![]() |
Fusion of Neural Networks, Fuzzy Systems and Genetic Algorithms: Industrial Applications
by Lakhmi C. Jain; N.M. Martin CRC Press, CRC Press LLC ISBN: 0849398045 Pub Date: 11/01/98 |
Previous | Table of Contents | Next |
B. Köppen-Seliger and P. M. Frank
Gerhard-Mercator-Universität - GH Duisburg
FB 9, Meβ- und Regelungstechnik
Bismarckstr. 81, BB
D-47048 Duisburg, Germany
This chapter introduces advanced supervision concepts and presents examples showing how fuzzy logic and neural networks can be applied in model-based fault diagnosis.
Emphasis is placed on a combined quantitative/knowledge-based concept incorporating fuzzy logic for residual evaluation and the application of certain types of neural networks for residual generation and evaluation. Realistic simulation studies at a wastewater plant and an actuator benchmark prove the applicability of the proposed schemes.
Due to the increasing complexity of modern control systems and the growing demand for quality, cost efficiency, availability, reliability, and safety, the call for fault tolerance in automatic control systems is gaining more and more importance. Fault tolerance can be achieved by either passive or active strategies. The passive approach makes use of robust control techniques to ensure that the closed-loop system becomes insensitive with respect to faults. In contrast, the active approach provides fault accommodation, i.e., the reconfiguration of the control system when a fault has occurred. While robust control can tolerate small faults to a certain degree, the reconfiguration concept is absolutely inevitable when serious faults occur that would lead to a failure of the whole system.
In order to accomplish fault accommodation a number of tasks have to be performed. One of the most important and difficult of these tasks is the early diagnosis of the faults. Besides this, fault diagnosis is needed as part of the supervision of complex control systems that incorporate artificial intelligence.
Fault diagnosis has thus become an important issue in modern automatic control theory and, during the last two and a half decades, an immense deal of research was done in this field, resulting in a great variety of different methods with increasing acceptance in practice. The core of the fault diagnosis methodology is the so-called model-based approach (see also Section 2). For literature on model-based techniques, the reader is referred to comprehensive survey papers as, for example, [5], [6], [7], [12], [35] and [43] or the book by [34].
In the case of fault diagnosis in complex systems, one is faced with the problem that no, or insufficiently accurate, mathematical models are available. The use of knowledge-model-based or data-model-based techniques, either in the framework of diagnosis expert systems or in combination with a human expert, is then the only feasible way to proceed.
This contribution provides a combined analytical- and knowledge-based approach applying fuzzy logic and a data-model-based approach based on neural networks as system models and pattern classifiers. Application results from realistic simulation studies at a wastewater plant and an industrial actuator benchmark problem are given to illustrate the different methods.
A permanent goal in operating technical processes is to ensure safety and reliability due to the general aim of increasing economic efficiency. This gives rise to the current demand for modern supervision concepts basically based on fault diagnosis schemes. In this context the term fault incorporates any kind of malfunctioning up to a complete failure of a system component, actuator, or sensor. The aim of fault diagnosis is to detect the faults of interest and their causes early enough so that failure of the overall system can be avoided.
The faults can be commonly described as additional inputs whose time of occurrence and size is unknown. In addition there is always modeling uncertainty due to unmodeled disturbances, noise, and model mismatch. This may not be critical for the process behaviour but may obscure the fault detection by raising false alarms.
Figure 1 General scheme for fault diagnosis.
The basic tasks of fault diagnosis are to detect and isolate faults and to provide information about their size and source. This has to be done on-line in the face of the existing unknown inputs and without, or with only very few, false alarms. As a result the overall concept of fault diagnosis consists of the three subtasks; fault detection, fault isolation, and fault analysis.
For the practical implementation of fault diagnosis, the following three steps are usually performed (see Figure 1) :
Residual generation
Signals, the so-called residuals, are generated which reflect the faults. The residuals should ideally be zero in the fault-free case and deviate from zero in case of an occurrence of a fault. In order to isolate different faults, properly structured residuals or directed residual vectors are needed.
Residual evaluation
Subsequent to residual generation, residual evaluation follows, with the goal of fault detection and, if possible, fault isolation. In a classification process a decision on the time of occurrence and the location of the possible fault is made.
Fault analysis
In this step the fault and its effect, as well as its cause, are analyzed.
In this chapter methods for the first two steps of residual generation and evaluation are discussed.
The commonly known approaches for residual generation can basically be divided into two categories of signal-based and model-based concepts with a further subdivision as shown in Figure 2. The main research emphasis of the last two decades has been placed on the development of model-based approaches starting from analytical models and leading to the recently employed data-based models, such as neural networks.
Residual evaluation techniques can be principally divided into threshold decisions, statistical methods, and classification approaches (Figure 3). The methods of fuzzy and neural classification will be introduced and applied in the following sections.
Figure 2 Classification of different residual generation concepts
Figure 3 Classification of different residual evaluation concepts
Previous | Table of Contents | Next |