Sensor fusion based on Extended and Unscented Kalman Filter for bioprocess monitoring

Abstract

Availability of measurements of key process variables is of great importance to allow real time monitoring and control applications. However, in microbial fermentation processes, the unavailability of sensors or their high cost is a major barrier in automated process control applications. In this study, we implemented an Extended Kalman Filter (EKF) and an Unscented Kalman Filter (UKF), both augmented to take state constraints into account, in order to estimate biomass formation, sugar consumption and CO2 formation in a fed-batch bacterial cultivation process. The filters use a simple monod-growth model combined with an in-situ absorbance probe and an infrared off-gas measurements device for monitoring of biomass production and substrate consumption. We tuned the covariance matrices of the filters by a dedicated experiment, and tested their performance on independent set of experiments. Our results demonstrate precise estimation of the biomass and glucose consumption during the batch and the feeding phases, particularly when the process covariance matrix is adapted to the phases to account for model inaccuracies during the feeding phase.