A Large Language Model-Driven Agile Reconfiguration Method for Virtual Reconfigurable Production Line

Compared to traditional fixed production lines, reconfigurable production lines can swap equipment and tooling, and alter line layouts as needed, significantly enhancing flexibility. Such production lines are better equipped to handle orders with diverse product varieties and fluctuating batch sizes, as well as unexpected disruptions during production. However, downtime caused by reconfiguration is inevitable and must be minimized to reduce economic losses. Some researchers have approached this from a task optimization perspective, developing techniques to generate reconfiguration plans more rapidly or to derive optimal line layouts. Yet, in practice, large-scale one-off reconfigurations are rare; most scenarios involve rapid, localized adjustments in specific segments due to order changes or equipment failures. Moreover, the process from receiving reconfiguration requirements, formulating a plan, validating it, to deployment relies on multidisciplinary expertise and collaboration among various personnel, considerably prolonging the reconfiguration timeline. With this in mind, this paper proposes a method for rapid production line reconfiguration driven by large language model (LLM) and simulation platforms. The study begins by modeling the production line from multiple dimensions, summarizing and integrating process knowledge. Subsequently, the LLM translates reconfiguration requirements in natural language into concrete reconfiguration plans, which are then deployed and validated on a simulation platform, enabling swift generation and verification of reconfiguration strategies.