Professional Documents
Culture Documents
a r t i c l e
i n f o
Article history:
Received 20 December 2009
Received in revised form 14 April 2010
Accepted 12 May 2010
Available online 6 July 2010
Keywords:
3-D separation
Wall-bounded ow
LES
Hybrid LES/RANS
a b s t r a c t
An incompressible fully-developed duct ow expanding into a diffuser, whose upper wall and one side
wall are appropriately deected (with the expansion angles of 11.3 and 2.56 respectively), and for
which reference experimental and DNS databases were provided by Cherry et al. (2008, 2009) and Ohlsson et al. (2009, 2010), was studied computationally by using a zonal hybrid LES/RANS (HLR) method,
proposed recently by Kniesner (2008) and Jakirlic et al. (2009). In addition a complementary Large-Eddy
Simulation (LES) method has been applied. The ow Reynolds number based on the height of the inlet
channel is Reh = 10,000. The primary objective of the present investigation was the comparative assessment of the computational models in this ow conguration characterized by a complex 3-D ow separation being the consequence of an adverse-pressure gradient evoked by the duct expansion. The focus of
the investigation was on the capability of different modelling approaches to accurately capture the size
and shape of the 3-D ow separation pattern and associated mean ow and turbulence features.
2010 Elsevier Inc. All rights reserved.
1. Introduction
Congurations involving 3-D boundary-layer separation are
among the most frequently encountered ow geometries in practice. Accordingly, the methods simulating them have to be appropriately validated using detailed and reliable reference databases.
However, the largest majority of the experimental benchmarks
being used for validating computational methods and turbulence
models relates to internal, two-dimensional ow congurations,
as e.g., ow in a 2-D diffuser (e.g., Obi et al., 1993), ow over a
backward-facing step and a forward-facing step, or ow over
fences, ribs, 2-D hills and 2-D humps mounted on the bottom wall
of a plane channel. In these examples it is assumed that the inuence of the side walls (according to Bradshaw and Wong (1972),
the minimum aspect ratio representing the ratio of the channel
height to channel width should be 1:10 in order to eliminate
the inuence of the side walls) is not felt at the channel midplane.
Consequently, within a computational framework, the spanwise
direction can be regarded as a homogeneous one, the fact enabling
the application of the periodic boundary conditions (even 2-D
computations when using the RANS approach). By doing so, the
3-D nature of the ow is completely missed: strong secondary mo* Corresponding author at: Institute of Fluid Mechanics and Aerodynamics,
Petersenstr. 30, D-64287 Darmstadt, Germany.
E-mail address: s.jakirlic@sla.tu-darmstadt.de (S. Jakirlic).
0142-727X/$ - see front matter 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.ijheatuidow.2010.05.004
tion across the inlet section of the channel induced by the Reynolds
stress anisotropy which is, as generally known, beyond the reach
of the eddy-viscosity RANS model group, complex 3-D separation
patterns spreading over duct corners (corner separation and corner
reattachment), etc.
These circumstances were the prime motivation for the recent
experimental study of the ow in a three-dimensional diffuser
conducted by Cherry et al. (2008, 2009). Such a diffuser conguration has also a high practical relevance. It mimics a diffuser
situated between a compressor and the combustor chamber in
a jet engine. Its task is to decelerate the ow discharging from
compressor over a very short distance to the velocity eld of
the combustor section. Typically a uniform inlet prole over
the diffuser outlet is desirable. Such a ow situation is associated
by a strong pressure increase. Cherry et al. provided a detailed
reference database comprising the pressure distribution along
the bottom non-deected wall, three-component mean velocity
eld and the streamwise Reynolds stress component eld within
the entire diffuser section. Recently Ohlsson et al. (2009, 2010)
have performed a complementary Direct Numerical Simulation
using a massively parallel high-order spectral element code.
The 3-D diffuser was meshed by approximately 172 million grid
points. In addition to the mean velocity eld, all six Reynolds
stress components were evaluated, as well as the surface
pressure distribution along the bottom wall. It should be noted
that in the experimental investigations two three-dimensional
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
821
imental and DNS database, the results obtained are comparatively assessed to a complementary LES simulation.
