You are on page 1of 13

JOURNALOF

ELSEVIER

Journal of Wind Engineering and Industrial Aerodynamics54/55 (1995) 657-669

~I/~1~

How can we simplify and generalize wind loads?


A.G. D a v e n p o r t
Boundary Layer Wind Tunnel Laboratory, The University of Western Ontario, Faculty of Engineering Science, London, Ont., Canada N6A 5B9

Abstract
The complexity of wind loading arises from their variations in space and time. The paper discusses three key spatial functions which control the magnitude of responses: the influence lines, the mode shapes and pressure distributions. The last can be efficiently described in terms of orthonormal expansions. The paper describes how the different time dependent loadings the mean, the background and resonant responses - are related to the spatial functions. These tools provide a promising route for generalization and simplification of wind loading.

1. Introduction
Notwithstanding its enormous fascination, wind loading is in fact a parasitic effect, and mostly an obstacle in the way of designing structures for their primary intended use. Without wind, structures - particularly large ones - would probably be a lot easier to design and cheaper. With this in mind, something we should try to do is to make sure that wind loading wherever possible is simple to use (so that errors are reduced) and 9eneral (so that there is a universal and consistent coverage). The aim of this paper is to describe several tools which can systematically lead to simplification and generalization.

2. Orthogonal and orthonormal functions


Let us consider the use of orthogonal and orthonormal functions. By way of an example, let us first consider the description of the mean wind speed with height on a tall mast (McNamara - 1975). Measurements of 10 rain average speed made on a 400 m mast over a period of several weeks are shown in Fig. la. For clarity, only the 100 and and 1200 ft measurements are shown; in fact the measurements were made at ten levels. Normally we assume that the average wind speed follows a power law or Elsevier Science B.V. SSDI 0 1 6 7 - 6 1 0 5 ( 9 4 ) 0 0 0 7 9 - S

658

A.G, Davenport/J. Wind Eng, Ind. Aerodyn. 54/55 (1995) 657-669

logarithmic velocity profile; this may be a satisfactory assumption for the lowest layers, but in considering a very tall mast or building this may be modified by thermal stability of storm systems passing by. How do we describe a wind speed profile which is constantly varying without getting overwhelmed by numbers and details? One approach is the use of orthonormal or orthogonal functions - a rather indigestible name for a very useful idea. We proceed as follows: first we form the covariance matrix between the wind speed at every height with the wind speed at every other height, as shown in Fig. lb. (The covariance is the average sum of the cross product of the windspeeds at each time step.) The rows and columns correspond to the ten heights on the masts.

-eo

.....

1200

FT.

300

FT.

;'

:,.

~,

.,0" .,0, -" .,,J

:~ '

! ~"~V ~ ~. '~'

)~

i; ~,'! ~

~ i~. [ I
'

,~:~'

;"

'~,,'~. ' ' ~,.~ '~,,


: I

.~..~ ' ~'i : i;


".

't ; ~r,i !

..t~h~ "': t' ;

t " "!

Till[

IN OIY$

~N

i
b
126.0 132.8 163.3 201.4 222.3 230.7 242.9 241.1 249.4 246.8 132.8 159.7 196.6 243.3 270.2 280.2 294.8 292.4 302.3 299.1 163.3 196.6 252.1 309.9 347.2 359.8 378.0 374.6 38Z0 382.5 201.4 243.3 309.9 410.0 444.2 460.4 483.4 478.8 494.5 488.8 222.3 270.2 34Z2 444.2 522.1 541.1 568.5 562.9 581.5 574.4 230.7 280.2 359.8 460.4 541.1 565.9 595.2 590.0 609.7 602.7 242.9 294.8 378.0 483.4 568.5 595.2 632.7 625.1 647.1 640.1 241.1 292.'4 374.6 478.8 562.9 590.0 625.1 626.6 645.7 639.0 249.4 302.3 387.0 494.5 581.5 609.7 647.1 645.7 676.2 665.0

/
246.8 299.1 382.5 488.8 574.4 602.7 640.1 639.0 665.0 660.8

ni

CO VARIANCE MA TRIX. R(zm.Zj) FOR PROFILES OF WIND SPEED

Fig. 1. (a) Record of 10 min average wind speeds and directions Cedar Hill. (b) Covariance matrix for profiles of wind speed (units: mph2). (c) Orthonormal expansions for wind speed. (d) Probability distribution of wind speed amplitude of the first two orthonormal functions.

