Académique Documents
Professionnel Documents
Culture Documents
a r t i c l e in fo
abstract
Article history:
Received 4 March 2009
Accepted 18 November 2009
Available online 23 December 2009
This paper considers aspects of the aerodynamic behaviour of high speed trains. It does not specically
address the many aerodynamic problems associated with such vehicles, but rather attempts to describe,
in fundamental terms, the nature of the ow eld. The rationale for such an approach is that the ow
elds that exist are the primary cause of the aerodynamic forces on the train and its components which
result in a whole range of aerodynamic issues. This paper thus draws on a wide range of model scale
and full scale experimental and computational work and attempts to build up a comprehensive picture
of the ow eld. Attention is restricted to trains in the open air (i.e. tunnel ows will not be considered)
for both still air conditions and crosswind conditions. For still air conditions the ow eld will be
described for a number of ow regions i.e.
Keywords:
High speed trains
Aerodynamics
Cross winds
Boundary layers
Wakes
Slipstreams
1. Introduction
This paper aims to set out a description of the ow eld around
high speed trains in the open air. It will approach this from a fairly
fundamental point of view, and will not specically address
practical issues and problems associated with the aerodynamic
behaviour of trains, although these will be briey discussed at the
end of the paper. Such an approach is adopted in the hope that
such a description will clarify the basic ow mechanisms that
exist around high speed trains, and will thus inform future
consideration of a range of more practical issues.
The basic tools in the study of train aerodynamics are full scale
testing, wind tunnel testing and CFD calculations, as indeed is the
case in other elds of aerodynamics. In the case of the study of
train aerodynamics, all of these approaches are fraught with
difculties. Full scale measurements have to be made in very
turbulent ows and very often a large number of runs have to be
carried out to enable the mean and unsteady ow patterns to be
elucidated. Baker et al. (2001) describes the technique of
ensemble averaging through which the results of a large
ARTICLE IN PRESS
278
Fig. 1. Velocities in the nose region of the ICE service train (z0 =0.5 m).
Fig. 2. Pressure time history measured during the passage of two ETR 500 trains
(x axis is an arbitrary time) (Mancini and Malfatti 2001).
Fig. 3. Results of the potential ow calculations of Sanz-Andres and SantiagoProwald (2002) (Pressure coefcient traces are shown for 2D and 3D computations. The x axis parameter T is the time from the passing of the nose of the train
normalised by train speed and distance from the centre of the train).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
ow regions the work of the author and his co-workers, and the
work of other investigators are considered to develop as complete
a picture as possible of the ow eld around the train.
Section 3 then goes on to consider the ow eld around trains
in a cross wind. This begins by a consideration of the nature of the
wind ow relative to the train in terms of the mean velocity
prole, turbulence prole and power spectrum. A qualitative
picture of the ow around trains is then developed from a
consideration of the work of a number of authors, and the nature
of the pressure distribution around high speed trains is also
discussed, in terms of both steady and unsteady surface pressures.
Finally the way in which these pressures sum to give cross wind
forces and moments is discussed in terms of the aerodynamic
admittances and aerodynamic weighting functions. Some concluding remarks are then made in Section 4 and the implications
of the results for current issues in high speed train aerodynamics
are set out.
279
a 14 car ICE service train. This data was obtained from trackside
anemometry in full scale experiments designed to measure the
slipstreams around such trains. The experiments are reported in
outline (RAPIDE Consortium, 2001) and discussed in considerable
length by the author and his co-workers in Sterling et al. (2008b).
Data from these experiments will be used extensively in what
follows to illustrate a number of effects. The data in Fig. 1 is an
ensemble average of the data from 17 train passes. This data was
aligned (at the point corresponding to the peak of the velocity trace
shown in the gure) and the data at all other points averaged over
all the runs. Thus x, the position along the train, is dened as
measured from this peak in velocity. The lateral distance y0 is
dened as the distance from the rail edge, and the vertical distance
z0 as the distance from the top of the rail. For the results shown the
velocities were measured at trackside with no platform present
(z0 =0.5 m). The air velocity data, u, is divided by the train speed, v,
to give the normalised value U. From Fig. 1 the velocity peak can be
seen to be sharply dened and, as would be expected, decreases
away from the train. The standard deviation of the ensemble is
small in all casesof the order of 0.020.03, which indicates that in
this ow region there is little run to run variation.
