Demagnetization Fault Diagnosis of PMSM Based on Fuzzy Extreme Learning Machine

To improve the accuracy of partial demagnetization fault diagnosis for permanent magnet synchronous motor (PMSM), an improved fuzzy extreme learning machine (F-ELM) algorithm is constructed by integrating fuzzy theory into the extreme learning machine (ELM) in this paper. Firstly, a PMSM field-circuit coupling simulation system under vector control is established by employing finite element analysis method. Secondly, the feature samples affecting classification accuracy can be obtained using wavelet packet decomposition (WPD). Thirdly, by taking the imbalance of demagnetization feature into consideration, a new type of improved ELM, i.e., F-ELM, is proposed by associating input layer with a fuzzy membership. Finally, a comparison with the accuracy of back-propagation neural network (BPNN), support vector machine (SVM), and ELM is performed. The experimental results show that the proposed method outperforms the existing machine learning methods, and can effectively diagnose the partial demagnetization fault of PMSM.