Control of Nonlinear Processes with Model Uncertainties

Several processes such as high purity distillation columns, chemical reactors, fermentors and neutralisation in chemical and process industries exhibit time-varying and nonlinear dynamic characteristics. Improved and robust control of these processes is becoming a necessity due to push factors such as increasing competition and environmental considerations, and pull factors such as availability of advanced technology and inexpensive computing power. Many model-based control (MBC) techniques have been proposed and analysed for nonlinear processes. These can be classified into two broad categories - internal model control (IMC) and model predictive control (MPC). For linear processes, MPC techniques such as dynamic matrix control has become a popular choice. Also, IMC as a general structure that uses a linear model in parallel with the (actual) process, has become popular among practicing engineers. For nonlinear processes, both IMC and MPC techniques are finding increasing applications in the industry.

An explicit model of the process is necessary for designing model-based controllers. Modeling errors areunavoidable in practice due to the simplifying assumptions in developing the model and/or parameter values used in the model, and these errors or uncertainties affect the performance of model-based control techniques. In our research pursued by a doctoral student (Hu Qiuping), two approaches - parameter adaptation and augmentation, to reduce the effect of model uncertainties on IMC were studied. An IMC controller is first designed via exact linearisation of a suitable model of the nonlinear process. In the parameter adaptation approach (Fig. 1), one or more uncertain parameters in the model are estimated on-line using the difference between process and model outputs, and then used in the controller and model. In the augmentation approach, difference between the process and model outputs is fed back with a gain, and added to the input of the controller (Fig. 2). Both the approaches were successfully applied to typical processes (such as for pH and level control in a neutralisation process, shown in Fig. 3) via simulation. Single-input single-output and multivariable processes, time delay, input constraints and experimental evaluation were also considered. Adaptation and augmentation techniques have produced markedly improved performance for pH control on an experimental neutralization process. Both the approaches have also been applied successfully to an industrial multi-stage evaporator system via simulation. The results of this application carried out in collaboration with Prof. M.O. Tade of School of Chemical Engineering at Curtin University of Technology, show significantly improved control due to adaptation and augmentation.

In summary, better control of uncertain nonlinear processes can be achieved by additional feedback via parameter adaptation and/or augmentation. A collaborative study on the application and evaluation of proposed techniques for uncertain nonlinear processes, to industrial processes will be of great interest and benefit to both the industry and academia. It will involve several steps such as selection of the process, development of a dynamic model, design of nonlinear controllers, application and evaluation of parameter adaptation and augmentation, and implementation aspects. Suggestions and proposals from the process industries and control companies, are solicited.

Figure 3: Control of Level and pH in a Neutralisation Process by IMC and Adaptive IMC; Augmented IMC (not shown in the figure) gave almost perfect control.