Table Of ContentAdvanced Information and Knowledge Processing
Series Editors
Professor Lakhmi Jain
[email protected]
Professor Xindong Wu
[email protected]
Also in this series
Gregoris Mentzas, Dimitris Apostolou, Andreas Abecker and Ron Young
Knowledge Asset Management
1-85233-583-1
Michalis Vazirgiannis, Maria Halkidi and Dimitrios Gunopulos
Uncertainty Handling and Quality Assessment in Data Mining
1-85233-655-2
Asunción Gómez-Pérez, Mariano Fernández-López and Oscar Corcho
Ontological Engineering
1-85233-551-3
Arno Scharl (Ed.)
Environmental Online Communication
1-85233-783-4
Shichao Zhang, Chengqi Zhang and Xindong Wu
Knowledge Discovery in Multiple Databases
1-85233-703-6
Jason T.L. Wang, Mohammed J. Zaki, Hannu T.T. Toivonen and Dennis Shasha (Eds)
Data Mining in Bioinformatics
1-85233-671-4
C.C. Ko, Ben M. Chen and Jianping Chen
Creating Web-based Laboratories
1-85233-837-7
Manuel Graña, Richard Duro, Alicia d’Anjou and Paul P. Wang (Eds)
Information Processing with Evolutionary Algorithms
1-85233-886-0
Colin Fyfe
Hebbian Learning and Negative Feedback Networks
1-85233-883-0
Yun-Heh Chen-Burger and Dave Robertson
Automating Business Modelling
1-85233-835-0
Dirk Husmeier, Richard Dybowski and Stephen Roberts (Eds)
Probabilistic Modeling in Bioinformatics and Medical Informatics
1-85233-778-8
Ajith Abraham, Lakhmi Jain and Robert Goldberg (Eds)
Evolutionary Multiobjective Optimization
1-85233-787-7
K.C. Tan, E.F. Khor and T.H. Lee
Multiobjective Evolutionary Algorithms and Applications
1-85233-836-9
Nikhil R. Pal and Lakhmi Jain (Eds)
Advanced Techniques in Knowledge Discovery and Data Mining
1-85233-867-9
Amit Konar and Lakhmi Jain
Cognitive Engineering
1-85233-975-6
Miroslav Kárny´ (Ed.)
Optimized Bayesian Dynamic Advising
1-85233-928-4
Yannis Manolopoulos, Alexandros Nanopoulos, Apostolos N. Papadopoulos and
Yannis Theodoridis
R-trees: Theory and Applications
1-85233-977-2
Sanghamitra Bandyopadhyay, Ujjwal Maulik, Lawrence B. Holder and Diane J. Cook (Eds)
Advanced Methods for Knowledge Discovery from Complex Data
1-85233-989-6
Marcus A. Maloof (Ed.)
Machine Learning and Data Mining for Computer Security
1-84628-029-X
Sifeng Liu and Yi Lin
Grey Information
1-85233-995-0
Vasile Palade, Cosmin Danut Bocaniala
and Lakhmi Jain (Eds)
Computational
Intelligence in
Fault Diagnosis
With 154 Figures
Vasile Palade,PhD Lakhmi Jain,PhD
Oxford University Computing Laboratory KES Center
Oxford University of South Australia
UK Australia
Cosmin Danut Bocaniala,Phd
Department of Communication Systems
Lancaster University
Lancaster
UK
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library ofCongress Control Number:2006922573
Advanced Information and Knowledge Processing ISSN 1610-3947
ISBN-10:1-84628-343-4 Printed on acid-free paper
ISBN-13:978-1-84628-343-7
© Springer-Verlag London Limited 2006
Apart from any fair dealing for the purposes ofresearch or private study,or criticism or review,
as permitted under the Copyright,Designs and Patents Act 1988,this publication may only be
reproduced,stored or transmitted,in any form or by any means,with the prior permission in
writing of the publishers,or in the case of reprographic reproduction in accordance with the
terms oflicences issued by the Copyright Licensing Agency.Enquiries concerning reproduction
outside those terms should be sent to the publishers.
The use of registered names, trademarks, etc. in this publication does not imply, even in the
absence of a specific statement, that such names are exempt from the relevant laws and
regulations and therefore free for general use.
The publisher makes no representation,express or implied,with regard to the accuracy of the
information contained in this book and cannot accept any legal responsibility or liability for any
errors or omissions that may be made.
