Accepted Manuscript

Seismic Analysis of Long Tunnels: A Review of Simplified and Unified Meth-

ods

Haitao Yu, Yong Yuan, Antonio Bobet

PII: S2467-9674(16)30030-7
DOI: http://dx.doi.org/10.1016/j.undsp.2017.05.003
Reference: UNDSP 22

To appear in: Underground Space

Received Date: 2 January 2017

Revised Date: 4 May 2017
Accepted Date: 14 May 2017

Please cite this article as: H. Yu, Y. Yuan, A. Bobet, Seismic Analysis of Long Tunnels: A Review of Simplified
and Unified Methods, Underground Space (2017), doi: http://dx.doi.org/10.1016/j.undsp.2017.05.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 SEISMIC ANALYSIS OF LONG TUNNELS:
A REVIEW OF SIMPLIFIED AND UNIFIED METHODS
Haitao Yua,b*, Yong Yuanc, Antonio Bobetd, e
a
Key Laboratory of Geotechnical and Underground Engineering of Ministry of Education, Tongji
University, Shanghai 200092, China
b
State Key Laboratory for GeoMechanics and Deep Underground Engineering, China University of
Mining and Technology, Beijing, 100083, China
c
State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, Shanghai
200092, China
d
High-End Foreign Expert at Tongji University, Shanghai 200092, China
e
Lyles School of Civil Engineering, Purdue University, West Lafayette, IN 47907, USA
*E-mail: yuhaitao@tongji.edu.cn
Abstract: Seismic analysis of long tunnels is important for safety evaluation of the tunnel
structure during earthquakes. Simplified models of long tunnels are commonly adopted in
seismic design by practitioners, in which the tunnel is usually assumed as a beam supported
by the ground. These models can be conveniently used to obtain the overall response of the
tunnel structure subjected to seismic loading. However, simplified methods are limited due to
the assumptions that need to be made to reach the solution, e.g. shield tunnels are assembled
with segments and bolts to form a lining ring and such structural details may not be included
in the simplified model. In most cases, the design will require a numerical method that does
not have the shortcomings of the analytical solutions, as it can consider the structural details,
non-linear behavior, etc. Furthermore, long tunnels have significant length and pass through
different strata. All of these would require large-scale seismic analysis of long tunnels with
three-dimensional models, which is difficult due to the lack of available computing power.
This paper introduces two types of methods for seismic analysis of long tunnels, namely sim-
plified and unified methods. Several models, including the mass-spring-beam model, and the
beam-spring model and its analytical solution are presented as examples of the simplified
method. The unified method is based on a multiscale framework for long tunnels, with coarse
and refined finite element meshes, or with the discrete element method and the finite differ-
ence method to compute the overall seismic response of the tunnel while including detailed
dynamic response at positions of potential damage or of interest. A bridging scale term is in-
troduced in the framework so that compatibility of dynamic behavior between the macro- and
meso-scale subdomains is enforced. Examples are presented to demonstrate the applicability
of the simplified and the unified methods.

1
Keywords: earthquake engineering; tunnel structure; seismic analysis; simplified method;
multiscale method.

2
1 INTRODUCTION
Tunnels constitute a major part of civil infrastructure and serve as public transportation facili-
ties, sanitation, irrigation utilities and storage infrastructure (Yu et al., 2013a). In seismically
active areas, these tunnels are under earthquake-induced risks. Recent events such as the Kobe
Earthquake in Japan (1995), the Duzce Earthquake in Turkey (1999), the Chi-Chi Earthquake
in Taiwan (1999), the Bam Earthquake in Iran (2003) and the Wenchuan Earthquake in China
(2008) have shown that tunnels are susceptible to irrecoverable damage due to seismic load-
ing (Yu et al., 2013b; Yu et al., 2016a), some quite extensively (Yu et al., 2016b). The ob-
served damage provides sufficient evidence to suggest that the safety of tunnels in seismically
active areas is still an important issue, but not well understood yet, or at least not well consid-
ered during design.
Tunnels are subjected to various forms of deformation under seismic loading (Wang, 1993),
namely: (a) ovaling or racking shear deformation of the tunnel cross section due to the shear
waves propagating normal or nearly normal to the tunnel axis; (b) axial compression and ex-
tension generated by the components of seismic waves that produce motions parallel to the
axis of the tunnel; and (c) longitudinal bending caused by the components of seismic waves
producing particle motions perpendicular to the longitudinal axis.
Shear deformation of tunnels induced by the vertically propagating shear waves has been
widely studied by a number of researchers (Bobet, 2003; Kattis et al., 2003; Amorosi &
Boldini, 2009), and it has been known to be the critical mode of deformation for tunnels under
seismic loading. The tunnel lining is generally simulated as a buried structure subjected to
ground deformations under a two-dimensional plane strain condition. Two basic approaches
are commonly used to estimate the response of tunnels under shear deformation. One ap-
proach is to carry out dynamic, nonlinear soil-structure interaction analysis using finite ele-
ment methods. The input motions in these analyses are time histories emulating design
response spectra, and the input motions are applied to the boundaries of a “soil island” to rep-
resent vertically propagating shear waves. The second approach assumes that the seismic
ground motions induce a pseudo-static loading condition to the structure. This approach al-
lows the development of analytical relationships to evaluate the magnitude of seismically in-
duced strains in the tunnel structure (Penzien & Wu, 1998; Penzien, 2000; Huo et al., 2006).
These relationships are based on the premise that tunnel structures under seismic loading will
tend to deform due to the demand from surrounding ground, and thus the structure is designed
to accommodate the imposed deformations without loss of its structural integrity.

