You are on page 1of 9

Journal of Wind Engineering and Industrial Aerodynamics, 46 & 47 (1993) 755-763

Elsevier

755

Numerical and experimental modelling of the three-dimensional


turbulent wind flow through an urban square
A. G A D I L H E
Ecole d'Architecture de Nantes - CERMA, Rue Massenet, 44300 Nantes, France
L. JANVIER
Electricit6 de France -DER/LNH - 6 Quai Watier, 78401 CMtou Cedex, France
G. B A R N A U D
C.S.T.B., Rue Henri Picherit, 44300 Nantes, France

Abstract
Urban de sign and planning may require detailed data on the wind flow around buildings, within
the streets and through the squares to ensure the safety and the comfort of pedestrian. Our aim was
to compare a numerical prediction of wind flow through a complex and realistic building site with
an analogic test in a boundary layer wind tunnel. The chosen site was a semi-circular square
located in downtown Nantes, France. Five streets lead to the square which is made up of four
buildings with double sloping roofs and a church with a dome.
The wind flow within the streets and through the square was calculated using a standard k-e
model of turbulence. The spatial finite elements discretization consisted of about 60, 000 nodes.
The results of the numerical simulation were compared with wind tunnel measurements made in
the CSTB large facilities. The wind velocities were in fairly good agreement. Nevertheless, the
complete experimental velocity field was not measured, especially in the wakes behind the
buildings or in the recirculating zones on the square, because of the limits of the triple hot wire
probe used. Therefore, the use of an other measuring technique, laser anemometry, is now being
considered.

1. INTRODUCTION
Town planners are interested in air flow through urban spaces to make pedestrian comfort
acceptable. In order to investigate the wind flow pattern through a planned urban area they can
carry out a wind tunnel test. If a well-designed mock-up is placed in a boundary layer wind tunnel,
an evaluation of this flow will be obtained, but parametric tests and optimization studies cannot
be easily performed. That is why the development of a numerical approach, including a turbulent
flow model, is proposed here. In this paper, to evaluate the validity of this approach, a numerical
prediction of the wind flow through a complex and realistic building site is compared with an
analogous test in a boundary layer wind tunnel.
0167-6105/93/$06.00 1993 - Elsevier Science Publishers B.V. All rights reserved.

756
2. THE EXPERIMENTAL SITE
The site of our experiment is located in downtown Nantes near the Loire River (figure 1). This
semi-circular square is a little complex with many courtyards, various roof slope angles, different
building height, a church with a dome, and many streets. In order to study the wind flow through
the streets and the square, the geometry of these elements was simplified and it was also decided
to study a symmetrical configuration to discretize only one half of the site (figure 2). Thus the
wind is supposed to blow parallel to the symmetry axis. It is a suburban wind whose characteristics are those presented by Counihan [1].

Figure 1 : The actual site of the Sanitat Square Figure 2 : The modelled site

3. NUMERICAL SIMULATION
3.1 Wind flow modelling
The equations governing the wind flow through urban spaces are based on the principles of
conservation of mass and momentum. The wind flow is assumed to be three-dimensional,
incompressible, turbulent and steady. It is thus possible to derive the Navier-Stokes equations and
the continuity equation from these principles. The average and fluctuating velocity components
at each point, the so-called Reynolds decomposition, can be further defined. In order to evaluate
the Reynolds stress tensor, the k-e model was chosen. The modelling of the turbulence
phenomena was completed by wall boundary conditions which result from an assumption of the
logarithmic velocity profile near wails. The values of numerical constants were assumed to be the
classical ones. The set of six equations was solved using a finite elements discretization method,
which is better able to model such a complex geometry of urban spaces, although this method
requires more storage space and larger computation times than a finite differences one. The N3S
code was used to solve the equations [2]. The space discretization method uses tetrahedra with
either parabolic or linear approximation for the velocity and linear approximation for the pressure
on each element. The time discretization is based on a fractional step method. The advection step

757
is solved by a characteristic method, the diffusion step is solved using a preconditioned
conjugated gradient algorithm and the Stokes problem by an Uzawa algorithm [3].
For the average flow the equations are as follows :
m
aUi

= 0

3xi
U--;DU'~= - l a p
axj
p axj

=--VT

UiUj

VT =

auiu j

0xj

' +-~i)

I-V-0xi0x j

ij

C v --=E

xjj

Uj 0xi

--

a~

0xj

t,':>, axj )

-- UiU j - -

0xj

E-~

-- E

~-- a~

with o~ = 0.42 ; Cv= 0.09 ; C 1 = 1.44

; C2=

1.92 ; o k = 1 ; o = 1.30

3.2 Numerical simulation of the wind flow through the square


The computational flow domain and the volumic mesh is the air located between the buildings
and a parallelepipedic box representing the wind tunnel boundaries. Because of the direction of
the incoming wind and the geometry of the
modelised square, the flow field was assumed
to be symmetrical in the longitudinal direction
and one half of the flow field was calculated.
This symmetry was experimentally checked.
The computational domain should be large
enough for the boundary conditions on the
upper and side walls to have little influence on
the predicted flow field, and to avoid any interaction between the recirculation zones and the
boundaries. The computational domain has a
Figure 3 : Surface mesh of the domain
downstream length of 399m, lateral width of
200m and a vertical height of 65m, compared
with the dome which is 40m high. The volumic meshing was generated with SIMAIL TM [4]. The
overall mesh contained 38,895 tetrahedra and 57,699 velocity nodes (figure 3).

