Abstract: The paper puts forward a set of faults diagnosis methods of testing the noise of the three-phase motor current which shows the characteristics of the motor faults. and multi-varieties time series models of the noise is established, the multi-varieties residual series is changed to the monistical as the fault detection index. In the fight of trait of the model parameter φi, presenting a faults classification recognition model based on the multi -layer NN structure, using APEX network extracts classification Information of the initial pattern vector,making use of feedforward network establishes the classification function. The diagnosis way is correct by practising.