INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS

Int. J. Numer. Anal. Meth. Geomech., 2004; 28:797–819 (DOI: 10.1002/nag.362)

Application of a constitutive model for concrete to the analysis

of a precast segmental tunnel lining

B. Winkler1,z, G. Hofstetter2,n,y and H. Lehar2

1
Hilti AG, New Business and Technology, Numerical Simulation, FL-9494 Schaan, Feldkircherstrasse 100,
Principality of Liechtenstein
2
Institute for Structural Analysis and Strength of Materials, University of Innsbruck, A-6020 Innsbruck,
Technikerstrasse 13, Austria

SUMMARY
A constitutive model for concrete, based on the smeared crack approach and formulated within the
framework of the theory of plasticity, is extended by coupling damage due to tensile stresses with damage
due to compressive stresses for mixed tension–compression loading and by introducing an isotropic scalar
damage model for unloading and reloading. Additionally, a uniaxial model for tension stiffening is
extended to reinforced concrete subjected to biaxial stress states. The constitutive model for plain and
reinforced concrete is validated by means of test data taken from the literature and by laboratory tests on
L-shaped panels.
Finally, the validated material model is used to perform a non-linear FE-analysis of a permanent tunnel
lining made of hexagonal precast concrete segments. During the construction work of the lining hairline
cracks were detected on the inner surface of some precast segments, running parallel to the longitudinal
axis of the tunnel lining. They were supposed to be mainly caused by the installation process of the lining.
In order to gain more information about the origin of these cracks, a non-linear numerical analysis of the
installation process of the lining is performed. The results of the numerical simulation, showing
under which conditions cracks are initiated, are presented and discussed. Copyright # 2004 John Wiley
& Sons, Ltd.

KEY WORDS: concrete; elastic–plastic material model; cracking; tension softening; tension stiffening; finite
element method; non-linear material behaviour

1. INTRODUCTION

The development of computer-aided analysis procedures for the numerical simulation of the
non-linear behaviour of plain and reinforced concrete structures by the finite element method
(FEM) has been the focus of extensive research for more than thirty years. Within this context
the employed constitutive model for plain and reinforced concrete plays an important role. The

n
Correspondence to: G. Hofstetter, Institute for Structural Analysis and Strength of Materials, University of Innsbruck,
A-6020 Innsbruck, Technikerstrasse 13, Austria.
y
E-mail: [email protected]
z
Formerly Research Associate, Institute for Structural Analysis and Strength of Materials, University of Innsbruck,
A-6020 Innsbruck, Technikerstrasse 13, Austria

Received 7 May 2003

Revised 12 December 2003
Copyright # 2004 John Wiley & Sons, Ltd. Accepted 15 December 2003
798 B. WINKLER, G. HOFSTETTER AND H. LEHAR

latter contains the mathematical description of the material behaviour up to failure, an essential
part of which is the model for cracking.
Surveys over the large number of smeared crack models, which have been developed in the
last three decades, can be found e.g. in References [1, 2]. However, the earlier crack models were
characterized by the deficiency associated with mesh dependent results. Especially for plain
concrete and slightly reinforced concrete refinement of the finite element mesh resulted in the
decrease of the numerically predicted ultimate load without achieving convergence. The reason
for this shortcoming is given by the fact that after attaining the tensile strength of concrete, in
the post-peak region of the material behaviour a unique stress–strain relationship does not exist.
In recent years improved crack models, ensuring objective results, have been proposed.
Referring to References [3–5] a relatively simple approach, which is suited for ultimate load
analyses of large concrete structures, is characterized by adjusting the constitutive relation for
the post-peak range according to the size of the finite elements by means of the specific fracture
energy. The concrete model, proposed in Reference [4] within the framework of the theory of
plasticity, will be extended by coupling damage due to tensile stresses with damage due to
compressive stresses for mixed tension–compression loading and by introducing an isotropic
scalar damage model for unloading and reloading.
In addition to cracking, for reinforced concrete the mechanism of the stress transfer between
the reinforcement and the surrounding concrete through bond is of great interest. A
mathematical description of the post-cracking behaviour of reinforced concrete is stated in
Reference [6] in terms of a uniaxial tension stiffening model. The latter will be extended to
reinforced concrete subjected to biaxial stress states. The transition between tension softening of
plain concrete and tension stiffening of reinforced concrete is modelled by taking into account
the layout of the reinforcement, the reinforcement ratio and the average crack spacing.
The constitutive model for plain and reinforced concrete is validated by means of well-known
tests taken from the literature and by laboratory tests on L-shaped panels, which have been
performed at the University of Innsbruck [7]. The motivation for the experimental investigation
is given by the fact that in recent years the L-shaped panel has become a popular benchmark test
for the validation of the mathematical description of the post-cracking behaviour of concrete
(see, e.g. References [8, 9]). However, for the evaluation of the numerical results no experiments
were available so far. Finally, the constitutive model for plain and reinforced concrete is
employed in a non-linear finite element analysis for the assessment of the structural behaviour of
a precast segmental tunnel lining.
Considering these issues the paper focuses on (i) the extension of the material model for
concrete by coupling damage due to tensile stresses with damage due to compressive stresses for
mixed tension–compression loading, (ii) the introduction of an isotropic scalar damage model
for unloading and reloading, (iii) the extension of the uniaxial model for tension stiffening to
reinforced concrete subjected to biaxial stress states, (iv) the validation of the employed material
model for plain and reinforced concrete by means of experimental investigations and (v) the
non-linear FE-analysis of a permanent tunnel lining made of hexagonal precast segments.

