Model Predictive Control in the Minerals Processing Industry (5f227471-d147-46e3-b94b-02751537edb1)

The control of minerals processing plants is by no means trivial. In order to address some of the difficulties, it was decided to develop and test a model predictive controller (MPC) for a milling circuit. A milling simulator was written to test the MPC controller. Due to the large number of tuning parameters in a controller of this type, an optimising algorithm was written that automatically tunes the controller. Good, robust control was obtained when the controller was tuned using this algorithm. The controller was tested for various scenarios that are often encountered in the minerals processing industry. When problems were encountered, the MPC algorithm was specially adapted to accommodate for these difficulties. Control was further improved by using feedforward control as well as quadratic programming. The movement of the manipulated variables was reduced by using a region of uncertainty that allows the controlled variable to vary within a specified region around the setpoint. Integrators were handled by adapting the algorithm slightly and adding an optional integral error portion in the MPC controller. The performance of the enhanced MPC controller was compared to PI controllers on both the simulator and a commercial plant. The MPC controller showed faster responses with less interaction in both cases.