An LLM-Enhanced and Element-Based Knowledge Graph Framework for Process Route Planning in Customized Production of Complex Heavy Equipment

Customized production of complex heavy equipment is challenged by fragmented production-element sources, heterogeneous data structures, and limited reuse of process knowledge. This study develops a large language model (LLM)-enhanced, element-based knowledge graph framework to support process-route knowledge organization and planning assistance. First, a domain ontology is defined to represent equipment structures, manufacturing processes, operational resources, and process parameters. DeepSeek-V3 is then used to extract entities, relations, and attributes from unstructured technical documents, while a rule-based parser converts structured process data into triples. Synonymous entities are subsequently aligned using Sentence-BERT, and the fused knowledge is stored in Neo4j. The framework is evaluated in the customized production of a 40 MN extrusion press, whose key components are primarily machined using an XK2860 CNC machine tool. The resulting graph contains 4,760 entity nodes and 5,428 relationships. Based on 150 CSV records held out before finalizing the parsing rules, rule-based conversion achieved a triple-level precision of 100.00%, a recall of 97.80%, and an F1 score of 98.89%. Within this controlled case-study evaluation, the results support the feasibility of using the framework to provide a structured representation of production elements and an evidence-traceable knowledge foundation for graph-based organization of candidate process routes.