3
Design considerations for axial and bending deformations are generally taken along the tunnel
axis (Wang, 1993). The behavior of a tunnel is sometimes approximated to that of an elastic
beam subjected to longitudinal deformations imposed by the surrounding ground. Closed-
form solutions have been proposed to estimate the strains and curvature of the tunnel subject-
ed to a harmonic motion propagating at an angle to the tunnel axis (St. John & Zahrah, 1987).
These solutions can be conveniently used to obtain the overall response of tunnel structures
subjected to seismic loading. However, the solutions ignore the inertia forces and the interac-
tion between the tunnel and the surrounding ground, and thus may overestimate or underesti-
mate the structure deformations, depending on the rigidity of the structure relative to the
ground, which has already been noted by a number of researchers (Bobet, 2003; Huo et al.,
2006; Hashash et al., 2001). Clearly, analytical formulations are limited due to the assump-
tions that need to be made to reach the solution. In most cases, the design will require a nu-
merical method that does not have the shortcomings of the analytical solutions, as it can
consider the structural details, non-linear behavior, etc. For the analysis of axial and bending
deformations of tunnels, it is most appropriate to utilize three-dimensional (3-D) models
(Stamos & Beskos, 1996; Hatzigeorgiou & Beskos, 2010; Li & Song, 2015). With high-
performance computers and large mass storage capabilities, it is now possible to run large-
scale 3-D dynamic FEM analyses for tunnels under seismic load (Ding et al., 2006). However,
if the entire tunnel is modeled including structural details such as connection bolts between
liner segments and joints, the finite element model will be extremely large and computation
will be very expensive, if not impossible, even with the most advanced supercomputer.
The objective of the paper is to illustrate the seismic analysis of long tunnels with two differ-
ent methods: the simplified method and the unified method. The mass-spring-beam model,
and the beam-spring model and its analytical solution for long tunnels are discussed, as ex-
amples of the simplified method. The unified method is based on two multiscale approaches,
i.e. one is a continuum-based multiscale method coupling both coarse and refined finite ele-
ment meshes, and the other is a discrete-continuum multiscale method coupling the discrete
element and the finite difference model, to reduce computational cost and improve accuracy
of the solution. Examples are given to demonstrate the applicability of the methods.

2 LONG TUNNELS IN CHINA

Three examples of tunnels in China, namely the Hongkong-Zhuhai-Macau (HZM) immersed
tunnel, the Qingcaosha water conveyor tunnel, and the Longxi tunnel, are used to demonstrate

4
the ongoing booming construction of long tunnels in China. Note that the three tunnels are
built using three different construction methods, the Immersed Tunnelling Method, the Shield
Tunnelling Method and the New Austrian Tunnelling Method, respectively, which are the
most widely used tunnelling methods in China. Given that all the tunnels are in seismic areas,
the earthquake demand on the tunnels should be evaluated to ensure the safety of the struc-
tures.

2.1 The HZM immersed tunnel

The Hongkong-Zhuhai-Macau linkage (HZM linkage) is a major infrastructure project start-
ing construction in China. Crossing the Lingding Sea, the project links Hongkong, with its
industry and tourist attractions, to Zhuhai city and Macao. The main structures of the HZM
linkage are composed of a cable stayed bridge and an immersed tunnel via two artificial is-
lands. The total length of the linkage is 35,600 meters and the immersed tunnel is 5,664 me-
ters. The tunnel is composed of 33 concrete tunnel elements (see Fig. 1 (a)), and each element
consists of 8 concrete tunnel segments, each 22.5 m long (see Fig. 1 (b)). The cross section of
the tunnel is a rough rectangle with 37.95 meters in width and 11.4 meters in height (see Fig.
1(c)) that provides two vehicle barrels, and one barrel for pipelines and ventilation. Segments
are prefabricated one by one using reinforced concrete, and then assembled into a tunnel ele-
ment in a dockyard before shipping and immersing them at the site. The sediments at the site
are of marine origin and include silt, clay and sand above bedrock.
This tunnel is, as of today, one of the largest long-distance immersed tunnels in the world.
Given that the tunnel area is prone to earthquake activity, seismic performance of the tunnel is
investigated for the 120-year design life of the structure. Furthermore, due to the spatial varia-
tion of earthquake motions along long tunnels (Yu et al., 2013a; Hwang & Lysmer,1981; Park
et al.,2009), the influence of non-uniform seismic excitations on the immersed tunnel needs to
be evaluated and considered for its seismic design (Yuan et al., 2016; Yu et al., 2016c).

5
 Fig. 1: Layout of the HZM immersed tunnel (Unit: m). (Yu et al., 2016c)

2.2 The Qingcaosha water conveyor tunnel

Water supply is one of the key issues in Megacities such as Shanghai, China. The Qingcaosha
water-conveyance tunnel is a double-line shield tunnel used for water supply for the city of
Shanghai, China. The tunnel has a total length of 14 km. Its layout can be divided into three
segments: the island, cross-river and the land segment. Three working shafts are used to con-
nect the different tunnel segments. The general plan and location of the shield tunnel, includ-
ing the working shafts, are shown in Fig. 2. The soils through which the tunnel is excavated
are mainly composed of saturated clay, silt seams and sands.
The tunnel has been constructed using the shield method and is, as of today, one of the largest
long-distance shield tunnels with medium-size diameter in the world. The tunnel is composed
of two lines: the East-line Tunnel and the West-line Tunnel, as shown in Fig. 2. Each ring
consists of six segments, each with a length of 1.5 m. The inner and outer diameters of the
tunnel, for the island and land segments, are 5.84 m and 6.8 m respectively, whereas for the
cross-river segment, the diameters are 5.5 m and 6.4 m. A typical cross section is shown in
Fig. 3. A staggered-jointed assembly is adopted between segments. Seismic demand of the
tunnel needs to be evaluated to ensure the safety of the tunnel structure under the expected 7-
degree seismic intensity (PGA of 0.1g) in Shanghai (Yu et al., 2013a). Therefore, the perfor-
mance of the tunnel, not only its seismic response along its full length, but also the response
of the segments and joints, should be fully evaluated.

