ANN Aplid To Seismic Assmnt of Deep Tunnels

Istituto Universitario
di Studi Superiori
Universit degli
Studi di Pavia
EUROPEAN SCHOOL FOR ADVANCED STUDIES IN

REDUCTION OF SEISMIC RISK
ROSE SCHOOL
ARTIFICIAL NEURAL NETWORKS APPLIED TO THE

SEISMIC DESIGN OF DEEP TUNNELS
A Dissertation Submitted in Partial

Fulfilment of the Requirements for the Master Degree in
EARTHQUAKE ENGINEERING
by
TERAPHAN ORNTHAMMARATH
Supervisor: Dr CARLO G. LAI, Dr. MIRKO CORIGLIANO

April, 2007
The dissertation entitled Artificial neural networks applied to seismic design of deep
tunnels, by Teraphan Ornthammarath, has been approved in partial fulfilment of the
requirements for the Master Degree in Earthquake Engineering.
Dr. Carlo G. Lai
Dr. Mirko Corigliano
Abstract
ABSTRACT
The underground structure responses in competent rock are widely accepted to conform to the
surrounding ground during the earthquake which is different from the aboveground structure
responses. Additionally, past study suggested that underground structures in general were less severely
affected than surface structures at the same geographic location. However, in the 1995 Kobe, 1999
Chi-Chi, and 1999 Kocaeli earthquakes, the damages of underground structures in these events show
that most tunnels were located in the vicinity of the causative fault. One of the main contributions of
these damages is the near-fault effect. From the past observations, the near-field ground motions
produce ground motion characteristic in the vicinity (<10-25 km) different from that in the far-field
because of the directivity and fling step effects. It is important from the practical design point of views
to evaluate the seismic performance of underground structures at a particular site, especially in near
field. This study presents a simplified method to predict the maximum shear strains around the fault by
using Artificial Neural Networks (ANNs). Since the deformation of underground structures, both
longitudinal and transversal, is mainly caused by the longitudinal and shear strains respectively in
terms of the whole cross section, the proposed method is then based on identification of these shear
strains by ANNs.
The proposed method is applied to the Ariano Irpino fault located in Southern Italy that was
subjected to the December 5, 1456 earthquake. The near-field ground motion model developed by
Hisada and Bielak [2003] had been performed as this fault with assumed ground profile in that area.
The observation point is the point where seismometers or accelerograms would be place to
record the ground motion characteristics. The observation points had been assumed to be laid
next to the fault in different directions. For this study, it was assumed that we have
observation points only in 100- and 600-meter depths. These synthetic data would be used as
a training data for ANNs to learn the near-field ground characteristics. From this assumption,
the trained ANNs would be able to predict the maximum shear strains in other different
directions and depths. The computed results show that the ANNs has a possible capability to predict
the maximum shear strains around the fault vicinity.
Keywords: deep tunnel; seismic design; artificial neural networks; near-field earthquake
Acknowledgement
ACKNOWLEDGEMENTS
I would like to express my sincere to a lot of people during my study in MEEES program both in
ROSE school, Italy, and UJF, Grenoble:
-
First of all, I would like to thank Dr. Carlo Lai for introducing to me the topic of seismic design of
tunnels, for his motivation during this work. I also appreciate the financial help provided by the
ROSE school during the last stage of this study.
To Dr. Mirko Corigliano for his guidance, discussion, and reviews of my works, and the good
marks in the homeworks during Prof. Penders course.
To all my MEEES and ROSE school students, for all times that we studied, traveled, and got
drunk together.
To my parents, for every phone call that they talked with me when I was home sick.
My final acknowledgement is to my grandmother, who is fighting her last stage of lung cancer,
while I am writing this dissertation. The woman that consoled me when I cried, the woman that
held my hand when I was just a little boy, the woman that gave / bought a candy to me when I just
came back from school and very hungry, and the woman that asked me about the school after the
first day of my primary school. She is always very kind and nice. She always loves and worries
about my personal life. I would like to tell her that Whatever would happen please do not worry I
am already a grown up man. I love you as much as you love me. If we had to be separate, I know
that you always stay with me and I would be alright because I have you as a mentor to impart
invaluable life knowledge; in spite of this, I would miss you the most.
ii
Index
TABLE OF CONTENTS
Page
ABSTRACT ............................................................................................................................................i
ACKNOWLEDGEMENTS....................................................................................................................ii
TABLE OF CONTENTS...................................................................................................................... iii
LIST OF FIGURES ...............................................................................................................................vi
LIST OF TABLES.................................................................................................................................ix
1 INTRODUCTION .............................................................................................................................1
1.1
Objective ....................................................................................................................................4
1.2
General outlines of study ...........................................................................................................5
2 THE DAMAGES TO UNDERGROUND STRUCTURES ..............................................................6

2.1
Typologies of underground structures .......................................................................................6

2.1.1 Support system properties................................................................................................6
2.1.2 Construction Methods......................................................................................................7
2.1.3 Sectional typologies .........................................................................................................7
2.2
The ground motion parameters ..................................................................................................8

2.2.1 Peak Acceleration ............................................................................................................8
2.2.2 Peak Ground Velocity (PGV) ..........................................................................................9
2.2.3 Earthquake magnitude .....................................................................................................9
2.2.4 Duration of earthquake ..................................................................................................10
2.2.5 Frequency-content effects ..............................................................................................10
2.2.6 Near-field ground motion ..............................................................................................11
2.3
The past studies........................................................................................................................14

2.3.1 Dowding and Rozen [1978] ...........................................................................................15
2.3.2 Owen and Scholl [1981] ................................................................................................17
2.3.3 Yoshikawa and Fukuchi [1984] .....................................................................................17
iii
Index
2.3.4 Sharma and Judd [1991] ................................................................................................18

2.3.5 Asakura and Sato [1998]................................................................................................20
2.3.6 American Lifeline Alliance (ALA) [2001] ....................................................................21
2.4
Lesson learns from the past damages.......................................................................................23

2.4.1 Geological settings.........................................................................................................23
2.4.2 Concrete lining...............................................................................................................24
2.4.3 Distance effect ...............................................................................................................24
2.4.4 Overburden depth...........................................................................................................25
2.4.5 Slope stability.................................................................................................................25
2.4.6 Duration of earthquake ..................................................................................................25
2.4.7 Frequency-content effect ...............................................................................................25
2.4.8 Peak ground motion parameters.....................................................................................25
2.4.9 Near-fault effect .............................................................................................................25
3 SEISMIC DESIGN AND ANALYSIS PROCEDURES FOR UNDERGROUND STRUCTURES26

3.1
Seismic behaviors of underground structures ..........................................................................26
3.2
Design and analysis methods ...................................................................................................29

3.2.1 The free field deformation method ................................................................................29
3.2.2 The soil-structure interaction method ............................................................................37
3.2.3 Numerical methods ........................................................................................................49
3.2.4 Conclusion .....................................................................................................................51
4 ARTIFICIAL NEURAL NETWORKS...........................................................................................52

4.1
Introduction to Artificial Neural Networks..............................................................................52

4.1.1 Artificial Neural Networks (ANNs)...............................................................................52
4.1.2 Biological neural networks ............................................................................................52
4.2
Neural Network Architectures .................................................................................................53

4.2.1 Single layer neural network ...........................................................................................54
4.2.2 Multiple layers neural network ......................................................................................54
4.2.3 Other neural network architectures ................................................................................55
4.3
Activation Function .................................................................................................................55

4.3.1 Binary sigmoid function ................................................................................................56
4.3.2 Bipolar sigmoid function ...............................................................................................56
4.4
Training Algorithm ..................................................................................................................56

4.4.1 Learning rule..................................................................................................................56
4.4.2 Generalized delta rule ....................................................................................................57
4.5
The Backpropagation Network ................................................................................................57
4.6
Deficiencies of Backpropagation .............................................................................................61

iv
Index
4.6.1 Network paralysis ..........................................................................................................61

4.6.2 Local minima .................................................................................................................61
5 NUMERICAL EXAMPLES ...........................................................................................................62
5.1
Introduction..............................................................................................................................62
5.2
Test Description and Data Analyzed .......................................................................................62

5.2.1 The case study................................................................................................................62
5.2.2 Near-field ground motion modeling ..............................................................................63
5.2.3 Identification of shear strains by ANNs.........................................................................66
5.3
Results and interpretation of numerical computations.............................................................68

5.3.1 The YZ-shear strain at 0- strike degree axis ..................................................................69
5.3.2 The YZ-shear strain at 270-strike degree axis ...............................................................71
5.3.5 The XY-shear strain at 0-strike degree axis...................................................................77
5.3.6 The XY-shear strain at 270-strike degree axis...............................................................79
6 CONCLUSION AND FUTURE RESEARCH ...............................................................................86

6.1
Introduction..............................................................................................................................86
6.2
Numerical examples.................................................................................................................86
6.3
Future research.........................................................................................................................87
REFERENCES .....................................................................................................................................88
Index
LIST OF FIGURES
Page
Figure 1.1 Shapes of underground structures [Kawashima, 2000]............................................3

Figure 2.1 An idealization of widely spread underground structure, Kawashima [2000]..........8
Figure 2.2 Different relationships between the pulse period of the velocigram and earthquake
magnitude , Corigliano, M., et al. [2007]..........................................................................11
Figure 2.3 Rupture-directivity effects in the recorded displacement time histories of the 1989
Loma Prieta earthquake, for the fault-normal (top) and fault-parallel (bottom)
components, EERI [1995].................................................................................................12
Figure 2.4 Schematic diagram showing the orientations of fling step and directivity pulse for
...........................................................................................................................................13
Figure 2.5 Schematic diagram of time histories for strike-slip and dip-slip faulting in which 14
Figure 2.6 Calculated peak surface responses with associated damage observations for
earthquakes, Owen and Scholl [1981] ..............................................................................15
Figure 2.7 Comparison of peak ground velocity measured at the free surface and observed
damage, Corigliano, M. [2007].........................................................................................16
Figure 2.8 Damage Statistics, Sharma and Judd [1991] ...........................................................20
Figure 2.9 Seismic forces and probable failure modes, Asakura and Sato [1998]. ..................21
Figure 3.1 Acceleration response of Underground and On-ground structures, [Kawashima,
2000] .................................................................................................................................27
Figure 3.2 Axial deformation along the tunnel, Wang [1993]..................................................28
Figure 3.3 Bending deformation along the tunnel, Wang [1993]. ............................................28
Figure 3.4 Ovaling deformation of a circular cross section [Owen and Scholl, 1981]...........29
Figure 3.5 Geometry of a sinusoidal shear wave oblique to axis of tunnel, Wang [1993].......31
Figure 3.6 Free-field shear distortion of ground Wang [1993].................................................35
vi
Index
Figure 3.7 The interaction between elastic waves and tunnel lining .......................................36
Figure 3.8 Relationship between stress and thickness of tunnel lining, Okamoto[1973].........37
Figure 3.9 Induced forces and moments caused by waves propagating along tunnel axis.......39
Figure 3.10 Induced circumferential forces and moments........................................................44
Figure 3.11 Lining response coefficients vs. flexibility ratio, full-slip interface, and circular
tunnel, Wang [1993] .........................................................................................................45
Figure 3.12 Normalized lining deflection vs. flexibility ratio, full slip interface, and circular
lining, Wang [1993] ..........................................................................................................46
Figure 3.13 Lining (thrust) response coefficient vs. compressibility ratio, no-slip interface,
and circular tunnel, Wang [1993] .....................................................................................48
Figure 3.14 Simplified three-dimensional model for analysis of the global response of an
immersed tube tunnel, Hashash et al. [1998]. ...................................................................50
Figure 4.1 Biological neuron ....................................................................................................53
Figure 4.2 Single layer network................................................................................................54
Figure 4.3 Multiple layer network ............................................................................................55
Figure 4.4 Binary Sigmoid Function ........................................................................................56
Figure 4.5 Bipolar Sigmoid Function .......................................................................................56
Figure 4.6 The diagram illustrates the process of minimizing the error of a function through
the set of empirical data ....................................................................................................58
Figure 4.7 Typical two hidden layers backpropagation neural networks .................................59
Figure 5.1 Location of the Serro Montefalco tunnel (dotted line) along the Caserta-Foggia
railway line (dark solid line). The nearby active faults retrieved from the DISS 3.0.2
database are superimposed. The Ariano Irpino fault (ITGG092), which is assumed as a
potential seismic source in the dynamic analysis of the tunnel, is highlighted. The short
segment perpendicular to the tunnel axis, denotes the cross-section of the tunnel,
Corigliano, M., et al. [2007]. ............................................................................................63
Figure 5.2 Geological profile along the Serro Montefalco tunnel, Barla et al. [1986] .........63
Figure 5.3 The crustal velocity profile adopted for the solution of the auxiliary problem,
Corigliano, M., et al. [2007] .............................................................................................64
Figure 5.4 The general outline of the studied fault and its subfaults, Hisada and Bielak [2003]
...........................................................................................................................................66
Figure 5.5 Methodology adopted for this study........................................................................68
Figure 5.6 The general outline of seismic source zone.............................................................69
vii
Index
Figure 5.7 The exact 0-degree YZ-shear strain computed by Hisada code ..............................69
Figure 5.8 The attenuation of PGVs at 0-degree axis...............................................................70
Figure 5.9 The comparison between 0-degree YZ-shear strain computed by Hisada and ANNs
at different depth ...............................................................................................................70
Figure 5.10 The exact 270-degree YZ-shear strain computed by Hisada code ........................71
Figure 5.11 The attenuation of PGVs at 270-degree axis.........................................................71
Figure 5.12 The comparison between 270-degree YZ-shear strain computed by Hisada and
ANNs at different depths ..................................................................................................72
Figure 5.15 The comparison between 280-degree YZ-shear strain computed by Hisada and
Figure 5.18The comparison between 315-degree YZ-shear strain computed by Hisada and
Figure 5.19 The exact 0-degree XY-shear strain computed by Hisada code............................77
Figure 5.20 The attenuation of PGVs at 0-degree axis.............................................................78
Figure 5.21 The comparison between 0-degree XY-shear strain computed by Hisada and
Figure 5.22 The exact 270-degree XY-shear strain computed by Hisada code........................79
viii
Index
LIST OF TABLES
Page
Table 2.1 Summary of Earthquakes and Lining/Support systems of the Bored tunnels, Power
et al. [1998] .......................................................................................................................22
Table 2.2 Statistics for all bored tunnels, ALA [2001].............................................................22
Table 2.3 Tunnel Fragility-Median PGAs-Ground shaking hazard only, ALA [2001]............23
Table 3.1 Strains and curvatures due to body and surface waves.............................................32
Table 5.1The studied ground profile.........................................................................................65
Table 5.2 The features of studied fault, DISS v. 3.0.2..............................................................65
ix
Chapter 1. Introduction
1 INTRODUCTION
Nowadays, underground tunnel becomes parts of concern to the populace, whether it is for
transport or facility purposes. The underground tunnels are still being one of the most
challenging problems especially in seismic regions, whose underground structures must
sustain both static and seismic phenomena. The seismic responses of tunnels and in general of
underground structures is considerably different from that of above-ground facilities since the
overall mass of the structure is usually small compared with the mass of the surrounding
ground, and the stress confinement provides high values of radiation damping. These two
effects make an underground structure to response in accordance with the response of
surrounding ground without the resonance. Historically, underground tunnels have
experienced a lower rate of damage than aboveground structures; nevertheless, recently
several large earthquakes resulted in heavy damage to underground structures in major urban
centers and mountain territories. Earthquake effects on underground structures can be grouped
into two categories [Hashash et al., 2001]:
1. Ground shaking, i.e. the deformation of the ground produced by seismic waves
propagating through the earth's crust.
2. Ground failure such as uplift due to soil liquefaction, fault displacement, and slope
instability.
The ground shaking damage, which is the main concern in this report, is caused by the
seismic wave propagating through the medium. Some previous studies indicate several
reasons for these subsurface facility collapses, which also depend on different earthquake
mechanisms, geological settings, and tunnel properties itself [Hashash et al., 2001]. Moreover,
different typologies and sizes of underground facilities (e.g. lifelines, repositories,
transportation tunnels, etc.) have also been widely acknowledged for their different structural
responses from the past damage. It therefore appears prudent to determine under what
conditions of anticipated ground shakings, site conditions and opening configurations which
are necessary to consider seismic design.
On the other hand, for ground failure damage, which is caused by the fault movement, special
design, e.g. a flexible joint, over-sizing the cross section of the tunnel, or reinforcing the shear
zone across the fault, needs to be proposed to accommodate the permanent displacement,
localized the damage, and provide the means to facilitate repairs. Apart from the direct effect
of earthquake ground shaking, the damages from soil liquefaction and slope instability have
1
damaged portals and shallow excavations. The main reasons of these damages are due to the
surrounding soil / rock, where the large permanent ground deformation takes place. The
ground stabilization techniques, such as soil reinforcement, drainage, grouting, may be
effective in preventing damages from liquefiable deposits and slope instability. However, this
type of damage is beyond the scope of the present study.
Peak ground motion parameters, such as acceleration and particle velocity, can be correlated
with the extent of damage. Duration of the earthquake motion also contributes to damage.
Besides earthquake parameters, other important parameters that affect tunnel stability are
tunnel support and in-situ stresses. A thorough evaluation of the relation between these
parameters, soil conditions, and the performance of the underground structures was not
possible because a complete suite of data could not be compiled, since many of the documents
citing the earthquake performance of underground structures do not provide details on all the
important parameters. Furthermore, many of the events occurred many years ago, and it is no
longer possible to obtain complete information on all the relevant factors. Consequently, some
empirical relations between various parameters (for example, PGA and PGV), and tunnel
damage are approximate and tentative. Then, a more detailed definition of the relationship
requires more comprehensive studies than currently available.
In design and analysis of seismic effects, underground structures are classified by Kawashima
[2000] into 3 categories (Figure 1.1) based on their structural responses to seismic waves.
1. A pipeline embedded in ground along the surface, most pipelines for utilities are in
this group.
2. An underground structure with a large cross section along the ground surface.
Underground roads, parking lots, subways and common utility ducts are in this group.
3. An underground structure deep in vertical direction. Large trenches and ducts for
ventilation and approaches to tunnels are in this group.
Figure 1.1 Shapes of underground structures [Kawashima, 2000]
From past investigations, the underground structures response in accordance to the

