Composites: Part A 56 (2014) 5163

Contents lists available at ScienceDirect

Composites: Part A
journal homepage: www.elsevier.com/locate/compositesa

Efcient progressive damage modeling of hybrid composite/titanium

bolted joints
G. Kolks, K.I. Tserpes
Laboratory of Technology and Strength of Materials, Dept. of Mechanical Engineering & Aeronautics, University of Patras, Patras 26 500, Greece

a r t i c l e i n f o a b s t r a c t

Article history: In an attempt to gain better understanding and predictability of the mechanical behavior of titanium-
Received 20 February 2013 lamella reinforced CFRP bolted joints, an efcient progressive damage model was developed in this work.
Received in revised form 31 July 2013 The model was used to represent bearing effects in a double-lap composite bolted joint subjected to ten-
Accepted 19 September 2013
sile loading. The progressive damage model relies on stress-based failure criteria accounting for three
Available online 2 October 2013
orthogonal fracture planes and a continuous degradation method depending on stress-state and stress
history. It is shown that the model, which for material specication uses available elastic moduli and
Keywords:
strength values, can be rened to reasonable accuracy with regard to experiments through a parameter
A. Polymermatrix composites (PMCs)
A. Hybrid
identication process. For pure CFRP bolted joints, the model predicts accurately both bolt strength and
B. Strength total displacement at nal failure. The predicted strength enhancement, achieved by replacing CFRP lay-
C. Finite element analysis (FEA) ers with titanium sheets, is in good agreement with experimental data obtained from the literature.
2013 Elsevier Ltd. All rights reserved.

1. Introduction straps [3] have been successfully tested in applications with

Ever since the introduction of carbon ber-reinforced plastics
(CFRPs) into aerospace design, signicant engineering efforts have
been employed to adopt existing metal-related designs to the new
requirements, induced particularly by the materials anisotropy
and brittleness. In doing so, the optimization of structural joints
has been a substantial eld of interest [1,2]. However, ber-com-
patible joint design often proved unsatisfactory in regard to
weight, required space or notably the ease of assembly and disas-
sembly, having crucial effects on the economical feasibility of a de-
sign-for-functionality approach.
Integral design featuring adhesive bonding can incorporate dif-
ferent functionalities in one part, thus being a benecial applica-
tion of CFRPs. In turn, it often evokes conicts with product
requirements regarding handling, assembly and maintenance. To
overcome these conicts, new non-permanent bonding techniques
are obligatory, since simple bolted laminate joints featuring drilled
holes (Fig. 1a) were widely considered not to be ber-compatible
due to notching and ber-rupture, and thus would not allow for
highly efcient primary structure joints. Bolted joints can generally
be distinguished in double-lap and single-lap congurations, the
latter inducing bending and extensive bearing stress peaks along
the outer laminate regions even under purely tensile load, due to
off-axis pulling. As opposed to the aforementioned inappropriate
designs with regard to CFRP, anchor-loaded ber-loop connector
Corresponding author. Tel.: +30 2610 969498; fax: +30 2610 997190.
E-mail address: kit2005@mech.upatras.gr (K.I. Tserpes).
mainly tensile loading, yet the required geometries and space devi-
ated signicantly from metal practice.
In order to make use of the advantages associated with bolted
joints, different approaches to reducing bearing stress have been
followed. Most common in design practice is the local reinforce-
ment of the base laminate with additional layers of CFRP. This,
though, does not overcome the materials brittle characteristic
and furthermore entails secondary bending as well as a consider-
able increase in total weight [4]. Sophisticated metal inserts are
supposed to combine the structural advantages of CFRPs and
the toughness of metal joints. Previous work has proved that
bonded metallic inserts can increase the joints efciency, though
it is necessary to carefully optimize load paths and bonding
strength [5].
In many applications it is, however, desired to avoid geometri-
cal transitions between the joint laps and the base laminate, be it
for space-restriction reasons or structural effects such as secondary
bending or notching. Kolesnikov et al. [6] have performed a lami-
nate reinforcement by locally substituting CFRP plies within the
laminate by equally thick titanium sheets as schematically shown
in Fig. 1b. In a multidirectional laminate, those plies contributing
least to the primary load direction, i.e. e.g. 90-layers, are preferen-
tially replaced and bonded by the matrix resin of the adjacent
remaining CFRP layers, resulting in a hybrid CFRP-titanium lami-
nate With reference to a laminate build-up with [50% 0; 40%
45; 10% 90], improvement factors in joint strength of 1.78 and
1.91 were achieved for single-row and three-row layout congura-
tions, respectively, in case of uniaxial loading.

