Multi-Frequency-Scale Distributed Recurrence Plot-Based Fault Diagnosis for PMSM.

Area of Science:

• Electrical Engineering
• Machine Learning
• Signal Processing

Background:

• Conventional permanent magnet synchronous motor (PMSM) fault diagnosis relies on 1-D time-series signals, facing challenges in feature extraction and noise immunity.
• Existing methods struggle with complex signal processing and limited effectiveness in identifying subtle fault characteristics.

Purpose of the Study:

• To develop a novel, efficient, and robust fault diagnosis method for PMSM.
• To overcome the limitations of traditional 1-D signal processing and recurrence plot techniques.

Main Methods:

• Utilized wavelet packet decomposition (WPD) for multi-frequency band signal representation.
• Employed distributed recurrence plot (DRP) generation and image transformation for enhanced feature extraction.
• Developed a lightweight multi-frequency-scale convolutional neural network (CNN) model incorporating a convolutional block attention module (CBAM) and global average pooling (GAP).

Main Results:

• The proposed method achieved high diagnostic accuracy and demonstrated strong noise immunity.
• Significantly reduced inference time compared to traditional recurrence plot-based CNN methods (12.35% and 50.03% of existing methods).
• Effectively represented signal features across multiple frequency bands, overcoming limitations of traditional recurrence plots.

Conclusions:

• The novel WPD-DRP-CNN approach offers a superior alternative for PMSM fault diagnosis.
• The developed lightweight model provides a computationally efficient and accurate solution for real-time applications.
• This method enhances diagnostic reliability and robustness in noisy environments.