Explainable Dynamic Graph Learning and Multi-Scale Feature Fusion for Hydraulic System Health Monitoring.

Area of Science:

• Engineering
• Artificial Intelligence
• Data Science

Background:

• Hydraulic systems are critical in aerospace and industry, requiring reliable health monitoring.
• Traditional methods struggle with dynamic sensor correlations and multi-resolution feature extraction, especially with limited data.

Purpose of the Study:

• To propose a novel Dynamic Multi-Scale Graph Neural Network (DMS-GNN) for accurate hydraulic system fault diagnosis.
• To address limitations of static graph structures and grid-based models in capturing evolving sensor data.

Main Methods:

• Developed a hierarchical multi-scale feature extraction module for diverse fault signatures.
• Introduced a self-attention-based dynamic graph learner to adaptively infer sensor topologies.
• Validated the approach on an electro-hydraulic test bench.

Main Results:

• The proposed DMS-GNN achieved a diagnostic accuracy of 98.47%.
• Outperformed state-of-the-art methods including GraphSAGE, Static GCN, and GAT.
• Demonstrated robust multi-sensor fusion diagnosis capabilities.

Conclusions:

• The DMS-GNN effectively combines multi-scale temporal learning with dynamic spatial reasoning.
• The method offers a significant advancement in hydraulic system health monitoring.
• Dynamic graph learning is crucial for evolving sensor correlations in fault diagnosis.