Professional Documents
Culture Documents
A Non Linear Thermodynamical Model For S PDF
A Non Linear Thermodynamical Model For S PDF
a r t i c l e i n f o a b s t r a c t
Article history: In Reinforced Concrete structures, the concrete damage configuration as well as the crack pattern distri-
Received 10 October 2011 bution depend directly on bonding, and its implementation for numerical analysis is still complicated. In
Accepted 9 April 2012 this work, a new thermodynamical non linear bond model is introduced taking into account tangential
Available online 9 May 2012
bond deterioration and coupling between tangential slip and normal decohesion/penetration. For FEM
implementation, the model (based on a stress–strain relationship) is supported by a non-width interface
Keywords: element. The calibrated material parameters are grouped in four sets: elastic behavior; tangential degra-
Steel–concrete bonding
dation; crack friction and confinement; and normal behavior. Finally, some numerical examples are pre-
Thermodynamical formulation
Reinforced Concrete
sented in this paper.
Interface element Ó 2012 Elsevier Ltd. All rights reserved.
Finite element method
0045-7949/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruc.2012.04.005
30 N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45
Interface material cracking due to shear efforts and/or The tangential degradation of bonding is assumed to be similar
decohesion. to a pseudo plastic/damage behavior.
Inelastic deformations due to sliding (incompatibility of A normal inter-penetration between materials is allowed, but it
deformations). is never greater than a maximum value of surface penetration.
Hysteresis cycles associated to the crack friction. The surface of the steel bar is considered completely stiff and
Coupling between tangential response and normal stresses. undeformable.
Instead of supposing an initial rigid connection between the
In what concerns the finite element implementation, instead of concrete and the steel rebar (without sliding), we assumed an
using a classical interface element based on a displacement jump, initial sliding as well as an initial elastic behavior for bonding
we created a degenerated non-width interface element using a stan- that is valid only if the local strains have not yet reached the
dard base of shape functions for a quadrilateral-four nodes element. yield strain limit, corresponding to the strain associated to the
This work is articulated in five sections outlined as follows: Section elastic shear resistance of the concrete.
1 explains the motivations as well as the general lines of this re- The behavior of the multiple micro-cracks formed in the inter-
search; in Section 2 we present in detail the thermodynamical for- face region is very similar to the behavior of a particular homo-
mulation of the non linear bond model, preceded by a brief geneous concrete zone, in which the stiffness is lower than the
description of the bond phenomena and the hypotheses that sup- stiffness of the rest of the concrete, but the ‘‘flexibility’’ of the
port our proposal. Section 3 is focused not only in the numerical zone makes easier the evolution of the sliding from small defor-
implementation of the constitutive model for bonding, but also in mations to large slips, without losing the transfer of efforts and
a brief description of the interface element that supports the bond- stresses between steel and concrete.
ing law behavior. In Section 4, the analysis and calibration of each The brutal degradation of bonding as soon as large slips are
parameter is fully described. Finally, in Section 5 we present some developed is due to the growing and coalescence of micro-
of the numerical applications in which the model has been used, iso- cracks which degenerate in the potential surface of cylindrical
lated or mixed with other non linear material models. failure previously defined.
In the same manner, the total tangential stress rT can be calculated Based on this separation, and associating it to the set of damage
by: equations, it can be observed:
@W
rT ¼ q ¼ Gð1 DT Þ T þ G DT T fT ð2:6Þ – A true separation between damage and crack friction.
@ T
– A partition of the isotropic hardening.
and the sliding tangential stress rfT by:
@W Hereafter, the isotropic hardening associated to the damage
rfT ¼ q f
¼ G DT T fT ð2:7Þ variable can be expressed as:
@ T
@W ZT1 if T < T 6 T 2
In the last expression, rfT is associated to the sliding friction strain ZT ¼ q ¼ H 0 ðzÞ ¼ 1
ð2:11Þ
fT . @z ZT1 ZT2 if T 2
< T
By the way, if we derivate the thermodynamical potential (2.4) In the last expression, Z T 1 and Z T 2 correspond to:
with respect to the tangential damage variable DT, we obtain the (
qffiffiffiffiffiffiffiffiffi rffiffiffiffi )2
restitution energy represented by Y: 1 G T
ZT1 ¼ Y T1 þ ln ð1 þ zT Þ ð2:12Þ
@W 1 1 ADT1 2 T 1
Y ¼ q ¼ T G T T fT G T fT
@ T 2 2 1 zT
ZT2 ¼ Y T2 þ ð2:13Þ
¼ Y DT þ Y fT ð2:8Þ ADT2 ð1 þ zT Þ
Subsequently, the damage yield function is established for the tan-
in which Y DT and Y fT are the damage restitution energy and the gential direction as:
rate of restitution energy associated to crack friction, respectively.
