Multivariable Model Predictive Control of a Simulated SAG plant

Abstract

This work presents the dynamic simulation and a multivariable predictive control strategy for a semi-autogenous grinding (SAG) plant. Models for SAG Mill, crusher and hidrocyclones have been integrated to simulate a SAG circuit. Data obtained from an industrial cooper concentrator has been used to estimate the main parameters of the dynamical models. The control strategy considers two levels: stabilizing and optimizing levels. The stabilizing level is based on a Model Predictive Control strategy with restrictions. In order to solve the optimization problem posed by this strategy, a standard Linear Programming solver has been selected. The optimizing level maximizes Power Draw using an online gradient estimation of the non-linear function between Power Draw and mill Hold Up. The results show that is possible to operate the SAG Mill in the optimal zone, compensate unmeasurable disturbances and maximize throughput without violating operational constraints. Moreover, the proposed control strategy has a modular structure; therefore its implementation can be carried out by parts. All the work has been done using Matlab® and the simulation platform Simulink®.