Investigation of turbulence model and numerical scheme combinations for practical finite-volume large eddy simulations

Abstract

Large eddy simulation (LES) is known to suffer from two primary error sources – the subgrid stress (SGS) model and the numerical discretization scheme. These cannot be separately quantified for finite-volume numerical simulations, but an appropriate combination can yield ‘engineering accurate’ prediction of turbulence dynamics. This paper seeks to evaluate combinations of commonly used second-order numerical schemes and Smagorinsky-type SGS models for practical LES. Error assessments are performed for isotropic decaying turbulence using both pseudospectral and finite-volume solvers, followed by validation for a complex turbulent flow of engineering interest. Error assessment using pseudospectral techniques is performed to isolate finite-differencing and modeling errors by explicitly adding numerical derivative error terms to the simulations. Error assessment using the finite-volume method focuses on identification of optimal model-discretization combinations for the best LES predictions using application solvers. The pseudospectral and finite-volume approaches show consistent predicted behavior of the interplay between numerical and modeling errors for different model-discretization combinations. Of those studied here, the two most optimal combinations are identified as: (1) standard or dynamic Smagorinsky SGS model with a bounded central difference scheme; and (2) Monotonic Integrated LES (MILES) with either a second-order upwind or QUICK scheme. These combinations were applied for an axisymmetric jet flow at Re ∼ 105, using an ‘engineering quality’ mesh, where the MILES model with either an upwind or QUICK scheme showed the best predictive capability.