A Novel Assembly Process Knowledge Graph Inference Method Integrating Logical Rules and Embedded Learning

The high complexity, repeatability, standardization, and quality requirements of the assembly process presently put forward raised requirements for the unified expression and organization of assembly process knowledge. As one of the core technologies supporting intelligent manufacturing, the construction of an assembly process knowledge graph (KG) becomes a viable method, the inference tasks of which include knowledge judgment, entity, and relation completion, all of which have also become research hotspots. However, most existing knowledge inference methods utilize the triplet form to express the KG, which cannot express all crucial information of assembly process KG construction, including entity, relation, entity type, attributes, etc. Therefore, a KG inference method integrating logical rules and embedded learning is proposed in this paper for KG inference tasks in KG construction. Comprehensively considering the semantic information of entity, relation, and entity type, an improved self-adversarial negative sampling method and logical rules are constructed, which are also introduced in the training process of existing embedded learning models. The proposed method could effectively solve the problems of incomplete assembly process KGs and low construction efficiency. Finally, based on three existing embedded learning models, including DistMult, ComplEx, and RotatE, this paper verifies the effectiveness of the proposed method relative to the above KG inference tasks.