A.G. Davenport/J. Wind Eng. lnd. Aerodyn. 54/55 (1995) 657-669


ONE

659

ON( 2

ON(

ONE 4,

3 t n Ca0 -0.1

6 1-0.99
0-

-0.5
-1 - 0.9

Rayteigh c = 44.85

"2-

~.Z = 1.0
' 0.5,_,

~ ' -3^

~ 2 = 5.9
~__normat probabitlty taw

a..
~-4I
.m. C

v
o_

-0.99 kw-2.07 Cw= 63.6

- -5-6I

]'0.1

"001 ^'

10

20 50 10~0 coeff, a t

400

-15

-10

-5

0 coeff a 2

10

15

Fig. 1. C o n t i n u e d .

If we invert this covariance matrix - as we do in finding modes of vibration for example - we end up with the eigen functions like code shapes), which are in fact orthogonal to one another. If their mean square amplitude is scaled to unity they are

660

A.G. Davenport/J. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669

known as "orthonormal shape functions or expansions - ONE's". The first four of these functions are shown in Fig. lc. These functions have the useful property that we can express the wind speed at timer t and height z as

V (z, t) = al (t)c~l (z) + a2(t)C~z(Z) + . . . .

(1)

The great advantage of the orthogonality is that the statistics of the functions are in "the mean square sense" independent of one another. This is exactly true if the process follows a normal distribution. Knowing these functions, the original record can be reconstructed in terms of the shape functions. The amplitudes at each time step are found by multiplying through by each of the shape functions in turn and integrating over the heigh of the mast,
l

ai(t) = ~ V(z, t)c~i(z) dz.


0

(2)

With a finite number of velocity readings we replace the integral with a summation over the ten instrumented heights. This gives us a continuous series of amplitudes at(t) which when multiplied by the shape functions reconstruct the complete velocity history. This is already a more compact, simpler representation. What makes it even simpler is that the higher order shape functions have progressively smaller amplitudes and after two or three terms they are no longer significant. The statistics of the amplitudes of the first two shape functions are given in Fig. ld. This indicates that the lowest orthonormal function for wind speed has a Rayleigh distribution (as might be expected), and the second (and higher) orthonormal functions are Gaussian. The relative amplitudes indicate the rapid decrease in importance. The first term captures roughly 95% of the "energy" of the fluctuations. These orthonormal functions in fact have the property that the first extracts the maximum amount of "energy" in the original velocity record. The second is the maximum energy in the remainder and so on. There are other orthonormal functions but these are the most "efficient" in extracting energy. These descriptions can also be applied to the pressure fields. One example of this is in the description of peak local pressure coefficients on an ensemble (population) of tall buildings - about 100 building facades overall. Here the covariance matrix of the pressure coefficients is assembled over a grid of points projected onto each facade and the orthonormal functions defining the pressure coefficients extracted. The lowest order terms then yield the approximations to the pressure field. A further application is to describing the instantaneous pressure patterns over a structure or portions of it. This has been applied by Ho, Surry and Davenport to roof pressures on low buildings (see Fig. 2), by King on suspension bridge design, by Vickery to the roof of a large stadium and Inculet to building facades. This latter is ideally suited to the solid-state pressure systems involving several hundred simultaneous measurements of pressure. This system replaces previous techniques using pneumatic averaging and porous surfaces for determining spatial

A.G. Davenport/.L Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669

661

%(,)

~,~

,(z)

~2j

~(*)

,,3j

~(~)

+...

={,
Amplitude:
Mode 1 Mode 2 Mode 3

Fig. 2. O r t h o n o r m a l shape functions for pressure of a roof of a low building.

effects of dynamic pressure fluctuations used by Surry, Stathopoulos and Davenport, by Rosales and Sato, and Davenport.

