Springer Series in Reliability Engineering

Mangey Ram
Hoang Pham Editors

Reliability and
Maintainability
Assessment
of Industrial Systems
Assessment of Advanced Engineering
Problems
Springer Series in Reliability Engineering

Series Editor
Hoang Pham, Department of Industrial and Systems Engineering, Rutgers
University, Piscataway, NJ, USA
More information about this series at https://link.springer.com/bookseries/6917
Mangey Ram · Hoang Pham
Editors

Reliability
and Maintainability
Assessment of Industrial
Systems
Assessment of Advanced Engineering
Problems
Editors
Mangey Ram Hoang Pham
Department of Mathematics, Computer Department of Industrial and Systems
Science and Engineering Engineering
Graphic Era University Rutgers University
Dehradun, Uttarakhand, India Piscataway, NJ, USA
Institute of Advanced Manufacturing
Technologies
Peter the Great St. Petersburg Polytechnic
University
Saint Petersburg, Russia

ISSN 1614-7839 ISSN 2196-999X (electronic)

Springer Series in Reliability Engineering
ISBN 978-3-030-93622-8 ISBN 978-3-030-93623-5 (eBook)
https://doi.org/10.1007/978-3-030-93623-5

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2022
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

Advanced technology instigated complexities in industries whatever in engineering

system or production, that resulting the increase risk factor. To control and manage
the risk factor, reliability analysis is necessary. This book entitled “Reliability and
Maintainability Assessment of Industrial Systems—Assessment of Advanced Engi-
neering Problems” enlightens the engineering problems related to their reliability
and maintenance due to increased complicacy into the production or manufacturing
in addition to superiority, general consciousness besides the perseverance of the level
of excellence of the system, updated rules and regulation regarding equipment relia-
bility, necessities for the specification of dependability and maintainability enactment
recommended by the government, and effects on income caused by the high cost of
failures and repairs.
Through this book, entitled “Reliability and Maintainability Assessment of Indus-
trial Systems—Assessment of Advanced Engineering Problems,” the engineers and
researchers have to lead a greater awareness and help them in the applications of reli-
ability and maintainability engineering. The book is meant for those who take relia-
bility engineering as a subject of study. The material is envisioned for the spectators
at the level of postgraduate or senior undergraduate students.

Dehradun, India Mangey Ram

Piscataway, USA Hoang Pham

Acknowledgements The editors acknowledge Springer for this opportunity and professional
support. Also, we would like to thank all the chapter authors and reviewers for their availability for
this work.

v
Contents

Dynamic Availability Analysis for the Flexible Manufacturing

System Based on a Two-Step Stochastic Model . . . . . . . . . . . . . . . . . . . . . . . 1
Wenbin Zeng, Guixiang Shen, Ilia Frenkel, Igor Bolvashenkov,
Jörg Kammermann, Hans-Georg Herzog, Lev Khvatskin,
and Anatoly Lisnianski
Integrated Reliability and Risk Assessments of Nuclear Facilities . . . . . . 21
Vanderley de Vasconcelos, Graiciany de Paula Barros,
Antônio Carlos Lopes da Costa, and Wellington Antonio Soares
Computational Tools of Media Analysis for Corporate Policy
Effectiveness Evaluation: Models and Their Reliability . . . . . . . . . . . . . . . 51
Gregory S. Khvatsky, Dmitry G. Zaytsev, Valentina V. Kuskova,
and Anna A. Sokol
Optimal Design of Checkpoint Systems with General Structures,
Tasks and Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
Kenichiro Naruse and Toshio Nakagawa
Series System Reliability: An Unified Approach Using Fatal Shock
and Stress Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Ricardo Puziol de Oliveira, Marcos Vinicius de Oliveira Peres,
Wesley Bertoli, and Jorge Alberto Achcar
The New Attempt of Network Reliability in the Assessment
of Business Activities for Precision Management . . . . . . . . . . . . . . . . . . . . . . 105
Shin-Guang Chen
Optimal Design of Reliability Acceptance Sampling Plans
for Multi-stage Production Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
M. Kumar

vii
viii Contents

The Importance of Technical Diagnostics for Ensuring

the Reliability of Industrial Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
D. Lj. Branković, Z. N. Milovanović, and V. Z. Janičić Milovanović
Reliability Assessment of Replaceable Shuffle Exchange Network
with an Additional Stage Using Interval-Valued Universal
Generating Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Amisha Khati and S. B. Singh
Modeling Software Vulnerability Injection-Discovery Process
Incorporating Time-Delay and VIKOR Based Ranking . . . . . . . . . . . . . . . 239
Mohini Agarwal, Deepti Aggrawal, Subhrata Das, Adarsh Anand,
and Navneet Bhatt
Assessment of Reliability Function and Signature of Energy Plant
Complex System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
Monika Negi, Megha Shah, Akshay Kumar, Mangey Ram,
and Seema Saini
Reliability Evaluation and Cost Optimization of Solar
Air-Conditioner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
Ashok Singh Bhandari, Mangey Ram, Akshay Kumar,
and Sushil Chandra Dimri
Analyzing Interrelationships Among Software Vulnerabilities
Using Fuzzy DEMATEL Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Misbah Anjum, P. K. Kapur, Vernika Agarwal, and Vivek Kumar
Universal Generating Function Approach for Evaluating
Reliability and Signature of All-Digital Protection Systems . . . . . . . . . . . . 301
Soni Bisht and S. B. Singh
Reliability Analysis of 8 × 8 SEN- Using UGF Method . . . . . . . . . . . . . . . . 329
Vaibhav Bisht and S. B. Singh
Editors and Contributors

About the Editors

Prof. Mangey Ram received the Ph.D. degree major in Mathematics and minor
in Computer Science from G. B. Pant University of Agriculture and Technology,
Pantnagar, India. He has been a faculty member for around thirteen years and has
taught several core courses in pure and applied mathematics at undergraduate, post-
graduate, and doctorate levels. He is currently Research Professor at Graphic Era
(Deemed to be University), Dehradun, India, and Visiting Professor at Peter the Great
St. Petersburg Polytechnic University, Saint Petersburg, Russia. Before joining the
Graphic Era, he was Deputy Manager (Probationary Officer) with Syndicate Bank
for a short period. He is Editor-in-Chief of International Journal of Mathematical,
Engineering and Management Sciences; Journal of Reliability and Statistical Studies;
Journal of Graphic Era University; Series Editor of six Book Series with Elsevier,
CRC Press-A Taylor and Frances Group, Walter De Gruyter Publisher Germany,
River Publisher, and Guest Editor and Associate Editor with various journals. He
has published 250 plus publications (journal articles/books/book chapters/conference
articles) in IEEE, Taylor & Francis, Springer Nature, Elsevier, Emerald, World Scien-
tific and many other national and international journals and conferences. Also, he
has published more than 50 books (authored/edited) with international publishers
like Elsevier, Springer Nature, CRC Press-A Taylor and Frances Group, Walter De
Gruyter Publisher Germany, River Publisher. His fields of research are reliability
theory and applied mathematics. He is Senior Member of the IEEE, Senior Life
Member of Operational Research Society of India, Society for Reliability Engi-
neering, Quality and Operations Management in India, Indian Society of Industrial
and Applied Mathematics. He has been a member of the organizing committee of a
number of international and national conferences, seminars, and workshops. He has
been conferred with “Young Scientist Award” by the Uttarakhand State Council for
Science and Technology, Dehradun, in 2009. He has been awarded the “Best Faculty
Award” in 2011; “Research Excellence Award” in 2015; “Outstanding Researcher
Award” in 2018 for his significant contribution in academics and research at Graphic

ix
x Editors and Contributors

Era Deemed to be University, Dehradun, India. Recently, he has been received the
“Excellence in Research of the Year-2021 Award” by the Honorable Chief Minister
of Uttarakhand State, India.

Dr. Hoang Pham is Distinguished Professor and former Chairman (2007–2013) of

the Department of Industrial and Systems Engineering at Rutgers University, New
Jersey, USA. Before joining Rutgers, he was Senior Engineering Specialist with the
Boeing Company and the Idaho National Engineering Laboratory. He has been served
as editor-in-chief, editor, associate editor, guest editor, and board member of many
journals. He is Editor of Springer Book Series in Reliability Engineering and has
served as conference chair and program chair of over 40 international conferences.
He is the author or coauthor of 7 books and has published over 200 journal articles, 100
conference papers, and edited 15 books including Springer Handbook in Engineering
Statistics and Handbook in Reliability Engineering. His numerous awards include
the 2009 IEEE Reliability Society Engineer of the Year Award. He is Fellow of the
Institute of Electrical and Electronics Engineers (IEEE) and the Institute of Industrial
Engineers (IIE).

Contributors

Jorge Alberto Achcar University of São Paulo, São Paulo, Brazil

Mohini Agarwal Amity School of Business, Amity University Uttar Pradesh,
Noida, India
Vernika Agarwal Amity International Business School, Amity University, Noida,
Uttar-Pradesh, India
Deepti Aggrawal University School of Management and Entrepreneurship, Delhi
Technological University, Delhi, India
Adarsh Anand Department of Operational Research, University of Delhi, Delhi,
India
Misbah Anjum Amity Institute of Information Technology, Amity University,
Noida, Uttar-Pradesh, India
Wesley Bertoli Federal University of Technology, Curitiba, Paraná, Brazil
Ashok Singh Bhandari Department of Mathematics, Graphic Era Hill University,
Uttarakhand, India
Navneet Bhatt Anil Surendra Modi School of Commerce, SVKM’s Narsee Monjee
Institute of Management Studies (Deemed to be University), Mumbai, India
Soni Bisht Department of Mathematics, Eternal University, Himachal Pradesh,
Baru Sahib, India
Editors and Contributors xi

Vaibhav Bisht Department of Mathematics, Statistics and Computer Science, G.B.

Pant University of Agriculture and Technology, Pantnagar, India
Igor Bolvashenkov Institute of Energy Conversion Technology, Technical Univer-
sity of Munich (TUM), Munich, Germany
D. Lj. Branković Department of Hydro and Thermal Engineering, Faculty of
Mechanical Engineering Banja Luka, University of Banja Luka, Banja Luka, Bosnia
and Herzegovina
Shin-Guang Chen Tungnan University, New Taipei City, Taiwan
Subhrata Das Department of Operational Research, University of Delhi, Delhi,
India
Antônio Carlos Lopes da Costa Centro de Desenvolvimento da Tecnologia
Nuclear - CDTN, Pampulha, Belo Horizonte, Minas Gerais, Brazil
Ricardo Puziol de Oliveira Maringá State University, Maringá, Brazil
Marcos Vinicius de Oliveira Peres Maringá State University, Maringá, Brazil
Graiciany de Paula Barros Centro de Desenvolvimento da Tecnologia Nuclear -
CDTN, Pampulha, Belo Horizonte, Minas Gerais, Brazil
Vanderley de Vasconcelos Centro de Desenvolvimento da Tecnologia Nuclear -
CDTN, Pampulha, Belo Horizonte, Minas Gerais, Brazil
Sushil Chandra Dimri Department of Computer Science and Engineering,
Graphic Era Deemed To Be University, Uttarakhand, India
Ilia Frenkel Center for Reliability and Risk Management, Shamoon College of
Engineering, Beer Sheva, Israel
Hans-Georg Herzog Institute of Energy Conversion Technology, Technical
University of Munich (TUM), Munich, Germany
Jörg Kammermann Institute of Energy Conversion Technology, Technical Univer-
sity of Munich (TUM), Munich, Germany
P. K. Kapur Amity Center for Inter-Disciplinary Research, Amity University,
Noida, Uttar-Pradesh, India
Amisha Khati Department of Mathematics, Statistics and Computer Science, G.
B. Pant University of Agriculture and Technology, Pantnagar, India
Lev Khvatskin Center for Reliability and Risk Management, Shamoon College of
Engineering, Beer Sheva, Israel
Gregory S. Khvatsky International Laboratory for Applied Network Research,
HSE University, Moscow, Russia
Akshay Kumar Department of Mathematics, Graphic Era Hill University,
Dehradun, Uttarakhand, India
xii Editors and Contributors

M. Kumar Department of Mathematics, National Institute of Technology, Calicut,

Kerala, India
Vivek Kumar Department of Operational Research, University of Delhi, Delhi,
India
Valentina V. Kuskova International Laboratory for Applied Network Research,
HSE University, Moscow, Russia
Anatoly Lisnianski Center for Reliability and Risk Management, Shamoon
College of Engineering, Beer Sheva, Israel
V. Z. Janičić Milovanović Banja Luka, Bosnia and Herzegovina
Z. N. Milovanović Department of Hydro and Thermal Engineering, Faculty of
Mechanical Engineering Banja Luka, University of Banja Luka, Banja Luka, Bosnia
and Herzegovina
Toshio Nakagawa Department of Business Administration, Aichi Institute of
Technology, Aichi, Japan
Kenichiro Naruse Faculty of Social and Environmental Studies, Josai International
University, Chiba, Japan
Monika Negi Department of Mathematics, Graphic Era Hill University, Dehradun,
Uttarakhand, India
Mangey Ram Department of Mathematics, Computer Science and Engineering,
Graphic Era Deemed To Be University, Uttarakhand, India;
Institute of Advanced Manufacturing Technologies, Peter the Great St. Petersburg
Polytechnic University, Saint Petersburg, Russia
Seema Saini Department of Mathematics, Graphic Era Deemed To Be University,
Dehradun, Uttarakhand, India
Megha Shah Department of Mathematics, Graphic Era Hill University, Dehradun,
Uttarakhand, India
Guixiang Shen School of Mechanical and Aerospace Engineering, Jilin University,
Changchun, China
S. B. Singh Department of Mathematics, Statistics and Computer Science, G.B.
Pant University of Agriculture and Technology, Pantnagar, India
Wellington Antonio Soares Centro de Desenvolvimento da Tecnologia Nuclear -
CDTN, Pampulha, Belo Horizonte, Minas Gerais, Brazil
Anna A. Sokol International Laboratory for Applied Network Research, HSE
University, Moscow, Russia
Dmitry G. Zaytsev International Laboratory for Applied Network Research, HSE
University, Moscow, Russia
Editors and Contributors xiii

Wenbin Zeng Fine Mechanics and Physics, Changchun Institute of Optics, Chinese
Academy of Science, Changchun, China
Dynamic Availability Analysis
for the Flexible Manufacturing System
Based on a Two-Step Stochastic Model

Wenbin Zeng, Guixiang Shen, Ilia Frenkel, Igor Bolvashenkov,

Jörg Kammermann, Hans-Georg Herzog, Lev Khvatskin,
and Anatoly Lisnianski

Abstract The paper proposes a dynamic availability analysis approach for the flex-
ible manufacturing system (FMS) under a stochastic environment that machines’
failure and starvation or blockage of production process occur randomly. Accurately
knowing the availability of the FMS, which is changing dynamically overtime in a
stochastic circumstance could benefit a lot for the improvement or re-design of the
system. A two-step stochastic model proposed in current paper that integrates the
intermediate buffers into associated workstations equivalently in terms of the rela-
tionships between upstream and downstream production rates. Calculation proce-
dures of relevant dynamic availability are established by using the Lz-transform
method, which conquer the states-explosion problem which is common in FMS
performance analysis. Meanwhile, the impacts of intermediate buffers on the FMS

W. Zeng
Fine Mechanics and Physics, Changchun Institute of Optics, Chinese Academy of Science,
Changchun 130022, China
e-mail: zengwb15@yeah.net
G. Shen
School of Mechanical and Aerospace Engineering, Jilin University, Changchun 130022, China
e-mail: shengx@jlu.edu.cn
I. Frenkel (B) · L. Khvatskin · A. Lisnianski
Center for Reliability and Risk Management, Shamoon College of Engineering, 84100 Beer
Sheva, Israel
e-mail: iliaf@frenkel-online.com
A. Lisnianski
e-mail: lisnianski@bezeqint.net
I. Bolvashenkov · J. Kammermann · H.-G. Herzog
Institute of Energy Conversion Technology, Technical University of Munich (TUM), 80333
Munich, Germany
e-mail: igor.bolvashenkov@tum.de
J. Kammermann
e-mail: joerg.kammermann@tum.de
H.-G. Herzog
e-mail: hg.herzog@tum.de

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 1

M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_1
2 W. Zeng et al.

dynamic availability also revealed, which assist to determine the appropriate volumes
of the buffers to satisfy the various FMS production demands. A numerical example
is presented for illustrating the effectiveness and rationality of the approach.

Keywords Flexible manufacturing system · Dynamic availability analysis ·

Two-step stochastic model · Intermediate buffers · Multi-state system ·
Lz-transform method

1 Introduction

With the increasing demand of customization and high-quality products, leveraging

the advantages of new technologies with the context of Industry 4.0 is seen as an
important factor for manufacturers to maintain competitiveness while responding
to the sustainability challenge [11, 27]. Industry 4.0, predicted as the fourth indus-
trial revolution, will implant the information technology deeper into the manufac-
turing automation [24]. Namely, Industry 4.0 will use the Internet of things (IoT)
in production systems to create a cyber-physical production system (CPPS) [17],
which is characterized by sharing production operation information over the Internet
with multiple systems inside the smart factory [16]. Flexible manufacturing systems
(FMS) is an inevitable choice for current companies to respond the rapid market
changes and customer customization requirements, which turns into a hot topic and
plays an important role in the realization of the Industry 4.0. It can achieve the goals
of Industry 4.0 that include a higher level of operational efficiency and productivity,
as well as a higher level of automatization [12, 35]. Therefore, the research on the
performance analysis of FMS is not only beneficial to its design, improvement and
control, but also can be used as one of the foundations for implementing intelligent
factory.
FMS is generally defined as an integrated computer-controlled system with a set
of workstations, interconnected by an automated material handling system (MHS)
and which can simultaneously perform several different tasks [3, 9]. It is learned
from Groover [9] and Jin and Liu [14] that the workstations (such as CNC machine
tools, industrial robots, washing and measuring machines) are processing stations
of FMS, where several machines are utilized at the same stage to meet the produc-
tivity and production balance. The automated material handling system is used for
transferring parts from one station to another, while the central computer is used
for controlling and coordinating the performance of the workstations and material
handling system. Moreover, the finite intermediate buffers are set between worksta-
tions for the completion of the processing task when the upstream or downstream
devices in a failed circumstance. Due to the essential and complicated nature of the
FMS, there have been various studies on the performance metrics analysis of FMS
as well as the flexible manufacturing cell (FMC).
Under the assumption that elements of the FMS are available at all times, Singholi
et al. [33] presented a framework based on the Taguchi experimental design for
Dynamic Availability Analysis for the Flexible Manufacturing … 3

evaluating the effects of varying levels of machine and routing flexibility on the
makespan, average waiting time (AWT), and average utilization of an FMS. Dosdogru
et al. [6] utilized an integrated genetic algorithm, the Monte Carlo method, to solve
the stochastic flexible job shop-scheduling problem and to measure the impact of
routing flexibility on shop performance. Jain and Raj [13] performed three different
approaches to analyze the intensity of performance variables in an organization and
proposed the FMS performance index to intensify the factors, which affect FMS. In
a multi-stage production system, Gyulai et al. [10] proposed a simulation-based opti-
mization method that utilizes lower-level shop floor data to calculate robust produc-
tion plans for final assembly lines of a flexible, multistage production system. Rybicka
et al. [26] demonstrated how discrete event simulation (DES) can address complexity
in an FMS to optimize the production line performance. However, the reliability of
machines has a considerable impact on the practical performance of manufacturing
systems. The disturbances caused by these breakdowns lead to scheduling prob-
lems, which decrease the productivity of the entire manufacturing process. This
issue points out an important tangible for the consideration of machine reliability in
the performance evaluation of FMS, especially in light of the increasing complexity
of such systems in recent years. Consequently, many researchers have realized the
importance of reliability features of the FMS and much work has been contributed.
Koulamas [15] developed a semi-Markov model to study the effects of tool failures
on the performance of an FMC. Das et al. [5] proposed an approach that provides
a flexible routing which ensures high overall performance of the cellular manu-
facturing system (CMS) by minimizing the impact of machine failure through the
provision of alternative process routes in case of any machine failure. Elleuch et al.
[8] proposed a Markov-based model for reducing the severity of breakdowns and
improving performances of the CMS with unreliable machines. Loganathan et al.
[23] suggested a methodology for availability evaluation of manufacturing systems
using semi-Markov model, which considers variable failure or repair rates. Zeng et al.
[38] developed a multi-state transition model for CNC machine tools that included the
setup of production, and the Lz-transform method was introduced to determine the
productivity and availability evaluation of the multi-state FMS. Savsar [28] developed
stochastic models to determine the performance of an FMC under random opera-
tional conditions, including random failures of cell components in addition to random
processing times, random machine loading and unloading times, and random pallet
transfer times. Moreover, Savsar [30] developed a stochastic multi-states model to
analyze performance measures of a cell, which is allowed to operate under degraded
mode and the results can be useful for practitioners to analyze performance of an FMC
at the design or operational stage. Chen et al. [4] proposed a process performance
evaluation chart to examine the manufacturing performance of bearing connectors,
which can also be used for multiple evaluations of manufacturing performance in
other manufacturing industries. Manocher and Hamid [25] developed a discrete-event
simulation model to investigate the effectiveness of a reliability plan focusing on the
most critical machines in manufacturing cell. More studies about the performance
analysis of manufacturing systems can be found in papers of Anupma and Jayswal
[2], Savsar [29], Li and Huang [18], Alhourani [1].
4 W. Zeng et al.

Although the aforementioned studies have been undertaken, only few studies that
include the influence of intermediate buffers on performance measures of an FMS.
Tan and Gershwin [34] proposed a tool for performance evaluation of general Marko-
vian continuous material flow of two-processing stage production systems with one
intermediate buffer. Liu et al. [22] modeled and analyzed the throughput of a two-
stage manufacturing system with multiple independent unreliable machines at each
stage and one finite-sized buffer between the stages. Duan et al. [7] suggested a reli-
ability modeling and evaluating methodology for a repairable, non-series, multistate
FMS with finite buffers by using an improved vector universal generation function to
satisfy the market demands on capacity and capability of the system. However, only
steady-state probabilities of FMS performance metrics such as steady-state avail-
ability, theoretical production rates, utilization rates, efficient production rates and
production loss are calculated. Therefore, a two-step stochastic model is presented in
current paper as an attempt to evaluate the dynamic performance of FMS, especially
the availability, and to meet various production demands through select appropriate
buffers volumes.
The remaining article is structured as follows. The description of the FMS and
corresponding hypothesis are shown in the section “Problem Description”. Then
the section “Two-Step Stochastic Model” introduces the procedure of establishing
the required model in detail. Subsequently, an exploratory “Numerical Example” is
utilized to present the effectiveness of the proposed approach and remarks as well as
future work discussed in the “Conclusion”.

2 Problem Description

The flexible manufacturing system (FMS) investigated in this paper consist of m

flexible manufacturing cells (FMCs) and corresponding intermediate buffers for a
part family, the schematic shown in Fig. 1. Each FMC contains various numbers of

Fig. 1 Schematic of the FMS consisting of multiple FMCs and corresponding buffers: F MC is
the abbreviation of the flexible manufacturing cell, M, R and B represent the processing machines,
robots and intermediate buffers, respectively
Dynamic Availability Analysis for the Flexible Manufacturing … 5

the identical processing machines and a robot for automatic loading and unloading.
Note that the machines process the same work independently while connect in series
with the robot separately. The buffers are situated between every two FMCs.
The (i)th FMC as the example to illustrate the multi-state characteristics of the
device, (i) = 1, . . . , m. Machine M (i)
j , j = 1, . . . , H is one machine of the (i)th
 
FMC that has K j performance rates, represented with vector g j = g j1 , . . . , g(i)
(i) (i)
jK
j

and robot R (i) is a binary-state

 device
 (working or total failure) with the relevant
performance vector r = r0 , r1 . Let g(i)
(i) (i) (i) (i)
j1 and r0 represent the worst performance

issue (total failure) for machines and robot while g(i) (i)
j K and r1 as the best. Therefore,
j

the performanceof the FMC G (i)

F MC (t) at any instant t ≥ 0 is a random variable where
(i) (i)
 H (i) 
G F MC (t) ∈  f r , j=1 g j ,  f is the structure operator. Including all devices
in this FMC, the state space of the FMC regarded as the product of all processing
machines and robot states.
In order to achieve the assessment of the dynamic availability of the FMS,
following assumptions are listed:
I. Strategy for all devices with corrective maintenance and repair “as bad as
old”, i.e. minimal repair. Moreover, repair time of all devices follow exponen-
tial distribution. Processing machines have aging characteristics and failure
rates obey Weibull distribution while failure rates of robots follow exponential
distribution.
II. The failure of any equipment is state-based, that is, the equipment only fails
during the processing of the blank, while does not fail during the shutdown
process caused by starvation or blockage.
III. Blank transportation time and robots loading and unloading time are negligible
compared to the machines’ processing time. There is no fault during the blank
transition.
IV. All machines located in the first FMC have enough blanks to be machined
and all machines located in the terminal FMC have enough storage space. In
addition, when the buffer B (i) is under blockage or starvation, all FMCs in the
upstream and in the downstream are shut down, respectively.

3 The Two-Step Stochastic Model

Duan et al. [7] studied the relationships between the buffers and the productivities
of upstream and downstream for establishing the steady-state transition equation
in order to obtain the probability of the blank remaining in the buffer. However,
productivities of the workstations are set as constant which entail the aging and
performance degradation of machines that apparently have influence on the produc-
tivity of system are excluded in that equation. Consequently, an attempt that brings
in such factors to explore the dynamic performance of FMS is made in current
6 W. Zeng et al.

paper, which summarizes a two-step stochastic model. The specific procedures are
as follows.

3.1 Step 1: Establishment of the Equivalent FMC Model

The FMS steady-state performance models had been established by Duan et al.
[7]. This model cannot be directly used for systems with stochastic characteristics.
However, the logic and ideas of analyzing that problem are still worth learning.
Zhou and Liu [39] proposed the method by that the FMS (shown in Fig. 1) can be
decomposed into (m − 1) processing domains with two FMCs and one buffer, which
is shown in Fig. 2.
Supposing the volume of buffer B (i) is bi , the production rate of upstream F MC (i)
is P R (i) (t) and the downstream F MC (i+1) is P R (i+1) (t). Then in a very short period,
at most one blank enters or leaves the buffer, and the intensity of the workpiece
entering and leaving the buffer area is determined by the productivity of the upstream
and downstream stations. Therefore, the change process of the number of blanks in
the buffer is shown in Fig. 3.
The change process of the blanks in the buffer B (i) could be regarded as a nonlinear
birth–death process or a discrete state continuous-time Markov process [37], the
corresponding state transition intensity matrix a(i) is

Fig. 2 Schematic of one processing domain

Fig. 3 The change process of the number of blanks in the buffer

Dynamic Availability Analysis for the Flexible Manufacturing … 7

⎡ ⎤
−P R (i) (t) P R (i) (t) ··· 0 0
⎢ PR  ⎥
⎢ (i+1) (t) − P R (i) (t) + P R (i+1) (t) P R (i) (t) ··· 0 ⎥
⎢ ⎥
⎢ . . . . . ⎥
a(i) =⎢ . . . . . ⎥
⎢ . . . . . ⎥
⎢  ⎥
⎣ 0 ··· P R (i+1) (t) − P R (i) (t) + P R (i+1) (t) P R (i) (t) ⎦
0 ··· 0 P R (i+1) (t) −P R (i+1) (t)

Moreover, Zhou and Liu [39] explained that the initial probability distribution
has no effect on the dynamic probability of the above process, then the dynamic
probability of the number of blanks in B (i) at time t can be calculated. Subsequently,
considering the influence of the front and rear buffers, the FMC has the following
five independent states and equivalent probabilities:
(1) The equivalent probability for F MC (i) operates normally is
B
p(i−1)0 (t)PV(i) (t) pib
B
(t);
i

(2) The equivalent probability for F MC (i) that lack of blanks while output normal
B
is p(i−1)0 (t)PV(i) (t) pib
B
(t);
i
(3) The equivalent probability for F MC (i) that input normal while output blocked
B
is p(i−1)0 (t)PV(i) (t) pib
B
i
(t);
(4) The equivalent probability for F MC (i) that input empty while output blocked
B
is p(i−1)0 (t)PV(i) (t) pib
B
i
(t);
∼
(5) The equivalent probability for F MC (i) that equipment failure is PV(i) (t)
where, PV(i) (t) is the dynamic probability of F MC (i) normal operation when
B
the impact of the buffer is excluded, p(i−1)0 (t) and p(i−1)0
B
(t) are the dynamic
probabilities of B (i−1) for non-starvation and starvation respectively, pib B
(t)
i
(i)
B
and pibi
(t) are the dynamic probabilities of B for non-blockage and blockage
respectively.
Therefore, the equivalent FMS can be established on the basis of these proba-
bilistic relationships, and the result is shown in Fig. 4.
According to the content in Fig. 4 and the above analysis results, it is necessary
to determine the dynamic productivity of each FMC and to determine the equivalent

Fig. 4 Schematic of the

equivalent FMS
8 W. Zeng et al.

system performance model for completing the dynamic performance analysis. Hence,
the Lz-transform method is introduced to finish the task and the specific analysis
procedure is as follows.
Machine M (i)
j , j = 1, . . . ., H in the (i)th FMC has K j states, where transition
intensities between states may be the function of time t, represented with a tran-
sition intensity matrix a j = a juv (t) , j = 1, . . . ., H, u, v = 1, . . . ., K j . Then the
machine can be regarded  as a discrete-state
 continuous-time (DSCT) Markov process
(i) (i) (i) (i)
[36] G j (t) ∈ g j = g j1 , . . . , g jK which is descripted by the following triplet
j
notation:
 
G (i) (i)
j (t) = g j , a j , p j (0) , (1)

where p j (0) is initial probability distribution of the machine which generally defines
the state K j with the highest performance level as the initial state, which is expressed
by (2).
  
p j (0) = p j1 (0) = Pr G (i) (i)
j1 (0) = g j1 , . . . ,
 
p jK j (0) = Pr G (i)
jK (0) = g (i)
jK = [0, . . . , 1] (2)
j j

Chapman-Kolmogorov differential equations should be solved in order to find

state probabilities for such a system. When the state-transition diagram for the
machine is built, Eq. (3) can be written as follows.

dp ju (t)  Kj

Kj
= p jv (t)a juv (t) − p ju (t) a juv (t), u, v = 1, . . . ., K j (3)
dt
v=1 v=1
v = u v=u
 
Introducing the row-vector p j (t) = p j1 (t), p j2 (t), . . . , p jK j (t) , then Eq. (3)
can rewrite in matrix notation:
dp j (t)
= p j (t)a j (4)
dt
Consequently, the probabilities of the states at the moment t should be found as
a solution of the system (4) under the initial conditions (2).
When the machine state transitions are caused by its failures and repairs events, the
corresponding transition intensities are expressed by the machine’s failure and repair
rates [20, 21]. Without lose generality, the stochastic process of the machine contains
minor and major failures and repairs. The state-transition process with corresponding
state performance rates of the machine is presented in Fig. 5.
Dynamic Availability Analysis for the Flexible Manufacturing … 9

Fig. 5 State-transition
process for repairable
machine with minor and
major failures and repairs

Once the dynamic probabilities of the machine’s states are obtained, the Lz-
transform of discrete-state continuous-time (DSCT) Markov process (1) can be
defined as:

  Kj
(i)
L z G (i)
j (t) = p ju (t) · z g ju (5)
u=1

where G (i) j (t) is the stochastic output performance process of the single machine,
p ju (t) is a probability that the process is in state u at instant t ≥ 0 for the given
previous initial states probability distribution p j (0), g(i)
ju is a performance in state u,
and z in general case is a complex variable. The definition and proof of Lz-transform
can be found in [19–21].
The Lz-transform of all processing machines, operated in parallel, can be obtained
by using the Universal Generating Operator (UGO)  f par [20, 21].

       KM
(i)
L z G (i)
M (t) =  f par
L z G (i)
1 (t) , . . . , L z G (i)
H (t) = PMU (t) · z g MU (6)
U =1

where G (i)
M (t) is the stochastic output performance process of the parallel operate
processing machines, K M is the possible states of the process, PMU (t) is a probability
that the process is in state U, U = 1, . . . , K M at instant t ≥ 0, g(i)
MU is a performance
in state U .
The Lz-transform of the whole (i)th FMC obtains by using the serial Universal
Generating Operator (UGO)  fser since the group of processing machines and robot
work in series
10 W. Zeng et al.

       
K (i)
(i) (i)
L z G (i) (t) =  fser L z G M (t) , L z G R (t) = PV(i) (t) · z g(i)V (7)
V =1

where G (i) (t) is the stochastic output performance process of the whole F MC (i) ,
G (i)
R (t) is the output performance process of the robot of the FMC, K (i) is the possible
states of the process, PV(i) (t) is a probability that the process is in state V, V =
1, . . . , K (i) at instant t ≥ 0, g(i)
V is a performance in state V .
The dynamic production rate of the (i)th FMC at instant t ≥ 0 can be calculated
by Eq. (8).


K (i)
P R (i) (t) = g(i) (i)
V · PV (t), i = 1, . . . , m (8)
V =1

Therefore, the P R (i) (t) results are utilized to calculate the dynamic probability
of the number of blanks in B (i) . Then, the equivalent probabilities of FMCs and the
equivalent FMS shown in Fig. 4 are obtained.

3.2 Step 2: Establishment of the Dynamic Availability

of the Equivalent FMS

Step 1 incorporates the influence of the intermediate buffers on the FMS into the
FMCs to generate the equivalent FMCs that are neither starvation nor blockage, and
an equivalent FMS that is neither starvation nor blockage is also generated. Moreover,
due to the existence of the intermediate buffers, the rigid connection of the production
system is transformed into a flexible connection [31]. When a station in the flexible
connection production line is shut down, the upstream and downstream stations can
continue processing because of the buffer for storing and supplying blanks, which
prevents the production line from being forced to shut down due to station shutdown.
From the perspective of reliability, the production line is transformed from a reliable
single-circuit series structure to a parallel redundant structure [32]. Therefore, the
dynamic availability A F M S (t) of the equivalent FMS is:
m 
 
∼
A F M S (t) = 1− A F M S (t)= 1 − 1 − A(i)
F MC (t) (9)
i=1

∼
where A F M S (t) is the unavailability of the equivalent FMS, A(i)
F MC (t) is the avail-
ability of the equivalent FMC, where A(i) (i)
F MC (t) = p(i−1)0 (t)A F MC (t) pib (t) and
B B
i

A(i)
F MC (t) is the availability of F MC
(i)
that excludes the impact of buffers and it
required to processes blanks more than production demands, otherwise it should be
Dynamic Availability Analysis for the Flexible Manufacturing … 11

shut down and repaired immediately. Therefore, A(i)

F MC (t) calculated as follows [20,
21]:

A(i)
F MC (t) = PV(i) (t) (10)
g(i)
V ≥d

where d is the production demands that could be constant or stochastic. Thus, the
dynamic performance measurements of the FMS can be calculated based on the
Eqs. (9) and (10), and the specific calculation index and approach will be presented
within the numerical example.

4 Numerical Example

This section introduces an exploratory numerical example to demonstrate the

proposed approach in detail, and the schematic of flexible manufacturing system
as depicted in Fig. 6. It is assumed that the blanks transportation process is failure-
free and all processing machine and robots are three-states and binary-state systems,
which is shown in Fig. 7, respectively. In particular, minor and major failures are
included in the machines’ state-transition process as well as aging characteristics that
failure rates follow the two-parameter Weibull distribution, while the repair rates of

Fig. 6 Schematic of the FMS

Fig. 7 State-transition process of machines (a) and robots (b)

12 W. Zeng et al.

machines obey exponential distribution. The failure rates and repair rates of robots
both follows exponential distribution. Note that the loading and unloading times of
robots are neglected compared to the processing times.
The parameters of each device are given in Table 1. Shape parameters β (i) and scale
parameters α (i) present the Weibull process of machines’ failures. The reciprocal of
the Mean time to repair (MTTR-1) and mean time to failure (MTTF) are given to
describe the exponential distribution. In addition, two constant production demands,
d1 = 12 units/h and d2 = 24 units/h are included to depict more about the impacts
of buffers on the FMS.
Therefore, taking constrain d1 as the detailed example, the first step to evaluate
the dynamic performance metrics of the FMS is to capture the dynamic production
rates of FMCs. Taking the F MC (2) as example, the differential system of the state
probabilities of machine M (2) j , j = 1, 2, 3, 4 obtained based on the Eq. (3) and
Fig. 7a. For simplicity, cancel the subscript indicating the FMC in the following
calculation.
⎧ dp1 (t)
⎨ = p3 (t)λ3,1 (t) + p2 (t)λ2,1 (t) − p1 (t)μ1,3 (t)
dt
dp2 (t)  
= p3 (t)λ3,2 (t) − p2 (t) λ2,1 (t) + μ2,3 (t) (11)
⎩ dp3 (t) dt  
dt
= p 1 (t)μ1,3 (t) + p 2 (t)μ2,3 (t) − p 3 (t) λ 3,1 (t) + λ 3,2 (t)

Initial conditions are: p1 (0) = 0, p2 (0) = 0, p3 (0) = 1.

Similarly, the differential system of state probabilities of the second robot R (2)
obtained based on the Eq. (3) and Fig. 7b.

d P0R (t)
dt
= P1R (t)λ1,0
R
(t) − P0R (t)μ0,1
R
(t)
d P1R (t) (12)
dt
= P0 (t)μ0,1 (t) − P1 (t)λ1,0
R R R R
(t)

Initial conditions are: P0R (0) = 0, P1R (0) = 1.

Then the Lz-transform of machine M j , j = 1, 2, 3, 4 and robot in the F MC (2)
obtained based on Eq. (5) and results of system (11) and (12).
  
L z G j (t) = p3 (t)z g3 + p2 (t)z g2 + p1 (t)z 0
R (13)
L z {G R (t)} = P1R (t)z g1 + P0R (t)z 0

The Lz-transform of the whole F MC (2) obtained through Eqs. (6) and (7), and
the dynamic production rate of F MC (2) are calculated through Eq. (8). The result
about F MC (2) is depicted in Fig. 8.
In the mimic procedures, the productivities of the remaining FMCs can be
evaluated as shown in Fig. 9.
Determining the dynamic productivities of FMCs, the next task is to calculate the
dynamic models of the equivalent FMCs. Based on the state transition intensity matrix
a(i) and the Chapman-Kolmogorov differential equations, the first state probability
of equivalent FMCs, i.e. the equivalent availability of F MC (1) is:
Table 1 Parameters of each device
Device Failure rates/(α (i) , β (i) ) Repair rates Performance Device Failure rates Repair rates Performance
(MTTR−1 /h) (Units/h) (MTTF/h) (MTTR−1 /h) (Units/h)
λ3,2 (t) λ3,1 (t) λ2,1 (t) μ2,3 μ1,3 g3 g2 λ1,0 μ0,1 g1
M (1) [3200, 2.2] [3500, 2.3] [2100, 1.9] 12 24 12 6 R(1) 2500 10 30
M (2) [1800, 1.8] [2300, 1.9] [1200, 1.6] 30 60 6 3 R(2) 2200 20 30
M (3) [2500, 2.7] [3100, 2.4] [1500, 1.8] 25 50 8 4 R(3) 2800 15 30
M (4) [2800, 2.4] [3400, 2.0] [1400, 1.5] 15 30 12 6 R(4) 3500 10 30
M (5) [2600, 3.2] [2900, 3.3] [1700, 2.3] 10 20 24 12 4000 8 30
Dynamic Availability Analysis for the Flexible Manufacturing …

R(5)
13
14 W. Zeng et al.

Fig. 8 Dynamic production rate of F MC (2) under constrain d1

Fig. 9 Dynamic production rates of F MC (1) , F MC (3) , F MC (4) and F MC (5) under constrain d1

A(1) (1)
F MC (t) = A F MC (t) p1b1 (t)
B
(14)

where A(1) (1)

F MC (t) is the equivalent availability of F MC , p1b1 (t) is the dynamic prob-
B

ability of B (i) non-blockage. Then, as the volume of b1 changes, the corresponding

shifts of A(1)
F MC (t) is depicted in Fig. 10.
Dynamic Availability Analysis for the Flexible Manufacturing … 15

Fig. 10 The equivalent availability of F MC (1) with the change of volume b1

The results in the figure shows that as b1 increases, the equivalent availability
of F MC (1) increases but gradually reaches the plateau. Note that, the availability
is chosen as the criterion for determining the appropriate buffer volume in current
paper, while other criterion in terms of system reliability associated cost (RAC)
will be discussed in further research, where the optimal volume of buffer could be
determined. Therefore, b1 = 40 is selected as an appropriate volume for buffer B (1) .
Consequently, the changes of equivalent availability of F MC (2) with the set
volume of b1 and variational b2 could also be obtained in the mimic procedure,
which shown in Fig. 11.
Therefore, an appropriate volume for buffer B (2) also confirmed as b2 = 30.
Similarly, the equivalent availability of F MC (3) , F MC (4) and F MC (5) with the
changes of volume of b3 and b4 depicted in Fig. 12, respectively. In addition, a group
of appropriate FMS intermediate buffers’ volumes can be obtained, as shown in Table
2.
In accordance with the aforementioned analysis results and the Eq. (9), the
dynamic availability of the FMS under production demands d1 and d2 plotted as
follow.
It can be concluded from Fig. 13 that the results in Table 2 could ensure the
FMS possess high availability under d1 while perform poorly under d2 , which entails
the production capacity requirements have a significant impact on the FMS avail-
ability. The approach proposed in current paper not only quantifies this impact, but
more importantly, provides a new solution for evaluating and improving the dynamic
availability of FMS during the life cycle. Nevertheless, it is inadequate to determine
the optimal buffers combination barely from the perspective of reliability because
production cost is an inevitable topic for the design, operation and optimization of
16 W. Zeng et al.

Fig. 11 The equivalent availability of F MC (2) with the volume of b1 = 40 and the variational b2

Fig. 12 The equivalent availability of F MC (3) , F MC (4) and F MC (5) with the changes of volume
of b3 and b4 , respectively

Table 2 An appropriate volume of each intermediate buffer

Buffer B (1) B (2) B (3) B (4)
Volume 40 30 30 40
Dynamic Availability Analysis for the Flexible Manufacturing … 17

Fig. 13 The availability of the FMS under d1 and d2

FMS. Therefore, it will be an interesting research to incorporate cost factors with the
proposed model for obtains a veritable optimal buffers combination.

5 Conclusion

The flexible manufacturing system consists of certain flexible manufacturing cells

and finite intermediate buffers considered in this paper. A two-step stochastic model
based on the Lz-transform method proposed to evaluate the dynamic availability of
the FMS and to determine the appropriate intermediate buffers volumes that realize
the system possess high availability throughout its life cycle under the premise of
meeting various production demands.
On one hand, Figs. 10, 11 and 12 manifest that the equivalent availability of
FMCs increases as the buffer’s volumes increase while gradually reaches the plateau,
which beneficial to select the appropriate intermediate buffer volume. Moreover,
the results in Fig. 13 also illustrate the efficiency and rationality of the proposed
method. On the other hand, the buffer optimization issue is to ensure and improve
manufacturing system productivity and efficiency during the design and operation
phases, then the system associated cost play a pivotal role in the decision-making
process. Therefore, it will have important theoretical and practical significance to
incorporate the cost factors with the proposed model to investigate the optimal buffers
arrangement systematically and exhaustive.
18 W. Zeng et al.

References

1. Alhourani F (2016) Cellular manufacturing system design considering machines reliability and
parts alternative process routings. Int J Prod Res 54(3):846–863
2. Anupma Y, Jayswal SC (2019) Evaluation of batching and layout on the performance of flexible
manufacturing system. Int J Adv Manuf Technol 101:1435–1449
3. Browne J, Dubois D, Rathmill K et al (1984) Classification of flexible manufacturing systems.
FMS Mag 114–117
4. Chen KS, Yu CM, Hus TH et al (2019) A model for evaluating the performance of the bearing
manufacturing process. Appl Sci 9(15):3105
5. Das K, Lashkari R, Sengupta S (2007) Reliability consideration in the design and analysis of
cellular manufacturing systems. Int J Prod Econ 105(1):243–262
6. Dosdogru AT, Gocken M, Geyik F (2015) Integration of genetic algorithm and Monte Carlo to
analyze the effect of routing flexibility. Int J Adv Manuf Technol 81:1379–1389
7. Duan JG, Xie N, Li LH (2019) Modelling and evaluation of multi-state reliability of repairable
non-series manufacturing system with finite buffers. Adv Mech Eng 11(6):1–13
8. Elleuch M, Bacha HB, Masmoudi F et al (2008) Analysis of cellular manufacturing systems
in the presence of machine breakdowns. J Manuf Technol Manag 19(2):235–252
9. Groover MP (2007) Automated assembly system, automation, production systems, and
computer-integrated manufacturing. Prentice Hall Press, Upper Saddle River, NJ
10. Gyulai D, Pfeiffer A, Monostori L (2017) robust production planning and control for multi-stage
systems with flexible final assembly lines. Int J Prod Res 55(13):3657–3673
11. Halse LL, Jæger B (2019) Operationalizing industry 4.0: understanding barriers of industry
4.0 and circular economy. In: Ameri F, Stecke K, von Cieminski G, Kiritsis D (eds) Advances
in production management systems. Towards smart production management systems. APMS
2019. IFIP advances in information and communication technology, vol 567. Springer, Cham
12. He C, Zhang SY, Qiu LM et al (2019) Assembly tolerance design based on skin model shapes
considering processing feature degradation. Appl Sci 9(16):3216
13. Jain V, Raj T (2016) Modeling and analysis of FMS performance variables by ISM, SEM and
GTMA approach. Int J Prod Econ 171(1):84–96
14. Jin R, Liu K (2013) Multimode variation modeling and process monitoring for serial-parallel
multistage manufacturing processes. IIE Trans 45(6):617–629
15. Koulamas CP (1992) A stochastic model for a machining cell with tool failure and tool
replacement considerations. Comput Oper Res 19(8):717–729
16. Kumar N, Kumar J (2019) Efficiency 4.0 for industry 4.0. Human Technol 15(1):55–78
17. Lee J, Bagheri B, Kao HA (2015) A cyber-physical systems architecture for industry 4.0 based
manufacturing systems. Manuf Let 3:18–23
18. Li JS, Huang NJ (2007) Quality evaluation in flexible manufacturing systems: a markovian
approach. Math Prob Eng. Article ID 057128, 24 pages. https://doi.org/10.1155/2007/57128.
19. Lisnianski A (2012) Lz-transform for a discrete-state continuous-time Markov process and its
application to multi-state system reliability. In: Lisnianski A, Frenkel I (eds) Recent advances
in system reliability. Springer-Verlag, London, pp 79–95
20. Lisnianski A, Frenkel I, Ding Y (2010) Multi-state system reliability analysis and optimization
for engineers and industrial managers. Springer, London
21. Lisnianski A, Frenkel I, Khvatskin L (2021) Modern dynamic reliability analysis for multi-state
systems. Springer series in reliability engineering. Springer, Cham
22. Liu JL, Yang S, Wu AG et al (2012) Multi-state throughput analysis of a two-stage manu-
facturing system with parallel unreliable machines and a finite buffer. Eur J Oper Res
219(2):296–304
23. Loganathan MK, Girish K, Gandhi OP (2016) Availability evaluation of manufacturing systems
using Semi-Markov model. Int J Comput Integ M 29(7):720–735
24. MacDougall W (2014) Industry 4.0: smart manufacturing for the future. Berlin, Germany,
GTAI
Dynamic Availability Analysis for the Flexible Manufacturing … 19

25. Manocher D, Hamid S (2019) Analysis of critical machine reliability in manufacturing cells.
J Ind Eng Manag 12(1):70–82
26. Rybicka J, Tiwari A, Enticott S (2016) Testing a flexible manufacturing system facility produc-
tion capacity through discrete event simulation: automotive case study. Int J Mech Aerosp Ind
Mechatron Manuf Eng 10(4):668–672
27. Sanghavi D, Parikh S, Raj SA (2019) Industry 4.0: tools and implementation. Manag Prod Eng
Rev 10(3):3–13
28. Savsar M (2000) Reliability analysis of a flexible manufacturing cell. Reliab Eng Syst Saf
67(2):147–152
29. Savsar M (2004) Performance analysis of an FMS operating under different failure rates and
maintenance policies. Int J Flex Manuf Sys 16:229–249
30. Savsar M (2011) Multi-state reliability modeling of a manufacturing cell. Int J Perform Eng
7(3):217–228
31. Shu S (1992) An analysis of the repairable computer integrated manufacturing system (CIMS)
with buffers and a study of the system reliability. Acta Automatica Sinica 18(1):15–22
32. Shu S, Zhang Y (1995) Reliability analysis of series production lines. Control Theory Appl
12(2):177–182
33. Singholi A, Ali M, Sharma C (2013) Evaluating the effect of machine and routing flexibility
on flexible manufacturing system performance. Int J Serv Oper Manag 16(2):240–261
34. Tan B, Gershwin SB (2009) Analysis of a general Markovian two-stage continuous-flow
production system with a finite buffer. Int J Prod Econ 120(2):327–339
35. Thames L, Schaefer D (2016) Software-defined cloud manufacturing for industry 4.0. Procedia
CIRP 52:12–17
36. Trivedi K (2019) Probability and statistics with reliability, queuing and computer science
applications. Wiley, New York
37. Zeng W, Chen P (2008) Volatility smile, relative deviation and trading strategies: a general
diffusion model for stock price movements based on nonlinear birth-death process. China
Econ Quart 7(4):1415–1436
38. Zeng W, Shen G, Chen B et al (2019) Lz-transform method and Markov reward approach for
flexible manufacturing system performance evaluation. Appl Sci 9(19):4153
39. Zhou J, Liu Z (2006) Relationship between machine utilization and buffer capacity. Tool Eng
40(9):24–26
Integrated Reliability and Risk
Assessments of Nuclear Facilities

Vanderley de Vasconcelos, Graiciany de Paula Barros,

Antônio Carlos Lopes da Costa, and Wellington Antonio Soares

Abstract Reliability is the probability that a system will perform its intended func-
tion satisfactorily during its lifetime under specified environmental and operating
conditions. Risk can be measured by assessing the probability of an undesired event
(e.g., a system failure) and the magnitude of its consequences. Therefore, risk and
reliability are complementary variables. Licensing of nuclear facilities requires safety
and risk assessment. Probability risk assessment implies carrying out a quantitative
assessment of the reliability of items important to safety (IISs). In addition, reliability
assessments are required during all lifetime phases for optimizing plant performance,
maintainability and safety. Thus, the outcomes of reliability and risk assessments are
interchangeable and complementary. This chapter proposes a framework for inte-
grating reliability and risk assessments of IISs of nuclear facilities using tools of
reliability engineering, as Markov models, fault tree analysis, event tree analysis,
reliability block diagrams and life data analysis. Based on frequency-dose limits to
the public, risk acceptance criteria are also suggested in the scope of the frame-
work. A case study demonstrating the advantages of using the integrated approach
applied to a preliminary plant design phase of a nuclear fuel fabrication, to meet risk
acceptance criteria and licensing requirements, is presented.

1 Introduction

Reliability is usually defined as the probability that an item, component, or engi-

neering system will perform its intended function without failing during a given

V. de Vasconcelos (B) · G. de Paula Barros · A. C. L. da Costa · W. A. Soares

Centro de Desenvolvimento da Tecnologia Nuclear - CDTN, Av. Presidente Antônio Carlos,
6.627 Campus da UFMG, Pampulha, Belo Horizonte, Minas Gerais 31270-901, Brazil
e-mail: vasconv@cdtn.br
G. de Paula Barros
e-mail: graiciany.barros@cdtn.br
A. C. L. da Costa
e-mail: aclc@cdtn.br

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 21

M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_2
22 V. de Vasconcelos et al.

period of time under operation conditions. Among several risk concepts used in
different scientific areas, risk can be defined as the potential of loss resulting from
exposure to hazards. Sometimes, the risk is measured by assessing the probability
(or frequency) of occurrence of an undesired event (e.g., an accident) and the mag-
nitude of its consequences (severity) [1]. Under this viewpoint, risk and reliability
can be considered as complementary variables. Thus, even if these analyses were
not performed at the same time during plant lifetime, or by the same team, their
outcomes should be interchangeable and complementary. This chapter proposes a
framework for integrating reliability and risk assessments applied to the design of
items important to safety (IISs) of nuclear facilities (NFs). This kind of integration
is a trend in advanced reactor designs within reliability assurance programs (RAPs),
which exploit the optimization of performance and safety of such plants using well-
established reliability, availability and maintainability (RAM) and probabilistic risk
assessment (PRA) tools [2]. The application of this framework for nuclear facilities,
which are simpler and less potentially hazardous than nuclear power plants (NPPs),
can also bring benefits during all plant life phases, supporting design, operation and
licensing processes.
Compared to NPPs, NFs differ in many aspects. These last facilities use a wide
variety of technologies and processes and, in addition to fissile material and waste,
they often handle, process and storage large quantities of toxic, corrosive, or com-
bustible chemicals. Furthermore, a common characteristic of these facilities is the
continuous improvement and innovations, research and development, as well as
greater dependence on human factors, which range the potential hazards and the
possibility of release of hazardous chemical or radioactive materials. Despite that,
the methodologies used for reliability and risk assessments of NPPs can be adapted
and integrated for use during all NF lifetime, in order to optimize the safety and
performance of such facilities [3].
An overview of an integrated approach for reliability and risk assessments of NFs
is shown in Fig. 1. Supported by reliability engineering tools and risk acceptance
criteria, the IISs are analyzed in an integrated way throughout the facility lifetime. The
optimization of plant performance and safety is carried out using well-established
techniques of PRA and reliability engineering, such as life data analysis (LDA),
reliability block diagrams (RBD), failure mode and effects analysis (FMEA), event
tree analysis (ETA), fault tree analysis (FTA), Markov models and Monte Carlo
simulation, among others. Special emphasis should be given to aspects of human
reliability analysis (HRA), sensitivity analysis of IISs and uncertainty assessment of
reliability and risk quantitative results.

2 Nuclear Facilities Under the Scope of Analysis

Nuclear-based electric power depends on heat generation within the reactor core,
which is usually an assembly of uranium fuel rods (nuclear fuel), the uranium some-
times being enriched in the naturally present uranium-235 isotope (fissile atom). The
Integrated Reliability and Risk Assessments of Nuclear Facilities 23

Fig. 1 Overview of integrated approach for reliability and risk assessments of nuclear facilities

heat results from a sustained and carefully controlled “chain reactions”, in which
neutrons cause fissions of fissile atoms within the core. The fission process involves
the splitting of the fissile atoms into lighter radioactive fission products, the produc-
tion of further neutrons, sustaining the chain reactions and releasing a considerable
amount of energy [4].
Spent nuclear fuel can be reprocessed to extract recyclable energy material, inside
the so-called “nuclear fuel cycle”. This cycle involves all industrial activities neces-
sary to produce electricity from uranium in NPPs, including “front-end” activities
(fuel fabrication), “service period” activities (electricity generation in NPPs) and
“back-end” activities (reprocessing, reuse and disposal). Fuel cycle is referred to as
“closed fuel cycle” if the spent fuel is reprocessed; otherwise, it is referred to as
“open fuel cycle”. Therefore, nuclear fuel cycle facilities (NFCFs) can include the
following industrial operations:
• Processing of uranium ores;
• Uranium conversion and enrichment;
• Nuclear fuel fabrication;
• Reprocessing of spent nuclear fuel;
• Nuclear waste management and disposal;
• Separation of radionuclides from fissile products;
• Research and development to support NFCFs.
NFCFs are installations other than NPPs, research reactors and critical assem-
blies, which pose potential hazards to workers, public and environment [5]. The
terms “nuclear fuel cycle facility” and “nuclear facility” and their initials are used
interchangeably in this chapter. The variety of processes associated with NFCFs
can result in a broad range of potential accidents. Some types of facilities with low
24 V. de Vasconcelos et al.

radioactive inventories and low criticality risk, or facilities handling natural uranium,
have a low hazard potential. Other NFCFs (e.g., reprocessing facilities) may have a
potential risk comparable with NPPs.
The following specific features need to be considered when performing a safety
assessment of NFCFs [6]:
• The nature and variety of technologies result in a broad range of hazardous con-
ditions;
• Presence of a wide diversity of radioactive, fissile, toxic, combustible, explosive
or reactive materials;
• Radioactive and other hazardous materials are present and handled throughout
different parts of the facility;
• The possibility of occurrence of nuclear criticality accidents;
• Need for frequent changes in equipment, processes, and operating procedures can
pose additional hazards;
• Operations of NFCFs usually require more operator actions than NPPs, making
operators more vulnerable;
• Many of NFCFs worldwide are obsolete and lack adequate design documentation
and information on operating experience, which require additional efforts and
specific approaches to safety assessment.
The loss of the main safety functions of NFCFs may lead to release of radiological
or hazardous materials. The most common safety functions required by NFCFs are
[7]:
• Prevention of criticality of fissile materials;
• Radiological protection;
• Confinement of radioactive or hazardous materials to prevent or mitigate unplanned
releases to the environment;
• Removal of decay heat.
This chapter presents a simplified or graded version of the complete probabilistic
risk assessment (PRA) traditionally used in NPPs, which is compatible with the
complexity and level of hazard presented by NFs. PRA is useful to complement the
conventional deterministic safety assessment (DSA), whereby plant safety can be
assessed and improved.
PRA provides not only a comprehensive and structured approach to identify fail-
ure scenarios and facility damages but also for estimating numerical values for the
risks. Likewise, it offers a systematic approach to verify whether the reliability and
independence of safety systems are adequate for implementing defence-in-depth
provisions and assessing whether the risks are “as low as reasonably achievable”
(ALARA criterion). It is a trend currently in such analyses to make a less conserva-
tive assumption and to consider the best-estimated values [6].
Both PRA applications to NPPs and NFCFs are based on common principles.
Nevertheless, there are some significant differences, mainly due to plant design and
processes. Compared to NPPs, NFCFs are usually characterized by a larger diversity
of technologies and processes. Apart from processing radioactive or fissile materials,
Integrated Reliability and Risk Assessments of Nuclear Facilities 25

Fig. 2 Integration of reliability and risk assessments during facility lifetime and licensing process

larger quantities of chemical materials are handled frequently. Consequently, these

materials must be taken into account appropriately in PRAs of NFCFs.

3 Reliability and Risk Assessments During Facility Lifetime

In relation to NFs, risk assessment is defined as “an assessment of the radiation risks
and other risks associated with normal operation and possible accidents involving
facilities and activities. This will normally include consequence assessment, together
with some assessment of the probability of those consequences arising.” Throughout
this chapter, safety assessment is used with the concept of “an assessment of all
aspects of a practice that are relevant to protection and safety”, a scope that encompass
risk assessment [5].
The integration of reliability and risk assessments of items important to safety
(IISs) can be carried out at all phases of the facility lifetime. At the beginning of a
lifetime, these assessments are almost independent of each other, although they can be
done together, at the same time, and by the same team in order to optimize resources.
Throughout the facility lifetime, this integration becomes even more necessary, not
only for cost-effectiveness reasons but also mainly for improving safety and quality
assurance of the outcomes [2, 8]. Figure 2 shows the integration of reliability and risk
assessments during facility lifetime and licensing process. The arrow width illustrates
the need or potential of integration of these assessments from initial phases of a
lifetime or licensing process to decommissioning of the facility.
26 V. de Vasconcelos et al.

Looking for the details of this integration, a table was built, listing the safety and
risk assessments typically required in the licensing process of NFs, as well as the
Reliability, Availability and Maintainability (RAM) assessments required by design,
operation and performance issues. Table 1 shows, for each phase in the lifetime of
an NF, some examples of typical documents and authorizations required during the
licensing process, relating them to safety, risk and RAM assessments.
Before the start of the licensing process itself (pre-licensing) in the phase of the
feasibility study and preliminary design, there is still no formal licensing process,
but some decisions are taken about reference facilities, strategic plan for licensing,
and acceptable levels of risk, safety and reliability. At this point, a definition and
categorization of a list of IISs, preliminary deterministic and probabilistic safety
assessments, a qualitative analysis of hazard and accident scenarios, as well as a
selection of design principles applied to safety are carried out. On the other hand,
reliability characteristics and functional requirements of IISs, as well as automation
and control philosophies are defined.
During siting and site evaluation phase, regulatory bodies usually require some
licensing documents, such as site evaluation reports, environmental impact assess-
ments, emergency preparedness and response plan. For supporting public inquiries,
identification of potential vulnerabilities and areas of importance to risk, demonstra-
tion of facility safety taking into account specific site characteristics and long-term
safety assessment are carried out. At the same time, the project team proceeds with
the availability and performance assessments taking into account specific site char-
acteristics.
Both deterministic and probabilistic safety assessments should be completed
before initiating the detailed design process, looking for any necessary changes
in the basic design philosophy, because future changes could become excessively
expensive. These assessments require technical specification of IISs, as well as defi-
nitions about maintenance, inspections, tests, requirements, man-machine interface
and human reliability, among others.
As the detailed design evolves to the final design, the importance of an integrated
assessment of safety, risk, maintainability and availability increase, not only for
economic reasons, but also for quality assurance requirements, which imposes high
reliability and safety and lower risks. In addition to technical specifications and PRA,
the elaboration of safety-related documents as physical protection and fire protection
plans are more effective when performed through integrated activities, using the same
tools as FTA, ETA and RBD, for example. Considerations about influence of human
factors on facility design and operation are also analyzed in this phase.
For granting a construction permit from a regulatory body, a preliminary safety
analysis report (PSAR) is one of the required documents. Therefore, an assessment of
the potential impact of construction on neighboring and demonstration of compliance
of IISs with safety requirements will need life data analysis and supplier assessment,
as well as the elaboration of pre-service inspection strategy. Integration of safety
assessment and procurement process is carried out to specify reliability standards
and purchase the most reliable components available.
Integrated Reliability and Risk Assessments of Nuclear Facilities 27

Table 1 Examples of typical safety, risk, reliability, availability and maintainability assessments
during facility lifetime and licensing process of nuclear facilities (NFs)
Facility lifetime Safety and risk RAM assessments Licensing process
phases assessments
1. Feasibility study Definition and Definition of Definition of reference
and preliminary design categorization of a list reliability facilities
of IISs characteristics of IISs Acceptable levels of
Preliminary Definition of risk, safety and
deterministic and automation and reliability
probabilistic safety control philosophies
assessments
2. Siting and site Vulnerabilities and Demonstration of Site approval
evaluation areas of importance to facility availability Site evaluation report
risk taking into account Emergency plan
Demonstration of specific site
facility safety characteristics
3. Design Deterministic safety RAM assessment of Technical
assessment (DSA) IISs specifications
Probabilistic risk Maintenance, Fire protection plan
assessments (PRA) inspection and test Technical design
Specifications for requirements documents
safety functions Human reliability
analysis (HRA)
4. Manufacturing and Assessment of Life data analysis Construction permit
construction potential impact of Suppliers assessment Preliminary safety
construction on analysis report
neighboring (PSAR)
5. Commissioning As-built safety Assessment of Authorization for use
assessment maintenance, of nuclear material
Safety assessment of in-service inspection Quality assurance
commissioning tests and testing strategies program
Analysis of risks and Unconformities
opportunities analysis
6. Operation Data specialization of Data specialization of Authorization for
failure rate of IISs reliability operation
Living PRA characteristics of IISs As-built final safety
Verification of Analyses of facility analysis report
operator fitness for performance using (FSAR)
duties related to safetyRBI and RCM Operating procedures
HRA for facility
operation
7. Technical Safety review of risks Evaluation of Approval of technical
modifications and and hazards reliability modifications
innovations Assessing of safety characteristics of IISs Periodic safety review
factors against current Evaluation of facility (PSR)
standards performance
8. Ageing Categorizations of IISs Understanding of Ageing management
management to ageing management ageing mechanisms plan and reports
Prioritization of Prioritization of
actions for safety corrective actions for
improvement RAM improvements
(continued)
28 V. de Vasconcelos et al.

Table 1 (continued)
Facility lifetime Safety and risk RAM assessments Licensing process
phases assessments
9. Decommissioning Safety assessment of Assessment of Authorization for
decommissioning progressive shutdown decommissioning
Assessment of safety and new Decommissioning
of workers, public and configurations of IISs plan and reports
environment

The authorization for use of nuclear material and commissioning require, among
others, the assessment of as-built safety, measures for controlling nuclear materials,
and commissioning tests. These assessments complement or require the definition of
in-service inspection and testing strategies (e.g., risk-based inspection—RBI), and
maintenance strategies (e.g., reliability-centered maintenance—RCM). On the other
hand, a definition of a quality assurance program requires an integrated unconfor-
mities analysis, analysis actions to address risks and opportunities, and elaboration
of training and qualification plans, to meet the current requirements of the ISO 9001
standard.
For granting authorization for operation from a regulatory body, a final safety
analysis report (FSAR), taking into account data specialization of the failure rate
of IISs and operating experience, is one of the required documents. Sometimes, the
safety assessment is complemented by a living PRA, helping to ascertain the com-
pliance of NF plans and procedures with safety requirements (e.g., related to the
operation, radiation protection, fire protection, emergency preparedness and waste
management). During the operation, a fully continuous assessment of the effective-
ness of preventive and corrective actions, availability of IISs, facility performance,
use of RBI and RCM strategies, and verification of operator fitness for duties should
be integrated and carried out by the same team, if possible, for cost-effectiveness
issues.
During facility lifetime, many technical modifications and innovations are nec-
essary, which should be approved by a regulatory body, sometimes under periodic
safety review plans (PSRs). These technical modifications require a safety review
of risks and hazards, assessing safety factors against current standards, and reval-
uation of facility safety, reliability characteristics of IISs, and facility performance
after modifications. These safety, risk and RAM assessments are complementary and
would become expensive and with a high possibility of inconsistencies if not carried
out in an integrated way.
Integrated Reliability and Risk Assessments of Nuclear Facilities 29

Ageing management is a phase of facility lifetime, which also requires simulta-

neous risk, safety, and RAM assessments. To determine the best strategy for ageing
management of IISs and their impact on risk changes, it is necessary to understand
how ageing affects the reliability of individual IISs. Examples of these analyses
include categorizations of IISs to ageing management, safety review of IISs under
current conditions, prioritization of corrective actions for safety improvement, mit-
igation of ageing effects, understanding of ageing mechanisms, review of reliabil-
ity characteristics of IISs under current conditions, and prioritization of corrective
actions for RAM improvement.
At the end of the facility lifetime, it is also necessary an authorization for decom-
missioning from the regulatory body. On the one hand, it is necessary a safety
assessment of decommissioning (potential new hazards), a demonstration of safe
management of radioactive waste, an assessment of proposed end-sate, including
radiation doses, and assessment of environmental impact and safety of workers and
public. On the other hand, it is necessary an assessment of progressive shutdown and
new configurations of IISs, an assessment of RAM during decommissioning (e.g.,
reliability of hoisting devices to prevent accidents and environmental impact), and
an assessment of knowledge management effectiveness. These are fully integrated
safety and reliability assessments, which should be carried out by the same team, at
the same time and using the same tools and data.

4 Reliability of Items Important to Safety

Reliability assessments of NFs are required, among others, by quality assurance (QA)
programs and technical specifications of IISs [2]. Safety and risk assessments of NFs
are licensing requirements [8, 9]. Two complementary methods (deterministic and
probabilistic) are usually required to assess and improve the safety during the lifetime
of these facilities.

4.1 Deterministic and Probabilistic Safety Analysis

According to International Atomic Energy Agency—IAEA [5] deterministic anal-

ysis, in the context of safety of NFs, “implies focusing on accident types, releases
of radioactive materials and consequences, without considering the probabilities of
different event sequences”. Deterministic safety analysis—DSA is performed to pre-
dict the facility response to postulated initiating event (e.g., design basis accidents—
DBAs) and demonstrate the adequacy of design of engineered safety features or other
IISs, both to control and mitigate their consequences [6]. DSA is based on conserva-
tive assumptions made at all steps of quantification of accident sequences, showing
that the facility and its safety systems meet the safety requirements, ensuring that
the end-states in terms of consequences are acceptable [10].
30 V. de Vasconcelos et al.

Probabilistic risk assessment—PRA (sometimes called probabilistic safety

analysis—PSA) evaluates the probabilities of occurrence of the accident sequences
and their consequences. This type of evaluation is used to identify risks, mitigation
measures, and weaknesses in facility design that might dominate the risks. PRA
is used in a complementary way with DSA, helping the selection of new accident
sequences to be included in safety analysis [10].

4.2 Reliability Characteristics

There are many reliability characteristics that are important to facility safety and
performance assessments. Table 2 summarizes some of these characteristics, with
their respective definitions. The selected reliability characteristics were probability
of failure given time, B(X) life, mean life, reliability given time, availability given
time, failure rate, and maintainability. These variables are interrelated in some way
and their definitions are complemented by the Eqs. 1–5 presented next [11].
Assuming an exponential life distribution for an item i with a constant failure
rate, λi , the reliability of item i as a function of time, t, Ri (t), is given by Eq. 1:

Ri (t) = e−λi t . (1)

Table 2 Selected reliability characteristics important to facility safety and performance assessments
[11]
Reliability characteristics Definitions
Probability of failure given time The probability that an item will be failed at a particular
point in time. The probability of failure is also known as
“unreliability” and it is the reciprocal of the reliability
B(X) life The estimated time when the probability of failure will
reach a specified point (X%)
Mean life The average time that the items in the population are
expected to operate before failure. This is often referred to
as “mean time to failure” (M T T F)
Reliability given time The probability that an item will operate successfully at a
particular point in time and under specified operational
conditions
Availability given time Probability that an item is operational at a given time (i.e.,
has not failed or it has been restored after failure)
Failure rate The number of failures per unit time that can be expected to
occur for the item
Maintainability Probability that an item is successfully restored after failure.
“Mean time to repair” (M T T R) is often a reliability
characteristic used to measure maintainability
Integrated Reliability and Risk Assessments of Nuclear Facilities 31

Mean time to failure, M T T F, can be obtained by integrating the reliability function

of item i from zero to infinity, as given by Eq. 2:
 ∞
MT T F = Ri (t) dt. (2)
0

As the reliability of an item is the probability that it will operate successfully by a

given time, in dealing with repairable items, some other definitions are needed, as
described below.
Maintainability, Mi (t), is defined as the probability of performing a successful
repair action within a given time, t. Then, it measures the ease and speed a system
can be restored to its operational status after a failure occurs. Considering an item i
in which repair times are distributed exponentially, with a constant repair rate, μi ,
its maintainability is given by Eq. 3:

Mi (t) = 1 − e−μi t . (3)

Mean time to repair, M T T R, can be obtained by integrating maintainability function

from zero to infinity, as given by Eq. 4:
 ∞
MT T R = Mi (t) dt. (4)
0

Availability can be defined as the probability an item/system is operational at a given

time (i.e., it has not failed or it has been restored after failure). Then, availability for
repairable systems takes into account both reliability and maintainability. In other
words, availability, A(t), is the probability a system is operating properly when it is
requested for use, and in the case of a single item, i, is given by Eq. 5:

Item uptime MT T F
Ai (t) = = . (5)
Item uptime + Item downtime MT T F + MT T R

In the case of systems with items in series or parallel configurations, the reliability
characteristics are calculated using concepts of Boolean algebra and probability
theory. For more complex configurations, reliability engineering tools and computer
codes are usually required [11, 12].

4.3 Reliability Engineering Tools

There are many tools of reliability engineering, usually supported by computer codes,
developed to solve reliability related problems. Some of the most important reliability
engineering tools are next shortly described.
32 V. de Vasconcelos et al.

Reliability block diagram (RBD) is a graphic tool that represents a system as

connections of blocks (system items) in series or in parallel, according to their relia-
bility relationships. RBD analyzes the effect of item failures on system performance,
assuming that each item is in an operating or failed state, and associating each item a
failure probability or a reliability characteristic. Its results are used to reliability and
availability estimations [11].
Fault tree analysis (FTA) is also a graphic and logical technique that analyzes
the probability of occurrence of an undesired state of a system (top event) from the
probability of occurrence of its causes (basic events). The top events are selected
taken into account the items that are critical to safety or reliability. A fault tree can
be described as logical combinations of events (e.g., item failures or human errors),
through “OR” or “AND” gates, that will result in the top event. Fault trees can be
analyzed qualitatively or quantitatively. Qualitative analysis involves the identifica-
tion of the combinations of basic events that lead to top event (minimal cut sets).
Quantitative analysis involves the estimation of the probability of occurrence of top
events, when the probabilities of the basic events are known. These results are used
both to risk and reliability analyses [12, 13].
Reliability analysis and PRAs of IISs require life data (also called lifetime), as
time, distance or cycles of operation before they fail. Life data analysis (LDA) studies
and models use observed item lifetimes to estimate failure rates of IISs and fit statisti-
cal distributions to available data. The most common distributions used for modelling
reliability and failure behavior are Weibull, normal, lognormal and Gamma, among
others. Computer programs are required to select and fit the adequate distributions
and estimate important reliability characteristics and probability of failure at specific
times. Sometimes, accelerated life testing (ALT) is needed in order to get lifetimes
of items of high reliability [12].
Markov models can be required to estimate reliability characteristics and failure
probabilities (e.g., piping rupture rates) from available data, as inspection and repair
rate, flaw occurrence rate and leak rates. They use decision trees for modeling inter-
mediate states of an item before its failure. Markov models are appropriate whenever
the stochastic behavior of an item depends on the states of their components. These
models assume that the probability an item will perform a transition from a state to
another at any time depends on the initial and final state of the transition. In general,
Markov models are more adequate than FTA for modeling reliability and failure
probability of items as function of time [14].

4.4 Reliability, Availability and Maintainability (RAM)

of Items Important to Safety

According to [5], an item important to safety (IIS) is an item whose failure led to
radiation exposure of workers or public and includes structures, systems, and com-
ponents (SSCs), which prevent that foreseeable operational events result in accident
Integrated Reliability and Risk Assessments of Nuclear Facilities 33

Table 3 Examples of design principles and criteria applied to reliability, safety and risk [9]
Design focuses Principles and criteria
Reliability Standby and redundancy, physical separation, functional independence,
diversity, safety factors, maintainability, availability
Safety Fail-safe design, double contingency, single failure design, ALARA (as low
as reasonably achievable), defence-in-depth, passive safety, fault tolerance,
inherently safe, licensing requirements
Risk Prevention principle, precautionary principle, protection principle,
radioprotection, limitation of risks to individuals and environment, design
basis accidents (DBAs), risk-based inspections (RBI), IAEA safety principles

conditions, as well as devices necessary to mitigate the consequences of failures or

malfunctioning of SSCs.
The reliability of IISs shall be in accordance with their safety significance. The
design and technical specification of IISs shall be such as to ensure that they can
be qualified, installed, operated, and maintained to be capable of withstanding, with
sufficient reliability and effectiveness, all conditions specified in their design basis.
Table 3 shows some examples of design principles and criteria, which are com-
monly used in NFs in order to increase safety and reliability of IISs, as well as reduce
risks [9].
The design principles and criteria cited in Table 3 are grouped in three focuses:
reliability, safety, and risk. For example, the design of IISs shall take into account
common mode failures (CCFs), using concepts of redundancy, diversity, physical
separation, and functional independence, in order to achieve the intended reliabil-
ities. The improvement of maintainability, e.g., including self-checking capability,
periodic testing, or adopting reliability centered maintenance (RCM), increases the
reliability and availability of IISs. The concepts of fail-safe design, defence-in-depth,
and single failure design shall be incorporated into the design of IISs, so that even
if they fail the safety functions are performed. In the same way, prevention, precau-
tionary, and protection principles, as well as radioprotection or risk-based inspection
(RBI), shall be used, as appropriate, to prevent, reduce and mitigate risks to indi-
viduals and the environment. Reliability engineering tools for assessing the reliabil-
ity characteristics are widely used to verify and prove these principles and criteria
[2, 9].

4.5 Human Reliability

Human reliability analysis (HRA) studies and models the interactions between
humans and systems to estimate human error probabilities (HEPs), taking into
account work environment, opportunities to recover from errors and their conse-
quences. The most known HRA techniques are technique for human error rate pre-
34 V. de Vasconcelos et al.

diction (THERP), holistic decision tree (HDT), cognitive reliability and error anal-
ysis method (CREAM), human error assessment and reduction method (HEART),
and standardized plant analysis risk-human reliability analysis (SPAR-H). Many of
these techniques use the so-called performance-shaping factors (PSFs) to help HEP
prediction from specific work situations [15].
PSFs are defined as variables that may affect human performance in an HRA and
are used for adjusting basic human error probabilities (HEPs) assumed as nominal
under certain conditions. PSFs usually adopted in HRA methods include available
time, stress, complexity of systems and tasks, training, worker experience, opera-
tional procedures, human factors, and workload, among others.
Some HRA methods, such as SPAR-H, categorize human tasks into two basic
types: action or diagnosis. Routine operation, starting equipment, calibration or main-
tenance, and other activities of following procedures are examples of action tasks.
On the other hand, diagnosis tasks depend on knowledge/experience for analyzing
current facility conditions, planning and prioritizing activities, and take decisions
about alternative actions. SPAR-H assigns a figure of 10−2 to nominal H E Ps for
diagnosis tasks, which is ten times the nominal H E Ps for action tasks (10−3 ).
Some PSF levels cause the increase of H E P (negative influence), while other
PSF levels cause the decrease of H E P (positive influence). Figure 3 illustrates this
relationship for diagnosis and action tasks, including 5th and 95th percentiles, rep-
resenting the uncertainties involved in the estimates of H E Ps. A lower bound of
10−5 is suggested [15].

Fig. 3 Influence of PSF on

nominal error rate (based on
[15])
Integrated Reliability and Risk Assessments of Nuclear Facilities 35

SPAR-H treats mathematically the influence of PSF levels on H E P using specific

multipliers for diagnosis and action tasks (Table 4), according to Eq. 6:


8
HEP = NHEP × Si , (6)
i=1

where N H E P is the nominal H E P (10−2 for diagnosis and 10−3 for action), and
Si is the multiplier associated with the corresponding PSFi level (Table 4). As an
example, in an accident situation, when a diagnosis task (N H E P = 10−2 ) of high
complexity (S3 = 5) is carried out under extreme stress (S2 = 5), considering the
other six levels as nominal (S1 = S4 = S5 = S6 = S7 = S8 = 1), the H E P value,
according to Eq. 7, would be:


8
HEP = NHEP × Si = 10−2 × 1 × 5 × 5 × 1 × 1 × 1 × 1 × 1 = 0.25. (7)
i=1

In this theoretical example, a diagnosis task that under normal conditions would have
a human error probability of 1% becomes 25% under accident conditions!
When three or more negative PSF influences are present (i.e., anytime a multiplier
greater than 1 is selected), when applying Eq. 6, SPAR-H method recommends the
use of an adjustment factor, according to Eq. 8 for computing H E P [16]. This helps
to reduce double counting interdependencies between PSF effects and avoids H E P
value from being greater than 1.
8
N H E P × i=1 Si
HEP = 8 . (8)
N H E P × ( i=1 Si − 1) + 1

5 Probabilistic Risk Assessment (PRA)

Mathematically, the risk from an undesired event can be expressed as a product of

the frequency of occurrence and severity, as shown in Eq. 9 [17]:
 consequence     consequence 
event
Risk = frequency × severity (9)
time time event

Risk assessment is the combination and integration of the probabilities (or fre-
quencies) and the severities (or consequences) for identified hazards, taking into
account the effectiveness of controls and barriers (defence-in-depth levels). It pro-
vides an input to risk evaluation and decisions about risk management, through the
adoption of adequate risk acceptance criteria [18].
36 V. de Vasconcelos et al.

