Professional Documents
Culture Documents
N U M E R I C A L S I M U L A T I O N OF T U R B U L E N T FLOW OVER
S U R F A C E - M O U N T E D O B S T A C L E S WITH S H A R P E D G E S A N D
CORNERS
Summary
Results from numerical simulations of turbulent flow around a surface-mounted cube and over
a surface-mounted square rib are presented. The first problem is treated by solving the Reynolds
equations together with the standard k-¢ turbulence model, for 0 ° and 45 ° orientation of the flow
obstacle, in uniform and boundary layer incident flow. The second problem is solved by applying
the concept of large-eddy simulation. In both calculation concepts, sufficient spatial resolution is
essential to get satisfying agreement with experimental data in the critical flow regimes, e.g. above
the top face of the flow obstacles. The computations based on large-eddy simulation are very
expensive but can provide excellent agreement of the numerical results with the experimental data
for the second-order statistics.
1. Introduction
The numerical simulation of high Reynolds number turbulent flow over and
around flow obstacles represents an important aspect of the activities in the
field of wind engineering. A surface-mounted rib or a cube, with sharp edges
and corners, can serve as a prototype of a building. To study the resulting
turbulent flow field, with separation and reattachment of the flow, numerical
simulations can be carried out using two basically different concepts to provide
a numerical solution. The first concept is based on the solution of the Rey-
nolds-averaged Navier-Stokes equations for the mean flow field together with
a turbulence model. A two-equation model of turbulence is considered to be
the minimum requirement for a sufficiently accurate representation of the to-
*Present address: CONVEX Computer G.m.b.H., Schatzbogen 54, D-8000 Munich 82, F.R.G.
tTo whom correspondence should be sent.
tal effect of turbulent velocity fluctuations on the mean flow. In this paper we
selected the standard k-~ turbulence model. The second concept, large-eddy
simulation (LES), is based on the solution of the basic Navier-Stokes equa-
tions, averaged over the minimum grid cell which can be resolved on a given
computer. The effect of the non-resolvable motion on the grid-scale flow field
is taken into account by a subgrid-scale (SGS) turbulence model. In this paper
we selected as the SGS model the so-called Smagorinsky-Lilly model. The
LES concept allows one to evaluate directly the Reynolds shear stresses on the
grid scale, and the SGS contributions are negligible away from rigid walls (with
diminishing importance near the wall if the grid resolution is improved).
Therefore the LES concept makes it possible to provide information for tur-
bulence modellers working with the first concept. In addition, LES always pro-
vides a three-dimensional and time-dependent database which can be evalu-
ated to give new insights into the complex interactions of large-scale structures
in turbulent flows around obstacles.
Previous numerical work by Hirt et al. [ 1 ] considered a cube normal and at
45 ° orientation to the incident flow, solving the Reynolds equations together
with an arbitrary eddy viscosity. A k-e turbulence model together with hybrid
discretization of the convective terms have been applied in the doctoral thesis
of Vasilic-Melling [2], and the results were compared with the experimental
data of Castro and Robins [ 3 ] for a cube normal to the upstream flow. A recent
paper by Paterson and Apelt [4 ] includes results from a computation using a
k-e turbulence model and hybrid discretization, and the results are compared
with the data in ref. 3 for a cube with a simulated atmospheric boundary layer
serving as incident flow normal to the front face of the cube.
The LES concept was applied for the first time by Murakami et al. [5] to
calculate the flow over a periodic arrangement of cubes in a simulated atmos-
pheric boundary layer, i.e. applying periodic boundary conditions in the main
flow direction.
Recent numerical solutions of the Reynolds equations together with a k-e
turbulence model for a square rib on the bottom of a plate channel are available
from Kessler et al. [6]. First-order upstream and central differencing (second
order) together with Richardson extrapolation are used to estimate numerical
solution errors. Then, in comparison with the experimental data [7], a clear
separation between pure numerical errors and turbulence model deficiencies
could be achieved.
Solutions from an LES for a periodic arrangement of square ribs have been
obtained by Kobayashi et al. [8].
