Neural network based on wavelet packet-characteristic entropy and rough set theory for fault diagnosis

A new method of vibrant fault diagnosis was proposed for electric locomotive traction motor based on wavelet packet transform, rough set theory and the back propagation neural network. Firstly, Energy analysis and symptom extraction are carried out by wavelet packet transform. Wavelet packet transform can pick up more comprehensive useful information of the traction motor in high frequency domain than wavelet transform, which is regarded as evidence to diagnose fault. Secondly, the fault information of wavelet packet-characteristic entropy is reduced by the rough set theory on the basis of classifying capability unchanged, then the information is diagnosed by improved BP neural network, which not only decreases the number of the network input number effectively, but also shortens the training time. Finally, the simulation results in electric locomotive traction motor indicated the high diagnosing accuracy and effectiveness of the presented net.