Tunnelling and Underground Space Technology 42 (2014) 161174

Contents lists available at ScienceDirect

Tunnelling and Underground Space Technology

journal homepage: www.elsevier.com/locate/tust

Numerical simulation of two trains intersecting in a tunnel

Chia-Ren Chu a,, Ssu-Ying Chien a, Chung-Yue Wang a, Tso-Ren Wu b
a
Department of Civil Engineering, National Central University, Taiwan, ROC
b
Institute of Hydrological and Oceanic Sciences, National Central University, Taiwan, ROC

a r t i c l e i n f o a b s t r a c t

Article history: This study uses a three-dimensional, compressible, turbulence model to investigate the pressure waves
Received 9 December 2013 generated by two trains passing each other in a tunnel. The turbulent ow around the train bodies is
Received in revised form 14 February 2014 computed by the RNG ke turbulence model; a sliding mesh method is utilized to treat the moving
Accepted 25 February 2014
boundary problem. The numerical results are veried through the results of laboratory experiment and
eld observation. Then, a series of numerical simulations are carried out to examine the inuences of
the tunnel length, the blockage ratio, the train speed and the intersecting location on the interactions
Keywords:
of aerodynamic waves generated by the trains. The simulation results reveal that the pressure and drag
High speed train
Compressible ow
coefcients of the trains reach a maximum when the two trains intersect at the mid-point of the tunnel
Train/tunnel interaction and the values of pressure and drag coefcients increase as the train speed and the blockage ratio
RNG ke model increase due to the train/tunnel interaction. However, the side force coefcient is dominated by the
Computational Fluid Dynamics train/train interaction and its maximum value occurs when the two trains are aligned side by side.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
The aerodynamic pressure generated by high-speed trains
travelling through a long tunnel is an important factor with regard
to the design of tunnel facilities and to the comfort of passengers
(Gawthorpe, 2000). Notably, the interaction of pressure waves
generated by two high-speed trains intersecting in tunnels is able
to cause fatigue load and signicant damage to the train bodies
(Raghunathan et al., 2002). The pressure variation and ow phe-
nomenon are much more complicated than those of a single train
in a tunnel and the rapidly uctuating aerodynamic loading on
the train bodies may result in instability and lead to an undesirable
snake-like motion (Hwang et al., 2001). With the advancement of
high speed trains in many countries, such an aerodynamic problem
becomes an important issue for the designers of the railway tech-
nology (Howe, 1998; Uystepruyst et al., 2011).
There are several studies on the aerodynamics of two trains
moving in opposite directions and which pass each other inside a
tunnel. Fujii and Ogawa (1995) solved the three-dimensional, com-
pressible Euler equations by using the domain decomposition
method and the Fortied Solution Algorithm to handle the moving
Corresponding author. Address: Department of Civil Engineering, National
Central University, 300 Jhong-Da Road, Jhong-Li, Taoyuan 32001, Taiwan, ROC. Tel.:
+886 3 4227151x34138; fax: +886 3 4252960.
E-mail addresses: crchu@cc.ncu.edu.tw (C.-R. Chu), armani.chien@gmail.com
(S.-Y. Chien), cywang@cc.ncu.edu.tw (C.-Y. Wang), tsoren@cc.ncu.edu.tw (T.-R. Wu).
boundary problem. Their results showed that when two trains
meet, the high pressure region around the train noses rst pushes
the trains laterally away from each other, but the lateral force
changes direction and pushes the trains toward each other when
they are aligned side by side. In addition, they found that the lift
force was small compared to the drag and side forces; therefore,
it could be neglected in this problem.
Hwang et al. (2001) applied a three-dimensional inviscid
numerical model and the domain decomposition method to inves-
tigate the inuences of several parameters, such as the shape of the
train nose, the location of the trains (inside or outside the tunnel),
the length of the trains and the gap between two trains on the
intersecting problem. Their results showed that the side force is
proportional to the square of the train speed and the variation of
the side force is primarily dependent on the nose shape. Also, the
maximum side force increased as the gap between two trains de-
creased and the drag force was mainly affected by the intersecting
location (inside/outside a tunnel) but insensitive to the train nose
shape.
Zhao and Sun (2010) used a two-dimensional incompressible
model and a rotational symmetric boundary system to simulate
the train-passing problems in open space. Their numerical results
show that the peak positive pressure generated by the train-pass-
ing event is larger than that of by a single train, while the peak neg-
ative pressure is nearly the same as that in the single train case.
However, the variation of their simulated pressures is too small
to be true.

http://dx.doi.org/10.1016/j.tust.2014.02.013
0886-7798/ 2014 Elsevier Ltd. All rights reserved.
162 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

Nomenclature

Atunnel cross-sectional area of tunnel s lateral distance between the two trains
Atrain frontal area of train Up mean velocity at point P
Br = Atrain/Atunnel blockage ratio of the train to tunnel Vt train speed
c speed of sound Xsc = x/Ltunnel intersecting location of two trains
CS side coefcient YM dissipation rate due to of the uctuating dilatation
CD drag coefcient yp the distance from point P to the wall
Cp = DP/0.5qV2train pressure coefcient y dimensionless distance
E total energy yT dimensionless thickness of thermal sublayer
k turbulent kinetic energy q air density
keff effective thermal conductivity e dissipation rate
kp turbulent kinetic energy at point P d Kronecker delta
Lr = Ltunnel/Ltrain ratio of the length of tunnel to train l dynamic viscosity of the air
m = M2Lr a parameter used to evaluate the compressible effect lt turbulent dynamic viscosity
M = Vtrain/c Mach number leff effective dynamic viscosity
Mt turbulent Mach number j von Karman constant
Prt turbulent Prandtl number

