Real-Time Nonlinear State Estimation in Polymerization Reactors for Smart Manufacturing

A high order Geometric Observer (GO) and a hybrid discrete-time extended Kalman filter (h-DEKF) are assessed in real-time towards data reconciliation and comprehensive polymer characterization. The nonlinear estimators are tested under a set of experiments using the data provided by the state-of-the-art smart sensor Automatic Continuous Online Monitoring of Polymerization reactions (ACOMP). To design the GO three different high-order structures are evaluated in terms of their robustness and performance. The gains of the GO are set following a systematic approach with a distinction between pure measurements and Lie derivatives. The h-DEKF design follows an auto-tuned error driven initialization of the free-parameters where different metaheuristic algorithms are employed for tuning. Both nonlinear estimators are contrasted for applications in the Smart Manufacturing of polymers. The aqueous polymerization of acrylamide using potassium persulfate as initiator demonstrates the effectiveness and flexibility of the estimator for real-time applications.