A Knowledge Graph and Large Language Model Framework for Customized Manufacturing: A Case Study in Furniture Panel Production

Compared to traditional mass production, customized discrete manufacturing—such as furniture panel production—generates massive, heterogeneous, and highly dynamic data throughout its lifecycle (order processing, process design, manufacturing execution). Storing this complex, interconnected lifecycle data in traditional relational databases suffers from high redundancy, weak relevance, dispersed distribution, and storage limitations. To address this, this paper proposes a generalizable knowledge graph-based method for knowledge modeling and fusion with semi-dynamic update capability in customized manufacturing, demonstrated through a case study in custom furniture panel production. Firstly, three ontology models (process, resource, feature) are constructed to structurally represent multi-source production knowledge, resolving issues of high redundancy and weak semantic associations found in traditional databases. Secondly, a bidirectional KG construction method is proposed, integrating top-down ontology modeling with bottom-up entity mining for effective dynamic knowledge integration. Thirdly, a graph embedding-based multi-dimensional knowledge fusion algorithm is designed, combined with order-based Knowledge Graph Fragment (KGF) models, to enable semantic association and reasoning of knowledge nodes across processes. A case study in a furniture factory validated the method, building a KG from the ontology models, confirming the fusion method's effectiveness in significantly improving production knowledge manageability and application efficiency. Furthermore, to overcome limitations (e.g., insufficient intent recognition, generalization) of pure knowledge graph-based QA systems, the constructed knowledge graph was integrated with the DeepSeek Large Language Model (LLM). This integration employs ontology-driven prompting and leverages the LoRA lightweight fine-tuning of the DeepSeek-LLM-7B-Chat. This KG–LLM fusion approach for customized manufacturing QA generates more domain-specific and professional responses than methods using solely LLM or KG. The proposed framework provides effective support for intelligent decision-making and optimization in mass-customization production environments, with validated efficacy in the furniture domain.