The velocity changes illustrated in Fig. 1 are accompanied by
pressure changes. Fig. 2 shows typical pressure changes caused by
an ETR 500 (Mancini and Malfatti, 2001). These measurements
were obtained from train passing tests carried out as part of the
major EU TRANSAERO project. It can be seen that there is a rapid
increase and then decrease in pressure around the train nose.
0.4
0.3
y' = 1.16m
0.2
y' = 1.5m
y'= 2.42m
0.1
0
-50
Fig. 5. ICE service car velocity measurements (z0 = 0.5 m; measurements made at trackside with no platform) (Sterling et al., 2008b).
0.3
0.25
x=50m
0.2
U
x=100m
0.15
x=200m
x=300m
0.1
x=350m
0.05
0
0
y (m)
Fig. 6. ICE service car velocity measurements (z0 =0.5 m; measurements made at trackside) (Sterling et al., 2008b).
ARTICLE IN PRESS
280
Again for any particular train, this effect is highly repeatable from
run to run.
As such ows are inviscid they can be well predicted by
reasonably simple calculation methodsas shown Fig. 3 below
from the potential ow calculations of Sanz-Andres and SantiagoProwald (2002). More complex panel methods can be used to
calculate the details of the pressure and velocity variations
around train nose shapes of different types (such as the results
for the Euler method shown in Fig. 2). The blunter the nose shape,
the higher are the velocity and pressure disturbances.
2.2. The boundary layer region
2.2.1. Train side
Over the last few decades a number of investigators have made
boundary layer measurements on trains, using conventional train
Fig. 7. Boundary layer parameters for ICE service train (Sterling et al., 2008b).
Turbulence intensity
0.3
0.2
Trackside y'=1.16m,
z'=0.5m
Platform y=1m,
z=1m
0.1
0
0
100
200
300
400
x (m)
Fig. 8. Turbulence intensity for boundary layers on the side of the ICE service train (Sterling et al., 2008b).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
0.8
Trackside y'=1.16m,
z'=0.5m
Platform y = 1m, z=1m
Autocorrelation
0.6
0.4
0.2
0
0
0.5
1.5
2.5
-0.2
Lag time (sec)
Fig. 9. Autocorrelations for boundary layers on the side of the ICE service train (Sterling et al., 2008b).
Fig. 10. Local skin friction coefcients for HST model and full scale tests of Brockie and Baker (1990).
Fig. 11. Boundary layer displacement thickness on 1/76th model scale HST roof from Brockie and Baker (1990).
281
ARTICLE IN PRESS
282
based pitot probes, hot wire probes etc. These tests have been
carried out at both full scale and model scale for a variety of train
types. From these experiments it is possible to derive standard
boundary layer parameters such as the displacement thickness
and the form parameter (although note that these are formally
derived for classical two dimensional boundary layers, rather than
the complex three dimensional ows found around trains). The
data from some of these experiments is summarised in Fig. 4
below, with data from the wind tunnel and full scale tests of
Brockie and Baker (1990) for the UK HST, and the data correlation
of model scale results given in Schetz (2001), reporting the earlier
Fig. 12. Boundary layer on train roofs from moving model tests for 1/25 scale ICE
model (Baker et al., 2001): (a) displacement thickness; (b) form parameter.
Fig. 13. Velocity traces beneath Korean high speed train (Kwon and Park, 2006).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
283
Fig. 14. Vertical and horizontal velocity proles beneath Korean high speed train (Kwon and Park, 2006).
ARTICLE IN PRESS
284
Fig. 16. Velocity proles measured below Shinkansen train (Ido et al., 2008). The
different curves indicate different sections along the train, with M1 being at the
front of the vehicle and M5 at the rear.
Fig. 15. Pressure coefcients beneath Eurostar train (Quinn and Hayward, 2008).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
285
Fig. 17. Helical vortices in the wake of trains (Baker et al., 2001; Sockel, 1996; FLUENT).
