基于大语言模型的航天器健康管理知识图谱构建与应用OA

To address the complexity of on-orbit operational data of spacecraft and the dispersion of expert knowledge,this paper constructs a spacecraft fault knowledge graph,which provides associated knowledge assistance and logical reasoning support for anomalous conditions.A knowledge-based question answering method for spacecraft health management is proposed by in-tegrating knowledge graph embedding with large language models(LLMs).First,knowledge graph embedding techniques are employed to perform vectorized representation of entities and re-lations,enabling the deep mining of complex semantic associations among equipment.Second,triple-based vector computation is combined with a Double-Array Trie(DAT)to optimized query efficiency,and domain knowledge alignment and fine-tuning of the LLM are performed.On this basis,a full-link fault QA process is designed,encompassing semantic parsing and candidate an-swer generation,while a vector database is utilized to achieve efficient retrieval and similarity matching of high-dimensional semantic information.Experimental and application analysis dem-onstrate that the proposed method significantly improves the accuracy and response efficiency of spacecraft fault diagnosis,providing a new technical approach for spacecraft health management.