surrounding soil / rock because of the small mass (inertia) of tunnel compared to the
surrounding ground and large radiational damping. Okamoto et al. [1973] measured the
seismic response of an immersed tube tunnel during several earthquakes show that the
response of a tunnel is dominated by the surrounding ground response not from the inertial
properties of the tunnel structure itself. Then, the focus of underground seismic design relies
on the estimation of the induced seismic strain and their interaction with the structures. From
many underground structural designs, [Wang 1993; Hashash et al. 2001; Corigliano, M., et al.
2007], the accuracy of the stress increment in the tunnel lining evaluated through pseudostatic approach are highly dependent on the prediction of maximum shear strain under the
free-field condition. The simplified methods to predict induced seismic strain in the ground
should be developed.
However, the general perception of structural and geotechnical engineers was that
underground structures presented minimal seismic risk unless they were intersected by active
faults, where slip could occur, or liquefaction of the surrounding ground could be triggered.
Even current design specifications in USA [AASHTO LRFD, 1998, Interim 2001] for
highway structures do not consider the seismic design in the transverse direction unless the
structure crosses an active fault. In seismic design for pipeline, the Eurocode 8 suggests the
use of simple approaches, Newmark and Kuesel-type of analyses, which are not considered
the uncertainty of different geological formations.
The damages of underground structures in the 1995 Kobe, 1999 Chi-Chi, and 2004 Niigata
earthquakes, show that most tunnels were located in the vicinity of the causative fault. One of
the main contributions of these damages is the near-fault effects, which their ground motions
are characterized by strong and coherent (narrow band) long period pulses. The consideration
of near-field effects to underground structural design would then be appropriated.
However, the near-fault ground motion is also strongly influenced by the fault geometries,
which make it even more difficult for the determination of ground motion characteristics in
the vicinity area. The direction of rupture propagation relative to the site, termed herein as the
rupture-directivity, and possible permanent ground displacements, fling step, are the major
effects in near fault region. For vertical strike-slip faults, the rupture directivity effects cause a
strong spatial variation in ground motions for a given closest distance to the fault in the
direction normal to the fault. In the parallel direction of the vertical strike slip fault, the fling
step effect is dominating the ground response. For dip-slip earthquake, however, the effects of
rupture directivity and fling step would be concentrated only in the direction normal to the
fault.
However, even with considerable knowledge of near-field ground motion, it is just after the
1999 Turkey and 1999 Chi Chi earthquakes that increase the ten-fold of near-field ground
motion recordings. For underground structures located in the vicinity of a fault rupture, it is
even more difficult to cope with the lacks of near-fault ground motion records and the
difficulty to adequately scale the time-histories recorded at the free surface. Synthetic records
are then the options for the area in which no properly near-field records. On the other hand, all
near-field ground motion records do not include forward rupture directivity effects. This
important effect gives even fewer proper data to use in the near-field structural design. This is
true even if time histories are being matched to a design spectrum because the spectral
matching process cannot build a forward rupture directivity pulse into a record where none is
present to begin with, Somerville, P. [2000]. Then, in this study, the semi-analytical near-field
ground motion model developed by Hisada and Bielak [2003], which had the capability to
generate directivity effect, had been chosen to use through this study. This model is based on
the computation of static and dynamic Greens function of displacements and stresses for a
viscoelastic horizontally layered half space.
The application of Artificial Neural Networks (ANNs), which considered as a non-parametric
approach, to compute the maximum shear strains in the near-fault ground motion condition
would be developed in this study. The ANNs is a powerful and viable tool in satisfactorily
emulating complex mapping functions between available and relevant inputs, i.e. acquirable
active fault parameters, ground profiles, and outputs, i.e. the maximum shear strains. For the
major advantage of ANNs over the physical based-model, considered as a parametric
approach, is that an investigated active fault may not behave within the class of models
initially assumed. This is also the main reason to hinder the practical engineers and decisionmaking people to recognize the complex behaviours of near-field ground motion.
1.1 Objective
Since the main reason of underground tunnels damages are near fault effect, which its ground
motion characteristic in the vicinity (<10-25 km) different from that in the far-field because of the
directivity and fling step effects, the consideration of near fault effect in underground structural design
is then crucial and inevitable. From many pseudo-static approaches to design underground
structures, [Wang 1993; Hashash et al. 2001; Corigliano, M., et al. 2007], it is able to predict
the seismic stress increment in the lining if the maximum shear strain is predicted correctly.
The application of ANNs to predict the maximum shear strains under free-field condition
4
around the fault would be developed in this study. The main objective of this study can be list
below.
1) Assessment of SOA (State Of the Art) on the seismic design of underground structures
with a detailed analysis of the typologies of damages caused by earthquakes

2) Development of ANNs for prediction of the maximum shear strains based on the
synthetic near-field ground motion generated using Hisada and Bielak [2003]
approach.
1.2 General outlines of study
The present thesis contains six chapters. It covers the damages to underground structures from
earthquakes (Chapter 2), the seismic design and analysis procedures for underground
structures (Chapter 3), the general summaries of Artificial Neural Networks (ANNs) (Chapter
4), the numerical examples of the proposed method to predict maximum shear strain using
ANNs (Chapter 5). Finally in Chapter 6 the conclusions and future research proposals are
presented.
1)
The literature of both damaged underground facility and underground seismic
design would be reviewed. Also the differences of site characteristics, ground
motion parameters, and fault mechanisms, which relate to different structural
responses, would then be pointed out. The general summaries about the lesson
learn from the past damages would then be provided at the end of the Chapter 2.
2)
The seismic behaviors of underground structures would be described at the

beginning of the Chapter 3. The current design and analysis of underground
structures in the simplified free field deformation and Pseudo-static methods
would be reviewed, and investigated their applicability. Some past examples in
different approaches would also be provided.
3)
The introduction of ANNs would be provided, along with its general procedure,
training algorithms, and the deficiencies of ANNs would be provided at the end of
Chapter 4.
4)
The capabilities of ANNs to reproduce the maximum near-field ground motion

shear strains would be performed based on the synthetic near-field ground motion
generated by Hisada and Bielak [2003] code. The seismic source would be based
on the Ariano Irpino fault geometries and characteristics in the Sannio region.
The data analyzed and the numerical computation procedures would be explained
in detailed. The ANNs would be given the sets of training data in order to let the
ANNs learning the near-fault characteristics from available data. To generalize and
expand its applicability, the input to the ANNs would be available field
measurement data which are the soil density, the maximum shear modulus, the
PGV, the distance in x and y directions from the fault origin, and the depth of the
observation points. At the end of Chapter 5, the trained ANNs would then be able
to regenerate the maximum shear strain in the different directions and depths from
the trained data.
5
Chapter 2. The Damages to Underground Structures
2 THE DAMAGES TO UNDERGROUND STRUCTURES

The response of underground excavation to earthquake shaking is influenced by many
variables. Some important factors of these are the structural typologies and depth of the
excavation, the properties of the soil or rock within which the excavation is constructed, the
properties of support systems, and the severity of the ground shaking. General concepts of
these variables would be provided, and also the past damaged studies to underground
structures are described with the explanation about the failure causes. The earthquake
characteristics related to damaged of underground structures would be also briefly discussed.
2.1 Typologies of underground structures
Since the growing and various use of subsurface facilities in urban area and different ground
conditions, their shapes and sizes would then be varied in a wide range resulting in their
unique structural responses. Some common types of these structures based on their support
systems, construction methods, and sectional typologies, which affect the structural responses
in design and analysis, are support system properties, construction methods, and sectional
typologies.
2.1.1 Support system properties
Underground structural responses within rock and soil can be quite different from each other
depending on the strength and quality of the surrounding ground, as well as on the size of the
opening. The rock mass can vary from very competent rock with massive blocks to very weak
and highly fractured rock. Thus the support requirements can also vary from no support at all
to fairly heavy steel sets.
Lined tunnel, normally, are lined with 0.2 0.5 cm of concrete or cast-in-place
concrete, where tunnels are excavated in soft rock or where the use of the tunnel requires high
safety and infrequent maintenance. The damages of lined tunnel from ground shaking include
cracking, spalling, and failure of the liner as a direct consequence of the shaking.
Alternatively, vibratory motion may reduce the strength of the ground thereby placing
additional loads on the tunnel support system.
Unlined tunnel will be used where the rock is sound and there is no or little water
infiltration. From the past records, however, this type of tunnels is more liable to damage than
lined and grouted tunnels even in rock zone, such damage occurs as rock fall, spalling, local
opening of rock joints, and block motion.
2.1.2 Construction Methods

The linear underground tunnel, which is the main concerned in this study, can be grouped into
three broad categorize, each having distinct design features and construction methods.
Generally, the reason of different construction methods comes from the different ground
conditions.
-
Bored or mined tunnel are unique because they are constructed without significantly
affecting the soil or rock above the excavation. Tunnels excavated using tunnel-boring
machines (TBMs) are usually circular. Situation where boring may be preferable to cutand-cover excavation include (1) significant excavation depths, and (2) the existence of
overlying structures.
Cut-and-cover structure are those in which an open excavation is made, the structure is
constructed, and fill is placed over the finished structure. This method is typically used
for tunnels with rectangular cross-section and only for relatively shallow tunnels (< 15
m of overburden). Example of these structures includes subway stations, portal
structures. From past experiences, this type of tunnels is more vulnerable than the other
methods, since its different soil-structure interaction between backfill and medium. In
terms of tunnel performance, the racking behavior of cut-and-cover tunnels appears to
be the seismic response most in need of careful attention.
Immersed tube tunnels are sometimes employed to traverse a body of water. This
method involves constructing sections of the structure in a dry dock, then moving these
sections, sinking them into position and ballasting or anchoring the tubes in place,
Hashash, et al. [2001].
2.1.3 Sectional typologies

The shape and size of underground structures can vary in a wide range. They may be
classified according to their sizes and shapes into three groups.
-
Laterally long (or linear) underground structures, e.g. pipeline utilities, underground
tunnel, are more affected to the axial deformation than flexural deformation, while the
effect of flexural deformation increases as the size increases, Kawashima [2000].
Large cross-sectional structures, e.g. underground roads, parking lots, subways, and
common utility ducts, are more subjected to in-plane deformation along the cross
section. And also because of its shape which is spread extensively in lateral direction as
well as longitudinal direction, the beam-type analysis is not realistic. This could be
extend to the property of subsurface ground varies in not only longitudinal direction but
also transverse direction. Then, an idealization of the structure by plate elements, Figure
2.1, may be more realistic, Kawashima [2000].
Figure 2.1 An idealization of widely spread underground structure, Kawashima [2000]
Vertically deep underground structures, e.g. large trenches and ducts for ventilation and
approaches to tunnels, are more subjected to in-plane deformation along the cross
section. In analyzing a vertically deep underground structure, three-dimensional and axisymmetric finite element idealizations are generally used. However, two-dimensional
analysis also provides sufficiently accurate results when a structure is sufficiently stiff
compared to the ground, Kawashima [2000].
2.2 The ground motion parameters

The ground motion parameters are essential for describing the important characteristics of
strong ground motion in compact, quantitative form. Many parameters have been proposed to
characterize the amplitude, frequency content, and duration of strong ground motions; some
describe only one of these characteristics, while others may reflect two or three. Because of
the complexity of earthquake ground motions, identification of a single parameter that
accurately describes all important ground motion characteristics is regarded as impossible,
[Jenning, 1985; Joyner and Boore, 1988].
2.2.1 Peak Acceleration
The most commonly used measure of the amplitude of a particular ground motion is the peak
horizontal acceleration (PHA). The PHA for a given component of motion is simply the
8
largest (absolute) value of horizontal ground acceleration obtained from the accelerogram of
that component. By taking the vector sum of two orthogonal components, the maximum
resultant Peak Horizontal Acceleration (PHA) (the direction of which will usually not
coincide with either of the measured components) can be obtained. For most earthquakes, the
horizontal acceleration is greater than the vertical acceleration, and thus the peak horizontal
ground acceleration also turns out to be the peak ground acceleration (PGA).
Ground motions with high peak accelerations are usually, but not always, more destructive
than motions with lower peak accelerations. Very high peak accelerations that last for only a
short period of time may cause little damage to many types of structures. Although peak
acceleration is a very useful parameter, it provides no information on the frequency content or
duration of the motion; consequently, it must be supplemented by additional information to
characterize a ground motion accurately, Kramer [1996].
While a surface structure responds as a resonating cantilevered beam, an underground
structure responds essentially with the ground, and then the PGA is not a good parameter to
describe the underground responses. The PGA seems to correlate with the extent of damage.
However, it is also dependent on the surrounding ground. A severe damage is often associated
with tunnels in soil and poor rock; where as damage to tunnels in competent rock is usually
(but not always) minor.
2.2.2 Peak Ground Velocity (PGV)
The peak ground velocity (PGV) is another useful parameter for characterization of ground
motion amplitude. Since the velocity is less sensitive to the high-frequency components of the
ground motions, the PGV is more likely than the PGA to characterize ground motion
amplitude accurately at intermediate frequencies. Moreover, the PGV is very importance for
underground structural design, since a rough estimate of the maximum shear strain could be
computed using the following well-known expression, Newmark [1967] relating the peak
ground strain (PGS) to the Peak Ground Velocity (PGV):
PGS =
PGV
C
(2.1)
where C denotes either the apparent speed of propagation velocity of S-waves in the
horizontal direction (VSapp) or the prevailing phase velocity of Rayleigh waves (VR).
2.2.3 Earthquake magnitude
The earthquake magnitude is a number characteristic of the earthquake depending on the
release of energy at the focus and independent of the location of the recording station. Several
different magnitudes scales are currently in use, the most common being the local magnitude,
ML; the surface wave magnitude, MS; the body wave magnitude, MB; and the moment
magnitude, MW. Physically, the magnitude has been correlated with the energy released by the
earthquake, as well as the fault rupture length, and maximum displacement, St. John and
Zahrah [1987]. A standard magnitude scale that is completely independent of the type of
instrument is the moment magnitude, and it comes from the seismic moment M0.
Mw =
log M 0
10.7
1.5
(2.2)
where M0 is
M 0 = Ad
(2.3)
where is the shear modulus of the faulted rock (about 3.31010N/m2), A is the area of the
fault (i.e. the product of its length and width), and d is the average displacement on the fault
(i.e. the slip which is the length of the slip vector of the rupture measured in the plane of the
fault).
2.2.4 Duration of earthquake
The level of earthquake damage is often strongly influenced by the duration of strong ground
motion. For the near-fault effect, the forward directivity time duration is short but with high
intensity.
Bommer and Martinez-Pereira [1999] review almost thirty different definitions of strongmotion duration, which have been proposed by various researchers since 1962. They identify
three generic groups: bracketed duration, uniform duration and significant duration. They
show that the use of different definitions can give rise to very different duration values for any
given strong-motion record. Selection of a specific definition should therefore depend on
purpose.
2.2.5 Frequency-content effects
The earthquake responses of structures and the ground are highly influenced by the frequency
content of the input motion. Frequency content is significant for buried structures in as much
as the response of the soil layers in which they are embedded is sensitive to frequency
content. It is therefore important to consider how the amplitude of ground motion is
distributed among the range of frequencies. For the near-fault effect, the frequency content of
the forward directivity effect is narrow band and low to intermediate frequency.
For the pulse period of the velocigram several authors proposed empirical correlations
between this quantity and moment magnitude (see Figure 2.2). These relations differ mainly
for the definitions used for the pulse period and for the database used in the regression
analysis.
10
Figure 2.2 Different relationships between the pulse period of the velocigram and earthquake magnitude ,
Corigliano, M., et al. [2006]
2.2.6
Near-field ground motion
In the immediate vicinity of a fault, ground motion exhibits various characteristics that can be
attributed to the orientation, direction and other features of propagation of the fault rupture.
These factors result in effects termed as rupture directivity and fling step effects. These
effects are significantly difference from those further away from the seismic source. The
estimation of ground motions close to an active fault should account for these characteristics
of near-field ground motions.
- Rupture directivity effect
The propagation of fault rupture toward a site at a velocity that is almost as large as the shear
wave velocity causes most of the seismic energy from the rupture to arrive coherently in a
single large long period pulse of motion which occurs at the beginning of the record. This
pulse of motion represents the cumulative effect of most of the seismic radiation from the
fault. The radiation pattern of the shear dislocation on the fault causes this large pulse of
motion to be oriented in the direction perpendicular to the fault, causing the strike-normal
peak velocity to be larger than the strike-parallel peak velocity. The enormous destructive
potential of near-fault ground motions was manifested in the 1994 Northridge and 1995 Kobe
earthquakes. In each of these earthquakes, peak ground velocities as high as 175 cm/s were
recorded, and the period of the near-fault pulse lie in the range of 1 to 2 seconds, comparable
to the natural periods of structures such as bridges and mid-rise buildings, many of which
were severely damaged, Somerville [2000].
Forward rupture directivity effects occur when two conditions are met: the rupture front
propagates toward the site, and the direction of slip on the fault is aligned with the site. The
conditions for generating forward rupture directivity effects are readily met in strike-slip
faulting, where the rupture propagates horizontally along strike either unilaterally or
bilaterally, and the fault slip direction is oriented horizontally in the direction along the strike
of the fault. The pulse of motion is typically characterized by large amplitude at intermediate
to long periods and short duration. However, not all near-fault locations experience forward
11
rupture directivity effects in a given event. Backward directivity effects, which occur when
the rupture propagates away from the site, give rise to the opposite effect: long duration
motions having low amplitudes at long periods, Somerville [2000]. Neutral directivity occurs
for sites located off to the side of the fault rupture surface (i.e., rupture is neither
predominantly toward nor away from the site).
The effects of rupture-directivity on ground displacements recorded during the 1989 Loma
Prieta earthquake are shown in Figure 2.3. The epicenter of the earthquake is near Corralitos
and Branciforte Drive, where the horizontal ground displacements are moderate on both faultnormal and fault-parallel components. This is attributed to backward directivity. At the ends
of the fault, however, at Lexington Dam and Hollister, forward directivity causes the
horizontal ground motions in the fault-normal direction to be impulsive and much larger than
the fault-parallel motions, which are similar to those near the epicenter. The large impulsive
motions occur only in the fault-normal direction and only away from the epicenter, Stewart et
al. [2001].
Figure 2.3 Rupture-directivity effects in the recorded displacement time histories of the 1989 Loma Prieta
earthquake, for the fault-normal (top) and fault-parallel (bottom) components, EERI [1995].
The conditions required for forward directivity are also met in dip slip faulting, including both
reverse and normal faults. The alignment of both the rupture direction and the slip direction
updip on the fault plane produces rupture directivity effects that are most concentrated updip
from the hypocenter near the surface exposure of the fault (or its updip projection if it does
not break the surface).
- Fling step effect
12
Moreover, the effects of surface faulting due to tectonic deformations have been recently
called fling step. These static displacements occur over a discrete time interval of several
seconds as the fault slip is developed. In contrast to forward directivity effects, which show
the large long-period pulse in the direction normal to the fault plane, the fling effects exhibit
long-period pulses and permanent static offsets in the direction parallel to the fault plane, and
therefore are not strongly coupled with the aforementioned dynamic displacements referred to
as the rupture-directivity pulse. In dip-slip faulting, both the fling step and directivity pulse
occur on the strike-normal component. The orientations of fling step and directivity pulse for
strike-slip and dip-slip faulting are shown schematically in Figure 2.4, and time histories in
which these contributions are shown together and separately are shown schematically in
Figure 2.5, Stewart et al. [2001].
Figure 2.4 Schematic diagram showing the orientations of fling step and directivity pulse for
strike-slip and dip-slip faulting, Stewart et al. [2001].
13
Figure 2.5 Schematic diagram of time histories for strike-slip and dip-slip faulting in which
the fling step and directivity pulse are shown together and separately, Stewart et al. [2001].
2.3 The past studies

