Optimization Strategies for Maintenance Schemes Based on Large Language Models and Prompt Engineering

With the advent of the intelligent decision-making era, the maintenance field is transitioning from traditional human-led approaches to collaborative human-machine methodologies. This evolution presents a significant challenge: effectively integrating the experience and creativity of human decision-makers with the computational capabilities, extensive knowledge storage, and logical reasoning capabilities of large language models (LLMs). This paper explores the use of LLMs and prompt engineering techniques to enhance maintenance strategies for mechanical equipment, aiming to foster cross-domain knowledge integration and develop dynamic, adaptable maintenance schemes. Our research involves collecting and preprocessing datasets for model fine-tuning, selecting pre-trained models, and applying fine-tuning techniques. We have also innovatively designed various categories of prompts, including few-shot and instructional prompts, to steer the model towards generating practical information and recommendations. Through iterative testing and adjustment, this study continuously refines these prompts. Strategies include rephrasing problem descriptions, modifying the granularity of the information, and incorporating new contextual data. These adjustments aim to increase the efficiency and precision of maintenance strategies informed by LLMs.