Study On Particle Swarm Optimization-based Neural Network For Fault Diagnosis

In essence, fault diagnosis of the transmission is a process of recognizing the fault pattern, and it has been challenging the existing intelligent fault diagnosis technology because of the complexity of transmission structure, the various operating conditions and the constant loud background noise.Particle Swarm Optimization (PSO) is a population-based stochastic optimization technique, has gained popularity in solving single objective or other optimization tasks. In order to avoid the problem of premature convergence of the PSO algorithm, a regrouping PSO is adopted in this paper. And BP neural network also has the local extreme problem which is one of the causes of poor diagnostic accuracy and stability.So this paper presents a novel regrouping PSO-based neural network in which parameters of neural network can be selected automatically. The fault diagnosis of bearing has shown that the proposed method is better than other methods, such as BP neural network, genetic neural network and basic particle swarm optimization-based neural network, in the aspect of training efficiency and fault identification rate. Based on the previous study, a regrouping PSO-based varistructure neural network algorithm (IPSOVNN) is proprosed, in the training process which can automatically choose the number of neurons in the hidden layer. Results of the fault diagnosis of bearing reveal the effectiveness of this method.After the transmission gear failure experiment, IPSOVNN is employed in the fault diagnosis of the gear samples under two different conditions, and the performance of IPSOVNN is discussed comparing with other methods such as BP neural network and firefly varistructure neural network.