Auto-balancing Ballbot Systems: A Fractional-Order Sliding Mode 271

controller must maintain the robot’s balance while adhering to the specified trajectory,
this characteristic complicates the control challenge. PD control [3], double-loop control
[4], and sliding mode control [5] have all been offered as solutions to this problem.
Sliding mode control (SMC) is a nonlinear control method that is known to be stable,
simple, and effective, and it no longer has the same severe standards as it did previously.
Qian’ hierarchical sliding mode control (HSMC) [6, 7] is an example of a contribution
that improves the quality of controller design for under-actuated mechatronic systems.
Prior research [8, 9] has highlighted the development and implementation of HSMC
for underactuated systems. The minimal susceptibility to changes in system parameters,
strong noise resistance, and quick dynamic reactions are all advantages of this tech-
nology. However, when the system tracks the sliding surface, the sign function in the
control signal can cause chattering in which high-frequency oscillation around the slid-
ing surface exists in the system’s trajectory. Furthermore, creating an SMC necessitates
knowledge of the system’s exact mathematical model as well as additional marginal
circumstances.
Neural networks (NNs) [10] is an intelligent method to compute the equivalent con-
trol of such partly known or unknown systems. In [11], the authors use radial-basis
function neural networks (RBFNNs) to constitute an approximation of the dynamics
function of the ballbot system. The fundamental drawback with such network architec-
ture is that as the number of inputs grows, it may become more difficult to calculate.
Using HSMC and estimating dynamical functions by RBFNNs are a solution that can
overcome the weakness of both HSMC and RBFNNs, which is applied in [11, 12].
The concept of fractional order derivatives has been applied in the construction of
controllers for some objects [13] but it has never been used in controlling the Ballbot
systems. Instead of using fixed integer order derivatives as usual, fractional order deriva-
tives are considered more favorable because of the possibility of adjustment [14, 15].
By adjusting the orders of derivatives, optimal responses can be obtained.
The main contribution of this paper is applying fractional derivatives to the con-
trol of ballbot systems, in combination with HSMC controllers and RBFNNs to form
thefractional order derivative hierarchical sliding mode control - radial basis function
(FoHSMC - RBF) controller for use in a ballbot system with friction, centrifugal force,
and gravity uncertainties. An controller not only adapts well to parameter deviations but
also can be flexibly adjusted to enhance the system control performance.

2 2D Modelling of the Ballbot System

2.1 Coordinate Systems and Assumptions

The ballbot system is composed of five rigid parts: the ball, the body, and three omni-
directional wheels (considered as a single wheel in the center). The 2D model of the
ballbot system on z-y plane possesses two degrees of freedom; therefore, the minimal
 T
coordinate vector is defined as qx = yk θx ∈ R× , where yk is the position of the
ball on the z-y plane; θx is the body tilt angle around the x-axes, respectively (Fig. 1).
For analytical modeling, the following assumptions are provided to idealize the
model, facilitate for calculations without causing any significant error for the system:
272 M. D. Pham et al.

Fig. 1. 2D model of ballbot system

(a) the body, omnidirectional wheels, and floor are rigid; (b) no slip occurs between the
ball and the floor and between the ball and the omnidirectional wheels; (c) the contacts
between the ball and the floor and between the omnidirectional wheels and the ball are
always point contacts; (d) the floor is flat and horizontal; and (e) no yaw or spinning
motion for either the ball or the body occurs.

2.2 Dynamic Model

The ballbot system dynamics are described by obtaining the kinetic and potential of the
ball, body, and omnidirectional wheels. The kinetic energy of the ball in the z-y plane
contains translational and rotational parts. The potential energy of the ball is zero, when
choosing a coordinate frame in the center of the ball.
 
1 Ik
Tkx = mk + 2 y˙k 2 , Vkx = 0 (1)
2 rk

where mk , Ik and rk are the mass, the moment of inertia, and the radius of the ball,
respectively. The rotation of the actuating wheel:
rk  
ϑ˙x = ϕ˙x − θ˙x − θ˙x (2)
rw
where θx and ϕx are the rotations of the ball and the body, respectively , rw is the radius of
the omnidirectional wheel. The omnidirectional wheels are attached to the body. Thus,
only the rotational motion of the omnidirectional wheels must be calculated.
 2
1 ˙ 1 rk  
Twx = Iw ϑx = Iw
2 ˙ ˙
ϕ˙x − θx − θx
2 2 rw
Iw  2
= 2 y˙k − (rk + rw )θ˙x , Vwx = 0 (3)
rw
where mw and Iw are the mass and the moment of inertia of an omnidirectional wheel,
respectively. For the body:
1 ˙2 1  2 1
Ix θx + ma y˙k + l θ˙x cosθx + ma l 2 θ˙x sinθx ,
2
Tax =
2 2 2
 Auto-balancing Ballbot Systems: A Fractional-Order Sliding Mode 273

Vax = ma glcosθx (4)

where ma and Ia are the mass and the moment of inertia of the body, respectively, l
indicates the distance between the center of the ball and the center of mass of the body,
and g is the gravitational acceleration. The friction vector is defined as

   T
D q˙x = bk y˙k brx θx (5)

where bk and brx are the viscous damping coefficients that model the spherical wheel–
floor friction and spherical wheel–body friction in the z–y plane, respectively.
The Euler - Lagrange equation that describes the motion of the ballbot system is

 
d δLx δLx 1 1  
− = τx − D q˙x (6)
dt δ q˙x δqx r w r k + rw
From the (6), the dynamic equations are expressed as follows
      
y¨k fx1 q˙x , qx  g q
¨ = + x1  x  τx (7)
θx fx2 q˙x , qx gx2 qx
where
  
(a4 − a3 cosθx ) a5 sinθx − brx θ˙x + a2 a3 θ˙x sinθx − by y˙k
2
 
fx1 q˙x , qx =
Ax
  
(a4 − a3 cosθx ) a3 θ˙x sinθx − by y˙k +a1 a5 sinθx − brx θ˙x
2
 
fx2 q˙x , qx =
Ax
  a2 + (rk + rw )(a3 cosθx − a4 )   a3 cosθx + a4 − a1 (rk + rw )
gx1 qx = ; gx2 qx = ;
Ax rw Ax rw
Ik Iw
Ax = a1 a2 − (a4 − a3 cosθx )2 ; a1 = mk + ma + 2 + 2 ;
rk rw
Iw (rw + rk )2 Iw (rw + rk )
a2 = Ia + ma l 2 + 2
; a3 = ma l; a4 = ; a5 = ma gl.
rw rw2

3 Control Design
3.1 Design of FoHSMC
A FoHSMC controller is developed for the 2D ballbot system, a type of single-input-
multiple-output (SIMO) system. It is assumed that all state variables are measurable.
The FoHSMC controller is designed to help yk tracks to ykd , and make sure θx converge
to 0. During the design process, we hypothesized that q˙x = 0 and q¨x = 0.
First, to design this controller in the z-y plane, a suitable pair of sliding mode surfaces
(SMSs) of the ball and body subsystems is defined as:

s1 = Dμ e1 + c1 e1 ; s2 = Dμ e2 + c2 e2 (8)
274 M. D. Pham et al.

where e1 = yk − ykd , e2 = θx − θxd = θx are tracking errors of the state variable, Dμ is

the fractional order derivation defined in [13], c1 , c2 are candidate constants. The second
layer SMS is constructed based on the first layer SMSs s1 and s2 is defined as follows:
 
s = αs1 + βs2 = α(Dμ e1 + c1 e1 ) + β Dμ e2 + c2 e2 (9)

where α, β are sliding mode parameters. From the definitions of e1 , e2 , and (7), then:

ṡ = αDμ−1 (fx1 + gx1 τx ) + αc1 e˙1 + βDμ−1 (fx2 + gx2 τx ) + βc2 e˙2 (10)

Based on the methodology of equivalent control of variable structure control, total

control law include two parts as follows:

τx = τxeq + τxsw (11)

where τxeq and τxsw are the equivalent law and switching law. In order to ensure the
stability of second-level sliding surface, let:
   
˙ 0
αDμ−1 fx1 + gx1 τxeq + αc1 e˙1 + βDμ−1 fx2 + gx2 τxeq + βc2 e2 = (12)

   
αDμ−1 gx1 τxeq + βDμ−1 gx2 τxeq = −Dμ−1 (εsgn(s)) (13)

From (12) to (14), the control law is deduced as follows

αfx1 + βfx2 + D1−μ (αc1 e˙1 + βc2 e˙2 ) + εsgn(s)

τx = − (14)
αg x1 + βgx2
Finally, the sgn function component in the control law (14) can cause chattering.
Therefore, it could be replaced by sat function.

3.2 Adaptive FoHSMC Based on RBFNN

In this section, the RBFNN is utilized to build up the adaptive FoHSMC for uncertain
ballbot system and the stability of closed-loop system is analyzed.
  
Separating two dynamical functions of system fx1 q˙x , qx and fx2 q˙x , qx , which are
unknown into two elements as follows:

fx1 =
fx1 + fcx1 , fx2 =
fx2 + fcx2 (15)

where fcx1 and fcx2 are the certainty elements and

fx1 = W1T h+ε1 and
fx2 = W2T h+ε2
are the uncertainty elements with ideal weight matrices W 1 and W2 , ε1 and ε2 are
approximated errors with positive constants ε1∗ , ε2∗ satisfying |ε1 | ≤ ε1∗ and |ε2 | ≤ ε2∗
and h is a nonlinear activation function of hidden layer of the two networks.
  


Defining W 1 and W 2 as estimations of unknown W1 and W2 , respectively,
f x1 and

f x2 as estimations of
fx1 ,
fx2 satisfying:
 

T 
T

f x1 = W 1 h,
f x2 = W 2 h (16)
 Auto-balancing Ballbot Systems: A Fractional-Order Sliding Mode 275

Accordingly, the errors of weight are identified as:

 

W̃1 = W1 − W 1 , W̃2 = W2 − W 2 (17)

The signal in (14) and the derivative of the second layer in (10) are rewritten as:
 
 
α(
f x1 + fcx1 ) + β(
f x2 + fcx2 ) + D1−μ αc1 y˙k + βc2 θ˙x + εsgn(s)
τx = − (18)
αg 1 + βg2
 
ṡ = αDμ−1
fx1 + f cx1 + gx1 τx + αc1 y˙k + βDμ−1 (
fx2 + gx2 τx ) + βc2 θ˙x (19)

Considering a Lyapunov function candidate as follows:

1 2
(s + W̃1T F1− W̃1 + W̃2T F2− W̃2 )
1 1
V = (20)
2
Differentiating the Lyapunov function combined with (16), (17), we obtain:

1 ˙


V̇ = sDμ−1 (−εsgn(s) + αε1 + βε2 ) + sDμ−1 (α W̃1T h) − W̃1T F1− W1


1 ˙


+ sDμ−1 (β W̃2T h) − W̃2T F2− W2 (21)

Equation (21) can be rewritten as follow:

1 ˙


V̇ = sDμ−1 (−εsgn(s) + αε1 + βε2 ) + sα(Dμ−1 (W̃1T h) − W̃1T h) + W̃1T (sαh − F1− W1 )
1 ˙


+ sβ(Dμ−1 (W̃2T h) − W̃2T h) + W̃2T (sβh − F2− W2 ) (22)

Choosing double layers RBF as following to ensure the asymptotical stability of the
second sliding surface:


˙ 

˙
W1 = αF 1 sh, W2 = βF 2 sh (23)

Then we obtain:

V̇ = sDμ−1 (−εsgn(s) + αε1 + βε2 ) + sα(Dμ−1 (W̃1T h) − W̃1T h)

+ sβ(Dμ−1 (W̃2T h) − W̃2T h) (24)

Subsequently, Eqs. (21)–(24) yields:

V̇ ≤ sDμ−1 (−εsgn(s))

+ |s| |Dμ−1 (α W̃1T h) − W̃1T h| + |Dμ−1 (β W̃2T h) − W̃2T h| + α|ε1 | + β|ε2 | (25)


It is obvious that W̃1T h = dfx1 − d f x1 + ε1 is difference between of dfx1 and its esti-
mation so W̃ T h is bounded and continuous. As a result, W̃ ˙ T h is bounded and continuous.
1 1
276 M. D. Pham et al.

˙ T h is also bounded and continuous. Therefore, there are

and
Similarly, W̃2T h and W̃ 2 1

2 satisfying:

|Dμ−1 (W̃1T h) − W̃1T h| ≤

1 < ∞, |Dμ−1 (W̃2T h) − W̃2T h| ≤
2 < ∞ (26)

Let γ = Dμ−1 (ε), (0 < γ < ε), we have:

 
V̇ ≤ −|s|γ + |s|Dμ−1 αε1∗ + βε2∗ + |s|α
∗1 + |s|β
∗2 (27)
 
From Eq. (27), if parameter γ is designed satisfying γ > Dμ−1 αε1∗ + βε2∗ +α
∗1 +
∗
β
2 , then V̇ ≤ 0.

4 Numerical Simulation
In this section, the simulations are carried out not only for the two control rules designed,
FoHSMC - RBF and FoHSMC, but also for the conventional SMC, which has the same
approach and establishment as in the study [5]. By analyzing the performances of these
controllers, it is possible to somewhat assess their quality and effectiveness. The system
parameters are provided in Table 1.

Table 1. Ballbot parameters

System parameter Variable Value Unit

Mass of the ball mk 2.29 kg
Mass of the actuating wheel mw 3 kg
Mass of the body ma 9.2 kg
Radius of the ball rk 0.125 m
Radius of the omniwheels rw 0.06 m
Radius of the body (cylinder) rk 0.1 m
Distance from the mass center of the body to that of the ball l 0.339 m
Moment of the inertia of the ball Ik 0.0239 kgm2
Moment of the inertia of the wheel Iw 0.00236 kgm2
Moment of the inertia of the body Ia 0.00945 kgm2
The spherical wheel–floor friction bk 5 Ns.m
The spherical wheel–body friction brx 3.68 Nms/rad

The sliding mode parameters are provided as α = 3, β = 10, c1 = 2, c2 = 5, and

ε = 4.
Balancing control with initial non-zero tilt and designed position
In the first simulation, the ballbot with initial angle value of 5.7◦ is designed to move
from position y = 0 to new position y = 5 and it is able to be self-stable in there.
 Auto-balancing Ballbot Systems: A Fractional-Order Sliding Mode 277

The simulation results of FoHSMC method in comparison of three differential orders

(k = 0.9, k = 1.0, k = 1.1) are shown in Fig. 2a, b and the torque input is shown in
Fig. 2c. As we can see, the performance of the FoHSMC methods shows better quality
in terms of setting time. However, the increase of the differential order can lead to the
change of the overshoot. Therefore, the FoHSMC methods can maintain the stability
with different orders of derivation and the order will be adjust depending on the purpose
of controllers.

(a)

(b)

(c)
Fig. 2. (a) Position tracking with initial tilt angles and designed position. (b) Balancing
performance with initial non-zero tilt angles and designed position. (c) The torque input signal
278 M. D. Pham et al.

Balancing control with mass variation

In practice, the ballbot may carry objects of unknown mass (with unknown moment of
inertia). Thus, the uncertainty mass of the body in third simulation is changing in order to
test the performance of radial basic function neurol network. The new mass is different
by 50%, compared to the previous. As can be seen from the Fig. 3a and b, the robot
using FoHSMC - RBF arrived at the expected position in nearly 25s, which is faster
than that using other controllers and the absolute value of the angle is less than 0.15 rad.
Therefore, the digital controller using FoHSMC - RBF law has good performance in the
actual environment which contains parametric uncertainties.

(a)

(b)
Fig. 3 (a) Control with mass variation. (b) Control with mass variation

5 Conclusion
In this research, the combination of FoHSMC and RBF method had solved the prob-
lem off controlling an inverted – pendulum ballbot with the model having uncertainty
elements and variable load parameters. Obviously, practical model is affected by many
uncertainties, including the nonlinearities and unmodeled components. Therefore, the
first main target of this research is developing an adaptive control law to tackle this prob-
lem via approximating the uncertainties with the RBF neural network. The second target
of this research is to develop an approach of FoHSMC not only enhances the stability of
the controller according to the Lyapunov criterion but also perform well in minimizing
 Auto-balancing Ballbot Systems: A Fractional-Order Sliding Mode 279

response time, settling time and overshoot. Consequently, numerical simulation results
are given in some cases to investigate the effectiveness of this control law. Besides, the
comparisons between SMC, FoHSMC and FoHSMC - RBF controllers are implemented
to explore the outstanding merit of using the adaptive law.

Acknowledgement. Van Trong Dang was funded by Vingroup JSC and supported by the Mas-
ter, PhD Scholarship Programme of Vingroup Innovation Foundation (VINIF), Institute of Big
Data, code VINIF.2021.ThS.26. This research is funded by the Hanoi University of Science and
Technology (HUST) under project number T2022-PC-003.

References
1. Mauro, R.P., Correlatore, S., Pastorelli, S., Paolo, I., Melchiorre, M., Chevalie, J.P.:
Preliminary study for the design of a Ballbot (2018)
2. Kumaga, M., Ochiai, T.: Development of a robot balanced on a ball—application of passive
motion to transport. In: Proceedings of IEEE International Conference Robot. Automation,
pp. 4106–4111 (2009). https://doi.org/10.1109/ROBOT.2009.5152324
3. Singhose, W., Kim, D., Kenison, M.: Input shaping control of double-pendulum bridge crane
oscillations. J. Dyn. Syst. Meas. Control. Trans. ASME 130(3), 1–7 (2008). https://doi.org/
10.1115/1.2907363
4. Dong, H.Q., Lee, S.G., Dinh Ba, P.: Double-loop control with proportional-integral and
partial feedback linearization for a 3D gantry crane. In: International Conference Control
Automations System, vol. 2017, October, no. ICCAS, pp. 1206–1211 (2017). https://doi.org/
10.23919/ICCAS.2017.8204411
5. Pham, D.B., Kim, J., Lee, S.G.: Combined control with sliding mode and Partial feedback
linearization for a spatial ridable ballbot. Mech. Syst. Signal Process. 128, 531–550 (2019).
https://doi.org/10.1016/j.ymssp.2019.04.008
6. Qian, D., Yi, J., Zhao, D.: Control and cybernetics hierarchical sliding mode control for a
class of SIMO under-actuated systems
7. Qian, D., Yi, J.: Hierarchical sliding mode control for under-actuated cranes design, analysis
and simulation
8. Pham, D.B., Lee, S.-G.: Aggregated hierarchical sliding mode control for a spatial ridable
ballbot. Int. J. Precis. Eng. Manuf. 19(9), 1291–1302 (2018). https://doi.org/10.1007/s12541-
018-0153-5
9. Hoang, Q.-D., Park, J.-G., Lee, S.-G., Ryu, J.-K., Rosas-Cervantes, V.A.: Aggregated hierar-
chical sliding mode control for vibration suppression of an excavator on an elastic foundation.
Int. J. Precis. Eng. Manuf. 21(12), 2263–2275 (2020). https://doi.org/10.1007/s12541-020-
00422-9
10. Liu, J.: Radial Basis Function (RBF) Neural Network Control for Mechanical Systems.
Springer, Berlin, Heidelberg (2013). https://doi.org/10.1007/978-3-642-34816-7
11. Le Xuan, H., et al.: Adaptive hierarchical sliding mode control using an artificial neural
network for a ballbot system with uncertainties. J. Mech. Sci. Technol. 36(2), 947–958 (2022).
https://doi.org/10.1007/s12206-022-0141-7
12. Le, H.X., Nguyen, T.V., Le, A.V., Phan, T.A., Nguyen, N.H., Phan, M.X.: Adaptive hierarchi-
cal sliding mode control using neural network for uncertain 2D overhead crane. Int. J. Dyn.
Control 7(3), 996–1004 (2019). https://doi.org/10.1007/s40435-019-00524-x
13. Huang, J., Li, H., Chen, Y., Xu, Q.: Robust position control of PMSM using fractional-order
sliding mode controller. Abstr. Appl. Anal. 2012 (2012). https://doi.org/10.1155/2012/512703
280 M. D. Pham et al.

14. Faieghi, M.R., Delavari, H., Baleanu, D.: Control of an uncertain fractional-order Liu system
via fuzzy fractional-order sliding mode control. JVC/J. Vib. Control 18(9), 1366–1374 (2012).
https://doi.org/10.1177/1077546311422243
15. Delavari, H., Ghaderi, R., Ranjbar, A., Momani, S.: Fuzzy fractional order sliding mode
controller for nonlinear systems. Commun. Nonlinear Sci. Numer. Simul. 15(4), 963–978
(2010). https://doi.org/10.1016/j.cnsns.2009.05.025
 Backstepping Sliding Mode Control Design
for Active Suspension Systems in Half-Car
Model

Viet Hung Pham , Duc Thinh Le , Nhu Toan Nguyen, Van Trong Dang ,
Van-Anh Nguyen-Thi , Danh Huy Nguyen, and Tung Lam Nguyen(B)

Hanoi University of Science and Technology, Hanoi 100000, Vietnam

lam.nguyentung@hust.edu.vn

Abstract. The suspension system in a car plays an important role in ride comfort,
road holding, and ride safety, as it transmits all forces from the road to the vehicle
body. The dynamics of the suspension system not only influence considerably
the handling capabilities of the car but also prevent physical fatigue of driver and
passengers and reduce the rate of fatality in traffic accidents. This paper concerns
with an active suspension system in a half-car model, if we consider the left and
right side of the car is symmetrical. We proposed a backstepping controller for
this system with a reference trajectory which improves the ride comfort of the
people sitting in the car. The system’s zero dynamics are mathematically proved.
Control quality is validated by simulations in different scenarios and compared to
the passive suspension system in MATLAB/Simulink.

Keywords: Active car suspension system · Half-car model · Backstepping

sliding mode control

1 Introduction
With recent development in the car industry, the requirements for car performance are
increasing day by day. A car suspension system, which is comprised of a wishbone,
spring, and shock absorber, is significantly important in handling cars, comforting people
occupying the car by isolating road noise, bumps, and vibrations from transmitting to
the vehicle body [1]. An active suspension system has to stabilize the vertical and pitch
motions of the vehicle body, provide good road holding, and keep suspension spaces and
output of actuators under their corresponding physical limits. Different control schemes
have been proposed to tackle the aforementioned problems of the active suspension
system. Linear Quadratic Gaussian (LQG) control [2], adaptive control [3], and robust
control [4] have been utilized for controlling the active suspension system. In addition,
some intelligent approaches such as Fuzzy Logic Control (FLC) [5], Genetic Algorithm
(GA) [6], radial basis function networks [7, 8] and the combination of the mentioned
methods have been applied for the active suspension system.
Backstepping sliding mode control is a popular control method for non-linear sys-
tems. It combines the immunity to uncertainties of the backstepping technique and

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 281–289, 2023.
https://doi.org/10.1007/978-3-031-22200-9_30
282 V. H. Pham et al.

the robustness of sliding mode control to external disturbances. Utilizing this property,
many researchers have implemented this control algorithm for a wide range of non-linear
objects which contain uncertainties in the model and are susceptible to disturbances [9–
11]. The nonlinear of the active suspension system for the half car model makes it highly
suitable for using a backstepping sliding mode approach to achieve great performance.
In this paper, a backstepping sliding mode is proposed to handle this active suspension
system in a non-linear half-car model. A stability analysis is performed on the zero
dynamics of the system to ensure all the safety performance indicators are all bounded
and the corresponding upper bounds are estimable. A reference trajectory is chosen
such that the driver and passengers always experience a smooth operation of this active
suspension system. Root mean square (RMS) values of heave and pitch accelerations
are greatly reduced compared to the passive suspension system and active suspension
system without a reference trajectory. Simulations in MATLAB/Simulink illustrate the
effectiveness of the proposed approach.

2 Mathematical Model

Fig. 1. Half-car suspension model [1].

As shown in Fig. 1, if the left and right parts of the 4-wheel vehicle are considered
symmetrical, a half-car model is obtained. The tire is modeled as a combination of
spring and damper. Using Newton’s second law, the dynamic equations of motion for
this system can be derived as follows:
⎧
⎪ M ż + F fd + Frd +
⎪
⎨  Ffs + Frs = uz ,
I ϕ̈ + a Ffd + Ffs − b(Frd + Frs ) = uϕ ,
(1)
⎪
⎪ m z̈ − Ffs − Ffd + Fft + Ffb = −uf ,
⎩ f f
mr z̈r − Frs − Frd + Frt + Frb = −ur .

where: Ffs = kfs1 yf + knfs1 yf3 , Frs = krs1 yr + knsr1 yr3 , (2)
 Backstepping Sliding Mode Control Design 283

 
bef 1 ẏf , ẏf ≥ 0 ber2 ẏr , ẏr ≥ 0
Ffd = , Frd = , (3)
bcf 1 ẏf , ẏf < 0 bcr2 ẏr , ẏr < 0
 
Fft = kft2 zf − zof , Frt = krt2 (zr − zor ), (4)

 
Ffb = bfb2 żf − żof , Frb = brb2 (żr − żor ). (5)

where kfs1 , krs1 , knfs1 , knsr1 are the stiffness coefficients of the linear and cubic terms;
bef 1 , ber1 , bcf 1 , bcr1 are the damping coefficient for the extension and compression
movements; kft2 , krt2 , bfb2 , brb2 are the stiffness and damping coefficients of the tires.
yf and yr stand for the front and rear suspension spaces:

yf = z + a sin ϕ − zf , yr = z − b sin ϕ − zr . (6)

Define state variables:

x1 = z, x2 = ż, x3 = ϕ, x4 = ϕ̇, x5 = zf , x6 = żf , x7 = zr , x8 = żr .

3 Synthesis of Backstepping Sliding Mode Controller

To stabilize the vehicle when it encounters a bump on the road, x1 , x3 → 0 as t → ∞.

The system can be controlled by exerting control forces uf , ur on the actuators of the
suspension system. By determining the control signals uz for the heave motion and uϕ
for pitch motion of the system, the real control inputs uf , ur can be calculated. First, the
control signal uz is designed so that tracking error converges to 0. Dynamic equations
for heave motion:

ẋ1 = x2 ,
(7)
ẋ2 = M1 (−Fz + uz ).

with Fz = Ffd + Frd + Ffs + Frs . First of all, let e1 = x1 − x1r where x1r is the desired
trajectory. Therefore, we have ė1 = ẋ1 − ẋ1r = x2 − ẋ1r Select Lyapunov function as
V1 = 21 e12 . Taking the derivative of V1 :

V̇1 = e1 ė1 = e1 (x2 − ẋ1r )

In order to obtain V̇1 ≤ 0, let x2 = s1 − c1 e1 + ẋ1r , i.e.

s1 = x2 + c1 e1 − ẋ1r = c1 e1 + ė1 , c1 > 0 (8)

If s1 = 0, then V̇1 ≤ 0. The following step is conducted. Next, select a Lyapunov

candidate function V2 = V1 + 21 s12 . Because.

1
ṡ1 = ẋ2 + c1 ė1 − ẋ1r = (−Fz + uz ) + c1 ė1 − ẍ1r ,
M
284 V. H. Pham et al.
 
we have V̇2 = V̇1 + s1 ṡ1 = e1 s1 − c1 e12 + s M1 (−Fz + uz ) + c1 ė1 − ẍ1r .
To have V̇2 ≤ 0, a controller is designed as follows:

uz = Fz + M (−c2 s1 − e1 − c1 ė1 + ẍ1r − η1 sgn(s1 )) (9)

where c2 , η > 0. This leads to V̇2 = −c1 e12 − c2 s12 − η1 |s1 | ≤ 0

Therefore, e1 → 0 and e2 → 0 as t → ∞. Similarly, the control signal uϕ for pitch
motion of the system is designed as:

uϕ = Fϕ + I (−c4 s2 − e2 − c3 ė2 + ẍ3r − η2 sgn(s2 )) (10)

   
with Fϕ = a Ffd + Ffs − b(Frd + Frs ) , where c3 , c4 , η2 are positive coefficients.
After obtaining uz and uϕ , we can calculate real control inputs uf , ur as
buz + uϕ auz − uϕ
uf = , ur = . (11)
a+b a+b

4 Zero Dynamics
The backstepping sliding mode controller yields a 4th order error dynamic, while the
original system is 8th order. Therefore, the zero dynamics consists of 4 states. Let
e1 = e3 = 0, we have:

uz = M ẍ1r + Fz
(12)
uϕ = I ẍ3r + Fϕ

Then, solve uf , ur based on the definitions of uz , uϕ and replace ones in ẋ6 , ẋ8 . The
following linearized zero dynamics is obtained:

ẋ = Ax + Bzo + Bref xref (13)

where:
⎡ ⎤ ⎡ ⎤
⎡ ⎤ 0 1 0 0 0 0 0 0
x5 ⎢ kft2 bfb2 ⎥
⎢ x6 ⎥ ⎢ − kft2 − bfb2 ⎥ ⎢m m 0 ⎥
⎥, A = ⎢ ⎥
0 0 0
x=⎢ ⎢ mf mf
⎥ , B = ⎢ f f ⎥,
⎣ x7 ⎦ ⎣ 0 0 0 1 ⎦ ⎢ 0 0 0
⎣ 0 ⎥
⎦
k
x8 0 0 − kmrt2r − bmrb2r 0 0 mfr2r brb2
mr
⎡ ⎤ ⎡ ⎤
zof 0 0
⎢ − bM − I ⎥  
⎢ żof ⎥ ⎢ mf (a+b) mf (a+b) ⎥ ẍ1ref
⎢ ⎥
zo = ⎣ ⎦, Bref = ⎢ ⎥, xref = .
zor ⎣ 0 0 ⎦ ẍ3ref
żor − mr bM I
(a+b) mr (a+b)

Define a positive function V = xT Px, with P > 0 is a positive matrix. Take the
derivative of V:
 
V̇ = ẋPx + xT P ẋ = xT AT P + AP x + 2xT PBzo + 2xT PBref xref
 Backstepping Sliding Mode Control Design 285

It is worth noting that matrix A has eigenvalues with negative real parts. Hence,
AT P + AP = −Q, where Q > 0 is a positive matrix. It is easy to point out that
1 T 1 T
2xT PBzo ≤ x PBBT Px + ηz1 zoT zo , 2xT PBref xref ≤ x PBr BrT Px + ηz2 xref
T
xref
ηz1 ηz2
where η1 , η2 are tuning positive values. We can properly choose matrices P, Q and
tuning values ηz1 , ηz2 so that there exists a positive value ε1 such V̇ ≤ −ε1 V + ε2 ,
where ε2 = η1 zo max + η2 zref max with zoT zo ≤ zo max , xref
T x
ref ≤ xref max . Which means

q
|xk | ≤ λmin (P) , (k = 5, 6, 7, 8).

5 Performance Constraints

We know that all signals are bounded within known ranges, and the bounds of dynamic
tire loads can be estimated as

      q    
Df  = Fft + Ffb  ≤ kft2 + bfb2 + kft2 zof ∞ + bfb2 żof ∞
λmin (P)

q
|Dr | = |Frt + Frb | ≤ (krt2 + brb2 ) + krt2 zot ∞ + brb2 żot ∞
λmin (P)

Furthermore, the bounds of suspension spaces can be calculated as

 
yf  ≤ |x1 | + a|sin x3 | + |x5 | ≤ |x1 | + a|x3 | + |x5 |
  
    q
   
≤ x1ref ∞ + 2V2 (0) + a x3ref ∞ + a 2V4 (0) + (14)
λmin (P)
  
    q
|yr | ≤ x1ref ∞ + 2V2 (0) + bx3ref ∞ + b 2V4 (0) + (15)
λmin (P)

If we choose properly
  the initial values and tuning parameters,  then
 these below can
always be obtained: Df  ≤ Ff , |D
 r | ≤ Fr , |yr | ≤ yr max , yf  ≤ yf max .
Similarly, bounds of |uz |, uϕ  can be estimated in the form of

|uz | ≤ |Fz | + M ((c1 c2 + 1)|e1 | + (c1 + c2 )|ė1 |) ≤ uzb , (16)

   
uϕ  ≤ Fϕ  + I ((c3 c4 + 1)|e2 | + (c3 + c4 )|ė2 |) ≤ uϕb , (17)
 
which leads to the upper bounds of uf , |ur | with

  buzb + uϕb auzb + uϕb

uf  ≤ , |ur | ≤ .
a+b a+b
If the initial values and tuning gains
 c1 , c2 , c3 , c4 are properly adjusted, then the
saturation conditions are satisfied:uf  ≤ uf max , |ur | ≤ ur max .
286 V. H. Pham et al.

6 Simulation Results
In this paper, a special polynomial is designed as a reference trajectory. Because of the
existence of the reference trajectory, the controller can lower the vertical and pitching
acceleration in a predetermined time and improve travelling comfort [12]:

aj0 + aj1 t + aj2 t 2 + aj3 t 3 + aj4 t 4 , t < Tjr
xjr = (18)
0, t ≥ Tjr

where j = 1, 3, coefficient vectors aji , i = 0, 1, 2, 3, 4, j = 1, 3 are determined as:

xjr (0) = aj0 = xj (0),

ẋjr (0) = aj1 = xj+1 (0),
 
xjr Tjr = aj0 + aj1 Tjr + aj2 Tjr2 + aj3 Tjr3 + aj4 Tjr4 = 0,
 
ẋjr Tjr = aj1 + 2aj2 Tjr + 3aj3 Tjr2 + 4aj4 Tjr3 = 0,
 
ẍjr Tjr = 2aj2 + 6aj3 Tjr + 12aj4 Tjr2 = 0.

The road profile of a single road bump event is given by:

⎧   
⎪
⎪ A 2π v L
⎨ 1 − cos t ,0≤t≤
2 L v
zroad =
⎪
⎪ L
⎩ 0, t >
v
where A represents the bump height, L is the bump length, and v is the velocity of the
passing vehicle. The half-car model parameters are:

M = 1200 kg, I = 600 kgm2 , mf = mr = 100 kg, kfs1 = krs1 = 15000 N/m,
knfs1 = knsr2 = 1000 N/m3 , kft2 = 200000 N/m, krt2 = 150000 N/m, bfb2 = 1500 Ns/m,
brb2 = 2000 Ns/m, bef 1 = ber1 = 1500 Ns/m, bcf 1 = bcr2 = 1200 Ns/m, a = 1.2 m, b = 1.5 m.

Initial values are set at x1 (0) = 3 cm, x3 (0) = 3 cm, x5 (0) = 1 cm , x7 (0) = 1 cm.
From Figs. 2 and 3, the passive suspension system exhibits oscillations considerably
much longer than active systems. Table 1 shows the proposed backstepping sliding
mode controller greatly improves both the position and angular response of the system
by smoothing the trajectory and significantly reducing accelerations acting on vehicle
occupants after a pre-determined time, in this case, 1 and 2 s.
At first, the controller exerts forces to stabilize the vehicle. When the car hits the
road bump at the third second, Figs. 4, 5 and 6 show that the controller manages to bring
the vehicle to stability in 1s. Control signals produced by the controller is in the range
of 4000N, which is suitable for applications. The form of control signals in Figs. 5 and
6 are similar due to the fact that both front and rear suspensions are introduced to the
same road condition.
 Backstepping Sliding Mode Control Design 287

Fig. 2. Position responses in different system configurations

Fig. 3. Angular responses in different system configurations

Fig. 4. Control signal amplitude in the case of no reference trajectory

288 V. H. Pham et al.

Fig. 5. Control signal amplitude in the case Tr = 1 s

Fig. 6. Control signal amplitude in the case Tr = 2 s

Table 1. RMS values of accelerations

   
z̈crms m/s2 ϕ̈rms rad/s2

PSS 4.38 3.04

ASS (No Ref) 3.05 3.52
ASS (Tr = 1 s) 0.84 0.83
ASS (Tr = 2 s) 0.84 0.79

7 Conclusion
A backstepping sliding mode controller is utilized to control the active suspension system
in a half-car model along a smooth reference trajectory. The system is mathematically
proved to be stable under certain constraints. The controller reduced RMS values of
heave and pitch accelerations significantly, ensuring ride comfort and improved car
 Backstepping Sliding Mode Control Design 289

handling and ride safety. The control quality of the proposed controller is verified by
simulations in MATLAB/Simulink. In the future, this control algorithm combined with
state observers will be applied for a full-car model under external disturbances.

References
1. Sun, W., Gao, H., Shi, P.: Advanced Control for Vehicle Active Suspension Systems, vol.
204, no. 2. Springer International Publishing, Cham (2020)
2. Chen, S., Cai, Y.-M., Wang, J., Yao, M.: A novel LQG controller of active suspension system
for vehicle roll safety. Int. J. Control Autom. Syst. 16(5), 2203–2213 (2018). https://doi.org/
10.1007/s12555-017-0159-2
3. Hao, R., Wang, H., Liu, S., Yang, M., Tian, Z.: Multi-objective command filtered adaptive
control for nonlinear hydraulic active suspension systems. Nonlinear Dyn. 105(2), 1559–1579
(2021). https://doi.org/10.1007/s11071-021-06559-0
4. Ovalle, L., Ríos, H., Ahmed, H.: Robust control for an active suspension system via continuous
sliding-mode controllers. Eng. Sci. Technol. Int. J. (xxxx) (2021). https://doi.org/10.1016/j.
jestch.2021.06.006
5. Bingül, Ö., Yıldız, A.: Fuzzy logic and proportional integral derivative based multi-objective
optimization of active suspension system of a 4 × 4 in-wheel motor driven electrical vehicle,
2022(0), 1–21 (2022), https://doi.org/10.1177/10775463211062691
6. Wang, Y., Chen, K., Dong, M.: Research on active suspension system of heavy commercial
vehicle controlled by PID controller based on genetic algorithm. Springer Proc. Phys. 270,
555–566 (2022). https://doi.org/10.1007/978-981-16-7258-3_53/TABLES/2
7. Papadimitrakis, M., Alexandridis, A.: Active vehicle suspension control using road preview
model predictive control and radial basis function networks. Appl. Soft Comput. 120, 108646
(2022). https://doi.org/10.1016/J.ASOC.2022.108646
8. Al Aela, A.M., Kenne, J.P., Mintsa, H.A.: Adaptive neural network and nonlinear electro-
hydraulic active suspension control system. JVC/J. Vib. Control 28(3–4), 243–259 (2022).
https://doi.org/10.1177/1077546320975979
9. Dang, V.T., et al.: Adaptive control for multi-shaft with web materials linkage systems.
Inventions 6(4), 76 (2021). https://doi.org/10.3390/inventions6040076
10. Le, D.T., Dang, V.T., Dinh, B.H.N., Vu, H.P., Pham, V.P., Nguyen, T.L.: Disturbance observer-
based speed control of interior permanent magnet synchronous motors for electric vehicles.
In: Inflammation, pp. 244–259 (2022)
11. Le, D.T., Nguyen, D.T., Le, N.D., Nguyen, T.L.: Traction control based on wheel slip tracking
of a quarter-vehicle model with high-gain observers. Int. J. Dyn. Control (2021). https://doi.
org/10.1007/s40435-021-00881-6
12. Pang, H., Zhang, X., Xu, Z.: Adaptive backstepping-based tracking control design for nonlin-
ear active suspension system with parameter uncertainties and safety constraints. ISA Trans.
88, 23–36 (2019). https://doi.org/10.1016/j.isatra.2018.11.047
 Balancing Composite Motion Optimization
and Artificial Neural Network for the Prediction
of Critical Load of Concrete-Filled Steel Tubes
Under Axial Compression

Huan Thanh Duong1 , Tien-Thinh Le2,3(B) , Xuan Son Nguyen2 , Minh Vuong Le4 ,
Hieu Chi Phan5 , Lu Minh Le1 , and Van Hai Nguyen2,3
1 Faculty of Engineering, Vietnam National University of Agriculture, Trau Quy, Gia Lam,
Hanoi 100000, Vietnam
2 Faculty of Mechanical Engineering and Mechatronics, PHENIKAA University, Yen Nghia,
Ha Dong, Hanoi 12116, Vietnam
thinh.letien@phenikaa-uni.edu.vn
3 PHENIKAA Research and Technology Institute (PRATI), A&A Green Phoenix Group JSC,
No. 167 Hoang Ngan, Trung Hoa, Cau Giay, Hanoi 11313, Vietnam
4 Laboratoire Modélisation et Simulation Multi Echelle, Université Paris-Est, MSME UMR
8208 CNRS, 5 bd Descartes, 77454 Marne-la-Vallée, France
5 Faculty of Engineering and Applied Science, Department of Civil Engineering, Memorial

University of Newfoundland, St. John’s, NL 1B 3X5, Canada

Abstract. In this study, the critical capacity of concrete-filled steel tubes (CFST)
under axial compression is investigated through a machine learning model (BCM-
NN) that is a combination of Balancing Composite Motion Optimization (BCM)
and Artificial Neural Network (NN). To develop the model, a database of experi-
mental results is gathered from literature works. Such database includes variables
related to the CFST geometry parameters and its mechanical behavior. Quality
assessments namely coefficient of determination (R2 ) and Root Mean Square
Error (RMSE) are used to evaluate the performance of the proposed model. The
value range of each model parameter has also been studied in this work. The
obtained results show that the hybrid BCM-NN model produced great perfor-
mance in predicting the critical load of CFST compared to conventional neural
network algorithms.

Keywords: Artificial Neural Network · Ultimate load · Balancing composite

motion optimization · CFST column · Prediction

1 Introduction

Due to their many beneficial properties, including high tensile and compressible strength,
the capacity to support significant plastic deformation, dependability, and high strength
safety, steels are now widely used as structural components in a variety of construction

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 290–296, 2023.
https://doi.org/10.1007/978-3-031-22200-9_31
 Balancing Composite Motion Optimization and Artificial Neural Network 291

industries, including civil, industrial, bridge, hydraulic, etc. [1, 2]. However, structural
steel has the drawbacks of being expensive and subject to rust and corrosion [3].
Artificial intelligence (machine learning) techniques have gradually established
themselves as a reliable solution in a number of industries in recent years [4, 5]. One
of the most well-known techniques in this category, particularly for predicting material
behavior, is the artificial neural network (NN) [6]. For instance, various studies on the
prediction of the mechanical behavior of CFST have been conducted on the subject
of the thermomechanical behavior of CFST columns [7], the biaxial bending charac-
teristics of CFST beam-columns, and the ultrasonic behavior of CFST columns. The
outcomes of these have demonstrated the NN technique’s promise because it performs
well in situations like these.
The recently released work by Sarir et al. [8], in which the authors presented two
NN-based hybrid metaheuristic models optimized by particle swarm optimization (PSO)
and whale optimization algorithm (WOA), respectively, provides a detailed and in-depth
state-of-the-art on the behavior and modeling of concrete-filled steel tube columns. The
effectiveness of the WOA as a powerful optimization method in the proposed NN-WOA
metaheuristic model’s ability to estimate the bearing capacity of concrete-filled steel
tube columns was proven by comparisons with existing metaheuristic computational
models.
In order to enhance the performance of traditional neural networks for the prediction
of the ultimate load of rectangular CFST members, this study offers a hybrid machine
learning method employing the Balancing Composite Motion Optimization (BCM) tech-
nique. Out of these, the BCM technique was employed to help the NN model’s train-
ing phase optimize the weights and biases in order to create a novel hybrid BCM-NN
model for more accurate prediction of the ultimate load of rectangular CFST members.
The geometrical characteristics of the structural members as well as the mechanical
properties of steel and concrete are among the variables in the database. Such a large
amount of data made it possible to get important results when training and validating the
models. Evaluation and comparison of the hybrid BCM-NN model with traditional gra-
dient descent methods were conducted including Levenberg-Marquardt (LM), One-step
secant (OSS), Scaled conjugate gradient (SCG), Gradient descent algorithm (GDA), and
Conjugate gradient algorithm (CGB).

2 Methods and Materials

2.1 Balancing Composite Motion Optimization
Le-Duc et al. [9] have developed the unique population-based optimization method
known as balancing composite motion optimization (BCM). The approach is character-
ized as straightforward and devoid of any human or natural behaviors. The fundamental
tenet of BCM is that population search motions are equalized in a composite way. In
other words, a candidate solution can travel in both the presumed Cartesian global space
and the Cartesian local space. While the other movement in the global space is helpful
in sufficiently examining the search space, the movement in the local space enables
the candidate to determine whether there are any superior answers in the local area. A
mathematical, probabilistic methodology for making sure the equalization is in control
292 H. T. Duong et al.

of these motions is the idea. Because the motions of a candidate solution are equalized
throughout both exploration and exploitation, this idea enables attaining a self-balance of
the approach during the optimization process. The effectiveness of BCM was evaluated
using a number of traditional benchmark functions, including unimodal, multimodal,
and noisy quartic [10]. The flowchart of different steps of BCM is described in Fig. 1.

Fig. 1. BCM’s flowchart.

2.2 Artificial Neural Network

Artificial neural network (NN) is a directed network of a system of nodes, which are
organized into layers, connected by weighted links [11]. The data is fed at the input
layers, of which the number of nodes equals to the features of each sample. The one
or more hidden layer(s) is consequently placed behind the input layers and their nodes
connected to the previous layer by directed-toward links. Each node transfers signal
to nodes in the next layer by directed-outward links. After the process of receiving
information from other nodes, the received information is combined and the signal of
this node is a result of the activation function. At the end of the network is the output
layer which provides the prediction of the current network. A bias node is consistently
added at all layers except for the input layer. The purpose of this addition is to increase
the flexibility of the overall signal received by nodes in the next layer (Fig. 2).
 Balancing Composite Motion Optimization and Artificial Neural Network 293

Fig. 2. Architecture of NN.

2.3 Monte Carlo Technique

Monte Carlo technique has been widely employed to consider the randomness of the
input space in a given problem [12], especially in civil and mechanical engineering. The
principle of this method is to simulate the results by randomly varying each input variable
in the database to investigate the influence of them on the output response. By this logic,
this technique will require a great number of simulations to be carried out. Therefore,
we have applied a parallel computing scheme in this work to reduce the computational
cost.

3 Results and Discussion

3.1 Convergence of Random Sampling

Using training and testing datasets, Fig. 3 illustrates the convergence of random sam-
plings in terms of R2 and RMSE. All designed configurations of the parametric investiga-
tion (i.e., in adjusting the number of neurons in the hidden layer as well as the population
size of the BCM) are represented by various curves in Fig. 3. It is demonstrated that when
the number of Monte Carlo simulations is relatively small, a large fluctuation was seen in
both the training and testing data situations (i.e., smaller than 40). This finding showed
that the input space is variable, and the Monte Carlo approach was helpful in taking this
variation into account. After 80 random samplings, the statistical convergence is finally
seen with respect to all criteria. As a result, statistical analysis used in post-processing
should be trustworthy.

3.2 Optimization of NN’S Weight Parameters by BCM

In this work, the BCM algorithm’s topping condition is a set number of 5000 iterations.
The statistical values of the quality assessment criteria in relation to the number of
iterations are shown in Fig. 4. This chart demonstrates that 5000 iterations were sufficient
to achieve convergence for all criteria.
294 H. T. Duong et al.

Fig. 3. Estimation of convergence using training data for (a) R2 , (c) RMSE; using testing data
for (b) R2 , (d) RMSE.

Fig. 4. Evolution of (a) R2 , (b) RMSE for both training and testing data during the training phase
(based on iteration).

3.3 Optimal Architecture and Population Size

Figure 5 presents the surface graph of average values of R2 and RMSE for training and
testing data, over 100 random samplings. In this figure, the optimal zone (i.e., highest
performance) was also highlighted. It is seen that the capacity of the NN model highly
depends on its architecture for a given BCM’s population size. This is the same remark,
as observed in the literature [8]. However, as the population size grows, the NN model’s
capability is not necessarily better. Based on the findings in Fig. 5, it can be concluded
that 260 is the right population size for the BCM and that 5 is the ideal number of neurons
for the hidden layer. Finally, the hybrid NN-BCM model’s final structure is determined.

3.4 Prediction Capability of BCM-NN and Comparison with Other Gradient

Descent Techniques

It is shown how BCM performs in comparison to other learning methods including LM,
OSS, SCG, CGB, and GDA. The values of R2 and RMSE are given as quality assessment
indicators for comparative purposes, as can be seen in Fig. 6. Figure 6a and b rank the
 Balancing Composite Motion Optimization and Artificial Neural Network 295

Fig. 5. Surface graphs in function of BCM’s population size and NN’s number of neurons for (a)
R2 using training data, (b) R2 using testing data, (c) RMSE using training data, (d) RMSE using
testing data.

effectiveness of leaning approaches according to R2 and RMSE, respectively, for further

analysis. The BCM technique demonstrated the maximum efficiency and performance
to train the NN model for forecasting the final load of the CFST, according to overall
statistical results.

Fig. 6. Graphs comparing prediction accuracy in relation to (a) R2 , (b) RMSE.

296 H. T. Duong et al.

4 Conclusion
This work’s major objective was to accurately forecast the ultimate load of rectangular
steel tubes filled with concrete by using a unique surrogate model. In order to achieve
better prediction performance compared to current predictive models, a new NN-based
hybrid model optimized by balancing composite motion has been created. The most
trustworthy estimation in terms of performance indicators is provided by the suggested
BCM-NN model.

Acknowledgements. This research is funded by Phenikaa University under grant number

PU2022–1-A-01.

References
1. Ali, F., Nadjai, A., Goodfellow, N.: Experimental and numerical study on the performance
of hollow and concrete-filled elliptical steel columns subjected to severe fire. Fire Mater. 40,
635–652 (2016). https://doi.org/10.1002/fam.2316
2. Le, T.-T.: Practical machine learning-based prediction model for axial capacity of square
CFST columns. Mech. Adv. Mater. Struct. 29, 1–16 (2020). https://doi.org/10.1080/15376494.
2020.1839608
3. Ho, N.X., Le, T.-T.: Effects of variability in experimental database on machine-learning-based
prediction of ultimate load of circular concrete-filled steel tubes. Measurement 176, 109198
(2021). https://doi.org/10.1016/j.measurement.2021.109198
4. Ahmadi, M., Naderpour, H., Kheyroddin, A.: ANN model for predicting the compressive
strength of circular steel-confined concrete. Int. J. Civ. Eng. 15, 213–221 (2017)
5. Duong, T.H., Le, T.-T., Nguyen, S.X., Le, M.V.: Adaptive-neuro-fuzzy-inference-system
model for prediction of ultimate load of rectangular concrete-filled steel tubular columns.
J. Intell. Fuzzy Syst. Preprint, 1–19 (2022). https://doi.org/10.3233/JIFS-201628
6. Sankar Jegadesh, J.S., Jayalekshmi, S.: Application of Artificial Neural Network for
calculation of axial capacity of circular concrete filled steel tubular columns
7. Al-Khaleefi, A.M., Terro, M.J., Alex, A.P., Wang, Y.: Prediction of fire resistance of concrete
filled tubular steel columns using neural networks. Fire Saf. J. 37, 339–352 (2002)
8. Sarir, P., Chen, J., Asteris, P.G., Armaghani, D.J., Tahir, M.M.: Developing GEP tree-based,
neuro-swarm, and whale optimization models for evaluation of bearing capacity of concrete-
filled steel tube columns. Eng. Comput. 37(1), 1–19 (2019). https://doi.org/10.1007/s00366-
019-00808-y
9. Le-Duc, T., Nguyen, Q.-H., Nguyen-Xuan, H.: Balancing composite motion optimization.
Inf. Sci. 520, 250–270 (2020). https://doi.org/10.1016/j.ins.2020.02.013
10. Yao, X., Liu, Y., Lin, G.: Evolutionary programming made faster. IEEE Trans. Evol. Comput.
3, 82–102 (1999). https://doi.org/10.1109/4235.771163
11. Ho, N.X., Le, T.-T., Le, M.V.: Development of artificial intelligence based model for the
prediction of Young’s modulus of polymer/carbon-nanotubes composites. Mech. Adv. Mater.
Struct. 1–14 (2021, in press). https://doi.org/10.1080/15376494.2021.1969709
12. Le, T.-T.: Probabilistic investigation of the effect of stochastic imperfect interfaces in
nanocomposites. Mech. Mater. 151, 103608 (2020). https://doi.org/10.1016/j.mechmat.2020.
103608
 Building Power Demand Forecasting Using
Machine Learning: Application for an Office
Building in Danang

Hoang-Anh Dang(B) and Van-Dung Dao

School of Electrical and Electronic Engineering, Hanoi University of Science and Technology,
Hanoi, Vietnam
anh.danghoang@hust.edu.vn

Abstract. Toward 2030, 20% of existing buildings and all new construction would
have to be zero-carbon ready for reaching the global goal of carbon neutrality by
2050. As a member state of UNFCCC, Vietnam has come up with mechanisms
and policies to ensure compliance with strong commitments at COP26. However,
energy efficiency design strategies for new construction would not guaranty the
energy reduction if buildings were not appropriately operated. To contribute the
objective of energy savings and lowering the building’s operational costs, this
paper presents a Machine Learning approach for an office building, located in
Danang, to forecast the power demand based on time, weather conditions and
historical energy consumption. For discovering energy consumption pattern and
correlation between features, exploratory data analysis has been performed to ini-
tially investigate data sets and summarize their main characteristics. After selecting
features that have strong relationship with the target, the power demand of this
office building has been predicted using different Machine Learning methods.
Forecasting results will be compared to define the relevant method.

Keywords: Building energy efficiency · Power demand forecasting · Machine

learning

1 Introduction

Over the past decade, global CO2 emissions from building operation have increased by an
average of 1% per year and plummeted up to 9 Gt in 2020. In Southeast Asia, building
sector is responsible for nearly a quarter of total final energy consumption and CO2
emissions due to urbanization and population growth across the region [1]. According
to the International Energy Agency (IEA), buildings are still falling short of carbon
neutrality by 2050. To reach this goal, 20% of existing buildings and all new construction
would have to be zero-carbon ready by 2030 [2]. In line with strong commitments,
Vietnam has come up with mechanisms and policies to ensure compliance with the
National Technical Regulation on Energy Efficiency Buildings (QCVN 09:2017/BXD)
and at least 80 Green Building projects by 2025 [3].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 297–304, 2023.
https://doi.org/10.1007/978-3-031-22200-9_32
298 H.-A. Dang and V.-D. Dao

Despite applying passive design strategies, energy-saving materials and modern

Mechanical, Electrical, and Plumbing (MEP) system for new building design, the
building operation could not be considered as high efficiency and reliability. Indeed,
numerous new buildings use more energy than expected consumption and have some
system faults after being placed in service. Building energy demands have always
been difficult for facility managers to handle effectively. For archiving energy savings
and lowering the building’s operational costs, engineers need to anticipate the energy
demand profile as well as the building’s energy consumption. The traditional approach
is to exploit building energy model in design phase for calculating building energy
demand in operation, however, detail of input data, model calibration and performance
computing are major challenges [4]. Another approach is to apply Machine Learning
(ML) to forecast building energy demand without requiring much experience of building
performance calculation [5].
The main objective of this study is to forecast building power demand based on
time, weather conditions and historical energy consumption, using Machine Learning
methods. These data sets have been respectfully extracted from a weather station of
international airport (weather data) and an energy monitoring system of an office building
(energy consumption data), located in Danang. For discovering energy consumption
pattern and correlation between features, Exploratory Data Analysis (EDA) has been
performed to initially investigate data sets and summarize their main characteristics.
After selecting features that have strong relationship with the target, the power demand
of this office building has been predicted using different Machine Learning methods.
Forecasting results will be compared to define the appropriate method for this research.

2 Case-Study Description
The FPT-Complex is one of the first buildings designed for the goal of saving energy as
well as reducing greenhouse effect compared to other current buildings of Danang city.
In the framework of the USAID Vietnam Urban Energy Security project, this building
was selected as the demonstration object for deploying a modern energy management
system and implementing energy efficiency measures.
From its historical operation data and weather data of Danang International Airport
(Fig. 1), the power demand will be assessed for the purpose of Validation, Verification and
Monitoring. In this paper, the dataset, which includes time-series data (from January 1 to
May 31, 2022) of building power demand (kW), outdoor air temperature (°C), outdoor
relative humidity (%) and wind speed (m/s) with timestep of 30 min, has been used for
predicting the baseline of energy consumption.

3 Exploratory Data Analysis

3.1 Building Energy Consumption Pattern
The graph in Fig. 1 shows that the FPT-Complex building uses much lower demand
during the holiday than any other time of the year. However, because of providing
cloud server services, the base of power consumption is always over 250 kW. In 2021,
 Building Power Demand Forecasting Using Machine Learning 299

Covid-19
May Day Lockdown May Day
New Year’s Day New Year’s Day Tet Holiday

Tet Holiday

Fig. 1. Energy demand and outdoor air temperature of FPT-complex building from January 1 to
May 31, 2022

authorities in Danang implemented strict lockdown from August 15 to September 16,

due to COVID-19. In this period, air conditioning and lighting demand had been reduced
(lower power in daytime), but remote computer demand had slightly increased (higher
power in nighttime and on weekend). Overall, the consumption profile exhibits annual
seasonality with high power demand in summer.

Fig. 2. Power demand distribution by month, hour of week and weekday

300 H.-A. Dang and V.-D. Dao

Figure 2 gives more insights about the dataset through power demand distribution
by month, hour of week and weekday. Although the power demand seems to be higher
in June and July, the annual seasonality shows the highest median value in August and
higher peaks in May. The building operator should pay attention to possible overload
faults. On the graph of weekly distribution, there is obviously a weekly seasonality with
higher consumption during weekdays and lower consumption on weekend. Moreover,
there is also a daily seasonality with demand increasing during working hours (from
8:00 to 16:00) and a clear correlation between power demand and time.

3.2 Correlation Between Power Demand and Weather Conditions

The correlation between variables is commonly measured through the Pearson’s correla-
tion coefficient and the p-value [6]. This coefficient is expressed as a number between −1
and 1, with 1 being the strongest positive correlation and −1 representing the strongest
negative correlation. The p-value is the probability for rejecting the null hypothesis.
Indeed, the correlation coefficient is said to be statistically significant if the probability
is less than the conventional 5% (p-value < 0.05).

Table 1. Correlation between the power demand and the weather conditions

Outdoor air Outdoor relative Wind speed (m/s)

temperature (°C) humidity (%)
Power demand Pearson’s 0.57067 −0.31896 0.00187
(kW) coefficient
p-value 6.43628e-46 1.14957e-13 0.96614

The Table 1 shows that outdoor air temperature and outdoor relative humidity
are associated with building power demand (high correlation coefficient and very
low p-value). Meanwhile, wind speed should be excluded in forecasting process (low
correlation coefficient and high p-value) (Fig. 3).

Fig. 3. Relationship between power demand and weather conditions (outdoor temperature and
outdoor humidity)
 Building Power Demand Forecasting Using Machine Learning 301

4 Building Energy Demand Forecasting

4.1 Machine Learning for Time-Series forecasting

Forecasting is the technique of estimating the future value of a time series by modeling
its pattern with or without external variables. To forecast using Machine Learning, the
time-series must be transformed into a matrix in which each value is associated to the
time window that precedes it in order. In this paper, the Skforecast library [7], which
contains the classes and functions of Scikit-learn regression model [8], has been exploited
to perform forecasting on time series of building power demand. There are two main
forecasting strategies implemented in Skforecast: Recursive autoregressive and Direct
multi-step (Fig. 4).

Fig. 4. Diagram of recursive autoregressive (a) and Direct multi-step (b) forecasting [7]

To serve the process of hyperparameter tuning and 24 h-prediction assessment, the

time-series data of FPT-Complex building are divided into 3 different datasets: training,
validation, and test. Primarily, the training and the validation datasets have been used to
identify the optimal parameters of regression model through a Grid Search with valida-
tion by Back Testing. This initial stage consists of training with parametric process and
evaluating prediction ability. After archiving the relevant model, the predicted demand
will be compared with real demand of the test period for verifying the model accuracy.
In this paper, 6 Machine Learning methods, which are widely used for regression, have
been tried: Ridge regression (Ridge), eXtreme Gradient Boosting (XGBoost), Forests
of randomized trees (RandomForest), Linear Support Vector Regression (LinearSVR),
K-Nearest Neighbors (K-NN) and Multi-Layer Perceptron Regression (MLPRegres-
sor). The Mean Absolute Percentage Error (MAPE) is used as a measure to assess the
prediction accuracy between different strategies and methods (Fig. 5).

4.2 Forecasting Result

Recursive Autoregressive Forecasting. The value tn−1 (unknown value) is required to

predict the value tn , therefore, each new prediction must be recursively based on the
previous one (Fig. 4a). thanks to one model trained to predict the next step value, this
strategy has low computational cost for model training and forecasting (Table 2).
302 H.-A. Dang and V.-D. Dao

Fig. 5. Power demand data split into training, validation, and test datasets

Table 2. Forecasting MAPE of different Machine Learning methods using recursive autoregres-
sive forecasting strategy

MAPE Ridge XGBoost Random Forest k-NN Linear SVR MLP regressor
Validation 10.50% 6.70% 8.60% 11.40% 34.59% 10.87%
Test 9.56% 8.03% 10.28% 11.49% 44.85% 11.23%

Because of high MAPE and resulting negative value of power demand, the Lin-
earSVR is not qualified. With slightly different between MAPE of validation and test,
the other methods seem to be good for forecasting the power demand of FPT-Complex
building. The accuracy of XGBoost (MAPE of 8.03%) is the highest for recursive autore-
gressive strategy and its parallel computing is worthy to accelerate learning and fore-
casting process. For fast computing, low resource requirements and good accuracy, the
Ridge (MAPE of 9.56%) should be considered. Although the accuracy of MLPRegressor
is not the best, the representative of neural network model shows its ability to predict
the peak of power demand (Fig. 6).
Direct Multi-step Forecasting. This strategy generates the training matrices for a dif-
ferent model and the prediction of each step is independent (Fig. 4b). thus, the higher
performance computing is necessary for the train of multiple models (Table 3).
In this scenario, the LinearSVR (MAPE of 11.37%) provides far better results than
using Recursive autoregressive forecasting strategy. However, due to a slightly increase-
ment of MAPE, some methods (Ridge, XGBoost and k-NN) are not recommended to
follow Direct multi-step forecasting strategy. Meanwhile, the MLPRegressor becomes
the best candidate for forecasting the power demand of FPT-Complex building with high
accuracy (MAPE of 7.64%) and impressive ability of peak prediction (Fig. 7).
 Building Power Demand Forecasting Using Machine Learning 303

Fig. 6. Comparison between real power demand and predicted results of different Machine
Learning methods for test period, using Recursive autoregressive forecasting

Table 3. Forecasting MAPE of different Machine Learning methods using Direct multi-step
forecasting strategy

MAPE Ridge XGBoost Random Forest k-NN Linear SVR MLP regressor
Validation 10.22% 7.67% 7.92 11.55% 12.12% 10.13%
Test 9.72% 10.08% 9.68 11.78% 11.37% 7.64%

Fig. 7. Comparison between real power demand and predicted results of different Machine
Learning methods for test period, using Direct multi-step forecasting
304 H.-A. Dang and V.-D. Dao

5 Conclusion
In this paper, the power demand forecasting of an office building, located in Danang,
has been based on time, weather conditions and historical energy consumption, using
Machine Learning methods. By performing the EDA technique, several features, which
are strongly related to the power demand, have been selected such as time lag, day of
week, day of month, holiday events, Covid-19 lockdown events, outdoor air temperature
and outdoor relative humidity. To archive an appropriate approach, the case-study has
been tested with Ridge regression, eXtreme Gradient Boosting, Forests of randomized
trees, Linear Support Vector Regression, K-Nearest Neighbors and Multi-Layer Percep-
tron Regression. Following the recursive autoregressive forecasting strategy, the eXtreme
Gradient Boosting is the relevant method for power demand forecasting with good accu-
racy and accelerated computing features. On the other hand, the Multi-layer Perceptron
Regression applied direct multi-step forecasting strategy is the best method with the
highest accuracy and impressive ability of peak prediction. However, the total power
demand forecasting is not enough to archive energy savings and lowering the building’s
operational costs. In further research, after receiving enough data from energy monitor-
ing system, the breakdown of power demand forecasting will be processed to provide
better insight analysis for more details of energy efficiency strategies.

References
1. IEA: Roadmap for Energy-Efficient Buildings and Construction in the Association of South-
east Asian Nations. IEA, Paris https://www.iea.org/reports/roadmap-for-energy-efficient-bui
ldings-and-construction-in-the-association-of-southeast-asian-nations. Accessed 7 June 2022
2. IEA: Tracking Buildings 2021. IEA, Paris. https://www.iea.org/reports/tracking-buildings-
2021. Accessed 7 June 2022
3. Prime Minister of Vietnam: National Program on Energy Efficiency 2019–2030 period.
Decision 280/QÐ-TTg (2019)
4. Le Gia, T.T., et al.: A simulation-based multi-objective genetic optimization framework for
efficient building design in early stages: application for Vietnam’s hot and humid climates. Int.
J. Build. Pathol. Adapt. 40(3), 305–326 (2022)
5. Wu, L., Kaiser, G., Solomon, D., Winter, R., Boulanger, A., Anderson, R.: Improving efficiency
and reliability of building systems using machine learning and automated online evaluation. In:
2012 IEEE Long Island Systems, Applications and Technology Conference (LISAT), pp. 1–6.
IEEE, NY, USA (2012)
6. The Data School, Fundamental of Analysis: Correlation and P value. https://dataschool.com/
fundamentals-of-analysis/correlation-and-p-value/. Accessed 16 June 2022
7. Skforecast: time series forecasting with Python and Scikit-learn. https://www.cienciadedatos.
net/py27-forecasting-series-temporales-python-scikitlearn.html. Accessed 18 June 2022
8. Pedregosa, et al.: Scikit-learn: machine learning in Python. JMLR 12, 2825–2830 (2011)
 Calculating Optimum Input Factors
for Maximum Material Removal Rate
in Wire-EDM SKD11 Tool Steel

Nguyen Hong Linh1 , Tran Huu Danh2 , Bui Thanh Danh3 , Tran Minh Tam4 ,
Nguyen Anh Tuan5 , Trinh Kieu Tuan5 , Nguyen Manh Cuong6 ,
and Nguyen Thi Thanh Nga6(B)
1 Electric Power University, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 University of Economics - Technology for Industries, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nguyennga@tnut.edu.vn

Abstract. This paper introduces a study on calculating optimum Wire Electrical

Discharge Machining (Wire-EDM) factors when cutting arc of SKD11 tool steel
to get the maximum Material Removal Rate (MRR). Seven input parameters were
examined in this work: the pulse on time, the pulse off time, the cutting voltage,
the serve voltage, the cutting speed, the wire feed, and the arc radius. The effect of
these factors on the MRR of the Wire-EDM process was investigated. Furthermore,
optimal input factors for achieving the highest MRR were suggested. It was also
reported that the optimal input parameters were appropriate for use.

Keywords: WEDM · Wire-EDM · Material Removal Rate · SKD11

1 Introduction
Wire-EDM or WEDM is a type of EDM which uses a thin wire as the electrode. This
wire is wound continuously running along a predetermined path to process conductive
materials. This type of machining is very widely used to process cylindrical holes with
any shapes, or narrow slits with arbitrary arcs. Consequently, research on optimizing
Wire-EDM process has attracted the attention of many scientists.
So far, there have been quite a few studies on Wire-EDM. Research often focuses
on optimizing input parameters to improve the machining productivity [1, 2], reduce the
surface roughness [3–5], or control the wire tension [6, 7], etc. Besides, many studies
have been conducted for various materials such as hot die steel [8], aluminum [9],
mild steel [9], Ni 55.8 Ti [10], 90CrSi [5], Inconel 625 [11], titanium [12], Ti-6Al-4V
[13], etc. The mechanism of how electromagnetic force applied to the wire electrode in

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 305–313, 2023.
https://doi.org/10.1007/978-3-031-22200-9_33
306 N. H. Linh et al.

Wire-EDM was learned in [14]. The real-time wire tension control to improve the wire
displacement accuracy was investigated in [15]. The effect of input parameters on the
formation and characteristics of recast layer of workpiece material was investigated in
[16]. In addition, the research on Wire-EDM is also carried out using various methods
such as the Taguchi method [8–11, 13, 17], the Taguchi method and Genetic algorithm
[8, 18], or Grey Relational Analysis [12, 13, 17].
This paper deals with a study on the effect of input factors on optimal Wire-EDM
process parameters when cutting circular arcs SKD11 steel. To do that, seven input
parameters were considered. The impact of the input factors on the MRR were discov-
ered. Also, a proposed model to determine the optimum Wire-EDM parameters was
given.

2 Experimental Work

To evaluate the effect of input process factors on the MRR, an experiment was imple-
mented. The workpiece used in the experiment was SKD11 and the machine used in this
work was shown in Fig. 1. In this experiment, 7 main process factors were investigated.
These parameters consist of the cutting voltage (VM), the pulse on time (Ton ), the pulse
off time (Toff ), the servo voltage (SV), the wire feed (WF), the arc radious (R), and the
cutting speed (SPD) (Table 1). In addition, the experimental plan with the design L18
(21 × 35 ) is created by using the Minitab R18 software and the Taguchi method. The
experimental plan and the response (the MRR) are listed in Table 2.

Fig. 1. WEDM machine used in the experiment

 Calculating Optimum Input Factors for Maximum Material Removal Rate 307

Table 1. Input parameters and their levels

No Parameter Code Unit Level

1 2 3
1 Cutting voltage VM V 6 9 12
2 Pulse on time Ton μs 6 9 12
3 Pulse off time Toff μs 8 12 16
4 Servo voltage SV V 24 29 34
5 Wire feed WF mm/min 8 10 12
6 Arc radious R mm 3 6 9
7 Cutting speed SPD mm/min 2 4 –

Table 2. Experimental plan and the MRR

No VM Ton Toff SV WF SPD R MRR (mm3 /min) S/N

1 3 6 8 24 8 2 3 1.560 3.8625
2 6 6 12 29 10 2 6 3.176 10.0376
3 9 6 16 34 12 2 9 8.180 18.2551
4 6 9 8 24 10 2 9 8.460 18.4835
5 9 9 12 29 12 2 3 10.793 20.6628
6 3 9 16 34 8 2 6 6.754 16.4473
7 9 12 8 29 8 2 6 13.298 22.4757
8 3 12 12 34 10 2 9 9.879 19.8059
9 6 12 16 24 12 2 3 8.783 18.4285
10 6 6 8 34 12 4 6 4.293 12.6552
11 9 6 12 24 8 4 9 3.777 11.5429
12 3 6 16 29 10 4 3 3.878 11.4829
13 3 9 8 29 12 4 9 6.298 16.0951
14 6 9 12 34 8 4 3 8.752 18.8421
15 9 9 16 24 10 4 6 12.115 21.6665
16 9 12 8 34 10 4 3 12.437 21.8943
17 3 12 12 24 12 4 6 14.142 23.0653
18 6 12 16 29 8 4 9 12.722 22.0281
308 N. H. Linh et al.

3 Results and Analysis

To evaluate the influence of input factors on the MRR, the ANOVA analysis method in
Minitab 19 software is used. Also, the objective function is the maximum MRR being
the best:
 
1 n 1
S/N = −10log 10 (1)
n i=1 y 2
i

Table 3 shows the results of ANOVA analysis. It is found from the results that
Ton has the greatest influence on MRR (68.55%) and after that is VM (12.00%). The
remaining parameters have a low influence on MRR. They are WF (3.49%), Toff (3.03%),
R (2.92%), SV (2.09%) and SPD (1.38%).

Table 3. Analysis of variance of input parameters on S/N

Analysis of variance for SN ratios

Source DF Seq SS Adj SS Adj MS F P C (%)
SPD 1 6.496 6.496 6.496 0.85 0.410 1.38
Ton 2 321.754 321.754 160.877 20.97 0.008 68.55
Toff 2 14.219 14.219 7.109 0.93 0.467 3.03
SV 2 9.822 9.822 4.911 0.64 0.574 2.09
WF 2 16.405 16.405 8.203 1.07 0.425 3.49
VM 2 56.310 56.310 28.155 3.67 0.124 12.00
R 2 13.706 13.706 6.853 0.89 0.478 2.92
Residual error 4 30.692 30.692 7.673 6.54
Total 17 469.404
Model summary
S R-Sq R-Sq (adj)
2.7700 93.46% 72.21%

Table 4 shows the order of influence of parameters on S/N: Ton , VM, WF, Toff , R,
SV và SPD. Figure 2 shows the evaluation of the influence of the two parameters with
the greatest influence on MRR: Ton and VM. From the Figure, it is noted that with Ton
value greater than 10 (μs) and VM greater than 8.5 (V) (corresponding to the dark green
area), the MRR value is the largest (estimated greater than 14 mm3 /min). In addition,
to get the maximum material removal rate, as described in Eq. 1, the S/N value of each
input parameter must be the maximum. Thus, based on the analysis of the influence of
the parameters on the S/N ratio in the chart in Fig. 3, it is possible to determine the
optimal set of Wire-EDM parameters to achieve the largest MRR. Therefore, the values
of optimum parameters are presented in Table 5.
 Calculating Optimum Input Factors for Maximum Material Removal Rate 309

Table 4. Order of influence of input parameters on S/N

Larger is better
Level SPD Ton Toff SV WF VM R
1 16.50 11.31 15.91 16.17 15.87 15.13 15.86
2 17.70 18.70 17.33 17.13 17.23 16.75 17.72
3 21.28 18.05 17.98 18.19 19.42 17.70
Delta 1.20 9.98 2.14 1.81 2.33 4.29 1.86
Rank 7 1 4 6 3 2 5

To predict the optimum MRR corresponding to this set of optimum parameters,

the analysis tool of Minitab 19 software was used and the results are shown in
Table 6. Thus, with the set of optimal parameters, the corresponding maximum MMR
is 21.2143 mm3 /min.

Fig. 2. Contour plot of MRR versus Ton and VM

Evaluation of experimental models

The appropriateness of the experimental model was evaluated by the Anderson-Darling
method (Fig. 4). From Fig. 4, it can be learned that the experimental points (the blue
points) on the histogram lying very close to the normal distribution line (the red solid
line). That means the deviation is very small.
The graph about the frequency of errors shows that most of the errors are in the range
−1 to 0. In the other two graphs, the experimental errors are randomly distributed. It
means that the built model is largely influenced by the selected input parameters and is
not affected by the order of experiments. Besides, the chart of reliability also shows that
the data of the experimental points (the blue dots) are all within the bounded area by 2
upper and lower limit lines with 95% significance level. Also, the P value of 0.515 is
310 N. H. Linh et al.

Fig. 3. Effect of input factors on S/N ratio

Table 5. Optimum wire-EDM parameters

No Factor Code Unit Level Value

1 Cutting voltage VM V 3 9
2 Pulse on time Ton μs 3 12
3 Pulse off time Toff μs 3 16
4 Servo voltage SV V 3 34
5 Wire feed WF mm/min 3 12
6 Arc Radius R mm 2 6
7 Cutting speed SPD mm/min 2 4

Table 6. Prediction of MMR

Prediction
S/N ratio Mean
31.9672 21.2143
Settings
SPD Ton Toff SV WF VM R
4 12 16 34 12 9 6

greater than the value of α = 0.05, which indicates that the applied experimental model
is suitable (Fig. 5).
 Calculating Optimum Input Factors for Maximum Material Removal Rate 311

Fig. 4. Residual plots for S/N ratios

Fig. 5. Probability plot of MRR

4 Conclusions
The present work introduces the determination of optimum Wire-EDM factors when
arc cutting SKD11 steel. In this work, seven input process parameters including the
cutting voltage, the pulse on time, the pulse off time, the serve voltage, the wire feed, the
cutting speed, the arc radius were investigated. The impacts of the input factors on the
312 N. H. Linh et al.

MRR were explored. It is noted that Ton has the greatest influence on MRR (68.55%)
and after that is VM (12.00%). The other input factors have a lower influence on MRR:
WF (3.49%), Toff (3.03%), R (2.92%), SV (2.09%) and SPD (1.38%). In addtion, the
following optimum input factors was suggested: VM = 12; Ton = 12; Toff = 8; SV =
24; WF = 12; and SPD = 4. It is also found that the proposed model is proper to use in
practice.

Acknowledgments. This work was supported by Thai Nguyen University of Technology.

References
1. Maher, I., et al.: Increasing the productivity of the wire-cut electrical discharge machine
associated with sustainable production. J. Clean. Prod. 108, 247–255 (2015)
2. Klocke, F., et al.: Model-based productivity analysis of wire EDM for the manufacturing of
titanium. Procedia CIRP 77, 594–597 (2018)
3. Spedding, T.A., Wang, Z.: Parametric optimization and surface characterization of wire
electrical discharge machining process. Precis. Eng. 20(1), 5–15 (1997)
4. Bobbili, R., Madhu, V., Gogia, A.: Effect of wire-EDM machining parameters on surface
roughness and material removal rate of high strength armor steel. Mater. Manuf. Processes
28(4), 364–368 (2013)
5. Thi Hong, T., et al.: Effects of process parameters on surface roughness in wire-cut EDM of
9CRSI tool steel. Int. J. Mech. Eng. Technol. 10(3) (2019)
6. Wentai, S., Zhidong, L., Mingbong, Q., Zongjun, T.: Wire tension in high-speed wire electrical
discharge machining. Int. J. Adv. Manuf. Technol. 82(1–4), 379–389 (2015). https://doi.org/
10.1007/s00170-015-7352-3
7. Habib, S., Okada, A.: Study on the movement of wire electrode during fine wire electrical
discharge machining process. J. Mater. Process. Technol. 227, 147–152 (2016)
8. Kuruvila, N.: Parametric influence and optimization of wire EDM of hot die steel. Mach. Sci.
Technol. 15(1), 47–75 (2011)
9. Tilekar, S., Das, S.S., Patowari, P.: Process parameter optimization of wire EDM on aluminum
and mild steel by using Taguchi method. Procedia Mater. Sci. 5, 2577–2584 (2014)
10. Magabe, R., et al.: Modeling and optimization of Wire-EDM parameters for machining of
Ni 55.8 Ti shape memory alloy using hybrid approach of Taguchi and NSGA-II. Int. J. Adv.
Manuf. Technol. 102(5), 1703–1717 (2019)
11. Subrahmanyam, M., Nancharaiah, T.: Optimization of process parameters in wire-cut EDM
of Inconel 625 using Taguchi’s approach. Mater. Today: Proc. 23, 642–646 (2020)
12. Sahoo, S.K., et al.: Analysis and optimization of wire EDM process of titanium by using
GRA methodology. In: Materials Science Forum. Trans Tech Publication (2019)
13. Kumar, S., Dhanabalan, S., Narayanan, C.: Application of ANFIS for the selection of optimal
wire-EDM parameters while machining Ti-6Al-4V alloy and multi-parametric optimization
using GRA method. Int. J. Decis. Supp. Syst. Technol. (IJDSST) 11(4), 96–115 (2019)
14. Tomura, S., Kunieda, M.: Analysis of electromagnetic force in wire-EDM. Precis. Eng. 33(3),
255–262 (2009)
15. Yan, M.-T., Huang, P.-H.: Accuracy improvement of wire-EDM by real-time wire tension
control. Int. J. Mach. Tools Manuf 44(7–8), 807–814 (2004)
16. Newton, T.R., et al.: Investigation of the effect of process parameters on the formation and
characteristics of recast layer in wire-EDM of Inconel 718. Mater. Sci. Eng., A 513, 208–215
(2009)
 Calculating Optimum Input Factors for Maximum Material Removal Rate 313

17. Huang, J., Liao, Y.: Optimization of machining parameters of wire-EDM based on grey
relational and statistical analyses. Int. J. Prod. Res. 41(8), 1707–1720 (2003)
18. Yan, M.-T., Fang, C.-C.: Application of genetic algorithm-based fuzzy logic control in wire
transport system of wire-EDM machine. J. Mater. Process. Technol. 205(1–3), 128–137 (2008)
 Calculating Optimum Main Design Factors
of a Two-Stage Helical Gearboxes for Minimum
Gearbox Mass

Do Quang Khai1 , Nguyen Hong Linh2 , Tran Huu Danh3 , Tran Minh Tan4 ,
Nguyen Manh Cuong5 , Bui Thanh Hien5 , Vu Ngoc Pi5 ,
and Nguyen Thi Quoc Dung5(B)
1 Vietnam Maritime University, Haiphong, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 Vinh Long University of Technology Education, Vinh Long, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

quocdung@tnut.edu.vn

Abstract. The goal of this study is to find the best main design parameters of
a two-stage helical gearbox in order to achieve the lowest gearbox mass. The
gear ratio of the first stage, the coefficient of wheel face width of stages 1 and 2,
and the allowable contact stress of stages 1 and 2 were all taken into account in
this work to determine their optimum values. To achieve the goal, a simulation
experiment was designed and implemented by a computer program. The Minitab
R19 software was also used to analyze the experimental results. The impact of key
design factors on gearbox mass was assessed. The best values for these parameters
were also proposed.

Keywords: Helical Gearbox · Main design parameters · Optimum gearbox

design · Gearbox mass

1 Introduction
A drive system is a system that controls a system’s speed, torque, and, ultimately, power.
There are various types of drive systems, including mechanical drive, electric drive,
pneumatic drive, hydraulic drive, and so on. Because of their simple structure, dependable
operation, and low cost, mechanical drive systems are the most commonly used.
A mechanical drive system is made up of a motor, a gearbox, and two couplings,
or a coupling and a V-belt or chain drive (Example shown in Fig. 1). The gearbox is
the most important component, as it reduces the speed and torque from the motor shaft
to the working shaft (conveyor drive shaft in Fig. 1). Therefore, many researchers have
been interested in the optimum design of the gearbox.
Up to now, there have been quite a few studies on the optimal design of gearboxes.
In [1] discusses the engineering design problem formulation for optimization as well as
the optimal design methodology. A procedure for establishing optimization statements

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 314–325, 2023.
https://doi.org/10.1007/978-3-031-22200-9_34
 Calculating Optimum Main Design Factors 315

from textual data is introduced in this study and applied to an example of a gearbox
design problem with step-by-step procedures. The general form of the objective func-
tion and design constraints for the volume of a gear train has been introduced in [2].
The overall volume of one, two, and three-stage gear trains is minimized in this study
by using a Matlab program. The authors in [3] describe a co-optimization study that
considers assemblability as a constraint during performance optimization. In this paper,
a framework is used to minimize the mass of a gearbox while also respecting a set of
assembly rules and constraints on axle stiffness and housing first resonance frequency. In
[4], modal-based design optimization of a gearbox housing using Finite Element Anal-
ysis is presented. In [5] presents a multi-objective optimization of a two-stage helical
gearbox with a wide variety of restrictions.

Fig. 1. Schema of a mechanical drive system: (1) Motor; (2) Coupling; (3) Gearbox; (4) Chain
drive; (5) Belt conveyor

Determining optimal gear ratios has received a lot of attention in gearbox design.
This is due to the fact that the gear ratio is the most important factor influencing the
size, mass, and cost of a gearbox. The optimum gear ratios have been carried out for
helical gear gearboxes [6–9], bevel gearboxes [10–12], and worm gearboxes [13–17]. In
addition, studies were carried out for two-stage [6–8, 13, 18], three-stage [10, 11, 19],
and four-stage gearboxes [20]. Although many studies on the optimization of gearbox
parameters have been conducted, they have only identified a few parameters, and no
optimization study has determined several main design parameters.
This paper describes a study to identify the optimum main design parameters for a
two-stage helical gearbox in order to achieve the smallest gearbox mass. To fulfill this,
a simulation experiment was carried out using the Taguchi method and the Minitab R19
software. The impact of main design parameters on gearbox mass was analyzed. The
best values for five main design factors have been proposed.
316 D. Q. Khai et al.

2 Methodology
2.1 Calculation of Gearbox Mass
See (Fig. 2).

Fig. 2. Calculating schema

The mass of a two-stage helical gearbox mgb is determined by (Fig. 1):

mgb = mg + mgh + ms (1)

where mgb , mgh , and ms represent the gear mass, gearbox housing mass, and shaft mass,
respectively. These mass components are defined in the following sub-sections.

2.1.1 Calculation of Gear Mass

The gear mass can be found by:

mg = mg1 + mg2 (2)

In which, mg1 and mg2 are the mass of gears of the first and the second stages:
 
π · e1· dw11
2 ·b
w1 π · e2· dw21
2 ·b
w1
mg1 = ρg · + (3)
4 4
 Calculating Optimum Main Design Factors 317

 
π · e1· dw12
2 ·b
w2 π · e2· dw22
2 ·b
w2
mg2 = ρg · + (4)
4 4

In Eqs. (3) and (4), ρg is the gear mass density (kg/m3 ); as gear material is steel, ρg
= 7.82 (kg/dm3 ); e1 and e2 are the volume coefficients of the driver gear and the driven
gear; In practice, e1 and e2 can be selected as 1 and 0.6, respectively. bw1 and bw2 are
the gear widths of stage 1 and stage 2; dw11 , dw12 , dw21 , and dw22 are the pitch diameters
of the pinion and the gear sets; These factors can be determined by [21]:

bw1 = Xba1 ·aw1 (5)

bw2 = Xba2 ·aw2 (6)

dw11 = 2 · aw1 /(u1 + 1) (7)

dw21 = 2 · aw1 · u1 /(u1 + 1) (8)

dw12 = 2 · aw2 /(u2 + 1) (9)

dw22 = 2 · aw2 · u2 /(u2 + 1) (10)

In the above equations, Xba1 and Xba2 are the wheel face width coefficients of stage
1 and stage 2; aw1 and aw2 are the center distances of stage 1 and stage 2 which can be
determined by the following equations [21]:

aw1 = ka · (u1 + 1) · 3 T11 · kH β /([σHi ]2 · u1 · Xba1 ) (11)

aw2 = ka · (u2 + 1) · 3
T12 · kH β /(AS22 · u2 · Xba2 ) (12)

where u1 is the gear ratio of stage 1; kH β = 1.01 ÷ 1.21 is the contacting load ratio for
pitting resistance; and it can be chosen as 1.11; [σH 1 ] is the allowable contact stress of
stage 1; ka = 43 (Mpa1/3 ) is the material coefficient; T11 is the torque on the drive shaft
(Nmm) of stage 1 which is determined by:

T11 = Tout /(ugb · η2hg · η3be ) (13)

In which, Tout is the output torque (Nmm); ηhg is the efficiency of a helical gear unit
(ηhg = 0.96 ÷ 0.98 [21]; ηb is the efficiency of a rolling bearing pair (ηh = 0.99 ÷
0.995 [21]).
318 D. Q. Khai et al.

2.1.2 Calculation of Gearbox Housing Mass

The mass of gearbox housing is found by the following Equation:
mgh = ρgh · Vgh (14)

where Vgh is the volume of the gearbox housing (m3 ) which can be determined by:
Vgh = 2 · Vb + 2 · VA1 + 2 · VA2 (15)
In which, Vb , VA1 , and VA2 are the volumes of bottom housing, sides A1 , sides A2 ,
respectively.
Vb = L · B1 · SG (16)

VA1 = L · H · SG (17)

VA2 = B1 · H · SG = (B − 2 · SG ) · H · SG (18)
In the above equations, L, H, B1 , and SG are calculated as:
dw11 dw22
L= + aw1 + aw2 + + 20 (19)
2 2
H = max(d w21 ; dw22 ) + 6.5 · SG (20)

B1 = bw1 + bw2 + 11 · SG (21)

SG = 0.005 · L + 4.5 [22] (22)

2.1.3 Calculation of Shaft Mass

The mass of shafts of the gearbox (kg) can be found by:
ms = ms1 + ms2 + ms3 (23)
In which,
ms1 = ρs · π · ds1
2
· ls1 /4 (24)

ms2 = ρs · π · ds2
2
· ls2 /4 (25)

ms3 = ρs · π · ds3
2
· ls3 /4 (26)
where, ms1 , ms2 , and ms3 are the mass of shafts 1, 2, and 3 (kg); ρs is the mass density
of shaft material (kg/m3 ); ls1 , ls2 , and ls3 are the length of shafts 1, 2, and 3 which are
determined by (see Fig. 1):
ls1 = bw1 + 6 · SG + 1.2 · ds1 (27)
 Calculating Optimum Main Design Factors 319

ls2 = B1 (28)

ls2 = B1 + 1.2 · ds3 (29)

In Eqs. (24)–(26), ds1 , ds2 , and ds3 are the diameters of shaft 1, 2, and 3 (mm) which
can be calculated by [21]:

ds1 = [T11 /(0, 2 · [τ ])]1/3 (30)

ds2 = [T12 /(0, 2 · [τ ])]1/3 (31)

ds3 = [T13 /(0, 2 · [τ ])]1/3 (32)

wherein, [τ] = 17 (MPa) is the allowable shear stress.

2.2 Optimization Problem

The optimization problem can be defined as follows given the previous analysis:

Minimize mgb (33)

With

mgb = f (u1 ; Xba1 ; Xba2 ; AS 1 ; AS 2 ) (34)

And with the following constraints:

1 ≤ u1 ≤ 9; 1 ≤ u2 ≤ 9 (35)

3 Simulation Experiment

A simulation experiment was conducted to investigate the impact of key design factors
on gearbox mass. Five major design parameters were considered: u1 , Xba1 , Xba2 , AS1 ,
and AS2 . These parameters and their levels are described in Table 1. In addition, the
Minitab R19 software and the Taguchi method were used for experimental design and
data analysis.
To reduce programming complexity, the impact of the main design parameters on
the gearbox mass was investigated for the following gearbox ratio values: 5, 10, 15, 20,
25, and 30. Because the 5-level Taguchi design type (L25) was chosen, 25 test runs for
the simulation experiment were performed for each above value of the gearbox ratio.
For the gearbox ratio is 10, Table 2 shows the experimental plan and the output results
(the gearbox mass).
320 D. Q. Khai et al.

Table 1. Main design factors and their levels

Factor Level
1 2 3 4 5
Total gearbox ratio u1 1 3 5 7 9
Coefficient of wheel face width of stage 1 Xba1 0.25 0.27 0.29 0.31 0.33
Coefficient of wheel face width of stage 2 Xba2 0.32 0.34 0.36 0.38 0.4
Allowable contact stress of stage 1 AS1 (Mpa) 350 368 386 404 420
Allowable contact stress of stage 2 AS2 (Mpa) 350 368 386 404 420

Table 2. Experimental plan and output results (mgb ) for ugb = 10

Trial u1 Xba1 Xba2 AS1 AS2 mgb (kg)

1 2.36 0.25 0.32 350 350 58.9
2 2.36 0.27 0.34 368 368 57.61
3 2.36 0.29 0.36 386 386 56.5
4 2.36 0.31 0.38 404 404 55.54
5 2.36 0.33 0.4 420 420 54.82
6 2.51 0.25 0.34 386 404 54.91
…
24 2.96 0.31 0.36 368 350 59.59
25 2.96 0.33 0.38 386 368 58.41

4 Results Discussion

The Analysis of Variance (ANOVA) method is used in conjunction with Minitab R19
software to assess the impact of the main design factors on mgb for ugb = 10. To determine
the effect of each main design factor on the output results, the signal-to-noise ratio, or
S/N number, must be calculated for each experiment. The calculated S/N value will
decide the best main design factor and which factor has the largest effect on gearbox
mass. To minimize performance characteristics, the S/N ratio equation is used:
 n 
1 2
S/N = −10log10 yi (36)
n
i=1

The average S/N ratio for each parameter and level was calculated after defining the
S/N ratio for each experiment. Table 3 and Fig. 2 identify the impact of the main design
factors on mgb . Figure 2 clearly shows that mgb is proportional to u1 , Xba1 , and Xba2
and inversely proportional to AS1 and AS2 . That is, mgb rises when u1 , Xba1 , and Xba2
rise, and falls when AS1 and AS2 rise.
 Calculating Optimum Main Design Factors 321

Furthermore, Table 3 shows that AS2 has the greatest influence on mgb (83.29%),
followed by AS1 (10.65%). Besides, Xba1 and Xba2 have little effect on mgb (2.97% and
2.76%). Moreover, u1 has almost no influence on mgb (0.3%). The order of the main
parameters’ influence on gearbox mass is shown in Table 4.
The S/N value is maximized for each main design factor to gain the smallest gearbox
mass using the objective function in Eq. (33). The optimum main design factors were
discovered as a result of the analysis of the effect of the factor on the S/N ratio in the
chart in Fig. 3: u1 = 2.36; Xba1 = 0.25; Xba2 = 0.32; AS1 = 420 (MPa); AS2 = 420
(MPa).

Table 3. Analysis of variance for means

Source DF Seq SS Adj SS Adj MS F P C (%)

Analysis of Variance for Means
u1 4 0.194 0.194 0.0485 13.08 0.014 0.30
Xba1 4 1.898 1.898 0.4745 127.9 0 2.97
Xba2 4 1.7663 1.7663 0.4416 119.02 0 2.76
AS1 4 6.803 6.803 1.7007 458.42 0 10.65
AS2 4 53.2286 53.2286 13.3072 3586.83 0 83.29
Residual Error 4 0.0148 0.0148 0.0037 0.02
Tota 24 63.9048
Model summary
S R-Sq R-Sq(adj)
0.0609 99.98% 99.86%

Proceed as before for the remaining values of ugb containing 5, 15, 20, 25, 30, and 35.
Table 5 shows the best values for the main design factors. The following observations
were reported in Table 5:

– The optimal Xba1 and Xba2 values are their minimum values: Xba1 = 0.25 and Xba2
= 0.32. This is because, in order to have a small gearbox mass, the coefficients Xba1
and Xba2 must be as small as possible in order to reduce the gear widths of stage 1
and stage 2 (Eqs. (5) and (6)).
– The optimal AS1 and AS2 values are also their maximum values. The reason behind
it is that, in sequence to have the smallest gearbox mass possible, the AS1 and AS2
values must be as large as possible to reduce the center distance of the gear stage 1
and stage 2 (Eqs. (11) and (12)).
– The optimal values of u1 have a first-order relationship with ugb (Fig. 4). Also, the
following regression model (with R2 = 0.9919) is proposed based on the experimental
data to determine the optimum values of u1 (see Fig. 5):
322 D. Q. Khai et al.

Fig. 3. Influence of main design factors on gearbox mass

Table 4. The order of impact of main design factors on gearbox mass

Response table for means

Level u1 Xba1 Xba2 AS1 AS2
1 56.67 56.36 56.36 57.54 58.95
2 56.68 56.54 56.57 57.08 57.71
3 56.70 56.77 56.75 56.70 56.59
4 56.78 56.93 56.94 56.33 55.63
5 56.91 57.14 57.11 56.08 54.85
Delta 0.23 0.78 0.75 1.46 4.10
Rank 5 3 4 2 1

u1 = 0.1863 · ugb + 0.6157 (37)

After obtaining u1 , the optimum values of u2 can be easily determined using u2 =

ugb /u1 .

5 Conclusions
This paper presents the findings of a study on optimizing a two-stage helical gearbox to
achieve the lowest gearbox mass. Five main design factors were optimized in this study.
They were the gear ratio of the first stage, the coefficient of wheel face width of stages
 Calculating Optimum Main Design Factors 323

Fig. 4. Main effects plot for S/N ratios

Table 5. Optimum values of main design factors

No ‘ ugb
5 10 15 20 25 30 35
u1 1.77 2.36 3.34 4.23 5.07 6.48 7.14
Xba1 0.25 0.25 0.25 0.25 0.25 0.25 0.25
Xba2 0.32 0.32 0.32 0.32 0.32 0.32 0.32
AS1 420 420 420 420 420 420 420
AS2 420 420 420 420 420 420 420

1 and 2, and the allowable contact stress of stages 1 and 2. To solve this optimization
problem, a simulation experiment with Taguchi L25 type of design was conducted. The
effect of key design factors on gearbox mass was also studied. Furthermore, the following
optimum values of the main design factors and a regression model for calculating the
optimum values of u1 were proposed: Xba1 = 0.25; Xba2 = 0.32; AS1 = 420 (MPa);
AS2 = 420 (MPa); u1 is calculated by Eq. (37).
324 D. Q. Khai et al.

Fig. 5. Optimum gear ratio of stage 1 versus gearbox ratio

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Maputi, E.S., Arora, R.: Design optimization of a gearbox: problem formulation procedure.
In: 2018 International Conference on Automation and Computational Engineering (ICACE).
IEEE (2018)
2. Golabi, S.I., Fesharaki, J.J., Yazdipoor, M.: Gear train optimization based on minimum
volume/weight design. Mech. Mach. Theory 73, 197–217 (2014)
3. Eremeev, P., et al.: Optimization of a gearbox taking into account dynamic performance and
assimilability. In: Proceeding of ISMA2020 including USD2020, pp. 2753–2762 (2020)
4. Korka, Z.I., Cojocaru, V., Miclosina, C.-O.: Modal-based design optimization of a gearbox
housing. Rom. J. Acoust. Vib. 16(1), 58–65 (2019)
5. Patil, M., Ramkumar, P., Shankar, K.:Multi-objective optimization of the two-stage helical
gearbox with tribological constraints Mech. Mach. Theory 138, 38–57 (2019)
6. Tuan, N.K., et al.: Determining optimal gear ratios of a two-stage helical reducer for getting
minimal acreage of a cross-section. In: MATEC Web of Conferences. EDP Sciences (2018)
7. Pi, V.N., et al.: A new study on determination of optimum gear ratios of a two-stage helical
gearbox. In: Advances in Material Sciences and Engineering, pp. 107–115. Springer (2020)
8. Pi, V.N., Thao, T.T.P., Tuan, D.A.: Optimum determination of partial transmission ratios of
mechanical driven systems using a chain drive and two-step helical gearbox J. Environ. Sci.
Eng. B 6, 80 (2017)
9. Hung, L.X., et al.: Calculation of optimum gear ratios of mechanical driven systems using a
two-stage helical gearbox with first stage double gear sets and chain drive. In: International
Conference on Engineering Research and Applications. Springer (2019)
10. Pi, V.N., et al.: Determination of optimum gear ratios of a three-stage bevel helical gearbox.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2019)
11. Thao, T.T.P., et al.: Determining optimum gear ratios of mechanical driven systems using
three-stage bevel helical gearbox and chain drive. In: International Conference on Engineering
Research and Applications. Springer (2019)
12. Tuan, N.K., et al.: Optimum calculation of partial transmission ratios of mechanical driven
systems using a V-belt and a three-step bevel helical gearbox. In: International Conference
on Engineering Research and Applications. Springer (2018)
 Calculating Optimum Main Design Factors 325

13. Tung, L.A., et al.: A study on determination of optimum gear ratios of a two-stage worm
gearbox. In: International Conference on Engineering Research and Applications. Springer
(2019)
14. Hong, T.T., et al.: Calculation of optimum gear ratios of mechanical driven systems using
worm-helical gearbox and chain drive. In: International Conference on Engineering Research
and Applications. Springer (2019)
15. Pi, V.N., et al.: Determining optimum gear ratios of a worm—helical gearbox for minimum
acreage of the cross section. In: Advances in Material Sciences and Engineering, pp. 89–96.
Springer (2020)
16. Le Hong Ky, T.T.H., et al.: Calculation of optimum gear ratios of mechanical driven systems
using worm-helical gearbox and chain drive (2020)
17. Pi, V.N., et al.: A study on determination of optimum gear ratios of a worm-helical gearbox.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2019)
18. Cam, N.T.H., et al.: Determining optimal partial transmission ratios of mechanical driven
systems using a V-belt drive and a helical reducer with second-step double gear-sets. In:
International Conference on Engineering Research and Applications. Springer (2018)
19. Hong, T.T., et al.: A study on calculation of optimum gear ratios for three stage bevel helical
gearboxes, vol. 500, p. 1 (2006)
20. Pi, V.N.: Optimal determination of partial transmission ratios for four-step helical gearboxes
with first and third step double gear-sets for minimal mass of gears, vol. 2, p. 2 (2008)
21. Chat, T., Uyen Van, L.: Design and Calculation of Mechanical Transmissions Systems 1
Educational Republishing House Hanoi (2007)
22. Römhild, I., Linke, H.: Gezielte Auslegung Von Zahnradgetrieben mit minimaler Masse auf
der Basis neuer Berechnungsverfahren. Konstruktion (1981) 44(7–8), pp. 229–236 (1992)
 Deep ANFIS for Understanding Emotions
in Movie Clips

Tuan-Linh Nguyen(B)

Thainguyen University of Technology, Tich Luong Ward, 3/2 Street, 666 Thai Nguyen City,
Vietnam
ntlinh@tnut.edu.vn

Abstract. Significant success of Deep learning (DL) in extracting features from

big data which led to an explosion of Artificial Intelligence (AI) applications.
However, most of those AI systems lack transparency and interpretability due to
the black-box nature of DL architecture, which causes major trust-related prob-
lems in AI applications where validation is essential. This necessitates the need
for interpretable AI systems that can explain their decisions. In this view, this
paper proposes a deep adaptive neuro-fuzzy inference system (Deep ANFIS), that
combines fuzzy inference with DL architecture to provide a more comprehensible
explanation of the decision processes. Deep ANFIS uses ANFIS based convolution
units in convolutional neural networks (CNNs) architecture that allows approxi-
mating the complex non-linear problems using fuzzy if-then rules. To verify the
performance of the proposed Deep ANFIS and analyze the essential features for
classification, we perform experiments with COGNIMUSE dataset for emotion
recognition.

Keywords: Adaptive neuro-fuzzy inference system (ANFIS) · Deep learning

(DL) · Convolutional Neural Network (CNN) · Deep ANFIS · Emotion
classification · Rules extraction

1 Introduction
Deep Learning (DL) has emerged as a family of efficient machine learning models
with fairly good classification accuracy in different application areas [1]. This can be
attributed to the hierarchical architecture that can effectively extract high-level features
from training data set in a layer-by-layer manner [2]. However, further information on
how this representation is distributed layer-wise, and how it arises from learning is not
very clear [3]. This lack of transparency in the training process often causes crucial trust-
related problems in critical application areas such as health care, finance, self-driving
cars, and different problem domains of artificially intelligent (AI) where reliability is
essential. A vital component of an AI system is the ability to explain its decisions. These
explanations offer an insight into why a particular decision has been taken.
Rule extraction is an approach to expose the model’s hidden knowledge. Simple rule-
based methods like decision lists, decision trees, or fuzzy IF-THEN rules provide a more

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 326–334, 2023.
https://doi.org/10.1007/978-3-031-22200-9_35
 Deep ANFIS for Understanding Emotions in Movie Clips 327

comprehensible explanation of the decision processes. Fuzzy logic can also deal with
many practical problems of uncertainties and ambiguities present in raw data. These
structured rule-based systems improve their ability to use available subjective infor-
mation. This enables the analyst to provide expert knowledge to the system, improve
classification, or change the system’s behavior. Both fuzzy systems and neural networks
have proved their effectiveness in data representation. The adaptive neuro-fuzzy infer-
ence system (ANFIS) [4] is an amalgamation of neural networks (NN) and a fuzzy
inference system (FIS) that allows for complex non-linear problems to be approximated
using fuzzy if-then statements with increased adaptation capability. ANFIS works well
for applications that require an extensive knowledge base to accommodate various sce-
narios that cannot be fully captured by domain experts. Although ANFIS has shown
significant improvement in classification accuracy, interpreting thousands of fuzzy rules
and parameters for high dimensional application problems remains a problem.
DL is a powerful paradigm that allows rich and discriminative feature learning from
big data [2]. This has attracted increasing attention among researchers in interpreting the
decisions of deep learning models. This has considered the visual analysis of intermedi-
ate deep layers in image classification problems [6] as well as providing an explanation of
the model’s decisions [7]. Techniques for explaining DNNs visualized the hidden states
of neural models [8], most notably of CNNs in image classification. In [9], layer-wise
relevance propagation (LRP) and sensitivity analysis (SA) that explained the model pre-
dictions was presented. In [10], deep Taylor decomposition efficiently back propagated
the explanations from the output of DNN to the input layer. The explanation process
provided the importance of each pixel for the prediction in terms of a heat map. How-
ever, for more complex or nested architecture, the LRP approach was not satisfying. In
[11], the model identified contents in the image and generated the caption. An expla-
nation for each word in the caption was provided by highlighting the relevant pixels of
the input image. In [12], Deep models were trained for visual question answering tasks
for communication with humans. The decisions were justified textually, and attention
mechanism was used to emphasize the evidence in the images. Textual explanations,
along with visual classification, were generated using DNN [13]. To produce better
explanations, reinforcement learning based discriminative loss was used.
Convolutional Neural Network (CNN) is the most popular DL methods with a large
number of linear (convolution operation) and nonlinear transformations (activation func-
tion) in each convolution layer enabling efficient big data feature extraction [5]. CNN
has proved its effectiveness in image recognition and classification. The ability of CNNs
to solve a variety of complex problems can be further improved by explainable rule-
based structures extracting understandable rules. Not many have attempted to study the
rule extraction algorithm from complex deep neural networks. Hybrid models combine
transparent methods with black-box models to achieve an appropriate trade-off between
accuracy and interpretability. In this view, this paper proposes the Deep ANFIS that suc-
cessfully integrates ANFIS with CNNs, by replacing the convolutions in hidden deep
layers with ANFIS regression units. Each ANFIS unit works as a convolution. The pro-
posed framework benefits from the rule-based structure of FIS to explain the high-level
features of CNN, offering both a label and an explanation. The proposed Deep ANFIS
extracts features from both audio and video modalities and inputs to an ANFIS classifier,
which is an integration of an ANFIS and a softmax layer.
328 T.-L. Nguyen

With a greater spread of AI applications, end-to-end learning interpretable DNN is of

interest. Since CNN benefits from the principles such as local receptive fields and shared
weights in extracting robust features from image inputs, we intend to integrate ANFIS
with CNNs, by replacing the convolutions in hidden deep layers with ANFIS regression
units. ANFIS works well for applications that require extensive knowledgebase [4].
Therefore, these integrated systems provide information-rich features obtained from
CNN that will be interpreted by fuzzy-inference systems offering explanations for how
the system has selected a classification label.
The important contributions of this research are:

• This paper proposes Deep ANFIS, which is an integration of ANFIS into the deep
CNN architecture that can explain the features extracted layer-wise using fuzzy if-then
rules.
• Our second contribution is to interpret the system’s conclusions based on fuzzy rules
extraction.
• Our third contribution is fuzzy rule-set simplification, which significantly reduces the
number of rules and antecedent features after training the model. The minimal rule-
set with minimized input features improves the interpretability of the model while
maintaining the performance.

In the next section, we present a review of techniques for interpreting machine

learning models with more focus on deep neural networks (DNNs). In Sect. 2, we
present the details of the proposed Deep ANFIS model. In Sect. 3, we present different
methods for audio, video feature extraction, and corresponding model configurations.
Section 4 presents the experimental results with detailed discussions on critical features
from different modalities that contribute to the classification. In the last section, we
provide conclusions based on our results.

2 Proposed Deep ANFIS

Figure 1 shows the conceptual configuration of the proposed Deep ANFIS. The Deep
ANFIS replaces the hidden convolution layers of the deep architecture by ANFIS regres-
sion units. Each ANFIS unit works as a convolution. Therefore, each ANFIS will output
one feature map of higher layer feature maps.

xjl = ANFISj (xil−1 ), i ∈ Mj (1)

where Mj represents a selection of jth input map.

For improving the generalization capabilities along with reducing the complexity, we
use the CNN architecture in which ANFIS convolutional layers are followed by structural
elements of pooling, batch normalization, and dropout. In the last layer, ANFIS classifier
is used to classify the features obtained from the final convolutional ANFIS layer which
into multiple classes. Section 2.2 gives details about the ANFIS classifier.
 Deep ANFIS for Understanding Emotions in Movie Clips 329

Fig. 1. The proposed Deep ANFIS framework.

2.1 ANFIS Convolutional Units

As shown in Fig. 2, for each input feature map, we slide a small window over the input
multiple times, each time we get an input feature set. Each feature in this set has a 2D
shape. To benefit from CNNs principles of local receptive fields, shared weights, feature
maps, and pooling (or down/sub-sampling), we flatten those feature sets and pass to
an ANFIS regression. We perform multiple ANFIS convolutions, each using distinct
ANFIS parameters and resulting in a different output feature map. Then, we stack all the
feature maps together and get the output of the ANFIS convolution layer. Incorporating
ANFIS convolutions into DL architecture gives the model the ability to extract fuzzy
spatial relation of input feature maps, as opposed to other hierarchical fuzzy models.
After the ANFIS convolution layer, we apply some traditional CNNs operators such
as pooling, dropout, and batch normalization. The pooling strategy used is Max-pooling,
which is a down-sampling method. This operator not only reduces the size of the feature
map, but also generalizes the results from the ANFIS convolution filter.

Fig. 2. ANFIS convolutional layer

330 T.-L. Nguyen

2.2 ANFIS Classifier

Finally, after several ANFIS convolutional layers, the last layer of the network is an
ANFIS classifier (clfANFIS). As shown in Fig. 3, the ANFIS classifier is an integration
of an ANFIS and a softmax layer. The consequence part of the ANFIS classifier includes
multiple consequence units. Each unit receives the same output from the premise part
and input variables, and computes output value corresponding to one class. The softmax
layer is a normalized exponential function that calculates each class’s probabilities in a
multi-class problem. By integrating with a softmax layer, the ANFIS classifier can be a
multi-class classifier so that the Deep ANFIS can be applied to any multi-class classifier
problem.

Fig. 3. ANFIS classifier.

3 Deep ANFIS for Classification

3.1 Deep ANFIS for Audio Emotion Classification
Feature Extraction in Audio Domain
The input to Deep ANFIS is usually 2-D. It is therefore important to transform the audio
segment in to 2-D by extracting the static and delta channels of log Mel-spectrogram
[14]. For each utterance, the log Mel-spectrogram is obtained using 32 Mel-filter banks
ranging from 20Hz to 8kHz using a 25ms Hamming window with out overlap. A context
window of 5 frames is applied to the whole log Mel-spectrogram to obtain the 2-D
Mel-spectrogram. Once the static 2-D segment of size 32 × 5 is extracted, the temporal
information of the segment is captured using a first-order frame-to-frame time derivative.
The Mel-spectrogram segment with two channels of size is 2 × 32 × 5 is obtained.
 Deep ANFIS for Understanding Emotions in Movie Clips 331

Model Configuration
The details of model’s configuration and the parameter setting of Deep ANFIS for audio
sentiment analysis is shown in Table 1. For fair comparison, the parameters used by
CNN are same as Deep ANFIS.

Table 1. Parameters of Deep ANFIS for emotion classification in Audio domain

Layer CNN Deep ANFIS

Input Train: 1914 × (2 × 32 × 5) Train: 1914 × (2 × 32 × 5)
Test: 759 × (2 × 32 × 5) Test: 759 × (2 × 32 × 5)
Conv1 96 × (2 × 5 × 1), ReLU 4 × ANFIS(2 × 5 × 1) (100 rules, 3 MFs)
Conv2 128 × (3 × 1), ReLU 6 × ANFIS(4 × 3 × 1) (120 rules, 3 MFs)
Conv3 192 × (3 × 1), ReLU 8 × ANFIS(6 × 3 × 1) (180 rules, 3 MFs)
Classifier 384 × 2 clfANFIS(16 × 2) (160 rules, 3 MFs)
Params ∼112,000 ∼95,000

3.2 Deep ANFIS for Video Emotion Classification

Feature Extraction in Visual Domain
Gabor function is used to extract the orientation information [15]. The orientation feature
tensor is the output magnitude obtained by group of multi-scale oriented filters whose
values include 0°, 45°,90°, and 135°. To obtain the color information, each RGB color
frame is transformed to HSI color space. To reduce the computational cost, we apply
downsampling to both color and orientation features. We then obtain a feature tensor of
size 32 × 32 × 4 by down-sampling all the orientation values from 320 × 240 to 32 × 32.
The H , S, I channels are also resized from 320 × 240 to 100 × 100. Each frame will
have a total of seven channels with three H , S, I images and four orientation matrices.
A 32-bin histogram is computed to construct the final input feature tensor for each of
seven channels, of size (7 × 32). For seven movies, which are 30 min long and having
24 frames/second sample rate, the size of the feature set is 2673 × (7 × 32).
Model Configuration
The hyper-parameters of Deep ANFIS for visual sentiment analysis and its structure are
given in Table 2. CNN uses the same setting as Deep ANFIS for comparison.

4 Experiments and Results

We verify the performance of Deep ANFIS for emotion recognition and perform experi-
ments on COGNIMUSE database [16]. COGNIMUSE dataset is a multimodal database
annotated with semantics, audio and visual events along with emotion. In experiments
332 T.-L. Nguyen

Table 2. Parameters of deep ANFIS for video emotion classification

Layer CNN Deep ANFIS

Input Train: 1914 × (7 × 32) Train: 1914 ×(7 × 32)
Test: 759 ×(7 × 32) Test: 759 ×(7 × 32)
Conv1 64 × (1 × 7), ReLU 8 × ANFIS(7 × 7 × 1) (150 rules, 3 MFs)
Conv2 96 ×(1 × 5), ReLU 12 × ANFIS(8 × 5 × 1) (150 rules, 3 MFs)
Conv3 128 ×(1 × 3), ReLU 16 × ANFIS(12 × 3 × 1) (120 rules, 3 MFs)
Classifier FC(128 × 2) clfANFIS(16 × 2) (120 rules, 3 MFs)
Params ∼120,000 ∼110,000

with audio and video modalities, the performance of Deep ANFIS with conventional
CNN, Support Vector Machine (SVM), and ANFIS model are compared. Grid search is
used to select the optimal hyper-parameters of the SVM as: kernel = ‘poly’; degree =
3; C = 0.35; coeff = 0.125, gamma = 0.0625. For comparisons with the fuzzy inference
system, we do the same experiments with the original ANFIS model. The emotion clas-
sification using the proposed Deep ANFIS significantly outperforms other comparison
models.

4.1 Audio Emotion Classification

The model is trained and tested by repeating the experiment for 10-times. For training
data, the excerpts from 5 movies (BMI, CHI, FNE, GLA, LOR) are extracted and tested
on excerpts from 2 movies (CRA, DEP). A CNN with similar structure as Deep ANFIS
is used for comparison. To avoid the inequality of comparison, we increase the number
of CNN’s hidden convolution layers to get its maximum performance. Table 3 gives
the comparison of statistical classification accuracy of Deep ANFIS with SVM, ANFIS
and CNN. The proposed Deep ANFIS can get the best accuracy with a smaller standard
deviation than the baseline CNN.

Table 3. The average emotion classification accuracy in audio domain

SVM(%) ANFIS(%) CNN(%) DeepANFIS(%)

Average 70.4 76.1 79.6 81.5
STDa 0.93 1.15 1.78 1.17
a Standard Deviation

4.2 Video Sentiment Analysis

We train the classifier on five movie clips and test on two others and repeat the experiment
ten times. This is compared with CNN having a similar structure and complexity [14].
 Deep ANFIS for Understanding Emotions in Movie Clips 333

For fair comparison, we increase the size of the convolution layers of the CNN and train
to get the best of its accuracy. For fair comparison, the hyper-parameters (number of
layers, dropout rate and pooling probabilities) for both proposed and baseline methods
are set to the same value during training.
For both positive and negative emotions, the average classification accuracies
obtained using CNN and Deep ANFIS are compared in Table 4. The comparison results
suggest that the average standard deviation of proposed Deep ANFIS outperforms the
CNN model. This necessitates the need to analyse the potential of ANFIS convolutional
layers in extracting higher features for classification. The feature set visualization of
video features extracted is performed in a similar way as in Sect. 5 and using t-SNE.

Table 4. The average emotion classification accuracy in video domain

SVM(%) ANFIS(%) CNN(%) DeepANFIS(%)

Average 72.3 75.5 80.8 82.6
STD 0.84 1.12 1.79 0.95

5 Conclusion

The proposed framework benefits from the advantages of fuzzy modeling such as robust-
ness against uncertainties and transparency of the decision processes. On the other hand,
it also benefits from the high-level feature abstraction of deep learning for big data. This
paper proposes a hybrid model which integrates CNN and fuzzy inference system called
Deep ANFIS for classifying emotions using audio and video modalities. The proposed
Deep ANFIS can interpret the inference process for audio and video emotion layer-wise
using fuzzy if-then rules and their contribution to the prediction performance. In future,
we focus on developing a hierarchical framework to further reduce the rule size. We also
intend to develop a deep recurrent neuro-fuzzy network to increase the interpretability.

Acknowledgment. This work was supported by Thainguyen University of Technology.

References
1. Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, V.V.D., Erhan, D., Rabinovich,
A.: Going deeper with convolutions. In: 2015 IEEE Conference on Computer Vision and
Pattern Recognition (CVPR), Boston, MA, vol. 2015, pp. 1–9 (2015)
2. Wang, W.N., Yu, Y.L., Jiang, S.M.: Image retrieval by emotional semantics: A study of
emotional space and feature extraction. In: Conference Proceedings-IEEE International Con-
ference on Systems, Man and Cybernetics, vol. 4, pp. 3534–3539 (2007). https://doi.org/10.
1109/ICSMC.2006.384667
3. Montavon, G.: On layer-wise representations in deep neural networks (2013)
334 T.-L. Nguyen

4. Jang, J.-S.: Anfis: adaptive-network-based fuzzy inference system. IEEE Trans. Syst. Man
Cybern. 23, 665–685 (1993)
5. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: 2016
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 770–778 (2016).
arXiv:1512.03385. https://doi.org/10.1109/CVPR.2016.90
6. Zeiler, M.D., Fergus, R..: Visualizing and understanding convolutional networks. In: ECCV
(2014)
7. Gan, C., Wang, N., Yang, Y., Yeung, D.-Y., Hauptmann, A.G.: Devnet: a deep event network for
multimedia event detection and evidence recounting. In: 2015 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), pp. 2568–2577 (2015)
8. Tzeng, F.Y., Ma, K.L.: Opening the black box-data driven visualization of neural networks.
In: Proceedings of IEEE Visualization-05 Conference, pp. 383–390 (2005)
9. Samek, W., Wiegand, T., Muller, K.R.: Explainable artificial intelligence: understanding,
visualizing and interpreting deep learning models. ITU J. ICT Discoveries-Special Issue
I-Impact Artificial Intelligence, pp. 1–10 (2017). arXiv:1708.08296
10. Montavon, G., Lapuschkin, S., Binder, A., Samek, W., Muller, K.R.: Explaining nonlinear
classification decisions with deep taylor decomposition. Pattern Recogn. 65, 211–222 (2017).
arXiv:1708.08296
11. Xu, K., et al.: Show, attend and tell: neural image caption generation with visual attention.
In: International Conference on Machine Learning, pp. 2048–2057 (2015)
12. Park, D.H., Hendricks, L.A., Akata, Z., Schiele, B., Darrell, T., Rohrbach, M.: Attentive
explanations: Justifying decisions and pointing to the evidence (2016).. arXiv:l61204757
13. Hendricks, L.A., Akata, M.R.Z., Donahue, J., Schiele, B., Darrell, T.: Generating visual
explanations. In: Computer Vision-ECCV 2016, pp. 3–19 (2016). arXiv:1603.08507
14. Huzaifah, M.: Comparison of time-frequency representations for environmental sound
classification using convolutional neural networks. https://arxiv.org/abs/1706.07156
15. Kumar, A., Pang, G.K.H.: Defect detection in textured materials using gabor filters. IEEE
Trans. Ind. Appl. 38(2), 425–440 (2002). https://doi.org/10.1109/28.993164
16. Zlatintsi, A., et al.: Cognimuse: a multimodal video database annotated with saliency, events,
semantics and emotion with application to summarization. EURASIP J. Image Video Process
2017(1), 54 (2017)
 Determination of Best Input Factors
for PMEDM 90CrSi Tool Steel Using MABAC
Method

Nguyen Hong Linh1 , Phan Dang Phong2 , Thangaraj Muthuramalingam3 ,

Tran Minh Tan4 , Tran Huu Danh5 , Vu Ngoc Pi6 , Hoang Xuan Tu6 ,
and Nguyen Van Tung6(B)
1 Electric Power University, Ha Noi, Vietnam
2 National Research Institute of Mechanical Engineering, Ha Noi, Vietnam
3 SRM Institute of Science and Technology, Kattankulathur, India
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Vinh Long University of Technology Education, Vinh Long, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

mr.nguyentung85@tnut.edu.vn

Abstract. This paper describes a multi-criteria decision making (MCDM) study

that was conducted to determine the best input process factors for powder-mixed
electric discharge machining (PMEDM) 90CrSi tool steel in order to achieve
minimal surface roughness (RS), maximum material removal speed (MRS), and
maximum electrode wear rate (EWR) all at the same time. To accomplish this,
the MCDM problem was solved using the MABAC (Multi-Attributive Border
Approximation Area Comparison) method, and the weights of the criteria were
calculated using the MEREC (Method based on the Removal Effects of Criteria)
method. In addition, five input factors were investigated: powder concentration,
pulse-on time, pulse-off time, pulse current, and servo voltage. Furthermore, the
experiment was designed using the Taguchi method with an L18 (21 + 34 ) orthog-
onal array. The best input process factors for PMEDM 90CrSi steel have been
proposed based on the study’s findings.

Keywords: PMEDM · MCDM · MABAC · TOPSIS · MEREC · Surface

roughness · Material removal speed · Electrode wear rate · 90CrSi

1 Introduction
MCDM is a technique used to analyze many different options to select the best one.
This technique is very widely used not only in medicine [1, 2], economics [3, 4],
social sciences [5, 6], or military, but also in mechanical machining processes. This
is because machining processes are often required to meet multiple criteria, such as
minimum machined surface roughness (SR), maximum material removal rate (MRR)
minimum manufacturing cost. In fact, machining process criteria frequently contradict
one another. A requirement for an increase in MRR, for example, will result in an increase

The original version of this chapter was revised: Author name has been corrected. The correction
to this chapter is available at https://doi.org/10.1007/978-3-031-22200-9_104

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023, corrected publication 2023
D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 335–344, 2023.
https://doi.org/10.1007/978-3-031-22200-9_36
336 N. H. Linh et al.

in depth of cut, an increase in feedrate, and, as a result, a decrease in SR. Therefore, it is

very appropriate to apply MCDM technique to choose the best solution to a machining
process.
Several studies on the use of MCDM in machining processes have been conducted
up to this point. Many studies have been conducted to solve MCDM problems in milling,
turning, grinding, electric discharge machining, and so on. For milling process, Trung
[7] conducted an MCDM study in which five methods including EDAS, MARCOS, PIV,
MOORA, and TOPSIS were used. The purpose of this work is to find the best cutting
parameters for both low SR and high MRR. In [8], an MCDM study to determine the
best dressing mode for internal grinding was proposed. Four MCDM methods including
TOPSIS, MARCOS, EAMR, and MAIRCA were chosen in this work to obtain the min-
imum SR and maximum MRR simultaneously. When processing silicon steel, Phan and
Muthuramalingam [8] used the MCDM technique to determine the best input parame-
ters in a low frequency vibration assisted EDM process. The DEAR (Data Envelopment
Analysis based Ranking) method and the Taguchi method were used in their study. The
DEAR method was also applied in [9] for EDM process when machining SKD61 die
steel. The results of an MCDM study when processing cylindrically shaped parts of
90CrSi tool steel for the PMEDM process were reported in [10]. In the study, MARCOS
(Measurement of Alternatives and Ranking according to Compromise Solution), TOP-
SIS (Technique for Order of Preference by Similarity to Ideal Solution), and MAIRCA
(Multi-Attributive Ideal-Real Comparative Analysis) were applied.
In this paper, the results of a study on the application of the MABAC method for
determining the best input process factors for PMEDM 90CrSi are introduced. In this
study, three criteria including SR, MRS, and EWR were selected for the investigation.
Besides, the MEREC method was chosen to calculate the weights of criteria. The best
input parameters for PMEDM 90CrSi tool steel has been proposed.

2 Methodology
2.1 Method for Multi-criteria Decision Making
Pamucar and Cirovic invented the MABAC method in 2015. The following are the steps
for implementing this method [11]:
Step 1: Creating the initial decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X = ⎢ r1 · · · rij · · · rin ⎥
⎢
⎥ (1)
⎢ . . . . . ⎥
⎣ . .
. .. ... . . ⎦
rm1 · · · rmj · · · rmn m×n

In which, m is the number of options; n is the number of criteria.

Step 2: Finding the normalized values r ij by the following Equation:

rij − ri−
rij∗ = (2)
ri+ − ri−
 Determination of Best Input Factors 337

rij − ri+
rij∗ = (3)
ri− − ri+

wherein, i = 1,2,…,m and j = 1,2,…,n. Equation (2) is used for criteria MRS, and
(3) is used for criteria SR and EWS. In addition, ri+ = max(r1 , r2 , . . . , rm ) and ri− =
min(r1 , r2 , . . . , rm ).
Step 3: Calculating the weighted matrix elements by:

vij = wj + wj × rij∗ (4)

Step 4: Determining the border approximation area matrix by:


1/m

m
gj = vij (5)
i=1

In which j = 1,2,…,n.
Step 5: Calculating the distance between the alternatives and the border approxima-
tion area by:

qij = vij − gi (6)

where, i = 1,2,…,m and j = 1,2,…,n.

Step 6: Determining the total distances of each alternative from the approximate
border area:

n
Si = qij i = 1, 2, . . . , m (7)
j=1

Step 7: Ranking the alternatives by maximizing S.

2.2 Method for Determining Criterion Weights

In this work, the MEREC method was used to determine the weight of the criteria. To
use this method, take the following steps [12]:
Step 1: As described in step 1 of the MABAC method, create the initial matrix:
Step 2: Calculating the normalized values by:

– If j is MRS:
minxij
hij = (8)
xij
– If j is SR or EWR:

xij
hij = (9)
maxxij
338 N. H. Linh et al.

Step 3: Finding the alternative performance S i by:

⎡ ⎛ ⎞⎤
1   
Si = ln⎣1 + ⎝ ln hij ⎠⎦ (10)
n
j

Step 4: Determining the performance of ith alternative Si.j ’ by:

⎡ ⎛ ⎞⎤
 1    
Sij = Ln⎣1 + ⎝ ln hij ⎠⎦ (11)
n
k,k=j

Step 5: Calculating the removal effect of the jth criterion Ej by the following equation:
  

Ej = Sij − Si  (12)
i

Step 6: Determining the weights of criteria by:

Ej
wj =  (13)
k Ek

3 Experimental Work
Figure 1 shows the set up of the experiment. A Sodick A30 EDM machine is used to
perform the PMEDM process. The samples are cylindrical in shape and made of 90CrSi
tool steel. Graphite electrodes, SiC powder with a particle size of 100 (nm), and Diel
MS7000 dielectric solution are used. Table 1 shows the input factors and their levels.
Furthermore, the Taguchi method and Minitab R19 software are used to design and
analyze the experiment with the L18 (21 + 34 ) design. As a result, a total of 18 test runs
were carried out. Table 2 shows the experimental results, which were measured.

4 Determining the Best Input Process Factors

This section explains how to use the MABAC method to solve the MCDM problem and
the MEREC method to calculate criteria weights to determine the best input factors for
PMEDM 90CrSi tool steel.

4.1 Determining the Criation Weights

The MEREC method can be used to estimate the weights of criteria in the following steps
(see Sect. 2.2): Use Eqs. (10) and (11), calculate Si and Sij ’; use Eq. (12) to calculate the
criterion removal effect; and use Eq. (13) to calculate the weight of the criteria wj . The
weights of Ra, MRS, and EWR were also discovered to be 0.3743, 0.3234, and 0.3023,
respectively.
 Determination of Best Input Factors 339

Fig. 1. Experimental setup

Table 1. Input dressing parameters

No Input parameters Code Unit Level

1 2 3
1 Powder concentration Cp g/l 0 0.5 1
2 Pulse-on time Ton µs 8 12 16
3 Pulse-off time Toff µs 8 12 16
4 Servo current IP A 5 10 15
5 Servo voltage SV V 4 5 -

4.2 Determining the Best Input Factors

The following are the steps for performing MCDM using the MABAC method (see
Sect. 2.1): The initial matrix is generated using Eq. (1). Then compute the normalized
values of rij *, keeping in mind that Eq. (2) is used for MRS criteria and Eq. (3) is used
for SR and EWS criteria. A formula is then used to compute the normalized weighted
values vij (4). The border approximation area matrix is then computed using Eq. (5).
The formula (6) is then used to calculate the distance between the alternatives and the
border approximation area qij . Finally, the total distances between each option and the
approximate border area Si are calculated using the formula (7). S is maximized to
determine the ranking of options. Table 3 shows several calculated parameters as well
as the option ranking.
Table 3 shows that Option 5 is the best option. This is because it has the highest Si
value (Si = 0.1751). As a result, the set of optimal input factors for PMEDM 90CrSi
involves the following input variables: Cp = 0.5 (g/l); Ton = 12 (s); Toff = 12 (s); IP =
15 (A); SV = 4 (V).
340 N. H. Linh et al.

Table 2. Experimental plan and output results

No Cp (g/l) Ton (µm) Toff (µm) IP (A) SV (V) Ra (µm) MRS (g/h) EWR (g/h)
1 0 8 8 5 4 1.965 0.731 22.305
2 0 12 12 10 4 2.928 1.248 25.572
3 0 16 16 15 4 6.930 3.820 33.078
4 0.5 8 8 10 4 2.069 2.552 74.008
5 0.5 12 12 15 4 5.371 7.069 78.978
6 0.5 16 16 5 4 5.054 1.428 16.957
7 1 8 12 5 4 2.041 3.889 114.339
8 1 12 16 10 4 2.771 1.398 30.264
9 1 16 8 15 4 7.704 4.411 34.357
10 0 8 16 15 5 2.909 6.969 143.748
11 0 12 8 5 5 3.723 2.088 33.656
12 0 16 12 10 5 3.565 0.799 13.450
13 0.5 8 12 15 5 2.869 7.100 148.474
14 0.5 12 16 5 5 4.126 1.943 28.264
15 0.5 16 8 10 5 2.952 0.814 16.534
16 1 8 16 10 5 1.674 2.328 83.435
17 1 12 8 15 5 4.841 7.105 88.055
18 1 16 12 5 5 4.568 1.344 17.656

The TOPSIS method was used to validate the results of the MABAC method. This
method was chosen because it is a very old method (it was introduced in 1993 by Ching-
Lai Hwang et al. [13]). Furthermore, this method has been used to determine the best
solution for a wide range of mechanical machining processes, including turning, milling,
grinding, electrical discharge machining, and so on [10]. The steps to use the TOPSIS
method to solve MCDM problem as shown in [14]. Table 4 shows the priority indexes and
rankings of alternatives when using the TOPSIS method to solve the MCDM problem
and the MEREC method to determine the weight of criteria.
Figure 4 also compares the results of the two MCDM methods for solving the MCDM
problem with the MEREC method for weight calculation. Figure 4 and Tables 3 and 4
show that both methods agree on the same result - the fifth solution is the best. This
result demonstrates that the MABAC method can solve the MCDM problem in PMEDM
90CrSi tool steel.
The Spearman’s rank correlation coefficient (R) was used to compare the degree
of association between ranks obtained using the MABAC and TOPSIS methods. The
following formula is used to calculate this coefficient [12]:

6 · ni=1 Di2
R=1−   (14)
n · n2 − 1
 Determination of Best Input Factors 341

Table 3. Several calculated results and ranking of options using MABAC method

Trial gij qij Si Rank

Ra MRS EWR Ra MRS EWR
1 0.3638 0.4330 0.4991 −0.0048 −0.1096 0.0856 −0.0288 15
2 0.3638 0.4330 0.4991 −0.0022 −0.0833 0.0783 −0.0073 12
3 0.3638 0.4330 0.4991 0.0084 0.0472 0.0615 0.1171 4
4 0.3638 0.4330 0.4991 −0.0045 −0.0172 −0.0302 −0.0518 16
5 0.3638 0.4330 0.4991 0.0043 0.2121 −0.0413 0.1751 1
6 0.3638 0.4330 0.4991 0.0034 −0.0742 0.0976 0.0268 5
7 0.3638 0.4330 0.4991 −0.0046 0.0507 −0.1204 −0.0743 17
8 0.3638 0.4330 0.4991 −0.0027 −0.0757 0.0678 −0.0106 14
9 0.3638 0.4330 0.4991 0.0105 0.0772 0.0586 0.1463 3
10 0.3638 0.4330 0.4991 −0.0023 0.2070 −0.1863 0.0184 9
11 0.3638 0.4330 0.4991 −0.0001 −0.0407 0.0602 0.0194 8
12 0.3638 0.4330 0.4991 −0.0005 −0.1061 0.1054 −0.0012 11
13 0.3638 0.4330 0.4991 −0.0024 0.2137 −0.1968 0.0144 10
14 0.3638 0.4330 0.4991 0.0009 −0.0480 0.0722 0.0252 6
15 0.3638 0.4330 0.4991 −0.0022 −0.1054 0.0985 −0.0090 13
16 0.3638 0.4330 0.4991 −0.0056 −0.0285 −0.0513 −0.0853 18
17 0.3638 0.4330 0.4991 0.0028 0.2139 −0.0616 0.1551 2
18 0.3638 0.4330 0.4991 0.0021 −0.0784 0.0960 0.0197 7

In which n is the number of options; D is difference between ranks.

The Spearman’s rank correlation coefficient for rankings obtained using the two
methods was calculated to be 0.9698. This is a very good result when compared to 0.96
in [15] and 0.83 in [16] (Fig. 2).

5 Conclusions

This paper presents the results of using the MCDM technique to find the best input
factors for PMEDM 90CrSi. To do this, the MABAC method was used to solve the
MCDM problem, and the MEREC method was used to calculate the weights of the cri-
teria. Furthermore, the following input factors were investigated: powder concentration,
pulse-on time, pulse-off time, pulse current, and servo voltage. Besides, the experiment
was designed with the L18 (21 + 34 ) orthogonal array using the Taguchi method. The
following conclusions can be drawn from the work’s findings:
342 N. H. Linh et al.

Table 4. Several calculated results and ranking using TOPSIS method

Trial Lij Di+ Di− Ri Rank

Ra T R
1 0.3000 0.1355 0.2466 1.1870 1.64717 0.5812 7
2 0.4469 0.2314 0.2827 1.1110 1.54503 0.5817 6
3 1.0576 0.7085 0.3657 1.0303 1.40366 0.5767 9
4 0.3158 0.4733 0.8183 1.0793 1.23761 0.5342 14
5 0.8197 1.3111 0.8732 0.9183 1.44891 0.6121 1
6 0.7713 0.2649 0.1875 1.1730 1.5149 0.5636 13
7 0.3115 0.7214 1.2642 1.2661 1.11026 0.4672 18
8 0.4230 0.2593 0.3346 1.0876 1.5134 0.5819 5
9 1.1758 0.8182 0.3799 1.0723 1.43459 0.5723 11
10 0.4440 1.2926 1.5894 1.4532 1.37008 0.4853 16
11 0.5682 0.3873 0.3721 1.0066 1.42978 0.5868 3
12 0.5442 0.1482 0.1487 1.2045 1.6211 0.5737 10
13 0.4379 1.3169 1.6416 1.5040 1.39287 0.4808 17
14 0.6297 0.3604 0.3125 1.0407 1.45445 0.5829 4
15 0.4506 0.1509 0.1828 1.1834 1.62921 0.5792 8
16 0.2555 0.4318 0.9225 1.1762 1.20492 0.506 15
17 0.7388 1.3177 0.9736 0.956 1.42642 0.5987 2
18 0.6971 0.2493 0.1952 1.157 1.52781 0.5691 12

1. The MABAC method was used to solve the MCDM problem in PMEDM 90CrSi,
and the results were obtained.
2. The TOPSIS was used to validate the results of the MABAC method. According
to the results, all of two methods identified Option 5 as the best for the PMEDM
process.
3. The Spearman rank correlation coefficient has been found to be highly useful in
determining the relationship between different ranking methods.
4. The best input fcators for PMEDM 90CrSi to achieve the minimum surface rough-
ness, the maximum wheel life and the minimum roundness simultaneously are: Cp
= 0.5 (g/l); Ton = 12 (s); Toff = 12 (s); IP = 15 (A); SV = 4 (V).
 Determination of Best Input Factors 343

Fig. 2. Comparing the results of MABAC and TOPSIS methods

Acknowledgment. Thai Nguyen University of Technology provided support for this research.

References
1. Ozsahin, D.U., et al.: Evaluation of migraine drugs using MCDM methods. In: Applica-
tions of Multi-Criteria Decision-Making Theories in Healthcare and Biomedical Engineering,
pp. 261–275. Elsevier (2021)
2. Chowdhury, N.K., et al.: Machine learning for detecting COVID-19 from cough sounds: an
ensemble-based MCDM method. Comput. Biol. Med. 145, 105405 (2022)
3. Govindarajan, U., Ramachandaramurthy, V.K., Selvi T, S.O.: Multiple criteria decision mak-
ing (MCDM) based economic analysis of solar PV system with respect to performance
investigation for Indian market. Sustainability 9(5), 820 (2017)
4. Łuczak, A., Just, M.: A complex MCDM procedure for the assessment of economic
development of units at different government levels. Mathematics 8(7), 1067 (2020)
5. Massam, B.H.: Multi-criteria decision making (MCDM) techniques in planning. Prog. Plan.
30, 1–84 (1988)
6. Wu, H.-Y., et al.: Ranking universities based on performance evaluation by a hybrid MCDM
model. Measurement 45(5), 856–880 (2012)
7. Trung, D.D.: Application of EDAS, MARCOS, TOPSIS, MOORA and PIV methods for
multi-criteria decision making in milling process. decision making. 4, 7 (2021)
8. Huu Phan, N., Muthuramalingam, T.: Multi criteria decision making of vibration assisted
EDM process parameters on machining silicon steel using Taguchi-DEAR methodology.
Silicon 13(6), 1879–1885 (2021)
9. Phan, N.H., et al.: Enhancing surface morphology of machined SKD61 die steel in EDM
process using DEAR approach based multi criteria decision making. Int. J. Interact. Des.
Manuf. (IJIDeM) 1–7 (2022)
10. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
344 N. H. Linh et al.

11. Alinezhad, A.: New methods and applications in multiple attribute decision making (MADM).
Springer (2019)
12. Keshavarz-Ghorabaee, M.: Assessment of distribution center locations using a multi-expert
subjective–objective decision-making approach. Sci. Rep. 11(1), 1–19 (2021)
13. Hwang, C.-L., Lai, Y.-J., Liu, T.-Y.: A new approach for multiple objective decision making.
Comput. Oper. Res. 20(8), 889–899 (1993)
14. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
15. Çalışkan, H., et al.: Material selection for the tool holder working under hard milling con-
ditions using different multi criteria decision making methods. Mater. Des. 45, 473–479
(2013)
16. Athawale, V.M., Chatterjee, P., Chakraborty, S.: Selection of industrial robots using com-
promise ranking method. In: Proceedings of the 2010 International Conference on Industrial
Engineering and Operations Management (2010)
 Determination of Optimum Main Design
Parameters of a Two-Stage Helical Gearbox
for Minimum Gearbox Cross-Section Area

Do Quang Khai1 , Tran Huu Danh2 , Bui Thanh Danh3 , Tran Minh Tan4 ,
Nguyen Manh Cuong5 , Le Xuan Hung5 , Vu Ngoc Pi5 , and Tran Thi Phuong Thao5(B)
1 Vietnam Maritime University, Haiphong, Vietnam
2 University of Transport and Communications, Hanoi, Vietnam
3 Vinh Long University of Technology Education, Vinh Long, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

tranphuongthao@tnut.edu.vn

Abstract. The purpose of this research is to determine the optimum main design
parameters for minimizing the cross-section area of a two-stage helical gearbox.
In this work, five main design parameters of the gearbox were taken into account to
find their optimum values, including the gear ratio of the first stage, the coefficient
of wheel face width of stages 1 and 2, and the allowable contact stress of stages
1 and 2. A simulation experiment was designed and implemented by a computer
program to achieve the goal. Furthermore, the Minitab R19 software was used
to analyze the experimental results. The impact of major design factors on cross-
section area was assessed. The optimum values for these parameters were also
suggested.

Keywords: Helical gearbox · Main design parameters · Optimum gearbox

design · Across-section area

1 Introduction
The most important component of a mechanical drive system is the gearbox. Its purpose
is to reduce the speed and torque transmitted from the motor shaft to the working machine
shaft. It is widely used in practice due to its simple structure, stable operation, reliability,
and low cost in comparison to electric, hydraulic, or pneumatic reduction systems. As a
result, the optimal design of the gearbox is a constant concern.
Until now, the optimal design of the gearbox has been carried out on a variety of
subjects. The created model is used in [1] to perform vibration and noise analysis on
the gearbox housing. The authors in [2] introduced an investigation into the potential
for improving energy efficiency and lowering lifecycle costs of electric city buses with
multispeed gearboxes. A two-speed dual clutch gearbox and a continuously variable
transmission were investigated and compared to a reference fixed gear ratio powertrain
in this study. In [3] investigates the optimization of gearbox geometric design parameters

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 345–353, 2023.
https://doi.org/10.1007/978-3-031-22200-9_37
346 D. Q. Khai et al.

to minimize rattle noise in an automotive transmission using an empirical model app-

roach. The selection of the coefficient of wheel face width for multi-speed helical gear
transmissions was introduced in [4]. However, the coefficient values are not optimal;
they were only reasonably chosen based on some advice. In [5], the modal characteristic
of gearbox housing with applied load was investigated. The problem of determining opti-
mum gear ratios has received considerable attention. It was done for helical gearboxes
with two [6, 7], three [8, 9] or four [10] stages. Although there have been numerous
studies on optimizing gearbox parameters, no research has been conducted to identify
the optimal design parameters simultaneously.
This paper introduces a study to determine the optimum main design parameters for
a two-stage helical gearbox. The objective function of the optimization problem is the
minimum gearbox cross-section. To do that, a simulation experiment with the application
of the Taguchi method and the Minitab R19 software was conducted. The influence of
main design parameters on the objective function was evaluated. Optimal values of main
design parameters have been proposed.

2 Methodology
2.1 Determining Gearbox Cross-Section Area

The cross-section area of the gearbox Agb can be calculated by (Fig. 1):

Agb = L · B1 (1)

In which, L, and B1 can be calculated by:

dw11 dw22
L= + aw1 + aw2 + + 20 (2)
2 2

B1 = bw1 + 4 · SG (3)

SG = 0.005 · L + 4.5 [11] (4)

In Eqs. (2) and (3), bw1 is the gear width of the first stage; dw12 , dw21 , and dw22 are
the pitch diameters of the pinion and the gear sets which can be determined by [4]:

dw21 = 2 · aw1 · u1 /(u1 + 1) (5)

dw12 = 2 · aw2 /(u2 + 1) (6)

dw22 = 2 · aw2 · u2 /(u2 + 1) (7)

bw1 = Xba1 · aw1 (8)

 Determination of Optimum Main Design Parameters 347

Fig. 1. Schema for determining cross-section gearbox area

In Eqs. (5)–(8), u1 and u2 are the gear ratios of the first and the second stages;
u2 = ug /u1 ; with ug is gearbox ratio; aw1 and aw2 are the center distances of the first
and the second stages which are calculated by [4]:
  
aw1 = ka · (u1 + 1) · 3 T11 · kH β / AS21 · u1 · Xba (9)
  
aw2 = ka · (u2 + 1) · 3 T12 · kH β2 / AS22 · u2 · Xba2 (10)
where, kH β is the contacting load ratio for pitting resistance; kH β = 1.05 ÷ 1.27 [4]
and it was chosen as kH β = 1.16; AS1 and AS2 is the allowable contact stress of the
first and the second stages (MPa); ka is the material coefficient; As the gear material is
steel, ka = 43 [4]; Xba1 and Xba2 are the wheel face width coefficients of the first and
the second stages; T11 and T12 are the torques on the pinions of the first and the second
stages (Nmm):
 
T11 = Tout / ug · ηhg2
· ηb3 (11)
 
T12 = Tout / u2 · ηhg · ηbe
2
(12)

wherein, Tout is the system output torque (N.mm); ηhg is the efficiency of a helical gear
unit (ηhg = 0.96 ÷ 0.98 [4]; ηb is the efficiency of a rolling bearing pair (ηh = 0.99 ÷
0.995 [4]).
348 D. Q. Khai et al.

2.2 Optimization Problem

Based on the preceding analysis, the optimization problem can be defined as follows:

Minimize Agb (13)

With

Agb = f (u1 ; Xba1 ; Xba2 ; AS 1 ; AS 2 ) (14)

And with the following constraints:

1 ≤ u1 ≤ 9; 1 ≤ u2 ≤ 9 (15)

3 Simulation Experiment

A simulation experiment was carried out to investigate the influence of main design
paremeters on the objective function (across-section area). Besides, the Minitab R19
software and the Taguchi method were applied for experimental design and analysis the
results. In addition, five main design factors including u1, Xba1, Xba2, AS1, and AS2
were investigated. Table 1 describes these parameters and their level.
For the convenience of the survey, the influence of these parameters on the across-
section area was evaluated with each value of the gearbox ratios: 5, 10, 15, 20, 25 and 30.
For each above value of the gearbox ratio, 25 test runs for the simulation experiment were
performed because the 5-level design type of Taguchi design (L25) was selected. Table 2
presents the experimental plan and the output response (the gearbox cross-section area)
when the gearbox ratio is 5.

Table 1. Main design parameters and their levels

Factor Level
1 2 3 4 5
Total gearbox ratio u1 1 3 5 7 9
Coefficient of wheel face width of stage 1 Xba1 0.25 0.27 0.29 0.31 0.33
Coefficient of wheel face width of stage 2 Xba2 0.32 0.34 0.36 0.38 0.4
Allowable contact stress of stage 1 AS1 (Mpa) 350 368 386 404 420
Allowable contact stress of stage 2 AS2 (Mpa) 350 368 386 404 420
 Determination of Optimum Main Design Parameters 349

Table 2. Experimental plan and output response (Agb ) when ugb = 5

Trial u1 Xba1 Xba2 AS1 AS2 Agb (mm2 )

1 1.8 0.25 0.32 350 350 70885.64
2 1.8 0.27 0.34 368 368 64257.65
3 1.8 0.29 0.36 386 386 58581.72
4 1.8 0.31 0.38 404 404 53680.36
5 1.8 0.33 0.4 420 420 49692.26
6 1.9 0.25 0.34 386 404 57562
…
24 2.2 0.31 0.36 368 350 60129.7
25 2.2 0.33 0.38 386 368 54857.51

4 Results Discussion
The Analysis of Variance (ANOVA) method is used with Minitab R19 software to
evaluate the influence of the main design parameters when ugb = 5 on Agb . The signal-
to-noise ratio, or S/N number, must be calculated for each experiment to determine
the effect of each parameter on the output. The calculated S/N number will determine
the best parameters and which parameter has the most influence on the outcome. The
following S/N ratio Eq. (16) should be used to minimize performance characteristics:
 n
1 2
S/N = −10log10 yi (16)
n
i=1

After calculating the SN ratio for each experiment, the average SN ratio for each
parameter and level is calculated. The effect of main design factors on Agb was described
in Table 3 and Fig. 2. From Fig. 2 it is easy to see that Agb is proportional to all five
main design parameters. Besides, Table 3 shows that AS2 has the greatest influence on
Agb (59.19%); followed by Xba2 (22.46 percent) and u1 (10.86%). AS1 and Xba1 have
little effect on Agb (4.62% and 2.81%). Table 4 shows the order of the main parameters’
influence on gearbox across-section area.
To obtain the smallest gearbox cross-section area using the objective function in
Eq. (13), the S/N value is maximized for each main design parameter. As a result of
the analysis of the effect of the parameters on the S/N ratio in the chart in Fig. 3, the
optimum main design parameters were discovered as follows: u1 = 2.22; Xba1 = 0.33;
Xba2 = 0.4; AS1 = 420; AS2 = 420.
For other values of ugb including 10, 15, 20, 25 and 30, proceed in the same way as
above. Table 5 shows the optimal values of main design parameters. From Table 5, the
following observations were reported:
350 D. Q. Khai et al.

Table 3. Analysis of variance for means

Analysis of variance for means

Source DF Seq SS Adj SS Adj MS F P C (%)
u1 4 56879084 56879084 14219771 208.26 0 10.86
Xba1 4 14732884 14732884 3683221 53.94 0.001 2.81
Xba2 4 117683210 117683210 29420803 430.9 0 22.46
AS1 4 24220444 24220444 6055111 88.68 0 4.62
AS2 4 310092572 310092572 77523143 1135.41 0 59.19
Residual error 4 273111 273111 68278 0.05
Total 24 523881305
Model summary
S R-Sq R-Sq(adj)
261.3001 99.95% 99.69%

Fig. 2. Effect of main design factors on Agb

– Both Xba1 and Xba2 optimal values are their maximum values. The reason for this is
that in order to have a small gearbox cross-section, the coefficients Xba1 and Xba2
must be as large as possible in order to reduce the center distance of the gear sets,
resulting in a smaller size L (Eq. (2) and Agb (formula (1)).
– The optimal values of AS1 and AS2 also take their maximum values. The reason is the
same as above, in order to have the minimum gearbox cross-section, the AS1 and AS2
values need to be taken as large as possible to reduce the axial distance of the gear
transmissions, leading to a reduction in the size L (formula (2)) and Agb ((formula (1)).
 Determination of Optimum Main Design Parameters 351

Table 4. The order of influence of the parameters on gearbox across-section area

Response table for means

Level u1 Xba1 Xba2 AS1 AS2
1 59420 58287 60281 58545 62340
2 57863 57434 58505 57735 59353
3 56905 56979 56829 56806 56765
4 55971 56387 55472 56362 54335
5 55072 56142 54144 55782 52438
Delta 4348 2145 6137 2763 9903
Rank 3 5 2 4 1

Fig. 3. Main effects plot for S/N ratios

Table 5. Optimum values of main design parameters

No ugb
5 10 15 20 25 30
u1 2.22 3.72 5.22 6.19 7.5 8.56
Xba1 0.33 0.33 0.33 0.33 0.33 0.33
Xba2 0.4 0.4 0.4 0.4 0.4 0.4
AS1 420 420 420 420 420 420
AS2 420 420 420 420 420 420
352 D. Q. Khai et al.

Fig. 4. Gear ratio of the first stage versus gearbox ratio

– The gear ratios of the first stage and the gearbox have a first-order relationship
(Fig. 4). From this relationship, the following regression formula (with R2 = 0.9947)
is proposed to determine the optimum values of u1 :

u1 = 0.2515 · ugb + 1.1673 (17)

After having u1 , the optimum values of u2 can be easily found by u2 = ugb /u1 .

5 Conclusions
The results of a study on optimization of a two-stage helical gearbox to get the smallest
gearbox cross-section are presented in this paper. In this study, the gear ratio of the first
stage, the coefficient of wheel face width of stages 1 and 2, and the allowable contact
stress of stages 1 and 2 were the five main design parameters of the gearbox that were
optimized. A simulation experiment with Taguchi L25 type was designed and carried
out to solve this work. The effect of main design factors on gearbox cross-section area
was also investigated. Furthermore, the optimum values of the main gearbox parameters
and a regression model to calculate the optimum gear ratios of the first stage have been
proposed as follows: Xba1 = 0.33; Xba2 = 0.4; AS1 = 420 (MPa); AS2 = 420 (MPa);
u1 is calculated by Eq. (17).

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Hajzman, M., Zeman, V.: Noise analysis and optimization of gearboxes. Eng. Mech. 13(2),
117–132 (2006)
2. Ritari, A., et al.: Energy consumption and lifecycle cost analysis of electric city buses with
multispeed gearboxes. Energies 13(8), 2117 (2020)
 Determination of Optimum Main Design Parameters 353

3. Bozca, M., Fietkau, P.: Empirical model based optimization of gearbox geometric design
parameters to reduce rattle noise in an automotive transmission. Mech. Mach. Theory 45(11),
1599–1612 (2010)
4. Chat, T., Van Uyen, L.: Design and Calculation of Mechanical Transmissions Systems, vol.
1. Educational Republishing House, Hanoi (2007)
5. Weis, P., et al.: Modal analysis of gearbox housing with applied load. Proc. Eng. 192, 953–958
(2017)
6. Pi, V.N., Tuan, N.K., Hung, L.X., Tung, L.A.: A new study on determination of optimum
gear ratios of a two-stage helical gearbox. In: Awang, M., Emamian, S.S., Yusof, F. (eds.)
Advances in Material Sciences and Engineering. LNME, pp. 107–115. Springer, Singapore
(2020). https://doi.org/10.1007/978-981-13-8297-0_14
7. Cam, N.T.H., Pi, V.N., Hong, T.T.: A study on calculation of optimum gear ratios of a two-stage
helical gearbox with second stage double gear sets, vol. 500, p. 1 (2019)
8. Vu, N.-P., et al.: The influence of main design parameters on the overall cost of a gearbox.
Appl. Sci. 10(7), 2365 (2020)
9. Pi, V.N., et al.: Determining optimum partial transmission ratios of mechanical driven systems
using a V-Belt drive and a three-stage helical reducer. In: Advances in Material Sciences and
Engineering, pp. 81–88. Springer (2020)
10. Pi, V.N.: Optimal calculation of partial transmission ratios for four-step helical gearboxes with
first and third-step double gear-sets for minimal gearbox length. In: American Conference on
Applied Mathematics (MATH 2008), Harvard, Massachusetts, USA (2008)
11. Römhild, I., Linke, H.: Gezielte Auslegung Von Zahnradgetrieben mit minimaler Masse auf
der Basis neuer Berechnungsverfahren. Konstruktion (1981), 44(7–8), pp. 229–236 (1992)
 Determination of the Location of the Photo-
and Video-Fixation System of Traffic Rules
in the Context of Vietnam

Nguyen Xuan Hien(B) , Vu Thi Van Anh(B) , and Krylov Grigory Alexandrovich(B)

Moscow Automobile and Road Construction State Technical University (MADI), 64,
Leningradsky Avenue, Moscow 125319, Russia
hienmadi@gmail.com, anhvu7587@gmail.com, grigory_a_krylov@mail.ru

Abstract. This paper is about the state of road safety in Vietnam, the accident rate
indicators are given and latest measure for reducing the impact of accidents and
suggestions to increase traffic flow - the use of Photo- and Video- Fixation system
for fixing traffic violations. The purpose of this article is to improve the efficiency
of this system and to improve the quality of traffic flow. In this case, enhancing
the traffic monitoring camera system solves two major problems: increasing road
safety by reducing the number of accident cluster, increasing the effectiveness of
traffic organization by increasing traffic capacity and identifying traffic violations.

Keywords: Road safety · Intelligent transport systems · Photo- and Video-

fixation system of traffic rules

1 Introduction
Vietnam is in a period of strong economic growth. One of the conditions for maintain-
ing the current rates of economic growth is to maintain the efficiency of the country’s
transport and road complexity, where the key area is the creation and maintenance in
the proper (normative) state of a road network that ensures the required level of traffic
safety. The issue of road safety in Vietnam is a problematic one: With approximately
7000 lives cut short yearly due to road traffic accidents and about 20 000–25 000 people
suffer non-fatal injuries, with many living with a disability as a result of their injuries
(Table 1) [1].
In order to improve road safety, reduce road accidents, improve the efficiency of
transport infrastructure, create more comfort for road users, reduce environmental pol-
lution without building new roads, bridges and overpasses in Vietnam, Intelligent Trans-
portation Systems (ITS) is one of the most optimal solutions, which includes the use
of Photo- and Video-Fixation system (PVF) for detecting traffic violations, which is an
urgent task for the current situations of Vietnam [2, 3]. By analyzing foreign experiences,
the authors suggest a technical and design solutions to the problem of installing PVF
system in traffic management in Vietnam.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 354–360, 2023.
https://doi.org/10.1007/978-3-031-22200-9_38
 Determination of the Location of the Photo- and Video-Fixation System … 355

Table 1. Statistics of traffic accidents in Vietnam

Year Number of traffic accidents Injuries Deaths

2015 22474 20970 8442
2016 21589 19280 8417
2017 20080 17040 8279
2018 18736 14802 8248
2019 17626 13624 7624
2020 14510 10804 6700
2021 11495 8018 5799

2 Study Essentials
To maintain the effectiveness of traffic management, constant monitoring of the param-
eters of the traffic flow and traffic conditions is needed, consisting of the following
components:

– Traffic monitoring cameras to identify general trends of the capacity of transport sys-
tem. They determine the sections of the transport network that require changes in
traffic management, depending on the inconsistencies identified, as well as monitor-
ing the effectiveness of the adopted and implemented facility in identifying in real
time accidents clusters and fix their causes. Informations from these cameras play an
important role in resolving existing problems of traffic conflicts.
– Situational surveillance cameras, to identify emerging in real time foci of accidents
and fix the causes of foci with the number of participants in the movement who
provoked them. Informations from these cameras allow to resolve conflicts of road
users using archives of records.
– Traffic violations cameras with the function of transmitting data to traffic departments
for fines registration.

In this case, a mathematical model was developed to definite the effectiveness of the
PVF system on road safety and traffic control, which, in this case, consists of two parts [2]:

– data obtained from the PVF system and data obtained by using the method of expert
assessments to determine significant factors;
– data from the analytic hierarchy method to determine the weighted values of factors
in analyzed intersection of the road network.

According to the objective function formulas, the total weighted value of the inter-
section was calculated before and after the installation of the PVF system, which will
356 N. X. Hien et al.

be carried out based on the number of traffic violations, the value of the ratio parameter
of each factor for each intersection before and after the installation of the PVF system:


N 
u
R= Ki · fin ; (1)
n=1 i=1

R—the total weighted value of the analyzed intersections before the installation of
the PVF system;
K i —weighted value of the ith factor at one intersection;
F in —number of ith violation at the nth intersection;
N—the total number of accident clusters;
n—the accident cluster, n = 1,…,N;
u—the number of traffic violations;
i—traffic violation, i = 1,…,u.
The total weighted value after the installation of the PVF system:


N 
M 
u
R∗ = Ki · fin · (1 − Dim ); (2)
n=1 m=1 i=1

R*—the total weighted value of the analyzed intersections after the installation of
the PVF system;
m—PVF device;
Dim —weighted influence of the mth device on the ith factor.
Maximum efficiency of the installed PVF system under the condition of limited
funding:


N
 
E = R − R∗ = Rn − R∗n
n=1
N u
  
= ϕ(Ki , fin ) − ϕ Ki , fin∗ → max; (3)
n=1 i=1

E—the efficiency of the installed PVF system.

Maximum reduced efficiency of the PVF system installation with maximum
profitability:
N  ∗

E R − R∗ n=1 Rn − Rn
= = M
Sy Sy m=1 Sm
N u   
ϕ(Ki , fin ) − ϕ Ki , fin∗
= n=1 i=1 M → max; (4)
m=1 Sm

S y —the cost of installing the PVF complex at an intersection:


M
S= Sm ≤ S; (5)
m=1
 Determination of the Location of the Photo- and Video-Fixation System … 357

S m —the cost of installing the mth complex PVF at the nth intersection:

Sm = S0m + Smn ; (6)

S 0m —the cost of the mth complex PVF;

S nm —the cost of communication of the mth-complex of the PVF at the nth -
intersection;
M—the number of PVF complexes;
S—the investment amount of the region (city, district) for the implementation of the
PVF system in total.
To obtain unavailable data, the method of expert assessments and the method of anal-
ysis of hierarchies were used [6]. Under the assessment of transport experts, information
of a qualitative characters is suggested and estimated.
In this work, the method of expert assessments is used to obtain a list of significant
factors of those analyzed intersections. The analytic hierarchy method is used to define
the weighted values of factors on the analyzed intersections. When applying these meth-
ods, it is necessary to verify the consensus of opinion among experts assessment to make
sure that the arithmetic average relatives to the true value, then the solution is considered
reliable.

3 Experimental

To conduct an experimental study, sections of the Hanoi City Road Traffic Road were
identified (see Pic. 1) with high traffic problems (centers of traffic accidents, traffic
violations, etc.) and data were collected on traffic violations, which were divided into
two main categories of traffic management: traffic safety and traffic organization.

Pic. 1. Selected analyzed intersections

As a result of analyzing the experts’ agreement, significant traffic violations at each

section of the road traffic system are defined as below (Table 2).
358 N. X. Hien et al.

Table 2. Fixed traffic violations at the selected intersections

Symbol of factor Factors

F1 Exceeding authorized speed limit
F2 Driving at the wrong side of the road
F3 Running red light
F4 Disregarding yield sign
F5 Improper turn/U-turn
F6 Driving in the public transit lane and roadside
F7 Disregarding stop sign
F8 Exit to the intersection of carriageways
F9 Parking and stopping in the wrong place
F10 Violation of road marking requirements
F11 Exit to a pedestrian or bicycle lane

For the experiment, data from 5 different camera complexes are used: Dahua, Hikvi-
sion, Osamic 2400, Vantech, Wisenet SNP. At each critical point, the intensity of the
total weighted value before the installation of the PVF system was determined according
to the formula (1).
The study examines the combinatorial placement of 5 different devices in 20 possible
locations—intersections. The enumeration is the combination of 5 different complexes
at 20 intersections, which gives the results of 15504 different ways of combinatorial
arrangements for PVF systems.
To simplify the process, the requirements for the software according to the research
results, is based on the programming language Microsoft Visual FoxPro 9.0 [7], as well
as for building an analyzing system to determine the optimal choice of the PVF system
were formulated.
Based on the calculations, a list of options was obtained at the output, which is sorted
in descending order by the difference in the indicators, based on which a decision on the
installation of the PVF system can be made (see Figs. 1 and 2).
By analyzing the obtained efficiency and the reduced efficiency, the optimal result
of installing the PVF system was specified. Within the limit of regional budget (for
example 275 thousand dollars), with a reduced efficiency of 0,00349 and an installation
cost of 274230 dollars, a complex Osamic 2400 must be installed at intersection 3,
a complex Wisenet SNP–intersection 4, a complex Dahua–intersection 9, a complex
Vantech–intersection 17, and a complex Hikvision–intersection 18.
 Determination of the Location of the Photo- and Video-Fixation System … 359

Fig. 1. Determination of the maximum efficiency

Fig. 2. Determination of the maximum reduced efficiency

4 Conclusion
Traffic accidents is a social problem throughout the world. The key role of intelligent
transport systems in improving the safety and organization of road traffic has been
proven. Check-up of the state of the application of intelligent transport systems and the
PVF system in Vietnam shows that the practice of implementing ITS in Vietnam has a
number of significant shortcomings and problems, which included: lack of standards for
the installation and use of PVF complexes; lack of principles of synchronization within
the system; limited functionality of PVF complexes, especially in traffic monitoring
and violation determination. The analysis of accident clusters of the road network of
Hanoi city was carried out, among 10 intersections with existing PVF systems and 10
intersections without cameras systems were selected. Based on the identification of
the total weighted value of the effectiveness when installing different combinations of
360 N. X. Hien et al.

the PVF systems, it is determined which type of traffic monitoring cameras should be
installed at each of the analyzed intersections. This encourages further development and
improvement in the approach of this study in below directions:

– development of a system of national standards for ITS as a whole and for the system of
photo and video recording, including in the implementation of scientifically grounded
methods for setting state priorities for the implementation of PVF system;
– development of principles for coordinating the actions of ministries and other execu-
tive authorities to complete the system for collecting and storing data on road safety,
in order to increase the accuracy of analysis and research in this subject area and the
use of reliable methods of calculation and analysis;
– development of software based on modern technical languages, for example, C++,
Python, SQL, which is easy to update and process.

References
1. General Statistics Office of Vietnam (2021). http://www.gso.gov.vn/default.aspx?tabid=721.
Accessed 20 Oct 2021
2. Nguyen, X.H. : To the question of improving the effectiveness of the system of photo-video
fixation of violations of road traffic in order to improve the motion quality. Advanced Science
and Technology for Highways, No 3, pp. 14–17 (2020)
3. Zhankaziev, S., Nguyen X. H.: Analysis of the condition of road traffic by the expert assessment
method. Advanced Science and Technology for Highways, No 1, pp. 7–10 (2019)
4. Nguyen H.D.: ITS development in Vietnam: some lessons learned and measures to ensure the
sustainable efficiency of the system. In: Traffic Safety Conference in Vietnam, Hanoi (2015)
5. Marusin, A.: Methodology for assessing the effectiveness of the functioning of systems for
automatic recording of traffic violations. PhD thesis, Saint-Peterburg, SPBGASU, p. 160 (2017)
6. Klepik, N.: The expert assessments method in the tasks of organizing traffic. Volgograd,
VollgGTU, p. 84 (2012)
7. Klepinin, V.: Visual FoxPro 9.0, Saint-Peterburg, p. 1216 (2008)
 Determining Best Dressing Parameters
for Internal Cylindrical Grinding Using
MABAC Method

Nguyen Hong Linh1 , Trieu Quy Huy2 , Tran Huu Danh3 , Tran Ngoc Huy Thinh4 ,
Bui Thanh Danh5 , Le Xuan Hung6 , Hoang Xuan Tu6 , and Do Thi Tam6(B)
1 Electric Power University, Hanoi, Vietnam
2 University of Economics - Technology for Industries, Hanoi, Vietnam
3 Vinh Long University of Technology Education, Vinh Long, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 University of Transport and Communications, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

dothitam@tnut.edu.vn

Abstract. This paper presents the findings of a study on determining the best
input dressing parameters for internal grinding SKD11 tool steel to achieve min-
imal surface roughness (RS), maximum material removal rate (MRR), and maxi-
mum wheel life (WL) simultaneously. To achive this, the multi-criteria decision-
making (MCDM) problem was solved using the MABAC (Multi-Attributive Bor-
der Approximation Area Comparison) method. In addition, the MEREC (Method
based on the Removal Effects of Criteria) method was applied to compute the
weights of the criteria. Furthermore, six input parameters were investigated includ-
ing coarse dressing depth, coarse dressing passes, fine dressing depth, fine dressing
passes, non-feeding dressing, and dressing feed rate. The experiment was designed
using the Taguchi method and an L16 orthogonal ar-ray. Based on the results of
a study, the best input dressing factors for internal cylindrical grinding have been
proposed.

