54 Moeller - Libero
54 Moeller - Libero
Sven Möller
l
i-1 i i+1
2006 Mitteilung 54
des Instituts für Geotechnik IGS
Herausgeber P. A. Vermeer
Herausgeber:
Prof. Dr.-Ing. P. A. Vermeer
Institut für Geotechnik
Universität Stuttgart
Pfaffenwaldring 35
70569 Stuttgart
Telefon 0711/685-62436
Telefax 0711/685-62439
e-mail: pieter.vermeer@igs.uni-stuttgart.de
ISBN-10: 3-921837-54-5
ISBN-13: 978-3-921837-54-2
Gegen Vervielfältigung und Übersetzung bestehen keine Einwände, es wird lediglich um
Quellenangabe gebeten.
vorgelegt von
Pieter A. Vermeer
*) for references the reader is referred to the list at the end of this dissertation study
i
Acknowledgments
The research presented in this thesis is a result of the work carried out in the years 2001-
2006 at the Institute of Geotechnical Engineering, University of Stuttgart. No doubt, this
result could only be achieved by the support of a number of people to whom I would
like to express my gratitude.
To my supervisor Professor Pieter Vermeer: I experienced the freedom I needed to be
creative while at the same time I was always helped to continue research in case of diffi-
culties. Most grateful I am to him knowing that the last years will be of great profit for
my further way. I am really looking forward to continue this valuable working relation-
ship.
To my colleagues: Thank you for being such a good team. I have always enjoyed work-
ing and having a good time with you.
To my wife Anna: I know that a man can only be as strong as the woman behind the
man. Thank you for all your energy and patience.
To my parents: Thank you Mom for correcting my English. After reading my thesis I
hope you are not afraid of driving through tunnels. Believe me when I say, most of them
are safe. Thank you Dad for supporting me and my family throughout my entire studies.
Sven Möller
iii
Contents
1 Introduction 1
v
Contents
vi
Contents
6 Conclusions 127
6.1 On elementary design methods . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.2 On the FEM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.3 Recommendations for further research . . . . . . . . . . . . . . . . . . . . . 130
Bibliography 141
vii
Zusammenfassung
Die vorliegende Arbeit verschafft einen Überblick sowohl über gängige vereinfachte
Tunnelberechnungsmethoden als auch über tunnelstatische Berechnungen mit Hilfe von
drei- und zweidimensionalen Finite-Element-Simulationen. Ein Hauptaugenmerk wird
dabei auf die Wahl und den Einfluss von statischen Systemen und Randbedingungen
zur Berücksichtigung unterschiedlicher Tunnelvortriebsmethoden gelegt. Darüber hin-
aus wird auch die Auswirkung unterschiedlicher Stoffgesetzte zur Modellierung des
Baugrundverhaltens untersucht. In diesem Zusammenhang werden folgende Kapitel
berücksichtigt:
ix
Zusammenfassung
Kapitel 2 Dieses Kapitel verschafft dem Leser einen Überblick über gängige moderne
Tunnelvortriebsmethoden. Dabei wird unterschieden zwischen Methoden bei denen
eine ungestützte Ortsbrust zur Ausführung gelangt und solchen, bei denen die Orts-
brust durch ein Stützmedium verschlossen ist.
Kapitel 4 Dieses Kapitel behandelt die Simulation von Tunnelvortrieben mit Hilfe der
Finite-Element-Methode. Um ein besseres Verständnis für das weitere Studium dieser
Arbeit zu fördern, werden zunächst einige grundlegende Begriffe sowie Modellierungs-
aspekte von zwei- und dreidimensionalen Finite-Element-Tunnelsimulationen aufge-
griffen und erläutert. Die Finite-Element-Methode ist durch eine Reihe von Faktoren
beeinflusst, welche bei Nichtbeachtung zu einem erheblichen Genauigkeitsverlust der
Berechnungsergebnisse führen können. Aus diesem Grunde werden zunächst die Be-
deutung und der Einfluss des tolerierten Gleichgewichtsfehlers und die Bestimmung
von ausreichenden Finite-Element-Netzabmessungen sowie Netzfeinheiten für die Be-
rechnung von zwei- und dreidimensionalen Setzungs- und Schnittkraftverläufen unter-
sucht. Die Auswirkung der Größe und Verteilung der Ausgangsbaugrundspannungen
auf die Ergebnisse von Oberflächensetzungen und Schnittkräften wird in einem geson-
derte Unterkapitel untersucht. Ein Schwerpunkt dieses Kapitels liegt auf der Berück-
sichtigung unterschiedlicher Tunnelvortriebsmethoden bei der numerischen Simulation
von Setzungen und Schnittkräften. In diesem Zuge werden sowohl zwei-, als auch drei-
dimensionale numerische Tunnelsimulationsmethoden anhand einer Literaturstudie zu-
sammengefasst. Zur Berechnung von geschlossenen Schildvortrieben wird sowohl ein
verbessertes drei- als auch ein zweidimensinales Verfahren vorgeschlagen, welche die
Simulation des Ortsbruststützdruckes oder/und der Zementmörtelverpressdrücke vor-
sieht. Eine gängige zweidimensionale Berechnungsmethode wird genauer untersucht,
indem Ergebnisse von systematischen Parameterstudien sowie Empfehlungen zur Größe
des sogenannten Vorentlastungsfaktors gezeigt werden.
x
Zusammenfassung
Kapitel 6 Im letzten Kapitel werden die Ergebnisse dieser Arbeit noch einmal zusam-
mengefasst und mit den wichtigsten Schlussfolgerungen bedacht. Anschließend wird
ein Ausblick auf fortführende Forschungsmöglichkeiten gegeben.
xi
Abstract
The design of tunnels requires a proper estimate of surface settlements and lining forces.
In engineering practice different design methods tend to be used, varying from simple
empirical and analytical formulations to advanced finite element analyses. Depending
on the tunnelling method, e.g. conventional or closed shield tunnelling, different proce-
dures are applied for the modelling of the excavation and support sequence. Such pro-
cedures have a significant effect on predicted/computed deformations and lining forces.
The method of analysis needs to take into consideration the effects of the installation on
the ground-lining interaction.
This thesis focuses on shallow tunnelling in soil. It begins with a summary of mod-
ern tunnelling methods, distinguishing between installation procedures of open face
and closed face tunnelling. Subsequently elementary methods of analysis for both set-
tlements and lining forces are reviewed, placing emphasis on installation procedures.
Hereafter the focus is on the use of three- and two-dimensional finite element analyses.
Besides the influence of some basic modelling aspects of finite element analyses, the im-
portance of adequately modelling the excavation and support sequence, in particular the
appropriate boundary conditions to be specified at the tunnel face and wall are exam-
ined. After a review on the aspects of such installation procedures, results for different
installation procedures and different constitutive models are presented on the basis of
two case studies. It is shown that installation procedures are most important to be con-
sidered in order to arrive at proper predictions for tunnelling settlements, horizontal
ground movements and lining forces.
xiii
Chapter 1
Introduction
Tunnel design requires a proper estimate of both ground deformations and pressures on
the tunnel lining. Depending on the method of excavation and support, tunnelling may
induce considerably different magnitudes of deformation to the surrounding ground,
resulting as well in different ground pressures on tunnel linings. For the design of the
tunnel lining and for the prediction of surface settlements, the excavation and support
sequence needs to be taken into consideration, in order to come to a reliable conclusion
whether or not the design is adequate. In the present thesis the use of appropriate 3D and
2D FE-simulation methods for both conventionally driven open face tunnels and closed
face shield tunnels will be analyzed to demonstrate their significant influence on the
prediction of surface settlements and lining forces. Moreover, the influence of different
constitutive models will be evaluated, to show improved predictions of surface settle-
ments and horizontal ground movements when accounting for the small-strain stiffness.
Tunnel engineers traditionally use a number of elementary methods of analysis, which
comprise a large variety of empirical, simple (mostly elastic or elastoplastic) analytical
or bedded beam models for the assessment of surface settlements and lining forces. On
reviewing the literature on the aspects of elementary design methods with respect to in-
stallation procedures, one gets the impression that peculiarities of support and excava-
tion of different tunnelling methods are hardly accounted for. Nevertheless, elementary
methods of analysis are still frequently used in engineering practice and they can not be
omitted as they reflect both tunnelling tradition and design experience.
However, with the rise of computer capacity, complex numerical methods came into
the realm of design practice and tunnelling can thus be simulated more realistically. Both
non-linear ground behavior and complex geometries, such as ground layering or non-
circular tunnel cross sections can easily be accounted for. Moreover, the effects of tunnel
support installation may be incorporated, in order to arrive at appropriate loads on the
lining and realistically estimate associated surface settlements.
The transition from elementary methods of analysis to advanced numerical analysis
should not be abrupt and a sufficient validation in terms of measurements and engineer-
ing experience should be gained before bidding farewell to a well proven approach. The
present thesis is intended to contribute to the effective application of numerical analysis
of tunnelling settlements and lining forces. In order to do so the following chapters will
be considered:
1
Chapter 1 Introduction
Chapter 4 This chapter deals with the finite element modelling of tunnels. In order
to provide a better understanding it starts by introducing some basic terms and mod-
elling aspects of two- and three-dimensional FEM tunnel analyses. Factors which may
significantly decrease the accuracy of the method when not considered properly, includ-
ing amongst others the influence of the tolerated equilibrium error, mesh dimensions
and the mesh coarseness, are discussed. The importance of the magnitude and orien-
tation of initial stresses is considered in a separate section. Special attention is given
to the numerical modelling of the tunnel excavation and support sequence. Both two-
and three-dimensional models of open and closed face tunnels are studied, giving a
detailed literature overview. For closed shield tunnelling an improved numerical ex-
cavation and support procedure is proposed, being named the grout pressure method.
The two-dimensional modelling of open face tunnels is studied in detail carrying out a
parametric study on the magnitude of the so-called unloading or β-factor.
Chapter 5 The importance of the numerical modelling of the excavation and support
sequence and the influence of constitutive models is evaluated in this chapter, consid-
ering measured surface settlements and horizontal ground deformations of two tunnel
case studies. Both two- and three-dimensional FE-analyses are carried out to demon-
strate the predominant importance of the excavation and support sequence. Besides
the influence of the constitutive model is significant. The well known Mohr-Coulomb
Model, the non-linear elastoplastic Hardening Soil Model and an extension of the latter,
which includes the small-strain stiffness and is named the Hardening Soil Small Model,
are all considered to demonstrate the importance of the small-strain stiffness for the
simulation of ground deformations. In particular it will be shown that the small-strain
stiffness significantly improves the steepness of the settlement trough.
Chapter 6 In this chapter the work of the present thesis is summarized, arriving at
conclusions and giving recommendations for future research.
2
Chapter 2
Modern tunnelling methods
Introduction
There is a basic distinction between simple cut-and-cover construction methods for shal-
low tunnels, where a trench is excavated and roofed over, and underground construction
methods, which are tunnelling methods to undermine without removing the overbur-
den ground. Whereas the first category of tunnels is reduced more or less to a general
type of excavation problem, the second category is related to what is usually understood
as tunnelling in the sense of classical mining techniques (trenchless tunnelling methods).
In this thesis attention will only be paid to the latter type of tunnels. Moreover, the focus
is on shallow tunnels in soil, where depending on the method of construction it will be
generally distinguished between open face tunnelling and closed face tunnelling.
3
Chapter 2 Modern tunnelling methods
anchors anchors
shotcrete
round length
Figure 2.1: Principals of conventional tunnelling to control ground stability and ground
deformation: Systematic use of shotcrete and anchors
4
2.1 Open face tunnelling
shield tail
shield tubings
1)
jacks
2)
3)
tunnelling methods (e.g. earth pressure balance machine, slurry machine or tunnel bor-
ing machine).
Until a final lining is placed, the primary shotcrete lining has to guarantee the stabil-
ity of the ground alone. The sealing of the ground with shotcrete after each excavation
is a stepwise procedure. In a first immediate action the ground is covered with a thin
shotcrete layer to protect against rock fall. In a following step a lattice girder is applied
and finally the full shotcrete lining is sprayed. Accompanied by deformation measure-
ments the shotcrete lining will be thickened and if necessary supplemented with steel
arches.
5
Chapter 2 Modern tunnelling methods
ground
tail of shield
grouted
grout gap
injection
tubbing
jack
sealing
Tunnelling with a shield in particular is well suited for softer grounds which need
continuous radial support. The shield is mostly a cylindrical construction out of steel.
The shield has to be designed to be able to take all ground and working loads with
relatively small deformations. It is usually made of thicker steel plates at its front, to
transfer the relatively high axial working forces of the jacks from the lining to the ground.
At the shield tail, the steel is not as thick as at the shield front because only radial ground
loading has to be accounted for. The inner shield diameter is somewhat larger than
the outer diameter of the lining, enabling the installation of tubbings (precast concrete
lining segments) in cases where tunnels have to undergo curvatures. Fig. 2.2 shows
the principal working stages of shield tunnelling. Jacks installed in the shield push the
shield away from the installed lining into the soil. Depending on the length of the tubing
segments, the equal length of one sequence of tunnel advance is usually in between
0.8m-2.0m.
After each sequential tunnel advance of one segment length, the jacks are released,
giving space for a new tubing ring to be built. As shown in Fig. 2.2 tubbings are installed
inside the tail of the shield, which keeps the ground from deforming or falling into the
excavated tunnel. Fig. 2.3 shows a detailed view of the shield tail. Inside the shield
tail, grout 3 is pumped into the gap between ground and tunnel lining, to limit further
radial ground deformation. To prevent the continuous grouting to flow into the shield,
between shield tail and tubing ring a sealing is installed. The sealing consists of steel
brushes filled with grease. During tunnel advance this sealing is sliding over the tubings.
An unstable tunnel face can be improved e.g. by applying steel plates which are con-
nected to hydraulic jacks, giving a certain face pressure. Alternatively the soil at the face
may be given its natural inclination, letting it roll into the shield, but ground deformation
will be significantly larger. Underneath the ground water table open face tunnelling is
problematic and therefore the ground water table should be lowered. When tunnelling
underneath the ground water table or with larger tunnel diameters it is more efficient to
apply closed face tunnelling methods, where a shield is combined with a cutting wheel.
Closed face tunnelling methods are to be considered in the following section.
3
For shield tunnelling in rock also gravel is used to fill the gound-lining gap.
6
2.2 Closed face tunnelling
installation of
anchors
cutting wheel
shotcrete lining
gripper
jacks
7
Chapter 2 Modern tunnelling methods
Compressed air As shown in Fig. 2.5b compressed air is used to stabilize the tunnel
face. The method is mainly applied to tunnels driven underneath the ground water
table to avoid water influx. To apply an effective pressure to the soil skeleton the surface
of the tunnel face needs to have a small permeability. In soils with permeabilities of
k < 10−6 the pores are small enough to prevent the air of flowing into the soil body
(B ABENDERERDE and H OLZH ÄUSER, 2000). In soils with high permeabilities the face
can be sealed with a filter cake. If erosion takes place there is the danger of a blowout.
For this reason one tends to use EPB or slurry shilds.
Earth pressure balance Earth pressure balance (EPB) shields are most commonly used
in soft grounds. As shown in Fig. 2.5c, the excavated soil is used to apply a support
pressure to the tunnel face. Various additives are often used to ensure appropriate muck
properties. This is of particular importance for the screw conveyer, which is extracting
the soil from the chamber behind the tunnel face. The screw conveyer is controlling the
pressure at the tunnel face by its advance rate. For tunnelling underneath the ground
water table, the length of the screw conveyer has to be designed for the hydrostatical
water pressure. It should be long enough in order to reduce the water pressure to atmo-
spheric pressure.
Slurry support Slurry shields stabilize the tunnel face by applying a pressurized ben-
tonite slurry, as illustrated in Fig. 2.5d. During operation soil is mixed into the slurry
and at the end the soil is removed from the slurry in a separation plant. For the slurry
pressure it is important that a more or less impermeable mud layer is formed, the so-
called filter cake, which is sealing the tunnel face. It ensures that the slurry does not
totally flow into the ground, keeping a certain pressure on the tunnel face. To control the
slurry pressure a chamber with air pressure is connected to the slurry, as shown in Fig.
8
2.2 Closed face tunnelling
steel plates
cutting wheel cutting arm
a)
b)
earth paste
screw conveyer
c)
suspension
d)
Figure 2.5: Shield tunnelling with a) mechanical support, b) compressed air, c) earth
pressure balance and d) slurry support
9
Chapter 2 Modern tunnelling methods
2.5d. In contrast to mechanical and EPB-shields, in slurry shields the excavated soil is
pumped away together with the slurry. Bigger stones need therefore to be crushed by a
special stone crusher before they enter the slurry chamber. Slurry shields are well suited
for almost all types of soil. In particular they can be well applied to sandy soils.
10
Chapter 3
Elementary computational methods for tunnels
Introduction
When the tunnelling engineer designs a tunnel structure, he guarantees that the struc-
ture is safe with respect to structural collapse and ground deformations during its pro-
jected lifetime. Depending on ground conditions and tunnelling method he must choose
an appropriate method of analysis and derive, or even invent, a structural model, i.e.
a structural idealization. By applying equilibrium and compatibility conditions to the
model, the engineer has to arrive at those criteria that are factors in deciding whether
or not the design is safe. Different structural design methods and design models have
been developed and they are used for different excavation and support sequences, for
the preliminary and the final tunnel lining, or for different ground behavior, e.g. in dis-
continuous rock or homogeneous soil.
There is no other section of geomechanics where structural design methods have
proven so controversial and debated as it is the case for tunnel constructions. There-
fore a condensed overview of relevant computational methods for settlements and lin-
ing forces will be given in the following. Papers that applied a broader approach to all
the complex aspects of tunnelling, including different structural design methods, have
been published e.g. by C RAIG and M UIR W OOD (1978) or E INSTEIN (1979-1980).
11
Chapter 3 Elementary computational methods for tunnels
extend of surface
settlement trough
x y
z
tunnel z0
advance Smax
Figure 3.1: Tunnel induced settlement trough after ATTEWELL et al. (1986)
mass may be modelled in the same way as soil. In the present thesis the focus is on
tunnelling in soil and soft rock rather than on tunnelling in hard rock, although some
of the structural design approaches may be generally applicable. In order to address
geomaterials uniformly, regardless of whether it concerns soil or rock, in the present
thesis the word ground will be used.
12
3.4 Deformations in open face tunnelling
2 3 anchors
tunnel face 5 C
with support
4 A B
1
round length
a) b)
For shield tunnelling with adequate face support (Fig. 3.2a), the first component of
ground deformation will be relatively small, but the second component may be appre-
ciable; in particular for a somewhat conical shield or in case of over-cutting, as well if
there are steering problems in maintaining the alignment of the shield. The third com-
ponent of ground deformation can be minimized by grouting, but this component is
strongly influenced by the experience of the crew and the ground pressure control being
implemented. This third component is usually the major cause of settlements. Compo-
nent four tends to be of minor importance in relation to conventional tunnelling. Com-
ponent five can be of importance for tunnelling in soft soils with low permeabilities.
In case of insufficient face pressure the pore water pressure dissipation/consolidation
phenomenon may take place in front of the tunnel face.
13
Chapter 3 Elementary computational methods for tunnels
14
3.6 The empirical assessment of settlements
-2i -i 0 i 2i y
Vs
0.6×Svmax
Svmax point of inflection
settlement
Figure 3.3: Gaussian curve for transverse settlement trough and ground loss Vt
In addition to the settlement volume Vs one has to consider the ground loss Vt . This is
the volume of the ground that has deformed into the tunnel after the tunnel has been
constructed, as illustrated in Fig. 3.3. For tunnelling in undrained ground, the settle-
ment volume is more or less equal to the ground loss, but the settlement volume tends
to be somewhat smaller for drained excavations. Indeed, dilation and swelling due to
unloading may result in soil expansion, such that Vs < Vt (C ORDING and H ANSMIRE,
1975). However, differences tend to remain small and it can be assumed that Vs ≈ Vt . As
the ground loss depends more or less linearly on the tunnel volume, it is convenient to
15
Chapter 3 Elementary computational methods for tunnels
Soft to stiff
clays
4
Sands below
2
groundwater
level
0
0 1 2 3 4
Relative width between inflection point and tunnel axis 2i/D
Figure 3.4: Relation between settlement trough width and tunnel depth for different
grounds (P ECK, 1969)
At
Svmax ≈ √ · GLR (3.4)
i · 2π
and
At y2
Sv (y) ≈ √ · GLR · e− 2i2 . (3.5)
i · 2π
Assuming the Gaussian curve to asses the distribution of transverse surface settlements,
one needs information on two input parameters, namely the distance to the point of
inflection i for the width of the settlement trough and the ground loss ratio GLR for the
depth of the settlement trough.
