Professional Documents
Culture Documents
2001 A Unified Approach For 3D Stability and Time Domain Response Analysis With Application of Quasi-Steady Theory
2001 A Unified Approach For 3D Stability and Time Domain Response Analysis With Application of Quasi-Steady Theory
2001 A Unified Approach For 3D Stability and Time Domain Response Analysis With Application of Quasi-Steady Theory
Abstract
Keywords: Quasi-steady theory; Turbulence; 3D response; Galloping; Buffeting; Bridges; Towers; Time
domain analysis; Generalised state equation; Numerically generated wind speeds; Frequency domain
method; Integration accuracy; Millennium Footbridge; I235 Pedestrian Bridge; Comparison with wind
tunnel
0167-6105/01/$ - see front matter r 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 6 7 - 6 1 0 5 ( 0 1 ) 0 0 1 5 7 - X
1592 S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606
(a) the instantaneous forces exciting the moving structure are almost equal to the
static forces measured at the same effective angle of wind attack; and
(b) the phase sift between the fluid force and the body velocity is negligibly small.
These conditions could be met at wind speeds where no vortex shedding lock-in and/
or flutter occur. Other important assumptions are:
(1) The structure is situated in an open terrain, where the influence of the other
adjacent structures on the fluctuating part of the wind is small, i.e., only the
effect of wind buffeting is considered.
(2) The cross-sectional dimensions of the structural elements are small compared to
the free lengths exposed to the wind and the aerodynamic influence of one part
of the structure on the others is not considered. Therefore, the line-strip
approach is applicable and permits a division of the exposed elements into a
finite number of sections and nodes. All properties, structural and aerodynamic
are assumed to be constant and lumped into the nodes.
(3) The aerodynamic properties of each section are governed by the wind field where
its modification effects due to the section geometry are incorporated in the wind
model by means of aerodynamic admittance. Thus the wind speeds are
converted into instantaneous wind forces and moments where the time lag
effects are neglected as small compared to the overall response time lengths.
(4) The mean wind direction is assumed to be approximately normal to the exposed
areas.
(5) The structure is expected to vibrate elastically around its mean deflected position
with small amplitudes. Hence the resulting structural nonlinear effects are
presumed to be insignificant.
(6) An internal reduction from Cartesian to generalized coordinates is applied
which enables analysis of large MDOF systems at reasonable time.
(7) The process involves step-by-step integration in time of buffeting responses.
Thus the wind load should be given explicitly as a function of time. Here the
wind load series are generated as multi-correlated stochastic processes by auto-
regressive (AR) filtering technique [22]. Aerodynamic admittance is incorporated
in the simulation using Irwin’s equation [12].
vector Urel and its effective angle of attack a can be presented [12] as
’
wðtÞ zðtÞ
ðtÞ ¼ U% þ 2UuðtÞ
% 2U% xðtÞ;
2
2
Urel ’ a0 ðtÞ ¼ ;
%
U ð1Þ
aðtÞ ¼ a0 ðtÞ þ yyy ðtÞ; yyy ðtÞ ¼ y% yy þ y* yy ðtÞ;
where u and w are the along- and across-wind fluctuating speeds, and U% is the mean
speed in the mean wind direction. Since the cross-section is free to oscillate, for
convenience its translation x can be taken in the direction of the mean wind, then z is
the across-wind translation, and yyy is the rotation also taken relative to the mean
wind direction where y% yy is the mean, and y* yy the fluctuating rotation angles.
