6610
SYNOPSIS
A number of civil engineering structures suchas 10%span suspension bridges,
tall masts and overhead power lines conform to a slender, linelike form. The
general expressions for the response of such structures to a gusty wind are
developed using the concepts of the stationary random series. Some empirical
data on atmospheric turbulence suitable for estimating the gust loading are
provided. In a worked example the response of a longspan overhead power
line is considered.
INTRODUCTION
INa previous paper’ the Author discussed the effects of a gusty wind on a
simple elasticstructure having one degreeof freedom and small enoughfor the
wind forces to be acting virtually at a point. In the present Paper the solution
is extended to certain flexible structures of slender, elongated, linelike form.
There are a number of civil engineering structures of this type: among them are
longspan bridgesin particular suspension bridgestall masts and stacks and
overhead power lines. All of them are particularly vulnerableto wind loading.
2. There are two principal differences between the response of “line” and
“point” structures to a natural wind and both relate to the effects of gusts.
First, a point structure is likely to have only one mode of vibration, whereas a
line structure may have many modes, each of which may be excited by the wind.
Second, a point structure is only affected by the temporal velocity fluctuations
of the wind at a point, while for a line structure the spatial variations of the
wind velocity acrossthe span are also important.
3. Although the dynamiceffectsof the gustinesshave not hithertobeen
fully appreciated (possibly becausethe standard practices in use today grew up
in an era of monumental structures such as the Forth Bridge and the Eiffel
Tower whose rigidity practically precluded any possibility of dynamic excitation),
some attempts have been made to allow for the fact that gusts are relatively
localized. In his notable experiments at the Forth Bridge,forinstance, Sir
Benjamin Baker2 (1884) found that thepressures on awindgaugeboard
300 sq. ft in area were some 50% less than those on a board 13 sq. ft in area.
More recently Sherlock’ (1 947) found by studying the force on a 120ft conductor
wire that the effective velocityat any one moment, averaged over the wholeof
this length was 3.1% less at 37 m.p.h. and 8.2% less at 45 m.p.h. than the
velocity at a point. The pressures were correspondingly some 6% and 16%
less. Bearing in mind that Baker’s results refer to areas and Sherlock’s to a
line, the results are in reasonable agreement. However, in spite of these fairly
positive indications, full advantage has never been taken of the possible reduc
tions. This maybe partly due to a contrary (but somewhat dubious) conclusion
reached by Stanton following his notable experiments at the Tower Bridge (see
the Author’s discussion4Davenport, 1960) but more particularly due to the
lack of a suitable statistical framework into which to fit these observations.
4. More recently this situation has improved. Liepmanns (1952) has pointed
out that the statistical concepts of the stationary time series and the power
spectrum approach, inconjunctionwith statistical theories of turbulence
developed earlier by G . I. Taylor (1935) could be usefully applied to the gust
loading of aircraft. The same techniques are used in this and the previous
paper to solve the gustloading problem of civil engineering structures.
5. In solving the overall wind loading problem (effects of the mean wind as
well as gusts) it was previously found practicable to relate the gust loading to
the concurrent mean wind load averaged over an hour or so. The statistical
properties of the loading could then be determined from the climatological
records of mean (hourly) wind velocity kept by meteorological stations. The
same approach is applicable here.
10. The coefficients P, can be found by multiplying both sides by p,(x) and
integrating overthe entire span (much as in Fourier analysis). Since a universal
property of natural modes is that they are orthogonal to one another
Thus
where
K  L/'K(x).p,(x)dx
=NI 0
. . . . .
The deflexion of the structure may then be written
where Sp,(n) is the spectrum of the rth mode load component and S& X ’ ; n)
the crossspectrum of the loads on the span at X and X’, for frequency n. The
crossspectrum is a complex quantity, that is to say it is defined by two com
ponents, one in phase and the other in quadrature. It can be normalized by
dividing by the spectrum at some reference point yielding, effectively, a cross
correlation coefficient between the loads at the two points X and X‘ for the
frequency n.
16. Hence
where Spo(n) =the spectrum of the force at a referencepoint and R(x, x’n) = the
crosscorrelation coefficient of the load at X and X’ at frequency n,
18. The latter is sometimes termed thejoint acceptance for the rth mode and
measures the correlation between the spatial distribution of pressure across the
span and the mode.
