The 12th International Conference of

International Association for Computer Methods and Advances in Geomechanics (IACMAG)

1-6 October, 2008
Goa, India

TIME DEPENDENT DEFORMATIONS IN SQUEEZING TUNNELS

G. Barla, M. Bonini and D. Debernardi

Department of Structural and Geotechnical Engineering, Politecnico di Torino, Italy

Keywords: tunnelling, squeezing, time dependent behaviour, numerical modelling

ABSTRACT: This paper deals with full face excavation of large size tunnels in rock masses of very poor quality
which exhibit squeezing behaviour. Time dependence is accounted for explicitly by using constitutive models
which have been validated with reference to laboratory testing and in situ performance monitoring. The case
study of the Saint Martin La Porte access adit (Lyon-Turin Base Tunnel) is taken as illustration. In this adit an
innovative excavation-construction method is being used to cope with the severely squeezing Carboniferous zone
encountered. An essential feature of this method is a two-stage excavation sequence with installation of a
composite pre-reinforcement/reinforcement system with deformable elements embedded in the shotcrete lining.
The tunnel response has been analysed by both semi-analytical and numerical solutions with close attention paid
to the constitutive models adopted. Results of modelling are compared with convergence monitoring data.

1 Introduction
Tunnel construction in squeezing conditions is very demanding due to the difficulty in making reliable predictions
at the design stage. During excavation such conditions are not easily anticipated, even when driving into a
specific geological formation and experience is gained on the squeezing problems encountered. Squeezing
conditions may vary over short distances due to rock heterogeneity and fluctuations in the mechanical and
hydraulic properties of the rock mass. Indeed, the selection of the most appropriate excavation-construction
method (i.e. mechanized tunnelling versus conventional tunnelling) is highly problematic and uncertain. Due to
the fixed geometry and the limited flexibility of the TBM (Tunnel Boring Machine) allowable space to
accommodate ground deformations is restricted. On the contrary, in conventional tunnelling a considerably larger
profile can be excavated initially in order to allow for large deformations. The obvious consequence is that in deep
tunnels, whenever severely squeezing conditions are anticipated, conventional tunnelling appears to be preferred
over mechanized tunnelling.
Squeezing is essentially a time dependent behaviour although for design purposes the rock mass which
undergoes squeezing is often represented as an equivalent elastic-plastic medium with strength and deformability
parameters which are down-graded based on observation and monitoring during excavation. The so called “short
term” and “long term” conditions are often invoked, characterized by different values of the parameters involved in
the constitutive model being used. However, there is no doubt that under the most severe squeezing conditions
an appropriate representation of the tunnel response is obtained only by using constitutive models which account
for time dependent behaviour. This originates from the fact that time dependent deformations are observed
whenever face advancement is stopped and these are likely to take place during excavation, when it is difficult to
distinguish the “face effect” from the “time effect”.
This paper deals with full face excavation of large size tunnels in rock masses which exhibit squeezing behaviour.
This is considered explicitly by using three constitutive models which have been validated during the years
following the 11th IACMAG Conference in Torino (Barla, 2005). The case study of the Saint Martin La Porte
access adit along the Lyon-Turin Base Tunnel is taken as illustration.

2 Squeezing behaviour
The term “squeezing” originates from the pioneering days of tunnelling through the Alps. It refers to the reduction
of the tunnel cross section that occurs as the tunnel is being advanced (Figure 1). Based on the work of a
Commission of the International Society for Rock Mechanics (ISRM), which has described squeezing and the
main features of this mechanism, it is agreed that “squeezing of rock” stands for large time dependent
convergence during tunnel excavation. This happens when a particular combination of material properties and
induced stresses causes yielding in some zones around the tunnel, exceeding the limiting shear stress at which
creep starts. Deformation may terminate during construction or continue over a long period of time.

1
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

The magnitude of tunnel convergence, the rate of deformation, and the extent of the yielding zone around the
tunnel depend on the geological and geotechnical conditions, the in situ state of stress relative to rock mass
strength, the groundwater flow and pore water pressure, and the rock mass properties. Squeezing is therefore
synonymous with yielding and time-dependence, and often is largely dependent on the excavation and support
techniques being used. If the support installation is delayed, the rock mass moves into the tunnel and a stress
redistribution takes place around it. On the contrary, if deformation is restrained, squeezing will lead to long-term
load build-up of the support system.

Figure 1. Squeezing rock reduces the tunnel cross section. This is shown in this photograph
where re-profiling of a highly deformed cross section is taking place
in the Saint Martin access adit (Lyon-Turin Base Tunnel).

3 Time-dependent constitutive models

In order to describe the tunnel response associated with severely squeezing conditions, time dependent
constitutive models need be used. In the following a viscoelastic-plastic model (CVISC), an elastic-viscoplastic
model (VIPLA) and a more complex elastic-plastic-viscoplastic model (SHELVIP) are briefly described.

