You are on page 1of 33

________

* Corresponding author. Present address: EGIS Industries, 4 rue Dolors Ibarruri, 93188 - Montreuil France.
Tel.: +33 6 19 71 33 14. Fax: +33 1 73 13 19 00. E-mail: gianluca.ruocci@egis.fr (G. Ruocci)
An improved damage modelling to deal with the variability of
fracture mechanisms in FRP reinforced concrete structures
Gianluca Ruocci
1,*
, Pierre Argoul
2
, Karim Benzarti
1
, Francesco Freddi
3

1
Universit Paris Est, IFSTTAR, MAST, 14-20 Boulevard Newton, F-77447 Marne la Valle Cedex 2, France
2
Universit Paris Est, Laboratoire Navier (UMR 8205), CNRS, ENPC, IFSTTAR, Ecole des Ponts ParisTech, 6&8
Avenue Blaise Pascal, F-77455 Marne la Valle Cedex 2, France
3
Universit di Parma, Dept. of Civil-Environmental Engineering and Architecture, Viale Usberti 181/A, I-43100
Parma, Italy
ABSTRACT A new way of modelling is developed and proposed to predict different damage scenarios of
concrete elements strengthened by externally bonded Fibre Reinforced Polymer (FRP) plates. The bonded assembly
is modelled as a three-domain system with concrete, glue and FRP reinforcement assumed as damageable materials
being connected together by two interfaces. Interaction between domain and interface damage is introduced.
Detachment between FRP reinforcement and concrete in a single lap shear test configuration is analysed by
implementing the equations governing the damage model obtained in a finite element code. The damage evolution is
characterised through various indexes, which makes it possible to discriminate the failure mechanism when varying
properties of the glue or interfacial characteristics. Comparison between simulations and experimental tests shows
the accuracy of the damage model prediction and its capability to detect different failure modes; in particular, this
new modelling approach allows distinguishing between an adhesive failure at a glue-substrate interface and a
cohesive failure of the glue layer.
KEYWORDS: epoxy (A), concrete (B), fracture mechanics (C), durability (D), damage modelling.
1. Introduction
The adhesive bonding of Fibre Reinforced Polymer (FRP) composites is one of the most effective
solutions for the rehabilitation of civil structures. This retrofit technique consists in the strengthening of
concrete structures by means of externally bonded composite plates or carbon fibre sheets. Nowadays,
both the repair of damaged structures and the upgrading of structurally deficient civil infrastructures are
carried out through the gluing of stiff external reinforcements. Despite the popularity of this promising
technology, some issues concerning the variability of the damaging behaviour and the durability of the
adhesive joint are still matters of concern.
2

The long-term performances of the glued assembly may be critically affected by several
environmental factors. For instance, large cycles of variation or consistently high values of humidity and
temperature are known to possibly affect the mechanical properties of the glue, and/or weaken the
physico-chemical linkages between the polymer adhesive and the substrates [1-4]. In this particular
context, the effectiveness of the stress transfer and consequently, of the strengthening becomes
undermined. Depending on the degradation mechanisms involved, failure modes are characterised by
cohesive cracking, adhesive debonding or intermediary situations. The bearing capacity and durability of
FRP-strengthened concrete elements can thus be significantly reduced according to the activated failure
mechanism.
Therefore, a reliable retrofit design of concrete structures reinforced by bonded composites plates
requires a model capable of satisfactorily predicting the multiple failure phenomena due to the different
states that the assembly may experience. The purpose of this research is to investigate the failure
mechanisms of FRP-strengthened concrete elements through the damage modelling which considers both
the interfacial debonding and cohesive fracture. The proposed methodology relies on the model proposed
by Frmond et al. [5], which considers damage as the result of microscopic motions whose power is
introduced in the principle of virtual power. On the basis of this theoretical framework, Freddi et al. [6]
developed a model to describe the damage behaviour of bonded assemblies. Both the damage in the
volume of the adherents and the failure of the adhesive interface, as well as their interactions, are taken
into account in order to deal with different failure phenomena.
In addition, Benzarti et al. [7] adopted a simplified version of the former model, whose validity is
restricted to the case of rate independent problems, e.g. quasi-static fracture tests, and which involves a
reduced set of physical parameters and damage coefficients. The simplified formulation proved to
satisfactorily describe the detachment process (fracture propagation in 2D) for a single-lap shear test
carried out on a FRP-concrete assembly. However, only a single condition was analysed, that referred to
an unaltered state of the constitutive materials (in particular the bulk epoxy adhesive) characterised by
means of some standard mechanical tests.
In the present study, different sets of mechanical characteristics and damage coefficients are
considered in order to evaluate the variation in the failure mechanism due to an alteration of the
3

constitutive material and/or interfacial properties. Unlike in former studies, the actual thickness of the
glue joint is considered in the damage modelling formulation. Indeed, the previously adopted interfacial
solution was unable to distinguish between the adhesive failure at the interface and the cohesive fracture
within the glue layer. Moreover, a further advantage of this new approach consists in the possibility to
differentiate the characteristics of the interfacial areas between the glue and specific substrates. As clearly
outlined in [810], the glue penetrates porous media with a small thickness acting as a reinforcement of
concrete and locally changing its characteristics.
The paper is organized as follows. In Section 2, the damage modelling is addressed and the basic
equations governing the problem are briefly reviewed. The simplified approach, assuming that the
adhesive bond is a simple contact surface, is first recalled to describe the theoretical framework. The
enhanced modelling of the glue layer is then presented, the computational differences with the former
model are pointed out and the advantages are discussed in details. Next, in Section 3, different indexes are
proposed to characterise the progression of damage in the assembly subjected to mechanical loading. A
numerical example of FRP-reinforced concrete specimen tested in a single shear configuration, as well as
the corresponding experimental tests, are presented in section 4 to investigate how an alteration of the
glue properties affects the fracture process. In addition, the numerical results obtained from the two
damage models are compared and discussed in the light of the experimental evidences. Finally, general
conclusions drawn from this study as regards to damage modelling in bonded assemblies are presented in
Section 5.
2. Damage Modelling
In the structural response of the FRP reinforcement of concrete elements a key role is played by the
adherence between concrete and composite material determined by the adhesive layer. Particularly, as
evidenced in Figure 1 the separation of the FRP from the support can take place in the concrete (failure
1), among concrete and the adhesive layer (failure 2) inside the glue (failure 3) or at the interface between
the glue and the reinforcement (failure 4).
4


Figure 1. Possible failure modes in concrete substrate reinforced by FRP element.
Usually, the debonding process is modeled concentrating and limiting the nonlinear effect in a
fictitious zero thickness cohesive interface between the concrete and the reinforcement (see [11] for an
exhaustive review). This approach revealed to correctly reproduce the macroscopic evidence and the
structural response of cohesive failure mode located in the substrate but, other more complex fracture
processes cannot be exactly reproduced by this simplified approach. In fact, in order to describe the
variety of fracture processes both the materials (substrate, glue, reinforcement) and the interfaces
(substrate-glue, glue-reinforcement) have to be modeled in such a way that rupture may initiate and
develop in different locations.
Starting from these physical evidences we adopted a damage model that is capable to simulate all
these failure modes. Indeed, all involved materials are damageable and cohesion can be lost at the
interfaces. For clarity, a model with two materials and one interface will be presented first in order to
illustrate the main basis and after the formulation will be extended to the case of three materials and two
interfaces.
2.1 Two-domains damage model
The theoretical fundamentals of the damage model adopted here to simulate the detachment between
FRP and concrete are briefly reviewed in this section. For a more detailed summary, the reader can refer
to [57, 12]. The equations of motion result from the principle of virtual power. In this formulation, the
virtual power is assumed to depend on the strain rate, damage velocity and its gradient. For the sake of
simplicity, a quasi-static isothermal problem is considered. As such, acceleration forces can be set to zero
and thermal effects can be neglected.
5