The constitutive function for the kinematics hardening leads to UDT ¼ Y DT ðY T1 þ ZT Þ 6 0 ð2:14Þ
the definition of the back stress function: Regarding to the yield points that control the tangential damage
@W evolution function, they will be expressed as a function of the en-
v ¼ q ¼ca ð2:9Þ ergy Y DT ; for that reason, the first expression associated to the per-
@a
fect bond will be written as:
while the constitutive function for the isotropic hardening is ex-
pressed as:
32 N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45
exp ADT1 ð Y DT Y DT1 Þ law of Drucker, that is, the linearity of the law: here, the non linear
: G ; terms are included in the dissipative potential. In which concerns
( ) to the sliding criteria, it takes the classical form of the Drucker–
1 Prager yield function taking into account the effect of radial con-
BD
ð2:18Þ
1 þ ADT2 hY DT Y T 1 iþ T2 finement on the sliding:
Uf ¼
rfT v
þ c I1 6 0 ð2:25Þ
responds to the transition region from small deformations to large
displacements with two parameters, ADT1 and BDT1 ; and the part of In this expression, v is the back stress, c is a material parameter
final damage corresponding to the crack friction, as well with two associated to confinement influence, and I1 is the first invariant of
parameters ADT2 and BDT2 . the stress tensor expressed as:
Finally, tangential isotropic hardening will be controlled by the
1 1
next functions: I1 ¼ Tr½r ¼ rN ð2:26Þ
3 3
ZT1 ¼ Y DT Y T1 ð2:19Þ
By the way, the initial yield for crack sliding is 0. Based on the
principle of maximal plastic dissipation, the evolution laws could
0; if Y T1 <Y DT 6Y T2
ZT2 ¼ ð2:20Þ be deduced from the plastic potential expression:
Y DT Y T2 if Y T2 <Y DT
3
Upf ¼
rfT v
þ c I1 þ a v2 ð2:27Þ
From these expressions, it can be remarked that Z T 2 is not taken into 4
account in the region of transition from small deformations to large Here, a is a material parameter, while the non linear effects in the
displacements. kinematics hardening are introduced thanks to the quadratic term
v2. Accordingly, the evolution laws associated to the sliding strain
2.3.4. Damage in the normal direction and kinematics hardening will be expressed as:
In the model, the behavior of bonding in the normal direction
@ Upf @ Upf
must be taken into account for two essential cases: the first one _ fT ¼ k_ f and a_ ¼ k_ f ð2:28Þ
is the decohesion between steel bars and concrete produced by @ rfT @v
tension forces, while the second case is the inter-penetration of
The sliding multiplier k_ f is calculated numerically by imposing a
surfaces induced by compression forces between both materials.
consistency condition.
In other words, these two mechanisms may be interpreted respec-
tively as an opening or a closing of multiple cracks, which can be
controlled by a constitutive function for the normal behavior 3. Numerical implementation in a finite element context
uncoupled from the tangential behavior. In compression, assuming
a small inter-penetration between steel and concrete -that means 3.1. Finite element support: the non-width interface element
N 6 0 -, we adopt a classical expression for an elastic behavior
law: Briefly, the proposal is based on the development of a new joint
element in order to simplify calculations and reduce meshing con-
rN ¼ E hN i if N 6 0 ð2:21Þ struction problems. This interface element will be adapted from a
For decohesion, instead of using a behavior law for cohesive cracks classical isoparametric finite element but without any dimension
(see [5]) or a Barenblatt law (see [6]), we preferred to use a tension or thickness, able to carry out tangential and normal stresses that
damage behavior law for concrete, that means: would be calculated from normal and tangential strains. The do-
main of a straight-edged quadrilateral element is defined by the
rN ¼ ð1 DN Þ E hN iþ if N > 0 ð2:22Þ location of its four nodal points xea ; a ¼ 1; . . . ; 4 in the 2D-space. In
N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45 33
order to control strains into the joint element, we introduce the @ Upf
hpen parameter which is a physical value that corresponds to the Dv ¼ vðiþ1Þ vðiÞ ¼ c Da ¼ c Dkf ð3:4Þ
@v
maximal penetration due to thickness of compressed-pulverized
concrete [8]; that will allow us to define a normal stress even if this ðiþ1Þ ðiÞ @ Upf
DrfT ¼ rfT rfT ¼ G DT DfT ¼ G DT Dkf ð3:5Þ
one does not have any normal dimension, avoiding the classical @ rfT
stability drawbacks when dealing with contact forces. For the joint
element, classical shape functions will be used. Even so, the ele- By combining these expressions with the yield surface equation
ment stiffness matrix is constructed with derivatives of shape and making Uf = 0, we may deduce the increment of the multiplier
functions; so in order to avoid undetermined Jacobian for the inter- Dkf for each iteration i:
face normal dimension, we affect the nodal coordinate functions
UfðiÞ
introducing hpen projections in calculations. Finally, in order to Dkf ¼ ð3:6Þ
p ðiÞ p ðiÞ
complete the joint element formulation, we have adopted two @ Uf ðiÞ @ Uf @ Uf ðiÞ @ Uf
f G DT f þ @v
c @v
@ rT @ rT
points of integration for tangential direction and only one for nor-
mal direction, using the Gaussian rules for integrals in several As soon as Dkf is obtained, it can be substituted in Eqs. (3.4) and
dimensions. For a more detailed explanation about the numerical (3.5) in order to update the thermodynamics forces rfT and v. These
implementation of the interface element, see [9]. iterations must continue until the consistency condition be verified
(by the respect of a given tolerance).
