You are on page 1of 6

Developing a test bed for the validation of

wall-modeled LES and RANS LES hybrid methods for


complex high Reynolds number flows
Francois Cadieux
Department of Aerospace & Mechanical Engineering
University of Southern California
Scott Murman
Fundamental Modeling & Simulation Branch
NASA Advanced Supercomputing Division
February 2, 2015
Abstract
Large eddy simulation (LES) has proven to be an accurate and relatively robust tool for
predicting transitional, separated and unsteady flows. Flows over complex geometries at high
Reynolds numbers such as full aircrafts at cruise speeds are out of reach for wall-resolved large
eddy simulation (LES) due to the stringent requirements to resolve thin turbulent boundary
layers (TBL). Wall-modeled LES (WMLES) and hybrid RANS LES (HRL) methods try to
bridge this gap by removing the need to resolve the flow near surfaces. A number of contenders
have made significant progress, but the wall models and HRL models developed have not been
scrutinized and validated to the same extent as LES models, which casts doubts over their
predictive power and accuracy. The first part of the proposed research project is to implement
the two most promising approaches dynamic slip wall model (LES-DSWM), dynamic hybrid
RANS LES (DHRL). The second goal is to validate and compare their performance on a number
of canonical and complex geometries for which either experimental or DNS data is readily
available. The use of a high order discontinuous Galerkin (DG) compressible Navier-Stokes solver
will allow the unambiguous assessment of the impact of these models without the influence of
unquantified numerical dissipation. The final goal is then to compare their accuracy and relative
computational cost to using a single of large layer of high order elements extending from the
surface to the beginning of the log-layer, making full use of the flexibility of a discontinuous
Galerkin CFD solver.

1
1.1

Introduction
Context

Large eddy simulation (LES) has proven to be an accurate and relatively robust tool for predicting
transitional, separated and unsteady flows despite its shortcomings. LES converges to DNS results
as the resolution is increased instead of grid-converged LES results because the filter width is
generally tied to the grid size instead of a physical quantity [15]. Its performance is also strongly
correlated with the level of numerical dissipation and thus the order of accuracy of the solver used
1

February 2, 2015

F. Cadieux, S. Murman

since subgrid scale (SGS) dissipation provided by LES models is often of the same order or smaller
than numerical dissipation [11, 5]. Flows over complex geometries at high Reynolds numbers
such as full aircrafts at cruise speeds remain out of reach for wall-resolved large eddy simulation
(LES) due to the stringent requirements to resolve thin turbulent boundary layers (TBL) [16].
Reynolds averaged Navier-Stokes (RANS) turbulence modeling approaches are more tractable but
provide limited insight into these types of flows because they preclude some important transient
and unsteady dynamics. Unsteady Reynolds averaged Navier-Stokes (URANS) is an improvement,
but it only provides partial unsteady information due to the large eddy viscosity damping the
growth of most instabilities. Wall-modeled LES (WMLES) and hybrid RANS LES (HRL) methods
try to bridge this gap by removing the need to resolve the flow near surfaces while providing
sufficient insight into the flow dynamics and unsteadiness. Much emphasis has been placed on the
development of WMLES and HRL methods in the last 20 years to enable more accurate treatment
of unsteady high Reynolds number flows over complex geometries.
Although successful in many ways, Spalarts detached eddy simulation (DES) approach has endemic
issues that may be fundamental to the method: log-layer mismatch caused by modeled stress
depletion (MSD), and non-monotonic grid dependence which can cause grid induced separation
[21, 22]. Attempts to address these issues without heavily modifying the original formulation like
delayed detached eddy simulation (DDES) [21], and zonal methods [7, 6] have had some success
but also discovered other issues in their application such as sensitivity to grid skewness [2].
Partially averaged Navier-Stokes (PANS) [10, 1] and other variable resolution approaches such as
the scale-adaptive simulation (SAS) [13, 9] and turbulence-resolving RANS (TRANS) [20] avoid
these issues by tying the eddy viscosity to physical quantities like energy grid density or an integral
length scale. Their turbulence modeling approaches and results are often closer to URANS than
LES and are thus limited in their ability to capture transient and unsteady effects accurately [13].
Their accuracy in pure LES mode has also not been thoroughly validated [17].
The dynamic hybrid RANS LES (DHRL) framework trades the simplicity of DES and PANS for
a more mathematically and physically grounded approach. The velocity is decomposed into three
components: the Reynolds averaged (
u), resolved fluctuating (u ), and unresolved fluctuating (u )
velocity.
ui = u
i + ui + ui = u
bi + ui

