Computers and Geotechnics 63 (2015) 229–243

Contents lists available at ScienceDirect

Computers and Geotechnics

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

Three-dimensional numerical analysis for the longitudinal seismic

response of tunnels under an asynchronous wave input
Peng Li, Er-xiang Song ⇑
Department of Civil Engineering, Tsinghua University, Beijing 100084, China

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

Article history: The numerical analysis of the longitudinal seismic response of tunnels to the spatial variation of earth-
Received 6 April 2014 quake ground motion is an important issue that cannot be ignored in the design and safety evaluation
Received in revised form 14 September 2014 of tunnel structures. In this paper, numerical modelling techniques for the analysis of the longitudinal
Accepted 7 October 2014
response of tunnels under an asynchronous seismic wave are extensively studied. The free field wave
Available online 25 October 2014
motion is calculated using the 1D time-domain finite element method and is then extended to 3D. Based
on the theory of wave input, three-dimensional numerical simulation techniques for a site response to an
Keywords:
asynchronous earthquake are developed. Then, a 3-D soil–tunnel structure interaction model is estab-
Asynchronous input
Longitudinal seismic response
lished to simulate the longitudinal response of the tunnel when subjected to an asynchronous earth-
Tunnel structure quake. These numerical modelling techniques are discussed in detail, and the analysis results prove
Numerical analysis their validity. The proposed method is applied to analyse the longitudinal seismic response of a tunnel
Travelling wave effect under the oblique seismic incidence of P, SV and SH waves from a semi-infinite foundation. The influence
of the incident angle and incident direction on the longitudinal seismic response is discussed. The study
outcome will be beneficial to further research on refined nonlinear numerical analysis and seismic
response mechanisms of tunnel structures that experience asynchronous earthquake motion.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction structures are relatively mature [3–6]. In contrast, less systematic

Underground facilities are an integral part of the infrastructure
of modern society and are used for a wide range of applications.
During recent strong earthquakes, especially during the 1989 Loma
Prieta, 1994 Northridge and 1995 Kobe earthquakes, a number of
underground structures experienced significant damage [1], and
since then, there has been a general understanding that the seismic
design and analysis of underground structures are of great
importance.
A tunnel structure is usually very long, passes through different
soils, and may be built using multiplex construction methods, all of
which make the dynamic analysis of tunnels a sophisticated prob-
lem. Numerical methods such as the Finite Difference Method,
Finite Element Method and Boundary Element Method have been
widely used for soil-structure interaction problems due to their
versatility and reliability. With the aid of these methods, the
dynamic response can be solved reasonably well under realistic
conditions [2]. Currently, the numerical analysis and simplified
design methods for the transverse dynamic behaviours of tunnel
⇑ Corresponding author. Tel.: +86 10 62773545; fax: +86 10 62771132.
E-mail address: songex@mail.tsinghua.edu.cn (E.-x. Song).
and thorough work has been reported on the numerical analysis
of the longitudinal seismic response of tunnels under an asynchro-
nous earthquake input. It has been noted that appropriate consid-
erations should be given to travelling waves [7], especially for
extended or embedded structures with a large span or size such
as tunnels, because spatial variation of earthquake motion has dra-
matic effects [8]. Consequently, it is of great importance to carry
out numerical analyses on the longitudinal seismic response of
tunnels, which will greatly facilitate the evaluation of the influence
of the wave-passage effect on the safety of tunnel structures.
For the analysis of longitudinal axial and bending deformations
of tunnel structures, several simplified design approaches have
already been proposed [9], such as the free field deformation
approach, the soil structure interaction approach and the mass-
spring system model [10]. The free field deformation method is a
simple and effective design tool. However, in many cases, espe-
cially in soft soils, this method provides an overly conservative
design. Although the soil structure interaction approach treats
the tunnel as a beam on an elastic foundation to account for the
interaction effects, the spring coefficients of the foundation are
not easy to determine. Both these methods depend on the selection
of the apparent velocity of the earthquake wave.

