Recurrent wavelet neural networks applied to fault diagnosis

The paper investigates the development of a new type of recurrent wavelet neural network and its application to fault detection and isolation (FDI) of a dynamic process. Hybrid learning based on c-means fuzzy clustering algorithm and the steepest-descent method, is used to train the proposed neural network. The experimental case study concerns the sensor and actuator fault diagnosis of a sub-system from the evaporation station of a sugar factory, namely the evaporator. A neural generalised observer scheme is used to generate the residuals (symptoms) in the form of one step-ahead prediction errors. These are then processed by a neural classifier in order to take the appropriate decision regarding the type of the behaviour of the process (normal or abnormal).