Table 4 Evaluation of PSFs for diagnosis and action portions of tasks according to SPAR-H method
[15]
PSFs PSF levels Diagnosis multiplier Action multiplier
1. Available time Inadequate time P (Failure) = 1.0 P (Failure) = 1.0
Barely adequate time 10 10
(≈2/3 × nominal)
Nominal time 1 1
Extra time (between 1 0.1 0.1
and 2 × nominal and >
than 30 min)
Expansive time (>2 × 0.01 0.01
nominal and >30 min)
Insufficient 1 1
information
2. Stress/stressors Extreme 5 5
High 2 2
Nominal 1 1
Insufficient 1 1
information
3. Complexity Highly complex 5 5
Moderately complex 2 2
Nominal 1 1
Obvious diagnosis 0.1 Not applicable
Insufficient 1 1
information
4. Experience/training Low 10 3
Nominal 1 1
High 0.5 0.5
Insufficient 1 1
information
5. Procedures Not available 50 50
Incomplete 20 20
Available, but poor 5 5
Nominal 1 1
Diagnostic/symptom 0.5 Not applicable
oriented
Insufficient 1 1
information
6. Ergonomics/HMI Missing/misleading 50 50
Poor 10 10
Nominal 1 1
Good 0.5 0.5
Insufficient 1 1
information
(continued)
Integrated Reliability and Risk Assessments of Nuclear Facilities 37

Table 4 (continued)
PSFs PSF levels Diagnosis multiplier Action multiplier
7. Fitness for duty Unfit P (failure) = 1.0 P(failure) = 1.0
Degraded fitness 5 5
Nominal 1 1
Insufficient 1 1
information
8. Work processes Poor 2 5
Nominal 1 1
Good 0.8 0.5
Insufficient 1 1
information

5.1 PRA Steps

PRA is a quantitative assessment of the risks in terms of hazards, such as hazardous

chemicals or a radioactive release and their effects on the health of the public and
environment, in the case of nuclear facilities (NFs). In addition to fissile materi-
als, NFs usually handle large quantities of toxic, corrosive, or flammable materials.
Therefore, these materials must be taken into account in typical PRAs. Furthermore,
the safety of NFs is more frequently influenced by human errors than NPPs. With
the intention of continually improve processes, operational procedures and techni-
cal equipment can undergo frequent changes. In contrast to PRAs for NPPs, which
have different levels of consequences (from core damage to environmental impact),
typical end-states of NFs accident sequences have only one stage (e.g., the release of
hazardous materials or criticality). Therefore, PRAs of NFs must take into account
these differences compared to traditional PRAs of NPPs. Figure 4 shows the main
steps necessary to perform a PRA of an NF.
Step 1 “Management and organization” covers the actions and activities neces-
sary to carry out the PRA, including the definition of objectives, scope, project man-
agement, team, training, budget, and schedule, as well as quality assurance (QA)
program.
Step 2 “Identification and selection of initiating events (IEs)” looks for a list of
internal and external events, as complete as possible, that upset the normal opera-
tion and require a facility response in order to avoid undesirable outcomes. Operator
errors, damaged process pipes or filters, transfer errors, ageing of SSCs, and degra-
dation of control functions can be mentioned as examples of typical IEs for NFs
[3].
Step 3 “Scenario modeling” intends to develop a model, which links the IEs and
the defence-in-depth levels available to prevent or mitigate the undesirable facility
end-states. This model describes the sequence of events as a consequence of an
IE, including operator actions, which are initiated for controlling the accident. The
38 V. de Vasconcelos et al.

Fig. 4 Main PRA steps of a nuclear facility

most common tools for modeling complex systems are ETA, FTA, and RBD, among
others. HRA methods are used for modeling human factors affecting the behavior
and performance of operators.
Step 4 “Data assessment and parameter estimating” consists of gathering the
information relevant for the quantification regarding the frequencies of occurrence
of operational sequences and the magnitude of consequences. Parameters such as
frequencies of IEs, component reliabilities, safety system unavailability, and human
error probabilities are necessary for frequency estimation. For consequence assess-
ment, specific parameters related to the amount, form, and transport of hazardous and
radioactive materials during postulated accidents are required. General data related
to effects to people and the environment are also required.
Step 5 “Sequence frequency” and step 6 “Consequence evaluation” consist of
quantification of scenarios using the models developed in step 3 and the data gathered
in step 4. These steps result in the assessment of the frequency of accident sequences
and the estimation of potential consequences, generally in terms of radiation doses.
There are many uncertainties related to both frequency and consequence assessments.
Assessment of uncertainties related to data, methods and models is of fundamental
importance to risk management. Uncertainty assessment usually involves the deter-
mination of the probability density function (pdf) of some parameters associated
with outcomes based on pdf of input data. For instance, uncertainty propagation of
failure probability in fault trees and event trees should be considered in PRAs, in
order to get the confidence bounds of the quantitative risk metrics.
Step 7 “Risk assessment” involves the integration (or product) of frequency and
severity assessments using the Eq. 9, and a comparison of estimated levels of risk
with defined risk acceptance criteria set by regulators. For risks involving radioac-
Integrated Reliability and Risk Assessments of Nuclear Facilities 39

tive materials, usually frequency-dose relationships are obtained and compared with
acceptance criteria (frequency-dose curves). Uncertainty assessment is also impor-
tant in this step to rank risks and verify compliance with standards and guidelines.
Where appropriate for risk management, sensitivity studies should be made for the
main assumptions and parameters to get their relative importance and guide the
decision-making processes.
Step 8 “Documentation” of the PRA, as part of the QA program, should cover
a compilation of assumptions, data, methods, detailed analyses, and interpretation
of the results. This is essential for future verification, validation, auditing, updating,
and improve the assessments.

5.2 PRA Tools

Many risk assessment tools to estimate both the frequency of accidents and their con-
sequences can be used to support the implementation of PRA steps. Some commonly
used tools are next shortly described [13, 19].
Failure mode and effect analysis (FMEA) is a qualitative technique for identifying
potential failure modes, causes, effects and control mechanisms of failures on pro-
cesses or systems. Sometimes, a semi-quantitative analysis of risks is carried out for
prioritizing corrective actions. This analysis is based on scores assigned to severity
of effects, likelihood of occurrence and detection of failures. FMEA is used at the
initial steps of PRAs for screening the postulated initiating events (IEs) and accident
scenarios.
Event tree analysis (ETA) uses a binary decision tree for modeling accident sce-
narios. It starts with an IE and proceeds through a sequence of successes or failures
of defence-in-depth levels until the end-states are reached. Each end-state is an acci-
dent sequence whose consequence should be quantitatively assessed. Each accident
sequence has its own frequency of occurrence, depending on the frequency of the IE
and on the probability of success or failure of defence-in-depth levels. These proba-
bilities can be estimated using logical tools as RBD and FTA, or analyzing available
life data, using LDA or ALT, for example.
Monte Carlo simulation (MCS) is a stochastic method for modeling based on
direct simulation of systems or processes. Through the use of random sampling of
probability density functions of input parameters, the uncertainties of output param-
eters are obtained. It is simple in principle, but requires the use of computer programs
to be implemented because the high number of samples necessary to get accurate
results. In PRA, MCS is mostly used for uncertainty propagation in FTAs and ETAs,
as well as consequence estimates of accident sequences [10, 20].
Analytical models are used in safety assessment mainly through the determina-
tion of phenomenological, statistical or empirical mathematical equations to model,
for instance, accident consequences. They are mainly based on mass, energy and
momentum conservation principles. Analytical models can represent process condi-
tions, atmospheric dispersions, explosion shockwave, radiation heat, or radiological
40 V. de Vasconcelos et al.

doses, among others, at various points in time after an accident. The consequences
can be expressed as fatalities, injuries, radiation doses, damage in structures, or
environmental impact [10].
Other tools used in PRAs, such as RBD, FTA and Markov models, are also used
in RAM analyses and have already been described in Sect. 4.3 of this chapter.

5.3 Risk and Reliability Acceptance Criteria

Risk acceptance criteria are limits and conditions defined by regulatory bodies to
ensure the adequate safety levels during licensing processes. In the case of NPPs,
these criteria can be, for example, core damage frequency (CDF) and large early
release frequency (LERF). In the case of NFs, risk acceptance criteria are usually
defined in terms of frequency-consequence (e.g., frequency-dose) ranges for workers
or the public. On the other hand, these criteria can be more subjective, for instance,
the demonstration by the licensee that IISs or other SSCs meet certain principles
and criteria, such as, ALARA principle, single failure criterion, fail-safe design or
double contingency principle [21].
NUREG 1860 [22] presents a frequency-consequence (F-C) curve based on the
principle that event frequencies and doses are inversely related, which is broadly
consistent with ICRP 64 [23]. Table 5 shows these proposed numerical criteria in
terms of frequency per reactor-year and the corresponding total effective dose equiv-
alent (TEDE) ranges for the public. These frequency-dose ranges for the public are
based on U.S. Code of Federal Regulations (CFR) Parts 20, 50 and 100, as well as
on Environmental Protection Agency (EPA) Action Guidelines. While these regula-
tory requirements provide some dose points for deriving the dose ranges, consider-
able engineering judgment was necessary to assign the corresponding frequencies.
Despite these ranges had been proposed for NPPs, their base lines and criteria can
be applied to NFs, e.g., ALARA principle, stochastic and deterministic effects of
radiation, radiological protection, and emergency preparedness.
Figure 5 shows a graphical comparison of the safety criteria of Germany, Great
Britain, ICRP 64, NUREG 1860, and Switzerland. The values plotted in Fig. 5 are
based on the data available in [3, 22]. Note that some frequency-dose curves are only
represented by step lines separating the acceptable and unacceptable regions (Ger-
man, Switzerland and NUREG 1860), while other criteria present ALARA regions
between acceptable and unacceptable regions (ICRP 64 and Great Britain).
Apart from risk acceptance criteria, there are reliability acceptance criteria to meet
the safety, performance, and QA requirements. These criteria are based primarily on
verifying the compliance of the reliability characteristics (Table 2) with the specified
values. For instance, the vendor must include documented bases for the estimated
M T T F and M T T R of IISs. Specifying RAM levels for IISs depends not only on
their safety significance, but on many factors, such as energy efficiency and economic
evaluation (e.g., pricing, maintenance, and inspection costs).
Integrated Reliability and Risk Assessments of Nuclear Facilities 41

Table 5 Proposed dose-frequency ranges for public, according to NUREG 1860 [22]
Dose range (mSv) Frequency Comment
(per reactor-year) (all doses are total effective dose equivalent—TEDE)
0.01–0.05 1.0 0.05 mSv/year is the ALARA dose in 10 CFR 50 App I
0.05–1 10−2 1 mSv/year is the public dose limit from licensed
operation in 10 CFR 20
1–10 10−3 10 mSv/event is the off-site trigger of EPA Protection
Action Guidelines
10–250 10−4 250 mSv/event triggers abnormal occurrence reporting
and is limit in 10 CFR 50.34 and in 10 CFR 100 for siting
250–1000 10−5 500 mSv is a trigger for deterministic effects (i.e., some
early health effects are possible)
1000–3000 10−6 In this range the threshold for early fatality is exceeded
3000–5000 5 × 10−7 Above 3–4 Sv, early fatality is quite likely
>5000 10−7 Above 5 Sv early fatality is very likely and curve shall be
capped

Fig. 5 Comparison of safety criteria (risk based) for nuclear facilities

42 V. de Vasconcelos et al.

The acceptable levels of complex IISs depend on the evaluation of the whole
system and interdependencies. RAM criteria and principles, such as redundancy,
diversity, physical separation, fail-safe, single failure and defence-in-depth, among
others (Table 3), should be carefully evaluated. Reliability tools, as described in
Sect. 4.3, are useful for this verification, mainly taking into account common mode
failures.
Double contingency is a requirement of particular relevance for NFs. According
to this requirement, a “criticality accident cannot occur unless at least two unlikely,
independent and concurrent changes in process conditions have occurred” [5]. This
implies, not only following the above mentioned criteria of Table 3, but also the
estimation of the probability of occurrence of failures, which can result in criticality
accidents, including human errors.

6 Integrated Approach of Reliability and Risk Assessments

Reliability, availability, and maintainability (RAM) assessments of items important

to safety (IISs) are carried out in quality assurance (QA) programs to define the
technical specifications and the required availability levels [2]. On the other hand,
safety assessment, conducted using both deterministic and probabilistic approaches,
intends to demonstrate that the facility complies with regulatory requirements. The
safety assessment shall also assure that defence-in-depth has been implemented in
the facility design [9].
RAM assessments, PRA, and definition of acceptance criteria are complementary
and can be carried out in an integrated way. Venn diagram of Fig. 6 illustrates this
integration, showing the interrelationships: technical specifications, reliability data,
risk criteria, and safety and availability analyses. The integration involves also the
use of common tools, as RBD, FTA, ETA, MCS, FMEA, LDA, and Markov models.
A framework to integrate RAM assessments and PRA is presented in Fig. 7.
According to this framework, the design basis accidents (DBAs) are defined ini-
tially through the analysis of postulated initiating events (IEs), which can include
equipment failures, operator errors, and human-induced or natural events that chal-
lenge the IISs. The accident scenarios analysis includes the modeling of accident
sequences, as well as estimating their probabilities of occurrence and consequences.
The quantitative risk assessment, integrating probability (or frequency) and severity
of consequences can identify important scenarios beyond DBAs, which in turn shall
be analyzed quantitatively adding new accident sequences to accident scenarios.
Based on risk assessment, the IISs are ranked according to their significance
in preventing or mitigating the consequences of accident sequences. PRA, using
reliability and maintainability data, taking into account the technical specifications
of IISs, shall include uncertainty assessment, in order to assure that the facility design
has adopted adequate safety margins and meets the defined risk acceptance criteria.
Frequency-consequence or frequency-dose curves, depending on regulatory bodies
of different countries (e.g., Fig. 5), are the most used criteria to evaluate whether
Integrated Reliability and Risk Assessments of Nuclear Facilities 43

Fig. 6 Venn diagram

illustrating the
interrelationships among
RAM assessments, PRA and
acceptance criteria

Fig. 7 Proposed framework

for integrating reliability and
risk assessments of nuclear
facilities
44 V. de Vasconcelos et al.

the facility complies with the acceptable levels of risk. At this point, it is important
to consider the confidence bounds of quantitative assessments (represented by the
5th and 95th percentiles of pdf of the estimated risks), obtained through uncertainty
assessment, in order to assure an adequate level of conservatism.

7 Case Study of Integrated Reliability and Risk

Assessments

In order to illustrate the use of the integrated approach to reliability and risk assess-
ments, a simple example involving analysis of a criticality accident in a nuclear fuel
fabrication plant (NFFP) was developed.
Nuclear facilities such as fuel fabrication plants handle fissile material and must
manage their processes and activities to ensure criticality safety. The nuclear subcrit-
icality depends on many parameters, such as mass, concentration, geometry, volume,
enrichment, and density of fissile material. It is also influenced by the presence of
other materials in the facility, such as moderators, absorbers, and reflectors. Sub-
criticality is then ensured through different defence-in-depth levels of protection for
preventing failures, ensuring detection, and mitigating the consequences of critical-
ity accident. They can be classified as passive, active, and administrative protection
levels [24].
Examples of passive defence-in-depth levels, which do not depend on control sys-
tems, or human interventions, are the designing of geometry and volume of pipes,
vessels, and structures inherently safe against criticality, limiting of the uranium
enrichment, presence of neutron absorbing materials, as well as shielding for mit-
igating accidental radiation doses. Examples of active defence-in-depth levels are
automatic process control, neutron and gamma monitors, and computer systems for
controlling the movement of fissile material. Examples of administrative defence-in-
depth levels are operating procedures for avoiding dangerous situations, controlling
of the isotopic composition, mass, density, concentration, chemical composition,
degree of moderation, and spacing between fissile material systems. Other examples
are defining of areas authorized to contain significant quantities of fissile material
and access control. Active components that require human actions in response to
indicators or alarms are better classified as administrative defence-in-depth levels.
Let’s consider a probabilistic risk assessment (PRA) of a nuclear fuel fabrication
plant (NFFP) carried out at the preliminary facility design phase. It was assumed
as design basis accident (DBA) a criticality excursion producing an initial burst of
1.0 × 1018 fissions in 0.5 s followed successively at 10 min intervals by 47 bursts of
1.9 × 1017 fissions, resulting in a total of 1.0 × 1019 fissions in 8 h. According to
reports of this kind of accident, this event produces little or no mechanical damage in
the structures, systems and components (SSCs) of the NFFP [25]. It is also assumed
that the system for diagnosis of criticality accidents and mitigating their consequences
Integrated Reliability and Risk Assessments of Nuclear Facilities 45

is moderately complex and there are only poorly elaborated checking procedures to
anticipate hazardous conditions.
The semi-empirical Eqs. 10 and 11, respectively, are recommended for estimating
the prompt gamma and neutrons doses following the postulated accident [25]:

Dγ = 2.1 × 10−19 N d −2 e−3.4d μc , (10)

where:
Dγ = prompt gamma dose (mSv),
N = number of fissions,
d = distance of source (km),
μc = dose reduction factor (μc = 2.5 for concrete wall thickness = 8 in. and
μc = 5 for concrete wall thickness = 20 in.).

Dn = 7.0 × 10−19 N d −2 e−5.2d μc , (11)

where:
Dn = prompt neutron dose (mSv),
N = number of fissions,
d = distance of source (km),
μc = dose reduction factor (μc = 2.3 for concrete wall thickness = 8 in. and
μc = 4.6 for concrete wall thickness = 20 in.).
To investigate design alternatives, defining reliability and risk acceptance crite-
ria, and make initial considerations about the influence of human factors on design
and operation, a preliminary PRA can be used in an integrated way with reliability
assessment techniques.
The generic event tree analysis shown in Fig. 8 was developed to investigate the
accident scenarios of criticality accident in an NFFP, where λ is the frequency of
initiating event, and P1 , P2 and P3 are the failure probabilities of passive, active and
administrative defence-in-depth levels, respectively.
An initiating event for this kind of accident could be an uncontrolled increase
of concentration of fissile material in piping and vessels of the facility. It will be
assumed conservatively a frequency of occurrence of such event as λ = 10−1 per
year [26], and P1 , P2 and P3 as 10−2 , and 10−1 , and 10−1 , respectively [27]. These
are very conservative figures since they are one order of magnitude greater than
the recommended upper limits for single passive and active defence-in-levels. The
figure of 10−1 assumed for administrative defence-in-depth, is also conservative and
is recommended for rare unplanned administrative events. It is also compatible with
SPAR-H estimates for moderately complex diagnosis tasks with poorly procedures.
Assuming that the administrative defence-in-depth levels are only to mitigate the
accident and not to prevent it, as well as the defence-in-depth levels are independent
and each one is capable of avoiding the accident, the only sequences in which the
criticality accident occurs are S1 and S2 .
46 V. de Vasconcelos et al.

Fig. 8 Generic event tree for the criticality accident analysis in an NFFP

Table 6 Estimated frequency-doses for accident sequences S1 and S2 of case study

Sequences S1 S2
Alternative 8” concrete wall 20” concrete wall 8” concrete wall 20” concrete wall
design
Freq. Dose Freq. Dose Freq. Dose Freq. Dose
year−1 (mSv) year−1 (mSv) year−1 (mSv) year−1 (mSv)
1 10−5 241 10−5 120 9× 46.9 9× 23.5
10−5 10−5
2 10−6 241 10−6 120 9× 46.9 9× 23.5
10−6 10−6
3 10−7 241 10−7 120 9× 46.9 9× 23.5
10−7 10−7
1: initial basic design 2: redundant active defence-in-depth 3: improved administrative defence-in-
depth

Doses and frequencies for sequences S1 and S2 were estimated for three alterna-
tives designs: the initial assumed basic design; the improvement on active defence-
in-depth level, adding redundancy; and additional improvements on administrative.
Table 6 presents these estimated values.
For simplicity, the external doses from dispersion of fission products were
neglected, and only prompt gamma and neutron doses were considered in calcu-
lations for the public individuals of the critical group, assuming they are located at
0.1 km of distance.
Integrated Reliability and Risk Assessments of Nuclear Facilities 47

Fig. 9 Frequency-dose curve (based on NUREG-860 [22]]) and accident sequences S1 and S2 ,
comparing different design alternatives

The total doses estimated for the sequence S1 , using Eqs. 10 and 11, were 241
mSv for 8 in. concrete wall thickness and 120 mSv for 20 in. concrete wall thickness,
respectively. These doses were estimated considering that the public individuals
remain throughout the course of the accident at the distance of 0.1 km. The doses
from sequence S2 , estimated supposing an evacuation time of one hour, were reduced
to 46.9 mSv for 8 in. concrete wall thickness and 23.1 mSv for 20 in. concrete wall
thickness, respectively.
The frequencies of occurrence of sequences S1 and S2 , estimated according to
equations presented in Fig. 8, were 10−5 and 9 × 10−5 , respectively. The estimated
risks are plotted in a frequency-dose curve, according to NUREG 1860 criteria [22],
taken as reference. Both sequence S1 and S2 , for the initial basic design defence-in-
depth levels, lie under the curve, in an acceptable region, as can be seen in Fig. 9.
However, the dots are very close to the curve, and if the uncertainties on estimating
frequency and doses were taken into account, certainly the dots would lie in an unac-
ceptable region [20]. Therefore, further design alternatives should be investigated.
Adopting redundancy at active defence-in-depth level, its reliability becomes 0.99,
which reevaluates P2 as 10−2 per year. Revising design for simplifying diagnosis
system and improving maintenance, checklist, and operation procedures, the nominal
value of 10−3 per year can be adopted to P3 , according to SPAR-H method. In this
way, new values estimated for criticality risks are achieved, which lie in the acceptable
region in Fig. 9, far from limits of the frequency-dose curve and with adequate safety
margins for the risks.
48 V. de Vasconcelos et al.

8 Concluding Remarks

A framework integrating reliability, maintainability and availability (RAM) assess-

ments of items important to safety and probabilistic risk assessment (PRA) of nuclear
facilities (NFs) is proposed in this chapter. The use of common reliability engineer-
ing tools and reliability data could optimize the safety and availability analyses. The
outcomes of this proposal can also support the licensees in achieving a high level
of safety and reliability through the integrated and complementary use of RAM and
PRA tools in all phases of NF lifetime and meeting licensing requirements.
A case study demonstrating the advantages of using the integrated framework to
a preliminary facility design phase of a nuclear fuel fabrication plant was presented.
Design and administrative control alternatives were analyzed to reduce the criticality
accident risk. This case study highlighted that the use of redundancy of active defence-
in-depth levels and considerations of human factors from the feasibility study and
preliminary design phase helps to meet the reliability and risk acceptance criteria, as
well as nuclear licensing requirements.

Acknowledgements The authors thank the following intitutions: CNEN (Brazilian Commission
of Nuclear Energy), CDTN (Center of Nuclear Technology Development) and FINEP (Funding of
Science, Technology and Innovation).

References

1. Christensen F, Andersen O, Duijm J (2003) Risk terminology: a platform for common under-
standing and better communication. J Hazard Mater 103:181–203
2. International Atomic Energy Agency (2001) Reliability assurance programme guidebook for
advanced light water reactors. IAEA-TECDOC-1264. IAEA, Vienna
3. International Atomic Energy Agency (2002) Procedures for conducting probabilistic safety
assessment for non-reactor nuclear facilities. IAEA-TECDOC-1267. IAEA, Vienna
4. Winteringham F, Peter W (1992) Energy use and the environment. Lewis Publishers Inc, London
5. International Atomic Energy Agency (2018) IAEA safety glossary terminology used in nuclear
safety and radiation protection. 2018th edn. Vienna
6. International Atomic Energy Agency (2020) Safety analysis and licensing documentation for
nuclear fuel cycle facilities. Safety Reports Series No. 102 IAEA, Vienna
7. International Atomic Energy Agency (2017) Safety of nuclear fuel cycle facilities. IAEA Safety
Standards Series No. SSR-4, IAEA, Vienna
8. International Atomic Energy Agency (2010) Licensing process of nuclear installations. Specific
Safety Guide No. SSG-12 IAEA, Vienna
9. International Atomic Energy Agency (2016) Safety of nuclear power plants: design, specific
safety requirements. SSR-2/1 (Rev. 1), IAEA, Vienna
10. Vasconcelos V, Soares W, Costa A et al (2019) Deterministic and probabilistic safety analyses.
In: Advances in system reliability engineering. Elsevier, London
11. Reliasoft Corporation (2015) System analysis reference: reliability, availability and optimiza-
tion. ReliaSoft, Tucson
12. Reliasoft Corporation (2015) Life data analysis reference. ReliaSoft, Tucson
13. Stamatelatos M (2002) Probabilistic risk assessment procedures guide for NASA managers
and practitioners: Version 1.1. NASA, Washington DC
Integrated Reliability and Risk Assessments of Nuclear Facilities 49

14. Fleming K (2004) Markov models for evaluating risk-informed in-service inspection strategies
for nuclear power plant piping systems. Reliab Eng Syst Saf 83:27–45
15. U.S. Nuclear Regulatory Commission (2004) The SPAR-H human reliability analysis method.
NUREG/CR 6883. USNRC, Washington DC
16. Park J, Jung W, Kim J (2020) Inter-relationships between performance shaping factors for
human reliability analysis of nuclear power plants. Nucl Eng Technol 52:87–100
17. U.S. Nuclear Regulatory Commission (1975) WASH-1400: reactor safety study, NUREG
75/014. USNRC, Washington DC
18. International Organization for Standardization (2009) Risk management: principles and guide-
lines. ISO 31000, 1st edn. ISO/IEC, Geneva
19. Calixto E (2013) Gas and oil reliability engineering, modeling and analysis. Elsevier, Amster-
dam
20. Vasconcelos V, Soares W, Costa A, Raso A (2019) Treatment of uncertainties in probabilistic
risk assessment. In: Reliability and maintenance-an overview of cases. IntechOpen, London.
https://www.intechopen.com/chapters/65179
21. International Atomic Energy Agency (2001) Applications of probabilistic safety assessment
(PSA) for nuclear power plants. TECDOC-1200, IAEA, Vienna
22. U.S. Nuclear Regulatory Commission (2007) Feasibility study for a risk-informed and per-
formance based regulatory structure for future plant licensing. NUREG 1860, v. 1, USNRC,
Washington DC
23. International Commission on Radiological Protection (1993) ICRP publication 64: protection
from potential exposure: a conceptual framework. Ann ICRP 23(1). Ottawa
24. International Atomic Energy Agency (2014) Criticality safety in the handling of fissile material.
Specific Safety Guide No. SSG-27, IAEA, Vienna
25. U.S. Nuclear Regulatory Commission (1979) Regulatory guide 3.34. Revision 1. Assumptions
used for evaluating the potential radiological consequences of accidental nuclear criticality in
a uranium fuel fabrication plant, Washington, DC
26. Laboratory Los Alamos National (2000) LA-13638. A review of criticality accidents (2000)
Revision. Los Alamos, New Mexico
27. Canadian Nuclear Safety Commission (2018) REGDOC-2.4.3. Nuclear criticality safety.
Ottawa, Ontario
Computational Tools of Media Analysis
for Corporate Policy Effectiveness
Evaluation: Models and Their Reliability

Gregory S. Khvatsky, Dmitry G. Zaytsev, Valentina V. Kuskova,

and Anna A. Sokol

1 Introduction

Human decision-making is becoming more complex due to increasing amount of

information t that needs to be analyzed in today’s modern systems. Decision-making
in business corporations is not an exception. Such complexity raises a question about
reliability or broader efficiency and effectiveness of corporate decision-making.
Corporate efficiency and effectiveness is strongly associated with inclusion of the
modern computational or data science tools into the strategic corporate decision-
making. A special concept describes this relationship—business intelligence (BI).
BI is “the use of all the organization’s resources: data, applications, people, and
processes in order to increase its knowledge, implement and achieve its strategy, and
adapt to the environment’s dynamism” [2, p. 6].
However, BI faces some challenges. Among them are the ability to identify “pat-
terns, trends, rules, and relationships” [2, p. 6]. Alnoukari and Hanano [2, p. 6] from
large amounts of data that are available, Most of that data are too complex for humans
to process. Data complexity leads to the problem where complex corporate efficiency
and effectiveness evaluation systems are limited to integration of computational tools

G. S. Khvatsky · D. G. Zaytsev (B) · V. V. Kuskova · A. A. Sokol

International Laboratory for Applied Network Research, HSE University, 11 Pokrovsky Bulvar,
Pokrovka Complex, Moscow 101000, Russia
e-mail: dzaytsev@hse.ru
G. S. Khvatsky
e-mail: gkhvatsky@hse.ru
V. V. Kuskova
e-mail: vkuskova@hse.ru
A. A. Sokol
e-mail: asokol@hse.ru

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 51

M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_3
52 G. S. Khvatsky et al.

into the simpler models and approaches such as SWOT analysis, scenario analysis,
and forecasting [2, 3, 36].
Nevertheless, efficiency is a concept that is universally used across all social
sciences—not only in management. In political science, efficiency is associated with
governments and states [34], in sociology—with various social institutions (such
as civil society) and the social environment [16]; in economics—with commerce
efficiency [39]; in psychology—with efficient behaviors of individuals [13]. This
multidimensional nature of efficiency has also led to the propagation of attenuating
concepts such as effectiveness [23] and efficacy [28].1
Therefore, to establish a reliable system of corporate efficiency evaluation that
is not limited to simpler models, one must take into consideration the increasing
complexity of decision-making in modern systems, including large corporations.
Outside the narrower field of management, this complexity is reflected in social
sciences, and mostly in policy science. Policy science is as a newer interdisciplinary
field, studying of policy-making, and it integrates contemporary developments in
economics, management, sociology, and political science [20]. The context-specific,
problem-oriented, and evidence-based approach of policy science also allows policy
analysts, corporate consultants, and business intelligence agents to better inte-
grate data science methods into their decision-making process, thereby increasing
corporate policy effectiveness.
Changing the focus from corporate efficiency to corporate policy effectiveness
embeds the necessary complexity into evaluation of decision-making. The concept
of “policy” provides a broader context of the social environment, which serves as a
ecological niche in which large corporations function and for which they carry certain
responsibilities.. Policy effectiveness focuses not only on the efficient results (any
outputs achieved with the minimum use of resources), but also on policy outcomes
(corporate policy effects that may be valuable for the general public and the society
at large).
Table 1 provides questions for corporate policy effectiveness evaluation, derived
from the stages of applied problem-solving approach, and the corresponding stages in
the policy cycle theory and practice. To increase the effectiveness, policy designers,
managers, practitioners, scientists, and social engineers usually have to answer
these questions using different methods and tools. Therefore, the table serves as
the summary of the commonly used methods and computational tools.
We argue that computational tools are not inferior, and in some cases, they may
be superior to the traditional methods of corporate policy effectiveness evaluation.
They could provide additional answers when decision-makers are dealing with big
corporate data wherehuman processing is impossible. Examples of such data are data
from old and new media (social networks), which lately became the source of many
important insights that affect corporate effectiveness.

1Despite the differences between these concepts (efficiency, effectiveness, and efficacy) exist, their
detailed analysis is beyond the scope of this chapter.
Computational Tools of Media Analysis … 53

Table 1 Corporate policy effectiveness evaluation: questions and methods

Applied Stages in policy Questions Traditionally Computational
problem-solving cycle used methods tools
Problem Agenda-setting Who is concerned Stakeholder Named-entity
recognition (problem about the analysis recognition (NER)
definition) problem? Mass survey Social networks
What is their Expert survey analysis (SNA)
status? Focus groups Natural language
Are there new processing (NLP):
ideas? topic modeling
Proposal of Policy What policy tools Experts NLP: topic
solution formulation are used? interviews modeling
(constructing the Are there new (brainstorming, NER & SNA with
policy actors with new consulting) NLP: topic
alternatives) ideas? modeling
Choice of Decision-making What are the Cost–benefit NLP: topic
solution (selection of effects of policy analysis modeling and
preferred policy constraints? Scenario sentiment analysis
option; policy What is the policy analysis
design) capacity level? SWOT analysis
Putting solution Policy What is the Interviews with SNA
into effect Implementation relationship of policy actors
policy means with
policy goals (both
outputs and
outcomes)?
Monitoring (and Policy What are the Program NER, SNA, NLP:
evaluation) of (monitoring and) learning monitoring topic modeling
results evaluation outcomes? Performance and sentiment
management analysis
Program
evaluation

Natural Language Processing in general, and Topic Modeling in particular,

became the key instruments of analyzing old and new media data for a large corpo-
ration (Table 1). Therefore, in this chapter we are focusing on different techniques
of topic mining and their reliability for policy-making in a large corporation. Our
main contribution is an integration of computational tools such as topic mining into
specific tasks of business intelligence, which then leads to an increase of corporate
policy effectiveness.
The context for this study came from the case of one of the largest transnational
corporations (TNK) based in Russia (“the Company”). The problem faced by the
Company was a concern with its reputation. The management wanted to implement
a new corporate policy aimed at building a more effective image and increase the
company’s reputation in old and new media.
To demonstrate how computational tools were used to improve business intelli-
gence, we first provide a short overview of the existing methods of topic mining.
54 G. S. Khvatsky et al.

Second, we describe how we use three popular data science methods of topic mining
to tackle a real-life business problem. Third, we provide an analysis of the advan-
tages and disadvantages of the topic mining methods related to the case of the Russian
TNK. In conclusion, we provide recommendations for development of a system of
monitoring and evaluation of corporate policy effectiveness using computational
policy tools, topic mining procedures in particular.

2 Theory and Methodology of Topic Mining

Text data is one of the most important sources of information and knowledge. In fact,
about 95% of big data come in text or some other unstructured form.2 Therefore, the
ability to get ideas and understanding of such data is an excellent advantage for the
successful possession and control of information in business [18, 27]. Social media
plays a decisive role in propagating textual information, having a great influence on
what information reaches the public [10, 31]. Today’s coverage of information has
greatly increased the ability of both ordinary people and large corporations to publish
their content and expand their reach.
At the same time, as information and knowledge from these media continue to be
digitized and stored in the form of news articles, web pages, social media posts and the
like, analyzing all this information will help businesses manage it in time to respond
to various events [19, 29]. That is why learning the characteristics of social media
content becomes important for a number of tasks, such as recent news discovery,
personalized message recommendation, friend recommendation, sentiment analysis,
and others.
However, analyzing social media content presents certain challenges. Typical
social media contains large amounts of data consisting of brief writings with serious
fragmentation. Many documents are “off-topic” to the subject of interest; excess
highlights often complicate readability. Users have no limitation on content they
post; their language is not sufficiently standardized, unlike news or scientific articles.
The language may have urban dialects and digitals forms and reductions, making it
difficult to recognize and understand. All of these issues create problem for data
reduction when processing social texts.
The challenge of identifying common topics is to detect “hot” topics in social
media data streams. In natural language processing, including topic modeling, there
are multiple approaches for detecting, labeling, and pattern detection. Most of these
tasks are done by machine learning algorithms. In this study we demonstrate three
different approaches to topic mining and outline their benefits and drawbacks.
Topic modeling became a popular analytical tool for evaluating text data.
Researchers use topic modeling for different reasons: to find similarity of post content
in social media [9, 15], explore environmental data [11], recommend scientific arti-
cles [37]. It also has a special place in management research [14]. There are several

2 https://builtin.com/big-data.
Computational Tools of Media Analysis … 55

definitions of “topic modeling;” in this study, we will use the definition of [7], who
define it as “an unsupervised machine learning method that learns the underlying
themes in a large collection of otherwise unorganized documents” [7, p. 419].
Topic modeling allows to capture the hidden semantic structures in a document.
Typical document is composed of mixtures of topics, and topics are composed of a
set of words [17, 32, 33]. Therefore, the methods used usually are aimed at separating
these sets from each other.
One popular topic modeling method is the Latent Dirichlet Analysis (LDA). Orig-
inally proposed for use with genetic data [30], the algorithm quickly found its way
into text analysis and machine learning [5], where it remains popular to this day.
LDA is a three-level hierarchical Bayesian model, where documents are represented
as random mixtures over latent topics. Each of the topics is a probability distribu-
tion over individual units of analysis (e.g. individual words.) Topics in LDA do not
mean the same as “topics” or “discourses” in qualitative text analysis, but rather
probability distributions over words that need additional interpretation. The model
has one hyperparameter—the number of topics to estimate. While there are methods
that allow for computational estimation of number of topics, in a particular, a corpus
[26], using them for large corpora requires a large amount of computational effort.
LDA for traditional text mining has made some progress and achieved good results
[5, 12]. Despite its wide popularity for complex text data, it is not very efficient at
analyzing short texts from social media, which have their own unique properties, so
special care is required when modeling social media data with LDA.
Another another popular method is Hierarchical Dirichlet Process (HDP). HDP is
an extension to LDA, designed to be used when the number of “topics” in inside the
document is not known [40]. Using this method, researchers can effectively improve
the efficiency of text segmentation.
Because both methods are the so-called these bag-of-words methods [8, 21], they
can be used without manual intervention since they are not concerned with text
structure. This is somewhat of a limitation, since the topics obtained automatically
by these methods can be correlated with each other, and this can complicate their
interpretation [6, 38].
There is another problem that complicates topic modeling. Sometimes, words
in the training data are spelled the same, but have different meaning and represent
different topics. Such words as “Elephant” or “Donkey” belongs to the “animal”
topic, but these words are also related to the “American politics” topic. Both LDA
and HDP are insensitive to the original topic of such words.
One way to solve the above limitations is the “manual labeling” method. Experi-
ments with real data sets have shown that manual labeling method is efficient and flex-
ible for solving problems related to brand image [1, 40]. In this study, we compare the
manual labeling method with several others for the use in corporate policy efficiency
evaluation.
56 G. S. Khvatsky et al.

3 Application of Topic Mining Algorithms to Real-Life

Problems in Business and Public Administration

3.1 Policy Problem: Corporate Reputation Vulnerability

Impact of social media on corporate reputation cannot be underestimated: with the

right words and enough reposts, reputation of a company can be damaged in a matter
of minutes [24]. By now, most companies know this very well. The case presented
in this chapter is based on a large Russian multinational corporation (the Company).
The Company was well aware of the effects that social media had on opinions of its
stakeholders, so it was working to design a corporate policy aimed at effective public
image management.
The company approached the authors of this study with several problems. First,
it wanted to have an understand of its then-current corporate image: the sentiment
around the company, topics that were generating the largest number of positive and
negative comments, and the overall sentiment of public discussion around itself.
Second, it wanted to design a media monitoring tool that would provide near-
immediate results of frequent media scanning for the purpose of evaluating any
changes in corporate reputation. This was a complex project consisting of several
parts; this paper reports on the topic mining component of it—the first problem that
the company posed before the authors.

3.2 Data

The data used for this project were collected via a media monitoring platform called
YouScan. This platform provides a tool that collects data from a variety of sources,
including social media websites, online news websites, web forums, blogs, and
posts made in public channels of messaging applications. YouScan also maintains
an archive of such data and provides access to it. The workflow for getting data
from YouScan is as follows: the user defines a “topic” that is denoted by several text
queries. The system searches the archive for posts matching the queries, and starts
monitoring for new data. The user can then export the data for offline processing via
a built-in API.
In total, we have collected 3,118,412 publications in Russian (texts and metadata
information), ranging in time from August 30, 2017 to October 29, 2019. Publications
we collected could be separated into multiple groups based on the role they played
in the discussions that happened on the online platforms monitored by YouScan. The
most common of these publications is a ‘post’ (1,270,434 publications). This type
of a publication describes a text that was published on an online platform by a user.
The descriptions of all publication types and their counts are provided in Table 2.
Computational Tools of Media Analysis … 57

Table 2 Publication types with counts and descriptions

Type Count Description
post 1,270,434 A text that was published on an online platform by a user
repost 841,506 A publication of someone else’s post made by another user
or community on their own site or page
comment 596,834 A comment on a post or a repost
replyComment 315,674 A comment on a post or a repost that is a reply to another
comment
Extended repost 93,964 A repost of a post made by another user or community that
has some content added by the user or community that
made the repost

Table 3 Top-10 data sources

Source Count Percentage of total (%)
and publication counts
vk.com 1,050,353 33.68
facebook.com 793,709 25.45
ok.ru 342,666 10.99
twitter.com 294,028 9.43
telegram.me 108,743 3.49
youtube.com 77,498 2.49
instagram.com 53,444 1.71
livejournal.com 38,632 1.24
talks.by 20,408 0.65
pikabu.ru 19,012 0.61
Other sources 319,919 10.26

YouScan collects data from a wide variety of sources, including, but not limited to,
social network websites, public Telegram channels, and a selection of news websites.
A more complete breakdown on data sources for our dataset is provided in Table 3.
As shown in Table 3, almost 80% of the data came from social network websites.
The most common of these and most common overall is VKontake (VK), one of the
most popular social network website in the Russian-language segment of the Internet.
It is followed by Facebook, Odnoklassniki (ok.ru), another social network targeted at
the Russian-language segment of the Internet, then Twitter and Instagram. Another
large source of data is Telegram, a messaging application that allows its users to
create public channels to which they can post. Other users can then subscribe to
these channels and discuss the posts in them. Telegram is also popular in Russia,
with more than 25% penetration rate.3 Among the top-10 most common data sources
are blog websites such as LiveJournal (similar to Blogspot) and Pikabu (similar to
Reddit), which are also targeted at the Russian-speaking audience. Also among the

3 https://www.statista.com/statistics/867549/top-active-social-media-platforms-in-russia/.
58 G. S. Khvatsky et al.

Table 4 Distribution of
Resource type Count Percentage of total (%)
publications across different
resource types social 2,628,923 84
news 219,663 7
blog 98,944 3
messenger 92,528 3
forum 76,852 2
reviews 1502 0

most common sources of data are YouTube and a large Belarus-based web forum
(talks.by) that is now defunct.
The distribution of publication count across different types of media indexed by
YouScan is presented in Table 4. As Table 4 shows, most popular type of resource
are also social network websites. Social media accounts for 84% of all publications
in the dataset.
YouScan collects a large amount of information on each of the publications. Of
prime interest for this study are the texts, but other information is also present.
YouScan collects data on the time and date the publication was made, the source
of the publication, information about the author of the publication (e.g. name, age,
gender of a person, and a URL and a name for a community) and the amount of
engagement (shares, likes, reposts) a publication has received. What information is
collected is dependent on the source of the publication. For example, it is not possible
to collect engagement rates from a web forum that does not allow its users to “like”
and “share” the discussions.

3.3 Preprocessing

Before doing any analysis, we applied multiple preprocessing steps to our data to
clean it and improve the interpretability of the results obtained later. Our first step
was text cleaning. Since the original corpus was in Russian, we removed English
words. We also removed emoji characters, hashtags, punctuation, and web URLs
from the texts in our database. The purpose of this step was to eliminate words that
could confuse the lemmatizer and to remove information that was not relevant to our
analysis.
Next, we performed the lemmatization procedure on all of the texts in the database.
Russian language has a fusional morphology, which means that words have different
forms according to their roles in a sentence. This can lead to problems with text
analysis, since bag-of-words based algorithms such as LDA may recognize different
grammatical forms of the same word as different words. Because of that, a lemma-
tization procedure that puts all words into their dictionary form is required when
performing bag-of-words based analysis.
Computational Tools of Media Analysis … 59

For this study, we used Mystem—a closed-source, but free to use lemmatizer for
Russian and Ukrainian languages. It uses a hybrid approach combining a suffix tree
and a dictionary. This approach allows for both accurate lemmatization of known
words and guessing dictionary forms of words not originally present in the dictionary
[35]. It should be noted that in our case, Mystem failed to properly lemmatize the
name of a large Russian state enterprise that was very common in our dataset, so we
had to create a custom dictionary in order to perform this step in a proper manner.
As a final step, we removed stopwords from our lemmatized texts. Stopwords are
words that carry no meaning and can be detrimental to bag-of-words based analysis.
Examples of such words for the English language are the verb “to be”, words such
as “and,” “or,” and “maybe”. There are lists of such words for almost all languages,
and we have used one of these lists for this study. After performing all of these steps,
the texts were finally ready for further analysis.

3.4 Latent Dirichlet Allocation

The first algorithm that we used for this study was Latent Dirichlet Allocation (LDA).
This algorithm has readily available open-source implementations, which makes it
particularly easy to use.
For our application of LDA, we used the implementation provided by the Gensim
Python library. The Gensim library is a collection of NLP algorithms and models.
The library strives to provide implementations of various models that are able to
process arbitrarily large corpora that might not fit into RAM of the computer running
the models. This property was quite useful for this study, as our corpus of 3,118,412
publications did not fit into RAM of the computer running the models.
We did not use algorithmic detection of the number of topics due to the extreme
computational complexity of having to fit a large number of models needed to esti-
mate the value of the hyperparameter. For this study, we chose to train an LDA model
with 100 topics and use its results for further analysis. While this number may appear
somewhat arbitrary, methodological studies aimed at detecting the number of topics
for topic modeling usually stop at about 100 (e.g., [41]). Therefore, we started with
this large number to make sure we picked up most of the topics.
With our particular implementation of LDA we faced a set of problems. First,
when building the dictionary of words that are present in the corpus, Gensim may
remove words that are not very common in the corpus. Such words, however, could
still be important for our analysis. One example is words that are present in less than
10 publications in the corpus. Second, Gensim may also remove the words that are too
common—for example, words that are present in more than 70% of the publications
in the corpus. One way to solve this problem was to tweak the parameters used for
this word removal. However, we decided to take an alternative approach, and created
a list of words that should not be removed from the dictionary even if they were
rare. These words included words that, for example, referred to events and people
important for the industry under investigation.
60 G. S. Khvatsky et al.

A third problem was the presence of publications that were not relevant to our
analysis. These publications (e.g. advertisements) have decreased the performance
of LDA: discovered topics often had words from the advertisements with very high
probabilities, which made interpretation of the discovered topics much more difficult.
The fourth problem we faced was a large number of topics that were highly
correlated with one another, thus making the model interpretation much harder. This
was most likely due to the large number of topics in the model. Also, it could be due
to the way that the original search query was constructed, because certain words (that
were part of the query itself) were present in a very large numbers of posts. Again,
this presented a problem for our analysis as it made it much harder for the training
algorithm to differentiate different topics present in the posts from one another.
Yet another problem that affected the model interpretability was the choice of the
procedure used to assign topics to individual publications. LDA assigns a probability
distribution of topics to each of the publications in the corpus. This means that
technically, all publications have all topics present within them with various degrees
of “affinity.” Determining which topics are present in which publication becomes
quite complicated as a result. An obvious choice would be to simply select the most
probable topic for each of the publications. However, doing so is based on an implicit
assumption that a publication can only have exactly one topic present within it. After
using this approach on our data, we have found that this assumption was not true for
many of the publications present in our dataset.
Thus, we decided to implement an alternative approach. We picked a threshold
value and selected all the topics with the probability equal to or higher than this
value. This approach allows for multiple topics to be present in one document, but
of course, is based on determining the appropriate threshold value. For this study,
set the probability threshold to the standard 0.05 or 5%.
After building the models, we visualized them using another Python library called
pyLDAvis. This library can convert LDA topic models to interactive visualizations
that can be used to analyze and interpret the meaning of topics discovered in the corpus
algorithmically. The library also visualizes the relationships between different topics
via a what is called a “topic map.” PyLDAvis builds these maps by first computing a
matrix of similarities between all possible pairs of topics and then embedding it into
a 2D space using the classic multidimensional scaling. An example of such map for
the LDA model we built on our corpus of 3,118,412 publications is present in Fig. 1.
It can be seen from the map that topics 93, 17, 8, 42 and 24 are the most common in
the corpus. After in-depth analysis, we have determined that they represent important
discourses present in the media.
Topic 93 represents the discourse that is related to the Company as well as its
subsidiaries. Publications related to this topic describe corporate activities of the
Company, such as public events that it organizes. A sizable part of publications
related to this topic were also related to companies that were important not only to
the Company itself, but to the industry as a whole. For example, certain environmental
protection agencies, societies and the events they organize became included in our
sample.
Computational Tools of Media Analysis … 61

Fig. 1 An LDA model topic map

Topic 17 represents the discourse that is related to the workers of the industry
that the Company operates in. The publications related to this topic had discussions
on the identity of typical industry workers and descriptions of their daily lives. This
discourse also contained the discussions of dangers that are inherent to the industry
for the workers themselves, the environment, and humanity as a whole.
Topic 8 represents the discourse related to one of the worst disasters to happen to
the industry in question, its short- and long-lasting effects on the lives of the affected
people, the environment, and the future of the industry. Another important aspect of
the discourse is the continued discussion of the ways the disaster was represented in
various media (e.g., TV shows and video games).
Topic 42 represents a discourse that is related to that of topic 8. The discourse for
this topic is also centered on the same disastrous event, but it centers more on the
media depictions of the event, the long-term effects on the environment surrounding
the epicenter of the event, and the exploration of this environment.
62 G. S. Khvatsky et al.

Topic 24 represents a discourse that is connected with the industry in question as

a whole. It also contains discussions on its relationship with other industries.
While we found the results of the analysis useful, because of the problems with
the analysis described above, we decided to apply other topic discovery algorithms
to our dataset. Our goal was to order to compare the results we obtained by using
different algorithms and models.

3.5 Hierarchical Dirichlet Process

Another method that we attempted to use for topic modeling and topic discovery
was to perform topic modeling using the Hierarchical Dirichlet Process (HDP) topic
model. While it is harder to use than LDA, the availability of free and open source
implementation makes the use of this model relatively easy.
For our application of HDP, we have used the implementation provided by the
Gensim Python library. The implementation of HDP provided by Gensim is based
on an online algorithm that allows for analysis of arbitrarily large text corpora. As in
the case with LDA, this property was very useful due to the large size of our corpus.
For this model, we focused only on some types of publications in our corpus. We
removed comments from our analysis since they did not add as much substance to
the discussion of the Company and industry as posts and (extended) reposts. The
size of our final corpus used to train the model was 2,205,904 publications, and it
consisted of posts, reposts, and extended reposts.
Since HDP can infer the number of topics from the data, we did not employ
any algorithmic procedures for topic count estimation. For this study, we set the
maximum possible number of topics in the model to 150.
HDP and LDA are both based on the bag-of-words approach and as such share a
common set of problems. We followed the same procedure for building the dictionary
as outlined in the section on LDA. We used the same list of words that were important
for our analysis to prevent them from being removed from the dictionary based on
being too common or too rare. When building the dictionary, we used the same
parameters for word exclusion, removing words from the dictionary if they were
present in less than 10 publications, or more than in 70% of publications.
Another issue that was shared to a degree between HDP and LDA modeling
approaches was that the topics were highly correlated and contained similar sets of
probable words. In the case of HDP, this problem was alleviated by the fact that
the number of topics inferred from the text by the model was much lower than the
number we used for the part of our analysis based on LDA. Also relevant to this
algorithm were the issues caused by the presence of irrelevant publications in the
dataset.
Assigning topics to individual publications has also proved problematic. As with
the case of LDA, HDP assigns a probability distribution over all the topics to each
of the documents. However, this issue affects HDP much more than LDA, because
the implementation of HDP we used generates a distribution not over the inferred
Computational Tools of Media Analysis … 63

number of topics, but over the maximum possible number of topics specified when
training the model. To remedy this, we decided to use an approach similar to what
we did for LDA: we set a probability threshold of 5%. As with LDA, we could assign
more than one topic to a document.
After training the model, we created a visualization procedure that allowed us
to see the relationships between different topics. While HDP models can be trans-
formed into LDA models, this transformation destroys a large portion of the available
data, such as the number of topics inferred by the model. This severely limits the
usefulness of converting HDP models to LDA models to aid in their visualization
and interpretation.
To visualize the model we used the following procedure. First, for each topic we
counted the number of documents related to the topic using the rules described above.
Next, for each of the topics inferred by the model, we obtained a sample of 1000 most
probable words with their probability scores. We then created a topic-word matrix
with rows representing topics and columns representing sampled words. If for some
topic a word was not in the 1000 most probable, we assigned to it a probability score
of 0. Next, we used t-SNE to embed this large matrix into a 2-dimensional space. An
example of a topic map created using this procedure is presented in Fig. 2. The size
of the bubble represents the number of documents related to each of the topics.
As Fig. 2 shows, the model has discovered 20 topics in our dataset of publication
texts. Topics 0, 1, 2, 3 and 6 have the largest number of documents related to them,
and further analysis suggests that they represent important discourses that exist in
social media.

Fig. 2 HDP model topic map

64 G. S. Khvatsky et al.

Topics 0, 3 and 6 represent a discourse that is centered around one of the largest
disasters to happen to the industry under investigation. This discourse also refers
to the geographical and geopolitical area where the event took place, as well as the
lives of people involved in the containment measures and otherwise affected by the
tragedy as well as its long-term effects. Topic 0 is focused on the lives of people
affected by the tragedy, and by its effects on the area surrounding the site. Topic 3
is concerned with the strain that this tragedy put on the industry under investigation
in the country where it happened, as well as how this strain affects large economic
sectors of this country. Topic 6 is concerned with the general dangers of the continued
development of the industry in question, and their connection with one of the largest
disasters that happened in connection with that industry.
Topic 1 represents a discourse centered around investments that the Company
makes in its home country and how it helps develop the remote parts of the country.
This discourse also contains discussions around new large-scale projects undertaken
by the corporation and the jobs it creates.
Topic 2 represents a discourse centered around new technological developments
in the industry in question and the role of the Company in these developments.
This discourse also includes discussions on the investments and job creation by the
corporation.

3.6 Manual Labeling

The third method that we used for topic analysis for this study is what we called
“manual labeling.” While this method required human expert involvement, it allowed
for simultaneous topic discovery and interpretation. It also allowed us to filter out
publications that were detrimental for analysis, such as advertisements.
The manual labeling method for topic analysis is centered around assigning topics
to publications based on a dictionary that is created by a panel of experts. A topic
can be assigned to a document based on the text of the publication or the set of other
topics assigned to the document (for example, if a word is present in the text or a
match of a regular expression).
For example, when analyzing a dataset related to computer hardware, one can
define a topic called “motherboards.” This topic will be assigned to a publication if
it contains the word “motherboard” or the word “mainboard.” Other topics might be
“graphics cards” (with the words being “graphics card” or “display adapter”) and
“keyboards” (with the word “keyboard”). It would then be possible to define two
other topics. One is “core components”, assigned to the publications that already
have topics “motherboards” or “graphics cards” assigned to them. Another is the
“peripherals,” assigned to publications that have the topic “keyboards” assigned to
them. The hierarchy of topics can be infinitely deep. It would also be possible to
define the topic “computers” that would be assigned only to those publications that
have both the “core components” and the “peripherals” topics assigned to them. This
topic hierarchy is represented in Fig. 3.
Computational Tools of Media Analysis … 65

Fig. 3 Example of a topic hierarchy

For our analysis, we created our own implementation of this method using the
Python programming language. In our implementation, the definition for each topic
contained sets of regular expressions (used to search for substrings in the text), sets
of other topics, or a combination of both.
In order to build the topic dictionary using expert opinion, we devised the
following cyclic procedure. (Experts were the people who had the subject matter
knowledge of the industry.) Starting with the initial empty dictionary, we sampled a
set of documents that did not have any topics assigned to them from our database.
After that, we had a panel of experts analyze the sample of the documents in order
to make additions and modifications to the dictionary so that all the sampled docu-
ments had topics assigned to them. In the case of our study, the experts identified
various discourses that existed in our dataset. They also identified the connections
that existed between the discourses that were already discovered. In this study, we
repeated this procedure until all unassigned documents remaining were not relevant
to our analysis (e.g., unclassified advertisements).
One of the biggest challenges inherent to this type of analysis is that it is both
time-consuming and susceptible to subjectivity of the experts. While the first problem
remains unsolved, we have tried to remedy the second problem by having a panel of
multiple experts. Additionally, we have used independent coding together with the
experts discussing the coding results to reach consensus among experts and to make
the coding as objective as possible.
Another problem related to this method is that it does not allow to determine the
level of affinity of a publication to a particular topic, as our implementation only
allows for binary topic assignment. Still, the method allows to assign arbitrarily
many topics from the dictionary to a single publication, so it does not require the
assumption of one topic per document to hold.
Another issue with using this method (that it shares with other bag-of-words text
analysis methods) is that it largely ignores the structure of the text of the publications.
What makes this issue important is that particular care should be taken when building
66 G. S. Khvatsky et al.

the dictionary, since ambiguous entries can greatly affect the quality of the extracted
topics. For example, if the dictionary is designed to use single words to assign topics
to documents, using this method can lead to assigning topics to documents that are
not relevant to the topic. A general recommendation in this case is to include as much
context as possible when adding text-based entries to the dictionary.
While it was possible, our implementation of this method did not make the use
of additional metainformation available for the publications in our dataset. Since the
analysis performed for this study was largely exploratory, we reserved the use of
metainformation for future work. However, after performing this analysis, we used
the metadata available for the publications for further analysis.
After performing all the procedures outlined above, we discovered 189 topics
in the texts of publications in our dataset. We visualized them using the following
procedure. First, for each of the topics in the dictionary, we created a sample of the
publications related to the topic. Next, we used a pre-trained word2vec model to
convert the text of each of the publications of our samples to a vector with 1536
components. Next, for each of the topics we computed the component average of
the vectors of all the documents related to the topic. Then, we used t-SNE to embed
the average vectors into a 2-dimensional space. This visualization (or topic map) is
presented on Fig. 4.
The sizes of the bubbles represent the number of documents in a dataset that are
related to each of the topics. As Fig. 4 shows, some topics are more prevalent in the
dataset than the others. We have determined that topics 159, 12, 17, 1 and 121 are
the most common in the full corpus.
Topic 159 is related to the corporation under investigation, its social media
accounts and publications, and social media discussions centered around its public
image. It should be noted that this topic was defined using a single word, namely the
name of the Company, so it is quite general and requires care when interpreted.

Fig. 4 Manual labeling model topic map

Computational Tools of Media Analysis … 67

Topics 12, 17 and 121 are related to each other. They describe a discourse centered
around one of the largest disasters that happened in the industry under investigation.
Topic 12 is more focused on the geographical area where the accident happened,
and the effects of the incident on the surrounding nature and environment. It also
contains publications dedicated to lives of those affected by the disaster. Topic 17
describes a discourse centered around both technical and cultural implications of the
accident and its containment, as well as related media. Topic 121 is focused on a
discourse surrounding the actual site of the accident, the present efforts for continued
containment, modern explorations of the site, and the ways the site of the accident
is represented in media.
Topic 1 describes a more general discourse that is focused on other disasters and
accidents related to the industry, and their cultural, environmental, and media impact.
It should be noted that the dictionary was constructed in a way that minimized the
intersection between this topic and topics 12, 17, and 121.

4 Comparing Three Different Topic Mining Algorithms:

Advantages and Disadvantages of Each Method
for Practical Problem-Solving in Business

Drawbacks and benefits of each method described above are summarized in Table 5.
This Table can be used as a so-called “roadmap” for building a system of monitoring
and evaluation of corporate policy effectiveness using computational tools of old and
new media data mining.
It should be noted that there are many possible avenues for improvement of the
analysis presented in this study. The first is to use other, more modern methods for
automated topic discovery. The methods used for this study only made use of the text
data available for publications but not the metainformation. For example, the topic
model proposed by [25] can take location and time data into account when modeling
topics from the corpus, and both time and location data is available in our dataset.
Another topic model proposed by [22] is a variant of LDA that can take connections
between documents into account. These data are also available in our dataset in the
form of connections between posts and reposts, posts and comments, comments and
replies to comments. It is entirely possible that including metadata into the topic
discovery process may allow for better, less correlated, and easier to interpret topics
to be discovered in our dataset.
Another possible avenue for improvement is through modifications of the manual
labeling topic discovery process. While it better than having the experts read millions
of publications one by one, the process is still laborious and time-consuming. One
possible way to improve this process is by using active learning. It may be possible to
start with an automatically generated grouping of publications. They can be generated
by one of the two methods. The first is via a clustering algorithm that allows for
detection of publications that do not belong to any of the clusters (e.g. DBSCAN).
68 G. S. Khvatsky et al.

Table 5 Comparison of topic modeling methods used in our study

LDA HDP Manual labeling
Determination of the The number of topics The number of topics The number of topics
number of topics is a hyperparameter, is inferred from the is inferred from the
for a large dataset, data, although the data by a panel of
determining the maximum number of experts, and can be
optimal number of topics is still a from 1 to an arbitrarily
topics requires a large hyperparameter of the large value. However,
amount of model the sampling approach
computational effort may still leave some
of the topics in the
data undiscovered by
the expert panel
Topic interpretability Discovered topics are Discovered topics are Topics are
not necessarily not necessarily human-defined, and
interpretable. An interpretable thus are interpretable
suboptimal choice of by definition
the number of topics
may further decrease
topic interpretability
Robustness to dirty The method is not The method is not The method is robust
data robust to dirty data. robust to dirty data. to dirty and unrelated
Presence of data not Presence of data not data in the sense that
related to the subject related to the subject the expert panel has a
of analysis in the of analysis in the means to move all the
dataset may dataset may negatively unrelated data to a
negatively affect the affect both the quality separate topics and
quality of discovered of discovered topics remove it from further
topics and the inferred analysis
number of topics
Inferring topics from This method allows This method allows for The method allows for
documents for inference, inference, however, assigning topics to
however, assigning assigning topics to individual documents.
topics to individual individual documents However, it does not
documents requires requires further allow for computation
further consideration consideration. In of a level of “affinity”
addition, difference of a document to a
between the maximum particular topic
number of topics and
the inferred number of
topics may further
complicate this
procedure
Using text structure to LDA is a HDP is a bag-of-words It is possible to
determine topics bag-of-words method method and does not integrate text structure
and does not take the take the structure of into the topic
structure of the text the text into account dictionary to a degree
into account
Computational Tools of Media Analysis … 69

The second is via a topic modeling algorithm. After automatic groups are generated,
the expert panel can work on interpreting and improving the discovered document
groupings.
It is also possible to combine both automated topic modeling approaches with
the manual labeling approach we applied in this study. This combined approach can
help with removing posts that are irrelevant for analysis, such as advertisements. It
can also help with manual labeling. As a result, it can help improve the performance
of the automated topic discovery models.
Continuing with recommendations of applying the methods described in this
chapter for future system of monitoring and evaluation of corporate policy effec-
tiveness, we would like to suggest the following improvements to our methodology.
Our first recommendation is to take special care when creating the search query to
obtain the initial dataset for analysis. Since the topic modeling methods we used
were affected by irrelevant publications obtained by the query we used, using a
better, narrower query has a large potential for improving the performance of all the
presented models. Another recommendation is to use an algorithm to determine the
number of topics for the LDA model. We did not use it due to lack of time (the project
had a tight deadline) and low computational power. However, in the future, using an
optimal number of topics for the LDA model will definitely improve the quality and
interpretability of the discovered topics.

Notes Funding: The article was prepared within the framework of the HSE University Basic
Research Program.

References

1. Allahyari M, Pouriyeh S, Kochut K, Arabnia HR (2017) A knowledge-based topic modeling

approach for automatic topic labeling. Int J Adv Comput Sci Appl 8(9):335–349. https://doi.
org/10.14569/IJACSA.2017.080947
2. Alnoukari M, Hanano A (2017) Integration of business intelligence with corporate strategic
management. J Intell Stud Bus 7(2). https://doi.org/10.37380/jisib.v7i2.235
3. Alnoukari M, Razouk R, Hanano A (2016) BSC-SI, a framework for integrating strategic
intelligence in corporate strategic management. Int J Strateg Inf Technol Appl (IJSITA) 7(1):32–
44. https://doi.org/10.4018/IJSITA.2016010103
4. https://builtin.com/big-data. What is big data?
5. Blei DM (2003) Latent Dirichlet allocation. J Mach Learn Res 3:993–1022
6. Boyd-Graber J, Blei DM (2010) Syntactic topic models. arXiv:1002.4665 [cs, math, stat].
http://arxiv.org/abs/1002.4665. Accessed 21 July 2021
7. Chaney A, Blei D (2012) Visualizing topic models. Proc Int AAAI Conf Web Soc Media
6(1):419–422
8. Chasin R et al (2014) Word sense disambiguation in the clinical domain: a comparison
of knowledge-rich and knowledge-poor unsupervised methods. J Am Med Inform Assoc
21(5):842–849. https://doi.org/10.1136/amiajnl-2013-002133
9. Cheng X et al (2014) BTM: topic modeling over short texts. IEEE Trans Knowl Data Eng
26(12):2928–2941. https://doi.org/10.1109/TKDE.2014.2313872
70 G. S. Khvatsky et al.

10. Drury G (2008) Opinion piece: social media: should marketers engage and how can it be done
effectively? J Direct Data Digit Mark Pract 9(3):274–277. https://doi.org/10.1057/palgrave.
dddmp.4350096
11. Girdhar Y, Giguère P, Dudek G (2013) Autonomous adaptive underwater exploration using
online topic modeling. In: Desai JP et al (eds) Experimental robotics: the 13th interna-
tional symposium on experimental robotics. Springer tracts in advanced robotics. Springer
International Publishing, Heidelberg, pp 789–802. https://doi.org/10.1007/978-3-319-00065-
7_53
12. Griffiths TL, Steyvers M (2004) Finding scientific topics. Proc Natl Acad Sci 101(suppl
1):5228–5235. https://doi.org/10.1073/pnas.0307752101
13. Haley KJ, Fessler DMT (2005) Nobody’s watching? Subtle cues affect generosity in an anony-
mous economic game. Evol Hum Behav 26(3):245–256. https://doi.org/10.1016/j.evolhumbe
hav.2005.01.002
14. Hannigan TR et al (2019) Topic modeling in management research: rendering new theory from
textual data. Acad Manag Ann 13(2):586–632. https://doi.org/10.5465/annals.2017.0099
15. Hong L, Davison BD (2010) Empirical study of topic modeling in Twitter. In: Proceedings of
the first workshop on social media analytics. Association for computing machinery (SOMA
’10), New York, NY, USA, pp 80–88. https://doi.org/10.1145/1964858.1964870
16. Hoxha G (2015) Limited efficiency of civil society during the democratization process of
Albania. Thesis. Epoka University. http://dspace.epoka.edu.al/handle/1/1762. Accessed 17
November 2020
17. Jelodar H et al (2019) Latent Dirichlet allocation (LDA) and topic modeling: models, appli-
cations, a survey. Multimed Tools Appl 78(11):15169–15211. https://doi.org/10.1007/s11042-
018-6894-4
18. Kaplan AM (2015) Social media, the digital revolution, and the business of media. Int J Media
Manag 17(4):197–199. https://doi.org/10.1080/14241277.2015.1120014
19. Kaplan AM, Haenlein M (2010) Users of the world, unite! The challenges and opportunities
of social media. Bus Horiz 53(1):59–68. https://doi.org/10.1016/j.bushor.2009.09.003
20. Lasswell HD (1970) The emerging conception of the policy sciences. Policy Sci 1(1):3–14
21. Lau JH et al (2012) Word sense induction for novel sense detection. In: Proceedings of the 13th
conference of the European chapter of the association for computational linguistics. EACL
2012. Association for computational linguistics, Avignon, France, pp 591–601. https://aclant
hology.org/E12-1060. Accessed 21 July 2021
22. Liu Y, Xu S (2017) A local context-aware LDA model for topic modeling in a document
network. J Am Soc Inf Sci 68(6):1429–1448. https://doi.org/10.1002/asi.23822
23. Magalhães PC (2014) Government effectiveness and support for democracy. Eur J Polit Res
53(1):77–97. https://doi.org/10.1111/1475-6765.12024
24. McCorkindale T, DiStaso MW, Carroll C (2013) The power of social media and its influence
on corporate reputation. In: The handbook of communication and corporate reputation, vol 9,
no. 1, pp 497–512
25. Mei Q, Zhai C (2006) A mixture model for contextual text mining. In: Proceedings of the 12th
ACM SIGKDD international conference on knowledge discovery and data mining—KDD ’06;
The 12th ACM SIGKDD international conference, Philadelphia, PA, USA. ACM Press, p 649.
https://doi.org/10.1145/1150402.1150482
26. Mimno D et al (2011) Optimizing semantic coherence in topic models. In: Proceedings of
the 2011 conference on empirical methods in natural language processing, EMNLP 2011.
Association for computational linguistics, Edinburgh, Scotland, UK, pp 262–272. https://acl
anthology.org/D11-1024. Accessed 13 July 2021
27. Mulhern F (2009) Integrated marketing communications: from media channels to digital
connectivity. J Mark Commun 15(2–3):85–101. https://doi.org/10.1080/13527260902757506
28. Niemi RG, Craig SC, Mattei F (1991) Measuring internal political efficacy in the 1988 national
election study. Am Polit Sci Rev 85(4):1407–1413. https://doi.org/10.2307/1963953
29. Pentina I, Tarafdar M (2014) From “information” to “knowing”: exploring the role of social
media in contemporary news consumption. Comput Hum Behav 35:211–223. https://doi.org/
10.1016/j.chb.2014.02.045
Computational Tools of Media Analysis … 71

30. Pritchard JK, Stephens M, Donnelly P (2000) Inference of population structure using multilocus
genotype data. Genetics 155(2):945–959
31. Qualman E (2012) Socialnomics: how social media transforms the way we live and do business.
Wiley
32. Ramamonjisoa D (2014) Topic modeling on users’s comments. In: 2014 third ICT international
student project conference (ICT-ISPC), pp 177–180. https://doi.org/10.1109/ICT-ISPC.2014.
6923245
33. Rani S, Kumar M (2021) Topic modeling and its applications in materials science and
engineering. Mater Today Proc 45:5591–5596. https://doi.org/10.1016/j.matpr.2021.02.313
34. Rayp G, Sijpe NVD (2007) Measuring and explaining government efficiency in developing
countries. J Dev Stud 43(2):360–381. https://doi.org/10.1080/00220380601125230
35. Segalovich I (2003) A fast morphological algorithm with unknown word guessing induced by
a dictionary for a web search engine. In: MLMTA, p 273
36. Tewolde MH, Gubán P (2010) the means of analysis and evaluation for corporate performances.
Ann Univ Apulensis Ser Oecon 1(12):738–749. https://doi.org/10.29302/oeconomica.2010.12.
1.43
37. Wang C, Blei DM (2011) Collaborative topic modeling for recommending scientific articles.
In: Proceedings of the 17th ACM SIGKDD international conference on knowledge discovery
and data mining. Association for computing machinery (KDD’11), New York, NY, USA, pp
448–456. https://doi.org/10.1145/2020408.2020480
38. Wang H et al (2019) Optimization of topic recognition model for news texts based on LDA. J
Digital Inf Manag 17(5):257. https://doi.org/10.6025/jdim/2019/17/5/257-269
39. Wen HJ, Lim B, Lisa Huang H (2003) Measuring e-commerce efficiency: a data envelopment
analysis (DEA) approach. Ind Manag Data Syst 103(9):703–710. https://doi.org/10.1108/026
35570310506124
40. Yau C-K et al (2014) Clustering scientific documents with topic modeling. Scientometrics
100(3):767–786. https://doi.org/10.1007/s11192-014-1321-8
41. Zhao W, Chen JJ, Perkins R, Liu Z, Ge W, Ding Y, Zou W (2015) A heuristic approach to
determine an appropriate number of topics in topic modeling. In: BMC Bioinf 16(13):1–10.
BioMed Central, 2015
Optimal Design of Checkpoint Systems
with General Structures, Tasks and
Schemes

Kenichiro Naruse and Toshio Nakagawa

Abstract This chapter proposes some kinds of checkpoint systems with general
structures, tasks and schemes. We have already considered redundancy techniques
which are duplex and majority systems, and have applied them to two checkpoint
models in which their interval times are constant and random. Giving overheads for
checkpoints, we have obtained the mean execution times until the process succeeds,
and have derived optimal checkpoint times to minimize them. In this chapter, we first
introduce the standard checkpoint model, and propose general checkpoint models
which include parallel, series and bridge systems. Furthermore, we consider tandem
and bulk tasks, and apply them to two schemes and compare optimal policies the-
oretically and numerically. Finally, as examples of the above models, we give four
models, obtain their mean execution times analytically and discuss which scheme is
better numerically.

1 Introduction

It is of great importance to develop the design of computer systems with high reliabil-
ity. Especially, human technologies have been recently growing up to build spacecraft
bound for Mars, International Space Station, auto-drive cars, and so on. All of them
mostly consist of computing units that need high reliability and high speed pro-
cessing. Therefore, we have to design previously such computer systems with high

K. Naruse (B)
Faculty of Social and Environmental Studies, Josai International University,
1 Gumyo, Togane, Chiba 283-8555, Japan
e-mail: knaruse@jiu.ac.jp
T. Nakagawa
Department of Business Administration, Aichi Institute of Technology,
1247 Yachigusa, Yakusa, Toyota, Aichi 470-0392, Japan
e-mail: toshi-nakagawa@aitech.ac.jp

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 73

M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_4
74 K. Naruse and T. Nakagawa

quality. However, we cannot eliminate some computer errors, and to mask these
errors, we have to design high reliable computers with multiple units and tolerant
performances.
Some errors may occur due to space radiation, electromagnetic waves, low qual-
ity of hardware, element overheat and over current, and so on. These errors lead
to faults or failures, and might cause serious damage to systems. To prevent such
faults, some types of fault-tolerant technologies including system redundancies and
configurations have been offered [1, 5, 13]. Using such techniques, we can achieve
high reliabilities and effective performances of objective computer systems in actual
fields.
Failures due to errors may be found and computer systems lose their consistency.
To protect such critical incidents, several recovery methods are required to restore
a consistent state just before failures. The most standard method is to taking copies
of normal states at suitable times, which is called checkpoint times. When errors of
the process occur and are detected, we execute the rollback operation to the nearest
checkpoint time and restore the consistent state of the process.
The simplest scheme for error detection recovery techniques is [6]: We execute two
independent modules which compare two states at checkpoint times. If two states of
each module do not match with each other, we go back to the newest checkpoint and
make their retrials. Furthermore, we recently considered six modules with duplex
and majority schemes, which consist of three clusters with two modules and two
clusters with three modules and compared them [9].
Several studies have been conducted to determine the optimal checkpoint fre-
quencies: The reliability and performance of a double modular system with one
spare module were evaluated [7, 11]. In addition, the performance of checkpoint
schemes with task duplication was evaluated [14, 15]. An optimal instruction retry
period that minimizes the possibility of dynamic failures using the triple modular
controller was derived [4]. Evaluation models with finite checkpoints and bounded
rollback were discussed [10].
In this chapter, we obtain the mean execution time to complete the computer
processing time for the standard checkpoint model, using renewal processes, and
derive optimal periodic checkpoint interval in Sect. 2. We make a generalization of
checkpoint models in Sect. 3: We consider a K-out-of-n structure in Sect. 3.1 and
obtain the reliability of general redundant systems which include parallel, series,
majority decision, and bridge systems in Sect. 3.2. We take up Tandem and Bulk tasks,
and compare which task is better numerically in Sect. 3.3. Section 3.4 introduces two
schemes of checkpoints such as compare-checkpoint and store-checkpoint. To apply
the above models to practical ones easily, we give four examples of random tasks in
Sect. 4: We consider a parallel structure with double modules in Sect. 4.1, 4 tasks with
3 schemes in Sect. 4.2, 6 tasks with 9 schemes with double modules in Sect. 4.3(1),
and triple majority modules in Sect. 4.3(2). We obtain the mean execution times of
each model, show their numerical examples, and discuss which model is better.
Optimal Design of Checkpoint Systems … 75

2 Standard Model

Suppose that S(0 < S < ∞) is a native execution time of processes and does not
include any checkpoint overheads. It is assumed that some errors occur according to
a general distribution F(t), and the failure rate is h(t) ≡ f (t)/F(t), where f (t) is a
density function of F(t) and Φ(t) ≡ 1 − Φ(t) for any function Φ(t). To detect such
errors, we divide S equally into N (N = 1, 2, . . .) time intervals and place periodic
checkpoints at planned times kT (k = 1, 2, . . .), where T ≡ S/N . If we detect errors
at checkpoint times, roll back to the previous checkpoint and re-execute the process,
where it is assumed that occur according to an identical distribution F(t) for each
checkpoint interval ((k − 1)T, kT ] and the failure rate h(t) increases with t.
Introduce constant overhead C1 of one checkpoint. Then, the mean time L(N ) to
complete the process with a native execution time S is the total of the execution times
and the overhead C1 for the checkpoint. From the assumption that if some errors are
detected at checkpoint times, the process is rolled back to the previous checkpoint,
the mean execution time for each checkpoint interval ((k − 1)T, kT ] is given in a
renewal equation [8, p. 126]:

L(1) = (T + C1 )F(T ) + [T + C1 + L(1)]F(T ). (1)

Solving (1) with respect to L(1),

T + C1
L(1) = . (2)
F(T )

Thus, the mean time to complete the process is

N (T + C1 ) S + N C1 S(1 + C1 /T )
L(N ) ≡ N L(1) = = = . (3)
F(T ) F(S/N ) F(T )

We find optimal N1∗ to minimize L(N ). For this purpose, we derive optimal
1 (0 < T
T 1 ≤ S) to minimize

 1 + C1 /T
L(T ) ≡ .
F(T )

Differentiating 
L(T ) with respect to T and setting it equal to zero,

h(T )(T 2 + C1 T ) − C1 = 0, (4)

whose left-hand side increases strictly with t from −C1 to ∞. Thus, there exists a
1 (0 < T
finite T 1 < ∞) which satisfies (4).
76 K. Naruse and T. Nakagawa

Therefore, we have the following optimal number N1∗ , using the partition method
[8, p. 42]:
(i) When T 1 < S, we set [S/T 1 ] ≡ N and calculate L(N ) from (3). If L(N ) ≤
L(N + 1), then N1 = N , and conversely, if L(N ) > L(N + 1), then N1∗ = N +
∗

1.
(ii) When T 1 ≥ S, N1∗ = 1, i.e., we do not make any checkpoint, and the mean time
is (S + C1 )/F(S).
Note that T 1 given in (4) does not depend on S. Thus, if S would be very large, be
changed greatly, or be uncertain, then we may adopt T 1 as an approximate checkpoint
time of T1∗ . For example, when S is a random variable with a distribution L(t) and
mean l,
l(1 + C1 /T )
L(N ) = .
F(T )

In particular, when F(t) = 1 − e−λt , (4) is

C1
T 2 + C1 T − = 0. (5)
λ
Solving (5) with respect to T ,
 
1 = C1 1
T 1+ −1 . (6)
λC1

If 1/λ > (S 2 /C1 ) + 2S, then N1∗ = 1.

3 General Checkpoint Models

3.1 General Stractures

We propose general structures with redundant modules to mask errors and detect them
at checkpoint times. As one of general structures, we consider the following K-out-
of-n system whose theoretical properties and practical applications were extensively
collected [12]: A K-out-of-n (1 ≤ K ≤ n) structure can operate if and only if at least
K modules of the total n units are operable, and their reliability characteristics were
investigated. When errors of each module occur according to an identical distribution
F(t), the reliability of the structure at time t is
n  
 n
R K (t) = F(t) j F(t)n− j . (7)
j
j=K
Optimal Design of Checkpoint Systems … 77

Furthermore, suppose that K is a random variable with a probability function

pk ≡ Pr{K = k}(k = 0, 1, 2, . . . , n) for a specified n(n = 1, 2, . . .) where p0 ≡
0 [2, 3]. It is denoted that the distribution of K is Pk ≡ Pr{K ≤ k} = kj=0 p j (k =
0, 1, . . . , n), where P0 ≡ 0, PN ≡ 1 and Pk increases with k from 0 to 1. Then, the
reliability of the structure at time t is, from (7),


n n  
 
n  
n n
R P (t) = pk F(t) j F(t)n− j = Pk F(t)k F(t)n−k , (8)
j k
k=0 j=k k=1

and the mean time is

∞ 
n   ∞
n
MP ≡ R P (t)dt = Pk F(t)k F(t)n−k dt. (9)
0 k 0
k=1

Note that Pk represents the probability that when k modules have no error, the struc-
ture is normal, i.e., the process executes correctly.

