Documente Academic
Documente Profesional
Documente Cultură
WIND LOADS
Actions on Structures
April 1996
O CIB
PREFACE
The Task of the commission is to develop a set of stochastic models for actions which are
mutually consistent and which can be used both in probabilistic design and analysis and as
a basis for deterministic models for actions.
This work will be described in a series of reports having a common general title "Actions
on Structures". There will be one report dealing with general principles which are common
for many kinds of actions and a number of reports each describing a specific action.
This report describes those aspects of windload that are considered to be important in
probabilistic load modelling. It has not been the intention that the report includes detailed
information on wind loads at specific sites since this can be obtained from literature and
windtunnelexperiments. An extensive list of references has been included for this purpose.
This Report seeks to review and summarise current methods of representing Wind Loads
in the design of ordinary buildings and structures. These methods, in many respects, are
also relevant to the design of more complex structures, but full and complete details for
the design of such special structures are outside the scope of this Report.
This report was prepared by E.C. Klaver of the Technical University Delft, Department of
Civil Engineering, taking into account an earlier draft of N.J. Cook, and contributions by
A. Vrouwenvelder, R. Rackwitz and A. Larsen.
Wind Engineering covers a wide field of engineering including the subject of the forces on
buildings and structures, wind flows around buildings (environmental winds) and the
dispersal of pollutants. In this CIB. report we shall be concerned only with the first of
these.
That part of the subject of Wind Engineering concerned with the design of structures to
resist wind forces can be divided into three distinct sub-parts corresponding to the physical
processes involved.
1. The first concerns the mechanisms which produce strong winds and the description
of the characteristics of these winds when blowing over the surface of the Earth.
This part of the subject is treated in Chapters 2 and 3.
The second concerns the forces and pressures which the wind exerts on structures
over which it flows. These forces are fluctuating, not only because the incident
flow is turbulent, but also because of the detailed flows induced by the structure
itself. Since the forces are intimately related to flows over the surface of the
structure, which are in turn determined both by the characteristics of the incident
flow and the shape of the structure, another large body of experimental data is
required to provide comprehensively for the many situations met with in design.
This part of the subject is treated in Chapters 4 and 5.
3. The third part concerns the static and dynamic response of the structure to these
wind forces. In many cases the dynamic part of the response is negligible and
design is then concerned only with ensuring the stability of the structure and its
component parts under the action of the most severe quasi static wind loading
expected in the lifetime of the structure. In a smaller number of cases the dynamic
response of the structure is not negligible. Then the design must achieve control of
the response so as to produce a serviceable structure. Here design may be concer-
ned with the problems of human comfort and tolerance, with compliance with
performance criteria and sometimes with fatigue. In a very small class of structures
the dynamic response is such that it actually alters significantly the forces which
are producing the response and then the designer is dealing with so-called 'Aeroe-
lastic' structures. The subject of structural response is covered in Chapter 6. This is
done by the description of a commonly used engineering model.
2. CLIMATIC BACKGROUND
The energy which produces winds comes ultimately from the Sun which heats the
atmosphere unevenly. This uneven heating occurs at several widely differing scales from a
global scale down to scales of a few metres, and it results in corresponding different
scales of atmospheric circulation. Beginning at the largest global scale, the atmosphere is
warmed more near the equator than near the poles. At the same time, due to the rotation
of the Earth, the angular momentum of the rotating atmosphere is greatest at the equator.
Conservation of mass and momentum is responsible for atmospheric circulation on a
global scale.
The basic circulation pattern in each hemisphere consists of three main cells as shown in
Figure 1 . (The depth of the atmosphere is not shown to scale.) Mean winds at the surface
2.2 Depressions
The instability of the cold and warm air masses along the line of the sub-polar low
pressure bands (Figure 1) gives rise to the frontal depressions which are 500 km to 1000
km in diameter. They are responsible for the strongest winds in these latitudes. The
depressions move with the mean circulation, tracking from west to east, but each individu-
al track is influenced by the position and strength of neighbouring weather systems.
Sometimes their progress is blocked by large stationary anti-cyclones.
Tropical cyclones (hurricanes or typhoons) are typically 200 km in diameter, with the
region of strongest winds around a nearly windless "eye" of between 10 km and 20 km in
diameter. They form over the sea when the surface temperature is in excess of 27" and
where there is a sufficient component of the Earth's rotation to drive the circulation.
Accordingly, these resmctions limit their range to the band of latitudes from 5" to 30"
either side of the Equator. Smaller and more intense than depressions, tropical cyclones
drift from east to west with the Trade Winds, gaining in strength over the warm sea, but
dissipating rapidly over land so that they are a problem only within a 70 km-wind strip
around the coast.
2.4 Tornadoes
Although tornadoes are much smaller than tropical cyclones, they can produce the
strongest and most destructive winds known. They can form when strong high-level winds
blow over the top of strongly convective cells (usually frontal thunderstorms). The high
altitude wind draws air upwards from the core of the cell thus reducing the internal static
pressure. A funnel-shaped cloud grows downward from the cloud-base as the reducing
pressure lowers the condensation level. The air drawn into the tornado at its base causes
strong convergence near the ground, so that the tornado rotates rapidly in the cyclonic
direction. The best known region for the occurrence of destructive tornadoes is probably
the central-southern United States.
Differential solar heating also occurs on much smaller scales due to the different absorpti-
ve characteristics of the Earth's surface. A local hot area may produce enough convection
to cause strong local convergence and the formation of cyclonic circulation. On the
smallest scale this may result in 'dust-devils'. These are not usually considered in
structural design although there is some evidence that they cause damage to small
buildings.
2.6 Thunderstorms
These two forms of wind climate are associated with mountains. A continuous range of
mountains act as a barrier to a dense mass of cold air as it attempts to displace a warmer
air mass. The cold air accumulates behind the mountain range until it is able to pour over
the top, accelerating under gravity to produce strong down-slope or katabatic winds.
These are significant in parts of Switzerland and Italy, along the Rocky Mountains and
other mountainous regions.
3. WIND STRUCTURE
3.1 Introduction
In this report the main subject is windloads on structures. Therefore only strong winds,
with average windspeeds > I0 rn/s are considered.
The majority of wind speed data throughout the world is acquired by measuring surface
winds within the boundary layer by means of anemometers mounted on masts. It is
standard practice to reduce these data to the common base of a standard height above a
standard terrain, usually a height of 10 m above flat or gently undulating open country
with very few surface obstructions (zo = 0.03 m, see 3.2.2). Where possible, the anemo-
meters are sited at this standard height in open country; for exampIe on airfields. In this
case, little or no correction is needed. Otherwise the data are corrected to the standard
exposure using the known characteristics of the boundary layer appropriate to the
particular exposure.
The wind speed data thus acquired contain contributions resulting from the passage of
large-scale weather systems and from the boundary layer itself. A record of the wind
speed will in principle look Iike the diagram shown in Figure 2. It has long term fluctuati-
ons, 1,
associated with the large-scale weather systems and short term fluctuations, g, due
to the turbulence in the boundary layer. The total windspeed can thus be broken down in
two parts. a mean windspeed 1 (averaged over At and in the mean winddirection) and
windfluctuations g = (u.v,w) around that mean. In vector notation:
These two parts can be readily identified from the data record because of the widely
differing physical scales which result in the production of wind speed fluctuations of
V tor
,Time
Time
v,
(b) 10-minute mean wind speed (moving window)
widely differing frequencies. This is best shown by displaying the data in the 'frequency
domain'. The first comprehensive "spectrum" of this type was compiled by Van der
Hoven 111 from wind records taken at Brookhaven, Long Island, NY, USA, and is
reproduced in Figure 3.
0.001 0.01 0.1 1 10 100 1000
Frequency [ 1Iuur ]
From the form of this spectrum it follows that in most cases the assessment of the
magnitudes of wind loads on structures in any particular storm, which depend on the
exposure of the structure (i.e. its situation) and the shape of the structure, can be separated
from the prediction of the severity and frequency of occurrence of the storms themselves.
