Está en la página 1de 16

ARTICLE IN PRESS

Journal of Wind Engineering


and Industrial Aerodynamics 95 (2007) 13841399
www.elsevier.com/locate/jweia

Wind-tunnel modelling of the Silsoe Cube


P.J. Richardsa,, R.P Hoxeyb, B.D. Connella, D.P. Landera
a

Department of Mechanical Engineering, University of Auckland, Auckland, New Zealand


b
Silsoe Research Institute, Silsoe, UK
Available online 13 March 2007

Abstract
1:40 scale wind-tunnel modelling of the Silsoe 6 m Cube at the University of Auckland is reported.
In such situations, it is very difcult to model the full turbulence spectra, and so only the highfrequency end of each spectrum was matched. It is this small-scale turbulence that can directly
interact with the local ow eld and modify ow behaviour. This is illustrated by studying data from
tests conducted in a range of European wind tunnels. It is recommended that spectral comparisons
should be carried out by using turbulence-independent normalising parameter, such as plotting
fS(f)/U2 against reduced frequency f nz/U. Using parameters such as the variance and integral
length scale can easily mask major differences. It is noted that it is the size of the tunnel that limits the
low-frequency end of the spectra, and so the longitudinal and transverse turbulence intensities were
lower than in full scale. In spite of this similar pressure distributions are obtained. Some differences
are observed and these are partially attributed to the reduced standard deviation of wind directions,
which affects both the observed mean and peak pressures by reducing the band of wind directions
occurring during a run centred on a particular mean direction. The reduced turbulence intensities
also affect the peak-to-mean dynamic pressure ratio. However, since the missing turbulence is at low
frequencies, the peak pressures appear to reduce in proportion. By expressing the peak pressure
coefcient as the ratio of the extreme surface pressures to the peak dynamic pressure observed during
the run, reasonable agreement is obtained. It is argued that this peakpeak ratio is also less sensitive
to measurement system characteristics or analysis method, provided the measurement and analysis of
the reference dynamic pressure is comparable with that used for the surface pressures.
r 2007 Published by Elsevier Ltd.
Keywords: Wind tunnel; Turbulence; Cube

Corresponding author. Tel.: +64 9 3737599; fax: +64 9 3737479.

E-mail address: pj.richards@auckland.ac.nz (P.J. Richards).


0167-6105/$ - see front matter r 2007 Published by Elsevier Ltd.
doi:10.1016/j.jweia.2007.02.005

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1385

1. Introduction
The wind-tunnel modelling of low-rise buildings introduces a number of challenges that
are less evident with larger buildings. In order to reproduce the details of the ow eld
around such buildings, a moderately large scale is required. Tieleman (2003), for example,
recommends utilising models of low-rise buildings that have a scale not smaller than 1:50.
The use of such a large scale inevitably means that the largest turbulence length scales in
the wind tunnel are much smaller than the scaled full-scale equivalents. In such situations,
the modeller must decide whether to match the turbulence intensity, the integral length
scale or neither. The wind-tunnel modelling reported in this paper set out to reproduce the
conditions experienced by the Silsoe 6 m Cube in the University of Auckland boundary
layer wind-tunnel.
The Silsoe Cube provides a facility for fundamental studies of the interactions between
the wind and a structure. It is situated in an open country exposed location at the Silsoe
Research Institute, UK. Surface pressure measurements have been made on a vertical and
on a horizontal centreline section with additional tapping points on the roof. Fig. 1 shows
the full-scale and wind-tunnel cubes.
2. The German comparative study
The approach taken in modelling the approach ow for the current study was inuenced
by the results of the German comparative study of wind-tunnel modelling of the pressures
on a cubic building reported by Holscher and Niemann (1998). That cubic building was
50 m high and was to be modelled in a neutral suburban boundary layer with a prole
exponent a 0.2270.02. A total of 15 simulations, from various European institutions,
were reported and these demonstrated signicant scatter. Richards et al. (2001) compared
this European wind-tunnel data with that from the Silsoe 6 m Cube, which could be
considered to be a 1:8.33 scale model of the 50-m cube except that the equivalent roughness
length of 83.3 mm is a little small for suburban terrain along with the prole exponent
which is approximately a 0.17. While it was noted that the wind-tunnel data
demonstrated the sensitivity of the pressures to approach ow conditions, no clear
pattern was discerned. Tieleman (2003), following the lead of Melbourne (1980), has
suggested that such pressure results are sensitive to the small-scale turbulence level which
can be characterised by a small-scale spectral density parameter S nSuu(n)/U2 evaluated
at n 10U/LB, where U is the mean wind speed at the relevant height and LB is a
characteristic length of the building. Melbourne (1980) notes that the spectral density is
normalised by the mean velocity, rather than the variance, so that the effects of turbulence
intensity can be included.
These principles can be demonstrated by considering six of the 15 European simulations
for which turbulence properties and spectra are available (Niemann, 2000). Table 1 gives
some of the ow characteristics of these six studies while Fig. 2(a) shows the pressures on
the cube from these studies along with the Silsoe data. It is clear that the ow over the roof
of the cube is being affected by differences in the ow and that in the centre of the roof the
mean pressure coefcients vary by up to a factor of 2.
Although the data available does not allow calculation of Melbournes small-scale
spectral density parameter, the general level of small-scale turbulence can be investigated.
Fig. 2(b) shows the European spectra plotted in von Karman form, where with the