3.2 Examples of Redundant Modules

(1) Parallel Structure

When Pk = 1 (k = 1, 2, . . . , n),

R P (t) = 1 − F(t)n . (10)

(2) Series Structure

When Pk = 0 (k = 1, 2, . . . , n − 1) and Pn = 1,

R S (t) = F(t)n . (11)

(3) Majority Decision Structure

When n = 2m + 1 and k = m + 1, P j = 0 ( j = 1, 2, . . . , m) and P j = 1 ( j = m +

1, . . . , 2m + 1),


2m+1  
2m + 1
R M (t) = F(t)k F(t)2m+1−k . (12)
k
k=m+1

In particular, when m = 1,

R M (t) = 3F(t)2 − 2F(t)3 .

78 K. Naruse and T. Nakagawa

Fig. 1 Bridge structure with

5 modules with 2-terminal
network

(4) Bridge Structure

When the structure with 5 modules is given in Fig. 1, P1 = 0, P2 = 1/5, P3 = 4/5,
P4 = P5 = 1,
     
1 5 4 5 5
F B (t) = F(t)2 F(t)3 + F(t)3 F(t)2 + F(t)4 F(t) + F(t)5
5 2 5 3 4
= F(t)2 [1 + 2F(t) + F(t)2 − 2F(t)3 ]. (13)

Example 1 Consider a majority decision structure with 3 modules when F(t) = 1 −

e−λt . If more than two states of three modules agree with each other, then the process
is correct, i.e., the structure can mask a single module error. Then, the reliability of
the structure is, from (12),

R M (t) = 3e−2λt − 2e−3λt . (14)

Assuming that C2 is the overhead of a majority comparison with three modules,

the mean time to complete the process is, from (3),

S + N C2 S(1 + C2 /T )
L M (N ) = = −2λT ≡ L M (T ). (15)
3e−2λS/N − 2e −3λS/N 3e − 2e−3λT

Differentiating L M (T ) with respect to T and setting it equal to zero,

 
λT C2 C2
(e − 1) T + C2 T −
2
= , (16)
2λ 6λ

whose left-hand side increases strictly with T from 0 to ∞. Thus, there exists a finite
and unique T2 (0 < T
2 < ∞) which satisfies (16), and using the partition method in
Sect. 2, we obtain optimal N2∗ and T2∗ which minimize L M (N ) in (15). Therefore,
for any structures with reliability R P (t) in (8), we obtain the mean time in (3) and
discuss optimal policies to minimize it.
Table 1 presents optimal N2∗ and the mean execution time L M (N2∗ ) in (15) for λ
and C2 . For example, when S = 1.0, λ = 0.150, C2 = 0.05, optimal N2∗ is 3. Thus,
Optimal Design of Checkpoint Systems … 79

Table 1 Optimal N2∗ and mean time L M (N2∗ ) when S = 1.0, G(t) = 1 − e−t
λ C2 = 0.001 C2 = 0.005 C2 = 0.05
N2∗ L M (N2∗ ) N2∗ L M (N2∗ ) N2∗ L M (N2∗ )
0.150 5 1.008 3 1.022 3 1.117
0.125 4 1.007 3 1.020 1 1.092
0.100 4 1.006 2 1.017 1 1.077
0.075 3 1.005 2 1.014 1 1.066
0.050 2 1.004 2 1.012 1 1.057
0.025 2 1.002 1 1.007 1 1.052
0.001 1 1.001 1 1.005 1 1.050

when we make 3 chechkpoints until time S, it is optimal and the mean execution
time is 1.117. This indicates that N2∗ increase with λ from 1 and decrease with C2 ,
and L M (N2∗ ) increases with λ and C2 from S = 1.0. This means that if error rate λ
is small and overhead C2 is large, we should not place any checkpoints until time S.

3.3 Two Kinds of Random Tasks

(1) Tandem task

Suppose that the process executes N tasks (N = 1, 2, . . .), each of which has a
processing time Y j ( j = 1, 2, . . . , N ) with an identical distribution G(t) ≡ Pr{Y j ≤
t} with finite mean 1/θ and is executed successively, which is called Tandem task in
Fig. 2.
Supposing that T is a random variable with G(t) in Sect. 2, the mean time to
complete the process is, from (1),
∞
L T (N ) ≡ (t + N C1 ) F(t) + [t + N C1 + L T (N )] F(t) dG (N ) (t), (17)
0

where G (N ) (t) is the N -fold Stieltjes comvolution of G(t) with itself, and G (0) (t) ≡ 1
for t ≥ 0. Solving (17) with respect to L T (N ), the mean time to complete the process
is
N (1/θ + C1 )
L T (N ) = ∞ . (18)
(N ) (t)
0 F(t)dG

(2) Bulk task

Suppose that the process executes N tasks simultaneously, which is called Bulk
task in Fig. 3. If some tasks with errors are detected, then the process returns to the
80 K. Naruse and T. Nakagawa

Fig. 2 Tandem task

Fig. 3 Bulk task

previous checkpoint and executes all of N tasks, and ends when all of N tasks have
no errors, Then, letting C N be the overhead for N tasks, the mean time to complete
the process is
∞
L B (N ) = (t + C N )F(t) + [t + C N + L B (N )] F(t) dG(t) N . (19)
0

Solving (19) with respect to L B (N ),

∞  
1 − G(t) N dt + C N
L B (N ) = 0
∞ . (20)
0 F(t)dG(t) N

We compare L T (N ) and L B (N ) when C N = C1 N : From (18) and (20),

∞  
N (1/θ + C1 ) 1 − G(t) N dt + N C1
L T (N ) − L B (N ) = ∞ (N ) − 0
∞ .
0 G (t)dF(t) 0 G(t) N dF(t)
 
Noting that G(t) N ≥ G (N ) (t) (N = 1, 2, . . .) and 1 − G(t) N /N decreases with
N from G(t) to 0, we easily have
∞ ∞  
N
0 G(t) dF(t) (1/N ) 1 − G(t) N dt + C1
∞ (N ) ≥ 0
.
0 G (t)dF(t) 1/θ + C1

Thus, if C N = N C1 , then Bulk task is better than Tandem one, because we can begin
to execute N tasks simultaneously for Bulk one. However, overhead C N might be
larger than N C1 .
Optimal Design of Checkpoint Systems … 81

Table 2 Overhead C 2 and mean time L B (2) when N = 2, L T (2) = L B (2), C1 = 1 and G(t) =
1 − e−θ t and F(t) = 1 − e−λt
λ θ = 1.00 θ = 0.75 θ = 0.50 θ = 0.25
2
C L B (2) C2 L B (2) 2
C L B (2) 2
C L B (2)
0.1 2.690 4.840 2.958 5.994 3.545 8.640 5.667 19.600
0.05 2.598 4.410 2.817 5.310 3.286 7.260 4.909 14.400
0.01 2.520 4.080 2.698 4.792 3.059 6.242 4.196 10.816
0.005 2.510 4.040 2.682 4.729 3.030 6.121 4.099 10.404
0.001 2.502 4.008 2.670 4.679 3.006 6.024 4.020 10.080
0.0005 2.501 4.004 2.668 4.673 3.003 6.012 4.010 10.040
0.0001 2.500 4.001 2.667 4.668 3.001 6.002 4.002 10.008

Example 2 When N = 2, G(t) = 1 − e−θt and F(t) = 1 − e−λt ,

2 (1/θ + C1 ) 3/ (2θ ) + C2
L T (2) = , L B (2) = 2 .
[θ/ (θ + λ)] 2 2θ / [(θ + λ) (2θ + λ)]

Thus, if
2θ (λ + 2θ ) C2 − 8θ (θ + λ) C1 > 5λ + 2θ,

then Tandem task is better than Bulk one.

Table 2 presents overhead C 2 and mean time L B (2) for λ and θ when N = 2,
L T (2) = L B (2) and C1 = 1. For example, when λ = 0.05 and θ = 1.00, C 2 = 2.598

and L B (2) = L T (2) = 4.410. Values of C2 and L B (2) increase with λ and 1/θ . This
2 , then Tandem task is better than Bulk one.
indicates that if C2 > C

3.4 Two Schemes of Checkpoints

We consider the following three schemes of checkpoints [8, p. 138]: Store-checkpoint

(SCP) stores the state of the process, Compare-checkpoint (CCP) compares the state
of the process that is functioning when the structure consists of more than two
modules, and Compare-and-store checkpoint (CSCP) stores and compares the state
of the process.
Suppose that S is a native execution time, and to detect errors, we provide the
structure with two independent modules, where errors of each module occur at con-
stant λ, i.e., F(t) = 1 − e−λt . Then, the probability that any errors do not occur
in (0, t] is given by F(t)2 = e−2λt . In addition, introduce the overhead C2 for the
comparisons of two states and C S for the comparison and the store of them where
Cs ≥ C2 .
82 K. Naruse and T. Nakagawa

Fig. 4 Two schemes

Under the above assumptions, we consider two schemes in Fig. 4: In Scheme 1,

the mean execution time for each interval ((k − 1)T, kT ] is, from (1),

L 1 (1) = (T + C2 + C S )e−2λT + [T + C2 + L 1 (1)] (1 − e−2λT ). (21)

Solving (21) with respect to L 1 (1),

T + C2
L 1 (1) = + CS.
e−2λT
Thus, the mean time to complete the process is
 
S + N C2 1 + C2 /T CS
L 1 (N ) ≡ L 1 (T ) ≡ N L 1 (1) = −2λS/N + N C S = S + . (22)
e e−2λT T

In Scheme 2, the mean time for ((k − 1)T, kT ] is

l2 (k) = C2 + T + e−2λT l2 (k + 1) + (1 − e−2λT )l2 (1) (k = 1, 2, . . . , N − 1),

l2 (N ) = C2 + T + e−2λT C S + (1 − e−2λT )l2 (1). (23)

Solving (23) with respect to l2 (1),

 
(C2 + S/N ) 1 − e−2λS
L 2 (N ) ≡ L 2 (T ) ≡ l2 (1) =  + CS
1 − e−2λS/N e−2λS
 
(C2 + T ) 1 − e−2λS
=   + CS, (24)
1 − e−2λT e−2λS

where note that L 1 (S) = L 2 (S).

Optimal Design of Checkpoint Systems … 83

Table 3 Optimal N1∗ , N2∗ and mean time L 1 (N1∗ ), L 2 (N1∗ ) when S = 1.0, Cs = 0.1, G(t) = 1 −
e−t/μ
λ C2 = 0.001 C2 = 0.005 C2 = 0.001 C2 = 0.005
N1∗ L 1 (N1∗ ) N1∗ L 1 (N1∗ ) N2∗ L 2 (N2∗ ) N2∗ L 2 (N2∗ )
0.200 2 1.424 2 1.434 14 1.365 6 1.409
0.150 2 1.364 2 1.373 12 1.295 6 1.331
0.125 2 1.335 2 1.344 11 1.262 5 1.294
0.100 2 1.307 2 1.316 10 1.229 5 1.258
0.075 1 1.263 1 1.268 9 1.198 4 1.221
0.050 1 1.206 1 1.211 7 1.167 3 1.185
0.025 1 1.152 1 1.157 5 1.136 2 1.149
0.001 1 1.103 1 1.107 1 1.103 1 1.107

We find optimal T1∗ and T2∗ to minimize L 1 (T ) in (22) and L 2 (T ) in (24). Differ-
entiating L 1 (T ) with respect to T and setting it equal to zero,

e2λT [2λT (T + C2 ) − C2 ] = C S . (25)

Thus, there exists a finite and unique T1∗ (0 < T1∗ < ∞) which satisfies (25).
Differentiating L 2 (T ) with respect to T and setting it equal to zero,

1 2λT
(e − 1) − T = C2 . (26)
2λ
Thus, there exists a finite and unique T2∗ (0 < T2∗ < ∞) which satisfies (26). Using
the partition method in Sect. 2, we can get optimal N1∗ and N2∗ to minimize L 1 (N )
and L 2 (N ), respectively.

Example 3 Table 3 presents optimal N1∗ and N2∗ and the mean time L 1 (N1∗ ) in (22)
and L 2 (N2∗ ) in (24) for λ and C2 when Cs = 0.1 and S = 1. For example, when
λ = 0.100 and C2 = 0.001, N1∗ = 2 and N2∗ = 10, i.e., we should 1 CSCP in Scheme
1 and 9 CCP’s in Scheme 2. This indicates that both N1∗ and N2∗ increase with λ and
decrease with C2 .

4 Four Examples of Random Tasks

We can give four checkpoint models with random tasks by combining structures,
tasks and schemes.
84 K. Naruse and T. Nakagawa

4.1 Model 1

We adopt a parallel structure with 2 modules, N (N = 1, 2, . . .) tandem tasks, and

Schemes 1 and 2 in Fig. 4, in which errors of each module occur at constant λ, i.e.,
F(t) = 1 − e−λt , and each task has an identical distribution G(t). Then, for Scheme
1 in Fig. 4, the mean execution time of task k is, from (21),
∞ ∞
L 1 (1) = (t + C2 + C S )e−2λt dG(t) + [t + C2 + L 1 (1)] (1 − e−2λt )dG(t).
0 0
(27)
Solving (27) with respect to L 1 (1),

C2 + 1/θ
L 1 (1) = + CS, (28)
G ∗ (2λ)
∞ −st
where G ∗ (s) ≡ 0 e dG(t) for Re(s) ≥ 0. Thus, the mean time to complete the
process is  
C2 + 1/θ
L 1 (N ) ≡ N L 1 (1) = N + CS . (29)
G ∗ (2λ)

Similarly, for Scheme 2,

1 ∞ 
l2 (k) = C2 + + l2 (k)e−2λt + l3 (1)(1 − e−2λt ) dG(t) (k = 1, 2, . . . , N − 1),
θ 0

1 ∞ 
l2 (N ) = C2 + + C S e−2λt + l2 (1)(1 − e−2λt ) dG(t). (30)
θ 0

Solving (30) with respect to l2 (1),

(C2 + 1/θ ) 1 − [G ∗ (2λ)] N

L 2 (N ) ≡ l2 (1) = + CS. (31)
[1 − G ∗ (2λ)] [G ∗ (2λ)] N

Therefore, the mean execution time per one task is

 
l (1) (C2 + 1/θ ) 1 − G ∗ (2λ) N CS

L 2 (N ) ≡
2
= + . (32)
N ∗ ∗
N [1 − G (2λ)] [G (2λ)] N N

We find optimal N2∗ to minimize 

L 2 (N ). Forming the inequality 
L 2 (N + 1) −

L 2 (N ) ≥ 0,
N 
1   ∗ j Cs
N +1
1 − G (2λ) ≥ , (33)
2 + 1/θ
∗
[G (2λ)] C
j=1
Optimal Design of Checkpoint Systems … 85

Table 4 Optimal N2∗ and mean time 

L 2 (N2∗ ) and L 1 (N2∗ ) when Cs = 0.1 and G(t) = 1 − e−t
λ C2 = 0.001 C2 = 0.005 C2 = 0.05
∗
N2  ∗ ∗ ∗
L 2 (N2 ) L 1 (N2 ) N2 
L 2 (N2 ) L 1 (N2 ) N2∗
∗ ∗ L 2 (N2∗ ) L 1 (N2∗ )
0.1 1 1.301 1.301 1 1.306 1.306 1 1.360 1.360
0.05 1 1.201 1.201 1 1.206 1.206 1 1.255 1.255
0.01 3 1.075 3.363 3 1.079 3.375 3 1.126 3.513
0.005 4 1.051 4.444 4 1.055 4.460 4 1.102 4.642
0.001 10 1.022 11.030 10 1.026 11.070 10 1.072 11.521
0.0005 13 1.016 14.326 13 1.020 14.378 13 1.065 14.964
0.0001 30 1.007 33.036 30 1.011 33.156 30 1.057 34.506

whose left-hand side increases strictly with N to ∞. Thus, there exists a finite and
unique minimum N2∗ (1 ≤ N2∗ < ∞) which satisfies (33). If

1 − G ∗ (2λ) Cs
≥
∗
[G (2λ)] 2 C2 + 1/θ

then N2∗ = 1.

Example 4 Table 4 presents optimal N2∗ and the mean time  L 2 (N2∗ ) in (32) and
∗ ∗  ∗ ∗
L 1 (N2 ) in (29). Compute N2 and compare L 2 (N2 ) and L 1 (N2 ) for λ and C S when
C2 = 1 × 10−4 , Cs = 0.1 and 1/θ = 1. For example, when λ = 0.005, C2 = 0.001,
optimal N2∗ is 4. Thus, when we make 3 CCPs and 1 CSCP for 4 tasks, it is optimal
and the mean execution time is 1.051. Clearly, Scheme 2 is better than Scheme 1
with 4 tasks.

4.2 Model 2

We consider 3 schemes with 4 tasks in Fig. 5 for double modules, and discuss which
schemes are better.
When G(t) = 1 − e−θt and F(t) = 1 − e−λt , the mean execution times of each
scheme are obtained as follows: Let lk (k = 1, 2, 3, 4) be the mean time for task k to
complete the final task 4. For Scheme 1,

Fig. 5 4 tasks with 2 CSCP

86 K. Naruse and T. Nakagawa

∞ ∞  
l1 = (Cs + t + l2 ) e−2λt dG(t) + (C2 + t + l1 ) 1 − e−2λt dG(t),
0 0
∞ ∞  
l2 = (Cs + t) e−2λt dG (3) (t) + (C2 + t + l2 ) 1 − e−2λt dG (3) (t).
0 0

For Scheme 2,
∞ ∞  
l1 = (Cs + t + l3 ) e−2λt dG (2) (t) + (C2 + t + l1 ) 1 − e−2λt dG (2) (t),
0 0
∞ ∞  
l3 = (Cs + t) e−2λt dG (2) (t) + (C2 + t + l3 ) 1 − e−2λt dG (2) (t).
0 0

For Scheme 3,
∞ ∞  
l1 = (Cs + t + l4 ) e−2λt dG (3) (t) + (C2 + t + l1 ) 1 − e−2λt dG (3) (t),
0 0
∞ ∞  
l4 = (Cs + t) e−2λt dG(t) + (C2 + t + l4 ) 1 − e−2λt dG(t).
0 0

Solving the above equations with respect to l1 , respectively,

 
1/θ + C2 [1 − G ∗ (2λ)] 3/θ + C2 1 − G ∗ (2λ)3
L 1 = L 3 = 2Cs + + ,
G ∗ (2λ) G ∗ (2λ)3
   
2/θ + C2 1 − G ∗ (2λ)2
L 2 = 2 Cs + .
G ∗ (2λ)2

Comparing L 1 and L 2 , we easily have L 2 < L 1 because

1/θ + C2 3/θ + C2 2/θ + C2

+ ∗ > .
G ∗ (2λ) G (2λ)2 G ∗ (2λ)

This means that we should place the checkpoint at the middle point of the number
of tasks.

4.3 Model 3

(1) Double Modules

We consider 6 tasks with 9 schemes in Fig. 6. The mean execution times of each
scheme for 6 tandem tasks with a distribution G(t) when F(t) = 1 − e−λt are
obtained as follows:
Optimal Design of Checkpoint Systems … 87

Fig. 6 6 Tasks with CCP

and CSCP

C2 + 6/θ
L 1 = Cs − C2 + ,
G ∗ (2λ)6
 
C2 + 3/θ
L 2 = 2 Cs − C2 + ∗ ,
G (2λ)3
 
C2 + 2/θ
L 3 = 3 Cs − C2 + ∗ ,
G (2λ)2
 
C2 + 1/θ
L 4 = 6 Cs − C2 + ,
G ∗ (2λ)
 
(C2 + 3/θ) 1 + G ∗ (2λ)3
L 5 = Cs − C2 + ,
G ∗ (2λ)6
 
(C2 + 2/θ) 1 + G ∗ (2λ)2 + G ∗ (2λ)4
L 6 = Cs − C2 + ,
G ∗ (2λ)6
 
(C2 + 1/θ) 1 + G ∗ (2λ) + G ∗ (2λ)2 + G ∗ (2λ)3 + G ∗ (2λ)4 + G ∗ 2λ)5
L 7 = Cs − C2 + ,
G ∗ (2λ)6
   
(C2 + 1/θ) 1 + G ∗ (2λ) + G ∗ (2λ)2
L 8 = 2 Cs − C2 + ,
G ∗ (2λ)3
  
(C2 + 1/θ) 1 + G ∗ (2λ)
L 9 = 3 Cs − C2 + .
G ∗ (2λ)2

For examples, L 1 and L 7 are obtained as follows: Let lk (k = 1, 2, 3, 4, 5, 6)) be

the mean execution time for task k to complete the final task 6. For Scheme 1,
88 K. Naruse and T. Nakagawa

∞ ∞  
l1 = (Cs + t) e−2λt dG(t)6 + (C2 + t + l1 ) 1 − e−2λt dG(t)6
0 0
6  
= (Cs − C2 ) G ∗ (2λ)6 + C2 + + l1 1 − G ∗ (2λ)6 .
θ
Solving it with respect to l1 ,

C2 + 6/θ
L 1 ≡ l1 = Cs − C2 + .
G ∗ (2λ)6

For Scheme 7,
∞ ∞  
lk = (C2 + t + lk+1 )e−2λt dG(t) + (C2 + t + l1 ) 1 − e−2λt dG(t)
0 0
(k = 1, 2, 3, 4, 5),
∞ ∞  
l6 = (Cs + t)e−2λt dG(t) + (C2 + t + l1 ) 1 − e−2λt dG(t).
0 0

Solving it with respect to l1 ,

L 7 ≡ l1 = Cs − C2
 
(C2 + 1/θ ) 1 + G ∗ (2λ) + G ∗ (2λ)2 + G ∗ (2λ)3 + G ∗ (2λ)4 + G ∗ (2λ)5
+ .
G ∗ (2λ)6

Example 5 Table 5 presents L 1 ∼ L 9 for λ when Cs = 0.1, C2 = 0.001 and G(t) =

1 − e−t , using double module. When λ = 0.1, L 4 is the smallest and when λ = 0.001,
L 7 is the smallest.
Table 6 presents L 1 ∼ L 9 for C2 when Cs = 0.1 and G(t) = 1 − e−t , Comparing
Tables 5 and 6, when λ becomes smaller, L 1 ∼ L 9 becomes smaller, and when C2
becomes smaller, L 1 ∼ L 9 becomes smaller. Thus, if we improve the mean execution
time, we need to decrease an error rate λ and overhead C2 .

(2) Triple Majority Module

Consider a majority decision structure with 3 modules, i.e., 2-out-of-3 structure.

When more than two states of three modules agree with each other, the process in
this interval is correct. Then, the correct probability of the process in the interval
((k − 1)t, kt] is given in (14). Thus, replacing G ∗ (2λ) in L k (k = 1, 2, . . . , 9) with
∞  −2λt 
3e − 2e−3λt dG(t) = 3G ∗ (2λ) − 2G ∗ (3λ),
0
Optimal Design of Checkpoint Systems … 89

Table 5 Mean times of L 1 ∼ L 9 when Cs = 0.1, C2 = 0.001 and G(t) = 1 − e−t

λ L1 L2 L3 L4 L5 L6 L7 L8 L9
0.1 18.018 10.570 8.941 7.801 14.246 13.105 12.027 8.943 8.225
0.05 10.730 8.187 7.561 7.201 9.410 8.995 8.595 7.487 7.234
0.01 6.857 6.567 6.543 6.720 6.663 6.600 6.540 6.447 6.484
0.005 6.469 6.382 6.421 6.660 6.377 6.347 6.319 6.325 6.393
0.001 6.172 6.236 6.324 6.612 6.155 6.150 6.147 6.228 6.321
0.0005 6.136 6.218 6.312 6.606 6.128 6.126 6.126 6.216 6.312

Table 6 Mean times of L 1 ∼ L 9 when Cs = 0.1, λ = 0.05 and G(t) = 1 − e−t

C2 L1 L2 L3 L4 L5 L6 L7 L8 L9
0.05 10.768 8.219 7.592 7.230 9.513 9.164 8.962 7.746 7.427
0.01 10.737 8.193 7.566 7.206 9.429 9.026 8.662 7.535 7.269
0.005 10.733 8.189 7.563 7.203 9.418 9.009 8.625 7.508 7.250
0.001 10.730 8.187 7.561 7.201 9.410 8.995 8.595 7.487 7.234

we have easily L k . For example,

 
C2 + 2/θ
L 3 = 3 Cs − C2 + ,
[3G ∗ (2λ) − 2G ∗ (3λ)]2
 
(C2 + 1/θ ) [1 + 3G ∗ (2λ) − 2G ∗ (3λ)]
L 9 = 3 Cs − C2 + .
[3G ∗ (2λ) − 2G ∗ (3λ)]2

For general modules given in (8), we may replace G ∗ (2λ) with


n   ∞
n
Pk F(t)k F(t)n−k dG(t). (34)
k 0
k=1

Example 6 Table 7 presents L 1 ∼ L 9 for λ when Cs = 0.1, C2 = 0.001 and G(t) =

1 − e−t , using triple majority modules. Table 8 presents L 1 ∼ L 9 for C2 when Cs =
0.1 and G(t) = 1 − e−t . Compairing Tables 5 and 7, all of mean times in Table 5 are
larger than those of Table 7, however, when λ is small, the mean times of two tables
are almost the same. In general, C2 for triple modules would be larger than those for
double modules. Comparing Tables 6 and 8, all mean times of Table 8 are smaller
than those in Table 6. We should select a triple module under these conditions.
90 K. Naruse and T. Nakagawa

Table 7 Mean times L 1 ∼ L 9 when Cs = 0.1, C2 = 0.001 and G(t) = 1 − e−t for triple majority
modules
λ L1 L2 L3 L4 L5 L6 L7 L8 L9
0.1 7.692 6.949 6.790 6.840 7.272 7.136 7.004 6.697 6.668
0.05 6.545 6.419 6.445 6.672 6.433 6.397 6.362 6.349 6.412
0.01 6.121 6.210 6.307 6.603 6.116 6.116 6.117 6.211 6.308
0.005 6.105 6.203 6.302 6.601 6.105 6.106 6.108 6.206 6.304
0.001 6.100 6.200 6.300 6.600 6.101 6.102 6.105 6.204 6.303
0.0005 6.100 6.200 6.300 6.600 6.101 6.102 6.105 6.204 6.303

Table 8 Mean times L 1 ∼ L 9 when Cs = 0.1, λ = 0.05 and G(t) = 1 − e−t for triple majority
modules
C2 L1 L2 L3 L4 L5 L6 L7 L8 L9
0.05 6.549 6.422 6.448 6.676 6.487 6.502 6.620 6.552 6.564
0.01 6.546 6.419 6.446 6.673 6.443 6.416 6.410 6.387 6.440
0.005 6.546 6.419 6.445 6.672 6.437 6.405 6.383 6.366 6.424
0.001 6.545 6.419 6.445 6.672 6.433 6.397 6.362 6.349 6.412

5 Conclusion

We have proposed the optimal design of checkpoint systems with general structures,
tasks and schemes for random checkpoint models. In particular, we have considered
several useful checkpoint systems, obtained the mean time execution time and com-
pared them theoretically and numerically. It would be useful for systems designers
that need high reliability and high speed computing systems. When system designers
choose an appropriate system from this chapter, they will be able to make the opti-
mal checkpoint systems. Furthermore, if the systems failure rate are known, they can
easily choose an optimal kind of checkpoint schemes and get its checkpoint times.
This study can be applied to several systems with high reliable requirements such
as spacecraft, stock exchange system, aircraft, drone and so on, because such systems
need to provide high reliability and performance by implementing recovery methods
and redundant techniques.

References

1. Abd-El-Barr M (2007) Reliable and fault-tolerant. Imperial Colledge Press, London

2. Ito K, Nakagawa T (2019) Reliability properties of K-out-of-N: G systems. In: Ram M, Dohi
T (eds) Systems engineering. CRC Press, pp 25–40
3. Ito K, Zhao X, Nakagawa T (2017) Random number of units for K-out-of-n systems. Appl
Math Model 45:563–572
Optimal Design of Checkpoint Systems … 91

4. Kim H, Shin KG (1996) Design and analysis of an optimal instruction-retry policy for TMR
controller computers. IEEE Trans Comput 45(11):1217–1225
5. Lee PA, Anderson T (1990) Fault tolerance principles and practice. Dependable computing
and fault-tolerant systems. Springer, Wien
6. Nakagawa S, Fukumoto S, Ishii N (2003) Optimal checkpointing intervals of three error detec-
tion schemes by a double modular redundancy. Math Comput Model 38:1357–1363
7. Nakagawa S, Okuda Y, Yamada S (2003) Optimal checkpointing interval for task duplication
with spare processing. In: Ninth ISSAT international conference on reliability and quality in
design, Honolulu, Hawaii, vol 2003, pp 215–219
8. Nakagawa T (2008) Advanced reliability models and maintenance policies. Springer, London
9. Naruse K, Nakagawa T (2020) Optimal checkpoint intervals, schemes and structures for com-
puting modules. In: Pham H (ed) Reliability and statistical computing. Springer, pp 265–287
10. Ohara M, Suzuki R, Arai M, Fukumoto S, Iwasaki K (2006) Analytical model on hybrid state
saving with a limited number of checkpoints and bound rollbacks (reliability, maintainability
and safety analysis). IEICE Trans Fundam Electron Commun Comput Sci 89(9):2386–2395
11. Pradhan DK, Vaidya NH (1992) Rollforward checkpointing scheme: concurrent retry with
nondedicated spares. IEEE Computer Society Press, pp 166–174
12. Ram M, Dohi T (2019) Systems engineering: reliability analysis using k-out-of-n structures.
CRC Press
13. Siewiorek DP, Swarz RS (eds) (1982) The theory and practice of reliable system design. Digital
Press, Bedford, Massachusetts
14. Ziv A, Bruck J (1997) Performance optimization of checkpointing schemes with task duplica-
tion. IEEE Trans Comput 46:1381–1386
15. Ziv A, Bruck J (1998) Analysis of checkpointing schemes with task duplication. IEEE Trans
Comput 47:222–227
Series System Reliability: An Unified
Approach Using Fatal Shock and Stress
Models

Ricardo Puziol de Oliveira, Marcos Vinicius de Oliveira Peres, Wesley Bertoli,

and Jorge Alberto Achcar

Abstract This chapter introduces a new approach to estimate the reliability and
series system reliability functions based on the structure of shock and stress mod-
els. Conceptually, shock may refer to a deterioration process that destroys the entire
system. Conversely, stress can be related to two components kept working inde-
pendently, both having maintenance scheduled at a fixed time. In this context, we
introduce multivariate models derived from those structures to evaluate the reliability
function of an n-component series system.

1 Introduction

Systems’ reliability is not limited to the study of individual components’ reliabil-

ity. Better models that imply better inferences to study the reliability of component
systems, often structured in different configurations, are important, especially in
applications such as in all manufacturing sectors, engineering and management sec-
tors, [1–11]. As emphasized by [12], a complex system is often used to perform a
task that simpler systems could have done but with lower reliabilities.
Depending on the assumed structure, if one has many components working simul-
taneously, then the system’s overall reliability may decrease, and the maintenance
would be more complex. According to [13], the calculations to assess the reliability
of a system should be based on two essential operations:

R. P. de Oliveira (B) · M. V. de Oliveira Peres

Maringá State University, Maringá, Brazil
W. Bertoli
Federal University of Technology, Curitiba, Paraná, Brazil
e-mail: wbsilva@utfpr.edu.br
J. A. Achcar
University of São Paulo, São Paulo, Brazil
e-mail: achcar@fmrp.usp.br

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 93

M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_5
94 R. P. de Oliveira et al.

1. It is needed a reliability measurement, as precise as possible, of the components

used in the system environment;
2. It is also needed the calculation of the reliability of some complex combinations
of these components.
In this way, our primary goal is to obtain better inferences for system reliability
originated by block diagrams that depict the functional relationship of various ele-
ments in a physical system. For example, a block diagram having two blocks is called
a series system, usually assumed in many engineering experiments. Such a configu-
ration has the following property: if a single component fails, then the entire system
fails. In probabilistic terms, we say that each system-related component response is
represented by a random variable that could be binary (fail/no-fail) or positive real-
valued (lifetime). In such a context, it is possible to evaluate the system reliability
by estimating the overall probability of failure or the survival function [12, 14].
Several studies related to series system reliability usually assume that the compo-
nents’ lifetimes are not dependent. However, in many practical engineering situations,
such an assumption is far from reasonable. For instance, two specific system compo-
nents may be subject to the same set of stresses/fatal shocks or may receive the same
load/voltages. Besides, the components may share the same working environment,
so one may expect them to perform similarly. Thus, when obtaining the reliability
of 2-component systems, it is crucial adopting statistical models that accommodate
dependence structures between the components’ lifetimes (e.g., bivariate models)
[15]. In many practical applications, not considering the dependence as a source
of variation could affect the assessment of the entire system’s reliability [16–19].
Therefore, various bivariate distributions could be assumed for modeling correlated
lifetimes of a series system. Notably, the Bivariate Exponential distributions have
been widely used for estimating systems’ reliability [20–28].
A system’s reliability is typically obtained as the product of reliabilities at a fixed
time ti = min{ti } (i = 1, . . . , n). Under the independence assumption, and consid-
ering a 2-component series system as a special case, one can assume two random
variables T1 and T2 following the same distribution, but with different parame-
ters. Usually, T1 and T2 are univarite lifetimes of a series system and so we have
ti = min{t1i , t2i }.
Several lifetime models could be chosen for each component of a series sys-
tem. For example, T1 and T2 could be Weibull or Exponential random variables. A
mixture of these distributions (T1 following a Weibull and T2 following an Expo-
nential distribution) or a random field discretization can also be considered [29–41].
However, there may be a deterioration process damaging both components simulta-
neously. In this context, the lifetimes’ dependence cannot be ignored as the lifetimes’
dependency is strongly related to the deterioration process. In this case, it is highly
recommended to adopt a bivariate distribution for the inferential process.
In the present study, we introduce a time-dependent framework for a deteriora-
tion process using multivariate models for an n-component series system. Such an
approach can provide better estimates for the system reliability instead of adopting
independent univariate distributions for each component. Besides, since obtaining a
Series System Reliability: An Unified Approach … 95

system’s reliability function based on component reliabilities poses a great challenge,

using probabilistic models, especially multivariate ones, can be helpful to determine
system’ failure rates and warranty periods accurately.

2 Methodology

2.1 Series Systems with n Components

Firstly, we consider a 2-component series system whose components’ lifetimes are

given, respectively, by T1 and T2 . Thus, the system’s reliability function at a fixed
time t is given by
R(t) = P(min{T1 , T2 } > t),

where T1 and T2 represent the lifetimes of components 1 and 2, respectively.

Typically, the lifetimes T1 and T2 are correlated, which imply some implica-
tions when evaluating the system reliability [13, 31, 32, 42]. Under the dependence
assumption, the reliability function of the system at a fixed time t is given by

R(t) = P(T > t) = P(T1 > t, T2 > t) = S(t, t),

where S(t, t) denotes the joint survival function (sf) and T = min{T1 , T2 }, that is,

S(t, t) = 1 − F1 (t) − F2 (t) + F(t, t),

where F j (t j ) = P(T j  t j ) ( j = 1, 2) is the marginal cumulative distribution func-

tion (cdf) of T j and F(t, t) is the joint cdf. By the induction principle, the reliability
function of the system at a fixed time t, assuming n dependent components, is simply
given by

R(t) = P(T > t) = P(T1 > t, . . . , Tn > t) = S(t, . . . , t),

where S(t, . . . , t) represents the multivariate sf and T = min{T1 , . . . , Tn }. By ignor-

ing the dependence structure, one can simply evaluate the system reliability by
product-law of the reliabilities.

2.2 Fatal Shock and Stress Methods

The shock model structure introduced by [22] presupposes three existing independent
shock sources in the environment where a 2-component system is placed. Following
[43], such a structure is defined as follows.
96 R. P. de Oliveira et al.

Definition 1 ([43]) Suppose independent Poisson processes W1 (t, δ 1 ),

W2 (t, δ 2 ), and W3 (t, δ 3 ) represent the occurrence number of fatal shocks for
a 2-component system. Thus,
1. W1 (t, δ 1 ) represents the fatal shocks transmitted to component 1;
2. W2 (t, δ 2 ) represents the fatal shocks transmitted to component 2;
3. W3 (t, δ 3 ) represents the fatal shocks transmitted equally and independently
to both components.
If X = min{W1 , W3 } and Y = min{W2 , W3 } denote, respectively, the lifetimes
of components 1 and 2, then the probability of the system is working until the
occurrence of an overall failure is given by

P(X > x, Y > y) = P(W1 (x) = 0, W2 (y) = 0, W3 (max{x, y}) = 0)

= P(min{W1 , W3 } > x, min{W2 , W3 } > y)
= P({W1 > x, W3 > x}, {W2 > y, W3 > y})
= P(W1 > x, W2 > y, W3 > z),

where z = max{x, y}. Once W j ( j = 1, 2, 3) are mutually independent, we

have that

P(X > x, Y > y) = P(W1 > x)P(W2 > y)P(W3 > z).

in which the dependence structure of the random variables X and Y is related

to the common source of shock.

By the induction principle, we can easily generalize the Definition 1 to a n-

component series system. However, let us first introduce a 3-component system in
Definition 2 (see [43]). In this case, one can notice that seven Poisson processes
representing the fatal shocks. If we consider a 4-component system, we will have 15
Poisson processes for the shocks and so on. In general, for a n-component system,
we can assume 2n − 1 Poisson processes for the shocks. Such a quantity denotes the
number of extra model parameters that need to be estimated to characterize the full
system’s reliability.

Definition 2 ([43]) For a 3-component system and following the same struc-
ture for a 2-component series system, we have:
1. W1 (t, δ 1 ), W2 (t, δ 2 ), and W3 (t, δ 3 ) represents the fatal shocks, respectively,
to components 1, 2, and 3;
2. W12 (t, δ 12 ), W13 (t, δ 13 ), and W23 (t, δ 23 ) represents the fatal shocks, respec-
tively, to the component pairs (1, 2), (1, 3), and (2, 3);
Series System Reliability: An Unified Approach … 97

3. W123 (t, δ 123 ) represents an overall fatal shock to components 1, 2, and 3.

If X = min{W1 , W12 , W13 , W123 }, Y = min{W2 , W12 , W23 , W123 }, and Z =
min{W3 , W13 , W23 , W123 }, then the probability of the system is working until
the occurrence of an overall failure is given by

P(X > x, Y > y, Z > z) = P(W1 (x) = 0, W2 (y) = 0, W3 (z) = 0,

W12 (max{x, y}) = 0,
W13 (max{x, z}) = 0,
W23 (max{y, z}) = 0,
W123 (max{x, y, z}) = 0).

Once the random variables W j ( j = 1, 2, 3), Wi j (i = 1, 2), and W123 are

mutually independent, we have that

P(X > x, Y > y, Z > z) = P(W1 > x)P(W2 > y)P(W3 > z)
× P(W12 > max{x, y})
× P(W13 > max{x, z})
× P(W23 > max{y, z})
× P(W123 > max{x, y, z}).

Since in the n-dimensional space, similar arguments hold, we have that:

P(X 1 > x1 , . . . , X n > xn ) = P(W1 > x1 ) · · · P(Wn > xn )

× P(W12 > max{x1 , x2 })
× ...
× P(W123 > max{x1 , x2 , x3 })
× ...
× P(W12...n > max{x1 , . . . , xn })

For the structure of a stress model, also introduced by [12, 22], it is assumed that
a 2-component system is subject to individual and independent stresses, say U1 and
U2 . Besides, the authors suppose that the system has an overall stress source, U3 ,
which is transmitted to both components equally and independently. The stresses for
the components are X 1 = max{U1 , U3 } and X 2 = max{U2 , U3 }. Similar arguments
hold for the multivariate sf of the shock model. Here, the difference in the estimation
of the reliability is, in the case of 2-component, that the reliability function of the
system (T = min{X 1 , X 2 }), under the dependence assumption, is given by

R(t) = R1 (t)R2 (t) + R3 (t) − R1 (t)R2 (t)R3 (t),

98 R. P. de Oliveira et al.

where Ri (t) (i = 1, 2, 3) is the sf of Ui , so defining the stress of each component.

Noticeably, the system’s reliability functions based on stress models are very different
from those derived from fatal shock.

2.3 Inference Methods for a Series System

Without loss of generally, suppose a series system with two components. Naturally,
the procedure for a n-component system is analogous. In general, the existing liter-
ature on the reliability of 2-component series systems is based only on the failure
times and the indication of which component had failed. Specifically, we have the
complete lifetime data for one component but only partial information for the other
since it may not have failed yet. In this case, we have that T = min{T1 , T2 }.
Now, given a random sample (T1 , . . . , Tm ) of size m from a series system with
two components, one can define the following variable

1, for T1i < T2i
δi =
0, for T1i  T2i .

Hence, the individual contribution of T1i and T2i for the likelihood function of
vector β is given by

∂ S(t1i , t2i )
P(T1i = t1i , T2i > t2i ) = − , if δi = 1,
∂t1i

and
∂ S(t1i , t2i )
P(T1i > t1i , T2i = t2i ) = − , if δi = 0.
∂t2i

Therefore, the full likelihood function of vector β (see [44]) is given by

m    
∂ S(t1i , t2i ) δi ∂ S(t1i , t2i ) 1−δi
L(β) = − − . (1)
i=1
∂t1i ∂t2i

Note that Eq. (1) could have no compact form depending on the probability dis-
tributions adopted for each component. This implies that the maximum likelihood
estimator and the Fisher information matrix of β could only be obtained using opti-
mization algorithms. Nevertheless, from a fully Bayesian perspective, one may use
a Gibbs Sampling (GS) algorithm, which is a Markov Chain Monte Carlo (MCMC)
method (see [45, 46]) for obtaining posterior inferences.
Series System Reliability: An Unified Approach … 99

We have considered the GS algorithm to generate pseudo-random samples from

the joint posterior distribution of model parameters. After that, we have obtained the
Monte Carlo (MC) estimates under the squared error loss function. For instance, let
Zi = (X i , Yi ) be a Markov Chain and f (x, y) be a given joint probability density.
Also, let f (x; y) and f (y; x) be the conditional probability density functions. In this
case, the steps of the adopted GS algorithm are:
1. Choose initial values Z 0 = (X 0 , Y0 ) and set i = 1;
2. Generate X i ∼ f (xi ; Yi−1 = yi−1 ) and Yi ∼ f (yi ; X i = xi );
3. Set i = i + 1 and return to Step 2.
Using such a procedure, we have generated chains with N = 210,000 values
for each parameter, where the first 10,000 values were discarded as burn-in period.
The chains’ convergence was monitored using trace plots and Geweke’s diagnostic
methods. Then, to obtain pseudo-independent samples from the joint posterior dis-
tribution, for every 100 values generated, one was kept, which resulted in chains of
size 2,000 for each model parameter. The analysis were made in the R software [47]
using the R2jags package [48].

3 Probabilistic Models

This section illustrates the proposed methodology using 2-component systems

described in Definition 1 to derive probabilistic models that can be used to describe
a system’s reliability. The obtained models are presented in Tables 1 and 2.

Table 1 Models derived from the fatal shock structure

Fatal shock models (z = max{x1 , x2 })
Model Components S(x1 , x2 ) Marginal (S X j (x j ), j = 1, 2)
W1 ∼ E x p(β1 )
 
BE-I W2 ∼ E x p(β2 ) exp{−β1 x1 − β2 x2 − β3 z} exp −(β j + β3 )x j
W3 ∼ E x p(β3 )
W1 ∼ Ray(β1 )    
x12 x22
z2 β12 + β32
BR-I W2 ∼ Ray(β2 ) exp − 2 − − exp − x 2j
2β1 2β22 2β32 β12 β32
W3 ∼ Ray(β3 )
W1 ∼ Lin(β1 ) 
2  
βi xi βjxj
BL-I W2 ∼ Lin(β2 ) 1+ e−βi xi e−β3 z 1+ e−β j x j −β3 x j
1 + βi 1 + βj
W3 ∼ E x p(β3 ) i=1
100 R. P. de Oliveira et al.

Table 2 Models derived from the stress structure

Stress models (z = min{x1 , x2 })
Model Components F(x1 , x2 ) Marginal (FX j (x j ), j = 1, 2)
W1 ∼ E x p(β1 )

2

BE-II W2 ∼ E x p(β2 ) 1 − exp{−β3 z} 1 − exp{−βi xi } FE (xi , βi )FE (xi , β3 )
W3 ∼ E x p(β3 ) i=1

W1 ∼ Ray(β1 )    
z2 
2 x2
BR-II W2 ∼ Ray(β2 ) 1 − exp − 1 − exp − i 2 FR (xi , βi )FR (xi , β3 )
2β32 i=1
2βi
W3 ∼ Ray(β3 )
W1 ∼ Lin(β1 )  2 
 
β x
BL-II W2 ∼ Lin(β2 ) 1 − e−β3 z 1− 1+ i i e−βi xi FL (xi , βi )FE (xi , β3 )
1 + βi
W3 ∼ E x p(β3 ) i=1

4 Data Application: Operation Time of a Series System

This section presents an application of the proposed methodology using data from a
2-component series system. For each component, we assume that the failure times can
be approximated by an Exponential distribution with parameters β = 0.07 for the first
component and β = 0.05 for the second one. The system’s reliability was obtained
for an initial operation time of 50 h using the classical methodology, the product-law
of reliabilities (PLR), that is, assuming independence between the lifetimes [13].
Here, the main goal is to evaluate the system reliability using the proposed models
and the classical PLR methodology. In this way, we may verify if the proposed
models better accommodate the dependence structure related to a fatal shock or
overall stress that affects both components simultaneously. To perform a Bayesian
analysis, we have chosen noninformative Gamma prior distributions (α j = κ j =
0.001) for β j ( j = 1, 2) and an informative Gamma prior distribution with α3 = T
and κ3 = V(T ), where T is the random variable related to the operation time for
the fatal shock models, T is the sample mean and V(T ) is the sample variance.
Moreover, we have adopt a noninformative Uniform prior distributions, U (0, 10),
for β j ( j = 1, 2, 3) for the stress models to prevent computational instability. To
evaluate the series systems reliability, we have considered the parameter estimates to
compute the full system’s reliability for an initial operation time of 50 h. The obtained
results are presented in Figs. 1, 2 and 3.
From Figs. 1, 2 and 3, one may notice that only fatal shock models provide accurate
estimates for the system reliability compared to the PLR methodology. This result
may be related to the fact that we have considered a series system, and the results
could be different considering a parallel one; that is, maybe only the stress models
could provide similar estimates for the system reliability due to the structure from
which the stress models are derived.
One can notice that the BL-I model and the PLR method provide approximate
results, which can be related to the flexibility of the Lindley distribution on reliability
analysis (see [49]). For the BE-I model, the obtained result is quite similar to the
Series System Reliability: An Unified Approach … 101

1.0 PLR 1.0 PLR

BE−I BE−II

0.8 0.8
R(t)

R(t)
0.5 0.5
^

^
0.2 0.2

0.0 0.0
0 12 25 38 50 0 12 25 38 50
Time (Hours) Time (Hours)

Fig. 1 System reliability for an initial operation time of 50 h considering the BE-I (left-panel), the
BE-II (right-panel), and the PLR methodology

1.0 PLR 1.0 PLR

BR−I BR−II

0.8 0.8
R(t)

R(t)

0.5 0.5
^

0.2 0.2

0.0 0.0
0 12 25 38 50 0 12 25 38 50
Time (Hours) Time (Hours)

Fig. 2 System reliability for an initial operation time of 50 h considering the BR-I (left-panel), the
BR-II (right-panel), and the PLR methodology

1.0 PLR 1.0 PLR

BL−I BL−II

0.8 0.8
R(t)

R(t)

0.5 0.5
^

0.2 0.2

0.0 0.0
0 12 25 38 50 0 12 25 38 50
Time (Hours) Time (Hours)

Fig. 3 System reliability for an initial operation time of 50 h considering the BL-I (left-panel),
BL-II (right-panel), and the PLR methodology
102 R. P. de Oliveira et al.

PLR methodology, which is expected since we consider similar modeling structures.

The results based on the Rayleigh distribution are good, but such a model is not
always as flexible as the other ones. Despite the difference in the obtained results,
one can notice that both methodologies are useful for such data analysis. Besides,
the proposed models have the advantage of incorporating dependence structures: a
fatal shock or stress that damages both components of the system simultaneously.

5 Conclusion

In this chapter, we considered both stress and fatal shock structures. Based on the
obtained results, we can notice that the fatal shock model led to accurate inferential
results for the full system reliability, even using noninformative priors in the Bayesian
analysis. Also, as the real-data application showed, the proposed methodology could
be applied in many industrial situations where a series system fails due to a common
shock source that destroys both components. Based on the reliability estimates, it is
possible to conclude that using any of the proposed models could be an excellent
alternative to the PLR methodology under the independence assumption. Besides,
other probabilistic models could be assumed for the random variables Wi and Ui (i =
1, 2, 3), which would generate even more flexible models for accommodating other
information of the entire system.

References

1. Hall PL, Strutt JE (2003) Probabilistic physics-of-failure models for component reliabilities
using Monte Carlo simulation and Weibull analysis: a parametric study. Reliab Eng Syst Saf
80(3):233–242
2. Xie L, Zhou J, Hao C (2004) System-level load-strength interference based reliability modeling
of k-out-of-n system. Reliab Eng Syst Saf 84(3):311–317
3. Nandal J, Chauhan SK, Malik SC (2015) Reliability and MTSF evaluation of series and parallel
systems. Int J Stat Reliab Eng 2(1):74–80
4. Chauhan SK, Malik SC (2016) Reliability evaluation of series-parallel and parallel-series sys-
tems for arbitrary values of the parameters. Int J Stat Reliab Eng 3(1):10–19
5. Zhou W, Xiang W, Hong HP (2017) Sensitivity of system reliability of corroding pipelines to
modeling of stochastic growth of corrosion defects. Reliab Eng Syst Saf 167:428–438
6. Chi Z, Chen R, Huang S, Li YF, Bin Z, Zhang W (2020) Multi-state system modeling and
reliability assessment for groups of high-speed train wheels. Reliab Eng Syst Saf 107026
7. Okabe T, Otsuka Y (2020) Proposal of a validation method of failure mode analyses based on
the stress-strength model with a support vector machine. Reliab Eng Syst Saf 107247
8. Torres-Alves GA, Morales-Nápoles O (2020) Reliability analysis of flood defenses: the case
of the Nezahualcoyotl Dike in the Aztec City of Tenochtitlan. Reliab Eng Syst Saf 107057
9. Stefenon SF, Ribeiro MHDM, Nied A, Mariani VC, Coelho LS, Rocha DFM, Grebogi RB,
Ruano AEB (2020) Wavelet group method of data handling for fault prediction in electrical
power insulators. Int J Electr Power Energy Syst 123:106269
10. Ribeiro MHDM, Coelho LS (2020) Ensemble approach based on bagging, boosting and stack-
ing for short-term prediction in agribusiness time series. Appl Softw Comput 86:105837
Series System Reliability: An Unified Approach … 103

11. Yousefi N, Coit DW, Song S (2020) Reliability analysis of systems considering clusters of
dependent degrading components. Reliab Eng Syst Saf 202:107005
12. Oliveira RP, Achcar JA, Mazucheli J, Bertoli W (2021) A new class of bivariate Lindley
distributions based on stress and shock models and some of their reliability properties. Reliab
Eng Syst Saf 211:107528
13. Aggarwal KK (2012) Reliability engineering, vol 3. Springer Science & Business Media
14. Jensen PA, Bard JF (2003) Operations research models and methods, vol 1. Wiley
15. Thoft-Christensen P, Murotsu Y (1986) Reliability of series systems. In: Application of struc-
tural systems reliability theory. Springer, Berlin, Heidelberg
16. Li J, Coit DW, Elsayed EA (2011) Reliability modeling of a series system with correlated
or dependent component degradation processes. In: 2011 international conference on quality,
reliability, risk, maintenance, and safety engineering. IEEE, pp 388–393
17. Hu L, Yue D, Li J (2012) Availability analysis and design optimization for a repairable series-
parallel system with failure dependencies. Int J Innov Comput Inf Control 8(10):6693–6705
18. Park JH (2017) Time-dependent reliability of wireless networks with dependent failures. Reliab
Eng Syst Saf 165:47–61
19. Jafary B, Mele A, Fiondella L (2020) Component-based system reliability subject to positive
and negative correlation. Reliab Eng Syst Saf 107058
20. Gumbel EJ (1960) Bivariate exponential distributions. J Am Stat Assoc 55(292):698–707
21. Freund JE (1961) A bivariate extension of the exponential distribution. J Am Stat Assoc
56(296):971–977
22. Marshall AW, Olkin I (1967) A generalized bivariate exponential distribution. J Appl Probab
4(02):291–302
23. Marshall AW, Olkin I (1967) A multivariate exponential distribution. J Am Stat Assoc
62(317):30–44
24. Downton F (1970) Bivariate exponential distributions in reliability theory. J R Stat Soc Ser B
(Methodol) 32(3):408–417
25. Hawkes AG (1972) A bivariate exponential distribution with applications to reliability. J R Stat
Soc Ser B (Methodol) 129–131
26. Block HW, Basu AP (1974) A continuous, bivariate exponential extension. J Am Stat Assoc
69(348):1031–1037
27. Sarkar SK (1987) A continuous bivariate exponential distribution. J Am Stat Assoc
82(398):667–675
28. Arnold BC, Strauss D (1988) Bivariate distributions with exponential conditionals. J Am Stat
Assoc 83(402):522–527
29. Burr IW (1968) On a general system of distributions. J Am Stat Assoc 63
30. Burr IW (1973) Parameters for a general system of distributions to match a grid of α3 and α4 .
Commun Stat Theory Methods 2
31. Singh C, Billinton R (1977) System reliability, modelling and evaluation, vol 769. Hutchinson
London
32. Blanchard BS, Fabrycky WJ, Fabrycky WJ (1990) Systems engineering and analysis, vol 4.
Prentice Hall Englewood Cliffs, NJ
33. Chao MT, Fu JC (1991) The reliability of a large series system under Markov structure. Adv
Appl Probab 23(4):894–908
34. Mori Y, Ellingwood BR (1993) Time-dependent system reliability analysis by adaptive impor-
tance sampling. Struct Saf 12(1):59–73
35. Hulting FL, Robinson JA (1994) The reliability of a series system of repairable subsystems: a
Bayesian approach. Naval Res Logist 41(4):483–506
36. Zhang T, Horigome M (2001) Availability and reliability of system with dependent components
and time-varying failure and repair rates. IEEE Trans Reliab 50(2):151–158
37. Rausand M, Arnljot HÃ (2004) System reliability theory: models, statistical methods, and
applications, vol 396. Wiley
38. Kołowrocki K (2008) Reliability of large systems. Wiley Online Library
104 R. P. de Oliveira et al.

39. Eryilmaz S, Tank F (2012) On reliability analysis of a two-dependent-unit series system with
a standby unit. Appl Math Comput 218(15):7792–7797
40. Zhen H, Mahadevan S (2015) Time-dependent system reliability analysis using random field
discretization. J Mech Des 137(10):101404
41. Oliveira RP, Achcar JA (2019) Use of Basu-Dhar bivariate geometric distribution in the analysis
of the reliability of component series systems. Int J Quality Reliab Manag
42. Chowdhury S, Kundu A (2017) Stochastic comparison of parallel systems with log-Lindley
distributed components. Oper Res Lett 45(3):199–205
43. de Oliveira RP (2019) Multivariate lifetime models to evaluate long-term survivors in medical
studies. PhD thesis, University of São Paulo
44. Lawless JF (1982) Statistical models and methods for lifetime data. Wiley Series in probability
and mathematical statistics. Wiley, New York
45. Gelfand AE, Smith AFM (1990) Sampling-based approaches to calculating marginal densities.
J Am Stat Assoc 85(410):398–409
46. Chib S, Greenberg E (1995) Understanding the Metropolis-Hastings algorithm. Am Stat
49(4):327–335
47. R Core Team (2020) R: a language and environment for statistical computing. R foundation
for statistical computing, Vienna, Austria
48. Su Y-S, Yajima M (2012) R2jags: a package for running jags from R. R package version
0.03-08. http://CRAN.R-project.org/package=R2jags
49. Ghitany ME, Atieh B, Nadarajah S (2008) Lindley distribution and its application. Math Com-
put Simul 78(4):493–506
The New Attempt of Network Reliability
in the Assessment of Business Activities
for Precision Management

Shin-Guang Chen

Abstract Network reliability has been applied to many real-world applications. The
popular areas are traffic planning, computer network planning, power transmission
planning, etc. However, in the social applications, there are still very limited cases
reported. This chapter is planned to introduce the new attempt of applying network
reliability in the assessment of business activities. Such attempt is a step toward
precision management in business administration. Based on the capability of con-
ducting precision management, a great deal of improvements in business practices
can be taken place. Precision performance evaluation for individuals or groups are
presented, and a case study is illustrated to show the approach proposed in this
chapter.

1 Introduction

Network reliability has been applied to many real-world applications. The popular
areas are traffic planning, computer network planning, power transmission planning,
etc. Since 1954, the problem of maximum flow has attracted widespread attention in
the world [22]. They have also expanded to many other application areas [13]. Firstly,
the reliability problem of no flow in the binary state network is discussed by Aggarwal
et al. [3]. Lee [16] extended the problem to include cases with streams. The binary-
state cases for network reliability problem were solved firstly with minimum path
(MP) by Aggarwal et al. [2]. The multi-state cases for reliability network problem
were solved firstly by Xue [24]. The multi-state network (MSN) is a network whose
flow has a random state. Lin et al. [17] use MP or minimum cut (MC) [14] to illustrate
the reliability calculation of MSN. They also set up three stages in these calculations:
(a) Search for all MPs [10, 12, 23] / MCs [1, 8, 15, 28]; (b) Search for all d-MP
[19, 25] / d-MC [26] from these MP/MC; (c) From these d-MPs [6, 7, 30] / d-MCs

S.-G. Chen (B)

Tungnan University, New Taipei City, Taiwan
e-mail: bobchen@mail.tnu.edu.tw

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 105
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_6
106 S.-G. Chen

[12, 18], we calculate the joint probability of them. The staged approach (SA) was
stream-lined by Lin [19] and proposed a simpler and better method to evaluate the
reliability MSNs.
Now, SA has been extensively studied in the literature. A comprehensive discus-
sion can be referred to the article [4]. A good improvement in algorithms can be
referred to the article [11]. Based on the progresses, we can now apply the network
reliability to larger area such as social applications to show the impressive power of
network reliability.
The following sections are arranged as follows. The mathematical requirements
for SA are presented in Sect. 2. The details of SA [11] is given in Sect. 3. The per-
formance evaluation for individuals and groups in a process is given in Sect. 4. The
bonus dispatching in a process is given in Sect. 5. A case study is presented in Sect. 6.
Then, we draw the conclusion of this chapter in Sect. 7.

2 Mathematical Preliminaries

Assume that the process is a MSN, we have (E, V ) been a MSN, where E is the set
of edges representing the precedence relationships between individuals in a process,
V = {ai |1 ≤ i ≤ N } is the set of nodes representing the individuals in a process. Let
M = (m 1 , m 2 , . . . , m N ) be a vector with m i (an integer) being the maximal capacity
of ai . The process is assumed to satisfy the following assumptions.
1. The capacity of ai is an integer-valued random variable which takes values from
the set {0, 1, 2, …, m i } according to a given distribution μi .
2. The edges are perfect.
3. Flow in the process satisfies the flow-conservation law [13].
4. The states in nodes are statistically independent from each other.

2.1 Process Modeling

From the input to the output of the process, we call ρ1 , ρ2 , …, ρz the MPs. The process
is described by two variables at time t: the status variable St = (s1t , s2t , . . . , s N t ) and
the stream variable Ft = ( f 1t , f 2t , . . . , f zt ), where sit denotes the current capacity
state of ai at time t and f jt denotes the current work flow on ρ j at time t. Then, Ft
is feasible if and only if


z
{ f jt |ai ∈ ρ j } ≤ m i , for i = 1, 2, . . . , N . (1)
j=1

The above equation states that the total stream of ai have a maximal limit. Then,
we have κ Mt ≡ {Ft |Ft is feasible under M}.
The New Attempt of Network Reliability … 107

At the same time, Ft exists under St = (s1t , s2t , . . . , s N t ) if and only if


z
{ f jt |ai ∈ ρ j } ≤ sit , for i = 1, 2, . . . , N . (2)
j=1

For clarity, let κ St =

{Ft |Ft is feasible under St }. The maximal flow under St is
defined as  St ≡ max{ zj=1 f jt |F ∈ κ St }.
Ft ∈ κ M also exists such that the sum of streams in Ft equals d. We have the
following equation,
z
f jt = d. (3)
j=1

Let Ft = {Ft |Ft ∈ κ M and complies with Eq. (3)}. The below lemmas [19] hold.

Lemma 1 Let St be a d-MP, then Ft ∈ Ft exists such that


z
sit = { f jt |ai ∈ ρ j }, for each i = 1, 2, . . . , N . (4)
j=1

Given Ft ∈ Ft , a state vector S Ft = (s1t , s2t , . . . , s N t ) can be built by Equation

(4). The set t = {S Ft |Ft ∈ Ft } is created. Let min,t = {St |St is a minimum vector
in t }. We have

Lemma 2 min,t includes all the members of d-MPs at time t.

2.2 Performance Evaluation of a Process at Time t

Given a demand dt at time t, the reliability (ωdt ) is named the probability of the
network such that the maximal flow is greater than or equal to dt at time t, i.e., ωdt ≡
Pr{St | St ≥ dt }. For obtaining ωdt , we search the minimum vectors in {St | St ≥
dt }. A minimum vector (St ) is a d-MP iff (a)  St ≥ dt and (b) Wt < dt , ∀Wt = St
such that Wt < St , where Wt ≤ St iff w jt ≤ s jt for each j = 1, 2, . . . , n and Wt < St
if and only if Wt ≤ St and w jt < s jt for at least one j. Suppose there are totally q
d-MPs for dt : S1t , S2t , . . . , Sqt . The probability (i.e., the reliability) is derived as


q
ωdt = Pr{ {St |St ≥ Skt }}. (5)
k=1

This value can be derived by the amazing method in [6].

108 S.-G. Chen

3 The SA Approach

The SA approach includes three stages of computation [4]. Stage one is to search for
all MPs in a network. Stage two is to search for all d-MPs under the MPs. Stage three
is to calculate the union probability (the reliability) under all d-MPs. The following
subsections will briefly introduce the stages listed above.

3.1 Stage One—Searching for MPs

The most popular method to search for MPs can be referred to [10]. It can be shown
in the following.
Lemma 3 ρ j ∪ v creates a cycle if and only if ∃vk ∈ ρ j and vk = v.
Lemma 4 (Minimal Path detection) ρ is a MP iff ρ has no cycles, and the cor-
responding w = s ∪ v1 ∪ v2 ∪ . . . ∪ vk ∪ t, where vk ∈ B − (S ∪ T ), s ∈ S, and
t ∈ T.
The algorithm is briefly described as follows.

3.1.1 Algorithm (for Minimum Paths)

Since all undirected networks can be transformed to directed ones, the following
efficient algorithm [10] for directed networks to find MPs is introduced here.
Algorithm 1: // for minimum paths
1. Get L. // Get the network data.
2. While v ∈ L(0) = S, and v = φ, select a candidate.
3. R = v ∪ L − S. // Temporary list R (starting from 0).
4. Set j = 0, ρ = w = φ. // Initialization variables.
5. If j < 0, then ρ = ∪2≤l {ρ(l)}, Got MP and do backtrack.
6. Else begin v = R( j).
7. If w(k) = v holds, backtrack.
8. Else if j ∈ v and { j} = φ, ρ = { j} ∪ ρ, w = {v} ∪ w, and run Step 6.
9. Else do backtrack.
End.
end While.

3.1.2 Explanation Example

A popular ARPA network [29] is given in Fig. 1. Table 1 shows the results. The
efficiency of the approach is very good.
The New Attempt of Network Reliability … 109

4
a c
1 8
3 5
s 7 t

2 9
6
b d
Fig. 1 The ARPA network

Table 1 The MPs searched The proposed approach

MPs {s, b, a, c, t} {s, b, a, c, d, t}
{s, a, c, b, d, t} {s, b, d, c, t}
{s, a, b, d, c, t} {s, a, b, c, d, t}
{s, a, b, c, t} {s, a, b, d, t} {s, b, c, d, t}
{s, a, c, t} {s, b, c, d, t} {s, b, c, t}
{s, b, d, t}
Cycles {a, c, b} {a, c, d, b} {b, d, c}

3.2 Stage Two—Searching for d-MPs

An elegant method has been reported in [9]. This method results in the birth of a
better efficiency approach [11]. This subsection will briefly describe the approach in
[11]. A non-negative linear Diophantine equation has the following form:

c1 f 1 + c2 f 2 + · · · + cz f z = d, ci , f i , d ∈ Z+ . (6)

It can be shown that the following lemmas hold [9].

 
q+d−1
Lemma 5 For Eq. (6), if ci = 1, ∀i, the number of solutions is d , and the
 
q+d−1
complexity is O( d ).
 
q+d−1
Lemma 6 The number of solutions is no greater than d , and the complexity
 
q+d−1
is no greater than O( d ).

Lin and Chen [20] pointed out a very efficient method for d-MPs verification,
which was pioneered by Yeh [27]. This creative method can be used along with the
exact enumeration method, and creates a better method to find out d-MPs in the
network. According to Lin and Chen [20], the following lemmas hold.
110 S.-G. Chen

Property 1 ai is crossed by a path, then we have si > 0.

Lemma 7 ρ j ∪ γ creates a cycle if and only if ∃γk ∈ ρ j with γk = γ .

One can find all cycles or loops under Property 1 and Lemma 7. The following
lemmas also hold [27].

Lemma 8 S has no cycle, then S is a d-MP.

3.2.1 Algorithm for Searching d-MPs

Assume that ⊕ is a union operation for vectors. Let L be the network’s linked path
structure (LPS) [10]. Chen and Lin proposed the following algorithm.
Algorithm 2:
1. Let D = {1, 1, . . . , 1}.
2. Set r1 ⊕ r2 ⊕ . . . ⊕ rq ≡ D.
3. { f i | f i = #(ri ) for all i}. // #(ri ) is the count
 of 1’s. 
4. if F meets the equations as follows, zj=1 { f j |ai ∈ ρ j } ≤ m i , zj=1 f j = d,
for 1 ≤ i ≤ N ,

then si = zj=1 { f j |ai ∈ ρ j }, for 1 ≤ i ≤ N ,
5. for 1 ≤ i ≤ N do
6. if si > 0, then I = I ∪ {i}.
endfor
7. for i ∈ I do
8. λ = φ. // the path trace.
9. if not(L i ∈ λ), then keep on tracing L i until to the sinks, then continue, else
λ = λ ∪ Li .
10. else backtrack to Step 2.
endfor
11. min = min ∪ {S}.
12. Backtrack to Step 2.
13. Return. //Finish search.

3.2.2 Illustrative Examples

A simple network in Fig. 2 is presented for the explanation of the proposed approach.
Four MPs are found: ρ1 = {a1 , a3 }, ρ2 = {a2 , a4 }, ρ3 = {a1 , a5 , a4 }, and ρ4 = {a2 ,
a6 , a3 }. The corresponding information flows are F = { f 1 , f 2 , f 3 , f 4 }. The observed
data transmission distribution of the links are listed in Table 2. The demand of data
sink is set to 6.
The New Attempt of Network Reliability … 111

Fig. 2 A simple network Data source

a1

a6 a2
a3
a5

a4
Data sink

Table 2 The observed data transmission distributions of the links for Fig. 2
Arcs The distributions
0 1 2 3 4 5 6
a1 0.000 0.000 0.001 0.008 0.069 0.316 0.606
a2 0.000 0.000 0.006 0.061 0.309 0.624 0.000
a3 0.000 0.000 0.000 0.006 0.055 0.292 0.647
a4 0.000 0.000 0.002 0.019 0.114 0.369 0.497
a5 0.000 0.000 0.008 0.073 0.328 0.590 0.000
a6 0.000 0.000 0.001 0.008 0.069 0.316 0.606

Based on Table 2, the 6-MPs is described as follows: min = {S1 = (0, 6, 0, 6, 0,

0), S2 = (0, 6, 1, 5, 0, 1), S3 = (0, 6, 2, 4, 0, 2), S4 = (0, 6, 3, 3, 0, 3), S5 = (0, 6, 4,
2, 0, 4), S6 = (0, 6, 5, 1, 0, 5), S7 = (0, 6, 6, 0, 0, 6), S8 = (1, 5, 0, 6, 1, 0), S9 = (1,
5, 1, 5, 0, 0), S10 = (1, 5, 2, 4, 0, 1), S11 = (1, 5, 3, 3, 0, 2), S12 = (1, 5, 4, 2, 0, 3),
S13 = (1, 5, 5, 1, 0, 4), S14 = (1, 5, 6, 0, 0, 5), S15 = (2, 4, 0, 6, 2, 0), S16 = (2, 4,
1, 5, 1, 0), S17 = (2, 4, 2, 4, 0, 0), S18 = (2, 4, 3, 3, 0, 1), S19 = (2, 4, 4, 2, 0, 2),
S20 = (2, 4, 5, 1, 0, 3), S21 = (2, 4, 6, 0, 0, 4), S22 = (3, 3, 0, 6, 3, 0), S23 = (3, 3,
1, 5, 2, 0), S24 = (3, 3, 2, 4, 1, 0), S25 = (3, 3, 3, 3, 0, 0), S26 = (3, 3, 4, 2, 0, 1),
S27 = (3, 3, 5, 1, 0, 2), S28 = (3, 3, 6, 0, 0, 3), S29 = (4, 2, 0, 6, 4, 0), S30 = (4, 2,
1, 5, 3, 0), S31 = (4, 2, 2, 4, 2, 0), S32 = (4, 2, 3, 3, 1, 0), S33 = (4, 2, 4, 2, 0, 0),
S34 = (4, 2, 5, 1, 0, 1), S35 = (4, 2, 6, 0, 0, 2), S36 = (5, 1, 0, 6, 5, 0), S37 = (5, 1,
1, 5, 4, 0), S38 = (5, 1, 2, 4, 3, 0), S39 = (5, 1, 3, 3, 2, 0), S40 = (5, 1, 4, 2, 1, 0),
S41 = (5, 1, 5, 1, 0, 0), S42 = (5, 1, 6, 0, 0, 1), S43 = (6, 0, 0, 6, 6, 0), S44 = (6, 0,
1, 5, 5, 0), S45 = (6, 0, 2, 4, 4, 0), S46 = (6, 0, 3, 3, 3, 0), S47 = (6, 0, 4, 2, 2, 0),
S48 = (6, 0, 5, 1, 1, 0), S49 = (6, 0, 6, 0, 0, 0)}.
112 S.-G. Chen

3.3 Stage Three—Evaluating the Network Reliability via

Union Probability

Deriving the union probability of d-MPs is a way to obtain the network reliability. We
usually employ the inclusion-exclusion principle (IEP) to do so. It can be backtracked
from the idea of Abraham de Moivre (1718) [21]. To explain this concept, Eq. (7)
shows the way of IEP to obtain the probability of {A1 ∪ A2 ∪ A3 }.

Pr{A1 ∪ A2 ∪ A3 } = Pr{A1 } + Pr{A2 } + Pr{A3 }

− Pr{A1 ∩ A2 } − Pr{A2 ∩ A3 } (7)
− Pr{A1 ∩ A3 } + Pr{A1 ∩ A2 ∩ A3 }

By IEP, it is to sum up each entity probabilities, subtract the interception of paired

probabilities, and adds the intersection probability of all entities. By set theory,
the complexity of IEP equals to the complexity of power set expansion. That is
O(2q − 1). The Recursive Inclusion-Exclusion Principle (RIEP) can do a better job
than IEP just by rearranging Eq. (7) to its recursive form as Eq. (8). The complexity
is still O(2q − 1), but it has 10 times faster by compared to the normal IEP cases.

Pr{A1 ∪ A2 ∪ A3 } = Pr{A1 }
+ Pr{A2 } − Pr{A1 ∩ A2 }
+ Pr{A3 } − Pr{A2 ∩ A3 }
− Pr{A1 ∩ A3 } + Pr{A1 ∩ A2 ∩ A3 } (8)
= Pr{A1 }
+ Pr{A2 } − Pr{A1 ∩ A2 }
+ Pr{A3 } − Pr{A2 ∩ A3 ∪ A1 ∩ A3 }

Chen [6] reported a reduction approach to speed up the derivation of union proba-
bility. He emphasized the good application for monotonic cases such as the network
reliability problems. That is, the network reliability is to derive the union probability
of d-MPs.

3.3.1 The Approach of IEP

Given {A1 , A2 , …, An }. The union probability by IEP is such that

⎛ ⎞

n 
n
⎜  ⎟
Pr{ Ai } = ⎝(−1)
k−1
Pr{ A i }⎠ . (9)
i=1 k=1 I ⊂{1,...,n} i∈I
|I |=k
The New Attempt of Network Reliability … 113

+ +

Fig. 3 Rearrangement of RIEP

3.3.2 The Approach of RIEP

A conceptual diagram of RIEP is presented in Fig. 3. Let π(X 1 , X 2 , . . . , X n ) ≡

Pr{X 1 ∪ X 2 ∪ · · · ∪ X n }. Then, RIEP can be expressed by the following equation.

π(A1 ) = Pr{A
n 1 }, when n = 1,
π(A1 , A2 , . . . , An ) = i=1 (Pr{Ai } (10)
−π(A1 ∩ Ai , A2 ∩ Ai , . . . , Ai−1 ∩ Ai )), when n > 1,

3.3.3 The Approach of RRIEP

We firstly denote what is a reduction for the union events.

Definition 1 {X 1 , X 2 , . . . , X k } is a reduction of {Y1 , Y2 , . . . , Yn }, if for any k, n ∈

N and k < n such that {X 1 ∪ X 2 ∪ . . . ∪ X k } = {Y1 ∪ Y2 ∪ . . . ∪ Yn }.

One have the following lemmas hold.

Lemma 9 {X 1 , X 2 , …, X k } is a reduction of {Y1 , Y2 , …, Yn }, then π(X 1 , X 2 , . . . ,

X k ) = π(Y1 , Y2 , . . . , Yn ).

Lemma 10 The complexity of π(X 1 , X 2 , . . . , X k ) is no greater than or equal to the

complexity of π(Y1 , Y2 , . . . , Yn ).
114 S.-G. Chen

We said {B1 , B2 , . . . , Bki } a reduction of {A1 ∩ Ai , A2 ∩ Ai , . . . , Ai−1 ∩ Ai }. By

Lemma 9, Equation (10) is equal to

π(A1 ) = Pr{A1 }, when n = 1,

n
π(A1 , A2 , . . . , An ) = i=1 (Pr{Ai }
−π(A1 ∩ Ai , A2 ∩ Ai , . . . , Ai−1 ∩ Ai )), (11)
n
= i=1 Pr{Ai }
−π(B1 , B2 , . . . , Bki )), when n > 1,

Apparently, RRIEP is more efficient than RIEP. To do the trimming of union

entities, one can refer to the monotonicity property.

Property 2 Given Q̄ = (q̄1 , q̄2 , . . . , q̄h ) and W̄ = (w̄1 , w̄2 , . . . , w̄h ) the state vec-
tors. Q̄ and W̄ comply with the following rules:
1. Q̄ ≤ W̄ if and only if q̄ j ≤ w̄ j for each j = 1, 2, . . . , h.
2. Q̄ < W̄ if and only if Q̄ ≤ W̄ and q̄ j < w̄ j for at least one j.

3.3.4 The Algorithm (for Union Probability)

Let Q̄ ⊕ W̄ ≡ {ḡi |ḡi = max(q̄i , w̄i ), q̄i ∈ Q̄, w̄i ∈ W̄ , ∀i}. ωdt can be derived by
the following steps.
Algorithm 3: // for union probability
1. d M P = {X 1 , X 2 , ...,X h }.
2. ω = 0.
3. If h = 0, then return ω,
4. For 1 ≤ i ≤ |d M P| do
5. ω2 = Pr {X ≥ X i }.
6. For 1 ≤ j < i do
7. X j,i = X j ⊕ X i .
End (for).
8. Let E = {X j,i |1 ≤ j < i} and Wl , Wk ∈ E.
9. For k ∈ / J and 1 ≤ k ≤ ||E|| do // Reduction process.
10. For l ∈ / J and k < l ≤ ||E|| do
11. If Wl ≤ Wk , then J = J ∪ {k} and go to Step 9.
Else, if Wl > Wk , then J = J ∪ {l}.
End (for).
12. d M P ∗ = d M P ∗ ∪ {Wk }.
End (for).
13. Run Step 1 employing d M P ∗ and got ω3 .
14. ω = ω + ω2 − ω3 .
End (for).
15. got ω.
The New Attempt of Network Reliability … 115

Table 3 The computation comparison (sec)

# of d-MPs IEP RIEP RRIEP
15 1.765 0.390 0.0003
18 19.891 1.891 0.0005
20 75.953 2.953 0.016
25 9256.750 9.359 0.016
30 –∗ 43.343 0.015
35 – 106.131 0.035
40 – 415.580 0.054
45 – 1109.876 0.066
50 – 2134.090 0.083
55 – 5884.508 0.108
59 – 8652.496 0.110
∗ Not stopped over 1 day

The reduction of union entities is begun in Step 9. Because RRIEP always

decreases the computation efforts in most cases, it is a good approach to shorten
the computation time for most applications. However, in the worst cases, it still
needs O(2 Q − 1) to obtain the reliability.

3.3.5 Explanation Example

One randomly generates vectors of length 20 as d-MPs, and makes comparison

with well-known methods. The results are listed in Table 3. We run these cases by
PROLOG on a laptop.

4 The Assessment of Individuals and Groups

When we think of MSN as a process network, we get precision tools to achieve

precision management practices. The first step is how to assess the performance of
the individuals in a process. This can be done by modeling the process by a MSN
such that the individuals are nodes, and their precedence relationship in the process
are edges. The work flow through the network can reflect the performance of the
process. Then, we can define the following indicators to reflect the performance of
individuals in a process.
116 S.-G. Chen

4.1 The Performance Indicators for Individuals

Since St = {s1t , s2t , . . . , s N t } are the capacity vector for the individuals at time t,
we can obtain the empirical throughput distribution t = {μ1t , μ2t , . . . , μ N t ]}. The
definition of the expected throughput at time t for individual is as follows.

ik |1≤k≤t}
max{s
E it = jμit ( j). (12)
j=1

Further, we can define the individual importance in the process. This can be done
by defining the covering set of the individual ai in the process. Let P j = {ρ j |ai ∈ ρ j }
be the MP set involving ai . The covering set ζi of ai is defined by the following.

Definition 2 Assume ai , a j ∈ V . j is in ζi iff P j ⊆ Pi .

Then, we define the individual importance ϒi by the following.

||ζi ||
ϒi = , (13)
N
where || • || is the number of the member in a set.
By the above definition, we observe the fact that if ϒi = 1.0, it means it is a
critical individual in the process, since it cover all the flows in the process, and ϒi
cannot be 0, because it must at least cover itself.
We can also define the individual contribution in the process. This can be done
by defining the Survivability ωdt,i of ai in the process.

Definition 3 The Survivability ωdt,i of ai is the process reliability at time t when ai

is unavailable (i.e., μit (0) = 1).

Then, we define the individual contribution Ci by the following.

t
j=1 (ωd j − ωd j,i )
Ci = t . (14)
j=1 ωd j

By the above definition, we observe the fact that if Ci = 1.0, it means ai having
most real contribution in the process, since all the time, the flows in the process
cannot exist without him. On the contrary, if Ci = 0.0, it means ai having no real
contribution in the process, since all the time, the flows in the process did not be
influenced without him.
The New Attempt of Network Reliability … 117

4.2 The Performance Indicators for Groups

Let the process output at time t be t , and the empirical distribution during this
period is μt , then the throughput of the group during this period is defined as the
expected value during this period:


max{i |1≤i≤t}

E t = jμt ( j). (15)

j=1

4.3 The Assessments

Along with the above definitions, we get the following assessments for individuals
as well as groups.

