Buridino and Encironmenr, Vol. 30, No. 2, DP. 267-276.

1995
Copyright 0 1995 E&&r Science Ltd
Pergamon Pnnted m Great Bntam. All riehts reserved

0360-1323(94)00038-7

Numerical Evaluation of Wind Pressures

on Flat Roofs with the I&C Model

T. STATHOPOULOS*
Y. S. ZHOU*

The results of an investigation regarding the numerical evaluation of wind pressures on flat roofs
by using the Reynolds averaged Navier-Stokes equations and the standard k--E turbulence model
are presented. A low and a taller building model are considered and wind blowing from different
directions is assumed. Data are presentedfor roof edges and corners. Experimental measurements
are also carried out in a boundary layer wind tunnel and results are compared with the computed
data. Over most of the roof surface area, the comparisons are goodfor normal wind conditions and
the low building model. However, the present approach does not seem to yield adequate rep-
resentations of available e.rperimental data in the high vorticity region near the windward roof
edges of the building, particularly for oblique wind conditions. The influence of proximity of grid
points to the roqf surface is examined in detail.

1. INTRODUCTION fications with respect to the solid boundary conditions of

turbulence properties k and E and the tangential flow
THE wind environment around buildings and the wind
velocities to account for the viscous effect in the immedi-
pressures on building surfaces are usually studied by
ate proximity of the solid boundaries were adopted by
scale-model experiments in wind tunnels and confirmed,
Stathopoulos and Baskaran [8]. This led to a better pre-
in some limited cases, by full-scale measurements. Such
diction of the mean pressures on the roof.
experiments are often expensive and time-consuming.
Murakami [9] studied this problem by adopting the
Computational wind engineering as a new branch of
RANS equations for unsteady flow and the standard
Computational Fluid Dynamics (CFD) has been
k--E model. Numerical results from two different grid
developed recently to evaluate the interaction between
arrangements were compared with experimental data for
wind and buildings numerically. Such investigations
a cubic building model. The numerical mean pressure
include the studies by Hanson et al. [l] and Summers et
coefficients over the top of a cube from a fine grid system
al. [2] for wind blowing normal to a building or to a two-
agreed well with experimental data. However, such suc-
building configuration.
tions obtained from a coarse grid arrangement with dis-
Vasilic-Melling [3] has perhaps pioneered the numeri-
tance from the first gridline to roof surface equal to 1.7
cal simulation of three-dimensional wind flow around a
cm (compared with the building height which is 20 cm)
cube by using the k--E turbulence model together with the
showed apparent underestimation of the experimental
Reynolds averaged Navier-Stokes (RANS) equations.
data. Wind blowing normal to a cubic building jn a group
However, the predicted mean pressure on the top surface
of uniformly distributed buildings of the same geometry
of the cube did not agree well with the experimental
was also simulated by Murakami et al. [lo] by using
results by Castro and Robins [4] which was attributed
the large eddy simulation (LES) but the computed roof
to the coarse grid arrangement used. Recent studies by
pressures did not show any apparent improvement com-
Paterson and Apelt IS,61 have also addressed the numeri-
pared with those from the k--E model.
cal evaluation of wind pressure on the roof of a cubic
Baetke et al. [ 1 I] evaluated pressures numerically on
building. Results compared fairly well with the exper-
the surface of a building with wind blowing at an oblique
imental data by Castro and Robins [4] except for the
direction (45”). These data taken on the central line of
region near the windward roof edge where suctions were
the roof have only been presented for the case of uniform
overestimated by the computation.
flow conditions. More recently, Selvam and Paterson [ 121
Baskaran and Stathopoulos f7] experimented with a
simulated the three-dimensional wind flow around a low
modified k--E model including curvature effect of stream-
rise building for which both full-scale and wind tunnel
lines to better predict the wind conditions around a rec-
data were available. They solved the RANS equations
tangular building. However, the pressure coefficients on
and the k--E model by applying both staggered and non-
the roof did not show much difference from the results
staggered grid arrangements in their calculation. Results
obtained with the standard k--E model. Further modi-
were presented along the central line of the roof for
normal wind direction. Data from the staggered grid
*Centre for Building Studies, Concordia University, arrangement case agreed fairly well with the wind tunnel
Montreal, Quebec, Canada H3G lM8. results by Surry [ 131, whereas the numerical results from
268 T. Stathopoulos and Y. S. Zhou

0.2<WL-zO.4

f--x l.D<WR11.5

- Selvem and Paterson (1991)