3.2. Numerical integration
3.2.2. Algorithm for numerical resolution
The numerical integration of the model has been widely simpli-
The implicit resolution algorithm for the ‘‘crack friction’’ part is
fied thanks to the uncoupling of the bonding behavior in both
presented as follows:
directions, diminishing the cost of calculations and avoiding the
use of a iterative method.
(i) Geometrical updating:
The behavior in the normal direction has been explicitly inte-
grated by the definition of an yield surface; in the tangential direc- ðT Þnþ1 ¼ ðT Þn þ rs uT
tion, the behavior has been separated in two parts: damage and
sliding crack friction. The first one is explicitly integrated by the
(ii) Elastic prediction:
definition of two limit surfaces, while the second part must be cal-
ð0Þ
culated by an implicit method. For that, we have chosen the classi-
fT ¼ fT ;
cal resolution known as ‘‘return-mapping algorithm’’, developed by nþ1 n
Ortiz and Simo [10]. In the next paragraph we will explain in a de-
e ð0Þ
tailed way how it was implemented this algorithm. T nþ1
¼ ðT Þnþ1 fT ;
nþ1
ð0Þ
3.2.1. The return mapping algorithm for the implicit analysis
ðaÞnþ1 ¼ ðaÞn ;
Crack friction effects on bonding can be calculated on the
ð0Þ ð0Þ ð0Þ
ðrÞnþ1 ¼ r eT nþ1 ; ðaÞnþ1
framework of a pseudo-plastic behavior with a non linear kinemat-
ics hardening included. According to the method proposed by Ortiz
and Simo [10], a decoupling between elastic and plastic strains for
each time instant may be assumed, which allows doing an elastic (iii) Evaluation of the yield surface:
prediction of strains and stresses: nonetheless this assumption 8
> if YES; ending of cycle and :
must be handled carefully, because there is really no physical >
>
>
> ð0Þ
>
> f f
decoupling of strains (for a more detailed explanation of these is- > T nþ1 ¼ T nþ1 ;
>
>
sues, see [11]). The predicted elastic stresses/strains will be useful >
>
>
> e ð0Þ
as initial conditions for the integration of plastic equations, from >
> T nþ1 ¼ eT nþ1 ;
>
>
which a plastic correction will be made, by relaxation of the elastic >
< ð0Þ
6 0 ? ðaÞnþ1 ¼ ðaÞnþ1 ;
ð0Þ
stresses using an iterative procedure, in order to update the yield ðUf Þnþ1
>
>
function: the limit surface is reached as soon as the value of the >
> ð0Þ
>
> ðrÞnþ1 ¼ ðrÞnþ1 ;
yield function is equivalent to zero. The convergence is always >
>
>
>
reached because the stress trajectories follow a ‘‘steepest descend >
>
>
>
path’’, defined by the relaxation equations. >
>
> if NOT; starting of iterations and :
>
>
The relaxation procedure for stresses and internal variables is :
i¼0
done by an step-to-step iterative method. For each iteration, the
yield surface is linearized with respect to the last state-variables
and this is written as it follows:
(iv) Plastic correction:
@U
ðiÞ @U
ðiÞ ðiÞ
Uf ¼ UðiÞ f
: rfT
ðiþ1Þ ðiÞ
rfT
f
: vf
ðiþ1Þ
vf
ðiÞ
0 ðUf Þnþ1
f þ þ ð3:1Þ Dkf ¼
ðiÞ
ðiÞ
p ðiÞ ;
@r f
T
@v p ðiÞ
@ Uf @U @ Uf @U
f G DT f
f þ @v c @vf
@ rT @ rT nþ1
Introducing Eq. (2.28) in both Eqs. (2.9) and (2.10), we will have: nþ1 nþ1 nþ1
p !