6
 Fig. 2: Map of the Qingcaosha tunnel. (Yu et al., 2013 a)

Fig. 3: Typical cross section of the Qingcaosha tunnel. (Yu et al., 2013 a)

2.3 The Longxi tunnel

In western areas of China, a number of tunnels have suffered severe damage during the
Wenchuan strong earthquake in 2008, some quite extensively. The Longxi road tunnel, locat-
ed 2 km from the epicenter of the Wenchuan earthquake, suffered significant damage and col-
lapsed when crossing a faulted zone, as shown in Fig. 4. The total length of the tunnel is 3,658
m, with a cross section composed of two parallel twin tunnels, one for each direction, separat-
ed 30 m between axes. The cross section of each of the twin tunnels is shown in Fig. 5. The
tunnel support was designed following the New Austrian Tunneling Method. It consists of a

7
primary and a secondary support, and an impermeable lining between the primary and sec-
ondary supports. The cross section of each of the twin tunnels has a total width of 9.20 m and
a maximum height of 8.04 m. The pavement is made of concrete with a thickness of 0.43 m.
Near the Eastern end, at a distance of 620 m from the portal, the tunnel crosses a pre-existing
fault (see Fig. 4). It is a reverse fault that, at the crossing of the tunnel, has a dip angle of 82 o
and has an angle with the tunnel axis of 85°, i.e. the fault is quasi-perpendicular to the tunnel.
The Longxi tunnel, one of the most damaged tunnels during the 2008 Wenchuan earthquake,
suffered from small to heavy damage both at the portal and inside the tunnel, while sections
close to the fault completely collapsed, as seen in Fig. 6. More detailed information of the
tunnel damage can be found in (Yu et al., 2016b). Current design codes for seismic design of
mountain tunnels focus on the stability of the portals and of sections of the tunnel near slope
surfaces, but do not consider potential damage that may occur at other sections of the tunnel,
except for ground failure (Wang et al., 2001). This is in contrast to the observations made af-
ter the Wenchuan Earthquake that reported damage to several tunnels that ranged from minor
cracking to even failure (Yu et al., 2016b). This conclusion should serve as a warning for the
need of better design of tunnels and other underground structures in seismic areas (Yu et al.,
2016b).

Fig. 4. Location of the Longxi tunnel and earthquake epicenter (not to scale). (Yu et al., 2016b)

8
 Fig. 5. Typical cross section of the Longxi tunnel (units in m). (Yu et al., 2016b)

Fig. 6: Collapse of the tunnel at the crossing of the fault. (Yu et al., 2016b)

3 SIMPLIFIED METHODS
Simplified methods for the seismic design of long tunnels are commonly favored by practi-
tioners. They can be conveniently used to obtain the overall response of a tunnel structure
subjected to seismic loading. Two simplified models, namely the mass-spring-beam model,
and the beam-spring model and its analytical solution for seismic analysis of long tunnels are
discussed.

3.1 Mass-spring-beam model

The mass-spring-beam model is widely used in seismic design and analysis of long tunnels,
especially for immersed tunnels (Kiyomiya, 1995). The mass-spring model was first proposed
for the earthquake response analysis of immersed tunnels based on shaking table model tests
and earthquake observations of an immersed tunnel in Japan (Okamoto & Tamura, 1973). It
relies on the assumptions that (1) the natural period of the ground is not influenced by the ex-

9
istence of the tunnel, and (2) the motion of the ground is induced by the shear vibration of soil
strata and only the fundamental mode is taken into consideration.
Fig. 7 is a sketch of the mass-spring model (Kiyomiya, 1995). The surface layer of the soil
along the tunnel axis is divided into a number of soil slices. Each slice is represented by an
equivalent mass-spring system that consists of a mass, a spring and a dashpot connecting the
mass to the base rock. The neighboring masses are then connected to each other along the
tunnel axis by springs and dashpots to simulate the connection between the adjacent soil slices.
By solving the dynamic equilibrium equation of the model, the ground displacements at the
positions of the soil masses and tunnel can be evaluated. Assuming the tunnel as a beam sup-
ported by soil springs, the seismic response of the tunnel subjected to the obtained ground
motions can be calculated through a dynamic time-history analysis. Note that the mass-spring
model is established with the assumption that the ground displacement is dominated by the
fundamental shear vibration. Clearly, the ground response calculated may not be accurate
when different soil layers are contained in a soil slice.
A new method (Li et al., 2016), i.e. one-dimensional multiple-degree-of-freedom (MDOF)
system, based on the equivalence of modes, has been developed to represent the one-
dimensional layered soil deposit. A series of one-dimensional MDOF systems are connected
to constitute a two-dimensional mass-spring model of the soil. The equivalent mass-spring
model is then used for the longitudinal seismic analysis of long tunnels, as shown in Fig. 8.
The modal equivalent MDOF system has the distinction and advantage that it can characterize
well the one-dimensional ground properties, under shaking, such as natural frequency and
hysteretic damping. Comparisons of the amplification function of the modal equivalent
MDOF system with the theoretical solution for the one-dimensional wave propagation pro-
vide excellent results (Li et al., 2016).

10
 Fig. 7: Mass-spring model for free field (Kiyomiya, 1995).

Fig. 8: Mass-spring-beam model for longitudinal seismic analysis of long tunnels.

3.2 Beam-spring model

The pseudo-static or dynamic analysis based on the beam-spring model is a simple method for
the seismic analysis of long tunnels (Hashash et al., 1998; Anastasopoulos et al., 2007). In this
model, the tunnel is assumed as a beam supported on the ground by soil springs (and dashpots)
representing the soil-structure interaction, as shown in Fig. 9. The seismic input is the free-
field motion at the depth of the tunnel. Generally, two basic approaches, analytical and nu-
merical, are employed for beams resting on elastic or viscoelastic foundations. Numerical
methods such as the finite element method are useful to analyze complex structures; however,
both time and spatial increments have to be considered in the dynamic algorithms, and thus
these methods are time-consuming and the numerical accuracy depends on the integration al-
gorithm. Computational efficiency can be improved if analytical solutions are available (Yu &
Yuan, 2014).

11
Yu et al. (2016d) proposed an analytical solution for the dynamic response of long lined tun-
nels subjected to travelling loads, taking into account both the inertia forces and the interac-
tion between the soil and the structure. Several assumptions were made: the tunnel was
assumed to be infinitely long with a uniform cross-section; its behavior was linear elastic; the
surrounding soil was assumed to be isotropic and homogeneous and behave as viscoelastic;
the travelling loads were plane harmonic loads and propagated parallel to the tunnel axis. A
solution was obtained using Fourier transform to simplify the governing equation in the space
domain, whereas the Laplace transform was employed to reduce the equation in the time do-
main. The governing equation, based on the integration transform, was expressed as an alge-
braic equation so that the solution was obtained in the frequency domain. Finally, the
convolution theorem was employed to convert the solution into the time domain.