http://dx.doi.org/10.1016/j.compgeo.2014.10.003
0266-352X/Ó 2014 Elsevier Ltd. All rights reserved.
230 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243
In the mass-spring system model, the surrounding soil is divided
into several strips, and each strip is considered to be a lumped mass
[10]. The lumped masses are connected to one another and con-
nected to the bedrock. Meanwhile, the tunnel beam is connected
to the lumped masses of the soil. All the connections mentioned
above are composed of a spring and a viscous damper connected
in parallel. The parameters of each strip are obtained from the basic
shear formation of the soil layers. The vertical shear force cannot be
transmitted between adjacent strips, which indicates the model’s
imperfection in modelling the cooperative work of the foundation.
Moreover, the wave input from the bedrock is also difficult to
simulate.
Because the above simplified approaches all have their short-
comings, it is desirable and most appropriate to utilise numerical
soil–tunnel structure interaction models that take into account
the wave-passage effect. Several studies have been performed on
this aspect. Stamos and Beskos [11] proposed a frequency domain
boundary element method to analyse the dynamic response of
three-dimensional large underground structures to external or
internal dynamic forces or to seismic waves. Then, assuming the
underground structure to be infinitely long with a uniform
cross-section, they employed a special direct BEM in the frequency
domain for both the structure and the soil, which effectively
reduces the three-dimensional problem to a two-dimensional
one [12]. Yang and Hung [13] proposed a 2.5D method finite/
infinite element procedure in the frequency domain for dealing
with the ground vibrations induced by moving loads, which can
be used in computing the steady-state responses of a tunnel
embedded in a half-space subjected to moving loads. These meth-
ods are performed in the frequency domain; therefore, they cannot
be used directly with time-step integration techniques in solving
nonlinear problems. Park et al. [14] computed the spatially
variable ground motion displacement time histories in terms of
the coherency function and evaluated the longitudinal tunnel
response by performing a series of pseudo-static three-
dimensional finite element analyses. Yu et al. [2] proposed a
time-domain multi-scale method for the dynamic analysis of
underground structures, which couples FEM calculations with
coarse and fine meshes. Ding et al. [15] have also performed a
three-dimensional large-scale seismic finite element analysis for
an immersed tube tunnel in Shanghai. In these analyses, however,
specific issues such as the asynchronous earthquake wave input,
tunnel boundary treatment, as well as the tunnel wave-passage
characteristic are not addressed systematically.
In light of this, the present study aims to develop a 3-D time-
domain simulation model that is based on a well-developed finite
element framework to provide a feasible computational modelling
technique for the longitudinal seismic response of tunnels under
an asynchronous earthquake wave input. The ground is assumed
to be a horizontally layered half-space. The procedure starts with
the computation of the free field wave motion and the realization
of wave input into the finite domain to be analysed. By compiling
a FORTRAN program with the finite element software ANSYS, 3-D
numerical simulations of an elastic layered half space site
response under oblique seismic incidence of plane body waves
(P, SV or SH wave) are achieved. The accuracy and validity of
the predictions are proven by numerical examples. Based on this,
a 3-D soil–tunnel structure interaction model is established to
simulate the longitudinal response of tunnels when subjected to
asynchronous earthquakes, which display the wave-passage effect
on the tunnel. The treatment of the tunnel boundary in the
numerical model is proposed, and its efficiency is demonstrated.
Based on these numerical examples, the influences of the incident
angle and incident direction on the tunnel’s longitudinal seismic
response are also discussed.
2. Computational modelling technique for the soil-structure
interaction
2.1. Problem statement
This paper studies a numerical modelling technique for the
analysis of the longitudinal response of tunnels under asynchro-
nous seismic waves. Several assumptions are made for the analysis.
First, the foundation is assumed to be a horizontally layered half-
space that can reflect the stiffness inhomogeneous feature in the
vertical direction. Second, the asynchronous seismic wave field is
induced by oblique seismic incidence of body waves (P, SV or SH
wave). In fact, for the near field seismic problems, the incident
direction is never vertical when the seismic waves arrive at an
engineering site because the earthquake wave does not experience
enough refractions during its propagation. As a result, the wave-
passage effect occurs, which is a key factor among the four distinct
phenomena that give rise to the spatial variability of earthquake-
induced ground motions [16]. Third, at a certain distance away
from the source, the earthquake wave can be seen as a type of
plane wave because the source distance is much larger than the
scale of the source dimension [17].
Consider the idealised model drawn out for a practical engi-
neering problem in which a tunnel of infinite length is buried in
a horizontally layered half-space foundation and the whole system
is set into motion by an oblique incident plane body wave propa-
gating from the infinite foundation. The last assumption is that
the material and geometric properties of the tunnel structure are
uniform along the tunnel axis, which is similar to the assumption
made in several previous analyses [12,13] performed in the
frequency domain.
When the finite element method is used to solve the wave prop-
agation problems, a finite computational area is truncated from the
unbounded media, and the proper artificial boundary designed to
eliminate wave reflections is imposed on the truncated boundary.
To simulate the 3-D longitudinal response of tunnel structures
under an asynchronous seismic wave, the asynchronous free field
ground motion must be properly solved and the wave input must
also be correctly modelled.
2.2. Calculation of the free field motion
There have been several approaches to calculate the free field
motion excited by an oblique body wave incidence. The earliest
approaches were the transfer matrix method and the stiffness
matrix method [18–20], which are performed in the frequency
domain. These methods can be directly used to calculate the
steady-state wave field, while for transient wave propagation prob-
lems, Fourier transformation is needed. Several studies have also
been performed on time-domain earthquake analysis in a stratified
soil foundation, such as the finite-difference method proposed by
Alterman and Karal [21], Boore [22] and Jean [23]. Note that Joyner
and Chen [24] also presented a method for calculating the seismic
response of a system of horizontal soil layers that considers the
nonlinear stress–strain behaviour of soils. To avoid the precision
inconsistency of the frequency-domain analysis with those
obtained from the widely used time-domain method, a 1D finite
element method in the time domain was recently proposed by Liu
[25,26] to calculate the free field wave motions excited by an
anti-plane SH wave or in-plane P-SV wave oblique incidence in an
elastic layered half-space. Numerical results have demonstrated
that the 1D time-domain method is highly accurate and has good
stability. Zhao [27] proposed an accurate artificial boundary for
the in-plane P-SV wave oblique incidence problem to improve the
method.
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243
This improved 1-D time domain method is directly applied to
solve the free field motion because of the convenience of extending
it to 3-D problems. Consider a horizontally stratified soil over an
elastic underlying half-space, and establish a Cartesian coordinate
system shown in Fig. 1(a), where axis r and z represent the hori-
zontal and vertical directions, respectively, and h is the incident
angle. The material properties are also listed in the figure.
The finite element discretisation scheme shown in Fig. 1(b) is
performed with rectangular grids that meets the following
requirements: the vertical grid size Dz (may be varied in different
soil layers) is determined by the simulation precision criterion of
the 1-D method; the horizontal grids Dr are equal in size and
determined by the following expression
Dr ¼ c r Dt
where cr is horizontal apparent velocity of the incident wave that
can be calculated through cr = c/sin h with c representing the inci-
dent body wave speed; the discrete time step Dt is chosen to satisfy
the stability condition of the central difference method as given by
 obtained through the 1D time-domain method and is expressed as
Dt  Dz=cp ; for P-SV wave oblique incidence problem
Dt  Dz=cs ; for SH wave oblique incidence problem
After the above discretisation both in time and space, the nodal
wave motions in the free field can be written as
upm;n ¼ uðmDr; nDz; pDtÞ
where upm;n denotes the displacement vector of the node at location
(mDr, nDz) and time t = pDt with m, n and p all integers. The nodal
displacement has three components upm;n ¼ ður ; uz ; uw Þ, where for
the SH wave incidence, only the anti-plane displacement uw exists,
and for the P-SV wave incidence, both the in-plane displacements ur
and uz are needed.
In the 1D time-domain method, the appropriate artificial
boundaries are adopted to simulate the process of the wave trans-
mission towards infinity. Meanwhile, the incident waves are trans-
formed into equivalent loads applied on the truncated boundary.
According to Snell’s law, the characteristics of the lumped mass
finite element method and the central difference method, the finite
element equation of the oblique incidence motion can be estab-
lished such that the unknown quantities are the motions of the
nodes and are in one column. In this way, the 2D problem is
reduced to a 1D problem. The details of the method can be found
in the literature (see [25–27]). It is noteworthy that for the in-
plane problem, the oblique incident angle of the SV wave should
not exceed the critical incident angle hcr to avoid the generation
of reflected non-uniform waves.
Fig. 1. Schematic plot of the layered half-space and the finite element discretisation scheme.
231
2.3. Asynchronous wave input method
By solving the 1D equations, the displacements of nodes in any
vertical line, for example at r = 0, can be obtained. Then, the 3D
wave motions in the layered half-space can be determined based
on the characteristics of the travelling plane body wave. With
regard to 3-D numerical simulation of the asynchronous free field,
the wave input problem can be converted into a source problem
[28].
2.3.1. Extension of the free field to 3D
Consider the 3D numerical model established under the Carte-
sian coordinate system oxyz. The incident angle of the oblique seis-
mic plane body wave is h, and assumes that the projection of the
ð1Þ direction of incident wave on the horizontal plane has an included
angle of u with the positive x direction, as shown in Fig. 2(a). Here,
the angle u is called the incident direction. Element sizes along the
three axes are Dx, Dy and Dz, respectively, where Dz should be the
same as that in the 1D time-domain method.
Assume the asynchronous free field of a line of nodes at r = 0 are
ð2Þ u0(zk, mDt) (m = 0, 1, 2 . . ., k = 0, 1, 2 . . .), in which u0 = {u0r, u0z,
u0w}T. According to the propagation characteristic of the travelling
waves, the motions on the virtual grid nodes that satisfy the dis-
cretisation rule of Dr = crDt can be expressed directly. For example,
the free field motion ur(ri, zk, mDt) of the virtual grid nodes that is
ð3Þ located at (ri, zk) can be written as:
 