3.1 CVISC model

The CVISC model (Itasca, 2006) is an analogical model which couples, in series, the Burgers viscoelastic model
(i.e. Kelvin and Maxwell models in series) with a plastic flow rule, based on the Mohr-Coulomb yield criterion, as
shown in Figure 2.
K

ε vol G εq
K c,φ,ψ,σt ηM G
M
c,φ,ψ,σt
p q

ηK
MOHR MOHR
ELASTIC COULOMB KELVIN MAXWELL COULOMB

(a) Volumetric behaviour (b) Deviatoric behaviour

Figure 2. Sketch of the CVISC model: (a) volumetric behaviour, and (b) deviatoric behaviour.

The volumetric behaviour is only elastic-plastic and is governed by the linear elastic law and the plastic flow rule
(Figure 2.a), while the deviatoric behaviour is viscoelastic-plastic and is driven by the Burgers model and the
same plastic flow rule (Figure 2.b). This means that the viscoelastic strains are deviatoric and depend only on the
deviatoric stress state; instead the plastic strains are both deviatoric and volumetric and depend on the global
stress in accordance with the chosen flow rule.

2
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

3.2 VIPLA model

The VIPLA model (Lemaitre & Chaboche, 1996) is based on Perzyna’s overstress theory (Perzyna, 1966) which
states that the strain rate tensor ε&ij can be split into elastic ε&ije and viscoplastic ε&ijvp components, to give:

ε&ij = ε&ije + ε&ijvp (1)

The viscoplastic strain rate tensor ε&

vp
ij can be calculated by the following flow rule:
∂g
ε&ijvp = γ ⋅ Φ ( F ) ⋅ (2)
∂σ ij

where γ is the fluidity parameter, F is the over-stress function, representing the distance from the yield surface
f = 0 , Φ ( F ) is the so-called viscous nucleus, g is the viscoplastic potential and σ ij the stress tensor.
The time-dependency is introduced in this model by modifying the classical flow rule of elastoplasticity and by
discarding the consistency rule ( df = 0, f ≤ 0 ) , thus allowing the yield function f to be positive or negative. The
viscoplastic potential g defines the direction of ε&ijvp , while F influences its modulus by means of the viscous
nucleus Φ .
In the VIPLA model F is assumed to be represented by the yield function f and Φ is assumed to be a power
law:
n n
Φ= F = f
(3)
where n is a constitutive parameter ( n ≥ 1) .
The yield function f is splitted into a part f , which depends only on the stress state, and a part κ , which
depends only on the viscoplastic strain, according to:

f (σ ij )
f = (4)
κ ( ε ijvp )

For the function f the Von Mises yield criterion is assumed:

f (σ ij ) = q (5)

where q is the stress deviator.

A potential hardening is introduced by means of the function κ :

κ ( ε ijvp ) = ( ε qvp )
−m
n
(6)

where m is a constitutive parameter (1 − n < m ≤ 0 ) and ε qvp is the deviatoric viscoplastic strain,
ε qvp = 4 3 ⋅ J 2,ε vp where J2,ε vp is the second invariant of the viscoplastic strain deviator.
Under these assumptions, the yield surface f = 0 is reduced to the hydrostatic axis and it does not change with
time.
The viscoplastic potential g is taken to be equal to f (i.e. the flow rule is associated). With these assumptions,
the viscoplastic strains depend only on the deviatoric stress state and do not induce volumetric strains. Therefore,
Eqn. (2) becomes:
3
( )
m
ε&ijvp = ⋅ γ ⋅ q n −1 ⋅ ε qvp ⋅ sij (7)
2
The constitutive parameters n and m define respectively the dependence of the viscoplastic strain rate tensor
on the deviatoric stress and on the equivalent viscoplastic strain, whereas the parameter γ defines the amplitude
of the viscoplastic strains.

3.3 SHELVIP model

The SHELVIP model (Stress Hardening ELastic VIscous Plastic model) is derived from the Perzyna’s overstress
theory, by adding a time independent plastic component (Debernardi, 2008). Therefore it is possible to split the
strain rate tensor ε&ij into elastic ε&ije , plastic ε&ijp , and viscoplastic ε&ijvp components, to give:
ε&ij = ε&ije + ε&ijp + ε&ijvp (8)

According to the classical theory of elastoplasticity, the time-independent plastic strains ε ijp develop only when
the stress point reaches the plastic yield surface fp = 0 (Figure 3), defined by the Drucker-Prager criterion:
fp = q − α p ⋅ p − k p (9)

3
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

The plastic strains ε ijp can be evaluated using the classical flow rule of elastoplasticity:
∂g p
ε ijp = λ ⋅ (10)
∂σ ij

where g p is the plastic potential, g p = q − ωp ⋅ p , that defines the direction of ε ijp , ωp is the plastic dilatancy and
λ is the plastic multiplier, that can be determined using the consistency condition dfp = 0, fp ≤ 0 .

q Plastic
fp = 0

ε ijp g p = const
P2
q2
Viscoplastic
ε ijvp
P1
q1 fvp = 0
fvp
gvp = const

σt p1,2 p

Figure 3. Schematization of the SHELVIP model in the p-q plane.

