A novel method of combining generalized frequency response function and convolutional neural network for complex...

Area of Science:

• Electrical Engineering
• Machine Learning
• Signal Processing

Background:

• Traditional fault diagnosis methods for Permanent Magnet Synchronous Motors (PMSM) suffer from low accuracy.
• Accurate characterization of system state information is crucial for effective fault diagnosis.
• Advanced feature extraction techniques are needed to improve diagnostic capabilities.

Purpose of the Study:

• To propose a novel fault diagnosis method for PMSM by integrating generalized frequency response function (GFRF) and convolutional neural network (CNN).
• To enhance the accuracy and reliability of PMSM fault diagnosis.
• To develop a robust system for characterizing normal and fault states of PMSM.

Main Methods:

• A variable step size least mean square (VSSLMS) adaptive algorithm was employed to compute second-order GFRF spectrum values for PMSM under normal and fault conditions.
• A convolutional neural network (CNN) architecture, featuring gradient descent learning rate and alternating convolution and pooling layers, was designed for fault feature extraction from GFRF spectra.
• Two-dimensional GFRF spectra were converted into image format for input into the CNN for training and diagnosis.

Main Results:

• The proposed method successfully extracted fault features from the GFRF spectra of PMSM.
• The fault diagnosis accuracy achieved by the combined GFRF and CNN method reached 98.75%.
• Experimental results validated the reliability and effectiveness of the proposed approach for PMSM fault diagnosis.

Conclusions:

• The integration of GFRF and CNN offers a significant improvement over traditional fault diagnosis methods for PMSM.
• The VSSLMS algorithm provides accurate system state characterization, forming a reliable basis for fault diagnosis.
• The developed CNN effectively extracts critical fault features, leading to high diagnostic accuracy and demonstrating practical applicability.