ri
ur ðr i ; zk ; mDtÞ ¼ u0 zk ; mDt  ð4Þ
cr
Consider the horizontal plane xoy that corresponds to the plane
z = zk as shown in Fig. 2(b). According to the propagation character-
istic of the plane body waves, by projecting any node of the 3D
model, for example, the node with the coordinate (xi, yj, zk), to
the r axis, with the projected radius rij expressed as
rij ¼ xi cos u þ yj sin u ð5Þ
and it can be deduced that the displacement of the node at time
t = pDt that is denoted by u(xi, yj, zk, pDt) is exactly the same as
the displacement at location (rij, zk) in the roz plane at time
t = pDt, which is denoted by ur(rij, zk, pDt). Particularly, for the con-
dition of the incident SH wave
 
r
ux ðxi ; yj ; zk ; pDtÞ ¼ u0w zk ; pDt  cijr sin u
 
r ð6Þ
uy ðxi ; yj ; zk ; pDtÞ ¼ u0w zk ; pDt  cijr cos u
uz ðxi ; yj ; zk ; pDtÞ ¼ 0
232 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243
Fig. 2. Schematic plot of the 3D expansion of the free field.
and for the condition of the incident P or SV wave
 
r
ux ðxi ; yj ; zk ; pDtÞ ¼ u0r zk ; pDt  cijr cos u
 
r
uy ðxi ; yj ; zk ; pDtÞ ¼ u0r zk ; pDt  cijr sin u ð7Þ
 
r boundary. Take node B on the viscous-spring boundary of the 3D
uz ðxi ; yj ; zk ; pDtÞ ¼ u0z zk ; pDt  cijr
Note that the projection of the nodes to the r axis always fall on
the locations between two virtual grid nodes, so linear interpolation
is needed to calculate the motions of the nodes in the 3-D model.
For example, for an incident P or SV wave, if (pDt–rij/cr)2[mDt,
(m + 1) Dt), the x displacement of node (xi, yj, zk) at time t = pDt
becomes
 
rij
ux ðxi ; yj ; zk ; pDtÞ ¼ u0r zk ; pDt  cos u
cr
 spring boundary under the free field motion, which can be deter-
ðpDt  r ij =cr Þ  mDt
¼ u0r ðzk ; mDtÞ þ
Dt

 ½u0r ðzk ; ðm þ 1ÞDtÞ  u0r ðzk ; mDtÞ cos u ð8Þ
Then, the oblique incidence wave motions in the 3-D free field
can be determined. The nodal velocity and acceleration can be
obtained using the central difference method.
2.3.2. Asynchronous wave input
When the finite element method is used to solve wave propaga-
tion problems in unbounded media, a proper artificial boundary is
imposed to eliminate the reflection of the out-going wave on the
truncated boundary of the finite computational area. Based on
the one-dimensional plane wave hypothesis, the viscous boundary
was first established by Lysmer and Kuhlemeyer [29]. To avoid the
stability problem when dealing with low frequency seismic loads,
the viscous-spring transmitting boundary is proposed [30] and has
been widely used in numerical analysis of dynamic problems
because of its good performance in absorbing the out-going wave
and simulating the elastic recovery capability of the exterior infi-
nite media. Here, the practical viscous-spring transmitting bound-
ary proposed by Liu et al. [31] is adopted; the coefficients of the
spring and dashpot are
( Recommended values of the correction factors for viscous-spring boundary in 3-D
K BT ¼ aT GR ; C BT ¼ qcs tangential direction
ð9Þ
K BN ¼ aN GR ; C BN ¼ qcp normal direction
where R is the distance between the scatter and the artificial bound-
ary; aT and aN are the correction factors introduced for the tangen-
tial and normal spring coefficients, respectively, that account for the
difference between the half space and whole space theories; their
recommended values are listed in Table 1.
Based on the conception of an isolated body, the incident waves
are transformed into equivalent loads applied on the artificial
finite element model for example (as shown in Fig. 3(a)), the free
field is assumed to be u0(xB, yB, zB, t) and the equivalent loads
excited by the free field that are applied on the truncated boundary
of the 3-D finite element model are expressed as Rf(xB, yB, zB, t). To
simulate the wave input accurately, the equivalent loads FB(xB, yB,
zB, t) applied on the artificial boundary should be given as
FB ðxB ; yB ; zB ; tÞ ¼ f b ðxB ; yB ; zB ; tÞ þ Rf ðxB ; yB ; zB ; tÞ ð10Þ
where fb(xB, yB, zB, t) represents the loads produced by the viscous-
mined directly through the spring-dashpot coefficients and the free
field.
Aiming to calculate the equivalent loads Rf(xB, yB, zB, t) gener-
ated by the free field, the consistent mass finite element equilib-
rium equations for undamped media are investigated such that
the inertial force and the elastic restoring force are in equilibrium.
Thus, the finite element equation of motion for node B can be writ-
ten as
€ Bo þ Mi u
ðMo u € Bi Þ þ ðKo uBo þ Ki uBi Þ ¼ 0 ð11Þ
where uBo represents the lateral elements associated with node B
around the truncated boundary; uBi represents the medial elements
associated with node B around the truncated boundary; Mo and Ko
represent the mass and stiffness matrixes, respectively, for the cor-
responding lateral elements, while Mi and Ki represent the mass
and stiffness matrixes, respectively, for the corresponding medial
elements. From Eq. (11), the equivalent loads excited by the free
field become
Rf ðxB ; yB ; zB ; tÞ ¼ Mi u
€ Bi þ Ki uBi ¼ ðMo u
€ Bo þ Ko uBo Þ ð12Þ
Table 1
problems.
Correction factors 3-D problems Used in this study
aT 0.5–1.0 0.67
aN 1.0–2.0 1.33
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243 233

Fig. 3. Schematic plot of the wave input method and calculation of the equivalent loads.