ARTICLE IN PRESS
1386

P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

Fig. 1. (a) The Silsoe 6 m Cube and (b) the 1:40 scale model.

variance and integral length scale determined from the turbulence it appears that all six
studies are similar. However, plotting the same data using turbulence-independent
normalising parameters in Fig. 2(c) reveals a different picture. It can now be seen that the
high-frequency small-scale turbulence levels vary signicantly. Wind-tunnels 4 and 5 had
the lowest levels of small-scale turbulence, but still larger than Silsoe, and these gave the
pressures closest to the Silsoe results. On the other hand, tunnel 10 has one of the highest
small-scale turbulence levels and has produced the least negative roof pressures. The
exception to this pattern is tunnel 11, which has high small-scale turbulence but gave
pressures that are in the middle of the bunch.
In the past it has often been stated that pressures are more sensitive to changes in the
total turbulence intensity than to changes in integral length scale. For example, Melbourne
et al. (1997) state: The length scale, while of importance, does not have a major inuence

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1387

Table 1
Characteristics of the ow for six of the wind-tunnel studies of ow around a cubic building reported by Holscher
and Niemann (1998) and Niemann (2000)
WT

Scale

U (0.6h)

Iu (0.6h)

Lux (0.6h)

Lux (30 m)

3
4
5
10
11
14

500
312.5
250
250
750
500

5.775
9.17
?
6.425
4.356
6.28

0.2067
0.162
0.166
0.227
0.224
0.217

0.22
0.3
0.38
0.338
0.077
0.295

110
93.75
95
84.5
57.75
147.5

3
5
11
Silsoe F-S

0.5

4
10
14

0
0

-0.5
-1
-1.5
Distance Over Cube (Cube Heights)

nSuu(n)/var(u)

Mean Pressure Coefficient

0.35
3
4
0.30
5
10
0.25
11
14
0.20
0.15
0.10
0.05
0.00
0.001 0.010 0.100 1.000 10.000 100.000
nLux/U(z)

fSuu(f)/U(z)^2

c
0.016
0.014
0.012
0.01
0.008
0.006
0.004
0.002
0
0.0001 0.001

3
4
5
10
11
14
Silsoe

0.01
0.1
f=nz/U(z)

10

Fig. 2. (a) Mean pressure coefcients on the vertical centreline of a cubic building, (b) longitudinal spectra at
z/h 0.6 plotted in von Karman form and (c) normalised by turbulence-independent parameters (the Silsoe
spectra is at z/h 0.5).

on wind load estimates and an error by a factor of 2 will introduce errors in the load of the
order of 10%. Of greater signicance is the intensity of turbulence, which denes the
magnitude of the wind spectrum.
As is the case with all the spectra shown in Fig. 2(c), it is quite common for wind-tunnel
spectra to be decient in low-frequency turbulence in comparison with full scale. This is
caused by the physical limits created by the tunnel walls that restrict the maximum eddy
size that can exist within the tunnel. Hence, in order to match the full-scale turbulence

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1388

intensity, it is necessary to increase the high-frequency content. It should be noted that if


both the model-scale and full-scale spectra are of the von Karman form and have the same
turbulence intensity, but the integral length scale in the tunnel is only one half of that
which the geometric scaling would suggest, then the low-frequency spectral density will be
half of the target value (obtained with the correctly scaled integral length scale) and the
high-frequency spectral densities will be 22/3 (1.59) times bigger than the target. While wind
tunnel engineers would normally question the turbulence intensity being nearly 26%
higher than the target (variance 59% larger), accepting a situation where the turbulence
intensity is matched and the integral length scale is only half the target implies accepting a
high-frequency spectral density 59% higher than target. It is therefore suggested that when
comparing full-scale and model-scale spectra, it is better to use full-scale turbulence
intensity and integral length scale data, together with the von Karman or similar spectral
equations, to create a target spectrum and to transform this into a form, such as that
shown in Fig. 2(c), which makes use of turbulence-independent normalising parameters to
carry out the comparison in that form. This will highlight where the measured wind tunnel
spectrum matches the target and where there are signicant differences.
Both the work of Castro and Robins (1977) and Ogawa et al. (1983) show that increased
turbulence tends to promote earlier reattachment of the ow on the roof of a cube. It is
probably such changes that lead to the pressure changes observed in Fig. 2(a). Earlier
reattachment has been promoted on the roof of the Silsoe Cube by pitching it into the
wind. Fig. 3 illustrates the changes in roof pressure brought about by pitching the cube
forwards by 2.51 and 51. It may be observed that when at (01 pitch) the roof suctions are
almost constant for the windward third of the roof and are generally more negative over
the centre of the roof. On the other hand, with the roof pitched 51, the pressures reach a
higher peak one-quarter of the way across and then become less negative more rapidly. It is
believed that this is associated with earlier ow reattachment on the roof.
It appears from Fig. 2 that in order to adequately model the ow over the Silsoe Cube, it
will be necessary to match the small-scale turbulence levels. However, as illustrated in
Fig. 2(c), all six European wind tunnels had low-frequency turbulence levels much less than
observed at Silsoe, and hence it is likely that this will also occur in the Auckland tunnel.

