Rolling bearing fault diagnosis method based on gramian angular difference field and dynamic self-calibrated...

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Traditional fault diagnosis methods struggle with limited data and high noise levels.
• Effective feature extraction is crucial for accurate rolling bearing diagnostics.

Purpose of the Study:

• To propose an advanced fault diagnosis method for rolling bearings.
• To enhance feature extraction capabilities under challenging conditions like sample scarcity and strong noise.

Main Methods:

• Utilized Gramian Angular Difference Field (GADF) to transform 1D signals into 2D images, capturing complex data relationships.
• Introduced Dynamic Self-Calibrated Convolution (DSC) to adaptively adjust convolution kernel weights for improved feature learning.
• Validated the method on public bearing datasets from HUST and HIT.

Main Results:

• Achieved classification accuracy consistently above 90% even with -8 dB Gaussian white noise.
• Demonstrated a 6%-15% improvement in accuracy compared to other state-of-the-art models.
• The DSC module significantly enhanced feature extraction and model generalization.

Conclusions:

• The proposed GADF and DSC-based method offers superior performance for rolling bearing fault diagnosis.
• The approach is robust in noisy environments and effective with limited training samples.
• This method provides a significant advancement for industrial machinery health monitoring.