Fault Diagnosis and Fault Frequency Determination of Permanent Magnet Synchronous Motor Based on Deep Learning

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Early diagnosis of motor faults is crucial for preventing failures and ensuring operational efficiency.
• Deep learning approaches are increasingly utilized for motor fault diagnosis.
• Permanent magnet synchronous motors (PMSMs) are widely used, necessitating reliable diagnostic methods.

Purpose of the Study:

• To propose a weakly supervised one-dimensional convolutional neural network (1D-CNN) for diagnosing permanent magnet synchronous motors.
• To develop a model capable of diagnosing motors under diverse operating conditions, including varying speeds, loads, and eccentricity.
• To enhance diagnostic accuracy and reduce model complexity compared to traditional methods.

Main Methods:

• A weakly supervised 1D-CNN model comprising multiple convolutional feature-extraction modules was developed.
• Analysis of motor torque and current signals was performed to extract multiscale features.
• A class feature map was introduced for automatic identification of frequency components contributing to classification.
• Model training parameters were reduced to mitigate overfitting.

Main Results:

• The proposed 1D-CNN model effectively diagnosed three motor states: healthy, demagnetization fault, and bearing fault.
• The model demonstrated capability in detecting eccentric effects in motors.
• Classification accuracy reached up to 98.85% when combining current and torque features.
• Performance surpassed classical machine learning methods like k-nearest neighbor and support vector machine.

Conclusions:

• The developed 1D-CNN model offers a robust and accurate solution for diagnosing permanent magnet synchronous motors.
• The multiscale feature extraction and parameter reduction contribute to effective fault diagnosis under complex conditions.
• This deep learning approach provides a superior alternative to conventional machine learning techniques for motor fault detection.