Hybrid Simulation Approach for Cavity Flows: Blending, Algorithm, and Boundary Treatment Issues

Abstract

The maturation of high-performance computer architectures and computational algorithms has prompted the development of a new generation of models that attempt to combine the robustness and ef ciency offered by the Reynolds averaged Navier-Stokes equations with the higher level of modeling offered by the equations developed for large eddy simulation. The application of a new hybrid approach is discussed, where the transition between these equation sets is controlled by a blending function that depends on local turbulent ow properties, as well as the local mesh spacing. The utilization of local turbulence properties provides added control in specifying the regions of the ow intended for each equation set, removing much of the burden from the grid-generation process. Moreover, the model framework allows for the combination of existing closure model equations, avoiding the dif culty of formulating a single set of closure coef cients that perform well in both Reynolds averaged and large eddy simulation modes. Simple modi cations to common second-order accurate Reynolds averaged Navier-Stokes algorithmsare proposed to enhance the capturing of large eddy motions. Incompressible Poiseuille ow, supersonic base ow, and supersonic ow over recessed cavities were considered to evaluate various aspects of the proposed model and computational framework. Calculations using another hybrid approach (detached eddy simulation) were also performed for comparison. Nomenclature a i j = matrix of normalized velocity variances and covariances b = law of the wall intercept C D k! = cross-diffusion term for the speci c turbulent dissipation rate C d = constant for the destruction term of k C DES = constant for the detached eddy simulation (DES) length scale C p = pressure coef cient C s = constant for the Smagorinsky viscosity coef cient C ¹ = constant for the turbulent viscosity coef cient C ! p = constant for the production term of ! c = constant for the gradient sensor D = cavity depth d = distance to the nearest solid surface F = blending function F c ; F d ; F u = central, dissipative, and upwind contributions to the inviscid uxes f = joint normal distribution function h = channel height k = turbulent kinetic energỳ = turbulent length scale m = normalized cavity mass n;¯= constants for the latency parameter of Speziale 6 P = pressure P k = production term for the turbulent kinetic energy P ! = production term for the speci c turbulent dissipation rate q = MUSCL extrapolated variable R = sting base radius r = numerical dissipation reduction factor r; µ = polar coordinates t = time u i = velocity u ¿ = friction velocity V = turbulent velocity scale v i = velocity normalized with the rms of its uctuation x; y; z = Cartesian coordinates 1 = subgrid length scale ± i j = Kronecker delta ± C ; ± ¡ = nite difference operators (forward, backward) ² = turbulent dissipation rate or generic small numbeŕ = blending function argument · = MUSCL interpolation constant or law of the wall slope ¹ = viscosity coef cient º = speci c viscosity coef cient ...

High-fidelity simulations of unsteady civil aircraft aerodynamics: stakes and perspectives. Application of zonal detached eddy simulation

“…Especially within ZDES, there is no need to switch off the subgrid-scale model to simulate accurately jet flows conversely to zonal RANS/implicit large eddy simulation (ILES), which have also proven to be efficient in certain situations like jet flows [39][40][41] and base flows [42][43][44]. The interaction of these instabilities with the pylon is captured by the simulation.…”

Simulation of Transient Dynamics of Shock Wave Boundary Layer Interactions Using Hybrid Large-Eddy/Reynolds-Averaged Navier-Stokes Models

“…Several approaches have been used in our earlier work to extend Menter's model to serve as the RANS component of a hybrid LES/RANS closure [9][10][11][12][13]. In this investigation, we have opted for a simple strategy motivated by Strehlets' [22] development of a two-equation detached-eddy simulation method.…”

“…A strong coupling of the model response with the boundary layer structure is therefore required, and this is facilitated first by the design of a RANS-to-LES transition function that is based on the modeled structure of the flow in the logarithmic region and second, by the use of "recycling / re-scaling" techniques that sustain the transport of large turbulent structures in the outer part of the boundary layer. While predictions of mean-flow properties have been evaluated in earlier works [9][10][11][12][13], this study also focuses on the prediction of second-moment quantities and on the dynamics of the shock / boundary layer interaction. In this scope and in its one-to-one comparisons with available experimental data at the same conditions, the present work is closely related to the Mach 3 LES investigation of flow over a compression corner conducted by Loginov, et al [14], the Mach 3 DNS compression-corner calculations of Wu and Martin [15], and the earlier "very" large-eddy simulations of Mach 3 flow over a compression corner of Hunt and Nixon [16].…”

“…As shown later, the RANS predictions for Dolling's compression-corner experiments (and similar shock / boundary layer interactions) are sensitive to the constant sst A in Eq. (9), which ranges between zero (for no SST modification) to one (for the full SST contribution).…”

“…The objective of the current work is to apply a hybrid LES/RANS framework developed at NCSU [9][10][11][12][13] to the Dolling Mach 5 compression-corner configuration to determine the model's ability to capture the time-dependent structure of this flow. The viewpoint taken in the NCSU model is that the RANS component functions as a form of near-wall closure in an otherwise LES description of the flow.…”