Keywords: Internal cylindrical grinding · MCDM · MABAC method · MEREC

method · Surface roughness · Material removal rate · Wheel life · SKD11

1 Introduction

MCDM is a popular method for analyzing multiple options and selecting the best one.
This technique is widely used in transportation [1, 2], economics [3, 4], military [5, 6],
social sciences [7, 8], and so on. It has been widely applied in mechanical manufacturing
processes in particular. This is due to the fact that machining processes are frequently
required to meet multiple criteria, such as minimum machined surface roughness (SR),
maximum material removal rate (MMR), maximum cutting tool life, and minimum
manufacturing cost. In fact, machining process criteria normally contradict one another.
For example, the requirement for a marginal surface roughness will result in a drop in the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 361–368, 2023.
https://doi.org/10.1007/978-3-031-22200-9_39
362 N. H. Linh et al.

cut depth and feed rate, lowering the MMR. As a result, many scientists are interested
in solving MCDM problems using various methods.
Till now, a number of studies on using MCDM in machining processes have been
conducted. There have been many studies on solving MCDM problems in milling, turn-
ing, grinding, electric discharge machining, etc. Singh [10] used the TOPSIS method
to determine the best input process parameters in end milling machining of AA6082T6
metal. Trung [9] describes the use of five methods for milling decision-making, includ-
ing EDAS, MARCOS, PIV, MOORA, and TOPSIS. The goal of this research is to find
the best cutting parameters for both low surface roughness and high MRR. The Additive
Ratio Assessment (ARAS) method was used in [10] to determine the best process param-
eters and coated tool for turning AISI 4340 steel. The results of an MCDM study when
powder-mixed electrical discharge machining (PMEDM) cylindrically shaped parts of
90CrSi tool steel were presented in [11]. The MARCOS (Measurement of Alternatives
and Ranking according to Compromise Solution), TOPSIS (Technique for Order of
Preference by Similarity to Ideal Solution), and MAIRCA (Multi-Attributive Ideal-Real
Comparative Analysis) MCDM methods were used in this study. The findings of an
MCDM study to determine the best dressing mode for internal grinding were reported
in [12]. TOPSIS, MARCOS, EAMR, and MAIRCA were chosen as methods for solv-
ing the MCDM problem in this work to obtain the minimum SR and maximum MRR
simultaneously.
The results of a study on the use of the MABAC method for determining the best
dressing mode for internal cylindrical grinding are presented in this paper. Three criteria
were chosen for the work: SR, MRR, and WL. Furthermore, the MEREC method was
used to compute the weights of criteria. The best dressing mode for internal cylindrical
grinding has thus been proposed.

2 Methodology

2.1 Method for Multi-criteria Decision Making

Pamucar and Cirovic pioneered the MABAC method in 2015. The steps for implementing
this method are as follows [13]:
Step 1: Making the initial decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X=⎢
⎢ r1 · · · rij · · · rin ⎥
⎥ (1)
⎢ . . . . . ⎥
⎣ .. . . .. . . .. ⎦
rm1 · · · rmj · · · rmn m×n

wherein, m is the number of options; n is the number of criteria.

Step 2: Calculating the normalized values rij by:

rij − ri−
rij∗ = (2)
ri+ − ri−
 Determining Best Dressing Parameters 363

rij − ri+
rij∗ = (3)
ri− − ri+

In the above Equations, i = 1,2,…,m and j = 1,2,…,n. Equation (2) is used for criteria
MRR and WL, and (3) is used for criterion SR. In addition, ri+ = max(r1 , r2 , . . . , rm )
and ri− = min(r1 , r2 , . . . , rm ).
Step 3: Determining the weighted matrix elements by:

vij = wj + wj × rij∗ (4)

Step 4: Finding the border approximation area matrix:


1/m

m
gj = vij (5)
i=1

In which j = 1,2,…,n.
Step 5: Determining the distance between the alternatives and the border approxi-
mation area by:

qij = vij − gi (6)

In which, i = 1,2,…,m and j = 1,2,…,n.

Step 6: Calculating the total distances of each alternative from the approximate
border area:

n
Si = qij , i = 1, 2, . . . , m (7)
j=1

Step 7: Ranking the options by maximizing S.

2.2 Method for Calculation of Criterion Weight

The MEREC method was used to determine the weight of the criteria in this work. The
following steps should be taken to employ this method [14]:
Step 1: Creating the initial matrix as described in step 1 of the MABAC method:
Step 2: Determining the normalized values by:

– If j is MRR or WL:
minxij
hij = (8)
xij
– If j is SR:

xij
hij = (9)
maxxij
364 N. H. Linh et al.

Step 3: Determining the alternative performance S i by:

⎡ ⎛ ⎞⎤
1   
Si = ln⎣1 + ⎝ ln hij ⎠⎦ (10)
n
j

Step 4: Calculating the performance of ith alternative Si.j ’ by:

⎡ ⎛ ⎞⎤
 1    
Sij = Ln⎣1 + ⎝ ln hij ⎠⎦ (11)
n
k,k=j

Step 5: Finding the removal effect of the jth criterion Ej by:

  

Ej = Sij − Si  (12)
i

Step 6: Calculating the criation weight by:

Ej
wj =  (13)
k Ek

3 Experimental Work
An experiment was carried out to solve the multi-objective optimization problem. It was
designed using the Taguchi method and the L16 orthogonal array (44 × 22 ). The input
factors and their levels are shown in Table 1. The experimental setup is demonstrated in
Fig. 1, and Tables 2 display the input dressing factors and their levels. The experiment
for the dressing process was carried out using the experimental matrix illustrated in
Table 2. After the dressing process, the grinding wheel has been used to grind test
samples at the following grinding regimes: wheel speed of 12000 (rpm); workpiece
speed of 150 (rpm); radial wheel speed of 0.0025 (mm/stroke); and axial feed speed
of 1 (mm/minute). Following the completion of the experiments, the SR (in this case,
Ra (µm)) was measured and the MRR (mm3 /s), and WL (minute) were calculated. The
output results are shown in the two right columns of Table 2.

4 Determining the Best Dressing Parameters

This section describes how to solve the MCDM problem with the MABAC method and
calculate criteria weights with the MEREC method to determine the best input dressing
parameters for internal grinding SKD11 tool steel.

4.1 Determining the Criation Weights

The following steps (see Sect. 2.2) can be taken when using the MEREC method for
estimating the weights of criteria: determining Si and Sij ’ using Eqs. (10) and (11),
calculating the criterion removal effect using Eq. (12), and determining the weight of
the criteria wj using Eq. (13). Ra , MRR, and Tw were also discovered to have weights
of 0.3184, 0.3198, and 0.3618, respectively.
 Determining Best Dressing Parameters 365

Fig. 1. Experimental setup for internal grinding

Table 1. Input dressing parameters

No Input parameters Levels

1 2 3 4
1 Coarse dressing depth (mm) 0.025 0.03 – –
2 Coarse dressing passes (times) 1 2 3 4
3 Fine dressing depth (mm) 0.005 0.01 0.015 0.02
4 Fine dressing passes (times) 0 1 2 3
5 Non-feeding dressing (times) 0 1 2 3
6 Dressing feed rate (mm/min) 1 1.2 – –

4.2 Determining the Best Input Dressing Parameters

The steps for performing MCDM using the MABAC method are performed as discussed
in Sect. 2.1. First, Eq. (1) is used to generate the initial matrix. Then, Eq. (2) is applied
to determine the normalized values of rij *, and Eq. (3) is utilized to create SR. The
normalized weighted values vij are then computed using a formula (4). Later, Eq. (5)
is used to compute the border approximation area matrix. The distance between the
alternatives and the border approximation area qij is then calculated using the formula
(6). Finally, using the formula (7), the total distances between each option and the
approximate border area Si are calculated. To determine the ranking of options, S is
maximized. Table 3 displays several calculated parameters as well as the ranking of
options.
366 N. H. Linh et al.

Table 2. Experimental plan and output results

No ar nr af nf n0 Sd Ra (µm) MRR (mm3 /s) Tw (min.)

1 0.025 1 0.005 0 0 1 0.3652 0.9186 0.9186
2 0.03 1 0.01 1 1 1 0.214 1.0413 12.733
3 0.025 1 0.015 2 2 1.2 0.195 1.1215 12.697
4 0.03 1 0.02 3 3 1.2 0.242 1.1111 11.423
5 0.03 2 0.005 1 2 1.2 0.185 1.2459 11.803
6 0.025 2 0.01 0 3 1.2 0.248 1.2667 12.430
7 0.03 2 0.015 3 0 1 0.252 1.1782 11.147
8 0.025 2 0.02 2 1 1 0.217 1.2251 12.077
9 0.03 3 0.005 2 3 1 0.306 1.4878 12.547
10 0.025 3 0.01 3 2 1 0.324 1.3724 12.077
11 0.03 3 0.015 0 1 1.2 0.341 1.2709 12.027
12 0.025 3 0.02 1 0 1.2 0.354 1.1988 12.503
13 0.025 4 0.005 3 1 1.2 0.318 1.2273 12.177
14 0.03 4 0.01 2 0 1.2 0.313 1.2647 10.620
15 0.025 4 0.015 1 3 1 0.326 1.1247 13.153
16 0.03 4 0.02 0 2 1 0.363 1.1898 12.230

Option 3 is observed to be the best option in Table 3 due to the fact that it has the
highest Si value (Si = 0.1383). As a result, the best dressing mode includes the following
input factors: ar = 0.025 (mm); nr = 1 (times); af = 0.015 (mm); nf = 2 (times); n0 =
2 (times); Sd = 1.2 (m/minute).

5 Conclusions

The results of using the MABAC method to solve the MCDM problem and the MEREC
method to calculate the criteria weights in internal grinding SKD11 tool steel are pre-
sented in this paper. To accomplish this, an experiment with an L16 orthogonal array
was designed and analyzed using the Taguchi method (44 x22 ). The study findings result
in the best input dressing factors to get the minimum surface roughness, the maximum
material removal rate, and the maximum wheel life as follows: ar = 0.025 (mm); nr =
1 (times); af = 0.015 (mm); nf = 2 (times); n0 = 2 (times); Sd = 1.2 (m/minute).
 Determining Best Dressing Parameters 367

Table 3. Several calculated results and ranking of options

No gij qij Si Rank

Ra MRR Tw Ra MRR Tw
1 0.3164 0.4741 0.4075 0.0020 −0.1543 0.3161 0.1638 16
2 0.3164 0.4741 0.4075 −0.0018 −0.0853 −0.0333 −0.1204 8
3 0.3164 0.4741 0.4075 −0.0022 −0.0403 −0.0322 −0.0747 7
4 0.3164 0.4741 0.4075 −0.0011 −0.0461 0.0054 −0.0417 6
5 0.3164 0.4741 0.4075 −0.0025 0.0296 −0.0058 0.0213 5
6 0.3164 0.4741 0.4075 −0.0009 0.0413 −0.0243 0.0161 4
7 0.3164 0.4741 0.4075 −0.0008 −0.0084 0.0136 0.0044 3
8 0.3164 0.4741 0.4075 −0.0017 0.0180 −0.0139 0.0024 2
9 0.3164 0.4741 0.4075 0.0005 0.1656 −0.0278 0.1383 1
10 0.3164 0.4741 0.4075 0.0010 0.1007 −0.0139 0.0878 15
11 0.3164 0.4741 0.4075 0.0014 0.0437 −0.0124 0.0327 14
12 0.3164 0.4741 0.4075 0.0017 0.0032 −0.0265 −0.0216 13
13 0.3164 0.4741 0.4075 0.0008 0.0192 −0.0169 0.0032 12
14 0.3164 0.4741 0.4075 0.0007 0.0402 0.0292 0.0701 11
15 0.3164 0.4741 0.4075 0.0010 −0.0385 −0.0457 −0.0832 10
16 0.3164 0.4741 0.4075 0.0020 −0.0019 −0.0184 −0.0184 9

Acknowledgment. Thai Nguyen University of Technology provided support for this research.

References
1. Nassereddine, M., Eskandari, H.: An integrated MCDM approach to evaluate public
transportation systems in Tehran. Transp. Res. Part A Policy Pract. 106, 427–439 (2017)
2. Kiciński, M., Solecka, K.: Application of MCDA/MCDM methods for an integrated urban
public transportation system–case study, City of Cracow. Arch. Transp. 46 (2018)
3. Govindarajan, U., Ramachandaramurthy, V.K., Selvi T, S.O.: Multiple criteria decision mak-
ing (MCDM) based economic analysis of solar PV system with respect to performance
investigation for Indian market. Sustainability 9(5), 820 (2017)
4. Łuczak, A., Just, M.: A complex MCDM procedure for the assessment of economic
development of units at different government levels. Mathematics 8(7), 1067 (2020)
5. Sánchez-Lozano, J., Serna, J., Dolón-Payán, A.: Evaluating military training aircrafts through
the combination of multi-criteria decision making processes with fuzzy logic. A case study
in the Spanish Air Force Academy. Aerosp. Sci. Technol. 42, 58–65 (2015)
6. Sennaroglu, B., Celebi, G.V.: A military airport location selection by AHP integrated
PROMETHEE and VIKOR methods. Transp. Res. Part D Transp. Environ. 59, 160–173
(2018)
368 N. H. Linh et al.

7. Massam, B.H.: Multi-criteria decision making (MCDM) techniques in planning. Prog. Plan.
30, 1–84 (1988)
8. Wu, H.-Y., et al.: Ranking universities based on performance evaluation by a hybrid MCDM
model. Measurement 45(5), 856–880 (2012)
9. Trung, D.D.: Application of EDAS, MARCOS, TOPSIS, MOORA and PIV methods for
multi-criteria decision making in milling process. Decis. Mak.4, 7 (2021)
10. Singaravel, B., Shankar, D.P., Prasanna, L.: Application of MCDM method for the selection
of optimum process parameters in turning process. Mater. Today Proc. 5(5), 13464–13471
(2018)
11. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
12. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
13. Alinezhad, A.: New methods and applications in multiple attribute decision making (MADM).
Springer (2019)
14. Keshavarz-Ghorabaee, M.: Assessment of distribution center locations using a multi-expert
subjective–objective decision-making approach. Sci. Rep. 11(1), 1–19 (2021)
 Determining Best Input Parameters for CBN
Grinding Al6106 T6 Using WASPAS Method

Do Quang Khai1 , Tran Huu Danh2 , Bui Thanh Danh3 , Nguyen Manh Cuong4 ,
Hoang Xuan Tu4 , and Nguyen Van Trang4(B)
1 Vietnam Maritime University, Haij i Phòng, Vietnam
2 Vinh Long University of Technology Education, Vinh Long, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nvtrang@tnut.edu.vn

Abstract. This paper describes the findings of a multi-criteria decision-making

(MCDM) study in CBN grinding Al6061 T6 to determine the best input process
parameters to achieve both minimal surface roughness (SR) and maximum mate-
rial removal speed (MRS). In order to do this, the WASPAS (Weighted Aggregates
Sum Product Assessment) method was used to solve the MCDM problem, and
the weights of criteria were calculated using the MEREC (Method based on the
Removal Effects of Criteria) method. Furthermore, an experiment was carried out
with four process input parameters containing wheel speed (WS), feed rate (Fe ),
depth of cut (aed ), and down feed (Df ). The experiment was designed by a 2-level
1/2 fraction type. The experimental output results, including the SR and MRS,
were measured and supplied into the MCDM problem. Based on the findings of
this study, the best input parameters for CBN grinding Al6061 T6 were proposed.

Keywords: CBN grinding · MCDM · WASPAS method · MEREC method ·

Surface roughness · Material removal speed · Al6061 T6

1 Introduction

CBN wheels are made of cubic boron nitride, which is second only to diamond in terms
of hardness. They are increasingly used in mechanical processing because of their high
abrasion resistance and thermal conductivity, as well as their ability to maintain sharp
cutting edges. As a result, the study of CBN grinding is a very interesting topic.
There have been numerous studies on CBN grinding up to this point. Jackson et al. [1]
reported on current research and future developments in the field of high-speed grinding,
as well as the development of high-speed CBN camshaft grinding. A comparative analy-
sis of the performance of two types of CBN grinding wheels in nodular cast iron plunge
grinding was presented in [2]. The impact of the metal removal rate on the selection of
abrasive in terms of wear and power consumed during grinding was reported in [3]. It
was noted that the abrasive used in the study has a significant impact on the grinding
process’s economics. The experimental and modeling characterization of wear and life

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 369–375, 2023.
https://doi.org/10.1007/978-3-031-22200-9_40
370 D. Q. Khai et al.

expectancy of electroplated CBN grinding wheels were investigated in [4]. The authors
in [5] investigated the effects of the cutting factors, including the cutting depth, the feed
rate, and the wheel speed on the grinding time when CBN grinding tablet shape punches
on a CNC (Computerized Numerical Control) milling machine. Besides, [6] introduced
a comparison study on the grindability of Ti-6Al-4V using CBN and diamond brazed
type monolayered grinding wheels in different environments. The influence of CBN grit
shape on the grindability of a low alloy chrome steel, 100Cr6, was shown in [7]. The
authors in [8] investigated the effect of different grinding fluids applied in a minimum
quantity cooling-lubrication mode on surface integrity in CBN grinding of Inconel 718.
A study on MCDM in CBN grinding 9CrSi tool steel was reported in [9]. The determi-
nation of the best dressing input parameters using TOPSIS method for CRN cylindrical
grinding En-31 samples was introduced in [10].
According to the results of the above analysis, although there have been various opti-
mization studies or MCDM studies to find the optimum or best input grinding parameters,
there is still a lack of MCDM studies to find the best input grinding factors for CBN
grinding Al6061 T6 to achieve both minimum SR and maximum MRS.
This paper introduces a study on MCDM in CBN grinding Al6061 T6, in which
two criteria, including the minimum SR and maximum MRS were taken into account.
Also, the WASPAS method was selected to solve the MCDM problem and the MEREC
method was used to find the weights of criteria. From the results of the study, the best
input factors for CBN grinding Al6061 T6 were proposed.

2 Methodology

2.1 Method for Multi-criteria Decision Making

The WASPAS method was used to solve the MCDM problem in this work. To use this
method, the following steps need to be taken [11]:
Step 1: Creating the first decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X=⎢
⎢ r1 · · · rij · · · rin ⎥
⎥ (1)
⎢ . . . . . ⎥
⎣ .. . . .. . . .. ⎦
rm1 · · · rmj · · · rmn m×n

In which, m is the alternative number; n is the criterion number.

Step 2: Determining the normalized values of the decision matrix by:
rij
r∗ij = (2)
max rij
i
min rij
r∗ij =
i
(3)
rij
 Determining Best Input Parameters for CBN 371

Noted that Eq. (2) is used for SR and the other is used for MRS.
Step 3: Finding the additive relative importance in the weighted normalized data of
each alternative:

n
Ai = rij∗ · wj (4)
j=1

where, wj is the weight of criterion j; i = 1 ÷ m.

Step 4: Calculating the multiplicative relative importance in the weighted normalized
data of each alternative:
n

wj
Mi = r∗ij (5)
j=1

Step 5: Determining the joint generalizad criterion Qi by:

Qi = λ · Ai + (1 − λ) · Mi (6)

Step 6: Ranking the options by sorting Qi in a descending order and the best option
has the highest rank.

2.2 Method for Determination of Criterion Weights

In this study, the MEREC method is used to determine the weights of the criteria. This
method can be used by following the steps [12]:
Step 1: Making the initial matrix by Eq. (1).
Step 2: Determining the normalized values by:
minxij
hij = (7)
xij
xij
hij = (8)
maxxij

wherein, Eq. (11) is used for criterion RS and Eq. (12) is used for criterion MRS.
Step 3: Calculating the alternative performance S i by:
⎡ ⎛ ⎞⎤
1    
Si = ln⎣1 + ⎝ ln hij ⎠⎦ (9)
n
j

Step 4: Finding the performance of ith alternative S ij by:

⎡ ⎛ ⎞⎤
 1    
Sij = Ln⎣1 + ⎝ ln hij ⎠⎦ (10)
n
k,k=j
372 D. Q. Khai et al.

Step 5: Determining the removal effect of the jth criterion E j by:

  

Ej = Sij − Si  (11)
i

Step 6: Determining the criterion weight by the following Equation:

Ej
wj =  (12)
k Ek

3 Experimental Work

An experiment was carried out to solve the MCDM problem. This experiment, which had
a 2-level 1/2 fraction design, was created using the Minitab R19 software. As a result,
there were 24−1 = 8 test runs. The input factors and their levels are described in Table 1.
Figure 1 illustrates the experimental setup, which includes a CNC milling machine
(Model M-V50C - Japan), a CBN grinding wheel, CBN wheel dresser equipment (V-
TDM-2 type), Al6061 workpiece samples, and a roughness meter (SURFTEST SV-
3100). Table 2 describes the experimental plan. The values of SR were measured and
those of MRS were calculated after the experiment. The average results of the SR and
MRS are given in the first and the second columns from the right of Table 2.

Fig. 1. Experimental setup

4 Determining the Best Input Parameters for CBN Grinding

Al6061 T6

This section describes how to choose the best experimental setup for CBN grinding
Al6061 T6 by solving the MCDM problem with the WASPAS method and estimating
criterion weights with the MEREC method.
 Determining Best Input Parameters for CBN 373

Table 1. Input parameters

Factor Unit Low High

Wheel speed (WS) rpm 4000 5000
Feed rate (Fe) mm/min 2000 3000
Depth of cut (aed ) mm 0.01 0.025
Down feed (Df ) mm/double journey 0.5 2

Table 2. Experimental plans and output results

Run order Rpm Fe aed Df SR (µm) MRS (g/h)

1 4000 2000 0.01 0.5 0.1497 3.4779
2 4000 3000 0.025 0.5 0.1510 4.3989
3 5000 2000 0.025 0.5 0.1257 3.5631
4 5000 3000 0.01 0.5 0.1300 4.0997
5 5000 2000 0.01 2 0.3377 10.6490
6 5000 3000 0.025 2 0.1900 16.2297
7 4000 2000 0.025 2 0.1857 11.6366
8 4000 3000 0.01 2 0.1673 13.9366

4.1 Calculating the Criation Weights

To calculate the weights for the criteria using the MEREC method, it was recommended
to follow the steps outlined in Sect. 2.2: Determine Si and Sij ’ using Eqs. (9 and 10);
calculate the criterion removal effect using Eq. (11); and find the weight of the criteria wj
using Eq. (12). Furthermore, the weights of RS and MRS were discovered to be 0.5358
and 0.4642, respectively.

4.2 Determining the Best Input Parameters

The WASPAS method was used in this work to solve the MCDM problem in the main
procedures (see Sect. 2.1). Equation (2) for RS and (3) for MRS are used to compute the
normalized values of the decision matrix rij * . Then, using Eq. (4), the additive relative
importance of each alternative in the weighted normalized data was calculated. Then,
using formula (5), the multiplicative relative importance in the weighted normalized data
was determined. Equation (6) is then used to calculate the joint generalized criterion Qi .
Finally, to rank the alternatives, Qi was sorted in descending order. Table 3 displays sev-
eral calculated results as well as the ranking of the alternative. Besides, Fig. 2 describes
the relation between solutions and the values of Qi .
From Table 3 and Fig. 2, it is reported that the best option is option 6. This is because
it has the highest Qi value (Qi = 0.500). Therefore, the best input process parameters for
374 D. Q. Khai et al.

CBN grinding Al6061 T6 are: WS = 4000 (rpm), Fe = 3000 (mm/min.), aed = 0.025
(mm), and Df = 2 (mm/double journey).

Table 3. Several calculated results and rankings of options

Solution Wij Ri Pi Qi Rank

Ra MRR
1 0.412 0.109 0.521 0.045 0.283 8
2 0.408 0.138 0.546 0.056 0.301 6
3 0.491 0.112 0.603 0.055 0.329 5
4 0.474 0.129 0.603 0.061 0.332 4
5 0.183 0.334 0.517 0.061 0.289 7
6 0.325 0.509 0.834 0.165 0.500 1
7 0.322 0.365 0.697 0.121 0.409 3
8 0.369 0.437 0.806 0.161 0.484 2

Fig. 2. Relation between solutions and the values of Qi

5 Conclusions
This paper describes the results of a study on MCDM in CBN grinding Al6061 T6 using
the WASPAS method to solve the MCDM problem and the MEREC method to calculate
the criterion weights. To achieve this, an experiment with a 2-level 1/2 fraction design
was conducted. From the results of the study, several conclusions were given:

– The WASPAS method was used for the first time to solve the MCDM problem in CBN
grinding Al6061 T6.
 Determining Best Input Parameters for CBN 375

– The best input process parameters for CBN grinding Al6061 T6 to get minimum SR
and maximum MRS were found: WS = 4000 (rpm), Fe = 3000 (mm/min.), aed =
0.025 (mm), and Df = 2 (mm/double journey).

Acknowledgment. Thai Nguyen University of Technology provided support for this work.

References
1. Jackson, M., et al.: High-speed grinding with CBN grinding wheels—applications and future
technology. J. Mater. Process. Technol. 110(1), 78–88 (2001)
2. de Martini Fernandes, L., et al.: Comparative analysis of two CBN grinding wheels perfor-
mance in nodular cast iron plunge grinding. Int. J. Adv. Manuf. Technol. 98(1–4), 237–249
(2018). https://doi.org/10.1007/s00170-018-2133-4
3. Jackson, M.J., Mills, B.: Materials selection applied to vitrified alumina & CBN grinding
wheels. J. Mater. Process. Technol. 108(1), 114–124 (2000)
4. Yu, T., Bastawros, A.F., Chandra, A.: Experimental and modeling characterization of wear
and life expectancy of electroplated CBN grinding wheels. Int. J. Mach. Tools Manuf 121,
70–80 (2017)
5. Vu, N.-P., et al.: Optimization of grinding parameters for minimum grinding time when
grinding tablet punches by CBN wheel on CNC milling machine. Appl. Sci. 9(5), 957 (2019)
6. Teicher, U., et al.: Performance of diamond and CBN single-layered grinding wheels in
grinding titanium. Mater. Manuf. Process. 23(3), 224–227 (2008)
7. Macerol, N., Franca, L.F., Krajnik, P.: Effect of the grit shape on the performance of vitrified-
bonded CBN grinding wheel. J. Mater. Process. Technol. 277, 116453 (2020)
8. Naskar, A., et al.: Effect of different grinding fluids applied in minimum quantity cooling-
lubrication mode on surface integrity in cBN grinding of Inconel 718. J. Manuf. Process. 36,
44–50 (2018)
9. Thinh, H.X., Trung. D.D.: A research on application of the measurement of alternatives and
ranking according to compromise solution method for multi-criteria decision making in the
grinding process. EUREKA: Phys. Eng. (2), 101–110 (2022)
10. Patil, S.S., Bhalerao, Y.J.: Selection of levels of dressing process parameters by using TOPSIS
technique for surface roughness of en-31 work piece in CNC cylindrical grinding machine.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2017)
11. Alinezhad, A.: New methods and applications in multiple attribute decision making (MADM).
Springer (2019)
12. Keshavarz-Ghorabaee, M.: Assessment of distribution center locations using a multi-expert
subjective–objective decision-making approach. Sci. Rep. 11(19461), 2021 (2021). https://
doi.org/10.1038/s41598-021-98698-y
 Determining Partial Gear Ratios of a Two-Stage
Helical Gearbox with First Stage Double Gear
Sets for Minimizing Total Gearbox Cost

Tran Huu Danh1 , Trieu Quy Huy2 , Bui Thanh Danh3 , Tran Minh Tan4 ,
Nguyen Van Trang5 , and Luu Anh Tung5(B)
1 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
2 University of Economics - Technology for Industries, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

luuanhtung@tnut.edu.vn

Abstract. The partial gear ratio is an essential factor in gearbox design because it
is directly related to the gearbox’s volume, size, and cost. A method for determining
the optimum gear ratio u1 of a two-stage helical gearbox with the first-stage double
gear set to achieve the lowest cost was presented in this work. To accomplish this,
a simulation experiment was designed and run by an optimization program with
the goal of minimizing gearbox costs. Ten input factors were considered in the
experiment to determine their influence on the optimal gear ratio of the first stage
u1 . Finally, the proposed regression model is validated by calculating the difference
between the experiment and response u1 using the residual evaluation distribution.

Keywords: Helical gearbox · Partial gear ratio · Optimum gearbox design ·

Gearbox cost

1 Introduction

A gearbox is a mechanical device that reduces the torque and speed transmitted from the
motor shaft to the working machine shaft. It is widely used in practice due to its simple
structure, stable operation, reliability, and low cost compared to electric, hydraulic, or
pneumatic reduction systems. As a consequence, optimal gearbox design is a major
concern.
To date, many studies on the optimal design of the gearbox have been conducted,
such as identifying the appropriate parameters of the gearbox’s gears [1, 2], designing the
gearbox housing [3, 4], investigating the reliability of gears [5, 6], finding the optimal
partial gear ratios [7, 8], etc. An optimization study for designing a tillage machine
gearbox for getting a minimum center diameter was introduced in [9]. Furthermore,
many researchers have worked to determine the best gear ratios. The reason for this is
that the gear ratio has a significant impact on the weight, size, and cost of the gearbox.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 376–388, 2023.
https://doi.org/10.1007/978-3-031-22200-9_41
 Determining Partial Gear Ratios of a Two-Stage Helical Gearbox 377

The optimum partial gear ratios have been found for different gearboxes, including
helical gearboxes [7, 10, 11], bevel gearboxes [8], and worm gearboxes [12–14]. With
helical gearboxes, the partial gear ratios have been calculated for two-stage [7, 15], three-
stage [16, 17] and four-stage [18] gearboxes. Besides, the gear ratios have been found
for helical gearboxes with the first stage [10, 19] or second stage [15, 20, 21] double
gear sets. The optimal gear ratios were found for the two-stage helical gearbox with first
stage double gear sets with many objective functions, including the minimum gearbox
length [11, 22], the minimum gearbox cost [19], or the minimum gearbox cross-section
area [10]. However, no study has been conducted to date with the objective function of
minimizing gearbox costs with the bearing cost component.
This paper presents a cost optimization study to find the optimal partial gear ratios of
a two-stage helical gearbox with the first-stage double gear sets. In this study, the impact
of the input parameters on the optimal gear ratio of the first stage u1 was investigated.
Moreover, a regression model to find u1 as an explicit equation was suggested.

2 Methodology
2.1 Determination of Gearbox Cost
The cost of a two-stage helical gearbox with first stage double gear sets Cgb is calculated
by (Fig. 1):

Cgb = Cg + Cs + Cgh + Cb (1)

In which, Cg , Cs , Cb , and Cgh are the cost of gears, the cost of shafts, the cost of
bearings, and the cost of gearbox housing which can be determined by:

Cg = Cg,m · mg (2)

Cs = Cs,m · ms (3)

Cgh = Cgh,m · mgh (4)

n
Cb = C (5)
i=1 b,i

where, Cg,m , Cs,m , Cgh,m are the cost per a kilogram of gears, sahfts, and gearbox housing
(USD/kg). Cb,i is the cost of a pair of bearing (USD/pair). mg, ms, and mgh are the mass
of gears, shafts, and gearbox housing (kg). These mass components will be calculated
in the following subsections.

2.2 Determination of Mass of Gears

The mass of gears in the gearbox can be found by:

mg = mg1 + mg2 (6)

378 T. H. Danh et al.

A1

SG

1.5SG

H
10 10
SG

5SG
L

SG
dw11 dw21

1.5SG
1a III A2
I II

dsb1 dsb3
dsb2
L2
bW1

aW1 aW2
bW2
B1
L3

L1
B
SG

2
bW1

1.5SG

dw12 dw22
2a
SG

Fig. 1. Calculated schema

where mg1 and mg2 are the mass of gears of stages 1 and 2 which are determined by the
following Equations:
 
π · e1 · dw11
2 ·b
w1 π · e2 · dw21
2 ·b
w1
mg1 = ρg · + (7)
4 4
 
π · e1 · dw21
2 ·b
w2 π · e2 · dw22
2 ·b
w2
mg2 = ρg · + (8)
4 4

In which, ρg = 7.82 (kg/dm3 ) is the density of gear material (for steel gears); e1 =
1 and e2 = 0.6 are the volume coefficient of stages 1 and 2, respectively; bw1 nad bw2
are the gear widths of stages 1 and 2.
bw1 = Xba1 · aw1 (9)

bw2 = Xba2 · aw2 (10)

In Eqs. (9) and (10), Xba1 and Xba2 are the wheel face width coefficients of stages
1 and 2. aw1 and aw2 are the center distances of stages 1 and 2 which are detemined by
 Determining Partial Gear Ratios of a Two-Stage Helical Gearbox 379

[23]:

aw1 = ka · (u1 + 1) · 3
T11 · kH β /(AS21 · u1 · Xba ) (11)

aw2 = ka · (u2 + 1) · 3 T12 · kH β2 /(AS22 · u2 · Xba2 ) (12)

where, T11 and T12 are the torque on the pinions of stage 1 and stage 2:
 
T11 = Tout / 2 · ug · η2hg · ηb3 (13)
 
T12 = Tout / u2 · ηhg · ηbe
2
(14)

wherein, Tout is the system output torque (N.mm); ηhg = 0.95 ÷ 0.97 is the efficiency
of a helical gear set [23]; ηb = 0.99 ÷ 0.995 is the efficiency of a rolling bearing pair
[23].

2.3 Determination of Mass of Shafts

The mass of shafts of the gearbox are determined by:

ms = ms1 + ms2 + ms3 (15)

where, ms1 , ms2 and ms3 are the mass of shafts 1, 2 and 3 of the gearbox which can be
calculated by:

ms1 = ρs · π · ds1
2
· ls1 /4 (16)

ms2 = ρs · π · ds2
2
· ls2 /4 (17)

ms1 = ρs · π · ds2
2
· ls2 /4 (18)

wherein, ds1 , ds2 , and ds3 are the diameters of shaft 1, 2, and 3 (mm); ls1 , ls2 , and ls3 are
the lengths of shaft 1, 2 and 3 of the gearbox (mm). These diameters and lengths can be
found by:

ds1 = [T11 /(0.2 · [τ ])]1/3 (19)

ds2 = [T12 /(0.2 · [τ ])]1/3 (20)

ds3 = [Tout /(0.2 · [τ ])]1/3 (21)

In which, [τ] = 15 ÷ 20 is the allowable shear stress (MPa) [23].

380 T. H. Danh et al.

2.4 Determination of Mass of Gearbox Housing

The gearbox housing mass of the gearbox is calculated by:

mgh = ρgh · Vgh (22)

where, ρgh = 7.2 (kg/dm3 ) is the density of the gearbox housing material; Vgh is the
gearbox housing volume (m3 ) which can be found by:

Vgh = 2 · (Vb + VA1 + VA2 ) (23)

In which, Vb , VA1 , VA2 are the volume of the bottom of the gearbox housing, the
volumes of sides A1 , and A2 respectively. These volume components are calculated by:

Vb = L · B · SG (24)

VA1 = L · (H − 2 · SG ) · SG (25)

VA2 = (H − 2 · SG ) · B1 · SG (26)

In the Eqs. (24)–(26), L, H, B1 and SG can be determined by:

dw21 dw12
L = dw11 + + + dw22 + 20 (27)
2 2

H = max(dw21 , dw22 ) + 8.5 · SG (28)

B1 = 2 · bw1 + bw2 + 5 · SG (29)

B = B1 + 2 · SG = 2 · bw1 + bw2 + 7 · SG (30)

SG = 0.005 · L + 4.5[24]. (31)

2.5 Determination of Bearing Cost

The cost of bearings of the gearbox (USD) was calculated by [25]:

Cb,i = kcb · Cb,mi (32)

In which, Cb,mi is the cost of a medium-sized deep groove ball bearing on the ith
shaft which has the inner diameter of 5Ne,i ; Cb,mi can be found by [25]:

Cb = 0.2707 · Ne,i
2
− 1.2566 · Ne,i + 3.3056 (33)

where, kcb is the bearing selling coefficient (kcb < 1).

 Determining Partial Gear Ratios of a Two-Stage Helical Gearbox 381

2.6 Optimization Problem

Based on the preceding analysis, the optimization problem can be described as follows:

Minimize Cgb (34)

With the following constraints:

1 ≤ u1 ≤ 9; 1 ≤ u2 ≤ 9. (35)

3 Experimental Work
To minimize costs, estimate the influence of the input variables on the reducer’s gear ratio
u1 , a simulation experiment was carried out. Table 1 shows a simulation experiment with
ten input variables and selected survey levels. Besides, the partial experimental planning
with 210–3 = 128 experimental runs was used to reduce the number of experiments while
still assessing the influence of the input variables. Also, the experiment was designed and
analyzed using the Minitab R18 software. The experimental planning matrix is described
in Table 2.

Table 1. Input parameters and their levels

No. Parameters Code Unit Level

Low High
1 Total gearbox ratio ug – 7 30
2 Coefficient of the face width of the first gear stage Xba1 – 0.25 0.3
3 Coefficient of wheel face width of the second gear stage Xba2 – 0.3 0.45
4 Allowable contact stress of stage 1 AS1 MPa 350 420
5 Allowable contact stress of stage 2 AS2 MPa 350 420
6 Output torque Tout Nm 5000 50000
7 Cost of a kilogram of gearbox housing Cgh USD/kg 1 5
8 Cost of a kilogram of gears Cg USD/kg 5 9
9 Cost of a kilogram of shafts Cs USD/kg 1.5 5
10 Bearing selling price coefficient kcb – 0.7 1

4 Results and Discussions

Figure 2 shows the effect of input factors on the optimum gear ratio of the first stage u1 .
From this Figure, it was noted that u1 is proportional to ug , AS1 , Tout , Cgh , Cg and it is
inversely proportional to AS2 , CS , and Kcb . Moreover, ug is the most influential factor
382 T. H. Danh et al.

Table 2. Experimental matrix and output results (u1 )

No. ug Xba1 Xba2 AS1 AS2 Tout Cgh Cg Cs kcb u1

1 7 0.3 0.45 350 420 5000 5 5 1.5 0.7 1.32
2 7 0.3 0.45 350 350 50000 5 5 5 1 1.38
3 30 0.25 0.3 420 350 5000 1 5 1.5 1 4.87
4 30 0.3 0.3 350 420 5000 1 9 5 0.7 4.39
5 30 0.3 0.45 420 350 50000 5 9 1.5 1 6.46
6 30 0.25 0.45 420 420 5000 1 9 1.5 0.7 5.39
…
127 7 0.3 0.3 420 420 5000 5 9 1.5 0.7 1.59
128 7 0.3 0.45 350 420 5000 1 9 1.5 0.7 1.45

on u1 ; followed by AS2 , Cs , Tout , and Kcb . In addition, Xba1 and Xba2 had no significant
effect on u1 .
The influence of input variables and their interactions on u1 can also be known
through the Pareto chart (Fig. 3). The length of the blue column indicates the variables’
influence level. Factors with magnitudes greater than the red reference line have a sig-
nificant effect on u1 , with a significance level of = 0.05. From this graph, it was reported
that A (ug ), H (Cg ), E (AS2 ), J (Cs ), F (Tout ), K (Kcb ), D (AS1 ), G (Cgh ), B (Xba1 ), and
the interactions AE, AH, AF, AD, AK, FJ, and DH have a significant effect on u1 .
The normal graph of normalized effects (Fig. 4) can be used to determine the trend
and influence of the input variables, as well as their interaction. The variables denoted
in red squares on this graph have a significant effect on u1 . The variables that are further
away from the standard line have a greater impact. Variables to the right of the standard
line, including A, H, F, G and the interactions AH, AF, DF, EJ, JK, HK have a positive
effect on u1 . Variables to the left of the standard line, such as: E, J, K and the interactions
AE, AJ, AK have a negative effect on u1 .
Using Minitab 18 software, a regression equation for determining u1 was proposed
after removing the parameters with little or no impact on u1 . Table 3 describes the
estimated coefficients of the input variables as well as their interactions for u1 .
It was found from the Table that the R-sq and R-sq (adj) are very high (99.93% and
99.91%). That proves that the model is well-matched with the data. The proposed model
to calculate the optimum gear ratio of stages 1 is in the form as follows:

u1 = 0.600 + 0.22996ug + 1.175Xba1 − 1.475Xba2 + 0.00038AS1

+ 0.00165AS2 − 0.000004 Tout + 0.00395 Cgh − 0.1032 Cg
− 0.0860 Cs + 0.439 kcb − 0.0644 ug ∗ Xba1 + 0.000157 ug ∗ AS1
− 0.000323 ug ∗ AS2 + 0.000000 ug ∗ Tout + 0.000764 ug ∗ Cgh
+ 0.004881 ug ∗ Cg − 0.005415 ug ∗ Cs − 0.02378 ug ∗ kcb − 0.252 Xba1 ∗ Cs
+ 0.00432 Xba2 ∗ AS1 − 0.000006 Xba2 ∗ Tout − 0.000009 AS1 ∗ AS2
 Determining Partial Gear Ratios of a Two-Stage Helical Gearbox 383

Fig. 2. The effect of input parameters on u1

Fig. 3. Pareto chart of standardized effects for u1

+ 0.000000 AS1 ∗ Tout + 0.000381 AS1 ∗ Cg − 0.002634 AS1 ∗ kcb

+ 0.000297 AS2 ∗ Cs − 0.000000 Tout ∗ Cg − 0.000001 Tout ∗ Cs
+ 0.000006 Tout ∗ kcb + 0.0604 Cs ∗ kcb (36)

The proposed model is confirmed by determining the deviation between the trial
and u1 through the residual evaluation distribution plot (Fig. 5). From this figure, it
can be seen that the errors (the blue dots) are distributed very close to the normal
distribution (solid lines). This is also achieved through the frequency plot - most errors
384 T. H. Danh et al.

Fig. 4. Normal distribution chart of effect of input parameters and interactions for u1

are close to zero. The other two graphs show the relationship between the error and the
corresponding values of the regression model, both randomly distributed. It proves that
u1 is not significantly affected outside of survey parameters. In addition, the adjustment
coefficients Rsq and Rsq_adj are both greater than 99% (Table 3). That shows that the
computational regression model is suitable.

5 Conclusions

This paper presents a cost optimization study of a two-stage helical gearbox with a first-
stage double gear set for finding the optimal partial gear ratios u1 . To do that, a simulation
experiment that has ten input parameters was designed and performed. The effect of the
input parameters on u1 was evaluated. It was reported that ug has the strongest effect
on u1 . Also, the influence of other factors and their interactions on u1 was investigated.
Moreover, a regression model was proposed for calculation of u1 . All of the values of
R-sq are more than 99%. That allows to conclude that the proposed regression model is
suitable.
 Determining Partial Gear Ratios of a Two-Stage Helical Gearbox 385

Table 3. Estimated coefficient of input parameters and their interaction for u1

Coded coefficients
Term Effect Coef SE coef T-value P-value VIF
Constant 3.26289 0.00504 647.20 0.000
ug 3.56172 1.78086 0.00504 353.24 0.000 1.00
Xba1 −0.04172 −0.02086 0.00504 −4.14 0.000 1.00
Xba2 0.00297 0.00148 0.00504 0.29 0.769 1.00
AS1 0.18172 0.09086 0.00504 18.02 0.000 1.00
AS2 −0.47391 −0.23695 0.00504 −47.00 0.000 1.00
Tout 0.36172 0.18086 0.00504 35.87 0.000 1.00
Cgh 0.07234 0.03617 0.00504 7.17 0.000 1.00
Cg 0.49047 0.24523 0.00504 48.64 0.000 1.00
Cs −0.39141 −0.19570 0.00504 −38.82 0.000 1.00
kcb −0.19922 −0.09961 0.00504 −19.76 0.000 1.00
ug *Xba1 −0.03703 −0.01852 0.00504 −3.67 0.000 1.00
ug *AS1 0.12641 0.06320 0.00504 12.54 0.000 1.00
ug *AS2 −0.25984 −0.12992 0.00504 −25.77 0.000 1.00
ug *Tout 0.14641 0.07320 0.00504 14.52 0.000 1.00
ug *Cg h 0.03516 0.01758 0.00504 3.49 0.001 1.00
ug *Cg 0.22453 0.11227 0.00504 22.27 0.000 1.00
ug *Cs −0.21797 −0.10898 0.00504 −21.62 0.000 1.00
ug *kcb −0.08203 −0.04102 0.00504 −8.14 0.000 1.00
Xba1 *Cs −0.02203 −0.01102 0.00504 −2.18 0.031 1.00
Xba2 *AS1 0.02266 0.01133 0.00504 2.25 0.027 1.00
Xba2 *Tout −0.02047 −0.01023 0.00504 −2.03 0.045 1.00
AS1 *AS2 −0.02172 −0.01086 0.00504 −2.15 0.034 1.00
AS1*Tout 0.03891 0.01945 0.00504 3.86 0.000 1.00
AS1 *Cg 0.05328 0.02664 0.00504 5.28 0.000 1.00
AS1 *kcb − 0.02766 − 0.01383 0.00504 − 2.74 0.007 1.00
AS2 *Cs 0.03641 0.01820 0.00504 3.61 0.000 1.00
Tout *Cg − 0.03609 − 0.01805 0.00504 − 3.58 0.000 1.00
Tout *Cs − 0.06359 − 0.03180 0.00504 − 6.31 0.000 1.00
Tout *kcb 0.03797 0.01898 0.00504 3.77 0.000 1.00
Cs *kcb 0.03172 0.01586 0.00504 3.15 0.002 1.00
(continued)
386 T. H. Danh et al.

Table 3. (continued)

Coded coefficients
Term Effect Coef SE coef T-value P-value VIF
Model summary
S R-sq R-sq(adj) R-sq(pred)
0.0570382 99.93% 99.91% 99.88%

Fig. 5. The distribution charts evaluate the residuals

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Reddy, B.S., Mahato, K.K.: Calculation, design and analysis of two stage single speed gearbox
for all terrain vehicle for baja sae. Mater. Today: Proc. (2021)
2. Marjanovic, N., et al.: A practical approach to the optimization of gear trains with spur gears.
Mech. Mach. Theory 53, 1–16 (2012)
3. Weis, P., et al.: Modal analysis of gearbox housing with applied load. Procedia Eng. 192,
953–958 (2017)
4. Sonsino, C.: Structural durability of cast aluminium gearbox housings of underground railway
vehicles under variable amplitude loading. Int. J. Fatigue 27(8), 944–953 (2005)
5. Nejad, A.R., Gao, Z., Moan, T.: On long-term fatigue damage and reliability analysis of gears
under wind loads in offshore wind turbine drivetrains. Int. J. Fatigue 61, 116–128 (2014)
 Determining Partial Gear Ratios of a Two-Stage Helical Gearbox 387

6. Nejad, A.R., Gao, Z., Moan, T.: Fatigue reliability-based inspection and maintenance planning
of gearbox components in wind turbine drivetrains. Energy Procedia 53, 248–257 (2014)
7. Pi, V.N., Tuan, N.K., Hung, L.X., Tung, L.A.: A new study on determination of optimum
gear ratios of a two-stage helical gearbox. In: Awang, M., Emamian, S.S., Yusof, F. (eds.)
Advances in Material Sciences and Engineering. LNME, pp. 107–115. Springer, Singapore
(2020). https://doi.org/10.1007/978-981-13-8297-0_14
8. Pi, V.N., et al.: Determination of optimum gear ratios of a three stage bevel helical gearbox.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2019)
9. He, H.-B., et al.: Optimal design of a tilling machine reduction gearbox using Matlab. Int. J.
Precis. Eng. Manuf. 10(2), 63–66 (2009)
10. Pi, V.N., et al.: Calculating optimum gear ratios of a two-stage helical reducer with first
stage double gear sets. In: IOP Conference Series: Materials Science and Engineering. IOP
Publishing (2019)
11. Hung, L.X., et al.: Calculation of optimum gear ratios of mechanical driven systems using
two-stage helical gearbox with first stage double gear sets and chain drive. In: International
Conference on Engineering Research and Applications. Springer (2019)
12. Tung, L.A., et al.: A study on determination of optimum gear ratios of a two-stage worm
gearbox. In: International Conference on Engineering Research and Applications. Springer
(2019)
13. Ky, L.H., et al.: Calculation of optimum gear ratios of mechanical driven systems using worm-
helical gearbox and chain drive. In: Sattler, K.-U., Nguyen, D.C., Vu, N.P., Tien Long, B.,
Puta, H. (eds.) ICERA 2019. LNNS, vol. 104, pp. 66–75. Springer, Cham (2020). https://doi.
org/10.1007/978-3-030-37497-6_7
14. Pi, V.N., Tuan, N.K., Hung, L.X., Dung, N.T.Q., Hien, B.T.: Determining optimum gear
ratios of a worm—helical gearbox for minimum acreage of the cross section. In: Awang,
M., Emamian, S.S., Yusof, F. (eds.) Advances in Material Sciences and Engineering. LNME,
pp. 89–96. Springer, Singapore (2020). https://doi.org/10.1007/978-981-13-8297-0_12
15. Cam, N.T.H., PI, V.N., Hong, T.T.: A study on calculation of optimum gear ratios of a two-stage
helical gearbox with second stage double gear sets. a a, 500, 1 (2019)
16. Vu, N.-P., et al.: The influence of main design parameters on the overall cost of a gearbox.
Appl. Sci. 10(7), 2365 (2020)
17. Pi, V.N., et al.: Determining Optimum Partial Transmission Ratios of Mechanical Driven
Systems Using a V-Belt Drive and a Three-Stage Helical Reducer. In: Advances in Material
Sciences and Engineering, pp. 81–88. Springer (2020)
18. Pi, V.N.: Optimal calculation of partial transmission ratios for four-step helical gearboxes with
first and third-step double gear-sets for minimal gearbox length. In: American Conference on
Applied Mathematics (MATH 2008), Harvard, Massachusetts, USA (2008)
19. Pi, V.N., et al.: Cost optimization study of two-step helical gearboxes with first stage double
gear sets. Int. J. Mech. Prod. Eng. Res. Dev. (IJMPERD) 10, 573–584 (2020)
20. Linh, N.H., et al.: Cost optimization of two-stage helical gearboxes with second stage double
gear-sets. EUREKA: Phys. Eng. 2021(6), 89–101 (2021)
21. Cam, N.T.H., et al.: Calculating optimum gear ratios of mechanical drive systems using
two-stage helical gearbox with second-stage double gear sets and chain drive for minimum
gearbox length. In: Sattler, K.-U., Nguyen, D.C., Vu, N.P., Tien Long, B., Puta, H. (eds.)
ICERA 2019. LNNS, vol. 104, pp. 155–163. Springer, Cham (2020). https://doi.org/10.1007/
978-3-030-37497-6_18
22. Van Cuong, N., Le Hong, K., Tran, T.H.: Splitting total gear ratio of two-stage helical reducer
with first-stage double gearsets for minimal reducer length. Int. J. Mech. Prod. Eng. Res. Dev.
(IJMPERD) 9(6), 608, 595 (2019)
23. Chat, T., Van Uyen, L.: Design and calculation of Mechanical Transmissions Systems, vol.
1. Educational Republishing House, Hanoi (2007)
388 T. H. Danh et al.

24. Römhild, I., Linke, H.: Gezielte Auslegung Von Zahnradgetrieben mit minimaler Masse auf
der Basis neuer Berechnungsverfahren. Konstruktion (1981), 44(7–8), 229–236 (1992)
25. Quang, N.H., et al.: Optimizing the partial gear ratios of the two-stage worm gearbox for
minimizing total gearbox cost. East.-Eur. J. Enterp. Technol. 1(1), 115 (2022)
 Determining the Best Alternative
in Powder-Mixed Electrical Discharge
Machining 90CrSi Tool Steel Using the TOPSIS
Method

Nguyen Anh Tuan1 , Trieu Quy Huy1 , Tran Huu Danh2 , Bui Thanh Danh3 ,
Do Thi Tam4 , Nguyen Van Tung4 , and Le Xuan Hung4(B)
1 University of Economics - Technology for Industries, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

lexuanhung@tnut.edu.vn

Abstract. This paper presents the results of a multi-criteria decision-making

(MCDM) study in powder-mixed electrical discharge machining (PMEDM) cylin-
drically shaped parts made of 90CrSi tool steel. In this study, five input factors
including powder concentration, pulse duration, pulse off time, pulse current, and
servo voltage were selected for the investigation. Besides, the Taguchi method
was applied for the experimental design. Furthermore, the TOPSIS (Technique
for Order of Preference by Similarity to Ideal Solution) method was used to solve
the MCDM problem in order to achieve both a low electrode wear rate (EWR)
and a high material removal speed (MRS). The Entropy method was also used to
compute the weights of the criteria. Based on the findings, the best solution for
the MCDM problem in PMEDM cylindrically shaped parts was proposed.

Keywords: PMEDM · MCDM · TOPSIS method · Electrode wear rate ·

Material removal speed · 90CrSi

1 Introduction

Multi-criteria decision making (MCDM) is used to evaluate various options in order

to select the best one. It is widely used in practice, not only in medicine, business, and
social sciences but also in mechanical processing. This is because the machining process
is frequently required to meet multiple criteria, many of which contradict each other.
For example, increasing the MRS will result in an increase in cut depth and feed rate, as
well as an increase in surface roughness and a decrease in cutting tool life. As a result,
many scientists are interested in solving MCDM problems in mechanical machining
processes.
Up to now, different MCDM methods have been applied to mechanical machining
processes. The TOPSIS method and ANOVA analysis have been used to find the best

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 389–394, 2023.
https://doi.org/10.1007/978-3-031-22200-9_42
390 N. A. Tuan et al.

cutting parameters in finishing the milling of Ti-6Al-4V titanium alloy with minimum
quantity lubricant (MQL) [1]. The TOPSIS method was used in [2] to select an industrial
robot for the arc welding operation. The end-milling operation of AISI 1522H steel grade
under MQL conditions was investigated in [3] using a novel blend of vegetable oils,
namely canola and olive oil. A combination method for multi-criteria decision making
problem in turning process was reported in [4]. To rank the options in this study, eight
MCDM methods were used, including SAW, WASPAS, TOPSIS, VIKOR, MOORA,
COPRAS, PIV, and PSI. To solve the MCDM problem in the external cylindrical grinding
process of 65G steel, the Weighted Aggregates Sum Product ASsessment (WASPAS)
and Proximity Indexed Value (PIV) methods were used [5]. The results of the MCDM
problem in the dressing process for internal grinding using the combinative distance-
based assessment (CODAS) method were introduced in [6]. The authors of [7] announce
the findings of an MCDM study performed on cylindrically shaped parts made of 90CrSi
tool steel using powder-mixed electrical discharge machining (PMEDM).
The results of an MCDM study in PMEDM cylindrically shaped parts are presented
in this paper. Minimum EWR and maximum MRS were chosen as investigation criteria
in the study. The TOPSIS method was also used to solve the MCDM problem, and the
Entropy method was used to calculate the weights of the criteria. The best option for
obtaining minimum EWR and maximum MRS at the same time was suggested.

2 Methodology
2.1 Method for Multi-criteria Decision Making
The TOPSIS method was used to handle the MCDM problem in this study.
To use this method, the following steps need to be taken [8]:
Step 1: Creating initial decision-making matrix:
⎡ ⎤
x11 · · · x1n
⎢ x21 · · · x2n ⎥
⎢ ⎥
X =⎢ . . ⎥ (1)
⎣ .. · · · .. ⎦
xmn · · · xmn
In which, m is the alternative number; n is the criterion number; x mn is the value of
the criterion n in the alternative m.
Step 2: Calculating the normalized values kij by
xij
kij =  (2)
m 2
i=1 xij

Step 3: Finding the weighted normalized decision matrix by the following equation:

lij = wj × kij (3)

Step 4: Determining the best alternative A+ and the worst s alternative A− by:


A+ = l1+ , l2+ , . . . , lj+ , . . . , ln+ (4)
 Determining the Best Alternative in Powder-Mixed Electrical Discharge 391


A− = l1− , l2− , . . . , lj− , . . . , ln− (5)

where, lj+ and lj− are the best and worst values of the j criterion (j = 1,2,…, n).
Step 5: Calculating the values of better options Di+ and worse options Di− by the
following equations:

n  2
+
Di = lij − lj+ (6)
j=1

n  2
Di− = lij − lj− (7)
j=1

In the above equations i = 1, 2, …, m.

Step 6: Determining closeness coefficient values Ri of each alternative by:
Di−
Ri = (8)
Di− + Di+
where, i = 1, 2, …, m; 0≤ Ri ≤ 1
Step 7: Ranking the order of options by maximizing the value of R.

2.2 Method for Determining the Weight of Criteria

The Entropy method was used to calculate the weights of the criteria in this work.
The steps outlined below can be used to implement this method [9]:
Step 1: Determining the indicator’s normalized values by:
xij
pij = m 2 (9)
m + i=1 xij
Step 2: Determining the Entropy for each indicator by the following Equation:
m     m   m 
mej = − pij × ln pij − 1 − pij × ln 1 − pij (10)
i=1 i=1 i=1
Step 3: Finding the weight of each indicator by:
1 − mej
wj = m  . (11)
j=1 1 − mej

3 Experimental Setup
An experiment was carried out in order to solve the proposed MCDM problem. Table 1
shows the experiment’s input process parameters. The Taguchi method with L18 (21 +
34) design was also selected for the experiment. Figure 1 demonstrates the experimental
setup. A Sodick A30 EDM machine (Japan), graphite electrodes (TOKAI Carbon Co.,
LTD, Tokyo, Japan), workpieces made of 90CrSi tool steel (China), 100 nm SiC powder
(China), and Total Diel MS 7000 dielectric solution (France) were used in the experiment.
Following the completion of the experiments, the EWR were measured and the MRS
were calculated. Table 2 shows the experimental matrix as well as the output results
(EWR and MRS).
392 N. A. Tuan et al.

Fig. 1. Experimental setup

Table 1. Input factors and their levels

No. Input parameters Unit Level

1 2 3
1 Powder concentration Cp g/l 0 0.5 1
2 Pulse on time Ton µs 8 12 16
3 Pulse off time Toff µs 8 12 16
4 Peak current Ip A 5 10 15
5 Servo voltage SV V 4 5 –

4 Determining the Best Alternative

This section discusses about the MCDM problem when the TOPSIS method is used and
the criteria weights are calculated using the Entropy method.

4.1 Calculating the Weights for the Criteria

The weights of criteria are determined using the Entropy method (see Sect. 2.2): The
normalized values pij are calculated using Eq. (9). The Entropy value for each indicator
mej was then calculated using Eq. (10). At last, using Eq. (11), compute the wj weight
of the criteria. The weights of EWR and MRS were calculated to be 0.3678 and 0.6322,
respectively.

4.2 Determining the Best Alternative by Using TOPSIS Method

Section 2.1 explains how to solve the MCDM problem using the TOPSIS method. As a
consequence, Eq. (2) is used to determine normalized kij values, while Eq. (3) is used
to predict normalized weighted lij values. The A+ and A− values of EWR and MRR
are determined by Eqs. (4) and (5). EWR and MRS are 0.0173 and 0.3020 for A+ ,
 Determining the Best Alternative in Powder-Mixed Electrical Discharge 393

Table 2. Experimental plan and output results

No. Input parameters Output results

Cp Ip Ton Toff SV EWR (g/h) MRS (g/h)
1 0.0 8 8 5 4 22.3049 0.7306
2 0.0 12 12 10 4 25.5724 1.3441
3 0.0 16 16 15 4 33.0783 3.8894
4 0.5 8 8 10 4 74.0077 0.7992
5 0.5 12 12 15 4 78.9780 7.0694
6 0.5 16 16 5 4 16.9566 0.8136
7 1.0 8 12 5 4 114.3392 1.3978
8 1.0 12 16 10 4 30.2637 3.8204
9 1.0 16 8 15 4 34.3570 4.4114
10 0.0 8 16 15 5 143.7481 6.9694
11 0.0 12 8 5 5 33.6556 1.9435
12 0.0 16 12 10 5 13.4498 1.4282
13 0.5 8 12 15 5 148.4736 7.1003
14 0.5 12 16 5 5 28.2640 2.3283
15 0.5 16 8 10 5 16.5338 1.2479
16 1.0 8 16 10 5 83.4345 2.0881
17 1.0 12 8 15 5 88.0550 7.1046
18 1.0 16 12 5 5 17.6557 2.5645

and 0.1913 and 0.0311 for A− , respectively. Moreover, the Di + and Di − values were
computed using Formulas (6) and (7). Finally, the ratio Ri was calculated using Eq. (8).
Table 3 displays the results of using the TOSIS method to calculate several parameters
and rank the alternatives.
According to Table 3, option 5 is the best option. This is due to the fact that it has the
highest closeness coefficient value (R5 = 0.7708). As a result, the best solution includes
the following parameters: Cp = 0.5 (g/l), Ton = 12, Toff = 12, Ip = 12 (A), and SV = 4
(V).

5 Conclusions
The article presents the findings of an MCDM study performed on PMEDM cylindrical
shaped parts made of 90CrSi tool steel. In this work, the TOPSIS method was used to
solve the MCDM problem, and the Entropy was used to calculate the weights of the
criteria. In addition, 100 nm SiC powder was mixed into the Diel MS 7000 dielectric. In
addition, five process factors were selected for the investigation: powder concentration,
pulse on time, pulse off time, pulse current, and servor voltage. The experiment was
394 N. A. Tuan et al.

Table 3. Calculated results and ranking of options

Trial kij lij Di + Di − Ri Rank

EWR MRS EWR MRS
1 0.0781 0.0492 0.0287 0.0311 0.2712 0.1626 0.3748 15
2 0.0896 0.0904 0.0330 0.0572 0.2453 0.1605 0.3955 13
3 0.1159 0.2617 0.0426 0.1654 0.1389 0.2004 0.5907 6
4 0.2593 0.0538 0.0954 0.0340 0.2791 0.0960 0.2559 17
5 0.2767 0.4756 0.1018 0.3007 0.0845 0.2841 0.7708 1
6 0.0594 0.0547 0.0219 0.0346 0.2674 0.1695 0.3880 14
… … … … … … … … …
17 0.3085 0.4780 0.1135 0.3022 0.0961 0.2821 0.7458 2
18 0.0619 0.1725 0.0228 0.1091 0.1930 0.1858 0.4904 8

designed using the Taguchi method with L18 (21 + 34 ) design. It has been reported that
the best input factor configuration for obtaining the lowest EWR and the highest MRS
simultaneously when processing cylindrically shaped parts is Cp = 0.5 (g/l); Ton = 12
(s); Toff = 12 (s); Ip = 12 (A), and SV = 4 (V).

Acknowledgments. This work was financially supported by the scientific project No. B2021-
TNA-04.

References
1. Nguyen, V., Nguyen, T., Tien, D.: Cutting parameter optimization in finishing milling of Ti-6Al-
4V titanium alloy under MQL condition using TOPSIS and ANOVA analysis. Eng. Technol.
Appl. Sci. Res. 11(1), 6775–6780 (2021)
2. Chodha, V., et al.: Selection of industrial arc welding robot with TOPSIS and entropy MCDM
techniques. Mater. Today: Proc. 50, 709–715 (2022)
3. Abas, M., et al.: Parametric study and optimization of end-milling operation of AISI 1522H
steel using definitive screening design and multi-criteria decision-making approach. Materials
15(12), 4086 (2022)
4. Do, D.T.: A combination method for multi-criteria decision making problem in turning process.
Manuf. Rev. 8 (2021)
5. Trung, D.D.: The combination of Taguchi–Entropy–WASPAS–PIV methods for multi-criteria
decision making when external cylindrical grinding of 65G steel. J. Mach. Eng. 21 (2021)
6. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
7. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
8. Hwang, C.-L., Lai, Y.-J., Liu, T.-Y.: A new approach for multiple objective decision making.
Comput. Oper. Res. 20(8), 889–899 (1993)
9. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Viet. J. Agricult. Sci. 17(4), 32–2331 (2019)
 Develop an Advanced Driver’s Behaviors
Detection System

Tung-Duong Pham, Hai-Hung Bui, Nhat-Linh Nguyen-Bui, Dinh-Tien Nguyen,

Trong-Phuoc Le, Chi-Quan Le, Thanh-Tung Nguyen(B) , and Hoang-Phuc Dam

Department of Vehicle and Energy Conversion Engineering, School of Mechanical Engineering,

Hanoi University of Science and Technology, Hanoi, Vietnam
tung.nguyenthanh@hust.edu.vn

Abstract. Due to the rising number of car accidents involving drowsy or dis-
tracted drivers, it is essential to develop a system to detect these drivers’ behaviors.
Multiple strategies were proposed and evaluated to determine whether the driver
is in those driving states. Based on data collected from a surveillance camera and
onboard diagnosis system, our research group has developed a method and system
to recognize those abnormal behaviors and alert the driver to focus on driving. An
onboard computer is used to evaluate the potential of the application system in an
actual driving situation. The result shows that the system could operate effectively.

Keywords: Behavior detection · Driver drowsiness · Driver distraction ·

Computer vision · ITS

1 Introduction

Driver behavior detection is a car safety technology that prevents accidents when the
driver is lacking concentration or getting drowsy. According to the National Safety Coun-
cil [1], 2.4 percent of all crashes (or over 2,400 crashes) are caused by drowsy driving
annually. In a report published in 2022 by National Highway Traffic Safety Admin-
istration (NHTSA) [2], 8.1 percent of all crashes are caused by distracted driving. For
these reasons, automotive manufacturers have been working on driver behavior detection
using different techniques such as Driver Alert Control from Volvo or Attention Assist
System introduced by Mercedes-Benz. In order to implement the system on vehicles,
the system needs to detect abnormal behaviors effectively at a reasonable price. Previ-
ous studies have applied the one of main approaches such as behavior, vehicular data,
or physiological data [3]. In order to create a more reasonable product with the actual
conditions in Vietnam, we have produced a suitable system.
In this study, our research group proposes a system that uses a hybrid strategy between
the behavioral and vehicular approaches to detect the driver’s drowsiness and inattention
patterns. The system will alert the driver to focus back on driving by giving an audible
signal. It can detect drivers who are drowsy or distracted driving through camera image
processing and compare them with a normal threshold. The threshold is calculated by

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 395–401, 2023.
https://doi.org/10.1007/978-3-031-22200-9_43
396 T.-D. Pham et al.

estimating head position, detecting eye blink, yawing action, and steering wheel move-
ment. The driving behaviors from on-board data provide more information to increase
the accuracy of the algorithm. The system was implemented and tested inside a vehicle
cabin to check its effectiveness and ease of use. The system has been deployed and
tested. The results show that this system can potentially help to reduce traffic accidents
and save the lives of many people by giving warnings at the right moments.

2 Methods
2.1 Perspective-N-Point (PnP) and Face Landmark

Perspective-N-Point problem is an issue in computer vision. It estimates the pose of

a calibrated camera from n 3D point coordinates and their corresponding 2D [4]. The
camera pose has 6 degrees of freedom (DOF) which consist of 3 DOFs of translation and
3 DOFs of rotation. This problem has applications in many fields of computer vision. In
the proposed system, the PnP problem is used to estimate the head pose.
⎡ ⎤ ⎡ ⎤⎡ ⎤⎡ ⎤ ⎡ ⎤
u r11 r12 r13 fx 0 cx X tx
s⎣ v ⎦ = ⎣ r21 r22 r23 ⎦⎣ 0 fy cy ⎦⎣ Y ⎦ + ⎣ ty ⎦ (1)
1 r31 r32 r33 0 0 1 Z tz

where the left part of equation is the 2D imagine coordinates, the first element of the
right part is the intrinsic properties (optical center), the second element is the extrin-
sic properties (camera rotation and translation) and the last element is the 3D world
coordinates.
Face landmark detection is a computer vision task used to detect and track the key
point on a human face. MediaPipe [5] is a collection of libraries, pertained models,
and methods to work with human face features. The pipeline for processing media
information (video/image flow) is built as a result of MediaPipe’s solution to the issue of
468 facial landmarks that are grouped in fixed quads and identified by their coordinates
(x, y, and z).

2.2 Head Pose Estimation

The main objective of this task is to find the relationship between the position of the
human’s head and the camera. It means detecting the head’s Euler angles. The Euler
angles include: yaw, pitch, and roll. The rotation of an object in 3D can be described
with these three values.
The PnP problem and the Face landmark identification challenge must be combined
to obtain these three values. 6 points from the face mesh (showed in Fig. 1), and their
corresponding 3D and 2D coordinates are obtained. The remaining points are two at the
corners of the eyes, one on the nose, two at the corners of the mouth, and one on the chin.
The internal camera matrix is established by the values of the focal length of the camera
and the image’s center coordinates. After having three input matrices, the human head’s
Euler angles are computed by using the PnP Eq. (1).
 Develop an Advanced Driver’s Behaviors Detection System 397

Fig. 1. MediaPipe Face Mesh Mask: (left) points to determine the Head pose, (right) points to
determine eyes and mouth

2.3 Eye and Mouth Detection

Eye detection is needed for symptom extraction. There are many methods to detect eyes,
MediaPipe FaceMesh provides information about this.
In some systems, lips are detected by filtering out red color, but this method is not
reliable in some low light conditions. Therefore, feature points extracted from MediaPipe
FaceMesh for mouth detection is our approach (as shown in Fig. 1).

2.4 On-Board Diagnostics Data

On-board diagnostics (OBD) is a term referring to a vehicle’s self-diagnostic and report-
ing capability. Modern cars have the OBD system which provides for a standardized
hardware interface with the female 16-pin (2x8) J1962 connector. The system uses
Python - OBD [6] which is a python module for handling real-time sensor data from
OBD-II vehicle ports. Data is streamed through an OBD-II adapter, and stored in the
Raspberry Pi. The driving data (turning signals, speed, steering wheel movement, etc.)
obtained from OBD allows the system more flexibility to set the warning conditions.

2.5 Inattention Detection

The driver’s heading conditions are categorized as: “Looking Forward”, “Looking Left”,
“Looking Right”, “Looking Up”, and “Looking Down”. Based on these data, the system
will warn the driver if the driver does not look forward and focus on driving. In case
of turning, the drivers have to look at the wing mirrors without looking forward to a
period that is longer than common, the system can acknowledge this by reading OBD
data of steering angle. Thus, the previous warning time is longer, and the system can
detect inattention more precisely.

2.6 Drowsiness Detection

The Steering Wheel Movement (SWM) is a method to detect drowsy states from driving
behavior. This index can be calculated from OBD data. In general, driving task character-
istics (e.g. speed, curvature, and lane width), driver attributes (e.g. driving experience),
398 T.-D. Pham et al.

and driver states affect steering behavior (e.g. laxness, distraction, or fatigue). Drivers
constantly assess the situation in front of them and make small, smooth steering adjust-
ments to adapt to minor road obstacles and crosswinds. According to [7], sleep-deprived
drivers made fewer steering wheel reversals than average drivers (Fig. 2).

Fig. 2. Steering Movement-Based Detection

Indices extracted from the eyes and mouth used in the system are Eye Aspect Ratio
(EAR) and Mouth Aspect Ratio (MAR). The eye aspect ratio is the estimation of the
eye-opening state. As mentioned in Sect. 2.3, the system will detect 6 points on each eye
(see Fig. 3). Then, it applies the EAR Eq. (2) to determine the problem [8]. Similarly,
the estimation of the mouth-opening state [9] is also determined by 6 points (Fig. 4) with
the following formula:

of Vertical Distance
EAR(MAR) = (2)
2 × Horizontal Distance of Eye
By conducting many experiments, the threshold values of EAR and MAR are esti-
mated for each individual and general use. For every frame captured, those values were
computed and timed the duration. If the duration is lower than the threshold [10], the
system will give a warning.

Fig. 3. Eye Aspect Ratio with 6 points on each eye

 Develop an Advanced Driver’s Behaviors Detection System 399

Fig. 4. Mouth Aspect Ratio with 6 points on each eye

3 Experiments
3.1 Experimental Setup
The experiment includes 3 main following devices: Webcam 1080P Rapoo C260 is used
as a surveillance camera, Raspberry Pi 3 is an embedded computer that collects and
processes all data and runs the detection algorithm and gives out the warning, OBD2
adapter is a device that has the function of reading parameters on the vehicle.
The algorithm is implemented in Python. It includes many methods for collecting
data, evaluation, visualization, and handling the communication between modules in the
experiment.

Fig. 5. Testing system on vehicle

3.2 Participants
For this study, 20 volunteers were recruited from the Hanoi University of Science and
Technology students, including 15 men and 5 women. Their ages range from 19–23.
The participants were asked to sleep no more than 4 h a day in the previous two days.
The experiment was conducted at around 12 p.m and 1 p.m (commonly nap time) to
ensure that the subject should be in a drowsy state. Figure 5 shows the test vehicle and
a subject.

3.3 Scenarios
For scenarios, we conduct to measure the accuracy of the experiment in normal light and
low light conditions. This experiment was conducted in our laboratory. During the test,
the device monitored continuously and gave a warning when needed. Then, it exports
data to make a graph similar (see Fig. 6).
400 T.-D. Pham et al.

Fig. 6. EAR and MAR result in different states

4 Results and Discussion

4.1 Experimental Results
Average accuracy of 82.5% indicated that the detecting system performed well. The
accuracy for detecting inattention and tiredness, respectively, was 95% and 80% in
favorable lighting settings where the camera can clearly scan the driver’s face and the
system provides some clever algorithms. Despite the correctness decreasing in low light,
it is still high at 85% and 75%. It is quite challenging for the camera to detect every
gesture the driver makes in poor light. Therefore, the accuracy will be incompletely
correct.

4.2 Discussion
Regarding the method of detecting inattention based on the driver’s head position, in
some situations the system found difficulties in detecting the head landmarks. Thus the
accuracy and efficiency sometimes are not high. Furthermore, the proposed method is
mainly handling images, the external conditions affect the effectiveness of the processing
process. This phenomenon can be seen in the different mean of accuracy in two lighting
conditions.
On the other hand, the method of detecting the non-concentrated status of the driver
based on biological signals [11] such as ECG, EEG, etc. has higher accuracy. These bio-
logical signals are relevant and affect the driver’s concentration levels. But the downside
of this method is that the devices will need to directly attach the driver to be able to
obtain the signals, and the motion artifacts can cause difficulties in identifying. Because
of the driver’s posture and behavior to detect the non-concentrated situation, there will be
differences between individuals. Further investigation of combining both image based
method and bio-signal based method could potentially improve the efficiency of the
system in different working conditions.

5 Conclusion
In this study, four deep called Steering Wheel Movement, Eye Aspect Ratio, Mouth
Aspect Ratio, and Head pose estimation are discussed for detecting drowsiness, distrac-
tion, or inattention. The proposed system uses the head pose algorithm because of its
 Develop an Advanced Driver’s Behaviors Detection System 401

effectiveness to detect inattention regarding the change of head pose, then monitoring
the drivers whether they are in a distracting state. To identify drowsiness, methods are
investigated including SWM, EAR, and MAR. The outcomes are both quick and reli-
able. Despite the need for light circumstances, the system has a reasonable price and is
simple to use. To effectively apply the limiting light in the future, many works need to be
done to improve the proposed methods and the system needs to extend the application
domain of the system.

Acknowledgments. The research group would like to give special thanks to the colleges in the
Group of Automotive Engineering (Hanoi University of Science and Technology) who supported
and gave us a lot of recommendations during the process of research.

References
1. National sleep foundation, Sleep first. Drive alert. Drowsy driving prevention Week® (2021)
2. U.S. Department of Transportation, Traffic Safety Facts (2022)
3. Ramzan, M., Khan, H.U., Awan, S.M., Ismail, A., Ilyas, M., Mahmood, A.: A survey on
state-of-the-art drowsiness detection techniques. IEEE Access 7, 61904–61919 (2019)
4. Li, S., Xu, C., Xie, M.: A Robust O(n) Solution to the Perspective-n-Point Problem. IEEE
Trans. Pattern Anal. Mach. Intell. 34(7), 1444–1450 (2012)
5. Lugaresi, C., Tang, J., Nash, H., McClanahan, C., Uboweja, E., Hays, M., Lee, Mediapipe,
J.: A framework for building perception pipelines. arXiv preprint arXiv:1906.08172 (2019)
6. OBD-Python, https://github.com/brendan-w/python-obd, last accessed 2022/06/28
7. Saini, V., Saini, R.: Driver drowsiness detection system and techniques: a review. Int. J.
Comput. Sci. Infor-Mation Technol. 5(3), 4245–4249 (2014)
8. Tuba, M., Akashe, S., Joshi, A.: ICT systems and sustainability. In: Proceedings of ICT4SD,
vol 1 (2019)
9. Kumar, A., Patra, R.: Driver drowsiness monitoring system using visual behaviour and
machine learning. In: 2018 IEEE Symposium on Computer Applications and Industrial
Electronics (ISCAIE), pp. 339–344. IEEE (2018)
10. Miah, A.A., Ahmad, M., Mim, K.Z.: Drowsiness detection using eye-blink pattern and mean
eye landmarks’ distance. In: Proceedings of International Joint Conference on Computational
Intelligence, pp. 111–121. Springer, Singapore (2020)
11. Jung, S.J., Shin, H.S., Chung, W.Y.: Driver fatigue and drowsiness monitoring system with
embedded electrocardiogram sensor on steering wheel. IET Intel. Transp. Syst. 8(1), 43–50
(2014)
 Developing an Intelligent Waste Sorting
with 6DOF Robotic Arm

Vo Thu Ha1 , Than Thi Thuong1 , and Vo Thanh Ha2(B)

1 Faculty of Electrical Engineering, University of Economics Technology for Industries, Hanoi,
Vietnam
2 Faculty of Electrical and Electronic Engineering, University of Transport and
Communications, Hanoi, Vietnam
vothanhha.ktd@utc.edu.vn

Abstract. The creation and management of an intelligent waste sorting system

using a robot arm are described in the study. This system comprises a 6DOF robot
manipulator, a machine vision block, and a control unit for rubbish sorting using
analytic pictures. The item will be recognized by the neural network using the
YOLOv4 software. This technique detects and identifies various types and kinds
of trash, including paper-based waste, metal waste, and plastic waste. Results of
offline testing on a library of more than 600 untrained photos reveal that the trained
model has an average classification accuracy of roughly 98.43%.

Keywords: Waste sorting · Robotic arm · YOLO · Machine vision · 6DOF

1 Introduction
Robots that have been enhanced with artificial intelligence (AI) technology are a mod-
ern innovation in waste categorization. Both the industrial and civic sectors utilize it
extensively [1]. Modern recycling facilities in wealthy nations like the US, Finland, and
Switzerland are increasingly using powerful waste sorting robots due to the advent of
AI [2]. According to analysts, garbage will be handled and categorized more quickly
and precisely due to the practical help provided by robots with AI technology embedded
into them [3].
Additionally, selective waste type sorting is necessary, necessitating robots and
machine vision technologies in an automated sorting system. The Robot system must be
able to detect, identify, and estimate the distance to the waste sorting item to approach
it precisely. For instance, Zen Robotics Recycler, a recycling tool, was demonstrated
by Lukka, 2014 [4]. Waste from construction and demolition sites may be collected
by this robot and placed on a conveyor belt. Without the assistance of humans, it can
distinguish between wood, stone, and metal. Paulraj [5] presented a method in another
study to identify material waste from a picture captured with a thermal camera. A robot
with a thermal camera, a proximity sensor, and a five-step robot was created.
Visual characteristics were utilized as a sorting criterion in papers [6, 7]. In such
situations, materials are detected using optical sensors and laser beams based on their

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 402–411, 2023.
https://doi.org/10.1007/978-3-031-22200-9_44
 Developing an Intelligent Waste Sorting with 6DOF Robotic Arm 403

shape, color, and texture and are classified using mechanical techniques. Utilizing three
commercially available scanners, a method based on optical detection of fluorescence
in the literature [8] has been used to trace and identify several kinds of plastic materi-
als. It does not, however, categorize any content outside of its purview. An automated
garbage sorting system was created and developed by Overmore, Chahine, and Ghazal
[9], utilizing an inductive proximity recognition sensor, a capacitive proximity sensor,
and a photoelectric sensor. A separate document [9] claims that a mobile robot delivery
system was included in a 2013 automated waste sorter. A garbage-picking robot that can
deposit rubbish in a trash can is part of the system. Over 80% of the time, the system
has gathered garbage successfully. Russel [10] created a different method to categorize
metals, paper, plastic, and glass. Sensors have been employed for metal classification
and traditional glass, and a sensor using LASER and LDR has been created to classify
paper and plastic. There is also a weight sensor and counter to calculate the quantity of
sorted material.
Garbage sorting robots are costly, difficult to run and maintain, focused on a certain
kind of garbage, and mainly produced for industrialized nations. They have been created
to separate or manage waste. Numerous sciences have been used in robotic arm research
and development. However, a garbage-sorting robot has not been successfully entering
the market as a brand-new product. On the other hand, there isn’t much-connected
research regarding this robot globally. Most of them are solely concerned with the issue
of detecting and classifying trash in lab experiments, such as paper, metal, and plastic
trash [11].
The article discusses the control of a 6DOF robotic arm, as shown in Fig. 1. This
robot can distinguish between several sorts of rubbish, including paper, metal, and plas-
tic waste. All garbage must be detected by the machine vision unit, which is also in
charge of locating it and relaying that information to the control unit so it may move the
approaching garbage collecting equipment. Six stepper motors power a six-degrees-of-
freedom robot that serves as the operational mechanism. Convolutional Neural Network
(CNN)-based identification software called YOLO (You only look once) is combined
with the gripper system to recognize different sorts of waste. The test results demonstrate
the usefulness of the acquired recognition training model and the precision with which
the control system classifies waste.

2 Waste Sorting Controller

2.1 Controller Hardware
The hardware structure of the control block is illustrated in Fig. 2. The core element of
the control block is an industrial computer that integrates the control peripherals and
application software for control and monitoring and supervising all pineapple harvesting
activities. The main task of the partnership is to control the movement of the tongs to
reach the identified garbage and turn in the pecking order.
The position and depths of the garbage types are transformed and calculated into the
number of pulses for the Step motor 57HS21A – 2.2Nm Leadshine 8m axis. The short size
42 Step motor (NEMA17HS3401S) moves the gripper in the 0XY plane parallel to the
surface containing the various types of garbage. In addition, this motor will reciprocate
404 V. T. Ha et al.

Fig. 1. The 6DOF robot arm model

the handle going down to approach different kinds of waste. The computer will transfer
this motion information to AI NVIDIA Jetson Nano Developer and Arduino mega 2560
and Kit via an RS-485 communication protocol. The operation of each gripper will be
handled by AI NVIDIA Jetson Nano Developer and Arduino mega 2560 responsible.
The output of the Arduino mega 2560 connects to the drive drivers for the stepper motors
that move the horizontal arm and the garbage collector motor. The hardware structure
of the control block of the waste sorting robot arm is shown in Fig. 2.

Fig. 2. The hardware structure of the control block of the waste sorting robotic arm
 Developing an Intelligent Waste Sorting with 6DOF Robotic Arm 405

2.2 Algorithm to Control the Waste Sorting Process

The algorithm to control the waste sorting process for the robot arm is shown in Fig. 3(a)
and (b).

Fig. 3. Flowchart of the waste classification control algorithm of AI robot with six degrees of
freedom

In Fig. 3 the flowchart of the waste classification control algorithm of AI robot with
6 degrees of freedom is operated three steps, as shown:
Firstly, reset the robot arm angles to the original position. At this point, the camera
recognizes the object in the left, right, or middle parts.
Secondly, the camera takes pictures of the thing and sends the data to the Jetson
nano. This Jetson nano is responsible for image processing. In addition, this Jetson nano
406 V. T. Ha et al.

detects the object’s label and location coordinates and calculates the distance and angle
of the object relative to the base joint.
Thirdly, the Jetson nano will transmit data to the Arduino Mega 2560 microcontroller
to calculate the robot arm’s joint angle values. At this, the Arduino mega 2560 will control
the matching angles by some steps. In the first step, standard 0 will hold motor joint 1 to
move the robot to the left or right, and the remaining joints will control the robot to the
object’s position and pick up the thing. In the second step, control the robot to move to
the work of releasing the object, and the robot proceeds to remove the object. Computer
vision block.

3 Computer Vision Block

The machine vision block is in charge of photographing and processing trash sites. This
block finds and identifies the junk classes present in the camera-received picture. Then
determine the exact coordinates of the rubbish classifications in the original coordinate
system and compute the distance between the camera and the garbage classifications
to direct the robot’s gripper as it approaches and collects the trash. To solve this issue,
sophisticated image processing technologies must be used to recognize and classify var-
ious sorts of garbage. To detect multiple kinds of waste, traditional image processing
techniques utilized attributes including color, shape, size, angle, texture, and classifi-
cation technology [14]. However, because there are so many various sorts of garbage
in cities, including paper waste, metal waste, plastic waste, etc., the technologies men-
tioned above are ineffectual. Therefore, new technology is needed to overcome these
constraints due to its excellent accuracy and capacity to tolerate shifting environmental
conditions. A recent sophisticated method for learning picture attributes is deep learning
technology. Convolutional neural networks (CNNs) are created and used to recognize
objects. The YOLO (You Only Look Once) [9], SSD (Single Shot Detector) [10], and
Faster R-CNN [11] are examples of standard networks. When used to identify various
waste categories, these neural networks all provide positive results [12–14], and [15].
This research uses YOLO version 4 to find, categorize, and detect rubbish. According to
the documents [16, 17], the YOLO is a sophisticated convolutional neural network that
detects things quickly and accurately in real-time. A loss function directly connected to
the training feature is used to train the YOLO. The "You only look once" technique uses
a single forward propagation throughout the network to identify and anticipate a picture.
The item that has bounding boxes will be the output. The algorithm applies a neural
network to the entire concept, splitting it into sections and forecasts bounding boxes and
probabilities for each portion. The prediction probability weights these bounding boxes.
The data was trained using YOLO version 4 (YOLOv4), utilizing the transfer learning
approach as a result of the research.

3.1 Hardware

Built-in hardware includes a DS-UO2 industrial camera. It is fixed on the Waste Sort-
ing Robot Arm framework to take pictures of the work area. These snapshots will be
transferred to the host computer via USB C for further processing.
 Developing an Intelligent Waste Sorting with 6DOF Robotic Arm 407

3.2 Image Analysis and Waste Classification Identification

The machine vision block performs image analysis. First, this block creates rectangular
contours around the object. Then it calculates their coordinates in the Robot coordinate
system. Next, the garbage classification identification phase is performed by YOLOv4.
The configuration of YOLOv4 is detailed in [10].
The training process is carried out through the following steps:
Step 1. Collect data for training
One of the critical stages of deep learning is collecting the training database. The nec-
essary condition for getting a good training model is that the database is extensive and
high-quality. In this study, the data contained a lot of garbage samples. About 600 images
were obtained, including different lighting conditions and angles. Each photo has an
average of 5 junk labels with trash like metal, plastic, paper, and vegetables. In addition,
to ensure high accuracy for testing, the camera must be installed on the experimental
circuit with the same position and tilt as the practical circuit during image acquisition.
Finally, the captured IPEG images with a resolution of 1280x720 and a depth image are
transferred to the computer.
Step 2. Waste labeling
In the step, rectangular boxes will be drawn around the garbage present in the image to
label garbage. Next, each image is a file.xml containing information about the types of
junk. The image includes the class names labeled for the garbage types with the x - and
y - (in pixels) coordinates of the center of the rectangle, or the corresponding width and
height (in pixels) of the envelope rectangle, and finally the respective size and width (in
pixels) of the entire image. These.xml files are converted to file.txt format to be suitable
for network training.
Step 3. Data collection
The data is then collected using the Robo flow tool. Finally, the images will be applied
with several functions to rotate, flip the image, change the brightness, and blur.
+ Upload the dataset to Robo flow.
+ Check the labels of the objects in the uploaded photos.
+ Choose image transformation functions, and create multiple post-processed images
to create a large enough dataset. Diversify the cases, increase model performance.
+ Download the dataset after it has been collected. If you need more, download another
version.
The data collection used Google Collaboratory to train on the Tesla K80 GPU con-
figuration in the article. The database is uploaded to Google Drive, and the database
directory path with Google Collaboratory is set up. The images in the database are
divided into two sets with a ratio of 75:25: (1) training set (train.txt) for model training
and (2) validating set (valid.txt) for testing paradigm.
408 V. T. Ha et al.

Step 4. Build a model to identify different types of garbage using transfer learning via
fine-tuning technology
At this step, the identification model is built using YOLOv4 software using transfer learn-
ing technology through fine-tuning. The pre-trained model Yolov4 on the COCO dataset
of Alexey AB is used as the pre-trained -model. First, the top CNN convolution layers are
frozen, so their weights remain constant and trained on only two fully-connected layers.
The purpose of freezing CNN layers is to ensure that the higher-order features learned
from the pre-trained model are not destroyed. The first training phase is performed over
30 epochs with a learning rate of 1e-3 with a batch size of 32. The purpose of the first
training phase is to achieve stability of the loss function, which helps the model converge
to the global optimum. After the model reaches the optimal threshold on fully-connected
layers, it isn’t easy to increase the accuracy. Next, we need to perform the second training
process fine-tuning for garbage classification. In this process, the CNN convolutional
layers of the base network are de-icing, and training is conducted on all layers of the
neural network through 60 epochs with a learning rate of 1e-4 and batch size of 8. The
training process ends when the loss function converges to the threshold after about 20
epochs. At the end of the training process, the model weights will be written to the
weights file on the host computer.
Step 5. Testing
The test will be performed with the training model just received in step 4. Here the
article’s research used 100 photos of different types of garbage. The recognition model
predicts the kinds of waste, draws a bounding box around the garbage types, classifies
the garbage types, and determines the expected confidence score.

Fig. 4. The image identifies the types of waste on the image

Figure 4 shows that the types of garbage in the image have been identified and
surrounded by rectangles. It is labeled according to the green classification with the
corresponding accuracy. In Table 1, the results of off-line recognition on 600 images are
 Developing an Intelligent Waste Sorting with 6DOF Robotic Arm 409

statistically, showing that the average recognition accuracy of MAP is up to 98.82%.

These results have created a premise for the experiment of automatic waste identification
and classification in the laboratory presented in Sect. 4.

Table 1. Identification results

No Name of Number of Number of Number of Number of Identification

types of test images predicted correctly wrongly accuracy
waste wastes predicted predicted
waste waste
1 Paper type 650 690 678 12 98.2
waste
2 Plastic 670 700 690 10 98.5
waste
3 Paper type 630 692 683 9 98.6
waste
Average accuracy (MAP) 98.43

4 Experiment Results

The machine vision block is integrated with the robot to pick up garbage. The system
performed automated waste sorting tests in the laboratory—garbage sorting robot in the
laboratory, as shown in Fig. 11. The results of the identification and classification of
waste from reality are transferred from the camera to the host computer. The entire robot
arm, 6 degrees of freedom, will be moved to the garbage sorting area. When the system
starts up, the robot arm’s 6 degrees of freedom is checked to ensure it has returned to the
starting position. The camera captures and sends the image to the host computer. The
monitoring model is created, and the host computer screen consists of two windows, one
showing pictures taken from the camera and one displaying image recognition results
corresponding to each type of garbage. These images are circled in rectangles with paper,
plastic, and metal classifications and their predictive confidence (Fig 5).

5 Conclusion

This project used an intelligent system to operate an autonomous trash sorting robot.
With an average accuracy of up to 98.43%, the YOLOv4 tool is used by the application
to train the Neural network to recognize various types of garbage, ensuring that the
identification is made online under changing circumstances. Experimental findings have
confirmed the effectiveness and precision of the system overall. To increase accuracy
and productivity, the authors’ following research will concentrate on improving the
mechanical and control components of the system.
410 V. T. Ha et al.

Fig. 5. Experimental results of identification and classification of wastes

Acknowledgements. This research is sponsored by the project at the University of Economics -

Technology for Industries, and University of Transport and Communications.

References
1. Poon, C., Ann, T., Ng, L.: On-site sorting of construction and demolition waste in Hong Kong.
Resour. Conserv. Recycl. 32(2), 157172 (2001)
2. Ahmad, S.: A new technology for automatic identification and sorting of plastics for recycling.
Environ. Technol. 25(10), 1143–1149 (2004)
3. Huang, J., Pretz, T., Bian, Z.: Intelligent solid waste processing using optical sensor based sort-
ing technology, in image and signal. In: Processing (CISP), 2010 3rd International Congress
on, IEEE, vol 4, pp 1657–1661 (2010)
4. Lukka, T.J., Tossavainen, T., Kujala, J.V., Raiko, T.: Zenrobotics recyclerrobotic sorting using
machine learning. In: Proceedings of the International Conference on Sensor-Based Sorting
(SBS) (2014)
5. Paulraj, S.G., Hait, S., Thakur, A.: Automated municipal solid waste sorting for recycling
using a mobile manipulator. In: ASME 2016, International Design Engineering Technical
Conferences and Computers and Information in Engineering Conference. American Society
of Mechanical Engineers, pp. V05AT07A045V05AT07A045 (2016)
6. Kiyokawa, T., Katayama, H., Ogasawara, T.: Robotic waste sorter with agile manipulation
and quickly trainable detector. In: Computer Science IEEE Access (2021)
7. Ang, F., Gabriel, M.K.A.R., Sy, J., Tan, J.J.O., Abad, A.C.: Automated waste sorter with
mobile robot delivery waste system. In: de la salle University Research Congress, p. 79
(2013)
 Developing an Intelligent Waste Sorting with 6DOF Robotic Arm 411

8. Russel, M.H., Chowdhury, M.H., Uddin, M.S.N., Newaz, A., Talukder, M.M.M.: Develop-
ment of automatic smart waste sorter machine. In: Conference: International Conference on
Mechanical, Industrial and Materials Engineering 2013 (ICMIME2013) (2013)
9. Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.Y., Berg, A.C.: SSD: Single shot
multibox detector. In: Proceeding of European Conference on Computer Vision, pp 21–37.
Springer, New York (2016)
10. Redmon, J., Divvala, S., Girshick, R., Farhadi, A.: You only look once: Unified, real -
time object detection. In: Proceeding of IEEE Conference on Computer Vision and Pattern
Recognition, pp 779–788 (2016)
11. Girshick R (2015) Fast R –CNN. In: Proceedings of IEEE International Conference on
Computer Vision, pp 1440–1448
12. Halstead, M., McCool, C., Denman, S., Perez, T., Fookes, C.: Fruit quantity and quality
estimation using a robotic vision system (2018). arXiv:1801.05560v1[cs.CV]
13. Zhang, J., Gao, G., Karkee, M., Zhang, X., He, L., Zhang, Q.: Branch detection for apple trees
trained in fruiting wall architecture using depth features and regions -convolutional neural
network (R -CNN). Comput Electron Agric 155, 386–393 (2018)
14. Diya1, S.Z., Proma, R.A., Islam, M.N., Anannya, T.T., Al Mamun, A., Arefeen, R„ Al Mamun,
S„ Rahman, I.I., Rabbi, M.F.: Developing an intelligent waste sorting system with robotic arm:
a step towards green environment. In: International Conference on Innovation in Engineering
and Technology (ICIET) (2018)
15. Kiyokawa, T., Katayama, H., Tatsuta, Y., Takamatsu, J., Ogasawara, T.: Robotic waste sorter
with agile manipulation and quickly trainable detector. IEEE Access 9 (2021)
16. Zhang, Z., Wang, H., Song, H., Zhang, S., Zhang, J.: Industrial robot sorting system for
municipal solid waste ICIRA 2019: In: Intelligent Robotics and Applications. Industrial Robot
Sorting System for Municipal Solid Waste, pp. 342–353
17. Gao, M., Cai, Q., Zheng, B., Shi, J.: A hybrid YOLO v4 and particle filter based robotic arm
grabbing system in nonlinear and non-gaussian environment. Electronics 10(10), 1140 (2021)
18. https://github.com/Harighs/Robot-Arm-for-Sorting-Mechanism-using-ROS-and-YOLOv4
 Discrete Element Modeling Based
Determination of Shear Behavior in a Granular
Medium Through Displacement Field

Tien-Thinh Le1,2(B) , Thong Chung Nguyen3 , Xuan Son Nguyen1 ,

Huan Thanh Duong3 , Lu Minh Le3 , Anh Tu Nguyen4 , and Van Hai Nguyen1,2
1 Faculty of Mechanical Engineering and Mechatronics, PHENIKAA University, Ha Dong, Yen
Nghia 12116, Vietnam
thinh.letien@phenikaa-uni.edu.vn
2 A&A Green Phoenix Group JSC, PHENIKAA Research and Technology Institute (PRATI),
Trung Hoa, No.167 Hoang NganCau Giay, Hanoi 11313, Vietnam
3 Faculty of Engineering, Vietnam National University of Agriculture, Trau Quy, Gia Lam,
Hanoi 100000, Vietnam
4 Faculty of Mechanical Engineering, Hanoi University of Industry, 298 Cau Dien Str, Bac Tu

Liem District, Hanoi, Vietnam

Abstract. In this study, a numerical Discrete-Element-Modeling (DEM) model

was created to analyze a displacement field to identify the shear behavior in a
granular media. The resultant shear stress/normal stress ratio was compared to
findings reported in the available literature to validate the DEM model. The sim-
ulation model’s displacement field enabled analysis of the mechanical properties
of the granular medium. Exploring the mechanical behavior of granular media
under various boundary conditions and/or loadings may be helpful with the DEM
model.

Keywords: Discrete Element Method · Shear test · Granular mechanics ·

Agricultural granular materials · Displacement field

1 Introduction
With the growing of computation effort, Discrete-Element-Modeling (DEM) has been
used to model granular media in considerable part of various fields [1, 2]. Indeed, in the
field of agricultural machinery, Fu et al. [3] have developed AgriDEM platform, which
is able to simulate, analyze and evaluate the working process of machinery, in order
to fast develop and design of processing equipment for granular materials. In the field
of construction and building materials, for the purpose of examining the behavior of
concrete under intense confining pressure, Tran et al. [4] used DEM. To study the plastic
reaction of moist granular medium under compression, Than et al. [5] built a DEM model.
In a numerical simulation setting, Lommen et al. [1] investigated the connection between
particle stiffness and bulk material behavior. By providing insight to the local mechanical
response in a granular media, research using the DEM technique have the potential

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 412–418, 2023.
https://doi.org/10.1007/978-3-031-22200-9_45
 Discrete Element Modeling Based Determination 413

to significantly enhance the characterization. When compared to complex trials, such

numerical techniques could also save money and time when designing and developing
machinery that uses particle systems.
Agricultural granular materials, however, are not suited for shear test investigations
made for soil mechanics due to the high particle sizes and low forces involved. Indeed,
shear properties of the materials must be known in order to apply existing theories for the
prediction of wall pressures in structures for the storage of agricultural granular materials.
Due to the lower stresses and larger particle sizes required in testing agricultural granular
materials, equipment currently utilized in soil engineering cannot be used. Therefore,
research into the shear characteristics of agricultural granular materials is essential for
both understanding their mechanical properties as well as designing and developing
processing machinery.
Through the use of a shear test, a DEM model was created in this work to assess the
mechanical behavior of agricultural granular material. The outcomes made it possible to
investigate the applicability of the DEM technique in such a mechanical environment.
Additionally, the advantages of the DEM approach were demonstrated in the analysis
and monitoring of all particle displacements in the granular media under consideration.

2 Methods and Materials

2.1 Discrete Element Method (DEM)

DEM is a numerical technique that is frequently used to simulate the impact of many
particles, particularly granular particles. The creation of DEM was founded on model-
ing the movement of individual particles inside a granular medium through the use of
simulation. In order to calculate this motion, the rotational and translational equations of
motion formulated by Newton for separate particles need to be solved. The use of DEM
on granular particles has been the subject of numerous studies. The literature contains
the foundations of this methodology [6].

2.2 Input Parameters for DEM Simulations

In order to create and construct the seeding machine, the mechanical behavior of agri-
cultural granular materials such as dry soybean grains was examined in this work. The
variety was Glycine max with moisture content smaller than 10%. The microscopic
characteristics of soybean particles are collected from the previous publication of the
authors [7]. The DEM simulations in this work were conducted using the LIGGGHTS®
code. The contact between a particle and a particle wall was simulated using a nonlinear
Hertz-Mindlin model with no cohesion, as advised by several studies, including Raji
et al. [8].
Particle sizes were quantified using a custom imaging platform (4.42 MP/cm2 pixel
density Fujifilm X-E2S camera with a Fujinon XF18-55mm F2.8–4 R LM OIS lens). This
allowed for a pixel density of 16 per millimeter in the resulting image (recommended
for characterizing particles greater than 3 mm in size). Random samples of soybean
grains were used for the photographic study (about 900 grains were tested). The raw
414 T.-T. Le et al.

image is shown in Fig. 1a, while the processed binary image showing the equivalent
diameter of each particle is shown in Fig. 1b. The diameter was determined by first
calculating the particle’s acquired area. Figure 1c shows the particle size distribution
as a function of 900 equivalent diameters, with a mean of 6.33 mm and a standard
deviation of 0.46 mm. Our examination showed that the mean diameter estimated by
digital analysis was very comparable to the findings published by Ghodki et al. [9] for
the same variety of soybeans. (i.e., 6.24 mm). It was the particle size distribution that
was used in DEM simulations to generate the particle diameter.

Fig. 1. Particle size characterization includes a the original picture, b the processed picture with
each particle’s diameter, and c the particle-size-distribution

2.3 Description of Shear Test

Figure 2 schematizes the shear test used in this work. a box with two parts, the bottom
of which is stationary and the top of which is capable of constant-velocity horizontal
translation along the x-axis. There is a servo-plate attached to the top wall. The latter
is unrestricted in its vertical movement (z-axis). A confinement force is given to the
sevo-plate, which, in conjunction with the ongoing loading, causes the granular media
to be compressed in a consistent manner. Compression force is a constant normal force
working on the particles, whereas the force operating on the top part in the x-direction
is perpendicular to the normal force that was previously described. Using a numerical
DEM technique, the normal and, in particular, the tangential forces enable describing
the local stress transmission in the granular medium. Many scholars used this type of
shear test design to study the mechanics of granular materials, including Schellart et al.
[10].

3 Results and Discussion

3.1 Validation of Numerical Model

The ratio of the shear stress to the normal stress, when the granular material is being
loaded, is what determines the coefficient of static friction particle/particle, as suggested
 Discrete Element Modeling Based Determination 415

Fig. 2. The shear test’s design in this study

by numerous publications in the literature [11]. In this instance of compression, the

corresponding wall reaction forces were used to compute the stresses in the x-axis
(tangential to the direction of load) and z-axis (normal to the direction of compression).
More specifically, the reaction forces in each triangle mesh element in contact with the
particles were used to determine these reactions. The evolution of the normal stress placed
on the servo-wall over time is depicted in Fig. 3a. Figure 3b depicts the progression of
the shear stress to normal stress ratio for the material under consideration. As shown
in Fig. 3b, when the compression plate makes contact with the particle assembly, the
normal stress on the wall begins to rise. Due to the initial contacts between the particles
and compression plate, a slight overshoot is also seen. Regarding discrete elements, the
particles are positioned such that they can react to the loading. Finally, the shear and
normal stresses in the granular medium converge. Such convergence demonstrates the
granular medium’s equilibrium under continual loading. Figure 4 depicts how granular
media changed over time during the shear test; the particles are colored to emphasize
movement.

3.2 Characterization of Shear Behavior by DEM

Figure 5 presents displacement vectors and horizontal displacement fields of particles,
respectively. More precisely, Figs. 5a, 5c, 5e and 5g show displacement vector at 0.01%,
2.5%, 5%, and 10% of strain, respectively. Figures 5b, 5d, 5f and 5h show horizontal
displacement field at 0.01%, 2.5%, 5%, 10% of strain, respectively. The scale of the
vectors and colorbar are the same in all figures. It is seen that at the beginning of
the shear test (very small strain), all particles have small and similar velocities, so the
displacement field (similar to a continuous domain). When the strain is higher, the
particles tend to move following the direction of movement of the wall, especially the
ones at the boundary between the upper and bottom parts. The higher the strain, the
larger the norm of the displacement vector and so the horizontal displacement.

4 Conclusion
An example of a numerical simulation of the shear test for granular medium has been
provided here so that the mechanical behavior can be analyzed. When calibrating the
416 T.-T. Le et al.

(a) (b)
1.2 0.3

1
0.25
Normal stress applied by servo-wall
(in ratio with normal stress target)

Ratio shear stress / normal stress

0.8
Ideal application
DEM simulation
0.6 0.2 Static friction coefficient
DEM simulation

0.4
Initiation Shear test 0.15
0.2 Initiation Shear test

0 0.1
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
Elapsed time (s) Elapsed time (s)

Fig. 3. Evaluation of: a normal stress applied by servo-wall and b shear stress / normal stress
ratio over time

Fig. 4. Evolution of granular media over time during the shear test (0, 5% and 10% of strain,
respectively), (particles are colored to highlight the movement)

numerical DEM model, we took into consideration the particle size distribution as well as
the mechanical characteristics, the coefficient of static friction, the coefficient of rolling
friction, and the coefficient of restitution. The simulation made it possible to determine,
from a mechanical standpoint, how discrete particles would behave when subjected to a
shear test.
 Discrete Element Modeling Based Determination 417

Fig. 5. Displacement vector at a 0.01%, c 2.5%, e 5%, g 10% of strain, respectively. Horizontal
displacement field at b 0.01%, d 2.5%, f 5%, h 10% of strain, respectively. The scale of the vectors
and colorbar are the same in all figures

Acknowledgements. This research is funded by Phenikaa University.

References
1. Lommen, S., Schott, D., Lodewijks, G.: DEM speedup: Stiffness effects on behavior of bulk
material. Particuology 12, 107–112 (2014). https://doi.org/10.1016/j.partic.2013.03.006
2. Le, T.-T.: Investigation of force transmission, critical breakage force and relationship between
micro-macroscopic behaviors of agricultural granular material in a uniaxial compaction test
using discrete element method. Part. Sci. Technol. 0, 1–18 (2021). https://doi.org/10.1080/
02726351.2021.1983904
3. Fu, H., Jin, C., Yu, J.: The DEM-based digital design platform for agricultural machinery—
AgriDEM. In: Li, X., Feng, Y., Mustoe, G. (eds.) DEM 2016. SPP, vol. 188, pp. 1253–1263.
Springer, Singapore (2017). https://doi.org/10.1007/978-981-10-1926-5_129
418 T.-T. Le et al.

4. Tran, V.T., Donzé, F.-V., Marin, P.: A discrete element model of concrete under high triaxial
loading. Cem. Concr. Compos. 33, 936–948 (2011). https://doi.org/10.1016/j.cemconcomp.
2011.01.003
5. Du, T.V., Saeed, K., Minh, T.A., Jean-Michel, P., François, C., Jean-Noël, R.: Basic mechanical
properties of wet granular materials: a DEM study. J. Eng. Mech. 143, C4016001 (2017).
https://doi.org/10.1061/(ASCE)EM.1943-7889.0001043
6. Benvenuti, L., Kloss, C., Pirker, S.: Identification of DEM simulation parameters by artificial
neural networks and bulk experiments. Powder Technol. 291, 456–465 (2016)
7. Nguyen, T.X., et al.: Characterization of soybeans and calibration of their DEM input
parameters. Part. Sci. Technol. (2020). https://doi.org/10.1080/02726351.2020.1775739
8. Raji, A.O., Favier, J.F.: Model for the deformation in agricultural and food particulate materials
under bulk compressive loading using discrete element method. I: Theory, model development
and validation. J. Food Eng. 64, 359–371 (2004). https://doi.org/10.1016/j.jfoodeng.2003.
11.004
9. Ghodki, B.M., Patel, M., Namdeo, R., Carpenter, G.: Calibration of discrete element model
parameters: soybeans. Comput. Particle Mech. 6(1), 3 (2018). https://doi.org/10.1007/s40
571-018-0194-7
10. Schellart, W.P.: Shear test results for cohesion and friction coefficients for different granular
materials: scaling implications for their usage in analogue modelling. Tectonophysics 324,
1–16 (2000). https://doi.org/10.1016/S0040-1951(00)00111-6
11. Zhou, Y., Wang, H., Zhou, B., Li, J.: DEM-aided direct shear testing of granular sands incor-
porating realistic particle shape. Granul. Matter 20(3), 1–12 (2018). https://doi.org/10.1007/
s10035-018-0828-8
 Double-Loop PID Control with Parameter
Optimization for an Autonomous Electric
Vehicle

Quoc-Dong Hoang1 , Le Dinh Nghiem1(B) , Nguyen Duc Toan1 ,

Pham Khanh Hung1 , Luu Quang Do1 , Dang Trung Hieu1 , Tung Lam Nguyen2 ,
and Le Sy Dung3
1 Institute of Mechanical Engineering, Vietnam Maritime University, Haiphong, Vietnam
{Nghiemldvck,donghqvck}@vimaru.edu.vn
2 Ha Noi University of Science and Technology, Dai Co Viet 01, Ha Noi, Vietnam
3 School of Mechanical Engineering, Hanoi University of Science and Technology,

Hanoi 100000, Vietnam

Abstract. An autonomous vehicle is a fascinating and trendy topic from both

research and practical perspectives. It can be said that this is a development ten-
dency for future automobiles and attracts the attention and a substantial amount
of finance from big automobile corporations worldwide. In terms of technology,
driving a driverless car puts a great deal of pressure since there is only one con-
trol input (steering angle) that must simultaneously control the position and yaw
angle of the vehicle. Moreover, the position of the vehicle over time is a function
of the yaw angle. In order to solve the problem, this study proposes a new app-
roach with the use of a dual-loop Proportional Integral Derivative (PID) controller
combined with optimizing parameters based on the Particle Swarm Optimization
(PSO) algorithm to improve the control performance. By simulating cars follow-
ing given trajectories based on the car’s kinematics and dynamics, the efficiency
and feasibility of the work can be assessed. The numerical simulation results of
both the established PID controller and PSO algorithm are given at the end of the
presentation.

Keywords: Self-driving car · Autonomous vehicle · PSO algorithm · Double

loop control · PID

1 Introduction
A self-driving car or an autonomous vehicle has been researching, developing, and
becoming a great tendency in the field of automobile manufacturing. Also, the number
of people owning them has increased significantly in recent years because it improves
comfort, luxury, and especially safety when traveling. In present-day technology, there
are many levels of unmanned driving, but to ensure high levels, precise control of the
trajectory according to the desired trajectory is still a core issue. Different from some
previous approaches focusing on analyzing the development trend [1], planning the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 419–424, 2023.
https://doi.org/10.1007/978-3-031-22200-9_46
420 Q.-D. Hoang et al.

trajectory [2], or precisely path following with complicated algorithms [3, 4], this study
proposes a simple approach to solve the problem of precise control of the car’s trajectory.
In general, this system poses significant challenges to the control as it can also be
considered an underactuated system [5, 6] (the system has a smaller number of control
signals than the number of controlled state variables [7]). However, there is a difference
with other systems [8, 9] where the un-actuated state variables are mainly secondary
oscillations arising around the equilibria; all the state variables in the driverless car
follow certain trajectories. The first theoretical analysis and practical application of PID
were in the field of autopilot systems for ships, developed in the early 20s of the 19th
century. It is simple but easy to apply for control and high efficiency. Therefore, it is
widely used in industry. In this study, based on the kinematic property of the car that
the precise control of the yaw angle will determine the accuracy of the side deflection,
we have built a controller with a double loop combining PID controllers. Moreover, the
controller parameters are optimized based on the PSO algorithm, first introduced by
Kennedy, Eberhart and Shi [10, 11], which reduces the controller design time, improves
the control quality, and gives relatively satisfactory results. The control structure and
built algorithms have been proposed and investigated for efficiency through simulations
given at the end of the paper.

2 Dynamic Model and Control Strategy

In Fig. 1, the main kinematic parameters of an autonomous vehicle operating on the road
 T
are given. For the system, the vector of state variables is defined as x = x1 x2 x3 x4 ,
where x1 and x2 are the yaw angle and the angular velocity of the vehicle; while, x3 and
x4 denote the lateral position and velocity, respectively.
The vehicle’s model [12] can be given in a simple form as follows

ẋ = Ax + Bu, (1)
⎡ ⎤ ⎡ ⎤
a11 0 a13 0 b1
⎢ 0 0 1 0⎥ ⎢0⎥
⎥, B = ⎢ ⎥,
where A = ⎢
⎣ a31 0 a33 0⎦ ⎣ b2 ⎦
1 a42 0 0 0
−2(Cβf +Cβr ) −2(Cβf lf +Cβr lr ) −2(Cβf lf +2Cβr lr )
and a11 = mVx , a13 = −Vx − mVx , a31 = − Iz Vx ,
−2(Cβf lf2 +Cβr lr2 ) 2Cβf 2Cβf lf
a33 = −Vx − Iz Vx , b1 = m , b3 = Iz .
Here, Cβf and Cβr denote the stiffness of the front and rear wheels. lf and lr arethe
distance from the center of the car to the front and rear wheels. Vx is the velocity of the
car. Iz and m are the moment of inertia and the mass of the car. The system is controlled
by the steering angle δf .
In fact, the lateral position of the vehicle is the function of yaw angle and time. In
other words, the precise control of the yaw angle will greatly determine the accuracy of
the lateral orbital tracking. Because the nature of the yaw angle change will determine
the lateral deflection of the vehicle. Based on the system characteristics, the control
strategy is developed as follows.
 Double-Loop PID Control with Parameter 421

Fig. 1. Autonomous vehicle

The inner PID control the output signal, χs (t), can be computed as

χs (t) = λ1a σ1 + λ1b σ̇1 + λ1c σ1 d τ , (2)

where σ1 = x3 − x3T with x3T being the reference position of the car.λ1a , λ1b , and λ1c
are the controller parameter.
By defining the function σ2 = x1 − χs = χ̃s , the outer loop PID can be calculated as

δf (t) = λ2a σ2 + λ2b σ̇2 + λ2c σ2 d τ , (3)

where λ2a , λ2b , and λ2c are the controller parameters.

The parameters λ1a , λ1b , λ1c ,λ2a , λ2b , and λ2c are determined by utilizing the PSO
method that is proposed in the next section. The over-all control structure with the
integration of the PID control and the PSO algorithm is provided in Fig. 2.
Firstly, the particles adjust their displacement by learning by themselves from their
best individual displacements and the best displacement in the global swarm. With
considering a problem of minimization with D-dimensional search space, the velocity
and displacement are updated by the formula as follows
   
Vijt+1 = ωVijt + c1 r1 Pijt − Xijt + c2 r2 Gjt − Xijt , Xijt+1 = Vijt+1 + Xijt , (4)

where 1 ≤ i ≤ D; 1 ≤ i ≤ M with M being the population size; ω is the inertia

weight to better balance the exploitation and exploration; c1 and c2 are the coefficients
of the acceleration; r1 and r2 in the interval (0,1) are the mutually independent random
numbers uniformly distributed; Pijt and Gjt are the best displacement of the particle i and
the global optimized displacement in the swam that satisfies the conditions
t    
Xi , if f Xit < f Pit   
Pij =
t
t−1  t  t  , G t = Pgt , g = arg min f Pit . (5)
Pi , if f Xi ≥ f Pi 1≤i≤M
422 Q.-D. Hoang et al.

Fig. 2. Physical model of a bridge crane

3 Numerical Simulation
In Fig. 3, the process of the weight convergence following the iteration. Obviously, the
weight is converged in the 6th iteration cycle, with the value being 4819.5. Consequently,
the lateral position has been compelled to its desired trajectory, as shown in Fig. 5. This
is coinciding and consistent with the lateral orbital tracking error (in Fig. 6) being forced
to zero. The control accuracy of the outer loop PID is achieved by the precise of the
inner loop PID with the yaw angle tracking error (Fig. 8) or the precise control of the
yaw angle is given in Fig. 7. Eventually, the simulated route of the vehicle on the lane
is shown in Fig. 4.

Fig. 3. The weight following the iteration

 Double-Loop PID Control with Parameter 423

Fig. 4. Vehicle’s position

Fig. 5. Control the lateral position

Fig. 6. Lateral position tracking error

Fig. 7. Control the Yaw Angle

4 Conclusion
The study proposed a control strategy with a combination of PID dual loop structure and
PSO algorithm for driverless cars. The dual PID loop is designed based on the kinematic
characteristic between yaw angle and lateral position. Meanwhile, PSO is applied to
optimize controller parameters to improve control quality and reduce execution time.
The simulation results partly confirm the feasibility and effectiveness of the controller,
which is the premise for applying to real self-driving car systems.
424 Q.-D. Hoang et al.

Fig. 8. Yaw angle tracking error

Acknowledgments. The study was supported by the Institute of Mechanical Engineering,

Vietnam Maritime University.

References
1. Hussain, R., Zeadally, S.: Autonomous cars: research results, issues, and future challenges.
IEEE Commun. Surv. Tutorials 21(2), 1275–1313 (2019)
2. Jo, K., Jo, Y., Suhr, J.K., Jung, H.G., Sunwoo, M.: Precise localization of an autonomous car
based on probabilistic noise models of road surface marker features using multiple cameras.
IEEE Trans. Intell. Transp. Syst. 16(6), 3377–3392 (2015)
3. Pham, T., Park, J., Lee, S.-G., Hoang, Q.-D.: Adaptive neural network sliding mode control for
an unmanned surface vessels. In: 2020 20th International Conference on Control, Automation
and Systems (ICCAS), pp. 519–523 (2020)
4. Huang, Y., Yong, S.Z., Chen, Y.: Stability control of autonomous ground vehicles using
control-dependent barrier functions. IEEE Trans. Intell. Veh. 6(4), 699–710 (2021)
5. Hoang, Q.-D., Park, J.-G., Lee, S.-G., Ryu, J.-K., Rosas-Cervantes, V.A.: Aggregated hierar-
chical sliding mode control for vibration suppression of an excavator on an elastic foundation.
Int. J. Precis. Eng. Manuf. (2020)
6. Hoang, Q.-D., Lee, S.-G., Dugarjav, B.: Super-twisting observer-based integral sliding mode
control for tracking the rapid acceleration of a piston in a hybrid electro-hydraulic and
pneumatic system. Asian J. Control 21(1), 483–498 (2019)
7. Hoang, Q.-D., Rosas-Cervantes, V.A., Lee, S.-G., Weon, I.-S., Choi, J.-H., Kwon, Y.-H.:
Robust finite-time convergence control mechanism for high-precision tracking in a hybrid
fluid power actuator. IEEE Access 8, 196775–196789 (2020)
8. Hoang, Q.-D., Park, J., Lee, S.-G.: Combined feedback linearization and sliding mode control
for vibration suppression of a robotic excavator on an elastic foundation. J. Vib. Control
1077546320926898 (2020)
9. Hoang, Q.-D., et al.: Robust control with a novel 6-DOF dynamic model of indoor bridge
crane for suppressing vertical vibration. J. Braz. Soc. Mech. Sci. Eng. 44(5), 1–12 (2022).
https://doi.org/10.1007/s40430-022-03465-3
10. Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of ICNN’95 -
International Conference on Neural Networks (1995)
11. Dorigo, M., et al. (eds.): ANTS 2010. LNCS, vol. 6234. Springer, Heidelberg (2010). https://
doi.org/10.1007/978-3-642-15461-4
12. Coppola, P.: Autonomous Vehicles and Future Mobility. Elsevier Science, Amsterdam (2019)
 Effect of Input Factors on Material Removal
Speed in Electrical Discharge Machining SKD11
Tool Steel

Nguyen Hong Linh1 , Bui Thanh Danh2 , Tran Minh Tam3 , Tran Huu Danh4 ,
Nguyen Huu Quang5 , Nguyen Thanh Tu6 , and Do Thi Tam6(B)
1 Electric Power University, Hanoi, Vietnam
2 University of Transport and Communications, Hanoi, Vietnam
3 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
4 Vinh Long University of Technology Education, Vinh Long, Vietnam
5 University of Economics-Technology for Industries, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

dothitam@tnut.edu.vn

Abstract. The results of a study on the influence of input factors on the material
removal speed (MRS) in electrical discharge machining (EDM) cylindrical shaped
parts made of SKD11 tool steel are presented in this paper. The Taguchi method
was used to design the experiment and analyze the results to solve that problem.
In addition, the input process factors such as the pulse time, the pulse off time, the
current, and the serve voltage were investigated. In addition, Analysis of Variance
(ANOVA) technique was used to assess the effect of the input factors on MRS.
Furthermore, a set of optimal input factors for achieving the maximum MRS has
been proposed.

Keywords: EDM · Electrical discharge machining · MRS · Cylindrical shaped

parts · SKD11

1 Introduction
Electrical discharge machining is a method of conductive material removal by electrical
pulses. It is the most commonly used non-traditional machining method. EDM is often
employed to process parts with complex shapes such as stamping dies, thin-walled parts
as well as small and deep holes, etc. Also, it can be used to process extremely hard
or difficult-to-machine material. Nevertheless, up to now, there have been quite a few
studies on the EDM process.
So far, the research on the EDM process has concentrated on determining the optimal
input parameters to improve MRR [1–3], reduce the surface roughness [1, 3–5], or limit
the electrode wear rate [1, 3, 6]. A number of studies have been conducted with different
electrode materials such as copper [4–6] or graphite. In addition, different methods are
also used to study EDM such as the Taguchi method [6–8], gray relational analysis
method [7, 9], or TOPSIS [10, 11]. Most of the research on EDM focuses on the study

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 425–432, 2023.
https://doi.org/10.1007/978-3-031-22200-9_47
426 N. H. Linh et al.

of cavity-shaped workpiece [8, 12, 13]. Recently, there have been a number of studies
on machining cylindrical shaped parts [4–6]. The above-mentionedstudies have focused
on machining cylindrical shaped partsbut for 90CrSi steel. There has not been much
research done on EDM the aforementioned parts with SKD11 steel.
This paper presents a study on optimization EDM process for maximum MRS when
processing cylindrical shaped parts SKD11 tool steel. In the paper, the impact of the EDM
parameters on the MRS was examined. Also, optimum EDM parameters for maximum
MRS was suggested.

2 Experimental Work
The experiment to investigate the influence of parameters on MRS was designed with 4
input parameters, which are shown in Table 1 together with their levels. Figure 1 displays
the experimental setup. The material of the electrode was copper and the workpiece
material was SKD11 tool steel. Also, a sinker EDM (model CNC-AG40L) was used
for the experiment. The Taguchi method with design L9 (34 ) was applied to design
experimental plan. Table 2 presents the plan of the experiment and the response (the
MRS).

Table 1. Input factors and their levels

No Parameters Code Unit Level

1 2 3
1 Pulse on time Ton μs 6 16 26
2 Pulse off time Toff μs 8 18 28
3 Peak current IP A 3 7 11
4 Servo voltage SV V 3 6 9

3 Results and Analysis

ANOVA analysis tool on Minitab 19 software was used to evaluate the impact of input
parameters on MRS. With the objective function as the maximum, MRS is the best:
 n 
1 1
S/N = −10log 10 (1)
n y2
i=1 i

From the results of ANOVA (Table 3 and Fig. 2), it is reported that IP has the greatest
effect on the MRS (48.83%). The second greatest effect is posed by Ton (29.62%). SV
accounts for 19.81%, and Ton appears with very little influence on MRS (1.75%). The
influence order of the input parameters on S/N is described in detail in Table 4, including
IP, Ton , SV, and Toff .
 Effect of Input Factors on Material Removal Speed 427

Fig. 1. Experimental setup

Table 2. Experimental plan and measured MRS

No SPD VM Ton Toff MRS (g/h) Mean S/N

1 6 8 3 3 0.7693 0.76928 −2.27837
2 6 18 7 6 1.1348 1.13483 1.09861
3 6 28 11 9 0.9634 0.96341 −0.32376
4 16 8 7 9 2.6826 2.68258 8.57106
5 16 18 11 3 2.8785 2.87847 9.18324
6 16 28 3 6 0.6974 0.69744 −3.12992
7 26 8 11 6 0.7622 0.76217 −2.35901
8 26 18 3 9 0.5853 0.58532 −4.65216
9 26 28 7 3 1.6701 1.67014 4.45508

The influence of the parameters on the S/N value is described in Fig. 3. Since the
goal is to obtain the MRS as large as possible, it is required to achieve the largest S/N
value. In this case, for Ton , the MRS increases when Ton increases from 6 to 16 μs.
However, the MRS decreases when Ton increases from 16 to 26 μs. For Toff , MRS
increases as Toff increases from 8 to 18 μs and MRS decreases as Toff increases from
18 to 28 μs. Besides, MRS increases sharply when IP increases from 3 to 7A, and MRS
decreases when IP continues to increase from 7 to 11A. For servo voltage, when SV
increases from 3 to 6V, MRS will decrease and then increases when SV increases from
6 to 9V. Thus, to achieve the maximum MRS, the S/N value needs to be maximum for
each input parameter. Based on Fig. 2, it is possible to determine the set of optimum
parameters for the largest MRS value as shown in Table 5. With the proposed optimum
input parameters, the predicted MRS can reach 3.17297 g/h, as shown in Table 6.
428 N. H. Linh et al.

Table 3. Analysis of Variance for S/N ratios

Source DF Seq SS Adj SS Adj MS F P C (%)

Ton 2 77.802 77.8019 38.9009 * * 53.60
Toff 2 23.467 23.4673 11.7337 * * 16.17
IP 2 35.893 35.8927 17.9463 * * 24.73
SV 2 7.982 7.9818 3.99.9 * * 5.50
Residual error 0 * * * *
Total 8 145.144

Fig. 2. Influence on MRS (%)

Table 4. The influence order of input parameters on S/N ratios

Larger is better
Level Ton Toff IP SV
1 −0.5012 1.3112 −3.3535 3.7867
2 4.8748 1.8766 4.7083 −1.4634
3 −0.8520 0.3338 2.1668 1.1984
Delta 5.7268 1.5428 8.0617 5.2501
Rank 2 4 1 3

Evaluation of experimental models:

Figure 4 describes the error distribution charts to evaluate the experimental model. From
the figure, the points represent the experimental errors (the blue points) appear close to
 Effect of Input Factors on Material Removal Speed 429

Fig. 3. Effect of input parameters on the S/N ratios

Table 5. Optimum process parameters for maximum MRS

Parameter Code Unit Level Value

Pulse on time Ton μs 2 16
Pulse off time Toff μs 2 18
Peak current IP A 2 7
Servo voltage SV V 1 3

Table 6. Predicted MRS

the normal distribution line (the red line), which indicates a small deviation. Besides, the
error frequency graph shows that the errors are mostly in the range −1 to 1. The other
430 N. H. Linh et al.

two graphs show a random distribution of mean experimental errors. It shows that the
proposed model is largely influenced by the selected input parameters and is not affected
by the order of experiments.

Fig. 4. Residual Plots for MRS

Figure 5 presents the graph which was built by using Anderson-Darling method in
Minitab 19 software. From the graph, the data corresponding to the experimental points
(the blue dots) are all within the area bounded by 2 upper and lower limits with 95%
significance level. Also, the P value is 0.411 and it is larger than α = 0.05. This indicates
that the applied experimental model is consistent with the above significance level.

4 Conclusions

This paper presents a study on the effect of the EDM parameters on the MRS when
processing cylindrical shaped parts made of SKD11 steel. To do that, four input process
parameters counting the pulse on time, the pulse off time, the current, and voltage were
investigated. Also, the Taguchi method was used to design and analysis experiments.
The influences of the input factors on the MRS were analyzed by ANOVA. Furthermore,
the optimum EDM process parameters were proposed for getting the maximum MRS.
 Effect of Input Factors on Material Removal Speed 431

Fig. 5. Probability plot of MRS

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Jabbaripour, B., et al.: Investigating the effects of EDM parameters on surface integrity, MRR
and TWR in machining of Ti–6Al–4V. Mach. Sci. Technol. 16(3), 419–444 (2012)
2. Pradhan, M.K., Biswas, C.K.: Modelling of machining parameters for MRR in EDM using
response surface methodology (2008)
3. Dwivedi, A.P., Choudhury, S.K.: Effect of tool rotation on MRR, TWR, and surface integrity
of AISI-D3 steel using the rotary EDM process. Mater. Manuf. Process. 31(14), 1844–1852
(2016)
4. Ky, L.H., et al.: Multi-objective optimization of surface roughness and electrode wear in EDM
cylindrical shaped parts. In: Materials Science Forum. Trans Tech Publ. (2021)
5. Hong, T.T., et al.: Evaluating influences of input parameters on surface roughness in sinking
EDM cylindrical shaped parts. In: Materials Science Forum. Trans Tech Publ. (2021)
6. Hong, T.T., et al.: Effects of input parameters on electrode wear rate when EDM cylindrical
shaped parts. In: Materials Science Forum. Trans Tech Publ. (2021)
7. Jung, J.H., Kwon, W.T.: Optimization of EDM process for multiple performance charac-
teristics using Taguchi method and Grey relational analysis. J. Mech. Sci. Technol. 24(5),
1083–1090 (2010)
8. Rahang, M., Patowari, P.K.: Parametric optimization for selective surface modification in
EDM using Taguchi analysis. Mater. Manuf. Process. 31(4), 422–431 (2016)
9. Roy, A.K., Kumar, K.: Effect and optimization of various machine process parameters on the
surface roughness in EDM for an EN41 material using Grey-Taguchi. Procedia Mater. Sci. 6,
383–390 (2014)
432 N. H. Linh et al.

10. Manivannan, R., Kumar, M.P.: Multi-attribute decision-making of cryogenically cooled

micro-EDM drilling process parameters using TOPSIS method. Mater. Manuf. Process. 32(2),
209–215 (2017)
11. Huo, J., et al.: Influence of process factors on surface measures on electrical discharge
machined stainless steel using TOPSIS. Mater. Res. Express 6(8), 086507 (2019)
12. Liu, K., Reynaerts, D., Lauwers, B.: Influence of the pulse shape on the EDM performance
of Si3 N4 –TiN ceramic composite. CIRP Ann. 58(1), 217–220 (2009)
13. Zeilmann, R., Ivaninski, T., Webber, C.: Surface integrity of AISI H13 under different pulse
time and depths by EDM process. Procedia CIRP 71, 472–477 (2018)
 Effect of Main Blade Configuration
on the Performance of the Optimized Multicurve
Savonius Wind Turbine

Banh Duc Minh1 , Le Dinh Anh1(B) , Tran Cuong Hung2 , Doan Viet Ha3 ,
Tran Cong Manh Hung3 , and Nguyen Thi Thu Phuong3
1 School of Aerospace Engineering, VNU-University of Engineering and Technology, Vietnam
National University, Hanoi, Vietnam
anh.ld@vnu.edu.vn
2 Faculty of Engineering Mechanics and Automation, VNU-University of Engineering and
Technology, Vietnam National University, Hanoi, Vietnam
3 VNU-University of Engineering and Technology, Vietnam National University, Hanoi,

Vietnam

Abstract. This study numerically investigates the effect of multi-curve blade

configuration on the performance of the optimized Savonius wind turbine. The
simulation is performed utilizing a sequence of unsteady 2D computational fluid
dynamics in the commercial software ANSYS Fluent 2021R2. The results state a
high influence of the turbine performance on the multi-curve shape. The highest
performance of the rotor is recognized on the blade configuration with the main
profile made by a quarter circular R3 * = 0.5 and a quarter elliptical section. For
which, the power coefficient C p is improved by 185.1% at the tip speed ratio
(TSR) higher than 1.0 and up to 5.5% at TSR of less than 0.8, making this design
better suited for wide working conditions over the previous configurations.

Keywords: Savonius · Renewable energy · Multi-curve blade · Realizable k-ε ·

Tip speed ratio

1 Introduction
Savonius wind turbine, built by engineer S. J. Savonius [1, 2], is a popular vertical
axis wind turbine (VAWT) used in both urban and rural areas. The advanced features
of the Savonius turbine include simple assembly, low maintenance costs, low starting
speed, omnidirectional operation, and low noise. However, the efficiency of the Savonius
turbine is still relatively low compared to other turbines.
Many studies have been carried out to improve the efficiency of Savonius turbines
including flow control by flow deflectors [3, 4] and improved blade profiles [5–13]. With
flow deflectors, the wind will be oriented to the advancing blade and blocked from the
returning blade. However, in this technique, the advanced features of the Savonius turbine
were eliminated such as its compactness size and the independence of wind direction.
Therefore, at the present, most of the research focuses on the optimizing blade shape such

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 433–438, 2023.
https://doi.org/10.1007/978-3-031-22200-9_48
434 B. D. Minh et al.

as an overlap blade type [5], a modified blade with variable thickness [6, 7], Bach-type
[8–10], elliptical and multi-curve type [11, 12], and multi-blade [13]. Compared with
the conventional turbine, the power coefficient C p of the Savonius rotor can be improved
up to hundreds of percent depending on the blade configuration [11]. However, previous
blade designs were only paid attention to specific operating conditions such as the low
TSR range of lower than 0.8 for rural application or high TSR > 1.2 for use in urban
environments.
In this study, various blade configurations for the Savonius rotor are investigated to
recognize the best configuration with the highest performance in wide operating ranges.
The multi-curve blade is designed, for which the main profile is made by a quarter circle
with a non-dimensional radius R3 * and a quarter elliptical section. The performance of
the optimized design and its advantage against the conventional one are analyzed utilizing
a sequence unsteady simulation in commercial software ANSYS Fluent 2021R2.

2 Geometry Design for Optimized Rotors with R3* Varies

Based on flow characteristics around the blade, and from previous research on the ellipti-
cal and multi-curve blade [11], the new multi-curve blade is designed. Figure 1 illustrates
a configuration of the conventional blade OB [14] (a) and the designed multi-curve blade
shapes in this study according to the non-dimensional radius of the quarter circular pro-
file R3 * . Here, the optimized blade is named MB02 to MB06 which is proportional
to the R3 * ranging from 0.2 to 0.6, respectively. The detailed designed parameters for
conventional and optimized blade configurations are presented in Table 1.

3 Numerical Simulation
3.1 Numerical Method
In this study, the flow around the rotor blade is numerically analyzed utilizing the
sequence unsteady simulation, which is performed using the commercial Computational
Fluid Dynamics software ANSYS Fluent 2021R2. For which, the governing equation
for a 2D incompressible unsteady Reynolds-Averaged Navier-Stokes equation is derived
as follows [11]:
∂ui
=0 (1)
∂xi
 
∂ui ∂   −1 ∂p ∂ ∂ui
+ ui uj = + v − ui uj (2)
∂t ∂xj ρ ∂xi ∂xj ∂xj

Here, p, u, u , and v are the pressure, the mean velocity, the fluctuation velocity, and
 
the kinematic viscosity of the wind flow, respectively. ui uj denoted the Reynolds stress
tensor. The realizable k-ε turbulence model and the enhanced wall function are selected
to account for the effect of the turbulence on the flow around the rotor [11].
The implicit pressure-based solver with the coupled method was used to solve the
governing equation above. The second-order upwind method was applied for the dis-
cretization of the convection equations. The least-square cell-based method was used
for the gradient. Besides, the second-order implicit method is used for the time variable.
The rotation of the rotor is simulated using the sliding mesh model.
 Effect of Main Blade Configuration on the Performance 435

Fig. 1. Geometry and parameters of blade configurations (a–f)

Table 1. Details of rotor configuration design parameters [m]

d [m] D [m] t [m] h* R1 * R2 * R3 *

MB02 0.5 1 0.004 0.25 0.5 0.55 0.2
MB03 0.3
MB04 0.4
MB05 0.5
MB06 0.6
OB 0.5 1 0.004 – – – –

3.2 Computational Domain and Simulation Conditions

Figure 2 illustrates the computational domain, boundary conditions, and mesh details
used for the simulation. The domain includes a rotating zone of size 1.2D placed at a
distance of 6D and 10D from the inlet and outlet boundaries, respectively. The stationary
zone has a size of 16D × 14D with two side boundaries located symmetry from the rotor
center. These two zones are connected by an interface model.
At the inlet boundary, the wind velocity U 0 = 7 m/s [14], and the turbulent quantities
are specified [11, 13]. The symmetry condition is applied at the two-side boundaries,
meanwhile, the no-slip condition is used for the rotor blade. Depending on the specific
TSR, the rotating zone has different rotational speeds ω, as shown in Table 2. The rotation
436 B. D. Minh et al.

of 20 /step is set for the time step size. And the simulation is performed for the completion
of 10 rotations.
The unstructured mesh is generated over the domain. Regarding the mesh indepen-
dency study, the mesh is generally fine at the rotor-stator interface with the minimum
size of 0.005m and at the blade shape with the minimum size of 0.001 m. The coarse
mesh is used for the stationary zone. Near the blade surface, the structured mesh with
20 layers, in which the first layer thickness is 10–5 corresponding to mesh resolution of
y+ < 1, is generated.

Fig. 2. Computational domain, boundary conditions and mesh detail around rotor

Table 2. Rotational speed of rotating zone, ω [rad/s]

λ 0.67 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
ω 9.38 11.2 12.6 14 15.4 16.8 16.8 19.6 21 22.4

4 Results and Discussions

4.1 Torque Distribution in One Period Rotating

Figure 3 illustrates the comparison of the averaged moment C m and power C p coefficients
between the conventional and the optimized blade configurations with different R3 * at
various TSRs. For which, C m decreases as TSR gradually increases. Rotor MB02 has the
lowest C m of configurations considered in all TSRs. Meanwhile, at TSR > 1.0, rotors
MB04, MB05, and MB06 perform significantly better than the other rotor. C m decreases
 Effect of Main Blade Configuration on the Performance 437

slowly with a gentler slope at TSR between 1.0 and 1.4 in these 3 rotors. Although the
optimized MB04 rotor has a lower C p than the previously studied MB05 rotor [11] at
TSR from 1.0 to 1.4, this rotor allows more efficient operation in wider wind conditions.
Specifically, the MB04 rotor gives greater efficiency at TSRs < 1.0 and TSRs > 1.4.
The best performance that the MB04 rotor offers improves up to 5.5% at TSR < 1.0 and
up to 185.1% at TSR > 1.4 when compared to the OB rotor. This greatly increases the
energy efficiency that this turbine produces at both low TSRs and high TSRs, thereby
demonstrating the applicability of the rotor MB04 under wider operating conditions than
in previous studies.

Fig. 3. Averaged C p and C m with R3 * varied

Fig. 4. Torque distribution in one period rotating

Figure 4 presents the distribution of torque in one rotation at TSR 0.8 and 1.5 between
the rotor with blades MB04, MB05, and MB06. At TSR 0.8, the rotor with blade MB04
produces a higher torque area than that with blades MB05 and MB06 at azimuths from
80° to 175° and from 260° to 355°. Similar behavior can be seen at TSR = 1.5. For
which the area with the higher torque in the rotor with blade MB04 is much larger than
the area with the higher torque in the rotor with blades MB05 and MB06, responding to
better performance on the MB04 rotor, as shown in Fig. 3.
438 B. D. Minh et al.

5 Conclusion
In this study, a multi-curve blade configuration that improves the performance of a
Savonius rotor is numerically recognized. The advanced feature of the optimized design
against the conventional one is analyzed through a sequence unsteady simulation using
the commercial CFD software Ansys Fluent 2021R2.
The results suppose a high dependency of the rotor performance on the main blade
configuration. The best performance is observed with blade MB04 with R3 * = 0.4 with
the increment in power coefficient by 185% at high TSR of greater than 1.0 and up to
5.5% at low TSR range of less than 0.8 over the other configurations. The result implies a
high potential application of the rotor with blade MB04 for energy harvesting application
in wide wind conditions.

References
1. Joselin Herbert, G.M., Iniyan, S., Sreevalsan, E., Rajapndian, S.: A review of wind energy
technology. Renew. Sustain. Energy. Rev. 11(6), 1117–45 (2007)
2. Savonius, S.J.: The S-rotor and its application. Mech. Eng. 53, 333–8 (1931)
3. Mohamed, M.H., Thevenin, D.: Performance optimization of a Savonius turbine considering
different shapes for frontal guiding plates. In: 10th International Congress of Fluid Dynamics,
Stella Di Mare Sea Club Hotel, Ain Soukhna, Red Sea, Egypt (2010)
4. El-Askary, W.A., Nasef, M.H., Abdel-Hamid, A.A., Gad, H.E.: Harvesting wind energy for
improving performance of Savonius rotor, J. Wind Eng. Ind. Appl. 139, 8–15 (2015)
5. Al-Faruk, A., Sharifian, A.: Blade overlap and blade angle on the aerodynamic coefficients
in vertical axis swirling type Savonius wind turbine. In: 19th Australasian Fluid Mechanics
Conference, Melbourne, Australia (2014)
6. Saeed, H.A.H., Elmekawy, A.M.N., Kassab, S.Z.: Numerical study of improving Savonius
turbine power coefficient by various blade shapes. Alex. Eng. J. 58, 429–441 (2019)
7. Kollmann, T.: J. In: Kollmann, T., Kuckertz, A., Stöckmann, C. (eds.) Gabler Kompakt-
Lexikon Unternehmensgründung, pp. 223–224. Springer, Wiesbaden (2021). https://doi.org/
10.1007/978-3-658-30901-5_10
8. Kamoji, M.A., Kedare, S.B., Prabhu, S.V.: Experimental investigations on modified Savonius
rotor. Appl. Energy 86(7–8), 1064–1073 (2009)
9. Kacprzak, K., Liskiewicz, G., Sobczak, K.: Numerical investigation of conventional and
modified Savonius wind turbines. Renew. Energy 60, 578–585 (2013)
10. Roy, S., Saha, U.K.: Wind tunnel experiments of a newly developed two-bladed Savonius-style
wind turbine. Appl. Energy 137, 117–125 (2015)
11. Anh, D.L., Minh, B.D., Tam, H.V., Hung, T.T.: Modified Savonius wind turbine for wind
energy harvesting in urban environments, ASME. J. Fluids Eng. 144(8), 081501 (2022)
12. Anh, D.L., Minh, B.D., Trinh, C.D.: High efficiency energy harvesting using a Savonius
turbine with multicurve and auxiliary blade, ASME. J. Fluids Eng. 144(11), 111207 (2022)
13. Hassanzadeh, R., Mohammadnejad, M., Mostafavi, S.: Comparison of various blade profiles
in a two-blade conventional Savonius wind turbine, ASME. J. Energy Resour. Technol 143(2),
021301 (2021)
14. Blackwell, B.F., Sheldahl, R.E., Feltz, L.V.: Wind tunnel performance data for two and three-
bucket Savonius rotors, SAND76-01321 UC-60, National Tech. Information Service, U. S.
Dept. Commerce, Springfield, VA 22161 (1977)
 Effect of Mass Reduction and Aerodynamic
Drag Coefficient on Fuel Consumption
Reduction on Student Fuel-Saving Car

Nguyen Thanh Quang(B) , Ngo Quang Tao, and Le Hong Quan

Hanoi University of Industry, Hanoi, Vietnam

nguyenthanhquang@haui.edu.vn

Abstract. This study analyzes scientific basis of fuel-saving racing vehicles.

Every year, Honda corporation organizes an Asian competition with the purpose
of choosing a model of the most efficiency. Students representing Asian countries
take part in this competition. Haui Super Cub’s of Hanoi University of Industry has
been highly appreciated in 2020 competition. The vehicle consumes 1352 km/liter.
Research method includes: Using drag coefficient C D to analyze the affect of
the air drag force L D to the fuel consumption S G and, analyzing the affect of the
composite material to vehicle mass.
The results of this research will be used to improve the car’s designing and
building in following competitions.

Keywords: Fuel–saving car · Aerodynamic CFD · Reducing mass · Composite

material

1 Introduction

Honda Corporation organizes competitions of fuel-saving cars (often called eco-car) with
slogan “the lowest fuel consumption”. International students, Vietnamese students and
students from Hanoi University of Industry actively take part in these competitions. Most
of students are inexperienced. Therefore, many difficulties are ahead. The tendency of
designing eco racing car focuses on following main issues: speed increasing and reducing
fuel consumption. The solution is to optimize aerodynamic shape and to reduce vehicle’s
weight by using lightweight materials.
Computer simulation is a key tool in designing, manufacturing and vehicle’s control.
There are different softwares and simulation methods, which analyze vehicle’s kinetics
and dynamics. A study chooses aluminum for vehicle’s body with the purpose to reduce
weight. Using RAPID PROTOTYPING technique in Hypermesh software can replace
the real manufacturing of prototype. In the results, the vehicle’s weight has reduced
by 5% while the durability (stress) has increased by 15% [1]. Calculating forces act-
ing on racing car with reasonable accuracy helps to reduce the motion drag. That is
important in designing suspension, driving and breaking systems. The research team
uses GUI Simmechanics to simulate 3-D vehicle in MATLAB/Simulink environment

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 439–447, 2023.
https://doi.org/10.1007/978-3-031-22200-9_49
440 N. T. Quang et al.

[2]. In recent years, computational fluid dynamics methods (CFD methods), supported
by algorithm development and big capability of computers, are popularly used. Students
have effectively used these methods in designing racing car [3]. Students also try to study
fuel-saving engines. A solution of adjusting carburetor injector has been considered and
is also effective [4, 5]. A study of kinetic characteristics of fuel injection regulator, by
simulating the calculation of the regulator’s parameters in an engine of new genera-
tion, using ANSYS software, has found that changes in the temperature of the oxidizing
agent, which is supplied to gas mixture generator from the fuel consumption generator,
affect the fuel pump power [6]. However, its hydrodynamics forces are not taken into
account and this fact reduces the accuracy of calculating parameters and characteris-
tics of the regulator. This study eliminates this distance by using package calculation
of hydrodynamics forces. The consumption regulator is assembled to the fuel supply
line in two-component AE gas generator. In mathematical model of the regulator, the
hydrodynamic force, generated during the fuel discharge through the delta window of the
differential valve, has considerable impact on the static characteristics of the regulator
and its stability. Let’s calculate the hydrodynamics forces acting on the differential valve.
A geometry model of the regulator has been developed to meet the requirements of the
calculating model. From that its hydraulic domain has been developed in CAD-module
of Solid Works.

2 Scientific Basis of Mass Reduction

2.1 Drag Forces in Racing Car

During moving in flat road, following drag forces act on racing car: air drag force (L D
= 0.5.ρ.CW.A.v3 ), inertial force Pj (Pj = δ.G.j.v), rolling force (Pf = f.G.v), Fig. 1.
Where: drag forces in (N), ρ–air density (kg/m3); C W –aerodynamic drag coefficient;
A–vehicle’ front drag area (m2 ); δ–coefficient affected by the mass translational and
rotational moving; G–vehicle’s weight (kg), j–moving acceleration (m/s2 ), v–vehicle’s
speed (km/h), f –rolling drag force, C the vehicle’s center of gravity, [7].

Fig.1. Drag forces in racing car

To achieve the goal of competition, cars are designed to increase speed and reduce
fuel consumption by following measures: reducing air drag force L D by aerodynamic
shape with smallest front area A and reducing C W coefficient; reducing inertial and
rolling drag force by using lightweight material to reduce the vehicle’s weight G.
 Effect of Mass Reduction and Aerodynamic Drag 441

2.2 Reducing Air Drag Force LD

During running at high speed, the air hitting vehicle’s body will be accelerated and
deflected, creating forces, jets, that creates turbulent air flow and air drag force L D .
There are 3 main factors affecting L D : vehicle’s dimensions and shape; air density;
vehicle’s speed. In designing, it is necessary to evaluate the influence of drag force L D
to vehicle’s speed and fuel consumption. Necessary engine power for a car to move is
proportional to the 3rd power of its speed and is approximated in Eq. (1).

CD Av3
P= (1)
1.225

P–Engine power (kW), A–front resistance area (m2 ), v–vehicle’s speed (m/s), [8, 9].
Engine power Pe , the hourly fuel consumption S G is calculated in Eq. (2).
1000 × SG 1
Pe = × (2)
ge k1 k2

S G (kg/h)–hourly fuel consumption at engine power Pe (HP); ge (g/ml.h)—fuel con-

sumption level at 85% of the max power; k 1 = 1.0–1.3—adjustment coefficient of fuel
consumption level at different power levels; k 2 = 1.0–1.05—adjustment coefficient of
fuel consumption according to the engine’s working time.
Combining (1) and (2) at engine power Pe , obtained Eq. (3) showing correlation of
hourly fuel consumption and vehicle’s speed.
1 CD A 3
SG = (ge k1 k2 ) × v (3)
1000 1.225
Specific fuel consumption is calculated by formula (4).
1
ge = (4)
QH ηe
where, QH is the fuel consumption in an unit of time, ηe is the engine output.
At a constant speed, the fuel consumption depends on drag coefficient C D . If C D
decreases, the fuel consumption reduces accordingly. Equation (3) can be applied to
choose engines for motobikes with fuel consumption S G in Table 1.
Air drag consists of 3 components: Lift force L f –which tends to lift the vehicle
and reduce the wheel’s grip to road surface; down force L S –which tends to press down
vehicle’s body and increase the wheel’s grip to road surface; front drag L D –which tends to
reduce movement speed and increase fuel consumption (designers should pay ettention
to this force). Forces are calculated by Eq. (5).

Lf = CL × 1/2ρAV 2
LS = CS × 1/2ρAV 2 (5)
LD = CD × 1/2ρAV 2
442 N. T. Quang et al.

Table 1. Fuel consumption of some motobike enginesa

No. Name of engines Fuel consumption G (L/100 km)

1 Honda Wave Alpha 110 1.6
2 Honda Wave RSX Fi 1.7
3 Honda Super Cub C125 1.73
5 Yamaha Jupiter Fi 1.55
6 Yamaha Sirius Fi 1.57
a Source Vietnam Registry Department [17]

Lift, down and drag coefficients are calculated by Eq. (6). [10]
Lf
CL = 0.5Lf /ρAV 2 = 1
2 ρAV 2

LS
CS = 0.5LS /ρAV 2 = (6)
2 ρAV
1 2

LD
CD = 0.5LD /ρAV 2 =
2 ρAV
1 2

In case of running at constant speed, drag force LD is a function of front resistance

area A and drag coefficient CD in Eq. (7).

LD (min) = 1/2ρV 2 × f (A, CD ) (7)

There are 3 ways to resolve Eq. (7) to find out the smallest aerodynamic frontal drag
force L D . The 1st is an algebraic method solving the conditional extreme problem of
a function of 2 variables. The 2nd is an experimental method. The 3rd is numeracal
simulation method using CFD tool in Ansys Workbench and, encourage this method
because of its modernity, [18].

2.3 Reducing the Body’s Mass

Today lightweight meterials are popularly used to reduce vehicle’s mass and reduce
inertial and rolling drag. In Fig. 1 above, if vehicle mass G reduces then inertial drag PJ
and rolling drag Pf in running.
Lightweight materials, such as composite, is usually used in manufacturing force-
bearing components and complete block units, such as body, frame and some components
in transmission system. Plastic is used in decorative components, seats, front and rear
bumpers, lights and air-conditioning system covers. Composite has following advan-
tages: easy fabrication, high aesthetics, high drag, anti-oxidation, anti-discoloration
under ultraviolet rays. The advantages of plastic are low cost and easy replace.
 Effect of Mass Reduction and Aerodynamic Drag 443

3 Results
3.1 Designing Using CAD
Total vehicle’s mass mn is equal to the sum of the vehicle’s body mass m1 with the
driver’s mass m2 and the mass of other additional equipment. Vehicle’s layout design
CAD is in Fig. 2.

Fig. 2. Designing vehicle’s layout

The coordinate of vehicle’s center of gravity can be determined by means of geome-

try, by experiments or by modeling in 3-D design software. The results of designing by
CAD show parameters in Table 2.

Table 2. Parameters of designing eco racing car

No. Complete block unit Unit Value Remark

1 Body and engine m1 kg 29 Body 7 kg, covers 7 kg, engine
and driving assistance 15kg
2 Driver m2 kg 55
3 Total mass mn kg 84
4 Dimension DxRxC mm 2600x550x70 Should not be over 3500 × 1700
× 1800
5 Wheel base L mm 1500 Should not be under 1000
7 Height of center hm mm 250 Equal to the height of driver’s
seat
8 Material of frame/cover Steel S355JR/composite

In the design, the vehicle’s coordinate origin is contact point of front tire with road
surface. Front wheel is the directive wheel. Point x is the center of gravity. The vehicle’s
height is enough for one driver. Other dimensions can be changed. Figure 3 [15].

3.2 Manufacturing
3.3 Prototyping Process
The prototyping process has following steps: Step 1: preparing materials including fuber-
glass, plastic, glues. Step 2: preparing manufacturing place. Step 3: creating open mould
444 N. T. Quang et al.

Fig. 3. Designing racing car by CAD

from foam blocks. Step 4: creating mould release layer with paraffin. Step 5: creating
reforcements of composite and fiberglass. Step 6: filling the composite into mould sur-
face. Step 7: mixing polyester 8201 with butanoc solidifier (ratio 1:1.5). Step 8: coating
the mixture on fiberglass reinforcement surface. Step 9: creating layers with dimensions
specified in the design. Step 10: drying in natural environment (about 10–12 h), releas-
ing the mould. Step 11: smoothing the cover with skim coat and paint. Step 12: final
assembly of frame and other components. Manufacturing process and complete car are
shown in Fig. 4.

a) Mould of body b) Vehicle’s frame

c) Vehicle’s body d) Complete vehicle

Fig. 4. Manufacturing process and complete vehicle

3.4 Assessing PRODUct’s Quality

The car has been assessed in Asian eco car competition 2016, organized by Honda
Corporation in Japan. The result was 2948 km/L.
From formula (3) above, provided that the aerodynamic drag coefficient C D in case
A = 0,3 m2 , vehicle’s velocity v = 60 km/h, S G = 0.000339 L/km, ge = 21 g/CVh, k 1
= 1.1, k 2 = 1.05 obtained by Eq. (8).
1000 ∗ 1.225 ∗ SG
CD = = 0.192 (8)
(ge k1 k2 )Av3
 Effect of Mass Reduction and Aerodynamic Drag 445

If C D reduces then the fuel consumption S G reduces accordingly.

3.5 Simulation of Eco Car

The car model participated and was highly evaluated in the eco-car competition. The
car has been assessed by numerical simulation method in the Fluid Ansys Workbench
2022R1 software, [16].

3.6 Reduce Mass Simulation

The 3-D model of eco racing car was designed by CAD. It is easy to change its dimensions
and shape. The model’s coordinate system is at the midpoint of front axis. The materials
of body and frame including composite, fiberglass, polyethylene 8210 and 2-component
glue. Body of eco car simulation in Fig. 5.

Fig. 5. Simulation eco car for mass reduce

Body mass calculation reduce with drag force L D and rolling force in Fig. 1 about
with difference materials result obtained in Table 3. Body mass reduces then the drag
force L D and Pf reduces and fuel consumption S G reduces accordingly.

Table 3. Reduce Drag force with difference materials

Density Body mass Drag force L D Rolling force Body mass

(kg/m3 ) (kg) (kG) Pf = f.G.v prototy (kg)
Equation (5) (kG)
Steel 7.85e−006 25.33 83.93 27.35 –
Composite 1.8e−006 21.38 19.35 23.09 29.0
PA6/glass
filber
Polyethylene 9.5e−007 20.83 10.16 22.50 –
446 N. T. Quang et al.

3.7 CFD Simulation

The CFD simulation can determine the front resistance area A (m2 ) and drag coefficient
C D in order to compare with that in the real car.
Parameters and commands simulation: Models/Visous/Model/k-epsilon (2 eqn); k-
epsilon Model/Realizable; Near-Wall Treatment/Enhanced Wall Treatment/OK. Mate-
rials/Fluid/air/Density 1.225 (kg/m3 ), Viscosity 1.7894e−05 (kg/m.s). Boundary Con-
ditions/Inlet/Velocity Inlet 16.667 m/s (60 km/h). Solution Methods/Coupled; Tur-
bulent Kinetic Energy: Second Order Upwind; Turbulent Dissipation Rate: Second
Order Upwind. Initialization. Calculation Activities/RunCalculation. Results/Graphics.
Reports. Meshing process is in automatic mode with a 4-node element type and 6 free-
dom degreee at each node. In the result, there are 321621 elements and 483361 nodes
obtained. The simulation determinded the front resistance area of 0.298 m2 (it is 0.3 m2
in the design) and drag force 1310 N.
Results of simulating racing car in Fig. 6.

a) Distribution of air presure Stream

lines around vehicle’s body
b) Graph of drag force (1310N)
Fig. 6. Results of the CFD simulation of eco-car

In the result of the simulation, the drag coefficient C D is shown in Eq. (17)
Drag 131
CD = = = 0.198 (17)
0.5ρAV 2 0.5 ∗ 1.225 ∗ (0.3e−6) ∗ 602

4 Conclusion
This study has analyzed the main scientifi basis in designing student’s eco racing car
Haui Super Cub of the Hanoi University of Industry. It, with fuel consumption of
906’880 km/L, was highly evaluated in Japan 2020.
The use of lightweight material, as composite, reduces vehicle’s mass and drag forces
as presented in basic movement equations. Analyzing the affect of the body mass to drag
force L D and rolling force Pf with difference materials steel, composite PA6/glass filber
and polyethylene. Body mass reduces then the drag force L D and Pf reduces and fuel
consumption S G reduces accordingly.
Analyzing the affect of the drag coefficient CD to fuel consumption SG. Determining
the actual drag coefficient of 0.192 while it is 0.198 as calculated in the simulation by
 Effect of Mass Reduction and Aerodynamic Drag 447

Ansys Worlbench, and hourly fuel consumption at the same vehicle’s speed reduce 4%
obtained. With this simulation tool, it is possible to adjust the designing of vehicle’s
body tio achieve more reasonable drag coefficient CD.
The achievements in this study will be used in designing and making new vehicles
for following competitions.

References
1. Prajwal, P.B.: Design optimization of formula one student sports car upright using hypermesh.
Int. J. Mech. Ind. Eng. (IJMIE) 2(2231–6477) (2012). https://doi.org/10.47893/IJMIE.2013.
1088
2. Madhu, A., Aravind, J.R.: Software tool development for estimating forces acting on a for-
mula student racing vehicle using simple vehicle dynamics models. Trans. Eng. 2, 100016
(2020)https://doi.org/10.1016/j.treng.2020.100016
3. Sverbilov, V., et al.: Study on dynamic behavior of a gas pressure relief valve for a big flow
rate. In: ASME/BATH, Symposium on Fluid Power and Motion Control FPMC (2013)
4. Goryachkin, A.A., Zhukovsky, A.E., Ignachkov, S.M., Shorin, V.P.: Flow regulators for fuel
systems of aircraft engines. Mech. Eng., 200 (2000)
5. Gimadiyev, A.G.: Choice of parameters, the calculation of the static and dynamic character-
istics for fuel regulator. State Aerospace, Samara, p. 63 (2007)
6. Lebedinsky, E.V., et al.: Computer models of liquid rocket engines. Mech. Eng., 376 (2009)
7. Milliken, W.F., Milliken, D.L.: Race Vehicle Dynamics, ISBN: 1-56091-526-9, SEA Order
No R-146 (1995)
8. McBeath, S.: Competition Car Data Logging, a Practical Handbook. Haynes Publishers, p. 25
(1998), ISBN 13: 9781844255658
9. McBeath, S.: Competition Car Aero Dynamics, a Practical Handbook, eBook edn. Veloce
Publishers, p. 57 (2017). ISBN: 978-1-787110-86-1
10. Katz, J.: Race Car Aerodynamics: Designing for Speed (Engineering and Performance), vol.
10, p. 47. Bentley Publishers.com (2003). ISBN: 0837601428, 9780837601427
11. Rat, iu, S.A, Zgaverdea, A.C.: The potential of using bio plastic materials in automotive appli-
cations. Mater. Plast. (2019). https://doi.org/10.37358/MP.19.4.5282. https://www.researchg
ate.net/publication/338940145
12. Tran, N.-T., Pham, N.T.-H.: Investigation of the effect of polycarbonate rate on mechanical
properties of polybutylene terephthalate/polycarbonate blends. Hindawi Int. J. Polym. Sci.
2021, 7 (2021). Article ID 7635048. https://doi.org/10.1155/2021/7635048
13. Sehanobish, K.: Engineering Plastics and Plastic Composites in Automotive Applications
(2009). ISBN 978-0-7680-1933-9, SAE Order No. T-122
14. Jackson, N.R., Finlayson, B.A.: Calculation of Hole pressure: I. Newtonian fluids. J. Non-
Newton. Fluid Mech. 10(1982), 55-69 (1982). https://doi.org/10.1016/0377-0257(82)850
04-0. Elsevier. All content following this page was uploaded by Bruce A. Finlayson on 30
March 2020
15. Seward, D.: Race car Design (2014). ISBN: 978-1-137-03014-6. Palgrave, UK
16. https://www.ansnys.com/academic/students
17. http://www.vr.org.vn
18. https://www.grc.nasa.gov/www/k-12/airplane/sized.html
 Effect of Composite Materials to Vehicle Body
Vibration

Nguyen Thanh Quang(B)

Hanoi University of Industry, Hanoi, Vietnam

nguyenthanhquang@haui.edu.vn

Abstract. Vibration is an important problem in automotive design and produc-

tion process, particularly in passenger bus. There are some factors that affect the
vibration behavior of the bus, one of which is the selection of main material for
the body. This paper presents analyses on how the passenger bus body vibration
behavior changes when composites are used as the main material. Finite Ele-
ment Method in Ansys Workbench to conduct the analyses used. Comparing the
frequency range and the magnitude of vibration in the two cases of using sheet
metals and composites as the main material for the bus body. The results have
contributed to the assisting apply of advanced materials in automotive design and
manufacturing.

Keywords: Bus body · Vibration · Steel · Composite material

1 Introduction

During the vehicle’s operation there are many vibrations generated by different sources.
Vibration can come from engine, powertrain, tires, and it generates a disturbing force.
The floor panel, gear shift and steering wheel transmit additional vibrations to the feet
and hands of the driver, [1]. The quality evaluation of vehicles is very important because
those parameters are quality vibration to the passengers [2, 3]. Vibration is considered
the most important one in determining the overall ride quality. Passengers in a moving
vehicle are continuously exposed to a vibratory environment caused by excitations from
road surface and from vehicle internal sources such as powertrain, the combination of
engine and transmission. The ride quality can also be evaluated by objective methods
including simulation and testing. Low frequency vibrations are perceivable to passengers,
while high-frequency vibrations involve the inner self of the parts or components. The
study low frequency vibrations often use suspension models of suspension systems such
as 1/4 model, 1/2 model and full 3D model. To improve these low frequency vibrations,
it is optimal to use the stiffness and damping coefficient parameters of the suspension
system. Also high-frequency vibrations give the parameters of stress and deformation
related to the durability of the parts as components and take them as a target function to
optimize the body design of vehicles. The low-frequency vibration of vehicles is usually
less than 100 Hz for vibration and 150 Hz for sound, [4, 5].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 448–454, 2023.
https://doi.org/10.1007/978-3-031-22200-9_50
 Effect of Composite Materials to Vehicle Body Vibration 449

2 Scientific Basis of Vibration on Body Vehicle

The bus vehicle assembled from six thin steel plates is formed a compartment with an
interior volume of 3.6 cubic meters, Fig. 1.

a) Bus b) Body structure

Fig. 1. Bus vehicle body

The body of vehicle is a complex parameter system, where the corresponding motion
equation of is a multiple-degree-of-freedom system can be represented as Eq. (1), [4, 5].
   
[Me ] Üe + [Ce ] U̇e + [Ke ]{Ue } = {Fe } (1)

In which [Me], [Ce], [Ke],{Fe} are respectively the structural mass matrix, structural
damping matrix, stiffness matrix, and external excitation vector.
The analysis method for a lumped-parameter system allows to establish the relation-
ship among system physical parameters, transfer functions, which helps to quantify the
properties of system vibrations. The finite element method is effective for the analysis
of structural vibrations of low frequency, Eq. (2).

{Ue } = [ϕ]{q} (2)

in which, [ϕ] is the modal matrix, {q} is the modal vector.