Mean Pressure Coefficient

1.0
zero pitch
2.5 deg pitch
5 deg pitch

0.5
0.0

-0.5
-1.0
-1.5
0

1
2
Distance Over Cube (Cube Heights)

Fig. 3. (a) Mean pressure coefcients on the vertical centreline of the Silsoe Cube when pitched forwards and
(b) the cube at 51 pitch.

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1389

As a result, matching the high-frequency spectrum will inevitably mean that the windtunnel turbulence intensities will be lower.
3. The full-scale facility
The Silsoe 6 m Cube has a plain smooth surface nish and has been instrumented with
surface tapping points on a vertical and on a horizontal centreline section with additional
tappings on one-quarter of the roof. Simultaneous measurements have been made of 32
pressures and of the simultaneous wind dynamic pressure and direction derived from a
sonic anemometer positioned upstream of the building at roof height. Tapping points are
constructed of simple 7 mm diameter holes (a size sufcient to prevent water blocking of
the tapping points) and the pressure signals transmitted pneumatically, using 6 mm
internal diameter plastic tube to transducers mounted centrally. Tube lengths of up to 10 m
are used in this system giving a frequency response of 3 dB down at 8 Hz.
This paper will consider the ring of taps on the vertical centreline and at mid-height,
together with the 30 tappings on one-quarter of the roof, as shown in Fig. 4. The corner
roof tappings are in a grid of ve columns and six rows with a spacing of 0.52 m (0.087h) in
both directions. The tappings nearest the roof edges are 0.4 m (0.066h) from the edge.
For the basic data recording, simultaneous measurements of the pressures were made at
a rate of 4.17 samples per second, together with the three components of the wind speed.
A 36-min record length was used (9000 samples) which was sub-divided into three 12 min
segments. The records were processed to give mean, peak and uctuating properties. For
some of the runs the cube was rotated 451 clockwise, relative to that shown in Fig. 4, so
that the instrumented corner was towards the prevailing winds. In order to fully investigate
the roof pressure distribution, it would have been necessary to carry out measurements
with the corner roof taps in a variety of orientations. However, due to a shortage of
suitable wind during the testing period, the only tests completed had the taps on the
windward corner.

b
a
3.48h
Reference Mast
(1.0h high,
1.04h to the side
of cube centre)
Roof Tap 6

Wind
Direction

0.087h
Wall Tap 17

0.066h
0.066h

0.087h

Fig. 4. (a) Plan view of wall and roof tappings on the Silsoe Cube and (b) taps on the Auckland model.

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1390

The velocity prole at the Silsoe Research Institute site has been measured at various
times and the recent measurements are well matched by a simple logarithmic prole with a
roughness length z0 0.0060.01 m. This means that the cube has a Jensen number (h/z0)
of 6001000. The longitudinal turbulence intensity at roof height is typically 1920%.
4. The wind-tunnel model
The University of Auckland modelling of the Silsoe Cube was conducted at a scale of
1:40. At this scale, the 150-mm edge length made for simple manufacture and manageable
location of taps. All of the tappings on the Silsoe Cube were modelled together with some
additional taps, such that both vertical centreline planes, all four faces at mid-height and
six rows of six taps on one-quarter of the roof were included. Since only one-quarter of the
roof was instrumented, tests were carried out with up to four orientations of the model in
order to give a full picture of the roof pressure distribution. Wind directions 901, 751, 601
and 451 were considered. The 88 taps were connected through 325 mm long, 1.5 mm
diameter plastic tubes to a four-transducer scannivalve. Sixty second records were
recorded at 200 samples per second, giving a total of 12,000 samples per record. Testing of
the tubing system showed that the gain was near unity for frequencies below 60 Hz and
reduced signicantly above this cut-off. For the analysis of peak positive and negative
pressures, each record was subdivided into ve segments lasting 12 s and containing 2400
samples. The Lieblein best linear unbiased estimator (BLUE) method (Cook, 1985) was
applied in order to give estimates of the expected extremes.
The University of Auckland boundary layer wind tunnel is 1.85 m wide and 1.1 m high.
Upstream of the model a 7.5 m fetch of roughness blocks, roughness strips and corrugated
cardboard mat were used to generate the boundary layer. The boundary layer up to two
building heights was well matched by a log law with a roughness length of 0.42 mm. This
corresponds to a full scale roughness length of 16.8 mm which is just slightly rougher than
recent measurements at Silsoe. The mean wind velocity at cube height was 6.4 m/s, which is
similar to the typical full scale velocity. In general, only full scale pressure measurements
associated with a mean wind dynamic pressure greater than 20 Pa (wind speed over 5.7
m/s) have been analysed, since at lower speed the measurements were inaccurate and
unreliable. The typical full scale wind speed analysed was about 7 m/s. Fig. 5 shows a
comparison of velocity and turbulence intensity proles. Although the velocity proles are

b
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

Height/Cube Height z/h

Height/Cube Height z/h

a
Wind-tunnel
Silsoe

0.2

1
0.4
0.6
0.8
Velocity Ratio V(z)/V(h)

