Fault Diagnosis

“Until quite lately the term diagnostics was invariably associated with medicine as its field concerning the ways of disease recognition on the basis of symptoms. The term is derived from the Greek word diagnosis, which means recognition, whereas diagnostikós represents the ability to recognize. During the last ten years technical development has caused, on the one hand, the growth of the complexity of technical means, and on the other, the growth of the responsibility of tasks that are carried out with the use of these means. Nowadays, we are witnesses of the establishment and development of a new knowledge domain — technical diagnostics, which arises as an object of the demand of the users of these complex technical means. The goal of this new domain is to determine the broadly understood technical state of objects with the use of objective methods and means.” Such a definition of technical diagnostics and its goals was included in the monograph by Cempel published in 1982.

Such a definition of technical diagnostics and its goals was included in the monograph by Cempel published in 1982.

Due to the existence of various classes of diagnosed systems, different kinds of models are used in diagnostics studies which are being developed. A general description of a system that takes the effects of faults into account is usually impossible, and even if such a description exists, the dependence that characterises particular faults cannot be defined on the grounds of it. Therefore, different kinds of simplified models are applied in diagnostics. The most important of them are described in the present chapter. Models used for fault detection as well as models applied to fault isolation or system state recognition are singled out. Among models used for fault detection, the most important are analytical, neural, and fuzzy ones. A great variety of models are applied to fault isolation or system state recognition. These models define the relationship that exists between diagnostic signals (symptoms) and faults or system states. Models that map binary, multi-value or continuous diagnostic signals into the space of system faults or states are described.

This chapter is an introduction to the problems of the diagnostics of processes or systems. A general methodology of system diagnostics is presented by the description of such vital elements as fault detection, isolation and identification as well as the monitoring of system states. While diagnosing the system, it is necessary to ensure adequate distinguishability of its faults or states. Problems associated with the evaluation of faults (states) distinguishability as well as the methods of choosing a detection algorithm set that ensures higher distinguishability are presented. The chapter should give the reader some understanding of different diagnostic methods and create a base for studying particular problems being the subject matter of the following chapters.

Information obtained as a result of observing objects that are subjects of a diagnostic process is a basis for inference in technical diagnostics. Depending on the kind of diagnostic observations considered, information can deal with physical quantities that are connected with the object operation (e.g., the flow intensity of a medium in a suction connector). Information can be also related to residual processes that are effects of each object operation (e.g., the level of acoustic emission during the operation of a cutting tool). To generalise the numerous kinds of observations, one can consider a signal (diagnostic signal) as a material carrier that makes it possible to transmit information about the observed object or process. In most cases this carrier is a set of any quantities. The application of information included in the signal requires its appropriate description. Signals may be effectively described by sets of values of features that are results of signal analysis. Examples of features of a stochastic signal can be the estimates of that process.

Methodologies of the technical diagnostics of dynarnic processes, represented by the acronym FDI referring to Fault Detection and Isolation (Chen and Patton, 1999; Gertler, 1995; 1998; Isermann, 1984; Frank, 1990; Patton and Chen, 1993; Willsky, 1976), commonly concern three principal diagnostic operations (performed in parallel or in sequence): detection (the discovery that something has gone wrong in the monitored/supervised process), isolation (the differentiation, separation, localisation of a fault, leakage, bias, etc., or the indication of a faulty element) and the identification of a fault (the determination of the value or magnitude of this error).

Analytical methods of detecting faults and failures constitute a most essential branch of current approaches to the technical diagnostics of dynamic processes (objects). A major difficulty encountered while solving the tasks of the synthesis of detection algorithms concerns the effective determination of multiple residues, referred to as residual vectors (Chow and Willsky, 1984; Gertler, 1998; Frank, 1990). Such vectors, computed by properly weighting the errors of process output estimates, which leads to a suitable exposition of the symptoms of faults, are principal premises for making diagnostic decisions (Chen and Patton, 1999; Chow and Willsky, 1984; Magni and Mouyon, 1994). Practical generators of residual vectors can be based, for instance, on detection observers of states supplying appropriate estimates of the internal states of the process. Certainly, apart from fulfilling their detective task, such observers should be both robust to the uncertainty of the model of the process under supervision and insensitive to the influence of unmeasurable disturbances and unmeasurable noise affecting the object and its measurement channels (Chen and Patton, 1999; Kowalczuk and Suchomski, 1998). In practice, these tasks should be fulfilled to a possibly high degree. The issue of the optimisation of residual signals generators can thus be accordingly expressed as a general problem of multi-objective optimisation (Chen and Patton, 1999; Kowalczuk and Białaszewski, 2000; Kowalczuk and Suchomski, 1999, Kowalczuk et al., 1999).