The natural frequency ωn of structure we obtained by (3)

n2 π 2 c EI
ωn = =n π
2 2
(3)
l 2 ρl 4

where n is coefficient, n = 1, 2, 3, …, E iselastic modulus of material, I is inertia moment

of structure, ρ is density, lis parameter of structure.
Each of the frequencies corresponds to a principal mode of body structure vibration.
The natural frequency is a function of elastic stiffness and the mass of the structure. The
relationship between a structure’s mass and its resonant frequencies is shown by Eq. (4).

k
ωn2 = (4)
M
where, ωn : Resonant frequency, k: Stiffness, M: Mass.
450 N. T. Quang

3 Material Properties
Using Ashby’s method in deriving the material selection, composite materials with low
specific gravities, offer a combination of strength and modulus to weight-ratios that are
either comparable better than metals. Automobile manufacturers and material engineers
have proposed several formulas and simulation results for testing material selection
for a dent resistance of vehicle. One of the most acceptable empirical formulas is that
presented by Eqs. (5) and (6), [2].

kσy2 t 4
R= (5)
S

S = Et α H (6)

where R is the denting energy, k is a constant, σ y is the material yield strength, t is the
panel thickness, S is the panel stiffness, α is the power adjustment of the thickness, E is
Young’s modulus of elasticity, and H is the shape factor. The original materials for the
construction of the bus body were 1.0 mm thin steel plates. The alternatives are three
types of composites (1) Composite, Epoxy/carbon fiber, resin infused fabric, biaxial
(abbreviated as Composite, E/c-f,fb); (2) Composite, Epoxy/glass fiber, UD prepreg, QI
(Composite, E/g-f, QI); (3) Epoxy E-glass UD is a multi-dimensional material, the main
characteristics of the materials are shown in Table 1, [2, 6].

Table 1. Properties of Materials

Str. steel Com.E/c.f,fb Com.E/g.f E-glass UD

Density (kg mm^−3) 7.850e−006 1.515e−006 1.904e−006 2.0e−006
Young’s modulus (MPa) 200000 60930 19970 X;Y;Z
Poisson’s ratio 0.3 7.616e−002 0.3065 XY;YZ;XZ
Tensile yield St(MPa) 250 642.1 479.9 X;Y;Z
Tensile ultimate St(MPa) 460 642.1 479.9 X;Y;Z

4 Simulation
4.1 Model of Body Structure
Automotive assembly lines need a high customization. Modular assembly is developed
in lieu of conventional assembly lines due to the following characteristics: (1) Can
be arranged as independent module assembly stations. Independent stations are only
affected by assembled products, regardless of the fixed transport system or the general
time of the line. (2) Use automated guided vehicles (AGV) in the module transport.
(3) Modular assembly lines can be moved by assembly stations.It is assumed that steel
 Effect of Composite Materials to Vehicle Body Vibration 451

plates, frames and glassis plates are rigidly assembled together and are not influenced by
auxiliary assemblies in the interior of the vehicle such as cladding, interior components,
ceilings, seats. And furniture. The solution of vibration requires computer calculation,
as the frequency calculation of each panel must take into consideration the other ele-
ments. Modeling of the entire vehicle structure applies the finite element method (FEM).
Transfer path analysis (TPA) is an effective approximation method for vehicle vibration
analysis by using vector manipulation. TPA can also be implemented by partly using the
results from finite element analysis. Meshed and boundary condition applied 3D model
is imported to the ANSYS Workbench software. The modelling of county bus structure,
model meshing shown in Fig. 2(a), strain and stress shown in Fig. 2(b) for elements size
is 10 mm, 54597 nodes and 51583 elements.

a) Meshing of model b) The strain and stress of model

Fig. 2. Modelling county bus body structure

The parameters of of county bus structure shown in Table 2.

Table 2. Parameters of of county bus structure

Properties Statistics
Volume (m3 ) 1.7157e−003 Bodies 19
Mass of steel (kg) 1350 Nodes 215154
Scale factor value 1 Elements 95308

4.2 Force and Boundary Condition

During vehicle’s running on roads there are impulces from road surface to tires front
left, front right, rear left and rear right, and to 4 points in chassis at suspension contact.
Load is sinusoidal excitation impulse, Fig. 3, [7].

4.3 Analysis of Result

Assessing the quality of vehicles according to vibration criteria by customers perception
is always a matter of need and requires quantification. Frequency vibrations in the range
below 100 Hz are assessed by the intensity of vibration, over 100 Hz to thousands of Hz
(10.0 kHz) is considered to be loud sound with sharp intensity [8].
452 N. T. Quang

Fig. 3. Load sinusoidal excitation impulse

Result of simulation is frequency range on bus body by color spectrum distribution

for easy-to-observe. On Fig. 4 is a result of vibration at 50 Hz with maximum amplitude
of vertical deformation is 69.155 mm.

Fig. 4. Vibrations at 50 Hz freq. For deformation 69.155 mm.

Considering which frequency range will follow the customer’s regional criteria, for
vibration survey Bus County in the frequency range from 0 to 100 Hz and receive
different types of vibration according to each type of material of vehicle’s body, Fig. 5.

Fig. 5. Amplitude of vertical vibrations to Y direction of 4 types of materials

 Effect of Composite Materials to Vehicle Body Vibration 453

5 Experiment
The rapid reciprocal method has been used in the interior vibration test. An accelerometer
is located on the floor of the vehicle. The measured results are the vehicle floor oscillation
acceleration signals over time, Fig. 6.
The experimental results were processed by spectral analysis and Fourier transform
to obtain the data in the frequency region. Test at least 3 measurements for each mode
and average for that mode.

Fig. 6. Setup acceleration sensors

The vehicle floor vibration acceleration value is evaluated by the mean squared RMS
(Root mean square) according to formula (7).
 

1 T  2 
XRMS = xi dt (7)
T 0

where: XRMS–mean square of the acceleration; x i –the value of the acceleration, with i
= 1,2,3,….,T. The relationship between vibration measurement data and displacement
is calculated by the formula (8).
j j
df = = (8)
ω2 10−3 10−3 (2π fn )2
where: d f Deformation (mm), jVibration accelaration (mm/s2), f n Frequency (Hz).
The vehicle floor vibration acceleration values are measured over time. Using the fast
Furie transform in the range of 100 (Hz) and taking the mean squared value according to
formula (7) we get the graph results of relationship between acceleration and frequency
of vibration. Figure 7.(a) is the test result of a case of thin sheet steel. Figure 7.(b) is
the result of vibration measured in the interval of 30 (Hz) compared with Fig. 4 about
simulation result shows the maximum value of vibration amplitude above 2.00e−02
(mm) with a difference of less than 10% is quite accuracy of the simulation method.

6 Conclusions

The main objective of this research work has been satisfied. The comparison among
vibration analysis results of existing Composite, Epoxy/carbon fiber, resin infused fabric,
454 N. T. Quang

Fig. 7. Results vibration of experiment measured

biaxial (abbreviated as Composite, E/c-f,fb); Composite, Epoxy/glass fiber, UD prepreg,

QI (Composite, E/g-f, QI); Epoxy E-glass UD, a multi-dimensional material, has been
performed and it is summarized that E-glass (UD) is proven to be better in frequencies
vibration behaviour. Body panels of muntidirectional materials Epoxy E-glass UD have
amplitude of vibration stable pattern in all frequencies range. Thus, using Epoxy E-glass
UD materials is used for high quality vibration of bus body.
Using the experiment gives less than a 10% deviation from the simulation, which
proves that the research results are reliable.
Using this material will be very beneficial due to the easy parameters change of
sheet thickness, reinforcement, hardened pieces on the surface and simply processing to
help avoid the vehicle with the nearby of frequency vibration excited forces. Analysis
Vibration of Bus Body Structure with using Difference Materials help design engineering
with composite which is considered to avoiding any resonance to occur and also increase
durability for the all of the bus body structure.

References
1. Burdzik, R., Konieczny, Ł.: Vibration Issues in Passenger Car, Silesian University of
Technology, Faculty of Transport, Krasiński 8, 40-019 Katowice, Poland, vol. 9, Iss. 3 (2014)
2. Njuguna, J.: Lightweight Composite Structures in Transport, Design, Manufacturing, Analysis
and Performance, Science and Engineering: Number 67 (2016). ISBN: 978-1-78242-325-6
(print), ISBN: 978-1-78242- 343-0 (online), Woodhead Publishing, Elsevier
3. Mallick, P.K.: Composites engineering handbook. In: Library of Congress Cataloging-in-
Publication Data (1997). ISBN 0-8247-9304-8
4. Sheng, G. (Gang Sheng Chen): Vehicle Noise, Vibration, and Sound Quality. SAE International
(2012). ISBN 978-0-7680-3484-4, SAE Order No. R-400. https://doi.org/10.4271/R-400
5. Timoshenko, S., Young, D.H.: Vibration Problems in Engineering. D. Van Nostrand Company
Inc., New Jersey (1955)
6. Kaw, A.K.: Mechanics of Composite Materials, 2nd edn. Taylor & Francis Group (2006). ISBN
0-8493-1343-0
7. Quang, N.T.: Finite element analysis in automobile chassis design. Appl. Mech. Mater. 889,
461–468 (2019). ISSN: 1662-7482. https://doi.org/10.4028/www.scientific.net/AMM.889.461
8. Zwicker, E., Fastl, H.: Psychoacoustics. Springer-Verlag Pub., Facts and Models, Berlin (1999)
 Effect of Printing Parameters on Characteristics
of PCL Scaffold Fabricated by Direct Powder
Extrusion

Quang Duy Do, Du Khac Nguyen, Tung Van Nguyen, Lan Xuan Phung,
and Trung Kien Nguyen(B)

School of Mechanical Engineering, Hanoi University of Science and Technology, Hanoi,

Vietnam
trung.nguyenkien@hust.edu.vn

Abstract. Nowadays, 3D printing technology is becoming more popular with

various applications. 3D printing is also a potential process in tissue engineering
applications for fabricating scaffold. The scaffold needs to have the proper struc-
ture and properties to initiate a favorable environment for cell growth and tissue
generation. In this study, the customized direct powder extrusion based on screw
extruder was developed and the influence of printing parameters including print-
ing speed and temperature on characteristics of scaffolds from PCL powder was
investigated. The result evaluation shows that the size of line width and pore size
are significantly affected by printing parameters. The optimal printing conditions
were obtained at print speed 6 mm/s, extrusion speed 50 rpm, and heating temper-
ature of 100 °C to fabricate scaffolds from PCL powder materials. The results are
the basis for further studies in optimizing the printing condition for PCL scaffolds
based on direct powder extrusion.

Keywords: PCL scaffold · 3D printing · Screw extruder

1 Introduction
Tissue engineering is an emerging and potential field to provide a replacement for dam-
aged tissues or regenerative medicine [1]. Three basic components of tissue engineering:
cells, supporting framework (scaffold), and biochemical active ingredients. The scaffold
is a framework for cells to adhere, grow and form functional tissues. The bioactive
molecules and growth ingredients support and regulate the desired tissue formation [2].
These three basic components can be used individually or in combination to regener-
ate organs or tissues. Among the three mentioned components, the scaffold plays an
important role for cell adhesion, proliferation, and new substrate creation. They also
are biodegradable to allow cells to produce their extracellular matrix. The scaffold must
have adequate mechanical properties for surgical handling of the implant. Depending on
the type of and characteristics of engineered tissue, these scaffolds are different in the
external and internal structure, mechanical and material properties corresponding to the
tissue to be created. For scaffold fabricated by 3D printing techniques, the printability

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 455–461, 2023.
https://doi.org/10.1007/978-3-031-22200-9_51
456 Q. D. Do et al.

of biomaterials is significantly important to control the accuracy of geometry character-

istics of scaffold such as line width and pore size. The line width ensures the surface
area for cell adhesion. The pore size needs to be large enough to allow cells to migrate
into the structure, where they eventually attach to the support scaffold structure to form
a particular tissue. The scaffold also needs to have an interconnected pore structure
and high porosity to ensure cell penetration and adequate diffusion of nutrients to the
intracellular cells and extracellular substrate [3].
Using 3D printing technology, the scaffolds are easily created with controllable geo-
metrical properties thereby the technical and biological properties can be modified [4].
The high-resolution scaffold can be made from stereolithography (SLA) by solidifying
the photopolymer liquid using UV light. The selective laser sintering (SLS) is a powder-
based prototyping technology in which SLS technology uses a laser to sinter different
types of powdered materials to create a solid pattern. Both SLA and SLS method has
low flexibility in combination of other printhead to fabricate the complex scaffold made
of multi-biomaterials for practical tissue engineering application. The scaffolds are also
made from fused deposition modeling (FDM) printing technology by extruding the melt
material through the nozzle to generate the solidified material on the substrate. FDM is a
low-cost 3D printing technology with a fast speed but low accuracy and low availability
of biomaterial in filament form. Overcoming the limitation of material availability, the
direct powder extrusion 3D printing method developed from the FDM method is devel-
oped. This method can be directly fabrication scaffold made of multi-materials without
filament preparation. This paper presents a customized direct powder extrusion based
on screw extruder to fabricate polycaprolactone (PCL) scaffolds and analyze the effect
of printing parameters on the geometry characteristics of scaffold.

2 Methods

2.1 Design PCL Powder-Based Printing Head

Since the powder extruder can print directly from commercially available powdered
or granular thermoplastic biomaterials, the 3D printing method from powder materials
presents a great commercial advantage of the material. Although the FDM based printing
technology of powder material is complicated with the strict requirement of heat control,
is possible to directly use the combination of other materials in order to change the
biological properties of the printing material.
The common design of the screw extrusion printhead includes the feed hopper,
barrel, extrusion screw, heater, heat sink, and nozzle as show in Fig. 1a. In this study, the
copper extrusion nozzle with a diameter of 0.3 mm is mounted on the heating chamber
including a heater and a temperature sensor. The barrel and feed hopper are responsible
for feeding the powder material to the heating region from the hopper at low temperature
to the melting zone near the nozzle area. The barrel is divided into two parts Teflon upper
side and Aluminum lower side and connected to the heating chamber through a threaded
hole. With this novel design structure, the material flow inside the barrel is kept in a
steady state, reducing the problem of heat transfer from the heating chamber to the feed
hopper.
 Effect of Printing Parameters on Characteristics of PCL Scaffold … 457

The extrusion screws are designed depending on the powder used, its transition
temperature and melting point. Each type of powder requires a different type of extrusion
screw with different diameters and lengths to control the feeding and heating region.
Another very important parameter for the extruder working is the gap between the
threads of the screw and the inner surface of the barrel since the material flow is highly
dependent on this gap. In this research, the screw is designed with total length 80 mm,
screw length 55 mm and diameter 8 mm.

2.2 Thermal Distribution Analysis

One heat source is supplied to the printhead by a 50 W heating wire and a temperature
sensor accuracy within ± 5 °C from the setup temperature. The heat transfer process
and temperatures of the plastic extruder were investigated by using Solid works software
with the FFEPlus solver. Figure 1b shows the thermal distribution simulation result
in the extruder at 300 s after the heating started. The heating area ensures a stable
temperature at 100 °C, while the feeding area is completely insulated. Table 1 summarizes
the temperature at the heating region in 300 s.

Temp (°C)
120

100

80

25
a) The extruder structure b) Thermal distribution

Fig. 1. Design structure and thermal simulation

Table 1. Heating process

Time (s) Temperature (°C)

Heating point Sensor point
50 80 50
100 120 80
150 110 90
200 105 95
250 100 100
300 100 100
458 Q. D. Do et al.

The temperature of the heating point comes up to 120 °C after about 100 s and starts
to gradually decrease to 100 °C when the heating point was reached 100±5 °C then
stabilizes at this temperature throughout the printing process. The thermal simulation
results show that the stability of the powder extruder ensures a continuous printing
process and eliminates the problem of heat transfer from the heating chamber to the feed
hopper causing clogging of the powder print head.

2.3 Experimental Work Design

In the experimental work, the printing tests were conducted on a nozzle diameter of
0.3 mm with white powder PCL material input. The scaffold layer size of 10 x 10 mm
structure, the number of printing layers is three layers alternating 0°−90°. The effect of
printing speed on fabricated scaffolds’ dimensions such as line width, pore size in upper
and lower layers is investigated through three areas as shown in Fig. 2.

Fig. 2. The analysis area on the printing sample

Figure 3 shows line width and pore size under different printing conditions. The
results show that the printing condition with T = 100 °C, V = 6 mm/s, and E = 8
rpm (Fig. 3b) provides the line width and pore sizes closest to the target size (0.3 mm).
Therefore, the printing temperature T = 100 °C and extrusion speed E = 8 rpm is kept
for further analysis with a variation on the printing speeds at 4 mm/s, 6 mm/s, 8 mm/s,
and 10 mm/s. The printed samples were measured using a Kruss microscope, and the
data for analysis is conducted with Imagefocus plus V2 measurement software. For each
print speed, three samples were measured, each sample will measure two consecutive
layers (L1, L2) and each print layer was measured in three areas (A1, A2 and A3 in
Fig. 2).

3 Result and Discussion

Figure 4 shows the influence of printing speed on the accuracy of the printed. The effect
of the printing speed on the line width and pore size is presented in Fig. 5 and Table 2. The
 Effect of Printing Parameters on Characteristics of PCL Scaffold … 459

a) T=80 °C b) T=100 °C c) T=120 °C d) T=100 °C e) T=100 °C

V= 6mm/s V= 6mm/s V= 6mm/s V= 6mm/s V= 6mm/s
E=8rpm E=8rpm E=8rpm E=5rpm E=12rpm

Fig. 3. Line width and pore size at printing conditions

result shows that the line width tends to decrease while the pore size tends to decrease
with increasing printing speed.

Fig. 4. The line width and pore size at different printing speeds

The printed line tends to be stretched when printing at high speed. The higher the
printing speed is, the more the material is pulled. The porosity result of printed samples
shows that the porosity increases as the printing speed increases. The printing speed V4
provides the highest porosity result on scaffold (71%). The printing speed V1 and V2
can obtain the better line width than printing speed V3 and V4. Both line width and pore
size are the smallest for printing layers at printing condition V2.
460 Q. D. Do et al.

Fig. 5. The distribution of line width and pore size at different printing speeds

Table 2. Experimental data of the sample at different printing speeds

Exp Speed Temp Extrusion Average Line Average Pore Mass of Porosity
(mm/s) (°C) speed width (mm) size (mm) sample (%)
(rpm) (g)
V1 4 100 8 0.311 ± 0.007 0.273 ± 0.008 0.07 32
V2 6 100 8 0.303 ± 0.013 0.284 ± 0.013 0.05 51
V3 8 100 8 0.264 ± 0.006 0.318 ± 0.008 0.04 61
V4 10 100 8 0.253 ± 0.037 0.327 ± 0.017 0.03 71

4 Conclusion
The direct powder extrusion based on screw extruder to fabricate PCL scaffolds is devel-
oped. The novel design of print head can solve heat transfer problem on the screw
extruder. The study also investigates the effects of temperature, and printing speed on
the line width, pore size, and porosity. The experimental results show that the optimal
printing conditions to fabricate scaffolds from powdered PCL materials at a temperature
of 100 °C, printing speed of 6mm/s, and extrusion speed of 8rpm. These results would
be the experimental recommendation for fabricating 3D scaffolds with controllable and
precise geometries in tissue engineering application.

Acknowledgments. This research is funded by the Vingroup Innovation Foundation (VINIF)

through project VINIF.2020.DA13.

References
1. Gleadall, A., et al.: Review of additive manufactured tissue engineering scaffolds: relationship
between geometry and performance. Burns Trauma 6, 1–16 (2018)
2. Jahani, B., Wang, X., Brooks, A.: Additive manufacturing techniques for fabrication of bone
scaffolds for tissue engineering applications. Recent Prog. Mater. 2(3), 1–41 (2020)
 Effect of Printing Parameters on Characteristics of PCL Scaffold … 461

3. O’Brien, F.J.: Biomaterials & scaffolds for tissue engineering. Mater. Today 14(3), 88–95
(2011)
4. Paula, A., et al.: Advances in additive manufacturing for bone tissue engineering scaffolds.
Mater. Sci. Eng., C 100, 631–644 (2019)
 Effect of Process Parameters on Material
Removal Speed in CBN Grinding Al6061 T6

Bui Thanh Danh1 , Tran Huu Danh2 , Nguyen Hong Linh3 , Nguyen Anh Tuan4 ,
Nguyen Van Tung5 , Hoang Xuan Tu5 , and Nguyen Van Trang5(B)
1 University of Transport and Communications, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vinh Long, Vietnam
3 Electric Power University, Hanoi, Vietnam
4 University of Economics-Technology for Industries, Hanoi, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nvtrang@tnut.edu.vn

Abstract. The results of a study on the effect of the process parameters on the
material removal speed (MRS) in CBN grinding Al6061 T6 are presented in this
paper. To do this, four input parameters were investigated in the study: wheel speed
(WS), feed rate (Fe ), depth of cut (aed ), and down feed (Df ). The impact of these
variables on the MRS of the grinding process was investigated. Furthermore, a
regression model was proposed to determine the relationship between the input
parameters and the MRS. It was also stated that the proposed model was suitable
for use.

Keywords: Grinding · CBN grinding · Material removal speed · Al6061 T6

1 Introduction
Grinding is a non-traditional machining method that uses a grinding wheel made up
of abrasives as a cutting tool. This technique is widely used for finishing and semi-
finishing. For the cutting process, CBN grinding employs a CBN wheel made of cubic
boron nitride. This material is indeed one of the hardest on the market, behind only just
to diamonds. The material’s main advantage is its high abrasion resistance and thermal
conductivity, which allows it to keep its sharp cutting edges. Therefore, CBN grinding
is widely used in mechanical processing.
There have been numerous studies on CBN grinding so far. Chen et al. [1] discussed
the advantages of grinding with a CBN wheel for precision manufacturing. Pavel and
Srivastave [2] described an empirical work of temperatures during conventional and
CBN grinding. Shi and Malkin [3] reported the results of a study on the wear of CBN
grinding wheels and how the wear process affects the wheel topography and grinding
behavior. The results of a study for improving CBN wheel tool life and optimizing the
grinding process while considering economic, environmental, and social sustainability
issues were introduced in [4]. The authors in [5] described the effects of compressed cold
air on spindle shaft materials (SCM4 and SCM21) when cylindrically ground with CBN

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 462–467, 2023.
https://doi.org/10.1007/978-3-031-22200-9_52
Effect of Process Parameters on Material Removal Speed in CBN Grinding Al6061 T6 463

wheels. The impact of vitrified bond structure on radial wear of CBN grinding wheels
was introduced in [6]. Vu et al. [7] presented a study on optimizing grinding parameters
for the minimum grinding time when grinding tablet punches with a CBN wheel on a
CNC milling machine. The authors in [8] analyzed of the various application and types
of cutting fluid used in plunge cylindrical grinding with conventional and CBN grinding
wheels. [9] reported on the grinding performance of AISI D6 steel using a vitrified and
resinoid bonded CBN wheel.
In this paper, the effect of the input parameters on MRS in CBN grinding Al6061
T6 were evaluated. In addition, a regression equation to express the relation between the
input grinding parameters and MRS was proposed.

2 Experimental Work
An experiment was carried out to solve the investigation of the effect of process param-
eters on MRS. This experiment was designed using the Minitab R19 software with a
2-level 1/2 fraction design. Therefore, 24–1 experimental runs were conducted. The input
factors and their levels are shown in Table 1. Figure 1 depicts the experimental setup.
Also, Table 2 shows the experimental plan as well as the output results (MRS)

Fig. 1. Experimental setup

3 Results and Analysis

The graph in Fig. 2 is used to assess the impact of the primary parameters on the MRS.
Figure 2 shows that all four input parameters, including Rpm , Fe , aed , and Df , positively
464 B. T. Danh et al.

Table 1. Input parameters and their levels

Factor Unit Low High

Wheel speed (WS) rpm 4000 5000
Feed Rate (Fe) mm/min 2000 3000
Depth of cut (aed ) mm 0.01 0.025
Down feed (Df ) mm/double journey 0.5 2

Table 2. Experimental plan and output results

Run order Rpm Fe aed Df MRS (g/h)

1 4000 2000 0.01 0.5 3.4779
2 4000 3000 0.025 0.5 4.3989
3 5000 2000 0.025 0.5 3.5631
4 5000 3000 0.01 0.5 4.0997
5 5000 2000 0.01 2 10.649
6 5000 3000 0.025 2 16.2297
7 4000 2000 0.025 2 11.6366
8 4000 3000 0.01 2 13.9366

affect MRS. That is, as Rpm , Fe , aed , and Df increase, so does MRS. Furthermore, the
effect of Df is the greatest, while the impact of Rpm is the smallest.

Fig. 2. Main effects plot for MRS

Figure 3 depicts the influence of the input parameters and their interactions on the
MRS using a Pareto chart. The length of the blue column indicates the effect of the
parameters. It was noted that the influence of D (Df ) on MRS is the greatest. Other
Effect of Process Parameters on Material Removal Speed in CBN Grinding Al6061 T6 465

parameters with decreasing impact include B (Fe ), AC (Rpm *aed ), C (aed ), AB (Rpm *Fe ),
AD (Rpm *Df ), and A (Rpm ). Figure 4 clearly shows the effect of the interactions.

Fig. 3. The Pareto chart of the effects

Fig. 4. Effect of two-way interactions on MRS

466 B. T. Danh et al.

Determining regression Equation to calculate MRS:

Minitab 19 software is used to calculate the regression equation for the MRS. Table 3
describes the estimated coefficients of the input parameters as well as their interactions.

Table 3. Estimated coefficients of input parameters and their interactions to determine MRS

Coded Coefficients
Term Effect Coef SE Coef T-value p-value VIF
Constant 8.499 * * *
Rpm 0.2729 0.1364 * * * 1.00
FE 2.335 1.167 * * * 1.00
Aed 0.9163 0.4581 * * * 1.00
Df 9.228 4.614 * * * 1.00
Rpm*Fe 0.7241 0.3620 * * * 1.00
Rpm*aed 1.6058 0.8029 * * * 1.00
Ppm*Df 0.3798 0.1899 * * * 1.00

The Minitab R19 software was used to determine the regression equation for MRS.
Table 3 describes the estimated coefficients of the input parameters as well as their
interactions. The regression equation is obtained as follows:

MRS = 28.68 − 0.007728 · Rpm − 0.004182 · Fe − 902.4 · aed + 3.873 · Df

+ 0.0000001 · Rpm · RS + 0.2141 · Rpm · aed + 0.000506 · Rpm · Df (1)

Evaluation of proposed model:

The Anderson-Darling method has been used to justify the fit of the experimental model
(Fig. 5). The results on this graph show that all of the experimental data (blue dots) are
within a 95 percent confidence interval defined by two limit lines. Moreover, P = 0.065
is more significant than =0.05, denoting that the proposed model is appropriate.

4 Conclusions

The findings of a study on the effect of process parameters on material removal speed in
CBN grinding Al6061 T6 were presented in this paper. The study looked at four input
parameters: wheel speed, feed rate, depth of cut, and down feed. The influence of these
factors on the MRS of the grinding process was evaluated. A regression model was also
proposed in order to determine the relationship between the input parameters and the
MRS. It was also stated that the proposed model could be used.
Effect of Process Parameters on Material Removal Speed in CBN Grinding Al6061 T6 467

Fig. 5. Probability plot of the fit of the proposed model

Acknowledgments. This work was supported by Thai Nguyen University of Technology.

References
1. Chen, X., Rowe, W., Cai, R.: Precision grinding using CBN wheels. Int. J. Mach. Tools Manuf.
42(5), 585–593 (2002)
2. Pavel, R., Srivastava, A.: An experimental investigation of temperatures during conventional
and CBN grinding. Int. J. Adv. Manufact. Technol. 33(3), 412–418 (2007)
3. Shi, Z., Malkin, S.: Wear of electroplated CBN grinding wheels (2006)
4. Caggiano, A., Teti, R.: CBN grinding performance improvement in aircraft engine components
manufacture. Procedia CIRP 9, 109–114 (2013)
5. Choi, H., Lee, S., Jeong, H.: A comparison of the cooling effects of compressed cold air
and coolant for cylindrical grinding with a CBN wheel. J. Mater. Process. Technol. 111(1–3),
265–268 (2001)
6. Herman, D., Krzos, J.: Influence of vitrified bond structure on radial wear of cBN grinding
wheels. J. Mater. Process. Technol. 209(14), 5377–5386 (2009)
7. Vu, N.-P., et al.: Optimization of grinding parameters for minimum grinding time when grinding
tablet punches by CBN wheel on CNC milling machine. Appl. Sci. 9(5), 957 (2019)
8. Monici, R.D., et al.: Analysis of the different forms of application and types of cutting fluid used
in plunge cylindrical grinding using conventional and superabrasive CBN grinding wheels. Int.
J. Mach. Tools Manuf. 46(2), 122–131 (2006)
9. Sato, B.K., et al.: Grinding performance of AISI D6 steel using CBN wheel vitrified and
resinoid bonded. Int. J. Adv. Manufact. Technol. 105(5–6), 2167–2182 (2019). https://doi.org/
10.1007/s00170-019-04407-5
 Effects of Voltage Unbalance on Matrix
Converter Induction Motor Drive

Nhu Y Do1 and Xuan Cuong Ngo2(B)

1 Hanoi University of Mining and Geology, Ha Noi, Viet Nam
2 School of Engineering and Technology, Hue University, Hue, Viet Nam

ngoxuancuong@hueuni.edu.vn

Abstract. Due to the significant share of the induction motor (IM) in the global
electricity consumption, the study of the IM and its drive systems in different
power supply modes is the aim of this paper. This paper presents the effects of
voltage unbalance considering the positive-sequence component on efficiency,
power factor, torque ripple, and current total harmonic distortion of the direct
matrix converter induction motor drive (DMCIMD). Simulation results have also
shown that torque ripple and average current total harmonic distortion increase
linearly with increasing voltage unbalance factor (VUF). In the case of voltage
unbalance with positive-sequence voltage less than phase voltage, the increase of
VUF reduces efficiency and increases the power factor of DMCIMD. Conversely,
in the case of voltage unbalance with positive-sequence voltage greater than phase
voltage, the increase in VUF increases efficiency and decreases the power factor
of DMCIMD.

Keywords: Voltage unbalance · Matrix converter · Induction motor

1 Introduction
Power quality (PQ) is an important aspect of electromechanical systems due to their
significant share in overall power consumption. Electric motors and their drive systems
are the largest end-users of electricity. It is estimated that electric motor-driven systems
account for more than half of the world’s electricity consumption [1]. Induction (IM)
motors account for approximately 70% of the industry’s energy consumption worldwide;
therefore, improving their efficiency will reduce power consumption globally [1].
The mechanical speed of the IM is directly controlled by the frequency of the supply
voltage, which limits its versatility when powered directly from the grid. Therefore,
many speed control systems have been researched and developed, including the direct
matrix converter induction motor drive (DMCIMD) [2–4].
The matrix converter is a single-stage converter that has a series of mxn bidirectional
power switches to directly connect an m-phase voltage source to an n-phase load. Direct
matrix converters offer many attractive features compared to conventional two-stage
conversion topologies, such as unlimited frequency; no DC link energy storage com-
ponents; bidirectional power transmission; and sinusoidal input and output current [5].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 468–476, 2023.
https://doi.org/10.1007/978-3-031-22200-9_53
 Effects of Voltage Unbalance on Matrix Converter Induction Motor Drive 469

The matrix converter of 3x3 switch has the highest practical interest because it connects
a three-phase voltage source to a three-phase load, typically an IM.
Voltage imbalance represents a condition in which three-phase voltages (or currents)
differ in amplitude or deviate from their normal 120˚ phase relationship, or both [6]. This
is a common situation in three-phase systems, and the operation of the IM and the three-
phase drive are affected by the positive and negative sequence components. The negative
sequence current produces a magnetic flux that rotates in the direction opposite to that
of the rotor. This flux causes heat gain, additional loss, and additional torque acting in
the opposite direction to the torque produced by the positive-sequence flux [6].
There are many studies on reducing the loss and increasing the effectiveness of IM.
For example, an increase in IM performance can be achieved at all machine design
parameters, and the quality of the power supply should also be considered. The impact
of voltage unbalance on the performance of IMs and their drive systems have been
extensively studied [1, 7, 8].
The voltage unbalance studies of the IM show that the positive-sequence voltage
level has more influence on the core loss of the IM than the voltage unbalance level [9].
In addition, the effects of positive and negative-sequence and zero sequence components
on the torque and performance of IM were evaluated and analyzed [10, 11].
Experimental evaluations of the effects of various abnormal voltage conditions like
unbalanced power supply, balanced non-sinusoidal power supply, input voltage sags,
and short time blackout of power supply on matrix converter fed induction motor drives
are presented in [12, 13].
The above analysis has shown that additional studies on the effect of voltage unbal-
ance on electric drive systems are needed. This paper presents the effects of voltage
unbalance on the efficiency, power factor, torque ripple, and current total harmonic dis-
tortion of direct matrix converter induction motor drive. The study also suggests various
phase difference cases for the research system.
Section 2 of the paper builds a research model. Section 3 presents the analysis of
simulation results of a DMCIMD operating under various unbalanced conditions. Finally,
conclusions and future research are presented in Sect. 4.

2 Research Model
2.1 Modeling of the Direct Matrix Converter Induction Motor Drive
The DMCIMD model under voltage unbalance is built using the MATLAB/Simulink
2015a software. We chose a single squirrel cage IM with a rated power of Pn = 250 kW
[8]. The stator winding is connected in a star to the internal neutral point. The parameters
of IM are presented in Table 1. The physical model is linear with respect to the inductance,
which is assumed to be constant, so saturation of the magnetic circuit is not considered
here.
The three-phase matrix converter is built from 9 IGBT switches in series using a
switching algorithm based on indirect spatial vector modulation described in the Matlab-
simulink library [14, 15].
The research model includes a three-phase 50Hz source and a line-to-line voltage of
6kV through an LC filter (L = H; C = 25 μF) to power the matrix inverter and the IM.
470 N. Y. Do and X. C. Ngo

Table 1. Motor parameter.

Parameter Value
Motor name A4-355L-4
Motor type Induction motor
Rotor type Squirrel cage
Rated power (Pn ) 250 kW
Rated line-to-line voltage (V n ) 6 kV
Frequency (f n ) 50 Hz
Rated speed (N n ) 1480 rpm
Pole pairs (p) 2
Stator full Load current (Is) 29 A
Rated efficiency (ηn) 94.2%
Starting/rated current (Ist/In) 6.0
Starting /rated torque (Tst/Tn) 1.2
Maximum pull out/rated torque (Tbr/Tn) 2.2
Power factor (PF) 0.88
Moment of inertia of the rotor (J) 6.25 kg.m2

The diagram is depicted in Fig. 1. Simulation was conducted with the full load (1613
Nm) at 0.4 s. The voltage unbalance source is constructed at the start of the simulation.

Fig. 1: Research model of matrix converter induction motor drive.

Calculate the power factor (PF), mechanical power and mechanical losses are
calculated according to the following formula:
Pin
PF =  ; P_M = Te .ωm ; P_loss_m = F.ωm
2
(1)
2
Pin + Qin
2
 Effects of Voltage Unbalance on Matrix Converter Induction Motor Drive 471

where Pin − the active input power, in W; Qin − the reactive input power, in Var; T e −
the electromagnetic torque, in Nm; ωm − the angular velocity of the rotor, in rad/s; F −
friction factor, in N.m.s.
We determine the efficiency of DMCIMD using the direct method as follows:
P_M − P_loss _m
η= .100 (2)
Pin
Torque ripple of electric motor in one cycle is calculated [16]:
Tmax − Tmin
RipT = .100 (3)
Te
where, T max − maximum torque value; T min − minimum torque value; T e − Average
torque value of electric motor, M.m; RipT − Torque ripple, %.

2.2 Voltage Unbalance

In this paper, the IEC or the “true” definition is used to define the degree of voltage unbal-
ance, as the voltage unbalance factor (VUF) definition is adequate in measuring asym-
metry. A voltage unbalance factor is defined as the ratio of the negative-sequence voltage
component (V neg ) to the positive-sequence voltage component (V pos ). This definition is
consistent with IEC standard 61000-4-30 [17]:
Vneg
KV = .100 (4)
Vpos

where Vneg is the negative-sequence voltage component; V pos is the positive-sequence

voltage component; K V is voltage unbalance factor, %.
We analyzed the effects of the unbalance regime on the DMCIMD for VUF from
0% to 5%. Because the magnitude of the positive-sequence component affects the motor
behavior differently for the same VUF. Therefore, for the same value of VUF, several
cases can occur, we considered four different conditions for each value of VUF. The
phase voltage unbalance cases considered for system simulation are presented in Table
2. In Table 2, the modules and displacement angles of the three phase-to-neutral voltages
are presented.

3 Research Results and Discussion

In this section, the simulation results of a DMCIMD running at 100% rated load with a
voltage bias source are presented. Accordingly, the total harmonic distortion of current,
torque ripple, power factor (PF), and system efficiency for the four unbalanced cases are
further analyzed.
472 N. Y. Do and X. C. Ngo

Table 2. Cases of phase voltage unbalance considered for system simulation.

Case Unbalance conditions KV (%) Va Vb Vc

1 Different amplitudes, equal 0 4900  0 4900  - 120° 4900  120°
phase shifts, and 1 4754.4  0° 4900  - 120° 4900  120°
positive-sequence voltages are
less than phase voltages 2 4611.8  0° 4900  - 120° 4900  120°
3 4471.8  0° 4900  - 120° 4900  120°
4 4334.6  0° 4900  - 120° 4900  120°
5 4200  0° 4900  - 120° 4900 120°
2 Different amplitudes, equal 0 4900  0 4900  - 120° 4900  120°
phase shifts, and 1 5048.5  0° 4900  - 120° 4900  120°
positive-sequence voltages are
greater than phase voltages 2 5200  0° 4900  - 120° 4900  120°
3 5354.6  0° 4900  - 120° 4900  120°
4 5512.5  0° 4900  - 120° 4900  120°
5 5673.7  0° 4900  - 120° 4900  120°
3 Different amplitudes, different 0 4900  0 4900  - 120° 4900  120°
phase shift, and the 1 4998 < 1.284 4900  - 120° 4900  120°
positive-sequence voltage is
larger than the phase voltage 2 4998 < 3.233 4900  - 120° 4900  120°
3 4998 < 5.01 4900  - 120° 4900  120°
4 4998 < 6.751 4900  - 120° 4900  120°
5 4998 < 8.478 4900  - 120° 4900  120°
4 Equal amplitudes, different 0 4900  0 4900  - 120° 4900  120°
phase shift, and 1 4900  1.719° 4900  - 120° 4900  120°
positive-sequence voltage
equals phase voltage 2 4900  3.437° 4900  - 120° 4900  120°
3 4900  5.154° 4900  - 120° 4900  120°
4 4900  6.869° 4900  - 120° 4900  120°
5 4900  8.581° 4900  - 120° 4900  120˚

3.1 Harmonic Distortion of the Current

Due to voltage unbalance, the current values of the three phases are different, so the
average value of total harmonic distortion (average THDI ) of the three phases is selected
for analysis.
Figure 2 shows the dependence of the average THDI for three phases of current
with VUF for unbalanced cases. Due to the unbalanced voltage system, the third, fifth,
and seventh harmonic currents are detected in the working current. It can be observed
from Fig. 2 that average THDI increases almost linearly with increasing VUF. Another
observation is that for the same degree of unbalance, the values of average THDI increase
with the magnitude of the positive-sequence voltage. For example, when KV = 5%:
average THDI = 19.05% for case 1 with Vpos < Vphn ; average THDI = 21.44% for
 Effects of Voltage Unbalance on Matrix Converter Induction Motor Drive 473

22 80
21 Case 1 Case 1
20 70
Case 2 Case 2
Average THDi (%)

Torque ripple (%)

19 Case 3 60 Case 3
18 Case 4 Case 4
17 50
16
15 40
14 30
13
12 20
0 1 2 3 4 5 0 1 2 3 4 5
Kv (%) a) Kv (%) b)

Fig. 2. The average THDI (a) and torque ripple (b) with VUF.

case 2 with Vpos > Vphn ; average THDI = 20.67% for case 3 with Vpos > Vphn ; average
THDI = 20.48% for case 4 with Vpos = Vphn .

3.2 Torque Ripple

When the system operates under balanced voltage, torque ripple is generated at a fre-
quency 6 times higher than the nominal frequency. Voltage unbalance produces elec-
tromagnetic torque oscillations at frequencies higher than 2 times the rated frequency.
This torque component at the 100 Hz frequency is due to the negative string currents,
which is 3.5 times more valuable than the torque component at 300 Hz (Fig. 3). Figure 2b
shows the torque ripple according to VUF. It can be observed that torque ripple increases
linearly with increasing VUF. It should be noted that for the same value of VUF, the
amplitude of the torque ripple is different for two different unbalanced cases. For exam-
ple, for Kv = 5%, torque ripple is 56.11%, when Vpos < Vphn (case 1) and 63.89%,
for Vpos > Vphn (case 2), which means a 13.87% difference between the two values of
torque ripple.

Fig. 3. Spectral analysis of the electromagnetic torque in the case VUF = 5%.
474 N. Y. Do and X. C. Ngo

3.3 Power Factor

Furthermore, we analyze the influence of the voltage unbalance on the PF of the matrix
converter induction motor drive. The PF variation with VUF is represented in Fig. 4a.
It can be seen that PF can increase or decrease depending on the specific case. It should
be noted that in case 2, when the positive-sequence voltage increases much compared
to the phase voltage, the PF of the system decreases. In other cases, the PF can be seen
to increase slightly.

1 87.4
87.2
0.995
Case 1 87

Efficiency (%)
0.99
Power factor

Case 2 86.8
0.985 Case 3 86.6
Case 4 86.4
0.98 Case 1
86.2 Case 2
0.975 86 Case 3
85.8 Case 4
0.97
0 1 2 3 4 5
0 1 2 3 4 5
Kv (%) a) Kv (%) b)

Fig. 4. Power factor (a) and efficiency (b) due to voltage unbalance factor.

3.4 The Efficiency of the Matrix Converter Induction Motor Drive

The efficiency of the DMCIMD running at 100% of the rated load due to voltage unbal-
ance is depicted in Fig. 4b. We see that, in the case that the positive-sequence voltage
is less than the phase voltage, this efficiency decreases with the increase of VUF. And
in the case where the positive-sequence voltage is greater than the phase voltage (case
2), the efficiency of DMCIMD increases to the VUF. This is explained by the fact that
the component causing the electromagnetic torque of the motor is the positive-sequence
component.

4 Conclusion
For a detailed analysis of the effect of voltage unbalance on the electrical and mechanical
parts of the DMCIMD, VUF value alone is not enough; it is also necessary to know the
magnitude of the positive-sequence component. Research results have shown that for
different unbalance cases, with the same VUF value and different positive-sequence
components, the torque ripple can vary in amplitude by up to 13.87% and the efficiency
values differ by 1.2%. In the case of lower values of the positive-sequence component,
a lower value of efficiency but a higher value of PF is obtained.
 Effects of Voltage Unbalance on Matrix Converter Induction Motor Drive 475

In the scenarios considered, when the matrix converter induction motor drive is
fed from balanced voltages, the electromagnetic torque vibrates at 300 Hz, and when
unbalanced voltages appear, the electromagnetic torque vibrates more at 100 Hz and has
a magnitude of about 3.5 times higher than the electro-magnetic torque component at
300 Hz.
The simulation results have also shown that torque ripple and average THDI increase
linearly with increas-ing VUF. Also, the efficiency of the DMCIMD decreases with the
increase of VUF in cases where the positive-sequence voltage is less than the phase
voltage, and vice versa. The power factor of the DMCIMD tends to be the opposite of
its efficiency.
The proposed method for estimating the effect of voltage unbalance on DMCIMD
behavior is the basis for a generally applicable method to evaluate the overall effect of
the considered PQ case for any application. The research results are the first step for
studies on power quality effects on drive systems with IM provided by power converters.

References
1. Beleiu, H.G., et al.: Effects of voltage unbalance and harmonics on drive systems with
induction motor. J. Taibah Univ. Sci. 16(1), 381–391 (2022)
2. Altun, H., Sünter, S.: Matrix converter induction motor drive: modeling, simulation and
control. Electr. Eng. 86(1), 25–33 (2003)
3. Lettl, J.: Matrix converter induction motor drive. In: 2006 12th International Power Electronics
and Motion Control Conference, IEEE (2006)
4. Vijayagopal, M., et al.: Control of a direct matrix converter induction motor drive with mod-
ulated model predictive control. In: 2015 IEEE Energy Conversion Congress and Exposition
(ECCE), IEEE (2015)
5. Wheeler, P.W., et al.: Matrix converters: a technology review. IEEE Trans. Industr. Electron.
49(2), 276–288 (2002)
6. Baggini, A.: Handbook of Power Quality. John Wiley & Sons (2008)
7. Do Nhu, Y., Cuong N.X.: Impact of Voltage Unbalance and Harmonics on Induction Motor
in Operation Mode. In International Conference on Engineering Research and Applications.
Springer (2021)
8. Ngo, X.C., Do, N.Y., Tran, Q.H.: The Influence of Voltage Quality on Asynchronous Motor
Performance of EKG Excavator in Open Pit Mines–Vinacomin. Inżynieria Mineralna (2020)
9. Donolo, P.D., et al.: Effects of negative sequence voltage on the core losses of induction
motors. In: 2016 IEEE ANDESCON, IEEE (2016)
10. e Silva, M.D.C., et al.: Effects of sequence voltage components on torque and efficiency of a
three-phase induction motor. Electr. Power Syst. Res. 140, 942−949 (2016)
11. Lee, C.-Y.: Effects of unbalanced voltage on the operation performance of a three-phase
induction motor. IEEE Trans. Energy Convers. 14(2), 202–208 (1999)
12. Kumar, V., Bansal, R.C., Joshi, R.R.: Experimental realization of matrix converter based
induction motor drive under various abnormal voltage conditions. Int. J. Control. Autom.
Syst. 6(5), 670–676 (2008)
13. Sun, K., et al.: Compensation control of matrix converter fed induction motor drive under
abnormal input voltage conditions. In: Conference Record of the 2004 IEEE Industry
Applications Conference, 39th IAS Annual Meeting, IEEE (2004)
14. Huber, L., Borojevic, D.: Space vector modulated three-phase to three-phase matrix converter
with input power factor correction. IEEE Trans. Ind. Appl. 31(6), 1234–1246 (1995)
476 N. Y. Do and X. C. Ngo

15. MathWorks.: Three-Phase Matrix Converter (2022). [cited 2022 1/7/2022]. https://www.mat
hworks.com/help/physmod/sps/ug/three-phase-matrix-converter.html
16. Zhang, W., et al.: Minimising torque ripple of SRM by applying DB-DTFC. IET Electr. Power
Appl. 13(11), 1883–1890 (2019)
17. Santoso, S., et al.: Electrical Power Systems Quality. McGraw-Hill Education (2012)
 Electrocardiogram (ECG) Circuit Design
and Using the Random Forest to ECG
Arrhythmia Classification

Van Nam Pham1(B) and Hoai Linh Tran2

1 Hanoi University of Industry, Hanoi, Vietnam
nampv@haui.edu.vn
2 Hanoi University of Science and Technology, Hanoi, Vietnam

Abstract. The paper presents a proposed hardware device for ECG signals mea-
surement with an integrated remote system for signal automatic analysis to help
doctors monitor health, diagnose cardiovascular diseases. This measuring device
can transmit the ECG signals online to the server, The server is equipped with a
software for analyzing ECG signals and classifying them to detect the arrhyth-
mias using a RF (Random Forest) network. The Hermite basis functions were
used to generate the feature vectors together with 2 time-based features: the last
R-R period and the average of the last 10 R-R periods. The proposed solution was
tested with ECG signals taken from databases MIT-BIH (Massachusetts Institute
of Technology, Boston’s Beth Israel Hospital). Seven types of ECG beats were
classified with an error of 2.28%. The proposed Random Forest algorithm works
very fast, which makes it suitable for the requirement of quick classification of
some cardiovascular diseases.

Keywords: Random forest · Decision tree · Hermite basis functions ·

Electrocardiogram (ECG) signals

1 Introduction
In society nowadays, the population aging is increasing. At first in developed countries
such as US, Japan, European countries, etc. then to developing countries like Vietnam,
Thailand, the number of cases of cardiovascular patients is growing up, greatly affected
to the health and quality of life of the elderly people. But if these diseases are detected
early, they are highly curable. Meanwhile, the more life develops the greater demands
of adults in the family. They are busy at work, want to take care of the health of elder
but they don’t have time to go to hospitals for regularly checking up.
This paper proposed the model of automatic diagnosis of ECG signals as shown in
Fig. 1. A measuring device will collect and transmit the ECG signals of the patient to
the server via WIFI/3G. The server equipped with a program of automatic analysis and
classification of ECG signals with high accuracy and reliability will produce the analysis
results. The results can be later transmitted to the mobile phone of users. In addition, the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 477–494, 2023.
https://doi.org/10.1007/978-3-031-22200-9_54
478 V. N. Pham and H. L. Tran

software also supports doctors to quickly filter the ECG data sets of users, monitor their
health status as well as provide patient treatment regimens remotely.
The advantages of this model for users: being convenient, comfortable for users, low
cost, quick results and avoid the problem of infection in crowded places such as medical
centers especially in COVID-19 disease era. In term of medical centers, hospitals can
reduce the pressure on the number of patients, reduce the cost for society.

Fig. 1. Transmitting ECG signals system via IoT, automatic recognition of ECG signals and
application for users.

Therefore, the research and development of the above models is necessary, especially
when we are developing in Industry 4.0, the main research content of the article:

• Build a handheld ECG device with function of transmitting ECG signals to the server
so that doctors can monitor the health of patients remotely, diagnose diseases early;
• Develop the identifying ECG signal software to assist doctors to quickly filter for
some cardiovascular diseases bases on the Random Forest (RF).

The simulation results were verified on the standard sample data sets of the MIT-
BIH database, which are the databases used by many domestic and international research
groups for reference.

2 Design of ECG Signals Device

2.1 Functions of Device

With the purpose of designing a personal device, which should be convenient to carry as
used in this paper, the hardware is designed with the following functional requirements:

• Can measure and acquire one selected channel of ECG signal (RA, LA, RL),
• The device should be portable, battery-powered;
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 479

• The measured ECG signals can be transmitted online to the server via GPRS/3G
network, with the purpose of supporting doctors to diagnose the health status of
patients remotely, when accessing the database of patients’ ECG signals from server;
• Device has LCD screen and this device stores data on a 8GB SD card;
• Device can be connected to PC through USB;
• The block design of basis functions of device as shown in Fig. 2.

Fig. 2. Block design of device.

2.2 Design Hardware

(a) Design of the circuit for measuring and normalizing of ECG signals

The ECG (Electrocardiogram) signals contains essential information to identify cardio-

vascular diseases. As shown in Fig. 3, the ECG signals is a curve that records the change
in voltage which is generated by the heart during operation, the main shape of ECG
signals is a combination of P waves, QRS complex and T waves. In which, the QRS
complex contains the most important information.
Collecting and standardizing signals block contains electrodes and circuits for low-
pass and high-pass filter blocks. The collecting circuit has the task of amplifying the ECG
signals at mV range to a suitable range for processing, with a following requirement:

• Eliminate interference that is out of the range of ECG frequency (1 ÷ 100 Hz);
• Signals Amplifier should be around 1000 times.
480 V. N. Pham and H. L. Tran

Fig. 3. Shape and main components of ECG signals.

The ECG signals that measured from the human body is in a form of voltage dif-
ference. In this paper, we only consider to measure D1, due to this measurement of the
differential voltage signals would be brought back to two electrodes which attached on
right arm (RA), left arm (LA) and right leg (RL) as shown in Fig. 4. In term of ECG
signals, the input impedance of circuit must be several times greater than the equivalent
output impedance of the signal. For main leads, the input impedance of circuit must be
at least 10 M.´ The ECG signals is a weak signal, easily affected by interference. The
requirement for amplifier circuit is to have a common-mode rejection ratio (CMRR)
greater than 60dB. Nowadays, in the market of component manufacturers, there are so
many types of IC amplifiers with better parameters. Based on the results of survey, we
decided to select IC AD8232.
The IC AD8232 already integrates the analog blocks for filtering and normalizing
the output signal, so there is little need for additional external analog elements. We just
needed to add a passive filter RC (as seen on Fig. 5), to reduce interference. The output
signals can be put into the converter block ADC to digitize.

Fig. 4. The positions of ECG electrodes on the human body

 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 481

Fig. 5. Signals after standardizing and a passive filter RC (a), Module AD8232.

The AD8232 module is designed specifically to collect ECG signals, directly con-
nected to the two electrodes attached at the Right Arm (RA), Left Arm (LA) position
into a amplifier with amplifier coefficients of G1 = 100 (times), and Right Leg (RL)
electrodes provide right foot feedback circuit. The AD8232 module has built-in low
pass filters and high pass filters. The output signal of the module has been filtered out,
and through the 2nd amplifier with amplifier coefficient G2 = 11, so in combination
with the built-in IC of G1 = 100, the total amplification of the ECG signal is:

G = G1 × G2 = 100 × 11 = 1100 (1)

(b) Others functional blocks

• Display block: The designed device uses a LCD TFT 3.2 Inch Screen, with the
resolution of 480x320 pixels;
• Data storage block: Device has the ability to store data which is measured as
well as the time measured into SD card. In order to store the measured data, in
this paper we decided to use an SD card with 8GB storage;
• WIFI communication block: Using module BLE SoC ESP32 (Fig. 6).

Fig. 6. LCD screen (a), SD card (b), Module WIFI ESP32 (c).
482 V. N. Pham and H. L. Tran

(c) Center control block

The central microprocessor unit is responsible for controlling all the functions of the
device: collecting ECG signals from the same module AD8232, sampling those signals,
preprocessing, storing and displaying or transmitting to the server for further processing
purposes. The microprocessor is selected for use in this paper is the STM32L from
STMicroelectronics. The STM32L has advantage of energy saving, so it is suitable for
handheld devices that use battery power. Paper decided to select the STM32L151RB
chip with some main specifications as follows:

• Operating voltage range: 1.65 to 3.6 V;

• Consumption voltage: 0.9 µA in sleep mode, 214 µA/MHz in running mode;
• Frequency: 32 MHz;
• Memory: 128 KB flash, 16 KB RAM, 4 KB EEPROM;
• Up to 24 ADC 12bit channels with speeds up to 1 Msps;
• There are full communication ports: USB, USART, SPI, I2C.

3 Develop the Software for Identification of ECG Signals

The general model of identifying ECG signal system is shown in Fig. 7.

ECG signal

Preprocessing stage

The detection of the R peak of

the QRS complexes;

Feature
vector

Neural classifiers

Heartbeat classes
Fig. 7. General model of identifying ECG signal system
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 483

3.1 Pre-processing

Due to the ECG signals through acquisition module AD8232 has been filtered and
standardized after being digitized, the pre-processing stage only needs to add a 50 Hz
digital filter, with bandwidth 4 Hz (width of blocking range), and 3 dB reduction, having
a following transmission function:

0, 9825 − 1, 2631z −1 + 0, 9825z −2

H (z) = (2)
1 − 1, 2631z −1 + 0, 9651z −2

3.2 Feature Extraction

Feature extraction is the process of finding the typical values of the original input sig-
nal, the amount of information of the feature signal is much more streamlined than the
original signal, so the calculation volume of the identification block will decrease. Fea-
ture choosing is usually based on its object separation. Therefore, analysis and feature
extraction play an important role, it allows to support to give identifiable results more
easily and accurately. The feature vector of each beat (QRS complex) consists of 18
components:

• 16 decomposition factors C i (i = 0, 1,…,15) of ECG signals based on Hermite

functions;
• 2 time-based features of ECG signals, is RRlast - the distance between peak R of beat
under-processing and the previous peak R and RRmean - the average value of 10 R-R
interval.

The Hermite functions are used to extract the features from ECG signals because of
the similarity between their shapes and the shapes of the ECG signals.

(a) Detect the location of the R peak and isolate the QRS complex from the ECG signals

The QRS complex in the ECG signal contains quite a lot of useful information for
identification. The paper decided to select a length of 250 ms around the peak R that is
wide enough to accommodate the QRS segments, assuming that the ECG signals were
sampled with 360 Hz frequency (250 ms would correspond to about 90 values around
the peak R). Using QRS complex decomposition algorithm by Hamilton and Tompkins
[1, 2], we used the following steps to detect the R peaks in the ECG signals (Fig. 8).
484 V. N. Pham and H. L. Tran

ECG - x(n)

Low pass filter

High pass filter

y(n)

|d[ ]/dt|

16
1
16 ∑
1
z(n)

The detection of the R peak

of the QRS complexes;

Fig. 8. Operation diagram of detecting peak R algorithm

The results tested for a sample record from the MIT-BIH database is shown on Fig. 9.

Fig. 9. Steps of detecting peak R: (a) original ECG signals, (b) the result after filtering, (c) take a
derivative, (d) take a absolute value, (e) the result after taking average, (f) the results of detected
R peaks
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 485

(b) Analyse QRS complex based on Hermite basis function

The paper decided to use Hermite basis function to develop QRS complex to take a
expansion factor of feature vector of ECG signals. On the other hand, we also use 2
time-base features: the distance between 2 consecutive peak R, the average values of 10
R-R interval.
Hermite functions have a following formula for n ≥ 0:
√ − 1 x2
ψn (x) = π · 2n · n! 2 e− 2 Hn (x) (3)

with Hn (x)is polynomial Hermite which is defined by:

Hn+1 (x) = 2x · Hn (x) − 2n · Hn−1 (x) (4)

for n ≥ 1, with H0 (x) ≡ 1; H1 (x) = 2x.

Fig. 10. Diagram of Hermite function n degree: a) n = 0; b) n = 1; c) n = 3; d) n = 10

As seen in Fig. 10, the higher the order of the Hermite function, the higher the rate
of variations of the function. Besides, the shape of the functions is quite similar to the
shape of the basic components in ECG signals. This is the basis of using the Hermite
function to analyze ECG signals.
To express the ECG signal s(t) according to the first N Hermite functions as in
formula (10), we use the Singular Value Decomposition algorithm to find the optimal
solution of the first order system of equations which has more equations than variables,
details can be found in [7] (Fig. 11).


N −1
s(t) ≈ ci · ψi (t) (5)
i=0
486 V. N. Pham and H. L. Tran

Fig. 11. ECG signals by N Hermite function: a) N = 5; b) N = 10; c) N = 12; d) N = 16

The choice of the number of Hermite functions also needs to be investigated in detail,
because if we use too little will cause the lack of information for identifying model so
the results are inaccurate. If we use too much, it will make the model become complex,
requiring high volume of calculations. Let’s investigate the error between original signal
and approximated signal according to the first N Hermite function:
 
 
N −1 
 
E = s(t) − ci ψi (t) ⇔ E = s(t) − b (6)
 
i=0

Fig. 12. Investigation diagram of error according to the number of Hermite basis functions

The error survey chart on Fig. 12 shows that when increasing the number of Hermite
functions to analyse ECG signals, the approximating error decreases, that is, the closest to
the original signal. Also seen on Fig. 12, the error for the number of Hermite functions (N
= 16) is quite close to min value (meaning if the number of Hermite functions continues
to increase, the error does not decrease much). So in this paper we proposed to use the
case of N = 16.
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 487

Continuing to investigate when approximating six different types of heart rate accord-
ing to the 16 Hermite functions, shown in Fig. 13 for the use of the Hermite function is
still pretty good approximation.

Fig. 13. Image of developing other types of heartbeat according to the first 16 Hermite function

(c) Creating feature vector of ECG signals

According to the process of feature extraction of ECG signals, feature vector.

x = [c0 , …, c15, RRlast , RRmean ] of each beat (QRS complex) consists of 18
components:

• 16 expansion factors Ci (i = 0, 1,…, 15) of ECG signals based on Hermite functions;

• 2 time-base features of ECG signal are RRlast and RRmean

3.3 Algorithm for Classifying ECG Signals

The paper used RF (Random Forest) identification model which is developed in [3]
and also used in [4, 5], the basic structure of RF model is sets of large number of M
non-pruning decision trees, the structure of each decision tree is randomly trained from
the bootstrap sample set (bootstrap − the method of taking samples randomly from a
known sample data set), steps of creating the structure of RF:

• Input is a data set used for training.

488 V. N. Pham and H. L. Tran

• Each RF model is a set of M decision trees (M is random).

• The structure of each decision trees is trained by one bootstrap sample set, which is
taken randomly from the general sample set.
• The single decision trees are not pruned.
• Summarizing result of identification of M decision trees using method of voting by
majority to give the final results of Random Forest model.

So for a new data sample need to be tested, at first, it is passed through M decision
trees for classification, each tree will have a unique result (maybe similar or different)
and these results will be put into the voting process to give the final result.
The random forest method is also a form of combination model, in which the basic
identifying models use the same decision tree algorithm (but being trained on different
samples), summarize the results by using the method of voting by majority, the number
of identifying models is quite large (usually have around several hundreds of decision
trees).
The process of building an RF identification model is as follow:

• Build a RF (which is shown in Fig. 14) to create set of M decision trees which are not
pruned, each tree is built by bootstrap sample set

Fig. 14. Process of building decision trees

• Summarizing identifying results from M decision trees by using method of voting by

majority to give the final result, as shown in Fig. 15.
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 489

Fig. 15. Process of testing of RF model

3.4 Data Base of ECG Signals

The first ECG signals’ data base which is used in this paper, was the famous data base
of MIT-BIH of arrhythmia [3, 8], select the record of 19 patients (ID in database are
respectively 100, 105, 106, 109,111, 114, 116, 118, 119, 124, 200, 202, 207, 208, 209,
212, 214, 221 and 222), classify 7 types of arrhythmia such as: Left Bundle Branch
Block Beat (L), Right Bundle Branch Block Beat (R), Premature Atrial Contractions
(A), Premature Ventricular Contraction (V), Ventricular Flutter Wave (I), Ventricular
Escape Beat (E), Normal beat (N). The number of samples which is taken from the
record of 19 patients is shown in Tables 1 and 2 below:

Table 1. Table of classification learn samples and test samples of 7 types of arrhythmia from
CSDL MIT-BIH

Beat type Total of samples Learn samples Test samples

N 2000 1065 935
L 1200 639 561
R 1000 515 485
A 902 504 398
V 964 549 451
I 472 271 201
E 105 68 37
Total 6643 3611 3068

From Fig. 16, it can be seen that the major obstacle in identifying ECG signals is
the huge variation in amplitude and shape of heartbeats of the same disease. Moreover,
some types of beat of different diseases can be similar in term of form. Therefore, it is
clear that the problem of classifying ECG signals is a hard problem.
490 V. N. Pham and H. L. Tran

Premature Atrial Contractions -A Ventricular Escape Beat -E

Left Bundle Branch Block Beat -L Right Bundle Branch Block Beat -R

Ventricular Flutter Wave -I Premature Ventricular Contraction -V

Fig. 16. The shape of QRS complex of 6 types of beat: A, E, L, R, I and V

Table 2. Table of classification learning samples and test sample of 2 beat types

Beat type Total of samples Learn samples Test samples

Normal 2000 1065 935
Abnormal 4643 2546 2133
Total 6643 3611 3068

4 Results
In this paper, we had designed and tested the ECG signals device with the following
functions:

• Can measure and acquire the selected channel of ECG signal (RA, LA, RL);
• Sample frequency: 360 Hz/720 Hz;
• Measure and transmit ECG signals to the server for automatically classification and
analysis of ECG signals via GPRS/3G, to help doctors to diagnose the health situation
as well as to give a regimen in order to treat patients remotely;
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 491

• Can be connected with PC via USB;

• Can display heart rate per minute, error messages, electrodes contact status;
• It can give warnings when heart rate is beating faster or slower than upper limit or
lower limit, disconnect with electrodes;
• It has a 3.2-inch LCD screen with the resolution of 480x320 pixels, to help doctors to
observe the ECG signal right on the device;
• It has an 8G SD memory card slot to store ECG signals;
• This device used Li-Ion Battery with voltage (root mean square) of 3.7V and capacity
of 2000mAh. The battery charger will use 5V power and can operate continuously
for 480 min;
• Dimensions of device: 170(L) x 135(W) x 50(H) mm;
• Weight: ~500 g (Fig. 17).

Fig. 17. Image of testing process on patients

Parameters are displayed in LCD screen: (Fig. 18).

The data of ECG signals in SD card, of in the server can import file in order to print
in the paper (Fig. 19).
Website has ability to access the data in server. Information which is stored in basis
data includes: ID, Name, ECG signals file and result of diagnose (Fig. 20).
Identifying ECG signals software:
This paper develops a model of identifying ECG signals based on the set of 3611 samples
(Table 1), identifying 7 types of beat. Applied by the method of Breiman (2001) [4],
built an RF identification model with 100 Decision trees (DT).
492 V. N. Pham and H. L. Tran

Fig. 18. Image of ECG signals is display in LCD screen with some parameters: Heart rate per
minutes is 85 bpm, ID’s patient is 665

Fig. 19. ECG signals of patient after importing

Fig. 20. Information of ECG signals of patient Hac Thi. Bui, 62 years old

Test results of identification models on 3068 test samples (Table 1). The results are
presented in the matrix as shown in Table 3 in which the column is the standard results,
the row is the identification results, the main diagonal of the matrix is the number of
correct identification samples, the value surroundings the main diagonal are the wrong
identification samples, the specific results are as follows:
 Electrocardiogram (ECG) Circuit Design and Using the Random Forest … 493

Table 3. Matrix of identification results of 7 types of sample by RF

Sample results N L R A V I E
N 914 6 3 11 0 0 0
L 1 547 0 0 4 0 1
R 1 0 478 4 0 0 0
A 18 4 1 382 3 0 0
V 1 3 1 1 443 3 0
I 0 1 2 0 1 198 0
E 0 0 0 0 0 0 36
Total error 21 14 7 16 8 3 1

The results of identifying ECG signal software by using RF network as follows:

• Number of error samples: 70 (samples),

Number of error samples 70

%Error = × 100% = × 100% = 2.28%
Number of test samples 3068
• FN (False Negative): Number of cases which is diagnosed false negative, it means
patients is diagnosed is diseased but being misdiagnosed is normal: 20 samples
• TN (True Negative): Number of cases which is diagnosed true negative: 914 samples
• FP (False Positive): Number of cases which is diagnosed false positive, it means
patients is diagnosed is normal but being misdiagnosed is diseased: 21 samples
• TP (True Positive): Number of cases which is diagnosed true positive: 2113 samples
• Sensitivity (True Positive Rate): Ratio of true positive diagnosis
TP 2113
Sensitivity = · 100% = · 100% = 99, 06%
TP + FN 2113 + 20
• Specificity (True Negative Rate): Ratio of true negative diagnosis
TN 914
Specificity = · 100% = · 100% = 97, 75%
TN + FP 914 + 21

Identification model get a good quality and reliability when:

• Number of samples which is identified wrong, number of FP diagnosis, number of

FN diagnosis is small;
• Sensitivity and Specificity is large.

In the criteria to evaluate the model quality listed above, FN (20 samples) is important
because if the FN error is large, it will be very dangerous for people due to the identifying
model will not detect the disease. So the less number of cases which is detected FN, the
494 V. N. Pham and H. L. Tran

greater quality model has. Therefore, the model of identifying ECG signals using RF
network is quite good.
Experiment 2: In the database of MIT-BIH (Table 2), identifying 2 types of beat
(normal and abnormal), has a following result:

• Number of error identification cases: 37 samples

• % Error: 1.21%
• FN (False Negative): 13 samples
• TN (True Negative): 911 samples
• FP (False Positive): 24 samples
• TP (True Positive): 2120 samples
• Sensitivity (True Positive Rate): 99.39%
• Specificity (True Negative Rate): 99.43%

5 Conclusions
This paper has designed ECG signals measurement device. The simulation tests show
that it is possible to integrate ARM micro-processor, with IC analog that collect ECG
signals (IC AD8232) to become handheld signal measurement device, additional func-
tion of transmitting data to the server via 3G/WIFI, to help doctor and medical center to
monitor the patients’ health status remotely, possible to invest and develop to become a
commercial product. On the other hand, paper has proposed a plan to use the RF net-
work (Random Forest) to build software for identifying ECG signals. This is not the best
solution, but it has an advantage is very simple, fast processing speed, suitable for large
data problems. The algorithms that were tested on data sheet of MIT-BIH, or used by
international research groups. The software has a high recognition result.
In the near future, we continue to optimize the measurement device and collaborate
with medical center to collect basis set of ECG signals. Optimize recognition device,
gradually testing on patients.

References
1. Hamilton, P.S., Tompkins, W.J.: Quantitative investigation of QRS detection rules using the
MIT/BIH arrhythmia database. IEEE Trans. Biomed Eng., BME-33, 1157−1165 (1986)
2. Thakor, N.V., Webster, J.G., Tompkins, W.J.: Optimal QRS detector. Med. Biol. Eng. 343−50
(1983)
3. và, G., Mark Moody, R.: The impact of the MIT-BIH arrhythmia database. IEEE Eng. Med.
Biol. 20(3), 45−50 (2001)
4. Breiman, L.: Random forests. Mach. Learn. 45, 5–32 (2001)
5. Emanet, N.: ECG beat classification by using discrete wavelet transform and Random Forest
algorithm. Decis. Control, 1 – 4 (2009)
6. Gao, P., Zhao, J., Wang, G., Guo, H.: Real time ECG characteristic point detection with ran-
domly selected signal pair difference (RSSPD) feature and random forest classifier. In: 2016
38th Annual International Conference of the IEEE Engineering in Medicine and Biology
Society (EMBC), pp. 732−735 (2016)
7. Tran, H.L., Van Pham, N., Nguyen D.T.: A hardware implementation of intelligent ECG
classifier. COMPEL: Int. J. Comput. Math. Electr. Electron. Eng. 34(3), 905−919 (2015)
8. http://www.physionet.org
 Electromagnetic Torque and Force Analysis
of Switched Reluctance Motors with Different
Rotor Pole Shapes

Bui Duc Hung, Dinh Bui Minh, and Vuong Dang Quoc(B)

Department of Electrical Engineering, Hanoi University of Science and Technology, Hanoi,

Vietnam
vuong.dangquoc@hust.edu.vn

Abstract. In this paper, analysis of the torque ripple and radial force reduction
by different rotor pole designs of a three-phase 6/4 switched reluctance motor is
presented. In order to improve torque performances by changing stator and rotor
angles for three-phase switched reluctance motors, many papers have recently used
the types of stator and rotor poles, with 6/4; 8/12; 18/12 and 24/18. Improvement of
the average torque and torque ripple have influenced directly on rotor poles designs,
because flux density distribution at the air gap will affect torque performance and
radial forces. Several rotor tapped-skew poles have been calculated for the three-
phase switched reluctance motor by the finite element method to give suitable
results. The efficiency, average torque and torque ripples have been simulated and
analyzed. A significant distribution of this research is to implement the skew-
tap rotor pole with the optimal tap-angle for maximizing torque ripple and radial
forces. The improvement of method is validated on the practical problem, with a
switched reluctance motor of 30 kW.

Keywords: Switched reluctance motors · Rotor pole shapes · Electromagnetic

torque · Efficiency · Finite element method

1 Introduction
Nowadays, switched reluctance motors (SRMs) have been applied as the best candidate
to drive this small-scale electric vehicle due to the advantages of simple construction,
such as wide constant power region; effective torque-speed characteristic in term of
material and manufacturing cost [2, 3]. The SRM used for a high electric vehicle (HEV)
is designed with the high starting torque and low speed, and the high efficiency and high
speed. Thus, the SRM can operate in a wide constant power region by applying optimal
turn-on and turn off angle co-ordinate with shaping current controller. The effective
torque and speed characteristic can be achieved by imposing the continuous current
with the higher speed and saturated mode. Many authors have recently applied a finite
element method (FEM) to improve torque performances by changing stator and rotor
angles for three-phase SRMs [1–4]. Or in [5–9], it has been shown that the average
torque and torque ripple influenced directly on rotor poles designs, because the torque

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 495–504, 2023.
https://doi.org/10.1007/978-3-031-22200-9_55
496 B. D. Hung et al.

performance and radial forces will be affected by the flux density distribution at the air
gap. The various fields for SRMs have been also proposed in [1–5].
In practice, a conventional design of SRM is considered as a reference to present the
efficient performance of the motor. Some design parameters must be considered such
as several phases, stator pole arc, air gap thickness and other SRM geometries. In this
paper, the influence of the SRM geometry on working performance has been to compute
the efficiency, torque, power and iron loss. The new contribution has pointed out the
suitable rotor topology with the modified skew tap pole to minimize radial forces. The
effect of pole embrace has been investigated on the performance of SRM by using the
ANSYS and SPEED tools [10]. The stator and rotor pole embraces can be changed from
0.2 to 0.5. According to the highest average torque and lowest torque ripple, proper pole
embraces have been selected for the optimum design of SRMs. The analyzed motor is
30 kW with the speed is high (27000 rpm) while the torque is only 25 Nm which can
apply for to electric vehicles with gear box or mid drive sport motorcycle.

2 Electromagnetic Torque Analysis

The arc angle (β) of rotor and stator poles is a critical part of the design of SRMs.
The rotor pole arc (β r ) has to be equal to or greater than stator pole arc (β s ), because
the number of rotor pole is generally less than a number of stator pole, i.e., β s ≤ β r .
Thus, when β s is selected as smaller than the step angle, none of the phases gets a rising
inductance slope. The term of β r should be greater than the sum of stator and rotor pole
arcs, i.e., [9, 10]:
360◦ 360◦
βr ≥ βs ; βr + βs < ; min(βr , βs ) ≥ (1)
Nr m.Nr
The static torque curves are performed by the FEM, where the current value is equal
to from 0 to 250A in the range of rotor angle.

1 2 dL(θ, i)
Tc = i (2)
2 dθ
where T c is the static torque, L is the inductance and θ is the phase angle arc between
rotor and stator. The static torque can be measured with the phase current and rotor
position from 0 to 360°/Nr (90°). The average torque is based on the different current
and speed (Fig. 1).
The embrace ratio of rotor (as ) and embrace ratio of stator (aR ) can be defined as
following equation:
βs βr
as = ; ar = (3)
τs τr
The average torque of the SRM is calculated as:

τk 
m
1
Tavg = Tk (θ, i)dt, (4)
τk
0 k=1
 Electromagnetic Torque and Force Analysis of Switched Reluctance … 497

Fig. 1. Cross-section of a three-phase SRM 6/4. Model 1: rotor round tooth βr = 42° (top left);
model 2: rotor straight tooth βr = 42° (top right); model 3: tap shoe βr = 42°; model 4: tap straight
shoe βR = 42°.

where τk is. The circle time, m is the number of phase SRM and Tk is the toque of one
phase.
The B-H curve of M270-35A is presented in Fig. 2. It has a high saturated flux density
up to 2 T. A full range of these standard grades is available for a steel grade that can be
selected according to the specific requirements of each application, from low alloy grades
offering the advantage of excellent magnetic permeability, thermal conductivity, and
punch ability to alloyed grades offering the advantage of very low losses, even at higher
frequencies. A wide range of coatings is available, allowing further enhancement of
punch ability, reduction of interlaminar losses and improvement of corrosion protection.
The rotors have no windings or permanent magnets. The SRM types are three-phase 6/4,
where 6 is the number of stator poles and 4 is the number of rotors. The detail winding
parameters are given in Table 2. The weight material of stator, rotor and total motor are
pre-determined in Table 3. The inductance, loss and power of the motor are given in
Table 4.

Table 1. Stator and rotor angle of SRM 6/4.

Ratio as Stator angle (°) Ratio ar Rotor angle (°)

0.30 18 0.30 27
0.35 21 0.35 31.5
0.40 24 0.40 36
0.45 27 0.45 40.5
0.50 30 0.50 45
0.55 33 0.55 49.5
0.60 36 0.60 54

In order to to be convenient, the ratio of pole angle/pole step is checked with the
pole angle. The stator pole angle value is calculated from 120 to 360, the value range
498 B. D. Hung et al.

Table 2. Winding parameters.

Series number 4
Parallel number 1
Number of tuns 15
Setting type Phase resistance
Slot-fill factor, % 71
Slot area, mm2 637.2
Conductor area, mm2 226.2
Phase resistance, ohm 0.02291
Winding Concentrated winding
Coil current density (@1A), A/mm2 0.06631

Table 3. Mass and weight of components.

Total Total weight, kg 22.34

Total volume, mm3 2.785e + 06
Stator Stator core−mass, kg 11.21
Stator core−volume, mm3 1.475e + 06
Coil−mass, kg 7.742
Coil−volume, mm3 8.641e + 05
Part weight (sr_stator) 18.95
Rotor Rotor core−mass, kg 3.391
Rotor core−volume, mm3 4.462e + 05
Part weight (sr001) 3.391
Inertia Rotor core, kg m2 3.984e-03
Total, kg m2 3.984e-03

of the ratio as is from 0.2 to 0.6 and the rotor pole angle is from 0.2 to 0.7 as in Table
1. Calculation results of stator and rotor are based on the pole angle to ratio stator pole
angle/stator pitch (as ) and rotor pole angle/rotor pitch (ar ) according to the Eq. (3).
For a certain ratio as stator pole angle, the ratio of rotor pole angle/rotor pitch ar
increases and the torque also increases and after reaching the pole value. Then the
average torque decreases. It means that the larger the rotor pole angle, the higher the
torque, especially in the range ar = 0.5 ÷ 0.7, the average torque decreases rapidly; the
average torque increases only with increasing rotor pole angle in the range ar = 0.2 ÷
0.40. From the value ar = 0.40 ÷ 0.5, the average torque starts to decrease.
On the above average torque calculation results, the maximum average torque points
can be reduced according to the ratio of stator pole angle/stator step as as shown in
 Electromagnetic Torque and Force Analysis of Switched Reluctance … 499

Table 4. Machine parameters.

Revolution speed N, rpm 2e + 04

Inductance Minimum inductance, H 2.878e−04
Maximum inductance, H 1.838e−03
Electric part Phase current (RMS) A 41.15
Wire current density, A/m2 1.364e + 06
Power Torque, Nm 4.956
Efficiency, % 67.08
Power, W 1.038e + 04
Power factor 0.3505
Loss Copper loss, W 38.84
Iron loss, W 5055

Fig. 2. B-H curve of M270-35A

Fig. 3. Average torque results.

500 B. D. Hung et al.

Fig. 3. In order to maximize the static electromagnetic torque, the term as should be in
range of 0.45; 0.5; 0.55; 0.6 or 13.50 ; 150 , 16.50 ; 180 . By simulating via the FEM, the
static torque and the rotor angle can be checked from 0 to 450 .

Fig. 4. Analysis of efficiency and power.

The torque and power of four designs have been calculated in Table 6. The model
has lowest torque ripple with 23.81% and radial force of 0.345 kN. Maximum efficiency
of 93.6% is quite high for electric vehicle applications. It can achieve high efficiency
with high current density of 20 A/mm2 and direct water cooling and controlling current
and turn-on/off angles. The electromagnetic performance results obtained by analytical
program coupling to Matlab and FEM to build those results. The analysis of efficiency
and power with the different models (from model 1 to model 4) are presented in Fig. 4
and Table 5. The current and iron loss are also given in Fig. 5.

Fig. 5. Current (left) and Iron loss (right).

3 Torque and Efficiency Performances

In order to compute the torque and efficiency performances, a MatLab model has been
proposed to validate the dynamic torque at speed up to 27000 rpm and phase current of
250A. The turn-on and turn-off angles are defined in Eq. (4). The torque and speed curve
 Electromagnetic Torque and Force Analysis of Switched Reluctance … 501

Table 5. Technical Parameters of SRM 6/4.

Technical parameters Model 1 Model 2 Model 3 Mode 4

Stator poles 6 6 6 6
Stator pole angle 30 30 30 30
Stator lamination (mm) 160 160 160 160
Stator bore (mm) 80 80 80 80
Stator pole depth (mm) 25 25 25 25
Pole number 4 4 4 4
Rotor pole angle 45 45 42 42
Skew rotor (mm) 0 0 -5 5
Pole Taper Angle -10 -5 -5 5
Airgap 0.3 0.3 0.3 0.3
Number Strands in Hand 22 22 22 22
Phases 3 3 3 3
Turns per phase 20 20 20 20
Rated Power (W) 30000 30000 30000 30000
Maximum Speed 27000 27000 27000 27000
Maximum Torque (N.m) 60 60 60 60
Rotor pole τR (360/Nr = 360/4) (degree) 90 90 90 90
Stator pole τR (360/Ns = 360/6) (degree) 60 60 60 60

plot was already shown in Fig. 2. The torque ripple obtained form 30N.m to 25 N.m is
expressed as [9, 10]

Ton = θon = −Nr ∗ Ns (βr− βs )
(5)
Toff = θoff = 1800 − Nr ∗ Ns (βr− βs )

The dynamic torque and efficiency with different rotor and stator angles have been
already compared in Table 2. The line current, power, losses, efficiency and torque has
been presented in Fig. 6.
From the results given in Table 5, the model with 4-rotor straight taped-skew pole
has many advantages and the highest average torque and efficiency. The distribution of
stator radial and tangential force and rotor position is pointed in Figs. 7 and 8. It can be
seen that the maximum force is equal to 1.2 Nm in the frame of angle from 0 to 45°.
In the same way, the distribution of rotor radial and tangential force and rotor position
is pointed in Fig. 9. It can be seen that the maximum force is equal to 1.09 Nm in the
frame of angle from 0 to 45°.
The phase current, flux linkage and torque waveforms are presented by the dynamic
Matlab simulation at a maximum speed of 27000 rpm. After running four simulation
502 B. D. Hung et al.

Fig. 6. Average torque and speed performances.

Fig. 7. Rotor radial and tangential force ad rotor position.

Fig. 8. Radial force harmonics of stator and rotor.

models, the comparison of the electromagnetic performances of torque ripple, efficiency,

radial and tangential forces for four models are given in Table 6.
 Electromagnetic Torque and Force Analysis of Switched Reluctance … 503

Table 6. SRM 6/4 electromagnetic performance comparison at the speed of 27000 rpm.

Parameters Model 1 Model 2 Model 3 Mode 4

Maximum torque possible (mm) 48.466 48.5649 48.6466 48.6722
Average torque (N.m) 21.642 22.9893 24.1424 25.0111
Torque ripple (N.m) 12.223 5.4121 5.9682 5.9376
Torque ripple [%] 57.93 29.622 23.958 23.812
Electromagnetic power (W) 916.1 1023.3 1032.4 1022.5
Input power (W) 301092 29895.1 31207.6 32198.8
Output power (W) 25220 23918.57 25821.95 26922.58
Total losses (on load) (W) 5872.7 5976.53 5385.65 5276.22
System efficiency (%) 91.11411 90.00833 92.74251 93.61361
Shaft torque (N.m) 26.945 25.771 28.8435 27.7685
Radial force max (kN) 0.37 0.35 0.36 0.345
Tangential force max (kN) 0.165 0.1540 0.155 152

4 Conclusions
The stator and rotor embraces/pitches influenced on average torque and torque ripples
have been successfully analyzed by the FEM to find out rotor/stator pole angles. Four
models of different normal, tapped-skew pole topologies for high speed switched reluc-
tance motor 30 kW–27000 rpm have been simulated to evaluate the torque and efficiency
performances. The tapped-skew pole with optimal tap angle of 450 mechanical degrees
has been also selected for the proposal design. The radial and tangential forces are also
compared and discussed.

References
1. Raj, M.A., Kavitha, A.: Effect of rotor geometry on peak and average torque of external-rotor
synchronous reluctance motor in comparison with switched reluctance motor for low-speed
direct-drive domestic application. IEEE Trans. Magn. 53(11), 1–9 (2017)
2. Han, J., Ge, B., Zhang, K., Wang, Y., Wang C.: Influence of control and structure parameters
on the starting performance of a 12/8 pole switched reluctance motor. Energies. 13(14) (2020)
3. Yueying, Z., Chuantian, Y., Yuan, Y., Weiyan, W., Chengwen, Z.: Design and optimization
of an In-wheel switched reluctance motor for electric vehicles. IET 13(1) (2019)
4. Yildirim, M., Kurum, H.: Influence of poles embrace on in-wheel switched reluctance motor
design. In: Proceedings of the 2018 IEEE 18th International Conference Power Electronics
Motion Control (PEMC), pp. 562–567 (2018)
5. Tap, A., Xheladini, L., Asan, T., Imeryuz, M., Yilmaz, M, Ergene, L.T.: Effects of the rotor
design parameters on the torque production of a PMaSynRM for washing machine applica-
tions. In: Proceedings of the 2017 International Conference Optimal Electrical Electronics
Equipment (OPTIM) & 2017 International Aegean Conference Electrical Machines Power
Electron (ACEMP), pp. 370–375 (2017)
504 B. D. Hung et al.

6. Siadatan, A., Roohisankestani, M., Farhangian, S.: Design and simulation of a new switched
reluctance motor with changes in the shape of stator and rotor in order to reduce torque
ripple and comparison with the conventional motor. In: SPEEDAM 2018-Proceedings of
the International Symposium Power Electronics, Electrical Drives, Automation and Motion,
pp. 353–358 (2018)
7. Rallabandi, V., Wu, J., Zhou, P., Dorrell, D.G., Ionel, D.M.: Optimal design of a switched
reluctance motor with magnetically disconnected rotor modules using a design of experiments
differential sevolution FEA-based method. IEEE Trans. Magn. 54(11), 1–5 (2018)
8. Siadatan, A., Fatahi, N., Sedaghat, M.: Optimum designed multilayer switched reluctance
motors for use in electric vehicles to increase efficiency. In: SPEEDAM 2018-Proceedings of
the International Symposium Power Electronics, Electrical Drives, Automation and Motion,
pp. 304–308 (2018)
9. Yueying, Z., Chuantian, Y., Yuan, Y., Weiyan, W., Chengwen, Z.: Design and optimisation of
an In-wheel switched reluctance motor for electric vehicles. IET Intell. Transp. Syst. 13(1),
175–182 (2019)
10. Bui, M.D., Schneider, S., Arnaout, S., Schaefer, U.: Torque maximization of a high-speed
switched reluctance starter in acceleration test. In: 2013 15th European Conference Power
Electronics Application, EPE 2013, vol. 1, pp. 1–5 (2013)
 Energy-Based Tracking Control with Vertical
Vibration Suspending for a 6-DOF Bridge Crane

Quoc-Dong Hoang1 , Le Anh Tuan1,1(B) , Do Duc Luu2 , and Le Sy Dung3

1 Institute of Mechanical Engineering, Vietnam Maritime University, Haiphong, Vietnam
{donghqvck,tuanla.ck}@vimaru.edu.vn
2 Institute of Research and Development, Maritime University, Haiphong, Vietnam
3 School of Mechanical Engineering, Hanoi University of Science and Technology,

Hanoi 100000, Vietnam

Abstract. A bridge crane, or 3D crane, is a classic research subject in the control

field and is considered a typical example in the group of underactuated systems.
This control problem is relatively fascinating and poses significant challenges
when the number of control signals is larger than that of state variables to be
controlled. Specifically, the task of the control algorithm is not only to ensure the
correct tracking with the desired trajectory but also to reduce and quickly extin-
guish unwanted oscillations. In this study, we develop a controller for the system
with a vertical-oscillation quenching capability, a problem that was rarely men-
tioned in previous studies. Based on the positive definite and the skewed symmetric
characteristics of the matrices of the mathematical model, a controller is designed
to guarantee the stability of the system according to Lyapunov criteria. Paralleling
the convergence of the tracking errors, the horizontal and vertical oscillations of
the cargo are suppressed. Simulation results are given at the end of the paper to
investigate the effectiveness and feasibility of the entire study.

Keywords: 3D crane · Overhead crane · Underactuated system · Energy-based

control · Nonlinear control · Vibration control

1 Introduction
3D crane is considered an underactuated system [1–4], which has attracted the attention
of numerous researchers over the past two decades [5, 6]. Most of the previous studies
focused on precisely controlling the trajectory and suppressing the oscillations of the
cargo being swing angles [7, 8]; whilst, the vertical vibration originating from the elas-
ticity of cable and steel structure is barely noticeable. The vertical oscillations, although
much smaller than the horizontal oscillations, have a significant negative impact on the
energy consumption and the life of the structure.
In some previous studies [9, 10], we proposed some control solutions with a new
model of a 6-degree-of-freedom crane and gave relatively positive results. In this study,
based on the newly built 6-degrees-of-freedom model, we will develop a simpler con-
troller, energy-based coupling control (EBC), to reduce computational resources and re-
experience the special characteristics of the dynamic model built based on the theory of

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 505–511, 2023.
https://doi.org/10.1007/978-3-031-22200-9_56
506 Q.-D. Hoang et al.

Lagrangian dynamics. Comparisons with classical PID controllers are also made through
simulation results. Eventually, the effectiveness of the whole work will be evaluated.

2 Dynamic Model

The physical model of an under-actuated indoor bridge crane is illustrated in Fig. 1.

Here, the six DOFs corresponds to six generalized coordinates, specifically the trolley
position along the girder x(t), the angle of the hoist cable ϕd (t), the displacement of the
bridge along the rail y(t), the swing motions of the cargo in the Oxz and Oyz planes, and
the vertical oscillation. Here, the angle of the angle ϕd (t) will determine the height of the
cargo. Moreover, the first three generalized coordinates, x(t), y(t), ϕd (t), are considered
as actuated states, and the last three, θx (t), θy (t), δ(t), are un-actuated.

a. Coordinate system in three-dimensional space

b. Sketch of physical model in Oxz plane

Fig. 1. Physical model of a bridge cran.
 Energy-Based Tracking Control … 507

The system consists of three masses: the cargo, the trolley with hoist, and the bridge,
which are denoted as mc , mt , and mb , respectively. And we set the non-negative function
E as the entire energy of the system. Depending on the particular working conditions and
type of handled cargo, the carrying equipment is adjusted and its mass can be significant.
In these cases, both the masses of the cargo and the lifting equipment are related to mc .
Here, δ denotes the initial elongation, ke refers to the equivalent elastic coefficient.
Jd , and rd represent the moment of inertia and radius of the payload-hoisting drum.
bm , bb , bt and br represent the damping factors of the hoisting mechanism, bridge,
trolley, and inside wire rope, respectively. τ1 , τ2 , and τ3 are active control inputs generated
by the trolley, bridge moving, and payload-lifting mechanisms, respectively.
The physical features of the indoor bridge crane are characterized for full-state
 T
variables χ s = x(t) y(t) ϕ(t) θx (t) θy (t) δ(t) .
The system dynamics are provided in matrix form as [9]:

M(χ s )χ̈ s +Bχ̇ s +C(χ s , χ̇ s )χ̇ s +G(χ s ) = U, (1)

 T  T
where χ̇ s = ẋ ẏ ϕ̇ θ̇x θ̇y δ̇ and χ̈ s = ẍ ÿ ϕ̈ θ̈x θ̈y δ̈ are the first-order and second-
order time derivatives of the system states, respectively; the input vector U is U =
 T  T
U s 0 0 0 with U s = τ1 τ2 τ3 ; the mass matrix M(χ s ) = M T (χ s ) is positive
definite (PD); C(χ s , χ̇ s ) represents a Coriolis and centrifugal matrix; B is a damping
coefficient matrix; and G(χ s ) denotes a gravity vector.

3 Controller Design
The system dynamic can be divided into actuated and un-actuated parts by decoupling
as

M 11 (χ s )χ̈ S1 +M 12 (χ s )χ̈ S2 +B11 χ̇ a +C 11 (χ s , χ̇ s )χ̇ S1 +C 12 (χ s , χ̇ s )χ̇ S2 +G1 (χ s ) = U s , (2)

M 21 (χ s )χ̈ S1 +M 22 (χ s )χ̈ S2 +B22 χ̇ S2 +C 21 (χ s , χ̇ s )χ̇ S1

+C22 (χ s , χ̇ s )χ̇ S2 +G2 (χ s ) = 0, (3)
 T  T  T
where χ S1 = x y ϕ , χ̇ S1 = ẋ ẏ ϕ̇ , and χ̈ S1 = ẍ ÿ ϕ̈ are the vectors of the
 T  T  T
actuated states; χ S2 = θx θy δ , χ̇ S2 = θ̇x θ̇y δ̇ , and χ̈ S2 = θ̈x θ̈y δ̈ are the
vectors belonging to the un-actuated parts; and the block matrices M 11 (χ s ), M 12 (χ s ),
M 21 (χ s ), M 22 (χ s ), B11 , B22 , C 11 (χ s , χ̇ s ), C 12 (χ s , χ̇ s ), C 21 (χ s , χ̇ s ), C 22 (χ s , χ̇ s ),
G1 (χ s ), and G2 (χ s ) are also given in [9].
The mass matrix M(χ S ) is positive definite, then we have
 
χ̈ S = M−1 (χ S ) U − Bχ̇ S − C(χ S , χ̇ S )χ S − G(χ S ) . (4)
  −1
Also, M(χ S ) > 0, χ TS M(χ S )χ S > 0, M −111 (χ S ) = det M 11 (χ S ) M phk3 (χ S ),
and M phk3 (χ S ) > 0 is the measurable auxiliary term
508 Q.-D. Hoang et al.

Let
 −1
χ̈ S1 = det(M(χ S )) (W a +P a U S ), (5)

where

W a = −M phk3 (χ S )Bχ̇ S − M phk3 (χ S )C(χ S , χ̇ S )χ S − M phk3 (χ S )G(χ S ), (6)

and

P a = M phk3 (χ S ). (7)

The mission of the controller is to force the system states χ s to their target points
 T
χ ST = xT yT ϕT θxT θyT δT .
 T
Define the tracking error being e(t) = χ S1 − χ S1T , where χ S1T = xT yT ϕT .
The controller of the system is computed as follow
  

1 d  −1
US = −kD χ̇ S1 − kp e − kp Pa−1 Wa − kp det(M(χ S ))Pa−1 χ̇ S1 ke +kp
2 dt
(8)
where, kD , ke , kp are the positive controller coefficients.
Selecting the Lyapunov function as
1 1 
V2 (t) = kp eT e + ke E + kp χ̇ TS1 det(M(χ S ))Pa−1 χ̇ S1 , (9)
2 2
where, E is the entire energy of the system
Its first-order derivative with respect to time is
  
d 1 T 1
V̇2 (t) = kp e e + kp χ̇ TS1 det(M(χ S ))Pa−1 χ̇ S1 + ke Ė
dt 2 2
 
1
= ke χ̇ TS M(χ S )χ̇ S + χ̇ TS Ṁ(χ S )χ̇ S +χ̇ TS G(χ S ) + kp χ̇ TS1 e
2
 

1 d
+ kp χ̇ TS1 det(M)Pa−1 χ̇ S1 + kp χ̇ TS1 det(M(χ S ))Pa−1 χ̇ S1 . (10)
2 dt
With assumption that the influence of the dampers is smaller in the system, then
⎧ T  ⎫
⎨ χ̇ S M(χ S )M−1 (χ S ) U − C(χ S , χ̇ S )χ S − G(χ S ) ⎬
V̇2 (t) =ke + kp χ̇ TS1 e
⎩ + 1 χ̇ T Ṁ(χ )χ̇ + χ̇ T G(χ ) ⎭
S S S
2 S S
1 d 

+ kp χ̇ TS1 det (M)Pa−1 χ̇ S1 + kp χ̇ TS1 det (M(χ S ))Pa−1 χ̇ S1
2 dt
   
1
= ke χ̇ TS −C(χ S , χ̇ S )χ̇ − G(χ S ) + U + Ṁ(χ S )χ̇ S + χ̇ TS G(χ S ) + kp χ̇ TS1 e
2
 

1 d
+ kp χ̇ TS det (M )Pa−1 χ̇ S1 + kp χ̇ TS1 det (M(χ S ))Pa−1 χ̇ S1 (11)
2 dt
 Energy-Based Tracking Control … 509

The matrix M c (χ S , χ̇ S ) = Ṁ(χ S )−2C(χ S , χ̇ S ) is the skew symmetric matrix.

Thus, we have
⎧   
⎫
⎪ 1
⎨ +kp det(M)χ S ))Pa−1 χ̇ S1 + kp
d
det(M(χ S ))Pa−1 χ̇ S1 ⎪
⎬
V̇2 (t) = χ̇ TS ke U + χ̇ TS1 2 dt (12)
⎪
⎩ ⎪
⎭
+kp e
⎧  ⎫
⎪ −1 1 ⎪
⎪ k U
⎨ e S + kp det(M(χ S ))P a (W a +P a S ⎪
U )⎬
det(M(χ S ))
V̇2 (t) = χ̇ S1
T  
. (13)
⎪
⎪ 1 d ⎪
⎪
⎩ + kp det(M(χ S ))P −1
a χ̇ S1 + kp e ⎭
2 dt

Then, the function V̇2 (t) is computed as V̇2 (t) = −kD χ̇ TS1 χ̇ S1 based on control
 2
law (8). Furthermore, V̇2 (t) = −kD χ̇ S1 2 ≤ 0, (kD > 0), V2 (t) < V2 (0), and the
function V2 (t) is stable
 in the sense of Lyapunov  since t → ∞, or
and convert to zero
lim V2 (t) = lim ke E + 21 kp eT e + 21 kp χ̇ TS1 det(M(χ S ))P −1
a χ̇ S1 = 0. It is inferred
t→∞ t→∞
that lim e(t) = 0, lim E(t) = 0, lim cos βm (t) = 1. Eventually, χ S1 is compelled to
t→∞ t→∞ t→∞
 T
χ S1T = ϕlT ϕrT βrT γrT , and χ S2 is converged to zero.

4 Simulation Implementation
The system parameters are obtained from a real crane system provided as: mc = 5,000 kg,
mb = 2,316.5 kg, mt = 371.9 kg, rd = 0.31 m, Jd = 180 kg m2 , bt = 310 Nm/s, bb =
350 Nm/s, bm = 170 Nm/s, br = 260 Nm/s, g = 9.81 m/s2 , ke = 300,000 N/m, δ =
0.01 m.
In order to clearly display the efficiency of the EBC controller, simulations are carried
out for the EBC controller (shown by the solid purple line) and compared with the PID
controller (shown by thes green dashed line). The simulation results are provided in
Figs. 2–3.
Different from almost previous studies, the simulations were carried out with the
parameters taken according to a real working crane with high lifting capacity and load.
To ensure that the motors operate without overload as well as reduce the impact, the
speed of the cargo movement in all directions should not exceed 0.6 m/s.
In Fig. 1, it can be seen that it takes about 18–25 s to drive the cargo to reach its
desired position according to the spatial coordinates when using the EBC controller. In
all three directions, these times are shorter than the PID control by more or less 5 s.
Furthermore, the cargo movement speeds in the EBC case are maintained at a stable
level (solid purple line), which can be compared with the PID controller (green dashed),
as shown in Fig. 2b, d, f.
The superiority of the EBC controller is more evident when observing the oscillations
in Fig. 3. Obviously, the amplitude of oscillation of the cargo when controlled by PID is
larger, and it remains for more than 30 s. Meanwhile, the oscillation only appears with
a small amplitude at the beginning and minimizes approximately 0 degrees during the
movement until the end.
With the use of the EBC controller, not only is the cargo towed to its correct trajectory
with minimized oscillations, but the travel speed is also limited to appropriate thresholds.
510 Q.-D. Hoang et al.

Fig. 2. Control the position of cargo

Fig. 3. The system’s vibrations

 Energy-Based Tracking Control … 511

5 Conclusion
This study proposed a control issue for a new model of the 3D crane with 6 degrees of
freedom, including the cargo’s vertical vibration. Based on the computational model’s
decoupling and the mathematical model’s special properties built on the basis of the
Lagrangian theory dynamics, a controller based on the energy functions and the Lya-
punov stability criterion was developed. The efficiency of the algorithm was investigated
through simulations and comparisons with traditional PID controllers. The study results
will be the premise to perform experiments on practical systems in the future.

Acknowledgment. Hoang Quoc Dong was funded by the Postdoctoral Scholarship Programme
of Vingroup Innovation Foundation (VINIF), code VINIF.2022.STS.01.

References
1. Hoang, Q.-D., Park, J., Lee, S.-G.: Combined feedback linearization and sliding mode control
for vibration suppression of a robotic excavator on an elastic foundation. J. Vib. Control (2020).
https://doi.org/10.1177/1077546320926898
2. Hoang, Q.-D., Park, J.-G., Lee, S.-G., Ryu, J.-K., Rosas-Cervantes, V.A.: Aggregated hierar-
chical sliding mode control for vibration suppression of an excavator on an elastic foundation.
Int. J. Precis. Eng. Manuf. 21(12), 2263–2275 (2020). https://doi.org/10.1007/s12541-020-
00422-9
3. Hoang, Q.-D., Tuan, L.A., Huynh, L.N.T., Viet, T.X.: Fractional Approach for Controlling the
Motion of a Spherical Robot BT—Intelligent Systems and Sustainable Computing. In: Reddy,
V.S., Prasad, V.K., Mallikarjuna Rao, D.N., Satapathy, S.C. (eds.), pp. 183–192. Springer
Nature Singapore (2022)
4. Xuan, H.L., et al.: Adaptive hierarchical sliding mode control using an artificial neural network
for a ballbot system with uncertainties. J. Mech. Sci. Technol. 36, 947–958 (2022)
5. Cuong, H.M., Dong, H.Q., Trieu, P.V., Tuan, L.A.: Adaptive fractional-order terminal sliding
mode control of rubber-tired gantry cranes with uncertainties and unknown disturbances.
Mech. Syst. Signal Process. 154, 107601 (2021)
6. Le, T.A., Kim, G.-H., Kim, M.Y., Lee, S.-G.: Partial feedback linearization control of overhead
cranes with varying cable lengths. Int. J. Precis. Eng. Manuf. 13, 501–507 (2012)
7. Tuan, L.A., Kim, J.-J., Lee, S.-G., Lim, T.-G., Nho, L.C.: Second-order sliding mode control
of a 3D overhead crane with uncertain system parameters. Int. J. Precis. Eng. Manuf. 15(5),
811–819 (2014). https://doi.org/10.1007/s12541-014-0404-z
8. Dong, H.Q., Lee, S., Ba, P.D.: Double-loop control with proportional-integral and partial
feedback linearization for a 3D gantry crane. In: 2017 17th International Conference on
Control, Automation and Systems (ICCAS), pp. 1206–1211 (2017). https://doi.org/10.23919/
ICCAS.2017.8204411
9. Hoang, Q.-D., et al.: Robust control with a novel 6-DOF dynamic model of indoor bridge
crane for suppressing vertical vibration. J. Braz. Soc. Mech. Sci. Eng. 44(5), 1–12 (2022).
https://doi.org/10.1007/s40430-022-03465-3
10. Le, H.X., et al.: Adaptive fuzzy backstepping hierarchical sliding mode control for six degrees
of freedom overhead crane. Int. J. Dyn. Control (2022). https://doi.org/10.1007/s40435-022-
00945-1
 Enhancing Total Transfer Capability
via Optimal Location of Energy Storage Systems
Using a Hybrid Improved Min-Cut Algorithm
and Genetic Algorithm

Tung Linh Nguyen1 , Ngoc Sang Dinh2 , Viet Anh Truong3 , Thanh Long Duong4 ,
Dao Huy Du5(B) , and Do Anh Tuan6
1 Hochiminh City, Vietnam
linhnt@epu.edu.vn
2 Hochiminh University of Architecture, Hochiminh City, Vietnam
3 Hochiminh University of Technology and Education, Hochiminh City, Vietnam
4 Industrial University of Hochiminh City, Hochiminh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen Province, Vietnam
daohuydu@tnut.edu.vn
6 Chau Industrial Techonology Joint Stock Company, Ha Noi, Vietnam

Abstract. This paper proposes a new optimal approach to determine the location
and size of the Energy Storage System (ESS) in transmission expansion planning
(TEP). The proposed method combines two steps in sequence: the improved Min-
Cut algorithm (IMCA) and the Genetic Algorithm (GA) algorithm to reduce the
search space and time for finding the optimal solution satisfying many constraints.
This algorithm reduces the computational volume significantly compared to pre-
vious algorithms and thereby shortens the calculation time leading to investors
making timely and competitive ESS system investment decisions. The research
results are applied in determining the position of the ESS in the IEEE 24-bus
system, which shows the possibility of a solution to the TEP problem.

Keywords: TEP · Energy storage systems · Min-cut algorithm · Genetic

algorithm

1 Introduction

In the problem of electricity system development planning, there are many problems
with planning for transmission expansion to achieve one or more specific goals. Among
them, efficiency optimization is considered one of the essential planning stages. There-
fore, critical planning solutions such as Transmission Line Expansion Planning (TEP),
Generation Expansion Planning (GEP) has been studied extensively in [1, 2].
TEP is still a vital solution for solving planning problems in the long run. Still, it is
not always a perfect solution to solve overload and congestion to achieve stability and
reliability [3]. The common disadvantage of TEP solutions is that they are expensive

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 512–524, 2023.
https://doi.org/10.1007/978-3-031-22200-9_57
 Enhancing Total Transfer Capability via Optimal Location 513

or have limitations on the number of additional towlines, constraints on the number of

lines per pole, or the cost of building a new towline. In this case, more powerful and
economical methods are often used to improve the system, such as restructuring the
power grid for the existing power source [4], controlling the power flow using FACTS
[5], or solutions to build new renewable energy sources [6, 7]. However, these methods
have pros and cons to varying degrees, and they are not perfectly interchangeable.
In recent years, the development of renewable energy sources is gradually replacing
fossil energies. The energy from solar and wind has developed rapidly in many countries
worldwide [6, 7].
The characteristic of these energy sources is that they cannot be controlled according
to the operator commands but depend on natural conditions at each time. Therefore, when
the electricity demand is lower than generating capacity, the energy transmitted to the
grid affects the security of the power system. In contrast, when electricity demand is high
at a time when the renewable energy plant is not generating electricity. For controlling
these power sources, one of the solutions is to use ESS in the power system. The ESS
will store the excess energy on the grid and transmit it back to the grid at the request of
the dispatcher to satisfy goals [1, 8].
Types of ESS sources are pretty flexible in construction but could redistribute the
power flow on the system to reduce the load on the part of the power transmission
system at the required location. However, like the fundamental problem of power system
planning, the issues of concern are determining the location and capacity of the necessary
ESSs to be installed on the power system, also known as the optimal problem when TEP
incorporates an energy storage system (ESSTEP).
The organization of the paper consists of six parts as follows. Part 1 introduces.
Part 2 introduces the theory of methods to solve the TEP problem and proposes a new
model to solve the ESSTEP programming problem. Part 3 Details IMCA algorithm and
its application in the search space reduction method in an optimization problem. Part 4
mathematical optimization model to solve the ESSTEP programming problem. Part 5
experiments on IEEE-RTS 24-bus system and gives discussion results. Section 6 offers
some concluding remarks.

2 ESSTEP Problem
In TEP planning, the optimization problem is one of the most important to be solved to
meet the operator’s goals. In studies on TEP, the many different models and algorithms to
solve optimization problems have been mentioned as AC and DC mathematical models,
meta-heuristic algorithms PSO, GA [2].
In the long-term options, the solution to expand the transmission line is one of the
more suitable options. However, this solution’s investment loss is relatively significant
compared with other aggressive solutions, such as using FACTs or ESS in the short term
[3, 5].
Active power sources such as ESS can also provide electricity when the power
system is congested at peak load and ensure stable operation of the power system without
expanding the transmission power system. Moreover, ESS also enhances the system’s
transmission capacity by injecting more power into the grid to reduce congestion and
pressure on existing power sources, ensuring the security of the power system [3, 4].
514 T. L. Nguyen et al.

In the recent commercial market with different energy storage technologies, the
battery is still the most popular storage device today [4]. In addition, to solve the problem
of short-term capacity shortage, there are other types of technologies such as flywheels,
supercapacitors. Energy storage by Hydrogen, compressed air are technologies with
larger capacity, and higher energy storage capacity, so it is more flexible in the power
system. Specific application conditions corresponding to each type of technology are
classified as shown in Fig. 1.
As stated in the introduction, when planning TEP, it is also necessary to consider
the issue of incorporating ESSTEP. The weaknesses in the ESS are also the weaknesses
of ESSTEP. Therefore, recent studies have developed large-capacity, high-performance
ESS tools in response speed and efficiency to solve the TEP problem [5]. Various methods
and models have been proposed in the literature in this field to solve the difficulties
mentioned above of the TEP planning problem. The models and techniques can be
organized as follows:

Fig. 1. Classification of ESSs [6]

Classical Optimization Algorithm: A commonly used analytic technique, and the

results are often optimal global solutions to the universal version of the initial integer
problem, where the initial data are integers, which are replaced by continuous variables.
The computational loads in these algorithms are often huge. Thus there are many obsta-
cles to solving extensive programming problems for medium and large power systems,
like most real power systems. In addition, this method can interfere with the convergence
problem in some cases, as in [6, 7].
Application of the minimum cross-sectional algorithm (MCA): different from other
algorithms, this algorithm always gives precise results, and the computational volume is
relatively low, so some studies use it to find the position of the minimum cross-section.
Optimal placement of FACTS installations [5, 8], solving bottlenecks in the power
 Enhancing Total Transfer Capability via Optimal Location 515

system [8], as well as studies on stability and reliability in transmission expansion

planning [9, 10].
Meta-heuristics Algorithms: These algorithms are inspired by optimizations in nat-
ural laws. They are extended to specific search procedures and can be used in almost
any case, and this is a particularly suitable solution for complex problems with many
uncertain constraints. Algorithms are often used to determine optimal or non-optimal
results of general issues, even for correspondingly large systems involving enormous
computational efforts [11].
Despite many advantages, the convergence probability and large computational vol-
ume of meta-heuristic algorithms are concerns of researchers, and this is an opportunity
to propose applying a hybrid method to overcome this problem. The defects in solving
optimization problems in TEP are presented in this paper; this hybridization leads to a
two-stage matching algorithm. First, improve the MCA algorithm to fit the objective of
the optimization problem, which is known as “limited search space. “The later stage,
called “Optimization of the Solution,” uses a genetic algorithm (GA) to find the opti-
mal global result in the reduced space of the previous stage. This process is perfectly
designed to overcome the disadvantage of the meta-heuristic algorithm as mentioned
above.

3 MCA Algorithm

3.1 Basic MCA Algorithm

In the 1950s, the problem of finding congestion of an undirected graph was proposed
by Ford and Fulkerson by determining the smallest slice of that undirected graph. This
solution is the premise for the MCA algorithm. Mechtild Stoer and Frank Wagner devel-
oped this algorithm to apply it to computer programs in 1997 and called it the Min-Cut
algorithm [12]. This model is also used to determine the optimal location of TCSC to
solve the problem of congestion in the power system [5]. The basic MCA algorithm
flowchart is shown in Fig. 2.
Subject to:
a matrix (n x n) of magnitude total [A] (this case power per capacity of lines).
n
is thenumber  of nodes.

sj= asj , jt= ajt - magnitude total with generators, loads of i cut and j branch.
S(i), T(i) - i cut with generators, loads.
u, v - buses.
{s}, {t} - set of nodes generators, loads.
MC(i) - mincut step i.
MF - maxflow.

3.2 Improved MCA Algorithm

The traditional MCA algorithm mentioned above is limited because it does not include
the distributed power flow on the transmission line, which leads to inadequate implemen-
tation of problem-solving objectives, such as when choosing a branch. In the minimum
516 T. L. Nguyen et al.

Fig. 2. Base MCA algorithm [5]

cut to solve congestion problems, transmission lines with large power distribution can
be ignored while these are critical locations that need to overcome congestion. Adding
branches with more significant power distribution to min-cut is an improvement of the
MCA algorithm to overcome the shortcomings of traditional MCA, which is proposed
in this paper.
The power flow distribution is the first step added to the Fig. 2 flowchart to check for
congested branches in the power system. When any branch is overloaded (congestion)
due to a high-power flow, it should be taken to update to the min-cut set during algorithm
execution.
Steps 6 and 2 of the added flowchart

sj (new) = sj + Kj

tj (new) = tj + Kj
 
Pflow = runopf (sys)
 Enhancing Total Transfer Capability via Optimal Location 517

Importance,
 
Pflow(j)
Kj = γ 1 −
Prate(j)

γ is an essential factor in this improvement. The value of γ is chosen according to the

contribution of power flow per branch. With the updates mentioned above compared
to the basic MCA algorithm, each branch is checked for power flow. The result is the
minimum possible cross-sectional set, including the units in the congested area in the
power system.

4 ESSTEP Mathematical Model

4.1 Function Fitness
Several recent studies on power system expansion planning have considered the instal-
lation of ESS. However, it is not yet widespread in the commercial market because the
investment capital of ESS is still relatively high in the market [13]. Therefore, the prob-
lem of optimizing the location and capacity of ESSs in the TEP planning to achieve a
reasonable cost when investing in a new ESS and the cost of operating efficient power
sources is expressed by the following objective function:
 
minimize C T = CiT + CjESS . (1)
i j

Subject to,

• CiT − generation cost of i source and calculated by:


CiT = ci (t).Ai (t) (2)
i∈G

• CjESS − investment cost of j ESS and calculated by:


CjESS = Piess .cess + C0 (3)
i∈Gess

With constrantion
 as.
• PGi − PDi − j∈bus Vi Vj (Gij cos δij + Bij sin δij ) = 0.(4)

• QGi − QDi − j∈bus Vi Vj (Gij sin δij + Bij cos δij ) = 0. (5)
• min ≤ V
Vi∈bus i∈bus ≤ Vi∈bus .(6)
max

• Si∈branch ≤ Si∈branch .(7)

max

• Location of ESSs ∈ mincut(8)

With,
cess – ESSs Fun investment.
C0 – an rate of ESSs cost.
518 T. L. Nguyen et al.

t – An evaluation cycle (years).

ci (t) – i generation cost in t cycle.
Ai (t) – i generation energy in t cycle.
Piess – i ESS capacity.
PGi , QGi , PDi , QDi – Power of generator and load.
max
Si∈branch – Capacity of i branch.
δij – Voltage angle.
min , V max – Min and max voltage.
Vi∈bus i∈bus

4.2 IMCA—GA Two Steps Algorithm

The optimization problem (1) with many solved methods has been studied previously
using heuristic, meta-heuristic algorithms, or mathematics [2]. The advantages and dis-
advantages of each method are not the same, but all the problems scientists and planners
are most interested in are related to finding the global extreme of the fastest problem,
especially for large power systems.
Among the more recently used algorithms, GA is a standard method used in TEP
programming to solve the dark problem. However, its characteristic is that computation
is too large in some systems like most other meta-heuristic methods, and sometimes
problem results can be trapped in local extremes.
Therefore, this paper proposes a two-stage combination between IMCA, an improved
MCA algorithm, and a GA algorithm to enhance the problem of computational volume
and local extremes of the GA algorithm, as mentioned.
The diagram in Fig. 3 depicts the hybrid problem as mentioned above in a five-step
sequence:

(1) Current capacity trend distribution of the operating power system.

(2) Use the IMCA algorithm in this paper to find the minimum cross-section set.
(3) Determine the minimum cross-sectional search space defined in step 2.
(4) Solve the optimization problem (1) by GA algorithm with the limit in space in step
3.
(5) Finally, the position and capacity of the ESSs.

5 24-Bus IEEE Test

They are using a standard IEEE RTS 24-bus power system to check the feasibility of the
proposed method in this paper. This system consists of 36 branches and 10 power sources,
and the data of this power system can be found in documents [14, 15], summarized in
Fig. 4.
According to the survey time, it is assumed that the power load capacity and power
source increase equally at all nodes. When the ratio increased to 160% of the standard
capacity (4,560MW and 5,448MW respectively according to Tables 1 and 2), the system
started to become congested at branches 6–10 and 7–8.
The power flow in step 1 (according to 4.2) is calculated using Matpower 6.0 software
according to the system data in this case and using the IMCA algorithm to determine
 Enhancing Total Transfer Capability via Optimal Location 519

Fig. 3. Hybrid IMCA and GA algorithm flowchart

the minimum cross-section by step 2 (in 4.2). The results are shown in Tables 3 and 4
below.
520 T. L. Nguyen et al.

Fig. 4. 24-bus IEEE-RTS system

Compares the results of two algorithms, MCA and IMCA. Overloaded branches
8–7 and 10–6 (with overload coefficients of 104 and 133%) were excluded from the
 Enhancing Total Transfer Capability via Optimal Location 521

Table 1. Power of generators (rate 160%)

Bus 1 2 7 13 15
Pg (MW) 307.2 307.2 480 945.6 344
Qg (MVar) 99.84 99.84 120 331.2 106.08
Bus 16 18 21 22 23
Pg (MW) 248 640 640 480 1,056
Qg (MVar) 86.88 160 160 96 397.76

Table 2. Power of loads (rate 160%)

Bus 1 2 3 4 5
Pd (MW) 172.8 155.2 288 118.4 113.6
Qd (MVar) 35.2 32 59.2 24 22.4
Bus 10 13 14 15 16
Pd (MW) 312 424 310.4 507.2 160
Qd (MVar) 64 86.4 62.4 102.4 32
Bus 6 7 8 9
Pd (MW) 217.6 200 273.6 280
Qd (MVar) 44.8 40 56 57.6
Bus 18 19 20
Pd (MW) 532.8 289.6 204.8
Qd (MVar) 108.8 59.2 41.6

Table 3. Min-cut result on 24-bus system

Basic MC New IMCA

Branches into min-cut 8–9 2–6
8–10 6–10
7–8
8–9
8–10

minimum cross-section in the MCA algorithm, while underloading branches (such as

9–8 and 10) −8 only 25 and 14% respectively) are included in the minimum cross-
section. In contrast, with the IMCA algorithm, implemented with the coefficient γ =
522 T. L. Nguyen et al.

Table 4. Power flow

Branch 10–8 8–7 9–8 10–6 6–2

Power flow (%) 14 133 25 104 51

175, the above-overloaded branches 6–10 and 7–8 are updated to the minimum cross-
section set in Table 4, and this is the congestion area that needs to be expanded of the
power system as stated [10].
According to the IMCA results, the set of nodes 10, 9, 8, 7, 6, and 2, which are
related to the min-cut found, are six favourable locations for installing ESSs to solve
the congestion of the power system at branch 9–8 and 10–8 as mentioned in [4]. This
implementation is step 3 of the problem.
The GA algorithm to solve the optimization problem in programming (1) is imple-
mented next in this step 4. The limit of the search space in the problem consists of six
locations defined in step 3 above. The positions of the ESSs can be located at nodes 2,
6, 7, 8, 9, and 10, respectively. The capacity of the ESSs is assumed to allow a stepwise
change of 10MW.
ESS capacity of 10MW at node 6 and 30MW at node 8 is the result of solving the
problem. And this result is entirely consistent with the results calculated and checked
by the corresponding AC mathematical method using Matpower 6.0 software.

6 Conclusion
Planning to expand the power system is indispensable in the process of socio-economic
development of a country, and it is the task of scientists as well as strategic planners. In
addition, in recent times, aiming at a strategic replacement for fossil fuels that pollute the
environment, renewable energy sources are of great interest in the world. These sources
and other canopies are widely used in TEP [8].
According to the research, the application of ESS source in power system planning
is a flexible and feasible solution to prolong the investment period to expand the system
and thus improve investment efficiency. This result brings economic benefits and has the
effect of levelling power flow between peak and off-peak load times, helping to operate
the power system more efficiently, making the most of renewable energy sources, like
wind and solar.
In this paper, the research results have pointed out two outstanding issues:
(i) Improve the MCA algorithm suitable for the optimization problem and propose
a hybrid method between IMCA and GA to reduce the search space of the optimization
problem, thus reducing the computational load compared to the traditional GA method.
(ii) The combination of the improved IMCA and the GA algorithm used to improve
the efficiency of the optimization problem and determine the result is a more powerful
tool because it can find the plan results, and the ESS resources are more efficient.
Therefore, in the future, this method can be applied in practice and research and
development more widely, specifically to continue testing other hybrid combinations, or
it can also be developed for multi-stage planning.
 Enhancing Total Transfer Capability via Optimal Location 523

Acknowledgments. This research is funded by Thai Nguyen University of Technology (TNUT).

References
1. Mahdavi, M., Monsef, H.: Review of static transmission expansion planning. J. Electr. Control.
Eng. 1(1), 11–17 (2011)
2. Lumbreras, S., Ramos, A.: The new challenges to transmission expansion planning. Survey
of recent practice and literature review. Electr. Power Syst. Res. 134, 19–29 (2016)
3. Lee, C., Ng, S.K., Zhong, J., Wu, F.F.: Transmission Expansion Planning From Past to Future.
Proc Power Systems Conf. Expo. PSCE 06(03), 257–265 (2006)
4. Ilic, M., Galiana, F., Fink, L.: Power systems restructuring: engineering and economics.
Springer Science and Business Media, LLC, New York (2013)
5. Duong, T.L., Yao, J., Truong, V.A.: A new method for secured optimal power flow under
normal and network contingencies via optimal location of TCSC. Int. J. Electr. Power Energy
Syst. 52(8), 2013
6. Herbert, G.M.J., Iniyan, S., Sreevalsan, E., Rajapandiand, S.: A review of wind energy
technologies. Renew. Sustain. Energy Rev. 11(6), 1117–1145 (2007)
7. Solangi, K.H., Islam, M.R., Saidur, R., Rahim, N.A., Fayaz, H.: A review on global solar
energy policy. Renew. Sustain. Energy Rev. 15(4), 2149–2163 (2011)
8. Akinyele, D., Rayudu, R.: Review of energy storage technologies for sustainable power
networks. Sustain. Energy Technol. Assess. 8, 74–91 (2014)
9. Makansi, J., Abboud, J.: Energy Storage: the missing link in the electricity value chain-An
ESC. White Paper, Energy Storage Counc 1–23 (2002)
10. Hemmati, R., Hooshmand, R.-A., Khodabakhshian, A.: Comprehensive review of generation
and transmission expansion planning. Inst. Eng. Technol. 7(9), 955–964 (2013)
11. Hadjipaschalis, L., Poullikkas, A., Efthimiou, V.: Overview of current and future energy
storage technologies for electric power applications. Renew. Sustain. Energy Rev. 13(6–7),
1513–1522 (2009)
12. Dunn, B., Kamath, H., Tarascon, J.-M.: Electrical energy storage for the grid: a battery of
choices. Science 334, 928–935 (2011)
13. Mendonça, I.M., Ivo, C.S., Marcato, A.L.M., Dias, B.H.: Transmission expansion optimiza-
tion via constructive heuristic technique. IEEE Grenoble PowerTech. 1–5 (2013)
14. Garver, L.L.: Transmission network estimation using linear programming. IEEE Trans. Power
Appar Syst. PAS-89(7), 1688–1697 (1970)
15. Sánchez, I.G., Romero, R., Mantovani, J.R.S., Rider, M.J.: Transmission—expansion plan-
ning using the dc model and nonlinear—programming technique. IEE Proc. Gener. Transm.
Distrib. 152(6), 763–769 (2005)
16. Duong, T.L., Yao, J., Truong , V.A.: Application of Min Cut Algorithm for Optimal Location
of FACTS Devices Considering System Loadability and Cost of InstallationInt. J. Electr.
Power Energy Syst. 63(12) (2014)
17. Tran, T., Choi, J., Park, J.K., Moon, S.I., El-Keib, A.A.: A fuzzy branch and bound-based
transmission system expansion planning considering ambiguities. IEEE Power Engineering
Society General Meeting (2004)
18. Choi, J.S., Tran, T.T., Kang, S.R., Jeon, D.H., Lee, C.H., Billinton, R., Jaeseok, C.: A study on
the optimal reliability criteria decision for a transmission system expansion planning. IEEE
Power Engineering Society General Meeting (2004)
19. Rocha, M.C., Saraiva, J.T.: Multiyear Transmission Expansion Planning Using Discrete
Evolutionary Particle Swarm Optimization. Energy Market (EEM), 2011—8th International
Conference on the European, pp. 802–807 (2011)
524 T. L. Nguyen et al.

20. Stoer, M., Wagner, F.: A simple min-cut algorithm. J. ACM 44(4), 585–591 (1997)
21. Teleke, S.: Energy storage overview: applications, technologies and economical evaluation.
White Paper, Quanta Technology, pp. 1–11 (2011)
22. IEEE Reliability Test System: IEEE Trans. Power Appar. Syst. PAS-98(6), 2047–2054 (1979)
23. The IEEE Reliability Test System: IEEE Trans. Power Syst. 14(3), 1010–1020 (1999)
 Evaluation of Model Order Reduction
Algorithms for Unstable High-Order System
Applied to Large Power System

Thanh-Tung Nguyen1 , Huy-Du Dao2(B) , Ngoc-Kien Vu2 , Van-Ta Hoang3 ,

and Thanh-Binh Nguyen Thi2
1 Thai Nguyen University of Information and Communication Technology, Thainguyen,
Vietnam
2 Thai Nguyen University of Technology, Thainguyen, Vietnam
daohuydu@tnut.edu.vn, nttung@ictu.edu.vn
3 College of Technology and Trace, Thainguyen, Vietnam

Abstract. This paper presents and evaluates the method of order of large-scale
unstable power systems using Modal truncation (MT), Balanced truncation (BT),
Positive-real balanced truncation (PRBT), Balanced stochastic truncation (BST)
and Linear-quadratic Gaussian balanced truncation (LQGBT). The results show
that the LQGBT and BT algorithm has the smallest order reduction error. The MT
method has the largest order of decreasing error. The BST algorithm gives the best
time domain response. The PRBT method preserves the passivity of the original
system. The simulation results show the advantages and disadvantages and the
application range of step reduction methods.

Keywords: Modal truncation · Balanced truncation · PRBT · BST · LQGBT ·

Large power systems · Model order reduction

1 Introduction
Considering a kinematic system G(s) invariant in linear time of the first order described
by the equivalent equations:
  
ẋ(t) = Ax(t) + Bu(t) AB
G(s) : ⇔ G(s) := ⇔ G(s) := C(sI − A)−1 B + D
y(t) = Cx(t) + Du(t) CD
(1)

where A ∈ Rn×n , B ∈ Rn×m , C ∈ Rp×n , D ∈ Rp×m

In many cases, such as large-scale circuit simulation, n can be very large, while
the number of inputs m, outputs p is much smaller than n. When installed, the higher-
order system makes computation impossible due to hardware limitations, memory, time,
reliability, and some other limitations. One approach to overcome this is Model Order
Reduction (MOR). The goal of MOR is to create a low-dimensional (low-order) sys-
tem with the same response characteristics as the original system, but with the storage

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 525–536, 2023.
https://doi.org/10.1007/978-3-031-22200-9_58
526 T.-T. Nguyen et al.

requirements, evaluation time, computing power, and deployment cost. Declaration, etc.
is much lower. The reduced-order model can be used to replace the high-order generation
or used to develop low-order controllers for real-time applications.
The problem of model reduction is stated as follows: Given the system G(s) in (1),
then its order reduction system Gr (s) is defined as:
  
ẋ (t) = Ar xr (t) + Br u(t) Ar Br
Gr (s) : r ⇔ Gr (s) := ⇔ Gr (s)
yr (t) = Cr xr (t) + Dr u(t) Cr Dr
:= Cr (sI − Ar )−1 Br + Dr (2)

where Ar ∈ Rr×r , Br ∈ Rr×m , Cr ∈ Rp×r , Dr ∈ Rp×m , R < n, and satisfy the following
basic requirements:

– the error is small.

– System properties, such as stability, passivity, etc., are preserved.
– Efficient in terms of calculation.

In order to investigate the possibility of reducing the complexity for large-scale to

investigate the possibility of reducing complexity for high-order systems, along with the
criterion of preserving some properties of the original system, the authors approach two
groups of methods: Modal Analysis (MA) and Singular Value Decomposition (SVD).
The Modal truncation algorithm represents a group of MA-based methods [1] that
will retain significant eigenvalues or significant poles. The modal truncation method has
been developed in many works such as: S. Zhang used a modal truncation method based
on an integrated structural-electromagnetic criterion is proposed for model reduction
of cable mesh reflector antennas [2]. It can reduce the order of finite element model
directly orientating to the electromagnetic performance with acceptable accuracy. An
offset cable mesh reflector and a boom are employed in this method to show its feasibility
and effectiveness in modal truncation. Modeling and trajectory tracking control of a new
parallel flexible link robot in the document [3]. These models are further reduced by
modal truncation and a Component Mode Synthesis minimizing the number of elastic
degrees of freedom, which is necessary for real-time control purposes. Document [4]
gives Modal analysis based on magnetic dipole theory brings out detailed explanations
for broadband circular polarization response which is obtained after truncating corners of
rectangle slot antenna. Orthogonal modes are excited by truncating corners of coplanar
waveguide fed rectangle slot antenna.
Balanced truncation (BT) represents SVD method group computes a lower-order
approximation of your model by neglecting states that have relatively low effect on the
overall model response. Using a lower-order approximation that preserves the dynamics
of interest can simplify analysis and control design [5]. The BT method is improved,
modified and applied in many cases such as: Authors in the document [6] introduce
a balanced stochastic truncation in model order reduction of a coupled 3D intercon-
nect to provide uniform approximation of the frequency response of the original sys-
tem over the whole frequency domain. The proposed approach can significantly reduce
the complexity of interconnect, while retaining high accuracy in comparison to the
original model. Paper [7] presents a model reduction approach to study the dynamic
 Evaluation of Model Order Reduction Algorithms for Unstable High-Order System 527

responses of inverter-dominated networked microgrids (MGs) under disturbance. The

developed reduced-order model represents a closed-loop simulation that combines both
linear model reduction for reduced system dimensions and detailed nonlinear model for
better accuracy. This article [8] considers balanced truncation of discrete-time Hankel
k-positive systems, characterized by Hankel matrices whose minors up to order k are
nonnegative. It provides a broad class of systems where balanced truncation is guaran-
teed to result in a minimal internally positive system. Article [9] is dedicated to model
order reduction of linear time-invariant passive systems, i.e., positive real (PR) systems,
based on the balanced truncation (BT) concept. The main feature of this article is that we
have used the phase angle of the transfer function, G(s), in model order reduction for the
first time in the proposed method, which results in more accurate reduced-order models.
In this paper [10], Olympia Axelou applied a computationally efficient low-rank BT
procedure based on the extended Krylov subspace method, that can handle large-scale
models and significantly accelerate the electromigration stress analysis.
In this paper, the authors apply 5 algorithms for unstable high-order systems and
for large-scale power systems [14]. The algorithms used are MT, BT along with those
with the ability to preserve the characteristic property of the original system that is
unstable (preserving the stable subsystem), and extended algorithms of BT to preserve
phase minima with Balanced stochastic truncation algorithm, direct order reduction for
unstable systems with Linear-quadratic Gaussian balanced truncation and conservation
complete circuit passivity with Positive-real balanced truncation.

2 Modal Truncation (MT) Method

The basic idea of MT is to project the LTI system onto an invariant subspace corre-
sponding to the constraints (poles, eigenvalues). This method will select eigenvalues
with non-negative or small negative real parts to dominate the long-term dynamics,
omitting the fast decaying and less important large negative real components [1]. The
method truncation algorithm based on eigenvalues is described as follows:
Input: The system (A, B, C, D) is asymptotically stable and minimally of order n
described as (1):

– Step 1: Calculate the state transformation matrix

T and T −1 such that:

⎡ ⎤ ⎡−⎤
λ1  
⎢ .. ⎥ −1 ⎢B1 ⎥; CT =
− −
T −1 AT = ⎣ . ⎦ ; T B = ⎣ − ⎦ C C (3)
1 2
λn B
2

where λ1 , ..., λn , … are the singular values of the system arranged in ascending order
0 > Re(λ1 ) ≥ Re(λ2 ) ≥ . . . ≥ Re(λn ).

– Step 2: Select the order r to reduce.

– Step 3: Determine the matrices V and W satisfying:
528 T.-T. Nguyen et al.

 
In
V =T = T (:, 1 : r) (4)
0
   T
I
W = T −T r = T −1 (1 : r, :) (5)
0

Output: The asymptotically stable order reduction system as (2) (Ar , Br , Cr , Dr )

is determined:

Ar = W T AV ; Br = W T B; Cr = CV ; Dr = D (6)

3 Balanced Truncation (MT) Method

Balanced truncation (BT) was introduced in the systems and control theory in the early
1980s. The concept of the BT depends on the controllability and observablity grami-
ans matrices which are the symmetric positive semi definite solution of the Lyapunov
equations. The BT algorithm is described as follows [5]:
Input: Asymptotic stable and minimally system (A, B, C, D) of order n is described
as (1):

– Step 1: Solve two Lyapunov equations to determine the control Gramian P and the
observed Gramian Q: (P > 0; Q > 0).

AP + PAT = −BBT (7)

AT Q + QA = −C T C (8)

– Step 2: Cholesky analysis for P and Q into:

P = RR (9)

Q = LL (10)

– Step 3: Singular Value Decomposition SVD:

L R = U V  (11)
 Evaluation of Model Order Reduction Algorithms for Unstable High-Order System 529

– Step 4: Calculate the transformation matrices:

1 1
T := RV  − 2 ; T −1 =  − 2 U  L (12)

– Step 5: Balance transformation:

     
−1 −1 A11 A12 B1  
T AT , T B, CT , D = , , C1 C2 , D (13)
A21 A22 B2

– Step 6: Select the order r to be reduced. (r < n).

Output: Asymptotically stable order reduction system (Ar , Br , Cr , Dr ) =

(A11 , B1 , C1 , D).

4 Positive-Real Balanced Truncation (PRBT) Method

An important dynamic factor is the positive (passive) real system class. In the physical
sense, positive realism means that the energy produced by the system cannot exceed
the energy it receives. The circuit is part of a positive real dynamics system. The
asymptotically stable system (1) is positive real if [11]:

m = p, DT + D ≥ 0;G ∼ (jw) + G(jw) ≥ 0(∀w ∈ R)

Gramian control P and Gramian observe Q are solved from 2 Riccati (14), (15)
equations satisfying: (where P = P T > 0; Q = QT > 0)
  −1  
A P + PA + PB − C  D + D B P − C = 0 (14)

  −1  
AQ + QA + QC  − B D + D CQ − B = 0 (15)

Then the extreme solutions of P and Q are solved from 2 Eqs. (14), (15), satisfying:
0 < Pmin ≤ P ≤ Pmax ; 0 < Qmin ≤ Q ≤ Qmax
−1
 
Pmin = Qmin = Pmax = Q−1max = diag π1 Is1 , . . . , πq Isq (16)

where 1 ≥ π1 > π2 > · · · > πq > 0 are the positive real singular values of the system
and s1 + · · · + sq = n
The Positive -real balanced truncation algorithm is presented as follows:
Input: An asymptotically stable system and a balanced minimum perception (A, B,
C, D) of order n are described as (1).
530 T.-T. Nguyen et al.

– Step 1: Determine the control Gramian P and the observed Gramian Q from 2 Eqs. (17),
(18).
– Step 2 ÷ Step 6: From P and Q determined from step 1, the remaining steps are
performed similarly to the Balanced Truncation Algorithm.

Output: Asymptotically stable order reduction system: (Ar , Br , Cr , Dr ) =

(A11 , B1 , C1 , D).

5 Balanced Stochastic Truncation (BST) Method

Considering a nonsingular, minimal, asymptotically stable, randomly balanced system,

m = p and det(D) = 0 are described as (1). The Balanced stochastic truncation algorithm
is presented as follows [12]:
Input: An asymptotically stable system and a balanced minimum perception (A, B,
C, D) of order n are described as (1):

– Step 1: Determine the control (approach) Gramian P from the Lyapunov Eq. (7) and
the observed Gramian Q from the Ricacti Eq. (17):

 T  −1  
 
AT Q + QA + C − BW Q DDT C − BW Q =0 (17)

where BW := PC T + BDT

– Step 2 ÷ Step 6: From P and Q determined from step 1, the remaining steps are
performed similarly to the Balanced Truncation Algorithm.

Output: Asymptotically stable order reduction system: (Ar , Br , Cr , Dr ) =

(A11 , B1 , C1 , D).

6 Linear-Quadratic Gaussian Balanced Truncation (LQGBT)

Method

Linear-quadratic Gaussian balanced truncation was developed for unstable standard-

state-space systems in [13]. The LQGBT method will determine the control Gramian P
and the observed Gramian Q based on the Riccati equations. The LQGBT algorithm is
described as follows:
Input: The system (A, B, C, D) of order n is described as (1):

– Step 1: Solve the two Riccati Eqs. (18), (19) to determine the control Gramian P and
the observed Gramian Q:
 Evaluation of Model Order Reduction Algorithms for Unstable High-Order System 531

  −1  
AP + PAT + BBT = PC T + BDT I + DDT CP T + DBT z (18)
  −1  
AT Q + QA + C T C = QT B + C T D I + DT D BT Q + D T C (19)

– Step 2 ÷ Step 6: From P and Q determined from step 1, the remaining steps are
performed similarly to the Balanced Truncation Algorithm.

Output: Reduction system: (Ar , Br , Cr , Dr ) = (A11 , B1 , C1 , D).

7 Reduce Model for Higher Order Power System

Consider a large electrical system described in document [14]. This is a model of the
New England Electrical System (CEPEL, Brazil). This benchmark is described by the
data file ww_36_pmec_36.mat, which is an SISO system of order 66, consisting of 4
matrices A66×66 , B66×1 , C 1×66 , D1×1 . This system is unstable because there is a pole
with a positive real part.
Apply the above 5 algorithms to reduce the complexity of the original system by
equivalent systems of lower order. Reduce the order of this system to 6th, 10th and 26th
order systems. Conducting simulations, determining impulse responses and bode plots,
we obtain the results as Figs. 1, 2, 3, 4, 5 and 6.

Fig. 1. Impulse responses of original system (66th order) and 6th order reduction systems using
algorithms MT, BT, PRBT, BST, LQGBT.

From the impulse responses of the 6th, 10th and 26th order reduction system when
using 5 methods compared to the original system, it can be seen that:

– In the 6th order reduction system, the BT and LQGBT methods give the closest
impulse response to the original system and the MT method deviates the most.
532 T.-T. Nguyen et al.

Fig. 2. Impulse responses of original system (66th order) and 10th order reduction systems using
algorithms MT, BT, PRBT, BST, LQGBT.

Fig. 3. Impulse responses of original system (66th order) and 26th order reduction systems using
algorithms MT, BT, PRBT, BST, LQGBT.

– In the 10th order reduction system, the BT, PRBT, BST and LQGBT methods give
impulse response that coincides with the original system while the MT method deviates
greatly.
– In the 26th order reduction system, the algorithms MT, BT, PRBT, BST and LQGBT
all give impulse response that coincides with the original system.

From the Bode plots of the 6th, 10th and 26th order reduction system when using 5
methods compared to the original system, it can be seen that:

– In the 6th order reduction system, the BST method gives the amplitude and phase
angle response closest to the original system, while the MT method has the most
deviation.
 Evaluation of Model Order Reduction Algorithms for Unstable High-Order System 533

Fig. 4. Bode plot of the original system (66th order) and the systems reduced to the 6th order
using algorithms MT, BT, PRBT, BST, LQGBT.

Fig. 5. Bode plot of the original system (66th order) and the systems reduced to the 10th order
using algorithms MT, BT, PRBT, BST, LQGBT.

– In the 10th order reduction system, the BT, PRBT, BST and LQGBT methods give
the amplitude and phase angle response closest to the original system, while the MT
method has the most deviation.
– In the 26th order reduction system, the algorithms BT, PRBT, BST and LQGBT all
give the amplitude response and phase angle coincident with the original system,
while the MT method has a response close to the original system.
534 T.-T. Nguyen et al.

Fig. 6. Bode plot of the original system (66th order) and the systems reduced to the 26th order
using algorithms MT, BT, PRBT, BST, LQGBT.

Absolute error according to Hinf standard of the reduced order system (6th, 10th
and 26th order) compared with the original system (66th order) when using 5 methods:
MT, BT, PRBT, BST, LQGBT are shown in Tables 1, 2 and 3, respectively.

Table 1. Error of the 6th order system relative to the original system.

Method Absolute error

MT 0.012944
BT 0.002082
PRBT 0.007050
BST 0.009023
LQGBT 0.002074

Table 2. Error of the 10th order system relative to the original system.

Method Absolute error

MT 0.012876
BT 0.000204
PRBT 0.008904
BST 0.000814
LQGBT 0.000291
 Evaluation of Model Order Reduction Algorithms for Unstable High-Order System 535

Table 3. Error of the 26th order system relative to the original system.

Method Absolute error

MT 7.665993 × 10–5
BT 2.193033 × 10–8
PRBT 0.010409
BST 4.639587 × 10–8
LQGBT 5.111156 × 10–6

– In the 6th order reduction system, the LQGBT method gives the smallest error,
followed by the BT, PRBT, BST and MT methods, respectively.
– In the 10th order reduction system, the method with the smallest error is BT, followed
by the methods LQGBT, BST, PRBT and MT respectively.
– In the 26th order reduction system, the method for the smallest error is BT, followed
by BST, LQGBT, MT and PRBT respectively.

Based on the properties of the order reduction methods and the comparison results,
we have the following comments:
-

– Before proceeding to reduce the order of the original system (unstable system) using
MT, BT, PRBT and BST methods, it is necessary to analyze split the system into 2
subsystems: stable system and unstable system, then proceed to reduce order in the
stable subsystem, while LQGBT method can reduce order directly for an unstable
system.
– The MT and BT methods preserve the stability (or instability) of the original system.
– The BST method preserves the minimal phase of the original system.
– The PRBT method preserves the passivity of the original system.
– If the requirements of the large-scale power system order reduction problem are
mainly concerned with the impulse response, the algorithm BT and LQGBT should
be selected.
– If the requirements of the large-scale power system order reduction problem are mainly
concerned with the frequency domain response, the BST algorithm should be selected.
– If the requirement of the large-scale power system order reduction problem [1] with
direct order reduction for unstable systems, and for response in time domain, frequency
domain, with minimum order reduction error, we should choose LQGBT algorithm,
with 6th order.

8 Conclusion
The paper has introduced 5 order reduction algorithms: Modal truncation (MT), Balanced
truncation (BT), Positive-real balanced truncation (PRBT), Balanced stochastic trunca-
tion (BST) and Linear-quadratic Gaussian balanced truncation (LQGBT). After that, the
536 T.-T. Nguyen et al.

authors reduce the order for the unstable power system with the order of 66th. From
the results of order reduction, each algorithm has its own advantages and disadvantages.
With the given evaluations, we can choose the order reduction method suitable to the
requirements of the order reduction problem for large power systems. The next research
direction of the authors is to evaluate and compare other order reduction algorithms with
algorithms or to improve and develop the methods mentioned in the article.

Acknowledgments. This research is funded by Thai Nguyen University of Technology (TNUT).

References
1. Benner, P., Quintana-Orti, E.S.: Model reduction based on spectral projection methods. In:
Benner, P., Mehrmann, V., Sorensen, D. (eds.) Dimension Reduction of Large-Scale Systems,
Vol. 45 of Lecture Notes in Computational Science and Engineering, pp. 5-45. Springer,
Berlin/Heidelberg, Germany (2005)
2. Zhang, S., Duan, B., Du, J., Zhang, Y.: Model reduction of cable mesh reflector antennas
with integrated structural-electromagnetic criterion. IEEE/ASME Trans. Mechatron. 23(2),
927–935 (2018)
3. Morlock, M., Meyer, N., Pick, M., Seifried, R.: Modeling and trajectory tracking control of a
new parallel flexible link robot. In: IEEE/RSJ International Conference on Intelligent Robots
and Systems (IROS) (2018)
4. Odhekar, A.A., Deshmukh, A.A.: Coplanar waveguide fed broadband circularly polarized
corner truncated slot antenna. In: 2021 4th Biennial International Conference on Nascent
Technologies in Engineering (ICNTE), pp. 1–6 (2021)
5. Safonov, M.G., Chiang, R.Y.: A Schur method for balanced model reduction. IEEE Trans.
Automat. Contr. 34(7), 729–733 (1989)
6. Zjajo, A., van der Meijs, N., van Leuken, R.:Balanced stochastic truncation of coupled 3D
interconnect. Proceedings of 2013 International Conference on IC Design and Technology
(ICICDT), , pp. 13–16 (2013)
7. Du, Y., Lu, X., Zhao, D.: Model reduction for inverter-dominated networked microgrids with
grid-forming inverters. In: IECON 2021—47th Annual Conference of the IEEE Industrial
Electronics Society, pp. 1–6 (2021)
8. Grussler, C., Damm, T., Sepulchre, R.: Balanced truncation of k-positive systems. IEEE Trans.
Autom. Control 67(1), 526–531 (2022)
9. Salehi, Z., Karimaghaee, P., Khooban, M.-H.: A new passivity preserving model order reduc-
tion method: conic positive real balanced truncation method. IEEE Trans. Syst. Man Cybern.
Syst. 52(5), 2945–2953 (2022)
10. Axelou, O., Floros, G., Evmorfopoulos, N., Stamoulis, G.: Accelerating electromigration
stress analysis using low-rank balanced truncation. In: 18th International Conference on
Synthesis, Modeling, Analysis and Simulation Methods and Applications to Circuit Design
(SMACD), pp. 1–4 (2022)
11. Gugercin, S., Antoulas, A.C.: A survey of model reduction by balanced truncation and some
new results. Internat. J. Control 77(8), 748–766 (2004)
12. Zhou, K.: Frequency-weighted model reduction with L∞ error bounds. Syst. Contr. Lett. 21,
115–125 (1993)
13. Benner, P., Stykel, T. (2017). Model order reduction for differential-algebraic equations: a
survey
14. Rommes, J., Martins, N.: Efficient computation of transfer function dominant poles using
subspace acceleration. IEEE Trans. Power Syst. 21(3), 1218–1226 (2006)
 Evaluation of Rollover Stability of Liquid Tank
Truck in the Lane Change

Xuan Ngoc Nguyen1 , Van Nhu Tran2(B) , Van Tan Vu2 ,

and Tien Phuc Dang1
1 Industrial University of Ho Chi Minh City, Hochiminh City, Vietnam
2 University of Transport and Communications, Hanoi, Vietnam

vannhu.tran@utc.edu.vn

Abstract. The tank truck had stability very poor during movement due to affecting
of internal vibration of the liquid in tank. So to assess the rollover stability of a
circular tank truck in case of double lane change maneuver corresponding to each
liquid level in the tank, the research investigated the load transfer ratio (LTR)
value and the roll angle of suspension. The paper used the Lagrange method for
the quasi-static model of the circular tank and the D’Alembert’s principle for the
roll model of the vehicle. Then the system of differential equations for rollover
stability was built from the two models above. The simulation results showed that
when the vehicle moves at a speed of 100 km/h corresponding to the liquid level
height of 0.4, 0.8 m compared to the tank diameter, the vehicle was rolled causing
danger to people and property. Therefore, during the tank truck operation, the
driver should noted the vehicle movement speed and it is necessary to propose
speed limit warning signs along the road.

Keywords: Rollover stability · Tank truck · Lane change

1 Introduction
Rollover of vehicle is a very dangerous phenomenon in the process of moving on the
road. According to statistics [1], the rollover accident in 2021 decreased by 4% com-
pared to 2020 due to the covid epidemic, but still higher than in 2019. Today, tank trucks
have worse movement stability than other trucks due to the influence of the liquid inside
the tank. Therefore, scientists focus on research on the stability properties of tank trucks
to have better improvement method. Typically, the research [2] uses Newton’s law and
Pasimodo software in combination with experimental data to analyze the rollover stabil-
ity of the B-train liquid tank truck corresponding to the volume percentage of 40, 60, 80,
100% in the tank in the case of lane change maneuver, it can be determined in the speed
range from 30 to 40 km/h which the vehicle is stability. The authors in [3] using genetic
algorithms to propose the optimal shape of the elliptical tank compared with the existing
shapes have shown that the vibration frequency of the liquid in the tank is considered
the factor to impact on rolling of the vehicle. In addition, the article [4] analyzed the
motion of the liquid in an elliptical tank with Ansys software and used Newton’s law to

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 537–542, 2023.
https://doi.org/10.1007/978-3-031-22200-9_59
538 X. N. Nguyen et al.

build a system of equations of motion for the vehicle dynamics model, the oscillation
of the liquid in the tank is replaced by the motion of a simple pendulum, from the input
signal the steering angle shows the filling rate of 0.4–0.6 in tank which vehicles tend to
unstable. Besides, the article [5] simulates the vibration amplitude of kerosene in tanks
with and without partitions using VOF liquid volume model and Ansys software to show
that the explosion noise of kerosene decreases in container with partitions.
Based on the above scientific works, the study of liquid tank trucks is of great interest.
Therefore, in this study, the authors propose a multi-object model of a liquid circular
tank truck, especially fire trucks, to analyze the vehicle’s rollover stability in the case of
double lane change maneuver.

2 The Dynamic Model of the Liquid Tank Truck

The dynamic model of the circular liquid tank truck is analyzed into two parts: the
circular tank and the vehicle. The circular tank is simulated by the movement of a
trammel pendulum depending on the liquid level in the tank [6]. The vehicle is analyzed
based on the roll model [7]. All are shown in Fig. 1.

Fig. 1. The dynamic modelling of the liquid tank truck with circular tank.

Applying the D’Alembert’s principle to the dynamic model of the vehicle (Fig. 1a)
which is described as follows:
(mS + mn )Z̈S + 2CR (ZS − ZU ) + 2KR (ŻS − ŻU )
(1)
+mn α̇ 2 l cos α + mn α̈l sin α = 0
IS ψ̈S + (2dR2 CR + CS )(ψS − ψU ) + 2dR2 KR (ψ̇S − ψ̇U )
−mS ay hS cos ψS − mS ghS sin ψS
(2)
−mn g(ht sin ψS + l sin α) − mn gay (ht cos ψS − l cos α)
−mn α̇ 2 lht sin(ψS + α) − mn α̈l(l − ht cos(ψS + α)) = 0
 Evaluation of Rollover Stability of Liquid Tank Truck 539

mU Z̈U − 2CR (ZS − ZU ) − 2KR (ŻS − ŻU )

(3)
+CT (2ZU − q1 − q2 ) + KT (2ŻU − q̇1 − q̇2 ) = 0
IU ψ̈U − (2dS2 CR + CS )(ψS − ψU ) − 2dS2 KR (ψ̇S − ψ̇U )
+CT dT (2dT ψU − q1 + q2 ) + KT dT (2dU ψ̇U − q̇1 + q̇2 ) (4)
−mu ay hU cos ψU − mu ghU sin ψU = 0

According to Fig. 1b, the equation of motion of the liquid mass (5) is built based on
the quasi-static model [6] and the Lagrange method (Table 1).

mn lht ψ̈S +mn l 2 α̈ + mn lht α α̇ ψ̇S + mn l 2 α α̇ 2 + mn gl sin α + Cl α̇ = Pltn l (5)

where:

Pltn = mn ay (6)
⎛   ⎞
2 arccos d −2h
.d 2  2   
d d d − 2h
mn = 1000.L⎝ ⎠− −h . tan arccos
8 2 d
(7)
 
d 3 . arccos d −2h   
d d3 d −2h
d 8 − 12 . sin arccos d
l= − 


2   
2 d 2 . arccos d −2h
− d2 − h . tan arccos d −2h
d
4 d
 2     
d −2h
2 − h . tan arccos 3 + 6
. 2h
d d
d
+ 
d −2h

 2    (8)
d 2 . arccos d d −2h
4 − d
2 − h . tan arccos d

The ay lateral acceleration in this case is a input signal function, it represents the
double lane change maneuver as mentioned in the study [8] and is shown in Fig. 2.
To evaluate the rollover stability of the liquid tank truck, the research uses the load
transfer ratio (LTR) value and the suspension roll angle value. In which, the LTR value
reaches 1 (or −1) or the suspension roll angle value is within the range of 7–8 degrees,
the vehicle will be unstable [9].

3 Simulation Results
When the liquid tank truck changes lanes, the fluid inside the tank will oscillate depending
on the liquid level in the tank. The vehicle moves at different speeds, in addition to the
vibration of the liquid, it will make the vehicle tend on rollover.
The simulation results are performed in the case of the double lane change maneuver
with the speed of 90 and 100 km/h corresponding to each ratio of liquid level height h to
diameter d of the tank is 0% (no load), 25, 50, 75, 100% (full load) (equate to the fluid
levels are high 0, 0.4, 0.8, 1.2, 1.6 m in tank) is presented as follows:
540 X. N. Nguyen et al.

Table 1. The symbols and parameter values of liquid tank truck (Do not take into account the
liquid mass)

Symbols Description Values Unit

mS Sprung weight 2305 kg
IS The sprung weight moment of inertia 11174 kgm2
mU Unsprung weight 1073 kg
IU The unsprung weight moment of inertia 1311.1 kgm2
hS Distance from the sprung weight center to the roll center 0.61 m
hU Distance from the ground to the unsprung weight center 0.525 m
ht Distance from the tank center to the roll center 0.9 m
CR Stiffness of suspension 724704.5 N/m
KR Damping of suspension 42629 Ns/m
dR Distance from the roll center to suspension 0.62 m
dT Distance from the wheel track to unsprung weight center 1.025 m
CT Tyre cornering stiffness 1015000 N/m
KT Tyre roll stiffness 25000 Ns/m
CS Stiffness of anti-inclination bar 35800 N/m
L The liquid tank length 2 m
d The liquid tank diameter 1.6 m
Cl Viscosity of liquid 0.5
g Gravity acceleration 9.81 m/s2
h The liquid level height in tank – M
l Distance from the tank center to the liquid mass center – M
mn The liquid mass – kg
ΨS Sprung weight roll angle – rad
ΨU Unsprung weight roll angle – rad

Figure 3 shows the liquid tank truck moves in double lane change maneuver, the
LTR value and the travel of suspension are within the allowed threshold corresponding
to different liquid levels in tank with speed 90 km/h. In addition, the LTR value at the
fill level of 25% is the highest which it mean that the vehicle has a tendency to rollover
in this case.
From simulation results in Fig. 4, the load transfer ratio value at the fill fluid of
25, 50% is beyond −1, so the tank truck is rolled in this case, although the travel of
suspension value is still in limits.
 Evaluation of Rollover Stability of Liquid Tank Truck 541

Fig. 2. The dynamic modelling of the liquid tank truck with circular tank.

Fig. 3. The load transfer ratio (LTR) (a) and the roll angle of suspension (b) of liquid tank truck
with speed 90 km/h.

Fig. 4. The load transfer ratio (LTR) (a) and the roll angle of suspension of liquid tank truck (b)
with speed 100 km/h.

4 Conclusions

The paper modeled the motion of a liquid in a circular tank using a trammel pendulum.
Then applying the D’Alembert’s principle, quasi-static model and the Lagrange method
to a multi-object model of the tank truck with circular tank in case of double lane
542 X. N. Nguyen et al.

change maneuver. Simulation results show that at the speed of 90 km/h (with different
liquid levels) and 100 km/h (corresponding to the liquid level height of 0 m (no load),
1.2, 1.6 m), the vehicle is stable. Therefore, drivers control the liquid tank truck which
should pay attention to vehicle’s velocity when changing lanes or changing direction.

References
1. National Center for Statistics and Analysis: Early Estimates of Motor Vehicle Traffic Fatalities
and Fatality Rate by Sub-Categories in 2021. National Highway Traffic Safety Administration,
Report no. DOT HS 813 118 (2022)
2. Barbieri, N., de Melo, R.P., de Lima, K.F., de Sant’Anna Vitor Barbieri, G.: Dynamical analysis
of B-train vehicle combination with liquid cargo. Proc. Inst. Mech. Eng. Part D: J. Automob.
Eng. 233(14), 3871–3881 (2019)
3. Shojaeefard, M.H., Talebitooti, R., Yarmohammadi Satri, S., Amiryoon, M.H.: Investigation
on natural frequency of an optimized elliptical container using real-coded genetic algorithm.
Lat. Am. J. Solids Struct. 11(1), 113–129 (2014)
4. Zheng, X.-L., Zhang, H., Ren, Y.-Y., Wei, Z.-H., Song, X.-G.: Rollover stability analysis of
tank vehicles based on the solution of liquid sloshing in partially filled tank. Adv. Mech. Eng.
9(6), 1–26 (2017)
5. Singal, V., Bajaj, J., Awalgaonkar, N., Tibdewal, S.: CFD analysis of a kerosene fuel tank to
reduce liquid sloshing. Procedia Eng 69, 1365–1371 (2014)
6. Kolaei, A.: Dynamic liquid slosh in moving containers. PhD Dissertation, Concordia
University, Quebec, Canada (2014)
7. Vu, V.T.: Enhancing the roll stability of heavy vehicles by using an active anti-roll bar system.
PhD Dissertation, Grenoble Alpes University, France (2017)
8. Soltani, A., Bagheri, A., Azadi, S.: Integrated vehicle dynamics control using semi-active
suspension and active braking systems. Proc. Inst. Mech. Eng. Part K J. Multi-Body Dyn.
232(3), 314–329 (2017)
9. Vu, V.T., Sename, O., Dugard, L., Gaspar, P: Enhancing roll stability of heavy vehicle by LQR
active anti-roll bar control using electronic servo-valve hydraulic actuators. Veh. Syst. Dyn.
55(9), 1405–1429 (2017)
 Factors Affecting Saving Energy Intention
of Urban Residents in Hanoi

Tung Thanh Nguyen1,2,3 , Nhung Tang Cam3,4(B) , Kien Trung Duong4 ,

and Tuan Anh Do5
1 Faculty of Engineering Physics and Nanotechnology, VNU University of Engineering and
Technology, Hanoi, Vietnam
2 National University, Hanoi, Vietnam
3 Thai Nguyen University of Technology, Thai Nguyen, Vietnam
tangcamnhung@tnut.edu.vn
4 Faculty of Industrial and Energy Management, Electric Power University, Hanoi, Vietnam
kiendt@epu.edu.vn
5 Chau Industrial Technology Joint Stock Company, Hanoi, Vietnam

Abstract. Vietnam is a dense population country. As to Vietnam’s economic

growth in the past years, energy has played an important role. With rapid urban-
ization and the improvement in people’s standard of living, the demand for energy
will keep growing. Promoting saving and efficient use of energy prioritize of urban
residents by implementation of energy-saving behaviors is the key to build a sus-
tainable urban environment, and reduce carbon emissions. This study establishes
a theoretical model of the factors affecting the energy-saving behavior of urban
residents in Hanoi based on the Theory of Planned Behavior and research method-
ologies in economics and psychology. In this study, we used 300 questionnaires for
Hanoi urban residents survey. The influence mechanism of energy-saving behav-
ior of urban residents is clarified through a structural equation based on partial
least squares (PLS-SEM). The result shows that factors such as subjective norm,
perceived behavioral control, perceived usefulness of energy-saving technology
have a direct effect on the intention to use energy-saving. This study did not find
an impact relationship between attitude and intention to use energy saving. These
results play an important part in the build of energy saving policies in urban Hanoi.

Keywords: Energy-saving intention · Household · Structural equation

modeling · PLS-SEM

1 First Section
Over the past few decades, Vietnam has made considerable developments in all aspects,
especially in the energy field. To meet society’s demand, Vietnam has adapted rapidly
growing economy and deep integration into the world economy and has become a middle-
income country [5]. The Vietnamese economy is growing; expressed economic indicators
such as Gross Domestic Product have grown rapidly from 115.9 billion USD in 2010 to
about 271.2 billion USD. In 2020 [4]. The average population growth rate in 2016–2020

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 543–557, 2023.
https://doi.org/10.1007/978-3-031-22200-9_60
544 T. T. Nguyen et al.

is 1.13%, of which the urban population increases by 3.1% due to the migration flow
from rural to urban areas [4]. Although the economy is strongly affected by the Covid-
19 pandemic, the 5-year average growth rate in 2016–2020 reached about 5.9% and
belonged to the group of fast-growing countries in the region [1]. Viet Nam is at risk of
facing energy shortages due to the huge demand for fuel consumption, especially when
fuel prices are high and climate change is present; The phenomenon of climate change
is becoming increasingly apparent, leading to difficulties in forecasting and causing a
decrease in hydroelectricity sources as well as exploitation and use of primary energy
sources.
The process of urbanization is happening quickly and the population of urban areas
has increased rapidly in Vietnamese cities over the years, leading to changes in urban
energy consumption models. Therefore, with the rapid economic development and the
increasing urbanization rate, it is important to improve people’s understanding of the
energy-saving characteristics of urban residents, the factors affecting intention and bind-
ing relationships. In 2010, the Law on Economical and Efficient Use of Energy was
promulgated and a 10-year review of Resolution No. 18-NQ/TW of the Xth Politburo
Committee on the orientation of the National Energy Development Strategy. Resolu-
tion No. 55-NQ/TW dated February 11, 2020 of the Politburo Committee was approved
on the orientation of Vietnam’s national energy development strategy to 2030, with a
vision to 2045. Resolution No. 55-NQ/TW emphasizes that “Effective and efficient use
of energy and environmental protection must be considered an important national policy
and the responsibility of the whole society.” Strengthening energy audits; developing
synchronous mechanisms and policies, strong and feasible sanctions to encourage invest-
ment in and use of energy-saving and environment-friendly technologies and equipment,
contributing to boosting labor productivity and innovating the growth model” [6].
The consistency of the strategic vision and policies of the Party, State and Government
of Vietnam have been demonstrated through the “National Program on economical
and efficient use of energy for the period of 2019–2030” with the goal of promoting
economical and efficient use of energy through the synchronous implementation of state
management solutions, research and development, scientific and technological products,
and development cooperation in the international arena. At the same time, forming a
lifestyle and habits of economical and efficient use of energy in all activities of society;
energy saving is a mandatory criterion in key energy-using facilities, towards the goal of
green development and reduction of greenhouse gas emissions. Especially in the period
of 2019–2025, the specific goals to be achieved are as follows: Achieving energy savings
from 5.0 to 7.0% of the total national energy consumption; 70% of industrial zones and
50% of industrial clusters will apply solutions for economical and efficient use of energy;
building an energy efficient urban model [2].
Deriving from the practice, research works, legal bases in the 2010 Law on Eco-
nomical and Efficient Use of Energy, and orienting the national strategy on economical
and efficient use of energy in the period of 2019–2030, the study on the intention to use
energy saving of the urban population is a topical study, which is very relevant to prac-
tical needs. The implementation of behaviour interventions to help promote economical
and efficient use of energy, promote energy conservation, develop sustainable energy in
urban areas, contribute to development goals sustainable energy of Vietnam.
 Factors Affecting Saving Energy Intention … 545

2 Research Hypotheses
2.1 Theoretical Basis
The Theory of Planned Behavior (TPB) is a modified model of the theory of rational
action (TRA), and it is widely used to explore predictions about actual behavior through
intention when people have the ability to control their actions. According to TPB, as can
be seen in Fig. 1, the intention to act will be the only and reliable psychological determi-
nant of actual behavior; Furthermore, attitudes toward behavior, subjective norms, and
perceived behavioral control are said to be pre-existing intentions.

Fig. 1. Planned behavior model. Source Ajzen [9]

Ajzen defined attitude as an individual’s feelings for performing a particular behavior;

Subjective norms such as perceived social pressure to act or not act, and perceived
behavioral control refer to difficulty performing the behavior. Ajzen argues that TPB
models the interaction between behavior and cognition and allows for irrational and
unconscious thinking through their influence on beliefs. Furthermore, the difficulty in
predicting intention from behavior and the fact that behavior is influenced by changing
circumstances such as the time from intention formation to action [8]. The following
three factors influence a person’s behavioral intentions:

1. Attitudes towards behavior - individual judgments are favorable or unfavorable when

performing the behavior.
2. Subjective norm - personal perception of social pressure to perform or not to perform
the behavior.
3. Perceived behavioral control - the ease or difficulty of engaging in the behavior.

2.2 Proposed Research Model and Hypotheses

Research on the intention to save energy has been interested in many scholars around
the world [11, 16, 19, 22, 25, 27]. This study combines two theoretical models TPB and
TAM model to examine the influence of factors on the intention to use energy saving of
people in urban areas, in the context of Vietnam. In the digital transformation process,
the acceptance or investment of new technology products can be one of the barriers
preventing people from realizing their energy-saving intentions. The combination theory
model is depicted in Fig. 2.
546 T. T. Nguyen et al.

– Using the theory of planned behavior (TPB) to study the influence of individual
psychological factors on the intention to use energy saving in the household.
– Using technology acceptance theory (TAM) to predict the intention of households in
accepting the use of energy-saving technologies.

Fig. 2. Proposed research model.

2.2.1 Perceived Behavioral Control (PBC)

Perceived behavioral control is more expression in the perception of external factors that
affect behavior.
Knussen et al. analyzed the results of 252 surveys in Scotland and showed that per-
ceived behavioral control is an important factor influencing the “accession” of household
waste recycled because of finding that the lack of household waste recycling facilities,
they are not conducive to the development of recycling behavior and insignificant behav-
ior motives. On the contrary, when objective factors are favorable for the behaviorist to
grasp external resources, the behavior motivation will be more powerful [20].
The definition of energy-saving perceived behavioral control is the degree of
difficulty that individuals perceive to implement the behavior of energy-saving use
H1: Perceived behavioral control has a positive effect on the intention to use energy
saving.