Eq. (12) implies the loads applied on node B due to free field
motion are determined by the motion of the elements associated
with it. Fig. 3(b) shows the explanation and schematic plot for cal-
culating this equivalent loads. This method can be compiled easily
and is more concise compared with the method in literature [32]
because a complicated derivation is avoided.
2.3.3. Numerical verification by simulation of an asynchronous free
field
The procedure is programmed in a FORTRAN compiler combined
with the finite element software ANSYS. Two dynamic problems
corresponding to the in-plane and anti-plane wave conditions,
respectively, are tested to prove the accuracy and validity of the
simulation method of the asynchronous free field. The horizontal
layered half-space foundation corresponds to a horizontal layer of
soil over an elastic underlying half-space. The material properties
of the soil and underlying half-space are listed in Table 2.
In both numerical examples, the displacement time history of
the incident body wave is a Dirac impulse; the time history
and amplitude spectrum are shown in Fig. 4.
Example 1: P wave oblique incidence with h = 45° and u = 30°.
In this example, the depth of the soil layer is 200 m, and the
plane P wave incidents from 350 m below the ground surface.
The size of the numerical model is 300 m  100 m  350 m
(length  width  depth), and the element size in all three
directions is 5 m, which meets the accuracy requirement.
Fig. 5 shows the time histories of the X displacement at observa-
tion point O (0, 0, 0) on the ground surface calculated using the 1D
time-domain method and 3-D numerical analyses. The theoretical
solution obtained by taking the inverse Fourier transform of the solu-
tion in frequency domain [18] is given as the exact solution. Clearly,

Table 2
The material properties of the soil and underlying half-space.

Layer Shear modulus (MPa) Poisson’s ratio Density (kg/m3) P wave speed (m/s) S wave speed (m/s)
Soil 40 0.25 1000 346 200
Underlying half-space 375 0.25 1500 866 500

1.0 0.8
Normalized amplitude

0.8
0.6
Amplitude (cm)

0.6
0.4
0.4

0.2
0.2

0.0 0.0
0.0 0.15 0.3 0.45 0.6 0 3.33 6.67 10 13.33 16.67
Time (s) Frequency (Hz)
( a ) ti m e h i s t o r y ( b ) a m p l i tu d e s p e c t r u m
Fig. 4. Dirac impulse displacement: (a) time history and (b) amplitude spectrum.
234 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243

X displacement at O(10-2m) 1.2 1.2

X displacement at O (10-2 m)
0.8 0.8

0.4 0.4

0.0 0.0

-0.4 -0.4
1D time-domain method 3D numerical analysis
exact solution exact solution
-0.8 -0.8
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) Time (s)
(a) comparison between 1D time-domain method (b) comparison between 3D numerical analysis
and exact solution and exact solution
Fig. 5. X displacement time history of the incident P wave at observation point O (0, 0, 0).

(a) t=0.30s (b) t=0.70s

(c) t=1.00s (d) t=1.40s

-0.66 -0.45 -0.24 -0.03 0.18 0.39 0.60 0.81 1.02 1.23

Fig. 6. X displacement field snapshot of the 3-D numerical example 1 (unit: cm).
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243 235

1.5 1.5

Y displacement at O1(10-2m)
Y displacement at O1(10-2m)
1.0 1D time-domain method 1.0 3D numerical analysis
exact solution exact solution

0.5 0.5

0.0 0.0

-0.5 -0.5
0.0 0.5 1.0 1.5 2.0 2.5 3.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0
Time (s) Time (s)

Fig. 7. Y displacement time history of the incident SH wave at observation point O1 (0, 0, 100).

(a) t=0.3s (b) t=0.6s

(c) t=1.00s (d) t=1.60s

-0.78 -0.42 -0.05 0.31 0.68 1.04 1.41 1.77 2.14 2.50

Fig. 8. Y displacement field snapshot of 3-D numerical example 2 (unit: cm).

the results of the 1D time-domain method and 3-D numerical anal-
yses agree well with the theoretical solution. The displacement field
snapshot of the 3-D numerical model shown in Fig. 6 displays the
reflection and refraction of the propagating waves.
Example 2: SH wave oblique incidence with h = 60° and u = 0°.
In this example, the depth of the soil layer is 100 m, and the
plane SH wave incidents are from 250 m below the ground surface.
The size of the numerical model is 300 m  60 m  250 m
(length  width  depth), and the element size for all three direc-
tions is 5 m, which meets the accuracy requirement.
236 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243

Viscous-spring transmitting
boundary

Tunnel structure
Viscous-spring transmitting
boundary
Incident wave

(a) global sketch of the finite element mesh (b) a quarter of the soil-structure interaction mesh
Fig. 9. The soil–tunnel structure interaction finite element model.

Fig. 7 shows the displacement time histories at observation
point O1 (0, 0, 100) for the y direction calculated using 1D time-
domain method and 3-D numerical analyses and compared with
the theoretical solution. Clearly, the results are in agreement with
the exact solution. The wave propagation process can be clearly
observed in the y displacement field snapshot shown in Fig. 8.
The efficiency of modelling the site asynchronous earthquake
responses induced by body waves with different wave types and
various incident angles and incident directions can be illustrated
clearly and concisely using the two numerical examples.
Consequently, the proposed procedure of the three-dimensional
numerical simulation technique is effective and reliable.
2.4. The soil–tunnel structure interaction model
In the finite element analysis of the soil-structure interaction
problem, the proper input of an asynchronous earthquake motion
is the key issue that influences the analysis accuracy.
2.4.1. Wave input method
For a normal soil-structure interaction analysis, the truncated
boundary is far enough from the structure, so the outgoing wave
scattered by the structure can be well absorbed by the viscous-
spring transmitting boundary; meanwhile, the equivalent loads
generated by the free field can be treated equal to the condition
when the structure is absent. The proposed methodology of 3-D
numerical simulation of the asynchronous free field could read-
ily be used for the soil-structure interaction analysis. However,
for the condition of a tunnel with an infinite length, the tunnel
structure is also truncated from the infinite foundation; there-
fore, the outgoing wave motion as well as the incident free field
near the tunnel truncated boundary become quite complicated.
In fact, the development of accurate artificial boundary
conditions and an accurate wave input for this model is difficult
and unnecessary.
It is known that the seismic response of a tunnel is dominated
by the surrounding ground response, that is, the longitudinal
response of a tunnel structure at a location is controlled by a
certain range of neighbouring ground motions. Because the 3-D
asynchronous free field is well simulated, directly applying the
wave input method to the infinite tunnel–soil interaction may be
acceptable to simulate the longitudinal seismic response of a tun-
nel structure within the middle range of the numerical model, as
long as the model is relatively long in the longitudinal direction.

