Simulating large-scale biotech bubble columns: challenges and insights from the parcel approach.

Abstract

The growing industrial demand for economically attractive large-scale bioprocesses has increased the need for reliable and computationally feasible modelling approaches. Gas fermentation plays a particular role as the targeted commodities ask for large-scale solutions using bubble-column type bioreactors that allow them to benefit from the economy-of-scale. To reduce development efforts, large-scale designs increasingly build on computational fluid dynamics (CFD) simulations. State-of-the-art approaches employ finite-volume Reynolds-averaged Navier-Stokes (FV-RANS) methods that require enormous computational efforts. In contrast, Euler-Lagrange (E-L) approaches offer detailed local resolution as a prerequisite for optimum bioreactor design. By tracking individual bubbles, they predict bubble population dynamics while computations are parallelized. However, to cope with realistic bubble numbers, large-scale simulations typically make use of the so-called 'parcel size' approach, i.e. multiple bubbles are represented by a single computational particle. This study applies Lattice Boltzmann large eddy simulations (LB-LES) to illustrate the observed consequences: Swarm-like bubble behavior is found that results in inhomogeneous bubble distributions. As a consequence, the volumetric mass transfer coefficient k_La, a key criterion for design, may be biased. Comparing results with state-of-the-art FV-RANS simulations revealed similarities of transient flow structures and averaged velocity values. However, local deviations are found that mirror handling differences of turbulence representation and the impact of the parcel approach. The findings demonstrate the potential and current limitations of parcel-based modelling for large-scale bioreactors.