Non-linear state estimation and control of emulsion polymerization

Efforts to automate product quality control in emulsion polymerization processes have languished due to the complexity of such processes and the lack of appropriate sensors. Recent advances in in-situ monitoring of emulsion polymerization reactors via near-infra-red (NIR) spectroscopy, together with improved understanding of the reaction itself, have laid a path towards practical on-line control. This study, for example, uses NIR composition measurements with an extended Kalman filter (EKF) to track latex particle growth in a fed-batch reactor. The issues examined include sensor calibration, process modeling, EKF tuning, EKF performance using simulated and off-line data, and EKF interaction with a molecular weight (MW) controller.; An off-line method for measuring residual monomer concentration via UV spectroscopy is presented. The UV data provides a calibration basis set for the NIR spectrometer. A-four-latent variable PLS regression on NIR data spanning 790 to 955 nm provides a linear calibration that typically predicts residual styrene concentration with less than 10% error. The calibration holds for continuous or pulsed addition of monomer.; A mechanistic process model is developed and optimized to fit experimental percent solids, monomer concentration, and particle size data. Optimization requires an empirical correction to the propagation rate constant (k{dollar}sb{lcub}rm p{rcub}{dollar}) that varies from 0.8 to 0.4 depending on the polymer volume fraction. Such a correction is indicative of a diffusion limited reaction such as the well known gel-effect seen in solution polymerizations.; In simulation studies, the EKF proves most sensitive to measurement bias and k{dollar}sb{lcub}rm p{rcub}{dollar} modeling errors. A 10% error in k{dollar}sb{lcub}rm p{rcub}{dollar} produces a 3.5% error in the particle size estimate. This scaling holds for measurement and k{dollar}sb{lcub}rm p{rcub}{dollar} errors ranging from 5 to 25%. The EKF convergence rate depends heavily on the number of latex particles in the reaction mixture--more particles leading to slower convergence. Modal analysis suggests that the cause lies in reduced system observability accompanying the increase in reaction loci. Simulation studies also showed that the EKF and a temperature-based MW controller work well together unless the controller sets the reaction temperature far from its nominal value of 50{dollar}spcirc{dollar}C for extended time intervals.; Off-line studies show that the EKF, using a fixed correction for diffusion limited reactions, performs adequately under "normal" conditions. When process upsets alter the particle formation phase, adequate EKF performance requires a k{dollar}sb{lcub}rm p{rcub}{dollar} correction factor that varies with polymer volume fraction. The off-line studies suggest that the EKF should perform well on-line.