16
3.6 The empirical assessment of settlements
a) b)
Figure 3.5: Observed width of surface settlement trough as a function of tunnel depth:
a) In clays, b) in sands and gravels (M AIR and TAYLOR, 1997)
depending on ground conditions, as shown in Fig. 3.4. After the suggestion by P ECK
many other authors have come up with similar relationships, e.g. C ORDING and H ANS -
MIRE (1975) or C LOUGH and S CHMIDT (1981). O’R EILLY and N EW (1982) presented
results from multiple linear regression analyses performed on field data, confirming the
strong correlation of i with tunnel depth, but showing no significant correlation of i with
tunnel diameter (except for very shallow tunnels, with a cover to diameter ratio less
than one) or method of construction. They stated, that for most practical purposes the
regression lines may be simplified to the form
i = K · z0 , (3.6)
where K is a trough width parameter, with K ≈ 0.5 for clayey grounds and K ≈ 0.25
for sandy grounds. The approach of Eq. 3.6 has been generally confirmed by R ANKIN
(1988), who presented a variety of tunnel case histories in clayey, sandy, residual and in
mixed grounds. M AIR and TAYLOR (1997) presented a large number of tunnelling data
with different linear regressions for tunnels in clays and tunnels in sands and gravels.
As shown in Fig. 3.5, the regressions confirm the findings of O’R EILLY and N EW (1982)
for clayey soils, with a trough width parameter ranging in between 0.4 and 0.6, with a
mean value of K = 0.5. However, for sandy soils they obtain a K ranging in between
0.25 and 0.45, with a mean value of 0.35, indicating somewhat wider settlement troughs.
Layered ground: Often tunnels are constructed in layered ground, including both clayey
and sandy ground layers. For tunnels in layered ground N EW and O’R EILLY (1991) pro-
17
Chapter 3 Elementary computational methods for tunnels
0.6
0.8 i = K (z0 - z)
K = increasing with depth
1.0
Figure 3.6: Variation of settlement trough width with depth of subsurface settlement pro-
files above tunnels in clays (M AIR et al., 1993)
i = K 1 · z1 + K 2 · z2 , (3.7)
where K1 is the trough width parameter for ground layer 1 with thickness z1 and K2
is the trough width parameter for ground layer 2 with thickness z2 respectively. M AIR
and TAYLOR (1997) also discussed this formula and they report that it agrees reasonably
well with field observations of tunnels in sands overlain by clay layers. However, when
sandy layers overlay clays they found Eq. 3.7 to be less evident. Nevertheless, in com-
bination with estimated ground losses it would seem that Eq. 3.7 may be used for a first
prediction of surface settlements, both for open and for closed face tunnelling.
Subsurface settlement When tunnelling in urban areas, one may have to consider the
interaction with deep foundations or existing tunnels. This leads to the need of having
information about the development of subsurface settlement profiles. M AIR et al. (1993)
analysed subsurface deformations from tunnels in clays as well as centrifuge tests in clay
(Fig. 3.6 and Fig. 3.7). They showed that subsurface deformations can also be reasonably
approximated by a Gaussian distribution. A possible extension of Eq. 3.6 for subsurface
settlement profiles is
i = K · (z0 − z), (3.8)
where z is the depth of the subsurface profile being considered. As shown in Fig. 3.6
M AIR et al. (1993) observed that the value of i for subsurface settlement profiles is sig-
nificantly larger than would be predicted with a constant K. To match the data of Fig.
3.6 and Fig. 3.7 for tunnels in clay, M AIR et al. proposed the expression
18
3.6 The empirical assessment of settlements
Figure 3.7: Variation of trough width parameter K with depth of subsurface settlement
profiles above tunnels after M AIR and TAYLOR (1997)
19
Chapter 3 Elementary computational methods for tunnels
0
0.2 0.4 0.6 0.8 1.0
LF [-]
Figure 3.8: GLR versus load factor LF in overconsolidated clay (M ACKLIN, 1999)
Several proposals are related to the stability number N , defined by B ROMS and B EN -
NERMARK (1967) as
pv − pt
N= , (3.10)
cu
for tunnelling under undrained conditions, where pv is the total overburden pressure at
tunnel axis level, pt is the tunnel face support pressure (if present) and cu is the undrained
shear strength of the ground. Here it should be noted that R USE (2004) defined the
stability number by the equation
p f = p v − c u · Nf , (3.11)
where pf is the minimum face support pressure at failure and Nf a given function of the
tunnel cover over tunnel diameter ratio H/D. In fact it would be better to refer to N as
the mobilized stability number and it should be obvious that N ≤ Nf .
For tunnels in undrained clays, C LOUGH and S CHMIDT (1981) proposed a relationship
between mobilized stability number N and ground loss ratio based on the closed form
solution for the unloading of a circular cavity in a linear elastic-perfectly plastic contin-
uum under axisymmetric conditions. According to C LOUGH and S CHMIDT for N less
than 2 the response is elastic with small ground movements and the tunnel face being
stable. For N between 2 and 4 loads increase and limited plastic yielding occurs, while
for N between 4 and 6 the yielding zone is spreading leading to larger movements. For
N greater than 6 the yielding zone is significant, leading to tunnel face instability with
large ground movements. From a mechanical point of view such findings should be gen-
eralized by considering the ratio of N/Nf rather than simply N , as Nf is not a constant
but is heavily dependent on tunnel depth.
20
3.6 The empirical assessment of settlements
horizontal disp.
(positive towards tunnel centre)
-2i -i i 2i
y
Shmax
point of inflection
Svmax
settlement
ATTEWELL et al. (1986) and U RIEL and S AGASETA (1989) presented field data of ground
loss ratios related to the mobilized stability number, based on C LOUGH and S CHMIDT’s
proposal. The results show a very wide scatter, which is probably associated with the
use of N rather than N/Nf .
Recent work of M ACKLIN (1999) on the assessment of Ground Loss Ratio (GLR) is
shown in Fig.3.8. He related measured GLR-data from different tunnelling projects in
overconsolidated clay to the load factor LF = N/Nf , which is the inverse of the factor of
safety. For LF ≥ 0.2 he proposed the linear regression
Considering the fact that measured GLR data in Fig.3.8 show a considerable scatter,
M ACKLIN emphasizes that for design purposes the range of values Fig.3.8 should be
considered, rather than just Eq. 3.12. However, ground movements are affected by a
large number of different factors and thus such relations on the assessment of GLR can
be indicative only. It would seem that Fig.3.8 ideas has not yet found its way into engi-
neering practice, but the idea of estimating settlements in relation to a factor of safety or
load factor would seem to be sound.
21
Chapter 3 Elementary computational methods for tunnels
Using Eqs. 3.1 and 3.5 in Eq. 3.13 the horizontal surface displacement can be written as
y y2 y At y2
Sh (y) = · Svmax · e− 2i2 ≈ · √ · GLR · e− 2i2 . (3.14)
z0 z0 i · 2π
Fig. 3.9 shows the distribution of the horizontal displacement together withe the Gaus-
sian settlement trough. Consistent with field observations by C ORDING and H ANSMIRE
(1975) the theoretical maximum horizontal displacement, Shmax , occurs at the point of
inflection of the settlement trough, where Sv (y) = 0.6 · Svmax . Hence Eq. 3.13 can be
written as
i
Shmax = · 0.6 · Svmax . (3.15)
z0
Using Eqs. 3.14 and 3.15 the horizontal displacement yields
Sh (y) y y2
= 1.65 · · e− 2i2 . (3.16)
Shmax i
When considering tunnel induced horizontal deformations on existing buildings, hori-
zontal strains are important. Horizontal strains may be obtained by differentiating the
horizontal displacement with respect to y.
where x is the distance from the tunnel face in the longitudinal direction of the settlement
trough, as shown in Fig. 3.1. ATTEWELL and W OODMAN (1982) have validated the
assumption of a cumulative probability function reasonably well by an examination of
several field study reports.
ATTEWELL et al. (1986) assumed that generally the settlement directly above the tun-
nel face (x = 0) coincides with 50% of the maximum settlement Svmax , as indicated in Fig.
3.10. This may be more appropriate in case of open face tunnelling. However, for closed
22
3.6 The empirical assessment of settlements
-x +x
0.25·Smax
Smax 0.5·Smax
settlement S
tunnel excavation
tunnel
face
Figure 3.10: Longitudinal settlement trough above tunnel center line after ATTEWELL
et al. (1986)
face tunnelling where significant face support is provided, settlements ahead of the tun-
nel face will reduce significantly. M AIR and TAYLOR (1997) concluded that for closed
face tunnelling much lower values of only 25% - 30% are to be obtained, which leads
to a translation of the longitudinal settlement profile as indicated by the dashed line in
Fig. 3.10. C RAIG and M UIR W OOD (1978) have reviewed shield tunnels in the United
Kingdom on this matter. They stated that the percentage of the maximum settlement
that occurs ahead of the shield, over the shield and behind the shield varies for different
grounds. In general the percentages fall into the ranges given in Table 3.1. C RAIG and
M UIR W OOD stated that in most instances 80% - 90% of the maximum settlement will
be complete when the face of the tunnel has travelled a distance equivalent to one, to
two times the depth of the tunnel past the point of observation. On the centreline of a
Table 3.1: Development of settlement profile (C RAIG and M UIR W OOD, 1978)
23
Chapter 3 Elementary computational methods for tunnels
tunnel the settlements commence at a distance from the face roughly equivalent to half
the width of the transverse settlement trough.
24
3.7 The ground response curve for loads on linings
ground pressure p´
1
ground pressure p
DR
lining
ground response curve
(convergence line)
a) tunnel contraction DR
normalized ground pressure p´/p0´
III S
lining
II
b) crown settlement S
Figure 3.11: Ground response curves after PACHER (1964) for a) deep tunnels, b) shallow
tunnels in different grounds
25
Chapter 3 Elementary computational methods for tunnels
concerns ground S
concerns lining
lining
B
b
a lining response curve
A
crown settlement S
b a
time
Figure 3.12: Ground response curve: Time dependent interaction between ground and
lining after PACHER (1964)
26
3.8 Structural design models for tunnels
By designing a tunnel engineers ”promise” that the lining will neither suffer structurally
nor collapse during its projected lifetime. Thus, models of the reality are necessary for
analysis in order to predict the behavior of a tunnel during the excavation and during its
lifetime.
Structural design models, including a variety of analytical closed form solutions and
bedded beam approaches, have been developed for the use in both conventional and
shield tunnelling. The different installation procedures of these tunnelling methods sig-
nificantly influence the magnitude and distribution of loads on tunnel linings and it has
been discussed throughout the literature, whether or not it is appropriate to incorporate
reduced primary stresses as ground loads. Special attention to loads on tunnel linings is
given in Section 3.8.3.
The excavation of a tunnel changes the primary stress field into a three-dimensional
pattern at the tunnelling face. Farther from the face, the stress field eventually will return
to an essentially two-dimensional system. Therefore, common structural design models
consider only two-dimensional stress-strain fields. Three-dimensional approaches have
been proposed e.g. by L OMBARDI (1971) or E RDMANN (1983), but it would seem that
such approaches are not widely-used in engineering practice. Instead three-dimensional
models tend to be analyzed on the basis of numerical finite element computations.
Because elementary models are easier to apply and they give qualitative insight into
the matter, they still are commonly used in engineering practice. But due to fast increas-
ing computer capacities it is not surprising that such models have been more and more
replaced by finite element approaches.
27
Chapter 3 Elementary computational methods for tunnels
sv
pr
wR pt
M
N
sh = K0 sv
28
3.8 Structural design models for tunnels
ER3
α= (3.18)
El Il
and
ER
β= , (3.19)
El Al
where El Al and El Il are the normal stiffness and the flexural rigidity of the lining respec-
tively and E is the elasticity modulus of the ground. E RDMANN (1983) supplemented the
findings of A HRENS et al. (1982) to obtain relatively simple solutions for normal forces
N and bending moments M
N = N0 + N2 and M = M2 , (3.20)
with
σv + σh
N0 = · R · n0 , (3.21)
2
and · ¸ · ¸
N2 σv − σh R · n2
= · · cos2ω. (3.22)
M2 2 R 2 · m2
Here the subscript 0 implies a constant loading and the subscript 2 indicates a non-
constant load that is changing with angle ω, as adopted in Fig. 3.13. Values of the
coefficients n0 , n2 and m2 for bending moments and normal forces are shown in Figs.
3.14 - 3.16 respectively, for full bonding, tangential slip and for different Poisson’s ratios
ν of the ground.
The coefficients n0 , n2 and m2 for bending moments and normal forces shown in Figs.
3.14 - 3.16 are obtained for a relative stiffness ratio of α/β = 3 · 103 , they increase with
decreasing relative stiffnesses α and β respectively. For tunnels with a radius of 5m one
tends to find ratios of α/β between 1000 and 5000, but such variations have little effect
on the curves in Figs. 3.14 - 3.16 (E RDMANN, 1983). Typical ratios of tunnel linings are
given in Tab. 3.2.
29
Chapter 3 Elementary computational methods for tunnels
n0
z0
z0
for
a/b = 3×10³
b
Figure 3.14: Coefficient n0 for constant part of normal force (E RDMANN, 1983)
n0
full bond
z0
for
tangential slip a/b = 3×10³
Figure 3.15: Coefficient n2 for non-constant part of normal force (E RDMANN, 1983)
30
3.8 Structural design models for tunnels
m2
full bond
z0
tangential slip
for
a/b = 3×10³
Bending moments depend linearly on the deviatoric stress (σv − σh )/2, i.e. on tunnel
depth and tunnel installation. To incorporate installation procedures the stress reduction
method may directly be applied to the stresses of Eqs. 3.21-3.22 of the analytical solution.
Tunnel installation might also be accounted for by a displacement approach, as done by
C HOU and B OBET (2002). They considered 28 shield tunnels to find values for the gap
between lining and ground ranging from 10mm to 288mm, depending on tunnel radius
and installation procedures. Information on the so-called gap might be transferred into a
stress reduction and then be applied to analytical solutions, but such an approach would
seem to be circumstantial. Instead, analytical solutions can more easily be combined
with the stress reduction method.
31
Chapter 3 Elementary computational methods for tunnels
sv = g z0
z0 2D
90°
R
ElI, ElA
sh = K0 sv
Kr = const.
sv
Figure 3.17: Example of a bedded-beam model for shallow shield tunnels in soils
(D UDDECK and E RDMANN, 1982)
proach. His approach divides the tunnel ring into 16 equal divisions with the external
ground loads combined to give 16 point loads, one acting upon each of the divisions.
The ground reaction forces are governed by spring constants. Further contributions were
given by R OZSA (1963), S ATTLER (1965), W INKLER (1970) and WAGNER et al. (1980).
Models which are still used in present engineering practice have been presented e.g.
by D UDDECK (1972). He distinguishes between shallow tunnels with z0 < 2D and deep
tunnels with z0 ≥ 3D, where D denotes the tunnel diameter. For shallow tunnels a model
as shown in Fig. 3.17 is proposed. Here bedding is only accounted for in regions, where
lining deflection is oriented outwards leading to ground compression. At the tunnel
crown, where the lining deforms inwards, tension bedding is not applied. Commonly a
non-bedded lining arc length with an angle of 90◦ - 120◦ is assumed. At the same time,
a ground load, having the magnitude of the full vertical overburden, is assumed to act
on the non-bedded tunnel crown.
According to the ITA (1988)-working group on General Approaches to the Design of Tun-
nels, a design model such as that shown in Fig. 3.17 may be particularly well-suited to
the design of linings of shallow shield-driven tunnels in soil. For deep tunnels a bedded
beam model as presented in Fig. 3.18 may be applied. Using supporting means like an-
chors the ground is well involved into the tunnel structure to carry a significant part of
the load. An idealized ground-lining ring, as shown in Fig. 3.18 is proposed. The thick-
ness of the incorporated ground ring depends on the active anchor length. This ring
needs not to be circular and non-circular cross-sections like a horse shoe profile may as
well be accounted for.
32
3.8 Structural design models for tunnels
sv
ground anchor
lining
sh = K0 sv bedding
Figure 3.18: Example of a bedded-beam model for deep tunnels (D UDDECK, 1979)
Spring constants for bedding Bedded beam approaches consider a ground behavior
according to the hypothesis by W INKLER (1867), describing the load-deformation rela-
tionship as
p = k · S, (3.23)
where p is the ground bedding pressure, S is the radial displacement of the tunnel and
k is the ground reaction modulus. This approach implies a pointwise consideration of
ground behavior (no coupling with neighbor regions).
As shown in Figs. 3.17 and 3.18, in numerical bedded beam approaches the ground
bedding is represented by a certain number of springs at regularly distributed locations
along the tunnel lining with a certain spring stiffness Kspring . These spring stiffnesses
have the correlation
Kspring = a · b · k, (3.24)
where a is the distance of springs in the transverse cross-section and b is the distance of
springs in the longitudinal cross-section (generally 1.0 per unit length of tunnel). The
spring stiffness Kspring has the unit [kN/m].
To involve radial bedding in a numerical bedded beam calculation, a radial ground
reaction modulus kr has to be determined. Tangential shear bedding may be modelled
by a separate ground reaction modulus kt with tangentially placed springs. For a circular
tunnel in elastic ground under axisymmetric loading, the ground reaction (bedding) is
only dependent on the radius of the tunnel and the ground elasticity parameters. In this
case the analytical solution for the radial ground reaction modulus yields
1 E
kr = · (3.25)
1+ν R
(e.g. K OLYMBAS, 1998), where ν is the Poisson’s ratio and E is the Young’s modulus
of the ground. Using the theory of elasticity it can be deduced that E = 2/3 · Eoed for
33
Chapter 3 Elementary computational methods for tunnels
ν = 1/3. Substituting this relation into Eq. 3.25, one obtains kr = 0.5 · Eoed /R, where
Eoed is the constrained modulus, also referred to as oedometer modulus. In practice one
often uses the similar relation
Eoed
kr = α · , (3.26)
R
but for tunnels with σh < σv α-values greater than 0.5 are often applied (K ATZENBACH,
1981). For engineering practice the factor α is difficult to assess, as kr is not a material
constant, but depends as well significantly on the geometry of the structure. In situ tests,
such as plate loading tests, although not representative for the final tunnel geometry, are
often used to estimate values of α (ITA, 1982). In German engineering practice, values
for α between 0.66 and 3.0 (M ÜLLER -S ALZBURG, 1978) have been proposed, often α = 1
is used (D UDDECK, 1980).
Advantages of bedded beam models The bedded beam model in particular is attrac-
tive to civil engineers because common rules of beam statics and the design codes of
reinforced concrete can be directly assigned. The numerical implementation of bedded
beam approaches significantly improves the handling of tunnel calculations. Uncertain-
ties such as the vertical and horizontal ground stresses, the spring constants for the bed-
ding of the lining ring or the elasticity parameters of the lining can be assessed at a
relatively low cost by varying them in additional calculations. The model can easily
be adjusted to measured data to match displacements and structural forces, such that
it can be calibrated for the design of further tunnel cross-sections. The model is ap-
plicable to tunnels in layered ground and non-circular cross-sections. For non-circular
cross-sections with corners and/or curvatures with smaller radius of the lining, bedding
may be increased at corner sections. For complex tunnel structures lining properties may
vary within a cross-section. Staged tunnel constructions may be accounted for.
34
3.8 Structural design models for tunnels
prerelaxation
DR
secondary
lining
final further relaxation
tunnel contraction DR
35
Chapter 3 Elementary computational methods for tunnels
to the surrounding ground and the secondary ground pressures are reduced down to
final pressures on the tunnel lining. For shield tunnelling the use of relatively stiff pre-
cast segmental linings (accounting for some stiffness reduction of joints between lining
segments) will generally show relatively small lining deformation but in conventional
tunnelling lining deformation may become relatively large. Fig. 3.19 shows that the
amount of the further stress relaxation resulting from tunnel lining deformation is rela-
tively small compared to the amount of the stress prerelaxation.
In order to assess structural forces in tunnel linings using a suitable structural model,
both the amount of stress prerelaxation and the amount of further stress relaxation have
to be accounted for. Analytical solutions of continuum models or bedded beam calcula-
tions incorporate lining deformations resulting from ground loading and hence the asso-
ciated further stress relaxation of the ground is automatically accounted for. In contrast
the effects of stress prerelaxation, i.e. tunnel installation procedures, are not automati-
cally accounted for and assumptions about its magnitude have to be made.
Besides stress prerelaxation the distribution of primary ground pressures is important
to be considered in tunnel analysis. In the following approaches for the distribution
of primary ground pressures and the amount of secondary ground pressures resulting
from stress prerelaxation will be briefly reviewed.
Distribution of primary ground pressures Fig. 3.20 shows a variety of different struc-
tural models with different distributions of primary ground pressures on the tunnel lin-
ing. The distribution of increasing primary horizontal pressures with depth as indicated
by Fig. 3.20a) is used for analyzing shallow tunnels, whereas constant horizontal pres-
sures as shown by Fig. 3.20b) are used for deep tunnels. The reason for reducing hor-
izontal stresses with depth as indicated in Fig. 3.20c) is not made clear in most of the
literature.