In view of the above studies, there is a need to use a true three- l is the dynamic (molecular) viscosity of the air and the subscripts i,
dimensional turbulence model to investigate the aerodynamic ow j = 1, 2, 3 represent the x, y, z directions, respectively. The Reynolds
eld induced by two trains passing each other inside a tunnel. stress tensors qu0i u0j are related to the mean velocity gradients via
Most of the previous numerical studies used symmetry effect in the Boussinseq approximation:
their calculation and assume a train intersecting with a virtual    
@ui @uj 2 @u
train under the same speed at the mid-point of the tunnel. How- qu0i u0j lt  qk lt k dij 3
ever, the intersecting location x may not be at the mid-point of @xj @xi 3 @xk
the tunnel (x/Ltunnel = 0.5) and the intersecting location may affect where lt is the turbulent viscosity and k is the turbulent kinetic
the pressure waves in the tunnel; the aerodynamic forces acting on energy. The turbulent viscosity can be computed from the kinetic
the trains may vary when two trains are traveling at different energy k and the dissipation rate e:
speeds.
2
Therefore, the objective of this study is to utilize a three-dimen- k
sional, compressible, turbulence model to investigate the variation lt qC l 4
e
of pressure waves and aerodynamic loading on the trains when
two trains are passing each other in a tunnel. The simulation re- where the model coefcient is Cl = 0.0845. The transport equations
sults were analyzed to understand the inuences of the length ra- of the turbulent kinetic energy and energy dissipation rate are:

tio, the train speed, the blockage ratio and the intersecting location @ @ @ leff @k 
on the interaction of pressure waves in the tunnel. qk qkui Gk  qe  Y M 5
@t @xi @xj ak @xj

2. Numerical model @ @ @ leff @ e  e e2
qe qeui C 1e Gk  C 2e q 6
@t @xi @xj ae @xj k k
2.1. Governing equations
where leff is the effective dynamic viscosity equal to the sum of the
The aerodynamics of the train/tunnel systems was solved by a molecular and turbulent viscosities. The model coefcients:
three-dimensional, compressible, RNG ke turbulence model. The C1e = 1.42 and C 2e is given by:
RNG ke model was developed by Yakhot and Orszag (1986) using
the Renormalization Group (RNG) method. It is similar to the stan- C l g3 1  g=go
C 2e C 2e 7
dard ke model but includes an additional term in the dissipation 1 bg3
rate e equation for interaction between the mean shear and turbu-
where C2e = 1.68, go = 4.38, b = 0.012, g = Sk/e, S is the skewness fac-
lence dissipation. It also improved the predictions of heat and mass
tor of turbulent velocity. The term Gk is the generation of turbulent
transfers near the wall. The governing equations are the continuity
kinetic energy by the mean velocity gradients, YM represents the
equation and the Reynolds-averaged NavierStokes equations:
contribution of the uctuating dilatation in compressible turbu-
@q @ lence and is calculated as follows:
qui 0 1
@t @xi
Y M 2M 2t 8
@qui @ @p where Mt (=k1/2/c) is the turbulent Mach number, c is the speed of
qui uj  qgdi3
@t @xj @xi sound. The transport equation of the total energy, E, is:
  
@ @u @u 2 @u @  
l i j  dij l @
qE
@
ui qE p
@
keff
@T
ui sij eff 9
@xj @xj @xi 3 @xl @xj
@t @xi @xj @xj
 q u0i u0j 2
where T is the temperature, keff is the effective thermal conductivity
where u and P are the mean velocity and pressure, q is the density (sum of the molecular and turbulent conductivities). The two terms
of the air, g is the gravitational acceleration, d is the Kronecker delta, on the right-hand side of Eq. (9) represent the energy transfer due
 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174 163

Region 2
Tunnel

Region 1 Train

Grid interface
Fig. 1. Schematic diagram of the sliding mesh method.

to thermal conduction and viscous dissipation. The term (sij)eff is the where Uc is the mean velocity at y yT and the term B was com-
deviatoric stress tensor, which is dened as: puted by (Jayatilleke, 1969):
  " #
@uj @ui 2 @uk 3=4
sij eff leff  dij leff 10 Pr
@xi @xj 3 @xk B 9:24  1 1 0:28e0:007Pr=Prt  14
Prt
In this study, the standard wall function is used to bridge the
wall-affected region and the fully-turbulent region (Launder and where Prt is the turbulent Prandtl number.
Spalding, 1974). The dimensionless velocity, U, at the near-wall The above governing equations were solved by the commercial
region is related to the dimensionless distance, y, to the wall: code Fluent 6.2.12. The computational domain was discretized by
qC 1=4 1=2 the Finite Volume Method (FVM). The convection and diffusion
l kp
y yp 11 terms were discretized using the second-order upwind scheme.
l The time derivative was discretized using the second-order impli-
cit scheme for unsteady ow calculation. The velocitypressure
u2
Up 1=4 1=2
U 12 coupling and overall solution procedure are based on the PISO
C l kp (Pressure Implicit with Splitting of Operators) algorithm. The con-
vergence criterion of the continuity and momentum equations
where kp and Up are turbulent kinetic energy and mean velocity at
were 105 and the convergence criterion of the energy equations
point P, yp is the distance from point P to the wall and u is the shear
was 106.
velocity.
The temperature in the near-wall region was inuenced by the
wall function. A linear law is described for the thermal conduction 2.2. Sliding mesh
sublayer and a logarithmic law is applied for the turbulent region.
However, since the thickness of the thermal conduction sublayer is The sliding mesh method was utilized to implement the moving
different from the thickness of the viscous sublayer, the dimen- train relative to the tunnel. The ow eld is divided into two com-
sionless thickness of the thermal sublayer, yT , is used to evaluate putational regions as shown in Fig. 1. Region 1 consists of the space
the near-wall mean temperature: around the train body, Region 2 consists of the region far away
8 1=4 1=2 from the train, and a grid interface is utilized to describe the inter-
< Pry 1 qPr C l kp U 2 y < yT 
 2 q_ p face of two regions. During the computational process, these two
T 1  1=4 1=2
: C k independent regions can slide relative to one another along the
Prt j lnAy B 12 q l q_ p fPrt U 2p Pr  Prt U 2c g y > yT
grid interface without mesh regeneration, and the grid faces do
13 not need to be re-aligned at the interface.

