重庆科技大学学报（自然科学版）2026，Vol.28Issue(1)：90-99,10.DOI:10.19406/j.issn.2097-4531.2026.01.010
基于LLM和GraphRAG的柴油发电机故障维保知识推理方法
A Knowledge Reasoning Method for Diesel Generator Fault Maintenance Based on LLM and GraphRAG
摘要
Abstract
To enhance the knowledge reasoning performance of diesel generator fault diagnosis,a fault maintenance knowledge reasoning method is proposed based on Large Language Models(LLM)and knowledge graph-enhanced Retrieval-Augmented Generation technology.The research data is extracted from enterprise diesel generator fault maintenance records,processed through LLM for semantic standardization to form an ontological knowledge graph covering fault systems,fault types,fault causes,and solution strategies.Specifically,the CRISPE framework prompt engineering is employed to drive the LLM to automatically extract entities and relationships,thereby build-ing a structured fault diagnosis knowledge graph thet is subsequently integrated into the graph database.By integra-ting semantic embedding technology with graph reasoning mechanisms,a knowledge indexing network is constructed to support functionalities such as natural language Q&A,fault logic tracing,and professional maintenance guid-ance.Experimental results demonstrate that the method achieves an overall answer accuracy rate of 94％,with accuracy rates exceeding 90％for core question types.This indicates that the method significantly outperforms tradi-tional RAG,as well as general LLM such as ChatGPT-4o and DeepSeek in terms of fault attribution precision,solu-tion professionalism,and adaptability to complex fault scenarios,exhibiting higher domain adaptability and reasoning capabilities.关键词
大语言模型/GraphRAG技术/知识图谱/故障诊断Key words
large language model/GraphRAG technology/knowledge Graph/fault diagnosis分类
信息技术与安全科学引用本文复制引用
何焱,雷思凡,郭梁柱,杨光富,牛炼..基于LLM和GraphRAG的柴油发电机故障维保知识推理方法[J].重庆科技大学学报（自然科学版）,2026,28(1):90-99,10.基金项目
四川长宁天然气开发有限责任公司科学研究与技术开发项目"场站应急电源远程启动的研究"(2025D716) （2025D716）
