Hybrid Methodologies For Multiscale Separated Turbulent Flow Simulations PDF Download

Abstract The goal of the present research is to develop and assess multiscale and hybrid turbulence models in simulation of separated turbulent flows at high Reynolds numbers in terms of grid refinement and CPU resources required for a certain level of spectral resolution of the separated flow. These investigated multiscale turbulence models include the DES (Detached Eddy Simulation), hybrid RANS (Reynolds-Averaged Navier Stokes) /LES (Large Eddy Simulation) and PANS (Partially Averaged Navier-Stokes) closure models. These techniques adapt a turbulence model, that function as a RANS model in regions where the grids are highly stretched and the high Reynolds number boundary layer is attached, to function as a sub-grid scale LES type model where the grid is nearly isotropic in the separated flow regions. This accomplished by reducing the turbulence eddy viscosity to promote resolution of more turbulence scales in these regions while still reverting to the original RANS behavior in attached flow and near wall regions. The sensitivity of the computed results to multiscale closure model parameters are compared for three developed formulations of the DES model, one variant of hybrid RANS/LES model, a proposed adaptation of the PANS model and the original multiscale SST-DES model for a number of problems involving unsteady separated high Reynolds number flow. The flow configurations include transonic flow over open cavity, subsonic flow over a back facing step and flow over wall-mounted hump. Simulation predictions are compared with experimental data and also equivalent LES simulations. Simulated results show that these models perform better when there is a distinct demarcation between the attached and separated regions and unsteady shear layers dominate the flow. Computed results show that multiscale methods based on the modification of the turbulent kinetic energy dissipation rate provide the most accurate results and the computed results for the unsteady spectra amplitude and frequency are significantly influenced by the model parameters and the grids. These models provide a useful tool for predicting complex 3-D separated unsteady flows over an expansive dynamic range at high Reynolds number and are comparable to LES predictions at (1/6)th - (1/10)th the corresponding LES CPU resources.