A Spatio-Temporal Joint Diagnosis Framework for Bearing Faults via Graph Convolution and Attention-Enhanced...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearing fault diagnosis is critical for industrial machinery maintenance.
• Existing methods struggle with inadequate modeling of time-space coupling and multi-modal correlations.

Purpose of the Study:

• To propose an advanced joint diagnosis framework for rolling bearing fault diagnosis.
• To enhance the modeling of spatio-temporal characteristics and multi-modal correlations.

Main Methods:

• Utilized graph convolutional networks (GCNs) for spatial feature extraction.
• Implemented attention-enhanced bidirectional gated recurrent units (BiGRUs) for temporal modeling.
• Developed a joint learning architecture integrating GCNs and BiGRUs with a spatio-temporal graph.

Main Results:

• Achieved a classification accuracy of 97.08% on public datasets (e.g., CWRU).
• Demonstrated effective decoupling of bearing signals via dynamic spatial topological modeling.
• Successfully combined multi-scale spatio-temporal features for accurate fault impact characterization.

Conclusions:

• The proposed GCN-BiGRU framework significantly improves rolling bearing fault diagnosis performance.
• The method offers a robust approach for deep fusion of spatio-temporal features.
• This framework accurately captures bearing fault impact characteristics, enhancing diagnostic capabilities.