The entirety of fundamental engineering issues (Chen and Patton, 1999a; Frank, 1990; Gertler, 1998; Isermann, 1984; Patton and Chen, 1993; Wilsky, 1976) is identified with the technical diagnostics of dynamic plants and processes. This domain encompasses three basic subdomains of diagnostic procedures known as the detection, isolation and identification of faults. Therefore, they are referred to as FDI (Fault Detection and Isolation).

One of the most important methodologies of designing Fault Detection and Isolation (FDI) systems is based on the model of a diagnosed system. In a general case, this concept can be realized using different types of analytical, knowledge-based, neural or fuzzy logic-based models (Köppen-Selinger and Frank, 1999). Models are constructed for systems working both in nominal and faulty conditions. Conventional fault diagnosis systems use analytical models, e.g., Luenberg observers or Kalman filters (Chen and Patton, 1999). The system’s dynamical properties are described by a set of differential equations or mapping functions with a suitable set of parameters.

In engineering research and design, evolutionary algorithms have found an increasing number of applications (Chen et al., 1996; Fogarty and Bull, 1995; Grefenstette, 1985; Huang and Wang, 1997; Kirstinsson, 1992; Kozieł and Kordalski, 1996; Kowalczuk et al., 1999a; Li et al., 1997; Linkens and Nyongensa, 1995; Man et al., 1997; Martinez et al., 1996; Park and Kandel, 1994; Sannomiya and Tatemura, 1996; Tanaka et al., 1996; Tang et al., 1996). The significance of such optimisation methods, which emulate the evolution of biological systems, is proven by their great usefulness and effectiveness. The well-known features of biological systems are their ability to re-generate, perform self-control and re-product as well as to adapt to the changeable conditions of existence. In a similar way, we also require that the designed technical systems be characterised by analogous features within the scope of adaptation, optimality and immunity. In particular, we can easily formulate tasks concerning the optimality of solutions and their robustness to small changes in environmental parameters and to disturbances that lead to more effective and reliable engineering systems. Moreover, in many practical decision-making processes it is essential to totally optimise several objective functions, and we often have to determine the relations between the partial objectives considered in order to integrate those objectives into one.

Any parametric description of a diagnosed object includes only a small part of all existing state parameters; in fact, it is only a simplified model of reality. The best description is that which is in equivalent relation to the given states of the object. It means that the value x(p) of a given physical quantity from the set X occurs if and only if the object is in the state m. In the case of real objects, it is often impossible to establish unambiguously these relations because the physical processes considered are not known adequately in their analytical form, or the parameter calculation is too complex computationally. It follows that fault diagnosis demands determining the relations existing between the measured symptoms (changes of the observed quantity over its face value) and the faults (Calado et al., 2001; Frank and Koppen-Seliger, 1997; Isermann and Balle, 1997).

Modern measurement technology makes it possible to continuously observe and record signals connected with the courses of technological processes, and machinery or devices which take part in these processes. Most often, the signals are supplied to modules which analyse them in order to estimate a set of their features forming the symptoms of the present technical state of the observed object. A particular property of the problems of technical diagnostics is that they are usually related to objects (e.g., machines) of different constructions. It requires a distinction between the forms of databases, and specialisation of rule sets applied within an inference process dealing with the technical state of the object. Additional difficulty is that the history of changes occurring in the observed objects (e.g., modernization) and the history of their maintenance (e.g., repair and control) must be recorded and taken into account in the inference process. The need for applying monitoring and diagnosing devices to complex technical objects is the main reason for research whose goal is to find proper tools for aiding the processes of the design and maintenance of such devices. Interpreting the results of signal analysis is a difficult task. It always requires some kind of experience regardless of the fact that the diagnosing is based on an exhaustive model of the object or on diagnostic rules which are considered to be valid for a determined class of machinery.

