Investigation of anisotropy-resolving turbulence models by reference to highly-resolved LES data for separated flow

The predictive properties of several non-linear eddy-viscosity models are investigated by reference to highly-resolved LES data obtained by the authors for an internal flow featuring massive separation from a curved surface. The test geometry is a periodic segment of a channel constricted by two-dimensional (2D) 'hills' on the lower wall. The mean-flow Reynolds number is 21560. Periodic boundary conditions are applied in the streamwise and spanwise directions. This makes the statistical properties of the simulated flow genuinely 2D and independent from boundary conditions, except at the walls. The simulation was performed on a high-quality, 5M-node grid. The focus of the study is on the exploitation of the LES data for the mean-flow, Reynolds stresses and macro-length-scale. Model solutions are first compared with the LES data, and selected models are then subjected to a-priori studies designed to elucidate the role of specific model fragments in the non-linear stress-strain/vorticity relation and their contribution to observed defects in the mean-flow and turbulence fields. The role of the equation governing the length-scale, via different surrogate variables, is also investigated. It is shown that, while most non-linear models overestimate the separation region, due mainly to model defects that result in insufficient shear stress in the separated shear layer, model forms can be derived which provide a satisfactory representation of the flow. One such model is identified. This combines a particular quadratic constitutive relation with a wall-anisotropy term, a high-normal-strain correction and a new form of the equation for the specific dissipation omega = epsilon/k.