One of the major challenges mankind faces nowadays is combating climate change. A substantial fraction of greenhouse gases are released by industrial processes, as steelmaking, (oil)refinery and waste processing. Emissions from these processes can partly be prevented with a recently developed technology called gas fermentation. Within this process, synthesis gas – amixture containing CO, CO2 and H2 – is converted into ethanol and acetic acid by bacteria such as Clostridium autoethanogenum. These products could be used in a wide range of applications, like fuels, plastics and cosmetics. Whilst gas fermentation is already applied at commercial-scale, challenges in scale-up persists due to complex multi-scale interactions among the bioreactor, gas bubbles, and bacteria. The poor solubility of CO and H2 alongside gas bubble coalescence, leads to low gas-to-liquid mass transfer rates (typically denoted via kLa). Slow mixing in industrial bioreactors (500m3), and high gas conversion rates, result in large spatial variations in dissolved gas concentrations. Bacteria experiencing concentration fluctuations have often been related to decreased process performance...@en

Presented works describe a novel approach to assess mixing in a stirred vessel using light-based tomography. The study is driven by key research questions: What are the obstacles in three-dimensional dynamic tracer distribution reconstruction? How should the experimental equipment be constructed to obtain the best possible data from three cameras recording the back-lit stirred tank? How can raw images be processed to isolate the ray-dye interaction? And, how can relevant mixing information be obtained from projections and reconstructed volumetric data? The research begins with a short introduction of traditional mixing measurement techniques, establishing the context and relevance of the work. The theoretical background of the study is then presented, including the principles of tomographic reconstruction and the main algorithms used in the process. The methodology involves the use of synthetic data obtained from LES for the framework and baseline creation followed by the acquisition of experimental data. A significant part of the methodology is dedicated to image pre-processing, which incorporates as main steps the inverted grayscale transformation, brightness normalization, background removal using image similarity metrics and object removal with the use of a neural network. The use of the simplistic forward model based on the Lambert-Beer law is described, followed by the implementation of the projection matrix-free Simultaneous Algebraic Reconstruction Technique. The outcomes of both synthetic and experimental data reconstruction are presented and despite the shortcomings of the used experimental setup the 2D and 3D mixing maps were created, supported by the local Coefficient of Variance calculation to gain further insight into the process. The conclusions highlight the potential of light-based tomography for evaluating mixing while acknowledging the need for significant refinement and validation of the methodology. Recommendations include the improvement of imaging, equipment modifications, and reconstruction implementation. ii

In recent years, biotechnological processes have gained increased interest due to their potential for high-value compound production and waste recycling. This shift towards biotechnology is driven by global challenges such as food security, climate change, and the transition to renewable resources. To address the limitations of large-scale fermentations, scale-down approaches have been recommended to minimize microbial performance losses during scale-up procedures. Computational fluid dynamics (CFD) coupled with omics-based technologies offer valuable insights into the environmental and intracellular
lifelines of cells. However, current laboratory-scale setups have certain limitations, emphasizing the need for dynamic microfluidic single-cell cultivation (dMSCC) devices. These devices enable the analysis of single-cell behavior in dynamic environments with high temporal resolution.

This thesis focuses on improving the amplitude control while maintaining temporal resolution in dMSCC devices. A new dMSCC device design was analyzed using a 2D model, which was experimentally validated. The results demonstrated that the design mechanism effectively generated concentration profiles resembling discrete and smooth lifelines, albeit with a relatively high response time (30 seconds). A mesh independence study indicated minimal deviations (2 %) in results for different mesh refinements, while complex geometric structures introduced greater variations.

