Wavelet Neural Network Approach for Fault Diagnosis to a Chemical Reactor

A fault diagnosis method of chemical reactor using wavelet neural network (WNN) is presented. A WNN with three-layer is constructed for fault diagnosis. The activation function of neuron in hidden layer of WNN is constructed by Morlet wavelet function. An improved adaptive BP learning algorithm of adjusting weights of WNN was given. The convergence speed of WNN is faster than the traditional BP neural networks. A representative chemical process in which heptane was converted to toluene aromatization via a catalyst was studied. Five fault types of this process were given, the relationship between the chemical reaction process and fault component was constructed. Using WNN as fault pattern recognition device, the fault diagnosis of this chemical reactor was achieved, and the recognition results by the WNN were compared with BP´s results. The validity and veracity of two methods are verified by simulation