Automatic tuning of a MIMO PI controller of a flotation bank

Abstract

The literature on the automatic tuning of PID controllers is surveyed. The automatic tuning methods are sorted into model based and model-free methods. The methods are further subdivided into the manner the system is perturbed. The method of Bayesian optimization is presented and discussed within the context of automatic controller tuning. The method used to constrain the Bayesian optimization is presented. The level flotation model is given and linearized. The controllers are given and discussed. The controller tuning strategies for both SISO and MIMO controllers are presented. A Bayesian optimization automatic tuner is implemented on SISO and MIMO PI controllers used to control the pulp levels in a flotation bank. The implemented automatic tuner achieves performance improvement for both SISO and MIMO cases without any noise present. The MIMO controller tuning is also implemented on a system with measurement noise present and the Bayesian optimization automatic tuner settings performed on par with a state of the art forward-feeding controller. The Bayesian optimization automatic tuner is constrained to ensure safety and stability. The constraints are found using a structured singular value analysis.