Printed in the United States ofAmerica (MVY)
9 8 7 6 5 4 3 2 1
Springer Science+Business Media
springer.com
Contributors
Viorel Ariton
“Danubius” University of Galati
Lunca Siretului no. 3, 800416
Galati, Romania
Email: [email protected]
Cosmin Danut Bocaniala
Computer Science and Engineering Department
“Dunarea de Jos” University of Galati
Domneasca 47, Galati, Romania
Email: [email protected]
João Calado
IDMEC/ISEL, Polytechnic Institute of Lisbon
Mechanical Engineering Studies Centre
Rua Conselheiro Emídio Navarro, 1950-062
Lisbon, Portugal
Email: [email protected]
Kok Yeng Chen
School of Electrical and Electronic Engineering
University of Science Malaysia
Engineering Campus, 14300
Nibong Tebal, Penang, Malaysia
Florin Ionescu
Department of Mechatronics
University of Applied Sciences in Konstanz
Brauneggerstraße 55, 78462 – Konstanz, Germany
Email: [email protected]
Weng Kin Lai
MIMOS Berhad
Technology Park Malaysia
57000 Kuala Lumpur, Malaysia
vi V Palade, CD Bocaniala and L Jain (Eds.)
Chee Peng Lim
School of Electrical and Electronic Engineering
University of Science Malaysia
Engineering Campus, 14300
Nibong Tebal, Penang, Malaysia
Email: [email protected]
Ar(cid:460)nas Lipnickas
Kaunas University of Technology
Department of Control Technology
Student(cid:464) 48-317, Kaunas LT-51367, Lithuania
Email: [email protected]
Luca Marinai
Department of Power, Propulsion & Aerospace Engineering
Cranfield University
Beds. MK43 OAL, United Kingdom
Email: [email protected]
Luis Mendonça
Technical University of Lisbon
Dept. of Mechanical Engineering, GCAR/IDMEC
Pav. Eng. Mecânica III, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
Email: [email protected]
Stephen Ogaji
Department of Power, Propulsion and Aerospace Engineering
School of Engineering
Cranfield University
Beds. MK43 OAL, United Kingdom
E-mail: [email protected]
Vasile Palade
Oxford University
Computing Laboratory
Wolfson Building, Parks Road
Oxford, OX1 3QD, United Kingdom
Email: [email protected]
José Sá da Costa
Technical University of Lisbon
Department of Mechanical Engineering, GCAR/IDMEC
Pav. Eng. Mecânica III, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
Email: [email protected]
Computational Intelligence in Fault Diagnosis vii
Riti Singh
Department of Power, Propulsion and Aerospace Engineering
School of Engineering
Cranfield University
Beds. MK43 OAL, United Kingdom
João Sousa
Technical University of Lisbon
Dept. of Mechanical Engineering, GCAR/IDMEC
Pav. Eng. Mecânica III, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
Email: [email protected]
Dan Stefanoiu
Department of Automatic Control and Computer Science
“Politehnica” University of Bucharest
313 Splaiul Independen(cid:288)ei, 060042–Bucharest, Romania
Email: [email protected]
Foreword
With the increased complexity of industrial machines and processes, the task of
fault diagnosis is becoming increasingly difficult and its complexity almost
unmanageable using conventional techniques. Therefore, in the past decade, intense
research was dedicated to find alternative solutions using methods that mirror
human reasoning as well as involve complex problem solving techniques inspired
from nature, to cope with the need for adaptation of the diagnostic methodology to
the inherent changes occurring in the diagnosed process.
The automatic diagnosis requires the ability to identify the symptoms
automatically and map them to their causes as well as, eventually, to prescribe
solutions for repairing/restoring the good functionality of the device, machine or
plant. Some methods can prove suitable for certain systems while being totally
inappropriate for others.
Computational intelligence attempts to emulate human and biological
reasoning, decision-making, learning and optimization via a series of techniques
that mirror the adaptive evolutionary nature of living beings. Such techniques can
be either used individually or combined into more complex hybrid methodologies,
resulting in systems with enhanced capabilities, e.g., the same system can benefit
from the decision-making under uncertainty enabled by fuzzy logic as well as from
learning and adaptation that neural networks provide, or from the evolutionary
optimization inherent in genetic algorithms.