Historically, underground facilities have experienced a lower rate of damage than
aboveground structures. However several large earthquakes resulted in damage to modern
underground structures both in mountain and urban areas.
Some investigators of the performance of underground excavations have attempted to develop
direct empirical relationships between damage levels and ground motion parameters. Such
attempts are fraught with difficulties since damage assessments may be highly subjective and
the peak ground motion experienced at a site must often be deduced from very incomplete
data. Therefore, it is desirable that arrays of strong instruments be deployed in and around
important underground structures.
Wang, et al. [2001] reported the various degrees of mountain tunnel damages after the 1999
Chi Chi earthquake. The most and often serious damage were found on the east of the
Chelungpu fault line (hanging wall) while damages on the footwall and other areas suffered
less. Then, the extent of damage to tunnel linings was influenced by the position of the
tunnels in relation to fault zones, ground conditions, and closeness to the epicenter and
surface slopes.
14
Information on the performance of underground openings during earthquakes is relatively

scarce, compared to information on the performance of surface structures. Therefore, the
summaries of published data presented in this section may represent only a small fraction of
the total amount of data on underground structures. There may be many damage cases that
went unnoticed or unreported. However, there are undoubtedly even more unreported cases
where little or no damage occurred during earthquakes, Indrawan [2001].
2.3.1 Dowding and Rozen [1978]
Dowding and Rozen [1978] identified three levels of damage for underground excavations in
rock due to ground shaking: these were no damage, minor damage, and damage. No damage
meant no new cracks or falls of rocks; minor damage meant new cracking and minor rock
falls; and damage included severe cracking, major rock falls, and closure. Dowding and
Rozen [1978] presented results of correlation of the estimated peak surface acceleration and
peak particle velocity with reported damage. Their correlations are reproduced in Figure 2.6.
The numbers on the ordinate axis are the designations of the cases tabulated in their paper.
The same numbering system also is used within the extensive tabulation of damage prepared
by Owen and Scholl [1981]. It should be noted that the peak ground motion parameters
(acceleration and velocity) were not recorded at the sites of the excavations but were
calculated using attenuation relationships. Free-field strong motion measurements from
instruments placed in and around tunnels could provide much more reliable data in the future.
(a)
(a) peak surface acceleration
(b)
(b) peak particle velocities
Figure 2.6 Calculated peak surface responses with associated damage observations for earthquakes, Owen
and Scholl [1981]
Review of data such as those presented by Dowding and Rozen [1978] suggests that no
15
damage should be expected if the peak surface accelerations are less than about 0.2g, and only
minor damage should be experienced between 0.2 and 0.4 g. The corresponding thresholds for
peak particle velocity are approximately 20 cm/s and 40 cm/s. Of these two correlations, the
one based on velocity is probably to be preferred as a design criterion because the peak
particle velocity resulting from an earthquake of a given magnitude can be predicted to fall
within reasonably narrow limits. Moreover, experience on the performance of mining
excavations adjacent to rock bursts has indicated that damage is better correlated with peak
velocity than peak acceleration McGarr [1983].
It should be emphasized that the above relationships hold for rock sites only, and may be
very different for underground structures in soil because the attenuation of motion with depth
and the confinement of the structure are very different than those for rock sites.
Unfortunately, similar relationships have not yet been derived for underground structures in
soil, St. John and Zahrah [1987].
Dowding and Rozen [1978] also summarized two relationships involving tunnel damage.
First, the observed damage is compared to Modified-Mercalli (MM) Intensity levels for
aboveground structures. Secondly, the damage level is correlated to Richter magnitude and
distance between epicenter and tunnel location. The no damage zone with acceleration up to
0.19g, is equivalent to MM VI-VIII; the minor damage zone with acceleration up to 0.5g is
equivalent to MM VIII IX. It is clear that at peak surface accelerations which are expected
to cause heavy damage to aboveground structures (MM VIII - IX) there is only minor damage
to tunnels. Comparatively, then, tunnels are less vulnerable to damage from shaking than
aboveground structures at the same intensity level as determined from surface motions.
However, the values of PGV suggested by Dowding and Rozen [1978] are typical for nearfault earthquake and for such events the predictions of PGV made by attenuation relations
carry a certain level of uncertainty. Bray and Rodriguez-Marek [2004] developed a more
reliable relation of PGV in the near fault region. This relation has been used to correlate the
PGV to the damage thresholds defined by Dowding and Rozen [1978] as illustrated in Figure
2.7, Corigliano, M. [2007].
Figure 2.7 Comparison of peak ground velocity measured at the free surface and observed damage,
Corigliano, M. [2006].
16
However, damage resulting from fault displacement must still be considered. Based on their
study, Dowding and Rozen [1978], concluded primarily for rock tunnels, that;
-
Tunnels are much safer than aboveground structures for a given intensity of shaking.
Tunnels deep in rock are safer than shallow tunnels
No damage was found in both lined and unlined tunnels at surface acceleration up to
0.19g
Minor damage consisting of cracking of brick or concrete or falling of loose stones was
observed in a few cases for surface accelerations above 0.25g and below 0.4g.
No collapse was observed due to ground shaking effect alone up to a surface

acceleration of 0.5g
Severe but localized damage including total collapse may be expected when a tunnel is
subject to an abrupt displacement of an intersecting fault.
2.3.2 Owen and Scholl [1981]

These authors documented additional case histories to Dowding and Rozen [1978]s, for a
total of 127 case histories. In addition, they suggested the following:
-
Little damage occurred in rock tunnels for peak ground accelerations below 0.4g.
Severe damage and collapse of tunnels from shaking occurred only under extreme
conditions, usually associated with marginal construction such as brick or plain concrete
liners and lack of grout between wood lagging and the overbreak.
Severe damage was inevitable when the underground structure was intersected by a fault
that slipped during an earthquake. Cases of tunnel closure appeared to be associated with
movement of an intersecting fault, landslide, or liquefied soil.
Deep tunnels were less prone to damage than shallow tunnels.
Duration of strong seismic motion appeared to be an important factor contributing to the

severity of damage to underground structures. Damage initially inflicted by earth
movements, such as faulting and landslides, may be greatly increased by continued
reversal of stresses on already damaged sections.
2.3.3 Yoshikawa and Fukuchi [1984]

These authors described the damage of railway tunnels in Japan from different earthquakes,
which magnitude ranging from 7.0 to 7.9. This paper reported a vague tendency of the
number of damages not dependent only on the earthquake magnitude, but also on the
geological settings of railway tunnels.
The ground failure, which is caused by slope stability, is the main reason of damages from
these records, since Japan mountain tunnels have been and are constructed around the sloping
17
area. The heavy damage could be observed at the intersection of the railway and fault lines.
The number of damaged tunnels decreased with respect to the farther distance from the
hypocentral zone coinciding with the acceleration attenuation. The high seismic vulnerability
of joint portion (e.g. portal part) had also been notified by the authors according to the statistic
records. Deformations were mostly tied to faults or where the strata show sudden change in
strength.
2.3.4 Sharma and Judd [1991]
The authors extended Owen and Scholl [1981]s work and collected qualitative data for 192
reported observations from 85 worldwide earthquake events. They correlated the vulnerability
of underground facilities with six factors: overburden cover, rock type (including soil), peak
ground acceleration, earthquake magnitude, epicentral distance, and type of support. It must
be pointed out that most of the data reported are for earthquakes of magnitude equal to 7 or
greater. Therefore, the damage percentage of the reported data may appear to be astonishingly
higher than one can normally conceive.
The results are summarized in the following paragraphs. These statistical data are of a very
qualitative nature. In many cases, the damage statistics, when correlated with a certain
parameter, may show a trend that violates an engineers intuition. This may be attributable to
the statistical dependency on other parameters which may be more influential.
-
The effects of overburden depths on damage are shown in Fiugre 2.8A for 132 of 192
cases. Apparently, the reported damage decreases with increasing overburden depth.
Figure 2.8B shows the damage distribution as a function of material type surrounding
the underground opening. In this figure, the data labeled Rock(?) were used for all
deep mines where details about the surrounding medium were not known. The data
indicate more damage for underground facilities constructed in soil than in competent
rock.
The relationship between peak ground acceleration (PGA) and the number of damaged
cases are shown in Figure 2.8C.
-
For PGA values less than 0.15g, only 20 out of 80 cases reported damage.
For PGA values greater than 0.15g, there were 65 cases of reported damage
out of a total of 94 cases
Figure 2.8D summarizes the data for damage associated with earthquake magnitude.
The figure shows that more than half of the damage reports were for events that
exceeded magnitude M =7.
The damage distribution according to the epicentral distance is presented in Figure

2.8E. As indicated, damage increases with decreasing epicentral distance, and tunnels
are most vulnerable when they are located within 25 to 50 km from the epicenter.
Among the 192 cases, unlined openings account for 106 cases. Figure 2.8F shows the
18
statistical damage data for each type of support. There were only 33 cases of concrete
lined openings including 24 openings lined with plain concrete and 9 cases with
reinforced concrete linings. Of the 33 cases, 7 were undamaged, 1 was slightly
damaged, 3 were moderately damaged, and 11 were heavily damaged.
It is interesting to note that, according to the statistical data shown in Figure 2.8F, the
proportion of damaged cases for the concrete and reinforced concrete lined tunnels appears to
be greater than that for the unlined cases. Sharma and Judd [1991] attributed this phenomenon
to the poor ground conditions that originally required the openings to be lined. Richardson
and Blejwas [1992] offered two other possible explanations:
-
Damage in the form of cracking or spalling is easier to identify in lined openings than
in unlined cases.
Lined openings are more likely to be classified as damaged because of their high cost
and importance
19
Figure 2.8 Damage Statistics, Sharma and Judd [1991]
2.3.5 Asakura and Sato [1998]

The authors provide an excellent compilation of past earthquake damage to Japanese tunnels
and also a description of damage due to the 1995 Hyogoken-Nanbu (Kobe) earthquake. There
are 24 damage tunnels out of 107 rock tunnels in the area, excluding cut-and-cover tunnels
and tunnels constructed by shield tunnelling. Twelve tunnels were reported as requiring repair
20
and 12 with minor damage not requiring substantial repair. Typical damage patterns were
cracking in the lining, spalling of concrete in the arch and the sidewalls, expansion of existing
cracks, heave and cracking of the invert, settlement of the arch crown, pounding of
construction joints, and collapse of portal. The closest tunnel, the Maiko tunnel under
construction, with an epicentral distance of 4 km received only slight damage while a fourstory building on the surface on top of the tunnel was completely destroyed. There was,
however, no definite regularity of damage relative to the epicentral distance for the 12 most
damaged tunnels. They were all within ~10 km from the presumed earthquake fault plane.
Tunneling method is not a predominant factor for tunnel performance either. A great part of
the damage was caused again where faults are crossing the tunnels.
In this paper, Asakura and Sato [1998] also proposed the damage by seismic force in the
tunnel cross-sectional direction and probable failure modes, Figure 2.9. If the tunnel cross
section is horizontally compressed, compressive failure at the arch crown and compressionshear failure at the arch shoulder may occur. On the other hand, if the tunnel cross section is
vertically compressed, spalling may occur at the arch-sidewall joint, which has a special
structure, for lining work convenience, to induce stress concentration. If horizontal shear acts
on the tunnel, longitudinal cracking around the arch shoulder is possible to occur, which was
observed in the Higashyama Tunnel.
Figure 2.9 Seismic forces and probable failure modes, Asakura and Sato [1998].
In 1995 Kobe earthquake, the tunnel damages can be classified as cracking and exfoliation of
lining concrete at portals, and at other places where the depth is shallow and where a fault
cross the tunnel. Record, where available, on damage of tunnel lining due to fault movement
show that damage took place within ~ 10 m from the faults.
From their conclusion, mountain tunnels may suffer some damage if the tunnel is located near
the epicenter of the earthquake fault, i.e. within 10 km for a magnitude 7 earthquake and 30
km for a magnitude 8 earthquake or, when the tunnel has special geological or construction
conditions, such as poor slope stability around tunnel portal, crossing existing faults or
fracture zones, poor lining with material and structural defects, or if collapse or water inflow
trouble occurred during construction.
2.3.6 American Lifeline Alliance (ALA) [2001]
From this report, a database of 217 bored tunnels that have experienced strong ground
21
motions in prior earthquake had been reviewed. It is composed of 204 entries based on work
by Power et al., [1998] and supplemented by case history data based on Asakura and Satio
[1998].
Table 2.1 Summary of Earthquakes and Lining/Support systems of the Bored tunnels, Power et al. [1998]
From this report, the defined damage states during difference levels of shaking had also been
provided. The four damage states are: DS=1 none; DS=2 slight; DS=3 moderate; and DS=4
heavy.
Table 2.2 Statistics for all bored tunnels, ALA [2001].
Using the above findings as a guide, judgments were made regarding median values of PGA
at ground surface at outcropping rock for the damage categories of slight, moderate and
heavy. Slight damage includes minor cracking and spalling and other minor distress to tunnel
liners. Moderate damage ranges from major cracking and spalling to rock falls. Heavy
damage includes collapse of the liner or surrounding soils to the extent that the tunnel is
blocked either immediately or within a few days after the main shock. These assessments are
made for tunnels in rock and tunnels in soil, in both poor-to-average construction and
conditions and in good construction and conditions, American Lifeline Alliance [2001].
Rock Tunnels with poor-to-average construction and conditions:
Tunnels in average or poor rock, either unsupported masonry or timber liners, or unreinforced
22
concrete with frequent voids behind lining and/or weak concrete.

Rock Tunnels with good construction and conditions:
Tunnels in very sound rock and designed for geologic conditions (e.g., special support such as
rock bolts or stronger liners in weak zones); unreinforced, strong concrete liners with contact
grouting to assure continuous contact with rock; average rock; or tunnels with reinforced
concrete or steel liners with contact grouting.
Alluvial (Soil) and Cut and Cover Tunnels with poor to average construction:
Tunnels that are bored or cut and cover box-type tunnels and include tunnels with masonry,
timber or unreinforced concrete liners, or any liner in poor contact with the soil. These also
include cut and cover box tunnels not designed for racking mode of deformation.
Alluvial (Soil) and Cut and Cover Tunnels with good construction:
Tunnels designed for seismic loading, including racking mode of deformation for cut and
cover box tunnels. These also include tunnels with reinforced strong concrete or steel liners in
bored tunnels in good contact with soil.
Table 2.3 Tunnel Fragility-Median PGAs-Ground shaking hazard only, ALA [2001].
The magnitudes of the median fragilities are about the same for tunnels of good quality
construction and somewhat lower for tunnels of lower quality construction. The heavy
damage state is provided only for tunnels with poor-to-average conditions. From the
observation of this database, no heavy damage has occurred to well-constructed tunnels in
good ground conditions.
2.4 Lesson learns from the past damages
The following general observation can be made regarding the seismic performance of
underground structures.
2.4.1
Geological settings
Mountain tunnels in rock and lined without material and structural defects are less affected by
an earthquake even if it is very large. Seismic waves propagate faster in hard and dense
materials, and thus less energy will be released at places where the tunnels lie in ground that
is harder than the tunnel structure, meaning that such tunnels will tend to deform with the
23
ground and suffer less damage. On the other hand, if the tunnels lie in relatively weaker
ground they will absorb larger amounts of energy and thus suffer greater damage. Concrete
linings can particularly be damaged easily by ground displacement or ground squeeze where
soft and hard grounds meet, as soft and hard grounds behave differently during earthquakes,
Hashash, et al. [2001].
2.4.2 Concrete lining
Okamoto [1973] reviewed damages to railway tunnels in 1923 Kanto earthquake. Based on
observed damages he concluded that there is a certain correlation between lining thickness
and damage, i.e. earthquake damage was greater in sections which thick lining than thin ones
as shown in the following:
Lining Thickness (cm)
57.1
45.7
34.3
22.9
Damage Rate
80 %
55 %
11 %
0%
Also, during the Kita-Minto earthquake, the rate of damage to waterway tunnels for
hydroelectric power generation is as high as 82 % for lining thickness of 40 cm, and only 16
% for thickness of 20 cm. The damage ratios are compared only be geological classification
without consideration of lining thickness, the rate is progressively reduced in the order of soil
or soil and gravel, rock with joints, soft rock, and hard rock as shown in following:
Type of soil
Hard rock
Soft rock
Rock with joints
Soil or Soil & Gravel
Damage Rate
16 %
40 %
44 %
61 %
The damage characteristics described above indicates the rate of damage is higher the poorer
the geology of the ground and also higher the thicker the lining. This shows the importance of
geological settings which can not be overcome by merely increasing the lining thickness,
Okamoto [1973].
Concrete linings can particularly be damaged easily by ground displacement or ground
squeeze where soft and hard grounds meet, as soft and hard grounds behave differently during
earthquakes. Any unfavorable events such as cave-in or collapse during tunneling would
extend the plastic zone around the tunnel, weaken the surrounding rock and cause excessive
vibration when seismic waves pass through. In addition, if the ground has previously
experienced vertical stress from loosening, plastic stress owing to squeezing, inclined stress or
any other weakening processes, tunnels in these areas will suffer greater damage to their
concrete linings during an earthquake [Wang et al., 2001].
Tunnels are more stable under a symmetric load, which improves ground-lining interaction.
Improving the tunnel lining by placing thicker and stiffer sections without stabilizing
surrounding poor ground may result in excess seismic forces in the lining. Backfilling with
non-cyclically mobile material and rock-stabilizing measures may improve the safety and
stability of shallow tunnels, Hashash, et al. [2001].
2.4.3 Distance effect
The intensity of seismic force experienced by each tunnel differs owing to their different
distances from the displaced fault zone and the direction from the epicenter of the earthquake.
24
Seismic waves propagate in the ground and lose energy because of dispersion and ground
resistance, causing tunnels to be under greater seismic forces if they are closer to the
displaced fault zone or the epicenter. The near-fault effect should be considered in this short
distance case.
2.4.4 Overburden depth
The distance to the ground surface also influences the seismic effect. When seismic waves
reach the ground surface, they release energy due to reflection or refraction, and thus tunnels
near the surface, and especially those near slope faces, will absorb a greater seismic energy.
High shaking intensity is also due to the lower stiffness of the soils and the site amplification
effect. The frequencies of damage reports decrease with depth, which is attributed to the lack
of surface waves, lower acceleration with depth, and increased strength of the rock with
depth. However, far more facilities exist at shallower depth a fact that may induce
considerable bias, [Sharma and Judd, 1991].
2.4.5
Slope stability
Damage at and near tunnel portals may be significant due to slope instability. The primary
failure mode tends to be slope failure. Particular caution must be taken if the portal also acts
as a retaining wall [St. John and Zahrah, 1987].
2.4.6 Duration of earthquake
Duration of strong-motion shaking during earthquakes is of utmost importance because it may
cause fatigue and therefore, large deformations, [Hashash, et al, 2001].
2.4.7 Frequency-content effect
High frequency motions may explain the local spalling of rock or concrete along planes of
weakness. These frequencies, which rapidly attenuate with distance, may be expected mainly
at small distances from the causative fault, [Hashash, et al, 2001].
2.4.8 Peak ground motion parameters
Damage may be related to peak ground velocity based on the magnitude and rupture distance
of the affected earthquake, [Hashash, et al, 2001].
2.4.9 Near-fault effect
For the near-fault zone, the directivity and fling step effects are obviously dominated the
tunnel responses. The underground structural design should also include the effect of high
intensity ground motion at intermediate to long period pulse.
25
Chapter 3. Seismic Design and Analysis Procedures for Underground Structures
3 SEISMIC DESIGN AND ANALYSIS PROCEDURES FOR