4.3.1 The Assessment of Individual Performance

Let the individual standard output δit at time t relative to the standard output dt , then
the performance appraisal λit of individual ai can be calculated by the following
formula:
E it
λit = t . (16)
j=1 δi j
t

Therefore, when λit > 1, the individual has over-standard performance; when
λit = 1, the individual has standard performance; when λit < 1, the individual has
below-standard performance.

4.3.2 The Assessment of Group Performance

The group performance assessment λt can be calculated using the following formula:

E t
λt = t . (17)
i=1 di
t

So when λt > 1, the group has over-standard performance; when λt = 1, the group
has standard performance; when λt < 1, the group has below-standard performance.
118 S.-G. Chen

5 The Bonus Dispatching for the Process

The purpose of bonus dispatching is to distribute the profit of the process at time t
fairly to all individuals in the process. Therefore, there are two calculation methods.
1. Taking the individual contribution as a coefficient, the following formula can be
used to calculate:
i1 = λit Ci . (18)

This method will be affected by the individual capacity, that is, if the individual
is not properly allocated, some bottlenecks will appear. This kind of allocation
can reflect the bottleneck contribution.
2. With individual importance as a coefficient, the following formula can be used to
calculate:
i2 = λit ϒi . (19)

This method will not be affected by the individual capacity, but only reflects the
importance of the process position, that is, the dynamic bottleneck contribution
will not be reflected.

6 The Case Studies

Figure 4 is the network of a distribution process in a well-known company. The

standard total output is dt = 4 deliveries at time t per unit time, t = 1, 2, . . . , 10.
The detailed information is listed in Table 4.
According to the above formulas, the group and individual performance indicators
at time t = 10 are shown in Table 5.

a2 Transporter
Salesman 1 a4

a1

Order Entry a5
Accountant
a3
Salesman 2
Fig. 4 A distribution process network in a well-known company
The New Attempt of Network Reliability … 119

Table 4 The initial throughput of the distribution process

t a1 a2 a3 a4 a5 λt ωdt
1 6 0∗ 3 2 5 5 0.1230
2 5 3 2 3 4 4 0.1728
3 5 3 3 3 5 5 0.3087
4 0 3 2 3 5 5 0.4368
5 6 3 0 3 3 3 0.3479
6 5 3 3 3 6 6 0.4928
7 5 3 3 3 6 6 0.6561
8 4 2 1 3 0 0 0.5616
9 5 3 2 0 6 6 0.5688
10 4 3 3 3 6 6 0.5760
Si 1.0 0.2 0.2 0.2 1.0
Ci 1.0 1.0 1.0 0.07 1.0
∗ Absence

Table 5 The group and individual performance indicators when t = 10

Items a1 a2 a3 a4 a5 E t λt ω4t
δit 4 2 2 2 4
E it 4.5 2.6 2.2 2.6 4.6
λit 1.13 1.3 1.1 1.3 1.15 4.6 1.15 0.58
i1 1.13 1.3 1.1 0.09 1.15
i2 1.13 0.33 0.28 0.33 1.15

6.1 Discussions of the Case Study on Precision Management

The following features can be learned from the results of the case study on precision
management.
1. The case of this chapter involves the random absence of individual.
2. The performance of individuals in this case has exceeded their standard perfor-
mances, but the probability of achieving the corporate goal is only 0.58. Therefore,
there is a great risk of process shutdown. For example, if someone is absent for a
long time, the process may be shut down.
3. According to the indicators in this case, it can be seen that the risk of the case is due
to improper capacity planning of individuals. It is known from the Ci index that
everyone except a4 has the greatest substantial contribution, which is the reason
for the high risk. According to Chen’s optimal capacity planning scheme [5], the
standard output of a2 , a3 , and a4 should be 4, 4, 1. Therefore, the two salesmen
activities must increase manpower, and the transporter only needs to select the
least capable person. Even abolishing a4 will not affect the overall performance
too much.
120 S.-G. Chen

4. It would be unfair to implement incentive measures in accordance with individual

performance λit . Because the truly important roles such as a1 and a5 are not
highlighted, while those with no substantive contributors such as a4 receive the
highest rewards. Fairer reward indicators can be i1 and i2 , the former can be
fairly rewarded based on the actual contribution; the latter can highlight the real
importance in the process.
5. The group performance has exceeded the standard performance by 15%, and
individual can be rewarded according to this ratio.

7 Conclusion

From the perspective of network reliability, this chapter makes an attempt to assess
the process performance by network reliability for precision management. One of
them is done by objectively evaluating the overall process performance, as well as
analyzing the different importance and contribution of each individual in the process.
The results show that the network reliability is very suitable to assess social activities
with process in them.
By the case study, we found that the improvement of process can be facilitated
by precising analysis of the performance of individuals as well as groups. This is the
basis of so-called precision management of business activities.
Future researches can be conducted by inspecting multi-commodities in a process
to show the merits of the theories in network reliability.

Acknowledgements This paper was supported in part by the Ministry of Science and Technology,
Taiwan, Republic of China, under Grant No. MOST 107-2221-E-236-004-MY3.

References

1. Abel U, Bicker R (1982) Determination of all minimal cut-sets between a vertex pair in an
undirected graph. IEEE Trans Reliab R-31:167–171
2. Aggarwal KK, Chopra YC, Bajwa JS (1982) Capacity consideration in reliability analysis of
communication systems. IEEE Trans Reliab 31:177–80
3. Aggarwal KK, Gupta JS, Misra KB (1975) A simple method for reliability evaluation of a
communication system. IEEE Trans Commun 23:563–565
4. Chen SG (2020) Computation in network reliability. In: Pham H (ed) Reliability and statistical
computing. Springer, pp 107–126
5. Chen SG (2012) An optimal capacity assignment for the robust design problem in capacitated
flow networks. Appl Math Model 36(11):5272–5282
6. Chen SG (2014) Reduced recursive inclusion-exclusion principle for the probability of union
events. In: 2014 IEEE international conference on industrial engineering and engineering man-
agement, Selangor, Malaysia, pp 1–3
7. Chen SG (2014) Reduced recursive sum of disjoint product in network reliability. In: 2014 the
20th ISSAT international conference on reliability and quality in design, Seattle, Washington,
USA, pp 170–173
The New Attempt of Network Reliability … 121

8. Chen SG (2015) Search for all MCs with backtracking. Int J Reliab Qual Perform 6(2):101–106
9. Chen SG (2018) A novel approach to search for all solutions of a non-negative linear Diophan-
tine equation. In: 24th ISSAT international conference on reliability and quality in design, pp
55–57
10. Chen SG, Lin YK (2012) Search for all minimal paths in a general large flow network. IEEE
Trans Reliab 61(4):949–956
11. Chen SG, Lin YK (2020) A permutation-and-backtrack approach for reliability evaluation in
multistate information networks. Appl Math Comput 373:125024
12. Colbourn CJ (1987) The combinatorics of network reliability. Oxford University Press, UK
13. Ford LR, Fulkerson DR (1962) Flows in networks. Princeton University Press, NJ
14. Jane CC, Lin JS, Yuan J (1993) On reliability evaluation of a limited-flow network in terms of
minimal cutsets. IEEE Trans Reliab 42:354–361, 368
15. Jasmon GB, Foong KW (1987) A method for evaluating all the minimal cuts of a graph. IEEE
Trans Reliab R-36:538–545
16. Lee SH (1980) Reliability evaluation of a flow network. IEEE Trans Reliab 29:24–26
17. Lin JS, Jane CC, Yuan J (1995) On reliability evaluation of a capacitated-flow network in terms
of minimal pathsets. Networks 25:131–138
18. Lin YK (2001) On reliability evaluation of a stochastic-flow network in terms of minimal cuts.
J Chin Inst Ind Eng 18:49–54
19. Lin YK (2001) A simple algorithm for reliability evaluation of a stochastic-flow network with
node failure. Comput Oper Res 28(13):1277–1285
20. Lin YK, Chen SG (2019) An efficient searching method for minimal path vectors in multi-state
networks. Ann Oper Res 1–12
21. Roberts FS, Tesman B (2009) Applied combinatorics, 2nd edn. CRC Press
22. Schrijver A (2002) On the history of the transportation and maximum flow problems. Math
Program 91(3):437–445
23. Shen Y (1995) A new simple algorithm for enumerating all minimal paths and cuts of a graph.
Microelectron Reliab 35:973–976
24. Xue J (1985) On multistate system analysis. IEEE Trans Reliab 34(4):329–337
25. Yeh WC (2001) A simple algorithm to search for all d-MPs with unreliable nodes. Reliab Eng
Syst Saf 73:49–54
26. Yeh WC (2001) A simple approach to search for all d-MCs of a limited-flow network. Reliab
Eng Syst Saf 71:15–19
27. Yeh WC (2002) A simple method to verify all d-minimal path candidates of a limited-flow
network and its reliability. Int J Adv Manuf Technol 20(1):77–81
28. Yeh WC (2006) A simple algorithm to search for all MCs in networks. Eur J Oper Res 174:1694–
1705
29. Yeh WC (2007) A simple heuristic algorithm for generating all minimal paths. IEEE Trans
Reliab 56(3):488–494
30. Zuo MJ, Tian Z, Huang HZ (2007) An efficient method for reliability evaluation of multistate
networks given all minimal path vectors. IIE Trans 39:811–817
Optimal Design of Reliability Acceptance
Sampling Plans for Multi-stage
Production Process

M. Kumar

Abstract One of the goals in manufacturing industry is to follow manufacturing

standards which ensure that the manufactured products meet expectations of con-
sumers. An Acceptance sampling plan is a tool to ensure that quality of products
meet the minimum expected standards. In this chapter, an attempt is made to derive
lot acceptance single and double sampling plans based on type ii censored data.
The units in the lot have multiple quality characteristics and are processed through
multi-stage process. We assume that the quality characteristics of units follow the
exponential distribution. The acceptance criterion for the given lot is derived based
on mean-life of units in the sample at every stage. Further, two non-linear opti-
mization problems, which minimize the expected total testing cost at the acceptable
quality level, are solved. In addition, sensitivity analysis studies are also conducted
to assess the behavior of total testing costs with respect to change in producer’s and
consumer’s risks and sample sizes. Several numerical examples and two case studies
are presented to illustrate our resulting sampling plans.

1 Introduction

Acceptance sampling plan is one of the oldest techniques in quality control, which
helps to make a decision on the acceptance of a lot which contains a finite or infinite
number of units. It occupies the middle ground between no inspection and 100%
inspection. A complete sampling is often not desirable when the cost of testing is
high, compared to that of passing a defective unit, or if the testing is destructive.
The sampling schemes protect both the producer and consumer. It enables the pro-
ducer to accept the good lot and the consumer to reject the bad lot. The use of
sampling schemes minimize the nonconforming products and maximize the profit
of the industry.

M. Kumar (B)
Department of Mathematics, National Institute of Technology, Calicut, Kerala, India
e-mail: mahesh@nitc.ac.in

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 123
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_7
124 M. Kumar

There exists a variety of sampling schemes in the literature. The acceptance sam-
pling plans which incorporate the lifetime information of a product are known as
reliability sampling plans. They are time censored (Type I), failure censored (Type
II), Progressive censored, both time and failure censored (Hybrid) or the mixture of
these censorings. Some of the earlier researchers have studied the design and devel-
opment of the life test sampling plans. The first attempt on the development of the life
test sampling plans was done by Epstein and Sobel [1]. They derived the sampling
plan for exponential distribution under Type I and Type II censoring. The expected
number of failures were obtained in Type I censoring and expected testing time in
Type II. Later, many acceptance sampling plans were considered by many authors
for different distributions. For example, Epstein [2], Jun et al. [3], Bartholomew [4],
Aslam and Jun [5], Azam et al. [6], Sai and Wu [7], Kantam et al. [8], Balamurali and
Usha [9], Srinivasa Rao [10], Aslam [11], Aslam and Jun [12] and Chien-Wei Wu
et al. [13], Vellaisamy and Sankar [14]. In all these works, they have considered the
number of failures. Kumar and Ramyamol [15] have considered reliability accep-
tance sampling plan for exponential distribution in which the lifetime of failed units
is taken into account under repetitive group sampling. Ramyamol and Kumar [16]
have also designed acceptance sampling plans for mixture distributions. However,
the above-mentioned works concentrated on the single stage processes.
Asokan and Balamurali [17] considered multi-attribute acceptance sampling plan
for multiple quality characteristics based on a single sample. Using acceptance sam-
pling plan, Duffuaa et al. [18] developed process targeting model for a product with
two dependent quality characteristics. They determined the first quality character-
istic by setting the first process, whereas the determination of the second quality
characteristic depends on the setting of the two processes. The quality of the prod-
uct is controlled by a single sample inspection plan, where inspection is assumed
to be error free. Moskowitz et al. [19] proposed a multi-stage Bayesian acceptance
sampling model for a serial production process incorporating multiple-type defects
that may occur in each stage. Lee and Tagaras [20] derived an economic acceptance
sampling plan in a complex multi-stage production system. The first stage can be
viewed either as the first manufacturing stage or an incoming lot of components or
raw materials on which an operation is performed at the second stage. The planning
and scheduling of production in a multi-task or multi-stage batch manufacturing
process (for example in the case of industries such as chemical manufacturing, food
processing, and oil refining) was developed and studied in the work carried out by
Francesco Gaglioppa et al. [21]. In this paper, a reliability acceptance sampling plan
is developed for a multi-stage production process, in which units pass sequentially
from Machine N to Machine 1. In real life, several products such as semiconductor
devices and television tubes are processed through a multi-stage production process,
and the literature has addressed acceptance sampling plans for these situations. By
exploring the literature, Aslam et al. [22] found that there is no study on the devel-
opment of an acceptance sampling plan for multiple quality characteristics from a
multi-stage process using the time-truncated life test. They considered the devel-
opment of this sampling plan for the product coming from a two-stage process by
assuming that the lifetime of the product follows the Weibull distribution with known
Optimal Design of Reliability Acceptance Sampling Plans … 125

or unknown shape parameters. They have taken the number of failures as the criterion
for accepting the lot and ignored the failure time of units tested. The cost of testing is
also not addressed in their work. The “number of failures criterion” is not enough to
judge whether a lot is good or bad. For example, an accepted lot (for instance, a lot of
certain brand of bulbs), based on this information, may actually perform bad when
the units are deployed in the market, and thereby, could receive negative feedbacks
from the consumers. This will result a huge loss to the industry which produces such
units. In the light of the above, we propose to develop a complete acceptance sampling
plan which overcomes the drawbacks in the existing sampling plan for multi-stage
process. We include the lifetime information of failed units, instead of just depending
upon “number of failures”, and make an attempt to develop a reliability acceptance
sampling plan for multi-stage process which minimizes the expected total testing
cost. The rest of the chapter is organized as follows: In Sect. 2, we define a single
sampling plan for multi-stage production process in which lifetime of units in each
stage follows the exponential distribution with different parameters. In Sect. 3, we
consider double sampling plan. Section 4 discusses the results obtained under single
and double sampling plans. In Sect. 5, comparisons of proposed single and double
acceptance sampling plans are made. Besides, we compare our double sampling plan
with the existing plan (see [22]). In Sect. 6, we have included sensitivity analysis of
expected total testing cost. Case studies are presented in Sect. 7, and in Sect. 8, some
conclusive remarks are presented.

2 Design of Single Sampling Plan (SSP) Under Type II

Censoring

Let N be the number of stages (quality characteristics) under consideration. A single

sample of size n i , i = 1, 2, . . . , N from the ith stage is taken and the quality checking
is carried out at the end of the ith stage. Throughout the paper, we assume that the
stages are independent, and the quality characteristics at the ith, i = 1, 2, . . . , N
stage follow the exponential distribution with different parameters. Let θi be the
parameter of the exponential distribution in the ith stage (ith quality characteristics)
of the process. Then the pdf of the exponential distribution for the ith stage is given
by 1
exp( −x i
), xi > 0, θi > 0
f (xi ) = θi θi (1)
0, other wise,

where xi is the value of the random variable X i , which represents the lifetime of ith
quality characteristics. The formulation of the acceptance sampling plan for the N
stage process is given in the following steps 1 and 2.
1. At the ith, i = 1, 2, . . . , N , stage select a random sample of size n i from the
submitted lot and put them on a test. Continue the test until the total number of
failures observed is ri . Let, X 1,ni , X 2,ni , . . . , X ri,ni be the lifetime of failed units
in the increasing order.
126 M. Kumar

2. Calculate θ̂ri,ni , the maximum

 i likelihood estimate (MLE) of θi . The expression
for the MLE is θ̂ri,ni = ( rj=1 X j,ni + (n i − ri )X ri ,ni )/ri (see [2]). If, θ̂ri,ni ≥ ti ,
accept the units processed in the ith stage and go to (i + 1)th stage. Otherwise,
stop testing at the ith stage and reject the entire lot.
Note that the entire lot will be accepted only when, θ̂ri,ni ≥ ti , ∀ i = 1, 2, . . . , N .
In the above defined single sampling plan, the parameters ri and ti , i = 1, 2, . . . , N
are determined by modeling the problem as an optimization problem which mini-
mizes the average total testing cost at acceptable quality level, satisfying the Type I
and Type II error constraints. The acceptance probability at the ith stage is given by
Pi = P(θ̂ri,ni ≥ ti ).
According to Epstein and Sobel [2], 2ri θθ̂iri,ni follows the chi-square distribu-
tion with 2ri degrees of freedom (χ 2 (2ri )). Hence Pi = P(χ 2 (2ri ) ≥ (2rθii ti ) ), i =
1, 2, . . . , N .
 i −1 e −a2 ( a2 ) j
Using the property, P( χ 2 (2ri ) ≥ a) = rj=0 j!
of the chi-square distribu-
−ri ti
ri −1 e θi
θi (
ri ti
)j
tion, we can write Pi = j=0 j!
.
For an N stage process, the final acceptance of the lot depends on the acceptance
at all the stages. Since the stages are independent, the final acceptance probability is
given by
−ri ti
  i −1
N r
N
e θi ( rθi iti ) j
P= Pi = .
i=1 i=1 j=0
j!

Let Ti denote the time to get ri failures from a sample of size n i at the ith stage.
Then Ti is a random variable with pdf

n!e(n−r +1)x/θi
[1 − e−x/θi ]ri −1 , x > 0, θi > 0
gri ,ni (x) = (r −1)!(n−r )!θi
0, otherwise.

Moreover, the expected testing time is given by


ri
1
E(Ti ) = θi
j=1
ni − j + 1

(See [2]). Next, we calculate the expected total testing time based on the Bernoulli
random variables as follows:
Let 
1, if the ith quality is acceptable
Zi = (2)
0, otherwise.

Also, the expression for the expected value of Z i is calculated as

Optimal Design of Reliability Acceptance Sampling Plans … 127

E(Z i ) = 1.P(ith quality is acceptable) + 0.P(ith quality is unacceptable)

= P(θ̂ri,ni ≥ ti ) = Pi . Now the total random testing time is
T = T1 + Z 1 T2 + Z 1 Z 2 T3 + · · · + (Z 1 Z 2 . . . , Z N −1 )TN
 i−1 
N 
= Z k Ti , Z 0 = 1.
i=1 k=0

N i−1
Hence the expected total testing time, E(T ) = i=1 k=0 E(Z k ) E(Ti )
N i−1  ri  
= i=1 k=0 Pk θi j=1 n i − j+1 , P0 = 1, Z 0 = 1, since Z k s and Ti s are inde-
1

pendent.
In the subsequent sections, we propose the design of optimal acceptance sampling
plan which minimizes the expected total testing cost. It is to be noted that the final
total cost in conducting the proposed acceptance sampling plan will generally consist
of installation cost of installing all the test units at the beginning of the test, test units
cost and operational cost (salary of operators, utility, depreciation of test equipment
and so on), in addition to expected total testing cost. Note that the final total cost can
be computed once the minimum expected total testing cost is obtained.

2.1 Formation of Optimization Problem and Its Solution

The acceptance sampling plan defined above involves parameters (ri , ti ), i =

1, 2, . . . , N , that need to be determined. In this section, we discuss the formula-
tion of an optimization problem which helps to find the parameters of the plan. The
major components of acceptance sampling plan are producer’s and consumer’s risk.
Therefore, we find (ri , ti ), i = 1, 2, . . . , N , which suitably controls the producer’s
and consumer’s risk, and minimizes the expected total testing cost. To accommodate
these pieces of information, we model our problem as an optimization problem with
objective function as the expected total testing cost at the acceptable quality level.
The constraints associated with the problem are Type I error at the acceptable quality
level and Type II error at the unacceptable quality level. If Ci is the cost of testing a
sample of size n i at the ith stage for unit time, then the expected total testing cost of
N i−1 ri
k=0 Pk θi j=1 n i − j+1 C i , P0 = 1.
1
the process is given by i=1
Then our optimization problem to obtain the plan parameters (ri , ti ), i = 1, 2, . . . ,
N , is given by Problem Q 1 :
⎡  ⎤

N i−1 
ri
1
Min ⎣ Pk θi ⎦ Ci (3)
i=1 k=0 j=1
ni − j + 1

such that
P(Type I error at acceptable quality level) ≤ α (4)
128 M. Kumar

P(Type II error at unacceptable quality level) ≤ β. (5)

The acceptable and unacceptable quality levels are defined as follows:

We have N stages (quality characteristics), and θi , i = 1, 2, . . . , N is the param-
eter of the exponential distribution which represents the lifetime of ith quality char-
acteristics. At acceptable quality level, all the θi should possess a minimum value
and is defined as, θi ≥ θi∗ , i = 1, 2, . . . , N . At unacceptable quality level, θi ≤ θi0 ,
for some i, 1 ≤ i ≤ N . We assume that (θi∗ , θi0 ), 1 ≤ i ≤ N to be known. In view of
this, our problem Q 1 can be rewritten as Problem Q 2 :
⎡  ⎤

N i−1 
ri
1
Min ⎣ Pk θi ⎦ (6)
i=1 k=0 j=1
ni − j + 1

such that

n
Pi ≥ 1 − α, θi ≥ θi∗ , ∀1 ≤ i ≤ N (7)
i=1


n
Pi ≤ β, θi ≤ θi0 , for some 1 ≤ i ≤ N . (8)
i=1

Observe that the above Problem Q 2 can be rewritten as Problem Q 3 :

⎡  ⎤

N i−1 
ri
1
Min ⎣ Pk θi ⎦ (9)
i=1 k=0 j=1
ni − j + 1

such that

N
Min Pi ≥ 1 − α, θi ≥ θi∗ , ∀1 ≤ i ≤ N (10)
(θ1 ,θ2 ,...,θn )
i=1


N
Max Pi ≤ β, θi ≤ θi0 , for some 1 ≤ i ≤ N , (11)
(θ1 ,θ2 ,...,θn )
i=1

where α and β are producer’s and consumer’s risk respectively. Note that for an
N stage process one has to find 2N parameters, (r1 , t1 ), (r2 , t2 ), . . . , (r N , t N ). The
problem is intractable when N ≥ 3. Hence we solve our problem for the case N = 2.
Hence Problem Q 3 can be written as Problem Q 4 as follows:
⎡  ⎤

2 i−1 
ri
1
Min ⎣ Pk θi ⎦ (12)
i=1 k=0 j=1
ni − j + 1
Optimal Design of Reliability Acceptance Sampling Plans … 129

such that

2
Min Pi ≥ 1 − α, θi ≥ θi∗ , ∀1 ≤ i ≤ 2 (13)
(θ1 ,θ2
i=1


2
Max Pi ≤ β, θi ≤ θi0 , for some 1 ≤ i ≤ 2. (14)
(θ1 ,θ2 )
i=1

2
Observe that, i=1 Pi on the left-hand side of the inequalities (13) and (14) are
all increasing in θi . The minimum i=1 2
Pi occurs at (θ1∗ , θ2∗ ). Consider the feasible
regiondescribed in (14). Divide it into  three parts
 as follows: 
A1 = (θ1 , θ2 ) : θ1 ≤ θ10 and θ2 ≤ θ20 , A2 = (θ1 , θ2 ) : θ1 ≥ θ10 and θ2 ≤ θ20 and
A3 = (θ1 , θ2 ) : θ1 ≤ θ10 and θ2 ≥ θ20 .
The regions A2 and A3 are unbounded, hence the Problem Q 4 cannot be solved.
Therefore, we consider only the bounded region A1 and note that the point of maxi-
mum is (θ10 , θ20 ). So the final optimization problem becomes Problem Q 5 :
⎡  ⎤

2 i−1 
ri
1
Min ⎣ Pk θi ⎦ Ci (15)
i=1 k=0 j=1
ni − j + 1

such that

2
Pi ≥ 1 − α at (θ1∗ , θ2∗ ) (16)
i=1


2
Pi ≤ β at (θ10 , θ20 ). (17)
i=1

The optimization problem for N = 2 is solved by genetic algorithm solver in

Matlab. We give several numerical examples in Table 1.

3 Construction of Double Sampling Plan (DSP) Under

Type II Censoring

A double sampling plan, in general, is more difficult to construct and hence need
not be easy for implementation as compared to the single sampling plan. However,
it may give similar levels of the consumer’s and the producer’s risk and requires
less sampling in the long run than a single sampling plan. We consider the double
sampling plan for multi-stage process which is defined as follows. Here one may
note that each stage can have two samples. The DSP algorithm is given as follows.
For 1 ≤ i ≤ N :
130 M. Kumar

1. Select a random sample of size n i .

2. Observe ri number of failures.
(1)
3. Calculate MLE θ̂ri,ni of θi .
(1)
a. If, θ̂ri,ni ≥ ti2 , accept the units produced in ith stage. Set i = i + 1, go to Step
1, else go to (b).
(1)
b. If, ti1 ≤ θ̂ri,ni < ti2 , take one more sample from the ith stage by fixing the
(2)
same sample size n i and observe ri failures. Calculate MLE θ̂ri,ni of θi using
the new sample and go to (c). Else go to (e).
(1) (2)
c. If, θ̂ri,ni + θ̂ri,ni ≥ 2ti2 , accept the units from the ith stage. Set i = i + 1, go
to Step 1. Else go to (d).
d. Reject the lot, go to Step 5
e. Reject the lot, go to Step 5.
4. If all the stages are accepted, then accept the entire lot
5. End.
The unknown quantities (ti1 , ti2 , ri ), i = 1, 2 . . . , N are determined by formulat-
ing an optimization problem which minimizes the expected total testing cost. In the
double sampling plan, the acceptance probability P is calculated as follows:
At the ith stage, we accept the units if
   
(1) (1) (1) (1)
θ̂ri,ni ≥ ti2 or ti1 ≤ θ̂ri,ni < ti2 and θ̂ri,ni + θ̂ri,ni ≥ 2ti2 .

Then the acceptance probability at the ith stage is given by

(1) (1) (1) (2)
Pi = P(θ̂ri,ni ≥ ti2 ) + P((ti1 ≤ θ̂ri,ni ≤ ti2 ) ∩ (θ̂ri,ni + θ̂ri,ni ≥ 2ti2 )).

(1) (2)
Here θ̂ri,ni and θ̂ri,ni are independent random variables.
(1) (2)
Denote θ̂ri,ni = X and θ̂ri,ni = Y . Then the pdf of X (see [1]) is given by

( 2rθii )ri 2xri Γ ri exp( −rθii x ), x > 0, θi > 0
ri−1

f (x) = (18)
0, otherwise.

Note that Y also has the same pdf. The joint pdf of X and Y is f (x) f (y) since X
and Y are independent.
Thus
Optimal Design of Reliability Acceptance Sampling Plans … 131

P((ti1 ≤ X ≤ ti2 ) ∩ (X + Y ≥ 2ti2 ))

 ti2  ∞
2ri x ri−1 −ri x 2ri ri y ri−1 −ri y
= ( )ri r exp( )( ) r exp( )d yd x
ti1 2ti2 −X θi 2 iΓr
i θi θi 2 iΓr
i θi
 ti2 r
i −1
2ri x ri−1 −ri x −ri (2ti2 − x) −ri (2ti2 − x) j 1
= ( )ri r exp( ) exp( )( )
ti1 j=0 θi 2 iΓr
i θi θi θi j!
i −1
r 
ri −2ri ti2 1 ti2
= ( )ri + j exp( ) x ri − (2ti2 − x) j d x
θ
j=0 i
θi j!Γ ri ti1

Observe that P(X ≥ ti2 )+ P((ti1 ≤ X ≤ ti2 ) ∩ (X + Y ≥ 2ti2 ))

 i −1 ri ri + j t
= P(χ 2 (2ri ) ≥ riθtii2 ) + rj=0 ( θi ) exp( −2rθii ti2 ) j!Γ1 ri ti1i2 x ri −1 (2ti2 − x) j d x
−ri ti2  −ri ti2 
ri −1 e θi ( θrii ) j e θi ( θi )ri  ti2
r
j
Hence Pi = j=0 j!
ti2 + Γ ri
i
ti1 x
ri−1
(2ti2 − x) . j

Now, the independence of all the stages gives the total acceptance probability, P =
N
i=1 Pi . As in the single sampling plan, we find the plan parameters (ri , ti1 , ti2 , ), i =
1, 2, . . . , N by formulating an optimization problem which minimizes the expected
total testing cost at acceptable quality level satisfying the Type I and Type II error
constraints.
(1) (1) (2)
We have, E(Z i ) = P(θ̂ri,ni ≥ ti2 ) + P((ti1 ≤ θ̂ri,ni ≤ ti2 ) ∩ (θ̂ri,ni + θ̂ri,ni ≥
2ti2 )) = Pi , where Z i is the Bernoulli random variable defined in Sect. 2.
Let Ti be the random testing time to get ri failures from a sample of size
n i . Then the random testing time for the ith stage, i = 1, 2, . . . , Nin the DSP
(1)
is (1 + Pi1 )Ti , where Pi1 = P((ti1 ≤ θ̂ri,ni ≤ ti2 ). Also E(Ti ) = θi rj=1 i 1
n i − j+1
.
Hence the total random testing time, T = (1 + P11 )T1 + (1 + P21 )Z 1 T2 + · · · +
(1 + PN 1 )(Z n−1 , . . . , Z 1 )TN .
If Ci is the cost of testing a sample of size n i for unit time in the ith stage, then the
N i−1 ri
k=0 Pk θi (1 + Pi1 )
1
expected total testing cost is i=1 j=1 n i − j+1 C i , where
 −ri ti1 −ri ti2 
 i −1 e θi ( riθtii1 )m r t
e θi ( iθ i2 )m
P0 = 1, Pi1 = rm=0 m!
− m!
i
.

Then our optimization problem becomes Problem Q 6 :

⎡  ⎤

N i−1 
ri
1
Min ⎣ Pi θi (1 + li ) ⎦ Ci (19)
i=1 k=0 j=1
ni − j + 1

such that

N
Pi ≥ 1 − α, θi ≥ θi∗ , ∀1 ≤ i ≤ N (20)
i=1
132 M. Kumar


N
Pi ≤ β, θi ≤ θi0 , for some 1 ≤ i ≤ N . (21)
i=1

Observe that the inequalities in (20) and (21) are equivalent to


N
Min Pi ≥ 1 − α, θi ≥ θi∗ , ∀1 ≤ i ≤ N (22)
(θ1 ,θ2 ,...,θn )
i=1


N
Max Pi ≤ β, θi ≤ θi0 , for some 1 ≤ i ≤ N . (23)
(θ1 ,θ2 ,...,θn )
i=1

For an N-stage process, one has to find 3N parameters namely (ri , ti1 , ti2 ), i =
1, 2, . . . , N for the implementation of DSP. The problem is hard to deal with for
N ≥ 3. Hence we solve our problem for the case N = 2. Hence the Problem can be
rewritten as Problem Q 7 :
⎡  ⎤

2 i−1 
ri
1
Min ⎣ Pk θi (1 + li ) ⎦ Ci (24)
i=1 k=0 j=1
ni − j + 1

such that

2
Min Pi ≥ 1 − α, θi ≥ θi∗ , ∀1 ≤ i ≤ 2 (25)
(θ1 ,θ2 )
i=1


2
Max Pi ≤ β, θi ≤ θi0 , for some 1 ≤ i ≤ 2. (26)
(θ1 ,θ2 )
i=1

Now  divide the feasible region of (26)

 into three parts as given in Sect. 2. 
A1 = (θ1 , θ2 ) : θ1 ≤ θ10 and θ2 ≤ θ20 , A2 = (θ1 , θ2 ) : θ1 ≥ θ10 and θ2 ≤ θ20 and
 
A3 = (θ1 , θ2 ) : θ1 ≤ θ10 and θ2 ≥ θ20 . If we choose, ti2 ≥ θ2i , then i=1 2
Pi in (25)
and (26) are increasing with respect to each θi . The minimum in (25) occurs at
(θ1∗ , θ2∗ ). Since the feasible region of inequality (26) is unbounded, we consider the
region A1 only. So Problem Q 7 becomes Q 8 :
⎡  ⎤

2 i−1 
ri
1
Min ⎣ Pk θi (1 + li ) ⎦ Ci (27)
i=1 k=0 j=1
ni − j + 1

such that

2
Pi ≥ 1 − α at (θ1∗ , θ2∗ ) (28)
i=1
Optimal Design of Reliability Acceptance Sampling Plans … 133


2
Pi ≤ β at (θ10 , θ20 ). (29)
i=1

The optimization can be solved using genetic algorithm solver in Matlab. The exam-
ples are mentioned in Table 2. It is observed from Tables 3 and 4 that a change in
sample size will effect expected total testing time and cost.

4 Numerical Results and Comparisons

In this section, we present some numerical results of single and double sampling
plans in Tables 1 and 2 respectively. Various comparisons of results based on SSP
(ri , ti ) and DSP (ri , ti1 , ti2 ) are shown in Tables 3 and 4.
First, we compare SSP (ri , ti ) and DSP (ri , ti1 , ti2 ). Next, DSP (ri , ti1 , ti2 ) is com-
pared with the existing plan in the literature [22]. One may observe that in Aslam et
al. [22], authors have considered the quality ratio, and for the different quality ratios,
minimum sample number is calculated. The main drawback of this work is that the
plan parameters are calculated using quality ratio which ignores the actual measured
lifetime of the product for a specified period. Our DSP (ri , ti1 , ti2 ) based on the life-
time of the product, one can choose the sample size, where the units are tested for their
failure, and the testing cost is changing according to the lifetime of the products. For
example, in Aslam et al. [22], when α = 0.05, β = 0.1, ratio = 2 and a = 0.5, their
plan gives an average sample number of 41.66, and the lot is accepted or rejected only
based upon the number of defectives obtained. Hence the test plan ignores the actual
lifetime of the units in the lot. The total testing time involved is 1200 units. Next,
we consider our DSP (ri , ti1 , ti2 ) problem. Let, α = 0.05, β = 0.1, θ1∗ = 100, θ10 =
30, θ2∗ = 300, θ20 = 100, n 1 = 15, n 2 = 18, will give an expected total testing cost
114.2. Note that the actual testing cost may even be less than 114.2. It is observed that
the proposed plan DSP (ri , ti1 , ti2 ) has an advantage of savings in expected testing
costs of about 50%.

5 Comparisons of Single Sampling and Double Sampling

Plans

In this section, we compare our single sampling and double sampling plans. A com-
parisions of our plans are done with existing sampling plans in the literature (see
[22]). It is seen that our plans perform better as compared to the existing plans in
terms of testing costs incurred (see Tables 3, 4, 5 and 6).
134 M. Kumar

Table 1 Examples of single sampling plan (SSP(ri , ti ))

The values of t1 , t2 and r1 , r2 for various choices of (θ1∗ , θ10 ), (θ2∗ , θ20 ), (α, β) and C1 = C2 = 1
(α, β) (θ1∗ , θ10 ) (θ2∗ , θ20 ) (n 1 , n 2 ) t1 t2 (r1 , r2 ) E T C at
θ1∗ , θ2∗
0.1, 0.25 1500, 700 900, 500 20, 20 850 400 12, 1 1711
0.1, 0.05 1500, 700 900, 500 20, 20 1000 500 14, 6 2439.4
0.1, 0.05 1000, 200 300, 100 15, 18 550 200 9, 10 1375.3
0.05, 0.01 100, 30 300, 100 15, 18 50 100 7, 5 208.58
0.05, 0.05 100, 30 300, 100 15, 18 55 111 7, 7 256.49
0.1, 0.05 100, 30 300, 100 15, 18 59 99 5, 2 118.5925
0.05, 0.06 500, 100 300, 100 10, 10 203 109 4, 3 535.1864
0.1, 0.1 500, 100 300, 100 10, 10 242 112 2, 3 371.9771
0.05, 0.05 500, 100 300, 100 10, 10 201 101 5, 3 636.4034

Table 2 Examples of double sampling plan (DSP (ri , ti1 , ti2 ))

The values of (t11 , t12 ), (t21 , t22 ) and r1 , r2 for various choices of (θ1∗ , θ10 ), (θ2∗ , θ20 ),
(α, β) and C1 = C2 = 1
(α, β) (θ1∗ , θ10 ) (θ2∗ , θ20 ) (n 1 , n 2 ) (t11 , t12 ) (t21 , t22 ) (r1 , r2 ) E T C at
θ1∗ , θ2∗
0.1, 0.05 1500, 500 900, 300 20, 20 275, 650 65, 485 4, 3 802.96
0.05, 0.1 1500, 500 900, 300 20, 20 150, 669 65, 372 4, 3 819.296
0.1, 0.25 1500, 500 900, 300 20, 20 40, 396 85, 381 2, 2 609.9
0.1, 0.05 1000, 300 300, 50 15, 10 99, 387 18, 98 2, 2 409.699
0.05, 0.1 1000, 300 300, 50 15, 10 57, 226 19, 97 2, 2 468.645
0.1, 0.25 300, 80 800, 350 10, 15 17, 96 71, 326 2, 2 412.442
0.1, 0.15 300, 80 800, 350 10, 15 14, 119 68, 365 2, 3 480.88
0.1, 0.25 300, 80 300, 80 10, 15 14, 97 11, 57 2, 1 204.76
0.2, 0.25 300, 80 300, 80 10, 15 19, 97 11, 58 1, 1 150.26
0.15, 0.15 700, 90 400, 100 10, 15 25, 154 11, 91 1, 1 267.05

Table 3 Comparison between single (SSP (ri , ti )) and double (DSP (ri , ti1 , ti2 )) sampling plans
Comparisons of SSP(ri , ti ) and DSP (ri , ti1 , ti2 ) for various choices of (θ1∗ , θ10 ),
(θ2∗ , θ20 ), (α, β) and C1 = C2 = 1
(α, β) (θ1∗ , θ10 ) (θ2∗ , θ20 ) (n 1 , n 2 ) ETC under ETC under
single double
sampling sampling
0.1, 0.25 1500, 700 900, 500 20, 20 1711 978.3
0.05, 0.05 100, 30 300, 100 15, 18 256.49 114.2
0.05, 0.05 500, 100 300, 100 10, 10 636.4034 233.544
Table 4 Comparison between double sampling plan DSP (ri , ti1 , ti2 ) and the existing sampling plan Aslam et al. [22]
Comparisons of double sampling plan with the existing plan for various choices of
(θ1∗ , θ10 ), (θ2∗ , θ20 ), (α, β) and C1 = C2 = 1
(α, β) (θ1∗ , θ10 ) (θ2∗ , θ20 ) (n 1 , n 2 ) ETC under double expected total testing cost Average sample number
sampling for existing plan, a=0.5 for existing plan
and ratio=2
0.05, 0.25 1500, 700 900, 500 20, 20 978.3 1200 40.13
0.05, 0.05 100, 30 300, 100 15, 18 114.2 200 65.66
Optimal Design of Reliability Acceptance Sampling Plans …

0.05, 0.1 1500, 700 900,500 20, 20 1015.4 1200 53.47

135
136 M. Kumar

Table 5 Effect of change in sample number on cost in single sampling plan (SSP (ri , ti ))
The values of ETC for various choices of (θ0 , θ1 ), (α, β) and C = 1
(α, β) (θ1∗ , θ10 ) (θ2∗ , θ20 ) (n 1 , n 2 ) t1 t2 (r1 , r2 ) E T C at
θ1∗ , θ2∗
0.1, 0.25 1500, 700 900, 500 15, 10 850 400 12, 1 3086.9
0.1, 0.25 1500, 700 900, 500 25, 10 850 400 12, 1 1392.6
0.1, 0.25 1500, 700 900, 500 15, 20 850 400 12, 1 2953.3
0.1, 0.05 1000, 200 300, 100 15, 25 550 200 9, 10 1272.8
0.1, 0.05 1000, 200 300, 100 25, 25 550 200 9, 10 711.6130
0.1, 0.05 1000, 200 300, 100 15, 15 550 200 9, 10 1474.9
0.05, 0.01 100, 30 300, 100 10, 12 50 100 7, 5 365.1710
0.05, 0.01 100, 30 300, 100 15, 20 50 100 7, 5 193.7672
0.05, 0.01 100, 30 300, 100 25, 20 50 100 7, 5 156.4667

Table 6 Effect of change in sample number on cost in double sampling plan DSP (ri , ti1 , ti2 )
The values of ETC for various choices of (θ1∗ , θ10 ), (θ2∗ , θ20 ), (α, β) and C1 = C2 = 1
(α, β) (θ1∗ , θ10 ) (θ2∗ , θ20 ) (n 1 , n 2 ) (t11 , t12 ) (t21 , t22 ) (r1 , r2 ) E T C at
θ1∗ , θ2∗
0.1, 0.05 1500, 500 900, 300 20, 20 275, 650 65, 485 4, 3 802.96
0.1, 0.05 1500, 500 900, 300 10, 20 275, 650 65, 485 4, 3 1439.30
0.1, 0.05 1500, 500 900, 300 10, 10 275, 650 65, 485 4, 3 1709.41
0.1, 0.05 1000, 300 300, 50 15, 10 99, 387 18, 98 2, 2 409.699
0.1, 0.05 1000, 300 300, 50 25, 10 99, 387 18, 98 2, 2 287.9
0.1, 0.05 1000, 300 300, 50 5, 15 99, 387 18, 98 2, 2 1003.8
0.1, 0.25 300, 80 800, 350 10, 15 17, 96 71, 326 2, 2 412.442
0.1, 0.25 300, 80 800, 350 15, 15 17, 96 71, 326 2, 2 365.79
0.1, 0.25 300, 80 800, 350 15, 25 17, 96 71, 326 2, 2 256.06
0.1, 0.25 300, 80 300, 80 10, 15 14, 97 11, 57 2, 1 204.76
0.1, 0.25 300, 80 300, 80 5, 5 14, 97 11, 57 2, 1 471.02
0.1, 0.25 300, 80 300, 80 5, 25 14, 97 11, 57 2, 1 323.4
0.15, 0.15 700, 90 400, 100 10, 15 25, 154 11, 91 1, 1 267.05
0.15, 0.15 700, 90 400, 100 10, 5 25, 154 11, 91 1, 1 416.5

6 Sensitivity Analysis of Sampling Plans

In this section, we focus on the sensitivity of some of the important parameters in

the derived sampling plans, and investigate the behavior of the expected total testing
cost due to the variations in producer’s and consumer’s risks and sample sizes. The
results of sensitivity analysis study are shown in Figs. 1, 2, 3 and 4.
Optimal Design of Reliability Acceptance Sampling Plans … 137

138

137 θ*1=150,θ*2=90
θ01=40,θ02=30
136 β=0.1
Expected total testing cost

135

134

133

132

131

130

129
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095
α

Fig. 1 Sensitivity of expected total testing cost to changes in α in DSP

We observe from the above sensitivity analysis studies that the expected total testing
cost is a decreasing function of α (see Fig. 1). On the other hand, ETC shows fluc-
tuating behavior, that it is neither decreasing nor increasing with respect to changes
in β. It can also be noted from Fig. 2 that ETC has an increasing trend in the interval
(0.06, 0.85).
The sensitivity analysis study of the expected total testing cost subjected to
changes in two parameters, namely n 1 and n 2 , is given in Figs. 3 and 4. It is clear
from the figures that ETC is a decreasing function of (n 1 , n 2 ), when all the other
parameters in the sampling plan are kept fixed.

7 Some Case Studies

Lets us consider the following examples to understand the results obtained in the
previous sections.
Example 1 We consider the example depicted in Aslam et al. [22], a two-stage
inspection process of ball bearings. The compressive strength and the maximum
stress of a ball bearing are first and second quality characteristics respectively.
Assume that the compressive strength and the stress of a ball bearing are inde-
138 M. Kumar

137.5

137 θ1*=150, θ2*=90

0 0
θ1=40, θ2=30
136.5 α=0.1
expected total testing cost

136

135.5

135

134.5

134

133.5
0.05 0.055 0.06 0.065 0.07 0.075 0.08 0.085 0.09 0.095
β

Fig. 2 Sensitivity of expected total testing cost to changes in β in DSP

Table 7 Failure times of 20 ball bearings related with strength (stage I)

Sample 0.2439 0.4728 0.1764 0.1239 0.8787 0.3995 0.1687 0.3858 0.4427
I:
0.5582
Sample 0.4447 0.3119 0.3733 0.7539 0.1624 0.5920 0.8900 0.5280 0.1079
II:
0.1050

pendent and that each follows the exponential distribution. Let us consider the single
and double sampling plan for the lot acceptance. Assume that α = 0.05, β = 0.25.
Let the specified mean failure time under the test compressive load or the stress load
of a ball bearing be θ1∗ = θ2∗ = 0.25 and θ10 = θ20 = 0.1. Table 7 and Table 8 give the
sample values.

Single sampling plan: Choose the first sample of each stage with n 1 = n 2 = 10.
Then the plan parameters are calculated using Problem Q 5 . Here t1 = 0.11, t2 =
0.115, r1 = 5 and r2 = 6. Consider sample 1 given in Table 7. We take first five
failure times, which give θ̂5,10 = 0.60554, and 0.60554 > t1 = 0.11. So we accept
the lot from the first stage and move to the second stage. In the second stage, first 6
failure times recorded in sample 1 given in Table 8 are under consideration. Then we
Optimal Design of Reliability Acceptance Sampling Plans … 139

θ* =100,θ* =300,
1 2
0 0
θ =30, θ =100,
1 2
α=0.05, β=0.01
500
expected total testing cost

400

300

200

100

0
40
30 40
20 30
20
10 10
n2 0 0
n1

Fig. 3 Sensitivity of expected total testing cost to changes in (n 1 , n 2 ) in SSP

Table 8 Failure times of 20 ball bearings related with stress (stage II)
Sample 0.3337 0.0379 0.3248 0.2935 0.3901 0.3608 0.5570 0.6922 0.0971
I:
0.3029
Sample 0.3042 0.2392 0.9623 0.4297 0.2629 0.6186 0.5351 0.6047 0.6976
II:
0.1603

have, θ̂6,10 = 0.5083 > t2 = 0.115. Hence we accept the product from the second
stage and hence this leads to the final acceptance of the entire lot.
Double sampling plan: Let n 1 = n 2 = 10. Then the decision making parameter
values are, t11 = 0.1, t12 = 0.2, t21 = 0.1, t22 = 0.184, r1 = 2 and r2 = 1. These
(1)
values are obtained using Problem Q 8 . Thus if the average failure time θ̂2,10 is greater
than 0.2, we accept the units from the first stage. In our example, from sample 1 given
(1)
in Table 7, θ̂2,10 = 0.567 > 0.2. So we accept the units from the first stage. In the
second stage, r2 = 1. From sample 1 given in Table 8, we have the first failure time
(1)
is 0.0379 and θ̂1,10 is 0.379 which is greater than t22 = 0.184. Hence we accept the
units from the second stage also. Thus we finally accept the entire lot.
140 M. Kumar

* *
θ1=1500,θ2=900
θ01=500, θ02=300
α=0.1, β=0.05
4000
expected total testing cost

3000

2000

1000

0
40
30 40
20 30
20
10 10
n2 0 0
n1

Fig. 4 Sensitivity of expected total testing cost to changes in (n 1 , n 2 ) in DSP

Table 9 Failure times of 10 units with mean life 1 (stage I)

0.0989 0.0906 2.3275 0.6035 0.0357
2.0637 0.4583 1.2783 0.0434 1.8476

Table 10 Failure times of 10 units with mean life 5 (stage II)

0.1492 3.6142 9.7636 0.4401 0.2068
0.2189 1.1140 4.3166 1.1647 2.1099

Example 2 In this second example, we have generated two samples of size 10 from
an exponential distribution, each with mean 1 and 5 respectively. The samples are
given in Tables 9 and 10.
We illustrate SSP and DSP for this example in the following.
Single sampling plan: Set α = 0.1, β = 0.2, n 1 = 10, n 2 = 10, θ1∗ = 1, θ10 = 0.5
and θ2∗ = 5, θ20 = 2. Our plan parameters are given by r1 = r2 = 5, t1 = 0.395 and
ˆ = 0.60368.
t2 = 3. From the Table 9, select the first 5 failure times, which gives θ5,10
Since 0.60368 > t1 = 0.395, we accept the product from the first stage. In the second
ˆ = 1.5398 < t2 = 3. Hence we reject the product
stage r2 = 5. From Table 10, θ5,10
from the second stage and hence this leads to the final rejection of the entire lot.
Optimal Design of Reliability Acceptance Sampling Plans … 141

Double sampling plan: Set α = 0.1, β = 0.2, n 1 = 10, n 2 = 10, θ1∗ = 1, θ10 = 0.5
and θ2∗ = 5, θ20 = 2. Our plan DSP (ri , ti1 , ti2 ) give r1 = r2 = 2, t11 = 0.35, t12 =
(1)
0.8, t21 = 2.12 and t22 = 4. From Table 9, θ̂2,10 = 0.375 > t11 = 0.35. Now we take
another sample of size 10, which is given below:
Sample II: 1.4491 1.0408 0.1970 4.1732 3.1460 1.7779 0.4321 0.3124 0.4343
0.7965.
(2) (1) (2)
From this sample θ̂2,10 = 1.5043, and (θ̂2,10 + θ̂2,10 )/2 = 0.93965 > 0.8. Hence
(1)
we accept the units from the first stage. In the second stage, θ̂2,10 = 1.0052 < t21 =
2.12. Therefore, we reject the units from the second stage which leads to the rejection
of the entire lot.

8 Conclusions

In this chapter, we developed two sampling schemes, namely single and double
sampling plans for the multistage process based on the information about the failure
times of units obtained under Type II censoring. The multistage production process in
which stages are independent is discussed by Aslam et al. [22]. They have considered
the number of failures as the criterion for accepting the lot and ignored the failure
time of units tested. Also the cost of testing is not addressed in their work. In this
work, we have removed the above drawbacks by considering the average-life of units
in the sample. It is noticed that when the sample size increases, the expected total
testing cost decreases which is clear from the sensitivity analysis study of ETC with
respect to changes in sample sizes (n 1 , n 2 ). Also, as α increases, ETC decreases.
However, it is observed that ETC shows a fluctuating trend subjected to changes
in β. It is inspected that our plan DSP (ri , ti1 , ti2 ) has an advantage of savings in
expected total testing cost of about 50% as compared to Aslam et al. [22]. The actual
testing cost may even be less than that reported. As a future scope of work one may
consider dependent production process with several stages.

Acknowledgements The author would like to express his gratitude to the editors for their con-
structive comments which improved the presentation of the chapter. The author would also like to
thank Dr. Ramyamol P C for her computational assistance.

References

1. Epstein B, Sobel M (1953) Life testing. J Am Stat Assoc 48:486–502

2. Epstein B (1954) Truncated life tests in the exponential case. Ann Math Stat 25:555–564
3. Jun CH, Balamurali S, Lee SH (2006) Variable sampling plans for Weibull distribution lifetimes
under sudden death testing. IEEE Trans Reliab 55:53–58
4. Bartholomew DJ (1963) The sampling distribution of an estimate arising in life testing. Tech-
nometrics 5:361–374
142 M. Kumar

5. Aslam M, Jun CH (2009) A group acceptance sampling plans for truncated life tests based on
the Inverse Rayleigh and Log-Logistic distributions. Pakistan J Stat 25:1–13
6. Azam M, Aslam M, Balamurali S, Javaid A (2015) Two stage group acceptance sampling plan
for half normal percentiles. J King Saud University - Sci 27:239–243
7. Tsai TR, Wu SJ (2006) Acceptance sampling based on truncated life test for generalized
Rayleigh distribution. J Appl Stat 33:595–600
8. Kantam RR, Rosaiah K, Srinivas Rao G (2001) Acceptance sampling plans based on life tests:
log- logistic model. J Appl Stat 28:121–128
9. Balamurali S, Usha M (2013) Optimal design of variable chain sampling plan by minimizing
the average sample number. Int J Manuf Eng 3:1–10
10. Srinivasa Rao G (2009) A group acceptance sampling plans for lifetimes following a general-
ized exponential distribution. Econ Quality Control 24:75–85
11. Aslam M (2007) Double acceptance sampling based on truncated life-tests in Rayleigh distri-
bution. Eur J Sci Res 17:605–611
12. Aslam M, Jun CH (2009) A group acceptance sampling plan for truncated life test having
Weibull distribution. J Appl Stat 36:1021–1027
13. Chien-Wei W, Aslam M, Jun C-H (2012) Variables sampling inspection scheme for resubmitted
lots based on the process capability index Cpk. Eur J Oper Res 217:560–566
14. Vellaisamy P, Sankar S (2005) A unified approach for modeling and designing attribute sam-
pling plans for monitoring dependent production processes. Methodol Comput Appl Probab
7:307–323
15. Kumar M, Ramyamol PC (2016) Optimal reliability acceptance sampling plan for exponential
distribution. Econ Quality Control 31:23–36
16. Ramyamol PC, Kumar M (2019) Optimal design of variable acceptance sampling plans for
mixture distribution. J Appl Stat 46:2700–2721
17. Asokan MV, Balamurali S (2000) Multi attribute single sampling plans. Econ Quality Control
15:103–108
18. Duffuaa SO, Al-Turki UM, Kolus AA (2009) Process-targeting model for a product with two
dependent quality characteristics using acceptance sampling plans. Int J Prod Res 47:4041–
4046
19. Moskowitz H, Plante R, Tang K (1996) Multi stage multi attribute acceptance sampling in
serial production systems. IIE Trans 130–137
20. Lee HL, Tagaras G (1992) Economic acceptance sampling plans in complex multi-stage pro-
duction systems. Int J Prod Res 30:2615–2632
21. Gaglioppa F, Miller LA, Benjaafar S (2008) Multi-task and multi stage production planning
and scheduling for process industries. Oper Res 56:1010–1025
22. Aslam M, Azam M, Jun C-H (2015) Acceptance sampling plans for multi-stage process based
on time–truncated test for Weibull distribution. Int J Adv Manuf Technol 79:1779–1785
The Importance of Technical Diagnostics
for Ensuring the Reliability of Industrial
Systems

D. Lj. Branković, Z. N. Milovanović, and V. Z. Janičić Milovanović

Abstract The success and sustainability of the business industrial system is largely
determined by the degree of effectiveness of the production system as its basic integral
part. The reliability of the technical system, i.e., the probability of performing the
projected goal function in the observed time period, along with the readiness and
functional convenience is a basic indicator of the effectiveness of the production
system. Maintenance, as an integral part of the production system, has the function
of providing the projected level of optimal reliability by implementing activities
aimed at ensuring the required level of technical readiness of parts of the technical
system. One of the ways to ensure the optimal level of reliability of the technical
system is the application of maintenance concepts according to the condition that
allows monitoring and control of technical parameters of the state of the elements of
technical systems within the production processes. The basis of the application of the
condition based maintenance is the introduction of procedures for monitoring and
control of condition parameters using technical diagnostic methods. Knowing the
current state of the parameters that determine the degree of success of the projected
function of the goal of the technical system gives the possibility of timely response
to the occurrence of malfunctions and avoid the entry of the system into failure. The
paper presents a systematic provision of regular monitoring and control of condition
parameters of parts of the technical system of paper machines, specifically vibration
levels on high voltage motors, within the industrial system for production of toilet
paper using technical diagnostic methods using portable vibration control devices.
By timely response to the observed occurrence of increased vibration levels at a
specific critical position with the application of technical diagnostic methods and

D. Lj. Branković · Z. N. Milovanović (B)

Department of Hydro and Thermal Engineering, Faculty of Mechanical Engineering Banja Luka,
University of Banja Luka, Stepe Stepanovića 71, Banja Luka, Bosnia and Herzegovina
e-mail: zdravko.milovanovic@mf.unibl.org
D. Lj. Branković
e-mail: dejan.brankovic@mf.unibl.org
V. Z. J. Milovanović
Solunska 8a, 78000 Banja Luka, Bosnia and Herzegovina
e-mail: valentina.mil@live.com

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 143
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_8
144 D. Lj. Branković et al.

complex maintenance intervention activities, it is possible to avoid unwanted failure,

long-term unplanned production downtime and very high maintenance costs. Along
with the description of the specific event and preventive action of the maintenance
service, the possibility of connecting the observed working position with other critical
positions provided with systems for continuous monitoring of operating parameters,
specifically the vibration levels of bearings, is indicated. By constantly striving and
finding ways to raise the overall level of reliability to a higher level, maintaining
the production system plays a major role in providing conditions for continuous
production, achieving planned results and profits and sustainability of production
and business industrial system in the increasingly demanding world market.

Keywords Technical diagnostic · Effectiveness · Reliability · Condition based

maintenance

1 Introduction

The production of goods with new use value as well as the provision of the required
quality of services within the production process is the basis of economic develop-
ment of social communities. The organization of production processes arises as a
consequence of the planned integration of human, material and energy resources and
their transformation into the desired form within the production industrial systems.
Production systems are composed of a series of complex technical systems, i.e., inte-
grated functional units by means of which the transformation of input quantities is
performed, i.e., material values into semi-finished or finished products. An integral
part of production processes are activities to ensure the continuity of the process of
technical systems in order to fulfill the projected function of the goal. These activ-
ities are planned, organized and managed by a special organizational unit of the
production system—maintenance. All maintenance activities within the production
industrial system are closely related to other units that participate in the production
process and form part of the integrated system support to production.

2 Theoretical Settings of Industrial Systems Maintenance

The need for maintenance in industrial systems appears from the very beginning
and application of technical systems during the first industrial revolution and the
introduction of mechanization in production processes. The feature of any technical
system is that under the influence of external and internal disturbances during oper-
ation, it can reach a state of failure which interrupts its function, which stops part
or the entire production process. In order to prevent the failure condition, or if it
occurs, an appropriate response from the maintenance service is required. Within the
period of the first industrial revolution, there was no need for a systematic approach
The Importance of Technical Diagnostics … 145

Fig. 1 Industrial revolutions

to maintenance. Maintenance was reduced to interventions after the occurrence of

failures and the occurrence of failures of parts of technical systems. With the devel-
opment of technical means, electrification during the second industrial revolution,
the introduction of electronics and automation during the third industrial revolution,
all the way to digitalization and application of modern software solutions in a new
understanding of the so-called. Industries 4.0 and 5.0 (Fig. 1), the importance of
maintenance has become one of the key factors for ensuring high performance of
efficient production systems.
The development of technological processes and industry in the second half of the
twentieth century conditioned a new way of understanding the maintenance process.
As this is a very complex concept, there are many ways to define the maintenance
process. Rejec [1] cites Congress OECD (Organization for Economic Co-operation
and Development) from 1963 and presents one of the first definitions of maintenance:
Maintenance is the function of the company entrusted with constant control over the plants
and the performance of certain repairs and audits, which enables constant functional capacity
and preservation of production plants, auxiliary plants and other equipment.

There are numerous scientific papers and researches that try to describe the concept
of maintenance in more detail. Papić and Milovanović [2] point out that maintenance
ceases to be seen as a need that has a random character and becomes a joint action of
a set of segments or elements in order to ensure the functional state of the technical
system in accordance with the requirements and criteria. The process of maintaining
assets, as one of the most important parts of the overall production process, has the
task of preventing and eliminating system failures, primarily through streamlining
and optimizing their use and increasing productivity and cost-effectiveness in the
production and operation process.
146 D. Lj. Branković et al.

The traditional understanding of maintenance related to the processes or activi-

ties of repairing damaged parts of a technical system is changing, Tsang [3]. The
new understanding is defined by the maintenance terminology EN 13,306: 2001
[4], Bengtsson [5] by CEN (Committee of Europeen de Normalisation, European
Committee for Standardization). Standard EN 13,306: 2001 and is partly based on
the standard IEC 60,050 (191) according to which the term maintenance is defined
as:
Maintenance is a combination of all the technical, administrative and management activities
during the life cycle of parts intended to maintain or restore the condition of the system in
which the required function can be provided.

Although all interpretations of the concept of maintenance have their specifics,

general maintenance can be understood as the implementation of all measures neces-
sary for a machine, plant or entire factory to function properly, developing perfor-
mance within the prescribed limits, or with the required effects and quality, without
failure and with the prescribed providing the environment, and under the assump-
tion of good security of all conditions, i.e., with the necessary logistical support,
Majdandžić [6].
As part of the business system, the maintenance system has the role of integrating
the organizational structure, technology and information system and activating all
available material and human resources within the design / redesign process. Main-
tenance is the process of providing conditions for fulfilling the function of the goal
in relation to defined costs, effectiveness, availability, reliability, etc.
The success of maintenance is determined by providing the greatest possible
degree of certainty over time. Ensuring the functional state of the system is possible
only with the interaction of the maintenance system with parts of the business system
and environment. In this case we are talking about a system approach to maintenance,
Fig. 2.
Maintenance is the process of engaging all the resources and activities necessary
to meet the predetermined characteristics and conditions of the observed system
within a given time interval.

Fig. 2 System approach to maintenance

The Importance of Technical Diagnostics … 147

With the construction and establishment of the production system, the need has
been imposed to define certain procedures that ensure the elimination of possible
causes of failures in the operation of the system. Working condition and condition in
failure are the main characteristics of the production system. The permanent goal of
the maintenance process is to minimize the time of failure and to provide working
conditions in which the production system fulfills the planned function of the goal.

2.1 Maintenance Strategies and Concepts

Ensuring the basic maintenance functions—security of operation of the technical

system by providing the highest possible level of reliability and availability, is
possible if there is a clear maintenance strategy. Maintenance Strategy is the
highest level of maintenance determined by the management in accordance with
the requirements:
• ensuring the maximum availability of the technical system which enables the
performance of the required function with minimal downtime/costs,
• meeting safety requirements for workers and equipment,
• achieving the necessary durability of the technical system,
• quality assurance of services provided by maintenance processes.
The maintenance strategy is determined based on the characteristics and needs of
other parts of the business system, Fig. 3.
Historically, the meaning and role of maintenance have changed over time. Unlike
the traditional understanding, in which the maintenance of parts of the technical

Fig. 3 Conditions for choosing a maintenance strategy, [2]

148 D. Lj. Branković et al.

system was entrusted to individuals and in which maintenance was seen as a non-
profit business entity, the basic characteristic of modern maintenance is the organi-
zation and implementation of team activities to prevent and eliminate failures at the
lowest possible cost gives the maintenance service the possibility of a great influence
on the economic results and a large share in the creation of the value of profit in the
operation of the system. In Table 1 presents the basic characteristics of the old /
traditional and new / modern understanding of maintenance.
There are a large number of ways in which the division of maintenance strategies
can be performed, but in each one a connection with the way of performing main-
tenance procedures which characterizes the basic division of maintenance strategies
can be seen, Fig. 4.
The most important characteristics of basic maintenance strategies are:
• Corrective maintenance strategy or so-called. “Wait and see” strategy—main-
tenance activities are performed after the technical system fails. There are no
previous planned interventions other than regular cleaning and lubrication oper-
ations. The justification of this strategy is in cases of simple, cheap and less
important elements of the technical system,

Table 1 Characteristics of traditional and modern understanding of maintenance organization

Characteristic Old (traditional way of New (moder way of
maintenance) maintenance)
Orijentation On repair On reliability
Akctivity Repair Improve
Concept “Firefighting” anticipate, plan and program
activities
Executor Repairman member of the business team
Occurrence of failure Failure elimination elimination of causes
Consequence in accordance Maintenance costs Increasing of working time
with the set goal reduction
Business advancement “Action program of the Continous improvement
week/month”
Belief Failures are inevitable Failures are only exceptions
Priorities defining Failures that occurred Eliminating the cause of the
failure
Number of failures Large Just a few failures
Share of planned operations Low High
Number of complaints Large Small
Reliability level Low High
Nivo troškova održavanja High Small
Planning period Short-term Long-term
Investments character Non-profit investment Attracts investment
The Importance of Technical Diagnostics … 149

Fig. 4 Basic division of maintenance strategies

• Opportunistic maintenance strategy—upgrade of the “Wait and see” strategy

which refers to the introduction of elements of preventive maintenance only in
cases of parts of technical systems where it is necessary,
• Preventive maintenance strategy—methods of prevention of failures based on
the scientific theory of reliability of the elements of the technical system. It is
used in cases of more expensive and responsible equipment while providing the
desired level of safety for both workers and equipment,
• Condition based maintenance strategy—by monitoring certain technical param-
eters of the condition of parts of the technical system, a decision is made on the
implementation of maintenance activities. It is used in cases of monitoring and
maintenance of very expensive, responsible and complex technical systems such
as aircraft, power plants, process industry and the like,
• Risk based maintenance strategy—upgrading the “Condition based mainte-
nance” strategy in the case of parts of complex high-risk technical systems such
as nuclear power plants, petrochemical plants and the like.
In addition to the above basic maintenance strategies, in practice the following
ways of organizing maintenance activities in industrial systems can be found:
• Logistics maintenance strategy—if the priority of the maintenance strategy is
based on the development of elements of logistics support such as administrative
tasks, ergonomic and environmental factors, realization and distribution of profits,
then we are talking about logistics maintenance (Logistics Maintenance). This
strategy basically supports the fully installed equipment with an emphasis on its
design, production of equipment and conditions of use. Within the basic activities,
the production and repair of spare parts in service or own workshops is realized,
as well as the procurement of raw materials and completely new spare parts,
• Total Productive Maintenance strategy (TPM)—higher maintenance strategies
from the late twentieth century that the staff, i.e., puts the individual in the produc-
tion process at the center of interest in relation to the disposal of equipment and
means of work. The common goal of all employees in the company is to raise
the performance and results with a new way of organizing the system. Compared
to the competition, which uses the same equipment, a better business result is
150 D. Lj. Branković et al.

achieved. Equipment used in production processes should have a projected level

of capability with the maximum possible return on investment. The application of
the concept of eliminating “small” losses and changing the overall philosophy of
work and business requires full commitment of all participants in the production
process and a high level of awareness and motivation to work,
• Terotechnological maintenance strategy—the greatest importance and role
in the processes is given to maintenance professionals during the design and
construction of production systems in all phases of the life cycle (lifespan). From
the very initial idea, through procurement, installation, operation and write-off,
i.e., withdrawal of equipment from use, qualified maintenance persons have an
active role in providing functional and economic support to all processes in the
system,
• Expert systems strategy—the basis of expert systems consists of large databases
supported by computer systems that have data, knowledge and reasoning
mechanisms, i.e., artificial intelligence, Fig. 5.
• Self-maintenance strategy—the highest level in the functioning of maintenance
characteristic of fully automated companies with a high degree of automation
such as e.g., technological systems with robotic lines,
• Maintenance-free maintenance so-called strategy—replacement of parts of
technical systems, subassemblies or assemblies after the expiration of the working

Fig. 5 Example of configuration and operation of an expert system

The Importance of Technical Diagnostics … 151

function, which means work until the failure condition occurs, without the inten-
tion to repair the failed element and bring it back to functional condition. The
strategy is acceptable in all cases in which it is economically unacceptable to
engage its own maintenance or services of external companies in relation to the
availability and price of a completely new part, subassembly or assembly. If the
lifespan of such a part or assembly is long enough to ensure the write-off of the old
part while achieving the planned productivity within the technological process,
this strategy provides for the complete replacement of the failed part with a new,
spare part or assembly.

2.2 Condition Based Maintenance Strategy

A large number of researchers have defined the concept of maintenance strategy

according to the condition. Bengtsson [5] states that maintenance according to condi-
tion belongs to the strategy of preventive maintenance based on performance and /
or parameters of subsequent diagnostics of technical condition. Butcher [7] defines
maintenance technology as a series of maintenance activities in terms of assessing the
condition of equipment in real or near real time. Condition assessment is obtained by
purpose-built sensors and/or external tests or measurements using portable measuring
devices/instruments, [7]. According to Kothamasu et al. [8], a state-of-the-art main-
tenance strategy is a decision-making strategy related to maintenance activities based
on observation of the state of the system and its constituent components, [8].
In contrast to the basic maintenance division shown in Fig. 4, in the Shin and
Jun [9] and Erbe at all studies [10], maintenance is divided into three basic cate-
gories: preventive, corrective, and condition-based maintenance. Preventive mainte-
nance refers to periodic scheduled inspections of elements of technical systems at
clearly defined time intervals. Although it has the properties of failure prevention,
the meaning of the maintenance strategy according to the condition is the fact that the
degradation of the condition does not occur immediately because usually processes
occur that translate the system into an abnormal state.
Condition based maintesnance according to EN 13,306 [4] is defined as:
Preventive maintenance based on monitoring of performance and / or parameters and
subsequent corrective activities.

Monitoring is defined according to standard EN 13,306 [4] as:

Activity realized manually or automatically, with the aim of observing the current state of
the system.

The main difference between monitoring and inspection is that monitoring is used
to assess any changes in system parameters over time. Monitoring can be continuous,
within time intervals or within a certain number of operations. It is usually realized
in the operating conditions of the system.
152 D. Lj. Branković et al.

Parameter and performance monitoring can be predetermined as continuous or

intermittent. Preventive maintenance strategy uses the conditions and means of diag-
nostic devices to analyze the current state of the observed parts of the system and
using the knowledge of technical diagnostics, operating conditions of equipment
and technological procedures within production processes, determines appropriate
preventive maintenance activities. By using maintenance intervals and condition-
based maintenance tasks, maintenance activities become predictable. In this sense,
standard EN 13,306 [4] defines predictive maintenance as:
Condition based maintenance carried out after the forecast obtained by analysis and
assessment of unambiguous parameters of system degradation.

The term system degradation is defined according to standard EN 13,306 [4] as:
An irreversible process of disrupting one or more characteristics of a system either by time
or by the action of an external.

The basic determinant of the term degradation is the possibility of the system
coming to a state of failure. Often the term degradation also refers to wear. In terms of
the previous definition, the notion of degradation can also be explained as a notion of
disruption of system operating conditions because the basic task of the maintenance
strategy according to the condition is to diagnose the disorder before it becomes an
irreversible process.
Looking at the financial aspects, it is very important to determine the economic
feasibility of a particular maintenance strategy. In the case of high capital importance,
criticality and safety of the complete system, safety of the production process, etc.,
condition based maintenance is a better solution, although it is a significant invest-
ment in diagnostic equipment. The basic precondition is that the observed system has
measurable parameters of the work process. There are numerous analysis decisions
on which maintenance method and strategy to apply for each case, Al-Najjar and
Alsyouf [11].
The basis of monitoring is the existence of the condition based maintenance,
which can be defined as:
A system that uses state-of-the-art maintenance to determine and plan predictive maintenance
activities independently or in interaction with other systems or humans.

Tavner et al. [12] emphasize the growing importance of the application of Condi-
tion Monitoring in the automation and reduction of human oversight within power
plants. They analyze the effects of monitoring the condition of electric rotating
machines with techniques that are applicable when these machines are in opera-
tion, neglecting the many off-line inspection techniques. The results of monitoring
the condition relate to: electromagnetic and dynamic behavior of electrical machines
(with special emphasis on control with modern power electronics) and the behavior
of the insulation system of electrical machines, [12].
Condition based maintenance has the function of converting measurable energy
input quantities (temperature, vibration, noise …) into the desired shape or effect
(condition of elements, comparison with allowed values for safe operation, forecasts
The Importance of Technical Diagnostics … 153

for failure, etc.) in space and time. The maintenance system according to the condi-
tion uses the characteristics and parameters of the monitoring technique (vibrations,
sound, heat …) in terms of detecting performance disturbances or changes in the
operating characteristics of the observed system.
Based on the previous definitions, techniques or activities of the monitoring system
can be:
• subjective—noticed and observed by workers,
• objective:
• use of portable measuring devices—manual,
• stand-alone continuous systems for measuring predefined parameters.
It should be noted that monitoring activities are usually carried out during the
normal operation of the system, which in many cases represents the ability to obtain
relevant data on the state of the system. Monitoring can be planned, on request or
continuous.
Following the characteristics and parameters of monitoring, the maintenance
system according to the condition should determine the subsequent activities that are
realized within the predictive maintenance. Advanced maintenance systems, using
the support of computer programs, have the ability to autonomously plan mainte-
nance activities. They provide input on the current state of the system and an estimate
of where the observed system should be in the future given the required operating
conditions. After predicting the state of the system, the role of man is reduced to
deciding on the schedule of maintenance activities. This integration of man into an
integral part of the system creates the possibility of realization of all set requirements
within the condition based maintenance.

3 Technical Diagnostics—Basis of Condition Based

Maintenance

3.1 The Concept of Technical Diagnostics

In the last few decades, the term technical diagnostics has penetrated into all branches
of technology, especially in electrical engineering and mechanical engineering. This
term means all measures used to assess the current condition or give a forecast of the
behavior of machinery, devices, equipment and the like, without disassembly / disas-
sembly or destruction in a certain period of time. The basic task accomplished by
technical diagnostic measures is the timely prediction of the occurrence of malfunc-
tions of parts of the technical system. The implementation of technical diagnostic
measures increases the overall degree of reliability, availability and effectiveness of
the complete technical system.
Technical diagnostics is the science of recognizing the condition of a technical
system and thanks to systems for monitoring the size of the condition, the occurrence
154 D. Lj. Branković et al.

of malfunctions can be predicted. By eliminating the factors of possible system

failure, the efficiency and reliability of the system increases [13].
Technical diagnostics provides a large number of benefits both for the normal and
safe operation of the production technical system and for achieving the function of
the maintenance goal. The most important characteristics of technical diagnostics
are:
• optimization of planning of preventive repairs and overhaul works of the plant
based on the assessment of the actual condition of the technical system,
• avoidance of conditions during operation that may cause the technical system to
fail,
• reduction of downtime and unplanned activities that result in optimization of
reliability and prevention of emergency situations,
• extension of the service life of the equipment as the results of technical diag-
nostics determine and indicate the terms of replacement of individual parts and
assemblies,
• forming a database suitable for easier viewing and access for the purpose of future
investments in critical places and obsolete equipment.

3.2 Condition, Changes and Condition Control

of the Technical System

Technical diagnostics is a kind of tool for assessing the condition of the components of
the technical system. The state of the system is the basic characteristic of the observed
technical system. The state of the technical system determines a set of physical
parameters whose monitoring and comparison in time can define the behavior of the
system at a given time and the forecast of the state of the system in the future after
the action of internal or external disturbances, Fig. 6.
The term system state can be described by the following features:
• working condition,
• state of dismissal,
• quality of functioning,
• safety and security, etc.

Fig. 6 General overview of the technical system [14]

The Importance of Technical Diagnostics … 155

The failure of a technical system is defined as the cessation of the ability of an

element or system as a whole to perform the functions for which it was designed.
The technical system may have impaired performance due to the action of various
factors that can cause different degrees of damage.
Causes of impairment of working ability can be:
• technological (conditions of technological processes),
• construction (errors during the process of construction of parts),
• exploitation (periods of operation / use, methods of use and maintenance),
• wear of parts (friction, wear …).
The assessment of the condition of the diagnosed object is defined by the limit
values of the corresponding characteristic or parameter. The values that determine
the normal functioning of the system are conditioned by the system design, method
of construction, mode of functioning and conditions, i.e., change of environmental
conditions (Design, Production Technology, Operational, Change of Condition—
DPTOCC). Comparing the measured values of the observed parameter (values that
condition normal functioning) with the limit prescribed values creates the basis for
deciding whether the observed object or part fulfills the designed objective function
or it is necessary to perform appropriate maintenance activity in terms of adjustment.
The state of an object is described by a certain set of parameters (characteristics),
which should perform the designed function, under certain conditions and in a certain
period of time. The characteristics of the monitoring object, depending on the basic
purpose of the observed system, can be: voltage, noise level, fluid flow, operating
temperature, bearing vibrations, wall thickness of a pressure vessel, etc. Changes
in the parameters can cause a decrease in the degree of functionality of the object,
and in a longer period of time it can lead to a complete interruption of the projected
function, i.e., occurrence of failure.
Parameters that affect the function of the object can be with:
• constant (monotonous),
• growing or
• by a sudden effect on the condition of the observed object.
The basic precondition for making any decision related to the further use of the
observed facility is knowledge of working conditions and allowable wear limits.
The wear limit of the observed part represents the boundary between the operational
usability of the elements or parts of the system and its damage. The task of tech-
nical diagnostics is to determine the technical condition of the system elements at
a precisely defined time and precisely defined limits of accuracy. Procedures for
determining the technical condition of the observed element of the system include
appropriate instrumentation or observations by the senses of a diagnostic specialist.
The state limits of the system elements in diagnostic procedures, which are of
special interest, refer primarily to the states of damage (fatigue, wear, wear, etc.).
The criteria for determining the boundaries are:
156 D. Lj. Branković et al.

• technical–technological (production, assembly, etc.),

• economic (replacement cost of system parts, risks, etc.),
• ergonomic,
• security, etc.
For example, technical–technological criteria can be the thickness of the ther-
mally treated layer of the shaft, the clearance of the shaft and bearing assembly, etc.
Economists take into account the cost of construction / repair, etc. Safety and secu-
rity must be considered in accordance with the possible consequences of the failure.
Ergonomic aspects have recently been increasingly analyzed due to the human–
machine relationship (e.g., vibrations, noise, temperature, various types of radiation,
required manpower for maintenance, etc.). Most often, several criteria are taken into
account when assessing the situation. Boundaries can be determined based on:
• empirical procedures or experiences,
• experimental research within real systems,
• tests on test benches,
• theoretical calculations.
Condition assessment is a very complex procedure and its definition depends on
a set of measures that need to be taken to maintain the functionality of the system in
the next period of operation.

3.3 Technical Diagnostic Processes

The existence of a technical system with measurable parameters is the first prereq-
uisite for establishing technical diagnostics. Technical diagnostics is a process
composed of certain activities or steps that include:
• installation of measuring sensors on the object (observed system),
• stabilization of the operating mode (systems and equipment) to obtain relevant
results,
• measurement of diagnostic signals,
• comparing the obtained values with the prescribed values from the norms,
• making a conclusion about the current state of the elements of the system under
observation,
• forecasts of the future state of the system, with the definition of recommendations
on the sustainability of that state.
Defining critical places in terms of system parts that by their function can cause
failures of parts or a complete system narrows the process of eliminating potential
risks that endanger the system. Each individual critical location should contain some
of the measurable physical quantities that change over time during operation. Sensors
for monitoring and measuring the characteristic values of the system state are placed
in critical places. The process involves measuring and converting these quantities
The Importance of Technical Diagnostics … 157

Fig. 7 Technical diagnostic processes

into values that can be displayed in a clear and concise manner. The measured values
are compared with the prescribed values from the standards, on the basis of which a
conclusion is drawn about the state of the operating characteristics of the system.
If we monitor the behavior of the observed part of the technical system by
collecting, measuring, converting and processing certain physical quantities that
describe the state of the technical system, we can talk about the diagnostic procedure
of the technical system, Fig. 7.
The processes of technical diagnostics are related to the monitoring of the
parameters of the state of the technical system by the workers / operators on
machines/devices, maintenance workers trained to handle diagnostic equipment or
by the equipment manufacturer through the service of technical support.

