Multi-source data fusion for demagnetization fault diagnosis of PMSM based on SDP and CNN

The Permanent Magnet Synchronous Motor (PMSM) is widely used in industrial production. However, the demagnetization fault of the PMSM significantly affects motor performance, making accurate demagnetization fault diagnosis particularly important. As demagnetization can have an impact on various electromagnetic parameters, the comprehensive utilization of multiple parameters for demagnetization fault diagnosis is particularly effective. This paper proposes a multichannel multisource data fusion method for demagnetization fault diagnosis based on the symmetric dot pattern (SDP) images. Firstly, the Ansys soft is used to map a kind of PMSM and obtain multisource data based on the mechanism and simulation experiment of the PMSM. That is, the stator current, electromagnetic torque, and back electromotive force (EMF). Secondly, those one-dimensional univariate data can be converted into two-dimensional image data based on the sliding window technique and the SDP technique. In particular, three kinds of methods based on the SDP technique for image generation are designed and compared to obtain the optimal method based on the same CNN. Finally, the experimental results show that the multichannel multisource data fusion method proposed in this paper achieves the highest classification accuracy of 99.07%, which significantly improves the reliability of PMSM demagnetization fault diagnosis.