ONLINE MODELING AND PREDICTIVE CONTROL OF AN INDUSTRIAL TERPOLYMERIZATION REACTOR

Large-scale industrial polymer reactors are typically multivariable, nonlinear, and often have significant time delays. Furthermore, key process measurements are sometimes not available at all, or they become available only after long laboratory analysis has rendered them obsolete. This paper reports on the development of a control system for such a reactor. The specific process in question is used to manufacture the polymer 'P' from three monomers 'A', 'B', and 'C', in a continuous stirred-tank reactor. Product quality measurements are available only by laboratory analysis from samples taken every two hours; however, the mole fraction of the monomers, catalyst, etc in the reactor are available every 5 minutes via chromatographic analysis. The control scheme involves a two-tier system in which the monomer, catalyst, etc, flow rates are used to regulate reactant composition in the reactor at the first tier level, every 5 minutes. At the second tier level, reactant composition target values are used to regulate final product properties. A dynamic kinetic model supplies on-lione estimates of the product properties between the two-hour samples. The entire control scheme, implemented on a real-time process control computer, has resulted in significant reduction in product variability, with the consequent benefits of improved yield and product quality. Pertinent facts regarding the design and implementation of the control scheme are summarized along with some results representative of typical performance.