Online Model-Adaptation for an Uncertain Liquid-Liquid Mixer System

Abstract

In this paper, a practical modeling approach for a batch-operated, stirred reactor is investigated that shows significant deviations in system behavior to the same actuation applied over multiple batches. Starting from an easily identifiable, sub-optimal, linear dynamical model to describe the basic dynamic behavior of the system, an augmented model reformulation is defined. This is used to model observed, nonlinear drift effects and static differences that can be seen in the system response over multiple batches, which can finally be estimated using an extended Kalman filter (EKF). The quality of this adaptive model approach that can be used in an online fashion is further assessed using a prediction error for all performed experiments.