This study presents a novel approach to construct a fault knowledge graph for weapon systems, aiming to enhance their health management capabilities. The proposed method leverages large-scale models to perform knowledge extraction, extracting critical information from system structure principles and fault case data. The extracted knowledge is then integrated and stored in a Neo4j graph database, facilitating visualization and query through a user-friendly interface. To enable reasoning within the graph, Cypher language is employed, reflecting the expert's logical reasoning process. The method is demonstrated through the construction and reasoning of a fault knowledge graph for a specific rotary-wing UAV, confirming its feasibility and effectiveness in enhancing the understanding and management of weapon system faults.