Study Of PSO-H-BP Neural Networks Application In Fault Diagnosis

With the raising of the complexity and the level of automation of the equipment, the equipment fault diagnosis is becoming more and more important. Imitating the human brain's physical structure and with its powerful parallel computing and the ability to think, artificial neural network is very suitable for the equipment fault diagnosis. The equipment failure diagnosis using of neural network can greatly enhance the accuracy and reduce the people's dependence.This paper mainly studied Hopfield neural network firstly, because of its advantages in associative memory and data optimized and its features such as robust, self-organizing adaptive, parallel processing, distributed storage and a high degree of fault-tolerant, combining with BP neural network which has features of local optimization and teacher-learning, and then the H-BP neural network is advanced. The H-BP neural network has the features of global optimization and non-teacher-learning which are an attribute of Hopfield neural network, and the features of BP neural network, that is local optimization and teacher-learning. The H-BP neural network can improve the determining rate of pattern recognition. Then in this paper, the weight of Hopfield neural network is modified using the particle swarm optimization, and the PSO-H neural network is proposed to optimize the original data, and then another network structure with higher determining rate is advanced: PSO-H-BP neural network.In this paper, simulation software Matlab is used. According to the result of diagnosis and analysis of centrifugal fan faults, it is verified that various types of common fans faults could be accurately identified by the H-BP neural network and PSO-H-BP neural network, and the effectiveness of the proposed network is proved.