Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43

Contents lists available at ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

An analytical model for the fluctuating wind velocity spectra MARK

of a moving vehicle
⁎
Xiao-Zhen Lia, Jun Xiaoa, , De-Jun Liua,b, Ming Wanga, De-Yi Zhanga,c
a
Department of Bridge Engineering, Southwest Jiaotong University, 610031 Chengdu, Sichuan, China
b
MOE Key Laboratory of High-Speed Railway Engineering, Southwest Jiaotong University, 610031 Chengdu, Sichuan, China
c
Department of Engineering Mechanics, Ontario Power Generation, Pickering, Ontario, Canada L1W 3J2

A R T I C L E I N F O A BS T RAC T

Keywords: An analytical model for both the longitudinal and lateral fluctuating wind velocity spectra of a moving train in
Moving train the random wind field has been developed based on Cooper's theory (Cooper, 1984). The fluctuating wind
Fluctuating wind velocity spectrum velocity spectrum of a moving point can be derived through the weighted linear superposition of the
Cooper's theory longitudinal and lateral spectra of the stationary point with both frequency and amplitude scaling. The
Taylor's frozen turbulence hypothesis
proposed analytical model has been validated systematically, and based on the Simiu spectrum (Simiu and
Doppler effect
Scanlan, 1996), parametric studies were conducted for the influence on the fluctuating spectra of a moving train
by wind flow angle from 0°, 30°, 60°, 90°, to 150° and varying speed ratios between vehicle and wind (mean
wind velocity of 20 m/s) from 1 to 4. It is revealed that the fluctuating wind velocity spectrum experienced by a
moving train exhibits a Doppler effect phenomena due to the train movement, especially in the longitudinal
case. The proposed spectrum model can provide an efficient simulation scheme for the fluctuating wind velocity
time series of the moving train with adequate accuracy and can provide a rigorous analytical scheme for the
subsequent aerodynamic response analysis of the moving train within the random wind field.

1. Introduction
With the increased tendency for higher speed of the public
transportation system like the high-speed trains, the aerodynamic
environment of the moving train has gradually become more of an
essential factor than the wheel-rail system in design and development
of those high-speed trains. Under a strong crosswind, the transverse
force of the train will experience an abrupt increase so that the train's
aerodynamic performance will degrade significantly, resulting in the
risk of train derailment or overturning and severe threats to the safe
operation of a high-speed train. There have been many severe train
derailment or overturning accidents worldwide owing to a strong cross
wind. For example, on February 28, 2007, a strong crosswind has
caused a serious train derailment accident in Xinjiang Province, China
(shown in Fig. 1a) and led to at least 4 fatalities and more than 30
injured. The accident region is known as a "hundred-kilometer wind
zone" (with at least level 8 Gale wind more than 320 days of each year)
where the traffic of the train can be affected significantly. Similarly, a
strong wind overturned three trucks driving on the Fujian Qingzhou
Bridge (in Qingzhou City, Fujian Province, China) on September 1,
2005 (shown in Fig. 1b).
To investigate the wind effects on the safe operability and service-
ability of moving train, it is essential to derive a random fluctuating
wind velocity model that can be used for the subsequent aerodynamic
analysis of a moving train. The majority of the existing studies tend to
focus on the simulation of the wind velocity field in the time domain
(wind loading) based on the fixed-point multivariable stationary
random process method.
By utilizing the simulated fluctuating wind velocity time series for a
moving train, the dynamic response analysis of wind-train-track
system or wind-train-bridge coupling system can then be performed
in the time domain (Cai and Chen, 2004; Li et al., 2005; Xu and Guo,
2003). The traditional way of obtaining the transient fluctuating wind
velocity time series is based on the discrete point extraction and
neighbor interpolation methods within the multivariable random
process simulation scheme (Shinozuka, 1971). However, the tradi-
tional simulation method usually has a drawback with simulation error
due to the artificial dispersion and therefore requires an increase of the
simulation points to compensate the error, as a result, significant
computation effort has been induced. Although some improvements on
the multivariable random process simulation method had been made
by several researchers (Ding et al., 2011; Huang et al., 2013; Chen

⁎
Corresponding author.
E-mail address: jun.xiao@my.swjtu.edu.cn (J. Xiao).

http://dx.doi.org/10.1016/j.jweia.2017.02.007
Received 25 August 2016; Received in revised form 5 February 2017; Accepted 6 February 2017
0167-6105/ © 2017 Elsevier Ltd. All rights reserved.
X.-Z. Li et al. Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43

Fig. 1. Vehicles overturn due to the strong wind [http://news.xinhuanet.com/society/2007-02/28/content_5781677.htm; http://news.sina.com.cn/c/2005-09-02/09236844314s.

shtml?qq-pf-to=pcqq.c2c]. a) Train overturn in Xinjiang Province, China. b) Lorry overturn on bridge in Fuzhou City, China.

