Maintenance Knowledge Inference based on Relational Graph Convolutional Networks

This paper discusses the optimization of maintenance strategy inference by constructing a comprehensive maintenance knowledge graph and applying Relational Graph Convolutional Networks (RGCN). A vast dataset encompassing equipment, their attributes, various fault types, and corresponding maintenance approaches was collected to build the initial maintenance knowledge graph. To further unveil the hidden maitnance link, the RGCN model is introduced. This model significantly enhances the graph’s capability by assigning specific matrices to the relationships between different types of equipment and fault categories. Such an approach enables more precise inference of the most suitable maintenance strategies for specific faults. The knowledge graph is implemented in the Neo4j graph database, which enhances the querying process and the intuitive understanding of fault diagnosis and maintenance strategies. Experimental evaluations demonstrate the effectiveness of RGCN in optimizing maintenance recommendations for equipment faults, thus promoting the advancement of intelligent maintenance practices in industrial settings.