CREATION OF A COMPUTATIONAL FLUID DYNAMICS MODEL IN MSTAR TO VALIDATE PREDICTIONS AGAINST KLA EXPERIMENTS AT VARIOUS IMPELLER SPEEDS AND GAS FLOW RATES

Abstract

In the scale-up of biopharmaceutical processes, a significant challenge arises from the heterogeneities introduced, which can substantially impact cell growth—a critical factor in biologic production. To address this, Computational Fluid Dynamics (CFD) modeling, combined with metabolic modeling, offers a way to capture the dynamic environment experienced by cells during scale-up. This thesis focuses primarily on the development of CFD models, emphasizing the importance of validating these models against experimental data, with a particular focus on the volumetric mass transfer coefficient (kLa).

MStar CFD software, which utilizes the Lattice-Boltzmann Method (LBM) with Large Eddy Simulation (LES), serves as a faster alternative to traditional CFD solutions. Our methodology includes mesh independence studies to ascertain optimal resolution, case study validation, and a thorough evaluation of the factors affecting kLa, all directed towards aligning the model with experimental findings. A notable challenge was the absence of essential experimental data, such as bubble size distribution, energy dissipation rates, and gassed power consumption which is key for accurately predicting kLa, thus complicating model validation. These difficulties are particularly pronounced in bioreactors of industrial scale, where obtaining such data can be challenging. This challenge was partly addressed by using available experimental data and cutoffs from literature for bioreactors.

Analysis revealed that drag and virtual mass forces are the main factors influencing bubble dynamics, with local turbulence-driven breakup models satisfactorily explaining bubble behavior. The drainage theory-based coalescence model overestimated coalescence due to insufficient experimental data for precise constant determination, while the Critical Reynolds model, benefiting from an adjustable critical reynolds number, showed improved accuracy. Models based on Kolmogrov’s isotropic theory of turbulence, proved most effective in predicting the mass transfer coefficient (kL) and Energy Dissipation Rates (EDR) in the range typical to industrial bioreactors. To accurately predict local specific surface area’s, it was necessary to reduce the initial bubble size with increasing flow rate, possibly reflecting an imprecise estimation of the Smagorinsky coefficient. This finding underscores the need for further investigation, including the collection of local energy dissipation data. Additionally, discrepancies in power dissipation were observed particularly at higher flow rates, not as a cause, but because the tank experiences flooding under these conditions. Such discrepancies are significantly exaggerated by MStar CFD’s limitations in accurately modeling the gas cavities beneath the impeller, a critical aspect for precise gassed conditions simulation.

Despite these challenges, the developed CFD model successfully validated kLa across various operational settings without requiring parameter adjustments for each case, demonstrating the model’s ability to capture essential mass transfer phenomena in bioreactors. This success sets the stage for future efforts to integrate CFD models with metabolic models, aiming to improve the understanding and optimization of conditions affecting cell growth in biopharmaceutical production.