UNDERGROUND STRUCTURES
3.1
Seismic behaviors of underground structures

In general, seismic design loads for underground structures are characterized in terms
of the deformations and strains imposed on the structure by the surrounding ground, often due
to the interaction between the two. For most underground structures, the inertia of the
surrounding soil is large relative to the inertia of the structure. The response of a tunnel is
neither resonance in the ground nor dominated by the inertial properties of the tunnel structure
itself, but it conforms to the surrounding ground response, Kawashima [2000]. The author
also gives two major reasons for this phenomenon. First, the mass effect is generally small in
an underground structure, because the gross unit weight of an underground structure is
generally 10.5 11 kN/m3 while the unit weight of the surrounding soils is generally in the
range of 14 18 kN/m3. Second, the damping of an underground structure is very high due to
the radiation of energy from the structure to the surrounding ground. In contrast, surface
structures are designed for the inertial forces caused by ground accelerations.
The comparison between underground and on ground structure behaviors also proves the
previous assumptions. Figure. 3.1 shows the peak response acceleration of the on ground
structure is 0.77g and this is larger than the acceleration of 0.6g at the ground surface; this
difference is caused by the amplified effect of the on ground structural response. On the other
hand, the response acceleration of the underground structure is quite close in both frequency
content and peak value to the ground acceleration at the same depth as the underground
structure, Kawashima [2000].
26
Figure 3.1 Acceleration response of Underground and On-ground structures, [Kawashima, 2000]
The major factors influencing shaking damage include: [Dowding and Rozen, 1978; St. John
and Zahrah, 1987]
The shape, dimensions, and depth of the structure
The properties of the surrounding soil or rock
The properties of the structure
The severity of the ground shaking, e.g. duration, PGA, PGV, and frequency effects.
The general behavior of the linear tunnel is similar to that of an elastic beam subject to
deformation or strains imposed by the surrounding ground. Owen and Scholl [1981] express
the response of underground structures to seismic motions into three types:
(1) Axial extension and compression
27
Axial deformations in tunnels are generated by the components of seismic waves that produce
motions parallel to the axis of the tunnel and cause alternating compression and tension. The
design of a tunnel lining to accommodate axial deformations generally concentrates on the
direction along the tunnel axis, Wang [1993]. This type of deformation is slightly complex
since there will be some interaction between the structure and the ground. This interaction
would be more important if the ground is soft and shear stress transfer between the ground
and the structure is limited by the interface shear strength, Hashash et al. [2001].
Figure 3.2 Axial deformation along the tunnel, Wang [1993].
(2) Curvature deformation or longitudinal bending

Bending deformations are caused by the components of seismic waves producing particle
motions perpendicular to the longitudinal axis. The design and analysis of bending
deformations are also in the longitudinal direction along the tunnel axis, Wang [1993].
Figure 3.3 Bending deformation along the tunnel, Wang [1993].
(3) Ovaling deformations

Ovaling deformation in a tunnel structure develops when shear waves propagate normal or
nearly normal to the tunnel axis, resulting in a distortion of the cross-sectional shape of the
tunnel lining. Design considerations for this type of deformation are in the transverse
direction. The general behavior of the lining may be simulated as a buried structure subject to
ground deformations under a two-dimensional plane-strain condition, Hashash et al. [2001].
Ovaling deformations may be caused by vertically, horizontally or obliquely propagating
seismic waves of any type. Many previous studies have suggested, however, that the
vertically propagating shear wave is the predominant form of earthquake loading that governs
the tunnel lining design against ovaling. The following reasons are given, Wang [1993]:
28
Ground motion in the vertical direction is generally considered less severe than its
horizontal component. Typically, vertical ground motion parameters are assumed to be 1/2
to 2/3 of the horizontal ones. (Note that a vertically propagating shear wave causes the
ground to shake in the horizontal direction.) This relation is based on observation of
California earthquakes, which are most commonly of the strike-slip variety in which
horizontal motion predominates, Wang [1993].
For thrust faults, in which one rock block overrides another, vertical effects may equal or
exceed the horizontal ones. The effects of thrust faulting are usually more localized, however,
than those of the strike-slip faulting, and they are attenuated more rapidly with distance from
the focus, Wang [1993].
-
For tunnels embedded in soils or weak media, the horizontal motion associated with
vertically propagating shear waves tends to be amplified. In contrast, the ground strains
due to horizontally propagating waves are found to be strongly influenced by the ground
strains in the rock beneath. Generally, the resulting strains are smaller than those
calculated using the properties of the soils, Wang [1993].
Figure 3.4 Ovaling deformation of a circular cross section [Owen and Scholl, 1981]
3.2 Design and analysis methods

Design methods have been developed to estimate seismic loads on underground structures.
These methods include free-field deformation motion, which the ground deformation/strains
caused by the travelling seismic waves without the structure being present is imposed on the
underground structure, as well as dynamic soil-structure interaction analysis.
3.2.1 The free field deformation method
This simplest method, the quasi-static method without soil-structure interaction, assumes
conservatively that the structure is flexible enough to follow the deformations of the
surrounding grounds. Thus by using the maximum values of amplitude and wavelength of the
soil seismic deformation, the maximum structural strains can be determined. This approach
shows satisfactory results when low levels of shaking are anticipated or the underground
structure stiffness is similar to the surrounding ground.
29
However, the applicability of this approach is quite limited, since it may overestimate or
underestimate structure deformations depending on the rigidity of the structure relative to the
ground. In many cases, especially in soft soils, this method gives overly conservative designs
because of free-field ground distortions in soft soils are generally large. For example,
rectangular box structures in soft soils are typically designed with stiff configurations to resist
static loads and are therefore, less tolerant to racking distortions, Hwang and Lysmer, [1981].
On the other hand, when the tunnel structure is flexible relative to the surrounding ground,
this method may also underestimate the seismic responses of the structure. Soil-structure
interaction effects have to be included for the design of such structures Wang [1993].
The free-field deformation method, nevertheless, has been used on many significant projects,
including the San Francisco BART stations and tunnels, Kuesel [1969] and the Los Angeles
Metro, Monsees and Merritt [1991], while Sakurai and Takahashi [1969] apply it to
underground pipelines. The newly built Daikai station in Kobe was also designed in this
criterion. Kuesel found that, in most cases, if a structure can absorb free-field soil distortions
elastically, no special seismic provisions are necessary.
(1) Closed form elastic solutions
This simplified method assumes the seismic wave field to be that of plane waves with the
same amplitudes at all locations along the tunnel, differing only in their arrival time. Wave
scattering and complex three-dimensional wave propagation, which can lead to differences in
wave amplitudes along the tunnel are neglected, although ground motion incoherence tends to
increase the strains and stresses in the longitudinal direction. Then, results of analyses based
on plane wave assumptions should be interpreted with care, Power et al. [1996].
Newmark [1968] and Kuesel [1969] proposed a simplified method for calculating free-field
ground strains caused by a harmonic wave propagating at a given angle of incidence in a
homogeneous, isotropic, elastic medium (Figure 3.5). The most critical incidence angle
yielding maximum strain is typically used as a safety measure against the uncertainties of
earthquake prediction. Newmarks approach provides an order of magnitude estimate of
wave-induced strains while requiring a minimal input, making it useful as both an initial
design tool and a method of design verification [Wang 1993; Hashash et al. 2001].
30
Figure 3.5 Geometry of a sinusoidal shear wave oblique to axis of tunnel, Wang [1993]
St. John and Zahrah [1987] used Newmarkss approach to develop solutions for free-field
longitudinal, normal, and shear strains due to compression (P-wave), shear (S-wave) and
Rayleigh waves Table 3.1. However, it is generally considered that the P-wave induced
response would not control the design. It is also difficult to determine which type of wave will
dominate due to the complex nature of the characteristics associated with different wave
types. Generally, strains produced by Rayleigh waves may govern only when the site is at a
large distance from the earthquake source and the structure is built at shallow depth, Wang
[1993].
31

Strain and curvature due to body and surface waves, St. John and Zahrah [1987]
Wave type
Longitudinal strain
P-wave
l =
VP
cos 2
CP
lm =
S-wave
l =
Normal strain
n =
VP
for = 0o
CP
nm =
VS
sin cos
CS
lm =
VP
sin 2
CP
n =
VS
for = 45o
2CS
VP
for = 90o
CP
VS
sin cos
CS
nm =
VS
for = 45o
2CS
Shear strain
VP
sin cos
CP
m =
=
VP
for = 45o
2CP
VS
cos 2
CS
m =
VS
for = 0o
CS
Curvature
K=
aP
sin cos 2
2
CP
K m = 0.385
K=
aP
for = 35o16'
2
CP
aS
cos3
2
CS
Km =
aS
for = 0o
2
CS
Rayleigh wave
Compressional
l =
VRP
cos 2
CR
component
lm =
Shear
component
VRP
for = 0o
CR
n =
VRP
sin 2
CR
nm =
n =
VRP
for = 90o
CR
VRS
sin
CR
nm =
VRS
for = 90o
CR
VRP
sin cos
CR
m =
=
VRP
for = 45o
2CR
VRS
cos
CR
m =
VRS
for = 0o
CR
K=
aRP
sin cos 2
2
CR
K m = 0.385
K=
aRP
for = 35o16'
CR2
aRS
cos 2
2
CR
Km =
aRS
for = 0o
2
CR
Table 3.1 Strains and curvatures due to body and surface waves
32
Notation:
The Poisson's ratio and dynamic modulus of a soil deposit can be computed from measured Pwave and S-wave propagation velocities in an elastic medium:
Cp 2
2
1 Cs
The Poissons ratio =
; Youngs modulus E = C 2p (1 + )(1 2 ) ; and Shear
2 Cp 2
(1 )
C 1
s
modulus
G = Cs2 ,
respectively.
where:
r:
radius of circular tunnel or half height of a rectangular tunnel
aP:
peak particle acceleration associated with P wave
aS:
peak particle acceleration associated with S wave
aRP:
peak particle acceleration associated with Rayleigh wave, compressional component
aRS:
peak particle acceleration associated with Rayleigh wave, shear component
angle of incidence of wave with respect to tunnel axis
Vp:
peak particle velocity associated with P wave
Cp:
apparent velocity of P wave propagation
Vs:
peak particle velocity associated with S wave
Cs:
apparent velocity of S wave propagation
VRP:
peak particle velocity associated with Rayleigh Wave, compressional component
VRS:
peak particle velocity associated with Rayleigh Wave, shear component
CR:
apparent velocity of Rayleigh wave propagation
mass density
Combined axial and curvature deformations can be obtained by treating the tunnel as an
elastic beam. Using beam theory, total free-field axial strains, ab, are found by combining the
longitudinal strains generated by axial and bending deformations, Power et al. [1996]
V
a
ab = P cos 2 + r P2 sin cos 2 for P-waves
(3.1)
CP
CP
33
VS
ab =
CS
VR
sin cos + r
aS
3
cos
for S-waves
CS2
aR
2
sin
cos
for Rayleigh-waves (compressional component)
CR2
CR
Application of the strain equation in Table 3.1 requires knowledge of:
ab =
cos 2 + r
The apparent wave propagation velocity
The peak particle velocity
The peak particle acceleration
(3.2)
(3.3)
The peak particle velocity and acceleration can be established through empirical methods,
field measurements, or site-specific seismic exposure studies. The apparent wave propagation
velocity in rock can be determined with reasonable confidence from in-situ and laboratory
tests. Estimating the apparent wave propagation velocity in soil overburden presents the major
difficulty. Previous studies have shown that, except possibly for vertically propagating shear
waves, the use of soil properties in deriving the wave velocity in soil overburden may be
overly conservative. Free-field ground deformations and velocities due to a seismic event
should then be estimated using site specific response analysis that accounts for local geology.
The contribution of bending deformation to axial strain increases as the radius of the tunnel
increases. However, calculations using the equations shown in Table 3.1 indicate that the
bending component of strain is, in general, relatively small compared to axial strains for
tunnels under seismic loading. The cyclic nature of the axial strains should also be noted.
Although a tunnel lining may crack in tension, this cracking is usually transient due to the
cyclic nature of the incident waves. The reinforcing steel in the lining may close these cracks
at the end of the shaking, provided there is no permanent ground deformation and the steel has
not yielded, Hashash et at. [2001]. Even unreinforced concrete linings are considered
adequate as long as the cracks are small, uniformly distributed, and do not adversely affect the
performance of the lining, Wang [1993]. According to this method moment and forces
generated in tunnel lining are in the following form:
2
2
2
3
M =
cos E1 I1 D sin
L
L / cos
(3.4)
2
2
4
V =
cos E1 I1 D cos
L
L / cos
(3.5)
2
Q=
L
(3.6)
2
cos sin E1 A1 D cos
L / cos
where M, bending moment; V, shear force; Q, axial force; , angle of wave impact; I1,
moment of inertia of tunnel lining; E1, modulus of elasticity of lining material; D, amplitude
of sine wave; L, shear wavelength; and A1, section area of lining.
34
(2) Ovaling deformation of circular tunnels

Ground shear distortions can be defined in two ways, as shown in Figure 3.8: (1) Nonperforated ground; and (2) Perforated ground. Typically two-dimensional, plane strain
conditions are considered. In the non-perforated ground, the maximum diametric strain d is a
function only of the maximum free-field shear strain max:
d =
d
= max
d
2
(3.7)
where; d is the diameter of the opening. The diametric strain in a perforated ground is related
to the Poissons ratio of the medium:
d =
d
= 2 max (1 )
d
(a)
(a) Non-perforated medium
(3.8)
(b)
(b) Perforated medium
Figure 3.6 Free-field shear distortion of ground Wang [1993]
Both equations assume that there is no liner, therefore ignoring structure-ground interaction.
The perforated ground yields a much greater distortion than the non-perforated ground, by a
factor of two to three. Results with the perforated ground may provide a reasonable distortion
criterion for a soft lining, while the non-perforated results may be appropriate when the lining
stiffness is similar to that of the ground. A lining with large relative stiffness should
experience distortions much smaller than those given by Eq. (3.8) Perforated, Wang [1993].
Okamoto [1973] also developed the analytical solution using the theory of elastic waves to the
dynamics of a tunnel to investigate the effect of lining thickness. The stress around a hole
subjected to shear wave coming from one side was calculated. In this case shear waves and
longitudinal waves are reflected from the surface of the hole, and a stress concentration is
produced around the hole, while in the tunnel lining a reaction force to the reflection of
seismic waves acts as seismic force.
35
Figure 3.7 The interaction between elastic waves and tunnel lining
As indicated in Figure 3.7, the calculations were made for sinusoidal shear waves,
a exp {ip ( t + x / )} , propagating along the x axis. The result shows that the following seismic
force acts normal to the surface of the lining when the diameter of the hole is small in relation
to the wavelength of the earthquake motion:
rr = rr0 sin 2
(3.9)
.The maximum seismic load is given by the following equation:

3
h
6av0
r
rr0 =
3
g h
16
+

Ec r
(3.10)
where
a
is the amplitude of incident seismic wave;
rr
is the intensity of seismic load per unit width of lining acting perpendicular on lining
surface;
rr0
is the maximum value of rr;
is the radius of tunnel;
is the thickness of lining;
is the unit weight of soil;
is the propagation velocity of shear wave;
v0
is the velocity amplitude of seismic wave; and
36
Ec
is the Youngs modulus of lining concrete.
Bending moment is produced in the lining and the maximum fiber stress is given by the
following equation:
h
r
=
g h
16
+