2. MATERIAL MODEL

The constitutive model for concrete subjected to biaxial stress states, proposed in Reference [4],
serves as the basis for the description of the material behaviour of plain and reinforced concrete.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 799

It is extended by taking into account the coupling of damage due to tensile stresses with damage
due to compressive stresses for biaxial tension–compression loading and considering unloading
and reloading in the post-peak range of the material behaviour.
The model is summarized briefly in this section. Based on the theory of plasticity, it is
characterized by a composite yield surface and suitable softening laws to describe tensile and
compressive failure. Hence, cracking is represented in a smeared manner. According to
Reference [4] the composite yield function consists of the Rankine principal stress criterion f1 to
limit the tensile stress and of a Drucker–Prager yield function f2 to describe the compression
regime (Figure 1). Assuming small strains, the strain rate vector can be subdivided into an
elastic part e’ e and a plastic part e’ p by
e’ ¼ e’ e þ e’ p ð1Þ
The evolution of the plastic part of the strain rate vector is determined by Koiter’s generalized
flow rule for multi-surface plasticity:
X @gi
e’ p ¼ l’ i ð2Þ
i
@r

Assuming associated plasticity, the plastic potential functions gi are equal to the yield functions
fi : The plastic multipliers l_ i and the yield functions fi have to fulfill the Kuhn-Tucker conditions
fi 40; l’ i 50; l’ i fi ¼ 0 ð3Þ

2 /fcm

f1= 0
0.2

1/fcm
0.0
-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2

-0.2

-0.4

-0.6

-0.8
f2 = 0
-1.0

-1.2

-1.4
Experimental results

Figure 1. Comparison of the composite yield function at ultimate strength with the results of the
experimental investigations of Reference [10].

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
800 B. WINKLER, G. HOFSTETTER AND H. LEHAR

For plane stress states with the stress vector rT ¼ fs11 ; s22 ; s12 g the two yield functions are given
as

1=2
f1 ¼ 12 rT P1 r þa1 pT r
b1 s% 1 ðk1 Þ
ð4Þ

1 T 1=2
f2 ¼ 2 r P2 r þa2 p r
b2 s% 2 ðk2 Þ
T

A comparison with the experimental data of Reference [10] shows that the two yield
functions match the ultimate strength data of concrete in the regions of biaxial tension and
biaxial compression. In Figure 1 the stresses are normalized with respect to the uniaxial
compressive strength fcm : The factors a1 and b1 in (4) result from the Rankine principal stress
criterion f1 to
a1 ¼ 12; b1 ¼ 1 ð5Þ
Using the ratio bc between the biaxial compressive strength and the uniaxial compressive
strength, the factors a2 and b2 for the Drucker–Prager yield function f2 can be determined by
bc
1 bc
a2 ¼ ; b2 ¼ ð6Þ
2bc
1 2bc
1
According to Reference [10] bc  1:16: In (4) the projection matrices P1 ; P2 and the projection
vector p are given by
2 1 3 2 3 2 3
2
12 0 1 2
1 0
6 7 6 7 6 7
6 1 1 7 6 7 6 7
P1 ¼ 6

6 2 2 0 7; p ¼ 6 1 7; P2 ¼ 6
1 2 0 7
7 6 7 6 7 ð7Þ
4 5 4 5 4 5
0 0 2 0 0 0 6
The equivalent stresses s% 1 ðk1 Þ and s% 2 ðk2 Þ stand for the uniaxial tensile strength and the uniaxial
compressive strength in terms of the internal variables k1 and k2 : The latter can be considered as
damage indicators.
Considering the experimental data of Reference [10] for combined tensile and compressive
loading, damage of the material due to compressive loading in one direction results
in a reduction of the maximum attainable tensile stress in lateral direction. This effect is
taken into account by coupling damage caused by compressive loading with damage caused
by tensile loading. The coupling of the internal variables k1 and k2 allows to model the
reduction of the tensile strength under mixed tension–compression loading. Beginning
with the initial yield surface under compressive loading f2 ; indicated by the broken curve in
Figure 1, upon further compressive loading the material will be damaged such that the
attainable tensile stress in lateral direction decreases from the uniaxial tensile strength to zero.
Figure 2 shows the comparison of the computed ultimate strength curve considering the
coupling effect of the damage parameters k1 and k2 with the experimental results of Reference
[10]. Concerning the evolution of the irreversible material behaviour within the framework of
the theory of plasticity, the internal variables k1 and k2 can be determined according to
Reference [4] by assuming that the equivalent stresses are related to energy dissipation via a
work-hardening hypothesis
W’ p ¼ rT e’ p  s% i ðki Þk’ i
i ð8Þ

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 801

2 /fcm

f1= 0
0.2

1/ fcm
0.0
-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2

-0.2

-0.4

-0.6

-0.8
f2 = 0
-1.0

-1.2

-1.4
Experimental results

Figure 2. Computed ultimate strength curve considering the coupling effect of the
damage parameters k1 and k2 :

Using the scalars zij for representing the coupling of damage in the material at different stress
states the work-hardening hypothesis can formally be expressed as
X s% i  @fj 
’ p
Wi ¼ zij rT l’ j ¼ s% i k’ i ð9Þ
j
s% j @r