3. Influence functions

The second tool I would like to refer to is the use of influence functions. These functions are well known but in practice are only used in specific applications. They deserve wider usage since they provide a direct link between the load pattern and the response. A variety of influence lines are shown in Figs. 3a-3c. Several of these have sign reversals which strongly suggest the susceptibility to "patch" loading due to gusts. (The following discussion indicates how the magnitude of these unbalanced loads can be determined.) The structural engineer is interested in a wide variety of structural responses ranging from member stress to bending moments and shears, as well as deflections. We will designate all of these responses as r. They can be related to the wind force through the appropriate influence line or surface denoted it(z), defining the magnitude of the response r due to a wind load at point z. Referring to Fig. 3, the mean reference, f, can then be written, = q f [Cfl,(z)] dA.
A

(3)

In the latter A represents the reference area over which the pressures act, and z is a point in it. The integral reflects the interaction of the force distribution with the influence function for the response. This quite general, straightforward procedure for

662
WIND SPEED

A.G. Davenport/J. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669


FORCE

(q

' p UZ 3.)

"~1 DH ~ "

INFLUENCE

LINES

,,M, OOE SHAPES I 2

SHEAR MOMENT

I
A(z)
(z) or m(z) ;~(z) i,o(Z ) ~l(z) ~2

J
WIND PRESSURE CROSS SECTION INFLUENC FUNCTION MODE SHAPES

Aj
C

B ~
BENOING MOMENT AT B SHEAR FORCE BELOW B

I~,

GUY REACTION AT B

Fig. 3. (a)-(c) Influence lines and mode shapes for freestanding and guyed structures. (d) Influence lines and surfaces for quarter point deflection in long span bridges. (e) Tension in a stay of tensioned fabric roof.

A.G. Davenport/,/. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669


NORMANDIE

663

(856 m)

i
LILLEgAI~LT

(600 ~)

I
-I

+t

+2 x tO 3 METER/T +3

avg. C I T H R U S T IN M A I N STAY 1.80 1.40 1.00 0.60 0.30 0.10 -0.10 -0.30 -0.60

' T E N S I O N IN L O W E R ; S T A Y

Fig. 3. Continued.

664

A.G. Davenport/J. WindEng. Ind. Aerodyn. 54/55 (1995) 657-669

finding the mean response using the notion of influence lines is also useful for finding the associated dynamic responses.

4. Dynamic response and mode shapes


The third tool are mode shapes of vibration. These are indicated for some of the structures in Fig. 3. Some structures such as guyed towers have closely spaced complex modes. The inertia loads on the structures are proportional to w 2 m ( z ) p j ( z ) where w2is the jth circular natural frequency, m(z) the mass and/tj(z) the jth mode shape. Thus the "convolution" of the mode shape and the influence function determines how much energy feeds into the mode response. Basically there are three sources of aerodynamic excitation causing dynamic response: (i) forces induced by turbulent fluctuations in the oncoming flow causing both background and resonant responses in the along-wind and cross-wind directions; (ii) forces induced by vortices shed in the wake of the structure; these affect primarily the resonant responses and occur primarily in the cross-wind direction; (iii) forces induced by motion of the structure; the most significant of these are aerodynamic damping forces which control the resonant response amplitude. These can add to or subtract from the available structural damping. Negative aerodynamic damping is primarily associated with the cross-wind motion and can cause large amplitude oscillation and in the extreme case lead to aerodynamic instability. The movements of the structure recorded in Fig. 4 are influenced by all three sources of excitation. The fluctuations consist of both an irregular, slowly varying component, termed here the "background response", and an oscillatory component, with a well defined frequency (or frequencies) termed the "resonant response", and occurring in bursts of several cycles. The mean response is predominantly in the along-wind direction, but the fluctuating responses occur both along-wind and cross-wind. The large cross-wind oscillations are of critical design importance both from the viewpoint of excessive motion as well as fatigue. The peak response, f, can be represented in the form
= f + Op,

(4)

in which f is the mean (time average) response, f is the rms of the fluctuating response and ~ is a statistical peak factor, generally in the range 3 to 4. The manner in which the energy in the fluctuating response of Fig. 4a is distributed with frequency is displayed by the spectrum shown in Fig. 4b. The concentration of energy of the r resonant response in the several resonant peaks is in contrast to that of the background response which is spread over a broad frequency range below the resonant peaks. The areas of the spectrum are proportional to the mean square of the response components in the frequency range and the overall mean square fluctuating response, f2, is equal to their sum. Hence
"~1/2

A.G. Davenport/J. WindEng. Ind. Aerodyn. 54/55 (1995) 657-669


A) RESPONSE 8) fSrlf) SPECTRUM

665

rRz _ 2
rR 3 I lit,

LOG. FREQUENCY

Fig. 4. Characteristicsof dynamicresponseto wind.