According to the quasi-steady fundamentals, the instantaneous forces and moments
acting on a unit length cross-section at different angles of attack can be estimated
from the corresponding static force and moment coefficients measured in the wind
tunnel as
FD ðtÞ ¼ 12rbUrel
2
ðtÞCD ðaðtÞÞ; FL ðtÞ ¼ 12rbUrel
2
ðtÞCL ðaðtÞÞ;
ð2Þ
MðtÞ ¼ 12rb2 Urel
2
ðtÞCM ðaðtÞÞ;
with r the air density, or in body coordinates as
FX ðtÞ ¼ FD ðtÞ cosðaðtÞÞ FL ðtÞ sinðaðtÞÞ;
MðtÞ ¼ MðtÞ:
It can be proved that for small angles of attack, up to about 710 degrees (with error
of about two percent)
FX ðtÞ ¼ 12rbUrel
2
ðtÞðCD ðaðtÞÞ CL ðaðtÞÞa0 ðtÞÞ;
FZ ðtÞ ¼ 12rbUrel
2
ðtÞðCD ðaðtÞÞa0 ðtÞ þ CL ðaðtÞÞÞ; ð4Þ
The force and moment coefficients can be then expanded in Taylor series around the
assumed mean angle a0 ¼ y% yy of attack which for many cases is negligibly small.
After retaining only the linear part of that expansion
dCD dC
aðtÞ ;
L
CD ðaðtÞÞ ¼ CD þ CL ðaðtÞÞ ¼ CL þ aðtÞ ;
da %yyy da %
yyy
ð5Þ
dCM
CM ðaðtÞÞ ¼ CM þ aðtÞ ;
da y% yy
which are the slopes of the corresponding force and moment coefficients [1]. Using
Eqs. (4) and (5) the load on each finite-strip j along the structure is
f f gj ¼ ½ca j fzg ’ j þ ½ka j fzgj þ ½o j fwgj ; ð6Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflffl{zfflfflfflffl}
Self-excited forces Buffeting forces
where j ¼ 1; 2; y; n strips or element section, and ½ca ; ½ka ; and ½o are the
aerodynamic damping, stiffness and buffeting load matrices
2 0
3 8 x’ 9
2CD ðCL CD Þ 0 < U% >
> =
6 7 z’
½ca j ¼ plj 4 2CL ðCD þ CL0 Þ 0 5 ; ’ j¼
fzg U%
;
0 :’ >
> ;
2bCM bCM 0 j yyy j
2 0
3 8 9
0 0 CD < x >
> =
6 7
½ka j ¼ plj 4 0 0 CL0 5 ; fzgj ¼ z ;
>
: >
;
0 0 bCM 0
yyy j ð7Þ
j
2 0
3 8 9
CD 2CD ðCL CD Þ <1>
> =
6 7 u
½o j ¼ plj 4 CL 2CL ðCD þ CL0 Þ 5 ; fwgj ¼ U% ;
0 :w>
> ;
bCM 2bCM bCM j U% j
dCDðL;MÞ
p ¼ 12rbU% j ;
2 0
CDðL;MÞ ¼ :
da
where lj is the exposed wind strip length. In terms of body coordinates these matrices
are
2 0
3 2 0
3
2CX CX 0 0 0 CX
6 7 6 7
½ca j ¼ plj 4 2CZ CZ0 05 ; ½ka j ¼ plj 4 0 0 CZ0 5 ;
0 0
2bCM bCM 0 j 0 0 bCM j
ð8Þ
2 0
3
CX 2CX CX
6 7 dCXðZ;MÞ
½o j ¼ plj 4 CZ 2CZ CZ0 5; 0
CXðZ;MÞ ¼ :
0
da
bCM 2bCM bCM j
1596 S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606
Considering that for vertical strips vðtÞ replaces wðtÞ; y z; yzz yyy ; and expanding
½ca j ; ½ka j ; ½oa j to 6DOF per node, the overall matrices assemble
2 3
½a 1 0 : 0
6 0 ½a 0 7
6 2 : 7
½A ¼ 6 7 ; ð9Þ
4 : : : : 5
0 0 : ½a n 6n 6n
where ½A ¼ ½Ca ; ½Ka ; or ½O ; and ½a ¼ ½ca ; ½ka ; or ½o :
The unloaded DOF are filled with zeroes and for those sheltered from wind strips,
f f gj ¼ 0: The overall vectors are
fFgT ¼ ff f g1 ; f f g2 ; y; f f gn gT ; fWgT ¼ ffwg1 ; fwg2 ; y; fwgn gT ;
ð10Þ
’ T ¼ ffzg
fZg ’ 1 ; fzg ’ n gT ;
’ 2 ; y; fzg fZgT ¼ ffzg1 ; fzg2 ; y; fzgn gT :
called the generalised state equation of the system [25]. Adopting the notations in the
braces, this equation takes the standard form
’
fyðtÞg ¼ ½S fyðtÞg þ fYðtÞg: ð15Þ
The homogeneous solution of Eq. (15) or when fYðtÞgE0 is in fact the well-
known linear stability problem. It is obtained from the eigenvalue solution as a set of
conjugated complex eigenvalues lk and vectors fUgk ; for the k ¼ 1; 2; y; 2m
retained modes
( ) ( )
’
fFðtÞg lffg
lt
fyðtÞg ¼ ¼ fUge ; fUg ¼ ; ð16Þ
fFðtÞg ffg
where ffg are the generalised eigenvectors of the transformed generalised system.
