MetaIndux-PLC: A Control Logic-Guided LLM for PLC code generation in industrial control systems

Programmable Logic Controllers (PLCs) are widely used for automation control, and they are well-suited for industrial systems control tasks. LLMs can assist engineers in streamlining the programming process and reducing development costs, and one of the key issues is the construction of the PLC code dataset. However, the lack of an open-source PLC code dataset in the research community, combined with the high complexity of industrial systems control logic, has caused most LLMs to struggle with generating accurate control code. This complexity arises from the need to manage real-time sensor data fusion, integrate various communication protocols, and ensure compliance with stringent safety and regulatory standards. In this study, we construct ST4Indux, a PLC code dataset specifically for industrial systems control. And we propose the Control Logic-Guided Iterative Fine-Tuning (CLIFT) method, which iteratively optimizes the model's generation capability. Based on these, we train a large language model named MetaIndux-PLC, to enable the generation of complex motion control code. Additionally, we propose a multi-dimensional evaluation and optimization method to systematically assess the model's performance in terms of task completion quality, efficiency, and user experience. The experimental results demonstrate that the proposed approach significantly enhances MetaIndux-PLC's performance and reliability in real-world engineering environments, providing a foundation for the future development of intelligent programming assistance systems.