In this paper the flow problems treated are the flow around a cube and the
flow over a square rib. These cases have been selected because detailed exper-
imental data are available from Castro and Robins [3] (for turbulent flow
around a cube in uniform as well as in simulated atmospheric boundary layer
flow for a Reynolds number Re between 3 X 104 and 105) and from Dimaczek
131
et al. [ 7 ] (for flow over a cross-wind square rib mounted on the floor of a plate
channel for a Reynolds number Re of about 42500, based on bulk velocity and
obstacle height). The first problem, a cube normal and at 45 ° to the incident
flow (Section 2.1 ), will be solved on the basis of the Reynolds equations to-
gether with the standard k-~ turbulence model. The second problem, a square
rib on the bottom of a plate channel (Section 2.2), will be solved by applying
the LES concept, using as inflow condition the results from a large-eddy sim-
ulation for a fully developed channel flow.
/
/
/ -----
R( i / . . . . . .
B
l ,tr,
]
q
::l
t
x
convex corners (see Fig. 3 ) must be formulated with care since they can strongly
influence flow separation. For example, the total flux across the bottom face
of the U-momentum cell above a convex corner (Fig. 3 ) was evaluated as the
sum of the diffusive flux across the wall part (which is taken into account by
an additional source term) and the diffusive flux across the open part. The
situation can be handled numerically by use of a harmonic averaging procedure
133
0 -- O 0
UT
ut
UW Fp _ _ Up FE _
UE UEE
--~ 0 ----~
WB -
- -- u b _ WEB
/ / / / / / . [//////, " / / / / / A / / / / / / / / / / / / / / / / / / / / / /
%
OFB 0 FEB 0
z ~ WBB
I UB B
O 0 O
~ I m - X
0 ~ 0 O
L--:---------JLit - -
UW [" Up U_ UEE
p- --+-~ ~ r E ~ o -
l
WB ~ ub ~ WEB
"/////////'//////// /
/'~ UB F
O rB ---,-- O EB O
/
¢ I
/~,UBB
O -~- O O
~-- x
Fig. 3. U - m o m e n t u m cell a t a c o n v e x c o r n e r .
for the turbulent exchange coefficient, which reads in the case demonstrated
in Fig. 3 as follows:
F~ F~ F~ F ~
F ~ - F ~ + F~ t F~ + F~B
If F is set to a very small negative value inside rigid walls (to characterize
134
the rigid walls), this averaging procedure allows autovectorization of the re-
sulting computer code for rather complicated building forms (for further de-
tails see ref. 10). This type of interpolation of exchange coefficients was orig-
inally proposed by Patankar [11 ] for thermal problems.
4.0
TMTr TT " F
I "I
3.5]
3.0 I :
-
2.S
i I • !I .......... t I
! i I
I ! I • IIIIllItHIltltIft ........ i !
2.0
I, Ii " 1• i IIII!HIItlIHII.I
IIII/IHlllII/ll
............
I . . . . . . . . . . I 4
I .S! I r ; ' - I 4 i i.l.l.i-{.I J I +.FI.l.I I i • I I
I : [ ! I: ! : : + :l:i t ":":" : ::: ":+':":I:::"":" : ' ! I
i !1 IIIH lt:I~ :• : i i I', I
: I
! t I I
.0
-4 0 -3.0 -~.0 -1.0 -.0 1.0 "20 3.0 4.0 S 0 6.0 X
• I i , L ¸ 11!IHH!lli~J ¸
_.3 J~ ~ ~lIIIllHliIIilIIJI--
ii , ILt
ltHlilltlLltl11L
+ + i ~-- ,_~___~._ t~.
-+.'{
3 01
-+.0 -3.0
+
-2.0
'
- 9 -o I.o
II
• --
2. 3.o +o
~i
so 0 ×
%, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(P
; o I / ' ~
~!!
.s31 i /
~+",-~'f " '
00
-.3
- S
-I 0
-~ .3 i
_,s . ! i
-~ .B i T
-2.0 FRO
0 G .$ 10 I S 20 2.S 3.0
FR0 ?0P BAC
ference between the two cases is that in case B the shear layer separating at
the leading horizontal edge reattaches on the roof and the pressure coefficient
recovers to a value close to zero.
