DNS-Based Analysis of RANS Predictions of a Low-Pressure Turbine Cascade

The state of the art design of turbomachinery components is based on RANS solutions. RANS solvers model the effects of turbulence and boundary layer transition and therefore allow for a rapid prediction of the aerodynamic behavior. The only drawback is that modeling errors are introduced to the solution. Researchers and CFD developers are working on reducing these errors by improved model calibrations which are based on experimental data. These experiments do not typically, however, offer detailed insight into three-dimensional flow fields and the evolution of model quantities in an actual machine. This can be achieved through a direct step-by-step comparison of model quantities between RANS and DNS. In the present work, the experimentally obtained model correlations are be recomputed based on DNS of the same turbine profile simulated by RANS. The actual local values are compared to the modeled RANS results, providing information about the source of model deficits. The focus is on the transition process on the blade suction side and on evaluating the development of turbulent flow structures in the blades wake. It is shown that the source of disagreement between RANS and DNS can be traced back to three major deficiencies that should be the focus of further model improvements.