Adaptive feature extraction based on Stacked Denoising Auto-encoders for asynchronous motor fault diagnosis

As the important power equipment in the mechanical system, fault diagnosis for asynchronous motor is helpful to monitor working status and prevent failure causing unnecessary loss. In the fault diagnosis domain, feature extraction is the key step which is related to the performance of diagnosis results. For the asynchronous motor, the motor current signature analysis (MCSA) is one of the most powerful diagnosis method with stator-current signals. However, MCSA has some shortcomings, which degrade performance and accuracy of a motor-diagnosis system. Therefore, advanced feature extraction algorithm of current signal using Stacked Denoising Auto-encoders (SDAE) is proposed in this paper. The method of SDAE and application in motor are discussed in detail. Then, the features learned from the SDAE is displayed and a softmax regression model is used to verify the discriminability of the features. The experiments show that SDAE is an effective feature extraction technique for asynchronous motor fault diagnosis.