Let us consider a system made of two domains
i
, i = 1, 2, subjected to mixed boundary conditions
and connected by a cohesive interface or contact surface denoted
2 1
=
s
. For the case at hand,
the two domains are the concrete specimen and the FRP reinforcement, and the zero thickness cohesive
interface is defined as the adhesive layer. For each material, the state quantities are the macroscopic
damage quantity
i
, its gradient
i
grad and the deformation tensor
i
. All of these variables depend on
time t and position x. The damage quantity can be interpreted as the volume fraction of active links or
undamaged material; thus its values vary between 0 and 1, where 1 represents the undamaged state and 0
corresponds to completely cracked zones. The gradient of
i
accounts for local interactions of damage at
a specific point with damage in the surrounding area. Similarly, the deformation
i
accounts for the local
interaction of displacement at a point with displacement in its neighbourhood.
Concerning the contact surface
s
, the state quantities are i) variables which describe the evolution of
the surface (or glue layer in the present case), such as the surface damage quantity
s
and its gradient
s
grad , the latter accounting for the local damage interaction on the surface, and ii) quantities which
describe the macroscopic and microscopic interactions between the domains and the contact surface. The
macroscopic mechanical interactions are described by the gap between the domains ( )
2 1
u u , where
i
u
represents the displacement field of each domain
i
.
Combining the principle of virtual power with the constitutive laws leads to three sets of equations of
motion as described in Freddi et al. [6] and in Benzarti et al. [7]. The first one (1) is the classical equation
of motion in classical mechanics, while the others are non-standard differential equations that describe the
damage evolution in the domains (2) and at the interface (3):
0
i i
= + f div , in
i
, (1)
( )
i i i i i
k F = , in
i
, (2)
( ) ( ) ( )
2 1 ,1 1 ,2 2 s s s s s s s s s
k G k k = u - u , on
s
(3)
6

where div and are the divergence and Laplace operator respectively,
i i i i
= C is the stress
tensor associated to the elasticity fourth order tensor
i
C , while
i
f represents the volume exterior force.
The function ( )
i i
F is the failure criterion that governs the rupture mechanisms in the domain and whose
expression can be specified for each material. Analogously, the function ( )
1 2
u - u
s
G is the failure
criterion for the interface.
The coefficient
i
k in eq. (2) is the damage extension parameter, which controls the size of the
transition zone between sound and damaged material. Increasing the value of
i
k leads to transitions from
concentrated damage (fracture) to widespread damage in the whole domain. The parameter
s
k of eq. (3)
assumes similar meaning for the bonded interface
s
. Moreover,
s,1
k and
s,2
k are the surface-volume
interaction parameters, which quantify the influence of damage in the domains on damage at the interface.
Setting this parameter to 0 creates a damage barrier at the interface. The physical meaning of the model
parameters is presented in [6] and a sensitivity analysis was presented in [13], where the influence of the
damage coefficients on the bulk and interfacial damaging was highlighted too.
The boundary conditions for the domains read:
i i i
g N = , in ( )
2 1
\
i
, (4)
0
i
i
=

N
i
k

, in ( )
2 1
\
i
, (5)
where
i
N is the outward normal to the domain
i
and
i
g represents the surface exterior force.
The boundary conditions on the contact surface are:
( ) ( )
i 1 2 1

s s
k = N x u - u , for
2 1
x , (6)
( ) ( )
i 2 2 1

s s
k = N y u - u , for
2 1
y , (7)
( )
1 1 ,
1
1
1

s s
k k
N
, (8)
( )
2 2 ,
2
2
2

s s
k k
N
. (9)
7

where
s

k is the surface rigidity of the bonded interface


s
. The adhesive source of damage is represented
by ( )
1 2
u - u
s
G .
The initial conditions:
( ) ( )
0
, 0 1
i i
= = x x , in
i
, (10)
( ) ( )
0
, 0 1
s s
= = x x , in
2 1
, (11)
simply assume the integrity of the volume and interface areas at the initial time t
0
= 0. Finally, in order to
prevent material healing, the value of
i
and
s
are updated to satisfy the irreversibility conditions for
two subsequent loading steps t and t + dt ( ) ( ) ,
i i
t t dt + x x, , ( ) ( ) ,
s s
t t dt + x x, .
This approach [6, 7] has proved to be sufficient in capturing the main physical phenomena. More
sophisticated constitutive laws, such as those considered in [14, 15], could be adopted; however the
enhancement of the results that could be derived is secondary for the purposes of this study.
The two domains here considered are the concrete element and the FRP reinforcement. In the
following equations, they are distinguished by the subscripts c and p, which refer to concrete and
composite, respectively (i=c,p). The subscript s refers to the adhesive layer, whose thickness is
neglected in order to simulate the joint as a contact surface. In the previous equations (2)-(3), the failure
criterions have to be specified. They are expressed in term of elastic energy and also represent the sources
of damage. In particular, in the case of FRP reinforcement, whose mechanical behaviour law is mainly
symmetric, the source of damage ( )
i i
F reported in (2) takes the form:
( )
p
p
p p p p p
w
F

= C : :
2
1
, (12)
while in the case of concrete, which exhibits asymmetric behaviour in tension and compression, damage
is mainly related to extension. Thus, the failure criterion takes only the positive part of the deformation
tensor
+
c
into account:
( )
c
c
c c c
w
F

=
+ +
c c
C : :
2
1
. (13)
8

The coefficient
i
w in equations (12) and (13) is the initial damage threshold expressed in terms of
volumetric energy. It derives from the rupture criterion, which is adapted for each material.
Analogously, the damage threshold
s
w , i.e., Duprs energy that accounts for the glue cohesion, permit to
introduce the source of damage assumed for the interface:
( ) ( )
s
s
c p
s
c p s
w k
G

=
2
2

u - u u - u . (14)
2.3 Three-domains damage model
An improvement in the way of modelling glued assemblies affected by damage is adopted here. Its
main characteristic consists in the introduction of a third damageable domain representing the glue layer,
and consequently the presence of two contact surfaces (interfaces) that connect this new domain to the
adjacent adherents. The glue layer is modelled as a volume and its actual thickness is taken into account
in order to capture the phenomena related to the cohesive failure of the adhesive joint. In this context, the
damage behaviour can be fully characterised by considering both the adhesive failure at the interface
between different components and the cohesive fracture within the glue layer.
This variant leads to a new geometry of the system that contains three domains
i
, i = 1, 2, 3,
corresponding to the concrete element, FRP reinforcement and glue joint, respectively. The interface
3 1 1
=
s
connects the concrete and glue domains, while the interface
2 3 2
=
s
creates
a mechanical link between the adhesive layer to the composite plate. An illustration of the difference
between the two versions of the model is given in Figure 2.

