| Two factors are important in the development and implementation of successful control strategies for polymerization reactors: (i) availability of kinetic models describing the underlying dynamics of the polymerization reaction and the properties of the resulting polymer, and (ii) availability of on-line sensors to characterize the polymer during the reaction. This work aims to establish appropriate methodologies to deal with these factors.; A mathematical model is developed for a batch methyl methacrylate (MMA) solution polymerization. It incorporates advanced theories of polymer diffusion in the form of size and concentration dependent termination rate constants. The model successfully fits monomer conversion and average molecular weight (MW) versus time data over a range of reactor conditions. Moreover, it has the capability of predicting the entire molecular weight distribution (MWD).; An experimental set-up is described, including a series of on-line instruments to measure polymer properties (density, viscosity and MWD). A process control computer is used for on-line data acquisition and implementation of control strategies. The data indicate that on-line density provides accurate estimates of monomer conversion and that on-line viscosity gives reliable estimates of average polymer MW.; A modified extended Kalman filter is developed, implemented and tested off-line. The filter incorporates fast on-line measurements (density and temperature) with time delayed MWD data to yield optimal estimates of monomer conversion, initiator conversion as well as the entire polymer MWD. The filter is shown to have excellent convergence properties in the face of parameter uncertainty, poor initial estimates and high measurement noises. The application of this two time scale filtering technique demonstrates the feasibility of real time control of unimodal and bimodal polymer MWD's.; Optimal temperature trajectories are derived with the mathematical model to produce a polymer with constant polydispersity. Guidelines on the real time implementation of these policies are discussed. In addition, various adaptive control algorithms are simulated for temperature control of a continuous MMA reactor. The performance of the controllers (self-tuning controller, pole placement based self-tuner and dead-beat controller) is compared to that of a standard PID controller. It is found that adaptive controllers offer better temperature control in the face of parameter uncertainty and large loads. |