Considering the Euler’s theorem of homogeneous functions the internal variables ki can be
determined by
X
k’ i ¼ zij bj l’ j ð10Þ
j

with b1 resulting from the Rankine principal stress criterion f1 and b2 from the Drucker–Prager
yield function f2 : In this way, the evolution of the internal variables is determined by the plastic
strain rate and can be considered as a measure of the progressive fracturing of the material. The
equivalent stress s% 1 ðk1 Þ is described by means of an exponential softening law for plain concrete
(Figure 3(a)) by
 
k1
s% 1 ðk1 Þ ¼ fctm exp
ð11Þ
k1u
with fctm as the uniaxial tensile strength. The parameter k1u is given in terms of the specific
fracture energy Gf for tensile failure and the equivalent length h as
Gf
k1u ¼ ð12Þ
h fctm

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
802 B. WINKLER, G. HOFSTETTER AND H. LEHAR

1
2

fctm fcm

Gc / h

Gf / h 1 fcm
3

1 2
1u 2e
(a) (b)
Figure 3. Uniaxial material behaviour of plain concrete: (a) tension; and (b) compression.

In a direct tension test the equivalent length represents the length of the specimen and in a FE-
analysis h is related to a representative length of the element. The equivalent stress s% 2 ðk2 Þ is
described by means of a parabolic hardening law and an exponential softening law (Figure 3(b))
and is determined for k2 5k2e by
 
fcm k2 k22
s% 2 ðk2 Þ ¼ 1þ4
2 2 ð13Þ
3 k2e k2e
and for k2 5k2e by
 
ðk2
k2e Þ2
s% 2 ðk2 Þ ¼ fcm exp
ð14Þ
ðk2u Þ2
with fcm as the cylindrical compressive strength. The parameters k2e and k2u are given in terms of
the specific fracture energy Gc for compressive failure, the equivalent length h and the modulus
of elasticity Ec as
fcm 2 Gc
k2e ¼ 0:0022
; k2u ¼ pffiffiffi ð15Þ
Ec p h fcm
In (15) the total strain at the uniaxial compressive strength is chosen to be equal to 0.0022
according to Reference [6]. The introduction of the specific fracture energies Gf and Gc for
tensile and compressive failure of concrete, respectively, and of the equivalent length h yields
objective results with respect to the employed mesh size. In the context of a finite element
analysis the equivalent length in (12) and (15) depends on the type, the size, the shape and the
integration scheme of the finite element. However, in this study the equivalent length is simply
determined as the square root of the area of the respective finite element.
The post-cracking behaviour of reinforced concrete differs considerably from the respective
behaviour of plain concrete. If cracking occurs tensile stresses are transferred between
reinforcement and concrete by bond, resulting in the tension stiffening effect. A uniaxial tension
stiffening relationship for an embedded reinforcing steel bar (Figure 4(a)) is specified in
Reference [6]. For a smeared crack model the tension stiffening effect can be considered by
modifying either the stress–strain relationship of steel or the one of concrete. In this study the
tension stiffening effect is included in the post-cracking behaviour of concrete. Tension stiffening
can be described as the stress difference ss;TSE between the steel stress ss of the reinforced

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 803

Figure 4. Constitutive relations to account for the tension stiffening effect: (a) stress–strain relationship for
an embedded reinforcing steel bar; (b) equivalent concrete stress–strain relationship; and (c) equivalent
concrete stress–strain relationship adapted to the elasto–plastic material model for concrete.

concrete member and the stress ss;II of a bare steel bar at a given strain. According to References
[11, 12] the stress increase ss;TSE can be replaced by an equivalent concrete stress sc;TSE :
Assuming that in a cracked reinforced specimen between the cracks tensile stresses will be
transmitted from the steel to the surrounding concrete, the equivalent concrete stress sc;TSE
(Figure 4(b)) can be determined as
sc;TSE ¼ reff ss;TSE ¼ reff ðss
ss;II Þ ð16Þ

with reff ¼ As =Ac;eff as the effective reinforcement ratio, defined by the sectional area As of the
steel and the area Ac;eff of the effective zone of the concrete. The effective zone of the concrete is
computed as proposed in Reference [6]. Finally, the stress–strain relationship of Figure 4(b) has
to be transformed to a practicable relationship within the theory of plasticity (Figure 4(c)). For
the latter the equivalent concrete stress is determined for k1 5k1srn by
 
k1
s% 1 ðk1 Þ ¼ fctm 1
ð1
bt Þ ð17Þ
k1srn

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
804 B. WINKLER, G. HOFSTETTER AND H. LEHAR

and for k1sry 4k1 5k1sy by

 
k1sy
k1
s% 1 ðk1 Þ ¼ bt fctm ð18Þ
k1sy
k1sry
For k1srn 4k1 5k1sry the equivalent concrete stress takes the constant value bt fctm : For k1 5k1sy
the equivalent concrete stress is set to zero. The parameters k1srn ; k1sry and k1sy depend on the
modulus of elasticity Es and the yield strength fy of the reinforcement as well as on the modulus
of elasticity Ec of the concrete. Making use of the basic assumptions of Reference [6] they are
obtained as
 
1 1
k1srn ¼ ð1:3
bt Þ fctm þ ð19Þ
Es reff Ec
 
1 1
k1sry ¼ k1sy
bt fctm þ ð20Þ
Es reff Ec
and k1y ¼ fctm =Es : According to Reference [6] bt is equal to 0.40.
The uniaxial tension stiffening model can be extended to reinforced concrete subjected to
biaxial stress states by assuming that the effective extensional stiffness per unit area of the
reinforcement in the direction normal to the crack can be computed by the transformation
X
Es reff ¼ Es;j cos4 gj reff;j ; j ¼ 1; 2; . . . ; n ð21Þ
j

