A novel assembly knowledge graph construction framework enhanced by large language model

With the rapid evolution of intelligent manufacturing technologies, the complexity of assembly processes has significantly increased. To address the growing demand for enhanced automation and intelligence in manufacturing systems, there is an imperative need for an efficient knowledge organization methodology, particularly one that leverages knowledge graph technology. Nevertheless, the efficient construction of assembly knowledge graphs that integrate multiple knowledge types remains a critical challenge in the field. In response to this challenge, this paper constructs quintuples format to express all significant information of assembly knowledge instead of triplets. Based on novel quintuples format, a novel framework for constructing a multi-type assembly knowledge graph (KG) augmented by large language model (LLM) technology is proposed for utilizing heterogeneous assembly data sources. The proposed framework systematically decomposes the assembly KG construction process into five distinct tasks, employing LLM fine-tuning techniques for text processing and KG inference mechanisms for knowledge integration. This integrated approach enables the automated extraction, consolidation, and construction of multi-type assembly KGs from extensive heterogeneous assembly process documentation. The methodological framework facilitates both structured storage of assembly knowledge and automated KG construction. Experimental validation demonstrates that the proposed framework significantly enhances assembly knowledge acquisition efficiency, thereby advancing the practical implementation of intelligent manufacturing solutions.