Assessment and Some Improvements of Hybrid RANS-LES Methods

Sammanfattning: Motivated by numerical analysis of a shock/boundary-layer interacting (SBLI) flow in a duct, where some typical existing hybrid RANS-LES methods become awkward to give accurate predictions of the shock-induced corner separation bubble, a new hybrid RANS-LES method, has been proposed. The model is based on a low-Reynolds-number (LRN) $k-omega$ RANS model, which is one of the few models that are capable of giving reasonable predictions for the SBLI flow in RANS computations. The proposed hybrid RANS-LES model is not only intended for this specific flow, but also for turbulent flows typically present in aeronautical applications, e.g flow in an air intake or SBLI flow. The model has thus undergone extensive calibration and validation in computations of fundamental flows, namely, Decaying Homogeneous Isotropic Turbulence (DHIT) and channel flows. Ultimately, the proposed model will be further verified in computations of complex aerodynamic flows, such as SBLI and bluff-body flows. In order to use the same base model in both RANS and LES modes, the LRN effect nestled in the near-wall RANS mode is diminished in the off-wall LES region by a correction function in the proposed model. With this correction function, a weak sensitivity to different grid resolutions is observed for the proposed model in DHIT. The constant $C_{LES}$ was calibrated to 0.70, which is comparable to other hybrid RANS-LES formulations. In the computation of turbulent channel flow at $Re_{ au} = 950$, the RANS-LES interface condition has been evaluated in comparison with available DNS data and different SGS turbulent length scales have been tested. It is suggested that a wall distance based length scale reasonably well supports the turbulence-resolving capability and modeling of near-wall turbulence properties, in comparison with LES using the dynamic Smagorinsky model and the WALE model. Moreover, the model has been validated further as a zonal approach (similar to embedded LES) in computations of turbulent channel flow at $Re_{ au} = 8000$. Different length scales for the LES mode, as well as different RANS-LES switch locations, have been evaluated. It is shown that the wall distance based LES length scale is able to give superior results with a diminishing log-layer mismatch. In general, the model has produced very encouraging predictions of the mean flow and resolved turbulent statistics.