Fig. 9: Long lined tunnel on a viscoelastic foundation subjected to dynamic loads.

Fig. 9 depicts the coordinate system and dimensions associated with an infinite long lined
tunnel. The tunnel has constant stiffness EI and mass per unit length  A , where
E  Young’s modulus; I  moment of inertia of the tunnel cross section;   density of the
tunnel liner; and A  area of the cross section of the tunnel. The tunnel is supported by a vis-
coelastic foundation with constant spring stiffness K and viscous damping C .
The solution is given by the function y( x, t ) , which is the vertical deflection of the tunnel
and F ( x, t ) , which is the harmonic travelling load propagating parallel to the tunnel axis (see
the x-axis in Fig. 9). The wave-passage loads can be expressed as:

0,  x  Vt 

F  x, t     x (1)
 P sin[2  t - V ],  x  Vt 
  

Where t, V ,  and P are time, velocity, frequency and amplitude of the loads, respectively.
The displacement response of the tunnel is (Yu et al., 2016d):

12
  cos u  r  x  
 
 t Vs
P  r   (t  s ) 
y  x, t   0   sin[2  s  ]e sin 
   u  t  s  
  drdsdu (2)
 A 
0 0 
 u   V  

where r , u and s are the intermediate variables of the integration transforms. The new vari-
ables  and  (u ) are defined as
C
 (3)
2 A

EI 4 K C2
 u   u   (4)
A  A 4(  A)2
The velocity v( x, t ) and acceleration a( x, t ) responses of the tunnel structure can be obtained
as the first and second derivatives of the displacement response with respect to time t . Ana-
lytical solutions for the bending moment M ( x, t ) and shear force Q( x, t ) of the tunnel can be
obtained taking the second and third derivatives of the displacement response with respect to
the coordinate x , and each multiplied by ( EI ) .

4 MULTISCALE METHODS
Simplified seismic analyses may not be able to incorporate enough details of the structure or
the damage at the location of potential failure. To address this problem, two multiscale ap-
proaches, namely the continuum-based multiscale method and the discrete-continuum
multiscale method, are developed to capture not only the seismic response along the entire
tunnel length, but also detailed structural response of the liner segments and their joints.

4.1 Continuum-based multiscale coupling method

The continuum-based multiscale approach proposed by Yu et al. (2013c) couples FEM calcu-
lations with coarse and fine meshes. The coarse-scale mesh is employed to capture seismic
response characteristics of the entire system, whereas the fine-scale mesh is employed to de-
scribe in detail the dynamic response at positions of potential damage or interest. This
multiscale method can significantly reduce the computational load. It adequately covers wide
areas of the entire tunnel-soil system, while including details in key locations of the tunnel.
Furthermore, the total number of finite elements can be kept within the range of computation
capacity of the machine used for the simulations.
The multiscale method consists of two steps. In the first step, a regular, e.g. coarse, 3-D Finite
Element Model is built based on geological data, tunnel geometry, construction process, etc.
Non-linear soil and/or structure models are included in the analysis as well as contact inter-

13
faces. The objective of this first step is to determine the seismic response characteristics of the
system and to identify areas where detailed, e.g. refined, analysis is needed. In the second step,
the coarse mesh in those areas of interest identified in the first step is replaced by a refined
mesh and a new simulation with the composite, i.e. coarse and fine, mesh is carried out. This
can be done while the coarse mesh still captures the seismic response characteristics of the
system.
A difficulty that arises in multiscale couplings is that high frequency waves may be spuriously
reflected at the fine/coarse interface, as noted by a number of researchers such as Holmes &
Belytschko (1976). This is because the time step used for the entire domain depends on the
size of the smallest element, and as a result one may observe spurious reflections of high fre-
quency waves at locations where there is a change of mesh density. An overlapping domain
between coarse and refined meshes is defined to solve the problem. A schematic of the cou-
pling method is shown in Fig. 10. The complete domain in the initial configuration is denoted

by  0 . The domain is subdivided into two subdomains: the refined subdomain denoted by  0 ,
R

and the coarse subdomain, denoted by  0 . The overlap of these two domains is denoted by
C

 
int
0 ; 0 denotes the boundary between the refined and the overlap subdomains and 1 de-

notes the boundary between the overlap and the coarse subdomains.

Fig. 10: The continuum-based multiscale model for long tunnels.

In the method, the total energy is taken as a linear combination of the refined and coarse mesh
energies. A scaling parameter  is introduced in the overlapping subdomain. The Hamiltonian
for the complete domain is taken as a linear combination of the refined and coarse subdomain.
That is,
H  (1  ) H R  H C (5)

14
To keep the continuity of the two divided subdomains, the refined model and the coarse mod-

el are constrained on the overlapping subdomain 0 by:

int

2
 
d I    diI    uiC ( X I )  uiR ( X I )     N JC ( X I )uiJC   N JR ( X I )uiJR   0 (6)
2 2

i i i  J J 

where d I is the representation of the displacement boundary conditions on the overlapping

subdomain; the superscripts C and R correspond to the coarse and refined subdomains, re-
spectively. The refined displacements are required to conform to the coarse displacements at
the discrete nodal positions of the refined model. The constraints are applied to all compo-
nents of the displacements.
In the coupling method, the constraints are appended to the dynamic equations with Lagrange
multipliers. The total Hamiltonian can be described as
HL  H  λTd  H  λTI dI (7)
I

where λ I   iI  is a vector of Lagrange multipliers, with components corresponding to the

components of the displacement of the refined node I .

The equations of motion with the Lagrange multipliers are
MIuCI fIextC fIintC fIdC fILC in C0
 (8)
 mIuI fI fI fI fI in R0
R extR int R dR LR

ext int d
where f I and f I are the external and internal nodal forces respectively, f I is the damping
L
node force, f I is the force due to the constraints enforced by the Lagrange multipliers; the

superscripts C and R correspond to the coarse and refined subdomains, respectively; M I and

mI are the lumped mass of coarse and refined subdomains at node I , respectively.
LC LR
The forces f I and f I are due to the constraints enforced by the Lagrange multipliers and
are given by
d J
f ILC   λ TJ   λ TJ DCJI (9)
J u CI J

d J
f ILR   λ TJ   λ TJ D RJI (10)
J u R
I J

where N IJ  N I ( X J ) , and

15
  d   d 
DCJI   CJ    NJIC I  , DRJI   RJ   NJIR I (11)
 uI   uI 

where N I is the shape function.