Based on the developed analytical model, parametric studies were

conducted to investigate the influence on the fluctuating wind velocity
spectrum of moving train by wind flow angle and speed of train. The
results drawn from this paper can be used to qualitatively explain the
frequency shift phenomenon on fluctuating wind velocity spectrum and
can also provide a rigorous analytical scheme for the subsequent
aerodynamic response analysis of the moving train within the random
wind field.

2. Analytical model and the validation

2.1. Problem statement

This section briefly presents the typical fluctuating wind velocity

time series simulation results for a moving train in the random wind
Fig. 2. Wind velocity extraction scheme for the moving train in the simulated fluctuating field based on multivariate random process (which is the stationary
wind field (Chen et al., 2014).
ergodic process) simulation scheme by Chen et al. (2014). This section
can be taken as a problem statement for the developed analytical model
et al., 2014); however, there are still challenging issues needing to be
of this paper. Details of the time domain simulation scheme by Chen
addressed. For example, the time domain method lacks of a compre-
et al. (2014) are not presented for brevity.
hensive systematic analytical scheme in accounting for the effect of
Fig. 2 presents a typical spatial-temporal wind-field simulation
different wind flow angle with the moving train and significant
result of the fluctuating wind velocity component and the wind velocity
computation efforts are induced by increase of the number of simula-
extraction scheme from the simulated fluctuating wind field for a
tion and discretion points.
moving train. When the train moves to the point P1/P2 at time t1/t2,
In addition to the time domain method, Cooper (1984) initially
respectively, the transient fluctuating wind velocities (loadings from
derived an analytical model in the frequency domain for the long-
the natural wind) exerting on the train equal to the transient fluctuat-
itudinal fluctuating wind velocity spectrum (normal to the train moving
ing wind velocity value of the fixed point P1/P2 at time t1/t2 in the
direction) of a moving point based on Von Karman spectrum. Using
spatial-temporal wind-field. In this paper, the equivalent wind velocity
Cooper's theory (Cooper, 1984), Wu et al. (2014) developed both the
resulting from the moving of the train (which is opposite to the velocity
longitudinal and lateral fluctuating wind velocity spectra of a moving
of train) is considered to be an average wind velocity component
point through the semi-analytical fit of the autocorrelation coefficient
without fluctuation. It is seen that the way of obtaining the wind
function corresponding to the Simiu spectrum (Simiu and Scanlan,
velocity time history of a moving train by simulating the entire
1996). The analytical method in the frequency domain differs from the
fluctuating wind velocity field is obviously inefficient.
time domain simulation method in that the multivariate random
Based on the scheme in Fig. 2, the wind velocity time history of a
process can be converted into a fluctuating wind velocity spectra (the
moving point will be extracted and the corresponding spectrum will be
longitudinal and lateral), then the simulation of a fluctuating wind
compared with that for the stationary point in the following example.
velocity time series become more efficient based on the univariate
Taking the average wind velocity of U=20 m/s, height of z=10 m,
random process. Based on the analytical fluctuating wind velocity
surface roughness of z0=0.03, Von Karman constant of k=0.4, train
model by Cooper (1984), both the steady and unsteady aerodynamic
speed of V=60 m/s, simulated time step of 0.1 s, simulated spacing
responses of the moving train within the random wind field have been
interval of 6 m, and wind field spatial and temporal simulation points
investigated (Baker, 1991a, 1991b, 2010; Cheli et al., 2003a, 2003b,
of 4096. Since the distance travelled by moving point at each time
2007; Yu et al., 2016).
interval equal the simulated spacing interval (60m / s × 0.1s = 6m ), the
Following these discussions, based on Cooper's theory, this paper
moving point is always located at the spatial simulation point at each
develops a systematic analytical model for both the longitudinal and
sampling time point, making the extracted instantaneous wind velocity
lateral fluctuating wind velocity spectra of moving train in the random
of the moving point being exactly same as the simulated value at each
wind field, considering the effects of different wind flow angle and
simulation point of the wind field. The time histories of fixed stationary
ratios of train speed to wind velocity. The proposed spectrum model
points (travel path of moving train) and moving point extracted are
can provide an efficient simulation scheme for the fluctuating wind
presented in Fig. 3. The comparison of estimated power spectrum
velocity time series acting on the moving train with adequate accuracy.
between the stationary point and moving point is given in Fig. 4.