The viscoplastic strain rates ε&ijvp develop only if the effective stress state exceeds a viscoplastic yield surface
fvp = 0 (Figure 3) which is also defined by the Drucker-Prager criterion. This surface is internal to the plastic yield
surface and intersects the p -axis at the same point as the plastic yield surface. Thus, it is possible to write:
⎛ k ⎞
fvp = q − α vp ⋅ ⎜ p + p ⎟ (11)
⎜ α p ⎟⎠
⎝
where αvp is a visco-hardening parameter that defines the internal viscous state of the material.
The viscoplastic ε&ijvp strain rate can be determined using the flow rule of Perzyna’s overstress theory:

∂gvp
ε&ijvp = γ ⋅ Φ ( F ) ⋅ (12)
∂σ ij

The overstress function F is assumed to be equal to the viscoplastic yield function fvp and the viscous nucleus
Φ to be a power law:
n n
Φ= F = fvp (13)

where n is a constitutive parameter.

The viscoplastic potential gvp is assumed to be gvp = q − ωvp ⋅ p , where ωvp is the viscoplastic dilatancy.
The hardening of the viscoplastic yield surface is governed by the differential equation:
n ⋅m
l fvp ⎛f ⎞
α&vp = ⋅ ⋅ ⎜ vp ⎟ (14)
m ⋅ n p + kp α p ⎝q ⎠
where m and l are constitutive parameters.

4 The Saint Martin La Porte access adit

The CVISC, VIPLA, and SHELVIP constitutive models have been used to back analyse the tunnel response
based on convergence monitoring in representative sections of the Saint Martin La Porte access adit. An

4
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

illustration of the geological conditions and of the excavation-support system adopted in this tunnel is given
below.
4.1 Geological conditions and excavation-support systems

The Saint Martin La Porte access adit (Figures 4.a and 4.b) is being excavated in the Carboniferous Formation,
“Zone Houillère Briançonnaise-Unité des Encombres“ (hSG in Figure 4.a), which is composed of black schists (45
to 55%), sandstones (40 to 50%), coal (5%), clay-like shales and cataclastic rocks. A characteristic feature of the
ground observed at the face during excavation (Figure 4.b) is the highly heterogeneous, disrupted and fractured
conditions of the rock mass which exhibits very severe squeezing problems. The formation is affected often by
faulting that results in a degradation of the rock mass conditions. The overburden along the tunnel in the zone of
interest ranges from 300 m to 550 m. Excavation takes place in essentially dry conditions.
In order to assess the rock mass quality during excavation, detailed mapping of the geological conditions at the
face was undertaken as depicted in Figure 4.b. This provides information to evaluate the percent distribution of
“strong” (sandstones and schists) and “weak” (coal and clay-like shales) rocks at the face.

(a) (b)
Figure 4. (a) Geological profile along the Saint Martin La Porte access adit and (b) typical geological
conditions at the face at chainage 1480 m (gps-sandstones, a-clay like shales, c-coal, etc.).

Several support systems were used in the Carboniferous zone. However, it soon became apparent that a stiff
support would not be feasible in the severely squeezing conditions encountered. The design concept finally
chosen (Figure 5.b) was based on allowing the support to yield while using full-face excavation with systematic
face reinforcement by fiber-glass dowels. The support system initially implemented (Figure 5.a) consisted of
yielding steel ribs with sliding joints (TH, Toussaint-Heintzmann type), rock anchors and a thin shotcrete layer in a
horseshoe profile. These sections of the tunnel underwent very large deformations with convergences up to 2 m
and later needed to be re-profiled.

10 Boulons Swellex Mn12

L= 4m 1u/m
Ligne B
X = 0.000
Y = 9.490

Ligne A X = 0.000
8.840
Y = 8.990
voute parapluie m
1. 00
+ 8.84
20 Boulons autoforants
L= 8m esp 50cm
Présoutènement
Boulons fibre de verre
54 unités
recouvrement 6m maxi.
Ligne F R5.40

m
50
55

0. (voir plan T240)

R5.

