Large-Eddy Simulation of Thrbulent Flow Over

and Around a Cube in a Plate Channel

H.Wemer and H.Wengle1

Institut fUr Stromungsmechanik und Aerodynamik, LRT, WE7, Universitlit der Bundeswehr
Milnchen, D-80l4 Neubiberg, Germany

Abstract

The solution concept of large-eddy simulation (LES) has been applied to simulate turbulent flow
over and around a single cube mounted on the bottom of a plate channel for a Reynolds number of
50000 (based on the incoming mean bulk velocity and obstacle height). Here we present, as a first
part of the evaluation of the data, a few interesting views of the time-dependent fields and results for
the three-dimensional mean fields (velocity, vorticity, Reynolds stress, ens trophy, helicity). Having
engineering application in mind, this flow problem represents an ideal case, which is very well
suited for testing and validating numerical simulation techniques and turbulence models.

1. Introduction

The purpose of our work is to apply the solution concept of large-eddy simulation (LES) to
turbulent flow over and around flow obstacles with sharp edges and comers. The work is
motivated by the need for applications ofLES to engineering problems involving (a) high
Reynolds number flow, (b) three-dimensionality of the mean flow and (c) non-periodic
boundary conditions in the main flow direction. In this paper, we present results from a
numerical simulation of turbulent flow over and around a cube mounted on the bottom
of a plate channel. For the example presented here, we selected (a) a Reynolds number
of about 50000, based on mean bulk velocity and obstacle height (which is equal to the
channel half width), (b) a three-dimensional obstacle, creating a flow field without any
homogeneous direction, making it necessary to provide the statistics by time-averaging
only, and (c) a single obstacle, i.e. not allowing the use of periodic boundary conditions
in the main flow direction.
Earlier related work on LES of turbulent flow over a periodic arrangement of cubes
in a simulated atmospheric boundary layer has been published by Murakami, Mochida &
Hibi (1987), results for the flow over a periodic arrangement of square ribs in a channel
have been presented by Kobayashi, Kano & Ishihara, (1985), and results from our own
earlier work on flow over a single square rib on the bottom of a plate channel can be
found in Werner & Wengle (1988, 1989).

1 to whom correspondence should be sent

F. Durst et al. (eds.), Turbulent Shear Flows 8

© Springer-Verlag Berlin Heidelberg 1993
156 H.Wemer and H.Wengle

2. Mathematical Models, Solution Technique and Evaluation of the

Statistics

The governing equations describing the resolvable flow quantities (grid scale, GS) are
derived from the integral conservation equations for mass and momentum applied to
a finite grid volume Ll V = Llx * Lly * Llz following the 'volume balance method' of
Schumann (1975). From this it follows that GS quantities of the three velocity components
UCt (a = 1,2,3) are defined as averages over the corresponding surfaces, .1ACt (a =
1,2,3) of the grid volumes, LlV.
The subgrid scale (SGS) stresses, arising from the nonlinear convection terms, are
evaluated by the Smagorinsky-Lilly model (with Cl = 0.1) which relates the SGS stresses
to the GS velocity field via an eddy viscosity model. In grid volumes next to walls, we
used for the mixing length the smaller value of '" * Xa and 0.1 * (Llx * Lly * LlZ)1/3
respectively (xa is the distance normal to a wall).
The geometry of the computational domain and of the flow obstacle is evident from
figure 1. Measured in units of the reference height H (height of the cube) of the problem,
the dimensions of the computational domain are (X, Y , Z) = (10,7,2) and, for the
results presented here, we used (NX, NY, NZ) = (144,92,58) grid points.
The governing equations are solved numerically on a staggered and non-uniform grid
using second order finite-differencing in time and space (explicit leap-frog for time dis-
cretization, central differencing for convection terms and time-lagged diffusion terms).
The problem of pressure-velocity coupling is solved iteratively (point-by-point relax-
ation).
The direct results from LES are the time-dependent and three-dimensional fields for
the GS quantities of the three velocity components and the pressure. Figure 2a shows
a sample of the instantaneous flow field in a vertical plane through the center of the
cube. By time-averaging of the instantaneous flow field, the mean flow field (fig. 2b) is
obtained, and as soon as the mean velocity field has reached stable (Le. time-independent)
values, the fluctuating velocity field can be evaluated as the difference of the two fields
(fig. 2c). This procedure leads to the classical Reynolds-decomposition of a turbulent flow
field. Finally, from the fluctuating fields, the root-mean-square values, e.g. for velocity,
vorticity and pressure fluctuations can be calculated, as well as the Reynolds-stresses and
other statistics desired.