1359-835X/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compositesa.2013.09.011
52 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163
a entailed effects (e.g. ber pull-out) in different strain regimes for
CFRP
Titanium
b The model involves a PDM approach based on constitutive
Fig. 1. Schematic sectional view of: (a) Single-lap bolted CFRP joint; (b) Double-lap
bolted hybrid composite joint with four composite layers locally substituted by
titanium plies.
In order to entirely utilize the proposed technique, optimization
for the most efcient layer build-up as regards strength enhance-
ment has to be performed. Thus, numerical models have to be
developed, capable of detecting the damage progress. A variety of
past works have established progressive damage models (PDMs)
based on nite element (FE) formulation for pure CFRP bolted
joints, but so far little effort has been done on the simulation of hy-
brid laminated materials.
Existing PDM approaches can be classied regarding the failure
criteria and degradation strategies adopted. Employing stress-
based criteria and directly coupling the failure variables to stiffness
degradation using material strength data is a self-evident method
that can exhibit accurate predictions of failure onset with compar-
atively low computational costs. Tserpes et al. [7] developed a 3D
model of a single-lap single-bolt joint of a CFRP-laminate that
was subjected to tensile loading. Hashin-type failure criteria [8]
were used for the computation of local failure variables, partly
due to their ease of implementation into FE applications. Eventu-
ally, the model produced overly conservative strength predictions
in consequence of the shear-stress contribution to ber failure in
the failure criteria used, as stated in [7]. In [9], the model was fur-
ther rened aiming for a model set-up that would meet the partic-
ular demands for simulating textile and cellular materials. Thereby,
the Hashin criteria were transposed to a strain based formulation,
which enabled them to account for strain softening. The constant
and sudden degradation method was replaced with an exponential
course of the damage variables, as well as a rened failure mode
interference treatment was introduced.
As opposed to the aforementioned class of PDM concepts based
on stress or strain state variables and strength data, other ap-
proaches involve the use of energy-based degradation strategies.
There, the dissipated energy from evolving cracks is evaluated con-
sidering the fracture toughness for each failure mode and the deg-
radation is carried out in compliance with the energy equilibrium.
Models making use of this technique (e.g. [10,11]) claim a decrease
in mesh dependency on the models characteristics. However, the
application of fracture toughness values oftentimes implies exces-
sive material testing, given that they are rarely available in the lit-
erature. Further renement of the energy-based method has been
carried out in [11,12]. A short introduction into the theory of com-
plimentary free energy, energy dissipation and their interference
with the tensor components in the constitutive equations is given
in the accompanying papers. The authors introduced damage acti-
vation functions based on the failure criteria LaRC03 and LaRC04
[13,14], including the effects of ber-kinking under longitudinal
compression and transverse compressive cracking on effective
planes in distinctive angles, also known as C-mode failure. The
damage activation functions exhibit hysteretic characteristics, en-
abling them to consider crack-closure effects on load reversal. In
order to represent the damage process accurately, the damage evo-
lution laws distinguish primary failure onset (e.g. ber failure) and
every failure mode.
The scope of this work is to develop a parametric model of a
symmetric double-lap hybrid CFRP/Ti composite bolted joint that
can be used to study the optimal reinforcement strategy for a given
laminate design. The emphasis is being put on an accurate repro-
duction of the characteristic of nominal bearing stress against lon-
gitudinal displacement under tensile loading. Moreover, the model
shall allow for analyses on the damage propagation within the
specimen and the global failure mode induced.
orthotropic composite strength data and a basic isotropic material
model that incorporates plasticity for the composite fraction and
the titanium fraction respectively. ANSYS implicit FE-formulation
is used for analysis computation and while the model is built up
of basic geometrical shapes, it is well suited for implementation
in ANSYS Parametric Design Language (APDL) implicit FE-code,
which allows for a very exible and fast model set-up.
2. Model topology
For a better overview and classication of the terms dened in
the subsequent sections, a brief introduction into the global model
topology is given here. It is schematically shown in Fig. 2.
The developed parametric model can be subdivided into two
substantial units, the static fraction and the dynamic fraction, with
their particular sets of parameters, the modeling parameters and
the identication parameters respectively. The FE entities used to
model the bolted joint are herein considered the static fraction,
since they are created in advance to the sequential PDM procedure
and not altered in either shape or structure throughout the process.
The PDM process is the dynamic fraction that accesses the static
model and manipulates its properties iteratively.
The model is implemented as a mixed physical and parametri-
cal formulation, often referred to as gray-box model. Therein, the
major behavior is specied by physically founded equations. A
mathematical model, relying on a set of parameters, is applied to
those parts of the model that are yet mostly unknown or where
the complexity of physical sub-modeling exceeds reasonable lim-
its. These parameters are the aforementioned identication param-
eters, which are gained in a process called parameter identication.
As a main component in the modeling process, during the param-
eter identication, the model output is iteratively matched with
measurement data, i.e. mathematically it is considered a curve t-
ting or optimization problem.
3. Stress-based PDM approach
In this section, the PDM strategy developed during this work as
a key component for the integral model will be described. It
Manipulate
properties
Identification
Dynamic fraction Static fraction
Modeling
parameters parameters
Load step
result
Fig. 2. Structure of the global progressive damage model.
 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163 53