.05
R6
eA

B
Lign

ne

Ligne F
Lig

50°
50 °
32°
64
°

Treillis soudé
maille (150x150Ø8)

X = -6.050 X = 6.050
Y = 3.440
Y = 3.440
X =X-5.550
= -5.400 6.55 6.55 X = 5.550 X = 5.400
Y =Y 3.440
= 3.440 Y = 3.440 Y = 3.440
+3.44

5 à 10 cm de béton projeté RIG

avec fibres métalliques Coque de beton projeté fibré Ligne F R11.95 Autoforants L=8.00m,
immédiatement en parement y compris en contre-voûte avec 1 tôle de blindage S4 sur 2
et 10 cm sur front de taille (épaisseur à définir)
24 u/ml en piedroit et voûte.
Ligne A R12.1
0

Ligne
Ligne
Ligne A
Ligne B R11.10

B R12
.60
Reference: +1.00m
F R1
R10.60

0.45

X = 4.740
2%
0,200

XY= =4.862
1%
X = -4.740
Y =X-0.475
= 4.862
-0.475
Y = -0.581 Y = -0.581

Treillis soudé
(150x150x8)
Coque de beton à définir
X = -5.267 X = -0.142
Y = -0.933 X = 0.000 Y = -1.762
Y = -1.612
X = 5.267
1 nappe de Y = -0.933
Treillis soudé ST30 - 1.612
en contre-voûte 10 Boulons Autoforants
L= 8m 1u/m²
10Largeur
Boulons Autoforants
radier 8.74 en contre-voûte
Plan du radier:
L= 8m 1u/m
X = 0.000
voir plan N° EXE2098RZ--PLTRS-GS--T232 Treillis soudé Y = -2.262

150x150 Ø8 + écarteurs

(a) (b)
Figure 5. Tunnel cross section showing the excavation-support systems adopted in the Saint Martin La Porte
access adit between chainage 1267 and 1324 m (P7.3, a) and chainage 1325 and 1700 m (DSM, b).

5
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

In order to improve the working conditions and to control deformations, a novel support system was implemented
with a near circular cross section. This can be summarized as follows (Figure 5.b):
• Stage 0: face pre-reinforcement, including a ring of grouted fiber-glass dowels around the opening, designed
to reinforce the rock mass ahead and around the tunnel perimeter over a 2 to 3 m thickness.
• Stage 1: mechanical excavation carried out in steps of one meter length, with installation of a support
system consisting of untensioned rock anchors (length 8 m) along the perimeter, yielding steel ribs with
sliding joints (TH type), and a 10 cm thick shotcrete layer. The tunnel is opened in the upper cross section to
allow for a maximum convergence of 600 mm.
• Stage 2: the tunnel is opened to the full circular section at a distance of 15-25 m from the face, with application
of 20 cm shotcrete lining, yielding steel ribs with sliding joints (TH type) with 9 longitudinal slots (one in the
invert) fitted with HiDCon (High Deformable Concrete) elements. The tunnel is allowed to deform in a controlled
manner to develop a maximum convergence which should not exceed 400 mm.
• Stage 3: installation of a coffered concrete ring at a distance of 80 m from the face.

4.2 Controlled response of the tunnel deformation

Systematic monitoring of tunnel convergence is underway along the tunnel. Convergences are measured by
means of optical targets placed along the tunnel perimeter. A number of special sections have been equipped
with multi-position borehole extensometers and strain meters located across the HiDCon elements. Extrusometer
monitoring has been used to measure the longitudinal displacement ahead of the tunnel face. In addition, the
strain level in the primary lining has been monitored (Barla et al., 2007).
In order to gain in the understanding of the tunnel response so far, it is of interest to consider Figure 6.a, which
shows the tunnel “deformation” that has occurred (i.e. the convergence divided by the length of each array
measured at the time of installation of the optical targets) in stage 1 along arrays 1-3, 3-5 and 1-5 (Δli-j) between
chainage 1200 m and 1700 m, with the tunnel face being 15 m ahead of the monitoring section. Also illustrated in
Figure 6.b are the convergences occurred along array 1-5 in stage 2 after 30, 80 and 120 days from the stage 2
installation. Three stops of face advance, represented in the figures by the vertical lines, took place along the
tunnel length with the DSM cross section implemented.
The following observations can be made:
• Large deformations are associated with cross section P7.3 between chainages 1200 and 1400 m; with cross
section DSM the convergences in stage 1 generally are smaller with the tunnel strain never in excess of 6-7 %.
• The 600-mm allowed convergence with cross section DSM has been exceeded locally (e.g., at chainage
1478 m due to the stop at chainage 1494 m and between chainages 1525 and 1550 m where the rock mass
quality was very poor) and required re-profiling of the tunnel cross section before installing the composite
lining adopted in stage 2.
• The tunnel deformation associated with cross section P7.3 appears to be rather different in one section with
respect to the neighbouring one, which is not the case for cross section DSM.
• With chainage 1550 m approximately the tunnel response appears to exhibit smaller convergences in line
with the improved rock mass conditions encountered, which appear even more encouraging around
chainage 1700 m.

array 1-3 array 3-5 array 1-5 600mm limit (stage1)

stage2-30d stage2-80d stage2-120d 400 mm limit (stage2)
12
P2→ P7.3 P7.3-DSM DSM 800
P7.3-DSM DSM
1494

1545

1605

10 700 6
1545

1605

3
2 4 600 5
8
500
Δ l1-5 (mm)

5
Δ li-j (%)

1
Δ l1-5 (%)

4
6 400
3
300
4
2
200
1494

2 1
100

0 0 0
1200 1300 1400 1500 1600 1700 1300 1400 1500 1600 1700
Chainage (m) Chainage (m)
(a) (b)
Figure 6. (a) Deformations measured along arrays 1-3, 3-5 and 1-5 at 15 m from face in phase 1
and (b) convergences at 30, 80 and 120 days following excavation with stage 2 installed.

