A bearing fault diagnosis method for unknown operating conditions based on differentiated feature extraction

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Domain generalization is crucial for rolling bearing fault diagnostics under unknown conditions.
• Existing methods may overlook domain-specific features by focusing only on transferable characteristics.
• This limitation hinders accurate fault detection in diverse operational environments.

Purpose of the Study:

• To develop a novel feature decomposition learning method for rolling bearing fault diagnostics.
• To simultaneously extract inter-domain transferable and domain-specific features.
• To enhance the robustness and accuracy of fault detection under unknown operating conditions.

Main Methods:

• A feature decomposition learning approach is proposed.
• Different feature extractors are constructed to capture diverse feature types.
• A joint metric method based on central moment differences extracts transferable features.
• A difference maximization method is employed for domain-specific feature extraction.

Main Results:

• The proposed method successfully extracts both transferable and domain-specific features.
• Experimental results demonstrate superior defect detection capabilities compared to existing approaches.
• The technique shows enhanced performance across two distinct datasets.

Conclusions:

• The feature decomposition learning method provides richer feature information for fault diagnostics.
• This approach effectively addresses the limitations of solely focusing on transferable features.
• The study advances the field of rolling bearing fault diagnosis with improved accuracy and reliability.