Macro- and micrometeorological fluctuations (or long and short term fluctuations), for this
reason, may be discussed separately.
When wind speed data are collected over a long period of time (usually at a height of 10
m), these data accumulate to form a statistical distribution in the 'amplitude domain'. The
large body of available data shows that for temperate latitudes generally the distribution of
average wind speeds is very well modelled by the Weibull distribution:
where FV(x) gives the probability of a At-mean wind speed less than x and Vo and k are
parameters defining the scale and shape of the distribution. The shape parameter k ranges
between 1.7 and 2.5 in most temperate areas, but values close to 1 have been measured in
areas of Italy dominated by katabatic winds. The solid line in Figure 4 shows the shape
of the Weibull distribution for the special case of the Rayleigh distribution, for which k =
2, corresponding to the middle of the observed range.
A possible approximation to take this persistency into account is shown in Figure 5(b),
where the windclimate is considered to consist of n independent stationary subprocesses
in a year, each of duration having a At-mean windspeed. The point is that when At is
chosen somewhere between 3 and 12 hours, the mean wind velocity various periods At
may in some calculations be considered as independent.
The windload on a structure depends on the windspeed and the wind direction. For
describing the statistics of the windload it is therefore necessary to take into account the
distribution of wind directions j, commonly divided in 12 sectors. A first approximation
for describing the windvelocities is a varying vector with an intensity V, described by (I),
and a direction statistically described by fraction of time that this vector is within a sector.
10 min
P I
The probability distribution for annual maxima FVm,() can be connected to the parent
distribution Fv() of the At-mean windspeed V as follows:
Unfortunately the value of n in (3) is not always known at a site, so that the distribution
function of Vmax, FVm,(u) is usually directly estimated from observations.
The essence of this method to estimate the annual maximum, is the extraction of indepen-
dent samples from the parent to recover maxima, in this case the maximum wind speed in
each year. However, most sites have fairly short data records, 10 to 50 years, so that the
distribution of annual maxima will only contain 10 to 50 values, insufficient to determine
the shape accurately. In order to make a reasonable estimation, the shape of the dismbuti-
on needs to be determined independently, so that the observations are used to fit only its
parameters. A Weibull parent distribution leads asymptotically to a Type 1 distribution
(Gumbel):
There is some indication that the Gumbel dis~ibution given by (4) overestimates the
probabilities in the very high windspeeds [49]. In case the distribution functions FVm,(x)
and FV(x) are known, the value of n can be estimated from (3).
Notes:
Seasonal and Directional Extreme Value Analysis.
Since the seasonal variations of climate can be averaged over a period of years, the
wind extremes for any one given month should belong to the same parent populati-
on and, being from separate storms, also be independent of the populations for
other months. Thus separate estimates of extreme winds for each calender month
can be obtained by taking all the maxima for a particular month, year by year, and
ranking these in the way already described for the overall annual maxima. This
yields separate estimates for the risk of extreme winds for each month of the year.
Such data are of use for the assessment of design loads on temporary structures,
structures during construction and for occasional structures such as marquees and
pavilions which are erected only for specific sub-annual periods. Since the winter
months are more susceptible to high winds in the temperate latitudes of Europe,
structures erected only for the summer would be significantly over-designed by
using annual extreme winds.
If continuous wind data are available, the basic data can be grouped in sectors with
respect to direction. For practical reasons these are usually 30" sectors. However,
the standard method of extracting annual extremes when applied to individual
sectors is no longer valid because the partition of the number of hours of wind is
different in each sector and also varies more from year to year. Hence there is no
longer a constant annual parent sectorial distribution from which the extremes can
be drawn. Attempts are frequently made to extract directional information from
discontinuous data; for example, from monthly (daily, or storm) maxima for which
the direction is also known - but this is invalid because it neglects the fact that
important data for other directions in the same month (day or storm) are excluded
by the process of recording only the maximum and discarding everything else.
However, the method of using the independent storm data can be applied if the
basic data are continuous; i.e. wind speeds and wind directions are available for
every hour. Each sector is treated separately, and individual sectorial storm maxima
are extracted. Most storms will contribute extremes to several sectors but this
correlation between sectors does not invalidate the method because extremes within
one particular sector will be selected from separate storms and thus remain
independent. The number of extremes in each sector will not be the same. The
average annual rate of occurrence of independent maxima is calculated for each
sector separately; this average may also differ from sector to sector. The CDF of
independent maxima for each sector is raised to the power of the annual rate of
occurrence for that sector thus yielding the CDFs of the independent sectorial
maxima which occur at an average rate of one per annum in each sector. These are
the required distributions for assessing risk within a sector and comparing this with
other sectors.
Rijkoort [49] studied the windclimate in the Netherlands and used subdivisions of
wind data in order to get more or less statistical homogeneous groups of data. He
distinguished day- and nighttime, 6 seasons of two month each and 12 30"-sectors
denoted by 0, 30, etc., where 0 refers to the sector 345"-15". In doing so, he
arrived at 26.12 = 144 groups of data that could be characterised by a frequency
distribution of the form (2) (for daytime) or a variant of (2) (for nighttime). From
there on Rijkoort could describe the windclimate taking into account the correlation
between the hourly mean of windspeed in succesive hours, the period (for instance
one year) and the winddirection. In [49] Rijkoort gives several parent distributions
and extreme value distributions of windspeed (see also 3.2.3.2.).
In the case of a meteorological station in a region where there are two major storm
producing mechanisms, the collection of annual maximum wind speed values and
the plotting of these extremes by the standard Gumbel method results in a graph
represented by the solid symbols in Figure 6. The points will not lie on a straight
line but curve upwards towards the largest extremes. Until recently, the pragmatic
approach was to fit the data to a different extreme value distribution, having a
more slowly decaying upper tail. The Fisher-Tippett Type I1 distribution was a
strong candidate and there were extensive debates about the validity of this
procedure. Simiu and Filliben [12] concluded that attempts to fit other model distri-
butions were not satisfactory. The problem was resolved in 1978 when Comes and
Vickery [13] demonstrated that when a mixed climate is separated into its compo-
nent storm mechanisms, the data from each separate mechanism converge to a
Fisher-Tippett Type I distribution.
The schematic example of Figure 6 shows two populations of maxima. The data
represented by the circular symbols come from one storm mechanism (A), while
those represented by the square symbols come from another (B). In any particular
year the annual maximum for the mixed climate (M) will be the larger of the
maxima from A or B occurring in the same year. The largest values of the mixed
distribution generally come from B and the smallest from A, but values from A
and B are interleaved in the region where the two Type I lines cross, so that the
mixed distribution follows the upper branch of each line.
v-11 v
Reduced variate y= 7
v
Figure 6: Gumbel plot for annual maximum wind speeds in a two population mixed
climate
When winds are produced by any of the climatic mechanisms described in the previous
Chapter, the air movement is resisted by the frictional effects of the rough surface of the
Earth: sea, sand, grass, trees, small and large buildings, and hills and mountains. This
frictional resistance ultimately dissipates the energy in the wind. The surface drag slows
down the air flow near to the ground thus forming a 'Boundary layer' extending from the
surface itself, where the flow velocity is zero, up to a height called the 'Gradient height'
where the ground friction ceases to have any effect and the flow is controlled solely by
the high-level pressure gradients. The depth of this boundary layer is dependent upon the
scale of the surface roughness being deeper for larger scales of roughness. The depth is
typically between 900m and 2000m. However, the wind must blow over a considerable
distance, called 'fetch', before the full depth of boundary layer is established in equilibri-
um with the ground roughness. Each change of roughness disturbs any established
boundary layer and begins the growth of a new boundary layer displacing the old one
from the ground upwards. Thus equilibrium boundary layers are rarely established to their
full depth unless there is a long fetch of uniform terrain. In particular, city centre
boundary layers seldom reach full depth since these need the longest fetch of all to
achieve equilibrium; up to lO0krn in extreme cases. However, in the majority of urban
areas, smaller fetches will usually produce part- depth boundary layers sufficiently deep
to immerse all but the tallest of buildings.