The explicit central difference method (CDM) is adopted to solve for the motion. First, trial
displacements and velocities are obtained neglecting the constraints. Second, Lagrangian mul-
tipliers are obtained to satisfy the constraints. Finally, updated final displacements and veloci-
ties are computed. The steps required for FEM multiscale coupling are as follows:
(1). Initial geostatic stress conditions are computed using the dynamic relaxation method
with the coarse mesh;
(2). Seismic simulation of the coarse model is performed;
(3). Sections of interest are identified after (2);
(4). Multi-scale FEM is implemented by replacing the coarse model with the refined model
in the sections of interest;
(5). Numerical simulation of the multi-scale model under seismic loading is carried out, us-
ing the dynamic explicit algorithm (CDM);
(6). Multi-scale FEM dynamic computation ends when the accumulated time is greater than

the maximum time tmax .

4.2 Discrete-continuum multiscale coupling method

Seismic analyses of long tunnels are generally performed using continuum-based methods
such as FEM or FDM (finite difference method) because of their suitability for a continuous
formulation of the problem considered. However, to resolve local phenomena such as crack
propagation and progressive failure at positions of potential damage, a very refined model is
desired. Furthermore, current constitutive models establish stress-strain relationships at the
macroscopic scale but may lose the capability of capturing response at the microscale. An al-
ternative to those methods is the discrete-based approach formulation, to describe the
mesomechanical or micromechanical response of materials and local failure processes. How-
ever, because of the demanding computational cost of discrete methods, it is impossible to
realistically model large engineering problems with the current capacity. Fortunately, in many
applications, the region of interest is localized, and complex or highly nonlinear response is
limited to well-defined areas. Therefore, multiscale methods coupling discrete-based ap-

16
proaches and continuum-based approaches have the advantage of both techniques and provide
an alternative solution to model problems efficiently.
A novel and reliable multiscale coupling method for dynamic simulations between continuum
and discrete models is proposed by Li et al. (2015). Two numerical methods, the continuum-
based FDM and the discrete-based DEM (discrete element method), are coupled. The process
starts with the discretization of the entire domain with finite difference meshes and discrete
particles. The meshes are employed to capture dynamic response characteristics of the system,
whereas the particles describe the dynamic response at the mesoscale or microscale. An over-
lapping domain is created between the finite difference meshes and the discrete particles, and
a bridging scale term is introduced to ensure the compatibility of energy in the overlapping
domain. In the overlapping domain (Fig. 11), the constraints between mesh and particle sub-
domains are imposed to the motion equations using Lagrange multipliers, which are devel-
oped in a consistent manner from the energy potential. The proposed method does not result
in spurious wave reflections and does not need additional filtering or damping in the overlap-
ping domain between the FDM meshes and the DEM particles.

Fig. 11: Bridging domain coupling discrete and continuum regions (Li et al., 2015).

The equations of motion for the coupling method, including the Lagrange multipliers are (Li
et al., 2015)

M J u J  f Jext C  f Jint C  f JLC , in 0C (12)

mI d I  fIext P  fIint P  fILP , in 0P (13)
where M J and mI are the lumped mass of node J and particle I , respectively; f Jext C and f Iext P

are the external nodal force and particle force, respectively; and f Jint C and f Iint P are the internal
nodal force and particle force, respectively.

17
The forces f JLC and f ILP are caused by the constraints, enforced by Lagrange multipliers, and
given by ( Li et al., 2015)

g K
f JLC   λ K   λ K G CKJ (14)
K u J K

g K
f ILP   λ K   λ K G PKI (15)
K u I K

where λ K is a vector of Lagrange multipliers, g K is the representation of the kinematic con-

straints in the overlapping domain and

 g   g 
G CKJ   K    N JK I  , G PKI   K     IK I  (16)
 u J   u I 
and N IK  N I  XK  is the interpolation function.
To solve the coupled dynamic system with the Lagrange multiplier method, an explicit algo-
rithm is developed based on the central difference method. The following describes the pro-
cedure to obtain displacements and velocities at time n+1, assuming that displacements and
velocities are known at time steps n and n+1/2. First, trial displacements and velocities are
obtained, neglecting the constraints. Second, Lagrangian multipliers are obtained to satisfy the
constraints. Finally, updated final displacements and velocities are computed. A user-defined
subroutine is developed based on the two commercial packages, fast Lagrangian analysis of
continua (FLAC) and particle flow code (PFC). The advantage of the multiscale continuum-
discrete coupling approach is that it allows modeling of local physical phenomena at the
mesoscopic or microscopic scales in very large systems.

5 EXAMPLES
Several examples using the simplified and unified seismic analyses of the Qingcaosha water-
conveyance tunnel (Fig. 2) are presented to illustrate the applicability of the methods. More
detailed information of the tunnel is provided in Section 2.2. To provide a better understand-
ing and yet limit the number of analyses, the following discussion is limited to the beam-
spring model and multiscale model, coupling coarse-refined meshes, for large-scale seismic
analyses.

