Fault Diagnosis of Permanent Magnet Synchronous Motor Based on Stacked Denoising Autoencoder

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Automotive Systems

Background:

• Permanent magnet synchronous motors in EVs face complex operating conditions and diverse failure modes.
• Existing intelligent fault diagnosis methods struggle with accuracy, generalization, and high-dimensional data.
• Fault signs in these motors can be overlapping and difficult to isolate.

Purpose of the Study:

• To develop an advanced fault feature extraction method for electric vehicle motors.
• To improve the accuracy and generalization of fault diagnosis in complex motor systems.
• To address the limitations of traditional and single autoencoder-based diagnosis methods.

Main Methods:

• Utilized a stacked denoising autoencoder (SDAE) for robust feature extraction from motor signals.
• Integrated SDAE with a support vector machine (SVM) classifier for fault identification.
• Employed noise injection during data processing to enhance model resilience.
• Optimized network parameters, including learning rate and noise reduction coefficients.

Main Results:

• The proposed SDAE-SVM method demonstrated superior accuracy compared to traditional techniques.
• The approach showed enhanced generalization ability in diagnosing motor faults.
• Effectively processed high-dimensional motor operational data for feature extraction.

Conclusions:

• The combined SDAE-SVM method offers a powerful solution for electric vehicle motor fault diagnosis.
• This technique overcomes limitations of existing methods in accuracy and data handling.
• The study highlights the potential of deep learning for complex electromechanical system diagnostics.