(1)

u
is the ongoing time average of u
b. The modeled Reynolds and sub grid scale stresses are computed
and balanced using local turbulent quantities produced by each model [2]. The formulation is thus
model independent any RANS and SGS model pair may be used and shows some significant
improvements over other HRL methods whilst avoiding some of their pitfalls [24]. The simulation
may start in pure LES mode until enough time averaging is obtained to have the RANS model
active, allowing better predictions of transition at the cost of a coarse LES. This difficulty may be
avoided in cases where transition is not present by starting the simulation from a RANS or URANS
solution.
Most LES wall models make one of two assumptions which are too strict to allow their widespread
use: they either assume an equilibrium boundary layer like in RANS, or they assume a specific
form of the mean velocity profile [14]. The recent dynamic slip velocity wall model makes no such
assumption and derives an off-wall boundary condition for LES based on a specific differential filter
form and a dynamic procedure [3]. The method is able to resolve a recirculating region at the end of
a NACA 4412 airfoil where most RANS and URANS models fail [3]. To achieve similar agreement
2

February 2, 2015

F. Cadieux, S. Murman

for the same case with a 3D URANS type wall model on an embedded grid, the turbulence model
coefficients have to be adjusted using a local and time averaged dynamic procedure [14].

1.2

Motivation

It is clear that a number of contenders to bridge the gap between wall-resolved LES and RANS
have made significant progress, but the LES wall models and HRL models developed have not
been scrutinized and validated to the same extent as LES models, which casts doubts over their
predictive power and accuracy [17]. HRL methods often require new turbulence models or modified
SGS models which have often not been thoroughly validated in pure LES mode, or even in pure
RANS mode. Yet, some important weaknesses have already been identified in a number of these
methods. New methods like the DHRL which claim to avoid these pitfalls deserve more scrutiny
[2, 24]. Similarly, a number of recent LES wall models aim to properly predict non-equilibrium flows
with pressure gradients and recirculation regions. Some extending RANS ideas, whereas others like
the dynamic slip velocity model take a different approach. The accuracy and computational cost
of these models remains to be explored and compared to HRL methods.

Proposed project

The proposed project is to validate the most promising hybrid RANS LES method, DHRL, and
LES wall-model, dynamic slip velocity (LES-DSWM), on a number of canonical and complex high
Reynolds number flows for which experimental and/or DNS data are available. In addition, a novel
method unique to the discontinuous Galerkin formulation is to be developed and tested. Each
methods performance will be compared on a number of relevant metrics: skin friction, pressure
coefficient, mean velocity profiles, RMS velocities, as well as computational cost.

2.1

Technical approach

The first step will be to implement the dynamic hybrid RANS LES method [2] and dynamic slip
velocity wall model [3] in an unstructured discontinuous Galerkin Navier-Stokes solver. This will
allow the unambiguous evaluation of each method without any effects from unquantified numerical
dissipation. To make sure of this, the numerical dissipation of the solver can be computed using
the method developed by Schranner et al [18] and compared to that of the models employed in
these methods. Each method should then be tested and validated on a simple canonical flow such
as turbulent channel flow at a Reynolds number for which both experimental and DNS data are
available, e.g. 590, to ensure it is performing as the authors of the method reported. The two
methods will then be applied to more complex flows to test the accuracy of their predictions, and
their convergence characteristics when the grid is refined as well as when the order of the polynomial
bases in the discontinuous Galerkin elements is increased. For example, the Ahmed body at a
Reynolds number of 7.68 105 with a sub-critical slant angle of = 25 remains an important case
where experimental data is available. Previous HRL methods like DES and wall-modeled LES have
improved upon RANS results for this case but failed to capture the reattachment observed near the
middle of the slant in experiments and high order LES [19]. Given the importance of the turbulent
boundary layers in this case, particular care should be devoted to verifying that the DHRL method
does not suffer from modeled-stress depletion (MSD) or the log-layer mismatch observed in DES.
3

