LLM Agents Automate PID Tuning for Chemical Processes

A new approach combines both large and small language models to enhance the tuning process of PID controllers in chemical operations. This technique mimics how plant engineers typically work by observing responses, pinpointing problems, adjusting gains, and verifying results. The large language models are equipped with data on closed-loop responses, diagnostics from control engineering, tuning preferences, and examples of internal model control, allowing them to generate and modify PID gains according to established criteria. For local applications, the Qwen3-0.6B model is fine-tuned using supervised methods on simulation-validated targets and focuses on ensuring stability and performance. Testing involved 100 first-order plus dead time and 100 second-order plus dead time systems. The findings are published in arXiv (2607.26594).