Professional Documents
Culture Documents
www.elsevier.com/locate/engstruct
Received 3 March 2005; received in revised form 9 May 2005; accepted 9 May 2005
Abstract
This research identifies the time-varying wind load from the responses of a structure. The fluctuating wind speed process is simulated
as an ergodic multivariate stochastic process. The identification equation is formulated in state space, and it is solved with regularization to
obtain the wind load from simulated structural responses. A modified iteration scheme is further proposed to improve the identified results.
A 50 m guyed mast is selected as an example to validate the effectiveness of the proposed method. Numerical simulations indicate that
the proposed method can be used to identify wind load accurately and effectively over the full height of the structure from the structural
displacement or strain responses, and only two displacement or strain responses in two orthogonal directions at a reference level are needed
to obtain accurate results.
© 2005 Elsevier Ltd. All rights reserved.
Keywords: Dynamic; Wind load; Guyed mast; Response; Finite element model; Equivalent model; Displacement; Regularization; Simulation; Identification
0141-0296/$ - see front matter © 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2005.05.007
S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598 1587
2α
process is simulated as an ergodic multivariate stochastic where Cm (z) = 12 ρ µs (z)A(z) zz is the coefficient of
process. The identification equation is formulated in state
mean wind load, which depends on the reference height
space, and it is solved with regularization to obtain the
wind load from simulated structural responses. A modified height of the selected level, C f 1 (z) =
and the vertical
α
iteration scheme is further proposed to improve the identified ρ µs (z)A(z) zz and C f 2 = 12 ρµs (z)A(z) are similarly
results. A 50 m guyed mast is selected as an example defined coefficients for the fluctuating wind load. Benfratello
for validating the effectiveness of the proposed method. et al. [10], after analyzing the stochastic response of a SDOF
The guyed mast is modeled by equivalent beam–column structure subject to wind action, concluded that neglecting
elements for the mast and truss elements for the guys. The the quadratic pressure term of the fluctuating wind speed
connections between the guys and beam–column elements could not lead to accurate results. The fluctuating wind
are assumed to be rigid links, which preserve the guys’ speed is simulated as an ergodic multivariate stochastic
eccentricities. And a static reduction technique is used process, and the fast Fourier transform is needed to
to condense the slave degrees of freedom. Numerical estimate the fluctuating wind speed components acting
simulations indicate that the proposed method can be on the structure [11]. When the mean wind speed v̄(z )
used to identify wind load over the full height of the corresponding to a reference level z and the time history of
structure accurately from the structural displacement or the fluctuating wind speed at all levels, v̂(z, t), are obtained,
strain responses. And only displacement or strain responses the wind load on the structure can be computed.
obtained in two orthogonal directions at a reference level are
required for an accurate identification.
3. Problem formulation
2. The wind load model in the time domain 3.1. The state space equation
The longitudinal wind load is considered in the following
The equation of motion of an n DOF damped structural
analysis neglecting the transverse and vertical wind load
system is expressed as
components. The wind speed at level z above the ground,
v(z, t), can be written as M Ẍ + C Ẋ + K X = P (5)
v(z, t) = v̄(z, t) + v̂(z, t) (1) where Mn×n , Cn×n and K n×n are the mass, damping and
where v̄(z, t) and v̂(z, t) denote the average wind speed and stiffness matrices. X, Ẋ and Ẍ are (n × 1) nodal displace-
fluctuating wind speed, respectively. The mean wind speed ments, velocities and accelerations vectors of the structure.
at a different level, v̄(z), may be calculated according to the Pn×1 is the wind load vector applied on the system.