2.2.2 Attitude
Attitude mentions a subjective assessment of a resident’s specific behavior toward
energy-efficient use, which results from a combination of behavior beliefs and out-
come evaluation views [9]. Behavior beliefs determine whether an individual engages
in a particular activity, and outcome judgments are positive or negative assessments of
 Factors Affecting Saving Energy Intention … 547

the consequences of that individual’s behavior. Therefore, the more positive attitude
towards energy-saving appliances, the stronger the willingness to chase energy-saving
appliances.
H2: Attitude has a positive impact on the intention to use energy saving.

2.2.3 Subjective Norm

For the subjective norm, the theory of planned behavior and the results of a number of
literature studies show that the subjective norm has a significant impact on energy-saving
behavior and actual behavior through the behavior intention. The factors affecting the
subjective criteria of the subject of behavior mainly come from the perceived social
norms and behavior of the reference group. This study determines that the subjective
indicator of energy-saving behavior is the performance of the role and energy-saving
effect of the reference group in the psychological perception of the behavior subject.
H3: Subjective norm positively affect on intention to use energy saving.

2.2.4 Realize the Usefulness and Ease of Use of Energy-Saving Technology

Davis proposed a Technology Acceptance Model (TAM). This model explains and pre-
dicts an individual’s acceptance of new technologies [14]. TAM has two main variables
perceived usefulness and perceived ease of use, which influence consumers’ motivation
and perception to applicate new technology. Perceived usefulness refers to the extent
to which individuals perceive any technology as being able to benefit them. Perceived
ease of use refers to the extent to which an individual believes it is easy to use new
technology. Some studies have shown that it is necessary to examine the two factors
perceived usefulness and perceived ease of use from many different aspects [7, 18].
Furthermore, scholars also suggest that including relevant variables in a technology
acceptance model may increase its explanatory power in a particular context [12, 13].
In order to understand the influence mechanism of the technology acceptance model
in the context of people’s energy-saving intentions, this study proposes the following
hypotheses:
H4: Perceived ease of use of energy-saving technology positively affect energy-
saving attitudes.
H5: Perceived ease of use of energy-efficient technology positively affect on
perceived usefulness of energy-saving technology.
H6: Perceived usefulness of energy-saving technology positively affects the energy-
saving attitude.
H7: Perceived usefulness of energy-saving technology positively affect using energy-
saving intention.
548 T. T. Nguyen et al.

3 Study Method
3.1 Questionnaire Survey

The questionnaire for each factor of the model is referenced from the study of Wang
et al. [27]. The questions have also been revised after discussions with experts on energy-
saving research and interviews with 20 test households. This study uses a five-point Likert
scale: (1) Strongly Disagree, (2) Disagree, (3) Neutral, (4) Agree, and (5) Strongly Agree.

3.2 Data Collection

In this paper, we use a questionnaire to survey the reality of the intention to use energy
saving of urban residents in Hanoi. With 300 questionnaires randomly distributed in
districts of Hanoi city. Data used in the paper served from January 2019.
The questionnaire consists of two main sections: Part A factors affecting energy-
saving behavior, Part B general information of respondents such as gender, age,
educational attainment, occupation, income…

3.3 Analytical Methods

This paper uses SEM structural equation model to analyze the relationship between
factors such as attitude (TD), subjective norm (CCQ), perceived behavioral control
(KS), perceived ease of use of energy-saving technology (DSD), perceived usefulness of
energy-saving technology (HI), energy-saving intention (YD). Use the Smart PLS-SEM
analysis support tool to perform data analysis. As shown in Fig. 1, the arrows show the
direction of influence of the factors.

4 Result and Discussions

See (Table 1).

4.1 Assessing the Quality of the Observed Variable

Hair et al. suggested that outer loading must be greater than or equal to 0.7 observed
variable that is quality [17] (Table 2).

4.2 Reliability Assessment

Cronbach’s Alpha method and the total variable correlation coefficient are used to eval-
uate the reliability of the scale. The analysis results show that all the analyzed variables
in the model have Cronbach’s Alpha coefficient greater than 0.6 and the total correlation
coefficient is greater than 0.3, so the scale of the model has internal consistency [3],
specifically in Table 3.
 Factors Affecting Saving Energy Intention … 549

Table 1. The statistical analysis-describing

No Classification Quantity Percent (%)

I. Age
1 Less than 25 125 41
2 26–35 60 19,6
3 36–45 45 14,75
4 45 above 75 24,65
Total 305 100
II. Gender
1 Male 141 46,2
2 Female 164 53,8
Total 305 100
III. Educational attainment
1 High school 22 7,2
2 College 19 6,2
3 Graduate 208 68,2
4 Post graduate 46 18,4
Total 305 100
IV. Monthly Income (millionVNÐ)
1 Less than 5 million 20 6,6
2 5–10 50 16,4
3 10–15 98 32,1
4 15 above 137 44,9
Total 1037 100

4.3 Convergence Test

To evaluate the convergence value of the scales, analysis on PLS-SEM software was
performed. With an index of factor loading greater than 0.5 and mean explanatory vari-
ance (AVE) greater than 0.5. The results show that all factor loading coefficients of the
factors are greater than 0.5 and AVE is greater than 0.5 as shown in Table 4.

4.4 Discriminant Validity

To evaluate the discriminant of the factors, this study uses the Fornell-Larcker criteria
to evaluate and compare. When the square root AVE of the factors is larger than the
corresponding correlation coefficient, the factors are discriminant. The specific results
are in Table 5.
550 T. T. Nguyen et al.

Table 2. Quality assessment of observed variables.

SN PEU PU PB BA BI
SN1 0.820
SN2 0.849
SN3 0.882
PEU1 0.876
PEU2 0.897
HI1 0.753
HI2 0.839
HI3 0.744
PB1 0.810
PB2 0.785
PB3 0.811
BA1 0.758
BA2 0.868
BA3 0.804
BI1 0.788
BI2 0.854
BI3 0.885
SN: Subjective norm; BI: Behavior Intention; PB: Perceived behavioral control; BA: Behavior
attitude; PEU: Perceived Ease of Use; PU: Perceived Usefulness

4.5 SEM Model Analysis Result

– Multicollinearity Testing

For each dependent variable, there will be VIF (Variance inflation factor) for each inde-
pendent variable respectively. The multicollinearity test results show that the model does
not exist or multicollinearity does not affect the analysis results. With the VIF values of
the independent variables all less than 5, the research concepts have no manifestation of
multicollinearity (Table 6).

– Hypothesis Testing.

The analysis results from Smart PLS-SEM show that the P-value of the relationship
between attitude and intention by 0.258 is larger than the theoretical level of 0.05.
Therefore, hypothesis H2 is not accepted. The rest accept hypotheses H1, H3, H4, H5,
H6, H7 (Fig. 3).
 Factors Affecting Saving Energy Intention … 551

Table 3. Evaluating the reliability of the scale.

Num. Observed Coefficient of Total variable Coefficient of Assessment

Variables variable Cronbach’s correlation Cronbach’s
alpha coefficient alpha without
variable
1. Group of Energy saving attitude variable - TD
n=3 TD1 0,673 0,442 0,668 Usable
TD2 0,610 0,432 consistency
scale
TD3 0,439 0,635
2. Group of Subjective norm variable - CCQ
n=3 CCQ1 0,807 0,622 0,771 Good scale
CCQ2 0,637 0,761
CCQ3 0,716 0,672
3. Group of Perceived behavioral control variable - KS
n=3 KS1 0,724 0,559 0,623 Usable
KS2 0,579 0,598 consistency
scale
KS3 0,506 0,682
4. Group of energy saving intention variable - YD
n=3 YD1 0,742 0,457 0,781 Usable
YD2 0,612 0,602 consistency
scale
YD3 0,644 0,570
5. Group of perceived usefulness of energy saving technology variable - HI
n=3 HI1 0,673 0,433 0,653 Usable
HI2 0,577 0,458 consistency
scale
HI3 0,455 0,616
6. Group of perceived ease of use of energy saving technology varia - DSD
n=2 DSD1 0,726 0,573 Temporarily
DSD2 0,573 acceptable
consistency
scale

SEM model result provided the values in Table 7 below.

This study analyzed the factors affecting the intention to use energy-saving of urban
residents in Hanoi. The factors that directly affect the intention to use energy-saving
include, CCQ, HI, and KS with impact coefficient β decreasing respectively: 0.347;
0.322; 0.160. It can be seen that the perceived ease of use of energy-saving technology
has the strongest and most positive impact on the perception of the usefulness of energy-
saving technology with the coefficient β = 0.443; next is the perceived usefulness of
energy-saving technology that has an impact on the attitude factor with β = 0.369.
552 T. T. Nguyen et al.

Table 4. Testing for convergence of factors

No Factors Label Outer loading Average Variance

Coefficient Extracted (AVE)
1 Attitude TD1 0,658 0,611
TD2 0,868
TD3 0,804
2 Subjective norm CCQ1 0,820 0,724
CCQ2 0,849
CCQ3 0,882
3 Perceived behavioral control KS1 0,810 0,643
KS2 0,785
KS3 0,811
4 Energy saving intention YD1 0,688 0,662
YD2 0,854
YD3 0,885
5 Perceived usefulness of HI1 0,753 0,608
energy saving technology HI2 0,839
HI3 0,744
6 Perceived ease of use of DSD1 0,876 0,786
energy saving technology DSD2 0,897

Table 5. Discriminant validity.

SN PEU PU PB BA BI
SN 0.851
PEU 0.447 0.887
PU 0.591 0.443 0.780
PB 0.573 0.428 0.515 0.802
BA 0.530 0.407 0.477 0.577 0.782
BI 0.663 0.388 0.640 0.561 0.494 0.814

However, this result shows that the attitude factor has no positive effect on the intention
to use energy saving.

(1) According to the analysis results in Table 7, the subjective norm is the most impor-
tant factor affecting the intention to save energy as well as the factor that directly
 Factors Affecting Saving Energy Intention … 553

Table 6. Multicollinearity.

SN PEU PU PB BA BI
SN 0.851
PEU 0.447 0.887
PU 0.591 0.443 0.780
PB 0.573 0.428 0.515 0.802
BA 0.530 0.407 0.477 0.577 0.782
BI 0.663 0.388 0.640 0.561 0.494 0.814
SN PEU PU PB BA BI
SN 1.898
PEU 1.000 1.244
PU 1.244 1.694
PB 1.828
BA 1.683
BI

Table 7. The SEM analysis result.

Original Sample Standard T statistics P-values Hypothesis

sample (β) mean (M) deviation (|β/STDEV|) evaluation
(STDEV)
SN -> BI 0.347 0.350 0.065 5.378 0.000 Accepted
hypothesis
PEU -> 0.443 0.445 0.054 8.152 0.000 Accepted
PU hypothesis
PEU -> 0.244 0.244 0.052 4.657 0.000 Accepted
BA hypothesis
PU -> 0.369 0.370 0.049 7.566 0.000 Accepted
BA hypothesis
PU -> BI 0.322 0.320 0.054 5.938 0.000 Accepted
hypothesis
PB -> BI 0.160 0.161 0.066 2.442 0.015 Accepted
hypothesis
BA -> BI 0.064 0.064 0.056 1.132 0.258 Nonaccepted
the hypothesis
P -value < 0,05: Accepted hypothesis

affects the intention to use energy saving of people. In Hanoi city (b = 0.347; P-
value = 0.000). It shows that the certainty induced by social norms significantly
554 T. T. Nguyen et al.

Fig. 3. Results of official model analysis of factors affecting the intention to use energy saving
of urban residents in Hanoi.

influences residential energy-saving intentions (consistent with TPB theory as the

norms are applicable and directional). When individuals live in an environment
where many individuals are aware of energy-saving, this individual will be greatly
affected by that environment. This is also consistent with some studies that suggest
that subjective norms are the motivating factors for the implementation of energy-
saving and long-term use behaviors, and explain the level of readiness when using
energy. Implement energy-saving behavior [10, 15]. The difference can be affected
the culture and lifestyle of Vietnamese who are easily influenced by social groups
and especially oral culture. Influencers for people were surveyed, such as: friends
and family members have a direct impact on the intention and behavior of the respon-
dents. Therefore, it will be easier to promote the intention to use energy-saving by
strengthening the subjective standards of energy-saving in Hanoi city. That is, to
strengthen the energy-saving propaganda to create an atmosphere where the whole
society realizes energy saving. Saving energy without reducing people’s quality
of life is an important challenge for policymakers. The government should pay
attention to how-to guide and promote energy-saving habits to form energy-saving
intentions in life.
(2) Perceived behavioral control has a direct influence on the intention to use energy
saving of people in urban areas with the coefficient (b = 0.160; P-value = 0.015).
This relationship seems to represent the weakest interaction in the model. This result
is consistent with related studies Wang et al., Lopes et al., and Ko et al. [21, 23, 26].
Perceived behavioral control is believed to be influenced by external factors. In this
 Factors Affecting Saving Energy Intention … 555

study, two external aspects can be considered: cost and time. Hanoi is the capital
of Vietnam, the economic center of Vietnam, so all living costs are expensive,
according to statistics from in-depth interviews, we find that when it comes to
saving energy more than 60% of respondents are interested in the first thing that
is cost (investment cost of equipment, payback period, benefits from savings, etc.).
Besides, time is also a big barrier to the implementation of energy-saving. With the
fast and fast pace of life in Hanoi, almost people pay little attention to taking time
to implement energy-saving measures, along with time to learn and be aware of the
problems of energy conservation. Energy saving. All of which leads to a limitation
in the energy-saving intentions of residents. Therefore, policy implications should
take this into account in order to formulate and implement suitable strategies.
(3) Energy-saving attitude: Although in some studies TD is thought to be an important
factor in predicting intention, the analysis results from the data of this study show
that TD does not have a positive effect on the intention to use energy saving.
(4) Perceived ease of use of energy-saving technology

This is the first study to investigate the impact of DSD and HI on the saving intention
of urban residents in Vietnam. The results of this study indicate that both of these
factors are important in the adoption and use of energy-saving technologies in urban
areas. DSD factor has an indirect impact on the intention to use energy saving through
perceived usefulness of energy-saving technology with impact coefficient β = 0.443.
This result is consistent with the results of some studies such as Liang and Yi-Hsuan
(2009), which argue that DSD is the most important factor in explaining why people
adopt energy-saving technologies. [28]. Venkatesh et al. also mentioned that DSD is a
strong factor for “intent to use” in technology adoption [24]. Similarly, Han points out
that in determining usage behavior, the importance of DSD should be emphasized. As
such, many studies and demonstrations have consistently demonstrated that perceived
ease of use of energy-saving technology is a strong predictor of “intent to use” in the
application of science and technology.

(5) Perceiving the usefulness of energy-saving technology

This paper has presented a combination of technology acceptance model and planned
behavior theory model.
The perceived usefulness of energy-saving technology is considered to have a direct
and positive impact on the intention to save energy with the value β = 0.322, which
explains that when people in urban areas perceived usefulness of energy-saving technol-
ogy increased by 01 point in the 5-point Likert scale, the intention to use energy-saving
increased by 0.322 points. The fact that Hanoi people are living in the period of scientific
and technological innovation, being aware of the effects of energy-saving technologies
will help them bring many benefits, thereby influencing their intention to accept and use
these technologies. This result is quite similar to the study of Davis [14].
556 T. T. Nguyen et al.

5 Conclusion
The article has shown the relationship between the factors affecting the intention to use
energy of urban people in Hanoi city. Based on the combination of two models, TPB
and TAM, to develop an analytical model that combines survey questionnaires for many
different energy users in Hanoi. The data analysis tool PLS-SEM is used to evaluate the
convergence value of the scales and the results show that the perceived ease of use of
energy saving technology has the greatest influence on the concept of energy efficiency.
Intended to use energy saving. Meanwhile, the subjective norm is also one of the factors
that promote the behavior of economical energy use due to cultural factors, especially
the mutual influence in the same social sphere, so if the policy makers to strengthen
propaganda will create certain values. As for the three factors Perceived behavioral
control, Attitude to save energy, Perceived usefulness of energy saving technology has
not really affected the intention to save energy due to some barriers. Such as: rapid socio-
economic development makes the pace of life and concern about issues closely related
to each individual’s life make the consideration of economical use of energy no longer
important. Weak, or helping the majority of people realize the usefulness of technology
when realizing the benefits, the intention to accept the technology will be voluntarily
accepted by the people.

Acknowledgments. This research is funded by Thai Nguyen University of Technology (TNUT).

References
1. Report No. 154/BC-TCTK (2020), General Statistics Office
2. National program on economical and efficient use of energy for the period of 2019–2030.
Chief author, Ministry of Industry and Trade (2018)
3. Hoang Trong, Chu Nguyen Mong Ngoc (2008), Analyze research data with SPSS. Vol. 2,
University of Economics Ho Chi Minh City, Ða.i ho.c Kinh tế TP Hồ Chí Minh, Hong Duc
Hanoi Publishing House
4. Statistical Yearbook: General Statistics Office. https://www.gso.gov.vn/du-lieu-va-so-lieu-
thong-ke/2021/07/nien-giam-thong-ke-2021/(2020). Accessed 25 June 2021
5. Overview Viet Nam: World Bank. https://www.worldbank.org/en/country/vietnam/overview
(2020). Accessed 26 Mar 2020
6. Economical and efficient use of energy towards the sustainable development of Viet-
nam’s energy industry: Web Portal Commission for The Management of State Capital at
Enterprises. http://cmsc.gov.vn/xem-chi-tiet/-/asset_publisher/WqvULR6gmpvh/Content/
su-dung-nang-luong-tiet-kiem-va-hieu-qua-huong-en-su-phat-trien-ben-vung-nganh-nang-
luong-viet-nam?1640959 (2021). Accessed 20 June 2021
7. Acheampong, P., et al.: Hybridizing an extended technology readiness index with technology
acceptance model (TAM) to predict e-payment adoption in Ghana. Am. J. Multidiscip. Res.
5(2) (2017)
ij
8. Ajzen, I.: Attitudes, personality, and behavior Berkshire, chu biên. Open University Press,
England (2005)
9. Ajzen, I.: The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50(2), 179–211
(1991)
 Factors Affecting Saving Energy Intention … 557

10. Black, J.S., Stern, P.C., Elworth, J.T.: Personal and contextual influences on househould
energy adaptations. J. Appl. Psychol. 70(1), 3 (1985)
11. Cai, S., et al.: Determinants of intention and behavior of low carbon commuting through
bicycle-sharing in China. J. Clean. Prod. 212, 602–609 (2019)
12. Chang, C.-T., Hajiyev, J., Su, C.-R.: Examining the students’ behavioral intention to use e-
learning in Azerbaijan? The general extended technology acceptance model for e-learning
approach. Comput. Educ. 111, 128–143 (2017)
13. Chin, J., Lin, S.-C.: A behavioral model of managerial perspectives regarding technology
acceptance in building energy management systems. Sustainability. 8(7), 641 (2016)
14. Davis, F.D.: Perceived usefulness, perceived ease of use, and user acceptance of information
technology. MIS Q. 319–340 (1989)
15. Fishbein, M., Ajzen, I.: Belief, attitude, intention, and behavior: An introduction to theory
and research. Philos. Rhetor. 10(2) (1977)
16. Ha, H.-Y., Janda, S.: Predicting consumer intentions to purchase energy-efficient products. J.
Consum. Mark. (2012)
17. Hair Jr., J.F., Hult, G.T.M., Ringle, C., Sarstedt, M.: A Primer on Partial Least Squares
Structural Equation Modeling (PLS-SEM). London (2016)
18. Hamari, J., Sjöklint, M., Ukkonen, A.: The sharing economy: why people participate in
collaborative consumption. J. Am. Soc. Inf. Sci. 67(9), 2047–2059 (2016)
19. He, X., Zhan, W.: How to activate moral norm to adopt electric vehicles in China? An empirical
study based on extended norm activation theory. J. Clean. Prod. 172, 3546–3556 (2018)
20. Knussen, C., et al.: An analysis of intentions to recycle household waste: the roles of past
behaviour, perceived habit, and perceived lack of facilities. J. Environ. Psychol. 24(2), 237–
246 (2004)
21. Ko, S.B., Jin, B.: Predictors of purchase intention toward green apparel products: a cross-
cultural investigation in the USA and China. J. Fash. Mark. Manag.: Int. J. 21(1), 70–87
(2017)
22. Li, D., et al.: Understanding energy-saving behaviors in the American workplace: a unified
theory of motivation, opportunity, and ability. Energy Res. Soc. Sci. 51, 198–209 (2019)
23. Lopes, J.R.N., et al.: A new model for assessing industrial worker behavior regarding
energy saving considering the theory of planned behavior, norm activation model and human
reliability Resour., Conserv. Recycl. 145, 268–278 (2019)
24. Venkatesh, V., et al.: User acceptance of information technology: Toward a unified view. MIS
Q 425–478 (2003)
25. Wang, B., et al.: Analysis of factors influencing residents’ habitual energy-saving behaviour
based on NAM and TPB models: Egoism or altruism? Energy Policy 116, 68–77 (2018)
26. Wang, S., Fan, J., Zhao, D., Yang, S., Fu, Y.: Predicting consumers’ intention to adopt
hybrid electric vehicles: using an extended version of the theory of planned behavior model.
Transportation 43(1), 123–143 (2014). https://doi.org/10.1007/s11116-014-9567-9
27. Wang, Z., Zhang, B., Li, G.: Determinants of energy-saving behavioral intention among
residents in Beijing: Extending the theory of planned behavior. J. Renew. Sustain. Energy.
6(5), 053–127 (2014)
28. Liang, T.-P., Yeh, Y.-H.: Situational Effects on the Usage Intention of Mobile Games, pp. 51–
59. Springer, Workshop on E-Business (2008)
 H∞ Optimal Tracking Control for Robot
Manipulators with Input Constraint
and Disturbances

N. D. Dien1(B) , N. T. Luy2(B) , and L. K. Lai3(B)

1 Faculty of Electrical Engineering, University of Economics—Technology for Industries, Ho
Chi Minh City, Vietnam
nddien@uneti.edu.vn
2 Faculty of Electronic Technology, Industrial University of Ho Chi Minh City, Ho Chi Minh
City, Vietnam
nguyentanluy@iuh.edu.vn
3 Faculty of Electrical Engineering, Thai Nguyen University of Technology, Thai Nguyen,
Vietnam
laikhaclai@tnut.edu.vn

Abstract. This paper introduces an H∞ optimal tracking controller for robot

manipulators with saturation control torque, and disturbances based on a reinforce-
ment learning (RL) method. The robot manipulators dynamics are transformed into
a strict-feedback nonlinear system with input constraint, and disturbances. The
feedforward control inputs are designed to transform a position tracking control
problem into an H∞ optimal control problem. The constrained Hamilton–Jacobi–
Isaac (HJI) equation is built, which is solved by the online RL algorithm using
only a single neural network (NN) to reduce computational cost. Then the optimal
control law with the input constraint and the worst disturbance law are determined.
The concurrent learning (CL) technique is used to relax the demand for the persis-
tence of excitation (PE) condition when updating the critic NN weights to global
optimal values. The control performance is given in the comparative simulation
results, showing the effectiveness of the proposed method.

Keywords: Robot manipulators · Reinforcement learning · Optimal control ·

Input constraints · Disturbances

1 Introduction

Robot manipulators have been used with great efficiency in the fields of manufacturing,
healthcare, and services. An important problem of the robot manipulators control system
is the control tracking the reference position [1, 2]. With the actuators of joints in robot
manipulators, the input that drives the motors of the joints is always torque-bounded.
S. Ling et al. [3] used an adaptive fuzzy controller and C. Yang et al. [4] used an RBF
NN controller to control the robot manipulators with input saturation. Q. Zhou et al. [5]
designed an adaptive NN controller and applied control to the robot manipulators with

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 558–572, 2023.
https://doi.org/10.1007/978-3-031-22200-9_61
 H∞ Optimal Tracking Control for Robot Manipulators … 559

dead zones. During the working process, the robot manipulators are affected by external
disturbances, nonlinear friction, and unexpected changes in model parameters (due to
equipment wear after a long working time). R.-D. Xi et al. [6] designed a robust adaptive
sliding controller and applied control to the robot manipulators with disturbances and
parameter uncertainties. Van and Ge [7] used an adaptive fuzzy sliding controller to
control the robot manipulators with disturbance observers. The above articles did not
minimize any long-term performance function and hence were not optimal.
Mathematically, the H∞ optimal control problem is solved when solving the solution
of the HJI equation. The nonlinear HJI equation is difficult to solve. RL is a useful method
used to approximate online solutions of the HJI equation [8, 9]. K. G. Vamvoudakis et al.
[8] used the control scheme with 3 NN to solve the H∞ optimal control problem. A. P.
Valadbeigi et al. [10] solved the H∞ optimal control problem in discrete-time. RL was
used to control the H∞ optimal tracking for robot manipulators [11]. In [11], Modares
et al. used an offline policy algorithm, which leads to a slow learning rate. L. Kong et al.
[12] used RL to control the H∞ optimal tracking for robot manipulators, but they didn’t
consider the problem of saturation control torque. Zhao et al. [13] have used RL for the
problem of saturated optimal tracking control for robot manipulators in discrete time. R.
Song et al. [14] have proposed an H∞ optimal controller for the robot manipulators with
constrained state and input. [15–18] have designed an H∞ optimal tracking controller
using RL for nonlinear system [15] and successfully controlled for a mobile robot [17],
a PMSM system [18] with input constraints and disturbances. To our best knowledge,
there is currently no studies for the H∞ optimal tracking problem for robot manipulators
with saturated inputs and disturbances in continuous time.
The main contributions of this article are summarized as follows. We propose a
design for feedforward control inputs to transform a position tracking control problem
into an H∞ optimal control problem for robot manipulators with saturated inputs and
disturbances. The constrained HJI equation is built, which is solved by the online RL
algorithm using only a single NN to reduce computational cost. The CL technique is
used to relax the demand for the PE condition.
The structure of the article is organized as follows. The content of Sect. 2 presents the
dynamics of the robot manipulator and the design problem of feedforward control inputs.
Section 3 presents a design of the H∞ optimal control law with the input constraint and
disturbances. In Sect. 4, the simulation results are presented, and Sect. 5 presents the
conclusion of the article.

2 Robot Manipulators and Feedforward Control

2.1 Robot Manipulators Nonlinear
The dynamics of the robot manipulators with saturation control torque and disturbances
are as follows [1]

M (q)q̈ + C(q, q̇)q̇ + G(q) + F(q̇) = τ − τ0 (1)

where q ∈ n×1 is the vector of joint angles, q̇ ∈ n×1 is the velocity vector, q̈ ∈ n×1
is the acceleration vector, M (q) ∈ n×n is the inertia matrices, C(q, q̇) ∈ n×n is the
560 N. D. Dien et al.

Coriolis-centripetal matrices, G(q) ∈ n is the dynamic friction, F(q̇) ∈ n is the static

friction, τ ∈ n×1 is the control input vector of joints torque, τ0 ∈ n×1 is the vector of
external disturbances.
Dynamics (1) has some properties [1]: M (q) is a symmetric matrix, positive definite,
and is bounded m1 I ≤ M (q) ≤ m2 I (m1 , m2 are positive constants). The matrix C(q, q̇)
is bounded. The matrix G(q) is bounded. Ṁ (q)−2C(q, q̇) is the skew-symmetric matrix.
τ0 is bounded.
We transform dynamics (1) into a strict-feedback nonlinear system with saturation
torques, and disturbances, as follows:

q̇ = fq (q) + gq (q)v
(2)
v̇ = fv (q, v) + gv (q, v)τ + kv (q, v)τ0

where fq (q) = 0nx1 , gq (q) = In , fv (q, v) = −M −1 (Cv + G + F) ∈ n×1 , gv (q, v) =

M −1 ∈ n×n , kv (q, v) = −M −1 ∈ n×n , and τ = [τ1 , τ2 , ..., τn ]T are the control
inputs, limited by τ  ≤ λ.

2.2 Feedforward Control

We will design the feedforward control inputs by applying the backstepping method in
this section. Then, we will transform (2) into an H∞ optimal control problem [19].
First, we define vd = vd∗ + vda and τ = τ ∗ + τa , where vd is the virtual control input
vector, vd∗ is the virtual H∞ optimal control input vector, vda is the feedforward virtual
control input vector, τa is the feedforward control input vector, τ ∗ is the H∞ optimal
control input vector. Then we take some steps to convert (2) to the form of tracking error
dynamics.
Step 1: Define eq = q − qd , we have

ėq = −q̇d + gq (q)vd + gq (q)(v − vd ) = fq (eq ) + gg (q)vd + gq (q)ev (3)

The feedforward virtual control input vector vda is designed as follows

 
vda = gq−1 (q) 1 ev + q̇d + fq (eq ) (4)

By defining f q (eq , ev ) = fq (eq ) + 1 ev , and substituting (4) into (3), one obtains

ėq = fq (eq ) + gg (q)vd∗ + 1 ev + gq (q)ev

(5)
= f q (eq , ev ) + gg (q)vd∗ + gq (q)ev

Step 2: By employing the sliding model surface, and defining ev = q̇ − vd , we have

ėv = −v̇d + fv (q, v) + gv (q, v)τ + kv (q, v)τ0 (6)

The feedforward control input vector τa is designed as

 
τa = gv−1 (q, v) fv (eq , ev ) + v̇d − fv (q, v) − gqT (q)eq − 2 ev (7)
 H∞ Optimal Tracking Control for Robot Manipulators … 561

where τa is designed with the following constraint [18]

τa  ≤ λ − ρ tanh(1) (8)

If τ ∗ is designed with the following constraint τ ∗  ≤ ρ, and combined with (8)

then the control inputs will be constrained τ  ≤ λ.
By defining f v (eq , ev ) = fv (eq , ev ) − 2 ev , and substituting (7) into (6), one obtains

ėv = f v (eq , ev ) + gv (q, v)τ ∗ − gqT (q)eq + kv (q, v)τ0 (9)

Lemma 1: Consider the tracking dynamics of the robot manipulators with saturation
control torque, and disturbances as follows

ż = f qv + gqv u∗ + kqv d (10)

T T
with z = [eqT , evT ]T ∈ 2n×1 , f qv = [f q (eq , ev ), f v (eq , ev )]T ∈ 2n×2n , u∗ =
 
[vd∗T , τ ∗T ] ∈ 2n×1 , ua = [vda
T , τ T ]T ∈ 2n×1 , g = diag g (q), g (q, v) ∈ 2n×2n ,
  a qv q v
kqv = diag kq (q), kv (q, v) ∈ 2n×2n , d = [01×n , τ0T ]T ∈ 2n×1 .
If u∗ is designed to stabilize the system (10), then the position tracking control
problem for (2) is equivalent to an H∞ optimal tracking control problem for (10).

Proof: For (2), choose a Lyapunov function as

1 T 1
V1 = eq eq + eTv ev (11)
2 2
Taking derivative (11) along with (5) and (9), we have

V̇1 = eqT f q (eq ) + eqT gq (q)vd∗ + eqT gq (q)ev + evT f v (eq , ev )

+evT gv (q, v)τ ∗ − evT gqT (q)eq + evT kv (q, v)τ0


(12)
= eqT f q (eq ) + gq (q)vd∗ + evT f v (eq , ev ) + gv (q, v)τ ∗ + kv (q, v)τ0

= z T f qv + gqv u∗ + kqv d

For dynamics (10), choose a Lyapunov function as

1 T
V2 = z z (13)
2
Taking derivative (13) along with (10) one obtains
1 T

V̇2 = z f qv + gqv u∗ + kqv d (14)

2
Comparing (12) and (14), it can be seen that if V̇1 < 0 implies V̇2 < 0. In the other
words, the stabilities of closed-loop of dynamics (2) and closed-loop dynamics (10) are
equivalent.
The Proof is completed.
562 N. D. Dien et al.

3 H∞ Optimal Tracking Control

We design the H∞ optimal tracking controller with input constraint and disturbances in
this section. First, a cost function is defined. Then the constrained HJI equation is built
and solved by the online RL algorithm using only one NN. Then the H∞ optimal control
law with the input constraint and the worst disturbance law are determined.
The cost function for system (10) is defined as
 ∞  ∞

J (z, u, d ) = r(z, u, d )dt = z T Qz + U (u) − ξ 2 d T d dt (15)

t t

with Q is a positive definite matrix and U (u) is non-negative, ξ > ξ ∗ > 0 is the
disturbance attenuation, where ξ ∗ is the minimum value of ξ for system (10) to ensure
stability.
We choose the energy cost function [20] as follows
 u
U (u) = 2ρ tanh−T (s/ρ)Rds (16)
0

with ρ is the saturating bound of u, R is a positive definite diagonal matrix, and define R
is a column vector of the diagonal elements of R. Using integral by parts, (16) become
   u

−T s
u
U (u) = 2ρ tanh Rs − 2ρR sd tanh−T (s/ρ) (17)
ρ 0 0

We have
 u

U (u)2 = −2ρR sd tanh−T (s/ρ) (18)

0

Define v = tanh−T (s/ρ), (18) become

 tanh−T (u/s)
U (u)2 = −2ρ 2 R tanh(v)dv (19)
0

Define  = tanh(v), 1 = [1, 1, · · · , 1]T , we have

d d
d  = (1 − tanh2 (v))dv → dv = = (20)
1 − tanh2 (v) 1 − 2
and (19) become
  u/ρ
u/ρ  −1/2
U (u)2 = −2ρ R̄ 2
d  = −2ρ R̄
2
d (1 − 2 )
0 1̄ − 2
0 1̄ − 2
  2  (21)
u
= ρ R̄ ln 1̄ −
2
ρ
 H∞ Optimal Tracking Control for Robot Manipulators … 563

Then (17) becomes

 
u u2
U (u) = 2ρ tanh−T Ru + ρ 2 R ln 1 − 2 (22)
ρ ρ
Define the Hamilton function for system (10) with a cost function (15):

H (z, u, d , Jz ) = r(z, u, d ) + JzT f qv + gqv u + kqv d (23)

where Jz = ∂J /∂z
Using zero-sum game theory [8], the optimal cost function is determined:

J ∗ (z, u, d ) = min max(J (z, u, d )) (24)

u d

Using the stationary condition of (18), the saturation optimal control law is
determined:

∗ 1 −1 T ∗
u = −ρ tanh R gqv Jz (25)
2ρ
and the worst disturbance law is determined:
1 T ∗
d∗ = k J (26)
2ζ 2 qv z
Substituting (25), and (26) into (23), we get the HJI equation:
   

H ∗ z, u∗ , d ∗ , Jz∗ = r z, u∗ , d ∗ + Jz∗T f qv + gqv u∗ + kqv d ∗ = 0 (27)

The constrained HJI equation is defined as (27). We solve this equation using the
online RL algorithm using only a single NN to reduce the computational cost. Ve∗ is
approximated as follows

J ∗ (z) = W T θ (z) + ε(z) (28)

with W ∈ N is the NN weight vector, θ (z) : n → N is the NN activation function

vector with θ (0) = 0, N is the number of neurons in the hidden layer, ε(z) is the NN
approximation error.

Assumptions 1 [18]: One can choose θ (z) such that θ(z) ≤ sθ , θz  =
∂θ (z)/∂z ≤ s∇θ , ε(z) ≤ sε , εz  = ∂ε(z)/∂z ≤ s∇ε , where sθ , s∇θ , sε ,
s∇ε are positive constants.

Taking derivative (28) we have

Jz∗ = W T θz + εz (29)

The Hamiltonian Eq. (27) becomes

 

H ∗ z, u∗ , d ∗ , W T θz = r z, u∗ , d ∗ + W T θz f qv + gqv u∗ + kqv d ∗ − εH = 0 (30)

564 N. D. Dien et al.

where εH = −εz f qv + gqv u∗ + kqv d ∗

The weight vector (28) is unknown, so the cost function is approximated by

Ĵ (z) = Ŵ T θ (z) (31)

where Ĵ (z) is the approximate cost function, Ŵ ∈ N is the approximate NN weight

vector. The approximate control law for (25) becomes

1 −1 T T
û = −ρ tanh R gqv θz Ŵ (32)
2ρ
and the approximate worst disturbance law for (26) becomes
1 T T
d̂ = k θ Ŵ (33)
2ζ 2 qv z
Define e1 is the error caused by the approximation of the function. Use (31), (32),
(33) for the Hamiltonian Eq. (27), we have

Ĥ z, û, d̂ , Ŵ T θz = r z, û, d̂ + Ŵ T θz f qv + gqv û + kqv d̂ = e1 (34)

Define W̃ = W − Ŵ is the approximate error NN weight vector, from (30) and (34),
we get

e1 = −W̃ T θz f qv + gqv û + kqv d̂ + εH (35)

For Ŵ → W , we need to adjust Ŵ to minimize the squared error E = 21 e1T e1 . Using the
algorithm Normalized gradient descent and the CL technique [8, 20], the update law Ŵ

is defined:

If z T f qv + gqv û + kqv d̂ < 0, then

˙ =−α
Ŵ σ T Ŵ + zT Qz + U (û) − ξ 2 d̂ T d̂
1
(σ T σ
+ 1) 2


P
σ (ti )

−α1 σ (t i ) T
Ŵ + z(t i ) T
Qz(t i ) + U (û(t i ) − ξ 2 T
d̂ (t i ) d̂ (t i )
(σ (ti )T σ (ti ) + 1)2
i=1
(36)

and if z T f qv + gqv û + kqv d̂ ≥ 0, then

˙ =Ŵ
Ŵ ˙ + α 1 θ (G − K)z (37)
2 z
2

where σ = θz f qv + gqv û + kqv d̂ , G = gqv R−1 gqv

T , K = 1 k k T , α > 0, α > 0
ζ 2 qv qv 1 2
In the update law, the CL technique is used to relax the demand for the PE condition
when updating the critic NN weights to global optimal values. σ (ti ), and r(ti ) = zT Qz +
 H∞ Optimal Tracking Control for Robot Manipulators … 565

U (û)−ξ 2 d̂ T d̂ are recorded and stored in {σ (ti )}Pi=1 , {r(ti )}Pi=1 , the number of P selected
is P ≥ N [20]. Note that {σ (ti )}Pi=1 must be linearly independent, which means its rank
must satisfy the condition rank(σ (t1 ), σ (t2 ), ..., σ (tP )) = N .
Theorem 1: Consider the dynamic (10), with the equation HJI (27), the critic NN
(31), the saturated optimal control law (32), the worst disturbance law (33), and the
tuning law for the NN (36), (37). Then, the tracking error and the approximation error
are UUB. Furthermore, the control law (32) and the worst disturbance law (33) converge
to the near-optimal value.
Proof: The proof is presented in the appendix.

4 Simulations
We perform simulations with a two-link robot manipulator [11]. The dynamics of the
two-link robot
 manipulator is given as  (1), with.  
h1 + 2h3 c2 h2 + h3 c2 −h3 s2 q̇2 −h3 s2 (q̇1 + q̇2 )
M (q) = , C(q, q̇) =
h2 + h3 c2 h2 h3 s2 q̇1 0
 T
G(q) = 8.45 tanh(q̇1 ), 2.35 tanh(q̇2 )

where c2 = cos (q2 ), s2 = sin(q2 ), h1 = 3.473kgm2 , h2 = 0.196kgm2 , h3 =

0.242kgm2 ,τ is limited τ  ≤ 5 kgm. The disturbance is assumed to be in random
value τ0 = [−0.1, 0.1].

Fig. 1. The weight values of the NN in the case of using only PE

 T
The desired trajectory qd is given as qd = qd 1 , qd 2 = [sin(0.1t), sin(0.1t)]T , the
initial value is chosen
 q(0) = [0.5, −0.5]T , q̇(0) = [0, 0]T . The activation function  is
chosen θ (z) = eq1 , eq1 eq2 , eq1 ev1 , eq1 ev2 , eq2 , eq2 ev1 , eq2 ev2 , ev1 , ev1 ev2 , ev2 . Select
2 2 2 2

α1 = 50, α2 = 0.01, R = 1, Q = I ∈ R4×4 , λ = 5, ρ = 0.6, ζ = 1.5.

First, we give simulation results of the convergence of NN weights using only PE
and using CL. The case of using only PE (with PE use time off 0 ≤ t ≤ 80s) has the
result as shown in Fig. 1, it shows that the weights get stuck in a local minimum because
σ becomes 0 after oscillation. The use case of CL (with PE use time off 0 ≤ t ≤ 0.04s)
566 N. D. Dien et al.

Fig. 2. The weight values of the NN in the case of using CL with 0 ≤ t ≤ 0.04s

has the result as shown in Fig. 2, it shows that the NN weights converge early. Figures 1
and 2 have shown the efficiency of using CL to update the weights NN.
The tracking position and tracking error of joints are illustrated as shown in Figs. 3, 4,
and 5. The results show that the controller which we designed has good tracking position
performance; the tracking error is 2.10−4 after learning. Figures 6, 7, and 8 show that
the feedforward control and the saturated optimal feedback control inputs satisfy the
condition τa  ≤ 5 − 0.6 tanh(1), u∗  ≤ 0.6, and the control inputs are constrained
τ  ≤ 5. To show the efficiency of the saturated H∞ optimal tracking control (SROTC),
we perform a comparative simulation with the PD gravity controller (PDGC). Figures 9,
and 10 show that the SROTC is smooth, unaffected by external disturbances, and has a
smaller error tracking the position than the PDGC, where the external disturbances are
given as follows: τ0 has random value in ±0.1 at t = 50s, and τd = 0 at t = 60s.

Fig. 3. Compare the position of the first joint with the reference position

5 Conclusion
This paper proposed an H∞ optimal tracking controller for robot manipulators with
saturation control torque and disturbances. A new design of the feedforward control
inputs were designed. The optimal value function is approximated based on the online
RL algorithm using only a single NN. The CL technique was used to relax the PE
 H∞ Optimal Tracking Control for Robot Manipulators … 567

Fig. 4. Compare the position of the second joint with the reference position

Fig. 5. The error tracking the position

Fig. 6. The saturated optimal feedback control inputs

condition when updating the NN weights. The tracking error and the approximation
error are UUB, the NN weights, the control law, and the disturbance law converge to a
near-optimal value.
568 N. D. Dien et al.

Fig. 7. The feedforward control inputs

Fig. 8. The control inputs

Fig. 9. Compare the error tracking the position of the first joint position
 H∞ Optimal Tracking Control for Robot Manipulators … 569

Fig. 10. Compare the inputs torque of the first joint position

Appendix
PROOF OF THE THEOREM 1.
The Lyapunov function is chosen for (10) as follows
1 1

V3 = V31 + V32 = α2 z T z + trace W̃ T W̃ (38)

2 2
We take the derivative of the function V31 V31

V̇31 = α2 z T ż = α2 z T f qv + gqv û + kqv d̂ (39)

If α2 z T f qv + gqv û + kqv d̂ < 0, then it is easy to see that ∃λ0 > 0 such that

α2 z T f qv + gqv û + kqv d̂ < −α2 λ0 z, (43) becomes

V̇31 < −α2 λ0 z (40)

Taking derivative (11), one obtains


P

V̇32 = −α1 Y σ T W̃ − εH − α1 Y (ti ) σ T (ti )W̃ − εH (ti ) (41)

i=1

σσT
with Y = (σ T σ +1)2
, and (41) becomes
 

P
V̇32 = −α1 W̃ ψ W̃ + α1 W̃
T T
Y εH + Y (ti )εH (ti ) (42)
i=1
P
where ψ = Y σ T + i=1 Y (ti )σ
T (t
i ), ψ > 0. Using Young’s inequality, one obtains
 2 α 2
 
V̇32 ≤ −(α1 − 1)λmin (ψ)W̃  + 1 (P + 1)εHm
2
(43)
4
where εHm is the upper bound of εH
570 N. D. Dien et al.

From (40) and (43), we have

 2
 
V̇3 ≤ −α2 λ0 z − β1 W̃  + β2 (44)

α2
where β1 = (α1 − 1)λmin (ψ), β2 = 41 (P + 1)εHm 2 . We choose α > 1 to β > 0.
1 1
V̇3 ≤ 0, if and only if

z ≥ αβ2 λ2 0 = sz
  
  β (45)
W̃  ≥ β21 = sW̃

In case z T f qv + gqv û + kqv d̂ ≥ 0, (41) becomes

1
V̇31 ≤ −α2 λ0 f qv − α2 z T (G + K)θzT Ŵ (46)
2
With the first terms (46), α2 z T f qv is bounded by

α2 z T f qv < −β3 z2 (47)

where β3 = α2 sf , sf is the upper bound of f qv . Substituting (37) into V̇32 of (39),

qv qv
and combining with (43), we have
 2
  1
J̇3 ≤ −β3 z2 − β1 W̃  + β2 − α2 z T (G + K)θzT W
2
  2 (48)
β4 2   β2
→ V̇3 ≤ −β3 z − − β1 W̃  + β2 + 42
2β3 4β3

where β4 = 21 α2 (sG + sK )sθ sW .

V̇3 ≤ 0, if and only if
⎧ 
⎪
⎪ 4β32 β2 +β42 β4
⎨ z ≥ 4β33
+ 2β = sz
   2
3
(49)
⎪
⎪   4β3 β2 +β42
⎩ W̃  ≥ = sW̃
4β 2 β
3 1

 
 
From (45) and (49), we can be seen that when z or/and W̃  are large, then
 
 
the updated law Ŵ (36) and (37) will make V̇3 ≤ 0, and then z or/and W̃  will be
small. According to the extended Lyapunov theorem, the tracking
 error and approximate
 
weight error of NN are UUB. Notice that z ≥ sz and W̃  ≥ sW̃ , where sz =
   
max sz , sz and
  sW̃ = max sW̃ , sW̃ .From (25), (26), (28), (31), (32), (33), we have
 ∗    
 ∗ 
J (z) − Ĵ (z) ≤ sW̃ sθ + sε = sv , u − û ≤ 2ρ 1
λmin (R)sgqv sW̃ s∇θ + s∇ε = su ,
   
 ∗ 
d − d̂  ≤ 2ζ1 2 skqv sW̃ s∇θ + s∇ε = sd and with sv ≥ 0, su ≥ 0, sd ≥ 0.. We see
 H∞ Optimal Tracking Control for Robot Manipulators … 571

that the error of the approximation of the cost function, the error of the approximation
of the optimal control law, and the error of the approximation of the disturbance law are
bounded by positive constants. Good convergence quality can be achieved by choosing
the appropriate α1 , α2 factor.
The Proof is completed.

References
1. Lewis, F.L., Dawson, D.M., Abdallah, C.T.: Robot Manipulator Control Theory and Practice.
CRC Press (2003)
2. Zeghloul, S., Laribi, M.A., Gazeau, J.P.: Robotics and Mechatronics, presented at the Pro-
ceedings of the 4th IFToMM International Symposium on Robotics and Mechatronics
(2015)
3. Ling, S., Wang, H., Liu, P.X.: Adaptive fuzzy dynamic surface control of flexible-joint robot
systems with input saturation. IEEE/CAA J. Autom. Sin. 6(1), 97–107 (2019)
4. Yang, C., Huang, D., He, W., Cheng, L.: Neural control of robot manipulators with trajectory
tracking constraints and input saturation. IEEE Trans. Neural Netw. Learn. Syst. 32(9), 4231–
4242 (2020)
5. Zhou, Q., Zhao, S., Li, H., Lu, R., Wu, C., Systems, L.: Adaptive neural network tracking
control for robotic manipulators with dead zone. IEEE Trans. Neural Networks 30(12), 3611–
3620 (2018)
6. Xi, R.-D., Xiao, X., Ma, T.-N., Yang, Z.-X.: Adaptive sliding mode disturbance observer based
robust control for robot manipulators towards assembly assistance. IEEE Robot. Autom. Lett.
7(3), 6139–6146 (2022)
7. Van, M., Ge, S.S.: Adaptive fuzzy integral sliding-mode control for robust fault-tolerant
control of robot manipulators with disturbance observer. IEEE Trans. Fuzzy Syst. 29(5),
1284–1296 (2020)
8. Vamvoudakis, K.G., Lewis, F.L., Control, N.: Online solution of nonlinear two-player zero-
sum games using synchronous policy iteration. Int. J. Robust 22(13), 1460–1483 (2012)
9. Van Der Schaft, A.J.: L 2-gain analysis of nonlinear systems and nonlinear state feedback
H∞ control. IEEE Trans. Autom. Control 37(6), 770–784 (1992)
10. Valadbeigi, A.P., Sedigh, A.K., Lewis, F.L.: H_ {\infty} static output-feedback control design
for discrete-time systems using reinforcement learning. IEEE Trans. Neural Netw. Learn.
Sys-Tems 31(2), 396–406 (2019)
11. Modares, H., Lewis, F.L., Jiang, Z.P.: H infinity tracking control of completely unknown
continuous-time systems via off-policy reinforcement learning. IEEE Trans. Neural Netw.
Learn. Syst. 26(10), 2550–2562 (2015). https://doi.org/10.1109/TNNLS.2015.2441749
12. Kong, L., He, W., Yang, C., Sun, C.: Robust neurooptimal control for a robot via adaptive
dynamic programming. IEEE Trans. Neural Netw. Learn. Sys-Tems 32(6), 2584–2594 (2020)
13. Zhao, X., Tao, B., Qian, L., Ding, H.: Model-based actor-critic learning for optimal tracking
control of robots with input saturation. IEEE Trans. Industr. Electron. 68(6), 5046–5056
(2021). https://doi.org/10.1109/tie.2020.2992003
14. Song, R., Liu, L. Xia, L., Lewis, F.L.: Online optimal event-triggered H∞ control for nonlinear
systems with constrained state and input. IEEE Trans. Syst., Man, Cybern.: Syst. (2022)
15. Xue, S. Luo, B. Liu, D., Yang, Y.: Constrained event-triggered H∞ control based on adaptive
dynamic programming with concurrent learning. IEEE Trans. Syst., Man, Cybern.: Syst.
(2020)
16. Tan, L.N.: Distributed H∞ optimal tracking control for strict-feedback nonlinear large-scale
systems with disturbances and saturating actuators. IEEE Trans. Syst., Man, Cybernet-Ics:
Syst. 50(11), 4719–4731 (2018)
572 N. D. Dien et al.

17. Nguyen Tan, L.: Distributed optimal control for nonholonomic systems with input constraints
and uncertain interconnections. Nonlinear Dyn. 93(2), 801–817 (2018). https://doi.org/10.
1007/s11071-018-4228-8
18. Tan, L.N., Pham, T.C.: Optimal tracking control for PMSM with partially unknown dynamics,
saturation voltages, torque, and voltage disturbances. IEEE Trans. Industr. Electron. 69(4),
3481–3491 (2021)
19. Zargarzadeh, H., Dierks, T., Jagannathan, S.: Adaptive neural network-based optimal control
of nonlinear continuous-time systems in strict-feedback form. Int. J. Adapt. Control Signal
Process. 28(3–5), 305–324 (2014)
20. Vamvoudakis, K.G., Miranda, M.F., Hespanha, J.P.: Asymptotically stable adaptive-optimal
control algorithm with saturating actuators and relaxed persistence of excitation. IEEE Trans.
Neural Netw. Learn. Syst. 27(11), 2386–2398 (2016)
 Hybrid Energy Storage Systems: A Brief
Overview

Nicola Campagna1 , Vincenzo Castiglia1 , Rosario Miceli1 , Fabio Viola1 ,

Alessandro Busacca1 , and Minh T. Nguyen2(B)
1 Department of Engineering, University of Palermo, Viale Delle Scienze, Parco d’Orleans,
90128 Palermo, Italy
2 Department of Electrical Engineering, Thai Nguyen University of Technology, Thai
Nguyen 240000, Vietnam
nguyentuanminh@tnut.edu.vn

Abstract. In this paper, a brief overview on the Hybrid Energy Storage Systems
(HESSs) is provided. In literature, different architectures are chosen to realize the
HESSs, and they are based on the principal aim of the HESSs employment. In
this paper, the most used HESS topologies are presented, with particular attention
to the active, passive and semiactive topologies, highlighting their characteris-
tics. To have a complete schematic idea of the HESSs application, a focus on
the principal sizing methodologies is provided, distinguishing the conventional
approaches and the advanced ones, exploiting their main applications. Together
with a proper sizing, a correct power-sharing strategy is one of the HESSs key
points. For this reason, several control strategies are described, focusing on the
energy management control and on the underlying control.

Keywords: Hybrid energy storage system · Sizing · Control

1 Introduction
The transition to a low-carbon and green economy includes the goals of a 40% reduc-
tion in greenhouse gas emissions, 32% of consumption provided by Renewable Energy
Sources (RES) and a 32.5% improvement in energy efficiency [1, 2]. In order to achieve
these objectives, the development of power generation systems from non-programmable
renewable sources, such as eolic and photo-voltaic (PV), will be of fundamental impor-
tance. Thanks to incentive policies and technological developments, the price of renew-
able technologies has been significantly reduced in recent years [3, 4], with an 80%
reduction for PV panels between 2009 and 2015, and a 30–40% reduction for wind
turbines in the same period. The typical generation prole of the PV panel repeats daily,
with a peak around noon and minimum values during the darkest hours. In addition,
cloudiness can cause sudden changes in power generation within a few minutes, posing
a challenge for balancing the power grid. In the case of wind-based generation, fluc-
tuations in the produced energy usually affect longer periods, typically days or weeks.
However, these fluctuations must be balanced by the power grid. In this scenario, energy
storage systems can be a turning point. The use of such systems can lead to:

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 573–579, 2023.
https://doi.org/10.1007/978-3-031-22200-9_62
574 N. Campagna et al.

• Improved network management

• Speed of reaction in case of rapid power loss or power surges
• Interfacing of different grid elements for cost-effective balancing of RES over different
time windows.

In automotive and grid connected applications, the Energy Storage Systems (ESSs)
usually experiences irregular and frequent discharging/charging pattern which truncates
the ESS lifespan; therefore, the replacement cost of the ESS increases significantly. High
energy density storage technologies such as batteries and fuel cells have limited power
capability. On the other hand, high power density technologies such as supercapacitors or
flywheels have limited energy storage capability. The drawback of each technology can
be overcome with the so-called Hybrid Energy Storage Systems (HESSs). Depending
on the purpose of the hybridization, different energy storages can be used as a HESS.
Generally, the HESS consists of high-power storage (HPS) and high-energy storage
(HES) where the HPS absorbs or delivers the transient and peak power while the HES
meets the long-term energy demand. HESSs provide many benefits: improving the total
system efficiency, reducing the system cost, and prolonging the lifespan of the ESS. Due
to the various types of energy storage technologies with different characteristics, a wide
range of energy storage hybridization can be realized.
Figure 1 shows an example HESS that is composed of batteries (high specific energy
storage) and supercapacitors (high specific power storage), and three possible power
flow management strategies. Figure 1 (a) shows the case of high-power demand from
the load. Both energy storage sources supply power to the load. Figure 1 (b) shows the
case of low power demand. The battery supplies power both to the load (continuous
arrow) and the supercapacitors (dashed arrow). Figure 1 (c) shows the case of negative
power. The peak power is absorbed by the supercapacitors, the remaining power can be
supplied to the batteries.

2 HESS Topologies

Based on the nature and on the dynamics of the power sources composing the HESSs,
different architectural structures can be realizable. The choice of the proper one prin-
cipally depends on the HESS purpose of use and on the control strategies. Principally,
depending on the connection of the different sources to the system, three main classes
can be defined: passive, semi-active and active topologies.

2.1 Passive Topology

The passive HESSs interface the different storage systems directly, without using addi-
tional converters. The terminal voltage of the sources is not regulated and the power flow
management is addressed only by the source’s internal resistance and voltage-current
characteristic.
The main advantages are related to the ease of implementation and the cost effec-
tiveness, while the main disadvantage is related to the limited power split management
[5]. A structure of a passive HESS is shown in Fig. 2 (a).
 Hybrid Energy Storage Systems: A Brief Overview 575

Fig. 1. Hybrid energy storage system power flow in case of (a) high power demand, (b) low power
demand, (c) negative power demand.

2.2 Semi-active Topology

The semi-active topology provides for the employment of a DC/DC converter able to
control one of the sources, as depicted in Fig. 2 (b). In this case, even if the power flow
management is improved, extra space is needed and, if compared to the passive ones,
costs are higher. In most cases, the sources to be controlled by the DC/DC converter
are chosen based on their dynamics [6]. In particular, the high energy density sources
are preferred, in order to facilitate the power delivery during the fast transients and the
reverse power flow through the high-power density sources connected directly to the
load.

Fig. 2. (a) Structure of a passive HESS. (b) Structure of a semiactive HESS.

2.3 Active Topology

In this topology, each energy source is connected to the system through a power converter.
In particular, two main configurations are available in literature: parallel and cascade, on
the base of the chosen control method as reported respectively in Fig. 3 (a) and (b). At
the expense of a lower efficiency and higher cost, the active HESSs allow for an accurate
power flow management [7]. In Fig. 4, a synoptic diagram is shown with the aim of
comparing the three different presented topologies.
576 N. Campagna et al.

Fig. 3. Structure of an active HESS: (a) parallel, (b) cascade.

Fig. 4. Comparison between HESSs topologies.

3 Sizing Methodologies
Literature shows different methodologies for the proper sizing of the energy sources
composing the HESSs. Different approaches are presented in literature, they can be
divided into two main classes: conventional and advanced methods.

3.1 Conventional Methods

Among the conventional methods, the analytic approach is one of the most used, it is
based on analyzing a series of operational configurations by varying the elements of the
system to be optimized, such as cost, volume and weight of the energy sources, cost and
size of the power converters and systems stability [8]. Besides, a probabilistic method
can be used, it is preferred for the Renewable HESS for which the uncertainty of the
renewable source must be taken into the account [9]. For this reason, they are mainly
used for the grid-connected applications, and they are based on the simultaneous analysis
of the load conditions and the energy generation probabilistic distribution.
 Hybrid Energy Storage Systems: A Brief Overview 577

3.2 Advanced Methods

The advanced methods are mainly based on the resolution of non-linear problems. They
allow for an accurate sizing at the expense of a high computational effort. One of the
most employed approaches is the Particle Swarm Optimization Method [10]. It is based
on a multi-objective optimization analysis for which the energy sources sizing is carried
out through the definition of cost function to be minimized. They are mostly employed
for the automotive application, for which different aspects such as the storage weight
and life-time are relevant. Other advanced methods are based on Pinch Analysis and
they are used mostly for the smart-grid applications.

4 Control Strategies

One of the main issues of a HESS is to design and implement a suitable control system.
The choice of the control method depends on several factors: the reason for using the
HESS (life extension, power quality, cost minimization, volume, weight), the speed of
response and the hybridization architecture. To achieve safe, stable and efficient opera-
tion, HESSs need a power-split (or power-sharing) strategy. Generally, the management
and control system of a HESS can be divided into two parts: energy management control
and underlying control.

4.1 Energy Management Control

The goal of the energy management control is the power allocation among the different
sources composing the HESS. Depending on the scope of the power-split strategy, differ-
ent approaches are proposed in literature. They can be divided into two main categories:
classical and intelligent controls. For both the categories, the output of the employed
algorithm is the reference power of the HESS. The latter is than employed as the input
of the underlying control to effectively act on the power converters. Among the classical
controls, the filtering-based one is one of the most used. This control method is based
on applying a filter (e.g., lowpass, moving average) to the power profile required by
the load [11]. In particular, the control acts in order to split the power profile into as
many components as there are the HESS sources. The low-frequency components are
than imposed as a reference to the low-dynamic high energy density sources (e.g., fuel
cells, batteries), while the high-frequency components are given as a reference to the
high-dynamic high power density sources (e.g., SCs).
In order to take into the account, the system parameters (e.g., SoC, maximum deliv-
ered current, temperature), a rule-based control can be used [12]. It is based on the
definition of a set of constraints related to the HESS source parameters (e.g., SoC,
maximum current, temperature) and a set of deterministic rules, then the power to be
delivered by each source is obtained by the simultaneous check of the employed rules.
The classical controls are characterized by their ease of implementation, but most of
them are designed to be performed online. In order to have a real-time control, advanced
methodologies can be employed. Among them, the Sliding Mode Control [13] and the
Model Predictive Control are the most used [14].
578 N. Campagna et al.

4.2 Underlying Control

The reference power, given as an output of the energy management control, has to
be effectively employed to act on the power converters. Among the different control
methods, the most used are the PI control [15], the non-linear control and the fuzzy logic
control.

5 Conclusions
In terms of power and energy, a single nature ESS cannot satisfy all an application’s
needs. Additionally, most applications require both quick reaction times and lengthy
discharge times, allowing for the provision of energy for periods ranging from minutes
to days. HESSs, which combine two or more ESS technologies, are a desirable option
for satisfying these needs since they even allow for a decrease in the system’s total cost
and lifespan. This paper analyzed the feasibility of the HESSs. It focused the different
topologies mainly employed in the literature and it also provided a brief overview of the
capacity sizing of the energy sources composing the HESSs. Different control strategies
have been described. As an added value, this paper helps to have a brief schematic idea
about the principal challenges in the field of the storage systems.

Acknowledgements. The authors would like to thank Thai Nguyen University of Technol-
ogy, Viet Nam for the support. This work was financially supported by PON R&I 2015–
2020 “Propulsione e Sistemi Ibridi per velivoli ad ala ssa e rotante-PROSIB”, CUP no:
B66C18000290005, by H2020-ECSEL-2017-1-IA-two-stage, by “rst and european sic eightinches
pilot line-REACTION”, by Prin 2017—Settore/Ambito di intervento: PE7 linea CAdvanced pow-
ertrains and—systems for full electric aircrafts, by PON R&I 2014-2020-AIM (Attraction and
International Mobility), project AIM1851228-1.

References
1. European Commission: 2030 Climate & Energy Framework (2021)
2. World Energy Council: Five Steps to Energy Storage (2020)
3. International Renewable Energy Agency: The Power to Change: Solar and Wind Cost
Reduction Potential to 2025 (2016)
4. Sprake, D., Vagapov, Y., Lupin, S., Anuchin, A.: Housing Estate Energy Storage Feasibility
for a 2050 Scenario (2017)
5. Zheng, J., Jow, T., Ding, M.: Hybrid power sources for pulsed current applications. IEEE
Trans. Aerosp. Electron. Syst. 37(1), 288–292 (2001)
6. Cao, J., Emadi, A.: A new battery/ultracapacitor hybrid energy storage system for electric,
hybrid, and plug-in hybrid electric vehicles. IEEE Trans. Power Electron. 27(1), 122–132
(2012)
7. Alkafaji, A.S., Al-Samawi, A.A., Trabelsi, H.: Hybrid energy storage review for renewable
energy system technologies and applications. In: 2021 18th International Multi-Conference
on Systems, Signals and Devices (SSD), pp. 1059–1067 (2021)
8. Liu, Y., Du, W., Xiao, L., Wang, H., Bu, S., Cao, J.: Sizing a hybrid energy storage system for
maintaining power balance of an isolated system with high penetration of wind generation.
IEEE Trans. Power Syst. 31(4), 3267–3275 (2016)
 Hybrid Energy Storage Systems: A Brief Overview 579

9. Abbassi, A., Dami, M.A., Jemli, M.: A statistical approach for hybrid energy storage system
sizing based on capacity distributions in an autonomous pv/wind power generation system.
Renew. Energy 103, 81–93 (2017)
10. Zhou, T., Sun, W.: Optimization of batterysupercapacitor hybrid energy storage station in
wind/solar generation system. IEEE Trans. Sustain. Energy 5(2), 408–415 (2014)
11. Kollimalla, S.K., Mishra, M.K., Narasamma, N.L.: Design and analysis of novel control
strategy for battery and supercapacitor storage system. IEEE Trans. Sustain. Energy 5(4),
1137–1144 (2014)
12. Chong, L.W., Wong, Y.W., Rajkumar, R.K., Isa, D.: Modelling and simulation of standalone
PV systems with battery-supercapacitor hybrid energy storage system for a rural household.
Energy Procedia, vol 107, pp 232–236 (2017), 3rd International Conference on Energy and
Environment Research, ICEER 2016, pp 7–11. Barcelona, Spain (2016)
13. Tan, S.C., Lai, Y.M., Tse, C.K.: General design issues of sliding-mode controllers in dc-dc
converters. IEEE Trans. Industr. Electron. 55(3), 1160–1174 (2008)
14. Hredzak, B., Agelidis, V.G., Jang, M.: A model predictive control system for a hybrid battery-
ultracapacitor power source. IEEE Trans. Power Electron. 29(3), 1469–1479 (2014)
15. Campagna, N., Castiglia, V., Damiano, A., Di Noia, L.P., Miceli, R., Di Tommaso, A.O.: A
hybrid energy storage sizing for a vertical take-off and landing electric aircraft. In: IECON
2021 47th Annual Conference of the IEEE Industrial Electronics Society, pp. 1–6 (2021)
 Improving the Surface Characteristics of Mg
AZ31B Alloy by RF Magnetron Sputtering

Anup Malik(B) , Harlal Singh Mali, and Arvind Kumar Singh

Department of Mechanical Engineering, Malaviya National Institute of Technology Jaipur,

Jaipur, Rajasthan 302017, India
anupmalik321@gmail.com, harlal.singh@gmail.com,
arvindsingh5710@gmail.com

Abstract. The abstract should summarize the contents of the paper in short terms,
i.e. 150–250 words. Magnesium (Mg) and its alloy continue developing in popu-
larity since they are the most lightweight structural metallic materials. Mg alloys
have several features that make them ideal for use as biodegradable prostheses in
clinical practice. Despite having excellent properties, the greater application of
Mg alloys is limited by various natural constraints, including a high degradation
rate and hence loss of mechanical strength. In addition to basic alloying, surface
treatment and functionality are prominent ways to improve resistance to corrosion
for Mg alloys. This research belongs to the improvement in corrosion resistance
of AZ31B Mg alloy by coating Zinc oxide (ZnO) and Silver (Ag) with the help
of an RF magnetron sputtering system. A comparison of the degradation rate is
made with and without coating. The degradation testing immersion test is done
in Hank’s Solution, a simulated body fluid (SBF) type. The biodegradable behav-
ior of AZ31B in SBF was examined for different immersing time intervals. The
corrosion rate was found to be reduced when coated with Zinc oxide (ZnO) and
Silver (Ag) with the help of RF magnetron sputtering. The coating also reduced
the surface roughness of samples.

Keywords: AZ31B Mg alloy · Surface coating · RF magnetron sputtering ·

Biodegradable implant materials

1 Introduction

In recent years, magnesium alloys have gotten a lot of interest, especially in the biomed-
ical industry, because of their desired biodegradability, mechanical characteristics, and
biocompatibility [1]. Magnesium (Mg) and its alloys, showing low density, good biocom-
patibility, and specific strength, are favorable in lightweight metals for transportation,
aerospace, hydrogen storage, and load-bearing biomedical implant [2]. Mg alloys used
as a biomedical implant material have Young’s modulus and density that are equiva-
lent to natural bone, reducing the stress shielding effect [3]. In the Mg implant material
such as bone screw, subsequent surgery is usually avoided because of the biodegradable
behavior of Mg alloys. One of the biggest disadvantages of Mg alloys is the corrosion

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 580–588, 2023.
https://doi.org/10.1007/978-3-031-22200-9_63
 Improving the Surface Characteristics of Mg AZ31B Alloy … 581

resistance is very low, and the material degrades quickly [4]. Mg alloys are degraded in
the presence of chloride ion (Cl− ) solutions present in the body. As a result, additional
surgeries will not need to remove the implant. Mg alloy’s mechanical properties deterio-
rate significantly before bone recovery due to the high corrosion rate. In the continuous
degradation of Mg, alloy hydrogen is evaluated which forms the air-cavity around the
implant, this is also adverse for the tissue and bone. Therefore, many researchers worked
on them to overcome the challenges of Mg alloy. A. Gungor et al. [5] investigated the
influence of alloys and thermo-mechanical procedures on the corrosion and mechanical
characteristics of biodegradable magnesium alloys. The theoretical findings were dis-
covered: grain size dropped drastically as the Zn level increased, corrosion resistance
improved when the Zn/Ca atoms proportion increased, and the tensile strength of hot
rolled alloys was substantially higher, but the toughness and ductility were substan-
tially lower. Zhensheng Lin et al. [6] have conducted a study on the impact of popular
antimicrobial metals, such as zinc (Zn), silver (Ag), and copper (Cu), in increasing the
antimicrobial characteristics and resistance to corrosion of Mg-based alloys. When a
Zn montmorillonite (Zn-MMT) covering was applied to the surfaces of an AZ31 alloy,
researchers discovered that it had greater corrosion protection and antimicrobial capa-
bilities than an unprotected AZ31 alloy. Xuannam Ly et al. [7] synthesized a composite
coating on bio-degradable Mg-4.71 wt % Zn-0.6 wt % Ca alloy by micro-arc oxida-
tion (MAO). The preprocessing of equal-channel-angular-pressing (ECAP) is done to
enhance the material and corrosion characteristics. The immersion test indicates that the
resistance to corrosion of the ECAP sample is superior to the as-cast sample. J. E. Gray
et al. [3] have done a study of protective coatings for magnesium and its alloys. For
improved wear and corrosion resistance, this analysis delves into coating and surface
alteration techniques such as conversion coatings, electroplating, gas phase deposition
methods, organic coatings anodizing, and laser surface cladding/alloying. Nanoindenta-
tion measurements were used to determine nanoscale hardness and Young’s modulus of
nanostructured HA coatings fabricated by RF magnetron sputtering on AZ31 alloy. On
both a micro- and nanoscale, the HA coatings increased the hardness of the magnesium
alloy during the Nanoindentation test [8]. So the magnesium alloys are lighter in weight,
have a high strength-to-weight ratio, have a high vibration damping capacity, and are
fully recyclable, good biodegradable implant material. The addition of elements such
as Zn, Al, Ag, Ca, Th, and Ce, which can form diverse microstructures and improve
the mechanical characteristics, is done to increase the strength of Mg-based alloys [9].
Magnesium alloy surface treatment for improved corrosion resistance has seen a lot of
development and research, but relatively little work has been done on Magnesium alloy
surface modification by RF Magnetron Sputtering [10]. The main goal is to change the
surface of a magnesium alloy to increase corrosion resistance and mechanical qualities
while reducing weight and increasing strength-to-weight ratio.

2 Materials and Methods

2.1 Sample Preparation
AZ31B Mg alloy was employed as experimental material in this study. The material is
procured from PLUS METALS, Mumbai. In this investigation, AZ31B Mg alloy was
582 A. Malik et al.

sliced into small sheets (20mm × 20mm × 10mm) and used as substrates. Before surface
treatment, the samples were manually polished with SiC abrasive paper, then cleaned
with acetone and deionized water before being dried in the air. AZ31B has composition
and mechanical properties as shown in Tables 1 and 2 respectively.

Table 1. Composition of AZ31B alloy

Si Fe Cu Mn Al Zn Ni Mg
MAX 0.08 MAX 0.003 MAX 0.01 0.20 2.5 0.60 MAX BAL
1.0 3.5 1.4 0.001
0.019 0.0012 0.0021 0.33 3.01 1.1 0.0007 BAL

Table 2. Mechanical properties of AZ31B alloy

Tensile Elongation (mm) Yield strength Applied standard Ultrasonic

strength (MPa) inspection
(MPa)
250 8 150 GB/T5153–2003 GB/T23600–2009
GB/T5154–2010 A Class
259 10 157 Good

2.2 RF Magnetron Sputtering

Magnetron sputtering (MS) is a fast vacuum process that belongs to the PVD cate-
gory. Target, Substrate holder, vacuum pump, power supply, electric coils, working gas,
cooling water and magnets are all components of a conventional magnetron sputtering
coating apparatus. The magnetron sputtering coating apparatus can generate coatings
of alloys, pure metals, and composites up to 5 μm thick. This machine’s power supply
can be divided into two categories: direct current (DC) and radio frequency (RF). The
capacity of sputtering the atom from nonconductor material is RFMS’s most significant
advantage over DCMS (Fig. 1).
A target intended to be coated on the substrate is bombarded with high-energy ions,
such as Ar or He, during the MS process. Whenever such high-energy ions hit the target
material, they provide the particles with the momentum and energy that already exist on
the target’s surface. The atoms that have gained momentum start vibrating and colliding
with those around them, resulting in a cascade. The atoms on the target material’s outer
surface sputtered outward when they received enough energy to break their bonds with
nearby atoms. The sputtering particles travel to the substrate surface which is mounted
on the holder through an ultraclean atmosphere created by the vacuum pump.
Coating the substrate after cleaning with the ethanol mounted on the RF magnetron
sputtering holder using double-sided tape. The substrate sticks to the holder. After that,
 Improving the Surface Characteristics of Mg AZ31B Alloy … 583

Fig. 1. (a) RF Magnetron Sputtering System, (b) RF Magnetron Sputtering System heating cham-
ber, (c) Mounting of Substrate on RF magnetron sputtering holder, (d) Creation of high energy
ionized Plasma

the gate of the system is closed and a vacuum is created. When the vacuum is created
there is a process gas valve and the fore line valve starts. After some time high-energy
ionized plasma is created.

2.3 Measurement

WENSAR® electronic balance weighing scale was used to compute the weight of the
uncoated and coated samples. It has the capacity of minimum measurement of 1mg to
maximum weight measurement of 310 g. It is used to calculate the weight which is
further used to calculate the corrosion rate and degradation rate.

2.4 Simulated Body Fluid (SBF)

SBF is fluid with an ion concentration equivalent to the blood plasma of a human,
held at the same pH and physiological temperature as human blood plasma approx 7.4.
Preparation of HBSS (Hank’s Balanced Salt Solution).

• In an appropriate container, prepare 800 mL of distilled water.

• To the solution, add 8 g of NaCl.
• 400 mg KCl is added to the solution.
• Add 140 mg of CaCl2 into the solution.
• Add 100 mg MgSO4 -7H2 O into the solution.
584 A. Malik et al.

• To the solution, add 100 mg MgCl2 -6H2 O.

• To the solution, add 60 mg of Na2 HPO4 -2H2 O.
• 60 mg KH2 PO4 should be added to the solution.
• To the solution, add 1 g of D-Glucose (Dextrose).
• 350 mg NaHCO3 should be added to the solution.
• Pour in distilled water until the volume reaches 1 L. Stir the solution with a hot plate
magnetic Stirrer. pH is calculated by pH meter and. pH should be maintained approx
7.4 and temperature remains approx human body temp which is nearly 37 °C.

2.5 Immersion Test

An immersion test was performed to observe the corrosion behavior of AZ31B alloy.
The initial weight of the sample before dipping into SBF was measured by a weighing
machine. After the preparation of the SBF solution, the samples both coated and uncoated
(ZnO and Ag sputtered) were immersed in the SBF for different time intervals. The
stirring was done by a hot plate magnetic stirrer. The temperature should be maintained to
approximately 37 °C near the human body temperature by using a temperature regulator
in a hot plate magnetic stirrer. The weight of the samples was observed for various time
intervals of 3, 5, and 7 days. The corrosion rate (CR) is calculated for various samples.
As per ASTM standard practice the following calculation is used to compute the average
corrosion rate (Table 3).
(K × W )
CR =
(A × T × D)
where, K = Constant
T = exposure time (hours).
A = Area (sq. cm).
W = mass loss (grams).
D = density (g/cm3 ).

Table 3. Values of constant K

Corrosion rate units desired K (constant)

Miles per year (mpy) 3.45 × 106
Inches per month (ipm) 2.87 × 102
Millimeter per year (mm/y) 8.76 × 104

The degradation rate (μg cm−2 day−1 ) is calculated by the following equation:
(Wo × Wt )
Degradation rate =
(A × T )
where, W o (μg) = Initial weight.
W t (μg) = Final weight after removal of the corrosion product.
A = Exposed area of the sample to solution in cm2 .
T = Exposure time in days.
 Improving the Surface Characteristics of Mg AZ31B Alloy … 585

2.6 Surface Roughness Test

Surface roughness of the sample before the sputtering and after the sputtering was
observed by using TAYLOR HOBSON® Profilometer. Surface roughness is a very
important criterion for using the AZ31B as a biodegradable implant material. Rough
surfaces wear out faster and show more corrosion.

3 Results and Discussions

The immersion test shows that the sample after the RF magnetron sputtering the sam-
ples having higher corrosion resistance. The corrosion rate of the coated sample is
significantly reduced (Figs. 2, 3 and Tables 4 and 5).

Fig. 2. Samples before immersion test (a) Uncoated (b) ZnO Coated (c) Ag coated

Fig. 3. Samples after immersion test (a) Uncoated (b) ZnO Coated (c) Ag coated

The degradation rate of the coated and uncoated sample is also calculated after the
immersion test which is shown below (Table 6).
Surface roughness testing by Profilometer shows average roughness values (Ra),
Maximum Height of the Profile (Rt), Maximum Profile Valley Depth (Rv), Maximum
Profile Peak Height (Rp), and Cutoff (λc) (Fig. 4 and Table 7).
586 A. Malik et al.

Table 4. Initial and final weight of various samples for the different time interval

Sample Initial weight (g) Final weight (g)

After 3 days After 5 days After 7 days
Uncoated 10.8250 10.8211 10.8180 10.8147
ZnO coated 10.8750 10.8740 10.8733 10.8720
Ag coated 10.9060 10.9032 10.9005 10.8978

Table 5. Corrosion rate of various samples

Sample Corrosion Rate (CR) in mm/y

After 3 days 5 days 7 days
Uncoated 0.46291 0.49851 0.52394
ZnO Coated 0.11869 0.12106 0.15261
Ag Coated 0.33234 0.39168 0.41712

Table 6. Degradation rate of various samples

Sample Degradation rate (CR) (μg cm−2 day−1 )

After 3 days 5 days 7 days
Uncoated 206.349 233.333 245.238
ZnO coated 55.555 56.667 71.428
Ag coated 148.148 183.333 195.238

4 Conclusions
In this research work coating Zinc oxide (ZnO) and Silver (Ag) with the help of an
RF magnetron sputtering system was performed. A comparison of the degradation rate
is made with and without coating. The degradation test is done in Hank’s Solution, a
simulated body fluid (SBF) type. The biodegradable behavior of AZ31B in SBF was
examined for different immersing time intervals. Following are the conclusions that can
be drawn from the experimental results.

• The corrosion rate was found to be reduced when coated with Zinc oxide (ZnO) and
Silver (Ag) with the help of RF magnetron sputtering.
• The coating also reduced the surface roughness of samples.
• The surface roughness curves show that the roughness value of the coated sample is
significantly reduced which will be beneficial in its medical applications.
• The immersion test shows that the ZnO-coated sample has a higher corrosion
resistance than the uncoated and Ag-coated materials.
 Improving the Surface Characteristics of Mg AZ31B Alloy … 587

Fig. 4. Surface Roughness Curves by TAYLOR HOBSON® Profilometer (a) Uncoated AZ31B
alloy (b) ZnO coated AZ31B alloy (c) Ag coated AZ31B alloy

Table 7. Surface roughness profile parameters

Sample Surface roughness profile parameters

Ra (μm) Rt (μm) Rv (μm) Rp (μm) λc (mm)
Uncoated 1.40 8.50 3.00 3.50 0.80
ZnO coated 1.10 7.50 3.00 3.00 0.80
Ag coated 1.10 8.50 2.50 4.00 0.80

• The degradation rate was also reduced in the ZnO-coated samples.

• AZ31B with slight modification in surface properties which makes it better choice for
biodegradable implant material.

Declaration of Competing Interest. No financial conflicts of interest or relationships between

the study’s authors might influence their research.

References
1. Ren, Y., Babaie, E., Bhaduri, S.B.: Nanostructured amorphous magnesium phosphate/poly
(lactic acid) composite coating for enhanced corrosion resistance and bioactivity of biodegrad-
able AZ31 magnesium alloy. Prog. Org. Coat. 118, 1–8 (2018). https://doi.org/10.1016/j.por
gcoat.2018.01.014
2. Panahi, Z., Tamjid, E., Rezaei, M.: Surface modification of biodegradable AZ91 magnesium
alloy by electrospun polymer nanocomposite: Evaluation of in vitro degradation and cyto-
compatibility. Surf. Coat. Technol. 386, 125461 (2020). https://doi.org/10.1016/j.surfcoat.
2020.125461
588 A. Malik et al.

3. Gray, J.E., Luan, B.: Protective coatings on magnesium and its alloys—a critical review. J.
Alloy. Compd. 336, 88–113 (2002). https://doi.org/10.1016/S0925-8388(01)01899-0
4. Zhang, B., et al.: Poly (dimethyl diallyl ammonium chloride) incorporated multilayer coating
on biodegradable AZ31 magnesium alloy with enhanced resistance to chloride corrosion and
promoted endothelialization. Chem. Eng. J. 421, 127724 (2021). https://doi.org/10.1016/j.
cej.2020.127724
5. Gungor, A., Incesu, A.: Effects of alloying elements and thermomechanical process on the
mechanical and corrosion properties of biodegradable Mg alloys. J. Magnes. Alloys 9, 241–
253 (2021). https://doi.org/10.1016/j.jma.2020.09.009
6. Lin, Z., Sun, X., Yang, H.: The role of antibacterial metallic elements in simultaneously
improving the corrosion resistance and antibacterial activity of magnesium alloys. Mater.
Des. 198, 109350 (2021). https://doi.org/10.1016/j.matdes.2020.109350
7. Ly, X., Yang, S., Nguyen, T.: Effect of equal channel angular pressing as the pretreatment on
microstructure and corrosion behavior of micro-arc oxidation (MAO) composite coating on
biodegradable Mg–Zn–Ca alloy. Surf. Coat. Technol. 395, 125923 (2020). https://doi.org/10.
1016/j.surfcoat.2020.125923
8. Surmeneva, M.A., et al.: Enhancement of the mechanical properties of AZ31 magnesium
alloy via nanostructured hydroxyapatite thin films fabricated via radio-frequency magnetron
sputtering. J. Mech. Behav. Biomed. Mater. 46, 127–136 (2015). https://doi.org/10.1016/j.
jmbbm.2015.02.025
9. Wang, S., Zhang, X., Li, J., Liu, C., Guan, S.: Investigation of Mg–Zn–Y–Nd alloy for potential
application of biodegradable esophageal stent material. Bioactive Materials. 5, 1–8 (2020).
https://doi.org/10.1016/j.bioactmat.2020.01.002
10. Surmeneva, M.A., Mukhametkaliyev, T.M. Khakbaz, H. Surmenev, R.A. Bobby Kannan,
M.: Ultrathin film coating of hydroxyapatite (HA) on a magnesium–calcium alloy using RF
magnetron sputtering for bioimplant applications. Mater. Lett. 152, 280–282 (2015). https://
doi.org/10.1016/j.matlet.2015.03.140
 Influence of Dressing Parameters on Surface
Roughness and Wheel Life in Internal Grinding
SKD11 Tool Steel

Tran Huu Danh1 , Nguyen Hong Linh2 , Bui Thanh Danh3 , Nguyen Anh Tuan4 ,
Nguyen Thi Quoc Dung5 , Hoang Xuan Tu5 , and Le Xuan Hung5(B)
1 Vinh Long University of Technology Education, Vinh Long, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 University of Economics—Technology for Industries, Ho Chi Minh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

lexuanhung@tnut.edu.vn

Abstract. This paper presents a multi-objective study on internal grinding to

achieve minimum surface roughness (SR) and maximum wheel life (WL) at the
same time. This study looked at six input parameters: coarse dressing depth ar ,
coarse dressing time nr , fine dressing depth af , fine dressing time nf , non-feeding
dressing n0 , and dressing feed rate Sd . The effect of these parameters on the SR
and WL of the grinding process was investigated. Furthermore, optimal input
parameters to achieve SR and WL at the same time were proposed. It was also
reported that the optimal set of input dressing parameters was appropriate for use.

Keywords: Dressing · Internal grinding · Multi-objective optimization · Surface

roughness · Wheel life · SKD11

1 Introduction
Grinding is a common machining process used in practice to achieve a high-quality sur-
face finish. It is especially useful for machining annealed products with high hardness
and strength. It accounts for approximately 20–25% of total mechanical part expen-
ditures in industries. Researchers have been interested in improving performance and
lowering machining costs for general and internal grinding while maintaining accuracy
requirements for these reasons.
There have been numerous studies on the optimization of the internal grinding pro-
cess to date. The monitoring and optimization of internal grinding process have been
done in [1]. To reduce production time while maintaining part quality requirements, an
online optimization system for cylindrical plunge grinding was developed [2]. A study
on optimal determination of replaced grinding wheel diameter in internal grinding was
presented in [3]. The effects of grinding process parameters such as wheel life, total
dressing depth, radial grinding wheel wear per dress, and initial grinding wheel diame-
ter on the exchanged grinding wheel diameter were investigated in the study. The effect

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 589–598, 2023.
https://doi.org/10.1007/978-3-031-22200-9_64
590 T. H. Danh et al.

of coolant parameters on surface roughness in internal cylindrical grinding of annealed

9CrSi steel was evaluated in [4]. The work’s findings led to the identification of optimal
input parameters for achieving the lowest surface roughness. The authors in [5] deter-
mined the optimum replaced wheel diameter in stainless steel internal grinding. The
effects of input process factors and cost components were investigated in this work. In
[6] presented a study on optimizing internal traverse grinding of valves based on wheel
deflection. Many single-objective [7] and multi—objective [8] optimization studies on
internal grinding have been conducted. The optimum parameters of dressing process for
internal grinding was proposed in several studies [9, 10]. Recently in [11] have applied
the multi-criteria decision making (MCDM) method to determine the best dressing setup
for internal grinding process.
This paper presents the results of a multi-objective optimization study on dressing
process for internal grinding. In this work, the minimum SR and maximum WL were
selected for the objectives of the optimization study. The effect of the input parameters
of the dressing process were evaluated. In addition, an optimum set of input dressing
factors for internal grinding SKD11 tool steel was proposed.

2 Experimental Work

Fig. 1. Experimental setup

An experiment was carried out in order to solve the multi-objective optimization problem.
This experiment was designed using the Taguchi method and the L16 orthogonal array
(44 × 22 ). The input factors and their levels are shown in Table 1. Figure 1 depicts
the experimental setup, with the dressing parameters shown in Table 1. The experiment
was carried out as follows: the dressing process was carried out according to the plan
shown in Table 2. After dressing, the grinding wheel was used to grind test samples at
 Influence of Dressing Parameters on Surface Roughness … 591

the following speeds: wheel speed of 12000 (rpm); workpiece speed of 150 (rpm); radial
wheel speed of 0.0025 (mm/stroke); and axial feed speed of 1 (mm/min.). Following the
completion of the experiments, the SR (in this case, Ra (μm)) was measured and WL
(min.) was calculated. Table 2 shows the experimental plan as well as the output results
(RS and WL).

Table 1. Input dressing parameters and their levels

No. Input factors 1 2 3 4

1 Coarse dressing depth (mm) 0.025 0.03 – –
2 Coarse dressing time (times) 1 2 3 4
3 Fine dressing depth (mm) 0.005 0.01 0.015 0.02
4 Fine dressing time (times) 0 1 2 3
5 Non-feeding dressing (times) 0 1 2 3
6 Dressing feed rate (mm/min) 1 1.2 – –

Table 2. Experimental plan and output results

Exp. No ar nr af nf n0 Sd Ra (μm) Tw (min.)

Trial Trial Trial Trial Trial Trial
1 2 3 1 2 3
1 0.025 1 0.005 0 0 1.0 0.365 0.367 0.364 11.27 11.15 11.38
2 0.030 1 0.010 1 1 1.0 0.214 0.212 0.215 12.73 12.64 12.83
3 0.025 1 0.015 2 2 1.2 0.198 0.194 0.192 12.70 12.82 12.57
4 0.030 1 0.020 3 3 1.2 0.242 0.240 0.243 11.47 11.48 11.32
5 0.030 2 0.005 1 2 1.2 0.185 0.187 0.183 11.80 11.77 11.84
6 0.025 2 0.010 0 3 1.2 0.249 0.248 0.246 12.43 12.52 12.34
7 0.030 2 0.015 3 0 1.0 0.252 0.250 0.254 11.15 11.08 11.21
8 0.025 2 0.020 2 1 1.0 0.2.17 0.214 0.219 12.10 11.97 12.16
9 0.030 3 0.005 2 3 1.0 0.307 0.305 0.307 12.55 12.45 12.64
10 0.025 3 0.010 3 2 1.0 0.324 0.322 0.326 12.10 11.97 12.16
11 0.030 3 0.015 0 1 1.2 0.343 0.340 0.339 12.03 12.09 11.96
12 0.025 3 0.020 1 0 1.2 0.354 0.356 0.352 12.50 12.43 12.58
13 0.025 4 0.005 3 1 1.2 0.320 0.318 0.316 12.18 12.28 12.07
14 0.030 4 0.010 2 0 1.2 0.313 0.310 0.315 10.62 10.73 10.51
15 0.025 4 0.015 1 3 1.0 0.327 0.325 0.326 13.15 13.02 13.29
16 0.030 4 0.020 0 2 1.0 0.363 0.365 0.362 12.29 12.18 12.22
592 T. H. Danh et al.

Multi-objective Optimization
For multi-objective optimization, the gray relational analysis (GRA) method combined
with the Taguchi method were used. In the study, two objectives including the minimum
surface roughness and the maximum durability of the grinding wheel were selected for
the investigation.
The signal to noise (S/N) ratio for each test can be calculated using the surface
roughness expectation that “the smaller the better”:
 n 
1 2
S/N = −10log10 yi (1)
n
i=1

conversely for WL is “the larger is better”:

 
1 1
n
SN = −10log10 (2)
n yi2
i=1

where n is the total of test runs, yi the observed data.

The normalized
 values of grey relation for both SR and WL can be determined by
Zij 0 ≤ Zij ≤ 1 :
 
SNij − min SNij , j = 1, 2, . . . k
Zij =     (3)
max SNij , j = 1, 2, . . . n − min SNij , j = 1, 2, . . . n

Table 3 displays the S/N ratio and the normalized grey relation values. The grey rela-
tion coefficient, (y(k)) which express the relationship between reference and experimental
data normalized data is found as follows:
min + ζ max
γ (k) = (4)
j (k) + ζ max

In which:
oj is thedeviation of reference
 data.
j (k) = Z0 (k) − Zj (k)

 and Z0(k) is the reference data or best data.

(+)  min = min minZ0 (k) − Zj (k) minimum value of i (k).
∀j∈i ∀k  
(+)  max = max maxZ0 (k) − Zj (k) maximum value of i (k).
∀j∈i ∀k
(+) ζ is distinguishing or identification coefficient; ζ = 0÷1 (In this case ζ = 0.5).
Calculating the gray relational grade:

1
m
γj = γij (5)
k
i=1

wherein k is the number of objectives optimized.

 Influence of Dressing Parameters on Surface Roughness … 593

Table 3. Grey relational coefficient and grey grade values

No S/N Zi i (k) Gray relational yi

values yi
Ra Tw Ra Tw Ra Tw Ra Tw
Reference value
1.000 1.000
1 8.7488 21.0350 0.000 0.276 1.000 0.724 0.333 0.409 0.371
2 13.4051 22.0984 0.788 0.849 0.212 0.151 0.702 0.768 0.735
3 14.2084 22.0730 0.924 0.835 0.076 0.165 0.868 0.752 0.810
4 12.3355 21.1553 0.607 0.341 0.393 0.659 0.560 0.431 0.496
5 14.6562 21.4400 1.000 0.494 0.000 0.506 1.000 0.497 0.749
6 12.1225 21.8890 0.571 0.736 0.429 0.264 0.538 0.654 0.596
7 11.9718 20.9426 0.546 0.227 0.454 0.773 0.524 0.393 0.458
8 13.2838 21.6384 0.768 0.601 0.232 0.399 0.683 0.556 0.619
9 10.2729 21.9701 0.258 0.790 0.742 0.220 0.403 0.694 0.548
10 9.7920 21.6384 0.177 0.601 0.823 0.399 0.378 0.556 0.467
11 9.3562 21.6027 0.103 0.582 0.897 0.418 0.358 0.545 0.451
12 9.0171 21.9402 0.045 0.763 0.955 0.237 0.344 0.679 0.511
13 9.9544 21.7099 0.204 0.639 0.796 0.361 0.386 0.581 0.483
14 10.1013 20.5216 0.229 0.000 0.771 1.000 0.393 0.333 0.363
15 9.7386 22.3798 0.168 1.000 0.832 0.000 0.375 1.000 0.688
16 8.7912 21.7483 0.007 0.660 0.993 0.340 0.335 0.595 0.465

Table 4. Effect of dressing parameters on the gray relation values yi .

594 T. H. Danh et al.

Table 5. Influence order of the dressing parameters on yi .

3 Results and Analysis

The Taguchi method is used to assess how dressing parameters affect mean gray relation
values. The S/N value of is calculated using formula (2) in order to maximize the average
gray relation value. Table 4 shows the calculated results of the effect of input parameters
on the average gray relation value using the ANOVA method.
According to Table 4, the two parameters that have the greatest influence on yi are
nf (39.98%) and n > (32.24%); followed by the influence of nr (16.71%), af (5.31%),
and ar (1, 79%). Furthermore, Sd has no discernible effect on yi (0.27%). The order of
influence of the parameters on the gray relation value is described in Table 5. This table
shows that the order of influence of input parameters is nf, n0, nr, af, ar, and Sd, in that
order.
Figure 2 depicts the influence of input parameters across levels. From Fig. 2 it is
easy to see that when n0 increases from level 1 to level 3 increases; However, when it
increases further to level 4, it reduces yi ; With nf : when nf increases from level 1 to level
2, yi increases sharply; however, when it continues to increase from level 2 to level 4, yi
decreases; With af : when af increases from level 1 to level 2, yi increases insignificantly;
af increases from level 2 to level 3 causing yi to increase; Nonetheless, if it increases
to level 4, yi decreases sharply; The impact of nr is described as follows: If it increases
from level 1 to level 2 and from level 3 to level 4, yi changes little; Nevertheless, when
nr increases from level 2 to level 3, yi decreases sharply; The effect of ar as follows:
when ar increases from level 1 to level 2, yi will decrease. Besides, when Sd increases
from level 1 to level 2, yi increase slightly.
To determine the best dressing mode, take into account the effect of the input param-
eters on the S/N ratio. As previously discussed, the set of parameters with the highest
value for each input factor is the most reasonable. The larger value S/N corresponds to yi
being as large as possible. This also implies that the effect of noise is reduced. Figure 3
depicts the effect of input parameters on the S/N value. Also, this Figure identifies the
optimal set of parameters for the multi-objective function (Table 7).
 Influence of Dressing Parameters on Surface Roughness … 595

Fig. 2. Effect of input parameters on yi .

Fig. 3. Effect of input parameters on S/N ratio

Using Minitab 19, the optimal set of parameters can be used to calculate and predict
the results of the objective functions (Table 6). The surface roughness Ra = 0.1821
596 T. H. Danh et al.

Table 6. Optimum dressing parameters

No Factor Level Value

1 Coarse dressing depth (mm) 1 0.025
2 Coarse dressing time (times) 2 2
3 Fine dressing depth (mm) 3 0.015
4 Fine dressing time (times) 2 1
5 Non-feeding dressing (times) 3 2
6 Dressing feed rate (mm/min) 2 1.2

Table 7. Prediction of values RS and WL corresponding to the optimal parameters

(μm) and the wheel life Tw = 12.8431 (min.) were obtained from the predictive analysis
results with the optimal set of dressing parameters.
Evaluation of Experimental Models:
The Anderson–Darling method was used to assess the suitability of the experimental
model (Fig. 4). Figure 4 shows that the experimental points (the blue points) on the
histogram are very close to the normal distribution line (the red solid line). It demonstrates
that the deviation is very small. The Histogram graph of error frequency shows that the
majority of errors are in the −0.1 to 0.1 range. The experimental errors in the other two
graphs are distributed at random. This means that the built model is heavily influenced
by the input parameters chosen and is unaffected by the order of experiments.
The Anderson–Darling method was used to validate the experimental model’s fit
(Fig. 5). The results on this figure show that the experimental points (blue dots) are all
within a 95% confidence interval bounded by two limit lines. Furthermore, the value of P
of 0.185 is greater than a = 0.05, indicating that the empirical model used is appropriate.
 Influence of Dressing Parameters on Surface Roughness … 597

Fig. 4. Residual plots for S/N ratios

Fig. 5. Probability plot of the fit of proposed model

598 T. H. Danh et al.

4 Conclusions
The results of a multi-objective optimization study in the dressing process for inter-
nal grinding SKD11 tool steel are presented in this article. For the investigation, six
input dressing parameters were chosen. The Taguchi technique is used in conjunction
with GRA to reduce the surface roughness and increase the wheel life. The following
conclusions can be drawn from the findings of this work:

– The influence of the dressing factors on the SR and WL of the grinding process was
investigated.
– The most affected factors on yi are nf (39.98%) and (32.24%); followed by nr (16.71%),
af (5.31%), ar (1, 79 %), and Sd (0.27%).
– The following input factors were proposed for the best dressing process when internal
grinding to achieve minimum SR and maximum MRR simultaneously: ar = 0.03
(mm/l); nr = 2 (times); ar = 0.025 (mm); af = 0.015 (mm); nf = 1 (times); n0 = 2
(times); and Sd = 1.2 (m/min).

Acknowledgments. This work was supported by Thai Nguyen University of Technology.

References
1. Inasaki, I.: Monitoring and optimization of internal grinding process. CIRP Ann. 40(1), 359–
362 (1991)
2. Xiao, G., Malkin, S.: On-line optimization for internal plunge grinding. CIRP Ann. 45(1),
287–292 (1996)
3. Hung, L.X., et al.: Determination of optimal exchanged grinding wheel diameter when inter-
nally grinding alloy tool steel 9CrSi. In: IOP Conference Series: Materials Science and
Engineering. IOP Publishing (2018)
4. Hung, L.X., et al.: A study on coolant parameters in internal grinding of 9CrSi steel. In:
Materials Science Forum. Trans Tech Publ (2019)
5. Tran, T.-H., et al.: Optimization of replaced grinding wheel diameter for minimum grinding
cost in internal grinding. Appl. Sci. 9(7), 1363 (2019)
6. Gao, S., Yang, C., Xu, J., Fu, Y., Su, H., Ding, W.: Optimization for internal traverse grinding
of valves based on wheel deflection. Int. J. Adv. Manuf. Technol. 92(1–4), 1105–1112 (2017).
https://doi.org/10.1007/s00170-017-0210-8
7. Hung, L.X., et al.: A study on calculation of optimum exchanged grinding wheel diameter
when internal grinding. Mater. Today: Proc. 18, 2840–2847 (2019)
8. Hung, L.X., et al.: Multi-objective optimization of dressing parameters of internal cylindrical
grinding for 9CrSi Aloy steel using taguchi method and grey relational analysis. Mater. Today:
Proc. 18, 2257–2264 (2019)
9. Nguyen, A.T., et al.: Optimization of dressing parameters for minimum surface roughness
and maximum material removal rate in internal grinding of SKD11 tool steel. In: E3S Web
of Conferences. EDP Sciences (2021)
10. Hung, L.X., et al. Determining the optimum set of dressing parameters satisfying minimum
surface roughness when conducting the internal grinding of hardened SKD11 steel. In: Solid
State Phenomena. Trans Tech Publ (2021)
11. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
 Influence of Input Factors on Surface Roughness
When Using Wire-EDM SKD11 Tool Steel

Tran Huu Danh1 , Nguyen Hong Linh2 , Bui Thanh Danh3 , Tran Minh Tam4 ,
Nguyen Huu Quang5 , Tran Ngoc Giang6 , Nguyen Manh Cuong6 ,
and Nguyen Thi Thanh Nga6(B)
1 Vinh Long University of Technology Education, Vinh Long, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 University of Economics - Technology for Industries, Ho Chi Minh City, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nguyennga@tnut.edu.vn

Abstract. This paper describes the results of a study to determine the optimal
process parameters when Wire Electrical Discharge Machining (Wire-EDM) pro-
cess was used to cut circular arcs of SKD11 steel. Several input parameters were
considered in this study, including the pulse on time, the pulse off time, the cutting
voltage, the servo voltage, the wire feed, the cutting speed, and the arc radius. The
effects of these parameters on the surface roughness of the workpiece after cutting
process were discovered. Furthermore, optimal process parameters for achieving
the best surface finish were discovered. It was discovered that the best experimental
mode was appropriate for use.

Keywords: WEDM · Wire-EDM · Surface roughness · Taguchi method · SKD11

1 Introduction
SKD11 steel is also known as SKD11 cold stamping die steel. It is one of the most
widely used steels in the mold making industry. This is because this steel has good
hardenability and high wear resistance. For that reason, up to now, there have been many
studies on SKD11 steel processing. To process SKD11, traditional machining methods
or non-traditional machining methods can be used. With traditional machining methods,
this steel has been studied when milling [1, 2], grinding [3–5], or turning [6, 7]. SKD11
steel is not only the object of study when using traditional machining methods but also
when using non-traditional machining methods. In [8] we studied the use of fiber laser
to polish surfaces of SKD11 workpieces machined by milling and WEDM. A laser heat
treatment system for SKD11 steel using a YAG laser was applied in [9]. EDM is widely
used to process conductive materials, including SKD11 steel. Up to now, there have been
many studies on machining SKD11 by EDM method [10, 11]. Wire EDM machining is a
type of EDM which is used very effectively to cut shaped holes, particularly narrow slits

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 599–606, 2023.
https://doi.org/10.1007/978-3-031-22200-9_65
600 T. H. Danh et al.

with curved arcs. Hence, there have been quite a few studies on this type of machining.
For example, the impact of WEDM’s parameters on wire crater sizes was examined in
[12]. The optimal process parameters have been investigated when processing hot die
steel [13], aluminium HE30 [14], and Ni 55.8 Ti [15]. SKD11 steel is also the object of
study when using this type of machining. For instance, the effect of input parameters on
the surface roughness was shown in [16, 17]. The optimal process parameters are also
recommended in [11, 18].
This work deals with the determination of optimal WEDM process parameters when
machining circular arcs SKD11 steel. In this work, seven input parameters were taken
into account. The effect of these parameters on the workpiece surface roughness was
explored. In addition, an experimental model for finding the optimum input parameters
was suggested.

2 Experimental Work

Fig. 1. Wire EDM machine for the experiment

 Influence of Input Factors on Surface Roughness … 601

Table 1. Input parameters and their levels

No Parameter Code Unit Level

1 2 3
1 Cutting voltage VM V 6 9 12
2 Pulse on time Ton μs 6 9 12
3 Pulse off time Toff μs 8 12 16
4 Servo voltage SV V 24 29 34
5 Wire feed WF mm/min 8 10 12
6 Arc radius R mm 3 6 9
7 Cutting speed SPD mm/min 2 4 –

To weigh the effect of the process parameters on the surface finish, an experiment
was conducted. In the experiment, seven WEDM process parameters were inspected.
These parameters are the pulse on time (Ton ), the pulse off time (Toff ), the cutting voltage
(VM), the servo voltage (SV), the wire feed (WF), the cutting speed (SPD), and the arc
radius (Table 1). Also, the Taguchi method with the design L18 (21 × 35 ) was used for
building the experimental plan. The machine used in this work is described in Fig. 1 and
the work material was SKD11. The experiment and the response (the surface roughness)
are shown in Table 2.

3 Results and Analysis

To weigh the effect of WEDM process parameters on the surface roughness, the Analysis
of Variance (ANOVA) method is used with the help of Minitab 18 software. Also, the
objective function is the smallest surface texture being the best:
 n 
1 2
S/N = −10log10 yi (1)
n
i=1

The results of ANOVA analysis (Table 3) show that: Ton has the greatest influence
on Ra (51.54%); followed by the influence of VM (19.34%). The remaining parameters
have the effect level of SV (11.92%), R (6.87%), Toff (4.22%) and WF (3.69%), SPD
has a negligible effect on Ra, respectively.
602 T. H. Danh et al.

Table 2. Experimental plan and the surface roughness

No VM Ton Toff SV WF SPD R Ra (μm) Mean S/N

Trial 1 Trial 2 Trial 3
1 3 6 8 24 8 2 3 1.719 1.720 1.700 1.7130 −4.6753
2 6 6 12 29 10 2 6 1.680 1.793 1.767 1.7466 −4.8472
3 9 6 16 34 12 2 9 1.414 1.321 1.238 1.3243 −2.4523
4 6 9 8 24 10 2 9 1.707 1.736 1.730 1.7243 −4.7324
5 9 9 12 29 12 2 3 1.643 1.620 1.656 1.6397 −4.2955
6 3 9 16 34 8 2 6 1.744 1.679 1.611 1.6781 −4.5008
7 9 12 8 29 8 2 6 1.845 1.848 1.858 1.8503 −5.3449
8 3 12 12 34 10 2 9 1.883 1.883 1.876 1.8806 −5.4861
9 6 12 16 24 12 2 3 1.958 1.974 1.962 1.9646 −5.8657
10 6 6 8 34 12 4 6 1.388 1.441 1.414 1.4147 −3.0143
11 9 6 12 24 8 4 9 1.538 1.516 1.466 1.5065 −3.5613
12 3 6 16 29 10 4 3 1.801 1.792 1.791 1.7946 −5.0794
13 3 9 8 29 12 4 9 1.621 1.719 1.706 1.6819 −4.5191
14 6 9 12 34 8 4 3 1.789 1.785 1.834 1.8026 −5.1188
15 9 9 16 24 10 4 6 1.617 1.657 1.645 1.6397 −4.2957
16 9 12 8 34 10 4 3 1.737 1.751 1.767 1.7515 −4.8683
17 3 12 12 24 12 4 6 2.016 2.184 2.193 2.1310 −6.5781
18 6 12 16 29 8 4 9 1.852 1.907 1.895 1.8846 −5.5051

Table 3. Analysis of variance for SN ratios

Table 4 shows the order of impact of input parameters on S/N as Ton , VM, SV, R, Toff ,
WF, and SPD, respectively. With the objective function as in Eq. (1), to get the minimal
 Influence of Input Factors on Surface Roughness … 603

surface finish, the S/N value is maximal for each input process parameter. Hence, based
on the analysis of the effect of the parameters on the S/N ratio in the chart in Fig. 2,
the optimal input parameters for the WEDM process were found to achieve the smallest
surface roughness, as shown in Table 5.
To predict the surface roughness corresponding to this set of optimum input param-
eters, the analysis tool to predict the results of the Taguchi method from Minitab 18
software was used and the results are shown in Table 6. In this case, the minimum
surface roughness was 1.2413 μm.

Table 4 Order of influence of input parameters on S/N.

Table 5. Optimum Wire-EDM parameters

No Parameter Code Unit Level Value

1 Cutting voltage VM V 2 9
2 Pulse on time Ton μs 1 6
3 Pulse off time Toff μs 1 8
4 Servo voltage SV V 3 34
5 Wire feed WF mm/min 3 12
6 Arc radius R mm 3 9
7 Cutting speed SPD mm/min 1 2

Table 6. Prediction of minimum surface roughness

604 T. H. Danh et al.

Fig. 2. Effect of WEDM parameters on S/N ratios

Fig. 3. Residual plots for S/N ratios

Evaluation of experimental models:

The proposed model is evaluated through error distribution charts (Fig. 3). From the
Figure, it is found that the errors of the experimental points corresponding to the blue
 Influence of Input Factors on Surface Roughness … 605

points on the graph are close to the normal distribution line (the red solid line). That means
the error level is very small. The remaining two graphs describe the random distribution
of the experimental errors. That means the optimal model is largely influenced by the
selected input parameters and it is not affected by the experimental order.
Figure 4 shows the suitability of the experimental model and the optimum set of
parameters. This graph was built by using the Anderson–Darling method and the Minitab
18 software. From the Figure, the experimental points (the blue dots) are all within the
bounded area by 2 upper and lower lines which are limited with 95% significance level.
In addition, the applied experimental model is suitable as the P value is 0.561—larger
than the value α = 0.05.

Fig. 4. Probability plot of the fit of the experimental model for Ra

4 Conclusions

This work deals with a study on the influence of WEDM process parameters when pro-
cessing circular arcs SKD11 tool steel. In this work, seven input parameters containing
the cutting voltage, the pulse on time, the pulse off time, the servo voltage, the wire feed,
the cutting speed, and arc radius were explored. The effects of these parameters on the
surface finish were studied. It was found that Ton has the greatest effect on Ra (51.54%);
followed by VM (19.34%), SV (11.92%), R (6.87%), Toff (4.22%), WF (3.69%) and
SPD has a negligible effect on Ra. Besides, a set of optimum input parameters was
proposed: VM = 9; Ton = 6; Toff = 8; SV = 34; WF = 12; R = 9, and SPD = 2. Also,
it is noted that the proposed optimum model is suitable for application.

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

606 T. H. Danh et al.

References
1. Jing, L.-L., et al.: Experiment on surface integrity of milling tool for hardened steel SKD11.
Trans. Nanjing Univ. Aeronaut. & Astronaut. 24(2), 157–163 (2007)
2. Zhao, R., Lv, X.J., Liu, J.: Simulation study of milling force and milling temperature for high
speed milling thin-wall SKD11 hardened steel part. In: 2016 5th International Conference on
Measurement, Instrumentation and Automation (ICMIA 2016). Atlantis Press (2016)
3. Tran, T.H., et al.: Improvement of wheel life by optimization of dressing parameters in surface
grinding of SKD11 steel. In: Materials Science Forum. Trans Tech Publ (2021)
4. Tran, T.H., et al.: Influence of dressing conditions on surface roughness when surface grinding
SKD11 steel. In: Materials Science Forum. Trans Tech Publ (2021)
5. Tran, T.H., et al.: Optimizing dressing conditions for minimum flatness tolerance when
grinding SKD11 tool steel. In: Materials Science Forum. Trans Tech Publ (2021)
6. Yih-fong, T.: Parameter design optimisation of computerised numerical control turning tool
steels for high dimensional precision and accuracy. Mater. Des. 27(8), 665–675 (2006)
7. Kumar, Y., Singh, H.: Multi-response optimization in dry turning process using Taguchi’s
approach and utility concept. Procedia Mater. Sci. 5, 2142–2151 (2014)
8. Dai, W., et al.: Evaluation of fluences and surface characteristics in laser polishing SKD 11
tool steel. J. Mater. Process. Technol. 273, 116241 (2019)
9. Hirogaki, T., et al.: In-situ heat treatment system for die steels using YAG laser with a
machining center. Precis. Eng. 25(3), 212–217 (2001)
10. Lee, H.-T., et al.: The study of EDM hole-drilling method for measuring residual stress in
SKD11 tool steel. J. Mater. Process. Technol. 149(1–3), 88–93 (2004)
11. Ming, W., et al.: Investigating the energy distribution of workpiece and optimizing process
parameters during the EDM of Al6061, Inconel718, and SKD11. Int. J. Adv. Manuf. Technol.
92(9–12), 4039–4056 (2017). https://doi.org/10.1007/s00170-017-0488-6
12. Tosun, N., Pihtili, H.: The effect of cutting parameters on wire crater sizes in wire EDM. Int.
J. Adv. Manuf. Technol. 21(10–11), 857–865 (2003)
13. Kuruvila, N.: Parametric influence and optimization of wire EDM of hot die steel. Mach. Sci.
Technol. 15(1), 47–75 (2011)
14. Reddy, V.C., Deepthi, N., Jayakrishna, N.: Multiple response optimization of wire EDM on
aluminium HE30 by using grey relational analysis. Mater. Today: Proc. 2(4–5), 2548–2554
(2015)
15. Magabe, R., et al.: Modeling and optimization of Wire-EDM parameters for machining of
Ni 55.8 Ti shape memory alloy using hybrid approach of Taguchi and NSGA-II. Int. J. Adv.
Manuf. Technol. 102(5), 1703–1717 (2019)
16. Bobbili, R., Madhu, V., Gogia, A.: Effect of wire-EDM machining parameters on surface
roughness and material removal rate of high strength armor steel. Mater. Manuf. Processes
28(4), 364–368 (2013)
17. Thi Hong, T., et al.: Effects of process parameters on surface roughness in wire-cut EDM of
9CRSI tool steel. Int. J. Mech. Eng. Technol., 10(3) (2019)
18. Chen, X., et al.: Investigation on MRR and machining gap of micro reciprocated wire-EDM
for SKD11. Int. J. Precis. Eng. Manuf. 21(1), 11–22 (2020)
 Influence of Input Process Factors on Surface
Roughness in CBN Grinding Al6061 T6

Nguyen Hong Linh1 , Bui Thanh Danh2 , Tran Huu Danh3 , Nguyen Anh Tuan4 ,
Nguyen Van Tung5 , Hoang Xuan Tu5 , and Nguyen Van Trang5(B)
1 Electric Power University, Hanoi, Vietnam
2 University of Transport and Communications, Hanoi, Vietnam
3 Vinh Long University of Technology Education, Hanoi, Vietnam
4 University of Economics - Technology for Industries, Hanoi, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nvtrang@tnut.edu.vn

Abstract. This paper presents the findings of a study on the influence of process
parameters on surface roughness (SR) in CBN grinding when processing Al6061
T6. The study looked at four input parameters including wheel speed (WS), feed
rate (Fe), depth of cut (aed ), and down feed (Df ). The effect of these variables on
the grinding process’s SR was evaluated. A regression model was also proposed
in order to identify the relationship between the input parameters and the SR. It
was also stated that the proposed model could be used.

Keywords: Grinding · CBN grinding · Surface roughness · Al6061 T6

1 Introduction
CBN grinding makes use of a cubic boron nitride CBN wheel. CBN wheels, which have
high abrasion resistance and thermal conductivity and can maintain sharp cutting edges,
are increasingly being used in mechanical processing. As a result, many scientists are
interested in CBN grinding research.
W. B. Rowe et al. [1] proposed in-process thermal damage prevention in CBN grind-
ing. According to the findings of this study, one significant advantage of CBN grinding
is that a significantly lower proportion of total grinding energy enters the workpiece
when compared to alumina wheel grinding. Rodrigo Daun Monici et al. [2] reported on
an analysis of the various forms of application and types of cutting fluid used in plunge
cylindrical grinding with conventional and superabrasive CBN grinding wheels. R. Pavel
and A. Srivastave [3] introduced an experimental study on effect of temperatures dur-
ing conventional and CBN grinding. In [4] investigated the wheel surface topography
during the grinding process with vitrified CBN wheels. This study also discussed the
relationships between these results and the grinding characteristic parameters. Z. Shi
and S. Malkin [5] published the findings of a study on the wear of CBN grinding wheels
as well as the influence of the wear on the topography and grinding behavior of the
wheel. [6] introduced a study to improve CBN wheel tool life and optimize the grinding

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 607–612, 2023.
https://doi.org/10.1007/978-3-031-22200-9_66
608 N. H. Linh et al.

process while taking economic, environmental, and social sustainability into account.
The results on optimization of grinding parameters for the minimum grinding time when
CBN grinding tablet punches on a CNC milling machine were reported in [7]. In [8] the
effect of vitrified bond structure on radial wear of CBN grinding wheels was discussed.
The grinding performance of AISI D6 steel using a vitrified and resinoid bonded CBN
wheel was reported in [9].
The effect of the input factors on the surface roughness in CBN grinding Al6061 T6
was investigated in this paper. In addition, a regression equation to identify the relation
between the input grinding factors and surface roughness was proposed.

2 Experimental Work

Fig. 1. Experimental setup

Table 1. Input factors

Factor Unit Low High

Wheel speed (WS) rpm 4000 5000
Feed Rate (Fe) mm/min 2000 3000
Depth of cut (aed ) mm 0.01 0.025
Down feed (Df ) mm/double journey 0.5 2

To conduct the investigation of the effect of process factors on RS, an experiment

was carried out. The Minitab R19 software was used to implement this experiment,
 Influence of Input Process Factors on Surface … 609

which had a 2-level 1/2 fraction design. As a result, 24–1 = 8 experimental runs were
completed. Table 1 describes the input factors and their levels. Figure 1 describes the
experimental setup, which consists of a CNC milling machine (Model M-V50C - Japan),
a CBN grinding wheel, CBN wheel dresser equipment (V-TDM-2 type), Al6061 work-
piece samples, and a roughness meter (SURFTEST SV-3100). Besides, the experimental
matrix is exhibited in Table 2. After conducting experiment, the surface roughness was
measured. The average results of the three measurements of SR are given in the first
column from the right of Table 2.

Table 2. Experimental matrix and output response

Run order Rpm Fe aed Df SR (µm)

1 4000 2000 0.01 0.5 0.1497
2 4000 3000 0.025 0.5 0.1510
3 5000 2000 0.025 0.5 0.1257
4 5000 3000 0.01 0.5 0.1300
5 5000 2000 0.01 2 0.3377
6 5000 3000 0.025 2 0.1900
7 4000 2000 0.025 2 0.1857
8 4000 3000 0.01 2 0.1673

3 Results and analysis

The effect of the input factors on the SR is described in Fig. 2. Figure 2 shows that all
four input parameters have an effect on the surface texture. Specifically, increasing Rpm
and Df parameters increases SR, while increasing Fe and aed parameters decreases SR.

Fig. 2. Main effects plot for surface roughness

610 N. H. Linh et al.

The Pareto chart (Fig. 3) illustrates how the input factors and their interactions affect
the Ra . The length of the blue column on this graph denotes the influence of the factors.
It is obvious that the effect of D (Df ) on Ra is the most significant. Other parameters with
decreasing influence include AD (Rpm *Df ), AC (Rpm *aed ), B (Fe ), C (aed ), A (Rpm ),
and AB (respectively Rpm *Fe ). Figure 4 clearly shows the effect of the interactions.

Fig. 3. The Pareto chart of the effects

Fig. 4. Effect of two-way interactions on surface roughness

 Influence of Input Process Factors on Surface … 611

Finding regression model to determine the surface roughness:

The regression equation for the RS is determined using Minitab 19 software. Table 3
shows the estimated coefficients of the input factors as well as their interactions. The
following is the regression equation:

Ra = −0.64415 + 0.0001989 · Rpm + 0.000244 · Fe + 23.55 · aed

− 0.2755 · Df − 0.0000001 · Rpm · Fe − 0.0057233 · Rpm · aed + 0.0000732 · Rpm · Df (1)

Table 3. Coded coefficients of regression model to determine RS

Coded coefficients
Term Effect Coef SE Coef T-Value P-Value VIF
Constant 0.1796 * * *
Rpm 0.03242 0.01621 * * * 1.00
FE −0.04012 −0.02006 * * * 1.00
Aed −0.03308 −0.01654 * * * 1.00
Df 0.08108 0.04054 * * * 1.00
Rpm*Fe −0.03158 −0.01579 * * * 1.00
Rpm*aed −0.04293 −0.02146 * * * 1.00
Rpm*Df 0.05493 0.02746 * * * 1.00

Evaluation of proposed model:

Fig. 5. Probability plot of the fit of the regression model

612 N. H. Linh et al.

The Anderson-Darling method was used to confirm the experimental model’s fit (Fig. 5).
The graph illustrates that all of the experimental data (blue dots) were located within a
95% confidence interval defined by two limit lines. Furthermore, P = 0.118 is greater
than 0.05, indicating that the proposed model is suitable.

4 Conclusions

This paper presents the results of a study on the influence of process parameters on
surface roughness in CBN grinding Al6061 T6. Four input parameters were examined
in the study including wheel speed, feed rate, depth of cut, and down feed. The impact of
these variables on the surface roughness of the CBN grinding process was investigated. In
order to determine the relationship between the input factors and the surface roughness,
a regression model was also recommended. The proposed model could also be used, as
it was stated.

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Rowe, W., Black, S., Mills, B.: Temperature control in CBN grinding. Int. J. Adv. Manuf.
Technol. 12(6), 387–392 (1996)
2. Monici, R.D., et al.: Analysis of the different forms of application and types of cutting fluid used
in plunge cylindrical grinding using conventional and superabrasive CBN grinding wheels. Int.
J. Mach. Tools Manuf 46(2), 122–131 (2006)
3. Pavel, R., Srivastava, A.: An experimental investigation of temperatures during conventional
and CBN grinding. Int. J. Adv. Manuf. Technol. 33(3), 412–418 (2007)
4. Fujimoto, M., et al.: Characterization of wheel surface topography in cBN grinding. JSME Int
J., Ser. C 49(1), 106–113 (2006)
5. Shi, Z., Malkin, S.: Wear of electroplated CBN grinding wheels (2006)
6. Caggiano, A., Teti, R.: CBN grinding performance improvement in aircraft engine components
manufacture. Procedia CIRP 9, 109–114 (2013)
7. Vu, N.-P., et al.: Optimization of grinding parameters for minimum grinding time when grinding
tablet punches by CBN wheel on CNC milling machine. Appl. Sci. 9(5), 957 (2019)
8. Herman, D., Krzos, J.: Influence of vitrified bond structure on radial wear of cBN grinding
wheels. J. Mater. Process. Technol. 209(14), 5377–5386 (2009)
9. Sato, B.K., et al.: Grinding performance of AISI D6 steel using CBN wheel vitrified and
resinoid bonded. Int. J. Adv. Manuf. Technol. 105(5–6), 2167–2182 (2019). https://doi.org/10.
1007/s00170-019-04407-5
 Influence of WEDM Process Parameters
on Material Removal Rate When Machining
Circular Profile of Hardened SKD11 Steel

Nguyen Anh Tuan1 , Nguyen Hong Linh2 , Tran Huu Danh3 , Nguyen Huu Luan4 ,
Bui Thanh Danh5 , Do Thi Tam6 , Nguyen Thi Quoc Dung6 ,
and Nguyen Dinh Ngoc6(B)
1 University of Economics - Technology for Industries, Hanoi, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 Vinh Long University of Technology Education, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 University of Transport and Communications, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

ngocnd@tnut.edu.vn

Abstract. The aim of the current study is to find the optimum set of WEDM
parameters which can maximize the material removal rate when machining circular
profile of hardened SKD11 steel. Six main parameters namely Cutting voltage,
pulse on time, pulse off time, server voltage, wire feed, cutting speed have been
investigated by using Taguchi technique. In order to find the significant level of
WEDM process parameters, analysis of variance has been adopted. The results
show that Toff has the most influence on MRR with 46.21%. Confirmation shows
that the proposed process for optimizing is reliable and can be extended to WEDM
machining of other hard work materials.

Keywords: WEDM · Wire-EDM · Surface Roughness · Material removal rate ·

SKD11

1 Introduction
In recent decades, the hard materials are widely utilized in various applications in indus-
try. In order to cut this kind of materials, using the conventional operations such as
milling, drilling, trimming, turning are difficult to get the effectiveness, quality as well.
Wire electrical discharge machining (WEDM) can machine the hard materials and com-
pletely meet the requirements of quality and effectiveness [1–4]. The working principle
of WEDM is that removed material process is taken place between the wire electrode
and workpiece because of energy thermally generated by control discharges. To remove
the microchip, a dielectric fluid is provided inside the machining area.
The important parameters have an effect on the machining qualities like Cutting
voltage, Pulse on time, Pulse off time, Server voltage, Wire feed, and Cutting speed [5–
8]. It should be profoundly investigated for the best results. Almost published research

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 613–622, 2023.
https://doi.org/10.1007/978-3-031-22200-9_67
614 N. A. Tuan et al.

work dealing with WEDM has been experimentally studied [9–14]. Taguchi method
has been perfectly applied so far when conducting the single objective function such
as increasing the material removal rate or decreasing the surface roughness. In case of
considering multi-objective functions, a grey relational analysis (GRA) method coupled
with Taguchi method is preferred to adopt [15, 16]. Optimization process to find the
optimum set of machining parameters for getting the objective functions is necessary.
Hence, it is mandatory to optimize the process characteristics using a suitable method.
Shukla et al. [16] evaluate the effects of process parameters (Peak current, Pulse on
time, Pulse off time and Wire speed) on the MRR when machining WEDM of Titanium
Alloy (6Al-4V). The orthogonal array of 33 = 27 is selected based on Taguchi method.
It reveals that Pulse current has clear impact on the response. When conducting the
influence of machining parameters on the surface roughness and MRR, Bobbili et al.
[9] found that pulse-on time, pulse-off time, and spark voltage are crucial variables to
MRR and surface roughness. The validation of the proposed method is conducted by
experiment resulting from Taguchi technique [17, 18]. Different from the previously
mentioned studies, Tomadi et al. [7] purposely chose three output responses at the same
time for their studies, e.g. reducing surface roughness and electrode wear and increasing
MRR. The design of experiments of (DOE) technique and ANOVA used statistical
software. The results exhibit that the main effects of the peak current and pulse on time
are dominant compared with the remaining parameters. MRR and surface roughness
have been typically selected as the output responses at the same time. The used methods
for optimizing have been adopted as mathematical model [5] or combining Taguchi
method and GRA [9, 10]. Shather et al. [15] carried out the WEDM process of medium
carbon steel 33 where pulse on time, pulse off time, and servo feed rate are considered
as input parameters. The multi-objective functions in this study are surface finish and
MRR. It was found that pulse on time has the greatest impact on surface roughness while
servo feed rate only affects crucially on MRR.
The results in literature have shown that there has been little research work dealing
with machining circular profile of hardened SKD11 steel, specially by using WEDM
process. In this study, optimization process to find a set of WEDM parameters will be
conducted in which maximizing the material removal rate is chosen as the response.
Taguchi method is used to setup the orthogonal array of experiments. Cutting voltage,
pulse on time, pulse off time, server voltage, wire feed, cutting speed are investigated
WEDM parameters as above mentioned. The experiments for confirming are executed
based on the statistical analysis.

2 Experimental Work

In order to evaluate the impact of machining parameters on MRR, six process parameters
such as cutting voltage, pulse on time, pulse off time, servor voltage, wire feed, cutting
speed are purposely selected. The investigated levels of the mentioned parameters are
detailed in Table 1. The number of degrees of freedom of the model is 11. The experi-
mental design is based on the survey levels of input parameter where Taguchi method is
applied with orthogonal array of L18 (21 + 35 ). This gives the condition that the number
of freedom degree is minor to those of experiments. The measured surface roughness
and the matrix of the experiment are presented in Table 2. The Wire-EDM process setup
 Influence of WEDM Process Parameters on Material Removal Rate … 615

can be referred in Fig. 1. The experimental setup included the following items: a Fanuc
Robocut -1 iA EDM machine (Fig. 1); brass wire with a diameter of 0.25 (mm) (Taiwan);
workpiece material using SKD11 tool steel and dielectric fluid or deionized water.
Table 1. Input parameters and the investigated levels

N0 Input parameters Notation Unit Levels

1 2 3
1 Cutting voltage VM V 6 9 12
2 Pulse on time Ton μs 6 9 12
3 Pulse off time Toff μs 8 12 16
4 Server voltage SV V 24 29 34
5 Wire feed WF mm/min 8 10 12
6 Cutting speed SPD m/min 2 4 –

Table 2. Input parameters and output results

Exp. No. Input parameters MRS (mm2 /min) Mean S/N

VM Ton Toff SV WF SPD Trial 1 Trial 2 Trial 3
1 6 6 8 24 8 2 8.000 8.152 8.241 8.1310 18.2009
2 9 6 12 29 10 2 28.066 27.864 27.991 27.9737 28.9349
3 12 6 16 34 12 2 32.897 33.056 33.203 33.0520 30.3838
4 9 9 8 24 10 2 15.007 14.987 15.149 15.0477 23.5491
5 12 9 12 29 12 2 30.431 30.683 30.952 30.6887 29.7389
6 6 9 16 34 8 2 31.825 32.019 32.343 32.0623 30.1193
7 12 12 8 29 8 2 26.366 26.135 25.719 26.0733 28.3226
8 6 12 12 34 10 2 41.423 41.173 40.964 41.1867 32.2949
9 9 12 16 24 12 2 26.061 25.986 26.172 26.0730 28.3237
10 9 6 8 34 12 4 11.387 11.631 11.375 11.4643 21.1856
11 12 6 12 24 8 4 20.664 21.056 21.149 20.9563 26.4250
12 6 6 16 29 10 4 28.851 28.352 28.649 28.6173 29.1319
13 6 9 8 29 12 4 23.304 22.982 22.864 23.0500 27.2526
14 9 9 12 34 8 4 39.433 38.682 38.144 38.7530 31.7637
15 12 9 16 24 10 4 27.758 28.032 27.988 27.9260 28.9199
16 12 12 8 34 10 4 29.926 30.363 30.141 30.1433 29.5834
17 6 12 12 24 12 4 39.092 38.821 38.619 38.8440 31.7862
18 9 12 16 29 8 4 26.213 26.011 26.694 26.3060 28.3996
616 N. A. Tuan et al.

Fig. 1. Wire-EDM machine used in the experiment

3 Results and Analysis

In order to evaluate the impact of input parameters on MRR, analysis of variance

(ANOVA) is used through Minitab@19 in which objective function is maximum material
removal rate.
 n 
1 1
S/N = −10 log10 (1)
n y2
i=1 i

The analyzed results of ANOVA process are exhibited in Table 3. It is observed

that Toff has the most influence on MRR (46.21%), followed by that of Ton (21.39%),
SV (15.08%), VM (5.04%) and WF (2.14%). The effect of SPD can be ignored when
showing the percentage of 0.14% (Table 4).
The influence order of input parameters on the response is detailed in Table 3. The
results presented in Table reveal that the effect of input parameters on the S/N ratio is in
the order of Toff , Ton , SV, VM, WF and SPD. Another way to show the effect of input
parameters on the response is using Main Effects Plot presented in Fig. 2.
 Influence of WEDM Process Parameters on Material Removal Rate … 617

Table 3. Analysis of variance for means

Source DF Seq SS Adj SS Adj MS F P C (%)

SPD 1 21.27 21.27 21.27 0.71 0.431 0.14
Ton 2 147.52 147.52 73.76 2.47 0.165 21.39
Toff 2 433.96 433.96 216.98 7.28 0.025 46.21
SV 2 353.26 353.26 176.63 5.92 0.038 15.08
WF 2 120.39 120.39 60.20 2.02 0.214 2.14
VM 2 120.25 120.25 60.12 2.02 0.214 5.04
Residual 6 178.93 178.93 29.82 10.01
error
Total 17 1375.57
Model summary
S R-Sq R-Sq(adj)
4.7699 90.00% 71.65%

Table 4. Influence order of the input parameters on the S/N ratio

Level SPD Ton Toff SV WF VM

1 26.70 21.70 18.98 22.83 25.38 28.65
2 27.34 27.92 33.07 27.12 28.48 24.27
3 31.44 29.01 31.11 27.20 28.14
Delta 0.64 9.74 14.08 8.28 3.10 4.38
Rank 6 2 1 3 5 4

Figure 2 shows the influence of input parameters viz., the investigated levels. Indeed,
when cutting speed (SPD) increases from level 2 to level 4 (m/min), there is an increase
in S/N ratio. In case of Ton parameter, Ton increases from level 6 to level 12 (μs) making
S/N sharply increase. When Toff increases from 8 to 12 (μs), the MRS increases rapidly.
However, as Toff increases from 12 to 16 (μs), the MRS decreases. The effect of the
WF on the MRS is similar to that of Toff on MRS but with a milder degree. MRS and
SV have a positive relation. As SV increases from 24 to 34, MRS increases. Besides,
the relation of VM and MRS is as follows: When VM increases from 6 to 9 (V), MRS
decreases. However, as the VM rises from 9 to 12 (V), the MRS increases. However, in
order to determine the optimal cutting mode, it is necessary to consider the S/N ratio of
the process parameters. The material removal rate is as large as possible since the higher
the S/N value is, the smaller the interference effect is. The influence of input parameters
on the S/N values is clearly shown in Fig. 3.
618 N. A. Tuan et al.

Fig. 2. Influence of process parameters on MRS

Fig. 3. Effect of input parameters on S/N ratios

It is seen that the information provides the optimal set of process parameter for
Wire-EDM when cutting curve profile R7 of hardened SKD11 steel. In this case, MRR
reaches the maximum value as shown in Table 5.
 Influence of WEDM Process Parameters on Material Removal Rate … 619

Table 5. Optimum machining condition

N0. Input parameters Notation Unit Optimum investigated Optimum values of

level response
1 Cutting speed SPD m/min 2 4
2 Pulse on time Ton μs 3 12
3 Pulse off time Toff μs 2 12
4 Cutting voltage VM V 3 12
5 Server voltage SV V 3 34
6 Wire feed WF mm/min 2 10

The value of material removal rate corresponding to a reasonable set of technolog-

ical parameters is calculated based on the prediction method using Minitab 19 soft-
ware. Calculated results are as shown in Table 6. Thus, through the prediction results in
Table 6, it shows that the MRR corresponding to the optimal set of parameters found in
the proposed experimental model is 36,0953 (mm2 /min). From the table of prediction
results, it is also shown that with 95% confidence, 95% of the removal rate obtained
with the optimal machining mode will be in the range of 26.6549 (mm2 /min) to 45.5358
(mm2 /min).

Table 6. Predicted MMR values corresponding to the optimal set of parameters

Variable Setting
SPD 4
Ton 12
Toff 12
SV 34
WF 10
VM 12
Prediction
Fit SE fit 95% CI 95% PI
36.0953 4.28919 (26.6549, 45.5358) (17.7672, 54.4235)

Evaluation of proposed experimental models

The proposed experimental model is evaluated through error distribution charts as shown
in Fig. 4. The information in Fig. 4 reveals that in the normal error distribution chart, the
errors of the experimental points corresponding to the blue points on the graph are close to
the normal distribution line (red solid line). It shows that the error level is very small. The
histogram of the frequency of errors shows that the errors occur in the range −5 to 5 and
620 N. A. Tuan et al.

Fig. 4. Error assessment distribution histograms.

Fig. 5. Probability graph of the fit of the empirical model to the MRR

follow the normal distribution. The remaining two graphs show the random distribution
of experimental errors, which means that the built model is largely influenced by the
selected input parameters and is not affected by the order of the experiments (Fig. 5).
 Influence of WEDM Process Parameters on Material Removal Rate … 621

The suitability of the experimental model and the optimal set of parameters was
checked by Anderson-Darling method using Minitab 19 software in Fig. 4. Figure 4
shows that the data corresponding to the experimental points (blue dots) are all within
the bounded area by 2 upper and lower limits with 95% significance level, and the P
value is 0.334, and is larger than α = 0.05. This indicates that the suggested experimental
model is consistent with the above significance level.

4 Conclusions
Machining curve profile of hardened SKD11 steel by using WEDM process is presented
in this study. An optimization process is carried out to find a set of WEDM parameters
which can maximize the material removal rate. Taguchi method and ANOVA are utilized
for optimizing. There are conclusions presented as flows:

• ANOVA results used to evaluate the significant level of input parameters on the
response reveal that Pulse off time (Toff ) has the most influence on MRR with 46.21%,
followed by Pulse on time with 21.39%. Inversely, SPD has minor effect on the
response.
• The optimal set of process parameters for WEDM when cutting curve profile R7 of
hardened SKD11 steel includes SPD = 4 m/min, Ton = 12 μs, Toff = 12, VM = 12 V,
SV = 34 V, and WF = 10 mm/min.
• Anderson-Darling method shows that the proposed optimization process is an efficient
method and it can be used to find the optimal values of the reasonable machining
mode in order to find the maximum MRR. This method can be extended to WEDM
machining of other hard work materials.

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Somashekhar, K.P., Mathew, J., Ramachandran, N.: A feasibility approach by simulated
annealing on optimization of micro-wire electric discharge machining parameters. Int. J.
Adv. Manuf. Technol. 61 (2012)
2. Ho, K.H., Newman, S.T., Rahimifard, S., Allen, R.D.: State of the art in wire electrical
discharge machining (WEDM). Int. J. Mach. Tools Manufact. 44 (2004)
3. Nithin Aravind, S.S.: Optimization of metal removal rate and surface roughness on wire EDM
using Taguchi method. IEEE-Int. Conf. Adv. Eng. Sci. Manag. (2012)
4. Sivakiran, S., Bhaskar Reddy, C., Eswara reddy, C.: Effect of process parameters on in wire
electrical discharge machining of En31 Steel. Int. J. Eng. Res. Appl. 2 (2012)
5. Scott, D.A.N., Boyina, S., Rajurkar, K.P.: Analysis and optimization of parameter com-
binations in wire electrical discharge machining. Int. J. Prod. Res. 29(11), 2189–2207
(1991)
6. H. Singh*, R.G., Effects of process parameters on material removal rate in WEDM, Journal
of Achivements in Materials and Manufacturing Engineering, 2009. 32(1)
622 N. A. Tuan et al.

7. S.H.Tomadi, M.A.H., Z. Hamedon, Member, IAENG R.Daud, A.G.Khalid, Analysis of the

Influence of EDM Parameters on Surface Quality, Material Removal Rate and Electrode Wear
of Tungsten Carbide, International MultiConference of Engineers and Computer Scientists,
2009. 2009
8. Huang, J.T., Liao, Y.S.: Optimization of machining parameters of Wire-EDM based on Grey
relational and statistical analyses. Int. J. Prod. Res. 41(8), 1707–1720 (2010)
9. Bobbili, R., Madhu, V., Gogia, A.K.: Effect of wire-EDM machining parameters on surface
roughness and material removal rate of high strength armor steel. Mater. Manuf. Processes
28(4), 364–368 (2013)
10. Garg, M.P., Jain, A., Bhushan, G.: Multi-objective optimization of process parameters in wire
electric discharge machining of Ti-6-2-4-2 alloy. Arab. J. Sci. Eng. 39(2), 1465–1476 (2013).
https://doi.org/10.1007/s13369-013-0715-x
11. Tonday, H.R., Tigga, A.M.: Analysis of effects of cutting parameters of wire electrical dis-
charge machining on material removal rate and surface integrity. In: National Conference on
Processing and Characterization of Materials (2013)
12. Fairuz, M.A., Hafiezal, M.M., Hussin, R., Adam, S.A. and Aiman, A.F.: Performance Study of
WEDMed on Inconel 718 by Using Response Surface Methodology. Int. Rev. Mech. Eng.7(4)
(2013)
13. Kumar, A., Kumar, V., Kumar, J.: Semi-empirical model on MRR and overcut in WEDM
process of pure titanium using multi-objective desirability approach. J. Braz. Soc. Mech. Sci.
Eng. 37(2), 689–721 (2014). https://doi.org/10.1007/s40430-014-0208-1
14. Parveen, Kr., Saini, M.V.: Experimental investigation of wire EDM process parameters on
MRR of Ti-6al-4v alloy. Int. J. Innov. Technol. Explor. Eng. 4(5) (2014)
15. Saad Kareem Shather, D.A.K.N.: Influence of intervening variables on surface roughness and
metal removal rate during WEDM process. Eng. & Tech. J. 32(5) (2014)
16. Venkat Lalit Shukla, D.S.K.S.Y.: Effect of wire cut EDM parameters on material removal
rate. Int. Res. J. Eng. Technol. 7(9) (2020)
17. Ali, M., Ibrahim, A.: Effect of machining parameters on surface roughness and metal removal
rate for AISI 310 L stainless steel in WEDM. Eng. Technol. J. 40(1), 181–188 (2022)
18. Perumal, A., et al.: Multiresponse optimization of wire electrical discharge machining parame-
ters for Ti-6Al-2Sn-4Zr-2Mo (α-β) alloy using Taguchi-grey relational approach. Adv. Mater.
Sci. Eng.2022, 1–13 (2022)
 Iterative Learning Control for Completely
Uncertain CSTR with Matched Disturbance

Trung Thanh Cao(B) , Nam Hoai Nguyen, and Phuoc Doan Nguyen

Hanoi University of Science and Technology, Hanoi, Vietnam

trung.caothanh@hust.edu.vn

Abstract. This paper proposes an intelligent controller for output tracking con-
trol for a completely uncertain two-state continuous stirred tank reactor (CSTR)
with disturbance on input. This control method is established by combining the
concept of iterative learning control (ILC) and a model-free disturbance estimator
for compensating purpose. Hence, the created controller does not use the orig-
inal nonlinear model of CSTR or linearize it around operating points as usual.
In consequence, all unexpected performances, which are inevitability caused by
switching the control between linear subsystems, are prevented. The effectiveness
of proposed approach had been authenticated by an illustrative simulation.

Keywords: Iterative learning control · Continuous stirred tank reactor ·

Disturbance estimation · Lagrange interpolation

1 First Section Introduction

A continuous stirred tank reactor (CSTR) is a batch reactor, which is equipped with a
mixed tank, pipes, tubes, and impeller device to efficiently produce chemical resultants
as desired [1, 2]. Figure 1 illustrates the physical framework of a CSTR, in which a
chemical material A is irreversibly converting into a material B by adjusting the flow
rate u(t) of the coolant, which is additionally disturbed by an uncertain disturbance d (t).
The chemical reaction A → B occurring in the stirred tank is demonstrated mainly
through the changing of its concentration Cx and temperature Tx .
The standard CSTR as depicted in Fig. 1 was widely used in chemical manufacture
industry [3]. Hence, it has been intensively studied and many control methods have
been available to ensure that their obtained closed-loop control systems will achieve
the required accuracy of output tracking errors. Traditionally, a CSTR was controlled
by using linear model-based design techniques [2, 3]. Because the usage of linear tech-
niques with only one representative linear model was very restrictive if the controlled
system is highly nonlinear, the utilization of multiple time invariant linear (LTI) models
had been investigated. These series of LTI models are obtained through a linear approx-
imation of the original nonlinear model around different operating points. Accordingly,
some appropriate control approaches have been also proposed for them, for example
the adaptive PID in [4–6], the gain-scheduling and exact linearization method in [7],
the multiple model adaptive control in [6, 8] and the model predictive control (MPC)

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 623–632, 2023.
https://doi.org/10.1007/978-3-031-22200-9_68
624 T. T. Cao et al.

techniques in [10]. Since these methods are created on a chain of approximated linear,
time-invariant sub-models around series of operating points of the original nonlinear
model, the unwanted oscillation would be inevitably happening there as the controller
switches from one LTI subsystem to another one. This is the main disadvantage of
multiple LTI model-based methods. Furthermore, they were limited to the operating
range.
Hence, for the scenario that CSTR is not only highly nonlinear, but also there
are external disturbances and model’s uncertainty affecting control performance, some
advanced controllers have been proposed to improve the system robustness, such as neu-
ral network-based approaches [9, 10] and sliding mode control (SMC) [11]. However,
one unavoidable disadvantage of the SMC is its inherent chattering phenomena.

Fig. 1. The standard structure of a CSTR [14].

An ordinary approach to overcome all disadvantages above of model-based control

methods is the usage of intelligent control concept, where the mathematical model of
controlled plants is needless. The iterative learning control (ILC) is one of the intelli-
gent concepts, which is suitable for batch process like CSTR [12, 13]. However, ILC
cannot be applied for all batch processes [13]. For such instable processes it often needs
a supplemental model-free controller to make the application of ILC possible. In our
recently published work [14] a model-free compensator for stabilizing uncertain nonlin-
ear systems has been proposed. It was used there as a supplement controller to tracking
control CSTR through the ILC concept, where almost CSTR parameters are assumed to
be unknown (except the gain factor of input δ).
As an extension, the article will modify this model-free compensator proposed in
[14], so that it would be applicable even for the circumstance of the uncertain gain
factor δ. We will call it the completely uncertain CSTR. Moreover, in this article the
Lagrange interpolation will be applied for estimating the derivative of a smooth function,
instead of backward differentiation as having done in [14], as well as the usage of a
modified approach presented in [15] for adaptively determining the learning parameters,
to improve the convergence rate of applied ILC controller.
 Iterative Learning Control for Completely Uncertain CSTR with Matched … 625

2 Main Results
2.1 Problem Formulation and Notations

Consider a CSTR illustrated in Fig. 1, which is described mathematically by [6]

   
−φx1 κ(x2 ) + q(x1f − x1 ) x1
ẋ = ,x= . (1)
βφx1 κ(x2 ) − (q + δ)x2 + qx2f + δ[u + d ] x2

where φ, q, β, δ, x1f , x2f are non-negative constant parameters; x1 , x2 represent system

states the concentration Cx and temperature Tx of the chemical material in the stirred
tank, respectively, u(t) is adjustable flow rate of the coolant (input) and
 
x2
κ(x2 ) = exp  witha positive parameter λ. (2)
1 + x2 λ

In the model all unwanted changes in plant parameters and inevitable effect of exoge-
nous signals on CSTR’s behavior had been captured entirely by a time function d (t) as
a matched disturbance.
The control aim here is an intelligent controller u(x) to be designed such that the
interesting output y = x1 or y = x2 will converge to a desired repetitive reference r(t),
where d (t) is time-varying disturbance and all constant parameters φ, q, β, δ, x1f , x2f , λ,
including the gain factor δ of input, are considered to be unknown. For solving this
control problem, the Lagrange interpolation technique will be used as follows. Con-
sider a vector of timecontinuous functions x(t). Denote m its collected values by
χ = x(t1 ), . . . , x(tm ) , then x(t) would be recovered approximately from χ with

m
w(t) (m+1)
x(t) = x(ti )Li (t) + x (ς ) (3)
(m + 1)!
i=i

where ς ∈ [t1 ,tm ] and

m
t − tj

m
Li (t) = , w(t) = (t − ti ). (4)
t i − tj
j=1,j=i i=1

The functions Li (t), i = 1, . . . , m are called the Lagrange polynomials.

Finally, in this article, the identity matrix of dimension n × n is symbolized by In ,
the n × n matrix with all zero entries is denoted with 0n . A column or line vector of
real numbers a1 , . . . , an is indicated by a or aT , respectively, and the column vector
  of
zeros is designated with 0. Any p-norm, 1 < p < ∞, of a is indicated with a. An
induced norm of a matrix A is denoted with A. The set of real number is symbolized
by R.
626 T. T. Cao et al.

2.2 Model-Free Compensation of Lumped Disturbance via Lagrange

Interpolation

Under the consideration that all system parameters φ, q, β, δ, x1f , x2f of CSTR are
unknown, the original model (1) of it would be expressed correspondingly as an uncertain
one as below
   
−θ1 x1 κ̂(x2 ) − θ2 x1 + θ3 x2
ẋ = , κ̂(x2 ) = exp (5)
−αx2 + u + η 1 + θ6 x2

where α > 0 is an arbitrarily chosen constant for ILC being applicable to CSTR, all
θ1 = φ, θ2 = q, θ3 = qx1f , θ4 = βφ, θ5 = q + δ, θ6 = λ−1 , θ7 = δ, θ8 = qx2f are
new assigned uncertainties for unknown system parameters, and

η = θ4 x1 κ(x2 ) + (α + θ5 )x2 + (θ7 − 1)u + θ7 d + θ8 (6)

is the lumped disturbance (summarized disturbance).

It is already shown in [14] that the ILC concept will be applicable for tracking control
for the modified system (5) if there is η = 0. Hence our task is now to estimate η by η̂
for compensating it through u = v − η̂.
In following, the lumped η(t) will be estimated in stepwise mode. It means the
obtained estimation result η̂(t) is stepwise constant η̂(t) = η̂j for all tj ≤ t < tj+1 .
Furthermore, it will be carried out with tj = jTs for all j = 0, 1, . . . where 0 < Ts  1
is an arbitrarily small update time. Hence, we have tj+1 − tj = Ts , ∀j.
Take the first derivative of both sides of Lagrange interpolation (3) at the current
time instant tj = jTs , we receive the  approximation of derivate ẋ(t) at this time instant
from recently collected data χj = xj−m+1 , . . . , xj , xi = x(iTs ) as follows

m
ẇ(t) (m+1)
ẋj = xj−m+i DiL (tj ) + x (ς )
(m + 1)!
i=1
 (7)

m
d Li (t) 
L L
≈ xj−m+i Di (tj ) where Di (tj ) =
dt t=jTs
i=1

This approximation is acceptable, because the last term on the right-hand side of (7)
is proportional to Tsm and hence it is very small. It means that we can ignore it. Moreover,
the smaller Ts is chosen, the more precise this approximation will be.
Substitute (7) in we get with u = v − η̂

ϕ(χ2,j ) ≈ −αx2,j + vj − η̂j−1 + ηj (8)

where x2,j = x2 (jTs ), vj = v(jTs ), η̂j−1 = η̂((j − 1)Ts ), ηj = η(jTs ) and

m
ϕ(χ2,j ) = x2,j−m+i DiL (tj ) (9)
i=1
 Iterative Learning Control for Completely Uncertain CSTR with Matched … 627

The quadratic optimization problem ε2 → min, where ε is the approximation error

between two sides of, provides uniquely the following solution η̂j

η̂j = η̂j−1 + αx2,j − vj + ϕ(χ2,j ) (10)

and this will be used for approximating iteratively η̂j from the previous value η̂j−1 and
from recently collected data x2,j−m+1 , . . . , x2,j , vj . The initial value η̂0 could be assigned
arbitrarily.

2.3 Iterative Learning Controller Design

After the lumped disturbance η had been compensated by u = v − η̂, where η̂ is obtained
iteratively from (10), the original CSTR (1) becomes
 
−θ1 x1 κ̂(x2 ) − θ2 x1 + θ3
ẋ = with output y = x1 or y = x2 , (11)
−αx2 + v + δη

where δη = η − η̂ denotes the remaining estimation error of η. It is obvious that the

smaller Ts is chosen, the smaller δη will be. Moreover, it is realizable from (7) that if the
(m+1)
second state x2 (t) satisfies x2 (ς ) = 0, then we have δη (jTs ) = 0, ∀j.
For completing the control problem, in following the compensated system will be
tracked y → r by using an ILC controller of D-Type as below

vk+1 (i) = vk (i) + Kk ek (i + 1) (12)

with vk (i) = v(tki ), yk (i) = y(tki ), tki =kT + iTs , ek (i) = r(i) − yk (i) for all k =
0, 1, . . . and i = 0, 1, . . . , N −1, N = T Ts . The time constant T denotes the repetitive
working interval of the batch-working system. Correspondingly, the desired reference
r(t) must be periodic with the same period T either. The index k in (12) indicates the
k th working period of the system, which is often called “a trial” in ILC concept. We
see from (12) that ILC uses the previous information of the system, i.e. of vk (i) and
ek (i) collected during the whole k th trial with i = 0, 1, . . . , N − 1, for updating the
control signal vk+1 (i), which will be used for the next trial. In addition, v(t) obtained
from (12) is a stepwise constant function with v(t) = vk (i) and vk (N ) = vk+1 (0), where
kT + iTs ≤ t < kT + (i + 1)Ts .
Based on method presented in [15], hereby the learning parameter Kk in will be
determined model-freely through the modified optimization problem.
 
Kk = arg min Kε k + (Kmax − K)εk−1 , k ≥ 2 (13)
0≤K≤Kmax

where ε l = (el (0), . . . , el (N − 1))T denotes the vector of all tracking errors belong to
the trial l. This modified Eq. (13) has two arbitrarily chosen initial values K0 , K1 and a
constraint Kmax , what would improve the learning convergence rate of ILC.
628 T. T. Cao et al.

3 Numerical Simulation
To illustrate the tracking performance of the proposed intelligent controller we will
hereafter apply it to control the completely uncertain CSTR (5). It is supposed that this
uncertain CSTR is working repetitively with T = 100s.
If the interesting output is y = x2 , i.e. the temperature Tx , then by implementing the
proposed intelligent controller with.

– x0 = (CA , TA )T , CA = 0.1, TA = 0.5 as starting states of each trial,

– TB = x2 (T ) = 5 as the end point to reach for every trial,
– Ts = 0.02s for collecting data,
– Kmax = 15 for determining model-freely the adaptive learning parameter Kk via
Eq. (13) under initial values K0 = K1 = 12.8,
– α = 5 for ILC being applicable,
– m = 2 for evaluating Lagrange interpolation,
– the desired reference
   
TA + (Tm − TA ) 1 − exp(−0.2t) if 0 ≤ t < T 2
r(t) =    
Tm + (TB − Tm ) 1 − exp(−0.2(t − T 2) if 0 ≤ t < T 2

where TA < Tm = 2 < TB ,

– the vague adjustment of coolant flow rate u(t) as matched disturbance

d (t) = x1 sin(θ9 x2 ) + θ10 x2 cos(x1 )

where θ9 , θ10 are considered randomly,

We obtain simulation results as exhibited in Figs. 2, 3, and 4, which are attained by
using the simulation program ICERA2022.m given in Appendix.

a) b)

Fig. 2. Output tracking performance y = x1 → r: a) after 5 trials (left) and b) after 10 trials
(right).

As expected, Fig. 2 shows that the system output y = x2 had tracked brilliantly its
reference r(t) in spite of completely unknown parameters 0.2 ≤ θi ≤ 1.2, i = 1, . . . , 8
of the CSTR and matched disturbance d (t). Concretely, after 10 trials both system
output y and the reference r are virtually overlapped. The remaining tracking error after
10 trials is approximately 0.028. Moreover, the proposed intelligent controller does not
 Iterative Learning Control for Completely Uncertain CSTR with Matched … 629

a) b)

Fig. 3. Simulation results: a) Corresponding response of concentration x1 = Cx (left) and b)

Adaptive learning parameter Kk .

cause oscillations around non-smooth point of r(t), which occurs definitely by using
model-based controllers.
Figure 3 demonstrates the corresponding behavior of material concentration x1 = Cx
and the adaptive learning parameter Kk depends on number of trials, which generates a
more quickly convergent tracking behavior to zero than by using the original formula
given in [15].

Fig. 4. Estimation result of lumped disturbance during 0.15s at the beginning of simulation.

Figure 4 illustrates the effectiveness of the integrated disturbance compensator. As

expected, just after a short time of the beginning (≈ 0.15s) of whole simulation time
(1000s) the lumped disturbance η and its estimated result η̂ are almost overlapping.

4 Conclusions

The article proposed a state feedback regulator for output tracking control of a com-
pletely uncertain CSTR without using both its original nonlinear model and multiple
linear models as usual. Therefore, it does not produce unwanted oscillations during the
630 T. T. Cao et al.

control. This controller was created through combining the model-free estimator created
with Lagrange interpolation for compensating the lumped disturbance and an D-Type
ILC controller. The disturbance compensation control methods have been utilized to
practically stabilize the nonlinear CSTR for applying the ILC controller. The D-Type
iterative learning controller is applied for robust tracking control of the CSTR output to
a designed reference. In addition, the here applied ILC controller possesses an adaptive
learning parameter, which is determined model-freely from the collected data of the
plant. Hence, the convergence rate of tracking error to zero was improved.
The illustrated simulation has authenticated the excellent control performance of
proposed controller as expected.

Appendix

The source code of simulation program ICERA2022.m written in MatLab is as below.

% Modified D-type ILC for completely uncertain CSTR

clc; clear;
global alph v i etah th eta
op=optimoptions(@fmincon,'algorithm','sqp');
alph=5; T=100; Ts=0.02; N=T/Ts; % simulation values
Ta=0.5; Ca=0.1; Tm=2; Tb=5; Kmax=15; % simulating parameters
r=[]; tp=[]; pd=[]; pdh=[]; td=[]; err=[]; tk=[]; Kp=[];
for i=1:10 % random constants
th(i)=0.2+rand;
end
for k=1:N % create the reference
if k<=N/2
r=[r;Ta+(Tm-Ta)*(1-exp(-0.2*Ts*(k-1)))];
else
r=[r;Tm+(Tb-Tm)*(1-exp(-0.2*Ts*(k-N/2-1)))];
end
tp=[tp;(k-1)*Ts]; % control time
end
v=r; e=r; etah=0; M=10; K=12.8; % initial values
for k=1:M % iterative trials
x0=[Ta;Ca]; x1=[]; x2=[]; t0=0; k
for i=1:N
[t,x]= ode45(@funICERA2022,[t0 t0+Ts],x0); % to the CSTR
h= -alph*x(end,2);
etah= ((x(end,2)-x0(2))/Ts-h)/th(7)-v(i)+etah;
pd= [pd eta]; pdh= [pdh etah];
x1= [x1;x(end,1)]; % output is Cx (concentration)
x2= [x2;x(end,2)]; % output is Tx (temperature)
t0= t(end); x0= x(end,:);
end
if k>2
K=fmincon(@(z) norm(z*(r-x2)+(Kmax-z)*e),
0.5,[],[],[],[],0,Kmax,[],op)
end
 Iterative Learning Control for Completely Uncertain CSTR with Matched … 631

Kp=[Kp;K]; err=[err;norm(e)]; tk=[tk;k-1];

td=[td;tp+(k-1)*T]; e=r-x2;
v=v+K*[e(2:end);e(end)]; % D learning
if k==M/2 ym=x2; em=e; end
end
figure(1);
plot(tp,r,tp,ym,tp,em);legend('Reference','Output','Error');
figure(2);
plot(tp,r,tp,x2,tp,e);legend('Reference','Output','Error');
figure(3);
plot(td,pd,td,pdh); legend('Disturbance','Estimation result');
figure(4); plot(tp,x1);legend('Concentration');
figure(5);
plot(tk,err,tk,Kp);legend('Tracking errors','Learning parameters');
The aforementioned simulation program uses following subprogram for the declaration
of CSTR dynamic.
% Dynamic of uncertain CSTR
function dx = funICERA2022(t,x)
global alph v i etah th eta
fK=exp(x(2)/(1+x(2)*th(6))); h=-alph*x(2);
f=[-th(1)*x(1)*fK-th(2)*x(1)-th(3);h];
u=v(i)-etah; % control signal
d=x(1)*sin(th(9)*x(2))+th(10)*x(2)*cos(x(1)); % disturbance
% lumped disturbance
eta=th(7)*d+(th(7)-1)*u+(alph+th(5))*x(2)+th(4)*x(1)*fK+th(8);
dx=f+[0;th(7)]*(u+eta);
end

The aforementioned simulation program uses following subprogram for the decla-
ration of CSTR dynamic.

% Dynamic of uncertain CSTR

function dx = funICERA2022(t,x)
global alph v i etah th eta
fK=exp(x(2)/(1+x(2)*th(6))); h=-alph*x(2);
f=[-th(1)*x(1)*fK-th(2)*x(1)-th(3);h];
u=v(i)-etah; % control signal
d=x(1)*sin(th(9)*x(2))+th(10)*x(2)*cos(x(1)); % disturbance
% lumped disturbance
eta=th(7)*d+(th(7)-1)*u+(alph+th(5))*x(2)+th(4)*x(1)*fK+th(8);
dx=f+[0;th(7)]*(u+eta);
end

References
1. M.E. Davis and R.J. Davis. Fundamentals of chemical reaction engineering. McGraw-Hill
Boston, 2003
2. L.D. Schmidt. The engineering of chemical reactions. New York-Oxford University Press,
1998
3. M. Morari and E. Zariou. Robust process control. Englewood CliMs, NJ, Prentice-Hall, 1989
4. Prakash, J., Srinivasan, K.: Design of nonlinear PID controller and nonlinear model predictive
controller for a continuous stirred tank reactor. ISA Trans. 48(3), 273–282 (2009)
5. Marin, A., Hernandez, J.A., Jimenez, J.A.: Tuning multivariable optimal PID controller for a
continuous stirred tank reactor using an evolutionary algorithm. IEEE Latin America Trans.
16(2), 422–427 (2018)
632 T. T. Cao et al.

6. Jingjing, D., Chunyue, S., Ping, L.: Modeling and control of a continuous stirred tank reactor
based on a mixed logical dynamical mode. Chin. J. Chem. Eng. 15(4), 533–538 (2007)
7. Doyle, F.J., Kwatra, H.P., Schwaber, J.S.: Dynamic gain scheduled process control. Chem.
Eng. Sci. 53(15), 2675–2690 (1998)
8. Ciccio, M.P.D., Bottini, M., Pepe, P.: Digital control of a continuous stirred tank reactor,
Mathematical Problems in Engineering, 2011. Article ID 439785, 1–18 (2011)
9. Deepa, S.N., Baranilingesan, I.: Optimized deep learning neural network predictive controller
for continuous stirred tank reactor. Comput. Electr. Eng. 71, 782–797 (2018)
10. Wang, G., Jia, Q.-S., Qiao, J., Bi, J., Zhou, M.: Deep learning-based model predictive control
for continuous stirred-tank reactor system. IEEE Trans. Neur. Netw. Lear. Syst. 32(8), 3643–
3652 (2021)
11. Sinha, A., Mishra, R.K.: Control of a nonlinear continuous stirred tank reactor via event
triggered sliding modes. Chem. Eng. Sci. 187, 52–59 (2018)
12. Wang, Y., et al.: Survey on iterative learning control, repetitive control and run to run control.
J. Process Control 19(10), 589–1600 (2009)
13. Bristow, D.A., et al.: A survey of iterative learning control: A learning-based method for
high-performance tracking control. IEEE Control Syst. Mag. 26, 96–114 (2006)
14. T.T. Cao, P.D. Nguyen, N.H. Nguyen and H.T. Nguyen. An indirect iterative learning controller
for nonlinear systems with mismatched uncertainties and matched disturbances. Int. Journal
of Systems Science, 2022
15. Nguyen, P.D., Nguyen, N.H.: An intelligent parameter determination approach in iterative
learning control. Eur. J. Control. 61, 91–100 (2021)
 Load Sharing Control of Two Ac Motors
with a Common Stiff Shaft by Using an Adaptive
Fuzzy Controller

Nguyen Trung Thanh and Dang Danh Hoang(B)

Thai Nguyen University of Technology, Hanoi, Vietnam

hoangktd1977@tnut.edu.vn

Abstract. The drive system using the large power AC motor to meet the heavy
load requirements leads to difficulty both economically and technically. Using
the large power means high manufacturing costs for a large-capacity inverter
(controlled inverter for 3-phase synchronous motor). Therefore, it is necessary
to propose solutions to control the drive system using two AC motors with a
standard stiff shaft to reduce each motor’s power and reuse the available motors
in the system. This article presents the fuzzy adaptive control method capable of
controlling load sharing for the drive system using two 3-phase synchronous AC
motors with a common stiff shaft. This method meets the control performance
requirements for the torque and speed of the drive system.

Keywords: AC motor · PID controller · Fuzzy logic · Adaptive control · Load

balance

1 Introduction

Today AC motors are widely used in industrial applications such as lifting machines,
production lines, pumps, crushers, etc. Speed and torque control frequency converters
for AC motors have also been developed almost perfectly to meet an AC motor’s start-
ing, speed, and torque control requirements. When upgrading the drive system for more
demanding applications, the usual solution is to replace the entire motor and inverter
system with a significantly larger capacity. This results in a much more expensive invest-
ment for the new drive system because the cost rate for the large-capacity drive system
is very high. In addition to transporting this engine from production to use and installing
them in the working position, it has many obstacles because of the weight and size
factors. On the other hand, this also leads to discarded or wasted old powertrains in the
system.
The solution to the above disadvantages when it comes to expanding the drive capac-
ity of the system is to supplement the extra power to be extended with an AC motor of
equal ability. The two motors are hard-coupled together. The power-sharing control strat-
egy will be applied to these two motors, each contributing a percentage of the required
power through the torque and current to respond to the load. The control ensures that the

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 633–648, 2023.
https://doi.org/10.1007/978-3-031-22200-9_69
634 N. T. Thanh and D. D. Hoang

current of each motor will vary in the same pattern as the required load capacity but with
different contribution ratios, such that the total power of the two motors is equal to the
required load capacity. This meets the load requirements while ensuring synchronous
speed.
Fuzzy control is utilized in many industrial processes, including electrical drives.
Electrical drive speed control is often made difficult by internal nonlinearities, diffi-
culties in the proper identification of system parameters, external load disturbances,
and unpredictable parameter variations. This problem arises especially when the load
machine inertia changes during the drive system operation [1–3]. The presence of sys-
tem nonlinearities increases the advantages of the use of fuzzy control as compared
with linear methods. Thus, the application of some more advanced techniques is favor-
able. Namely, the sliding mode control is robust to plant uncertainties and insensitive to
external disturbances [4–6]. It is widely used to ensure the good dynamic performance
of controlled systems. Another method, popular in recent years, is based on a fuzzy
neural network. Fuzzy-neural networks combine the capability of fuzzy reasoning and
the ability of neural networks to learn from processes [7–11]. In [6], the comparison
between mentioned methods was presented. It was proven that both ways ensure good
characteristics, yet the fuzzy-neural network requires less control effort.
Additionally, speed oscillations cause a decrease in the quality of the product and
can influence the stability of the control system. Several control structures have been
developed [12–15] to suppress torsional vibrations. The predictive control method based
on the properties of the converter is also implemented in [16–29]. Traditionally, a PI con-
troller is used for both current and speed regulation because of its simplicity. However,
the performance of the PI controller for speed regulation degrades under external dis-
turbances and machine parameter variations [30, 31]. Parameter adaptation of the ac
machine [32–36] is one way to overcome the problem outlined above because exter-
nal disturbances remain an issue. Another approach to addressing the issue is using
intelligent controllers like fuzzy logic [37–43], which is the direction adopted in this
work.
In this study, we propose a solution that is two small-capacity AC motors to perform
the same power supply for a large load. One engine acts as the master, and one motor
acts as the slave. The master motor current serves as the reference current for the slaver
motor current controller. The speed controller is common to both motors made in the
out-loop. For the current control problem, we use the fuzzy adaptive rule to adjust the
parameters of the current controller for a slaver motor.

