Study on Fault Diagnosis Method Based on Recommender System

To improve the efficiency and accuracy of fault diagnosis in complex industrial systems, this paper proposes a recommender-system-based method that integrates Knowledge Graphs (KG) and Graph Convolutional Networks (GCN). A multi-layer fault knowledge graph is constructed using “fault phenomenon - fault cause - solution” triples to represent structured semantics and enable multi-hop reasoning. By leveraging the propagation mechanism of GCN, high-order neighbor relationships within the graph are effectively captured, allowing the model to infer distant semantic correlations. In the context of recommender systems, fault phenomena and potential causes are analogized to user-item interactions, and personalized preference modeling is performed by constructing a collaborative knowledge graph, which integrates the knowledge graph with the user-item interaction graph, recursively propagating embedding vectors to capture high-order connectivity, and uses an attention mechanism to distinguish the contribution of different neighbors, thereby capturing high-order correlations among entities and the potential causes of fault phenomena. Experimental results show that the proposed method outperforms traditional models in fault identification accuracy on the turbine and its subsystems dataset. It also demonstrates superior robustness and generalizability in scenarios characterized by sparse data and complex failure chains.