5.1 Case I: Simplified seismic analyses

Due to the aforementioned limitations of the simplified method, only the cross-river segment
of the tunnel (Fig. 2) is considered here, which is assumed as an infinitely long lined tunnel

18
with external and inner diameters 6.4 m and 5.5 m, respectively. The tunnel is assumed linear
elastic, homogeneous and isotropic, with mass density  =2500kg/m3 and Young’s modulus

E  3.55 1010 Pa . The foundation is linear elastic, with spring stiffness K  3.5 107 N / m2
and viscous damping C  0 . A travelling harmonic load is imposed to the tunnel structure
from the origin of the coordinate system x  0 (see in Fig. 9), propagating along the positive
axis. The travelling harmonic load applied at time t is described in Eq. (1). The parameters of
the load are as follows: amplitude P  21kN ; frequency   2Hz ; and wave velocity
V  200m / s .
The analytical solution is provided in integral form in Eq. (2), which can be solved in combi-
nation with the Gaussian integral. Fig. 12 shows the comparison of the bending moment and
shear force responses at point x  500m of the tunnel under travelling and uniform harmonic
loads. The figure indicates that both the bending moment and shear force responses of the
tunnel under travelling loads are significantly larger than those under uniform loads. This can
be taken as an indication that, for long tunnels, the wave passage effect would dramatically
amplify the structural response; therefore, uniform loads should not be used in the seismic de-
sign, but rather appropriate consideration should be given to travelling loads.
It is of interest to observe how the velocity and frequency of the travelling load influence the
response of the tunnel. Fig. 13 shows the bending moments and the shear forces at point
x  500m for the travelling load, with velocity 200m/s, 400m/s, 600m/s, 800m/s and constant
frequency 2Hz. It can be seen from the figure that the velocity of the travelling loads has sig-
nificant influence on the response of the tunnel. In this case, the dynamic response for the
wave velocity 200 m/s is much larger than the other waves, and thus it can be considered as
the critical condition for the tunnel design.
Fig. 14 is a plot similar to Fig. 13, also showing the bending moments and shear forces at
x  500m , for the travelling load with frequency 2Hz, 4Hz, 6Hz, 8Hz and constant wave ve-
locity 200m/s. The figure indicates that the bending moment of the tunnel structure increases
with decreasing wave frequency. This result suggests that the influence of low frequency
waves should be considered for the seismic design of long tunnels.
Regarding the advantages (and limitations) of analytical solutions over numerical routines ap-
plied to practical engineering problems, the following comments apply: although the assump-
tion of linear elasticity limit the applicability of the analytical solution presented in this article,

19
it has the distinct advantage of providing a rational framework to carry out a systematic para-
metric analyses with the most relevant parameters.

(a) (b)
Fig. 12. Response of the tunnel liner under travelling and uniform harmonic loads, at x  500m : (a) bend-
ing moment; and (b) shear force.

(a) (b)
Fig. 13. Response of the tunnel liner for travelling harmonic waves with velocity 200m/s, 400m/s, 600m/s,
800m/s and constant frequency 2Hz, at x  500m : (a) bending moment; and (b) shear force.

(a) (b)
Fig. 14. Response of the tunnel liner for travelling harmonic waves with frequency 2Hz, 4Hz, 6Hz, 8Hz
and constant velocity 200m/s, at x  500m : (a) bending moment; and (b) shear force.

20
5.2 Case II: Unified seismic analyses
The multiscale method, coupling coarse-refined meshes, is used to estimate the large-scale
seismic response of the Qingcaosha water-conveyance tunnel (Fig. 2) in Shanghai. To high-
light the resolution of the method, the dynamic behavior of the tunnel segments and joints is
obtained as part of the solution.
Fig. 15 depicts the coarse-scale soil and tunnel models. The coarse-scale discretization is done
with eight-node hexahedral solid elements. The full size of the soil-tunnel model is 12,660
m  2,509 m  300 m (length  width  depth), and comprises a total of 1,323,978 nodes and
3,515,026 elements. The bottom boundary is allowed to have horizontal displacements, but
not vertical displacements; the lateral boundaries are taken in the model as free boundaries. In
the model, the closest distance from the tunnel to the lateral boundary is more than ten times
the spacing between the east and the west tunnels. Because the lateral boundaries are far from
the tunnel structure, the influence of wave reflection caused by the free boundaries on the
tunnel is negligible. The adequacy of the type of boundary and mesh size was verified by run-
ning a number of preliminary numerical tests, where the lateral boundaries of the discretiza-
tion were placed at different distances from the tunnel. It was decided that the size of the
discretization was acceptable when free-field conditions were recovered in the area between
the tunnel and the boundaries.
The multi-scale model is implemented by replacing the coarse model with the refined model
at positions of potential damage; these include the liner segments at particular cross sections
and the bolts that connect the segments (Fig. 16). In the refined model, each of the six seg-
ments in each ring is connected by M36 bolts (circumferential bolts), with a total of 24 bolts;
the joints between the segments of two adjacent rings have M30 bolts with a total of 16 bolts
(longitudinal bolts). Both the circumferential and longitudinal bolts are placed in tension. The
pre-tension force is about 176 kN. The refined mesh for the liner segments is composed of
eight-node hexahedral elements, the same elements used for the coarse mesh. The length of
the overlap domain between the coarse and refined discretization is 1.5 m (Fig. 16). The pre-
stressed bolts are simulated with two-node link elements (Fig. 16). The force generated by the
bolt is nonzero if and only if the bolt is in tension. Since the nodes of the link element have
the same degrees of freedom as the solid elements, the connection between bolts and tunnel
segments is consistent.

21
 Fig. 15. Coarse-scale soil and tunnel model.

(a) (b)
Fig. 16. Mesh of the multiscale tunnel model: (a) coarse-refined mesh; and (b) detail of the refined
mesh for a liner segment and bolts.

An elastic material model is assumed for both the concrete (liner segments) and steel (connec-
tion bolts). The nonlinear soil model (R-O model) is used for the entire ground, i.e. for both
the coarse and refined meshes. The material parameters for segments, bolts and soil properties
used in the model can be found in Yu et al., 2013c.
The Shanghai artificial seismic wave record has been imposed to the bottom of the discretiza-
tion, which corresponds to bedrock (Fig. 17). In the numerical model, the seismic input is ap-
plied to all the nodes at the bottom of the discretization, i.e. all the nodes, at a given time,
have acceleration in both the longitudinal and transverse directions. The simulations have
been carried out using 64 CPUs of the Magic Cube supercomputer at the Shanghai Super-
computer Center. Each run takes about 75 hours to complete. The following is a discussion of
the results obtained with the refined model at one of the potentially most critical sections
along the tunnel.

22
 0.10

0.05

A c c e le ra tio n (g )
0.00

-0.05

-0.10
0 5 10 15 20
Time (s )

Fig. 17. Acceleration time history of the Shanghai artificial seismic wave.