Amount of secondary ground pressures The amount of the secondary ground pres-
sure is influenced by the sum of all stress redistributions which have been caused dur-
ing tunnel excavation. Before the tunnel lining is installed, ahead of the tunnel face and
around the shield machine (or around the unsupported cut-stretch in conventional tun-
nelling) some stress redistribution inevitably takes place resulting in stress prerelaxation.
Depending on tunnel installation procedures, tunnel depth and ground properties, stress
prerelaxation may become relatively large and secondary ground pressures may reduce
significantly.
D UDDECK and E RDMANN (1982) distinguish between shallow tunnels with z0 ≥ 2D,
moderately deep tunnels with 2D ≤ z0 ≤ 3D and deep tunnels with z0 ≥ 3D. For shal-
low and moderately deep tunnels they propose that no stress prerelaxation takes place at
the crown of the tunnel, applying full primary stresses on top of the tunnel, as indicated
in Fig. 3.17. Hence, it is assumed that in the final state (some years after the construction
of the tunnel), the ground eventually will return to nearly the same condition as before
the tunnelling. Changes in ground water levels, traffic vibrations, etc., may provoke this
readjustment. Indeed, C RAIG and M UIR W OOD (1978) report that the instrumentation of
36
3.8 Structural design models for tunnels
sv sv sv
a) b) c)
existing shallow tunnels 50 to 75 years old, which have been required to be dismantled
during the construction of new works, has shown combined hoop and bending stresses
in the lining equivalent to the overburden pressure. For tunnels in sands below the
water table they state that measurements have shown combined stresses between 80%
and 100% of the equivalent overburden stress, which may develop within the first few
months.
For deep tunnels it is obvious that some stress prerelaxation needs to be accounted
for to reduce the loads on the lining. D UDDECK and E RDMANN (1982) argue that no
matter what tunnel depth, allowance should be made for a tendency towards larger or
lower ground stresses, acting on the lining in regard to, at least, cohesion, stiffness of the
ground, time to closure of the tunnel ring, excavation procedure, erection method for the
lining, time-dependent behavior of the ground and the lining and effects of groundwater.
Thus the transition from shallow to deep tunnels is not sharp and the three cases overlap.
To account for installation of closed face tunnelling, M UIR W OOD (1975) proposed to
take only 50% of the initial ground stresses into consideration. Indeed, in present two-
dimensional numerical analyses of open face tunnelling, a stress reduction factor, being
referred to as unloading or beta factor, of around 50% is commonly used, but this value
would seem to be rather low for modern closed face tunnelling (see Section 5.2). Because
of the relatively high mobilization of the grounds shear strength in open face tunnelling,
this method requires a ground with a pronounced cohesion and therefore a significant
stress prerelaxation may generally be justified.
The topic of ground pressures on tunnel linings with regard to different structural
design models, tunnel depths and ground stiffnesses has also been reviewed by the ITA
(1988)-working group on General Approaches to the Design of Tunnels. Fig. 3.21 categorizes
four different approaches of structural design models:
37
Chapter 3 Elementary computational methods for tunnels
1 2 3 4
z0
empirical
z0 approach
z0
h
sv = g zz00
sh = K0 sv
K0 = 0.5 sv = g h
sv sv sv
sv
sh sh
Figure 3.21: Plane-strain design models for different depths and ground stiffnesses (ITA,
1988)
2. continuum model for tunnels at shallow depth and moderately stiff ground,
3. continuum model for deep tunnels in stiff ground,
4. continuum model for deep tunnels, empirical approach for ground pressures.
For tunnels at shallow depth in soil, immediate support must be provided by a rela-
tively stiff lining. Here it is agreed that the three-dimensional stress release at the face
of the tunnel during excavation may be neglected. Therefore in cases (1) and (2) of Fig.
3.21 no stress prerelaxation is taken into account incorporating full primary ground pres-
sures.
Case (3) assumes that some stress prerelaxation is caused by deformations that oc-
cur before the lining participates. In rock or in highly cohesive soil, the ground may be
strong enough to allow a certain unsupported section at the tunnel face. Stress prere-
laxation is also assumed for tunnels having a high overburden, and a reduction of the
acting crown pressure (as represented in Fig. 3.21 by h < z0 ) is taken into account. Con-
firming these recommendations, C RAIG and M UIR W OOD (1978) discuss measurements
of tunnels in rock, where readings have been taken of the stresses in the arch ribs prior
to the casting of a cast in-situ lining. Their presented results generally show relatively
low stresses.
38
3.8 Structural design models for tunnels
In case (4), the ground stresses acting on the lining are determined by an empirical
approach, which may be based on previous experiences with the same ground and the
same tunnelling method, on in-situ observations and monitoring of initial tunnel sec-
tions, on interpretation of the observed data and on continuous improvements of the
design model. Here, some reduction of stresses may generally be incorporated.
39
Chapter 4
Finite element modelling of tunnels
Introduction
Starting with the 1960ies the last forty years have lead to a significant development and
advance in the application of numerical methods to tunnelling. Whereas in the begin-
ning of its development, numerical analysis as a design tool was often criticized, nowa-
days the increase of computer capacity has caused a revolution within the field of tun-
nelling. There are no significant tunnelling projects any more, which are carried out
without the support of full numerical analyses. No doubt, simlified methods as dis-
cussed in the previous section still play an important role and they can not be omitted,
as they reflect both tunnelling tradition and experience. But the days are gone in which
tunnel design was based on experience, intuition and analytical solutions of simple con-
tinuum models alone. Todays tunnelling engineers are provided with a wide range
of various modern numerical tools: Finite Element Method, Finite Difference Method,
Boundary Element Method, Discrete Element Method, etc. Cumbersome data input and
viewing of calculation results may soon be remembered as a thing of the past, as mod-
ern user-friendly data pre- and post-processing tools are being developed. Automatic
mesh generation and colored output graphs make such calculations even more attrac-
tive to the engineer. Thanks to powerful computer capacity and user friendly software,
numerical analyses that once took weeks are being performed within a few days and
in future within a few hours. The advantages of numerical analysis are obvious. Both
complex material behavior and boundary conditions can be taken into account, whilst
parameteric studies to improve the design can be easily carried out.
But advantages of numerical analyses are not achieved without their costs. The proper
use of numerical tools requires sufficient background knowledge, not only in geotech-
nical engineering, but also a good basic understanding of the numerical method itself.
Unlike analytical solutions, results from numerical analyses often can be hardly verified
and only good engineering judgement may estimate whether or not results can be be-
lieved as plausible. On the other hand, the need for a high specialization in numerical
analysis is often a communication hinderance between designers and analytical model-
ers. When looking at design practice, the role of numerical analysis seems to be more
controversial than ever. Numerical analysis has found itself torn between the temptation
to quantify and dimension every single support measure and the fear of being held for
nothing but a number crunching exercise. Being so, the success of such calculations is
often judged, depending on whether you are a protagonist or an antagonist of numerical
41
Chapter 4 Finite element modelling of tunnels
a) b)
analysis.
For a condensed overview of numerical methods applicable to geotechnical problems
the reader is referred to S CHWEIGER (1995). The most relevant numerical method for tun-
nelling applications is the Finite Element Method (FEM). The method has been presented
throughout the literature and detailed descriptions are available e.g. by Z IENKIEWICZ
and TAYLOR (1991) or B ATHE (1982). In the following a contribution will be given to the
modelling of tunnels with the help of the FEM. The present thesis is intended to discuss
some of the points which are of major importance for FE-analysis of tunnelling settle-
ments and lining forces. Besides the consideration of different constitutive models for
use in FEM as explained in Appendix A, the influence of the tolerated equilibrium error
in numerical analysis, the influence of FE-mesh dimensions and the mesh coarseness as
well as the modelling of initial ground stresses will be analyzed in this section. Here-
after the focus will be on both three- and two-dimensional FE-installation procedures
for conventional driven tunnels and shield tunnelling. All results presented have been
obtained by using the two- and three-dimensional versions of the FE-code P LAXIS.
42
4.1 Basic terms and modelling aspects of FEM tunnel analysis
uy = free t=0
ux = 0 s = free
uy = free N = free
fz = 0 M = free
ux = 0
y uy = 0 s = free
ux = 0
t = free
x
Figure 4.2: Boundary conditions of bottom, surface and vertical boundaries of symmet-
rical half
43
Chapter 4 Finite element modelling of tunnels
44
4.1 Basic terms and modelling aspects of FEM tunnel analysis
2m
2 Smax
3
4
steady-state
5
d = 2m
above the steady-state solution. In both cases the disturbance extends over a consider-
able length of about 35m. Towards the left model boundaries, settlements become too
large because of the influence of the vertical fixities which are sliding contact bearings
(compare Section 4.1.2). The significant initial disturbance implies that one has to use
large mesh dimensions to simulate tunnel excavations over a considerable length with
many excavation phases, in order to arrive at a reliable steady-state solution. The use of
sufficient FE-mesh dimensions for the steady-state solution will be discussed in Section
4.2.
45
Chapter 4 Finite element modelling of tunnels
-1800
normal force [kN/m]
N+
-1400
-1000
-600
-200
0
0 20 40 60 80 100
Figure 4.5: Zigzagging of normal forces in the direction of the tunnel axis
-10 M+
-20
-30
-40
-50
Figure 4.6: Zigzagging of bending moments in the direction of the tunnel axis
46
4.2 On FE-mesh dimensions
Just like the normal forces the bending moments show a zigzagging pattern
that matches the step-by-step installation with d = 2m. For convenience, the focus is
on the average value as indicated by the solid line. Near the tunnel heading, vanishing
small bending moments of about −3kN m/m are found. However with the advance of
the tunnel face, the bending moment in Fig. 4.6 reaches an average steady-state value of
about −22kN m/m. Again on the left in Fig. 4.6, the lining is more heavily loaded up to
−43kN m/m, indicating the significant disturbance of the mesh boundary.
47
Chapter 4 Finite element modelling of tunnels
3) × D from the tunnel center point to the bottom boundary. B LIEHM (2001) considers
a different criterion relating the choice of mesh dimensions to the magnitude of strains.
According to his recommendation the dimensions should be chosen such that the strains
perpendicular to the mesh boundaries do not exceed 0.005% after construction of the
tunnel.
In the present thesis, two different criteria for the vertical and the bottom boundaries
respectively were used. For the bottom boundary the magnitude of stress rotation was
considered, which is a criterion almost independent of the constitutive model being
used. The dimension of the bottom boundary were taken such that the maximum ro-
tation of primary stresses at the bottom boundary did not exceed 2.5◦ after construction
of the tunnel. For the dimension of the vertical boundaries, however, the magnitude
of surface settlements was considered, which is directly influenced by the constitutive
model being used 2 . The dimension of the mesh width and mesh length were taken
such that the maximum settlement at the boundary did not exceed 1% of the maximum
centerline settlement. In the following a condensed overview on the results of both 2D
and 3D mesh variations will be presented. For more details on the analyses of sufficient
mesh dimensions the reader is referred to M ÖLLER (2006).
Bottom boundary The results obtained for the bottom boundary h can be summarized
as:
h = (1.3 − 2.2) × D f or D = 4m − 12m, (4.1)
comparing well to the recommendations of (2 − 3) × D from the tunnel center point to
the bottom boundary, i.e. h = (1.5 − 2.5) × D, as given by M EISSNER (1996).
Mesh width After the evaluation of the bottom mesh dimension these results were
incorporated into the mesh variations for the evaluation of sufficient mesh widths w.
The results obtained for the mesh width w can be well approximated by the equation
µ ¶
H
w = 2D 1 + . (4.2)
D
2
For the present studies the elastoplastic HS-Model (see Appendix A.2) has been used
48
4.2 On FE-mesh dimensions
The strong correlation with the ratio H/D is logical: the deeper the tunnel the wider
the surface settlement trough and vice versa. Compared to the recommendations of
w = (4−5)×D by M EISSNER (1996), the present criterion leads to considerably wider FE-
meshes for ratios H/D ≥ 1.5. The indication of Eq. 4.2 that surface settlement troughs
will become very wide for very deep tunnels may need a further consideration. The
criterion that the boundary settlement should not exceed 1% of the maximum center
line settlement may not be required for relatively wide settlement troughs of relatively
deep tunnels, as here the magnitudes of resulting surface settlements will generally be
relatively small. Therefore an upper bound of the mesh width w when approaching
relatively deep tunnels might be considered.
Bottom boundary The results of the analyses for the bottom boundary h of 3D analyses
can be expressed as:
Mesh width The results for the three-dimensional mesh width are almost the same as
for the two-dimensional analyses, being again well described by the equation
µ ¶
H
w = 2D 1 + . (4.4)
D
Mesh length The results of the mesh length l which were obtained are almost inde-
pendent of the round length. But as for the analyses of the mesh width they show a
strong correlation to the cover to diameter ratio, which can be well approximated by the
equation
µ ¶
11 H
l = D 13 + · . (4.5)
3 D
49
Chapter 4 Finite element modelling of tunnels
50
4.3 The influence of the mesh coarseness
8D
2D
35 elements D = 8m
1.5 D
3D mesh coarseness
2D mesh coarseness
2D: 176 elements
16,25 D
round lenght: d = 2m
51
Chapter 4 Finite element modelling of tunnels
Smax [cm]
Smax
d = 2m
0
0 1 2 4
element slices / round length
Table 4.1: Ground parameters of the HS Model (see AppendixA.2) as used to study mesh
coarseness
4.1. The lining was modelled weightless and linearly elastic using El A = 3.75GN , El I =
19.53M N m2 and νl = 0.15.
Surface settlements Fig. 4.8 shows that the analysis with one element per round length
slightly underestimates the settlement, giving a maximum surface settlement of Smax =
3.7cm. The mesh refinements with two and four elements per round length result in a
settlement increase of around 10% giving Smax = 4.1cm. Fig. 4.8 is clearly indicating that
the mesh dependency is becoming vanishingly small beyond a number of two elements
per round length.
Structural forces Fig. 4.9 presents results of normal forces for different mesh coarse-
ness for a local tunnel stretch of two round lengths, showing the zigzagging pattern
which results from step-by-step installation (compare Section 4.1.4), with high forces at
the front and lower forces at the rear of a lining ring. As indicated by the curves of
Fig. 4.9, each lining shell element contributes two Gaussian integration points. Whereas
the coarse mesh results in two Gaussian integration points per round length, the finer
meshes comprise six and eight Gaussian integration points respectively.
The finer meshes of 2 and 4 elements per round length appear to give considerably
larger maximum normal forces at the front of a lining ring than the coarse mesh with
a single element per round length. Compared to the coarsest mesh, the increase of the
52
4.3 The influence of the mesh coarseness
24 25 26 27 28
-200
-600 N+
-1400 1
2
round length 4
-1800
d = 2m
rear
rear
front
Figure 4.9: Influence of 3D mesh coarseness on distribution of normal forces. Normal front
forces are shown in the Gaussian integration points of the shell elements,
being used for the lining.
Nfront [kN/m]
18
1800
N
16
d = 2m
M
1600
N+
14
1400
0 1 2 4
element slices / round length
53
Chapter 4 Finite element modelling of tunnels
h4
sv < g×h4
h3
sv = g×h1 h1 sv = S gi×hi h1 smax > g×h3
h2
sv > g×h2
a) b) c)
normal force is 26%, as indicated in Fig. 4.10. Fig. 4.9 shows that the results for the coarse
meshes produce more or less the correct forces in the integration points considered, but
one would have to extrapolate from these data the extremes at the joint between two
adjacent round lengths. For the bending moment a similar trend is observed, giving an
increase of 28% towards the joint between two adjacent round lengths.
K0 -procedure Fig. 4.11 shows magnitudes and possible orientations of initial stresses.
The ground shown in Fig. 4.11a is homogeneous with a horizontal ground surface. In
such a case, the effective initial ground stresses around a tunnel may be assumed to
54
4.4 On the initial stress
σv0 = σv − u = γ · h − u (4.6)
σh0 = K0 · σv0 , (4.7)
where σv0 is the effective vertical stress in depth h, σv is the corresponding total stress,
u is the pore water pressure, γ is the unit ground weight, σh0 is the effective horizontal
stress and K0 is the coefficient of lateral earth pressure at rest. The use of Eqs. 4.6 and 4.7
is often referred to as K0 -procedure. It is not a real numerical calculation phase because
it is used to prescribe initial stresses as a starting point for numerical analysis, i.e., no
deformations are computed.
For tunnels in ground water the effective stresses σv0 have to be considered for a de-
formation analysis by subtracting the pore water pressures u from the total stresses.
For tunnels which are constructed above the ground water table and are governed by
drained ground behavior, the pore water pressures are zero and the effective stresses
equal the total ones.
Many geological profiles will have a horizontally layered ground as shown in Fig.
4.11b. Here Eq. 4.6 may be modified adopting a summation of vertical stresses of the
respective ground layers i to arrive at the equation
X
σv0 = γi · ∆hi − ui . (4.8)
i
Gravity loading Fig. 4.11c shows magnitudes and orientations of initial stresses in a
ground with non-horizontal ground surface. In such a case, initial stresses around a tun-
nel can not be described by the K0 -procedure, as they may differ considerably within
different regions around the tunnel. Indeed, the vertical stress in depth h2 may be sig-
nificantly larger than assumed by Eq. 4.6, whereas the vertical stress in depth h4 may be
significantly smaller. The maximum stress at depth h3 is not even coinciding with the
vertical direction and can be considerably larger than γ · h3 (see also M ÜLLER -S ALZBURG
(1978)).
The numerical procedure commonly adopted to compute initial stresses for non-hori-
zontal ground situations may be referred to as the gravity loading method. Starting from
zero stresses in this method, initial stresses are computed by applying the ground self
weight γ in a calculation phase.
55
Chapter 4 Finite element modelling of tunnels
sv
smax B
K0NC
OCR - 1 s
1 1 OCR
max
nur
(1 - nur)
sv0 C
1 1 s
max
OCR
A K0
OC
sh
Figure 4.12: Relationship between K0OC and OCR for overconsolidated grounds
(B RINKGREVE and V ERMEER, 2001)
sidering a friction angle of ϕ0 = 30◦ , as an average value for many grounds, V ERMEER
56
4.4 On the initial stress
Another approach for assessing K0OC -values is shown in Fig. 4.12. Here the initial state
of stress of overconsolidated grounds is achieved by considering the stress history of
overconsolidated grounds. As shown in Fig. 4.12, it is assumed that the loading up to the
maximum effective vertical stress σmax
0
was a one-dimensional compression with σh0 =
NC
K0 · σv . Starting from point A the stress ratio of the vertical and the horizontal stresses
0
up to point B is taken according to Eq. 4.7 and K0 for normally consolidated grounds.
From point B to point C the ground is unloaded from σmax 0
down to σv00
according to the
magnitude of OCR. While unloading it is assumed that the ground behaves linearly
elastic, using Eq. 4.7 and K0 = ν/(1 − ν) with a Possons’s ratio νur for unloading-
reloading (the indices ur stand for unloading-reloading). The resulting inclination from
point A to point B can then be described by the coefficient
σh0 νur
K0OC = = (1 − sin ϕ0 ) · OCR − (OCR − 1) · . (4.11)
σv0 1 − νur
Fig. 4.13 shows a comparison between Eqs. 4.9, 4.10 and 4.11 for the formulation of K0OC
considering OCR and different friction angles. The maximum K0OC -value is governed
by the Mohr-Coulomb failure criterion (Eqs. A.1-A.3), which is incorporated in all con-
stitutive models explained in Appendix A. While Eqs. 4.9 and 4.10 give a non-linear
relation, the graph of Eq. 4.11 is a straight line. For a Poisson’s ratio of νur = 0.2 and
a friction angle of ϕ = 35◦ resulting K0OC -values of all three equations compare well to
each other. However, for a friction angle of ϕ = 25◦ and larger OCR the resulting K0OC -
values of Eq. 4.11 are somewhat larger than the ones predicted by Eq. 4.10. Eq. 4.11
appears to give values that lie in between the results of Eq. 4.11 and Eq. 4.10. Even such
slight variations of the coefficient of lateral pressure are of considerable importance for
tunnelling settlements and lining forces as will be shown in the following section.
57
Chapter 4 Finite element modelling of tunnels
OCR
MC-failure criterion (c = 0)
limitation of K0 by
2 j = 35°
1
0 1 2 3 4
K0OC
3
OCR
MC-failure criterion (c = 0)
limitation of K0 by Eq. 4.9
2 Eq. 4.10
Eq. 4.11, nur = 0.2
j = 25°
1
0 1 2 3
OC
K0
Figure 4.13: Formulation of K0OC : comparison between Eqs. 4.9, 4.10 and 4.11
by G UILLOUX et al. (1998) and G UEDES and S ANTOS P EREIRA (2002). Many computa-
tions performed involve two-dimensional analyses. Several authors have noticed that
the settlement trough obtained from such analyses is too wide and too flat compared to
measured data in particular when large K0 -values are used for the initial stresses.