- Murakaml *t al (1990)
--. Baskann PM Stath0pou109 (19SD)
----~-Ssetkeetal(l9m)
- All Qastm *t 111(2-D) (1991)
- - - Patwso” B”d Apolt (1990)
1.4 - 0 surv(1989)
0 Castro and Robins (1977)
LI SbaIhopouIW and Dumltfescu-BruloM (1999)
+ StathopOulos et (11,198,)

X/L
Fig. 1. Computed and experimentally measured pressure coefficients in previous studies.

the non-staggered grid failed to predict the pressure on numerical values. The limitations of the k--E model in
the roof. An unsuccessful attempt to compute roof corner simulating the wind conditions in separated flow regions
pressures for critical wind directions was attributed by have been identified.
Selvam [14] to coarse discretization or to the limitation
of the k--E model.
2. COMPUTATIONAL APPROACH
The numerical evaluation of wind-induced pressures
on flat roofs is most challenging because of the complex The original Navier-Stokes equations have been aver-
turbulence flow regime conditions resulting from the aged over time to obtain the Reynolds averaged Navier-
wind flow, particularly for oblique wind directions. How- Stokes (RANS) equations for mean velocities. The Reyn-
ever, the previous review clearly shows that most of the olds stresses appearing as extra terms in the RANS equa-
studies dealing with the computation of wind pressures tions are determined by the widely used k--E turbulence
on roofs have generally reported data for normal wind model. The final governing equations for mean flow are
conditions only. Such numerical suctions along the cen- as follows :
treline of a flat roof have been plotted along with avail-
able experimental data (discrete points) for both lower
buildings (H/L < 0.4) and taller buildings (H/L > 0.85)
in Fig. 1. For low-rise buildings, considering the differ-
ences in the geometries and the exposure characteristics
of the various cases, the agreement between the computed
and the experimental data is fairly good with the excep-
tion of a two-dimensional simulation that gives an overall
overestimation of the suctions. For taller buildings, the
numerical results agree reasonably well with the exper-
imental data except in the windward edge region where
the numerically predicted negative pressures show an
apparent overestimation of the experimental pressure
data. In fact, the curvatures of experimental and com-
putational curves have opposite signs in this region.
The present paper presents the results of a study on
the numerical evaluation of wind-induced suctions on a
where
flat roof for both normal and oblique wind directions.
The effect of distance between the first gridline and the
roof surface on the computation of wind pressure has p,P+;k (6)
P
been investigated. Square (15.2 cm x 15.2 cm) low and
taller building models (3 and 13.5 cm high, respectively) is the augmented pressure and p is the mean pressure. In
were considered with the suctions evaluated on both cen- equations (I)-(5), U; represent the mean velocity com-
trelines and edgelines. Detailed experimental measure- ponents along X, y and z axis ; k is the turbulence kinetic
ments of the high suctions near the windward roof edges energy; E is the isotropic dissipation of turbulence kinetic
and corners were also carried out in the boundary layer energy ; v, is the eddy viscosity ; p is the fluid density and
wind tunnel of the Centre for Building Studies in order C,, C,, Cz, ok and ot are constants equal to 0.09, 1.44,
to make thorough comparisons with the respective 1.92, 1.O and 1.3, respectively.
 Evaluation of Wind Pressures on Fiat Roofs 269

treatment method proposed by Stathopoulos and Bas-

karan [K] is applied for the formulation of conditions
near solid boundaries.
For the calculation of tangential velocity components
on the first gridline near a solid boundary, a correction
in the source term of the discretized RANS equations is
needed to account for the friction effect from the bound-
line
ary [I 11. This friction effect is evaluated by the well
known wall functions [ 16] shown as folIows :
line

u; = for Y+ > 11.6 (8)

