Professional Documents
Culture Documents
1 Introduction
The lid driven cavity flow is widely considered as one of the benchmark problems
for closure model development and code validation in computational fluid dynamics.
The relatively complex flow phenomena that appear in a seemingly simple geometry
have attracted much work on this problem in the past few decades. The flow geometry
for the lid driven cavity is shown in Fig. 1. The lid moves in the X direction, Y is
in the vertical direction and the spanwise direction is Z . The spanwise aspect ratio
of the cavity (SAR) is defined as the ratio of spanwise length (L) to the height
(H ) of the cavity. Koseff and Street [1] performed experiments for SAR 3:1:1 at
Re = 10,000 and measured mean velocity profiles close to the symmetry plane.
They also performed measurements for turbulent shear stress at selected locations.
Prasad and Koseff [2] measured second moment of velocities at the symmetry plane
for different SAR and Reynolds numbers. As most of the experimental data and
numerical simulations in literature are available at the SAR of 3:1:1 and 1:1:1 and
for Re = 10,000, these flow conditions are considered for current investigation.
B. Akula P. Roy P. Razi S. Anderson S. Girimaji (B)
Texas A&M University, College Station, TX, USA
e-mail: girimaji@tamu.edu
Springer International Publishing Switzerland 2015
S. Girimaji et al. (eds.), Progress in Hybrid RANS-LES Modelling,
Notes on Numerical Fluid Mechanics and Multidisciplinary Design 130,
DOI 10.1007/978-3-319-15141-0_29
359
360
B. Akula et al.
The objective of this work is to compare the PANS method against URANS and
LES for lid driven cavity flow simulation. PANS is a hybrid model that is intended to
bridge smoothly between URANS and LES/DNS. In PANS, the accuracy of results
can be optimized based on available computational resources. In the URANS method,
only the unsteady mean flow i.e. scales that are comparable to the geometry of the
flow are resolved, whereas all other scales are modeled. In LES, all the large scale
motions or energy carrying eddies are computed, and the small scale motions, that
are more universal in nature, are modeled [3]. To illustrate the operative regions of
PANS, a typical spectrum of energy as a function of wavelength for turbulent flow is
shown in Fig. 2. The relative cut-off for unresolved flow scales is shown for RANS,
PANS, and LES. The cut off parameters f k and f for PANS are defined as the ratio
of the unresolved to resolved turbulent kinetic energy and dissipation, respectively.
Value of f k close to zero indicates the limit of DNS (Direct Numerical Simulation)
and the value of unity is essentially a URANS simulation. As an important part
of the closure development process, it is essential to perform PANS simulations on
benchmark flow problems in order to verify its ability to resolve wider range of scales
than URANS, and closely reproduce experimental data at higher degree of fidelity
on much coarser grids than LES/DNS.
361
In this paper, we simulate lid driven cavity problem using PANS k turbulence
model. The parameter f k is varied between simulations and its effect on mean flow
and turbulence quantities is examined. Section 2 presents the PANS formulation and
equations. Numerical simulation setup is presented in Sect. 3. The results are presented in Sect. 4 and we conclude in Sect. 5 with a brief discussion. In the companion
paper [4], we examine the vortical structures within the cavity flow as computed
by PANS.
2 Hybrid Models
The development of all the hybrid models commences from the incompressible
Navier-Stokes equations and continuity equation:
Vi
Vi
p
2 Vi
=
+
,
+ Vj
t
x j
xi
x jx j
Vi
= 0.
xi
(1)
(2)
The difference between URANS and PANS/ LES lies in how the averaged or filtered
velocity equations are obtained from Eq. 1. URANS uses an averaging operator
leading to equations that describe the mean velocity field. On the other hand, PANS/
LES uses a generalized homogeneous filter to decompose the velocity into resolved
and unresolved part [5]:
Vi = Ui + u i ,
Ui = Vi ,
u i = 0
(3)
By applying the filter to the Navier-Stokes equations, the momentum equations for
the resolved field are given as:
(Vi , V j ) p
Ui
2 Ui
Ui
+ Uj
=
+
,
t
x j
x j
xi
x jx j
(Vi , V j ) = Vi V j Vi V j .
(4)
(5)
The term (Vi , V j ) in Eq. 4 represents the sub-filter stress, given by Eq. 5 [6].
The sub filter stress term is modeled differently in various turbulence models.