February 2, 2015

F. Cadieux, S. Murman

Another interesting case would be to simulate an airfoil near stall at a moderately high Reynolds
number of 2 106 such as to investigate the capability of the DHRL and LES-DSWM approaches
to capture the successively occurring separation bubble, reattachment, and turbulent separation on
the suction side of the airfoil [12]. Grid refinement studies would need to be performed to ensure the
methods converge to a solution monotonically and that they do not exhibit particular sensitivity to
grid-induced separation. This test will also serve as a good indication of the performance to expect
from these methods in massively separated flows.
Discontinuous Galerkin (DG) methods allow another unique solution avenue to reduce computational requirements near the wall: a single layer of large higher order (N=32 to N=64 in the vertical)
elements could be used instead of a wall model or RANS hybrid model. Such large elements near
the wall can mitigate time step restrictions. Significant savings can also be achieved by using
higher order polynomials: a laminar separation bubble flow was resolved accurately in an LES with
only 32 Chebyshev points in the vertical for a domain twice as large as the maximum turbulent
boundary layer thickness at the outflow [4]. The computational cost of a single layer of large high
order elements spanning y + = 0 to y + = 80 may be comparable to HRL method, but will benefit
from remaining closer to a pure LES. Coupled with the truncated Navier-Stokes (TNS) approach
[8, 23], the proposed method could even side-step assumptions made in LES, i.e. that the filtered
Navier-Stokes equations are the appropriate equations when the smallest eddies are not resolved.
By using DG with TNS, the pure Navier-Stokes dynamics are preserved, and filtering is applied to
remove energy accumulated in the small scales only when it reaches unphysical levels. This third
approach requires few modifications, and should provide an interesting benchmark to compare the
DHRL and LES-DSWM approaches to. Although it is specific to discontinuous and finite element
formulations, this method may allow wall-resolved LES to reach higher Reynolds numbers that
would otherwise be prohibitively expensive to compute.

2.2

Expected Outcomes

Two novel approaches to simulate high Reynolds number flows over arbitrary geometry will be
implemented in a discontinuous Galerkin solver: the dynamic hybrid RANS LES method, and the
LES with dynamic slip wall modeling. Each will be validated on a canonical flow, and then further
scrutinized through simulations of complex high Reynolds number flows for which experimental
data is available. Their grid convergence properties, sensitivity to grid-induced separation, and
their ability to capture turbulent and massively separated flows will be investigated. Comparisons
to pure LES with large high order elements near the surfaces will help better understand the
effects of the wall-models on the flow. Knowledge gained from this endeavor will help better our
understanding of these methods and how to best apply them. Ultimately, this effort is to help
simulations at even higher Reynolds numbers succeed where other HRL methods have only had
mitigated success.

