| Optimization of fed-batch fermentations is usually addressed by applying optimal control theory to determine the nutrient feed rate that minimizes a cost function. The optimal feed rate is calculated for the idealized situation where the process model is accurate, the initial conditions are known, and no disturbances are foreseen. However, when applied to the real process in an open-loop manner, this feed rate does not yield the optimal performance because these ideal conditions are not satisfied. It is therefore desired to develop feedback policies which reduce the effect of model-plant mismatch on process performance.;In this work, we examined three feedback feed policies. The first policy assesses the fermenter state periodically and refines the optimal feed rate accordingly. This approach can suffer from a high computational load, and it relies on its estimation scheme. The second method controls the feed rate so that the profile of a measured or estimated process variable tracks a pre-calculated trajectory. In contrast with the NMPC method, the need for repeated optimizations is eliminated. A drawback of this method is that a variable whose optimal profile is not sensitive to model error is difficult, if not impossible, to determine.;The third feedback method approximates the optimal feed policy by a two-phase feeding strategy. During the first phase, one manipulates the feed rate so as to maximize the specific rate of cell growth. During the second phase, the feed rate is manipulated so as to obtain a high cellular yield or high metabolite formation rate.;A MinMax approach that takes account of model uncertainty, during feed policy design, is described and used in two case studies. The problem of designing a feedback policy is thus transformed into one of defining a set of models which can represent the process, and then determining the element of this set to use in optimal feed design.;Finally, the question of run-to-run improvements in the outcome of a fermentation is discussed and implemented on a baker's yeast process. An extensive review of baker's yeast growth models is presented. |