p ðiÞ
!ðiÞ
@U f @ Uf @ Upf
v_ ¼ c a_ ¼ c k_ f ð3:2Þ ðr
ðiþ1Þ
Þnþ1 ¼ ðr
ðiÞ
Þnþ1 G DT Dkf c Dkf ;
@v @ rfT @v
p
@U f !ðiÞ
r_ fT ¼ G DT fT ¼ G DT k_ f f
ð3:3Þ @ Upf
@r T
ðiþ1Þ
ðaÞnþ1 ¼ ðaÞnþ1 þ Dkf
ðiÞ
@v
The last ones can be discretized in the next way:
34 N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45
(v) Convergence requirement: T Þ
G 1 @gð
@ T
fT
8 H¼ @ Up ð3:8Þ
>
> if YES; ending of cycle and : @ Uf f
>
> @ rT @ rT
>
> ðiþ1Þ
ðrÞnþ1 ¼ ðrÞnþ1 ; 1 þ G DT @Up
>
>
>
>
@ Uf
@ 2 ðqWÞ
f
>
> ðiþ1Þ
ðaÞ ¼ ðaÞnþ1 ; @v @ a2 @v
>
> nþ1
>< e ¼ e ððrÞ ;ðaÞ Þ; In this equation, the partial derivative corresponds to:
ðiþ1Þ
ð0Þ
T T nþ1 nþ1
ðUf Þnþ1
6
TOLðUf Þnþ1
? nþ1 e
>
> f
@gðT Þ
>
> T nþ1 ¼ ð T Þ nþ1 T nþ1 ; ¼
@DT @Y DT
>
>
>
> @ T @Y DT @ T
>
>
>
> 0
>
> if NOT; keep iterating fromðivÞ : f g h f 0 g h f g0 h
>
: ¼ 2
G T ð3:9Þ
i¼iþ1 h
where f,g and h correspond to the known functions:
The procedure is as follows: sffiffiffiffiffiffiffiffiffiffi
Y T1
f ¼
1. Initially, reading of bonding data, that is, geometrical and Y DT
material parameters. ( "rffiffiffiffi ! )
2. Predictive calculations of elastic strains and stresses. 2 qffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffi BD
g ¼ exp ADT1 ð Y DT Y T 1 T1
3. Evaluation of damage yield limit. G
4. Non-linear calculations of damage and cracking friction BD
h ¼ 1 þ ADT2 hY DT Y T 1 iþ
T2
contribution:
(a) In normal direction: explicit calculation of the damage
scalar variable. 3.3. Final implementation in a standard finite element code
(b) In tangential direction:
(i) definition of the type of damage by evaluating two The proposed formulation was implemented in two standard fi-
limit surfaces for damage; nite element codes, FEAP [13] and Code_Aster [14], taking care of
(ii) explicit calculation of the damage variable; respecting the global architecture of each one of them. In both
(iii) implicit calculation of the hysteresis effects pro- codes, the interface element and the non linear constitutive func-
duced by the sliding crack friction; tion were implemented separately.
In the case of the non-width interface element, this one was
5. Tangent matrix calculation: introduced as a new type of finite element, according to the
r_ T requirements of each code. Nevertheless, because the interface ele-
H¼ ment has no width (that means two nodes in the same coordi-
_ T
nates), in FEAP it was necessary to modify previously a global
6. Construction of the elementary matrix and the residual for controller which verifies the coherence of nodes coordinates,
the interface element: avoiding to erase immediately the interface element; in Code_Aster
Z this was not necessary because the program is based in objects in-
eðiÞ
BT H nþ1 B dXe
ðiÞ
K nþ1 ¼ and stead of logical tasks. About the implementation of the constitutive
Xe
eðiÞ e;ext e;intðiÞ
function for bonding, in both codes it was simply introduced as a
Rnþ1 ¼ f nþ1 f nþ1 new independent material library that is called for the element.
7. Assembling of the global matrix: In other words, the bonding model can be used in a standard finite
element code even if this one does not allow to manipulate a new
Nelem n o Nelem n o
eðiÞ ðiÞ e;ext e;intðiÞ interface element, being possible to use a typical four-nodes quad-
A K nþ1 Dunþ1 ¼ A f nþ1 f nþ1
e¼1 e¼1 rilateral element with a very small width.
Finally, the system matrix resolution is done by a Newton–Raph-
son iterative method, without affecting the global architecture of 4. Calibration of bond model parameters
any finite element code.