The experimental validation of the 2D COMSOL Multiphysics model highlighted discrepancies between the experimental data and model predictions, both at the outlets of the microfluidic concentration gradient generator (μCGG) and inside the chamber (RMSE=0.1-0.75; >10% of experimental data). However,
the observed trends of the concentration profiles inside the chamber were well-captured. Optimization studies were conducted based on these findings, leading to valuable conclusions. Narrowing the chamber width increased the chip’s response time. Moreover, increasing the space between μCGG outlets
as well as increasing fluid velocity inside the μCGG (while keeping the maximum velocity constant) improved gradient width. The latter approach is preferred to maintain temporal resolution. A comparison between COMSOL Multiphysics (RMSE=0.14) and Ansys Fluent (RMSE=0.15) models revealed that Ansys Fluent better captures experimental trends but has lower prediction accuracy. Further investigations involved a Design of Experiments (DoE), which indicated that the current μCGG design is suitable for fluid velocities preferably lower than 1 · 10−5 m/s and tracers with high diffusion coefficients. These conclusions provide insights into optimizing dMSCC devices and contribute to the broader understanding of mimicking microbial lifelines.

Fermentation processes are considered to be essential to decrease our reliance on fossil fuel based products. However, the scale-up from lab-scale to industrial-scale has proven to be difficult. Computational Fluid Dynamics (CFD) has the potential to be a tool to optimize the scale-up and to help engineers understand the relevant hydrodynamics inside such reactors. However, traditional CFD simulations are computationally intensive and it is not uncommon that simulations can take several months to compute a few minutes of flow-time. Due to the recent development of GPU-based hardware, the Lattice-Boltzmann method (LBM) has been gaining much interest as this meant an orders-of-magnitude decrease of needed computational time compared to conventional CFD methods. In order to calibrate the models underlying the simulations, the obtained results of said simulations should be validated against experimentally obtained data, which is exceptionally scarce for industrial-scaled reactors. Hence the aim of this thesis is to to investigate the applicability of the LBM in high gas-flow, industrial sized reactors by studying three benchmark cases. Using Large Eddy Simulations (LES) and Euler-Lagrangian tracking of bubble parcels, the models provided by M-Star (MStar Simulations, LLC) are validated by replicating a small-scale reactor of which the LBM has already been successfully applied to. The industrial-scaled case consists of the simulation of the 22m3 industrial reactor located in Stavanger, which is an exception regarding the scarcity of experimental data for industrial sized reactors. As this data set did not encompass all the relevant data for aerated stirred tanks, the suitability of the LBM and provided models are also tested for a smaller scaled tank of which the Bubble Size Distribution (BSD) throughout the vessel is known. The applicability of the provided models will be tested by comparing the obtained gas hold-up, BSD and power consumption with experimental data sets. Although the LBM is suitable for industrial-scaled reactors, the provided models by M-Star are not sufficient to predict the gas hold-up and power consumption at high superficial gas velocities. The provided Free Particle drag correlation is not sufficient to describe the dispersion of bubbles throughout the vessel, leading to a significantly under-estimated gas hold-up. In addition, the parcel approach leads to the formation of large bubbles in the vessel. And although enforcing a maximum coalescence diameter does improve the BSD, the gas hold-up is not significantly influenced by the presence of large bubbles. Furthermore, the Euler-Lagrangian way of modeling bubble particles did not lead to the formation of gas cavities, which resulted in an insignificant drop in power consumption. Nevertheless, the current work provides a foundation for subsequent research.

The traditional Eulerian view of biomass in bioprocess modelling results in issues when modelling large-scale bioreactors in which heterogeneous conditions are common. As the field increasingly moves from “scale-up” to “scale-down” philosophy, in which such heterogeneities are included from the start of process design, accurate modelling of the biomass response to these varying conditions is essential. Euler-Lagrange (EL) simulations provide a means of modelling the microbial lifelines of cells traversing heterogenous conditions of a bioreactor. Lapin et al. were the pioneers of EL simulation in their 2004 paper where a metabolic model of glycolysis is coupled to a Lagrangian biomass phase. This BSc thesis focusses on reproducing their model using modern computational fluid dynamics (CFD) techniques. Specifically, by using a dynamic Lattice Boltzmann Method using Large Eddy Simulation model for CFD as opposed to a frozen-flow Finite Volume Reynolds Averaged Navier- Stokes model. The resulting differences in the overall behaviour of the cell metabolism through the lens of glycolytic oscillations are discussed. In addition, possible pitfalls in model validity such as grid dependence, the effects of heterogeneous particle distributions and the effects of particle numbers were explored. The synchronisation and desynchronisation of glycolytic oscillations as observed in Lapin et al. 2004 were able to be reproduced.