The angle gj denotes the angle between the direction normal to the crack and the direction of the
reinforcing bar j (Figure 5); n is the number of reinforcing bars with different directions. Taking
into account that the tension stiffening effect will be neglected when the yield strength of the
reinforcing steel has been reached, k1sy has to be modified by k1sy ¼ ðfy =Es Þ=cos2 gj :

Figure 5. Geometrical interpretation for the computation of the effective extensional stiffness per unit area
of the reinforcement in the direction normal to the crack.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 805

Finally, the transition between tension stiffening of reinforced concrete and tension softening
of plain concrete is modelled by using the minimum reinforcement ratio according to Reference
[6]. For reinforced concrete with less than the minimum reinforcement ratio the steel is not able
to carry the tensile forces after the tensile strength of concrete has been attained. The failure of
such slightly reinforced concrete members is governed by the material behaviour of plain
concrete. Including the minimum reinforcement ratio in the computation of the effective
extensional stiffness per unit area of the reinforcement as a lower limit for tension stiffening
allows to model the transition from plain to reinforced concrete, taking into account the
direction of the crack with respect to the direction of the reinforcement. This approach is
consistent with the computation of the average crack spacing of reinforced concrete members
according to Reference [6], which secures the transition between tension stiffening and tension
softening in a similar manner.
During ultimate load analyses of plain and reinforced concrete structures material points can
undergo cyclic response due to stress redistribution. Referring to Reference [9] these local stress
relief states are described by an isotropic scalar damage model for unloading and reloading
conditions. Within the framework of damage mechanics the progressive loss of material
integrity due to crack propagation leads to the degradation of the material properties. It can be
described in terms of the damage parameter d; which is related to the ratio between the degraded
modulus of elasticity Ed and the modulus of elasticity Ec for the undamaged material.
Consequently, the degraded modulus of elasticity is determined by
Ed ¼ ð1
dÞEc ; 04d41 ð22Þ

with d ¼ 0 for undamaged material and d ¼ 1 for totally damaged material. Within the
framework of the smeared crack approach the current state of damage can be estimated
according to Reference [13] by the relation between the equivalent stress and the equivalent
strain. Figure 6(a) shows the related equivalent stress-equivalent strain diagram. Using the
equivalent stress s% n and the internal variable kn at the beginning of the unloading process the

Figure 6. Relation between the equivalent stress and the equivalent strain for computation of the damage
parameter d in consideration of: (a) no; and (b) partly irreversible plastic strains.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
806 B. WINKLER, G. HOFSTETTER AND H. LEHAR

damage parameter d can be determined by

Ed s% n
d ¼1
¼1
ð23Þ
Ec s% n þ Ec kn
Computing the damage parameter according to (23) leads to a complete closure of the crack as
indicated in Figure 6(a). Such a complete crack closure can be avoided by introducing a
permanent crack width, which is defined in terms of kn and the scaling factor bd : Consequently,
the damage parameter d can be determined by
Ed* s% n
d ¼1
¼1
ð24Þ
Ec s% n þ Ec ð1
bd Þkn
with Ed* as the unloading and reloading modulus of elasticity as shown in Figure 6(b).
According to Reference [14] the scaling factor bd is set to 0.20.
The described material model was implemented into the finite element program ABAQUS
within the framework of the return mapping algorithms. To this end a user-defined subroutine,
defining the constitutive behaviour at the integration point level, was written.

3. VALIDATION

The computational model for plain and reinforced concrete is validated by means of well-known
tests taken from the literature. As an example for the validation the experimental investigations
of Reference [10] on plate-type specimens (Figure 7) with the dimensions of 200  200  50 mm
will be discussed. These specimens were tested under monotonically increasing biaxial loadings
with constant stress ratio s1 =s2 for each test. The results of the experimental investigations
contain the respective stress–strain curves for concrete under biaxial compression, combined
tension and compression and biaxial tension. In the following the combined tension and
compression load cases will be discussed. According to the experimental results the uniaxial
tensile strength is set to 2:84 N=mm2 and the uniaxial compressive strength to 31:10 N=mm2 :

Figure 7. Geometry of the specimens used for the experimental investigations of Reference [10].

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 807

The Poisson’s ratio amounts to 0.195 and the modulus of elasticity to 33000 N=mm2 : The
fracture energies for tensile and compressive failure are assumed to 0.10 and to 10 N=mm;
respectively.
The results of the numerical analyses are shown in Figure 8 in terms of stress–strain curves.
Figure 8(a) shows the stress–strain curves related to the compressive stress. Figure 8(b) contains
the stress–strain curves for the tensile stress. The comparison of the experimentally and
numerically determined compressive stress–strain curves demonstrates good correspondence
between numerical and experimental results. Beginning with the stress ratio s1 =s2 ¼ 0:0=
1:0
and ending with the ratio s1 =s2 ¼ 0:204=
1:0 the ultimate compressive stress decreases as the
applied tensile stress in lateral direction is increased (Figure 8(a)). The influence of damage

Figure 8. Computed stress–strain curves for concrete subjected to combined tension

and compression in terms of: (a) compressive stresses; and (b) tensile stresses compared
with the experimental results of Reference [10].

