Fault Diagnosis of AC Squirrel-cage Asynchronous Motors based on Wavelet Packet-Neural Network

Abstract

Squirrel-cage asynchronous motors are used widely in industry production process. It is significant to improve squirrel-cage asynchronous motors diagnosis technique in application. It helps to decrease the occurrence of accident and reduce the cost of maintenance. Based on the wavelet packet-neural network the scheme on the real-time diagnosis of the stator, bearing, and eccentricity fault of squirrel-cage asynchronous motors is presented in this paper. The electric machines stator current signal is analyzed and processed, with wavelet packet decomposition algorithm. By this method the problem that conventional FFT algorithm can't provide any partial information in time domain is solved. The eigenvectors of bearing and eccentricity fault that processed by wavelet packet decomposition algorithm. The eigenvectors are tiny. The neural networks are not convergent easily when they are trained, with these eigenvectors as above. To this problem, this paper gives a novel scheme based on logarithm fault eigenvectors extraction. By the logarithm fault eigenvectors proposed, the fault of squirrel-cage asynchronous motors can be real time diagnosed and categorized precisely. This scheme is feasible, which is proved by the test experiment.