Cell zone 1

V VI VII VIII

A B C D E interface zone 1

a b c d e f g h i j

F G H I J
interface zone 2
V VI VII VIII

Cell zone 2

Fig. 2. Schematic diagram of sliding mesh method for computing the ux across the two non-conformal interface zones of each grid interface.
164 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

Fig. 3. Grid arrangement. (a) Side view and (b) cross-sectional view.

time step in order to compute the uxes. The number of intersec-

600 tion grids will vary as the interface zones move relative to one an-
C1
other. The uxes across the grid interface are computed using
400 E1
intersection grids rather than the grids on each interface zone.
N For example, the interface zones are composed of faces A-B, B-C,
200 C-D and D-E as well as faces F-G, G-H, H-I and I-J in Fig. 2. Faces
Pressure (Pa)

C2
d-e and e-f are used, instead of face G-H, to compute the ux across
0 the interface and bring information from Cells II and III into Cell VI.
Therefore, the physical information can be communicated between
-200 two adjacent zones.
Experiment (Ricco et al., 2007) Both Regions 1 and 2 were divided into several smaller zones
-400 Compressible model and discretized by non-structured and non-uniform grids (see
Incompressible model E2 Fig. 3). The smallest grid size around the train body was 0.3 m
-600 and the total number of the grid was 1,071,139. The maximum
-0.02 -0.01 0.00 0.01 0.02 0.03 0.04 0.05
stretching ratio of the adjacent grid was 1.08. The grid indepen-
Time (s) dence was checked by comparing the simulation results of the grid
Fig. 4. Comparison of simulated and measured pressure variation at x = 0.9 m as of 1,924,331 against 1,071,139. The difference of the positive pres-
regards the laboratory experiment of Ricco et al. (2007), Vtrain = 30.6 m s1, length sure peak of the compression wave was 4.6%. Therefore, the grid of
ratio Lr = 10, blockage ratio Br = 14.4%. 1,071,139 was adopted for the rest of the simulation.

The sliding mesh method requires a technique with which to 3. Model verication
compute the uxes across each grid point inside the non-confor-
mal interface zones. The intersection grids (grid points: b, c, d, In order to demonstrate the accuracy of the present model, the
. . ., i in Fig. 2) between the interface zones are determined at every simulation result was compared with the results of laboratory

BC: P = 0 BC: P = 0
Pressure sensor:

