Extended Kalman filtering for particle size control in a fed-batch emulsion polymerization reactor

An approach to the analysis of convergence problems in extended Kalman filtering (EKF) is described. In general, filter convergence depends on the state trajectory of the process. If it traverses regions where the states are only weakly observable, the EKF gains may become too small for rapid correction of estimation errors. Such problems were encountered in an emulsion polymerization process application. Reasons for slow convergence were determined through examination of the Kalman gains and the observability of system modes, in combination with physical arguments. A modified EKF was then successfully implemented