BP Network for Diagnosing Rotor Broken Fault Based on a New PSO

A new BP neural network is proposed in this paper, the weights and thresholds of the network are optimized by an improved PSO algorithm instead of gradient descent method. The strategy of the algorithm is that at each iteration loop, on every dimension of particle swarm, choose the particle whose acceleration is biggest and the velocity is small enough to mutate its velocity according to some probability. The strategy could improve the global exploring ability of the particle, and could avoid the particle sticking to the local minimum. Also the improved PSO algorithm could effectively improve the convergence performance of conventional BP neural network. The new BP network is applied to diagnosing the rotor broken fault of motor in this paper. Experiment results show that the new algorithm enhance the diagnosis accuracy and improve the performance of network effectively.