G UNN (1993) modelled tunnel construction in London Clay, involving K0 = 1, and
different soil models. First of all he used the elastic perfectly plastic Tresca model and
secondly a non-linear elastic perfectly plastic Tresca model, which accounts for the small-
strain stiffness of soils. He carried out two-dimensional computations and compared his
results to the Gaussian distribution curve. Although the ground loss ratios, which he
obtained from his numerical analyses, were quite close to common values for tunnels
in London Clay, he obtained much wider settlement curves by a factor of two, when
compared to a Gaussian curve with the same ground loss ratio. It would seem, however,
that such wide predicted settlement troughs have also been caused by the use of elastic-
ity models, which tend to cause too large swelling in the ground layers underneath the
tunnel resulting in too large settlement troughs (compare Section 5).
A DDENBROOKE et al. (1997) presented a series of two-dimensional FE analyses on
the construction of the Jubilee Line Extension at St James’s Park in London. They used
linear-elastic and non-linear-elastic soil models with a Mohr-Coulomb yield surface. In-
serting K0 = 1.5 into their analyses they concluded that the computed surface settlement
58
4.4 On the initial stress
trough is too wide when using appropriate soil parameters for London Clay.
It has been suggested that 3D effects could account for this discrepancy. Tunnel con-
struction clearly is a three-dimensional problem and one would thus expect that 3D FE-
analysis improves the surface settlement predictions, when compared to 2D modelling.
L EE and N G (2002) adopted a linear-elastic perfectly-plastic soil model to compare re-
sults of 3D analysis to the results of A DDENBROOKE et al. (1997). Besides a variation of
the soils anisotropy they also carried out a variation of K0 . Although the tunnel diameter
and the tunnel depth were different in the two studies, L EE and N G arrive at the con-
clusion that the shape of the surface settlement curve improves in the three-dimensional
FE-analysis.
This statement, however, is in contrast to findings of other authors. G UEDES and S AN -
TOS P EREIRA (2002) presented results of two- and three-dimensional FE-analyses, adopt-
ing an elastic soil model and a linearly with depth increasing Young’s modulus. Com-
paring results of the transverse surface settlement trough they show that for K0 = 0.5
and K0 = 1.0 both 2D and 3D analyses give almost the same settlement shape. In the
present thesis similar results will be shown. For lining forces they showed that with in-
creasing K0 -values the normal forces in the lining were also increasing, whereas bending
moments obviously decreased for a variation from K0 = 0.5 up to K0 = 1.0, to approach
a value being very close to zero. The two-dimensional analyses they performed showed
a slight deviation from the three-dimensional ones for normal forces and good agree-
ment to 3D results of bending moments. Again a similar trend will be shown in the
present thesis.
D OLE ŽALOV Á (2002) compared results of numerical analyses to measurements of the
Mrazovka Exploratory Gallery in Prague, using the Mohr-Coulomb Model and a non-
linear elastoplastic model. Considering different K0 -values of 0.5, 1.0 and 1.5 she con-
cluded that only the results of K0 = 0.5 were satisfactory when compared to the shape
of the measured surface settlement trough. She stated that these findings were practi-
cally independent of the type of analysis, observing no difference between 2D and 3D
computations.
F RANZIUS et al. (2005) carried out two- and three-dimensional FE-analysis for the Ju-
bilee Line Extension at St James’s Park, adopting both isotropic and anisotropic non-
linear-elastic perfectly plastic soil models, including a formulation for the small-strain
stiffness and Mohr-Coulomb’s failure criterion. Comparing their results of surface set-
tlements to field data, neither the 2D analyses nor the 3D analyses were steep enough to
match the measurements. The difference observed between 2D and 3D computations of
the transverse settlement trough was negligible.
In the present studies the non-linear elastoplastic Hardening Soil Model (see Ap-
pendix A.2) was applied, to investigate the influence of K0 on surface settlements and
lining forces both for 2D and 3D analyses of conventional driven open face tunnels.
Whereas the 2D analyses were carried out using the stress reduction method (see Sec-
tion 4.6.1.3) the 3D analyses were performed according to the installation scheme of the
step-by-step method (see Section 4.5.1) with a round length of d = 1.5m.
It will be shown that the shape of the transverse settlement trough of a 2D analy-
sis compares well to the one of a 3D analysis, at least when using appropriate unloading
59
Chapter 4 Finite element modelling of tunnels
8D
2D
D = 8m
1.5 D
18.75 D
Table 4.2: Ground parameters of the HS Model (see Appendix A.2) as used to study K0
factors (compare Section 4.8). Moreover, it will be shown that the settlement shape of the
two-dimensional analyses depends significantly on the magnitude of the unloading fac-
tor for K0 > 1. The results of the analyses will show (when choosing a common unload-
ing factor of 0.5) a considerable difference between the two- and the three-dimensional
analyses. Beyond the consideration of surface settlements the influence of K0 on struc-
tural forces will be analyzed.
In order to study the influence of K0 on the results of tunnel analysis, the two-dimen-
sional and three-dimensional FE-mesh dimensions as shown in Fig. 4.14 have been con-
sidered. The ground parameters of the HS Model have been taken as presented in Table
4.2 and the weightless circular lining was modelled linearly elastic using El A = 3.75GN ,
El I = 19.53M N m2 and νl = 0.15.
Shape of settlement trough Fig. 4.15 shows results of three-dimensional analyses after
76.5m of step-by-step tunnel excavation. The different longitudinal surface settlement
curves have been obtained for K0 = 0.5, 1.0, 1.25, 1.5 and 2.0 respectively. A tremen-
dous influence of K0 on the magnitude of the maximum steady-state settlement is ob-
served. The analysis with K0 = 0.5 shows the largest settlement of 2.3cm and consider-
ably smaller settlements of 1.6cm and 0.3cm are obtained from K0 = 1.0 and 1.25. The
analyses with K0 = 1.5 and 2.0 give heaves of 0.3cm and 1.6cm instead of settlements.
Fig. 4.16 shows the normalized longitudinal settlement curves to demonstrate the
influence of K0 on the shape. As indicated by the vertical axis they have been normalized
by the maximum centerline settlement Smax . The graphs show that K0 = 0.5 and 1.0 give
60
4.4 On the initial stress
-2
K0 = 2.0
K0 > 1.0
-1
K0 = 1.5
0
K0 = 1.25
settlement [cm]
1
K0 = 1.0 K0 < 1.0
2
K0 = 0.5
Figure 4.15: Surface settlement above tunnel axis for different K0 -values
-1
K0 = 2.0
K0 = 1.5
0
K0 = 1.25
K0 = 1.0
K0 = 0.5
S / Smax
61
Chapter 4 Finite element modelling of tunnels
the steepest settlement curves, showing relatively little influence of K0 . However, K0 >
1.0 significantly decreases the steepness, as demonstrated by the curve for K0 = 1.25.
Even for the heave curves this effect is further continued, giving a less steep curve for
K0 = 2.0 than for 1.5.
At first the effect of heave instead of settlements may seem somewhat peculiar and it
may be doubted, whether or not it is realistic. Indeed, additional computations with the
HS-Small Model (for model formulation see Appendix A.3) confirmed the phenomenon.
Using Gref0 = 100M pa and γ0.7 = 2 · 10−4 in addition to the parameters from Tab. 4.2,
the small-strain analysis still yields surface heave. But this time the maximum centerline
heave was significantly smaller, giving only 0.7cm instead of 1.6cm. In the present anal-
yses OCR = 1 for normally consolidated grounds was considered and another analysis
was carried out to investigate the influence of OCR on the mechanism of surface heave.
Increasing OCR up to a value of five, it appeared to have almost no influence at all.
This is reasonable because the ground around the tunnel is unloading and the mechan-
ical mechanism is predominantly a function of the soils elasticity rather than a function
of plasticity associated with material hardening. However, to the authors knowledge in
practice so far no surface heave has been measured according to tunnels in grounds with
large K0 -values and the effect of heave is a topic of further research.
From a mechanical point of view surface heave is logical. With K0 larger than one the
major principal stress is no longer in the vertical direction but in the horizontal one. As
indicated in the insert of Fig. 4.15 for K0 < 1.0, the tunnel will be mainly horizontally
ovalized, whereas for K0 > 1.0 it changes to a vertical ovalization. Fig. 4.17 shows
displacements from two-dimensional analyses for K0 = 0.5 and 2.0. In the first case
the total displacements indicate that the ground moves into the crown of the tunnel
pushing some ground away sideways at the tunnel wall. This type of mechanism is
what is causing a surface settlement trough. In the latter case the total displacements
show that the ground is moving inwards towards the tunnel wall, leading also to some
ground heave above the tunnel crown with related surface heave. Considering these
mechanisms it also becomes clear why the transition from small K0 -values to larger ones
is causing less deep and wider settlement troughs.
Fig. 4.18 shows transverse settlement troughs from 3D and 2D analyses for different
K0 -values. The trough becomes shallower and wider with increasing horizontal initial
stress. For K0 = 1.5 and 2.0 again surface heave is observed but the transverse heave
curves show that heave is primarily concentrated over the tunnel center line. For K0 =
1.5 only the center part of the curve is giving heave and at some distance from the center
line settlement is observed again.
The two-dimensional analyses shown in Fig. 4.18 have been carried out with different
unloading factors (see Section 4.6.1.3), as also indicated in this figure, in order to match
the maximum settlement of the three-dimensional analyses. The conclusion given by
L EE and N G (2002) that larger K0 -values would give steeper three-dimensional settle-
ment curves when compared to the ones of two-dimensional analysis, is not confirmed
by the present studies. On the contrary, a slightly steeper 3D-curve is only observed for
the smallest K0 -value of 0.5. For K0 = 1.0 and 1.25, the three-dimensional analyses are
62
4.4 On the initial stress
K0 = 0.5 K0 = 2.0
3D
2D -1
K0 = 1.5
b = 0.46
K0 = 1.25
b = 0.43
settlement [cm]
K0 = 1.0
b = 0.41
2
K0 = 0.5
b = 0.66
Figure 4.18: Transverse settlement troughs from 2D and 3D steady state analyses for dif-
ferent K0 -values
63
Chapter 4 Finite element modelling of tunnels
S / Smax
S / Smax
2D Smax = 4.7 cm 2D Smax = 1.0 cm
1 1
K0 = 0.5 K0 = 1.0
3D
3D Smax = 0.4 cm 2D
S / Smax
b = 0.5
2D Smax = 0.2 cm
1
K0 = 1.25
well matched by the two-dimensional ones. For K0 = 1.5 and 2.0, the heave curves of
the two-dimensional analyses become significantly steeper than those from the three-
dimensional ones.
Fig. 4.19 shows results of normalized settlement troughs of both 3D and 2D analyses,
using a constant unloading factor of 0.5. The analyses with K0 = 0.5 and 1.0 give three-
dimensional curves which are slightly steeper than the two-dimensional ones, but still
are well matched. For K0 = 1.25 the shape of the 2D curve does not match the results of
the 3D analysis at all and obviously for K0 > 1.0 and β = 0.5 the two-dimensional analy-
sis is no longer predicting a Gaussian shape. Instead, a similar shape as observed for the
heave curves in Fig. 4.18 is computed. Apart from the fact that a constant unloading fac-
tor of 0.5 does not match the magnitude of the maximum three-dimensional settlement
value, the present data show that the shape of the two-dimensional settlement analysis
is affected by the magnitude of the unloading factor as well.
Structural forces In the following results of structural forces from shallow tunnels will
be shown, knowing that they are of relatively low magnitudes and thus of minor im-
portance. Nevertheless, results will be discussed in detail to point out the mechanical
effect of K0 on tunnel linings. Fig. 4.20 shows results of bending moments from three-
64
4.4 On the initial stress
K0 = 0.5
20
K0 = 2.0 K0 = 1.0
K0 = 1.5
K0 = 1.5
0
K0 = 2.0
K0 = 1.0
-20 M+
K0 = 0.5
-40
-60
M [kNm/m]
65
Chapter 4 Finite element modelling of tunnels
K0 = 0.5
-500
K0 = 1.0
K0 = 1.5
-1000
K0 = 1.0
K0 = 2.0
K0 = 0.5
-1500 K0 = 1.5
K0 = 2.0
N+
-2000
-2500
-3000
N [kN/m]
ring(lower rows of data points) of the normal force show a different behavior than the
rear ones. For K0 = 0.5 the normal force is about −1500kN/m, decreasing to a value of
−1250kN/m for K0 increasing to 1.0. When further increasing the K0 -value, however, the
normal force is not dropping further but is increasing to −1625kN/m and −1875kN/m
for K0 = 1.5 and 2.0 respectively.
Figs. 4.22 and 4.23 show a comparison of structural forces from three-dimensional and
two-dimensional analyses for different K0 -values. The 3D results relate to two adjacent
steady-state tunnel cross-sections, one in the front of a lining ring and one in the rear of
the same lining ring. The two-dimensional results were taken from the earlier analyses
of surface settlements, adopting the same unloading factors as presented in Fig. 4.18.
As reported earlier, the zigzagging of 3D bending moments along the tunnel longitu-
dinal axis is small, as shown by the generally small differences between front and rear
values of the three-dimensional bending moment. For K0 = 0.5 the lining is deflect-
ing inwards at the tunnel crown and invert, giving a positive bending moment of about
10kN m/m. At the tunnel wall it is deflecting outwards as indicated by the negative
bending moment of about −25kN m/m. For K0 = 1.0 the 3D bending moments obvi-
ously decrease to a relatively small negative value of around −6kN m/m. For K0 = 1.5
the 3D bending moment at the tunnel wall is almost zero and the values at the crown and
invert increase to about −15kN m/m, indicating that the crown and invert are deflecting
outwards now. For K0 = 2.0 the tendency observed for K0 = 1.5 is further increased,
giving a 3D bending moment of about −30kN m/m at the tunnel crown and invert.
The two-dimensional bending moments match the three-dimensional ones quite well
66
4.4 On the initial stress
10 2D -400
2D
0
-800
K0 = 0.5
-10 2D b = 0.66
-1200
-20
-30 -1600
M [kNm/m] N [kN/m]
M [kNm/m] N [kN/m]
3D front
3D rear
2D
M+ N+
w
90
180
Figure 4.22: Comparison of bending moments and normal forces from 3D and 2D anal-
yses for K0 = 0.5 and 1.0. Unloading factor β in 2D-analysis is chosen such
that it matches the maximum settlement from a 3D analysis.
67
Chapter 4 Finite element modelling of tunnels
0 -1000
2D
K0 = 1.5
2D b = 0.46
-10
-2000
2D
-20
M [kNm/m] N [kN/m]
20
-1000
0 K0 = 2.0 2D
2D b = 0.5 -2000
-20
2D
-40 -3000
M [kNm/m] N [kN/m]
3D front
3D rear
M+ 2D
N+
w
90
180
Figure 4.23: Comparison of bending moments and normal forces from 3D and 2D anal-
yses for K0 = 1.5 and 2.0. Unloading factor β in 2D-analysis is chosen such
that it matches the maximum settlement from a 3D analysis.
68
4.5 Review of 3D FEM installation procedures
for K0 = 0.5. For the other K0 -values of 1.0, 1.5 and 2.0, results from three-dimensional
analyses are only matched at the crown and at the invert. Around the region of the
tunnel wall, the two-dimensional analyses tend to predict considerably smaller nega-
tive bending moments or even are giving positive values. The fact that 3D bending
moments are well matched by two-dimensional analysis as reported by G UEDES and
S ANTOS P EREIRA (2002) is thus only observed for a small value of K0 = 0.5. For higher
values of K0 , however, the good agreement between 3D and 2D results of bending mo-
ments is not observed. However, bending moments are typically low and thus not as
important as normal forces.
The three-dimensional normal forces generally show a large difference of around
1000kN/m between front and rear values, as has already been demonstrated previously
by the zigzagging pattern of Fig. 4.21. For K0 = 0.5 the wall is giving the maximum nor-
mal forces of about −1600kN/m at the front and −400kN/m at the rear of a lining ring.
At the crown and the invert the normal forces are about −1000kN/m and −200kN/m at
the front and the rear respectively. For K0 = 1.0 the normal force along the cross-section
is becoming almost constant, giving about −1200kN/m at the front and −300kN/m at
the rear of a lining ring. When further increasing K0 , the normal force both at the front
and the rear is mainly increasing at the tunnel crown and invert, reaching finally values
of about −2500kN/m and −1000kN/m for K0 = 2.0.
The two-dimensional normal forces are well in between the upper and lower bounds
of the three-dimensional results, showing always a very similar shape. Whereas the 2D
solution for K0 = 0.5 is almost exactly in the middle between front and rear values, the
other analyses for K0 = 1.0, 1.5 and 2.0 tend to predict somewhat too low magnitudes,
being more close to the results of the rear values.
69
Chapter 4 Finite element modelling of tunnels
Portal
Tunnel connection
‘Trouser’
Connection with shaft
a short overview of 3D FE-installation procedures for both open face and closed face
tunnelling will be given.
70
4.5 Review of 3D FEM installation procedures
i-1 i i+1
to strokes of the tunnel boring machine this method converges to the real process of
shield tunnelling. In fact, closed shield tunnelling is a continuous process with a contin-
uous support pressure and installation of lining segments. However, for the reason of
numerical discretization a step-by-step approach needs to be adopted.
It would seem that D IJK and K AALBERG (1998) use a step-by-step approach, but in-
stead of prescribing a contraction of the lining as done by A UGARDE et al., they prescribe
a hydrostatic pressure at the tunnel face as well as directly behind the shield, whilst the
shield is modelled by elastic shell elements. H OEFSLOOT and V ERWEIJ (2005) refine this
model by prescribing in the tail void a grout pressure as a function of the distance behind
the shield. In addition they model the conicity of the shield by applying a certain contrac-
tion, but they report that ground deformations are best matched when modelling only
a stress boundary condition. Indeed, the contraction type of approach does not seem to
give accurate results, as also reported by D IJK and K AALBERG (1998) and M AIDL et al.
(2005). The latter consider the pure pressure type of simulation and they slightly simplify
the D IJK and K AALBERG model by prescribing pressures over the entire tunnel heading
including the tunnel face, the shield and some lining rings behind the shield. This would
seem to be most realistic for high tail and/or face pressures, at least for slurry shields.
71
Chapter 4 Finite element modelling of tunnels
l
i-1 i i+1
The preceding condensed literature review emphasizes the need for modelling a stress
boundary condition in order to match ground deformations. Whereas H OEFSLOOT and
V ERWEIJ (2005) model a slurry shield tunnel, D IJK and K AALBERG (1998) and M AIDL
et al. (2005) advocate a pressure boundary condition regardless of the type of shield
machine, i.e. both for earth pressure balance and slurry shields. Although only the
simulation of a slurry shield has been considered in detail by comparing results of FE-
analyses to measurements (see Section 5.2), it would seem that a pressure boundary
condition is also valid to model earth pressure balance shields.
Slurry shield The step-by-step pressure method considered in the present thesis is sim-
ilar to the pressure method by M AIDL et al. Fig. 4.26 shows details of the step-by-step
pressure method. The slurry at the face of the tunnel is simulated by a relatively high ax-
ial pressure, the shield is simulated by a somewhat lower radial pressure and the fresh
72
4.5 Review of 3D FEM installation procedures
Detail:
ground shield tail
approx.
8cm - 20cm
tail of shield gap with
grouted bentonite
gap suspension
grout
tubbing injection
Figure 4.27: Details of the shield tail of slurry shields after B ABENDERERDE (2000)
grout in the tail void behind the shield is also simulated by a radial pressure. Hence,
instead of modelling the shield by stiff shell elements a pressure controlled boundary
condition is applied. This is certainly realistic for modern machines which use injections
around the shield to control ground loss, to compensate overcutting and to reduce shield
friction (M AIDL et al., 2005).
Moreover, it would also seem realistic for slurry shields where the pressure is so large
that slurry will flow from the face to the back of the shield. Fig.4.27 shows details of
the ground-shield interaction of slurry shields. Between ground and shield there is a
steering gap due to overcutting and conicity of the shield machine. The final tail void
amounts up to approximately 8cm - 20cm, as reported by B ABENDERERDE (2000). The
filling with grout in order to compensate this final tail void is achieved by an injection
through the shield tail as illustrated in Fig.4.27. This tail void grout is partly flowing
back to correspond with the slurry pressure and hence, according to B ABENDERERDE, a
pressure condition for the entire shield applies.