I Ill I C1/4kWK
P

0 nOdesforP,
k, 6
0 5 y+
P ncdas for velcclty components u, = ___ for Y+ < 11.6, (9)
CWk1/2
-T direction of velocity components
P
where lJ, represents the component of tangential velocity
Fig. 2. Staggered grid arrangement in the corner of a building.
along the xi direction, z, is the corresponding shear stress
component on the wall, Y+ is the dimensionless distance
from the wall to the first gridline given by equation (7)
These steady state equations are nonlinear and are
with U* = CL“’ kj”, K is the von KgrmPn constant, E = 9
solved by iteration. They are first integrated over appro-
is the roughness parameter and d, = do/2 is the distance
priate control volumes on a staggered grid system and
from the first gridline to the wall-see Fig. 2.
then the resulting discretized equations are solved by
Equations (7), (8) and (9) are used to evaluate the
using the well known SIMPLE method [ 151. It should be
components of wall friction (T,) which will correct the
noted that because of the staggered grid system used, the
source term of the discretized RANS equations for the
numerica pressures on flat roofs are not evaluated on
calculation of tangential velocities on the first gridline
the roof surface, but on the first gridline above the roof
near this wall. Compared with Paterson and Apelt [5]
as shown in Fig. 2. Some researchers linearly extrapolate
and Baetke et al. [l l] who have only adopted the log-
the pressure on the first and second gridlines to the roof
arithmic law, the current study uses either the logarithmic
surface [lo, 121. Others apply the values for first gridline
law-equation @-or the linear law--equation (9)-to
to roof surface [6, 71. Regardless of the procedure used,
evaluate the wall friction according to the magnitude
the distance between this first gridline and the roof
of the dimensionless distance (Y+). The caIculation of
(denoted as d,) appears to be an important parameter
velocities farther from the solid wall uses the finite differ-
for the computation. Compared with previous works in
ence form of RANS equations without correction in the
which this distance was taken between 5 and 15 mm, the
source terms.
present study has adopted 0.7, 1.5and 3.0 mm to examine
As mentioned previously, the zonal treatment method
the effect of this distance on the values of the computed
[8] has been applied for the formulation of turbulence
suctions. In fact, the dimensionless distance
conditions (k and e) near solid boundaries to bridge the
d U* flow properties in the turbulence field with those in the
y+ =P
(7) viscous sublayer near solid boundaries. According to this
v ’
method, the wind flow near a solid boundary is composed
in which U* is the friction velocity and v is the kinematic of two regions. One is the fully turbulent flow region far
viscosity of the fluid, is around 5 for d, = 0.7 mm, 11 for from solid boundaries and the other is the viscous sub-
d, = 1.5 mm and 25 ford, = 3.0 mm. layer (VSL) near the walls, the roof and the ground. In
the fully turbulent flow region, the k--E model is applied
3. BOUNDARY CONDITIONS to obtain the turbulence properties, while in the VSL the
turbulence kinetic energy k and dissipation E are deter-
The boundary conditions of velocity at free boundaries mined by the equations
are fixed at the initial power-law velocity profile (i.e.
u/ ug = (ZjZg)Q.‘S, V = 0 and W = 0). The corre-
(10)
sponding initia1 turbulence properties (k and E) are also
used as the free boundary conditions.
For solid boundary conditions, the problem is a little (11)
more difficult. In the region near solid boundaries where a-m
velocity gradients are very high so that the viscous effects where k,, and k, are the turbulence kinetic energy at the
are important, the fully turbulent flow conditions are not edge and within the VSL, respectively, considered at the
satisfied and the k--E model is no longer applicable. The distances cd, and d, from the solid boundary. In equation
conventional method to solve this problem is to use the (1 l), E, is the turbulence energy dissipation at distance d,
semi-empirical wall function to connect the solid bound- from the wall.
ary conditions to the corresponding tangential velocities Although the above-mentioned boundary conditions
near solid boundaries. In the present study, the zonal are only valid for the turbulent flow over a large plane,
 T. Stathopoulos
they also seem to work well for the flow near a small
wind tunnel building model when d,,as well as the dimen-
sionless distance Y+ is small. For the computation of
flow conditions around a full-scale building, however,
these boundary conditions should be represented by the
usual log-law wall function method as in [l I] since the
distance between first gridhne and solid boundary will
normally be much larger than the viscous sublayer thick-
ness.
To apply the relevant boundary conditions, the pro-
gram used in this study evaluates the position of each
grid point. If this point is on a free boundary, the cal-
culation will be omitted to keep the initial values as free
boundary conditions. When the point is inside the build-
ing, the variables (velocities, pressures, turbulence
properties, etc.) will be fixed at an almost zero value by
using the Payne-Iron method, i.e. by multiplying the
variable with a higher order large number on both sides
of the equation concerned so that the other terms appear-
ing in this equation can be ignored 1151. If this grid point
is on a solid wall, the velocity component normal to this