3. Inflow and Wall Boundary Conditions

At the inflow section, we used at each time step the instantaneous flow field of a LES
result of the corresponding (fully developed) channel flow (see figure 1). Boundary
conditions at horizontal and vertical walls were specified by assuming that at the grid
points (P) closest to the wall, (a) the instantaneous velocity components tangential to the
wall (up, vp) are in phase with the instantaneous wall shear stress components (Tub, T.b)
and (b) the instantaneous velocity distribution is assumed to follow the linear law-of-
 LES of turbulent flow over a cube 157

":I
tftHffW\O---~1 ~ ;]1
~ ~
.; 0

.,.c:
a::

- I .0 0.0 1.0 2 .0 3.
2 .0

Fig. 2. Reynolds-decomposition of flow field: ii =< ii > +u'

instantaneous flow (2a) = mean flow (2b) + fluctuating flow (2c) in
a vertical plane at Y=O through center of the cube

:!:

..
o
a::
::l
ou
c:;
"
C
.!'!
~

Fig. 1. Geometry of computational domain and inflow boundary

conditions
......
~ ?
.... ~
~
left: channel flow NX * NY * NZ =
54 * 92 * 58 grid points
.; ~
right: flow over cube NX * NY * NZ = 144 * 92 * 58 grid points
158 H.Werner and H.Wengle

the-wall u+ = z+ for z+ ~ 11.81, and for z+ = Zm > 11.81 it is continued by a

power-law description of the fonn u+ = A(Z+)B (with A = 8.3 and B = 1/7). The
velocity components tangential to a wall at the grid point next to the wall (up, vp) can
be related to the corresponding wall shear stress components by integrating the velocity
distribution over the height of the first grid element, and the resulting expression can be
resolved analytically for the wall shear stress component, i.e.

ITub I = 2JLlupi for lupl < _JL_A~ (la)

Llz - 2pLlz

1 -B 1+8 ( - JL )1+B +
ITub I = p [ -Ar=B
2 pLlz A
- ( -pLlz
1 +B JL)B
lupl
Yia 1
for lupl > _JL_A~ (lb)
2pLlz
Llz is the vertical width of the grid volume next to the wall, Zm is the upper boundary of
validity of the linear-law-of-the-wall. Though theoretically not fully satisfying, from an
application point of view this procedure offers the advantage that the averages < Tub >
and < Up > are not required (in particular in flow situations in which these variables
may be slowly varying in time), and numerical problems are avoided using the relation
Tub = Up < Tub > / < up > in reattachment regions. In consequence of the experimental
results from Ruderich & Fernholz (1986) we have abandoned the use of the logarithmic-
law-of-the-wall.

4. Discussion of Results

The results shown in this paper are made dimensionless using a reference height Lre! = H
(cube height) and a reference velocity UTe! = Ub (mean bulk velocity, see fig. 1). Note,
that in our nomenclature, Z is the coordinate nonnal to the walls of the plate channel.
The great advantage of LES is that it provides insight into the time-dependent and
three-dimensional large-scale structure of a turbulent flow field. An instantaneous view of
the velocity vectors projected onto a vertical plane through the center of the cube (fig. 2a)
exhibits an extremly complicated flow field. The interaction of different processes like
the development of a three-dimensional shear layer, the reattachment of flow on the
bottom plate behind the flow obstacle, the recirculation of highly turbulent flow and
its reentrainment into the free shear layer takes place within a spatial regime which
is significantly smaller compared to the case of turbulent flow over a ,square-rib, see
Werner & Wengle (1988, 1989): the mean recirculation length is about 2.0 for the flow
over a cube, and about 7.0 for the square-rib. Strong horizontal fluctuations in the lateral
direction close to the walls can be observed in the small recirculation regimes in front
and on top of the cube and in the large recirculation regime behind the flow obstacle. The
creation of strong horizontal fluctuations by splashing down of tongues of fluid material
can be seen in figure 3.
An ideal way of presenting time-dependent results would be a video film, but in a
printed medium the only way is to present a series of snapshots from flow regions of
 LES of turbulent flow over a cube 159

