Doctoral Dissertations
Date of Award
12-2024
Degree Type
Dissertation
Degree Name
Doctor of Philosophy
Major
Food Science
Major Professor
Jiajia Chen
Committee Members
Doris D'Souza, Scott Lenaghan, Madhu Dhar, Simon Kahan
Abstract
Cultivated meat production, driven by increasing meat consumption worldwide, requires optimized bioreactor designs for high cell densities and yields. Computational Fluid Dynamics (CFD) models are invaluable for analyzing bioreactor performance, such as shear stress, turbulence, and oxygen transfer rates. Traditional and recent modeling approaches were reviewed and compared for seven types of cell culture bioreactors. This comparison allowed the identification of some limitations in the CFD models; namely, the considerable computational resources required and lack of accuracy due to assumptions taken to reduce computational time. This body of work reviews CFD models for various bioreactors, explores machine learning (ML) approaches to enhance CFD modeling, and investigates the impact of dynamic viscosity on mixing performance.
An Artificial Neural Network (ANN)-based ML model was developed to predict and correct coarse-mesh-induced errors in CFD modeling of a spinner flask bioreactor. The ANN model improved the Root Mean Square Error (RMSE) values of nodal velocities by an average of 20% at different rotational speeds. The effect of ANN structure, input data normalization, and training dataset combinations on prediction performance was evaluated. However, the model had limited generalization capabilities especially when correcting lower mixing speeds when a higher speed was used during training.
A three-dimensional Convolutional Neural Network (CNN) ML model was then developed to address ANN model’s generalization limitation by predicting high-resolution fluid profiles from a low-resolution counterpart directly. This model’s structure was optimized and was able to improve the coarse shear stress profile by 67%. The model's generalization capabilities were evaluated across bioreactors with different impeller geometries, showing proper generalization for Kolmogorov length but not shear stress when trained with a single dataset but when including data from all impellers in training, the model was able to improve the coarse shear stress profile by an average of 57% regardless of impeller shape.
Finally, the dynamic viscosity of Human Embryo Kidney (HEK) 293T cell cultures was characterized, revealing shear thinning behaviors and an increase with microcarrier concentrations and cell culture stages. Incorporating dynamic viscosity data into the CFD model showed significant influence on shear stress and Kolmogorov length profiles. The results highlight the importance of monitoring dynamic viscosity and controlling mixing parameters for optimized cell growth, especially during scale-up production operations. These methods reduce computational expenses and increase accuracy of CFD models and lay the groundwork for the development of digital twins of tissue engineering bioreactors and advancing the commercialization of cultivated meat.
Recommended Citation
Cantarero-Rivera, Fernando J., "Computational fluid dynamics modeling of cell culture bioreactors. " PhD diss., University of Tennessee, 2024.
https://trace.tennessee.edu/utk_graddiss/11312
