An automated recipe generator for semi-batch solution radical copolymerization via comprehensive stochastic modeling and derivative-free algorithms

Abstract

Knowledge of the average composition and molecular weights of copolymers containing functional groups is often not enough to ensure product quality, as the distribution of the reactive moieties among the chains affects performance. This work couples a kinetic Monte Carlo model, previously verified as providing an accurate description of the semi-batch radical solution copolymerization of butyl methacrylate with 2-hydroxyethyl methacrylate under constant-feed higher-temperature conditions, with an optimization procedure designed to maintain the robust features of starved-feed operation (with no on-line measurements of reactant concentrations or polymer properties) while reducing batch time and keeping product properties at target values. The optimizer calculated piece-wise time-varying dosing strategies for monomer, comonomer, and initiator feeds that reduced reaction time by 75% while improving the uniformity of the product molecular weight, composition, and non-functional fraction. The strategy was successfully tested in a 1 L lab-reactor, and then scaled to a larger test system of 5 L to demonstrate the feasibility of significantly reducing the total reaction time while simultaneously improving the quality of the resin.