LES: The generalized sub-filter stress in LES is modeled via Boussinesq-type
approximation [68]:
i j = (Vi , V j ) =
2
ki j T Si j
3
(6)
where T , and k are the unresolved eddy viscosity and the unresolved kinetic
energy respectively. Si j is the resolved strain-rate tensor defined as:
362
B. Akula et al.
Si j =
1
2
U j
Ui
+
x j
xi
(7)
(8)
Here, is the grid spacing and C is the model coefficient. If the filtercut-off is in
the inertial range, then the values of the Smagorinsky constant (Cs = C) usually
lie between 0.18 and 0.23. In the dynamic Smagorinsky model, the model parameter
C is not constant, rather it is calculated from the energy content of the smallest
resolved scales [10]. In order to make the model self-consistent, an additional test
> ) is introduced and C = Cd (x, y, z, t) is dynamically adjusted based
filter (
on the following identity:
L i j = Ti j
ij
(9)
(10)
(11)
2 |
S| S
i j = 2
ij
(12)
i j = 2 |S|Si j
(13)
where,
PANS: The filtering procedure in PANS modeling is similar to LES, which separates
the flow into resolved and unresolved features. However, while LES uses explicit filtering, PANS uses implicit filtering via the parameters f k and f (or f k and f ) in such
a way that the governing equations resolve a selected portion of the unsteady scales
of motion. Thus, the degree of PANS resolution is dependent upon the parameters
f k,, . These parameters designate the ratio of unresolved to resolved kinetic energy
(k), dissipation (), or turbulence frequency (), which are defined as:
fk =
ku
k
f =
363
f =
(14)
In PANS, the sub-filter stress term in Eq. 5 is also modeled with Boussinesq approximation:
U j
Ui
2
i j = (Vi , V j ) = ku i j u
+
.
(15)
3
x j
xi
Where u is the unresolved eddy viscosity defined as u = ku /u = C ku2 /u .
Based on the cut-off parameters defined in Eq. 14, the governing equations for PANS
k model are given by
ku
ku
ku
,
+ Uj
= Pu ku u +
( + u /ku )
t
x j
x j
x j
u
u
u
u
+ u /u
+ Uj
= Pu u2 +
,
t
x j
ku
x j
x j
ku u
+ ku u ,
Pu = f k P
f fk
=
+
.
f
f
(16a)
(16b)
(16c)
(16d)
fk
,
f
u =
fk
f
(17)
The values of the closure coefficients are as follows: = 0.09, = 5/9, = 0.075,
k = 2.0, and = 2.0.
URANS: In URANS, Eq. 5 becomes the familiar Reynolds stress term [1113],
where the filter denotes an ensemble average. With two-equation URANS models,
the Reynolds stress term is given by:
ui u j =
2
ki j T
3
U j
Ui
+
x j
xi
(18)
364
B. Akula et al.
365
366
(a)
B. Akula et al.
(b)
Fig. 4 Comparison of a U mean velocity along the vertical center line of the symmetry plane, b
V mean velocity profile across horizontal center line of the symmetry plane for SAR 3:1:1 cavity
at Re = 10,000
(a)
(b)
Fig. 5 Comparison of a U mean velocity along the vertical center line of the symmetry plane b V
mean velocity profile across horizontal center line of the symmetry plane for SAR 1:1:1 cavity at
Re = 10,000
for SAR 1:1:1 cavity. This can be attributed to the breaking effect [16] caused by
the closer wall distances at lower aspect ratio of cavity. In contrast to the SAR =
3:1:1 case, LES performs well in predicting the peaks and gradients of the mean
velocity profiles for the SAR 1:1:1. In fact, it matches well with the experimental
measurements closer to the upstream wall and bottom wall for the mean vertical and
horizontal velocity profiles, respectively. This may be due to the lower grid resolution
needed to resolve the low turbulent flow features at SAR 1:1:1. In addition, PANS
and RANS calculations predict the mean velocity profiles and their gradients accurately at this low Reynolds number simulation. This is consistent with the conclusions
of SAR = 3:1:1 at Re = 10,000. Figures 4 and 5 demonstrate that for the current aspect
ratios studied at Re = 10,000, URANS is able to predict the mean velocity profiles
adequately well. The results indicate that PANS simulations yield a slight improvement in the mean velocity profiles over URANS calculation. The difference between
URANS and PANS calculations might be more apparent at high Reynolds numbers
367
simulations where a wider range of turbulence length scales are present in the flow.
Unfortunately, experimental data for lid driven cavity flow is not available at higher
Reynolds numbers and limits the study to the low Reynolds number of 10,000.
(19)
where < U > and < V > are the time averaged resolved velocities. The unresolved part < u v > is calculated from Boussinesqs approximation and is time
averaged during the simulation. Only the case of SAR = 1:1:1 is considered here as
experimental data is not available for the other case.
Figure 6 shows the total shear stress profiles along the cavity centerlines for different models. As seen from the experimental data, the shear stress peaks close to the
walls. Variation of shear stress along the vertical line is shown in Fig. 6a. The results
show that none of the simulations capture the peak precisely. However, the trends for
shear stress variation along the centerline are well captured by both the PANS cases
shown. For the variation along the horizontal centerline, shown in Fig. 6b, the peak
near the upstream wall is negative and small compared to the peak near the downstream wall. PANS model f k = 0.5 for 653 grid predicts the peaks well compared to
other models. In addition, URANS calculation is not able to capture the peak values
of shear stress close to the walls. It is interesting to note that, URANS does capture
all the peaks and valleys in the interior of the domain reasonably well.
(a)
(b)
Fig. 6 Comparison of total turbulent shear stress < u v >=< u v >r esolved + < u v >u a along
the vertical center line of the symmetry plane b along the horizontal center line of the symmetry
plane, for SAR 1:1:1 cavity at Re = 10,000
368
B. Akula et al.
The results presented here for shear stress profiles indicate that overall, PANS performs adequately in predicting the trends and peak values throughout the domain.