120000 15000

10000
60000 Maximum values of Fx
Minimum values of Fx
Fz (kN)

Maximum values of Fz
Fx (kN)

5000 Minimum values of Fz

0
0
-60000
-5000
-120000
-300 -150 0 150 300 -300 -150 0 150 300
Tunnel longitudinal position (m) Tunnel longitudinal position (m)
(a) axial force Fx (b) shear force Fz
0.8
10000
0.6
5000
Mz (kNm)

Time (s)

0 0.4
Minimum Fx
Maximum values of Mz Maximum Fx
-5000 Minimum values of Mz Minimum Fz
0.2 Maximum Fz
Minimum Mz
10000 Maximum Mz
0.0
-300 -150 0 150 300 -300 -150 0 150 300
Tunnel longitudinal position (m) Tunnel longitudinal position (m)
(c) bending moment Mz (d) time when peak values appear
Fig. 10. Peak values of tunnel longitudinal internal forces and their appearing moments.
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243 237

120000
10000
X=-150m
60000 X=-75m
X=0m
Fx (kN) 5000 X=75m

Fz (kN)
0 X=150m
X=-150m 0
X=-75m
-60000 X=0m
X=75m
X=150m -5000
-120000
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Time (s) Time (s)
(a) axial force Fx (b) shear force Fz
8000

4000
(kNm)

0
Mz

X=-150m
-4000 X=-75m
X=0m
-8000 X=75m
X=150m
-12000
0.0 0.2 0.4 0.6 0.8 1.0
Time (s)
(c) bending moment Mz
Fig. 11. Time histories of the tunnel longitudinal internal forces at different positions.

2.4.2. Numerical verification
The validity of this proposed method is proven by the following
numerical example. The material properties of the semi-infinite
foundation are homogeneous without material damping and non-
linearity, which are the same as the underlying half-space listed in
Table 2. The material properties of the tunnel structure are linear
elastic; Young’s modulus, E = 3.0  104 MPa; Poisson’s ratio,
t = 0.2; and density, q = 2600 kg/m3. The cross section of the tunnel
is a 10 m  10 m square, with its walls, top plate and bottom plate
all 1.0 m thick. The centre of the tunnel cross section is 15 m below
the ground surface. Plane P wave incidents from 60 m below the
ground surface with an incident angle h = 45° and u = 30° with a
time history of its displacement shown in Fig. 4.
The ANSYS finite element software is used to perform the soil–
tunnel structure interaction analysis. The foundation is modelled
using 8-node solid elements, and the tunnel structure is modelled
using 4-node shell elements. The full scale of the numerical model
is 600 m  50 m  60 m (length  width  depth), and the ele-
ment size for all three directions is 5 m, which meets the accuracy
requirement. For simplicity, the interface between the soil and
structure is not modelled, but their nodes coincide. The soil ele-
ments within the tunnel structure are removed, except for the ones
that are adjacent to the truncated boundary for convenience of the
wave input. The soil–tunnel structure interaction finite element
model is illustrated in Fig. 9. The internal forces of the longitudinal
tunnel structure are calculated by using the weighted sum of the
nodal average stresses of the shell elements of the cross section.
Fig. 10(a)–(c) display the maximum and minimum values of the
longitudinal axial force Fx, shear force Fz and bending moment Mz,
respectively, along the tunnel structure during the whole analysis
time. The peak internal forces are uniform along a large portion of
the tunnel structure, although disturbance exists within a short
length near both boundaries of the model. Indeed, for this idealised
model, in which the tunnel is of infinite length and the oblique

0.8
1.2

0.6
uy (10-2m)
uz (10-2m)

0.8
0.4
0.4
0.2
0.0
0.0

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Time (s) Time (s)
(a) Vertical displacement uz (b) Horizontal displacement uy

Tunnel response at X=-100 Tunnel response at X=0 Tunnel response at X=100

Free field at X=-100 Free field at X=0 Free field at X=100

Fig. 12. Time histories of the tunnel displacements compared with the free field motion.
238 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243

-0.03 0.13 0.28 0.44 0.59 0.75 0.90 1.06 1.22 1.37 -0.02 0.12 0.27 0.42 0.56 0.71 0.86 1.00 1.15 1.30
Soil-structure interaction model Tunnel structure

(a) t=0.20s

(b) t=0.34s

(c) t=0.47s

(d) t=0.60s
Fig. 13. Z displacement field snapshot of the 3-D soil–tunnel structure interaction numerical model (unit: cm).

incident plane wave is propagating from the infinite half-space, the

Earthquake wave type
physical processes of the problem are symmetric on the timeline.
Parameters
That is, the seismic reaction processes at each location on the tun- Displacement history
nel are exactly the same, the only difference is when they occur. of input wave
Material properties
Therefore, the peak response of the tunnel should be the same, Material properties of and geometry of
which is illustrated by Fig. 10(a)–(c). Fig. 10(d) displays the the stratified soil tunnel structures
moments when the peak longitudinal internal forces of the tunnel
occur. This result implies that the peak responses are produced
successively along a wide portion of the tunnel. The projection of 1D FEM
the speeds at which the peak responses transmit along the Free field motion
longitudinal tunnel structure in the wave propagation direction in one column Soil-tunnel structure model
(r direction) is exactly equal to the horizontal apparent velocity Characteristics of
cr of the incident wave. For instance, from Fig. 10(d), the maximum traveling wave Apply the
Fz transmitting speed is (150(150))/(0.545–0.335)=1429 m/s, 3-D free field viscous-spring
motion transmitting
and its projection in the r direction, 1429  cos 30° = 1237 m/s, is
boundary
almost equal to the apparent velocity, cr = 866/sin 45 = 1225 m/s. Wave input method
Fig. 11 shows the time history of the longitudinal axial force Fx,
Boundary forces Boundary nodes
shear force Fz and bending moment Mz, at different locations in the
tunnel. Fig. 12 shows the vertical displacement and y displacement FORTRAN program compiled
histories of locations (100, 0, 10), (0, 0, 10) and (100, 0, 10), 3D FEM simulation
which are in the middle of the tunnel top plate compared with the
ANSYS modeling
displacement histories when the tunnel structure is absent. From
by means of APDL
Figs. 11 and 12, the propagation of the tunnel vibration along the
longitudinal direction can be clearly observed. Fig. 13 shows a ver- Fig. 14. Schematic flow-chart of the proposed soil–tunnel structure interaction
tical displacement field snapshot of the 3-D soil–tunnel structure analysis approach.
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243 239