19. The response to the rth mode load component Pr(t) can be expressed
generally by the equation
STRUCTURES TO A GUSTY WIND 393
in which m,,c, and K,are generalized expressionsfor the mass, velocity damping
and stiffness inthe rth mode of vibration and y, is the deflexionof the reference
point.
20. The spectrum of the deflexion in the rth mode can now be determined
by the same method used previously in determining the response of the one
degreeoffreedom structure (Davenport', 1961), yielding
. . . . (17)
in which
= S ~ l X r ( n ) l 2 . l ~ , ( n ) l 2 . s p ~ ( n ) . d n. . . . . . . (20)
23. In the case of a beam, thespectra of the shear force and bending moment
at a point X are given by
sq(x, n) = 2 r
Ixr(n)l2.sP,(n).4r2(x) . . . (21)
where m&) and qr(x) are the bending momentand shear force at station due to
unit rth mode load component.
24. The variances of shear force and bending moment at X may be written
and um2(x)= 2r
a,TP dyn).mr2(x) . . . . . (24)
25. It is evident that the data required for evaluating the above expressions
are :
(1) the spectrum of the load
(2) the cross spectrum of the load for pairs of points across the span
(3) the modes and natural frequencies of the structure, and
(4) the damping.
~ ( t=) P+ U .
V
+l(n). ?.sin GTnt+e,) . . . . (26)
L(t) = L+2L.+z(n).=.sin(2rrnt+82)+~.$3(n)..sin(2?mt+83)
V dL W (27)
V
V dM W
~ ( t=) f i + 2 ~ T . + ~ ( n )  s i n ( 2 ~ n r + e ~ ) +   + ~ ( n ) .  =sin (2wnt+B5) (28)
8 dol V
where 01.....05 arephaseangles;
$1 . . .
are aerodynamic admittances;
4 and are variations of lift and torque withangle of attack;
and E and m a r e the lift and torque due to the mean wind. (These expressions
are written for a horizontal line structure: for a vertical linestructure of w must
be interpreted as a cross wind velocity).
35. In terms of the spectra of the vertical and horizontal velocity components
S&) and Sw(n) and cross correlation spectrum S&) the spectra of the drag,
lift and overturning torque S&), S&) and S&) are
S&)
Sp(n) = 4F2l$~(n)l2
82
. . . . . . . . . . (29)
37. Somecommentisnecessaryregardingtheaerodynamicadmittances
.
+ l . . .d5. These quantities must allow for two effects: the first relates to the
possibility that the resistance in fluctuating flow is not quasistatic and varies
with frequency (asin fact it has been found to do). Secondly, it must allow
for the spatial variation of the flow in the sense that the forces are generated
not by the velocity fluctuationsat a point in the flow but over somefinite region
surrounding the cross section of the structure.
38. Both these effects are only likely to be important in structures having
wider cross sections. In a thin cable structure the natural frequencies of the
structure are such that the important gust wave lengths are of the order of 1,OOO
times the cable diameter. Under these circumstances the resistance is almost
certain to be quasistatic and the aerodynamic admittme therefore effectively
unity. In a suspensionbridge or tall mast,however, the diameterismuch
larger (compared with the important gust wave lengths) and the matter may
require consideration.
39. In viewof the “strip assumption”referred to above, the spanwise
crosscorrelation of pressures is the same as the spanwise crosscorrelation of
velocities. Thus it is apparent that estimates can be made of the spectra of the
loads, provided that expressions can be found for :
(1) the spectrum of horizontal velocity;
(2) the spectrum of vertical velocity;
(3) the crosscorrelation spectra of both vertical and horizontal velocities
for cross wind separations (relevant to horizontal structures); and
(4) the crosscorrelation spectraof horizontal velocities for vertical separa
tions (relevant to vertical structures).
Some estimatesof these quantities are now provided.
PROPERTIES OF ATMOSPHERICTURBULENCE
PI2  4K +
(1 x2)4/3
. . . .
where X = 4,000 n/ V, (with n/rlin cycles/ft)
K = surface drag coefficient
= velocity at reference height of 33 ft, (10 m)
This expression was utilized in the previous Paperl.
where K is the drag coefficient at the surface referred to the mean velocityat the
10 m height, 71 as in the horizontal spectrum.
Rewriting
This function is shown in Fig. 2. The form differs somewhat from the hori
zontal spectrum insofar as the vertical scale of the disturbances appears to be
governed by the height above ground.