35
X.-Z. Li et al.
Fig. 3. Simulated sample wind time histories (Chen et al., 2014).
Fig. 4. Estimated longitudinal fluctuating wind velocity spectrum for stationary and
moving points (Chen et al., 2014).
It is seen from Fig. 4 that the energy of the high frequency
component for the moving point wind velocity spectrum is greater
than that of the stationary point, while the energy in the low frequency
range is smaller than that of the stationary point. This frequency shift
phenomenon on the fluctuating wind velocity spectrum will be
modelled and explained in the following sections. In addition, the time
domain simulation results will also be employed to validate the
proposed analytical model of this paper. It should be noted that, since
the extracted instantaneous wind velocity of the moving point is exactly
same as the simulated value at each simulation point of the wind field,
the simulation error owing to artificial dispersion has been avoided.
2.2. Analytical model of fluctuating wind velocity spectrum of moving
train
Taylor's frozen turbulence hypothesis was primarily used to model
the turbulence correlation between two stationary points in a random
wind field (Taylor, 1938). Taylor made the hypothesis that if the mean
wind velocity U is sufficiently larger than the velocity of the turbulence
u(t), the turbulence flow pattern does not have time to change and
passes a stationary fixed point with the mean wind speed (i.e., the
turbulence field can be considered to be frozen in space). The changes
in fluctuation at the fixed point is simply due to the passage of an
unchanging pattern of turbulent motion over the point (ESDU, 2010).
Fig. 5 presents a mean wind velocity of U in parallel with x-y plan with
turbulence components of u, v, and w in x, y, and z direction
respectively. The correlation between stationary physical point P at
36
Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43
Fig. 5. Physical and equivalent points in a frozen turbulence field (Cooper, 1984).
time t and point P′ at t + τ can be transformed into the correlation
between point P and the equivalent point P′e in frozen field (with zero
time lag).
Based on the Taylor's frozen turbulence hypothesis and the
isotropic turbulence assumption, the cross-correlation coefficient func-
tion (for i = u , vorw ) of two fixed points P and P′ at any position can be
expressed as the cross-correlation coefficient function of points P and
P′e (ESDU, 2001) by
f−g 2
ρii (Δr; τ )sym = Δsi + g
Δre2 (1)
where Δr is the distance between two fixed separate points P and P′;
f , g are the longitudinal and lateral auto-correlation coefficient func-
tion, respectively; Δre is the equivalent separation distance for non-zero
time lag (i.e., the distance between points P and P′e); Δsi is component
of Δre resolved in direction in which gust component i is acting.
The turbulence cross-correlation coefficient function between P and
P′ (i.e., Eq. (1)) can be alternatively and approximately expressed as
ρ (Δr; τ ) = c f (r∼) + (1 − c ) g (r∼)
ii ii ii (2)
where f (r∼) and g (r∼), respectively, represent the longitudinal and lateral
auto-correlation coefficient function; when i = u ,cuu = (Δxe / Δre )2 ; when
i = v , cvv = (Δye / Δre )2 (the correlation between different turbulence
components in u, v, w directions are not modelled for simplicity).
As shown in Fig. 6, a train is moving at a constant speed V within
the fluctuating wind velocity field with an average wind velocity of U,
the angle between the moving direction of train and the direction of
average wind velocity is θ (wind flow angle), and P denotes the fixed
point attached to the train.
Based on Eq. (2) and Taylor's frozen turbulence hypothesis, the
auto-correlation coefficient function of point P considering the train
movement can be expressed as
ρiiM (P; τ ) = ρii (P, Pe; τ = 0) = cii f (r∼) + (1 − cii ) g (r∼) (3)
where the superscript M denotes the case considering the train
Fig. 6. Geometry of two points moving with the vehicle and equivalent point (Cooper,
1984).
X.-Z. Li et al. Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43

Fig. 8. Comparison of the longitudinal wind velocity power spectrum with Cooper
(1984).

+∞ +∞
Fig. 7. The scaling factor α . 4σu2 ∫0 f (r∼′)cos(2πnτ ) dτ = 4σu2 ∫0 f (αr∼)cos(2πnτ ) dτ
(9-1)

