Research on the Intelligent Fault Diagnosis Method for the Rolling Bearing System of an Aeroengine Based on GNN and Multi-Source Information Fusion Theory
Guoying Pang, Wenwei Yang, Jun S. Liu, Yangyang Yu, Zhiru Liu, Rui Wang · 2025
Based on the theory of GNN and multi-source information fusion, an intelligent fault diagnosis method under different working conditions is studied. The research focuses on a dataset comprising diverse sensor signals collected from the rotor and shell of an actual aero-engine. GNN network diagrams are utilized in fault diagnosis applications. Eight different network architectures and three pooling schemes of both CNN and GNN are systematically compared and analyzed. Research findings indicate that GNN surpasses CNN regarding generalization, fit degree, prediction accuracy, classification ability, computational speed, and recognition rate. The optimal combinations identified are GraphSage-EdgePool and MLP-EdgePool. Compared to TopkPool and SAGPool, EdgePool demonstrates lower error rates, higher fit-recognition rates, and superior classification capabilities. The solid theoretical foundation for the diagnosis of intelligent faults is provided in the rolling bearing system.