interaction numerical model, from which the wave-passage effect
can be observed.
2.4.3. Summary of the procedure
All numerical results coincide with the traditional explanation
of the wave-passage effect and demonstrate that the proposed
method is able to reproduce most of the basic features of an
underground tunnel subjected to an asynchronous earthquake
motion and is an effective way to evaluate asynchronous
earthquake engineering problems. The schematic flow-chart of
the proposed method is shown in Fig. 14.
The method can provide satisfactory results for a wide range of
tunnel structures because the variability of the seismic response at
the boundary of the model is restricted within a certain distance.
This distance, called the boundary influence range, is needed to
fully deform the tunnel structure and the surrounding soil collab-
oratively. In fact, from the theory of soil-structure interaction, it is
easy to deduce that the stiffer the tunnel structure is compared
with the soil, the longer the distance needed to make the tunnel
structure and soil work cooperatively. Further modification of the
tunnel boundary treatment as well as determination of the bound-
ary influence range can be investigated to make the proposed
method more practical and save time for analysing tunnel
structure earthquake engineering problems.
3. Numerical applications
In this section, the proposed 3-D soil–tunnel structure interac-
tion method is used to analyse the influence of the incident angle
as well as incident direction on the tunnel longitudinal response
under an asynchronous earthquake motion. The numerical model
as well as the material properties are the same as Section 2.4.2.
Plane body waves (SH, P and SV waves) incident from 60 m below
the ground surface with various incident angles h and incident
directions u, and the time histories of the incident displacements
are chosen as shown in Fig. 4. The maximum and minimum values
of longitudinal internal forces of the tunnel structure, which are
evaluated at the location of X = 0, at the centre of the model, are
investigated for each case to gain a comprehensive insight into
the seismic behaviour of the longitudinal soil-structure interaction
system.
3.1. Asynchronous SH wave input
Fig. 15(a)–(e) shows maximum and minimum values of longitu-
dinal axial force Fx, shear force Fy and Fz, bending moment My and
Mz, respectively, during the whole analysis time. Interestingly,
although there are only horizontal displacement components in
the free field motion produced by the SH oblique incident wave,

160000 50000

80000 25000
Fy (kN)
Fx (kN)

0 0

-80000 -25000

-160000 -50000
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Incident angle (°) Incident angle (°)
(a) peak values of Fx (b) peak values of Fy
16000 30000

8000
15000
My (kNm)

0
Fz (kN)

-8000 0

-16000
-15000
-24000

-32000 -30000
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Incident angle (°) Incident angle (°)
(c) peak values of Fz (d) peak values of My
150000
Minimum value when =0°
75000 Maximum value when =0°
Minimum value when =15°
Mz (kNm)

0 Maximum value when =15°

Minimum value when =30°
-75000 Maximum value when =30°
Minimum value when =45°
-150000 Maximum value when =45°
Minimum value when =60°
-225000
Maximum value when =60°
-300000 Minimum value when =75°
10 20 30 40 50 60 70 80 Maximum value when =75°
Incident angle (°)
(e) peak values of Mz

Fig. 15. Peak values of the tunnel longitudinal internal forces produced by the incident SH wave.
240 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243

a vertical shear force and bending moment are generated when the
incident directions u is not zero. This phenomenon is shown in
Fig. 15(c) and (d), although Fz and My are generally much smaller
than Fy and Mz. This phenomenon cannot be predicted by
traditional simplified approaches where the tunnel structures are
idealised as beams; therefore it demonstrates the necessity of a
3-D meticulous structure-soil interaction analysis. Fig. 15(a) indi-
cates that the peak values of axial force increase with the incident
angle h for a given incident direction u and reach their maximum
when u is approximately 60°. Fig. 15(b) shows that peak values of
horizontal shear force Fy increase with the incident angle h for a
given u, while at a given incident angle, the peak values of Fy first
decrease and then increase with u, and their minimum is when u
is approximately 45°. Fig. 15(e) shows that the peak values of the
bending moment Mz increase with the incident angle h for a given
u and decrease with the incident direction u for a given h.
In Fig. 12, the displacements of the tunnel structure are close to
the free field motion, so it can be deduced that the ground is very
stiff, which is a condition where the free field deformation method
is a simple and effective design tool. However, this observation is
not completely in agreement with the established results based
on the free-field deformation method [9,33]. For example, for the
incident shear wave, the longitudinal strain and normal strain
obtained from the free-field deformation method reach their max-
imums when u is 45°, while numerical analysis Fx reaches its max-
imum when u is approximately 60° for each incident angle h.
Moreover, the shear strain and curvature reach their maximum
when u is 0°, while numerical analysis Fy decreases first and then
increases with u, and Mz decreases with the incident direction u.
These differences are important because the equation in the liter-
ature [9] calculates the strain and curvature of the tunnel using
body waves. This equation is derived on the assumption that a har-
monic wave propagates at a given angle of incidence in a homoge-
neous, isotropic, elastic medium, without taking the soil-structure
interaction into account. In earthquake engineering problems, the
oblique incident wave is propagating from the underlying half-
space to the stratified soil resting on it with an arbitrary transient
displacement history. The 3-D soil–tunnel structure interaction
numerical method has an advantage in this aspect and can
simulate the dynamic process much more accurately.
3.2. Asynchronous P wave input
Fig. 16(a)–(e) shows the maximum and minimum values of the
longitudinal internal forces during the whole analysis time.
Fig. 16(a) and (d) shows the peak values of the longitudinal axial

140000 18000

70000
9000
Fx (kN)

Fy (kN)

0
0
-70000

-9000
-140000

-210000 -18000
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Incident angle (°) Incident angle (°)
(a) peak values of Fx (b) peak values of Fy

40000
12000
20000
8000
0
My (kNm)
Fz (kN)

4000
-20000
0 -40000

-4000 -60000

-8000 -80000
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Incident angle (°) Incident angle (°)
(c) peak values of Fz (d) peak values of My

10000 Minimum value when =0°

Maximum value when =0°
5000 Minimum value when =15°
Maximum value when =15°
Mz (kNm)