1

s

2

1

1 s

3

2 s

2

Figure 2: The geometries of the two versions of the bonded assembly modelling: two domains and one interface
(left) and three domains and two interfaces (right).
9

The damage model is based on the same set of state quantities presented before for both of the
domains and contact surfaces, but with the numbering adapted for the new geometry of the bonded
assembly. We assume 3 , 2 , 1 = i equivalent to g p c i , , = , where g represents the glue domain. The
subscript 2 , 1 s s s = stands for the interface between the concrete and glue, and the interface between the
glue and composite, respectively. The three sets of equations of motion that describe the damage
evolution are obtained again by combining the principal of virtual power with the constitutive laws and
considering the same simplifying assumptions of the first version of the model. The equations referring to
the mechanical equilibrium and damage evolution for the domains (12, 13) are kept unchanged. The
equation of the interface damage evolution (14) is doubled so as to describe the behaviour of both
interfaces:
( ) ( ) ( )
g s g s c s c s c g s
s
s s s s
k k
k
w k =
1 , 1 1 , 1
2
1
1
1 1 1
2

u - u , (15)
( ) ( ) ( )
p s p s g s g s p g s
s
s s s s
k k
k
w k =
2 , 1 2 , 2
2
2
2
2 2 2
2

u - u . (16)
The boundary conditions of this problem are similar to the ones reported into equations (4-9).
The initial damage threshold
g
w for the glue domain is derived from the rupture criterion ( )
g g
F that
assumes symmetric mechanical behaviour of the adhesive layer and takes the form:
( )
g
g
g g g g g
w
F

= C : :
2
1
, (17)
where the rank four elasticity tensor
g
C accounts for the mechanical behaviour of the glue layer. In the
former model, the rigidity of the bond between the two solids, i.e., the rigidity of the interface, was
represented by the parameter
s
k

. In the new version of the damage model, it is necessary to introduce the


surface rigidities of the two interfaces,
1

s
k and
2

s
k . Analogously, all of the damage parameters related to
the cohesive interface in the first model are doubled so as to take the two adhesive interfaces into account.
This leads to two damage extension parameters
1 s
k and
2 s
k and two damage thresholds
1 s
w and
2 s
w ,
which are derived from the sources of interface damage (18) and (19), respectively:
10

( ) ( )
1
1
2
1
1
2

s
s
c g
s
c g s
w k
G

= u - u u - u , (18)
( ) ( )
2
2
2
2
2
2

s
s
p g
s
p g s
w k
G

= u - u u - u . (19)
Finally, the surface-volume interaction parameters
c s1,
k and
g s1,
k are introduced to respectively
quantify the influence of the damage in concrete on the damage at the first contact surface and that of the
glue on that of the same interface. Similarly,
g s2,
k accounts for the damage relationship between the glue
and second contact surface, while
p s2,
k links damage in composite with that on the second interface.
The most important result that we obtain from taking the actual thickness of the joint into account is
the uncoupling of the adhesive and cohesive failure mechanisms. The former is assigned to the two
contact surfaces, while the latter is related to the glue layer. This variant of the damage model also gives
the opportunity to choose the values of the interface characteristics in order to simulate the creation of a
thin transition zone (also called interphase) with characteristics differing from that of the glue and the
substrate, and resulting either from the penetration of the glue into the porosity of the substrate or from
physico-chemical interactions between the glue components and the surface of the adherent [16].
However, it is a hard task to select the characteristics of these transition zones (thickness and mechanical
properties) which are very dependent on the kinetics of diffusion phenomena and requires sophisticated
characterization tools to be properly assessed. For the sake of simplicity, in the following, the interfaces
are imagined to be part of the glue layer and their mechanical properties are selected accordingly. These
additional advantages provided by the new damage model will be evaluated in a forthcoming work.
3. Damage characterisation
The characterisation of damage in bonded assemblies deals with the fundamental issues listed below:
- damage occurrence (failure criterion),
- damage intensity,
- damage location and extension,
- damage evolution in time and space.
11

The first item represents the basis of the damage assessment process. A threshold checking approach
based on one or a combination of parameters is commonly used towards this aim. Maximum stress or
strain are the quantities generally adopted, with the latter more effective when the adhesive joint exhibits
significantly nonlinear behaviour. In both cases, a reference value is selected and failure is assumed to
have occurred when the maximum strain or stress reaches its ultimate value. The drawbacks of this
approach are clearly evident: the structural integrity is related to the maximum value of a single feature,
thereby discarding its distribution within the specimen. Moreover, the selection of the threshold value is a
difficult task to accomplish because of its dependency on the mechanical properties of materials; only the
condition of complete failure is considered, whereas the transition through intermediate damage states is
neglected.
The state quantity introduced in the proposed damage model is well-suited for quantifying the
extent of damage and describing its evolution since it is a function of time and position. Its physical
meaning refers to the volume fraction of active links, or undamaged material; thus this variable is directly
related to the local structural integrity of the specimen. Compared to the classical threshold approaches,
the damage parameter provides a fuzzy description of the damage state, which also involves partially-
failed situations. The null value represents open fractures, or complete detachment between two bonded
components, while values in the range 1 0 < < define partially-cracked zones, or incipient
delamination phenomena. Neither thresholds nor equivalent features need to be calculated to define the
damage state. In the next paragraphs new damage and failure indexes are presented to deal with damage
characterisation of FRP-reinforced concrete elements.
3.1 Average damage index
A first parameter that describes the damaging mechanisms of glued assemblies subjected to rupture
tests is defined. It is hereinafter referred to as average damage index since it averages the damage
quantity over the number of nodes
d
N affected by damage, i.e., those nodes whose value is lower
than unity:

=
d
N
k
k
d
N

1
. (20)
12

The average damage index provides a quantification of the damage extent, which is a function of the
damage intensity (the value of
k
) and the extension in the domain, or along the interface (expressed in
terms of the number of affected nodes
d
N ). The occurrence of damage in a quasi-static rupture test leads
to an index value lower than 1. As damage increases, the index decreases because of the addition of new
damaged nodes and the lowering of the values of those already affected. Unlike the simple damage
quantity , complete failure does not necessarily correspond to the 0 value of the average damage index.
Indeed, some portions of the specimen could be only partially damaged when rupture occurs, and the
contribution of their nodes in (20) would prevent the index from reaching 0. The unchanged value of the
damage index throughout two or a few more consecutive time steps can be assumed as reliable evidence
of complete failure occurrence. Moreover, the availability of the two damage quantities
i
and
s

allows the damaging process to be followed separately in the domains and at the interface.
3.2 Damage distribution vectors
Although the average damage index provides an effective estimate of the overall damage extent at
each step of the rupture process, no indication is given about the position of the failure within the
specimen and its evolution in space. A new damage feature is introduced to deal with damage
localization. Depending on the dimension (2D or 3D) of the problem being analysed, a different amount
of distinct quantities is defined: one for each coordinated direction. Each quantity is a vector (or a matrix
in the 3D case) representing the distribution of damage along an axis (or over a plane in the 3D case).
Values at nodes are averaged along the direction of each coordinate axis, which gives
X
representing
the damage distribution over the X axis and
Y
that over the Y axis. The valleys in the damage
distributions identify the areas where damage is more extended and pronounced. An explanatory example
is shown in Figure 3, where the damage pattern on a two-dimensional domain is depicted in terms of a set
of level curves whose colours assume the coldest tonalities for the lowest values of the damage parameter
(failure). The shape of the two distributions (Figure 3b-c) provides a clear idea of the damage location
within the domain and along the axes, as well as the amount of damage, which was depicted through the
amplitude. The evolution of damage is reflected in the spreading of the distribution along the axes and the
13

increase of the amplitude of the distribution. Some examples of the evolution of the damage distributions
will be provided in the next section with reference to the present case study.

Figure 3: (a) level curves of a damage pattern defined on a two-dimensional domain, (b) damage distribution
X

along the horizontal axis and (c) damage distribution
Y
along the vertical axis.
3.3 Failure indexes
A final effort in the characterisation of the damaging process is the fusion of the features concerning
the damage extent and its propagation in time. The aim is to condense into a single scalar value those
characteristics that are most significantly affected by the parameters variation in the modelling of damage
behaviour. The so-called failure index is derived accounting for three different sources: the amount of
energy released throughout the damaging process, the time lapse needed to reach complete rupture and
the overall damage amount at failure. With reference to Figure 4, which shows the evolution of the
average damage index throughout the rupture test of a FRP-strengthened concrete element, the first
quantity is identified by the coloured area

A , the second refers to the interval of time t between the


damage triggering
t
t and the instant of the rupture occurrence
f
t and the third is given by the ultimate
reduction of the average damage index, noted . The variables (excluding the last one) are normalized
in order to assume values in the range [0, 1] and then combined according to:
f f
t
A
t
t
t
A
I


= , with [ ] 1 , 0

I . (21)
a)
b)
c)
14

Therefore, the failure index

I can be defined as the value of the average damage index that


equals the surface of the two shaded areas split by the horizontal dash-dotted line in Figure 4. Values of
the failure index

I close to 1 represent a detachment process protracted in time and extended in space.