wall will also be set to an almost zero value. If this point
is on the first gridhne near a solid boundary, the solid
boundary conditions for velocity components, i.e. equa-
tion (8) or (9) will be used depending on the dimen-
sionless distance Y+. If this point is inside the VSL,
equations (10) and (11) will provide k and E.
4. MESH REFINEMENT
As mentioned previously, the staggered grid system
has been adopted in the present study. Although the non-
staggered grid system is a little easier to use, the staggered
grid is chosen since the evaluation of the boundary con-
dition for pressures is avoided [15]. In fact the pressures
near solid surfaces can be obtained directly without the
extrapolation required for non-staggered grids [ 171. The
gridnodes for normal wind velocity components are set
on the solid boundary and the high density gridnodes are
only used in the region where they are required, i.e. near
the building envelope and near the ground. Several com-
putational grid arrangements of various densities have
been tested and a sensitivity analysis has been carried out
with d,, fixed at 0.7 mm. The same dimension of the
computational domain has been maintained in all these
computations. The results presented in Fig. 3 for the
case of the taller building model show that when grid
arrangement is reasonably dense, the effect on wind pres-
sures on the roof surface is very limited, whereas a coarse
grid arrangement, say 22 x 2 1 x 16, naturally causing sev-
ere non-uniformity in the mesh, apparently affects the
numerical results. Based on the results of this analysis, a
computational grid arrangement of 46 x 38 x 34 has been
used for the taller building model whereas a smaller grid
44 x 34 x 26 has been selected for the low building model.
All computations were carried out by using a VAX
station 3100 model 76, in which a typical run takes about
70 minutes CPU time.
All experimental measurements were made in the
boundary layer wind tunnel of the Centre for Building
Studies, Concordia University. The tap locations for the
building models tested in the wind tunnel for normal and
oblique wind directions are shown in Fig. 4. Data shown
and Y. S. Zhou
in this paper are for simulated wind flow in an open
country terrain exposure with a velocity profile described
by a power-law with exponent equal to 0.15. Results are
presented in terms of mean pressure coefiicients CC,,), i.e.
pressures normalized by the dynamic pressure at roof
height. The Reynolds number of the flow is of the order
of 10’ based on the width of the model. More details
about the experimental measurement procedures can be
found in [IS].
5. RESULTS AND DISCUSSION
Previous studies on the computational evaluation of
wind pressures on roofs have reported results mainly
along the central axis of the roof such as those presented
in Fig. 1. Numerically evaluated pressures along the edge-
line of a flat roof have seldom appeared in the literature.
In the present study, roof pressures have been computed
at a distance d, = 3 mm from the roof edge for both low
and tall building models. Data have also been obtained
at the same location from wind tunnel measurements and
results are compared in Fig. 5 for normal wind direction.
It should be noted that the first gridhne above the roof
has been placed at a distance d, = 5 mm from the roof
surface and the computed pressures on the roof have
been found by linear extrapolation. Experimental data
from a previous study by Stathopoulos et al. [19] have
also been included in the comparison although these data
have been measured at a distance d, = 6 mm from the
edge. Computational results have been denoted by a con-
tinuous line whereas experimental data have been indi-
cated by discrete points throughout this paper.
The comparison between computed and measured data
shown in Fig. 5 clearly shows that in the case of the low
building model, the computed results underestimate the
measured values significantly near the windward edge
where the highest suctions occur. For the higher building
model for which the distance d,, = 5 mm is relatively
smaller in terms of building height, the computed data
are near the measured values at the windward edge but
they drop sharply causing underestimation of suctions
farther downstream. On the other hand, the lack of even
qualitative agreement between computed and measured
results is noted as also previously (Fig. 1).
In the light of the above comparisons and previous
findings as well, it was decided to investigate more thor-
oughly the effect of distance between the first com-
putational gridline and the roof surface on the computed
pressure coefficients. Although different wind directions
from 0 to 180” have been examined, only the typical
results for 0 and 45” are presented in this paper along
with the limited experimental data available for compari-
son. The simulation of oblique wind cases is achieved by
rotating the flow field rather than the building itself to
avoid the difficulty in re-arranging the gridnodes.
Figure 6 shows the results of this investigation for both
centreline and edgeline (de = 3 mm) on the roof of the
low building model by using three distances (d,) between
the first gridline and the roof surface, namely 0.7, 1.5 and
3.0 mm. For such small values of d,,, differences between
the results computed with linear extrapolation to the roof
surface and those computed without extrapolation are
very small. Therefore, the pressure coefficients presented
 Evaluation of Wind Pressures on Flat Roofs

