×

Fault diagnosis. Models, artificial intelligence, applications. (English) Zbl 1074.93004

Berlin: Springer (ISBN 3-540-40767-7/hbk). xxix, 920 p. (2004).
[The articles of this volume will not be indexed individually.]
This book presents descriptions of modern diagnosis methods which are based on control engineering. To the area of such methods belong control theoretical methods, artificial neural networks in fault diagnosis, genetic algorithms. So the goal of the book is to present in a coherent way various research methods such as control engineering, artificial intelligence, machinery and device design and diagnostics of industrial processes. In the book are described the most important models which are applied in diagnostics. Among them the analytical models such as physical models, input-output-type linear models, state linear equations, and state observers which use Kalman filters. For fault detection, beside analytical models, fuzzy and neural models which provide information redundancy and guarantee the reliability of diagnostic systems are frequently used. The process diagnostic methodology is based on the calculation of residuals, fault detection, isolation and identification. Fault detection extracts symptoms which contain information on faults. Physical models are often nonlinear. The nonlinear models reflect static and dynamic properties of the system in the whole range of operation, while linear models can be used only in a neighborhood of the nominal point of operation. But generation of models for calculating residuals on the ground of physical equations is extremely difficult and sometimes impossible. So, it is very important to elaborate methods which overcome the difficulties with the use of physical equations.
In the book are described methods of signal analysis which are widely used in control processes and which by analogy are used in fault diagnosis. As examples of such methods let us mention analogue-to-digital conversion of signals, filtering to avoid noise, smoothing, analysis using the wavelet transform.
The eigenvalue analysis of complex systems, which is represented in the book, is one of the perspective methods of fault diagnosis. This method may be employed for diagnosis of systems which are described by tens thousands of equations.
For fault diagnosis, minimax optimization of system input-output relations may be used. Detection, isolation and identification of faults in system design may be considered via optimal filtering in the \(H_\infty\) sense which is reduced to solving one discrete-time algebraic Riccati equation. In searching for an optimal design method of residual-vector generators, a problem of disturbance decoupling and the use of a dual model of the monitored dynamic object are solved.
As was mentioned before, the application of analytical models is usually limited by linear systems or systems in which the linearization process causes small errors. In cases when there are no mathematical models or the complexity of the model makes its implementation impossible, artificial intelligence techniques based on knowledge of input-output information become an alternative tool for the construction of the desired models, and such an approach is described in the reviewed book. For example, the authors present genetic algorithms, evolutionary search with soft selection, symptom extraction such as the choice of a gain matrix for a robust nonlinear observer via genetic programming, design of the robust residual generator, the use of multiobjective optimization and evolutionary algorithms, the employment of evolutionary algorithms in the design of neural models. Artificial neural networks provide an alternative solution to the classical methods such as Luenberger observers or Kalman filters and can model any nonlinear process with high accuracy. A neural network examines a residual evolution and gives a fault classification signal to declare whether the system is faulty or not.
In the book applications of fuzzy logic to diagnostics which use expert knowledge are considered. A vital advantage of fuzzy techniques is the ability of modeling nonlinear systems and using learning methods, initially developed for neural networks. The large growth of fuzzy rules when the number of inputs grows, may be overcome by means of special methods of aggregation of variables. In the book genetic algorithms in multiobjective optimization of fault detection observers are described. Such algorithms are very useful and effective because of an analogy with biological systems, their ability to re-generate, perform self-control and reproduce as well as to adapt to the changeable conditions of existence. Such methods like minimal-distance methods are very effective. They are based on the concept of distance in functional space, which model the symptom space and permits the employment of functional analysis methods, for example, the contraction mapping principle for estimation of the iteration processes convergence.
Expert systems are described. The expert systems are computer programs designed for the use of professional knowledge and experience. Such programs are able to apply knowledge acquired from a team of specialists.
An application consists in the detection and location of leaks in transport pipelines. Large pipeline network are widely used to transport oil and gases. It is known that the occurrence of a leak is accompanined with a sudden pressure drop which develops at the leak location resulting in a rapid re-pressurization wave. It contains the information about the leak and may be used for fault detection in a pipeline.
The book contains most of the modern control methods which are used in fault diagnosis and wide bibliographical information.
This book is a good introduction to fault diagnosis and may be very useful for students, post graduate students, engineers and scientists who deal with industrial control systems to guarantee their safety.

MSC:

93-02 Research exposition (monographs, survey articles) pertaining to systems and control theory
00B15 Collections of articles of miscellaneous specific interest
93-06 Proceedings, conferences, collections, etc. pertaining to systems and control theory
93A30 Mathematical modelling of systems (MSC2010)
93C42 Fuzzy control/observation systems
93B51 Design techniques (robust design, computer-aided design, etc.)
68T05 Learning and adaptive systems in artificial intelligence
68T20 Problem solving in the context of artificial intelligence (heuristics, search strategies, etc.)
68T35 Theory of languages and software systems (knowledge-based systems, expert systems, etc.) for artificial intelligence
90B25 Reliability, availability, maintenance, inspection in operations research
92B20 Neural networks for/in biological studies, artificial life and related topics
PDFBibTeX XMLCite