Evaluating Large Language Models for Graphical PLC Programming: A Temperature Control Case Study

This study investigates the use of large language models (LLMs), specifically ChatGPT-4o, to support Programmable Logic Controller (PLC) programming using graphical languages. Existing solutions focus mainly on Structured Text (ST), with limited exploration of Ladder Diagram (LAD) and Function Block Diagram (FBD). This work addresses the gap in evaluating LLM-generated control logic for complex systems, such as temperature cascade control. Two strategies for interacting with the model were tested: incremental refinement and full task representation. A laboratory test bench with a PLC controller was used to implement and verify the generated programs. Results show that LLM adequately handles single-loop control but encounters challenges with cascade control, including logical inconsistencies. Manual feedback and corrections were necessary to achieve a working code. The results suggest that while LLMs can support PLC development, they require human oversight and could benefit from improved adaptation to graphical languages.