Fault Diagnosis of Bearing in PMSM Based on Calibrated Stator Current Residual Signal and Improved Symmetrized Dot Pattern

Permanent magnet synchronous motor (PMSM) is widely used in nearly all industrial applications, as key components, the performance degradation of bearings in PMSM-driven-systems poses threats to operation safety. However, because of environmental conditions, space, and weight limits, additional vibration sensors are unacceptable in some engineering scenarios; therefore, this study proposes a fault diagnosis method for bearings coupled with PMSM based on stator current signal. First, the current signals are calibrated based on current phases. Then, to improve the robustness of neural network, the training samples are extended based on data enhancement, considering the influencing factors of current signal. After that, calibrated residual signals (CRSs) are calculated to enhance the subtle fault information in current signals, and the CRSs are transformed into symmetrized dot pattern (SDP) images with color information. Finally, these images are employed for fault pattern recognition based on a pretrained 2-D convolution neural network (2D-CNN). The proposed method was validated based on stator current signals acquired from a test rig, and the accuracy of fault diagnosis was 96.67%. The advantages of the proposed method were also demonstrated based on comparative case studies.