2. 3-D diffuser: case description
The diffuser shape, dimensions and the coordinate system are
shown in Fig. 1. The ow in the inlet duct (height h = 1 cm, width
B = 3.33 cm) corresponds to fully-developed turbulence (enabled
experimentally by a development channel being 62.9 channel
heights long). The L = 15h long diffuser section is followed by a
straight outlet part (12.5h long). Downstream of this the ow goes
through a 10h long contraction into a 1 in. diameter tube. The curvature radius at the walls transitioning between diffuser and the
straight duct parts are 6 cm. The bulk velocity in the inow duct
is Ubulk = Uinflow = 1 m/s in the x-direction resulting in the Reynolds
number based on the inlet channel height of 10,000. The origin of
the coordinates (y = 0, z = 0) coincides with the intersection of the
two non-expanding walls at the beginning of the diffusers expansion (x = 0). The working uid is water ( q = 1000 kg/m3 and
l = 0.001 Pas).
3. Computational method
The continuity and momentum equations governing the incompressible ow in the present 3-D diffuser conguration read:
@U i
0
@xi
@U i @U i U j
1 @P
@
sm st
@xj
@t
q @xi @xj ij ij
Here, smij 2mSij (with Sij 0:5 @U i =@xj @U j =@xi being the rate of
strain tensor) represents the viscous stress tensor, whereas the turbulent stress tensor stij is to be modelled. The rationale and the most
Fig. 1. Geometry of the 3-D diffuser considered (not to scale), Cherry et al. (2008).
822
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
"
!#
DU i
1 @p
@
@U i @U j
m m t
Dt
q @xi @xj
@xj @xi
The coupling of the instantaneous LES eld and the ensembleaveraged RANS eld at the interface is realized via the turbulent viscosity, which makes it possible to obtain solutions using one system
of equations. This means practically that the governing equations
are solved in the entire solution domain irrespective of the ow
sub-region (LES or RANS). Depending on the ow zone, the hybrid
model implies the determination of the turbulent viscosity mt either
from a near-wall ke RANS model: mt = Clflk2/e or from the subgrid-scale (SGS) model in the LES formulation (standard Smagorinsky model was applied, 1963): mt mSGS C S D2 jSj. The Smagorinsky constant CS takes the value of 0.1. D = (Dx Dy Dz)1/3
represents the lter width and jSj Sij Sij 1=2 the strain rate modulus.
The near-wall variation of the turbulent viscosity mt is obtained from
a two-equation, ke RANS model, implying solution of the transport
equations for kinetic energy of turbulence k and its dissipation rate e.
The near-wall and viscous damping functions and the relevant
source terms are presently modelled in line with the proposal of
Launder and Sharma (LS, 1974), with e representing a homogeneous part of the total viscous dissipation rate e ! ~e e 2m
1=2
1=2
@k =@xk @k =@xk , taking zero value at the wall (~ewall 0). This
near-wall model is a straightforward extension of the standard
high-Reynolds number ke model aimed at capturing the enhanced
viscosity inuence on turbulence in the immediate wall vicinity
(viscous sub-layer and buffer layer). Solving the equations for k
and e in the RANS sub-domain (being necessary for the turbulent
viscosity determination) requires appropriate boundary conditions
at the RANS-LES interface. The approach used presently is based on
the equality of the RANS interface values for k and e with corresponding SGS values. This condition, originating from the equivalence of the total stresses at both sides of interface (the RANS
total stress, similar as the LES one, comprises both resolved and
RANS
modelled parts) - sLES
tot stot , reduces to the equality of their modelled contributions by assuming the continuity of the resolved fractions across the interface (ifce), Temmerman et al. (2005):
2mSGS Sij 2mt Sij . Finally, the condition applied at the interface
implies the equality of the (modelled) turbulent viscosities:
kSGS
C S D2 jSj2
p ;
Cl
eSGS C S D2 jSj3
RANS
LES
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
The present HLR model has been intensively validated in congurations of different geometrical complexity featuring ow separation (from sharp-edged and continuous curved surfaces) and
strong heating under the conditions of constant and variable uid
properties. Also, different sub-grid-scale (SGS) model formulations
in LES (besides the Smagorinsky model the transport model for the
residual kinetic energy from Yoshizawa and Horiuti (1985) has
been tested) were combined with different eddy-viscosity-based
RANS models. More details about the present HLR method are given in Jakirlic et al. (2006, 2009) and Kniesner (2008).