power law [8] Eq. (5) can be rewritten in state space as
α
z Ż = K ∗ Z + B̂ P (6)
v̄(z) = v̄(z ) (2)
z X
where Z= ;
Ẋ 2·n×1
in which z and v̄(z ) are the reference height and average
wind speed at the reference level, respectively. z and v̄(z) 0 I
K∗ = ;
are the arbitrary height and its corresponding average wind −M −1 K −M −1 C 2·n×2·n
speed. The power law exponent α is determined from the
0
terrain roughness [8]. B̂ = . (7)
M −1 2·n×n
Since the wind speed is low, the aeroelastic forces arising
out of the interaction between air and the structure are If the response of the structure is represented by Ns
so small that they can be neglected. The aerodynamic output quantities in the output vector y(t) from sensors
forces due to wind turbulence are expressed as follows with such as accelerometers, velocity transducers, displacement
no lifting effect in the present case of vertical cantilever transducers or strain gauges, etc., the output equation can be
structure. The fluctuating wind load F(z, t) on the structure expressed as
at level z can be written as [9]
y = Ra Ẍ + Rv Ẋ + Rd X (8)
1
F(z, t) = ρµs (z)A(z)v 2 (z, t) (3) where Ra , Rv and Rd are output influence matrices
2
where ρ is the density of air, A(z) is the orthogonal exposed for the acceleration, velocity, displacement measurements,
wind area at level z and µs (z) is the drag coefficient of the respectively. Removing the term Ẍ from Eqs. (5) and (8)
structure at level z. yields
Substituting Eqs. (1) and (2) into (3) yields y = RZ + DP (9)
F(z, t) = Cm (z)v̄ (z ) + C f 1 (z)v̄(z )v̂(z, t)
2
where R = Rd − Ra M −1 K Rv − Ra M −1 C , D =
+ C f 2 (z)v̂ (z, t)
2
(4) Ra M −1 .
1588 S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598
Eqs. (6) and (9) are converted into discrete equations pseudo-inverse solution of Eq. (13). It is clear from the
using the exponential matrix, and the final discrete model second term that the non-negative regularization parameter
is λ has the effect of forcing a bounded solution. The optimal
Z ( j + 1) = AZ ( j ) + B P regularization parameter is determined by minimizing the
(10) identified error calculated by Eq. (16). In practice, the
y( j ) = R Z ( j ) + D P ( j = 1, 2, . . . , N)
real value is not known and an L-curve method [14] is
where N is the total number of sampling points. τ is the used to solve for the optimal regularization parameter.
time step between the state variables Z ( j + 1) and Z ( j ), If the measured displacements, velocities, strains and
and A = exp(K ∗ τ ), B = K ∗ (A − I ) B̂. accelerations (or their combinations) are used, each response
Solving for the output y( j ) with zero initial conditions component in the vector y(t) in Eq. (8) should be scaled by
from Eq. (10) in terms of the previous inputs P(i ) (i = its respective norm to have dimensionless units.
1, 2, . . . , j ) yields
j
4. Numerical simulation
y( j ) = Hi P( j − i ) (11)
i=0
A 50 m high guyed mast is studied as an example to
where H0 = D and Hi = R A(i−1) B. check the validity of the proposed wind load identification
The constant matrices in the series in Eq. (11) are known method. White noise is added to the calculated responses of
as system Markov parameters. The Markov parameters are the system to simulate the polluted measurements.
commonly used as the basis for identifying mathematical
y = ycalculated + E P Noise σ (ycalculated) (15)
models in linear dynamic systems. The Markov parameters
represent the response of the discrete system to unit impulse, where y is the simulated measured response, ycalculated is the
and they are properties of the structural system [12]. calculated response, E P is the noise level, Noise is a standard
normal distribution vector with zero mean value and unit
3.2. Wind load identification with regularization standard deviation and σ (ycalculated) is the standard deviation
of the calculated response.
We rewrite Eq. (11) to give the matrix convolution The relative percentage error in the identified wind load
equation as can be calculated as
H̄ P̃ = Y (12) Relative Percentage Error
in which P identified − P true
= × 100% (16)
H0 0 ··· 0 P true
H1 H0 ··· 0 where • is the norm of the wind load vector. P identify
H̄ = .. .. ,
. . ··· 0 and P true are the identified and true wind load time histories,
H N−1 H N−2 · · · H0 respectively.
