Fault detection and isolation with RBF neural network

Fault detection and diagnosis have been actively studied during recent years. Estimation methods, rule-based reasoning and pattern recognition techniques are the most common methods used to solve the above issues. In recent years artificial neural networks have been successfully used in pattern recognition tasks and their suitability for fault diagnosis problem has also been demonstrated. In this paper the use of RBF neural network in this area is proposed. Firstly a neural network can be used instead of a mathematical model for residual generation. Secondly another neural network can be trained to perform the classification task for residual evaluation and fault isolation. Thirdly a one step diagnosis (OSD) is used, where a neural network is directly trained to detect the possible faults from input - output measurements without the need for intermediates signals as residuals. Results obtained from the application of RBF neural network to the fault detection problem for an industrial plant (boiler-drum) are presented. Copyright (C) 1998 IFAC.