(95b) Real Time Optimal Control and State Estimation in Semi-Batch Free-Radical Polymerization Reactors

State estimation is a powerful tool for calculating indirectly important on-line properties of products during polymerization reactions. In free-radical polymerization processes, the trajectory of weight average molecular weight permits to define and control the ending characteristics of a polymer as well as its molar mass distribution. The Automatic Continuous Online Monitoring of Polymerization reactions (ACOMP) provides online measurements of weight average molecular weight and monomer concentration during polymerization reactions. However, for estimating all other process variables involved in the process behavior, it is required to implement real time state estimation for tracking their evolution while the reaction takes place.

In this study, a validated nonlinear dynamic process model is combined with a discrete-time extended Kalman filter (DEKF) and a feedback controller. The DEKF is employed to improve the accuracy of the measurements and compute a full state estimation while the controller regulates the flow of monomer/initiator to follow an optimal trajectory. As it is known, the efficiency and stability of a DEKF depends highly on the proper tuning of its parameters. Therefore, in this work a novel approach for improving the performance and stability of the DEKF is presented. The filter free parameters, which are the covariance matrix of the process, covariance matrix of the noise and error covariance of the initial states, are tuned by using a metaheuristic stochastic global optimization technique.

The feasibility of the approach was validated using a fully automated pilot scale experimental unit linked with ACOMP and a tailor made filter/controller module for on-line state estimation and optimal control. The free-radical polymerization reactions for synthetizing poly-acrylamide using KPS as initiator was utilized as case study. Results illustrate excellent performance and robustness of the filter and controller for achieving the optimal trajectory. Furthermore, real-time MMD is monitored using the DEKF state estimation.