movement. Assuming the train is moving on the x-y plane, i.e., Δze = 0 , +∞ +∞
based on Fig. 6, we have Δxe = −τV cos(θ ) − τU , Δye = −τV sin(θ ), 4σv2 ∫0 g (r∼′)cos(2πnτ ) dτ = 4σv2 ∫0 g (αr∼)cos(2πnτ ) dτ
(9-2)
Δr = Δx 2 + Δy 2 + Δz 2 , z = (z + z′)/2 , r∼ = Δr / z (where z is the
e e e e m e m
height of point P in Fig. 6 above ground at time t and z′ is the height Eqs. (9–1) can then be re-written as
of point P at time t + τ ), the following equations can be then derived as +∞ 1 +∞ ⎛ n ⎞
4σu2 ∫0 f (αr∼)cos(2πnτ ) dτ = 4σu2 ∫0f (αr∼)cos ⎜2π ⋅ατ ⎟ d (ατ )
V2 cos(θ ) + U2
+ 2VU cos(θ ) α ⎝ α ⎠
cuu = (Δxe / Δre )2 =
V 2 + U2 + 2VU cos(θ ) (4) 1 +∞ ⎛ n ⎞ 1 ⎛ n ⎞
= 4σu2 ∫0 f (αr∼)cos ⎜2π ⋅ατ ⎟ d (ατ ) = Su ⎜ ⎟
α ⎝ α ⎠ α ⎝α⎠
V 2 sin2 (θ )
cvv = (Δye / Δre )2 = (10)
V2 + U2 + 2VU cos(θ ) (5)
Similarly, Eqs. (9–2) can then be expressed as
Based on Wiener–Khinchin theorem provided in Appendix A, a
correlation function will remain back to the original form after a +∞ 1 ⎛n⎞
4σv2 ∫0 g (r∼′)cos(2πnτ ) dτ = Sv ⎜ ⎟
Fourier transform and then a subsequent inverse Fourier transform. α ⎝α⎠ (11)
The semi-analytical models of the longitudinal and lateral fluctuating
wind velocity spectra of a moving point (in Appendix B) were Based on Eqs. (7) to (11), the longitudinal fluctuating wind velocity
developed by numerical fit of the autocorrelation function derived by spectrum of moving train can be derived as
inverse transform of the Simiu power spectrum (Wu et al., 2014). The
drawback of the semi-analytical results can be resolved by the full 1⎡ ⎛n⎞ σ2 ⎛ n ⎞⎤
SuM (n ) = ⎢cuu Su ⎜ ⎟ + u2 (1 − cuu ) Sv ⎜ ⎟ ⎥
analytical scheme presented in the following. α⎣ ⎝ α ⎠ σv ⎝ α ⎠⎦ (12)
Based on relationship between bilateral and one-side power spec-
trum and Wiener–Khinchin theorem in Appendix A, the auto-power Similarly, the lateral fluctuating wind velocity spectrum of moving
spectrum of longitudinal fluctuating wind velocity for moving train can train is
be written as
1 ⎡ σv2 ⎛n⎞ ⎛ n ⎞⎤
SvM (n ) = ⎢ 2 cvv Su ⎜ ⎟ + (1 − cvv ) Sv ⎜ ⎟ ⎥
SuM (n ) = 4 ∫
+∞
RuM (τ )cos(2πnτ ) dτ = 4(σuM )
2 +∞ M α ⎣ σu ⎝ α ⎠ ⎝ α ⎠⎦ (13)
0
∫0
ρuu (τ )cos(2πnτ ) dτ
+∞
[cuu f (r∼′) + (1 − cuu ) g (r∼′)]cos(2πnτ ) dτ
2
= 4(σuM ) ∫ It is seen from Eqs. (12) and (13) that the longitudinal and lateral
0
fluctuating wind velocity spectra for the moving train can be derived by
(6)
a weighted linear superposition of both longitudinal and lateral
where r∼′ = r∼ V 2 + U2 + 2VU cos(θ )
. fluctuating wind velocity spectra of the stationary fixed point, along
U2 with the scaling for both amplitude and frequency by parameter α . The
Eq. (6) can be alternatively expressed by separating the longitudinal
scaling factor α in Eq. (8) is a function of the ratio between train
and lateral auto-correlation functions as
moving speed and mean wind velocity (i.e., speed ratio V/U) and the
⎧ +∞ ⎫ wind flow angle with direction of the moving train (i.e., wind flow angle
SuM (n ) = cuu ⎨4σu2 ∫0 f (r∼′)cos(2πnτ ) dτ ⎬ θ ). Fig. 7 presents a plot of α . It reveals from Fig. 7 that (1) when θ is
⎩ ⎭
smaller than 90°, the frequency content of the power spectrum shifts to
⎧ +∞ ⎫
+ (1 − cuu ) ⎨4σu2 ∫0 g (r∼′)cos(2πnτ ) dτ ⎬ the high-frequency (i.e., α is greater than 1); (2) when V/U > 1, the
⎩ ⎭ (7) extent the frequency shift to the higher range increases with the
increase of the train speed. It is interesting that when the train moving
where r∼′ = αr∼ representing the time domain variation of the
direction aligns with the direction of the wind (i.e.,θ = 90°) and the
correlation function for a moving point in comparison to that of a
train speed equals wind velocity (i.e., V/U=1), there is no fluctuating
stationary fixed point, and r∼ = z is a dimensionless parameter in the
τU
wind velocity (wind loading) experienced on the train and the
time domain, and the scaling factor of α is
fluctuating wind appears stationary relative to the moving train
U2 + V 2 + 2UV cos(θ ) (pattern of the turbulent motion assumed unchanged as discussed
α= above).
U2 (8)
According to ESDU 86010 (ESDU, 2001), the standard deviation of
The right part of Eq. (8) can be separated into two parts as each fluctuation component has the following forms

37
X.-Z. Li et al. Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43

Fig. 9. Comparison of longitudinal fluctuating wind velocity spectrum with Wu et al. (2014).