In the process of technical systems fault diagnosis man uses different methods of knowledge representation and inference paradigms. The most common scenario of such a process consists in the detection of faulty behavior of the system, classification of that kind of behavior, search for and determination of causes of the observed misbehavior, i.e., the generation of potential diagnoses, verification of diagnoses and selection of the correct one, and, finally, the repair phase. There exist a number of approaches and diagnostic procedures having their origin in very different branches of science, such as mechanical engineering, electrical engineering, electronics, automatics, or computer science. In the diagnosis of complex dynamical systems an important role is played by approaches originating in automatics (Kościelny, 2001) and computer science (Frank and Köppen-Seliger, 1995; Hamscher et al., 1992; Johnson and Keravnou, 1985; Korbicz and Cempel, 1993; Liebowitz, 1998), and they seem to be the most interesting ones. A fine comparative analysis of some selected approaches was presented in (Cordier et al., 2000a; 2000b). The present chapter is devoted to the presentation of some selected approaches originating in computer science.

Contemporary technical means, especially machinery and equipment, become more and more complex. Therefore, one may see the growth of requirements for persons and organizations whose responsibility is the building (that includes designing and manufacturing) and operation of machinery and equipment. To efficiently perform in each of the above-mentioned zones of activity, one needs suitable knowledge and skills. Both of them are possessed by domain experts, usually several in each domain. They acquire knowledge and skills during their studies, long-term professional activity, by observations or from technical literature.

In the case of industrial processes (systeIlls), diagnoses should be formulated on-line and in real time. Such a method of diagnosing is called system state monitoring (Kościelny, 2001). This chapter presents state monitoring algorithms for complex industrial installations.

The diagnostic system is most often integrated with the automatic control system. The structures of modern computer-based automatic control systems that control industrial processes are decentralised and space-distributed (Fig. 19.1). It is advisable that diagnostic functions that constitute an integral part of control tasks and process protection be also realised in decentralised structures.

Tracking filters are the principal part of radar data processing systems. In fact, the tracking filter is a state estimator (Anderson and Moore, 1979) of the object (target) being tracked. Its task is to process radar measurements of motion parameters of such a target in order to reduce measurement errors by means of time averaging, to estimate the object’s velocity and acceleration and to predict its future positions (Blackman and Popoli, 1999).

Large pipeline networks are widely used to transport various fluids (liquids and gases) from production sites to consumption ones. Transmission pipelines are the most essential parts of such networks. They are used to transport fluids at very long distances, which may vary in length from several kilometres to hundreds or thousands of kilometres. They have few branches and no loops, operate at relatively high pressure and need compressor stations, which generate the energy the fluid needs to move through the pipeline. Despite huge initial establishment costs, in the long run transmission pipelines are convenient and economical for the transportation of large volumes of fluid. In various industries transmission pipelines are often compulsory.

The requirements concerning the industrial application of fault detection and isolation seem to be quite natural taking into account process safety, end product quality and process economy factors. The pilot industrial implementations of chosen diagnostic methods will be presented and described in this chapter. Those methods are principally based on fuzzy or fuzzy neural network models used for the fault detection and isolation purposes. The presented examples deal with very complex systems, e.g., the steam-water line of the power boiler, the sugar manufacturing evaporator unit, as well as simple ones, namely, final control elements. To demonstrate the role and importance of fault detection and isolation one can present an example of a particular fault of the final control element controlling the inflow rate of thin juice to the first evaporator apparatus of a sugar factory. This fault may completely stop the process if not detected within 30 s. An example of the application of diagnostics to the development of the fault tolerant condensation power turbine controller is also briefly described.

In the last few years, there has been observed a rapid growth of interest in industrial applications of diagnostic systems. It results from potentially high economic profits which can be brought about by industrial applications, as well as from the rise of a new generation of control systems which facilitate the application of advanced software tools to the supervisory control and diagnostics of industrial processes.