758

The vertical profile of the wind velocity was set


to follow the relation (figure 4) :

5(]

20
10

with ~ = 0.20
U ~ = lrn/s
Z 0 = 10m

0
0,50

0,75

1,00

1,25

1,50
U/U30

Figure 4 : Vertical wind velocity profiles


As the experiment had not yet been set up at the
time of the beginning of the N3S computation, the
inlet boundary conditions for k and e were specified
via vertical profiles like those encountered in any
boundary layer.

~_

6O
513

*2

40

.~,

3I

*3

~=u

20
/Z

10.

: N~R
where 1(was the von Karman's constant ( = 0.4) and
u" the friction velocity. The friction velocity was
calculated with the Reichardt law.

0,0

t
0,1

i
0,2

I
0,3

I
0,4

0,5
I

Figure 5 : Vertical turbulence intensity


profiles
In a forthcoming simulation, we shall take into
account the measurements to prescribe the turbulence intensity at the inlet (figure 5). Free slip
conditions are prescribed for the velocity on the
side and upper walls.
3.3 Results [5]
Before discussing the results it may be interesting to recall the computational time and the memory
space required for the numerical simulation. The
time required to achieve a converged solution (figure 6) was about 15 hours CPU on a CRAY-YMP
computer and the storage allocation required 15
Megawords.

dF

T .... ~..... F

>
.E
~: ~
0

100 200 300 400 500


Number of iterations

Figure 6: Evolution of the wind velocity


versus the number of iterations

759
By means of the GRAFN3S postprocessor, the velocity field in different planes or surfaces
was visualized (figure 7) and some particle trajectories were calculated to analyse the flow
pattern. Thus, the location of the stagnation points on the windward sides of the first building and
of the church were found, and the recirculation zones in the front of the church and the wakes
behind the church and the second building were investigated.

"

"

.'-,

.t

Q h,~d'l

." ."
, ..,

'

"-

'

","

--

.
I?"~L~J
~.V---------.-,-/lll~."Ic-"T.,,.v"

.
..'//

:. .-'J: >,,,- - t.
4"."

J"~$.P~,,~.,

".~

#a "

"

.
"%i'6 ' r l l ~

__-',II

.
,

.
__'~

. ~

I,.
"

" -/

"

--

.
. . . . .

I"P

-,

fJPF~

---

.
"-

...

~1.,/.,f-...,~,.,'-...

IIHI I I

:::::

"--

".'.,

.i

IpIpl~l

|lI ........

Figure 7 : Velocity vectors in the plane z = 1.5m

4. COMPARISON WITH WIND TUNNEL EXPERIMENT


4.1 Wind tunnel experiment [6]
The experiments used for comparison were performed by the C.S.T.B. in a boundary layer
tunnel [7]. The test section was 4m wide and 2m high. A suburban wind was reproduced using
various roughnesses in the upwind part of the tunnel. A model was built in roughmatt on a scale
of 1/100 . Velocity measurements were performed with a triple hot wire anemometer at 60 points
located in the vicinity of the areas where the numerical results had been investigated (figure 8).
The sensors of the constant-temperature anemometer used were thin wires of tungsten about 51arn
in diameter and the precision of the probe was 2% as far as it had been placed within a solid angle
of 15 relatively to the wind direction. The Reynolds numbers, V D h , were about 106 in the wind
v
tunnel experiment (D h was equal to the hydraulic diameter of the wind tunnel cross section) and
9 10 6 in the numerical simulation (D hwas equal to the hydraulic diameter of the numerical domain
cross section). It must be said that the effects of the Reynolds number are relatively small for sharp
edges buildings.

760

47

39

43

36 3738

41

42

40

34[

1920 tl
t_
Church
Bloc B

1617 118
31

30

15
14
13

Street 2

33 k

ll12
lO
k

[--

56

57 411t 49

54

58 5150 59

\52

32
55

53

Bloc A

78

27

28

~5 6

1 2

25

24

23

Figure 8 : Location of the measurement points

22

26

761
4.2. Comparison between the numerical and measurement results [8]
4.2.1 Comtmrison of the wind velocity at the 1.5m olane
In this paper the flow is analysed with a focus on the symmetry (figure 9). An exhaustive
comparison is presented in [8]. On the symmetry axis it can seen that the transversal and vertical
velocities are close to zero both in the wind tunnel experiment and in the numerical simulation.
In street nl ( -100m < Y < -50m) the wind velocity is quite constant according to the numerical
simulation but, according to the measurements, it decreases as Y increases. At the inlet of this
street, the agreement between measurement and numerical results is not good : it is better at its
outlet. In the square, the church and a backward flow make the wind velocity decrease, although
it must be said that this point is more apparent on the numerical results than on the measurements.

