Evaluation of turbulence models on roughened turbine blades

The accuracy of turbulence models for the Reynolds-Averaged Navier-Stokes (RANS) equations in rough-wall flows is evaluated by comparing the model predictions with the data obtained from large-eddy simulations (LES). We have considered boundary layers in favourable and adverse pressure gradients mimicking those encountered in hydroturbines. We find that some features of the flow cannot be captured accurately by any model, due to the fundamental modelling assumptions. An example is the flow reversal that occurs in the roughness sublayer prior to separation, which cannot be predicted by the commonly used approaches, which bypass the roughness sublayer while modifying the boundary conditions. In mild pressure gradients most models are sufficiently accurate for engineering applications, but if strong favourable or adverse pressure gradients are applied (especially those leading to separation) the model performance rapidly degrades. A particularly difficult problem (both for rough- and smooth-wall cases) is the return to equilibrium after a strong perturbation, a known limitation of RANS models. Simulations of real configurations using commercial codes are also considered.