A fault diagnosis method for rolling bearing based on gram matrix and multiscale convolutional neural network

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearings are critical for rotating machinery safety and reliability.
• Intelligent fault diagnosis algorithms for bearings have advanced, but noise interference remains a challenge.
• Complex working environments introduce noise into vibration signals, complicating fault diagnosis.

Purpose of the Study:

• To propose an end-to-end bearing fault diagnosis method robust to noise.
• To enhance the accuracy and generalization capabilities of bearing fault diagnosis models.
• To address the limitations of existing methods in noisy environments.

Main Methods:

• Gram Matrix (GM) for noise reduction of vibration signals.
• Multi-scale Convolutional Neural Network (MSCNN) for feature extraction at various scales.
• Feature enhancement branches utilizing undenoised signals to enrich model representations.

Main Results:

• The proposed GMSCNN method demonstrated strong noise robustness.
• Experimental analysis on two bearing datasets validated the effectiveness of the approach.
• The method successfully captured vibration signal characteristics across different frequencies and time scales.

Conclusions:

• GMSCNN offers a powerful solution for intelligent bearing fault diagnosis in noisy conditions.
• The integration of GM and MSCNN with feature enhancement improves diagnostic accuracy and reliability.
• This approach contributes to enhanced safety and maintenance of rotating machinery.