The variable
lk ¼ ðzk þ iÞok ; ð17Þ
where zk is the total damping ratio (comprising aerodynamic plus structural, d ¼ 2pz
is the logarithmic decrement) and ok is the circular frequency of response. The
criterion for stability of any structure is the condition if
(k: lk p0 divergent or non-decaying oscillations (i.e., galloping) or if
8k: lk > 0 decaying oscillations of a dynamically stable system.
That is, a system is regarded as dynamically stable if and only if the lk > 0 stability
conditions are fulfilled for all modes.
One can then find the left ½V and ½U right complex generalised eigenmatrices
½U ¼ ½fUg1 ; fUg2 ; y; fUg2m ; ½V ¼ ½fVg1 ; fVg2 ; y ;
ð20Þ
½l ¼ diag½lk ;
where fyð0Þg is the initial state vector and et½L is a diagonal matrix. Provided the load
vector is given in discrete-time form at equal Dt time intervals, t ¼ pDt; p ¼ 0; 1; 2; y
the response can be calculated step by step, based only on the previous step, not as
required by the integral of Duhamel-type given in Eq. (18). The discrete-time
formula response calculation is
Z Dt
T T
½V fyðp þ 1Þg ¼ e Dt½L
½V fyðpÞg þ et½L ½V T fYðp þ 1Þg dt: ð23Þ
|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} 0 |fflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflffl}
fWðpþ1Þg fWðpÞg fQðpþ1Þg
Assuming that the load changes linearly from time p to time p þ 1; the exact solution
of Eq. (23) is
fWðp þ 1Þ ¼ eDt½L fWðpÞg þ ½L 1 ðeDt½L IÞfQðp þ 1Þg; ð24Þ
of size 2n and 2n 2m: From the given initial conditions, the starting state is
½Wð0Þ ¼ ½V ½*w 1 ½w T fZð0Þg;
T
½*w ¼ ½w T ½w ¼ diag½xk ;
ð26Þ
½*w 1 ¼ diag½1=xk ; xk ¼ xkþm ¼ fXgTk fXgk ;
which completes all necessary steps required for obtaining the response to wind
buffeting.
S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606 1599
4. Examples
Fig. 2 shows a typical deck of the Millennium Footbridge over Thames River in
London which is used to illustrate the method. The bridge main span is 144 m with
total length of 325 m linking St. Paul’s Cathedral and the new Tate Gallery of
Modern Art. This unique elegant structure features a light-weight curved deck
supported on high-strength cables with variable width end elevation relative to the
deck curve. Force and moment coefficients of representative cross-sections were
obtained in the wind tunnel laboratory of RWDI, Guelph, Ontario, Canada. Wind
loads required for design of this bridge were obtained by applying the well
established FDM buffeting procedure of Irwin [12]. In parallel, the present method
was applied on a 41 nodes discrete model with typical strip length lj ¼ 8 m (Fig. 3).
