Generative AI-Assisted PID Controller Tuning Under Performance and Actuator Constraints

Generative artificial intelligence tools are increasingly being used to support engineering tasks; however, their applicability to process identification, controller tuning, and control performance prediction remains insufficiently investigated. This article evaluates the capabilities of GenAI systems to identify dynamic process models, design PI and PID controllers, and predict closed-loop performance metrics. Five benchmark process models representing first-order, delayed, integrated, and second-order dynamics were considered. Open-loop and closed-loop datasets were provided to two AI systems, which were then requested to identify process models, propose controller parameters, and predict overshoot, settling time, integral of time-weighted absolute error, and disturbance rejection time. The generated controller settings were validated using MATLAB/Simulink simulations, and both absolute and relative prediction errors were calculated. The results show that both AI systems could generate feasible controller parameters that produced stable closed-loop responses for most of the processes investigated. However, significant discrepancies were observed between predicted and simulated performance indicators. The integrating process exhibited the highest prediction errors, indicating substantial difficulties in interpreting integrating dynamics. The results suggest that current GenAI systems can support preliminary controller design and process analysis but cannot yet replace simulation-based validation for accurate quantitative performance assessment.