Crosscorrelation coeficients
43. The crosscorrelation coefficients whichexpress the phase relationship
and randomness between different velocity componentsat the same or different
398 DAVENPORT ON RESPONSE OF SLENDER, LINELIKE
STRUCWRES TO A GUSTY WIND 399
points in the flow, are normally complexquantities: that is to say they contain
two components, one measuring the inphase correlation and the other the
quadrature. However, because the maximum force on a horizontal line struc
ture is almost certain to be felt in a beam wind, it follows that only the span
wise crosscorrelation is of importance,i.e. in the crosswinddirection. By
symmetry the quadrature component is zero in this case. For a verticalstructure
this is not so, as is discussed later.
44. Over the region of high correlation, the crosswind crosscorrelation, it
seems, can be quite adequately represented by a simple exponential expression
of the type
R,,‘($ = elxx‘l/L . . . . . . . (34)
(7
where L ;is termed the ‘‘scale” of the turbulence and is a function of the
wavelength
45. Putting the separation X  x ’ = A x it is seen that
/;~,,(n)d(Ax) = L . . . . . . (35)
The scalecan therefore be thought of or the average dimensionof a disturbance
of given wavelength.
46. Some resultsquoted by Cramerg (1958) for the “scales” of the horizontal
F
components of turbulence are shown in Fig. 3 as functions of 7 the wavelength.
The results were obtained in open grassland. Only crosswind scales for stable
stratification are shown:forunstableconditions the crosswindscales are
,
’ l
 
U component
0 ’/ Along
wind:
(also cmsmlnd In
unstable
conditlonr)
  .  V component

WAVELENGTH 5: FEET
FIG. 3.sCALES OF TURBULENCE FOR HORIZONTALCOMPONENTS OF WIND VELOCITY AS A
. FUNCTION OF WAVELENGTH
(AFTER CRAMER)
400 DAVENPORT ON RESPONSE OF SLENDER, LINELIKF.
stated to be more or less the same magnitude as the downwind. From the
Figure it would appear
__ that the cross wind “scales” can be represented quite
adequately by
P in
stable
conditions
and
1P
L(:) = 7n inunstableconditions
47. The question which of these is the more appropriate in high winds needs
to be considered. Although there is not yet much information available the
indications are that inhighwinds the turbulence structure tendsmore to
resemble that in fairly stable conditions insofar as the disturbances tend on the
whole to be elongated in the direction of the wind. Crosscorrelation results
calculated by the writer from Bailey and Vincent’s10 (1939) SevernBridge
records ofwindvelocity at various points across the span, suggest that the
lateral scale (of the longitudinal velocity component) is about onethird of its
longitudinal scale.
48. However, in the present state of knowledge it seems advisable to adopt
the more conservative estimate (i.e., the value producingthe higher correlation)
and assumethevalue for the crosswindscale in unstableconditions. The
crosscorrelation coefficient is then
RA,(n) = e7Ax.nlF . . .
. . . . (36)
The same expression is assumed for the lateral correlationof the vertical velocity
component.
49. Some estimates of the cross correlation of the horizontal velocities in
the verticaldirectionhavebeenmade bySinger1’frommeasurements
on a 400ft mast situated in wooded country and by Davenport7 on a 500ft
mast in open grassland. From these it was apparent that the quadrature
correlation, although small, was not zero. In fact themaximumcorrelation
between the velocity, at two pointsseparated
vertically,
occurred not
simultaneously but after the wind had travelled downwind a distance roughly
equal to the vertical separation. There was no suggestion that the vertical scale
of the horizontal fluctuations varied with height above ground.
50. In spite of the nonzero quadrature component it is nevertheless small,
and it seems adequate for practical purposes to use the square root of the
“coherence”as ameasureof the crosscomllation. Thecoherenceis the
absolute square of the inphase and quadrature correlations.Measurements
of the coherence during occasions of strong windmade bySinger and
Davenportare shown in Figs 4aand 4b.Markeddifferences in surface
roughness do not seem to have any appreciable effect.
51. It is evident that the same expression usedfor the other crosscorrelation
coefficients is also adequate for the vertical crosscorrelation: that is
R&) W dCoherence

. . . . .(37)
eln.Az/V
401
0 150300het
(average runs
023. 926.927)
10
00 1
08 06 00 04 02 I.o I.2
4z.n
RATIO OF VERTICALSEPARATION TO WAVELENGTH Y
VI0
09
07
E@ "= \
04 
0 3.