3.4 Technical Diagnostic Methods

Measurement method, i.e., technical condition identification is a set of procedures

that define the relationships between measured quantities. There are several different
criteria on the basis of which the division of measurement methods is performed.
If the measured quantity is currently read as an absolute value then we are talking
about an absolute method. If the ratio of measured and predefined other quantities
is determined, then it is a relative method. Considering whether it is a measure-
ment on the object of measurement without taking a sample or measurement in an
experimental way through the examination of the characteristics of the materials
that represent the object of measurement, we are talking about methods with or
without destruction of the material. If the measured quantity is read directly, these
are direct methods. If the measured quantity is obtained by calculation through a
predefined relation, we are talking about indirect methods. The contact measure-
ment method uses measuring devices that are in direct contact with the measuring
medium. Otherwise, if the measurement is realized through the media or wave, then
we can talk about non-contact methods. If the measurement refers to a large number
of individual results on the basis of which a conclusion is made, it is a matter of differ-
ential methods. Simultaneous measurement of several parameters defines complex
measurement methods.
158 D. Lj. Branković et al.

When choosing a method for identifying the state of the technical system, it is
essential that the measurement procedures, i.e., diagnostics or tests to check the tech-
nical condition of the elements or the overall system are realized without destroying
the elements and dismantling the technical system, because with each disassembly
and reassembly there is a change in the geometry and conditions of contact. This
leads to an increased intensity of wear due to the repetition of the running-in phase.
In general, all procedures or methods of diagnosing the state of technical systems
can be divided into: subjective and objective procedures.
Subjective methods of technical diagnostics are based primarily on the expe-
rience of responsible technical persons in charge of monitoring and forecasting
the current state of the technical system. Certainly, the level of education with the
necessary experience is a key parameter of a successful diagnosis.
The main task when choosing an objective measurement method is to observe
the characteristics of the system and the physical quantities that can be monitored.
The use of objective methods to identify the condition can be based on:
• temperature recording (using thermometers, thermocouples, thermistors, thermo-
sensitive paint, infrared radiation and heat flow),
• vibration recording (speed accelerator and shock pulse displacement meter,
stethoscope) and length measurement: mechanical, electrical, laser rangefinder,
etc.,
• registration of defects in the material (magnetism, penetrating colors, eddy
currents, radiography, ultrasound, hardness, sound resonance, corona emission
and optical fiber),
• registration of deposition, corrosion and erosion (ultrasound, radiography,
cathodic potential and using weight),
• flow and pressure recording (using a neon-Freon detector, manometer and micro
turbine meter),
• registration of electrical parameters (cable fault meter, frequency meter, phase
angle meter, universal instruments for checking electrical circuits and voltage
meter in transients),
• load recording (dynamometers: mechanical, hydraulic and electrical),
• registration of physical and chemical quantities (spectrograph, humidity meter,
O2 and CO2 gas meter, pH meter, viscosity meter and meter of the presence of
metal particles in the lubricant), etc.
Objective methods of identifying the state of the technical system cover only a
certain segment of measurable values and determining their relationship with changes
in system state parameters, characteristics of their change, diagnosing the state of
system parts in a certain time interval and predicting future state of performance.
In practice, there is no unambiguous criterion for the introduction of a single
diagnostic system that would be appropriate for all technical production systems.
The choice of the control-diagnostic system concept to be proposed and selected
depends on a number of factors, namely:
The Importance of Technical Diagnostics … 159

• the nature of the technological or production process,

• technical characteristics of production capacities (media, machines, equipment
…),
• availability of measured quantities or diagnostic parameters,
• availability of measuring equipment on the market,
• the existence of trained workers for the implementation and monitoring of the
diagnostic system,
• financial capabilities of the business system, etc.
Depending on whether the definition of the cause of failure of the technical system
is realized during the work or operation or within the downtime of the plant, we can
talk about exploitation and overhaul diagnostics. Both terms are an integral part of
the condition based maintenance.

3.5 Methods of Measurement and Analysis of Vibrations

The basic characteristic of technical systems that have moving parts is the creation of
mechanical forces that are the source of the excitation of oscillation of the complete
system or its individual parts. This oscillation, if it is continuous for a long period
of time and if it goes beyond the limits that disrupt normal and uninterrupted work,
can lead to:
• wear of system elements,
• changes in the geometry and position of system parts,
• failure of electronic components,
• damage to the insulation of cables in contact,
• disturbance of workload balance,
• disturbance of lubrication between moving parts,
• the appearance of noise and causing damage and disease in humans,
• material fatigue, damage to parts and eventually fracture which causes the
complete system to fail.
Precisely for the purpose of recognizing and monitoring the state of the system
in the conditions of vibration occurrence and avoiding harmful events, they have
developed methods of vibration monitoring, which we often abbreviate as vibro-
diagnostics (Fig. 8).
The main characteristic of vibration systems is the intense interaction of forces
and vibrations, Fig. 9.
There are numerous reasons for the development and application of vibration
analysis methods in industrial practice. Randal [15] points out that vibration analysis
is by far the most common method for defining the state of a machine because it
has a number of advantages over other methods. The main advantage relates to the
immediate response to external change and can therefore be used for continuous or
intermittent monitoring. Modern signal processing techniques enable the detection
160 D. Lj. Branković et al.

Fig. 8 Reasons for vibration monitoring

Fig. 9 Vibration conditions

and filtering of very weak, masking signals that can be indicators of damage to the
monitored elements of the technical system, [15].
The main benefit that can be expected after the introduction of vibro diagnostics
can be described through:
• elimination or minimization of unplanned downtimes that cause the technical
system to fail,
• minimizing corrective maintenance and transition to preventive maintenance,
• reduction of the risk of material damage,
• increasing the operational reliability of the technical system,
• increase in the mean time between cancellations,
The Importance of Technical Diagnostics … 161

• introduction of the “Just-In-Time” model in the spare parts business,

• increasing the operational reliability of the technical system,
• improving maintenance management.

3.6 Concept, Classification and Digital Signal Processing

In science, the term signal is related to a physical quantity that depends on time,
spatial coordinates or some other independently variable. A signal is the carrier of
a message that was created in the process of signal generation. Given the current
level of development of technology and especially electronics, when we say a signal
in most cases we mean electrical signals that are suitable for further processing and
presentation.
There are two basic signal classes: stationary and non-stationary, Fig. 10.
If we can predict the value of the signal size at any given time then we can talk about
deterministic signals. Random (stochastic) signals are characterized by the property
of uncertainty, i.e., their occurrence and value cannot be accurately predicted at some
point in time in the future. To represent such a signal using a time function, we can
only use known values from the past and in that case we are talking about random
functions.
Non-stationary continuous signals can be viewed as random signals or can be
divided into smaller intervals and viewed as transient. If the beginning and end of
the signal observation are related to a constant value, e.g. most often zero value, then
we can talk about transient signals.
By nature, signals can be divided into: continuous and discrete. Continuous signals
correspond to one of the continuous functions in time. Discrete signals appear as
arrays of separate elements and can have a finite number of different values.

Fig. 10 Signal classifications

162 D. Lj. Branković et al.

3.6.1 Frequency Signal Analysis

Frequency domain signal analysis is one of the most acceptable methods today. As
the signal recording is always in the time domain, this means that the signal recorded
in this way should be transformed into a frequency domain.
The theoretical basis of the previously mentioned transformation is the possibility
of decomposing a complex function of time into a set of sine functions:

x(t) = A1 sin(w1 · t + ϕ1 ) + · · · + An sin(wn · t + ϕn ). (1)

where:
Ai —amplitude of sinusoidal function,
ωi —circular frequency of the i-th sinusoidal function,
ϕi —phase of the i-th sinus function
The decomposition of the time function x(t) into a continuous complex spectrum
in the frequency domai F( jw) is performed using the Fourier integral which reads
(Fig. 11):

Fig. 11 Fundamentals of oscillation theory

The Importance of Technical Diagnostics … 163
 +∞
F( jw) = x(t) · e− jwi · dt.) (2)
−∞

The Fourier integral is derived based on the concept that a periodic function
expressed in the Fourier order can fully represent an aperiodic function if its period
tends to infinity. In expression (2) the function F( jw) is called the spectral density
of amplitudes.
In general, there are two ways to transform a signal from a time domain to a
frequency domain. The first way is transformation using specialized devices, spec-
trum analyzers that at first glance resemble classical oscilloscopes. By establishing
signal digitization, spectrum analyzers immediately perform Furier signal integra-
tion and plot the result on a monitor that is an integral part of the device. Another
way is to digitize the help of analog-to-digital (A/D) converters that are built into the
computer.
Both signal transformations from the time domain to the frequency domain use the
developed Furier Transformation FFT algorithms, which, depending on the number
of measurement results, save up to 99% of the time during the calculation (Fig. 12).
There are numerous scientific papers that show the advantages of applying
advanced methods of frequency signal processing. Lee [16] describes the devel-
oped concept of the Acceleration Enveloping method and the result of applying the
method to early detection of rolling element damage in a bearing. By proper anal-
ysis of the Wave form using Enveloping, it is possible to determine the cause of the
potential problem that can arise from the bearing cage, the outer or inner ring or the
rolling element of the bearing itself. In this particular case, the detection of problems
on the outer ring of the bearing is described, showing how this observed frequency
could be isolated and differentiated in the total vibration spectrum, [16]. Kazzaz and

Fig. 12 Furier signal processing (FFT) vibration monitoring

164 D. Lj. Branković et al.

Fig. 13 Principle of frequency analysis [18]

Sigh [17] cite an example of on-line digital signal processing obtained from a moni-
toring system for monitoring process state parameters using the C++ programming
language. The obtained signals are analyzed in detail through specially developed
modified algorithms in order to obtain the most accurate state of technical systems.
By using quality techniques to protect the input signals that come into the monitoring
system, it is possible to timely warn operators of the possibility of serious equipment
failures, [17].
Figure 13 shows the principle of frequency analysis in the case of monitoring the
operating parameters of the electric motor assembly and the actuator (reducer, mixer,
mill, etc.). Vibration level / amplitude measurement and frequency spectrum analysis
can provide an answer to the condition of the operating element and the forecast of
the behavior of operating parameters in the future.
All physical structures and machines, which are connected to the rotating compo-
nents, stimulate vibration. Veltkamp [19] emphasizes that vibrations generated on
machines have become a well-used parameter for assessment within condition moni-
toring. It is one of the most versatile techniques that can detect about 70% of common
mechanical faults associated with rotating machines [19].
The measured values are compared with the values prescribed by the standard for
each type of equipment or working element. An example of ISO 10816–3 defining
oscillation amplitude values and allowable limit values for rotating elements is shown
in Fig. 14.
In addition to accelerometers, there are other ways to measure and analyze vibra-
tion signals such as ultrasonic defect detection. In their study, Kim et al. [21]
give the results of testing low-velocity bearing vibration signals where significantly
The Importance of Technical Diagnostics … 165

Fig. 14 Limit values of oscillation amplitudes of rotating parts [20]

better results of ultrasonic measurement techniques are shown compared to vibration

acceleration measurements, [21].

3.7 Importance of Technical Diagnostics on the Reliability

of Technical Systems

The application of technical diagnostics within the concept of maintenance according

to condition is a set of activities that actively positively affect the effectiveness of
technical systems: availability, reliability, sustainability, safety, efficiency, etc., as
well as reducing energy consumption and environmental pollution. The result of
condition-based maintenance activities usually allows the user sufficient time to
plan, schedule, and make necessary repairs before the equipment reaches a technical
failure condition. This can avoid major failures and costly repairs.
Monitoring, collection and analysis of signals that characterize the behavior of
the technical system provide the possibility of early detection of the potential cause
of future failure because it is possible to signal the occurrence of the cause of failure
long before the situation of increased noise or temperature that inevitably precedes
Fig. 15.
Previous long-term practice in the maintenance of technical systems has pointed
out that the best way to optimize the performance of system elements has been
166 D. Lj. Branković et al.

Fig. 15 Example of early warning of the possibility of failure

overhaul or replacement at a fixed interval. This is based on the assumption that

there is a direct relationship between the amount of time (or number of cycles) of
equipment being monitored and the likelihood of failure. Moubray [22] emphasizes
that the relationship between working time (age) and failure is correct for some types
of failure, but that it is no longer credible because the equipment is now much more
complex and sophisticated than it was fifteen years ago. This way of understanding
maintenance has led to significant changes in techniques for predicting equipment
failure dates. Unless there is a clear indication of the existence of age and equipment
obsolescence factors, overhaul at fixed intervals or replacement may have very little
or no significant effect on improving the reliability of rotating equipment. In this case,
it is more realistic to develop maintenance strategies that will assume that equipment
failure can occur at any time, and not at a fixed time.
Choosing the right maintenance strategy is a very professional and sensitive
process. In the long run, it can result in a series of events that have a decisive impact
on the overall result of the business system. To choose an appropriate maintenance
strategy within the production system in continuous working conditions is to plan
the projection of the operation of the complete system and the conditions in which it
is provided. Proper choice of maintenance strategy within the continuous production
process means providing the following conditions:
• optimal number of qualified maintenance workers deployed within shift and
intervention work,
• the optimal number of spare parts that can be found,
The Importance of Technical Diagnostics … 167

• availability and application of modern technical systems for monitoring operating

parameters,
• positive impact on the degree of effectiveness of the production system (reliability,
availability, readiness),
• planning and implementation of repairs in terms of failure prevention,
• cost tracking control,
• application of safety measures at work,
• positive impact on environmental protection.
Providing the mentioned conditions means optimizing maintenance procedures.
Wrong or inadequate choice of maintenance strategy means in the short term perhaps
a better financial result, considering that it is mostly thought of avoiding finan-
cial investments in the maintenance service (personnel, equipment, organization,
management…). In the long run, the wrong strategy inevitably requires large invest-
ments. The term investment refers to the investment in equipment, professional staff
and the improvement of technical means used by the maintenance service in terms of
providing the projected strategy. Today, the modern understanding of maintenance
does not view this phenomenon as a “repair” but maintenance should be approached
as a way of providing targeted activities in terms of failure prevention. The correct
choice of maintenance strategy is a rational planning of operating costs of the entire
production system over a long period of time.

4 Application of Technical Diagnostics in Industrial

Practice

Technical diagnostics is a very suitable and desirable methodology for application

in industrial production systems in terms of control and insight into the state of the
elements of technical systems and prediction of their state and expected behavior in
the future. As an illustration of the manner of application of technical diagnostics
and significance, i.e., effects that are achieved in the process of providing normal
conditions for the functioning of technical systems, describes a specific event of
observation and repair of a critical situation in the working position of high voltage
motor technical system paper machine within the industrial system for production of
toilet paper from Banja Luka, Republika Srpska, Bosnia and Herzegovina, Fig. 16.
Paper machine is a complex technical system whose ultimate function is the
production of low-gram hygienic paper with a grammage of 15–40 g/m2 .
The complete technological process of preparation and creation of paper tape
consists of the following technical units:
• paper pulp preparation plant—as an input semi-finished product paper pulp ground
in water in the percentage of dry matter up to max. 5% is used,
168 D. Lj. Branković et al.

Fig. 16 Paper machine of an industrial plant in Banja Luka

• machine part of the plant paper machines—mixing devices (mixing tubs, pumps,
fresh and regenerated water tanks, …), homogenization (equalizers, mills, concen-
tration meters,…), transport (pumps, pipelines and related instrumentation),
purification (vertical and centrifugal purifiers) and central pump for delivery of
prepared paper pulp to paper machine,
• paper machine—a technical system consisting of several functionally dependent
units from which can be distinguished:
– wet part—subsystem for injecting paper pulp (flow), a system of rollers that
move the sieves through which water is separated from the pulp by pressing
and presses (rotary rollers for additional squeezing of water from the pulp by
pressure and vacuum),
– dry part—cylinder for drying paper with steam and creping with the help of a
system of knives, system for winding already formed paper tape created in the
process of creping,
– recuperation—a system related to the provision of conditions for drying paper,
which includes: steam–air tube exchangers, high-voltage motors for hot air
fans, hot air drying hood and all accompanying instrumentation (condensers,
air flow regulators, thermocouples, etc.)
The output of the described technological process is a semi-finished product—a
paper strip that is wound on winding shafts and further handed over or on a cutting
machine to appropriate formats needed for processing in own paper confectionery
plants or for sale as a final output semi-finished product with defined specification
of quality parameters.
The Importance of Technical Diagnostics … 169

4.1 Description of the Reference Event

In order to analyze the importance of the application of technical diagnostics within

the production system that operates in a continuous mode, a specific event that
occurred during one of the regular diagnostic inspections of technical assemblies of
a paper machine from 2019 is described.
The maintenance service of this production system uses three ways of preventive
diagnostic examinations and condition control of the characteristic elements of the
technical system of the paper machine, as follows:
• on-line diagnostic system for monitoring the state of vibration (at critical positions
presses the paper machine, Fig. 17) and
• periodic measurement of a number of predefined elements of the technical system
of the paper machine with a portable vibrodiagnostic device and subsequent
analysis of the measurement value by an external specialized company,
• use of own portable measuring device for instant control of vibration condition
by shift maintenance workers.

Fig. 17 Structure of the on-line diagnostic system at paper machine press positions
170 D. Lj. Branković et al.

In the case of the described technical system of the paper machine, one of the
regular preventive activities of the maintenance service is the contracted vibrodi-
agnostic examinations, which are realized by an external service from the Czech
Republic. Experts from a specialized external company come to the site once or
twice a month and inspect the previously defined measuring points of technical
functional elements of the paper machine (drive and guide rollers, presses, drying
cylinder) and auxiliary units (high voltage recuperation motors with couplings and
fan bearings, mill motors and vacuum plants). Examples of parts of the diagnostic
report with the values of the measured parameters are shown in Figs. 18, 19 and 20.
Portable vibration measuring devices such as Mikrolog CMVA 65 and Microlog
CMVA55 are used as on-site vibration diagnostic measuring equipment by the
external company, Fig. 21.
These devices have the ability to measure, archive and instantly analyze the
vibration spectrum.
During the regular control and measurement of the vibration condition parameters
of the bearings at the position of the high-voltage motor for the drive of the hot air
recirculation fan for drying the paper strip on the dry side of the hood, on March
21, 2019, the measured values of bearing oscillation vibrations at position L2 are
significantly higher than allowed, Fig. 22.
Two methods were used to evaluate the motor bearings condition implemented
in SKF Microlog frequency analyzers: “Enveloping” and HFD (High Frequency
Detection) methods (Fig. 23).
On botht trends diagrams is a visible increase of the overall values that indicate
worsening of the bearing condition. On the Fig. 24 is frequency spectrum measured
with Enveloping method and on the Fig. 25 there is time domain spectrum of the
measured signal.
In time domain is visible impact occurrence with frequency approximately
575 rpm. This frequency is also clearly visible and dominate in the frequency spec-
trum. Frequency 575 rpm is 0.40 multiplies of the rotary speed of the motor, what
correspond to cage fault frequency of the bearing. Appearance of this frequency
indicates a problem with the cage of the L2 bearing.
Cage problem of the bearing is a severe problem because of the cage separates
the rolling element. In the case of the brake of the cage, the rolling element changes
its position and the bearing is immediately blocked. Rolling element are not rolling
but sliding on the bearing races what lead to high increase of the temperature and in
most cases to the damage of the rotor shaft. Possible is also physical contact of the
rotor and stator, which lead to total destruction of the motor.
Comparing the obtained results with previous measurements (trend/measurement
history) it could be clearly seen that there was a sudden jump in vibration values
at the position of the front bearing of the high voltage motor L2 (side to the motor
fan) and a dangerous situation that can cause a serious problem at any time. on the
bearing itself but also on the motor and the associated motor-coupling-fan bearing
assembly.
The Importance of Technical Diagnostics … 171

Fig. 18 Example of a part of a diagnostic report - values of roller bearing vibrations

172 D. Lj. Branković et al.

Fig. 19 Example of a part of a diagnostic report—analysis of roller bearing condition

The Importance of Technical Diagnostics … 173

Fig. 20 Example of a part of a diagnostic report—condition of engine bearings

174 D. Lj. Branković et al.

Fig. 21 Microlog CMVA65 portable vibration meter and analysis

Fig. 22 Trend of vibration value measurement at the position of bearing L2 of a high voltage hot
air fan motor on recuperation with the Enveloping method

Figure 26 shows the working position of the high-voltage motor for driving the
fan for recirculation of hot air on the recuperation—“dry” side of the hood for drying
paper.
Taught by experience from previous years, when due to sudden bearing failure,
brush problems or voltage problems, on several occasions there were more serious
failures and burns of both high-voltage motors (Figs. 27 and 28) and other motor
units (Fig. 29), a state of high risk for the occurrence of serious emergency situations
was stated.
The Importance of Technical Diagnostics … 175

Fig. 23 Trend of measuring oscillation amplitude values with HFD method

Fig. 24 Frequency spectrum measurement and analysis results

Fig. 25 Measured signal values in the time domain

The conclusion of the expert diagnostic measurement finding is that the cage
problem of the bearing is a severe problem that can lead to immediate damage of the
bearing and after all to the severe motor damage. Based on the expert opinion of the
specialist who performed the measurement and who recommended the replacement
of the L2 bearing of the high-voltage motor for hot air recirculation on recuperation
176 D. Lj. Branković et al.

Fig. 26 High voltage recuperation motor operating position

Fig. 27 Breakdown of the winding of the high-voltage fan motor from 2017

Fig. 28 High vibrations and cracks of the high-voltage motor support, 2018
The Importance of Technical Diagnostics … 177

Fig. 29 DC motor drive failure

in the shortest possible time, the responsible maintenance persons adopted executive
measures and assessed the terms for specific preventive and corrective activities:
• decision to extend operation in high-risk mode with intensive, half-hour temper-
ature measurement at a critical place—high voltage motor bearing by shift main-
tenance with clear instruction on immediate shutdown of the entire production
process if the temperature value rises above 55 °C because it would mean imme-
diate entry into the accident emergency condition. In addition to the measure of
constant temperature control, a written order was issued for intensive lubrication
of this position (grease). The decision and risk-taking for the continuation of work
was made on the basis of a technical assessment of the situation and taking into
account the needs of the production cycle, fulfillment of the production plan and
compliance with obligations to customers,
• preparation of emergency maintenance activities during the next day which were
related to:
– providing the required number of executors (locksmiths and electricians of
the maintenance service, auxiliary technological staff and support of logistics
workers),
– providing the necessary tools and devices—various types of manual chain
hoists, certified ropes, devices for dismantling bearings and couplings and
forklifts,
– providing a crane—the service of an external person since the company does
not have its own means of lifting heavy loads. The crane had to be urgently
announced and delivered to the scene. The position of the high-voltage motor
requires the engagement of a crane from the outside, i.e., outside the produc-
tion hall, so that the manipulation (extraction of the engine from the working
position) is realized in the shortest possible time,
• realization of high voltage motor replacement—emergency maintenance activity.
178 D. Lj. Branković et al.

4.1.1 Remediation of a Critical Situation in the Work Position

Maintenance activities are often very complicated and specific for certain types
of industrial production plants due to numerous external and internal factors that
affect the conditions and manner of their implementation. The concrete case is a
real example of very demanding preparation and realization with the appearance of
several unpredictable situations that needed to be solved during the elimination of
the observed problem. There are several specifics that were present in the preparation
of the implementation of maintenance activities:
• careful planning of the most favorable period of shutdown of the plant with regard
to the current production plan and compliance with the already defined deadlines
for delivery of semi-finished products to the customer,
• available working space at the working position of the high-voltage electric motor,
• use of the presence of a technician for diagnostic measurement,
• the required number of workers for the intervention,
• crane engagement—external company and availability of equipment,
• preparation of the necessary tools (based on events in the history of replacement
of this engine), …
Realization of repair, i.e., replacement of the problematic high-voltage motor
and elimination of unforeseen problems that occurred during the performance of
maintenance activities is shown in Figs. 30, 31, 32 and 33.
The dismantling of the lightweight partition wall made of 120 mm thick panels
was necessary in order to facilitate the access of the crane and significantly shorten the
time of pulling the high-voltage motor out of the working position. It takes an average
of up to 6 h to remove the engine from the outside and insert the spare engine, while
the same activity realized in the direction of the interior of the production hall was
performed in 16 h due to numerous physical obstacles such as the inner panel wall,
steam and condensate piping. impossibility of reaching the crane into the extraction
space, etc.

Fig. 30 Working and spare high-voltage electric motor of the recuperative fan
The Importance of Technical Diagnostics … 179

Fig. 31 Dismantling a light partition panel wall on recuperation

Fig. 32 Disassembly of the coupling from the working motor, finishing of the spacer ring and
installation of the coupling on the spare motor

Fig. 33 Setting the spare high-voltage motor to the working position using a truck crane

The most complex problem that arose during the realization of the replacement
of the high-voltage motor is the observation of the inappropriate position of the
opening on the spare motor in relation to the place of reliance on the metal structure
and the concrete base. The reason for this problem was the fact that the spare engine
was not manufactured by the same manufacturer and was intended as a replacement
for two similar positions, where the structural measures of the support point were
provided for the second position. After defining the necessary measures for position
correction, new corresponding holes were cut in the metal support, Fig. 34. This
180 D. Lj. Branković et al.

Fig. 34 Correcting the location of the motor support on the concrete base

Fig. 35 Centering the connector and connecting the cables to the high voltage motor

necessary intervention required a significant amount of time relative to the total time
spent on engine replacement.
By installing and centering the coupling as well as connecting the motor to the
existing cables, the basic maintenance activities were completed, Fig. 35 (Fig. 36).
After almost 14 h of continuous work on the replacement of the high-voltage motor
with the successfully completed testing of the motor with parameters that correspond
to the operating conditions, the conditions for the continuation of the operation of
the plant have been created. After the resumption of production, the installation of a
light panel wall was completed, thus completing the complete maintenance activity.
Vibration control measurement after engine replacement at this position showed
satisfactory values.

5 Discussion of Results

The application of technical diagnostic methods has a very positive impact on the
degree of realization of planned production activities of industrial systems. In the
case of the previously described industrial system for the production of toilet paper,
the following characteristic indicators can be highlighted:
The Importance of Technical Diagnostics … 181

Fig. 36 Spare high-voltage motor installed and tested in working position

1. Reduction of the total downtime—by preventing failure states of the elements

of technical systems with greater negative consequences for the technical condi-
tion, there is an increase in the total downtime, which increases the reliability
and overall effectiveness of the technical production system. In the specific
case, the total production stop caused by the decision on the planned shut-
down of the production plant based on the results of the technical diagnostics is
14 h, including unplanned and so-called. hidden problems that had to be solved
operatively during the performance of maintenance works and which obviously
estimated, extended the expected duration of the downtime by some 4 h. If we
compare an almost identical event from the past, which lasted a total of 22 h,
when there was a serious accident (engine burnout, bearing damage to the side
of the fan coupling, waiting for the organization and engagement of an external
company that should provide a crane and the absence of unforeseen / unplanned
additional problems), a total positive reduction of production downtime by 8 to
10 h was obtained in the case of only one accident. If we count from the average
9 h of shortening, i.e., time savings we can estimate the possibility of realization
of production (Po ) in that time from

t
Po = 9h · 4, 8 = 43, 2t (3)
h
where:
• 4,8 tons—the average production of a technical paper machine system for one
hour of operation.
If, given the diversity of the product range, an average gross margin—GM of 290
e/ton of paper produced is assumed, a direct potential production loss (T p ) can be
182 D. Lj. Branković et al.

calculated, i.e., realization of potential profit (Pr ) by selling the realized production
from:

e (4)
T p = Pr = G M · Po = 290 · 43, 2t = 12.580e
t

which is a significant financial amount, given that the estimate relates to only one
accident event in a system with a constant production cycle.
2. Maintenance cost savings—the estimate of maintenance cost savings can be
reported on the basis of a comparison of the costs of preventive planned repairs
and the costs of a possible emergency situation. The common costs in both
cases are: the cost of labor of the maintenance worker during the downtime
and engagement to replace the high-voltage motor and the cost of the crane.
Additional costs in case of an accident
• burnt engine winding cost (C v ):

Cv = PV M · C PV M = 132K W · 100e/K W = 13.000e (5)

where:
PV M —is high voltage motor power (KW ),
C PV M —is estimated cost of engine winding per unit power.
• transport cost (C t ) to the nearest service center (in this case abroad) and the value
of customs duties/taxes = e 1,500 (estimate based on previous cases).
Total savings of currently identifiable maintenance costs (C m ):

Cm = Cv + Ct = 13.000e + 1.500e = 14.500e (6)

Total savings in production costs (avoiding loss of production) and maintenance

(C t ):

Cu = T p + Cm = 12.580e + 14.500e = 27.080e (7)

3. Integration of vibration level monitoring into a single on-line monitoring

system—in this case there is a possibility of connecting vibration monitoring
on the bearings of high-voltage electric motors for driving air fans on recuper-
ation with the already installed on-line diagnostic system Octavis, Fig. 17. The
existing online diagnostic system Octavis, manufactured by IFM Electronic, is a
simple modular system that has the ability to monitor the condition of bearings
and vibration values using high-frequency signal analysis. The four-channel
PLC unit type VSE100 forms the basis of the diagnostic system and has the
The Importance of Technical Diagnostics … 183

possibility of continuous measurement and evaluation of the signal obtained

from the VSE001 acceleration sensor, Fig. 37.
All collected values are stored on a pre-set structure of the software system. As
two four-channel PLC units are currently installed with a total of 6 installed vibration
monitoring sensors at critical positions of the paper machine press (Fig. 38), two free
channels remain currently available for connection and condition diagnostics that
could be used with another PLC four-channel unit to secure both critical positions
of high-voltage motors on recuperation.
The measured results are forwarded with the help of the Ethernet network to the
control system with the VES003 software installed, where they are archived in the
form of long-term trends of individual measurements. The diagnostic system enables
the creation of real-time oscillation frequency analysis on any sensor connected to
the system.

Fig. 37 Electrical cabinet of the vibrodiagnostic system with VSE001 PLC measuring units
manufactured by IFM electronic

Fig. 38 Vibration sensors at measuring points of critical positions presses the paper machine
184 D. Lj. Branković et al.

Fig. 39 Opening of the scolding high-voltage motor recovery

4. Fire protection prevention—the working position of the high voltage motor of

the air blower for drying on the dry side of recuperation is especially interesting
from the point of view of fire protection because a large amount of paper dust
often accumulates in the working environment, Fig. 39.
Certainly, the accident situation followed by the combustion of a high-voltage
electric motor at this position can cause a fire and ignition of paper dust, so moni-
toring the condition of the motor would significantly increase the level of safety and
protection of a particular position in relation to increased fire risk,
5. Technical diagnostics and classification of downtime—a specific event
opened a controversy about the categorization of the resulting downtime of
production, i.e., correct understanding of the character of the maintenance action
after the indication of the technical diagnostics on the occurrence of the viola-
tion of the optimal condition of the monitored component of the element of the
technical system. Based on the established system of evaluation of the activi-
ties of the maintenance service and measuring the overall effectiveness of the
production system of the paper machine, the question arose whether in the
case of action, i.e., maintenance interventions after the results of diagnostic
measurement, analysis and decision to stop the production plant, the production
downtime is treated as a maintenance downtime due to failure, i.e., cancellation
because the production process is currently stopped or should be considered as
a preventive maintenance downtime. The condition-based maintenance concept
belongs to the form of preventive maintenance when decisions on subsequent
maintenance activities are made on the basis of periodic or continuous control
of the technical condition of the elements of the technical system. The form
The Importance of Technical Diagnostics … 185

and manner of realization of planned activities depend on the measurement

results and the degree of violation of the optimal technical characteristics of the
observed components of the technical system. On the other hand, there is a clear
difference between scheduled maintenance and condition-based maintenance.
In planned maintenance, all subsequent maintenance activities are carried out
according to a precisely defined time schedule, while in condition-based mainte-
nance, the current technical condition and the forecast of the technical system’s
behavior in the near future are taken as the most important factors. If the data
and values of measuring the parameters of the technical system are outside
normal operating tolerances or if the assessment of the development of moni-
tored quantities has a markedly negative trend with a tendency to move from
an inappropriate state to an increased or high risk of emergencies, a decision is
made. In this regard, subsequent maintenance interventions, conditioned by the
action of technical diagnostics and according to the results determined by deci-
sions, are treated as a form of preventive maintenance because their application
prevents far greater and more serious consequences that can result in significant
disruption of production processes (high maintenance costs, extremely expen-
sive spare parts, engagement of specialized firms, etc.). In this particular case as
well, the production stoppage was accepted and recorded as the use of planned
time for preventive repairs of the maintenance service,
6. Optimization of spare parts/units—the specificity of the working positions of
high-voltage recuperation motors (dry and wet side) is that in the recuperation
area there are two motors of the same operating characteristics with a certain
difference in the construction of the metal frame constructions with a chain
crane to assist in the manipulation / extraction of the high-voltage motor on the
“wet” side of the air recirculation. In this regard, the need to have at least one
spare engine with identical operating characteristics is defined, which provides
conditions for minimal downtime in case of unforeseen, accident situations or
replacement activities in case of need to overhaul these work units. As there is
a real possibility of a simultaneous problem in both work positions, there is a
justified need to secure these jobs with an online diagnostic system with only one
existing spare unit, which raises the reliability of this part of the technical system
paper machine to a higher level by ensuring process continuity production with
simultaneous optimization of the stock of spare aggregates and their spare parts.

6 Conclusion

Technical diagnostics within the monitoring of technical system elements proves to

be a particularly suitable tool for failure prevention in industrial systems operating in
a continuous mode where the imperative of high reliability and efficiency of produc-
tion capacity utilization does not allow frequent stops and control of technical system
parts preventive controls and repairs were implemented. By monitoring the status
of system elements in the work using computer technology and modern software
186 D. Lj. Branković et al.

solutions it is possible to obtain high-quality, timely and objective information on

the current state of the elements of technical systems in operation as well as very
precise prognosis assessment in future period based on which it can make timely
executive decisions and ensure the optimal level of utilization of production capaci-
ties. Ensuring a high degree of reliability is a prerequisite for optimal time utilization
of production capacities and maximum realization of production plans with minimal
costs, which means that industrial systems try to stand out from the competition and
survive in the increasingly demanding global world market.

References

1. Rejec E (1974) Terotechnology: modern organization of maintenance of funds, Informator,

Zagreb, p 134
2. Papić L, Milovanović Z (2007) Systems maintainability and reliability, DQM monograph
library quality and reliability in practice, Book 3, Prijevor, 501 p
3. Tsang A, Jardine A, Kolodny H (1999) Measuring maintenance performance: a holistic
approach. Int J Oper Prod Manage 19(7):691–715. https://doi.org/10.1108/014435799102
71674
4. EN 13306:2001 (2001) Terminologie maintenance
5. Bengtsson M (2004) Condition based maintenance systems—an investigation of technical
constituents and organizational aspects, Malardalen University Licentiate Thesis, No.36
6. Majdandžić N (1999) Maintenance strategies and maintenance information systems, JJ
Strossmayer University, Mechanical Engineering Faculty in Slavonski Brod, 349 p
7. Butcher SW (2000) Assessment of condition-based maintenance in the department of defense,
technical report
8. Kothamasu R, Huang SH, VerDuin WH (2006) System health monitoring and prognostics—a
review of current paradigms and practices. Int J Adv Manuf Technol 28:1012–1024. https://
doi.org/10.1007/s00170-004-2131-6
9. Shin J, Jun H (2015) On condition based maintenance policy. J Comput Des Eng 2(2):119–127
10. Erbe H, Morel G, Lee J, Iung B, L’eger JB, Seliger E et al (2005) Infotronic technologies for
e-maintenance regarding the costs aspects. Proc IFAC 38(1):1–12
11. Al-Najjar B, Alsyouf I (2003) Selecting the most efficient maintenance approach using fuzzy
multiole criteria decision making. Int J Prod Econ 84(1):85–100. https://doi.org/10.1016/
S0925-5273(02)00380-8
12. Tavner P, Ran L, Penman J, Sedding H (2008) Condition monitoring of rotating electrical
machines, IET power and energy series 56, The institution of engineering and technology.
London, United Kingdom, p 282
13. Sebastijanović S (2002) Machine constructions basic maintenance, JJ Strossmayer University,
Mechanical Engineering Faculty in Slavonski Brod, p 278
14. Jeremić B, Todorović P, Mačužić I, Koković V (2006) Technical diagnostic. University of
Kragujevac, Faculty of Mechanical Engineering in Kragujevac, Serbia, p 186
15. Randal R (2011) Vibration-based condition monitoring: industrial, aerospace and automotive
applications. A John Wiley and Sons, Ltd., Publication, ISBN 978-0-470-74785-8, p 289
16. Lee G (2006) Acceleration enveloping, a key element in aggressive condi-tion monitoring.
uptime, the magazine for PdM and BBM professionals, pp 38–42
17. Kazzaz S, Singh G (2003) Experimental investigations on induc-tion machine condition moni-
toring and fault diagnosis using digital signal processing techniques, Electric Power Syst Res
65(3):197–221. Elsevier
18. Bilošova A, Biloš J (2012) Vibration diagnostic, investments in education development, p 113.
Ostrava
The Importance of Technical Diagnostics … 187

19. Veltkamp A (2001) Vibration introduction course: SKF condition monitoring, in computer-
ized maintenance management system and enterprise resource planning, Nigerian Society of
Engineers, Lagos, Nigeria, pp 1.1–4.5
20. ISO 10816–3 (1995) Displacement, International standard
21. Kim YH., Tan ACC, Mathew J, Yang BS (2006) Condition monitoring of low speed bearings:
a comparative study of the ultrasound technique versus vibration measurements. In: Mathew J,
Kennedy J, Ma L, Tan A, Anderson D (eds) Engineering asset management. Springer, London.
https://doi.org/10.1007/978-1-84628-814-2_21
22. Moubray J (2000) Maintenance management: a new paradigm, strategic technologies, Inc.,
Aladon Ltd, UK, pp 7–11. http://www.maintenanceresources.com/RCM/Maintparadigm
Reliability Assessment of Replaceable
Shuffle Exchange Network
with an Additional Stage Using
Interval-Valued Universal Generating
Function

Amisha Khati and S. B. Singh

Abstract In this paper, a Shuffle Exchange Network with an additional stage (SEN+)
is inspected, the probabilities of whose components aren’t known with accuracy.
To overcome this difficulty we determined its reliability by using the method of
Interval-valued universal generating function (IUGF) and thus the transient state
probabilities are obtained in intervals. The reliability has been analyzed in terms
of three parameters: Terminal, Broadcast, and Network reliability. Also within the
observed network if any of the Switching Element (SE)comes up short and the system
quits working then we will replace that SE by a fixed replacement rate to continue the
operation of the network. A numerical example is also provided to offer a practical
clarification of the proposed technique.

Keywords Terminal reliability (TR) · Broadcast reliability (BR) · Network

reliability (NR) · Mean time to failure (MTTF) · Interval-valued Universal
generating function (IUGF)

1 Introduction

In parallel computer systems, interconnection networks offer communication among

memory modules, processors, and other devices by interconnecting them. Intercon-
nection systems are at present being utilized for various one-of-a-kind applications,
extending from very-large-scale integration (VLSI) circuits to wide-area computer
networks. For certain applications, interconnection systems have been considered in
profundity for quite a long time. In any case, over the most recent 10 years, we’ve
seen a fast development of the interconnection network innovation that is as of now
being injected into the modern era of multiprocessor frameworks. The innovation is
sufficiently experienced to discover its way into business items, while continually
introducing new roads for development and application. There are abundant features
pivotal in the progression of the interconnection networks.

A. Khati · S. B. Singh (B)

Department of Mathematics, Statistics and Computer Science, G. B. Pant University of
Agriculture and Technology, Pantnagar, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 189
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_9
190 A. Khati and S. B. Singh

To gauge the performance of a network, reliability is often taken as a vital

factor. The reliability and accordingly the overall performance of the interconnection
networks rely upon how its components are interconnected with one another.
Reliability of Multistage Interconnection Networks (MIN) can be thought-out as
a scale of the network’s ability to transform the message or data from input to output
depots. One of the most broadly used MIN is Shuffle Exchange Network (SEN) which
comprises a remarkable way between any source node and sink node and if any way
in the system becomes defective the whole system will get abolished. To conquer
this problem, engineers designed SEN+, which is a double-path MIN, by adding one
more stage to the existing SEN. The reliability of the SEN+ alongside the SEN has
been assessed by numerous specialists with the aid of various approaches. Fard and
Gunawan (2005) brought in a modified SEN comprising of 1 × 2 switching elements
at the source node, 2 × 2 at the intermediate stages, and 2 × 1 at the sink nodes, and
estimated the terminal reliability of SEN and modified SEN. The results obtained
say that the terminal reliability is higher for the network of size greater than 4 ×
4. Yunus and Othman (2011) scrutinized six different types of SENs having further
stages. The reviewed SENs are SEN+, Irregular Augmented SEN (IASEN), Irregular
Augmented Shuffle Network (IASN), Generalized SEN (GSEN), Improved Irreg-
ular Augmented Shuffle Multistage Interconnection Network (IIASN), and Irregular
Modified Alpha Network (ALN). It was detected that a more redundant path is
conquered by augmenting the stages and it also lessens the latency and upsurges the
fault-tolerance by permitting auxiliary links. However, increasing the network size
increases the price and complexity of the network. Bistouni and Jahanshahi [4] intro-
duced an approach, called rearranging links, to perk up the reliability of MINs. They
implemented this approach on SEN+ and ASEN and then evaluated their terminal
reliability. It was found that the procedure of rearranging the links is an efficient
one to spice up the reliability of MINs. Also, it had been observed that utilizing this
approach leads to a less steeply-priced MIN. Bistouni and Jahanshahi [2] presented
how to boost up the fault-tolerance and reliability of SEN by expanding the number of
stages and deduced that the reliability of SEN+ is better than both SEN or SEN having
two extra stages (SEN + 2). The reliability of SEN + has always been found more
than the reliability of SEN + 2. Bistouni and Jahanshahi [3] evaluated the reliability
of SENs by employing the technique of Reliability Block Diagrams (RBDs). Yunus
and Othman [14] proposed a new network SEN with minus one stage (SEN-) and
then compared it with SEN, SEN+, and SEN + 2. The comparison was made based
on three reliability indices, viz., terminal, broadcast, and network reliability. The
three aforesaid reliability parameters of SEN- were found higher than the rest three
considered networks. Bisht and Singh [1] figured out the terminal, broadcast, and
network reliability of SENs by applying the method of UGF. Bistouni and Jahanshahi
[5] analyzed the reliability importance of the switching elements in SEN, SEN+, and
SEN + 2. They concluded that a high-reliable network can be reached by replacing the
sensitive switching elements and by using switching elements having high-reliability.
From the above discussion it can be observed that none of the researchers examined
the networks incorporating the uncertainties in them while in the real-world situation,
it’s pretty common to go through the networks possessing many uncertainties. Thus
Reliability Assessment of Replaceable Shuffle Exchange Network … 191

it becomes very vital to analyze the reliability of SEN+ incorporating their uncer-
tainty which will prominently upgrade the believability of the reliability analysis of
the network. The motives causing the uncertainties in the complex systems are:
(a) Temperature, strain, moistness, vibration, stress, and so forth are the natural
highlights that would end in the uncertainties within the device and its
segments.
(b) When evaluating a posh system, in some cases it turns out to be very brain
desensitizing to obtain exact reliability data.
(c) If we routinely utilize a system, its overall performance depreciates with time,
and consequently the probability of its components shifts with time.
Interval-valued universal generating function (IUGF) is a way to investigate the
reliability of a network having uncertainties. Li et al. [8] advised a way to appraise
the reliability of the Multi-state system (MSS) when the prevailing data about the
component isn’t enough. In such cases rather than considering the precise values of
the probabilities, the interval-valued probabilities of the components can be consid-
ered. To acquire the interval-valued reliability of MSS, an IUGF was built up and
from the results it can be observed that the proposed method is efficient when the
state probabilities of the components are imprecise (or uncertain). Pan et al. [10] gave
an approach for the assessment of interval-valued reliability of MSS incorporating
uncertainty. They constructed the algorithm for the IUGF method and authorized
their method by encountering examples. Kumar et al. [7] considered a 2-out-of-4
system comprising of two components organized in a series pattern and evaluated
its interval-valued reliability using the IUGF technique. Singh [9] appraised the
reliability and MTTF of a non-repairable MSS by using the IUGF approach. They
analyzed the system’s reliability by considering the uncertainties in the probabilities
and the failure rates of the components of the considered system.
Also, all the researchers considered the SEN+ in which if any of the components
fizzles then one can’t replace it by a decent one. Nonetheless, in practice, we’d like
a network in which we can replace any broken segment with the goal that the entire
network doesn’t quit working.
Keeping all these facts in mind, we’ve considered a SEN+, the probability of
whose parts aren’t known with accuracy, and determined its reliability by using the
method of IUGF. Thus the transient state probabilities are evaluated in intervals.
The reliability is examined regarding three parameters: Terminal, Broadcast, and
Network reliability. Also, in the considered network if any of the SE fails and the
network stops operating, at that point we can replace that SE to proceed with the
operation of the network.
The SEN+ proposed is of size 8 × 8. The differential equations overseeing the
network’s conduct have been obtained by applying the method of supplementary
variable technique. The probabilities of all the components are obtained by using the
Laplace transform and henceforth the reliability of the network is computed. Also,
the MTTF of the SEN+ is analyzed. At last, we demonstrated our model by taking a
numerical example.
192 A. Khati and S. B. Singh

2 Assumptions

1. To begin with, the network is in worthy condition i.e., all the hubs and
connections are working appropriately.
2. The network considered is an 8 × 8 SEN+ wherein each of the SE is of size 2
× 2.
3. All the network’s components can be in a working stage or in a failed stage.
4. On the off chance that the network comes up short totally then only will go for
replacement and after replacement, the network will become as good as new.
5. The failure rates of different components are different while the replacement
rates of all the components are supposed to be the same.

3 Notations

x: Elapsed replacement time

λi : Failure rate of the ith component
η: Replacement rate for the entire network’s component
Pi : Lower bound of the probability of ith component, represented by a real number
Pi : Upper bound of the probability of ith component, represented by a real number
[R]: Interval-valued reliability of the network

4 Definitions

(i) Terminal Reliability: The terminal reliability (TR) of a network is defined

as the probability of the existence of at least one fault-free path between a
source–destination pair.
(ii) Broadcast Reliability: The broadcast reliability (BR) of a network is defined
as the probability of fruitful communication between a single source hub and
all the destination hubs.
(iii) Network Reliability: The network reliability (NR) of a network is defined as
the probability that all the source nodes are associated with all the destination
nodes.
Reliability Assessment of Replaceable Shuffle Exchange Network … 193

5 Terminal Reliability of SEN+

The RBD for the terminal reliability of SEN+ is shown in Fig. 1 and the transition
state diagram for the considered SEN+ is shown in Fig. 2.
(a) Formulation of mathematical model:
Applying the supplementary variable technique to the proposed model, the set of
difference-differential equations for lower bound probabilities of the components of
the model are obtained as:
  ∞ ∞
d
+ D P 0 (t) = P 1 (x, t)η(x)d x + P 8 (x, t)η(x)d x +
dt
0 0
∞ ∞
P 9 (x, t)η(x)d x+ P 10 (x, t)η(x)d x +
0 0

Fig. 1 RBD for the TR of

an 8 × 8 SEN+

Fig. 2 Transition state

diagram for Fig. 1
194 A. Khati and S. B. Singh

∞ ∞
P 11 (x, t)η(x)d x + P 12 (x, t)η(x, t)d x (1)
0 0
 
∂ ∂
+ + η(x) P1(x, t) = 0 (2)
∂x ∂t
 
d
+ λ4 P2(t) = λ2P0(t) (3)
dt
 
d
+ λ2 P3(t) = λ4P0(t) (4)
dt
 
d
+ λ5 P4(t) = λ3P0(t) (5)
dt
 
d
+ λ3 P5(t) = λ5P0(t) (6)
dt
 
d
+ λ2 + λ4 P6(t) = λ5P4(t) + λ3P5(t) (7)
dt
 
d
+ λ3 + λ4 P7(t) = λ2P3(t) + λ4P2(t) (8)
dt
 
∂ ∂
+ + η(x) P8(x, t) = 0 (9)
∂x ∂t
 
∂ ∂
+ + η(x) P9(x, t) = 0 (10)
∂x ∂t
 
∂ ∂
+ + η(x) P 10 (x, t) = 0 (11)
∂x ∂t
 
∂ ∂
+ + η(x) P11(x, t) = 0 (12)
∂x ∂t
 
∂ ∂
+ + η(x) P12(x, t) = 0 (13)
∂x ∂t

Boundary conditions:

P 1 (0, t) = λ1 P 0 (t) (14)

P 8 (0, t) = λ2 P 6 (t) (15)

Reliability Assessment of Replaceable Shuffle Exchange Network … 195

P 9 (0, t) = λ4 P 6 (t) (16)

P 10 (0, t) = λ3 P 7 (t) (17)

P 11 (0, t) = λ5 P 7 (t) (18)

P 12 (0, t) = λ6 P 0 (t) (19)

Initial conditions:
P 0 (t) = 1 att = 0 and is zero at all other values of t.
(b) Solution of the model:
On taking Laplace transform of Eqs. (1) to (13) along with the boundary conditions
(14) to (19) and using the initial conditions, we get:

∞ ∞
(s + D)P 0 (s) = 1 + P1(x, s)η(x)d x + P1(x, s)η(x)d x
0 0
∞ ∞
+ P8(x, s)η(x)d x + P9(x, s)η(x)d x (20)
0 0
∞ ∞
+ P10(x, s)η(x)d x+ P11(x, s)η(x)d x
0 0
 
s + λ4 P 2 (s) = λ2 P 0 (s) (21)

 
s + λ2 P 3 (s) = λ4 P 0 (s) (22)

 
s + λ5 P 4 (s) = λ3 P 0 (s) (23)

 
s + λ3 P 5 (s) = λ5 P 0 (s) (24)

 
s + λ2 + λ4 P 6 (s) = λ5 P 4 (s) + λ3 P 5 (s) (25)

 
s + λ3 + λ5 P 7 (s) = λ4 P 2 (s) + λ2 P 3 (s) (26)
 
∂
s+ + η(x) P 8 (s, x) = 0 (27)
dx
196 A. Khati and S. B. Singh

 
∂
s+ + η(x) P 9 (s, x) = 0 (28)
dx
 
∂
s+ + η(x) P 10 (s, x) = 0 (29)
dx
 
∂
s+ + η(x) P 11 (s, x) = 0 (30)
dx
 
∂
s+ + η(x) P 12 (s, x) = 0 (31)
dx

Boundary conditions:

P 1 (0, s) = P 0 (s)λ1 (32)

P 8 (0, s) = P 6 (s)λ2 (33)

P 9 (0, s) = P 6 (s)λ3 (34)

P 10 (0, s) = P 7 (s)λ3 (35)

P 11 (0, s) = P 7 (s)λ5 (36)

P 12 (0, s) = P 0 (s)λ6 (37)

(c) Transition state probabilities:

The equations from (20) to (37) lead to the following transition state probabilities of
the considered model:
1
P 0 (s) = (38)
(s + D)
λ1 (1 − S(s))
P 1 (s) = P 0 (s) (39)
s
λ2
P 2 (s) = P0(s) (40)
(s + λ4 )
λ4
P 3 (s) = P (s) (41)
(s + λ2 ) 0
Reliability Assessment of Replaceable Shuffle Exchange Network … 197

λ3
P 4 (s) = P0(s) (42)
(s + λ5 )

λ5
P 5 (s) = P 0 (s) (43)
(s + λ3 )
 
λ5 λ3 1 1
P 6 (s) = + P 0 (s) (44)
(s + λ2 + λ4 ) (s + λ5 ) (s + λ3 )
 
λ2 λ4 1 1
P 7 (s) = + P 0 (s) (45)
(s + λ3 + λ5 ) (s + λ2 ) (s + λ4 )
λ1 (1 − S(s))
P 8 (s) = P 6 (s) (46)
s
λ4 (1 − S(s))
P 9 (s) = P 4 (s) (47)
s
λ3 (1 − S(s))
P 10 (s) = P 3 (s) (48)
s
λ5 (1 − S(s))
P 11 (s) = P 5 (s) (49)
s
λ6 (1 − S(s))
P 12 (s) = P 0 (s) (50)
s

where D = λ1 + λ2 + λ3 + λ4 + λ5 + λ6 .
Similarly, we can find the expressions for upper bounds of the transition state
probabilities of the network’s components by replacing P i (s) by P i (s) (i = 0 to 12)
and λ j by λ j (j = 1 to 6) in Eqs. (38) to (50).
(i) Interval-valued Reliability of the network:
The interval-valued terminal reliability of the replaceable SEN+ is given as:

R = [ P 0 + P 5 + P 6 + P 8 + P 9 + P 11 + P 12 + P 13 + P 16 + P 18 , P 0 + P 5
+P 6 + P 8 + P 9 + P 11 + P 12 + P 13 + P 16 + P 18 ]
(51)
198 A. Khati and S. B. Singh

(ii) Mean time to failure:

The lower bound of the MTTF for the TR of SEN+ is obtained as:
 
1 λ λ λ λ λ3 λ5 1 1
(MTTF)TR = + 2 + 4 + 3 + 5 + +
D Dλ4 Dλ Dλ5 Dλ3 D(λ2 + λ4 ) λ5 λ3
 2 
λ2 λ4 1 1
+ + (52)
D(λ3 + λ5 ) λ2 λ4

The upper bound of the MTTF can be found by replacing λi by λi (i = 1 to 6) in

Eq. (52).

6 Broadcast Reliability of SEN+

The RBD for the BR of SEN+ is shown in Fig. 3 and the transition state diagram for
the broadcast reliability of SEN+ is shown in Fig. 4.
(a) Formulation of mathematical model:
By applying supplementary variable technique to the planned model, the difference-
differential equations for lower bound probabilities of the components of the network
in different states, governing the network’s behavior are as:

  ∞ ∞
d
+ E P 0 (t) = P 3 (x, t)η(x)d x + P 10 (x, t)η(x)d x
dt
0 0
∞ ∞
+ P 11 (x, t)η(x)d x + P 12 (x, t)η(x)d x
0 0
∞ ∞
+ P 13 (x, t)η(x)d x + P 14 (x, t)η(x)d x
0 0

Fig. 3 Broadcast RBD for

an 8 × 8 SEN+
Reliability Assessment of Replaceable Shuffle Exchange Network … 199

Fig. 4 Transition state diagram for Fig. 3

∞ ∞
+ P 17 (x, t)η(x)d x + P 20 (x, t)η(x)d x
0 0
∞ ∞
+ P 21 (x, t)η(x)d x + P 22 (x, t)η(x)d x (53)
0 0
 
d
+ λ3 + λ4 + λ5 P 1 (t) = λ2 P 0 (t) (54)
dt
 
d
+ λ2 + λ6 + λ7 P 2 (t) = λ3 P 0 (t) (55)
dt
 
d
+ λ5 P 4 (t) = λ4 P 1 (t) (56)
dt
 
d
+ λ4 P 5 (t) = λ5 P 1 (t) (57)
dt
 
d
+ λ3 + λ6 + λ7 P 6 (t) = λ4 P 5 (t) + λ5 P 4 (t) (58)
dt
200 A. Khati and S. B. Singh
 
d
+ λ7 P 7 (t) = λ6 P 2 (t) (59)
dt
 
d
+ λ6 P 8 (t) = λ7 P 2 (t) (60)
dt
 
d
+ λ2 + λ4 + λ5 P 9 (t) = λ7 P 7 (t) + λ6 P 8 (t) (61)
dt
 
∂ ∂
+ + η(x) P10(x, t) = 0 (62)
∂x ∂t
 
∂ ∂
+ + η(x) P11(x, t) = 0 (63)
∂x ∂t
 
∂ ∂
+ + η(x) P12(x, t) = 0 (64)
∂x ∂t
 
∂ ∂
+ + η(x) P13(x, t) = 0 (65)
∂x ∂t
 
∂ ∂
+ + η(x) P14(x, t) = 0 (66)
∂x ∂t
 
d
+ λ7 P 15 (t) = λ6 P 6 (t) (67)
dt
 
d
+ λ6 P 16 (t) = λ7 P 6 (t) (68)
dt
 
∂ ∂
+ + η(x) P17(x, t) = 0 (69)
∂x ∂t
 
d
+ λ5 P 18 (t) = λ4 P 9 (t) (70)
dt
 
d
+ λ4 P 19 (t) = λ5 P 9 (t) (71)
dt
 
∂ ∂
+ + η(x) P 20 (x, t) = 0 (72)
∂x ∂t
 
∂ ∂
+ + η(x) P 21 (x, t) = 0 (73)
∂x ∂t
 
∂ ∂
+ + η(x) P 22 (x, t) = 0 (74)
∂x ∂t
Reliability Assessment of Replaceable Shuffle Exchange Network … 201

Boundary conditions:

P 3 (0, t) = λ3 P 1 (t) + λ2 P 2 (t) (75)

P 12 (0, t) = λ9 P 0 (t) (76)

P 11 (0, t) = λ8 P 0 (t) (77)

P 13 (0, t) = λ10 P 0 (t) (78)

P 22 (0, t) = λ2 P 9 (t) (79)

P 17 (0, t) = λ7 P 15 (t) + λ6 P 16 (t) (80)

P 20 (0, t) = λ5 P 18 (t) + λ4 P 19 (t) (81)

P 21 (0, t) = λ3 P 6 (t) (82)

P 22 (0, t) = λ2 P 9 (t) (83)

Initial conditions:
P 0 (t) = 1 at t = 0 and is zero at all other values of t.
(b) Solution of the model:
By taking Laplace transform of the Eqs. (53) to (83) and using the initial conditions,
we get:

∞ ∞
 
s + F P 0 (s) = 1 + P 1 (s, x)η(x)d x + P 2 (s, x)η(x)d x
0 0
∞ ∞
+ P 3 (s, x)η(x)d x + P 4 (s, x)η(x)d x
0 0
∞ ∞
+ P 7 (s, x)η(x)d x + P 10 (s, x)η(x)d x
0 0
∞ ∞
+ P 14 (s, x)η(x)d x + P 15 (s, x)η(x)d x
0 0
202 A. Khati and S. B. Singh

∞ ∞
+ P 19 (s, x)η(x)d x + P 20 (s, x)η(x)d x (84)
0 0
 
s + λ3 + λ4 + λ5 P 1 (s) = λ2 P 0 (s) (85)

 
s + λ2 + λ6 + λ7 P 2 (s) = λ3 P 0 (s) (86)
 
∂
s+ + η(x) P 3 (s, x) = 0 (87)
dx
 
s + λ5 P 4 (s) = λ4 P 1 (s) (88)

 
s + λ4 P 5 (s) = λ5 P 1 (s) (89)

  λ λ P (s) λ4 λ5 P 1 (s)
s + λ3 + λ6 + λ7 P 6 (s) = 4 5 1 + (90)
(s + λ4 ) (s + λ5 )
 
s + λ7 P 7 (s) = λ6 P 2 (s) (91)

 
s + λ6 P 8 (s) = λ7 P 2 (s) (92)

  λ λ P (s) λ7 λ6P 1 (s)

s + λ2 + λ4 + λ5 P 9 (s) = 7 6 1 + (93)
(s + λ7 ) (s + λ6 )
 
∂
s+ + η(x) P 1 (s, x) = 0 (94)
dx
 
∂
s+ + η(x) P 2 (s, x) = 0 (95)
dx
 
∂
s+ + η(x) P 3 (s, x) = 0 (96)
dx
 
∂
s+ + η(x) P 4 (s, x) = 0 (97)
dx
 
∂
s+ + η(x) P 11 (s, x) = 0 (98)
dx
 
∂
s+ + η(x) P 12 (s, x) = 0 (99)
dx
Reliability Assessment of Replaceable Shuffle Exchange Network … 203
 
∂
s+ + η(x) P 13 (s, x) = 0 (100)
dx
 
∂
s+ + η(x) P 14 (s, x) = 0 (101)
dx
 
s + λ7 P 15 (s) = λ6 P 6 (s) (102)

 
s + λ6 P 16 (s) = λ7 P 6 (s) (103)
 
∂
s+ + η(x) P 17 (s, x) = 0 (104)
dx
 
s + λ5 P 18 (s) = λ4 P 9 (s) (105)

 
s + λ4 P 19 (s) = λ5 P 9 (s) (106)
 
∂
s+ + η(x) P 20 (s, x) = 0 (107)
dx
 
∂
s+ + η(x) P 21 (s, x) = 0 (108)
dx
 
∂
s+ + η(x) P 22 (s, x) = 0 (109)
dx

Boundary conditions:

P 3 (0, s) = P 1 (s)λ3 + P 2 (s)λ2 (110)

P 10 (0, s) = P 0 (s)λ1 (111)

P 11 (0, s) = P 0 (s)λ8 (112)

P 12 (0, s) = P 0 (s)λ9 (113)

P 13 (0, s) = P 0 (s)λ10 (114)

P 14 (0, s) = P 0 (s)λ11 (115)

P 17 (0, s) = P 15 (s)λ7 + P 16 (s)λ6 (116)

204 A. Khati and S. B. Singh

P 20 (0, s) = P 18 (s)λ5 + P 19 (s)λ4 (117)

P 22 (0, s) = P 9 (s)λ2 (118)

(c) Transition state probabilities:

On solving the equations from (84) to (109) and using the boundary conditions from
(110) to (118), the following transient state probabilities are obtained:

1
P 0 (s) = (119)
(s + E)
λ2 P 0 (s)
P 1 (s) = (120)
(s + λ3 + λ4 + λ5 )
λ3 P 0 (s)
P 2 (s) = (121)
(s + λ2 + λ6 + λ7 )
 
  1 − S(s)
P 3 (s) = λ3 P 1 (s) + λ2 P 2 (s) (122)
s
λ2 λ4 P 0 (s)
P 4 (s) = (123)
(s + λ5 )(s + λ3 + λ4 + λ5 )
λ2 λ5 P 0 (s)
P 5 (s) = (124)
(s + λ4 )(s + λ3 + λ4 + λ5 )
λ2 λ4 λ5 λ6 P 0 (s)
P 6 (s) =
(s + λ4 )(s + λ7 )(s + λ3 + λ4 + λ5 )
λ2 λ4 λ5 λ6 P 0 (s)
+ (125)
(s + λ5 )(s + λ7 )(s + λ3 + λ4 + λ5 )
λ3 λ6 P 0 (s)
P 7 (s) = (126)
(s + λ7 )(s + λ2 + λ6 + λ7 )
λ3 λ7 P 0 (s)
P 8 (s) = (127)
(s + λ6 )(s + λ2 + λ6 + λ7 )
λ3 λ6 λ7 P 0 (s)
P 9 (s) =
(s + λ6 )(s + λ2 + λ4 + λ5 )(s + λ2 + λ6 + λ7 )
λ3 λ6 λ7 P 0 (s)
+ (128)
(s + λ7 )(s + λ2 + λ4 + λ5 )(s + λ2 + λ6 + λ7 )
 
  1 − S(s)
P 10 (s) = λ1 P 0 (s) (129)
s
Reliability Assessment of Replaceable Shuffle Exchange Network … 205
 
  1 − S(s)
P 11 (s) = λ8 P 0 (s) (130)
s
 
  1 − S(s)
P 12 (s) = λ9 P 0 (s) (131)
s
 
  1 − S(s)
P 13 (s) = λ10 P 0 (s) (132)
s
 
  1 − S(s)
P 14 (s) = λ11 P 0 (s) (133)
s
λ2 λ4 λ5 λ6 P 0 (s)
P 15 (s) =
(s + λ4 )(s + λ7 )(s + λ3 + λ4 + λ5 )
λ2 λ4 λ5 λ6 P 0 (s)
+ (134)
(s + λ5 )(s + λ7 )(s + λ3 + λ4 + λ5 )
λ2 λ4 λ5 λ7 P 0 (s)
P 16 (s) =
(s + λ4 )(s + λ6 )(s + λ3 + λ4 + λ5 )
λ2 λ4 λ5 λ7 P 0 (s)
+ (135)
(s + λ5 )(s + λ6 )(s + λ3 + λ4 + λ5 )
 
  1 − S(s)
P 17 (s) = λ7 P 15 (s) + λ6 P 16 (s) (136)
s
λ3 λ4 λ6 λ7 P 0 (s)
P 18 (s) =
(s + λ5 )(s + λ7 )(s + λ2 + λ4 + λ5 )(s + λ2 + λ6 + λ7 )
λ3 λ4 λ6 λ7 P 0 (s)
+ (137)
(s + λ5 )(s + λ6 )(s + λ2 + λ4 + λ5 )(s + λ2 + λ6 + λ7 )
λ3 λ5 λ6 λ7 P 0 (s)
P 19 (s) =
(s + λ4 )(s + λ7 )(s + λ2 + λ4 + λ5 )(s + λ2 + λ6 + λ7 )
λ3 λ5 λ6 λ7 P 0 (s)
+ (138)
(s + λ4 )(s + λ6 )(s + λ2 + λ4 + λ5 )(s + λ2 + λ6 + λ7 )
 
  1 − S(s)
P 20 (s) = λ5 P 18 (s) + λ4 P 19 (s) (139)
s
 
  1 − S(s)
P 21 (s) = λ3 P 6 (s) (140)
s
 
  1 − S(s)
P 22 (s) = λ2 P 9 (s) (141)
s

where E = λ1 + λ2 + λ3 + λ8 + λ9 + λ10 + λ11 .

206 A. Khati and S. B. Singh

Similarly, we can find the expressions for upper bounds of the transition state
probabilities of the network’s components by replacing P i (s) by P i (s) (i = 0 to 22)
and λ j by λ j (j = 1 to 11) in Eqs. (119) to (141).
(i) Interval-valued reliability:
The interval-valued BR of the SEN+ is given as:

R = [P 0 + P 1 + P 2 + P 4 + P 5 + P 6 + P 7 + P 8 + P 9 + P 15 + P 16
+ P 18 + P 19 , P 0 + P 1 + P 2 + P 4 + P 5 + P 6 + P 7
+ P 8 + P 9 + P 15 + P 16 + P 18 + P 19 ] (142)

(ii) Mean time to failure:

The lower bound of the MTTF for the BR of SEN+ is given as:

(MTTF)BR =
1 λ2 λ3 λ2 λ4
+ + +
E E(λ3 + λ4 + λ5 ) E(λ2 + λ6 + λ7 ) Eλ5 (λ3 + λ4 + λ5 )
λ2 λ5 λ2 λ4 λ5 λ2 λ4 λ5
+ + +
Eλ4 (λ3 + λ4 + λ5 ) Eλ4 (λ3 + λ4 + λ5 ) Eλ5 (λ3 + λ4 + λ5 )
λ3 λ 6 λ3 λ7
+ +
Eλ7 (λ2 + λ6 + λ7 ) Eλ6 (λ2 + λ6 + λ7 )
λ3 λ6 λ7 λ3 λ6 λ7
+ +
Eλ7 (λ2 + λ6 + λ7 )(λ2 + λ4 + λ5 ) Eλ6 (λ2 + λ6 + λ7 )(λ2 + λ4 + λ5 )
λ2 λ4 λ5 λ6 λ2 λ4 λ5 λ6
+ +
Eλ4 λ7 (λ3 + λ4 + λ5 ) Eλ5 λ7 (λ3 + λ4 + λ5 )
λ2 λ4 λ5 λ7 λ2 λ4 λ5 λ7
+ +
Eλ4 λ6 (λ3 + λ4 + λ5 ) Eλ5 λ6 (λ3 + λ4 + λ5 )
λ3 λ4 λ7 λ6 λ3 λ4 λ7 λ6
+ +
Eλ5 λ7 (λ2 + λ6 + λ7 )(λ2 + λ4 + λ5 ) Eλ5 λ6 (λ2 + λ6 + λ7 )(λ2 + λ4 + λ5 )
λ3 λ5 λ7 λ6 λ3 λ5 λ7 λ6
+ +
Eλ4 λ7 (λ2 + λ6 + λ7 )(λ2 + λ4 + λ5 ) Eλ4 λ6 (λ2 + λ6 + λ7 )(λ2 + λ4 + λ5 )
(143)

We can determine the upper bound of the MTTF for BR of SEN+ by replacing λi
by λi (i = 1 to 11) in Eq. (143).
Reliability Assessment of Replaceable Shuffle Exchange Network … 207

7 Network Reliability of SEN+

The network RBD for an 8 × 8 SEN+ is shown in Fig. 5 and the transition state
diagram for the network reliability of SEN+ is shown in Fig. 6.
(a) Formulation of mathematical model:
By applying supplementary variable technique to the assumed model, the difference-
differential equations for lower bound probabilities of the components of the network
in different states, governing the network’s behavior are as:

  ∞ ∞
d
+ F P 0 (t) = P 1 (x, t)η(x)d x + P 2 (x, t)η(x)d x
dt
0 0

Fig. 5 Network RBD for an 8 × 8 SEN+

Fig. 6 Transition state diagram for Fig. 5

208 A. Khati and S. B. Singh

∞ ∞
+ P 3 (x, t)η(x)d x + P 4 (x, t)η(x)d x
0 0
∞ ∞
+ P 7 (x, t)η(x)d x + P 10 (x, t)η(x)d x
0 0
∞ ∞
+ P 14 (x, t)η(x)d x + P 15 (x, t)η(x)d x
0 0
∞ ∞
+ P 19 (x, t)η(x)d x + P 20 (x, t)η(x)d x (144)
0 0
 
d
+ λ6 P5 (t) = λ5 P 0 (t) (145)
dt
 
d
+ λ5 P6 (t) = λ6 P 0 (t) (146)
dt
 
∂ ∂
+ + η(x) P 7 (x, t) = 0 (147)
∂x ∂t
 
d
+ λ8 P8 (t) = λ7 P 0 (t) (148)
dt
 
d
+ λ7 P9 (t) = λ8 P 0 (t) (149)
dt
 
∂ ∂
+ + η(x) P 10 (x, t) = 0 (150)
∂x ∂t
 
d
+ λ10 P11 (t) = λ9 P 0 (t) (151)
dt
 
d
+ λ9 P12 (t) = λ10 P 0 (t) (152)
dt
 
d
+ λ11 + λ12 P13 (t) = λ9 P 12 (t) + λ10 P11 (t) (153)
dt
 
∂ ∂
+ + η(x) P 14 (x, t) = 0 (154)
∂x ∂t
 
∂ ∂
+ + η(x) P 15 (x, t) = 0 (155)
∂x ∂t
Reliability Assessment of Replaceable Shuffle Exchange Network … 209

 
d
+ λ12 P16 (t) = λ11 P0 (t) (156)
dt
 
d
+ λ11 P17 (t) = λ12 P0 (t) (157)
dt
 
d
+ λ9 + λ10 P18 (s) = λ12 P16 (s) + λ11 P17 (s) (158)
dt
 
∂ ∂
+ + η(x) P 19 (x, t) = 0 (159)
∂x ∂t
 
∂ ∂
+ + η(x) P 20 (x, t) = 0 (160)
∂x ∂t
 
∂ ∂
+ + η(x) P 21 (x, t) = 0 (161)
∂x ∂t
 
∂ ∂
+ + η(x) P 22 (x, t) = 0 (162)
∂x ∂t
 
∂ ∂
+ + η(x) P 23 (x, t) = 0 (163)
∂x ∂t
 
∂ ∂
+ + η(x) P 24 (x, t) = 0 (164)
∂x ∂t

Boundary conditions:

P 1 (0, t) = λ1 P 0 (t) (165)

P 2 (0, t) = λ2 P 0 (t) (166)

P 3 (0, t) = λ3 P 0 (t) (167)

P 4 (0, t) = λ4 P 0 (t) (168)

P 7 (0, t) = λ6 P 5 (t) + λ5 P 6 (t) (169)

P 10 (0, t) = λ8 P 8 (t) + λ7 P 9 (t) (170)

P 14 (0, t) = λ11 P 13 (t) (171)

210 A. Khati and S. B. Singh

P 15 (0, t) = λ12 P 13 (t) (172)

P 19 (0, t) = λ9 P 18 (t) (173)

P 20 (0, t) = λ10 P 18 (t) (174)

P 21 (0, t) = λ13 P 0 (t) (175)

P 22 (0, t) = λ14 P 0 (t) (176)

P 23 (0, t) = λ15 P 0 (t) (177)

P 24 (0, t) = λ16 P 0 (t) (178)

Initial conditions:
P 0 (t) = 1 at t = 0 and is zero at all other values of t.
(b) Solution of the model:
By taking Laplace transform of the Eqs. (144) to (178) and using the initial conditions,
we get:

∞ ∞
 
s + F P 0 (s) = 1 + P 1 (s, x)η(x)d x + P 2 (s, x)η(x)d x
0 0
∞ ∞
+ P 3 (s, x)η(x)d x + P 4 (s, x)η(x)d x
0 0
∞ ∞
+ P 7 (s, x)η(x)d x + P 10 (s, x)η(x)d x
0 0
∞ ∞
+ P 14 (s, x)η(x)d x + P 15 (s, x)η(x)d x
0 0
∞ ∞
+ P 19 (s, x)η(x)d x + P 20 (s, x)η(x)d x (179)
0 0
 
∂
s+ + η(x) P 1 (s, x) = 0 (180)
dx
Reliability Assessment of Replaceable Shuffle Exchange Network … 211
 
∂
s+ + η(x) P 2 (s, x) = 0 (181)
dx
 
∂
s+ + η(x) P 3 (s, x) = 0 (182)
dx
 
∂
s+ + η(x) P 4 (s, x) = 0 (183)
dx
 
s + λ6 P 5 (s) = λ5 P 0 (s) (184)

 
s + λ5 P 6 (s) = λ6 P 0 (s) (185)
 
∂
s+ + η(x) P 7 (s, x) = 0 (186)
dx
 
s + λ8 P 8 (s) = λ7 P 0 (s) (187)

 
s + λ7 P 9 (s) = λ8 P 0 (s) (188)
 
∂
s+ + η(x) P 10 (s, x) = 0 (189)
dx
 
s + λ10 P 11 (s) = λ9 P 0 (s) (190)

 
s + λ9 P 12 (s) = λ10 P 0 (s) (191)

 
s + λ11 + λ12 P 13 (s) = λ9 P 12 (t) + λ10 P 11 (t) (192)
 
∂
s+ + η(x) P 14 (s, x) = 0 (193)
dx
 
∂
s+ + η(x) P 15 (s, x) = 0 (194)
dx
 
s + λ12 P 16 (s) = λ11 P 0 (s) (195)

 
s + λ11 P 17 (s) = λ12 P 0 (s) (196)

 
s + λ9 + λ10 P 18 (s) = λ10 P 16 (t) + λ11 P 19 (t) (197)
212 A. Khati and S. B. Singh

 
∂
s+ + η(x) P 19 (s, x) = 0 (198)
dx
 
∂
s+ + η(x) P 20 (s, x) = 0 (199)
dx
 
∂
s+ + η(x) P 21 (s, x) = 0 (200)
dx
 
∂
s+ + η(x) P 22 (s, x) = 0 (201)
dx
 
∂
s+ + η(x) P 23 (s, x) = 0 (202)
dx
 
∂
s+ + η(x) P 24 (s, x) = 0 (203)
dx

Boundary conditions:

P 1 (0, s) = λ1 P 0 (s) (204)

P 2 (0, s) = λ2 P 0 (s) (205)

P 3 (0, s) = λ3 P 0 (s) (206)

P 4 (0, s) = λ4 P 0 (s) (207)

P 14 (0, s) = λ11 P 13 (s) (208)

P 15 (0, s) = λ12 P 13 (s) (209)

P 19 (0, s) = λ9 P 18 (s) (210)

P 20 (0, s) = λ10 P 18 (s) (211)

P 7 (0, s) = λ6 P 5 (s) + λ5 P 6 (s) (212)

P 10 (0, s) = λ8 P 8 (s) + λ7 P 9 (s) (213)

Initial conditions:
Reliability Assessment of Replaceable Shuffle Exchange Network … 213

P 0 (t) = 1 at t = 0 and is zero at all other values of t.

(c) Transition state probabilities:
On solving the equations from (179) to (203) and using the boundary conditions
from (204) to (213), the following transient state probabilities are obtained:

1
P 0 (s) = (214)
(s + F)
 
  1 − S(s)
P 1 (s) = λ1 P 0 (s) (215)
s
 
  1 − S(s)
P 2 (s) = λ2 P 0 (s) (216)
s
 
  1 − S(s)
P 3 (s) = λ3 P 0 (s) (217)
s
 
  1 − S(s)
P 4 (s) = λ4 P 4 (s) (218)
s
λ5 P 0 (s)
P 5 (s) = (219)
s + λ6
λ6 P 0 (s)
P 6 (s) = (220)
s + λ5
 
  1 − S(s)
P 7 (s) = λ6 P 5 (s) + λ5 P 6 (s) (221)
s
λ7 P 0 (s)
P 8 (s) = (222)
s + λ8
λ8 P 0 (s)
P 9 (s) = (223)
s + λ7
 
  1 − S(s)
P 10 (s) = λ8 P 8 (s) (224)
s
λ9 P 0 (s)
P 11 (s) = (225)
(s + λ10 )
λ10 P 0 (s)
P 12 (s) = (226)
s + λ9
λ9 λ10 P 0 (s) λ9 λ10 P 0 (s)
P 13 (s) = + (227)
(s + λ9 )(s + λ11 + λ12 ) (s + λ10 )(s + λ11 + λ12 )
214 A. Khati and S. B. Singh

 
  1 − S(s)
P 14 (s) = λ11 P 13 (s) (228)
s
 
  1 − S(s)
P 15 (s) = λ12 P 13 (s) (229)
s
λ11 P 0 (s)
P 16 (s) = (230)
s + λ12
λ12 P 0 (s)
P 17 (s) = (231)
s + λ11
λ11 λ12 P 0 (s) λ11 λ12 P 0 (s)
P 18 (s) = + (232)
(s + λ12 )(s + λ9 + λ10 ) (s + λ11 )(s + λ9 + λ10 )
 
  1 − S(s)
P 19 (s) = λ9 P 18 (s) (233)
s
 
  1 − S(s)
P 20 (s) = λ10 P 18 (s) (234)
s
 
  1 − S(s)
P 21 (s) = λ13 P 0 (s) (235)
s
 
  1 − S(s)
P 22 (s) = λ13 P 0 (s) (236)
s
 
  1 − S(s)
P 23 (s) = λ13 P 0 (s) (237)
s
 
  1 − S(s)
P 24 (s) = λ13 P 0 (s) (238)
s

where F = λ1 + λ2 + λ3 + λ4 + λ5 + λ6 + λ7 + λ8 + λ9 + λ10 + λ11 + λ12 +

λ13 + λ14 + λ15 + λ16 .
Similarly we can find the expressions for upper bounds of the transition state
probabilities of the network’s components by replacing P i (s) by P i (s) (i = 0 to 24)
and λ j by λ j (j = 1 to 13) in Eqs. (214) to (238).
(i) Interval-valued reliability:
The interval-valued NR of the considered network is given as:

R = [P 0 + P 5 + P 6 + P 8 + P 9 + P 11 + P 12 + P 13 + P 16 + P 18 ,
P 0 + P 5 + P 6 + P 8 + P 9 + P 11
+ P 12 + P 13 + P 16 + P 18 ] (239)
Reliability Assessment of Replaceable Shuffle Exchange Network … 215

(ii) Mean time to failure:

The lower bound of the MTTF for the NR of the considered SEN+ can be determined
by using Eq. (240) as:

1 λ λ λ λ λ λ
(MTTF)NR = + 5 + 6 + 7 + 8 + 9 + 10
F Fλ6 Fλ5 Fλ8 Fλ7 Fλ10 Fλ9
λ10 λ9 λ λ
+ + + 11 + 12
F(λ11 + λ12 ) F(λ11 + λ12 ) Fλ12 Fλ11
λ12 λ11
+ + (240)
F(λ9 + λ10 ) F(λ9 + λ10 )

One can determine the upper bound of the MTTF for NR in a similar manner by
replacing λi by λi (i = 1 to 16) in Eq. (240).

