Industrial Control Systems, Neural Network Fault Diagnosis

Fault detection with neural network is one of the major courses of intelligent fault detection theory and technology. Studying the major researches relating neural network fault detection, this paper divides the faults into two group, measure level faults and system level faults, and deal with them separately. To take advantage of redundancy information, this paper gives a new method that implement the output data fusion of integrated neural network based on the D-S evidence theory. To view each neural network as evidence, it can carry on the output data fusion of neural network on both the time and space regions, and thus can improve the precision of the diagnosis result. Also the fusion can be applied to between neural network and fuzzy neural network, and by simulation its feasibility is proved.To deal with the defects existed in the neural network fault diagnosis algorithm, an improved GA-ANN algorithm has been given which applied genetic algorithm to optimization of the neural network's weights. It combines the BP algorithm with genetic algorithm by utilizing the advantages of both algorithms. Fast approaching speed and whole space optimizing are two features. In this paper, this algorithm has been compared with BP algorithm, GA algorithm, general GA-ANN algorithm by simulation. Results show that this algorithm outgoing the other three.