Multiobjective Tuning Technique for MPC in Grinding Circuits

Abstract

We investigate the control challenges in grinding circuits—slow dynamics, long dead times, variable coupling— and the controller tuning challenge, that is, the difficulty in translating operating goals into tuning goals and closed-loop performance. A tuning algorithm for DMC (dynamic matrix control), suitable for the mineral processing industry, is proposed. The tuning problem is posed as a multiobjective optimization problem, in which the tuning goals are directly related to the desired closed-loop performance of process variables. The problem is solved using a compromise optimization, which minimizes the Euclidian distance between a feasible solution and the Utopia solution. Three case studies are presented, which validate the tuning algorithm for DMC in linear and non-linear grinding circuit models. The closed-loop performance obtained with the proposed tuning algorithm is compared to the one obtained through a benchmark tuning technique from the literature. The proposed tuning method has the following features: i) it shapes the closed-loop response according to the goal definitions for linear systems; ii) it requires tailored initial guesses and search spaces to converge to a stabilizing solution in non-linear applications; and iii) it allows the user to specify the desired closed-loop performance behavior in the tuning procedure, allowing the implementation of an adequate controller for each situation.

Keywords

• Grinding circuit
• model predictive control
• dynamic matrix control
• controller tuning
• multiobjective optimization
• compromise optimization