Working-Condition-Indexed Generative Domain Generalization for Intelligent Fault Diagnosis Under Unseen Conditions

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Domain generalization is crucial for cross-domain fault diagnosis, enabling models to perform under unknown conditions without target data.
• Existing methods often neglect inter-domain operating condition relationships, leading to suboptimal generalization.
• Accurate fault diagnosis in rotating machinery is vital for industrial reliability.

Purpose of the Study:

• To propose a novel working-condition-indexed generative domain generalization (WCIGDG) method for data-free fault diagnosis.
• To address the limitation of directionless domain generalization by modeling relationships between operating conditions.
• To enhance the adaptability of fault diagnosis models to target operating conditions.

Main Methods:

• Developed a working-condition-indexed generative domain generalization (WCIGDG) framework.
• Employed a domain index to model operating condition relationships across domains.
• Utilized a domain index predictor to guide data generation and adversarial training for enhanced adaptation.

Main Results:

• WCIGDG demonstrated effectiveness in bearing and gear fault diagnosis across multiple cross-speed generalization tasks.
• The proposed method showed improved average accuracy and F1-Score compared to the state-of-the-art SDGN baseline.
• Experiments on CWRU and WT datasets validated the method's performance in cross-domain fault diagnosis.

Conclusions:

• The WCIGDG method successfully improves fault diagnosis performance under unknown operating conditions by leveraging working-condition parameters.
• Modeling inter-domain operating condition relationships is key to achieving effective domain generalization in fault diagnosis.
• WCIGDG offers a promising approach for robust and adaptable fault diagnosis systems in industrial applications.