The flow in the atmospheric boundary layer is turbulent, with components in the three
principle orthogonal directions. The range of eddy size extends from the largest which can
be accomodated within the boundary layer of the order of 1000 m down to the small
vortices which whirl leaves about of the order of 1 m and beyond. A cascade is formed
which transfers energy from larger to progressively smaller eddies right down to the point
where the final energy dissipation is at a molecular level in the form of heat. This
specuum of turbulence produces the familiar gustiness of the wind which changes both
speed and direction from moment to moment. The level of gustiness corresponds to the
turbulence intensity, and is greatest in regions of the roughest terrain and least in regions
of the smoothest terrain.
A full-depth boundary layer is not established in case of tornadoes and thunderstorms due
to their small scale relative to the length of fetch needed to establish equilibrium. These
wind producing mechanisms destroy the existing boundary layer in the localised region
that they influence.
A good mathematical model of an equilibrium boundary layer is that due to Harris and
Deaves [14]. The model is based upon the principle of 'asymptotic similarity'. The model
has been verified against experimental data. Harris and Deaves [14] have also extended
their model to cover the transitional conditions associated with abrupt changes of surface
roughness.
Most investigations of wind speeds have been conducted in the lowest 200 m of the
atmosphere since this region is most accessible to measurement. In this layer it has been
found that the systematic change of mean wind direction with height, (the Ekman spiral),
is negligible so that the the mean velocity vector can be considered in terms of its scalar
magnitude (i.e. the mean wind speed).
Log-law Model.
Boundary layer theory indicates that in the lowest layer of the atmosphere a logarithmic
profile law is to be expected. In measurements of mean wind profiles in the atmosphere, it
has been found that over level terrain of uniform roughness, in suong winds (i.e. neutral
stability), the results can be represented by:
where
V(z) is the mean wind speed at height z,
do is the zero-plane displacement
u* is the friction velocity ( = \ l ( ~ p ) )
p airdensity
Z~
shear stress at ground level
K von Karmans' constant (=0.4)
z is the height above ground
In practice u* and zg are measured from a log/linear plot of the observed profiles.
Note: The logarithmic-law model does not apply above the surface layer. Ln order to
extend it higher into the boundary layer an additional term is needed:
where
z
S
is the gradient height zS=u*/6fcor
fcor Coriolis parameter (=lo4 rad/s for temperate latitudes)
KI is a constant in the range between 4 and 7.
With this correction term the profile is accurately represented to a height of about 300 m.
In the full model of the boundary layer developed by Harris and Deaves [14] there are
several more additional terms involving higher powers of z/z
S'
Power-law Model.
For many years the velocity profile has been empirical represented by a power-law model.
where
V(z) is the mean wind speed at height z,
z is the height above ground,
Although a popular model for engineering applications owing to its simplicity, the power
law has two major disadvantages: (1) it has no theoretical justification, and (2) it models
the upper part of the boundary layer quite well, but it is a poor fit near the surface which
is where it is most needed.
The fluctuating parts of the wind u represents the wind turbulence. The turbulence is a
three dimensional rotational air movement resulting from the roughness of the earth
surface. For relatively long fetches and shon periods the turbulence can be described as a
zero mean stationary and homogeneous Gaussian random velocity field. A full description
of such fields requires the definition of a 3*3 mamx of covariance functions (or alternati-
vely spectra), each covariance function depending on the time gap At and the spatial
distance h-. In descriptions of this type usually the assumption of a "frozen isotropic
field" is made (see [53] and annex B). In reality the wind velocity field, however, is far
from istropic.
In this document we maintain the assumption of the "moving frozen field", but isotropy is
not assumed. We shall, however, in this main text not present a full description, but
resmct ourselves to the "autospectra" for the individual components of the vector u. Only
for the inwind velocity, information on the spatial correlation pattern shall be given. For a
more general model. reference is made to Annex B.
Standard deviations
The amount of wind turbulence in respectively the x-, y- and z direction is expressed by
the standard deviations 0, , ov and ow . In first approximation holds (Lurnleyl Panofksy
[541)
which demonstrates the anisotropy of the turbulent wind. The smaller values are valid for
rough surfaces (See Table 1). There is also a tendency to decrease with height above the
ground. Figure 7 gives a more detailled picture of pi as a function of the height.
height z (m)
-
Figure 7a: Standard deviations (oi = piu*) (ref. [58])
3
=:5
CE
L Pu
0
+
0
a j - /--- \
\
-
3 - \
1.5 -
5
t - A
V,,=20m/s; z ,=0. 1m
1 . '
10 20 JO 80 10 320
height z-do in meter
Figure 7b: Standard deviations (o, = piu*) (according to Deaves and Harris)
Turbulence intensity
The longitudinal turbulence intensity I, is defined by:
Approximately holds:
Herein is Sii ( f ) the variance spectrum of the inwind fluctuations in i- direction, defined in
such a way that:
Table 1: Characteristic roughness and profil parameters
*) Eurocode [56]
Note: various authors may give different values for some of the variables.
Von Karman 1101 proposed spectra equations which are generally accepted as the best
analytical representation of isotropic turbulence. The effect of departure from isotropic
turbulence near the ground is allowed for by the variation with height and surface
roughness of the appropriate variance oi2 and length parameter Li since they typify the
intensity and size of eddies constituting turbulence. The von Karrnan spectral equation for
the in-wind each velocity is given by (see Annex B for the mathematical background):
fS,,(f) 4xu
FDu = F D = - 2
= (Von Karman) (13)
ou 2 516
(1 +70.8xu)
The corresponding spectral equations for the lateral and vertical velocities are:
2
fSvv(f) 2xJI +188.6xv)
FDv =- =
2 2 1116 (Von Karman)
ov (1 +70.8xv)
2
fsww(f) -
FDw = - 2
-
2xW(1+188.6xw)
(Von Karman)
Ow (1 +70.8x;)l
Note that the longitudinal length scales vary appreciably up to a factor of 2. The ESDU
formula appears to reflect observations on the lower side and therefore is considered as
conservative with respect to the velocities spectra in the high frequency range.
For the inwind fluctuations u of the windspeed, a number of empirical spectra are com-
monly used. The most popular spectra are:
(Davenport)
(Harris)
(Kaimal)
6.8~
FD(x) = (Eurocode)
(1 + 1 0 . 2 ~ ) ~ / ~
Herein is:
x a dimensionless frequency (fL/Vref)
f the frequency in Hz
L Characteristic length:
Davenport: L = 1200 m, Vref = V10
Harris: L = 1800 m, Vref = VI0
Kaimal: L = 50z, Vref = V,
Eurocode: L = 300 ( ~ 1 3 0 0 ),~for E see table 1. z and L are in [ml
Vref = average windspeed at equivalent height zequ (see [56])
All these spectra are autospectra. They all have the same characteristic at higher frequen-
cies:
This follows from theoretical considerations [52] and is fairly well confirmed by experi-
ments [53], [59].
Figure 8 shows the typical form of the spectra for the inwind, crosswind and vertical
turbulence components u, v and w, taken from [58]. The empirical Davenport spectrum
(20) and the more scientific Von Karman spectrum (13) compare very well. Differences
must be seen in the light of the limitations of the engineering model (Chapter 6) where
these spectra are used.
-
-
von Karman
-+- Davenport
8 %
Harris
'\
- \ , Simiu
*, a * Eurocode
V,t =25m/s
frequency in Hz
-
.
-
-
von Karman
Davenport
Harris
----~t----Simiu
S Eurocode
zO=0.05m
eps=0.26
0.001
0.2 0.4 0.6 0.8 1 1.2 1.4 1.b 1.8 2 3 3 2.4 2.6 1 8 3 2.3 3.4 3.6 2.3
frequency in Hz
Figure 8a.2: Windspectra in terms o f f S(f), logscale, z = 10 m.
