Model Predictive Control for an Industrial SAG Mill

We discuss Model Predictive Control (MPC) based on ARX models and a simple lower order disturbance model. The advantage of this MPC formulation is that it has few tuning parameters and is based on an ARX prediction model that can readily be identied using standard technologies from system identication.

When applied to MIMO systems we call this controller a MIMO-ARX based MPC. We use an industrial Semi-Autogenous Grinding (SAG) mill to illustrate the performance of this controller. SAG mills are the primary units in a grinding chain and also the most power consuming units. Therefore, improved control of SAG mills has the potential to signicantly improve eciency and reduce the specic energy consumption for mineral processes.

Grinding circuits involving SAG mills are multivariate processes. Commissioning of a control system based on a classical single-loop controllers with logic is time consuming, while MPC has the potential to both improve the control performance and the commissioning time and expertise required. The simulation results demonstrate that the MPC based on a MIMO-ARX model is able to provide nice control performance measured by its ability to track an output reference and reject unknown disturbances.

Furthermore, the method used to design the controller represents a systematic method that can be automatized for wide-spread deployment in industrial environments.