Numerical and modelling aspects of large-eddy and hybrid simulations of turbulent flows
Time: Thu 2019-12-19 10.15
Subject area: Engineering Mechanics
Doctoral student: Matteo Montecchia , Mekanik, Linne' FLOW Centre, KTH Mechanics
Opponent: prof. Dominic Von Terzi, TU Delft, Dept. of Aerodynamics, Flight Performance and Propulsion
Supervisor: prof. Arne V. Johansson, Mekanik, Linné Flow Center, FLOW; Dr. Gert Brethouwer, Turbulens, Linné Flow Center, FLOW, SeRC - Swedish e-Science Research Centre; Dr. Stefan Wallin, Turbulens, Linné Flow Center, FLOW, Mekanik
In this study, the explicit algebraic sub-grid scale (SGS) model (EAM) has been extensively validated in wall-resolved large-eddy simulations (LES) of wall-bounded turbulent flows at different Reynolds numbers and a wide range of resolutions. Compared to eddy-viscosity based models, the formulation of the EAM is more consistent with the physics and allows to accurately capture SGS anisotropy,which is relevant especially close to walls.The present work aims to extend the validation of the EAM to larger Reynolds numbers using codes with different orders of numerical accuracy.The first simulations, performed by using a pseudo-spectral code, show that the use of the EAM, compared to the dynamic Smagorinsky model (DSM), leads to significant improvements in the prediction of the first-and second order statistics of turbulent channel flow.These improvements are observed from relatively low to reasonably high Reynolds numbers and with coarse grids.The evaluation of the EAM was continued by implementing and testing of the EAM in the general-purpose finite-volume code OpenFOAM.Several tests of LES of turbulent channel flow have shown thatthe use of the Rhie and Chow (R&C) interpolation in OpenFOAM induces significant numerical dissipation.A new custom-built solver has been utilized in order to minimize the dissipation without generating significant adverse effects. The use of the EAM, together with the new solver, gives a substantially improved prediction of the mean velocity profiles as compared to predictions using the DSM, resulting in roughly 50% reduction in the grid point requirements to achieve a given degree of accuracy. In periodic hill flow, LES with the EAM agreed reasonably well with the reference dataat different bulk Reynolds numbers and reduced the misprediction of the first- and second order statistics observed in LES with DSM.The reduction of the R&C filter dissipation was also shown to be beneficial for the prediction of the mean quantities. An analysis of the skin friction along the lower wall reveals spanwise-elongated, almost axi-symmetric vortical structures generated by the Kelvin-Helmholtz instability. The structures introduced a significant amount of anisotropy.The last part of the study involved the development of a novel hybrid RANS-LES model where explicit algebraic Reynolds stress modelling is applied in both RANS and LES regions.Validations have been conducted on turbulent channel and periodic hill flows at different Reynolds numbers.The explicit algebraic Reynolds stress model for improved-delayed-detached-eddy simulation (EARSM-IDDES) gives reasonable predictions of the mean quantities and Reynolds stresses in both the geometries considered.The use of EARSM-IDDES, compared to the k-omega SST-IDDES model, improves the estimation of the quantities close to the wall.The present work has proven that the use of EAM in wall-resolved LES of wall-bounded flows in simple and complex geometries leads to a substantial reduction of computational requirements both in high-accuracy and general-purpose codes, compared to the use of eddy-viscosity models.In hybrid simulations the EARSM-IDDES shows a clear potential in capturing the physics of wall-bounded flows.