Abstract: To address the issues of high costs in manual methods, accuracy dependence on personnel quality, and ineffective utilization of historical maintenance experience in advanced equipment maintenance scenarios, an improved retrieval-augmented and LLM chain-of-thought maintenance strategy generation algorithm is proposed. This algorithm introduced an intent recognition module to optimize processing paths. A hierarchical routing mechanism was incorporated to enable multi-layer classification based on key factors. A multi-query retriever was integrated to enhance retrieval effectiveness, and a context compression module was included to reduce redundant information and avoid attention interference. A chain-of-thought reasoning method guided the large language model in progressively generating refined maintenance strategies. In the field of advanced equipment fault diagnosis and maintenance decision-making, this efficient maintenance strategy generation algorithm was validated through experiments and applications in typical aircraft maintenance decision scenarios, demonstrating significant improvements in both the accuracy and timeliness of maintenance strategy generation for advanced equipment.