Multi-Source Information-Based Bearing Fault Diagnosis Using Multi-Branch Selective Fusion Deep Residual Network

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearings are critical in industrial machinery, but traditional fault diagnosis struggles with noisy and attenuated sensor signals.
• Existing multi-source methods face challenges with signal redundancy and fail to account for varying signal characteristics, limiting deep learning performance.
• Adaptive feature extraction tailored to individual signal sources offers potential for enhanced diagnostic precision.

Purpose of the Study:

• To propose a novel deep learning model for robust rolling bearing fault diagnosis.
• To address challenges of signal noise, transmission attenuation, and inter-signal redundancy in multi-source data.
• To improve diagnostic accuracy by adaptively processing diverse signal characteristics.

Main Methods:

• A multi-branch selective fusion deep residual network was developed.
• Each signal source utilizes a unique feature processing channel to prevent blind coupling.
• Adaptive convolution kernel sizes and dropout techniques were employed to extract multiscale features and reduce redundancy.

Main Results:

• The proposed model demonstrated superior performance compared to other intelligent methods.
• Experimental validation on two public bearing datasets confirmed its feasibility and effectiveness.
• The multi-branch approach successfully mitigated issues of information redundancy and signal characteristic differences.

Conclusions:

• The multi-branch selective fusion deep residual network offers a superior solution for rolling bearing fault diagnosis.
• Adaptive feature extraction and selective fusion enhance diagnostic reliability and precision.
• The model effectively overcomes limitations of traditional and existing multi-source fault diagnosis techniques.