454 230
0.452
Ltrain= 114
200
Train z Br = 11.2 % 9.074
x

234 150 Ltunnel = 1140 150

Br = 12.56 % Br = 12.56 %
unit: m
Fig. 5. Computational domain and boundary condition of Glockle and Pfretzschner (1988). The pressure sensor is located 200 m from the tunnel entrance.
 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174
CN ET N T
1600
1200
800
Pressure (Pa)
400
0
-400
-800
-1200 Measured (Glockle & Pfretzschner, 1988)
Predicted (Present study)
-1600
0 1 2 3 4 5
Time (sec)
Fig. 6. Comparison of simulated and measured pressure variation at x = 200 m for
the eld observation of Glockle and Pfretzschner (1988), Vtrain = 67.2 m s1, length
ratio Lr = 10.
experiment of Ricco et al. (2007) and eld observation of Glockle
and Pfretzschner (1988). The experiment of Ricco et al. (2007)
was conducted by using circular cylinders (diameter 0.038 m) trav-
eling in a circular tube (diameter 0.099 m). The cylinders have dif-
ferent nose angles between the axis and directrix. The length of the
tube was Ltunnel = 6.0 m, the length of the cylinder was Ltrain = 0.6 m
and the length ratio Lr = Ltunnel/Ltrain = 10. The blockage ratio of the
tunnel was Br = Atrain/Atunnel = 14.4%, and the train speed Vt = 30.6 -
m s1. A pressure sensor was installed on the wall of the tube at
x = 0.9 m from the tunnel entrance. All the boundaries were no-slip
boundaries except for the zero pressure at the far-eld outside the
tunnel. The total number of grids was 274,012. The computation
was carried out with a constant time step Dt = 5  104 s over a
time interval of 0.07 s and the train started x = 1.0 m from the
tunnel entrance. The time of the nose entry is set as t = 0.
Fig. 4 shows the time history of pressure variation at x = 0.9 m
computed by the present compressible model and an incompress-
ible turbulence model. The nose angle between the axis and direc-
trix was 60. The symbols are the experimental results of Ricco
et al. (2007), and the solid and dash lines are the simulation results
of the compressible and incompressible models, respectively. It
shows that the results of the compressible model compared favor-
ably with the experimental results, while the incompressible mod-
el failed to predict the negative peak pressure caused by the
expansion wave at t = 0.032 s. In other words, the compressibility
effect must be taken into account for the simulation of high speed
trains in tunnel.
Fig. 7. Geometry of the Taiwan High Speed Train 700T.
165
The present model was also applied to simulate the eld obser-
vation of Glockle and Pfretzschner (1988). They measured the
aerodynamic pressure generated by a train ICE/V (Inter-City Ex-
press V) entering into the Einmalberg tunnel. The tunnel length
was Ltunnel = 1140 m, train length was Ltrain = 114 m and the length
ratio Lr = 10. The cross-sectional area of the ICE/V was A = 10.3 m2
and the train speed Vt = 242 km h1 (=67.2 m s1). The cross-sec-
tional area of the tunnel expanded at x = 454684 m from the tun-
nel entrance (see Fig. 5) and the blockage ratio became Br = 11.2%.
At other locations, the blockage ratio was Br = 12.56%. The total
number of the computational grids was 1,143,757. The computa-
tion time step Dt = 0.018 s over a time interval of 12.654 s, and
the train started 150 m from the tunnel entrance.
6 7 8 The simulated pressure variation at x = 200 m from the tunnel
entrance is shown in Fig. 6. The empty symbols are the measure-
ments of Glockle and Pfretzschner (1988), and the solid line repre-
sents the simulation result. It shows that the simulated pressures
compared satisfactorily with the eld observations. On the top of
the gure, marker CN represents the time t = 0.58 s (=200 m/
347 m s1) that the compression wave generated by the nose entry
reach the sensor location x = 200 m, and c = 347 m s1 is the speed
of sound; marker ET at t = 2.27 s (=114 m/67.2 m s1 + 200 m/
347 m s1) represents the expansion wave caused by the tail entry
to the tunnel; marker N (at t = 200 m/67.2 m s1 = 2.98 s) is the
time that the train nose passed x = 200 m; and T (at t = 200 m/
67.2 m s1 + 114 m/67.2 m s1 = 4.67 s) is the time that the train
tail passed x = 200 m. Note that the pressure peak P = 834 Pa
(Cp = 0.295) at t = 0.58 s was caused by the compression wave CN.
The pressure coefcient is dened as:
P  Po
CP 15
1
2
qV 2t
where Po is the reference pressure.
4. Results and discussion
This study used the RNG ke turbulence model to investigate
the inuences of the length ratio Lr, the train speed Vt, the blockage
ratio Br and the intersecting location Xsc (=xsc/Ltunnel) on the pres-
sure waves generated by two trains passing each other in a tunnel.
The dimensionless distance X is dened as X = x/Ltunnel. The aerody-
namic drag coefcient CD and side (lateral) force coefcient CS are
dened as:
Fx
CD 1 16
2
qV 2t Atrain
166 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

(a) side view

BC: P = 0 BC: P = 0
Pressure sensor:

Intersecting location

456
Ltrain= 304
z
152 8.625 VB
Train A Train B
x 2.5

1062 400 Ltunnel = 912 400 1062

tunnel entrance x = 0 unit: m

(b) cross-sectional view

40

12
28
3.38

8.625
z 3.65
2.5
0.8

Pressure sensor: 2.25 unit: m

Fig. 8. Schematic diagram of computational domain and boundary condition. (a) Side view and (b) cross-sectional view.

Fy t = tNT, two trains are aligned side by side; at t = tTT, the tails of both
CS 1 17
2
qV 2t Atrain trains are at the position Xsc and the intersection is nished. In this
study, the time of the nose intersection was set as tNN = 0 s, the
where Atrain is the frontal area of the train. The direction of the drag intersection of the train nose and the tail was tNT = 1.83 s and the
force is opposite to the travelling direction; the positive CS indicates time of the tail intersection was tTT = 3.65 s (=Ltrain/Vt).
that the lateral forces push two trains toward each other.

4.1. Reference case Top view

The train parameters were based on the Taiwan High Speed
Train 700T (see Fig. 7). The train was 304 m in length, 3.38 m in CS
t = tNN = 0.0 sec B
width and 3.65 m in height. The train nose (length is 8.0 m) has
the same shape as the tail. The tunnel was a double-tracked tunnel A CD
with the cross-sectional area Atunnel = 86.9 m2, the blockage ratio Pressure sensor:
was Br = 14.2% and the train runs with an offset from the centerline
of the tunnel. The gap (lateral distance) between two trains was
s = 1.12 m. The tunnel length Ltunnel = 912 m, the length ratio B
Lr = 3.0, and the train speed VA = VB = 300 km h1 (=83.3 m s1). t = tNT
This ow condition is hereinafter referred to as the reference case.
A
Fig. 8 presents the schematic diagram of the computational do-
main and the boundary condition. The numerical sensors for mea-
suring pressure are located on the train nose A and on the tunnel
wall at X = 0.167, 0.333, 0.50, 0.667 and 0.833 (x = 152 m, 304 m, B
456 m, 608 m and 760 m) from the tunnel entrance at a height of t = tTT
A
2.5 m from the ground.
Fig. 9 portrays the top view of the relative position of two trains
at three important times: the intersection of the train noses (NN), Xsc
the intersection of the train nose and the tail of other train (NT) Fig. 9. Schematic diagram of two trains pass each other, tNN, tNT and tTT are the
and the intersection of the train tails (TT). At t = tNN, the noses of times of the intersection of the train noses, the intersection of the train nose and the
both trains come to the intersection point Xsc in the tunnel; at tail of other train, and the intersection of the train tails, respectively.
 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174 167