Ec r
12av0
(3.11)
According to this equation, both rr and are proportional to v0. Therefore, it is the peak
ground velocity of the earthquake motion which affects the stress of the lining.
m kg / cm 2 , rr t / m 2
Figure 3.8 Relationship between stress and thickness of tunnel lining, Okamoto[1973]
Form previous equations, the seismic stresses produced in the lining depending on the lining
thickness. Figure 3.8 shows the seismic forces acting on the lining and the seismic stresses
produced thereby for cases of various earthquake wave propagation velocities. According to
this, the seismic load acting on the lining and the seismic stress increase with increased lining
thickness. In general, increased thickness tends to have a detrimental effect on the lining.
Some general conclusion can be given as:
(a)
The seismic load acting on the lining was increased and the seismic stress also became
greater the slower the seismic wave propagation velocity
(b)
With the exception of very slow seismic wave propagation velocity, the seismic load
and seismic stress were increased with increased lining thickness.
(c)
However, when the propagation velocity of seismic waves was extremely slow in a
range above a certain lining thickness, increase in the lining thickness reduced seismic
stress. This reduction in seismic stress was not, however, very great considering the
increase in lining thickness.
3.2.2 The soil-structure interaction method

A pseudo-static analysis method is commonly used to account for soil-structure interaction
37
effects. However, the pseudo static approach may be valid for weak ground motion where
deformations are elastic.
In this approach it is assumed that the presence of an underground structure modifies the
deformations of the ground around the structure, and thus the free-field ground deformations
are not experienced by the structure [Wang, 1993; Penzien and Wu, 1998; Penzien, 2000; and
Hashash et al. 2001]. The ground and the underground structure have a complex interaction
during seismic motions. In general, when an underground structure is stiffer than its
surrounding soils, the structure resists, rather than conforms to the deformations imposed by
the ground. On the other hand, when an underground structure is more flexible than its
surrounding ground, the deformations of the ground surrounding the structure may be larger
than those of the free-field ground. Relative stiffness (stiffness ratio) between an underground
structure and the surrounding ground plays a key role in the response of the structure.
(1) Elastic solutions of longitudinal deformations for circular tunnels
The beam-on-elastic foundation approach is used to model pseudo-static ground structure
interaction effects. The solutions ignore dynamic inertial interaction effects. Under seismic
loading, the cross-section of a tunnel will experience axial bending and shear strains due to
free field axial, curvature, and shear deformations (Figure 3.9). St. John and Zahrah [1987]
proposed that the maximum structural strains are caused by a wave at an incident angle of 45.
The maximum axial strain amax used by a 45 incident shear wave (Figure 3.7) is:
a
max
A
fL
L
2
4 El Ac
E A 2
2+ l c
Ka L
(3.12)
where
L
Wavelength of an ideal sinusoidal shear wave
Ka
Longitudinal spring coefficient of ground medium (see Eq. 11); in units of

force per unit deformation per unit length
Free-field ground displacement response amplitude for an ideal sinusoidal

shear wave
Ac
Cross-sectional area of tunnel lining
El
Elastic modulus of the tunnel lining
Maximum friction force per unit length between tunnel and surrounding soil
Eq. 3.7 states that the maximum axial strain shall be smaller than the one induced by the
38
maximum shear force that can be developed between tunnel and surrounding soil. The
maximum friction shear is dependent on the roughness of the ground-tunnel interface and the
normal force applied to the tunnel from the ground.
Figure 3.9 Induced forces and moments caused by waves propagating along tunnel axis
The maximum bending strain occurs when the incident angle of shear wave is equal to 0
(Figure 3.5):
2
b
max
A
L
r
=
4
El I c 2
1+
Kl L
(3.13)
Ic: moment of inertia of the tunnel section

Kl: transverse spring coefficient of the ground medium (see Eq. 3.18); in force per unit
deformation per unit length of tunnel
r: radius of circular tunnel or half height of a rectangular tunnel
The maximum shear force on the tunnel cross-section can be written as a function of this
maximum bending strain:
3
Vmax
L
=
EI
1+ l c
Kl
El I c A
b
2 El I c max
=
4
r
L
2
(3.14)
The maximum bending moment is:
39
M max
=
EI
1+ l c
Kl
El I c A
2
(3.15)
and the maximum axial force is (from Eq. 3.12):
Qmax
El I c A
L
=
2
E I 2
2+ l c
Ka L
(3.16)
A conservative estimate of the total maximum axial strain is obtained by combining the axial
and bending strains since both the liner and the ground are assumed as linear elastic, Power et
al., [1996]:
ab
a
b
max
= max
+ max
(3.17)
Note that in the above equations, the interaction between ground and structure is modeled by
including springs with the spring coefficients Ka and Kl for longitudinal and transverse soil
response. St. John and Zahrah [1987] suggested that the spring coefficients Ka and Kl are
functions of the incident wavelength:
K a = Kl =
16 G (1 ) d
( 3 4 )
(3.18)
where G is the shear modulus and is the Poissons ratio of the ground; d is the diameter of
circular tunnels or the height of rectangular tunnels; and L is the incident wavelength.
Wang [1993] indicated that the springs differ from those of a conventional beam analysis on
an elastic foundation. Not only must the coefficients be representative of the dynamic
modulus of the ground, but the derivation of these constants must consider the fact that the
seismic loading is alternately positive and negative due to the sinusoidal wave.
The input wavelengths for underground structure design have been investigated by a number
of researchers. The incident wavelength of a ground motion may be estimated as:
L = T Cs
(3.19)
where T is the natural period of a shear wave in the soil deposits and Cs is the shear wave
velocity. Idriss and Seed [1968] recommended that:
T=
4h
Cs
(3.20)
40
where h is the thickness of the soil deposit.

The ground displacement response amplitude A, in Eqs (3.12) through (3.16), represents the
spatial variation of ground motions along a horizontal alignment and should be derived by
site-specific subsurface conditions. The displacement amplitude generally increases with
increasing wavelength. In case of a sinusoidal shear wave with a displacement amplitude A
and a wavelength L, the displacement amplitude A can be calculated from the following
equations:
For free-field axial strains:
2 A Vs
=
sin cos
L
Cs
(3.21)
4 2 A as
=
cos3
2
Cs
L
(3.22)
For free-field bending strains:
For a sinusoidal compression wave with a displacement amplitude A and a wavelength L, the
displacement amplitude A can be calculated from the following equations:
For free-field axial strains:
2 A V p
cos 2
=
L
Cp
(3.23)
4 2 A a p
=
sin cos 2
2
Cp
L
(3.24)
For free-field bending strains:
(2) Ovaling deformations of circular tunnels

Peck et al. [1972] characterized tunnel liners into two groups: flexible liners and rigid liners.
A liner is said to be "flexible" if it interacts with the ground in a way that the pressure
distribution on the liner and the corresponding deflected shape result in negligible bending
moments in the lining. A "rigid" liner is a liner which deflects insignificantly under the loads
imposed by the ground; and thus it has to support large bending moments.
Actual tunnel linings are neither perfectly flexible nor perfectly rigid. A tunnel that may be
rigid in a soft ground may behave as a flexible liner in a very stiff ground. In order to
quantitatively describe the relative stiffness between tunnels and ground, based on earlier
work by Burns and Richard [1964] and Hoeg [1968], Peck et al. [1972] proposed that the
relative stiffness of a tunnel-ground system can be divided into two separate and distinct
types. The first type is extensional stiffness, which is a measure of the equal uniform pressure
41
to cause a unit diametric strain of the tunnel without changing the shape of the tunnel. The
second type is flexural stiffness, which is a measure of the non-uniform pressure to cause a
unit diametric strain resulting in a change in shape or an ovaling of the tunnel. The relative
stiffness between the tunnel and surrounding ground is hence characterized by two
dimensionless ratios designated as the compressibility ratio and flexibility ratio (C and F).
The compressibility ratio, a measure of the extensional stiffness of the ground to that of the
liner, is obtained by considering an infinite, elastic, homogeneous and isotropic ground
subjected to a uniform external pressure. The compressibility ratio is equal to the ratio
between the pressure required to cause a unit diametric strain (contraction) of the free-field
ground and the pressure required to cause a unit diametric strain (contraction) of the liner.
Note that in order to obtain the diametric strain of the free-field ground, a circle with its size
identical to the liner is assumed. The compressibility ratio can be expressed as:
C=
E 1 l2 r
El t (1 + )(1 2 )
(3.25)
where E is the Young's modulus of the ground; v is the Poisson's ratio of the ground; I is the
moment of inertia of the tunnel lining per unit width; r is the radius of the tunnel; and t is the
thickness of the tunnel lining.
The flexibility ratio, a measure of the flexural stiffness of the ground to that of the liner, is
obtained by considering an infinite, elastic, homogeneous and isotropic ground subjected to a
pure shear loading. The flexibility ratio is equal to the ratio between the shear stress required
to cause a unit diametric strain (ovaling) of the free-field ground and the shear stress required
to cause a unit diametric strain (ovaling) of the liner. Note that in order to obtain the diametric
strain of the free-field ground, a circle with size identical to the liner is assumed. The
flexibility ratio is:
F=
E 1 l2 r 3
6 El I (1 + )
(3.26)
It is often suggested that the flexibility ratio is the most important because it is related to the
ability of the lining to resist distortion imposed by the ground. Burns and Richard [1964] have
shown that the forces and deformations of ground and structure depend on (1) the
compressibility ratio, C; (2) the flexibility ratio, F; and (3) the slippage at the interface
between the ground and the liner. The interface between ground and support has often been
assumed to be frictional, i.e. the shear stress and normal stress developed at the interface
follow the Coulomb friction law. In other words, the maximum shear stress at the interface is
equal to the normal stress times the friction coefficient between ground and support. Two
extreme cases are considered: full-slip and no-slip.
The full-slip case assumes that the friction coefficient is zero and no shear force develops at
the interface. The ground may detach from the tunnel during an earthquake. In the no-slip
case, the friction coefficient is such that the ground and structure are tied together. The ground
42
and structure cannot be separated. Actual conditions may be in between these two extreme
cases. However, due to the complexity of the problem, most of the work has focused on either
full-slip or no-slip interface conditions.
- Full-slip conditions
Peck et a1. [1972] provided closed-form solutions for diametric strain d, thrust T, and
bending moment M, for the full-slip case under static loading conditions, i.e. tunnels under
overburden and lateral earth pressures. The displacements and forces are functions of the
compressibility ratio C, flexibility ratio F and the in-situ overburden pressure of the soil , H.
At the crown and invert of a circular tunnel:
d 1 l Hr
2 1
=
(1 )(1 + K0 ) bl C +
(1 K0 ) b2 F
2 MC
3 1 2
d
d =
1
1
T = t Hr (1 + K 0 ) b1 (1 K 0 ) b2
2
3
d 1 t Hr
2 1
=
(1 )(1 + K0 ) b1C +
(1 K0 ) b2 F
2 MC
3 1 2
d
d =
1
M = t Hr 2 (1 K 0 ) b2
6
(3.27)
(3.28)
(3.29)
(3.30)
At the springline of a circular tunnel;
d =
d 1 t Hr
2 1
=
(1 )(1 + K0 ) bl C
(1 K0 ) b2 F
2 MC
3 1 2
d
(3.31)
1
1
T = t Hr (1 + K 0 ) b1 + (1 K 0 ) b2
2
3
(3.32)
1
M = t Hr 2 (1 K 0 ) b2
6
(3.33)
where
t = Total unit weight of the soil;
Ko= Lateral earth pressure coefficient;
H = Burial depth of the tunnel, measured from the free surface to the center of the tunnel
Mc = Constrained modulus of the soil which is given by:
43
Mc =
E (1 )
(1 + )(1 2 )
(3.34)
E = Youngs modulus of the ground;

= Poissons ratio of the ground;
r = radius of the tunnel;
(1 2 )( C 1)
(1 2 ) C + 1
(3.35)
2 F + 1 2
2F 1
4
2 F + 5 6
2 F + 5 6
(3.36)
b1 = 1
and
b2 = 1 + 3
The solution can be used to obtain deformations and forces due to a shear wave; this can be
done by using Ko = -1, which replaces the far field normal stresses , and h = Ko by a far
field shear stress , Wang [1993]. After some mathematical manipulations, the diametric
strain d, maximum thrust Tmax, and bending moment Mmax can be presented in the following
forms (see Figure 3.10):
Figure 3.10 Induced circumferential forces and moments
caused by waves propagating perpendicular to tunnel axis
d =
d
1
= K1F max
d
3
(3.37)
44
1
E
Tmax = K1
r max
6 (1 + )
(3.38)
1
E
M max = K1
r 2 max
6 (1 + )
(3.39)
where
K1 =
12 (1 )
2 F + 5 6
(3.40)
E and are the Youngs modulus and Poissons ratio of the ground, respectively; r is the
radius of the tunnel; max is the maximum free-field ground shear strain and and F is the
flexibility ratio defined in Eq. 3.26. Kl is the full-slip lining response coefficient and is
determined by Eq. 3.39. The relationship between K1 and F is shown in Figure 3.11, Wang
[1993].
Figure 3.11 Lining response coefficients vs. flexibility ratio, full-slip interface, and circular tunnel, Wang
[1993]
45
dlining
d free field
2
K1F
3
(3.41)
The normalized lining deformation provides an indication of the importance of the flexibility
ratio in lining response, Wang [1993]: According to this equation and Figure 3.12, a tunnel
lining will deform less than the free field when the flexibility ratio is less than one, i.e. a stiff
lining in soft soil. As the flexibility ratio increases, the lining deflects more than the free field
and may reach an upper limit equal to the deformations of an opening without support.
Figure 3.12 Normalized lining deflection vs. flexibility ratio, full slip interface, and circular lining, Wang
[1993]
- No-slip conditions
Slip at the interface is only possible for tunnels in soft soils or cases of severe seismic
intensity, Hashash et al.[2001]. Full-slip assumptions under simple shear may significantly
underestimate the maximum thrust. Hoeg [1968] supported this conclusion and recommend
the no-slip assumption is made in assessing the lining response. For no-slip:
46
Tmax = K 2 r max = K 2
E
r max
2 (1 + )
(3.42)
where
1
(1 2 )2 + 2
2
K2 = 1 +
5
F ( 3 2 ) + (1 2 ) C + C 8 + 6 2 + 6 8
2
F (1 2 ) (1 2 ) C
(3.43)
K2 is defined as the no-slip lining response coefficient. Expressions for deformations and
maximum moment were not provided by the author. The relationship between K2 and C is
shown in Figure 3.13, Wang [1993]. The figure shows that seismically-induced thrust
increases with decreasing compressibility and flexibility ratios when the Poisson's ratio of the
surrounding ground is less than 0.5. As the Poisson's ratio approaches 0.5, the thrust response
is independent of compressibility because the soil is considered incompressible, Wang [1993].
47
Figure 3.13 Lining (thrust) response coefficient vs. compressibility ratio, no-slip interface, and circular
tunnel, Wang [1993]
Figure 3.13 (Continued) Lining (thrust) response coefficient vs. compressibility ratio, no-slip interface,
and circular tunnel, Wang [1993]
Corigliano et al. [2007] also developed a closed form solution for the pseudo-static analysis of
the transversal response. The relationships for the thrust force Q and bending moment M per
unit length of tunnel lining for seismic design associated with the no-slip condition between
the lining and the surrounding ground are given by the following relations:
Q=

E

max R 1 cos 2 +
2 (1 + )
4
3

(3.44)
48
M=

1
E

max R 2 1 + + cos 2 +
2 2 (1 + )
4
3
(3.45)
where the parameters , , , C*, F* are defined by:
{
{
} {
}
3C * / F * C * (1 ) 2 3 C * (1 ) + 2 + C * (1 ) 2 + C * (1 ) + 2 1 + C * (1 )
=
*
C * (1 ) C * (1 ) + 4 2 C * (1 ) + 2 + 3C * + C * (1 ) + 4 2 C * (1 ) + 2
F
(3.46)
C * (1 ) 2 C * (1 ) + 4

=
*
C (1 ) + 2
(3.47)
= C * (1 ) + 4 6 C * (1 ) + 2 (1 )
(3.48)
C =
( )
E A (1 )
(3.49)
( )
E I (1 )
(3.50)
ER 1 l2
l
F =
ER 3 1 l2
l l
where R is the average tunnel radius, A, and Il are area and moment of inertia per unit length
of the lining respectively, E, El, , and l are the Youngs modulus and Poissons ratio of
ground and lining respectively. The parameter C* and F* are the compressibility and
flexibility ratios. They represent a measure of the relative stiffness of the ground with respect
to the supporting system (i.e. the lining) under a symmetric and antisymmetric loading
respectively, Einstein and Schwartz [1979]. Finally max is the maximum shear strain (in
absolute value) calculated in free-field conditions. Knowing max, the imposed stress can be
easily computed as follows:
= =
E
max
2 (1 + )
(3.51)
The maximum shear strain is then the key parameter to determine the stress, thrust force, and
moment in the tunnel lining.
3.2.3 Numerical methods
The analysis of underground structures is complicated due to their interaction with the
surrounding soil, especially under dynamic conditions, cut-and-cover structures, mined
tunnels with non-circular shapes, and non-uniform properties of circular linings that preclude
the use of simple close-form solutions. As opposed to close form analytical solutions,
numerical methods can be used for analysis and design of complex structures. The numerical
49
methods of analysis fall under one of the following categories:

(a)
The lumped mass / stiffness methods
In the lumped mass method, the tunnel is divided into a number of segments with masses /
stiffness, which are connected by springs representing the axial, shear, and bending stiffness
of the tunnel. The soil reactions are represented by horizontal, vertical, and axial springs,
Hashash et al. [1998], Figure 3.17.
Figure 3.14 Simplified three-dimensional model for analysis of the global response of an immersed tube
tunnel, Hashash et al. [1998].
Free-field displacement time histories are first computed at selected locations along the tunnel
length. The computed free-field displacement time histories are then applied at the ends of the
springs representing the soil-tunnel interaction. If a dynamic, time-history analysis is desired
appropriate damping factors have to be incorporated into the springs and the structure.
However, this method has some drawbacks since it accounts for soil-structure interaction in
an approximate manner through springs and dashpots, whose characteristics are usually
difficult to determine and primarily used for preliminary design.
(b)
The Finite Difference Method (FDM)
The FDM analysis involves a discretization of the governing equations of motion for the
soil/structure system. The discretization is based on replacing the continuous derivatives in
the governing equations by the ratio of changes in the variables over a small, but finite
increment. The differential equations are, thus, transformed into difference equations. The
method of solution of these equations for transient analysis can be based on either an implicit
or an explicit scheme. The implicit scheme requires the solution of a set of simultaneous
equations and large storage may be needed. Explicit schemes are relatively straightforward
and may require less effort than implicit schemes. For certain types of problems, it is possible
to obtain unconditionally stable implicit schemes. The choice of the best solution scheme
depends on the particular application. These methods, FDM and FEM, are used both for
modeling the structure and the soil in two- or three-dimensional problems of underground
structures. For analyzing axial and bending deformations, it is most appropriate to utilize
three-dimensional models.
However, this method can be difficult to apply when nonhomgeneity and nonlinearities exist;
however, this difficulty can be overcome using the so called integrated finite difference
techniques. Another situation common in wave propagation problems involves infinite media.
Accordingly, there is a need to crate appropriate boundary conditions that will simulate the
50
physical behavior of the actual problem. The most popular approach is the use of viscous
dashpots to eliminate boundary reflections.
(c)
The Finite Element Method (FEM)
For FEM, the continuum is discretized into an equivalent system of smaller continua, which
are called finite elements. Each element is assigned constitutive or material properties and its
equations of state are formulated. Subsequently the elements are assembled to obtain
equations for the total structure. As in the case of the finite difference method, the solution
scheme can be based on either an implicit or an explicit formulation. In either case, a finite
difference approximation is used to represent the time dimension. The main advantage of the
finite element method is that arbitrary boundaries and material inhomogeneity can be
accommodated easily. As in the finite difference method, energy absorbing boundaries are
used to approximate the wave propagation in an infinite medium.
In comparison between FEM and simplified tunnel models, Gomez-Masso and Attalla [1984]
performed an extensive study comparing these two different methods, and found that, with
few exceptions, simplified methods tend to be very conservative. One reason for this finding
is that the simplified methods they used fail to consider structure-to-structure interaction
effects through the soil, which are important in this case.
Also, the nonlinear analyses of the Los Angeles Metro system, Sweet [1997] displayed
structural racking greater than the free-field, though previous linear analyses showed smaller
racking. This supports the assertion that both the nonlinear structural behavior and the
frequency content of the free-field environment contribute to the structural-racking behavior.
These elastic assumptions limit their applicability given the significant nonlinearity associated
with the soil behavior and associated near field ground motion, which generally contain high
velocity intensity.
(d)
The Boundary Element Method (BEM)
The boundary element method involves numerical solution of a set of integral equations that
relate the boundary or surface tractions to the boundary displacements. The method is based
on solution of integral rather than differential equations. It requires the discretization of only
the surface of the body into a number of segments or elements. The solution is first obtained
at the boundary, and then the solution at points within the medium based on the solution at the
boundaries. The method is mostly used for the analysis of linear, static problems. It has not
been widely utilized to handle material nonlinearities and non-homogeneities, St. John and
Zahrah [1987].
3.2.4 Conclusion
Since the deformation of underground structures, both longitudinal and transversal, is mainly
caused by the longitudinal and shear strains respectively in terms of the whole cross section,
the accuracy in determination of concrete lining stresses is a key parameter, Equation (3.44)
and (3.45) in case of quasi-static and no-slip condition. The application of ANNs to define the
maximum shear strain around the interested fault would be studied in this work.
51
Chapter 4. Artificial Neural Networks
4 ARTIFICIAL NEURAL NETWORKS

4.1
Introduction to Artificial Neural Networks
4.1.1
Artificial Neural Networks (ANNs)
ANNs is an information-processing system that has certain performance characteristics in

common with biological neural networks. Artificial neural networks have been developed as
generalizations of mathematical models of human cognition or neural biology, based on the
assumptions that:
1) Information processing occurs at many simple elements called neurons.
2) Signals are passed between neurons over connection links.
3) Each connection link has an associated weight, which, in atypical neural net, multiplies
the signal transmitted.
4) Each neuron applies an activation function (usually non-linear) to its net input (sum of
weighted input signals) to determine its output signal.
A neural network is characterized by the pattern of connections between the neurons (called
its architecture), the method of determining the weights on the connections (called its training
or learning, algorithm) and the activation function. A neural net consists of a large number of
simple processing elements called neurons, units, cells or nodes.
Each neuron is connected to other neurons by means of directed communication links, each
with an associated weight. The weights represent information being used by the net to solve a
problem. Each neuron has an internal state, called its activation or activity level, which is a
function of the inputs it has received. Typically, a neuron sends its activation as a signal to
several other neurons. It is important to note that a neuron can send only one signal at a time,
although that signal is broadcast to several other neurons.
4.1.2 Biological neural networks
A biological neuron has three types of components that are of particular interest in
understanding an artificial neuron: its dendrites, soma and axon. The many dendrites receive
52
signals from other neurons. The signals are electric impulses that are transmitted across a
synaptic gap by means of a chemical process. The action of the chemical transmitter modifies
the incoming signal (typical, by scaling the frequency of the signals that are received) in a
manner similar to action of the weights in an artificial neural network.
The soma, or cell membrane, sums the incoming signals. When sufficient input is received,
the cell fires; that is, it transmits a signal over its axon to other cells. It is often supposed that a
cell either fires or does not at any instant of time, so that transmitted signals can be treated as
binary. However, the frequency of firing varies and can be viewed as a signal of either greater
or lesser magnitude. This corresponds to looking at discrete time steps and summing all
activity (signals received or signals sent) at a particular point in time.
Second, we are able to tolerate damage to the neural system itself. Humans are born with as
100 billion neurons. Most of these are in the brain, and most are not replaced when they die.
transmitted signals can be treated as binary. However, the frequency of firing varies and can
be viewed as a signal of either greater or lesser magnitude. This corresponds to looking at
discrete time steps and summing all activity (signals received or signals sent) at a particular
point in time.
A generic biological neuron is illustrated in Figure. 4.1, together with axons from other
neurons (from which the illustrated neuron could receive signals) and dendrites for other
neurons. Another important characteristic that artificial neural networks share with biological
neural systems is fault tolerance. Biological neural systems are fault tolerant in two respects.
First, we are able to recognize many input signals that are somewhat different from any signal
we have seen before. An example of this is our ability to recognize a person in a picture we
have not seen before or to recognize a person after a long period of time. Second, we are able
to tolerate damage to the neural system itself. Humans are born with as 100 billion neurons.
Most of these are in the brain, and most are not replaced when they die.
Figure 4.1 Biological neuron
4.2 Neural Network Architectures

Each unit can be combined into a network in numerous fashions. The most common of these
is the multilayer perceptron (MLP) network. The basic MLP network architecture is
arrangement of neurons into layers of input units and the connection patterns weight between
layers, which are often classified as:
53
4.2.1 Single layer neural network

A single layer network has only one layer of connection weight. The input units receive
signals from the outside world, multiplies an associated weight, sum its weight input signal,
apply activation function to output unit. The weight for one output unit does not influence the
weights for other output unit.
Typically, the input units of single layer net are fully connected to output and these output
units are not connected to the other output unit. The weight for one output unit does not
influence the weights for other output unit.
Figure 4.2 Single layer network
This neural network is very simple; the presence of a hidden unit together with a non-linear
activation function gives it the ability to many more complicated problems. That is discussed
in Multi-Layer network below.
4.2.2 Multiple layers neural network
A multi-layer network is a network with one or more layers of hidden units between the input
units and the output units. There is a layer of weights between hidden layers, too. As shown in
Figure. 4.3, The output layer is a layer whose output is the network output. The other layers
are hidden layers.
54
Figure 4.3 Multiple layer network
Multi-layer networks are more powerful than single layer networks. For instance, multi-layer
network can be trained to approximate most functions arbitrarily well. Single-layer network
cannot do this. The most practical neural networks have just two or three layers.
4.2.3 Other neural network architectures
The previous describe network architecture is just one out of many neural networks, but it is
the only one of covered here. Several textbooks are available that offer introductions to the
most common network types. Most of these books also provide the historical background for
neural networks and how they relate to biological neural networks. See for example Hertz et
al. [1991], Haykin [1998], and Zurada [1992]. Not all networks are equally suitable for
modeling and control of dynamic systems. For these applications the most common
alternative to the multilayer percepteron (MLP) network is probably the Radial Basis
Function (RBF) networks [Sanner and Slotine 1992, Tzirkel-Hancock and Fallside 1992].
Studying and comparing the performance of different network types is however beyond the
scope of this study.
4.3 Activation Function
The activation function is the same for all neurons in any particular layer of a neural net. In
most cases, a non-linear activation function is chosen to satisfy some specification of the
problem that the neuron is attempting to solve. For the results of feeding a signal through two
or more layers of linear processing element.
In identification-related problem, the useful activation function is sigmoid function (S-shape
curves). The logistic function and the hyperbolic tangent functions are the most common,
which are Binary sigmoid function and Bipolar sigmoid function.
55
4.3.1 Binary sigmoid function

The binary sigmoid function is shown in Figure. 4.4. This activation function takes input,
which may have any value between plus and minus infinity and squashes the output into the
range 0 to 1.
Figure 4.4 Binary Sigmoid Function
The binary sigmoid function is the logistic function, which commonly used in multi-layer
networks that are trained using the backpropagation algorithm, in part because this function is
differentiable.
4.3.2 Bipolar sigmoid function
The bipolar sigmoid is closely related to the hyperbolic tangent function. The application of
this function is similar as the binary sigmoid function; this function is illustrated in Figure.
4.5. The input can be any value between plus and minus infinity. But the desire ranges of
output values are between 1 and 1, (-1,1). The bipolar sigmoid output is wider than output
from binary sigmoid function (0,1).
Figure 4.5 Bipolar Sigmoid Function
4.4
Training Algorithm
4.4.1 Learning rule

Learning rule means a procedure for modifying the weights and biases of the network. The
purpose of the learning rule is to train to perform some task. There are many types of neural
network learning rules. They fall into two broad categories: supervised learning and
unsupervised learning.
In supervised learning, the learning rule is provided with a set of examples (the training set) of
proper network behavior:
56
(y1, o1), (y2, o2), , (yn, on)

where yn is input to the network and on is the corresponding correct (target) output. As the
inputs are applied to the network, the network outputs are comparing to the targets. The
learning rule is then used to adjust the weights and biases of the network in order to move the
network outputs closer to the targets.
In unsupervised learning, the weights and biases are modified in response to network inputs
only. There are no target outputs available. Most of these algorithms perform some kind of
clustering operation. They learn to categorize the input patterns into a finite number of
classes. This is especially useful in such applications as vector quantization.
4.4.2 Generalized delta rule
In the Widrow-Hoff rule, or delta rule, the amount of learning is represented as the difference
(or delta) between the desired and computed outputs. Purposed by Rumelhart, Hinton, and
Williams [1986], backpropagation (BP) is an error-correcting learning procedure that
generalized the delta rule to multi-layer feedforward neural networks with hidden units
between the input and output units. The goal of the learning procedure is to update the
weights of the links connecting the nodes, and to minimize the average squared system error
between the desired and the computed outputs. The error term from one output neuron is
defined as:
E=
1
( t o )2
2
(4.1)
where t is the desired output, o is the network output.

We adjust the weights and threshold value in proportion to
E
or in order to reduce the
w j
system error:
w j =
E
w j
(4.2)
where is called the learning rate, which determines what amount of the calculated error
sensitivity to, weight change will be used for the weight correction.
4.5 The Backpropagation Network
The backpropagation network (BPN) is currently the most general-purpose and commonly
used neural- network paradigm. The BPN achieves its generality because of the gradientdescent technique used to train the network.
Gradient descent is analogous to an error- minimization process. Error minimization, as the
term implies, is an attempt to fit a closed-form solution to a set of empirical data points, such
that the solution deviates from the exact value by a minimal amount. Figure 4.6 illustrates the
error-minimization concept.
57
Figure 4.6 The diagram illustrates the process of minimizing the error of a function through the set of
empirical data
The learning process begins with the presentation of an input pattern to the BPN. That input
pattern is propagated through the entire network, until an output pattern is produced. The BPN
then makes use of what is called the generalized delta rule to determine the error for the
current pattern contributed by slightly in a direction that reduces its error signal, and the
process is repeated for the next pattern.
In this project, the backpropagation neural network is adapted for the identification of nonlinear. The principles of the backpropagation neural network are introduced in the following.
Figure 4.7 the typical two hidden layers backpropagation neural network: the input layer with
n nodes, hidden layer 1, 2 with q, p nodes respectively and the output layer with 1 node. There
are weights uih, vhj, wj layers, represent the strength of connection of the nodes in the network.
58
Figure 4.7 Typical two hidden layers backpropagation neural networks
The first type of operation of bakpropagation neural network is called feed forward and it
shown as solid lines with arrows. Starting from an input-output pair, each input neuron
receives an input signal and broadcasts this signal to the connected neurons Z1 ,, Zq in the
first hidden layer. The total input to the Zh neuron from the input layer is
n
z _ inh = u0 h + yi uih
(4.3)
i =1
Each of these neurons then computes its activation
zh = f ( z _ inh )
(4.4)
and sends its result to the connected neurons K1, , Kp in the second hidden layer. The total
input to the Kj neuron from the first hidden layer is
q
k _ in j = v0 j + zh vhj
(4.5)
h =1
Next, each neuron in the second hidden layer computes its activation
k j = f k _ in j
(4.6)
and sends its result to the output neuron. The total input to the output neuron O from the
second hidden layer is
p
o _ in = w0 + k j w j
(4.7)
j =1
Finally, the output neuron yields the network output according to
o = f ( o _ in )
(4.8)
59
In the equations above uih, vhj, and wj are the connection weights between the layers, whereas
f() is the activation function, Haykin [1999]. For example, a bipolar sigmoid activation
function, Figure 4.5, is defined as
f ( x) =
2
1
1 + e x
(4.9)
During training, the network output o is compared with its target or sample output t to
determine the error e associated with the output neuron.
e = (t o)
(4.10)
The update of the connection weights aims at minimizing this discrepancy. Consequently, an
objective function to be minimized is defined as
E = 0.5 ( t o )
(4.11)
Using the generalized delta rule, the update of the weights connecting the second hidden layer
with the output layer is given by
w j ,n = w j ,o + w j
w j =
E
w j
(4.12)
(4.13)
= ( t o ) f ' ( o _ in ) k j
(4.14)
k j
(4.15)
where = learning rate; wj,n = new weight, wj,o = old weight.

For the weights connecting the first hidden layer to the second hidden layer, the update is
done in the same way
vhj ,n = vhj ,o + vhj
(4.16)
E
vhj
(4.17)
= f ' ( k _ in j ) w j zh
(4.18)
j zh
(4.19)
vhj =
Finally, the update for the weights connecting the input layer to the first hidden layer is
uih ,n = uih ,o + uih
(4.20)
60
uih =
E
uih
(4.21)
p
= f ' ( z _ inh ) j vhj yi
(4.22)
h yi
(4.23)
j =1
After finishing the first sample input-output pair, the procedure repeats from (4.3) to (4.23)
for each consecutive sample pair. When there is no further improvement in the discrepancy
reduction, the procedure is stopped. Besides consecutively updating the connection weights
as described, the update can be done in a batch mode manner. In the latter case the objective
function (9) is defined as the summation of all discrepancy from individual sample inputoutput pair.
4.6 Deficiencies of Backpropagation
Despite the apparent success of the backpropagation learning algorithm, there are some
aspects, which make the algorithm not guaranteed to be universally useful. Most troublesome
is the long training process. This can be a result of a non-optimum learning rate. A lot of
advanced algorithm based on backpropagation learning have some optimized method to adapt
this learning rate but it is beyond the scope of this study. Outright training failures generally
arise from two sources: network paralysis and local minima.
4.6.1 Network paralysis
As the network trains, the weight can be adjusted to very large values. The total input of a
hidden unit or output unit can therefore reach very high (either positive or negative) values,
and because of the bipolar sigmoid activation function the unit will have an activation very
close to minus one or very close to one. As is clear from Eq. (4.16), the weight adjustments
which are proportional to (1+f(x))(1-f(x)) will be close to minus one, and the training
process can come to a virtual standstill.
4.6.2 Local minima
The error surface of a complex network is full of hills and valleys. Because of the gradient
descent, the network can get trapped in a local minimum when there is a much deeper
minimum nearby. Probabilistic methods can help to avoid this trap, but they tend to be slow.
Another suggested possibility is to increase the number of hidden units. Although this will
work because of the higher dimensionality of the error space, and the chance to get trapped is
smaller, it appears that there is some upper limit of the number of hidden units which, when
exceeded, again results in the system being trapped in local minima.
61
Chapter 5. Numerical Examples
5 NUMERICAL EXAMPLES
5.1 Introduction
From the past observations, most rock tunnel damages were caused by near-fault earthquakes,
which produce ground motion characteristic in the vicinity (<10-25 km) different from that in
the far-field. Since the characteristics of underground structures are depended to the
deformations and strains of the surrounding grounds, the method to describe the ground
behaviors especially near-field earthquake is crucial and necessary. From the past researches
[Wang 1993; Hashash et al. 2001; Corigliano, M., et al. 2007], the seismic designed loads of
the underground structures from both the simplified soil structure interaction analysis and
advanced numerical methods (i.e. Finite Element Method, Finite Difference Method, or
Spectral Element Method) provide a good agreement if the maximum shear strains could be
defined correctly. The idea of this study is to apply the synthetic near fault ground motion
data for ANNs to predict the maximum shear strains under free-field conditions around the
fault, the ANNs-based attenuation of near-fault ground motions.
5.2
Test Description and Data Analyzed
5.2.1 The case study