Fig. 18 illustrates the ratio between maximum principal stress and tensile strength of the bolts
connecting the segments at the most critical section of the east-line tunnel. What is shown is
the maximum stress at any given time in any of the bolts. The maximum tension, of about 240
MPa, occurs at about 8.2 s within the earthquake, which can be considered acceptable given
that the maximum allowable stress is 480 MPa to 640 MPa, depending on the grade of the
bolts used (Yu et al., 2013a). It is interesting to note that the largest values of the principal
stresses, i.e. maximum tension and maximum compression, occur at the contact between seg-
ments, as shown in Fig. 19, which is a contour plot of the maximum and minimum principal
stresses. The extreme values are due to the stress concentration induced by contact between
tunnel segments. Also, since the tensile strength of the concrete is reached during the earth-
quake, cracking may be expected.
0.55

0.50
Ratio between max. principal stress

0.45
and tensile strength

0.40

0.35

0.30

0.25

0.20
0 2 4 6 8 10 12 14 16 18 20
Time (s)

Fig. 18. Maximum tension in bolts.

23
 (a) (b)
Fig. 19. Contour plots of stresses in the tunnel liner segments at time 5.2 s. (a) maximum principal stress;
(b) minimum principal stress. Max. and Min. ratio in the figures denote the maximum and minimum ratio
of maximum principal stress / tensile strength or the maximum and minimum ratio of minimum principal
stress / compressive strength.

In reality, not only the ovaling deformation of the cross sections but also the deformation of
joints is of great concern. Fig. 20 and Fig. 21 are plots of the maximum opening and the max-
imum shearing dislocation, respectively, of any of the circumferential joints. The figures show
that the maximum opening of the circumferential joint is 2.3 mm and the maximum shear dis-
location of the circumferential joint is 2.4 mm. These values are smaller than the allowable
upper limits of 3-5 mm for joint opening and 4- 5 mm for joint dislocation, based on the
Shanghai code for foundation design (DGJ08-11-1999), and so the performance of the tunnel
under the seismic acceleration input is acceptable.
3

2
circumferential joints (mm)
Maximum opening of

-1

-2

-3
0 2 4 6 8 10 12 14 16 18 20
Time (s)
Fig. 20. Maximum opening of circumferential joints.

24
 2.5

2.0

Maximum dislocation amount of

circumferential joints (mm)
1.5

1.0

0.5

0.0

-0.5

-1.0
0 2 4 6 8 10 12 14 16 18 20
Time (s)
Fig. 21. Maximum shearing dislocation of circumferential joints.

6 CLOSING REMARKS
Two different methods, the simplified and the unified method, are introduced, which are em-
ployed for seismic analysis of long tunnels. The simplified methods such as the beam model
provide effective tools for practitioners. They allow readily identification of the variables con-
trolling the magnitude of the distortions and thus provide an insight into the behavior of the
structure. Furthermore, the simplified method and its solutions are invaluable to obtain a bet-
ter understanding of the interplay that exists between dynamic loads, viscoelastic foundation
and tunnel structure, to identify what are the most critical parameters for the problem, and to
provide first estimates or even a preliminary design. An added advantage is that they can be
used with little cost to conduct sensitivity analysis. However, the simplified method may not
be able to capture response and damage in structural details, in elements or at locations of po-
tential failure, due to the simplified assumptions taken to include the tunnel structure and the
soil-tunnel interaction.
Thus, a novel and reliable multiscale coupling method is proposed to solve the problem. The
multiscale approach couples continuum-based FEM calculations with coarse and fine meshes.
The coarse-scale mesh is employed to capture seismic response characteristics of the integral
system, whereas the fine-scale mesh describes in detail the dynamic response at the locations
of potential damage or of interest. An overlapping subdomain is created between the refined
and the coarse meshes, where Lagrange multipliers are used to enforce kinematic constraints.
The coupling method bridging domain is implemented to minimize, or at least reduce, spuri-
ous wave reflections at the coarse-fine mesh interface. The multiscale approach is further ex-
tended to couple discrete-based DEM and continuum-based FDM for dynamic analysis. The

25
advantage of the multiscale continuum-discrete coupling approach is that it allows modeling
of local physical phenomena occurring at the meso- or micro-scales in very large systems.
This multiscale modeling method can significantly reduce the computation load. It adequately
covers wide simulation areas of the entire tunnel-soil system and may also include element
details in key locations of the tunnel. Furthermore, the total number of elements can be kept
within the range of computation capacity of the supercomputer used for the simulations. Ex-
amples are presented to demonstrate the applicability of the multiscale method in large-scale
seismic analysis of long tunnels. The success using the method in a long and complex tunnel
reinforces the notion that full 3D simulations can be performed; with an effective selection of
the areas where a refined mesh is needed, results can be obtained even if the computational
capacity available is limited.

ACKNOWLEDGEMENTS
The research has been supported by the National Natural Science Foundation of China (grant
No. 51678438 & 51478343), the Shanghai Rising-Star Program (17QC1400500), and the
Shanghai Committee of Science and Technology (grant No. 16DZ1200302 & 16DZ1201904).
The authors acknowledge the support from the Fundamental Research Funds for the State Key
Laboratory for GeoMechanics and Deep Underground Engineering, China University of Min-
ing & Technology (SKLGDUEK1723).

REFERENCES
Anastasopoulos, I., Gerolymos, N., Drosos, V., Kourkoulis, R., Georgarakos, T., & Gazetas,
G. (2007). Nonlinear response of deep immersed tunnel to strong seismic shaking. Jour-
nal of Geotechnical and Geoenvironmental Engineering, 133(9), 1067-1090.
Amorosi A, & Boldini D. (2009). Numerical modeling of the transverse dynamic behavior of
circular tunnels in clayey soils. Soil Dynamics and Earthquake Engineering, 29, 1059-
1072.
Bobet A. (2003). Effect of pore water pressure on tunnel support during static and seismic
loading. Tunneling and Underground Space Technology, 18, 377-393.
DGJ08-11-1999, Shanghai code for design of foundation, Management Office for Shanghai
Technical Standard Project: Shanghai, 1999. (in Chinese)
Ding JH, Jin XL, Guo YZ, & Li GG. (2006). Numerical simulation of large-scale seismic re-
sponse analysis of immersed tunnel. Engineering Structures, 28, 1367-1377.
Hashash, Y.M.A., Tseng, W.S., & Krimotat, A. (1998). Seismic soil-structure interaction
analysis for immersed tube tunnels retrofit. Geotechnical Earthquake Engineering Soil
Mechanics III 2, ASCE Geotechnical Special Publication No.75, 1380-1391.

