Hybrid semantic-structural graph neural network for fault knowledge graph completion in railway operational equipment

Incomplete fault knowledge graphs in railway operational equipment hinder effective fault diagnosis, prediction, and maintenance planning. This study addresses the challenge of completing fault knowledge graph by proposing a hybrid semantic-structural graph neural network (Hybrid S-GNN) that integrates both semantic and structural information for knowledge graph completion (KGC) in railway fault scenarios. The Hybrid S-GNN comprises four key modules: a semantic encoding module that enhances textual fault data representations through contextual enhancement and dynamic weight allocation; a structural encoding module that captures graph topology using multi-view structure encoding combining local aggregation, global path encoding, and relation-aware adjustment; a semantic-structural fusion module leveraging attention mechanisms to balance semantic and structural signals; and an optimization-prediction module employing margin-based ranking loss and context-aware negative sampling for accurate triple prediction. Experiments on real-world railway fault knowledge graphs demonstrate that Hybrid S-GNN achieves a Hits@10 of 80.5 % and mean reciprocal rank (MRR) of 0.640, outperforming state-of-the-art baselines by 5.8 % and 6.1 %, respectively. Ablation studies confirm the critical contributions of each module, validating the necessity of jointly modeling semantic and structural features. This work provides an effective solution to enhance railway fault knowledge graphs and paves the way for advanced fault management applications in railway operations.