-A
von Karman
Davenport
Harris
Simiu
Eurocode
V,, =25m/s
m
'* r
z=90m
~ - .m 2, =0.05m
u
\
j.2 0.4 0.b 0.8 1 1.2 1.4 1.6 1.8 1- 2 2 L4 2 6 2 8 3 3.2 2.4 3.6 3.8
frequency in Hz
t
-
-
-
F von Karman
-
',
P - Davenport
--c- Harris
Simiu
Eurocode
z, =0.05m
eps=0.26
0.001
0.9 0.4 0.6 0.8 1 1.1 1.4 1.6 1.8 2 L2 2 4 2.6 3 8 3 2.1- 3.4 3.6 3.8
frequency in Hz
Figure 8b.2: Windspectra in t e r n of f S(f), z = 90 m, logscale
Cmss spectra
The cross spectrum for windspeeds in the points i and j is mostly written in the form:
Herein is:
The coherence function takes into account the windspeed fluctuations at different points.
Coherence data can be sufficiently accurate represented by a relatively simple exponential
expression of the form:
Herein is
y.z the coordinates
coherence constant z-direction (in the order of 6-10)
Cy
coherente constant y-direction (in the order of 6-16)
windspeed fluctuation u at yi , zi
U.
1
windspeed fluctuation u at y. zj
1
.
In (26) sometimes J[vi(z) vj(z)] is used in stead of V l o . Annex B gives a number of
alternative expressions for coherence functions based on the von Karman model (equations
(15) - (19). In figure 9 examples of these coherence functions are given and compared
with the simplified form (26).
0.S
-
0. - a,
t
I I + ESDU
--t ESDU (appr.)
0.0 - :, -A- E-power
A
8
0.5 - *.
frequency in Hz
-*I
-
m
I )
,
-
+E-power
ESDU
+ESDU (appr.)
\ \
I\.
- '4,
- A \ z, om
Z , =0,1
=50m
'& Z, =40m
-
;\\ \Q
y , =o
Y, =o
-
0.75
- -
_ _
0.85
,
-
-
0.95
-
-
frequency in Hz
A
-
- ESDU
ESDU(appr.)
E-power
8
- A
\
- A
* A
z,=O,lOm
. Z i =50m
-
A
z, =40m
- L,
y i =o
- Y, =o
-
0.05 0.15 0.3 0.35 0.45 0.55 0.6 0.75 0.85 0.95
frequency in Hz
The range of real structures for which the lattice plate model is a good model includes
lattice masts and towers, scaffolding and building frames during construction. The lattice
plate model is also extended to less favourable situations, subject to appropriate correcti-
ons.
Strip model
Two forms of strip model are relevant. The first is obtained when a lattice plate is
elongated so that the plate tends towards a line, as with a tall guyed mast. The lattice
plate model applies, but now only the dimension along the line is relevant when conside-
ring the action of gusts and summation of the elemental forces.
The second form is obtained when an elongated structure is not a lattice, but is solid, like
a tall chimney stack or long bridge deck. The structure sheds a turbulent wake and vortex
shedding is very likely to occur unless steps are taken to prevent it. Independence is
maintained between any plane along the length of the structure, so that the structure may
be divided into short sections or "strips" and the loading of each strip considered separate-
ly. This is a standard technique in the design of road decks for long-span suspension
bridges and is called "strip theory".
Structures of both these types that have their structural properties as well as their
aerodynamic properties concentrated along a line are often called "line-like structures". A
characteristic of line-like structures is that their structural stiffness is low, unless externally
braced or guyed, making them likely candidates for large dynamic response or aeroelastic
instability.
Numerical computation of flow around bluff objects have received a fair amount of
attention over the past decade. The main features of the flow (i.e. mean wind loading on
walls and wake properties) can be calculated with fair accuracy for the 2D as well as the
3D case. Finer details such as the peak suction at the eaves and comers of low pitch roofs
are not yet satisfactorily accounted for. References are [62] to [64] and [67].
where V@ is the mean undisturbed windspeed at point and p is the mass density of air.
Note that in the definition (27) c is considered as a function of time. The characteristics
P
of this function are very complicated. In general it will depend on the point of application
and for higher frequencies the behaviour of p and that of c are not necessarly the same.
P
However, the (statistical) characteristics of c are considered as being independent of the
P
mean wind speed.
The same procedure is used to define the force coefficient cF and the moment coefficient
chq (both dimensionless):
where %ef is a reference area, most conveniently the area under load, and 4ef is a
reference length.
Conventionally, the force coefficient in the direction of the wind is called the drag
coefficient, c, , and the force coefficients normal to the wind are called lift coefficients,
cL . both terms having been borrowed from aircraft aerodynamics.
4.3 Models for building aerodvnamics
It is convenient to model the range of structures by these three categories of their response
to the fluctuating turbulent wind forces:
The vast majority of all structures are static and are designed for strength with a check on
maximum deflection and, possibly, on low-cycle fatigue. A proportion of the taller
buildings, tall chimneys and light flexible structures such as masts, towers and long-span
bridges, will be dynamic and are designed for deflection with checks on acceleration (for
comfort of occupants) and on high-cycle fatigue. Line-like structures are the most
susceptible to aero elastic effects, typically tall chimneys, suspension bridges and other
similar specialised structures.
5 THE BLUFF BODY MODEL
The windloading (p, F. M) on the structure will be described by the At-reference wind-
2 a loading coefficient c (cp , cF , cM ) defined by (see also 4.2.):
pressure q=0.5 p ~ and
Due to windspeed fluctuations within a period At the load coefficient will also fluctuate. A
maximum (minimum) loading coefficient can now be defined by
where pE , FE and ME are relative maxima (minima) taken from a continuous wind loa-
dingrecord (load effect p, F, or M).
Figure 10 shows typical results for measurements of pressure coefficients for minima
(peak suctions) of durations Is, 4s, and 16s. This shows that a fit to a Fisher Tippett Type
I distribution is excellent with observation periods as short as ten minutes. We assume
here that this also holds for the maxima (peak pressures).
0.0
-0.5
d
C
.$ -1.0
-a
.-
0
-1.5
?!
7
ln
ln
2? -2.0
P
f
E -2.5
.-
C
5
-3.0
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Reduced variate y : a(V-U)
The distribution of the relative maxima (minima) of the loading coefficients CE,] .
( c ~ ~ , ~
c ~ or ~c , ~ )~ in
~ a stationary
, ~ subprocess of duration At and winddirection j can thus be
expressed by:
Herein is:
UCE,] and acEj can be determined for instance in a windtunnel as a function of the flow
direction with respect to the orientation of the structure.
Consider now the At reference wind pressure q; and describe the upper tail of the
probability density function, fq,,(x), of the At-mean reference pressure ( 0 . 5 ~ in
~ wind
~ ) di-
rection j, by the following Gumbel distribution:
Herein is
where
p, factor depending on the distribution of wind directions; pj=1/12 (12 wind directions
are considered)
Note:
For fg,,, (T=l year) measurements in the Netherlands (At= 1 hour and wind directions
not distinguished) give values of the order of l/aq = 70 ~ / m U,
* = 325 ~ / [50].
m ~
The probability that the load effect F=cE q exceeds the value x in a period At follows now
from:
or:
Let us now assume that (the upper tail of) FF(x) resembles a Gumbel distribution,
characterised by UF (the modus) and aF (a measure of the dispersion), derived by for
instance a least square method:
The modus of the load effect with a reference period of R year follows then from (remem-
ber that At=l/n year):
For describing the forces (moments) due to wind just the same approach as given above
can be followed. For forces (or moments) (36) must be read as the distribution of the
maxima (or minima) of forces (or moments). The expressions (29) and (30) give the
appropriate forms of the force and moment coeficient respectively. The bluff body theory
can play a role to a better understanding of the pressure coefficient c and the factor B to
be used in the engineering model which is described in chapter 6.
The bluff body model is a complete model, however it requires wind tunnel experiments
in order to get information on FCEj(see (33)).
