随着挖掘机等工程机械在我国基础设施建设中的不断普及和应用，对于这些机械设备的高效、智能维修也日益受到工业界的关注与重视。远程智能诊断与维修指导是挖掘机故障维修业务中的重要环节，是加快故障解除速度、提高故障维修效率、降低企业人力成本的必要手段。然而现有的挖掘机远程故障维修业务中仍具有如下问题和挑战：（1）维修指导效率较低：装修手册等理论知识不包含故障的维修方案操作指导，现场维修往往需要依赖专业维修人员；（2）维修方案知识匮乏：历史维修工单中记录了过往发生的挖掘机故障维修过程的全部信息，包括故障起因及维修步骤，这部分维修信息却一直未被利用；（3）行业知识来源有限：现有知识全部来自挖掘机厂商提供的指导手册，范围比较有限，而互联网中存在着大量维修知识未被整理和利用。基于以上问题，本文进行如下重点研究：1.基于历史维修工单的维修知识抽取算法。梳理分析历史维修工单数据，结合挖掘机故障诊断与维修业务流程，构建挖掘机故障维修领域知识图谱本体模型；实现维修知识抽取算法，完成工单维修知识抽取；实现维修知识与原始挖掘机故障维修知识图谱的知识融合，完成对挖掘机故障维修领域知识库的扩充。2.基于人工智能大模型的维修知识生成技术。基于开源的基础大模型开展研究，提出了面向大模型领域化微调的领域知识数据组织方法；设计了面向不同大模型与不同领域知识数据集的微调方法和技术架构；提出大模型领域化微调效果的评价方法；完成挖掘机故障维修场景下知识问答效果最佳的领域专业大模型构建，实现基于大模型的维修知识生成技术。3.知识抽取算法验证与领域专业大模型的微调效果评估。针对历史工单维修知识抽取算法，设计并完成对比实验，完成维修知识抽取算法的实验验证；针对大模型维修知识生成技术，依据大模型领域化徽调效果评价方法，完成挖掘机故障维修领域大模型的问答效果评估，为工业领域大模型应用效果评估提供参考。4.挖掘机智能故障维修系统设计与实现。分析挖掘机故障维修系统的业务和技术需求，完成系统架构设计，划分系统核心模块并进行详细设计；实现基于知识图谱的故障维修引擎与基于大模型的维修服务引擎的系统集成，满足实际挖掘机故障维修业务需求。

With the continuous popularization and application of construction machinery such as excavators in China's infrastructure construction, the efficient and intelligent maintenance of these machinery and equipment has increasingly attracted the attention of the industry. "Remote intelligent diagnosis and maintenance guidance is an crucial component in the excavator fault repair business, serving as essential means to expedite fault resolution, enhance maintenance efficiency, and reduce enterprise labor costs.However, the existing remote fault maintenance of excavators still faces the following problems and challenges: (1) Inefficient reapir guidance: theoretical knowledge such as the repair manuals lacks operational guidance for troubleshooting, often requiring reliance on professional maintenance personnel for on-site repair; (2) Lack of repair scheme knowledge: historical maintenance work orders record all information about past excavator fault repair processes, including the causes and repair steps of the faults. However, this maintenance information has not been utilized; (3) Limited sources of industry knowledge: existing knowledge comes entirely from guidance manuals provided by excavator manufacturers, which are relatively limited in scope, while there is a vast amount of repair knowledge on the Internet that has not been organized and utilized. Based on the above questions, this paper focuses on the following key research areas:1. An algorithm for extracting repair scheme knowledge based on historical maintenance work orders. Analyzing historical maintenance work orders, constructing an ontology model of excavator fault repair domain knowledge graph combining with excavator fault diagnosis and repair business processes; implementing a repair knowledge extraction algorithm to extract repair knowledge from work orders; integrating repair knowledge with the original excavator fault repair knowledge graph, and expanding the knowledge base of excavator fault repair domain.2. Repair knowledge generation technology based on artificial intelligence large models. Based on open-source large models, research was conducted to propose a method for organizing domain knowledge data for domain-specific fine-tuning of large models. A fine-tuning method and technical architecture were designed for different large models and domain knowledge datasets. An evaluation method for the effectiveness of domain-specific fine-tuning of large models was proposed. The construction of domain-specific professional large models with the best knowledge question-answering effects in the excavator fault repair scenario was completed, achieving repair knowledge generation technology based on large models.3. Validation of knowledge extraction algorithm and evaluation of fine-tuning effects on domain-specific large model. Regarding the algorithm for extracting repair knowledge from historical work orders, comparative experiments were designed and conducted to complete the experimental verification of the repair knowledge extraction algorithm. For the large-model repair knowledge generation technology, based on the evaluation method for the effectiveness of domain-specific fine-tuning of large models, the question-answering effectiveness of large models in the excavator fault repair domain was evaluated, providing reference for the evaluation of the application effectiveness of large models in the industrial domain.4. Design and implementation of intelligent excavator fault repair system. Analyzing the business and technical requirements of the excavator fault repair system, completing system architecture design, dividing core modules of the system, and conducting detailed design; implementing system integration of the fault repair engine based on knowledge graph and the maintenance service engine based on large models to meet the practical requirements of excavator fault repair business.