Fast Online Fault Diagnosis for PMSM Based on Adaptation Model

Permanent magnet synchronous motors (PMSMs) are widely used as the key equipment in devices due to their superior performance, and their health status is closely related to the reliability of the devices' operation. Therefore, it is extremely necessary to monitor the health status of PMSM in real time. Deep learning algorithms, as one of the most popular applications for diagnosing faults in motors, are primarily utilized on computers or servers. There are also a number of related online diagnostic methods whose reasoning is relatively time-consuming and only suitable for stationary signals. In this article, a fast online fault diagnosis method utilizing time-frequency domain adaptation is proposed. First, a fault diagnosis model is proposed utilizing wavelet transform and mask impute (WTMI). After using wavelet transform (WT) to analyze and process nonstationary data, the model uses mask impute to maintain temporal consistency, thereby effectively completing training. The trained model needs to be verified by the theoretical simulation platform to verify the diagnostic accuracy, noise immunity, and execution time. Second, a chunked processing framework and hardware platform are designed to make the WTMI model fast and convenient. Finally, the proposed domain adaptation-based diagnostic model is transplanted onto the edge computing (EC) platform, enabling online fault diagnosis of PMSM. The reliability of the proposed method is verified on the test platform across various online scenarios. The test results demonstrate that the proposed method is 4-10 times faster than the comparison methods while ensuring accuracy.