A theoretical and experimental optimization of fed-batch fermentation processes

Jayant Moreshwar Modak, Purdue University

Abstract

A systematic approach to cycle-to-cycle optimization of fed-batch cultures for determining the optimal substrate feed strategies is developed. The proposed optimization strategy involves three key steps: developing a reliable mathematical model, computing the optimal control policy, and experimentally verifying the effectiveness of the optimal policy. Saccharomyces cerevisiae, more commonly known as baker's yeast, is chosen as a model system. An unstructured model for aerobic growth of baker's yeast is developed. An attractive feature of the model is that it accounts for three major metabolic pathways of yeast growth, yet it is simple enough to be used in optimization studies. The optimal glucose feed policy achieves a balance between the growth rate of yeast cells and the cellular yield on glucose. The optimal feed rate sequences are identified for four different initial culture conditions. The optimal glucose feed policy is implemented in a computer controlled fermentor system. The parameters in the model are determined from the experimental data of one cycle. Then, the optimal glucose feed rate is computed and applied in the next cycle. The optimal results, obtained in two cycles of operation, show a 10 to 15% improvement over the suboptimal results. The effectiveness of the optimization scheme is demonstrated under conditions of poor parameter estimates and incorrect model assumptions. A new nonsingular control approach, in which the fermentor volume is treated as a control variable, is developed for optimization of fed-batch cultures. The proposed approach is not limited by the dimensionality of the model, and therefore, it can be used in optimization studies of complex fermentation processes. An optimization strategy is developed for problems involving two control variables. The strategy is demonstrated for determining optimal operating conditions for a bioreactor by controlling (i) substrate feed rate and temperature, (ii) feed rates of two substrates, and (iii) inlet and outlet feed rates. The uncoupled nature of the controlled variables allows a direct extension of the computational algorithms of single control variable problems to a two control variable problem.

Degree

Ph.D.

Advisors

Lim, Purdue University.

Subject Area

Chemical engineering

COinS