2 Control Structure

The proposed solution of the paper is given with the cascade control structure of two
control loops as shown in Fig. 1, in which the outside is a common speed loop for both
motors with fixed parameters, and controllers. This speed controller can use the PID
control rule and two current control loops for two motors. One current control loop is
called a master current circuit, and the other is a slaver circuit, usually for larger power
motors often used as a master loop. The controller for these two loops uses a PI control
rule, where the parameters of the current controller for the master are kept constant.
 Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 635

The gain of the PI controller for the slaver loop is corrected by an adaptation rule based
on fuzzy logic. This fuzzy adaptation law fixes so that the current ratios of the master
and slave circuits are kept constant. The process of keeping these two currents in a ratio
maintains the power contribution of each motor to the load.

Fig. 1. The control structure of two AC motors with a common stiff

636 N. T. Thanh and D. D. Hoang

The fuzzy controller parameters will be corrected based on the current difference
between the two motors, as shown in Fig. 2 [44, 45]. With this structure during operation,
the first motor current is considered the model line. The second motor current always
sticks to motor line 1 with the smallest error. That means the armature flow of both
engines is always the same, which is also what we would expect. We have built up the
control principle diagram of two hard-coupled motors (in common stiff) with motor 1
as the model used to create Issue the adaptive rule for motor controller 2 as shown in
Fig. 2.

Fig. 2. Diagram of adaptive control principle of two ac motors with a common stiff

Consider a plan described by the equation in the Eq. (1)

dy
= −ay + bu (1)
dt
and the reference model is
dym
= −am ym + bm uc (2)
dt
The control law is u = K1 uc − K2 y and the control error is defined as ε = ym − y.
There are two unknown parameters K1 , K2 in the control law. We need to find out an
adaptive structure to adjust the parameters K1 , K2 to the desired value so that the error
ε approaches 0. To find this adaptive structure, we can use Lyapunov’s stability theory
or gradient method as follows.

2.1 Law of Adaptation According to LYApunov’s Theory

Assume bη > 0 and choose the Lyapunov function of the form:
⎛ ⎞
1
ε2 + (−bK2 − a + am )2
1⎜ bγ ⎟
V (ε, K1 , K2 ) = ⎜ ⎟ (3)
2⎝ 1 ⎠
+ (−bK1 + bm ) 2
bγ
 Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 637

The regulation of the parameters K1 , K2 so that ε → 0 is:

dK1 dK2
= γ uc e, = −γ ye (4)
dt dt
If only one parameter is variable, the adaptive parameter rule (4) becomes:
dK
= γ uc e (5)
dt

2.2 Law of Adaptation According to Gradient Method

Suppose K a parameter vector needs to be determined, and it relies on the error between
the plan’s output y and the model’s output ym . Criteria of error response of the controlled
system is
1 2
J (K) = ε →0 (6)
2
The adjustment rule in the direction of the gradient J (K) is:
dK ∂J ∂ε ∂y
= −γ = −γ ε = γε (7)
dt ∂K ∂K ∂K
There is generally no need for a perfect model in classical adaptive control. However,
the difference between the model and the object and its nonlinearity is within some limits.
If it exceeds this limit, the regulator will not work effectively anymore. The subsequent
research uses an adaptive fuzzy control system with a parallel model to overcome that.
This paper uses the adaptive method according to the Lyapunov and Gradient law
to adjust the adaptive stabilization coefficient K of the fuzzy controller. From Fig. 2,
we have an adaptive fuzzy control structure using the parallel model (MRAFC—Model
Reference Adaptive Fuzzy Controller) as shown in Fig. 3.

Fig. 3. Model Reference Adaptive Fuzzy Controller for adjusting the output gain

The control object has the transmission function G, the model has the transmission
function Gm , the fuzzy controller includes the basic fuzzy controller combined with the
638 N. T. Thanh and D. D. Hoang

K amplifier. It is necessary to find the regulation of the K coefficient similar to the fuzzy
control structure adapted to the linear pattern model so that the difference between the
model and the object reaches 0 (ε → 0).
The closed-loop system around equilibrium becomes linear with the equation for the
closed-loop as
KFG
y= uc (8)
1 + KFG
Suppose that y approaches ym , we can approximate Eq. (8) as Eq. (9)
KFG
≈ Gm (9)
1 + KFG
So we have
∂ε ∂y KFG e e
=− =− ≈ −Gm (10)
∂K ∂K 1 + KFG K K
Then the adaptive adjustment rule for the output amplification coefficient of FLC
according to Gradient can be determined as:
1 εe
K = γ Gm (11)
s K
or Eq. (12) based on the Lyapunov theory

dK 1
= γ uc e (12)
dt s
The parameter γ in (11) and (12) denote the convergence rate of the adaptive
algorithm (Fig. 4).
There is generally no need for a perfect model in classical adaptive control. However,
the difference between the model and the object and its nonlinearity is within some
limits. If it exceeds this limit, the regulator will not work effectively anymore. Based
on the design method of the adaptive fuzzy controller presented above, we have built
up a control structure for two AC motors systems with a common stiff shaft using the
adaptive fuzzy controller.

3 Control Design

3.1 PID Controller Design for the Motor 1

The control structure for the motor 1 is described in the Fig. 5

To design the controller, we choose the motor with the parameters given in the
following Table 1.
From the motor parameters, we can determine the parameters of the current con-
trollers with 2 components isd and isq of motor 1 based on Ziegler Nichols method as
follows:
 Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 639

Fig. 4. Control structure for two AC motors system with a common stiff shaft using the adaptive
fuzzy controller

Fig. 5. The diagram of a current control loop

The controller PID for the current isd is

1 1 + 0.5s
RIsd = 0.5 + = (13)
s s
and the current isq is formed by

10 10 + 100s
RIsq = 100 + = (14)
s s
640 N. T. Thanh and D. D. Hoang

Table 1. The physical motor parameter

Parameter Meanings Values

P Rated power 45 KW
- ,
Udm Rated phase voltages m´uc 220 V
f Grid frequency 50 Hz
Rs Stator coil resistance 0.08 
Lsd Axial inductance d and q 4.09 mH
Lsq 5.13 mH
J Moment of inertia 0.0002 kg m2
ψp Flux magnetic 0.51 Wb
pc Number of magnetic poles 1

The current controllers for motor 2 will be proposed as adaptive fuzzy controllers
and will be implemented later.

3.2 Design the Adaptive Fuzzy Controller

To design the adaptive fuzzy controller using the parallel model, we conduct two parts
as follows.
Part 1: Design of fuzzy controller.
We perform the steps of setting the fuzzy controller like a normal fuzzy controller:
– Step 1: Select membership functions.
– Step 2: Select the control rule.
– Step 3: Fuzzy operations in inference cells:

• Fuzzification
• Fuzzy inference
• Defuzzification

Part 2: Use adaptive rules based on Eq. (15) or (16) to adjust the fuzzy controller output.
The adaptive adjustment rule for the output amplification coefficient of FLC
according to Gradient can be determined as:
1 εe
K = γ Gm (15)
s K
Equation (16) is based on the Lyapunov theory
dK 1
= γ uc e (16)
dt s
 Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 641

3.3 Design the Speed Control Loop

Since the current loop acts very quickly compared to the reaction of the speed loop, for
simplicity, we can approximate the closed loop transfer function of the current loop into
a first-order model [44]:
K 1 1
G(s) = = ≈ (17)
TR s(Ts s + 1) + K 2Ts s + 2Ts s + 1
2 2 0.8s + 1

We design a common speed controller for 2 two motors:

5 5 + 0, 5s
RIsd = 0, 5 + = (18)
s s
Before designing the adaptive controller to adjust the current for the second motor,
we design the PID controller for motor 1. The algorithm to build the PID controller of the
paper is based on the Ziegler-Nichols method. Moreover, the fuzzy controller is adapted
according to the model of motor 2 to bring the closed system to the same as the closed
system of motor 1. Therefore, we consider the equivalent transfer function of the whole
motor closed system 2 to be the same as motor 1 when designing the speed controller
(Figs. 6, 7, 8, 9, 10, 11, 12, 13,14, and 15).

4 Simulation Results

Fig. 6. The simulation control diagram of 2 AC motor with common stiff using two control loops
(the speed and current loop)

The speed and current responses of the two motors show that when the rate changes,
the system still guarantees the quality.
642 N. T. Thanh and D. D. Hoang

Fig. 7. The simulation control diagram of motor 2 with adaptive fuzzy controllers (the motor 1
is supposed the reference model)

Fig. 8. The speed responses of two motors with no load

5 Conclusions

The article deals with two AC motor systems with joint stiff to ensure load balance by
the adaptive fuzzy controller with the parallel model. The simulation results show that
the quality of motor control is guaranteed when there is no load.
The control method proposed in this paper can be fully applied in practice. The
article deals with the design of the controllers for two AC motors systems with common
stiff to ensure load balance by the adaptive fuzzy controller according to the parallel
model. Simulation results show that the quality of system control has been guaranteed
through current values (load balancing) and the speed of the motor. The speed and current
 Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 643

Fig. 9. The isd current responses of two motors with no load

Fig. 10. The isq current responses of two motors with no load

responses of the two motors show that when the load change, the system still guarantees
the quality in this case.
644 N. T. Thanh and D. D. Hoang

Fig. 11. The torque responses with no load

Fig. 12. The speed responses of two motors at load increase

Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 645

Fig. 13. The isd current responses of two motors at load increase

Fig. 14. The isq current responses of two motors at load increase
646 N. T. Thanh and D. D. Hoang

Fig. 15. The torque responses at load increase

Acknowledgments. The authors gratefully acknowledge Thai Nguyen University of Technology

for supporting this work.

References
1. Orlowska-Kowalska, T., Szabat, K., Jaszczak, K.: The influence of parameters and structure
of PI-type fuzzy-logic controller on DC drive system dynamics. Fuzzy Sets Syst. 131(2),
251–264 (2002)
2. Orlowska-Kowalska, T., Szabat, K.: Optimization of fuzzy-logic speed controller for DC drive
system with elastic joints. IEEE Trans. Ind. Appl. 40(4), 1138–1144 (2004)
3. Pajchrowski, T., Zawirski, K.: Robust speed control of servodrive based on ANN. In: Pro-
ceedings of the IEEE International Symposium on Industrial Electronics, vol. 1, pp. 81–86.
ISIE 2005. IEEE (2005)
4. Ferrara, A., Incremona, G.P., Cucuzzella, M.: Advanced and Optimization Based Sliding
Mode Control: Theory and Applications. Society for Industrial and Applied Mathematics
(2019)
5. Koronki, P., Hashimoto, H., Utkin, V.: Direct torsion control of flexible shaft in an observer-
based discrete-time sliding mode. IEEE Trans. Industr. Electron. 45(2), 291–296 (1998)
6. Lin, F.J., Fung, R.F., Wai, R.J.: Comparison of sliding-mode and fuzzy neural network control
for motor-toggle servomechanism. IEEE/ASME Trans. Mechatron. 3(4), 302–318 (1998)
7. Lin, F.J., Lin, C.H., Shen, P.H.: Self-constructing fuzzy neural network speed controller for
permanent-magnet synchronous motor drive. IEEE Trans. Fuzzy Syst. 9(5), 751–759 (2001)
8. Barrero, F., Gonzalez, A., Torralba, A., Galvan, E., Franquelo, L.G.: Speed control of induction
motors using a novel fuzzy sliding-mode structure. IEEE Trans. Fuzzy Syst. 10(3), 375–383
(2002)
9. Jaszczak, K., Orlowska-Kowalska, T.: Adaptive fuzzy-neuro control of DC drive system. In:
Proc. 9th Int. Conf. Opt. Electr. Electron. Equipments, Vol. 3, pp. 55–62 (2004)
10. Wai, R.J.: Robust fuzzy neural network control for nonlinear motor-toggle servomechanism.
Fuzzy Sets Syst. 139(1), 185–208 (2003)
11. Fischle, K., Schröder, D.: Stable model reference neurocontrol for electric drive systems. In:
European Conference on Power Electronics And Applications, vol. 2, pp. 2–432. Proceedings
Published by Various Publishers (1997)
 Load Sharing Control of Two Ac Motors with a Common Stiff Shaft … 647

12. Sugiura, K., Hori, Y.: Vibration suppression in 2-and 3-mass system based on the feedback of
imperfect derivative of the estimated torsional torque. IEEE Trans. Industr. Electron. 43(1),
56–64 (1996)
13. Vukosavic, S.N., Stojic, M.R.: Suppression of torsional oscillations in a high-performance
speed servo drive. IEEE Trans. Industr. Electron. 45(1), 108–117 (1998)
14. Zhang, G., Furusho, J.: Speed control of two-inertia system by PI/PID control. IEEE Trans.
Industr. Electron. 47(3), 603–609 (2000)
15. Szabat, K., Orlowska-Kowalska, T.: Comparative study of the twomass system dynamics with
different speed controllers. In: Proc. 9th Int. Conf. Opt. Electr. Electron. Equipments, vol. 3,
pp. 97–102 (2004)
16. Quang, N.H., Quang, N.P., Nam, D.P., Binh, N.T.: Multi parametric model predictive control
based on laguerre model for permanent magnet linear synchronous motors. Int. J. Electr.
Comput. Eng. (IJECE) 9(2), 1067–1077 (2019)
17. Nguyen, H.Q., Nguyen, P.Q., Nguyen, N.H., Nguyen, T.B.: Min max model predictive control
for polysolenoid linear motor. Int. J. Power Electron. Drive Syst. 9(4), 1666 (2018)
18. Nguyen, Q.H., Dao, N.P., Nguyen, T.T., Nguyen, H.M., Nguyen, H.N., Vu, T.D.: Flatness
based control structure for polysolenoid permanent stimulation linear motors. SSRG Int. J.
Electr. Electron. Eng. 3(12), 31–37 (2016)
19. Nguyen, Q.H., Dao, N.P., Nguyen, H.M., Nguyen, H.N., Nguyen, T.T., Nguyen, C.P.:
Design an exact linearization controller for permanent stimulation synchronous linear motor
polysolenoid. SSRG Int. J. Electr. Electron. Eng. 4(1), 7–12 (2017)
20. Quang, N.H., Quang, N.P., Hien, N.N.: On tracking control problem for polysolenoid motor
model predictive approach. Int. J. Electr. & Comput. Eng. (2088–8708), 10(1) (2020)
21. Quang, N.H.: Multi parametric programming based model predictive control for tracking
control of polysolenoid linear motor. Spec. Issue Meas., Control. Autom. 19, 31–37 (2017)
22. Ty, N.T., Hung, N.M., Nam, D.P., Quang, N.H.: A Laguerre model-based model predictive
control law for permanent magnet linear synchronous motor. In: Information Systems Design
and Intelligent Applications (pp. 304–313). Springer, Singapore (2018)
23. Nguyen, H.Q.: Observer-based tracking control for polysolenoid linear motor with unknown
disturbance load. In: Actuators, vol. 9, no. 1, p. 23. Multidisciplinary Digital Publishing
Institute (2020, March)
24. Nam, D.P., Quang, N.H., Hung, N.M., Ty, N.T.: Multi parametric programming and exact
linearization based model predictive control of a permanent magnet linear synchronous motor.
In: 2017 International Conference on System Science and Engineering (ICSSE), pp. 743–747.
IEEE (2017, July)
25. Quang, N.H.: Controlling permanent magnet polysolenoid linear motor based on the nature
of voltage source inverter Part 1: Two-phase voltage source inverter in polysolenoid motor
control structure. Turk. J. Comput. Math. Educ. (TURCOMAT) 12(12), 1324–1329 (2021)
26. Quang, N.H., Quang, N.P., Hoang, D.D.: Finite control set model predictive control for
polysolenoid linear motor. Turk. J. Comput. Math. Educ. (TURCOMAT) 12(13), 4374–4379
(2021)
27. Quang, N.H.: Continuous control set model predictive control for polysolenoid linear motor.
Turk. J. Comput. Math. Educ. (TURCOMAT) 12(13), 2473–2478 (2021)
28. Quang, N.H.: Tracking control for polysolenoid linear motor base-on model predictive con-
troller: a comparative study of finite control set and continuous control set. Turk. J. Comput.
Math. Educ. (TURCOMAT) 12(13), 4851–4360 (2021)
29. Quang, N.H.: Dead-beat Control for Polysolenoid Linear Motor. Turk. J. Comput. Math.
Educ. (TURCOMAT) 12(13), 3293–3299 (2021)
30. Nordin, K.B., Novotny, D.W., Zinger, D.S.: The influence of motor parameter deviations in
feedforward field orientation drive systems. IEEE Trans. Ind. Appl. 4, 1009–1015 (1985)
648 N. T. Thanh and D. D. Hoang

31. Krishnan, R., Doran, F.C.: Study of parameter sensitivity in high-performance inverter-fed
induction motor drive systems. IEEE Trans. Ind. Appl. 4, 623–635 (1987)
32. Schierling, H.: Self-commissioning—a novel feature of modern induction motor drives. In:
Proceedings of the IEE Conference on Power Electronics and Variable Speed Drives, pp. 287–
290 (July 1988)
33. Schierling, H.U.B.E.R.T.: Fast and reliable commissioning of AC variable speed drives by
self-commissioning. In: Conference Record of the 1988 IEEE Industry Applications Society
Annual Meeting, pp. 489–492. IEEE (1988, October)
34. Garces, L.J.: Parameter adaption for the speed-controlled static AC drive with a squirrel-cage
induction motor. IEEE Trans. Ind. Appl. 2, 173–178 (1980)
35. Lorenz, R.D.: A simplified approach to continuous on-line tuning of field-oriented induction
machine drives. IEEE Trans. Ind. Appl. 26(3), 420–424 (1990)
36. Krishnan, R., Bharadwaj, A.S.: A review of parameter sensitivity and adaptation in indirect
vector controlled induction motor drive systems. IEEE Trans. Power Electron. 6(4), 695–703
(1991)
37. Heber, B., Xu, L., Tang, Y.: Fuzzy logic enhanced speed control of an indirect field-oriented
induction machine drive. IEEE Trans. Power Electron. 12(5), 772–778 (1997)
38. Uddin, M.N., Radwan, T.S., Rahman, M.A.: Performances of fuzzy-logic-based indirect
vector control for induction motor drive. IEEE Trans. Ind. Appl. 38(5), 1219–1225 (2002)
39. Lai, Y.S., Lin, J.C.: New hybrid fuzzy controller for direct torque control induction motor
drives. IEEE Trans. Power Electron. 18(5), 1211–1219 (2003)
40. Zhen, L., Xu, L.: On-line fuzzy tuning of indirect field-oriented induction machine drives.
IEEE Trans. Power Electron. 13(1), 134–141 (1998)
41. Bim, E.: Fuzzy optimization for rotor constant identification of an indirect FOC induction
motor drive. IEEE Trans. Industr. Electron. 48(6), 1293–1295 (2001)
42. Zhen, L., Xu, L.: Fuzzy learning enhanced speed control of an indirect field-oriented induction
machine drive. IEEE Trans. Control Syst. Technol. 8(2), 270–278 (2000)
43. Ibrahim, Z., Levi, E.: A comparative analysis of fuzzy logic and PI speed control in high-
performance AC drives using an experimental approach. IEEE Trans. Ind. Appl. 38(5), 1210–
1218 (2002)
44. Nguyen, D.C.: Ministry Science Project “The load balancing for 2 motors with common stiff”
(2014)
45. Åström, K.J., Wittenmark, B.: Adaptive control. Courier Corporation (2013)
 Mathematical Model of the Workflow
of a Modern Spark-Ignition Engine Running
on Ammonia

Andrey Y. Dunin1(B) , Dung Van Nguyen1 , and Nguyen Khac Minh2

1 Department of Heat Engineering and Automotive Engines, Moscow Automobile and Road
Construction State Technical University (MADI), Moscow, Russia
a.u.dunin@yandex.ru
2 Faculty of Automotive and Power Machinery Engineering, Thai Nguyen University of

Technology, No. 666 Tich Luong Ward, Thai Nguyen, Vietnam

Abstract. Internal combustion engines (ICEs) that use petroleum-based fuels are
the most commonly used engine types in vehicles. During the operation of ICEs, a
large amount of harmful emissions is generated, which are the leading causes that
adversely affect the quality of the environment and human health. Moreover, over-
reliance on fossil fuels leads to an imbalance between energy sources (electricity,
solar energy, and new energy sources). In this paper, the author presents a method
to use Ammonia to replace traditional fuel. Ammonia is one of the most promising
carbon-free alternative fuels. High-octane, low-viscosity ammonia protects against
detonation and is also an effective hydrogen carrier: 1.5 mol of hydrogen per
mole of ammonia. For research purposes, a mathematical model has been built to
describe the working process of ICE using Ammonia. The research results show
that the feasibility of using alternative fuel Ammonia and the cycle specification
of the internal combustion engine is significantly improved.

Keywords: Internal combustion engines · Ammonia · Alternative fuels · Spark

ignition engine

1 Introduction
Diesel engines are considered one of the largest sources of environmental pollution
caused by emissions of harmful compounds with exhaust gases. Carbon (C) and hydrogen
(H2 ) form the basis of diesel fuel, as do most other fossil fuels. The main combustion
products of petroleum fuels formed in the combustion chambers of an engine with
an ideal thermodynamic cycle are carbon dioxide (CO2 ), water (H2 O) and nitrogen
(N2 ) [1]. Under real conditions, the reactions do not proceed completely, therefore,
intermediate compounds are present in the exhaust gases. This is influenced by many
reasons (the ratio of air and fuel, the gorengs, turbulence in the combustion chamber, the
form of combustion, the burning temperature, etc.), so, a number of harmful products are
formed during the combustion process. The most significant harmful products are carbon
monoxide (CO), hydrocarbons (HC), nitrogen oxides (NOX ) and soot solids (PM) [1, 2].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 649–661, 2023.
https://doi.org/10.1007/978-3-031-22200-9_70
650 A. Y. Dunin et al.

In the modern world, in all spheres of the national economy, environmental protection
has become one of the main topics.
In Europe, EURO standards have been developed for vehicles, which since 1993 have
been constantly being tightened from the EURO I level to the EURO level currently in
force. Table 1 shows the European standards for heavy-duty vehicles.

Table 1. EU emission standards for heavy-duty vehicles [1].

CO (g/kW·h) HC (g/kW·h) NOx (g/kW·h) PM (g/kW·h)

EURO-1 4,5 1.1 8.0 0.61
EURO-2 4 1.1 7.0 0.15
EURO-3 2,1 0.66 5.0 0.13
EURO-4 1,5 0.46 3.5 0.02
EURO-5 1,5 0.46 2.0 0.02
EURO-6 1,5 0.13 0.4 0.01

Decades of research have focused on engine modifications, electronically controlled

fuel injection systems, and improved fuel properties. The limited oil resources lead to
the need to use alternative motor fuels of non-oil origin [3–5]. These fuels are obtained
from various feedstocks and have very diverse physical and chemical properties.
One of the most promising alternative fuels that do not contain carbon is ammonia
(NH3 ). Ammonia occurs naturally in the form of ammonium salts, and its purest form
is anhydrous NH3 . It is an effective hydrogen carrier: 1.5 mol of hydrogen per mole
of ammonia. Ammonia is lighter than air and dissolves well in water, which ensures
leakage control, avoiding the risk of fire or explosion.
The low viscosity contributes to the spraying of NH3 and the formation of droplets
during fuel injection. Ammonia has a high octane number (RON = 130), which makes
it suitable for engines with a high compression ratio, since it reduces the probability of
detonation.
The main advantages of using ammonia as a fuel: High energy density (33% higher
than that of liquefied H2 ), high octane number, easy handling and storage (liquid ammo-
nia is stored at a temperature of 298 K and a pressure of up to 1 MPa, as well as at a
temperature of 240 K (−33 °C) and at atmospheric pressure) and a carbon-free product
during combustion.
Disadvantages of ammonia for use in engines: Its low flame speed (0.07 m/s at
atmospheric conditions); high minimum ignition energy; high self-ignition temperature
(930 K); narrow flammability limits, making it difficult to use as a single fuel; and high
NOx emissions.
In this article, a supercharged gasoline engine was simulated, operating on a cycle
with continued expansion (Miller cycle), and in this case, two programs were involved:

• a program for calculating the working cycle of a gasoline engine with spark ignition
(GESI);
 Mathematical Model of the Workflow … 651

• a program for calculating the combined operation of a gasoline engine and a

turbocharger (TR).
At the same time, the engine performance parameters and indicators necessary to
calculate the combined operation of GESI and TR are: filling factor (ηv ), exhaust gas
temperature in front of the turbine (TT ), effective torque (Mtor .) and average effective
fuel consumption (ge ).
The simulation of a supercharged GESI operating in a cycle with continued
expansion is carried out in the following sequence:
• at the beginning, for a given value of the engine crankshaft speed (n) and the moment of
the end of the intake (ϕde.(inj.) ), according to the program (2), the values of pressures
and temperatures at the inlet and outlet (pk and pT , Tk and TT , respectively) are
determined;
• then the obtained values pk , pT , Tk , TT are entered into the program (1) and the duty
cycle is calculated.
To simulate the combined operation of the GESI, operating on a cycle with con-
tinued expansion, and the TR, the program for calculating the combined operation of
a gasoline engine and a turbocharger, developed at the Department of Thermal Engi-
neering and Automotive Engines of MADI, was taken as the initial one. The program
was based on the calculation algorithm developed by A.A. Frizin [6].
The program provides ample opportunities for varying the parameters of the
engine, the charge air cooler and the TR and allows to solve such tasks as:
• selection in the first approximation of the TR when the engine is supercharged;
• assessment of the impact of the design parameters of the TR on the operation of the
engine when turbocharged;
• evaluation of the use of a charge air cooler with a given efficiency;
• obtaining an approximate law of opening the bypass valve depending on the engine
shaft speed with the throttle fully open;
• determination of the surge reserve when the engine is running with the TR, the
parameter values of which have been set.

The information scheme of the model, taking into account the sequential application
of two programs, is shown in Fig. 1.
When constructing the model, a number of polynomials were used to calculate both
the performance indicators of the TR and the performance indicators of the engine,
depending on the group of influencing factors.
In general, these polynomials have the following:

ηv , TT , Mk , ge = f (ϕde.(inj .) , ϕc , ε, pk , Tk , pT ), (1)

where: ϕde.(inj.) —the moment of the end of the intake, deg. CAD to BDC; ϕk— the
angle of the beginning of the heat release process, deg. CDA to TDC; ε—compression
ratio; pk , Tk —conditions at the inlet (before the intake valve), MPa and K, respectively;
rT —pressure at the outlet before the turbine, MPa.
When calculating the operation of a vehicle using polynomials, the following
indicators are calculated:
652 A. Y. Dunin et al.

Fig. 1. Information diagram of a supercharged gasoline engine model operating in a cycle with
continued expansion

• the adiabatic efficiency of the compressor (ηk.ad. );

• the degree of increase in air pressure in the compressor (πk ) when calculated according
to the dimensional characteristics of the compressor;
• compressor head ratio (ψ0) when calculated from the dimensionless characteristic of
the compressor;
• the given second gas flow rate through the turbine (Gt. Etc.) when calculated according
to the dimensionless characteristic of the turbine;
• effective turbine efficiency (ηte );
 Mathematical Model of the Workflow … 653

• the degree of decrease in the pressure of gases in the turbine (πT ) when calculated
according to the dimensional characteristic of the turbine.

2 Research Methodology for Increasing the Average Effective

Pressure

The main purpose of the study is to determine the value of the average effective pressure
(pe ), which can be achieved through the use of turbocharging and the implementation
of a cycle with continued expansion on a gasoline engine.
The results of the study will be compared with the limiting (pe ) results that can be
achieved on a gasoline engine running on a conventional cycle both with and without
turbocharging.
This problem is solved on a specific mode determined by the external speed
characteristic and the engine crankshaft rotation speed (n) equal to 3200 min−1 .
When solving the problem, the following concept of the engine under study was
adopted: a gasoline engine equipped with a TR an electrohydraulic valve drive system.
The 4CHN 8.2/7 engine served as a prototype of the experimental engine, focusing
on which the basic geometric parameters were taken, the setting of which was necessary
as initial data both in the cycle calculation program and the joint operation of the engine
and the TR. Below, Table 2 shows these parameters.
The duration of the intake valve opening (ϕ0 (intake) ) was not included in Table 2,
since this parameter is variable for the experiment.
Before starting the study, it was necessary for a given mode to determine the value
of ϕivc (and therefore ϕ0 (intake) ), which, other things being equal, would ensure the
maximum filling of the cylinder (ηv max ) and, as a consequence, the highest value of pe .
The search for the specified value of ϕz (let’s denote it as ηv max ) was carried out
according to the cycle calculation program for a naturally aspirated engine equipped
with a hydraulic valve drive and operating according to a normal cycle.
The intake parameters corresponded to standard atmospheric conditions and had the
values: pk = 0.1 MPa, Tk = 298 K.
The composition of the mixture in this search and further in the study is taken
stoichiometric (α = 1), necessary for the effective operation of the neutralizer.
The pressure at the outlet (Tt) was taken to be 0.102 MPa, which corresponds to the
values that take place for the MZMA-412 engine.
The value of the outlet temperature (TT ) was approximately (according to experi-
mental data for the MZMA-412 engine) set to 843 K and then refined by the program
by repeated calculation.
The parameters of the combustion process had the following values:

• duration of heat release: ϕT = 40 deg. CA;

• angle of the beginning of the heat release process: ϕc = 15 deg. TDC
• heat release curve indicator: β = 2;
• heat release curve indicator: m = 2.
The average temperatures of the surfaces of the engine parts were assumed to be
equal:
654 A. Y. Dunin et al.

Table 2. Basic geometric parameters of the experimental engine.

№ Title Designation Dimension Value

1 Cylinder diameter D mm 82
2 Piston stroke S mm 70
3 Number of cylinders i – 4
4 Compression ratio ε – 9
5 Ratio of combustion chamber surface to Fcb /Fp – 2.3
piston crown area
6 Engine displacement i·Vh l 1.478
7 Ratio of crank radius to connecting rod λ – 0.280
length
8 Maximum intake valve lift Hvalve (intake) max mm 8.7
9 Intake valve throat diameter Dg (intake) mm 36
10 Angle of advance opening of the intake ϕadvance (intake) deg 21
valve after TDC
11 Inlet valve chamfer angle αf (intake) deg 45
12 Maximum lifting height of the exhaust Hvalve (inlet) max mm 8.7
valve
13 Diameter of the exhaust valve neck Dg (exhaust) mm 31
14 Angle of advance opening of the ϕ advance (exhaust) deg 63
exhaust valve befor BDC
15 Duration ϕ0 (exhaust) deg. CA 263
exhaust valve openings
16 Chamfer angle of the exhaust valve αf (exhaust) deg 45
17 Minimum engine crankshaft speed Nmin min−1 1400
18 Engine crankshaft speed at nominal Nnom min−1 5400
mode

• piston bottom temperature: Tp = 573 K;

• the temperature of the inner surface of the combustion chamber located in the cylinder
head: Tg = 548 K;
• cylinder wall temperature: Tc = 518 K.

The results of the search for ϕivc ηv max are presented in the form of a graph (Fig. 2),
repeating the experimental dependence for a given engine.
Based on the results obtained, the value of ϕivc ηv max : 26 deg. Was determined after
BDC, which, taking into account ϕ0 (intake) corresponds to the value ϕ0 (intake) equal to
227 deg.
Since neither the boost nor the geometric compression ratio affect ηv , the resulting
value of ϕivc ηv max will also provide ηv max at a given mode for a supercharged engine.
 Mathematical Model of the Workflow … 655

Fig. 2. Dependence of the filling coefficient (ηv ) and the average effective pressure (pe ) on the
moment of closing of the intake valve (ϕivc )

In the case of a naturally aspirated engine, the selected value of ϕivc ηv max will allow,
all other things being equal, to obtain the maximum value of pe for it.
Thus, during the study, ϕivc will take the values indicated in Table 3, which correspond
to ϕivc (in the order of their location in the table): 60, 40, 20 deg. Before BDC; at BDC;
26 after BDC ((ϕivc ηv max ).

Table 3. The value of ϕ0 (intake) taken by him at the entrance of the study

ϕ0 (intake) , [deg. CA] 141 161 181 201 227

In the cycle calculation program, the current flow areas of the intake and exhaust
valves (fvalve (intake) and fvalve (exhaust) respectively) are determined based on the premise
that they have an electrohydraulic drive. At the same time, the current flow coefficients
through the intake and exhaust valves (μvalve (intake) and (μvalve (exhaust) , respectively) are
determined depending on the current valve lift and have values in magnitude similar to
those of the MZMA-412 engine.
For the specified value of ϕivc ηv max , the value of the moment of the beginning of
heat release (ϕc ) will be selected, providing the highest value of pe . The values of ϕc are
taken in the range from 4 to 22 deg. to the TDC in increments of 1 deg.
It is assumed that the parameters of engines operating on a normal cycle, both with
and without supercharging, have values equal to the values indicated in Table 2. This
is done in order to exclude the influence of the design features of the engines on the
obtained comparison results.
Since the results obtained on the experimental engine model will be compared with
the marginal results obtained on a turbocharged engine operating on a conventional cycle,
it is necessary to select a TR for this engine before starting the study. The selection of
the transport vehicle will be carried out based on the condition that they provide the
656 A. Y. Dunin et al.

maximum possible value of pe at a given operating mode. The method of selection of

turbocharger is described below.
For the experimental engine, in order to obtain the maximum value of pe, the tur-
bocharger is also selected. The selection procedure in this case remains the same, but
the selection itself takes place for each combination of the values of the duration of
the opening of the intake valve (with the exception of the duration corresponding to
ϕivc ηv max ) from those indicated in Table 3.
When selecting the TR for both the conventional and the engine under study, as
well as when conducting a study, the following values of the excess air coefficient
(α), combustion process parameters (ϕT , β, m) and average surface temperatures of
engine parts are set in the cycle calculation program as they were set when searching
for ϕivc ηv max (see above). This is done in order to exclude their influence. At the same
time, the moment of ignition onset (ϕc ) will be varied in the range from 10 to 22 deg. to
the TDC in increments of 1 deg. BTDC. The calculation step of the cycle also remains
the same: 1 deg. CA.
Table 4 shows the values (with the exception of i:Vh , nmin , nnom , indicated in Table 2),
the values of which must be set in the calculation program for the joint operation of the
engine and the TR as initial data, and at the same time these values will remain unchanged
both during the selection of the turbocharger and during the study.
Dependencies constructed at values ϕ0 (intake) , which correspond to the closure of the
intake valve 60, 40, 20 deg. up to BDC, characterize an engine operating in a cycle with
continued expansion. The dependences constructed at values ϕ0 (intake) , corresponding
to the closure of the intake valve in the BDC and at ϕivc ηv max is characterized by
a turbocharged engine operating on a regular cycle. The pe values obtained at pk =
0.100 MPa correspond to a naturally aspirated engine.
When constructing each of the dependencies of pe on pk , the point on the detonation
line will correspond to the limit value of pe, which can be achieved with the selected
turbocharger and at which the detonation is characterized as strong.
The estimation is based on the proportion of detonating fuel (X). At the same time,
it is known from experience that detonation is possible only if it occurs in the unburned
part of the charge, before 90% has time to burn. Thus, in this paper, we will take the
detonation at which X will be 0.1 as the initial level.

3 Turbocharger Selection Methodology

In the conditions of the study, the main task of selecting the TR is to determine such TR
parameters that would provide the pe limit values at a certain engine operating mode: at
the specified values of ϕc and ϕ0 (intake).
As well as research tasks, the task of selecting a TR will be solved on a mathe-
matical model of a supercharged gasoline engine operating in a cycle with continued
expansion. Therefore, all the results obtained will be valid only within the framework
of the assumptions that were made when constructing this model.
When selecting a turbocharger, the following parameters will be varied:
 Mathematical Model of the Workflow … 657

Table 4. The initial data of the program for calculating the joint operation of the engine and the
TR, the values of which remain unchanged when selecting the TR and conducting the study

№ Title Designation Dimension Value

1 Compressor performance reduction temperature T0k K 288
2 Compressor performance reduction pressure p0C MPa 0,1
3 Stoichiometric amount of air l0 kg/kg of fuel 14.96
4 Gas constant R DJ/(kg K) 287
5 Charge air density ρB kg/m3 286
6 Air adiabatic index k – 1.4
7 Gas adiabatic index k1 – 1.35
8 Atmospheric air pressure p0 MPa 0.1
9 Atmospheric air temperature T0 K 298
10 A coefficient that takes into account the pressure loss a MPa 0.002
at the compressor inlet
11 A coefficient that takes into account the pressure loss b MPa 0.002
in the section from the compressor to the intake valve
12 Coefficient that takes into account the pressure c MPa 0.002
increase behind the turbine
13 Coefficient that takes into account the loss of gases in m – 0.98
the section up to the turbine
14 Coefficient for calculating air cooling efficiency E – 0.25
15 Excess air ratio α – 1
16 Hourly air consumption by the engine at nominal GB nom kg/h 300
mode
17 Dimensionless air velocity at the compressor inlet λC – 0.2
18 Dimensionless gas velocity at the turbine inlet λt – 0.2
19 Efficiency of the air cooler at nominal mode Enom – 0.65

• the outer diameter of the compressor wheel (Dk2 );

• the diameter of the compressor wheel at the inlet (Dk1 );
• the minimum cross-sectional area of the turbine nozzle (fca );
• the outer diameter of the turbine wheel (Dt1 ).

The variation will be carried out according to the size range of turbochargers given
in the work of A.A. Frizin [6]. This series is presented in Table 5.
The choice of this series is explained by the fact that it is for such sizes that the
characteristics of turbochargers are incorporated into the calculation program for the
joint operation of a gasoline engine and a turbocharger. In addition, it is precisely such
turbochargers that are able to provide effective supercharging of a subcompact gasoline
engine.
658 A. Y. Dunin et al.

Table 5. Size range of turbochargers

No. Dk2 = Dt1 , mm Dk1 , mm

1 50 28
2 55 30
3 60 33
4 65 36

The range of variation of the aca value will be taken from 0.0003 to 0.0012 m2 .
When selecting a turbocharger, the equivalent flow section of the bypass valve is set
to zero (bypass is not carried out).
The selection of the turbocharger at the specified values ϕc and ϕivc will be carried
out in the following order:

• at the beginning, for each set of wheel diameters from Table 5 such a value of fca will
be selected to provide the highest value of pe ;
• then, from the four turbochargers selected in this way (one with each set of wheel
diameters), the one providing the highest re value will be selected.

When selecting turbochargers, the achieved pe value will be evaluated for the
probability of detonation.
In addition to the restriction on detonation, restrictions are imposed on:

• the value of the temperature in front of the turbine (TT ), no more than 1230 K;
• surge margin (Gk pom.), should be positive (better if more than 10%);
• the rotation speed of the turbocharger rotor (pr ), not more than nominal (nr nom).

4 Research Results
In the first part of the study, for a naturally aspirated engine, the intake valve closing
moment was selected in order to achieve maximum filling (ηv ) and average effective
pressure (pe ) at a certain moment of ignition onset (ϕc = 15 deg. Before the TDC)
according to the methodology given in Sect. 3.1. The results for this part of the study
are shown in Fig. 3.
Thus, it was possible to select ϕivc equal to 26 deg. After the BDC, which provided
the following values of the indicators of interest to us:

pe = 1.062MPa;
ηv = 0.893.

With further variation by the value of ϕs , it was not possible to exceed the value of
1.062 MPa in re, and the value of the angle of the beginning of heat release equal to
15 deg. Before the TDC, it was considered optimal. The results of the variation of the
value ϕc are shown in Fig. 4.
 Mathematical Model of the Workflow … 659

In the second part of the study, based on the methodology of searching for the
maximum value of pe and the associated selection of turbochargers, turbochargers were
selected for four values of ϕivc : 26 deg. After BDC, at BDC, 10 deg. up to BDC and
20 deg. Before BDC.

Fig. 3. Dependence of the average effective pressure (pe ) on the moment of ignition initiation
(ϕc )

The geometric parameters of the selected turbochargers are given in Table 6, and
the re values that they were able to provide are in Table 7. Also, Table 7 shows the
corresponding boost pressure values.

Table 6. Geometric parameters of selected turbochargers.

ϕivc , deg. After BDC Dk2 = Dt1 , mm Dk1 , mm fca , m2

26 50 28 0.001120
0 50 28 0.000724
−10 50 28 0.000589
−20 50 28 0.000542

Table 7. Values of pk , pe , obtained as a result of the selection of turbochargers

ϕivc , deg. After BDC pcom , MPa pe , MPa

26 0.126 1.366
0 0.139 1.393
−10 0.149 1.401
−20 0.155 1.348

The pe values indicated in Table 6 were obtained at the moment of ignition onset
equal to 10 deg. BTDC.
660 A. Y. Dunin et al.

The dependences of pe on pk obtained at different values of ϕivc are shown in Fig. 4.

As can be seen from the Table 7 and Fig. 4, for the value of ϕivc equal to 20 deg. up
to BDC, it is not possible to reach the limit value for re due to insufficient boost pressure
and turbocharger efficiency in Fig. 4. This limit value is set approximately and is equal
to 1.409 MPa.
For a similar reason, the values of ϕivc equal to 40 and 60 deg. Were excluded from
consideration before BDC.

Fig. 4. Dependence of the average effective pressure (pe ) on the boost pressure (pk ) and the
closing moment of the intake valve (ϕivc )

26 − ϕivc = 26 deg. After BDC;

0 − the intake valve is closed in the BDC;
−10 − ϕivc = 10 deg. Befor BDC;
−20 − ϕivc = 20 deg. Befor BDC.

5 Conclusions
Analyzing the results of the study, the following conclusions can be drawn:

1. Due to the implementation of a cycle with continued expansion on a gasoline engine

in combination with the introduction of supercharging, there is a real possibility
of a significant increase in the average effective pressure lying at the detonation
boundary. In our case, such an increase compared to the naturally aspirated version
was 31,9% with a change in the closing moment of the intake valve from 26 deg.
After BDC up to 10 deg. Before BDC.
2. The implementation of a cycle with continued expansion makes sense on a super-
charged gasoline engine. In our case, the value of the average effective pressure was
raised by 2,6% with a change in the closing moment of the intake valve from 26 deg.
After BDC up to 10 deg. Before BDC.
 Mathematical Model of the Workflow … 661

3. To get a greater effect from the implementation of the cycle with continued expansion,
the introduction of earlier angles of closing the intake valve, high values of boost
and efficiency of turbochargers are required.

References
1. Taranushich, V.A., Savostyanov, A.P., Sulima, S.I., Zemlyakov, N.D., Bakun, V.G., Narochny,
G.B., Ilyin, V.B., Ponomarev, V.V.: Technology of Catalysts: Textbook South-Russian State
Technical University. un-t (NPI). Novocherkassk: YURSTU (NPI), 96 p. (2012)
2. Baryshnikov, S.A.: Requirements for diesel fuel supply systems/Baryshnikov, S.A., Devyatkin,
D.E., Mashrabov, A.N.: Service of technical systems—agro-industrial complex of Russia:
Materials of the international scientific and practical conference, Chelyabinsk, March 29–31
(2017). Ministry of Agriculture of the Russian Federation, Department of Scientific and Tech-
nological Policy and education; South Ural State Agrarian University, pp. 30–36. Chelyabinsk:
South Ural State Agrarian University (2017)
3. Berridge, G.R., Lloyd, L.: F. In: Barder, B., Pope, L.E., Rana, K.S. (eds.) The Palgrave Macmil-
lan Dictionary of Diplomacy, pp. 147–160. Palgrave Macmillan UK, London (2012). https://
doi.org/10.1057/9781137017611_6
4. Markov, V.A., Devyanin, S.N., Semenov, V.G., Shakhov, A.V.: The use of vegetable oils and
fuels based on them in diesel engines. In: M.: LLC SIC “Engineer”, LLC “Oniko-M”, 536 p.
(2011)
5. Markov, V.A., Bebenin, E.V., Pozdnyakov, E.F.: Comparative evaluation of alternative fuels
for diesel engines. Transp. Altern. Fuel. (5), 24–29 (2013)
6. Frizin, A.A.: Computational and experimental determination of performance indicators of a
turbocharged gasoline engine. In: Dis. Candidate of Technical Sciences, 222 p. MADI, Moscow
(1987)
 Mixed Convection of a Non-Newtonian Fluid
in a Square Enclosure with Rotating Cylinder

Swaranjali Maurya(B) , Ram Dayal(B) , and Manish Kumar(B)

Malaviya National Institute of Technology Jaipur, Jaipur, Rajasthan 302017, India

swaranjalimaurya@gmail.com, {ramdayal.mech,
manish.mech}@mnit.ac.in

Abstract. Two-dimensional simulations of laminar mixed convection in a square

enclosure subjected to constant wall temperature with a heated rotating cylinder
have been carried out. The enclosure is filled with non-Newtonian fluid. Due to the
rotation of the cylinder, forced convection effects are set up along with the natural
convection. The effect of different parameters, viz: Rayleigh number (103 to 106 ),
and angular velocity on non-Newtonian fluid enclosed in a square enclosure are
investigated. Observations are also made on the dynamic behavior of the flow
features and temperature fields due to the rotation of hot cylinder inside a square
enclosure. Overall, the present study focuses on how the rotating cylinder affects
the thermo-fluid behavior of non-Newtonian fluids and quantifies heat transfer
characteristics.

Keywords: Mixed convection · Heat transfer · Non-Newtonian fluid · Rotating

cylinder

1 Introduction
Due to frequent interaction with non-Newtonian fluid behavior in wide-ranging geo-
physical and engineering systems, considerable effort has been made to understand
the natural, forced, and mixed convection in recent years. Typical examples are ocean
currents, thermal storage, and electronic cooling devices, drilling of oil wells, nuclear
reactor fuel rods, environmental control, solar collectors, and food processing.
Several studies have been done in the literature for free convection inside a square
enclosure filled with Newtonian fluids without cavity reviewed by Davis [1], with square
cavity House et al. [2], and Bhave et al. [3], several disconnected blocks Merrikh et al.
[4], and Lage et al. [5], heated cylinder Das et al. [6], Kim et al. [7], and Hussain et al.
[8], rotating cylinder Costa et al. [9], Hussain et al. [10], Liao et al. [11].
The non-Newtonian fluids have received less attention in the literature than Newto-
nian fluids. However, the literature on non-Newtonian fluids has some investigations on
free convection in an enclosure with or without stationary cylinder.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 662–670, 2023.
https://doi.org/10.1007/978-3-031-22200-9_71
 Mixed Convection of a Non-Newtonian Fluid … 663

Isothermal cold top wall, TC

Isothermal cold right wall, TC

Isothermal cold left wall, TC

R
g
L

Isothermal
hot wall,

Isothermal cold bottom wall, TC

L

(a) (b)
Fig. 1. (a) Physical Domain of Present study√(b) Vertical velocity at mid height along horizontal
line for Ra = 106 , and Pr = 0.71, where, V = gβTL, validated numerical [20] and experimental
[21]

This work is extended by Turan [12] for CWHT. In a square enclosure that was
cooled from the sidewalls and heated from the bottom, Hassan et al. [13] examined
the free convection of viscoplastic fluid numerically. Hassan et al. [14] expand this
work to include partially heated bottom walls at their center. Furthermore, there are also
other investigations on natural convection for heated cylinders in an enclosure. Sairamu
et al. [15] and Baranwal et al. [16] presented their research on natural convection heat
transfer in a Bingham plastic fluid. Mishra et al. [17] evaluate the cylinder’s vertical
and diagonal orientations with Bingham plastic fluid. They found that heat transmission
for the vertically positioned cylinder is maximized when the two cylinders are located
either at the enclosure’s axis or extremely close to one of the side walls. Additionally,
heat transfer is greatest for cylinders oriented diagonally when the space between them is
as small as possible. The heat transfer for power-law fluid has been studied by Chhabra
et al. [18]. When the Ra is low, the flow shows dual recirculating cell regions, and
as the Ra rises, only one recirculating cell region is generated. Recently, Kumar et al.
[19] investigated for Newtonian fluid presented their analysis of the free convection of
stationary, and oscillating cylinders in an isothermal square enclosure. They investigated
the various cylinder diameter, frequency, amplitude, oscillation axis and Ra. The Nu
along enclosure’s wall and the cylinder did not considerably increase compared to the
stationary cylinder. However, with increased oscillation, amplitude and frequency, the
average Nu dramatically rose at the bottom wall.
However, there have been fewer studies done on convection heat transfer of non-
Newtonian fluid. The purpose of the present work is to investigate 2D laminar mixed
convection in an isothermal cold square enclosure with a heated rotating cylinder. The
enclosure is filled with non-Newtonian fluid. Due to the rotation of the cylinder, forced
convection effects are set up along with the natural convection. Furthermore, observations
are made on the dynamic behavior of the flow features and temperature fields due to the
rotation of a hot cylinder inside a square enclosure. Hence, numerical simulations are
carried out.
664 S. Maurya et al.

Fig. 2. (a) Comparison of present simulation result of mid-height temperature profile, θ vs. x for
different number of cells with [27] (b) Dimensional less velocity (V = vLϑ ) at mid plane along
horizontal line (x) for R = 0.1 and Ra = 106 , validated with [27]

2 Problem Definition and Mathematical Description

The domain of the 2D square enclosure with the concentric revolving cylinder, width and
height L, is shown in Fig. 1(a). In contrast to the cylinder, which is isothermally heated at
constant high-temperature TH (320 K), the square enclosure’s walls are isothermally cold
at constant low-temperature TC (300 K). The fluid temperature (T0 = 300 K) is the same
as the cold wall temperature. Gravitational acceleration (g) is in negative y-direction.
An incompressible non-Newtonian fluid’s unsteady-state flow is taken into account
in this study. For incompressible fluids under an unsteady situation, mass, momentum
and energy conservation equations take the following form of Eqs. (1), (2) and (3)
respectively:
∂uj
=0 (1)
∂xj
∂uj ∂ui uj ∂ρ ∂ 2 ui
+ =− + v 2 + βg(T − T0 ) (2)
∂t ∂xj ∂xi ∂xj

∂T ∂T ∂ 2T
+ uj =α 2 (3)
∂t ∂xj ∂xj

where ui , p and T denotes velocity field, pressure and temperature respectively.

The governing non-dimensional parameters i.e., Rayleigh number (Ra), Grashof
number (Gr) and Prandtl number (Pr) are defined as-
 
gβT 2−n 4
ρf2 R R
Gr = (4)
m2
  2
ρC m (n+1) 2(n−1) 3(n−1)
Pr = R (n+1) (gβTR) 2(n+1) (5)
k ρ
 Mixed Convection of a Non-Newtonian Fluid … 665

Ra = Gr · Pr (6)

The rate of heat transfer is computed at each wall of the cold square enclosure and
cylinder wall is expressed in terms of average Nusselt number (Nu).

hL
Nu = (7)
k
dT
hT = k (8)
dn

hL
Nu = (9)
k
Power Law fluid used in the experimental work of Dale and Emery [22] and numerical
simulation Rath et al. [23] is considered in this study. A power-law fluid in present
investigation is a water solution containing 0.05% Carboxy-Methyl-Cellulose (CMC)
powder. The power-law index n = 0.8 is used. Table 1 shows the properties of fluid used
throughout the research work.
Modified power-law rheological model [24], [25], [26], [23] is used to simulate
the power-law fluid being discussed here. According to Eq. (10), this model has four
parameters. It removes physiologically unreasonable restrictions of the fluid’s zero and
infinity shear rate viscosity by setting the lower and upper Newtonian viscosity limits
as ϑmin = 0.0001 and ϑmax = 1. The upper and lower bounds of these restrictions are
set by two orders higher and lower than the flow consistency index, respectively. These
bounds are not set arbitrarily; rather, they are compared by examining how it affects the
numerical results.
⎧
⎨ ϑmin
⎪ if γ̇ → 0
ϑ = mγ̇ n−1
if 0 ≤ γ̇ ≤ ∞ (10)
⎪
⎩
ϑmax if γ̇ → ∞

where, the local shear rate, γ̇ , is:

∂Vj ∂Vi
γ̇ = +
∂xi ∂xj

OpenFOAM® v2112, a free and open-source computational fluid dynamics (CFD)

tool, is used in the current numerical investigation. The buoyantBoussinesqSimpleFoam
solver is used for rotating cylinder.
666 S. Maurya et al.

Table 1. Properties of 0.05% Carboxy Methyl Cellulose (CMC) solution in water

Properties: 0.05% carboxy methyl cellulose solution in water

ρ 998.1 kg/m3
cp 4.187 × 103
k 0.6111 W/m K
m 1.066 × 10–2

Table 2. Comparison of present simulation result for Newtonian fluid (air) for Rayleigh number
range 104 to 106 and Pr = 0.71

Ra De vahl Davis [1] Merrikh et al. [4] House et al. [2] Present
104 2.243 2.244 2.254 2.250
105 4.519 4.536 4.561 4.619
106 8.811 8.860 8.923 8.971

3 Grid Independency Test

The grid Independency test with concentric cylinder is conducted for Ra = 106 and Pr =
0.71 with four different cell number 3200, 9600, 19200 and 32000. For the different cells
3200, 9600, 19200 and 32000, the variations in the average Nu are −1.43%, −1.34%, −
1.09%, and −0.91%, respectively. For different grids, variation of the temperature profile
along the mid-height of the enclosure is shown in Fig. 2(a). At all grids, it shows the
similar trend but with 32000 cells, having less error in average Nu. Hence, cell 32000 is
used throughout this present study. Figure 2(b) shows the velocity profile at mid height
of the enclosure for cell 32000.

4 Model Validation

The present study validates the numerical model with results obtained by previous study.
This validation has been done for the 2D model of a square enclosure filled with New-
tonian fluid. In this validation of Newtonian fluid, Fig. 1(b) shows the vertical velocity
at y = 0 that compared with numerical [20] and experimental [21] results and Table 2
compared the average Nu number to [1, 4] [2]. The validation of the numerical model
has been done for Ra = 104 to 106 and Pr = 0.71. The validation also has been done for
non-Newtonian fluid filled in square enclosure with stationary cylinder where Ra = 103
to 105 , Pr = 0.71, and R = 0.2 L.
 Mixed Convection of a Non-Newtonian Fluid … 667

5 Result and Discussion

5.1 Natural Convection of Non-Newtonian Fluid Filled in Square Enclosure
with Stationary Cylinder

First, the impact of stationary cylinders for various Ra is examined. The isotherms and
streamlines are displayed in Fig. 3(a)–(c) for Ra range 103 to 105 . Throughout the
research work, Pr = 10, R = 0.2 and n = 0.8.

(a) (d) (g)

(b) (e) (h)

(c) (f) (i)

Fig. 3. Isotherms (left) and streamlines (right) at RPM = 0 [(a)–(c)], RPM = 1 [(d)–(f)] and RPM
= 5 [(g)–(i)] with Ra = 103 to 105 , Pr = 10, n = 0.8 and R = 0.2

Effect of Ra
Free convection heat transfer dominates enclosure at Ra = 103 as shown in Fig. 3(a), and
the streamlines and isotherms resemble those at Ra = 105 when an enclosure is filled
with Newtonian fluid. Hence, it can be inferred that shear-thinning power law fluid (n =
0.8) are more effective at transferring heat. As the isotherms plume move upward, the
enclosure’s thermal gradients are stronger in the upper half and noticeably lower in the
lower half. The center of the recirculating eddies is exclusively in the enclosure’s upper
half that interpret that the enclosure’s upper half has the dominant flow. Convection heat
transmission further strengthened by raising Ra = 104 as seen in Fig. 3(b). The flow at
the bottom of the enclosure is growing weaker, and the isotherms travel upward with
the appearance of the plume on top. At the right, left, and top walls, the flow becomes
stronger. Recirculating eddies only occur in the enclosure’s top part. At Ra = 105 , on the
surface of the inner cylinder, the thermal boundary layer gets thinner, that illustrated in
668 S. Maurya et al.

Fig. 3(c). The separation of the recirculating eddies in upper half of enclosure resulted
in the creation of two vortices at the enclosure’s bottom.

5.2 Mixed Convection in Square Enclosure with Concentric Rotating Cylinder

Filled with Non-Newtonian Fluid

The mixed convection of non-Newtonian fluid filled square enclosure with a rotating
cylinder moving at various RPMs has been simulated for Ra range 103 to 105 , Pr = 10,
n = 0.8, R = 0.2 and RPM = 1 and 5. The Ra is the dimensionless parameter used to
evaluate the overall significance of mixed convection.
Effect of Ra at Different Rotational Speed
At RPM = 1, mixed convection dominates at low Ra = 103 . The Fig. 3(d) illustrates that
the isotherms are concentric to the cylinder, and at each corner of the wall, the revolving
eddies are formed. That means forced and natural convection occurs at the cylinder and
corner of the enclosure, respectively. As fluid is revolving with the cylinder, heat gets
distributed throughout the fluid, but at the corner, the cold fluid comes in contact with
the heated fluid and forms eddy. At Ra = 104 , Fig. 3(e) illustrates that mixed convection
heat transfer dominates. The isotherms are cylinder-centric. The recirculating eddies
are visible only at the bottom and top corners of the right wall. At the bottom wall,
the cold fluid zone increases as the isotherms squeeze due to an increase in natural
convection. Furthermore, Ra increases to 105 , where natural convection increases as
shown in Fig. 3(f). The mechanism involves fluid revolving with the cylinder; and fluid
around the cylinder trying to move upward due to thermal gradient. As a result, cold
fluid beneath the cylinder is not participating in the heat transfer process and impedes
the heat transfer at the bottom wall compared to the heat transfer at the Ra = 103 and
Ra = 104 .
As rotational speed of cylinder increases from RPM = 1 to RPM = 5, heat transfer
increases. At Ra = 103 and 104 , isotherms are concentric to cylinder, and recirculating
eddies are created at all corners of the wall that seen in Fig. 3(g) and (h). Heat transfer
is uniform at all walls of the enclosure due to the forced convection domination over
natural convection. Similarly, at Ra = 105 , Fig. 3(i) shows isotherms are concentric to
cylinder, and recirculating eddies are created at top and bottom corner of the right wall.
And, isotherms expand towards the left and top wall, hence heat transfer increases at the
left and top wall.

6 Conclusion

In this study, the heat transfer for 2D natural and forced convection of non-Newtonian
fluid in a square enclosure with heated rotating cylinder has been conducted at different
rotational speed, RPM = 0, 1, and 5. The influence of Ra and rotational speed has been
observed. The numerical study predicts the following results:

1. In compared to Newtonian fluid, heat transfer for stationary cylinders with a shear
thinning fluid (n = 0.8) is higher.
 Mixed Convection of a Non-Newtonian Fluid … 669

2. At RPM = 0, with the increase of Ra, natural convection heat transfer increases and
decreases at the top and bottom wall respectively.
3. At RPM = 1, mixed convection heat transfer dominates, but as we increase Ra from
103 to 105 , due to natural convection, heat transfer impedes at bottom wall and hence
average Nu decreases at Ra = 105 .
4. At high rotational speed i.e., RPM = 5, effect of Ra neglects as forced convection
dominates and heat transfer at all the wall of enclosure approximately same.

References
1. Davis, G.D.V.: Natural Convection of air in a square cavity a bench mark Numerical Solutions.
Int. J. Numer. Methods Fluids 249–264 (1983)
2. House, J.M., Beckermann, C., Smith, T.F.: Effect of a centered conducting body on natural
convection heat transfer in an enclosure. Int. J. Comput. Methodol. 213–225 (1990)
3. Bhave, P., Narasimhan, A., Rees, D.A.S.: Natural convection heat transfer enhancement using
adiabatic block: optimal block size and Prandtl number effect. Int. J. Heat Mass Transf.
3807–3818 (2006)
4. Ali, J.L., Merrikh, A.: Natural convection in an enclosure with disconnected and conducting
solid blocks. Int. J. Heat Mass Transf. 1361–1372 (2005)
5. Lage, J.L., Junqueira, S.L.M., De Lai, F.C., Franco, A.T.: Aspect ratio effect on the prediction
of boundary layer interference in steady natural convection inside heterogeneous enclosures.
Int. J. Heat Mass Transf. 940–947 (2016)
6. Roychowdhury, D.G., Das, S.K., Sundararajan, T.: Numerical simulation of natural convective
heat transfer and fluid flow around a heated cylinder inside an enclosure. Heat Mass Transf.
156–173 (2002)
7. Kim, B.S., Lee, D.S., Ha, M.Y., Yoon, H.S.: Numerical study of natural convection in a
square enclosure with a circular cylinder at different vertical locations. Int. J. Heat Mass
Transf. 1888–1906 (2008)
8. Hussain, S.H., Hussein, A.K.: Numerical investigation of natural convection phenomena in
a uniformly heated circular cylinder immersed in square enclosure filled with air at different
vertical locations. Int. Commun. Heat Mass Transf. 1115–1126 (2010)
9. Costa, V.A.F., Raimundo, A.M.: Steady mixed convection in a differentially heated square
enclosure with an active rotating circular cylinder. Int. J. Heat Transf. Eng. 1208–1219 (2010)
10. Hussain, S.H., Hussein, A.K.: Mixed convection heat transfer in a differentially heated square
enclosure with a conductive rotating cylinder at different vertical locations. Int. Commun. Heat
Mass Transf. 263–274 (2011)
11. Liao, C.C., Lin, C.A.: Mixed convection of a heated rotating cylinder in a square enclosure.
Int. J. Heat Mass Transf. 9–22 (2014)
12. Turan, O., Sachdeva, A., Poole, R.J., Chakraborty, N.: Laminar natural convection of power
law fluids in a square enclosure with differentially heated sidewalls subjected to constant wall
heat flux. J. Heat Transf. 122504–122519 (2012)
13. Hassan, M.A., Pathak, M., Khan, M.: Natural convection of viscoplastic fluids in a square
enclosure. J. Heat Transf. 122501–122511 (2013)
14. Hassan, M.A., Pathak, M., Khan, M.K., Khan, N.H.: Natural convection of viscoplastic fluids
in an enclosure with partially heated bottom wall. Int. J. Therm. Sci. 106527 (2020)
15. Sairamu, M., Nirmalkar, N., Chhabra, R.P.: Natural Convection from a circular cylinder in
confined Bingham plastic fluids. Int. J. Heat Mass Transf. 567–581 (2013)
670 S. Maurya et al.

16. Baranwal, A.K., Chhabra, R.P.: Effect of fluid yield stress on natural convection from
horizontal cylinders in a square enclosure. Int. J. Heat Transf. Eng. 557–577 (2017)
17. Mishra, L., Chhabra, R.P.: Combined effects of fluid yield stress and geometrical arrangement
on natural convection in a square two differentially horizontal cylinders. J. Therm. Sci. 1049–
1063 (2019)
18. Shyam, R., Sairamu, M., Nirmalkar, N., Chhabra, R.P.: Free convection from a heated circular
cylinder in confined power law fluids. Int. J. Therm. Sci. 13603–13615 (2013)
19. Kumar, M., Roy, S.: Immersed boundary method simulation of natural convection over fixed
and oscillating cylinders in square enclosure. Int. J. Heat Fluid Flow 122–132 (2016)
20. Mark, A., Svenning, E., Edelvik, F.: An immersed boundary method for simulation of flow
with heat transfer. Int. J. Heat Mass Transf. 56, 424–425 (2013)
21. Krane, J.: Some detailed field measurements for a natural convection flow in a vertical square
enclosure. In: ASME-JSME Thermal Engineering Joint Conference Proceedings, vol. 1,
pp. 323–329 (1983)
22. Dale, J.D., Emery, A.F.: The free convection of heat from a vertical plate to several non-
Newtonian pseudoplastic fluids. In: ASME (1972)
23. Rath, S.: Complex interplay of power-law rheology and non-Oberbeck-Boussinesq effects
on natural convection heat transfer in a confined domain. Int. J. Heat Mass Transf. 176,
121462–121479 (2021)
24. Boger, D.: Demonstration of upper and lower Newtonian fluid behaviour in a pseudoplastic
fluid. Nature 126–128 (1977)
25. Moulic, S.G., Yao, L.S.: Non-Newtonian natural convection along a vertical flat plate with
uniform surface temperature. J. Heat Transf. 131, 062501–062508 (2009)
26. Molla, M.M., Yao, L.S.: Non-Newtonian natural convection along a vertical heated wavy
surface using a modified power-law viscosity mode. J. Heat Transf. 012501–012506 (2009)
27. Moukalled, F., Acharya, S.: Natural convection in the annulus between concentric horizontal
circular and square cylinders. J. Thermophys. Heat Transf. 10, 524–531 (1996)
 Modeling and Dynamic Simulation
of a Multi-slope Vibrating Screen to Determine
the Basic Working Parameters

Tan Nguyen Dang(B)

Thuyloi University, Hanoi, Vietnam

nguyendangtan@tlu.edu.vn

Abstract. Currently, there are many researches to improve the efficiency and
capacity of the vibrating screen. Usually, multi-slope vibrating screens use an
unbalanced exciter to produce a circulating radial force of the same frequency on
both drive shafts. This produces an alternating force in the direction of the exciter
stand due to the circulation of the imbalance weights. The basic working parame-
ters, such as the forces acting on the bearings of the drive shafts and the supporting
springs, the amplitude, velocity, acceleration, and orbit of the screen surface, have
an important role in the design of a vibrating screen. These parameters are usually
determined by establishing a vibrating model and solving the technical vibrating
problem. However, this method does not give high accuracy because it simplifies
the vibration model of the screen. To make simulation results more accurate, this
study builds a 3D model of the vibrating screen and uses a dynamic multi-body
simulation to determine the basis working parameters of the screen. The correct-
ness of dynamic simulation research is verified through experiment. This work
would provide some guidance for designing and studying vibrating screen with
flexible construction and working parameters.

Keywords: 3D model · Dynamic simulation · Multi-body simulation ·

Multi-slope vibrating screen · Dynamic model

1 Introduction

A wide variety of vibrating screens are operated in the raw material processing and
aggregate industries. Vibrating screen is a common equipment in mining machinery and
equipment, its vibration is mainly generated by the exciter to achieve, and the use of
the exciter working conditions are harsh, and with strong vibration impact [1]. Vibrating
screens have a wide range of power, design, number of decks, and they can work on from
10 to over 1000 tons of material per hour. Cyclic excitation of the screen surface can
be accomplished by an unbalanced rotary shaft, hydraulic cylinder or electromagnetic
actuator. Based on the orbit of screen surface, vibrating screens are classified into circular,
elliptical, or linear types, which depend on the direction and value of the oscillating force
(see example in Fig. 1).

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 671–684, 2023.
https://doi.org/10.1007/978-3-031-22200-9_72
672 T. N. Dang

Fig. 1. Circular motion screen operated by single eccentric shaft [2]

During the working process, some problems may occur with the screen such as
material clogging on the screen surface, screen surface being worn, electrical motor
overloaded, failure of support springs and bearings. The factors that leads to bearing
damage mainly have 3 main points: design of vibrating screen; processing, and working
condition; and maintenance of vibrating screens [1]. To monitor the working condition
of the vibrating screen, monitoring devices are installed to find out the causes of bro-
ken springs and damaged bearings, incorrect rotational speed, and unbalanced masses
settings [3]. The multi-slope vibrating screen, which is an innovation, has a series of
continuous slopes of the screen surface, so that a constant padding thickness can be
maintained during screening. There is information that multi-slope vibrating screens
can double the processing capacity [4]. In order to optimize the vibrating screen, which
requires a better understanding of screening processes, in addition to setting up technical
vibrating equations, Discrete Element Method (DEM) software are used and increasingly
developed. The particle flow on a multi-slope screen is numerically studied [5].
There are also a few studies related to the new vibration mode. Inspired by manual
screen, a new vibrating screen with rotating vibration is proposed [6]. Compared with
translational vibration, its movement changes in amplitude and direction as the screen
surface of the rotating screen moves while swinging and shaking. The results show that
this type of movement can effectively improve the screening performance. However,
when the particles move to the vicinity of the center of the screen surface, where there
is slight vibration, it will sometimes gather and resist penetration; so screen rotation is
generally not applicable to fine particles [7]. In addition, several innovative vibrating
screens have been studied to improve their performance in terms of amplitude. A novel
variable-amplitude screen is proposed to drive externally by an unbalanced two-axle
excitation [8] and has large production capacity and good screening performance. Fur-
thermore, a novel equal-thickness screen with variable amplitude was investigated using
the vibration test method [9]. Modeling and dynamic analysis are studied according
to two-degree of-freedom systems theory. The effects of vibration parameters, includ-
ing swing angle, swing frequency, vibrating direction angle, and translation frequency,
on the screening efficiency were researched by experiment which is designed based
on the vibrating mode [10]. In order to improve the dynamic simulation efficiency of
vibration screen and reduce the computer calculation complexity, the model of vibration
screen is simplified. The vibrations of five different positions are recorded with different
amplitudes [11]. At present, the simplified 2-degree or 3-degree freedom mass-spring
 Modeling and Dynamic Simulation … 673

vibration model is often used in dynamic analysis of the vibrating screen [12]. How-
ever, the 2-degree vibration model can only represent the linear motion of the vibrating
screen in two directions, and the 3-degree freedom vibration model can represent the
swing around a certain axis in addition, and it is difficult to fully reflect the complex
motion of the vibrating screen [13].
Instead of setting up vibrating mathematical models or simplifying 3D models, this
study establishes 3D model and simulates multi-body dynamic of multi-slope vibrating
screen with 3-degree freedom vibration model by using Creo Parametric software. The
dynamic simulation results show the force acting on the bearings of the drive shafts, on
the supporting spring as well as the amplitude, velocity, acceleration, and orbit of the
screen surface. This work would also provide some guidance for designing and studying
vibrating screen with flexible construction and working parameters.

2 Modeling and Dynamic Analysis

2.1 Dynamical Model

In many vibrating machines, several unbalanced exciters are used to excite vibrations,
which rotate at the same angular speed and with defined phase differences. A directed
excitation force is often generated by two unbalanced rotors, whose synchronous move-
ment is forced by a pair of gears. A softly positioned vibrator body is considered as the
supporting system, which is excited with high excitation frequencies so that the inertial
forces dominate and the restoring forces are negligible. The rotors must therefore run
through the low critical speeds in order to reach the operating state. The axes of rotation
of the two identical vibration exciters lie parallel to a central main axis of inertia of the
vibrator body and are equidistant at a distance ξ from the center of gravity CG of the
body (see Fig. 2). It is a plane problem, the centers of gravity of the unbalanced masses
move in the same plane. It follows from the theory that in this case the synchronization
is faster and more stable, the expression is mξ 2 /Js . This means that the distance between
the centers of rotation of the imbalance exciters and the center of gravity CG has a great
influence. Where Js is the moment of inertia of the screen, m is the vibrating mass.
It follows that in the first case the rotations with in-phase unbalances are stable and
in the second case the rotations with anti-phase unbalances are stable. If the rotors move
in the same direction of rotation and if the condition is met, a circular translational
movement (circular displacement) occurs of the vibrator body mξ 2 > 2Js (see Fig. 2a).
If the opposite inequality (mξ 2 < 2Js ) is fulfilled, torsional vibrations of the vibrator
body occur (see Fig. 2b). If the rotors rotate in the opposite direction and mξ 2 > Js
applies, the opposite rotation with the phase difference π is stable. Linear translational
thrust vibrations of the vibrator body are caused (see Fig. 2c), the amplitudes are x̂ =
0 and ŷ = 2m1 e/m.
674 T. N. Dang

2 mξ 2
Fig. 2. Vibrator body with two unbalanced exciters [14]. (a) mξJS > 2; circular shift; (b) JS > 2;
torsional vibration around CG; (c) parallel shear oscillation with opposite rotation

2.2 The Vibration Model of Multi-slope Vibrating Screen

Theoretically, the vibrating screen can move in space with 6 degrees of freedom. How-
ever, rotations about the x, y-axes and translation along the z-axis are constrained. There-
fore, the vibrating screen usually only moves in the xy-plane with 3 degrees of freedom,
which is translational motion in the x, y-axes, and rotational motion around the z-axis.
The mathematical vibrating model of a multi-slope vibrating screen is shown in Fig. 3.
Where the vibrating mass m is placed on spring supports with the distance L. Circular
or oval vibration is induced mechanically by the rotation of eccentric weights (fly-
weights/flywheels) attached to a single concentric drive shaft. Two radial forces F1 and
−
→
F2 are generated by twovibration exciters.
 The sum of two radial force vectors F1 and
−
→ −
→ −
→
F2 is the force vector F F = F1 + F2 . The angle between the force vector F and the
y-axis is 0 . The drive shaft has an eccentric mass m1 and eccentric radius e, angular
velocity .
When the drive shaft rotates, radial force is generated with the same magnitude but
in the opposite phase [15]:
F1 = F2 = m1 · e · 2 (1)
According to x- and y-axis, the radial forces can be given
Fx = 2 · m1 · e · 2 · sin t · sin0 (2)

Fy = 2 · m1 · e · 2 · sin t · cos0 (3)

Applying the Newton’ second law, the differential equations of motion of the screen
in the x and y-axes are
m · ẍ + cx · x = 2m1 · e · 2 · sin t · sin 0 (4)
 Modeling and Dynamic Simulation … 675

Fig. 3. Structure schematic of a multi-slope vibrating screen

m · ÿ + cy · y = 2 · m1 · e · 2 · sin t · cos 0 (5)

where: ẍ, x; ÿ, y are the acceleration and displacement of screen in the x and y-axes. The
stiffness of springs according to x- and y-axes is cx , cy .
The solution of the differential equation (Eqs. (4) and (5)) is written under the
following equations

x(t) = x̂ · sin(t + ϕx ) (6)

 
y(t) = ŷ · sin t + ϕy (7)

where x̂ and ϕx are the vibration amplitude and phase offset in the x-axis; ŷ and ϕy are the
vibration amplitude and phase offset in the y-axis. These values are given below [10]:
2 · m1 · e · cos 0 · sin ϕx
x̂ = (8)
−m + 2 · m0
2 · m1 · e · sin 0 · cos ϕx
ŷ = (9)
−m + 2 · m0
 
−cx
ϕx = arctan (10)
m + 2 · m0 · 2
 
−cy
ϕy = arctan (11)
m + 2 · m 0 · 2
676 T. N. Dang

Because the screen moves in the xy-plane, the cylindrical spring bears the longitu-
dinal force according to y-axis, it also bears the shear force according to x-axis. The
longitudinal stiffness cy is used for cylindrical coil springs [14]:
G · d4
cy = (12)
8 · i · D3
where: d —wire diameter, (m); D—mean coil diameter (m); i—number of effective
spring coils; G slip modulus (applies to hardened rolled steel: G = 7.9 · 1010 N/m2 );
l0 —free spring length (m); l—spring length after loading (m) (see Fig. 4).

Fig. 4. Coil spring, coordinates and dimensions [14]

The coil spring can be designed for any desired ratio of longitudinal stiffness cy to
transverse stiffness cx . y = Fys /cy is the displacement due to the longitudinal force Fys ,
x = Fxs /cx is the spring deflection transverse to the spring axis due to the lateral force
Fxs (see Fig. 4). The ratio of lateral stiffness cx to longitudinal stiffness cy and correspond
to the formula:
 

cy D l0 y l0 l0
= 1.0613 − 0.6142 · tan 0.9422 − 0.6142 + 1 − (13)
cx y y D y y
The proportional deformation of springs is generally in the range l0 /D = 1 ÷
4 and y/l0 = 1/10 ÷ 2/3.

3 Dynamic Simulation Model of Multi-slope Vibrating Screen

3.1 Establish the Constraints of Simulation
To simulate multi-body dynamic, a 3D model of the screen is preliminarily designed. This
3D model will be used to manufacture and experiment. Because the screen body moves
 Modeling and Dynamic Simulation … 677

in the xy-plane of a coordinate system, the translational and rotational motions must be
defined in Fig. 5. This screen body is placed on four support springs; each support spring
has three cylindrical coil springs, in which geometry and steel properties are gradually
deteriorated. The basic parameters of a spring are: d = 0.030 m; D = 0.210 m; i =
9, l0 = 0.410 m. In fact, the force acting on the cylindrical spring is in the longitudinal
axial and transverse directions. In this simulation model, the cylindrical springs only have
longitudinal stiffness, to be equivalent to the actual working conditions of the cylindrical
spring, one actual spring is replaced by two cylindrical springs (see Fig. 5). According
to Eq. (12), the longitudinal stiffness of the spring is cy = 95967 N/m. According to
c
Eq. (13), the ratio between cy /cx are: cyx = 1.53 → cx = 95967
1.53 = 62895 N/m.

Fig. 5. Simulation model of multi-slope screen

The two drive shafts contain eight eccentric weights and are mounted on the screen
body on four bearings. In the simulation model, the rotational connection of these shafts
to the screen body is defined by pin and cylinder joints. The initial angle between
eccentric weight and y-axis (see Fig. 6) as well as the rotational direction of drive shafts
affect the amplitude and orbit of the screen. Therefore, they must be defined in the
simulation model. The vibration amplitude can be modified by adjusting the unbalanced
weights (by increasing or decreasing mass of the unbalanced weights). To generate
vibration in a vibration system, one of the shafts of the unbalanced exciter is driven by
a stationary drive motor through a propeller shaft (a shaft fitted with universal joints at
each end). Each drive shaft is directly driven from the electric motor with a motor speed
of 980 rpm. Therefore, in the simulation model, these shafts are defined servo motors
with rotational speed also equal to 980 rpm.
678 T. N. Dang

Fig. 6. Initial angle of vibration exciter and vertical plane

To determine the mass, coordinates of the center of gravity and the moment of inertia
of the screen body, normally the screen body is divided into n relatively small elements
so that each element can be considered as a point particle. The multi-slope screen is
composed of many parts with different shapes and masses. Therefore, the determination
of the above parameters is relatively complicated and the accuracy is not high. An
outstanding advantage of using 3D model can quickly determine the mass, volume,
coordinates of the center of gravity, and the moment of inertia of the object. Figure 7a
shows an illustrative example for the screen body with the following parameters: a
volume of 9.8938 · 108 m3 , a surface area of 1.599 · 108 mm2 , a mass of 7766.6 kg.
The distance of center of drive shaft and center of gravity of screen body along x, y and
z-axes are 40.96 mm, 483.24 mm and 46.39 mm, respectively. Similarly, the parameters
of the eccentric weight are shown in Fig. 7b. The unbalanced mass (m1 ) weighs 52.82 kg
and the eccentric radius (e) is 0.08546 m. By determining these parameters, it is possible
for the designer to evaluate and check the results of the dynamic simulation.

3.2 Dynamic Simulation Results

At 4 positions of the spring supports, the longitudinal forces acting on the springs in the
x, y-axes are shown in Fig. 8 and denoted by Fx1 , Fy1 ; Fx2 , Fy2 ; Fx3 , Fy3 and Fx4 , Fy4 .
The longitudinal forces acting on the springs change in time when the multi-slope
vibrating screen works (see Fig. 9). The values of longitudinal force Fx1 , Fx3 are smaller
than Fx2 , Fx4 . The graph of the longitudinal force of Fx1 is the same as Fx3 . Similarly,
the graph of the longitudinal force of Fx2 is the same as Fx4 . The different longitudinal
forces on these springs are explained by the fact that the center of gravity of the screen
and the center of the drive shaft do not coincide. The maximum value of Fx1 and Fx2 are
24545 N and 40802 N (see Fig. 9a).
 Modeling and Dynamic Simulation … 679

Fig. 7. Determination of the mass properties of screen body and eccentric weight

Fig. 8. Longitudinal forces on cylindrical springs at 4 spring supports

The longitudinal force diagrams in the y direction are similar to those in the x
direction. The graph of the longitudinal force of Fy1 is the same as Fy3 , and Fy2 is the
same as Fy4 . The maximum longitudinal forces of Fy1 and Fy2 have a value of 227839
and 220199 N (see Fig. 9b).
According to Fig. 7b, the eccentric mass (m1 ) weighs 52.82 kg and the eccentric
radius (e) is 0.08546 m. The drive shaft rotates at 980 rpm, so the corresponding angular
velocity  is 102.57 rad/s. The sum of the radial forces of the four eccentric masses is
F1 = F2 = 4 · m1 · e · 2 = 4 · 52.82 · 0.08546 · 102.572 = 189959 N. The drive
shaft is supported by two shaft ends, the reaction force acting on one end of the shaft is
R = F1 /2 = 94979.5 N. Simulation results of reaction forces on bearing supports are
shown in Fig. 10, these forces have the same value, and they are denoted as R1 , R2 , R3 , R4 .
When the drive shaft rotates, the screen body oscillates on the supporting springs,
the reaction forces acting on the supports are not constant (see Fig. 11). As shown in
Fig. 11, the increase in the reaction force occurs in the first second after starting the
drive, and then the screen stabilizes its operation. The full stabilization of the vibrations
680 T. N. Dang

Fig. 9. Value of the longitudinal forces acting on the springs

Fig. 10. Reaction forces acting on four bearing supports

allows the determination of the amplitude in the steady state. The graph of the radial
force of R1 is the same as R2 . Similarly, the graph of R3 is the same as R4 . The maximum
value of R1 is 470371 N, while the maximum value of R3 is 236524 N.
The cause of the different reaction forces between the left and right side of the
screen body (R1 , R2 = R3 , R4 ) is that the center of gravity of the screen body is not in
the vertical plane of symmetry along the side of the screen. In other words, the mass of
the two sides of the screen is different.
Dynamic simulation can determine the amplitude, velocity, and acceleration of the
multi-slope vibrating screen at any point. This example examines the point at the center
of the drive shaft with the coordinate system as shown in Fig. 12.
The vibration amplitude, velocity, and acceleration directly affect the movement of
the material on the screen surface as well as the capacity and efficiency of the screen.
The determination of these parameters at different positions of the screen surface is
usually carried out by measurement. By using dynamic simulations, these parameters
can be quickly determined. Figure 13 shows the vibration amplitude (x̂, ŷ), velocity
(ẋ, ẏ), and acceleration (ẍ, ÿ) at the center of the total radial force in x and y-direction.
 Modeling and Dynamic Simulation … 681

Fig. 11. Value of radial forces acting on bearings of eccentric block shaft

Fig. 12. Coordinate system at the center of the drive shaft

Their maximum values are determined as follows: x̂ = 0.5 mm; ŷ = 2 mm; ẋ =

150 mms , ẏ = 350 s ; ẍ = 38797 s2 ; ÿ = 86625 s2 .
mm mm mm

Theoretically, the screen body oscillates in the y-axis. In this example, the simulation
model has different center of gravity and center of drive shaft, so the orbit of the body
screen is shown in Fig. 14. For easier observation, this orbit has been magnified 500
times.
In order to verify the correctness of dynamic model and simulation model of multi-
slope vibrating screen, experimental prototype of vibration screen which takes the same
parameters to 3D model is manufactured (see Fig. 15). Due to the lack of equipment to test
force, velocity, and acceleration, the vibrating screen is tested to find out the amplitude
and orbit of the screen surface. Experimental results show that the screen body moves
smoothly, there is no impact between the screen body and the screen support, and there
682 T. N. Dang

Fig. 13. Vibration amplitude, velocity, and acceleration at the center of the drive shaft

Fig. 14. The orbit of the body screen

is no noise at the bearings. The motion of the screen is inclined at a small angle to the
y-axis, the vibration amplitude is from 2–4 mm. For easier observation, the orbit of the
screen surface is also magnified as shown in Fig. 15.

4 Conclusions and Discuss

By building 3D models and simulating dynamic instead of mathematical vibration mod-
els, the model of 3D dynamic simulation has approached closer to the actual working
conditions of the vibrating screen. The main results obtained in this study are:
 Modeling and Dynamic Simulation … 683

Fig. 15. Measurement the vibration amplitude and orbit of the screen body

• The vibration model of the multi-slope vibration screen has been simulated based
on the multi-body dynamic simulation, and the simulation results are presented with
clear images as well as graphs.
• The forces, acting on the bearing springs of the drive shaft, change during the working
cycle of the screen, the values of these forces are not the same. Accurate determination
of the maximum value of these forces allows the selection of the appropriate support
spring and bearing, increasing their service life.
• The values of vibration amplitude, velocity, acceleration as well as the orbit of the
screen surface at any point can be quickly and easily determined. The simulation
model allows to change the unbalanced mass on the drive shaft and the rotational
speed of the shaft to find out the vibration amplitude, velocity, acceleration, and orbit
of the screen surface as desired by the designer.
• The multi-slope vibrating screen is manufactured and measured the vibration ampli-
tude and orbit of screen surface. It is necessary to be equipped with force measuring
devices, so that the force at the bearings and supporting springs can be measured. That
is the basis for evaluating the difference between dynamic simulation and experiment.

References
1. JXSC: How to easily solve the problem of vibrating screen bearing heating and damage.
https://mineraldressing.com/blog/. (2021)
2. Shah, K.P.: Construction, working and maintenance of electric vibrators and vibrating screen.
www.practicalmaintenance.net. (2018)
3. Krot, P., Zimroz, R., Michalak, A., Wodecki, J., Ogonowski, S., Drozda, M., Jach, M.: Devel-
opment and verification of the diagnostic model of the sieving screen. Shock Vib. 2020, Article
ID 8015465 (2020)
4. Meyers, A., Mylec, A.B.: Performance of banana screens in D and R applications. Tech. Rep.
C7048, ACARP, Brisbane City, QLD, Australia (2000)
5. Dong, K.J., Yu, A.B., Brake, I.: DEM simulation of particle flow on a multi-deck banana
screen. Miner. Eng. 22(11), 910–920 (2009)
6. Xiao, J.Z., Tong, X.: Screening characteristics and screening efficiency of a vibrating screen
with a swing trace. Particuology 11(5), 601–606 (2013)
684 T. N. Dang

7. Li, Z.F., Tong, X., Zhou, B., Wang, X.Y.: Modeling and parameter optimization for the design
of vibrating screens. Miner. Eng. 83, 149–155 (2015)
8. Jiang, H., Zhao, Y., Qiao, J., et al.: Process analysis and operational parameter optimization
of a variable amplitude screen for coal classification. Fuel 194, 329–338 (2017)
9. Jiang, H., Qiao, J., Zhou, Z., et al.: Time evolution of kinematic characteristics of variable-
amplitude equal-thickness screen and material distribution during screening process. Powder
Technol. 336, 350–359 (2018)
10. Li, Z., Tong, X., Zhou, B., Ge, X., Ling, J.: Design and efficiency research of a new composite
vibrating screen. Shock Vib. 2018, Article ID 1293273 (2018)
11. Du, C., Gao, K., Li, J., Jiang, H.: Dynamics behavior research on variable linear vibration
screen with flexible screen face. Adv. Mech. Eng. 2014, Article ID 957140 (2014)
12. Liu, Y., Meng, G., Suo, S., Li, D., Wang, A., Cheng, X., Yang, J.: A theoretical rigid body
model of vibration screen for spring failure diagnosis. Mathematics (Switzerland) 7, 246, 16
pages (2019)
13. Yang, X., Wu, J., Jiang, H., Qiu, W., Liu, C.: Dynamic modeling and parameters optimization
of large vibrating screen with full degree of freedom. Shock Vib. 2019, Article ID 191578
(2019)
14. Dresig, H., Holzweißig, F.: Maschinendynamik. Springer-Verlag Berlin Heidelberg (2011)
15. Van Khang, N. Dao dong ky thuat. Nha xuat ban khoa hoc va ky thuat, Hanoi (2004)
 Module Integrated Converters and Independent
MPPT Technique

Bui Van Hien1 , Truong Viet Anh2 , Nguyen Duc Minh3 , Trinh Trong Chuong4(B) ,
Y. Do Nhu5 , and Trieu Viet Phuong6
1 Faculty of Mechanical—Electrical and Computer Engineering, School of Engineering and
Technology, Van Lang University, Ho Chi Minh City, Vietnam
2 Faculty of Electronics and Electrical, HCMC University of Technology and Education, Ho
Chi Minh City, Vietnam
3 Institute of Energy Science, Vietnam Academy of Science and Technology, Hanoi, Vietnam
4 Faculty of Electrical Engineering, Hanoi University of Industry, Hanoi, Vietnam
chuonghtd@haui.edu.vn
5 Faculty of Electro—Mechanics, Hanoi University of Mining and Geology, Hanoi, Vietnam
6 Vietnam Standards and Quality Institute, Hanoi, Vietnam

Abstract. As the irradiance changes, the current at the maximum power gen-
erating point (MPP) of the photovoltaic (PV) cell system fluctuates. But when
the temperature changes, its voltage is affected more. In actual conditions when
operating in a partial shade (PSC) condition, fluctuation of these two parame-
ters makes it difficult to determine the global maximum power point tracking
(GMPPT) accurately. The compatibility between the PV system topology and the
algorithm plays an important role in improving the performance of the whole
system. This paper proposes a PV module architecture combined with inverters
and a standalone MPPT technique operating under non-uniform temperature and
radiation conditions. A modified P and O algorithm with multiple inrush current
values as an input parameter is simulated in the PSC. The research results in this
paper have shown that the proposed solution not only has outstanding speed but
also has the ability to significantly improve performance compared to published
methods under the same operating conditions.

Keywords: Partial shade · Photovoltaic (PV) · P–V characteristics · Integrated

inverter module · Standalone MPPT

Abbreviations

PV Photovoltaic
SC Serial Configuration
PC Parallel Configuration
SPC Serial–Parallel Configuration
PSC Partial Shading Conditions

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 685–698, 2023.
https://doi.org/10.1007/978-3-031-22200-9_73
686 B. Van Hien et al.

GMPPT Global Maximum Power Point Tracking

LMPPT Local Maximum Power Point Tracking
MPPT Maximum PowerPoint Tracking
PSO Particle Swarm Optimization

1 Introduction
This paper is an extension of the work presented at the International Conference on
Smart power and internet energy systems [1]. Which has been rated to deliver outstanding
performance in non-uniform operating conditions compared to several recently proposed
solutions. Its outstanding advantage is that it does not produce many extremes in PV
strings under all working conditions. From there, it is easy to choose the MPPT solution
to achieve the highest efficiency with significantly improved speed. This is also the
main reason to improve the performance of PV systems in series strings. Practice shows
that Series Connection (SC) gives us an advantage in voltage but limits current. Which
causes the entire PV series to operate the same current value even when they receive
different radiation levels. This is the main cause leading to the phenomenon of heating,
reversing the working state in some positions, and even destroying the entire chain link
[2, 3]. In addition, the Parallel Configuration (PC) will give a fairly stable voltage value,
producing the least extremes in all operating conditions, but the disadvantage is that the
output current value is quite large. It makes it difficult to design and select control keys
and has high power losses [4, 5]. Meanwhile, the demand for the installation and use
of grid-tied solar power systems is growing rapidly [6–8]. In order to achieve applied
power values in these cases requires PV systems sufficient to generate the appropriate
voltage and current.
Several publications have proposed PV interconnection configurations to improve
efficiency, increase convergence speed, reduce costs and reduce losses under all operating
conditions. However, most of them are still based on two basic types of links, SC and PC.
Among them, the most widely used type is Serial–Parallel Configuration (SPC). Here,
PV modules are linked together in series to achieve the desired voltage value. Then, the
strings are connected in parallel to form an array to increase the system current [9]. In
actual operation, when the radiation changes, the surface temperature of the PVs also
fluctuates. It is one of two factors that make it difficult to GMPPT, which has not been
mentioned in [1].
In this paper, a system of 3 × 3 PV modules linked in series–parallel configuration is
proposed. It is operated not only under varying radiation conditions but also with varying
operating temperatures. In which, shading patterns at all different positions are tested in
PSIM platform. Simultaneous PSC scenarios are operated with the bypass-diode method
and an improved optimization algorithm under the same operating conditions [10, 11].
The simulation results are compared with some of the most recently proposed methods
[12, 16–19].
 Module Integrated Converters and Independent MPPT Technique 687

2 Environmental Influence on PV Properties

2.1 Photovoltaics
The model for each photovoltaic cell that can be found in [13] includes a light-controlled
current source in parallel with a diode. A resistor in parallel has a fairly large value (Rp )
and a resistor in series (Rs ) has negligible resistance. The relationship between the output
current and the voltage of the PV cell is shown by Eq. (1).
 q(V +I .Rs )  V + I .R
s
I = ISC − I0 e kT −1 − (1)
Rp
A PV system includes a plurality of cells interconnected to achieve a desired voltage
and current. At that, the output current of a system of N s module in a sequence and N p
parallel sequences is presented in Eq. (2).
 
q(V +Ns IRs ) V + Ns IRs
I = Np ISC − Np I0 e Ns kT − 1 − (2)
Ns Rs

2.2 Effect of Operating Conditions

PV plants spread over a fairly large area are often operated in a heterogeneous situation
due to PSC. This is the cause of changing the shape of the overall characteristic curves,
causing the appearance of many MPPs in the strings, making it difficult to accurately
determine the GMPP. Figure 1 presents the P–V curve of the system under two different
operating conditions.
Even if the working condition is uniform whole system, it is not possible to assure
stability throughout the operation period. Published studies have shown that when irradi-
ance varies—often below standard conditions, the MPP coordinates are greatly affected
by fluctuations in the current. Meanwhile, if the surface temperature changes—usually
higher than the standard condition temperature, then the voltage parameter will have
more impact on the stability of the MPP (Fig. 2) [14, 15].
Under real conditions, when the radiation increases, the temperature also increases
and vice versa. This causes the MPP coordinates to fluctuate and is affected by both
current and voltage parameters. Moreover, when these two parameters change hetero-
geneously, it will be more difficult to determine the GMPP among many extremes. It
can be explained in detail from Fig. 1 that if only the irradiance is fluctuating, then the
GMPP will be equivalent to the MPP3 point. However, as the radiation on a module
increases, so does its surface temperature. This makes the power increase undesirable
and leads to a change in the structure of the curve. If the temperature fluctuations are
more than the radiation then it is likely that the GMPP will move to the MPP2 position.
It shows that the GMPPT in real conditions is much more complicated.

2.3 Proposed Solution

Operating conditions greatly affect the performance, speed and accuracy of MPPT solu-
tions. To solve this problem, a system of 3 modules with three different irradiance levels
is proposed. Their P–I characteristic curves are presented in Fig. 3.
688 B. Van Hien et al.

Fig. 1. P–V characteristics under operating conditions

Fig. 2. Effect of radiation and temperature on the P–V and I–V curves

If the start current value I0 is the same as the current position (Fig. 3), it means that
only the two modules receiving the higher radiation are capable of emitting power. The
module that receives the least energy will be ignored, so the performance is reduced.
Moreover, at this boot site, the I03 series is quite far from its MPP so it takes a long time
to converge. If the start value I0 > I2 is selected to increase the speed for the largest
module in this case, the two modules corresponding to lines I1 and I2 will be ignored.
Conversely, if you want to obtain the full power of all three modules, the starting
value I0 < I1 . But with a low boot value will increase the system’s MPPT search time.
If the step size is increased to overcome this disadvantage, the system performance is
reduced again.
 Module Integrated Converters and Independent MPPT Technique 689

Fig. 3. P–I characteristics of PV module under different conditions

To resolve this difficulty, the article proposes a standalone MPPT solution for mod-
ules that simultaneously use multiple boot values at the same time in order to find the
optimal value for each module and increase convergence speed. The structure of the
proposed solution is presented in Figs. 4 and 5.

Fig. 4. Proposed solution structure

690 B. Van Hien et al.

Fig. 5. Simulated shading states

The proposed solution consists of two parts; the first is to limit the search range
through initialization values. Then there is the basic GMPPT part based on traditional P
and O algorithm. The proposed solution flowchart shown in Fig. 6 includes the following
steps:

• Step 1: Select 4 random D[i] start values in the search range. Thereby, the values of
V[i] and I[i] are determined respectively.
• Step 2: define the MPP search scope. Calculate the power P[i] corresponding to the
values D[i] in step 1. Select the largest value among the P[i] to be the reference value
for the next iteration. The values D, V and I corresponding to this power level are also
selected as reference values and initialized for the MPPT solution when changing a
number of D increments.
• Step 3: determine power deviation P and current I. The MPPT unit will measure
the current and voltage values to calculate the power and current after having changed
a quantity D. Then calculate the power and current error compared to the reference
value in step 2.
Module Integrated Converters and Independent MPPT Technique 691

Fig. 6. Flowchart of the proposed algorithm

692 B. Van Hien et al.

• Step 4: check the convergence of the algorithm. The solution is considered convergent
when P < ε. Conversely, the reference current value will be adjusted via parameter
adjustment D as follows.
• If P.I > 0, then increase the reference current value Iref.
• If P.I < 0, then reduce the reference current value Iref .

Repeat the loops from step 3.

3 Results and Discussion

The proposed solution is tested on a configuration of 3 × 3 modules of type HTM330PA-

72 (Fig. 4) with the following parameters: Isc = 9.29 A, Voc = 46.15 V, Impp = 8.69 A,
Vmpp = 37.89 V. Shading states are listed in Fig. 5 from left to right and from top to
bottom: column shading, row shading, corner shading, double corner shading, diagonal
shading, and center shading. The first six cases only change irradiance when partially
shaded. The scenario is repeated in the following cases but both radiation and temperature
parameters are changed. With the performance shown in our previous study. Simulation
states that will be performed on different structures to compare results include:

(i) Use bypass-diode for protection in cases of partial shade (PSC).

(ii) Use the proposed configuration to MPPT the system (PMPPT ).
(iii) Using bypass-diode and combined with the improved PSO algorithm in the
proposed study introduced previously.

Table 1 presents the MPPT results when simulating the proposed configuration com-
pared with the traditional method in the above cases. The obtained results show that in all
working conditions, the proposed solution always proves to be capable of significantly
improving power generation efficiency. Specifically, in the case of row shading (No. 2),
it was possible to increase the efficiency by more than 28% compared to the traditional
method. This can be explained as when shading in rows, the operating conditions of the
modules in a chain will be different, so there are many extremes. Which directly affects
the ability to generate electricity in the chain as well as the whole system. Likewise, in
the case of variations in both irradiance and temperature in rows (No. 8), the performance
improvement is also quite impressive.
This result is confirmed again when the improved performance of the diagonal shad-
ing case (No. 5 and No. 11) is 21.51% and 20.21%, respectively. This advantage has also
been demonstrated in our previous study when considering SC in PSC. Meanwhile, the
column shading cases (No. 1 and No. 7) have insignificant performance improvement.
The cause of this result is that when shading in columns, the operating conditions of the
modules in a series are identical, so both solutions can easily approach MPP equally.
In addition, the data in Table 1 also shows that the convergence speed of the proposed
solution is quite impressive at the fastest level of 0.006 s. The more complicated cases
make the search time last in the range of 0.024 s (No. 11).
When comparing the performance of the proposed solution compared with the
improved PSO method, the results obtained in Fig. 7 show that: when the operating
 Module Integrated Converters and Independent MPPT Technique 693

Table 1. MPPT results in cases

Pmppt −Psc
No Power (W) P = Psc × 100% Speed (s)
Psc Prompt
1 1440.55 1441.45 00.06 0.008
2 1624.66 2083.26 28.23 0.007
3 2028.78 2275.44 12.16 0.013
4 1815.50 2144.42 18.12 0.013
5 1590.23 1932.29 21.51 0.007
6 1997.45 2022.41 01.25 0.006
7 1316.35 1317.21 00.07 0.008
8 1461.94 1795.49 22.82 0.020
9 1747.38 1993.17 14.07 0.020
10 1629.55 1787.99 09.72 0.023
11 1469.59 1766.54 20.21 0.024
12 1546.45 1774.34 14.74 0.006

conditions on each series are homogeneous (shading in columns), the solutions are quite
different. The similarity in MPPT performance. They differ only in the speed of con-
vergence. With three different strings, the proposed solution gives three representative
results about the value of D. In the first group (column 1) the most shaded also corre-
sponds to the longest search time (D1 = D2 = D3 ) is 0.006 s. Meanwhile, the group
that received the highest radiation (column 3) had the fastest convergence rate (D7 = D8
= D9 ). With the same operating conditions, the solution using PSO shows that it takes
more time to converge at 0.036 s.
In another scenario, when the operating conditions are more complex (No. 11),
the results of comparing the MPPT capabilities of the solutions shown in Fig. 8 show
that the proposed solution not only has convergence speed. Outstanding but also has
quite impressive power generation efficiency. Its performance improvement over the
traditional method is up to 20.21%. Although the energization ratio D stops at 0.008 s
and the output stabilizes later at 0.024 s, it is still far ahead of the PSO solution when it
converges at 0.037 s. In this case, the proposed solution has only two representative D
values because this diagonal shading produces two extrema at each series. The values
D1 = D5 = D12 , and the rest are equal to D6 .
When changing operating conditions continuously during the simulation, the results
obtained in Figs. 9, 10 and 11 show that: the proposed solution always gives the perfor-
mance greater than or equal to the other method. Its response speed is about 0.01 s on
average, which is also a prominent advantage compared to 0.38 s for the improved PSO
algorithm. Thereby, the proposed solution not only significantly improves the MPPT
performance but also increases the convergence speed in the continuously changing and
complex PSC.
694 B. Van Hien et al.

Fig. 7. Comparison of MPPT performance in No. 1

Fig. 8. Comparison of MPPT performance in No. 11

The MPPT efficiency of the solutions for all shading cases is presented in Fig. 12.
From the column chart, it can be seen that: the proposed solution has superior power
generation ability compared to the previously proposed methods in all operating condi-
tions. Especially this ability is best demonstrated in the cases of shading in rows, corners,
and diagonals. Which is 100% efficient in previous research when surveying on a series
[1]. Meanwhile, under these conditions, the PSO algorithm proved to be less efficient.
When operating conditions are uniform across the series, all solutions have the same
power generation efficiency (No. 1 and No. 7) as analyzed above.
In addition, the performance, convergence speed, and convergence speed of the
proposed solution are also compared with some optimized algorithms that have been
improved recently. The figures shown in Table 2 show outstanding MPPT capability
with impressive convergence time.
From the obtained simulation results, it is shown that:
 Module Integrated Converters and Independent MPPT Technique 695

Fig. 9. Comparison of MPPT performance when changing from No. 7 to No. 4

Fig. 10. Comparison of MPPT performance when changing from No. 6 to No. 10

• The GMPPT efficiency of the proposed solution can be improved by over 28% com-
pared to the traditional method. Especially in complex shading states and changing
operating conditions, both radiation and temperature parameters.
• Convergence speed is also an advantage of the solution as most cases show that
MPPT takes only a few milliseconds. This can be explained from Figs. 7 and 8, that
by comparing the initial D values to find the most potential location, the number
of unnecessary iterations will be greatly reduced and the convergence speed will be
greatly reduced. It also reflects the comparison results between the proposed solution
and the improved PSO algorithm in one of our previous studies. Which proved to be
quite effective when it reached 0.08 s. But when operating under the same conditions
in this study, it only reached 0.036 s compared to 0.006 s (No. 1).

In addition, it also presents the possibility to choose different initialization values for
modules with different solar irradiance. This greatly reduces the search time compared
696 B. Van Hien et al.

Table 2. Comparison of the proposed solution with some other solutions

Algorithms Converter type Speed of tracking (s) Efficiency (%) Complexity

Proposed Buck-boost 0.006 100 Simple
E P and O Boost 0.12 100 Simple
[12]
M INC Boost 0.05 – Moderate
[16]
I GWO SEPIC 0.4 100 Complicated
[17]
N INC Boost 0.3 98 Moderate
[18]
M ABC Buck-boost 0.39 99.91 Complicated
[19]

Fig. 11. Comparison of MPPT performance under continuous change

to the traditional method when all are selected for a fixed reference value. Which can
cause a drop in system capacity by ignoring the capacity of some modules that receive
less power.

4 Conclusion
The paper proposes a standalone GMPPT solution based on the improved traditional
MPPT method to enhance the performance and convergence speed. The obtained results
proved that the solution can increase the maximum efficiency by 28.23% and achieve
the fastest convergence speed of 0.006 s. It also shows that simply by limiting the exact
search scope without using complex algorithms, it is still possible to improve the system’s
power generation efficiency at a reasonable cost.
 Module Integrated Converters and Independent MPPT Technique 697

Fig. 12. Comparison of GMPPT performance

The obtained simulation results show the outstanding advantages of the proposed
solution compared with some recently published solutions. It helps to improve the
efficiency of solar power generation in more realistic conditions when both radiation
and temperature parameters are considered. In short, this is a reliable MPPT improve-
ment solution that helps the PV system operate stably and effectively in a non-uniform
operating environment.

Acknowledgements. This research is funded by Van Lang University, website: https://www.van

languni.edu.vn. The authors would like to show their gratitude to the support from Van Lang Uni-
versity, Ho Chi Minh City University of Technology and Education, Vietnam, which coordinates
the research project T2021-64TÐ.

Appendix

I0 —saturation current of the diode (a);

q—charge of the electron (1,602.10–19 C);
k—Boltzman constant (1,381.10–23 J/K);
T—contact layer temperature (K);
I—PV output current (A);
ISC —PV short circuit current (A).

References
1. Chuong, T.T., Minh, N.D., Hien, B.V., Yang, G., Anh, T.V.: Optimizing the performance of
the photovoltaic system using the micro DC–DC converter. In: 3rd International Conference
on Smart Power and Internet Energy Systems, pp. 392–397 (2021)
2. Ramaprabha, R., Mathur, B.L.: A comprehensive review and analysis of solar photovoltaic
array configurations under partial shaded conditions. Int. J. Photoenergy 2012, 120214 (2012).
https://doi.org/10.1155/2012/120214
698 B. Van Hien et al.

3. Gao, L., et al.: Parallel—connected solar PV system to address partial and rapidly fluctuating
shadow conditions. IEEE Trans. Ind. Electron. 56(5), (2009)
4. Alik, R., Sutikno, T., Jusoh, A.: A review on perturb and observe maximum power point
tracking in photovoltaic system. TELKOMNIKA 13(3), 745–751 (2015). https://doi.org/10.
12928/TELKOMNIKA.v13i3.1439
5. Belhachat, F., Larbes, C.: Modeling, analysis, and comparison of solar photovoltaic array
configurations under partial shading conditions. Sol. Energy 120(2015), 399–418 (2015)
6. Minh, N.D., Chuong, T.T., Huy, B.V., Cuong, Q.D., Thanh, B.D.: Research and design of
inverter applied in solar PV systems connected to distribution grid. J. Electr. Eng. 7(1), (2019)
(Serial Number 26)
7. Minh, N.D., Do Nhu Y., Chuong, T.T.: Development of algorithm to identify the global
optimized point of solar Photovoltaics panel under the condition of non-uniform solar array
on the surface. J. Sci. Technol. 56(6), (2020)
8. Sacli, M., Ayan, O., Silsupur, M., Empre turkay, B.: Investigation of power quality analysis
of single-phase inverter topologies for solar energy systems. In: International Conference on
Condition Monitoring, Diagnosis and Maintenance. Bucharest, Romania (2017)
9. Pachauri, R., Mahela, O.P., Sharma, A., Bai, J., Chauhan, Y.K., Khan, B., Alhelou, H.H.:
Impact of partial shading on various PV array configurations and different modeling
approaches: a comprehensive review. IEEE Access 8, 181375–181403 (2020)
10. Bui, H.V., Truong, V.A., Quach, T.H.: Optimizing the maximum power generation point of
PV cells under shading conditions. J. Sci. Technol. Dev. 3(1), 326–338 (2020)
11. Luong, X.T., Bui, V.H., Do, D.T., Quach, T.H., Truong, V.A.: An improvement of maximum
power point tracking algorithm based on particle swarm optimization method for photovoltaic
system. In: 5th International Conference on Green Technology and Sustainable Development,
HCM City, Vietnam, pp. 53–58 (2020). https://doi.org/10.1109/GTSD50082.2020.9303110
12. Mohammad, A.N.M., Radzi, M.A.M., Azis, N., Shafie, S., Zainuri, M.A.A.M.: An enhanced
adaptive perturb and observe technique for efficient maximum power point tracking under
partial shading conditions. Power Electron. Appl. Power Energy Syst., Sci. 10(11), 3912
(2020)
13. Baba, A.O., Liu, G., Chen, X.: Classification and evaluation review of maximum power point
tracking methods. Sustain. Futur. 2, (2020)
14. Kadri, R., Andrei, H., Gaubert, J.-P., Ivanovici, T., Champenois, G., Andrei, P.: Modelling of
the photovoltaic cell circuit parameters for optimum connection model and real-time emulator
with partial shadow conditions. Energy 42, 57–67 (2012)
15. John, R., Sheik Mohammed, S., Zachariah, R.: Variable step size Perturb and observe MPPT
algorithm for standalone solar photovoltaic system. In: IEEE International Conference on
Intelligent Techniques in Control, Srivilliputhur, India (2017)
16. Shang, L., Guo, H., Zhu, W.: An improved MPPT control strategy based on incremental
conductance algorithm. Prot. Control Mod. Power Syst. 5, 14 (2020)
17. Debnath, D., et al.: Improved grey wolf assists MPPT approach for solar photovoltaic system
under partially shaded and gradually atmospheric changing condition. Int. Energy J. 20,
87–100 (2020)
18. Xu, L., Cheng, R., Yang, J.: A new MPPT technique for fast and efficient tracking under fast
varying solar irradiation and load resistance. Int. J. Photoenergy 2020, (2020)
19. Li., N., et al.: Maximum power point tracking control based on modified ABC algorithm
for shaded PV system. In: AEIT International Conference of Electrical and Electronic
Technologies for Automotive, 2–4 July 2019, Turin, Italy (2019)
 Multi-criteria Decision Making in Wire-EDM
Process Using MARCOS Method

Trieu Quy Huy1 , Tran Huu Danh2 , Bui Thanh Danh3 , Nguyen Van Cuong3 ,
Nguyen Manh Cuong4 , and Nguyen Thanh Tu4(B)
1 University of Economics—Technology for Industries, Hanoi, Vietnam
2 Vinh Long University of Technology Education, Vinh Long, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nguyenthanhtucnvl@tnut.edu.vn

Abstract. This paper present the results of a study on the application of the Multi-
Criteria Decision Making (MCDM) technique to select the best input parameters in
Wire Electrical Discharge Machining (WEDM) 90CrSi tool steel. In the study, the
Measurement of Alternatives and Ranking according to COmpromise Solution
(MARCOS) method was used to solve the MCDM problem, and the Entropy
method was used to calculate the weights of the criteria. In addition, six input
parameters are required: the cutting voltage VM, the pulse on time Ton , the pulse
off time Toff , the servo voltage SV, the wire feed WF, the feed speed SPD, and
the workpiece cutting radius R. In addition, a 27–2 design experiment was carried
out and a total of 32 different experimental setups were performed. The MCDM
problem has been solved. The best experimental setup, according to the results of
this work, is experiment no. 7 with the following input parameters: VM = 9 V,
Ton = 12 s, Toff = 13 s, SV = 25 V, WF = 8 mm/min, SPD = 4.5 mm/min, and
R = 9 mm.

Keywords: WEDM · MCDM · MARCOS method · Surface roughness · Cutting

speed · 90CrSi

1 Introduction
In mechanical processing, it is frequently necessary to determine the process input param-
eters to meet multiple criteria at the same time, but these criteria, such as minimum sur-
face roughness (SR), maximum material removal rate (MRR), or minimum tool wear,
contradict each other. To achieve a low SR, the depth of the cut and the feed rate must
be reduced, resulting in a lower MMR. Increasing the MMR will also necessitate an
increase in the depth of the cut and feed rate, as well as an increase in the SR and tool
life. Solving the MCDM problem to select the best solution for a mechanical machin-
ing process is very common in this case. This problems have been solved in milling
[1, 2], turning [1, 3, 4], grinding [1, 5], electrical discharge machining (EDM) [6], and
powder-mixed EDM [7].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 699–706, 2023.
https://doi.org/10.1007/978-3-031-22200-9_74
700 T. Q. Huy et al.

WEDM is an unconventional machining process used to process conductive materials

and parts with narrow slots. Determining the best cutting mode for WEDM is difficult
due to the large number of input parameters such as VM, Ton , Toff , SV, WF, SPD, and
so on. As a result, many studies have used the MCDM problem to solve this problem.
To date, various MCDM methods have been used to determine the best alternative
in WEDM. The Multi-Objective Optimization Ratio Analysis (MOORA) method was
used in [8] to find the optimum input factors for wire-EDM Inconel 718. The MOORA
was also used in [9] when cutting D3 die steel and in [10] when machining magnesium
AZ91 alloy. In this study, four input factors including pulse-on-time, pulse-off-time,
servo voltage, and wire tension were investigated. In [11] the Technique for Order of
Preference by Similarity to Ideal Solution (TOPSIS) method was used to determine the
best cutting mode when working with magnesium AZ91 alloy. In [12] two methods
including MOORA and TOPSIS were applied to solve MCDM problem. In [13], the
Weighted Aggregates Sum Product Assessment (WASPAS) method was chosen to solve
the MCDM problem when machining Inconel 718 with three criteria: kerf width, material
removal rate, and tool wear rate. In [14] the Operational competitiveness rating analysis
(OCRA) method was used to get optimum input factors when cutting aluminium metal
matrix material.
This paper presents the results of an MCDM study in WEDM 90CrSi. In the study,
the MARCOS method was used as the MCDM method, and the Entropy method was
used to calculate the weights of the criteria. Furthermore, SR and CS were chosen as
criteria. In addition, seven input parameters were investigated: VM, Ton , Toff , SV, WF,
SPD, and R. Furthermore, 32 different experiential runs were carried out. The MCDM
problem has been solved, and the best experimental configuration has been proposed.

2 Methodology

2.1 Method for MCDM

The MARCOS method was used to solve the MCDM problem in this work. To use this
method, the following steps must be taken [15]:
Step 1: Creating the first decision-making matrix:
⎡ ⎤
x11 · · · x1n
⎢ x21 · · · x2n ⎥
⎢ ⎥
X =⎢ . . ⎥ (1)
⎣ .. · · · .. ⎦
xmn · · · xmn

In which, m is the number of the alternative; n is the criteria number.

 Multi-criteria Decision Making in Wire-EDM … 701

Step 2: Adding an ideal (AI) and anti-ideal solution (AAI) to the initial decision-
making matrix to create an extended initial matrix:
⎡ ⎤
AAI xaa1 · · · xaan
A1 ⎢⎢ x11 · · · x1n ⎥
⎥
⎢ ⎥
A2 ⎢ x21 · · · x2n ⎥
X = . ⎢ . . . ⎥
⎢ (2)
.. ⎢ .. .. .. ⎥ ⎥
⎢ ⎥
Am ⎣ xm1 · · · xmn ⎦
AI xai1 · · · xain
 
In the preceding equation, AAI = min xij and AI = max xij if the requirement

set with criterion j is as large as possible; AAI = max xij and AI = min xij if the
requirement set with criterion j is as small as possible; I = 1, 2,…, m; j = 1, 2,…, n.

Step 3: The extended initial matrix is normalized (X). The normalized matrix N =
nij m×n can be calculated using:

uij = xAI /xij (3)

uij = xij /xAI (4)

Equation (3) is used when the criterion j is as small as possible, and Eq. (4) is used
when the criterion j is as large as possible.
Step

4: Using the following equation to calculate the weighted normalized matrix
C = cij m×n :

cij = uij · wj (5)

In which, wj is the weight coefficient of criterion j.

Step 5: Calculating the utility of alternatives K i − and K i + by:

Ki− = Si /SAAI (6)

Ki+ = Si /SAI (7)

In (6) and (7), S i can be found by:


m
Si = cij (8)
i=1

Step 6: Calculating the utility function f (K i ) of alternatives by:

Ki+ + Ki−
f (Ki ) =   (9)
1−f Ki+ 1−f Ki−
1+  +
f Ki
+  −
f Ki
702 T. Q. Huy et al.

where f (K i − ) represents the utility function associated with the anti-ideal solution and
f (K i + ) represents the utility function associated with the ideal solution. These functions
can be discovered by:
 
f Ki− = Ki+ / Ki+ + Kii (10)

 
f Ki+ = Ki− / Ki+ + Kii (11)

Step 7: Ranking the alternatives based on the final utility function values to find the
alternative with the highest utility function value.

2.2 Method for Calculation of the Weight of Criteria

The Entropy method was used in this work to determine the weights of the criteria. The
following steps can be used to implement this method [16].
Step 1: Calculating indicator normalized values:
xij
pij = m (12)
m+ 2
i=1 xij

Step 2: Calculating the Entropy for each indicator:

   
m

  m 
m
mej = − pij × ln pij − 1 − pij × ln 1 − pij (13)
i=1 i=1 i=1

Step 3: Calculating the weight of each indicator:

1 − mej
wj = m  (14)
j=1 1 − mej

3 Experimental Setup

For wire-EDMing 90CrSi steel, an experiment was carried out to determine the best
solution that meets two criteria at the same time: minimum surface roughness SR and
maximum cutting speed CS. For this experiment, seven input parameters were specifi-
cally selected (Table 1). A 2-levels 1/4 factorial experimental design was also chosen.
Therefore, 27–2 = 32 test runs will be conducted. The following items were included in
the experimental setup: a Fanuc Robocut-1 iA EDM machine (Fig. 1); brass wire with a
diameter of 0.25 mm (Taiwan); workpiece material 90CrSi; samples with a cross section
of 22 × 22 mm2 ; dielectric fluid: deionized water; surface roughness tester: Mitutoyo
178-923-2A, SJ-201 (Japan).
Following the experiment, the surface roughness of the workpieces was measured
and the cutting speed was calculated. Table 2 displays the various levels of input factors
as well as the results of the output response (Ra and CS).
 Multi-criteria Decision Making in Wire-EDM … 703

Table 1. Input factors and their levels

Factor Code Unit Low High

Cutting voltage VM V 3 9
Pulse on time Ton µs 8 12
Pulse off time Toff µs 13 18
Server voltage SV V 25 35
Wire feed WF mm/min 8 12
Feed speed SPD mm/min 2.5 4.5
Workpiece cutting radious R mm 3 9

Fig. 1. WEDM machine for experiment

4 Determining the Best Alternative in WEDM 90CrSi Tool Steel

This section discusses determining the best experimental setup when using the MARCOS
method for MCDM problem and calculation the criteria weights when using the Entropy
method.

4.1 Determining the Weights for the Criteria

When using the Entropy method, the weights of criteria are calculated as follows (see
Sect. 2.2): Eq. (12) is used to calculate the normalized values pij . Using Eq. (13), calculate
the Entropy value for each indicator mej . Finally, using Eq. (14), calculate the weight
704 T. Q. Huy et al.

Table 2. Experimental plan and output results

Run order VM Ton Toff SV WF SPD R Ra (µm) CS (mm/min)

1 9 8 18 35 12 2.5 3 3.877 2.08
2 9 8 13 25 8 2.5 3 3.412 2.18
3 9 8 18 25 8 4.5 3 3.410 1.98
4 9 8 13 35 8 4.5 9 2.800 1.81
5 3 8 13 35 12 2.5 9 3.453 1.34
6 3 8 18 35 8 4.5 3 3.962 1.63
7 9 12 13 25 8 4.5 9 3.067 2.50
8 3 8 13 25 8 4.5 9 3.066 1.51
…
31 3 8 13 35 8 2.5 3 3.992 1.83
32 9 8 18 25 12 4.5 9 2.870 1.68

of the criteria wj . The weights of Ra and CS were calculated to be 0.4664 and 0.5336,
respectively.

4.2 Determining the Best Experimental Setup Using MARCOS Method

The steps for multi-objective decision making in the MARCOS method are as described
in Sect. 2.1. In particular, as follows: Using the formula (2), calculate the ideal solution
(AI) and the anti-ideal solution (AAI). Ra and CS were calculated to be 2.8 m and
2.55 mm/min with AI, and 3.96 m and 1.19 mm/min with AAI. The following step is
to compute the normalized values uij using the formulas (3) and (4). The formula (5)
was then used to determine the normalized values while accounting for the weight cij .
− +
Furthermore, Eqs. (6) and (7) yield the coefficients K+i and K+i . Equations (10) and
 (11)
−
were used to calculate the values of f Ki and f Ki . f Ki = 0.4906 and f Ki+ =
0.5094 were discovered. Finally, using formula (9), the values of f (Ki ) are calculated.
Table 3 displays the results of some parameters as well as the ranking of the alternatives.
From Table 3, option 7 is the best alternative. This is due to the fact that it has
the highest utility function value (f (K i ) = 0.1205). As a result, the optimal solution
includes the following parameters: VM = 9 V; Ton = 12 s; Toff = 13 s; SV = 25 V; WF
= 8 mm/min; SPD = 4.5 mm/min; R = 9 mm.

5 Conclusions
In this paper the MARCOS method was applied to optimize the various input factors
of the wire-EDM process when cutting 90CrSi tool steel. From the results of the study,
it is recommended that using alternative 7 can achieve the lowest surface roughness
and highest cutting speed the same time. Experiment No. 7 demonstrated the best per-
formance feature among the 32 trials, with maximum utility function value (f (K i ) =
 Multi-criteria Decision Making in Wire-EDM … 705

Table 3. Several calculated results and ranking of alternatives

Trial K− K+ f (K − ) f (K + ) f (K i ) Rank
1 0.149861 0.144316 0.149861 0.509425 0.0980 17
2 0.162832 0.156807 0.162832 0.509425 0.1065 11
3 0.154664 0.148941 0.154664 0.509425 0.1012 13
4 0.164041 0.157972 0.164041 0.509425 0.1073 9
5 0.127635 0.122912 0.127635 0.509425 0.0835 30
6 0.130178 0.125362 0.130178 0.509425 0.0851 29
7 0.184186 0.177371 0.184186 0.509425 0.1205 1
8 0.144158 0.138824 0.144158 0.509425 0.0943 20
…
31 0.137626 0.132534 0.137626 0.509425 0.0900 25
32 0.156645 0.150849 0.156645 0.509425 0.1024 13

0.1205). The MARCOS technique has identified the best experimental setup for getting
minimum SR and maximum CS is: VM = 9 V; Ton = 12 s; Toff = 13 s; SV = 25 V; WF
= 8 mm/min; SPD = 4.5 mm/min; R = 9 mm.

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. Do, D.T.: Multi-criteria decision making under the MARCOS method and the weighting
methods: applied to milling, grinding and turning processes. Manuf. Rev. 9, (2022)
2. Çalışkan, H., et al.: Material selection for the tool holder working under hard milling con-
ditions using different multi criteria decision making methods. Mater. Des. 45, 473–479
(2013)
3. Trung, D.D., Thinh, H.: A multi-criteria decision-making in turning process using the
MAIRCA, EAMR, MARCOS and TOPSIS methods: a comparative study. Adv. Prod. Eng.
Manag. 16(4), 443–456 (2021)
4. Do, D.T.: A combination method for multi-criteria decision making problem in turning
process. Manuf. Rev. 8, (2021)
5. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
6. Huu Phan, N., Muthuramalingam, T.: Multi criteria decision making of vibration assisted
EDM process parameters on machining silicon steel using Taguchi-DEAR methodology.
Silicon 13(6), 1879–1885 (2021)
7. Nguyen, H.-Q., et al.: Multi-criteria decision making in the PMEDM process by using
MARCOS, TOPSIS, and MAIRCA methods. Appl. Sci. 12(8), 3720 (2022)
8. Bavche, A.L., Valekar, M., Padaseti, B.K.: Application of MOORA to optimize WEDM
process parameters: a multi-criteria decision making approach. In: ICRRM 2019–System
Reliability, Quality Control, Safety, Maintenance and Management: Applications to Civil,
Mechanical and Chemical Engineering, p. 73 (2019)
706 T. Q. Huy et al.

9. Jaiswal, A., et al.: Multi response optimization of wire EDM process parameters. In: IOP
Conference Series: Materials Science and Engineering. IOP Publishing (2018)
10. Muniappan, A., et al.: Optimization of WEDM process parameters on machining of AZ91
magnesium alloy using MOORA method. In: IOP Conference Series: Materials Science and
Engineering. IOP Publishing (2018)
11. Muniappan, A., et al.: Parametric optimization of WEDM control variables on magne-
sium AZ91 alloy by TOPSIS method. In: IOP Conference Series: Materials Science and
Engineering. IOP Publishing (2018)
12. Sreeraj, P., et al.: Application of MCDM based hybrid optimization of WEDM process
parameters. Mater. Today: Proc. 50, 1186–1192 (2022)
13. Bagal, D.K., Mahapatra, S.K., Barua, A., Jeet, S., Pattanaik, A.K., Patnaik, D.: Multi-
parametric optimization of wire-EDM of Inconel 718 super alloy using Taguchi-coupled
WASPAS method. In: Pant, P., Mishra, S.K., Mishra, P.C. (eds.) Advances in Mechanical
Processing and Design. LNME, pp. 459–467. Springer, Singapore (2021). https://doi.org/10.
1007/978-981-15-7779-6_40
14. Patel, J.D., Maniya, K.D.: Optimization of WEDM process parameters for aluminium metal
matrix material Al+ SiC using MCDM methods. In: Advances in Manufacturing Processes,
pp. 59–70. Springer (2021)
15. Stević, Ž, et al.: Sustainable supplier selection in healthcare industries using a new MCDM
method: measurement of alternatives and ranking according to COMPROMISE solution
(MARCOS). Comput. Ind. Eng. 140, 106231 (2020)
16. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Vietnam J. Agric. Sci. 17(4), 32–2331 (2019)
 Multi-objective Optimization in Internal
Grinding SKD11 Tool Steel in Terms of Surface
Roughness, Material Removal Rate and Wheel
Life

Tran Huu Danh1 , Nguyen Hong Linh2 , Bui Thanh Danh3 , Nguyen Anh Tuan4 ,
Do Thi Tam5 , Hoang Xuan Tu5 , and Le Xuan Hung5(B)
1 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 University of Economics—Technology for Industries, Hanoi, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

lexuanhung@tnut.edu.vn

Abstract. This paper presents a multi-objective study on the determination of

optimum dressing parameters for internal grinding. The aim of this study is to
achieve minimum surface roughness (SR), maximum material removal rate (MRR)
and maximum wheel life (WL) concurrently. To accomplish the stated goal, six
input dressing parameters including coarse dressing depth ar , coarse dressing
time nr , fine dressing depth af , fine dressing time nf , non-feeding dressing n0 , and
dressing feed rate Sd were taken into account. The effect of the parameters on
the SR, MRR and WL of the grinding process was evaluated. Furthermore, the
optimal input parameters to achieve the mentioned objectives were proposed. It
was also reported that the proposal optimal dressing parameters were suitable for
use.

Keywords: Dressing · Internal grinding · Multi-objective optimization · Surface

roughness · Material removal rate · Wheel life · SKD11

1 Introduction
Internal grinding is a common type of grinding used for machining holes that require
high precision and low roughness. However, due to limited machining space, the wheel
diameter is frequently small, making lubrication of the grinding process difficult. As a
result, productivity, machining quality, and wheel life in internal grinding are frequently
low. Therefore, many scientists are interested in studying the optimization of the internal
grinding process.
There have been numerous studies on the optimization of the internal grinding pro-
cess to datewhich have been done with different single-objective and multi-objective
functions. Among the single-objective functions chosen are minimum SR [1, 2], max-
imum MRR [3, 4], minimum grinding cost [5–7], minimum manufacturing time [8],

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 707–717, 2023.
https://doi.org/10.1007/978-3-031-22200-9_75
708 T. H. Danh et al.

or stability of the grinding process [9]. There have been also various studies on multi-
objective optimization of internal grinding process [10, 11]. Studies have been con-
ducted with different workpiece materials such as C40E steel [12], 90CrSi steel [5, 10],
SKD11 tool steel [1, 11], etc. The determination of coolant parameters has also received
great attention [13]. Using a multi-criteria decision making technique, the best dressing
parameters for internal grinding were discovered in [14].
This paper presents the findings of a multi-objective optimization study on the dress-
ing process for internal grinding in order to achieve the lowest SR, the highest MRR
and highest WL all at the same time. The effects of the input dressing parameters of
the grinding process were investigated. An optimal set of input dressing parameters for
internal grinding SKD11 tool steel was also proposed.

2 Experimental Work

Fig. 1. Experimental setup

To solve the multi-objective optimization problem, an experiment was carried out. The
Taguchi method and the L16 orthogonal array (44 × 22 ) were employed to design this
experiment. Table 1 shows the input factors and their levels. The experimental setup is
depicted in Fig. 1, with six dressing parameters shown in Table 1. The dressing procedure
was followed exactly the experimental plan as shown in Table 2. After dressing, the
grinding wheel was used for grinding the workpiece samples with the following grinding
mode: the workpiece material of SKD11 tool steel; the wheel speed of 12000 rpm; the
workpiece speed of 150 rpm; the radial wheel speed of 0.0025 mm/stroke; and the axial
 Multi-objective Optimization in Internal Grinding … 709

feed speed of 1 mm/min. Following the completion of the experiments, the SR (in this
case, Ra (μm)) was measured, and the MMR and WL (min) were calculated. Table 2
displays the experimental strategy as well as the output results (RS, MRR and WL).

Table 1. Input dressing factors

No. Input factors Levels

1 2 3 4
1 Coarse dressing depth (mm) 0.025 0.03 – –
2 Coarse dressing time (times) 1 2 3 4
3 Fine dressing depth (mm) 0.005 0.01 0.015 0.02
4 Fine dressing time (times) 0 1 2 3
5 Non-feeding dressing (times) 0 1 2 3
6 Dressing feed rate (mm/min) 1 1.2 – –

Table 2. Experimental plan and output results

No. ar nr af nf n0 Sd Mean of Ra Mean of MRR Mean of Tw

(μm) (mm3 /s) (min)
1 0.025 1 0.005 0 0 1.0 0.3652 0.9186 11.2667
2 0.030 1 0.010 1 1 1.0 0.2137 1.0413 12.7333
3 0.025 1 0.015 2 2 1.2 0.1948 1.1215 12.6967
4 0.030 1 0.020 3 3 1.2 0.2417 1.1111 11.4233
5 0.030 2 0.005 1 2 1.2 0.1850 1.2459 11.8033
6 0.025 2 0.010 0 3 1.2 0.2477 1.2667 12.4300
7 0.030 2 0.015 3 0 1.0 0.2520 1.1782 11.1467
8 0.025 2 0.020 2 1 1.0 0.2167 1.2251 12.0767
9 0.030 3 0.005 2 3 1.0 0.3064 1.4878 12.5467
10 0.025 3 0.010 3 2 1.0 0.3239 1.3724 12.0767
11 0.030 3 0.015 0 1 1.2 0.3406 1.2709 12.0267
12 0.025 3 0.020 1 0 1.2 0.3541 1.1988 12.5033
13 0.025 4 0.005 3 1 1.2 0.3179 1.2273 12.1767
14 0.030 4 0.010 2 0 1.2 0.3126 1.2647 10.6200
15 0.025 4 0.015 1 3 1.0 0.3259 1.1247 13.1533
16 0.030 4 0.020 0 2 1.0 0.3634 1.1898 12.2300
710 T. H. Danh et al.

Multi-objective optimization
The gray relational analysis (GRA) method, in conjunction with the Taguchi method, was
used for multi-objective optimization work. In this study, three objectives were chosen
for investigation: the minimum SR, the maximum MMR, and the maximum WL.
The S/N index of Ra, MRR and Tw is calculated according to the following formulas.
For the minimum SR:
 n 
1 2
S/N = −10 log10 yi (1)
n
i=1

For the maximum MRR and WL:

 
1 1
n
SN = −10log10 (2)
n yi2
i=1

The results of the calculation of the S/N index of the output targets are described in
Table 3.

Table 3. S/N values of output parameters

No. Input parameters Ra (μm) MRR (mm3 /s) Tw (min)

ar nr af nf n0 Sd Mean S/N Mean S/N Mean S/N
1 0.025 1 0.005 0 0 1.0 0.3653 8.7488 0.9186 −0.7372 11.2667 21.0350
2 0.030 1 0.010 1 1 1.0 0.2137 13.4051 1.0413 0.3511 12.7333 22.0984
3 0.025 1 0.015 2 2 1.2 0.1947 14.2084 1.1215 0.9958 12.6967 22.0730
4 0.030 1 0.020 3 3 1.2 0.2417 12.3355 1.1111 0.9143 11.4233 21.1553
5 0.030 2 0.005 1 2 1.2 0.1850 14.6562 1.2459 1.9091 11.8033 21.4400
6 0.025 2 0.010 0 3 1.2 0.2477 12.1225 1.2667 2.0528 12.4300 21.8890
7 0.030 2 0.015 3 0 1.0 0.2520 11.9718 1.1782 1.4244 11.1467 20.9426
8 0.025 2 0.020 2 1 1.0 0.2167 13.2838 1.2251 1.7630 12.0767 21.6384
9 0.030 3 0.005 2 3 1.0 0.3063 10.2729 1.4878 3.4510 12.5467 21.9701
10 0.025 3 0.010 3 2 1.0 0.3240 9.792 1.3724 2.7494 12.0767 21.6384
11 0.030 3 0.015 0 1 1.2 0.3407 9.3562 1.2709 2.0825 12.0267 21.6027
12 0.025 3 0.020 1 0 1.2 0.3540 9.0171 1.1988 1.5744 12.5033 21.9402
13 0.025 4 0.005 3 1 1.2 0.3180 9.9544 1.2273 1.7788 12.1767 21.7099
14 0.030 4 0.010 2 0 1.2 0.3127 10.1013 1.2647 2.0400 10.6200 20.5216
15 0.025 4 0.015 1 3 1.0 0.3260 9.7386 1.1247 1.0204 13.1533 22.3798
16 0.030 4 0.020 0 2 1.0 0.3633 8.7912 1.1898 1.5094 12.2300 21.7483
 Multi-objective Optimization in Internal Grinding … 711

To analyze the gray relationships based on S/N values, it is necessary to convert them
to comparison series with dimensionless quantities or to normalize the data. In this case,
the S/N values are normalized in the range (0 ÷ 1). They are normalized by the value of
Z ij with 0 ≤ Z ij ≤ 1; Z i is determined by the following formula:
 
SNij − min SNij , j = 1, 2, . . . k
Zij =     (3)
max SNij , j = 1, 2, . . . n − min SNij , j = 1, 2, . . . n

where n is the total of test runs, yi is the observed data.

The grey relation coefficient, (γ (k)) which expresses the relationship between
reference and experimental data is found as follows:
min + ζ max
γ (k) = (4)
j (k) + ζ max
 
In which, j (k) is the deviation of the reference data; j (k)  
 = Z0 (k) −Zj (k) and
 
Z 0 (k) is the reference data or best data.  min = min min Z0 (k) − Zj (k) minimum
 ∀j∈i ∀k
value of j (k).  max = max maxZ0 (k) − Zj (k) maximum value of j (k). ζ is the
∀j∈i ∀k
distinguishing or identification coefficient; ζ = 0 ÷ 1 (In this case ζ = 0.5).
The gray relational grade is determined by the following Equation:

1
m
γj = γij (5)
k
i=1

where k is the number of objectives.

3 Results and Analysis

The Taguchi method is used to determine how different dressing parameters affect mean
gray relation values. In order to maximize the average gray relation value, the S/N value
is calculated using formula (2). Table 4 displays the ANOVA-calculated results of the
effect of input parameters on the average gray relation value.
712 T. H. Danh et al.

Table 4. Influence order of dressing factors on yi

Analysis of Variance for Means

Source DF Seq SS Adj SS Adj MS F P
n0 3 0.063995 0.063995 0.021332 2.77 44.09
nf 3 0.052646 0.052646 0.017549 2.28 36.27
af 3 0.006252 0.006252 0.002084 0.27 4.31
nr 3 0.014393 0.014393 0.004798 0.62 9.92
ar 1 0.000142 0.000142 0.000142 0.02 0.10
Sd 1 0.000026 0.000026 0.000026 0.00 0.02
Residual Error 1 0.007687 0.007687 0.007687 5.30
Total 15 0.145140
Model Summary
S R-Sq R-Sq(adj)
0.08 94.70 20.55%
77 %

From Table 4 it is obvious to see that the two parameters with the greatest influence
on yi are n0 (44.09%) and nf (36.27%), followed by the influence of nr (9.92%), af
(4.31%). The effect of ar (0.1%) and Sd (0.02%) on yi is insignificant.
Table 5 describes the order in which the parameters influence the gray relation value.

Table 5. Effect of dressing factors on the gray relation values yi

Response Table for Means

Level n0 nf af nr ar sd
1 −7.06 −6.421 −5.281 −5.634 −5.242 −5.331
2 −5.11 −4.323 −5.172 −4.656 −5.330 −5.241
3 −4.43 −4.467 −4.970 −4.962
4 −4.55 −5.933 −5.721 −5.892
Delta 2.62 2.098 0.750 1.236 0.087 0.091
Rank 1 2 4 3 6 5
Average gray relational value: 0.553

The influence order of the input parameters on yi in this table is nf , n0 , nr , af , ar , and

Sd .
 Multi-objective Optimization in Internal Grinding … 713

Fig. 2. Effect of input factors on yi

Figure 2 shows that as n0 increases from level 1 to level 3, yi increases; when n0

rises further to level 4, yi begins to decrease slightly. Furthermore, as nf rises from level
1 to level 2, yi grows sharply. If nf rises from level 2 to level 3, yi falls slightly. However,
as nf develops from level 3 to level 4, yi drops precipitously. yi decreases slightly as
af increases from level 1 to level 2. If af increases from level 2 to level 3, yi increases;
however, if af increases to level 4, yi decreases. The relationship between nr and yi is as
follows: as nr increases from level 1 to level 2 and from level 3 to level 4, yi increases
sharply. However, as nr increased from 2 to 3 and 4, yi fell precipitously. The influence
of ar and Sd on yi is insignificant.
The effect of the input parameters on the S/N ratio is considered when determining
the best dressing mode. The set of parameters with the highest value for each input factor,
as previously discussed, is the most reasonable. The large value of S/N, the better. This
also implies that the noise effect is reduced. The effect of the input parameters on the
S/N value is depicted in Fig. 3. This figure also identifies the best set of parameters for
the multi-objective function (Table 6).
714 T. H. Danh et al.

Fig. 3. Effect of input factors on S/N ratio

Table 6. Optimum dressing factors

No Factor Level Value

1 Coarse dressing depth (mm) 1 0.025
2 Coarse dressing time (times) 2 2
3 Fine dressing depth (mm) 3 0.015
4 Fine dressing time (times) 2 1
5 Non-feeding dressing (times) 3 2
6 Dressing feed rate (mm/min) 2 1.2

The optimal set of parameters can be used in Minitab 19 to calculate and predict
the results of the objective functions (Table 7). The predictive analysis results with the
optimal set of dressing parameters yielded the surface roughness Ra = 0.1821 μm, MRR
= 1.1691 mm3 /s, and wheel life Tw = 12.8431 min.
 Multi-objective Optimization in Internal Grinding … 715

Table 7. Prediction of values of RS, MRR and WL corresponding to the optimal factors

Settings
n0 nf af nr ar sd
2 1 0.015 2 0.025 1.2
Prediction
Output S/N Ratio Mean stDev Ln(StDev)
Ra 14.6735 0.182146 0.0025014 −6.01621
MRR 1.46616 1.16906 0.0062689 −5.08516
TW 22.1889 12.8431 0.0713222 −2.85528

Experiment model evaluation

The Anderson–Darling method was applied to evaluate the experimental model’s suit-
ability (Fig. 4). Figure 4 indicates that the experimental dots on the histogram (the blue
points) are very close to the normal distribution line (the red solid line). It shows that the
deviation is very small. The error frequency Histogram graph indicates that the majority
of errors are in the −0.15 to 0.1 range. From the other two graphs, the experimental
errors are distributed at random, which implies that the built model is greatly influenced
by the input parameters selected but has no effect on the order of experiments.

Fig. 4. Residual plots for S/N ratios

716 T. H. Danh et al.

The Anderson–Darling technique was used to verify the experimental model’s suit-
ability (Fig. 5). The results on this model indicates that the experimental points (blue
dots) are all within a 95% confidence interval defined by two limit lines. Moreover,
the value of P of 0.509 is greater than 0.05, denoting that the empirical model used is
suitable.

Fig. 5. Probability plot of the fit of proposed model

4 Conclusions
This paper presented the results of a multi-objective optimization study in the dressing
process for internal grinding SKD11 tool steel to achieve the lowest SR, highest MRR,
and highest WL. Six input dressing parameters were chosen for the study. The Taguchi
technique was used in conjunction with GRA to reduce SR, increase MRR, and increase
WL. The following conclusions can be drawn from the findings:

– The effect of the dressing factors on the SR, MRR, and WL of the grinding process
was investigated.
– The most affected parameters on yi are n0 (44.09%) and nf (36.27%), followed by nr
(9.92%), af (4.31%), ar (0.1%) and Sd (0.27%).
– The following input factors were proposed for the best dressing process when internal
grinding to achieve minimum SR, maximum MRR, and minimum WL simultaneously:
ar = 0.03 mm/l; nr = 2 (times); ar = 0.025 mm; af = 0.015 mm; nf = 1 (times); n0
= 2 (times); and Sd = 1.2 m/min.

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

 Multi-objective Optimization in Internal Grinding … 717

References
1. Hung, L.X., et al.: Determining the optimum set of dressing parameters satisfying minimum
surface roughness when conducting the internal grinding of hardened SKD11 steel. In: Solid
State Phenomena. Trans Tech Publications (2021)
2. Zhang, K., Yu, Y., Wang, W., Li, H.: Experimental study on internal cylindrical grinding
of bearing steel with electrolytic in-process dressing grinding. Int. J. Adv. Manuf. Technol.
81(5), 1175–1185 (2015). https://doi.org/10.1007/s00170-015-7297-6
3. Hung, L.X., et al.: Optimum dressing parameters for maximum material removal rate when
internal cylindrical grinding using Taguchi method. Int. J. Mech. Eng. Technol. (IJMET)
9(12), 123–129 (2018)
4. Jeevanantham, S., et al.: A study on characteristics of parameters influencing internal grinding
process with MRR. Indian J. Sci. Technol. 9(37), 1–7 (2016)
5. Hung, L.X., et al.: Determination of optimal exchanged grinding wheel diameter when inter-
nally grinding alloy tool steel 9CrSi. In: IOP Conference Series: Materials Science and
Engineering. IOP Publishing (2018)
6. Tran, T.-H., et al.: Optimization of replaced grinding wheel diameter for minimum grinding
cost in internal grinding. Appl. Sci. 9(7), 1363 (2019)
7. Hung, L.X., et al.: A study on calculation of optimum exchanged grinding wheel diameter
when internal grinding. Mater. Today: Proc. 18, 2840–2847 (2019)
8. Hung, L.X., et al.: Optimization of manufacturing time in internal grinding. In: Sattler, K.-
U., Nguyen, D.C., Vu, N.P., Tien Long, B., Puta, H. (eds.) ICERA 2019. LNNS, vol. 104,
pp. 557–565. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-37497-6_63
9. Akintseva, A., Pereverzev, P.: Modeling the influence of the circle overtravel on the stability
of internal grinding process. Procedia Eng. 206, 1179–1183 (2017)
10. Hung, L.X., et al.: Multi-objective optimization of dressing parameters of internal cylindrical
grinding for 9CrSi alloy steel using Taguchi method and grey relational analysis. Mater.
Today: Proc. 18, 2257–2264 (2019)
11. Nguyen, A.T., et al.: Optimization of dressing parameters for minimum surface roughness
and maximum material removal rate in internal grinding of SKD11 tool steel. In: E3S Web
of Conferences. EDP Sciences (2021)
12. Jeevanantham, S., et al.: Optimization of internal grinding process parameters on C40E steel
using Taguchi technique. Int. J. Appl. Eng. Res. 12(19), 8660–8664 (2017)
13. Hung, L.X., et al.: A study on coolant parameters in internal grinding of 9CrSi steel. In:
Materials Science Forum. Trans Tech Publications (2019)
14. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
 Novel Energy Efficient Schemes for Wireless
Sensor Networks Utilizing Mobile Sensors

Minh T. Nguyen1(B) , Cuong V. Nguyen2 , Quynh C. Le3 , Binh D. Ha4 ,

Hoang T. Tran4 , Vuong Q. Dang5 , Mien L. Trinh6 , and Fabio Viola7
1 Thai Nguyen University of Technology, Thai Nguyen City 240000, Vietnam
nguyentuanminh@tnut.edu.vn
2 Thai Nguyen University of Information and Communication Technology, Thai
Nguyen 240000, Vietnam
3 Electric Power University, Ha Noi 100000, Vietnam
4 Duy Tan University, Da Nang 550000, Vietnam
5 Ha Noi University of Science and Technology, Ha Noi 100000, Vietnam
6 University of Transport and Communications, Ha Noi 100000, Vietnam
7 University of Palermo, Viale Delle Scienze, Parco d’Orleans, 90128 Palermo, PA, Italy

Abstract. Wireless sensor networks (WSNs) facilitate many applications in dif-

ferent fields and are well-known in both academic and industrial areas. The net-
works mainly include sensors, networking parts and a data processing center to
collect data for a variety of purposes. However, the energy consumption problem
in the networks is always a critical problem that many researchers are looking
for good solutions. In this paper, we propose a data collection method that uses
mobile sensors attached with bigger energy storages to collect data from static
sensors. In addition, we propose a hybrid energy harvesting scheme to harvest
energy from ambient environments to support static sensors to prolong the net-
work lifetime. Some algorithms are provided to support the networks either with
data collection or harvesting energy. We outline various possibilities for power-
ing the gadgets with solar and RF energy gathering. In particular, all the circuits
that are constructed using mathematical equations are offered. All of the network
possibilities are clarified using simulation and experimental results. The outcomes
are encouraging and useful.

Keywords: Wireless sensor networks · Mobile sensors · Data collection ·

Energy harvesting · Radio frequency (RF) · Solar energy · Energy efficiency

1 Introduction
1.1 Motivation
Traditionally, wireless sensor networks (WSNs) include a lot of sensors for measuring
sensing data from ambient environments. The sensors are small, inexpensive and have
limited capacity of communication and energy storage [1, 2]. They collect sensory data
to be transferred to a base station (BS) or a data center. They operate in the networks

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 718–732, 2023.
https://doi.org/10.1007/978-3-031-22200-9_76
 Novel Energy Efficient Schemes for Wireless Sensor 719

until they deplete all the energy storage in their pre-charge batteries. The networks may
stop working if some sensors disconnected to each other. Hence, energy saving in such
networks is always a critical issue that needs to improve [3, 4].
There are many energy efficient mechanisms to improve the network lifetime based
on different points. Many data collection methods are proposed to save energy consump-
tion for the networks [5–8]. In clustered WSNs, cluster heads (CHs) can help non-CH
sensors saving energy from sending directly to the BS. In addition, the role of being
CHs can be changed to balance energy storage for sensors [5]. In tree-based WSNs,
sensors are connected as tree with the root as the BS. The multi-hop routing commu-
nication through the tree can significantly save energy due to the short communication
range between sensors [6]. Random walk routing [7] and Gossiping [4, 8] sensing data
in WSNs also support energy saving and balancing problems in the networks that can
prolong the network lifetime.
Data processing techniques are also useful to support energy saving in WSNs. Sens-
ing data in almost applications are usually high correlated or compressible that can be
sampled or compressed to be a certain number of measurements, instead of being as raw
data. There are many techniques that can support to process data in the networks. A lot
of results using frequency domains such as Wavelet or discrete cosine transform (DCT)
to compress data [9, 10]. Compressed sensing (CS) techniques are also combined to
routing methods to reduce data transmission in WSNs [11, 12]. The smaller the number
of the sending measurements, the lower the energy consumption from sensors or the
networks. Hence, the network lifetime can be improved significantly.
However, all of the efforts also have some limited capacity to solve the energy saving
problems. The networks need to be more active to operate long and to be more suitable
to many applications while the demand for the network services is increasing. Currently,
energy harvesting techniques are proposed to support the network directly. The sensors
can harvest energy from their surrounding environments, since there are energy resources
available right at the sensing areas. This could be a promising point for now and the future
development of WSNs.

1.2 Related Work

Traditional WSNs collect data based on different topologies [4–8]. However, static sen-
sors with very limited capacities cannot deal with long distance communication and also
long-time operation demand. Hence, mobile sensors are often deployed in a sensing area
to either collect data or support the static sensors. Normally, static sensors measure data
from the fields and send the data to the mobile ones. Finally, the data are sent to the BS
in long communication ranges by the mobile sensors [13, 14]. In different scenarios, to
save energy, mobile sensors do not move to all positions in a sensing field, the only move
in some certain regions and share their collected data to each other, then finally send the
whole data to the BS [15–17]. Based on the previous work, the mobile sensor can save
significant energy for static sensors, and definitely prolong the network lifetime.
Energy harvesting methods have been widely used in a variety of systems depending
on the available natural resources, such as wind, solar, RF, thermal, flow (wind, hydro),
mechanical (vibration, pressure), human (activity, physiological), etc. [18–20]. These
resources can be categorized as predictable and controlled or not. The most effective
720 M. T. Nguyen et al.

ways to provide systems that require more energy or function depending on the harvested
energy may be found by selecting appropriate energy harvesting technologies based on
the characteristics of the various resources.
Since the sensors in WSNs are normally very small and they do not spend much
energy, it seems quite possible to harvest energy from ambient environments. Depend-
ing on specific sensing fields, the sensors can harvest energy from different resources,
such as RF, solar, thermal or wind. In papers [16, 17], the authors provide possible
energy consumptions for sensors in different status. Based on the levels, we can design
energy harvesting models to support the sensors effectively. In papers [18–21], differ-
ent resources are exploited separately that provide significant energy for the sensors.
Each resource has either advantages or disadvantages that depend on specific condi-
tions. Hence, choosing suitable resources to harvest energy is important to support static
sensors.
Based on the ideas of previous work, including the motivation, we propose a
hybrid energy harvesting mechanism that can combine energy harvested from differ-
ent resources. In our work, solar and RF energy are collected and then combined to
charge sensor batteries. The RF energy may not provide much energy but quite stable
while the solar resource is only available in day time but unlimited. We either propose
data collection algorithms or energy harvesting algorithms for the networks. New system
models are provided in our work. Circuits are designed and associated with simulation
results to clarify the models.
The rest of the paper is organized as follows. Section 2 provides system models
and problem formulation. Section 3 addresses details of our algorithms with the energy
harvesting methods, including circuit designs. Some simulation results are provided
in Sect. 4 to clarify our proposed problems. Finally, conclusions and future work are
provided in Sect. 5.

2 System Model and Problem Formulation

2.1 System Model

We assume that many static sensors are randomly deployed in a sensing area that needs
to be observed. The sensors can communicate wirelessly to each other within their
communication ranges, denoted as Rc . They measure sensing data to be sent to a data
processing center or a base-station (BS). In order to support the static sensors, a certain
number of mobile sensors are also randomly deployed in the sensing field. They have
higher profiles compared to the static sensors to deal with longer communication ranges.
In addition, they are expected to support the static sensors with either data collection or
energy supply. The network is illustrated in Fig. 1.
The mobile sensors are designed with circuits that can harvest energy from solar and
RF from the sensing field. All the energy collected is combined to charge the batteries
attached to them. The mobile sensors later can either collect data from sensor nodes or
transmit energy to static sensor nodes. Based on this model, the mobile sensors can save
energy for static sensors and share their harvested energy to supply the sensors.
 Novel Energy Efficient Schemes for Wireless Sensor 721

Static sensors are also designed to harvest energy from the ambient environments.
However, their profile is small and inexpensive. Hence, it would be better if they can
receive directly energy from mobile sensors in short communication distances.

Fig. 1. A certain number of static sensors (in blue) and mobile sensors (in red) are deployed
randomly in a sensing area for monitoring purposes.

2.2 Energy Model

Almost energy consumption of sensors focuses on transmitting, receiving and processing
data, as shown in Table 1. Hence, different operating modes are designed appropriately
to save energy for the sensors. Normally, K-bit packets are considered and associated
with the transmitting distances (d) in such network to calculate energy consumption for
static sensors as follows.

Eelec K + fs Kd 2 , d < do
ETX (K, d )static = , (1)
Eelec K + amp Kd 4 , d ≥ do
ERX (K)static = Eelec K (2)

where Eelec is the energy necessary for sending 1-bit and ERX presents the energy usage
to receive data. fs and amp are the amplified power in the radio model. d0 is the threshold
obtained using Eq. (3) as follows.

fs
d0 = (3)
amp
722 M. T. Nguyen et al.

The energy consumption models for the mobile sensors include receiving data from
static sensors and transmitting data to the BS that are also calculated as follows.

ERX (K)mobile = Eelec K (4)


Eelec K + fs Kd 2 , d < do
ETX (K, d )mobile = (5)
Eelec K + amp Kd 4 , d ≥ do

In addition, when the mobile sensors exchange data with each other, the energy
model can be calculated as follows.

ERX (K)mobile = Eelec KM (6)

where, M is the convergence time that the mobile sensors exchange completely their
data with each other.

Table 1. Energy consumption in a MICA2 sensor nodes with a CC1000 transceiver in different
modes that can be flexibly scheduled.

Mode Time Energy consumption

Sleep …… 90 µW
Initiate radio 0.35 ms 18 mW
Turn radio 1.50 ms 3 mW
Switch Tx/Rx 0.25 ms 45 µW
Receive 1 byte 0.4 ms 45 µW
Transmit 1 byte 0.4 ms 606060W

2.3 Data Collection Method

In this paper, we propose an algorithm that can support either static sensors or mobile
sensors to collect data in WSNs with energy saving manners. With the mobile collabo-
rations, each mobile sensor can be considered as a distributed one. One mobile sensor
can have all the data from all static sensors if it has enough time for exchanging data.
This point can support the network in case of some mobile sensors become failure, the
others can transmit data to the BS with all data from the network. The data collection
algorithm is addressed as follows.
Algorithm 1: Hybrid data collection for mobile sensors to support WSNs
Phase 1: Initialization

1. N sensors are randomly deployed in a sensing area;

2. M mobile sensors are also randomly deployed in the sensing area;
 Novel Energy Efficient Schemes for Wireless Sensor 723

3. All the sensors can communicate wirelessly within a communication range Rc ;

Phase 2: Data communications between the sensors

4. M mobile sensors move to the sensing area among static sensors;

5. N static sensors transmit their sensing data to mobile sensors within their Rc ;

Phase 3: Exchange data between mobile sensors

6. M mobile sensors may have different number of collected measurements from static
sensors;
7. M mobile sensors share the collected data to each other within their Rc ;

Phase 4: Sending data to the Base-station (BS)

8. After a convergence time (T), each mobile sensor can have all sensing data from the
network;
9. M mobile sensors finally forward the whole data to the BS.

2.4 Energy Harvesting Scheme

In this paper, we propose that either static sensors or mobile sensors can harvest energy
from ambient environments. However, the static sensors are small that cannot guarantee
to collect sufficient energy to supply themselves. Hence, the mobile sensors with higher
profiles can harvest and transmit energy wirelessly to support the static sensors.
Table 2 provides possible energy that can be harvested from ambient environments.
Based on the table, we choose to combine energy from solar and RF into mobile sensors.

Table 2. Possible energy harvesting from different resources.

Energy sources Conditions Performance

Solar Outdoor 7500 µW/cm2
Solar Indoor 100 µW/cm2
RF Wi-Fi 0.001 µW/cm2
RF GSM 0.1 µW/cm2
Vibration 1 m/s2 100 µW/cm3
Thermal T = 5 0 C 60 µW/cm2

We propose a scheme to harvest energy for either static sensors or mobile sensors as
shown in Fig. 2. The steps for designing the harvesting system are addressed as follows.
724 M. T. Nguyen et al.

Algorithm 2: Hybrid energy harvesting Solar and RF resources to power WSNs

Phase 1: Initialization

1. N sensors are designed to be able to harvest Solar and RF;

2. M mobile sensors are designed to be able to harvest Solar and RF;
3. Assume that the all conditions that support the process without loss or disturbance.

Phase 2: Solar Energy Harvesting

4. M mobile sensors can harvest energy from solar separately;

5. N static sensors can harvest energy from solar separately;

Phase 3: RF Energy Harvesting

6. N static sensors can harvest energy from RF separately;

7. M mobile sensors can harvest energy from RF separately;

Phase 4: Combine RF and Solar energy to power WSNs

8. Design circuits to combine both the energy resources;

9. Test and compare in different conditions;
10. Evaluate, stabilize the hybrid system.

3 Circuit Design for the Harvesting Systems

The Fig. 3 shows the detailed diagram of the proposal. For each energy source, we will
have a separate method and technique to harvest, the ultimate goal is to create a DC power
source to put into the charger to charge the battery. In the Fig. 3, the solar energy source
uses the MPPT algorithm and the boost converter to raise the voltage to the required
level of 18.5V. The RF energy source is harvested from the surrounding environment at
the 900MHz frequency range, then through the converter circuits to obtain a DC voltage
of 5V corresponding to 5 stages connected in series. The detailed design of the system
components is described in the following sequence.

3.1 RF Energy Harvesting

Impendence matching: In this paper, we use inductor (L) and capacitor (C) elements
to perform impedance matching.
We assume the design requirement of an impedance matching circuit with source
impedance ZS = 50 , load impedance ZL = 60 − 45j (), line impedance Z0 = 50 ,
operating at 900MHz. Parameters L and C are calculated based on the Smith graph.
We choose S-Parameters with a hop of 0.1 GHz, starting frequency of 0.1 GHz, and
ending frequency of 2 GHz as shown in Fig. 4.
 Novel Energy Efficient Schemes for Wireless Sensor 725

Fig. 2. A hybrid harvesting design to supply static sensor nodes and mobile sensors.

Solar energy harvesting

DC
DC

V
Duty Cycle

MPPT controller
Charging system
RF energy harvesting
Impendence Voltage
Transmit Receive matching multiplier
antenna Battery
antenna
RF input

Fig. 3. Proposed energy harvesting system

Voltage multiplier: In Fig. 5, the RF energy through the impedance matching circuit,
this energy will go to the next circuit called rectifiers, to convert from AC source to DC
source. The output voltage depends on the number of stages in the circuit. In this paper,
we use 5 stages connected in series, and the output voltages are different from each state.
726 M. T. Nguyen et al.

C = 3.78 pF
L= 9.86 nH
Source impedance
Z= 50 Ω
S-Parameters Load impedance
Start = 0.1 GHz Z= 60-j*45 Ω
Stop = 2.0 GHz

Fig. 4. Impedance matching circuit design

Impendence
Matching Voltage
multiplier

Receive
antenna
V_RF

Fig. 5. The full multistage voltage multiplier circuit

3.2 Solar Energy Harvesting

The process of building and simulating the MPPT algorithm is shown in Fig. 6. In Fig. 7a,
the relationship between output current and voltage of the solar harvesting system using
the MPPT algorithm is presented. The graph shows that thanks to the MPPT algorithm,
the current only fluctuates in the range from 0.25 A to 0.3 A and the output voltage
always fluctuates in the range from 18 V to 20 V.
In Fig. 7b shows the relationship between voltage and maximum output power of
the solar harvesting system, the power is in the range from 4.6 W to 5 W. However, the
disadvantage of the solar energy system is that it is highly dependent on the surrounding
environmental conditions. Therefore, we need to combine with another energy source
that is less dependent on environmental conditions, which is RF energy.

3.3 Charging System

We designed the detailed charging circuit diagram as shown in Fig. 8, one of the important
elements of this charging circuit is the LM317. LM317 is a voltage regulator IC. Pin 1
(ADJ) of the IC is the control pin used to control the charging voltage. Pin 2 (VO) is
the output pin where the charging voltage appears. Pin 3 (VI) is the input pin to which
the regulated DC supply is supplied. The output voltage can be adjusted to nearly 40 V.
The voltage regulation of the LM317 is in the range from 1.25 V to 37 V, the maximum
output current is 1.5 A, and the internal resistance is as small as 0.05 .
 Novel Energy Efficient Schemes for Wireless Sensor 727

Continuous
I-V

x P-V

Vin1
Vin
VC

PV
MPPT

C1 C R

25

Fig. 6. Harvest solar energy using the MPPT algorithm

I(A) P (W)
Maximum Power Point
0.8
5
0.7

0.6 4
0.5
3
0.4

0.3 2
0.2
1
0.1
V(v) V(v)
0 5 10 15 20 0 5 10 15 20

(a) (b)
Fig. 7. The output of solar harvester circuit using MPPT algorithm.

+
DC
Source
-
Battery

Fig. 8. Proposed charging system

Because the maximum output current of the LM317 is only 1.5 A. so we need to
use transistor Q2 to amplify the current, we have a higher charging current to help the
728 M. T. Nguyen et al.

battery charge process faster. The regulation of charging voltage and charging current is
dependent on rheostat and resistor R2 .

4 Simulation Results

In this section, we work on different schemes with energy efficient manners for WSNs
including data collection and energy harvesting methods for mobile sensors.

4.1 Data Collection Methods

As shown in Fig. 9, there are 20 mobile sensors are randomly deployed in a square sensing
area to collect data from 500 static sensors. The sensors have a limited communication
range as Rc that also can be changed to be able to achieve different results. In our
simulations, for simplicity, we assume there are no obstacles, no loss. Hence, the power
consumption is only based on the distances between sensors.

Fig. 9. 20 mobile sensors with communication range Rc = 50 [unit] are deployed to support 500
static sensors in a square sensing area of 100*100 [unit].

As mentioned in Algorithm 1, Fig. 10 shows the convergence time for exchanging

data between the mobile sensors. As shown in the figure, as the communication range
Rc increased, the convergence time is reduced. It means that increasing the connections
between mobile sensors can make the exchange data processes go faster. Regarding the
power consumption for the networks, Fig. 11 provides the total power consumption for
the network to collect data.
 Novel Energy Efficient Schemes for Wireless Sensor 729

Fig. 10. The convergence time for mobile sensors exchanging data completely with different
communication ranges in a square sensing area of 100*100 [unit].

Fig. 11. The total power consumption (energy consumption) for data transmission in a square
sensing area of 100*100 [unit].

4.2 Charging System for Mobile Sensors

Unlike sensor nodes, our mobile sensors use an energy harvesting system that combines
solar energy and RF energy. Because the mobile sensors require quite high charging
current and charging voltage, so each RF power source could not be enough to supply.
Solar energy is highly dependent on the surrounding electrical conditions, so we need to
730 M. T. Nguyen et al.

hybridize these two sources together to charge the mobile sensor battery. Therefore, the
DC power source on the diagram in Fig. 12 will now be replaced by the output voltage
of the energy harvesting system hybrid RF—Solar energy as shown in Fig. 12.

RF-energy

Solar-energy Battery

Fig. 12. Charging system for mobile sensors

The voltage after passing through the output of the charging system will be 20 V and
the current is 6.8 A.

Charging for Battery of Mobile Robots

100
SOC (%)

50

0 1000 2000 3000 4000 5000 6000 7000

Charging Voltage (V)

20

19

0 1000 2000 3000 4000 5000 6000 7000

Charging Current (A)

6 CC Charge CV Charge
4

0 1000 2000 3000 4000 5000 6000 7000

Time (s)

Fig. 13. The battery charging process of mobile sensors

The charging process of devices in the network is shown in Fig. 13. With the mobile
sensor, we use a battery with a capacity of 7.66 Ah, the charging current is 6.8 A, the
charging voltage is 20 V, and starts to recharge when the SoC of the battery drops to 30%.
The charging process is shown in Fig. 13. The battery charging process goes through
two stages, phase one from time t = 0 s to time t = 2600 s, the battery is charged by
 Novel Energy Efficient Schemes for Wireless Sensor 731

a current source (CC charge). When SoC = 80%, the battery goes to the second stage.
That is, from the time interval t = 2600 s to 3000 s as shown in the Fig. 13. Charging
phase by a voltage source (CV charge) for the SoC of the battery to reach 100%.

5 Conclusion and Future Work

In this paper, we propose an energy efficient framework for mobile sensors to support
WSNs to deal with energy saving problems. Data collection methods are proposed for
both the mobile sensors and static sensors. Simulation results are provided to clarify
our data collection algorithm. Regarding the energy harvesting problems, we also have
proposed an efficient energy collection system for mobile sensors and static sensor nodes.
We have detailed design issues with devices with different power consumption to use the
power accordingly. Our proposed system can generate a fairly high charging current I =
6.8 A. Therefore, it is possible to reduce the charging time for the battery significantly.
When the SoC (State of Charge) parameter of the battery drops to 30%, the system will
activate the charging process for the Battery. Therefore, the system work can prolong
the working time for the devices in the network.
In our future work, the problems should be presented with real devices in both data
collection and energy harvesting. Optimization problems should be defined and solved
to support the network to be able to provide the best performance.

Acknowledgements. The author would like to thank Thai Nguyen University of Technology
(TNUT), Viet Nam for the support.