Fig. 3. Fluctuating flow field in a hor-

izontal plane close to the bottom plate
(Z = 0.01)

particular interest. Figure 4 shows ten snapshots of the flow in a vertical plane in front
of the cube. The time difference between the snapshots is 0.4 * Trej. One reference time
Tref is equivalent to the time a tracer particle needs to travel with bulk velocity Ub (see
fig. 1) over a distance of one obstacle height H. One or somtimes two large vortices
(rotating in the same direction) can be observed in front of the cube: they are pushed
toward the front face, and later pushed back again by a very violent stream of fluid
from above. In front of the horse-shoe vortex and close to the bottom plate a very calm
region in the flow field can be observed. Figure 5 shows snapshots close to the top face
of the cube, demonstrating the complicated separation process which not always starts
immediately at the sharp leading edge. Finally, figure 6 presents a few snapshots from
the three components of the time-dependent vorticity field (wx, wy , wz ) in the vertical
center plane, illuminating the recirculating region behind the cube which is bounded by
a free shear layer reattaching again on the bottom plate (the mean reattachment length is
about 2.25 measured from the center of the cube).
The mean structure of the flow field (e.g. mean velocity, mean vorticity, second-order
statistics, mean ens trophy, mean helicity) can be provided by time-averaging only. For
the results shown here, we started averaging for the first-order statistics after 60 reference
times Tref; the first-order statistics was sufficiently stable after about 100 reference times
(taking one sample at every 40th time step), samples for the second-order statistics have
been collected over the latest 140 reference times. Figure 1 shows vertical profiles of the
mean U-velocity component; the profiles are quite smooth, indicating that a sufficient
number of samples has been taken for the first-order statistics. From distributions of
the mean U-velocity component close to the bottom plate (Z = 0.01 =location of first
vertical grid point) and close to the top face of the cube (Z = 1.01), the different
mean recirculation lenghts can be approximately determined to XF = 0.9, XR 1 = 1.75,
XR2 = 0.2 and XT = 0.75 (fig. 7a,b).
In a vertical cut normal to the main flow direction at a dimensionless position X =
0.88 behind the cube (fig. 7d), two pairs of counter-rotating mean vortices can be observed
(one pair very close to the side walls of the cube, and the other one at Y = +1.5 and
Y = -1.5 from the center of the cube). In that vertical plane, the lateral shape of the
mean shear layer also becomes visible (fig. 7d).
 160 H.Wemer and H.Wengle

0.6
0.4-
~~"'7=;r_
_~-?-)rn7'=;r
~~"':J
~7""J
i'
,.
~
:1"......
.. ,
..... l=a=r
...

'iI'''i
0.6r::::::::::::::::::::::::~:=~~~~~
0.4
';!It:1r7S''''~~3P'iI''9''i
~;rsP"~1: J' ....
)P)rnlS'=.=-1rP'7"~~)
>?=:r:;;r~;r9'~>iI?O!
r _ ...-" •••.•.

"1
_.,I~,,_'I'

N N

0.2 0.2

-\.O
o~~~~~~
-I.S -1.0 -O.S
X X
•W TIMEPH • O.233753E+OJ ,W O.235752E·"3

0.6E::===~~
.... ~~:=:=

~--=-~.,.;r;Jlr'!l"";,-""
,...,.w~

;;;r?t" ... ,... .. :a-;r7 ~;

o.~ 0.4 :3*:ir;'7:;;r-:;m7;n; ..

N N

-\.O -O.S
X
,II TtMEPH_ 0.234153E·03 ,w TlMEPHI 0.236152£+03

0'6§~~"
...., ..
;.0"0,............ ,
;.s:;!lm:;r- "';'-!lI":r3Fr';lrll
.,.. S!m =-:r~
S'3P!"
J Cwo
!: ... " .. /"""'\" .... - - - . , . .
I
"'7'
)r"''3I",.;r~3 \,.,~.,.,
0.4 $I" '$.!!""~;t=;:;:,; f' •• ~Hu/"
:;. )' 3'" 7'":;;r-;or ;r ;II Jr !II " • ! : ~ ...... .

N ;r ~ 'V . ,
31" .'P.:~!" iI ; ... " ~.....
' ...............

0.2 :- :;~: :.::: ::~~~~\\;:~

;; ! :::.::..... ::! ~ ~~~...
~:":'~~''':
::::.:::!
I,
~;---I,T'.".:.,:),~~ , ~8
~::-=;~::-~~~:!""~".\.'~~~ ....
o
-) ,0 -o.S -1.5 -1.0 -0.5
X X
,II TIHEPHI O.234553E+03 ,w TlHEPHI 0.23655IE+03

O. 6 j=J..........., . . . . , . - - -.......--~~"'P?";"A
~~..- .
... ;r",,:;,-;t:.,.1"""'JSi' ;.~__",_,. ...... ,
~~ ... ;e:;r?"?' ~ i' 'i' p ~ ... ,_~",,,,,~ .. .
0.4 ,.;-,.:a-;;.n:J=;:¥~;~~_,,#,_~ .... .
!Om;r 3E;r-.,
;mi' i'"P J iC " , _ , , , , _ , , ••

N --""'~~) " rp" ........~~.......