02
01
U
0 A A
0.01 I I
06 0.00 4 02 0.8 I .o I2 'l'4
RATIO OF VERTICAL SEPARATION TO WAVELENGTH
VI0
FIG.4 . a H E R E N C E OF WINDSPEED IN STRONG WIND AS A FUNCTION OF VERTICAL
SEPARATION TO WAVELENGTH RATIO 
(A;)
(top) for wooded countryside(after Singer) (bottom) for open grassland
30
402 DAVENPORT ON RESPONSE OF SLENDER, LINELJKE
mechanical damping at the joints and in the material of a structure, and from
the aerodynamic damping due to the motion of the object through the strong
current of air. Some estimation of these damping parameters is vital on account
of the control they exert over the amplitudes of oscillation. It is not possible
to determine the mechanical damping in general terms since it depends entirely
on the type, material and construction of the structure: it is possible, however,
to derive some general expressionsfor the aerodynamic damping as shown below.
Aerodynamic damping
53. A structure vibrating in air will be subject to certain aerodynamic forces
tending to damp the vibration. In still air, the forces will result mainly from
the viscosity of the air. The damping is not likely to be large, and probably
insignificantcompared to the mechanicaldamping. In a strong wind, the
principal aerodynamic forces acting will be form drag and lift. If the object is
moving, fluctuating components of these lift and drag forces will be induced,
tending to oppose, and hence, damp out the motion.
54. An approximate expression for the logarithmic damping decrement is
where AE is the work done per cycle against the drag (or lift) and E is the total
energy stored in the system.
55. Suppose the mode of the beam (i.e. mast or suspension bridge) is pI(x),
then the velocity of the beam at station X is p= 2mz,p1(x)sin 2rn,t where n, is
the natural frequency in the rth mode.
56. The drag on aslice of the beam of length dx, inasteadywind of
velocity Tx,is
Fx.dx = 3 p . C ~Fx2.D.dX
.
where D is the diameter of the beam.
The fluctuating component of this drag, when the velocity of the object is j , is
2y 
P . d x
P
The work done per cycle on the slice is then
2y 
d(AE) = dxlp=P,.dy
V
X
dL
where  is the rate of change of the lift force perunit length of the deck with
da
the angle of attack a (in radians).
59. Evidently, the damping varies directly as the meanwindvelocity and
inversely as the frequency of vibration. The latter implies that the higher modes
are more lightly damped.
APPLICATIONS
Gust factors
60. The product of the analysis so far has been to enable estimates to be
made of certain statistical properties ofthe deflexions, bending moments, shear
forces, etc., at points across the span of a linelike structure in a strong wind.
It remains to beshownhowtheseresults can beuseful in determining the
specific quantities required in the design. Probably the most useful statements
that can be made concernthe maximum deflexions and stress levels at points in
the structure having a given probabilityof occurrence (e.g. the oncein100year
peak stress). At a later stage, when the evaluation of fatigue life can be under
taken, it may be valuable to be able to estimate the number of applications of
stress ofgiven magnitude. These and other questions can be answered using
certain timehistory relationships some ofwhichweregiven in the previous
Paper'.
61. Many of these rely on the assumption that the statistical distribution of
the variable is normal or Gaussian. Justification for assuming that the wind
turbulence was Gaussian was given in the previous Paper': from the further
assumptions that the variables(velocity,pressure,deflexion,etc.) are linear
functions of one another it follows that they, too, are Gaussian.
62. For such a distribution it was shown that the largest instantaneous value
of a stationary random variable X during a period T averages
Xpcak = . 3 . (42)
404 DAVENPORT ON RESPONSE OF SLENDER, LINELIKE
where
and R, u(x) and &(n) are respectively the mean, standard deviation and spectrum
of X. Values of g(vT) as a function of vT are shown in Fig. 5.
63. It was also shown previously that if the period T was chosen large com
1
pared to the characteristic response time ;then the largest instantaneous peak
value would not vary much from one period to the next. Thus not much error
would be incurred if the largest value is taken equal to the average given in
eq. (42).
64. In the case of the wind, the mean value P is also a chance event with a
frequency distribution of its own. Because of the narrowness of the peak value
distributions it follows that the overall distribution of peak values (taking into
account the mean value variation) will besimilar to the mean value distribution
but with the variable augmentedby the factor
(1 +g(vT) F).