~1,0

~1'0 t

0,6

0'6 I

0,2

0,2~...~
I

-OA 5-100-85 -70 -55 -40 -25y-tO)


-0,6
-1,0

N3S
CSTB
Variation of U with Y

_0,~I5-100-85--70 -55 -40 - 2 ~ O )

-0'6 f
-1,0

-'~@B
Variation of Ux with Y

1006i
0,2~

_0:f5-10--40
5
0'6I
- 1,0 ~

3,

N3S
o CSTB
Variation of Uy with Y

_0,~I5-100-85--70-5'-40-~5y2~O)

-0'6 f
- 1,0

N3S
o CSTB
Variation of Uz with Y

Figure 9 : Wind velocity at the 1.5m plane on the symmetry axis

762
4.2.2 Comp0xison of the wind velocity gradints
At some points chosen in the recirculating zones we measured some vertical gradients of the
velocity. Here, we will discuss three gradients values, two on the square and one behind the
church, all of them are analysed in [8]. On the square, in front of the church (figure 10) the
computed vertical velocity component is negative (downward motion), while the measured one
is positive. At a height of 1.5m a backward motion was numerically predicted but was not
measured. However the wind intensities are in good agreement.

20

~-" 20
N
N 15 f

15

/,:
/

10

,/

10

N3S
CSTB
I

-1,0

!1 I
,
I

-0,6

-0,2

0,2

N3S
CSTB
I

0,6

1,0
U
Varition of the wind velocity U
with the height - Point 48

:0

,
I

I~

-1,0 -0,6 -0,2' 0,2

1,0
Ux
Variation of the wind velocity Ux
with the height - Point 48

0,6

"~'20 7
N
15

//
i

lo

1,

f,

CSTB
N3s /#i
I

-1,0 -0,6 -0,2

CSTB
I

0,2

0,6

1,0
Uy
Variation of the wind velocity Uy
with the height - Point 48

-1,0-0,6-0,2

0,6

1,0
Uz
Variation of the wind velocity Uz
with the height - Point 48

Figure 10 : Wind velocity gradient in the square near the church

0,2

763

5. CONCLUSION
Numerical simulation for the investigation of the wind flow pattern through urban space has
already enabled to provide results which are in an acceptable agreement with wind tunnel
measurements. However, because of the approximation of the turbulence intensity inlet conditions, the simulation presented here should be run again to be as close as possible to the wind
tunnel conditions. To account for the greatest numerical to experimental discrepancies encountered in the present study, in the wakes and recirculating zones, it can be argued that the wind
tunnel measurements themselves may not be completely reliable owing to measurement
difficulties. That is why we are considering the improvement of the hot wire technique or
resorting to laser anemometry. The comparison we have described here represents an early stage
in what should ultimately constitute a complete validation : further comparative studies are
required.

6. ACKNOWLEDGEMENTS
This work was supported by funds from the Architectural Research Council (Minist~re de
l'Equipement, du Logement et des Transports - Direction de l'Architecture et de l'Urbanisme).
This study was carried out as a part of the co-operative work being done by L.N.H./E.D.F.,
C.S.T.B. which realized the wind tunnel experiment and C.E.R.M.A. The authors wish to tank
P. SIMON who ran the numerical simulations during his training course.

7. REFERENCES
1 Counihan J., Adiabatic atmospheric boundary layers : A review and analysis of data from the
period 1880-1972, Atmospheric Environment, Vol. 9, p.871-905
2 Chabard J.P., Projet N3S de mrcanique des fluides : Volume 1 - Manuel throrique de la version
3.0, Rapport EDF-DER-LNH, n HE-41/91.30, 1991
3 Labadie G., Lasbleiz P., Quelques mrthodes de rrsolution du probl~me de Stokes en 616ments
finis., Rapport EDF-DER-LNH, n HE-41/83.01
4 SIMULOG, Documentation technique du logiciel de maillage SIMAIL, Mars 1991
5 Simon P., Simulation numrrique d'un 6coulement de vent en milieu urbain., Rapport de stage
de D.E.A., Octobre 1991, E.N.S.M.
6 Barnaud G., Simulation arrodynamique de la place du Sanitat h NANTES, Report EN-ECA
91.22C, 1991
7 Bamaud G., Solliec C., La soufflerie h atmosphrrique du C.S.T.B., Report EN-ECA
92.3L,1992
8 Gadilhe A., Ecoulement du vent sur la place du Sanitat ~tNantes : Comparaison des simulations
numrrique et analogique. Rapport interne CERMA n98, Avril 1992

You might also like