-r-",,"~;Jr i" ;!C-=9 . . . . . .~ .... ",~~

:::;;::~-=-~ ::;::.::~~..:~~~~\
. '.
O. 2
~=:::::~';-~~- :~~~~'":',\:.tt~ 9
::::::::-:::::;'?~':;; Ii;::::: y
0~~~~~~~~~~L2 o :-- '7 =-~'.. /.,;_. . ~ - :.. ;.: -O.S
-) ,S -1,0 -O,S -I .S -1.0
X X
,W TIMEPH. O.234952E+03 ,II TlHEPHI 0.23695IE+03

7e ;t: ~

~~\¥~~.,..._

~
. . ::=:.-jir
.........,............_ _ __

r;r-3'"er'!!r)r;r-?,.
'P
0.6 r-::::::======~~
I-::;r
~
;r )In:r .,.~; ;r;JmJ
,.. ,. .. :r:Jn;r~a::,., .. ';'-_/'."
1"" "~-.'" .
;r ,... >~JIn~"'''' J ;0.'" - .. , _. . . . . . . ,
.

0.4- ?":;e;..." l3E1!P";;e:;ri; , .. o.• '$I: :;.sr;r",,"3' ;m;rg P:;'~_'~II'u,,,

)Ie 7r .... 3" it' "?:? .. ~ 'i • ~ =-!7'Jt- )ln3i'"¥"_ ....................... " ••••
N N
~
. . ;,r 7"P''J- . . .
!!r ,. r _%~
,
3!:'" ~\\ . . . _ ....
_____, ........,\\\ •.
1 ••

0,2 O.2-;11-~""J'/llffll'
::::::::.;--=-~~ ... ~.--;;:~~,\: ;:\~~~:.
---~"'--"nl-\\"'''''10
;: 'f ~ ~--("~. '.....~;=. ~.~~~~~=);::
O~~~~~~~~~~~~~~
-\.o -1.5 -1.0 -O.S
X X
.W TIHEPHI O,2J5J52f.·OJ ,II TIHEPHt 0.2373SIE"'OJ

Fig. 4. Snapshots of instantaneous velocity field ( Lit = 0,4 * Tref ). in front of the cube. in a
vertical plane through the centerline of the cube (Y =0)
 LES of turbulent flow over a cube 161

:::: ::
-------
:::::::::-:-:-::
1 .•
........---~--------

N 1.2 N 1.2

1.0 1 .0

-0.5 0.0 0.5 -0.5 0.0

X X
0.20gI671;+03 .W TIHEPH' 0.210166E+03
.W TlHEPHI

--------
------...---
1 .• 1 .•

N 1.2 N 1.2
,.. ---------
1 .0 1 .0

-0.5 0.0 0.5 -0.5 0.0 0.5

X X
.W TIHEPHI O. 209366E +03 .W TIHEPHI 0.210366E+03

--------
1.4
-----------
I..
:;..----r
------
___________ - - -
~----....------- ..-

N 1 .2 N 1.2

1.0 1.0

-0.5 0.0 0.5 -0.5 0.0 0.5

X X
.W TIHEPHI 0.209566£ +OJ .W TIHEPHI 0.210566E+03

1 .•
-------
---------
1 .•

,.....------_-....---
--------
_ _.-JI!:r".
~----

N 1.2 -----...
............... N 1.2

1.0 1 .0

-0.5 0.0 0.5 -0.5 0.0 0.5

X X
.W TINEPHI O. 209766E +03 .11 T1HEPH I 0.210766E+03

...........-----
---- --------
----, , ,----------
1.4 1.4
-"'----...---------- ----------
~-_/
.---'.......-
/ -........... -----
--
N 1.2 N 1.2

~~
..~~---..
........... --:-;;-:--:
, ...
:
1.0 fif-£:~ --~:; -"""==T-: ; ~ : : : 1.0
..
/1111