1.2

2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

Silsoe U
Silsoe V
Silsoe W
Wind-tunnel U
Wind-tunnel V
Wind-tunnel W

10 15 20 25 30 35
Turbulence Intensity Ia (%)

Fig. 5. Full-scale and wind-tunnel (a) velocity and (b) turbulence intensity proles.

40

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1391

0.012

0.4

0.01
-fCuw(f)/u2

fSaa(f)/U(z)2

0.3
0.008
0.006
0.004
0.002
0
0.0001

0.001

0.01

0.1

Reduced Frequency f=nz/U(z)

10

0.2
0.1
0
0.0001

0.001

0.01

0.1

10

-0.1
Reduced Frequency f =nz/U(z)

Fig. 6. Comparison of (a) spectra and (b) cospectra for the Silsoe site and the Auckland wind-tunnel at half cube
height.

quite similar, there are signicant differences to the turbulence intensities, which is further
illustrated by the spectra and cospectra in Fig. 6.
A number of points may be noted from Fig. 6:







The Auckland wind-tunnel spectra match the full-scale spectra in the high-frequency
range.
At both full-scale and model scale, the vertical (w) spectral density becomes much
smaller than that of longitudinal (u) or transverse (v) components at reduced frequencies
below 0.3.
The wind-tunnel longitudinal (u) and transverse (v) spectra are much smaller than the
full-scale spectra at reduced frequencies below 0.03.
Both cospectra, which indicate the frequencies contributing to the uw Reynolds shear
stress, have their peaks at about f 0.1 and are quite small below f 0.01.
In spite of the signicant differences in the u spectra, the two cospectra are very similar.

It appears that although the wind-tunnel model has not matched the low-frequency end
of the full-scale spectra, it has matched the medium to high-frequency bands and has hence
been able to reproduce the 3D turbulence effects which result in the uw Reynolds shear
stress. The missing turbulence is primarily horizontal and has large effective length scales.
In the wind-tunnel, a reduced frequency of 0.03, at a height of 0.075 m, means that such
frequencies are associated with longitudinal length scales of the order of 2.5 m, which is
slightly larger than the 1.8 m width of the tunnel. It is therefore not surprising that
uctuations at reduced frequencies below 0.03 are relatively suppressed. In full scale, at a
height of 3 m, the longitudinal length scale associated with f 0.03 is 100 m, which can
easily exist but will be constrained to be primarily horizontal by the ground.
The nature of the low-frequency 2D turbulence has been studied at Silsoe by
simultaneous measurements at heights of 1, 3, 6 and 10 m with sonic anemometers
(Richards et al., 2003). Cross-spectral analysis of the time series showed that both
horizontal components were well correlated for frequencies below 0.01 Hz and that at these
frequencies the spectral density was proportional to the mean velocity squared. This
indicated that these low-frequency uctuations were affected by processes similar to those
creating the mean velocity prole and hence are effectively low-frequency uctuations in

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1392

the mean wind speed and direction. Cross-correlation analysis between pressures on the
Silsoe Cube and the dynamic pressure at the reference anemometer shows high correlations
for frequencies below 0.01 Hz. Further quasi-steady predictions of the pressure
uctuations, which account for both changes in wind speed and direction, at these low
frequencies closely match the measured pressures.
In summary, it should be noted that an approach has been taken where the wind-tunnel
model reasonably matches the velocity prole but not the turbulence intensities. The
measured turbulence spectra do match full scale in the high-frequency end of each
spectrum but do not include all of the low-frequency uctuations observed in full scale.
A similar approach has previously been taken by Irwin (2004), who reports using a partial
simulation approach in studies of bridge decks. In that study, the vertical turbulence was
normalised using the mean velocity, only the high-frequency part of the spectrum was
matched and the turbulence intensity was less than half the full-scale value. In the
following sections, the impact of this approach on the measured pressures will be
considered.
5. Mean pressure distributions
In Section 2, the European variation in vertical centreline mean pressure distributions
was attributed to the differences in high-frequency turbulence. Since the University of
Auckland wind-tunnel tests have high-frequency turbulence levels similar to those at
Silsoe, one may expect the pressure distribution to match.
Fig. 7 shows that in general there is considerable similarity between the wind-tunnel
and full-scale vertical centreline mean pressure distributions at both 901 and 451. In this
and subsequent gures, the various Test cases represent experiments carried out with the
corner pressure taps oriented in various ways. It does appear that these roof tappings have
some affect on the ow over the roof. In Fig. 7(a) the two wind-tunnel tests, one with the
corner taps to windward and the other to leeward, produce distributions that are similar to
the 01 and 51 tilted cube results shown in Fig. 3(a). This suggests that the positioning of the
corner roof taps may be slightly modifying the ow reattachment behaviour and hence
the mean pressure distribution.
In Fig. 8, the wind-tunnel results from the various tests have been combined to give
contour maps for the complete roof. These contour diagrams do not include the edge strip

b
1.0
Full-scale
Test A
Test B

0.5
0.0
0.0

0.5

1.0

1.5

2.0

2.5

-0.5
-1.0
-1.5
Distance OverCube (Cube Heights)