In the step-by-step pressure method all prescribed pressures increase hydrostatically
with depth according to a unit weight of the slurry and the grout respectively. The radial
shield pressure is taken equal to the grout pressure. It is applied along the shield and
along two lining rings directly behind the shield, as shown in Fig.4.26. Here a ground-
lining gap is simulated by deactivating ground elements with a thickness of around
8 − 20cm (B ABENDERERDE, 2000) and a length of approximately l = 1.5m according to
the width of a lining ring. Within this gap the ground may deform until contact to the
lining is made. Hence ground displacements are controlled not to exceed the ground-
lining gap. For subsequent lining rings, the grout is assumed to be hardened and the
radial pressure is switched off. Each time radial pressures are switched off, volume
elements are activated to fill the gap, assuming linear elastic lining properties for the
hardened grout.
In reality one may expect a smooth transition from the axial face pressure to the radial
shield pressure, rather than a pressure jump, as indicated in Fig. 4.26. Similarly the grout
pressure may also be modelled with a smooth transition to the hardened grout (TALMON
and B EZUIJEN, 2005). However, in the present analysis the focus is on surface settlements
and such abrupt transitions can be accepted, as surface settlements are hardly effected.
73
Chapter 4 Finite element modelling of tunnels
In order to model structural forces in linings most accurately one might have to consider
a more detailed modelling of shield and grout pressure distributions.
During excavation the shield and the cutting wheel will impose some shear stresses
to the ground (K ASPAR and M ESCHKE, 2004). The present pressure approach slightly
simplifies the interaction between ground and shield machine by neglecting such shear
stresses, but it would seem that they have little impact on surface settlements.
To simulate excavation for slurry shields, again for numerical reasons the step-by-step
pressure approach as indicated in Fig. 4.26 is adopted. Starting from initial geostatic
stresses, the following excavation sequence is carried out. In each computational phase
i a tunnel advance of one segment length l is simulated, in which one slice of ground
elements is switched off. At the same time pressure is applied in the new tunnel slice
and in five subsequent shield slices. Behind the shield the full lining is modelled by
shell elements. The first two rings of shell elements are surrounded by a pressurized
gap, as explained earlier. Calculation phases are repeated in steps i + 1 to i + n until a
representative steady-state solution is obtained.
Earth pressure balance shield For earth pressure balance shields, one might assume
a contraction type of boundary condition along the shield to model overcutting and/or
a conical shield. Near the tail, however, it is plausible that grout will enter into the
shield-soil interface. For high grout pressures the greater part of the shield may even
be in contact with grout and a pressure boundary condition may prevail. It is felt, how-
ever, that analyses with displacement boundary in the front of the shield and pressure
boundary at the rear of the shield make the matter unnecessary complex. For the sake of
convenience EPB shields may also be modelled by using a pressure boundary condition,
but it is debatable whether or not the same boundary conditions as for slurry shields
applies.
74
4.6 Review of 2D FEM installation procedures
pg pg
pg
ps ps
p0
stress reduction
lining
method ps
DR
gap method
75
Chapter 4 Finite element modelling of tunnels
pa
a×E pa = 0
As demonstrated by Fig. 4.28b, two-dimensional analyses can not simulate the three-
dimensional arching effect. The missing 3D arch is compensated by involving an artifi-
cial support pressure ps . When reading the literature on 2D approximations for tunnel
analyses, this artificial support may not necessarily be incorporated by means of a pres-
sure approach but may as well be accounted for by a displacement approach.
Fig.4.29 illustrates tunnel construction, adopting the ground response curve. As shown
in this figure, most 2D approximations can either be classified as a stress reduction
method where the initial ground pressure p0 inside the tunnel is reduced down to a
support pressure ps , or as a displacement/gap method, where the amount of support
is controlled by simply prescribing a certain tunnel contraction ∆R or ground loss. In
the following both literature on conventional and closed shield tunnelling will be briefly
reviewed on the aspects of such 2D approximations.
76
4.6 Review of 2D FEM installation procedures
p0 pd
d×El
soil inside the tunnel is removed and the lining is activated. S CHIKORA and F INK (1982)
report realistic values of α < 0.2 for tunnels with 2 < H/D < 4. For partial excavations
without immediate closure of the lining ring they recommend values of α ranging in
between 0.3 and 0.5.
77
Chapter 4 Finite element modelling of tunnels
sb
support pressure
p0
b·p0
A
B
b×p0
ing
lin
sb stot settlement
Figure 4.32: Display of stress reduction method adopting ground response curve
often taken around 0.5. In this thesis β will simply be referred to as the unloading fac-
tor. To this point only the ground is loaded and settlements of the amount Sβ occur, as
indicated in Fig. 4.32.
In the second calculation phase the lining is installed and the remaining load β · p0
is divided over the lining and the ground. Final settlements of the amount Stot occur
in this phase due to the combined loading of ground and lining. Structural forces in
the lining occur only during this second phase. For stiff linings the remaining load will
largely go into the lining. Small β-factors correspond to large round lengths and/or late
installation of tunnel lining. In this case ground deformations will be relatively large,
whilst structural forces in the lining will be relatively low. Vice versa a larger factor β
leads to smaller ground deformations and larger structural forces in the lining. In the
later Section 4.8 an extensive parameter study on the unloading factor β will be given.
It will be shown that the stress reduction method gives reasonable structural forces and
settlements when used in combination with appropriate unloading factors.
78
4.6 Review of 2D FEM installation procedures
ground
gap
tunnel
lining
79
Chapter 4 Finite element modelling of tunnels
ground loss
support pressure
p0
pA
A
B
pA
g
in
lin
ground loss
prescribed
ground loss
Figure 4.34: Stress reduction method with ground loss control by A DDENBROOKE et al.
(1997) adopting ground response curve
The stress reduction method has also be proposed to model installation of shield tun-
nelling. As the magnitude of unloading factors are uncertain and shield tunnelling is
controlled by the amount of ground loss, unloading factors have been replaced by a
control of ground loss (A DDENBROOKE et al., 1997).
Fig.4.34 illustrates the two calculation phases of the stress reduction method with a
ground loss control, adopting the ground response curve. Starting from initial stresses,
the stresses inside the tunnel are reduced stepwise in a first calculation phase and the re-
sulting ground loss is calculated after each increment. As soon as the prescribed ground
loss is reached in point A at corresponding pressure pA , the lining is activated. Hence,
the stress reduction factor βg is chosen such that it matches a given ground loss.
A DDENBROOKE et al. (1997) assume undrained material behavior, i.e. constant vol-
ume deformation, for which the volume of the surface settlement trough is equal to the
ground loss. By doing so, they prescribe a ground loss for the monitored volume of the
settlement trough.
When the focus is on settlements and no structural forces are analyzed, the calculation
can be stopped after reaching the prescribed ground loss. For analyzing structural forces,
the load transfer to the lining has to be calculated in a subsequent calculation phase.
After placing the lining in a second calculation phase, some further ground loss will
take place until final equilibrium is reached in point B. But this additional ground loss
will be small compared to the ground loss of the first calculation phase, as lining stiffness
is considerably higher than ground stiffness.
80
4.7 Improved installation procedure for shield tunnels: The grout pressure method
ground ground
resulting
ground loss
tunnel contracted
lining tunnel
lining
81
Chapter 4 Finite element modelling of tunnels
pcrown
gap
ggrout
1 tunnel
lining
a) b)
considerably from the distribution of initial stresses. In the following the grout pressure
method will be proposed as an improved installation procedure for shield tunnels.
In the grout pressure method the lining is considered to be surrounded by a thin grout
layer with a known grout pressure. The thin grout layer may be modelled with so-called
interface elements between the ground and the lining. Within these interface elements
the normal grout stress σg is assumed to be given, e.g. with an hydrostatical increase
with depth as illustrated in Fig. 4.36a, such that the lower pressure pcrown acts at the top
of the lining. Before tunnel excavation the interface elements have the initial normal and
shear stress
σv − σh
σ0 = σh · sin2 ω + σv · cos2 ω and τ0 = · sin 2ω, (4.12)
2
where ω is a rotation angle as illustrated in the insert of Fig. 4.37. The transition of the
interface stress σ from σ0 to σg can be written as
σ = (1 − λ) · σ0 + λ · σg and τ = (1 − λ) · τ0 , (4.13)
where λ is a pseudo time parameter, which increases from zero to unity. For λ = 0 the
interface elements have ground properties and transfer the initial geostatic normal and
shear stress σ0 to the lining. Then the geostatic stresses in the interface elements are
changed to the normal grout pressure with no shear stresses by increasing λ up to 1.
For a linear elastic ground and lining this can be done in a single computational step,
but incremental procedures are required when adopting non-linear ground behavior.
Depending on the difference between σ0 and σg the interface elements will expand or
contract. The interface contraction should not exceed a prescribed value. This prescribed
value is the difference between the excavation radius and the outer radius of the lining,
i.e. the gap as indicated in Fig. 4.36b.
82
4.7 Improved installation procedure for shield tunnels: The grout pressure method
45 0 12,5 10
w z
m
10
90 15
=
D
wall 90 wall 15
20
180
135 17,5
w [°] z [m]
a) b)
sv = g × z
z
sh = K0 × sv initial radial stress Eq. 4.15
H = 10m
Figure 4.37: Difference in pressure distributions of stress reduction and grout pressure
methods. Display of radial stresses: a) as a function of rotation angle ω, b)
as a function of depth z.
Comparison between grout pressure and stress reduction methods When comparing
the stress distribution of the grout pressure method to the one of the stress reduction
method, there is a considerable difference. To describe the stress distribution of the stress
reduction method, Eq. 4.12 may be extended to
where βg is the unloading factor (compare Section 4.6.2.2) to reach a prescribed ground
loss. Hence, the normal and shear stresses have a similar distribution as the initial
stresses, but are reduced by a factor 0 < βg < 1. In contrast the grout pressure method
considers only normal stresses and no shear stresses (no bonding between ground and
lining), which is more realistic for shield tunnelling. Moreover, the grout pressure meth-
ods results in a considerably different distribution of the radial pressure on the tunnel
lining.
Fig. 4.37a) and b) shows the difference of radial stresses of the grout pressure and the
stress reduction method, displayed both as a function of the rotation angle ω and depth
83
Chapter 4 Finite element modelling of tunnels
z respectively. The radial stresses shown here are calculated for a particular tunnel with
a diameter of D = 10m and a cover to diameter ratio of H/D = 1. The unit weight
of the ground is γ = 20kN/m3 and the coefficient of lateral earth pressure was taken as
K0 = 0.5. In order to compare the two methods, the crown pressure of the grout pressure
method was taken the same as the reduced crown stress of the stress reduction method.
The unit weight of the grout pressure was taken as γgrout = 15kN/m3 . Fig. 4.37 shows
that the radial stresses of the two methods are almost the same at the tunnel invert, but
due to K0 = 0.5 the radial stress of the stress reduction method at the tunnel wall is
around two times smaller than the radial stress of the grout pressure method. Moreover,
the stress reduction method assumes an uniform reduction of initial stresses, whereas
the grout pressure method reduces initial pressures at the tunnel crown and invert, but
leads to increased radial stresses at the tunnel wall.
84
4.8 Evaluation of unloading factors for use in the stress reduction method
d = round length
β = unloading factor
D = excavation diameter
Table 4.3: Unloading factors β for full excavation of a horse shoe profile and different
round lengths d after B AUDENDISTEL (1979)
85
Chapter 4 Finite element modelling of tunnels
57m
120m
60m
explained by considering the ground response curve as shown in Fig. 4.32. The com-
plete ground response curve can be computed from a two-dimensional FE-analysis of
an unlined tunnel and the settlement Stot is obtained from a full three-dimensional FE-
analysis.
Assuming an elastic tunnel lining, it is straightforward to compute the inclination of
line B − Sβ in Fig. 4.32 and to find point B with the corresponding β-value. Indeed, for
deep tunnels with K0 = 1, there is a more or less axis symmetric state of stress around
the tunnel giving a cylindrical compression of the lining. As long as this lining behaves
elastically line B−Sβ in Fig. 4.32 will be linear. For shallow tunnels in non-linear ground,
however, circular tunnel linings deform into ellipses and line B − Sβ is no longer linear.
In such cases point B with its corresponding β-value has to be computed iteratively.
The above β-finding procedure applies to settlements. Instead of considering settle-
ments, one may also compute ground response curves for bending moments and/or
normal forces in linings. The settlement in Fig.4.32 might for instance be replaced by
the maximum bending moment or the maximum normal force, as both of them increase
when reducing the support pressure β · p0 . No doubt, all three curves will be somewhat
different and one will find three different factors: β-settlement, β-bending moment and
β-normal force. In the following results of β-values of bending moments and normal
forces will not be plotted as discrete curves but are shown as a band width for structural
forces in general.
86
4.8 Evaluation of unloading factors for use in the stress reduction method
models, which account for stress and strain-level depending stiffnesses and which dis-
tinguish between loading and unloading, computations were carried out for relatively
simple material models with a constant Young’s modulus E both for loading and un-
loading. Such simple models will give appropriate results for structural forces but sur-
face settlements will by no means be computed realistically when not accounting for the
unloading stiffness. In order to account to some extend for the unloading and the small
strain in the deep ground below the tunnel, a two-layer mesh as shown in Fig. 4.38 was
used for all analyses and the deep ground layer was taken much stiffer than the upper
layers. For the present case it was appropriate to increase the stiffness by a factor of
eight.
The lining shell elements were simulated by using a linearly elastic model. The in-
crease of the shotcrete lining stiffness El with time has been accounted for by using
a stepwise increase with excavation phases. In practice time effects of shotcrete such as
creep are commonly accounted for by reducing the final lining stiffness to 50% of the stiff-
ness of a 28 days old concrete. In Germany often a concrete C20/25 (German design code
DIN 1054-1) is used, which has a 28-days Young’s modulus of around 30, 000M N/m2 . To
account for the low stiffness of the very fresh shotcrete, usually the stiffness is once more
reduced by a factor of two.
For the three-dimensional analyses an initial stiffness of E1l = 7500M N/m2 in the first
lining ring behind the tunnel heading was applied, which was then increased to E2l =
15000M N/m2 for all subsequent lining rings. These Young’s moduli were converted
into data of the normal stiffness and the flexural rigidity to obtain E1l A = 2250M N ,
E2l A = 4500M N and E1l I = 16.875M N m2, E2l I = 33.75M N m2. In the two-dimensional
analyses no increase of lining stiffness was modelled, considering only the final flexural
rigidity E2l I and final normal stiffness E2l A.
87
Chapter 4 Finite element modelling of tunnels
500
1000
1500
2000
N
2500
3000
3500
N [kN/m]
a)
-200
-400
-600
M
-800
-1000
-1200
M [kNm/m]
b)
88
4.8 Evaluation of unloading factors for use in the stress reduction method
0.9
bending moment, normal force
d = 0.5m
0.8
N S
M
0.7
round length = 1.0m
settlement
N
0.6 S
M
d = 1.5m S
N
0.5
Figure 4.40: Influence of ground stiffness and round length for linear elastic ground
factor is decreasing with increasing round length. This seems logical as a tunnel without
a lining, i.e. d = ∞, has an unloading factor of β = 0 (no forces in the lining). Because
of different unloading factors for bending moments and normal forces a band width for
structural forces is plotted.
With increasing ground stiffness a decrease of unloading factors is observed. The β-
values for settlements always remain within the bandwidth for structural forces. The
elasticity calculations show the considerable influence of the round length on β. On
considering for example a ground stiffness of E = 20M N/m2 , for a very short round
length of d = 0.5m β = 0.85 is found, but a much lower value of β = 0.65 is found for a
long round length of d = 1.5m.
Elasticity computations give qualitative information on the β − d relationship and pos-
sibly quantitative information for tunnels in rock. Indeed, rock has a considerable shear
strength and tunnelling will be dominated by the quasi elastic rock properties. On the
other hand, soil has a relatively low shear strength and conventional tunnelling will be
dominated by plastic deformation rather than quasi elastic deformation. Therefore plas-
tic soil behavior is considered in the following section.
89
Chapter 4 Finite element modelling of tunnels
0.8
d = 1.0m
Case Study
0.6
d = 1.5m
cohesion c´ [kN/m²]
Figure 4.41: Influence of friction angle, cohesion and round length for elasto-plastic
ground
unloading factor b [-]
0,6
c´ = 100 kN/m²
N
M
0,4
c´ = 20 kN/m² S
0,2
0,5 1 1,5
90
4.8 Evaluation of unloading factors for use in the stress reduction method
was neglected by using a dilatancy angle equal to zero. The coefficient of lateral earth
pressure was taken as K0 = 0.67.
A variation of the friction angle was carried out by considering ϕ0 = 20◦ , 30◦ , and
40◦ . For this realistic range of friction angles, their influence appeared to be relatively
small. Fig.4.41 shows computed unloading factors as a function of the effective cohesion
for different round lengths. With increasing cohesion the computed β-values approach
a horizontal plateau that corresponds to the linear elastic solution for E = 20M N/m2
shown in Fig. 4.40, indicating that the analyses with c ≥ 200kN/m2 behave fully elastic.
Up to c0 = 200kN/m2 Fig.4.41 shows a significant influence of the cohesion, i.e. for
tunnels in soft to hard soils. Here, the c − β-relationship is nearly linear. Beyond an
effective cohesion of 200kN/m2 , i.e. for tunnels in rock, the cohesion plays no role at all.
The information of Fig.4.41 is also plotted in Fig.4.42, but this time β is plotted as a
function of the round length. Again one observes that the unloading factor decreases
when cohesion decreases and when the round length increases.
91
Chapter 4 Finite element modelling of tunnels
Moreover, it was found that the reduction factor decreases with decreasing cohesive
strength and increasing round length. In contrast to these findings, it was somewhat
surprising to find that the soil friction angle has relatively little influence on the load
reduction factor, but this is an issue of further research as relatively few variations were
carried out.
In the presented studies the coefficient of lateral earth pressure at rest was modelled
by the correlation K0 = 1 − sin ϕ0 . As has been discussed in Section 4.4.2, larger K0 -
values have a considerable effect on two- and three-dimensional bending moments and
thus different unloading factors for higher K0 -values would have to be considered. Such
effects will be more significant for overconsolidated ground, but this has not been con-
sidered in these studies.
In the present studies effects of tunnel depth H and tunnel diameter D have not been
discussed in detail. However, variations that were carried out on the relative tunnel
depth H/D showed that unloading factors slightly decrease with increasing H/D.
Considering load reduction factors it has been shown that it is not suitable to use one
single β-value for bending moments, normal forces and settlements. Instead one needs
to use three different values in order to compute both structural forces and settlements
precisely.
The load reduction factor for settlements of the Steinhaldenfeld tunnel with a non-
circular cross-section in layered ground perfectly matches findings for circular tunnels,
whereas load reduction factors of structural forces were found to be somewhat larger.
This makes it clear that β-values should be handled with care as they vary with tun-
nel geometries and/or ground conditions. In particular structural forces will be heavily
influenced by the geometry of the tunnel cross-section. Moreover, influences of secur-
ing means like anchors etc. or complex ground layering as adopted for the present case
study may play an important role, but they have not been evaluated. Indeed, the aim
of the present study on β-values was to give a qualitative insight into the matter, rather
than coming up with quantitative numbers. For practice, it may be advisable to per-
form at least a single 3D analysis. For further two-dimensional analyses such a three-
dimensional analysis may serve as a reference in order to assess appropriate unloading
factors.
For circular tunnels in homogeneous ground with round lengths of 0.5 - 1.5m the
settlements are well matched when using β-values between 0.3 and 0.4. For bending
moments approximately the same values are found, as long as the ground cohesion is
below 50kN/m2 . For a particular non-circular tunnel however, the bending moment
was matched with β = 0.54 and it would seem that low β-values are not appropriate
for bending moments. A conservative approach for structural forces would be to use
β-values of at least 0.5 − 0.7 in soil.
92
4.8 Evaluation of unloading factors for use in the stress reduction method
xc
x
DSi-1(c)
S DSi-1(c-d) DSi(c)
S
DSi-1(c-2d) DSi+1(c)
DSi-1(c-3d) DSi+2(c)
i-1 i i+1 n
tors for settlements, bending moments and normal forces respectively. In order to do
so, a full three-dimensional analysis is required. Even with the increase of present com-
puter capacities such three-dimensional analyses are still time consuming. When ana-
lyzing settlements separately from lining forces, e.g. for an estimate of tunnel induced
deformation to existing structures, the unloading factor for surface settlements may be
assessed in a fast way. To avoid long lasting computations as are needed to arrive at
the steady-state settlement solution in a 3D calculation, a fast way of settlement analysis
was developed. Instead of performing the time consuming method of step-by-step in-
stallation this analysis requires only two calculation phases. This way of analysis makes
it possible to arrive at the three-dimensional steady-state solution in a smart way, at least
for settlements.
