LLM-based intelligent Q&A system for railway locomotive maintenance standardization

Area of Science:

• Artificial Intelligence
• Mechanical Engineering
• Data Science

Background:

• Standardizing locomotive maintenance data is crucial for effective reliability-centered maintenance (RCM).
• Traditional manual data standardization methods face challenges including small sample sizes, nonstandardized formats, complex analysis, and high labor costs.

Purpose of the Study:

• To leverage large language models (LLMs) for efficient and accurate standardization of locomotive maintenance data.
• To develop customized LLMs and associated tools to address the limitations of manual data standardization in RCM.

Main Methods:

• A framework combining quality data generation, universal LLMs, and fine-tuning was employed.
• Custom scripts were used to generate high-quality data, and customized LLMs (e.g., UIE, ChatGLM) were developed for standardization.
• An auxiliary tool and an intelligent question and answer (Q&A) system were built upon the customized LLM.

Main Results:

• The customized LLM demonstrated significant capabilities in standardizing locomotive maintenance data.
• The Q&A system achieved high scores (e.g., Rouge-L: 94.26%) on the locomotive maintenance dataset.
• The auxiliary tool exhibited high efficiency, with a processing time of 18 ms per data piece.

Conclusions:

• Customized LLMs can significantly enhance locomotive data standardization performance.
• The developed LLM serves as a foundation for auxiliary tools and intelligent Q&A systems, simplifying the process.
• This approach offers substantial time and cost savings in locomotive maintenance data management.