_ .- .~ x/L
x/L
Fig. 3. Pressure coefficients under various mesh sizes.

hereafter are those found on the first gridline without

extrapolation. Experimentally determined pressure
coefficients have also been included in the graphs for
comparison purposes. However, it should be noted that
only those referred to as “present study” correspond to
UNIT,mm
the exact geometrical configuration and terrain exposure
76 assumed in the computation.
Two matters are worthy of consideration in these com-
00
parisons. First, the similarity of the three computational
-) **.........
.. . . . . . .
data sets for both centreline and edgeline of the low roof
except near the windward edge-in a zone equal to about
10% of the width of the model-where significant dis-
76 crepancies occur. In particular, the lower the distance
(d,), the higher the computed suction coefficients,
whereas for the same d, the suctions computed on the
,,,$ ........... . . . . . . . . .
: edgeline are lower than the corresponding values com-
puted on the centreline. The other point to be made
here is the general similarity between the computed and
measured data although the computed values seem to
Fig. 4. Tap locations for the building model tested in the wind
tunnel. underestimate the experimental results except near the

2.6- , ( , , , I , , ,

1.1;g===jp dp=5.0mm

t'
2.8 -

Gye- 4 ISI
** Ill
.. ..._.
4.6 - w
+ Stathopoulos et a1 (1981)(d,=6.0mm)
0 Present study(d,=d.Omm) 1.4 - + Stathopoulos et al (19Sl)(d,=6.Omm) _
0 Present study(de=3.0mm)

Fig. 5. Pressure coefficients along the edgeline of the roof (0” azimuth).
272 T. Stathopoulos and Y. S. Zhou

2.6 , , / , , > , , (
I , I ,
, ( 2.6 ,,, , ,,,,r,,,I ,// ,

K B===i!Z dp (mm)
2.4: p’ig dp (mm)

ql ui
I.2 - - 0.7 2.2 - - 0.7
t-x 1.5 t” 1.5
-cp -
2.9 -
4 152 3.0 - _Q 2.0 : I 152 3.9

1.e - 1 B- a* -------
w w
I.6 - 1.6 -
+ Stathapoulos et al (1961)(4,=6.0mm)
1.4 - 0 Present study(d,=3.0mm)
+ SBthoPo”los et aI (lssl) 4

A Sbnhopoulos and DumltrercwBrulotte (rses) _

0 Prb¶wdstUdy

,, ,2 .3 .I .5 .b .7 .B .9 1.0 B .I .2 ,3 ,I .5 .6 ,7 .a .9 1.B
X/L x/L
Fig. 6. Pressure coefficients along the centreline and edgeline (low building, 0’ azimuth)

+ Sbthopoulor
et sl(1991) + Stathopoulo9 et al (1961)(de=6.0mm) -
0 Castro nnd Roblnl) (1977)
o Present study(d,=3.0mm)

x/L x/L
Fig. 7. Pressure coefficients along the centreline and edgeline (tall building, 0’ azimuth).

centre of the windward edge, for which the opposite trend
is observed.
Such investigations for the taller building were also
attempted and the results are shown in Fig. 7, in the same
format as that of Fig. 6 and compared with available
experimental data including those obtained in the present
study with taps placed as close as 0.5 mm to the windward
edge-see Fig. 4 for tap locations. The comparison shows
that the suction curves along both the centreline and
edgeline of the roof do not follow the same trend with
the experimental data. In particular, the computed results
generally underestimate the experimental data, with the
exception of a small region near the windward edge where
the suctions on the roof are apparently overestimated.
should be noted that the proximity of the first gridline to
the roof surface does not improve the computed data, in
fact, the smallest dp causes the maximum overestimation
of suctions in this windward edge region. This is attri-
buted to the limitations of the k--E model which fails to
predict the reverse flow over the roof [12, 161 by over-
estimating the k values and, consequently, the skin fric-
tion values [16, 201. Furthermore, the wall-function-
based boundary conditions used are deduced from a two-
dimensional fully developed turbulent flow over a large
plane and these are not fully applicable even to the region
far from the windward edges. The application of these
coarse boundary conditions in the windward edge region
is not expected to provide accurate predictions. However,
it is of interest to know the error included by estimating
the discrepancy from the experimental results. Some
other facts which may also affect the quality of the com-
parisons need to be mentioned. The experimental data
shown in this paper are only average values whose vari-
It ances account for about 20% of their mean values. The
mean pressures measured in different experiments show
also differences among themselves as, for instance, dis-
played in Fig. 1. This is not surprising considering the
differences of physical simulation in the various wind
tunnels.
The influence of the magnitude of the distance d,, on
 Evaluation of Wind Pressures on Flat Roofs 273

.,,.
.I,, -1.. -II ..“..‘L”“““““‘-

Fig. 8. Computed
-1.0 -,., -a,, -,..