(a) NN NT TT
6
912
TB CBN2 CAT
NB 4
760
2
608 EAN2

Pressure (kPa)
0
x (m)

456
-2 x = 152 m
x = 304 m
304 EBT CAN2 EBN2 -4
CAN CBN x = 456 m
EBN EAT EAN x = 608 m
152 -6
NA CBT x = 760 m
TA
0 -8
-6 -4 -2 0 2 4 -6 -5 -4 -3 -2 -1 0 1 2 3 4
Time (sec) Time (sec)

CBN CBN2 CAN2 CAT CBT EAN2

Fig. 11. Time history of wall pressures at different locations for the two trains
EBN EAN EAT EBT
10 intersect at Xsc = 0.5 (x = 456 m).
NN NT TT
8
of train A passed the nose of train B at t = 0.0 s, pressure PA was af-
6 fected by the low pressure region near the shoulder of train B and
decreased more rapidly than the pressure drop caused by the
Pressure (kPa)

4
2
0 that, due to the compression waves CBN2, CAN2 and CAT (generated
-2
-4
-6 -4 -2 0 2
Time (sec)
Fig. 10. (a) Train displacement and the propagation lines of the pressure waves. (b)
Pressure variation on the nose of train A when the two trains intersect.
Fig. 10(a) shows the propagation line of the pressure waves as
well as the displacements of the train nose and tail. The horizontal
dot lines represent the positions of the sensors on the tunnel wall,
the black lines (marked by NA and TA) and red lines (marked by NB
and TB) are the positions of the noses and tails of train A and B at
different times. The dash lines with steeper slope represent the
propagation of pressure waves (CAN and CBN are compression
waves; EAN and EBN are expansion waves). The propagation lines
have steeper slope is due to fact that the speed of sound
c = 347 m s1 is faster than the train speed Vt = 83.3 m s1. Each
intersection of the horizontal lines and the propagation lines
(CAN, CBN, EAN, EBN) correspond to a pressure wave passing through
the pressure sensors. Notably, these two trains enter the tunnel
simultaneously and the nose-induced compression waves CAN
and CBN are generated by the train nose A and B, respectively.
The compression waves CAN and CBN are propagated along the tun-
nel and reected back from the other end of the tunnel as expan-
sion waves EAN and EBN (Achetz, 2001). Because part of the wave
energy is released to the exterior as micro-pressure waves, the
strength of the expansion waves EAN and EBN are always weaker
than the rst compression waves CAN and CBN (Baron et al., 2006).
Fig. 10(b) shows the time variation of pressure PA on the nose of
train A during the intersection. The vertical dot lines correspond to
the times of the compression waves CAN and CBN, as well as the
expansion waves EAN and EBN. The maximum positive pressure
on the train nose was Pmax = 7010 Pa (Cpmax = 1.70), which occurred
when the train nose A met the compression wave CBN at
t = 1.95 s. The following pressure drop at t = 2.0 to 0.0 s was
caused by the expansion waves EAN, EBN EAT and EBT. After the nose
expansion waves. Then, the expansion waves EAT and EBT generated
by the train tails further reduced the pressure PA until t = 0.79 s.
The minimum nose pressure was P = 2670 Pa (Cp = 0.65). After
by the entry of the tail of train A), the nose pressure PA started to
increase. After the nose of train A passed the tail of train B at
tNT = 1.83 s, the compression waves further increased the pressure
4
PA. However, the intersection of the train tails at tTT = 3.65 s has no
inuence on the nose pressure PA because the nose of train A is far
away from the tails.
Fig. 11 presents the time variations of wall pressure at
X = 0.167, 0.333, 0.50, 0.667 and 0.833 from the tunnel entrance.
The results demonstrate that the pressure at X = 0.167 was in good
agreement with the pressure at X = 0.833; and the pressure at
X = 0.333 agreed with the pressure at X = 0.667, i.e., two trains with
the same speed passing each other in the middle of the tunnel has
a symmetric effect. In addition, the maximum positive and nega-
tive pressures at the mid-point of the tunnel X = 0.50 was larger
than the pressures at other locations in the tunnel. In other words,
the maximum wall pressure occurred at the intersecting location.
In Fig. 11, the pressure rise between t = 4.3 to 1.6 s (before
the intersection) at X = 0.50 was attributed to three processes.
The rst pressure peak Pn = 3437 Pa (CPn = 0.83) at t = 4.3 s was
purely due to the nose-induced compression waves and the
pressure slightly decreased at t = 4.0 s because of the low pres-
sure region of the train shoulder. The pressure slightly increased
between t = 3.8 to 2.6 s, owing to the viscous effect (Ogawa
and Fujii, 1997). As two trains move toward each other, the de-
crease of the space between the two trains caused the pressure
to rise to the maximum pressure Pmax = 4000 Pa (CPmax = 0.97)
around t = 1.6 s. Note that this maximum wall pressure was
smaller than the maximum nose pressure on train A but the max-
imum pressure coefcient induced by two intersecting trains was
larger than that of a single train, as compared to the eld measure-
ments of Glockle and Pfretzschner (1988).
The pressure variations along the tunnel axis at different times
are shown in Fig. 12. When the compression waves CAN and CBN
met at x = 456 m (X = 0.5), the interaction of two pressure waves
reinforced the pressure inside the tunnel. The maximum pressure
P  4000 Pa occurs in the middle of the tunnel x = 456 m at time
t = 1.51 s. After t > 1.0 s, the expansion waves EAN and EBN (the
reection waves of CAN and CBN) and the tail-induced expansion
waves EAT and EBT propagated toward x = 456 m. The resulting
168 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