In general, the bond model is controlled by one geometrical
3.2.3. Consistent tangent matrix parameter (hpen) and 14 material parameters, distributed in the
In order to ensure the robustness and efficiency of the proposed next way:
model for the global analysis of massive structures, it is necessary
to calculate the consistent tangent matrix, which can be deter- Two general elastic parameters, Eb and mb;
mined from the next expression: nine parameters for behavior on the tangential direction,
three of them associated to the frictional effects and con-
r_ T ¼ H _ T ð3:7Þ
finement; and
The implementation of a consistent tangent matrix for the local for- three parameters for behavior on the normal direction.
mulation of the bond-slip model is needed for quadratic conver-
gence in Newton iterations. Nevertheless, it must be remarked These parameters -or in other words, the analytical expressions
that if the bond-slip model is combined with a nonlocal model for that allow to calculate them - have been developed from the
concrete, the global consistent tangent stiffness matrix cannot be numerical simulation of some experimental tests described in the
assembled from elementary contributions solely (see [12]). By next paragraph.
doing some analytical calculations, it is possible to deduce the
expression for the tangent matrix by using the consistency condi- 4.1. The experimental tests of reference
tion as well the respective evolution laws:
In order to calibrate these parameters, we have studied and
compared several pull-out experimental test that have been re-
N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45 35
Force (kN)
sponse. Among these tests, we have chosen as experimental refer- 8
ence the works of Eligehausen et al. [15] and Laborderie et al. [16] 6
based on the next requirements:
4
The rigorousness of the experimental procedures. 2
The quantity and variety of results.
The exhaustive study and vulgarization of data and 0
0 2 4 6 8 10 12 14
parameters. Displacement (mm)
300
lated rib area asR, using the expression (4.1):
with a residual variation close to 5%. Hence, assuming that con- 4.3.3. The exponential parameter of brittleness damage transition BDT1
crete will be the first material in being damaged, for bonding we This parameter affects the degree of curvature of the behavior
propose to use the same value of G estimated for the concrete. Nev- relationship, immediately in the transition between the elastic
ertheless, it is always possible to combine higher values for G and slope and the non linear region. Its value goes from 0.1 to 0.5,
hpen but we recommend not to adopt for bonding a value of G and it is recommended to adopt a value of 0.3 for ordinary calcula-
greater than the corresponding value of G for steel. tions. In Fig. 4 it could be observed that any reduction of this value
produces a drastic curvature as well as an increasing of the shear
stress.
4.3. Damage parameters on the tangential direction
The behavior of bond in the tangential direction can be sepa- 4.3.4. The post-coalescence yield strain T 2
rated in three major regions: elastic or perfect bond region, transi- The third region of the tangential bond behavior starts as soon
tion region (from small to large sliding), and post-coalescence as the sliding reaches the post-coalescence yield strain. In order to
region. The elastic region is controlled by the modulus of Coulomb calculate the value of this parameter, which is very close to the
G and limited by the perfect bond yield strain T 1 . The transition re- peak of the bond resistance (see Fig. 5(a)), we propose the next
gion is controlled by two parameters, ADT1 and BDT1 , while the third expression (4.4) that is based on a sigmoid function:
region is controlled by two similar parameters, ADT2 and BDT2 . the
boundary between these regions is defined by the post-coales- 1 ðADT1 Þn
cence yield strain T 2 . At the same time, the confinement and crack T ¼ 2 1 6 1:0 ð4:4Þ
2
hpen ðC þ ADT1 Þn
friction effects are acting in both damage regions. ref
hpen
ref
In the last expression, hpen must be expressed in mm and hpen is
4.3.1. The perfect bond yield strain T 1
a reference value of 1 mm. In which concerns to the sigmoid func-
In order to define the perfect bond limit,we consider that shear
tion, this one is necessary for adjusting the kinematics effect of ADT1
damage starts as soon as a perfect bond yield strain is reached.
on the sliding, which commonly diminishes as bond stiffness
Based on it, we propose to adopt the typical yield strain of concrete
grows. Adopting the values of C = 9 and n = 4 (but any other set
in traction, that means, a value between 1.e4 and 1.5e3, which
of values can be proposed), the last form of expression (4.4) will
corresponds to a shear stress between 0.5 and 4.0 MPa for perfect
be:
bond.
