Prior-Knowledge-Guided Graph Attention Network for Fault Diagnosis of Engine Valve Clearance

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Diesel engine fault diagnosis is crucial for equipment maintenance.
• Deep learning excels at feature learning but struggles with signal relationships.
• Existing methods lack interpretability and struggle with sparse data.

Purpose of the Study:

• To develop an interpretable diesel engine valve-clearance fault diagnosis method.
• To integrate domain knowledge with data-driven deep learning.
• To improve fault diagnosis accuracy, especially in low-data scenarios.

Main Methods:

• Converting engine signals into graph data based on spatiotemporal event relationships.
• Utilizing a graph convolutional attention network for feature extraction.
• Implementing a hybrid knowledge and data-driven deep learning model.

Main Results:

• The proposed model demonstrated superior diagnostic performance and interpretability.
• Achieved high accuracy in fault diagnosis even with scarce operational data.
• Effectively captured intrinsic structural information through graph representation.

Conclusions:

• The combined knowledge and data-driven approach offers significant advantages over traditional deep learning.
• This method enhances understanding of fault mechanisms in diesel engines.
• It provides a robust solution for valve-clearance fault diagnosis in challenging conditions.