8 Numerical Illustration

8.1 Terminal Reliability of the Replaceable SEN +

under Consideration by Using the IUGF Approach

Let the upper and lower bounds of the failure rates of the SEN + be λ1 = 0.01, λ2 =
0.03, λ3 = 0.05, λ4 = 0.07, λ5 = 0.09, λ6 = 0.11, λ1 = 0.02, λ2 = 0.04, λ3 = 0.06,λ4
= 0.08,λ5 = 0.1,λ6 = 0.13.
On substituting the assumed values of failures rates in Eqs. (15) to (17) and
Eqs. (25) to (27) and on taking their inverse Laplace transform, we get the expressions
for various transition state probabilities. Tables 1 and 2 show the changes in the
transition state probabilities of the operating state with respect to time.
From Eq. (29), we can determine the variation on the reliability of the replaceable
SEN with time which is presented in the Table 3 and is shown in the Fig. 7.

8.2 MTTF of the SEN + under Consideration

By using Eq. (52), we can find the bounds of MTTF by changing the bounds of
failure rates which is presented in the Tables 4, 5, 6 and 7 and the changes in the
bounds of MTTF with different failure rates are displayed in the Figs. 8 and 9, where
B = λ1 ,λ2 ,λ3 ,λ4 ,λ5 ,λ6 and B = λ1 ,λ2 ,λ3 ,λ4 ,λ5 ,λ6.
216 A. Khati and S. B. Singh

Table 1 Variations in the transition state probabilities of operating states w. r. t. time

t P 0 (t), P 0 (t) P 2 (t), P 2 (t) P 3 (t), P 3 (t) P 4 (t), P 4 (t)
0 [1] [0,0] [0,0] [0,0]
1 [0.657047, [0.024281, 0.031302] [0.05786, 0.06394] [0.040047, 0.04641]
0.697676]
2 [0.431711, [0.03958, 0.049463] [0.096518, 0.10345] [0.06454, 0.072566]
0.486752]
3 [0.283554, [0.048723, 0.059173] [0.121829, 0.127003] [0.078479, 0.085718]
0.339596]
4 [0.186374, [0.053675, 0.063503] [0.137877, 0.141062] [0.085324, 0.09704]
0.236928]
5 [0.122456, [0.055799, 0.064455] [0.146584, 0.147584] [0.087468, 0.090764]
0.165299]
6 [0.08046, [0.05604, 0.063332] [0.148616, 0.152667] [0.08656, 0.087816]
0.115325]
7 [0.052866, [0.055052, 0.060982] [0.147985, 0.154875] [0.083128, 0.083797]
0.08046]
8 [0.034735, [0.053284, 0.057948] [0.14553, 0.154961] [0.077736, 0.079744]
0.056135]
9 [0.022823, [0.051044, 0.05458] [0.142074, 0.153621] [0.07195, 0.07512]
0.039164]
10 [0.014996, [0.048544, 0.051098] [0.137963, 0.151347] [0.066166, 0.070231]
0.027324]

Table 2 Variations in the transition state probabilities of operating states w. r. t. time

t P 5 (t), P 5 (t) P 6 (t), P 6 (t) P 7 (t), P 7 (t)
0 [0,0] [0,0] [0,0]
1 [0.073612, 0.079088] [0.003779, 0.004894] [0.003779, 0.047245]
2 [0.12138, 0.126447] [0.012749, 0.016056] [0.00845, 0.0102818]
3 [0.151291, 0.153227] [0.024289, 0.0]29788 [0.015577, 0.0172214]
4 [0.16673, 0.16891] [0.036702, 0.043891] [0.022809, 0.0263046]
5 [0.17113, 0.178767] [0.048924, 0.057113] [0.029502, 0.0386]
6 [0.171495, 0.181595] [0.060315, 0.068806] [0.035341, 0.0556191]
7 [0.167228, 0.181828] [0.07052, 0.078689] [0.040203, 0.0795118]
8 [0.162312, 0.178236] [0.079374, 0.086707] [0.044083, 0.035263]
9 [0.155548, 0.173745] [0.086831, 0.092934] [0.047039, 0.038385]
10 [0.148282, 0.16857] [0.092925, 0.097513] [0.049162, 0.040892]
Reliability Assessment of Replaceable Shuffle Exchange Network … 217

Table 3 Variations in the

t R(t), R(t)
terminal reliability bounds
with time 0 [1]
1 [0.86045, 0.912732116]
2 [0.7254864, 0.795321]
3 [0.6929866, 0.7198718]
4 [ 0.6774008, 0.695321]
5 [0.670961, 0.68424]
6 [0.6591051, 0.667854]
7 [0.6014996, 0.6331808]
8 [0.54823, 0.5994108]
9 [0.5304517, 0.5601235]
10 [0.48002479, 0.5189737,]

Fig. 7 TR versus time 1.2

0.8
Lower bound of
0.6 TR
Upper bound of
0.4
TR
0.2

0
0 5 10 15

Table 4 MTTF w. r. t. λ1 ,λ2 ,λ3 λ3

MTTF w. r. t.λ1 MTTF w. r. t.λ2 MTTF w. r. t.λ3
0.1 [13.03921569,18.25581396] [13.87868,18.25581396] [13.121886,18.25581396]
0.2 [10.901639344, 18.25581396] [10.702502,18.25581396] [12.338651,18.25581396]
0.3 [9.3661972, 18.25581396] [10.06095,18.25581396] [11.763298,18.25581396]
0.4 [8.20987654,18.25581396693] [9.7852875,18.25581396] [10.343975,18.25581396]
0.5 [7.3076923077, 18.25581396] [9.006275,18.25581396] [9.2865307,18.25581396]
0.6 [6.581584158, 18.25581396] [8.87429610,18.25581396] [8.6520746,18.25581396]
0.7 [5.990990991, 18.25581396] [7.7942064,18.25581396] [7.5090764,18.25581396]
0.8 [5.4958677686, 18.25581396] [6.99432086,18.25581396] [6.201986,18.25581396]
0.9 [5.0763358779, 18.25581396] [6.55872122,18.25581396] [5.6768021,18.25581396]
0.91 [5.0378787879, 18.25581396] [6.51860825,18.25581396] [5.641976,18.25581396]
218 A. Khati and S. B. Singh

Table 5 MTTF w. r. t. λ4 ,λ5 ,λ6 λ6

MTTF w. r. t.λ4 MTTF w. r. t.λ5 MTTF w. r. t.λ6
0.1 [12.742349,18.25581396] [15.9032,18.25581396] [16.625,18.25581396]
0.2 [11.138954,18.25581396] [14.5412,18.25581396] [13.3,18.25581396]
0.3 [10.347424,18.25581396] [13.985,18.25581396] [11.08333,18.25581396]
0.4 [9.1762086,18.25581396] [12.98642,18.25581396] [9.5,18.25581396]
0.5 [8.88542087,18.25581396] [11.54345,18.25581396] [8.3125,18.25581396]
0.6 [8.175201,18.25581396] [10.878202,18.25581396] [7.388889,18.25581396]
0.7 [7.5560428,18.25581396] [9.6541509,18.25581396] [6.65,18.25581396]
0.8 [6.1753207,18.25581396] [9.0095414,18.25581396] [6.04545455,18.25581396]
0.9 [5.34730725,18.25581396] [8.7341765,18.25581396] [5.5416667,18.25581396]
0.91 [5.3110843,18.25581396] [8.706428,18.25581396] [5.4958677,18.25581396]

Table 6 MTTF w. r. t. λ1 ,λ2 ,λ3

MTTF w. r. t.λ1 MTTF w. r. t.λ2 MTTF w. r. t.λ3
0.09 [6.6500002,18.70851] [6.6500002, 17.70692] [6.6500002, 20.29365]
0.091 [6.6500002,18.66609] [6.6500002, 17.68255] [6.6500002, 20.26063]
0.092 [6.6500002,18.62386] [6.6500002, 17.65889] [6.6500002, 20.22849]
0.093 [6.6500002, 18.58182] [6.6500002, 17.63593] [6.6500002, 20.19709]
0.094 [6.6500002, 18.53997] [6.6500002, 17.61974] [6.6500002, 20.16652]
0.095 [6.6500002, 18.49831] [6.6500002, 17.59199] [6.6500002, 20.13742]
0.096 [6.6500002, 18.45683] [6.6500002, 17.59024] [6.6500002, 20.10765]
0.097 [6.6500002, 18.41554] [6.6500002, 17.55059] [6.6500002, 20.0793]
0.098 [6.6500002, 18.37443] [6.6500002, 17.53271] [6.6500002, 20.03863]
0.099 [6.6500002, 18.33351] [6.6500002, 17.51156] [6.6500002, 20.02469]

Table 7 MTTF w. r. t. λ4 ,λ5 ,λ6

MTTF w. r. t.λ4 MTTF w. r. t.λ5 MTTF w. r. t.λ6
0.09 [6.6500002, 22.92815] [6.6500002, 25.35714] [6.6500002, 24.21102]
0.091 [6.6500002, 22.93929] [6.6500002, 25.37771] [6.6500002, 24.14002]
0.092 [6.6500002, 22.95099] [6.6500002, 25.3987] [6.6500002, 24.06943]
0.093 [6.6500002, 22.96324] [6.6500002, 25.42009] [6.6500002, 23.99926]
0.094 [6.6500002, 22.97531] [6.6500002, 25.444186] [6.6500002, 23.92949]
0.095 [6.6500002, 22.98927] [6.6500002, 25.467032] [6.6500002, 23.86013]
0.096 [6.6500002, 22.99413] [6.6500002, 25.48649] [6.6500002, 23.79117]
0.097 [6.6500002, 23.01724] [6.6500002, 25. 512802] [6.6500002, 23.72261]
0.098 [6.6500002, 23.03191] [6.6500002, 25.53246] [6.6500002, 23.65444]
0.099 [6.6500002, 23.04701] [6.6500002, 25.548714] [6.6500002, 23.58666]
Reliability Assessment of Replaceable Shuffle Exchange Network … 219

Fig. 8 MTTF versus λ1 ,λ2 ,λ3 ,λ4 ,λ5 ,λ6

Fig. 9 MTTF versus failure rate

220 A. Khati and S. B. Singh

8.3 Broadcast Reliability of the Considered Replaceable sEN

+ by Using the Method of IUGF

Let the upper and lower bounds of the failure rates of the proposed SEN + be λ1 =
0.02, λ2 = 0.04, λ3 = 0.06, λ4 = 0.08, λ5 = 0.1, λ6 = 0.12„ λ7 = 0.14,λ8 = 0.16,λ9
= 0.18,λ10 = 0.2,λ11 = 0.22, λ12 = 0.24, λ1 = 0.01, λ2 = 0.03, λ3 = 0.05,λ4 =
0.07,λ5 = 0.09,λ6 = 0.11,λ7 = 0.13,λ8 = 0.15,λ9 = 0.17,λ10 = 0.19,λ11 = 0.21,λ12
= 0.23.
Substituting the values of failures rates in Eq. (119) to (141) and taking their
inverse Laplace transform, we get the expressions for various transition state proba-
bilities. Tables 8, 9 and 10 shows the changes in the transition state probabilities of
the operating states of the SEN with variation in time.
Using the Eq. (142), we can evaluate the variation on the reliability of the consid-
ered replaceable SEN w. r. t. time which is given in the Table 11 and is depicted in
the Fig. 10.

Table 8 Variations in the transition state probabilities of operating states w. r. t. time

t P 0 (t), P 0 (t) P 1 (t), P 1 (t) P 2 (t), P 2 (t) P 4 (t), P 4 (t)
0 [1] [0,0] [0,0] [0,0]
1 [0.414783, [0.018286, 0.02324] [0.029493, 0.033728] [0.000734, 0.06951]
0.444858]
2 [0.1720449, [0.022957, 0.027921] [0.035634, 0.08977] [0.001566, 0.014795]
0.1978986]
3 [0.07136127, [0.022228, 0.025962] [0.033039, 0.034677] [0.003448, 0.02263]
0.0880368]
4 [0.029599, [0.019672, 0.022081] [0.028096, 0.029818] [0.004556, 0.030231]
0.039163]
5 [0.0122773, [0.016626, 0.018057] [0.021812, 0.022391] [0.005376, 0.037623]
0.0174224]
6 [0.00509243, [0.013795, 0.01449] [0.016573, 0.017606] [0.005928, 0.044966]
0.0077505]
7 [0.00211225, [0.011324, 0.011516] [0.012449, 0.013669] [0.006256, 0.052368]
0.003448]
8 [0.000876, [0.009242, 0.009108] [0.009294, 0.010536] [0.00602, 0.059987]
0.0015338]
9 [0.0003634, [0.007519, 0.007185] [0.006915, 0.008088] [0.00641, 0.06796]
0.00068233]
10 [0.000150733, [0.006108, 0.00566] [0.00515, 0.006195] [0.006312, 0.076418]
0.00030354]
Reliability Assessment of Replaceable Shuffle Exchange Network … 221

Table 9 Variations in the transition state probabilities of operating states w. r. t. time

t P 5 (t), P 5 (t) P 6 (t), P 6 (t) P 7 (t), P 7 (t) P 8 (t), P 8 (t)
0 [0,0] [0,0] [0,0] [0,0]
1 [0.000951, 0.00136] [0.000217, 0.000612] [0.001857, 0.00235] [0.00221, 0.002758]
2 [0.002741, 0.003802] [0.000604, 0.001073] [0.005115, 0.00626] [0.006137, 0.00742]
3 [0.004542, 0.006126] [0.000968, 0.001415] [0.00807, 0.009608] [0.00977, 0.011487]
4 [0.006059, 0.007967] [0.001253, 0.001664] [0.010228, 0.01186] [0.012502, 0.01432]
5 [0.007223, 0.009278] [0.001452, 0.001838] [0.011561, 0.01309] [0.01428, 0.001598]
6 [0.008053, 0.010122] [0.001575, 0.001954] [0.012202, 0.01351] [0.015241, 0.01668]
7 [0.008597, 0.010587] [0.001638, 0.002025] [0.012316, 0.01336] [0.015565, 0.01669]
8 [0.008905, 0.010758] [0.001654, 0.002061] [0.012053, 0.01282] [0.015423, 0.01623]
9 [0.009029, 0.00709] [0.001636, 0.002071] [0.011536, 0.01204] [0.014953, 0.01545]
10 [0.009008, 0.010499] [0.001594, 0.002062] [0.01086, 0.011136] [0.014268, 0.01449]

8.4 MTTF of the SEN +

Using Eq. (143), we can determine the upper and lower bounds of MTTF of the SEN
+ corresponding to its BR. By varying the failure rate, the various bound obtained
are presented in Tables 12, 13, 14, 15, 16, 17, 18 and 19 and their graphical version
are given in Figs. 11 and 12, where C = λ1 ,λ2 ,λ3 ,ss λ4 ,λ5 ,λ6 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11 and C
= λ1 ,λ2 , λ3 ,λ4 , λ5 , λ6 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11 .

8.5 Network Reliability of the Replaceable SEN + using

IUGF Approach

Let the upper and lower bounds of the failure rates of the proposed SEN + be λ1 =
0.01, λ2 = 0.03, λ3 = 0.05, λ4 = 0.07, λ5 = 0.09, λ6 = 0.11„ λ7 = 0.13,λ8 = 0.15,λ9
= 0.17,λ10 = 0.19,λ11 = 0.21,λ12 = 0.23, λ13 = 0.25, λ14 = 0.27, λ15 = 0.28, λ16 =
0.29, λ 1 = 0.02, λ 2 = 0.04, λ 3 = 0.06,λ 4 = 0.08,λ 5 = 0.1,λ 6 = 0.12,λ 7 = 0.14,λ 8
= 0.16,λ 9 = 0.18,λ 10 = 0.2,λ 11 = 0.22,λ 12 = 0.24, λ 13 = 0.26,λ 14 = 0.28,λ 15 =
0.285,λ 16 = 0.295.
Putting the values of failures rates in Eqs. (214) to (238) and taking their inverse
Laplace transform, we get the expressions for various transition state probabilities.
Tables 20, 21 and 22 show the changes in the transition state probabilities of the
operating states with respect to time.
Using the Eq. (239), we can find the variation on the network reliability of the
replaceable SEN+ with time which is given in the Table 23 and is shown in the
Fig. 13.
 222

Table 10 Variations in the transition state probabilities of operating states w. r. t. time

t P 9 (t), P 9 (t) P 15 (t), P 15 (t) P 16 (t), P 16 (t) P 18 (t), P 18 (t)
0 [0,0] [0,0] [0,0] [0,0]
1 [0.000172, 0.000235] [0.0000051, 0.0000092] [0.00000536, 0.000012] [0.0000055, 0.0000106]
2 [0.001009, 0.001332] [0.0000309, 0.0000541] [0.0000363, 0.0000654] [0.0000594, 0.0007022]
3 [0.002526, 0.003237] [0.0000797, 0.0001361] [0.0000952, 0.0001629] [0.0002202, 0.0002248]
4 [0.004491, 0.005593] [0.0001461, 0.0002439] [0.000152, 0.00029242] [0.00016302, 0.000479]
5 [0.006644, 0.00805] [0.00022296, 0.000362] [0.0002706, 0.0003762] [0.000849, 0.00100913]
6 [0.00877, 0.010351] [0.0003039, 0.0004871] [0.00037144, 0.000589] [0.0013171, 0.001853]
7 [0.010722, 0.012338] [0.0003839, 0.0005937] [0.0004724, 0.0006210] [0.001942, 0.00233001]
8 [0.012408, 0.013929] [0.0004593, 0.0007098] [0.00056908, 0.000989] [0.0024625, 0.003154]
9 [0.013779, 0.015102] [0.0005277, 0.0008124] [0.0006624, 0.0009891] [0.003127, 0.003924]
10 [0.014824, 0.015871] [0.00058772, 0.000877] [0.00073805, 0.001090] [0.0037298, 0.0047326]
A. Khati and S. B. Singh
Reliability Assessment of Replaceable Shuffle Exchange Network … 223

Table 11 Changes in the

T R(t), R(t)
broadcast reliability bounds
with time 0 [1]
1 [0.468719, 0.57868]
2 [0.41220131, 0.54691109]
3 [0.3860208, 0.3992534]
4 [0.36142308, 0.3714442]
5 [0.3129225950.312922595]
6 [0.3119157, 0.31232434]
7 [0.271104265, 0.301650249]
8 [0.247908, 0.26147536]
9 [0.1913648, 0.21088568]
10 [0.1378953, 0.1680076]

Fig. 10 BR versus time 1.2

0.8
Reliability

0.6 Lower bound of

BR
0.4 Upper bound of
BR
0.2

0
0 5 10 15
Time

Table 12 MTTF w. r. t. λ1 ,λ2 ,λ3

MTTF w. r. t.λ1 MTTF w. r. t.λ2 MTTF w. r. t.λ3
0.1 [3.8774968, 27.954032] [4.604626, 27.954032] [5.687244, 27.954032]
0.2 [3.6580158, 27.954032] [6.683774, 27.954032] [6.652804,27.954032]
0.3 [3.1534619, 27.954032] [8.438454, 27.954032] [7.653553, 27.954032]
0.4 [2.5027476, 27.954032] [9.930167, 27.954032] [8.582255,27.954032]
0.5 [1.8402555, 27.954032] [10.46414, 27.954032] [9.417767,27.954032]
0.6 [1.2604490, 27.954032] [12.31958, 27.954032] [10.16279,27.954032]
0.7 [0.8079801, 27.954032] [13.2896, 27.954032] [10.82652,27.954032]
0.8 [0.4867350, 27.954032] [14.14345, 27.954032] [11.54981,27.954032]
0.9 [0.2765539, 27.954032] [14.90108, 27.954032] [11.95038,27.954032]
0.91 [0.1562452, 27.954032] [15.04965, 27.954032] [12.06321, 27.954032]
224 A. Khati and S. B. Singh

Table 13 MTTF w. r. t. λ4 ,λ5 ,λ6 λ6

MTTF w. r. t.λ4 MTTF w. r. t.λ5 MTTF w. r. t.λ6
0.1 [20.39979, 27.954032] [15.9842, 27.954032] [12.984207,27.954032]
0.2 [18.9774, 27.954032] [13.97621, 27.954032] [11.41526, 27.954032]
0.3 [15.34459, 27.954032] [12.98246, 27.954032] [10.85541, 27.954032]
0.4 [13.00737, 27.954032] [11.41096, 27.954032] [9.983425607,27.954032]
0.5 [11.37702, 27.954032] [10.39325, 27.954032] [8.298473,27.954032]
0.6 [10.17459, 27.954032] [9.52351, 27.954032] [8.119777,27.954032]
0.7 [9.753145, 27.954032] [8.20653, 27.954032] [8.0094112, 27.954032]
0.8 [8.518797, 27.954032] [8.001438, 27.954032] [7.8984723,27.954032]
0.9 [7.924162, 27.954032] [7.923206, 27.954032] [7.798434,27.954032]
0.91 [7.870775, 27.954032] [7.902426, 27.954032] [7.698818,27.954032]

Table 14 MTTF w. r. t. λ7 ,λ8 ,λ9 λ9

MTTF w. r. t.λ7 MTTF w. r. t.λ8 MTTF w. r. t.λ9
0.1 [7.853127., 27.954032] [4.539508, 27.954032] [4.652996, 27.954032]
0.2 [7.513118, 27.954032] [4.0460836,27.954032] [4.135997, 27.954032]
0.3 [7.3555276 27.954032] [3.6494087, 27.954032] [3.722396, 27.954032]
0.4 [7.2048652, 27.954032] [3.323569, 27.954032] [3.49723, 27.954032]
0.5 [7.118563, 27.954032] [3.051145, 27.954032] [3.101997, 27.954032]
0.6 [7.0101625, 27.954032] [2.819998, 27.954032] [2.872102, 27.954032]
0.7 [6.90373453, 27.954032] [2.621406„ 27.954032] [2.658855, 27.954032]
0.8 [6.881042,27.954032] [2.448945, 27.954032] [2.484309, 27.954032]
0.9 [6.7151872,27.954032] [2.297776, 27.954032] [2.326498, 27.954032]
0.91 [6.7019423,27.954032] [2.283679, 27.954032] [2.312048, 27.954032]

Table 15 MTTF w. r. t.
MTTFw. r. t.λ10 MTTF w. r. t.λ11
λ10 ,λ11
0.1 [4.772304, 27.954032] [4.897891, 27.954032]
0.2 [4.229997, 27.954032] [4.328369, 27.954032]
0.3 [3.798364, 27.954032] [3.877497, 27.954032]
0.4 [3.560932, 27.954032] [3.511695, 27.954032]
0.5 [3.154574, 27.954032] [3.208963, 27.954032]
0.6 [2.896932, 27.954032] [2.934981, 27.954032]
0.7 [2.6557518, 27.954032] [2.737057, 27.954032]
0.8 [2.402058, 27.954032] [2.549587, 27.954032]
0.9 [2.355947, 27.954032] [2.386152, 27.954032]
0.91 [2.34113, 27.954032] [2.370953, 27.954032]
Reliability Assessment of Replaceable Shuffle Exchange Network … 225

Table 16 MTTF w. r.
MTTF w. r. t.λ1 MTTF w. r. t.λ2 MTTF w. r. t.λ3
t.λ1 ,λ2 ,λ3
0.1 [1.26434352, [1.26434352, [1.26434352,
3.09359758] 1.654956] 1.536492]
0.2 [1.26434352, [1.26434352, [1.26434352,
2.78423782] 1.761904] 2.390001]
0.3 [1.26434352, [1.26434352, [1.26434352,
2,531125] 2.009773] 3.199891]
0.4 [1.26434352, [1.26434352, [1.26434352,
2.320198] 2.298437] 3.961509]
0.5 [1.26434352, [1.26434352, [1.26434352,
2.1513129] 2.595035] 4.676214]
0.6 [1.26434352, [1.26434352, [1.26434352,
1.988741] 2.886925] 5.346969]
0.7 [1.26434352, [1.26434352, [1.26434352,
1.8307822] 3.168934] 5.977091]
0.8 [1.26434352, [1.26434352, [1.26434352,
1.740149] 3.43903] 6.569825]
0.9 [1.26434352, [1.26434352, [1.26434352,
1.637787] 3.696612] 7.128206]
0.91 [1.26434352, [1.26434352, [1.26434352,
1.628209] 3.721682] 7.182264]

Table 17 MTTF w. r. t. λ4 ,
MTTF w. r. t.λ4 MTTF w. r. t.λ5 MTTF w. r. t.λ6
λ5 , λ6 λ6
0.1 [1.26434352, [1.26434352, [1.26434352,
2.8468719] 2.78429] 3.85526]
0.2 [1.26434352, [1.26434352, [1.26434352,
2.4268227] 2.68398] 3.81627]
0.3 [1.26434352, [1.26434352, [1.26434352,
2.2214942] 2.2845] 3.757654]
0.4 [1.26434352, [1.26434352, [1.26434352,
2.10470249] 2.14352] 3.6225]
0.5 [1.26434352, [1.26434352, [1.26434352,
2.03287682] 2.04436] 3.516455]
0.6 [1.26434352, [1.26434352, [1.26434352,
1.9860925] 1.983529] 3.475505]
0.7 [1.26434352, [1.26434352, [1.26434352,
1.82286785] 1.897982] 3.388433]
0.8 [1.26434352, [1.26434352, [1.26434352,
1.488691] 1.769752] 3.287852]
0.9 [1.26434352, [1.26434352, [1.26434352,
1.27589] 1.668726] 3.161679]
0.91 [1.26434352, [1.26434352, [1.26434352,
1.255834] 1.6483829] 3.14492]
226 A. Khati and S. B. Singh

Table 18 MTTF w. r. t.
MTTF w. r. t.λ7 MTTF w. r. t.λ8 MTTF w. r. t.λ9
λ7 ,λ8 ,λ9 λ9
0.1 [1.26434352, [1.26434352, [1.26434352,
3.92946] 3.66342118] 3.01864]
0.2 [1.26434352, [1.26434352, [1.26434352,
3.87847] 3.42744197] 2.887845]
0.3 [1.26434352, [1.26434352, [1.26434352,
3.7123] 3.30020844] 2.78429]
0.4 [1.26434352, [1.26434352, [1.26434352,
3.68763] 3.19098693] 2.67742]
0.5 [1.26434352, [1.26434352, [1.26434352,
3.55954] 3.0881076] 2.55070]
0.6 [1.26434352, [1.26434352, [1.26434352,
3.42497] 2.9854911] 2.44289]
0.7 [1.26434352, [1.26434352, [1.26434352,
3.383862] 2.84319] 2.376763]
0.8 [1.26434352, [1.26434352, [1.26434352,
3.292675] 2.74945] 2.209894]
0.9 [1.26434352, [1.26434352, [1.26434352,
3.15458] 2.686855] 2.16682]
0.91 [1.26434352, [1.26434352, [1.26434352,
3.126243] 2.643192] 2.13875]

Table 19 MTTF w. r. t.
MTTF w. r. t.λ10 MTTF w. r. t.λ11
λ10 ,λ11
0.1 [1.26434352, 2.8634218] [1.26434352, 2.209894]
0.2 [1.26434352, 2.74945] [1.26434352, 2.198549]
0.3 [1.26434352, 2.642497] [1.26434352, 2.142497]
0.4 [1.26434352, 2.559854] [1.26434352, 2.024987]
0.5 [1.26434352, 2.320989] [1.26434352, 1.985416]
0.6 [1.26434352, 2.242455] [1.26434352, 1.820984]
0.7 [1.26434352, 2.17494] [1.26434352, 1.74945]
0.8 [1.26434352, 2.0985413] [1.26434352, 1.64247]
0.9 [1.26434352, 2.042497] [1.26434352, 1.498519]
0.91 [1.26434352, 2.009894] [1.26434352, 1.474945]

8.6 MTTF of the SEN+

The bounds of the MTTF at various failure rates are given in the Tables 24, 25, 26,
27, 28, 29, 30, 31, 32, 33, 34 and 35 and are represented graphically in Figs. 14 and
15, where in Fig. 14 λ A = λ1 ,λ2 ,λ3 , λ4 ,λ5 ,λ6 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11 ,λ12 , λ13 , λ14 ,λ15 , λ16
and in Fig. 15 λ A = λ1 , λ2 , λ3 , λ4 , λ5 , λ6 ,λ7 , λ8 , λ9 , λ10 , λ11 , λ12 , λ13 , λ14 , λ15 , λ16 .
Reliability Assessment of Replaceable Shuffle Exchange Network … 227

Fig. 11 MTTF versusλ1 ,λ2 ,λ3 ,λ4 ,λ5 ,λ6 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11

Fig. 12 MTTF versus Failure rates

9 Result and Discussion

The proposed model studies how one can analyze the bounds of reliability of a
replaceable shuffle exchange network incorporating uncertainties by using the IUGF
technique. The reliability bounds of the proposed SEN+ have been determined based
on three reliability parameters: TR, BR, and NR. From the outcomes accomplished
it can be concluded that the reliability bounds of all the three reliability indices of the
SEN+ decrease with increasing time, which can be observed from Figs. 7, 10 and 13.
 228

Table 20 Variations in the transition state probabilities of operating states w. r. t. time

t P 0 (t), P 0 (t) P 5 (t), P 5 (t) P 6 (t), P 6 (t)
0 [1] [0,0] [0,0]
1 [ 0.068563154, 0.079659] [0.0303536, 0.0319571] [0.037611, 0.0388965]
2 [0.004700906, 0.006346] [0.0296097, 0.031967] [0.03737, 0.037861853]
3 [0.00032231, 0.000505] [0.026718034, 0.0272403] [0.034392, 0.03444167]
4 [0.0000222098, 0.000040266] [0.02395027, 0.0241703] [0.031451, 0.031176649]
5 [0.000001515, 0.000003207] [0.02143789, 0.021456692] [0.028745, 0.0282106]
6 [0.000000104, 0.00000025555] [0.01190137, 0.0192217] [0.026271, 0.025526118]
7 [0.000000007122, 0.00000002035] [0.0168637, 0.0172195] [0.023296, 0.02401]
8 [0.000000000488, 0.000000001621] [0.014957, 0.015426] [0.020899, 0.021944]
9 [0.000000000033, 0.0000000001291] [0.01326545, 0.013819] [0.01891022, 0.020055]
10 [0.00000000000229, 0.0000000000103] [0.0117654, 0.0123795] [0.017110672, 0.018329]
A. Khati and S. B. Singh
Reliability Assessment of Replaceable Shuffle Exchange Network … 229

Table 21 Variations in the transition state probabilities of operating states w. r. t. time

t P 8 (t), P 8 (t) P 9 (t), P 9 (t) P 11 (t), P 11 (t) P 12 (t), P 12 (t)
0 [0,0] [0,0] [0,0] [0,0]
1 [0.042997, [0.049902, 0.050444] [0.054291, 0.054448] [0.061337, 0.061509]
0.04353225]
2 [0.040080452, [0.047312, 0.047794] [0.048311, 0.049221] [0.055438, 0.056793]
0.04015]
3 [0.034359, [0.041368, 0.042234] [0.03981, 0.041048] [0.046594, 0.048304]
0.034803]
4 [0.029292796, [0.03598, 0.037155] [0.032611, 0.033973] [0.038938, 0.040784]
0.02997]
5 [0.0249626, [0.031281, 0.032628] [0.026701, 0.028096] [0.032525, 0.03441]
0.025809]
6 [0.02127821, [0.027194, 0.02865] [0.021861, 0.023235] [0.027168, 0.029031]
0.22208]
7 [0.018226655, [0.023642, 0.025158] [0.017898, 0.019214] [0.022692, 0.024492]
0.01911]
8 [0.015446517, [0.020553, 0.022091] [0.014654, 0.015889] [0.018954, 0.020663]
0.01645]
9 [0.013162653, [0.017868, 0.019398] [0.011997, 0.01314] [0.015832, 0.017433]
0.01416]
10 [0.011216473, [0.015534, 0.017033] [0.009823, 0.010866] [0.013224, 0.014708]
0.01218]

Thus, the performance of the network is crumbling with time. From the outcomes, it
was also observed that among all the three reliability bounds, the bounds of TR are
the best followed by the BR bounds which are trailed by the NR bounds.
In addition to the bounds of reliability of the network, the bounds of the MTTF
of the network at different failure rates were also evaluated and the results obtained
can be summed up as:
230 A. Khati and S. B. Singh

Table 22 Variations in the transition state probabilities of operating states w. r. t. time

T P 13 (t), P 13 (t) P 16 (t), P 16 (t) P 17 (t), P 17 (t) P 18 (t), P 18 (t)
0 [0,0] [0,0] [0,0] [0,0]
1 [0.012796, 0.013627] [0.064744, [0.071605, 0.072462] [0.019407, 0.020271]
0.065271]
2 [0.024539, 0.025431] [0.055368, [0.062374, 0.064509] [0.037725, 0.038341]
0.05706]
3 [0.029866, 0.030354] [0.043858, [0.050393, 0.05275] [0.046426, 0.46418]
0.04575]
4 [0.031016, 0.03102] [0.034521, [0.040464, 0.042795] [0.047581, 0.048573]
0.036383]
5 [0.029389, 0.029827] [0.027157, [0.032475, 0.034692] [0.045223, 0.046888]
0.02891]
6 [0.02666, 0.027441] [0.021362, [0.026062, 0.028121] [0.041009, 0.43151]
0.02297]
7 [0.02355, 0.024552] [0.016804, [0.020195, 0.022794] [0.03606, 0.038486]
0.018251]
8 [0.020423, 0.021562] [0.013219, [0.016785, 0.018477] [0.031033, 0.033582]
0.014501]
9 [0.017491, 0.018693] [0.010398, [0.01347, 0.014977] [0.026289, 0.028835]
0.011521]
10 [0.014848, 0.016057] [0.008179, [0.01081, 0.01214] [0.022008, 0.024459]
0.009154]

Table 23 Changes in the

t R(t), R(t)
network reliability of the
SEN+ with time 0 [1]
1 [0.462659, 0.575439]
2 [0.442011, 0.4598109]
3 [0.41389608, 0.423992534]
4 [0.38614208, 0.39962971442]
5 [0.37131528, 0.3802322595]
6 [0.329157, 0.3427192434]
7 [0.3027110465, 0.31260249]
8 [0.27097618, 0.299147536]
9 [0.2598855568, 0.265413648]
10 [0.249997855, 0.25802036]
Reliability Assessment of Replaceable Shuffle Exchange Network … 231

Fig. 13 Network Reliability 1.2

versus time 1

Reliability
0.8

0.6
Lower bound of NR
0.4
Upper bound of NR
0.2

0
0 5 10 15
Time

Table 24 MTTF w. r. t.λ1 ,λ2 ,λ3

MTTF w. r. t.λ1 MTTF w. r. t.λ2 MTTF w. r. t.λ3
0.1 [3.724784,4.39604205] [3.751972,4.39604205] [3.77956,4.39604205]
0.2 [3.594547,4.39604205] [3.61986,4.39604205] [3.645533,4.39604205]
0.3 [3.473109,4.39604205] [3.496736,4.39604205] [3.840953,4.39604205]
0.4 [3.359609,4.39604205] [3.381712,4.39604205] [3.821614,4.39604205]
0.5 [3.253292,4.39604205] [3.274014,4.39604205] [3.295001,4.39604205]
0.6 [3.153498,4.39604205] [3.172964,4.39604205] [3.158427,4.39604205]
0.7 [3.059644,4.39604205] [3.083559,4.39604205] [3.096507,4.39604205]
0.8 [2.971245,4.39604205] [2.998489,4.39604205] [3.005966,4.39604205]
0.9 [2.887754,4.39604205] [2.904069,4.39604205] [2.920569,4.39604205]
0.91 [2.886943, 0.39604205] [2.903249,4.39604205] [2.91974,4.39604205]

Table 25 MTTF w. r. t. λ4 ,λ5 ,λ6

MTTF w. r. t.λ4 MTTF w. r. t.λ5 MTTF w. r. t.λ6
0.1 [3.807557,4.39604205] [3.984653,4.39604205] [3.852282,4.39604205]
0.2 [3.621883,4.39604205] [3.93228,4.39604205] [3.893866,4.39604205]
0.3 [3.544967,4.39604205] [4.009738,4.39604205] [4.049092,4.39604205]
0.4 [3.426801,4.39604205] [4.12914,4.39604205] [4.121983,4.39604205]
0.5 [3.372291,4.39604205] [4.24622,4.39604205] [4.194467,4.39604205]
0.6 [3.212626,4.39604205] [4.265932,4.39604205] [4.233761,4.39604205]
0.7 [3.497356,4.39604205] [4.281708,4.39604205] [4.253427,4.39604205]
0.8 [3.023648,4.39604205] [4.290619,4.39604205] [4.296045,4.39604205]
0.9 [2.937258,4.39604205] [4.298344,4.39604205] [4.318242,4.39604205]
0.91 [2.92889,4.39604205] [4.321165,4.39604205] [4.356728,4.39604025]
232 A. Khati and S. B. Singh

Table 26 MTTF w. r. t. λ7 ,λ8 ,λ9

MTTF w. r. t.λ7 MTTF w. r. t.λ8 MTTF w. r. t.λ9
0.1 [3.972555,4.39604205] [4.275923,4.39604205] [4.253029,4.39604205]
0.2 [3.763703,4.39604205] [4.123972,4.39604205] [4.246949,4.39604205]
0.3 [3.757352,4.39604205] [4.148283,4.39604205] [4.235865,4.39604205]
0.4 [3.796785,4.39604205] [4.210882,4.39604205] [4.21873,4.39604205]
0.5 [3.851167,4.39604205] [4.244789,4.39604205] [4.21601,4.39604205]
0.6 [3.910577,4.39604205] [4.286146,4.39604205] [4.20352,4.39604205]
0.7 [3.971024,4.39604205] [4.316264,4.39604205] [4.118623,4.39604205]
0.8 [4. 030,703,4.39604205] [4.338612,4.39604205] [4.035371,4.39604205]
0.9 [4.088757,4.39604205] [4.351192,4.39604205] [3.072834,4.39604205]
0.91 [4.10 5164,4.39604205] [4.359421,4.39604205] [3.044328,4.39604205]

Table 27 MTTF w. r. t. λ10 ,λ11 ,λ12

MTTF w. r. t.λ10 MTTF w. r. t.λ11 MTTF w. r. t.λ12
0.1 [4.19842,4.39604205] [4.08431,4.39604205] [3.796355,4.39604205]
0.2 [4.001618,4.39604205] [3.84139,4.39604205] [3.59327,4.39604205]
0.3 [3.808442,4.39604205] [3.764321,4.39604205] [3.43058,4.39604205]
0.4 [3.729194,4.39604205] [3.683219,4.39604205] [3.37327,4.39604205]
0.5 [3.659392,4.39604205] [3.55942,4.39604205] [3.282759,4.39604205]
0.6 [3.550143,4.39604205] [3.42964,4.39604205] [3.166469,4.39604205]
0.7 [3.492753,4.39604205] [3.37275,4.39604205] [3.103403,4.39604205]
0.8 [3.329802,4.39604205] [3.247428,4.39604205] [3.004065,4.39604205]
0.9 [3.272829,4.39604205] [3.188528,4.39604205] [2.929254,4.39604205]
0.91 [3.249143,4.39604205] [3.129543,4.39604205] [2.910376,4.39604205]

Table 28 MTTF w. r. t. λ13 ,λ14 ,λ15

MTTF w. r. t.λ13 MTTF w. r. t.λ14 MTTF w. r. t.λ15
0.1 [4.079525,4.39604205] [4.112161,4.39604205] [4.12042,4.39604205]
0.2 [3.921838,4.39604205] [3.951001,4.39604205] [3.96162,4.39604205]
0.3 [3.77956,4.39604205] [3.807550,4.39604205] [3.86296,4.39604205]
0.4 [3.69259,4.39604205] [3.671573,4.39604205] [3.678141,4.39604205]
0.5 [3.66836,4.39604205] [3.544967,4.39604205] [3.503427,4.39604205]
0.6 [3.404107,4.39604205] [3.426801,4.39604205] [3.432522,4.39604205]
0.7 [3.393737,4.39604205] [3.316259,4.39604205] [3.395374,4.39604205]
0.8 [3.219265,4.39604205] [3.212626,4.39604205] [3.217654,4.39604205]
0.9 [3.096597,4.39604205] [3.115274,4.39604205] [3.120001,4.39604205]
0.91 [3.095575,4.39604205] [3.105862,4.39604205] [3.110561,4.39604205]
Reliability Assessment of Replaceable Shuffle Exchange Network … 233

Table 29 MTTF w. r. t. λ16

MTTF w. r. t.λ16
0.1 [3.455598,4.39604205]
0.2 [3.343221,4.39604205]
0.3 [3.275208,4.39604205]
0.4 [3.139055,4.39604205]
0.5 [3.07426,4.39604205]
0.6 [2.95839,4.39604205]
0.7 [2.89842,4.39604205]
0.8 [2.797389,4.39604205]
0.9 [2.653007,4.39604205]
0.91 [2.646179,4.39604205]

Table 30 MTTF w. r. t.λ1 ,λ2 ,λ3

MTTF w. r. t.λ1 MTTF w. r. t.λ2 MTTF w. r. t.λ3
0.1 [3.835971515, 4.245033] [3.835971515, 4.27769] [3.835971515, 4.48467]
0.2 [3.835971515, 4.08897] [3.835971515, 4.11925] [3.835971515, 4.31085]
0.3 [3.835971515, 3.94397] [3.835971515, 3.97214] [3.835971515, 4.14999]
0.4 [3.835971515, 3.8089] [3.835971515, 3.83517] [3.835971515, 4.00071]
0.5 [3.835971515, 3.726472] [3.835971515, 3.70733] [3.835971515, 3.73221]
0.6 [3.835971515, 3.56474] [3.835971515, 3.58774] [3.835971515, 3.61103]
0.7 [3.835971515, 3.42315] [3.835971515, 3.47562] [3.835971515, 3.49708]
0.8 [3.835971515, 3.35] [3.835971515, 3.3703] [3.835971515, 3.39085]
0.9 [3.835971515, 3.25204] [3.835971515, 3.27117] [3.835971515,3.29053]
0.91 [3.835971515, 3.24256] [3.835971515, 3.26158] [3.835971515, 3.28082]

Table 31 MTTF w. r. t. λ4 , λ5 , λ6
MTTF w. r. t.λ4 MTTF w. r. t.λ5 MTTF w. r. t.λ6
0.1 [3.835971515, 4.34453] [3.835971515, 4.36641] [3.835971515, 4.40186]
0.2 [3.835971515, 4.1812] [3.835971515, 4.33703] [3.835971515, 4.48618]
0.3 [3.835971515, 4.02971] [3.835971515, 4.44362] [3.835971515, 4.6746]
0.4 [3.835971515, 3.88881] [3.835971515, 4.57498] [3.835971515, 4.87625]
0.5 [3.835971515, 3.75743] [3.835971515, 4.70988] [3.835971515, 5.00642]
0.6 [3.835971515, 3.65638] [3.835971515, 4.84193] [3.835971515, 5.26432]
0.7 [3.835971515, 3.51962] [3.835971515, 4.9689] [3.835971515, 5.44485]
0.8 [3.835971515, 3.42843] [3.835971515, 5.09006] [3.835971515, 5.67248]
0.9 [3.835971515, 3.31011] [3.835971515, 5.20528] [3.835971515, 5.77759]
0.91 [3.835971515, 3.30029] [3.835971515, 5.21647] [3.835971515, 5.72164]
234 A. Khati and S. B. Singh

Table 32 MTTF w. r. t. λ7 ,λ8 ,λ9

MTTF w. r. t.λ7 MTTF w. r. t.λ8 MTTF w. r. t.λ9
0.1 [3.835971515, 4.54641] [3.835971515, 4.54378] [3.835971515, 4.74465]
0.2 [3.835971515, 4.33949] [3.835971515, 4.41388] [3.835971515, 4.33062]
0.3 [3.835971515, 4.33309] [3.835971515, 4.32902] [3.835971515, 4.24552]
0.4 [3.835971515, 4.38629] [3.835971515, 4.58728] [3.835971515, 4.25222]
0.5 [3.835971515, 4.42506] [3.835971515, 4.61997] [3.835971515, 4.29603]
0.6 [3.835971515, 4.48312] [3.835971515, 4.72925] [3.835971515, 4.35682]
0.7 [3.835971515, 4.57598] [3.835971515,4.79535] [3.835971515, 4.42547]
0.8 [3.835971515, 4.60006] [3.835971515, 4.82485] [3.835971515, 4.49736]
0.9 [3.835971515, 4.65637] [3.835971515, 4.95941] [3.835971515, 4.56997]
0.91 [3.835971515, 4.66189] [3.835971515, 4.96724] [3.835971515, 4.57721]

Table 33 MTTF w. r. t. λ10 ,λ11 ,λ12

MTTF w. r. t.λ10 MTTF w. r. t.λ11 MTTF w. r. t.λ12
0.1 [3.835971515, 4.70832] [3.835971515, 4.8825] [3.835971515, 4.86329]
0.2 [3.835971515, 4.35621] [3.835971515, 4.41449] [3.835971515, 4.43705]
0.3 [3.835971515, 4.30557] [3.835971515, 4.31137] [3.835971515, 4.35555]
0.4 [3.835971515, 4.33882] [3.835971515, 4.32792] [3.835971515, 4.3658]
0.5 [3.835971515, 4.40521] [3.835971515, 4.33622] [3.835971515, 4.41197]
0.6 [3.835971515, 4.48602] [3.835971515, 4.38514] [3.835971515, 4.4741]
0.7 [3.835971515, 4.57281] [3.835971515, 4.44231] [3.835971515, 4.54334]
0.8 [3.835971515, 4.66133] [3.835971515, 4.5031] [3.835971515, 4.61526]
0.9 [3.835971515, 4.75942] [3.835971515, 4.56502] [3.835971515, 4.6875]
0.91 [3.835971515, 4.73825] [3.835971515, 4.57122] [3.835971515, 4.69469]

Table 34 MTTF w. r. t. λ13 ,λ14 ,λ15

MTTF w. r. t.λ13 MTTF w. r. t.λ14 MTTF w. r. t.λ15
0.1 [3.835971515, 4.6731] [3.835971515, 4.92123] [3.835971515,4.73276]
0.2 [3.835971515, 4.48467] [3.835971515, 4.7127] [3.835971515, 4.53959]
0.3 [3.835971515, 4.31085] [3.835971515, 4.34452] [3.835971515, 4.34484]
0.4 [3.835971515, 4.14999] [3.835971515, 4.21095] [3.835971515, 4.19757]
0.5 [3.835971515, 4.00071] [3.835971515, 4.0297] [3.835971515, 4.04436]
0.6 [3.835971515, 3.87352] [3.835971515, 3.89945 [3.835971515, 3.91162]
0.7 [3.835971515, 3.73221] [3.835971515, 3.75743] [3.835971515, 3.77016]
0.8 [3.835971515, 3.56542] [3.835971515, 3.70842 [3.835971515, 3.66255]
0.9 [3.835971515, 3.49748] [3.835971515, 3.63464] [3.835971515, 3.53019]
0.91 [3.835971515, 3.48652] [3.835971515, 3.6228] [3.835971515, 3.51962]
Reliability Assessment of Replaceable Shuffle Exchange Network … 235

Table 35 MTTF w. r. t. λ16

MTTF w. r. t.λ16
0.1 [3.835971515, 4.75298]
0.2 [3.835971515, 4.55819]
0.3 [3.835971515, 4.37873]
0.4 [3.835971515, 4.21557]
0.5 [3.835971515, 4.05912]
0.6 [3.835971515, 3.82498]
0.7 [3.835971515, 3.78299]
0.8 [3.835971515, 3.65855]
0.9 [3.835971515, 3.54203]
0.91 [3.835971515, 3,35079]

Fig. 14 MTTF versus failure rates

(a) MTTF with respect to Terminal Reliability:

(i) It can be observed from Fig. 8 that as the values of λ1 ,λ2 ,λ3 ,λ4 ,λ5 and λ6
increase the lower bound of the MTTF of the considered SEN+?decrease while
the upper bound of MTTF remains consistent for all values of the mentioned
failure rates.
(ii) Also Fig. 9 shows that the upper bound of MTTF of the network decrease slowly
with increasing values of λ1 ,λ2 ,λ3 and λ6 while increase with increasing values
of λ4 ,λ5 . Also, for all failure rates the lower bound stays unaltered.
(b) MTTF with respect to Broadcast Reliability:
236 A. Khati and S. B. Singh

Fig. 15 MTTF versus failure rate

(i) By observing Fig. 11, it can be examined that with increasing values of
λ1 ,λ4 ,λ5 ,λ6 ,λ7 ,λ8 ,λ9 ,λ10 and λ11 the lower bounds of MTTF decrease while
they increase with increasing λ2 ,λ3 . The upper bound of the MTTF remains
constant concerning all the parameters.
(ii) From Fig. 12, it can be observed that the upper bound of the MTTF
of the considered network decrease with increase in the values of
λ1 ,λ4 ,λ5 ,λ6 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11 and increase with increasing values of λ2 ,λ3 .The
lower bound of the MTTF remains unchanged for all the mentioned parameters.
(c) MTTF with respect to Network Reliability:
(i) From Fig. 14 it can be detected that on increasing the values of the failure rates
λ1 ,λ3 ,λ4 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11 ,λ12 ,λ13 ,λ14 ,λ15 and λ16 ,lower bound of MTTF of the
proposed SEN decrease whereas with increasing value of λ6 , lower bound of
MTTF increase. Also, as λ2 and λ5 increase the lower bounds of the MTTF first
decreases and then increases. Here, also the upper bound remains constant.
(ii) On examining Fig. 15 it can be visualized that the upper bound
of the MTTF decrease with increasing values of the parameters
λ1 ,λ2 ,λ3 ,λ4 ,λ7 ,λ8 ,λ9 ,λ10 ,λ11 ,λ12 ,λ13 ,λ14 ,λ15 and λ16 while increase with
increasing λ6 . It can also be observed that if we increase λ5 then the upper
bound of MTTF first decrease then increase. Also, the lower bound of MTTF
remains the same for all the failure rates.
Reliability Assessment of Replaceable Shuffle Exchange Network … 237

One can observe from the results that the lower bound of the MTTF of the network
is lowest with respect to λ1 and is highest with respect to λ4 , and both are corre-
sponding to the NR of the network. The lowest and the highest value of the MTTF
are 0.1562452 and 20.39972 respectively. Also, the upper bound of the MTTF is
lowest with respect to the parameter λ4 with value 1.255834 and is highest with
respect to λ2 with value 27.954032. The lowest and highest values of the upper
bound of the MTTF are also obtained corresponding to the BR of the SEN+.

10 Conclusion

In this paper, we have considered a SEN+ the probability of whose components are
not known with accuracy and determined its reliability using the method of IUGF.
The transient state probabilities are obtained in intervals. The reliability is acquired
regarding three parameters: Terminal, Broadcast, and Network reliability. Also in the
observed network if any of the SE fails and the network quits operating at that point we
are able to replace that SE to hold the operation of the network. The SEN+ proposed
is of length 8 × 8. The RBD for the Terminal, Broadcast, and Network reliability
of SEN+ have been presented. The differential equations overseeing the network’s
behavior have been obtained with the aid of the method of the supplementary variable
technique. The probabilities of all the components are obtained by using the Laplace
transform method and hence the reliability of the network is computed. Furthermore,
the MTTF is examined and at last, we established our model by taking a numerical
example.

References

1. Bisht S, Singh SB (2020) Assessment of reliability and signature of Benes network using
universal generating function. Life Cycle Reliab Saf Eng
2. Bistouni F, Jahanshahi M (2014) Improved extra group network: a new fault- tolerant multistage
interconnection network. The J Supercomput 69(1):161–199
3. Bistouni F, Jahanshahi M (2014) Analyzing the reliability of shuffle-exchange networks using
reliability block diagrams. Reliab Eng Syst Saf 132:97–106
4. Bistouni F, Jahanshahi M (2018) Rearranging links: a cost-effective approach to improve
the reliability of multistage interconnection networks. Int J Internet Technol Secured Trans
8(3):336–373
5. Bistouni F, Jahanshahi M (2019) Determining the reliability importance of switching elements
in the shuffle-exchange networks. Int J Parallel Emergent Distrib Syst 34(4):448–476
6. Fard NS, Gunawan I (2005) Terminal reliability improvement of shuffle-exchange network
systems. Int J Reliab Qual Saf Eng 12(01):51–60
7. Kumar A, Singh SB, Ram M (2016) Interval-valued reliability assessment of 2-out-of-4 system.
In: 2016 international conference on emerging trends in communication technologies (ETCT).
IEEE, pp 1–4
8. Li CY, Chen X, Yi XS, Tao JY (2011) Interval-valued reliability analysis of multi-state systems.
IEEE Trans Reliab 60(1):323–330
238 A. Khati and S. B. Singh

9. Singh SB (2017) Reliability analysis of multi-state complex system having two multi-state
subsystems under uncertainty. J Reliab Stat Stud 10(1):161–177
10. Pan G, Shang CX, Liang YY, Cai JY, Li DY (2016) Analysis of interval-valued reliability of
multi-state system in consideration of epistemic uncertainty. In: International conference on
P2P, parallel, grid, cloud and internet computing. Springer, Cham, pp 69–80
11. Rajkumar S, Goyal NK (2016) Review of multistage interconnection networks reliability and
fault-tolerance. IETE Tech Rev 33(3):223–230
12. Sharma S, Kahlon KS, Bansal PK (2009) Reliability and path length analysis of irregular fault
tolerant multistage interconnection network. ACM SIGARCH Comput Architecture News
37(5):16–23
13. Yunus NAM, Othman M (2011) Shuffle exchange network in multistage interconnection
network: a review and challenges. Int J Comput Electr Eng 3(5):724
14. Yunus NAM, Othman M (2015) Reliability evaluation for shuffle exchange interconnection
network. Procedia Comput Sci 59:162–170
Modeling Software Vulnerability
Injection-Discovery Process
Incorporating Time-Delay and VIKOR
Based Ranking

Mohini Agarwal, Deepti Aggrawal, Subhrata Das, Adarsh Anand,

and Navneet Bhatt

Abstract Researchers have widely focused on catering software quality attributes

viz. reliability and maintainability. However, the reliance on software and software-
based products have redirected the focus of researchers/ engineers towards security.
The vulnerability in software can be due to design flaws, implementation errors,
configuration errors, etc., making it prone to attacks and can be used for malicious
activities. Timely detection and fixation of these loopholes can enhance the devel-
opment of safe and secure software thereby minimizing the efforts and resources
required to fix them afterwards. With the aim of modeling the discovery process of
vulnerability, in this chapter time delay-based formulation for vulnerability injection
and discovery has been proposed which has been modeled by considering the infi-
nite server queuing theory. For the empirical validation two vulnerability discovery
data has been used. Further VIKOR, a well-known Multi Criteria Decision Making
(MCDM) technique has been used to rank the different proposed models.

Keywords Discovery · Injection · Security breaching · VIKOR · Vulnerability

M. Agarwal
Amity School of Business, Amity University Uttar Pradesh, Noida 201303, India
D. Aggrawal
University School of Management and Entrepreneurship, Delhi Technological University, Delhi,
India
e-mail: deeptiaggrawal@dtu.ac.in
S. Das · A. Anand (B)
Department of Operational Research, University of Delhi, Delhi 110007, India
N. Bhatt
Anil Surendra Modi School of Commerce, SVKM’s Narsee Monjee Institute of Management
Studies (Deemed to be University), Mumbai 400056, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 239
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_10
240 M. Agarwal et al.

1 Introduction

Software security is crucial to ensure data security and prevent the misuse of the
software [28]. Software vulnerabilities are bugs/loopholes in the software system
which makes it prone to software breaching [22]. The presence of vulnerability is
a potential risk to the software whereas an exploit is the set of code to take advan-
tage of a security flaw. Vulnerability discovery process is like security check for
programmers/developers and assists in improving software quality. From the view-
point of software security, it is very important to review the bugs that are discovered
might be exploited by any hacker. Basically, when a software vulnerability is discov-
ered it follows a life cycles that consist of injection, discovery, disclosure, patch and
exploitation [8]. All software vulnerabilities are not equally harmful and are analyzed
based on their ease of detection, their exploitability and damage they can cause.
In the arena of software engineering, fault removal phenomena and vulnerability
handling has been given the utmost importance [10]. The most convenient way to
deal with vulnerabilities is to provide software patches or the corrective code to over-
come the loophole. The software firms tend to deploy software updates as soon as
possible and most of the vulnerability exploitations can be avoided by applying these
patches. Common vulnerabilities and Exposure (CVE), the vulnerability database
provides instances of publicly disclosed loopholes present in any software. Also,
many instances of vulnerabilities getting exploited can also be viewed. As quoted by
[32], the first half of the year 2019 has seen more than 3800 cases of publicly disclosed
software breaches uncovering 4.1 billion compromised records. It is even more aston-
ishing that out of this 3.2 billion of these records were uncovered by only eight
breaches. Moreover, in 2017 the infamous WannaCry vulnerability, a ransomware
affected the software community on a massive scale [23]. These breaching affects
the software and its dependent processes as well. Thereby making it a prime respon-
sibility to discover the loopholes before their exploitation by bad guys. Even though,
huge number of resources will be required for finding, disclosing, and fixing the attack
prone areas in susceptible software. The discovery process is being undertaken both
internally and externally which comprises of both testers (internal part of devel-
oping team and users (external, jointly engaged in discovering the vulnerabilities
throughout the lifespan of the software.
As described above, life cycle of vulnerability consist of injection, discovery,
disclosure, patch release and exploitation [4] and researchers have extensively
worked on each stage of the Vulnerability life cycle, be it working on induction,
be it its discovery and disclosure and be it the patch scheduling policies. The present
framework is different from some of the previous works in many aspects, which can
be understood as below (Table 1).
The chapter has been chronologically arranged as follows: after the brief discus-
sion on importance and introduction to vulnerabilities in Sect. 1, a brief literature on
vulnerability discovery process has been supplemented in Sect. 2. Methodology and
model development have been discussed in Sect. 3 followed by data analysis and
Modeling Software Vulnerability Injection-Discovery Process … 241

Table 1 Contrast between current and previous research

Present Research [4] [15] Research on Concept of VDM*
Role of Infinite Server Yes No No No
Queuing theory
Utility of any MCDM Yes Yes No No
technique for model selection
Time delay concept in VDM Yes No No No
Modeling vulnerability Yes Yes Yes Yes
discovery process
* Studies like [1, 2] and [27]

model validation in Sect. 4. In Sect. 5 the conclusion and lastly list of references are
provided.

2 Literature Review

Vulnerability Discovery Models (VDMs) are probabilistic models which helps in

mathematically assessing and quantifying the vulnerabilities present in the software.
Several practitioners have worked for developing mathematical structures based on
different distribution functions [2, 4, 6, 7, 17, 18, 27]. The basic software vulnerability
discovery model was the Anderson’s Thermodynamic Model [5]. In 2005, Rescorla
presented a linear and exponential vulnerability detection process whereas Alhazmi
and Malaiya model was an S-shaped, logistic VDM which defines the vulnerability
discovery process can be depicted as three phases as linear, learning and saturation
[2]. Alhazmi and Malaiya [2] also proposed an effort-based model for modeling
the effort consumed in terms of resources and budget in vulnerabilities detection.
Hanebutte and Oman [13] analyzed the program’s software faults and vulnerabilities
wherein they comprehended that these two are not distinct categories, but rather two
interrelated sets that should not be referred independently.
Woo et al. [33], quantitatively examined the Vulnerability Discovery Process
(VDP) for two main servers viz. Apache and IIS. The assessment and feasibility
based on time as well as effort. Younis et al. [35], investigated the applicability of a
VDM called Folded model, based on the Folded normal distribution, and compare
it with Alhazmi and Malaiya’s Logistic (AML) model [2]. The work of [21], is
based on exploring the various vulnerability techniques, basically the static anal-
ysis, fuzzing, penetration testing. Massacci and Nguyen [24] proposed an empirical
framework for the evaluation of existing VDMs performance in terms of quality and
their predictability and demonstrated superiority of their work over the traditional
approaches. Further, established the fact that simplest linear model works best for
the first 6–12 months after release date, apart from that logistics-based models are
also better choice. [11] suggested that the probability of exploitation rises with the
242 M. Agarwal et al.

vulnerabilities that are inherent to a software organization, the effectiveness of cyber-

attacks, and hacker’s advantage, but reduces with the allocation of resources for the
security and associated organization’s financial loss. Additionally, they investigated
three central decision instances in which security investment, information sharing,
or both would be respectively regulated by a social planner. Like ways,a lot has been
studied in depth about security [12, 20, 31, 34].
Kapur et al. [17] proposed a VDM based on logistic rate whereas the model
proposed by [4] utilized and proposed a hump-shaped vulnerability detection process.
Schatz and Bashroush [29] conducted a systematic literature review to identify and
analyze the state-of-the-art in economic assessment of information security invest-
ments in organizations. Kaur et al. [18] have discussed a vulnerability correction
model which models the time gap between the correction of leading and dependent
vulnerabilities. Kudjo et al. [19], examined the statistical significance of the annual
seasonal patterns and trends in vulnerability discovery using the weighted moving
window. They further worked on regression-based models as vulnerability predictors.
On examining the literature on vulnerability, it can easily be articulated that above
research focuses on assessment of vulnerabilities based on different distribution
patters. However, the work presented in this chapter quantifies vulnerabilities present
in software system, on the foundation of injection-based discovery process. This is
done to keenly understand the rationale behind introduction of vulnerabilities and its
impact on their discovery pattern. The concept used here has its roots in the field of
software reliability where the fault removal phenomenon is significantly dependent
on the fault detection process and the time delay between detection and correction
[16].

3 Methodology

In the chapter, an integrated modeling approach is proposed based on the concept of

infinite queuing server theory and which is as per the idea given by [14]. Here, we
have formulated different sets of models to illustrate several scenarios of vulnerability
discovery by considering the unification approach. Furthermore, due to availability of
several developed models it is essential for selecting the best model and for the precise
selection we have considered a ranking i.e., VIKOR approach [25, 26, 30]. In VIKOR
approach, several comparison criterions have been calculated for all the models. Each
alternative is compared with other alternatives based on the best calculated value of
each attributes.

3.1 Modeling Framework

Infinite Server Queuing Theory has been exhaustively utilized in various field like that
of software engineering [16] and Managerial decision making [3] to name a few. The
Modeling Software Vulnerability Injection-Discovery Process … 243

proposed modeling framework is formed by considering the Infinite Server Queuing

Theory used by [3] which helps in describing leap time between vulnerability getting
injected and its discovery. Assuming, the case when the vulnerability discovery
process is done in accordance with Non-Homogeneous Poisson Process (NHPP).
The present modeling framework is based on the following assumptions:
• Injection is synonymous to vulnerabilities introduction as the part of software
launch.
• Vulnerability Injection and Discovery process follows Non-homogeneous Poisson
Process (NHPP).
• Initial count of vulnerabilities present is constant, and the discovery process has
no influence on initial number.
• The count of non-discovered vulnerabilities affects the discovered vulnerabilities.
Following are the set of notations:

Vi (t) Mean Value function for vulnerability injection process

Vd (t) Mean Value function for vulnerability discovery process
v Number of vulnerabilities in the software
λ, β Shape and scale parameters

Let the counting process {M(t), t ≥ 0}, {N (t), t ≥ 0} represents number vulner-
abilities injected and discovered up to time  t  and the process started at initial time

t = 0 . Then the distribution of N (t) is given by [3]:
∞  
 Vi (t)e−Vi (t)
Pr{N (t) = n} = Pr{N (t) = n|M(t) = m } (1)
j=0
m!

Assuming the initial count of vulnerabilities introduced as  m  then the probability

that  n  loopholes will be discovered while the system experienced the discovery
process, can be given as follows:
 
m
Pr{N (t) = n|M(t) = m } = ( p(t))n (1 − p(t))m−n (2)
n

where p(t) represents the probability that an arbitrary vulnerability is getting discov-
ered at time  t  , which can be defined using steiltjes convolution and the concept of
the conditional distribution of arrival times, given as:

t
d Vi (u)
p(t) = Fd (t − u) (3)
Vi (t)
0
244 M. Agarwal et al.

Probability distribution function (pdf) of cumulative count of vulnerabilities

discovered by time  t 
t 
⎛ ⎞
t − Fd (t−u)d Vi (u)
e 0
Pr{N (t) = n} = ⎝ Fd (t − u)d Vi (u)⎠ (4)
m!
0

 t Equation (4) describes

 that Vd (t) follows an NHPP with MVF
Fd (t − u)d Vi (u) .
0
Therefore,
⎛ t ⎞

Vd (t) = ⎝ Fd (t − u)d Vi (u)⎠ (5)
0

Henceforth, the mean value function for vulnerability injection and distribution
pattern of vulnerabilities discovery can assist in determining the count of vulnerabil-
ities present in the system at any instant of time. For analytical understanding four
different cases have been considered.
VDM-I: Assuming the injection of vulnerability to follow the constant detection
pattern whereas the discovery process follows exponential pattern i.e., Vi (t) ∼ 1(t)
and Fd (t) ∼ exp(λ). On substitution the functional form for the same in Eq. (5) leads
to:
 
Vd (t) = v 1 − e−λt (6)

The function form given by Eq. (6) is like Rescorla’s Exponential (RE) model
(2005).
VDM-II: Considering another case of where injection happen to follow the
constant pattern whereas the discovery followed s-shaped pattern encompassed with
the learning i.e., Vi (t) ∼ 1(t) and Fd (t) ∼ logistic(λ, β). Putting the functional
form in Eq. (5) leads to:
 
1 − e−λt
Vd (t) = v (7)
1 + βe−λt

Equation (7) defines the logistic distribution-based vulnerability discovery model

being dependent on constant pattern of vulnerability injection which is in line with
Alhazmi and Malaiya’s Logistic (AML) model (2005).
VDM-III: Here we assumed that the vulnerability injection remains same over
the time and PDF for vulnerability discovery process to be Weibull behavior i.e.,
Vi (t) ∼ 1(t) and Fd (t) ∼ W eibull(λ, k). Putting the functional form in Eq. (5)
leads to:
Modeling Software Vulnerability Injection-Discovery Process … 245

Vd (t) = v 1 − e−λt
k
(8)

The function for VDM as stated in Eq. (8) is equivalent to the model proposed
by [15] where the researcher worked with Weibull distribution for vulnerability
discovery.
VDM-IV: If injection and discovery distribution function to follow one-stage
Erlang distribution with same rate i.e., Vi (t) ∼ exp(λ) and Fd (t) ∼ exp(λ). On
substituting in Eq. (5) gives:
 
Vd (t) = v 1 − (1 + λt)e−λt (9)

VDM-V: Considering the case when rates of vulnerability injection and discovery
both follow exponential pattern with different rates i.e.,Vi (t) ∼ ex p(λ1 ) and Fd (t) ∼
ex p(λ2 ). Using Eq. (5) we get:
 
1  −λ2 t 
Vd (t) = v 1 − λ1 e − λ2 e−λ1 t (10)
(λ1 − λ2 )

VDM-VI: Consider vulnerability injection intensity to be exponentially changing

over time and vulnerability discovery process follows Logistic distribution i.e.,
Vi (t) ∼ ex p(λ) and Fd (t) ∼ logistic(λ, β); and again, making use of Eq. (5)
we get,
 
(1 + β)e−λt
Vd (t) = v(1 − e−λt ) + v(1 + β)e−λt log (11)
1 + βe−λt

VDM-VII: Assuming intensity of vulnerability injection and discovery process

both to follow Logistic distribution i.e. Vi (t) ∼ logistic(λ, β) and Fd (t) ∼
logistic(λ, β); and again making use of Eq. (5) we get,
    
1 − e−λt (1 + β)2 e−λt (1 + β)e−λt
Vd (t) = v + v 2 λt + 2log (12)
1 − β 2 e−λt 1 − β 2 e−λt 1 + βe−λt

VDM-VIII: Yet another case that can be on constant pattern for injection intensity
and vulnerability discovery process follows hump shaped rate of detection Vi (t) ∼
1(t) and Fd (t) ∼ hump Shape distribution (λ, β). Putting the functional form in
Eq. (5) leads to:
    
−λ 1
− 1
Vd (t) = v 1 − e 1+βe−λt 1+β
(13)

Equation (13) defines the Hump shaped based vulnerability discovery model being
dependent on constant pattern of vulnerability injection which is in line with [4].
246 M. Agarwal et al.

Further, the Eqs. (6)–(13) can assist in determining the count of discovered vulnera-
bilities based on the pattern of vulnerability injection which can be useful instruments
for framing patching and update schedules.

4 Data Analysis and Model Validation

Model competency and suitability can be demonstrated by its behavior to repli-

cate and forecast based on observed behavior. To assess the performance of
proposed VDMs, the data of two operating systems obtained from the CVE
Details which is a Common Vulnerabilities and Exposures (CVE) security vulner-
ability database/information source [9] i.e., Mozilla Firefox (VDS-I) and Windows
Server 2008 (VDS-II) has been utilized. Further, various comparison criterions are
calculated to compare different models.

4.1 Estimation of Parameters

For parameter estimation a statistical software, SAS has been used which is a compre-
hensive and flexible software used for data management and analytics. Non-Linear
Regression (NLR) modules have been used, it is an analytical method which classi-
fies the variables as predictor and response variable. Here, we have considered time
as the predictor and vulnerabilities as the response variable.

4.2 Goodness-of-Fit Criteria

To examining the predictive nature of the developed model’s different compar-

ison criteria have been computed to assess the models numerically. Six predictive
criterions are used viz. SSE, MSE, Variance, Root MSE, R-Square, Adj. R-square
[3].

4.3 Graphical Approach (Goodness of Fit)

Figures 1 and 2 depicts the goodness-of-fit curves, i.e., it displays the how the
observed vulnerability and the model predicted vulnerabilities variate over the time.
Classification of models based on pictorial representation sometimes becomes
ambiguous. Hence, a need for an approach for better classification is necessary.
Modeling Software Vulnerability Injection-Discovery Process … 247

Fig. 1 Goodness of fit for VDS-I

Fig. 2 Goodness of fit for VDS-II

4.4 VIKOR Analysis

VIKOR, a very important multi criteria decision making approach, was proposed
by Serafim Opricovic [25, 26, 30] and is also known as the compromise ranking
method. The technique helps choose the best option out of the various alternatives
by considering the conflicting criteria. The utility lies in the fact that it facilitates
decision making in situations where decision makers are unable to express their
preferences. It makes uses of an aggregate function by expressing closeness to the
ideal solution and to rank the various alternatives. Along with the ranking, VIKOR
also provides a compromise solution with an advantage rate which maximizes the
group utility for the majority and minimizes the individual regret for the opponent.
The optimal solution is closest to the ideal solution and farthest from the nadir
solution. Following are the steps involved in its implementation:
Step-1: Establish a matrix of criteria and different alternatives.
Step-2: Normalization of decision matrix.
248 M. Agarwal et al.

Step-3: Calculate the weight of the normalized decision matrix.

m   ∗ 
xi − xi j
Si = wj ∗
j=1
xi∗ − xi−
  ∗ 
xi − xi j
max w j ∗
Ri = 
x∗ − x− i i
j

Step-4: Determine the ideal solution and nadir solutions (negative ideal solution)

I deal Solution N adir Solution

S ∗ = 
min Si R ∗ = 
min Ri
i i
S − = 
max Si R ∗ = 
max Ri
i i

Step-5: Compute the distance for each alternative.

Step-6: Calculate the relative closeness to the ideal solution.
   
∗ Si − S ∗ Ri − R ∗
Qi = v + (1 − v)
S− − S∗ R− − R∗

Step 7: Rank the Preference Order. The alternative with the smallest value is the
best solution.

4.5 Illustrated Examples

The idea behind this presentation is to evaluate the appropriateness of the VIKOR
method that an all-inclusive classification of the alternative models can be directed
by considering several attributes related to the models for a given dataset.

Example 1: Vulnerability discovery data for Mozilla Firefox (DS-I) has been
collected from the CVE details [9] and for the assessment, optimal model selec-
tion and ranking on the models on the bases of six different performance criteria’s
i.e., SSE, MSE, Variance, Root-MSE, R2 and Adj.-R2 . The parameter values for the
eight proposed models have been identified by applying the non-linear least square
(NLLS) regression technique and the parameter is provided in Table 2.

The value of the six performance criteria have been obtained. The value of the
attributes for each VDM is given in Table 3.
Based on the above comparison table, it is noticed that the selection of best model
becomes difficult. To evade this difficulty, we have applied VIKOR method to classify
Table 2 Parameter estimates for VDS-I
Parameter VDM-I VDM-II VDM-III VDM-IV VDM-V VDM-VI VDM-VII VDM-VIII
v 9850 2670.246 4707.227 5761.476 6524.02 2870.237 4818.152 14,853.58
λ1 0.010351 0.210592 0.003688 0.067987 0.101144 0.199337 0.089582 0.210377
λ2 – – – – 0.040864 – – –
β – 14.67787 – – 4.21 0.23 16.25926
k – – 1.734185 –
Modeling Software Vulnerability Injection-Discovery Process …
249
 250

Table 3 Comparison criteria for VDS-I

Criteria VDM-I VDM-II VDM-III VDM-IV VDM-V VDM-VI VDM-VII VDM-VIII
SSE 502,610 33,948 24,580.3 24,346.5 24,213.1 31,697.4 24,894.4 33,957.3
MSE 31,413.1 2424.9 1755.7 1623.1 1614.2 2113.2 1659.6 2425.5
Variance 163.65 45.52 39.09 39.00 38.88938 44.26052 39.44352 45.52526
Root MSE 177.2 49.2428 41.9015 40.2877 40.1772 45.9692 40.738 49.2496
R2 0.916 0.994 0.996 0.996 0.996 0.994 0.995 0.994
Adj.R 2 0.916 0.993 0.995 0.996 0.995 0.994 0.995 0.993
M. Agarwal et al.
Modeling Software Vulnerability Injection-Discovery Process … 251

Table 4 VIKOR analysis for VDS-I

Models Si Ri Qi Rank
VDM-I 1 0.166667 1 8
VDM-II 0.038227 0.011027 0.05159777 6
VDM-III 0.005368 0.002097 0.00834962 3
VDM-IV 0.000378 0.000148 0 1
VDM-V 0.002083 0.002083 0.0066646 2
VDM-VI 0.027952 0.007175 0.03489317 5
VDM-VII 0.00608 0.002083 0.00866386 4
VDM-VIII 0.038249 0.011035 0.05163355 7

the models by considering all six criteria taken together and the respective ranking
of the model is shown in Table 4.
The ranking is outlined on the relative closeness value of the models that is estab-
lished by considering six performance measures as contributing attributes together
used in VIKOR analysis. The model with the lowest value is given rank 1, that
with second lowest as rank 2, and so on. As per the results found after applying the
VIKOR method, it can be seen that the VDM-IV is ranked as first position followed
by VDM-V. Hence, the model that depicts the injection and discovery distribution
function follow one-stage Erlang distribution with same rate attain 1st rank.

Example 2: The vulnerability discovery data of Windows Server 2008 (DS-II) has
been collected from the CVE details [9] and for the assessment, optimal model selec-
tion and ranking on the models on the bases of six different performance criteria’s
i.e., SSE, MSE, Variance, Root-MSE, R2 and Adj.-R2 . The parameter values for the
eight proposed models have been identified by applying the non-linear least square
(NLLS) regression technique and the parameter is provided in Table 5.

The value of the six performance criteria have been obtained and the value of the
attributes for each VDM is given in Table 6.
Similarly, based on the above table it is difficult to judge which model is
performing best. Thus, VIKOR approach has been used to rank the models based on
all six criteria taken together and shown in Table 7.
The model with the lowest value is given rank 1, that with second lowest as rank
2, and so on. The results obtained depict that VDM-II is ranked as first position
followed by VDM-VIII. The outcomes of above examples show that not all the
scenarios can be best explained by all models and their predictive capabilities are
influenced accordingly. Also, the VDM-I is ranked 8 on both the data sets implying
exponential growth in vulnerability discovery is least suited to understand its pattern.
 252

Table 5 Parameter estimates for VDS-II

Parameter VDM-I VDM-II VDM-III VDM-IV VDM-V VDM -VI VDM-VII VDM-VIII
v 8402.65 2996.618 4519 9614 8364.35 6143.2 3787.1 9314.05
λ1 0.010504 0.253477 0.001766 0.049576 0.065066 0.150076 0.187941 0.296056
λ2 – – – – 0.045
β 29.40423 – 6.696046 2.019402 36.593999
k 2.069643 –
M. Agarwal et al.
Table 6 Comparison criteria for VDS-II
Criteria VDM-I VDM-II VDM-III VDM-IV VDM-V VDM-VI VDM-VII VDM-VIII
SSE 359,416 18,222.6 28,684.3 36,223 39,928.1 24,074.5 31,869.6 24,474.7
MSE 29,951.3 1656.6 2607.7 3018.6 3327.3 2188.6 2897.2 2225
Variance 162.9228 38.96507 48.43078 54.6645 57.27339 43.9689 50.80567 44.9426
Root MSE 173.1 40.7014 51.0653 54.9416 57.6831 46.7824 53.826 47.1696
R2 0.853 0.992 0.988 0.985 0.983 0.990 0.987 0.990
Adj.R 2 0.853 0.991 0.987 0.985 0.983 0.989 0.985 0.989
Modeling Software Vulnerability Injection-Discovery Process …
253
254 M. Agarwal et al.

Table 7 VIKOR analysis for VDS-II

Models Si Ri Qi Rank
VDM-I 1 0.166667 1 8
VDM-II 0 0 0 1
VDM-III 0.046113 0.013046 0.0621956 4
VDM-IV 0.07149 0.021109 0.09907072 6
VDM-V 0.08689 0.024616 0.11729418 7
VDM-VI 0.025189 0.007655 0.03555907 2
VDM-VII 0.059657 0.016522 0.07939336 5
VDM-VIII 0.027395 0.008142 0.03812452 3

5 Conclusion

This chapter has been focused on modeling the impact of vulnerability injection
on the discover process. As quoted “Vulnerability finding increases total software
quality”, it would be better to discover the vulnerabilities before being discovered by
exploiters. Here, we have formulated a unification strategy based on infinite server
queuing theory for the modeling of eight different VDMs which are based on different
injection and discovery patterns. These models were tested for prediction capability
on two different data sets collected from CVE repository. Moreover, for in depth
understanding a ranking procedure VIKOR analysis has been used and ranks are
assigned based on six comparison criterions. It was observed that the predictive capa-
bilities of the proposed models vary with the change in shape of the data. According
to authors knowledge, this is the first time that the VDM is being studied based on
the injection pattern of vulnerabilities.
In future, we wish to work on extending the field of vulnerability discovery
modeling to incorporate more flexibility in locating loopholes because of random
lag function approach and thereby establishing the equivalence with the proposed
approach.