σv ⎛π z ⎞ fluctuating wind velocity spectrum of moving train under different

= 1 − 0.22cos4 ⎜ ⎟
σu ⎝2 h⎠ speed can be obtained. The comparison between the results of the
⎛π z ⎞ developed analytical model and the one by Cooper (1984) is presented
σw
= 1 − 0.45cos4⎜ ⎟ in Fig. 8.
σu ⎝2 h⎠ (14) It is observed from Fig. 8 that, the results from the proposed model
where h is the atmospheric boundary layer height that can be are in good agreement with the results by Cooper (1984). As the speed
approximated as h = (1/6) u* × 10 4 , z is the height above plane, u* is of the train increases, the longitudinal fluctuating wind velocity
the friction velocity that can be obtained as u* = U10 /(2.5 ln(10/ z 0 )), z0 is spectrum shifts toward the high frequency, at the same time, the peak
surface roughness parameter, and U10 is the mean wind velocity at amplitude increases to some extent. The observations are consistent
z=10 m over the site. with those obtained from Fig. 7.

2.3. Validation of the proposed analytical model 2.3.2. Example 2: Simiu spectrum
Wu et al. (2014) presented the semi-analytical models of the
2.3.1. Example 1: Von Karman spectrum longitudinal and lateral fluctuating wind velocity spectrum for the
Cooper (1984) derived the analytical model of the longitudinal moving train based on Simiu spectrum (Simiu and Scanlan, 1996)
fluctuating wind velocity spectrum of moving train (i.e.,θ = 90°) based having the expressions below
on the Von Karman spectrum (Karman, 1937) with the following
n⋅Su (n ) 200nˆ
expression =
u*2 (1 + 50nˆ)5/3 (17)
n⋅Suu 4nu
= ; nu = xLun / Uz
σu2 [1 + 70.8nu2]
5/6
(15) n⋅Sv (n ) 15nˆ
=
u*2 (1 + 9.5nˆ)5/3 (18)
n⋅Sii 4ni (1 + 755.2ni2 )
2
= 11/6
; ni = xLin / Uz; i = v or w where Su (n ) and Sv (n ) are the longitudinal and lateral fluctuating power
σi (1 + 283.2ni2 ) (16)
spectrum, respectively, for the stationary point, n the Hertz frequency
The parameter values of this validation example include the mean (Hz), nˆ = nz / U the non-dimensional frequency, and u* the friction
wind velocity of U = 10m / s and the turbulence length scale of u- velocity (m/s).
component in x-direction of xLu = 45m , based on which the longitudinal In this validation example, we choose the wind flow angle of

Fig. 10. Comparison of lateral fluctuating wind velocity spectrum with Wu et al. (2014).

38
X.-Z. Li et al.
Fig. 11. Comparison with the time domain modelling results by Chen et al. (2014).
θ = 90°, height of z = 10m , average wind velocity of U = 10m / s , the
frequency range for n of [0.001,10], and the ratio of train speed to wind
velocity of 1 and 2, respectively. The calculated results with comparison
are presented in Figs. 9 and 10.
As seen from Figs. 9 and 10, one can conclude that the results by
the proposed model agree well with those by Wu et al. (2014) for
modelling of both longitudinal and lateral wind velocity power
spectrum of the moving train. The difference between these two
methods is that, the proposed model is a fully analytical one while
the one by Wu et al. (2014) is based on the numerical fitting (see the
small peaks and troughs of curves in Figs. 9 and 10).
2.3.3. Example 3: Time domain modelling approach
The fluctuating wind velocity time series for the moving train in the
spatial-temporal wind field are simulated based on multivariate ran-
dom process simulation scheme by Chen et al. (2014) (as discussed in
Section 2.1). The simulated time series are used to validate the
proposed analytical models in the frequency domain. The estimated
power spectra of the simulated wind velocity time series, the predicted
power spectrum from the proposed model of this paper, and the similar
results for the stationary fixed points are provided in Fig. 11 for V/U=1
and 2, respectively.
It reveals from Fig. 11 that the proposed analytical model can
provide an accurate modelling of the fluctuating wind velocity of
moving train, and it can also account for the frequency shift phenom-
ena of fluctuating wind velocity power spectrum (as discussed in
Section 2.1). The proposed spectrum model can provide an efficient
Fig. 12. The influence of wind flow angle on longitudinal fluctuating wind velocity spectrum.
39
Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43
simulation scheme for the fluctuating wind velocity time series acting
on the moving train with adequate accuracy, and can provide a rigorous
analytical scheme for the subsequent aerodynamic response analysis of
the moving train within the random wind field.
3. Numerical examples
Based on the developed analytical model, parametric studies were
conducted to investigate the influence on the fluctuating wind velocity
spectrum of moving train by wind flow angle and speed of train. In the
numerical example, the average wind velocity U=20 m/s, the height of
simulated point z=10 m, the roughness coefficient z0=0.03 and Karman
constant k=0.4 are used to calculate both the longitudinal and lateral
fluctuating wind velocity spectrum under different wind flow angles
and speed ratios in the following.
3.1. Influence of wind flow angle with train moving direction
Due to the complex travelling environment, high-speed train often
encounters inclined natural wind relative to its movement direction. As
discussed, the wind flow angle directly affects the fluctuating wind
velocity spectrum experienced by the moving train, at the same time
under the same speed ratio, the wind flow angle will affect the actual
yaw angle of the train. Therefore, the wind flow angle needs to be
taken into account in the modelling of the fluctuating wind velocity
spectrum. The influence of wind flow angle on both the longitudinal
and lateral fluctuating wind velocity spectra, with the ratio of
X.-Z. Li et al. Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43