3.0

Mean Pressure Coefficient

Mean Pressure Coefficient

1.0
Full-scale
Test A
Test B
Test C

0.5
0.0
0.0

0.5

1.0

1.5

2.0

2.5

-0.5
-1.0
-1.5
Distance OverCube (Cube Heights)

Fig. 7. Vertical centreline mean pressure distributions at (a) 901 and (b) 451.

3.0

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1393

a
90

75

Key for meanCp


contour plots
-0.4 to -0.2
-0.6 to -0.4
-0.8 to -0.6
-1.0 to -0.8
-1.2 to -1.0
-1.4 to -1.2
-1.6 to -1.4
-1.8 to -1.6
-2.0 to -1.8
-2.2 to -2.0
-2.4 to -2.2
-2.6 to -2.4

Key for meanCp


contour plots
-0.4 to -0.2
-0.6 to -0.4
-0.8 to -0.6
-1.0 to -0.8
-1.2 to -1.0
-1.4 to -1.2
-1.6 to -1.4
-1.8 to -1.6
-2.0 to -1.8
-2.2 to -2.0
-2.4 to -2.2
-2.6 to -2.4

c
60

45

Key formeanCp
tour plots
-0.4 to-0.2
-0.6 to-0.4
-0.8 to-0.6
-1.0 to-0.8
-1.2 to-1.0
-1.4 to-1.2
-1.6 to-1.4
-1.8 to-1.6
-2.0 to-1.8
-2.2 to-2.0
-2.4 to-2.2
-2.6 to-2.4

Key for meanCp


contour plots
-0.4 to -0.2
-0.6 to -0.4
-0.8 to -0.6
-1.0 to -0.8
-1.2 to -1.0
-1.4 to -1.2
-1.6 to -1.4
-1.8 to -1.6
-2.0 to -1.8
-2.2 to -2.0
-2.4 to -2.2
-2.6 to -2.4

Fig. 8. Roof mean pressure contours for wind directions: (a) 901, (b) 751, (c) 601 and (d) 451. In each case, the
diagram to the left is the combined wind-tunnel result and to the right is the full-scale results for windward quarter
of the roof.

1
Full-scale
Test A
Test B

0.5
0
0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

-0.5
-1
-1.5
Distance Around Cube (Cube Heights)

4.0

Mean Pressure Coefficient

Mean Pressure Coefficient

1
0.5
0
0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

-0.5
Test A
Test B
Test C
Full-scale

-1
-1.5

Distance Around Cube (Cube Heights)

Fig. 9. Mid-height mean pressure distributions at (a) 901 and (b) 451.

which is 0.066h wide. Both the wind-tunnel and full-scale results show a similar evolution
of the contours with direction and similar ranges of pressures occurring in each case.
In Fig. 8(a), with a 901 wind direction, the most negative mean pressures lie in the 1 to
1.2 range, whereas in Fig. 8(d), for a 451 wind direction, mean pressure coefcients in the
2 to 2.2 range were recorded both in full-scale and the wind-tunnel.
More obvious differences between full-scale and wind-tunnel were observed with the
mid-height ring of tappings. Fig. 9 shows that there are noticeable differences on the
sidewalls with a wind direction of 901 and on the leading edges of both windward walls
with a 451 wind direction. Unfortunately, the wind-tunnel results do show that the model
was not exactly perpendicular to the wind for the 901 case. The effects of this can be seen in