6
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

It also is important to consider the tunnel convergence versus time in stage 2 as depicted in Figure 6.b along
array 1-5. This occurs at a significant distance from the advancing face and when the yielding support has been
active for a certain time and the final concrete lining has not yet been installed. It is noted that between chainages
1450 and 1525 m the tunnel cross section experienced deformations in excess of that allowed (400 mm). In such
a case the HiDCon elements on the right wall (looking at the tunnel face) attained 40% limit strain and visible
overstressing occurred in them. This did not cause any significant problem as no difficulties were encountered
before installing the final lining.
A back analysis of the monitored convergence data between chainages 1394 and 1507 m has been performed,
using the following time-dependent relationship (Sulem et al., 1987):

⎡ ⎛ X ⎞2 ⎤ ⎪⎧ ⎡ ⎛ T ⎞0.3 ⎤ ⎪⎫
C ( x, t ) = C∞, x ⋅ ⎢1 − ⎜ ⎟ ⎥ ⋅ ⎨1 + m ⋅ ⎢1 − ⎜ ⎟ ⎥⎬ (15)
⎢⎣ ⎝ x + X ⎠ ⎥⎦ ⎩⎪ ⎢⎣ ⎝ t + T ⎠ ⎥⎦ ⎭⎪

where C(x,t) is the convergence at the distance x from the tunnel face and at the time t, C∞,x is the convergence
at distance x obtained in the case of an infinite rate of face advance (no time dependent effect), m is a non
dimensional parameter which depends on the ground conditions, X is a distance related to the distance of
influence of the face (for an elastic plastic model of behaviour X = 0.84⋅Rpl, with Rpl taken as the plastic radius of
the tunnel), T is a characteristic parameter of the rock mass time dependent properties.
The results obtained are reported in Table 1. It is noted that the distance of influence of the face (this length may
be estimated to be four times the value of X, Sulem et al., 1987) is considerable; also, the time dependent
properties of the rock mass, which are related to the T parameter, are representative of the severity of the
squeezing conditions encountered. The variability of parameters found out for C∞,x and X are due to the different
stiffness of the support installed more than a substantial change of geomechanical conditions of the rock mass.
Moreover the similarity of time related parameters (T and m) signifies that the time-dependency is comparable.

Table 1. Representative parameters for the Saint Martin La Porte tunnel.

C∞x X T m
[mm] [m] [days] [-]
P7.3 profile* (1272-1322 m) 1563.1 81.1 25.0 1.29
DSM profile - phase 1* (1394-1507 m) 602.6 41.2 35.3 1.09
* the parameters have been evaluated according to the monitoring data
between the chainages shown.

5 Modelling the tunnel response

The three constitutive models (CIVISC, VIPLA, and SHELVIP) described above have been adopted in order to
analyse the tunnel response in terms of convergence monitored during excavation. As an illustration of the
application of the VIPLA model the semi-analytical solution due to Nguyen-Minh and Pouya (1992) is discussed
first. Then, numerical modelling by the Finite Difference Method and the FLAC code is applied in conjunction with
the CIVISC and the SHELVIP constitutive models respectively. It is underlined that the purpose here is to see
how three different constitutive models with different levels of complexity reproduce the time dependent
deformations of the tunnel in different cross sections. It is implied that the experience gained is essential in order
to be able to predict the tunnel behaviour with the expected higher overburden to be encountered as the
excavation proceeds. The information provided is of relevance if one is to optimize the type of support adopted
and the excavation/support sequence.
The numerical analyses presented are for two different cross sections, namely at chainage 1311 m and 1406 m,
where the P7.3 and DSM support type were installed. One may question why so distant sections characterised by
different support types were taken for testing respectively the CIVISC and the SHELVIP constitutive models. It is
noted that for P7.3 the influence of the preliminary lining stiffness is negligible, so that axisymmetric modelling
with total stress release on the tunnel boundary can be favourably implemented and interpreted. On the other
hand, the study of the DSM cross section calls for an accurate representation of geometry, construction stages
and properties of the structural elements adopted, which may be better achieved by modelling the problem in
plane strain conditions.