ARTICLE IN PRESS
286
Fig. 19. ICE 1/25th model scale wake measurements (y is the distance from the
side of the train, h is the train height) (Baker et al., 2001).
Fig. 18. Computations of wake oscillations for the ICE (Schulte-Werning et al., 2003).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
into the vehicle wake. Schulte-Werning et al. (2003) takes the CFD
work somewhat further and investigates the unsteadiness of
these helical structures through the use of unsteady RANS
techniques. The results are illustrated in Fig. 18. A well dened
oscillation can be seen with a Strouhal number of 0.14.
Now the full scale slipstream data presented in Sterling et al.
(2008b) has been analysed intensively to attempt to determine
the wake characteristics. Fig. 5 shows the ensemble average
velocity traces, and for those nearest to the train there can be
seen to be a noticeable peak in the velocity in the near wake (i.e.
350400 m). These velocities were measured 1 m above the track
with no platform in place. The equivalent model scale experiments of Baker et al. (2001), measured half way up body height,
do not show this peak. The experiment and simulation of Fig. 17
indicate that the helical vortex occurs close to the ground, and it
thus seems likely that this is what is observed in the full scale
slipstream measurements. Further investigation of the full scale
data however showed that the technique of ensemble averaging
was, in this case, actually hiding a physical effect, and that this
peak did not appear on around half of the individual velocity
traces that were used. Careful re-analysis revealed that the
slipstream measurements were picking out some type of vortex
shedding oscillation with a Strouhal number of 0.11, which is
close to the computational value of 0.14 reported above. It thus
seems that the helical vortices in the train wake undergo
y/h = 0.033
0.1
0.05
0
0
50
100
287
150
T
3. The ow around trains with a cross wind
y/h = 0.1333
0.1
0.05
0
0
50
100
150
T
y/h = 0.40
0.1
0.05
0
0
50
100
150
T
Fig. 20. Best t curves to wake velocities using model of Baker (2001a)
((a) y/h= 0.033, (b) y/h= 0.2, (c) y/h=0.533).
ARTICLE IN PRESS
288
Height (m)
5
v = 0 m/s
v = 20 m/s
v = 40 m/s
v = 60 m/s
v = 80 m/s
1
0
0
0.5
1
1.5
Normalised velocity relative to train
Height (m)
5
v = 0 m/s
v = 20 m/s
v = 40 m/s
v = 60 m/s
v = 80 m/s
0
0
20
40
60
80
100
Height (m)
5
v = 0 m/s
v = 20 m/s
v = 40 m/s
v = 60 m/s
v = 80 m/s
1
0
0
0.1
0.2
0.3
0.4
Turbulence intensity relative to train
Fig. 21. The wind characteristics seen by a moving train over a at ground
(u= 20 m/s).
uniform, and the yaw angle is low and varies somewhat over the
height of the train. The turbulence intensity is lowof the order
of a few percent. Clearly these results have implications for the
type of wind tunnel or computational simulation that is usedif
there is a desire to model trains moving at high speeds, then these
tests can reasonably be carried out in low turbulence wind
tunnels, with no velocity shear (although the yaw angle twist
would not be simulated). If low vehicle speed conditions are
required, then atmospheric turbulence and shear needs to be
simulated, although it needs to be recognised that any simulation
will only model one specic set of wind speed/vehicle speed
conditions. The same comments can be made for the critical case
of a train on an embankment, where the wind prole speeds up
close to the ground, although in this case the wind shear and
turbulence intensities are rather less in all cases.
As well as the mean velocity proles and the turbulence
intensity, the turbulence spectrum experienced by a moving train
will be different to that experienced by a stationary train. This
was investigated in Cooper (1985) and typical results are shown
in Fig. 22. In this gure the spectra are shown for a range of train
speed/wind speed ratios from 0 to innity, for a pure cross wind.
The x axis is a frequency normalised with the atmospheric
turbulence length scale and the wind velocity relative to the train,
and the y axis is the spectral density normalised with the
frequency and the wind velocity variance. Plotted in this way the
spectra show a remarkable level of similarity, which thus implies
that, to a rst approximation, they scale on the wind velocity
relative to the train.
