Large language model assisted interactive design of PID controllers

Proportional-Integral-Derivative (PID) controllers remain widely used in industrial applications due to their simplicity and effectiveness. However, designing and tuning PID parameters to meet specific performance objectives often requires expert knowledge and iterative adjustment. Recent advances in large language models (LLMs) present new opportunities for interactive and knowledge-driven support in engineering design tasks. This bachelor’s thesis investigates the feasibility of using LLMs as interactive assistants for PID design and tuning. Two frameworks were considered: SmartControl, with a structured graphical interface and MATLAB Co-Pilot, guided through carefully designed prompts. Simulation-based benchmarking was conducted on a second-order, moderately underdamped system, comparing LLM-assisted tuning against analytical and classical PID methods. Key performance metrics, including rise time, settling time, overshoot, PID gain values and reproducibility across repeated tuning runs, were evaluated. The results indicate that LLM-assisted tuning can achieve comparable and in some cases, faster nominal transient responses with reduced overshoot relative to conventional analytical tuning rules for the considered benchmark system. However, these performance improvements are often obtained through comparatively high controller gains and exhibit variability between repeated tuning runs. This raises important considerations regarding robustness, control effort and practical implementability. Overall, these findings demonstrate that LLMs can serve as valuable tools in the controller design process, while reinforcing the continued necessity of human oversight and constraint-aware validation to ensure that the resulting controllers are both feasible and reliable in real-world applications.