References
1. Quang, V.T., Miyoshi, T.: Adaptive routing protocol with energy efficiency and event
clustering for wireless sensor networks. IEICE Trans. Commun. 91(9), 2795–2805 (2008)
2. Nguyen, M.T., Teague, K.A.: Neighborhood based data collection in wireless sensor networks
employing compressive sensing. In: 2014 International Conference on Advanced Technolo-
gies for Communications (ATC 2014), pp. 198–203 (2014). https://doi.org/10.1109/ATC.
2014.7043383
3. Nguyen, M.T., Teague, K.A., Rahnavard, N.: CCS: energy-efficient data collection in clustered
wireless sensor networks utilizing block-wise compressive sensing. Comput. Netw. 106, 171–
185 (2016). https://doi.org/10.1016/j.comnet.2016.06.029
4. Tran-Quang, V., Miyoshi, T.: A novel gossip-based sensing coverage algorithm for dense
wireless sensor networks. Comput. Netw. 53(13), 2275–2287 (2009)
5. Li, Y., Liang, Y.: Compressed sensing in multi-hop large-scale wireless sensor networks based
on routing topology tomography. IEEE Access 6, 27637–27650 (2018)
6. Nguyen, M.T.: Minimizing energy consumption in random walk routing for wireless sensor
networks utilizing compressed sensing. In: 2013 8th International Conference on System of
Systems Engineering, pp. 297–301 (2013). https://doi.org/10.1109/SYSoSE.2013.6575283
7. Nguyen, M.T., Teague, K.A.: Tree-based energy-efficient data gathering in wireless sensor
networks deploying compressive sensing. In: 2014 23rd Wireless and Optical Communication
Conference (WOCC). IEEE (2014)
732 M. T. Nguyen et al.

8. Altoaimy, L., Kurdi, H., Alromih, A., Alomari, A., Alrogi, E., Ahmed, S.H.: Enhanced
distance-based gossip protocols for wireless sensor networks. In: 2019 16th IEEE Annual
Consumer Communications & Networking Conference (CCNC), pp. 1–4 (2019). https://doi.
org/10.1109/CCNC.2019.8651701
9. Nguyen, M.T., Teague, K.A.: Distributed DCT based data compression in clustered wire-
less sensor networks. In: 2015 11th International Conference on the Design of Reliable
Communication Networks (DRCN). IEEE (2015)
10. Natsume, K., Ohta, M., Adachi, K., Takyu, O., Fujii, T.: DCT compression scheme for wireless
sensor information considering spatial correlation. Int. Conf. Inf. Netw. (ICOIN) 2021, 192–
196 (2021). https://doi.org/10.1109/ICOIN50884.2021.9333961
11. Nguyen, M.T., Teague, K.A.: Compressive sensing based random walk routing in wireless
sensor networks. Ad Hoc Netw. 54, 99–110 (2017)
12. Nguyen, M.T., Teague, K.A.: Compressive sensing based energy-efficient random routing in
wireless sensor networks. In: 2014 International Conference on Advanced Technologies for
Communications (ATC 2014), pp. 187–192 (2014). https://doi.org/10.1109/ATC.2014.704
3381
13. Nguyen, M.T., Teague, K.A.: Compressive and cooperative sensing in distributed mobile
sensor networks. In: MILCOM 2015—2015 IEEE Military Communications Conference,
pp. 1033–1038 (2015). https://doi.org/10.1109/MILCOM.2015.7357581
14. Nguyen, M.T., Boveiri, H.R.: Energy-efficient sensing in robotic networks. Measurement,
vol. 158 (2020)
15. Nguyen, M.T., Teague, K.A.: Random sampling in collaborative and distributed mobile sensor
networks utilizing compressive sensing for scalar field mapping. In: 2015 10th System of
Systems Engineering Conference (SoSE), pp. 1–6 (2015). https://doi.org/10.1109/SYSOSE.
2015.7151962
16. Nguyen, M.T., La, H.M., Teague, K.A.: Collaborative and compressed mobile sensing for data
collection in distributed robotic networks. IEEE Trans. Control Netw. Syst. 5(4), 1729–1740
(2018). https://doi.org/10.1109/TCNS.2017.2754364
17. Nguyen, M.T., Teague, K.A., Bui, S.: Compressive wireless mobile sensing for data collection
in sensor networks. Int. Conf. Adv. Technol. Commun. (ATC) 2016, 437–441 (2016). https://
doi.org/10.1109/ATC.2016.7764822
18. Ren, Q., Yao, G.: Enhancing harvested energy utilization for energy harvesting wireless sensor
networks by an improved uneven clustering protocol. IEEE Access 9, 119279–119288 (2021).
https://doi.org/10.1109/ACCESS.2021.3108469
19. Tran, H.T., Nguyen, M.T., Nguyen, C.V., Ala, G., Viola, F., Colak, I.: Hybrid solar-RF energy
harvesting mechanisms for remote sensing devices. Int. J. Renew. Energy Res. IJRER 12(1)
(2022)
20. Jadhav, S.B., Lambor, S.M.: Hybrid solar and radio frequency (RF) energy harvesting. In: 2017
IEEE International Conference on Power, Control, Signals and Instrumentation Engineering
(ICPCSI), pp. 1975–1980 (2017)
21. Nguyen, M.T., Tran, H., Nguyen, C.V., Ala, G., Viola, F., Colak, I.: A novel framework of
hybrid harvesting mechanisms for remote sensing devices. In: 2022 IEEE 21st Mediterranean
Electrotechnical Conference (MELECON 2022), (Palermo, Italy) (2022)
 Numerical Analysis of Capacity Augmentation
of Finned-Tube Type Waste Heat Recovery
Systems

Amit Arora1(B) and P. M. V. Subbarao2

1 Malaviya National Institute of Technology, Jaipur 302017, India
amit.mech@mnit.ac.in
2 Indian Institute of Technology, Delhi, New Delhi 110016, India

Abstract. Compact heat recovery systems require effective heat exchange pro-
cess. Placing winglets on the gas side of finned-tube type heat recovery systems
promotes bulk flow mixing. In this study, delta-shaped winglets are placed in the
tube aft region. Under fixed operating conditions, winglets’ geometry is varied to
assess the change in performance. The study considers winglets with four distinct
aspect ratios (AR = 0.5, 0.75, 1.0, 1.5). For the thermal features, local Nusselt num-
bers and fin temperature profiles are compared; and for the flow characteristics,
velocity profiles, pathlines, and velocity vectors are examined. Finally, average
Nusselt number and pressure drop are studied to measure the thermo-hydraulic
performance. Despite drops in local Nusselt numbers at some spots on the fin, the
average value of relative Nusselt number is always larger than unity, and the degree
of augmentation grows with the reducing aspect ratio, while the tube wake region
shrinks. The wake-affected fin is significantly benefitted as the relative Nusselt
number equals 134.9% at the studied Reynolds number.

Keywords: Waste heat recovery · Finned-tube array · Vortex generator · Heat

transfer augmentation · Nusselt number · Friction factor

1 Introduction
Finned-tube arrays are the thermal systems that involve heat exchange between a liquid
and a gas [1]; and they are preferred for waste heat recovery applications. The total
thermal resistance in plain finned-tube arrays is dominated by the gas side [2]. Generation
of longitudinal vortices on the fin side, by minor design modifications, bring about such
boundary layer modifications which produce significant decline in the said resistance [2].
Accordingly, their heat exchange capacity can be improved substantially. The aforesaid
design modification involves incorporating vortex generators suitably around the tubes.
The generated vortices facilitate enhanced vorticity, which augments the temperature
gradients [3].
Among different types of generators, the delta winglets are the preferred ones due to
superior performance efficacy [4]. Although winglet height is limited by the fin space,
its length can be varied over a wide range. Aspect ratio is a dimensionless parameter

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 733–738, 2023.
https://doi.org/10.1007/978-3-031-22200-9_77
734 A. Arora and P. M. V. Subbarao

that is used to describe the winglet geometry [5]. While existence of tube wakes is a
common phenomenon in plain finned tube arrays, placement of delta winglets with toe-
out arrangement can help in suppressing those wakes by redistributing the approaching
flow [6]. If the tube wakes are left unattended, they have an adverse effect on the heat
transfer coefficients.
An experimental study reports that the heat exchange capacity of finned tube arrays
improves with generators’ count [7]. However, the augmentation comes with a greater
rise in pressure drop. Furthermore, placement of generators should not be random, as
their position has a significant impact on the desired changes [8]. When generators hide
behind the tubes, their ability to produce thermal augmentation is virtually lost. The
attack angle of the generators is another principal design parameter. In a computational
study, the effect of generators attack angle was investigated between 10-degree and 20°
[6]. It is reported that both Nusselt number and pressure drop increase with the said angle,
where the latter undergoes greater increase. Another computational study compared the
potential of interrupted winglets with the baseline solid configuration [9]. It is reported
that the former design offers higher heat transfer augmentation with a relatively lesser
pressure loss.
For optimum performance augmentation, mindful deployment of the vortex gener-
ators around the tubes is highly desirable. Moreover, the generators’ location(s) that
facilitate extreme performances are likely to be disregarded due to manufacturing and/
or design constraints. But since there is a strong correlation between the geometry of the
generators and the resulting flow modifications, it is possible to compensate the loss in
performance by varying the geometric design, which underlines the novelty of current
study. Accordingly, a 3D computational study is conducted to investigate the outcome
of geometric changes of generators on the thermo-hydraulic augmentations.

2 Numerical Simulation
Previous article [1] gives details regarding the numerical methods, governing equa-
tions, boundary conditions, mesh selection, model validation, and computational domain.
Therefore, only key details are given here for brevity. The fluid flow and heat transfer are
considered to be steady turbulent, and the influence of turbulence is simulated by RNG
k-ε model [10]. According to the literature, the steady state assumption can predict the
average Nusselt number and friction factor in unsteady periodic flows [11]. The physical
dimensions of the system are; D = 0.038 m, S = 0.012 m, T = 0.001 m, and Ps = Pc = 0.1
m. The winglets are positioned inclined to make them produce the longitudinal vortex
system. In order to check the rise in pressure loss, the said angle equals 15 degree. The
winglets are placed at a fixed location, and they undergo a stepped geometric change.
The details are given in Sect. 3. The winglets’ aspect ratio ‘AR’ equals unity [5], where
AR = 2H/L, ‘H’ is the winglet height and ‘L’ is its length. Also, inlet and exit flow
lengths of 10H and 30H are provided to capture entrance and exit effects. Only a sym-
metric section is considered because of flow symmetry in Z-direction [11], as illustrated
in Fig. 1a. By modeling a periodicity in Y-direction, the calculation overhead is further
decreased.
While using enhanced wall treatment, the y-plus value at the wall’s nearby nodes is
≤1 [9]; and graded mesh is used to resolve the wall effects. The solution-independent
 Numerical Analysis of Capacity Augmentation of Finned-Tube Systems 735

Fig. 1. Fin-tube region.

Fig. 2. Model validation. a Excess temperature. b Friction factor.

mesh density was obtained using an iterative test. Over 4.2M 4.3M cells are present in
plain and modified finned-tube array, respectively. The validation of the model is shown
in Fig. 2. The convergence was concluded when the normalized residuals of the energy
equation and rest of the equations were ≤10–8 and ≤10–5 respectively. The friction factor
is defined as fapp = 2p.Amin / ρ.Ao Uin 2 , and the Reynolds number ranges from 1415
to 7075. The inlet flow velocity varies in the range 1–5 m/s, and the hydraulic diameter
is treated as the characteristic length [12].

3 Results and Discussions

For delta winglets making 15-degree attack angle, multiple locations are recommended
for their deployment in finned tube arrays [1]. But only a few are effective in curtailing
the tube wakes. In this study, the winglets are placed at a selected location, which is
illustrated in Fig. 3a. The winglet’s geometry is defined based on the aspect ratio ‘AR’,
and for a detailed analysis, four distinct aspect ratios are considered (i.e. AR = 1.5, 1.0,
0.75, 0.5). A flow visualisation study is carried out, before accounting the changes in
thermal characteristics, to gain insights about local flow modifications.
736 A. Arora and P. M. V. Subbarao

Fig. 3. a Fin-and tube with delta winglets at X = 0.3D, Z = 0.7D. b Various Aspect ratios.

3.1 Flow Field Visualization

The velocity vectors corresponding to the winglets bearing 0.5 aspect ratio are presented
in Fig. 4a, b. Those vectors are drawn in three projection planes in the winglet downstream
(i.e. X = 0.5D, 1.0D, and 1.5D, where ‘X’ is measured w.r.t. tube center). There are two
vortex systems, main vortex and corner vortex. Evidently, the main vortex is initially
circular, but it eventually becomes elliptical due to the presence of two sandwiching
fins. Furthermore, the spread of both vortices grows in the streamwise direction, thereby,
imparting enhanced bulk mixing over the affected fin surface on account of secondary
flow structures. The effect of winglet geometry on the flow structures is illustrated in
Fig. 4c. The vortex spread grows with reducing aspect ratio, and its axis shifts, away
from the tubes, in cross-stream direction. The growing span of main vortex is likely to
impart large scale mixing over a substantial fin surface as it prevails over a significant
length in the streamwise direction. The pathlines in baseline and modified finned tube
arrays are presented in Fig. 4d. For brevity, only a selected geometric design (i.e. AR
= 0.5) is considered. It is encouraging to notice that the prescribed winglets are quite
effective in generating three-dimensional longitudinal vortex, and the tube wakes too
are suppressed substantially.

3.2 Flow Field Visualization

The incorporation of winglets is justified, if the average Nusselt number of the finned
tube array undergoes reasonable augmentation. Nonetheless the average value(s) shall
be contributed by various local augmentations. The change in average Nusselt number,
and corresponding pressure loss, resulting due to the geometric change is illustrated
in Fig. 5. For a relative assessment, average values are scaled by the baseline values
(i.e. NuVG /NuO and PVG /PO ). Here, subscript ‘VG’ refers to the fins with vortex
generators and ‘O’ refers to the plain fins. Evidently, both performance parameters obey
a second order polynomial function with high R-square value. It may further be noticed
that the performance augmentation is an inverse function of winglets’ aspect ratio. The
highest augmentation in the average Nusselt number of the finned tube array equals
125.1% whereas the pressure loss rises by 176.9%. Furthermore, highest rise in the
average Nusselt number and the pressure loss, due to reduction in the aspect ratio, is
14.6% and 34.1% respectively. These values are obtained by treating highest aspect ratio
(i.e. shortest winglet) as the reference.
 Numerical Analysis of Capacity Augmentation of Finned-Tube Systems 737

(a) (b)
X = 0.5D

X = 1.0D

X = 1.5D

Projection planes
(1) X = 0.5D
(2) X = 1.0D
(c)
(3) X = 1.5D AR=1.5

AR=1.0

(d) AR=0.75

AR=0.5

Note: Above vectors are in the projection plane that

lies at 0.5D in VG downstream

Fig. 4. Velocity vectors. a Fin with toe-out winglets. b Secondary flow generation by the winglets
bearing AR = 0.5. c Effect of winglet aspect ratio. d Pathlines: Plain fin, AR = 0.5

Fig. 5. Effect of winglet geometry on thermo-hydraulic performance.

4 Conclusions
This three dimensional computational study attempts to compensate the degradation in
the heat exchange capacity of finned tube arrays, arising due to compromised positioning
of the vortex generators, by varying the geometric design of those generators. The vortex
generators were erected at a preferred location, and their four different geometric designs
were considered for a detailed study. The changes in the flow structures are analysed,
which manifest in the form of higher Nusselt numbers. Additionally, it has been observed
that the spread of the tube wakes contracts with the decreasing aspect ratio. Such a change
is instrumental in making the isotherms approach concentricity. As a result, the highest
738 A. Arora and P. M. V. Subbarao

augmentation in the average Nusselt number of the finned tube array equals 125.1% at
the specified Reynolds number.

References
1. Arora, A., Subbarao, P.M.V., Agarwal, R.S.: Numerical optimization of location of ‘common
flow up’ delta winglets for inline aligned finned tube heat exchanger. Appl. Therm. Eng. 82,
329–340 (2015). https://doi.org/10.1016/j.applthermaleng.2015.02.071
2. Webb, R.L., Kim, N.H.: Principles of Enhanced Heat transfer. Taylor & Francis, NY (2005)
3. Gorji, M., Mirgolbabaei, H., Barari, A., Domairry, G., Nadim, N.: Numerical analysis on
longitudinal location optimization of vortex generator in compact heat exchangers. Int. J.
Numer. Meth. Fluids 66, 705–713 (2011). https://doi.org/10.1002/fld.2278
4. Wu, J.M., Tao, W.Q.: Numerical study on laminar convection heat transfer in a channel with
longitudinal vortex generator, Part B: parametric study of major influence factors. Int. J.
Heat Mass Transf. 51, 3683–3692 (2008). https://doi.org/10.1016/j.ijheatmasstransfer.2007.
03.031
5. Fiebig, M., Valencia, A., Mitra, N.K.: Wing-type vortex generators for fin-and-tube heat
exchangers. Exp. Thermal Fluid Sci. 7, 287–295 (1993). https://doi.org/10.1016/0894-177
7(93)90052-k
6. Pal, A., Bandyopadhyay, D., Biswas, G., Eswaran, V.: Enhancement of heat transfer using
delta-winglet type vortex generators with a common-flow-up arrangement. Numer. Heat
Transfer, Part A 61, 912–928 (2012). https://doi.org/10.1080/10407782.2012.677322
7. Kwak, K.M., Torii, K., Nishino, K.: Simultaneous heat transfer enhancement and pressure
loss reduction for finned-tube bundles with the first or two transverse rows of built-in winglets.
Exp. Thermal Fluid Sci. 29, 625–632 (2005). https://doi.org/10.1016/j.expthermflusci.2004.
08.005
8. Huisseune, H., T’Joen, C., Jaeger, P.D., Ameel, B., Schampheleire, S.D., Paepe, M.D.: Influ-
ence of the louver and delta winglet geometry on the thermal hydraulic performance of a
compound heat exchanger. Int. J. Heat Mass Transf. 57, 58–72 (2013)
9. Wu, X., Yuan, P., Luo, Z., Wang, L., Lu, Y.: Heat transfer and thermal resistance characteristics
of fin with built-in interrupted delta winglet type. Heat Transf. Eng. 37(2), 172–182 (2015)
10. Lu, G., Zhou, G.: Numerical simulation on performances of plane and curved winglet type
vortex generator pairs with punched holes. Int. J. Heat Mass Transf. 102, 679–690 (2016).
https://doi.org/10.1016/j.ijheatmasstransfer.2016.06.063
11. Tian, L., He, Y., Tao, Y., Tao, W.: A comparative study on the air-side performance of wavy fin-
and-tube heat exchanger with punched delta winglets in staggered and in-line arrangements.
Int. J. Therm. Sci. 48, 1765–1776 (2009). https://doi.org/10.1016/j.ijthermalsci.2009.02.007
12. Sinha, A., Chattopadhyay, H., Iyengar, A.K., Biswas, G.: Enhancement of heat transfer in a
fin-tube heat exchanger using rectangular winglet type vortex generators. Int. J. Heat Mass
Transf. 101, 667–681 (2016). https://doi.org/10.1016/j.ijheatmasstransfer.2016.05.032
 Numerical Investigation into the Effects of Fuel
Injection Parameters and Piston Bowl Geometry
on Exhaust Pollutants and Performance
of Diesel Engine

Doan Nguyen Cong1(B) and Khanh Nguyen Duc2

1 University of Transport Technology, Hanoi 100000, Vietnam
doannc@utt.edu.vn
2 Hanoi University of Science and Technology, Hanoi 100000, Vietnam

Abstract. For diesel engines, the fuel injection parameters and piston bowl geom-
etry (PBG) significantly influence the working process and exhaust emissions. This
paper presents the research results on the effects of the fuel injection parameters
such as the diameter of the spray nozzles, injection pressure, the start of injection
(SOI), and diameter of fuel particles injected into the cylinder in combination with
optimizing PBG on the performance and exhaust emissions of the diesel engine
ISUZU 4BD1T. The study procedure was conducted using the specialized sim-
ulation software Diesel-RK. The study results showed that the optimization of
fuel injection parameters, including reducing the diameter of the spray nozzles,
decreasing the fuel jet inclination angle, reducing the SOI, and combined with
the proposed PBG, significantly reduces fuel consumption by 2.0 g/kWh, NOx by
10.1%, and soot by 29.5%.

Keywords: Fuel injection parameters · Piston bowl geometry · Exhaust

emissions · Diesel-RK

1 Introduction
At present, the development of new and improved manufactured internal combustion
engines (ICE) cannot be imagined without computational studies on a computer. Math-
ematical modeling and computer optimization of ICE are essential in tightly-designated
standards for harmful emissions from exhaust gases, high specific power, and efficiency
requirements, and when the volume and cost of experimental work increase dramatically.
Leading research centers are conducting large-scale research to optimize the organiza-
tion of engine workflows, including diesel fuel equipment control algorithms that would
provide standardized levels of harmful emissions. The hardest part is reducing both par-
ticulate matter and nitrogen oxides simultaneously because most ways to do this have a
positive effect on reducing the level of one component but a negative effect on another.
The sprayer design and location, injection profile, piston bowl shape, swirl, compo-
sition, pressure, and charge temperature significantly impact the combustion process in

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 739–751, 2023.
https://doi.org/10.1007/978-3-031-22200-9_78
740 D. N. Cong and K. N. Duc

a direct injection diesel engine. The development of a high-accuracy combustion model

taking into account the characteristics mentioned above is taking place to optimize
combustion and enhance emissions of a direct injection diesel engine with the thermo-
dynamic engine model. Because there are up to 30,000 engine simulation sessions in
optimization over the whole operating range, the choice of thermodynamic engine model
is constrained by the need for a high computational rate.
In recent years, a few experimental and numerical studies have been conducted
relating the effects of PBG and the fuel injection parameters on the performance and
emissions characteristics of diesel engines fuelled with conventional and alternative
fuels. The ordinary PBG was transformed into a toroidal and hemispherical particle by
Kumar V et al. [1]. It was found that the thermal brake efficiency rose by 1.9% for the
hemispherical piston and 2.9% for the toroidal piston. In contrast, the NOx emissions
increased for the hemispherical piston by 4.2% and the toroidal piston by 9.1%. The
brake-specific fuel consumption (BSFC) fell by 8.7% for the toroidal piston and 5.9%
for the hemispherical piston. Two distinct piston geometries, deep bowl and toroidal
bowl operated with grape seed oil methyl ester were introduced by Sankar Ganesh R
et al. [2]. According to the experiments, toroidal bowl geometry produced engines with
average engine characteristics for all mixtures. Brake thermal efficiency increased by up
to 4.7% with 25% mixed fuel. Smoke, HC, and CO emissions were reduced by around
9.2%, 30.2%, and 4.6%, respectively. Jaichandar and Annamalai [3] conducted engine
experimental testing using three different combustion chamber geometries: hemispher-
ical, toroidal, and shallow without changing the engine’s compression ratio. According
to their analysis, the toroidal combustion chamber outperformed the other two due to its
geometric design, which improved performance and emission.
Venkateswaran et al.[4] demonstrated that the swirl and turbulent kinetic energy
with bowl geometry have more re-entrance than the baseline bowl, resulting in better
combustion. Consequently, better engine performance, adequate fuel consumption, and
soot emissions were enhanced. Park [5] revised bowl shapes and fuel injection parame-
ters, resulting in lower engine emissions and better engine performance than the original
combustion chambers. A high re-entrant bowl produced better swirl and turbulent kinetic
energy and reduced NOx and soot emission levels, according to the numerical simula-
tion of different swirl-inducing piston bowls in diesel engines [6]. In order to achieve
a constant compression ratio of 17.5 with biodiesel mixes with diesel and various pis-
ton bowl shapes for the same bowl volume, Shahanwaz Khan et al. [7] undertook their
research. The AVL FIRE simulation tool was used to simulate the in-cylinder flow and
combustion process, and experimental data from a baseline hemispherical bowl was uti-
lized to validate the numerical model. The simulation results demonstrate that the piston
bowl arrangement in a diesel engine significantly impacts the flow behavior inside the
combustion chamber. The engine combustion chamber was tuned by Subramanian et al.
[8], who concluded that the CFD findings had assisted in determining the best piston
bowl profile for generating the additional swirl and turbulent kinetic energy needed for
improved combustion.
However, the previous research works have not considered the design features of the
fuel equipment, atomization fineness, features of the injection characteristics, direction-
ality and dynamics of the fuel jets, interaction of the jets with the air vortex and with
 Numerical Investigation into the Effects of Fuel Injection Parameters 741

the combustion chamber walls, the conditions for the development of near-wall flows
(NWF) from each fuel jet under the conditions of a particular combustion chamber, as
well as the interaction of these flows among themselves. Hence, the objective of the
research is to investigate the combined effect of optimization fuel injection parameters
and PBG of diesel engines using the specialized simulation software Diesel-RK which
is commonly used in the field of ICE research to reduce exhaust emissions and fuel
consumption.

2 The Multi-zone Diesel Spray Combustion Model

The multi-zone diesel spray combustion model [9], also known as the RK-model, incor-
porates two distinct emission sub-models: the NOx formation sub-model based on the
Zeldovich’s scheme [10], and the soot formation sub-model. Razleytsev [12] developed
the latter, whereas Zvonov [11] developed the former.
After combustion modeling, distinct techniques are used to simulate the generation
of soot and NOx . Razleytsev created the RK-fundamental model’s equations between
1990 and 1994. Kuleshov later improved and supplemented this approach [9].
The technique considers the evolutionary conditions of each fuel spray, the NWF
produced by sprays, and the interaction between sprays and the swirl and the NWF cre-
ated by nearby sprays. These RK-model features enable predictions of diesel combustion
and emissions across the whole operating range for the various parameters listed below:

• piston bowl shape and sprayer location;

• swirl intensity;
• number, diameter, and direction of sprayer nozzles;
• the shape of the injection profile, including different strategies for multiple injections.

According to the Razleitsev model, the jet in its development goes through three stages:

• The formation of an initial dense axial flow of air and drops.

• The main development stage of a cumulative jet with the deceleration and destruction
of the axial flow in the leading front.
• Period of spray interaction with the combustion chamber walls and fuel distribution
on the walls. The shape and speed of the parietal spot spread in different directions
depend on the angle of the jet with a wall and the influence of an air vortex.

2.1 Distribution of Fuel in a Spray

In the RKmodel, a spray injected into the combustion chamber is divided into 7 charac-
teristics (Fig. 1). Each zone has its own evaporation and burning conditions, which are
uniform within the whole zone.
While the spray is free (before impingement) there are only three characteristic
zones: dilute outer sleeve, dense axial core, and dense forward front. The NWF being
formed after impingement is inhomogeneous in structure, density, and temperature,
which makes the calculation of fuel evaporation difficult; it is therefore, expedient to
742 D. N. Cong and K. N. Duc

Fig. 1. Diagram of a diesel spray divided into zones. 1—dilute outer sleeve; 2—dense axial core;
3—dense forward front; 4—the axial conical core of a NWF; 5—the dense core of a NWF on a
piston bowl surface; 6—dense forward front of a NWF; 7—dilute outer sleeve of NWF; bm —depth
of spray forward front; hwfr —the height of NWF forward front; I—cylinder head; II—cylinder
liner; III—piston.

distinguish typical zones with averaged heat and mass transfer coefficients in the NWF
by analogy with the free spray. After wall impingement, a new set of zones is taken into
account: 4—the axial conical core of a NWF, 5—dense core of a NWF on a piston bowl
surface, 6—dense forward front of a NWF, 7—dilute outer sleeve of NWF.
Assumptions to be used when determining the distribution of sprays:

• Every spray is simulated on the assumption that the spray is alone.

• The fuel fractions are summed in each zone for all sprays.

2.2 Visualization of the Mixing Process

The program for visualizing the mixture formation process in a diesel engine, Fuel Jet
Visualization (FJV), is a part of the Diesel-RK software package. The program visually
displays the results of the calculation performed when simulating the engine workflow.
The program displays on the monitor screen a moving picture of the development of
fuel jets and their NWF, as well as graphs of the distribution of atomized fuel over the
characteristic zones of each jet (Fig. 2). The program allows to follow step by step the
rotation angle of the crankshaft:

• development of fuel jets and their NWF in the combustion chamber, taking into account
the portable influence of the vortex,
• deformation of jets by a vortex,
• the formation of NWF depends on the angles of the jets meeting the walls,
• development of NWF under the influence of a vortex and their interaction with each
other,
• distribution of fuel over characteristic zones (Fig. 2),
• injection speed and heat release.

At the end of the calculation, a summary table is displayed indicating the amount of
fuel that fell into each characteristic zone for each jet and all jets in general. Based on
 Numerical Investigation into the Effects of Fuel Injection Parameters 743

Fig. 2. Graphical display of the results of calculating the mixture formation of test engine ISUZU
4BD1T in the basic configuration.

the analysis of these data, as well as the analysis of the integral indicators of the diesel
engine obtained in the calculation and visual analysis of the development of jets using
the FJV program, it is possible to refine the orientation of each particular jet in the space
of the combustion chamber, modify the shape of the chamber in the piston, ensuring that
would be the maximum amount of fuel fell into zones with good evaporation conditions:
the jet shell, the core of the NWF; and, if possible, exclude fuel from getting into zones
with a low evaporation rate: on the surface of the cover and cylinder liner, as well as in
the intersection zones of NWF from neighboring jets.
The use of fuel distribution data, integrated indicators and the FJV program makes
it possible not only to compare various design options with each other and evaluate
their strengths and weaknesses but also to select such combinations of PBG and fuel jet
orientation that would allow achieving a reduction in emissions of harmful substances
and the lowest fuel consumption.

3 Test Engine Specification and Model Validation

3.1 Test Engine Specification

The simulation model was built based on structures of the test engine ISUZU 4BD1T
[13], which is commonly used in light-duty trucks in Vietnam, corresponding to the
specifications listed in Table 1.
744 D. N. Cong and K. N. Duc

Table 1. Specifications of the test engine

Item Technical parameters Unit

Engine name and branch 4BD1T—ISUZU –
Type of engine 4-strokes, 4-cylinder, direct injection, turbo-charger, –
firing order 1–3-4–2
Bore x stroke 102x118 Mm
Displacement 3856 cm3
Compression ratio 17.5 –
Max. power/engine speed 78/2500 kW/rpm
Max. torque/engine speed 325/1600 Nm/rpm
Specific fuel consumption 234 g/kWh
Start of injection 16 BTDC Deg
Injection pressure 185 kg/cm2

3.2 Model Validation

A numerical simulation was carried out at full load and a constant speed of 2500 rpm.
The anticipated results with baseline diesel were compared with the experimental find-
ings carried out on the test engine with the original combustion chamber to verify the
calculations’ accuracy. In-cylinder pressure profile was measured to assess the com-
bustion process using the AVL Indiset 620 device. Figure 3 clearly shows that, notably
during the compression and expansion strokes, simulated and experimental in-cylinder
pressure curves correspond well, indicating that the combustion process has been well
modeled.

Fig. 3. Validation results of cylinder pressure at full load operation.

 Numerical Investigation into the Effects of Fuel Injection Parameters 745

4 Simulation of Sprays Behavior and Fuel Distribution in Engine

with Direct Injection System
In this work, we studied the possibility of reducing the NOx emission and smoke level
of engine ISUZU 4BD1T by changing the following parameters:

• diameter of spray nozzles (dn);

• piston bowl geometry;
• fuel jet inclination angles (γ);
• start of injection (SOI).

The possibility of a significant increase in injection intensity has not been investigated
since implementing measures to increase the injection pressure radically requires a
serious revision of the design of the diesel engine as a whole.

4.1 Decrease the Diameter of the Spray Nozzles

When the diameter of the spray nozzles is reduced from 0.32 mm (basic configuration)
to 0.25 mm (configuration 1), the injection characteristic of fuel equipment changes
significantly. The maximum injection pressure Pinj_max increases from 387 to 577 bar;
injection duration (ID) increases from 21.5 to 28.3 degrees of crank angle (dCA).
The calculation results of the working process of the diesel engine in the basic con-
figuration, but with reduced nozzle diameter (Table 2) showed a significant improvement
in the quality of fuel atomization (the average droplet diameter d32 radically decreases
from 25 μm to 19 μm). In addition, a prolonged injection leads to a favorable fuel
distribution:

• 74.07% of the fuel in a rarefied shell of the jet and the shell of the NWF, where there
are the best evaporation conditions (for comparison, with nozzles of 0.3 mm in these
zones, the figure is 67.33%),
• the share of fuel in the core of NWF on the piston wall, where there are also
good evaporation conditions, is 22.45% (whereas with nozzles of 0.3 mm it reaches
26.18%),
• 3.49% of the fuel in the core of free spray where the evaporation rate is low (for
comparison, with nozzles of 0.3 mm in these zones, the figure is 6.49%),
• in both cases, the fuel does not get into areas with bad evaporation conditions: on the
cylinder head wall, the cylinder liner, and in the intersection areas of NWF formed by
adjacent sprays.

In conclusion, a decrease in the diameter of spray nozzles increased the fuel injec-
tion pressure, which in turn significantly improves the quality of fuel atomization and
consequently leads to a high rate of evaporation as a result of:

• BSFC decreases by 2 g/kWh;

• emission of NOx practically does not change;
• Hartridge smoke level decreases by 29.5% (Table 2).
746 D. N. Cong and K. N. Duc

Table 2. Results of working process calculation decreasing the diameter of the spray nozzles

Configuration Basic No.1

Pinj_max (bar) 387 577
ID (deg) 21.5 28.3
SOI (deg) BTDC 16 16
d32 (μm) 25 19
Ne (kW) 78.2 78.8
be (kg/kWh) 0.234 0.232
NOx (ppm) 1431 1425
Hartridge smoke level 8.8 6.2

To reduce the fuel consumption and soot emissions, it is necessary to change the shape of
the combustion chamber and the direction of the nozzles in such a way as to provide fuel
jets with greater freedom for development, which will increase the rate of heat release
in the afterburning section.

4.2 Improving Piston Bowl Geometry and Optimizing the Fuel Jet Orientation
Further research was carried out to improve the piston bowl geometry and optimize the
fuel jet orientation. It is known that the inclination angle of the spray holes (the angle
between the axis of the cylinder and the axis of the jet), especially for diesel engines
with volumetric mixing, should have an optimal value at the best performance of the
internal combustion engine is achieved.
Satisfactory results are shown in the proposed combustion chamber shown in Fig. 4.
The angles between the jet axes and the cylinder axis (γ) are decreased from 70 to
65 deg while the outer recess diameter is increased. These measures make it possible
to increase the length of the free development of the jets. The inclination angle of the
generatrix of the chamber in the piston is chosen so that the main development of the
NWF is directed deep into the chamber, and not on the surface of the piston crown in
the over-piston gap. The described combustion chamber and nozzles 4 × 0.25 with a
decreased cone angle, make up the configuration No.2.
Table 3 shows the indicators for typical diesel configurations:

• Basic—basic configuration with nozzles 4 x 0.32;

• No.1—basic configuration with nozzles 4 x 0.25;
• No.2—proposed combustion chamber, decreased fuel jet inclination angles, nozzles
4 x 0.25.
 Numerical Investigation into the Effects of Fuel Injection Parameters 747

Fig. 4. Graphical display of the results of calculating the mixture formation of test engine in
configuration №2.

Table 3. Results of working process calculation of configurations

Configuration Basic No.1 No.2

Pinj_max (bar) 387 577 577
ID (deg) 21.5 28.3 28.3
SOI (deg) BTDC 16 16 16
d32 (μm) 25 19 19
Ne (kW) 78.2 78.8 78.9
be (kg/kWh) 0.234 0.232 0.232
NOx (ppm) 1431 1425 1524
Hartridge smoke level 8.8 6.2 5.2

The distribution of the injected fuel in the proposed combustion chamber is more
favorable than in the base one due to the following:

• in the rarefied shell of the jet and the shell of the NWF, where there are the best condi-
tions for evaporation, 76.38% of the fuel are distributed, while in the base combustion
chamber, the figure is only 67.33%;
• in the core of the NWF, where there are also good evaporation conditions, there is
20.14% of the fuel in the proposed combustion chamber, and in the base combustion
chamber, it reaches 26.18%;
748 D. N. Cong and K. N. Duc

• in both chambers, the fuel does not enter areas with poor evaporation conditions: on
the cylinder head wall, on the cylinder liner, and in the zones of the intersection of
NWF.
• small average droplet diameter d32 (19 μm) causes a high evaporation rate.

Those decreasing diameter of the spray nozzles combination with improving the chamber
shape in the piston and optimizing the fuel jet orientation, such significant improvement
in the quality of fuel atomization causes a high rate of evaporation as a result of:

• BSFC decreases by 2 g/kWh;

• emission of NOx increases by 6.5%;
• Hartridge smoke level decreases by 40.9%.

Despite the reduction in soot emissions, a similar result cannot be considered satisfactory
all due to an increase in NOx emissions. The reserve for reducing NOx emissions of the
engine is to change the SOI.

4.3 Optimizing the Start of Injection

The described combustion chamber and nozzles 4 x 0.25 with a decreased cone angle,
in combination with the SOI of 14 dCA, make up the configuration No.3. Injection
profiles and characteristics of the combustion zone temperature of basic configuration
and configuration No.3 are described in Fig. 5. SOI equal to 14 dCA was chosen based
on the results presented in Fig. 6.

Fig. 5. Injection profiles and combustion zone temperature in the basic configuration and
configuration No.3.
 Numerical Investigation into the Effects of Fuel Injection Parameters 749

Figure 6 shows the dependence of specific effective fuel consumption, NOx , and
soot emissions on the SOI.

Fig. 6. Influence of SOI on the specific fuel consumption, NOx and soot emissions.

Table 4 compares the engine parameters obtained in configuration No.3 and the
basic configuration. The diameter of the nozzles is reduced from 0.32 to 0.25 mm, and
simultaneously the SOI should be reduced from 16 to 14 dCA. Wherein:

• the level of NOx emissions decreases from 1431 to 1286 ppm (by 10.1%);
• the level of soot emissions decreases from 8.8 to 6.2 Hartridge units (by 29.5%);
• BSFC decreases by 2.0 g/kWh.

It is known that the rising of NOx level is the result of a higher in-cylinder temperature
and the appearance of oxygen atoms in fuel molecular [14, 15]. Hence, the injection
extended in time, a small delay in self-ignition, and a small amount of fuel evaporated
during the delay period cause a decline in the combustion zone temperature (Fig. 4) and,
consequently leads to a decrease in NOx emissions.
750 D. N. Cong and K. N. Duc

Table 4. Results of working process calculation of basic and No.3 configurations

Configuration Basic No.3

Pinj_max (bar) 387 577
ID (deg) 21.5 28.3
SOI (deg) BTDC 16 14
d32 (μm) 25 19
Ne (kW) 78.2 79
be (kg/kW.h) 0.234 0.232
NOx (ppm) 1431 1286
Hartridge smoke level 8.8 6.2

5 Conclusion
An analysis of the distribution of air-fuel mixture vapors and the interaction of the fuel
jet with the combustion chamber walls indicates that the injection parameters and the
shape of the chamber in the piston, incorporated in the existing design of the test engine,
do not provide efficient operation in the full load mode and must be optimized.
The fuel jet inclination angle, PBG, and SOI significantly influence the quality of the
air-fuel mixture, and an increase in injection pressure contributes to a rise in the disper-
sion of fuel droplets and accelerates their evaporation. Optimization of these parameters
significantly reduces the impact of the fuel jet on the combustion chamber walls.
Thus, the use of a pilot combustion chamber, in combination with the optimization of
fuel injection parameters, including reducing the diameter of the spray nozzles, decreas-
ing the fuel jet inclination angle, and reducing the SOI, leads to a reduction in NOx by
10.1%, in soot emissions by 29.5% and even in fuel consumption by 2.0 g/kWh.
To find ways to further reduce fuel consumption and smoke emissions, it is necessary
to conduct additional in-depth numerical studies on the selection of the PBG, as well as
the number, diameter, and direction of nozzle holes, in order to provide fuel jets with
greater freedom for development and increase the rate of heat release in the afterburning
section.
The use of the Diesel-RK software package and the FJV program built into it makes
it possible to link often opposite trends caused by the need to reduce fuel consumption
and emissions of harmful substances and to find profitable compromises.

References
1. Kumar, V.: Experimental investigation of piston bowl geometry effects on performance and
emissions characteristics of diesel engine at variable injection pressure and timings. Int. J.
Ambient Energy 39(7), 685–693 (2017)
2. Ramaraj, S.G., Balakrishnan, G.B., Chidambaram, G.S.: Influence of combustion chamber
bowl geometry on combustion, performance and emission characteristics of a diesel engine
using grape seed oil methyl ester. Therm. Sci. 21(2), 443–451 (2017)
 Numerical Investigation into the Effects of Fuel Injection Parameters 751

3. Jaichandar, S., Annamalai, K.: Effects of open combustion chamber geometries on the
performance of pongamia biodiesel in a DI diesel engine. Fuel 98, 272–279 (2012)
4. Venkateswaran, S.P., Nagarajan, G.: Effects of the re-entrant bowl geometry on a DI tur-
bocharged diesel engine performance and emissions – A CFD approach. J. Eng. Gas Turbines
Power 132(12), 122803–122811 (2010)
5. Park, S.W.: Optimization of combustion chamber geometry and engine operating conditions
for compression ignition engine fueled with dimethyl ether. Fuel 97, 61–71 (2012)
6. Prasad, B.V.V.S.U., Sharma, C.S., Anand, T.N.C., Ravikrishna, R.V.: High swirl inducing
piston bowls in small diesel engines for emission reduction. Appl. Energy 88(7), 2355–2367
(2011)
7. Shahanwaz, K., Rajsekhar, P., Probir, K.B.: The impact of combustion chamber configuration
on combustion and emissions of a single cylinder diesel engine fuelled with soybean methyl
ester blends with diesel. Renew. Energy 143(C), 335–351 (2019)
8. Subramanian, S., Rathinam, B., Lalvani, J., Annamalai, K.: Piston bowl optimization for
single cylinder diesel engine using CFD. SAE Paper 2016-28-0107 (2016)
9. Kuleshov, A.S.: Model for predicting air-fuel mixing, combustion and emissions in DI diesel
engines over whole operating range. SAE Paper No 2005-01-2119 (2005)
10. Zeldovich, Ya.B., Raizer, Yu.P.: Physics of shock waves and high-temperature hydrodynamic
phenomena. Nauka, Moscow, 686 p. (1966)
11. Zvonov, V.A.: Internal combustion engines toxicity. Mashinostroenie, Moscow, 200 p. (1973)
12. Razleytsev, N.F.:Combustion simulation and optimization in diesels. Vischashkola, Kharkov,
169 p. (1980)
13. Catalog diesel engine ISUZU 4BD1T
14. Raine, R.R., Stone, C.R., Gould, J.: Modeling of nitric oxide formation in spark ignition
engines with a multizone burned gas. Combust. Flame 102(3), 241–255 (1995)
15. Velmurugan, K., Sathiyagnanam, A.P.: Effect of biodiesel fuel properties and formation of
NOx emissions: a review. Int. J. Ambient Energy 38(6), 644–651 (2017)
 Obstacle Avoidance Algorithm for Autonomous
Mobile Robots in the Indoor Environment

Nguyen Hong Thai1(B) , Trinh Thi Khanh Ly2 , Nguyen Thanh Long1 ,
and Than Thi Thuong3
1 School of Mechanical, Hanoi University of Science and Technology (HUST), Hanoi, Vietnam
thai.nguyenhong@hust.edu.vn
2 Electric Power University (EPU), Hanoi, Vietnam
3 University of Economics—Technology for Industries (UNETI), Hanoi, Vietnam

Abstract. Autonomous mobile robots (AMR) are commonly used in intelligent

manufacturing systems. For these robots to operate autonomously in the working
environment, in addition to being equipped with a sensor system, algorithms are
required to process information collected from the working environment and make
control decisions to control the robot’s movement in different situations. In this
paper, the authors present an algorithm to avoid static, dynamic obstacles and
generate local guidance trajectories for the robot to overcome the barriers on its
way based on information from the sensor network equipped with the robot. In
this, dynamic obstacles are understood as objects moving in the same direction
opposite to the direction of the robot or crossing the motion of the robot. The
proposed algorithm has been experimentally verified on a differential drive mobile
robot combined with a mecanum wheel mobile robot. The results of this research
have practical implications for motion control of AMR in industries.

Keyword: Autonomous mobile robots · Differential drive mobile robot · Static

obstacles · Dynamic obstacles

1 Introduction

Nowadays, wheel mobile robots have been widely utilized for many different applica-
tions of life, from serving people in industrial logistics services to military fields such
as floor cleaning [1], Supermarket [2, 3], Hospital transportation [4, 5]; Autonomous
guided vehicles in smart factories [6, 7], Military robot multi-purpose wheeled (UGVs)
[8], Borehole drilling robot in tunnelling construction [9], etc. In the indoor environ-
ment, Autonomous Mobile Robots (AMR) have been developed to replace Automated
Guided Vehicles (AGVs). They can be integrated into the system through IoT connec-
tions to become intelligent transport and service systems. In order for the AMRs to work
effectively, much effort has been made to control the desired trajectory with different
controllers such as the linear state feedback controller [10–13], the fuzzy-proportional
integral (fuzzy-PI) controller [14, 15], the time-varying parameter PID controller [16–
18]. In addition, path planning and obstacle avoidance are essential issues for AMRs.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 752–763, 2023.
https://doi.org/10.1007/978-3-031-22200-9_79
 Obstacle Avoidance Algorithm for Autonomous Mobile Robots 753

This is a subject of interest to many researchers, such as: using Dijkstra’s algorithm for
global path planning and static obstruction avoidance [19, 20]; Improving path planning,
and optimizing the path length by the combination of long short-term memory (LSTM)
neural network, fuzzy logic control, and reinforcement learning [21, 22] or genetic algo-
rithm [23]. However, most static and dynamic obstacle avoidance studies’ limitations
are based on the distance from the robot to the obstacle configuration [7, 25] without
considering the moving obstacle’s speed and the obstacle’s direction of motion. That
will not match the reality in the industrial environment when the dynamic obstacle has
the speed moving in the same direction or opposite to the robot’s movement.
To overcome the above problems in this study, we propose an algorithm to avoid
static and dynamic obstacles in the indoor environment based on the distance and relative
motion velocity between the robot and the dynamic obstacles. The robot’s sensor network
detects the dynamic obstacles’ distance and velocity.

2 Static and Dynamic Obstacle Avoidance Algorithms in the Indoor

Environment
2.1 Description of Obstacle Avoidance Situations in the Indoor Environment
Typically, obstacles in the industrial environment have two main objects: (1) Static
obstructions, including trolleys, machinery and equipment and (2) Dynamic obstacles,
including transport vehicles and people moving in production lines. Thus, for dynamic
obstacles, there are the following situations: The moving obstacle moves in the same
direction as the robot. In this case, there are two possibilities: (a) The moving object
has a faster moving speed than the robot, and (b) The moving object is moving at a
slow-moving speed than the robot’s movement speed; (ii) Dynamic obstacles move in
the opposite direction of the robot’s movement; (iii) A moving obstacle cuts across the
robot’s path at intersections, and (iv) A dynamic obstacle stops in the robot’s path for
a short period to perform some work. In order for the robot to recognize static and
dynamic obstacles from the environment, the robot is equipped with the following types
of sensors: (a) Ultrasonic sensor, (b) Lidar scanner and (c) Depth camera. In addition, to
instantly locate the robot on the moving map, the robot is equipped with an automatic
6-DoF IMU sensor combined with angular velocity sensors at the driving wheels.
Hence, for the robot to handle the above situations, it is necessary to have processing
algorithms to decide on motion control. This problem is presented in Sect. 2.2 below.

2.2 Static and Dynamic Obstacle Avoidance Algorithms in the Indoor

Environment
As analyzed and described in Sect. 2.1, there are five possible situations of static and
dynamic obstacles that the robot must pass during its movement. For simplicity, Fig. 1
below depicts the motion relationship of two robots to illustrate five situations in which
the case of a moving obstacle is considered a static obstacle when the second robot has
a velocity V 2 = 0.
From Fig. 1, if we call V 1 the instantaneous motion speed of the robot determined
by the IMU sensor, V 2 is the speed of the obstacle (Mecanum wheel mobile robot)
754 N. H. Thai et al.

Fig. 1. Static and dynamic obstacle avoidance situations on the robot’s path with (a) static and
dynamic obstacle avoidance in the path and (b) the moving, dynamic obstacle horizontally in the
robot’s path at the intersection.

determined by the Camera. L is the safe distance installed for the robot to make navigation
decisions and generate a local path to overcome the obstacle. L t is the instantaneous
distance from the robot to the obstacle determined by the ultrasonic sensors and the
Lidar scanner. Then the relative velocity of the robot relative to the obstacle is given by:
V = V1 − V2 (1)
From Eq. (1) corresponding to the obstacle situations described in Sect. 2.1, we have:

(1) V = V 1 the obstacle is a static obstacle (V 2 = 0), and the safe distances are
determined:
L − Lt = 0 (2)
(2) 0 < V < V 1 then the obstacle moves in the same direction but has a lower velocity
than the robot’s speed, the safe distance for the robot to decide to generate a local
trajectory over the obstacle is expressed by:
L − Lt μ = 0 (3)
wherein μ = g(L, V1 , V2 ) is the coefficient of safety determined from the
relative motion between the robot and the obstacle in the same direction.
(3) V < 0 the dynamic obstacle moves in the same direction as the robot but has a
speed faster than the actual speed of the robot. Thus, the robot moves at a steady
speed.
 Obstacle Avoidance Algorithm for Autonomous Mobile Robots 755

(4) V > V 1 the dynamic obstacle moves in the opposite direction to the robot’s
direction of motion, and the safe distance for the robot to avoid the obstacle is
determined as follows:

L − Lt β = 0 (4)

Here β = f (Lt , V1 , V2 ) is the coefficient of safety determined from the relative

motion between the robot and the dynamic obstacle in the opposite direction.
(5) V = V 1 and there is a gradual shift in the frame size in the horizontal direction
(recognized by the camera) when the dynamic obstacle horizontally moves the
robot’s path at the intersection. Thus, the robot will stop at a safe distance given in
Eq. (2) until the dynamic obstacle passes.

Figure 2 shows the robot’s static and dynamic obstacle avoidance algorithm during
motion.

Fig. 2. Static and dynamic obstacle avoidance algorithm.

756 N. H. Thai et al.

In Fig. 2, d l , d r are the distances from the obstacle to the robot path at the right and
left, respectively. Df , Dt are the given safety distance Df = Dt = d + d with Δd being
the safe distance next to the vehicle to avoid collision with the left and right flanks of
the robot. V Cy is the horizontal velocity of the moving obstacle.
Algorithm to overcome obstacles to the global path of the robot
After overcoming obstacles (Static/Dynamic), the robot moves to the global trajectory in
the robot’s movement map using the algorithm given in Fig. 3 relies on a sensor network
behind the robot (Ultrasonic Sensors and Lidar scanners).

Fig. 3. The robot’s static and dynamic obstacles are overcoming algorithms with (a) An illustration
of a robot overcoming an obstacle and (b) An algorithm of overcoming an obstacle.

Also, a problem is related to the local path when the robot crosses an obstacle to
return to the global path. Turns must be smooth at the bends to avoid sudden speed
changes. We used Dubins’ method [24] to solve this problem, as illustrated in Fig. 3a.
Where the radius of the arc is defined by:
y
ρ= (5)
2(1 − cos α)
Here α is the angle subtended by an arc, and y = d /2 + d .

3 Experimental Results
3.1 Sensor Network Equipped on the Robot
Sensors are equipped on the robot to recognize the environment and locate the robot
in the moving map as described in Fig. 4. In which (a) The sensor system to recognize
the moving environment includes: Four Ultrasonic sensors (URM37) to measure the
distance of obstacles, a Depth camera (IMX219-83 stereo) to recognize moving images,
 Obstacle Avoidance Algorithm for Autonomous Mobile Robots 757

two Lidar scanning (RPLIDAR A1-360) detects dynamic and static obstacles; (b) The
INS inertial navigation system consists of an IMU (SEN0386 Serial 6-Axis) combined
with two encoders (LPD3806-400BM-G5-24C) mounted on the two-wheel-drive; (c)
(c) Jetson Nano AI Embedded Computer is used as the robot’s central controller for
data processing and motion control. Figure 5 describes the arrangement of the sensor
network on the robot to locate the robot’s position in the moving map and recognize the
environment.

Fig. 4. Sensors identify obstacles and locate of the robot.

3.2 Results and Discussion

Experimental parameters setup
The robot dimensions are length x width x height (1.05 m × 0.65 m × 0.32 m); The
mecanum-wheeled mobile robot has dimensions length × width × height (0.5 m ×
0.35 m × 0.18m); Static obstruction has dimensions length × width × height (0.49 m
× 0.36 m × 0.28 m); The distance is safe from the robot to an obstacle L = 1.85 m; The
safe area around the robot Δd is 0.15m; The width of the robot’s movement path D is
1.7 m.
Static obstacle avoidance experiment
Figure 6 is the snapshot of the actual robot at three times: (a) Detects an obstacle and
starts navigating to generate a local path to avoid obstacles using Dubins’ algorithm
when it detects enough distance to pass, (b) Robot move over the obstacle, and (c) The
detection robot has passed the obstacle safely and returns to path planning. In this case,
the robot’s speed is set to V 1 = 0.1 m/s.
758 N. H. Thai et al.

Fig. 5. Sensor layout on the robot with (a) Photograph of the robot, (b) Positioning and navigation
sensors, (c) In front of the robot and (d) Behind the robot.

Fig. 6. Robot overcomes static obstacles.

 Obstacle Avoidance Algorithm for Autonomous Mobile Robots 759

Figure 6a shows that the error between the set value and the actual measured value
is 0.02 m, which is caused by the delay in the control signal from the time the obstacle
is detected at t = 20.1 s to the obstacle avoidance.
Overcoming obstacles in the same direction as the robot’s movement
In this case, the parameter is set as follows: the speed of the moving obstacle (The
mecanum-wheeled mobile robot) is 0.1 m/s, while the robot’s speed is set to 0.3 m/s.
Figure 7 is a photo taken during the robot’s process, avoiding moving obstacles in
the same direction as the moving direction of the robot. Figure 7a describes the robot
that detects dynamic obstacles (in this case, μ = 1.85) and performs obstacle avoidance.
Figure 7b the robot overcomes dynamic obstacles along a local path. Meanwhile, Fig. 7c
robot starts to exit the local path to return to path planning with a safe distance L = f
(V 1, V 2, t) = 0.72 m.

Fig. 7. Robot overcomes dynamic obstacles moving in the same direction.

From Fig. 7a, the error between the calculated and measured values is 0.02 m. The
reason is that from detecting an obstacle to navigating to avoid an obstacle, the dynamic
obstacle has moved.
Overcoming obstacles in the opposite direction to the robot’s movement
In this case, the robot’s speed is set to V 1 = 0.2 m/s, and the dynamic obstacle’ speed is
V 2 = 0.1 m/s. The calculated safe distance value is L t = f (V 1, V 2, t) = 2.74 m with the
coefficient β = 0.675. Figure 8 shows the robot overcoming dynamic obstacles opposite
to the robot’s movement. Nevertheless, in this case, the error of the safe distance between
calculation and experiment is 0.02m. The safe distance is smaller than calculated because
760 N. H. Thai et al.

from the time the robot detects an obstacle at time t = 9.7 s to the time of avoiding the
obstacle, the robot and dynamic obstacle moving with a relative speed is 0.3 m/s.

Fig. 8. Robot overcomes dynamic obstacles moving in the opposite direction.

The dynamic obstacle horizontally moves the robot’s path at the intersection
In this experiment, the robot speed is set to 0.2 m/s, and the dynamic obstacle is set to
0.1 m/s. Figure 9 describes the waiting process for the dynamic obstacle robot to pass
through the intersection. However, in this situation, the error between the calculated and
set values is 0.01 m. The reason is that when the obstacle detects a moving object passing
through the inertial intersection, the robot moves a certain distance without being able
to make an emergency stop.
The experimental results in Fig. 9 show that from the time the robot detects a dynamic
obstacle passing until the safe horizontal distance it can travel is 9.2 s.
Discussion
From the experimental situations, the robot’s motion error compared with the set and the
calculated value is from 0.01 m to 0.02 m. It shows that the robot’s motion and stopping
accuracy depends on factors such as the robot’s speed, the speed of the dynamic obstacle,
the Inertia forces (the robot’s load), the signal delay, the traction robot’s wheels etc.
Therefore, to improve the robot’s motion accuracy, it is necessary to develop algo-
rithms to estimate the safe distance and change the speed according to the safe distance
to reduce the inertia force and the sliding phenomenon between the wheel and the road
surface.
 Obstacle Avoidance Algorithm for Autonomous Mobile Robots 761

Fig. 9. The robot overcomes the move dynamic obstacle horizontally the robot’s path at the
intersection.

4 Conclusion

From the proposed algorithm, experimental results and discussion this research achieves
the following points: (1) A static/dynamic obstacle avoidance algorithm has been pro-
posed in real situations by generating a local path based on the feedback signals from the
sensor network on the robot; (2) The algorithm’s reliability has been verified through
experiments based on manufactured AMR and mecanum-wheeled mobile robots by
authors. The results show that the robot’s motion error ranges from 0.01 to 0.02 m.
These errors are perfectly acceptable in practice in the industrial environment. That
shows that this algorithm can be applied to installing the AGVs in the field of transporta-
tion in industrial environments. However, this study still has some limitations, such as
not mentioning the load, the inertia force and the friction between the wheel and the
road surface. This is also the cause of the robot’s motion error. Thus, these problems
will target our future research to improve the motion accuracy of autonomous mobile
robots.

Acknowledgments. This research was funded by the Ministry of Industry and Trade in
a ministerial level scientific and technological research project, conducted in 2020, code:
DTKHCN.076/20.
762 N. H. Thai et al.

References
1. Singh, R., Bera, T.K., Chatti, N.: A real-time obstacle avoidance and path tracking strategy
for a mobile robot using machine-learning and vision-based approach. Simul. Trans. Soc.
Model. Simul. Int. 1–17 (2022)
2. Kirsch, C., Rohrig, C.: Global localization and position tracking of an automated guided vehi-
cle. In: Proceedings of the 18th World Congress The International Federation of Automatic
Control Milano (Italy), pp. 14036–14041 (2011)
3. Ly, T.T.K., Thai, N.H., Dzung, L.Q., Thanh, N.T.: Determination of kinematic control param-
eters of omnidirectional AGV robot with Mecanum wheels track the reference trajectory and
velocity. In: Sattler, K.U., Nguyen, D.C., Vu, N.P., Long, B.T., Puta, H. (Eds.), Advances
in Engineering Research and Application. ICERA 2020. Lecture Notes in Networks and
Systems, vol. 178, pp. 319–328. Springer, Cham (2020)
4. Bačík, J., Ďurovský, F., Biroš, M., Kyslan, K., Perduková, D., Padmanaban, S.: Pathfinder
development of automated guided vehicle for hospital logistics. IEEE Access 5(1), 26892–
26900 (2017)
5. Pedan, M., Gregor, M., Plinta, D.: Implementation of automated guided vehicle system in
healthcare facility. Proc. Eng. 192, 665–670 (2017)
6. Qian, J., Zi, B., Wang, D., Ma, Y., Zhang, D.: The design and development of an omni-
directional mobile robot oriented to an intelligent manufacturing system. Sensors 17(9), 2073
(2017)
7. Thai, N.H., Ly, T.T.K., Dzung, L.Q.: Roadmap, routing and obstacle avoidance of AGV robot
in the static environment of the flexible manufacturing system with matrix devices layout.
Sci. Technol. Dev. J. 24(3), 2091–2099 (2021)
8. Qian, Y., Wang, F., Lin, C., Huang, S., Guo, X.: Development of military unmanned ground
vehicles. Sens. Imaging Technol. 12065, 62–68 (2021)
9. Thai, N.H., Thai, N.Q.: A kinematic control algorithm for Blasthole drilling robotic arm in
tunneling. Sci. Technol. Dev. J. 20(K5), 13–22 (2017)
10. Thai, N.H., Ly, T.T.K., Long, N.T., Dzung, L.Q.: Trajectory tracking using linear state feed-
back controller for a Mecanum wheel omnidirectional. In: Khang, N.V., Hoang, N.Q., Cec-
carelli, M. (Eds.), Advances in Asian Mechanism and Machine Science. ASIAN MMS 2021.
Mechanisms and Machine Science, vol. 113, pp. 411–421. Springer, Cham (2021)
11. Thai, N.H., Ly, T.T.K.: NURBS curve trajectory tracking control for differential-drive mobile
robot by a linear state feedback controller. In: Nguyen, D.C., Vu, N.P., Long, B.T., Puta, H.,
Sattler, K.-U. (eds.) ICERA 2021. LNNS, vol. 366, pp. 685–696. Springer, Cham (2022).
https://doi.org/10.1007/978-3-030-92574-1_71
12. Thai, N.H., Thien, H., Ly, T.T.K.: NURBS curve trajectory tracking control for differential-
drive mobile robot by a linear state feedback dynamic controller. In: Le, A.T., Pham, V.S., Le,
M.Q., Pham, H.L. (Eds.), The AUN/SEED-Net Joint Regional Conference in Transportation,
Energy, and Mechanical Manufacturing Engineering. RCTEMME 2021. Lecture Notes in
Mechanical Engineering. Springer, Singapore, pp. 610–623 (2022)
13. Ly, T.T.K., Thien, H.: Bézier trajectory tracking control of the omnidirectional mobile robot
based on a linear time-varying state feedback controller. Sci. Technol. Dev. J. 25(2), 2444–
2452 (2022)
14. Masmoudi, M.S., Krichen, N., Masmoudi, M., Derbel, N.: Fuzzy logic controllers design for
omnidirectional mobile robot navigation. Appl. Soft Comput. 49, 901–919 (2016)
15. Alshorman, A.M., Alshorman, O., Irfan, M., Glowacz, A., Muhammad, F., Caesarendra,
W.: Fuzzy-based fault-tolerant control for omnidirectional mobile robot. Machines 8(3), 55
(2020). https://doi.org/10.3390/machines8030055
 Obstacle Avoidance Algorithm for Autonomous Mobile Robots 763

16. Thai, N.H., Ly, T.T.K., Dzung, L.: Q: trajectory tracking control for Mecanum wheel mobile
robot by time-varying parameter PID controller. Bull. Electr. Eng. Informat. 11(4), 1902–1910
(2022)
17. Thai, N.H., Ly, T.T.K., Dzung, L.: Q: trajectory tracking control for differential-drive mobile
robot by a variable parameter PID controller. Int. J. Mech. Eng. Robot. Res. 11(8), 614–621
(2022)
18. Thai, N.H., Ly, T.T.K.: Path tracking control for car-like robots by PID controller with time-
varying parameters. Sci. Technol. Dev. J. 5(3) (2022)
19. Brandão, A.S., Sarcinelli-Filho, M., Carelli, R.: An analytical approach to avoid obstacles in
mobile robot navigation. Int. J. Adv. Robot. Syst. 10(6), 278 (2013). https://doi.org/10.5772/
56613
20. Liu, L.-S., Lin, J.-F., Yao, J.-X., He, D.-W., Zheng, J.-S., Huang, J., Shi, P.: Path planning
for smart car based on Dijkstra algorithm and dynamic window approach. Wirel. Commun.
Mobile Comput. 1–12 (2021). https://doi.org/10.1155/2021/8881684
21. Guo, N., Li, C., Gao, T., Liu, G., Li, Y., Wang, D.: A fusion method of local path planning
for mobile robots based on LSTM neural network and reinforcement learning. Math. Probl.
Eng. 2021, 1–22 (2021). https://doi.org/10.1155/2021/5524232
22. Song, Q., et al.: Intelligent optimization algorithm-based path planning for a mobile robot.
Comput. Intell. Neurosci. 2021, 1–17 (2021). https://doi.org/10.1155/2021/8025730
23. Li, K., Qianqian, H., Liu, J.: Path planning of mobile robot based on improved multiobjective
genetic algorithm. Wirel. Commun. Mob. Comput. 2021, 1–12 (2021). https://doi.org/10.
1155/2021/8836615
24. Cook, G.: Mobile Robots Navigation Control and Remote Sensing. John Wiley & Sons Inc,
Hoboken, NJ, USA (2011)
25. Duguleana, M., Mogan, G.: Neural networks based reinforcement learning for mobile robots
obstacle avoidance. Expert Syst. Appl. 62, 104–115 (2016). https://doi.org/10.1016/j.eswa.
2016.06.021
 Optimization of Central Pattern
Generator-Based Locomotion Controller
for Fish Robot Using Deep Deterministic Policy
Gradient

Quoc Tuan Vu1 , Minh Huy Pham1 , Van Dong Nguyen1 , Van Tu Duong1,2,3 ,
Huy Hung Nguyen1,4 , and Tan Tien Nguyen1,2,3(B)
1 National Key Laboratory of Digital Control and System Engineering (DCSELab), Ho Chi
Minh City University of Technology (HCMUT), Ho Chi Minh City, Vietnam
nttien@hcmut.edu.vn
2 Ho Chi Minh City University of Technology (HCMUT), Ho Chi Minh City, Vietnam
3 Vietnam National University Ho Chi Minh City, Ho Chi Minh City, Vietnam
4 Saigon University, Ho Chi Minh City, Vietnam

Abstract. This paper presents the optimization of a CPG-based locomotion con-

troller for a fish robot using Deep Deterministic Policy Gradient (DDPG). Firstly,
the rhythm of swimming of an elongated undulating fin-like black Knife fish is
generated by Central Pattern Generator (CPG). In the CPG network, the Hopf
oscillators are employed to provide the rhythmical output and ensure continuous
sinusoidal oscillation even when the CPG parameters are abruptly changed. The
smooth transition output of the CPG is dependent on an intrinsic parameter of the
oscillator called the convergence speed. This parameter is optimized by a combi-
nation of Deep Q-Network (DQN) and Policy Gradient (PG), which overcomes
the drawback of traditional DQN, such as providing stable learnings to adapt
specifically to dynamic environments. The simulation results demonstrate that the
convergence speed of the modified CPG network based on DDPG is improved by
about 2.2%. It also indicates that the rhythmical output of the CPG integrated with
the DDPG optimizer can provide higher accuracy of oscillatory amplitude (about
1,6%) than do the traditional DQN, leading to high efficiency in controlling the
swimming gait of the robotic fish.

Keywords: Central pattern generator · Deep deterministic policy gradient

1 Introduction
In nature, the swimming gait of fish is divided into two main categories, body and/or
caudal fin locomotion (BCF) and median and/or pair fin locomotion (MPF) [1]. An
estimated 15% of fishes do not use the BCF as primary propulsion, while the rest use
BCF for thrust and MPF for navigation and counterbalance [2].
To mimic the fish-like locomotion, the CPG network is used for the generation of the
rhythmic movement of the fish robot [3–5]. So far, a large number of methods have been

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 764–770, 2023.
https://doi.org/10.1007/978-3-031-22200-9_80
 Optimization of Central Pattern Generator-Based Locomotion 765

proposed to simulate CPG behavior, such as: the PID algorithm [6], Matsuoka oscillator
[7], and genetic algorithm (GA) [8]. In these references, the Hopf oscillator is used to
generate the motion of CPG network due to the capability of generating uninterrupted
sinusoidal oscillation in spite of the abrupt change of CPG parameters [9, 10]. To improve
the performance of the modified CPG model, many researchers have introduced different
optimization methods, such as GA [11], particle swarm optimization (PSO) [12], and
reinforcement learning (RL) [13].
In this paper, a deep deterministic policy gradient (DDPG) is used to seek the optimal
parameters of the CPG model for the elongated undulating fin for the first time.

2 Materials and Methods

2.1 Equation of Oscillation
The fish robot comprises 16 adjacent fins and a flat rubber screen to stimulate the swim-
ming gait of a fish robot. Each fin is controlled by an RC servo. The distance between the
fin rays is 32mm, and the fin width is 150mm. The 3D design diagram of the undulating
fin is shown in Fig. 1.

Fig. 1. 3D design diagram of the undulating fin

To generate the rhythm locomotion of the fish robot, the CPG network is widely
used due to its advantages. To generate the motion of CPG, the Hopf model is applied
to allow the continuous sinusoidal oscillation when changing the parameters.
The equation of the Hopf oscillator is given by:
 
u̇ = k A2 − u2 (t) − v2 (t) u(t) − 2π fv(t) (1)
 
v̇ = k A2 − u2 (t) − v2 (t) v(t) + 2π fu(t) (2)

where: u(t), v(t) are the states of the oscillator; A, f is the amplitude and frequency of
oscillators; and k is the convergence speed (k > 0).
766 Q. T. Vu et al.

2.2 Modified CPG Network

The biomimetic robot uses a simple neural network, which is generated from the asso-
ciation of CPGs together, as shown in Fig. 2. The proper coupling configuration of
oscillators is known as a noise vector.

uu11 uu22 uu33 uu...... uu15

15 uu16
16

vv11 vv22 vv33 vv...... vv15

15 vv16
16

CPG1 CPG2 CPG3 CPG... CPG15 CPG16

Fig. 2. Structure of the CPG network with bidirectional couplings

By introducing a noise vector, the Hopf equation is adjusted to fit the CPG as follows:
    2    
u̇i k A − u2 (t) − v2 (t)u(t) − 2π fv(t) pu,i
=  2 + (3)
v˙i k A − u2 (t) − v2 (t) v(t) + 2π fu(t) pv,i

The amplitude of oscillation is chosen: A = 1, then the output θ is determined as

follows:

θi = Gi ui (4)

where: Gi is the oscillator amplitude of fin ray i.

It can be noted that the oscillatory frequency f only affects the obliquity speed of fin
rays, and hence it is essential to optimize the convergence speed k in order to seek the
best swimming pattern.

3 Developed DDPG to Optimize CPG Model

3.1 DDPG
DDPG is an improved version of deep Q-network (DQN) algorithms proposed by the
reference [14] in 2015. It is combined by DNQ and PG to hand the low-dimensional
action spaces.
The DDPG network consists of two main structures, the Actor-network and the Critic
network. Assume the values st , at , rt , st+1 , which are the state at the time t, the action
taken, and the reward of performing action at at the statest . In addition, there are two
sub-networks, Target Actor and Target Critic, to guide Actor and Critic when using data
from the memory buffer.
 Optimization of Central Pattern Generator-Based Locomotion 767

We call the output of the Actor-network is μ(s|θ μ ), Critic network is Q(s, a|θ Q ),
 
Target Actor-network is μ (s|θ μ ) and Target Critic network is Q (s, a|θ Q ), where θ is
the corresponding weight of each network.
The weights of the network are trained by minimizing the loss value of critic Lcritic
as follows:
1   2
Lcritic = (yi − Q si , ai |θ Q ) (5)
N
i


  
where: yi = ri + γ Q si+1 , μ (si+1 |Qμ )|θ Q ); γ is the discount factor, 0 < γ < 1;
ri is the reward obtained when performing an action ai in state si ; N is the total training
episodes.
The Actor-network is updated according to the Policy Gradient equation as follows:
1     
∇θ μ J ≈ ∇a Q s, a|θ Q | ∇θ μ μ s|θ μ |s (6)
N s=si ,a=μ(si ) i
i

The Target Actor-network and Target Critic network are updated based on the weights
of the Actor and Critic network:
 
θ μ ← τ θ μ + (1 − τ )θ μ (7)

 
θ Q ← τ θ Q + (1 − τ )θ Q (8)

where: τ  1

3.2 Optimize the Coefficient k

As mentioned in Subsect. 2.2, the CPG output depends on the value of convergence
speed k. Thus, it is essential to choose an appropriate convergence speed so that the last
fin ray can move in the shortest time with the smallest amplitude error. The optimization
diagram of parameter k is shown in Fig. 3.

+ -
Timeold Timenew
Neural Network
k CPG_Calculation
Errorold Errornew
+ -
Fig. 3. Algorithm diagram of DDPG optimize k

The reward function is employed by trading-off between the transient-state time and
the amplitude error. Mathematically, the reward function is given as follows:

Reward = timeold − timenew + (errorold − errornew ) ∗ 10 (9)

768 Q. T. Vu et al.

In Eq. (9), the coefficient of 10 demonstrates that the amplitude error is a more
significant minimum than that of the transient-state time.
The steps of DDPG are expressed as follows:

Step 1 Initialize st = [00]; ep = 0: num_episodes.

Step 2 k = at = μ(st |θ μ )
Step 3 Calculate CPG.
Step 4 Time = thetimewhenlastCPGstartsmove
Step 5 Error = |1 − max(u)|
Step 6 Timeold = state[0]; Errorold = state[1]
Step 7 Calculate reward: rt = timeold − timenew + (errorold − errornew ) ∗ 10
Step 8 Update state: st = [TimeError]
Step 9 Store (st , at , rt , st+1 ) in buffer memory R.
 2
Step 10 Update Critic: Lcritic = N1 i (yi − Q si , ai |θ Q ) , where:yi = ri +

   
γ Q si+1 , μ (si+1 |Qμ )|θ Q ).
 
Step 11 Update Actor: ∇θ μ J ≈ N1 i ∇a Q s, a|θ Q |s=s ,a=μ(s ) ∇θ μ μ(s|θ μ )|si
i i
   
Step 12 Update Target network θ μ ← τ θ μ + (1 − τ )θ μ ; θ Q ← τ θ Q + (1 − τ )θ Q

4 Results and Discussions

In this paper, the proposed DDPG based CPG is performed on an elongated undulating
fin model to optimize the convergence speed k according to the minimum transient-state
time and the oscillator error. The parameters of Hopf-CPG model are chosen as A = 1,
f = 1, to allow the fin-rays to mimic the swimming gait of the black knife fish. After
the number of training episodes, the rewards will be accumulated together, called the
cumulative rewards. The graph of the cumulative reward value obtained is given in Fig. 4.
It can be observed from Fig. 4 that before reaching 500 training episodes, the rewards
were very low and fluctuated a lot because the model was being updated to find the
right weights. After that, the cumulative rewards increased steadily and then gradually
converged to the steady-state.
After DDPG algorithm is applied to optimize the CPG parameters, the biggest reward
is achieved, corresponding to the value of k of 87.86. By using the newly found k value,
we can determine the transient-state time is 0.61 s, and the maximum amplitude error of
one CPG is 0.00103 rad. Figure 5 illustrates the output of CPG in the case of k = 87.8
and f = 1.
To demonstrate the efficiency of the proposed method, a comparison of DDPG with
DNQ introduced in [13] is performed in Table 1.
From Table 1, we can clearly see that DDPG is better than DNQ. In addition, the
accumulative rewards and the amplitude error of DDPG are gradually reduced for its
training episodes.
Furthermore, using Based Reinforcement Learning (Actor/Critic, SAC) or other
upgrades (DPG, TRPO, PPO), it is impossible for the robotic fish model to train the
parameter of k converging accurately to the optimal value because these methods are
only applied for a finite number of actions. Due to an indefinite increase in the number
 Optimization of Central Pattern Generator-Based Locomotion 769

Fig. 4 Accumulated reward after training episodes optimize k

Fig. 5. CPGs output with k = 87.86175 and f = 1

Table 1. Comparison of DDPG with DNQ

Method Evaluation standard

The value of k The transient-state time (s) The amplitude error (%)
DDPG 87.86 0.61 3.92
DNQ 86 1.41 5.52

of action values when the value of k changes in the range of [1; 150], DDPG is the
most promising tool to find the optimal value of k corresponding to the maximum total
accumulative rewards.
770 Q. T. Vu et al.

5 Conclusion
This paper proposed a novel elongated undulating fin and adopted DDPG to optimize the
controlling parameters of its swimming posture. The CPG network, including sixteen
coupled Hopf oscillators for gait generation, was analyzed first, and then the principle
of DDPG was proposed to enhance the performance of the modified CPG.
The simulation results show that the accuracy of the oscillatory amplitude of the
CPG model based on DDPG is improved by about 1.6%, compared with the convenient
DQN method. Furthermore, there is a significant increase in the convergence speed after
using DDPG (approximately 2.2%) for optimizing the parameters of CPG.

Acknowledgment. This research is funded by Vietnam National University Ho Chi Minh City
(VNU-HCM) under grant number TX2022-20b-01. We acknowledge the support of time and facil-
ities from the National Key Laboratory of Digital Control and System Engineering (DCSELab),
Ho Chi Minh City University of Technology (HCMUT), VNU-HCM for this study.

References
1. Sfakiotakis, et al.: Review of fish swimming modes for aquatic locomotion. IEEE J. Ocean.
Eng. 24(2), 237–252 (2016)
2. Wang, R. et al.: Development and motion control of biomimetic underwater robots: a survey.
IEEE Trans. Syst. Man Cyber. Syst. 52(2), 833–844 (2022)
3. Yu, J., et al.: Amphibious pattern design of a robotic fish with wheel-propeller-fin mechanisms.
J. Field Robot 30(5), 702–716 (2013)
4. Bliss, T. et al.: Central pattern generator control of a tensegrity swimmer. IEEE/ASME Trans.
Mechatron. 18(2), 586–597 (2013)
5. Crespi et al.: Controlling swimming and crawling in a fish robot using a central pattern
generator. Auto. Robots 25(1–2), 3–13 (2008)
6. Zhao et al.: Development of multi-mode biomimetic robotic fish based on central pattern
generator. IEEE/RSJ Int. Conf. Intell. Robot. Syst. (2006)
7. Dong, X., et al.: CPG based motion control for an underwater thruster with undulating long-fin.
IFAC Proc. 41(2), 5433–5438 (2008)
8. Zhou, C., Low, K.H.: On-line optimization of biomimetic undulatory swimming by an
experiment-based approach. J. Bionic Eng. 11(2), 213–225 (2014). https://doi.org/10.1016/
S1672-6529(14)60042-1
9. Hu, Y., et al.: Parameter synthesis of coupled nonlinear oscillators for CPG-based robotic
locomotion. IEEE Trans. Ind. Electron 61(11), 6183–6191 (2014)
10. Dong, N.V. et al.: Force optimization of elongated undulating fin robot using improved PSO
based CPG. Comput. Intell. Neurosci. 1–11 (2022)
11. Jose, H.B.Z.: Two-phase GA parameter tunning method of CPGs for quadruped gaits. In: The
2011 International Joint Conference on Neural Networks (2011)
12. Wang, M., et al.: Control and optimization of a bionic robotic fish though a combination of
CPG and PSO. Neurocomputing 337, 144–152 (2019)
13. Dong, N.V., et al.: Reinforcement learning-based optimization of locomotion controller using
multiple coupled CPG oscillators for elongated undulating fin propulsion Mather. Biosci.
Engineer. 19(1), 738–758 (2021)
14. Mnih, V. et al.: Human-level control through deep reinforcement learning, pp 529–533 (2015)
 Optimization of Main Design Parameters
for a Two-Stage Helical Gearbox Based
on Gearbox Volume Function

Tran Huu Danh1 , Trieu Quy Huy2 , Bui Thanh Danh3 , Nguyen Van Trang4 ,
and Le Xuan Hung4(B)
1 Vinh Long University of Technology Education, Vinh Long, Vietnam
2 University of Economics - Technology for Industries, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

lexuanhung@tnut.edu.vn

Abstract. This paper describes a study that was carried out to determine the
optimum main design parameters for minimizing the gearbox volume of a two-
stage helical gearbox. The gear ratio of the first stage, the coefficient of wheel
face width of stages 1 and 2, and the allowable contact stress of stages 1 and 2
were all considered carefully in this work to determine their optimum values. A
simulation experiment was designed and implemented by a computer program to
solve this work. Minitab R19 software was also used to design the experiment and
analyze the results. The effect of main design factors on the gearbox volume was
analyzed. The optimum value for these parameters was also proposed.

Keywords: Helical gearbox · Main design parameters · Optimum gearbox

design · Gearbox volume

1 Introduction

A gearbox is a mechanical device that reduces the torque and speed transmitted from
the motor shaft to the working machine shaft. Because of its simple structure, stable
operation, reliability, and low cost in comparison to electric, hydraulic, or pneumatic
reduction systems, it is widely used in practice. As either a result, optimal gearbox
design is a significant concern.
Until now, numerous studies on the optimal design of gearboxes have been conducted.
The optimal design of a tillage machine gearbox was reported in [1]. In this study, the
minimum center diameter was chosen as the objective function. From the results it was
noted that the center distance of the gearbox decreased by about 10%. The authors in [2]
uses MATLAB/Simulink to model the dynamic loads experienced by main mechanical
components within the wind turbine gearbox under various operational conditions. In
[3] presented multi-objective optimization of a 2-stage spur gearbox using NSGA-II and
decision-making methods. Three objectives are investigated concurrently in this work:

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 771–779, 2023.
https://doi.org/10.1007/978-3-031-22200-9_81
772 T. H. Danh et al.

volume, power output, and center-distance. The module, pinion tooth number, and face-
width variables were found to have a greater effect on volume than power output and
centre-distance. The selection of the wheel face width coefficient of multi-stage helical
gearbox was shown in [4]. The coefficient values, however, are not optimal; they are
only chosen reasonably based on some advice. The authors of [5] conducted research
into the ability to improve energy efficiency and reducing lifecycle costs of electric
city buses with multi-stage gearboxes. Using an empirical model approach, [6] explores
the optimization of gearbox geometric design parameters to minimize rattle noise in an
automotive transmission. The determination of optimum gear ratios was found for helical
gearboxes with two [7, 8], three [9, 10] or four [11] stages. Despite numerous studies
on optimizing gearbox parameters, no research has been conducted to simultaneously
identify the optimal main design parameters.
A study to determine the optimum main design parameters for a two-stage helical
gearbox was presented in this paper. The objective function of the optimization problem
is to find the smallest gearbox volume. A simulation experiment was carried out using
the Taguchi method and the Minitab R19 software. The effect of the primary design
parameters on the objective function was analyzed. The best values for the main design
parameters have been proposed.

2 Methodology

2.1 Determination of Gearbox Volume

The gearbox volume of a two-stage helical gearbox Vgb is determined by (Fig. 1):

Vgb = L · B1 · H (1)

In which, L, B1 , and H are determined by:

dw11 dw22
L= + aw1 + aw2 + + 20 (2)
2 2

B1 = bw1 + bw2 + 4 · SG (3)

H = max(dw21 ; dw22 ) + 8.5 · SG (4)

SG = 0.005 · L + 4.5 (5)

In [12] above Equations, bw1 , and bw2 are the gear width of stages 1 and 2; dw11 ,
dw21 , dw12 , and dw22 are the drive and driven diameters of stage 1 and stage 2 which can
be found by [4]:

dw11 = 2 · aw1 /(u1 + 1) (6)

dw21 = 2 · aw1 · u1 /(u1 + 1) (7)

 Optimization of Main Design Parameters for a Two-Stage 773

Fig. 1. Schema for determination of gearbox volume

dw12 = 2 · aw2 /(u2 + 1) (8)

dw22 = 2 · aw2 · u2 /(u2 + 1) (9)

bw1 = Xba1 · aw1 (10)

bw2 = Xba2 · aw2 (11)

In Eqs. (6) to (11), u1 and u2 are the gear ratios of stages 1 and 2; u2 = ug /u1 ; with
ug is the gearbox ratio; aw1 and aw2 are the center distances of stages 1 and 2 which are
detemined by [4]:

aw1 = ka · (u1 + 1) · 3 T11 · kH β /(AS21 · u1 · Xba ) (12)

aw2 = ka · (u2 + 1) · 3
T12 · kH β2 /(AS22 · u2 · Xba2 ) (13)
774 T. H. Danh et al.

where, kH β = 1.05 ÷ 1.27 is the contacting load ratio for pitting resistance [4]; it was
chosen as kH β = 1.16; AS1 and AS2 is the allowable contact stress of stages 1 and 2
(Mpa); ka = 43 is the material coefficient [4]; Xba1 and Xba2 are the wheel face width
coefficients of stages 1 and 2; T11 and T12 are the torques on the pinions of the first and
the second stages (Nmm):
 
T11 = Tout / ug · η2hg · ηb3 (14)
 
T12 = Tout / u2 · ηhg · ηbe
2
(15)

wherein, Tout is the system output torque (N.mm); ηhg = 0.95 ÷ 0.97 is the efficiency
of a helical gear set [4]; ηb = 0.99 ÷ 0.995 is the efficiency of a rolling bearing pair [4].

2.2 Optimization Problem

The optimization problem can be defined as follows based on the preceding analysis:

Minimize Vgb (16)

With

Vgb = f (u1 ; Xba1 ; Xba2 ; AS 1 ; AS 2 ) (17)

And with the following constraints:

1 ≤ u1 ≤ 9; 1 ≤ u2 ≤ 9 (18)

3 Experimental Work

A simulation experiment was performed to investigate the effect of main design param-
eters on gearbox volume. In addition, the Minitab R19 software and the Taguchi method
were used for experimental design and data analysis. Moreover, five major design factors
were investigated: u1 , Xba1 , Xba2 , AS1 , and AS2 . These parameters and their levels are
described in Table 1.
To make programming easier, the effect of these parameters on gearbox volume was
investigated for each gearbox ratio value: 5, 10, 15, 20, and 25. Because the 5-level
Taguchi design type (L25) was chosen, 25 test runs for the simulation experiment were
performed for each above value of the gearbox ratio. When the gearbox ratio is 5, Table 2
shows the experimental matrix and the output results (the gearbox volume).
 Optimization of Main Design Parameters for a Two-Stage 775

Table 1. Main design parameters and their levels

Factor Level
1 2 3 4 5
Total gearbox ratio u1 1 3 5 7 9
Coefficient of wheel face width of stage 1 Xba1 0.25 0.27 0.29 0.31 0.33
Coefficient of wheel face width of stage 2 Xba2 0.32 0.34 0.36 0.38 0.4
Allowable contact stress of stage 1 AS1 (Mpa) 350 368 386 404 420
Allowable contact stress of stage 2 AS2 (Mpa) 350 368 386 404 420

Table 2. Experimental matrix and output results (Vgb ) when ugb = 5

Trial u1 Xba1 Xba2 AS1 AS2 Vgb (mm3 )

1 1.78 0.25 0.32 350 350 18339935.91
2 1.78 0.27 0.34 368 368 16613972.96
3 1.78 0.29 0.36 386 386 15133766.38
4 1.78 0.31 0.38 404 404 13853821.22
5 1.78 0.33 0.4 420 420 12845229.45
6 1.89 0.25 0.34 386 404 15387731.39
…
24 2.22 0.31 0.36 368 350 17788967.53
25 2.22 0.33 0.38 386 368 16211694.93

4 Results Discussion
The Analysis of Variance (ANOVA) method and the Minitab R19 software were used
to evaluate the influence of the main design parameters on gearbox volume Vgb when
ugb = 5. To determine the effect of each parameter on the output results, the signal-to-
noise ratio, or S/N number, must be calculated for each experiment. The calculated S/N
number will determine which parameters are best and which have the most influence
on the outcome. To minimize performance characteristics, use the following S/N ratio
equation:
 n 
1 2
S/N = −10log10 yi (19)
n
i=1

The average S/N ratio for each parameter and level is calculated after calculating
the S/N ratio for each experiment. Table 3 and Fig. 2 show how the main design factors
affect Vgb .
From Fig. 2 it is noted that the impact of the main design factors on Vgb is as follows:
Vgb is proportional to u1 and Xba2 and inversely proportional to Xba1 , AS1 and AS2 .
776 T. H. Danh et al.

Besides, Table 3 shows that AS1 has the greatest influence on Vgb (74.21%); followed
by AS2 (14.99%). u1 and Xba1 have little effect on Vgb (5.42% and 4.47%). Also, Xba2
has almost no effect on Vgb . The order of the main parameters’ influence on gearbox
volume is shown in Table 4.
The S/N value is maximized for each main design parameter to obtain the smallest
gearbox volume using the Eq. (19). From the analysis of the effect of main design
parameters on the S/N ratios in the chart in Fig. 3, the optimal main design parameters
were found as follows: u1 = 1.78; Xba1 = 0.33; Xba2 = 0.32; AS1 = 420; AS2 = 420.

Table 3. Analysis of variance for means

Source DF Seq SS Adj SS Adj MS F P C(%)

u1 4 2.83E + 12 2.83E + 12 7.06E + 11 247.7 0 5.48
Xba1 4 2.31E + 12 2.31E + 12 5.77E + 11 202.25 0 4.47
Xba2 4 4.24E + 11 4.24E + 11 1.06E + 11 37.16 0 0.82
AS1 4 3.83E + 13 3.83E + 13 9.57E + 12 3354.51 0.002 74.21
AS2 4 7.73E + 12 7.73E + 12 1.93E + 12 677.65 0 14.99
Residual error 4 1.14E + 10 1.141E + 10 2.852E + 09 0 0.02
Total 24 5.16E + 13
Model summary
S R-Sq R-Sq(adj)
53406.0534 99.98% 99.87%

Fig. 2. Effect of main design factors on Vgb

 Optimization of Main Design Parameters for a Two-Stage 777

Table 4. The order of influence of main design factors on gearbox volume

Level u1 Xba1 Xba2 AS1 AS2

1 15357345 16263050 15641291 17668034 16661129
2 15551981 16027540 15708156 16629637 16159026
3 15803701 15758985 15807732 15708206 15771687
4 16052822 15607269 15897367 14865468 15367641
5 16294502 15403507 16005805 14189005 15100869
Delta 937157 859543 364514 3479029 1560261
Rank 3 4 5 1 2

Fig. 3. Main effects plot for S/N ratios

Table 5. Optimum values of main design factors

No ut
5 10 15 20 25
u1 1.78 3.28 4.78 6.28 7.58
Xba1 0.33 0.33 0.33 0.33 0.33
Xba2 0.32 0.32 0.32 0.32 0.32
AS1 420 420 420 420 420
AS2 420 420 420 420 420
778 T. H. Danh et al.

Fig. 4. Gear ratio of the first stage versus gearbox ratio

For other values of ugb including 10, 15, 20, and 25, proceed in the same way as
above. Table 5 shows the optimal values of main design parameters. From Table 5, the
following observations were reported:

– The optimum values of Xba1 are their possible maximum values. In contrast to Xba1 ,
the optimal values of Xba2 are their minimum values. The reason can be explained
as follows: Xba1 needs to take a largest possible value and Xba2 needs to take the
smallest possible value to reduce the difference of drive diameter dw21 and dw22 to
reduce height H and leads to a decrease in Vgb .
– The optimal values of AS1 and AS2 take their maximum values. The reason is the
same as above, to have the minimum gearbox volume, the AS1 and AS2 values need
to have the maximum values to reduce the axial distance of the gear set, leading to a
reduction in the size L (formula 2) and Vgb ((formula 1).
– u1 and ug have a first-order relationship (Fig. 4). Also, from the data, the following
regression model (with R2 = 0.9992) was proposed to determine the optimum values
of u1 :

u1 = 0.292 · ugb + 0.36 (20)

After having u1 , the optimum values of u2 is determined by u2 = ugb /u1 .

5 Conclusions
This paper presents the findings of a study on optimizing a two-stage helical gearbox to
achieve the smallest gearbox volume. To solve this problem, five main design parameters
of the gearbox that were optimized in this work were the gear ratio of the first stage,
the coefficient of wheel face width of stages 1 and 2, and the allowable contact stress
of stages 1 and 2. Also, a simulation experiment with Taguchi L25 type was designed
and carried out. The effect of key design factors on gearbox volume was also studied.
Furthermore, the following are the optimum values of the main gearbox parameters and
 Optimization of Main Design Parameters for a Two-Stage 779

a regression model to find the optimum gear ratios of the first stage: Xba1 = 0.33; Xba2
= 0.32; AS1 = 420 (MPa); AS2 = 420 (MPa); u1 is calculated by Eq. (20).

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

References
1. He, H.-B., et al.: Optimal design of a tilling machine reduction gearbox using Matlab. Int. J.
Precis. Eng. Manuf. 10(2), 63–66 (2009)
2. Al-Hamadani, H., An, T., King, M., Long, H.: System dynamic modelling of three different
wind turbine gearbox designs under transient loading conditions. Int. J. Precis. Eng. Manuf.
18(11), 1659–1668 (2017). https://doi.org/10.1007/s12541-017-0194-1
3. Maputi, E.S., Arora, R.: Multi-objective optimization of a 2-stage spur gearbox using NSGA-
II and decision-making methods. J. Braz. Soc. Mech. Sci. Eng. 42(9), 1–22 (2020). https://
doi.org/10.1007/s40430-020-02557-2
4. Chat, T., Van Uyen, L.: Design and Calculation of Mechanical Transmissions Systems, vol.
1. Educational Republishing House, Hanoi (2007)
5. Ritari, A., et al.: Energy consumption and lifecycle cost analysis of electric city buses with
multispeed gearboxes. Energies 13(8), 2117 (2020)
6. Bozca, M., Fietkau, P.: Empirical model based optimization of gearbox geometric design
parameters to reduce rattle noise in an automotive transmission. Mech. Mach. Theory 45(11),
1599–1612 (2010)
7. Pi, V.N., Tuan, N.K., Hung, L.X., Tung, L.A.: A new study on determination of optimum
gear ratios of a two-stage helical gearbox. In: Awang, M., Emamian, S.S., Yusof, F. (eds.)
Advances in Material Sciences and Engineering. LNME, pp. 107–115. Springer, Singapore
(2020). https://doi.org/10.1007/978-981-13-8297-0_14
8. Cam, N.T.H., Pi, V.N., Hong, T.T.: A study on calculation of optimum gear ratios of a two-stage
helical gearbox with second stage double gear sets. a a 500, 1 (2019)
9. Vu, N.-P., et al.: The influence of main design parameters on the overall cost of a gearbox.
Appl. Sci. 10(7), 2365 (2020)
10. Pi, V.N., et al.: Determining optimum partial transmission ratios of mechanical driven systems
using a V-Belt drive and a three-stage helical reducer. In: Advances in Material Sciences and
Engineering, pp. 81–88. Springer (2020)
11. Pi, V.N.: Optimal calculation of partial transmission ratios for four-step helical gearboxes with
first and third-step double gear-sets for minimal gearbox length. In: American Conference on
Applied Mathematics (MATH 2008), Harvard, Massachusetts, USA (2008)
12. Römhild, I., Linke, H.: Gezielte Auslegung Von Zahnradgetrieben mit minimaler Masse auf
der Basis neuer Berechnungsverfahren. Konstruktion (1981) 44(7–8), 229–236 (1992)
 Pavement-Friendly Performance Analysis
of an Agricultural Truck with Two Types
of Hydro-Pneumatic Suspension Struts

Ta Tuan Hung1 , Le Xuan Long2(B) , Nguyen Van Tuan2 , Hoang Anh Tan2 ,
and Vu The Truyen1
1 Faculty of Mechanical Engineering, University of Transport Technology, Hanoi, Vietnam
2 Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
lexuanlong@tnut.edu.vn

Abstract. The purpose of this study is to analyze the pavement-friendly per-

formance of an agricultural truck with two types of hydro-pneumatic suspension
(HPS) struts on dynamic load coefficient (DLC). The nonlinear dynamical models
of two different types of HPS struts with one gas chamber and two oil chambers
(Model 1) and with one gas chamber and three oil chambers (Model 2) are set up
to calculate their nonlinear vertical forces. A quarter–vehicle mathematical model
of an agricultural truck is set up for the analysis of the nonlinear vertical forces of
two proposed suspensions which is implemented in MATLAB/Simulink platform.
The obtained results indicated that the pavement-friendly performance of Model
2 is better than performance of Model 1. Especially, the values of DLC at axle of
vehicle with Model 2 reduce by 6.97% in comparison with Model 1 when vehicle
moves on ISO class C road surface at vehicle speed of 40 km/h and full load.

Keywords: Agricultural truck · Hydro-pneumatic suspension · Dynamic tire

load · DLC

1 Introduction
Hydro-pneumatic suspension (HPS) system is applied widely in engineering fields
because the gas absorbs excessive force while the oil is nonlinear elasticity and good
vibration damping perforation. A multi body co-simulation approach was proposed to
investigate the effects of HPS parameters on the ride safety [1]. A semi-active HPS based
on the electro-hydraulic proportional valve control was proposed using the control strat-
egy to adjust the damping force of the semi-active hydro-pneumatic suspension [2]. A
HPS model based on fractional order was built with the advantage of fractional order
in viscoelastic material modeling considering the mechanics property of multiphase
medium of hydro-pneumatic suspension system [3]. The ride comfort of vehicle using a
HPS applied with the semi-active control was analyzed and compared with a passive cab
suspension [4]. A hydro-pneumatic inerter-based suspension system theoretical model
was developed to analyze its performance which was controlled to improve vehicle ride

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 780–789, 2023.
https://doi.org/10.1007/978-3-031-22200-9_82
 Pavement-Friendly Performance Analysis of an Agricultural 781

comfort based on the mechanical network theory of inerter and semi-active control [5].
The performance of the hydro-pneumatic suspension system of heavy truck on the ride
quality of road surfaces was proposed to compare to the rubber and air spring suspen-
sion systems based on a full dynamic model of a heavy vehicle [6]. The performance of
the hydro-pneumatic suspension system of heavy truck on vehicle ride comfort was pro-
posed and compared to those of rubber, leaf and air springs of the suspension systems [7,
8]. A hydro-pneumatic suspension strut concept with integrated two gas chambers was
proposed to realize nearly symmetric stiffness properties in compression and rebound,
and progressively hardening roll stiffness characteristics [9]. The nonlinear stiffness and
damping properties of a simple and low cost design of a hydro-pneumatic suspension
(HPS) strut that permits entrapment of gas into the hydraulic fluid were analyzed by
both experimental and numerical methods [10]. The semi-active hydro-gas suspension
was proposed for a tracked vehicle to improve ride comfort performance, without com-
promising the road holding and load carrying capabilities of the passive suspension
[11]. The HPS optimal parameters of a multi-axle heavy motorized wheel dump truck
were found out based on virtual and real prototype experiment integrated Kriging model
[12]. A dual-chamber HPS was proposed and analyzed the effect of various suspension
parameters on the storage stiffness and damping coefficient [13]. This proposal study
is to analyze the performance of an agricultural truck with two types of HPS struts in
terms of the negative impact on road surface. The nonlinear models of two types of
hydro-pneumatic suspension (HPS) struts which consist of two oil chambers and one
gas chamber (Model 1) and three oil chambers and one gas chamber (Model 2) are set up
to determine their nonlinear vertical forces and then it is connect with a quarter–vehicle
mathematical model of an agricultural truck to analyze and evaluate their effectiveness
on the dynamic load coefficient (DLC).

2 Mathematical Model of HPS Strut

Structural schematic of two types of HPS struts with two oil chambers and one gas
chamber (Model 1) and three oil chambers and one gas chamber (Model 2) is show
in Figs. 1 and 2 which consist of the main oil chamber (1), oil chamber (2), ring oil
chamber (3), the gas chamber (4), the oil pressures of the oil chambers (1), (2), (3) and
gas chamber (4) p1 , p2 , p3 and pa , the effective areas of cylinder, rod and floating piston
and gas chamber, A1 , A2 , and A3 , the area of the orifices between oil chamber (1) and
chamber (2) and the area of the orifices between oil chamber (1) and chamber (3) A12
and A13 , the vertical displacements of vehicle axle, floating piston, and vehicle body za ,
zd and zb . Nonlinear dynamic model of two types of HPS struts is shown in Fig. 3 which
consist of the stiffness and damping coefficients of Model 1 kh and ch13 , the stiffness
coefficient and two damping coefficients of Model 2 kh , c12 and c13 , vehicle axle and
vehicle body masses ma and mb and the vertical forces of Model 1 and Model 2 Fht and
Fhn .
From Fig. 3a, the vertical dynamic force of Model 1 is determined by
Fht = Fk + Fc (1)
where, Fk is the elastic force of the gas chamber and Fc is the damping force of the oil
chambers.
782 T. T. Hung et al.

The elastic force of the gas chamber of Model 1 is determined by

Fk = (pa − p0 )A3 (2)

The pressure in the gas chamber is defined as an adiabatic process following the laws
of thermodynamics

p0 V0n = pa Van (3)

where p0 and V0 are the initial absolute pressure and volume in the gas chamber, pa and
Va are the absolute pressure and volume in the gas chamber, n is the polytrophic rate (1
< n < 1.4).
 n
V0
pa = p0 , Va = V0 + A3 (zb − za ) (4)
Va

Combination of Eqs. (4) and (2), the elastic force of the gas chamber of Model 1 is
achieved by
 
V0
Fk = p0 − 1 A3 (5)
V0 + A3 (zb − za )

The damping force of Model 1 is determined by

 
md g
Fc = p1 (A1 − A3 ) − p3 (A1 − A2 ) − p0 − (A2 − A3 ) (6)
A3

where A1 , A2 , and A3 are the area of cylinder, rod and floating piston, p1 and p3 are the
pressures in the oil chamber (1) and chamber (3).
The flow rate through the orifice is determined by:

(p3 − p1 )
Q = Cd A13 2 sign(żb − ża ) (7)
ρ

where, Cd is coefficient of discharge, A13 is the area of the orifices and ρ is the density
of oil.
On the basic of volume balance, the flow rate is inferred by

Q = (A1 − A2 )(żb − ża ) (8)

Combination of Eqs. (7) and (8), the relationship p1 and p3 is achieved by

 
ρ(A1 − A2 )2 żb − ża (żb − ża )
p3 = p1 + (9)
2Cd2 A213

The differential motion equation of floating piston can be determined by

(p1 − pa )A3 + md g = md z̈d (10)

 Pavement-Friendly Performance Analysis of an Agricultural 783

Fig. 1. Model 1

The Eq. (10) can be rewritten by

md (z̈d − g)
p1 = pa + (11)
A3
Combination of Eqs. (11), (9) and (6), the damping force of Model 1 is achieved by
   n 
ρ(A1 − A2 )3 żb − ża (żb − ża ) V0
Fc = − + p0 (A2 − A3 ) −1
2Cd2 A213 V0 + A3 (zb − za )
md
− (z̈d A3 − gA2 )
A3
(12)
From Fig. 3b, the vertical dynamic force of Model 2 can be determined by

Fhn = Fk + Fcn (13)

where, Fk is the elastic force of the gas chamber, Fcn is the damping force of the oil
chambers and it is achieved by
    
ρ żb − ża (żb − ża ) A1 (A1 − A2 )2 A23 (A1 − A2 )2
Fcn = − + A2 2 −
2Cd2 A213 A12 A213
 n (14)
V0 md
+p0 (A3 − A2 ) −1 − (z̈d A3 − gA2 )
V0 + A3 (zb − za ) A3
784 T. T. Hung et al.

Fig. 2. Model 2

Fht Vehicle body Fhn Vehicle body

zb zb
mb mb

kh ch13 kh ch12 ch13

za za
ma ma
Fht Vehicle axle Fhn Vehicle axle
(a) Model 1 (b) Model 2
Fig. 3. Nonlinear dynamic model of two types of HPS struts

3 Quarter–vehicle Dynamic Model of a Mining Dump Truck

The schematic diagram of 2-DOFs quarter–vehicle dynamic model of an agricultural

truck with two models of HPS struts is shown in Fig. 4 which consists of the tire stiffness
and damping coefficients kt and ct , the random road surface roughness q.
 Pavement-Friendly Performance Analysis of an Agricultural 785

Fig. 4. 2-DOFs quarter–vehicle dynamic model

The equations of vehicle motion of Fig. 4 are written as follows

mb z̈b = −Fhi (15)


ma z̈a = Fhi − kt (q − za ) + ct (q̇ − ża ) (16)

Road surface excitations: The random road surface roughness of random white noise
is used as excitation source waveform for vehicle [18], the random road profile is pro-
duced by filtering the white noise using the following mathematical model of the road
roughness.

q̇(t) + 2πf0 q(t) = 2πn0 Gq (n0 )v(t)w(t) (17)

where, Gq (n0 ) is the road roughness coefficient which is defined for typical road classes
from A to H according to ISO 8068(2016) [17], n0 is a reference spatial frequency which
is equal to 0.1 m, v(t) is the speed of vehicle; f0 is a minimal boundary frequency with a
value of 0.0628 Hz, n0 is a reference spatial frequency which is equal to 0.1 m, w(t) is
a white noise signal.
Dynamic load coefficient (DLC): To analyze the pavement-friendly performance
of two types of hydro-pneumatic suspension systems, the DLC is chosen as objective
function which is defined by a ratio of the root mean square of the vertical dynamic tire
force over static load [14–16] as follows:
FT ,RMS
DLC = (18)
Fs
where, FT,RMS , Fs are the root mean square of the vertical dynamic and the static tire
force.
786 T. T. Hung et al.

4 Results and Discussion

In order to analyze the pavement-friendly performance of two HPS struts (Model 1 and
Model 2), the quarter–vehicle mathematical model of an agricultural truck is solved
in Matlab/Simulink software under the road surface excitations and a set of vehicle
parameters in Table 1. The simulation results of the vertical dynamic tire load at axle
with two types of HPS struts when vehicle moves on the ISO class C at the speed of
40 km/h and full load are shown in Fig. 5. The results of Fig. 5 reflect that the peak
amplitudes of dynamic tire force with Model 2 are smaller than Model 1.

Table 1. Parameters of the vehicle and HPSs.

Parameter Value Parameter Value Parameter Value Parameter Value

ma /kg 145 mb /kg 1000 kt /(N/m) 185500 ct /(Ns/m) 2100
p0 /Mpa 1.5 V0 /m3 1.3 × A1 /m2 2.5 × A2 /m2 1.6 ×
10–4 10–3 10–3
A3 /m2 1.3 × 103 Cd 0.62 A12 /m2 6× A13 /m2 7.5 ×
10–4 10–4

Model 1
Dynamic tire force (N)

Model 2
2000

-2000

0 10 20 30 40 50
Time(s)

Fig. 5. Dynamic tire forces at axle of vehicle with two types

The pavement-friendly performance of two HPS struts of two types of HPS struts
will be analyzed and compared in next sections when vehicle moves under the different
operating conditions.
Road surface condition: In order to evaluate the pavement-friendly performance of
two HPS struts, the road surface conditions from ISO class A road surface to ISO class
E road surface are considered when vehicle moves at vehicle speed of 40 km/h and
full load. The DLC values in comparison with both two types of Models under variable
road surface conditions are shown in Fig. 6. The results of Fig. 6 shows that the DLC
values with both two types of Models increase rapidly when the road surface quality
deteriorates. The DLC values with Model 2 respectively reduce 2.69%, 4.41%, 6.97%,
11.54% and 19.27% in comparison with Model 1 when vehicle moves on from ISO class
A road surface to ISO class E road surface which prove that the negative impact on road
surface with Model 2 has improved significantly in comparison with Model 1.
 Pavement-Friendly Performance Analysis of an Agricultural 787

Vehicle speed conditions: The different speed conditions from vehicle speed of
20 km/h to vehicle speed of 70 km/h are chosen to compare the pavement-friendly
performance of two types when vehicle moves on ISO class D road surface and full
load. The DLC values in comparison with both two types of Models at different speed
conditions are shown in Fig. 7. The results of Fig. 7 shows that the DLC values with
Model 2 are increase with the increase of vehicle speed. The DLC values with Model
2 respectively reduce 5.74%, 5.88%, 6.97%, 7.96%, 8.68% and 8.61% in comparison
with Model 1 with the change of vehicle speed from 20 km/h to 70 km/h. These results
show that the pavement-friendly performance of Model 2 is improved significantly in
the high vehicle speed region.
Vehicle body mass conditions: To analysis the pavement-friendly performance of
two HPS struts when the value of vehicle body mass changes, the mb values from
25% mb to 200% mb are chosen when vehicle moves on ISO class C road surface at
vehicle speed of 40 km/h. The DLC values in comparison with both two types of Models
under different vehicle body mass mb are shown in Fig. 8. The results of Fig. 8 show
that the DLC values of both two types of Models quickly decrease when the mb value
increases. However, mb value increases, it will affect on the durability and safety of
vehicle components. The DLC values with Model 2 respectively reduce 3.11%, 4.86%,
6.78%, 6.97%, 4.23%, 4.18%, 5.75% and 4.8% in comparison with Model 1 with the
change of mb value from 25% mb to 200% mb . The negative impact on road surface with
Model 2 is also significantly improved in comparison with Model 1 with the change of
vehicle body mass.

0.4
Model 1
Model 2
DLC

0.2

0
A B C D E
Different road surface conditions
Fig. 6. DLC values of both two models under different road surface conditions

5 Conclusions
In this study, a quarter–vehicle mathematical model of an agricultural truck was set
up to analyze and evaluate the pavement-friendly performance of two HPS struts. The
major conclusions drawn from the analysis can be summarized as follows: (1) The DLC
values with both two types of Models increase rapidly when the road surface quality
deteriorates. The DLC values with Model 2 respectively reduce 2.69%, 4.41%, 6.97%,
11.54% and 19.27% in comparison with Model 1; (2) The DLC values with Model 2
respectively reduce 5.74%, 5.88%, 6.97%, 7.96%, 8.68% and 8.61% in comparison with
788 T. T. Hung et al.

0.15
Model 1
0.1 Model 2

DLC 0.05

0
20 30 40 50 60 70
Different speed conditions, v (km/h)
Fig. 7. DLC values of both two models at different vehicle speed conditions

1.5
Model 1
1 Model 2
DLC

0.5

0
25 50 75 100 125 150 175 200
Different vehicle body mass, mb(%)

Fig. 8. DLC values of both two models under the different vehicle body mass

Model 1 with the change of vehicle speed from 20 km/h to 70 km/h; and (3) The DLC
values with Model 2 respectively reduce 3.11%, 4.86%, 6.78%, 6.97%, 4.23%, 4.18%,
5.75% and 4.8% in comparison with Model 1 with the change of mb value from 25%
mb to 200% mb .

Acknowledgment. The work described in this paper was supported by Thai Nguyen University
of Technology.

References
1. Han, S., Chao, Z., et al: Research on the effects of hydropneumatic parameters on tracked
vehicle ride safety based on cosimulation. In: Shock and Vibration, pp. 1–10 (2017)
2. Yue, W., Li, S., et al: Investigation of semi-active hydropneumatic suspension for a heavy
vehicle based on electro-hydraulic proportional valve. World J. Eng. Technol. 696–706 (2017)
3. Zhang, J., Chen, S., et al: Research on modeling of hydropneumatic suspension based on
fractional order. Math. Probl. Eng. (2015)
4. Sim, K., et al: Effectiveness evaluation of hydro-pneumatic and semi-active cab suspension
for the improvement of ride comfort of agricultural tractors. J. Terramech. 69, 23–32 (2017)
5. Yin, Y., et al.: Multi-performance analyses and design optimisation of hydro-pneumatic sus-
pension system for an articulated frame-steered vehicle. Int. J. Vehicle Mech. Mob. 57,
108–133 (2019)
6. Long, L.X., et al: Performance analysis of the hydro-pneumatic suspension system of heavy
truck. Int. J. Mech. Eng. Technol. (IJMET) 9, 1128–1139 (2019)
 Pavement-Friendly Performance Analysis of an Agricultural 789

7. Long, L.X., Van Quynh, L., et al: A comparison of ride performance of hydro-pneumatic
suspension system with those of rubber and leaf suspension systems. IOP Conf. Ser.: Mater.
Sci. Eng. 914, 012037 (2020)
8. Long, L.X., Van Quynh, L., et al: Ride performance evaluation of air and hydro-pneumatic
springs of suspension system. IJARET 12(1), 439–447 (2021)
9. Cao, D., et al: Analysis of a twin-gas-chamber hydro -pneumatic vehicle suspension. In:
Advances in Dynamics, Instrumentation and Control (2007)
10. Yin, Y., Rakheja, S., et al: Effects of entrapped gas within the fluid on the stiffness and damping
characteristics of a hydro-pneumatic suspension strut. SAE Int. J. Commer. Veh. 10(1) (2017)
11. Solomon, U., et al.: Semi-active hydro-gas suspension system for a tracked vehicle. J.
Terrramech. 48(4), 225–239 (2011)
12. Gong, B., et al.: Ride comfort optimization of a multi-axle heavy motorized wheel dump
truck based on virtual and real prototype experiment integrated Kriging model. Adv. Mech.
Eng. 7(6), 1–15 (2015)
13. Sang, Z., et al.: Numerical analysis of a dual-chamber hydro-pneumatic suspension using
nonlinear vibration theory and fractional calculus. Adv. Mech. Eng. 9(5), 1–13 (2017)
14. Quynh, L.V., Tuan, N.V., et al: Optimal design parameters of air suspension systems for
semi-trailer truck. Part 1: modeling and algorithm. Vibroengineering PROCEDIA 33, 72–77
(2020)
15. Buhari, R., Md Rohani, M., et al: Dynamic load coefficient of tyre forces from truck axles.
Appl. Mech. Mater. 405(408), 1900–1911 (2013)
16. Quynh, L.V., Tuan, N.V., et al: Optimal design parameters of air suspension systems for
semi-trailer truck. Part 2: results and discussion. Vibroengineering PROCEDIA 33, 147–152
(2020)
17. ISO 8608: Mechanical Vibration-Road Surface Profiles-Reporting of Measured Data.
International Organization for Standardization (2016)
18. Long, L.X., Quynh, L.V., Cuong, B.V.: Study on the influence of bus suspension parameters
on ride comfort. Vibroengineering PROCEDIA 21, 77–82 (2018)
 Performance Analysis of a Mining Dump Truck
Ride Comfort with a Hydro-Pneumatic
Suspension System under Different Operating
Conditions

Dang Viet Ha1 , Le Van Quynh2(B) , and Le Xuan Long2

1 Vietnam Register, Hanoi 12059, Vietnam
2 Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
lequynh@tnut.edu.vn

Abstract. This study proposes a quarter-vehicle dynamic model of a mining dump

truck using hydro-pneumatic suspension system (HPSs) suspension system to
analyze the ride comfort performance of a mining dump truck under different
operating conditions. The root mean square (RMS) acceleration response of vehi-
cle body (awz ) according to international standard ISO 2631-1:1997 is selected
for performance analysis. The numerical simulation results indicate that awz val-
ues increase very quickly under bad operating conditions. The study results are
useful reference for designers and manufacturers in the field of off-road vehicle
suspension systems.

Keywords: Earth-moving equipment · Hydro-pneumatic suspension · Vehicle

dynamic model · Power spectral density · Ride comfort

1 Introduction
Hydro-pneumatic suspension systems are more and more widely used in mining dump
trucks. The performance of vehicle ride comfort as well as the friendliness of road surface
and parts of vehicles is greatly improved by using hydro-pneumatic suspension systems.
An optimal method for an off-road vehicle hydro-pneumatic suspension parameters using
a VRPEI-Kriging model was proposed by Gong et al. [1]. A novel suspension configura-
tion with both hydraulically interconnected suspension (HIS) and electronic controlled
air spring (ECAS) was proposed by Qi et al. [2]. The effects of three characteristics of
the off-road vehicle suspension systems such as the rubber vehicle suspension (RVS),
air vehicle suspension (AVS) and hydro-pneumatic vehicle suspension (HVS) on the
friendly road surface through dynamic load coefficient (DLC) were analyzed and evalu-
ated by Long et al. [3]. An effective analysis of hydro-pneumatic suspension system for
three-axle heavy commercial vehicle on dynamic load coefficient (DLC) was analyzed
by Saglam and Unlusoy [4]. A hydro-pneumatic suspension system (HPSs) was simu-
lated using a quarter vehicle suspension model by Badway et al. [5]. The performance

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 790–797, 2023.
https://doi.org/10.1007/978-3-031-22200-9_83
 Performance Analysis of a Mining Dump Truck Ride 791

of HPS was analyzed by Long et al. [6, 7] to compare with other suspension system
in terms of ride comfort of vehicle. Ride effectiveness of the air suspension systems of
universal truck was compared by Le [8], controlled by Liem et al. [9]. The main pur-
pose of this study is to propose a quarter-vehicle dynamic model of a mining dump truck
using hydro-pneumatic suspension system (HPSs) suspension system to analyze the ride
comfort performance of a mining dump truck under different operating conditions

2 Quarter-vehicle Dynamic Model

A quarter-vehicle dynamic model: A quarter-vehicle dynamic model under random road
surface roughness is established to verify their effectiveness, as shown in Fig. 1. Expla-
nation of Fig. 1, kt and ct are the tire stiffness and damping coefficients, respectively, Fh
is the vertical dynamic force of novel HPS and q is road surface roughness.

Vehicle body zb
mbg
mb
Vehicle body

HPSs Fh

za
mag
ma

Vehicle axle
kt ct
Ft
q

Road surface roughness

Fig. 1. A quarter-vehicle dynamic model

From Fig. 4, the motion equations of vehicle axle mass and vehicle body mass using
Newton’s second law are written as follows
mb z̈b = mb g − Fh (1)
792 D. V. Ha et al.

 
ma z̈a = Fh − kt (q − zb ) + ct (q̇ − żb ) + ma g (2)

Mathematical modeling of novel HPS: The main idea of this study is to add an
auxiliary reservoir which is connected with the air chamber A through surge pipe that
leads to the variable values of both the stiffness and damping coefficients of air chamber
A though the change in air state in the two volumes, as shown in Fig. 2. A nonlinear
dynamic model of the novel HPSs is shown in Fig. 3. It consists of a novel hydro-
pneumatic spring and a novel hydro-pneumatic shock absorber. Explanation of Fig. 3,
khn and kvn is the static and viscous stiffness constants of air chamber A, respectively, cw
is the nonlinear viscous damper of air chamber A, ws is the displacement of air in surge
pipe, M is the air mass of the surge pipe, As is the cross section area of the pipeline, ls
is the length of the surge pipe, Vr is the air volume of the auxiliary reservoir, pr is the
air pressure of the auxiliary reservoir.
From Fig. 3, The vertical dynamic force of novel HPS is defined as

Fh = Fan + Fhn (3)

where, Fan is the total vertical force of air chamber A of the novel HPSs, Fhn is the
resistance force of the traditional HSPs.
The total vertical force of the air chamber A of the novel HPSs is defined as

Fan = khn (zb − za ) + kvn (zb − za − ws ) (4)

The static, viscous stiffness constants of air chamber A khn , kvn and the air mass of
the surge pipe M can be respectively written by
 2
p0 A2p n Vr0 p0 A2p n Vr0 Ap Vr0
khn = , kvn = ken = , and M = As ls ρ ,
V0 + Vr0 V0 V0 + Vr0 V0 As V0 + Vr0
(5)

where, ρ is the air density; V 0 and V r0 are the initial volume of air chamber A and the
auxiliary reservoir, respectively.
The nonlinear viscous damper cw is related to the velocity over the damper and not
related to the velocity in the surge pipe. Therefore, the vertical viscous force is expressed
as

Fvz = kvn (z − ws ) = cw |ẇ|w sign(ẇs ) + M ẅs (6)

The Eq. (6) can be rewritten as

M ẅs = kvn (z − ws ) − cw |ẇs |w sign(ẇs ) (7)

The relationship between the nonlinear damping cw and the damping cs caused by
the flow in the surge pipe is
 1+w
Ap Vr0 1 1
cw = cs , in which cs = ρAs cst = ρAs (cfr + cen + cc + cb )
As V0 + Vr0 2 2
(8)
 Performance Analysis of a Mining Dump Truck Ride 793

Combined piston
zb Vehicle body

Floating piston
Main oil chamber Air chamber B pb

Ring oil chamber

Auxiliary reservoir
zd C md pc
Vr
pr
A pa ws

za (As,ls, M)
Surge pipe
Fig. 2. Structural schematic of the novel HPSs

Fhn Vehicle body

zb
mb

kvn ws
khn ch
M
cw
za
ma
Fhn Vehicle axle
Fig. 3. Nonlinear dynamic model of the novel HPSs

where cst is total loss coefficient, cfr is the loss coefficient due to friction, cen is the loss
coefficient due to enlargement, cc is the loss coefficient due to contraction and cb is the
loss coefficient due to bends in the pipe, respectively.
794 D. V. Ha et al.

Combined Eqs. (6) and (4), the total vertical force of the air chamber A of the novel
HPSs can be achieved by

p0 A2p n
Fan = z + cw |ẇs |w sign(ẇs ) + M ẅs (9)
V0 + Vr0
The resistance force of the novel HSPs [12] is defined as
 n 
ρ(Ab − Ac )ż 2 sign(ż) V0
Fhn = − + p (A
0 c − A p ) − 1
2Cd2 A2 V
(10)
md z̈d
− (Ac − Ap ) + μv ż + Fc sign(ż) + Fs e−k|ż| sign(ż)
Ap

Combined Eqs. (9), (10), and (3) the vertical dynamic force of the novel HPSs can
be achieved by

p0 A2p n ρ(Ab − Ac )ż 2 sign(ż)

Fh = z + cw |ẇs |w sign(ẇs ) + M ẅs −
V0 + Vr0 2Cd2 A2
 n 
V0 md z̈d (11)
+p0 (Ac − Ap ) −1 − (Ac − Ap )
V Ap
+μv ż + Fc sign(ż) + Fs e−k|ż| sign(ż)

Road surface roughness: The road input is simulated for the random road surface
roughness according to ISO 8608(2016) [10]: The random white noise is selected as
excitation source waveform for vehicle suspension [11], the random road profile is
produced by filtering the white noise using the following mathematical model of the
random road surface roughness.

q̇(t) + 2πf0 q(t) = 2πn0 Gq (n0 )v(t)w(t) (12)

where, Gq (n0 ) is the road roughness coefficient which is defined for typical road classes
from A(Gq (n0 ) = 4 × 10−6 m3 ) to H (Gq (n0 ) = 16384 × 10−6 m3 ) according to ISO
8608(2016) [10], n0 is a reference spatial frequency which is equal to 0.1 m; v(t) is the
speed of vehicle; f0 is a minimal boundary frequency with a value of 0.0628 Hz; w(t) is
a white noise signal.

3 Results and Discussion

The quarter-vehicle dynamic model under the random road surface roughness is simu-
lated by the Matlab/Simulink software with the parameters in Ref. [12] when vehicle
moves on from ISO class C road surface at speed of 40 km/h and full load. The vertical
acceleration of vehicle body is shown in Fig. 4. From the achieved results in Fig. 4, we
could determine the value of the RMS acceleration of the vertical vehicle body (awb )
through Eq. (13) according to the international standard ISO 2631-1 [13] is 1.122 m/s2 .
 Performance Analysis of a Mining Dump Truck Ride 795

The time domain root mean square (r.m.s) acceleration responses of vehicle body (awz )
based on the ISO 2631:1997(E) standard [13, 14] is defined as
⎡ ⎤ 21
T
1
awz =⎣ az2 (t)dt ⎦ (13)
T
0

where, az (t) is the measured or simulated acceleration (translational and rotational) as a

function of time, m/s2 or rad/s2 ; T is the duration of the measurement or the simulation
time.

2
az/(m/s2)

-2

-4
0 5 10 15 20 25 30
Time/s
Fig. 4. The vertical acceleration of vehicle body

To analyze the ride comfort performance of a mining dump truck, the operating
conditions such as road surface, vehicle speed will be presented in the next sections.
Road surface conditions: Five road surface conditions from ISO class A (Gq (n0 ) =
4 × 10−6 m3 ) to ISO class E Gq (n0 ) = 1024 × 10−6 m3 based on the road data of ISO 8608
[6] are selected for analysis. The values of the RMS acceleration of the vertical vehicle
body (awb ) with original suspension system and novel HPSs when vehicle moves under
different road surface conditions at vehicle speeds of 40 km/h and full load are shown
in Fig. 5. The results in Fig. 5 show that the awz value with both two types of suspension
increases more quickly when vehicle move on the worse road surface conditions. The
awz value with novel HPSs reduce by 51.78%, 30.85%, 18.16 %; 15.45% and 20.87%.
The performance of novel HPSs has significantly improved vehicle ride comfort in
comparison with original suspension system under very good road conditions.
Vehicle speed conditions: The values of vehicle speed from 20 km/h to 80 km/h are
chosen are selected for analysis. The aw value with original suspension system and novel
HPSs when vehicle moves on ISO class C at different vehicle speed conditions and full
load is shown in Fig. 6. The result of Fig. 6 show that the awz value with both two types
of suspension increases more quickly when the value of vehicle speed increases.
796 D. V. Ha et al.

6
Original suspension
Novel HPSs

awz/(m/s )
4

2
2

0
A B C D E
Different road surface conditions
Fig. 5. awb value with two types of suspension under different road surface conditions

2
Original suspension
1.5 Novel HPSs
awz/(m/s )
2

0.5

0
20 30 40 50 60 70 80
Different vehicle speed conditions (km/h)
Fig. 6. awb value with two types of suspension under different vehicle speed conditions

4 Conclusion

In this paper, a quarter-vehicle dynamic model of a mining dump truck using hydro-
pneumatic suspension system (HPSs) suspension system is proposed to analyze the ride
comfort performance of a mining dump truck under different operating conditions. The
main study conclusions drawn from the analysis results are: (1) The awz value with both
two types of suspension increases more quickly when vehicle move on the worse road
surface conditions and (2) the awz value with both two types of suspension increases
more quickly when the value of vehicle speed increases. The results of this study are a
useful theoretical basis for researchers and designers to renovate the design of vehicle
suspension systems.

Acknowledgment. The work described in this paper was supported by Thai Nguyen University
of Technology for a scientific project.
 Performance Analysis of a Mining Dump Truck Ride 797

References
1. Gong, B., Guo, X., et al.: Ride comfort optimization of a multi-axle heavy motorized wheel
dump truck based on virtual and real prototype experiment integrated Kriging model. Adv.
Mech. Eng. 7, 168781401558425 (2015)
2. Qi, H., Chen, Y., Zhang, N., et al: Improvement of both handling stability and ride comfort of
a vehicle via coupled hydraulically interconnected suspension and electronic controlled air
spring. Proc. Inst. Mech. Eng., Part D: J. Automob. Eng. 095440701985653 (2019)
3. Long, L.X., Hong, T.T., et al.: Performance analysis of the hydro-pneumatic suspension
system of heavy truck. Int. J. Mech. Eng. Technol. (IJMET) 9, 1128–1139 (2019)
4. Saglam, F., Samim Unlusoy, Y.: Analysis of interconnected hydro-pneumatic suspension
system for load sharing among heavy vehicle axles. In: Proceedings of the 3rd International
Conference on Control, Dynamic Systems, and Robotics, Ottawa, Canada (2016)
5. Badway, I., et al.: Simulation and control of a hydro-pneumatic suspension system. Int. J. Sci.
Eng. Res. 8(7), 930–935 (2017)
6. Long, L.X., Van Quynh, L., et al.: A comparison of ride performance of hydro-pneumatic
suspension system with those of rubber and leaf suspension systems. IOP Conf. Ser.: Mater.
Sci. Eng. 914, 012037 (2020)
7. Long, L.X., Van Quynh, L., et al.: Ride performance evaluation of air and hydro-pneumatic
springs of suspension system. IJARET 12(1), 439–447 (2021)
8. Le, V.Q.: Comparing the performance of suspension system of semi-trailer truck with two air
suspension systems. Vibroengineering PROCEDIA 14, 220–226 (2017)
9. Van Liem, N., Jianrun, Z., et al.: Performance analysis of air suspension system of heavy
truck with semi-active fuzzy control. J. Southeast Univ. 33, 159–165 (2017)
10. ISO 8608: Mechanical Vibration-Road Surface Profiles-Reporting of Measured Data.
International Organization for Standardization (2016)
11. Long, L.X., Quynh, L.V., Cuong, B.V.: Study on the influence of bus suspension parameters
on ride comfort. Vibroengineering PROCEDIA 21, 77–82 (2018)
12. Long, L.: Performance analysis of the hydro-pneumatic suspension system on vehicle ride
comfort. Science Project Thai Nguyen University of Technology. Thai Nguyen, Vietnam
(2019)
13. ISO 2631-1. Mechanical Vibration and Shock-Evaluation of Human Exposure to Whole-Body
Vibration, Part I: General Requirements. ISO (1997)
14. Long, L.X., Ha, D.V., et al.: Performance evaluation of a novel hydro - pneumatic suspension
system of a heavy truck on ride comfort. In: E3S Web of Conferences, vol. 304, 01009 (2021)
 Problematic Issues of Improving the Quality
of Reception in VHF SDR

I. Yu. Malevich1 , Minh Tuan Nguyen2(B) , P. V. Zayats3 , Thuong TK. Nguyen2 ,

and Quynh C. Le4
1 Belarusian State University of Informatics and Radioelectronics, Minsk, Belarus
2 Thai Nguyen University of Technology, Thai Nguyen City, Vietnam
nguyentuanminh@tnut.edu.vn
3 LLC, Analog Systems, Minsk, Belarus
4 Electric Power University, Hanoi 100000, Vietnam

Abstract. The current stage in the development of very high frequency (VHF)
radio receiving equipment is the widespread introduction of Software-defined
radio (SDR) platforms. Such devices allow software control of operating radio
frequency parameters of the radio channel, providing both versatility and solutions
to the problems of specific radio information systems. The operation of VHF SDR
takes place in a complex electromagnetic environment with a lot of in-band and
out-of-band interference, and a noticeable level of the radio background, the level
of which in industrial areas is several times higher than the intrinsic noise of the
radio receiving path. The paper discusses the problems and modern approaches
to improve the quality of reception in VHF SDR. In addition, we provide recent
developments of VHF SDR in different scenarios and results in some specific
applications. The results show promise and practical uses in various fields for
communication purposes.

Keywords: Software defined radio (SDR) · Very high frequency (VHF) · Radio
channel · Electromagnetic environment · Transmitter · Receiver

1 Introduction

Radio reception in the VHF band plays a key role in ensuring the effective function of
zonal radio information systems [1]. Many years of experience in the use of VHF radio
channels confirms the significant advantages of such systems, which are characterized
by the easiness of organization and operation, stable performance on rough terrain and in
high-rise urban areas, compactness of technical means, manufacturability in production,
and relatively low cost [2].
Software Defined Radio (SDR) does not have a single, consistent definition that is
accepted everywhere. The players in the field have somewhat various interpretations.
Additionally, a number of terminologies that are similar have been developed and are
used in varying degrees. Software-based radio [3], reconfigurable radio [4], and radio
with flexible architecture are some of these. How flexible the radio waveform can be

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 798–805, 2023.
https://doi.org/10.1007/978-3-031-22200-9_84
 Problematic Issues of Improving the Quality of Reception 799

modified via altering software (SW) and without changing the SDR Platform (the mix
of hardware and operating environment where the waveform application is running)
is the primary parameter in many interpretations and associated words. Being able to
communicate at any suitable frequency, bandwidth, modulation, and data rate by simply
loading the proper SW is undoubtedly the ideal but impractical aim. Typically, a more
realistic understanding of SDR is offered, meaning just that significant portions of the
waveform are defined in SW, allowing for some freedom in waveform modification
within constraints set by the actual system. Most people presume that the flexibility
includes at least multi-band and multi-modulation.
Utilizing most software innovations, SDR provides a simple method for designing
and implementing flexible radiofrequency (RF) transceiver systems [5]. As a result,
it enables quick familiarization with common design challenges for receivers (RF
electronics flaws, asynchronously sampled signals, etc.) without being constrained by
the materials often associated with full-hardware implementations (i.e., cost of hard-
ware components to be soldered together, need of various development and testing
instruments…).
The idea of using a cost-effective, mass-market technology, such as IEEE 802.11 at
very high frequencies (VHF), taking advantage of the recent IEEE 802.11af variant, may
be interesting for applications such as large volume data exchange from sensing devices
[6], with the goal of avoiding proprietary solutions. With GNURadio, custom-designed
physical (PHY) and medium access control (MAC) layers tailored to the propagation
medium can be created using the new technology of software-defined radio (SDR), which
transfers all signal processing operations such as sampling, modulation, and channel cod-
ing to software. SDR enables the IEEE 802.11 standard to be used at lower frequencies
[7].

Digital Sample at Intermediate Frequency

Antenna
Digital Sample at baseband
Analog Signal
DDC
at Intermediate Frequency

Digital Low-Pass
RF Tuner ADC DSP
Mixer Filter

Digital
Analogic Components (in Red) Osc.
Digital Components (in Blue)

Fig. 1. Block diagram of the SDR receiver

A SDR receiver’s block diagram is shown in Fig. 1. The first three blocks of the
superheterodyne receiver’s first three blocks are duplicated in the RF tuner’s first con-
version of the analog signal to IF. After that The ADC converter receives the IF signal
and is responsible for modifying the signal’s domain so that its output offers digital sam-
ples. Then the samples are sent into the input of the digital down converter stage (DDC).
800 I. Y. Malevich et al.

The DDC, which is frequently a monolithic chip, is the essential component of the SDR
system. A digital mixer, a digital local oscillator, and a Finite Impulse Response (FIR)
low-pass filter make up its three primary parts. The IF digital samples are converted to
baseband by the digital mixer and local oscillator, and the resulting signal’s bandwidth
is constrained by the FIR low-pass filter [8]. The DDC has a significant number of
multipliers, adders, and shift registers for the implementation of each of its components.
At the output of the digital mixer, the signals are converted to their baseband counter-
part by being split into the I and Q counterphase components [9]. The converted signal
can be moved away from or toward the point when it hits 0 Hz by changing the tuning
of the digital local oscillator. Which portion of the reception recognized as a meaningful
signal depends on this fluctuation and the low-pass filter’s bandwidth adjustment.

Analog Signal
Digital Samples at Intermediate Frequency
at Intermediate Frequency
Antenna
DUC

Interpolation Digital A/D RF Power

DSP Filter Mixer Amplifier
Conv. Conv.

Digital
Osc. Analogic Components (in Red)
Digital Components (in Blue)

Fig. 2. Block diagram of the SDR transmitter

As illustrated in Fig. 2, a baseband signal, normally produced by a DSP, is received

as an input by SDR transmitters. The baseband signal is converted to IF, a digital up
converter (DUC). The samples are converted to the analog domain by the next DAC.
The signal is then shifted toward higher frequencies by the RF converter. The signal is
then boosted and sent to the antenna.
The DUC’s Interpolation Filter is in charge of increasing the baseband signal’s
sample rate to the following components’ operational frequency. It does the opposite
action of the Decimator in the receiver’s design. The samples are then shifted to IF by
the digital mixer and local oscillator, with the local oscillator controlling the shift.
Like other technologies, SDR has migrated from military to civilian settings. The
United States Navy created Speakeasy [10], the first operational SDR, from 1991 to
1995.Unfortunately, the program was only compatible with the hardware for which it
was designed. The fact that the equipment completely took up the rear of a transport
vehicle was still another drawback. The development of wireless communications cir-
cuits, reusable and modular programming approaches, and breakthroughs in electronics
all contributed to his younger brother Speakeasy II’s [11] much larger success.
SDR hardware as well as software are currently both relatively affordable (the major-
ity of software implementations are really free), which encourages thinking about using
the paradigm in radio solutions. In this paper, the authors specifies that the examination
 Problematic Issues of Improving the Quality of Reception 801

of SDR technology, including hardware designs and application domains, will be the
article’s main goal. Additionally, we demonstrate how a software-defined radio device
works in order to highlight job prospects for low-cost solutions that can address latent
issues in typical radio contexts.
The rest of this work is organized as follows. Section 2 provides some modern
problems of VHF SDR developments in which discussions are provided in details. Some
results of recent VHF SDR developments are addressed in Sect. 3. Finally, conclusions
and future work are provided in Sect. 4.

2 Modern Problems of VHF SDR Developments

The current stage in the development of VHF radio receiving equipment is the widespread
introduction of SDR platforms. Such devices allow software control of operating radio
frequency parameters of the radio channel, providing both versatility and solving the
problems of specific radio information systems [12–14].
The operation of VHF SDR takes place in a complex electromagnetic environment
with a lot of in-band and out-of-band interference and a noticeable level of the radio
background, the level of which, in industrial areas, is several times higher than the
intrinsic noise of the radio receiving path.
For example, Fig. 3 shows the experimental results of measuring the electromagnetic
environment in a radio channel with vertical polarization and a bandwidth of 1 MHz for a
sparsely populated countryside, remote from railways and high-power radio transmitters
(zone A), an industrial area of a large city (zone B) and a cluster with a powerful pulse
radio transmitter (zone C).

Fig. 3. A figure of experimental results of measuring the electromagnetic environment in a radio

channel with vertical polarization and a bandwidth of 1 MHz for different zones

As can be seen, the problem of improving the quality of reception in VHF SDR is
largely determined by the noise immunity of the radio path.
802 I. Y. Malevich et al.

An obvious way to reduce the impact of intense out-of-band interference, including

intentional, is to use a fast-tunable narrow-band filter with a bandwidth of BW ≤ (0.02
… 0.03)f 0 or a filtering system with a switched bandpass filter bank.
However, realization of the high selectivity fast-tunable narrow-band VHF filter with
low in-band losses (less than 1.5 … 2 dB) is a difficult technical task.
Systems with switched VHF bandpass filters are characterized by high speed, how-
ever, the number of selective channels does not usually exceed 4 … 6, and their
bandwidths are rarely less than 0.1f 0 .
Due to the limited selectivity of VHF filter systems, the problem of achieving a
wide dynamic range of the radio frequency path and its individual components become
important. So, for example, the required dynamic range of a broadband LNA included
in a typical VHF SDR preselector operating in region B (Fig. 1) must be at least 80 …
100 dB, and in region C—at least 110 … 120 dB. In this case, the LNA noise figure
should not usually exceed 3 … 4 dB.
In a non-stationary electromagnetic environment, it is promising to increase the noise
immunity of VHF SDR with the help of adaptive systems for adjusting the dynamic
range of the radio path. At the same time, it is important to take into account that the
common attenuator adjustments, traditionally used to ensure the stable operation of the
radio channel under the influence of intense interference in VHF, have a rather low
efficiency. Therefore, adaptive systems for adjusting the dynamic range of the radio
path, used to improve the noise immunity of VHF SDR, require the development of
new structural solutions (for example, relay systems or adaptive LNAs) that provide an
effective exchange of sensitivity for linearity.
Among the system-wide issues of improving the quality of reception in VHF SDR,
the structural-parametric approach to optimizing the radio path, taking into account real
operating conditions, is of practical interest.
An important task is also the development of general principles for the operation
of VHF SDRs, the dynamic range of which varies in accordance with the changing
electromagnetic environment.
Thus, the problematic issues of improving the quality of reception in VHF SDR,
including the development of a set of methods and technical implementations of the
radio path, are considered.

3 Recent VHF SDR Developments

3.1 Development of a Passive VHF Radar System Using SDR for Equatorial
Plasma Instability Studies
There has been a lot of research done on passive radar systems to track non-geophysical
objects like aircraft [15]. The “transmitters of opportunity” used in this research included
satellite, analog television, and FM radio broadcasts, among other types [16]. However,
for radar applications, the waveform chosen should meet two criteria. The light must be
strong enough and it must offer a helpful self-ambiguity reaction. Range and Doppler
estimates are excellent uses of GPS satellite waveforms [17]. Passive radar systems, in
general, are less expensive to operate and maintain, making them a viable option for
long-term studies of ionospheric abnormalities.
 Problematic Issues of Improving the Quality of Reception 803

3.2 Underwater IEEE 802.11 Networks at VHF Using SDR

In the study, software defined radio platforms are used to assess the performance of
widely available, cost-effective IEEE 802.11 networks operating at various sub-GHz
frequencies underwater [7]. The gr-ieee802.11 implementation was improved, and spe-
cially made antennas permitted testing of frequencies in the 70–700 MHz range in
real-world situations, and similar to [18].

Fig. 4. Seawater tests between two SDR at 5 MHz channel bandwidth

The dipole antenna is deployed with a 70 MHz operating frequency in the saltwater
testbed. Figure 4 shows the results that are attained. The jitter of 3 ms up to 1.2 m and
average throughputs of roughly 300 kbit/s is attained. The longest distance is 1.8 m. It
should be noted that because there is a different optimal value for each frequency and
distance, we have to manually adjust the SDR receiver gain. Further details can be found
in [7].

3.3 SDR Based Data Link Design for VHF Band Wireless Sensor Networks
Wireless sensor networks (WSNs) have been considered as emerging technologies for
remate sensing in a various field [19, 20]. In this research, a VHF 144–147 MHz band
operational tactical wireless sensor network is considered. Urban, rural, and wooded
contexts are all primarily covered by works on VHF channel propagation modeling [21].
To the best of our knowledge, there aren’t many articles in the literature that address
the problem of VHF band channel modeling in mountainous, uneven terrain with fixed
terminals. In order to create a computer-based propagation model that could forecast
route loss by providing the terrain profile between the transmitter and receiver, VHF
propagation measurements at a frequency of 110.6 MHz at low altitude over mountainous
804 I. Y. Malevich et al.

and wooded terrain should be carried out. In addition, VHF propagation models for
multi-media WSNs [22] could be also considered.

4 Conclusions and Future Work

In this work, we provide a framework of Software-defined radio (SDR) platforms based
on very high frequency (VHF) communications in different applications. The creation of
a multifrequency passive radar system for equatorial ionospheric research is discussed
in this work. The use of IEEE 802.11 networks in the sub-GHz band is sufficient for a
cost-effective, low power broadband communications solution, both in freshwater and
seawater, hence expanding the use of IEEE 802.11af in underwater environments. This
study also presents the overall architecture of a data connection for tactical VHF wireless
sensor networks, together with physical and MAC layer designs and empirical channel
propagation models.
In future work, we want to reduce the quantity of air in the air-water barrier by
studying the radiation patterns of the antennas and evaluating the SNR gain of employing
the antenna outside the cylinder. Additionally, it will be investigated whether using
frequencies between 10 and 70 MHz could extend the range of communications even
further.

Acknowledgements. The authors would like to thank Thai Nguyen University of Technology
(TNUT), Vietnam for the support.

References
1. Wildemeersch, M., Fortuny-Guasch, J.: Radio frequency interference impact assessment
on global navigation satellite systems. EC Joint Research Centre, Security Technology
Assessment Unit, Tech. Rep (2010)
2. Hamidou, K.: Radio and TV broadcasting in Algeria from origins to the present day. Mass
Communication in the Modern Arab World: Ongoing Agents of Change Following the Arab
Spring, 35 (2022)
3. Tuttlebee, W.: Software Defined Radio: Enabling Technologies. Wiley, Chichester (2002)
4. Kenington, P.B.: RF and Baseband Techniques for Software Defined Radio. Artech House,
Boston (2005)
5. Chamaillard, B., Lastera, M., Roque, D.: A flexible VHF-band aeronautical datalink receiver
based on software defined radio. IEEE Aerosp. Electron. Syst. Mag. 33(1), 58–61 (2018).
https://doi.org/10.1109/MAES.2018.170131
6. Celik, G., et al.: A software defined radio based data link design for VHF band wireless sensor
networks. Phys. Commun. 30, 167–178 (2018)
7. Teixeira, F., Santos, J., Pessoa, L., Pereira, M., Campos, R., Ricardo, M.: Evaluation of
underwater IEEE 802.11 networks at VHF and UHF frequency bands using software defined
radios. In: Proceedings of the 10th International Conference on Underwater Networks &
Systems, pp. 1–5 (2015)
8. Giannini, V., Craninckx, J., Baschirotto, A.: Baseband Analog Circuits for Software Defined
Radio. Ed: Springer (2008)
9. Anderson, J.B., Rolf, J.: Understanding Information Transmission. IEEE Press (2005)
 Problematic Issues of Improving the Quality of Reception 805

10. Lackey, R.J., Upmal, D.W.: Speakeasy: the military software radio. IEEE Softw. Defin. Radio:
Basic Princ. Appl. Commun. Magaz. 33, 56–61 (1995)
11. Budroweit, J.: Software-defined radio with flexible RF front end for satellite maritime radio
applications. CEAS Space J. 8(3), 201–213 (2016). https://doi.org/10.1007/s12567-016-
0121-9
12. Collins, T., Getz , R., Pu, Di, Wyglinski, A.: Software-Defined Radio for Engineers. Artech
House (2018)
13. Kenington, P.B.: RF and baseband techniques for software defined radio, ser. Mob. Commun.
Artech House 19, 317–320 (2005)
14. Rouphael, T.J.: RF and Digital Signal Processing for Software-Defined Radio: A Multi-
standard Multi-mode Approach. Newnes (2009)
15. Griffiths, H.D., Baker, C.J.: Passive coherent location radar systems. Part 1: performance
prediction. IEEE Proc.-Radar, Sonar Navig. 152(3), 153–159 (2005)
16. Tuysuz, B.U.R.A.K., Urbina, J., Lind, F.D.: Development of a passive VHF radar system using
software-defined radio for equatorial plasma instability studies. Radio Sci. 48(4), 416–426
(2013)
17. Sethi, H. S., Dashora, N.: Novel meteor observations using FM radio broadcast over Gadanki,
India. IEEE Trans. Geosci. Remote Sens. (2022)
18. Ribeiro, L.M.B., Müller, I., Buss Becker, L.: Communication interface manager for improving
performance of heterogeneous UAV networks. Sensors 21(13), 4255 (2021)
19. Nguyen, M.T., Teague, K.A.: Neighborhood based data collection in wireless sensor net-
works employing compressive sensing. In: 2014 International Conference on Advanced
Technologies for Communications (ATC 2014). IEEE (2014)
20. Nguyen, M.T., Teague, K.A.: Compressive sensing based random walk routing in wireless
sensor networks. Ad Hoc Netw. 54, 99–110 (2017)
21. Dias, M.H.C., Rotava, A., Andrade, F.G., Alem, R.A., Melo, M.A.K., Santos, J.C.A.: Path
loss measurements of HF/VHF land links in a Brazilian Atlantic rainforest urban site, IEEE
Antennas Wirel. Propag. Lett. (2011)
22. Nguyen, M.T.: An energy-efficient framework for multimedia data routing in internet of things
(IoTs). EAI Endorsed Trans. Ind. Netw. Intell. Syst. 6(19), e1–e1 (2019)
 Radial Basis Function Based Finite Element
Method for Bending, Vibration and Buckling
Analysis of Laminated Composite
Mindlin-Reissner Plates

D. Nguyen Kien1(B) , Xuefeng Chen1 , Xiaoying Zhuang1,2 , and Timon Rabczuk3

1 Department of Geotechnical Engineering, College of Civil Engineering, Tongji University,
Shanghai 200092, China
kiendung.nguyen@tongji.edu.cn, nkdung@hotmail.com
2 Chair of Computational Science and Simulation Technology, Department of Mathematics and
Physics, Leibniz University Hannover, 30167 Hannover, Germany
3 Chair of Computational Mechanics, Bauhaus University Weimar, Weimar, Germany

Abstract. The study introduces the formulation of the radial basis function-based
finite element method (RBF-FEM) for bending, free vibration, and buckling anal-
ysis of Mindlin-Reissner laminated composite plates by first-order shear deforma-
tion theory. The method utilizes the radial basis functions to construct the shape
functions from the finite nodes. Compared with the conventional finite element
method (FEM), the present RBF-FEM obtains the shape functions with simplicity,
especially when the number of element nodes is increasing. The proposed RBF-
FEM eight-node element (RBF-FEM-8) is utilized. By applying the Gauss point
reduction technique, the method could solve the thick and thin laminated compos-
ite plates with high accuracy, quick convergence, reducing mesh size in discretiza-
tion, and could remove the shear locking phenomenon. Several numerical example
studies of the thick and thin Mindlin-Reissner plates with the difference in geome-
tries (including line and curve boundary edges), oriented layers, edge boundary
conditions, Young’s modulus ratio, length-to-thickness ratio, are carried out. The
results are then compared with other numerical methods and analytical solutions.

Keywords: Finite element method · Radial basis function · Laminated

composite plates · Mindlin-Reissner plates · First-Order shear deformation
theory

1 Introduction

In various engineering structures, such as building, aircraft, vehicles and etc., the fiber
reinforcement composite materials have been used widely. In design, the laminated
composite materials are made up of stacked lamina layers in order to reach the desired
properties, such as fatigue life, high stiffness, damping, etc. [1]. Several approaches

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 806–822, 2023.
https://doi.org/10.1007/978-3-031-22200-9_85
 Radial Basis Function Based Finite Element Method 807

to modeling the laminated composite plates exist. The popular approach is the equiv-
alent single-layer models (ESL). Classical laminated plate theory (CLPT) [2–4], first-
order shear deformation theory (FSDT) [5–7], and high-order shear deformation theory
(HSDT) [8, 9] are in this classification.
The formulation of the laminated composite plates using FSDT is discussed in the
study. The FSDT could be implemented more easily and obtain better results than the
CLPT because it does not require the derivatives of the generalized displacement field
but requires the transverse shear strains. Another advantage of using FSDT is the com-
putational cost in comparison to the HSDT. The FSDT requires only a C 0 -continuity
approximation of the generalized displacement. Nevertheless, the FSDT has to include
the shear correction factors to take into consideration to the non-linear shear stress
terms. A layer-wise model was recommended in order to improve the shear stress accu-
racy when the laminated composite plates get thicker [10]. The ESL model produces
rather accurate results in the analysis of thin or quite thick laminated composite plates
[11].
In the study, the proposed method for structural analysis, which is a hybrid of the finite
element method and radial basis function, is presented. The method is the so-called radial
basis function-based finite element method (RBF-FEM). The method utilizes the radial
basis functions to construct the shape functions from the finite nodes. Compared with
the conventional finite element method (FEM), the present RBF-FEM obtains the shape
functions with simplicity, especially when the number of element nodes is increasing.
The study introduces the formulation of the radial basis function-based finite element
method for bending, free vibration, and buckling analysis of Mindlin-Reissner laminated
composite plates by FSDT of ESL. The extension of the method to analysis of the
Mindlin-Reissner laminated composite plates is to understand more about the behavior
of the laminated composite thin plates by the method. From the original study regarding
the accuracy, convergence, and efficiency of the RBF-FEM approach, the eight-node
element (RBF-FEM-8) showed the best choice for the solution of solid engineering
problems [12]. In our another study, the Gauss point reduction technique is employed
to avoid the shear locking phenomenon from the thin isotropic FSDT plate element by
RBF-FEM. The accuracy and efficiency of the RBF-FEM-8 solution with the Gauss
point reduction technique to the problems of out-plane laminated composite plates by
FSDT consumption are shown, and the results are then compared with other numerical
and analytical solutions.

2 Formulation of Radial Basis Function Based Finite Element

Method

Formulation of the radial basis function for two-dimension physical degree of freedom is
introduced. It is supposed that a convex four-edge element with np nodes is constructed
(Fig. 1). The radial basis function interpolation of the value s(x) at node is obtained as

n p
s(x) = ai φ(||x − xi ||), (1)
i=1
808 D. N. Kien et al.

Fig. 1. The 2D element in global coordinates.

where x = [x, y], and xi = [xi , yi ] denotes the i th node vertex (Fig. 1), and ai represents
the weight coefficient. In the study, the multiquadrics
 (MQ) radial basis function is used
and its formulation is φ(||x − xi ||) = φ(r, ε) = 1 + (εr)2 , where r = ||x − xi || and ε
denote the distance and the shape parameter, respectively. These two parameters can be
used to scale the inputs of the radial kernel and can influence the fitting accuracy. The
value at each node could be obtained in generalized coordinates as
np

s1 (x) = ai φ(||x1 − xi ||)
i=1
np

s2 (x) = ai φ(||x2 − xi ||) (2)
i=1
np
  
snp (x) = ai φ xnp − xi 
i=1

or it could be written in tensor notation as

s = a, (3)
 T  T
where s = s1 (x)s2 (x) . . . snp (x) , a = a1 a2 . . . anP ,
and
⎡ ⎤
φ(||x1 − x1 ||) · · · φ(||x1 − xnp ||)
⎢ .. .. .. ⎥
⎢ . . . ⎥
⎢ ⎥
⎢ ⎥
 = ⎢ φ(||xi − x1 ||) · · · φ(||xi − xnp ||) ⎥. (4)
⎢ .. .. .. ⎥
⎢ ⎥
⎣ . . . ⎦
φ(||xnp − x1 ||) · · · φ(||xnp − xnp ||)
The solution for a (from Eq. 3) could be obtained as
a = −1 s. (5)
 Radial Basis Function Based Finite Element Method 809

Expressing the term in Eq. (1) as the row vector is multiplied by the column vector,
we then have

s(x) = φ(x)a, (6)

 
where φ(x) = φ(||x − x1 ||) . . . φ(||x − xi ||) . . . φ(||x − xnp ||) .
From Eqs. (5) and (6), the interpolation s(x) could, therefore, be obtained as

s(x) = φ(x)−1 s. (7)

Consequently, the radial basis function based finite element method (RBF-FEM)
shape function is

N R = φ(x)−1 , (8)

while its gradients are

∇N R = ∇φ(x)−1 , (9)
 
where ∇φ(x) = ∇φ(||x − x1 ||) . . . ∇φ(||x − xi ||) . . . ∇φ(||x − xnp ||) .
The RBF-FEM shape function and its derivatives are vividly depicted in Fig. 2 (for
1D element). It is shown that the RBF-FEM shape function is C 0 continuous.
In discretization, the displacement approximation in terms of RBF-FEM shape
function could be obtained in finite subdomain (e ) as

u = u(x) ≈ ue (x) = N eR (x)d e , x ∈ e . (10)

Therefore, the element governing equation could be expressed as

ke ue (x) = f e in e . (11)

(a) (b)
Fig. 2. The one-dimension MQ RBF-FEM shape functions of four-node element (a) and their
derivatives (b).
810 D. N. Kien et al.

3 Formulation of Laminated Composite Mindlin-Reissner Plates

3.1 Principles of Virtual Work for Laminated Composite Plates
Assuming the mid-plane of the plate is represented by the domain  in R2 . The displace-
ment field can be expressed, in the middle plane in terms of first order shear deformation
theory assumption, as

u(x, y, z) = u(x, y) + zθx (x, y),

v(x, y, z) = v(x, y) + zθy (x, y), (12)
w(x, y, z) = w(x, y),

where u, v, w denote the displacement in the z, y, z and θx , θy denote the rotation of the
normal vector to the middle plane w.r.t. y, x axes respectively (Fig. 3). The displace-
 T
ment vector is then given as u = u, v, , w, θx , θy . The vectors of in-plane strain and
transverse shear strain are written as
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
xx ∂u/∂x ∂θx /∂x
 p = ⎣ yy ⎦ = ⎣ ⎦ + z⎣ ⎦ =  m + z b , (13)
 

∂v/∂y ∂θy /∂y

γxy ∂v/∂x + ∂u/∂y ∂θy /∂x + ∂θx /∂y
   
γxz ∂w/∂x + θx 

s = = = s, (14)
γyz ∂w/∂y + θy

where the hat denotes the generalized strain vectors [13].

The laminated plates are made of orthotropic layers, and the layer is arbitrarily
oriented in comparison to the reference axes. So the stress-strain relationship for the
k-th layer (k-th lamina) in matrix form becomes
⎧ ⎫(k) ⎡ ⎤(k) ⎧ ⎫(k)
⎪ σxx ⎪
⎪ ⎪ Q11 Q12 Q16 0 0 ⎪
⎪ xx ⎪
⎪
⎪ ⎪ ⎢Q ⎪ ⎪
⎪
⎪ σ
⎨ yy ⎬
⎪
⎪ ⎢ 21 Q22 Q26 0 0 ⎥ ⎪
⎪
⎥ ⎨ yy  ⎪
⎪
⎬
⎢ ⎥
τxy = ⎢ Q61 Q62 Q66 0 0 ⎥ γ , (15)
⎪
⎪ ⎪ ⎢ ⎥ ⎪ xy ⎪
⎪ τxz ⎪
⎪ ⎪
⎪ ⎣ 0 0 0 Q55 Q54 ⎦ ⎪ ⎪
⎪ γxz ⎪
⎪
⎪
⎪
⎩τ ⎭ ⎪ ⎪
⎩γ ⎪
⎭
yz 0 0 0 Q45 Q44 yz

where Qij denotes the k-th laminae transformed material constant [1].
The weak form of the static model for Mindlin-Reissner composite plate can be
expressed as
 
 

T 
T  

δ p D p d  + δ s Ds  s d  = δuT pd , (16)
  

where p the applied force vector (in case of the transverse load per unit area p =
 T
0, 0, pw , 0, 0 ), and
   

A B


D= .  
(17)
B Db
 Radial Basis Function Based Finite Element Method 811

Fig. 3. Geometric definition of Mindlin-Reissner plate.

   

The symbols A, B, Db and Ds denote the extensional stiffness, bending extensional

coupling stiffness, bending stiffness and transverse shear stiffness matrices, respectively.
They are defined, based on the transformed material constant, as [1, 11]
    h/2
 

Aij , Bij , Dbij = (1, z, z 2 )Qij dz, i, j = 1, 2, 6, (18)

−h/2
 h/2

∗
Dsij = Qij dz, i, j = 4, 5. (19)
−h/2

The free vibration dynamics weak form and buckling weak form for initially stressed
Mindlin-Reissner plates under FSDT assumption, assuming pre-buckling stresses σ 0 ,


could be written as
  
 

δ Tp D p d  + δ Ts Ds  s d  +
    

δuT müd  = 0, (20)

         
   σ̂ 0 0 ∇u
δ Tp D p d  + δ Ts Ds  s d  + h
   

∇ T δwσ̂ 0 ∇wd  + h ∇ T δu∇ T δv d

  0 σ̂ 0 ∇v
 
    (21)
h3   σ̂ 0 0 ∇θx
+ ∇ T δθx ∇ T δθy d  = 0,
12 0 σ̂ 0 ∇θy

 
 operator ∇ = ∂/∂x∂/∂y and the in-plane pre-buckling stresses
wherethe gradient T

σxx
0 τ0
σ0 =

xy
0 σ0 .
τxy yy

3.2 Radial Basis Function Based Finite Element Method Discretization

The plate domain is discretized into subdomains. Each domain (finite mesh) has a np
node. The displacement field interpolated by RBF-FEM shape function is obtained as
⎡ ⎤ ⎡ ⎤⎡ ⎤
u NRie 0 0 0 0 ui
⎢ v ⎥ ⎢ e 0 0 0 ⎥ ⎢ ⎥
⎢ ⎥ n ⎢ 0 NRi ⎥⎢ vi ⎥ n
⎢ ⎥ p ⎢ ⎥ ⎢ ⎥ p
u=⎢w ⎥= ⎢ 0 0 NRi e 0 0 ⎥⎢ wi ⎥ = Ne de . (22)
⎢ ⎥ i=1 ⎢ ⎥ ⎢ ⎥ i=1 Ri i
⎣ θx ⎦ ⎣ 0 0 0 NRi 0 ⎦⎣ θxi ⎦
e

θy 0 0 0 0 NRi e θyi
812 D. N. Kien et al.

Then the generalized membrane, bending, shear strains, and geometrical trains are
written as

np 
np np 
np 

m = Bemi d ei ,  b = Bebi d ei ,  s = Besi d ei ,  g = Begi d ei , (23)

i=1 i=1 i=1 i=1

where
⎡ ⎤ ⎡ ⎤
e
NRi,x 0 000 e
0 0 0 NRi,x 0
⎢ 0 0 0⎥ ⎢ e ⎥
= ⎣ 0 NRi,y ⎦, Bbi = ⎣ 0 0 0 0 NRi,y ⎦,
e
Bemi e
e e
NRi,y NRi,x 0 0 0 e e
0 0 0 NRi,y NRi,x
 
e
0 0 NRi,x e 0
NRi
Bsi =
e
e e ,
0 0 NRi,y 0 NRi

and
⎡ ⎤
e
NRi,x e
NRi,y 0 0 0 0 0 0 0 0
⎢ ⎥
⎢ 0 e
0 NRi,x e
NRi,y 0 0 0 0 0 0 ⎥
⎢ ⎥
Begi = ⎢
⎢ 0 0 0 e
0 NRi,x e
NRi,y 0 0 0 0 ⎥ ⎥,
⎢ 0 e e 0 ⎥
⎣ 0 0 0 0 0 NRi,x NRi,y 0 ⎦
0 0 0 0 0 0 0 e
0 NRi,x e
NRi,y
e and N e represent the derivatives of shape function at i th node over x and
where NRi,x Ri,y
y, respectively.
Therefore, the formulation of plates for static, free vibration and buckling analysis
could be written as

Kd = F, (24)
 ∼
K − ω2 M d= 0, (25)

 
K − λK g d = 0, (26)
∼
where d denotes the set of constant values at nodes in terms of vibration analysis, d the
buckling mode represented by a vector notation, and K = Ane=1 e
ke , F = Ane=1
e
f e, M =
ne n
Ae=1 m , K g = Ae=1 kg (A denotes the assembly to the global matrix). Then the element
e e e

stiffness matrix, load vector, mass matrix and geometrical stiffness matrix are written as
  
ke = BeT m ÂBe
m d  + B eT
m B̂B e
b d  + b B̂Bm d 
BeT e

  
  (27)
+ b D̂b Bb d  +
BeT e
s D̂s Bs d ,
BeT e

 
  
f =
e
ρN eR d , m =
e
R m̂N R d ,
N eT e
keg = g SBg d ,
BeT e
(28)
  
 Radial Basis Function Based Finite Element Method 813

where
⎡ ⎤ ⎡ 0  ⎤
h00 0 0 hσ 0 0 0 0
⎢0 h 0 0 0⎥ ⎢ 0 hσ 0


0 ⎥
⎢ ⎥ ⎢ 0 0 ⎥
⎢ 0 0⎥ ⎢ ⎥
hσ 0
 

m = ρ⎢ 0 0 h ⎥, S = ⎢ 0 0 0 0 ⎥,
⎢ 3 ⎥ ⎢ h3 0 ⎥
⎣0 0 0 h ⎦ ⎣ 0 0 ⎦


12 0 0 0 12 σ
h3 h3 0


12 σ
000 0 12 0 0 0 0

and ρ denotes the mass density, h the thickness, ω the natural frequency and λ the critical
buckling load.

4 Numerical Studies

4.1 Static Analysis of Laminated Composite Plates

In the study, the three-layer simply supported square sandwich square plates (length a,
thickness h) subjected to a uniform load pw is considered [14]. The plate has the length-
to-thickness ratio of 10. We use the RBF-FEM eight-node element with multiquadric
RBF shape function ε = 100. The laminated plate has one inner layer and two skin
layers where Q{skin} = RQ{core} . The plate normal displacement and stress are written
as

w = 0.99781w(a/2, a/2, 0)/(hpw ), σ̃xx

1
= σxx (a/2, a/2, h/2)/pw

σ̃xx
2
= σxx (a/2, a/2, 2h/5)/hpw , σ̃xx
3
= σxx (a/2, a/2, 2h/5)/pw

The convergence of normalized deflection with different values R is investigated in

this problem in a comparison between the present method with other numerical methods
and exact solutions. The normalized deflection for the value R = 5, 10, 15 is shown in
Table 1. The present results converge quickly and have high accuracy in comparison
to other reference numerical methods. Figure 4a shows the normalized deflection with
different R values versus the mesh size. And Fig. 4b–c depicts the deflection relative
errors with the exact solution [14]. It is shown that the results are very much in agreement
with the wavelet method [5]. They produce very small errors. The other methods reported
to compare with the present method are exact solutions [14], isogeometric analysis by
first-order shear deformation FSDT [11], (RBF-PS) method [15], wavelets method [5],
classical laminate plate theory (CLT) [11], and results based on FSDT and high-order
shear deformation HSDT [16] that are shown in Table 1.

4.2 Free Vibration Analysis of Laminated Composite Plates

Next, the free vibration analysis of laminated composite plate is studied. The accuracy
and efficiency of the method with eight-node element (RBF-FEM-8) are examined. The
plate changes in different boundary conditions, length-to-thickness ratios, modulus ratios
and fiber orientation angles are taken into account. The plate has its properties which
814 D. N. Kien et al.

Table 1. The normalized displacement and stresses of the three-layer simply supported square
sandwich plate.

R Method w 1
σ̃xx 2
σ̃xx 3
σ̃xx
5 RBF-FEM-8 259.2627 59.0982 47.2786 9.4557
HSDT [16] 256.13 62.38 46.91 9.382
FSDT [16] 236.10 61.87 49.50 9.899
CLT [11] 216.94 61.141 48.623 9.783
RBF-PS [15] 257.2869 60.4558 46.9146 9.3829
Wavelets [5] 259.278 59.1202 47.2962 9.4592
IGA-Quadratic [11] 259.8258 59.083 47.2664 9.4533
Exact [14] 258.97 60.353 46.623 9.34
10 RBF-FEM-8 159.6065 63.1134 50.4907 5.0491
HSDT [16] 152.33 64.65 51.31 5.131
FSDT [16] 131.095 67.80 54.24 4.424
CLT [11] 118.87 65.332 48.857 5.356
RBF-PS [15] 155.8623 65.5951 49.6095 4.9609
Wavelets [5] 159.611 63.137 50.5096 5.051
IGA-Quadratic [11] 159.9411 63.0971 50.4777 5.0478
Exact [14] 159.38 65.332 48.857 4.903
15 RBF-FEM-8 121.9296 63.6320 50.9056 3.3937
HSDT [16] 110.43 66.62 51.97 3.465
FSDT [16] 90.85 70.04 56.03 3.753
CLT [11] 81.768 69.135 55.308 3.687
RBF-PS [15] 116.8911 67.2306 49.5823 3.3055
Wavelets [5] 121.9329 63.6557 50.9246 3.395
IGA-Quadratic [11] 122.181 63.616 50.892 3.393
Exact [14] 121.72 66.787 48.299 3.238

are given as G12 = G13 = 0.6E2 , G23 = 0.5E2 , Young’s modulus ratios are used until
stated E1 /E2 = 10, 20, 30, 40, Poisson ratio ν12 = 0.25 and mass density ρ = 1. For
all computations, we apply the shear correction factors α1 = α2 = π 2 /12.
Square Laminated Composite Plates
The free vibration of a four-layer [0◦ /90◦ /90◦ /0◦ ] SSSS square plate is studied in this
example. It is assumed that the plate’s layers have the same thickness, mass density
and elastic composite material. The plate has the thickness-to-length ratio of a/h =
0.2. Table 2 shows the convergence of the first normalized frequency with different
Young’s modulus ratios. The RBF-FEM-8 results are then compared with the exact
solution (HSDT consumption [1, 9]), Liew et al. method [7], RBFs [17], wavelets [5],
 Radial Basis Function Based Finite Element Method 815

(a) Normalized deflection .