5.1 Semi-analytical solution for the P7.3 cross section (VIPLA model)
A simple time dependent analysis of a tunnel which applies the VIPLA model can be obtained with the solution
proposed by Nguyen-Minh and Pouya (1992). The main assumptions of this solution are as follows: (a) the tunnel
is of circular section, is not lined and is sufficiently deep; (b) the problem can be treated in plane strain conditions;
(c) the initial stress state is isotropic and homogeneous; the ground is homogeneous, isotropic, and
incompressible (ν=0.5). Under these assumptions the radial viscoplastic strain rate ε&rvp at the distance r from the

7
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

tunnel centre at time t is given by the following expression:

β
⎧1 ⎧ 1
⎫⎫α
∂ ⎛ε ⎞
vp
⎪α 1 1
1 ⎪ ⎡ β −1
⎤ 1− β
⎪⎪
⎟ = ⎨u ⎡⎣1 + ( β − 1) u τ ⎤⎦ + u ⎨1 − ⎢1 + ( β − 1) u τ ⎥ ⎬⎬
β −1 α 1− β β β α
⎜
r
(16)
∂τ ⎝ ε ⎠
e
R ⎪ β ⎪ ⎣⎢ ⎦⎥ ⎪⎪
⎩ ⎩ ⎭⎭

1
β +1
1 n ⎛γ ⎞
α ⎛ ε Re ⎞α 3 σ0 ⎛ 3⎞ ⎛R⎞
2

α= ; β = ; a=⎜ ⎟ ; τ = t ⎜ ⎟ ; ε Re = ; acyl = a ⎜ ; u=⎜ ⎟

⎜ 2 ⎟⎟
(17)
1− m 1− m ⎝α ⎠ ⎜ a ( 2σ ) β ⎟ 2 E ⎝ ⎠ ⎝r ⎠
⎝ cyl 0 ⎠

where R is the tunnel radius, σ 0 the isotropic stress state, ε Re the elastic radial strain at tunnel contour, E the
elastic modulus, γ , m and n the constitutive parameters of the VIPLA model. With a double numerical
integration, over time and over the tunnel radius, it is possible to obtain the tunnel radial displacement.
In order to account for the influence of the face on the tunnel radial displacement, the isotropic stress state need
be reduced according to (Panet, 1995):

σ0 ⎛ 0.84 ⎞
= 0.28 + 0.72 ⎜ 1 − ⎟ (18)
σ0 ⎝ 0.84 + x R ⎠
where x is the distance from the face and σ 0 the reduced isotropic stress state which accounts for the face
influence.
By fitting with the least square algorithm the radial displacements computed from the monitored tunnel
convergences at chainage 1311 m the constitutive parameters of the VIPLA model are obtained as shown in
Table 2. Figure 7 shows the comparison between the computed and the monitored radial displacements.

1200 2-3

1000 1-3
Radial displacement [mm]

1-5
2-4
800
Table 2. Constitutive parameters 3-5
of the VIPLA model 3-4
600
(time in year and pressure in kPa)

E 1180 MPa 400 VIPLA model

γ 2.25E-25
m -1.32 200
n 6.16
0
0 20 40 60 80 100
days

Figure 7. Computed (mean value) versus monitored radial

displacements at chainage 1311 m.

5.2 Numerical modelling of the DSM cross section (CVISC model)

For the purpose of numerical modelling with the CVISC model the DSM cross section at chainage 1406 m has
been chosen. The overburden is 400 m and the initial stress state is assumed to be isotropic and equal to
10 MPa. The analyses were performed with the Finite Difference Method and the FLAC code (Itasca, 2006).
Figure 8.a shows the finite difference grid adopted in close proximity to the tunnel. The grid is made of square
elements which increase in size away from the tunnel in order to minimize error propagation and solution time.
Symmetry is assumed and half the tunnel section is modelled in plane strain conditions. The constitutive
parameters for the CVISC model and the lining in stage 2 are shown in Table 3.
The model is to allow for simulation of the excavation/construction sequence in two stages. Overall five simulation
steps have been implemented in order to represent this as accurately as possible. Following simulation of the
initial stress state, the excavation of the top heading is performed with a stress release equal to 88% of the initial
stress sate. The presence of the rock dowels around the tunnel and of the primary lining is simulated with a
200 kPa uniform pressure applied on the same tunnel boundary. The ground is initially taken as elastic perfectly

8
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

plastic and the time dependent behaviour during stage 1 is simulated by running the CVISC model for 23 days
with the intent to match the tunnel response as observed through monitoring.
Following the invert excavation, where the rock mass behaviour is elastic perfectly plastic, the composite
deformable lining is placed in the model, the applied pressure on the tunnel boundary is removed and the stress
reduction process is completed. The composite deformable lining is modelled by beam elements, with a
deformation modulus equal to 30000 MPa and a Poisson’s ratio equal to 0.2. The presence of the HiDCon
elements incorporated in the lining is simulated by beam elements with a linearly elastic ideally plastic behaviour
and the mechanical properties listed in Table 3, where the mechanical parameters of the rock mass are also
given.
An interface is introduced between the lining and the rock surround with normal (kn) and shear (ks) stiffness equal
to 3.45⋅103 MPa/m and 3.45⋅102 MPa/m respectively. The analysis of stage 2 is obtained by running the CVISC
model for 77 days with the intent to reproduce the observed tunnel response versus time with the convergence
curve shown in Figure 8.b taken as target.

