Multi-bearing fault diagnosis method based on convolutional autoencoder causal decoupling domain generalization.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Current transfer learning (TL) for bearing fault diagnosis often overlooks multi-bearing systems.
• Challenges include limited fault samples and distribution shifts due to changing working conditions.
• Existing domain generalization methods may oversimplify data generation, introducing bias.

Purpose of the Study:

• To develop a robust transfer learning method for diagnosing faults in multi-bearing systems.
• To enhance the generalization ability of fault diagnosis models under domain shifts.
• To accurately diagnose and locate faults using vibration signals from auxiliary bearings.

Main Methods:

• Proposes a causal decoupling network based on convolutional autoencoder (CDN-CAE).
• Utilizes maximum entropy optimization and mutual information neural estimation to decouple time-series data into causal and non-causal factors.
• Introduces aggregation loss for factor separation and reconstruction loss for information completeness and robustness.

Main Results:

• The CDN-CAE method effectively diagnoses and locates faults in multi-bearing systems.
• Demonstrates improved generalization ability compared to existing methods.
• Achieves high accuracy and overall stability in experimental validation.

Conclusions:

• The proposed CDN-CAE method offers a comprehensive approach to multi-bearing fault diagnosis.
• Successfully addresses limitations of existing TL and domain generalization techniques.
• Provides a robust and accurate solution for practical engineering scenarios.