A data-driven intelligent fault diagnosis framework for permanent magnet in PMSM

Due to operating conditions and natural aging, the demagnetization of permanent magnets (PM) can significantly impact the performance of permanent magnet synchronous motors (PMSM), potentially leading to reduced output and motor damage. In response to this challenge, a comprehensive PM health management framework is proposed, applying data-driven algorithms to monitor the state of PMs based on raw stator current data. First, the proposed approach anomaly detection requires only normal data during the training phase, addressing the practical issue of limited fault samples. Furthermore, ensemble learning integrates the outputs of multiple base learners to enhance generalization. Second, fault diagnosis is carried out using the Genetic Algorithm-Support Vector Machine (GA-SVM) to accurately classify demagnetization faults. Finally, the Pauta criterion is applied to process the intermediate Support Vector Domain Description (SVDD) variables, generating anomaly scores that classify PMSM health states based on segmented thresholds. Compared to existing methods, this system provides a more comprehensive and autonomous PMSM state recognition, reducing the reliance on expert judgment and minimizing the risk of failure. Through FEM+MATLAB simulations under various demagnetization and load conditions demonstrate the framework's effectiveness in monitoring PM health. This approach offers a cost-effective and intelligent maintenance solution, enhancing the reliability and longevity of PMSM systems.