Here rB denotes the rms background response and rRj the rms of the resonant response peak in the jth mode of vibration. In design our final objective would be to define the response f, f in terms of the aerodynamic forces. The mean responses have already been defined by Eq. (3). There are generally speaking three main approaches available to the structural engineer to evaluate the dynamic response. All involve wind tunnel and mathematical models and differ only in where the division is made. The approaches are: (a) Aeroelastic modelling: in this all the significant dynamic and aerodynamic properties of the natural wind and the structure are modelled. The response of the full scale structure is directly inferred from measurements of the model. (b) Force measurements: in this direct measurements the steady and dynamic forces are measured on rigid models in a correctly scaled flow. A mathematical model of the structure is used to infer the responses. (c) Analytical model: in this the dynamic responses are inferred from general information on the turbulence properties and elementary aerodynamic characteristics. The first of these approaches requires an understanding of the scaling laws (relating to mass, length and time) but only a rough understanding of the precise aerodynamic mechanism. Both the latter approaches require some understanding of the causes of aerodynamic excitation.

5. Background response
The background responses at frequencies below the natural frequency are essentially quasi-static and the structure's response is unaffected by its dynamic behaviour. If the fluctuating component of force at position z is F'(z) the mean square background response can be written as

= ff
A

dA2.

(6)

666

A.G. Davenport/,].. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669


t

,.[

---.----,-..-

'

0.01

0.1

1,0 L./X

IO

I00

Fig. 5. Spatial correlation of gusts and the joint acceptance. If we express the rms force at z in the coefficient form as p(z)= qCp (z) and if R(z, z + Az) is the correlation coefficient of the forces at position z and z + Az then we can write

~2 B = q2 f~R(z,z + Az)ECpi,(z)]ECpi,(z + Az)-[ dA(z) dA(z + Az).


A

(7)

With slender structures useful simplifications can be made. For example it is often reasonable with drag forces to "scale" the force coefficients by Cp ~ 2Iu Cp and to treat the cross correlation as a function of a turbulence scale Z. We then can write ~R = q(2Iu) [Jr(L/Z)] 1/2 ~ EC, ir(z)] dA
A

(8) (9)

= ff(2I,) EJ,(L/2)] 1/2.

Here J(L/2) is a "joint acceptance factor" and depends on the relative size of the "span" to the "scale of turbulence" in the spanwise direction. It is an important function and its connection to the spatial correlation of turbulence is illustrated in Fig. 5. As it is indicated the span refers roughly to the length of structure with the deformation of similar sign. The scale of turbulence across the wind is roughly of the order of 50-60 m. The magnitude of the joint acceptance function also depends on whether the response is symmetrical or anti-symmetrical. The difference lies at values when the "span" is small compared to the turbulence scale (L/Z < 1). When the deflections are anti-symmetrical the gust action on the two span cancels out, the response is small. When the turbulence scale is less and approaches the span length there is an opportunity for the gust to envelope one span of the structure and not the other. This produces the maximum response. When the span is larger than the scale the disorganization again reduces the response. When the response has the same sign throughout the scale of turbulence is large compared to the span, the response (Z/L) reaches its full value. For structures which are broad as well as high (high aspect ratio) the joint acceptance falls off more closely as (Z/L) 2.

A.G. Davenport/J. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669

667

The background response in the cross-wind direction is generally less than half the response along-wind. Values of the joint acceptance for structures such as guyed towers and chimneys can be calculated quite accurately and account for the patch loading effects needed on these structures.

