GAN-DRSN based Inter-turn Short Circuit Fault Diagnosis of PMSM

Abstract

Due to the small number of samples of the inter-turn short-circuit fault of the current permanent magnet synchronous motor, and the motor working in a high-noise environment, the collected data contains complicated noise. So first the deep residual shrinkage network is pre-trained on the big data simulation dataset. And then to avoid imbalances between real data sets, GAN network is adopted to generate more datasets in this paper. Based on the aforementioned data set, the pretrained network is proposed to denoise the environment and other noise in the data set. And Spatial Dropout layer into the network is introduced to improve the accuracy and convergence speed of fault diagnosis. Experiments show that by combining GAN and DRSN methods for fault diagnosis of unbalanced samples, disturbances such as datasets and reducing environmental noise can be effectively balanced. The diagnostic accuracy is as high as 97.5%.