600 3
1-3
2 4

1 5
CVISC model
1-5
convergence [mm]

400
stage 1 Sulem et al., 1987 3-5

2-3
200

stage 2
0
0 20 40 60 80
days
(a) (b)

Figure 8. (a) Cross section of Saint Martin La Porte tunnel in the numerical model.
(b) Computed (CVISC model) versus monitored convergence at chainage 1406 m.
Also shown is the convergence curve according to Sulem et al., 1987.

Table 3. Constitutive parameters for CVISC model and lining in stage 2.

Ground (CVISC model) Units Value Lining in stage 2 Units Value
Deformation modulus, E MPa 1441 Thickness m 0.2
Poisson’s ratio, ν
4 -4
- 0.25 Moment of inertia, I m 6.67⋅10
Cohesion, c MPa 0.61 Yielding stress, σy MPa 8.5
Friction angle, ϕ ° 28 Tensile strength, σt MPa 0.85
Kelvin viscosity, ηK MPa⋅year 4.26 Deformation modulus, E MPa 5000
Kelvin shear modulus, GK MPa 498.1 Poisson’s ratio, ν - 0.2
Maxwell viscosity, ηM MPa⋅year 27.98

Figure 8.b shows the computed convergence plot versus time obtained for the cross section of interest and for a
total duration of 80 days. Also shown in the same figure is the plot of the interpolation function (15) computed with
the following characteristic parameters: T = 34.2 days, X = 23.7 m and m = 1.36.
It is shown that when the model simulates the tunnel “short term” response, the convergence along array 1-5,
namely 314 mm, is very similar to the in situ value (336 mm) and to the interpretation according to Sulem et al.
1987 (320 mm). The tunnel convergence of stage 1 computed when the CVISC model is activated, reproduces
quite well that observed in situ. It can be seen from Figure 8.b that the convergence versus time plot finally
obtained with the numerical analyses compares well with the observed response up to 50 days duration time. As
expected, the model does not predict the observed deformation thereafter, when the tunnel is shown to exhibit a
gradual decrease in the rate of convergence, reaching a near stable condition.

9
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

5.3 Numerical modelling of the P7.3 cross section (SHELVIP model)

Particular attention has been paid to the numerical analysis of sections that have been excavated with the
P7.3 support system, because they underwent very large deformations, with convergences up to 2 m. For this
reason the novel SHELVIP model, that has been formulated particularly for numerical analysis of tunnel under
severely squeezing conditions, has been adopted. The constitutive parameters obtained by fitting the results of
creep and relaxation tests has been used as starting parameters for the back-analysis carried out.
The numerical analyses of the tunnel response were performed by the FDM and the FLAC code (Itasca, 2006)
with the new SHELVIP model implemented (Debernardi, 2008). Axisymmetric conditions have been adopted in
order to reproduce the three-dimensional influence of the tunnel face, which is known to play a significant role in
squeezing conditions.
The tunnel cross section is assumed to be circular, with an equivalent radius of 5.5 m. Figure 9.a shows the FDM
mesh adopted. The mesh is composed of perfectly square elements, with size increasing gradually from 0.5 m to
4 m when moving from the near vicinity of the tunnel outwards, in order to minimize error propagation and
solution time. The total size of the mesh (184 m - 92 m) is very large in order to minimize the boundary effects
that are very significant in the case of large deformations. The influence of the primary lining and the
reinforcement system has been neglected.
As shown in Figure 9.b, particular attention has been posed on the chronological sequence of excavation (face
advancement) which is considered to influence the time-dependent deformational response. The complex
excavation sequence requires 245 computational steps, in order to closely follow the real chronological sequence.