The so-defined failure index is able to discriminate different failure mechanisms. Depending on the
mechanical properties of the components, their aging and the efficacy of the bond, a FRP-strengthened
concrete element shows types of rupture differently localised in the components and damaging processes
spread over a variable lapse of time. For instance, rupture can be abrupt, which signifies a fragile
behaviour, or prolonged in time, as typical of ductile materials. Both rupture time protraction and space
extension are phenomena that indicate a pronounced capacity of the assembly to redistribute stress. The
failure index implicitly takes into account redistribution and dissipative capacity (the amount of released
energy

A ) and quantifies the tightness of the assembly. The highest the index value the highest the
amount of energy which is needed to achieve complete failure.
The formulation presented in (21) applies to all the elements of the assembly depending on the average
damage index that in the present case is selected among
c
,
g
,
p
,
1 s
and
2 s
. Thus, the failure
index can be computed separately for each adherent, adhesive and interface, leading to
c
I

or
p
I

,
g
I

,
1 s
I

or
2 s
I

, respectively. The three domains can be grouped in a single index


V
I

, while the two


interfaces considered as a whole are represented by
S
I

. In addition, the combination of failure indexes


related to different components of the assembly provides the means to identify the type of rupture as a
function of a scalar value. In particular, for the improved damage model, the distinction of the glue layer
from the interfaces and substrates allows to compute a global bulk index

that is expressed as in (22):


s V
c
I I
I
I

+
=

, with [ [ 1 , 0

I , (22)
where
c
I

is the failure index of the concrete substrate. A value close to 1 for the global bulk index


refers to a type of rupture that mainly involves the concrete substrate. Conversely, very low values of the
global bulk index identify the case of adhesive failure. More specifically, the occurrence of the interfacial
15

failure is pointed out by using a second combination of the interface and volume failure indexes. Thus,
the global interface index
s
I

is defined as:
s V
s
s
I I
I
I

+
=

, with [ ] 1 , 0


s
I

. (23)
High values for the global interface index
s
I

define the adhesive failure at the interface. For more


detailed studies that adopt the three-domains damage model, the global interface index can be split in two
by distinguishing the failure index for the interface
1 s
from that for
2 s
. In the following sections, the
failure indexes and their combinations will be used as criteria for assessing the influence of the
mechanical and bonding characteristics of the epoxy on the fracture behaviour of a glued assembly.
0 10 20 30 40 50 60 70
0.4
0.5
0.6
0.7
0.8
0.9
1
Time [s]
A
v
e
r
a
g
e

d
a
m
a
g
e

i
n
d
e
x

A

t
t
t
f
t
0

I

Figure 4: The three components (

A , t and ) of the failure index

I characterizing the damage evolution.


4. Experimental and numerical tests
4.1 FRP-reinforced concrete specimen
The accuracy of the damage model presented in Section 2 is assessed through its comparison with the
results of an experimental campaign carried out jointly by IFSTTAR and the Department Laboratoire
dAutun in France. A large number of FRP reinforced concrete elements were tested under varying
characteristics of the epoxy adhesive. Each specimen consisted of a concrete block of dimensions 210
210 410 mm
3
cast using CEM II 32.5 cement and silico-calcareous aggregates (610 and 1020 mm)
and a water-to-cement ratio of 0.55. A maturation period of 28 days was enforced in order to obtain a
16

compressive strength of 32 4 MPa and a tensile strength of 3.6 0.1 MPa. A 100 mm wide and 1.5 mm
thick strip of CFRP plate was bonded using a two-component epoxy adhesive on the upper side of the
concrete substrate after a sand-blasting treatment. The adhesive layer measures 200 mm in length and 0.75
mm in thickness and it was positioned 50 mm away from the front side of the concrete block in order to
prevent edge effects (Figure 5a). This procedure is commonly adopted in experimental campaign of
debonding tests as in [19] and [20]. Former studies [7] have shown the influence of the geometric
configuration of the composite gluing on the observed failure mechanism for a single lap joint shear test.
The configuration chosen here, from those tested in [7], leads to cohesive fracture in the adherent near the
bonded surface. A nearly uniform concrete cover is detached from the block. Conversely, when the glued
surface is prolonged up to the edge of the specimen, a concrete wedge is removed from the blocks
corner. The gap between the edge of the concrete block and the adhesive layer allows a more regular
detachment growth along the specimens length, which is desirable when the aim is to characterize
different failure mechanisms. For more details about the physical properties and mechanical compositions
of the specimen constituents, the reader can refer to [1].
410
210
200 70
50
15
50
90
135
180
410
210
grips
F
strain gauges
170
X
Y


(a) (b)

(c)
Figure 5: (a) Geometry, (b) test setting and (c) mesh of the FRP-reinforced concrete specimen: units in mm.
17

The mechanical characterisation of the bonded assembly was performed through a single lap joint
shear test carried out using a hydraulic jack with a 100 kN maximum force capacity. The boundary
conditions of the test are shown in Figure 5a. The specimen is placed in a steel frame specifically
designed to carry out this kind of test [7]. Thick rigid walls around the concrete prism impede any
translation (Figure 5b). A specific support equipped with adjustable screws acting on the specimens
position was used to ensure a good alignment of the jack with the reinforcement and to prevent any
rotation of the specimen during the test. The clamping of the free end of the reinforcement is made with
conical grips 70 mm away from the glue joint (this distance is more or less arbitrary, but the main thing is
that it remains constant during al the test campaign). The test was carried out at a constant displacement
rate of 6 m/s until the complete detachment of the FRP overlay.
The shear tests were repeated on several sets of specimens with varying properties of the epoxy
adhesive. The experimental campaign showed a variety of failure mechanisms that can be grouped into 3
main types. The upper and lower faces of fractured surfaces obtained from the shear test for these 3
groups are depicted in Figure 6. The first fractured surface refers to debonding at the interface between
the FRP and adhesive (Figure 6a), the second shows pure concrete cracking (Figure 6b) and the third
involves cohesive failure within the epoxy joint (Figure 6c). In a first step, we will focus on the last two
failure modes, i.e. the cohesive failures within concrete and glue, which were obtained for states of the
epoxy adhesive related to different ageing times. The mechanical properties of the epoxy adhesive were
obtained through tensile-loading tests performed on dumbbell glue samples. Exposure to hydrothermal
ageing at 40C and 95% R.H. for 8 months revealed a decrease by a factor of 4 compared to the initial
state for both the tensile strength (from 251 MPa to 6.60.5 MPa) and the Youngs modulus (from
4.90.2 MPa to 1.30.2 MPa) of the glue. Moreover, the extensive plasticization of the polymer network
by water molecules from the ageing environment enhanced the ductility of the second state of glue
(reference to Figure 9a in [1]). Therefore, in the following, the first state of epoxy will be referred as the
elastic and fragile initial state, whereas the second as the elasto-plastic and ductile state observed after 8
months hydrothermal ageing.
18




(a) (b) (c)
Figure 6: Failure surfaces for shear tests carried out on specimens with different adhesive characteristics: (a) FRP-
glue interface debonding for poor adhesive properties of the glue, (b) concrete cracking for elastic and fragile glue
and (c) complete cohesive failure of the adhesive layer for elasto-plastic and ductile glue; dimensions are in mm.
4.2 Numerical simulations
The numerical simulations of the experimental shear test are carried out by implementing the damage
model in a FE code based on the Open Source package Deal.II [17]. Both of the 2D models presented in
Section 2 are considered. In the following, the model where the thickness of the adhesive layer is
neglected will be referred as Model 1 and the other as Model 2. In both models, the concrete and FRP
plate are represented by quadrilateral elements in plane strain condition with polynomial approximation
of degree 1. The introduction of the actual thickness of the adhesive layer in the second model entails the
adoption of a further class of quadrilateral elements for the third domain and two interfaces at the upper
and lower sides of the bond. Interface elements with one infinitesimal dimension are used to split the
domains and to model the interface damage coupling effect. The mesh of the first model, consisting of
4996 nodes and 5222 elements, is illustrated in Figure 5c. The mesh of the second model, consisting of
5398 nodes and 6022 elements, differs from the former only in the elements used to model the adhesive
layer and second interface. Convergence was studied at each time step as difference between the results of
two consecutives iterations and it was considered achieved for values lower than 1e-5. Convergence was
achieved for both models throughout the whole shear test, with slightly higher computation time for
Model 2 due to its larger mesh. The boundary conditions introduced in the simulations are represented in
Figure 5 and reproduce the clamping conditions adopted in the experimental tests.
200
100
200
100
200
100
19