T
P̃ = P(0)T , P(1)T , . . . , P(N − 1)T 4.1. The 50 m high guyed mast
T
Y = y(0)T , y(1)T , . . . , y(N − 1)T . The 50 m high guyed mast has an equilateral triangular
cross-section with each side 1.2 m long. The three vertical
This is an ill-posed problem due to the lack of continuous legs are of 219.1 mm diameter, 10 mm thick hollow steel
dependence of the solution on the data. A straightforward tubes for the lowest 5.8 m length and 6.35 mm thick for
least-squares solution produces an unbounded result. the upper length of the legs. The diagonals and horizontal
Regularization is usually used to circumvent the problem of members are 5 mm thick 60 mm steel tubes. The mast is
the lack of continuous dependence, and would provide an divided into 43 vertical segments with the first 40 segments
analysis of the ill-posed problem. One approach is proposed each 1.16 m high and the others 1.2 m high. It is tied to the
by Tikhonov [13]: replacing Eq. (12) with the associated ground with cables at two levels of 20.88 and 46.40 m above
equation the ground level with three cables at each level as shown in
H̄ T H̄ + λI P̃ = H̄ TY. (13) Fig. 1. The cables are of 17.5 mm diameter each, consisting
of 19 galvanized steel wires 3.55 mm in diameter, and the
Solving Eq. (13) is equivalent to solving the following initial tension in each cable is 23 520 N. The elastic mod-
unique problem: ulus of the material is 2.05 × 1011 and 1.2 × 1011 N/m2
for the mast and cables, respectively. The mass density and
min J ( P̃, λ) = H̄ P̃ − Y 2 + λV 2 . (14)
Poisson ratio for the mast are 7850 kg/m3 and 0.25 respec-
For λ > 0, the matrix operator H̄ T H̄ + λI is tively while those for the cables are 7820 kg/m3 and 0.20
mathematically unique, and its inverse is continuous in the respectively. Three 5 m × 4 m × 2.5 m deep reinforced
S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598 1589
Table 1
Properties of the equivalent beam–columns
Table 2
The first eight natural frequencies of the guyed mast
1 2 3 4 5 6 7 8 9 10
Original
model (Hz) 1.053 1.844 2.324 3.130 5.455 5.678 9.879 9.958 15.425 15.525
Reduced
model (Hz) 1.053 1.845 2.324 3.131 5.456 5.679 9.885 9.964 15.464 15.565
Error (%) −0.00 −0.08 −0.00 −0.03 −0.02 −0.02 −0.06 −0.06 −0.25 −0.26
Note: Error(%) = ( f original − f reduced )/ f original ∗ 100%.
Fig. 1. The equivalent properties of the spatial beam–column shown in the Appendix A. For the procedure for calculating
elements [19] for a segment of the mast with an equilateral the varying wind load time histories, refer to Appendix A.
triangular cross-section are listed in Table 1 [20]. In the Fig. 3 shows the simulated fluctuating wind speed time
following studies, the DOFs correspond to the x- and y- histories at Nodes 1, 4, 7 and 10 for a 10 min interval.
directions of all the mast nodes are taken as the master A comparison of the power spectral density of these wind
DOFs. Rayleigh damping is assumed and the first two speeds with those obtained from the Kaimal spectrum
damping ratios are taken equal to 0.02. Table 2 gives the is plotted in Fig. 4. The power spectral density of the
first ten natural frequencies of the system before and after simulated fluctuating wind speed agrees well with the
reduction. The relative percentage errors of the eigenvalues Kaimal spectrum [21] which is based on the mean wind
are very small, and the error increases with the order number speed. The variations in the spectrum are contributions from
with the largest error equal to 0.26%. It may be concluded the fluctuating wind components.
that the Guyan static reduction technique is accurate enough
for analyzing the lower eigenvalues and eigenvectors of the
system as well as the time response history under wind load. 4.3. Identified results and discussion
4.2. Simulation of the fluctuating wind load The orthogonal exposed wind area at a level is calcu-
lated [17] with µs (z) = 1.6, and the wind load is assumed
The highest frequency of interest of the fluctuating wind to be concentrated forces applied at the nodes at different
component is taken as 2.5 Hz and the size of fast Fourier levels. The responses of the system are calculated using a
transform is 2048. The time step of data is 0.2 s. The time increment of 0.02 s, and the wind data is interpolated
roughness length z 0 is taken as 0.02 m, and the exponential to have the same time increment, namely, with a sampling
decay coefficient C z is taken as 8.0 [9] for calculating the frequency of 50 Hz. The non-linearity of the system under
wind power spectrum; for the coherence function between the wind load has been checked to be very small, justifying
the wind speeds at two different levels, refer to Appendix A. the use of Newmark method in the response analysis [20].
