A Large Language Model-Based Question Answering System for Unmanned Ground Vehicles Maintenance Knowledge

With the continuous advancement in the design sophistication of unmanned ground vehicles (UGVs), the exponentially increasing volume of maintenance documentation poses substantial challenges to the maintenance operations. Aiming at tackling the problems of low knowledge utilization and insufficient interactivity in the existing UGV maintenance processes, this study introduces a Large Language Model (LLM)-based Question Answering (QA) system specifically designed for UGV maintenance knowledge. The system utilizes the LLM as an interactive bridge between maintenance personnel and maintenance knowledge repositories, allowing users to engage with the knowledge base via a natural language interface. Through the application of LoRA (Low-Rank Adaptation) fine-tuning and the implementation of Reranker enhanced Retrieval-Augmented Generation (Re-RAG), the system significantly improves the LLM's proficiency in understanding and applying UGV maintenance knowledge. The efficacy of the proposed approach was verified using real-world QA datasets. Experimental results show that the method achieves excellent performance in statistical metrics, semantic similarity metrics, and accuracy metrics, confirming that the developed system can effectively enable UGV maintenance knowledge question answering.