The mechanical properties of the constitutive materials and the damage coefficients used in Model 1
and Model 2 are collected in Table 1. For each damage model, both states of epoxy glue are considered in
order to assess the capability of dealing with various failure behaviours according to the weakening of the
glue bonding. The damage model parameters are derived starting from standard mechanical tests and
commercial data provided by the glue and FRP plates suppliers and applying formulas that follow.
Table 1. Mechanical and damage parameters used in Model 1 and 2 for the two states of the epoxy adhesive
Material Parameter Description
Model 1 Model 2
Glue state 1 Glue state 2 Glue state 1 Glue state 2
Concrete
E
c
[N/mm
2
] Young modulus 3.2 x 10
4
3.2 x 10
4
3.2 x 10
4
3.2 x 10
4

c
[-] Poisson coefficient 0.21 0.21 0.21 0.21
k
c
[N] damage extension 0.1 0.1 0.1 0.1
w
c
[N/mm
2
] damage threshold 0.9765 x 10
-5
0.9765 x 10
-5
0.9765 x 10
-5
0.9765 x 10
-5

FRP
E
p
[N/mm
2
] Young modulus 1.6 x 10
5
1.6 x 10
5
1.6 x 10
5
1.6 x 10
5

p
[-] Poisson coefficient 0.4 0.4 0.4 0.4
k
p
[N] damage extension 0.05 0.05 0.05 0.05
w
p
[N/mm
2
] damage threshold 24.5 24.5 24.5 24.5
Epoxy
adhesive
E
g
[N/mm
2
] Young modulus 4.9 x 10
3
1.3 x 10
3
4.9 x 10
3
1.3 x 10
3

g
[-] Poisson coefficient 0.4 0.4 0.4 0.4
k
s
[N mm] damage extension (interfacial) 0.01 0.01 0.01 0.01
w
s
[N/mm] damage threshold (interfacial) 0.134 0.003 1.79 x 10
-3
4.31 x 10
-5

k
g
[N] damage extension (volume) - - 0.01333 0.01333
w
g
[N/mm
2
] damage threshold (volume) - - 0.179 0.004
T
s
k

[N/mm
3
] interface rigidity (tangential) 2.33 x 10
3
6.19 x 10
2
1.75 x 10
5
4.64 x 10
4

N
s
k

[N/mm
3
] interface rigidity (normal) 6.53 x 10
3
1.73 x 10
3
4.9 x 10
5
1.3 x 10
5

k
s,c
[N/mm]
damage interaction
epoxy - concrete
0.2 0.01
- -
k
s,p
[N/mm]
damage interaction
epoxy - FRP
1.0 x 10
-10
1.0 x 10
-10

- -
k
s1,c
[N/mm]
damage interaction
interface 1 - concrete
- - 0.2 0.01
k
s1,g
[N/mm]
damage interaction
interface 1 - epoxy
- - 0.2 0.06
k
s2,g
[N/mm]
damage interaction
interface 2 - epoxy
- - 0.2 0.06
k
s2,p
[N/mm]
damage interaction
interface 2 - FRP
- - 1.0 x 10
-10
1.0 x 10
-10


The rigidity of the interface(s) is expanded into tangential
T
s
k

and normal
N
s
k

components which are


respectively related to the components of the displacement vector u in the x and y directions, respectively.
These tangential and normal components are calculated by respectively dividing the shear and Youngs
moduli by the actual thickness of the glue layer, i.e., 0.75 mm for the first model and 0.01 mm for the
second. The second value has been arbitrarily chosen to be small enough to reproduce the infinitesimal
20

dimension of the interface. Smaller values would have provided equivalent results. The damage threshold
parameters w are obtained by adopting the expression of the Duprs energy, which at glue interface and
domain reads respectively:
T
s
eg
s
k
w

2
1
2

= , (24)
g
eg
g
G
w
2
2
1

= , (25)
where
g
G is the shear modulus of the glue and
eg
corresponds to the stress at the end of the elastic
domain; in the case of the epoxy glue in the state 1 (brittle elastic behaviour),
eg
is assumed equivalent
to the mean tensile strength, whose value is equal to 25 MPa. As regards the glue in the state 2 (elasto-
plastic behaviour), the elastic limit
eg
derived from the tensile tests is reduced to 2 MPa, which
corresponds to 1/3 of the measured tensile strength (see Figure 9a in [1]). It is worthy to highlight that the
two damage thresholds as defined in (24) and (25) have the dimensions of energy per unit volume and
area, respectively. To give a physical interpretation, damage thresholds can be considered as the energy
that has to be produced to break the microscopic links in a unit volume (or area) of material.
Similarly, damage thresholds in the adherents (concrete and CFRP plate) are respectively given by:
c
ec
c
E
w
2
2
1
= and
p
ep
P
E
w
2
2
1

= (26)
As both adherents exhibit brittle elastic behaviours, elastic limits can be replaced by mean tensile
strengths, which take values of 2.5 and 2800 MPa for concrete and composite, respectively.
The damage extension parameters k are set according to previous studies [6, 7] and kept unchanged
for both the cases. In Model 2, the damage extension coefficient
g
k referring to the volume of the glue is
obtained by multiplying the interfacial parameter
s
k of Model 1 by the actual thickness of the adhesive
layer. In order to simplify the comparison between the models, both the interfaces of Model 2 are
assumed to be part of the glue layer and their mechanical properties and damage features are selected
accordingly. Consequently, the surface-volume interaction parameters
g s1,
k and
g s2,
k for Model 2 are
21

equal to
c s1,
k , which corresponds to
c s,
k for Model 1. This value can be numerically estimated by
considering the bond strength of the glue on the concrete, as was done in [6]. In the case of the second
state of adhesive, the value of
c s1,
k is lowered in order to consider the reduction of the interaction between
the interface and the volume of the adhesive due to the weakening of the glue bonding. Finally, since the
interaction between the glue and FRP plate is negligible, the interaction parameter
p s,
k for Model 1 and
p s2,
k for Model 2 are set to be practically zero for both states of adhesive.
4.3 Results and discussion
The damage patterns obtained at the end of the shear test for the two damage models and the two
states of epoxy are collected in Figure 7. For the sake of clarity in the representation, only the volume
damage
i
is portrayed in Figure 7, being the one that affects more the actual failure mechanism for both
the states of epoxy. The nodes where
i
is equal to 0 represent the complete material failure that leads to
opening of macro-cracks. The portions of the specimen characterised by values of
i
smaller that 1 are
partially damaged, e. g., occurrence of micro-cracking. The fracture within concrete produced by the stiff
adhesive (glue state 1) is accurately depicted by both of the two damage models (Figure 7a and Figure
7c). Conversely, only the improved damage model is capable of predicting the cohesive failure within the
adhesive layer when the weak glue is involved (Figure 7d). The simplified damage model provides a
misleading prediction that entails the crack of concrete immediately below the interface with the glue
(Figure 7b). The limits of Model 1 in predicting the fracture mode can be ascribed to the lack in the
description of the three-phase assembly. For this reason, in the following sections, the characterisation of
the damage behaviour for both adhesive states is carried out only by means of the second damage model.
The position of the fracture at the end of the shear test can be determined also by the displacement
field. A sudden change of displacement over a small area (large gradient) identifies a detachment within
the material, i.e., a fracture. The horizontal displacements illustrated in Figure 8 are in perfect agreement
with the volume damage patterns and show clearly the position of the fracture line in different domains
according to model and state that are considered.
22