In this study, the Sannio region, the northern sector of the southern Apennines, had been
used for this study. It is among the most active seismic regions in Italy. In this area five large
earthquakes with IMCS > X occurred in 1456, 1688, 1702, 1735, and 1805, causing several
victims and severe damage. A long seismic quiescence since 1805 event makes the area
highly susceptible to a new earthquake. In this area, there is the new Caserta-Foggia railway
line which has been designed in the late 80s and includes 17 tunnels. The Serro Montefalco
tunnel which belong to the latter section of the Caserta-Foggia. This is a 11.9 km long
tunnel, with a maximum depth of 225 m. It represents one of the most relevant structures of
the entire railway line due to the complexity of the geological context. The lithotypes include
varicoloured clay-shales, marl and marly limestone and clay andmarl intercalated with
limestone, Barla et al. [1986]. The varicoloured clay-shales (the so-called Argille Scagliose)
include expansive clay minerals which exhibit significant swelling behaviour. Previous
excavations of tunnels in this weak rock formation were characterized by severe squeezing
and swelling problems which lead to face instability, large convergences, invert-heave and
critical loading of the tunnel support.
62
Figure 5.1 Location of the Serro Montefalco tunnel (dotted line) along the Caserta-Foggia railway
line (dark solid line). The nearby active faults retrieved from the DISS 3.0.2 database are superimposed.
The Ariano Irpino fault (ITGG092), which is assumed as a potential seismic source in the dynamic
analysis of the tunnel, is highlighted. The short segment perpendicular to the tunnel axis, denotes the
cross-section of the tunnel, Corigliano, M., et al. [2007].
Figure 5.2 Geological profile along the Serro Montefalco tunnel, Barla et al. [1986]
5.2.2 Near-field ground motion modeling

The near-fault ground motion model developed by Hisada and Bielak [2003] had been used
through out this study. This model is accounting for the effects of fling step and rupture
directivity effects based on the computation of static and dynamic Greens function of
displacements and stresses for a viscoelastic horizontally layered half space. It takes
advantage of an analytical expression for the asymptotic solutions of the integrands of the
63
Greens functions, stemming from the generalized R/T reflection and transmission coefficient
method and of the stress discontinuity representations for boundary and source (i.e. kinematic
model of the source) conditions respectively. However, in this study, only the directivity
effect, the dynamic Greens function, had been considered. In order to running the analysis,
this code requires the ground profile and the seismic source model, which had been used
through out this study, are shown below:
-
Ground profile
The Sannio regions geological structure is rather complex and characterized by strong
lateral heterogeneities in the upper 4 km of the earth crust. Improta et al [2000] give an
interpretation of the crustal seismic refraction data from the Northern Sector of the Southern
Apennines thrust belt. This profile has been adopted as a generalized crustal model for the
Sannio region. Since this model is too rough in the shallow part of the earth crust (due to
the fact that only two layers in the first 5 km from the free surface are used), it has been
adapted to fit the soil profile proposed by Cotton et al. [2006] based on the Vs30 parameter. In
this case it was selected a value of Vs30 = 600 m/s to gradually merge with the Vs profile at
greater depths, Corigliano, M., et al. [2007]. The one-dimensional ground response analyses
are assumed and tabulated in the following table and figure. Since the active fault considered
in this study reaches a depth of 25 km and the adopted crustal model is defined only down to
13 km depth, the latter has been extended in depth following the less detailed model proposed
by other authors, Chiarabba and Amato [1997].
Figure 5.3 The crustal velocity profile adopted for the solution of the auxiliary problem, Corigliano, M., et
al. [2007]
64
No. of
Layer
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Density
(t/m3)
1.8
1.9
2.1
2.3
2.4
2.4
2.4
2.5
2.6
2.6
2.6
2.6
2.7
2.7
2.6
2.7
Vp
(m/s)
1,039
1,386
2,252
3,031
3,464
3,724
4,070
4,503
4,850
5,196
5,543
5,785
6,062
6,582
5,023
6,582
Vs
Thickness
(m/s)
(m)
600
25
800
25
1,300
100
1,750
100
2,000
150
2,150
100
2,350
500
2,600
500
2,800
500
3,000
1,000
3,200
1,000
3,340
1,000
3,500
2,000
3,800
4,000
2,900
8,450
3,800
-
Depth
(m)
25
50
150
250
400
500
1,000
1,500
2,000
3,000
4,000
5,000
7,000
11,000
19,450
19,450
Table 5.1The studied ground profile
Seismic source model
In this study, the Ariano Irpino fault has been selected as an only seismic source zone in
this study since it is the closest one to the tunnel and it is characterized by an expected
maximum magnitude of 6.9. It was the source of the December 5, 1456 earthquake, one of the
most significant seismic events of the Italian seismic history, Corigliano, M., et al. [2007].
The main parameters for the fault used in this study are summarized and visualized in the
following.
M0
(Nm)
Mw
2.541019
6.9
L
W
Slip Strike Dip Rake Min. Max. Hypo. Rupture Max. Rise
(Km) (Km) (m)
()
()
()
Depth Depth Depth Velocity Freq. Time
(Km) (Km) (Km) (Km/s) (Hz)
(s)
30
14.9
2
277
70
230
11
25
22.7
2.8
5
1.8
Table 5.2 The features of studied fault, DISS v. 3.0.2
65
X (North)
Free Surface
Strike Angle
Y (East)
strike
(NW-1)*NL+1
Z (Down)
NW*NL
+1
3NL+1
(NW-1)*NL
2NL+1
NL+1
fault
origin
rake
hypocenter
NL+2
NL+3
NL+4
2NL
dip
NL
strike
Figure 5.4 The general outline of the studied fault and its subfaults, Hisada and Bielak [2003]
5.2.3 Identification of shear strains by ANNs

In this study the maximum strain tensors can be obtained by the displacement time histories at
six points around the observation points which calculated by Hisada and Bielak [2003]s
code. The maximum shear strains in yz- and xy- directions had been chosen for ANNs to
predict. These shear strains can be computed by the displacement response from the following
equations:
w
1
1
( y0 , z0 + z ) ( y0 , z0 z ) +
w ( y0 + y, z0 ) w ( y0 y, z0 )
+
2y
z y 2z
1
1
u
( y0 , x0 + x ) ( y0 , x0 x ) +
u ( y0 + y, x0 ) u ( y0 y, x0 )
=
+
2y
x y 2x
(5.1)
yz =
xy
The choosing of the inputs for ANNs are also very important, since they are directly related to
the quality of ANNs computed values, and the amount of time that need for ANNs to learn
the given ground motion characteristics. There are six inputs for ANNs in this study
composing of
66
the soil density,
the maximum shear modulus,
the Peak Ground Velocity (PGV),
the distance in x and y directions from the fault origin, and
the depth of the observation points.
The PGV had been choosing as an input, since the peak strain (PGS) is dependent on peak
particle velocity (PGV) as:
PGS =
PGV
C
(5.2)
, which is crucial for ANNs to determine the PGS. The other reason for choosing these input
parameters lies in the fact that these inputs are general and easy to be acquired in the field
experiments. The following flow chart will explain briefly the framework, adopted for this
study.
67
Geotechnical, and fault parameter

estimation
Near fault responses from observed data
by Hisadas code
Re-training process by changing
ANNs training process
Verification of ANNs model using
ANNs parameters or acquiring

more training data
No
the lowest error prediction criteria
Yes
ANNs testing process
No
Result comparison between exact
solutions and ANNs solutions
Yes
ANNs based identification of near fault shear strains.
Figure 5.5 Methodology adopted for this study
5.3 Results and interpretation of numerical computations

The observation point is the point where seismometers or accelerograms would be place to
record the ground motion characteristics. The observation points had been assumed to be
positioned in North and West direction (0 and 270 degree strike angle, respectively), and the
280 degree as shown in the Figure 5.5. The distance between observation points equals to 1
km from 1 km to 40 kms. For this study, it was assumed that we have observation points only
in 100- and 600-meter depths. These synthetic data would be used as a training data for ANNs
to learn the near-field ground characteristics. From this assumption, the trained ANNs will be
used to predict the maximum shear strains in different directions at 400-, 800-, and 1,000meter depths.
68
North
Observation points
Active Fault
West
Observation points
The observation points with available filed-experiment data

The observation points with ground characteristics predicted by ANNs
Figure 5.6 The general outline of seismic source zone
The YZ-shear strain at 0- strike degree axis

Exact solution, 0-degree axis
6.E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1000-meter depth
YZ - Shear strain
5.E-05
4.E-05
3.E-05
2.E-05
1.E-05
0.E+00
0
10
15
20
25
Distance (Km s)
30
35
40
Figure 5.7 The exact 0-degree YZ-shear strain computed by Hisada code
PGV, 0-degree axis
PGV (m/s)
5.3.1
0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1000-meter depth
10
15
20
25
30
35
40
Distance (Km s)
69
Figure 5.8 The attenuation of PGVs at 0-degree axis

Depth 400 meters, 0-degree
4.0E-05
ANNs
YZ - Shear strain
3.5E-05
Hisada
3.0E-05
2.5E-05
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
40
Depth 800 meters, 0-degree axis
6.E-05
ANNs
5.E-05
YZ - Shear strain
35
Hisada
4.E-05
3.E-05
2.E-05
1.E-05
0.E+00
0
10
15
20
25
Distance (Km s)
30
40
Depth 1,000 meters, 0-degree axis
6.E-05
ANNs
5.E-05
YZ - Shear strain
35
Hisada
4.E-05
3.E-05
2.E-05
1.E-05
0.E+00
0
10
15
20
25
Distance (Km s)
30
35
40
Figure 5.9 The comparison between 0-degree YZ-shear strain computed by Hisada and ANNs at different
depth
From the computation, it can be seen clearly that ANNs can learn and simulate the near-field
YZ-shear strain at different depths but in the same direction as the training data sets.
Moreover, the effect of PGV to the maximum shear strains can be seen clearly by the
70
attenuation relationship of PGV and maximum shear strain.

The YZ-shear strain at 270-strike degree axis
1.6E-04
YZ-Shear strain
1.4E-04
1.2E-04
1.0E-04
8.0E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1000-meter depth
6.0E-05
4.0E-05
2.0E-05
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
PGV, 270-degree axis
0.8
PGV (m/s)
0.7
0.6
0.5
0.4
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1000-meter depth
0.3
0.2
0.1
0
0
10
15
20
25
Distance (Km s)
30
35
40

Depth 400 m eters, 270 degrees
1.4E-04
1.2E-04
YZ-Shear strain
5.3.2
1.0E-04
8.0E-05
6.0E-05
4.0E-05
ANNs
2.0E-05
Hisada
0.0E+00
0
10
15
20
25
30
35
40
Distance (Kms)
71

1.6E-04
YZ-Shear strain
1.4E-04
1.2E-04
1.0E-04
8.0E-05
6.0E-05
4.0E-05
ANNs
2.0E-05
Hisada
0.0E+00
0
10
15
20
25
Displacem ent (Km s)
30
35
40

1.6E-04
YZ-Shear strain
1.4E-04
1.2E-04
1.0E-04
8.0E-05
6.0E-05
4.0E-05
ANNs
2.0E-05
Hisada
0.0E+00
0
10
15
20
25
Displacem ent (Km s)
30
35
40
Figure 5.12 The comparison between 270-degree YZ-shear strain computed by Hisada and ANNs at
different depths
The ANNs and Hisadas computation shows a good agreement in the 400-meter depth
prediction; however, for the 800- and 1,000-meter depth, the ANNs loss its capability to
predict the maximum shear strain at farther distance from the fault origin. This effect comes
from the limitation of ANNs to predict values which exceed its upper bound in the training
data sets, i.e. in the 800-and 1,000-meter depths. This effect can be seen as the requirement in
choosing the appropriate observation points in the future and practical works. Moreover, the
effect of PGV to the maximum shear strains can be seen clearly by the attenuation
relationship of PGV and maximum shear strain.
72

1.E-04
YZ-Shear strain
8.E-05
6.E-05
4.E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
2.E-05
0.E+00
0
10
15
20
25
Distance (Km s)
30
35
40
PGV (m/s)

0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
0
10
15
20
25
Distance (Km s)
30
35
40

8.0E-05
7.0E-05
YZ-Shear strain
5.3.3
6.0E-05
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
73

1.0E-04
9.0E-05
YZ-Shear strain
8.0E-05
7.0E-05
6.0E-05
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
Depth 1,000 m eters, 280 degrees

1.0E-04
9.0E-05
YZ-Shear strain
8.0E-05
7.0E-05
6.0E-05
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
Figure 5.15 The comparison between 280-degree YZ-shear strain computed by Hisada and ANNs at
different depths
The ANNs and Hisadas computation at 280-degree direction and at 400-meter depth
provides a good agreement. However, the ANNs prediction accuracy is lower at deeper depth
and at higher distance from the fault origin, caused by the effect of exceeding the upper bound
of training data set in deeper depth. This effect can be seen as the requirement in choosing the
appropriate observation point depth in the future and practical works. Moreover, the effect of
PGV to the maximum shear strains can be seen clearly by the attenuation relationship of PGV
and maximum shear strain in each direction.
74

6.0E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
YZ-Shear strain
5.0E-05
4.0E-05
3.0E-05
2.0E-05
1.0E-05
0.0E+00
0
10
15
20
25
30
35
40
Distance (Km s)
0.25
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
PGV (m/s)
0.2
0.15
0.1
0.05
0
0
10
15
20
25
Distance (Km s)
30
35
40

3.5E-05
ANNs
3.0E-05
YZ-Shear strain
5.3.4
Hisada
2.5E-05
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
75

4.5E-05
ANNs
4.0E-05
Hisada
YZ-Shear strain
3.5E-05
3.0E-05
2.5E-05
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

5.0E-05
ANNs
4.5E-05
Hisada
YZ-Shear strain
4.0E-05
3.5E-05
3.0E-05
2.5E-05
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

6.0E-05
ANNs
YZ-Shear strain
5.0E-05
Hisada
4.0E-05
3.0E-05
2.0E-05
1.0E-05
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
76

7.0E-05
ANNs
YZ-Shear strain
6.0E-05
Hisada
5.0E-05
4.0E-05
3.0E-05
2.0E-05
1.0E-05
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
Figure 5.18The comparison between 315-degree YZ-shear strain computed by Hisada and ANNs at
different depths
For 315-degree computation, which is completely different testing data from the training data
set, the ANNs still can simulate the near-field YZ-shear strain at different depths. However,
the ANNs computation at 315- degree YZ-shear strains and at 800- and 1,000-meter depth
shows the limited of ANNs to predict the higher shear strain value than the ones that had been
used as the training data. This effect can be seen as the requirement in choosing the
appropriate observation points in the future and practical works. Moreover, the effect of PGV
to the maximum shear strains can be seen clearly by the attenuation relationship of PGV and
maximum shear strain in each direction.
The XY-shear strain at 0-strike degree axis
4.E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
3.E-05
XY - Shear strain
5.3.5
3.E-05
2.E-05
2.E-05
1.E-05
5.E-06
0.E+00
0
10
15
20
25
Distance (m eters)
30
35
40
Figure 5.19 The exact 0-degree XY-shear strain computed by Hisada code
77
PGV (m/s)
PGV, 0-degree axis

0.2
0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1000-meter depth
10
15
20
25
30
35
40
Distance (Km s)

3.0E-05
XY-Shear strain
2.5E-05
2.0E-05
1.5E-05
1.0E-05
ANNs
5.0E-06
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

2.5E-05
XY-Shear strain
2.0E-05
1.5E-05
1.0E-05
5.0E-06
ANNs
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
78

2.5E-05
XY-Shear strain
2.0E-05
1.5E-05
1.0E-05
5.0E-06
ANNs
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
Figure 5.21 The comparison between 0-degree XY-shear strain computed by Hisada and ANNs at
different depths
The ANNs and Hisadas computed shear strains show a good agreement, even at higher
depth, i.e. 800- and 1,000-meter depth, since the training data sets provide the highest upper
bound for shear strain in XY direction. Moreover, the effect of PGV to the maximum shear
strains can be seen clearly by the attenuation relationship of PGV and maximum shear strain
in each direction.
1.8E-04
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
1.6E-04
XY-Shear strain
5.3.6
1.4E-04
1.2E-04
1.0E-04
8.0E-05
6.0E-05
4.0E-05
2.0E-05
0.0E+00
0
10
15
20
25
Distance (m eters)
30
35
40
79

0.8
PGV (m/s)
0.7
0.6
0.5
0.4
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1000-meter depth
0.3
0.2
0.1
0
0
10
15
20
25
Distance (Km s)
30
35
40

1.6E-04
XY-Shear strain
1.4E-04
1.2E-04
1.0E-04
8.0E-05
6.0E-05
4.0E-05
ANNs
2.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

1.2E-04
XY-Shear strain
1.0E-04
8.0E-05
6.0E-05
4.0E-05
ANNs
2.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
80

1.0E-04
9.0E-05
XY-Shear strain
8.0E-05
7.0E-05
6.0E-05
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
different depths
It can be seen clearly that ANNs can learn and simulate the near-field XY-shear strain at
different depths at 270-degree direction. The ANNs computation provides a good result when
the highest maximum shear strain had been chosen in the training data set. Moreover, the
relationship of PGV and maximum shear strain in each direction.
1.0E-04
9.0E-05
XY-Shear strain
5.3.7
8.0E-05
7.0E-05
6.0E-05
5.0E-05
4.0E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
3.0E-05
2.0E-05
1.0E-05
0.0E+00
0
10
15
20
25
Distance (m eters)
30
35
40
81
PGV (m/s)

0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
0
10
15
20
25
Distance (Km s)
30
35
40

9.0E-05
XY-Shear strain
8.0E-05
7.0E-05
6.0E-05
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

7.0E-05
XY-Shear strain
6.0E-05
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
82

6.0E-05
XY-Shear strain
5.0E-05
4.0E-05
3.0E-05
2.0E-05
ANNs
1.0E-05
Hisada
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
different depths
It can be seen clearly that ANNs can learn and simulate the near-field XY-shear strain at
different depths at 280-degree direction. The ANNs computation provides a good result when
the highest maximum shear strain had been chosen in the training data set. Moreover, the
relationship of PGV and maximum shear strain in each direction.
3.5E-05
3.0E-05
XY-Shear strain
5.3.8
2.5E-05
2.0E-05
1.5E-05
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (m eters)
30
35
40
83

0.25
100-meter depth
400-meter depth
600-meter depth
800-meter depth
1,000-meter depth
PGV (m/s)
0.2
0.15
0.1
0.05
0
0
10
15
20
25
Distance (Km s)
30
35
40

4.0E-05
ANNs
XY-Shear strain
3.5E-05
Hisada
3.0E-05
2.5E-05
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

3.5E-05
ANNs
XY-Shear strain
3.0E-05
Hisada
2.5E-05
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
84

