Modelling gas-liquid flow and local mass transfer in stirred tanks

This doctorial thesis offers a guideline for modelling gas-liquid flow in stirred tanks with computational fluid dynamics (CFD). Particularly the effect of varying physical properties and industrial operating conditions is highlighted. The most important thing in modelling mass transfer in stirred vessels is the accurate prediction of local bubble size. Population balances for bubbles are needed for accurate description of the local mass transfer rate. There are many pitfalls in gas-liquid modelling at the transitional turbulence regime, and they need to be recognised and dealt with at a reasonable computational cost. Details of the work are presented in the included publications, this thesis sums up the findings. Backbone of this thesis is the experimental work done on 14 and 200 dm³ vessels. Experimental techniques were compared in making bubble size distribution (BSD) measurements. A variety of experiments were made to investigate: physical properties, vapour-liquid equilibrium, gas hold-up, gas-liquid mass transfer, bubble size distributions, local mixing times, flow fields and bubble swarm interactions. Parameters for a number of phenomenological models were fitted with a computationally less demanding multiblock model and were then used to simulate stirred reactors with CFD. The early systems were lean dispersions of low viscosity; at the end of this work opaque shear thinning G-L dispersions were modelled. The effect of impeller geometry on G-L mass transfer was studied by simulating three impeller geometries. There were no differences in the volumetric mass transfer rate between the impellers, although the flow patters and gas hold-up showed clear differences between the impellers. Heterogeneous behaviour like gas slug creation and reactor dead-spaces were successfully modelled. The simulated dispersions were highly heterogeneous: 50% of mass transfer took place in less than 10% of the reactor volume. A xanthan fermentation batch lasting for days was modelled; the reaction speed was bottlenecked by both mixing and mass transfer. These findings strongly support the use of spatially detailed models over ideal mixing assumption