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1394

the asymmetry of the windward wall pressure distribution and in the difference between
the two sidewall distributions. Both of these results suggest that the actual wind angle was
slightly less than 901. This results in a slightly more positive pressure between position 0
and 0.5 than between 0.5 and 1. The change in sidewall pressures is again similar to that on
the roof of the cube when it was tilted forwards. The wind approaches the sidewall 12 at a
slightly bigger angle and hence the ow will be slightly more separated, resulting in a atter
distribution with less suction at the windward end and slightly more at the leeward end. At
the same time, the lower angle of attack for sidewall 34 means that the ow is slightly
more attached and has a higher maximum suction about one-quarter of the way across the
face and then rapid recovery. The misalignment is thought to have affected all the data,
since the 451 results in Fig. 9(b) also show some asymmetry. Fig. 10 shows the effects of
altering the angle in 151 steps. Comparing the asymmetry of Fig. 9 with the data in Fig. 10
suggests that the misalignment was of the order of a few degrees.
Although the difference in sidewall pressure distributions in Fig. 9(a) may be simply
caused by incorrect modelling of the ow reattachment on these walls, this cannot explain
the differences in Fig. 9(b). With a wind direction of 451 it might be expected that the ow
is primarily attached to both windward walls. However, Fig. 10(b) shows that the pressures
on these walls are highly sensitive to wind direction, for example, for locations nearer
position 4, a change of wind direction of only 301 can alter the pressure coefcient from
about 0.7 at 451 to 0.9 at 751.
One of the consequences of the lower turbulence intensity in the wind-tunnel is a lower
standard deviation of wind directions. Fig. 11(a) shows typical examples of the distribution
of wind directions in the wind tunnel and in full scale. At the Silsoe site during a typical
12-min run the standard deviation of the wind direction was around 101, whereas in the
wind tunnel it was only 5.61. If the ow eld responds to these direction changes in a quasisteady manner, then the observed mean pressures will be weighted averages of the values
associated with particular wind directions. In the wind-tunnel, the range of wind directions
is approximately 7151 around the nominal value, so when the nominal wind direction is
451, it may be expected that the pressure on most of the windward faces of the cube would
remain positive at all times. In contrast, in full scale the range of wind directions is about
7301, and so at times the leading edge pressures may become quite negative as a
consequence of the instantaneous wind direction swinging around to 751 or 151. This can
be seen to be the case for Tap 17 in Fig. 12(a), where at 451 the wind-tunnel peak minimum

Mean Pressure Coefficient

1
0.5
0
0

0.5

1.5

2.5

-0.5
-1
-1.5
Distance Over Cube (Cube Heights)

Mean Pressure Coefficient

1
0.5
0
0

0.5

1.5

2.5

3.5

-0.5
-1
-1.5
Distance Around Cube (Cube Heights)

Fig. 10. The effect of wind direction on wind-tunnel mean pressure distributions for (a) the vertical centreline
section and (b) mid-height.

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

b
600

0.08
0.06
0.04
0.02

-20

-10

400
300
200
100
0

0
-30

y = 2.7788x

500
Max Dynamic
Pressure qmax

PDF

0.1

-40

1395

10

20

Wind Direction (degrees)

30

40

50

100

150

200

Mean Dynamic Pressure q

Fig. 11. Comparison of full-scale and wind-tunnel ow properties, at cube height, related to the turbulence levels.
(a) The distribution of instantaneous wind direction during typical runs. (b) The relationship between maximum
dynamic pressure and mean dynamic pressure.

pressure coefcient is only just negative whereas the full-scale peak minimum pressures are
consistently lower. It should be noted that in Fig. 12 the full-scale peak minimum and
maximum pressure coefcients are the ratios of the single most extreme pressures recorded
during a run to the peak roof-height dynamic pressure recorded during the same run. As a
result, there is a longer tail of negative pressures in the full-scale situation and so the
observed mean pressure is lower. This effect will be most signicant for positions such as
Tap 17 (see location in Fig. 4(a)), which at 451 is close to the leading edge of the windward
wall and as shown in Fig. 10(b) has the greatest sensitivity to wind direction.
6. Peak pressures
Changes to the standard deviation of wind direction also affect the distribution of peak
pressures. Richards and Hoxey (2004) show that with a quasi-steady model extreme
pressures are expected when a high dynamic pressure combines with an instantaneous wind
direction which is associated with either a high or low pressure coefcient. For Tap 17, the
highest mean positive pressure coefcient occurs at about 651 whereas the lowest occurs at
about 51. With mean directions around say 301, the expected maximum and minimum
pressure coefcients will depend on the likelihood of occurrence of instantaneous
directions of 651 or 51, respectively. This is illustrated in Fig. 12(b) where the measured
wind-tunnel mean pressure coefcient distribution has been combined with both a narrow
band of wind directions (standard deviation 51) and a wide band (standard deviation 101)
to give two sets of quasi-steady maximum and minimum pressure coefcients. For both
maximum and minimum pressures, the higher standard deviation of wind directions leads
to a broadening of the range of mean directions where high extremes are expected.
Fig. 12(b) also shows the wind-tunnel maximum and minimum coefcients derived from
Lieblein analysis (Cook, 1985). In general, the wind-tunnel extremes are closer to the
narrow quasi-steady model, which is appropriate since the wind-tunnel standard deviation
of wind directions was 5.61.
This broadening of the mean wind direction bands is also apparent in Fig. 12(a) where
the full-scale bands are broader than those from the wind tunnel. Both the full-scale and
wind-tunnel results show that in the ranges 101 to 301 and 1701901 the measured

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1396

Pressure Coefficient

3
Cp max WT

Cp mean WT

Cp min WT

Cp max FS

Cp mean FS

Cp min FS

-3
0

45

90
135
180
225
270
Mean Wind Direction (degrees)

