Mechanism-Based Fault Diagnosis Deep Learning Method for Permanent Magnet Synchronous Motor

Area of Science:

• Electrical Engineering
• Machine Learning
• Signal Processing

Background:

• Permanent magnet synchronous motors (PMSMs) are crucial in industrial applications.
• Harsh operating environments necessitate accurate fault diagnosis for PMSMs.
• Existing methods may not fully capture complex fault signatures.

Purpose of the Study:

• To develop an intelligent fault diagnosis method for PMSMs.
• To accurately identify inter-turn short-circuit and demagnetization faults.
• To leverage time-frequency analysis and deep learning for enhanced diagnosis.

Main Methods:

• Mechanism analysis of PMSM faults (inter-turn short-circuit, demagnetization).
• Application of continuous wavelet transform (CWT) for time-frequency feature extraction.
• Development and implementation of a convolutional neural network (CNN) for fault classification.
• Visualization of results using t-distributed stochastic neighbor embedding (t-SNE).

Main Results:

• Identified key frequency ranges for specific PMSM faults in time-frequency domain.
• CNN model effectively extracted features from time-frequency images.
• Achieved diagnostic accuracy exceeding 98.6% for various fault severities.
• t-SNE visualization confirmed effective feature separation and classification.

Conclusions:

• The proposed CWT-CNN method provides a highly accurate and intelligent approach for PMSM fault diagnosis.
• This technique is effective for detecting inter-turn short-circuit and demagnetization faults.
• The findings have significant implications for improving the reliability and maintenance of PMSMs in industrial settings.