Buffeting responses during storms with various durations of up to 54:61 min were
simulated. Fig. 4 shows a sample of longitudinal and vertical wind speeds generated
in the bridge middle (node 19). The turbulence model proposed by ESDU [24] was
used for the TDM analyses. Note that the FDM procedure was based on Irwin’s [12]
turbulence model. The aerodynamic admittance 2D function proposed by Irwin was
Fig. 2. Typical cross-section of Millennium Footbridge (courtesy of the Ove Arup & Partners, London,
UK).
l 17 = 8 m
Thames River
40
30
20
10
0
0 10 20 30 40 50
Time (min)
10
Velocity w(t) (m/sec)
10
0 10 20 30 40 50
Time (min)
Fig. 4. Wind speeds at the bridge middle, elevation 15 m; time step Dt ¼ 0:1 s; turbulence properties:
Iu ¼ 18:9%; Iw ¼ 10:4%; x Lu ¼ 146:3 m; x Lw ¼ 12:2 m:
10 10
Normalized Power Spectra
Normalized Power Spectra
1 1
0.1 0.1
0.01 0.01
0.01 0.1 1 10 0.01 0.1 1 10
(a) Frequency (Hz) (b) Frequency (Hz)
Fig. 5. Power spectrum of longitudinal wind turbulence. (a) Admittance included, (b) Unity admittance.
also incorporated when simulating the wind series as a modification to the power
spectrum functions. Fig. 5 demonstrates the effect of including the aerodynamic
admittance on the target and simulated power spectra of the wind speeds. The
spectral folding above 2:5 Hz is due to the time step 0:1 Hz used for wind series
generation. However its effect was judged small for this study where the frequency of
Mode 20 is at 1:434 Hz: Table 1 shows a detailed comparison between FD and TD
methods.
Comparing the responses in single, uncoupled vibration modes, it can be judged
that the two methods (based on two different turbulence models) gave reasonably
S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606 1601
Table 1
FDM vs. TDM peak dynamic responses, solid barriers, mean speed U% ¼ 23:3 m=s; time 13:65 min; time
step Dt ¼ 0:05 s; structural damping ratio z ¼ 0:5%
0.05
0.05
0 10 20 30 40 50
Time (min)
Fig. 6. Lateral response at node 19, low solidity barrier, 20 modes, U% ¼ 23:3 m=s; total time ¼
54:61 min; Dt ¼ 0:05 s; z ¼ 0:5%:
close results. The TDM tended predicting higher responses in the lateral direction.
This was attributed to increased correlations in the wind loads. Simulations using the
first 20 modes and all cross-components showed no significant increases in lateral
and vertical predictions due to the fact that most of the added modes with
contributions in those directions had higher frequencies. Although the torsional
response predictions were small, the 20-mode simulation gave higher predictions
because most of the lower frequency modes had torsional components contributing
to the overall response.
Further analysis on the performance of the FD method was carried out using the
wind tunnel data of the actually built deck with low solidity barriers. Fig. 6 presents
54:61 min sample of lateral response at node 19. The power spectrum given in Fig. 7
1602 S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606
0.001
Lateral displacements (m)
8 .10
4
6 .10
4
4 .10
4 - X - , ∆ t = 0.025 sec
-+- , ∆ t = 0.050 sec
2 .10
4
1100.2 1100.3 1100.4 1100.5 1100.6
Time (sec)
shows that the dominating mode in the lateral response is the corresponding lowest
frequency mode (see Table 1). The histograms display Gaussian distributions with
peak factors of about 3.3. The numerical stability and accuracy of the FD method
was verified exploring in detail the motions predicted at node 19. Although the
response solution as provided by Eq. (24) is exact, its integration time step Dt is
preferable to be at least two times smaller than the wind simulation time step, 0:1 s in
this case, to preserve the spectral properties of the loads. A numerical experiment
was carried out where the time series step was reduced by splining the loads from
0.05 to 0:025 s:
Fig. 8 reveals that a double reduction of the time step was sufficient to achieve an
adequate response representation.
