Teaching Multivariable Model Predictive Control in a Laboratory Scale Three-tank Process

Abstract

This paper proposes to study the potential of use a three-tank system in laboratory scale to teach how to design a predictive controller (MPC) applied to a system with multiple inputs/multiple outputs (MIMO). An algorithm that predicts the future behavior of the plant characterizes the controllers (MPC). With a representative model of the process, the algorithm calculates the future optimal control actions that will minimize the error between the controlled variables and their respective reference values, then, only the first values calculated for the plant inputs is sent. These controllers have a high popularity in the academy and in the industry because they provide high performance control systems without requiring interventions of operators for many hours. These important controllers were emerged initially in the 70’s, in order to overcome occurring difficulties in oil refineries and in power plants. Today, MPCs are also in food processing, aerospace industry, and automotive industry among others. The system used in this study can provide students with practical knowledge of automation and process control that it is not possible to be acquired using only the process simulators or theory. For this purpose, a predictive controller based on a linear state-space model was developed and evaluated to check the influence of tuning parameters in the response of variables controlled and the control actions. The mainly parameters studied were the prediction horizon, control horizon and weight of future increases in the system actuators.