PMSM Stator Winding Fault Detection and Classification Based on Bispectrum Analysis and Convolutional Neural Network

• Published in: IEEE transactions on industrial electronics (1982) Vol. 70; no. 5; pp. 5192 - 5202
• Main Authors: Pietrzak, Przemyslaw, Wolkiewicz, Marcin, Orlowska-Kowalska, Teresa
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Artificial neural networks; Bispectrum; Classification; convolutional neural network (CNN); Convolutional neural networks; deep learning; Fault detection; Fault diagnosis; Neural networks; permanent magnet motors; Permanent magnets; Phase current; Programming languages; stator fault diagnosis; Stator windings; Stators; Synchronous motors; Training; Winding; Windings
• ISSN: 0278-0046; 1557-9948
• DOI: 10.1109/TIE.2022.3189076

The diagnosis of permanent magnet synchronous motor (PMSM) faults has been the subject of much research in recent years, due to the growing reliability and safety requirements for drive systems. This article concerns PMSM stator winding fault detection and classification. A novel intelligent diagnosis approach is proposed, based on the bispectrum analysis of a stator phase current and the convolutional neural network (CNN). Rather than using raw phase current signals, bispectrum is applied for symptom extraction and utilized as the input for a pretrained CNN model. The CNN model is used for automatic inference on the winding condition of the PMSM stator. Experimental results are presented to validate the proposed algorithm. The classification effectiveness of the developed CNN is as high as 99.4%. This article also presents the possibility of improving the accuracy of the CNN model and reducing the training time by properly tuning the training parameters. The CNN model learning time is only about one minute. The fault classifier model is developed in Python programming language, avoiding the cost of purchasing additional software.