5.2 Codification
with
where Uq and aq are determined from the distribution of the annual maximum of At-mean
wind speed independent of the direction of the wind (In this case At=10 rnin). With the
above given definitions the design windload can be expressed as follows:
herein is
T o calculate the load effect (pressure) due to wind on a relatively small area the following
linearisation is commonly used
The mean b,the variance op2and the spectrum Spp of the loadeffect p (in this case a
pressure p=cq) are then given by:
p, = -1 p c v
2
The pressure coefficient c depends on many factors for instance: point of application, the
inluence of the building itself and of the buildings nearby, the frequency-content of the
wind speed fluctuations etc). Yet a common assumption is still a pressure coefficient
constant in time.
An estimate of the pressure p, with a frequency of exceedence once in At, indicated by pAt
, is given with
Wind
spectrum
1 Aerodynamic
adrniTnce 1
' spectrum
Load
Structural
transfer
function
Dynamic
response
spectrum
For larger areas A loaded by wind (figure 12) the wind effect F is expressed as
with
The fluctuating part of the windload effect for dAj at some point j (j = 1,2) follows from
d F . = pc J.VJ,uJ.iJ.dA.J = (pcJ.VJ.iJ.dAJ,)u,
J (51)
J
Herein is:
with
and
Substituting (53) and (54) in (52) and knowing that Vh is a function of V10 gives:
Note: In the integration (55) the total area of interest is involved. This could be only the
front area (in this case for the pressure coefficient is taken cfrO,pcbxk) or the total area
(front and back =2A). In the last case four coherence functions have to be distinguised
depending on where dA1 and dA2 are taken. (dAl, dA2 both on the front; dA1 on the
front and dA2 on the back; dA2 on the front and dA1 on the back; dA1 and dA2 both on
the back).
where A is the front area. In figure 13 the empirical expression of x() (aero dynamic
admittance) is compared with the joint admittance (57). Hereby is (57) simplified to
.
with i=l; cl=c2=ch ; VI=V2=Vh=VI0 and oul=o,2=~uh
frequency in Hz
Figure 13: Aerodynamic admittance; comparison of (59) and (60).
In case of assessing dynamic structures (figure 14) equation (56) changes into:
with Ha is the aerodynamic transfer function given by (58) and H, is the structural
frequency response function. For a single degree of freedom system holds:
In (64) F, is the equivalent quasi static load on the structure. (For the static case: F=Fv),
f, the natural frequency of the structure and 5 the damping ratio (clc,,). The dampingratio
c includes aerodynamic damping as well: c=c,+c, with 6, as the structural damping ratio
and Ca the aerodynamic damping depending on V and the building mass.
The variance of F, follows from:
where
Figure 14: Mass and spring system loaded by wind.
and
with
I-,.,.
Herein is:
According to (75) the frequency fo depends on several factors such as the dimensions of
area A (b,h). See figure 15 taken from the Eurocode 1, part 2.3 [56] as an example.
It must be kept in mind that due to a frequency depending c the dominant frequency of
the quasi static windload will be influenced.
Figure 15: Expected frequency of gust loading of rigid structures
p = ( b h ~ ' . ~/ L , V = mean wind speed at the equivalent height zeg,L =
length of turbulence at the equivalent height zeq,b = width and h = height.
Given (b+h)/L and p results in an intermediate variable S (see left part of
the figure); S and V/L results in the dominant frequency f, .
The engineering model described above can be seen as a reference model. It is thought,
however, that the approach given in chapter 5 (bluff body model) gives the possibility to
arrive at a better understanding of the pressure coefficient c and the factor P to be used in
the engineering model.
The general equation for the linear elastic behaviour of a structure under mechanical loads
is given by:
In here M is the mass matrix, C the damping matrix, K the stiffness matrix and F denotes
the external load vector; y, y and y are respectively the structural displacement, velocity
and acceleration vector. If, in the case of wind loading, the external load depends on the
movements of the structure, aero- elastic interaction is said to be present. If the structure is
very rigid, the interaction terms in F(t) can be neglected. If not so, a fruitful strategy in
many cases is to separate the various interaction terms as follows:
%
The interaction terms can then be transferred to the other side of the equation:
In general, structural mass and stiffness are large, while structural damping is small. As a
consequence the aerodynamic damping is of most practical significance. If the aerodyna-
mic damping is positive, it will help to stabilize the structure. However, aerodynamic
damping may also be negative, and have an increasing effect on the structural movements.
Problems may occur if the absolute value of the (negative) aerodynamic damping is
greater than the structural damping. In that case the system is highly unstable. Damping is
then only to be expected from nonlinear effects at larger displacements. As a result often
catastrophic damage is to be expected.
The most simple aero-elastic phenomenon occurs during the inwind loading of structures.
If we accept Morrison's approach we may write, as an extension of the load calculation
(48):
Here CD, CM1 and CM2 are the dimensionless drag and inertia coefficients which should
follow from wind tunnel experiments and which can be found in many textbooks and
publications. Vo is the volume of the structural element. The point is that the force in
reality depends on the velocity difference between wind and structure (the relative wind
velocity), rather then on the wind velocity alone. As stated before, the most important
term is the linear velocity term [-p A CD V y], which can be interpreted as a damping
term.
6.3.2. Divergence
Divergence is a purely static form of instability. Consider a beam under constant windloa-
ding as indicated in figure 16. If the beam has a small torsional rotation 0, the wind may
produce a torsional couple per unit length fT, which may be written as:
with
V = mean wind speed
B = width of the beam
6.3.3. Galloping
Galloping is a dynamic form of beam instability in a constant wind stream. Consider the
same beam of figure 16. Lf the beam moves up and down, the relative wind velocity has a
vertical component depending on the vertical velocity of the beam. The resulting vertical
force per unit length of the beam may be written as:
The quotient (y/V) is in fact the relative angle of the wind velocity. We see that the load
now has a pure damping term nature. In most cases the coefficient CG is negative and
stability is assured. However, for some cross sections CG is positive. In those cases, given
that the wind velocity is high enough, a small initial movement will cause large vertical
vibrations.
A clear trend as to the effect of turbulence on galloping and flutter is not eminent.
Experience indicates that turbulence have a slightly destabilizing effect on the flutter of
semi-streamlined bridge deck sections, but a stabilizing effect on H- and prismatic
sections. Galloping of structures composed of square, octagonal and D-type sections is in
general stabilized by the presence of turbulence in the flow. Rain fall on the other hand
has got a pronounced effect on the aeroelastic haviour on the galloping of bridge cables.
Evidence is found in [66].
6.3.4. Flutter
Flutter is a similar phenomenon as galloping, however, the beam performs both vertical
and rotational movements. This requires, among others, that the natural frequencies of the
rotational and vertical vibration modes have the same order of magnitude. Flutter may
occur for slender cable stayed bridges. The at that time in civil engineering unknown
flutter mechanism destroyed the Tacoma Narrow bridge in 1940.
Vortex shedding is a phenomenon that is a particular important for tower and chimney
type structures. Due to the development of a Van Karman vortex trail, these structures
may suffer from loads in directions perpendicular to the direction of the wind velocity (see
figure 17). This loading (per unit height) can be denoted as:
1
fy(z,t) = -~v(z)' D CL sin(2xfWt + cp) with f, = SV(z)/D (82)
2
Figure 17: Vortex shedding.
The coefficient CL depends on Reynolds number and the surface roughness; cp is a phase
angle and fw the vortex frequency, S is the Strouhall number which equals 0.2 for circular
cross sections and may vary between 0.1 and 0.2 for other cross sections. As long as the
structure reacts as a star object, the loads are relatively harmless: the vortices leave the
structure in small packages with a thickness of 2 to 6 times the diameter D; the various
packages need not to be vertical, need not to be in phase and may even have different
frequencies due to the presence of a vertical wind profile and turbulence, so cp and fw are
time dependent functions of the vertical coordinate z.
However, if the structure starts vibrating, the movements of the structure tends to have a
coordinating effect on the packages: all packages may then show a preference for the
frequency of the vibrating system (lock in). Given the fact that the structure has a
preference for vibrating in its natural frequency, this effect will lead to a frequency fw(z) =
f, over the total height of the structure. For this to happen, the mean wind speed has to be
approximately equal to the critical wind speed V,, = DfJS. In addition, while increasing
the amplitude of the vibration, all packages will show a preference for the same phase in
time and space. This will lead to an increase in the "effective" load. It is often assumed
that the effective load is proportional to the vibration amplitude:
As the preferred phase angle is such that the load is in phase with the velocity, we may
also write:
Of course this load reaches a maximum as soon as all packages leave the structure
periodically as one coordinated column over the complete height.
