An interpretable fault diagnosis method for aeroengine bearings based on belief rule based with a dynamic power set

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Data Science

Background:

• Accurate aeroengine bearing fault diagnosis is vital for operational safety and cost reduction.
• Current black-box models, like deep neural networks, lack interpretability, hindering trust in their predictions.
• Real-world industrial data often presents challenges such as imbalanced sample categories and local ignorance, reducing model accuracy.

Purpose of the Study:

• To propose a novel fault diagnosis method for aeroengine bearings that enhances interpretability and accuracy.
• To address limitations of existing models, including lack of transparency, data imbalance, and local ignorance.
• To develop a diagnostic model that provides credible and reliable fault identification.

Main Methods:

• A diagnostic model based on an interpretable belief rule base with a dynamic power set (D-HBRBP-I) was developed.
• The P-CMAES algorithm was employed to optimize the model, incorporating interpretability constraints.
• The proposed method was evaluated using an aeroengine-bearing dataset.

Main Results:

• The D-HBRBP-I model effectively addressed issues of sample category imbalance and local ignorance.
• Optimization with the P-CMAES algorithm ensured the interpretability of the diagnostic model.
• The proposed method achieved a high accuracy of 99% in aeroengine bearing fault diagnosis.

Conclusions:

• The developed interpretable belief rule base with a dynamic power set (D-HBRBP-I) offers a reliable solution for aeroengine bearing fault diagnosis.
• The method successfully overcomes the limitations of traditional black-box models in industrial applications.
• The study demonstrates the potential of interpretable AI models in critical engineering systems.