(b) Comparison of relative error of central (c) The close-up of comparison of relative
deflection among methods. error of central deflection among methods.
Fig. 4. Relative error of central deflection.

isogeometric analysis [11], polygonal finite element method (PFEM) and FEM-Q4. The
presented results are close to the moving least squares differential quadrature method
[7] and the exact values [1, 9] based on HSDT (Fig. 5).
Next, the free vibration of a three-layer [0◦ /90◦ /0◦ ] laminated CCCC square plate
is carried out. In this example, the Young’s modulus ratio sets to E1 /E2 = 40. The
length-to-thickness ratio is a/h = 10, 100. It is defined the normalized frequencies as
parameter ω = (ωa2 /π 2 )(ρh/D0 )1/2 with D0 = E2 h3 /12(1 − ν12 ν21 ). The results are
then compared with other numerical methods, such as RBF-PS [15], wavelets [5] and
the isogeometric analysis method (IGA) based on FSDT [11]. Table 3 shows the first
six normalized frequencies of the RBF-FEM-8 in comparison to the other reference
methods. They convergence quickly with high accuracy, even with coarse mesh.
Circular Laminated Composite Plates
The next study is the vibration analysis of a four-layer [45◦ / − 45◦ / − 45◦ /45◦ ] clamped
circular laminated composite plate. The geometry of the circular plate is illustrated
in Fig. 6. Table 4 presents the first six normalized frequencies of the plate with the
geometry ratio 2R/h = 10. The results in different mesh size are then compared with
other numerical methods, MISQ20 element method using FSDT [18], MLSDQ method
by Liew et al. [7], isogeometric analysis method [11]. The table shows that the results
816 D. N. Kien et al.

Table 2. The non-dimensional frequency parameter ω = (ωa2 /h)(ρ/E2 )1/2 of the four-layer
[0◦ /90◦ /90◦ /0◦ ] simply supported square laminated plate with different Young’s modulus ratios.

Method Points per side E1 /E2

10 20 30 40
RBF-FEM-8 9 8.3194 9.5880 10.3526 10.8794
17 8.3164 9.5873 10.3474 10.8752
PFEM 9 8.5103 9.7667 10.5117 11.0265
17 8.3667 9.6391 10.3930 10.9135
FEM-Q4 9 8.5234 9.7789 10.5235 11.0380
17 8.3734 9.6453 10.3988 10.9191
RBF [17] 9 8.2540 9.4986 10.2320 10.7341
17 8.2526 9.4974 10.2308 10.7329
Wavelets [5] 9 12.2487 13.1468 13.6379 13.9596
17 8.2794 9.5375 10.2889 10.8117
IGA-quartic [11] 9 8.2807 9.5388 10.2902 10.8130
17 8.2790 9.5376 10.2890 10.8118
Liew et al. [7] – 8.2924 9.5613 10.320 10.849
Exact [1, 9] – 8.2982 9.5671 10.326 10.854

Fig. 5. Comparison of methods on relative error of non-dimensional frequency versus the Young’s
modulus ratio E1 /E2 .

quickly converge and are quite close to the reference results. Figure 7 depicts the contour
of the first six modes of the plate.
 Radial Basis Function Based Finite Element Method 817

Table 3. The first six non-dimensional frequency parameters ω of a three-layer [0◦ /90◦ /0◦ ]
clamped square laminated plate.

a/h Method Mesh size Modes

1 2 3 4 5 6
10 RBF-FEM-8 25 7.3988 10.4747 13.6227 15.6037 16.0477 19.6899
100 7.4014 10.4527 13.6151 15.6768 15.7777 19.7244
400 7.4016 10.4512 13.6145 15.6835 15.7573 19.7332
IGA-quartic [11] – 7.4133 10.4016 13.9149 15.4463 15.8121 19.5863
RBF-PS [15] – 7.4727 10.2544 14.2440 14.9363 15.9807 19.4129
Wavelets [5] – 7.4106 10.3944 13.9128 15.4403 15.8061 19.5797
100 RBF-FEM-8 25 14.6417 18.8352 28.5288 38.5496 43.1384 43.6365
100 14.4217 17.3916 24.3462 35.5228 37.7107 39.3197
400 14.4203 17.3805 24.2728 35.2254 37.6929 39.3072
IGA-quartic [11] – 14.6003 17.7822 25.1828 36.9392 38.1465 39.8788
RBF-PS [15] – 14.4305 17.3776 24.2662 35.5596 37.7629 39.3756
Wavelets [5] – 14.4455 17.5426 25.1868 37.8851 39.5489 39.6519

Fig. 6. The circular plate and a typical mesh.

4.3 Buckling Analysis of Laminated Composite Plates

This section analyzes the buckling problem of the laminate composite plates. The
square plate under uniaxial and biaxial compression are taken into account. The dif-
ference in oriented layers, edge boundary conditions, Young’s modulus ratios, length-
to-thickness, are carried out. Let us define the buckling load factor λ = λa2 /(E2 h3 ),
where λ is critical buckling load. We configure the material parameter as follows:
818 D. N. Kien et al.

Table 4. The non-dimensional frequency parameter ω = (ω(2R)2 /h)(ρ/E2 )1/2 of the four-layer
[45◦ / − 45◦ / − 45◦ /45◦ ] clamped circular laminated composite plate.

Method Mesh size Modes

1 2 3 4 5 6
RBF-FEM-8 145 24.8541 39.6191 43.9359 58.0875 58.4270 66.5745
353 24.7420 39.1019 43.5852 56.9620 57.2690 65.5143
897 24.6740 38.9945 43.4764 56.8148 57.1201 65.3586
1581 24.6569 38.9675 43.4491 56.7779 57.0827 65.3196
IGA-quartic [11] – 24.6622 38.9814 43.4559 56.9205 56.9844 65.3320
MISQ20 [18] – 24.766 39.441 43.817 57.907 57.945 66.297
MLSDQ [7] – 24.752 39.181 43.607 56.759 56.967 65.571

(a) Mode 1. (b) Mode 2. (c) Mode 3.

(d) Mode 4. (e) Mode 5. (f) Mode 6.

Fig. 7. The first six mode shapes of the four-layer [45◦ / − 45◦ / − 45◦ /45◦ ] clamped circular
laminated composite plate.

G12 = G13 = 0.6E2 , G23 = 0.5E2 , Young’s modulus ratios are used until stated
E1 /E2 = 3, 10, 20, 30, 40 and Poisson ratio ν12 = 0.25. For all computations, we
apply the shear correction factors as α1 = α2 = π 2 /12. The plate could have the edge
boundaries that are free (F), simply supported (S) or clamped (C).
 Radial Basis Function Based Finite Element Method 819

Square Plates Under Uniaxial Compression

A four-layer cross-ply [0◦ /90◦ /90◦ /0◦ ] SSSS square laminated plate under uniaxial
compression is studied in this example. The study of the length-to-thickness effect on
the uniaxial buckling compression load is taken into account. Table 5 shows the critical
buckling load with a/h = 10, 20, 30, 40. The results of the present method quickly
converge with the mesh sizes. They are close to the reference methods shown in the
table.
The following study is considering the influence of the mixed boundaries on the uni-
axial buckling load of the 10-layer cross-ply square plate [0/90◦ ]5 . In the computation,
we configure length-to-thickness ratio of a/h = 10 and modulus ratio of E1 /E2 = 40.
Six different examples of edge boundary conditions carried out are given in Table 6. The
results are compared with the reference solution shown in the table. Again, they quickly
converge and are close to the reference solutions. Figure 8 depicts the contour of the first
buckling mode of the plate with various boundary edge conditions.

Table 5. A normalized critical buckling load of the four-layer cross-ply [0◦ /90◦ /90◦ /0◦ ] simply
supported square plate with various a/h ratios.

Method Mesh size a/h

10 20 30 40
RBF-FEM-8 25 23.3388 31.6381 35.3289 35.9552
100 23.3489 31.6558 35.3451 35.9519
400 23.3495 31.6568 35.3464 35.9531
FSDT [8] – 23.471 31.707 35.356 35.955
HSDT [8] – 23.349 31.637 35.419 35.971
IGA-quartic [11] – 23.1322 31.5498 35.3258 35.9499

Square Plates Under Biaxial Compression

Finally, the three-layer cross-ply [0◦ /90◦ /0◦ ] SSSS plate under the biaxial load are taken
into account. The plate has the thickness-to-length a/h = 10. This study examines the
effect of Young’s modulus ratio E1 /E2 on the critical buckling load. The normalized
critical buckling load of the present method with different mesh size in comparison
to other reference solutions is shown in Table 7. The biaxial buckling load increases
when the ratio E1 /E2 increases. The table shows that the present method (RBF-FEM-8)
produces results that are quite close to the reference solutions.

5 Conclusion

The study presented the applications of the radial basis function-based finite element
method (RBF-FEM) to the bending, free vibration, and buckling analysis of laminated
composite Mindlin-Reissner plates. The mathematical theory of the laminated composite
820 D. N. Kien et al.

 
Table 6. A normalized critical buckling load of cross-ply ten-layer cross ply 0/90◦ 5 square
plate with various types of boundaries (E1 /E2 = 40, a/h = 10).

Method Mesh size Boundary conditions

SSSS SSFF SSCC SSSC SSFC SSFS
RBF-FEM-8 25 25.3389 12.0522 34.4098 32.2940 14.3028 12.4085
100 25.3506 12.0494 34.4559 32.3561 14.3065 12.4096
400 25.3513 12.0492 34.4547 32.3583 14.3067 12.4096
PFEM 25 28.0260 13.1031 38.9850 36.0609 15.4962 13.5728
100 26.0625 12.2678 35.6997 33.3472 14.5639 12.6688
400 25.5281 12.0988 34.7640 32.6047 14.3751 12.4794
FEM-Q4 25 28.1229 13.4593 39.4960 36.2297 15.7953 13.8500
100 26.0345 12.3966 35.6873 33.3148 14.6780 12.7762
400 25.5217 12.1357 34.7607 32.5967 14.4000 12.5034
FSDT [20] – 25.450 12.092 34.837 32.614 14.358 12.524
HSDT [20] – 25.423 12.077 35.376 32.885 14.351 12.506
MISQ20 [21] – 25.525 12.131 35.105 32.870 14.352 12.541
MLSDQ [22] – 25.338 12.030 34.604 – – –

(a) SSSS. (b) SSFF. (c) SSCC.

(d) SSSC. (e) SSFC. (f) SSFS.

 ◦

Fig. 8. Fundamental buckling modes of 10-layer 0/90 5 composite plate.
 Radial Basis Function Based Finite Element Method 821

Table 7. The normalized critical biaxial buckling load of the three-layer cross ply [0◦ /90◦ /0◦ ]
simply supported square plate with various Young’s modulus ratios.

Method Mesh size E1 /E2

10 20 30 40
RBF-FEM-8 25 4.8871 7.3640 8.8987 10.2001
100 4.8895 7.3674 8.9070 10.2100
400 4.8896 7.3676 8.9075 10.2106
MISQ20 [21] 4.939 7.488 9.016 10.252
FSDT [23] 4.963 7.588 8.575 10.202
HSDT [9] 4.963 7.516 9.056 10.259

plate by the first-order shear deformation theory was presented. The method with eight-
node elements and the Gauss point reduction technique was utilized to avoid the shear
locking phenomenon without any shear strain modification. Several numerical example
studies of the plates with a difference in geometries (including line and curve boundary
edges), oriented layers, edge boundary conditions, Young’s modulus ratio, and length-
to-thickness ratio are carried out. The results were then compared with other numerical
methods and exact solutions. It is shown that the method could solve the problem of
laminated composite thin plates with quick convergence and with high accuracy. Fewer
meshes were required. However, the method has not yet tested the problems with skew
elements. In the near future, it should take some time to implement the method with
skew elements and distorted meshes. An extension of the method with the theory of
deep learning to engineering problems is interesting to carry out [19].

References
1. Reddy, J.N.: Mechanics of Laminated Composite Plates and Shells: Theory and Analysis.
CRC Press, New York (2004)
2. Reissner, E., Stavsky, Y.: Bending and stretching of certain types of heterogeneous aeolotropic
elastic plates. J. Appl. Mech. 28(3), 402–408 (1961)
3. Stravsky, Y.: Bending and stretching of laminated aeolotropic plates. J .Eng. Mech. Div. 87(6),
31–56 (1961)
4. Dong, S.B., Pister, K.S., Taylor, R.L.: On the theory of laminated anisotropic shells and plates.
J. Aerosp. Sci. 29(8), 969–975 (1962)
5. Ferreira, A., Castro, L.M., Bertoluzza, S.: A high order collocation method for the static
and vibration analysis of composite plates using a first-order theory. Compos. Struct. 89(3),
424–432 (2009)
6. Ferreira, A., Roque, C., Jorge, R.: Free vibration analysis of symmetric laminated composite
plates by FSDT and radial basis functions. Comput. Methods Appl. Mech. Eng. 194(39–41),
4265–4278 (2005)
7. Liew, K., Huang, Y., Reddy, J.: Vibration analysis of symmetrically laminated plates based
on FSDT using the moving least squares differential quadrature method. Comput. Methods
Appl. Mech. Eng. 192(19), 2203–2222 (2003)
822 D. N. Kien et al.

8. Reddy, J., Phan, N.: Stability and vibration of isotropic, orthotropic and laminated plates
according to a higher-order shear deformation theory. J. Sound Vib. 98(2), 157–170 (1985)
9. Khdeir, A., Librescu, L.: Analysis of symmetric cross-ply laminated elastic plates using a
higher-order theory: Part II—Buckling and free vibration. Compos. Struct. 9(4), 259–277
(1988)
10. Ferreira, A.J.M.: Analysis of composite plates using a layerwise theory and multiquadrics
discretization. Mech. Adv. Mater. Struct. 12(2), 99–112 (2005)
11. Thai, C.H., Nguyen-Xuan, H., Nguyen-Thanh, N., Le, T.-H., Nguyen-Thoi, T., Rabczuk, T.:
Static, free vibration, and buckling analysis of laminated composite Reissner-Mindlin plates
using NURBS-based isogeometric approach. Int. J. Numer. Meth. Eng. 91(6), 571–603 (2012)
12. Kien, D.N., Zhuang, X.: Radial basis function based finite element method: formulation and
applications. Eng. Struct. (Submitted.) (2022)
13. Onate, E.: Structural Analysis with the Finite Element Method. Linear Statics. Volume 2:
Beams, plates and shells. Springer (2013)
14. Srinivas, S.: A refined analysis of composite laminates. J. Sound Vib. 30(4), 495–507 (1973)
15. Ferreira, A., Fasshauer, G.: Computation of natural frequencies of shear deformable beams
and plates by an RBF-pseudospectral method. Comput. Methods Appl. Mech. Eng. 196(1–3),
134–146 (2006)
16. Pandya, B., Kant, T.: Higher-order shear deformable theories for flexure of sandwich plates—
Finite element evaluations. Int. J. Solids Struct. 24(12), 1267–1286 (1988)
17. Ferreira, A.: A formulation of the multiquadric radial basis function method for the analysis
of laminated composite plates. Compos. Struct. 59(3), 385–392 (2003)
18. Nguyen-Van, H., Mai-Duy, N., Tran-Cong, T.: Free vibration analysis of laminated plate/shell
structures based on FSDT with a stabilized nodal-integrated quadrilateral element. J. Sound
Vib. 313(1–2), 205–223 (2008)
19. Kien, D.N., Zhuang, X.: J. Zhejiang Univ,-SCIENCE A 22(8), 609–620 (2021). https://doi.
org/10.1631/jzus.A2000380
20. Reddy, J.N., Khdeir, A.A.: Buckling and vibration of laminated composite plates using various
plate theories. AIAA J. 27(12), 1808–1817 (1989)
21. Nguyen-Van, H., Mai-Duy, N., Karunasena, W., Tran-Cong, T.: Buckling and vibration analy-
sis of laminated composite plate/shell structures via a smoothed quadrilateral flat shell element
with in-plane rotations. Comput. Struct. 89(7–8), 612–625 (2011)
22. Huang, Y., Li, Q.: Bending and buckling analysis of antisymmetric laminates using the moving
least square differential quadrature method. Comput. Methods Appl. Mech. Eng. 193(33–35),
3471–3492 (2004)
23. Fares, M., Zenkour, A.: Buckling and free vibration of non-homogeneous composite cross-ply
laminated plates with various plate theories. Compos. Struct. 44(4), 279–287 (1999)
 Thermal Analysis of Drum Brake Mechanism
for a Heavy Truck

Bui Van Cuong(B) , Do Van Quan, Le Van Quynh, and Nguyen Thanh Cong

Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
cuongbui@tnut.edu.vn

Abstract. The temperature of the brake mechanism greatly affects the braking
performance as well as the safety of vehicle movement. In order to evaluate the heat
transfer process of drum-brake mechanism, a model of the drum brake mechanism
of heavy truck was built by Inventor software. The file of CAD model is then saved
as an IGS for import into the CAE environment. The finite element model of the
drum was established by Ansys software to analyze the heat of the brake drum
during the vehicle braking process. The analysis results show that the temperature
distribution on the drum is highly dependent on the braking process. Especially,
the values of the highest and lowest temperatures on brake drum are 333 °K and
295 °K when vehicle brakes on a long slope.

Keywords: Brake drum · Heavy truck · CAD · CAE · Heat

1 Introduction
Heavy vehicles service brakes are subject to severe loading conditions that can exceed
the design conditions of safe operating performance in critical situations. The thermal
loads of brake drums, when subjected to severe service condition were evaluated using
a FEA simulation [1]. A transient numerical heat transfer model was developed for the
purpose of estimating the time-dependent temperature distribution in a heavy duty drum
brake system [2]. The finite element model of the brake drum was established by using
ANSYS software [3]. A thermal analysis of a sub-set drum/shoe brake lining of a braking
system in a rear drum brake for a lightweight passenger vehicle was proposed to analyze
the thermal properties of the brake shoe lining [4]. A finite difference numerical model
was proposed to investigate the effects of wall thickness and thermal conductivity on
the temperature distribution in heavy duty drums during simulated braking conditions
[2]. The heat of heavy truck brake drum was proposed to investigate the thermal loads
of brake drums, when subjected to severe service conditions, in order to allow further
analyses, mainly related to their dimensional and form variation [1]. The objective of
this paper is to analyze the heat of the drum brake mechanism during the vehicle braking
process. Firstly, a CAD model of brake drum is established using Inventor software, then
the file of CAD model is then saved as an IGS for import into the CAE environment.
Finally, the heat of the brake drum is analyzed by Ansys software.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 823–830, 2023.
https://doi.org/10.1007/978-3-031-22200-9_86
824 B. Van Cuong et al.

2 Theoretical Basis
The kinetic energy of the moving mass of the vehicle is converted into the heat energy
at the brake mechanism during the vehicle braking process. Some cases are considered
below.
Case 1: Emergency brake.
The heat balance equation in case 1 could be written as follows

tp
G v02 − v2
= mt cT + At kt Tdt (1)
g 2
0

where, G is the vehicle weight, c is the specific heat capacity of the material of the
drum; mt is the brake drum mass; T is the degree of temperature increase relative to the
surrounding environment; At is the cooling area of the brake drum; kt is the heat transfer
coefficient from brake drum to air, tp is the braking time; v0 is the speed of the vehicle
starting to brake; v is the speed of the vehicle at the end of braking.
The braking time is short in case 1, the heat is released very small in the air, Eq. (1)
could be rewritten as follows
G.(v02 − v2 )
T= (2)
2gmt c
Case 2: Braking on a long slope.
The forces acting on the vehicle in case 2 are shown in Fig. 1.

G.sin

G.cos
G
Pp P
2
Pf2

Pp 24°
1 Pf1

Fig. 1. Diagram of forces acting on the vehicle when braking on a slope

 Thermal Analysis of Drum Brake Mechanism for a Heavy Truck 825

The vehicle speed is considered to be stable during braking on a long slope, based
on Fig. 1, the force balance equation could be rewritten as Eq. (3).

G. sin α = Pp1 + Pp2 + Pf 1 + Pf 2 + Pω (3)

where, Pp = Pp1 + Pp2 is the total braking forces; Pf = Pf1 + Pf2 is the rolling resistance
forces; Pω is the air resistance force.
Equation (3) could be rewritten as follows

Pp = G. sin α − Pf − Pω (4)

The heat balance equation is defined as

Pp dS
= mt cdt + At kt vTdt (5)
427

where, S is the braking distance.

The heat balance Eq. (5) could be solved by integrating both sides with respect to t

mt c Pp
S= ln( − At kt T ) + C (6)
At kt 427

where C is the time constant.

With S = 0 and T = 0, C is determined as follows
mt c Pp
C=− ln (7)
At kt 427
Combining Eqs. (6) and (7), the degree of temperature increase relative to the
surrounding environment could be determined as
 
Pp SAt kt
T = 1−e t
m c (8)
427At kt

3 Results and Discussion

A CAD model of brake drum is established using Inventor software, as shown in Fig. 2.
And then the file of CAD model is then saved as an IGS for import into Ansys software,
as shown in Fig. 3. The heat transfer coefficient of the brake drum is 62 W/m2 .K;
The heat transfer coefficient of the drum surfaces in contact with the environment is
44 W.m−1 .K−1 ; The environment temperature is 24 °C and the temperature of brake
drum surface in contact with brake pad is 40 °C in case 1. The heat gradient and flux
vectors at brake drum are shown in Figs. 4 and 5.
The results of Figs. 4 and 5 indicate the highest and lowest values of the heat gradient
and flux vectors are 2.974 and 0.128 x 10–3 , and 129.37 and 5.566 x 10–3 .
826 B. Van Cuong et al.

Fig. 2. CAD model

Fig. 3. CAE model

The temperature distribution of brake drum is shown in Fig. 6. The results of Fig. 6
indicate that the values of the highest and lowest temperatures on brake drum are 313
°K and 295 °K.
Similarly, the temperature of brake drum surface in contact with brake pad is 600 C
in case 2. The heat gradient and flux vectors at brake drum are shown in Figs. 7 and 8.
The results of Figs. 7 and 8 indicate the highest and lowest values of the heat gradient
and flux vectors are 6.278 and 0.121 x 10–4 , and 273.11 and 0.526 x 10–3 .
 Thermal Analysis of Drum Brake Mechanism for a Heavy Truck 827

Fig. 4. Heat gradient Vector in case 1

Fig. 5. Heat flux vector in case 1

Similarly, the temperature distribution of brake drum is shown in Fig. 9. The results
of Fig. 9 indicate that the values of the highest and lowest temperatures on brake drum
are 333 °K and 295 °K.
828 B. Van Cuong et al.

(a) Temperature distribution (b) Temperature distribution graph

Fig. 6. Temperature distribution of brake drum in case 1

Fig. 7. Heat gradient Vector in case 2

4 Conclusions
In this study, the heat of the drum brake mechanism of a heavy vehicle is analyzed using
CAD/CAE software. The major conclusions drawn from analysis results are: (1) The
values of the highest and lowest temperatures on brake drum are 313 °K and 295 °K in
case 1 and (2) The values of the highest and lowest temperatures on brake drum are 333
°K and 295 °K. The study results are the basis for the optimal design of the drum brake
mechanism for heavy trucks in the field of reducing the heat of the brake mechanism.
 Thermal Analysis of Drum Brake Mechanism for a Heavy Truck 829

Fig. 8. Heat flux vector in case 2

Fig. 9. Temperature distribution of brake drum in case 2

Acknowledgment. The authors wish to thank the Thai Nguyen University of Technology for
supporting this work.

References
1. Umaras, E., Barari, A., Tsuzuki, M.S.G.: Heavy vehicles brake drums—an accurate evaluation
on thermal loads in severe service conditions. Int. J. Automot. Technol. 22(2), 371–382 (2021)
830 B. Van Cuong et al.

2. Parmigiani, J.P., Ovaert, T.C.: The Transient Temperature Distribution in a Heavy Duty Brake
System During Fatigue Crack Testing. SAE Technical Paper Series (2000)
3. Yang, D., Yu, Z., Zhang, L., Cheng, W.: Modal Analysis of Automobile Brake Drum Based on
ANSYS Workbench. MCEI 2017
4. Chiaroni, A.B., Silveira, Z.C.: Thermal Analysis of a Rear Drum Brake for Lightweight
Passenger Vehicles. SAE Technical Paper Series, 2014
 Research on Building the System
of the Identification and Detection Traffic
Violations Vehicles at Vietnamese Intersections

Trinh Luong Mien(B) and Vu Van Duy

University of Transport and Communications, Hanoi, Vietnam

mientl@utc.edu.vn

Abstract. The article presents a solution to build a system to identify and detect
traffic violations at intersections, applying IoT technology and image processing.
The system uses traffic cameras to collect input data for software programs to
identify and detect traffic violations on Apps, Websites. The article focuses on
building an algorithm to recognize vehicle number plates, detect traffic viola-
tions, and warn of collisions of vehicles participating in traffic. Thereby helping
managers effectively respond and handle traffic situations in real time.

Keywords: ITS · IoT · AI · YOLO · Deep SORT · Image processing

1 Introduction
The Vietnamese uses many motorbikes and many types of transportation, so the traffic
flow in Vietnam is very complicated. Public transport in Vietnam is not efficient. Traffic
congestion at rush hour, especially in big cities such as Hanoi and Ho Chi Minh City, has
remained a problem for many years. Although there are traffic lights and traffic signs on
the streets in many places, a small part of people in traffic still do not have the sense of
strictly obeying the order.
Nowadays, image processing technology, internet of things has been applied in solv-
ing a number of transportation problems [1–6]. Scientific works show that the application
of image processing technology, IoT allows identification and detection of vehicle num-
ber plates [2, 3, 5, 6, 9] thereby to allow monitoring and managing vehicles. However,
almost scientific works are either packaged products (cannot be interfered with, difficult
to upgrade) or are simple simulations and tests, basic image processing algorithms, and
low efficiency [1–9].
This paper presents a system for analyzing and handling traffic situations including
image processing software to detect vehicles, analyze traffic violations and statistics
on vehicle traffic moving through intersections. Data from traffic cameras will be sent
to websites/apps to manage, operate and monitor vehicles participating in traffic. From
there, it supports traffic management, combats traffic jams, and improves people’s aware-
ness in traffic through reminders and cold penalties based on real-time data extracted
from the system.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 831–838, 2023.
https://doi.org/10.1007/978-3-031-22200-9_87
832 T. L. Mien and V. V. Duy

2 Solution for Building the System of the Identification

and Detection Traffic Violations Vehicles at Vietnamese
Intersections
The system of the identification and detection of traffic violations at intersections is
developed based on IoT technology and image processing technology, which are divided
into two parts as follows: (1). Software program to identify and detect traffic violations
at intersections, applied image processing, connected to traffic cameras; (2). The website
program provides information and warnings to users. The equipment used in the system
includes: traffic camera, IoT-based processing board, data server.
Operating principle of the system is that: The software program to identify and detect
violating traffic vehicles will connect to traffic cameras to receive image data. After that,
the software program will analyze the image, identify and detect traffic violations such
as: running red lights, exceeding the allowed speed; calculate the traffic volume through
the intersection. Useful information and data from the program to identify and detect
traffic violations are sent to the website to provide information and warnings to users.
The website interface also provides analysis and traffic charts of vehicles flowing through
the intersections.

3 Design of System Model of the Identification and Detection

Traffic Violations Vehicles at Vietnamese Intersections
3.1 System Structure Block Diagram
The structure block diagram of the system of the identification and detection of traffic
violations at Vietnam’s intersections is shown in Fig. 1.

Fig. 1. Structure of the system of the identification and detection of traffic violations at Vietnam
intersections

The proposed system consists of the following three subsystems:

 Research on Building the System of the Identification and Detection Traffic 833

(1) The hardware devices: traffic cameras; IoT-based processing board, data server.
(2) The application computer program (App): receives traffic image data from traffic
cameras, then identifies traffic vehicles, calculates traffic flow through intersections.
The resulting data will be sent to the webserver.
(3) The webserver program (Web): receives data from the App program, stores it in the
database, and interacts with the user.

3.2 The Algorithm Flowchart of the System of the Identification and Detection
Traffic Violation Vehicles

Fig. 2. The algorithm flowchart of the system

The algorithm flowchart of the system is shown in Fig. 2. After connecting the
camera successfully to the system, the camera collects video image data of the traffic
834 T. L. Mien and V. V. Duy

flow, thereby providing it to apps and computer programs to identify the means of
traffic, determine the volume and density of traffic on the street, and send data and
update information to the web server.

3.3 Design the IoT-Based Processing Board

Based on the requirements of the system, the authors designed the principle circuit, the
printed circuit of the processing board as shown in Fig. 3.

Fig. 3. The printed circuit board

3.4 Development of the Application Computer Program

The application computer program (App) is developed based on YOLOv5 and Deep
SORT with outstanding advantages in object detection, tracking and high accuracy.
The authors have developed the application program to meet the requirements:
identification of license plates of vehicles participating in traffic through intersections;
identification and detection of traffic violations at Vietnam’s intersections (Fig. 4).
 Research on Building the System of the Identification and Detection Traffic 835

Fig. 4. The App’s interface for vehicle identification and monitoring

3.5 Development of the Webserver Program

Webserver program (Web) is developed to manage and monitor vehicles. Web interface of
the system (Fig. 5) has basic functions such as system login screen, system management
interface, violation vehicle management interface, interface extracting images of vehicles
violating traffic, the vehicle statistical interface through the intersection, the screen of
the graph of the number of each vehicle passing through the intersection.

Fig. 5. System management website interface

836 T. L. Mien and V. V. Duy

4 Implementation and Results

The system was experimentally conducted with the actual data of the intersection of
the Nguyen Trai-Khuat Duy Tien str., Me tri str., Thai Ha str., in Hanoi, Vietnam. The
IoT-based processing board is shown in Fig. 6.

Fig. 6. The IoT-based proccessing board

The input data will be collected continuously every 30 s and every 20 min will give
an average result of the media density at the intersection.
Some results obtained when testing with actual data at the intersection are as follows
(Figs. 7, 8 and 9).

Fig. 7. Management interface for traffic violation vehicles

 Research on Building the System of the Identification and Detection Traffic 837

Fig. 8. Web interface that extracts videos and images of traffic violation vehicles

Fig. 9. Statistics of the number of each vehicle on an intersection

5 Conclusion
The system of identification and detection of traffic violations at Vietnamese intersec-
tions has been researched and developed in this paper. The research results obtained
initially show that the solution to build the system proposed in the paper is reasonable,
allowing identification, detection, traffic violation and measurement of traffic density
passing through intersection. However, this work also encountered certain difficulties
in the complicated traffic situation, location setting and server structure, besides the
geographical and urban factors in Vietnam, there are still many limitations in terms of
infrastructure should face many difficulties in the arrangement of camera systems.
In the future, the research team will continue to develop software to recognize traffic
collisions and arising traffic situations to ensure traffic safety, avoid traffic jams and take
emergency measures in time. Furthermore, traffic density statistics in locations with
traffic lights will be used for future project development.

References
1. Solawetz, J.: YOLOv5 New Version—Improvements and Evaluation, Rob flow (2020)
2. Cao, J.: Research on urban intelligent traffic monitoring system based on video image
Processing. Int. J. Signal Pro. Image Pro. Pattern Recognit. (2016)
838 T. L. Mien and V. V. Duy

3. Garg, A.: A novel approach to improve intelligent traffic management system. In: International
Conference Green Computing and Internet of Things on IEEE (2015)
4. Xu, X.P., Yin, Y.Y., Yan, J.H.: Research report on mass rule parallel processing technology in
intelligent traffic system. J. Parallel Cloud Comput. Res. 3 (2015)
5. Al-Sakran, H.O.: Intelligent traffic information system based on integration of internet of things
and agent technology. J. Int. J. Adv. Comput. Sci. Appl. 6(2) (2015)
6. Wen, X., Shao, L., Fang, W.: Efficient feature selection and classification for vehicle detection.
J. Circuits Syst. Video Technol. IEEE Trans. 25(3) (2015)
7. Athmaraman, N.: Adaptive predictive traffic timer control algorithm. In: Proceedings of the
2005 Mid-Continent Transportation Research Symposium, Ames, Iowa (2005)
8. Hawi, R.: Techniques for smart traffic control: an in-depth review. Int. J. Comput. Appl.
Technol. Res. 4(7), 566–573 (2015)
9. Eom, M., Kim, B.-I.: The traffic signal control problem for intersections: a review. Eur. Transp.
Res. Rev. 12(1), 1–20 (2020). https://doi.org/10.1186/s12544-020-00440-8
 Modified Skyhook Control for Semi-active
Electric Vehicle Suspension

Le Van Quynh(B) , Bui Van Cuong, Hoang Anh Tan, and Canh Chi Huan

Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
lequynh@tnut.edu.vn

Abstract. In order to improve comfort and ride quality of electric vehicle (EV), a
modified Skyhook controller is used for controlling the damping coefficient of EV
suspension system. A three-degree-of-freedom EV dynamic model is established
under two excitation sources such as road surface excitation and in-wheel electric
motor (IWM) excitation to evaluate the controller’s efficiency. The weighted root-
mean-square (r.m.s.) acceleration of the vertical vehicle body (aw ) according to the
international standard ISO 2631–1 (1997) is selected as an objective function. The
study results indicate that the design controller for the semi-active EV suspension
system has significantly improved the vehicle ride comfort. Especially, the aw
value with modified Skyhook controller for semi-active EV suspension system
reduces by 15.8% in comparison with passive suspension system when the vehicle
moves on ISO class B road surface at vehicle speed of 60 km/h and full load.

Keywords: Electric vehicles · In-wheel electric motor · Semi-active suspension

system · Skyhook · Ride comfort

1 Introduction
Suspension systems of electric vehicle (EV) are one of the important parts to improve
vehicle ride comfort. In order to improve EV ride comfort, the Macpherson suspension
of an electric vehicle was taken as the research object, the suspension model was estab-
lished by ADAMS/Car to analyze and optimize the key parameters variation of camber
angle, toe angle, caster angle, kingpin inclination angle and scrub radius [1]. An active
suspension was proposed to improve the EV conversion’s ride comfort performance
using a validated 7 degrees of freedom of vehicle’s ride model [2, 4]. The semi active
suspension system was proposed to improve the EV conversion ride comfort perfor-
mance using a validated 7 degrees of freedom of ride model of a passenger vehicle [3].
In-wheel electric motor (IWM) suspension system and its parameters were analyzed and
evaluated its influence on EV ride comfort using a dynamic model of quarter vehicle
[5, 6]. Three control strategies of linear quadratic control (LQ), improved LQ (ILQ)
and wheelbase preview LQ (WLQ) were proposed to control EV suspension control
system using an improved half-vehicle model. An in-wheel vibration absorber for in-
wheel-motor electric vehicles (IWM EVs) was proposed and optimized via a particle

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 839–845, 2023.
https://doi.org/10.1007/978-3-031-22200-9_88
840 L. Van Quynh et al.

swarm optimization (IPSO) and then a linear quadratic regulator (LQR) algorithm was
developed to control suspension damper to improve vehicle ride comfort [8].
The major goal of this paper is to propose a modified Skyhook controller for control-
ling the damping coefficient of EV suspension system using a dynamic model of quarter
vehicle with the combination of IWM vertical exciting force and road surface rough-
ness excitations. The performance of the proposed controller is analyzed and evaluated
according to the international standard ISO 2631–1 (1997). The ride performance of the
proposed controller is compared to Skyhook controller and passive suspension system
under different operating conditions.

2 Electric Vehicle Dynamic Model

Electric vehicle dynamic model: A quarter-vehicle model of an in-wheel motor (IWM)
electric vehicle with three degrees of freedom is established under the combination of
IWM vertical exciting force and road surface roughness excitations, as shown in Fig. 1.
In Fig. 1, mb , mm and ma are the sprung mass of EV body, IWM mass, and EV unsprung
mass, respectively; k t and ct are the stiffness and damping coefficients of the tire; k and
c are the stiffness and damping coefficients of the EV suspension system; k m and cm are
the IWM suspension system; za , zb and zm are the vertical displacements of the axle,
vehicle body and IWM mass, respectively; q is road surface excitation.

csky
zb mb zb mb zb mb
zm mm
k
zm mm zm mm
cm c
cm c k cm k csemi
km km km
za ma za ma za ma
ct kt ct kt ct kt
q q q

a) Electric Vehicle model b) Skyhook model c) Modified Skyhook

model

Fig. 1. One-quarter electric vehicle dynamic model

The equations of vehicle motion of Fig. 1:

The equations of vehicle motion of Fig. 1a are written as follows
⎧
⎪ mb z̈b = −[k(zb − za ) + c(żb − ża )]
⎪
⎪
⎨ m z̈ = F − [k (z − z ) + c (ż − ż )]
m m mz m m a m m a
(1)
⎪
⎪ ma z̈a = [k(zb − za ) + c(żb − ża )] + [km (zm − za ) + cm (żm − ża )]
⎪
⎩
−[kt (za − q) + ct (ża − q̇)]
 Modified Skyhook Control for Semi-active 841

The equations of vehicle motion of Fig. 1b are written as follows

⎧
⎪ mb z̈b = −[k(zb − za ) + c(żb − ża )] - Fsky
⎪
⎪
⎨ m z̈ = F − [k (z − z ) + c (ż − ż )]
m m mz m m a m m a
(2)
⎪
⎪ ma z̈a = [k(zb − za ) + c(żb − ża )] + [km (zm − za ) + cm (żm − ża )]
⎪
⎩
−[kt (za − q) + ct (ża − q̇)] + Fsky

The equations of vehicle motion of Fig. 1c are written as follows

⎧
⎪ mb z̈b = −[k(zb − za ) + csemi (żb − ża )]
⎪
⎪
⎨ m z̈ = F − [k (z − z ) + c (ż − ż )]
m m mz m m a m m a
(3)
⎪ ma z̈a = [k(zb − za ) + csemi (żb − ża )] + [km (zm − za ) + cm (żm − ża )]
⎪
⎪
⎩
−[kt (za − q) + ct (ża − q̇)]

where,

Fsky = csky żb żb (żb − ża ) ≥ 0
(4)
Fsky = 0 żb (żb − ża ) < 0
⎧   
⎪
⎨ csemi csky żb
= max cmin , min , cmin żb (żb − ża ) ≥ 0
żb − ża (5)
⎪
⎩c
semi = cmin żb (żb − ża ) < 0

Combination of IWM vertical exciting force and road surface roughness excitations:
The excitation force of the in-wheel motor [9] can be rewritten as:

Fmz = ms eωR2 sin ωR t (6)

where, ms is the total mass of the tire, the rim and the motor rotor; e is the eccentricity
of the rotor; ωR is the angular velocity of the rotor.
Road surface roughness excitation [6, 10, 11, 14, 15]: The road surface roughness
is usually assumed to be a zero-mean stationary Gaussian random process and can be
generated through an inverse Fourier transformation based on a power spectral density
(PSD) function. The time domain excitation of the uneven road surface is generated as
the sum of a series of harmonics:
N

q(t) = 2Gq (fmid −k )fk sin(2π fmid−k t + ϕk ) (7)
k=1

where, Gq (fmid −k ) is power spectrum according to ISO 8608:2016 [12], m2 /Hz; δf is

the frequency range, Hz; t is time, s; ϕk is the random phase uniformly distributed from
0 to 2π.
842 L. Van Quynh et al.

3 Results and Discussion

In order to evaluate the ride performance of the modified Skyhook controller for a
semi-active EV suspension system, Matlab/Simulink software is used to simulate the
equations of vehicle motion of Fig. 1 with the parameters in the reference. The time-
domain responses of body vertical acceleration with semi-active suspension system with
modified Skyhook controller compared to semi-active suspension system with Skyhook
controller and passive suspension system are shown in Fig. 2 when vehicle moves on
the ISO class B road surface at the vehicle speed of 60 km/h and full load. From the
results of Fig. 2, we show that the performance of modified Skyhook controller for semi-
active suspension system of EV is better than semi-active suspension system and passive
suspension system towards improving the vehicle ride comfort.

1.5
Electric Vehicle model
1 Skyhook model
Modified Skyhook model
0.5
a (m.s-2)

0
w

-0.5

-1

-1.5
0 5 10 15 20
Time(s)
Fig. 2. The time-domain responses of body vertical acceleration with three types of suspension
systems of EV.

The time-domain method can be applied to evaluate the vehicle ride comfort accord-
ing to ISO 2631–1 (1997) [13, 16], in this study, the vibration evaluation based on the
basic evaluation methods including measurements of the weighted root-mean-square
(r.m.s) acceleration defined as
⎡ ⎤1/2
T
1
aw = ⎣ az2 (t)dt ⎦ (8)
T
0

where, az (t) is the weighted acceleration (translational and rotational) as a function of

time, m/s2 ; T is the duration of the measurement, s.
From the results of Fig. 2, the values of root-mean-square (r.m.s) of acceleration
responses of the vertical vehicle body are respectively determined through Eq. (8) such as
aw = 0.3147 m/s2 with semi-active suspension system with modified Skyhook controller,
aw = 0.3353 m/s2 semi-active suspension system with Skyhook controller and aw =
0.3738 m/s2 passive suspension system. The aw value with modified Skyhook controller
reduces by 15.81% in comparison with passive suspension system and the aw value with
 Modified Skyhook Control for Semi-active 843

Skyhook controller reduces by 10.30% in comparison with passive suspension system.

It shows that the performance of semi-active suspension system with modified Skyhook
controller is better than the other two suspension systems towards improving the vehicle
ride comfort.
In order to continue evaluating the performance of semi-active suspension system
with modified Skyhook controller compared to semi-active suspension system with Sky-
hook controller and passive suspension system, five road surface conditions from ISO
class A road surface to ISO class E road surface according to ISO 8068 (2016) are
selected for evaluation when vehicle moves at speed of 60 km/h and full load. The aw
values with three types of suspension system when EV moves under the above conditions
based on the International Standard ISO 2631–1: 1997 which are determined through
Eq. (8), as shown in Fig. 3.

0.4
0.3
0.2
0.1
0
A B C D E

Electric Vehicle model Skyhook model

Modified Skyhook model

Fig. 3. aw values of with three types of suspension systems of EV under variable road surface
conditions

From the results of Fig. 3, it shows that the aw values with three types of suspension
system increase height fast when EV moves on poor and very poor road surface condi-
tions, especially ISO class D road surface and ISO class E road surface. The aw values
with three types of suspension system respectively increase when EV moves on from
ISO class A road surface to ISO class E road surface which make the vehicle comfort
worse, when the road surface conditions become worse. The aw values with modified
Skyhook controller respectively reduce 15.87%, 15.81%, 16.91%, 15.72%, and 15.73%
in comparison with passive suspension system when EV moves on from ISO class A
road surface to ISO class E road surface. It shows that EV ride comfort with modified
Skyhook controller for semi-active suspension system is significantly improved in com-
parison with the other two suspension system when EV moves on the different road
conditions.

4 Conclusions
In this paper, a modified Skyhook controller is proposed to control the damping coef-
ficient of semi-active suspension system of EV using a three-degree-of-freedom EV
844 L. Van Quynh et al.

dynamic model under the combination of two excitation sources. The major conclusions
drawn from the analysis can be summarized as follows: (1) The aw value with modified
Skyhook controller reduces by 15.81% in comparison with passive suspension system
and the aw value with Skyhook controller reduces by 10.30% in comparison with passive
suspension system, and (2) The aw values with modified Skyhook controller respectively
reduce 15.87%, 15.81%, 16.91%, 15.72%, and 15.73% in comparison with passive sus-
pension system when EV moves on from ISO class A road surface to ISO class E road
surface. The study results are the theoretical basis for optimal design and optimal control
for semi-active or active suspension system of the electric vehicles.

Acknowledgment. The authors wish to thank the Thai Nguyen University of Technology for
supporting this work.

References
1. Chen, Q., Kang, S., et al.: Research on modeling and optimization simulation analysis of
micro electric vehicle suspension. J. Vibroengineering 22(4), 923–932 (2020)
2. Abu Bakar, S.A., Masuda, R., et al.: Ride Comfort performance improvement of Electric Vehi-
cle (EV) conversion using SAS-controlled active suspension system. Jurnal Teknologi 71(2)
(2014)
3. Bakar, S.A.A., et al.: Ride comfort evaluations on electric vehicle conversion and improvement
using Magnetorheological semi active suspension system. SICE Ann. Conf. 2011, 305–311
(2011)
4. Abu Bakar, S.A., Abdul Aziz, A.: Active suspension system to improve ride comfort
performance of Electric Vehicle (EV) Conversion. Appl. Mech. Mater. 663, 208–212 (2014)
5. Tan, H.A., Cong, N.T. et al.: Effects of design parameter of in-wheel motor suspension system
on electric vehicle ride comfort. World J. Res. Rev.12(6), 29–32 (2021)
6. Quynh, L.V., Cuong, B.V., et al.: Effect of in-wheel motor suspension system on electric
vehicle ride comfort. Vibroengineering PROCEDIA 29, 148–152 (2019)
7. Cao, J.-T., Liu, H.-H., et al.: A study of electric vehicle suspension control system based on
an improved half-vehicle model. Int. J. Autom. Comput. 4(3), 236–242 (2007)
8. Liu, M., Gu, F., Zhang, Y.: Ride comfort optimization of inwheel-motor electric vehicles with
in-wheel vibration absorbers. Energies 10, 1647 (2017)
9. Yuewei, Y., Leilei, Z., Changcheng, Z.: Influence of rotor-bearing coupling vibration on
dynamic behavior of electric vehicle driven by in-wheel motor. IEEE Access 7, 63540–63549
(2019)
10. Van Quynh, L.: Influence of semi-trailer truck operating conditions on road surface
friendliness. Vibroengineering PROCEDIA 16, 67–72 (2017)
11. Tan, H.A., Quynh, L.V., et al.: Influence of damping coefficient into engine rubber mounting
system on vehicle ride comfort. Vibroengineering PROCEDIA 29, 112–117 (2019)
12. ISO 8608: Mechanical Vibration-Road Surface Profiles-Reporting of Measured Data.
International Organization for Standardization (2016)
13. ISO 2631–1: Mechanical Vibration and Shock-Evaluation of Human Exposure to Whole-
Body Vibration, Part I: General Requirements. The International Organization for Standard-
ization (1997)
14. Van Quynh, L., Duy, N.T., Van Liem, N., Van Cuong, B., Long, L.X.: Optimal Design of
Cab’s Isolation System for a Single-Drum Vibratory Roller. In: Sattler, K.-U., Nguyen, D.C.,
Vu, N.P., Long, B.T., Puta, H. (eds.) ICERA 2020. LNNS, vol. 178, pp. 619–627. Springer,
Cham (2021). https://doi.org/10.1007/978-3-030-64719-3_68
 Modified Skyhook Control for Semi-active 845

15. Le, V.Q., Nguyen, V.L.: Ride comfort performance of heavy truck with three control cases of
semi-active isolation systems. Vibroengineering PROCEDIA 22, 93–98 (2019)
16. Van Quynh, L., Viet Ha, D., et al.: Improvement of ride comfort quality for an earth-moving
machinery with semi-active cab isolation system. E3S Web Conf. 304, 02012 (2021)
 Ride Comfort Evaluation for a Wheel Loader
with Cab’s Hydraulic Isolation System

Canh Chi Huan1 , Dang Viet Ha2 , Le Anh Vu3 , Tran Van Thoan3 ,
and Le Van Quynh1(B)
1 Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
lequynh@tnut.edu.vn
2 Vietnam Register, 12059 Hanoi, Vietnam
3 Faculty of Automobile Engineering, Hung Yen University of Technology and Education,

Hung Yen, Vietnam

Abstract. In order to evaluate a wheel loader ride comfort using hydraulic iso-
lation system (HIS) of cab with the orifice and the annular orifice, a half-vehicle
dynamic model of a wheel loader is established under different driving conditions.
Time domain and frequency domain acceleration responses of the vertical driver’s
seat and cab pitching angle are selected as the objective functions to analyze and
evaluate vehicle ride comfort. The study results show that the amplitude values
of the PSD acceleration responses of the vertical driver’s seat with 1.5c0 value
of HIS respectively reduce in comparison with 0.5c0 value of HIS at the low fre-
quency region from 2.0 Hz to 20 Hz. Finally, the different operating conditions are
investigated and evaluated their effects on vehicle ride comfort with HIS, and the
evaluation results indicate that the values of the root mean square (RMS) acceler-
ation responses of the vertical driver’s seat (aws ) and cab pitching angle (awphi )
change greatly when operating conditions get worse, which means that operating
conditions greatly affect vehicle ride comfort.

Keywords: Wheel loader · Cab · Hydraulic Isolation System (HIS) · Dynamic

model · Ride comfort

1 Introduction

Wheel loader is off-road vehicle and works on uneven terrain, unexpected banks or
steep slopes. In order to reduce the adverse effects on the operator and vehicle durability
caused by vibrations, the researchers have been constantly looking for ways to improve
the design of the vehicle cab and driver’s seat suspension system. A five degrees of
freedom (5DOF) dynamic model of a tractor front-end loader system was established
for vehicle ride comfort analysis and the simulation results were verified by the experi-
mental ride comfort results [1]. Some results for the analysis of the influence of design
parameters of cab’s isolation system on construction machine ride comfort under differ-
ent conditions using dynamic models such as the half-vehicle and full-vehicle dynamic

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 846–854, 2023.
https://doi.org/10.1007/978-3-031-22200-9_89
 Ride Comfort Evaluation for a Wheel Loader with Cab’s 847

models were presented in the references [2–5]. In order to improve the cab of con-
struction machinery and reduce the harm of vibration to the driver, the optimal design
parameters of cab’s main and auxiliary isolation systems were found by using various
methods such as genetic algorithm, a multi-objective genetic algorithm, etc. [6–10]. In
addition, the cab hydraulic isolation system is being widely applied to cab isolation
system for construction machinery. The lumped parameter model of passive hydraulic
damping rubber mount (PHDRM) considering the nonlinearity of multi-inertial was
tracks analyzed and evaluated using both numerical and experimental methods [11]. A
novel liquid-filled isolator with an annular orifice was investigated at low frequency by
using experimental and analytical methods [12]. In recent years, some study results have
shown that the suspension system was applied to wheel loader to improve the vehicle
ride comfort. A multibody simulation model of wheel loader was used to investigate the
possibility to reduce driver vibrations by introducing suspended wheel axles [13]. The
multibody model of the wheel loader with hydropneumatic suspension was developed
by RecurDyn in co-simulation with MATLAB/Simulink to find out the design param-
eters of the hydropneumatic suspension to obtain the minimum root mean square of
vertical accelerations under different driving conditions [14]. Three different layouts of
hydropneumatic suspension, were installed on the wheel loader to improve ride com-
fort using the multibody model [15]. The main aim of this study is to evaluate the ride
performance of a wheel loader with the hydraulic isolation system (HIS) of cab with
the orifice and the annular orifice of wheel loaders under different driving conditions.
First, a nonlinear dynamical model of hydraulic isolation with the piecewise damping
characteristic is set up to determine the vertical nonlinear force of HIS which connects
to a half-vehicle dynamic model of a wheel loader under larger amplitude excitation of
road surface roughness. Time domain and frequency domain acceleration responses of
the vertical driver’s seat and cab pitching angle are selected for the objective function
to evaluate the ride performance of a wheel loader with HIS under different driving
conditions.

2 Nonlinear Dynamical Model of Hydraulic Isolation System (HIS)

A schematic of cab’s hydraulic isolation with the orifice and the annular orifice is pro-
posed by the references [16, 18] in Fig. 1. A nonlinear dynamic model of cab’s hydraulic
isolation based on the schematic of Fig. 1 is set up with the following assumptions: The
liquid is a viscous fluid, the elastic properties of silicone oil are not considered and
the gap is not closed, as shown in Fig. 2, where, mc and mb are vehicle body and cab
masses, respectively, p1 and p2 are the upper liquid chamber and the lower liquid cham-
ber, respectively, p0 , A0 , ż0 and pa , Aa , ża are the pressure loss at the orifice, the
effective area of the orifice, the average flow velocity in the orifice and the pressure loss
at the annular gap, the effective area of the annular orifice, the average flow velocities in
the annular orifice, respectively, Ac and Ad are the effective area of the chamber and the
effective area of the damping plate, zc and zb are the vertical displacements of cab mass
and vehicle body mass, respectively, and Fc is the total vertical forces of cab’s hydraulic
isolation.
848 C. C. Huan et al.

Cab
Rubber Part Upper chamber
zc
Orifice
Vehicle Frame
zb

Annular orifice
Liquid
DampingPlate
Lower chamber

Fig. 1. A schematic of cab’s hydraulic isolation

Fc
zc
mc Cab
Annular orifice
pa, Aa, za p1
kr cr Orifice
p0, A0, z0 p2,Ad, Ac
Damping plate
zb
mb Vehicle body
Fc
Fig. 2. Nonlinear dynamic model of cab’s hydraulic isolation

Based on references [14, 16] and Fig. 2, the total vertical forces of cab’s hydraulic
isolation could be achieved by

Fc = Fr + Ff
⎧ 11
⎨ cf ż + cf |ż|ż with ż > 0
12
⎪
= kr z + cr ż + c0f |ż|ż +  2 (1)
⎪ żc
⎩ cf21 ż + cf22 |ż|ż + cg ż with ż ≥ 0
δ

where, c0f , cf, and cg are the constants which are defined in the reference [16].

3 Half-Vehicle Dynamic Model of a Wheel Loader

The structure of 2-axle wheel loader includes a rear-engine articulated tractor with a
front mounted lifting implement. In order to evaluate a wheel loader ride comfort using
hydraulic isolation system (HIS) of cab with the orifice and the annular orifice, based on
the schematic of the elastic systems of a wheel loader as in Fig. 3, a half-vehicle dynamic
model of a wheel loader is established, as shown in Fig. 4. In this model, ms , mc , mb and
mf are driver’s seat mass, cab mass, vehicle body mass; and Ic and Ib —the mass moment
 Ride Comfort Evaluation for a Wheel Loader with Cab’s 849

of inertia about the pitch axis of the vehicle body and cab respectively; ks and cs —
the stiffness and damping coefficients of driver’s seat suspension system respectively;
kti and cti —thoses of tires respectively; zs , zc and zb —the vertical displacements of
driver’s seat mass, cab mass and vehicle body mass respectively; ϕc and ϕb —the pitch
angle displacements of cab mass and vehicle body mass, and the mass of a front mounted
lifting implement; qti —road surface roughness excitations; Fci —the vertical forces of
cab’s hydraulic isolation system; Fg —an equivalent alternative force of a front mounted
lifting implement, lk —the distances (I = 1 ÷ 2, k = 1 ÷ 8).

Driver'
seat
ms
mc
mb Front
Rear Cab m
frame f
frame

Fig. 3. A structure of 2-axle wheel loader

l8
zs
ms
ks cs
zc
mc, Ic c
l7 l6 Fg
Fc2 zb Fc1
m b, I b b
l5
l4 l3 ct1
kt2 kt1
ct2 qt1
qt2
l2 l1
Fig. 4. Half-vehicle dynamic model of a wheel loader
850 C. C. Huan et al.

The equations of vehicle motion: The equations of motion for a wheel loader using
Newton’s second law of motion are written in two operating conditions below. From
Fig. 4, the motion equations of vehicle mass are written as follows:
⎧
⎪
⎪ ms z̈s = −Fs
⎪
⎪
⎪
⎪ 
i=2
⎪
⎪
⎪
⎪ m z̈ = F − Fci
⎪
⎪
c c s
⎪
⎪ i=1
⎪
⎪
⎪
⎪ 
i=2
⎪
⎪
⎨ Ic ϕ̈c = (−1)i+1 Fci li+5 − Ft l8
(2)
⎪
⎪
i=1
⎪
⎪ 
i=2 
i=2
⎪
⎪
⎪
⎪ mb z̈b = Fci − Fti + Fg sin α
⎪
⎪
⎪
⎪
⎪
⎪
i=1 i=1
⎪
⎪ 
i=2 
i=2
⎪
⎪
⎪
⎪ I ϕ̈ = (−1) i+1
F l + (−1)i Fci li+2 − Fg sinα.l5
⎩ b b ti i
i=1 i=1

where, Fti , Fci , Fs and Fg are the vertical forces of tires, cab’s hydraulic isolation system
and the equivalent alternative force of a front mounted lifting implement, respectively.
Vibration excitations for a wheel loader: Low frequency vibrations in vehicles are
usually caused by road surface roughness which plays an important role in analyzing
driver ride comfort. It is assumed to be a zero-mean stationary Gaussian random process
[19–21] which can be generated through an inverse Fourier transformation

N

2vn20 Gq (n0 ) 
q(t) =  f . cos 2π fmid ,i t + ϕi (3)
2
fmid ,i
i=1

where, Gq (n0 ) is the road roughness coefficient which is defined for typical road classes
from A (Gq (n0 ) = 4 × 10–6 m3 ) to H (Gq (n0 ) = 16384 × 10–6 m3 ) according to ISO
8608(2016) [21], n0 is a reference spatial frequency; v is the speed of vehicle; ϕi is
an uniform probabilistic distribution within the 0 − 2π range; fmid,i is the temporal
frequencies of the road roughness.

4 Results and Discussion

The nonlinear equation of HIS Eq. (1) is written by a program in Matlab software
environment which is connected to the simulation blocks describing the equations of
motion Eq. (2) in Matlab/simulink software environment under road surface random
excitation function Eq. (3) and with the parameters in the reference [24]. The results of
the time domain and power spectral density (PSD) acceleration responses of the vertical
driver’s seat with hydraulic isolation system (HIS) when vehicle moves on ISO class
C road surface (Gq(n0) = 64 × 10–6 m3− Average condition) at the vehicle speed of
15km/h and full load are shown in Fig. 5.
 Ride Comfort Evaluation for a Wheel Loader with Cab’s 851

4 1

PSD as/ [(m/s)2/Hz]

X: 6.199
as/(m/s2) 2 Y: 0.8355

0 0.5

-2 X: 4.399 X: 11.8 X: 27.2 X: 29

Y: 0.06929 Y: 0.08023 Y: 0.06595 Y: 0.05694

-4 0
0 10 20 30 0 10 20 30
Time/s Frequency/Hz
(a) Time domain acceleraƟon response (b) PSD acceleraƟon response
Fig. 5. The results of the time domain and power spectral density (PSD) acceleration responses
of the vertical driver’s seat

From the results of Fig. 5a, we show that the values of root-mean-square (RMS)
of acceleration responses of the vertical driver’s seat (aws) are respectively determined
through Eq. (4) as aws = 0.8428 m/s2 , with HIS of cab. This result, compared with
Table 1, shows that human may feel uncomfortable. From the results of Fig. 5b, we
show that the resonance amplitude peaks of PSD acceleration response of the vertical
driver’s seat appear at frequencies such as 4.399 Hz, 6.199 Hz, 11.8 Hz, According to
the International Standard ISO 2631 [23], the resulting motions affect the ride quality
perceived by the occupants and their comfort as at these frequencies humans are most
sensitive to vertical vibrations from 4 Hz to 10 Hz.
The root mean square (RMS) acceleration responses according to ISO standard
2631-1:1997 [23] are chosen as the objective functions which are determined by the
formula:
⎡ ⎤ 21
T

aw = ⎣1 T a2 (t)dt ⎦ (4)
0

where a(t) is the weighted acceleration (translational and rotational) as a function of time
in m/s2 ; T is the duration of the measurements in s.
The values of c0 change: Three c0 values from 0.5c0 to 1.5 c0 are selected for
evaluation when vehicle moves on ISO class C road surface at the vehicle speed of
15 km/h and full load. The results of the time domain and power spectral density (PSD)
acceleration responses of the vertical driver’s seat with three c0 values of HIS are shown
in Fig. 6.
From the results of Fig. 6a, we show that the peak amplitude values of as decrease
when c0 values of HIS increase. The result reflects that the amplitude values of the PSD
acceleration responses of the vertical driver’s seat with 1.5c0 value of HIS respectively
reduce in comparison with 0.5c0 value of HIS at the low frequency region from 2.0 Hz
to 20 Hz which is a sensitive human frequency region for human health. Especially, the
PSD value of acceleration response of the vertical driver’s seat with 1.5c0 value of HIS
reduces by 30.77% in comparison with 0.5c0 value of HIS at frequency about 6.199 Hz
(see Fig. 6b).
Different road surface conditions: Five road surface conditions from ISO class A
road surface to ISO class E road surface according to ISO 8068(1995) are selected
852 C. C. Huan et al.

5 1

PDS a /[(m/s)2/Hz]
0.5c 1.0c 1.5c 0.5c0 1.0c0 1.5c0
0 0 0 X: 6.199
X: 6.199 Y: 0.855
as/(m/s2)

Y: 0.6538

0
0.5

s
X: 11.8
-5 Y: 0.093
0 10 20 30
Time/s 0
0 10 20 30
Frequency/Hz
(a) Time domain acceleraƟon response (b) PSD acceleraƟon response
Fig. 6. The results of as and PSD as with three c0 values of HIS

for evaluation when vehicle moves at speed of 15 km/h and full load. The values of
the root mean square (RMS) acceleration responses of the vertical driver’s seat (aws )
and cab pitching angle (awphi ) with the different vehicle speed conditions based on
the International Standard ISO 2631-1: 1997 which are determined through Eq. (4), as
shown in Fig. 6. From the results of Fig. 6, we show that the aws and awphi values increase
height fast when the vehicle moves on poor and very poor road surface conditions,
especially ISO class D road surface and ISO class E road surface. The aws and awphi
values respectively increase when the vehicle moves on from ISO class A road surface
to ISO class E road surface which make the vehicle comfort worse.

3
aws/(m/s2) awphi/(rad/s2)
2
aw

0
Class A Class B Class C Class D Class E

Fig. 7. aws and awphi values with the different road surface conditions

1.5
2 2
aws/(m/s ) awphi/(rad/s )
1
aw

0.5

0
5 10 15 20 25
v/(km/h)

Fig. 8. aws and awphi values with the different vehicle speed conditions

Different vehicle speed conditions: Five vehicle speed conditions from 5 km/h to
25 km/h are selected for evaluation when vehicle moves on ISO class C road and full load.
The values of the root mean square (RMS) acceleration responses of the vertical driver’s
 Ride Comfort Evaluation for a Wheel Loader with Cab’s 853

seat (aws) and cab pitching angle (awphi) with the different vehicle speed conditions
are shown in Fig. 7. From the results of Fig. 7, we show that the aws and awphi values
increase when vehicle speed value increases which make the vehicle comfort worse
(Fig. 8).

5 Conclusions
In this paper, a wheel loader ride comfort using hydraulic isolation system (HIS) of
cab with the orifice and the annular orifice was proposed for the evaluations when
vehicle operates under the different survey conditions. The major conclusions drawn
from evaluation results are: (1) The PSD value of acceleration response of the vertical
driver’s seat with 1.5c0 value of HIS reduces by 30.77% in comparison with 0.5c0 value
of HIS at frequency about 6.199 Hz; (2) The aws and awphi values respectively increase
when the vehicle moves on from ISO class A road surface to ISO class E road surface
which make the vehicle comfort worse; and (3) The aws and awphi values increase when
vehicle speed value increases which make the vehicle comfort worse.

Acknowledgment. The authors wish to thank the Thai Nguyen University of Technology for
supporting this work.

References
1. Pavlov, N., Sokolov, E. et al.: Study of the wheel loader vibration with a developed multibody
dynamic model. In: MATEC Web of Conferences, vol. 133, p. 02007 (2017)
2. Liem N.V., Run Z.J. et al.: Vibration analysis and modeling of an off-road vibratory roller
equipped with three different cab’s isolation mounts. Shock Vib. (2018)
3. Van Quynh, L., Thao, V.T.P., et al.: Influence of design parameters of cab’s isolation system
on vibratory roller ride comfort under the deformed ground surfaces. Int. Res. J. Eng. Technol.
(IRJET) 6(6), 1974–1978 (2019)
4. Le, V.: Ride comfort analysis of vibratory roller via numerical simulation and experiment.
DEStech Transactions on Engineering and Technology Research (2017)
5. Le V.Q., Zhang J.R., et al.: Ride comfort evaluation of vibratory roller under different soil
ground. Tran. Chinese Soc. Agri. Eng. 29, 39–47 (2013)
6. Quynh, L.V., Jianrun, Z., et al.: Experimental modal analysis and optimal design of cab’s
isolation system for a single drum vibratory roller. Vibroengineering PROCEDIA 31, 52–56
(2020)
7. Quynh, L.V., Zhang, J.R., et al.: Vibration analysis and optimal design for cab’s isolation
system of vibratory roller. Adv. Mater. Res. 199(200), 936–940 (2011)
8. Quynh, L.V., Thao, V.T.P., et al.: Optimal design parameters of drum’s isolation system for a
double-drum vibratory roller. Vibroengineering PROCEDIA 31, 74–79 (2020)
9. Van Quynh, L., Duy, N.T., Van Liem, N., Van Cuong, B., Long, L.X.: Optimal design of
cab’s isolation system for a single-drum vibratory roller. In: Sattler, K.-U., Nguyen, D.C., Vu,
N.P., Long, B.T., Puta, H. (eds.) ICERA 2020. LNNS, vol. 178, pp. 619–627. Springer, Cham
(2021). https://doi.org/10.1007/978-3-030-64719-3_68
10. Le, V.Q., Nguyen, K.T.: Optimal design parameters of cab’s isolation system for vibratory
roller using a multi-objective genetic algorithm. Appl. Mech. Mater. 875, 105–112 (2018)
854 C. C. Huan et al.

11. Sun, X., Liao, X., et al.: Modeling and dynamic analysis of hydraulic damping rubber mount
for cab under larger amplitude excitation. J. Vibroengineering 23(3), 542–558 (2021)
12. Sun, X., Zhang, C., Fu, Q., et al.: Measurement and modelling for harmonic dynamic char-
acteristics of a liquid-filled isolator with a rubber element and high-viscosity silicone oil at
low frequency. Mech. Syst. Signal Process. 140, 106659 (2020)
13. Rehnberg, A., Drugge, L.: Ride comfort simulation of a wheel loader with suspended axles.
Int. J. Veh. Syst. Model. Test. 3(3), 168–188 (2008)
14. Zhao, H., Wang, G., et al.: Optimization of hydropneumatic suspension for articulated wheel
loader based on kriging model and particle swarm algorithm. Adv. Mech. Eng. 10(11),
168781401881064 (2018)
15. Zhao, H., Li, X., et al.: Research on dynamic behaviors of wheel loaders with different layout
of hydropneumatic suspension. J. Vibroengineering 19(7), 5388–5404 (2017)
16. Sun, X., Zhang, J.: Performance of earth-moving machinery cab with hydraulic isolations in
low frequency. J. Vibroengineering Control 20, 724–735 (2012)
17. Ushijima, T., Dan, T.: Nonlinear B.B.A. for Predicting Vibration of Vehicle with Hydraulic
Engine Mount. SAE Technical. Paper Series (1986)
18. Higuchi, T., Miyaki, K.: Work machine with operator’s cabin. US Patent No. 5984036 (1998)
19. Van Quynh, L., Van Cuong, B., et al.: Effects of suspension design parameters of a semi-trailer
truck on vehicle ride comfort and road surface friendliness. In: Lecture Notes in Networks
and Systems, vol. 104 (2020)
20. Van Cuong, B., Van Quynh, et al.: Influence of heavy truck operating condition on dynamic
load coefficient. In: Lecture Notes in Networks and Systems, vol. 63 (2019)
21. Dodds, C.J., Robson, J.D.: The description of road surface roughness. J. Sound Vib. 31(2),
175–183 (1973)
22. ISO 8608, Mechanical Vibration-Road Surface Profiles-Reporting of Measured Data. ISO
(2016)
23. ISO 2631-1, Mechanical Vibration and Shock-Evaluation of Human Exposure to Whole-Body
Vibration, Part I: General Requirements. ISO (1997)
24. Quyet, T.T.: Study on the ride effectiveness of hydraulic isolation system for construction
machine cab. Master of Science thesis, Thai Nguyen University of Technology, Thai Nguyen,
Vietnam (2022)
 Robust Hierarchical Sliding Mode Control
for Planar Flexible-Joint Robot

Nguyen Thi Thu Hien1 , Tran Xuan Minh2 , and Dang Danh Hoang2(B)
1 TNU - University of Information and Communication Technology, Thai Nguyen, Vietnam
2 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

hoangktd1977@tnut.edu.vn

Abstract. In this paper, a nonlinear Planar Flexible-Joint Robot system with two
degrees of freedom is discussed. Due to the nonlinearities properties and under-
actuated system, it is essential to design a compatible control scheme. Therefore,
a robust control structure relies on the hierarchical sliding mode technique. Using
Lyapunov’s theory and Hurwitz’s polynomial property, stability of the sliding
surface as well as the convergence of the state variable can be ensured. In addi-
tion, numerical simulation results using Matlab/Simscape Multibody Link plug-in
on Solidworks are also presented and show excellent performance compared to
traditional linear controllers. In particular, the results are more obvious when
considering the model uncertainty parameter.

Keywords: Quasi-physical planar flexible-joint robot · Robust controller ·

Hierarchical Sliding Mode Control (HSMC) · Lyapunov’s theory

1 Introduction
Nowadays, research to develop algorithms for underactuated systems such as Inverted
Pendulum [1, 2], Wheeled Mobile Robots [3], Crane [4, 5], etc. have been attracting
more and more attention. In particular, the Planar Flexible-Joint Robot 2-DOF is a typical
underactuated system because only the first link can be directly controlled. Throughout
the years, there are a variety of works in relation to the Planar Flexible-Joint Robot,
which is considered to be a challenging task.
To control both links of this object, an energy approach and the passivity properties
are presented in the study [6]. However, the author has not considered the robustness
of the two-link robot with uncertainty elements. In [7], a new approach with great
performance is based on an equivalent-input-disturbance (EID) method to design a con-
troller that makes the robot with two-link asymptotically track the desired trajectory. In
addition, robust controllers are also considered an effective approach for underactuated
Flexible Joint Robot as in studies [8–13]. A practical robust controller which has simple
structure, high tracking performance, and more easy tuning factor using the nonlinear
H-infinity control in [8]. By a different approach, a robust controller is created by solving
the state-dependent Riccati equation for uncertain robot in [9]. It is noticeable that the
sliding mode control method is considered a typical robust controller and it is widely

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 855–863, 2023.
https://doi.org/10.1007/978-3-031-22200-9_90
856 N. T. T. Hien et al.

applied to robot manipulators. In [11], Conventional Sliding Mode Control (SMC) and
Quasi-Sliding Mode Control are presented, compared tracking performance, and sta-
bilized analysis for sliding mode surface. With the self-learning ability of radial basis
function neural network and SMC technique, a controller has the advantage of tuning
the parameter mentioned in the study [12]. In the study [13], a terminal sliding mode
control (TSMC) is proposed effectiveness in dealing with the lumped disturbance of the
system and ensuring all output convergence with fine-time. To avoid the measurement
of high order signal, a dynamic surface controller (DSC) is also mentioned in the study
[10].
Relied on the analysis of previous studies mentioned above, a Hierarchical sliding
mode control structure is proposed as well as demonstrate the output stability according to
the zero-dynamic method in this paper. The rest of this paper is presented in four sections.
The following part theoretically introduces model for Planar Flexible-Joint Robot that is
served for the design. In Sect. 3, a cascade control structure based on a sliding controller
and a proof of output stability is presented. The results of numerical simulation using
Matlab/Simscape Multibody Link plug-in on Solidworks and comparison of tracking
performance are given in Sect. 4. Finally, conclusions are made to show the achievements
of this work and the direction of development in the future.

2 Mathematical Model
In Fig. 1, the planar flexible-joint robot with two degrees of freedom consists of two
main elements: link 1 and link 2. The two output variables corresponding to the two links
are q1 and q2 , respectively. Besides, mi , Ii , li , and ωio represent the mass, the moment
of inertia, the length, and the angular velocity of the link i (I = 1, 2), respectively.

Fig. 1. The planar flexible-joint robot model

Using the Lagrange equation and relying on study [14], the mathematical model of
the planar flexible-joint robot is obtained in matrix form as follows

M(q)q̈ + C(q, q̇)q̇ + Kq = Bu (1)

 Robust Hierarchical Sliding Mode Control 857

in which M(q), C(q, q̇), K, and B represent the inertial matrix, the matrix of Cori-
olis and centrifugal forces, the stiffness matrix of the spring, and the input matrix,
respectively. Equation is rewritten in detail as follows
     
θ1 + θ2 + 2θ3 cos q2 −(θ2 + θ3 cos q2 ) q̈1 0 (θ3 sin q2 )q̇2 q̇1
+
−(θ2 + θ3 cos q2 ) θ2 q̈2 (θ3 sin q2 )q̇1 0 q̇2
    
0 0 q1 1  
+ = τ0
0 k q2 0
(2)

The planar flexible-joint robot has only one control input to control two outputs.
Therefore, the mathematical model (2) should be decomposed into two subsystems as

m11 q̈1 + m12 q̈2 + c11 q̇1 + c12 q̇2 = τ (3)

and

m21 q̈1 + m22 q̈2 + c21 q̇1 + c22 q̇2 + kq2 = 0 (4)

By the method of substitution and mathematical calculation, the two subsystems are
given as follows

m1 q̈1 + c11 q̇1 + c12 q̇2 + k 1 q2 = τ (5)

and
m21 τ
m2 q̈2 + c21 q̇1 + c22 q̇2 + k 2 q2 = − (6)
m11
     
m12 m21 m12 c21 m12 c22
where m1 = m11 − m22 , c11 = c11 − m22 , c12 = c12 − m22 , k1 =
− mm1222k
     
m21 m12 m21 c11 m21 c12
and m2 = m22 − m11 , c21 = c21 − m11 , c22 = c22 − m11 , k 2 = k.

Remark 1 Clearly, the mathematical model in Eq. (1) is transformed into the explicit
relationship in Eqs. (5) and (6) which is necessary for the control design process.

3 Control Design
 T
To ensure the control objective for the object, HSMC is expressed so that q = q1 q2
 T
converges to the desired value qd = q1d q2d . Besides, stability of the sliding surface
as well as the convergence of the output variable can be ensured.
858 N. T. T. Hien et al.

3.1 Hierarchical Sliding Mode Control Design

First of all, a hierarchical sliding surface is defined as follows

s = ė1 + k1 e1 + k2 ė2 + k3 e2 (7)

in which e1 = (q1 − q1d ), e2 = (q2 − q2d ) are the output errors of the first and second
joints, respectively. k1 , k2 , k3 are designing control parameters. With the assumption
that the first and second derivatives of the desired value are zero, taking the derivative
of the Eq. (7):

ṡ = ë1 + k1 ė1 + k2 ë2 + k3 ė2 = q̈1 + k1 q̇1 + k2 q̈2 + k3 q̇2 (8)

Get q̈1 and q̈2 from Eqs. (5), (6) to replace Eq.
m21 τ
τ − c11 q̇1 − c12 q̇2 − k 1 q2 − m11 − c21 q̇1 − c22 q̇2 − k 2 q2
ṡ = + k1 q̇1 + k2 + k3 q̇2
m1 m2
(9)

The control signal component that helps drive the sliding surface to zero is designed
as follows
 
m11 (−m2 c11 + m1 m2 k1 − m1 k2 c21 ) m 11 −m 2 k 1 q2 − m1 k2 k 2 q2
τeq = − q̇1 −
(m11 m2 − m1 k2 m21 ) (m11 m2 − m1 k2 m21 )
m11 (−m2 c12 − m1 k2 c22 + m1 m2 k3 )
− q̇2
(m11 m2 − m1 k2 m21 )
(10)

In addition, the control signal component that helps leads the system state towards
the sliding surface is as
k4 sgn(s)
τsw = − (11)
(m11 m2 − m1 k2 m21 )

where k4 is the control parameter. Combining the two control components in Eqs. (10)
and (11), the control signal for HSMC is obtained as follows:
 
m11 (−m2 c11 + m1 m2 k1 − m1 k2 c21 ) m 11 −m k q
2 1 2 − m k k
1 2 2 2q
τ =− q̇1 −
(m11 m2 − m1 k2 m21 ) (m11 m2 − m1 k2 m21 ) (12)
m11 (−m2 c12 − m1 k2 c22 + m1 m2 k3 ) k4 sgn(s)
− q̇2 −
(m11 m2 − m1 k2 m21 ) (m11 m2 − m1 k2 m21 )
Theorem 1 Consider the nonlinear planar flexible-joint robot system with two degrees
of freedom described in (1)–(6), the control law of the robot manipulator (12) guarantee
the global asymptotic stability of the control system.
Stability analysis is proven in the following subsystem.
 Robust Hierarchical Sliding Mode Control 859

3.2 Stability Analysis

Proof of Theorem 1 In order to consider the stability of the sliding surface, a Lyapunov
function is chosen as follows.
1 2
V = s (13)
2
Substituting the control signal in (12) into the derivative of Eq. (13), we have.
V̇ = sṡ = −k4 ssgn (s) (14)

If we choose k4 ≥ 0 then V̇ ≤ 0, however, for an underactuated system, the sliding

surface stabilization does not reflect the output stability of the system. Therefore, we will
prove the stability of the system output based on the zero-dynamic method. Replacing
Eq. (10) into Eq. (6):
q̈2 = P1 q̇1 + P2 q̇2 + P3 (15)

where:
     
−d11 c21 −d21 c11 +d21 d 1 k1 −d11 c22 −d21 c12 +d21 d 1 k3 −d11 k 2 q2 −d21 k 1 q2
P1 =   , P2 =   , P3 =  
d11 d 2 −d 1 k2 d21 d11 d 2 −d 1 k2 d21 d11 d 2 −d 1 k2 d21

Assuming s = 0 in Eq. (8), then Eq. (8) is rewritten as follows

q̇1 = −k1 (q1 − q1d ) − k2 q̇2 − k3 q2 (16)
 T  T
Let x = x1 x2 x3 = q2 q̇2 q1 − q1d , the Eqs. (15) and (16) are rewritten as

ẋ2 = P1 ẋ3 + P2 x2 + P3 (17)

ẋ3 = −k1 x3 − k2 x2 − k3 x1 (18)

Substitute Eq. (18) into Eq. (17)
ẋ2 = P1 (−k1 x3 − k2 x2 − k3 x1 ) + P2 x2 + P3
= −P1 k3 x1 − (P1 k2 − P2 )x2 − P1 k1 x3 + P3 (19)
Rewrite the equations x2 = ẋ1 , (18), and (19) as a matrix as follows
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
ẋ1 0 1 0 x1 0
⎣ ẋ2 ⎦ = ⎣ −P1 k3 P2 − P1 k2 −P1 k1 ⎦⎣ x2 ⎦ + ⎣ P3 ⎦ (20)
ẋ3 −k3 −k2 −k1 x3 0
The stability of the nonlinear system in (16) was evaluated using Lyapunov’s lin-
earization theorem. Equation (16) can be linearized to the equilibrium position x = 0 as
follows:
⎡ ⎤ ⎡ ⎤⎡ ⎤
ẋ1 0 1 0 x1
⎣ ẋ2 ⎦ = ⎣ A21 A22 A23 ⎦⎣ x2 ⎦ (21)
ẋ3 −k3 −k2 −k1 x3
860 N. T. T. Hien et al.

2 +θ3 )k1 k3 (θ2 +θ3 )(k3 −k1 k2 )

where: A21 = −P1 k3 |q=qd = − θ(θ
2 −k2 (θ2 +θ3 )
, A22 = P2 − P1 k2 |q=qd = θ2 −k2 (θ2 +θ3 )

(θ2 + θ3 )k12
A23 = −P1 k1 |q=qd = − .
θ2 − k2 (θ2 + θ3 )
To ensure the stability of the linearized system given by Eq. (21), the linearized
system with a matrix of state A must be Hurwitz. Therefore, the characteristic polynomial
of the linear system from Eq. (21) is given as follows
⎛⎡ ⎤⎞
s −1 0
(s) = det(sI3 − A) = det ⎝⎣ −A21 s − A22 −A23 ⎦⎠ (22)
k3 k2 s + k1

Finally, Eq. (22) is written as

⎛ ⎞
(θ2 + θ3 )(k1 k2 − k3 ) (θ2 + θ3 )k12 k2
⎜s + s−
2
s⎟
⎜ θ2 − k2 (θ2 + θ3 ) (s + k1 )(θ2 − k2 (θ2 + θ3 )) ⎟
(s) = (s + k1 )⎜ ⎟ (23)
⎝ (θ2 + θ3 )k1 k3 +(θ2 + θ3 )k12 k3 ⎠
+ −
θ2 − k2 (θ2 + θ3 ) (s + k1 )(θ2 − k2 (θ2 + θ3 ))
Sufficient conditions for stability can be determined by choosing control parameters
{k1 , k2 , k3 }, such that the polynomials (s) are Hurwitz. Using Routh-Hurwitz criteria,
we get:
θ2
k1 > 0, k2 < , (θ2 + θ3 )k2 < 0, (θ2 + θ3 )k3 > 0. (24)
θ2 + θ3
Remark 2 The control signal (12) with sufficient condition (24) helps the system to
 T
stabilize in state x = 0. Thus, the response of the state variable q = q1 q2 approaches
 T
the desired value qd = q1d q2d as t approaches infinity.

4 Simulation Evaluation

Matlab/Simscape Multibody Link plug-in on Solidworks are expressed to validate the

robust HSMC of the nonlinear planar flexible-joint robot system with two degrees of
freedom. By using the quasi-physical model, the simulation evaluation represent more
accurately the robot manipulator than traditional mathematical model. Considering the
robot manipulator with the following parameters θi,1≤i≤3 as: θ1 = 0.0799, θ2 = 0.0244,
θ3 = 0.0205. . The controller parameter based on the energy approach, i.e., the robot’s
passivity-based control method (PD) is kP = 1, kD = 1, kE = 1, for the stiffness constant
of the spring k = 1. The initial position initialized as: q1 = 2π/3, q2 = π/8, q̇1 =
0, q̇2 = 0 that shown in Fig. 2a.
In the first simulation scenario, the rotation angle response of both joints according
to the PD controller and the robust HSMC has response time at 4 s as shown from Figs. 3,
 Robust Hierarchical Sliding Mode Control 861

(b)

(a)

Fig. 2. The quasi-physical robot: a The initial position. b The derised position.

(a) (b)
Fig. 3. a Rotational response of the first joint. b Response of the rotation angle of the second
joint.

(a) (b)
Fig. 4. a Derivative response of the first joint rotation angle. b Response of the second joint
rotation angle derivative.

4, 5 and 6. In the range of from 0 s to 4 s, it was found that the response of the sliding
mode controller shows less oscillation than that of the PD controller.
In the second simulation scenario, Figs. 7, 8, 9 and 10 represent the output response of
the 2-DOF flexible-joint robot in case the model contains uncertainties. The simulation
results show the response of the robust HSMC when it has the ability to be stable with
862 N. T. T. Hien et al.

(a) (b)
Fig. 5. Considering the uncertainty parameter a Rotational response of the first joint. b Rotational
response of the second joint.

(a) (b)
Fig. 6. Considering the uncertainty parameter. a Response of the derivation of the first-joint
rotation angle. b Response of the second joint rotation angle derivation.

the uncertainties of the model, which the robot’s passivity-based control method has not
been able to solve.

5 Conclusion

The article performs modeling and control of the 2-DOF flexible-joint robot. With the
robot’s characteristic that the system is underactuated system, the robust HSMC is pro-
posed with the Lyapunov stability theory to prove the stability of the system including
the underactuated joint. The Matlab/Simscape Multibody Link plug-in on Solidworks
simulation is used to demonstrate the effectiveness of the controller in case the model
considers the uncertainty component. The controller simulation results are compared
with the robot’s passivity-based control method, obtaining significantly improved control
results especially when the system is affected by the uncertain components.

Acknowledgments. The authors gratefully acknowledge the Thai Nguyen University of Tech-
nology for supporting this work.
 Robust Hierarchical Sliding Mode Control 863

References
1. Yang, C., Li, Z., Cui, R., Xu, B.: Neural network-based motion control of an underactuated
wheeled inverted pendulum model. IEEE Trans. Neural Netw. Learn. Syst. 25(11), 2004–2016
(2014). https://doi.org/10.1109/TNNLS.2014.2302475
2. Jain, A., Sharma, A., Jately, V., Azzopardi, B., Choudhury, S.: Real-time swing-up control
of non-linear inverted pendulum using lyapunov based optimized fuzzy logic control. IEEE
Access 9, 50715–50726 (2021). https://doi.org/10.1109/ACCESS.2021.3058645
3. Nam, D.P., Ha, N.H., Tran, V.A., Khanh, D.D., Khue, N.D., Van Trong, D.: About model sep-
aration techniques and control problems of wheeled mobile robots. In: Intelligent Computing
in Engineering, pp. 1119–1127 (2020)
4. Tuan, L.A.: Neural observer and adaptive fractional-order backstepping fast-terminal sliding-
mode control of RTG cranes. IEEE Trans. Ind. Electron. 68(1), 434–442 (2021). https://doi.
org/10.1109/TIE.2019.2962450
5. Van Trieu, P., Cuong, H.M., Dong, H.Q., Tuan, N.H., Tuan, L.A.: Adaptive fractional-order
fast terminal sliding mode with fault-tolerant control for underactuated mechanical systems:
Application to tower cranes. Autom. Constr. 123(December 2020), 103533 (2021). https://
doi.org/10.1016/j.autcon.2020.103533
6. Fantoni, I., Lozano, R., Mazenc, F., Annaswamy, A.: Stabilization of a two-link robot using
an energy approach. In: European Control Conference ECC 1999—Conference Proceedings,
pp. 2886–2891 (2015). https://doi.org/10.23919/ecc.1999.7099766
7. Zhang, A., Lai, X., Wu, M., She, J.: Stabilization of underactuated two-link gymnast robot
by using trajectory tracking strategy. Appl. Math. Comput. 253, 193–204 (2015). https://doi.
org/10.1016/j.amc.2014.12.065
8. Yeon, J.S., Park, J.H., Lee, S.H.: Practical robust control for flexible joint robot manipulators.
IFAC Proc. 17(1 PART 1), 3377–3382 (2008). https://doi.org/10.3182/20080706-5-KR-1001.
3841
9. Nasiri, N., Fakharian, A., Menhaj, M.B.: Observer-based robust control for flexible-joint robot
manipulators: a state-dependent Riccati equation-based approach. Trans. Inst. Meas. Control
42(16), 3135–3155 (2020). https://doi.org/10.1177/0142331220941653
10. Jia, P.: Control of flexible joint robot based on motor state feedback and dynamic surface
approach. J. Control Sci. Eng. (2019). https://doi.org/10.1155/2019/5431636
11. Suklabaidya, S., Lochan, K., Roy, B.K.: Modeling and sliding mode control of flexible link
flexible joint Robot Manipulator. In: The ACM International Conference Proceeding Series.
https://doi.org/10.1145/2783449.2783509
12. Yadav, A., Kumar, A., Bhushan, B.: Sliding mode control with RBF neural network for two
link robot manipulator. Int. J. Comput. Appl. 178(52), 31–36 (2019). https://doi.org/10.5120/
ijca2019919408
13. Rsetam, K., Cao, Z., Man, Z.: Design of robust terminal sliding mode control for underactuated
flexible joint robot. IEEE Trans. Syst. Man Cybern. Syst. 1–14 (2021). https://doi.org/10.1109/
TSMC.2021.3096835
14. Fantoni, I., Lozano, R.: Non-linear Control for Underactuated Mechanical Systems. Springer,
London, London (2002)
 Selection of Input Parameters in External
Cylindrical Grinding Using EDAS Method

Tran Huu Danh1 , Bui Thanh Danh2 , Trieu Quy Huy3 , Tran Minh Tan4 ,
Nguyen Hong Linh5 , Luu Anh Tung6 , and Hoang Xuan Tu6(B)
1 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
2 University of Transport and Communications, Hanoi, Vietnam
3 University of Economics - Technology for Industries, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Electric Power University, Hanoi, Vietnam
6 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

hxt_dhktcn@tnut.edu.vn

Abstract. This paper presents the results of a multi-criteria decision-making

(MCDM) study in cylindrical external grinding SKD11 tool steel. The goal of
this research is to find the best input process parameters for dressing process in
external grinding to achieve minimal surface roughness (RS) and maximum wheel
life (T). First, the EDAS (Evaluation based on Distance from Average Solution)
method was used to solve the MCDM problem and the Entropy method was
employed to calculate the weights of criteria. Furthermore, an experiment was
conducted with six input parameters: fine dressing depth, fine dressing passes,
coarse dressing depth, coarse dressing passes, non-feeding dressing, and dressing
feed rate. The experiment was also designed using the Taguchi method and an
L16 orthogonal array. Based on the results of the work, the best dressing mode for
external cylindrical grinding has been proposed.

Keywords: External grinding · MCDM · EDAS method · Entropy method ·

Surface roughness · Wheel life · SKD11

1 Introduction
Grinding is a type of machining that uses hard abrasive particles as the cutting medium.
It is often the key to achieve the required quality [1] so it is widely used for finishing and
semi-finishing machine parts. It accounts for approximately 20–25% of total industry
spent on mechanical parts [2].
In practice, the grinding wheel is continuously worn while working, greatly affecting
the process quality and productivity. As a result, the dressing process is required to ensure
the grinding wheel’s original specification. The dressing process impacts on the profile,
topography, and wear behavior of the grinding wheel. Therefore, it has a significant
influence on the grinding process efficiency and accuracy. For those reasons, researchers
have been always interested in improving the quality, productivity of grinding process
as well as determining the best grinding mode.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 864–872, 2023.
https://doi.org/10.1007/978-3-031-22200-9_91
 Selection of Input Parameters in External Cylindrical 865

Daneshi et al. [3] conducted a study to identify appropriate dressing strategies for
dressing in internal cylindrical grinding. In this study, two different types of wheels,
including CBN (Cubic Boron Nitride) and corundum wheels, and three different dress-
ing rollers, including an electroplated, a vitrified bond form roller and a cup-dresser
were applied. Liu et al. [4] used kinematic simulations to predict surface roughness in
grinding. Three different abrasive grain shapes (sphere, truncated cone, and cone) and
a single-point diamond dressing model were investigated in their study. Also, the pro-
posed surface roughness model was experimentally validated with a difference of 7–11
percent. Klocke and Linke [5] investigated the mechanisms resulting in the formation
of grinding wheel topography by dressing. The effect of kinematical dressing factors
on wheel wear behavior has been computed in this work. Walter et al. [6] discussed
the dressing and truing of hybrid bonded (metal-vitrified) CBN grinding wheels using
a short-pulsed fiber laser. Several studies have been conducted to determine optimum
dressing mode. Optimal dressing parameters have been suggested for cylindrical grind-
ing [7–9], surface grinding [10–12], and internal grinding [13, 14]. Kozuro et al. [15]
proposed the following dressing mode for external grinding to achieve the surface rough-
ness Ra = 0.32–1.25 (µm): longitudinal feed rate of 0.4 (m/min); dressing four passes
with a dressing depth of 0.03 (mm); and four passes non-feeding dressing.
From the above analyses, it can be noted that there have been quite a few studies
on the dressing mode and on the determination of the optimum or reasonable wheel
dressing mode so far. Besides, it is clear that different dressing modes such as coarse
dressing, fine dressing, and non-feeding dressing have an influence on the efficiency of
the dressing process. However, to date, there has been no research on determining the best
dressing mode among many different dressing setups. Therefore, the best dressing mode
in external cylindrical grinding SKD11 tool steel has been taken into great consideration.
Multi-criteria decision making (MCDM) is a common problem when it is neces-
sary to analyze multiple options in order to choose the best one. Numerous studies
on MCDM have been conducted for various mechanical machining processes such
as milling, turning, grinding, electrical discharge machining (EDM), etc. [16]. This
is because the machining process is frequently required to meet many criteria, such as
maximal material removal rate (MMR), minimal surface roughness (SR), or maximal
tool wear rate. So far, a number of studies have been conducted to determine the best
dressing mode using MCDM methods. Patil and Bhalerao [17] used TOPSIS (Technique
for Order of Preference by Similarity to Ideal Solution) method to select levels of dress-
ing parameters for a better surface finish in cylindrical angular grinding of En-31 parts.
In this study, four input parameters, including the dressing cross feed rate, the dressing
depth of cut, the width of the diamond dresser and drag angle of the dresser were inves-
tigated. However, only three solutions with a surface finish criterion were provided for
the MCDM problem. Nguyen et al. [16] carried out a comparison study on MCDM for
the dressing process when internal grinding SKD11 steel. Four MCDM methods, TOP-
SIS, MARCOS (Measurement of Alternatives and Ranking according to Compromise
Solution), EAMR (Evaluation by an Area-based Method of Ranking), and MAIRCA
(Multi-Attributive Ideal–Real Comparative Analysis), as well as two weight calculation
methods, Entropy and MEREC, were used in this study. It can be recognized from the
above analysis that while there have been various applications of MCDM methods for
866 T. H. Danh et al.

different mechanical machining processes, no studies have been conducted to determine

the best dressing mode for external cylindrical grinding.
This paper, hence, presents a study on MCDM for the dressing process in cylindrical
external grinding. In this work, two criteria were chosen for the investigation including
the surface roughness RS and the wheel life T. In addition, to solve the MCDM prob-
lem, the EDAS method was selected. Furthermore, the Entropy method was applied to
determine the weights of criteria. As a result, the SAW method has been successfully
used to determine the best dressing mode in external cylindrical grinding.

2 Methodology

2.1 Method for Multi-criteria Decision Making

The EDAS method was employed to solve the MCDM problem in this work.
To use this method, the following steps must be taken [18]:
Step 1: Creating the initial decision-making matrix:
⎡ ⎤
r11 · · · r1j · · · r1n
⎢ .. . . .. . . .. ⎥
⎢ . .. .. ⎥
⎢ ⎥
X=⎢
⎢ 1 r · · · r ij · · · r ⎥
in ⎥ (1)
⎢ . . . . . ⎥
⎣ .. . . .. . . .. ⎦
rm1 · · · rmj · · · rmn m×n

In which, m is the number of the alternative; n is the criteria number.

Step 2: Determining the average solution of each criterion:
m
rij
AV j = i=1 (2)
m

Step 3: Calculating the positive and negative distances from average solution by:
For the criterion RS:

max 0, rij − AV j
PDAij = (3)
AV j

max 0, AV j − rij
NDAij = (4)
AV j

For criterion T:

max 0, AV j − rij
PDAij = (5)
AV j

max 0, rij − AV j
NDAij = (6)
AV j
 Selection of Input Parameters in External Cylindrical 867

Step 4: Computing the weighted PDA and NDA:


n
SP i = PDAij · wj (7)
j=1


n
SN i = NDAij · wj (8)
j=1

Step 5: Calculating weighted normalized PDA and NDA by:

SP i
NSP i = (9)
max(SP i )
i
SN i
NSN i = (10)
max(SN i )
i

Step 6: Determining appraisal score for each alternative by:

AS i = (NSP i + NSN i )/2 (11)

Step 7: Rank the alternatives according to the rule that the option with the highest AS i
is the best.

2.2 Method for Calculatiing Weight of Criteria

The Entropy method was used to calculate the weights of the criteria in this work.
The steps outlined below can be used to implement this method [19].
Step 1: Finding the indicator’s normalized values by:
xij
pij = m (12)
m+ 2
i=1 xij

Step 2: Calculating the Entropy for each indicator by the following Equation:
m 
 m   m 
mej = − pij × ln pij − 1 − pij × ln 1 − pij (13)
i=1 i=1 i=1

Step 3: Determining the weight of each indicator:

1 − mej
wj = m
(14)
j=1 1 − mej
868 T. H. Danh et al.

3 Experimental Work
An experiment was carried out in order to solve the MCDM problem. Figure 1 illus-
trates the experimental setup for external grinding, which includes the following equip-
ment: grinding machine CONDO-Hi-45 HTS (Japan); grinding wheel: Ct80MV1-G
400x40x203 (Vietnam); dresser: 3908-0088C type 2 (Russia); surface roughness tester:
Mitutoyo 178-923-2A, SJ-201 (Japan). In addition, Table 1 describes the dressing process
input parameters and their levels. These variables will influence the grinding process,
which has outcomes such as surface roughness SR and wheel life T. Furthermore, the
experiment was designed using the Taguchi method and the L16 orthogonal array design
(44 × 22 ). Table 2 illustrates the experimental strategy.
The experiment starts with the wheel dressing experiment. After dressing, the grind-
ing wheel is used to process the samples in the following grinding mode where the
workpiece was made of SKD11 tool steel, and the grinding wheel speed was 29.3 (m/s).
The longitudinal feed rate was 1.8 (m/min), and the total depth of cut was 0.05 (mm).
The values of SR and T can be found by carrying out the external grinding process
with the above grinding mode for each wheel dressing mode. Then the grinding time
and surface roughness of the machined part are measured after each test run. The pre-
ceding procedure is repeated indefinitely until the surface roughness of the machined
part exceeds the allowable value (0.5 µm in this study). At this point, the total grinding
time of all machined parts is equal to the wheel life T with the specified dressing mode.
Table 2 shows the SR and T values in the two left columns.

Fig. 1. Setup of external grinding

 Selection of Input Parameters in External Cylindrical 869

Table 1. Input dressing parameters and their levels

No Input factors Unit Levels

1 2 3 4
1 Coarse dressing depth ar mm 0.03 0.04 – –
2 Coarse dressing passes nr Times 1 2 3 4
3 Fine dressing depth af mm 0.005 0.01 0.015 0.02
4 Fine dressing passes nf Times 0 1 2 3
5 Non-feeding dressing n0 Times 0 1 2 3
6 Dressing feed rate Sd m/min 1 1.5 – –

Table 2. Experimental matrix and output results

Exp. no Input parameters Output results

af nf n0 nr ar Sd Ra (µm) T (min)
1 0.005 0 0 1 0.03 1 0.496 9.29
2 0.005 1 1 2 0.03 1.5 0.421 8.98
3 0.005 2 2 3 0.04 1 0.23 16.48
4 0.005 3 3 4 0.04 1.5 0.34 8.52
5 0.01 0 1 3 0.04 1.5 0.254 9.86
6 0.01 1 0 4 0.04 1 0.322 9.55
7 0.01 2 3 1 0.03 1.5 0.211 9.04
8 0.01 3 2 2 0.03 1 0.293 9.37
9 0.015 0 2 4 0.03 1.5 0.321 12.81
10 0.015 1 3 3 0.03 1 0.23 10.82
11 0.015 2 0 2 0.04 1.5 0.325 9.62
12 0.015 3 1 1 0.04 1 0.292 9.19
13 0.02 0 3 2 0.04 1 0.326 9.00
14 0.02 1 2 1 0.04 1.5 0.353 10.32
15 0.02 2 1 4 0.03 1 0.228 11.21
16 0.02 3 0 3 0.03 1.5 0.325 8.22

4 Determining the Best Experimental Setup for External Grinding

SKD11 Tool Steel
This section describes how to find the best experimental setup for external grinding
SKD11 tool steel using the EDAS method to solve the MCDM problem and the Entropy
method to determine criteria weights.
870 T. H. Danh et al.

4.1 Determining the Weights for the Criteria

The weights of criteria are calculated using the Entropy method (see Sect. 2.2). To start,
Eq. (12) is used to calculate the normalized values pij . Then, Eq. (13) is used to calculate
the Entropy value for each indicator mej . Finally, using Eq. (14), he criteria weight wj is
computed. The weights of Ra and T are determined to be 0.4983 and 0.5017, respectively.

4.2 Determining the Best Experimental Setup Using EDAS Method

The EDAS method is employed to solve the MCDM problem in the following manner
(see Sect. 2.1). The first, the mean solution of each criterion using Eq. (2) is determined.
After that, the positive and negative distances from the average solution are calculated
using Eqs. (3) and (4) for the surface roughness and (5) and (6) for the wheel life.
Then, using Eqs. (7) and (8), the weighted PDA and NDA are computed. The weighted
normalized PDA and NDA are then quantified using Eqs. (9) and (10). Afterwards,
the appraisal score ASi for each option is calculated using fomula (11). Finally, the
alternatives are arranged in such a way that the option with the highest ASi is the best.
Table 3 shows several calculated results as well as alternative rankings.

Table 3. Several calculated results and ranking of alternatives

Trial SPi SNi NSPi NSNi ASi Rank

1 0.0000 0.3402 0.0000 0.0000 0.0000 16
2 0.0000 0.2344 0.0000 0.3109 0.1554 15
3 0.4422 0.0000 1.0000 1.0000 1.0000 1
4 0.0000 0.1284 0.0000 0.6225 0.3113 14
5 0.0906 0.0140 0.2049 0.9589 0.5819 6
6 0.0000 0.0484 0.0000 0.8577 0.4288 9
7 0.1596 0.0544 0.3609 0.8401 0.6005 5
8 0.0285 0.0382 0.0645 0.8876 0.4761 7
9 0.1321 0.0164 0.2987 0.9517 0.6252 4
10 0.1624 0.0000 0.3674 1.0000 0.6837 3
11 0.0000 0.0487 0.0000 0.8569 0.4284 10
12 0.0296 0.0471 0.0669 0.8615 0.4642 8
13 0.0000 0.0810 0.0000 0.7620 0.3810 12
14 0.0088 0.0683 0.0198 0.7992 0.4095 11
15 0.1847 0.0000 0.4178 1.0000 0.7089 2
16 0.0000 0.1190 0.0000 0.6501 0.3251 13

It is recognizable from Table 3 that option 3 is the best optionas it has the highest ASi
value (Qi = 1.0000). As a result, the optimal experimental setup includes the following
 Selection of Input Parameters in External Cylindrical 871

input parameters: ar = 0.04 (mm); nr = 3 (times); af = 0.005 (mm); nf = 2 (times); n0

= 2 (times); Sd = 1 (m/min)

5 Conclusions

The results of using the EDAS method to solve the MCDM problem and the Entropy
method to calculate the weight of the criteria in external grinding SKD11 tool steel
are presented in this paper. The Taguchi method was used in the research to design
an experiment with an L16 orthogonal array (44 × 22 ). The study findings lead to the
following conclusions:

– For the first time, the EDAS method has been used to solve the MCDM problem in
external grinding.
– The following are the optimal wheel dressing parameters for achieving the lowest
surface roughness and longest wheel life: ar = 0.04 (mm); nr = 3 (times); af = 0.005
(mm); nf = 2 (times); n0 = 2 (times); Sd = 1 (m/min.).

Acknowledgment. Thai Nguyen University of Technology provided support for this research.

References
1. Rowe, W.B.: Principles of Modern Grinding Technology. William Andrew, USA (2009)
2. Malkin, S., Guo, C.: Grinding Technology: theory and Application of Machining with
Abrasives: Industrial Press Inc. (2008)
3. Daneshi, A., Jandaghi, N., Tawakoli, T.: Effect of dressing on internal cylindrical grinding.
Procedia CIRP 14, 37–41 (2014)
4. Liu, Y., et al.: Investigation of different grain shapes and dressing to predict surface roughness
in grinding using kinematic simulations. Precis. Eng. 37(3), 758–764 (2013)
5. Klocke, F., Linke, B.: Mechanisms in the generation of grinding wheel topography by dressing.
Prod. Eng. Res. Dev. 2(2), 157–163 (2008)
6. Walter, C., et al.: Dressing and truing of hybrid bonded CBN grinding tools using a short-
pulsed fibre laser. CIRP Ann. 61(1), 279–282 (2012)
7. Prusak, Z., Webster, J., Marinescu, I.: Influence of dressing parameters on grinding perfor-
mance of CBN/seeded gel hybrid wheels in cylindrical grinding. Int. J. Prod. Res. 35(10),
2899–2916 (1997)
8. Aleksandrova, I.: Optimization of the dressing parameters in cylindrical grinding based on
a generalized utility function. Chin. J. Mech. Eng. 29(1), 63–73 (2016). https://doi.org/10.
3901/CJME.2015.1103.130
9. Tu, H.X., et al.: Influence of dressing parameters on surface roughness of workpiece for grind-
ing hardened 9XC tool steel. In: IOP Conference Series: Materials Science and Engineering.
IOP Publishing (2019)
10. Hong, T.T., et al.: Multi-criteria optimization of dressing parameters for surface grinding
90CrSi tool steel using taguchi method and grey relational analysis. Mater. Sci. Forum. (Trans
Tech Publ) (2020)
872 T. H. Danh et al.

11. Hong, T.T., et al.: Multi response optimization of dressing conditions for surface grinding
SKD11 steel by HaiDuong grinding wheel using grey relational analysis in Taguchi method.
In: International Conference on Engineering Research and Applications. Springer (2020)
12. Tran, T.H., et al.: Influence of dressing conditions on surface roughness when surface grinding
SKD11 steel. Mater. Sci. Forum. (Trans Tech Publ) (2021)
13. Hung, L.X., et al.: Optimum dressing parameters for maximum material removal rate when
internal cylindrical grinding using Taguchi method. Int. J. Mech. Eng. Technol. (IJMET)
9(12), 123–129 (2018)
14. Hung, L.X., et al.: Determining the optimum set of dressing parameters satisfying minimum
surface roughness when conducting the internal grinding of hardened SKD11 steel. Solid
State Phenom. (Trans Tech Publ)
15. Kozuro, L.M., Panov, A.A., Remizovski, E.I., Tristosepdov, P.S.: Handbook of Grinding (in
Russian). Publish Housing of High-education, Minsk (1981)
16. Nguyen, H.-Q., et al.: A comparative study on multi-criteria decision-making in dressing
process for internal grinding. Machines 10(5), 303 (2022)
17. Patil, S.S., Bhalerao, Y.J.: Selection of levels of dressing process parameters by using TOPSIS
technique for surface roughness of en-31 work piece in CNC cylindrical grinding machine.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2017)
18. Alinezhad, A.: New Methods and Applications in Multiple Attribute Decision Making
(MADM). Springer (2019)
19. Hieu, T.T., Thao, N.X., Thuy, L.: Application of MOORA and COPRAS models to select
materials for mushroom cultivation. Vietnam J. Agricult. Sci. 17(4), 32–2331 (2019)
 Stability of Microgrids: An Application
of Virtual Synchronous Generator

Guido Ala1 , Nicola Campagna1 , Massimo Caruso1 , Vincenzo Castiglia1 ,

Antonino Oscar Di Tommaso1 , Rosario Miceli1 , Claudio Nevoloso1 ,
Giuseppe Schettino1 , Fabio Viola1 , and Minh Nguyen2(B)
1 Department of Engineering, University of Palermo, Viale Delle Scienze, Parco d’Orleans,
90128 Palermo, PA, Italy
2 Department of Electrical Engineering, Thai Nguyen University of Technology, Thai
Nguyen 240000, Vietnam
nguyentuanminh@tnut.edu.vn

Abstract. The objective of this paper is to illustrate an alternative control algo-

rithm for power converters which have the task of introducing the energy generated
by systems based on the use of renewable sources such as, photovoltaic or wind
power, in to small-sized networks, microgrids. Virtual Synchronous Generator can
operate both in parallel to the main network as well as in isolated and autonomous
conditions (islanding-mode). The contribute of this work is to drive a consoli-
date approach, made in time domain, in a more flexible and less time-consuming
approach based on Park transformation.

Keywords: Virtual synchronous generator · Synchronverter · Park

transformation

1 Introduction
The term microgrid refers to a portion of the network characterized by the presence of
an energy generation system, generally from renewable sources, and a system of loads;
this network can be operated both in parallel to the main distribution network, and in
a completely autonomous way (islanding mode). The penetration of generation from
renewable sources is characterized by undoubted advantages: the possibility to reduce
the use of fossil sources and therefore reducing the environmental impact associated with
energy production, and in recent years it has almost reached parity with fossil fuel-based
systems. Problems arise due to the uncertainty of the renewable sources during different
seasons, so the grid has to properly plan the use of its own resources, as first point of
view.
Synchronous generators which, with their inertia and the damping effect of mechan-
ical and electromagnetic friction, help to mitigate the effects of frequency or voltage
transients due to faults, maneuvers or simple load variations, are increasingly replaced
by green generators, so their absence is a weak point for the grid as a second point of
view.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 873–880, 2023.
https://doi.org/10.1007/978-3-031-22200-9_92
874 G. Ala et al.

These consequences are even more evident within the microgrids operating in iso-
lation which are generally characterized by a strong penetration of generation from
renewable sources.
It is more than evident that the use of renewable sources must be accompanied by
the use of inverters that can contribute to the stability of the grid.
The inverter operates in parallel to a stator generator (sometimes it can also replace
it) to mimic the behavior of synchronous generator, injecting energy from a capacitor
bank, also defining its own digital inertial properties, called virtual inertia. Algorithms,
drives the injection in conventional way, if the frequency or voltage decreases there is a
major injection of energy, as the flow generated by prime movers [1–5].
The Virtual Synchronous Generator (VSG) aims to emulate the mechanical inertia,
and therefore also the dynamics, of a traditional synchronous generator to deal with
stability problems with the advantage of being able to change the parameters of the
virtual machine according to the needs of the network. Smart inverters, can be operated in
grid-following or grid-forming mode. In the first case (grid-following mode) the inverter
is controlled in such a way as to chase the grid and support the ancillary grid services.
Operation in grid-forming mode is typical of isolated microgrids or small networks,
which due to the absence of inertia of great rotating generators are more subjected to
disturbances and instability.
Among the main models that can be found in the literature, undoubtedly stand out:
the VISMA (Virtual Synchronous Machine) and the Synchronverter [1].
Both types are characterized by the ability to generate virtual inertia and participate
in the absorption of negative effects due to changes in structure, load or failure within the
network thanks to their versatility. The VISMA model approach involves measuring the
voltage at the point of connection with the network (Point of Common Coupling) in order
to calculate the phase of the currents in real time; these are used as a reference for the
output currents from the inverter, making it behave like an impressed current converter
(or Current Source Inverter). A Virtual Synchronous Generator scheme is presented in
Fig. 1. It is capable of emulating the dynamics and behavior of a physical synchronous
generator, whose model is implemented within the inverter control logic.
In [5], the Synchronverter model and its operation do not depend on the measured
voltage or current values and therefore set as a reference, the only need is that of an
energy storage system necessary to provide for the lack of inertia of the converter.
The greater complexity in the implementation of a VISMA and its need to control
the inverter as a CSI, which does not match the needs of the network, have led to the
choice of the Synchronverter which, on the contrary, is controlled as an impressed voltage
(Voltage Source Inverter) it is capable of keeping itself synchronized with the grid just
like a traditional synchronous generator. In [1] the control in based in time domain
equations, but the authors propose a control based on Park transformation, for which the
complexity level is reduced [6, 7].
 Stability of Microgrids: An Application 875

Fig. 1. Typical topological scheme of a VSG, VSG control is based on Park transform

2 Dynamic Model of the Synchronous Generator

and Synchronverter
One of the main advantages deriving from the use of the synchronverter is that of being
able to vary the parameters of the virtual machine such as the moment of inertia, the
friction coefficient and the inductance of the stator and field windings even during the
operating phases of the system. The dynamic model of the synchronous machine, a
subject widely treated and present in the literature, and which in this case is referred to
in studies [8, 9].
The basic hypotheses taken into consideration are: machine consisting of a smooth
rotor and a polar pair; stator windings characterized by constant self-inductance; absence
of damper windings; negligible saturation effects in the ferromagnetic material as are any
eddy currents. The dynamic model of the machine is complete by making the appropriate
replacements and adding the equation that establishes the mechanical equilibrium, also
called swing equation:
dω
J = Tm − Te − Dn ω, (1)
dt
in which J is the moment of inertia, Tm and Te are, respectively, the mechanical and
electromagnetic torque and Dn the friction coefficient (damping factor).
It is known that, within an electrical system, as the reactive part of impedance is
greater than the resistive one„ it is possible to express a direct relationship between
the active power supplied from the inverter and the frequency, and the reactive power
supplied with the voltage at the output terminals. These relationships can be summarized
as follows: P ≈ EV X ϑ, Q ≈ X (V − E) where: E represents the voltage of the network
E

bus, V the voltage at the output terminals of the converter and ϑ its phase, X the reactance.
876 G. Ala et al.

By considering the droop control approach [10, 11] which allows adjustment of the
system frequency and the amplitude of the line voltage,
   
ω = ω∗ − mp P − P ∗ V = V ∗ − mq Q − Q∗ (2)

in which ω* and V* represent the nominal values of the pulsation and of the amplitude
of the mains voltage, mp and mq are the droop coefficients, while P* and Q* represent the
reference values for the powers supplied by the inverter, values limited by the maximum
deliverable power, as illustrated in Fig. 2.

Fig. 2. Drooping characteristics for frequency and voltage

In synchronous generators the rotor speed is kept in equilibrium by the prime mover
and by the effect due to friction. If there is an increase in load inside the network, this
will affect the machine as an increase in the electromagnetic torque Te , which will be
opposite to the driving one, and therefore the rotor will tend to slow down; at this point
the speed regulation system will act on the turbine, opening the lights and letting more
steam flow in order to increase the mechanical torque Tm and restore balance.
In a real synchronous generator mp and mq depend on the inertia of the mechanical
machine, in a synchronverter named Dp and Dq can be fixed by the algorithm.

3 Analysis of Dynamic Behavior

The parameters of the Synchronverter are: an inverter whose nominal power is 65 kVA
and which is able to guarantee an availability of active and reactive power equal to Pset
= 65 kW and Qset = 65 kVAR, for needs of the load, assuming that the converter is
forced to meet the entire energy demand and have a certain amount of reserve power;
Inductance (Ls ) 2.5 mH.
By following CEI EN 50160 which, for island operation, establishes that: (1) Varia-
tions of nominal frequency (50 Hz) ± 15% (42.5–57.5 Hz) are allowed for 100% of the
time, and of ±2% (49–51 Hz) for the entire week; (2) Short-term variations for nominal
voltage (400 VRMS) of ±5% (380–420 V) are allowed. Therefore, due the previous
considerations, the values of Dp and Dq are:
 Stability of Microgrids: An Application 877

Pmax 65000 s
Dp = = = 10345 W · ,
ωmax 0.02 · 2π50 rad
Qmax 65000
Dq = = √ = 3996.7 A (3)
Vmax 0.05 · 230 2
Finally, having chosen the time constants of the regulation loops respectively equal to
τf = 0.02 s and τv = 0.002 s, the latest parameters of synchronverter becomes: Moment
of inertia, J = Dp τf , 206.9 kg m2 ; and, Constant K, (which has a dual role to J in the
frequency regulation ring) K = τv ωn Dq = 2511.2 A rad.
The scenario evaluates the behaviour of the virtual synchronous generator when
reactive power is required. In fact, the initial condition a PV generator is the only source
of active power and which, alone and for the entire duration of the simulation, powers
the 20 kW resistive load; in the following instants an inductive load and a capacitive
load will be connected alternately, which can be powered only thanks to the support of
the Synchronverter, the only available energy source capable of supplying both active
and reactive power. The response of the system was analysed in relation to a time t =
1 s of overall duration of the phenomenon.
It is known, from the theory of electrical machines, that an inductive load leads to
a lowering of the voltage due to the demagnetizing effect generated by the alternator
armature reaction; dually, a load of a capacitive nature produces the exact opposite effect.
Within this simulation, the resistive load of 20 kW is powered exclusively by the
photovoltaic generator for 0 ≤ t ≤ 0.4 s, which provides the power request autonomously
without the aid of the synchronverter. By considering Fig. 3, it is required to deliver a
non-zero power contribution for 0.4 ≤ t ≤ 0.8 s in which for half the time it will be
ohmic-inductive and for the subsequent half ohmic-capacitive, as can be seen from the
following profile of Fig. 3.
It is possible to notice the effect that the load has on the voltage: in fact, precisely
at the instants t = 0.4 s and t = 0.6 s, i.e. when the inductive load and the capacitive
load are respectively connected, there is a sudden variation of the voltage, in particular
a lowering and a raising. The same effect, by the reasoning described above, can be
highlighted by the analysis of the trend of the electromotive force induced by the virtual
synchronous generator and for the frequency of the grid, as shown in Fig. 4.

4 Conclusions
The inexorable, and now indispensable, energy transition that involves a strong penetra-
tion of renewable sources within the electrical systems has given rise to a new problem
related to the absence of the inertia of the rotating masses characteristic of traditional
synchronous generators. In a global context in which the goal is independence from
fossil sources, the work carried out is part of an approach capable of encouraging the
integration of alternative generation systems such as wind, photovoltaic and energy stor-
age systems; in fact, thanks to the fact that the Synchronverter is capable of emulating
the dynamics of a synchronous generator, a network characterized by an important pres-
ence of these systems generates an equally important contribution to the inertia of the
network and its stability.
878 G. Ala et al.

Fig. 3. Active and reactive power supplied by the synchronverter, Voltage and current variations
 Stability of Microgrids: An Application 879

Fig. 4. Electromotive force induced by the virtual synchronous generator. Frequency variations
limited by the virtual synchronous generator, tend to suddenly return to the nominal value

Acknowledgements. The author would like to thank Thai Nguyen University of Technology
(TNUT), Viet Nam for the support.

References
1. Zhong, Q., Weiss, G.: Synchronverters: inverters that mimic synchronous generators. IEEE
Trans. Ind. Electron. 58(4), 1259–1267 (2011). https://doi.org/10.1109/TIE.2010.2048839
2. Zhong, Q.: Virtual synchronous machines: a unified interface for grid integration. IEEE Power
Electron. Mag. 3(4), 18–27 (2016). https://doi.org/10.1109/MPEL.2016.2614906
3. Mirafzal, B., Adib, A.: On grid-interactive smart inverters: features and advancements. IEEE
Access 8, 160526–160536 (2020). https://doi.org/10.1109/ACCESS.2020.3020965
4. Zhong, Q., Nguyen, P., Ma, Z., Sheng, W.: Self-Synchronized synchronverters: inverters
without a dedicated synchronization unit. IEEE Trans. Power Electron. 29(2), 617–630 (2014).
https://doi.org/10.1109/TPEL.2013.2258684
880 G. Ala et al.

5. Driesen, J., Visscher, K.: Virtual synchronous generators. In: 2008 IEEE Power and Energy
Society General Meeting—Conversion and Delivery of Electrical Energy in the 21st Century,
Pittsburgh, PA, USA, pp. 1–3 (2008). https://doi.org/10.1109/PES.2008.4596800
6. Viola, F., et al.: Virtual synchronous generator, an application to microgrid stability. 11th
International Conference on Renewable Energy Research and Applications, Istanbul/Turkey
7. Nguyen, M.T., Tran, H.T., Nguyen, C.V., Ala, G., Viola, F., Colak, I.: A novel framework of
hybrid harvesting mechanisms for remote sensing devices. In: 2022 IEEE 21st Mediterranean
Electrotechnical Conference (MELECON), pp. 1007–1012 (2022). https://doi.org/10.1109/
MELECON53508.2022.9842963
8. Grainger, J.J., Stevenson, W.D.: Power System Analysis. McGraw-Hill, New York (1994)
9. Fitzgerald, A.E., Kingsley, C., Umans, S.D.: Electric Machinery. McGraw-Hill, New York
(2003)
10. Liserre, M., Mastromauro, R.A., Nagliero, A.: Universal operation of small/medium-sized
renewable energy systems. Chapter 9. In: Abu-Rub, H., Malinowski, M., Al-Haddad, K.,
(eds) Power Electronics for Renewable Energy Systems, Transportation and Industrial
Applications. Wiley (2014)
11. Tran, H.T., Nguyen, M.T., Nguyen, C.V., Ala, G., Viola, F.: Hybrid solar-RF energy harvesting
mechanisms for remote sensing devices. Int. J. Renew. Energy Res. (IJRER) 12(1), 294–304
(2022)
 Effects of Design Parameters of Cab’s
Suspension System on an Agricultural Tractor
Ride Comfort

Bui Van Cuong(B) , Canh Chi Huan, Le Van Quynh, and Doan Thanh Binh

Faculty of Vehicle and Energy Engineering, Thai Nguyen University of Technology, Thai
Nguyen, Vietnam
cuongbui@tnut.edu.vn

Abstract. Cab’s suspension system plays an important role to improve vehicle

ride comfort. Therefore, a half-vehicle dynamic model of an agricultural tractor
is established to analyze the effect of the design parameters of cab’s suspension
system on an agricultural tractor vehicle ride comfort. The stiffness and damping
coefficients are analyzed respectively based on ISO 2631-1997 standard which
defines methods for the measurement of periodic, random and transient whole-
body vibration. The analysis results show that the stiffness and damping coeffi-
cients of cab’s suspension system has a great effect on vehicle ride. In addition,
the values of the stiffness and damping coefficients are respectively evaluated to
affect vehicle ride comfort.

Keywords: Agricultural tractor · Cab · Suspension system · Dynamic model ·

Ride comfort

1 Introduction
Agricultural tractors often operate under harsh environments. The source of vibration
transmitted to the body of the operator of the agricultural tractor was measured and
evaluated the impact of the vibration on the risks to health and occupational diseases
[1]. The vibration levels were emitted by agricultural tractors with different capacities,
comparing the results with the current norms in Brazil. The measurement results were
analyzed and evaluated the effect of agricultural tractors on the health and comfort
of the driver according to the international standard ISO 2631-1 [2]. Two agricultural
tractors with suspension and without suspension were measured to evaluate the effect
of ride comfort when vehicle operated under the different conditions [3]. The dynamic
model of the wheeled tractor/implement system with hydropneumatic suspension on the
front axle was proposed to evaluate the impacts of initial nitrogen volume, pressure of
accumulators, and orifice diameter of throttle valves on the vibration characteristics of
the tractor system [4]. Concept of a cab suspension system for improving the ride quality
of off-road tractors in the bounce, longitudinal, lateral, pitch and roll modes was proposed
to the effect of the cab suspension parameters on the ride performance [5]. The effects
of an active hydro-pneumatic cab-suspension system fitted to a telehandler on a driver’s

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 881–886, 2023.
https://doi.org/10.1007/978-3-031-22200-9_93
882 B. Van Cuong et al.

vibration exposure along the x-, y- and z-axes were evaluated trough both objective and
subjective assessments [6]. A vehicle-road coupled interaction model including vehicle
body, cab body and driver seat masses was proposed to analyze the performance of
a semi-active hydraulic cab isolation system (SHCIs) on an earth-moving machinery
ride comfort [7]. A nonlinear dynamic model of a double-drum vibratory roller was
proposed to control the value of the damping force of semi-active hydraulic cab mount
system in the direction of improving vehicle ride comfort [8]. The main idea of this
paper is to establish a half—vehicle dynamic model of an agricultural tractor under the
different operating conditions. The values of stiffness and damping coefficients of cab’s
suspension system are analyzed respectively based on ISO 2631-1997 standard such as
the weighted r.m.s acceleration responses of the vertical driver’s seat and the pitch angle
of the cab.

2 Vehicle Dynamic Model

To evaluate effects of the design parameters of cab’s suspension system of vehicle ride
comfort, a half-vehicle dynamic model with 6 degrees of freedom is established under
the different operating conditions, as shown in Fig. 1.

zs
ms
ks cs
zc
ϕc ls
mc Jc
lc1 lc2
kc1 cc1 kc2 cc2
ϕb zb
mb Jb
lbc1 lbc2
kf cf
z1 mf c t2
kt2
qt1 kt1 ct1 qt2
a b
Fig. 1. Half-vehicle dynamic model of an agricultural tractor

In Fig. 1, ms , mc , mb , mf are the masses of driver’s seat, cab, vehicle body and front
axle, respectively; ks , kci , kf , kti and cs , cci , cf , cti are the stiffness and damping coefficients
of suspension (isolation) systems of driver’s seat, cab, front axle and tires, respectively;
zs , zc , zb and zf the displacements of the driver’s seat, cab, vehicle body and front axle,
respectively; ϕc and ϕb are the pitch angles of cab and vehicle body, respectively; Jc and
Jb are the mass inertia moments of cab and vehicle body, respectively; a, b and l are the
distances qt1 , and qt2 are the road surface excitations, respectively (i = 1 ÷ 2).
 Effects of Design Parameters of Cab’s Suspension System 883

Based on half-vehicle dynamic model of an agricultural tractor in Fig. 1, the motion

equations are written by Newton’s law using as follows
⎧
⎪ ms z̈s = −Fs
⎪
⎪
⎪
⎪
⎪
⎪  2
⎪
⎪
⎪
⎪ m z̈
c c = Fs − Fci
⎪
⎪
⎪
⎪ i=1
⎪
⎨ J ϕ̈ = F l + F l − F l
c c c1 c1 s s c2 c2
(1)
⎪
⎪ 2
⎪
⎪
⎪
⎪ mb z̈b = Fci − Ff − Ft2
⎪
⎪
⎪
⎪ i=1
⎪
⎪
⎪
⎪ Jb ϕ̈b = Ff a − Ft2 b + Fc2 lbc2 − Fc1 lbc1
⎪
⎪
⎩
mf z̈1 = Ff − Ft1

where, Fs, Fci, Ff và Ft are the vertical forces of the suspension (isolation) systems of
driver’s seat, cab, front axle and tires.
Agricultural tractors usually work on deformed road surfaces as references [13,
15]. However, the random road surface roughness of random white noise is selected as
excitation source waveform for vehicle suspension [11, 12], the random road profile is
produced by filtering the white noise using the following mathematical model of the
road roughness

q̇(t) + 2πf0 q(t) = 2πn0 Gq (n0 )vw(t) (2)

where, Gq (n0 ) is the road roughness coefficient which is defined for typical road classes
from A to E according to ISO 8608 (2016) [9], n0 is a reference spatial frequency which
is equal to 0.1 m; v is the speed of vehicle; f0 is a minimal boundary frequency with a
value of 0.0628 Hz; n0 is a reference spatial frequency which is equal to 0.1 m; w(t) is
a white noise signal.

3 Simulation and Discussion

The motion equations of Eq. (1) are solved by using Matlab/Simulink software with
the reference parameters [4]. The acceleration responses of the vertical driver seat and
pitching angle of cab (azs and aφc ) are shown in Fig. 2 when vehicle moves on ISO class
C road surface at vehicle speed of 35 km/h.
The values of awzs and awφc according to ISO standard 2631-1:1997 [9] are chosen
as the objective functions which are defined by the formula below
⎡ ⎤ 21 ⎡ ⎤ 21

T
T
1 1
awzs = ⎣ 2
azs (t)dt ⎦ andawzs = ⎣ 2
azs (t)dt ⎦ (3)
T T
0 0

where, T is the duration of the measurements.

884 B. Van Cuong et al.

4 4

2 2

aφc(rad.s )
azs(m.s-2)

0 0

-2 -2

-4 -4
0 5 10 15 20 0 5 10 15 20
Time(s) Time(s)

(a) Vertical driver seat (b) Pitching angle of cab

Fig. 2. Acceleration responses of the vertical driver seat and pitching angle of cab

The results in Fig. 2 could be determined the values of the values of awzs and awφc
such as awzs = 1.31 m/s2 , awφc = 1.15 rad/s2 and this values satisfy uncomfortable
conditions for driver. Both stiffness and damping coefficients of cab’s suspension system
have important effect on vehicle ride comfort [14]. In order to analyze the effect of cab
suspension parameters such as stiffness and damping coefficients on vehicle ride comfort,
discussions will present in the following section.
Effect of cab suspension stiffness coefficients: To evaluate the effects of cab suspen-
sion stiffness on vehicle ride comfort, the stiffness coefficient values of kc = [0.2 0.4 0.6
0.8 1 1.2 1.4 1.6 1.8 2.0] × kc0 where kc0 = [kc1, kc2] are the stiffness coefficient values
of the original suspension system of cab were selected to evaluate their effect on vehicle
ride comfort with three damping coefficient values of cc = [0.5 1 1.5] × cc0, where cc0
= [cc1, cc2,] are the damping coefficient values of the original suspension system of cab
when vehicle moves on ISO class C road surface at vehicle speed of 35km/h. The effect
of cab suspension stiffness coefficients on awzs and awφc values are shown in Fig. 3.

1.4 1.5
cc=0.5xcc0 (N.s/m)
1.4 cc=1xcc0 (N.s/m)
1.3
awφc(rad.s )
awzs (m.s )
-2

1.3 cc=1.5xcc0 (N.s/m)

1.2
cc=0.5xcc0 (N.s/m) 1.2
1.1 cc=1xcc0 (N.s/m)
1.1
cc=1.5xcc0 (N.s/m)
1 1
0 0.5 1 1.5 2 xk 0 0.5 1 1.5 2 xkc0
kc(N.m-1) c0 kc(N.m-1)

(a) awzs values (b) awφc values

Fig. 3. Effect of cab suspension stiffness coefficients on awzs and awφc values

From Fig. 3a shows that when the value of cab suspension stiffness coefficient
increases, the awzs values tend to increase, which makes the negative effects on ride
comfort of driver. The awzs values increase quickly when the values of the damping
coefficients are c = 0.5c0 and the awzs values do not change much when the values of
the damping coefficients are achieved by great values. From Fig. 3b shows that when the
value of cab suspension stiffness coefficient increases, the awφc values tend to decrease,
which makes vehicle ride comfort of driver considerably improve. The awzs values
 Effects of Design Parameters of Cab’s Suspension System 885

increase quickly when the cc values are c = 0.5c0 and the awzs values do not change
much when the values of the damping coefficients are achieved by great values.
Effect of cab suspension damping coefficients: To evaluate the effects of cab sus-
pension damping coefficient on vehicle ride comfort, the damping coefficient values cc
= [0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2.0] × cc0, where cc0 = [cc1, cc2] are the damping
coefficient values of the original suspension system of cab the original engine cabin
suspension system are selected to evaluate their effect on the ride comfort with three
stiffness coefficient values of kc = [0.5 1 1.5] × kc0, where kc0 = [kc1, kc2,] are the
stiffness coefficient values of the original suspension system of cab when vehicle moves
on ISO class C road surface at vehicle speed of 35km/h. The effect of cab suspension
damping coefficients on awzs and awφc values are shown in Fig. 4.

1.4 1.8
kc=0.5xkc0 (N.m-1) kc=0.5xkc0 (N.m-1)
-1 1.6 kc=1xkc0 (N.m-1)
awφc (rad.s )
kc=1xkc0 (N.m )
-2
awzs (m.s )

1.35
2

kc=1.5xkc0 (N.m ) -1
1.4 kc=1.5xkc0 (N.m-1)

1.3
1.2

1.25 1
0 0.5 1 1.5 2 xcc0 0 0.5 1 1.5 2 xc
c0
cc(N.s.m-1) cc(N.s.m-1)

(a) awzs values (b) awφc values

Fig. 4. Effect of cab suspension damping coefficients on awzs and awφc values

Figure 4 shows that cab suspension damping coefficient increases, the awzs and awφc
values tends to decrease, which makes driver’s ride comfort considerably improve. And
then the awzs and awφc values tends to increase quickly, which has the negative effects
on driver’s ride comfort.

4 Conclusions
A half—vehicle dynamic model under absolutely hard road surface condition was pro-
posed in this paper. The values of stiffness and damping coefficients of cab’s suspension
system are analyzed respectively based on ISO 2631-1997 standard such as the weighted
r.m.s acceleration responses of the vertical driver’s seat and the pitch angle of the cab. The
main study conclusions drawn from the comparative evaluation results are: (1) In the cab
suspension system, the reasonable matching of the stiffness and damping coefficients of
cab suspension system can efficiently improve seat ride comfort and (2) Both stiffness
and damping coefficients of cab suspension system have important effects on vehicle
ride comfort. Especially, the stiffness and damping coefficient values of cab suspension
system are within the value ranges (0.25kc0 ≤ k ≤ 0.5 kc0 ) and (1.5cc0 ≤ c ≤ 2.0cc0 )
to improve the ride comfort of driver.

Acknowledgment. The authors wish to thank the Thai Nguyen University of Technology for
supporting this work.
886 B. Van Cuong et al.

References
1. Cutini, M., Brambilla, M., Bisaglia, C.: Whole-Body vibration in farming: background doc-
ument for creating a simplified procedure to determine agricultural tractor vibration comfort.
Agriculture 7(10), 84 (2017)
2. Clovis, F., Guilherme, A.M., Emerson, T., Murilo, M.B., Alcir, J.M.: Vibration levels emitted
by agricultural tractors. Afr. J. Agric. Res. 12(39), 2919–2925 (2017)
3. Claar, P.W., II., Sheth, P.N., Buchele, W.F., Marley, S.J.: Simulated tractor chassis suspension
system. Trans. ASAE 25(3), 0590–0594 (1982)
4. Zheng, E., Cui, S., Yang, Y., Xue, J., Zhu, Y., Lin, X.: Simulation of the vibration characteristics
for agricultural wheeled tractor with implement and front axle hydropneumatic suspension.
Shock. Vib. 2019, 1–19 (2019)
5. Rakheja, S., Sankar, S.: Suspension designs to improve tractor ride: II passive cab suspension.
SAE Trans. 93, 1105–1112 (1984)
6. Caffaro, F., Preti, C., Micheletti Cremasco, M., Cavallo, E.: Three-axial evaluation of whole-
body vibration in agricultural telehandlers: the effects of an active cab-suspension system. J.
Occup. Environ. Hyg. 14(10), 758–770 (2017)
7. Van Quynh, L., Ha, D.V., Van Cuong, B., Vu, L.A. Van Thoan, T.: Improvement of ride
comfort quality for an earth-moving machinery with semi-active cab isolation system. In:
E3S Web of Conferences, vol. 304, p. 02012 (2021)
8. Duy, N.T., Ha, D.V., Van Cuong, B.: Ride comfort evaluation for a double-drum vibratory
roller with semi-active hydraulic cab mount system. In: E3S Web of Conferences, vol. 304,
p. 01008 (2021)
9. ISO 8608: Mechanical Vibration-Road Surface Profiles-Reporting of Measured Data.
International Organization for Standardization (2016)
10. ISO 2631-1: Mechanical Vibration and Shock-Evaluation of Human Exposure to Whole-Body
Vibration, Part I: General Requirements. The International Organization for Standardization
(1997)
11. Quynh, L.V., Long, L.X., Cuong, B.V.: Study on the influence of bus suspension parameters
on ride comfort. Vibroeng. Procedia 21, 77–82 (2018)
12. Ha, D.V., Van Cuong, B., Hien, V.T.: Performance evaluation of a novel hydro-pneumatic
suspension system of a heavy truck on ride comfort. In: E3S Web of Conferences, vol. 304,
p. 01009 (2021)
13. Quynh, L.V., Thao, V.T.P., Phong, T.T.: Optimal design parameters of drum’s isolation system
for a double-drum vibratory roller. Vibroeng. Procedia 31, 74–79 (2020)
14. Le, V.Q., Zhang, J.R., et al.: Ride comfort evaluation of vibratory roller under different soil
ground. Trans. Chin. Soc. Agric. Eng. 29, 39–47 (2013)
15. Le, V.Q., Nguyen, K.T.: Optimal design parameters of cab’s isolation system for vibratory
roller using a multi-objective genetic algorithm. Appl. Mech. Mater. 875, 105–112 (2018)
 Study the Impact of Cutting Edge Number
Engaged on the Surface Integrity During
Machining Laminate Composites

Dinh-Ngoc Nguyen and Anh-Tuan Dang(B)

Faculty of International Training, Thai Nguyen University of Technology, Thai Nguyen, Vietnam
anhtuanck@tnut.edu.vn

Abstract. Machining CFRP composite normally has two issues concerned,

including minimizing machining defects which strongly influence on working
ability during the services and reducing particle dust emitted in the air which is
inhaled by operators. This article presents a study of investigating the effect of
tool geometry or a number of cutting edges engaged on the surface integrity of
trimming CFRP composites. Three carbide tool geometries, namely 2 helix flutes,
3 helix flutes, and 4 helix flutes, are selected for the experimental study. The sur-
face integrity given by each cutting tool is qualitatively evaluated by the scanning
electron microscope technique. The results show that when machining length is
still small, 4HF tools exhibit the better machining quality compared with those
given by other tool geometries. However, when machining reaches longer, it is
observed that due to the stick of microchip in the grooves between two consec-
utive cutting edges, 4HF tools generate worse machining quality, characterized
by matrix smearing and holes across machined surface, than those given by 2HF
and 3HF tools. The result of the current study can be concerned as a guidance for
selecting tool geometries for getting better surface integrity of trimming CFRP
composites.

Keywords: Laminates · Trimming · Cutting edge · Surface integrity · Surface

roughness

1 Introduction
The secondary machining operations for carbon fiber reinforced plastics (CFRPs) com-
posite are typically necessary to remove the excess materials after demolding. These
operations involve drilling, milling, trimming, turning, and grinding [1–4]. Due to the
difficulty in disposal of microchip, machining operations of CFRPs are typically carried
out without lubrication, e.g. dry machining. This is an important reason for appearing
damage, such as delamination, matrix-fiber debonding, and fiber pullout, in the machined
surfaces [5–7]. These defects can harmfully influence on the residual strength of com-
posite structures [3, 8, 9]. In order to minimize machining damage previously mentioned,
there have been methods used so far by the research community. Optimizing machining
quality, especially designing tool geometries, and reducing cutting forces have widely

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 887–894, 2023.
https://doi.org/10.1007/978-3-031-22200-9_94
888 D.-N. Nguyen and A.-T. Dang

considered by some researchers [5, 10–13]. Optimizing machining quality that attracts
many researchers [14] shows that when machining with small cutting speed and high
feed rate, occurrences of the machining damage are crucially observed, compared with
machining of high cutting speed and low feed rate. This phenomenon can be interpreted
that the combination between the high cutting speed and the small feed rate gener-
ates smaller lever of the theoretical chip thickness which makes the machining process
easy. Consequently, smoother machining surface is obtained. This result is prsesented
in different documents by other researchers [5, 14, 15].
However, the impact of the machining parameters on the surface integrity or machin-
ing surface, as well as defects has been an open issue because of various results opposite.
Indeed, compared with the results documented by [14] and several studies mentioned
above, [8, 12] have revealed that the previously mentioned tendency is only true for
machining at small machining distance. Nevertheless, at longer machining distance, the
cutting edge radii increase due to friction and temperatures in machining. This gener-
ates machined surfaces that become irregularity. At high cutting speed, the machining
area is wider, compared with small cutting speed. With the same machining time, the
higher speed leads the longer contacting time between the workpiece and the cutting
edge. Moreover, with the same machining time, cutting tool fed with small rate will
create higher. This leads to an increase in contacting time. Hence, machining with the
high cutting speed coupled with small feed speed makes the severe tool wear directly
influencing on machining quality.
Due to the natural brittle property of plastic matrix, an important component of
CFRPs, dry machining of CFRP composite naturally creates the emission of dust parti-
cles. The particles after leaving machining area can be emitted in the air and inhaled by
operators in machining area. For that reason, in addition to reducing machining damage,
minimizing dust particles in the air has attracted research community so far [8, 12].
Haddad et al. [12] shows that machining with the high cutting speed and low feed speed
generates high harmful particles, but low emitted particles are generated by high feed
speed and low cutting speed. This is related to the theoretical chip thickness firstly pro-
posed by [10]. Nguyen-Dinh [12] proposed a new tool geometry, 4SSF, which includes
the serration along the cutting edge. The results show that 4SSF tool exhibit the smaller
number of harmful particles, compared with 2SF and 2HF tool. Prakash [13] investigated
the influence of tool geometries on the machining damage. It was found that cutting tool
with the shape of trapezoid and the shape of pyramidal cutting edge creates the smallest
delamination level. The four fluted helical end mill generates maximum delamination
depth, compared with two remain tools.
The above analysis shows that tool geometries are important to the generation of
machining damage and dust particles. Balancing these issues is significant. In this paper,
the impact of cutting edge number on the surface integrity when trimming CFRP com-
posites will be experimentally investigated. Machined surfaces are generated by three
tool geometries representing by difference cutting edge engaged when trimming. For
each tool geometry including two, three and four flutes, three machining conditions
combined by on spindle speed and three feed speed will be considered for comparing
machining surface quality in both qualitative and quantitative evaluation.
 Study the Impact of Cutting Edge Number Engaged 889

2 Experimental Setup
Trimming tests are carried out by using a Mazak vertical center smart 530C as shown
in Fig. 1. The microstructures of machined surface and cutting edges are visualized by
a Scanning Electron Microscopes (SEM) machine referenced under “JEOL-JSM 5410
LV”. CFRPs in this study are laminated by using P2352 prepregs under the layer layout
of [90°/90°/−45°/0°/45°/90°/−45°/90°/45°/90°]s. Each plate is cut by the dimensions
of 300 mm × 150 mm × 5.2 mm, and the cutting direction is along 150 mm length. The
fabrication of and properties of CFRPs herein can be referred in [16, 17].
Three values of feed speed Vf , (500 mm/min, 1000 mm/min, and 1500 mm/min)
combined with only one value of spindle speed of 12000 rpm, maximum value of spindle
speed available in the selected machine tool. All cutting conditions in this study will
be carried out by using constantly 2 mm radial depth of cut. The reason for choosing
spindle speed is to make difference on machining damage in the machined surfaces. This
creates three cutting conditions for each tool geometry.
Correspondingly, there are three helix mill cutters with difference in cutting edges,
namely two helix flutes (2HF), three helix flutes (3HF), and four helix flutes (4HF)
chosen in this study as seen in Fig. 2. Each type of the cutting tool will be used three times
corresponding to the number of cutting condition selected. For each cutting condition,
twenty paths will be taken place. This means that each cutting condition corresponds to
3 m long of machining distance. The setup of evaluation of microstructures of machined
surfaces can be detailed in [16, 17].

2nd 1st

Fig. 1. Experimental setup showing devices used during trimming tests.

890 D.-N. Nguyen and A.-T. Dang

6 mm 6 mm 6 mm
(a) (b) (c)
Fig. 2. Carbide milling cutter with a two, b three, and c four helix cutting edges.

3 Results and Discussions

In order to evaluate the surface integrity resulting from three tool geometries, i.e. 2HF,
3HF, and 4HF, qualitative and qualitative comparisons will be conducted. The former
is representative by SEM observations where microstructures is considered to evaluate
machining defects. For the last, the parameter of Ten-point average roughness, Rz , is uti-
lized for characterizing machined surface in term of quantitative evaluation. To eliminate
the impact of tool wear on the experimental results, evaluating effects of tool geome-
tries will be considered at two values of machining length, Lc , 0.2 m and 3.0 m. SEM
images showing microstructures of machined surfaces given by three tool geometries
are presented in Figs. 3 and 4 for machining length of 0.2 m and Fig. 5 for machining
length of 3.0 m. It is noticed that only several representative machining conditions are
shown the SEM images. This is because of the fact that the remaining cutting conditions
similarly exhibit the topography. Based on the results presented in Fig. 3, it is seen that
machined surfaces are relatively clear, regardless of tool geometries. This is due to the
fact that when machining length is still small the cutting edges can be considered as
new or radius is small. Consequently, the trimming process easily happens, materials
are mainly cut by shearing instead of breaking because of pushing of worn cutting edge
[2]. Furthermore, it is observed that the damage level in the machined surfaces given by
2HF tools and 3HF tools is identically visualized, while those generated by 4HF tools
are minor better than the former. The mentioned phenomenon is attributed to that the
higher number of engaged cutting edge, the smaller theoretical chip thickness which can
be determined as Eq. (1). This has been documented in the research work of Haddad [8]
and Janardhan [10].
4HF tools exhibit the smaller theoretical chip thickness compared with those given
by the other tool geometries. Hence, the machining quality generated by 4HF tools is
seem better than those of the remains. One more thing should be concerned is that there
 Study the Impact of Cutting Edge Number Engaged 891

1000 µm 1000 µm
(a) (b) 1000 µm (c)
Fig. 3. SEM observations of machined surfaces for various tool geometries, a two helix flutes, b
three helix flutes, c four helix flutes when using feed speed of 500 mm/min and spindle speed of
12000 rpm at machining length of 0.2 m.

is a difference between machining quality obtained by 4HF tools and those generated by
2HF and 3HF tools, why there is no difference between machining quality resulted by
2HF and 3HF? Actually, there may be existing a limiting value of the engaged cutting
edges which can make differently machining quality.
ath = ae (Vf /Vc ) (1)
where: Vc is cutting speed (m/min), Vf is feed speed (mm/min), ac is radial depth of cut.
Figure 4 shows the microstructures of three tool geometries when using the feed
speed of 1000 mm/min and the spindle speed of 12000 rpm at the machining length of
0.2 m. It is observed that the tendency is identical to the results given by machining with
the feed speed of 500 mm/min. It means that the observed damage levels generated by
2HF and 3HF tools are smaller than those created by 4HF tools. The explanation for this
is similar to for those cut by feed speed of 500 mm/min. The pure effect of machining
parameters is confirmed when machining length is still small.
The influences of tool wear, indirectly representative by machining length, on the
microstructures of machined surfaces for three tool geometries are shown in Fig. 5. It
can be seen that the machining defects observed are clearly distinguished between those
given by three tool geometries. Indeed, in Fig. 5a there are some holes locating at the
fiber direction of −45°. These holes are characterized by fiber pullout, matrix cracking
or matrix-fiber debonding. However, matrix smearing damage is not seen in this case.
Nevertheless, this kind of damage is observed with increasing in levels in Fig. 5b, c.
Additionally, defects in form of holes at −45° fiber layer are also observed, but due to
the covering of matrix smearing phenomenon the hole defects not clearly visualized.
The discrepancy between the defects degree at machining length of 0.2 m and 3.0 m
892 D.-N. Nguyen and A.-T. Dang

1000 µm (a) 1000 µm 1000 µm (c)

(b)

Fig. 4. SEM observations of machined surfaces for various tool geometries, a two helix flutes, b
three helix flutes, c four helix flutes when using feed speed of 1000 mm/min and spindle speed of
12000 rpm at machining length of 0.2 m.

is attributed to tool wear or increasing in cutting edge radius as shown in Fig. 6. It

should be noticed that because of the tool wear, trimming process is hard happened;
hence, friction and machining temperature are also increased in this case. Consequently,
microchip is stuck in the cutting edges as shown in Fig. 6. 4HF tools, characterized by
small spaces of grooves between two next cutting edges, generate the machined surface
with highest level of matrix smearing due to the effect of stuck microchip in the grooves.
The removing microchip out machining area is difficult. Finally, the machining defects
including holes and matrix smearing given by 4HF are higher than those generated by
2HF and 3HF tools. This trend is opposite, compared with that exhibiting at machining
length of 0.2 m. Similar results are also observed for other cutting conditions of three
tool geometries which are not presented SEM images herein.

4 Conclusions

The influence of the number of the cutting edges engaged resulting from three tool
geometries of CFRP trimming on the surface integrity is investigated in this study.
Machining defect levels are considered at different machining lengths to purely ana-
lyze the influence of process parameters, number of cutting edge and tool wear on the
outcome. Scanning electron microscope technique is adopted for evaluating machining
damage. From the experimental results, several conclusions can be given as follows:

• When the machining length is still small (0.2 m), the machined surface given by three
tool geometries is clean. 4HF tools exhibit the better machining quality than those
generated by other tool geometries. This is attributed to the fact that the theoretical
 Study the Impact of Cutting Edge Number Engaged 893

Matrix smearing

Matrix smearing
Holes
Holes

Holes

1000 µm (a) 1000 µm (b) 1000 µm (c)

Fig. 5. SEM observations of machined surfaces for various tool geometries, a 2HF, b 3HF, c 4HF
when using feed speed of 500 mm/min and spindle speed of 12000 rpm at machining length of
3.0 m.

New cutting edge Worn cutting edge

Emitted particles

200 µm (b) 200 µm

(a)

Fig. 6. SEM images of cutting edge of 4HF tool at cutting distance of a 0.2 m, b 3.0 m when
using feed speed of 1000 mm/min and spindle speed of 12000 rpm.

chip thickness created by 4HF is smaller than those generated by 2HF and 3HF. Hence,
trimming process is happened more easily than machining with 2HF and 3HF.
• Trimming at the longer machining length (3.0 m) generates several kinds of machining
defects such as holes, matrix cracking, fiber-matrix debonding, and matrix smearing.
This is because of the effect of the tool wear which quickly accelerates increasing in
friction and temperatures in machining areas.
• Because of the difference from tendency at small machining length, 4HF tools gen-
erate machining surfaces are more irregular, compared with those given by 2HF and
3HF. This is because of the smaller spaces in the grooves between next cutting edges
engaged leading the removing chip process difficult. As a result, friction and temper-
atures increase lead the microchip particles stuck in the grooves and the formation of
matrix smearing in across machined surfaces as observations.
894 D.-N. Nguyen and A.-T. Dang

Acknowledgments. The author wish to thank Thai Nguyen University of Technology for
supporting this work.

References
1. König, W., et al.: Machining of fibre reinforced plastics. CIRP Ann. 34(2), 537–548 (1985)
2. Ghidossi, P., El Mansori, M., Pierron, F.: Edge machining effects on the failure of polymer
matrix composite coupons. Compos. Appl. Sci. Manuf. 35(7–8), 989–999 (2004)
3. Ghidossi, P., Mansori, M., Pierron, F.: Influence of specimen preparation by machining on
the failure of polymer matrix off-axis tensile coupons. Compos. Sci. Technol. 66(11–12),
1857–1872 (2006)
4. Azmi, A.I.: Chip formation studies in machining fibre reinforced polymer composites (2013)
5. Haddad, M., Zitoune, R., Eyma, F., Castanié, B.: Machinability and surface quality during
high speed trimming of multi directional CFRP. Int. J. Mach. Mach. Mater.
6. Sheikh-Ahmad, J., Shahid, A.H.: Effect of edge trimming on failure stress of carbon fiber
polymer composites (2013)
7. Ghafarizadeh, S., Lebrun, G., Chatelain, J.-F.: Experimental investigation of the cutting tem-
perature and surface quality during milling of unidirectional carbon fiber reinforced plastic.
J. Compos. Mater. 50(8), 1059–1071 (2015)
8. Haddad, M., et al.: Study of the surface defects and dust generated during trimming of CFRP:
influence of tool geometry, machining parameters and cutting speed range. Compos. Appl.
Sci. Manuf. 66, 142–154 (2014)
9. Nguyen-Dinh, N., Bouvet, C., Zitoune, R.: Influence of machining damage generated during
trimming of CFRP composite on the compressive strength. J. Compos. Mater. 54(11), 1413–
1430 (2019)
10. Janardhan, P., Sheikh-Ahmad, J., Cheraghi, H.: Edge trimming of CFRP with diamond
interlocking tools (2006)
11. Nguyen-Dinh, N., et al.: Surface integrity while trimming of composite structures: X-ray
tomography analysis. Compos. Struct. 210, 735–746 (2019)
12. Nguyen-Dinh, N., et al.: New tool for reduction of harmful particulate dispersion and to
improve machining quality when trimming carbon/epoxy composites. Compos. Part A: Appl.
Sci. Manuf. 131 (2020)
13. Prakash, R., et al.: High-speed edge trimming of CFRP and online monitoring of performance
of router tools using acoustic emission. Materials (Basel) 9(10) (2016)
14. Sheikh-Ahmad, J., Urban, N., Cheraghi, H.: Machining Damage in Edge Trimming of CFRP.
Mater. Manuf. Processes 27(7), 802–808 (2012)
15. Ucar, M., Wang, Y.: End-milling machinability of a carbon fiber reinforced laminated
composite-M-Ucar.pdf (2003)
16. Ngoc, N.D., Hue, N.T.: A Study on Qualitative and Quantitative Characterization of Machin-
ing Quality of Aerospace Composite Structures, in Advances in Engineering Research and
Application, pp. 94–101 (2020)
17. Ngoc, N.D., Truong, P.V., Hue, N.T.: Effects of machining configurations and process param-
eters on the machining damage generated during milling CFRP structures. In: Advances in
Engineering Research and Application, pp. 400–406 (2022)
 Studying Optimal Set of Input Parameters
for CBN Grinding Aluminum 6061T6 on CNC
Milling Machine

Bui Thanh Danh1 , Luu Anh Tung2 , Tran Huu Danh3 , Tran Minh Tan4 ,
Trinh Kieu Tuan5 , Duong Van Phuong2 , and Nguyen Dinh Ngoc2(B)
1 University of Transport and Communications, Hanoi, Vietnam
2 Thai Nguyen University of Technology, Thai Nguyen, Vietnam
ngocnd@tnut.edu.vn
3 Vinh Long University of Technology Education, Vı̃nh Long, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 University of Economics - Technology for Industries, Hanoi, Vietnam

Abstract. This article presents a multi-objective optimization process of surface

grinding for Aluminum alloy 6061 in CNC machine tools. The aim is to find the
optimal set of grinding parameters that can simultaneously satisfy maximizing
material removal speed (MRS) and minimizing surface roughness. The partial fac-
tor method 24–1 is used to determine the number of tests. Four process parameters
are chosen as input parameters, namely spindle speed (Rpm ), feed rate (Fe ), depth
of cut (aed ), down feed (Df ). The multi-objective function optimization becomes
only the optimization of Composite Desirability function (CDF). Regression func-
tions to predict both surface roughness and material removal rate are constructed
based on the experimental results. The results reveal that when CDF reaches the
value of 0.669, it can be found the minimum surface roughness of 0.19 µm and
MRS reaches a maximum value of 16.23 (g/h).

Keywords: Al 6061 · CBN grinding · Surface roughness · Material removal

speed · Multi-objective optimization

1 Introduction

Grinding is an important operation in machining to get the finished machined surface

ensuring final dimension requirements before assembly. The accuracy and surface finish
resulting from grinding can be up to ten times compared with those obtained by milling
or turning operations. The grinding mechanism can be constructed based on the rotating
of grinding wheels. Typically, abrasive grains which are held with a binder are always
needed. The cutting tool in this context is the mentioned abrasive grains which can
remove microchips from the workpiece. In order to get the effectiveness, there are some
conditions for the grinding process that should be followed as the hardness of abrasive
grains is higher than that of a workpiece, abrasives are friable, and heat-resistant abrasive

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 895–902, 2023.
https://doi.org/10.1007/978-3-031-22200-9_95
896 B. T. Danh et al.

wheels are required [1–4]. The fracture of abrasive grains can be replaced when they
become worn and dull.
In grinding Aluminum oxide, cubic boron nitride (CBN) which is super abrasive
using a percentage of grinding is about 5%. The hard materials of a workpiece are
contacted with abrasive grains with CBN materials [5]. Normally, the machining quality
is also controlled by the grain size of abrasive materials due to the removing microchip
from the machining area. The grinding process is quick if coarse and larger grains are
used, oppositely it can be obtained a better surface finish if smaller abrasive grains
are adopted [6–8]. In addition to abrasive grains, binders, a component in grinding
wheel material, are used to keep the abrasive grains bonding to each other. In industry,
depending on the shape required, there have been several kinds of grinding as cylindrical
grinding, internal grinding, surface grinding, profile grinding, or combination grinding.
Regarding surface grinding operation, generating mostly external surfaces, the straight
surface, tapered surface, or contoured surface may be produced.
Researchers have documented that surface grinding is typically impacted by some
factors such as abrasive kind, grain size, binder, shape, and dimension of workpiece,
mechanical properties, and chemical components, work speed, depth of cut, feed speed,
dressing condition, etc. Recently, there have been researchers focusing on reducing
surface roughness and increasing the material removal rate (MRR) by determining an
optimum set of grinding parameters [9, 10]. The combination of Taguchi method and
genetic algorithm (GA) is mostly used [11, 12]. Silicon carbide grinding was studied
by [2] in which Particle Swarm Optimization technique is utilized for identifying the
optimal set of machining parameters through the objective function of increasing MRR
[13–15]. The same studying objectives were also conducted by Kumar [16]. The support
tool in this study was Minitab software. A similar group of authors introduced the
empirical models to predict surface roughness and MRR by using response surface
methods (RMS). Additionally, an integrated Genetic Algorithm-Neutral Network (GA-
NN) has been widely utilized by various researchers [11, 17].
The aim of this article is to find the optimum grinding parameters for machining
Aluminum 6061. The critical factors influencing the responses are purposely chosen
including spindle speed, feed rate, depth of cut, and down feed. The composite desir-
ability function approach is applied to the optimization process. A software package
Minitab@19 is used to integrate various models and implement the experimental design.

2 Experimental Procedure
2.1 Design of Experiment
The optimal set of grinding parameters can be found by considering the effect of the
following parameters: spindle speed (Rpm ), feed rate (Fe ), depth of cut (aed ), down
feed (Df ) on surface roughness and MRS. The survey levels of input parameters and
details are exhibited in Table 1. Experimental planning of 2k including 4 to evaluate the
influence of the main parameters on Ra and MRS is utilized herein. In order to reduce
the number of experiments, the partial factor method 24–1 = 8 experiments are used. For
this reason, at a resolution degree of 4, there will be no overlapping each other between
the key factors or other two-way interactions. Minitab@19 software is used to design
 Studying Optimal Set of Input Parameters 897

the experimental matrix. The procedure for declaring experimental variables is shown
in Fig. 1. The grinding process setup can be shown in Fig. 2. The experimental matrix
and the measured MRS results are listed in Table 2.

Table 1. Input parameters and survey levels

Factor Code Unit Low High

Rpm Rpm rpm 4000 5000
Feed rate Fe mm/min 2000 3000
Depth of dressing cut aed mm 0.01 0.025
Down feed Df mm/double journey 0.5 2

Fig. 1. Declaration procedure in minitab software

Fig. 2. Experimental setup

898 B. T. Danh et al.

Table 2. Experimental matrix and measured results

Std Order Rpm Fe aed Df Means of Ra (µm) Means of MRS (g/h)

1 4000 2000 0.01 0.5 0.1497 3.4779
7 4000 3000 0.025 0.5 0.1510 4.3989
6 5000 2000 0.025 0.5 0.1257 3.5631
4 5000 3000 0.01 0.5 0.1300 4.0997
2 5000 2000 0.01 2 0.3377 10.6490
8 5000 3000 0.025 2 0.1900 16.2297
5 4000 2000 0.025 2 0.1857 11.6366
3 4000 3000 0.01 2 0.1673 13.9366

2.2 Optimization Process of Multi-objective Function

The multi-objective optimization problem is expected to simultaneously satisfy two
objectives, namely the minimum surface roughness and the maximum material removal
speed. For this reason, the multi-objective optimization problem is considered based on
the Composite Desirability function (CDF) which is synthesized from the Desirability
function of each singular objective. In this case, the multi-objective function optimization
becomes only the optimization of CDF. The CDF is calculated by the formula (1):

D = (r1 +r2 +...+rn ) d1r1 · d2r2 . . . dnrn (1)

where: di is CDF value of the ith objective, n is the number of output indicators (n = 2);
ri is the importance of the ith indicator. In case of the criteria have the same importance,
the expected function is calculated according to the formula (2):

D = n d1 · d2 . . . dn (2)

CDF values are to get all objectives when D reaches the maximum value. For the
objective of the best minimum surface texture, the CDF is calculated by the formula (3):
⎧
⎪
⎨  1 w yi ≤ Ti
Ui −yi
di = U −T T i < yi < Ui (3)
⎪
⎩ i i
0 yi ≥ Ti

In which: Ui is the upper limit of the acceptable region when minimizing; w is the
weight of the objective function yi .
For the objective of the maximum MRS as the best, the CDF is calculated by the
formula (4):
⎧
⎪
⎨  0 w yi ≤ Li
yi −Li
di = T −L Li < yi < Ti (4)
⎪
⎩ i i
1 yi ≥ Ti
 Studying Optimal Set of Input Parameters 899

where: Li is the lower limit of the maximum acceptable region; w is the weight of the
objective function yi .
In order to determine the expectation function of each individual objective, it is
necessary to determine the extreme value of each output objective by defining a regres-
sion equation describing the relationship between the objective function and the input
variables.
The second order regression model for k variables has the form:

i−1
y = b0 + bi xi + xi2 + bij xi xj (5)
i=1 i=1 i=1 j=1

where: xi , xj are the input variables, bi are the coefficients.

The extreme point of the function y is the point at which partial y partial derivatives
are equal and zero, as described in Eq. (6).
∂y ∂y ∂y
= = ... = =0 (6)
∂x1 ∂x2 ∂xk

3 Results and Discussion

Due to a large number of input variables, it is difficult to manually calculate, hence,
Minitab software is used to determine the regression equations with the output criteria Ra
and MRS. Based on the experimental results which can help construct a regression model
to predict the surface roughness and material removal rate. The determined regression
equation is described in Table 3.

Table 3. Regression equation describing the influence of input variables on output indicators

Regression equation in uncoded units

Ra = − 0.6441 + 0.000199 Rpm + 0.000244 Fe +
23.55 aed – 0.2755 Df – 0.000000 Rpm * Fe – 0.005723 Rpm * aed
MRS = 28.68 – 0.007727 Rpm – 0.004182 Fe – 902.4 aed + 3.873 Df + 0.000001 Rpm * Fe
+ 0.2141 Rpm * aed

In order to get the optimal values of each output objective, it should solve the
mentioned regression equations. Due to a large number of variables in the equa-
tions, Minitab@19 is adopted to get the solution. The smallest surface roughness value
determined is described in the analysis results in Table 4.
From the analysis results in Table 4, it is observed that the smallest possible roughness
value is 0.054 µm corresponding to the value of CDF is 0.64. The results of the analysis
to determine the maximum material removal rate are described in Table 5. The expected
value of MRS, in this case, is 16.2297 corresponding to CDF value of 0.8198.
900 B. T. Danh et al.

Table 4. Analytical results to determine the minimum surface roughness value

Response optimization: Ra
Parameters
Response Goal Lower Target Upper Weight Importance
Ra Minimum 0 0.15 1 1
Solutions
Solution Rpm Fe aed Df Ra fit Composite desirability
1 5000 3000 0.025 0.5 0.054 0.64

Table 5. Analytical results determine the value of the MRS

Response optimization: MRS

Parameters
Response Goal Lower Target Upper Weight Importance
MRS Maximum 15 16.5 1 1
Solutions
Solution Rpm Fe aed Df MRS fit Composite desirability
1 5000 3000 0.025 2 16.2297 0.819802

It is noticed that the objectives of minimizing surface roughness and the maximum
MRS. However, for the first goal, the allowable limit of the extreme region depends
on the required surface quality of the workpiece. Therefore, it is possible to expand
the extreme area of the surface texture to increase the material removal rate as much as
possible. For this reason, the surface texture can still be ensured according to the technical
requirements. The calculation and determination of the optimal mode and CDF of the
multi-objectives are done using Minitab@19.
After expanding the extreme area of the surface texture, the analysis results are
described in Table 6. Detailed information of the extreme value of CDF and the optimum
set of machining parameters are presented in Table 7. It is visualized that the optimal
cutting mode is Rpm = 5000 rpm, Fe = 3000 mm/min, aed = 0.025 mm, Df = 2 mm/
double journey. With the optimal technology mode determined, the output parameters
corresponding to this technology mode are Ra = 0.19 µm and MRS = 16.2297 g/h. The
value of the multi-objective expectation function is d = 0.6687.

4 Conclusion
The optimum set of grinding parameters that can simultaneously reduce the surface
roughness and increase material removal rate during grinding of Aluminum alloy 6061
is considered in this paper. The investigated parameters are spindle speed (Rpm ), feed
 Studying Optimal Set of Input Parameters 901

Table 6. Optimization analysis results to achieve Ra min and MRS max

Response optimization: Ra, MRS

Parameters
Response Goal Lower Target Upper Weight Importance
Ra Minimum 0.14 0.25 1 1
MRS Maximum 15 16.50 1 1
Solutions
Solution Rpm Fe aed Df Ra fit MRS fit Composite
desirability
1 5000 3000 0.025 2 0.19 16.2297 0.819802

Table 7. Optimum set of machining parameters

Multiple response prediction

Variable Setting
Rpm 5000
Fe 3000
aed 0.025
Df 2
Response Fit
Ra 0.1900
MRS 16.23

rate (Fe ), depth of cut (aed ), down feed (Df ). Thanks to using the CDF, the multi-objective
problems can be equivalently changed to find the unique value of CDF. There are some
conclusions as follows can be made:

• Regression functions to predict both surface roughness and material removal rate are
constructed based on the experimental results.
• The optimum set of grinding parameters are Rpm = 5000 rpm, Fe = 3000 mm/min,
aed = 0.025 mm, and Df = 2 mm/double journey.
• A single CDF value gives Ra min, and another CDF gives MRS max. Hence, to
agree on a common value of CDF, it should be expanded the boundary. This leads to
the result that when the CDF approaches 0.669, Ra reaches the maximum value of
0.054 µm and the maximum MRS reaches 16.23 (g/h).