Whilst the above gures give some indication of the wind
statistics relative to the train, it is not unusual when looking at
train cross wind stability to specify, in varying degrees of detail,
an extreme wind gust, on the basis that such wind gusts will
cause train stability problems. The question thus arises as to the
nature of these gusts, which vary both in space and in time. The
author and his co-workers have in recent years investigated a
number of velocity and surface pressure datasets obtained on the
University of Birmingham Wind Engineering eld site at Silsoe in
Bedfordshire, and, of particular relevance to the current discussion, have adopted the technique of ensemble averaging of
extreme gusts i.e. identifying the gusts in the time series and
Fig. 22. Wind spectra relative to a moving train over a at ground (Cooper, 1985).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
289
Fig. 23. Extreme gust proles (Sterling et al., 2006) (ds indicates dataset number;
velocities normalised by the corresponding standard deviation): (a) streamwise
velocity component; (b) lateral velocity component; (c) vertical velocity
component.
ARTICLE IN PRESS
290
Fig. 25. Wind tunnel tests and CFD computations of wake vortex ows behind trains in a cross wind. (aidealised train (Copley, 1987); b2 car ICE (Hemida, 2006);
cICE (Diedrichs, 2005)).
Fig. 26. Computed and measured surface streamline patterns (Chiu, 1991;
Hemida, 2006).
wind tunnel tests and the CFD resultssee the surface streamline
patterns from Chiu (1991), Hemida (2006) in Fig. 26. There is
some evidence, particularly from the LES results of wake
unsteadiness (Hemida, 2006) that indicates two modes of wake
unsteadinessa horizontal wake oscillation with a Strouhal
number of 0.1, and a weak vortex shedding motion with a
Strouhal number of 0.150.2. It has to be said however that these
frequencies are not always well dened in the LES simulations,
and do tend to vary somewhat with the type of calculation used.
Measurements of a different kind are reported in Baker et al.
(2007) which presents data from moving model experiments on a
four car ICE train, with rather a crude cross wind generator placed
normal to the track. The x axis unit is time normalised by train
velocity and carriage lengthwith zero being the point at which
the train nose passes the measurement point, and 4 being the
point at which the train tail passes that point. The y axis is the
slipstream velocity normalised by the train velocity. The slipstream velocities were measured in the wake of the vehicle and
ensemble averaged in a manner that has been previously
described. These results are shown in Fig. 27 in two
formatsone for the velocities themselves (Fig. 27a) and one
for the velocities minus the velocities measured with no cross
wind (Fig. 27b). In the rst of these the crosswind magnitude can
be seen before the train nose has passed, and the inviscid nose
peak is clearly visible. This is followed by a dip in the velocities,
due to the sheltering effect of the train. A maximum in the
velocity can then be seen before a gradual decay. In the wake the
crosswind velocities are again seen. In the alternative method of
presentation any value of the relative velocity that exceeds the
upstream wind speed indicates an enhancement of that wind
speed by the train wake. This can be seen to occur at a position
that corresponds to the passing of the second car of a four car
vehicle, with lower values elsewhere. Baker et al. (2007) argues
that this is consistent with the presence of inclined vortices in the
wake of the train. Finally Fig. 28 shows the maximum wake
velocities (one second averages) that were measured for a wide
variety of high speed trains in full scale test in the UK, normalised
by the train speed. There is, inevitably, a great deal of scatter but it
can be seen that there is a general increase in the maximum
normalised velocities as the cross wind speed increases.
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
291
0.3
0.75m
0.2
2.0m
0.1
0
-2
0.3
Normalised velocity
0.2
0.125m
0.1
0.75m
2.0m
0
-2
-0.1
Normalised time
Fig. 27. Wake velocity measurements from moving model experiments for ICE model (Baker et al., 2007).