Fig. 13. The influence of wind flow angle on lateral fluctuating wind velocity spectrum.

train speed to wind velocity of 1 and 4, respectively, is presented in
Figs. 12 and 13.
It is seen from Fig. 12 that, under the same ratio of train speed to
wind velocity, the longitudinal fluctuating wind velocity spectrum shifts
toward the high frequency as the wind flow angle increases. The greater
the wind flow angle, the more the frequency shifts. While Fig. 13
reveals that the lateral fluctuating wind velocity spectrum shifts toward
the low frequency with the increase of wind flow angle.
3.2. Influence of speed ratio between moving train speed and the
mean wind velocity
The influence of speed ratio (the ratio of train speed to wind velocity
varying from 1 to 4) on both the longitudinal and lateral fluctuating
wind velocity spectra, with the wind flow angles of 30–150°, respec-
tively, is presented in Figs. 14 and 15.
Fig. 14 indicates that, the movement of train can cause the
longitudinal fluctuating wind velocity spectrum shift obviously toward
the high frequency and the amplitude of the power spectrum exhibits
various degree of change. However, the overall trend reveals that the
power spectrum shifts toward the high frequency, and the greater the
wind flow angle close to 90°, the more the frequency shifts.
When the wind flow angle is 90° (i.e., the movement of train is
normal to the wind direction), the predominant frequency in the
longitudinal fluctuating wind velocity spectrum of the moving train is
close to 1 (shown in Fig. 14c), which is close to the natural frequency of
the typical train suspension system in China (Zhai, 2016). For the

Fig. 14. The influence of speed ratio on longitudinal fluctuating wind velocity spectrum.

40
X.-Z. Li et al.
Fig. 15. The influence of speed ratio on lateral fluctuating wind velocity spectrum.
lateral fluctuating power spectrum (shown in Fig. 15), when the wind
flow angle is less than 90° (i.e., the train moves against the wind), the
power spectrum will shift toward the high frequency and the higher the
speed, the more the frequency shifts. When the wind flow angle is more
than 90° (i.e., the train moves away from the wind), the lateral
fluctuating wind power spectrum will shift toward the low frequency
and the larger the wind flow angle, the more the frequency shifts. The
lateral fluctuating wind power spectrum will shift toward the high
frequency as the speed of the train increases. It is clearly observed that,
when the train is moving in a random wind field, there is a similar
Doppler effect, which is useful in qualitative understanding of the
fluctuating wind velocity spectrum experienced by the high-speed train
in a random wind field.
4. Conclusions
Based on the Taylor's frozen turbulence hypothesis and the
isotropic turbulence assumption, an analytical model for both the
longitudinal and lateral fluctuating wind velocity spectra of moving
train in the random wind field has been developed. The following
conclusions can be drawn from this research:
(1) The fluctuating wind velocity spectra of a moving point can be
derived through the weighted linear superposition of the long-
itudinal and lateral wind velocity spectra of the stationary point
with both frequency and amplitude scaling. During the model
derivation, the auto-correlation function has been separated into
two parts with the respective Fourier transform without the
intermediate transform performed by the semi-analytical ap-
proach. The proposed spectrum model can provide an efficient
41
Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43
simulation scheme for the fluctuating wind velocity time series
with adequate accuracy, and can provide a rigorous analytical
scheme for the subsequent aerodynamic response analysis of the
moving train within the random wind field.
(2) The proposed analytical model has been systematically validated.
Parametric studies were conducted to investigate the influence on
the fluctuating wind velocity spectra of moving train by wind flow
angle and speed ratio between vehicle and wind.
(3) When wind flow angle is smaller than 90°, the frequency content of
the power spectrum shifts to the high-frequency; when speed ratio
is greater than 1, the extent the frequency shifts to the higher range
increases with the increase of the train speed; when the train
moving direction aligns with the direction of the wind and the train
speed equals wind velocity, there is no fluctuating wind velocity
experienced on the train and the fluctuating wind appears sta-
tionary relative to the moving train.
(4) The fluctuating wind velocity spectrum experienced by a moving
train exhibits a phenomena of Doppler effect due to the train
movement, especially for the longitudinal one.
(5) When the train speed reaches 60 m/s, the predominant frequency
in the longitudinal fluctuating wind velocity spectrum of the
moving train is close to 1, which is close to the natural frequency
of the typical train suspension system in China and has significant
effect on the design of the train.
Acknowledgments
The research reported in this paper was supported in part by the
National Key Basic Research Program of China (973 Program) under
Grant No. 2013CB036301, the National Natural Science Foundation of
X.-Z. Li et al. Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43

China under Grant No. U1434205, and Fundamental Research Funds for thank the anonymous reviewers for their constructive comments and
the Central Universities under Grant No. 2682016CX020. The authors advice, which greatly improved the quality of this manuscript.