The fast settlement analysis Observations from practice have shown that the distribu-
tion of the longitudinal settlement curve has the shape of an s-curve (M AIR and TAYLOR,
1997) as shown in Fig. 4.43. The shape of this curve remains constant but is translated
by one round length d during every excavation. In such a case the position of the settle-
ment curve can be related to the respective excavation step i, denoting the corresponding
surface settlement profile as Si (x), with i = 1, 2, 3, ..., n.
If settlements have reached the steady-state, Si is not further increasing. Hereafter
every additional excavation step leads to the same settlement increment. Considering
xc as a control point, located at the surface before tunnel excavation, the settlement
increments during excavations i in this point occur in the order ∆S1 (xc ), ∆S2 (xc ),...,
∆Si−1 (xc ), ∆Si (xc ), ∆Si+1 (xc ),..., ∆Sn (xc ) caused by the translating settlement curve
S1 (x), S2 (x),...,S1−1 (x), Si (x), Si+1 (x),...,Sn (x).
93
Chapter 4 Finite element modelling of tunnels
100m
55m
H = 16m
d = 2m 28m
D = 8m
Fig. 4.43 illustrates that all respective settlement increments 1, 2, ..., i − 1, i, i + 1, ..., n
at location xc can also be obtained from one single increment i − 1 at translated locations
of the order
∆Si (xc ) = ∆Si−1 (xc − d)
∆Si+1 (xc ) = ∆Si−1 (xc − 2d)
.. (4.15)
.
∆Si+n (xc ) = ∆Si−1 (xc − (n + 1) · d).
For a general use the settlement at any location x after a number of i excavations can
be obtained from the settlement increment i − 1 (or any other steady state increment
accordingly). This can be expressed as
i
X
Si (x) = [∆Si−1 (x − (k + 1) · d)] (4.16)
k=1
For performing the fast 3D analysis the steady-state settlement increment i − 1 may be
obtained from a 3D FE-analysis.
The all-in-once installation The first phase is used to install a complete tunnel. To this
end soil elements are switched off and lining elements are switched on up to the esti-
mated steady-state length of the longitudinal settlement trough. The steady-state length
is the length of the tunnel that has to be excavated in order to reach the steady state
depth of the settlement trough. The second phase is used to model a single excavation
with one unsupported round length d and all previous displacements are reset to zero.
One will now compute a more ore less circular settlement crater as indicated in Fig. 4.44.
Finally it should be noted that the all-in-once installation does not require such a fine
mesh as needed for the step-by-step analysis. When computing the settlement crater af-
ter two computational phases as described above, the fine mesh is only needed around
94
4.8 Evaluation of unloading factors for use in the stress reduction method
the tunnel heading in the middle of the mesh block. Towards the mesh boundaries the
element size can be gradually increased.
In order to obtain results from the all-in-once installation, the circular tunnel as shown
in Fig.4.44 was analyzed. To simulate the settlement trough of this tunnel a block of
100m × 55m × 28m was divided into 4300 volume elements with a total of 13151 nodes.
The parameters for the Mohr-Coulomb Model were taken E = 42M P a, ν = 0.25, c0 =
20kP a, ϕ0 = 20◦ , ψ = 0 and K0 = 1 − sin ϕ0 . A circular tunnel with a diameter of
D = 8m and a cover of H = 16m was modelled in a symmetric half with an unsupported
excavation length of d = 2m. The shotcrete lining had a thickness of 30cm, a Young’s
Modulus of E = 20GP a and a Poisson’s ratio of νl = 0. The steady state settlement for
the circular tunnel computed from the fast 3D analysis was S = 3.92cm which compares
well to the finding of S = 3.81cm from a full 3D step-by-step installation including a
difference of only 2.9%. The step-by-step installation (Section 4.5.1) may be used, to
proof Eqs. 4.16 and 4.17.
Fig. 4.45 shows results of steady-state analysis, using the three-dimensional step-by-
step installation. Settlement curves have been obtained from a sufficiently large 3D mesh
and it has been checked, whether or not steady-state was reached. Independent of the
constitutive model being used in the analysis, steady-state should be reached after a
certain number of excavations. For higher order constitutive models, which adopt non-
linear ground behavior associated with a relatively high portion of plasticity, steady-
state might need considerably more excavation phases than e.g. linear elasticity or the
Mohr-Coulomb Model.
After reaching steady-state in the present analysis, all displacements were reset to
zero. The next excavation phase resulted in a single settlement increment, having the
95
Chapter 4 Finite element modelling of tunnels
96
Chapter 5
Tunnel case studies
Introduction
This chapter describes two case studies. The first case study considers a conventionally
driven tunnel and the second case study analyzes a slurry shield tunnel. The particu-
lar importance of installation procedures will be highlighted, whilst constitutive models
also play an important role. Conclusions will be given by comparing results of numeri-
cal analyses to measured data of both surface settlements and horizontal ground move-
ments.
Steinhaldenfeld
Steinhaldenfeld station
cemetery
Hauptfriedhof
station
m
l = 94 0
NATM,
cu t a
cut and n
cov d
co ve r l = 1 er
l = 110m 10 m
97
Chapter 5 Tunnel case studies
5m
6.2 m
top
heading
invert
940m, followed later by the excavation of the invert. An initial length of 120m will be
considered in the numerical analyses for the simulation of the top heading excavation.
5.1.1 3D FE-analyses
The three-dimensional FE-analyses of the Steinhaldenfeld tunnel were carried out for
a mesh block of 120m length, 40m width and 26m height, as shown in Fig. 5.3. The
ground is described with 15-noded wedge elements with a length of 1.2m, being equal
to the unsupported round length of the tunnel. The top layer is a manmade fill under-
lain by two layers of Keuper Marl. This heavily weathered soft rock is underlain by
strong limestone, which stiffness is approximately ten times larger than the stiffness of
the marl layers on top. Considering its high stiffness the FE-mesh was not extended into
the limestone. Mohr-Coulomb parameters as listed in Tab. 5.1 were available from site
investigation (M ÖLLER et al., 2004; R OGOWSKI, 2004). The Stuttgart Keuper Marl might
be classified both as a residual soil or as a (sometimes heavily) weathered soft rock. In
such a ground measurements of the initial horizontal stress are extremely difficult to
perform; both in case of in situ borehole measurements as well as in case of laboratory
tests on (disturbed) samples. Moreover, correlations as presented in Section 4.4.1 came
from data of sedimentary clays rather than residual soils. Existing data for Stuttgart
Keuper Marl would seem to suggest that horizontal stresses are well above those of a
normally consolidated soil, but with K0 around unity also well below those of a heavily
overconsolidated sedimentary clay.
To simulate drained ground behavior in the 3D analyses the HS Model (see Appendix
A.2) was considered. Because the HS-Small Model (see Appendix A.3) was not yet op-
erational in 3D, it was only considered in the 2D analyses (Section 5.1.2). Ground pa-
rameters of the HS and the HS-Small Models are listed in Tab. 5.21 . The Poissons ra-
tio νur for unloading/reloading and the stiffness exponent m have been obtained from
1
For a full description of the HS- and the HS-Small models and their performance in drained and
undrained triaxial tests, the reader is referred to B ENZ (2006).
98
5.1 Steinhaldenfeld conventionally driven tunnel
26m
H = 16m
D = 6.2 m
120m
40m
H ORNIG and B UCHMAIER (2005), who performed both triaxial and oedometer labora-
ref
tory tests on Stuttgart Keuper Marl. The stress dependent Youngs modulus Eur for
2
unloading/reloading has been calculated from the Youngs modulus in Tab. 5.1 , con-
sidering the stresses acting in the middle of ground layers and using Eq. A.16 from Ap-
pendix A.2. The laboratory tests on weathered Keuper Marl carried through by H ORNIG
ref ref
and B UCHMAIER (2005) showed a stiffness ratio of Eur /Eoed of three and this was used
ref
to obtain the stress dependent oedometer modulus Eoed . In addition it was assumed that
ref ref
E50 ≈ Eoed for hard soils S CHANZ (1998). For the parameters of the HS-Small Model a
review of correlations is given bypB ENZ (2006). For the small strain stiffness in Tab.5.2
the correlation E0 [M P a] = 140/e· p0 /pref by B IAREZ and H ICHER (1994) has been used,
where e is the void ratio, p0 is the mean effective stress in kP a and pref = 100kP a. Data
on the void ratio e have been obtained from laboratory tests on Stuttgart Keuper Marl by
H ORNIG and B UCHMAIER (2002). For p0 = pref the correlation of B IAREZ and H ICHER
reduces to E0 [M P a] = E0ref = 140/e, and E0ref can be converted into Gref 0 using linear
elasticity with G = E/(2 + 2νur ). For the threshold shear strain γ0.7 a correlation chart to
the soils plasticity index (B ENZ, 2006) has been used.
The tunnel was built above the groundwater level and therefore groundwater was
not considered in the FE-analyses. Directly around the tunnel the ground was im-
proved by nails and this was numerically accounted for by an increase of the cohe-
sion of 25kP a. The sprayed concrete lining was modelled linear elastically with an unit
weight of γ = 24kN/m3 and a thickness of 25cm. To account for hardening of the lin-
ing, the stiffness was increased stepwise, using a Young’s modulus for fresh shotcrete of
E = 7, 5GN/m2 for the first three lining rings behind the tunnel face and E = 15GN/m2
for the older shotcrete of the rest of the lining. The lining was modelled by shell elements
2
Please note that Young’s moduli in Tab. 5.1 are for unloading/reloading.
99
Chapter 5 Tunnel case studies
layer γ ν E c0 ϕ0 K0
[kN/m3 ] [−] [M P a] [kP a] [◦ ] [−]
1 20 0.37 15 10 25 0.57
2 24 0.2 100 25 25 0.9
3 23 0.35 60 25 25 0.9
4 23 0.2 750 200 35 0.6
ref ref
layer νur Eoed E50 ref
Eur Gref
0 γ0.7 m
[−] [M P a] [M P a] [M P a] [M P a] [%] [−]
1 0.2 10 10 30 75 2 · 10−4 0.5
2 0.2 33 33 100 250 2 · 10−4 0.4
3 0.2 16 16 48 120 2 · 10 −4
0.4
4 0.2 190 190 575 1438 2 · 10−4 0.3
Table 5.2: Additional ground parameters as used for the HS- (see Appendix A.2) and the
HS-Small Model (see Appendix A.3). Gref
0 and γ0.7 are parameters used in the
HS-Small model.
and previous lining data was converted into corresponding values of the flexural rigidity
El I and the normal stiffness El A. To simulate the excavation of the hoarse shoe shaped
top heading, the step-by-step-excavation was applied (Section 4.5.1) with a round length
of d = 1.2m 3 .
Structural forces Fig. 5.4 shows results of bending moments and normal forces at the
upper part of the tunnel wall after 57 excavation steps. Normal forces show a zigzagging
between about 1000kN/m at the front of a lining ring and 600kN/m at the back of a lining
ring. At the tunnel heading the normal force has not yet reached the average value of
about 800kN/m. Instead a lower value of about 300kN/m is obtained. This decrease of
the normal force towards the tunnel heading is logical, as lining forces build up due to
further cutting, while the three-dimensional state of stress gradually changes into a two-
dimensional state. The highest normal forces are observed at the entrance of the tunnel.
Here arching around the initial excavation could hardly occur as the left side of the mesh
was free to displace and a horizontal active earth pressure (earth pressure according to
3
A numerical ϕ-c-reduction analysis gave a factor of safety against face instability of around η = 1.4.
100
5.1 Steinhaldenfeld conventionally driven tunnel
N [kN/m] M [kNm/m]
-1600
M+ N+
-1200
N
-800
-40
-400
M -20
0 0
60 40 20 0
German design code DIN 4085) as shown in Fig. 5.5 was applied to simulate a soldier
pile wall.
Just like the normal forces the bending moments in Fig. 5.4 show a zigzagging pat-
tern that matches the step-by-step excavation, but the magnitude of this zigzagging is
considerably smaller than for normal forces. This is most probably related to the use of
the relatively high K0 -values, as has been discussed in Section 4.4.2. Near the tunnel
heading vanishing small bending moments around zero are found. However, with the
advance of the tunnel face the bending moment reaches a relatively small value of about
15kN m/m.
Surface settlements Fig. 5.6 shows measured settlements above the tunnel center line
as well as computational results. For this particular tunnel, the measurements do not
show a horizontal steady-state line, as the ground profile changes significantly along
the axis of the tunnel. Moreover, the construction of a soldier-pile wall prior to tunnel
excavation has lead to large additional settlements, as shown in the left of Fig. 5.6.
The HS Model involves a yield cap which position is controlled by the preconsolida-
tion pressure as explained in Appendix A.2. In addition there is a more or less conical
shear yield surface around the space diagonal in principal stress space, as illustrated in
Fig. A.3. For a normally consolidated soil with OCR = 1 the model will predict plastic
yielding from the onset of loading. For OCR > 1 and K0 > K0N C both the cap and the
shear yield surface will be initially positioned some distance away from the initial stress,
and the model predicts initially elastic strains. As a consequence OCR is an important
input parameter. Weathered Keuper Marl may have characteristics of a heavily overcon-
101
Chapter 5 Tunnel case studies
0.00
depth below
137.24
21.50
2.0
2D cross section
3.0
settlement [cm]
4.0
5.0
6.0
102
5.1 Steinhaldenfeld conventionally driven tunnel
1.0
2.0
measurement
settlement [cm]
3D HS (OCR = 2)
3D HS (OCR = 1) 3.0
4.0
solidated soil, but in many aspects it behaves rather like a normally consolidated soil.
Indeed, relative to unloading/reloading initial loading gives a soft response and one is
tempted to model this behavior by initializing OCR as unity. However, on the very onset
of loading most in situ soils (B ENZ, 2006) tend to show a high ”small-strain stiffness”,
being not directly incorporated in the HS Model. This may be modelled by expanding
the initial position of the yield surfaces somewhat beyond the existing initial stress. In
the present analysis this will be done by using OCR = 2.
However, to demonstrate the importance of OCR the settlement analyses shown in
Fig. 5.6 were also carried out for a normally consolidated ground (OCR = 1). In the
HS Model the initially stiffer response of overconsolidated grounds is accounted for by
the unloading-reloading Young’s modulus Eur . The softer ground response, associated
with material hardening takes place, when stresses exceed the preconsolidation pressure
(compare Section A.2).
Fig. 5.6 shows that the analysis for a normally consolidated ground (OCR = 1) clearly
yields too large settlements but good agreement with measurements is obtained for the
overconsolidated ground with OCR = 2. Fig. 5.7 shows the computed cross-section of
the settlement trough. This figure also shows that the settlements of a normally con-
solidated ground overestimate the measurements, whereas the analysis with OCR = 2
yields basically the correct settlements.
5.1.2 2D FE-analyses
Fig. 5.8 shows a two-dimensional FE-mesh of the Steinhaldenfeld tunnel. The mesh con-
sists of 6-noded triangular elements and its dimensions correspond to the cross-section
of the three-dimensional FE-mesh in Fig. 5.3. Drained ground behavior was modelled
103
Chapter 5 Tunnel case studies
0
1. Cover layer
5.4m
13.4m
with the MC, the HS and the HS-Small Model 4 with parameters given in Tables 5.1 and
5.2. Please note that for the MC Model the stiffnesses for unloading-reloading have been
considered as the ground was heavily overconsolidated. Because stiffness is stress level
dependent and this is not accounted for by the MC Model, the values of E given in Table
5.1 have been adjusted to the stress level acting in the middle of the respective layers
ref
(compare Section 4.8.3). The values of Eur as used for the HS and the HS-Small Model
(Table 5.2) are reference values and are thus different from the absolute stiffnesses of
the MC Model. The shell elements have a normal stiffness of El A = 10.5GN , a flexural
rigidity of El I = 26.78M N m2 and a unit weight of γ = 24kN/m3 .
Surface settlements Fig. 5.9 shows results of measured and computed settlement
troughs by using the stress reduction method. The best fit for the settlement analysis
was found for unloading factors of β = 0.28, β = 0.36 and β = 0.3, meaning a stress
reduction of around 70% for this particular conventionally driven tunnel, which is well
above the usual assumption of a stress reduction of 50% (β = 0.5). The HS-Small Model
gives the best prediction, but the measured settlement trough is still slightly steeper.
Fig. 5.10 shows similar results for the normalized settlement troughs using the same
unloading factors of β = 0.28 for all three constitutive models. The HS-Small Model
still gives the best prediction but the HS-curve considerably improves. Using a smaller
unloading factor of β = 0.28 for the HS Model increases the steepness of its settle-
ment curve, but it also has a considerable effect on the magnitude of settlements, giving
Smax = 4.0cm instead of 2.0cm. This dependency of the shape of the settlement trough
on unloading factors is also observed for the MC Model. Like the HS Model its settle-
ment curve is becoming slightly steeper, but this time the change of the magnitude of
4
For the elemental stress-strain behavior of the HS- and the HS-Small models, the reader is referred to
B ENZ (2006).
104
5.1 Steinhaldenfeld conventionally driven tunnel
1.0
Measurement
HS-Small Model, b = 0.28
HS Model, b = 0.36
settlement [cm]
MC Model, b = 0.3
2.0
Figure 5.9: Settlement troughs from 2D analyses using the stress reduction method. The
β-values are chosen such that 2D analyses match the measured settlement.
K0 as indicated in Table 5.1.
Measurement
S / Smax [-]
Figure 5.10: Normalized settlement troughs from 2D analyses with constant unloading
factor
105
Chapter 5 Tunnel case studies
1.0
measurement
2D HS, b = 0.36
settlement [cm]
3D HS
2.0
the settlement is somewhat smaller. Instead of the correct maximum settlement of 2.0cm
a value of Smax = 2.4cm is observed.
The effect that the measured settlement curve is more steep than the computed one has
been discussed in Section 4.4.2. The statement by L EE and N G (2002) that 2D analyses
in combination with large K0 -values tend to predict too wide settlement troughs when
compared to 3D analyses was not observed in the studies of Section 4.4.2. In the present
study, however, this effect is observed when comparing results of the HS Model from 2D
and 3D analyses. Fig. 5.11 shows that the 3D analysis improves and gives a considerably
steeper settlement trough, but it is believed that this is rather an effect of the non-constant
3D soil profile as well as the non-circular tunnel cross section being used in the present
studies.
Moreover, the present results show that the use of the small-strain stiffness consid-
erably improves results for the surface settlement trough, but the measured settlement
curve still remains slightly steeper. Most probably this is related to the fact, that the ma-
terial behavior of the Keuper Marl layers contains some anisotropy. However, this was
not accounted for by the present studies and only isotropic constitutive models were
applied to model the ground behavior. It is thus believed, that the additional considera-
tion of the grounds anisotropy will further improve the results of the surface settlement
trough.
Structural forces Fig. 5.12 shows results of structural forces of both 2D and 3D analyses
using the HS Model. As three-dimensional structural forces oscillate within one ring of
lining, maximum values from the front of a lining ring are plotted on the right side of
the tunnel and rear-values are plotted on the left side of the tunnel. The 2D settlement
analysis with ß=0.36 yields too low bending moments and normal forces almost by a
106
5.2 Second Heinenoord slurry shield tunnel
800 1260
60 69
rear front
3D 3D
2D, b = 0.36 2D, b = 0.36
2D, b = 0.49 2D, b = 0.71
d=1.2m
Figure 5.12: Comparison of structural forces of 2D and 3D analyses using the HS Model
factor of two. To match the average maximum 3D bending moment of 64.5 kNm/m
a considerable higher unloading factor of 0.49 had to be applied and for the average
maximum normal force of 1030kN/m an unloading factor of even 0.71 was necessary.
The need for different unloading factors for surface settlements, bending moments and
normal forces has been discussed in Section 4.8. The present analyses make it clear that
for structural forces the required amount of stress reduction is significantly smaller than
for surface settlements.
107
Chapter 5 Tunnel case studies
5.2.1 3D FE-analyses
Fig. 5.14 shows a three-dimensional mesh block, with a length of 120m, a width of 40m
and a height of 27.5m as used for the analyses of the Second Heinenoord tunnel. The
ground water table is at 1.5m below ground surface. The top ground layer is a fill un-
derlain by two layers of sand and a subsequent sand-clay layer. All layers are assumed
to behave drained. Mohr-Coulomb soil parameters as listed in Tab. 5.3 were available
from the site report (B AKKER, 2000). For the 3D analyses only the HS Model (see Ap-
pendix A.2) was applied. Because the HS-Small Model (see Appendix A.3) was not yet
operational in 3D, it was only considered in the 2D analyses (Section 5.2.2). Ground
parameters of the HS and the HS-Small Models are listed in Tab. 5.45 .