Results of numerical simulations of the flow fields just described are com-
pared with the experimental data in Fig. 5. Taking advantage of the symmetry
of the problem, the grid shown in Fig. 4 has been used for the present case
(cube normal to upstream flow) with Nx×Nv×Nz=30× 1 4 × 19 grid points
136
(a)
Z
4.00
3.00
2.00
1.00
Y
-~,,00 -3.00 -2.00 -I.O0 ,00 1,00 2,00 3.00 A,O0 5.00 8,00 X
U PROFILES IN THE PLANE Y = .00 {J=TB} - - = 2.00
- I .00
-2.00
-5.00
~4.00
-¢,00 -5.00 -2.00 - I .00 .00 1.00 2.00 3.00 C.O0 5.00 5.00 X
U PROFILES IN THE PLANE Z = .55 {K= BI - - = 2.00
CP
1.00
.50
\
t
• O0
-.50
-1.00
-1.50
-2.00
I
• 00 .50 .00 I .50 2.00 2.50 3.00
FRO TOP SAC
CP I
1.00
.50
I
.00
1
-.50
I
- I .00
-I .So
FRO
-2.00
.00 .50 1.00 1.50 2.oo 2.50 3.0o
FRO RGT 8AC
Fig. 5 (continued).
137
(b)
--4- - -
4
+
1 I,
-3.0 " -2.0 -I .0 -.0 .0 2,0 3.0 &.O S,O 6.0 X
PROFILES IN THE PLANE Z • 55 (K- 81 - - ' ~ . 0 0
rP
i 0
T
:i::l:+
+ *~~
- S
-I .0 - -
-I S
t-Ro
-7.0
0.0 .S 1.0 I ,5 ?.0 ? S 3.0
FRO TOP BAC
CF
1.o I
.S
I
O.0
-.S
- I .0
- I .S
FRO
-?.0
0.0 .S I .0 I .S ?.0 ? S
FRO RGT BAC
Fig. 5. Cube normal to incident flow: (a) case A-uniform incident flow with low turbulence inten-
sity; (b) case B-boundary layer incident flow with high turbulence intensity.
138
(not counting grid points in the additional two layers of basic cells for the
boundary conditions). For case A, excellent agreement of the numerical sim-
ulation with the experimental data could be reached for the dimensionless
pressure difference C~= 2 (p--Pref)/flUref2, with the reference pressure Pref and
the reference velocity Ureftaken from the inflow cross-section at cube height.
From the results of case B it is obvious that the calculated pressure coefficient
does not recover as quickly as the measured one. The calculated flow field does
not separate at all from the top surface of the cube. The experiment shows that
the separation bubble is very small (its thickness is about one-tenth of the
cube height), and obviously the spatial resolution chosen for the numerical
simulation is still too coarse to capture this small recirculation zone on the top
surface of the cube.
Cube at 45 ° to the upstream flow. When the flow approaches the cubic build-
ing at an angle of 45 ° to the vertical building faces, two strong counter-rotating
vortices are generated at the swept-back leading edges. These vortices are sim-
ilar to the vortices shed by a delta wing and have a dominant influence on the
mean flow field above the top surface and in the wake of the cube. Just behind
each vertical trailing edge of the cube, two vertical counter-rotating vortices
are generated by rolling-up of the shear layers separated from the side surfaces.
The horizontal and vertical vortices together create an extremely complex wake
structure. Unfortunately, from the experiments of Castro and Robins it is dif-
ficult to distinguish any region of reattachment downstream of the cube.
The numerical simulation of the 45 ° case is much more difficult to accom-
plish than the simulation of the cube with an approach flow normal to the
frontal face. To resolve properly the vortices generated at the swept-back lead-
ing edges of the top surface, a dense distribution of the grid points immediately
above the top surface had to be chosen. This special choice was essential to get
satisfactory agreement between experiment and numerical simulation. Figure
6 shows a vertical plane of the distribution of Nx X Ny X N z = 33 X 33 X 24 basic
cells in the flow field which we finally used for the calculations (case A only).