cc
T test
Ud
i Uj Ui Uj
ij
SGS
ij
1
test
b2 b c
T test
dij T test
ij
kk 2C g D j Sj Sij 2C g aij
3
1
2
SGS
sSGS
dij sSGS
ij
kk 2C g D jSjSij 2C g aij
3
^ are obtained by
The (test-) ltered quantities denoted with
applying the test-lter to each control volume (CV) as the volumeweighted average of the CV value itself and its twenty-six (26)
neighbours. Beside the cell volumes additional weighting factors
are taken into account. The variable arrangement and the test-lter
with corresponding weighting factors are depicted in Fig. 3. The resolved part of the turbulent stresses Lij is computed in accordance
with Germano (1992):
cc
Lij T test
sSGS
Ud
i Uj Ui Uj
ij
ij
8
9
dSGS
Lij 2C g atest
ij 2 C g aij
10
11
SGS
Lij 2C g ad
atest
2C g M ij
ij
ij
12
Ui Uj Ui Uj
823
C g ~
x; t
Lij M ij
2M ij M ij
13
n1
~
C
x; t n bC g ~
x; t n1
1 bC g ~
g x; t
14
Since C
g can become negative, leading to a negative effective
viscosity and to unphysical behaviour and numerical instabilities,
the negative values of Cg are clipped. Furthermore, an upper bound
of Cg is introduced to ensure the model does not become too diffusive. This nally yields the model coefcient in the corresponding
expression for the turbulent viscosity enabling a converged solution at the new time level tn1
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
824
i o
C g ~
x; t n1 min max C
g ;0 ;1
15
Fig. 4. Solution domain and computational mesh used in the LES simulation (every 5th grid line is plotted).
Fig. 5. Near-wall resolution expressed in wall units along the upper and lower walls in the central xy plane (z/B = 1/2) in the HLR simulation.
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
825
Fig. 7. Proles of the kinetic energy of turbulence at two selected streamwise locations (x/h = 2 inlet duct and x/h = 10 diffuser section) obtained by the present HLR
method (vertical lines denote the LES/RANS interface position).
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
826
Fig. 8. The instantaneous velocity eld obtained by LES (small gure on the upper left illustrates the turbulent inlet data generation through a separate inow duct simulation
utilizing a simultaneously running periodic channel simulation).
(a) experiment:
(b) HLR:
0.1 m/s
0.1 m/s
Fig. 10. Semi-log plots of axial velocity component across the inow duct (x/h = 2) and at a cross-section in the interior of the diffuser (x/h = 10).
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
827
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
x/h=15
x/h=12
x/h=8
x/h=5
x/h=2
828
Experiment
LES
HLR
Fig. 12. Comparison between experimentally and computationally obtained iso-contours of the axial velocity eld in the cross-planes yz at selected streamwise locations
within the diffuser section (the thick line denotes the zero-velocity line).
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
829
830
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
Fig. 14. Evolution of the axial velocity proles in the vertical plane xy at three spanwise locations z/B = 1/4, 1/2 and 3/4.
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
831
Fig. 15. Evolution of the turbulent streamwise stress component proles in the vertical plane xy at three spanwise locations z/B = 1/4, 1/2 and 3/4.
832
S. Jakirlic et al. / International Journal of Heat and Fluid Flow 31 (2010) 820832
5. Conclusions
Two eddy-resolving modelling approaches: Large-Eddy Simulation (LES) and a zonal hybrid LES/RANS scheme (HLR) were used to
predict the ow in a 3-D diffuser, measured recently by Cherry et
al. (2008, 2009) and computed by means of Direct Numerical Simulation by Ohlsson et al. (2009, 2010), aiming at a comparative
analysis of their features and performance. The latter method,
developed by the authors, merging a low-Reynolds-number ke
Reynolds-Averaged NavierStokes (RANS) model with Large-Eddy
Simulation (LES) in the framework of a two-layer scheme, was
the focus of the present investigation, keeping in mind that the basic motivation behind the HLR scheme is to obtain results on a
coarser grid, whose quality is comparable with that of a well-resolved LES. In order to check its feasibility, an intensive validation
has been performed. The ow conguration exhibited a 3-D recirculation pattern was thought to represent a suitable benchmark.
