Experimental investigations of a robust control strategy applied to cultures of S. cerevisae Laurent Dewasme, Frederic Renard and Alain Vande Wouwer Abstract— In this paper, two robust RST control schemes for the regulation of either the ethanol concentration or the dissolved oxygen concentration in cultures of S. cerevisae are presented and illustrated with various simulation and experimental results. Both controllers only require the prior knowledge about one stoichiometric coefficient and only one on-line measurement signal, making them easily implementable in an industrial environment. Disturbance rejection is ensured thanks to an on-line parameter adaptation procedure, which delivers as a side product an estimate of the growth rate that can be used for process monitoring purposes. The robustification of the controllers is achieved in a simple way, using the observer polynomial. Additional issues, such as the presence of a delay in the system response (as observed in real-life experiments) or non negligible sensor dynamics, are also addressed. I. INTRODUCTION S. cerevisae is one of the most popular host microorganism for vaccine production. The possibility to easily express a va- riety of different recombinant proteins explains its important role in the pharmaceutical industry. In order to maximize productivity, a common strategy is to regulate the ethanol concentration at a low value, thus ensuring an operating point close to the edge between the respirative and respiro- fermentative regimes where the yeast respirative capacity is exactly filled (bottleneck assumption of Sonnleitner’s model [1]). Several applications of this principle can be found, for instance in [2], [3], [4], [5], [6], [7]. However, these control schemes all require the on-line measurement of the ethanol concentration, implying the availability of an (unfortunately quite expensive) ethanol probe. This explains that alternative strategies based on more basic measurement signals, such as the dissolved oxygen concentration, have been proposed, e.g. in [8], [9], or that software sensors reconstructing ethanol from the measurements of basic signals have been designed [10]. In a recent study led by the same authors [7], a RST controller with Youla parametrization is developed for the regulation of the ethanol concentration and tested success- fully in real-life experiments. One of the main advantage of this approach is that it is based on a simple linear model linking the feed flow rate to the ethanol concentration, and a simple linear model of the disturbance, which represents the substrate demand for cell growth. The prior knowledge of only one stoichiometric coefficient is required, whereas the apparent growth rate can be easily estimated on-line (in order to ensure a good disturbance rejection). L. Dewasme, F. Renard and A. Vande Wouwer are with Service d’Automatique, Facult´ e Polytechnique de Mons, 7000 Mons, Belgium, Laurent.Dewasme; Alain.VandeWouwer@fpms.ac.be The objective of the present study is to propose an alternative, simpler, controller design based on the observer polynomial, and to consider both the ethanol regulation and the dissolved oxygen regulation problems. Several additional issues are also addressed, including the presence of a delay in the system response (as observed experimentally), and the influence of the probe dynamics when using relatively fast sampling. Finally, new experimental results are discussed, which illustrate the performance of the proposed scheme in various conditions. This paper is organized a follows. The next section in- troduces the nonlinear model of Sonnleitner [1] and, using the singular perturbation principle, derives two linear models linking either the feed flow rate to the ethanol concentration or the dissolved oxygen concentration. The controller design and the on-line parameter adaptation are presented in Section 3. Section 4 discusses several simulation results demonstra- ting the efficiency of the proposed control strategy, whereas Section 5 illustrates the performance of the ethanol regulation in a series of real-life experiments. The last section is devoted to general conclusions and comments. II. MODELING YEAST FED-BATCH CULTURES A. Nonlinear dynamic model The kinetic model considered in this study is based on Sonnleitner’s bottleneck assumption [1]. During a culture, the yeast cells are likely to change their metabolism because of their limited oxidative capacity. When glucose is in excess (concentration S>S crit ), the yeast cells produce ethanol through fermentation, and the culture is said in respiro- fermentative (RF) regime. On the other hand, when glucose becomes limiting (concentration S<S crit ), the available glucose, and possibly ethanol (as a substitute carbon source), if present in the culture medium, are oxydized. The culture is then said in respirative (R) regime. Component-wise mass balances give the following equa- tions : d(VX ) dt =(k 1 r 1 + k 2 r 2 + k 3 r 3 )XV - DV X (1a) d(VS ) dt = -(r 1 + r 2 )XV + F in S in - DV S (1b) d(VE) dt =(k 4 r 2 - r 3 )XV - DV E (1c) d(VO) dt = -(k 5 r 1 + k 6 r 3 )XV - DV O + V OTR (1d) dV dt = F in (1e) Proceedings of the European Control Conference 2007 Kos, Greece, July 2-5, 2007 ThA13.4 ISBN: 978-960-89028-5-5 4785