| Chemical and petroleum processes are nonlinear and have been controlled using linear systems. However, controllers based on linear systems do not perform well for highly nonlinear processes. Methods are needed that directly take into account the nonlinearity of the processes. Model Predictive Control (MPC) has been popular in the industry because of its ability to handle process interactions, dead time, unusual dynamic response and process constraints, and because it does not require rigorous models derived from first principles. Therefore, there have been efforts in the recent past to extend MPC for control of nonlinear processes. Most of the current techniques need a nonlinear model of the process in the form of differential equations. However, for industrial processes, only the step response models have usually been available.;In this thesis, an MPC approach to handle the nonlinearity of processes is presented that can work with step response models. Nonlinearity causes problems if and when the process is highly nonlinear or is operating over a wide range. In such cases the controller model needs to be updated as the process moves from one place to another in the operating region. In the proposed approach, updated controller models for different, operating regions are used. The various models that are needed are obtained through linear interpolation because in industry the plant is not available for extensive testing. A procedure for selecting the regions where experimental models are needed is proposed. The proposed approach is tested on SISO and MIMO example problems for servo and regulatory control. The Dynamic Matrix Control (DMC) algorithm is used as the predictive control algorithm. The mismatch between the controller and process models is considered. The proposed approach shows an improved control performance as compared with an approach where a fixed controller model is used. |
