(289a) Nonlinear Model-Predictive Control of An Industrial Polymerization Reactor In the Laboratory

Model-predictive control has been intensively researched since the eighties and numerous implementations of linear MPC have already been realized for stationary processes in industry. However, the promising technology of nonlinear MPC (NMPC), which is applicable in nonlinear and transient processes, has only been implemented in very few chemical processes so far. Although the processes can be operated more efficiently by optimization-based control, the complexity of NMPC is often prohibitive. Improvements of the solution algorithms, methods for developing models with reasonable effort and accuracy, as well as a high reliability of the NMPC control strategy and software are required in order to increase the tolerance towards this technology.

In this work we present an experimental application of nonlinear model-predictive control of a semi-batch polymerization reactor in the laboratory. The objective for process operation consists of the maximization of profit while certain product specifications must be fulfilled at the end of the batch. The resulting nonlinear and complex optimization problem is solved by the dynamic optimization software DyOS on a time-scale of 2 minutes. An underlying time-variant MPC-controller is tracking the optimal trajectory on a fast time-scale of 10 seconds. The experimental setting, where only measurements of temperature, pressure and flow rates are available, is comparable to the setting of an industrial production process. The unmeasurable process quantities are estimated with the help of an extended Kalman Filter.

Besides targeting different product qualities during process operation, an experiment with a pump failure of 10 minutes was successfully carried out. Technical problems and open research questions related to a robust implementation of the control strategy at the reactor are discussed in this contribution. In this aspect, this experimental work contributes to building a bridge between the theoretical research work carried out in the area of NMPC and the implementation of NMPC in a practical application.