A Large Language Model-Based Method for Generating Assembly Process Instructions for Complex Aerospace Products

To address the challenges in compiling assembly instructions for complex aerospace products, including difficulties in multi-source knowledge integration, low efficiency in standard language matching, and reliance on manual experience for process element correction, an intelligent instruction generation framework integrating knowledge graphs and large language models is proposed. Firstly, a dedicated knowledge graph for the assembly process domain is constructed to achieve unified semantic representation and structured storage of multi-source heterogeneous knowledge. Secondly, a hierarchical pruning VF3-DE subgraph matching algorithm is designed to support efficient reuse of standard assembly instruction language. Furthermore, a RoBERTa-BiSRU-CRF-based model for process element recognition and correction is introduced to automatically generate compliant assembly instructions. Validation confirms the method's stability and accuracy, greatly improving the efficiency and quality of assembly instruction authoring.