Fig. 9 compares the power spectra of the two responses where, as it could be
expected, the two power spectra collapse up to about 2:5 Hz (not shown here) since
S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606 1603
there is no loading frequency components higher than 5 Hz (see Fig. 5). The
frequency of the highest mode used for analysis, mode 20 at f ¼ 1:434 Hz was
accurately estimated using either time step and the difference in the root-mean-
square (rms) response was about 0.5%.
The TD method relies on a discrete technique, where all aerodynamic and
structural properties are assumed piece-wise constant and lumped into a chain of
nodes. Consequently, a 79 node model was also created to estimate the possible
deviations in the simulation. Compared to the 41 node model, the predicted
responses showed differences of about 3–5%. Considering that the bridge geometry,
mass, and force and moment coefficients vary along the bridge, these response
differences were attributed to input data extrapolation rather than to the accuracy of
the discrete model. The typical strip length lj was 8 m for the 41 node model and
about 4 m for the 79 node model. Thus it can be concluded that a model with lengths
lj in the range of b to 2b (where b ¼ 5:3 m is the deck width in this case) enables
realistic predictions.
The peak deflections at node 19 were found creating response histograms (Fig. 7)
for various sample lengths. Fig. 10 shows that more than 10 min of response was
required for stable predictions. Therefore, considering the lowest structural
frequency, about 29 response cycles were needed.
Fig. 11 shows peak vertical response of the I235 Pedestrian Arch Bridge (span
L ¼ 80 m). The site at Des Moines, Iowa, USA, was characterized with extreme ice
accumulations of 40 mm once every 50 years. It was, therefore, found that the
pedestrian fences could be solid frozen, which raised concerns for galloping
instability. The RWDIs wind tunnel study, however, showed that the onset of
galloping speed is higher than the required speed criterion. Using a simplified
1604 S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606
0.5
0.3
st
1 Vertical f=0.4117 Hz
0.2
st
1 Lateral f=0.4401 Hz
0.1
0
0 10 20 30 40 50 60
Time (min)
Iced Fences
d = 2.715 m
0.010 b=6.0 m
f = 1.157 Hz
m=6765 kg/m
ζ =0.55% C =0.64 D
0.005
dC L /dα=-4.53
0.000
0 1 2 3 4 5 6 7U/fb
0 10 20 30 40 (m/sec)
Wind velocity
Fig. 11. Peak vertical response at mean angle 0 degree (courtesy HNTB Corporation, Kansas City,
Missouri, USA).
with Sc the Scruton number, mEq ; the equivalent mass of the first vertical mode,
Fz ðsÞ its mode shape, mðsÞ the deck mass, and s the coordinate along the deck; the
onset of galloping speed was found as UR ¼ 5:68 ðUR ¼ UGal =ðfbÞ; UGal ¼
39:4 m=sÞ: Exactly the same onset speed was also predicted by the proposed method
on a 10 node numerical model. This example was used as a simple means of checking
the galloping procedure.
S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606 1605
5. Conclusions
Acknowledgements
The author is grateful to Drs. D. Novak, H. Herda, and K. Beronow, all visiting
scholars at Kyoto University back in 1992, for their help in creating the basics of this
work. Thanks to Dr. P. Irwin for his support and guidance of the RWDI’s Research
Project 94-052 where the method was largely developed. Part of that project was
supported by the National Science and Engineering Research Council of Canada
(NSERC). For the I235 Pedestrian Bridge example, thanks are due to the HNTB
Corporation, Kansas, Missouri, and to the Iowa Department of Transportation,
USA. The designers of the Millennium Footbridge, Ove Arup & Partners Consulting
Eng., London, UK are also acknowledged.
References
[1] J.P. Den Hartog, Transmission lines vibration due to sleet, AIEE 51 (1932) 1074–1076.
[2] G.V. Parkinson, N.P.H. Brooks, On the aeroelastic instability of bluff cylinders, ASME Trans.