Appendix A. Wiener–Khinchin theorem and bilateral and one-side power spectrum

Wiener–Khinchin theorem states that the autocorrelation function of a wide-sense-stationary random process has a spectral decomposition
given by the power spectrum of that process. According to the theorem, the relationship between bilateral power spectrum GX (ω) and auto-
correlation function RX (τ ) can be obtained as
+∞
G X (ω ) = ∫−∞ RX (τ ) e−jωτ dτ
(A.1)

1 +∞
RX (τ ) =
2π
∫−∞ GX (ω) e jωτ dω
(A.2)
Through Euler's formula, Eq. (A.1) can be rewritten as
+∞ +∞
G X (ω ) = ∫−∞ RX (τ )(cos ωτ − j sin ωτ ) dτ = 2 ∫0 RX (τ )cos ωτdτ
(A.3)
Since one-side spectrum model usually has the engineering significance, the engineering one-side spectrum can be expressed as
⎧ 2G (ω) ω ≥ 0
SX (ω) = ⎨ X
⎩ 0 ω<0 (A.4)
Furthermore,
+∞ +∞
SX (ω) = 4 ∫0 RX (τ )cos ωτdτ = 4σX2 ∫0 ρX (τ )cos ωτdτ
(A.5)

Appendix B. Semi-analytical wind velocity spectra model of moving train

Based on Eq. (3), the longitudinal power spectrum of moving point can be derived by following equation,
+∞ 2 +∞
SuM (n ) = 4 ∫ RuM (τ )cos(2πnτ ) dτ = 4(σuM ) ∫ ρuuM (τ )cos(2πnτ ) dτ
0 0
2 +∞
= 4(σuM ) ∫
[cuu f (r∼′) + (1 − cuu ) g (r∼′)]cos(2πnτ ) dτ (B.1)
0

V 2 + U2 + 2VU cos(θ )
where r∼′ = r∼ .
U2
According to Wiener–Khinchin theorem, power spectrum of stationary random process is obtained by the Fourier transform of its auto-
correlation function and the auto-correlation function can be derived through the inverse Fourier transform of power spectrum density function.
Thus, the longitudinal auto-correlation function of stationary point can be derived by
∞
ρuu (τ ) = f (r∼) =
1
σu2
∫0 Su (n)cos(2πnτ ) dn
⎛ 11 1 1 ⎞
=1− 2500 π 2r∼2 + ⋅52/3π 7/6r∼7/6LommelS 2 ⎜ ,
9 9
, πr ⎟
500 ⎝ 6 2 25 ⎠ (B.2)
where LommelS 2(μ, ν, x ) is the second kind of solution of Lommel differential equation (Gradsbteyn and Ryzbik, 2000).
Similarly, the lateral auto-correlation function of stationary point is
∞
ρvv (τ ) = g (r∼) =
1
σv2
∫0 Sv (n)cos(2πnτ ) dn
⎛ 11 1 4π ∼⎞
⋅195/6⋅21/3⋅π 7/6⋅r∼7/6LommelS 2 ⎜ ,
36 9
=1− 361 r̃ 2 + , r⎟
361 ⎝ 6 2 19 ⎠ (B.3)
Since the energy of the fluctuating component remains unchanged with the movement of the train, we have (σuM )2 = σu2 , and Eqs. (B.1), (B.2) and
(B.3) can finally result in the longitudinal auto-power spectrum of the moving train as
+∞
SuM (n ) = 4σu2 ∫ [cuu f (r∼′) + (1 − cuu ) g (r∼′)]cos(2πnτ ) dτ
0
⎧ +∞ ⎛ τ U2 + V 2 ⎞
⎫ ⎧ +∞ ⎛ τ U2 + V 2 ⎞
⎫
=cuu ⎨4σu2 ∫ f⎜ ⎟ cos(2πnτ ) dτ ⎬ + (1 − cuu ) ⎨4σu2 ∫ g⎜ ⎟ cos(2πnτ ) dτ ⎬
⎩ 0 ⎝ z ⎠ ⎭ ⎩ 0 ⎝ z ⎠ ⎭ (B.4)
∞
where σu2 = ∫0 Su (n ) dn = 6u*2 ,u* = kU / log(z / z 0 ).

References
Baker, C.J., 1991a. Ground vehicles in high cross winds part I: steady aerodynamic forces.
J. Fluids Struct. 5, 69–90. http://dx.doi.org/10.1016/0889-9746(91)80012-3.
Baker, C.J., 1991b. Ground vehicles in high cross winds part II: unsteady aerodynamic
forces. J. Fluids Struct. 5, 91–111. http://dx.doi.org/10.1016/0889-9746(91)80013-4.
Baker, C.J., 2010. The simulation of unsteady aerodynamic cross wind forces on trains. J.
Wind Eng. Ind. Aerodyn. 98, 88–99. http://dx.doi.org/10.1016/j.jweia.2009.09.006.
Cai, C.S., Chen, S.R., 2004. Framework of vehicle–bridge–wind dynamic analysis. J.

