Experimental study of an active grid-generated shearless mixing layer and comparisons with large-eddy simulation

Abstract

A shearless mixing layer characterized by interactions between two regions with different turbulence intensities but without mean shear is investigated experimentally in a wind tunnel. Reynolds numbers higher than those of prior studies [B. Gilbert, “Diffusion mixing in grid turbulence without mean shear,” J. Fluid Mech. 100, 349 (1980); S. Veeravalli and Z. Warhaft, “The shearless turbulent mixing layer,” J. Fluid Mech. 207, 191 (1989); B. Knaepen, O. Debliquy, and D. Carati, “Direct numerical simulation and large-eddy simulation of a shear-free mixing layer,” J. Fluid Mech. 514, 153 (2004); D. Tordella and M. Iovieno, “Numerical experiments on the intermediate asymptotics of shear-free turbulent transport and diffusion,” J. Fluid Mech. 549, 429 (2006); D. A. Briggs, J. H. Ferziger, J. R. Koseff, and S. G. Monismith, “Entrainment in a shear-free turbulent mixing layer,” J. Fluid Mech. 310, 215 (1996)] are achieved by using an active grid with rotating winglets on one-half of its cross section. Stationary flow-conditioning fine meshes are used to avoid mean velocity gradients. Measurements are performed at five different downstream wind-tunnel locations using an X-type hot-wire probe and a stereoscopic particle image velocimetry system. The Reynolds numbers based on the Taylor microscale in the high- and low-kinetic energy regions are 170 and 88, respectively. The energy and integral length-scale ratios between the two regions are 4.27 and 1.73, respectively. The inlet turbulence in the upper and lower portions of the shearless mixing layer is not fully isotropic, with the streamwise velocity fluctuations being between 6% and 13% higher than the cross-stream ones. Fundamental statistical properties of the flow are documented and analyzed at various scales using band-pass box-filtered velocities. Downstream evolution of variance and half-width of the mixing layer, skewness and flatness factors, as well as the statistics of two-point velocity increments at various displacements are presented. It is found that much of the deviations from Gaussian statistics originate from large-scale motions. The data are well suited to be used as initial condition for simulations and as test for large-eddy simulation (LES) models and codes. Comparison studies for three LES models including Smagorinsky, dynamic Smagorinsky, and dynamic mixed nonlinear models are implemented in simulations of temporally decaying shearless mixing layer using a pseudospectral code. Initial conditions are prescribed by matching the longitudinal energy spectra at all heights across the layer for both streamwise and cross-stream velocity components. LES with all three subgrid scale models tested underpredicts the kinetic energy and exhibits deviations from the measured non-Gaussian behaviors. Overall, the dynamic Smagorinsky model predicts statistics slightly better than the other two models.