Implementing a model predictive controller in an operator training simulator

Abstract

The goal of this thesis was to develop a model predictive control (MPC) based advanced process control system and integrate it into an existing operator training system for a grinding process. The function of the advanced process control system is to optimize a single variable or variables. MPC is based on model-based predictions. Control moves are optimized based on these predictions, so that the target setpoint is reached. In this thesis, the controlled variable was the P80 value, which had not been controlled earlier. Each type of ore has an optimal particle size for the separation process in order to maximize recovery of the ore. Therefore, it is crucial that the P80 can be controlled accordingly.

The experimental part of the thesis started by integrating Metso advanced process control platform and open platform communication (OPC) in order to collect data. Next, an experimental design for data collection was made and 22 data sets in total were collected. These data sets were preprocessed, and first order transfer function models for the model predictive controller were identified and validated. According to the validation, all the models were found to be applicable. The models were configured into MPC, after which some scenarios were simulated. The behaviour of the controlled and manipulated variables during the simulations of various scenarios was analysed.

MPC successfully controlled the P80 to the target setpoint in every scenario and behaved in a robust manner in the disturbance scenarios. Moreover, the developed disturbance model decreased the number of errors in the disturbance scenarios.