Applied Research on an Aircraft Maintenance Assistant Based on a Large Language Model

In recent years, significant progress has been made in the application of LLMs in various specialized fields. However, in the field of aircraft maintenance, intelligent systems face significant challenges due to the high complexity and private data involved. This paper describes the Zhihang Aircraft Maintenance Assistant, an LLM-based intelligent system designed to assist in aircraft maintenance tasks. The system integrates domain-specific knowledge on various aspects of aircraft maintenance and builds a vectorized knowledge base through advanced semantic understanding. Using RAG (Retrieval Augmented Generation) technology, it extracts relevant knowledge from the database based on user input and generates accurate responses to assist maintenance personnel with tasks such as fault diagnosis and troubleshooting. By using a specialized maintenance knowledge base, the system solves the problem of hallucination and ensures greater reliability in generating responses. Zhihang Assistant also provides one-click generation of maintenance record orders and concise summaries of policies and regulations. These features improve compliance with standard operating procedures, streamline workflow efficiency and minimize errors caused by inaccurate information retrieval. Test results show that Zhihang Assistant's average scores for answer relevance, content fidelity and contextual relevance were 0.973, 0.807 and 0.917 respectively, demonstrating its ability to provide highly accurate, relevant and contextually appropriate answers based on real-world maintenance needs.