6. Resonant response to turbulence and vortex shedding

Both turbulence in the oncoming flow as well as vortex shedding in the wake can cause significant resonant response. The latter can be particularly important for lighter structures for which the aerodynamic damping can be large and negative. The general expression for the resonant response parallels certains features of the background response,
, , ~ =(~fjSGvj(fj)~ rRj \ 4-(~7-~.j /

1/2

SA[~ml~fi,(z)]dz Ia eckm~f(z)]dz"

(10)

In this fj SGF(f j ) represents the spectrum of the generalized force of thejth mode of the structure at the jth natural frequency fj. (s and (, are the structural and aerodynamic critical damping factors; the latter is explained below. The distribution of mass is defined by m(z) = ~bm(Z)mo. The spectrum of the generalized force at the natural frequency can usually be estimated if the correlation (2s) at the resonant frequency is small compared to the height H or span L. The result for slender structures then takes the form

x/f*SGF(f*)=(J(L/X:) f f*j SF(f j*)p 2(Z)dA )1/2.


A

(11)

Here fjSF(fj)/le2 represents the spectrum of the force - a measure of the concentration of the excitation - at the natural frequency. The general form of the spectra for the turbulence excitation is shown in Fig. 6. This shows that the intensity of the spectrum is greater for structures operating at lower reduced frequencies (fjL/U). This implies an increase in the spectrum at increasing windspeed. When the increase in the joint acceptance with windspeed is taken into account (increase in 2:/L the excitation at the natural frequency, [ f j S~F(fj )] 1/2, tends to increase with the reduced windspeed U* = U/fjH raised to some power fl where theoretically ~6< fl < 4.3 depending largely on the aspect ratio. The cross-wind resonant response to turbulence is generally less than half the longitudinal one. The wakes behind buildings, bridges, chimneys and other structures contain vortices shed at regular intervals. These produce fluctuations concentrated over a fairly narrow range of frequencies with a center frequency f~ related through the diameter, D (or B), and windspeed, U, to a characteristic Strouhal number S =feDU.

668
A]

A.G. Davenport/J. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669


TURBULENCE FORCES Fr B) WAKE FORCES Fw

f S F t f ) / ~ '2
RANGE OF STRUCTURAL FREQUENCIES

fSFw(f)/~ 't
TURBULENCE INTENSITY z

LOW - -

O.l

I0 fL/U

lOO

0.5 (fB/U)/S

L
2
1% 5% 30% > 100%

Fig. 6. Force spectra due to atmospheric turbulence and wake vortex shedding.

Table 1 Aerodynamic drag damping for typical structures (U = 100 ft/s) Structure Tall building Long span bridge Transmission line cable Air supported roof
Pair/Pstr fo (Hz) B U* (pair/pstr) U* [(~l for ICal = 1

1/100 1/20 1/3000 4

0.2 0.2 0.2 0.2

1000 100 0.1 100

5 5 5000 5

0.05 0.25 1.7 20

7. Aerodynamic damping
The aerodynamic damping ~a generally takes the form
Ca = (Pair/Pstr)Ca(U*),

(12)

where Ca is a function of the reduced velocity U* = U (fD). If ~a is small or positive its role is simply to augment the available structural damping; if negative it reduces the available damping and can lead to instability if it is less than the structural damping. Details are not described but relative magnitudes of aerodynamic drag damping are given in Table 1. F o r vertical structures there are insignificant modifying influences by vertical variations in the mean velocity, the cross section and the turbulence intensity. Changes in mean wind speed and diameter frequency run counter to one another - as in tapered chimneys. This diffuses the peak excitation over a range of frequencies, lowers the spectral amplitude at any given frequency and it also reduces the effective (negative) aerodynamic damping averaged over the height of the structure.

A.G. Davenport/J. Wind Eng. Ind. Aerodyn. 54/55 (1995) 657-669

669

8. Conclusions The paper has discussed some powerful tools which enable the wind loading to be defined in general terms and where necessary simplified. Three kinds of shape function can be of paramount importance: the influence functions relating the responses to the load distribution; the mode shapes describing the distribution of inertia loading and finally; the description of the loading patterns. The last are most efficiently described in terms of orthogonal functions. Although unfamiliar they represent a promising avenue for systematically simplifying and generalizing wind pressures. In considering the other side of the loading - the time dependences the paper outlines how the mean, background and resonant responses can be expressed in terms of these shape functions.

You might also like