pressure
-9.2

CP
-1.a -I

coefficients
-.a -.* -.2

for various
,

gridline
L -1..

distances
-IA -1,, -7,. -1..

from the roof surface

4..

cp
-1.1 -t,, -,,

(0” azimuth).
-,, .., -1 ,

,I *, I., . *, *. I, I .I
.I., -2.4 -1.1 -*., -,., .I,, -3,. -4.2 -I., -.a -, .., ..I ,

cp cp
Fig. 9. Computed pressure coefficients for various gridline distances from the roof surface (45” azimuth)

the numerical roof suctions has been examined in detail presented for both the centreline and the edgeline of the
at four typical positions on roof surface. Results of this roof in Fig. 10 and are compared with the previously and
investigation are presented in Fig. 8 for both low building currently obtained experimental data. The similarity of
and taller building configurations. Numerically evaluated the three computational data sets for the centreline is
pressure coefficients are almost d,-independent at pos- apparent except near the windward edge where dis-
itions far enough from the windward edge (points 2 and crepancies occur; in particular, the smaller the d, the
4). However, at positions close to the windward edge higher the computational suctions. Discrepancies, how-
(points 1 and 3), suction coefficients increase dramatically ever, exist among these three computed data sets for
with decreasing d,, particularly for the taller building. the edgeline with the smallest d, providing the highest
The positions examined represent two distinct regions, suctions. The agreement between computed and exper-
namely: the low pressure gradient zone far from the imental data for the centreline is generally good except
windward edge-say x is larger than 10% of building in the windward edge region where the computed values
width-in which the numerical pressures are almost d,- underestimate the measured data. However, the pre-
independent ; and the high pressure gradient region next diction of pressure coefficients along the edgeline is far
to the windward edge which shows the drastic influence from satisfactory. In particular, regardless of the mag-
of the d, on the computed results. nitude of d,,, the computed results show a smooth change
Pressure coefficients at these four positions for the of suction coefficients while the experimental data follow
oblique wind condition (45’) have also been examined a different pattern. This discrepancy is possibly caused
and are presented in Fig. 9 in the same format as that of by the k--E model which is not representative for this
Fig. 8. Under this wind direction, position 4 is no longer high vorticity region, at least in its standard form. The
an inside roof point. Therefore, pressure coefficients at numerical false diffusion due to the skewness of the main
this position are also d,-dependent similar to those com- flow to the gridlines described by Patankar [ 151 may be
puted at position 1. However, the pressure gradients at another reason for this discrepancy.
these positions (1 and 4) are relatively smaller than that Figure 11 shows comparisons for the case of a taller
at position 3 (corner point). For the position far from building. The data show similar trends to those discussed
both windward edges (point 2), the pressure coefficients in Fig. 10. In particular, the numerically evaluated suc-
are still almost d,,-independent. tions along the centreline agree fairly well with the exper-
For oblique wind directions, the flow conditions above imental data only for X/L greater than about 0.2, con-
the roof are more complex with conical vortices develop- trary to the region near the windward edge where the high
ing along the roof edges. Pressure coefficients for 45’ suctions are clearly underestimated by the computational
oblique wind direction have been computed for three approach. Regarding the windward edge region of the
different d,s on the roof of a low building. Results are roof, the present methodology does not provide appro-
274 T. Stathopoulos and Y. S. Zhou

2.6, , ( , , , , , , , , , , , , , , , 2.6 I , , , , , , , I ! I I

!&r=iB 1 !gA_%
I.

dp (mm) 2.4 dp (mm)

24-
- t’ -
----
o., 2.2 - 0.7
2.2 -
I-‘: - -- 1.,

IS’
- -- I.,

-cp,.,_
qI....
3.c 2.0 - ---- 3.0
2.8 -
152 -cp,,, - dl*01
.......
;/-
1.6 - + Stathopoulos et al (1961)(d,=6.0mm) -
0
1.4 - 0 Present rtudy(d,=3.0mm)
0
I 2r
0
1 .B D

.B
_ _ -+ -
n
- - - -cl-
- - __ i
‘1
-&
.6 . - ___~__e___*___7_____________~_____
00 _----*---- +
i----
_m-
_-+ I
A
‘“I-_ + 4

,a 0 .I .2 .3 ,a .5 .6 .T .9 .9 1.B

X/L X/L
Fig. 10. Pressure coefficients along the centreline and edgeline (low building, 45” azimuth).

0
+ Stathopouloa et al (1961)(de=6.0mm)
2.6
-Q (3 Present study(d,=t.Omm)
2.4 0
0
+ 6mhapo”lo* et Ill (leal) i !

