Multivariable adaptive optimization of a continuous bioreactor with a constraint

Yong Keun Chang, Purdue University

Abstract

A single-variable on-line adaptive optimization algorithm using a bilevel forgetting factor was developed. Also a modified version of this algorithm was developed to handle a quality constraint. Both algorithms were tested in simulation studies on a continuous bakers' yeast culture for optimization speed and accuracy, reoptimization capability, and long term operational stability. The above algorithms were extended to a multivariable on-line adaptive optimization and tested in simulated optimization studies with and without a constraint on the residual ethanol concentration. The dilution rate (D) and the temperature (T) were manipulated to maximize the cellular productivity (DX). It took about 80 hours to optimized the culture and the attained steady state was very close to the optimum. When tested with a big step change in the feed substrate concentration it took 60 to 80 hours to drive and maintain the cellular productivity close to the new optimum value. Long term operational stability was also tested. The developed algorithm is capable of stably keeping the culture around the optimum for an extended period, at least for 500 hours after a convergence is attained. For the constrained optimization in which the cellular productivity was maximized while maintaining a sufficient level of ethanol concentration for a good baking quality of the yeast cells, the developed algorithm showed a good optimization speed and was accurate. The multivariable algorithm was experimentally applied to an actual bakers' yeast culture. Only unconstrained optimization was carried out. The optimization required 50 to 90 hours. The attained steady state was D = 0.301 1/hr, T = 32.8$\sp\circ$C, and DX = 1.500 g/l/hr. A fast inferential optimization algorithm based on one of the fast responding off-gas data, the carbon dioxide evolution rate (CER), was proposed. In simulation and experimental studies this new algorithmn is 2 to 3 times faster in optimization speed than the algorithm based on cell mass concentration measurements. A simple on-line estimation scheme was proposed for the ethanol concentration. But, the proposed estimation scheme was inaccurate and significantly underestimated the ethanol concentration. No plausible explanation for such inaccuracy was found. It was shown that the proposed scheme yielded fairly good results when the CER readings were arbitrarily increased by 15%.

Degree

Ph.D.

Advisors

Lim, Purdue University.

Subject Area

Chemical engineering

COinS