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
808 B. WINKLER, G. HOFSTETTER AND H. LEHAR

coupling can be seen in Figure 8(b) in terms of the maximum attainable tensile stress, which
decreases with increasing compressive stress. If the coupling of compressive and tensile damage
was neglected, then in any case of biaxial loading the uniaxial tensile strength would be reached,
which would be in contradiction with the observed material behaviour.
Additionally, the employed material model is validated by means of laboratory tests on L-
shaped panels performed at the University of Innsbruck. They are documented in References
[7, 15]. In Figure 9 the geometric properties of the L-shaped panels and the boundary conditions
are shown. The long and the short edges are given as 500 and 250 mm; the thickness is 100 mm:
The lower horizontal edge of the vertical leg is fixed. A vertical line load, acting opposite to the
direction of gravity, was applied on the lower horizontal surface of the horizontal leg at a
distance of 30 mm from the vertical end face. Actually, the experiments were conducted by
controlling the vertical displacement of the load point and the corresponding force was
measured. The constant displacement rate was chosen as 0:02 mm=min: As shown in Figure 9
the displacements were measured at four different points by inductive pick-ups during the
execution of the tests.
Four different series of experiments with three tests on identical panels for each series were
performed. In the first series, denoted series A, the post-cracking behaviour of plain concrete

500
250 250

pk No.2
20

figure A
280
250
500

pk No.3
bar 20mm
pk No.1 100
200
250

210 20 20
figure A
pk No.4
b=100mm

dimensions in [mm]
110 250 230
22 226

Figure 9. Test setup for the laboratory tests on L-shaped panels, made of plain and reinforced concrete.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 809

was investigated whereas in the other three series reinforced L-shaped panels with different
layout of the reinforcement were tested. In test series B, the reinforcement consisted of four steel
bars, in test series C a welded orthogonal reinforcing grid was used with the bars aligned to the
edges of the panel and in test series D the bars of the reinforcing grid enclosed an angle of 458
with the edges of the panel. The nominal diameter of the applied reinforcing steel bars was
6 mm: The wire meshes, made of heat-treated, ductile ribbed steel, had a 50 mm grid spacing.
The reinforcement was placed in the middle surface of the L-shaped panels with a concrete cover
of 25 mm with respect to the edges. The test setup and the experimental results for the four test
series are given in Reference [7]. Additionally, the data for the L-shaped panel made of plain
concrete can be downloaded from the web (http://nw-ialad.uibk.ac.at/Wp2/Tg2/Se2).
In the following, results of the numerical simulations for two of the four tests series are
presented and compared with the experimental results. The first one refers to test series A on L-
shaped panels made of plain concrete and the second one to test series D on L-shaped panels
reinforced with a welded orthogonal grid enclosing an angle of 458 with the edges of the panel
(Figure 11(b)).
For the numerical simulations three different meshes are generated, the second and the third
mesh representing consistent refinements of the first and the second mesh, respectively. For the
L-shaped panel made of plain concrete the first mesh consists of 600 three-node triangular
elements with one integration point per element and, consequently, the second and third mesh
contain 2400 and 9600 elements.
For the reinforced L-shaped panels coarser finite element meshes are employed, consisting of
75, 300 and 1200 four-node quadrilateral elements with four integration points per element. The
reinforcement is modelled by means of rebar-elements, i.e. one-dimensional elements, taking
into account elastic–plastic material behaviour.
The required material parameters were determined experimentally by means of three samples
for each test. For the L-shaped panel made of plain concrete the uniaxial tensile strength was
determined to 2:70 N=mm2 and the uniaxial compressive strength to 31:00 N=mm2 ; respectively.
The Poisson’s ratio amounted to 0.18 and the modulus of elasticity to 25850 N=mm2 : According
to References [5, 6] the fracture energies for tensile and compressive failure are assumed to be
0:065 and 10:80 N=mm: Concerning the reinforced L-shaped panel the uniaxial tensile strength
was determined to 2:65 N=mm2 ; the uniaxial compressive strength to 29:45 N=mm2 and the
modulus of elasticity to 26075 N=mm2 : The fracture energies for tensile and compressive failure
are assumed to be 0:063 and 10:30 N=mm: For the applied reinforcing grids the modulus of
elasticity was determined to 179073 N=mm2 ; the yield strength to 526:3 N=mm2 ; the ultimate
strength to 584:5 N=mm2 and the ultimate strain to 23:23%:
The comparison of experimental and numerical results is given in terms of load–displacement
curves, plotting the experimentally and numerically determined loads as a function of the
vertical displacement of the point of load application. The respective curves are shown in
Figures 10(a) and 11(a). In addition, the experimentally determined crack patterns are
compared with the computed crack propagation. A further possibility to describe the formation
of cracks is given by the internal variable k1 ; which represents an indicator for tensile damage.
The material behaviour of those elements, which are shown in black colour in Figure 10(c), is
characterized by tension softening, thus representing the crack.
Concerning test series A with the L-shaped panels made of plain concrete, the ultimate load
was reached when the first crack was formed. Any further increase of the vertical displacement
of the point of load application led to a decrease of the load. The ultimate load, computed on

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
810 B. WINKLER, G. HOFSTETTER AND H. LEHAR

12.0

10.0

8.0 Mesh No3

Load [kN]

6.0 Mesh No1, No2

4.0

2.0

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
(a) Vertical displacement of the point of load application [mm]

Experimental investigation (3 tests)

Numerical investigation (3 meshes)

