Real time estimations of kLa and viscosity in a fermentation pilot plant

Sammanfattning: Fermentation is a process that utilizes live microorganisms to produce different biological products such as enzymes. The oxygen mass transfer rate is often the limiting factor in aerobic fermentations and thereby an important process parameter. The initial aim of this MSc thesis was therefore to develop a soft sensor for the volumetric oxygen mass transfer coefficient, kLa, in order to monitor it online in real time. However, the oscillation in the kLa estimation had to be reduced to get smoother plots by first filtering the process data. The soft sensor was finished at an early stage and the scope of the thesis project was therefore expanded. Previous studies have shown that kLa is dependent on the viscosity of the fermentation broth and that higher viscosities leads to lower values of kLa. It was therefore of interest to monitor the viscosity online as well, and to explore if it would be possible to use the viscosity for implementations of new control concepts. The thesis further explored the relationship between the kLa and the viscosity, and investigated if it would be possible to describe the viscosity as a function of the kLa using rearranged empirical correlations. The viscosity estimation was validated versus offline measurements and new parameter sets for the correlation were also created and evaluated. The investigation resulted in another soft sensor which estimated the viscosity online with a standard deviation of 0.09 Pa s. It should be noted that the dataset generated in this thesis was not ideal for parameter estimation due to low variance in the process variables. It should therefore be possible to improve the estimation further, for example using previous datasets with larger variance in the process variables. The substrate is fed in pulses instead of a continuous stream since previous studies have shown that this reduces the viscosity. The third objective of the thesis was therefore to explore how the viscosity was influenced by the changes is paus time between the pulses, i.e., the cycle time, and whether it would be possible to base new control concepts on this relationship. The final objective was to see if it was possible to correlate biomass concentration with rheology parameters including viscosity to explore the possibility of estimating the biomass from online process data. Both of these objectives were investigated but did not yield promising results since no clear correlations could be observed in either case. The relationship between viscosity and cycle time therefore needs to be investigated further before implementation of new control concepts. Further investigation of the relationship between biomass and rheology parameters is also needed to get a better insight in the processes. This could result in relevant insight in how the rheology of a specific host organism and process is correlated to the biomass. It does however seem unlikely that a general correlation could be found for any host organism and process.