Extremum-seeking Control of Industrial-scale Fermentation Processes

Abstract: The work presented here is based on two papers, both pertaining to perturbation-based control strategies in industrial fed-batch fermentation processes. The first paper describes a new control strategy for avoiding overflow metabolism in the exponential growth phase of a fermentation, based on analysis of the frequency spectrum of the dissolved oxygen measurement following a periodic perturbation in the feed rate. A controller based on this strategy was tested in pilot scale, where it gave higher specific growth rates and lower concentrations of overflow metabolites during the exponential growth phase of the process compared to a reference strategy currently used to control the process, resulting in approximately 30 % higher biomass concentrations (w/w) 8 h after inoculation in two different processes utilizing different strains. Adding excess substrate at different points in time showed that the controller can detect and respond to excess substrate. In a set-up with inoculum volume decreased to 1/3 of its normal value, the controller compensated for the decreased feed demand whereas the reference strategy caused excessive accumulation of overflow metabolites leading to process failure. The second paper is based on an experimental study in an industrial production-scale (> 100 m3) process, in which sinusoidal perturbations in the feed rate were applied to evaluate the applicability of perturbation-based control strategies and to model the process for the purpose of such strategies. The results indicated that perturbations in the feed rate of the process can give rise to measurable responses in dissolved oxygen measurements without decreasing process productivity and that a second-order model can be used to describe feed and oxygen dynamics in the process. The perturbation frequency range 3.33-5 mHz was identified as suitable for utilization of the model for perturbation-based control and a simple example of an observer is given to illustrate how the model can be used in on-line control.