Note that in (84) damping (aerodynamic, structural) is not yet taken into account. Note
further that the lift coefficient CL depends to some extend on ymax. For further informati-
on see [67].
In structural analysis the wind velocity itself is widely recognised as a random variable.
However, in the total model of the wind loading on a structure many more uncertainties
are present. Some of these are of the "parameter uncertainty type", some of the "model
inaccuracy type". Examples are "the value of the pressure coefficient C" and the "assump-
tion of sufficient fetch and a homogeneous wind field" respectively.
Systematic research into all these uncertainties has never been made. In the following
table a number of the main model parameters have been listed and some subjective
quantification mainly based on engineering judgement have been given. In most cases a
lognormal distribution seems to be the most appropriate.
Parameter CoV
roughness z, 0.10
Van der Hoven I. Power spectrum of horizontal wind speed in the frequency range
form 0.0007 to 900 cycles per hour. Journal of Meteorology, (1957) 14, 160-164.
Fisher RA and Tippett LHC. Limiting forms of the frequency distribution of the
largest or smallest member of a sample. Proceedings of the Cambridge Philosophi-
cal Society, (1928) 24, 180-190.
Gumbel EJ. Statistics of extremes. New York, Columbia University Press, 1958.
Shellard HC. Tables of surface wind speed and direction over the United King-
dom. Meteorological Office, Met 0.792. London, HMSO, 1968.
Cook NJ. The designer's guide to wind loading of building structures, Part 1.
London, Butterworths, 1985.
Cook NJ. Towards better estimation of extreme winds. Journal of Wind Enginee-
ring and Industrial Aerodynamics, (1982) 9, 295-323.
Simiu E and Filliben JJ. Probabilistic models of extreme wind speeds: uncertain-
ties and limitations. Proc. 4th Internat. Conf. on Wind Effects on Buildings and
Structures, Heathrow, 1975. Cambridge, Cambridge University Press, 1977, pp
53-62.
Gomes L and Vickery BJ. Extreme wind speeds in mixed climates. Journal of
Indistrial Aerodynamics, (1978) 2, 331-344.
Harris RI and Deaves DM. The structure of strong winds. Wind Engineering in
the Eighties. London, Consauction Industq Research and Development Associati-
on, 1981.
BSI. Code of basic data for the design of buildings: Chapter V Loading, Part 2
Wind loads. London, British Standards Institution, 1972.
Rice SO. Mathematical analysis of random noise, PtIII. Bell System Technical
Journal, (1945) 19, 46-158.
Davenport AG. Note on the distribution of the largest value of a random function
with application to gust loading. Proceedings of the American Society of Civil
Engineers, Structural Division, (1964) 28, 187- 196.
Davenport AG. Gust loading factors. Proceedings of the American Society of
Civil Engineers, Structural Division, (1967) 93, 11-34.
Baker B. The Forth Bridge. Engineering. (1884) 38, 213-215 and 223-225.
Irminger JOV. Nogle forsag over trykforholderne paa planter og legemer paavirke-
de af luftstrflmniger. Ingeniaren, 1894 17 and 18.
Jensen M. The model law for phenomina in natural wind. Ingeniaren, (internatio-
nal edition), (1958) 2, 121-128.
Flachsbart 0. Model experiments on wind loadings on lattice girder structures.
Single lattice beams in one plane. Die Bautechnik, (1934) 7(9), 65-69.
Cook NJ. Simulation techniques for short test-section wind tunnels: roughness,
barrier and mixing-device methods. Wind Tunnel Modeling for Civil Engineering
Applications. (Ed: Reinhold TA), pp 126-136. Cambridge, Cambridge Univeristy
Press, 1982.
Cook NJ. The designer's guide to wind loading of building structures, Part 2.
London, Butterworths, 1990.
Bergh H and Tijdeman H. Theoretical and experimental results for the dynamic
response of pressure measuring systems. Report NLR-TR F238. National Aero-
space Laboratory (The Netherlands), 1965.
SIA. Actions sur les structures. Norme Suisse, SIA 160. Zurich, Societt Suisse
des Ingenieurs et des Architectes, 1988.
Mayne JR and Cook NJ. On design procedures for wind loading. Current Paper
CP25P8. Garston, UK, Building Research Establishment, 1978.
Cook NJ and Mayne JR. A novel working approach to the assessment of wind
loads for equivalent static design. Journal of Industrial Aerodynamics, (1979) 4,
149-164.
Cook NJ and Mayne JR. A refined working approach to the assessment of wind
loads for equivalent static design. Journal of Wind Engineering and Industrial
Aerodynamics, ( 1 980) 6, 125-137.
Homes JD. Discussion of "A refined working approach to the assessment of wind
loads for equivalent static design". Journal of Wind Engineering and Industrial
Aerodynamics, (1981) 8, 295-297.
Harris RI. An improved method for the prediction of extreme values of wind
effects on simple buildings and structures. Journal of Wind Engineering and
Industrial Aerodynamics, (1982) 9, 343-379.
Gumley SJ and Wood CI. A discussion of extxeme-loading probabilities. Journal
of Wind Engineering and Industrial Aerodynamics, (1982) 10, 31-45.
Rijkoort PJ, A compound Weibull model for the description of surface wind
velocity distributions, W.R. 83-13, KNMI, De Bilt, The Netherlands, 1983
Mann. J., Kristensen, L., Courtney, M.S., The Great Belt Coherence Experiment,
Risb National Laboratory, Roskilde 1991.
Proceedings of the 1st IAWE European and African Regional Conference, Guern-
sey, September 1993.
A recent conference held by CSIRO, Highett, Australia was devoted to the subject
of computational wind engineering.
[67] Simiu, E. and R.H. Scanlan, Wind effects on Structures, John Wiley and Sons,
New York, 1985.
APPENDIX A MEASUREMENT OF LOADING DATA
The earliest measurements of wind loading of stuctures were made exclusively at full
scale. Initially, the aim was to provide reliable values directly for use in design. In 1884,
reporting his experiments made for the design of the Forth Rail Bridge, and with the Tay
Bridge disaster still of major concern, Baker [23] wrote that the design:
Early measurement equipment was quite crude, but ingenuity of approach was certainly
not lacking as when Irrninger [24] measured the wind loads on a large gas holder of the
floating type by tethering the gas holder gainst the wind and measuring the overall drag by
the tension in the tether and the overall lift by the change in gas pressure. Later experi-
ments were progressively better equipped and measurements became more extensive and
accurate. By the 1930s experiments were beginning to be made at model scale in wind
tunnels, and the role of full-scale tests had changed to providing "benchmark" data to
verify developing theories and models. The peak of full-scale experimentation occurred in
the decade 1970-80. A survey [25] in 1974 showed 103 current of recently completed
experiments on a wide range of structures.
Full-scale experiments are very expensive to conduct, especially in comparison with wind
tunnel or numerical modelling, and the number of current experiments has declined
considerably from the peak. Nevertheless, the need for full-scale data always remains
because new theories and analysis techniques must be verified by full-scale measurements
using the same contemporary analysis techniques.
In 1893, just before his full-scale experiment with the gas holder, lnninger constructed
what was probably the world's first wind tunnel by attaching a duct to the base of a 30m
high chimney. The updraft of the chimney drew air through the duct at speeds between
7.5 and 15 4 s . In 1927, in his eightieth year, he began a comprehensive series of tests
on building shapes in collaboration with Npkkentved which took a decade to complete [26,
271. Unfortunately, all these data were collected in smooth uniform flow and it is now
known from the later work of Jensen [28] are that results from wind tunnels without a
good simulation of the atmospheric boundary layer are very different from what actually
occurs in nature. The only good data that survive from this time is the work by Flachsbart
[29, 30, 311 on lattice frames in uniform flow, because of the properties of the lattice plate
model described earlier in nr. 4.1, these data can be transcribed to apply the the turbulent
atmospheric boundary layer.
