Key points are not available for this paper at this time.
The standardization of locomotive maintenance data is a critical step in facilitating reliability centered maintenance (RCM) data analysis for locomotive maintenance. The performance of this analysis directly impacts the overall effectiveness of the RCM approach. However, challenges such as small sample sizes, nonstandardized data formats, complex analyses, and high labor costs make it difficult to standardize data via traditional manual methods. To address these challenges, we leverage the outstanding performance and unique learning capabilities demonstrated by large language models (LLMs), as extensively documented in academic research and industrial applications, to standardize the data related to locomotive maintenance data. This paper adopts a framework based on the premise of "quality data + universal LLMs + fine-tuning". We utilize custom scripts to generate high-quality locomotive maintenance data, integrate the distinct characteristics of such data, and develop customized LLMs specifically designed to standardize locomotive maintenance data via models such as UIE and ChatGLM. Furthermore, we present an auxiliary tool for locomotive maintenance data standardization, along with an intelligent question and answer (Q&A) system, both of which are based on the customized LLM. The proposed Q&A system achieves scores of 86.87% for Bleu-4, 89.60% for Rouge-1, 87.54% for Rouge-2, and 94.26% for Rouge-L on the locomotive maintenance dataset and demonstrates impressive performance, with an auxiliary tool efficiency of only 18 ms per piece. Consequently, the customized LLM can not only enhance the performance of locomotive data standardization but also serve as the basis for developing auxiliary tools and intelligent Q&A systems, simplifying the data standardization process and saving time and costs.
Chen et al. (Tue,) studied this question.