-0.5 0.0 0.5 -0.5 0.0 0.5

X X
.W TtHEPHI O.209966E+03 .W TIMEPHI 0.2' 0966E .. OJ

Fig. 5. Snapshots of instantaneous velocity field ( Llt = 0.2 * Tre/ ), at the top face of the cube, in
a vertical plane through the centerline of the cube (Y = 0)
162 H.Werner and H.Wengle

z z z

o
~

o
m
m

o
w

o
z z z
o
'" .0
~

o o o
~ ~ ~
o
o ~ ~
w
m m m
m m m
~ ~ ~

o o o
w

0
~," 0
W

z z o
z
o o
o o
'" '" '"
>-'
D ~-

~. ~::. ,-: O·
- Y, ~ " <"°0

o ,,'CS
m
m
~
,-,
o
w " w

'0
'''.~
.', ~ ' ..
.... ,.,'
_"~I

Fig. 6. Snapshots of instantaneous vorticity field ( Llt = 0.4 * Trej ), in the recirculation regime
behind the cube in a vertical plane through the centerline (Y = 0), dashed lines: negative values,
fuUtines: positive values, column on the left: wx, column in the middle: Wy, column on the right:
wz.
 LES of turbulent flow over a cube 163

1.5.[+0

1.01£1>0

• . • t- I I~""'''''_''''' __..
•.••.• t-------------------~~~--_j----~--~t_------_=~~~~------------------------_1

<U>

,..... 1r-----------------------------''---i---i-------------------------------------------,

.",-
-----+--- z - 1.010o!:..fCl
I ..... ir<..............................._--..._
S.U-I

•.•• ~ t---------------------------~~--~~T-------------------------------------------_1
-5.('(- 1

Fig. 7. Detennination of mean recirculation lengths, Fig. 7a,b: from profiles of mean velocity at
heights Z = 0.01 and Z = 1.01 at Y = 0, Fig. 7c: mean velocity vectors at Z = 0,0}, Fig. 7d:
mean velocity field (directions of velocity vectors) at X = 0.88Fig. 7e: surface flow pattern from
the experiment of Martinuzzi, see Larousse, Martinuzzi & Tropea (1991)
164 H. Werner and H. Wengle
Z
2.'(,' l r - - - r --."...;-----,

S.OIl:-1
<ti>
O. Of .O. I.'.•\;""=~,.';:====:=1.7,===,"':!.o _I .0 2.0 -1. 0 0.0 1.' 2.'
L
I_----- - - J
•

- I .0 0.0 .1 .0 2.0 3.0 4 .0 5.0 6.0

r_------ .- --
l.. '----
J ~

--- --,
z •
'.Of_O

1.5[·0

~
ptP?-!!- 0 oo~
1.0f·O

S .a! -l
- <
I I
u w >
0.0[-0
-2.0[-2 0 . 0 '.Z.0(-2 4.0(-2 e .O(-2 -2.0f-2 0.10 VO(-, '.0[ "" 1,0[- 2

Fig. 8. Comparison with experiment

above: vertical profiles of Mean V-velocity component
below: vertical profiles of Reynolds shear stress - < UW >
000: experimental data from Martinuzzi, see Larousse, Martinuzzi & Tropea (1991)

In the meantime. experimental data are available for the same flow conditions from
Larousse. Matinuzzi & Tropea (1991). Figure 7e shows the experimental surface flow
pattern which can be compared with the computed mean velocity field close to the bottom
plate (figure 7c). In addition, figure 8 presents a comparison of measured and computed
I
vertical mean velocity profiles and profiles of the Reynolds shear stress - < u w >.
I

From that comparison it can be concluded that the next large-eddy simulation of turbulent
flow over a cube needs an improved spatial resolution immediately above the top face of
the cube, and also close to the two side faces of the cube. From the comparison of the
measured and computed shear stress profiles (figure 8), and from the fact that the profiles
of the second-order statistics are still not sufficiently smooth in certain regions of space
(in contrast to those of the first-order statistics), it can be concluded that the second-order
statistics is very slowly converging in particular flow regions. One of these critical flow
 LES of turbulent flow over a cube 165

b
Fig. 9. Mean vorticity field
Fig. 9a: right half: vorticity vectors projected onto vertical planes at X = -0.3, +1.0, +2.5, left half:
isolines of total mean ens trophy < wi + w~ + WE >
Fig. 9b: isosurface of total mean enstrophy (isosurface cut open at Z = 1.4)