Normalised velocity
0.400
0.300
0.200
0.100
0.000
0
4
Wind speed (m/s)
Fig. 28. Maximum slipstream velocities for high speed passenger trains (Baker et al., 2007).
model, and sample results are shown in Fig. 29a Chiu (1991)
together with the results of simple panel method calculations
Fig. 29b shows equivalent measurements for a 2 car ICE 2 model
(Wu, 2004), which also shows the LES calculations of Diedrichs
(2005). Pressure coefcients are shown on loops around the
vehicle, for a variety of different distance (x) from the train nose,
normalised with the model diameter D or length L. The results are
plotted with a negative pressure coefcient in the positive
direction. For all values of x that are not near the nose, it can be
seen there is a suction peak on the windward roof corner (451 in
Fig. 29a, 315 degrees in Fig. 29b), small suctions over the rest of
the roof, leeside and underside, and a positive pressure coefcient
on the windward wall. Near the nose however, for small values of
x there is a suction peak on the leeward side, that can be expected
ARTICLE IN PRESS
292
Fig. 29. Pressure distributions around trains (aIdealised train, lines are panel method results and points are experimental results (Chiu, 1991), front of windward face is
at 01; b2 car ICElines are LES results of Hemida (2006), points are experimental results of Wu (2004), front of windward face is at 2701).
suction peak on the windward corner can again be seen for all of
the loops at both yaw angles. Fig. 30 also shows the standard
deviation of the pressure coefcients, which gives an indication of
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
293
ARTICLE IN PRESS
294
Fig. 30. Mean and standard deviations of pressure coefcient distributions around a model class 365 e.m.u: (a) 901 yaw; (b) 451 yaw.
Fig. 31. POD analysis of uctuating pressures around a model class 365 e.m.u. (Baker et al., 2008a).
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
295
6
5
4
Side force
Lift force
2
1
0
0.2
0.4
0.6
Time (sec)
0.8
6
Fig. 32. Aerodynamic admittances from Cooper (1985) (L = vehicle length,
H= vehicle height, xLu =turbulent length scale, VT = train speed, x axis is frequency
normalised with turbulence length scale and wind velocity relative to the train).
5
4
Side force
Lift force
2
1
0
0.2
0.4
0.6
Time (sec)
0.8
Fig. 34. Aerodynamic weighting functions for the class 365 e.m.u (left v= 55.9 m/s,
u= 15 m/s, right v= 17.9 m/s, u= 15 m/s).
they are only non-zero for values less than around 0.5 s i.e. the
train side and lift force coefcients are fully determined by the
values of the relative wind velocity over the previous 0.5 s. These
effects are illustrated in the results of Fig. 35, which show wind
time series generated by the method of Ding et al. (2008) and the
side forces and the lift forces for the class 365 corresponding to
the weighting functions shown in Fig. 34. The ltering effect of the
higher frequency uctuations in velocity is very clear, particularly
at the higher vehicle speeds.
Finally if calculations are carried out in the amplitude
domain, then the extreme values of the force coefcients,
based on extreme values of the forces and extreme values of the
upstream velocities, might also be expected to be less than the
mean values of these coefcients, as the high frequency uctuations will have been ltered out. This effect is illustrated in Fig. 36,
for the static wind tunnel model tests carried out on a UK Class
390 Pendolino train (Baker et al., 2004) for the extreme
coefcients based on a gust averaging time of 1 s. At rst sight
the fact that the extreme/mean ratio is signicantly below unity
seems inconsistent with the admittance and weighting function
results above which suggest that such effects due to lack of
correlation of surface loads, should be conned to averaging times
less than 0.5 to 1.0 s. However more recent calculations carried
out by the author suggest that the results of Fig. 36 are affected, to
a signicant degree, by the specication of wind velocities at the
reference streamline rather than on the signicant streamline
dened above. Nonetheless, as with the admittances, if the same
reference height is used in any stability calculations as was used
in the experiments that measured extreme values, then the
results will still be valid.
ARTICLE IN PRESS
296
Fig. 35. Wind velocity and force time histories for the class 365 e.m.u (left v= 55.9 m/s, u= 15 m/s, right v= 17.9 m/s, u= 15 m/s).
Fig. 36. Ratio of extreme to mean side force coefcients for the Class 390
Pendolino full scale experiments (Baker et al., 2004)legend indicates different
train/track congurations.
overall drag on the train, and the nature of any model scale
slipstream measurements in the boundary layer region. More
work is required in this area to determine the appropriate
way of representing the train boundary layer at model scale in
both wind tunnel and moving model experiments.
