On the modelling of turbulent flow and mixing in stirred reactors

Abstract: This thesis concerns the numerical simulations and modelling of the flow structure and scalar transport in stirred reactors. For this Large Eddy Simulations (LES) have been used. The flow situation in a stirred reactor includes several flow phenomena that are not well predicted by "standard" two-equation models, e.g. swirl, stagnation points and anisotropic turbulence. Also, using LES one can obtain data unavailable when using a Reynolds Averaged Navier-Stokes (RANS) approach, such as the spectral content of the solution. However, LES requires a high degree of resolution in time and space which, combined with long sampling times to obtain converged statistics, put higher demands on the computer hardware. One major difficulty when simulating flow in stirred reactors is how to incorporate the effects of the moving boundaries, i.e. the turbine blades. Several approaches have been proposed in the past spanning from deforming and/or sliding grids to stationary boundary conditions based on LDA measurements and the use of moving momentum source terms in the equations of motion. The numerical aspects of representing complex moving boundaries have been investigated and a model based on local velocity differences is presented. The results show that complex moving boundaries can be represented on a Cartesian grid in this way with reasonable numerical accuracy and efficiency. The influence of several Sub-Grid Scale (SGS) models, both for turbulence and mixing, on the flow and the transport of inert additives has been studied. However, for this the geometrically simpler case of a circular jet impinging on a flat plate is considered. The main reasons being that simulating the complex flow structure of stirred reactors is quite time consuming, in terms of CPU time. Also, the additional geometrical complexity may mask some physical factors/problems. The study of SGS model effects suggests that the choice of SGS model has a significant effect on the results and that a eddy viscosity based dynamic model could be the best choice.