The vertical wind profile is assumed to follow the power law As a matter of fact, the responses of the system in
with an exponent α of 0.16. For convenience of computation, the field measurement cannot be obtained with zero initial
the reference height is assumed as 50 m, which is at the conditions because the wind is continuous. The velocity
top of the structure. The mean wind speed is 30 m/s at this and acceleration responses of the structure cannot reveal the
reference level, and the angle β between the wind direction effect caused by the mean wind speed components with non-
and the positive direction of the x-axis is assumed to be zero initial conditions; therefore the wind load identification
30 degrees. The density of air ρ is taken as 1.23 kg/m3 shown below is performed on the basis of measured
in the wind load calculation. The fluctuating wind speed is displacement responses with non-zero initial conditions. The
simulated as an ergodic multivariate stochastic process as “measured” displacement responses are calculated from the
S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598 1591
Fig. 4. Comparison of power spectral density between simulated and Kaimal spectra (— Kaimal, – – – simulated).
1592 S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598
Table 3
Error of identification from different sets of measured responses (Eq. (12))
Node No. Wind load Displacement responses from different sets of nodes
1–10 1–5 6–10 1, 3, 5, 7, 9 2, 4, 6, 8, 10 10
(1) ∗ refers to nodes at the guy levels. (2) Fx and Fy are wind load components in x- and y-directions, respectively.
structure under the wind load excitation specific to the z can be approximately formulated with respect to the wind
simulation. load at the reference level. The system force vector Pn×1 is
defined as the following in the inverse problem:
4.3.1. Identified wind load from structural displacements T
The noise-free displacement response components in P = C̄1 C̄2 · · · C̄ j · · · C̄n P z , t = C̄ P z , t (19)
both the x- and y-directions at each nodal level are used to T
identify the wind load. Six sets of measured displacement in which C̄ = C̄1 C̄2 · · · C̄ j · · · C̄n is the force
responses are used as shown in Table 3, with responses coefficient vector. P(z , t) = 2 ρµs (z )A(z )v (z , t) is the
1 2
ranging from a single node to all the nodes above ground. wind load at the reference level z calculated from Eq. (3).
2α
Table 3 gives the Relative Percentage Error (RPE) of the A z
C̄ j = A j z j ( j = 1, 2, . . . , n), A j and z j are the
z
identified results and Fig. 5 shows the identified wind load
exposed wind areas and height at the j -th node; A z is the
components at Nodes 1, 5 and 10 obtained from the
exposed wind area at the reference level.
displacements of all nodes.
If the wind attack angle β (angle between the
It is noted from Table 3 and Fig. 5 that accurate nodal
positive direction of the x-axis and the wind direction)
wind loads can be identified when all the nodal displacement
is not zero, the force vector may be written as P =
responses are included in the identification, while other
C̄x1 0 · · · C̄x j 0 · · · C̄xn 0 T Px (z , t)
sets of responses with incomplete measurement give 0 C̄ y1 · · · 0 C̄ y j · · ·
Py (z , t)
= C̄ P, with
0 C̄ yn
unacceptable results at most of the nodal points. This may be
P = PPxy (z , t)
(z , t)
; C̄ x j and C̄ y j are the force coefficients along
due to the solution of the underdetermined set of equations
in Eq. (12). In practice, measuring all nodal displacements the x- and y-directions. Px (z , t) = P(z , t) × cos β and
is impossible or very costly. And the following alternative Py (z , t) = P(z , t) × sin β are wind load components in the
approach is proposed to overcome this problem. x- and y-directions, respectively, at the reference level.