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

902 B. T. Danh et al.

References
1. Gopal, A.V., Rao, P.V.: Selection of optimum conditions for maximum material removal rate
with surface finish and damage as constraints in SiC grinding. Int. J. Mach. Tools Manuf.
43(13), 1327–1336 (2003)
2. Ting, T.O., Lee, T.S., Htay, T.: Performance analysis of grinding process via particle
swarm optimization. In: Proceedings of the Sixth International Conference on Computational
Intelligence and Multimedia Applications (2005)
3. Abdur-Rasheed, A., Konneh, M.: Optimization of precision grinding parameters of silicon for
surface roughness based on taguchi method. Adva. Mater. Res. 264–265, 997–1002 (2011)
4. Wu, S., et al.: A simulation platform for optimal selection of robotic belt grinding system
parameters. Int. J. Adv. Manuf. Technol. 64(1–4), 447–458 (2012)
5. Hoang, T.D., Tran, T.H., Cuong, N.V., Le, H.K., Nga, N.T.T.: An optimization study on surface
grinding stainless steel. Int. J. Eng. Technol. 7 (2018)
6. Pai, D., Rao, S., D’Souza, R.: Application of response surface methodology and enhanced
non-dominated sorting genetic algorithm for optimisation of grinding process. Procedia Eng.
64, 1199–1208 (2013)
7. Kumar, P., Kumar, A., Singh, B.: Optimization of process parameters in surface grinding
using response surface methodology. IJRMET 3(2) (2013)
8. Periyasamy, S., et al.: Optimization of surface grinding process parameters for minimum
surface roughness in AISI 1080 using response surface methodology. Adv. Mater. Res. 984–
985, 118–123 (2014)
9. Karande, M.R.J.: Optimization of cylindrical grinding machine parameters for minimum
surface roughness and maximum MRR. GRD J. Eng. 2 (2017)
10. Selvaraj, D.P., Chandramohan, P.: Optimization of surface roughness of AISI 304 austenitic
stainless steel in turning operation using Taguchi method. IJARIIE (2017)
11. Sedighi, M., Afshari, D.: Creep feed grinding optimization by an integrated GA-NN system.
J. Intell. Manuf. 21(6), 657–663 (2009)
12. Samuel, A.U., Araoyinbo, A.O., Elewa, R.R., Biodun, M.B.: Effect of machining of alu-
minium alloys with emphasis on aluminium 6061 Alloy—A review. In: International
Conference on Engineering for Sustainable World (2020)
13. Li, F., et al.: Optimization of grinding parameters for the workpiece surface and material
removal rate in the belt grinding process for polishing and deburring of 45 steel. Appl. Sci.
10(18) (2020)
14. Samuel, A.U., et al.: Effect of machining of aluminium alloys with emphasis on aluminium
6061 alloy—A review. IOP Conf. Ser.: Mater. Sci. Eng. 1107(1) (2021)
15. Tamil Vanan, S.K., et al.: Evaluation of surface grinding of AISI 304 stainless steel using dry
and compressed air cooling techniques. SN Appl. Sci. 3(3) (2021)
16. Kumar, M., Singh, S., Goyal, K.: To study the effect of grinding parameters on surface
roughness and material removal rate of cylindrical grinding of heat treated en 47 steel. J.
Mech. Eng. 45 (2015)
17. Lee, P.H., Chung, H., Lee, S.W.: Optimization of micro-grinding process with compressed air
using response surface methodology. Proc. Inst. Mech. Eng. Part B: J. Eng. Manuf. 225(11),
2040–2050 (2011)
 Studying the Braking Efficiency of the Tractor
Semi-trailer on a Straight Road with Road
Surface Profiles According to ISO 8608–2016

Nguyen Thanh Tung(B) and Luong Van Van

Vinh Long University of Technology Education, 73 Nguyen Hue St, Vinh Long City, Vietnam
tungnt@vlute.edu.vn

Abstract. The tractor semi-trailer often causes accidents when braking. In this
paper, the author uses the multi-body system (MBS) method and the Newton-Euler
method to establish a system of vehicle dynamics equations. Matlab-Simulink
software is used to survey the braking efficiency of vehicles on a straight road
with road surface profiles according to ISO 8608–2016. When a tractor semi-
trailer is run straight at 80 km/h on road surface profiles A, B, and C, the slip ratio
of wheels is less than 10%; the braking acceleration of the vehicle is 6.2 m/s2 ,
meeting the requirements of ECE-R13 standard [1]. If the tractor semi-trailer is
run on road surface profiles D, E, and F, the braking force value is reduced by
about 23 ÷ 44%; the slip ratio of the middle and rear axle wheels reaches 100%;
the braking acceleration of the vehicle is less than 5.8 m/s2 , which does not meet
the requirements of ECE-R13 standard [1].

Keywords: Braking efficiency · Tractor · Semi-trailer · ISO 8608:2016

1 The Dynamics Model

In this paper, the authors studied the braking efficiency of the six-axle tractor semi-
trailer. The tractor semi-trailer consists of a tractor with three axles and a semi-trailer
also with three axles. The tractor and semi-trailer are linked together by a fifth wheel.
The tractor semi-trailer has a large size and weight, and a complex structure. To build
a dynamic model, the multi-body system method is used to build a three-dimensional
dynamic model of the tractor semi-trailer as shown in Fig. 1 [2–4].
The system of Newton-Euler equations is used to build the system of dynamics
equations for the tractor semi-trailer as follows [5–7]:


3
(mc1 + mAi )(ẍc1 − ψ̇c1 ẏc1 ) = (Fx11 − FR11 )cosδ11 + (Fx12 − FR12 )cosδ12
1
(1)
−Fy11 sin δ11 − Fy12 sin δ12 − (FR2j + FR3j ) + (Fx2j + Fx3j ) − Fwx1 − Fkx1

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 903–909, 2023.
https://doi.org/10.1007/978-3-031-22200-9_96
904 N. T. Tung and L. Van Van

zc2

c2

C2
xc2
Fkz2 c2

K2 c2
Fkx2
Mkx2
Fky2 yc2
Fw1 yc1 zc1
c1
Fkz1 M51 M61
Fky1 M41
Fy51 Fy61
hw1 1 hc2 Fy41
K1 Fkx1
Mkx1
C1 Fx61
xc1 c1
Fx41
Fx51

Fz51 Fz61 A6
Fz41 A4 A5 M62
hc1 M31 M42 M52
M21
M11 Fy31 Fy42 Fy52 Fy62
Fx62
Fy21 Fx52
Fy11
Fx31 Fx42 Fz52 Fz62
Fx11 A3 Fz42
Fx21 A2 Fz31 M32
Fz21 M22 Fy32
A1 l4
Fz11
M12 Fy22
Fx32
zAi ij
l5
Fy12 Fx22 Fz32
Fz22 l6
l2 l3
lk2
Fx12 l1
Fz12 lk1

Fig. 1. The dynamics model of the tractor semi-trailer


3
(mc1 + mAi )(ÿc1 + ψ̇c1 ẋc1 ) = (Fx11 − FR11 ) sin δ11 − Fky1
1
(2)
+(Fx12 − FR12 ) sin δ12 + Fy11 cosδ11 + Fy12 cosδ12 + (Fy2j + Fy3j )
Jzc1 ψ̈c1 = [(Fx1j − FR1j ) sin δ1j + Fy1j cosδ1j ]l1 + (Fx12 cosδ12 − Fx11 cosδ11 )b1
+(Fy11 sin δ11 − Fy12 sin δ12 + FR11 cosδ11 − FR12 cosδ12 )b1 − Fy2j l2 − Fy3j l3 (3)
+(Fx22 − Fx21 + FR21 − FR22 )b2 + (Fx32 − Fx31 + FR31 − FR32 )b3 + Fky1 lk1
 
6
mc2 + mAi (ẍc2 − ψ̇c2 ẏc2 ) = (Fx4j − FR4j ) + (Fx5j − FR5j ) +
4
(Fx6j − FR6j ) + Fkx2 (4)


6
(mc2 + mAi )(ÿc2 + ψ̇c2 ẋc2 ) = Fky2 + Fy4j + Fy5j + Fy6j (5)
4

Jzc2 ψ̈c2 = (Fx42 − Fx41 + FR41 − FR42 )b4 + (Fx52 − Fx51 + FR51 − FR52 )b5
(6)
+(Fx62 − Fx61 + FR61 − FR62 )b6 + Fky2 lk2 − Fy4j l4 − Fy5j l5 − Fy6j l6

mc1 (z̈c1 − ϕ̇c1 ẋc1 ) = FCij + FKij − Fkz (i = 1 ÷ 3) (7)

 Studying the Braking Efficiency of the Tractor Semi-trailer … 905

mc2 (z̈c2 − ϕ̇c2 ẋc2 ) = FCij + FKij − Fkz (i = 4 ÷ 6) (8)


Jyc1 ϕ̈c1 = −(FCij + FKij )li + Fxij (hc1 − ri ) + Fkxi (hc1 − hk1 )
+ Fkz1 lk1 + Mij (i = 1 ÷ 3) (9)


Jyc2 ϕ̈c2 = −(FCij + FKij )li + Fxij (hc2 − ri ) + Fkxi (hc2 − hk2 )
+ Fkz2 lk2 + Mij (i = 4 ÷ 6) (10)


i=3 
i=3
Jxc1 β̈c1 = (FC2i + FK2i − FC1i − FK1i )wi + Fi (hc − hBi ) − Mkx1 (11)
i=1 i=1


i=6 
i=6
Jxc2 β̈c2 = (FC2i + FK2i − FC1i − FK1i )wi + Fi (hc − hBi ) − Mkx2 (12)
i=4 i=4

mAi (z̈Ai + β̇Ai ẏAi ) = FCLij + FKLij − FCij − FKij (i = 1 ÷ 6; j = 1 ÷ 2) (13÷18)

mAi (ÿAi − β̇Ai żAi ) = Fi + Fyij (i = 1 ÷ 6; j = 1 ÷ 2) (19÷24)

JAxi β̈Ai = (FCi1 + FKi1 − FCi2 − FKi2 )wi

+ (FCLi2 + FKLi2 − FCLi1 − FKLi1 )bi − Fyij (rij + ξAij ) + Fi (hBi − ri )
(25÷30)
JAyij ϕ̈ij = MAij − MBij − FRij rdij (i = 1 ÷ 6; j = 1 ÷ 2) (31÷42)

where i = 1 ÷ 6 is axle number of the tractor semi-trailer; j = 1 is left wheel; j = 2

is right wheel; xc , yc , zc are displacement in the B (Cxc yc zc ) coordinate system; δ is
front-wheel turn angle; li is longitudinal distance of axle i from the mass center; hc1,
hc2 are height of mass center from the ground; bi is lateral distance of a wheel from the
longitudinal x-axis; ϕij is the rotation angle of the ij wheel; β, ϕ, ψ are the rotation angle
of the tractor body around the x, y, z axis; m, mA are sprung mass and un-sprung mass
of the tractor semi-trailer; Jx , Jy , Jz are the moment of inertia about the x, y, z axis of the
sprung mass; JAyij is the moment of inertia about the y-axis of the ij wheel; FCij is the
suspension elastic force of the ij wheel; FKij is the suspension damping force of the ij
wheel; FCLij is the tire elastic force of the ij wheel; Fxij is the longitudinal force of the
ij wheel; Fyij is the lateral force of the ij wheel; Fzij is the vertical force of the ij wheel;
FGij is the static weight of the ij wheel; MAij is the driving torque of the ij wheel; MBij
is the braking torque of the ij wheel.

2 The Results and Discussions

This paper used Matlab-Simulink software to investigate the braking efficiency of the
tractor semi-trailer. The tractor semi-trailer is run straight at 80km/h on road surface pro-
file according to ISO 8608:2016 [8] and then braked with a torque of MB = 80%MBmax
906 N. T. Tung and L. Van Van

Fig. 2. The braking torque of the tractor semi-trailer

= 80%FZ ϕxmax rd . Lines 1, 2, 3, 4, 5, and 6 correspond to road surface profile F, E, D,

C, B, and A of ISO 8608–2016. The braking torque of the tractor semi-trailer is shown
in Fig. 2.
The braking force of the tractor is shown in Fig. 3. On 6 types of road surface profile
A, B, C, D, E and F, the braking force of front axle has the same value FX1 = 19.5kN.
For axles 2 and 3, when the tractor semi-trailer is run on road surface profile A, B, C,
and D, FX2 ≈FX3 ≈26kN; on road surface profile E, FX2 ≈FX3 ≈20kN, the braking force
value is reduced by about 23%; on road surface profile F, FX2 ≈FX3 ≈16kN, the braking
force value is reduced by about 38.5%.

Fig. 3. The braking force of the tractor (x104 N)

Figure 4 shows the braking force of the semi-trailer. On three types of road sur-
face profile A, B, and C, the braking force of the semi-trailer has the same value
 Studying the Braking Efficiency of the Tractor Semi-trailer … 907

FX4 ≈FX5 ≈FX6 ≈25kN; on road surface profile D, FX4 ≈FX5 ≈FX6 ≈20kN, the braking
force value is reduced by about 20%; on road surface profile E, FX4 ≈FX5 ≈FX6 ≈17kN,
the braking force value is reduced by about 32%; on road surface profile F,
FX4 ≈FX5 ≈FX6 ≈14kN, the braking force value is reduced by about 44%.

Fig. 4. The braking force of the semi-trailer (x104 N)

The slip ratio of the tractor is shown in Fig. 5. For the front axle, on six types of road
surface profiles A, B, C, D, E, and F, the slip ratio has the value S1 ≈3 ÷ 5%, and the
wheels work stably. For axles 2 and 3, when tractor semi-trailer is run on road surface
profile A, B, and C, S2 ≈S3 ≈5%, the wheels work stably; on road surface profile D, E,
and F, S2 ≈S3 ≈100%, the wheels are completely slipped on the road surface.

Fig. 5. Slip ratio of the tractor

908 N. T. Tung and L. Van Van

Figure 6 is a graph showing the slip ratio of the semi-trailer. On three types of road
surface profile A, B, and C, the slip ratio has the value S4 ≈S5 ≈ S6 ≈10%, the wheels
work stably. On 3 types of road surface profile D, E, and F, S4 ≈S5 ≈S6 ≈100%, the
wheels are completely slipped on the road surface.

Fig. 6. Slip ratio of the semi-trailer

The braking acceleration of the tractor and the semi-trailer are shown in Fig. 7.

Fig. 7. The braking acceleration of the tractor semi-trailer

We see that the braking acceleration of the the tractor and the semi-trailer has the
same shape and value. When the tractor semi-trailer is run on road surface profile A,
B, and C, the braking acceleration of the vehicle is ax1 ≈ ax2 ≈6.2 m/s2 , meeting the
requirements of ECE-R13 standard [1]. If the tractor semi-trailer is run on road surface
profile D, the braking acceleration of the vehicle is ax1 ≈ ax2 ≈5.6 m/s2 , and on road
surface profile E, the braking acceleration of the vehicle is ax1 ≈ax2 ≈4.9 m/s2 and on
road surface profile F, ax1 ≈ ax2 ≈4.0m/s2 , which does not meet the requirements of
ECE-R13 standard [1].

3 Conclusions
When the tractor semi-trailer is run straight at 80km/h on road surface profile A, B
and C according to ISO 8608:2016, the braking force of axles is FX1 = 19.5kN,
FX2 ≈FX3 ≈26kN, FX4 ≈FX5 ≈FX6 ≈25kN; the slip ratio of wheels is less than 10%; the
braking acceleration of the vehicle is ax1 ≈ ax2 ≈6.2m/s2 , meeting the requirements of
ECE-R13 standard [1]. If the tractor semi-trailer is run on road surface profile D, E and
F, the braking force value is reduced by about 23 ÷ 44%; the slip ratio of the middle and
rear axle wheels reaches 100%, the wheels are completely slipped on the road surface;
the braking acceleration of the vehicle is less than 5.8m/s2 , which does not meet the
requirements of ECE-R13 standard [1].
 Studying the Braking Efficiency of the Tractor Semi-trailer … 909

References
1. The international Braking Regulation of the UNO-Economic Commission for Europe in Geneva
- ECE-Regulation No. 13, ECE-R13, Vehicle Regulations 2004
2. Jazar, R.N.: Vehicle Dynamics Theory and Application. Springer, Switzerland (2017)
3. Tung, N.T., Huong, V.V, Kiet, P.T.: Experimental research on determining the vertical tyre
force of a tractor semi-trailer. Int. J. Mod. Phys. B (2020). https://doi.org/10.1142/S02179792
20401633
4. Thanh Tung, N: Experimental analysis for determination of longitudinal friction coefficient
function in braking tractor semi-trailer. Int. J. Eng. (2021). https://doi.org/10.5829/ije.2021.
34.07a.24
5. Tung, N.T., Van Van, L., Quang, N.T.: A survey on the effects of bumpy road on the vibration
of multi-purpose forest fire fighting vehicle. Eng. Solid Mech. (2021). https://doi.org/10.5267/
j.esm.2021.3.001
6. Tung, N.T., Van Van, L.: A study on the effects of tire vertical stiffness on dynamic load of
DVM 2.5 truck. In: Nguyen, D.C., Vu, N.P., Long, B.T., Puta, H., Sattler, K.-U. (eds.) ICERA
2021. LNNS, vol. 366, pp. 128–134. Springer, Cham (2022). https://doi.org/10.1007/978-3-
030-92574-1_13
7. Tung, N.T., Van Van, L.: A research on the effects of suspension stiffness on dynamic load on
Dongfeng DVM 2.5 truck. Lecture Notes in Mechanical Engineering (2022). https://doi.org/
10.1007/978-3-030-99666-6_27
8. International standard ISO 8608: Mechanical vibration - Road Surface Profiles - Reporting of
Measured Data (2016)
 The Design of the Compound Non-circular Gear
Train Has an Internal Gears Pair
with an Improved Cycloid Tooth Profile

Nguyen Thanh Trung1 , Phung Van Thom2 , and Nguyen Hong Thai2(B)
1 National Research Institute of Mechanical Engineering, HaNoi, Vietnam
2 School of Mechanical Engineering, Hanoi University of Science and Technology, Hanoi,
Vietnam
thai.nguyenhong@hust.edu.vn

Abstract. This paper presents a design method for a compound non-circular gear
train (CNCG) with a pair of internal gears and mathematical model synthesis of
the mating centrode for the CNCG train according to the mating centrode of gear 1
and gear 4. A CNCG train design with a variable speed and torque range from 0.4
to 1.8, after evaluating and discussing the kinematic characteristics of the CNCG
train with the criterion is that the mating centrode of the non-circular gears in the
CNCG train is convex curves. In the CNCG train, the outer ring gears are shaped
by the rack cutter, while the shaper cutter shapes the inner ring gear.

Keywords: Mating centrode · Gear ratio function · Non-circular gear ·

Compound non-circular gear train · Internal gears

1 Introduction
Creating variable transmission functions with compact size and simple mechanical struc-
ture are the advantages of the non-circular gear (NCG). In recent years, it has become
the research topic of scientists, especially the research on the application of NCG in the
mechanism design and equipment such as: Ottaviano et al. apply a pair of non-circular
gears to design a volumetric pump blood circulation in cardiac surgery [1]; Bair et al.
researched the correction of elliptical gear pairs in hydraulic pump design [2]; The design
of an NCG pair for modifying the crank-slider kinematics of a nail machine [3]; Man-
ufacturing a pair of the elliptical gears in the flow meter [4]; Application of oval gear
pairs in Roots blower design [5]. In addition to the researches applying a pair of NCG
above, there are also studies on the application of the NCG train, such as: Applying the
CNCG train in combination with cylindrical gears in designing the gear drive of the new
coal seam gas (CSG) drainage machine in CSG mining industry [6]; Application of oval
gear train to design rice transplanters [7]; Zhao et al. [8, 9] Who improved the CNCG
train of Guo’s rice transplanter and developed it to transplant other leafy vegetables.
In addition, there are other studies on planetary NCG train like the followings: Volkov
et al. [10] application of planetary NCG train for a rotary hydraulic machine; D. Mundo
synthesized a planetary gear train with the center of movable planet gear [11], etc.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 910–923, 2023.
https://doi.org/10.1007/978-3-031-22200-9_97
 The Design of the Compound Non-circular Gear Train … 911

However, most of the above research applies two types of curves as tooth profiles,
namely using the involutes curve of a circle and the Novikov type arc, while other curves
such as the cycloid curve [12–14], the ellipse involute [15, 16] or an improved cycloid
curve formed from an ellipse generating rolling without slipping on the circle [17–19],
etc., has not applied in the design of tooth profiles of the NCG.
Another problem is that in most machine mechanisms, there is a working stroke and
an idle stroke. Therefore, it is necessary to synthesize the link sizes when designing the
machine to achieve the desired kinematic characteristics [20–22]. To solve the above
problem in this research, the authors design a CNCG train with the scheme described in
Fig. 1. To meet the above kinematic characteristics with the tooth profile of the NCG in
a CNCG train is the improved cycloid curve of the ellipse generating.

Fig. 1. Scheme of the compound non-circular gear train.

In the scheme Fig. 1, gear 1 is elliptical gear, compound gear 2–3 are non-circular
gears, while gear 4 is eccentric elliptical gear.

2 Design of the Mating Centrodes for the Compound Non-circular

Gear Train

To establish the mathematical model of the mating centrodes of the CNCG train as
described in Fig. 1, we proceed to develop each NCG pair as shown in the following
section.
912 N. T. Trung et al.

2.1 A Mathematical Model Describing the Mating Centrodes of Internal

Non-circular Gears Pair 1–2

In this case, the mating centrode of gear 1 is oval, and the equation is written in polar
coordinates [23]:
2a1 b1
ρ1 (ϕ1 ) = (1)
(a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )
wherein: a1 , b1 are the semi-major and semi-minor axis of mating centrode oval Σ 1 and
the polar angle ϕ 1 ∈ [0 ÷ 2π] (rad), respectively.

Fig. 2. Geometric relationship of mating centrodes pair 1–2.

Thus, from Fig. 2, if Σ 2 is called the mating centrode of NCG 2 conjugation with oval
gear 1, the calculation method is used in [24] then the mathematical equation describing
mating centrode Σ 2 is determined by:
⎧ 2a1 b1
⎪
⎪ ρ2 (ϕ2 (ϕ1 )) = a12 +
⎪
⎪ (a + ) − (a
⎨ 1 b1 1 − b1 ) cos(2ϕ1 )
ϕ 1 (2)
⎪
⎪ 2a1 b1
⎪
⎪ ϕ (ϕ ) = d ϕ1
⎩ 2 1
a12 ((a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )) + 2a1 b1
0

where: ρ2 (ϕ2 (ϕ1 )), ϕ2 (ϕ1 ) are the polar radius and polar angle of Σ 2 at point P2 ,
respectively.
 The Design of the Compound Non-circular Gear Train … 913

The following relationship determines the a12 axis distance:

2π  
2a1 b1
f (n1 , a1 , b1 ) = 2π − n1 d ϕ1 = 0
a12 ((a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )) + 2a1 b1
0
(3)
With n1 is the number of revolutions of oval gear 1 so that NCG 2 can rotate one
revolution. Solving Eq. (3) by integrating Dwight [25], the axis distance a12 (n1 , a1 , b1 )
is given by:


−(a1 + b1 ) + (a1 + b1 )2 − 4a1 b1 (1 − n21 )
f (a12 ) = a12 (n1 , a1 , b1 ) = (4)
2

2.2 A Mathematical Model Describing the Mating Centrodes of External Geared

Non-circular Gears Pair 3–4
In this case, gear 4 has a mating centrode Σ 4 which is an eccentric ellipse and the
mathematical equation as polar coordinates described by [26]:
1
ρ4 (ϕ4 ) = a4 (1 − E42 ) (5)
1 − E4 cos ϕ4

b24
wherein: E4 = 1− a42
is the eccentricity with a4 , b4 are the semi-major and semi-minor
axis of the mating centrode Σ 4 , and the polar angle ϕ4 ∈ [0 ÷ 2π ] (rad), respectively.

Fig. 3. Geometric relationship of mating centrodes pair 3–4.

According to [27, 28], from Eq. (5) and Fig. 3, the equation of the mating centrode
Σ 3 conjugation to Σ 4 is determined by:
⎧
⎪
⎨ ρ3 (ϕ3 (ϕ4 )) = a34 − a4 (1 − E4 ) 1−E4 cos
2 1
 ϕ4
ϕ4 a4 (1−E42 ) (6)
⎪
⎩ ϕ3 (ϕ4 ) = d ϕ4
a34 (1−E4 cos ϕ4 )−a4 (1−E42 )
0
914 N. T. Trung et al.

wherein: ρ3 (ϕ3 (ϕ4 )), ϕ3 (ϕ4 ) are the pole radius and pole angle of the mating centrode
Σ 3 respectively.
The a34 axis distance is determined as follows:

2π  
a4 (1 − E42 )
f (n4 , a4 , b4 ) = 2π − n4 d ϕ4 = 0 (7)
a34 (1 − E4 cos ϕ4 ) − a4 (1 − E42 )
0

With n4 is the number of eccentric elliptical gear 4 so that NCG 3 can rotate one
revolution. Same with gear pair 1–2, after solving Eq. (7), determine the axis distance
a34 (n4 , a4 , b4 ) as follows:
 
f (a34 ) = a34 (a4 , b4 , n4 ) = a4 1 + 1 + (1 − E42 )(n24 − 1) (8)

3 Kinematic Analysis of the Compound Non-circular Gear Train

To analyze the kinematics of the CNCG train, replace &#x006A; as the pole angle in the
mating centrode design of the train with φ the rotation angle of the gear shafts. Thus,
the gear ratio function of the CNCG train is given by:

ω1 d ϕ1
i14 (ϕ1 ) = = = i12 (ϕ1 )i34 (ϕ4 (ϕ1 )) (9)
ω4 d ϕ4
wherein:
ω1
i12 (ϕ1 ) =
ω2
 
d ϕ1 ρ2 (ϕ2 (ϕ1 )) a12 ((a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )) + 2a1 b1
= = = and
d ϕ2 ρ1 (ϕ1 ) 2a1 b1
 
ω3 d ϕ3 ρ4 (ϕ4 ) a4 (1 − E42 )
i34 (ϕ4 (ϕ1 )) = = = =
ω4 d ϕ4 ρ3 (ϕ3 (ϕ4 )) a43 (1 − E4 cos ϕ4 ) − a4 (1 − E42 )

After transforming, Eq. (9) is rewritten as:

i14 (a1 , b1 , a4 , b4 )
  
a12 ((a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )) a4 (1 − E42 )
= +1
2a1 b1 a43 (1 − E2 cos ϕ4 ) − a4 (1 − E42 )
(10)

Thus, the relationship of angular velocity ω2 of the shaft two according to the input
angular velocity ω1 is given by:

2a1 b1
ω2 (ω1 ) = ω1 (11)
a12 ((a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )) + 2a1 b1
 The Design of the Compound Non-circular Gear Train … 915

And the angular velocity ω4 of the eccentric oval gear four according to the input
angular velocity ω1 :

ω4 (a1 , b1 , a4 , b4 )
  
a34 (1 − E4 cos ϕ4 ) 2a1 b1
= − 1 ω1
a4 (1 − E42 ) a12 ((a1 + b1 ) − (a1 − b1 ) cos(2ϕ1 )) + 2a1 b1
(12)

4 Influence of Centrode Coefficient on Kinematic Characteristics

of the CNCG Train

From Eqs. (10) and (12), it can be seen that the gear ratio function i14 and the output
angular velocity ω4 are functions that depend on the design parameters of Σ 1 and Σ 4 .
To evaluate the influence of the parameters of Σ 1 and Σ 4 on the gear ratio function
characteristics of the CNCG train, set the centrode coefficient λ1 = b1 /a1 and λ2 =
b4 /a4 , then investigate the following cases:

4.1 Case 1: Influence of the Centrode Coefficient λ1 on Kinematic Characteristic

for the CNCG Train
With a4 = 45 mm, b4 = 35 mm, n4 = 3, &#x0077;1 = 5 rad/s, a1 = 50 mm, b1 =
45 mm, n1 = 3 the coefficients λ1 is calculated so that the circumference &#x0053;1
constant and satisfy Eq. (3). After numerical calculations are a program in Matlab, the
design parameters of the mating centrode for the CNCG train are statistics in Table 1
below.

Table 1. The mating centrodes design parameters of the CNCG train according to λ1 .

The design λ1 a1 [mm] b1 [mm] a12 [mm] a34 [mm]

1 0.90 50 45.00 94.76 153.74
2 0.70 55 38.75 91.17 153.74
3 0.51 60 30.75 82.02 153.74

From Table 1, we have Fig. 4a as a graph of the gear ratio function and Fig. 4b as a
graph of the output angular velocity ω4 according to ω1 corresponding to λ1 .
From the data in Table 1, we have the design mating centrodes of the CNCG train
according to coefficient λ1 as described in Fig. 5.
916 N. T. Trung et al.

Fig. 4. Influence of coefficient λ1 on kinematic characteristics of the CNCG train with (a) Gear
ratio i14 and (b) Output shaft angular velocity ω4 .

Fig. 5. Design the mating centrodes for the CNCG train according to coefficient λ1 .

4.2 Case 2: Influence of the Centrode Coefficient λ2 on Kinematic Characteristic

for the CNCG Train

IN this case a1 = 50 mm, b1 = 45 mm, n1 = 3, n4 = 2, ω1 = 5 rad/s, a4 = 40 mm, b4 =

35 mm, the coefficients λ2 is calculated so that the circumference Σ 4 constant and satisfy
Eq. (7). After numerical calculations is a program in Matlab, the design parameters of
the mating centrode for the CNCG train are statistics in Table 2.

Table 2. The mating centrodes design parameters of the CNCG train according to λ2 .

The design λ2 a4 [mm] b4 [mm] a12 [mm] a34 [mm]

1 0.875 40 35.00 94.76 146.77
2 0.630 45 28.34 94.76 136.94
3 0.380 50 19.12 94.76 123.65
 The Design of the Compound Non-circular Gear Train … 917

From Table 2, we have Fig. 6a as a graph of the gear ratio function and Fig. 6b as a
graph of the output angular velocity ω4 according to ω1 corresponding to λ2 .

Fig. 6. Influence of coefficient λ2 on kinematic characteristics of the CNCG train with a Gear
ratio i14 and b Output shaft angular velocity ω4 .

Figures 4 and 6 show that when reducing the coefficients λ1 and λ2 , the axis distance
decreases while the speed variation and gear ratios increase. But there is a sudden speed
variation at the output shaft (see Figs. 4b and 6b). From here, an interesting thing can be
seen in this design that in a working cycle, the gear drive will have two working modes:
speed increase and speed reducer.
From the data in Table 2, we have the design mating centrodes of the CNCG train
according to coefficient λ2 as described in Fig. 7.

Fig. 7. Design the mating centrodes for the CNCG train according to coefficient λ2 .

Figures 5 and 7 show that: (i) When the coefficient λ1 is small, then Σ 1 strain. As
a result, the mating centrode Σ 3 is a concave curve and causes sudden speed changes.
When λ1 = 0.51, the gear ratio function also splits at the top (see the dotted rectangle
on Fig. 4a); (ii) When the coefficient λ2 is small, the center of rotation moves to the
semi-major axis of the ellipse Σ 4 , resulting in not formed elliptical gears (see Fig. 7c).
918 N. T. Trung et al.

5 Discuss and Evaluate Calculation Results

In this section, to evaluate more accurately, call: δi14 = δi14 max − δi14 min , δω4 =
δω4 max − δω4 min respectively the amplitude of the gear ratio function i14 and the output
angular velocity ω4 , with the mating centrode designs of the CNCG train according to
the coefficients λ1 and λ2 in Tables 1 and 2. As a result, we have a graph describing the
amplitude of the gear ratio and the output angular velocity, as shown in Fig. 8.

Fig. 8. Amplitude of the gear ratio and output shaft speed for the designs with (a) according to
the coefficient λ1 and (b) according to the coefficient λ2 .

From Fig. 8 with the condition that the mating centrode circumference Σ 1 and Σ 4
are constant, then: (i) When the coefficient λ1 increases from 0.51 to 0.9, the amplitude
of the gear ratio for the CNCG train increases, while the amplitude of the speed the
output shaft angle again decreases. Specifically, with the design at λ1 = 0.9 compared to
the plan at λ1 = 0.51, the amplitude of the gear ratio δi14 increases by 18.09%, while the
amplitude of the angular velocity δω4 decreases by 17.79%; (ii) When the coefficient λ2
decreases from 0.875 to 0.38, the amplitude of the gear ratio and the output shaft angular
velocity increases, specifically with the design at λ 2 = 0.875 compared to the plan at
λ2 = 0.38, the amplitude is of gear ratio δi14 increased by 575.17%, and amplitude of
angular velocity δω4 increased by 572.94%.

6 Experimental Design

From the above discussion and evaluation, a mating centrode design of a CNCG train
with the design parameters was chosen as statistics in Table 3. At the same time, Fig. 9
shows the mating centrodes and gear ratio function of the CNCG train.
From the centrodes design parameter data in Tables 3 and Fig. 9, the method of gear
shaping the NCG in the CNCG train is as follows:
Gear pair 1–2: Oval gear 1 (external ring gear) gear shaping by a rack cutter 1 with
tooth profile is an improved cycloid, while NCG 2 (inner ring gear) gear shaping by gear
shaper 2 with tooth profile is an improved cycloid as shown on Fig. 10. The ellipse Σ E
generates an improved tooth profile with parameters aE = 1.6 mm, bE = 1.36 mm.
 The Design of the Compound Non-circular Gear Train … 919

Table 3. Design parameters for the mating centrodes of the CNCG train.

Parameters Notation Gears pair 1–2 Gears pair 3–4

Elliptical gear 1 NCG 2 NCG 3 Eccentric elliptical
gear 4
The semi-major axis a1 55.00 – – –
Σ 1 [mm]
The semi-minor axis b1 38.75 – – –
Σ 1 [mm]
Circumference n 3.00 3.00
coefficient
The semi-major axis a4 – – – 40.00
Σ 4 [mm]
The semi-minor axis b4 – – – 35.00
Σ 4 [mm]

Fig. 9. The CNCG train with a The mating centrodes and b The gear ratio.

Gear pair 3–4: Because gear pairs 3–4 are NCGs have external ring gear, both gears
3 and 4 are gears shaping with rack cutter 2 which has design parameters given in Table 4.
The parameter of the ellipse Σ E generates the tooth profile of the rack cutter 2, in this
case aE = 1.7 mm, bE = 1.42 mm.
Tables 3 and 4 have the CNCG train design described in Fig. 10, and the set of design
parameters of the train are statistics in Table 5 (Fig. 11).

7 Conclusion

From the above theoretical research results and experimental design, this research has
the following new points:
920 N. T. Trung et al.

Fig. 10. Gear shaping tool parameters with a The rack cutter and b The gear shaper.

Table 4. Design parameters of rack cutter.

Parameter Notation Rack cutter 1 Gear shaper 2 Rack cutter 2

Module [mm] M 5.95 5.95 6.26
Tooth thickness [mm] sc 9.34 9.34 9.83
Width of space [mm] wc 9.34 9.34 9.83
Tooth pitch [mm] pc 18.68 18.68 19.66
Tooth addendum [mm] ha 3.20 3.20 3.40
Tooth dedendum [mm] hf 3.20 3.20 3.40
Whole depth [mm] H 6.40 6.40 6.80
Radius of centrode circle [mm] Rgs – 35.70 –

(1) In the design process of the CNCG train, depending on the operational charac-
teristics of the system, the design coefficients λ1 and λ2 are selected to reconcile
 The Design of the Compound Non-circular Gear Train … 921

Fig. 11. The experimental design of the CNCG train.

Table 5. Design parameters of NCG in the CNCG train.

Parameter Notation Gear pair 1–2 Gear pair 3–4

Oval gear 1 NCG 2 NCG 3 NCG 4
Module [mm] m 5.95 5.95 6.26 6.26
Circumference coefficient n 3.00 – – 3.00
Number of teeth z 16 48 36 12.00
Tooth pitch [mm] p 18.68 18.68 19.66 19.66
Tooth thickness [mm] s 3.20 3.20 3.40 3.40
Width of space [mm] w 3.20 3.20 3.40 3.40
Whole depth [mm] h 6.40 6.40 6.80 6.80
Tooth addendum [mm] ha 3.20 3.20 3.40 3.40
Tooth dedendum [mm] hf 3.20 3.20 3.40 3.40

the mechanical structure and working features. So, for example, when you want to
increase the speed and torque range, reduce λ2 rather than reduce λ1 , but reducing
λ2 too much results in not forming elliptical gear 4.
(2) A numerical calculation program has been written on Matlab to investigate and
design a CNCG train with at least a pair of internal gears. The gear drive has an
exceptional feature of working in two modes: speed increase and speed reducer. In
addition, the tooth profile of the NCG in the CNCG train is an improved cycloid
generated from an ellipse.
922 N. T. Trung et al.

These research results have important implications for the creation of mechanical
drives in machines and equipment with working stroke and idle stroke without needing
to coordinate many mechanisms.

Acknowledgements. This research was funded by the Hanoi University of Science and Tech-
nology, code: T2022-PC-030. We would like to thank the editorial board of the ICERA 2022
conference.

References
1. Ottaviano, E., Mundo, D., Danieli, G.A., Ceccarelli, M.: Numerical and experimental analysis
of non-circular gears and cam-follower systems as function generators. Mech. Mach. Theory
43(8), 996–1008 (2008)
2. Bair, B.W., Sung, M.H., Wang, J.S., Chen, C.F.: Tooth profile generation and analysis of oval
gears with circular-arc teeth. Mech. Mach. Theory 44, 1306–1317 (2009)
3. Niculescu, M., Andrei, L., Cristescu, A.: Generation of noncircular gears for variable motion
of the crank-slider mechanism. In: 7th International Conference on Advanced Concepts in
Mechanical Engineering, Materials Science and Engineering, p. 147 (2016)
4. García Hernández, C., Gella Marín. R.M., HuertasTalón, J.L., Efkolidis, N., Kyratsis, P.:
WEDM manufacturing method for noncircular gears, using CAD/CAM software. Strojniški
Vestn.-J. Mech. Eng. 62, 137–144 (2016)
5. Trung, N.T., Viet, N.H., Thai, N.H.: Research and manufacturing of oval gear pair applied in
rotor of a roots type compressor. Eng. Technol. Sustain. Dev. 31(5), 31–38 (2021)
6. Xu, G., Hua, D., Dai, W., Zhang, X.: Design and performance analysis of a coal bed gas
drainage machine based on incomplete non-circular gears. Energies 2–19 (2017)
7. Guo, L.S., Zhang, W.J.: Kinematic analysis of a rice transplanting mechanism with eccentric
planetary gear trains. Mech. Mach. Theory 36, 1175–1188 (2001)
8. Zhao, X., Chu, M., Ma, X., Dai, L., Ye, B., Chen, J.: Research on design method of non-
circular planetary gear train transplanting mechanism based on precise poses and trajectory
optimization. Adv. Mech. Eng. 10(12), 1–12 (2018)
9. Zhao, X., Ye, J., Chu, M., Dai, L., Chen, J.: Automatic scallion seedling feeding mechanism
with an asymmetrical high-order transmission gear train. Chin. J. Mech. Eng. 33(1), 1–14
(2020). https://doi.org/10.1186/s10033-020-0432-9
10. Volkov, G.Y., Kurasov, D.A., Gorbunov, M.V.: Geometric synthesis of the planetary mecha-
nism for a rotary hydraulic machine. Russ. Eng. Res. 38(1), 1–6 (2018). https://doi.org/10.
3103/S1068798X18010161
11. Mundo, D.: Geometric design of a planetary gear train with non-circular gears. Mech. Mach.
Theory 41, 456–472 (2006)
12. Thai, N.H., Giang, T.C.: Influence of geometrical dimensions on the profile slippage in the
hypogerotor pump. Vietnam J. Sci. Technol 56(4), 482–491 (2018)
13. Thai, N.H., Trung, N.T., Nghia, L.X., Duong, N.T.: Synthesis of the external non-circular
gear-train with cycloid profile. J. Sci. Technol. Tech. Univ. 145, 033–039 (2020)
14. Thai, N.H., Trung, N.T., Nghia, L.X., Duong, N.T.: Profile sliding phenomenon in the external
non-circular gear-train with cycloidal profile. Eng. Technol. Sustain. Dev 31(2), 053–057
(2021)
15. Thai, N.H.: Shaping the tooth profile of elliptical gear with the involute ellipse curve. Sci.
Technol. Dev. J.-Eng. Technol. 4(3), 1048–1056 (2021)
 The Design of the Compound Non-circular Gear Train … 923

16. Thai. N.H., Thom. P.V, Lam. D.B.: Effects of pressure angle on uneven wear of a tooth profile
of an elliptical gear generated by ellipse involute. Sci. Tech. Dev. J. 24(3), 2031–2043 (2021)
17. Hsieh, C.F.: A new curve for application to the rotor profile of rotary lobe pumps. Mech.
Mach. Theory 87, 70–81 (2015)
18. Thai, N.H., Trung, N.T.: A novel curve for the generation of the non-circular gear tooth profile.
Int. J. Eng. Trans. B: Appl. 35(5), 1024–1036 (2022)
19. Thai, N.H., Ly, T.T.K., Trung, N.T.: Research design and experimental manufacturing of com-
pound non-circular gear train with an improved cycloid profile of the ellipse. In: International
Conference on Engineering Research and Applications, pp. 813–825 (2021)
20. Uicker, J.J., Pennock, G.R., Shigley, J.E., Mccarthy, J.M.: Theory of Machines and
Mechanicsms, Oxford University Press (2003)
21. Kolovsky, M.Z., Evgrafov, A.N., Semenov, Y.A., Slousch, A.V.: Advanced Theory of
Mechanisms and Machines. Springer, Berlin (2000)
22. Thai, N.H., Trung, N.T.: Pitch line synthesis of noncircular planetary gears. J. Sci. Technol.
Tech. Univ 140, 05–010 (2020)
23. Thai. N.H., Thom. P.V., Trung. N.T.: Influence of centrodes coefficient on the characteristic
of gear ratio function of the compound non-circular gear train with improved cycloid tooth
profile. In: IFToMM Asian Conference on Mechanism and Machine Science, pp. 204–214
(2021)
24. Thai. N.H., Thom. P.V.: Research on the characteristics of tooth shape and size of the oval
gear drive with an involute profile. In: The AUN/SEED-Net Joint Regional Conference in
Transportation, Energy, and Mechanical Manufacturing Engineering, pp. 167–184 (2021)
25. Litvin, F.L., Fuentes-Azna, A., Gonzalez-Perez, I., Hayasaka, K.: Noncircular Gears Design
and Generation. Cambridge University Press (2009)
26. Thai. N.H., Thom. P.V., Trung. N.T.: Experimental design and manufacture a pair of the
internal non-circular gears with an improved cycloid profile. In: The AUN/SEED-Net Joint
Regional Conference in Transportation, Energy, and Mechanical Manufacturing Engineering,
pp. 118–134 (2021)
27. Litvin, F.L., Fuentes, A.: Gear Geometry and Applied Theory. Cambridge University Press
(2004)
28. Viet, N.H., Thai, N.H.: Geometric design and kinematics analysis of non-circular planetary
gear train with cycloid profile. Eng. Technol. Sustain. Develop 31(3), 105–112 (2021)
 Effect of Main Design Factors on Two-Stage
Helical Gearbox Length

Tran Huu Danh1 , Nguyen Hong Linh2 , Bui Thanh Danh3 , Tran Minh Tan4 ,
Nguyen Van Trang5 , Tran Thi Phuong Thao5 , and Nguyen Manh Cuong5(B)
1 Vinh Long University of Technology Education, Vinh Long, Vietnam
2 Electric Power University, Hanoi, Vietnam
3 University of Transport and Communications, Hanoi, Vietnam
4 Nguyen Tat Thanh University, Ho Chi Minh City, Vietnam
5 Thai Nguyen University of Technology, Thai Nguyen, Vietnam

nmcuong@tnut.edu.vn

Abstract. This paper reports the results of an optimization study on the effect
of major design factors on the length of a two-stage helical gearbox. Five major
design factors were investigated in the study including the first stage gear ratio, the
coefficient of wheel face width of stages 1 and 2, and the allowable contact stress
of stages 1 and 2. A simulation experiment was also designed and accomplished by
a computer program. Furthermore, Minitab R19 software was applied to analyze
the experimental results. The impact of key design factors on gearbox length was
investigated. The optimum values of the parameters to obtain the smallest gearbox
length were also proposed.

Keywords: Helical gearbox · Main design factor · Optimum gearbox design ·

Gearbox length

1 Introduction
Mechanical drive systems are the most common type among many types of drive systems
such as electric drive, pneumatic drive, hydraulic drive, and so on. This is due to its
straightforward structure, dependable operation, and low cost. A typical mechanical
drive system includes a motor, a gearbox, and two couplings, or a coupling and a V-
belt or chain drive (Fig. 1). Of the mechanical drive system elements, the gearbox is
undeniably the most important because it is the main component to reduce speed and
torque from the motor shaft to the working shaft. Therefore, optimal design of gearboxes
is an urgent research topic.
There have been numerous studies on the optimal design of gearboxes up to this
point. In [1] gearbox geometric design parameters were optimized to reduce rattle noise
in an automotive transmission using a torsional vibration model approach. The authors of
[2] presented a study on multi-objective optimization for the drivetrain design and gear
shifting control of internal combustion engine vehicles, with the goal of minimizing fuel
consumption, exhaust emissions, and gearbox power losses. The optimum partial gear

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 924–932, 2023.
https://doi.org/10.1007/978-3-031-22200-9_98
 Effect of Main Design Factors on Two-Stage Helical Gearbox Length 925

ratios to minimize the cost of a three-stage helical gearbox were determined in [3]. The
problem of constrained multi-objective non-linear optimization of planetary gearboxes
using a hybrid metaheuristic algorithm was reported in [4]. In [5] a modal-based design
optimization of a gearbox housing using Finite Element Analysis was reported. In [6] a
multi-objective optimization of a two-stage helical gearbox with a variety of constraints
was described. The optimum gear ratios of different types of gearboxes have been found
such as helical gearboxes [6–9], bevel gearboxes [10–12] and worm gearboxes [13–17].
According to the results of the above analysis, despite the fact that many studies on the
optimization of gearbox parameters have been conducted, no optimization study has
determined key design parameters (Fig. 1).

Fig. 1. Schema of a mechanical drive system: (1) Motor; (2) Coupling; (3) Gearbox; (4) Chain
drive; (5) Belt conveyor

This paper presented an optimization study to determine the optimum main design
parameters for a two-stage helical gearbox to obtain the shortest gearbox length. A
simulation experiment was carried out using the Taguchi method and the Minitab R19
software. The effect of the main design parameters on gearbox mass was investigated.
The best values for the five most important design factors have been assigned.

2 Methodology
2.1 Calculation of Gearbox Length

The length of a two-stage helical gearbox Lgb can be detemined by (see Fig. 2):

dw11 dw22
Lgb = + aw1 + aw2 + + 20 (1)
2 2
In Eq. (1), dw11 is the diameter of the drive gear of stage 1; dw22 is the diameter of
the driven gear of stage 2; These diameters are found by [18]:

dw11 = 2 · aw1 /(u1 + 1) (2)

926 T. H. Danh et al.

Fig. 2. Calculating schema

dw22 = 2 · aw2 · u2 /(u2 + 1) (3)

In Eqs. (2) to (3), u1 and u2 are the gear ratios of stage 1 and stage 2; u2 = ug /u1 ;
with ug as the gearbox ratio; aw1 and aw2 are the center distances of stage 1 and stage 2
which are determined by [18]:

aw1 = ka · (u1 + 1) · 3 T11 · kH β /(AS21 · u1 · Xba ) (4)

aw2 = ka · (u2 + 1) · 3
T12 · kH β2 /(AS22 · u2 · Xba2 ) (5)
where, kH β is the contacting load ratio for the pitting resistance; kH β = 1.05 ÷ 1.27 [18]
and it was chosen as kH β = 1.16; AS1 and AS2 is the allowable contact stress of the
 Effect of Main Design Factors on Two-Stage Helical Gearbox Length 927

first and the second stages (MPa); ka = 43 is material coefficient (for steel gear) [18];
Xba1 and Xba2 are the wheel face width coefficients of stage 1 and stage 2; T11 and T12
are the torques on the drive gear of stage 1 and stage 2 (Nmm):
 
T11 = Tout / ug · η2hg · ηb3 (6)
 
T12 = Tout / u2 · ηhg · ηbe
2
(7)

In which, Tout is the output torque (N.mm); ηhg= 0.96 ÷ 0.98 is the efficiency of a
helical gear unit [18]; ηbe = 0.99 ÷ 0.995 is the rolling bearing efficiency [18].

2.2 Optimization Problem

From the above analysis, the optimization problem is describes as:

MinimizeLgb (8)

With

mgb = f (u1 ; Xba1 ; Xba2 ; AS 1 ; AS 2 ) (9)

And with the following constraints:

1 ≤ u1 ≤ 9; 1 ≤ u2 ≤ 9 (10)

3 Simulation Experiment
To investigate the impact of main design factors on gearbox length, a simulation exper-
iment was carried out. The following design parameters were investigated in this exper-
iment: u1 , Xba1 , Xba2 , AS1 , and AS2 . Table 1 defines these parameters and their levels.
For the experimental design and data analysis, the Minitab R19 software and the Taguchi
method were exploited.
To reduce computer programming workload, the impact of the main design param-
eters on the length of the gearbox was investigated using gear ratio values of 5, 10, 15,
20, 25, and 30. Furthermore, for this experiment, a 5-level for 5 factors Taguchi design
(L25) was chosen, so the simulation experiment was conducted based on 25 test runs
with each of the above values of gear ratios. Table 2 shows the test plan and the output
results (the gearbox length) for the gear ratio of 5.

4 Results Discussion
To evaluate the effect of the main design factors on Lgb for ugb = 20, the Analysis
of Variance (ANOVA) method is used in accordance with Minitab R19 software. The
signal-to-noise ratio, or S/N number, is calculated for each experiment to find the impact
928 T. H. Danh et al.

Table 1. Main design parameters and their levels

Factor Level
1 2 3 4 5
Total gearbox ratio u1 1 3 5 7 9
Coefficient of wheel face width of stage 1 Xba1 0.25 0.27 0.29 0.31 0.33
Coefficient of wheel face width of stage 2 Xba2 0.32 0.34 0.36 0.38 0.4
Allowable contact stress of stage 1 AS1 (Mpa) 350 368 386 404 420
Allowable contact stress of stage 2 AS2 (Mpa) 350 368 386 404 420

Table 2. Experimental plan and output results (Lgb ) for ugb = 5

Trial u1 Xba1 Xba2 AS1 AS2 Lgb (kg)

1 1.78 0.25 0.32 350 350 348.93
2 1.78 0.27 0.34 368 368 331.11
3 1.78 0.29 0.36 386 386 315.14
4 1.78 0.31 0.38 404 404 300.74
5 1.78 0.33 0.4 420 420 288.55
6 1.89 0.25 0.34 386 404 317.84
…
24 2.22 0.31 0.36 368 350 325.67
25 2.22 0.33 0.38 386 368 310.21

of each main design factor on the output results. The S/N ratios are calculated to minimize
gearbox length by:
 n 
1 2
S/N = −10log10 yi (11)
n
i=1

The average SN ratio for each parameter and level is calculated after calculating
the SN ratio for each experiment. Table 3 and Fig. 2 show how the main design factors
affect Lgb . Figure 2 shows that Lgb is inversely proportional to all five major design
parameters. Furthermore, Table 1 displays that AS2 has the greatest influence on Lgb
(49.09%), followed by AS1 (20.2%), Xba2 (18.22%), and Xba1 (17.22%). (11.59%).
Additionally, u1 has almost no effect on Lgb (0.89%). The influence order of the main
design factors on the gearbox length is shown in Table 4.
Using the objective function in Equation, the S/N value is maximized for each major
design factor to obtain the shortest gearbox length (8). The best main design factors are
discovered after analyzing the effect of each factor on the S/N ratio in the plot in Fig. 3:
u1 = 2.2; Xba1 = 0.33; Xba2 = 0.4; AS1 = 420 (MPa); AS2 = 420 (MPa) (Fig. 4).
 Effect of Main Design Factors on Two-Stage Helical Gearbox Length 929

Table 3. Analysis of variance for means

Analysis of variance for means

Source DF Seq SS Adj SS Adj MS F P C(%)
u1 4 27.72 27.72 6.929 94.75 0 0.89
Xba1 4 360.78 360.78 90.196 1233.36 0 11.59
Xba2 4 566.91 566.91 141.726 1938 0 18.22
AS1 4 628.55 628.55 157.137 2148.73 0 20.20
AS2 4 1527.57 1527.57 381.892 5222.1 0 49.09
Residual error 4 0.29 0.29 0.073 0.01
Total 24 3111.81
Model summary

Table 4. The order of impact of main design parameters on gearbox length

Response table for means

Level u1 Xba1 Xba2 AS1 AS2
1 316.9 320.8 322.0 322.5 326.7
2 315.4 317.4 318.4 318.5 320.2
3 314.8 314.9 314.6 314.5 314.5
4 314.3 312.1 311.7 311.4 309.0
5 313.8 310.1 308.6 308.4 304.8
Delta 3.1 10.7 13.5 14.1 22.0
Rank 5 4 3 2 1

Continue as before for the remaining ugb values of 10, 15, 20, 25, and 30. The best
values for the main design parameters are shown in Table 5. The following findings were
obtained from Table 5 and Fig. 5:

• The optimal Xba1 and Xba2 values are their maximum values: Xba1 = 0.33 and Xba2
= 0.4. In order to obtain a minimum gearbox length, the coefficients Xba1 and Xba2
must be as large as possible to minimize the center distances of stage 1 and stage 2
(Eqs. (4) and (5)).
• The optimal AS1 and AS2 values are also their maximum values. The reason for this
is that in order to have the shortest possible gearbox length, the AS1 and AS2 values
must be as large as possible to minimize the center distance of the gear stage 1 and
stage 2 (Eqs. (4) and (5)).
930 T. H. Danh et al.

Fig. 3. Influence of main design parameters on gearbox length

Fig. 4. Main effects plot for S/N ratios

• The optimal values of the gear ratio of stage 1 (u1 ) have a first-order relationship
with ugb (Fig. 4). Also, the following regression equation (with R2 = 0.9967) is for
determining the optimum values of u1 :

u1 = 0.2559 · ugb + 1.0453 (12)

After obtaining u1 , the optimum values of u2 is calculated by u2 = ugb /u1 .

 Effect of Main Design Factors on Two-Stage Helical Gearbox Length 931

Table 5. Optimum values of main design parameters

No ugb
5 10 15 20 25 30
u1 2.2 3.72 4.78 6.28 7.58 8.56
Xba1 0.33 0.33 0.33 0.33 0.33 0.33
Xba2 0.4 0.4 0.4 0.4 0.4 0.4
AS1 420 420 420 420 420 420
AS2 420 420 420 420 420 420

Fig. 5. Optimum gear ratio of stage 1 versus gearbox ratio

5 Conclusions
The findings of a study on optimizing a two-stage helical gearbox to obtain the smallest
gearbox length are introduced in this paper. The gear ratio of the first stage, the coefficient
of wheel face width of stages 1 and 2, and the allowable contact stress of stages 1 and
2 were investigated in this study. In addition, a simulation experiment with a Taguchi
L25 type of design was carried out to solve the optimization problem. The effect of
main design factors on gearbox length was also investigated. Furthermore, the following
optimum values of the main design factors were proposed, as well as a regression model
for calculating the optimum values of u1: Xba1 = 0.33; Xba2 = 0.4; AS1 = 420 (MPa);
AS2 = 420 (MPa); u1 is calculated by Eq. (12).

Acknowledgment. This work was supported by Thai Nguyen University of Technology.

932 T. H. Danh et al.

References
1. Bozca, M.: Torsional vibration model based optimization of gearbox geometric design param-
eters to reduce rattle noise in an automotive transmission. Mech. Mach. Theory 45(11),
1583–1598 (2010)
2. Eckert, J.J., et al.: Multi-speed gearbox design and shifting control optimization to minimize
fuel consumption, exhaust emissions and drivetrain mechanical losses. Mech. Mach. Theory
169, 104644 (2022)
3. Vu, N.-P., et al.: The influence of main design parameters on the overall cost of a gearbox.
Appl. Sci. 10(7), 2365 (2020)
4. Sedak, M., Rosić, B.: Multi-objective optimization of planetary gearbox with adaptive hybrid
particle swarm differential evolution algorithm. Appl. Sci. 11(3), 1107 (2021)
5. Korka, Z.I., Cojocaru, V., Miclosina, C.-O.: Modal-based design optimization of a gearbox
housing. Romanian J. Acoust. Vib. 16(1), 58–65 (2019)
6. Tuan, N.K., et al.: Determining optimal gear ratios of a two-stage helical reducer for getting
minimal acreage of cross section. In: MATEC Web of Conferences. 2018. EDP Sciences
7. Pi, V.N., Tuan, N.K., Hung, L.X., Tung, L.A.: A new study on determination of optimum
gear ratios of a two-stage helical gearbox. In: Awang, M., Emamian, S.S., Yusof, F. (eds.)
Advances in Material Sciences and Engineering. LNME, pp. 107–115. Springer, Singapore
(2020). https://doi.org/10.1007/978-981-13-8297-0_14
8. Pi, V.N., Thao, T.T.P., Tuan, D.A.: Optimum determination of partial transmission ratios of
mechanical driven systems using a chain drive and two-step helical gearbox. J. Environ. Sci.
Eng. B 6, 80 (2017)
9. Hung, L.X., et al.: Calculation of optimum gear ratios of mechanical driven systems using
two-stage helical gearbox with first stage double gear sets and chain drive. In: International
Conference on Engineering Research and Applications. Springer (2019)
10. Pi, V.N., et al.: Determination of optimum gear ratios of a three stage bevel helical gearbox.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2019)
11. Thao, T.T.P., et al.: Determining optimum gear ratios of mechanical driven systems using
three stage bevel helical gearbox and chain drive. In: International Conference on Engineering
Research and Applications. Springer (2019)
12. Tuan, N.K., et al.: Optimum calculation of partial transmission ratios of mechanical driven
systems using a V-belt and a three-step bevel helical gearbox. In: International Conference
on Engineering Research and Applications. Springer (2018)
13. Tung, L.A., et al.: A study on determination of optimum gear ratios of a two-stage worm
gearbox. In: International Conference on Engineering Research and Applications. 2019.
Springer
14. Ky, L.H., et al.: Calculation of optimum gear ratios of mechanical driven systems using worm-
helical gearbox and chain drive. In: Sattler, K.-U., Nguyen, D.C., Vu, N.P., Tien Long, B.,
Puta, H. (eds.) ICERA 2019. LNNS, vol. 104, pp. 66–75. Springer, Cham (2020). https://doi.
org/10.1007/978-3-030-37497-6_7
15. Pi, V.N., Tuan, N.K., Hung, L.X., Dung, N.T.Q., Hien, B.T.: Determining optimum gear
ratios of a worm—helical gearbox for minimum acreage of the cross section. In: Awang,
M., Emamian, S.S., Yusof, F. (eds.) Advances in Material Sciences and Engineering. LNME,
pp. 89–96. Springer, Singapore (2020). https://doi.org/10.1007/978-981-13-8297-0_12
16. Le Hong Ky, T.T.H., et al.: Calculation of Optimum Gear Ratios of Mechanical Driven Systems
Using Worm-Helical Gearbox and Chain Drive (2020)
17. Pi, V.N., et al.: A study on determination of optimum gear ratios of a worm-helical gearbox.
In: IOP Conference Series: Materials Science and Engineering. IOP Publishing (2019)
18. Chat, T., Van Uyen, L.: Design and calculation of Mechanical Transmissions Systems, vol.
1. Educational Republishing House, Hanoi (2007)
 The Experimental Research on Mechanical
Properties and Microstructural Characteristics
of Narrow Gap Welding Carbon Steel

Minh Hung Ha1,2 , Trong Binh Ngo1 , and Duong Vu3(B)

1 National Research Institute of Mechanical Engineering, Hanoi, Vietnam
2 Institute for Development of Advanced Technical and Technology, Hanoi, Vietnam
3 Duy Tan University, Da Nang, Vietnam

duongvuaustralia@gmail.com

Abstract. This paper introduces the results of experimental research on process

Gas Metal Arc Welding SS 400 carbon steel thick plates 50 mm with narrow
gaps and small chamfer angles using welding equipment PLEXTEC 500X of
LINCOLN corporation (USA) in an industrial production scale in Vietnam. The
quality of SS 400 carbon steel welded plates was evaluated at three levels as
25%, 50% and 75% of the weld height in combination with the result of measur-
ing microstructural characteristics in different specific zones: base metal; head-
affected zone (HAZ), the boundary between the weld metal and the HAZ such as
the weld center zone.

Keywords: Narrow gap · Small chamfer angle · Gas Metal Arc Welding
(GMAW)

1 Introduction
The butt welding to join the thick plates with a small gap and a chamfer angle using
GMAW process is one of the advanced technologies in the world and Vietnam also. It
helps to improve the productivity of welding process and saves the materials, provided
the stable mechanical properties of the connection.
In recent publications, the research team from the National Key Lab of the surface
treatment and Welding (Research Institute of Mechanical Engineering) published some
results such as the innovative design of the automatic feeding of the welding wire and
the possibility of the integration with the available power source D 500 [1], the charac-
teristics of the MAG welding joint of the carbon steel with the narrow gaps and small
chamfer angles [2], some microstructural characteristics of the welding carbon steel
material using the narrow gaps and small chamfer angles [3], the microhardness in some
specific structural sub-zones in welding carbon steel joint using the narrow gaps and
small chamfer angles [4]. The authors in [5] published the results of the study of the
microstructure and mechanical properties of ultra-narrow gap weld from high-strength
steel. But they used the mixture of 80% Argon + 20% CO2 for shielding gas. It is noted

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 933–941, 2023.
https://doi.org/10.1007/978-3-031-22200-9_99
934 M. H. Ha et al.

that the GMAW joint demonstrated better impact energy than the ultra-narrow gap laser
weld joint. Similarly, in [6], the shielding gas mainly consists of Ar with a small addi-
tive of CO2 for welding alloy structural steel. The quality of weld joint is evaluated by
using the metallography and the lateral bending test. The experiment with the flux-cored
narrow gap welding of low carbon steel showed the advantage over the hard facing due
to fewer spatters [7]. It is suggested the new narrow-gap welding- the laser-arc hybrid
technique [8] for mild steel, but it caused the complex equipment and the production
cost.
That is the reason the main target of this paper focuses on a more general overview
of the characteristics of the GMAW joint based on the study of some particular samples
(increased thickness up to 50 mm) to propose the set of criteria for the evaluation of the
quality of the welding including the mechanical properties and metallography.

2 Methodology
THE experimental samples are carbon steel plates with the chemical composition:
0,18 ÷ 0,2%C; 0,1 ÷ 0,2% Si; 0,06% Mn; 0,03% P and 0,04% S, the remain is Fe;
the strength stress: 373 ÷ 481 MPa; the yield stress: 206 ÷ 245 MPa, the comprehensive
elongation: 24 ÷ 27% and the microhardness: 206 ÷ 245 MPa after heat treatment
(equivalent to the brand SS 400) [3, 4]. The nominal thickness of the welding pieces is
50 mm, the width is 100 mm and the length is 300 mm. The chamfer angle chosen in
the range of 5,0 ÷ 15 ° (cover both sides of connection) is illustrated in Fig. 1.

Fig. 1. View on the surface of the sample in the experiment for welding carbon steel plates with
narrow gaps and small chamfer angles.

The welding equipment GMAW for this experiment is the system PLEXTEC 500
x from LINCOLN (USA) on the industrial scale (Fig. 2). The shielding gas was 100%
CO2 . The sub-zones are nominated according to the actual height of the welding joint,
from which the tensile test specimens are cut (see Fig. 3a). The specific points for
the investigation of the microstructural metallography are given in Fig. 3b. The optical
microscope Axiovert 25 integrated the digital picture analysis (see Fig. 3). In order to
eliminate the change in the structure due to the heat impact, the wire cutting method is
used to split layers in the horizontal direction of welding and shifting according to its
height.
 The Experimental Research on Mechanical Properties 935

Fig. 2. System PLEXTEC 500x from LINCOL (USA) in the experiment

Fig. 3. Side view on layers according to the height of welding joint to prepare the tensile test (a);
nomination of the specific sub-zone in welding joint for the picture in study microstructure (b);
optical microscope Axiovert 25 in experiment (c)

In Fig. 3, a, the picture taken from the welding samples is cut off by the wire
(Electrical Discharge Machining—EDM) to split 3 layers counting from bottom up
nominated accordingly: (1)—bottom part of welding joint; (2)—middle part of welding
joint (3)—top part of welding joint. These parts are prepared to obtain the tensile samples
for measuring the tensile strength according to about 25%; 50% and 75% in direction
of the height of welding joint. Therefore, it is allowed to calculate the average value of
the tensile strength in joint according to the welding mode set by the experiment design
in the investigated ones.
In Fig. 3b, the mapping scheme for all 3 layers of the welding metal, from which
it might investigate and analyze the microstructure metallography (the welding sample
had the thickness = 50 mm, the root gap b = 10 mm, the chamfer angle = 15 O )
including positions (1), (2) and (3) chosen in the central part of weld metal according
to 3 mentioned layers above. Positions (4), (5) and (6) chosen in the boundary zone
between weld metal and heat affected zone on the right side (HAZp). Positions (7), (8)
and (9) in the boundary zone between the welding metal and heat affected zone on the
left side (HAZt). All remain positions (10), (11), (12) are chosen in HAZ and nearby
HAZ from the both sides of welding metal with the relative distance so that it is expected
to recognize the change in the structure of welding material.
936 M. H. Ha et al.

3 The Result of Experiment and Discussion

In Table 1, the result of testing based on the value of the tensile strength from 09 optional
experiments with 3 layers and 3 sub-zones according to the condition of the orthogonal
plan N 27, changing welding modes and chamfer angle (see Fig. 3a). From Table 1, the
analysis is shown:

– When the chamfer angle is small (α = 5o), the welding speed vh = 6 ÷ 8 m/h
and the welding current Ih = 160 ÷ 220 A, the average tensile strength (σk ) tends
to proportional elevation, the welding current in range of Ih = 160 ÷ 190 A, the
tensile strength changes from σk = 469,526 MPa (sample number 02, code 010) to
σk = 491,623 MPa (sample number 04, code 100), then σk ightly decreases to σk =
475,82 MPa (sample number 08, code 210) in range of Ih = 190 ÷ 220 A. Should the
welding speed vh slow down, the σk reaches the bigger value in comparison with vh
chosen at the higher level.
– When the chamfer angle is average (α = 10o): there are 3 variable modes according to
Ih = 160; 190; 220 A and vh = 6; 8; 10 m/h. The average tensile strength of welding
joint σk also tends to change similarly the samples in case of α = 5o above, but their
value slightly increases: sample number 11 (code 011) with Ih = 160 A and vh =
8 m/h, σk = 1,633 MPa; sample number 15 (code 121) with Ih = 190 A and vh =
10 m/h, σk = 83,703 MPa, the remain sample number 16 (code 201) with Ih = 220
A and vh = 6 m/h, σk = 528,413MPa.
– In the case of the maximum chamfer angle (α = 15°), it is shown a similar relation,
namely the average tensile strength of welding joint changes with a proportional
increase with the increase of Ih and inversely decrease with the increasing of vh:
sample number 21 (code 022) with Ih = 160 A and vh = 10 m/h, σk = 515.703 MPa;
sample number 23 (code 112) with Ih = 190 A and vh = 8 m/h, σk = 28,403 MPa,
the remain sample number 25 (code 202) with Ih = 220 A and vh = 6 m/h, σk =
540,066 MPa.

The results of analysis and microstructure of welding metal in the specific structural
sub-zones shown in Figs. 4 ÷ 5.
The analysis of the results of the experiment from Figs. 4, 5 shown:
(1) From Fig. 4a ÷ d, the microstructure of welding metal obtaining in the experiment
sample number 23: in the structural zone right middle of the bottom part (position 1),
the structure of welding metal is wellfined grain and homogeneous, providing the high
hardness (Fig. 4a). Moving to position 4 in the boundary of welding metal on the right
side (B.G.Hp) counting from the left side to the right side, the structure of the material
is distinguished by the structural sub-zone with the size of fined grain similar to the
structure in the central part of welding metal (Fig. 4b). The next position number 7 is
specified by the bigger size of the grain in the structural zone approaching the boundary
zone B.G.Ht (Fig. 4c). In position number 12 (Fig. 4d) there is the most specific structural
zone under the heat impact on the right side (HAZt), namely the size of the grain in the
microstructure of welding metal is more coarse in comparison with the structural zones
in positions 1 and 4. It happened due to the heat during welding process which is not only
 The Experimental Research on Mechanical Properties 937

Table 1. Result of experiment with narrow gap and the value of tensile strength

No. Code Investigated Tensile strength in transverse Average Regression

√
parameter direction of welding joint, y1 = σk , regression S =2 S2
MPa square S2
Ih, vh, α,o (1) (2) (3) Trung
A m/h bình
02 010 160 8 5 468,25 469,51 470,82 469,526 1,10095 1,04926
04 100 190 6 5 491,62 493,22 490,03 491,623 1,69602 1,30231
08 210 220 8 5 477,12 475,81 474,53 475,820 1,11806 1,05738
11 011 160 8 10 492,75 491,60 490,55 491,633 0,80722 0,89845
15 121 190 10 10 482,55 484,86 483,70 483,703 0,88035 0,93826
16 201 220 6 10 528,42 527,94 528,88 528,413 0,14729 0,38378
21 022 160 10 15 516,75 514,66 515,70 515,703 0,72803 0,85324
23 112 190 8 15 527,38 528,40 529,43 528,403 0,70042 0,83691
25 202 220 6 15 540,00 539,05 541,15 540,066 0,73722 0,85861

transferred into the air environment, but also significant direct spreading through the steel
plates, in result the structure from central part of welding metal to the boundary zones
B.G.Hp and B.G.Ht, close to heat affected zones HAZp and HAZt, finally the parent
metal (in both side of welding joint) caused the development of grain size. Especially in
the zone of the parent metal, the microstructure of metal is similar to the initial tempering
carbon steel used in the experiment. The result of experiment in sample number 23 shows
the concordance with the basic theory for butt weld of steel plates.
(2) The microstructure of welding metal obtained in the sample number 23 (Fig. 4, e
÷ k) shows the following: in position number 3, the microstructure in the central zone
from the investigated layer (3) in the top part (Fig. 4e) is similar to position number 1
in the bottom part (1), the size of the grain in phases is relatively fine and homogeneous
due to not strong heat affect during welding process left the temporary cooling time,
providing the high hardness of welding metal. When moving to the position number 6
in zone B.G.Hp (Fig. 4f), the microstructure of welding metal is similar to position 4 in
the bottom part (2) mentioned above. Moving to position number 9 in the boundary zone
on the left side (B.G.Ht - Fig. 4g), the microstructure of the welding metal is similar
to position number 6 where the two sub-zones are distinguished with a strong phase
picture. In position number 10 (HAZp- Fig. 4h), the microstructure is distinguishable
when the phase grain size reaches a significantly high value in comparison with ones
in the central part and the boundary zones. Position number 13 (Fig. 4k) is far enough
from zone B.G.Hp, the microstructure here is similar to the initial tempered carbon steel
pieces used in the experiment.
(3) Figure 5a ÷ d, there are the microstructures of the welding metal taken in sample
number 21, and they show: in the right middle bottom part (position number 1), the
938 M. H. Ha et al.

a) Position 1, x200 b) Position 4, x200 c) Position 7,x200

d) Position 12, x200 e) Position 3, x200 f) Position 6, x200

g) Position 9, x200 h) Position 10, x200 k) Position 13, x200

Fig. 4. Microstructure of welding metal of steel plates SS 400 with the narrow gap (b = 10 mm),
chamfer angle = 15° in different positions on the sample number 23.

microstructure of the welding metal is specified by the coarse grain size and elongated
form (Fig. 5a), what is quite different from the sample number 23 mentioned above
(Fig. 4a). This happened probably due to the overheating caused by continuous welding
process without the cooling. In position number 4 in the boundary zone on the right
side (B.G.Hp), the microstructure of the welding metal is distinguished by the sub-zone
where the phase size is comprehensive on the left side and the phase structure elongated,
getting more coarse closing to sub-zone B.G.Hp (Fig. 5b) in comparison with the specific
structure in position number 1. The next position number 7 (Fig. 5c), the microstructure
of the welding metal in sub-zone B.G.Ht on the left side is similar to the position number
4 in the sub-zone B.G.Hp on the right side in the position number 12 (Fig. 5d) where
 The Experimental Research on Mechanical Properties 939

a) Position 1, x200 b) Position 4, x200 c) Position 7, x200

d) Position 12, x200 e) Position 2, x200 f) Position 5, x200

g) Position 8, x200 h) Position 11, x200 k) Position 13, x200

Fig. 5. Microstructure of welding metal of steel plates SS400 with narrow gap: (b = 10 mm),
chamfer angle = 15° in sample number 21.

there is a clear heat affect (HAZp), the phase size in the microstructure of the welding
metal is comprehensive and bigger in comparison with the positions 1 and 4. This can
be happened under the heat transfer from the molten pool to the parent metal on both
sides and its emission into the air environment as for the sample number 23 mentioned
above. It caused the ascending grain size from central part through the boundary zone,
heat affected zone close to the parent metal. The same character is repeated when the
research conducting in the opposite direction according B.G.Ht; HAZt.
940 M. H. Ha et al.

(4) The picture of microstructure of the welding metal taken in Fig. 5e ÷ k with the
sample number 21 shows: in the central part of welding metal (position number 2), the
microstructure is specified by not too coarse and elongated form (Fig. 5e). This happened
under the quick cooling due to the wind flow during the operation and discontinuous
welding from the bottom part (1) to the middle part (2). In position number 5 in sub-
zone B.G.Hp (Fig. 5f), the microstructure of the welding metal is distinguished by the
two different sub-zones on the left, right side,the grain size is big and elongated enough
closing to the boundary sub-zone B.G.Hp. In position number. 8 on the left side (Fig. 5g),
the microstructure of the welding metal in the boundary on the left side (B.G.Ht) is similar
to one in the position number 5 on the right side (B.G.Hp), but the difference between
the two these sub-zones is light distinguishable. Finally, in the position number 13*
(Fig. 5k), the structural zone of the parent metal is far from heat affected zone HAZp,
that is the reason it is similar to the structure of tempered carbon steel. The difference in
form and size of phases can be explained similarly to the specific positions investigated
in the sample number 21,23.

4 Conclusion
(1) In general, the tensile strength in the transverse direction of GMAW narrow gaps
and small chamfer angles conducting from different samples with changing welding
mode, satisfies the required nominal value σk = 469,526 ÷ 540,066 MPa, depending
on the initial conditions such as the welding current, the welding speed, the chamfer
angle.
(2) The quality of the welding joint is qualified based on the combination of the mechan-
ical properties (the microhardness, the tensile strength) and the metallography.
Based on the analysis of the character of the microstructure in some specific
positions, the quality of the welding joint is well provided, satisfying the main
requirements for thick plate steel constructions in production activities in Vietnam.
(3) To forecast the changing relation in the average tensile strength of the welding joint,
depending on the different orthogonal experiment design, the research team intend
to establish the mathematical model for the relation between 3 main parameters (Ih,
vh, α), the target function is (σk ), the result of which will be published in the next
issue.

References
1. Binh, N.T. et al.: Study on the narrow gap welding thick steel plates, using GMAW for joint
without chamfer. In: Report of scientific research project granted by the Ministry of Industry
and Commerce, code 105.14 RD/HD-KHCN, Insitute of Mechanical Research, Hanoi (2019)
2. Trong, B.N., Ha, M.H., Duy, T.D., Van, D.N.: The characteristics of butt weld with chamfered
gap using MAG method. Vietnam J. Mechnical Eng. 12, 15–22 (2020). http://cokhivietnam.vn,
ISSN 2615-9910/E
3. Ha, M.H., Trong, B.N., Duy, T.D., Van, D.N.: Study on microsdtructural properties of butt
weld material for thick plate carbon steel with narrow openings and small chamfer angle. J.
Sci. Technol. 58(1), 65–60 (2022); Hanoi University of Industry, P-ISSN 1859-3585/E-ISSN
2615- 9619)
 The Experimental Research on Mechanical Properties 941

4. Vu, D., Ha, M.H., Ngo, B.T., Truong, T.T., Van, D.N.: A study on mechanical properties of
MAG Welding Joints Of Steel Plates With Large Wall Thickness And Small Chamfer Angle.
In: 5th International Conference on Coltril, Robotics and Informatics (ICCRI 2022), Danang
City, Vietnam, April 2–4, 2022, iccri@iap.org,
5. Gu, V., Li, L., Dong, S., Crowther, D., Thompson, A.: Comparison of microstructure and
mechanical properties of ultra-narrow gap laser and ga-metal-arc welded S960 high strength
steel. Opt Lasers Eng 91:1–15. https://doi.org/10.1016/j.optlaseng.2016.11.011
6. Chen, B., Zhu, L., Xiao, W., Zhang, F.: Analysis on the microstructure and mechanical prop-
erties of welding joint of low alloy structural steel plate by narrow gap MAG. In: International
Conference on Mechatronics, Electronic, Industrial and Control Engineering (MEIC 2015),
pp.84–87
7. Dashuang, L., Ping, W., Long, W., Yucheng, W., Wenbin, H.: Narrow gap space contributes to
chemical metallurgy of self-shielded arc welding. China Weld. 30(3), 12–19 (2021)
8. Zhang, C., Li, G., Gao, M., Zeng, X.Y.: Microstructure and mechanical properties of narrow
gap laser- arc hybrid welded 40 mm thick mild steel. Materilas 10, 106 (2017). https://doi.org/
10.3390/ma10020106
 The Experimental Research on Plasma Welding
of SUS 304 Stainless Steel Thin Plates

Minh Hung Ha1,2 , Hai Tinh Do3 , Duong Vu4(B) , Duy Trung Dao1 , and Thu Quy Le1
1 National Research Institute of Mechanical Engineering, Hanoi, Vietnam
2 Institute for Development of Advanced Technical and Technologym, Hanoi, Vietnam
3 Ministry of Science and Technology, Hanoi, Vietnam
4 Duy Tan University, Danang 550000, Vietnam

duongvuaustralia@gmail.com

Abstract. The article introduces some experimental research results on the appli-
cation of plasma welding technology for SUS 304 stainless steel thin plates, using
welding equipment of LINCOLN CB-MATIC SF 32–33 (USA) with a production
industry scale in local conditions of Vietnam. The quality of the weld is prelimi-
narily assessed through the criteria of its surface morphology and the maximum
deformation of the weld structure in the state after natural environmental cool-
ing. The experimental results help to set the limitation of the domain of in-depth
investigation on the influence of key technological parameters on the mechanical
properties and microscopic structure of the weld material, using an experimental
design method as a scientific platform to determine the welding mode in accor-
dance with the technical requirements of the mechanical product, which needs to
be manufactured.

Keywords: Plasma welding · Stainless steel SUS 304 · Experiment design

1 Introduction
In recent mechanical engineering production, especially in the area of the food processing
industry and the chemical industry, the localization for many types of products made of
the stainless steel is needed. Some of mechanical engineering companies imported the
advanced welding and forming equipment [1] to make big tanks with different sizes, for
example the tank for storage of fresh milk installed in the shuttle transport from the farm
to the processing facilities and packing, the tank for storage of the drinking water, or the
device for the distillation of margarine oil,…[2, 3].
Upon the recommendations of the suppliers, the thickness of stainless steel SUS
304/304L plates better chosen in the range: 1.0 to 6.0 mm [4, 5], and also based on some
international publications [6–8].
The serial welding equipment in industry scale from LINCOLN, brand CB-MATIC
SF 32–33, model 450Plus /Plasma–SP7 Turnmatic [5] is the main tool, providing the
regulation all plasma welding modes with the automatic setting, having the real time
monitor on the welding process, is the well feasible condition for the deeply specialized
research in terms of the optimization the technology of production on the demand from
the market.It also spesify the scientific and practical values.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 942–951, 2023.
https://doi.org/10.1007/978-3-031-22200-9_100
 The Experimental Research on Plasma Welding of SUS 304 943

2 Methodology
– The material for the experiment: steel plate SUS 304 and SUS 304L, the thickness of
2,0 mm and 1,5 mm. The size of welding specimen B x L = 150 mm x 400 mm. The
chemical composition shown in the Table 1.

Table 1. The chemical composition of steel plate SUS 304 [5].

Brand The content of the main element, % wt Weldability, application

304 ≤ 0,08 C; ≤ 1,0 Si; ≤ 2,0 Mn; 18 ÷ 20 Good weldability, chemical and food
Cr; 8 ÷ 12Ni industry

• Joint design: the butt weld in the position G1. In 4 corners, the location is provided
by the clamp and wedge pin for easy release the pieces after the welding (Fig. 1f).

Fig. 1. Panorama view of plasma welding equipment in the experiment LINCOLN CB-MATIC
SF 32–33 (model 450Plus /Plasma –SP7 Turnmatic TT) (a); unit for setting the automatic welding
program and real time monitoring in front of welding seam (b); unit for feeding the generation
plasma gas (c); unit for automatic feeding the wire (d); fixture for automatic welding (e); view of
fixture the pieces in the experiment (f)
944 M. H. Ha et al.

The image of some main parts from welding plasma equipment CB-MATIC SF 32-
33 used in the experiment was shown in Fig. 1, the experiment process was illustrated in
Fig. 2. Whereby: the setting input parameters were selected by the monitoring program
in Table 2. Through the monitoring screen, it can be watched directly from the activated
plasma until the end of the automatic cycle (Fig. 2a, b).

Fig. 2. Shooting scene in the experiment for butt plasma welding of the thin plates steel SUS
304: Setting the automatic welding program (a, b); Some moments during the automatic welding
in front of welding seam (c–e); The connection process behind the welding seam (f)

3 Experiment, Results and Discussion

3.1 The Evaluation the Quality of Welding Joint Using the Criteria of Surface
Morphology (Appearance).
According to some publications on the plasma welding, microplasma welding the
stainless steel [4] ÷ [7] for probing in the butt weld of stainless steel thin plates
SUS 304, having the thickness above mentioned, the preliminary limitation of the
input parameters done as follows: 1) The upper current of plasma pulse: Ih = 80 ÷
100 A; 2) The undercurrent of plasma pulse: Ib = 50 ÷ 60 A; 3) The velocity of
welding tip along side of the connection (welding velocity) vh = 25 ÷ 40 cm/ph;
4)Therateofwirefeedingwiththediameterdb = 1,0 mm is vc.dh = 80 ÷ 90 cm/ph; 5) The
flow rate of gas: Ar + 5%H2 . Based on some recommendations from the supplier of
plasma welding equipment CB-MATIC SF 32–33 [2], the experiments setting upon the
 The Experimental Research on Plasma Welding of SUS 304 945

automatic program and the results of measurement welding size by the metallography
shown in Table 2.

Table 2. The setting of some plasma automatic welding modes in the experiment and weld size
measurement using the metallography

No The parameters of welding technology The size taken from macrostructure of welding
in the experiment metal, mm
Ih , Ib , vh , vc.dh , Gk.p , Diameter Thickness Max. Bead Root
A A cm/ cm/ l/min of wire, of weld Reinforcement, width, width,
min min db pieces, δ hmax bT bD
01 100 60 34 80 2.0 1.0 2.0 2.63948 5.40457 3.50761
02 80 50 40 90 2.0 1.0 2.0 3.27142 4.04399 1.48758
03 90 50 30 80 2.0 1.0 2.0 2.52897 4.53180 1.40253
04 90 50 25 80 2.0 1.0 2.0 2.68782 5.34280 3.50235
05 90 50 30 90 2.0 1.0 2.0 2.90773 5.21014 2.76776
06 90 50 28 90 2.0 1.0 2.0 2.78724 5.54351 3.26626
07 95 50 32 90 2.0 1.0 2.0 2.74218 5.30470 2.67924
08 95 50 30 90 2.0 1.0 2.0 2.84285 5.49962 3.71705
09 100 50 30 90 2.0 1.0 2.0 2.91839 5.45970 3.59594
11 60 30 30 75 1.5 1.0 1.5 2.19211 4.84359 3.90848
12 50 25 25 65 1.5 1.0 1.5 2.33874 3.95090 2.84103
13 70 35 30 75 1.5 1.0 1.5 1.51707 4.57737 3.82412

In Figs. 4, 5 bellow, there are the images of the surface morphology of welding seam
in the state after cooling in the quite open- air conditions. The letter “T” assign for picture
taken from the upper surface, the letter “D” for the under surface (Fig. 3).

3.2 The Image of the Macrostructure with the Result of Measurement Weld Size
in the Transverse Direction of Fusion Line
The result of measurement of weld size in welding thin plates by stainless steel SUS
304 for different modes using the metallography shown in Fig. 6 ÷ Fig. 7.
The analysis of the experiment result in Tab 3 and images from Fig. 4 ÷ Fig. 7 find
out:

– Sample 01: alongside the upper and under surface of welding seam, the welding metal
is obtaining very evenly and the roughness is well smooth (Fig. 3a, b). It does not
detect the infusion in whole length of welding seam, providing the high quality. Some
correlations about the form of welding joint: the maximum reinforcement of welding
metal hmax = 2.63948 mm, the bead width bT = 5.40457 mm, the root width bD =
946 M. H. Ha et al.

a) Sample 01 T b) Sample 01 D

c) Sample 02 T d) Sample 02 D

f) Sample 03 D
e) Sample 03 T

g) Sample 04 T h) Sample 04 D

k) Sample 05 T l)Sample 05 D
→
Welding direction alongside of fusion line→

Fig. 3. The image from the surface morphology of welding joint depending on the different
welding modes (steel SUS 304, thickness of 2.0 mm) [8]

a) Sample 06 T b) Sample 06 D

c) Sample 07 T d) Sample 07 D

e) Sample 08 T f) Sample 08 D

g) Sample 09 T h) Sample 09 D
Welding direction alongside the fusion line →
Fig. 4. The image from the surface morphology depending on the different welding modes (steel
SUS 304, the thickness of 2.0 mm) [8]

3.50761 mm, using its macrostructure in transverse direction of fusion line (Fig. 6a).
The rate between the bead width and the root width bT /bD = 1.54081. The rate between
 The Experimental Research on Plasma Welding of SUS 304 947

a) Sample 11 T b) Sample 11 D

c) Sample 12 T d)Sample 12 D

e) Sample 13 T f) Sample 13 D
Welding direction alongside the fusion line →
Fig. 5. The image from surface morphology depending on the different welding modes (steel
SUS 304, the thickness of 1.5 mm)

the bead width and the root width to the maximum reinforce accordingly: bT /hmax =
2.04758 và bD /hmax = 1.3289;
– Sample 02: In the upper and under surface alongside the fusion line, it is easy to note
some areas where the melting additive material is solidified in form of drip adjoining
evenly in order to make it difficult to recognize the infusion defect (Fig. 3c). In the
under surface of welding metal also appears the drip of solidified welding additive
but their size is well smaller in comparison with the upper one. The correlation about
form of welding metal: hmax = 3.27142 mm, bT = 4.04399 mm, bD = 1.48758 mm
(Fig. 6b); bT /bD = 2.7185; bT /hmax = 1.23615 and bD /hmax = 0.45471. The welding
joint in sample 02 failed;
– Sample 03: The morphology of the upper and under surface alongside the fusion
line is well even, the roughness is fine (Fig. 3e, f). The defect of infusion is not
detected in whole and the welding seam, the morphology of the surface is fine. Some
correlations about the form of welding joint: hmax = 2.52897 mm; bT = 4.5318 mm,
bD = 1.40253 mm (Fig. 6c); bT /bD = 3.23116; bT /hmax = 1.791954 and bD /hmax =
0.55458. Sample 03 passed;
– Sample 04: the upper and under surface of welding metal shows the event distribution
of welding metal with the fine roughness (Fig. 3g, h).The defect of infusion is not
detected in whole and the welding seam, the morphology is fine. The approximate
size of HAZ decreases. The correlations about the form of welding joint: hmax =
2.68782 mm , bT = 5.3428 mm; bD = 3.50235 mm (Fig. 6d); bT /bD = 1.52549;
bT /hmax = 1.98778 and bD /hmax = 1.30304. At this sample the quality of the welding
is good;
– Sample 05: the morphology of the upper and under surface of welding seam shows
the good state of the solidified welding metal, the roughness is fine and there is not the
infusion in whole and the length of welding seam, the morphology of the surface is the
finest (Fig. 3k, l). The size of the welding joint in the sample 05 is similar to the sample
04 above mentioned. All HAZ have the notably decreased size. All relations about
the form of welding joint: hmax = 2.90773 mm, bT = 5.21014 mm, bD = 2.76776 mm
948 M. H. Ha et al.

a) Sample 01 b) Sample 02

c) Sample 03 d) Sample 04

e) Sample 05 f) Sample 06

g) Sample 07 h) Sample 08

k) Sample 09

Fig. 6. The image of macrostructure of welding joint stainless steel SUS 304 (a ÷ k). The
thickness of welding piece δ = 2.0 mm)
 The Experimental Research on Plasma Welding of SUS 304 949

a) Sample 11
c) Sample 12

b) Sample 13
Fig. 7. The image of macrostructure of welding metal stainless steel SUS 304 (a ÷ c) (a thickness
of welding piece δ = 1.5 mm)

(Fig. 6e); bT /bD = 1.88243; bT /hmax = 1.79182 and bD /hmax = 0.95186. The welding
joint at the sample 05 meets the quality requirement;
– Sample 06: the metal solidified layer in the upper and under surface of this sample
shows the good appearance, the roughness is fine, no detectable defect type of infusion
in whole and the length of welding seam, the morphology is the finest (Fig. 4a, b).
The size of HAZ is not big. All correlations about the form of the welding joint are:
hmax = 2.78724 mm, bT = 5.54351 mm, bD = 3.26626 mm (Fig. 6f); bT /bD = 1.6972;
bT /hmax = 1.98888 and bD /hmax = 1.17186. The quality of the welding joint at the
sample 06 is good.
– Sample 07: the morphology of the upper and under surface of welding seam shows
the good state of the solidified welding metal, the roughness is fine and there is not
the infusion in whole and the length of welding seam, the morphology of the surface
is the finest (Fig. 4c, d).All correlations about the form of the welding joint are:
hmax = 2.74218 mm, bT = 5.3047 mm, bD = 2.67924 mm (Fig. 6g); bT /bD = 1.97992;
bT /hmax = 1.93448 and bD /hmax = 0.97704. The quality of the welding joint at the
sample 07 is good, similar sample 06;
– Sample 08: the morphology of the upper and under surface of welding seam shows
the good state of the solidified welding metal, the roughness is fine and there is not the
infusion in whole and the length of welding seam, the morphology of the surface is the
finest (Fig. 4e, f). The size of HAZ is small, similar samples 06 and 07. All relations
about the form of the welding joint are: hmax = 2.84285 mm, bT = 5.49962 mm, bD =
3.71705 mm (Fig. 6h); bT /bD = 1.47956; bT /hmax = 1.93454 and bD /hmax = 1.3075.
The quality of welding joinr at sample 08 is good;
– Sample 09: the morphology of the upper and under surface of welding seam shows
the good state of the solidified welding metal, the roughness is fine and there is not
950 M. H. Ha et al.

the infusion in whole and the length of welding seam, the morphology of the surface
is the finest (Fig. 4g, h). An approximate size of HAZ is similar at the sample 08.
All correlations about the form of the welding joint are: hmax = 2.91839 mm, bT =
5.4597 mm, bD = 3.59594 mm (Fig. 6k); bT /bD = 1.51829; bT /hmax = 1.87079 and
bD /hmax = 1.23216. The quality of the welding joinr at the sample 09 is good;
– Sample 11: The upper and under surface at the sample 11 which are distinguished by
the welding metal solidified layers are adjoining, making it rough alongside the weld-
ing seam. That is why the morphology of the welding metal surface is not fine, leading
to decrease the tensile strength of the welding joint (Fig. 5a, b). All correlations about
the form of the welding joint, using the metallography from the macrostructure, are as
follows: hmax = 2.19211 mm, bT = 4.84359 mm, bD = 3.90848 mm (Fig. 7a); bT /bD =
1.84359; bT /hmax = 2.20955 and bD /hmax = 1.78297.The quality of the welding joint
at sample 11 is considering an average level in dependence of context;
– Sample 12: In the upper and under surface of sample 12 there is the same image as
for sample 11: metal solidified layers are adjoining, making the notable drip rough
surface alongside the welding seam. Although the morphology of welding metal
surface is improving, it can decrease the tensile strength of the welding joint (Fig. 5c,
d). All correlations about the form of the welding joint: hmax = 2.33874 mm, bT =
3.9509 mm, bD = 2.84103 mm (Fig. 7b); bT /bD = 1.39065; bT /hmax = 1.68932 and
bD /hmax = 1.21476. The quality of the welding joint at sample 12 is qualified;
– Sample 13: The upper and under surface in the sample 13 is distinguished by solidified
welding metal, the roughness is fine and there is not the infusion in whole and the length
of welding seam, the morphology of the surface is the finest. There is not the infusion
defect in whole and the length of the welding seam (Fig. 5e, f). All correlations
about the form of the welding joint: hmax = 1.51707 mm, bT = 4.57737 mm, bD =
3.82412 mm (Fig. 7c); bT /bD = 1.19697; bT /hmax = 3.01724 and bD /hmax = 2.5272.
The quality of welding joint at sample 13 is qualified.

Based on the above analysis, it can outline the main parameters on the form of plasma
welding stainless steel plate SUS 304 as follows:

(1) For all samples with the thickness of 2.0 mm, the form of the welding metal in
transverse direction of welding seam, is specified such as: hmax = 2.52897 ÷
3.27142 mm; bT = 4.04399 ÷ 5.54351 mm; bD = 1.40253 ÷ 3.71705 mm; bT /bD =
1.47956 ÷ 3.23116; bT /hmax = 1.23615 ÷ 2.04758 and bD /hmax = 0.45471 ÷
1.3289;
(2) For all samples with the thickness of 1.5 mm, the form of the welding metal shows
the following characters: the maximum reinforce in the limitation hmax = 1.51707
÷ 2.33874 mm; bT = 3.9509 ÷ 4.84359 mm; bD = 2.84103 ÷ 3.90848 mm; bT /bD =
1.19697 ÷ 1.84359; bT /hmax = 1.68932 ÷ 3.01724 and bD /hmax = 1.21476 ÷
2.5272.

4 Conclusion
(1) The welding equipment LINCOLN CB-MATIC SF
32–33 (model 450Plus/Plasma—SP7 Turnmatic) is adopted to set the automatic
 The Experimental Research on Plasma Welding of SUS 304 951

welding program according to different plasma welding modes and the inline mon-
itoring in front of the welding seam, which is applied in the industrial scale in
Vietnam. The research team successfully finished some experiments for the butt
weld plasma stainless steel plate SUS 304, having the thickness of 1.5 mm and và
2 mm with the reasonable result;
(2) Through the morphology criteria of the surface and the form of the welding metal
made in the transverse direction of welding seam, the quality of the welding joint
meets the standard and keeps the state stable along side the fusion line. The form of
plasma welding stainless steel SUS 304 is specified by the basic sizes and the cor-
relation, including: maximum reinforce (hmax ); ; bead width (bT ); root width (bD );
rates: bT /bD , hbmax
T
andbD /hmax , measuring in the optical microscope integrated the
digital image software with more higher resolution in comparison to the mechanical
tool;
(3) The research team will publish the mechanical properties and microstructure from
plasma welding of the stainless steel’s plate in the next issue./.

References
1. Zhang, Y.M., Yang, Y.-P., Shang, W., Na, S.-J.: Advanced welding manufacturing: a brief
analysis and review of challenges and solutions. J. Manuf. Sci. Eng. 142(11), 110816(33pages)
(2020)
2. Dewangan A.K., Patel, A.D., Bhadania, A.G.: Stainless steel for dairy and foodindustry: a
review. J. Mater. Sci. Eng. 4(5) (2015). https://doi.org/10.4172/2169-0022.1000191
3. Brian Rothstein: How is stainless steel used in the food service industr? Mead Metall, Inc.
https://www.meadmetals.com/blog/stainless-steel-food-industry,(2019)
4. Batool, S., Khan, M., Jaffery, S.H.I., Khan, A., Mubashar, A., Ali, L., Khan, N., Anwar, M.N.:
Analysis of weld characteristics of micro-plasma arc welding and tungsten inert gas welding
of thin stainless teel (304L) sheet. J. Mater.: Des. Appl. 230(6), 1–13 (2015). https://doi.org/
10.1177/1464420715592438
5. Lincoln CB Matic SF 32–33/Pilot Pro, Nertamatic 450Plus/Plasma SP7 Turnmatic. www.air
liquidewelding.com
6. Prasad, K.S., Rao, C.S., Rao, D.N.: Prediction of weld pool geometry in pulsed current micro
plasma arc welding of SS304L stainless steel sheets. Int. Trans. J. Eng. Manag. Appl. Sci.
Technol. 2(3), 325–336 (2011)
7. Prasad, K.S., Rao, C.S., Rao, D.N.: A study on weld quality characteristics of pulsed current
micro plasma arc welding of SS304L sheets. Int. Trans. J. Eng. Manag. Appl. Sci. Technol.
2(3), 437–446 (2011)
, ,
8. Tı̃nh, Ð.H., Hùng, H.M., Trung, Ð.D.: Nghiên cúu thu. c nghiê.m hàn plasma giáp mối thép
ij
ij ,
không gi tấm mong SUS 304, Ta.p chí Co khí Viê.t Nam, ISSN 2615–9910, số 6,(2022) (The
Experimental Researching on plasma butt weld of SUS 304 Stainless Steel Thin Plates, Vietnam
Mechanical Engineering Journal, ISSN 2615–9910, No. 6 (2022)
 The Orbit Tracking Adaptive Control
for Omnidirectional Mobile Robot Based
on RBF Neural Network

V. T. T. Linh1 , V. V. Thong1 , R. V. Hoa1 , V. D. Hung1 , T. D. Chuyen1(B) ,

Tran Manh Hung2 , and Pham Ngoc Dien2
1 Faculty of Electrical, University of Economics - Technology for Industries, Hanoi, Vietnam
tdchuyen@uneti.edu.vn
2 College of Technology and Trace, Hanoi, Vietnam

Abstract. This paper presents the design of an adaptive tracking controller based
on the RBF neural network to control the trajectory tracking of a three-wheeled
omnidirectional mobile robot (TWOMR) taking into account the uncertainty non-
linear factor. When the robot moves, it will be affected by some uncertain com-
ponents of the model, unknown noises, the interaction of parameters in the robot
system. The Radial Basis Function (RBF) neural network is used here to esti-
mate the uncertain components and unknown noises of the robotic system. The
adaptive traction controller is proven stable through the Lyapunov benchmark.
The effectiveness of the controller is verified by simulation results on Matlab
software – Simulink.

Keywords: Intelligent control · Mobile robot · TWOMR · RBF neural network ·

Adaptive control

1 Introduction
In the scope of the article, the holonomic self-propelled robot is built based on the Omni
omnidirectional tricycle. With omnidirectional wheel structure capable of 360-degree
rotation, the robot can easily move to any position in the plane (Fig. 1).

Fig. 1. Three-wheeled omnidirectional mobile robot and model using Omni wheels

Unlike traditional wheeled robots (standard wheels), mobile robots using omnidi-
rectional wheels have additional advantages such as the ability to change position and

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 952–967, 2023.
https://doi.org/10.1007/978-3-031-22200-9_101
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 953

orientation flexibly because they have the ability to move forward and rotate simulta-
neously or independently. The wheels are usually arranged along the axis of the robot.
However, in omnidirectional mobile (OMR) robots, the wheels are arranged in a circle
circumscribing the robot to take advantage of the degrees of freedom of the omnidirec-
tional wheels. In the motion control technique of mobile robots, the problem of trajectory
tracking and fast impact is the most necessary requirement.
In the world, mobile robot control has been widely studied and developed in recent
decades and many methods including classical control, modern control have been pro-
posed to be applied to mobile robots. Previously, most of the publications used the
structure of two control loops: the outer kinematic loop uses the Lyapunov function to
synthesize the position tracking controller, and the inner dynamic loop control the speed
tracking. Many dynamic loop control methods have been proposed such as slip control
[1, 2]; backstepping control [3, 4]. When the dynamic equation has an uncertain param-
eter, adaptive control is included in the design [5, 6]. In the document [7–9], the adaptive
controller based on fuzzy logic has been applied to control the nonlinear system with
uncertain structure and presence of external noise in the control process. Here fuzzy
logic is used to approximate the unknown dynamics of the nonlinear system. However,
fuzzy controllers when building control rules often depend a lot on the experience of the
designer.
Based on the above analysis, the paper builds an orbital-like controller for a three-
wheeled omnidirectional mobile robot based on the RBF neural network, in which the
controller is designed when affected by some uncertain components of the model and
unknown perturbation factors.

2 Dynamics for a Three-Wheeled Omnidirectional Mobile Robot

The geometric structure of TWOMR is presented based on the document [10]:

Fig. 2. The geometric structure of TWOMR

Figure 2 depicts the geometric structure of TWOMR consisting of 3 omnidirectional

wheels, equidistant from each other at an angle of 120 degrees; r is the radius of the
wheel, the moving center of the robot is located at C, we can consider this as the center
of the robot; L is the distance from the center of the wheel to the point C; (OXY)
954 V. T. T. Linh et al.

is the global coordinate reference system; (CX0 Y0 ) is the local coordinate reference
system associated with the robot. The robot in the global reference coordinate system
 T  T
is defined by the vector qC = XC YC θC ∈ R3×1 ; vector PC = XC YC ∈ R2×1
which is defined as the position vector of the robot’s center C with origin O; Pwi ∈
R2×1 (i = 1, 2, 3) is the relative position vector of each wheel with respect to the local
coordinate reference system; Dwi ∈ R2×1 is the direction vector of each wheel with the
coordinate reference system (OXY).

2.1 The Kinematic Equation of TWOMR

The rotation matrix R(θC ) converted from the displacement coordinate system associated
with the robot to the global coordinate system is shown as follows:
 
cos(θC ) −sin(θC )
R(θC ) = ∈ R2×2 (1)
sin(θC ) cos(θC )

The position vectors Pw1 , Pw2 , Pw3 ∈ R2×1 of the wheels in the displacement
coordinate system attached to the robot center are expressed as follows:
 
1
Pw1 = L (2)
0
    2π    
2π cos 3 −sin 2π 1
Pw2 = R × Pw1 = 2π 2π
3 ×L
3 sin 3 cos 3 0

√     
− − 2
1 3
1 L −1
√
= √32 ×L = (3)
2 − 1
2
0 2 3
    4π    
4π cos 3 −sin 4π3 1
Pw3 = R × Pw1 = × L
3 sin 4π
3 cos 4π
3 0

√     
−√2 2
1 3
1 L 1
= ×L =− √ (4)
− 3 −1 2 2
0 2 3

The vectors indicating the direction of motion A, B of the wheel are calculated as
follows:
1
Dwi =
R(θC ) × Pwi (5)
L
           
1 π 1 cos π2 −sin π2 1 0 −1 1 0
Dw1 = R × Pw1 = π π × L = × =
L 2 L sin 2 cos 2 0 1 0 0 1
(6)
  π π   
1 π 1 cos 2 −sin 2 L −1
Dw2 = R × Pw2 = π π × √
L 2 L sin 2 cos 2 2 3
    √ 
1 0 −1 −1 1 3
= × √ =− (7)
2 1 0 3 2 1
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 955

     
1 π 1 cos π2 −sin π2 L
Dw3 = R × Pw3 = π  π × − √1
L 2 L sin 2 cos 2 2 3
    √ 
1 0 −1 1 1 3
=− × √ = (8)
2 1 0 3 2 −1

The position and velocity vectors of the wheels in the global coordinate system are
represented by the following equations:

Pi = PC + R(θC ) × Pwi (9)

Vi = ṖC + Ṙ(θC ) × Pwi (10)

⇒ ViT = ṖCT + ṘT (θC ) × Pwi

T
(11)

The angular velocity of the wheels is calculated as follows:

1 T
ωi = V × R(θC ) × Dwi
r i
1 
= ṖCT × R(θC ) × Dwi + Pwi
T
× ṘT (θC ) × R(θC ) × Dwi (12)
r
The kinematic equation of a three-wheeled omnidirectional mobile robot is written
in detail as follows [11]:
⎡ ⎤ ⎡ ⎤⎡ ⎤
ω1 −sin(θ
 C) cos(θ
 C) L ẊC
1
⎣ ω2 ⎦ = ⎣ −sin π/3 − θ C −cos π/3 − θ C L ⎦⎣ ẎC ⎦ (13)
r  
ω3 sin π/3 + θ C −cos π/3 + θ C L θ̇C

The equation is rewritten in vector form:

1 −1
z= H × q̇C (14)
r
 T
Accordingly, z = ω1 ω2 ω3 is the wheel’s angular velocity vector, and the matrix
H −1 ∈ R3×3 is expressed as follows:
⎡ ⎤
−sin(θ
 C) cos(θ
 C) L
H −1 = ⎣ −sin π/3 − θ C −cos π/3 − θ C L ⎦ (15)
 
sin π/3 + θ C −cos π/3 + θ C L

2.2 The Dynamics Equation of TWOMR

The dynamic model of TWOMR uses the Euler Lagrange method for the orderless
system. The dynamic equation [12] has the following form:

M q̈C + C q̇C + ud = u (16)

956 V. T. T. Linh et al.
⎡ ⎤ ⎡ ⎤
m1 0 0 0 −θC 0
where, M = ⎣ 0 m2 0 ⎦; C = a⎣ θC 0 0 ⎦;
0 0 m3 0 0 0
⎡   ⎤
−sin(θC ) −sin π/3 − θ C sin π/3
 + θC
B = 1r ⎣ cos(θC ) −cos π/3 − θ C −cos π/3 + θ C ⎦; m1 = m2 = Mp +
L L L
3Ir 3Ir L 2 3Ir
+ 2r 2 ; m3 = Ip + r2
; a = 2r 2 ; u = Bτ . With Mp is the mass of the robot; Ir is
the moment of inertia of the wheel; Ip is the moment of inertia of the robot; ud is the
 T
disturbance torque, u = u1 u2 u3 is the torque acting on each motor.
Note: The dynamic equation TWOMR in (16) has the following characteristics [12]:
Feature 1: The matrix M is invertible and positive definite. Feature 2: The matrix Ṁ −
2C(qC , q̇C ) has skewed symmetry.

3 The Controller Design

3.1 Model Structure of Neural Network Structure
The RBF neural network is used to approximate f (x), described in Fig. 3 as referenced
from the literature [13].

σ1

x1 w1

w2
x2 σ2 ∑
f(x)
wm

xn

σm
k j

Fig. 3. The neural network structure RBF

The RBF neural network structure consists of 3 layers: The first layer is the input
layer including input variables x1 , x2 , · · · , xn ; The second layer is the hidden layer, the
output of the hidden layer is calculated according to the following formula:

x − ck 2
σj = exp , j = 1, 2, ..., m (17)
b2j

The third layer is the output layer, which is calculated as follows:

f (x) = W T σ + ε (18)
 T
where x is the input state of the network, σ = σj is a Gaussian function, m is the
number of hidden layer neurons, ck is the center of the function, bj is the width of the
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 957

function, ε is the approximate error of the neuron network which will be then studied,
W = [w1 , w2 , ..., wm ]T is the weight vector of the RBF neural network.
 
T
Then we have : f (x) = W σ (19)


With W̃ = W − W , W F ≤ Wmax . From (18) and (19), we have:

 
T
f − f = W T σ + ε − W σ = W̃ T σ + ε (20)

From the expression f(x), the input of the RBF neural network is chosen as:
 
x = eT ėT qdT q̇dT q̈dT (21)

3.2 The Controller Structure

We need to control so that qC → qd ; q̇C → q̇d ; q̈C → q̈d , with component qd , q̇d , q̈d
being the desired values.
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
xC xd ẋC ẋd
In there: qC = ⎣ yC ⎦, qd = ⎣ yd ⎦, q̇C = ⎣ ẏC ⎦, q̇d = ⎣ ẏd ⎦, q̈C =
θC θd θ̇C θ̇d
⎡ ⎤ ⎡ ⎤
ẍC ẍd
⎣ ÿC ⎦, q̈d = ⎣ ÿd ⎦.
θ̈C θ̈d
Then we define the tracking error as follows:
⎡ ⎤
xd − xC
e = qd − qC = ⎣ yd − yC ⎦, e ∈ R3×1 (22)
θd − θC

and we define the tracking error filter function:


s(t) = ė + αe e + αI edt, s ∈ R3×1 (23)

In which: αe = diag(αei ), αI = diag(αIi ), i = 1, 2, 3. Then, with definitions in (22)

and (23), we give the robot model (16) in the following form:
  
M ṡ + Cs = M (ë + αe ė + αI e) + C ė + αe e+α I edt
  
= M (q̈d − q̈C + αe ė + αI e) + C q̇d − q̇C + αe e+α I edt
  
= M (q̈d + αe ė + αI e) + C q̇d + αe e+α I edt − M q̈C − C q̇C
  
= M (q̈d + αe ė + αI e) + C q̇d + αe e+α I edt + ud − u (24)
958 V. T. T. Linh et al.

  
We set : f = M (q̈d + αe ė + αI e) + C q̇d + αe e+α I edt (25)

Rewrite the equation we have:

M ṡ + (K + C)s = f + ud − u + Ks (26)

with, K = diag(Ki ), i = 1, 2, 3.
If the right side of the equation is set to zero, the system on the left side of the
equation will stabilize and converge to zero, then the tracking error will approach zero.
Therefore, we propose the following control law:

u∗ = f + ud + Ks (27)

It can be seen that the Ks part is a PID controller. However, to implement this control
law in the best way, the noise and the model parameter of the robot which are ud , f, have
to be given in advance. In this case, these values are unknown. So we will use the RBF
network to estimate these values. We propose an adaptive tracking control rule using a
neural network based on Eq. (27), then we have (28) as follows:



u = f + ud + Ks (28)



In there: ud is the noise compensation and estimated error, f ∈ R3×1 is the output of
the neural network, this component is also the estimated value of f. Then we can write


the component f represented as follows:


T


f = W σ (x) (29)


In which: W is the estimated value of W. We set the estimated error parameters as

follows:
 

f˜ = f − f ; W̃ = W − W (30)

From Eq. (28), we have an adaptive controller structure diagram based on RBF neural
network for TWOMR as shown in Fig. 4.

3.3 The Proof of Stability

The Lyapunov stability criterion is used to consider the stability of the control system and
to determine how to update the weights for the neural network. The Lyapunov function
is as follows:
 1 1 
−1
V s(t), W̃ = sT Ms + tr W̃ T KW W̃ (31)
2 2
where, KW ∈ Rm×m is a constant matrix, tr(.) 
is a matrix
trace.
For convenience, we put the component V s(t), W̃ = V .
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 959

Ŵ = K Wσ sT

s
uˆd = ( ε d + udb )
s

fˆ uˆd

qd , qd , qd e s (t ) = e + α e e + α I ∫ edt s Ks u
- K + TWOMR

qC , qC , qC

Fig. 4. The recommended controller structure for TWOMR

Derivative of both sides of Eq. (31) with respect to time, we have:

1 −1 ˙
V̇ = sT M ṡ + sT Ṁ s + tr(W̃ T KW W̃ ) (32)
2
˙
˙ = −W


With W being the ideal constant W̃ , we rewrite Eq. (32) as follows:

1 −1 ˙


V̇ = sT M ṡ + sT Ṁ s − tr(W̃ T KW W) (33)
2
Substituting (26) into (33), we have:

1 −1 ˙


V̇ = sT (−Cs + f + ud − u) + sT Ṁ s − tr W̃ T KW W
2
1 T −1 ˙


= s (Ṁ − 2C)s + sT (f + ud − u) − tr(W̃ T KW W) (34)

2
Using the skewed symmetry matrix property of the robot, Eq. (34) becomes:
−1 ˙


V̇ = sT (f + ud − u) − tr(W̃ T KW W) (35)

Substituting the proposed control rule from (28) to (35) we have:

 
−1 ˙
 


V̇ = sT f + ud − f − ud − Ks − tr W̃ T KW W
−1 ˙
 


= −sT Ks + sT (f + ud − f − ud ) − tr(W̃ T KW W)
T −1 ˙
 


= −sT Ks + sT (ε + ud ) − sT ud + sT (W T − W )σ − tr(W̃ T KW W)
T −1 ˙



= −s Ks + s (ε + ud ) − s (ud ) + s (W̃ )σ − tr(W̃ KW W )

T T T T T
(36)

Then, sT (W̃ T )σ = tr(W̃ T σ sT ) should (36) become:

−1 ˙



V̇ = −sT Ks + sT (ε + ud ) − sT (ud ) + tr[W̃ T (σ r T − KW W )] (37)

960 V. T. T. Linh et al.

Choose the weight update rule for the network as:

˙


W = KW σ sT (38)

Then the expression (37) becomes:



V̇ = −sT Ks + sT (ε + ud ) − sT (ud ) (39)

Suppose if we set the intercept on the estimation error and the noise: ε ≤ εd ,
ud  ≤ udb where εd and udb are positive constants, then (38) can be converted to:
 

V̇ ≤ −sT Ks + sT (εd + udb ) − sT ud (40)

Therefore, for simplicity, we can choose the noise component compensation control
law as follows:
s 

ud = (εd + udb ) (41)

s
Now the expression (40) will become:
T
V̇ ≤ −sT Ks + sT (εd + udb ) − sss (εd + udb )=

= −sT Ks + sT (εd + udb ) − s(εd + udb ) ≤ 0 (42)

This proves that the component V (s(t)) is a negative semi-deterministic function,

V (s(t)) ≤ V (s(0)), if s(t) is bounded at time t = 0, it is still bound at time t ≥ 0.
By defining (t) = −sT Ks, we get (t) ≤ −V̇ , and the product (t) divides with a
particular time:
t
(v)dv ≤ V (s(0)) − V (s(t)) (43)
0

Since V(s(0)) is a bounded function and V(s(t)) is a bounded and non-incrementing

function, the following result can be concluded:
t
lim (v)dv < ∞ (44)
t→∞
0

Therefore, by applying the Barbalat lemma [14] with (t) ˙ bounded, it can be shown
that lim (t) = 0. This implies that s(t) will converge to zero as time approaches infinity.
t→∞
As a result, semi-universal stability of the proposed control system can be guaranteed.

4 The Simulation Results

To test the theory and algorithm proposed above, we conduct simulation, with robot
parameters given in Table 1.
 T  T
Robot initial position state: qC = XC YC θC = 6 6 π2 .
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 961

Table 1. The robot model parameters

Parameter Value Unit

r 0.03 m
L 0.1 m
Mp 2.5 kg
Ip 0.12 kgm2
Ir 0.001 kgm2

4.1 Case 1: The Robot Works Without Interference

Let the robot move in a circular trajectory with center O(6,6), radius R = 1 m, we have
a system of equations.

xd = 6 + sin(t)
(45)
yd = 6 + cos(t)

We have some simulation results in the Matlab/simulink environment as follows.

To record the respective control performance, the mean square error (MSE) of the
respective gripping position is determined as follows:

1 
T
2
MSE = qd (i) − qC (i) (46)
T
i=1

and T is the total sampling time, qd (i), qC (i) is the set position and the real position.

Fig. 5. Robot trajectory when moving in a circle case 1

The PID and AD-RBF controllers have followed the trajectory very well when mov-
ing with no noise conditions, as shown in each figure: Fig. 5 is the tracking trajectory of
962 V. T. T. Linh et al.

Fig. 6. Stick to the x axis when the robot moves according to case 1

Fig. 7. Stick to the y axis when the robot moves according to case 1

the robot using two controllers PID and AD-RBF compared to the set trajectory. We can
see that both controllers are well attached to the preset trajectory. Next are Figs. 6, 7,
and 8 showing the robot’s response through following the x, y coordinate axes and theta
tracking angle. Here, we see that the good tracking error of the two controllers is also
shown in Fig. 9, Fig. 10 shows the control torque u1, u2, u3 to control the three motors
of the robot, the control torque is seen as stable oscillation, so it will be good for the
engine when operating. However, looking at the mean square error results in Table 2,
we find that the AD-RBF controller has a smaller error.
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 963

Fig. 8. Stick to theta angle when the robot moves according to case 1

Fig. 9. Error of tracking x, y, theta of the robot when moving according to case 1

4.2 Case 2: The Robot Works with Noise

Let the robot move in a circular trajectory with center O(6,6), radius R = 1 m, we have
the following system of equations:

xd = 6 + sin(t)
(47)
yd = 6 + cos(t)

Noise ud = [0.1sint; 0.1cost; 0]; At the 6th second, we increase the robot’s mass
from 2 to 5 kg.
In case 2, we add interference ud to the x and y axes, at the 6th second onwards,
the robot mass increases from 2 to 5 kg. The AD-RBF controller is proved to be more
adaptive than the PID controller. Specifically, Fig. 11 is the tracking trajectory of the
robot using two PID and AD-RBF controllers, we see that the PID controller has not
964 V. T. T. Linh et al.

Fig. 10. Control torque of the robot when moving according to the case 1

Table 2. The MSE of robot moving according to case 1

Error Controller
PID Adaptive-RBF (AD-RBF)
x 0.0018 7.2628 × 10–5
y 0.0184 0.0167
Theta 0.0489 0.0445

Fig. 11. Trajectory of the robot when moving according to the case 2

been able to follow the trajectory well, and is deviated from the trajectory much, while
the controller AD-RBF grips well. Figures 12, 13, and 14 show the robot’s response
through the x, y coordinate axes and theta tracking angle, at the 6th second the robot’s
mass increases, the PID controller has not yet adapted. Thus, the trajectory has been
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 965

Fig. 12. Stick to the x axis when the robot moves according to the case 2

Fig. 13. Stick to the y axis when the robot moves according to case 2

Fig. 14. Stick to theta angle when the robot moves according to case 2
966 V. T. T. Linh et al.

Fig. 15. Error of tracking x, y, theta of the robot when moving according to case 2

Fig. 16. Control torque of the robot when moving according to the case 2

Table 3. The MSE of robot moving according to case 2

Error Controller
PID Adaptive-RBF (AD-RBF)
x 0.0036 8.8367 × 10–5
y 0.0217 0.0169
Theta 0.0491 0.0446

deviated a lot. Figures 15 and 16 depict the robot’s tracking error in terms of x, y, theta
and robot control torque. Thereby, we see that the robot error goes to zero, and the control
torque is not continuously reversed, which helps the robot to preserve the equipment for
a longer time.
Looking at the mean square error (MSE) results in Table 3, we see that the AD-RBF
controller works better.
 The Orbit Tracking Adaptive Control for Omnidirectional Mobile 967

5 Conclusions
The adaptive controller built on the basis of the RBF neural network to control the three-
wheeled omnidirectional mobile robot has achieved good quality of tracking control,
with high accuracy in noisy working environment and take into account the uncertainty
factor of the model. The built controller is also proven to be stable throughout the working
area on the basis of the theory of stability of the Lyapunov function. The efficiency of
the controller has been verified through simulation on Matlab - Simulink. From the
simulation results, more research can be done to carry out experimental applications on
real robots.

References
1. Goswami, N.K., Padhy, P.K.: Sliding mode controller design for trajectory tracking of a
non-holonomic mobile robot with disturbance. Comput. Electr. Eng. 72, 307–323 (2018)
2. Alipour, K., Robat, A.B., Tarvirdizadeh, B.: Dynamics modeling and sliding mode control of
tractor-trailer wheeled mobile robots subject to wheels slip. Mech. Mach. Theory 138, 16–37
(2019)
3. Swadi, S.M., Tawfik, M.A., Abdulwahab, E.N., Almgotir-Kadhim, H.: Fuzzy-Backstepping
controller based on optimization method for trajectory tracking of wheeled mobile robot. In:
Proceedings - 2016 UKSim-AMSS 18th International Conference on Computer Modelling
and Simulation, UKSim 2016, pp. 147–152 (2016)
4. Pham, T.T., Thanh, M., Nguyen, C.-N.: Omnidirectional mobile robot trajectory tracking
control with diversity of inputs. Int. J. Mech. Eng. Robot. Res. 10(11), (2021)
5. Yoo, S.J.: Adaptive tracking control for a class of wheeled mobile robots with unknown
skidding and slipping. IET Control Theory Appl. 4(10), 2109–2119 (2010)
6. Yoo, S.J.: Adaptive tracking and obstacle avoidance for a class of mobile robots in the presence
of unknown skidding and slipping. IET Control Theory Appl. 5(14), 1597–1608 (2011)
7. Shuanghe, Y., Shuang, L., He, U.: Adaptive fuzzy trajectory tracking control of uncertain
nonholonomic mobile robots. In: IEEE International Conference on Industrial Informatics
(INDIN), pp. 481–486 (2008)
8. Yue, M., Wang, S., Zhang, Y.: Adaptive fuzzy logic-based sliding mode control for a non-
holonomic mobile robot in the presence of dynamic uncertainties. Proc. Inst. Mech. Eng.,
Part C: J. Mech. Eng. Sci. 229(11), 1979–1988 (2015)
9. Peng, S., Shi, W.: Adaptive fuzzy output feedback control of a nonholonomic wheeled mobile
robot. IEEE Access 6(c), 43414–43424 (2018)
10. Hung, N., Viet, T.D., Im, J.S., Kim, H.K., Kim, S.B.: Motion control of an omnidirectional
mobile platform for trajectory tracking using an integral sliding mode controller. Int. J. Control
Autom. Syst. 8(6), 1221–1231 (2010)
11. DaneshPanah, M., Abdollahi, A., Ostadi, H., Aghaebrahimi, H.: Comprehensive omni-
directional soccer player robots. In: Robotic Soccer (2007)
12. Vázquez, J.A., Velasco-Villa, M.: Path-tracking dynamic model based control of an omnidi-
rectional mobile robot. IFAC Proc. Vol. 41(2), 5365–5370 (2008)
13. Liu, J.: Intelligent Control Design and MATLAB Simulation. Springer (2018)
14. Slotine, J.-J.E., Li, W.: Applied Nonlinear Control. Prentice Hall, Englewood Cliffs, N.J
(1991)
 Trends in the Development of Battery Thermal
Management Systems on Electric Vehicles

S. Bazavluk1 , R. Kurmaev1 , K. E. Karpukhin1 , N. K. Minh2 , and N. K. Tuan2(B)

1 NAMI Russian State Scientific Research Center, 2 Avtomotornaya Street, Moscow, Russia
2 Faculty of Automotive and Power Machinery Engineering, Thai Nguyen University of
Technology, No. 666 Tich Luong Ward, Thai Nguyen, Vietnam
tuannkcn@tnut.edu.vn

Abstract. Recently, there has been an active growth in the number of electrified
vehicles. At the end of 2021, the number of electric vehicles was 16.5 million, three
times more than in 2018. To increase competitiveness, manufacturers of electric
vehicles increase plants’ power and battery capacity. This leads to an increase in the
release of heat into the cooling system. This article discusses the main directions
for improving the temperature control systems of electrified vehicles. A literature
review is carried out, problems that have not been studied before are noted, and
possible ways to solve them are presented. The influence of the organization of
the airflow in the air path of the cooling system on the aerodynamic resistance
is considered. A solution for the operation of electric vehicles under conditions
of extreme ambient temperatures for the battery, both positive and negative, is
proposed. The use of an immersion thermal control system on vehicles is analyzed.

Keywords: Battery thermal management systems · Electric vehicle · Vehicle ·

Traction battery · Radiator

1 Introduction

Using battery thermal management systems (BTMS) on electric vehicles (EV) makes it
possible to maintain a predetermined temperature range, on which their service life and
optimal operating parameters of electrical components [1–4].
Increasing the energy of power plants leads to increasing in the release of heat into
the cooling system. Due to the dense arrangement of components and assemblies in
modern vehicles, it is not possible to install radiators of large geometric dimensions to
ensure the necessary heat removal in all operating modes of the power plant. Therefore,
it is necessary to increase the efficiency of the structures used without changing the
weight and size parameters by improving the organization of the airflow in the air path
of the BTMS, and it is also necessary to reduce the energy consumption of the BTMS
components in order to increase the operating range of electric vehicles [5–7].

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 968–976, 2023.
https://doi.org/10.1007/978-3-031-22200-9_102
 Trends in the Development of Battery Thermal Management Systems 969

Another task of the BTMS on EV is to heat the working fluid to the optimum operating
range at low ambient temperatures. This task is relevant, since negative temperatures
negatively affect the service life of batteries, as well as the power reserve of the EV. At
the same time, a large amount of energy is expended on the operation of the heater [8,
9].
One of the promising areas for the development of thermal control systems for
electrical components of the vehicle is immersion cooling systems that can maintain the
working temperature of the traction battery within an optimum range.

2 Main Part
2.1 Airflow Management
One of the main directions for improving temperature control systems is the organization
of the flow in the air path of the radiator, as well as reducing the resistance of the radiator,
which also contributes to the overall reduction in the aerodynamic drag of the vehicle.
Petrov in his articles [10–12] considered the influence of various factors on the
aerodynamic resistance of the radiator under the influence of an uneven airflow. Two
main factors were identified that increase the aerodynamic resistance of the radiator:
the nonlinearity of the aerodynamic characteristic of the radiator (the dependence of the
aerodynamic drag coefficient of the radiator ζ on the air velocity V0 passing through it)
and the nonlinearity of the pressure drop characteristic P.
The aerodynamic resistance of the radiator is determined by the formula:

P = P0 · kζ , (1)

where P0 is the aerodynamic resistance of the radiator in a uniform field of airflow
velocities along its frontal surface; kζ is the coefficient of change in the aerodynamic
resistance of the radiator.
The unevenness of the airflow is estimated by the standard deviation of the air velocity
along the front of the radiator:

(Vi − V0 )2
σ= (2)
m−1

where Vi is the local air velocity on the i – element; V0 is the average air speed along the
radiator front m is the number of elementary sections where the speed was measured.
When conducting a numerical experiment in a wind tunnel, which is a channel with
a cross section of (0.34 × 0.48) m, corresponding to the cross section of the radiator
core, the unevenness of the airflow along the front of the radiator was set using a round
hole with a diameter of 0.25 m, installed in the chamber on a transverse partition. Flow
deformation was eliminated by the large length of the chamber −4.8 m and matching the
cross section of the channel and the core of the radiator. As a result of this experiment, the
values of the pressure drop on the radiator were obtained due to the uneven distribution
of the airflow velocity over the front surface of the radiator, while the reference point
was the option in which the diaphragm was maximally removed from the radiator and
970 S. Bazavluk et al.

its effect on the distribution of airflow along the front of the radiator was equal to zero.
The dependences of the coefficient of change in aerodynamic resistance on the root-
mean-square deviation of air velocity along the front of the radiator for radiators with
characteristics that have fundamental differences from each other are shown in Figs. 1
and 2.

Fig. 1. Dependence of the coefficient of change in aerodynamic resistance on the root-mean-

square deviation of the air velocity along the radiator front [10].

Fig. 2. Dependence of the coefficient of change in the aerodynamic resistance of the radiator
on the root-mean-square deviation of the air velocity along the front of the radiator at different
average air velocity [10].

In real conditions, the radiator installed on the vehicle is blown by an uneven airflow,
which is associated with the design of the air path, and obstacles located in the air
path. This article assumes the installation of a fan in front of the radiator, therefore, in
the experimental part, the flow unevenness was set by a round hole. In reality, objects
having a different cross section, so it is necessary to take into account the design of
the air path when determining the aerodynamic resistance of the radiator can create the
non-uniformity of the flow.
 Trends in the Development of Battery Thermal Management Systems 971

In another article [11] Petrov determined the effect of the invention according to
author’s certificate No. 1451300 on the passage of airflow through the radiator. The
negative effect of the device on the distribution of air over the surface of the radiator
is indicated, which contributes to an increase in the airflow rate within the ring located
opposite the radiator fan, while the speed decreases in the rest of the radiator (Fig. 3).

Fig. 3. Installation of the ring in front of the radiator according to the author’s certificate №
1451300 [11].

It is also noted that this solution contributes to an increase in aerodynamic resistance

to the flow of air in front of the radiator due to the restriction of the movement of the
airflow by the ring along the trajectory of least resistance. In this case, a negative impact
will be observed both when using the oncoming airflow and when the fan is running.
As a solution to this problem, the authors of the article propose the use of a partition
installed in front of the radiator in the plane of the casing edge. This will reduce air
recirculation through the radiator during fan operation due to an increase in aerodynamic
resistance when rounding this partition. In the case of free flow, the height of this baffle
must be limited for unhindered flow through the free zone of the radiator, since the fan
installation has a greater resistance.
To reduce the recirculation of air blown by the fan, it is more efficient to use a casing
with dampers that open under the action of the oncoming airflow and close when the
fan is running (Fig. 4). At the same time, the resistance of the radiator slightly increases
due to the dampers, but the distribution of airflow over the surface of the radiator will
be uniform.

2.2 Study of the Effect of Low Ambient Temperatures on the State of Traction
Battery (TB)
In the articles [13] Biksaleev and co-authors calculated the thermal state of TB under
the influence of extreme temperatures in large cities of various regions of the Russian
Federation. The value of −35.3 °C, recorded in Yekaterinburg in 2014, was used as the
972 S. Bazavluk et al.

Fig. 4. Application of a casing with dampers: a - incoming flow, b - fan operation.

minimum temperature, while the temperature of the modules was assumed to be −20 °C,
since this value was obtained by modeling the temperature of the TB modules for 16 h.
The article states that a longer downtime of the EV in operation is not provided. In this
case, +25 °C is taken as the initial cooling temperature of the modules (Fig. 5).

Fig. 5. Temperature ranges of traction battery modules [13].

The module-cooling graph is shown in Fig. 6. Taking into account the fact that
optimal operating temperatures are not reached during movement, even when using a
heater (Fig. 7), it is necessary to use control algorithms for maintaining the temperature
of the cells when the EV is parked at low ambient temperatures. This will eliminate the
 Trends in the Development of Battery Thermal Management Systems 973

risk of freezing of the battery modules, and will allow reaching the optimal temperature
range when starting to move after a long period of inactivity quickly.

Fig. 6. Graphic of cooling of the traction battery modules during the day [13].

Fig. 7. Temperature variation of traction battery during the day [13].

2.3 Immersion Cooling

Falcone and co-authors [14] consider a battery thermoregulation system based on a phase
change material, which is paraffin (RT35) with a phase transition temperature from solid
to liquid in the range of +29 °C–+36 °C. The main goal of the experiments described in
this paper is to reduce the maximum temperature inside the battery, as well as equalize
974 S. Bazavluk et al.

the temperature between the cells, which makes it possible to increase the service life
of the entire battery pack and individual cells.
For experimental studies, a battery module was used that simulates 21,700 cylindrical
lithium-ion cells, consisting of 4 cylindrical aluminum cells in which heating elements
are immersed to simulate battery operation. The 4-cell module is immersed in a container
filled with paraffin (RT35). The temperature is recorded using thermocouples at three
locations in each cell, at the top and bottom of the module, and the ambient air (Fig. 8).

Fig. 8. Location of the module in the box with paraffin [14].

The experiment consisted in carrying out 2 cycles of six tests, simulating battery
charge cycles, at an ambient temperature Tamb = +23.5 °C. In the first cycle, pure
paraffin was used and the results presented in Table 1 were obtained.

Table 1. Maximum cell temperatures during the first test cycle [14].

Units Cell 1 Cell 2 Cell 3 Cell 4

Tmax , Test 1.1 (°C) 36.52 36.40 36.16 36.48
Tmax , Test 1.2 (°C) 36.83 37.68 37.43 36.81
Tmax , Test 1.3 (°C) 36.88 37.68 37.43 36.87
Tmax , Test 1.4 (°C) 37.14 37.90 37.64 37.12
Tmax , Test 1.5 (°C) 36.51 37.36 37.11 36.51
Tmax , Test 1.6 (°C) 36.53 37.41 37.17 36.50
Temax (°C) 36.73 37.57 37.32 36.71

The authors note that all temperatures lie within the operating range, but the temper-
ature difference between the cells is not uniform, this is especially noticeable in the time
interval t = 15,000–40,000 s, when convective processes are observed. To reduce them
 Trends in the Development of Battery Thermal Management Systems 975

in the second cycle, a copper foam filler is used, which allows to reduce the maximum
temperature of the cells. The test results for the second cycle are presented in Table 2.

Table 2. Maximum cell temperatures during the second test cycle [14].

Units Cell 1 Cell 2 Cell 3 Cell 4

Tmax , Test 2.1 (°C) 35.70 34.47 36.18 35.75
Tmax , Test 2.2 (°C) 35.11 35.97 35.24 35.03
Tmax , Test 2.3 (°C) 34.60 35.63 35.58 34.52
Tmax , Test 2.4 (°C) 34.63 35.68 35.63 34.53
Tmax , Test 2.5 (°C) 34.88 35.97 34.58 35.57
Temax (°C) 34.98 35.93 35.44 35.08

Because of using copper foam as a filler, the maximum temperature of the cells is
reduced by about 2 °C there is better temperature uniformity within the module, although
there is still a temperature difference between the cells.
However, the use of such a system of thermal control of electrical components of
the vehicle may be limited due to the effect of low ambient temperatures on paraffin,
which will be in the solid phase and its melting time will be increased. In this case,
the temperature distribution between the cells will be uneven. One solution to this issue
could be the use of copper foam as a paraffin heater.
At ambient temperatures close to +40 °C, the influence of convection will be greater,
while the temperature difference between the cells will increase. For sealed systems, a
more efficient system will be required in this case.

3 Conclusions
Thus, as a result of the analysis of the literature on the battery thermal management
systems on electric vehicles, it was found that:

• when calculating the aerodynamic resistance of a radiator, uneven airflow in front of

the radiator is not mentioned;
• the use of dampers on the radiator casing to reduce air recirculation through the
radiator has not been considered;
• the influence of extreme temperatures is taken into account, but at negative air tem-
peratures there is a risk of overcooling of the traction battery modules, and also during
the movement of the traction battery does not reach the optimal temperature regime;
• the use of immersion thermal control systems on EV in conditions of negative ambient
temperatures is little studied in the literature.
976 S. Bazavluk et al.

References
1. Kim, J., Oh, J., Lee, H.: Review on battery thermal management system for electric vehicles.
Appl. Therm. Eng. 149, 192–212 (2019)
2. Kurmaev, R., Struchkov, V., Novak, V.: Experience in the development of an effective thermal
management system for the high-voltage battery of the vehicle. In: IOP Conference Series:
Materials Science and Engineering: International Automobile Scientific Forum, IASF 2019
“Technologies and Components of Land Intelligent Transport Systems”, Moscow, 16–18 Oct
2019. Institute of Physics Publishing, Moscow (2020)
3. Guodong, X., Cao, L., Guanglong, B.: A review on battery thermal management in electric
vehicle application. J. Power Sources 367, 90–105 (2017)
4. Slabosritsky, R., Khazhmuradov, M., Lukyanova, V.: Analysis and calculation of battery
cooling system. Radioelectron. Inform. 3, 3–8 (2011)
5. Tuan, N., Karpukhin, K., Terenchenko, A., Kolbasov, A.: World trends in development of
vehicles with alternative energy sources. ARPN J. Eng. Appl. Sci. 7(13), 2535–2542 (2018)
6. Kolbasov, A., Karpukhin, K., Sheptunov, D., Andrey, P., Minh, N., Tuan, N.: Analytical
study of the power parameters of electric traction drive for modern vehicles. In: Advances
in Engineering Research and Application. Lecture Notes in Networks and Systems, vol. 178,
pp. 200–209. Springer Nature Switzerland (2021)
7. Javani, N., Dincer, I., Naterer, G.: Thermodynamic analysis of waste heat recovery for cooling
systems in hybrid and electric vehicles. Energy 46(1), 109–116 (2012)
8. Karpukhin, K., Terenchenko, A., Shorin, A., Bakhmutov, S., Kurmaev, R.: Temperature
control of the battery for hybrid or electric vehicle. Biosci., Biotechnol. Res. Asia 12(2),
1297–1301 (2015)
9. Struchkov, V., Kurmaev, R.: Development of a promising temperature control system for
electric vehicles. Trudy NAMI, 29–35 (2019)
10. Petrov, A.P.: Aerodynamic resistance of the radiator in an uneven airflow. Izv. Mosc. STATE
Tech. Univ. MAMI 1(15), 158–162 (2013)
11. Petrov, A.P.: Correction of the airflow in the air path of the engine cooling system. Izv. Kurgan
State Univ. 4, 113–114 (2005)
12. Petrov, A.P.: Using a fan to measure air flow through the radiator of the cooling system. Izv.
Samara Sci. Cent. Russ. Acad. Sci. 4–2, 534–537 (2013)
13. Biksaleev, R., Karpukhin, K., Malikov, R., Klimov, A.: Simulation model of the temperature
control system of a traction battery with passive cooling. Trudy NAMI 4(283), 42–51 (2020)
14. Falcone, M., Russi, L., Pulvirenti, B.: Experimental analysis of phase-change-material-based
battery thermal management systems. Batteries 1, 0 (2022)
 Wear Analysis and Microstructural
Characterization of Al-MMC Manufactured
by Stir Casting

Pankaj Kumar Gupta1(B) , Mayank Verma1 , and Umesh Vishwakarma2

1 Malaviya National Institute of Technology, Jaipur 302017, Rajasthan, India
pankaj.mech@mnit.ac.in
2 O. P. Jindal University, Raigarh, Chhattisgarh, India

Abstract. Aluminium use has expanded recently due to its excellent strength-to-
weight ratio and strong erosive and corrosive resistance. The use of aluminium
matrix composites has grown in various industries, such as aerospace, automo-
biles, ships, and spacecraft. This paper selected Aluminium 6061 for matrix and
silicon carbide for reinforcement for better mechanical properties. Aluminium
metal matrix composite (Al-MMC) was manufactured by the stir casting method.
Cow dung ash and buffalo dung ash of 150 µm was prepared and added as rein-
forcement to decrease the density of the composite. Pin on disc tribological tests
was performed by changing the load from 10 to 40 N with a 10 N step. The wear
rate of the composite was observed at different loading conditions. Microstruc-
tural characterization of worn surfaces was analyzed using optical and scanning
electron microscopes. It was used to observe microstructure, reinforcing particle
size, shape, and dispersion in the composite to determine the physical qualities.

Keywords: Al-MMC · Cow dung ash · Buffalo dung ash · Stir casting · Wear
rate

1 Introduction
Composites, in comparison to metals, need a concerted effort spanning design, materi-
als, method, equipment, quality control, manufacturing, and process improvement [1].
The composite material consists of two phases, continuous and discontinuous. The dis-
continuous phase is harder, also termed ‘Reinforcement’ and the continuous phase is
termed ‘Matrix’ [2]. The constituent materials’ characteristics and their dispersion and
interactions considerably influence the composite’s quality [3]. The reinforcing structure
(form, spatial, size and shape) considerably impacts the composite’s characteristics in
relation to the nature of the individual components [4]. The properties of composite also
depend on the concentration and direction of reinforcement [5]. The stiffness-to-weight
ratio of composite structures is a distinct advantage over more conventional materi-
als [6]. Aluminium matrix composites have been used in various applications where
low weight and thermal stability are essential [7]. Aluminium matrix composites are
employed in multiple industries, including automotive, critical domains such as defense

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 977–983, 2023.
https://doi.org/10.1007/978-3-031-22200-9_103
978 P. K. Gupta et al.

and aerospace, and many other technology segments [8]. Various researchers have carried
out the investigations on wear and other properties of Al-MMC are following.
Poria et al. did morphology at the microstructure level using a scanning electron
microscope (SEM). It was observed that a more refined grain structure was owing to the
bonding of matrix and reinforcing materials. Wear track can also be identified using this
microstructure analysis [9]. Reddy et al. observed that composite reinforcement with
large voids, scars, cracks, and delamination was due to improper mixing, poor stirring
or insufficient wettability with 5% Tungsten carbide [10]. Vipin et al. performed a wear
test and wear was measured according to mass loss from composite. It was found that
the maximum wear was measured for unreinforced composite, and minimum wear was
observed for 15 wt% of tungsten in a matrix material. The hard particles placed into
the matrix material make the composite harder. Greater hardness reduces the wear rate
of the composite [11]. The results demonstrated that mass loss decreased as the sliding
velocity increased from 0.5 to 1.5 m/s. Further increase in velocity will result in other
mass loss as the temperature rises [12]. Suresh and Sudhakara performed a wear test
on aluminium7075 with Al2 O3 /SiC as reinforcement. It was concluded that composite
weight loss is a function of composite sliding velocities and stress applied to compos-
ite [13]. Arik et al. performed an experiment on wear tests using a pin on-disc wear
tester under dry sliding condition and concluded that the formation of stable iron based
mechanical mixed layer on the surface of Al–Al4 C3 composite during the dry sliding at
low applied loads resulted in low wear of MMCs [14]. Based on these investigations, this
article presents experimentation performed to study tribological behavior of fabricated
composite such as wear and characterization of worn surfaces.

2 Materials and Methods

2.1 Material Selection
Matrix and reinforcement are the key elements for a composite. The 6000 families of
aluminium alloys include the Al6061, one of the most widely used. An alloy of this
series has an appreciable strength-to-weight ratio, resistance to corrosion and wear, and
improved formability and weldability, providing features for manufacturing lightweight
composites.
In a composite material, one of the most significant tasks of reinforcement is to
increase mechanical properties. The reinforcement used in this research is silicon carbide.
Cow dung ash and buffalo dung ash were used as filler with reinforcement. Cow dung
and buffalo dung ash are reinforcement due to easy availability and cost-effectiveness.
It is organic in nature and eco-friendly. Raw Cow Dung Cakes and Buffalo Dung Cakes
were made and dried in sunlight for seven days and then burnt. Fine ash was collected
and 150 µm-sized powders were separated using a sieve. Figure 1 (a)–(c) show the
images of cow dung ash, buffalo dung ash and silicon carbide, respectively.

2.2 Fabrication of Al-MMC

The main furnace, graphite crucible, and other equipment are employed in this experi-
mental setup, as shown in Fig. 2. Preheating the materials used in composite fabrication
 Wear Analysis and Microstructural Characterization of Al-MMC 979

Fig. 1. a Cow dung ash, b buffalo dung ash, c silicon carbide

was the initial step in the experiment. Aluminium Al6061 matrix is melted inside the
graphite crucible. The ingot was initially warmed at 500 °C for 3–4 h. Silicon carbide,
cow dung ash, and buffalo dung ash all were preheated to 250 °C in preheater at the same
time. The graphite crucible containing the Al6061 was then heated to 820 °C. Warmed
powders were blended below their melting temperatures and melted. After that, the stirrer
is stirred into the crucible for 10 min period at 400 rpm. The aluminium matrix and the
reinforcing particles were evenly disseminated. The final furnace temperature was kept
at 820 ± 10 °C. This way, four composite samples were prepared as per the composition
based on past research given in Table 1.

Fig. 2. Stir casting setup

Table 1. Sample composition of reinforcement in Al6061 metal matrix

Sample name Al6061 (%) Silicon carbide (%) Cow dung ash (%) Buffalo dung ash (%)
S1 100 0 0 0
S2 80 10 3 3
S3 82 10 4 4
S4 84 10 5 5
980 P. K. Gupta et al.

2.3 Wear Test

The ASTM G99 standard for Pin-on-Disk equipment was followed when conducting
the wear testing. To slide against the pin, and EN31 steel disc (approximately 165 mm
diameter) was employed. The specimen had a diameter of 8 mm and a height of 33 mm.
The volumetric loss (V) is obtained by multiplying the specimens’ height loss by the
test pin’s cross-sectional area. Wear rate (Wr) = V/L (mm3 /m) is derived by dividing
volumetric loss by the sliding distance (L) [15]. Figure 3 shows the samples for the wear
test performed on the pin on the disc setup.

Fig. 3. Samples for the wear test

2.4 Surface Morphology

2.4.1 Scanning Electron Microscopy
An SEM was used to examine the worn surface. SEM can produce images at nanoscale
scales and under various pressure and temperature settings. Surface images were created
using focused electrons that generate signals when they contact atoms.

2.4.2 Optical Microscope

The microstructure has a massive effect on the overall execution of a composite. The
alloy’s microstructure, reinforcing particle size, shape, and dispersion determines the
physical qualities. The reinforcement distribution in the matrix was investigated using
an optical microscope on the manufactured samples. An inverted optical microscope
was used to study the surface of samples.

3 Results and Discussion

3.1 Wear Analysis of Composite
The dry wear rate was calculated using a pin-on-disc set up with a pin diameter of 10 mm,
a height of 33 mm and a 50 mm track radius. Sliding distance and sliding velocity were
kept constant at 1200 m and 1.2 m/s. Wear was calculated for different loads, i.e., 10,
20, 30 and 40 N. Wear rate was minimum for sample S2 as its hardness was maximum
compared to unreinforced sample S1 and reinforced samples S3 and S4, as shown in
 Wear Analysis and Microstructural Characterization of Al-MMC 981

Fig. 4. This is attributed to hard particles of the silicon carbide and the absence of ash.
The addition of ash harms hardness. This increases the porosity and the formation of
cracks reduces the mechanical property. It was also observed that the wear rate increased
as the applied load increased from 20 to 30 N and then 40 N. It is due to an increase in
the load increases the heating, making the composite surface soft. Therefore, the wear
rate increased significantly with an increase in applied load.

Fig. 4. Wear rate for different load

3.2 Micro-structural Characterization

3.2.1 Optical Microscope Characterization
An optical microscope was used on the fabricated samples to examine the reinforce-
ment distribution in the matrix. Proper mixing of reinforcement depends on the stir-
ring speed and size of reinforcement. Figure 5 shows the optical microscope images of
unreinforced and reinforced samples with a magnification of 1000×. Sample S1 seems
brighter compared to other samples showing the absence of filler material. The image
has no ash particles and silicon carbide. Sample S2 image shows that uniform mixing
of reinforcement has occurred. This attributes to the lowest wear rate. Sample S3 shows
the presence of ash particles which seems to be softer compared to S2. In sample S4,
it’s visible that improper mixing of ash particles and silicon carbide occurs, making the
composite porous due to this reason wear rate is higher compared to all other reinforced
composites.

3.2.2 SEM Characterization

A SEM examination was carried out on the worn surfaces of the reinforced Al6061
composite, as shown in Fig. 6. It was found that the composite material’s worn surface is
rougher than that of unreinforced matrix metal. Large grooves and cavities were found
982 P. K. Gupta et al.

Fig. 5. Optical microscope images of samples S1, S2, S3, and S4 (1000×)

in reinforced composite due to the hardness of reinforcement particles. Abrasive wear

mechanics was observed due to silicon carbide reinforcement. Wear tracks are also
visible in SEM images. The carbide particles resist the delamination process. Wear was
always less in composite material as compared to the unreinforced material. The worn
surface of S1 shows mico pits and wear track of unreinforced aluminium matrix. Its
shows that cracks formed are fewer and grooves are not significant due to the ductile
nature of aluminium matrix. S2 shows lesser worn surface and micro-cracks due to the
hard nature of silicon carbide. The S3 sample clearly shows a larger groove and more
cracks than the S2 sample. This shows the presence of ash which reduces hardness. S4
sample image shows a high amount of loose abrasive wear particles. This is attributed
to high porosity, which offers a high percentage of ash.

Fig. 6. SEM images of worn surface

4 Conclusion
Al6061-MMC reinforced with SiC, cow dung ash, and buffalo dung ash was fabricated
using the cost-effective stir casting method. Wear test results showed that wear rate
increased with the addition of cow dung ash and buffalo dung ash at higher loading
 Wear Analysis and Microstructural Characterization of Al-MMC 983

amounts (30 & 40 N). But minor change was observed at a lower part of loading (10 &
20 N). The reduction has occurred because of the availability of hard particles of carbides.
Optical microscopic images show the distribution and presence of reinforcement and
voids. SEM characterization represents the wear track, abrasive’s shape and size, surface
integrity, plowing, grooves and type of wear.

References
1. Suresh, S., Shenbaga Vinayaga Moorthi, N.: Aluminium-titanium diboride (Al-TiB2 ) metal
matrix composites: challenges and opportunities. Procedia Eng. 38, 89–97 (2012)
2. Prashantha, K.H.G., Anthony, X.M.: Graphene reinforced metal matrix composites. Procedia
Eng. 97, 1033–1040 (2014)
3. Stephen, B.F., et al.: Graphene–aluminium nanocomposites. Mater. Sci. Eng. A, 7933–7937
(2011)
4. Sasha, S., Dmitriy, D.A., Geoffrey, D.H.B., Kevin, K.M., Eric, J.Z.: Graphene-based
composite materials. Nature 442, 282–286 (2006)
5. Alireza, B.F., et al.: Enhanced tensile properties of aluminium matrix composites reinforced
with graphene encapsulated SiC nanoparticles. Compos. Part A: Appl. Sci. Manuf. 68,155–
163 (2015)
6. Dolatkhah, A., Golbabaei, P., Besharati Givi, M.K., Molaiekiya, F.: Investigating effects of
process parameters microstructural and mechanical properties of Al5052/sic metal matrix
composite fabricated via friction stir processing. Mater. Des. 37, 458–464 (2012)
7. Sedat, O., Recep, E., Femi, N.: Investigation of impact behavior of aluminium based SiC
particle reinforced metal matrix composites. Compos. Part A: Appl. Sci. Manuf. 38, 484–494
(2007)
8. Soon, H.J., Hong, K.M., Dae, H., Suryanarayana, C., Byong, C.S.: Effect of clustering on
the mechanical properties of SiC particulate reinforced aluminium alloy 2024 metal matrix
composites. Mater. Sci. Eng. A 347, 198–204 (2003)
9. Poria, S., Sutradhar, G., Sahoo, P.: Tribological characterization of stir-cast aluminium-TiB2
metal matrix composites. Silicon 8(4), 591–599 (2016)
10. Reddy, P.V., Prasad, P.R., Krishnudu, D.M., Goud, E.V.: An investigation on mechanical and
wear characteristics of Al 6063/TiC metal matrix composites using RSM. J. Bio- Tribo-Corros.
5, 90 (2019). https://doi.org/10.1007/s40735-019-0282-0
11. Vipin, S., Ujjwal, P., Manoj, K.B.V.: Surface composites by friction stir processing: a review.
J. Mater. Process. Technol. 224, 117–134 (2015)
12. Abdul, K.R.R., Fazlur, R.K., Asif, A., MaughalAhmed, B.A., Manzoore, S.E.M., Suvanjan,
B.: Fabrication characteristics and mechanical behavior of aluminium alloy reinforced with
Al2 O3 and coconut shell particles synthesized by stir casting. IOP Conf. Ser. Mater. Sci. Eng.
1057, 12–17 (2021)
13. Suresh, S., Sudhakara, D.: Investigations on machining and wear characteristics of Al
7075/nano-SiC composites with WEDM. J. Bio- Tribo-Corros. 5(4), 1–13 (2019). https://
doi.org/10.1007/s40735-019-0293-x
14. Arik, H., Ozcatalbas, Y., Turker, M.: Dry sliding wear behavior of in situ Al–Al4 C3 metal
matrix composite produced by mechanical alloying technique. Mater. Des. 27, 99–804 (2006)
15. Manikandan, R., Arjunan, T.V., Nath O. P, A.R.: Studies on microstructural characteris-
tics, mechanical and tribological behaviors of boron carbide and cow dung ash reinforced
aluminium (Al 7075) hybrid metal matrix composite. Compos. Part B: Eng. 183, (2020)
 Correction to: Determination of Best Input
Factors for PMEDM 90CrSi Tool Steel
Using MABAC Method

Nguyen Hong Linh, Phan Dang Phong, Thangaraj Muthuramalingam,

Tran Minh Tan, Tran Huu Danh, Vu Ngoc Pi, Hoang Xuan Tu,
and Nguyen Van Tung

Correction to:
Chapter “Determination of Best Input Factors for PMEDM
90CrSi Tool Steel Using MABAC Method” in:
D. C. Nguyen et al. (Eds.): Advances in Engineering Research
and Application, LNNS 602,
https://doi.org/10.1007/978-3-031-22200-9_36

In the original version of the book, the following belated corrections have been
incorporated: The author name “Muthuramalingam Thangaraj” has been changed to
“Thangaraj Muthuramalingam” in the Frontmatter, Backmatter and in Chapter 36. The
book has been updated with the change.

The updated original version of this chapter can be found at

https://doi.org/10.1007/978-3-031-22200-9_36

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023

D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, p. C1, 2023.
https://doi.org/10.1007/978-3-031-22200-9_104
Author Index

A D
Ala, Guido, 873 Dam, Hoang-Phuc, 395
Alexandrovich, Krylov Grigory, 354 Dang, Anh-Tuan, 93, 208, 887
Anh, Le Dinh, 433 Dang, Hoang-Anh, 297
Anh, Le Hoang, 116, 254 Dang, Tan Nguyen, 671
Anh, Pham Hoang, 177 Dang, Tien Phuc, 537
Anh, Truong Viet, 685 Dang, Van Trong, 42, 281
Arora, Amit, 733 Dang, Vuong Q., 718
Danh, Bui Thanh, 1, 99, 106, 116, 227, 244, 254,
B 262, 305, 345, 361, 369, 376, 389, 425,
462, 589, 599, 607, 613, 699, 707, 771,
Bazavluk, S., 968
864, 895, 924
Binh, Doan Thanh, 881
Danh, Tra n Huu, 227
Bui, Hai-Hung, 395
Danh, Tran Huu, 1, 99, 106, 116, 234, 244, 262,
Bui, Hai-Le, 157
305, 314, 335, 345, 361, 369, 376, 389,
Bui, Ngoc Tuyen, 128
425, 462, 589, 599, 607, 613, 699, 707,
Bui, Ngoc-Tam, 34
771, 864, 895, 924
Busacca, Alessandro, 573
Dao, Duy Trung, 942
Dao, Huy-Du, 525
C Dao, Phuong Thao, 270
Cam, Nhung Tang, 543 Dao, Quy-Thinh, 34, 150
Campagna, Nicola, 573, 873 Dao, Thi-Kien, 188
Cao, Trung Thanh, 623 Dao, Van-Dung, 297
Caruso, Massimo, 873 Dat, Duong Tan, 84
Castiglia, Vincenzo, 573, 873 Dayal, Ram, 662
Chen, Xuefeng, 806 Di Tommaso, Antonino Oscar, 873
Chuong, Trinh Trong, 685 Diem, Thuy T. T., 166
Chuyen, T. D., 952 Dien, N. D., 558
Cong, Doan Nguyen, 739 Dien, Pham Ngoc, 952
Cong, Nguyen Thanh, 823 Dinh, Ngoc Sang, 512
Cuong, Bui Van, 823, 839 Dinh, Van-Vuong, 150
Cuong, Dang Quoc, 234 Do Nhu, Y., 685
Cuong, Nguyen Manh, 254, 305, 314, 345, 369, Do, Hai T., 13
599, 699, 924 Do, Hai Tinh, 942

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2023
D. C. Nguyen et al. (Eds.): ICERA 2022, LNNS 602, pp. 985–988, 2023.
https://doi.org/10.1007/978-3-031-22200-9
986 Author Index

Do, Luu Quang, 419 Hung, Ta Tuan, 780

Do, Nhu Y, 468 Hung, Tran Cong Manh, 433
Do, Quang Duy, 455 Hung, Tran Cuong, 433
Do, The-Vinh, 188 Hung, Tran Manh, 952
Do, Thu Giang, 270 Hung, Tran Vinh, 177
Do, Trong-Hieu, 150 Hung, V. D., 952
Do, Tuan Anh, 543 Huy, Trieu Quy, 106, 227, 234, 244, 254, 262,
Doan, Cong Thang, 217 361, 376, 389, 699, 771, 864
Du, Dao Huy, 512
Duc, Khanh Nguyen, 739 K
Dung, Le Sy, 419, 505 Karpukhin, K. E., 968
Dung, Luong Viet, 116, 166 Khac Minh, Nguyen, 649
Dung, Nguyen Thi Quoc, 314, 589, 613 Khai, Do Quang, 314, 345, 369
Dunin, Andrey Y., 649 Kien, D. Nguyen, 806
Duong, Huan Thanh, 290, 412 Kumar, Manish, 662
Duong, Kien Trung, 543 Kurmaev, R., 968
Duong, Minh-Duc, 150 Ky, Le Hong, 84, 177, 254
Duong, Thanh Long, 512
Duong, Van Tu, 217, 764
L
Duy, Vu Van, 831
Lai, L. K., 558
Duyen, Le Hong Ky, 84, 177
Lam, Pham Duc, 99
Le, Chi-Quan, 395
G
Le, Duc Thinh, 42, 281
Giang, Tran Ngoc, 262, 599
Le, Lu Minh, 290, 412
Goetz, Stefan, 23
Le, Minh Vuong, 290
Gupta, Pankaj Kumar, 977
Le, Quynh C., 718, 798
H Le, Thu Quy, 942
Ha, Binh D., 718 Le, Tien-Thinh, 290, 412
Ha, Dang Viet, 74, 790, 846 Le, Trong-Phuoc, 395
Ha, Doan Viet, 433 Liem, Hoang Cong, 138
Ha, Minh Hung, 933, 942 Liem, Nguyen Ba, 234
Ha, Vo Thanh, 402 Linh, Nguyen Hong, 1, 99, 227, 244, 262, 305,
Ha, Vo Thu, 402 314, 335, 361, 425, 462, 589, 599, 607,
Hau, Tran Quang, 234 613, 707, 864, 924
Hien, Bui Thanh, 99, 106, 314 Linh, V. T. T., 952
Hien, Nguyen Thi Thu, 855 Long, Le Xuan, 74, 780, 790
Hien, Nguyen Xuan, 354 Long, Nguyen Thanh, 752
Hien, Pham Xuan, 201 Luan, Nguyen Huu, 613
Hien, Vu Thi, 74 Luu, Do Duc, 505
Hiep, Nguyen Hoang, 84, 177 Luy, N. T., 558
Hieu, Dang Trung, 419 Ly, Trinh Thi Khanh, 752
Hoa, R. V., 952
Hoang, Anh, 23 M
Hoang, Dang Danh, 633, 855 Malevich, I. Yu., 798
Hoang, Quoc Dong, 270 Mali, Harlal Singh, 580
Hoang, Quoc-Dong, 419, 505 Malik, Anup, 580
Hoang, Thang, 13 Maurya, Swaranjali, 662
Hoang, Tien-Dat, 166 Miceli, Rosario, 573, 873
Hoang, Van-Ta, 525 Mien, Trinh Luong, 831
Huan, Canh Chi, 839, 846, 881 Minh, Banh Duc, 433
Hung, Bui Duc, 495 Minh, Dinh Bui, 495
Hung, Le Xuan, 227, 345, 361, 389, 589, 707, Minh, N. K., 968
771 Minh, Nguyen Duc, 685
Hung, Pham Khanh, 419 Minh, Tran Xuan, 855
Author Index 987

N Pham, Minh Huy, 764

Nevoloso, Claudio, 873 Pham, Tu M., 23
Nga, Nguyen Thi Thanh, 234, 244, 305, 599 Pham, Tuan V., 23
Nghiem, Le Dinh, 419 Pham, Tung-Duong, 395
Ngo, Trong Binh, 933 Pham, Van Nam, 477
Ngo, Xuan Cuong, 468 Pham, Viet Hung, 281
Ngoc, Nguyen Dinh, 613, 895 Phan, Hieu Chi, 290
Nguyen, Anh Tu, 412 Phong, Phan Dang, 335
Nguyen, Cuong V., 718 Phung, Lan Xuan, 455
Nguyen, Danh Huy, 281 Phuong, Nguyen Thi Thu, 433
Nguyen, Dinh-Ngoc, 93, 208, 887 Phuong, Trieu Viet, 685
Nguyen, Dinh-Tien, 395 Pi, Vu Ngoc, 106, 234, 244, 314, 335, 345
Nguyen, Du Khac, 455
Nguyen, Duc-Canh, 150 Q
Nguyen, Huy Hung, 217, 764 Quan, Do Van, 823
Nguyen, Minh T., 13, 573, 718 Quan, Le Hong, 439
Nguyen, Minh Tuan, 798 Quang, Nguyen Huu, 425, 599
Nguyen, Minh, 873 Quang, Nguyen Thanh, 439, 448
Nguyen, Nam Hoai, 623 Quoc, Vuong Dang, 53, 495
Nguyen, Nga T. T., 166 Quynh, Le Van, 823, 839, 846
Nguyen, Ngoc-Kien, 65
Nguyen, Nhu Toan, 42, 281 R
Nguyen, Phuoc Doan, 623 Rabczuk, Timon, 806
Nguyen, Son T., 23
Nguyen, Tan Tien, 217, 764 S
Nguyen, Thanh-Tung, 395, 525 Schettino, Giuseppe, 873
Nguyen, Thong Chung, 412 Singh, Arvind Kumar, 580
Nguyen, Thuong TK., 798 Subbarao, P. M. V., 733
Nguyen, Trinh-Dong, 188
Nguyen, Trong Hai, 128 T
Nguyen, Trong-The, 188 Tam, Do Thi, 227, 361, 389, 425, 613, 707
Nguyen, Trung Kien, 455 Tam, Tran Minh, 305, 425, 599
Nguyen, Tuan-Linh, 326 Tan, Hoang Anh, 780, 839
Nguyen, Tung Lam, 42, 270, 281, 419 Tan, Tran Minh, 1, 244, 262, 314, 335, 345, 376,
Nguyen, Tung Linh, 512 864, 895, 924
Nguyen, Tung Thanh, 543 Tao, Ngo Quang, 439
Nguyen, Van Dong, 764 Thai, Lam Cuong Quoc, 217
Nguyen, Van Hai, 290, 412 Thai, Nguyen Hong, 752, 910
Nguyen, Van Trong, 217 Thangaraj, Muthuramalingam, 335
Nguyen, Van-Anh, 34 Thanh, Nguyen Trung, 633
Nguyen, Van-Tinh, 65 Thao, Tran Thi Phuong, 116, 345, 924
Nguyen, Vinh-Tiep, 188 Thi, Hue Luu, 42
Nguyen, Xuan Ngoc, 537 Thi, Thanh-Binh Nguyen, 525
Nguyen, Xuan Son, 290, 412 Thieu, Nguyen Ngoc, 234, 244
Nguyen-Bui, Nhat-Linh, 395 Thinh, Tran Ngoc Huy, 116, 361
Nguyen-Dinh, Ngoc, 166 Thoan, Tran Van, 846
Nguyen-Thi, Van-Anh, 281 Thong, V. V., 952
Nguyen-Van, Sy, 166 Thuong, Than Thi, 402, 752
Niem, Vu Thanh, 74 Toan, Nguyen Duc, 419
Tran, Dong L. T., 13
P Tran, Hoai Linh, 477
Pham, Cong Minh, 270 Tran, Hoang T., 13, 718
Pham, Duc-Binh, 34 Tran, Van Nhu, 537
Pham, Duy-Thi, 34 Trinh, Mien L., 718
Pham, Minh Duc, 270 Trinh, Minh-Chien, 150
988 Author Index

Trung, Nguyen Thanh, 910 Van Nguyen, Tung, 455

Truong, Viet Anh, 512 Van Phuong, Duong, 895
Truyen, Vu The, 780 Van Quynh, Le, 790, 881
Tu, Hoang Xuan, 106, 262, 335, 361, 369, 462, Van Thom, Phung, 910
589, 607, 707, 864 Van Trang, Nguyen, 1, 369, 376, 462, 607, 771,
Tu, Nguyen Thanh, 254, 425, 699 924
Tuan, Do Anh, 201, 512 Van Tuan, Nguyen, 74, 780
Tuan, Le Anh, 505 Van Tung, Nguyen, 99, 335, 389, 462, 607
Tuan, N. K., 968 Van Van, Luong, 903
Tuan, Nguyen Anh, 305, 389, 462, 589, 607, 613, Van, Trong Dang, 270
707 Verma, Mayank, 977
Tuan, Trinh Kieu, 116, 305, 895 Viet, Tan Nguyen, 53
Tung, Luu Anh, 1, 227, 376, 864, 895 Viola, Fabio, 573, 718, 873
Tung, Nguyen Thanh, 903 Vishwakarma, Umesh, 977
Vu, Duong, 933, 942
V Vu, Le Anh, 846
Van Anh, Vu Thi, 354 Vu, Ngoc-Kien, 525
Van Cuong, Bui, 881 Vu, Quoc Tuan, 764
Van Cuong, Nguyen, 699 Vu, Thanh Nguyen, 53
Van He, Ngo, 138 Vu, Van Tan, 537
Van Hien, Bui, 685
Van Hien, Ngo, 138 Z
Van Khoa, Vu, 99, 106 Zayats, P. V., 798
Van Nguyen, Dung, 649 Zhuang, Xiaoying, 806