The latter is termed the “gust factor” since it indicates the additional stress,
deflexion etc. attributable to the gustiness.
65. It should be pointed out that the gust factor is not necessarily constant
RSWNSE FACTOR V T
FIG. 5.&AK GUST FACTOR AS A FUNCI‘ION OF RESPONSEFACTOR
STRUCTURES To A GUSTY WIND 405
and may vary with changes in mean wind velocity. This variation is not likely
to be large however,and the adoption of some value corresponding
to the design
mean wind velocity shouldbe satisfactory.
Simplifcations
66. Thenumericalevaluationof the expressionsdeveloped in previous
sections is made considerably easier by one or two simplications which are
possible under most circumstances. The frst of these relates to the calculation
of the “joint acceptances” of the various modes. When the derived expressions
for the crosscorrelation coefficient of pressures and the modes are substituted
into the expression for the joint acceptance (eq.15) it is found that the latter
can be expressed in termsof integrals of the type
n
where ~ ~ 7 7 1It .is found that over the important range of the spectra (at or
near the resonant frequency), c is generally greater that about 5 for most long
span suspensionbridges,cablesandmasts.Consequently,theregion of sig
nificant correlation is confined to fairly short sections of thespan. For the
larger values of c the average correlation over these intervals tends towards 2/c.
Furthermore, since these sections are short X and X’ will not be widely separated
over regions of significant correlation and the above integral tends towards
67. Thus the joint acceptance at the natural frequency will have a value given
approximately by
(45)
TABLE
1.RESPONSE
OF CABLE SPAN TO GUSTY WIND
Horizontal
1st
2nd
VerticaI
0.00067 04055
0.00125 10.0193 1 i
I l l l
1.3
0.6
_ _ _ _ ~ ~~ ~~
Notes
(1) From Fig. 1. (2) From eq. (45). (3) From eq. (40). (4From eqs (14) and (29).
(5) From eq. (46). (6) From eq. (19). (7) From eq. (33). (8) From eq. (41).
ACKNOWLEDGEMBNTS
74. The work described in the paper formed part of a Ph.D. Thesis entitled
“A statistical approach to the treatment of windloading of the tall mast and
suspension bridge” submitted in 1961 to the University of Bristol. During the
period of the research the Author was awarded a D.S.I.R. research fellowship.
75. Thewriterwishes to record his greatindebtedness to Professor Sir
Alfred Pugsley,O.B.E., D.Sc., F.R.S. under whose general supervision this work
was carried out, for his interest and encouragement.
REFERENCES
1. A. G. Davenport, “The application of statistical concepts to the wind loading
of structures”, Proc. Instn of civ. Engrs, vol. 19, (Aug. 1961), pp. 449472.
2. B. Baker, “The Forth Bridge”, Engineering, vol. 38, 1884, p. 213.
3. R. H. Sherlock, “Gust factors for design of buildings”, Int. Assn Bridge L Struct.
Engg, vol. 8, 1947, pp. 204236.
4. A. G. Davenport, 1960, “Wind loads on structures”, National Research Council
of Canada, Division of Building Research, Technical Paper No. 88.
5. H. W. Liepmann,“Ontheapplicationofstatisticalconcepts to the buffeting
problem”, J. Aer. Sc’s, vol. 19, No. 12, 1952, pp. 793.
6. A.C.Eringen,“Responseofbeamsandplates to random loads”, J. Applied
Mechanics, vol. 20, 1953, pp. 461.
7. A. G . Davenport, “The spectrum of horizontal gustiness near the ground in high
winds”, Q. J. roy. Met. Soc., vol. 87, 1961, pp. 194211.
8. H. A. Panofsky and R. A. McCormick, “The spectrum of vertical velocity near
the surface”, College of Mineral Industries, Penn. State University, 1959, (un
published Paper).
9. H. E. Cramer, “Use of power spectra and scales of turbulence in estimating wind
loads”, unpublished Paper presented at the 2nd National Conference on Applied
Meteorology, Ann Arbor, Michigan, 1958.
10. A.Baileyand N. D. G. Vincent,“Windpressureexperiments atthe Severn
Bridge”, J. Instn civ. Engrs, vol. 11, 1939, pp. 363380.
408 DAWNPORT ON RESPONSE OF SLENDER, LINELIKE
STRUCTURES TO A GUSTY WIND