I + I
2.2
0 Came .nd Robins (1977)

0 PIwed study 2..

Fig. 11. Pressure coefficients along the centreline and edgeline (tall building, 45” azimuth)

priate values of pressure coefficients. Numerically-evalu- Although the solution of RANS equations with the k--E
ated data either underestimate or overestimate the exper- model seems to be computationally attractive, significant
imental results more significantly than in the case of the discrepancies in the evaluation of flat-roof pressures, par-
lower building. Numerical false diffusion and the inability ticularly in high vorticity areas, have also been found.
of the k--E model to describe the flow conditions near the Other approaches such as the large-eddy simulation and
windward roof edges and corners are the main reasons algebraic stress models are currently under investigation
for this discrepancy. However, numerically-evalu- for the computational evaluation of wind-induced pres-
ated pressures on the other region of the roof agree well sures on flat roofs [21, 221. Direct simulation might be
with the experimental data, particularly for small d, adopted in the future when more powerful super-
values. computers become available to scientists and engineers.
 Evaluation of Wind Pressures on Flat Roofs 275

6. SUMMARY AND CONCLUSIONS probably due to the limitation of the standard k--E model
which overpredicts the skin friction and has the tendency
The wind flow over the roof of a rectangular building
to keep the calculated flow attached on the roof in the
is very complex since it involves high turbulence, severe
areas where the experiments indicate separation [20]. For
pressure gradients, separation and possible reattachment.
low-rise buildings, this separation region is small, so that
This makes the numerical modelling and evaluation of
the numerical results from the k--E model can still show
wind-induced pressures on roofs very difficult.
good agreement with the experimental data. For taller
Computations of wind pressures on flat roofs have
buildings, however, such separation regions become
been made for three different distances (d,) of the first
much larger [23] and the failure of the k--E model in
gridline to the roof surface, namely, 0.7, 1.5 and 3.0 mm.
predicting the reverse flow region over the building roof
The pressure coefficient variation versus d, has also been
is more critical. It should also be noted that the boundary
investigated at four typical roof locations. It has been
conditions, e.g. the wall functions used for this region,
shown that the roof surface can be divided into two
are only valid for the turbulent flow near a large plane
subregions, i.e. the high pressure gradient region where
under zero pressure gradient [20]. Such boundary con-
the computed pressure is highly dependent on the dis-
ditions are approximately accurate in the region far from
tance d,, and the low pressure gradient region in which
the windward edge and too coarse to provide good pre-
the computed pressure is nearly d,-independent. Under
dictions elsewhere. The deviation of numerical pressure
normal wind conditions, the high pressure gradient
from the experimental data at the roof edge regions is
region is small-approximately 10% of the roof sur-
more apparent under oblique wind direction especially
face-whereas the high pressure gradient region for
for the taller building model. This may be attributed to
oblique wind conditions is larger-around 20% of the
the high vorticity in this separation region which is not
roof surface. The influence of d, on the predicted pressure
well represented by the standard k--E model. The numeri-
near the windward edge is dramatic especially for the
cal false diffusion caused by the skewness of wind velocity
oblique wind cases whereas the pressure outside the sep-
to gridlines described by Patankar [I 51 is possibly another
aration region is not apparently affected. The com-
reason for this discrepancy. Better prediction may be
parisons with experimental data have indicated that the
achieved by a more detailed mathematical model such as
computed results in the d,-dependent region always show
large eddy simulation or even direct simulation. More
larger differences, whereas grid refinement can give
accurate three-dimensional boundary conditions for the
improved results only for the low building model.
roof corner and the windward roof edge regions are also
In general, the commonly used RANS equations with
needed.
the k--E model for closure can predict the wind pressures
on the flat roof of a low building especially for normal
wind conditions when adequate dp is used. For a taller Acknowledgements-This study has been supported by the Natu-
building, the k--E model can be used with high res- ral Sciences and Engineering Research Council of Canada
(NSERC) and the Fonds pour la Formation de Chercheurs et
ervations since it fails to predict, even qualitatively, the
I’Aide ii la Recherche (FCAR) of the province of Quebec. The
trend of the experimental data. In particular, the suctions authors are grateful to one of the reviewers for the comments
at flow separation areas cannot be predicted properly, and the suggestions made for the improvement of this paper.