References

1. Alhazmi OH, Malaiya YK, Ray I (2007) Measuring, analyzing and predicting security
vulnerabilities in software systems. Comput Secur 26(3):219–228
2. Alhazmi OH, Malaiya YK (2005) Modeling the vulnerability discovery process. In: Proceed-
ings of the 16th IEEE international symposium on software reliability engineering. IEEE,
Chicago, IL, pp 138–147
3. Anand A, Agarwal M, Aggrawal D, Singh O (2016) Unified approach for modeling innovation
adoption & optimal model selection for the diffusion process. J Adv Manage Res-An Emerald
Insight 13(2):154–178
4. Anand A, Bhatt N (2016) Vulnerability discovery modeling and weighted criteria based ranking.
J Indian Soc Probab Stat 1–10
Modeling Software Vulnerability Injection-Discovery Process … 255

5. Anderson R (2002) Security in open versus closed systems—the dance of Boltzmann, Coase
and Moore. Cambridge University, England, Technical report, pp 1–15
6. Arora A, Nandkumar A, Telang R (2006) Does information security attack frequency increase
with vulnerability disclosure? An empirical analysis. Inf Syst Front 8(5):350–362
7. Bhatt N, Anand A, Aggrawal D (2019) Improving system reliability by optimal allocation of
resources for discovering software vulnerabilities. Int J Qual Reliab Manage
8. Bhatt N, Anand A, Yadavalli VSS, Kumar V (2017) Modeling and characterizing software
vulnerabilities. Int J Math, Eng Manage Sci (IJMEMS) 2(4):288–299
9. CVE (2019) https://www.cvedetails.com/. Accessed 20 Jan 2020
10. Chatterjee S, Saha D, Sharma A (2021) Multi-upgradation software reliability growth model
with dependency of faults under change point and imperfect debugging. J Softw: Evol Process
e2344
11. Gao X, Zhong W, Mei S (2015) Security investment and information sharing under an alternative
security breach probability function. Inf Syst Front 17(2):423–438
12. Garg S, Singh RK, Mohapatra AK (2019) Analysis of software vulnerability classification
based on different technical parameters. Inf Sec J: A Glob Perspect 28(1–2):1–19
13. Hanebutte N, Oman PW (2005) Software vulnerability mitigation as a proper subset of software
maintenance. J Softw Maint Evol Res Pract 17(6):379–400
14. Inoue S, Yamada S (2002) A software reliability growth model based on infinite server queuing
theory. In: Proceedings 9th ISSAT international conference on reliability and quality in design.
Honolulu, HI, pp 305–309
15. Joh H, Kim J, Malaiya YK (2008) Vulnerability discovery modeling using Weibull distribution.
In: 2008 19th international symposium on software reliability engineering (ISSRE). IEEE, pp
299–300
16. Kapur PK, Pham H, Gupta A, Jha PC (2011) Software reliability assessment with OR
applications. Springer, London Limited
17. Kapur PK, Sachdeva N, Khatri SK (2015) Vulnerability discovery modeling. In: International
conference on quality, reliability, infocom technology and industrial technology management,
pp 34–54
18. Kaur J, Anand A, Singh O (2019) Modeling software vulnerability correction/fixation process
incorporating time lag. In: Boca Raton FL (ed) Recent advancements in software reliability
assurance. CRC Press, pp 39–58
19. Kudjo PK, Chen J, Brown SA, Mensah S (2019) The effect of weighted moving windows
on security vulnerability prediction. In: 2019 34th IEEE/ACM international conference on
automated software engineering workshop (ASEW). IEEE, pp 65–68
20. Kumar A, Ram M (2018) System reliability analysis based on Weibull distribution and hesitant
fuzzy set. Int J Math Eng Manag Sci 3(4):513–521. https://doi.org/10.33889/IJMEMS.2018.3.
4-037
21. Liu B, Shi L, Cai Z, Li M (2012) Software vulnerability discovery techniques: a survey. In: 2012
fourth international conference on multimedia information networking and security. IEEE, pp
152–156
22. Liu Q, Xing L (2021) Survivability and vulnerability analysis of cloud RAID systems under
disk faults and attacks. Int J Math Eng Manag Sci 6(1):15–29. https://doi.org/10.33889/IJM
EMS.2021.6.1.003
23. MSRC Team (2017) Customer Guidance for WannaCrypt attacks. Accessed 25th Jan 2020
24. Massacci F, Nguyen VH (2014) An empirical methodology to evaluate vulnerability discovery
models. IEEE Trans Softw Eng 40(12):1147–1162
25. Opricovic S (1998) Multicriteria optimization of civil engineering systems. Faculty Civ Eng,
Belgrade 2(1):5–21
26. Opricovic S, Tzeng GH (2004) Compromise solution by MCDM methods: a comparative
analysis of VIKOR and TOPSIS. Eur J Oper Res 156(2):445–455
27. Rescorla E (2005) Is finding security holes a good idea? IEEE Secur Priv 3(1):14–19
28. Ryan KT (2016) Software processes for a changing world. J Softw: Evol Process 28(4):236–240
256 M. Agarwal et al.

29. Schatz D, Bashroush R (2017) Economic valuation for information security investment: a
systematic literature review. Inf Syst Front 19(5):1205–1228
30. Tong LI, Chen CC, Wang CH (2007) Optimization of multi-response processes using the
VIKOR method. The Int J Adv Manuf Technol 31(11–12):1049–1057
31. Verma R, Parihar RS, Das S (2018) Modeling software multi up-gradations with error genera-
tion and fault severity. Int J Math Eng Manag Sci 3(4):429–437. https://doi.org/10.33889/IJM
EMS.2018.3.4-030
32. Winder D (2019) https://www.forbes.com/sites/daveywinder/2019/08/20/data-breaches-exp
ose-41-billion-records-in-first-six-months-of-2019/#6e20808bd549, Accessed 25th Jan 2020
33. Woo SW, Joh H, Alhazmi OH, Malaiya YK (2011) Modeling vulnerability discovery process
in Apache and IIS HTTP servers. Comput Secur 30(1):50–62
34. Yang SS, Choi H, Joo H (2010) Vulnerability analysis of the grid data security authentication
system. Inf Secur J: A Glob Perspect 19(4):182–190
35. Younis A, Joh H, Malaiya Y (2011) Modeling learning less vulnerability discovery using a
folded distribution. In: Proceedings of SAM, vol 11, pp 617–623
Assessment of Reliability Function
and Signature of Energy Plant Complex
System

Monika Negi, Megha Shah, Akshay Kumar, Mangey Ram, and Seema Saini

Abstract The objective of this chapter is to calculate the performance of a complex

system of an energy plant. In this model, deals with the universal generating function
technique (UGF) for analyzing the performance of the complex fixable condensate
system of an energy plant, the process of analyzing the reliability has been used
for exaggerate the capability of the complex system by reducing the failure rate.
The illustrated system is divided into six subsystems those are arranged into series
parallel combination and by using the z transformation with UGF technique we
find the reliability, signature, tail signature, expected time and expected cost and
these resulting outcomes help to get the highly beneficial probabilistic approach that
analyzed the signature and the reliability of the system in effective manner.

Keywords Structure function · Signature · Tail signature · Thermo plant energy

system · Universal generating function

M. Negi · M. Shah · A. Kumar (B)

Department of Mathematics, Graphic Era Hill University, Dehradun, Uttarakhand, India
M. Ram
Department of Mathematics, Computer Science and Engineering, Graphic Era Deemed To Be
University, Uttarakhand, India
Institute of Advanced Manufacturing Technologies, Peter the Great St. Petersburg Polytechnic
University, 195251 Saint Petersburg, Russia
S. Saini
Department of Mathematics, Graphic Era Deemed To Be University, Dehradun, Uttarakhand,
India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 257
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_11
258 M. Negi et al.

1 Introduction

Reliability is the capability of a system or a model to determine the performance of

the entire system as well as the elements such as perfectly function or totally failed.
Osaki and Nakagawa [30] gave a short assessment about stochastic systems on the
lists of reliability of models through chosen reference on the system dependability
models with the help of Markov techniques. Dhillon [4] considered two stochastic
systems for the study of the accessibility and reliability of k-out-of-n component
group of systems having critical human errors (where an unnecessary model stops
working because of human error) and common-cause failures by making the use of
additional variables and Markov method. Sahner and Trivedi [33] described about
the software instrument for the study of stochastic model. SHARPE which stands
for Symbolic Hierarchical Automated Performance Evaluator is a software system
which helps in the examination of stochastic systems. This software was developed
for the designing of real systems and has been used for this very purpose for so
many years. Another model inexact two-stage stochastic programming (ITSP) is
offered by Huang et al. [10] for water resources under unpredictability. Authors had
analysed about the suppositional case study that illustrated the usage of the advanced
system and the solutions that had got are then evaluated and explicated for producing
decision substitutes and recognising consequential constituents of those which are
influencing the system’s efficiency. Weng [37] explored about the merged usage
of satellite remote sensing, stochastic model technologies and geographic informa-
tion systems (GIS). The investigation reveals that the unification of satellite remote
sensing and GIS were an efficient technique for evaluation of land use change’s rate,
direction and spatial pattern. Zacharof and Butler [38] supposed a model and the chief
aim was to have the ability to control landfill sites, subsequent leachate, and biogas
production for the optimisation of disposal process and the reduction of environ-
mental impact. A common analytic model-based technique was determined which
helps in a cloud service’s end-to-end performability study in which their approach is
explained from Infrastructure taken as Service cloud. On the contrary to a sole one-
level monolithic model, their method yielded a soaring precision of the model that is
tractable and scalable. This method had been easily expanding to other kinds of cloud
services and are appropriate for the private, public and hybrid clouds by [7]. Michael
et al. [25] explained about Markov and Discrete event simulation (DES) model and
these models are useful for failure interaction. Experiment’s designs had also been
introduced by them which was performed on both the approaches and DES model
was as precise as Markov model came out to be the conclusion. Machida et al. [21]
determined an automatic model constitution approach to a hierarchical stochastic
model from a system design which was the sensible union of combinatorial and
state-space models. The attained Hierarchical stochastic model was then calculated
analytically and significantly quickly. Andrzejczak [1] determined the most general
methods which are used for designing of fixable systems. Author explained stochastic
Assessment of Reliability Function and Signature … 259

procedures as instances of reliability-based systems for the support of the continua-

tion linked decisions. The availability of the casting procedures in foundry work is
well explained through Laplace transformation and supplementary variable method
and by the application of probability approach the defects can be overcome. The
outcomes were defined and illustrated by graphs from which authors can readily
envision the nature of the model after the appearance of any error by [16]. Malik
and Munday [22] illustrated to magnify the productivity of the models having hard-
ware redundancy which can be enhanced by taking correct repairing policies and
elements of low failure percentages by improving the repair time. The method of
redundancy is also regarded as an instrument to intensify longevity of the systems.
Manatos et al. [23] studied a multi-state rectifiable production model that is discussed
in preventive maintenance has been proposed and the reason behind its implemen-
tation is to enhance its dependability and performance. They have given a generic
model for the production system with the help of a semi-Markov procedures, for
scrutinizing the limiting behaviour of systems. Ram and Goyal [31] examined the
performance of ATM channel and also inspected the dependability measures of a
fixable ATM system which helps the reliability engineers or designers for the deter-
mination of how reliability could be enhanced using suitable alterations. It also seeks
the sensitive scanning of different variations in reliability attributes along with the
modifications in accurate values of input parameters. Goyal and Ram [8] gave the
assessment of how a wind electricity generated power plant performs by examining
the reliability measures. The research aimed at improvement of wind power plant’s
performance with the aid of different techniques. To secure farm-based biogas plant
projects for the long-run. Dennehy et al. [3] examined the necessity of it and had
objected to attain co-substrate’s stabled stocks and to classify plants volume which
is grounded on the accessibility of the co-substrates that drives methane production.
Komal [11] discussed the fuzzy reliability of coal fired thermo plant using fuzzy
methods having membership function from triangular fuzzy numbers etc. Gupta [9]
determined about the approach to illustrate the condensate system’s performance of
thermal power plant based on coal. The illustration was completed through avail-
ability matrices and these are evolved from various merger of repair plus failure rates
of each and every sub system to maximise the generation of power and to make it
working for a long-term. A two-stage stochastic degradation model is proposed by
[5], in the first stage, degeneration procedures of two correlated execution proper-
ties were explained by correlated bivariate Wiener, second stage was designed by
univariate Wiener method for adopted modelling. When both the degradation levels
attained its thresholds then the model stopped working absolutely in the last stage.
Explicit statements of system performance are grounded on maintenance policies and
non-identical inspection. Kumar et al. [17] illustrated an unconventional approach
for the accessibility scanning of a nuclear power plant’s Digital Feed Water Control
System, which supposed the preservation and restoration of the main-steam safety
valves. Stochastic Petri Net is used to model the system which took over all model
prerequisites, the partial failures of its subsystems along with general reasons of
failures.
260 M. Negi et al.

In signature reliability’s context for a multi-state system, Levitin [18] elucidated a

redundancy optimization problem. Their principal aim was to reduce the total invest-
ment costs while contenting demands, presented by an increasing demand curve, with
stated probability. Levitin [19] illustrated reliability examination and optimization of
different models which generalize successive k-out-of-r-from-n-F model comprising
of n consecutive multi state ordered unit where single component has its own state
using UGF. Levitin [20] gave the comprehensive description of UGF technique and
its application in both the field of binary and multi state system reliability anal-
ysis. Navarro et al. [28, 29] explained about the coherent system and its dependent
component. Also hyper minimal and hyper maximal distributions are two families of
univariate distributions which was introduced by the writers. In the context of reli-
ability theory, these families have engrossing applications in those elements which
having coherent system lifetime distributions. Samaniego [34] and Samaniego et al.
[35] computed dynamic signatures along with their usage for the comparison of
the reliability of used and new system. Authors discovered how one can compare a
fresh model to the another efficiently burned-in to the kth element failure, elements
also spot the situations in which burn-in is inferior (or is superior) to the fielding
of a fresh model. Coherent systems are pertinent subjects in reliability and survival
studies and is being examined by [26], it also functions as a handy apparatus to
calculate the model dependability and anticipated life-duration to compare various
models with the help of stochastic orderings. Here the elements are independent
and identically distributed (i.i.d.) with a common continuous distribution. Navarro
and Rychlik [27] regulated bounds and comparisons for the anticipated lifespan of
mixed or coherent systems having independent units that had unspecified distri-
butions depending on the elements anticipated lifespans and on their domination
polynomials which is only based on the system structure function. Using reliability
functions of the given models, Kumar and Singh [13] evaluated signature reliability of
complex and k-out-of-n coherent system. Kumar and Singh [14] computed expected
price and Barlow-Proschan index which were discovered through reliability or struc-
ture function of the linear multi-state sliding window system by the method of Owen
and UGF technique. Kumar and Singh [12, 15] supposed sliding window coherent
model in the event of multiple failures and also had analyzed about A-within-B-from-
D/G SWCS incorporating multiple failures. Owen’s method and UGF method is used
for the assessment of signature and its measures in the proposed model. Rushdi and
Ghaleb [32] studied the weighted k-out-of-n complex coherent system and discussed
system’s characteristic properties with some effective techniques. Triantafyllou [36]
evaluated the signature of complex (k,d)-out-of-n:F Constrained system with the help
of some proposed algorithm. Mean life time and conditional mean residual lifetime
also computed based on signature expressions.
In this studied we have considered thermo power plant complex system combina-
tion of series parallel manner using UGF method and find tail signature, signature,
mean time to failure and cost. In Sect. 1 discussed review and literature based on
Assessment of Reliability Function and Signature … 261

reliability and signature analysis. Section 2 studied some numerical formula for
computing signature factor. Section 3 concluded model configuration. In Sect. 4
having numerical result on the basis of Owen’s and UGF method. Section 5 and 6
describe result and discussion and conclusion part.
Notation
n = total number of components in the multi-state system
p = the probability of system
q = failure probability of the system
ϕ& ⊗ = shifting or multiplication operator
R = reliability of the system
E(X) = expected X of the components
E(T ) = excepted working time of units
S = signature of the system with l components
s = tail signature of system with l components
C = cost of the system having i units
H = number of components which is associated with n units.

2 Evolution of the Signature, Minimal Signature, Tail

Signature and Cost from Structure Function
of the Proposed Model

The signature and its measures of the various coherent systems which are i.i.d.
elements [for instance, order statistics and reliability function methods [2, 28, 29]
can be evaluated in the following manner.
Signature and its measures are defined as

1  1 
Sl =   ϕ(H ) −   ϕ(H ) (1)
n H ⊆[n]
n H ⊆[n]
n − l + 1 | H |=n−l+1 n − 1 | H |=n−1

From systems structure function which are of polynomial’s form and having i.i.d.
components
 

m
m e n−e n
H ( p) = Cj p q and Ce = si , e = 1, 2, .., n (2)
e=1
e i=n−e+1
262 M. Negi et al.


n 
The system is Sl = si = ⎛ 1 ⎞ ϕ(H ) with n-tuples set function
n
i=l+1 ⎝ ⎠ | H |=n−l
n−1
such as S = (S0 , ..., Sn ) and we have to find the tail signature along with signature
of the system we considered. Using Taylor’s expansion and signature is

n−1 l
Sl = D P(1), l = 0, 1, ..., n (3)
n!

to transform polynomial function into P(X ) = X n H 1

X
of proposed method [24]
which is

s = Sl−1 − Sl , l = 1, ..., n (4)


n 
n
and E(T ) = μ Ci
i
and E(X ) = i.si , i = 1, 2, ..., n are defined as to determine
i=1 i=1
the cost and anticipated lifespan of the system from reliability function [6, 26] which
are based on minimal signature and quantity of failed components of the system with
one as the mean value.

3 Model Description

Thermo power plant system is the generalization of coal fired power plant system.
Thermo-plant model is an arrangement of apparatus and complex manner which is
helpful in the conversion of fossil fuel’s chemical energy into thermal energy. Then
transform it into steam it is shifted to water at high temperature and pressure and
it is utilized for the evolution of power in a turbine or steam engine. The water is
transformed into steam with the assistant of heat generated by burning of coal in the
boiler. Then, the exhaust steam is liquefied to water as it passes through a condenser.
The condensed steam (condensate) is driven into X 4 (GS condenser) through X 2 (CEP
1), X 3 (CEP 2) from X 1 (condenser). Then it proceeds through X 5 (drain cooler) and
eventually to X 9 (deaerator) after the elevation of its temperature in X 6 (heater 1),
X 7 (heater 2), X 8 (heater 3). The complete working of condensate system has been
discuss in the diagram given below in (Fig. 1).

4 Numerical Example

A Thermo-plant complex structure is comprised of nine components. Among these

nine components 1, 2 and 3 are associated in series manner and 4 to 9 components are
arranged in combination of series–parallel manner. The performance of the system
Assessment of Reliability Function and Signature … 263

Fig. 1 Block diagram of thermo-plant

depends upon working culture of components which is attached in system. The

system structure function is defined as max(X 1 X 2 X 3 , max(X 4 X 5 , X 6 , X 7 , X 8 ), X 9 ).
Each unit will either work or fail and 0, 1 are given as the rate of performance of
working and non-working components shown as Fig. 1.
Now, calculate the expected reliability function of the consider system shown as
in the form of u-functions such as
u j (z) = (1 − p j )z 0 + p j z 1 , 1 ≤ j ≤ 9,
The system structure function is defined as

X = ϕ(X 1 , X 2 , ...X 9 ) = max(X 1 X 2 X 3 , max(X 4 X 5 , X 6 , X 7 , X 8 ), X 9 ).

From the composition operator, get u-function using structure of the system such
as

U (z) = ⊗(u 1 (z), ..., u 9 (z)).

ϕ

The reliability function can be obtaining repeated as

X 10 = X 1 X 2 X 3

X 11 = X 4 X 5

X 12 = max(X 11 , X 6 , X 7 , X 8 )
264 M. Negi et al.

X 13 = X 10 X 12

X = max(X 9 , X 13 )

Now, calculation of the system reliability is as follows

U10 = u 1 (z) ⊗ u 2 (z) ⊗ u 3 (z)

min min
U10 (z) = [(1 − p1 )z 0 + p1 z 1 ] ⊗ [
min
(1 − p2 )z + p2 z ] ⊗ [(1 −
0 1
p3 )z 0 + p3 z 1 ]
min
U10 (z) = [(1 − p1 )(1 − p2 ) + p1 (1 − p2 ) + p2 (1 − p1 )
+ p1 p2 (1 − p3 )]z 0 + p1 p2 p3 z 1 .
U11 (z) = u 4 (z) ⊗ u 5 (z)
min
U11 (z) = [(1 − p4 )z 0 + p4 z 1 ⊗ (1 − p5 )z 0 + p5 z 1 ]
min
U11 (z) = [(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )]z 0 + p4 p5 z 1
U12 (z) = U11 (z) ⊗ u 6 (z) ⊗ u 7 (z) ⊗ u 8 (z)
max max max
U12 (z) = [(1 − p4 )((1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )]z 0
+ p4 p5 z 1 ⊗ [(1 − p6 )z 0 + p6 z 1 ] ⊗
max max
[(1 − p7 )z 0 + p7 z 1 ] ⊗ [(1 − p8 )z 0 + p8 z 1 ]
max
U12 (z) = [(1 − p4 )(1 − p5 ) + p4 (1 − p5 )
+(1 − p4 ) p5 ](1 − p6 )(1 − p7 )(1 − p8 )z 0 + {[(1 − p4 )(1 − p5 )
+ p4 (1 − p5 ) + (1 − p4 ) p5 ](1 − p6 )[ p7 (1 − p8 ) + (1 − p7 ) p8 + p7 p8
+[[(1 − p4 )(1 − p5 )
+ p4 (1 − p5 ) + (1 − p4 ) p5 ] p6 + p4 p5 ][(1 − p7 )(1 − p8 ) + [ p7 (1 − p8 )
+(1 − p7 ) p8 + p7 p8 ]]}z 1 .
Assessment of Reliability Function and Signature … 265

U13 (z) = U12 (z) ⊗ U10 (z)

min
U13 (z) = {[(1 − p4 )(1 − p5 ) + p4 (1 − p5 )
+(1 − p4 ) p5 ](1 − p6 )(1 − p7 )(1 − p8 ){[(1 − p1 )(1 − p2 )
+ p1 (1 − p2 ) + p2 (1 − p1 ) + p1 p2 (1 − p3 )] + p1 p2 p3 }
+[(1 − p4 )(1 − p5 ) + (1 − p5 ) p4
+(1 − p4 ) p5 ](1 − p4 )[(1 − p8 ) p7 + (1 − p7 ) p8 + p7 p8 ]
+[[(1 − p4 )(1 − p5 ) + p4 (1 − p5 )
+(1 − p4 ) p5 ] p6 + p4 p5 ][(1 − p7 )[(1 − p8 ) + [ p7 (1 − p8 ) + (1 − p7 ) p8 + p7 p8 ]]
[(1 − p1 )(1 − p2 ) + p1 (1 − p2 ) + p2 (1 − p1 ) + p1 p2 (1 − p3 )]}z 0
+{[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )[ p7 (1 − p8 ) + p8 (1 − p7 )
+ p7 p8 ] + [[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )] p6 + p4 p5 ][(1 − p7 )(1 − p8 )
+ p7 (1 − p8 ) + p8 (1 − p7 ) + p7 p8 ]]} p1 p2 p3 z 1 .
U (z) = U13 (z) ⊗ u 9 (z)
max

U (z) = {[(1 − p4 )(1 − p4 ) + p4 (1 − p5 )

+(1 − p4 ) p5 ]{(1 − p6 )(1 − p7 )(1 − p8 )[(1 − p1 )(1 − p2 )
+ p1 (1 − p2 ) + (1 − p1 ) p2 + (1 − p3 ) p2 p1 ] + p1 p2 p3 }
+[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )](1 − p6 )[ p7 (1 − p8 )
+(1 − p7 ) p8 + [[(1 − p4 )(1 − p5 ) + p4 (1 − p5 )
+ p5 (1 − p4 )] p6 + p4 p5 ][(1 − p7 )(1 − p8 ) + [(1 − p8 ) p7 + (1 − p7 ) p8 + p7 p8 ]]
[(1 − p1 )(1 − p2 ) + p1 (1 − p2 ) + p2 (1 − p1 ) + p1 p2 (1 − p3 )]}(1 − p9 )z 0 +
{[[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )][(1 − p6 )(1 − p7 )(1 − p8 )[(1 − p1 )(1 − p2 )
+ p1 (1 − p2 ) + p2 (1 − p1 ) + (1 − p3 ) p1 p2 + p1 p2 p3 ] + [(1 − p4 )(1 − p5 ) + p4 (1 − p5 )
+(1 − p4 ) p5 ](1 − p6 )[ p7 (1 − p8 ) + (1 − p7 ) p8 + p7 p8 ] + [[(1 − p4 )(1 − p5 ) + (1 − p5 ) p4
+(1 − p4 ) p5 ] p6 + p4 p5 ][(1 − p7 )(1 − p8 ) + [ p7 (1 − p8 ) + p8 (1 − p7 ) + p7 p8 ]]
[(1 − p1 )(1 − p2 ) + p1 (1 − p2 ) + p2 (1 − p1 ) + (1 − p3 ) p1 p2 ]] p9
+{[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + p5 (1 − p4 )](1 − p6 )[ p7 (1 − p8 ) + p8 (1 − p7 )
+ p7 p8 ] + [[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + (1 − p4 ) p5 ] p6 + p4 p5 ][(1 − p7 )(1 − p8 )
+[ p7 (1 − p8 ) + p8 (1 − p7 ) + p7 p8 ]]} p1 p2 p3 }z 1 .
(5)

Hence, reliability function of u-function, expected value can be obtained (see

[20]) as
266 M. Negi et al.

R = [[(1 − p4 )(1 − p5 ) + (1 − p5 ) p4 + (1 − p4 ) p5 ][(1 − p6 )(1 − p7 )(1 − p8 )

[(1 − p1 )(1 − p2 ) + p1 (1 − p2 ) + p2 (1 − p1 ) + p1 p2 (1 − p3 )] + p1 p2 p3 ]+
[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + (1 − p4 ) p5 ](1 − p6 )] p7 (1 − p8 ) + (1 − p7 ) p8 +
p7 p8 ] + [[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + (1 − p4 ) p5 ] p6 + p4 p5 ][(1 − p7 )(1 − p8 )+
[ p7 (1 − p8 ) + (1 − p7 ) p8 + p7 p8 ]][(1 − p1 )(1 − p2 ) + p1 (1 − p2 ) + (1 − p1 ) p2 +
(1 − p3 ) p1 p2 ]] p9 + {[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + (1 − p4 ) p5 ](1 − p6 )
[(1 − p8 ) p7 + (1 − p7 ) p8 + p7 p8 ] + [[(1 − p4 )(1 − p5 ) + p4 (1 − p5 ) + (1 − p4 )
p5 ] p6 + p4 p5 ][(1 − p7 )(1 − p8 ) + [ p7 (1 − p8 ) + p8 (1 − p7 ) + p7 p8 ]]} p1 p2 p3
(6)

If all the probabilities be i.i.d. components of each other, then structure function
of the consider system is
p1 = p2 = ... = p9 = p.
The required reliability function of the proposed system is

R = p + 3 p4 − 5 p5 + 5 p7 − 4 p8 + p9 . (7)

Calculate minimal signature from the Eq. (7) such as

Min. Signature = (1, 0, 0, 3, -5, 0, 5, -4, 1).
Hence, calculate the tail signature from reliability function using Eq. (3) as
Tail signature S = (1, 1, 66/72, 396/504, 1920/3024, 708/1512, 2016/6048,
4032/18144, 4032/36288, 0).
In case of Signature using s = S l−1 − Sl , l = 1, 2, ..., 9 and value of tail
signature, we get
s = (0, 6/72, 4752/36288, 229,824/1524096, 762,048/4572288,
1,233,792/9144576, 12,192,768/109734912, 73,156,608/658409472, 4032/36288).
1
Now, determine the Expected time with the help of Eq. (7) using E(T ) have F(t)
0
is
1

= p + 3 p 4 − 5 p 5 + 5 p 7 − 4 p 8 + p 9 dp?
0
1 (8)
= e(−t) − 3e(−4t) − 5e(−5t) + 5e(−7t) − 4e(−8t) + e(−9t) dt
0
= 1.075.

And Expected X can be obtained by using system’s signature and formula


9
E(X ) = isi , i = 1, 2, ..., 9. such as = 5.472. (9)
i=1
Assessment of Reliability Function and Signature … 267

For expected cost of the system using Eqs. (8–9) as

Expected cost rate = E(X)/E(T ) = 5.09.

5 Result and Discussion

In this work, signature used for comparison of system with respect to probability and
failure probability of working elements. System have in good state according to mean
time to failure, cost analysis and failure probability of the elements. Furthermore,
most of the components are not in substandard state and they will execute effectively.
These results are good enough for a system to perform excellently.

6 Conclusion

This chapter aims at the efficacy of various components of the thermo plant
energy system in which UGF and system structure have been used to calcu-
late the reliability function. We evaluated both signature (0, 6/72, 4752/36288,
229,824/1524096, 762,048/4572288, 1,233,792/9144576, 12,192,768/109734912,
73,156,608/658409472, 4032/36288) and tail signature (1, 1, 66/72, 396/504,
1920/3024, 708/1512, 2016/6048, 4032/18144, 4032/36288, 0) of the system which
came out to be less and hence making our model more productive. From the outcomes
of the case study we can say that system’s mean time to failure is 1.075 high but
the expected cost 5.09 of the system is moderate. If water supply are tends to fail,
the system is will to be collapse. That is the one of the major limitation of consider
system.

References

1. Andrzejczak K (2015) Stochastic modelling of the repairable system. J KONBiN 35(1):5–14

2. Boland PJ (2001) Signatures of indirect majority systems. J Appl Probab 38(2):597–603
3. Dennehy C, Lawlor PG, Gardiner GE, Jiang Y, Shalloo L, Zhan X (2017) Stochastic modelling
of the economic viability of on-farm co-digestion of pig manure and food waste in Ireland.
Appl Energy 205:1528–1537
4. Dhillon BS (1992) Stochastic modelling of k-out-of-n units family of systems. Int J Syst Sci
23(8):1277–1287
5. Dong Q, Cui L, Si S (2020) Reliability and availability analysis of stochastic degradation
systems based on bivariate Wiener processes. Appl Math Model 79:414–433
6. Eryilmaz S (2012) The number of failed elements in a coherent system with exchangeable
elements. IEEE Trans Reliab 61(1):203–207
7. Ghosh R, Trivedi KS, Naik VK, Kim DS (2010) End-to-end performability analysis for
infrastructure-as-a-service cloud: an interacting stochastic models approach. In: 2010 IEEE
16th Pacific rim international symposium on dependable computing, pp 125–132
268 M. Negi et al.

8. Goyal N, Ram M (2017) Stochastic modelling of a wind electric generating power plant. Int J
Qual Reliab Manage 34(1):103–127
9. Gupta S (2019) Stochastic modelling and availability analysis of a critical engineering system.
Int J Qual Reliab Manage 36(5):782–796
10. Huang GH, Loucks DP (2000) An inexact two-stage stochastic programming model for water
resources management under uncertainty. Civil Eng Syst 17(2):95–118
11. Komal S (2019) Fuzzy reliability analysis of the compressor house unit (CHU) system in a coal
fired thermal power plant using TBGFLT technique. Int J Qual Reliab Manage 36(5):686–707
12. Kumar A, Ram M (2019) Computation interval-valued reliability of sliding window system.
Int J Math, Eng Manage Sci 4(1):108–115
13. Kumar A, Singh SB (2017) Computations of the signature reliability of the coherent system.
Int J Qual Reliab Manage 34(6):785–797
14. Kumar A, Singh SB (2018) Signature reliability of linear multi-state sliding window system.
Int J Qual Reliab Manage 35(10):2403–2413
15. Kumar A, Singh SB (2019) Signature A-within-B-from-D/G sliding window system. Int J Math,
Eng Manage Sci 4(1):95–107
16. Kumar A, Varshney A, Ram M (2015) Sensitivity analysis for casting process under stochastic
modelling. Int J Ind Eng Comput 6(3):419–432
17. Kumar P, Singh LK, Chaudhari N, Kumar C (2020) Availability analysis of safety-critical and
control systems of NPP using stochastic modeling. Ann Nuclear Energy 147:107657
18. Levitin G (2001) Redundancy optimization for multi-state system with fixed resource-
requirements and unreliable sources. IEEE Trans Reliab 50(1):52–59
19. Levitin G (2002) Optimal allocation of elements in a linear multi-state sliding window system.
Reliab Eng Syst Saf 76(3):245–254
20. Levitin G (2005) The universal generating function in reliability analysis and optimization, p
442. Springer, London. https://doi.org/10.1007/1-84628-245-4
21. Machida F, Xiang J, Tadano K, Maeno Y (2013) Composing hierarchical stochastic model
from Sys ML for system availability analysis. In: 2013 IEEE 24th international symposium on
software reliability engineering (ISSRE), pp 51–60
22. Malik SC, Munday VJ (2014) Stochastic modelling of a computer system with hardware
redundancy. Int J Comput Appl 89(7):26–30
23. Manatos A, Koutras VP, Platis AN (2016) Dependability and performance stochastic modelling
of a two-unit repairable production system with preventive maintenance. Int J Prod Res
54(21):6395–6415
24. Marichal JL, Mathonet P (2013) Computing system signatures through reliability functions.
Stat Probab Lett 83(3):710–717
25. Michael S, Mariappan V, Kamat V (2011) Stochastic modelling of failure interaction: Markov
model versus discrete event simulation. Int J Adv Oper Manage 3(1):1–18
26. Navarro J, Rubio R (2009) Computations of signatures of coherent systems with five
components. Commun Stat-Simul Comput 39(1):68–84
27. Navarro J, Rychlik T (2010) Comparisons and bounds for expected lifetimes of reliability
systems. Eur J Oper Res 207(1):309–317
28. Navarro J, Ruiz JM, Sandoval CJ (2007) Properties of coherent systems with dependent
components. Commun Stat—Theory Meth 36(1):175–191
29. Navarro J, Rychlik T, Shaked M (2007) Are the order statistics ordered? A survey of recent
results. Commun Stat—Theory Meth 36(7):1273–1290
30. Osaki S, Nakagawa T (1976) Bibliography for reliability and availability of stochastic systems.
IEEE Trans Reliab 25(4):284–287
31. Ram M, Goyal N (2016) Automated teller machine network inspection under stochastic
modelling. J Eng Sci Technol Rev 9(5):1–8
32. Rushdi AMA, Ghaleb FAM (2021) Reliability characterization of binary-imaged multi-state
coherent threshold systems. Int J Math, Eng Manage Sci 6(1):309–321
33. Sahner RA, Trivedi KS (1993) A software tool for learning about stochastic models. IEEE
Trans Educ 36(1):56–61
Assessment of Reliability Function and Signature … 269

34. Samaniego FJ (2007) System signatures and their applications in engineering reliability, vol
110. Springer Science & Business Media. ISBN 978-0-387-71796-8
35. Samaniego FJ, Balakrishnan N, Navarro J (2009) Dynamic signatures and their use in
comparing the reliability of new and used systems. Naval Res Logistics (NRL) 56(6):577–591
36. Triantafyllou IS (2021) On the lifetime and signature of constrained (k, d)-out-of-n: F reliability
systems. Int J Math, Eng Manage Sci 6(1):66–78
37. Weng Q (2002) Land use change analysis in the Zhujiang Delta of China using satellite remote
sensing, GIS and stochastic modelling. J Environ Manage 64(3):273–284
38. Zacharof AI, Butler AP (2004) Stochastic modelling of landfill leachate and biogas production
incorporating waste heterogeneity. Model formulation and uncertainty analysis. Waste Manage
24(5):453–462
Reliability Evaluation and Cost
Optimization of Solar Air-Conditioner

Ashok Singh Bhandari, Mangey Ram, Akshay Kumar,

and Sushil Chandra Dimri

Abstract As we all know synergy files are the channel for aspiring engineers
and technologists that are striving for a better and more sustainable world. During
summers when the temperature can soar to mid 40 °C in many places around the
world, air conditioning becomes more than just a luxury however it is costly to cool
spaces particularly in areas with a high level of humidity. One of the advantages of
cooling instead of heating is that cooling is required more when there is more heat
energy around to tap into or in other words there is more energy available. To be
more specific we have more solar energy available to us. Here we are interested in
the technology that uses heat from the sun to provide energy directly in the thermo-
dynamic circle of an air-conditioner. This research work is dedicated to evaluate the
reliability measures of solar air-conditioners which include availability, mean time to
failure (MTTF), and sensitivity analysis with their graphical representation by using
the Markov process. Along with reliability assessment, Particle Swarm Optimization
(PSO) technique is applied with the objective to find the minimum cost of the system
while taking maximum reliability as a constraint.

Keywords Solar air- conditioners · Solar energy · Markov process · Reliability ·

Availability · Mean time to failure · Sensitivity analysis · Particle swarm
optimization

A. S. Bhandari (B) · A. Kumar

Department of Mathematics, Graphic Era Hill University, Uttarakhand, India
M. Ram
Department of Mathematics, Computer Science and Engineering, Graphic Era Deemed To Be
University, Uttarakhand, India
Institute of Advanced Manufacturing Technologies, Peter the Great St. Petersburg Polytechnic
University, 195251 Saint Petersburg, Russia
S. C. Dimri
Department of Computer Science and Engineering, Graphic Era Deemed To Be University,
Uttarakhand, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 271
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_12
272 A. S. Bhandari et al.

1 Introduction

The need for air conditioning has been increased rapidly in recent times due to global
warming and climate change. The simultaneous processing of temperature, humidity,
purification, and distribution of air current in compliance with the requirement of
space needing air conditioning is defined as air conditioning [18]. In general, air
conditioning is a system for controlling the humidity, ventilation, and temperature in
a building or vehicle, typically to maintain a cool atmosphere in warm conditions. All
these types of processors need a continuous energy supply to operate. However, air
conditioning would commonly take up half of building electricity consumption [19].
Limited availability and environmental concerns of conventional sources of energy
such as coal, petroleum, natural gas, etc. are some serious challenges in the twenty-
first century. Some renewable sources of energy like hydroelectric power plants and
wind turbines are not that feasible to set up as per requirement. Critical worries of the
present energy area comprise of issues such as nonstop expansion in energy interest,
quick exhaustion of ordinary energy assets, and the impact on the climate [2, 3, 17,
20, 26–29]. By 2050 the interest in energy gracefully could double or even triple as
the worldwide population develops and developing nations extend their economies.
This has just raised worries over potential power supply difficulties, exhaustion of
fuel sources, and assisting ecological effects.
The primary need for solar air conditioners is to reduce the use of electricity and
use a renewable clean source of energy more and more [7]. Though, expanding carbon
impression because of the developing utilization of air conditioners had expanded
interest in sunlight-based cooling. The utilization of solar air conditioners is broadly
found in business, private, and modern arrangements. Developing endeavors to spare
power and expanding ecological concerns has additionally enlarged interest in solar
air conditioners. Easy maintenance and cost adequacy are some other benefits of solar
air conditioners. After the installation of solar air conditioners at home, electricity
consumption can drop by 50% or more. The solar panels at the rooftop do not require
much care and maintenance, sometimes they might need some repair work which is
far less than the traditional air-cooling systems.
Solar air conditioners have the great advantage of the excess of solar energy, as
air conditioning is mostly used when the temperature is high outside. Solar panels
work best with shining bright daylight and can also be used with lower sunlight, as
they are equipped with batteries. These batteries can last for 12–24 h and can be
recharged by sunlight. Therefore, it can also be money efficient to use solar AC.
A Markov model consisting state transition diagram of a solar-powered air condi-
tioner has been designed to study reliability measures, availability, MTTF, cost, and
sensitivity analysis. In this regard, Lameiro and Duff [14] presented a model approach
to the generalized solar energy space heating performance analysis problem. The
authors developed a stochastic model for the performance analysis of solar energy
space heating systems and implemented that model into a working FORTRAN
computer program. Gangloff [5] had described intractable common mode failures,
which were not easy to find in the regular Markov block diagram approach. To
Reliability Evaluation and Cost Optimization … 273

evaluate the common mode of failure, the author extended the fault tree analysis
techniques. In the domain of failure and repair of a system, several authors [8, 9,
12] (Cui and Li 2007) calculated the reliability of complex systems having different
modes of failure and repair techniques. Tushar et al. [25] introduced a model that
can capture the intertemporal dependency of solar irradiance using Markov chain
segmentation. Various state transitions were created for different intervals of time
for the used model to generate the solar states during daytime. By using a numerical
example, the interdependency of solar states was established with proof that solar
state transition from state to state is time-dependent. They calculated a probability
transition matrix by using real solar data sets and discussed the improvement. Ram
[16] used the copula technique to calculate reliability measures of three state systems,
and the findings of the work were significantly improved. Singh et al. [23] analyzed
the cost for a machine that consisted of two subsystems. Along with cost analysis
authors discussed various measures of reliability for the used system. The authors
concluded the relationship between availability and time for a complex repairable
system. Ram and Singh [21] improved the availability. MTTF and expected profit for
a complex system. The authors implemented the Gumbel-Hougaard family copula
technique in their work to conclude that in the long-run availability of the system
becomes constant and the expected profit decreases rapidly.
Particle swarm optimization (PSO) is a populace-based stochastic improvement
strategy created by [13]. It shows some developmental calculation properties: 1.
It is introduced with a populace of arbitrary arrangements that are called random
solutions. 2. It looks for optima by updating generations. 3. Updating generations
depends on past generations. In PSO, the possible arrangements, called particles, fly
through the issue space by following the current ideal particles [10].
The updates of the particles are cultivated by the accompanying conditions. Condi-
tion (A) ascertains new velocity for every particle dependent on its past velocity, the
particle’s area at which the best fitness (pid ) has been accomplished up until now,
and the best particle among the neighbors at which the best fitness (pgd ) has been
accomplished up until now. Condition (B) updates every particle’s position in the
solution hyperspace.
 
v(i+1)d = wI vid + c1rand(v)(pid − xid ) + c2Rand(v) pgd − xid

x(i+1)d = xid + v(i+1)d

where d is the dimension, positive constants c1 and c2 are learning factors, rand(v)
and Rand(v) are random functions, and the inertia weight ‘w’ has provided improved
performance in several applications. So,

NewV = (ParticleInertia) + (CognitiveTerm) + (SocialTerm)

Kennedy and Eberhart [13] introduced a methodology for nonlinear optimization,

named particle swarm optimization. This metaheuristic technique evolved through
274 A. S. Bhandari et al.

the simulation of a basic social model that describes connections between PSO and
genetic algorithms. The authors concluded PSO is a very simple and strong technique
for the optimization of several types of functions. Shi and Eberhart [22] experimen-
tally studied the efficiency of PSO. The authors showed the high rate of conver-
gence toward the optimal solution in PSO which makes the technique dependable.
The authors also suggested a new technique to make advancements in PSO. Some
authors [1, 4] also discussed the comparison of genetic algorithm and PSO tech-
niques and advancement in PSO as well. Liu et al. [15] introduced a hybrid particle
swarm optimization technique as up-gradation of PSO with the help of chaos, which
is a type of nonlinear system’s property. The authors concluded that the hybrid
PSO performs exploration as well as exploitation by implementing the chaotic local
searching performance. Jiang et al. [11] discussed an improvement in PSO by parti-
tioning the swarms into several sub-swarms, where each sub-swarm performs PSO
and after a period the whole population is shuffled, and new sub-swarm are created to
perform PSO again. This technique remarkably improved the ability to explore and
exploit the swarms. The authors concluded that the improved PSO greatly benefits
calculation accuracy and effective global optimization.

2 Working of Solar Air-Conditioner

Solar air conditioners can be distinguished in many types by their arrangement of

components and working procedure.
The most popular type of solar air conditioner consists of the panels placed on
the top of the building. The panels heat the water flowing in pipes and turn it into a
gas state, which takes out the ventilating air’s hotness and cools it down. Then the
cool air is flown to the building. The produced air is very cool and dry, therefore, the
efficiency of cooling down the building becomes very high.
The second type of solar air conditioner works on photovoltaic solar energy. In
these types of systems, photovoltaic panels are placed around the building to generate
electricity. The produced electricity runs the conventional air conditioner for cooling.
Extra or returned electricity is stored on a grid with inverters or by batteries. But the
batteries can lose about 20% of power due to the chemical inefficiency of batteries.
The third type of solar air conditioner works on a completely different procedure.
In the 1880s, ammonia and water were mixed in sealed steel chambers to create a
cooling effect in the water. This chilled water was pumped into the air handler to
absorb the heat from the space, where cooling is required. With the limited capacity
of space, there rises the need of separating ammonia from water to operate again,
otherwise, unlimited availability of space is needed. To recycle the chemical, the
mixture is heated by 250F hot water, which separates ammonia from the water. Here
evacuated tube solar panels come into the role of heating the water to 250 + F, to
separate the mixture of ammonia and water. After separation, both can be reused for
the chilling process.
Reliability Evaluation and Cost Optimization … 275

Fig. 1 Block diagram of a solar air conditioner

There are no destructive impacts or side-effects of solar air conditioner. They use
water as a refrigerant rather than some other harmful coolants. Additionally, on the
grounds that they are solar powered, they are autonomous of the poer grid and ration
energy by dispensing heat and friction created by the wires of a conventional air
conditioner.
Solar air conditioners come with no harmful effects and zero waste products, as
instead of any harmful coolant, they use water as a refrigerant. These solar air systems
work best in radiant, dry atmospheres, for example, the South-western United States
or Northern Africa. As these systems can also work with battery power, they can
perform well in rainy, cloudy weather. Following Fig. 1 describes the block diagram
of solar air conditioner.

3 Assumptions and Notations

The proposed system is structured by making the following assumptions:

(i) The system is assumed to be in satisfactory working condition in the primary
state.
(ii) Degraded and failed states are repairable with different rates.
(iii) After fix, the unit is treated all around great.
(iv) Constant disappointment and fix rates are accepted.
The notations used for the solar air conditioner are described in Table 1.

4 Mathematical Modelling

4.1 Formulation of the Model

The accompanying differential equations have been drawn from the state transition
diagram (Fig. 2) of the solar air conditioner.
276 A. S. Bhandari et al.

Table 1 Notations
t Time
s Variable for Laplace transformation
x Supplementary variable
λ1 /λ2 /λ3 /λ4 Hazard rate of unit inverter/ battery/ solar panel/ charge controller
μ Repair rate of the system from the degraded state P1 (t) to P0 (t)
P0 (t) Probability of initial (good) state
P1 (t) Probability of the degraded state when the system has one broke down unit of solar
panel
P2 (t) Completely failed state probability caused by battery failure
P3 (t) Completely failed state probability caused by inverter failure
P4 (t) Completely failed state probability caused by failure of both solar panels
P5 (t) Completely failed state probability caused by failure of charge controller
φ(x) Rate of repairing all the failed states

Fig. 2 State transition diagram

Reliability Evaluation and Cost Optimization … 277

 
d
+ λ1 + λ2 + 2λ3 + λ4 P0 (t) = μ(P1 (t))
dt
 ∞
+ φ(x)[P2 (x, t) + P3 (x, t) + P4 (x, t) + P5 (x, t)]d x
0
(1)
 
d
+ λ2 + λ3 + λ4 + μ P1 (t) = 2λ3 t P0 (t) (2)
dt
 
∂ ∂
+ + φ(x) P2 (x, t) = 0 (3)
∂x ∂t
 
∂ ∂
+ + φ(x) P3 (x, t) = 0 (4)
∂x ∂t
 
∂ ∂
+ + φ(x) P4 (x, t) = 0 (5)
∂x ∂t
 
∂ ∂
+ + φ(x) P5 (x, t) = 0 (6)
∂x ∂t

Boundary conditions

P2 (0, t) = λ2 [P0 (t) + P1 (t)] (7)

P3 (0, t) = λ1 [P0 (t) + P1 (t)] (8)

P4 (0, t) = λ3 P1 (t) (9)

P5 (0, t) = λ3 [P0 (t) + P1 (t)] (10)

Initial Conditions

P0 (0) = 1 (11)

and all other state probabilities are zero at t = 0.

4.2 Solution of the Model

By taking the Laplace transformation from Eqs. (1) to (10) and then using Eq. (11),
the solution obtained for the model is given below:
278 A. S. Bhandari et al.

 
[s + λ1 + λ2 + 2λ3 + λ4 ]P0 (s) = 1 + μ P1 (s)
 ∞
 
+ φ(x) P2 (x, s) + P3 (x, s) + P4 (x, s) + P5 (x, s) d x
0
(12)

[s + λ2 + λ3 + λ4 + μ]P1 (s) = 2λ3 P0 (s) (13)

 
∂
+ s + φ(x) P2 (x, s) = 0 (14)
∂x
 
∂
+ s + φ(x) P3 (x, s) = 0 (15)
∂x
 
∂
+ s + φ(x) P4 (x, s) = 0 (16)
∂x
 
∂
+ s + φ(x) P5 (x, s) = 0 (17)
∂x

Rewriting (14), (15), (16) and (17) as

 
∂
+ s + φ(x) Pi (x, s) = 0 (18)
∂x

For i = 2, 3, 4, 5.
Boundary condition
 
P2 (0, s) = λ2 P0 (t) + P1 (s) (19)

 
P3 (0, s) = λ1 P0 (t) + P1 (s) (20)

 
P4 (0, s) = λ3 P1 (s) (21)

P5 (0, s) = λ4 [[P0 (t) + P1 (s)] (22)

1
P0 (s) = (23)
D(s)

where

2λ3
D(s) = (s + +λ1 + λ2 + 2λ3 + λ4 ) − μ
s + λ2 + λ3 + λ4 + μ
Reliability Evaluation and Cost Optimization … 279
 
λ 1 + λ2 + λ3 + λ4
− + λ1 + λ2 + λ4 sφ (s) (24)
s + λ1 + λ2 + λ3 + μ

From (13)

2λ3
P1 (s) = P0 (s) (25)
s + λ2 + λ3 + λ4 + μ

From (14) and (25)

1 − sφ (s)  
P2 (s) = λ2 P0 (s) + P1 (s) (26)
s
1 − sφ (s)
P3 (s) = λ3 P1 (s) (27)
s
1 − sφ (s)
P4 (s) = λ3 P1 (s) (28)
s
1 − sφ (s)  
P5 (s) = λ4 P0 (s) + P1 (s) (29)
s

Pup (s) = P0 (s) + P1 (s) (30)

Pdown (s) = P2 (s) + P3 (s) + P4 (s) + P5 (s) (31)

It is noticed that
1
Pup (s) + Pdown (s) = (32)
s

4.3 Availability

Availability of a system is a characteristic of a system that explains the working

condition of the system at a particular time. Availability is the sum of probabilities
of the working conditions for any system. Authors have calculated the availability
of solar air conditioner by substituting the failure and repair rates as λ1 = 0.8,
λ2 = 0.03, λ3 = 0.05,λ4 = 0.09, μ = 0.02 and φ(x) = 0.05 [6].

Pup (t) = 0.1463142691e−1.207641122t cos(0.1282345336t)

+ 0.1825319221e−1.207641122t sin(0.1282345336 ∗ t) + 0.8536857307e−0.5471775631t
280 A. S. Bhandari et al.

Table 2 Availability of solar

Time (t) Pup (t)
air conditioner
0 0.9999999998
1 0.8585806188
2 0.7819738439
3 0.7298988710
4 0.6876053199
5 0.6498918490
6 0.6149406606
7 0.5820896673
8 0.5510623139
9 0.5217098429
10 0.4939271627

1.0

0.9

0.8
Availability

0.7

0.6

0.5

0 1 2 3 4 5 6 7 8 9 10 11 12
2 13 14 1
15
5
Time (t)

Fig. 3 Availability of solar air conditioner

Table 2 shows the availability of solar air conditioner at different times and
following Fig. 3 is the demonstration of the table.

4.4 Reliability Analysis

The reliability of the solar air conditioner’s is one of the basic quality attributes that
manages the conduct of every component and can be characterized as the likelihood
Reliability Evaluation and Cost Optimization … 281

Table 3 Reliability of solar

Time(t) Pup (t)
air conditioner
0 1.000000000
1 0.8199309868
2 0.6741644717
3 0.5557434437
4 0.4592117352
5 0.3802722923
6 0.3155266205
7 0.2622755327
8 0.2183662793
9 0.1820748625
10 0.1520151112

of the solar air conditioner to perform adequately for a specified time stretch in a
determined climate. With the unavailability of a repair facility i.e., μ = 0, φ(x) = 0
and λ1 = 0.8, λ2 = 0.03, λ3 = 0.05, λ4 = 0.09 [6], the reliability

− 1+ 0.10
s+0.17
R(s)=
s + 0.30

Now, by taking the Laplace inverse authors have calculated the reliability of the
model in terms of time (t). Which is

R(t) = e(−0.235t) ((cosh(0.65t) + (0.538sinh(0.65)t))

Above Table 3 represents the reliability of the model for the time t = 0 to t = 10
and Fig. 4 is the graphical representation of the reliability of solar air conditioner
with respect to time.

4.5 Mean Time to Failure

MTTF represents the average time between the failure of components of the system.
Following Table 4 and Fig. 5 describes the MTTF of the solar air conditioner with
respect to the variation of failure.
282 A. S. Bhandari et al.

1.0

0.8
Reliability

0.6

0.4

0.2

0.0
0 2 4 6 8 10
Time(t)

Fig. 4 Reliability of solar air conditioner

Table 4 MTTF of solar air conditioner

Variation in λi MTTF with respect to failure rates
λ1 λ2 λ3 λ4
0.1 1.714285714 1.464404814 2.003284072 2.032085562
0.2 1.481481481 1.442170916 1.690912924 1.884781885
0.3 1.303571429 1.391527600 1.464404814 1.730909091
0.4 1.163101604 1.330581356 1.291470434 1.589526093
0.5 1.049382716 1.267454350 1.154719937 1.464404814
0.6 0.9554655868 1.205844801 1.043740573 1.354861355
0.7 0.8766233765 1.147376450 0.9518375888 1.259037887
0.8 0.8095238095 1.092666091 0.8744797033 1.174949495
0.9 0.7517482518 1.041835358 0.8084774818 1.100803958

4.6 MTTF Sensitivity

This characteristic is calculated by partial differentiation of MTTF with respect to

the input variables and then varying the input by 0.1, 0.2, …, 0.9 respectively and
fixing the failure rates as λ1 = 0.8, λ2 = 0.03, λ3 = 0.05, λ4 = 0.09 [6] in partial
derivatives, one may get Table 5 and demonstrated graphically in Fig. 6.
Reliability Evaluation and Cost Optimization … 283

Fig. 5 Mean time to failure of solar air conditioner

Table 5 MTTF sensitivity of solar air conditioner

∂ MT T F ∂ MT T F ∂ MT T F ∂ MT T F
Variation in λi ∂λ1 ∂λ2 ∂λ3 ∂λ4

0.1 −15.51011029 −2.785142478 1.383559623 4.858441875

0.2 −9.003601439 −1.640794726 −0.80 7,076,446 3.066658569
0.3 −5.873651235 −1.137909405 −9.182736449 2.046115319
0.4 −4.131725531 −0.8601343456 −1.200072965 1.440297957
0.5 −3.063725490 −0.6846626419 −1.092408624 1.056832959
0.6 −2.362039402 −0.5638594440 −0.9682831634 0.8009739836
0.7 −1.876459468 −0.4756696512 −0.8521394955 0.6229670216
0.8 −1.526562182 −0.4085276152 −0.7501425268 0.4948867964
0.9 −1.266131453 −0.3557837044 −0.6625703911 0.4001512852

4.7 Reliability Sensitivity

Sensitivity in the reliability of the solar air conditioner can be analyzed by taking
partial derivative of the reliability function with respect to their input parameters, after
taking the inverse Laplace transformation. By fixing the values of input parameters as
λ1 = 0.8, λ2 = 0.03, λ3 = 0.05,λ4 = 0.09 [6] in the partial derivatives of reliability
284 A. S. Bhandari et al.

Fig. 6 MTTF sensitivity of solar air conditioner

expression. Table 6 contains the numerical values of the reliability sensitivity which
are demonstrated graphically in Fig. 7.

Table 6 Reliability sensitivity of solar air conditioner

∂ R(t) ∂ R(t) ∂ R(t) ∂ R(t)
Time (t) ∂λ1 ∂λ2 ∂λ3 ∂λ4

0 0 0 0 0
1 −0.3850328887 −0.4174266903 0.3341754247 −0.4174266903
2 −0.3099803012 −0.3969378011 0.6618856904 −0.3969378011
3 −0.2007384159 −0.3360541034 0.7658142659 −0.3360541034
4 −0.1274532735 −0.2980814046 0.7267230318 −0.2980814046
5 −0.08521134701 −0.2783179595 0.6278076002 −0.2783179595
6 −0.06124393423 −0.2661764893 0.5158654217 −0.2661764893
7 −0.04689343589 −0.2554315687 0.4116244354 −0.2554315687
8 −0.03750226945 −0.2435809364 0.3221518873 −0.2435809364
9 −0.03077464807 −0.2300916402 0.2483868571 −0.2300916402
10 −0.02560774555 −0.2152497664 0.1888986146 −0.2152497664
Reliability Evaluation and Cost Optimization … 285

Fig. 7 Reliability sensitivity of solar air conditioner

5 Optimization of Cost of a Solar Air Conditioner Using

PSO

The aim of using PSO is to get minimized the cost of the system with the required
reliability of the whole system as well as the components. The following nonlinear
programming problem is solved to get the results:
Minimize

C = K 1 R1 α1 + K 2 R2 α2 + K 3 R3 α3 + 2K 4 R4 α4 (33)

Subject to the constraints

 
R2 R3 R4 2R1 − R1 2 < 1

 
−[R 2 R3 R4 2R1 − R1 2 ] ≤ 0.3

Ri < 1fori = 1, 2, 3, 4 and

−Ri < 0fori = 1, 2, 3, 4

286 A. S. Bhandari et al.

Fig. 8 Cost convergence curve of solar air conditioner

The following values are assigned to the constants K i and αi for i = 1, 2, 3, 4,

K 1 ≡ 200, K 2 ≡ 100, K 3 ≡ 100, K 4 ≡ 150 and αi = 0.6 for i = 1,2, 3, 4 [24].
Here C represents the total cost of the system and Rs = R 2 R3 R4 2R1 − R1 2 is
the reliability of the solar air conditioner. R1 , R2 , R3 , and R4 are the reliabilities
of solar panels, battery, charge controller, and inverter, respectively. The optimal
solution was achieved by using the Particle Swarm Optimization algorithm where
100 random particles were initiated in the solution space. The inertia weight w =
0.9, the acceleration constants c1 = 2 and c2 = 2.05 are used and no. of iterations
= 1000.
The optimized cost for the above nonlinear problem is 365.0355 with system
reliability above 0.3 and R1 = 0.2278, R2 = 0.9088, R3 = 0.9020, and R4 =
0.9065. The following Fig. 8 is PSO convergence curve which converges to the
optimized cost of the system.

6 Results Analysis

In this research work, the authors have studied the numerous reliability characteristics
such as obtainability, dependability, MTTF, and sensitivity of reliability and MTTF of
the system under the consideration of four types of failures and employing the Markov
process. Through the general investigation on the solar air conditioning system,
Reliability Evaluation and Cost Optimization … 287

the authors mentioned the accompanying observable facts. Figure 3 addressing the
diagram of the availability of the solar air conditioner versus time. Initially, the
availability of the system decreases smoothly with increase in time but later, the value
of the availability of the system attains a constant value with the increase in time. The
general pattern shows that the availability of the system decreases with the increase
in time, and maintain a constant value (after long time) of 0.5 approximately for the
defined values of the failure rates. Figure 4 communicates the pattern of reliability
of the solar air conditioner with respect to time. From the graph, the reliability of
the solar air conditioner also decreases with time and tends to zero as time increases.
The pattern of MTTF of the solar air conditioner is illustrated in Fig. 5. Here L1,
L2, L3, and L4 represents the failure rates λ1 , λ2 , λ3 and λ4 respectively. Figures 6
and 7 are behavioral explanations of MTTF sensitivity and reliability sensitivity of
the solar air conditioner with respect to time. MTTF sensitivity is increasing while
there is variation in failure rates λ1 and λ2 , decreasing with variation in λ4 and
rapidly increasing after a quick fall while varying the failure rate λ3 . From Fig. 6, the
sensitivity in the reliability of the solar air conditioner is observable, here also L1,
L2, L3, and L4 represents the failure rates λ1 , λ2 , λ3 and λ4 respectively. Figure 8 is
the PSO convergence curve for the cost of solar air conditioner which converges to
365.0355.

7 Conclusion

The current work proposes the Markov model of solar air conditioner. Overall assess-
ment presumes that with the standard upkeep of solar air conditioners, availability
remains steady after some time. It is beneficial to comment here that the solar air
conditioner is generally delicate in worry of component failure. In this way, it is
important to deal with component failure rate, to achieve an exceptionally dependable
solar air conditioner. The optimization technique is capable to determine the number
of redundant components and their location for higher reliability and lower cost. The
consequences of this examination are a lot useful for technologists, engineers, and
plant managers.

References

1. Angeline PJ (1998) Using selection to improve particle swarm optimization. In: IEEE
international conference on evolutionary computation. Anchorage, Alaska, May 4–9, 1998
2. Chai B, Yang Z (2014) Impacts of unreliable communication and modified regret matching
based anti-jamming approach in smart microgrid. Ad Hoc Netw 22:69–82
3. Chai B, Chen J, Yang Z, Zhang Y (2014) Demand response management with multiple utility
companies: a two-level game approach. IEEE Trans Smart Grid 5(2):722–731
4. Eberhart RC, Shi YH (1998) Comparison between genetic algorithms and particle swarm
optimization. In: 1998 annual conference on evolutionary programming. San Diego
288 A. S. Bhandari et al.

5. Gangloff WC (1975) Common mode failure analysis. IEEE Trans Power Apparatus Syst
94(1):27–30
6. Goyal N, Ram M, Kaushik A (2017) Performability of solar thermal power plant under
reliability characteristics. Int J Syst Assurance Eng Manage 8(2):479–487
7. Gugulothu R, Somanchi NS, Banoth HB, Banothu K (2015) A review on solar powered air
conditioning system. Procedia Earth Planetary Sci 11:361–367
8. Gupta PP, Agarwal SC (1984) A parallel redundant complex system with two types of failure
under preemptive-repeat repair discipline. Microelectron Reliab 24(3):395–399
9. Gupta PP, Sharma MK (1993) Reliability and MTTF evaluation of a two duplex-unit standby
system with two types of repair. Microelectron Reliab 33(3):291–295
10. Hu X, Eberhart RC, Shi Y (2003) Engineering optimization with particle swarm. In: Proceed-
ings of the 2003 IEEE swarm intelligence symposium. SIS’03 (Cat. No. 03EX706). IEEE, pp
53–57
11. Jiang Y, Hu T, Huang C, Wu X (2007) An improved particle swarm optimization algorithm.
Appl Math Comput 193(1):231–239
12. Ke J-B, Lee W-C, Wang K-H (2007) Reliability and sensitivity analysis of a system with
multiple unreliable service stations and standby switching failures. Phys A 380:455–469
13. Kennedy J, Eberhart R (1995) Particle swarm optimization. In: Proceedings of ICNN’95-
international conference on neural networks, vol 4. IEEE, pp 1942–1948
14. Lameiro GF, Duff WS (1979) A Markov model of solar energy space and hot water heating
systems. Sol Energy 22(3):211–219
15. Liu B, Wang L, Jin YH, Tang F, Huang DX (2005) Improved particle swarm optimization
combined with chaos. Chaos, Solitons Fractals 25(5):1261–1271
16. Ram M (2010) Reliability measures of a three-state complex system: a copula approach. Appl
Appl Math: An Int J 5(10):1483–1492
17. Hassan NU, Pasha MA, Yuen C, Huang S, Wang X (2013) Impact of scheduling flexibility
on demand profile flatness and user inconvenience in residential smart grid system. Energies
6(12):6608–6635
18. Ochi M, Ohsumi K (1989) Fundamental of refrigeration and air conditioning. Ochi Engineering
Consultant Office
19. Pachauri RK, Reisinger A (2007) 2007. IPCC fourth assessment report. IPCC, Geneva
20. Khan RH, Brown J, Khan JY (2013) Pilot protection schemes over a multi-service WiMAX
network in the smart grid. In: 2013 IEEE international conference on communications
workshops (ICC). IEEE, pp 994–999
21. Ram M, Singh SB (2010) Availability, MTTF and cost analysis of complex system under
preemptive-repeat repair discipline using Gumbel-Hougaard family copula. Int J Qual Reliab
Manage
22. Shi Y, Eberhart RC (1999) Empirical study of particle swarm optimization. In: Proceedings of
the 1999 congress on evolutionary computation-CEC99 (Cat. No. 99TH8406), vol 3. IEEE, pp
1945–1950
23. Singh VV, Ram M, Rawal DK (2013) Cost analysis of an engineering system involving
subsystems in series configuration. IEEE Trans Autom Sci Eng 10(4):1124–1130
24. Tillman FA, Hwang CL, Fan LT, Lai KC (1970) Optimal reliability of a complex system. IEEE
Trans Reliab 19(3):95–100
25. Tushar W, Huang S, Yuen C, Zhang JA, Smith DB (2014) Synthetic generation of solar states
for smart grid: A multiple segment Markov chain approach. In: IEEE PES innovative smart
grid technologies. IEEE, Europe, pp 1–6
26. Tushar W, Chai B, Yuen C, Smith DB, Wood KL, Yang Z, Poor HV (2014) Three-party energy
management with distributed energy resources in smart grid. IEEE Trans Industr Electron
62(4):2487–2498
27. Tushar W, Yuen C, Chai B, Smith DB, Poor HV (2014) Feasibility of using discriminate pricing
schemes for energy trading in smart grid. In: 2014 IEEE global communications conference.
IEEE, pp 3138–3144
Reliability Evaluation and Cost Optimization … 289

28. Khalid YI, Hassan NU, Yuen C, Huang S (2014) Demand response management for power throt-
tling air conditioning loads in residential smart grids. In: 2014 IEEE international conference
on smart grid communications (SmartGridComm). IEEE, pp 650–655
29. Liu Y, Hassan NU, Huang S, Yuen C (2013) Electricity cost minimization for a residential
smart grid with distributed generation and bidirectional power transactions. In: 2013 IEEE
PES innovative smart grid technologies conference (ISGT). IEEE, pp 1–6
Analyzing Interrelationships Among
Software Vulnerabilities Using Fuzzy
DEMATEL Approach

Misbah Anjum, P. K. Kapur, Vernika Agarwal, and Vivek Kumar

Abstract The increasing data infringement is pressuring software organizations to

create and maintain secure software. Although, not all assaults can be anticipated or
prevented, many can be avoided by eliminating vulnerabilities in software. Security
teams must discover a mechanism to understand the interdependence of vulnerabil-
ities after their discovery to identify which security vulnerability pose the greatest
risk and demand immediate attention. The limited revenue and the time taken by
the software testers put additional pressure to select those with high severity and
are being a cause for other severe flaws. Addressing such defects is the main step
towards understanding the cause-effect relationship among the vulnerabilities. The
present objective is to propose a framework for assessing these interrelationships of
various software vulnerabilities by utilizing Fuzzy Decision-making Trial and Eval-
uation Laboratory (F-DEMATEL). The fuzzy theory assists in the decision mapping,
whereas the DEMATEL technique is utilized to illustrate the contextual link between
the kinds of vulnerability. The data is validated for software testing company placed
in northern India.

Keywords Vulnerabilities · Multi-Criteria Decision Making (MCDM) ·

Prioritization · Fuzzy DEMATEL

M. Anjum
Amity Institute of Information Technology, Amity University, Noida, Uttar-Pradesh, India
P. K. Kapur
Amity Center for Inter-Disciplinary Research, Amity University, Noida, Uttar-Pradesh, India
V. Agarwal (B)
Amity International Business School, Amity University, Noida, Uttar-Pradesh, India
V. Kumar
Department of Operational Research, University of Delhi, Delhi, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 291
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_13
292 M. Anjum et al.

1 Introduction

The technological innovation has grown across the globe, with the growing popularity
of electronic equipments, elevated sensors and other kinds of networking technolo-
gies are generally included in the Internet of Things (IoT) umbrella [1]. A statistical
forecast shows that the penetration of these intelligent devices in global infrastruc-
ture is anticipated to rise from 26 billion devices to over 75 billion by 2025 [2]. This
rapid growth of the use of networked intelligent devices leads to a growing danger of
cybersecurity [3]. Computer users may have been the only individuals to care about
vulnerabilities in the past [4], with the changing communication dynamics, anyone
using smartphones, smart watches, smart TVs, or any other connected device is now
vulnerable to the theft of their information [5]. Present organizations and companies
have been quite careful to safeguard their networks. However, even with sustain-
able and accountable defense expenditure, they are still at risk [6]. This occurs as
assailants can circumvent the security of organizations via unknown vulnerabili-
ties that security individuals do not list [7]. A flaw can be uncovered by a repeated
test of a certain hacker in a well-guarded network [8]. Intruders might use network
configuration flaws to breach the intended target [6].
Historical statistics demonstrate that computer threats with different
complexity/lethality exist in various forms. Computer vulnerability has also
been widely documented and a group effort was made to compile lists of vulner-
abilities such as the Database for National Vulnerabilities (NVD) and Common
Vulnerabilities and Exposures (CVE) [9]. In recent years, the amount of risk
assessments filed by NVD has continued to rise. In 2019 the NVD was reported
with 17,302 vulnerabilities [10] while as a total of 16,555 security vulnerabilities
were reported in the year 2018. IT management needs to detect and evaluate
vulnerabilities on various hardware and software platforms in order to prevent
vulnerability losses [11].
In the existing literature lot of quantitative models have been proposed by several
authors. Such quantitative models allow developers to allocate resources for testing
process, planning and security patch creation [12–16]. In addition, developers can use
Vulnerability Discovery Models (VDM) to assess risk and estimate the redundancy
needed in resources and procedures to deal with potential breaches [17]. As the repair
of all the vulnerabilities within the given time and money is a highly tiresome process,
the security team must prioritize these vulnerabilities and remedy the riskiest ones
[18]. Techniques for prioritization are divided into two major groups: quantitative
and qualitative. Qualitative systems offer a grading method for defining the degree of
vulnerabilities in software, whereas quantitative scoring systems link each vulnera-
bility with a number result [19]. Multicriteria (MCDM) decision-making approaches
have recently become a prominent tool for assessing vulnerabilities [20]. It is regarded
a complicated decision-making technique that incorporates both quantitative and
qualitative elements [21].
The literature has offered a broad range of MCDM procedures. The concepts and
applications of Fuzzy MCDM are always evolving [22] as they help to remove the
Analyzing Interrelationships Among Software Vulnerabilities … 293

vagueness in the data. In recent years, certain fuzzy MCDM techniques have been
suggested and utilized frequently, such fuzzy TOPSIS [23], fuzzy ELECTRE [24]
and FBWM [25]. In the context of the use of MCDM in prioritizing vulnerability,
a hybrid strategy has been provided in one recent study by Sibal based on Normal-
ized Criteria Distance (NCD), analytical hierarchy Process (AHP), and DEMATEL.
[26]. Also, Narang et al. [27] have proposed DEMATEL method to show the inter-
dependence in various vulnerabilities by suggesting a cause–effect theory. In the
conventional DEMATEL, measures are based on clear, ambiguous standards. The
following research targets can be selected from the previous discussion:
• Identifying vulnerability types of a software.
• To evaluate the interrelationships in terms of cause and effect among the identified
software vulnerability types.
To address the aforementioned research questions, the manuscript proposes a
framework for assessing these interrelationships of various software vulnerabili-
ties by employing “Fuzzy Decision-making Trial and Evaluation Laboratory (F-
DEMATEL).” DEMATEL methods assist to show a contextual link among the sorts
of vulnerabilities, while ambiguity in the decision-making process is incorporated
into the fuzzy set theory. The manuscript is organized in the following way. The
research technique utilized in this study is described in Sect. 2. Section 2.1 provides
the dataset in this article and Fuzzy DEMATEL is discussed in Sect. 2.2. Section 3
represents the data validation. Section 4 discusses the conclusion of the study.

2 Research Methodology

The present section concentrates on the methods of research. This study identi-
fies the vulnerabilities first, and then uses Fuzzy DEMATEL MCDM technique to
discover interconnections between categories of vulnerability and also to evaluate
what influence one type has on another.

2.1 Dataset Description

The data set for this study was collected from the National Vulnerability Database
(NVD) [28] and CVE [29] which are legitimate databases for data gathering. This
data collection comprises of nine categories of vulnerabilities, which are shown in
the following table along with their notations (Table 1).
294 M. Anjum et al.

Table 1 Software
Notation Vulnerability type
vulnerability types
G1 SQL injection (SQLI)
G2 Cross site scripting (XSS)
G3 Buffer overflow (BO)
G4 Cross site request forgery (CSRF)
G5 File inclusion (FI)
G6 Code execution (CE)
G7 Information gain (IG)
G8 Gain of privileges (GP)
G9 Race condition (RC)

2.2 Fuzzy DEMATEL Approach

To analyze the linguistic assessment by DMs on selected vulnerabilities, we have

combined fuzzy set theory with DEMATEL [30]. The steps involved are:
Step 1: “The decision-making body and short-listing elements are recognized”.
A decision-making committee is established to help identify the study objective,
and the vulnerabilities are highlighted.
Step 2: “Fuzzy language scale detection”.
The metrics assessment scale is being determined with the matching fuzzy
numbers in this phase. Table 2 shows the Fuzzy scale used.
Step 3: “Determining the assessments of decision-making body”.
The relationship between short listed metrics was determined to generate initial
direct matrix (IDM) in this step.
Step 4: “IDM matrix defuzzification”.
The verbal evaluations by the DMs are de-fuzzified into crisp value [31]. Let
Z i j = (ail j , bil j , cil j ) be the effect of metric i on metric j for lth DM. Z i j = (ail j , bil j , cil j )
represents a triangular fuzzy number (TFN) where ail j , bil j , cil j are left, middle and
right values. The steps of CFCS method are as follows:
Step 4.1: “Normalization”:

Table 2 Fuzzy scale

Linguistic term Notation Triangular fuzzy number
Very high influence VH (0.75, 1.0, 1.0)
High influence Hi (0.5, 0.75, 1.0)
Low influence Li (0.25, 0.5, 0.75)
Very no influence VLi (0, 0.25, 0.5)
No influence Ni (0, 0, 0.25)
Analyzing Interrelationships Among Software Vulnerabilities … 295
 
ail j − min ail j
xail j = (1)
max cil j − min ail j
 
bil j − min ail j
xbil j = (2)
max cil j − min ail j
 
cil j − min ail j
xcil j = (3)
max cil j − min ail j

Step 4.2: “Compute the normalized left (lxa) and right (lxc) value”:

xbil j
lxail j =   (4)
1 + xbil j − xail j

xcil j
lxcil j =   (5)
1 + xcil j − xbil j

Step 4.3: “Total normalized crisp value needs to be calculated”

   
lxail j 1 − lxail j + lxcil j lxcil j
xil j =   (6)
1 − lxail j + lxcil j

Step 4.4: “Calculate crisp value”

 
kil j = min ail j + xil j max cil j − min ail j (7)

Step 4.5: “Determine the integrated value”

1 1 
ki j = ki j + ki2j + ... + kil j (8)
l

Step 5: “Determination of Normalized direct relation matrix”

⎛ ⎞
 A
n n
K = ki j nxn
= wher e s = max⎝ max ki j , max ki j ⎠
s 1≤i≤n
j=1
1≤ j≤n
i=1

Step 6: “Total relation matrix T is calculated”

  
T = ti j nxn
= lim K + K 2 + K 3 ....... + K m = K (I − K )−1
m→∞
296 M. Anjum et al.

Step 7: “Building the casual diagram”.

Let the matrices for Ri and Cj be n × 1 and 1 × n indicating sum of the T row
and T column. Ri is the overall influence on another factor by factor I whereas Cj is
the overall effect by factor j. (Ri + Ci) thus reflects the importance of the ith element
in the decision-making process. On the other side, the resulting impact of ith factor
is shown by (Ri-Ci). The DMs can set a threshold limit in order to decrease the
factors which have a minimal influence on others. The values over the limits of the
thresholds will be mapped as a casual diagram in the dataset (Ri + Ci, Ri − Ci).

3 Data Analysis

The objective of this analysis is to investigate the interdependency among software

vulnerability types implying F-DEMATEL. The whole idea is to help security team to
remove such vulnerabilities which are related to other vulnerabilities. In this context,
the main step is to list and subsequently partition the software vulnerability categories
into the cause and effect divisions. The team consists of experts with considerable
development and testing skills of over 5–10 years. For assessing these indicators,
decision makers utilize a five-pointer fuzzy linguistic scale as indicated in Table 2
above. The initial direct matrix (IDM) is created for the identified vulnerability types
as given by the decision makers and is given in Table 3.
The normalized direct-relation (NDR) matrix is calculated following the steps as
mentioned in step 5 of Sect. 2.2. The NDR matrix is shown in Table 4. Then the
total-relation matrix is obtained using step 6 and is represented in Table 5.
Once the normalized direct matrix has been produced, the total relationship matrix
is calculated after the step 6 above. The total relation matrix is given in Table 5.
The complete matrix of cause and effect identifying segments, as specified in step
7, is then computed and presented in Table 6.

Table 3 Initial direct matrix

G1 G2 G3 G4 G5 G6 G7 G8 G9
G1 Ni VLi Li Li VLi VLi Hi VHi VLi
G2 Li Ni VLi VHi Li VLi VHi Li VLi
G3 VLi Li Ni Hi Li VLi Li Li VLi
G4 VLi Hi Li Ni Li VLi Li VLi VLi
G5 Li Li Hi Li Ni VLi Li Hi Li
G6 VLi Li Hi Li VLi Ni Li Li VLi
G7 Li VLi VLi Li Hi Li Ni Li Hi
G8 Li Hi Li VLi Li VLi Li Ni VLi
G9 VLi Li VLi Li Hi Li Li VLi Ni
Analyzing Interrelationships Among Software Vulnerabilities … 297

Table 4 Normalized direct matrix

G1 G2 G3 G4 G5 G6 G7 G8 G9
G1 0.065 0.052 0.100 0.100 0.052 0.052 0.149 0.186 0.052
G2 0.100 0.065 0.052 0.186 0.100 0.052 0.186 0.100 0.052
G3 0.052 0.100 0.065 0.149 0.100 0.052 0.100 0.100 0.052
G4 0.052 0.149 0.100 0.065 0.100 0.052 0.100 0.052 0.052
G5 0.100 0.100 0.149 0.100 0.065 0.052 0.100 0.149 0.100
G6 0.052 0.100 0.149 0.100 0.052 0.065 0.100 0.100 0.052
G7 0.100 0.052 0.052 0.100 0.149 0.100 0.065 0.100 0.149
G8 0.100 0.149 0.100 0.052 0.100 0.052 0.100 0.065 0.052
G9 0.052 0.100 0.052 0.100 0.149 0.100 0.100 0.052 0.065

Table 5 Total-relation matrix

G1 G2 G3 G4 G5 G6 G7 G8 G9
G1 0.393 0.466 0.476 0.539 0.471 0.321 0.610 0.602 0.357
G2 0.464 0.522 0.473 0.675 0.566 0.353 0.701 0.568 0.396
G3 0.366 0.498 0.429 0.575 0.500 0.309 0.549 0.501 0.344
G4 0.351 0.519 0.442 0.480 0.481 0.297 0.530 0.438 0.330
G5 0.468 0.567 0.573 0.606 0.538 0.356 0.630 0.623 0.441
G6 0.361 0.493 0.506 0.529 0.450 0.321 0.545 0.498 0.340
G7 0.449 0.497 0.466 0.575 0.594 0.391 0.566 0.554 0.474
G8 0.420 0.542 0.466 0.493 0.504 0.313 0.560 0.480 0.348
G9 0.369 0.497 0.424 0.534 0.549 0.361 0.553 0.462 0.362

Table 6 Identification of cause and effect

Ri Ci Ri + Ci Ri − Ci
G1 4.235 3.641 7.875 0.594 Cause
G2 4.718 4.601 9.319 0.118 Cause
G3 4.069 4.254 8.323 −0.185 Effect
G4 3.869 5.005 8.874 −1.136 Effect
G5 4.802 4.653 9.455 0.149 Cause
G6 4.043 3.023 7.066 1.020 Cause
G7 4.567 5.244 9.811 −0.678 Effect
G8 4.126 4.726 8.851 −0.600 Effect
G9 4.111 3.392 7.502 0.719 Cause
298 M. Anjum et al.