evolves from a foundation consisting of APDL-code developed in Delamination in compression (DC):

[7,15]. 8 r
 2  2
The considered approach relies on stress-state-determined fail- < rz sxz syz
dDC
j S?;C  qz j Sjj?  qxz S??  qyz : rz < 0 4
ure criteria and degradation concept. On behalf of the minimiza- :
tion of computational cost, the FE topology of the model remains 0 : rz P 0
static throughout the entire simulation process. Therefore, damage Fiber tensile failure (FTF):
is represented by altering the constitutive stiffness data according 8 r
>  
to the detected failures only, rather than by inserting structural >
> r2x r2y r2z  rx ry  rx rz  ry rz 3 s2xy s2xz s2yz
<
entities (e.g. crack bands, imperfect bonding elements, etc.) into dFTF : rx > 0
the FE-model. The material characteristics are represented by >
> Sjj;T
>
:
means of continuum mechanics, which implicates that the accu- 0 : rx 6 0
rate reproduction of behavior is limited to a scale that its magni- 5
tude are larger than the ber diameter, where averaging bears
Fiber compressive failure (FCF):
negligible decits. However, as damage in ber reinforced plastics 8 r
>  
needs to be treated on micro-scale, the failure criteria and degrada- >
>
< r2x r2y r2z  rx ry  rx rz  ry rz 3 s2xy s2xz s2yz
tion strategy have to be carefully chosen in order to compensate dFCF : rx < 0
for the structural shortcomings. As a result, the failure criteria rely >
> Sjj;C
>
:
on macroscopic strength data of unidirectional CFRP specimens. 0 : rx P 0
6
3.1. Failure criteria
The above equations include material strength values
explicated in Table 1 and softening compensation ratios dened as
To any portion of a unidirectional CFRP laminate a local orthog-
onal coordinate system can be assigned. It is convenient to declare Einit
i
the x-axis to represent the ber direction, z-axis the laminate-nor- qi ; i 2 fx; y; zg 7
Ei
mal and the y-axis the in-plane transverse direction. The failure
criteria detect failure in planes that are normal to each base-axis Ginit
i
of this coordinate system, in both tensile and compressive mode. qij ; i 2 fx; y; zg; j 2 fx; y; zg; ij 8
Gij
The predecessor version [7] of the presented model made use of
Hashin-type failure criteria. As pointed out there, the consideration These are the initial-to-current stiffness ratios of the currently
of shear stresses in the criteria terms representing ber failure considered element introduced in order to alleviate the impact of
leads to an overestimation of damage in regions that are exposed softening effects due to previous degradation. Especially in cases
to high shear rates. Considerable shear stresses occur near the con- that imply low stiffness levels and little degradation effects, such
tact zone between bolt and CFRP-specimen as well as in regions of as degradation due to matrix failure, softening effects lead to fur-
high stiffness gradients resulting from previous degradation. ther impeding of damage progress, depending on the degradation
Therefore, a different set of failure criteria with the aim of mitigat- strategy involved. The presented compensation ratios qi and qij in-
ing this shortcoming is proposed in the present work, formulated crease the respective terms of the damage variable reciprocally to
as the respective damage variables di. Damage is assumed to occur the related stiffness degradation, thus leading to a quasi-strain
for di > 1. The failure criteria for the different failure modes are de- based failure criteria formulation, making the model more prone
scribed below. to fast damage propagation once failures occurred. Actual element
Matrix tensile failure (MTF): stiffnesses are not immediately affected by this technique within
8 r one particular load step. The criterion for bermatrix shear-out,
<  ry 2 
sxy
2 
syz
2
originally included in [7], is omitted in the taken approach for
dMTF S ?;T
 q y S k?
 q xy S ??
 q yz : ry > 0 1
: the benet of minimizing the number of identication parameters
