Hybrid Modelling for Enhanced Bioreactor Performance

The topic of this thesis is the modeling of industrial process data from bioreactors for analysis, monitoring, control and optimization of cultivations for production of insulin precursors using genetically modified strains of Saccharomyces cerevisiae. The aim of this work is to develop models in order to facilitate and support the analysis of production data from industrial cultivation processes, which can be applied in both offline and online process analysis. Soft sensors are first developed for obtaining quantitative information from an existing industrial cultivation process. Two similar models are developed for monitoring of the biomass and product concentrations using first principle engineering modeling (FPEM). Application of the two simple soft sensors on industrial data provide reasonable descriptions of the general biomass and product concentration trajectories. Implementation and use of the soft sensors will enable a very simple yet highly attractive way of providing online information of the two key process variables. Process Chemometrics provide an alternative approach for monitoring the product concentration. A multiway projection to latent structures (MPLS) regression model is presented, providing both one-step ahead and end point predictions of the product concentration within 5-10 % of analytical offline measurements. Comparison of the two soft sensors for describing the product concentration indicates that the MPLS-predictor for the one-step ahead prediction gives a slightly better description of the variations in the product concentration. Attempts to use the two FPEM models mentioned above on the cultivation of a similar recombinant strain of S. cerevisiae failed, since unanticipated production of acetate indicates a different metabolic response to the growth conditions. Literature reviews on the transport and effect of organic acids in cultivation processes as well as the genetic engineering performed on the strain are presented. For modeling using mass balances the elemental composition of the biomass is required and since it has not been reported if the elemental composition of the recombinant strain has been influenced by the genetic engineering, this is investigated using macroscopic mass balances. The elemental composition of the biomass is found to be similar to what is reported in the literature. An indicator of the onset of oxidoreductive growth is presented. The indicator is based on comparing the measured ammonia flow rate to a modelbased estimate of the ammonia flow rate during oxidative growth. The model of the ammonia flow rate takes into account the effects from changes in the volume of the culture broth, effluent flow rate and glucose syrup feed rate. The indicator not only facilitates the monitoring of the cultivation process for onset of oxidoreductive growth. Together with the modelbased residual between measured and estimated ammonia flow rate, the indicator provides the foundation for extended modeling of the cultivation process. Using the information provided by the indicator of the onset of oxidoreductive growth, a soft sensor is constructed for the estimation of the conversion rate of acetate as well as the production rate of biomass. The soft sensor is based on a combination of a proton balance and a carbon balance, and using these rates in dynamic mass balances online estimations of acetate and biomass concentrations are provided as well. Applying the model on data from a number of cultivations, provides a surprising observation namely that acetate is being produced in large amounts 1-2 hours before formation of ethanol occurs. The reason for the onset of the acetate formation has not yet been determined. A small metabolic flux model is proposed using calculated and estimated conversion rates of substrate, biomass and key metabolites combined with physiological parameters reported in the open literature on another strain of S. cerevisiae. The model is used to illustrate and discuss observations from cultivations showing both normal and abnormal process behavior. The model illustrates how acetate is produced prior to ethanol formation. The model also shows how the activity of the oxidative phosphorylation changes extensively as ethanol formation starts and as ethanol consumption ends, which is interpreted as effects from repression/derepression of the oxidative phosphorylation. It has not been possible to explain what mechanism is responsible for this control of the oxidative phosphorylation, although it is discussed that it could not be a fixed limitation in the capacity of the oxidative phosphorylation, since an experiment using closed loop control of the ethanol concentration in the offgas shows even higher activity of the oxidative phosphorylation than are seen in similar open loop experiments. Finally a simple model is proposed to describe the specific productivity of the product. The description is based on a first order model expression for the dependence of production rate and biomass synthesis rate, with a time constant proportional to the specific glucose uptake rate provided by the metabolic flux model presented above. The model gives a reasonable description of the observed trajectories of product concentration in a normal cultivation and by a small extension of the model, it is also able to provide a reasonable estimation of the product concentration profile during process upset in the form of acetate formation.