Succinct Aero Engine-KG: A Knowledge Graph Framework for Pre-Retrieval in Fault Diagnosis

The increasing structural and functional complexity of aero engine systems, harsh operating environments, and high maintenance and repair costs have posed significant challenges to fault diagnosis. The development of LLM (Large Language Model), KG (Knowledge Graph), and RAG (Retrieval-Augmented Generation) technologies offers new prospects for formulating fault diagnosis solutions. In response, this paper proposes a collaborative RAG pre-knowledge enhancement architecture based on LLM and KG. This architecture is designed to improve the accuracy and completeness of fault diagnosis knowledge acquisition. Consequently, it will support the subsequent efficient construction of fault resolution plans. The architecture incorporates innovative methods such as Using LoRA fine-tuning instead of traditional KG ontology modeling, optimized entity-relationship extraction, and KG essential feature representation, enabling end-to-end knowledge extraction from unstructured text to structured triples. This enhancement of the domain adaptability and semantic coverage of the QA system is a significant contribution of this study. The experimental results demonstrate the capacity of the proposed framework to accurately identify user intent and retrieve relevant terms from the knowledge graph for subsequent RAG-based retrieval.