" #
1 ðADT1 Þ4
4.3.2. The parameter of brittleness damage transition ADT1
T ¼ 2 1 6 1:0 ð4:5Þ
ð9 þ ADT1 Þ4
2
hpen
In the second region, the damage evolution law controls the href
pen
shear stress as a function of the sliding and it is expressed in terms
of strains and its construction depends on the elastic slope defined The choice of the post-coalescence yield strain is very important
by the shear linear behavior. For our proposal, this parameter is a because it defines the level of the fragility in the bond response
key-point because the evolution of the damage must verify some (see Fig. 5(b)), which is directly associated to the concrete com-
particular conditions indicated by different researchers, for pression strength by the parameter ADT1 .
example:
The bonding resistance is directly proportional to the con- 4.3.5. The parameter of post-coalescence damage ADT2
crete compression strength. On the other hand, it should In this model, the tangential damage is controlled by two differ-
be taken into account that if the concrete strength becomes ent evolution laws expressed by just one classic scalar variable in
higher, the stiffness too, and by consequence the bond order to guarantee the damage continuity. Nevertheless, even if
might fail due to the associated fragile rupture. the parameters of both laws are independent, a bad selection of
The steel bar stiffness is related to its diameter and its them might produce serious effects on the shape of the local
quantity of ribs, therefore this condition must raise the stress–strain curve. as it is shown in Fig. 6(a). Being very difficult
bonding resistance. to build a simple general expression for this parameter, we propose
The relationship between the elastic modulus of both to adopt a value between 1.0e3 and 9.0e2 MPa1 based on the ac-
materials (steel and concrete) must define directly the quired experience. From Fig. 6(a), it can be derived that as soon as
kinematics of bonding. ADT2 grows, the maximal stress as well as the residual stress tend to
diminish.
Based on our numerical experiences, we have observed that the
value of this parameter should be placed between 1 et 5, and it 4.3.6. The exponential parameter of post-coalescence damage BDT2
must be adjusted according to the experimental test adopted as This parameter is necessary for controlling not only the growth
reference. In spite of this, we propose a conservative expression of the bond resistance or the shape of the behavior curve before
(4.3), that allows adopting an initial value, which depends on the and after the peak, but also the kinematics of the response. The
own characteristics of the concerned materials: parameter BDT2 determines the sliding associated to the maximal
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi shear stress as well as the amplitude of the curve in the behavior
1 f 0c Esteel peak. As it can be seen in the Fig. 6(b), this parameter is directly re-
ADT1 ¼ ð4:3Þ
ð1 þ asR Þ 30 Econcrete lated to the parameter ADT2 , and they should be calibrated at the
same time. Even though, it is important to note that the value of
In the last expression, f0 c corresponds to the concrete compression BDT2 varies inversely to the sliding amplitude at the peak, or in
strength; Esteel corresponds to the elastic modulus of steel; and Econ- other words, a value of 0.8 for BDT2 produces larger sliding than a
crete is the elastic modulus of concrete; all of them expressed in MPa. value of 1.2, for example. Based on our experience, it is recom-
The Fig. 3 gives a graphic comparison of the influence of this param- mended to use a value between 0.8 and 1.1 in order to produce a
eter, showing that as ADT1 raises, the maximal shear stress grows. coherent behavior curve.
N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45 37
4.4. Parameters for frictional effects and confinement bond is reached, there is not any other contribution to the resis-
tance associated to the sliding, only the friction between the sur-
4.4.1. The parameters c and a for the crack friction effects face of the concrete linked to the steel bar and the surface of the
One of the most important contributions of this model is the concrete body.
capacity of taking into account the crack friction effects in the bond
behavior: for monotonic loading, this contribution is appreciated
as a growing of the bond shear resistance; for cyclic loading, the
width and shape of the hysteresis loops are determined for the va- 4.4.2. The lateral pressure parameter c
lue of both parameters. Because the experimental tests taken as Normally, the values of the maximal bond shear resistance and
reference have not included the study of these cyclic effects, we the corresponding sliding increase as the confinement or lateral
have chosen for c a value lower than 10 MPa and a maximal value pressure raises around the steel bar. In this model, this influence
of 1.0 MPa1 for a. From Fig. 7 it is possible to appreciate that a is taken into account by the parameter c. For calibration, we have
reduction of the value of a produces an increasing of dissipation, done numerical simulations with confinement values of 0, 5, 10
shear resistance and the residual pseudo-plastic strain. The impor- and 15 MPa considering a value of 1.0 for the parameter c (see
tance of this lies on the fact that once the maximal resistance of Fig. 8).
38 N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45
In order to model the bond behavior on the normal direction, Fig. 6. Influence of parameters of post-coalescence damage.
each bond mechanism (closing or opening) has been represented
in an independent way. In the case of compression (closing), we
just allow a non-limited small deformation controlled by a linear 4.5.2. The parameter for traction damage ADN
elastic behavior, while in the case of traction (opening), a damage The parameter ADN controls the peak magnitude and shape of
criteria has been adopted for simulating decohesion between the the post-peak phase. A minimal value of 10.0 MPa1 produces a
steel bar and the concrete. Because there are not experimental coherent damage state similar to concrete damage. Nevertheless,
studies focused in this phenomenon, we propose hypothetical val- if it is necessary to model a much more fragile bond behavior, it
ues for the three parameters that give numerical stability and is enough by increasing this value, as it is shown in Fig. 9(a).
coherence to the model.
