Introduction

At the heart of bioprocess control is the ability to monitor important process variables such as biomass concentration [43]. The lack of reliable on-line sensors, which can accurately detect the important state variables, is one of the major challenges of controlling bioprocess accurately, automatically and optimally in biochemical industries [2,7,14,85]. Softsensors (also called software sensors) have been considered as alternative approaches to this problem [10,40,72,73,74,86]. In this chapter, RNNs with both activation feedback and output feedback connections are used for on-line biomass prediction of fed-batch baker's yeast fermentation. The information that is required by the softsensor involves the concentration of DO, feed flow rate and the reaction volume.

Softsensors work in a manner of cause and effect, the inherent biologic relation between measured and unmeasured states could affect the prediction accuracy significantly. DO, pH values, concentrations of carbon dioxide and ethanol are the most commonly selected process variables, which can be readily measured on-line in a research laboratory using standard sensors. Among them, DO concentration, which reflects the fundamental level of energy trans-duction in bioreaction, is intricately linked to cellular metabolism. It changes about 10 times faster than the cell mass and substrate concentrations during the reaction course. Some researchers showed that controlling DO at or above

L. Z. Chen et al.: Modelling and Optimization of Biotechnological Processes, Studies in Computational Intelligence (SCI) 15, 41-56 (2006)

www.springerlink.com © Springer-Verlag Berlin Heidelberg 2006

a critical value could enhance the performance of the bioreactor [87,88]. Nor et al. studied the on-line application of DO concentration [89]. They estimated the specific growth rate of fed-batch culture of Kluyveromyces fragilis based on the measurement of the maximum substrate uptake rate (MSUR) and online DO concentration using mass balance equations. The main assumptions made were that the specific growth rate and the cell yield remained constant during each feeding interval and that the culture was carbon-source limited. The information of on-line DO concentration was also employed to detect the acetate formation in Escherichia coli cultures [90]. Acetate accumulation in fed-batch cultivations is detrimental to the recombinant protein production. On-line detection of acetate enables the development of feedback control strategies for substrate feeding that avoids acetate accumulation, thus increasing the production of recombinant protein. Therefore, there is no doubt that DO dynamics are strongly related to the environmental conditions, and thus an appropriate process variable for on-line inferential estimation of biomass concentration.

The main features of the proposed on-line softsensor are: i) only the DO concentration, feed rate and volume are required to be measured; ii) RNNs are used for predicting the biomass concentration; iii) Neither a priori information, nor a moving window technique is necessary.

The layout of the remainder of the chapter is as follows: in Section 4.2, the recurrent neural softsensor structure is given and the simulation studies are described; in Section 4.3, the experimental investigation is detailed and the results are discussed; conclusions are drawn in Section 4.4.

Was this article helpful?

0 0

Post a comment