Session 9 - Theme: Bioreactors and Bioprocesses (1)

Estimation of power number and mixing characteristics lab-scale stirred reactors without wall-mounted baffles remains a major challenge in computational fluid dynamics (CFD) because of the complex intimate interaction between impeller-induced flow structures, free-surface deformation, a more or less central precessing vortex, and turbulence. In this study, CFD simulations were performed to evaluate the impact of vortex formation, impeller diameter, and Reynolds-Averaged Navier–Stokes (RANS) turbulence models on power number and mixing characteristics of a 100 mL stirred reactor operating in the turbulent regime at 400 rpm. A four-blade pitched blade turbine (4PBT45) with varying diameters (25 mm, 30 mm, 34 mm, and 38 mm) was analysed as the impeller of the reactor. Comparisons were conducted between two different modelling approaches, SST k–ω and k–ε, to assess their predictive capability for vortex dynamics and fluid flow characteristics. All the simulations are performed using sliding mesh (SM) and the volume of fluid (VOF) method to simulate liquid-air surface. The results demonstrate that vortex formation notably changes flow topology, especially near the free surface and in the impeller discharge region. Larger impeller diameters increased power consumption, but power number decreased with the increase in D/T. RANS models showed very similar power number, vortex depth and flow velocities.

In vitro cultivation of human inducible pluripotent stem cells (hiPSCs) for the production of Red Blood cells (RBCs) is a promising therapeutic alternative to donor-based blood cell transfusions. The sensitivity of human stem cell aggregates requires a tight balance between hydrodynamic stresses and suspension of the 3D aggregates, while maintaining sufficient oxygen transfer. We performed CFD-DEM simulations to characterize a 250 mL MiniBio® bioreactor for shear sensitive microcarrier cultures. Our model uses the Lattice Boltzmann method with a large eddy simulation (LES) turbulence model to solve the fluid flow, combined with a discrete element method (DEM) for microcarrier particle motion. We extensively validated our simulations using mixing time, kLa, and just-suspension (Njs) experiments, demonstrating that our model can predict these engineering parameters with a 20% accuracy. Additionally, 4D-particle tracking velocimetry measurements showed a good agreement between measured and predicted particle velocity and acceleration profiles. Inclusion of the Hertzian particle-particle interactions was essential to correctly capture particle suspension behaviour. The validated model was used to study the influence of important operating conditions, including stirrer speed, pumping-mode, stirrer height and filling volume. Furthermore, the Lagrangian nature of the simulated microcarriers was used to characterize the spatial-temporal fluctuations in fluid strain and energy dissipation rate from the microcarrier perspective to quantify the hydrodynamic stresses on microcarriers for different operating conditions. Future work will focus on applying this model to a 3L bioreactor and will help accelerating the scale-up of shear-sensitive stem-cell microcarrier cultures.

Single-use stirred tanks are increasingly adopted in biopharmaceutical manufacturing, however, despite their industrial relevance their flow behaviour has received limited detailed investigation. The gas-liquid hydrodynamics of a laboratory-scale single-use stirred tank with square cross-section and a bottom-mounted radial impeller is investigated using a Detached-Eddy Simulation coupled with a Volume-of-Fluid formulation (DES–VOF). Numerical predictions are quantitatively validated against stereoscopic PIV measurements, showing good agreement for both mean and fluctuating velocities. A harmonic-regression-based decomposition is introduced to separate the velocity field into mean, impeller-locked periodic, and stochastic turbulent components without requiring phase-synchronised data. The method provides a geometry-independent alternative to classical phase-averaging and is benchmarked against Proper Orthogonal Decomposition. The analysis confirms that impeller-locked fluctuations are confined to the immediate vicinity of the blades and account for only a minor fraction (≈3%) of the total kinetic energy, whereas stochastic turbulence dominates the region beneath the free surface. This separation clarifies the relative roles of coherent forcing and turbulence in gas–liquid mixing. Free-surface fluctuations are strongly correlated with residual turbulent intensity, while no measurable contribution from the periodic component is observed, indicating that surface dynamics is driven by turbulence, rather than by deterministic impeller forcing. The proposed framework enables a physically consistent partition of flow contributions in multiphase single-use systems and is directly applicable to both numerical simulations and experimental velocity datasets, supporting improved analysis and scale-down strategies for bioprocessing applications.

“Smarter mixing” is increasingly about decision efficiency as much as hydrodynamic efficiency: reducing energy intensity, shortening development cycles, and avoiding excessive experimentation while preserving product quality. In bioprocessing, these goals converge in the tension between gas–liquid mass transfer and shear sensitivity, where the bottleneck is often design exploration across feasible hardware and operating combinations. We present an AI-augmented modeling workflow that couples a first-principles mixing tool with small-data optimization for constrained design-space exploration. KaeMix provides physics-based predictions of power and gas–liquid mass-transfer performance for impeller and gassing configurations. OASIS.AI learns from a limited number of evaluations and proposes the next designs to test while enforcing engineering constraints. Here, “AI” means embedded, constraint-aware search guidance; it is not a language model and does not replace mechanistic simulation. The method is demonstrated on a 2000-L mammalian cell-culture bioreactor. The workflow searches over impeller configuration and operating conditions to satisfy a mass-transfer requirement (kLa ≥ 0.04 s⁻¹) and power limits while minimizing EDCF as a proxy for turbulent shear exposure. The AI-guided exploration identifies feasible, non-obvious solutions that meet the kLa target at reduced power and lower predicted shear—for example, combining higher gas throughput with fewer, larger impellers at low RPM—thereby improving sustainability and biological risk posture. The sustainability benefit is twofold: reduced operating power in the selected envelope and reduced development waste by reducing the iterations needed to converge on viable, constrained solutions. The approach generalizes to other mixing applications with large constrained design spaces.