FLIGHT CONTROL SYSTEM KNOWLEDGE GRAPH CONSTRUCTION BASED ON AERONAUTICAL DOMAIN KNOWLEDGE AUGMENTED LARGE LANGUAGE MODEL

In the complex competitive aerospace environment, the continuous intelligent assurance of the Flight Control System (FCS) is essential for the successful execution of aerospace missions. Monitoring and maintaining the FCS under dynamic and complex conditions rely heavily on empirical expert knowledge. Domain Knowledge Graphs (KG) serve as an efficient mechanism for FCS status management within expert systems, enabling effective value extraction. This paper outlines the construction of an FCS KG through large language model (LLM) fine-tuning, enriched with aeronautical domain knowledge derived from multi-source heterogeneous FCS texts. The process begins with the creation of an FCS ontology, integrating aeronautical domain knowledge and defining entity-relationship types. Subsequently, the Llama3-8B LLM was fine-tuned using methods such as LoRA, QLoRA, and AdaLoRA for parameter-efficient tuning, and prior aeronautical knowledge was incorporated via a chain of thought (COT) prompt template to facilitate intelligent discovery from a common FCS dataset. Experimental results indicate that the aeronautical domain knowledgeaugmented LLM method achieved an F1 score of 97.86%, representing a 6.48% improvement over traditional methods. Finally, the FCS KG was visualized using the graph database Neo4j, demonstrating the effectiveness and superiority of this approach in constructing FCS KG.