One measure can be influenced over the others by defining threshold levels in the
overall matrix of the relationship that is 0.476 in this scenario. (R + C) indicates
the strength index of effects exercised or received as indicated in Table 6. Increased
the (R + C) value, larger is the degree of effect on other characteristics. If (R −
C) is positive, then A affects other attributes, and if value is negative, A is being
affected. G7 (Information Gain) has become the most prominent attribute with the
greatest (R + C) value based on the aforementioned information. Once the threshold
value is calculated we can easily calculate which vulnerability is influencing other
vulnerability from Table 5. The values greater than the threshold value are known to be
influencing other vulnerability types accordingly. The vulnerability G1 is influencing
G3, G4, G7 and G8 with intensities of 0.476, 0.539, 0.610 and 0.602 respectively.
Vulnerability type G2 is influencing G4, G5, G7 and G8. Type G3 is influencing
G2, G4, G5 and G7. Vulnerabilities G2, G5 and G7 are being influenced by G4.
Likewise, vulnerability G6 influences all the identified vulnerability types excluding
G5 and G6. G7, G8 and G9 are influencing vulnerabilities G2, G4, G5, G8 and G7
respectively. The values inside the table represent the intensity of the effect. we also
obtain the causal diagram by mapping a dataset of (Ci + Ri, Ci − Ri) as given in
Table 6 is represented in Fig. 1.
On analyzing the above diagram given in Fig. 1, it is clear that vulnerability
metrices are visually divided into the cause and effect group. The vulnerabilities G1,
G2, G5, G6 and G9 are included in the cause group and vulnerability types G3, G4,
G7 and G8 fall under the effect group.

Fig. 1 IRD diagram

Analyzing Interrelationships Among Software Vulnerabilities … 299

4 Conclusion

In our everyday lives, software systems perform important and versatile functions.
In almost every advanced and complicated system, the software components are
the heart and soul. Thus, several organizations are widely using interdependent and
networked software systems for their business decisions. However, organizations
have lost large amounts of money and reputations in contemporary systems owing
to security violations and software vulnerabilities. This work aims at identifying and
assessing contextual connections between the categories of vulnerabilities in software
that assist the Security Team in minimizing the risk utilizing the fuzzy theory and
DEMATEL methodology.
The nine selected vulnerabilities are classified into cause and effect group.
According to the Table 6, it is evident that the vulnerabilities SQLI, XSS, FI, CE and
RC are the causative vulnerabilities while as BO, CSRF, IG and GP are the affected
vulnerabilities. G7 > G5 > G2 > G4 > G8 > G3 > G1 > G9 > G6 is the priority
of vulnerabilities based on the values of (R + C). The most critical vulnerability with
the greatest (R + C) value turned out to be Vulnerability G7 which requires quick
attention. When we look at the equivalent G3 (Ri-Ci) value, we discover it works as
an effect and therefore may decrease the effects of this vulnerability type by working
directly with the causes. Similarly, the following vulnerabilities are found in the list
G4, G7 and G8 and also the effect of additional vulnerabilities. Security managers
need to focus on such causative vulnerabilities so that the other vulnerabilities can
be controlled and loss is minimized.

References

1. Gourisetti SNG, Mylrea M, Patangia H (2019) Cybersecurity vulnerability mitigation frame-

work through empirical paradigm (CyFEr): prioritized gap analysis. IEEE Syst J 14(2):1897–
1908
2. Statista L. Internet of things (iot) connected devices installed base worldwide from 2015 to
2025 (in billions), Statista (2019). https://www.statista.com/statistics/471264/iot-number-of-
connected-devices-worldwide
3. Gubbi J, Buyya R, Marusic S, Palaniswami M (2013) Internet of Things (IoT): a vision,
architectural elements, and future directions. Future Gener Comput Syst 29(7):1645–1660
4. Kapur PK, Pham H, Gupta A, Jha PC (2011) Software reliability assessment with OR
applications. Springer, London, p 364
5. Carnegie Mellon University, Security Vulnerabilities. https://www.sei.cmu.edu/our-work/sec
urity-vulnerabilities/index.cfm
6. Singh UK, Joshi C, Kanellopoulos D (2019) A framework for zero-day vulnerabilities detection
and prioritization. J Inf Secur Appl 46:164–172
7. Huang CC, Lin FY, Lin FYS, Sun YS (2013) A novel approach to evaluate software vulnerability
prioritization. J Syst Softw 86(11):2822–2840
8. Shrivastava AK, Sharma R, Kapur PK (2015) Vulnerability discovery model for a software
system using stochastic differential equation. In: 2015 international conference on futuristic
trends on computational analysis and knowledge management (ABLAZE). IEEE, pp 199–205
300 M. Anjum et al.

9. Dondo MG (2008) A vulnerability prioritization system using a fuzzy risk analysis approach.
In IFIP international information security conference. Springer, Boston, MA, pp 525–540
10. National Vulnerability Database, published on January 1, 2020. https://nvd.nist.gov/general/
news
11. Liu Q, Zhang Y, Kong Y, Wu Q (2012) Improving VRSS-based vulnerability prioritization
using analytic hierarchy process. J Syst Softw 85(8):1699–1708
12. Kimura M (2006) Software vulnerability: definition, modelling, and practical evaluation for
e-mail transfer software. Int J Press Vessels Pip 83(4):256–261
13. Okamura H, Tokuzane M, Dohi T (2013) Quantitative security evaluation for software system
from vulnerability database. J Softw Eng Appl 06:15
14. Kapur PK, Garg RB (1992) A software reliability growth model for an error-removal
phenomenon. Softw Eng J 7(4):291–294
15. Kansal Y, Kapur PK, Kumar U, Kumar D (2017) User-dependent vulnerability discovery model
and its interdisciplinary nature. Int J Life Cycle Reliab Saf Eng, Springer 6(1):23–29
16. Younis A, Joh H, Malaiya Y (2011) Modeling learningless vulnerability discovery using a
folded distribution. In: Proceedings of SAM, vol 11, pp 617–623
17. Arora A, Krishnan R, Nandkumar A, Telang R, Yang Y (2004). Impact of vulnerability disclo-
sure and patch availability-an empirical analysis. In: Third workshop on the economics of
information security vol 24, pp 1268–1287
18. Anjum M, Agarwal V, Kapur PK, Khatri SK (2020) Two-phase methodology for prioritization
and utility assessment of software vulnerabilities. Int J Syst Assurance Eng Manage 11(2):289–
300
19. Liu Q, Zhang Y (2011) VRSS: a new system for rating and scoring vulnerabilities. Comput
Commun 34(3):264–273
20. Kazimieras Zavadskas E, Antucheviciene J, Chatterjee P (2019) Multiple-criteria decision-
making (MCDM) techniques for business processes information management
21. Govindan K, Rajendran S, Sarkis J, Murugesan P (2015) Multicriteria decision making
approaches for green supplier evaluation and selection: a literature review. J Clean Prod
98:66–83
22. Mardani A, Jusoh A, Zavadskas EK (2015) Fuzzy multiple criteria decision-making techniques
and applications–two decades review from 1994 to 2014. Expert Syst Appl 42(8):4126–4148
23. Zhang X, Xu Z (2015) Soft computing based on maximizing consensus and fuzzy TOPSIS
approach to interval-valued intuitionistic fuzzy group decision making. Appl Soft Comput
26:42–56
24. Chen N, Xu Z (2015) Hesitant fuzzy ELECTRE II approach: a new way to handle multi-criteria
decision-making problems. Inf Sci 2015(292):175–197
25. Anjum M, Kapur PK, Agarwal V, Khatri SK (2020) A framework for prioritizing software
vulnerabilities using fuzzy best-worst method. In: 2020 8th international conference on relia-
bility, infocom technologies and optimization (trends and future directions) (ICRITO). IEEE,
pp 311–316
26. Sibal R, Sharma R, Sabharwal S (2017) Prioritizing software vulnerability types using multi-
criteria decision-making techniques. Life Cycle Reliab Saf Eng 6(1):57–67
27. Narang S, Kapur PK, Damodaran D, Majumdar R (2018) Prioritizing types of vulnerability
on the basis of their severity in multi-version software systems using DEMATEL technique.
In: 2018 7th international conference on reliability, infocom technologies and optimization
(trends and future directions) (ICRITO). IEEE, pp 162–167
28. National Vulnerability Database, nvd.nist.gov/, 2020
29. CVE Details, The Ultimate Security Vulnerability Data source, www.cvedetails.com. 2020
30. Agarwal V, Govindan K, Darbari JD, Jha PC (2016) An optimization model for sustainable
solutions towards implementation of reverse logistics under collaborative framework. Int J Syst
Assurance Eng Manage 7(4):480–487
31. Opricovic S, Tzeng GH (2003) Defuzzification within a multicriteria decision model. Int J
Uncertainty, Fuzziness Knowl-Based Syst 11(05):635–652
Universal Generating Function Approach
for Evaluating Reliability and Signature
of All-Digital Protection Systems

Soni Bisht and S. B. Singh

Abstract Reliability analysis of various components of all-digital protection

systems (ADPS) is one of the key parts of the system reliability quantification process.
This paper focuses on the reliability indices and signature evaluation of the ADPS.
Here, first measure is to find the reliability of the ADPS using UGF having both inde-
pendent identically and non-identically distributed components. Second measure is
to evaluate the signature with the help of Owen’s method using different algorithms
where all the system components are coherent. Lastly, we have calculated the mean
time to failure with the help of minimal signature. This paper associates with the
reliability block diagram (RBD) as well as the system reliability and signature of the
considered systems.

Keywords Digital protection system · Reliability block diagram · Coherent

system · Universal generating function (UGF) · Signature reliability · Minimal
signature

1 Introduction

In an electrical power system, power distribution is an important factor in the trans-

mission of desirable electricity to the consumers. The manufacturing of reasonable
electric power systems has been changed due to the various environmental factors,
different economic problems and customer demands which need more reliable and
efficient electricity. To recognize the theory of the future grid, the current demand is
to develop advanced grid resources, various communication techniques, and infor-
mation technologies in the electric power system. A future grid is associated with
modernized power system having intelligent technologies which significantly help to

S. Bisht (B)
Department of Mathematics, Eternal University, Himachal Pradesh, Baru Sahib, India
S. B. Singh
Department of Mathematics, Statistics and Computer Science, G.B. Pant, University of
Agriculture and Technology, Pantnagar, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 301
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_14
302 S. Bisht and S. B. Singh

control and monitor the power system. There are many extensive controlling systems
in the future grid which deals with control of different aspects such as voltage, current,
energy distribution at transformers, substations transformers, distribution in different
switching devices and smart meters [4, 21].
The latest development of non-conventional devices and widespread use of
different digital relays to analyze the performance of an ADPS. An IEC61850
digital process system is connected to digital relays, through the output of the
non-conventional device. IEC 61,850 substations have been found to be more reli-
able than traditional substations, which have been in use for decades. Many of the
features mentioned in the IEC 61,850, such as fibre optics, redundancy settings, and
a self-testing and monitoring system, all contribute to the IEC 61,850’s increased
reliability. Usually, all conventional protection system is less reliable than digital
protection systems. Digital protection systems contain more electronic devices than
the conventional one, viz. MUs, Ethernet switches, time sources and PRs. The relia-
bility of the system can be improved in two ways, one by replacing the copper wires
with the fiber optics as proposed by the IEC 61,850 process bus. Another way is
redundancy, viz. (i) by adding redundant components in the selected components
of MUs and redundant MUs, named as MU1 and MU2; (ii) a communication link
to protective relay (PR) and their redundant components, recognized as PR1 and
PR2; and (iii) Ethernet communication media (EC) and their redundant components,
named as EC1, EC2, EC3, EC4, EC5, and EC6 [2].
In the past, many techniques are used for the analysis of different reliability aspects
of conventional protection systems, one of them is Markov process. Singh and Patton
[17] discussed a model of a system and its related protection system. The protection
systems recognized the faults and faulty components in the system to prevent the
damage and also minimize the effect of fault component in the whole operation of
the system. Schweitzer et al. described the reliability analysis of protective systems
using fault tree analysis. Many techniques applied for the reliability analysis of
conventional protection systems are also applied to the digital protection system.
Schweitzer et al. first introduced the unconventional architectures of ADPS and found
the reliability with the help of minimal path method. Here, researchers analyzed the
component importance of all alternative architectures. Chauhan et al. [2] discussed
the ranking of all components present in digital protection system using cost-based
heuristic redundancy importance measure.
The reliability analysis with the help of UGF is a novel approach used in reliability
engineering, especially in the digital protection system. Several authors applied UGF
in many systems such as series, parallel, l-out-of-m, consecutive m-out-of-n etc. with
identical and different elements [7, 8, 18] analyzed the reliability of binary state
and multistate systems with the help of UGF using different algorithms. Rausand
and Arnljot [13] also studied the different types of the probabilistic and statistical
reliability of different complex systems. Ushakov [19] evaluated the unreliability of a
weighted (w-l + 1)-out-of-m: system. Negi and Singh [12] computed the reliability,
expected lifetime and Birnbaum measure of the non- repairable complex systems
which have two subsystems A and B having weighted u-out-of-v: G and weighted
m-out-of-n: G configurations respectively with the help of UGF.
Universal Generating Function Approach for Evaluating Reliability and Signature … 303

Da et al. [3] examined the importance of component in a coherent system with

the help of Birnbaum Measure and the structural importance. Boland et al. [1] intro-
duced the new component importance called redundancy importance in the coherent
systems. Samaniego and Francisco [15] discussed the concepts of signature which
is an essential tool for the study of stochastic and aging properties of the coherent
systems. It gives the limited information about the failure status of the system lifetime
of the coherent systems. Navarro and Rychlik [11] obtained the upper and lower limits
for the reliability functions and expected lifetimes of the coherent systems based
on the dependent exchangeable absolutely continuous component with the help of
the signature. Kumar and Singh [5] evaluated the reliability of l-out-of-m coherent
system with the help of UGF. Here authors assessed the tail signature, signature
reliability, Barlow Proschan index, mean time to failure and estimated cost. Navarro
and Rubio [10] previously developed an algorithm for determining the signature
reliability and MTTF of a five-component coherent system. Marichal and Mathonet
[9] examined the reliability function, tail signature, signature, and minimum signa-
tures of the coherent system using the structural function in the literature. Kumar
and Singh [6] calculated the reliability, signature reliability, mean time to failure,
Barlow Proschan index and predictable cost of sliding window systems with the use
of different algorithms.
From above discussion, it is noticeable that lots of work have been done for the
reliability improvement of the non-conventional protection systems using different
techniques. Keeping this fact in the view the current paper proposes to study the
models corresponding to different architectures of digital protection systems which
is composed of non-conventional devices: external time sources, MUs, Ethernet
switches, EC and PR. Here we also analyze the reliability of independent and identi-
cally distributed and non-identically distributed components. The main focus of our
study is to estimate the reliability and signature of digital protection system with the
help of UGF and Owen’s method to show the impact of various components in the
system.

1.1 Definitions

1.1.1 Structure–function

The mapping ϕ : {0, 1}m → {0, 1} of a system consisting of m components is called

the structure–function of the coherent system. A system works if certain combinations
of components work:

1, if the system works
ϕ(x) =
0, if the system fails
304 S. Bisht and S. B. Singh

1.1.2 Coherent Systems

A system is considered to be coherent if each of its components is significant (a

component is relevant when the state of the system is determined by whether the
components are operating or not) and its structure function is monotone (a structure
function is monotone when the state of the system will never improve after the failure
of the component).

1.1.3 Structure Design of ADPS

Digital protection system is constructed with the help of MUs, time sources, Ethernet
switches, EC and PR.

1.1.4 Merging Unit

MU acquires AC current and voltage from conventional current and voltage trans-
formers. MU digitizes the AC current and voltage signals and sends them over the
Ethernet network in the form of sampled values.

1.1.5 Time Sources

Time source is an external time source but an external synchronization source needed
by the system for the large time synchronization. A time source is required to manage
the overall system reliability.

1.1.6 Ethernet Switch

This device is used to develop a network connection between the attached compo-
nents. Ethernet switch understand the packets addressing scheme and send any data
packet only to its destination ports.

1.1.7 Ethernet Communication Media (EC)

EC plays a very significant role in all-digital protection systems. The EC is a part of

the intelligent electronic device (IED). The IED is used in the electric power industry
to describe the power system equipment, such as circuit breakers.
Universal Generating Function Approach for Evaluating Reliability and Signature … 305

Structure design 1 Structure design 2

Structure design 3 Structure design 4

Structure design 5

Fig. 1 Structure designs (SDs) of digital protection system

1.1.8 Protective Relay

It is a relay tool measured to trip a circuit breaker when a fault is identified. Electro-
magnetic devices are the first PRs, relying on coils to recognize the irregular oper-
ating conditions such as overflow of current and voltage and reverse flow of voltage
and current. The need of protective relay is to protect from damage of circuits and
equipment within a few seconds.

1.2 Alternative System Structure Designs

Zhang et al. described various types of alternative structure designs of the ADPS.
In the first structure design, we considered the redundancy only on PRs whereas in
second structure design redundancies are there in MUs and PRs which are dependent
on each other. The third structure design has two redundant and independent PRs
and in the fourth structure design, the switches are in cascade form so as to provide
cross backup in the merging units. Fifth architecture adds redundancy in Ethernet
switches and EC (Fig. 1).

1.3 System Reliability Block Diagram

RBD is the graphical version of the components of a system and the associations
between them, which can be used to establish the overall system reliability, even
for a large-scale system. Therefore, it can be accomplished that the RBD method is
a precise method for the analysis of such systems. Many researchers observed that
the use of RBDs is one of the very important ways for system reliability analysis.
306 S. Bisht and S. B. Singh

(i) RBD of structural design 1

(ii) RBD of structural design 2

(iii) RBD of structural design 3

(iv) RBD of structural design 4

(v) RBD of structural design 5

Fig. 2 RBD of the different structure designs

Here, we consider the five-reliability block diagram in Fig. 2 which is related to the
structure designs of ADPS.

1.4 Universal Generating Function (UGF)

In reliability engineering, there are various methods for the reliability assessment of
the systems. The universal generating function is one of the useful tools due to its
difficulty and time reducing capability. The UGF is firstly introduced by Ushakov
[18].
The UGF of an independent discrete random variable l is expressed in the
polynomial form as:


M
U (z) = pm z l m (1)
m=1
Universal Generating Function Approach for Evaluating Reliability and Signature … 307

where the variable X has m has possible values and pm is the probability that X is
equal to lm and z is any variable.
Consider r independent discrete random variable X 1 , X 2 , …, X r . Let the UGF of
random variable X 1 , X 2 , …, X r be U1 (z), U2 (z)........., Ur (z) respectively and f (X 1 ,
X 2 , …, Xr) be an arbitrary function. Further, combination of r UGF is defined by
composition operator ⊗ f , where the properties of the composition operator strictly
depend on the properties of the function f (X 1 , X 2 , …, X r) . Therefore, the composition
U r (z) is defined as:

Ur (z) = ⊗ f U1 (z), U2 (z), . . . , Ur (z)

The UGFs’ of the structure design containing two types of components are
expressed as follow:
For the series arrangement, the UGF can be expressed as:


M 
N 
M 
N
Uc (z) ⊗ Ud (z) = kcm z gcm ⊗ kdn z gdn = kcm kdn z ser (gcm , gdn ) (2)
ser ser
m=1 n=1 m=1 n=1

UGF of a parallel system is given by:


M 
N 
M 
N
Uc (z) ⊗ Ud (z) = kcm z gcm ⊗ kdn z gdn = kcm kdn z par (gcm , gdn ) (3)
par par
m=1 n=1 m=1 n=1

where, ⊗ and ⊗ are the composition operators over u-function associated with
ser par
series and parallel system respectively. ser (gam , gbn ) and par (gam , gbn ) yield the
entire performance rates of binary state elements c and d at state m and n respectively.

1.5 Assessment of System Reliability

1.5.1 Reliability of Structural Design 1

The reliability evaluation of structure design 1 by UGF can be done using following
two steps:
(i) Consider the structure design that contains non-identical distributed compo-
nents.
(ii) Assume the architectures that contain independent and identically distributed
components.
Reliability expression of structure design 1 (RSD1 ) with the use of UGF can be
obtained as:
308 S. Bisht and S. B. Singh

RSD1 = min(max(min(MU, EC, E S) min(P R1, EC1)

(4)
min(P R2, EC2)T S1, T S2))

(a) Reliability computation of structure design 1, when all components in the

system are non-identically distributed.

The UGF of different components present in the design can be expressed as:

u i (z) = pi z 1 + (1 − pi )z 0 (5)

Here, pi is the probability of i th component, where i = TS1, TS2, MU, EC, ES,
PR1, PR2, EC1, EC2.
Now, apply the composition operator ⊗(uTS1 (z), uTS2 (z), uMU (z), uEC (z), uES (z),
uPR1 (z), uEC1 (z), uPR2 (z), uEC2 (z)) to obtain the system structure function.
If X 1 is the state variable corresponding to the subsystem consisting of elements
TS1 and TS2, then structure function is obtained as:

X 1 (z) = max (T S1, T S2)

When X 2 is the state variable related to the subsystem containing the elements
MU, EC and ES, then structure function can be written as:

X 2 (z) = min (MU, EC, ES)

When X 3 is the state variable corresponding to the subsystem consisting of

elements PR1 and EC1, then the structure function is given by:

X 3 (z) = min (P R1, EC1)

If X 4 is the state variable corresponding to the subsystem consisting of elements

PR2 and EC2, then structure function can be written as:

X 4 (z) = min (P R2, EC2)

When X 5 is the state variable related to the subsystem containing the elements X 3
and X 4 , then the structure function can be evaluated by:

X 5 (z) = max(X 3 , X 4 )

If X is the state variable corresponding to the subsystem containing the elements

X 1, X 2 , and X 5 , then the structure function can be computed as:
 
X (z) = min X 1, X 2 , X 5
Universal Generating Function Approach for Evaluating Reliability and Signature … 309

Assuming the values of pi ’s, i = TS1, TS2, ES, PR1, PR2, EC1, EC, and
substituting the same in Eq. (5), we have UGFs’ of the components of the system as:

u T S1 (z) = 0.9z 1 + 0.1z 0 , u T S2 (z) = 0.95z 1 + 0.05z 0 ,

u E S (z) = 0.95z 1 + 0.05z 0 , u P R1 (z) = 0.8z 1 + 0.2z 0 ,

u P R2 (z) = 0.85z 1 + 0.15z 0 , u EC1 (z) = 0.85z 1 + 0.15z 0 , u EC (z) = 0.8z 1 + 0.2z 0 .

To find the system reliability of the structure design 1, we apply composition

operators to UGFs of different components. By doing so, we have

U1 (z) = u T S1 (z) ⊗ u T S2 (z)

max
 
= 0.9z + 0.1z 0 ⊗ (0.95z 1 + 0.05z 0 ) = (0.995z 1 + 0.005z 0 ),
1
max

U2 (z) = u MU (z) ⊗ u EC (z) ⊗ u E S (z) = (0.684z 1 + 0.32z 0 ),

min min

U3 (z) = u P R1 (z) ⊗ u EC1 (z) = (0.68z 1 + 0.32z 0 ),

min

U4 (z) = u P R2 (z) ⊗ u EC2 (z) = (0.595z 1 + 0.405z 0 ),

min

U5 (z) = U3 (z) ⊗ U4 (z) = (0.8704z 1 + 0.1296z 0 ).

max

Finally, the reliability of the structure design 1 is computed as:

U (z) = U1 (z) ⊗ U2 (z) ⊗ U5 (z) = (0.592376832z 1 + 0.407623168z 0 )

min min

(b) When components, namely merging units, Ethernet communication media,

Ethernet switch, time source and protective relay having identical probabilities
of structure design 1, i.e., the probabilities of all the components are same, then
the structure function becomes:

RSD1 = 2 p 9 − 4 p 8 − p 7 + 4 p 6 (6)

Taking into account the different components reliabilities, one can get the
reliability of structure design 1 as listed in Table 1 and depicted in Fig. 3.
310 S. Bisht and S. B. Singh

Table 1 Values of reliability

Component reliability Reliability of SD1 (RSD1 )
of structure design 1
0.90 0.7004392
0.91 0.7295654
0.92 0.7590206
0.93 0.7887547
0.94 0.8187153
0.95 0.8488473
0.96 0.8790933
0.98 0.9396874
0.99 0.9699172

0.8
Reliability of SD 1

0.6

0.4

0.2

0
0.88 0.9 0.92 0.94 0.96 0.98 1
Component Reliability

Fig. 3 Component reliability versus reliability of SD1

1.5.2 Reliability of Structural Design 2

The reliability R S D2 of the structural design 2, with the help of UGF method is
expressed as:

RSD2 = min(max(min(MU 1, EC1)min(MU 2, EC2)E S)

(7)
(min(P R1, EC3)min(P R2, EC4)T S1, T S2)

(a) Reliability of structure design 2, when the components in the system are non-
identically distributed. In this case, the system structure functions are as follow:

where,
Universal Generating Function Approach for Evaluating Reliability and Signature … 311

State variable Subsystem consisting of elements

X1 max (TS1, TS2)
X2 min (MU1, EC1)
X3 min (MU2, EC2)
X4 max (X 2 , X 3 )
X5 min (PR1, EM3)
X6 min (PR2, EM4)
X7 max (X 5 , X 6 )
X8 min (X 1 , X 4, ES and X 7 )

Substituting the assumed values of pi ’s in Eq. (5), we have UGFs’ of components

of the structure design 2 as:

u T S1 (z) = 0.9z 1 + 0.1z 0 , u T S2 (z) = 0.95z 1 + 0.05z 0 ,

u MU 1 (z) = 0.8z 1 + 0.2z 0 , u MU 2 (z) = 0.85z 1 + 0.15z 0

u P R1 (z) = 0.8z 1 + 0.2z 0 , u P R2 (z) = 0.85z 1 + 0.15z 0 ,

u E S (z) = 0.9z 1 + 0.1z 0 , u EC1 (z) = 0.7z 1 + 0.3z 0 ,
u EC2 (z) = 0.75z 1 + 0.25z 0 , u EC3 (z) = 0.85z 1 + 0.25z 0 , u EC4 (z) = 0.7z 1 + 0.3z 0 .

Now to evaluate the reliability of structure design 2, firstly compute the following
by applying composition operators to the UGFs of components.
 
U1 (z) = u T S1 (z) ⊗ u T S2 (z) = 0.9z 1 + 0.1z 0 ⊗ (0.95z 1 + 0.05z 0 )
max max
= (0.995z 1 + 0.005z 0 ),

U2 (z) = u MU 1 (z) ⊗ u EC1 (z) = 0.56z 1 + 0.44z 0 ,

min

U3 (z) = u MU 2 (z) ⊗ u EC2 (z) = 0.6375z 1 + 0.3625z 0 ,

min

U4 (z) = U2 (z) ⊗ U3 (z) = 0.8405z 1 + 0.1152z 0 ,

max

U5 (z) = U P R1 (z) ⊗ U E M3 (z) = 0.68z 1 + 0.32z 0 ,

min

U6 (z) = U P R2 (z) ⊗ U E M4 (z) = 0.595z 1 + 0.405z 0 ,

min

U7 (z) = U5 (z) ⊗ U6 (z) = 0.8704z 1 + 0.1296z 0 ,

max
312 S. Bisht and S. B. Singh

Table 2 Values of reliability

Component reliability Reliability of SD2
of structure design 2
0.90 0.82783096
0.91 0.85007248
0.92 0.87148760
0.93 0.89196890
0.94 0.91140759
0.95 0.92969391
0.96 0.94671769
0.98 0.97653804
0.99 0.98911714

U E S (z) = 0.9z 1 + 0.1z 0 .

Finally, the reliability of the structure design 2 is obtained as:

U (z) = U1 (z) ⊗ U4 (z) ⊗ U E S (z) ⊗ U7 (z) = 0.6559z 1 + 0.3441z 0

min min min

(b) When components are independent and identically distributed ( pi = p), then
the structure function of reliability of the structure design 2 is expressed as:

RSD2 = − p 11 + 2 p 10 + 4 p 9 − 8 p 8 − 4 p 7 + 8 p 6 (8)

Table 2 lists the reliability of structure design 2 corresponding to different

component reliability and the same is represented graphically by Fig. 4.

1
0.9
0.8
Reliability of SD 2

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.88 0.9 0.92 0.94 0.96 0.98 1
Component Reliability

Fig. 4 Component reliability versus reliability of SD2

Universal Generating Function Approach for Evaluating Reliability and Signature … 313

1.5.3 Reliability of Structural Design 3

The reliability of structure design 3 with the help of UGF is given by:

RSD3 = min(max(T S1, T S2)

max(min(MU 1, EC1, E S1, P R1, EC3)min(MU 2, EC2, E S2, P R2, E S4))
(9)

(a) When all components in the system are non-identically distributed, then the
reliability of structure design 3 is computed as:
If X 1 is the state variable corresponding to the subsystem containing the elements
TS1and TS1, then the structure function can be expressed as:

X 1 (z) = max (T S1, T S2)

If X 2 is the state variable related to the subsystem containing the elements MU1,
EC1, ES1, PR1and EC3, then the structure function can be computed as:

X 2 (z) = min(MU 1, EC1, E S1, P R1 and EC3)

When X 3 is the state variable corresponding to the subsystem containing the

elements MU2, EC2, ES2, PR2 and EC4, then the structure function is given by:

X 3 (z) = min(MU 2, EC2, E S2, P R2 and EC4)

If X 4 is the state variable related to the subsystem containing the elements X 2 and
X 3 , then the structure function is computed by:

X 4 (z) = max(X 2 , X 3 )

When X is the state variable corresponding to the subsystem containing the

elements X 1 and X 4 , then the structure function becomes:

X (z) = max(X 1 , X 4 )

Putting the assumed values of pi ’s in Eq. (5), the UGFs’ of components can be
expressed as:

u T S1 (z) = 0.9z 1 + 0.1z 0 , u T S2 (z) = 0.95z 1 + 0.05z 0 ,

u MU 1 (z) = 0.8z 1 + 0.2z 0 , u MU 2 (z) = 0.85z 1 + 0.15z 0
314 S. Bisht and S. B. Singh

u P R1 (z) = 0.8z 1 + 0.2z 0 , u P R2 (z) = 0.85z 1 + 0.15z 0 ,

u E S1 (z) = 0.9z 1 + 0.1z 0 , u E S2 (z) = 0.95z 1 + 0.05z 0

u EC1 (z) = 0.7z 1 + 0.3z 0 , u EC2 (z) = 0.75z 1 + 0.25z 0 ,

u EC3 (z) = 0.85z 1 + 0.25z 0 , u EC4 (z) = 0.7z 1 + 0.3z 0

Now applying the composition operators, we have

 
U1 (z) = u T S1 (z) ⊗ u T S2 (z) = 0.9z 1 + 0.1z 0 ⊗ (0.95z 1 + 0.05z 0 )
max max
= (0.995z + 0.005z )
1 0

U2 (z) = u MU 1 (z) ⊗ u EC1 (z) ⊗ u E S1 (z) ⊗ u P R1 (z) ⊗ u EC3

min min min min
= 0.34272z + 0.65728z
1 0

U3 (z) = u MU 2 (z) ⊗ u EC2 (z) ⊗ u E S2 (z) ⊗ u P R2 (z) ⊗ u EC4

min min min min
= 0.3603468z 1 + 0.63965312z 0

U4 (z) = U2 (z) ⊗ U3 (z) = 0.579568774z 1 + 0.420431206z 0

max

At last, the reliability of the structure design 3 is given by:

U (z) = U1 (z) ⊗ U4 (z) = 0.576670z 1 + 0.42333z 0

max

(b) It is worth mentioning that when all components are identical ( pi = p), then
the structure function of reliability becomes:

RSD3 = p 12 − 2 p 11 − 2 p 7 + 4 p 6 (10)

Table 3 lists the reliability of structure design 3 corresponding to different

components reliability and also represented graphically by Fig. 5.

1.5.4 Reliability of Structural Design 4

The reliability of structural design 4 by the proposed method is:

RSD4 = min(max(T S1, T S2)(max(min(MU 1, EC1)min(MU 2, EC2)))

(11)
min(E S1, E S2)min(P R1, EC3)min(P R2, EC4))
Universal Generating Function Approach for Evaluating Reliability and Signature … 315

Table 3 Values of reliability

Component reliability Reliability of SD3
of structure design 3
0.90 0.823978544
0.91 0.851693121
0.92 0.878118434
0.93 0.902948198
0.94 0.925847836
0.95 0.946452873
0.96 0.964367298
0.98 0.990372479
0.99 0.997498265

1
0.9
0.8
Reliability of SD 3

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.88 0.9 0.92 0.94 0.96 0.98 1
Component Reliability

Fig. 5 Component reliability versus reliability of SD3

System structure function of the structure design 4, when all components in the
system are non-identically distributed is given by:
where,

State variable Subsystem consisting of elements

X 1 (z) max (TS1, TS2)
X 2 (z) min (MU1, EM1)
X 3 (z) min (MU2, EM2)
X 4 (z) max (X 2 , X 3 )
X 5 (z) min (ES 1 , ES 2 )
X 6 (z) min (PR1, EM3)
(continued)
316 S. Bisht and S. B. Singh

(continued)
State variable Subsystem consisting of elements
X 7 (z) max (PR2, EM4)
X 8 (z) max (X 6 , X 7 )
X(z) min (X 1 , X 4, X 5 and X 8 )

Again the UGFs’ of the components for the considered values of pi ’s are obtained
as:

u T S1 (z) = 0.9z 1 + 0.1z 0 , u T S2 (z) = 0.95z 1 + 0.05z 0 ,

u MU 1 (z) = 0.8z 1 + 0.2z 0 , u MU 2 (z) = 0.85z 1 + 0.15z 0

u P R1 (z) = 0.8z 1 + 0.2z 0 , u P R2 (z) = 0.85z 1 + 0.15z 0 ,

u E S1 (z) = 0.9z 1 + 0.1z 0 , u E S2 (z) = 0.95z 1 + 0.05z 0
u EC1 (z) = 0.7z 1 + 0.3z 0 , u EC2 (z) = 0.75z 1 + 0.25z 0 ,
u EC3 (z) = 0.85z 1 + 0.25z 0 , u EC4 (z) = 0.7z 1 + 0.3z 0

To evaluate the system reliability of the structure design 4 let us apply the
composition operators:
 
U1 (z) = u T S1 (z) ⊗ u T S2 (z) = 0.9z 1 + 0.1z 0 ⊗ (0.95z 1 + 0.05z 0 )
max max
= (0.995z + 0.005z ),
1 0

U2 (z) = u E M1 (z) ⊗ u MU 1 (z) = 0.56z 1 + 0.44z 0 ,

min
U3 (z) = u E M2 (z) ⊗ u MU 2 (z) = 0.6375z 1 + 0.3625z 0 ,
min

U4 (z) = U2 (z) ⊗ U3 (z) = 0.8405z 1 + 0.1595z 0 ,

max

U5 (z) = u E S1 (z) ⊗ u E S2 (z) = 0.855z 1 + 0.145z 0 ,

min

U6 (z) = U P R1 (z) ⊗ U E M3 (z) = 0.68z 1 + 0.32z 0 ,

min

U7 (z) = U P R2 (z) ⊗ U E M4 (z) = 0.595z 1 + 0.405z 0 ,

min

U8 (z) = U6 (z) ⊗ U7 (z) = 0.8704z 1 + 0.1296z 0 .

max

At last, the reliability of the structure design 4 can be computed as:

Universal Generating Function Approach for Evaluating Reliability and Signature … 317

Table 4 Values of reliability

Component reliability Reliability of SD4
of structural design 4
0.90 0.745047864
0.91 0.773565963
0.92 0.801768596
0.93 0.829531084
0.94 0.856723137
0.95 0.883209219
0.96 0.908848990
0.98 0.957007283
0.99 0.979226854

U (z) = U1 (z) ⊗ U4 (z) ⊗ U5 (z) ⊗ U8 (z) = 0.62236z 1 + 0.37764z 0

min min min

(b) Reliability of structural design 4 when the components are independent and
identically distributed ( pi = p) are having the same probabilities then the
structure function is given by:

RSD4 = − p 12 + 2 p 11 + 4 p 10 − 8 p 9 − 4 p 8 + 8 p 7 (12)

Considering the different components reliabilities, one can get the reliability of
structure design 4 as listed in Table 4 and depicted in Fig. 6.

1
0.9
0.8
Reliability of SD 4

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.88 0.9 0.92 0.94 0.96 0.98 1
Component Reliability

Fig. 6 Component reliability versus reliability of SD4

318 S. Bisht and S. B. Singh

1.5.5 Reliability of Structural Design 5

The reliability of the structure design 5 with the help of UGF is given by:

RSD5 = min(max(T S1, T S2)

(max(min(MU 1, MU 2)
max((min(EC1, E S1, EC3)min(EC2, E S2, EC4)
(min(EC5, E S3, EC7)min(EC6, E S4, EC8))max(P R1, P R2))
(13)

When X 1 is the state variable corresponding to the subsystem containing the

elements TS1 and TS2, then the structure function can be written as:

X 1 (z) = max (T S1, T S2)

If X 2 is the state variable related to the subsystem containing the elements EC1,
ES1and EC3, then the structure function can be expressed as:

X 2 (z) = min (EC1, E S1 and EC3)

If X 3 is the state variable corresponding to the subsystem containing the elements

EC2, ES2 and EC4, then the structure function is given by:

X 3 (z) = min (EC2, E S2, EC4)

If X 4 is the state variable related to the subsystem containing the elements X 2 and
X 3 , then the structure function can be computed as:

X 4 (z) = max(X 2 , X 3 )

If X 5 is the state variable corresponding to the subsystem containing the elements

EC5, ES3 and EC7, then the structure function can be obtained as:

X 5 (z) = min(EC5, E S3, EC7)

If X 6 is the state variable corresponding to the subsystem containing the elements

EC6, ES4 and EC8, then the structure function becomes:

X 6 (z) = min(EC6, E S4, EC8)

If X 7 is the state variable related to the subsystem containing the elements X 5 and
X 6 , then the structure function can be obtained as:

X 7 (z) = max(X 5 , X 6 )
Universal Generating Function Approach for Evaluating Reliability and Signature … 319

When X 8 is the state variable related to the subsystem containing the elements X 4
and X 7 , then the structure function can be expressed as:

X 8 (z) = max(X 4 , X 7 )

When X is the state variable corresponding to the subsystem containing the

elements X 1 andX 8 , then the structure function becomes:

X (z) = max(X 1 , X 8 )

UGFs’ of the structure design 5 corresponding to considered values of pi ’s using

Eq. (5) are given by:

u T S1 (z) = 0.9z 1 + 0.1z 0 , u T S2 (z) = 0.95z 1 + 0.05z 0 ,

u MU 1 (z) = 0.8z 1 + 0.2z 0 , u MU 2 (z) = 0.85z 1 + 0.15z 0 ,

u P R1 (z) = 0.8z 1 + 0.2z 0 , u P R2 (z) = 0.85z 1 + 0.15z 0 ,

u E S1 (z) = 0.9z 1 + 0.1z 0 , u E S2 (z) = 0.95z 1 + 0.05z 0 ,

u EC1 (z) = 0.7z 1 + 0.3z 0 , u EC2 (z) = 0.75z 1 + 0.25z 0 ,

u EC3 (z) = 0.85z 1 + 0.25z 0 , u EC4 (z) = 0.7z 1 + 0.3z 0 ,
u EC5 (z) = 0.9z 1 + 0.1z 0 , u EC6 (z) = 0.95z 1 + 0.05z 0 ,
u EC7 (z) = 0.88z 1 + 0.12z 0 , u EC8 (z) = 0.8z 1 + 0.2z 0

u E S3 (z) = 0.98z 1 + 0.02z 0

u E S4 (z) = 0.95z 1 + 0.05z 0 .

Applying composition operators to UGFs, we have.

U1 (z) = u T S1 (z) ⊗ u T S2 (z) = (0.9z 1 +0.1z 0 ) ⊗ (0.95z 1 +0.05z 0 ) = (0.995z 1 +
max max
0.005z 0 ),

U2 (z) = u EC1 (z) ⊗ u E S1 (z) ⊗ u EC3 (z) = 0.5355z 1 + 0.4645z 0 ,

min min

U3 (z) = u EC2 (z) ⊗ u E S2 (z) ⊗ u EC4 (z) = 0.49875z 1 + 0.50125z 0 ,

min min

U4 (z) = U2 (z) ⊗ U3 (z) = 0.767169375z 1 + 0.232830625z 0 ,

max

U5 (z) = u EC5 (z) ⊗ u E S3 (z) ⊗ u EC7 (z) = 0.77616z 1 + 0.22384z 0 ,

min min
320 S. Bisht and S. B. Singh

Table 5 Values of reliability

Component reliability Reliability of SD5
structure design 5
0.90 0.557636753
0.91 0.600090709
0.92 0.643788337
0.93 0.688492578
0.94 0.733922695
0.95 0.779751827
0.96 0.825604915
0.98 0.915632991
0.99 0.921059188

U6 (z) = u EC6 (z) ⊗ u E S4 (z) ⊗ u EC8 (z) = 0.722z 1 + 0.278z 0 ,

min min

U7 (z) = U5 (z) ⊗ U6 (z) = 0.68z 1 + 0.32z 0 ,

max
U8 (z) = U4 (z) ⊗ U7 (z) = 0.719430327z 1 + 0.280569673z 0 .
max
Finally, the reliability of structure design 5 can be computed as:

U (z) = U1 (z) ⊗ U8 (z) = 0.715833175z 1 + 0.284166825z 0

min

(b) The structure function of the structure design 5 where components are
independent and identically distributed is expressed as:

RSD5 = − p 18 + 2 p 17 + 4 p 15 − 8 p 14 − 4 p 12 + 8 p 11 (14)

Table 5 lists the reliability of structure design 5 corresponding to different

components reliability and the same is represented graphically by Fig. 7.

1
0.8
Reliability of SD 5

0.6
0.4
0.2
0
0.88 0.9 0.92 0.94 0.96 0.98 1
Component Reliability

Fig. 7 Component reliability versus reliability of SD 5

Universal Generating Function Approach for Evaluating Reliability and Signature … 321

1.6 Signature Reliability

The idea of the signature was firstly introduced by Samaniego [14] for the systems
whose components have continuous and i.i.d lifetimes. In recent decades, the signa-
ture has proved to be one of the powerful tools for the quantification of the reliability
of the coherent system. The signature is very effective tool in optimal design and
reliability economics of the systems. Let G 1 , G 2 , ....., G v be i.i.d. component with
continuous distribution function. Let T be the lifetime of this network, then the
reliability at a time t ≥ 0 is given by
v

P(T > t) = su P(G u:v > t) (15)
u=1

where, G 1:v ≤ G 2:v ≤ ........ ≤ G v:v are the ordered lifetime of the components
and signature of uth components su = P(T = X u:v ), u = 1, 2, ....., v.
The vector s = (s1 , s2 , ...., sv ) is called the signature of the system.

1.6.1 Algorithm for Evaluating the Signature of All Digital Protection

System with the Help of Structure Function

Step 1: Analyze the signature of the structure–function by

 
Sm = 1
φ(h) − 1
φ(h) (16)
(m
m−n+1) (m
m−n )
h⊆[m] h⊆[m]
|h|=m−n+1 |h|=m−n

and compute a polynomial function of the digital protection system by

 

b
b
h(u) = Ca pa q b−a (17)
a=1
a

where,


b
Cf = S f a = 1, 2, ..., b
f =b−a+1

Step 2: Calculate the tail signature of the system S = (S0 , ......., Sm ) using


u 
Sn = se = 1
φ(h) (18)
(m
m−n )
e=n+1 |h|=m−n
322 S. Bisht and S. B. Singh

Step 3: With the help of Taylor expansion about y=1, we find the reliability function
in the form of a polynomial.

p(x) = x u h( x1 ) (19)

Step 4: Estimate the tail signature of ADPS reliability function with the help of
Eq. (18) by
(m−n)! n
Sn = n!
D p(1) , n = 0, 1, ..., m (20)

Step 5: Find the signature of the ADPS using Eq. (20).

s = S n−1 − S n , n = 1, ..., m (21)

1.7 Signature of ADPS

1.7.1 Signature of Structure Design 1

Using Eqs. (6) and (17), one can get the structure function of structure design 1 as:

h(u) = 2u 9 − 4u 8 − u 7 + 4u 6 (22)

Computing the tail signature (S S D1 ) and the signature (SS D1 ) of the structure
design 1 by Eqs. (20) and (21) respectively as:

2 11 1
S S D1 = 1, , , , 0, 0, 0, 0, 0, 0
3 36 21

1 13 195 1 287
s S D1 = , , , , , 0, 0, 0, 0, 0
3 36 756 21 4410

1.7.2 Signature of Structure Design 2

With the help of Eqs. (8) and (17), we get the structure function of structure design
2 as:

h(u) = −u 11 + 2u 10 + 4u 9 − 8u 8 − 4u 7 + 8u 6 (23)

The tail signature (S S D2 ) and the signature (SS D2 ) of the structure design 2 can
be computed by Eqs. (20) and (21) respectively as:
Universal Generating Function Approach for Evaluating Reliability and Signature … 323

10 36 56 6 4
S S D2 = 1, , , , , , 0, 0, 0, 0, 0, 0
11 55 165 55 231

1 14 2860 38 1166 4
s S D2 = , , , , , , 0, 0, 0, 0
11 55 9075 165 12705 231

1.7.3 Signature of Structure Design 3

With the help of Eq. (10) and using Eq. (17) we get the structure function h(u) from
RBD of structure design 3 as:

h(u) = u 12 − 4u 11 − 2u 7 + 4u 6 (24)

The tail signature (S S D3 ) and the signature (SS D3 ) of the structure design 3 can
be obtained by Eqs. (20) and (21) respectively as:

20 3 10 1 1
S S D3 = 1, 1, , , , , , 0, 0, 0, 0
33 11 99 36 231

13 11 17 261 195 1
s S D3 = 0, , , , , , , 0, 0, 0, 0, 0
33 33 99 3564 8316 231

1.7.4 Signature of Structure Design 4

Using Eqs. (12) and (17), we get the structure function of structure design 4 as:

h(u) = −u 12 + 2u 11 + 4u 10 − 8u 9 − 4u 8 + 8u 7 (25)

The tail signature (S S D4 ) and the signature (SS D4 ) of the structure design 4 can
be obtained by Eqs. (20) and (21) correspondingly as:

5 6 14 4 1
S S D4 = 1, , , , , , 0, 0, 0, 0, 0, 0, 0
6 11 55 55 99

1 19 16 10 341 1
s S D4 = , , , , , , 0, 0, 0, 0, 0, 0
6 66 55 55 5445 99
324 S. Bisht and S. B. Singh

1.7.5 Signature of Structure Design 5

With the help of Eqs. (14) and (17), we get the structure function of the structure
design 5 as:

h(u) = −u 18 + 2u 17 + 4u 15 − 8u 14 − 4u 12 + 8u 11 (26)

The tail signature (S S D5 ) and the signature (SS D5 ) of the structure design 5 can
be obtained by Eqs. (20) and (21) respectively as:

7 8 43 53 2 1 1
S S D5 = 1, , , , , , , , 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
9 17 204 765 119 357 3978

1 13 195 1 287
s S D5 = , , , , , 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
3 36 756 21 4410

1.8 Algorithm for Finding Expected Lifetime of the ADPS

with Minimal Signature

Step 1: Determine the MTTF of the ADPS which has i.i.d components with mean
μ.
Step 2: Evaluate the minimal signature of the ADPS with an expected lifetime of
reliability function by using


n
HT (w) = Cr H1:r (w) (27)
r =1

where,H1:r (w) = Pu (z 1:r > w) for r = 1,2, …, v

Step 3: Calculate the expected lifetime E (T ) of the ADPS which has i.i.d
components by


n
E(T ) = μ Cr
r
(28)
r =1

where,C = (C1 , C2 , . . . , Cv ) is a vector coefficient of minimal signature.

Universal Generating Function Approach for Evaluating Reliability and Signature … 325

1.9 Mean Time to Failure (Expected Lifetime)

of the All-Digital Protection System

1.9.1 MTTF of Structure Design 1

Using Eq. (24), the expected lifetime from the minimal signature of structure design
1 is obtained as:

h(u) = 2u 9 − 4u 8 − u 7 + 4u 6

The minimal signature of the structure design 1 (M SS D 1 ) with the help of Eq. (27)
is given by

MSSD1 = (0, 0, 0, 0, 0, 4, −1, −4, 2).

Hence, the expected lifetime of the structure design 1 is evaluated by Eq. (28) as:

E(T ) S D1 = 0.246031

1.9.2 MTTF of Structure Design 2

The expected lifetime from the minimal signature of structure design 2 is computed
by Eq. (24) as:

h(u) = −u 11 + 2u 10 + 4u 9 − 8u 8 − 4u 7 + 8u 6

The minimal signature of the structure design 2 (M SS D 2 ) with the help of Eq. (27)
is computed as

MSSD2 = (0, 0, 0, 0, 0, 8, −4, −8, 4, 2, −1).

Hence, the expected lifetime of the structure design 2 is estimated by Eq. (28) as:

E(T ) S D2 = 0.252453

1.9.3 MTTF of Structure Design 3

Using Eq. (18), the expected lifetime from the minimal signature of the structure
design 3 is expressed as:
326 S. Bisht and S. B. Singh

h(u) = u 12 − 4u 11 − 2u 7 + 4u 6

The minimal signature of the structure design 3 (M SS D 3 ) using Eq. (27) is

specified as.

MSSD3 = (0, 0, 0, 0, 0, 4, −2, 0, 0, 0, −4, 1).

Hence, the expected lifetime of the structure design 3 is calculated by Eq. (28) as:

E(T ) S D3 = 0.100649

1.9.4 MTTF of Structure Design 4

Using Eq. (18), the expected lifetime from the minimal signature of the structure
design 4 is obtained as:

h(u) = −u 12 + 2u 11 + 4u 10 − 8u 9 − 4u 8 + 8u 7

The minimal signature of the structure design 4 by Eq. (27) is computed as

MSSD4 = (0, 0, 0, 0, 0, 0, 8, −4, −8, 4, 2, − 1).

Hence, the expected lifetime of the structure design 4 is evaluated by Eq. (28) as:

E(T ) S D4 = 0.254310

1.9.5 MTTF of Structure Design 5

With the help of Eq. (18), the expected lifetime from the minimal signature of the
structure design 5 is expressed as:

h(u) = −u 18 + 2u 17 + 4u 15 − 8u 14 − 4u 12 + 8u 11

The minimal signature of the structure design 5 (M SS D 5 ) with the help of Eq. (27)
is determined as

MSSD5 = (0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8, −4, 0, −8, 4, 0, 2, −1).

Universal Generating Function Approach for Evaluating Reliability and Signature … 327

Hence, the mean time to failure of the structure design 5 is estimated by Eq. (28)
as:

E(T ) S D5 = 0.151268

2 Conclusion

The present paper analyzed the reliability of the all-digital protection system having
non-identically and identically distributed components. Here, we computed the reli-
ability of all the structure designs with the help of the UGF. Also, unlike done in
the past the signature reliability of the all-digital protection systems with the help of
Owens method has also been studied in this paper. The expected lifetime of digital
protection systems with the help of minimal signature is also studied for the first
time. In these systems, we have considered many redundant components to increase
the system reliability.
It is also revealed in the study that when the systems have non-identical compo-
nent, then the architecture 5 is found to be most reliable having value 0.715833175
while the structure design 2 is the second most reliable having value 0.6559.
The signature analysis has been done to examine the impact of the failure proba-
bilities of components in the architectures which will help the engineers and system
engineers in designing. MTTF of the all architectures with the help of minimal signa-
ture has also been obtained with respect to all five proposed aspects. We also found
that MTTF of structure design 2 is the highest while the structure design 3 has attained
the lowest value.

References

1. Boland PJ, El Neweihi E, Proschan F (1988) Active redundancy allocation in coherent systems.
Probab Eng Inf Sci 2(3):343–353
2. Chauhan U, Singh V, Rani A, Pahuja GL (2015) Ranking of all digital protection system compo-
nents using cost-based heuristic redundancy importance measure. In: International conference
on recent developments in control, automation and power engineering (RDCAPE), pp 141–145
3. Da G, Zheng B, Hu T (2012) On computing signatures of coherent systems. J Multivariate
Anal 103(1):142–150
4. Djekic Z, Portillo L, Kezunovic M (2008) Compatibility and interoperability evaluation of
all-digital protection systems based on IEC 61850–9–2 communication standard. In: Power
and energy society general meeting-conversion and delivery of electrical energy in the 21st
century. IEEE, pp 1–5
5. Kumar A, Singh SB (2017) Computations of the signature reliability of the coherent system.
Int J Qual Reliab Manage 34(6):785–797
6. Kumar A, Singh SB (2017) Signature reliability of sliding window coherent system.
In: Mathematics applied to engineering, pp 83–95
328 S. Bisht and S. B. Singh

7. Levitin G (2005) The universal generating function in reliability analysis and optimization, vol
6. Springer, London
8. Lisnianski A, Levitin G (2003) Multi-state system reliability: assessment, optimization, and
applications, vol 6. World Scientific Publishing Co Inc.
9. Marichal JL, Mathonet P (2013) Computing system signatures through reliability functions.
Stat Probab Lett 83(3):710–717
10. Navarro J, Rubio R (2009) Computations of signatures of coherent systems with five
components. Commun Stat-Simul Comput 39(1):68–84
11. Navarro J, Rychlik T (2007) Reliability and expectation bounds for coherent systems with
exchangeable components. J Multivariate Anal 98(1):102–113
12. Negi S, Singh SB (2015) Reliability analysis of the non-repairable complex system with
weighted subsystems connected in series. Appl Math Comput 262:79–89
13. Rausand M, Arnljot HÃ (2004) System reliability theory: models, statistical methods and
applications, vol 396. Wiley
14. Samaniego FJ (2007) System signatures and their applications in engineering reliability, vol
110. Springer Science & Business Media
15. Samaniego FJ (1985) On closure of the IFR class under formation of coherent systems. IEEE
Trans Reliab 34(1):69–72
16. Scheer GW, Dolezilek DJ (2000) Comparing the reliability of Ethernet network topologies in
substation control and monitoring networks. In: Western power delivery automation conference.
Spokane, Washington
17. Singh C, Patton AD (1980) Protection system reliability modeling: unreadiness probability and
mean duration of undetected faults. IEEE Trans Reliab 29(4):339–340
18. Ushakov I (1986) Universal generating function. J Comput Sci Syst 24:118–129
19. Ushakov IA (1994) Handbook of reliability engineering. Wiley
20. Wu JS, Chen RJ (1994) An algorithm for computing the reliability of weighted-k-out-of-n
systems. IEEE Trans Reliab 43(2):327–328
21. Zhang P, Portillo L, Kezunovic M (2006a) Compatibility and interoperability evaluation for
all-digital protection system through automatic application test. In: Power engineering society
general meeting. IEEE
Reliability Analysis of 8 × 8 SEN- Using
UGF Method

Vaibhav Bisht and S. B. Singh

Abstract The aim of this article is to assess the reliability of an 8 × 8 shuffle

exchange network with one stage less (SEN-). The reliability of the 8 × 8 SEN-
has been assessed using three criteria: terminal, broadcast, and network reliability.
The current paper offers an effective algorithm for calculating the reliability of an 8
× 8 SEN- using the universal generating function (UGF) method. With the help of
UGF, the reliability is estimated for both, the same and distinct switching element
probabilities. This article also investigates the impact of removing one stage from
the usual 8 × 8 shuffle exchange network (SEN).

Keywords Shuffle Exchange Network (SEN) · 8 × 8 SEN- · Universal Generating

Function (UGF) Method · Terminal reliability · Broadcast reliability and Network
reliability

Paper type Research paper

1 Introduction

In this modern era of communication technology, there is necessity to form a network,

in which a large number (which may be hundreds or even thousands) of proces-
sors are being linked. Various complications have been aroused in the field of
air traffic control, weather- forecasting, and even in the area of defence, which
can only be resolved by high amount of computational power. As a result, future
system performance will be able to significantly improve principles associated to a
parallel processing system. When the several processors are being connected with
the memory modules with the help of interconnected network, then this network is
called the interconnection network. It is worth mentioning that a whole computer
network system plays a key role to transmit signals/messages to the destinations
(Trivedi 2008). Furthermore, these networks depend upon two types of switching

V. Bisht (B) · S. B. Singh

Department of Mathematics, Statistics and Computer Science, G.B. Pant University of
Agriculture and Technology, Pantnagar, India

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 329
M. Ram and H. Pham (eds.), Reliability and Maintainability Assessment
of Industrial Systems, Springer Series in Reliability Engineering,
https://doi.org/10.1007/978-3-030-93623-5_15
330 V. Bisht and S. B. Singh

topologies for the transmission of signals/messages from source to the destination,

namely, packet switching and circuit switching. Switching in the circuit completely
depends on the physical paths that lie between the source and the sink.

2 Preliminaries

2.1 Interconnection Network

For over the recent ten years, we’ve gradually seen the development in the field of
interconnection networks which is now being injected into this contemporary age of
multiprocessor frameworks. This development is introducing new roads for devel-
opment and application in numerous fields. There are different components of a
computer system interconnected by communication networks, and these networks
are referred to as interconnection networks. An interconnection network is simply
the interconnection of various networks of connecting nodes. In the interconnec-
tion network, switching elements are used to connect the destination node from
the source node. The reliability of interconnection networks depends on the reli-
ability of switching element and their interconnection network. Direct and indi-
rect networks are the two types of interconnection networks. The network is sent
across multiple edges of the direct network via point-to-point connections with the
processing nodes. The router-based network or static network is another name for
the direct network. Direct networks include the star graph, hypercube, torus, mesh,
and the trees. Messages between any two separate nodes in an indirect network are
routed through the network’s switches. In parallel computing, indirect interconnec-
tion networks are widely employed for switching and routing dynamic nodes. The
communication of signals/messages in an indirect network is accomplished with the
help of switches [4, 16]. Negi and Singh [15] assessed the reliability of complex
systems and its subsystems, which were not repairable and was connected in series.
The main purpose behind the construction of interconnection network is that when
the single processor is not able to handle the task involving a huge amount of data,
then task is broken into different parallel tasks, which are performed simultaneously,
resulting into reduction of processing time. Hence, these interconnection network
plays very important part in constructing large parallel processing system. Intercon-
nection networks are widely used in many real-life applications such as telephone
switches, networks in industries, supercomputers and many more.

2.2 Multistage Interconnection Network (MIN)

Since the dynamic topology consists of numerous links that may be reconfigured
by setting with the movable switching components, the multistage interconnection
Reliability Analysis of 8 × 8 SEN- Using UGF Method 331

networks (MINs) play an essential role in the dynamic network system. MINs are
made up of several sheets of linked switching components organised in a prede-
termined architecture. The integrated circuits, telecommunication switches, multi-
processor systems and computer communications are all examples of uses for these
networks. Since these networks are of big size and have a complicated topology, it
is necessary to enhance the performance and hence to increase their reliability. It is
important to note that these switching elements (SEs) are often linked in different
number stages. Blocking, non-blocking and the rearrangeable non-blocking MINs are
three different types of MINs. Since it mixes with the different connections already
existing in the network, there is a very low likelihood of communication between
a free source/destination pair in blocking networks. In non-blocking networks,
messages are communicated from every input node to every output node without
affecting the network’s pre-defined topology, resulting in various pathways between
each source and destination node, which results extra stages in the system. Gunawan
[9] discussed the basics of reliability engineering, distribution of probability and
certain fundamentals of probability theory.
A multistage interconnection network (MIN) typically has m inputs and m outputs,
as well as m (=log2 m) stages and every stage has m/2 switching components, where m
is the size of network. Shuffle exchange network (SEN), SEN with extra stages, Benes
network, Gamma interconnection network (GIN), Extra-stage Gamma interconnec-
tion network GIN, Clos network, Omega network, Multistage cube network, and
many more multistage interconnection networks are frequently used. It is important to
note that network reliability is determined not only by the components in the network,
but also by the topology of the network. The failure of the MINs is mostly due to
inadequate network architectural selection and insufficient routing methods. Many
engineers and academics have proposed different methods to enhance the perfor-
mance of MINs and make them more dependable in the past. Trivedi [17] used the
continuous Markov chains technique to assess the dependability of MINs. Rajkumar
and Goyal [16] attempted to connect and investigate different MIN network topolo-
gies in terms of reliability, fault-tolerance, and cost-efficiency. Blake and Trivedi
[5] investigated the single path MIN’s dependability and focused on fault tolerant
schemes for improving network resilience. Both achieved reliability derivations for
8 × 8 shuffle exchange networks and 16 × 16 shuffle exchange networks. Bistouni
and Jahanshahi [3] proposed a new technique to increase the reliability and fault-
tolerance of the SEN by augmenting the switching stages. They both found out that
SEN with one extra stage (SEN + 1) was more reliable than SEN or SEN with two
more stages (SEN + 2). Fard and Gunawan [8] estimated the terminal reliability
of a modified SEN with 2 × 2 SEs at the intermediate stages, 1 × 2 at the source
nodes, and 2 × 1 at the terminal nodes, and compared it to the conventional shuffle
exchange network. Chinnaiah [7] proposed a new MIN called replicated SEN, which
he compared to the SENs and Benes networks. Bisht and Singh [2] calculated the
reliability of 4 × 4 SEN, SEN + 1 and SEN + 2 by UGF method and found that 4
× 4 SEN is most reliable and SEN + 2 is least reliable and reliability of SEN + 1
lies between these two networks.
332 V. Bisht and S. B. Singh

2.3 Universal Generating Function (UGF)

The reliability of complex networks and engineering systems is calculated using a

variety of approaches/methods [11, 14]. Because of its simplicity and time-saving
nature, UGF is one of the most commonly utilised techniques among them. Ushakov
provided the foundation for this technique [18]. Levitin and Lisnianski [13] proposed
a method for calculating the relevance of element dependability in different Multi-
State Systems (MSS) based on the UGF process. Levitin [12] extended the universal
generating function approach for multi-state system analysis to the case when the
performance distributions of some components are reliant on the states of another
element or the collection of elements. Bisht and Singh [1] proposed utilising UGF to
assess reliability indices such as signature reliability and MTTF of complex bridge
networks with lifetime components which are distributed independently and identi-
cally. Chacko [6] used the named UGF technique to quickly assess the reliability of
a group of continuous MSSs. The suggested novel UGF technique can swiftly assess
the reliability factors for a variety of structures.
The Universal Generating Function (UGF) method is a generalisation of the ordi-
nary generating function. The probability mass function of its variables is represented
by a polynomial version of the moment generating function. If there are m potential
values for k independent variables and rm is the probability of k equaling km , then
the polynomial gives the UGF of k as follows:


M
U (z) = r m z km (1)
m=1

Consider the discrete random variables Y1 , Y2 , . . . Y p which are all independent.

Let U1 (z), U2 (z), . . . , U p (z) be the UGF of the random variables Y1 , Y2 . . . Y p , and
f (Y1 , Y2 , . . . Y p ) be an arbitrary function. Furthermore, the composition operator is
used to describe the combination of r UGF, and the properties of the composition
operator are dependent on the properties of f (Y1 , Y2 , . . . . . . Y p ). As a result, Ur (z)
is written as follows:
 
Ur (Z ) = ⊗ f U1 (Z ), U2 (Z ), . . . U p (Z )

The UGF for two components in a network is given as:

For two components connected in series, the UGF is given by:
  K1 
K2 
k1 
k2
u 1 (z) ⊗ u 2 (z) = p1k 1 z g1k1 ⊗ p2k 2 z g2k2 = p1 k 1 p2 k2 z ser (g1 k1 ,g2 k2 )
ser ser
k1 =1 k2 =1 =1 k2 =1
(2)

The UGF for two components connected in parallel is given by:

Reliability Analysis of 8 × 8 SEN- Using UGF Method 333

  K1 
K2 
k1 
k2
u 1 (z) ⊗ u 2 (z) = p1k 1 z g1k1 ⊗ p2k 2 z g2k 2 = p1 k 1 p2 k2 z par (g1 k1 ,g2 k2 )
par par
k1 =1 k2 =1 =1 k2 =1
(3)

where, ⊗ and ⊗ signify series and parallel composition operators, respectively, in

ser par
the network.
Equations (4) and (5) may be used to calculate the structural functions for the
series and parallel systems, respectively as:

ϕser ( p1 , . . . . pn ) = min{ p1 , . . . . pn }, (4)

ϕ par ( p1 , . . . . pn ) = max{ p1 , . . . . pn }, (5)

where, p1 , . . . ., pn are the elements connected in the system.

3 Shuffle Exchange Network with One Stage Less (SEN-)

Figure 1 depicts a SEN- of size 8 × 8.

The 8 × 8 SEN- Multistage interconnection network (MIN) does not meet the
MIN’s primary criterion of full connection i.e., some of the source nodes are not
connected to every destination node. For example, the source 000 is connected to
outputs 000, 001, 010 and 011, but the same source 000 is not connected to output
nodes 100, 101, 110 and 111. Since by the main requirement of MIN, all input nodes

Fig. 1 The 8 × 8 SEN-

334 V. Bisht and S. B. Singh

should transmit networks to all destinations at least by one of its paths. As a result,
this network does not meet the requirements for MIN approval.

3.1 New SEN-

In New 8 × 8 SEN- structure the MUX has been used at source and DEMUX have
been used at destination end, so that the signal from each input can be transmitted
to each output node. For the sake of convenience, 2 × 1 MUX and 1 × 2 DEMUX
have been used to transmit network from one input to all outputs [10]. Bigger size of
MUX and DEMUX can be also used to make system more redundant. This novel 8 ×
8 SEN provides two pathways between each and every source and destination. The
primary benefit of this strategy is that it creates totally independent paths between
the source and destination, making the entire system fault resilient at both ends. Now,
there are m MUX and m DEMUX at both, source, and destination node in this 8 × 8
SEN-, for a total of log 2 m− 1 stages in this network and m/2 SEs per stage. Figure 2
depicts the new SEN of size 8 × 8.

4 Reliability Analysis of SEN-

With the aid of UGF, the reliability of this SEN- may be assessed based on the three
reliabilities: (1) terminal reliability (TR) (2) broadcast reliability (BR) (3) the network
reliability (NR). To calculate reliability, we assume the following assumptions:

Fig. 2 New 8 × 8 SEN-

Reliability Analysis of 8 × 8 SEN- Using UGF Method 335

• The reliability of the 2 × 2 switching element is to be taken to be p.

• The reliability of the 2 × 1 MUX and 1 × 2 DEMUX is taken to be p 1/2 .

4.1 Terminal Reliability

A network’s terminal reliability (TR) may be expounded as the possibility of

minimum one fault-less path between the network’s source and its destination pairs.
Figure 3 shows block diagram of terminal reliability for SEN- of size 8 × 8.
With the aid of UGF, the terminal reliability of SEN- may be determined as
follows:

RT R (S E N -) = max(min( p1 , p3 , p5 , p7 ), min( p2 , p4 , p6 , p8 ))

where, p1 , p2 ,….., p8 are the probabilities of the switching element existing in this
8 × 8 SEN-.
(1) If the SEN- elements are not the same and the probabilities of the network
components are distinct, the UGFs of different SEs are represented by:
 
u s j (z) = ps j z 1 + 1 − ps j z 0

where, ps j is the switch’s probability, j = 1, 2, . . . ..,8.

Assume that the network’s UGFs u s j (z) of the switches s j , j = 1, 2, . . . .., 8
respectively, are given as follows:

u s1 (z) = 0.99z 1 + 0.01z 0 u s2 (z) = 0.98z 1 + 0.02z 0

u s3 (z) = 0.97z 1 + 0.03z 0 u s4 (z) = 0.96z 1 + 0.04z 0

u s5 (z) = 0.95z 1 + 0.05z 0 u s6 (z) = 0.94z 1 + 0.04z 0

1 3 5 7

2 4 6 8

Fig. 3 TR of 8 × 8 SEN-
336 V. Bisht and S. B. Singh

u s7 (z) = 0.93z 1 + 0.03z 0 u s8 (z) = 0.92z 1 + 0.02z 0

We get UGFs as follows after applying composition operators to distinct SEs

based on their combination:

U A (z) = u s1 (z) ⊗ u s3 (z) ⊗ u s5 (z) ⊗ . . . u s7 (z)

min min min
= 0.84842505z + 0.15157495z 0
1

U B (z) = u s2 (z) ⊗ u s4 (z) ⊗ u s6 (z) ⊗ . . . u s8 (z)

min min min
= 0.81360384z + 0.18639616z 0
1

Finally, the reliability of all SEN- terminal structures is determined as follows:

U (z) = U A (z) ⊗ U B (z)

max

U (z) = 0.971747011z 1 + 0.028252988z 0

Hence, the terminal reliability (TR) of this 8 × 8 SEN- is calculated as:

TR (SEN−) = 0.971747011

(2) The structural function, when the SEN- elements are identical and all SEs have
the same probability is given by:

RT R (S E N −) = 2 p 3 − p 6

With the help of the suggested UGF technique, the TR of 8 × 8 SEN- is assessed
for distinct switching element reliability and being compared with 8 × 8 SEN, as
shown in Table 1.

Table 1 TR of 8 × 8 SEN-
Switching TR of 8 × 8 SEN- by TR of 8 × 8 SEN [16]
reliability UGF
0.90 0.926559 0.72900
0.95 0.979658 0.85737
0.96 0.986714 0.88473
0.99 0.999118 0.97029
Reliability Analysis of 8 × 8 SEN- Using UGF Method 337

4.2 Broadcast Reliability

The likelihood of a network/message being sent from a single source to all destination
pairs is said to be broadcast reliability (BR). Figure 4 shows the block diagram for
broadcast reliability for SEN-, which is of the size 8 × 8.
The SEN- broadcast reliability may be computed using the UGF technique as
follows:

R B R (SEN−) = max(min( p1 , p3 , . . . .. p17 ), min( p2 , p4 , . . . .. p18 ))

where, p1 , p2 ,….., p18 are the switching elements probabilities of these components
in the network.
(1) If all the SEN- elements are distinct and the probability of the components of
the network is not the same, then the UGFs of the various SEs are provided by:
 
u s j (z) = ps j z 1 + 1 − ps j z 0

where, ps j is the probability for the switch s j , j = 1, 2, . . . , 18.

Consider the network’s UGFs u s j (z) for switches s j , j = 1, 2, . . . , 18 as:

u s1 (z) = 0.99z 1 + 0.01z 0 u s2 (z) = 0.98z 1 + 0.02z 0

u s3 (z) = 0.97z 1 + 0.03z 0 u s4 (z) = 0.96z 1 + 0.04z 0

u s5 (z) = 0.95z 1 + 0.05z 0 u s6 (z) = 0.94z 1 + 0.06z 0

u s7 (z) = 0.93z 1 + 0.07z 0 u s8 (z) = 0.92z 1 + 0.08z 0

u s9 (z) = 0.91z 1 + 0.09z 0 u s10 (z) = 0.90z 1 + 0.10z 0

1 3 5 7 9 11 13 15 17

4 6 8 10 12 14 16
2 18

Fig. 4 BR of 8 × 8 SEN-
338 V. Bisht and S. B. Singh

u s11 (z) = 0.89z 1 + 0.11z 0 u s12 (z) = 0.88z 1 + 0.12z 0

u s13 (z) = 0.87z 1 + 0.13z 0 u s14 (z) = 0.86z 1 + 0.14z 0

u s15 (z) = 0.85z 1 + 0.15z 0 u s16 (z) = 0.84z 1 + 0.16z 0

u s17 (z) = 0.83z 1 + 0.17z 0 u s18 (z) = 0.82z 1 + 0.18z 0

After applying composition operators to different SEs based on their combination,

we get UGFs like this:

U A (z) = u s1 (z) ⊗ u s3 (z) ⊗ u s5 (z) ⊗ . . . u s17 (z)

min min min
= 0.421755886z + 0.578244114z 0
1

U B (z) = u s2 (z) ⊗ u s4 (z) ⊗ u s6 (z) ⊗ . . . u s18 (z)

min min min
= 0.38170668z + 0.61829332z 0
1

Finally, the broadcast reliability 8 × 8 SEN- network is given as:

U (z) = U A (z) ⊗ U B (z)

max

U (z) = 0.642475527z 1 + 0.357524473z 0

As a result, the 8 × 8 SEN- network’s broadcast reliability (BR) is given by:

BR (SEN−) = 0.642475527

(2) The structural function is given, when the SEN- components are alike and all
SEs have the same probability:

R B R (S E N −) = 2 p 13/2 − p 13

With the help of the suggested UGF technique, the BR of 8 × 8 SEN- is assessed
for distinct switching element reliability and being compared with 8 × 8 SEN, as
shown in Table 2.
Reliability Analysis of 8 × 8 SEN- Using UGF Method 339

Table 2 Broadcast reliability

Switching BR of 8 × 8 SEN- by BR of 8 × 8 SEN [16]
of 8 × 8 SEN-
reliability UGF
0.90 0.754152 0.478297
0.95 0.919616 0.698337
0.96 0.945684 0.751447
0.98 0.984856 0.868126
0.99 0.996000 0.932065

1 3 5 7 9 11 13 15 17 19 21 23

2 4 6 8 10 12 14 16 18 20 22 24

Fig. 5 NR of the 8 × 8 SEN-

4.3 Network Reliability

The likelihood of successful signal transmission from all source pairs to all sink
nodes is said to be network reliability (NR). Figure 5 shows the block diagram for
network reliability for SEN-, which is of the size 8 × 8.
The network reliability of SEN- may be estimated using the UGF technique as
follows:

R N R (SEN-) = max(min( p1 , p3 , . . . .. p23 ), min( p2 , p4 , . . . .. p24 ))

where, p1 , p2 ,….., p24 are the switching element’s probabilities present in this
network.
(1) When the SEN- elements are not the same and the probabilities of components
of the network are distinct, the UGFs for these different switching parts are
provided by:
 
u s j (z) = ps j z 1 + 1 − ps j z 0

where, ps j is the probability of the switch s j , j = 1, 2, . . . ..0.24.

Assume that the UGFs u s j (z) for this network for the switches s j , j = 1,2,…..,24
are as follows:

u s1 (z) = 0.99z 1 + 0.01z 0 u s2 (z) = 0.98z 1 + 0.02z 0

340 V. Bisht and S. B. Singh

u s3 (z) = 0.97z 1 + 0.03z 0 u s4 (z) = 0.96z 1 + 0.04z 0

u s5 (z) = 0.95z 1 + 0.05z 0 u s6 (z) = 0.94z 1 + 0.06z 0

u s7 (z) = 0.93z 1 + 0.07z 0 u s8 (z) = 0.92z 1 + 0.08z 0

u s9 (z) = 0.91z 1 + 0.09z 0 u s10 (z) = 0.90z 1 + 0.10z 0

u s11 (z) = 0.89z 1 + 0.11z 0 u s12 (z) = 0.88z 1 + 0.12z 0

u s13 (z) = 0.87z 1 + 0.13z 0 u s14 (z) = 0.86z 1 + 0.14z 0

u s15 (z) = 0.85z 1 + 0.15z 0 u s16 (z) = 0.84z 1 + 0.16z 0

u s17 (z) = 0.83z 1 + 0.17z 0 u s18 (z) = 0.82z 1 + 0.18z 0

u s19 (z) = 0.81z 1 + 0.19z 0 u s20 (z) = 0.80z 1 + 0.20z 0

u s21 (z) = 0.79z 1 + 0.21z 0 u s22 (z) = 0.78z 1 + 0.22z 0

u s23 (z) = 0.77z 1 + 0.23z 0 u s24 (z) = 0.76z 1 + 0.24z 0

We get UGFs as follows after applying composition operators to distinct SEs

based on their combination:

U A (z) = u s1 (z) ⊗ u s3 (z) ⊗ u s5 (z) ⊗ . . . u s23 (z)

min min min
= 0.207808825z + 0.792191174z 0
1

U B (z) = u s2 (z) ⊗ u s4 (z) ⊗ u s6 (z) ⊗ . . . u s24 (z)

min min min
= 0.181020576z 1 + 0.818979423z 0

Finally, the broadcast reliability 8 × 8 SEN- network is given as:

U (z) = U A (z) ⊗ U B (z)

max

U (z) = 0.351211727z 1 + 0.648788272z 0

Reliability Analysis of 8 × 8 SEN- Using UGF Method 341

Table 3 Network reliability

Switching NR of 8 × 8 SEN- by NR of 8 × 8 SEN [16]
of 8 × 8 SEN-
reliability UGF
0.90 0.675632 0.2824295
0.95 0.886714 0.540360
0.96 0.922376 0.612709
0.98 0.977728 0.7847147
0.99 0.994032 0.8863849

Hence, the network reliability (NR) of 8 × 8 SEN- is given by:

NR (SEN-) = 0.351211727

(2) When all of the components of 8 × 8 SEN- are same and the probability of all
switching elements is same, and the structural function is written as:

R N R (SEN-) = 2 p 8 − p 16

With the use of the suggested UGF technique, the NR of 8 × 8 SEN- is assessed
with regard to the distinct switching element reliability and being compared to NR
of 8 × 8 SEN and is shown in Table 3.

5 Conclusion

Reliability is a one of major concerns for most networks, particularly in the realm of
communication networks. This paper demonstrates how we may use the proposed
UGF to assess the reliability of 4 × 4 SEN-. This paper emphasizes the fact that 4
× 4 SEN- is not possible and use of MUX and DEMUX at source and destination
makes this network possible. This study examines the reliability of 4 × 4 SEN- from
three perspectives: TR, BR, and NR. A critical analysis of the findings of various
reliabilities derived from the suggested technique, followed by a comparison reveals
that the terminal reliability (TR), broadcast reliability (BR) and network reliability
(NR) of 4 × 4 SEN- is way higher than that of 8 × 8 SEN. Hence the 4 × 4 SEN- is
more reliable than 8 × 8 SEN.

References

1. Bisht S, Singh SB (2019) Signature reliability of binary state node in complex bridge networks
using universal generating function. Int J Qual Reliab Manage 36(2):186–201
2. Bisht V, Singh SB (2021) Reliability estimation of 4 × 4 SENs using UGF method. J Reliab
Stat Stud 173–198
342 V. Bisht and S. B. Singh

3. Bistouni F, Jahanshahi M (2014) Analyzing the reliability of shuffle-exchange networks using

reliability block diagrams. Reliab Eng Syst Saf 132:97–106
4. Bistouni F, Jahanshahi M (2014) Improved extra group network: a new fault-tolerant multistage
interconnection network. The J Supercomput 69(1):161–199
5. Blake JT, Trivedi KS (1989) Multistage interconnection network reliability. IEEE Trans Comput
38(11):1600–1604
6. Chacko VM (2018) Continuous multistate system universal generating function. Reliab: Theory
Appl 13(4):51
7. Chinnaiah B (2019) Multistage interconnection networks in reliability shuffle exchange
networks. In: First international conference on artificial intelligence and cognitive computing.
Springer, Singapore, pp 571–581
8. Fard NS, Gunawan I (2005) Terminal reliability improvement of shuffle-exchange network
systems. Int J Reliab, Qual Saf Eng 12(01):51–60
9. Gunawan I (2014) Fundamentals of reliability engineering: applications in multistage inter-
connection networks. Wiley
10. Gupta S, Pahuja GL (2016) A new SEN minus: design and reliability measures. Int J Reliab,
Qual SafEng 23(04):1650012
11. Kumar A, Singh SB (2019) Signature A-within-B-from-D/G sliding window system. Int J
Math, Eng Manage Sci 4(1):95–107
12. Levitin G (2004) A universal generating function approach for the analysis of multi-state
systems with dependent elements. ReliabEng Syst Saf 84(3):285–292
13. Levitin G, Lisnianski A (1999) Importance and sensitivity analysis of multi-state systems using
the universal generating function method. ReliabEng Syst Saf 65(3):271–282
14. Singh SB (2017) Reliability analysis of multi-state complex system having two multi-state
subsystems under uncertainty. J Reliab Stat Stud 161–177
15. Negi S, Singh SB (2015) Reliability analysis of non-repairable complex system with weighted
subsystems connected in series. Appl Math Comput 262:79–89
16. Rajkumar S, Goyal NK (2016) Review of multistage interconnection networks reliability and
fault-tolerance. IETE Tech Rev 33(3):223–230
17. Trivedi KS (2008) Probability & statistics with reliability, queuing and computer science
applications. Wiley
18. Ushakov IA (1986) A universal generating function. Soviet J Comput Syst Sci 24(5):118–129