0 : ry 6 0 of the model, since it strongly correlates with ber compressive
Matrix compressive failure (MCF): failure and shear-dominated criteria. The criteria for failure in
8 r the xz-plane, i.e. through-thickness matrix cracking (Eqs. (1)
<  ry 2 
sxy
2 
syz
2
and (2)), share their characteristic with the Hashin criteria. All
dMCF S?;C
 qy Sjj?  qxy S??  qyz : ry < 0 2
: stress-terms with y-component are divided by their corresponding
0 : ry P 0 strengths and superposed in the second power. For failure in the x
y-plane, called delamination within the context of this PDM ap-
Delamination in tension (DT):
proach (Eqs. (3) and (4)) the criteria are dened in a similar way
8 r
 2  2
< syz as the aforementioned matrix cracking criterion. Here, the involved
j Sr?;Tz  qz j Ssjj? xz
 qxz S??  qyz : rz > 0 stress-terms all feature a z-component along with the correspond-
dDT 3
:
0 : rz 6 0 ing strengths. In order to increase the relative inuence of the nor-
mal stress in ply-orthogonal direction, the power of this term in
the superposition is reduced to the order one. The criteria for
delamination and matrix failure are illustrated by means of failure
Table 1 envelopes in Fig. 3, in compressive and tensile principle stress do-
Denition of strength terms. main. Delamination is not modeled at the ply interfaces, where it
S||,T Tensile strength in ber direction actually occurs, but is rather modeled through stiffness degrada-
tion inside the solid layered in order to reduce model complexity
S||,C Compressive strength in ber direction
S\,T Tensile strength in ber-normal direction and computational cost. Interfacial damage is thus not accurately
S\,C Compressive strength in ber-normal direction represented by creating discontinuities in the material. However,
S||\ Shear strength of a unidirectional laminate ply in rectangular the macroscopic stiffness response of the global laminate achieved
orientation by this method is similar to the effect of physical delamination,
S\\ Shear strength on a ber-normal plane
since laminar shear and separation resistance are reduced. The cri-
54 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163

a b 3.2.1. Stiffness degradation

For the ease of implementation, the present model comprises a
linear degradation characteristic. Allocated to each failure criterion
i, there exists the set of the following degradation variables
8
>
> 1 : dmax
i 61
< max
wij di d  1 10
>
i
a
 r ij  1 1 : 1 < dmax
i 61a
>
:
r ij : 1 a < dmax
i

as illustrated in Fig. 4, in which j denotes the six principal normal

Fig. 3. Failure envelopes of the applied failure criteria: (a) Through-thickness/ and shear stiffness moduli. Notably, the superscript index max
matrix failure; (b) Delamination. indicates the use of maximum value of di that occurred throughout
the analysis for the element at hand. The length of the domain of
the descending slope is referred to as the damage transition width
teria for delamination and matrix failure are illustrated by means a. Stiffness residuals rij dene the minimal stiffness ratios possible.
of failure envelopes in Fig. 3 in compressive and tensile principle These serve as the main degradation parameters of the model.
stress domain. The degradation variables operate as stiffness factors and each
In signicant contrast to the Hashin criteria, the ber failure modulus is nally generated as a multiplicative superposition of
criteria of the herein presented set do not compare a subset of the degradation variables of each failure criterion:
stress components with their individual corresponding strength Y
values. This strategy entails an overestimation of ber failure if K j K init
j  wij ; with i 2 fMTF; MCF; DT; DC; FTF; FCFg
secondary stresses are involved [16] that states that a sufcient i 11
accuracy of the ber tensile criterion is achieved by use of the and j 2 fx; y; z; xy; yz; xzg
Maximum Stress criterion if moderate shear stresses occur. In
Herein, Kj represents the moduli:
cases that are accompanied by extreme transverse compressive