CAN CBN
NA NB
TA TB

NA NB
EAN + EAT EBN + EBT

CAN CBN

TA TB
EAN+EAT+EBN+EBT
NA NB

NA NB

NA NB CAN2+EAT+EBT+ CBN2
EBN EAN

TA TB
Pmin

Pmax
NA NB
TA TB CAN2+EAT+EBT+ CBN2
CAN+CBN EAN
EBN

TA TB

Fig. 12. Pressure variations along the tunnel axis at different times for the two trains intersect at Xsc = 0.5 (x = 456 m).

minimum pressure Pmin  6000 Pa (CPmin  1.46) occurs at which was smaller than the pressure rise induced by the rst
t = 0.89 s. Then the next compression waves CAN2 and CBN2 (the compression waves CAN and CBN. This is because part of the wave
reective waves of EAN and EBN) raised the pressure to P 2000 Pa, energy was released to the exterior as a micro-pressure wave, so
 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174 169

NB TA TB NA
CBN2 CAN2

EBN2 EAN2

TA TB

CBN2 CAN2
EBN2 EAN2
CAT + CBT

NA TB TA
NB NB NA
TA TB

CAT CBT TB TA
NB NA

NB TA TB NA EAN2 EBN2

Fig. 12 (continued)

the pressure variation in the tunnel diminished with time. Fig. 12
also demonstrates that the pressure value inside the tunnel is all
negative between 0.09 and 1.85 s due to the expansion waves
EAN, EBN, EAT and EBT; the pressure inside the tunnel is all positive
between 2.81 and 3.61 s due to the new compression waves CAT,
CBT, CAN2 and CBN2.
Fig. 13 shows the top view of the pressure distribution at height
z = 2.5 m during the intersection of the train noses. Before the train
noses intersected (t = 0.55 s), the pressure is positive in the space
between the two trains. Note that the train noses have a maximum
positive pressure, while the pressures around the shoulder were
negative. As the train noses pass each other, the stagnation pres-
sure at the train nose decreased rapidly between t = 0.090.41 s,
owing to the expansion waves. Fig. 14 depicts the pressure distri-
bution during the intersection of the nose of train A and the tail of
train B. After the nose of train A passed the tail of train B, the pres-
sure at the nose of train A became positive again at t = 1.85 s.
Finally, the pressure distribution during the intersection of train
tails is shown in Fig. 15. When the train tails are moving toward
each other during t = 2.813.61 s, the pressure in the tunnel are
all positive, owing to the inuences of the compression waves
CAN2 and CBN2. At time t = 3.77 s (the tails have passed each other),
the un-occupied area near the train tails suddenly increased, the
ow deceleration in the un-occupied area caused the pressures
at the train tails to increase. When the two trains moved further
away from each other (t = 3.93 s), the pressure at the train tails de-
creased gradually.
Fig. 16 shows the time history of drag coefcient experienced by
train A during the intersection. It is highly correlated with the pres-
sure variation PA of the train nose. The maximum drag coefcient
CDmax = 2.20 occurring at t = 1.81 s results from the nose-induced
compression waves; it is close to the time (t = 1.95 s) of the
maximum nose pressure. As the nose pressure PA weakens by the
expansion waves EAN, EBN, EAT and EBT, the drag coefcient CD
170 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

(Pa) Unit: Pa
train B

train B

train A

(Pa)

train B
train B

train A train A

(Pa)

train B train B

train A train A

(Pa)

train B
train B

train A train A

(Pa)

train B
train B

train A train A

(Pa)

train B

train A train A
Fig. 13. Pressure distribution during the intersection of the train noses (NN). Fig. 14. Pressure distribution during the intersection of the train nose and the tail of
other train (NT).

gradually decreases between t = 1.81 and 0 s. Once the two noses

passed each other after tNN = 0.0 s, the drag coefcient became rel-
atively small until the nose of train A intersected the tail of train B at
t = tNT. Although the drag coefcient rose after t > tNT, its value was
still smaller than the drag coefcient CDmax before two trains rst
meet (t = 0); this is resulted from the diminished compression
waves CBN2 and CAN2.
The time evolution of the side force coefcient CS of train A was
presented in Fig. 17. When the two trains moved toward each
other, the high pressure region around the train noses push the
trains laterally away from each other and the side force coefcient
is negative (CSmin = 0.72). When the two trains were aligned side
by side (the nose of train A intersected the tail of train B), the side
force coefcient changed direction and pushed the two trains to-
ward each other. The maximum suction force (CSmax = 0.96) occurs
when tNT = 1.83 s. Once the train tails passed each other at
t = 3.65 s, the side force coefcient became negative again but the
pulling side force (CS = 0.58) was less than that of the nose
intersection.
Following are four simulation series to examine the inuences
of the length ratio, the blockage ratio, the train speed, and the
intersecting location on the maximum pressure and force coef-
cients. In these simulation series, only the key parameter was var-
ied, while all other parameters were are identical to the reference
case.
4.2. Length ratio effect
This section investigated the effect of the tunnel length on the
pressure and force coefcients. Three different tunnel lengths Ltun-
nel = 912 m, 1520 m and 2128 m, with the length ratio Lr = 3, 5 and
7, respectively. Two trains intersected at the mid-point of the
tunnel Xsc = 0.5 with the same train speed. Table 1 shows that
 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174 171

Unit: Pa NN NT TT
1.5

1.0
train B
0.5

train A 0.0

Cs
-0.5

train B -1.0

-1.5
train A -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Time (sec)

Fig. 17. Time history of side force coefcient during the train intersection.

train B Table 1
Parameters of simulation series for different tunnel length.