3.0E-05
ANNs
XY-Shear strain
2.5E-05
Hisada
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

3.0E-05
ANNs
XY-Shear strain
2.5E-05
Hisada
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40

2.5E-05
ANNs
Hisada
XY-Shear strain
2.0E-05
1.5E-05
1.0E-05
5.0E-06
0.0E+00
0
10
15
20
25
Distance (Km s)
30
35
40
different depths
For 315-degree computation, which is completely different testing data from the training data
set, the ANNs still can learn and simulate the near-field YZ-shear strain at different depths.
85
Chapter 6: Conclusion and Future Research
6 CONCLUSION AND FUTURE RESEARCH

6.1 Introduction
Since the effects of near fault effect owned to the forward directivity and fling step effects
contribute to the underground structural damages, which are largely different from that in the
far-field case. Corigliano, M., et al. [2007] developed the pseudo-static equations to predict
the seismic stress increment in the lining. The comparison between the pseudo-static and
numerical examples showed a reasonable result from his study. From the equation (3.44) and
(3.45), the seismic forces can be predicted correctly from the prediction of the maximum
shear strain that is the key parameter and it is the main concern in this study. The simplified
tool, the ANNs, had been developed and calibrated to compute the maximum shear strain
under free-field in near-fault region.
To develop an ANNs-based method to determine the maximum shear strain, the Ariano
Irpino fault in the Sannio region had been used as the seismic source in this study which
its near-fault ground motion record was generated by semi analytical method developed by
Hisada and Bielak [2003]. To compute the earthquake-induced shear strain field in the
vicinity of a causative fault, the displacement at six points around the interested observation
points is calculated, and the shear strains can be calculated based on equation (5.1).
Since the ANNs need training data set in order to learn the near-field ground motion
characteristics, the 100- and 600- meter depth and at 0-, 270-, and 280- strike degree data set
had been used for this purpose. The trained ANNs would then be tested to predict the
maximum shear strain at 400-, 800-, and 1,000-meter depth and at 0-, 270-, 280-, and 315strike degree).
6.2 Numerical examples
The choosing of observation points as a training data set is very important, since the ANNs is
highly affected to the quality of training data set. From Figure 5.12, in 270 degrees at 800and 1,000- meter depths, the computed maximum YZ-shear strain are clearly over predict the
target result. The main reason is the training data set, 100- and 600-meter depths have lower
maximum shear strain ordinate than that at800- and 1,000-meter depths. This observation
should be considered for installation of the seismograms or accelerometers in order to
improve the ANNs computation accuracy.
From the numerical calculation, the ANNs can predict the maximum shear strain in the same
86
Chapter 6: Conclusion and Future Research
direction as using in training data set but at different depth, with a reasonable accuracy as long
as the limit of upper bound of training data set does not exceed, Figure 5.9, 5.21, 5.24.
The computed result from ANNs in different strike angles and depths from the training data
set give a rationally acceptable as long as the limit of upper bound of training data set does
not exceed, Figure 5.30.
6.3 Future research
From this study, the prediction of the earthquake-induced maximum shear strain in the ground
under free-field condition had been proposed. The numerical examples show a possible
capability of ANNs in computing the maximum shear strains around the studied fault.
The applicable of ANNs to determine the maximum free-field shear strain should be applied
to other fault geometries and earthquake magnitudes to test the versatility of ANNs.
Moreover, the choosing of observation points in order to optimize and increase the predicted
accuracy should be also investigated.
87
References
REFERENCES
AASHTO [1998] LRFD Bridge Design Specifications, Second Edition, 2001 Interim Revisions,
American Association of State Highway and Transportation Officials, Washington DC.
AKI, K. [1993]. Local site effects on weak and strong ground motion, Tectonophysics, 218, pp. 93111.
American Lifelines Alliance [2001], Seismic Fragility Formulations for Water Systems.
Asakura T and Sato Y, [1998] Mountain Tunnels Damage in the 1995 HYOGOKEN-NANBU
Earthquake, QR of Railway Technical Research Institute (RTRI), 39(1) pp. 916.
Bommer, J.J. & Martinez-Pereira, A. [1999]. The effective duration of earthquake strong motion,
Journal of Earthquake Engineering, 3 (2), pp. 127-172.
Bray and Rodriguez-Marek [2004] Characterization of forward directivity ground motion in the near
fault region. Soil Dynamic and Earthquake Engineering, 24, pp. 815-828.
Burns, J. Q., and Richard, R. M. [1964] Attenuation of stresses for buried cylinders, Proceeding of
the Symposium on Soil-Structure Interaction, University of Arizona at Tempe, Arizona.
Hoeg, K. [1968] Stresses against underground structural cylinders, Journal Soil Mechanics Found.
Div., ASCE 94 (SM4), pp. 883-858.
Corigliano, M., Scandella, L., Barla, G., Lai, C. G., Paolucci, R. [2007] Seismic analysis of deep
tunnels in weak rock: a case study in southern Italy, Procceding of 4th International Conference on
Earthquake Geotechnical Engineering, Paper No. 1616, Thessaloniki, Greece
Cotecchia V, [1986] Ground deformations and slope instability produced by the earthquake of 23
November 1980 in Campania and Basilicata, International symposium on Engineering Geology in
Seismic Areas. Bari, Italy: Institution of Engineering Geology and Geotechnics, University of Bari,
Italy.
Cotton, F., Scherbaum, F., Bommer, J. J., and Bungum, H. [2006] Criteria for selecting and adjusting
ground motion models for specific target regions: Application to Central Europe and rock sites,
Journal of Seismology, 10, pp.137-156
DISS v. 3.0.1 [2006]: Database of Individual Seismogenic Sources: A compilation of potential sources
for earthquakes larger than M 5.5 in Italy and surrounding areas. http://www.ingv.it/DISS/.
Dowding, C.H., Rozen, A. [1978] Damage to Rock Tunnels from Earthquake Shaking, Journal of
88
References
Geotechnical Engineering Division, ASCE 104 GT2, pp. 175- 191

Einstein H. H. and Schwartz C. W. [1979] Simplified analysis for tunnel support, Journal of the
geotechnical engineering division, 105, pp. 499-518.
Japanese Society of Civil Engineers, JSCE [1975] Specification for Earthquake Resistant Design of
Submerged Tunnels.
Jenning, P. C. [1985] Ground motion parameters that influence structural damage, in R.E. Scholl
and J.L. King, eds. Strong Ground Motion Simulation and Engineering Applications, EERI
Publication 85-02, EERI, Berkeley, 298 pp.
Joyner, W.B. and Boore, D.M. [1988] Measurement, characterization, and prediction of strong
ground motion, in Earthquake Engineering and Soil Dynamics II Recent Advances in Ground
Motion Evaluation, Geotechnical Special Publication 20, ASCE, New York, pp. 43-102.
Gomez-Masso, A. and Attalla, I. [1984] Finite element vs. simplified methods in the seismic analysis
of underground structures, Earthquake Engineering and Structural Dynamics, 12, pp. 347-367.
Hashash, M. A. Youssef, Hook, J., Jeffrey, Schmidt Birger, Yao, John I-Chiang [2001] Seismic
design and analysis of underground structures, Tunneling and Underground Space Technology
Vol. 16, pp. 247-293.
Hashash. Y.M.A., Tseng, W.S. and Krimotat, A. [1998] Seismic Soil-structure Interaction Analysis
for Immersed Tube Tunnels Retrofit, Geotechnical Earthquake Engineering and Soil Mechanics.
Vol. I11 2, pp. 1380-1391. ASCE Geotechnical Special Publication no. 75.
Haykin, S. [1998] Neural Networks: A Comprehensive Foundation. Prentice Hall, 2nd edition.
Hertz, J., Krogh, A., and Palmer, R. G. [1991], An Introduction to the Theory of Neural
Computation. Lecture Notes, Volume I. Addison-Wesley, Redwood City, CA.
Hisada Y. and Bielak J., [2003] A theoretical method for computing near fault ground motion in a
layered half-spaces considering static offset due to surface faulting, with a physical interpretation
of fling step and rupture directivity, Bulletin of the Seismological Society of America 93(3), pp.
1154-1168.
Hwang, R.N., and Lysmer, J. [1981] Response of Buried Structures to Traveling Waves, Journal of
the Geotechnical Engineering Division, ASCE, Vol. 107, No. GT2.
Idriss, I.M. and Seed, H.B. [1968] Seismic response of horizontal soil layers, J. Soil mech. Found.
Div., ASCE 94 (SM4), pp. 1003-1031.
Improta, L., Iannaccone, G., Capuano P., Zollo, A., Scandone, P., [2000] Inferences on the upper
crustal structure of Southern Apennines (Italy) from seismic refraction investigations and
subsurface data, Tectonophysics, 317, pp. 273-297.
Indrawan, B., [2001] Lessons learned from past Earthquake Damages and seismic Observation of
Tunnel Structures, Tunneling seminar, University of Pelita Harapan. Karawachi, Indonesia.
Kawashima, K. [2000] Seismic Design of Underground Structures in Soft Soil Ground: A Review,
Key Note Presentation, Geothechnical Aspects of Underground Construction in Soft Ground,
Balkema, Rotterdam, 3-19, 2000
Kramar, S. L. [1996]. Geotechnical Earthquake Engineering. New Jersey: Prentice Hall.
89
References
Kuesel, T.R. [1969]. Earthquake Design Criteria for Subways, Journal of Structure Division, ASCE
ST6, pp. 1213- 1231.
McGarr, A. [1983]. Estimated ground motion for small near-by earthquakes, Seismic Design of
Embankment and Caverns, pp. 113-127. New York: ASCE.
Monsees, J.E. and Merritt J.L., [1988] Seismic modeling and design of underground structures,
Numerical Methods in Geomechanics, Innsbruck, pp. 1833-1842.
Monsees, J.E. and Merritt J.L., [1991] Earthquake considerations in design of the Los Angeles
Metro, Proceeding of the ASCE Conference on Lifeline Earthquake Engineering.
Newmark, N. M. [1967] Problems in wave propagation in soil and rocks, Proceedings of the
International Symposium on Wave Propagation and Dynamic Properties of Earth Materials,
University of New Mexico Press, pp.7-26.
Okamoto, S. [1973], Introduction to Earthquake Engineering, University of Tokyo Press, Tokyo,
Japan, 1973.
Owen, G.N. and Scholl, R.E. [1981] Earthquake Engineering of Large Underground Structures.
Report no. FHWARD-80-195, Federal Highway Administration and National Science Foundation.
Peck , R.B., Hendron, A.J., Mohraz, B.,[1972] State of the art in soft ground tunneling, Proceeding
of the Rapid Excavation and Tunneling Conference. American Institute of Mining, Metallurgical
and Petroleum Engineers, New York, pp. 259-286.
Power, M.S., Rosidi, D., and Kaneshiro, J., [1996], Vol. III Strawman: screening, evaluation, and
retrofit design of tunnels, Report Draft National Center for Earthquake Engineering Research,
Buffalo, New York.
Power, M.S., Rosidi, D., and Kaneshiro, J., Gilstrap, S.D., and Chiou, S.J. [1998], Summary and
Evaluation of Procedures for the Seismic Design of Tunnels: Multidisciplinary Center for
Earthquake Engineering Research, Draft Technical Report, FHWA Contract No. DTFH61-92-C0012 Task 112-D-5.3c, Sept. 1998.
Richardson, A., and Blejwas T. E. [1992] Seismic Design of Circular-Section Concrete-Lined
Underground Openings Preclosure Performance Considerations for the Yucca Mountain Site,
ASCE Symposium on Dynamic Analysis and Design Considerations for High-Level Nuclear Waste
Repositories.
Rumelhart, D. E., Hinton, G. E., and Williams, R. J. [1986] A Learning representations by backpropagating errors, Nature 323: pp. 533-536.
Sanner, R.M. and Slotine, J-J.E. [1992] Gaussian networks for direct adaptive control, IEEE
Transcations on Neural Network 2(6):pp. 837-863.
Sakurai, A. and Takahashi, T. [1969] Dynamic stresses of underground pipeline during earthquakes,
Proceeding of the Fourth World Conference on Earthquake Engineering.
Tzirkel-Hancock, E. and Fallside, F. [1992] Stable control of nonlinear systems using neural
networks, International Journal of Robust and Nonlinear Control, 2(1): pp. 63-86.
Sharma S and Judd W R, [1991] Underground opening damage from earthquakes, Engineering
Geology, Vol. 30: pp. 263276.
90
References
Simpson, K.A. [1996]. Attenuation of strong ground-motion incorporating near-surface foundation

conditions. Doctoral Thesis, University of London.
Somerville, P.G., Smith, N.F., Graves, R.W., and Abrahamson, N.A. [1997] Modification of
empirical strong ground motion attenuation relationships to include the amplitude and duration
effects of rupture directivity, Seismological Research Letters, 68 (1), pp.199-222.
SOMERVILLE, P. [1998] Ground motion attenuation relationships and their application to aseismic
design and seismic zonation, In: Irikura, Kudo, Okada & Sasatani (eds) The effects of Surface
Geology on Seismic Motion. pp.35-49.
Somerville, P. [2000] Characterization of Near-Fault Ground Motions, Proceeding of the U.S.-Japan
Workshop on the Effects of Near-Field Earthquake Shaking, Pacific Earthquake Engineering
Research Center, pp. 21-30.
St. John, C.M. and Zahrah, T.F. [1987] Aseismic Design of Underground Structures, Tunneling
Underground Space Technology. Vol. 22, pp. 165 - 197.
Stewart, J. P., Chiou, S. J., Bray, J. D., Graves, R. W., Somerville, P. G., Abrahamson, N. A., [2001],
Ground Motion Evaluation Procedures for Performance-Based Design, Report PEER 2001/09.
Sweet, J. [1997] Los Angeles Metro Red Line project: seismic analysis of the Little Tokyo Subway
Station, Report no. CAI-097-100. Engineering Management Consultants.
Tso, W.K., Zhu, T.J. & Heidebrecht, A.C. [1992] Engineering implication of ground motion A/V
ratio, Soil Dynamics and Earthquake Engineering. 11, pp.133-144.
Yoshikawa, K., Fukuchi, G. [1984] Earthquake Damage to Railway Tunnels in Japan, Advances in
Tunneling Technology and Subsurface Use, Vol.4 Nr.3, pp. 75-83
W.L. Wang, T.T. Wang, J.J. Su, C.H. Lin, C.R.Seng, T.H.Huang, [2001] Assessment of damage in
mountain tunnels due to the Taiwan Chi-Chi Earthquake, Tunneling and Underground Space
Technology, Vol 16, pp. 133-150
Wang, J. N. [1993] Seismic Design of Tunnels, A State-of-the-Art Approach, Monograph 7,
Parsons Brickerhoff Quade & Douglas, Inc., New York.
Zurada, J. M. [1992], Artificial Neural Systems. West Publishing Company, St. Paul, MN.
91

ANN Aplid To Seismic Assmnt of Deep Tunnels

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

ANN Aplid To Seismic Assmnt of Deep Tunnels

Hochgeladen von

Copyright:

Verfügbare Formate

Istituto Universitario

EUROPEAN SCHOOL FOR ADVANCED STUDIES IN

ARTIFICIAL NEURAL NETWORKS APPLIED TO THE

A Dissertation Submitted in Partial

Supervisor: Dr CARLO G. LAI, Dr. MIRKO CORIGLIANO

Dr. Carlo G. Lai

Dr. Mirko Corigliano

General outlines of study ...........................................................................................................5

2 THE DAMAGES TO UNDERGROUND STRUCTURES ..............................................................6

Typologies of underground structures .......................................................................................6

The ground motion parameters ..................................................................................................8

The past studies........................................................................................................................14

2.3.4 Sharma and Judd [1991] ................................................................................................18

Lesson learns from the past damages.......................................................................................23

3 SEISMIC DESIGN AND ANALYSIS PROCEDURES FOR UNDERGROUND STRUCTURES26

Seismic behaviors of underground structures ..........................................................................26

Design and analysis methods ...................................................................................................29

4 ARTIFICIAL NEURAL NETWORKS...........................................................................................52

Introduction to Artificial Neural Networks..............................................................................52

Neural Network Architectures .................................................................................................53

Activation Function .................................................................................................................55

Training Algorithm ..................................................................................................................56

The Backpropagation Network ................................................................................................57

Deficiencies of Backpropagation .............................................................................................61

4.6.1 Network paralysis ..........................................................................................................61

Test Description and Data Analyzed .......................................................................................62

Results and interpretation of numerical computations.............................................................68

6 CONCLUSION AND FUTURE RESEARCH ...............................................................................86

Figure 1.1 Shapes of underground structures [Kawashima, 2000]............................................3

Figure 1.1 Shapes of underground structures [Kawashima, 2000]

From past investigations, the underground structures response in accordance to the

with a detailed analysis of the typologies of damages caused by earthquakes

The seismic behaviors of underground structures would be described at the

The capabilities of ANNs to reproduce the maximum near-field ground motion

Chapter 2. The Damages to Underground Structures

2 THE DAMAGES TO UNDERGROUND STRUCTURES

Chapter 2. The Damages to Underground Structures

2.1.2 Construction Methods

2.1.3 Sectional typologies

Chapter 2. The Damages to Underground Structures

Figure 2.1 An idealization of widely spread underground structure, Kawashima [2000]

2.2 The ground motion parameters

Chapter 2. The Damages to Underground Structures

Chapter 2. The Damages to Underground Structures

Chapter 2. The Damages to Underground Structures

Near-field ground motion

Chapter 2. The Damages to Underground Structures

Chapter 2. The Damages to Underground Structures

Chapter 2. The Damages to Underground Structures

2.3 The past studies

Chapter 2. The Damages to Underground Structures

Information on the performance of underground openings during earthquakes is relatively

(a) peak surface acceleration

(b) peak particle velocities

Chapter 2. The Damages to Underground Structures

Chapter 2. The Damages to Underground Structures

Tunnels deep in rock are safer than shallow tunnels

No collapse was observed due to ground shaking effect alone up to a surface

2.3.2 Owen and Scholl [1981]

Deep tunnels were less prone to damage than shallow tunnels.

Duration of strong seismic motion appeared to be an important factor contributing to the

2.3.3 Yoshikawa and Fukuchi [1984]

Chapter 2. The Damages to Underground Structures