The Poissons ratio νur for unloading/reloading and the stiffness exponent m have been
assumed, considering that for many soils νur = 0.2 and typically m = 1.0 for a clay and
ref
m = 0.5 for a sand. The stress dependent oedometer modulus Eoed was calculated from
the oedometer modulus in Tab. 5.3, considering the stresses acting in the middle of
ground layers and using Eq. A.27 from Appendix A.2. For the first three layers it was
ref ref
assumed that E50 ≈ Eoed as found for sands by (S CHANZ, 1998), for the clayey bottom
ref ref
layer it was assumed that E50 ≈ 2 · Eoed S CHANZ (1998). The stress dependent Youngs
ref
modulus Eur for unloading/reloading has been calculated from the secant modulus
ref ref ref
E50 assuming a ratio of Eur /E50 = 3. The parameters of the HS-Small Model were not
directly available, but were estimated to be Gref 0 = 175M P a and γ0.7 = 5 · 10−4 .
5
For a full description of the HS- and the HS-Small models and their performance in drained and
undrained triaxial tests, the reader is referred to B ENZ (2006).
108
5.2 Second Heinenoord slurry shield tunnel
H = 12.5m 27.5m
D = 8.3m
l=1.5m
120m
40m
l=1.5m
i-1 i i+1
The tunnel lining has been modelled linear elastically, using shell elements with a
flexural rigidity of El I = 26.78M N m2 , a normal stiffness of El A = 10.5GN and a lining
weight of γ = 24kN/m3 . The flexural rigidity has been reduced to account for joints
between segments. Following M UIR W OOD (1975) a reduction factor of four has been
applied to the flexural rigidity of the tubings. The TBM and the shield was modelled
according to the step-by-step pressure method (Section 4.5.2.2). The slurry at the face
of the tunnel is simulated by an axial pressure, the shield is simulated by a radial pres-
sure and the fresh grout in the tail void behind the shield is also simulated by a radial
pressure.
All prescribed pressures increase hydrostatically with depth according to a unit weight
of 15kN/m3 of the slurry and the grout. The face pressure of 230kP a at the tunnel crown
was chosen according to measurements of the Heinenoord tunnelling project (B AKKER,
1999). The radial shield pressure was taken equal to the grout pressure of 125kP a at
the crown. This pressure is applied along the shield and along two lining rings directly
behind the shield, as shown in Fig. 5.14. Here a ground-lining gap is simulated by
deactivating ground elements with a thickness of 20cm and a length of 1.5m according
to the width of a lining ring. Within this gap the ground may deform until contact to
the shield or the lining is made. Hence ground displacements were controlled not to
exceed the ground-lining gap of 20cm. For subsequent lining rings, the grout is assumed
109
Chapter 5 Tunnel case studies
ref ref
layer νur Eoed E50 ref
Eur Gref
0 γ0.7 m OCR
[−] [M P a] [M P a] [M P a] [M P a] [%] [−] 1
1 0.2 14 14 42 52 5 · 10−4 0.5 1
2 0.2 35 35 105 175 5 · 10−4 0.5 1
3 0.2 35 35 105 175 5 · 10 −4
0.5 1
4 0.2 7 12 35 88 5 · 10−4 0.9 1
Table 5.4: Additional ground parameters as used for the HS- (see Appendix A.2) and the
HS-Small Model (see Appendix A.3). Gref
0 and γ0.7 are parameters used in the
HS-Small model.
to be hardened and the radial pressure is switched off. Each time radial pressures are
switched off, volume elements are activated to fill the gap, assuming linear elastic lining
properties for the hardened grout.
To simulate the excavation and support sequence a step-by-step pressure approach as
described in Section 4.5.2.2 and as indicated in Fig. 5.14 has been applied. In the present
study a steady state solution of surface settlements was reached after 30 excavations of
segmental lengths of l = 1.5m, resulting in a tunnel length of 45m.
Surface settlements Fig. 5.15 shows results of the longitudinal settlement trough by
the HS-Model and different grout pressures. Measured settlements appear to be well
matched by a crown pressure of 125kP a. The largest maximum settlement of 3.25cm is
obtained for a radial pressure of pcrown = 120kP a and a significantly smaller maximum
settlement of 2.0cm is obtained for a crown pressure of p = 130kP a. From these analyses
it is obvious that settlements are extremely sensitive to grout pressures and in addition
to the pressure boundary condition a limitation of tunnel convergence is needed. This
limitation of tunnel convergence is the outer diameter of the TBM or the tunnel lining re-
spectively as explained in Section 4.7. However, such a restriction was not implemented
110
5.2 Second Heinenoord slurry shield tunnel
measurement
1.0
settlement [cm]
pcrown = 125 kPa
3.0
pcrown = 120 kPa
TBM
Method A Method B
l=1.5m l=1.5m
in the present analyses and a check was carried out after the analyses to find that for all
analyses the maximum tunnel convergence was well below this restriction.
Please note that Fig. 5.15 shows two nearly coinciding settlement lines for the crown
pressure of 125kP a, as two slightly different computations were carried out. Besides
the standard analysis (Method A) as explained above, another analysis (Method B) has
been carried out. In this B-analysis a shield has been simulated by shell elements, as
shown in Fig. 5.16. Between shield and ground, volume elements of width 20cm were
switched off, to simulate a gap as also used around the lining. Inside this gap radial
pressures were applied both to the shield and to the ground. The difference between
results from Method A and Method B is an extra settlement of only 1mm as indicated by
the thin shaded zone in Fig. 5.15. It is also observed from Fig. 5.15 that the steepness of
the longitudinal settlement curve is matched well, but compared to measurements the
111
Chapter 5 Tunnel case studies
1.0
settlement [cm]
2.0
measurement
pcrown = 125, with shield, method B
pcrown = 125, without shield, method A
3.0
Structural forces Fig. 5.18 and 5.19 show results of 3D bending moments and nor-
mal forces. The longitudinal distributions show results of both Method A and Method
B. Both results of bending moments and normal forces indicate convergence towards
a steady-state solution. The longitudinal distributions show a zigzagging pattern that
matches the step-by-step installation of l = 1.2m.
The analyses with a shield appear to have slightly different values than the ones with-
out a shield. The graphs indicate that due to structural forces in the shield (first five
slices behind the tunnel face as indicated by the inserts of these figures) the distribu-
tion of the subsequent lining forces start with a value equal to the shield forces, whereas
in the case of no shield they start with a value close to zero. In Method B forces are
transferred from the shield to the lining as the interaction of shield and lining has been
modelled fully rigid. In reality the outer shield diameter is somewhat smaller than the
inner lining diameter, leaving some possibility for the shield to freely deform. One may
thus argue that the simplification of full bound in the transverse section is not appro-
priate. However, this simplification can be accepted as it slightly increases the value of
steady-state bending moments whereas at the same time normal forces are slightly de-
creased. The transverse distribution of bending moments, shown in Fig. 5.18, indicates
relatively little difference between front and rear values of a lining ring. The maximum
bending moment of about −25kN m/m is observed at the tunnel wall. The positive value
112
5.2 Second Heinenoord slurry shield tunnel
M+
-40
10
0
rear w
0
90
-10 front
180
-20
without shield
M [kNm/m]
113
Chapter 5 Tunnel case studies
N+
-400
-800
-1200
N [kN/m]
with shield, method B
without shield, method A
-1000
180
front
-1400
N [kN/m]
114
5.2 Second Heinenoord slurry shield tunnel
0
1.5m 1. fill
4m
2. sand
D=8.3m
19.75m
3. sand
23.25m
4. clay, local parts of sand
27.5m
5. stiff sand
indicates inward deflection of the lining. The difference between front and rear values of
normal forces is significant, as shown in Fig. 5.19. It is interesting to note that the front
normal force increases from the crown value of −1000kN/m down to a maximum at the
invert of about −1400kN/m, whereas the rear normal force shows a reversed distribu-
tion decreasing from a crown value of about −450kN/m down to the invert normal force
of about −300kN/m.
5.2.2 2D FE-analyses
Fig. 5.20 shows a two-dimensional FE-mesh of the Heinenoord tunnel. Again 6-noded
triangular elements were used for the mesh with a width of 40m and a height of 27.5m.
The soil parameters to model drained ground behavior with the MC, HS and the HS-
Small Model are listed in Tables 5.3 and 5.4. Please note that for the MC Model the
stiffnesses given in Table 5.3 are absolute stiffnesses which have been adjusted to the
stress level acting in the middle of the respective layers (compare Section 4.8.3). The val-
ref
ues of Eur as used for the HS and the HS-Small Model (Table 5.4) are reference values
and are thus different from the absolute stiffnesses of the MC Model. Whereas the HS
and the HS-Small Model automatically make distinction between primary loading and
unloading-reloading stiffnesses, the MC Model uses a constant stiffness, no matter the
loading history. In order to account to some extend for the unloading behavior in the MC
Model, the layers underneath the tunnel have been assigned the unloading-reloading
stiffness, whereas the layers on top were assigned the stiffness for primary loading. Be-
sides the evaluation of the influence of constitutive models, the influence of different
installation procedures for shield tunnels, namely the stress reduction method, the con-
traction method and the grout pressure method, was studied. In the latter method, the
ground-lining interaction described in Section 4.7 was not yet implemented and there-
fore a slightly simplified approach was chosen. In order to model the Heinenoord shield
tunnel only the grout pressure distribution was assigned to the excavated ground mod-
elling no lining. Doing so, the results of ground displacements were still controlled not
115
Chapter 5 Tunnel case studies
1.0
measurement
2D HS
settlement [cm]
3D HS
2.0
3.0
Surface settlements The results of the 2D grout pressure method are shown in Fig.
5.21. Although the 2D grout pressure method does not model a face pressure, results
compare well to the 3D analysis of Heinenoord. This indicates that the face pressure
of the 3D analysis does not much contribute to the development of surface settlements.
Indeed, as discussed previously in Section 3.3, ground deformations due to shield tun-
nelling are predominantly determined by the process of tail void grouting, causing the
largest deformations at the shield tail. If appropriate face pressures are assigned almost
no deformations will occur around the tunnel heading.
Fig. 5.22 shows results of the transverse settlement trough of the grout pressure
method using different constitutive models. The best agreement to measurements is
obtained from the HS-Small Model with a crown grout pressure of 128.5kP a and a unit
grout weight of γ = 15kN/m3 . Considering the initial total overburden stress of 225kP a,
the present analysis renders a stress reduction of 43%, being slightly less than the stress
reduction of 50% as proposed by M UIR W OOD (1975). The difference to the settlement
trough of the MC and the HS Model is moderate, but slightly higher crown pressures of
135kP a and 133kP a respectively had to be applied to match the maximum settlement.
Fig. 5.23 shows results of normalized transverse settlement troughs of different consti-
tutive models, using the same crown pressure of 128.5kP a for all analyses. The results
of maximum settlements as indicated in this figure, demonstrate the fact that ground
displacements are very sensitive to a variation of the grout pressure, giving now signif-
icantly larger settlements of Smax = 4.2cm and Smax = 3.6cm for the MC and the HS
Model respectively. Moreover, Fig. 5.23 demonstrates the large influence of a pressure
variation on the steepness of the settlement curves. Whereas the steepness of the HS
curve has not changed much, the steepness of the MC curve is heavily influenced, al-
116
5.2 Second Heinenoord slurry shield tunnel
1.0
measurement
HS-Small pressure, pcrown = 128.5
settlement [cm]
HS pressure, pcrown = 133 2.0
MC pressure, pcrown = 135
3.0
Figure 5.22: Transverse settlement trough using the grout pressure method and different
constitutive models
measurement
HS-Small pressure, Smax = 2.6cm
HS pressure, Smax = 3.6cm
S / Smax [-]
117
Chapter 5 Tunnel case studies
1.0
measurement
HS-Small pressure
settlement [cm]
HS-Small contraction 2.0
HS-Small stress reduction, b = 0.69
3.0
most matching now the solution of the HS-Small Model. The fact that the MC Model is
giving a steeper normalized settlement curve than the HS Model may seem somewhat
surprising. Most probably this is also related to the fact that the layers underneath the
tunnel had been assigned the unloading-reloading stiffness of the ground. However,
interrelations are complex and no definite explanation of this effect was found. But the
clear statement shall be given here that by no means it is meant to recommend the use
of the MC Model superior to the HS Model for tunnel analysis.
The fact that the shape of the settlement trough is first of all affected by the tunnel in-
stallation procedure is demonstrated by Fig. 5.24. Indeed, it is shown by this figure that
the shape of the transverse settlement trough is more affected when applying different
installation procedures rather than different constitutive models. Using the HS-Small
Model the settlement trough of the stress reduction and the contraction method is far
more flat than for the grout pressure method. Obviously numerical installation proce-
dures are most important for the shape of the settlement trough.
Installation procedures and tunnel deformation Fig. 5.25 shows the influence of nu-
merical installation procedures on the displacements around the tunnel lining. The tun-
nel convergence of both the grout pressure method and the stress reduction method is
predominantly located at the tunnel crown whilst at the tunnel invert almost no con-
vergence occurs. The relatively small tunnel uplift in the region between tunnel invert
and tunnel wall is logical, as for the ground underneath the tunnel the much higher
unloading-reloading stiffness of the HS-Small model is involved. The contraction method
imposes far more tunnel uplift than observed for the grout pressure and the stress reduc-
tion methods, regardless of the high unloading-reloading stiffness being used. Moreover,
the contraction method contains a circular tunnel shape whereas the other methods al-
low for a concentration of deformations at the tunnel crown.
Fig. 5.25 also shows the influence of the installation procedure on the shape of the
118
5.2 Second Heinenoord slurry shield tunnel
Figure 5.25: Deformed FE-mesh scaled up 100 times from different installation proce-
dures using the HS-Small Model
surface settlement trough. Whilst the grout pressure method has the most localized
crown convergence and therefore gives the steepest and narrowest settlement curve, the
stress reduction and the tunnel contraction methods show considerably wider and more
flat troughs.
The difference between the crown convergence of the grout pressure method and the
stress reduction method is explained by Fig. 5.26. From the tunnel wall up to the tun-
nel crown the ground tends to deform into the tunnel and the resulting convergence is
governed by the radial stress distributions of the methods. Fig. 5.26 shows the total ra-
dial initial stresses before construction of the tunnel and the reduced total radial stresses
after construction of the tunnel of the grout pressure and the stress reduction methods
respectively. The stress reduction method was carried out with β = 0.69, i.e. a stress
reduction of 31%. As illustrated in Fig. 5.26, the stress reduction from total radial initial
down to reduced total radial stresses applies only to the effective radial initial stresses.
The pore water pressures u are not reduced.
Fig. 5.26 shows that in the region from the tunnel wall up to the tunnel crown, the
grout pressure method renders the highest stress reduction with the maximum stress
reduction at the crown. This explains why the grout pressure method gives the largest
displacements at the tunnel crown. The stress reduction method on the contrary renders
a uniformly distributed stress reduction and therefore tunnel contraction is less concen-
trated at the tunnel crown, giving as well some contraction at the tunnel wall.
119
Chapter 5 Tunnel case studies
crown
3m
8.
wall
=
D
invert
wall D
Figure 5.26: Difference in radial stresses from grout pressure and stress reduction meth-
ods of Heinenoord
19,75
measurements
HS-Small
depth [m]
HS 23,25
MC
27,5
Figure 5.27: Horizontal ground movements scaled up 100 times of the grout pressure
method using different constitutive models
120
5.2 Second Heinenoord slurry shield tunnel
19,75
measurements
HS-Small pressure
depth [m]
23,25
HS-Small contraction
2D stress reduction method,
b = 0.69
27,5
Figure 5.28: Horizontal ground movements scaled up 100 times of different installation
procedures using the HS-Small Model
izontal movements will strongly be influenced by the grouting pressure assigned and
insufficient grouting may have lead to inward directed movements. However, applied
grout pressures are difficult to measure and they remain uncertain. Fig. 5.28 shows hori-
zontal displacements of different installation procedures. Using the HS-Small Model the
stress reduction method predicts far too large inward horizontal displacements in the
region closest to the tunnel, as well as the contraction method which predicts even more
inward movement. For simulating horizontal displacements Fig. 5.27. and Fig. 5.28
show clearly that numerical installation procedures are more significant to the simula-
tion of horizontal deformations than constitutive models. In the present case, the best
prediction is obtained from the grout pressure method and the HS-small Model.
Structural forces Figs. 5.29 and show a comparison for bending moments and normal
forces of the grout pressure method, the stress reduction method and the contraction
method using the HS-Small model. For all methods the bending moments are relatively
small. The contraction method gives the largest bending moment of around 100kN m/m
at the tunnel crown and wall, the stress reduction gives the smallest one, with around
50kN m/m at the tunnel crown and wall. The grout pressure method is well in between
the results of these two methods, giving crown and wall values of around 75kN m/m.
Higher and thus more important than bending moments are normal forces, giving a
121
Chapter 5 Tunnel case studies
-100
M [kNm/m]
Figure 5.29: Comparison of bending moments from different 2D installation methods
using the HS-Small model
N+
-600
0
w
-1000 90
-1400 180
N [kN/m]
Figure 5.30: Comparison of normal forces from different 2D installation methods using
the HS-Small model
122
5.2 Second Heinenoord slurry shield tunnel
100
-50
-100
M [kNm/m] M+
0
w
90
180
200
150
-50
-100
M [kNm/m]
Figure 5.31: Comparison of computed and measured bending moments using the HS-
Small model
123
Chapter 5 Tunnel case studies
-1400
-1800
-2200
N [kN/m] N+
0
w
90
180
-600
2D grout pressure method
2D volume loss control method
-1000 2D contraction method
330th day measurement (left lining half)
330th day measurement (right lining half)
-1400
-1800
N [kN/m]
Figure 5.32: Comparison of computed and measured normal forces using the HS-Small
model
124
5.2 Second Heinenoord slurry shield tunnel
crown value of around −600kN/m with very similar distributions for both the stress
reduction and the contraction method. The grout pressure method gives considerably
higher normal forces with a crown value of around −800kN/m.
In order to compute realistic values of bending moments and normal forces, it can not
be expected that the stress reduction method and the contraction method give accurate
solutions as they tend to predict ground movements incorrectly. This fact is also demon-
strated by Figs. 5.31 and 5.32, by showing 2D bending moments of different installation
methods and comparing them to measurements of Heinenoord. Please note that mea-
surements of structural forces were taken after 9 days (upper figures) as well as after 330
days (lower figures). Both measurements of bending moments and normal forces show
a slight increase from the first to the second measurement. Although measurements
show a relatively high scatter both Figs. 5.31 and 5.32 show that measurements are best
matched by the grout pressure method. In particular this is demonstrated for measured
normal forces, being significantly higher than predicted by the stress reduction and the
contraction method. The grout pressure method clearly improves, giving higher normal
forces much closer to the measured ones.
No doubt, bending moments and normal forces in tunnel linings are difficult to mea-
sure and good quality data for validating magnitudes of structural forces are difficult
to obtain. Nevertheless, the presented results of the grout pressure method indicate
that this method gives better predictions than the stress reduction and the contraction
method for both tunnel displacements and structural forces.
125
Chapter 6
Conclusions
Introduction
The objective of this thesis was to analyze tunnel induced settlements and structural
forces in linings, considering both elementary methods of analysis and the Finite Ele-
ment Method. The most important conclusions will be given in the following sections,
followed by some recommendations for further research.
127
Chapter 6 Conclusions
tunnels they state that some stress reduction should generally be valid, but again mag-
nitudes are not discussed. For shield tunnels it would seem that mostly full primary
stresses are incorporated into structural design models, not accounting for effects of tun-
nel installation.
Influence of K0 : From a parametric study on the magnitude of the lateral earth pres-
sure it was observed that both surface settlements and structural forces are heavily influ-
enced by the value of K0 . For surface settlements much wider and shallower settlement
troughs were computed for larger K0 .
The above observations hold both for two- and three-dimensional settlement analyses,
which appear to show good agreement for settlements up to a value of K0 ≤ 1.25, at least
as long as appropriate unloading factors are used in the 2D analyses. This is conform the
findings of several other authors, but in contrast to the findings by L EE and N G (2002).
For K0 ≥ 1.25 both 2D and 3D analyses predicted unrealistic surface heave instead of
settlement. It may therefore be concluded that an isotropic hardening model as used
in the present studies, gives a reasonable prediction of settlements for approximately
K0 ≤ 1.0. For K0 values being beyond one, however, an isotropic hardening model with
elastic unloading behavior would not seem to be appropriate anymore.
Similar to surface settlements, structural forces are heavily influenced by a change of
K0 . For K0 ≥ 1 the bending moment at the tunnel wall changes from an outward deflec-
tion to an inward deflection. Results of 2D analyses of structural forces are generally well
in between the values at the front and at the rear of a lining ring of three-dimensional
normal forces. 2D analyses of bending moments reasonably match results from 3D anal-
yses for the case of K0 = 0.5. For K0 > 0.5 2D analyses of bending moments show some
deviation from results of 3D analyses. However, this may be considered not so impor-
tant as in comparison to normal forces magnitudes of bending moments appear to be
relatively small.