Ltunnel (m) Lr CPmax CPmin CDmax CSmax

train A
912 3.0 0.97 1.46 2.24 0.96
1520 5.0 1.01 1.43 2.33 0.96
2128 7.0 0.90 1.10 2.08 0.98

train B
Table 2
Parameters of simulation series for different blockage ratio.
train A
Atunnel (m2) Br (%) CPmax CPmin CDmax CSmax
102.8 12.0 0.76 1.15 1.90 0.95
86.9 14.2 0.97 1.46 2.24 0.96
68.5 18.0 1.46 2.07 2.96 1.00
train B

train A Table 3
Parameters of simulation series for different train speed.

Vt (km h1) CPmax CPmin CDmax CSmax

Case B1 200 (=56 m s1) 0.81 0.80 1.84 0.96
train B Case B2 250 (=69 m s1) 0.90 1.12 2.00 0.93
Case B3 300 (=83 m s1) 0.97 1.46 2.24 0.96

train A
Fig. 15. Pressure distribution during the intersection of the train tails (TT). the values of the maximum and minimum wall pressures CPmax,
CPmin and drag coefcient of the train CDmax for the length ratio
Lr = 7 are all somewhat smaller than the results of Lr = 3 and 5. This
NN NT TT is because the interactions of compressive waves and reective
2.5
expansion waves are less concentrated and intense in long tunnels
(Lr P 7). Moreover, the maximum side force coefcients CSmax for
2.0
all three cases are about the same. This indicates that the inuence
1.5 of the length ratio on the side force coefcient CSmax is insigni-
cant. The reason is that the side force is closely related to train/
train interaction, while the train/tunnel interaction (and the rela-
CD

1.0
tive length of the tunnel) only has minor inuence on the side force
0.5 coefcient.

0.0 4.3. Blockage ratio effect

-0.5 The inuence of the blockage ratio on the pressure and force
-6 -5 -4 -3 -2 -1 0 1 2 3 4
coefcients is examined in this section. The blockage ratio was
Time (sec)
set to 12%, 14.2% and 18.0%, by changing the cross-sectional area
Fig. 16. Time history of drag coefcient during the train intersection. of the tunnel Atunnel. The simulated results are summarized in
172 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

CAN CBN + CAN2

TB
NB
NA
NA
TA B EBT

Pmax
CAN

TA
EAT CAT +CAN2 NA
NB
TA NA TB

CBN
TB , NA
TA , NB

EAT
TA NA
EBN + EAN2

CBN EAT
NA
NB
B
NB B TB , TA

TA NA B
Pmin

Fig. 18. Pressure distribution along the tunnel axis for the two trains intersect at Xsc = 0.67 (xsc = 1015 m), tunnel length L = 1520 m.

