Two-stage robust estimators for product concentration in a penicillin fermentation

Abstract

Two strategies are proposed for inferential estimations of penicillin concentration in a fed-batch fermentation process. The first scheme employs a partial least squares (PLS)-based estimator to directly estimate the intended product concentration. The second introduces a two-step estimation where biomass concentration is first estimated and used as the input to the final estimator along with other process variables. The results confirmed the strength of the two-stage approach. To further extend the capability of the estimation scheme, a wavelet-based filter is added to the estimation model. The improved performance obtained proved the efficacy of the wavelet filter.