An intelligent Q&A system for vessel equipment faults based on large language models and retrieval augmented generation technology

Objectives To address the low efficiency in fault diagnosis for vessel equipment and the difficulty in effectively communicating with commanders, a solution is proposed. This solution employs natural language interaction to rapidly pinpoint the cause of faults and suggest maintenance plans.

Methods Firstly, based on the domain-oriented design concept, an intelligent question answering system was constructed utilizing a Large Language Model and Retrieval Augmented Generation technology. Subsequently, a set of document preprocessing methods and comprehensive retrieval strategies were introduced to enhance system performance. Finally, a comprehensive evaluation scheme was devised to thoroughly assess the system.

Results Experimental results demonstrate that, by merely using natural language to describe the fault phenomenon, the system can quickly locate the cause of the fault and provide maintenance solutions, significantly improving the efficiency of fault diagnosis. Compared to basic question answering systems, the optimized question answering system has doubled its ROUGE score, achieved a nearly 30% increase in BERTScore, received a 1.5-fold increase in expert ratings, and a 95% reduction in system response time compared to traditional manual retrieval methods.

Conclusions  This offers robust technical support for the rapid recovery of equipment performance on coast guard vessels in complex mission environments, effectively enhancing their combat effectiveness and mission execution capabilities.