26
Hashash, Y.M.A., Hook, J.J., Schmidt, B., & Yao, J.I. (2001). Seismic design and analysis of
underground structures. Tunneling and Underground Space Technology,16, 247-293.
Hatzigeorgiou G.D., & Beskos D.E. (2010). Soil-structure interaction effects on seismic ine-
lastic analysis of 3-D tunnels. Soil Dynamics and Earthquake Engineering, 30, 851-861.
Holmes N., & Belytschko T. (1976). Postprocessing of finite element transient response cal-
culations by digital filters. Computer & Structures, 6, 211-216.
Huo H., Bobet A., Fernández G., & Ramírez J. (2006). Analytical solution for deep rectangu-
lar structures subjected to far-field shear stresses. Tunnelling and Underground Space
Technology, 21, 613-625.
Hwang R.N., & Lysmer J. (1981). Response of buried structures to traveling waves. Journal
of Geotechnical Engineering Division, 107(2), 183-200.
Kattis S.E., Beskos D.E., & Cheng A.H.D. (2003). 2D dynamic response of unlined and lined
tunnels in poroelastic soil to harmonic body waves. Earthquake Engineering and Struc-
ture Dynamics, 32, 97-110.
Kiyomiya O. (1995). Earthquake-resistant design features of immersed tunnels in Japan. Tun-
neling and Underground Space Technology, 10(4), 463-475.
Li C., Yuan J.Y., Yu H.T., & Yuan Y. (2016). Mode-based equivalent ground model for lon-
gitudinal seismic analysis of long tunnels. Soil Dynamics and Earthquake Engineering.
(Submitted)
Li M.G., Yu H.T., Wang J.H., Xia X.H., & Chen J.J. (2015). A multiscale coupling approach
between discrete element method and finite difference method for dynamic analysis. In-
ternational Journal for Numerical Methods in Engineering, 102, 1-21.
Li, P., & Song, E.X. (2015). Three-dimensional numerical analysis for the longitudinal seis-
mic response of tunnels under an asynchronous wave input. Computers and Geotechnics
63, 229-243.
Okamoto, S., & Tamura, C. (1973). Behaviour of subaqueous tunnels during earthquakes.
Earthquake Engineering and Structural Dynamics, 1(3), 253-266.
Park D., Sagong M., Kwak D.Y., & Jeong C.G. (2009). Simulation of tunnel response under
spatially varying ground motion. Soil Dynamics and Earthquake Engineering, 29, 1417-
1424.
Penzien J., & Wu C.L. (1998). Stresses in linings of bored tunnels. Earthquake Engineering
and Structure Dynamics, 27, 283-300.
Penzien J. (2000). Seismically induced raking of tunnel linings. Earthquake Engineering and
Structure Dynamic, 29, 683-691.
St. John, C.M., & Zahrah, T.F. (1987). Aseismic design of underground structures. Tunneling
and Underground Space Technology ,2(2),165-197.
Stamos, A.A., & Beskos, D.E. (1996). 3-D seismic response analysis of long lined tunnels in
half-space. Soil Dynamics and Earthquake Engineering,15, 111-118.
Wang J.N. (1993). Seismic design of tunnels: a state-of-the-art approach. New York: Parsons
Brinckerhoff Quade & Douglas, Inc.

27
Wang W.L., Wang T.T., Su J.J., et al. (2001). Assessment of Damage in Mountain Tunnels
due to the Taiwan Chi-Chi Earthquake. Tunneling and Underground Space Technology,
16(3), 133-150.
Yuan Y. Yu H.T., Li C., Yan X. & Yuan J.Y. (2016). Multi-point shaking table test for long
tunnels subjected to non-uniform seismic loadings - Part I: theory and validation. Soil
Dynamics and Earthquake Engineering. (Article in Press)
Yu H.T., Yuan Y., Qiao Z.Z., Gu Y., Yang Z.H., & Li X.D. (2013a). Seismic analysis of a
long tunnel based on multi-scale method. Engineering Structures, 49, 572-587.
Yu H.T., Yuan Y., Liu X., Li Y.W., & Ji S.W. (2013b). Damages of the Shaohuoping road
tunnel near the epicentre. Structure and Infrastructure Engineering, 9, 935-951.
Yu H.T., Yuan Y., & Bobet A. (2013c). Multiscale method for long tunnels subjected to
seismic loading. International Journal for Numerical and Analytical Methods in
Geomechanics, 37(4), 374-398.
Yu H.T., & Yuan Y. (2014). Analytical solution for an infinite Euler-Bernoulli beam on a vis-
coelastic foundation subjected to arbitrary dynamic loads. Journal of Engineering Me-
chanics, ASCE, 140(3), 542-551.
Yu H.T., Chen J.T., Yuan Y., & Zhao X. (2016a). Seismic damage in mountain tunnels due to
the Wenchuan strong earthquake. Journal of Mountain Sciences, 13(11), 1958-1972.
Yu H.T., Chen J.T., Bobet A., & Yuan Y. (2016b). Damage observation and assessment of
the Longxi tunnel during the Wenchuan earthquake. Tunnelling and Underground Space
Technology, 54, 102-116.
Yu H.T., Yuan Y., Xu G.P., Su Q.K., Yan X., & Li C. (2016c). Multi-point shaking table test
for long tunnels subjected to non-uniform seismic loadings - Part II: application to the
HZM immersed tunnel. Soil Dynamics and Earthquake Engineering. (Article in Press)
Yu H.T., Cai C., Guan X.F., & Yuan Y. (2016d). Analytical solution for long lined tunnels
subjected to travelling loads. Tunnelling and Underground Space Technology, 58, 209-
215.

28