February 2, 2015

F. Cadieux, S. Murman

References
[1] Branislav Basara, Sinisa Krajnovic, Sharath Girimaji, and Zoran Pavlovic. Near-wall formulation of the partially averaged navier stokes turbulence model. AIAA Journal, 49(12):26272636,
Dec 2011.
[2] S. Bhushan and D. K. Walters. A dynamic hybrid reynolds-averaged navier stokeslarge eddy
simulation modeling framework. Physics of Fluids (1994-present), 24(1):, 2012.
[3] S. T. Bose and P. Moin. A dynamic slip boundary condition for wall-modeled large-eddy
simulation. Physics of Fluids (1994-present), 26(1):, 2014.
[4] F. Cadieux and J. A. Domaradzki. Performance of subgrid scale models in coarse large eddy
simulation of a laminar separation bubble over a flat plate. submitted to Phys. Fluids, 2015.
[5] F. Cadieux, J. A. Domaradzki, T. Sayadi, and T. Bose. DNS and LES of laminar separation
bubbles at moderate Reynolds numbers. ASME J. Fluids Eng., 136(6), 2014.
[6] Sebastien Deck. Recent improvements in the zonal detached eddy simulation (zdes) formulation. Theoretical and Computational Fluid Dynamics, 26(6):523550, 2012.
Weiss, Mathieu Pami`es, and Eric Garnier. Zonal detached eddy
[7] Sebastien Deck, Pierre Elie
simulation of a spatially developing flat plate turbulent boundary layer. Computers & Fluids,
48(1):1 15, 2011.
[8] J. Andrzej Domaradzki, Kuo Chieh Loh, and Patrick P. Yee. Large eddy simulations using
the subgrid-scale estimation model and truncated navier-stokes dynamics. Theoretical and
Computational Fluid Dynamics, 15(6):421450, 2002.
[9] Y. Egorov, F.R. Menter, R. Lechner, and D. Cokljat. The scale-adaptive simulation method for
unsteady turbulent flow predictions. part 2: Application to complex flows. Flow, Turbulence
and Combustion, 85(1):139165, 2010.
[10] Sharath Girimaji and Khaled Abdol-Hamid. chapter Partially-Averaged Navier Stokes Model
for Turbulence: Implementation and Validation. Aerospace Sciences Meetings. American Institute of Aeronautics and Astronautics, Jan 2005. 0.
[11] A.G. Kravchenko and P. Moin. On the effect of numerical errors in large eddy simulations of
turbulent flows. Journal of Computational Physics, 131(2):310 322, 1997.
[12] Ivan Mary and Pierre Sagaut. Large eddy simulation of flow around an airfoil near stall. AIAA
Journal, 40(6):11391145, 2002.
[13] F.R. Menter and Y. Egorov. The scale-adaptive simulation method for unsteady turbulent
flow predictions. part 1: Theory and model description. Flow, Turbulence and Combustion,
85(1):113138, 2010.
[14] George Ilhwan Park and Parviz Moin. An improved dynamic non-equilibrium wall-model for
large eddy simulation. Physics of Fluids (1994-present), 26(1):, 2014.
[15] U Piomelli. Large eddy simulations in 2030 and beyond. Philosophical Transactions of the
Royal Society of London A: Mathematical, Physical and Engineering Sciences, 372(2022), 2014.

February 2, 2015

F. Cadieux, S. Murman

[16] Ugo Piomelli. Wall-layer models for large-eddy simulations. Progress in Aerospace Sciences,
44(6):437 446, 2008. Large Eddy Simulation - Current Capabilities and Areas of Needed
Research.
[17] Pierre Sagaut and Sebastien Deck. Large eddy simulation for aerodynamics: status and perspectives. Philosophical Transactions of the Royal Society of London A: Mathematical, Physical
and Engineering Sciences, 367(1899):28492860, 2009.
[18] F.S. Schranner, J. A. Domaradzki, S. Hickel, and N.A. Adams. Assessing the numerical
dissipation rate and viscosity in numerical simulations of fluid flows. submitted to Computers
& Fluids, 2015.
[19] Eric Serre, Matthieu Minguez, Richard Pasquetti, Emmanuel Guilmineau, Gan Bo Deng,
Michael Kornhaas, Michael Sch
afer, Jochen Fr
ohlich, Christof Hinterberger, and Wolfgang
Rodi. On simulating the turbulent flow around the ahmed body: A frenchgerman collaborative evaluation of {LES} and {DES}. Computers & Fluids, 78(0):10 23, 2013. {LES} of
turbulence aeroacoustics and combustion.
[20] Mikhail L. Shur, Philippe R. Spalart, Mikhail Kh. Strelets, and Andrey K. Travin. A hybrid rans-les approach with delayed-des and wall-modelled {LES} capabilities. International
Journal of Heat and Fluid Flow, 29(6):1638 1649, 2008.
[21] Philippe R. Spalart. Detached-eddy simulation. Annual Review of Fluid Mechanics, 41(1):181
202, 2009.
[22] PhilippeR. Spalart. Topics in detached-eddy simulation. In Clinton Groth and DavidW. Zingg,
editors, Computational Fluid Dynamics 2004, pages 312. Springer Berlin Heidelberg, 2006.
[23] T. Tantikul and J.A. Domaradzki. Large eddy simulations using truncated navier-stokes equations with the automatic filtering criterion. J. Turb., 11(21):124, 2010.
[24] D.K. Walters, S. Bhushan, M.F. Alam, and D.S. Thompson. Investigation of a dynamic
hybrid rans/les modelling methodology for finite-volume cfd simulations. Flow, Turbulence
and Combustion, 91(3):643667, 2013.

You might also like