Teodor Marcu

Model-based fault diagnosis in technical processes

In this paper the state-of-the-art developments model-based fault diagnosis in technical processes are reviewed. Attention is focused upon both the analytical approaches that make use of the quantitative models and the knowledge-based approaches using qualitative models. Basic concepts and the advantages as well as disadvantages of different model-based fault diagnosis schemes are outlined.

SYSTEM IDENTIFICATION USING FUNCTIONAL - LINK NEURAL NETWORKS WITH DYNAMIC STRUCTURE

Abstract The paper considers the development of a new type of artificial neural network and its applicability to non-linear system identification. This is the functional-link neural network with internal dynamic elements. The net consists of a single layer where the non-linearity is firstly introduced by enhancing the input pattern with a functional expansion. The internal dynamic elements are auto-regressive moving average filters that implement local activation feedback and local output feedback, respectively. Experimental results demonstrate a better capability of generalisation of the suggested neural network in comparison with the functional-link net with static structure and external dynamic elements, used so far to perform system identification.

Diagnosis strategies and systems: principles, fuzzy and neural approaches

Fault tolerance of automatic control systems is gaining increasing importance. This is due to the increasing complexity of modern control systems and the growing demands for quality, cost efficiency, availability, reliability and safety. The use of knowledge based systems and of various“intelligent technologies” demonstrated significant improvements over the classic techniques. In this chapter, we review the state of this development along with the enumeration of some successful applications.

A Multiobjective Evolutionary Approach to Pattern Recognition for Robust Diagnosis of Process Faults

Abstract The problem of robust model-based diagnosis of process faults is addressed in the framework of pattern recognition. Evolutionary algorithms of genetic type are used to solve both problems of feature selection and classifier design by means of multiobjective optimization. Process coefficients are directly identified by an on-line procedure. Symptoms are then evaluated by a non-parametric classifier. Application to a laboratory process is included. A diagnosis subsystem is designed and implemented in real-time to detect incipient faults in the components of a three-tank system.

Three—tank Control Reconfiguration

This chapter describes benchmark problem for controller reconfiguration. It is based on a laboratory process consisting of tanks with fluid flow. Many schemes for controller reconfiguration have been investigated and compared.

Miscellaneous Neural Networks Applied to Fault Detection and Isolation of an Evaporation Station

Abstract The problem of robust diagnosis of process faults in an evaporation station of a sugar factory is addressed by means of a neural-network approach. The main emphasis is placed upon the development of generalised observer schemes. These are designed based on neural nets with internal dynamics. The goal is to achieve an adequate approximation of process outputs corresponding to the normal behaviour of the plant. Symptoms characterising the current state of the process are obtained based on the prediction errors. Static artificial networks further evaluate these. Appropriate decision mechanisms lead to fault detection and isolation. Experimental results using real data supplied by industry assess the efficiency of the approach.

EVOLUTIONARY DESIGN OF DYNAMIC NEURAL NETWORKS APPLIED TO SYSTEM IDENTIFICATION

Abstract The problem of system identification is addressed by means of general neural networks with locally distributed dynamics. These networks are based on both multilayer perceptron and radial basis function structures. Evolutionary algorithms are suggested to select the optimal neural topologies and parameters. The accuracy of the neural models and the complexity of their architectures are evaluated by considering six objective functions organised on a two-level priority hierarchy. The multiobjective optimisation is solved in the Pareto-sense. Special mechanisms are developed, in order to encourage a rapid improvement of the genetic material. Application to a laboratory three-tank system illustrates the approach.

Parallel Evolutionary Approach to System Identification for Process Fault Diagnosis

Abstract The robustness issue in model-based diagnosis of process faults is addressed by means of parameter estimation. System identification is fonnulated as a problem of multiobjective optimization. The solution is based on parallel evolutionary algorithms of genetic type. Process coefficients are directly identified by a generalized on-line procedure. A simplification of the stage of symptom evaluation results. Application to a laboratory process is included. A diagnosis subsystem is designed to detect incipient faults in the components of a three-tank system.

Neural Approaches to Observer-Based Diagnosis of Faults in Dynamic Systems

Abstract The problem of robust model-based diagnosis of process faults is addressed by means of artificial neural networks. They are investigated for both approaches to function approximation and pattern classification. Main emphasis is placed upon generalized dynamic neural nets used as predictors of nonlinear models for symptom generation. Application to a laboratory process is presented. A diagnosis subsystem is designed to detect incipient faults in the components of a Three-Tank System. It is implemented in real-time using the SIMULINK/MATLAB environment.