Figure 7: Volume damage patterns
i
for (a) Model 1 and epoxy state 1, (b) Model 1 and epoxy state 2, (c) Model 2
and epoxy state 1, (d) Model 2 and epoxy state 2. Fracture line corresponds to
i
equal to 0
The investigation of the damage evolution throughout the whole rupture test would require time-
consuming analysis of all the steps that compose the numerical simulation. However, the identification of
the most significant stages of the damage evolution may rely on the computation of the average damage
index introduced in Section 3. In the case of the improved damage model, this index can be distinguished
between the volume and surface components. The first, which are related to the volume of concrete
c

and volume of glue
g
, are depicted in Figure 9a. The indexes of the interface between concrete and
adhesive layer
1 s
and that between adhesive layer and composite
2 s
are shown in Figure 9b.
Both the volume and interface average damage indexes show extremely different evolutions of the
damage behaviour according to the adhesives properties. For the first state of glue, the failure mainly
concerns the concrete specimen, whose brittle behaviour is outlined by the initial rapid trend of the
c

evolution. Afterwards, the
c
index shows a relatively long phase of asymptotic decrease, which is
finally interrupted by a sharp drop to unchanging values that marks complete rupture. The adhesive layer
is slightly affected and only at the first stages of concrete cracking: the
g
index decreases to a minor
(a)
(b)
(c)
(d)
23

extent a few seconds after the damage triggering in concrete. The adhesive failure at the interfaces does
not occur at all for the first state of glue ( ( ) ( ) t t t
s s
= = 1
2 1
in Figure 9b).




Figure 8: Horizontal displacements for (a) Model 1 and epoxy state 1, (b) Model 1 and epoxy state 2, (c) Model 2
and epoxy state 1, (d) Model 2 and epoxy state 2 (displacements in mm).
The weak adhesive shows a failure mechanism that is diametrically opposed to the former one. The
dashed line of
g
in Figure 9a decreases smoothly from the very first stages of the test. A small rise of
the index occurs towards the end, when a new large portion of the glue layer is affected by damage but to
a minor extent. In fact, the average in equation 20 is shifted towards higher values because a large number
of new nodes affected by minor damage (value of
k
close to 1) is taken into account. This behaviour
can be physically interpreted by micro-cracking, i.e. formation of microscopic cracks whose coalescence
leads to a macroscopic crack. It is authors belief that the loss of monotonic decrease with loading of the
average damage index is not misleading because it can be referred to a unique and well-defined
phenomenon, i.e., minor damage propagation.
The trend of the
c
index proves that concrete is practically unaffected by the cohesive failure within
the glue layer. Unlike,
1 s
and
2 s
indexes portrayed in Figure 9b show the same smooth evolution
of
g
. This is the result of the enhanced surface-volume interaction that is gained for the weak epoxy.
(a)
(b)
(c)
(d)
24

The circles depicted in Figure 9 identify the time instants that exhibit significant changes in the
damage evolution, such as damage triggering, turning points in the trend and final failure. They find
correspondence in the evolution of the damage patterns as a function of time, portrayed in Figures 10-11,
which clarify the process of the failure mechanisms observed in the experimental tests for the strong and
weak adhesive, respectively. In the first case (Figure 10), the crack in the concrete block originates a few
millimetres under and away from the front side of the interface (t = 27s). The fracture proceeds
backwards along the adhesive layer until it reaches the unglued concrete surface when complete failure
occurs (t = 58s). When the weak glue is involved (Figure 11), the fracture initiates within the adhesive
layer at the front side of the interface (t = 52s) and proceeds backwards with a constant increase rate.
When the crack entails about 80% of the length of the adhesive layer (t = 59s), damage rapidly propagates
in the rest of the joint and a sudden rupture occurs (t = 60s).
0 10 20 30 40 50 60 70 80 90 100
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time steps [s]
A
v
e
r
a
g
e
d
a
m
a
g
e

in
d
e
x

c
glue state 1

g
glue state 1

c
glue state 2

g
glue state 2
A
v
e
r
a
g
e

d
a
m
a
g
e

i
n
d
e
x
Time steps [s]

c
glue state 1

g
glue state 1

c
glue state 2

g
glue state 2

0 10 20 30 40 50 60 70 80 90 100
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time steps [s]
A
v
e
r
a
g
e

d
a
m
a
g
e

in
d
e
x

s1
glue state 1

s2
glue state 1

s1
glue state 2

s2
glue state 2
Time steps [s]
A
v
e
r
a
g
e

d
a
m
a
g
e

i
n
d
e
x

s1
glue state 1

s2
glue state 1

s1
glue state 2

s2
glue state 2

(a) (b)
Figure 9: Evolution of throughout the shear test: (a) concrete and glue volume indexes for strong (solid lines) and
weak (dashed lines) epoxy; (b) interface indexes for strong (solid lines) and weak (dashed lines) epoxy.
The propagation of damage in space is analysed by means of the vectorial damage indexes related to
the distribution of damage along the coordinated axes (see 3.2). The distributions presented in Figure 12
refer to the three significant stages of the damaging process already pointed out in the case of the stiff
glue (Figure 10). The propagation of the crack along the adhesive layer can be clearly recognized from
the spreading of the damage distribution
X
along the X axis (in Figure 12a, the range of X coordinates
whose
X
values is less than 1 widen with the test). Differently, the extension along the vertical
25

direction does not take place (in Figure 12b, Y coordinates affected by damage remain unchanged
throughout the test), while an increase in the extent of
Y
is observed.




Figure 10: Evolution of the volume damage pattern
i
sampled at the most significant stages (see circles of Fig.8a)
of the shear test carried out on the strongly bonded assembly (glue state 1).




Figure 11: Evolution of the volume damage pattern
i
sampled at the most significant stages (see circles of Fig.8b)
of the shear test carried out on the weakly bonded assembly (glue state 2).
The distributions depicted in Figure 13 describe the spatial evolution of the damaging process
observed for the weakly bonded specimen (Figure 11). Analogously to Figure 12, the longitudinal
propagation of fracture is shown by the spreading of the damage distribution
X
along X (Figure 13a)
and the increasing extent of
Y
for the other direction (Figure 13b). The discrepancy between the final
values attained by the damage distributions
X
and
Y
proves that the failure propagation takes place
t = 58 s
t = 27 s
t = 54 s
t = 60 s
t = 52 s
t = 59 s
26

mainly along the horizontal axis. Indeed, in Figure 13a the portion of failed material along Y is so limited
that averaging over the whole height of the specimen leads to values of
X
very close to 1. Therefore,
one can deduce that damage is spread over a large length along X, but not in the other direction.
Moreover, in Figure 13b, small values of
Y
and its sharp distribution along Y suggest that the damaged
zone is limited to the thickness of the adhesive layer for all of the three time steps. In conclusion,
combining the analysis of the two distributions provides a clear outlook of the area affected by damage.
0 50 100 150 200 250 300 350 400 450
0.92
0.94
0.96
0.98
1
Abscissas [mm]
D
a
m
a
g
e

