Modeling and Control of Continuous Free Radical Polymerization Reactors

Abstract

Nonlinear dynamics of continuous stirred tank reactors for free radical polymerization of styrene have been studied with complex initiator systems such as a binary mixture of monofunctional initiators and bifunctional initiators. The regions of reactor operating conditions which give rise to steady state multiplicity, multiple Hopf bifurcation points, isolas, period doubling bifurcations leading to period-doubling cascades and homoclinics have been identified. The parametric sensitivity of the reactor has also been investigated during the start-up and the steady state operations. Emphasis of the reactor analysis has been placed on the elucidation of the effect of initiator characteristics on the reactor dynamics and resulting polymer molecular weight properties. It has been observed that the presence of more than one monofunctional initiator or dual initiator functionalities makes the reactor dynamics more complex than a single monofunctional initiator. When the heat transfer coefficient of the reactor wall changes during the polymerization because of viscosity increase, the reactor exhibited simpler dynamic behavior than the case of constant heat transfer coefficient. The presence of reactive impurities such as inhibitors in the feed steams also influenced reactor dynamics significantly, depending on their reactivities and concentrations. A two-time scale extended Kalman filter has been used for on-line estimation and control of polymer molecular weight in continuous and batch polymerization reactors. The effects of model uncertainty and measurement time delay on the filter performance have been investigated through numerical simulations. In the presence of moderate model errors and unknown process disturbances in the continuous polymerization process, the filter showed robust estimation performance in predicting the polymer molecular weight properties when frequent molecular weight measurements are provided. It has been illustrated that the polymer grade change policy can be obtained effectively by using the filter and the steady state process model in the continuous process. The overall filter performance in the batch process was quite similar to that of the continuous process. With relatively large model errors or long measurement time delays, the filter converges slowly. Since the batch processes are operated in finite reaction time, more frequent molecular weight measurements than in continuous processes are required for fast filter convergence.