1380

1370 57 days

1360

m
X

ay /
Chainage [m]

1350

1d
1340 19 days
mesh 4 x 4 m
40

y/m
1330
da

32 days
92

1320
1.

mesh 2 x 2 m
24

face excavation
ay /

1310
7 7 14

mesh 1 x 1 m
mesh 0.5 x 0.5 m numerical analysis
1d

TUNNEL 1300
Y
67 25 92 3-Oct 12-Nov 22-Dec 31-Jan 12-Mar
184 Date
(a) (b)
Figure 9 (a) Sketch of the finite difference grid of Saint Martin La Porte tunnel and
(b) adopted chronological sequence of excavation

The overburden is approximately 300 m and the initial vertical stress is assumed to be 8.4 MPa, with the stress
ratio (K0) equal to 1 (i.e. hydrostatic conditions). The constitutive parameter obtained by numerical back-analysis
of the section at chainage 1311 m are reported in Table 4. Figure 10 shows the comparison of computed and
measured values in terms of radial displacement for the section at chainage 1311 m. The agreement of the
numerical results with the mean curve is excellent, notwithstanding the scattering of the monitoring data due to
the heterogeneity and anisotropy of the rock mass. It is worth to notice that with the same constitutive model the
radial displacements monitored in cross sections which exhibit a similar deformational response can be well
represented. Also the displacements around the tunnel monitored with the multi-position borehole extensometers
can be reproduced satisfactorily.

6 Conclusions
Three constitutive models (CVISC, VIPLA, SHELVIP), each one with different degrees of complexity for
representing the time dependent behaviour of rock, have been discussed. Among them, of particular interest is
the newly developed Stress Hardening ELastic VIscous Plastic model (SHELVIP), which is shown to describe the
main features of behaviour observed during excavation of large size tunnels which exhibit severely squeezing
conditions.
The SHELVIP model has been derived from the Perzyna’s overstress theory, by adding a time independent
plastic component. According to the classical theory of elastoplasticity, the time-independent plastic strains

10
 The 12th International Conference of
International Association for Computer Methods and Advances in Geomechanics (IACMAG)
1-6 October, 2008
Goa, India

develop only when the stress point reaches the plastic yield surface defined by the Drucker-Prager criterion. The
viscoplastic strain rates develop only if the effective stress state exceeds a viscoplastic yield surface which is also
defined by the Drucker-Prager criterion.
An innovative excavation-construction method has been described, which has been implemented in order to cope
with the severely squeezing Carboniferous zone encountered in the Saint Martin access adit, along the Lyon-
Turin Base Tunnel. This method utilises a two-stage excavation sequence which is based on the installation of a
composite lining which incorporates a number of yielding elements formed of highly deformable concrete.
The tunnel response which is continuously monitored during face advance has been analysed. As an illustration
of the application of the VIPLA model a semi-analytical solution has been used. Numerical modelling by the Finite
Difference Method and the FLAC code has been applied in conjunction with the CIVISC and the SHELVIP
constitutive models, with a plane strain and an axisymmetric model respectively. In all cases the computed and
monitored deformations around the tunnel during face advance have been compared, showing the advantages
and disadvantages of each model in describing the deformational response.

1400 2-3
1-3
Table 4. Constitutive parameters of the
SHELVIP model for Saint Martin La Porte 1200 1-5
tunnel (time in day and pressure in kPa)
Radial displacement [mm]

1000 2-4
E 600 MPa
3-5
ν 0.3 800
αp 1.03
600 3-4
kp 0.93 MPa
ωp 0 SHELVIP model
400
γ 0,00008
m 2,2 200
n 0.18
0
l 0.01
αv 0 0 20 40 60 80 100 120 140 160 180
days
ωv 0.735
Figure 10. Computed (SHELVIP model) versus monitored
convergence at chainage 1311 m model.

7 References
th
Barla G. 2005. Design analyses for tunnels in squeezing rock. 11 International Conference of IACMAG. Torino, June 19-24,
2005, 4, 3-22. ISBN/ISSN: 88-555-2812-2. Overview Lecture.

Barla G., Bonini M., Debernardi D. 2007. Modelling of tunnels in squeezing rock. Proceedings of the ECCOMAS Thematic
Conference on Computational Methods in Tunnelling (EURO:TUN 2007), Vienna. Eberhardsteiner et al. (eds.). Vienna
University of Technology, Austria, ISBN: 978-3-9501554-7-1.

Debernardi D. 2008. Viscoplastic behaviour and design of tunnels. PhD Thesis, Politecnico di Torino, Department of Structural
and Geotechnical Engineering, Italy.

Itasca Inc. 2006. Flac2D 5.0, User’s Manual. Minneapolis, USA.

Lemaitre J., Chaboche J.L. 1996. Mécanique des matériaux solides. Dunod, 253-341.

Nguyen-Minh, D. & Pouya, A. 1992. Une méthode d'étude des excavations souterraines en milieu viscoplastique. Revue
Française de Géotechnique 59, 5-14.

Panet M. 1995. Le calcul des tunnels par la méthode convergence-confinement. Presse de l’école des Ponts et chausses,
Paris

Perzyna P. 1966. Fundamental Problems in Viscoplasticity. Advances in Applied Mechanics, Academic Press, 9, 243-377.

Sulem J., Panet M., Guenot A. 1987. Closure Analysis in Deep Tunnels. International Journal of Rock Mechanics Mining
Sciences & Geomechanics Abstracts, 24(3), 145-154.

11