The early work in wind tunnels was done in tunnels designed for aircraft tests which
produced smooth uniform flow everywhere in the test section, except for a very thin
turbulent boundary layer next to the tunnel walls and floor. The effects of these thin
boundary layers confused early researchers, since each wind tunnel produced a floor
boundary layer of a different thickness, it was found that the same building model tested
in each wind tunnel produced different results. Unique among the early data is the work
by Bailey and Vincent [32], who found that the floor boundary layer of their wind tunnel
was a good scale model of the atmospheric boundary layer at 11240 and their data
compares well with modern experiments, except that only mean values were measured.
Inappropriate model tests using smooth uniform flow continued well into the 1960s and
most of these data found their way into codes of practice which were still in force in
1990. It was the work of Jensen [28], who demonstrated the effect of surface roughness
on the wind speed profile and its consequent effects on building pressures, that led to the
the accurate scale models of the atmospheric boundary layer in use today.
Most aeronautical wind tunnels are too short to develop a deep enough floor boundary
layer to give a good simulation of the atmospheric boundary layer, so special "bounda-
ry-layer wind tunnels" were developed for this purpose. The first of these tunnels were
very long in order to grow a deep enough boundary layer naturally over scaled surface
roughness. In other laboratories, where access to such long tunnels was not possible,
attempts were made to accelerate the growth of the floor boundary layer, leading to the
contemporary techniques called "roughness, barrier and mixing-device methods" [33].
Development of these methods has continued to increase their accuracy and the trend
towards their use in longer wind tunnels has tended to blur the distinction between
completely natural and accelerated growth methods.
A typical arrangment of hardware is shown in Figure 18. The role of the roughness is the
same as in the atmospheric boundary layer; it represents the roughness of the ground
surface to the correct linear scale. The roughness is the most important in that it establis-
hed the values of the three log-law parameters in Equation (3,
z,, d, and u*. The barrier
and mixing-device are the "artificial" part of the simulation: the barrier giving an initial
deficit of momentum and depth to the boundary layer, which is mixed into the developing
simulation by the turbulence generated by the mixing device which, in this case, is a grid
of slats. The flow is tricked by the barrier into believing the fetch of roughness to be
longer, and by the mixing device into believing the banier is not there at all!
W inN
-dr
I
3Jurbulence grid
Figure 18: Typical hardware used to simulate the atmospheric boundary layer.
The random or regular arrays of roughness elements used in the simulation methods
produce the correct general flow conditions, both mean velocity profile and turbulence
spectra, approaching the site. However, the conditions at a particular site are more
strongly influenced by the local terrain than by the terrain far upwind. At some distance
from the site, it is necessary to change from a general simulation to a detailed representa-
tion of the site. usually called the "proximity model". The construction of detailed
proximity models is time consuming and expensive, particularly for urban areas. If too
small an area is represented, the site conditions will not be accurately reproduced. A
rational approach to this problem is given in reference [34] along with a more detailed
description of the methodology and criteria for quality assurance.
Measurements of wind loading and response at full and model scale are made in four main
areas:
1. Measurements of wind speed and turbulence: the agent,
2. Measurements of surface pressures: the actions,
3. Measurements of forces and moments: the action effects, and
4. Measurement of the response in deflection, velocity or acceleration: the effects.
In full scale, cup anemometers give the magnitude of the horizontal wind vector and
weathercock vanes give the corresponding direction and are the standard for meteoro-
logical stations, resolving gusts down to about Is duration in strong winds. If the compo-
nents of turbulence are required at full scale, propellor anemometers mounted in groups of
three (one for each orthogonal axis) will resolve frequencies up to about 2Hz. Even
higher frequencies, up to 30Hz, may be resolved using ultrasonic anemometers which have
become available in recent years at prices comparable with the older techniques.
Most model-scale methods can also be used at full scale, although the instruments may not
be rugged enough for long-term exposure. The hot-wire anemometer works on the
principle that the heat lost by a fine wire kept at a constant high temperature is a function
of the wind speed normal to the wire. Used in pairs (X-probes), they can resolve the
turbulence components i n a plane which includes the mean wind speed (ie. u with v, or u
with w, but not v with w). Used in threes, they can resolve the instantaneous velocity
vector in all three dimensions. The frequency response of hot-wire anemometers is very
high, up to lOOkHz typically. A completely different approach is used by the pulsed-wire
anemometer: here a sudden pulse of electricity down a central wire heats the air around
the wire which advects with the flow. Another pair of wires, either side of the heated
wire and aligned normal to it, are fast-acting thermometers which detect the pulse of
heated air. The value of the wind speed vector normal to the heated wire is proportional
to the time delay between pulse and detection, and the direction of the flow is given by
whichever of the two detectors sense the pulse. Unlike the conventional hot-wire
anemometer, which gives a continuous analogue signal, the pulsed-wire anemometer is a
digital instrument passing inividual values at a rate up to 50Hz. A recently-developed
optical technique uses two intersecting beams from a single laser to illuminate small dust
particles in the flow. Interference fringes in the region of intersection result in periodic
illumination of the moving particles. The frequency of illumination is proportional to the
velocity vector.
In the earliest experiments at full and model scale pressures were measured using liquid
manometers which were read by eye. One side of the manometer was connected to the
pressure to be measured and the other to a static datum. The simplest manometer is a
U-tube containing water or alcohol, resolving to about 1Pa. The easiest to read is the
"Betz" or projection manometer in which a scale under a float projects the value onto a
screen, resolving to about O.1Pa. The most accurate are the zero-balancing types such as
the "Chattock", resolving to 0.02Pa, but these require the pressure to remain steady while
the manometer is adjusted to balance. Recent developments in transducers have enabled
direct-reading electronic manometers with a resolution comparable to the "Betz", about
0.1Pa.
Pressure transducers are devices which generate a voltage signal in response to applied
pressure, ideally linearly with amplitude and constant with frequency. When the UK
Building Research Establishment embarked on its programme of full-scale pressure
measurements in the early 1960s, there were no suitable transducers in existence so it was
necessary to develop one specifically for the purpose [35]. In essence, the active pressure
was applied to one side of a diaphragm and a steady reference pressure to the other.
Deflection of the diaphragm, proportional to applied pressure, was detected by strain
gauges. Modem transducers working by similar diaphragm principles but exploiting
modern electronics are now widely available, ranging in size and sensitivity, suitable for
measurements at both full and model scale.
At full scale, many pressure transducers are distributed over the surface of the building,
with either the diaphragm flush with the building surface or connected to a hole in the
building surface by a short length of tube. Multiple pressure transducers are required
because the changing weather conditions in storms require simultaneous measurements at
all locations. At model scale, the transducer is generally too large to mount flush with the
model surface, unless the average pressure over a large area is required, so the active side
is usually connected by tubing to a hole in the surface called a "tapping". As incident
wind conditions can be kept constant in the wind tunnel, simultaneous measurements are
not usually necessary and are often made sequentially using a pressure-scanning switch to
connect tubes in turn to a single transducer. In recent years, trandsucer technology has
advanced so that as many as 100 tranducers, each with an individual connection tube on
the active side, can be made on a single silicon chip 25mm square. This has allowed
many measurements to be made simultaneously at model scale.
The use of tubes to duct the pressure from the tapping to the transducer "colours" the
signal in that the frequency response of the tube is not constant, but acts like an organ
pipe imposing harmonic resonances on the signal. A theory was developed in 1965 which
enables the transfer function of any arbitrary tubing system to be calculated [36]. At
about the same time, physical methods of modifying the transfer function of the tubing to
give a constant magnitude and linear phase with frequency (constant time delay) were
being developed and the theory helped in the development and verification of the methods.
The most common method is to insert a restriction about half-way down the tube.
Electronic or numerical correction methods are also possible, but for cost effectiveness a
small brass restrictor in a length of plastic tube is hard to beat!