4.5.1. The elastic limit strain yN 4.5.3. The exponential parameter for traction damage BDN
Similar to the elastic limit strain in the tangential direction, we Combined with the previous parameter, BDN controls the magni-
have considered that decohesion should start as soon as as a con- tude and the shape of the normal behavior curve in the post-peak
crete limit strain is reached, assuming a typical value between phase. An non-dimensional value between 1.0 and 1.2 is conve-
1.0e4 and 1.0e3. nient for the model (see Fig. 9(b)).
N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45 39
ADN = 5e+1
ADN = 1e+2
15 15 1DN = 5e+2
5 5
6
Values for a lateral pressure of 2.5 MPa
4
2.5 10
4
4
2 10
2
4
1.5 10
Force (N)
0
4
0 0 .0 0 5 0.01 0.015
1 10 Normal Strain
c = 0.0
c = 0.1
(b) Identification of parameter BDN .
5000 c = 0.5
c = 1.0 Fig. 9. Influence of parameters on normal direction.
0
0 0.5 1 1.5 2 2.5 3
Displacement (mm) concrete and steel (see Table 1). By the way, we simulate the two
principal mechanisms of failure in bonding: a pure decohesion be-
Fig. 8. Combined influence of parameter c related to the lateral pressure. tween steel and concrete (see Fig. 10(b)), and the tangential sliding
or bond-slip behavior (see Fig. 10(c)). For the first mechanism, the
5. Numerical examples boundary conditions are assigned in the next way: a vertical trac-
tion displacement is applied in the nodes of the concrete element
Briefly, in the next paragraphs we present some numerical while the displacements in the bottom of the steel element are
examples in which the proposed bond model has been used either fully-restrained ( see Table 2). As it is shown in Fig. 11(a) and (b),
alone or combined with other non linear material behavior models. the normal resistance of the bond reaches the peak and immedi-
The examples are grouped in three sets: In the first one, a simple ately it descends until a value of zero. For compression, a simple lin-
validation test for a one interface element is done in order to verify ear elastic law is considered. Regarding to the second mechanism,
the mathematical coherence of our proposal. The aim of the second the only variation is the direction of the displacement over the con-
set is to validate the bond model by reproducing the experimental crete element, which is applied horizontally by little increments,
response of different pull-out tests extracted from literature (see and developing large displacements, activating the non linear tan-
[16,19]). In the third set, the numerical simulations include the gential behavior of the bond. This is shown in Fig. 11(c) and (d),
use of three non linear models for the material behavior of the steel where it is possible to observe the three zones of the bond behavior:
bar, concrete and bond (see the experimentaltest tie tests [20,21]). a region of perfect bond, followed by a region of damage with tran-
For a wider description of each experimental study, it is recom- sition from small to large displacements and finally, a region of
mended to consult the corresponding reference. post-coalescence damage and residual resistance produced by fric-
tion. In Table 3, the numerical values obtained for Tangential and
Normal behavior are listed respectively.
5.1. Test 1: A simple validation test for sliding and decohesion
For a simple validation of the interface model as well as for the 5.2. Validation based on experimental tests
bond thermodynamical model implementation, in FEAP we analyze
a simple mesh of three finite elements (see Fig. 10-(a)) based on a 5.2.1. Test 2: The pull-out validation test of Laborderie and Pijaudier-
fictitious behavior for each material: concrete, steel and bond. In or- Cabot
der to focus to the non linear behavior of bonding (see parameters In this numerical simulation we analyze a simple pull-out
in Table 1), we assigned a typical elastic model to both materials: experimental test carried out by La Borderie and Pijaudier-Cabot
40 N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45
Table 2
Material parameters for all tests (Continuation).
Table 3
any particular interface element on the global results is reduced
Tangential and normal stress–strain relationship for test 1.
significantly.