Eq. (12) may be rewritten as
4.3.2. Approximate formulation for wind load identification H̄c P̃c = Y (20)
We assume that the wind speed (including the fluctuating
components) along the vertical direction of the structure with the wind load and wind attack angle at the reference
follows the power law, and the wind load at a certain level level as the unknowns, in which
S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598 1593
Fig. 5. Identified results from all nodal displacements (— true wind loads, . . . . . identified results).
H0C̄ 0 ··· 0 It is interesting to note that the identified wind load data
H1C̄ H0C̄ ··· 0 is less than the true values over the whole time history.
H̄c = .. .. , The accuracy becomes better when the measured station
. . ··· 0
is closer to the reference level with the displacements at
H N−1 C̄ H N−2 C̄ · · · H0C̄
T Node 10 giving the better results. When the number of
P̃c = P(0)T , P(1)T , . . . , P(N − 1)T measurement stations increases, the identified results do not
T become better; on the contrary, they become slightly worse.
Y = y(0)T , y(1)T , . . . , y(N − 1)T . This may be due to the assumption of the same fluctuating
wind speed at different levels above ground. This is contrast
At another level z above the ground, the wind load is to the real case where the vertical correlation between two
1 measured fluctuating wind velocities is governed by the
P(z, t) = ρµs (z)A(z)v 2 (z, t) (21) coherence function in Eq. (A.11) in Appendix A, while the
2
α horizontal correlation is too small to neglect. Subsequently,
where v(z, t) = v̄(z ) zz + v̂(z, t), v̄(z ) is the mean the following iterative approach is further developed to
wind speed calculated from the identified wind load at the improve the identified results.
reference level, and v̂(z, t) is the simulated fluctuating wind
speed corresponding to a level z [21]. It is noted that the 4.3.3. Modified iterations for an improved identification
power law exponent α is a function of the terrain roughness Eq. (19) assumes a power law distribution of the wind
and is independent of the wind speed. It is known before the speed based on the wind speed at the reference level. In fact,
wind load identification. The wind attack angle β is known there is difference in the fluctuating wind speeds at different
when both the x- and y-components of the wind load are levels, even though they have the same power spectral
identified. density. This difference may induce a large identified
The same simulated sets of responses as for last study error. The iterative approach is therefore modified as
are used. The identified results from eight noise-free follows:
displacement sets are tabulated in Table 4 and Fig. 6 shows Step 1: Identify the wind load at the reference level from
the identified results from displacements at Node 10 or 8. the measured responses from Eq. (20). The wind speed
1594 S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598
Table 4
Error of identification from different sets of measured responses (Eq. (20))
Node No. Wind load Displacement responses from different sets of nodes
10 8 6 4 2 9, 10 8–10 5–10
Fx 7.84 11.40 18.20 22.79 31.33 8.41 9.36 10.96
1
Fy 9.02 11.14 16.99 20.63 27.30 9.31 9.81 10.74
Fx 7.75 11.29 18.05 22.61 31.09 8.32 9.26 10.85
2
Fy 8.92 11.02 16.85 20.47 27.08 9.21 9.71 10.63
Fx 7.76 11.29 18.06 22.62 31.09 8.33 9.27 10.86
3
Fy 8.92 11.03 16.86 20.47 27.09 9.21 9.71 10.63
Fx 7.77 11.31 18.08 22.65 31.13 8.34 9.28 10.88
4*
Fy 8.94 11.05 16.88 20.50 27.12 9.23 9.73 10.65
Fx 7.80 11.34 18.12 22.70 31.19 8.37 9.31 10.91
5
Fy 8.97 11.08 16.92 20.54 27.18 9.26 9.76 10.68
Fx 7.89 11.47 18.21 22.88 31.44 8.47 9.42 11.03
6
Fy 9.07 11.20 17.06 20.71 27.40 9.38 9.44 10.80
Fx 7.91 11.49 18.30 22.92 31.48 8.51 9.89 11.05
7
Fy 9.09 11.22 17.09 20.75 27.44 9.33 9.37 10.82
Fx 7.85 11.41 18.28 22.80 31.33 8.42 9.82 10.97
8
Fy 9.02 11.14 17.00 20.64 27.30 9.32 9.46 10.75
Fx 7.93 11.51 18.34 22.96 31.53 8.48 9.72 11.07
9*
Fy 9.11 11.24 17.12 20.78 27.48 9.40 9.91 10.85
Fx 7.86 11.43 18.23 22.82 31.35 8.43 9.38 10.99
10
Fy 9.03 11.16 17.01 20.66 27.32 9.33 9.83 10.76
Note: (1) ∗ refers to nodes at the guy levels. (2) Fx and Fy are wind load components in x- and y-directions, respectively.
