Coupling Fault Diagnosis Based on Dynamic Vertex Interpretable Graph Neural Network.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Mechanical equipment failures often involve coupled faults due to component interactions throughout their lifecycle.
• Independent fault diagnosis is insufficient for comprehensive equipment health management in real-world conditions.

Purpose of the Study:

• To propose a novel Dynamic Vertex Interpretable Graph Neural Network (DIGNN) for accurate mechanical equipment coupling fault diagnosis.
• To enhance the interpretability and efficiency of fault diagnosis models in industrial environments.

Main Methods:

• Utilized wavelet transform for interpretable feature extraction and reduced training uncertainty during data preprocessing.
• Developed a fault topology with dynamic vertices, connecting them to all other nodes based on fault coupling information.
• Implemented a DIGNN model where time series data is fed only into dynamic vertices during testing for classification and analysis.

Main Results:

• Achieved 100% accuracy in diagnosing independent faults within the dataset.
• Successfully determined coupling modes of faults with a comprehensive accuracy of 88.3%.
• Demonstrated the interpretability of DIGNN by analyzing features extracted across different network layers.

Conclusions:

• The proposed DIGNN method enables accurate and interpretable diagnosis of both independent and coupling faults in mechanical equipment.
• The dynamic vertex approach facilitates efficient fault diagnosis in industrial production environments.
• DIGNN offers a promising solution for advanced equipment health management systems.