Figure 6 also shows the pressure distribution on the top surface on a line from
one corner to the other, for different cases of spatial resolution NTX X NTV
above the top surface. For the case with NTX X NTy = 8 X 8 cubic control cells
above the top surface, no horizontal vortices developed and the corresponding
pressure distribution did not even qualitatively agree with the experiment. With
an increasing number NTx X NTy of control cells the pressure distribution im-
proved significantly. Finally, a further improvement could be reached by using
NTX X Nwv = 14 X 14 grid points above the top surface and by lowering the ver-
tical distance of the grid points next to the top surface to about 50% of the
horizontal distance. In the experiments of Castro and Robins the frontal stag-
nation point was switching intermittently from one side of the vertical leading
edge to the other side. Because of this unstable situation there was a distinct
asymmetry in the pressure distribution along lines indicated in Fig. 6 and
139
Z
4.00
3.00
2.00 , ,
I I
II I I ! :!:]:!:!~]:!:!: !: !
i i
1 .00 I . . . . . . . . . . . . . . .
II
.00
-2:oo :oo 2:oo 4:oo x
BACK
t
X /
VALUE5 OF C?
-.io00
-.SO00
\4 //
. . . . . 5000
. . . . . .7000
. . . . . . .6000
.......... .~000
// \
\ /-~/
; / 1 --~
\\:/,~-,
\ \
............
-I.0000
1.!000
-I.2000
-}.3000
" ~ ~ , r?__-~-~.~ j.
Cp
1.00
.50
L."-->-7
.00
-.50
// I
- I .00 -~.~+ ~ -..
-I ,50
-2.00
-~50 O0 ~50 .00 ,50 1.00 1.50 2.00 2.50 3.QO
m ~ : FRO TOP BAC
LFT RGT
therefore both experimental sets of data are included. Isolines of equal pressure
on the top face (see Fig. 6 ) give some indication of the strong influence of the
two horizontal vortices generated at the two swept-back leading edges.
140
2.2. Turbulent flow over a square rib on the bottom of a plate channel
x
t.O
×
a~
.±
~'~t- ~ ~'~
z×
I I
142
Fig. 8. Instantaneous flowfield: top, in vertical middle section of channel; bottom, in a horizontal
plane at height Z= 1.12.
(a) (b)
z
i.o+*o
;.ii*o
i.+~.o
\ - I t ee ,o
16{,a
/o
,+[+o
; . ++*o
t+e.o
LeE'O
l.+e+o
0 o~-,
6 o(-,
o~-~
'c
: i o~ +o
, o(-o
t o ~(,o
-, ~¢.o
i l(.o +
i e(.0
+*e*o
~~--o °°°o°°:++,_/
l+(*o
1 .ilt*O ++ i+
I.IL-I
l.Olt-+
+o+-i
z (c) 2~L,o
(d)
, e~.o soE.o
!/
/
7°
i ++~,o
Le~+o
A
4QE-I
~eL
. . . . . . e ee.o ........ i
*~\ • =~ \\\
/I I + ~\ • .~ \
1
..... I~l :/
I'+t °°""~ °°9 \ e e "
..... ':1/
.... lb" ~.OE*O 20E~I
"g
~ 0(-I E*.GE-t | OE-I i OE*O 00E+I2 2 QE-L + OE+I $.0E-I |.OE-i I.OE*G
is XR (exp.) = 7.6 (measured in units of obstacle height from the middle of the
cross-section of the rib; see Fig. 7).
(2) If the LES results in the recirculation zone are compared with the ex-
periment (see Fig. 9) at the same relative position, X/XR(num.) and X/
XR (exp.) respectively, surprisingly good agreement is obtained for the r.m.s.
values of the velocity fluctuations and for the Reynolds stress (even in calcu-
lations with a significantly smaller number of grid points).
(3) There is some indication of a three-dimensional mean structure in the
flow field (e.g. in front of the obstacle ) which is confirmed by the experiment
[7]. Further details and additional results can be found in refs. 15 and 16.