On most bluff bodies, the normal pressure forces far exceed the shear stresses from
friction, so that overall forces and moments can be reliably obtained by pneumatic or
numerical summation of surface pressures. In 1971 Surry and Stathopoulos [37] proposed
that if a number of pressure tappings were connected by tubing to a common point, the
pressure at that point would be the instantaneous average of the individual pressures,
supporting the case with experimental measurements. More recently, Gurnley [38]
developed the pressure tube theory [36] to account for multiple tubes and verified the
method analytically. With pneumatic averaging a different set of specially spaced tap-
pings is required for each measurement of force or moment. The advent of the mul-
ti-channel pressure transducers in conjuction with microcomputers has enabled numerical
averaging using arbitrary weighted averages to synthesise forces and moments from a
single set of tappings.
The alternative to integration of surface pressures is direct measurement using some form
of balance or dyanamometer. This is the only practical solution when friction forces are
significant, when the structure is a lattice, or there is some other reason that pressure
tappings are unsuitable. A high frequency range is necessary in order to study peak loads
on static structures and load spectra for dynamic structures, and this demands that the
balance be stiff. The majority of balances use strain gauges as the transducers and this
stiffness mitigates against sensitivity. Typically, strain-gauge balances have a range of
three orders of magnitude, so that a 10N range balance will resolve about 0.1N. Transdu-
cers based on piezo-electric crystals are many orders of magnitude stiffer and offer a
major advantage in range and frequency response. For the same 0.1N resolution a
ION-range strain-gauge balance is easy to damage, whereas the equivalent piezo-electric
balance 1391 would have a range of 25kN and be virtually unbreakable.
Three methods are used to measure or infer the dynamic response of models:
1. Rigid models. Dynamic response calculated from force spectra using equations of
motion of building.
2. Semi-rigid models. Response in fundamental modes represented by rigid model of
required mass mounted on gibals and springs.
3. Fully-dynamic models. Models consEucted to behave as near full-scale as
possible, often by using the same materials scaled down precisely.
In the second and third cases the motion of the model is measured in terms of deflection
or acceleration (occasionally, velocity). With semi-rigid models, the motion can be
inferred from strains in the spring system or directly from the underside of the model.
With fully-dynamic models, detection of the motion is more of a problem: accelerometers
may be mounted in the model (and their mass accounted for in the modelling) or the
motion may be detected optically using laser interfereometry or video tracking techniques.
Of these techniques, semi-rigid models were developed first, fully-dynamic models are
extremely expensive and time-consuming, whereas rigid model methods are quick and
cheap. With rigid models the motion is not represented, so that any aerodynamic feedback
is suppressed and this would not be acceptable for aeroelastic structures. However, the
response can be calculated for a range of dynamic characteristics from one set of load
spectra. With semi-rigid models, the support stiffness and model mass must be changed
and the measurements repeated for each set of dynamic characteristics. A separate
fully-dynamic model is required for each set of dynamic characteristics. At first sight, it
might be thought that calculating the response from the load spectra would be less accu-
rate than direct measurements, but the reality is that the semi-rigid and fully-dynamic
models are only physical representations of the equations of motion deduced from
knowledge of the structure, so are inherently less accurate since the physical representation
will be less than perfect.
APPENDIX B THE VON KARMAN TURBULENCE SPECTRUM
Isotropic Turbulence
The three-dimensional velocity wind field g(rJ on a certain point in time, given the
assumptions of homogenenity and normality, is fully described by the covariance-functi-
ons.
q j ( h J = E {ui@ ujQ + hJ) (B1)
ui@ = wind velocity component i at point = {x,, x2, x3]; it is assumed that x, is the
direction of the average velocity field.
Using Fourier transforms the nine covariance functions can be transferred into a set of
nine spectra:
For incompressible isotropic turbulence (that is: o(u,) = o(u2) = cT(u3)) this spectrum can
be written as [53]:
k = ~(k: + 6 +
2
k3 ) = the wave number vector
Von Karman [lo] proposed in 1948 an expression for a function E(k) which fulfils the
theoretically correct behaviour for extreme frequencies, i.e. that for k +0 the energy
spectrum grows from E(0) = 0 with k4 and falls as k-5/3 for large k:
where:
a = a dimensionless constant (- 1.7)
L = turbulence length scale.
T dV - u.2
E = the rate of viscious energy dissipation (=- - - -1
p dz kz
Given the expressions (3) and (4) one may derive the xl-one-dimensional spectral matrix
from:
In the above description the time variabiIity has not been considered. We now assume that
we may convert the space variabiIity into a time variability by means of Taylor's
hypothesis [61] of the "moving frozen field". In that case we have:
It can be shown that for the time spectra the following relation hold:
21~ 2xf
Sll(f) = - Fll(kl) with kl =
v
-I
+a0
4n 2nf
Sll(f) = - F I 1 (kl) with k l = -
v v
I = -
2
I 1
0
Rll(Ax) dAx with Rll(Ax) =
i kl Ax
~ ~ ~ e ( k ~dkl)
In a similar way:
The cross spectra of u , , u2 and u3 for different locations can be defined in terms of the
coherence function. For example:
with p = r/(aP). The autocorrelation function in time is again obtained by making use of
In the atmospheric boundary layer isotropy can at most be assumed in heights above 500
m or in the high frequency range. It is however, possible to adjust the spectra obtained
under isotropic conditions. Anisotropy means that o(ul) + o(u2) + o(u3) and that there is
a non zero cross spectrum FI3(kl) + 0. It is just the reason for the height dependent mean
velocity profile. As a consequence the turbuIent eddies are stretched in the wind direction.
One observes o(ul) > o(u2) > o(u3) , i.e. o(ul)/o(u3) = 2 and o(ul)/o(u2) = 1.25. One
also has to expect that the turbulence length scales are different for the longitudinal and
the lateral directions. Furthermore these length scales as well as the turbulence intensities
should be height dependent.
which demonstrates the pronounced anistropy of the turbulent wind. The small values are
valid for rougher surfaces. Recommended values are collected in table 1 in the main text.
Also, there is (see (5) in the main text):
with K = 0.4 (the Von Karman constant) and zg and do from table 1 in the main text. In
principle any other reference height can be chosen. The longitudinal turbulence intensity is
defined by
According to measurements cr,(z) grows f i s t rapidly with height and decays gradually
above the height of the roughness elements. This dependency usually is not taken into
account and is experimentally not always verified. Even if a, is assumed independent of
The vertical length scale L: = L: is about half of this value and the lateral length scale;^:
is somewhere in between.
In first approximation <iiw> = 0.3 a, a, is found for the cross correlation and there is a
cross spectrum Suw (0. In aeronautics the following formulae is proposed [lo]:
Wei .
d ) = d i d i i i c M~= 1eSQi
vol
ldvOl (C1)
Herein is:
d the displacement function
c the pressure coefficient as function of r, constant in time
and
(C4)
where
with
where
Herein is:
The engineering model described above can be seen as a reference model. It is thought
that the approach given in chapter 5 gives the possibility to arive at a better understanding
of the pressure coefficient c and the factor B to be used in this model.
International Council for Building Research Studies and
Documentation
m ~ c address:
e Kruisplein 25-G
3000 BV Rotterdam
The Netherlands
Tel: +31.10.41 1 02 40
Fax: +3 1 .10.433 43 72
E-mail: secretariat@cibworld.nl
W http:/~cn.arch.ufl.edukib.html
CIB is a world wide network of over 5000 experts from about 500 organisations, who actively
cooperate and exchange information in over 50 Working Commissions and Task Groups covering
These Commissions initiate projects for R&D and information exchange, organise workshops,
WE I is one of these and its broad terms of reference are outlined in the Preface on page 2
CIB Members come from institutes, companies, partnerships and other types of organisations as
well as individual experts involved in research or in the transfer or application of research results.
CIB is an Association that utilises the collective expertise of its membership to foster innovations and
to create workable solutions to technical, economic, social and organisational problems within its
competence.
Details on Membership and Activities are obtainable from the General Secretariat at the address
above.