Feature decoupling and cross domain alignment with transfer learning for cross working condition mechanical fault...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Cross-working-condition mechanical fault diagnosis faces challenges with heterogeneous signals and feature distribution shifts.
• Existing methods struggle with adaptability and interference from varying operational conditions.

Purpose of the Study:

• To propose the WDH-Net model for robust cross-working-condition mechanical fault diagnosis.
• To enhance adaptability and reduce interference from working condition variations.

Main Methods:

• The WDH-Net model integrates time-frequency feature enhancement, dual-stream decoupling encoding, and heterogeneous feature alignment.
• It utilizes Wavelet-ConvNet+Cross-Attention for feature extraction and a dual-stream encoder to separate fault and condition features.
• Heterogeneous multi-kernel maximum mean discrepancy is employed for cross-domain feature alignment.

Main Results:

• WDH-Net demonstrates excellent performance on cross-working-condition fault diagnosis tasks.
• The model effectively enhances fault-sensitive features and reduces working condition interference.
• Experimental validation on CWRU and PU datasets confirms the model's superior generalization ability.

Conclusions:

• The proposed WDH-Net model offers a novel and effective solution for intelligent fault diagnosis in industrial mechanical equipment.
• It provides a methodological reference for addressing challenges in cross-working-condition fault diagnosis.
• The approach significantly improves the generalization capability of fault diagnosis systems.