and shear stresses, however, these can cause ber failure in con- Ei : j 2 fx; y; zg
Kj 12
sideration of the ber as a macroscopic continuum. Neglecting Gj : j 2 fxy; yz; xzg
the morphologic anisotropy of carbon bers, here a von
Mises-type equivalent stress is employed and compared to the
macroscopic longitudinal laminate strengths in tensile and com- 3.2.2. Dealing with Poissons ratios
pressive mode (Eqs. (5) and (6)). As to the inuence of the Poissons ratio on the degradation, no
control of material behavior is intended by manipulating these val-
3.2. Degradation strategy ues. Nevertheless, as normal stiffness ratios can deviate largely
upon degradation, the positive denite property of the materials
Damage simulation of the brous material is performed by stressstrain matrix must be maintained in order to enable the
degrading the constitutive stiffness values of the material. A FE solver to formulate the problem. Adapting the Poissons ratios
stress-based continuous degradation strategy was used. In the deg- to this requirement is obligatory.
radation strategy each stiffness term generally depends on the cor- According to the ANSYS Product Documentation [17], the
responding initial stiffness and the set of maximum damage stressstrain matrix is positive denite, if the inequality
variables as follows: mxy myz myz mzy mxz mzx 2mxy myz mzx < 1 13
 
Ei f Einit
i ; fd
max
g 9 is met. This constraint stems from the restrictions on engineering
constants of orthotropic materials used to model the composite lay-
All data involved in the degradation of an element are intrinsic ers in ANSYS [17]. Due to the relation
to this particular element, i.e. state variables of surrounding ele-
ments are not accounted for. mij Ei
14
The continuous degradation method is chosen for two reasons. mji Ej
Primarily, it is intended to alleviate convergence difculties that stiffness degradation may lead to violation of Eq. (13) assuming
occur in regions of high stiffness gradients near contact denitions. constant major Poissons ratios. In order to meet Eq. (13), the
The resulting numerical instabilities cause the simulation to stop constraint
prematurely. By using continuous degradation rather than the sud-
den type, damage fronts spread more steadily. Secondly, the char- mij mji minit init
ij mji 15
acteristic of degradation may be used to consider matrix plasticity
and damping effects. is applied by dening
v
u
Eu  Einit
init t x y
mxy m 16
ij xy
Ey  Einit
x

1 v
u init
u
init tEy  Ez
myz mzy 17
Ez  Einit
rij y

s
init Ex  Einitz
1 1+ i mxz m zx 18
Ez  Einit
x

Fig. 4. Variation of degradation variable wij with respect to the damage variable di. in compatibility with Eq. (15).
 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163 55

3.3. PDM procedure

In contrast to the procedure suggested in [7], the herein pro-

posed PDM sequence limits the number of analysis iterations with-
in one displacement load step by a parameter nd. With regard to
accuracy of material behavior simulation, representing the dy-
namic fracturing process in quasi-static analyses implies an itera-
tive procedure of damage analyses within one load step until no
further damage occurs and a steady-state is reached. However, this
strategy entails extensive computational cost. Hence, a parameter
is introduced that limits the number of damage iterations without
a load increment. The simulation procedure stops as soon as either Fig. 6. Schematic of the bolted joint with nb = 2 and two layers locally replaced with
the maximum total number of iterations is reached or the simula- Ti layers.
tion fails to converge prematurely. A nal failure criterion can be
applied a posteriori, based on the course of bearing stress against
displacement and damage accumulation. An overview of the bolt
PDM procedure is given by means of a ow chart shown in Fig. 5. washer
pulling strap

4. FE (static) model

Prior to the application of the PDM modules described in Sec-

tion 3, the parametric FE model of the joint has to be created. This specimen
include the mesh, the denition of the shape and structure of the
model, the element types, the external constraints and loads, the
internal boundary conditions, i.e. contact elements and the mate-
bonded contact
rial models employed.