REFERENCES

I. T. Hanson, D. M. Summers and C. B. Wilson, A three-dimensional simulation of wind flow around

buildings. International Journalfor Numerical Methods in Fluids 6, 113-127 (1986).
2. T. Summers, T. Hanson and C. B. Wilson, Validating of a computer simulation of wind flow over a
building model. Building and Environment 21, 97-l 11 (1986).
3. D. Vasilic-Melling, Three-dimensional turbulent flow past rectangular bluff bodies, Ph.D. Thesis,
University of London (1977).
4. I. P. Castro and A. G. Robins, The flow around a surface-mounted cube in uniform and turbulent
streams. Journal ofFluid Mechanics 19(2), 307-335 (1977).
5. D. A. Paterson and C. J. Apelt, Computations of wind flows over three-dimensional buildings. Journal
of Wind Engineering and Industrial Aerodynamics 24, 19332 13 (1986).
6. D. A. Paterson and C. J. Apelt, Simulation of flow past a cube in a turbulent boundary layer. Journal
of Wind Engineering and Industrial Aerodynamics 35, 149- 176 (1990)
7. A. Baskaran and T. Stathopoulos, Computational evaluation of wind effects on buildings. Building
and Environment 24,325-333 (1989).
8. T. Stathopoulos and A. Baskaran, Boundary treatment for the computation of three dimensional
wind flow conditions around a building. Journal of Wind Engineering and Industrial Aerodynamics 35,
177-200 (1990).
9. S. Murakami, Computational wind engineering. JournaI of Wind Engineering and Industrial Aero-
dynamics36, 517-538 (1990).
10. S. Murakami, K. Hibi and A. Mochida, Three dimensional analysis of turbulent flowfield around
street blocks by means of large eddy simulation (part-l). Journal of Architecture, Planning and
Environmental Engineering, Architectural Institute of Japan 412, l-10 (1990).
il. F. Baetke, H. Werner and H. Wengle, Numerical simulation of turbulent flow over surface-mounted
obstacles with sharp edges and corners. Journal of Wind Engineering and Industrial Aerodynamics 35,
129-147 (1990).
12. R. P. Selvam and D. A. Paterson, Computation of pressure on and velocities near the Texas Tech
building : using staggered and non-staggered grids, Ninth ASCE Structures Congress, Indianapolis,
IN, 29 April-l May (1991).
276 T. Stathopoulos and Y. S. Zhou

13. D. Surry, Pressure measurement on the Texas Tech building-II : wind tunnel measurements and
comparison with full scale, Proceedings Eighth Colloquium on Industrial Aerodynamics, pp. 2355247,
Aachen, West Germany, 4-7 September (1989).
14. R. P. Selvam, Roof corner pressures on the Texas Tech building : numerical and field results. Tenth
AXE Structures Congress, San Antonio, TX, 13-15 April (1992).
15. S. V. Patankar, Numerical Heat Transfer and Fluid Flow, McGraw-Hill, New York (1980).
16. S. Murakami, A. Mochida and Y. Hayashi, Examining the k-e model by means of a wind tunnel test
and large-eddy simulation of the turbulence structure around a cube. Journal of Wind Engineering
and Industrial Aerodynamics 35, 87- 100 ( 1990).
17. M. Peric, R. Kessler and G. Scheuerer, Comparison of finite volume methods with staggered and
collocated grids. Computers and Fluids 16(4), 3899403 (1988).
18. T. Stathopoulos and M. Dumitrescu-Brulotte, Design recommendations for wind loading on buildings
of intermediate height. Canadian Journal of Civil Engineering 16,9 10-9 16 (1989).
19. T. Stathopoulos, D. Surry and A. G. Davenport, Effective wind loads on flat roofs. Journal af’ the
Structural Division, American Society of Civil Engineers 107(ST2), 281-300 (1981).
20. W. Rodi and G. Scheuerer, Scrutinizing the k-e turbulence model under adverse pressure gradient
conditions. Transactions qf the American Society ofMechanical Engineers 108, 174-179 (1986).
21. S. Murakami, Comparison of various turbulence models applied to a bluff body. Journal of Wind
Enaineerinq 52, 1644179 (1992).
22. S. Kawamoto, S. Kawabata and T. Tawahashi, Numerical analysis of wind around building using
high-speed GSMAC-FEM. Journal of Wind Engineering 52, 2555260 (1992).
23. D. J. Wilson, Flow patterns over flat-roofed buildings and application to exhaust stack design.
ASHRAE Transactions 85(2), 284-295 (1979).
24. A. Qasim, T. T. Maxwell and S. Parameswaran, Numerical predictions ofthe flow over a 2-D building,
Ninth AXE Structures Congress, Indianapolis, IN, 29 April-1 May (1991).