Test No1 Mesh No3

(b) (c)

Crack propagation Damage

Experimental investigation undamaged
partly damaged

totally damaged
Figure 10. Results of tests on L-shaped panels, made of plain concrete: (a) load–displacement curves;
(b) experimentally determined crack pattern; and (c) computed tensile damage.

the basis of the three meshes, is insensitive to the element size. In addition, the computed post-
peak behaviour shows both good correspondence with the experimental results and good
correspondence between the computed results. Hence, the numerical results can be viewed as

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 811

24.0

20.0

16.0
Load [kN]

12.0

8.0

Mesh No1
4.0 Mesh No2
Mesh No3
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
(a) Vertical displacement of the point of load application [mm]

Experimental investigation (3 tests)

Numerical investigation (3 meshes)

Geometry Test No1 vs. Mesh No1

6/50

6/50

(b) (c)

Crack propagation
Experimental investigation
Numerical investigation
Figure 11. Results of tests on reinforced L-shaped panels: (a) load–displacement curves;
(b) specimen geometry and layout of the reinforcement; and (c) computed versus
experimentally determined crack pattern.

objective with respect to the chosen discretization. Concerning the crack propagation the
employed FE-meshes are sufficiently fine to obtain good correspondence of the computed region
of damaged concrete (Figure 10(c)) with the crack, observed in the experiments (Figure 10(b)).

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
812 B. WINKLER, G. HOFSTETTER AND H. LEHAR

In contrast to the L-shaped panels made of plain concrete the steel bars of the reinforced
panels of test series D (Figure 11(b)) carried the tensile forces after the first crack in the concrete
had developed. Subsequently, further cracks formed until the ultimate load was attained by
yielding of the reinforcement. The computed crack propagation agrees well with the
experimental one (Figure 11(c)). The L-shaped panels of test series D failed by rupture of a
reinforcing steel bar, the obtained ultimate load being lower than for the L-shaped panels
reinforced with an orthogonal grid aligned to the edges of the panel (test series C). As soon as
the first crack was formed, the structural behaviour of the panels of test series D was softer than
for the panels of test series C. This effect, which is due to the acute angle between the reinforcing
bars and the cracks in test series D, is well represented by the employed material model for
reinforced concrete. In addition, the material model is validated by means of laboratory tests on
reinforced concrete bars, subjected to uniaxial loading [16] and by selected tests of the
experimental investigations for biaxial stress states, documented in References [17, 18]. The
results of these validation tests are reported in Reference [15].
Both in the experimental and the numerical investigation the formation of a crack is indicated
by a small decrease of the load. As one would expect, the related material points undergo cyclic
response due to stress redistribution. The introduced isotropic scalar damage model for
unloading and reloading helps to overcome numerical problems as well as to prevent incorrect
transfer of compressive stresses across open cracks.

4. NON-LINEAR FE-ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING

The constitutive model for concrete is used to perform an analysis of a permanent tunnel lining
made of hexagonal precast segments. Referring to Reference [19] this tunnel lining was part of
an upgrading project for two hydropower plants in Slovenia, where two new pressure tunnels
were required. Within the scope of this project the so-called double shield TBM technology,
combined with a monoshell segmental lining built of hexagonal precast segments, was used.
The lining is composed of four segments per ring of uniform hexagonal shape with a thickness
of 200 mm; which were fabricated in two Slovenian factories. Depending on the specific
requirements the used segments differed in small details. The invert segment contains an invert
culvert, invert pads and an invert grouting hole. The side wall segments and the roof segment are
nearly identical, differing only in the arrangements of the required grouting holes (Figure 12).
The segments are installed under the protection of the tail shield of the TBM in a circular
form and are kept in position by the help of the contact pressure of the TBM. To achieve
stabilization of the circular lining ring the invert section of the lining is backfilled with mortar.
Afterwards the gap between the remaining part of the lining and the rock mass is backfilled
with pea gravel (Figures 12 and 13). The backfilled gravel provides a certain instantaneous
bedding stiffness between the installed segments and the rock mass. Such a segmental lining
system is designed to act as an arch load bearing system, which pre-dominantly carries normal
compression forces. Hence, only a minimum quantity of reinforcement is needed for the loads,
carried during production, handling, transport and installation. A detailed description of the
construction work can be found in Reference [20].
During the construction work of the permanent tunnel lining hairline cracks were detected on
the inner surface of some precast segments, running parallel to the longitudinal axis of the
tunnel lining. They were supposed to be mainly caused by the installation process of the lining.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 813

pea gravel

135
200
roof segment

3200
°
47.5

side wall segment

6400
axis of excavation
axis of tunnel lining

45

3020
invert segment

180
200
45
diameter of excavation 6980 mm
Figure 12. Cross section of the tunnel lining (Reference: Archives of the Vorarlberger Illwerke AG).