Fig. 6. Identified results from noise-free displacements at Node 8 or 10 (— true wind loads, – – – Node 10, . . . . . Node 8).
S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598 1595
Fig. 7. Identified results from noise-free displacements at Node 10 (— true wind loads, – – – with modified iterations, . . . . . without modified iterations).
at the reference level, including the mean wind speed and where P̄ j (z ) and P̄ j (z ) are the mean values of the time
the fluctuating wind speed, is calculated from Eq. (3). The histories P j (z , t) and P j (z , t), respectively, and cc are
identified wind speed is then averaged. the coefficients of correlation between the identified wind
Step 2: Simulate the fluctuating wind speed v̂(z, t) [11] loads at the reference level in two successive iterations,
corresponding to a different level z on the basis of the P j (z , t) and P j −1 (z , t). The convergence criteria Tolerance1
identified mean wind speed from Step 1 and the power and Tolerance2 are selected as 10−3 in this study. The last
law. P j (z , t) after convergence can be taken as the wind load at
Step 3: Calculate the total wind load from Eq. (4) with the reference level. And for other level, z above the ground,
the identified mean wind speed from Step 1 and simulated the wind loads can be computed from Eq. (21) with the last
fluctuating wind speed data from Step 2. This is to include v̂(z, t) obtained after convergence as the latest simulated
the second order term of the fluctuating wind speed for better fluctuating wind speed.
accuracy. Then obtain the responses at the measured DOFs The identified results from displacements at Node 10
by solving the equation of motion of the system using the with and without using the modified iterations are plotted
Newmark method. in Fig. 7. It can be seen clearly that the new scheme of
Step 4: Identify the wind load increment P(z , t) from iterations can improve the identified results. The modified
Eq. (20) on the basis of the difference between the measured results match the true time history very well except at one or
responses and responses calculated from using the identified two peaks.
wind loads.
Step 5: On the basis of the newly identified wind load, 4.3.4. Effect of measurement noise
P j (z , t) = P j −1 (z , t) + P j (z , t), obtain the new wind Displacement responses at Node 10 with different noise
speed data v j (z , t) from Eq. (3). Repeat Steps 2 to 5 until levels are used to identify the wind load, and the modified
the following conditions are satisfied: iteration steps are employed. The identified results are
tabulated in Table 5, and Fig. 8 shows the identified wind
P̄ (z )
j load components from responses with 1% and 10% noise
≤ Tolerance1 and (1 − cc ) ≤ Tolerance2
P̄ j (z ) levels.
1596 S.S. Law et al. / Engineering Structures 27 (2005) 1586–1598
Table 5 Acknowledgements
Error of identification from different sets of measured responses (Eq. (20)
with the modified iteration procedure) The work described in this paper was supported by a grant
Node No. Wind load Noise level (%) from the Hong Kong Research Grant Council Project No.
0 1 5 10 PolyU 5043/02E. Acknowledgement is made to the High-
Fx 4.77 7.74 9.94 14.97 ways Department of the Hong Kong Special Administrative
1 Region Government for its permission to make use of the
Fy 3.52 6.49 8.79 14.06
guyed mast for the present research and this publication.