42
X.-Z. Li et al.
Wind Eng. Ind. Aerodyn. 92, 579–607. http://dx.doi.org/10.1016/
j.jweia.2004.03.007.
Cheli, F., Corradi, R., Diana, G., Tomasini, G., 2003a. A numerical experiments approach
to evaluate the aerodynamic effects on rail vehicle dynamics. In: Proceedings of the
18th IAVSD Dynamic of Vehicles on Roads and Tracks, Atsugi, Japan.
Cheli, F., Corradi, R., Diana, G., Tomasini, G., Roberti, R., Cheli, R., 2003b. Cross wind
aerodynamic forces on rail vehicles-Part II numerical model for vehicle dynamic
response computation. In: Proceedings of the World Congress on Railway Research,
Edinburgh.
Cheli, F., Corradi, R., Diana, G., Ripamonti, F., Rocchi, D., Tomasini, G., 2007.
Methodologies for assessing trains CWC through time-domain multi-body
simulations. In: Proceedings of the 12th International Conference on Wind
Engineering, Cairns, Australia.
Cooper, R.K., 1984. Atmospheric turbulence with respect to moving ground vehicles. J.
Wind Eng. Ind. Aerodyn. 17, 215–238. http://dx.doi.org/10.1016/0167-6105(84)
90057-6.
Chen, N., Li, Y., Xiang, H., 2014. A new simulation algorithm of multivariate short-term
stochastic wind velocity field based on inverse fast Fourier transform. Eng. Struct.
80, 251–259. http://dx.doi.org/10.1016/j.engstruct.2014.09.012.
Ding, Q., Zhu, L., Xiang, H., 2011. An efficient ergodic simulation of multivariate
stochastic processes with spectral representation. Probab. Eng. Mech. 26 (2),
350–356. http://dx.doi.org/10.1016/j.probengmech.2010.09.006.
ESDU, 2001. Characteristics of Atmospheric Turbulence Near the Ground. Part III:
Variations in Space and Time for Strong Winds (Neutral Atmosphere). Item NO.
86010, Engineering Sciences Data Unit, London.
ESDU, 2010. Characteristics of Atmospheric Turbulence Near the Ground. Part I:
Definitions and General Information. Item No.74030, Engineering Sciences Data
43
Journal of Wind Engineering & Industrial Aerodynamics 164 (2017) 34–43
Unit, London.
Gradsbteyn, I.S., Ryzbik, L.M., 2000. Table of Integrals Series and Products 6th ed.
Academic Press, New York.
Huang, G., Liao, H., Li, M., 2013. New formulation of Cholesky decomposition and
applications in stochastic simulation. Probab. Eng. Mech. 34, 40–47. http://
dx.doi.org/10.1016/j.probengmech.2013.04.003.
Karman, T.V., 1937. The fundamentals of the statistical theory of turbulence. J. Aero. Sci.
4, 131–138. http://dx.doi.org/10.2514/8.350.
Li, Y.L., Qiang, S.Z., Liao, H.L., Xu, Y.L., 2005. Dynamics of wind-rail vehicle-bridge
systems. J. Wind Eng. Ind. Aerodyn. 93, 483–507. http://dx.doi.org/10.1016/
j.jweia.2005.04.001.
Shinozuka, M., 1971. Simulation of multivariate and multidimensional random
processes. J. Acoust. Soc. Am. 49, 357–368. http://dx.doi.org/10.1121/1.1912338.
Simiu, E., Scanlan, R.H., 1996. Wind Effects on Structures: Fundamentals and
Application to Design. John Wiley & Sons, New York.
Taylor, G.I., 1938. The spectrum of turbulence. Proc. R. Soc. A. 164, 476–490. http://
dx.doi.org/10.1098/rspa.1938.0032.
Wu, M., Li, Y., Chen, X., Hu, P., 2014. Wind spectrum and correlation characteristics
relative to vehicles moving through cross wind. J. Wind Eng. Ind. Aerodyn. 133,
92–100. http://dx.doi.org/10.1016/j.jweia.2014.08.004.
Xu, Y.L., Guo, W.H., 2003. Dynamic analysis of coupled road vehicle and cable- stayed
bridge systems under turbulent wind. Eng. Struct. 25, 473–486. http://dx.doi.org/
10.1016/S0141-0296(02)00188-8.
Yu, M., Liu, J., Liu, D., Chen, H., Zhang, J., 2016. Investigation of aerodynamic effects on
the high-speed train exposed to longitudinal and lateral wind velocities. J. Fluids
Struct. 61, 347–361. http://dx.doi.org/10.1016/j.jfluidstructs.2015.12.005.
Zhai, W., 2016. Vehicle-Track Coupled Dynamics, 4th ed. Science Press, Beijing.