Dual-branch spatio-temporal graph network for bearing fault diagnosis

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Bearing failures in rotating machinery cause significant damage and safety risks.
• Traditional vibration analysis struggles with noisy, irregular signals for effective fault feature extraction.
• Advanced methods are needed to improve bearing fault diagnosis.

Purpose of the Study:

• To propose a novel bearing fault diagnosis method using dual-branch spatio-temporal graph networks (DBSGN).
• To address the limitations of traditional methods in extracting high-dimensional features from complex vibration signals.
• To enhance the accuracy and reliability of bearing fault diagnosis.

Main Methods:

• Vibration signals modeled using spectrum theory to construct spatio-temporal graphs.
• Laplace-based spectral decomposition employed for feature vector extraction.
• A dual-branch fusion network with a dynamic attention mechanism designed for model training and verification.

Main Results:

• The proposed DBSGN method demonstrated superior performance compared to traditional models.
• Experimental results on three benchmark datasets confirmed DBSGN's enhanced stability and accuracy.
• The method effectively extracts features from noisy and irregular vibration signals.

Conclusions:

• DBSGN offers a robust and accurate solution for bearing fault diagnosis.
• The approach improves the availability, reliability, and safety of rotating machinery.
• Spatio-temporal graph networks show significant potential for complex machinery diagnostics.