Additionally, URANS calculation is able to resolve the shear stress components in
the interior of the domain faily well. As discussed in the previous section, this performance of URANS simulation is attributed to the low turbulence levels presented
in this flow. On the other hand, LES fails to reproduce the right behavior of shear
stress in both near wall regions and interior of the domain for the grid resolutions
studied here.
5 Conclusion
Numerical simulations of lid driven cavity flow for two aspect ratios of SAR =
3:1:1 and SAR = 1:1:1 at Re =10,000 are performed using PANS k model. The
results from PANS simulations are compared against URANS, LES Smagorinsky
model and LES dynamic Smagorinsky model. It is important to reiterate that, the
near wall resolution in the present simulations may be quite inadequate for LES
computations. Nonetheless, it is the objective of the paper to compare PANS and LES
on identical grids. LES Smagorinsky model fails to capture the mean velocity profiles
accurately for the current grid resolutions and only LES dynamic Smagorinsky model
is primarily used for comparison purposes. Overall, PANS simulations adequately
capture the behavior of mean velocity profiles and peak values of shear stress along
the cavity centerlines compared to other models. The total turbulent kinetic energy
obtained by different simulations is shown to be very small compared to the total
energy imparted, indicating low Reynolds number nature of the flow. Under the
current conditions, the flow is highly unsteady and complex, but does not exhibit a
wide range of turbulence scales. Therefore, URANS appears to be nearly as good as
PANS or LES simulations in predicting the mean velocity and shear stress profiles.
Acknowledgments The authors would like to thank Kyle Knox for his assistance in the initial
phase of this project. The Texas A&M Supercomputing Facility (http://sc.tamu.edu/) is gratefully
acknowledged for providing computing resources useful in conducting the research reported in this
paper. The work was supported by NASA NRA #NNXA161A.
References
1. Koseff, J.R., Street, R.L.: The lid-driven cavity flow: a synthesis of qualitative and quantitative
comparisons. Trans. ASME 106, 390398 (1984)
2. Prasad, A.K., Koseff, J.R.: Reynolds number and end-wall effects on a lid-driven cavity flow.
Phys. Fluids 1(2), 208218 (1989)
3. Piomelli, U.: Large-eddy simulation: achievements and challenges. Prog. Aerosp. Sci. 35(4),
335362 (1999)
369
4. Razi P., Venugopal, V., Jagannathan, S., Girimaji, S.: Partially-averaged Navier-Stokes (PANS)
simulations of lid-driven cavity flowpart ii: flow structures. In: 5th Symposium on Hybrid
RANS-LES Methods (2014)
5. Lakshmipathy, S., Girimaji, S.S.: Partially-averaged Navier-Stokes method for turbulent flows:
k- model implementation. AIAA Paper 119:2006 (2006)
6. Germano, M.: Turbulence: the filtering approach. J. Fluid Mech. 238, 325336 (1992)
7. Murthi, A., Reyes, D., Girimaji, S., Basara, B: Turbulent transport modelling for pans and other
bridging closure approaches. In: Proceedings of V European Conference on CFD, ECCOMAS
CFD (2010)
8. Lakshmipathy, S.: Partially averaged Navier-Stokes method for turbulence closures: characterization of fluctuations and extension to wall bounded flows. Ph.D. dissertation, Texas A&M
University (2009)
9. Germano, M., Piomelli, U., Moin, P., Cabot, W.H: A dynamic subgrid-scale eddy viscosity
model. Phys. Fluids A: Fluid Dyn. (19891993), 3(7), 17601765 (1991)
10. Ghosal, S., Lund, T.S., Moin, P., Akselvoll, K.: A dynamic localization model for large-eddy
simulation of turbulent flows. J. Fluid Mech. 286, 229255 (1995)
11. Girimaji, S.S., Abdol-Hamid, K.S.: Partially-averaged Navier-Stokes model for turbulence:
implementation and validation. Number 0502. AIAA (2005)
12. Girimaji, S.S.: Partially-averaged Navier-Stokes model for turbulence: a Reynolds-averaged
Navier-Stokes to direct numerical simulation bridging method. J. Appl. Mech. 73(3), 413421
(2006)
13. Girimaji, S.S., Suman, S.: Partially averaged Navier Stokes (PANS) method for turbulence
simulations: theory and practice. In: Progress in Hybrid RANS-LES Modelling, pp. 2943.
Springer, New York (2012)
14. OpenCFD: OpenFOAMThe Open Source CFD ToolboxUsers Guide, 1.4 edn. OpenCFD
Ltd., Bracknell (2007)
15. Jordan, S.A., Ragab, S.A.: On the unsteady and turbulent characteristics of the threedimensional shear-driven cavity flow. J. Fluids Eng. 106, 386389 (1984)
16. Shankar, P.N., Deshpande, M.D.: Fluid mechanics in the driven cavity. Ann. Rev. Fluid Mech.
32, 93136 (2000)