Table 2. The increase of the blockage ratio led to the increase of the CSmax only slightly increased as the blockage ratio increased. Fur-
maximum and minimum pressure coefcients as well as the max- thermore, the maximum side force coefcient CSmax = 0.95 (block-
imum drag coefcient CDmax. However, the side force coefcient age ratio Br = 12.0%, train speed Vt = 300 km h1 and the gap
 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174
between two trains s = 1.12 m) compared favorably with the
numerical results of Fujii and Ogawa (1995): CSmax = 1.09
(Br = 12.0%, Vt = 270 km h1 and s = 0.858 m). This also demon-
strated that the side force coefcient CSmax is related to the train/
train interaction, rather than the train/tunnel interaction.
4.4. Train speed effect
In order to investigate the train speed effect, three different
train speeds VtA = VtB = 200, 250 and 300 km h1 were simulated.
The length ratio, blockage ratio and the intersecting location are
the same as the reference case. The simulated results (listed in Ta-
ble 3) revealed that the maximum and minimum pressure coef-
cients as well as the maximum drag coefcient CDmax increased
as the train speed increased. However, the inuence of the train
speed on the maximum side force coefcient CSmax was trivial. This
is in good agreement with the nding of Hwang et al. (2001) that
the side force coefcient is independent of the speed of the train.
4.5. Intersecting location
Finally, the effect of the intersecting location on the pressure
and force coefcients is studied by comparing the results of two
different intersecting locations: Xsc = 0.5 and 0.67. The tunnel
length Ltunnel = 1520 m and the length ratio Lr = 5, train speed
VtA = VtB = 300 km h1. Fig. 18 shows that the pressure distribution
along the tunnel axis for the two trains intersect at Xsc = 0.67
(xsc = 1015 m). Note that the pressure distribution is asymmetric
and the maximum pressure Pmax = 2850 Pa (CPmax = 0.69) occurred
at x = 608 m; while the minimum pressure Pmin = 5610 Pa
(CPmin = 1.35) occurred at the intersecting location xsc = 1015 m.
Table 4 shows that the magnitude of maximum and minimum
pressure coefcients of Xsc = 0.67 where both are smaller than
the corresponding values of Xsc = 0.50. This is because the maxi-
mum pressure coefcient CPmax resulted from the interaction of
the nose-induced compression waves CAN and CBN when the inter-
secting location was Xsc = 0.50. Notably, the maximum pressure
CPmax was caused by the compression waves CAT and CAN2 when
Xsc = 0.67. These reective waves CAT and CAN2 were weaker than
the rst pressure waves CAN, CBN. Similarly, for the minimum
pressure CPmin, it was under the inuence of EBN, EAN2 and EAT2,
as can be seen in Fig. 18. However, the simulated results reveal that
the inuences of the intersecting location Xsc on the maximum
drag coefcient CDmax and the side force coefcient CSmax were
insignicant.
5. Summary
A three-dimensional, compressible, turbulence model was uti-
lized to investigate the pressure waves generated by two high-
speed trains passing each other in a tunnel. The turbulent ows
around the train bodies were computed by the RNG ke turbulence
model. The sliding mesh method was utilized to implement the
moving train relative to the tunnel. The numerical model was ver-
ied by comparisons with the results of the laboratory experiment
of Ricco et al. (2007) and eld observations of Glockle and
Pfretzschner (1988). Then, the numerical model was used to exam-
ine the interaction of the pressure waves caused by two trains
Table 4
Parameters of simulation series for different intersecting location.
xsc (m) Xsc CPmax CPmin CDmax
755 0.50 1.01 1.43 2.33
1015 0.67 0.69 1.35 2.31
173
passing each other in a long tunnel. In addition, a series of numer-
ical simulations were performed to study the effects of the length
ratio Lr of the tunnel, the blockage ratio Br, the train speed Vtrain
and the intersecting location Xsc on the interactions of the train/
tunnel system.
The parametric study revealed that the maximum and the min-
imum wall pressure coefcients CPmax and CPmin occurred at the
mid-point of the tunnel when two trains with the same speed
intersect at Xsc = 0.5. Furthermore, the magnitude of CPmax, CPmin
and drag coefcient CDmax increased as the train speed and the
blockage ratio increased. However, when the two trains inter-
sected each other at Xsc = 0.67, the pressure coefcients CPmax
and CPmin became smaller, as compared to that of Xsc = 0.5. This is
because the pressure distribution in the tunnel is asymmetric
and the maximum and minimum pressures were generated by
the reective waves, rather than the rst waves, which were stron-
ger. Also, it was found that the maximum side force coefcient
CSmax occurs as two trains are aligned side by side and the side
force pushes two trains toward each other. Nevertheless, the mag-
nitude of CSmax is insensitive to the tunnel length, blockage ratio
and train speed because the side force was dominated by the
train/train interaction. Whereas the pressure coefcients CPmax,
CPmin and drag coefcient CDmax were controlled by the train/tunnel
interaction.
Notice that the maximum and minimum pressure and force
coefcients are functions of the length of tunnel and trains, the
train speed and blockage ratio. A different setting, for example,
two trains entering a long tunnel right after a third train just left
the tunnel could lead totally different extreme pressure. Hence,
the peak pressure coefcients shown in the paper only applied to
the described setting. However, the present numerical model could
be applied to simulate the aerodynamic forces and pressure waves
under other scenarios.
Acknowledgment
The nancial support from the China Engineering Consultants
Inc. (CECI) of Taiwan, R.O.C. under grant no. 00940 is gratefully
appreciated.
References
Achetz, J.A., 2001. Aerodynamics of high-speed trains. Ann. Rev. Fluid Mech. 33,
371414.
Baron, A., Molteni, P., Vigevano, L., 2006. High-speed trains: prediction of micro-
pressure wave radiation from tunnel portals. J. Sound Vibr. 296, 5972.
Fujii, K., Ogawa, T., 1995. Aerodynamics of high speed trains passing by each other.
Computers Fluids 24 (8), 897908.
Gawthorpe, R., 2000. Pressure effects in railway tunnels. Rail International, Monthly
Review of the IRCA/UIC, April 2000, 1017.
Glockle, H., Pfretzschner, P., 1988. High speed tests with ICE/V passing through
tunnels and the effect of sealed coaches on passenger comfort. In: 6th
International Symposium on the Aerodynamics and Ventilation of Vehicle
Tunnels, Durham, England, pp. 2344.
Howe, M.S., 1998. Mach number dependence of the compression wave generated by
a high-speed train entering a tunnel. J. Sound Vibr. 212, 2326.
Hwang, J., Yoon, T.S., Lee, D.H., Lee, S.G., 2001. Numerical study of unsteady ow
eld around high speed trains passing by each other. JSME Int. J. Ser. B. Fluids
Therm. Eng. 44 (3), 451464.
Jayatilleke, C., 1969. The inuence of Prandtl number and surface roughness on the
resistance of the laminar sublayer to momentum and heat transfer. Prog. Heat
Mass Transfer 1, 193321.
Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent ows.
Comput. Meth. Appl. Mech. Eng. 3, 269289.
Ogawa, T., Fujii, K., 1997. Numerical investigation of three-dimensional
compressible ows induced by a train moving into a tunnel. Comput. Fluids
26 (6), 565585.
Raghunathan, R.S., Kim, H.D., Setoguchi, T., 2002. Aerodynamics of high-speed
CSmax
railway train. Progr. Aerosp. Sci. 38, 469514.
0.96 Ricco, P., Baron, A., Molteni, P., 2007. Nature of pressure waves induced by a high-
0.97 speed train travelling through a tunnel. J. Wind Eng. Indus. Aerodyn. 95 (8),
781808.
174 C.-R. Chu et al. / Tunnelling and Underground Space Technology 42 (2014) 161174

Uystepruyst, D., William-Louis, M., Creus, E., Nicaise, S., Monnoyer, F., 2011. Zhao, X., Sun, Z., 2010. A new method for numerical simulation of two trains passing
Efcient 3D numerical prediction of the pressure wave generated by high-speed by each other at the same speed. J. Hydrodyn., Ser. B 22 (5), 697702.
trains entering tunnels. Computers Fluids 47 (1), 165177.
Yakhot, V., Orszag, S.A., 1986. Renormalization group analysis of turbulence: I. Basic
theory. J. Scientic Computing 1 (1), 151.