Since the early 1990s, attempts to apply various computational intelligence
methods to fault diagnosis, sometimes used to augment traditional methods, were
made mainly in research laboratories. Given their success, these are now moving
into industrial settings. Big companies such as Siemens and ABB have embraced
such novel technologies very early.
Most successful attempts proved that fault diagnosis can greatly benefit
from computational intelligence techniques. Neural networks can ease fault
identification through model matching and learning of new symptoms. Fuzzy logic
can improve the diagnostic decision-making under the uncertainty inherent in the
diagnostic information: vague symptoms, ambiguous mapping of symptoms to their
causes as well as capturing the gradual degradation of systems and processes in
appropriate (fuzzy) models. Genetic algorithms are capable of optimizing the
diagnostic models as well as the diagnostic process itself by tracking the
(sometimes gradual) changes occurring in the diagnosed system in various ways.
We welcome this new book for offering us a very good overview of the
state of the art in the development of computational intelligence techniques
pertaining to fault diagnosis. Covering all computational intelligence techniques
both in theory as well as illustrating how they work by clear examples and/or
x V Palade, CD Bocaniala and L Jain (Eds.)
practical applications on a relatively broad range of problems, the book gradually
exposes the reader to these various methods in its eleven chapters.
Structurally, the book is a comprehensive collection of works arranged in a
progressive manner, to ease the gradual grasping of concepts. Starting with a very
good overview of computational intelligence and its suitability to the difficult task
of fault diagnosis, in Chapter 1, it continues (in Chapters 2 to 5) with four
applications involving fuzzy logic to solve various real-world diagnosis problems,
then Chapters 6 and 7 illustrate successful neural network-based diagnostic models,
to progress in Chapter 8 to a generic computational intelligence approach. Hybrid
neuro-fuzzy diagnostic approaches are further illustrated in Chapters 9 and 10. The
last chapter presents a novel distributed causal model for diagnosing complex
systems.
Overall, I salute this work for marking the progress made in this significant
area of fault diagnosis, which can be very useful to a broad audience, ranging from
industrial users to graduate students. Enabling the use of these techniques in
industrial applications as well as for training and teaching purposes, the book can be
regarded as both a repository of knowledge for practitioners and a basis for a course
on computational intelligence in diagnosis.
Professor Mihaela Ulieru,
Canada Research Chair
Preface
In one of his recent commentaries, called “Integration automation”, Mark Venables,
editor of the IEE Manufacturing Engineer Journal, predicts that “there are five
technologies that will drive the future of industrial automation. These are control
and diagnosis, communication, software, electronics, and materials – with the
former trio being the most important” (http://www.iee.org/oncomms/
sector/computing/commentary.cfm). Indeed, one of the main current trends in
solving problems in manufacturing industry is developing fault-tolerant control
schemes. Fault-tolerant control is concerned with making the controlled system able
to maintain control objectives, despite the occurrence of a fault. Hence, fault
diagnosis represents the main ingredient of a fault-tolerant control system.
Diagnosing the faults that occurred in a system permits triggering control
mechanisms to keep a plant working sufficiently well until the necessary
maintenance may be performed. In practice, this feature results in a significant
improvement in industrial plant safety, productivity and time in service.
There are two main categories of fault diagnosis techniques currently in
use and each has its own basic support theory. The first class of methodologies used
for fault diagnosis-related problems were based on mathematical models of the
monitored plant. The differences between the plant model and its actual behaviour
are called residuals and form the basis for deciding if a fault did or did not occur;
and if a fault has occurred, deciding which particular fault occurred. Unfortunately,
these techniques provide satisfactory results only when plants exhibit linear
behaviour or when the modelling errors can be kept within acceptable limits.
Accurate mathematical models can be obtained only for plants with low behavioural
complexity.
Recent research efforts have concentrated on finding suitable techniques to
model plants with high nonlinear behaviour, noise and uncertainty. These three
characteristics have been successfully mastered by using computational intelligence
methodologies. These solutions are based on models such as fuzzy systems, neural
networks, and genetic algorithms, to name only the most important of them. The
above methods are commonly combined to give the desired result. Besides using
residuals for diagnosis purposes, the computational intelligence methods may also
be used to directly map the sensor measurements to the faults’ space. These
methods allow an understanding of plant behaviour using rules obtained directly
from sensor measurements. However, even if these techniques can solve the
difficult problems posed by nonlinearity, noise and uncertainty, if the complexity of
the plant behaviour is very high, the computational load becomes too large for
practical purposes.