3. G e n e r a l c o n c l u s i o n s and o u t l o o k on f u t u r e w o r k
Acknowledgments
T h i s w o r k is c u r r e n t l y s u p p o r t e d b y t h e G e r m a n R e s e a r c h Society ( D F G ) ,
P r i o r i t y R e s e a r c h P r o g r a m , P r o j e c t NO. Ro 4 9 7 / 5 ( R S m e r / W e n g l e ) . We also
gratefully a c k n o w l e d g e the s u p p o r t b y t h e c o m p u t i n g c e n t e r of the Universit~it
der B u n d e s w e h r M i i n c h e n a n d b y t h e L e i b n i z - C o m p u t i n g C e n t e r of the Ba-
v a r i a n A c a d e m y of Sciences.
References
1 C.W. Hirt, J.D. Ramshaw and L.R. Stein, Numerical simulation of three-dimensional flow
past bluff bodies, Comput. Methods Mech. Eng., 14 (1978) 93-124.
2 D. Vasilic-Melling, Three-dimensional turbulent flow past rectangular bluff bodies, Ph.D.
Thesis, Imperial College, London, 1977.
3 I.P. Castro and A.G. Robins, The flow around a surface-mounted cube in uniform and tur-
bulent streams, J. Fluid Mech., 79 (1977) 307-335.
4 D.A. Paterson and C.J. Apelt, Computation of wind flow over buildings, J. Wind Eng. Ind.
Aerodyn., 24 (1986) 193-213.
5 S. Murakami, A. Mochida and K. Hibi, Three-dimensional simulation of air flow around a
cubic model by means of large eddy simulation, J. Wind Eng. Ind. Aerodyn., 25 (1987) 291
305.
6 R. Kessler, M. Peric and G. Scheuerer, Solution error estimation in numerical prediction of
turbulent recirculating flow, CP 437, Vol. II, AGARD Validation of Computational Fluid
Dynamics, 1987.
7 G. Dimaczek, C. Tropea and A.B. Wang, Turbulent flow over two-dimensional, surface-
mounted obstacle: plane and axisymmetric geometries, Proc. 2nd Eur. Turbulence Conf. Ber-
lin, August 30-September 2, 1988, Springer, Berlin, 1989, pp. 114-121.
8 T. Kobayashi, M. Kano and T. Ishihara, Prediction of turbulent flow in two-dimensional
channel with turbulence promoters, Bull. JSME, 27 (231) ( 1984 ) 1893-1898; 28 (246) ( 1985 )
2948-2953.
9 B.E. Launder, A.P. Morse, W. Rodi and D.B. Spalding, Proc. NASA Langley Free Turbulent
Shear Flow Conf., Vol, 1, NASA SP 320, 1973.
10 F. Baetke, H. Werner and H. Wengle, Computation of turbulent flow around a cube on a
vector computer, Proc. 6th Symp. on Turbulent Shear Flows, Toulouse, September 7-9, 1987.
11 S.V. Patankar, Numerical Heat Transfer, Hemisphere, New York 1980.
12 U. Schumann, Subgrid scale model for finite difference simulations of turbulent flows in
plane channels and annuli, J. Comput. Phys., 18 (1975) 376-404.
13 R. Ruderich and H.H. Fernholz, An experimental investigation of turbulent shear flow with
separation, reverse flow, and reattachment, J. Fluid Mech., 163 (1986) 383.
14 I.P. Castro and A. Haque, The structure of a shear layer bounding a separation region, J.
Fluid Mech., 179 (1987) 439-468; 192 (1988) 577-595.
15 H. Werner and H. Wengle, Large-eddy simulation of turbulent flow over a square rib in a
channel, Proc. 2nd Eur. Turbulence Conf. Berlin, August 30-September 2, 1988, Springer,
Berlin, 1989, pp. 418-423.
147
16 H. Werner and H. Wengle, Large-eddy simulation of turbulent flow over a square rib in a
channel, Proc. 7th Symp. on Turbulent Shear Flows, Stanford University, August 21-23,
1989.