Abstract:

In order to achieve the diagnosis of fault categories of permanent magnet synchronous motor, fault feature is extracted by using wavelet function according to the different frequency band, then normalized date sample as the SOM(Self Organizing Map) network input, the SOM field function is constructed with the wavelet function and the second excited neurons are formed to update weights, so the local optimization of SOM is avoided. The fault data extracted is regarded as the input samples of SOM neural networks in order to train the network, and appropriate neuron index stimulated when a specific fault occurs is obtained. the feasibility has been demonstrated by simulation results.

Key words: neural network, fault diagnosis, neuron index, permanent magnet synchronous motor (PMSM)