4.1. Meshing Fig. 7. FE mesh of the composite bolted joint.

The geometry of the static model consists of the rectangular at

hybrid specimen with nb holes in one equally spaced row, nb bolts, actually modeled entities are reduced to half, i.e. one half speci-
2nb washers and pulling straps with matching hole patterns on men, nb half bolts, nb washers and one pulling strap. A symmetry
either side of the structure. Since the modeling purpose is limited constraint completes the model as illustrated in Fig. 6 along with
to specimens that are symmetric to their mid-x0 y0 -plane, the the global coordinate system spanned by the base vectors

Analysis start

Read parameters

Create FE-entities

Set initial
displacement

Increase
displacement value

Apply YES
displacement load

NO Maximum number of
damage iterations per load Degradation
step exceeded?

Solve FE problem
YES
NO
Maximum number of NO Read Evaluate New maximum damage
iterations exceeded? stresses failure criteria variables > 1 detected?

YES
Analysis end

Fig. 5. Flow chart of the PDM procedure.

56 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163
 
~ex0~
ey0~
ez0 . Fig. 7 provides an overview over the mapped meshing
applied to all of the models parts.
In favor of control and reproducibility, mapped parametric
meshing is applied to the entire model. In both pulling strap and
specimen, square shaped areas surrounding the bolt holes, in
which the transition from polar to rectangular mesh is conducted,
were dened. Washers and bolts were entirely meshed in polar
direction resulting in wedge-shaped elements along the bolt sym-
metry axes.
Meshing of the pulling strap is carried out creating SOLID46 ele-
ments [17]. These are brick-shaped 8-node element with three
translational degrees of freedom each that allow for a representa-
tion of layered solid structure within one element. Thus, it is a con-
venient way of modeling the global stiffness behavior of plane
layered ber composites. Compared to a layer-wise modeling with
solid elements, the featured integration in thickness-direction dur-
ing the composition of the element stiffness matrix reduces the re-
quired memory and computing costs signicantly. On the contrary,
this element type does not allow layers with bilinear plasticity def-
inition, which is the required material model for titanium layers.
The specimen region around the bolt hole, which will face deg-
radation during the analysis, is modeled using one layer of SOLID45
elements [17] per laminate ply. SOLID45 elements share properties
with the above mentioned SOLID46 without the ability to repre-
sent multiple layers, yet they feature plasticity models. By model-
ing each single composite layer, large shear-deformations with a
component in thickness-direction of single layers can be accounted
for and contact in the boltspecimen interface is represented more
accurately than is the case with elements that integrate several
layers, due to a close-meshed distribution of nodes on the outer
surface. With the objective of reducing the total number of degrees
of freedom, the inactive area of the specimen is modeled with SO-
LID46 elements and joined with the active part by bonded contact
elements.
Bolts and washers are built of SOLID45 elements. Degenerating
these to wedges is not recommended since they are then less accu-
rate in high stress gradient regions [17], but it is tolerated in case of
the bolts which are not subject to stress evaluation and created of
regular 8-node elements towards their outside.
4.2. Contact denitions
Interference of separate solid body FE entities, which do not
share nodes, is made possible by dening contact elements. The
nodes that embody the outer surface of the volumes are shared
by the solid elements and contact and target elements for this pur-
pose. In absence of contact pairs, different separate volumes geo-
metrically penetrate unrestrained. Surface-to-surface contact
pairs that need to be dened in this model are: bolt shaft & washer
(radial), bolt head & washer face (axial), washer face & pulling
strap face, bolt shaft & pulling strap hole surface, pulling strap face
& specimen face, bolt shaft & specimen hole surface and SOLID45
part of specimen & SOLID46 part of specimen.
ANSYS provides a variety of contact algorithms which differ in
stiffness behavior both lateral and normal, penetration tolerance,
stability and convergence rate. As a default setting, most likely to
be applicable in various cases, ANSYS offers the augmented La-
grange method. Analyses carried out on an early version of this
model, incorporating only penalty method contact denitions with
default contact stiffness, exhibited extensive penetration between
the bolt shaft and both specimen and pulling strap. Another con-
tact algorithm provided by ANSYS is the Lagrange multiplier meth-
od. It enforces zero penetration by applying a geometrical
constraint relation and introducing the contact forces as additional
degrees of freedom, that are solved explicitly, rather than implic-
itly through contact stiffness and penetration. However, if a
marginal penetration and tensile pressure upon existing gaps are
allowed (chattering control), convergence behavior can be im-
proved signicantly. In case too little chattering control is dened,
convergence may be impeded by abruptly changing contact status
between successive iterations.
The use of the Lagrange multiplier method in the presented
model results in accurate modeling of contact behavior with an
acceptable convergence rate. It is applied to all penetration-sensi-
tive contact pairs, i.e. those involving the bolt shaft due to their sig-
nicance for the joint stiffness, the interface between pulling-strap
and specimen, in case bolt pretension studies are performed, as
well as the bonded contact between layered and non-layered ele-
ment specimen region. In order to reduce computational costs,
ANSYS features a contact algorithm that employs Lagrange multi-
pliers in the contact normal and penalty method in the tangent
direction. This is implemented instead of the pure Lagrange multi-
plier method, with the exception of the bonded specimen contact.
As a measure to further improve the stability of the contact
algorithm, initial contact closure is allowed over small gaps and
penetration. In order to avoid problems arising from discontinuous
stiffness distribution, the specimen volume is covered by very thin
layers of solid elements (see Fig. 8) which are not subjected to deg-
radation, but other than that feature the same characteristics as the
next underlying elements. These steady layers transmit the contact
reaction to the degradation-active elements and are thin enough to
not signicantly redistribute stresses around damaged regions.
4.3. Constraints and body loads
The most direct way of dening load to a model that outputs
nominal bearing stress against displacement is to impose displace-
ment. For this purpose, the particular nodes that embody the long
ends face of the pulling strap have all their degrees of freedom
constrained to zero. The outer specimen face is treated equally, ex-
cept for an offset in longitudinal direction amounting to the re-
quired displacement. With the specimen constrained in this
manner, rigid body motion is prevented. Further stabilization to
the model is given through xation of the bolt in the symmetry
plane by very compliant springs. These are implemented as linear
LINK8 elements [17]. A symmetry constraint is applied to all nodes
belonging to the bolt and the specimen volumes that are located in
the symmetry plane. Since the used element types only feature
translational degrees of freedom, only one degree of freedom nor-
mal to the symmetry plane is locked.
Depending on the employed contact algorithms, a restriction
concerning degree of freedom constraints is necessary: If surface
body nodes are shared by a contact element that uses the Lagrange
multiplier method, further constraining these nodes will lead to an
over-constraint, due to conicting external boundary conditions
Fig. 8. Detail of the specimen highlighting the radial and axial steady element
additional layers.
 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163 57