Good results were obtained by both computational models with
respect to the characteristics of the duct ow expanding into a diffuser section, the consequent separation ow region (onset, shape
and size), the mean velocity eld and associated integral parameters (pressure distribution), as well as the turbulence quantities.
More importantly, a close agreement was achieved between the
hybrid LES/RANS model and both experimental and DNS reference
databases with respect to the near-wall behaviour of the timeaveraged quantities, demonstrating its high potential for computing such complex, wall-bounded turbulent ows.
Acknowledgements
The work of G. Kadavelil and M. Kornhaas is nancially
supported by the Deutsche Forschungsgemeinschaft (DFG) within
the German Collaborative Research Center (SFB 568) Flow and
Combustion in Future Gas Turbine Combustion Chambers. Our
special thanks goes to the authors of the reference results J. Eaton
and E. Cherry (experimental database) and J. Ohlsson (DNS database) for making their data available.
References
Bradshaw, P., Wong, F.Y.F., 1972. The reattachment and relaxation of a turbulent
shear layer. J. Fluid Mech. 52 (1), 113135.
Cherry, E.M., Elkins, C.J., Eaton, J.K., 2008. Geometric sensitivity of threedimensional separated ows. Int. J. Heat Fluid Flow 29, 803811.
Cherry, E.M., Elkins, C.J., Eaton, J.K., 2009. Pressure measurements in a threedimensional separated diffuser. Int. J. Heat Fluid Flow 30, 12.
Durst, F., Schfer, M., 1996. A parallel block-structured multigrid method for the
prediction of incompressible ows. Int. J. Numer. Methods Fluids 22, 549565.
Ertem-Mueller, S., 2003. Numerical efciency of implicit and explicit methods with
multigrid for Large Eddy Simulation in complex geometries. Dissertation.
Technische Universitt Darmstadt.
FASTEST-Manual, 2005. Institute of Numerical Methods in Mechanical Engineering,
Department of Mechanical Engineering, Technische Universitt Darmstadt,
Germany.
Frhlich, J., von Terzi, D., 2008. Hybrid LES/RANS methods for the simulation of
turbulent ows. Prog. Aerosp. Sci. 44, 349377.
Germano, M., 1992. Turbulence: the ltering approach. J. Fluid Mech. 238, 325336.
Germano, M., Piomelli, U., Moin, P., Cabot, W.H., 1991. A dynamic subgrid-scale
eddy viscosity model. Phys. Fluids A 3 (7), 17601765.
Jakirlic, S., Kniesner, B., aric, S., Hanjalic, K., 2006. Merging near-wall RANS models
with LES for separating and reattaching ows. In: ASME Joint USEuropean Fluids
Engineering Summer Meeting: Symposium on DNS, LES and Hybrid RANS/LES
Techniques, Paper No. FEDSM2006-98039, Miami, FL, USA, July 1720.
Jakirlic, S., Kniesner, B., Kadavelil, G., Gnir, M., Tropea, C., 2009. Experimental and
computational investigations of ow and mixing in a single-annular combustor
conguration. Flow Turbul. Combust. 83 (3), 425448.
Jakirlic, S., Kniesner, B., Kadavelil, G., 2010. A method for interface-turbulence
forcing in hybrid LES/RANS simulations. In: 8th Int. ERCOFTAC Symp. on
Engineering Turbulence Modelling and Measurements, Marseille, France, June
911.
Jakirlic, S., Kadavelil, G., Sirbubalo, E., von Terzi, D., Breuer, M., Borello, D., 2010.
Report on 14th ERCOFTAC Workshop on Rened Turbulence Modelling.
September 18, 2009. Sapienza University of Rome, ERCOFTAC Bulletin, No.