Fx 4.65 7.62 9.82 14.85
2
Fy 3.41 6.38 8.68 13.95
Fig. 8. Identified results from displacements at Node 10 with 1% and 10% noise (— true wind loads, – – – 1% noise, . . . . . 10% noise).
where γ j k (z, ω) is the coherence function between two [4] Specification for high-rise structure design (GBJB5-90). Beijing
variables v̂ j (t) and v̂k (t), and is defined as (PCR): China Construction Industry Publishing House; 1991.
! " [5] Chen J, Li J. Study on the inverse wind load identification of tall
ω C z z building. Chinese Quarterly of Mechanics 2001;22(1):72–7.
γ j k (z, ω) = exp − (A.11) [6] Law SS, Chan THT, Zeng QH. Moving force identification—time
2π 12 v̄(z j ) + v̄(z k ) domain method. Journal of Sound and Vibration 1997;201(1):1–22.
[7] Zhu XQ, Law SS. Moving force identification on a multi-span
in which C z is the exponential decay coefficient, and z = continuous beam. Journal of Sound and Vibration 1999;228(2):
|z j − z k |. 377–96.
The auto-spectral density is proposed by Kaimal et al. [8] Simiu E, Scanlan RH. Wind effects on structures. 3rd ed. New York:
[21], and the values of the auto-spectral density vary with Wiley; 1996.
the height above the ground as [9] Liu H. Wind engineering—A handbook for structural engineers.
Englewood Cliffs, NJ; 1991.
1 105u 2∗ z 1 [10] Benfratello S, Falsone G, Muscolino G. Influence on the quadratic
S(z, ω) = (A.12)
2 2π v̄(z) 5 term in the along-wind stochastic response of SDOF of structures.
3
1 + 33 2πzω
v̄(z)
Engineering Structures 1996;18(8):685–95.
[11] Deodatis G. Simulation of ergodic multivariate stochastic processes.
Journal of Engineering Mechanics 1996;22(8):778–87.
where v̄(z) is the mean speed at the z level, and the other
[12] Juang JN. Applied system identification. New Jersey; 1994.
parameters are the same as in previous section. u ∗ is the wind [13] Tikhonov AN, Arsenin VY. Solution of ill-posed problems. New
friction speed calculated as York: Wiley; 1977.
k v̄(z ) [14] Hansen PC. Analysis of discrete ill-posed problems by means of the
u∗ = (A.13) L-curve. SIAM Review 1992;34(4):561–80.
ln(z /z 0 ) [15] Wu HO, Ren X. Finite element analysis for structure. Beijing (PCR):
China Railway Publishing House; 2000.
where k is a constant, which can be approximately taken as
[16] Ernst HJ. Der E-Modul von Seilen unter Berucksichtigung des
0.4; z 0 is the roughness length which varies with the type of Durchhanges. Der Bauingenieur 1965;40(2):52–5.
terrain. [17] Raman NV, Surya Kumar GV, Sreedhara Rao VV. Large
displacement analysis of guyed towers. Computers & Structures 1987;
28(1):93–104.
[18] Guyan RJ. Reduction of stiffness and matrices. AIAA Journal 1965;
References
3(2):380.
[19] Ben Kahla N. Equivalent beam–column analysis of guyed towers.
[1] Zhang XT. A handbook for wind load computation in engineering Computers & Structures 1995;55(4):631–45.
design. Beijing (PCR): China Construction Publishing House; 1998. [20] Law SS, Bu JQ, Zhu XQ, Chan SL. Wind characteristics of typhoon
[2] Kareem A, Kijewski T. 7th US national conference on civil Dujuan and analytical and experimental studies of its effects on a 50 m
engineering: a summary of papers. Journal of Engineering and guyed mast. Engineering Structures [under review].
Industrial Aerodynamics 1996;62:81–129. [21] Kaimal JC, Wyngaard JC, Izumi Y, Coté OR. Spectral characteristics
[3] Kolousek V, Pirner M, Fischer O. Wind effects on civil engineering of surface-layer turbulence. Journal Royal Meteorology Society 1972;
structures. Amsterdam: Elsevier; Prague: Academia; 1984. 98:563–89.