regions can be identified immediately behind the flow obstacle (at about X = 1.0), and
another one is the reattaclunent region (at about X = 2.0).
The evaluation of the mean vorticity field in figure 9 gives an indication of the
geometry of the mean shear layer behind the cube. Figure 9a shows, on the right hand
side, the mean vorticity vectors projected onto vertical planes normal to the main flow
direction (the length of a vorticity vector is proportional to the strength of the local
mean vorticity) and, on the left hand side, isolines of the mean total enstrophy. An
interesting perspective view of the shape of the 3D mean recirculation zone is given in
fig. 9b showing an isosurface of the mean total enstrophy < Wx 2 + Wy 2 + W, 2 > (for an
arbitrarily chosen value of 20.0).
w
Finally, we evaluated the so-called helicity v· of the flow field. Fig. lOa shows a
perspective view of a mean helicity iso-surface for a value of 0.25. If the instantaneous
166 H. Werner and H. Wengle

helicity density is nonnalized as cos() = v· w/lVllwl, it can be interpreted as the cosine

of the angle between velocity vector and vorticity vector. For example, fig. lOb gives a
perspective view of the instantaneous locations of cos() = 0.90«() = 25°). Regions of
w
small helicity values are equivalent to regions of large values of v x (the non-linear
tenn of the Navier-Stokes equation in rotation fonn), in regions of small angles between
w
v and it is reasonable to expect that the local structure will be helical (an ideal case
w
would be cos() = 1.0, Le. v and are aligned).

/
Fig. 10. Isosurfaces of helicity
w
Fig. lOa: mean helicity < yo. >
Fig. lOb: instantaneous helicity density cosO = yo . w/lVllwl = ±O.9
 LES of turbulent flow over a cube 167

5. Conclusions

The most valuable result from a large-eddy simulation is the time-dependent and three-
dimensional large-scale structure of a turbulent flow field, and a lot of additional effort
seems to be necessary to present such data to the research community in a proper way.
[t is always very helpful and, at least for some time in the future, it will be necessary to
compare LES data with corresponding experimental data for the first-order and, most im-
portant, for the second-order statistics. Then, the two methods will be able to supplement
each other, and LES will develop to a valuable tool for a sufficiently accurate prediction
of turbulent flows of engineering interest. For a flow case without any homogeneous
direction, as presented in this paper, about 300 CPU hours on a CRAY/y-MP must be
spent to provide sufficiently stable second-order statistics, and the number of grid points
used here should be considered to be a minimum to reach satisfying agreement with the
experiment. The case of turbulent flow over a cube in a plate channel at high Reynolds
number represents a flow problem which is very well suited for testing and validating
numerical simulation techniques and turbulence models.

Acknowledgements

This work is currently supported by the German Research Society, Priority Research
Program, Project No. Ro 497/5 (Rl5mer/Wengle). We also gratefully acknowledge the
support by the computing center of the Universitat der Bundeswehr Miinchen, by the
Leibniz Computing Center of tlle Bavarian Academy of Sciences, and by Convex Com-
puter GmbH, GS Bayem.

References

Baetke, F., Werner, H. Wengle, H., (1987): "Computation of turbulent flow around a cube on a vector
computer", in Peoc.6th Symposium on Turbulent Shear Rows, Sept 7-9, Toulouse, France.
Kobayashi,T., Kano,M., Ishihara,T. (1985): "Prediction of turbulent flow in two-dimensional channel
with turbulence promoters", Bull. JSME,Vol. 28, No. 246, 2948-2953.
Larousse, A., Martinuzzi, R., Tropea, C. (1991): "Flow around surface-mounted three-dimensional
obstacles", in Proc.8th Symposium on Turbulent Shear Flows, Sept 9-11, Technical University
of Munich, see also this volume of selected papers from TSF8.
Murakami, S., Mochida, A., Hibi, K. (1987): "Three-dimensional numerical simulation of air flow
around a cubic model by means of large eddy simulation", J.Wind Engng. and Ind. Aerody-
namics, 25, 291-305.
Ruderich, R., Fernholz, H.H. (1986): "An experimental investigation of a turbulent shear flow with
separation, reverse flow, and reattachment", J.Ruid Mech. 163, 283-322.
Schumann, U., (1975): "Subgrid scale model for finite difference simulations of turbulent flows in
plane channels and annuli", J.Comp.Phys.18, 376-404.
168 H.Werner and H.Wengle

Werner, H., Wengle, H. (1988, 1989): "Large-eddy simulation of turbulent flow over a square rib
in a channel", in Proc.2nd European Turbulence Conference, August 30 - September 2, 1988,
Berlin, and in: Advances in Turbulence 2, Springer., Proc. 7th Symposium on Turbulent Shear
Flows, August 21-23, 1989, Stanford University, USA.