Large eddy simulation of channel flow with and without periodic constrictions using the explicit algebraic subgrid-scale model

Abstract

We analyse the performance of the explicit algebraic subgrid-scale (SGS) stress model (EASSM) in large eddy simulation (LES) of plane channel flow and the flow in a channel with streamwise periodic hill-shaped constrictions (periodic hill flow) which induce separation. The LESs are performed with the Code_Saturne which is an unstructured collocated finite volume solver with a second-order spatial discretisation suitable for LES of incompressible flow in complex geometries. At first, performance of the EASSM in LES of plane channel flow at two different resolutions using the Code_Saturne and a pseudo-spectral method is analysed. It is observed that the EASSM predictions of the mean velocity and Reynolds stresses are more accurate than the conventional dynamic Smagorinsky model (DSM). The results with the pseudo-spectral method were, in general, more accurate. In the second step, LES with the EASSM of flow separation in the periodic hill flow is compared to LES with the DSM, no SGS model and a highly resolved LES data using the DSM. Results show that the mean velocity profiles, the friction and pressure coefficients, the length and shape of the recirculation bubble, as well as the Reynolds stresses are considerably better predicted by the EASSM than the DSM and the no SGS model simulations. It was also observed that in some parts of the domain, the resolved strain-rate and SGS shear stress have the same sign. The DSM cannot produce a correct SGS stress in this case, in contrast to the EASSM.

Keywords:

• large eddy simulation
• explicit algebraic subgrid-scale model
• periodic hill flow
• turbulence

Acknowledgements

Dr Michael Breuer is acknowledged for providing the reference LES data for the periodic hill flow. We would also like to thank Dr Jing Gong from the PDC for helping with the installation of the Code_Saturne, and Dr Stefano Rolfo and Dr Alistair Revell from the University of Manchester for technical helps about the code.

Additional information

Funding

Support from the Swedish Research Council [grant number 621-2010-6965] and computer time provided by the Swedish National Infrastructure for Computing (SNIC) are gratefully acknowledged.