(b) The integral time scale at the train side is less than 0.1 s, and
thus ow unsteadiness in this region is unlikely to have any
effect on the stability of passengers or trackside workers, who
have a minimum response time of around 0.3 s (Jordan et al.,
2008). Instability in this region will rather be caused by mean
ow effects. (Note however that this conclusion is only true
for smooth high speed passenger trainSterling et al. (2008b)
shows that for freight trains the integral time scales on the
train side are rather larger and turbulent ows along such
trains could affect passenger and trackside worker stability.)
(c) The integral time scales in the underbody gap are very short at
0.02 to 0.03 s, with associated integral length scales of the order
of 2 m. Unpublished calculations of ballast ight paths beneath
the train by the author suggest time scales of the order of 0.3 s
and path lengths of the order of 3 m. This mismatch of times and
scales suggest that the ight path of train ballast will be
primarily determined by the mean ow eld beneath the train.
(d) The large scale unsteady ow structures in the near wake of the
train have time and length scales that are potentially hazardous
to waiting passengers, and in addition are a potentially large
ARTICLE IN PRESS
C. Baker / J. Wind Eng. Ind. Aerodyn. 98 (2010) 277298
Acknowledgements
Much of the material presented in this paper has been taken
from the results of the authors collaborators over the last two
decades (research students and fellows, former and present
colleagues), who are too numerous to name individually. Nonetheless their implicit contribution to this work is gratefully
acknowledged.
References
Baker, C.J., 1991a. Ground vehicles in high cross windsPart I Steady aerodynamic
forces. Journal of Fluids and Structures 5, 6990.
Baker, C.J., 1991b. Ground vehicles in high cross windsPart 2 Unsteady
aerodynamic forces. Journal of Fluids and Structures 5, 91111.
Baker, C.J., 1991c. Ground vehicles in high cross windsPart 3 The interaction of
aerodynamic forces and the vehicle system. Journal of Fluids and Structures 5,
221241.
Baker, C.J., 2000. Aspects of the use of the technique of orthogonal decomposition
of surface pressure elds. Wind and Structures 3, 2.
Baker, C.J., 2001a. Flow and dispersion in vehicle wakes. Journal of Fluids and
Structures 15 (7), 10311060.
Baker, C.J., 2001b. Unsteady wind loading on a wall. Wind and Structures 4 (5),
413440.
Baker, C.J., Dalley, S.J., Johnson, T., Quinn, A., Wright, N.G., 2001. The slipstream and
wake of a high speed train. Proceedings of the Institution of Mechanical
Engineers F Journal of Rail and Rapid Transit 215, 8399.
Baker, C.J., Lopez-Calleja, F., Jones, J., Munday, J., 2004. Measurements of the cross
wind forces on trains. Journal of Wind Engineering and Industrial Aerodynamics 92, 547563.
Baker, C.J., Sterling, M., Johnson, T., Figura-Hardy, G., Pope, C., 2007. The effect of
crosswinds on train slipstreams. In: International Conference on Wind
Engineering, Cairns, Australia.
Baker, C.J., Sterling, M., Bouferrouk, A., ONeil, H., Wood, S., Crosbie, E., 2008a.
Aerodynamic forces on multiple unit trains in cross winds. In: Proceedings of
the BBAA VI, Milano, Italy, July 2024.
Baker, C.J., Bouferrouk, A., Perez, J., Iwnicki, S.D., 2008b. The integration of cross
wind forces into train dynamic calculations. World Congress on Rail Research,
Seoul, S Korea.
Brockie, N.J.W., Baker, C.J., 1990. The aerodynamic drag of high speed trains.
Journal of Wind Engineering and Industrial Aerodynamics 34, 273290.
297
ARTICLE IN PRESS
298
Sterling, M., Baker, C.J., Jordan, S.C., Johnson, T., 2008b. A study of the slipstreams
of high speed passenger trains and freight trains. Proceedings of the Institute
of Mechanical Engineers Part F: Journal of Rail and Rapid Transport 222 (2),
177193.
Vino, G., Watkins, S., Mousley, P., Watmuff, J., Prasad, S., 2005. Flow structures in the
near wake of the Ahmed model. Journal of Fluids and Structures 20, 673695.