Large Eddy Simulation of Turbulent Flows in Combustor Related Geometries

Abstract: Numerical simulations using Large Eddy Simulation (LES) are applied to turbulent swirling flow fields. The swirling motion is often introduced into combustors to act as flame holders or enhance the mixing between species. Different turbulence models capture the swirling motion more or less accurately. LES is well suited for understanding details of swirling flows. It resolves all the large scales in the flow field and only the small scales have to be modeled. The small unresolved scales are the Subgrid Scales (SGS) and the model must take into account the interaction between the small scales and their influence on the resolved scales. In order to separate the effects from the SGS models and the numerical scheme, the problem must be well resolved and be of high order. SGS models have been applied, investigated and compared in swirling flow fields. Four SGS models are considered: an implicit, a stress similarity, a dynamic divergence and an exact differential model. The implicit model uses no SGS model. For the stress similarity model, similar behaviour between the resolved and unresolved stresses is assumed. The model parameter in the dynamic divergence model are depending upon both space and time and it is recalculated during the whole simulation. If a particular form of differential filter function is applied, an explicit expression of the SGS stress tensor can be received. This is the exact differential model. In the simulations, the stress similarity model is shown to have the largest effect on the results. Otherwise, the SGS models only show minor effects on both mean velocities and turbulence intensities. A high order Cartesian grid method have been proposed and employed in the simulations. Cartesian grids have features that are very suitable for LES. The grid generation is simple and fast, it does not require a lot of computational storage and the discretized governing equations can be easily extended to higher orders. The drawback of Cartesian grids is that it does not represent complex geometries correctly. The boundary conditions can be misplaced by as much as a cell size and this reduces the order of the solution. A high order wall treatment is proposed to handle the low order wall problem and it is incorporated into the Cartesian grid method. The high order Cartesian grid method is shown to maintain the order of the discretization.