J. Appl. Mech. 83 (1961) 252–258.
1606 S. Stoyanoff / J. Wind Eng. Ind. Aerodyn. 89 (2001) 1591–1606
[3] M. Novak, A.G. Davenport, Aeroelastic instability of prisms in turbulent flow, ASCE, EM1 96
(1970) 17–39.
[4] W.D. Iwan, Galloping oscillations of hysteretic structures, ASCE, EM99 6 (1973) 1129–1146.
[5] R.D. Blevins, W.D. Iwan, The galloping response of a two-degree-of freedom system, ASME,
J. Appl. Mech. 75 (1974) 1113–1118.
[6] K.F. Jones, Coupled vertical and horizontal galloping, ASCE, J. Eng. Mech. 118 (1992) 92–107.
[7] A. Laneville, G.V. Parkinson, Effects of turbulence on galloping of bluff cylinders, in: Conf. Proc.
Wind Effects Building and Structures, Tokyo, Japan, 1971, pp. 787–797.
[8] H. Lindner, Simulation of the turbulence influence of galloping vibrations, BBAA3, UWO, London,
Canada, 1990.
[9] H.W. Liepmann, On the application of statistical concepts to the buffeting problem, J. Aeronaut. Sci.
19 (1952) 793–822.
[10] A.G. Davenport, The response of slender, line-like structures to a gusty wind, Inst. Civ. Eng. 23 6610
(1962) 389–408.
[11] J.D. Holmes, Prediction of the response of a cable stayed bridge to turbulence, Fourth Wind Effects
Building Structures, Heathrow, UK, 1975.
[12] P.A. Irwin, Wind tunnel and analytical investigation of the response of Lions’ Gate Bridge to a
turbulent wind, NRCC Report LTR-LA-210, 1977.
[13] R.H. Scanlan, The action of flexible bridges under wind II: buffeting theory, Sounds Vibrations 60
(1978) 201–211.
[14] B.J. Vickery, K.H. Kao, Drag or along-wind response of slender structures, ASCE, ST1 98 (1972)
21–36.
[15] RWDI, Wind engineering studies for Cape Girardeau Bridge (steel alternate), RWDI Report
No. 94–197, 1994.
[16] J.-G. Beliveau, R. Vaicaitis, M. Shinozuka, Motion of suspension bridges subjected to wind loads, St.
Div., ASCE, ST6 103 (1977) 1189–1205.
[17] S. Arzoumanidis, G. Bieniek, M.P., Finite element analysis of suspension bridges, Comput. Struct. 21
(1985) 1237–1253.
[18] N.A. Newmark, A method for computation for structural dynamics, Trans. ASCE 127 (1962)
1406–1435.
[19] P. Spinelli, G. Solari, Time domain analysis of tall building response to wind action, Third
International Conference on Tall Buildings, Hong Kong, 1984, pp. 278–284.
[20] H.A. Buchholdt, S. Moossavinejad, A. Iannuzzi, Non-linear dynamic analysis of guyed masts
subjected to wind and guy ruptures, Inst. Civ. Eng. 2 (81) (1986) 353–395.
[21] D. Novak, S. Stoyanoff, H. Herda, Error assessment for wind histories generated by autoregressive
method, Struct. Safety 17 (1995) 79–90.
[22] S. Stoyanoff, P.A. Irwin, Simulation of wind loads in time domain, RWDI Research Project 94-052,
1995.
[23] E. Naudascher, D. Rockwell, Flow-Induced Vibrations: An Engineering Guide, A.A. Balkema,
Rotterdam, 1994.
[24] ESDU, Engineering Data Science Unit, Item Numbers ESDU 74030, 85020, 86010, London, UK,
1976–93.
[25] L. Meirovitch, Elements of Vibration Analysis, 2nd Edition, McGraw-Hill, New York, 1986.
[26] N.F. Morris, The use of modal superposition in nonlinear dynamics, Comput. Struct. 7 (1977) 65–72.