Fault diagnosis using hybrid wavelet/ Elman neural networks

The paper addresses the problem of process fault diagnosis using a new kind of neural network, namely the hybrid wavelet/ Elman neural network. Hybrid learning based on a subclustering algorithm and the steepest-descent method is used to train the proposed neural network. The experiments refer to the sensor and actuator fault diagnosis of a sub-system from the evaporation station of the Lublin sugar factory, namely the evaporator. A generalised neural network based observer scheme is used to generate the residuals (symptoms) in the form of one step-ahead prediction errors. These are then classified by a Multi-Layer Perceptron neural network in order to take the appropriate decision regarding the type of the behaviour of the process (normal or abnormal).