In order to gain more information about the origin of these cracks, a non-linear numerical
analysis of the installation process of the lining is performed.
Taking advantage of symmetry only half of the permanent tunnel lining is modelled using
continuum elements for the precast segments and the backfilled gravel (Figure 14). Concerning
the width of the precast segments of 1600 mm and the gaps between the single segments in the
longitudinal direction of the tunnel lining a plane stress state is assumed. The diameter of the
excavation amounts to 6:98 m; the inner diameter of the tunnel lining to 6:40 m: The axis of the
excavation is located 45 mm above the axis of the tunnel lining.
The generated mesh consists of 1158 finite elements, using 822 elements for the discretization
of the precast segments and 336 finite elements for the discretization of the backfilled gravel. The

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
814 B. WINKLER, G. HOFSTETTER AND H. LEHAR

pea gravel

roof segment

construction process
side wall segment

invert segment

1600 800 800 invert mortar

Figure 13. Longitudinal section of the tunnel lining (Reference: Archives

of the Vorarlberger Illwerke AG).

employed elements are eight-node quadrilateral elements with reduced integration. The inner
contour of the rock mass is modelled by a rigid surface. The mechanical connection between the
precast segments and the backfilled gravel and between the backfilled gravel and the rock mass,
respectively, is modelled by contact surfaces, allowing relative displacements between the
precast segments and the gravel and between the gravel and the rock mass, respectively,
disregarding possible effects due to friction. Assuming a considerable stiffness and stabilization
effect due to the gearing of the segments with respect to the direction parallel to the longitudinal
axis of the tunnel lining, the joints between the roof segments and the side wall segments and
between the side wall segments and the invert segments are not modelled.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 815

CL

contact pea gravel / rock

contact segment / pea gravel

precast segment

pea gravel

contact segment / rock

Figure 14. Discretization of the permanent tunnel lining.

The reinforcement of the segments is represented by means of rebar-elements. These are one-
dimensional elements taking into account elastic–plastic material behaviour. However, since
only a minimum quantity of reinforcement, required for the loads, carried during production,
handling, transport and installation, is provided, failure is governed by the material behaviour
of plain concrete.
For the precast segments concrete qualities B35 and B45 according to Reference [21] are used.
The uniaxial tensile strength is set to 3:00 N=mm2 : Referring to Reference [6] the fracture energy
for tensile failure is assumed to 0:095 N=mm:

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
816 B. WINKLER, G. HOFSTETTER AND H. LEHAR

Concerning the installation process there are two essential loadings, namely the dead weight
of the segments, given as 25 kN=m3 ; and the dead weight of the backfilled gravel, given as
19 kN=m3 :
Considering the installation procedure of the segments, in a first step the segments are kept in
position by the help of the contact pressure of the TBM. This is modelled by fixing the
continuum elements, representing the precast segments. During this fixation step the dead
weight of the segments and the dead weight of the backfilled gravel are activated. In the
following step the fixations of the segments are removed. Thus, the dead weight of the segments
and the dead weight of the backfilled gravel are stressing the already stabilized tunnel lining.
Neglecting geometric imperfections of the lining this load case results only in small bending
moments acting in the invert segment and in the roof segment, respectively. Assuming a
completely backfilled and compacted gravel layer and the regularly bedding of the segments the
predicted tensile stresses are smaller than the uniaxial tensile strength. Hence, no cracks should
appear.
However, the observed hairline cracks on the inner surface of some precast segments, running
parallel to the longitudinal axis of the tunnel lining, indicate additional loads acting during the
installation process, which cause the mentioned cracks.
Thus, a further numerical analysis is performed dealing with the situation before the gap
between the lining and the rock mass will be backfilled with gravel. Concerning the installation
of the segments they are kept in position by the help of the contact pressure of the TBM. Due to
this pressure and the gearing of the segments with respect to the direction parallel to the
longitudinal axis of the tunnel lining a considerable part of the dead weight of the segments will
be transferred to the neighbouring ring of segments, which already has been stabilized by the
backfilled gravel. The remaining part of the dead weight acts on the invert segment, which is not
embedded at that stage. Considering this load case, additional bending moments will arise in the
invert segment.
For the numerical simulation of this load case the already described finite element model for
the tunnel lining is used, just disregarding the backfilled gravel and the mortar bedding of the
invert segment. Neglecting the load transfer to the neighbouring ring of segments due to the
gearing effects with respect to the longitudinal direction, the structure is loaded by the total dead
weight of the segments, applying the latter in incremental steps.
In Figure 15 the material damage after application of the total dead weight is displayed in
terms of the internal variable k1 ; which represents the indicator for tensile damage. The material
behaviour of those elements, which are shown in black colour in Figure 15 is characterized by
tension softening, i.e. k1 > 0:
In the numerical simulation the first cracks are predicted on the inner surface of the invert
segment. These cracks are observed at 52% of the total dead weight of the segments and are
located in the cantilevering part of the invert segment. At 55% of the total dead weight
additional cracks are predicted. These cracks appear on the inner surface of the invert segment
in the region of the invert culvert. Since they are caused by stress concentrations at the corner of
the invert culvert, they will be prevented in practice by rounding off the corner region. Finally,
at 95% of the total dead weight of the segments cracks are predicted on the outer surface of the
side wall segment.
Concerning the results of the FE-analysis the numerically determined crack propagation
demonstrates good correspondence with the cracks observed at the construction site. The
computed cracking loads are in agreement with the crude assumption, that approximately half

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 817

C
L

cantilevering part
of the invert segment
invert culvert

contact segment / rock

Figure 15. Computed tensile damage.

of the dead weight of the segments will be transferred to the already stabilized neighbouring ring
of segments, otherwise cracks will occur. This explains, why hairline cracks are detected only for
some precast segments.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
818 B. WINKLER, G. HOFSTETTER AND H. LEHAR