d
is
t
r
i
b
u
t
i
o
n


t = 27 s
t = 54 s
t = 58 s
X coordinates [mm]
D
a
m
a
g
e

d
i
s
t
r
i
b
u
t
i
o
n

(a)
0 50 100 150 200 250
0.5
0.6
0.7
0.8
0.9
1
Ordinates [mm]
D
a
m
a
g
e

d
i
s
t
r
i
b
u
t
io
n


t = 27 s
t = 54 s
t = 58 s
Y coordinates [mm]
D
a
m
a
g
e

d
i
s
t
r
i
b
u
t
i
o
n

(b)
Figure 12: Evolution of the damage distribution
X
along the X axis (a) and
Y
along the Y axis (b) for the strongly
bonded assembly: comparison between the time steps t = 26 s, t = 52 s and t = 56 s. Origin of coordinates located at
the bottom-left corner of the concrete prism (Figure 5a).
0 50 100 150 200 250 300 350 400 450
0.9975
0.998
0.9985
0.999
0.9995
1
Abscissas [mm]
D
a
m
a
g
e

d
is
t
r
i
b
u
t
i
o
n


t = 52 s
t = 59 s
t = 60 s
X coordinates [mm]
D
a
m
a
g
e

d
i
s
t
r
i
b
u
t
i
o
n

(a)
195 200 205 210 215
0.6
0.7
0.8
0.9
1
Ordinates [mm]
D
a
m
a
g
e

d
is
t
r
i
b
u
t
i
o
n


t = 52 s
t = 59 s
t = 60 s
Y coordinates [mm]
D
a
m
a
g
e

d
i
s
t
r
i
b
u
t
i
o
n

(b)
Figure 13: Evolution of the damage distribution
X
along the X axis (a) and
Y
along the Y axis (b) for the weakly
bonded assembly: comparison between the time steps t = 52 s, t = 59 s and t = 60 s. Origin of coordinates located at
the bottom-left corner of the concrete prism (Figure 5a).
27

Finally, the failure index is computed for the concrete substrate
c
I

, adhesive layer
g
I

and interface
s
I

. These indexes quantify to what extent each component of the glued assembly is affected both in time
and space. Moreover, the combination of the failure indexes related to different components according to
(22) and (23) provides the global bulk

and interface
s
I

indexes, respectively.
The values obtained for the two states of epoxy are summarised in Table 2. These results quantify the
change in the failure mechanism depending on the mechanical and bonding properties of the adhesive. In
particular, the stiff and resistant epoxy leads to concrete fracture (high values for
c
I

and

), whereas
the weak and ductile glue moves the failure towards an adhesive decohesion that involves also slight
interface separation (high values for
g
I

,
s
I

and
s
I

). The comparison between the numerical and


experimental results in terms of failure mode and maximum shear load is presented in Table 3. A good
agreement is observed for both the types of glue state.
Table 2. Failure indexes and their combinations for the two adhesive states

c
I


g
I


s
I


s
I


Glue state 1 0.277 0.004 0 0.986 0
Glue state 2 4.67 x 10
-5
0.505 0.245 6.2 x 10
-5
0.333


Table 3. Comparison between numerical and experimental results for the two adhesive states
Numerical Experimental
Failure mode Max shear load [kN] Failure mode Max shear load [kN]
Glue state 1 concrete fracture 51 concrete fracture 54
Glue state 2
cohesive crack
within the glue layer
47
cohesive crack
within the glue layer
46
The deformation of the composite plate throughout the shear test was measured by means of five strain
gauges placed in the locations that are shown in Figure 5a. In Figure 14, the experimental measurements
are compared with the results of the numerical simulations that are carried out for both types of epoxy. A
good agreement between the outcomes of the damage model and the experimental evidence is observed in
both cases. The shape of the strain curves changes from a piecewise linear elastic-pseudo plastic profile
for the stiff epoxy (Figure14a) to a more rounded trend for the ductile epoxy (Figure 14b). This result
confirms the evolution of the load transfer mechanism outlined in [1] as a function of the stiffness of the
28

glue. The stiffness reduction and enhanced ductility of the adhesive lead to plastic deformation and
cohesive fracture within the joint.
0 10 20 30 40 50 60
0
500
1000
1500
2000
2500
Force [kN]
S
t
r
a
in

[
m
ic
r
o
-
e
p
s
ilo
n
]


S1 numerical
S2 numerical
S3 numerical
S4 numerical
S5 numerical
S1 experimental
S2 experimental
S3 experimental
S4 experimental
S5 experimental
S
t
r
a
i
n
[
x
1
0
-
6
]
Force [kN]

0 5 10 15 20 25 30 35 40 45 50
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Force [kN]
S
t
r
a
in

[
m
ic
r
o
-
e
p
s
ilo
n
]


S1 numerical
S2 numerical
S3 numerical
S4 numerical
S5 numerical
S1 experimental
S2 experimental
S3 experimental
S4 experimental
S5 experimental
S
t
r
a
i
n
[
x
1
0
-
6
]
Force [kN]

(a) (b)
Figure 14: Strain at the gauges locations vs. applied force: comparison between the numerical and experimental
results for the strongly (a) and weakly (b) bonded assemblies.
As final proof of the capability of the damage model to accurately reproduce different detachment
mechanisms, other two interfacial failure modes are considered. The first is the failure mode 4 of Figure
1, which finds experimental evidence in the fractured surface shown in Figure 6a. The adhesive failure at
the polymer/FRP interface can be ascribed to a poor adherence of the epoxy on the composite plate,
probably due to its incomplete curing. The second failure mode is the number 2 of Figure 1 and involves
a poor adherence between epoxy and concrete substrate that leads to adhesive/concrete separation. This
last interface fracture mode has not occurred in the experimental campaign, however, it has already been
observed in association with absent or ineffective sand-blasting treatment of concrete surface and
moisture degradation, as pointed out in [18].
No experimental data directly related with the interfacial failure modes (for instance, a measure of
penetration of the glue into the porosity of the matrix of the adherents) were available. Therefore, the
purpose in this part is simply to show that the improved damage model is able to reproduce all possible
failure mechanisms the bonded assembly may experience when an appropriate selection of models
parameters is made. We may suppose that the materials properties are not affected when interfacial
failure modes occur. Thus, the values of bulk parameters are kept the same as for the glue state 1 of Table
1. Debonding between glue and adherents is ascribed to poor adhesion. Lack of bonding is here modelled
29

by decreasing the values of damage parameters related to the cohesion of the interface
s
w and the
interaction between interface and adherent/adhesive
i s,
k . The damage threshold and damage interaction
parameters of the upper and lower interfaces are changed in turn in order to obtain the adhesive failure at
FRP/epoxy and epoxy/concrete interface, respectively. In this way, the former assumption that considered
both the interfaces as part of the glue layer is released to imagine them as different entities separately
affected by different phenomena. Obviously, the impossibility to distinguish between the two interfaces
prevents Model 1 to discriminate between the two interfacial failures.
Table 4 shows the values of the damage parameters that were selected to achieve the two interfacial
failure modes (modes 2 and 4 in Figure 1). The values used for the cohesive cracking mode (failure mode
1 of Figure 1) are given as well for comparison. Large variations of the parameters are necessary to
achieve failure at the interfaces. One possible explanation is that the whole set of material parameters as it
was derived from available experimental measurements is much more sensitive to cohesive cracking than
to interfacial failures. Probably, smaller changes would have been obtained for a set of material
parameters less prone to cracking in the adherent. More rigorous approaches that aim at identifying
parameters by taking their interaction into account are envisaged and will be pursued in future works.
Table 4. Damage parameters used in Model 2 to simulate interfacial failures.
Parameter Description
Failure mode 1
Concrete craking
Failure mode 2
Concrete/epoxy
interface
Failure mode 4
FRP/epoxy
interface
w
s1
[MPa mm] damage threshold (concrete/epoxy interface) 0.3 3.0 x 10
-4
0.3
w
s2
[MPa mm] damage threshold (FRP/epoxy interface) 0.3 0.3 3.0 x 10
-5

k
s1,c
[MPa mm] damage interaction interface 1 - concrete 0.2 1.0 x 10
-3
0.2
k
s1,g
[MPa mm] damage interaction interface 1 - epoxy 0.2 1.0 x 10
-3
0.2
k
s2,g
[MPa mm] damage interaction interface 2 - epoxy 0.2 0.2 2.0 x 10
-6


The volume and interfacial damage patterns that are obtained for the two adhesive failures are
depicted in Figure 15. In both cases, the improved damage model manages to accurately reproduce the
separation between adherents and adhesive through continuously failed interface. The interfacial fracture
is clearly represented by the straight line along the joint where the surface damage quantity
s
vanishes
(Figure 15b and zoom). Damage in the volume is limited to a small propagation of the fracture in concrete
at the end of the glue joint when the lower interface is affected (Figure 15a). The FRP/epoxy separation
30

(Figure 15d and zoom) that is coupled with concrete cracking at the front side of the joint (Figure 15c) is
comparable to the interface debonding observed in Figure 6a.