0 Minimum value when =30°

Maximum value when =30°
-5000
Minimum value when =45°
-10000 Maximum value when =45°
Minimum value when =60°
-15000 Maximum value when =60°
Minimum value when =75°
-20000 Maximum value when =75°
10 20 30 40 50 60 70 80
Incident angle (°)
(e) peak values of Mz

Fig. 16. Peak values of tunnel longitudinal internal forces produced by incident P wave.
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243 241

force Fx. The bending moment My increases and then tends to
decrease as the incident angle h increases for a given u and reaches
its maximum when h is approximately 60°, while for each given
incident angle h, the values decrease as u increases. Fig. 16(b)
and (e) shows the peak values of the horizontal shear force Fy.
The bending moment Mz increases as the incident angle h increases
for a given u, while for a given h, the peak values first increase and
then decrease as u increases and reach their maximum when u is
approximately 45°. Fig. 16(c) shows that the peak values of the ver-
tical shear force Fz increase first and then decrease with h for a
given u. The vertical shear reaches its maximum when h is approx-
imately 30° and decreases as the incident direction u for a given h
increases.
The most unfavourable conditions of u obtained from these
results are generally in agreement with the established results
based on the free-field deformation method. For example, for an
incident P wave, the shear strain and curvature reach their
maximum when u is 45° and 35°160 , respectively. From the
numerical analysis Fy and Mz, first increase and then decrease with
the incident direction u, reaching their maximum when u is
approximately 45°. However, the proposed method can simulate
the process of the oblique incident wave propagating into the
semi-space and analyse the wave field, which is the superposition
of the incident P wave together with the reflected and refracted P-
SV waves, rather than the single plane P wave. The incident angle h
can also be varied to obtain the most unfavourable conditions. The
above advantages cannot be achieved by the traditional simplified
approaches.
3.3. Asynchronous SV wave input
To avoid the generation of reflected non-uniform waves, the
incident angle of the SV wave does not exceed hcr = arcsin
(cs/cp) = 35.26°. Fig. 17(a)–(e) shows the maximum and minimum
values of longitudinal internal forces during the whole analysis
time. As shown in Fig. 17(a) and (d), the peak values of the longi-
tudinal axial force Fx and bending moment My increase as the inci-
dent angle h increases for a given incident direction, while for each
given incident angle, the peak values decrease as u increases.
Fig. 17(b) and (e) indicate that the peak values of the horizontal
shear force Fy and bending moment Mz increase with the incident
angle h for a given incident direction u, while for a given h, the
peak values first increase and then decrease with u, reaching their
maximum when u is approximately 45°. Fig. 17(c) shows that the

300000 36000

200000
18000
100000
Fx (kN)

Fy (kN)

0 0

-100000
-18000
-200000

-300000 -36000
5 10 15 20 25 30 35 5 10 15 20 25 30 35
Incident angle (°) Incident angle (°)
(a) peak values of Fx (b) peak values of Fy
20000 80000

10000
40000
My (kNm)

0
Fz (kN)

0
-10000
-40000
-20000

-30000 -80000
5 10 15 20 25 30 35 5 10 15 20 25 30 35
Incident angle (°) Incident angle (°)
(c) peak values of Fz (d) peak values of My
45000
Minimum value when =0°
Maximum value when =0°
30000 Minimum value when =15°
Maximum value when =15°
Mz (kNm)