315

360

Pressure Coefficient

3
Cp max WT

Cp mean WT

Cp max QS Narrow

Cp max QS Wide

Cp min QS Narrow

Cp min QS Wide

Cp min WT

-3
0

45

90
135
180
225
270
Mean Wind Direction (degrees)

315

360

^ q),
^ peak minimum (p=
 q)
^ and mean (p=q) pressure coefcients from full-scale and
Fig. 12. (a) Peak maximum (p=
wind-tunnel data for Tap 17. (b) The wind-tunnel data together with quasi-steady expectations for the peak
maximum and minimum pressure coefcients with either a 51 (narrow) or 101 (wide) standard deviation of wind
directions.

extremes are slightly larger than would be predicted by a quasi-steady model. This is
probably due to building-induced turbulence being created in the ows that are separating
and reattaching to the sidewalls at these angles.
The form of Fig. 12, with the peak pressures ratioed to the peak dynamic pressure,
partially masks the fact that at cube height the ratio of maximum dynamic pressure to
mean dynamic pressure is higher in full scale than in the wind-tunnel. Fig. 11(b) shows that
in full scale this ratio has a broad range of values when the wind is light, but is consistently
near 2.78 in stronger winds. In comparison the wind-tunnel ratio is only 1.91. These are
both close to that expected if the wind speed is normally distributed. The sought peak
value has a probability of the order of 1 in 3000. For a normal distribution, this occurs

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1397

3.4 standard deviations above the mean. Hence, the expected peak-to-mean dynamic
pressure ratio is given by
q^ 1 3:4I u 2

,
q
1 I 2u

(1)

which with the typical full-scale turbulence intensity Iu(h) 0.21 gives a ratio of 2.8,
whereas in the wind-tunnel with typically Iu(h) 0.11 the expected ratio is 1.86.
With a higher peak-to-mean dynamic pressure ratio in full scale, it may be expected that
the ratio of peak pressure to mean dynamic pressure would also be greater. This is
illustrated in Fig. 13(a) for roof tapping 6 (the location of this tapping is marked in
Fig. 4(a)). In Fig. 13(a) both the mean and negative peak pressures are normalised by the
mean dynamic pressure. It may be observed that there is reasonable agreement between the

a
0

60

120

180

240

300

360

Pressure Coefficient

-1
-2
-3
-4
-5
Mean p/Mean q Full-scale

-6

Min p/Mean q Full-scale


Mean p/Mean q Wind-tunnel

-7

Min p/Mean q Wind-tunnel

-8
Mean Wind Direction (Degree)

60

120

180

240

300

360

Pressure Coefficient

0
-0.5
-1
-1.5
-2
-2.5
-3
-3.5

Mean p/Mean q Full-scale


Min p/Max q Full-scale
Mean p/Mean q Wind-tunnel
Min p/Max q Wind-tunnel

-4
Mean Wind Direction (Degree)
Fig. 13. Roof Tap 6 mean and peak minimum pressures. The peak minimum pressures are shown normalised by
either (a) the mean dynamic pressure at cube height for each run or (b) the maximum dynamic pressure at cube
height that occurred during the run.

ARTICLE IN PRESS
1398

P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

full-scale and wind-tunnel mean pressure variations but the full-scale peak pressures are
markedly larger than those from the wind-tunnel. However, as illustrated in Fig. 13(b),
much better agreement is obtained if the peak pressures are normalised by using a peak
dynamic pressure.
It is recognised that there are differences in the methods used to process the wind-tunnel
and full-scale extreme values as well as slight differences in the sampling periods and
sampling frequencies. The differences in data analysis are partially driven by the
circumstances. Hoxey et al. (1996) discussed how in the full-scale situation each record is
statistically slightly different, and so it is impossible to use analysis methods such as the
Lieblein BLUE method (Cook, 1985). In these circumstances, a large number of records
are required in order to provide data on both the typical values and the range. On the other
hand, in the wind-tunnel stationarity can be achieved and so multiple runs and the
associated extreme value analysis are appropriate, whereas to carry out a large number of
runs would be both expensive and unnecessary. Since different methods are necessary, it
makes sense to ratio the peak pressures measured on the building to the peak dynamic
pressure measured in the approach ow, both of which can be analysed in the same way for
a particular testing situation, thus removing the sensitivity of the ratio to the analysis
method. Using peak pressure/peak dynamic pressure ratio also minimises sensitivity to
slight differences in sampling period or frequency provided both the surface pressures and
reference dynamic pressure measuring systems have similar frequency responses.
As illustrated in Fig. 13(b) another advantage of using the peak-to-peak ratio is that it
minimises the sensitivity to low-frequency turbulence. As noted earlier, the primary
deciency in the wind tunnel is the lack of low-frequency turbulence. Full-scale coherence
analysis between roof tapping pressures and the dynamic pressure at the upstream
reference mast show near unity coherence for all frequencies below 0.01 Hz (corresponding
to reduced frequencies o0.01). This high coherence suggests that the ow eld is
responding to these uctuations in a quasi-steady manner. Such low-frequency uctuations
elevate the ratio of the peak to the mean but do not signicantly alter the character of the
ow. Hence, by using the peak dynamic pressure as the reference, the results become far
less sensitive to the level of very low frequency uctuations.
7. Conclusions
In order to compare wind-tunnel turbulence spectra with full scale, normalising
parameters that are independent of the turbulence should be used. One suitable form is to
plot nS(n)/U(z)2 against reduced frequency f nz/U(z), where the normalising parameters
are the mean wind speed (U(z)) and height (z) of the measuring point. Using turbulencedependant parameters, such as the variance and integral length scale, can easily mask
differences.
In situations where it is not possible to model the full turbulence spectra, such as
the large-scale modelling of low-rise buildings, care should be taken to correctly model the
high-frequency end of each spectrum. It is this turbulence that can directly interact with the
local ow eld and modify ow behaviour. This has been illustrated by studying data from
tests conducted in a range of European wind tunnels.
The approach taken at the University of Auckland in wind-tunnel modelling the Silsoe
6 m Cube at a scale of 1:40 was to match the velocity prole and the high-frequency
turbulence as closely as possible. Similar mean pressure distributions were obtained as a

