CWMS-GAN: A small-sample bearing fault diagnosis method based on continuous wavelet transform and multi-size kernel...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Industrial bearing fault diagnosis relies heavily on sufficient data, which is often scarce.
• Traditional deep learning models suffer performance degradation with limited fault data.
• Generative Adversarial Networks (GANs) show promise for data augmentation but sample quality is critical.

Purpose of the Study:

• To propose an effective GAN-based method for small-sample bearing fault diagnosis.
• To enhance the accuracy and authenticity of generated fault signals.
• To address the challenge of insufficient bearing fault data in industrial settings.

Main Methods:

• Implemented a Continuous Wavelet Convolution (CWCL) strategy within the GAN framework to capture frequency domain features.
• Introduced a Multi-Size Kernel Attention Mechanism (MSKAM) for adaptive feature extraction across different scales.
• Utilized the Structural Similarity Index (SSIM) for quantitative evaluation of generated signal quality in time and frequency domains.

Main Results:

• The proposed CWCL and MSKAM significantly improved the quality and authenticity of generated bearing fault signals.
• Experimental results on CWRU and MFPT datasets demonstrated superior performance compared to existing small-sample fault diagnosis methods.
• The SSIM metric effectively validated the enhanced signal generation capabilities.

Conclusions:

• The novel GAN-based approach with CWCL and MSKAM offers a robust solution for small-sample bearing fault diagnosis.
• This method effectively overcomes the limitations of insufficient data in industrial fault detection.
• The enhanced signal generation improves the reliability of deep learning models for bearing health monitoring.