Figure 15: Volume and interface damage patterns for the two adhesive failures: (a) volume damage
i
at the
concrete/epoxy interface, (b) surface damage
s
at the concrete/epoxy interface, (c) volume damage
i
at the
FRP/epoxy interface, (d) surface damage
s
at the FRP/epoxy interface.

The results obtained for the computation of the failure index for concrete
c
I

, adhesive layer
g
I

and
interfaces
1 s
I

and
2 s
I

, together with those of the global bulk

and interfaces
1

s
I

and
2

s
I

indexes
are summarised in Table 5. As expected, the adhesive failure modes lead to a diametrically opposed result
compared to the cohesive fracture modes of Table 2. The global bulk index

is drastically reduced,
while the split of the global interface index between the surfaces
1 s
and
2 s
provides a clear distinction
of the detached interface.
Table 5. Failure indexes and their combinations for the two adhesive failures

c
I


g
I


1 s
I


2 s
I

s
I

s
I


Failure mode 2
Concrete/epoxy
interface

0.007 0.046 0.448 0 0.015 0.858 0
Failure mode 4
FRP/epoxy
interface
0.047 0.003 0 0.132 0.257 0 0.739
(a)
(d)
(b)
(c)
31


5. CONCLUSIONS
An improved damage modelling approach model is proposed to deal with different failure modes
depending on the properties of the adhesive. The model involves a three-domain representation of FRP
reinforced concrete structures and accounts for the interaction between domain and interface damage to
predict different failure mechanisms for a single lap joint shear test.
The damage model proves to be capable of accurately reproducing the failure mechanism and fracture
line, outperforming a former two-domain version of the damage model over different debonding modes.
The modelling of the thickness of the adhesive layer allows detecting the occurrence of cohesive failure
within the glue when its mechanical and bonding properties weaken. The sensitivity of the damage
models response to the variation of the epoxy parameters is highlighted. Indeed, the damage behaviour
significantly changes when the weakening of the glue is considered. Adhesive failure modes are
investigated too. The proposed damage model is capable to discriminate between the concrete/epoxy and
FRP/epoxy separation. The application of this new model, once confirmed by more exhaustive studies,
would offer promising prospects to investigate failure modes related to poor adherence of the epoxy on
the substrates.
Original damage and failure indexes are introduced to locate damage in time and space, quantify the
evolution of the detachment behaviour and distinguish different types of failure. Moreover, the distinction
of the glue layer from the interfaces and substrates in the improved damage model allows the computation
of the bulk and interface global indexes that proved to be suited to quantify the contribution of each type
of damage mechanism to the global failure. In future works, the proposed failure indexes will be used as
criteria to guide the selection of the damage model parameters within an optimisation framework aimed at
the definition of parameters sets related to the different failure modes that may occur.
By way of conclusion, the proposed damage model is an effective and robust tool for providing an
accurate description of the fracture behaviour of FRP-reinforced structures when they undergo states
which are capable to activate different damage mechanisms, such as glue weakening and lack of
adherence. Further studies are currently carried out focusing on the relationship between the damage
model parameters and their interactions and the evolution of different failure modes.
32

References
[1] K. Benzarti, S. Chataigner, M. Quiertant, C. Marty, C. Aubagnac, Accelerated ageing behaviour of the
adhesive bond between concrete specimens and FRP overlays, Construction and Building Materials, 25(2),
2011, 523538.
[2] M.A.V. Silva, H. Biscaia, Degradation of bond between FRP and RC beams, Composite Structures, 85(2),
164 (2008), 164-174.
[3] I. Nishizaki, Y. Kato, Durability of the adhesive bond between continuous fibre sheet reinforcements and
concrete in an outdoor environment, Construction and Building Materials, 25(2), 2011, 515-522.
[4] K. Benzarti, M. Quiertant, C. Aubagnac, S. Chataigner, I. Nishizaki, Y. Kato, Durability of CFRP
strengthened concrete structures under accelerated or environmental ageing conditions, Concrete Repair,
Rehabilitation and Retrofitting, M.G . Alexander et al. ed., CRC Press, 2009, 421-422.
[5] M. Frmond, B. Nedjar, Damage, gradient of damage and principle of virtual power, Int. J. Solids Structures
8 (1996), 10831103.
[6] F. Freddi, M. Frmond, Damage in domains and interfaces: a coupled predictive theory, Journal of Mechanics
of Materials and Structures, 1(7), 2006, 1205-1233.
[7] K. Benzarti, F. Freddi, M. Frmond, A damage model to predict the durability of bonded assemblies Part I:
debonding behaviour of FRP strengthened concrete structures, Construction and Building Materials, 25(2),
2011, 547-555.
[8] K. Benzarti, C. Perruchot, M. Chehimi, Surface energetics of cementitious materials and their wettability by
an epoxy adhesive, Colloids Surf A Physicochem Eng Aspects, 286(13), 2006, 7891.
[9] I. Neves, M. Chabut, C. Perruchot, M. Chehimi, K. Benzarti, Interfacial interactions of structural adhesive
components with cement pastes studies by inverse gas chromatography (IGC), Appl Surf Sci, 238(14),
2004, 5239.
[10] F. Djouani, C. Connan, M. Chehimi, Interfacial chemistry of epoxy adhesives on hydrated cement paste, Surf
Interface Anal, 40(34), 2008, 14650.
[11] F. Freddi, M. Savoia, Analysis of FRP-concrete delamination via boundary integral equations, Engineering
Fracture Mechanics, 75(6), 2008, 16661683.
[12] M. Frmond, K.L. Klutter, B. Nedjar, M. Shillor, One-dimensional damage model, Adv Math Sci Appl, 8(2),
1998, 54170.
33

[13] P. Argoul, K. Benzarti, F. Freddi, M. Frmond, T.H.T. Nguyen, A damage model to predict the durability of
bonded assemblies Part II: Parameter identification and preliminary results for accelerated ageing tests,
Construction and Building Materials, 25(2), 2011, 556-567.
[14] B. Nedjar, Elastoplastic-damage modelling including the gradient of damage: formulation and computational
aspects, Int. J. Solids Structures, 38, 2001, 54215451.
[15] B. Nedjar, A theoretical and computational setting for a geometrically nonlinear gradient damage modelling
framework, Comp. Mech, 30, 2002, 6580.
[16] F. Djouani, C. Connan, M. Delamar, M. Chehimi, K. Benzarti, Cement pasteepoxy adhesive interactions,
Construct Build Mater 25, 2010, 411423.
[17] W. Bangerth, R. Hartmann, G. Kanschat, Deal. II differential equations analysis library, technical reference,
2011. http://www.dealii.org.
[18] C. Au, Moisture degradation in FRP bonded concrete systems: an interface fracture approach, Sc.D. Thesis,
Massachusetts Institute of Technology, Cambridge, MA, 2004.
[19] D. Ferretti, F. Freddi, G. Rosati, P. Carrara Snap-back shear tests of FRP plates bonded to concrete
Engineering Fracture Mechanics 78, 2663-2678, 2011.
[20] P. Carrara, D. Ferretti, F. Freddi, Debonding behavior of ancient masonry elements strengthened with CFRP
sheets, Composites Part B 45, 800810, 2013

You might also like