A.W. Vreman

The filtering analog of the variational multiscale method in large-eddy simulation

The variational multiscale method introduced by Hughes et al. [Comput. Visual. Sci. 3, 47 (2000)] is extended to the classic filtering approach in large-eddy simulation. The role of the Germano identity in the formulation is precisely indicated. Multiscale methods based on standard eddy-viscosity models are related to (anisotropic) hyperviscosity models under certain conditions. Several models are tested and found to be as accurate as the standard dynamic model, while the implementations are more simple. Finally, the turbulent stress tensor is reformulated, such that filter and derivative in the filtered equations can be treated as a single operator.

Comparison of direct numerical simulation databases of turbulent channel flow at Re τ = 180

Direct numerical simulation (DNS) databases are compared to assess the accuracy and reproducibility of standard and non-standard turbulence statistics of incompressible plane channel flow at Re τ = 180. Two fundamentally different DNS codes are shown to produce maximum relative deviations below 0.2% for the mean flow, below 1% for the root-mean-square velocity and pressure fluctuations, and below 2% for the three components of the turbulent dissipation. Relatively fine grids and long statistical averaging times are required. An analysis of dissipation spectra demonstrates that the enhanced resolution is necessary for an accurate representation of the smallest physical scales in the turbulent dissipation. The results are related to the physics of turbulent channel flow in several ways. First, the reproducibility supports the hitherto unproven theoretical hypothesis that the statistically stationary state of turbulent channel flow is unique. Second, the peaks of dissipation spectra provide information on length scales of the small-scale turbulence. Third, the computed means and fluctuations of the convective, pressure, and viscous terms in the momentum equation show the importance of the different forces in the momentum equation relative to each other. The Galilean transformation that leads to minimum peak fluctuation of the convective term is determined. Fourth, an analysis of higher-order statistics is performed. The skewness of the longitudinal derivative of the streamwise velocity is stronger than expected (−1.5 at

Dynamic inverse modeling and its testing in large-eddy simulations of the mixing layer

y^{+}

Large-Eddy simulation of a particle-laden turbulent channel flow

=30). This skewness and also the strong near-wall intermittency of the normal velocity are related to coherent structures.

Noncommuting Filters and Dynamic Modelling for Les of Turbulent Compressible Flow in 3D Shear Layers

We propose new identities for dynamic subgrid modeling in large-eddy simulation involving an explicit filter and its inverse. Exact defiltering of a class of numerical realizations of the top-hat filter is developed. The approach is applied to large-eddy simulation of the temporal mixing layer. Smagorinsky’s model is adopted as base model and the results are compared to the standard dynamic eddy-viscosity model as well as to filtered DNS (direct numerical simulation) results. The difference between the results of the two models for the present application is found to be quite small. This is explained by performing a sensitivity analysis with respect to the dynamic coefficient, which hints towards a “self-restoring” response underlying the observed robustness of the physical predictions. Using DNS data the validity of the assumption that the model coefficients are independent of filter width is tested and found to favor the inverse modeling procedure. The computational effort of the dynamic inverse model is 15% smaller than of the standard dynamic eddy-viscosity model.

Development and applications of a 3D compressible Navier-Stokes solver

Large-eddy simulations of a vertical turbulent channel flow with 420,000 solid particles are performed in order to get insight into fundamental aspects of a riser flow The question is addressed whether collisions between particles are important for the ow statistics. The turbulent channel ow corresponds to a particle volume fraction of 0.013 and a mass load ratio of 18, values that are relatively high compared to recent literature on large-eddy simulation of two-phase ows. In order to simulate this ow, we present a formulation of the equations for compressible ow in a porous medium including particle forces. These equations are solved with LES using a Taylor approximation of the dynamic subgrid-model. The results show that due to particle-uid interactions the boundary layer becomes thinner, leading to a higher skin-friction coefcient. Important effects of the particle collisions are also observed, on the mean uid prole, but even more o on particle properties. The collisions cause a less uniform particle concentration nand considerably atten the mean solids velocity prole.

LES modeling errors in free and wall bounded compressible shear layers

Large-eddy simulation of complex turbulent flows involves the filtering and modelling of small scale flow structures whose intensity shows large spatial variations in the flow domain. This suggests the use of filters with nonuniform filterwidth. Such filters fail to commute with spatial derivatives and give rise to additional ‘noncommutation’ terms in LES. We construct higher order filters and show that the subgrid terms and the new noncommutation terms are a priori of comparable magnitude. We apply these filters to DNS data of the temporal mixing layer. The magnitude of the noncommutation terms and their contribution to the kinetic energy dynamics is determined. Finally, we show that LES predictions significantly depend on the specific explicit filter used in dynamic subgrid modelling.

Dynamic self-organization in particle-laden turbulent channel flow

As a part of the Dutch ISNaS project our group and NLR jointly develop a flow solver for compressible, turbulent flow. This flow solver is especially aimed at applications on the industrial level: the nm]ti-element airfoil and wing/body combination, both at transonic flow conditions. The flow solver is based on the Reynolds-averaged Navier-Stokes equations, in which presently the algebraic Baldwin-Lomax turbulence model is adopted. In reference [1] the first results, for laminar and turbulent flow around ~ single airfoil and over a finite flat plate have been shown. In the present paper recent developments in the solver are discussed. In section 2 the numerical method used in the ISNaS solver is briefly described. Section 3 discusses the role of the numerical, or artificial dissipation in relation to the physical dissipation. In section 4 numerical aspects of the extension of the monoblock solver to a multiblock solver are described. The numerical method used in the ISNaS solver serves as a basis for many CFD programs used in our group. These programs are not only intended for the tw o applications mentioned above, but also for more fundamental studies of turbulence (with the help of large eddy simulation (LES) and direct numerical simulation (DNS)) and for the simulation of viscous water waves. In section 5 of this paper the use of the numerical method in large eddy simulation is discussed.

Comparison of subgrid-models in LES of the compressible mixing layer

We present direct numerical simulation results (DNS) for a compressible mixing layer and a compressible boundary layer on a flat plate in 2D. The exact filtered Navier-Stokes equations are derived and the contribution of the subgrid-terms and the discretisation errors is calculated by filtering the DNS data to a coarser grid. It will be shown that the discretisation errors are larger than the dominant subgrid terms if the filterwidth is comparable to the grid spacing. This holds for second and fourth order accurate spatial discretisations.

A FINITE VOLUME APPROACH TO LARGE EDDY SIMULATION OF COMPRESSIBLE, HOMOGENEOUS, ISOTROPIC, DECAYING TURBULENCE

We study fundamental aspects of turbulent riser-flow which contains large numbers of interacting particles. We include particle-particle as well as particle-fluid interactions. These interactions and the flow-forcing are the source for the dynamic formation and destruction of large-scale coherent particle swarms in the flow. We establish the basic scenario of this self-organization and investigate the dominant aspects of the resulting turbulence modulation. Large-eddy simulations with different subgrid models, and large numbers of particles at a significant volume fraction and realistic mass load ratio, indicate the development of a thinner boundary layer and an accumulation of particles near the walls. At an average volume fraction of ≈ 1.5% it was found that neglecting particleparticle interactions leads to an unphysical modulated flow.