This paper reviews and compares artificial intelligence (AI) methods for the automatic tuning of multi-input-multi-output (MIMO) proportional-integral-derivative (PID) controllers in industrial process applications. The study focuses on fixed-parameter PID tuning and introduces a generalised procedure that unifies diverse AI methods within a single autotuning framework. A Pareto-front-based weighting strategy is proposed to balance performance and actuator usage, enabling fair comparison of tuning outcomes across different algorithms. Within this framework, three representative approaches, particle swarm optimisation (PSO), proximal policy optimisation (PPO), and Bayesian optimisation (BO), are implemented and evaluated against the defining criteria of an ideal autotuner: versatility, global optimality, data efficiency, and safety. The analysis bridges computational intelligence and machine learning perspectives, providing a structured benchmark for assessing AI-based tuning performance. Results show that all AI-based tuners successfully identify high-performing controller parameters for multivariable nonlinear systems, confirming their applicability to industrial processes. Among them, BO achieves the best overall performance, offering superior convergence speed and data efficiency through surrogate-driven optimisation. By maximising information gained from each plant trial, BO provides a safe, robust, and computationally efficient tuning method ideally suited to practical industrial deployment.
Niekerk et al. (2026) studied this question.