Tangential behavior Normal behavior
Strain Stress (MPa) Strain (e4) Stress (MPa) 5.3. Combined effect of three non linear models (steel–concrete-bond)
0.00 0.00 0.00 0.00 on tie tests
0.25 10.20 1.00 1.53
0.50 12.4 2.00 2.60
Tie tests are widely used to study the effort transfer between
0.75 14.50 3.00 3.90
0.95 16.20 4.00 5.28
steel bars and the concrete body, focusing particularly on the
1.00 16.20 5.00 6.58 crack pattern distribution on concrete, as well as the changes
1.25 15.70 6.00 7.96 on the force - displacement response at each time that a crack
1.50 14.90 7.00 4.37 appears on the tie. The effects of bonding deterioration are
1.75 14.00 8.00 0.80
clearly identified in this kind of test. In the next cases, concrete
2.00 13.10 – –
was modeled using the damage model proposed for Mazars [22],
while the steel behavior is modeled using a classical elasto-plas-
tic model based on Mises criteria; bonding behavior is modeled
the ND3 pull-out test (diameter of 18 mm) while the correspond- with this proposal using non-width interface elements. By the
ing experimental and numerical results are compared in way, in order to verify the effectiveness of mixing three non lin-
Fig. 14(b), as well as in Table 5. Despite the parameter sensitivity, ear models and before doing any realistic prediction, we simu-
a very good approximation between experimental and numerical lated a long concrete tie (length of 150 cm) in which a
results is observed in 14(b), and this could be explained as fol- maximal displacement of 3 mm was imposed, finding that the
lows: because the loading is applied monotonically, the damage model with interface elements included is able to reproduce
in each interface element occurs gradually and not at the the different jumps in force that should appear as soon as a
same time step and, as a consequence, the local influence of crack arises in concrete - in contrast with the model without
42 N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45
Table 4
Force–displacement data for test 2.
Displacement (mm) Force (kN) wo/conf Force (kN) w/conf 5 MPa Force (kN) w/conf 10 MPa Force (kN) w/conf5 MPa
0.0 0.00 0.00 0.00 0.00
0.1 2.97 2.98 2.95 2.92
0.2 5.71 5.99 5.96 5.93
0.4 10.10 11.10 11.90 12.00
0.6 12.70 13.80 15.30 16.80
0.8 14.00 15.10 16.60 18.50
1.0 14.30 15.40 17.00 18.80
1.2 14.10 15.20 16.80 18.70
1.4 13.70 14.80 16.40 18.30
1.6 13.10 14.30 15.90 17.80
1.8 12.60 13.70 15.30 17.20
2.0 12.00 13.10 14.80 16.70
2.5 10.70 11.90 13.60 15.50
3.0 9.60 10.80 12.60 14.50
3.5 8.70 10.00 11.70 13.60
4.0 8.00 9.30 11.10 13.00
Fig. 13. Comparison between experimental and numerical results of the structural response of the Laborderie/Pijaudier-Cabot pull-out test.
N. Domínguez, A. Ibrahimbegovic / Computers and Structures 106–107 (2012) 29–45 43
STRESS 4
-3.48E+01
-2.50E+01
-2.25E+01
-2.00E+01
-1.75E+01
-1.50E+01
-1.25E+01
-1.00E+01
-7.50E+00
-5.00E+00
-2.50E+00
0.00E+00
3.02E+00
Time = 1.60E+00
(a) Damage distribution for an ap-
(a) Shear stress distribution near to the resistance peak. plied displacement of 3 mm, with-
out and with interface elements.
140
40
120
35
100
Force (KN)
80 30
60 25
Force (KN)
40 Experimental
Numerical simulation 20
20
15
0
0 0.5 1 1.5 2 2.5 3 3.5 4
10
Displacement (mm)
Without interface elements
5 With interface elements
(b) Comparison between experimental and numerical force-
displacement relationship. 0
Fig. 14. Simulation of the ND3 specimen subjected to pull-out loading according to
0 0.5 1 1.5 2 2.5 3
Bamonte [19]. Displacement (mm)
(b) Comparison of numerical results with and without
Table 5
interfaces elements.
Force–displacement data for test 3.
Fig. 15. Numerical study of an hypothetic tie of 150 mm length.
Displacement (mm) Force–experimental (kN) Force–numerical (kN)
0.00 0.00 0.00
0.06 39.30 39.30
0.11 49.60 47.30
0.60 105.00 109.00 30
1.10 120.90 134.50
1.64 133.00 134.50
2.10 130.70 126.50 25
3.10 109.30 104.40
4.00 93.90 90.10
20
Force (KN)
[21] A. Daoud, Étude expérimentale de la liaison entre l’acier et le béton auto- [23] Dominguez N, Brancherie D, Davenne L, Ibrahimbegovic A. Prediction of crack
plaçant – contribution à la modélisation numérique de l’interface, Thèse de pattern distribution in reinforced concrete by coupling a strong discontinuity
l’INSA de Toulouse, 2003. model of concrete cracking and a bond-slip of reinforcement model.
[22] Mazars J. A description of micro- and macroscale damage of concrete Engineering Computations 2005;22(5/6):558–82.
structures. Journal of Engineering Fracture Mechanics 1986;25(5/6):729–37.