Robot Fault Knowledge Graph Completion Based on Relational Graph Convolutional Network

Abstract

The internal mechanism of industrial robots is complex. When a fault occurs because knowledge is too scattered, relying solely on traditional text search methods often leads to inefficient fault diagnosis. This article designs a knowledge graph based on text data and uses the knowledge graph to integrate fragmented knowledge. Because the existing data cannot completely cover the entire natural world, and there are problems in the correctness and timeliness of knowledge, a knowledge graph completion method is proposed that combines knowledge graphs and relationship graph convolutional networks incorporating attention mechanisms(A-RGCN). The attention mechanism it contains is to overcome the fixed selection of normalization constants for adjacent node message aggregation in the RGCN and capture the relationship attribute characteristics between different entities. This method combines the advantages of the knowledge graph, attention mechanism and RGCN. It captures the relationship attribute characteristics between different entities, extracts the internal correlation between entities, and finally integrates all knowledge. The experimental results validate the superior performance of the proposed model.