An experimental study of adaptive Kalman filtering in emulsion copolymerization

Abstract

Monitoring and control of emulsion polymerization reactors presents several difficulties because of profound modeling limitations and on-line measurement problems. The use of Kalman filtering to obtain optimal estimates of the process states in the presence of modeling inaccuracies, process disturbances and limited measurements is investigated. Different filter designs were developed and experimentally tested. Adaptive filtering was found to be extremely useful in tracking time-varying model parameters of large uncertainty and more robust in the case of unknown process noise statistics. The model used in all Kalman filter designs describes accurately the polymerization kinetics but grossly oversimplifies the thermodynamic relationships for the monomer distribution between the aqueous and polymer phases as well as the time dependency of the average number of radicals per particle. Despite this apparent process/model mismatch acceptable estimates of the process states were obtained.