Unloading factor β for open face tunnelling: 3D step-by-step modelling of tunnel ex-
cavation and support is straight forward, but still engineering time consuming. There-
fore the 2D stress reduction method is frequently used. A parametric study on the mag-
nitude of the unloading factor being used in the stress reduction method has been car-
ried out by a calibration to results of 3D analyses. It has been shown that three different
128
6.2 On the FEM
3D shield tunnelling: 3D shield tunnel analyses are still under debate, but the ten-
dency is to use step-by-step pressure simulations. The simulation of a slurry shield tun-
nel by the step-by-step pressure method showed that these simulations are extremely
sensitive to the magnitude of grout pressures. On the contrary, an alternative analysis
to model as well the shield by stiff shell elements showed that almost no difference in
surface settlement and structural forces occurred as compared to an analysis without a
shield.
129
Chapter 6 Conclusions
130
Appendix A
On constitutive models as used in this thesis
In order to discuss the influences of three different constitutive models on the results
of FE-tunnel analysis (see Section 5) first of all the simple MC Model will shortly be
described. Hereafter main features of two advanced models will be described, namely
the Hardening-Soil (HS) and the Hardening-Soil-Small (HS-Small) Model. As the focus
is on tunnelling settlements and lining forces, i.e. on ground deformation, emphasis will
be placed on the formulation of the ground stiffness. For more details on the formulation
of these two models the reader is referred to other publications.
131
Appendix A On constitutive models as used in this thesis
s s1
sf
e s3
s2
a) b)
Figure A.1: Basic ideas of MC-Model: a) liner elastic perfectly plastic material behavior,
b) yield surface in principal stress space with c0 = 0
where σ̇ is the stress rate, ε̇e the corresponding elastic strain rate and De is the elastic
material stiffness matrix. The latter one is formulated using the two elastic material
constants E and ν. The formulation of perfect plasticity decomposes strain rates ε̇ into
an elastic and a plastic part
ε̇ = ε̇e + ε̇p , (A.5)
where ε̇p is the plastic strain rate. Combining Eqs. A.4 and A.5 yields
σ̇ 0 = De (ε̇ − ε̇p ). (A.6)
If associated plasticity is assumed, the plastic strain rates can then be formulated using
the yield function introduced in Eqs. A.1 - A.3. Doing so, the plastic strain rates become
vectors perpendicular to the surface of the yield function. However, the use of MC type
of plastic potential functions would lead to a considerable overprediction of the angle of
dilatancy. Therefore, in addition to yield functions, different plastic potential function g
are employed. Using the plastic potential functions, non-associated plasticity is adopted
and the plastic strain rates are formulated as
∂g1 ∂g2 ∂g3
ε̇p = λ1 + λ2 0 + λ3 0 , (A.7)
∂σ 0 ∂σ ∂σ
where λ1 , λ2 and λ3 are plastic multipliers. The plastic potential functions are
1 1
g1 = · |σ20 − σ30 | − · (σ20 + σ30 ) · sin ψ (A.8)
2 2
1 1
g2 = · |σ30 − σ10 | − · (σ30 + σ10 ) · sin ψ (A.9)
2 2
1 1
g3 = · |σ10 − σ20 | − · (σ10 + σ20 ) · sin ψ (A.10)
2 2
132
A.2 The Hardening-Soil Model
The dilatancy angle ψ is used to model positive plastic volumetric strain increments
(dilatancy) in case of plastic yielding. Using the consistency condition
∂f 0
f˙ = σ̇ = 0, (A.11)
∂σ 0
and employing Eqs. A.6, A.7 and A.11, the plastic multipliers λ1 , λ2 and λ3 are solved
from the equation
∂fi e ∂g1 ∂g2 ∂g3
f˙i = D (ε̇ − λ1 0 − λ2 0 − λ3 0 ) = 0, (A.12)
∂σ 0 ∂σ ∂σ ∂σ
where i runs from 1 to 3.
For more details on the formulation and implementation of the MC Model the reader is
referred to S MITH and G RIFFITH (1982), VAN L ANGEN and V ERMEER (1990) and
B RINKGREVE and V ERMEER (2001).
133
Appendix A On constitutive models as used in this thesis
q = s1 - s3
asymptote
qa
failure line
qf
1
E50
Eur
e1
Figure A.2: Drained triaxial test: hyperbolic relationship for primary loading between
deviatoric stress and axial strain
where Rf = 0.9 for many soils. While the maximum stress is determined by the Mohr-
Coulomb failure criterion, the hyperbolic part of the curve can be defined using a single
secant modulus as additional input parameter. In the HS Model this is the stress depen-
dent modulus E50 , as used in Eq. A.13, which is defined as
µ 0 ¶m
ref c · cot ϕ0 + σ30
E50 = E50 · 0 , (A.15)
c · cot ϕ0 + pref
ref
where E50 is a reference stiffness modulus, corresponding to the reference confining
pressure pref . Following the ideas by O HDE (1951) the amount of stress dependency is
governed by the exponent m, which can be measured both in oedometer tests and in
triaxial tests. One tends to find values between 0.4 and 1.0. A value of 0.5 is typical for
sands and clays tend to have m = 1.0.
In contrast to E50 , which determines the magnitude of both the elastic and the plastic
strains, Eur is a true elasticity modulus. In conjunction with a Poisson’s ratio νur it de-
termines the ground behavior under unloading and reloading; the indices ur stand for
unloading/reloading. As the average primary loading modulus E50 the unloading modu-
lus Eur is stress-level dependent. For the HS Model it yields
µ 0 ¶m
ref c · cot ϕ0 + σ30
Eur = Eur · 0 , (A.16)
c · cot ϕ0 + pref
ref
where Eur is the reference Young’s modulus, corresponding to the reference confining
pressure pref .
When comparing the hardening model to the previous elastic prefectly-plastic MC
Model another significant difference is that plastic strains may already occur before the
limit MC-failure stress is reached. This implies that the HS Model incorporates another
yield surface, which is not fixed in principal stress space, but it may expand and soil
hardening is simulated due to plastic straining. As shown in Fig. A.3a, distinction is
134
A.2 The Hardening-Soil Model
q = s1 - s3 s1
e
lin
re
ilu
C -fa
M
fs
fc
elastic region
pp s3
p = 1/3 (s1 + s2 + s3) s2
a) b)
Figure A.3: Yield surface of the HS Model for c = 0: a) successive yield loci for shear
hardening and compression hardening in p-q-space b) total yield contour in
principal stress space
made between two types of hardening, namely shear hardening and compression hard-
ening. For the shear hardening law a yield function f s is introduced, which is a function
of the triaxial loading stiffness E50 and for the compression hardening a yield function f c
is formulated, being governed by the oedometer loading stiffness Eoed . As also indicated
in Fig. A.3a for unloading-reloading elastic soil behavior is assumed, adopting Hook’s
law with Young’s modulus Eur . Fig. A.3b shows the total contour of the HS yield surface
in principal stress space.
Yield function fs The yield function fs adopted in the HS-Model has the formulation
f s = f¯ − γ p , (A.17)
where
µ ¶m µ ¶m
1 c0 · cot ϕ0 + σ30 q 2q c0 · cot ϕ0 + σ30
f¯ = ref
· · − ref · (A.18)
E50 c0 · cot ϕ0 + pref 1 − q/qa Eur c0 · cot ϕ0 + pref
is a function of stress and the hardening parameter
κs = γ p = εp1 − εp2 − εp3 = 2 · εp1 − εpv ≈ 2 · εp1 (A.19)
is a function of plastic strains. Similar to the MC-Model the HS-Model adopts non-
associated plasticity to determine the rates of plastic strain with the plastic potential
g s = (3 − sin ψm ) · q − 6 · sin ψm · p, (A.20)
with p = 1/3 · (σ1 + σ2 + σ3 ). The mobilized angle of dilatancy ψm is calculated according
to the so-called stress-dilatancy equation of R OWE (1962)
sin ϕm − sin ϕcv
sin ψm = , (A.21)
1 − sin ϕm · sin ϕcv
135
Appendix A On constitutive models as used in this thesis
s1
Eoed
1
e1
sin ϕ − sin ψ
sin ϕcv = . (A.23)
1 − sin ϕ · sin ψ
The dilatancy angle is thus positive as soon as ψm exceeds a constant-volume angle ϕcv .
Considering dense materials contraction is excluded by taking ψm = 0 for a mobilized
friction angle ϕm < ϕcv .
136
A.2 The Hardening-Soil Model
formulates the relation between the plastic volumetric cap-strain εpc v and the preconsoli-
dation stress pp . pref is an isotropic reference pressure. The cap parameter β is not used
as an direct input parameter. Instead the odometer stiffness Eoed is used as an input
parameter which is linked to β. Fig. A.4 shows the typical characteristic curve of an oe-
dometer test. In the HS Model the virgin oedomenter stiffness obeys a stress dependency
according to the formula
µ 0 ¶m
ref c · cot ϕ0 + σ10
Eoed = Eoed · 0 , (A.27)
c · cot ϕ0 + pref
ref
where Eoed is the reference Oedometer Modulus for the axial reference pressure pref .
In the special case of m = 1 one obtains a linear stress-dependency as usual for a clay.
In addition to the moduli E50 and Eur , the oedometer modulus Eoed is also an input
modulus for the HS Model. Together with the parameters m, νur , c0 , ϕ0 and the dilatancy
angle ψ, there are a total of eight input parameters. To determine the rates of plastic
volumetric strains associated plasticity, i.e. g c = f c is adopted.
Plastic multipliers The strain rates are decomposed into an elastic part ε̇e and into a
plastic shear hardening part ε̇ps and/or volumetric hardening part ε̇pc
∂f s e ∂f s e ∂g s ∂f s ∂κs ∂g s
D ε̇ − λs D + λs s ps =0 (A.31)
∂σ ∂σ ∂σ ∂κ ∂ε ∂σ
1 ∂f s e
⇒ λs = D ε̇ (A.32)
H s − ds ∂σ
with the shear hardening modulus
∂f s ∂κs ∂g s
Hs = (A.33)
∂κs ∂εps ∂σ
and
∂f s e ∂g s
ds = D , (A.34)
∂σ ∂σ
137
Appendix A On constitutive models as used in this thesis
∂f c ∂f c c ∂f c ∂f c ∂κs pc
f˙c = σ̇ + κ̇ = σ̇ + ε̇ = 0 (A.35)
∂σ ∂κc ∂σ ∂κc ∂εpc
∂f c e ∂f c e ∂g c ∂f c ∂κc ∂g c
D ε̇ − λc D + λc c pc =0 (A.36)
∂σ ∂σ ∂σ ∂κ ∂ε ∂σ
c 1 ∂f c e
⇒λ = c D ε̇ (A.37)
H − dc ∂σ
with the compression hardening modulus
∂f c ∂κc ∂g c
Hc = (A.38)
∂κc ∂εpc ∂σ
and
∂f c e ∂g c
dc = D , (A.39)
∂σ ∂σ
Over Consolidation Ratio (OCR) Another input parameter which is embedded in the
HS Model is the Over Consolidation Ratio (compare Section 4.4.1). When using the
HS Model to consider over consolidated grounds (OCR > 1), initial stresses may be
modelled according to Eq. 4.11. The elastic unloading from the maximum vertical over-
burden stress down to the actual initial stress implies that the actual stress state is not
on the yield surface. Contrary to a normally consolidated initial stress state, with the
actual stress state located directly on the HS-yield surface, the initial stress state of an
overconsolidated ground is in the elastic region, shown in Fig. A.3. Due to the higher
unloading-reloading stiffness Eur , which is used to model the elastic region, overconsol-
idated grounds in the HS Model have an initially much stiffer response than normally
consolidated grounds.
For more information on the formulation of yield functions and for a more detailed
description of the implementation of the HS-Model, the reader is referred to S CHANZ
(1998) and B RINKGREVE and V ERMEER (2001).
138
A.3 The HS-Small Model
q = s1 - s3
1
E50
Eur E0
1
e1
Figure A.5: HS-Small Model: extension of the HS Model incorporating small strain stiff-
ness
soil particles or frictional particle forces exceeding their elastic limit. Thus, a drop of stiff-
ness can be observed whenever inter-particle forces are reorganized and concentrated.
As shown in Fig. A.6, for strains higher than 10−5 a rapid drop of small strain ground
stiffness is measured, considering shear modulus G. The strain levels obtained here,
are far below conventional laboratory testing, requiring special measuring devices such
as dynamic methods or local strain gauges. To incorporate small strain stiffness effects
into the HS Model a relatively simple expression for the small strain stiffness decay of the
shear modulus, similar to the one suggested by S ANTOS and C ORREIA (2001), is adopted
G0
G= γ , (A.40)
1 + 0.43 · γ0.7
where G is the actual shear modulus at shear strain γ, G0 is the initial shear modulus
and γ0.7 is the shear strain at which the initial shear modulus has reduced to 0.7 · G0 , as
shown in Fig. A.6. For general states of stress the shear strain is expressed using the
strain invariant
1 p
γ = √ · (ε1 − ε2 )2 + (ε2 − ε3 )2 + (ε3 − ε1 )2 , (A.41)
2
which in the special case of triaxial loading reduces to γ = |ε1 − ε3 |. While reducing the
shear modulus with increasing shear strain, the Poisson’s ratio νur is kept constant, such
that the resulting bulk modulus is not a constant but is also reducing as a function of
shear strain.
Conform the ideas by O HDE (1951) and the formulation of stiffnesses in the HS Model,
the initial shear modulus G0 is pressure dependent according to the equation
µ 0 ¶m
ref c · cot ϕ0 + σ30
G0 = G0 · 0 , (A.42)
c · cot ϕ0 + pref
The magnitude of Gref0 is strongly correlated to the porosity of the soil. A typical corre-
ref √
lation being used is G0 = 450 · pref by B IAREZ and H ICHER (1994).
139
Appendix A On constitutive models as used in this thesis
Dynamic methods
Gur
Figure A.6: Small strain stiffness curve for a particular soil adopting shear modulus
Fig. A.6 shows the stiffness degradation curve, reaching far into the plastic material
behavior at larger strains. According to the formulation of the HS Model, stiffness degra-
dation due to plastic straining is modelled by involving material hardening. Therefore,
before reaching plastic material behavior, the formulation of the small strain stiffness
curve is cut off at the unloading-reloading shear modulus Gur , defined as
Eur
Gur = . (A.43)
2 · (1 + νur )
The elastic constants Eur and νur have already been introduced in the HS Model.
Eq. A.43 indicates that Gur is the shear modulus in complete deviatoric unloading as
illustrated in Fig. A.5.
Besides the input parameters as introduced for the HS Model, there are a set of two
additional input parameters for the HS-Small Model: the elastic small strain shear mod-
ulus Gref
0 at reference pressure pref and the threshold value γ0.7 in primary loading. A
more detailed explanation of the HS-Small Model can be found in B ENZ (2006).
140
Bibliography
S. B ABENDERERDE and J. H OLZH ÄUSER. Taschenbuch für den Tunnelbau 2000, vol-
ume 24, chapter Betriebszustand Druckluftstützung beim Hydroschild, pages 231–
252. Glückauf, Essen, 2000.
K. J. B AKKER. Soil Retaining Structures, Development of Models for Structural Analysis. Phd
thesis, Technical University of Delft, 2000.
M. B AUDENDISTEL. Rock Mechanics, chapter Zum Entwurf von Tunneln mit großem
Ausbruchsquerschnitt, pages 75–100. Number Suppl. 8. 1979.
141
Bibliography
T. B ENZ. Small-Strain Stiffness of Soils and its Numerical Consequences. PhD thesis, Institute
of Geotechnical Engineering, University of Stuttgart, Stuttgart, 2006.
J. B IAREZ and P.-Y. H ICHER. Elementary mechanics of soil behaviour. Technical report,
Ecole Centrale de Paris, Cedex, France, 1994.
R. B. J. B RINKGREVE and P. A. V ERMEER. P LAXIS Finite Element Code for Soil and Rock
Analyses. A. A. Balkema, Rotterdam, 2001.
E. T. B ROWN, J. W. B RAY, B. L ADANYI, and E. H OEK. Ground response curves for rock
tunnels. Journal of Geoechnical Engineering ASCE, 109(1):15–39, 1983.
A. B ULL. Stresses in the linings of shield-driven tunnels. ASCE Soil Mechanics and Foun-
dations Devision, pages 1363–1394, 1944.
G. W. C LOUGH and B. S CHMIDT. Soft Clay Engineering, chapter Design and performance
of excavations and tunnels in soft clay, pages 569–634. Elsevier, 1981.
R. N. C RAIG and A. M. M UIR W OOD. A review of tunnel lining practice in the united
kingdom. Supplem. 335, Transport and Road Research Laboratory, 1978.
D. J. C URTIS. Correspondence on muir wood, a.m. : The circular tunnel in elastic ground.
Geotechnique, 25(1):231–237, 1975.
142
Bibliography
H. D UDDECK and J. E RDMANN. Structural design models for tunnels. Tunnelling 82,
pages 83–91, 1982.
J. M. D UNCAN and C.-Y. C HANG. Nonlinear analysis of stress and strain in soils. Journal
of the Soil Mechanics and Foundations Division ASCE, (96, SM 5):1629–1653, 1963.
143
Bibliography
R. J. G UEDES and C. S ANTOS P EREIRA. The role of the soil k0 value in numerical analysis
of shallow tunnels. In Proc. Int. Symp. Geotechnical Aspects of Underground Construction
in Soft Ground, IS-Tokyo ’99, pages 379–384, Tokyo, 2002.
E.-D. H ORNIG and R. F. B UCHMAIER. Laboratory and field tests investigating the stress
strain behaviour of weathered Keuper Mudstone. In Proc. of Int. Conf. on Problematic
Soils, volume 1, pages 389–396, Famagusta, N. Zypern, May 2005. Eastern Mediter-
ranean University.
ITA. Views on structrural design models for tunnelling. Advances in tunnelling technology
and subsurface use, 2(3):153–229, 1982.
ITA. Guidelines for the design of tunnels. Tunnelling and Underground Space Technology,
3(3):237–249, 1988.
T. K ASPAR and G. M ESCHKE. A 3d finite element simulation model for tbm tunnelling
in soft ground. Int. J. Numer. Anal. Meth. Geomech., 28:1441–1460, 2004.
144
Bibliography
B. M AIDL, U. M AIDL, and N. R USE. Erfahrung mit der FEM-Simulation im Rahmen des
Prozesscontrollings beim Schildvortrieb. Bauingenieur, 80:337–342, 2005.
145
Bibliography
146
Bibliography
J. A. S ANTOS and A. G. C ORREIA. Reference threshold shear strain of soils. its appli-
cation to obtain an unique strain-dependent shear modulus curve for soil. In 15th
International Conference on Soil Mechanics and Geotechnical Engineering, volume 1, pages
267–270, Istanbul, Turky, 2001.
B. F. S CHMIDT. Settlements and ground movements associated with tunnelling in soils. Phd
thesis, University of Illinois, Urbana, 1969.
H. S CHULZE and H. D UDDECK. Stresses in shield driven tunnels. Beton und Stahlbeton-
bau, (8):169–175, 1964b.
147
Bibliography
I. M. S MITH and D. V. G RIFFITH. Programming the Finite Element Method. J. Wiley & Sons,
Chisester, UK, 2 edition, 1982.
A. M. TALMON and A. B EZUIJEN. Grouting the tail void of bored tunnels: the role of
hardening and consolidation of grouts. In 5th Int. Symposium on Geotechnical Aspects of
Underground Construction in Soft Ground, pages 319–325, 2005.
H. VAN L ANGEN and P. A. V ERMEER. Automatic step size correction for non-associated
plasticity problems. International Journal for Numerical Methods in Engineering, 29:579–
598, 1990.
E. W INKLER. Die Lehre von der Elasticitaet und Festigkeit, volume 1. Theil. Verlag v. H.
Dominicus, Prag, 1867.
W. W INKLER. Anwendung der Federanalogie bei der statischen Berechnung von kreis-
förmigen Tunnelprofilen. Schweizerische Bauzeitung, 88(44):991–993, 1970.
O. C. Z IENKIEWICZ and R. L. TAYLOR. The Finite Element Method. McGraw Hill, London,
4 edition, 1991.
148
Curriculum Vitae
Nationality: German
Education: 09/82-07/92
Free Waldorf School Rengoldshausen/Überlingen
08/92-01/93
The Toronto Waldorf School, Canada
02/93-07/95
Free Waldorf School Rengoldshausen/Überlingen
Studies: 04/97-07/01
Civil Engineering, University of Stuttgart
Specialization in: Geotechnical engineering,
structural engineering and construction operation
149
Mitteilungen des Instituts für Geotechnik
ISBN-10: 3-921837-54-5
ISBN-13: 978-3-921837-54-2