5. CONCLUSIONS

The elastic–plastic constitutive model for plain concrete, proposed in Reference [4], was
extended by coupling damage due to tensile stresses with damage due to compressive stresses for
mixed tension–compression loading and by introducing an isotropic scalar damage model for
unloading and reloading. The model improvement with respect to mixed tension–compression
loading allows to represent the decreasing tensile strength with increasing compressive stresses
in lateral direction. The introduction of the scalar damage model for unloading and reloading
contributes to prevent numerical problems, associated with unloading and reloading due to
stress redistribution in the context of non-linear structural response and to prevent the
erroneous prediction of compressive stresses across open cracks.
In addition, the uniaxial model for tension stiffening, given in Reference [6], was extended to
reinforced concrete subjected to biaxial stress states. The tension stiffening effect is attributed to
the concrete taking into account the influence of the angle enclosed by a crack and the
reinforcement.
The extended material model was validated by means of well-known tests taken from the
literature and by tests on L-shaped panels. Panels made of plain concrete as well as panels made
of reinforced concrete were investigated, the latter with different layout of the reinforcement [7].
The comparison between experimental and computed results, including the load at the initiation
of cracking, the load–displacement curves in the pre- and post-peak region and the crack
propagation, demonstrates good correspondence between numerical and experimental results.
In the last part of the paper, the validated material model was applied to the assessment of the
structural performance of a permanent tunnel lining made of hexagonal precast concrete
segments. To this end non-linear FE-analyses were performed in order to predict the load at the
initiation of cracking and the potential crack pattern for the expected total loads.
The numerically determined tensile damage shows good correspondence with the hairline
cracks, observed at the construction site. The numerical simulation provides insight about the
origin of the observed cracks and confirms the presumed minor importance of the cracks for the
structural behaviour of the tunnel lining.

REFERENCES
1. Isenberg J. Finite Element Analysis of Reinforced Concrete Structures, II. American Society of Civil Engineers: New
York, 1993.
2. Hofstetter G, Mang HA. Computational Mechanics of Reinforced Concrete Structures. Vieweg & Sohn:
Braunschweig/Wiesbaden, 1995.
3. Feenstra PH, de Borst R. Aspects of robust computational modelling for plain and reinforced concrete. Heron 1993;
38(12).
4. Feenstra PH, de Borst R. A composite plasticity model for concrete. International Journal of Solids and Structures
1996; 33:707–730.
5. Feenstra PH. Computational aspects of biaxial stress in plain and reinforced concrete. PhD-Thesis 1993: Delft
University of Technology, Netherlands.
6. CEB-FIP. Model Code 1990, Bulletin d’information. Comit!e Euro-International du B!eton (CEB): Lausanne.
7. Winkler B, Hofstetter G, Niederwanger G. Experimental verification of a constitutive model for concrete cracking.
Proceedings of the Institution of Mechanical Engineers, Part L, Materials: Design and Applications 2001; 215:75–86.
8. Lackner R, Mang HA. Adaptive FEM for the analysis of concrete structures. In Computational Modelling of
Concrete Structures (EURO-C 1998): 897–919.
9. Menrath H. Numerische Simulation des nichtlinearen Tragverhaltens von Stahlverbundtr.agern. PhD-Thesis 1999:
University of Stuttgart, Germany.
10. Kupfer H, Hilsdorf HK, Ru. sch H. Behavior of concrete under biaxial stresses. ACI Journal 1969; 66:656–666.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819
 NON-LINEAR ANALYSIS OF A PRECAST SEGMENTAL TUNNEL LINING 819

11. Meiswinkel R, Neubauer R, Ji A. Mitwirkung des Betons auf Zug zwischen den Rissen - Berechnung nach EC2.
Beton und Stahlbetonbau 1995; 90(10):261–265.
12. Meiswinkel R, Rahm H. Modelling tension stiffening in RC structures regarding non-linear design analyses.
European Conference on Computational Mechanics (ECCM 1999); CD-ROM, 20pp.
13. Crisfield MA, Wills J. Analysis of R/C panels using different concrete models. Journal of Engineering Mechanics
1989; 115:578–597.
14. Dahlblom O, Ottosen NS. Smeared crack analysis using generalized fictious crack model. Journal of Engineering
Mechanics 1990; 116:55–76.
15. Winkler B. Traglastuntersuchungen von unbewehrten und bewehrten Betonstrukturen auf der Grundlage eines
objektiven Werkstoffgesetzes fu. r Beton. PhD-Thesis 2001: University of Innsbruck, Austria.
16. Hartl G. Die Arbeitslinie eingebetteter St.ahle bei Erst- und Kurzzeitbelastung. PhD-Thesis 1977: University of
Innsbruck, Austria.
17. Vecchio FJ, Collins MP. The response of reinforced concrete to in-plane shear and normal stresses. Technical Report
Publication No. 82-03 1982; University of Toronto, Department of Civil Engineering.
18. Bhide SB, Collins MP. Reinforced concrete elements in shear and tension. Technical Report Publication No. 87-02
1987; University of Toronto, Department of Civil Engineering.
19. Vigl A. Honeycomb segmental tunnel linings}simple, economical, successful. Felsbau 2000; 18(6):24–31.
20. Vigl A, J.ager M. Tunnel Plave-Doblar, Slowenien, 7 m Doppelschild-TBM mit einschaligem, hexagonalen
Volltu. bbing-Auskleidungssystem. In O. sterreichischer Betontag 2000: 97–100.
21. DIN 1045. Beton und Stahlbeton, Bemessung und Ausfu.hrung. Deutsches Institut fu. r Normung, Ausgabe Juli 1988.

Copyright # 2004 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. 2004; 28:797–819