Coupling Fault Diagnosis of Bearings Based on Hypergraph Neural Network

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Fault Diagnosis

Background:

• Coupling faults, involving multiple base fault types and high-order relationships, are common in mechanical equipment.
• Existing diagnostic methods struggle to effectively model these complex, multi-fault interactions.

Purpose of the Study:

• To propose novel hypergraph neural network (HNN) architectures for diagnosing high-order coupling faults.
• To evaluate the effectiveness of two distinct HNN-based frameworks for fault diagnosis.

Main Methods:

• Developed two HNN frameworks: one for feature generation (using HNN to generate negative samples) and another for feature extraction (using multi-head attention and HNN aggregation).
• In the feature generation framework, base faults are nodes and hyperedges connect them; the HNN generates coupling faults.
• In the feature extraction framework, nodes represent fault modes, hyperedges link common failure modes, and inner product correlation diagnoses faults.

Main Results:

• The feature generation framework achieved a diagnostic accuracy of 88.6% for coupling faults.
• The feature extraction framework reached a diagnostic accuracy of 86.76% for coupling faults.
• Both proposed frameworks demonstrated strong performance in diagnosing and analyzing high-order coupling faults.

Conclusions:

• Hypergraph neural networks provide a robust approach for modeling and diagnosing complex coupling faults.
• The two proposed HNN architectures offer effective solutions for mechanical equipment fault diagnosis, achieving high accuracy.