ARTICLE IN PRESS
P.J. Richards et al. / J. Wind Eng. Ind. Aerodyn. 95 (2007) 13841399

1399

result. Although the high-frequency end of each spectrum was matched, the size of the
tunnel limited the low-frequency end and so the longitudinal and transverse turbulence
intensities were lower than in full scale. This has the effect of reducing the standard
deviation of wind directions and hence affects both the observed mean and peak pressures
by reducing the band of wind directions occurring during a run centred on a particular
mean direction.
The reduced turbulence intensities also affect the peak-to-mean dynamic pressure ratio,
which in the Auckland wind tunnel was 1.91 in comparison with 2.78 in full scale.
However, since the missing turbulence is at low frequencies, the peak pressures appear to
reduce in proportion. By expressing the peak pressure coefcient as the ratio of the extreme
surface pressures to the maximum dynamic pressure observed during the run, reasonable
agreement is obtained. It is believed that the peakpeak ratio is a more reliable measure of
peak pressures, since it is less sensitive to spectral differences, measurement system
response characteristics and analysis methods, provided the reference dynamic pressure
and the surface pressures are measured and analysed in similar ways. It is also the
peakpeak ratio that is used in most wind loading codes.
References
Castro, I.P., Robins, A.G., 1977. The ow around a surface-mounted cube in uniform and turbulent streams.
J. Fluid Mech. 79 (pt 2), 307335.
Cook, N.J., 1985. The Designers Guide to Wind Loading of Building Structures, Part 1. Butterworth, London.
Holscher, N., Niemann, H.-J., 1998. Towards quality assurance for wind tunnel tests: a comparative testing
program of the Windtechnologische Gesellschaft. J. Wind Eng. Ind. Aerodyn. 7476, 599608.
Hoxey, R.P., Richards, P.J., Richardson, G.M., Robertson, A.P., Short, J.L., 1996. The folly of using extremevalue methods in full-scale experiments. J. Wind Eng. Ind. Aerodyn. 60, 109122.
Irwin, P., 2004. Bluff body aerodynamics in wind engineering. In: Proceedings of the fth International
Colloquium on Bluff Body Aerodynamics and Applications, Ottowa, Canada, 1115 July 2004, pp. 5157.
Melbourne, W.H., 1980. Turbulence effects on maximum surface pressures a mechanism and possibility of
reduction. Wind Engineering, vol. 1, Pergamon Press, pp. 541552.
Melbourne, W.H., Holmes, J.D., Vickery, B.J., 1997. Wind Engineering Course Notes. Monash University,
Melbourne, Australia.
Niemann, H.-J., 2000. Personal communication of the Windtechnologische Gessellschaft comparative testing
programme results provided in CD Rom.
Ogawa, Y., Oikawa, S., Uehara, K., 1983. Field and wind tunnel studies of the ow and diffusion around a model
cube (a)I. Flow measurements and (b)II. Neareld and cube surface ow and concentration patterns.
Atmos. Environ. 17 (6) (a) 11451159 and (b) 11611171.
Richards, P.J., Baker, C., Hoxey, R.P., 2003. Turbulent event sequencing in the atmospheric surface layer. In:
Proceedings of the 11th International Conference on Wind Engineering, Lubbock, TX, 25 June 2003,
pp. 21012108.
Richards, P.J., Hoxey, R.P., 2004. Quasi-steady theory and point pressures on a cubic building. J. Wind Eng. Ind.
Aerodyn. 92 (1415), 11731190.
Richards, P.J., Hoxey, R.P., Short, J.L., 2001. Wind pressures on a 6m cube. J. Wind Eng. Ind. Aerodyn. 89
(1415), 15531564.
Tieleman, H.W., 2003. Wind tunnel simulation of wind loading on low-rise structures: a review. J. Wind Eng. Ind.
Aerodyn. 91, 16261649.

También podría gustarte