15000 Minimum value when =30°

Maximum value when =30°
0 Minimum value when =45°
Maximum value when =45°
Minimum value when =60°
-15000
Maximum value when =60°
Minimum value when =75°
-30000
5 10 15 20 25 30 35 Maximum value when =75°
Incident angle (°)
(e) peak values of Mz
Fig. 17. Peak values of the tunnel longitudinal internal forces produced by an incident SV wave.
242 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243
peak values of vertical shear force Fz generally decrease as h
increases for a given u and decrease with the incident direction
u for a given h.
3.4. Comparison and discussion
A number of factors contribute to the soil-structure interaction
effect, including the relative stiffness between the soil and struc-
ture, structure geometry, input earthquake motions and tunnel
embedment depth. From the above analyses, the capability of the
proposed 3-D numerical modelling technique in simulating the
seismic response of tunnels under an asynchronous earthquake
wave input is examined. It is observed by comparison that for
plane SH, P and SV waves incidence to a half-space foundation with
the same displacement amplitude, the most unfavourable condi-
tions of longitudinal internal forces in the tunnel are induced by
shear waves. In particular, the maximum values of the longitudinal
axial force Fx and bending moment My are produced by the SV
wave when the incident angle h is close to hcr and the incident
direction u is 0, as shown in Fig. 17(a) and (d). The maximum val-
ues of the horizontal shear force Fy and bending moment Mz are
produced by the SH wave when the incident direction u is 0 and
the incident angle h is relatively large, that is, nearly the grazing
incidence, as shown in Fig. 15(b) and (e). The most unfavourable
conditions of the vertical shear force Fz is produced either when
the incident SH wave has h approaching 0 and u approaching 90
or the incident SV wave has h approaching 0 and u approaching
0, as shown in Fig. 15(c) and Fig. 17(c), respectively. The wave field
motions in these two conditions are essentially consistent, and the
peak values obtained from the two conditions also coincide. These
results obtained by the proposed 3-D method are more precise and
credible than the traditional simplified approaches because the
process of the oblique incident plane body wave with an arbitrary
transient displacement history propagating from the underlying
half-space to the soil layers with a tunnel structure buried in it is
more accurately simulated.
4. Conclusions
In this paper, a 3-D numerical modelling technique for analysing
the longitudinal seismic response of tunnels under an asynchro-
nous earthquake wave input is extensively studied and proposed.
The 1D time-domain approach is applied to calculate the free field
motion. Based on the numerical approach for the external source
wave motion problems, the wave input is determined by applying
equivalent nodal force at the truncated boundary, and its accuracy
and validity are proved by numerical examples. This wave input
method is directly applied to the 3-D soil–tunnel structure interac-
tion model to simulate the longitudinal response of tunnels when
subjected to asynchronous earthquakes. The results indicate its
validity for modelling the seismic response of a tunnel with an infi-
nite length as long as the computational model is relatively long in
the tunnel’s longitudinal direction. Numerical examples are per-
formed to study the influences of the incident angle and incident
direction on the longitudinal seismic response of the tunnel. Several
preliminary conclusions are drawn regarding further study and
research. The proposed method is superior to the traditional simpli-
fied methods in the following aspects: (1) it can reflect the horizon-
tal layered half-space feature of the foundation; (2) it can simulate
the process of the oblique incident wave with various incident
angles and propagating directions in the analysis domain; (3) the
incident wave displacement can be an arbitrary transient time his-
tory; and (4) the soil and tunnel structure are fully interacted in the
time marching algorithms. Consequently, the results obtained by
the proposed method are more precise and credible than the
traditional approaches. Further improvements of the tunnel
boundary treatment method can also be investigated. The study
may facilitate further refined nonlinear numerical analysis as well
as a simplified analysis of the longitudinal seismic response of
tunnels.
Acknowledgements
The work reported in this paper is financially supported by the
National Natural Science Foundation of China (Project Nos.
51038007, 51408331) and the Self Initiative Research Program of
Tsinghua University (Project No. 2012THZ02-2). Useful suggestions
from the reviewers and the editor are also gratefully acknowledged.
References
[1] Uenishi K, Sakurai S. Characteristic of the vertical seismic waves associated
with the 1995 Hyogo-ken Nanbu (Kobe), Japan earthquake estimated from the
failure of the Daikei underground station. Earthquake Eng Struct Dyn
2000;29:813–21.
[2] Yu HT, Yuan Y, Bobet A. Multiscale method for long tunnels subjected to
seismic loading. Int J Numer Anal Methods Geomech 2013;37:374–98.
[3] Chen CH, Wang TT, Jeng FS, et al. Mechanisms causing seismic damage of
tunnels at different depths. Tunn Undergr Space Technol 2012;28:31–40.
[4] Amorosi A, Boldini D. Numerical modeling of the transverse dynamic
behaviour of circular tunnels in clayey soils. Soil Dyn Earthquake Eng
2009;29:1059–72.
[5] Kiryu S, Murono Y, Morikawa H. Seismic response of a cut-and-cover tunnel
isolated by polymer material. Earthquake Eng Struct Dyn 2012;41:2043–57.
[6] Liu HB, Song EX. Seismic response of large underground structures in
liquefiable soils subjected to horizontal and vertical earthquake excitations.
Comput Geotech 2005;32:223–44.
[7] Hwang RN, Lysmer J. Response of buried structures to traveling waves. J
Geotechn Eng Div 1981;107(2):183–200.
[8] Raheem SEA, Hayashikawa T, Dorka U. Ground motion spatial variability
effects on seismic response control of cable-stayed bridges. Earthquake Eng
Eng Vib 2011;10:37–49.
[9] Hashash YMA, Hook JJ, Schmidt B, et al. Seismic design and analysis of
underground structure. Tunn Undergr Space Technol 2001;16(4):247–93.
[10] Kiyomiya O. Earthquake-resistant design features of immersed tunnels in
Japan. Tunn Undergr Space Technol 1995;10(4):463–75.
[11] Stamos AA, Beskos DE. Dynamic analysis of large 3-D underground structures
by the bem. Earthquake Eng Struct Dyn 1995;24(6):917–34.
[12] Stamos AA, Beskos DE. 3-D seismic response analysis of long lined tunnels in
half-space. Soil Dyn Earthquake Eng 1996;15:111–8.
[13] Yang YB, Hung HH. A 2.5D finite/infinite element approach for modeling visco-
elastic bodies subjected to moving loads. Int J Numer Meth Eng
2001;51:1317–36.
[14] Park D, Sagong M, Kwak DY, et al. Simulation of tunnel response under
spatially varying ground motion. Soil Dyn Earthquake Eng
2009;29:1417–24.
[15] Ding JH, Jin XL, Guo YZ, et al. Numerical simulation for large-scale seismic
response analysis of immersed tunnel. Eng Struct 2006;28:1367–77.
[16] Kiureghian AD, Neuenhofer A. A coherency model for spatially varying ground
motions. Earthquake Eng Struct Dyn 1996;25:99–111.
[17] Newmark NM, Rosenblueth E. Fundamentals of earthquake
engineering. Prentice-Hall; 1971.
[18] Thomson WT. Transmission of elastic waves through a stratified soil medium. J
Appl Phys 1950;21(1):89–93.
[19] Wolf JP, Obernhuber P. Free-field response from inclined SV- and P-waves and
rayleigh-waves. Earthquake Eng Struct Dyn 1982;10:847–69.
[20] Wolf JP, Obernhuber P. In-plane free-field response of actual sites. Earthquake
Eng Struct Dyn 1983;11:121–34.
[21] Alterman Z, Karal FC. Propagation of elastic waves in layered media by finite
difference methods. Bull Seismol Soc Am 1968;58:367–98.
[22] Boore DM. Finite difference methods of seismic wave propagation in
heterogeneous material. Methods Comput Phys 1972;11:1–37.
[23] Jean V. SH-wave propagation in heterogeneous media, velocity-stress finite
difference method. Geophysics 1984;49(11):1933–57.
[24] Joyner WB, Chen ATF. Calculation of nonlinear ground response in
earthquakes. Bull Seismol Soc Am 1975;65(5):1315–36.
[25] Liu JB, Wang Y. A 1D time-domain method for in-plane wave motions in a
layered half-space. Acta Mech Sin 2007;23:673–80.
[26] Liu JB, Wang Y. A 1D time-domain method for 2D wave motion in elastic
layered half-space by antiplane wave oblique incidence. Chin J Theor Appl
Mech 2006;38(2):219–25 (in Chinese).
[27] Zhao M, Du XL, Liu JB, et al. Time-domain method for free field in layered half
space under P-SV waves of oblique incidence. China Earthquake Eng J
2013;35(1):84–90 (in Chinese).
 P. Li, E.-x. Song / Computers and Geotechnics 63 (2015) 229–243 243

[28] Liu JB, Lu YD. A direct method for analysis of dynamic soil-structure
interaction based on interface idea. Dyn Soil-Struct Interact – Curr Res China
Switz 1998;83:261–76.
[29] Lysmer J, Kuhlemeyer RL. Finite dynamic model for infinite media. J Eng Mech
ASCE 1969;95:759–877.
[30] Deeks AJ, Randolph MF. Axisymmetric time-domain transmitting boundaries. J
Eng Mech ASCE 1994;120:25–42.
[31] Liu JB, Gu Y, Du XL. Consistent viscous-spring artificial boundaries and viscous-
spring boundary elements. Chin J Geotech Eng 2006;28(9):1070–5 (in
Chinese).
[32] Zhao JF, Du XL, Han Q, et al. An approach to numerical simulation for external
source wave motion. Eng Mech 2007;24(4):52–8 (in Chinese).
[33] John CM, Zahrah TF. Aseismic design of underground structures. Tunn Undergr
Space Technol 1987;2(2):165–97.