84, 2010.
Kniesner, B., 2008. Ein hybrides LES/RANS Verfahren fr konjugierte Strmung,
Wrme- und Stoffbertragung mit Relevanz zu Drallbrennerkongurationen (A
hybrid LES/RANS method for conjugated ow, heat and mass transfer with
relevance to swirl combustor congurations). PhD Thesis, Technische
Universitt Darmstadt. <http://www.tuprints.ulb.tu-darmstadt.de/950/>.
Launder, B.E., Sharma, B.I., 1974. Application of the energy dissipation model of
turbulence to the calculation of ow near a spinning disc. Lett. Heat Mass
Transfer 1, 131138.
Lilly, D.K., 1992. A proposed modication of the Germano subgrid-scale closure
method. Phys. Fluids A 4 (3), 633635.
Mason, P.J., Callen, N.S., 1986. On the magnitude of the subgrid-scale eddy
coefcient in large-eddy simulation of turbulent channel ow. J. Fluid Mech.
162, 439462.
Nikitin, N., 2008. On the rate of spatial predictability in near-wall turbulence. J.
Fluid Mech. 614, 495507.
Obi, S., Aoki, K., Masuda, S., 1993. Experimental and computational study of
turbulent separating ow in an asymmetric plane diffuser. In: Proc. 9th Symp.
on Turbulent Shear Flows, Kyoto, Japan, August 1619, pp. 305.1305.5.
Ohlsson, J., Schlatter, P., Fischer, P.F., Henningson, D.S., 2009. DNS of threedimensional separation in turbulent diffuser ows. In: Advances in Turbulence
XII. Proceedings of the 12th EUROMECH European Turbulence Conference, vol.
132. Springer Proceedings in Physics, Marburg. ISBN: 978-3-642-03084-0.
Ohlsson, J., Schlatter, P., Fischer, P.F., Henningson, D.S., 2010. DNS of separated ow
in a three-dimensional diffuser by the spectral-element method. J. Fluid Mech.
650, 307318.
Rhie, C.M., Chow, W.L., 1983. A numerical study of the turbulent ow past an
isolated airfoil with trailing edge separation. AIAA J. 21, 15251532.
Schneider, H., von Terzi, D., Bauer, H.-J., Rodi, W., 2010. Reliable and accurate
prediction of three-dimensional separation in asymmetric diffusers using largeeddy simulation. ASME J. Fluids Eng. 132 (3), 031101031111.
Smagorinsky, J., 1963. General circulation experiments with the primitive
equations: I. The basic experiment. Mon. Weather Rev. 91 (3), 99164.
Spalart, P.R., 2009. Detached-Eddy simulation. Annu. Rev. Fluid Mech. 41, 181202.
Spalart, P.R., Jou, W.-H., Strelets, M., Allmaras, S., 1997. Comments on the feasibility
of LES for wings and on a hybrid RANS/LES approach. In: 1st AFOSR Int. Conf. on
DNS and LES, Columbus, OH, USA, August 48.
Steiner, H., Jakirlic, S., Kadavelil, G., aric, S., Manceau, R., Brenn. G., 2009. Report on
13th ERCOFTAC Workshop on Rened Turbulence Modelling. September 2526,
2008. Graz University of Technology, ERCOFTAC Bulletin, No. 79, pp. 2429.
Sternel, D.C., Junglas, D., Martin, A., Schfer, M., 2004. Optimisation of partitioning
for parallel ow simulation on block structural grids. In: Topping, B.H.V., Mota
Soares, C.A. (Eds.), Proc. of the 4th Int. Conf. on Engineering Computational
Technology, Stirling: Paper 93. Civil-Comp Press, Stirling.
Temmerman, L., Hadziabdic, M., Leschziner, M.A., Hanjalic, K., 2005. A hybrid twolayer URANS-LES approach for large eddy simulation at high Reynolds numbers.
Int. J. Heat Fluid Flow 26, 173190.
Yoshizawa, A., Horiuti, K., 1985. A statistically-derived subgridscale kinetic energy
model for the large-eddy simulation of turbulent ows. J. Phys. Soc. Jpn. 54,
28342839.