and internal contact constraint conditions [17]. This effect is

avoided by neglecting the nodes that are meshed with contact ele-
ment when applying constraints. However, the accompanying tar-
get element side of the contact zone is not affected by this
restriction.
Bolt pre-load is simulated by dening a thermal expansion coef-
cient as in [18]. Here, a positive thermal expansion coefcient is
given to the bolt material and a temperature drop is simulated in
a rst load step as a body load. The thermal contraction leads to
compression of the pulling strap and specimen and given all axial
stiffness values, the equivalent clamping torque can be
approximated.
Fig. 9. Approximation of non-linear stressstrain curve by means of degraded
linear segments. (For interpretation of the references to color in this gure legend,
the reader is referred to the web version of this article.)
4.4. Material models

The quasi-static material behavior of the CFRP plies is described

by an orthotropic material model dened by three elastic moduli
(EX, EY, EZ), three shear moduli (GXY, GYZ, GXZ) and three major
Poissons ratios (PRXY, PRYZ, PRXZ). Due to orthotropy, the values
of EY and EZ, GXY and GXZ as well as PRXY and PRXZ are pair-wise
equal in their initial form. As soon as degradation takes place,
orthotropy is no longer maintained. The ber orientation is re-
garded by the alignment of the element coordinate systems with
their respective layer orientation. Since in [4] data are provided
for the obligatory parameter identication and validation, the
material used there for testing is simulated in this model. The com-
posite material used is the M40-J/CYCOM 977-2 having its proper-
ties listed in Tables 2 and 3. Due to the linearity of the CFRPs
material model, complex material response curves cannot be fea-
tured directly. Descending stressstrain gradients for example,
can only be approximated by a series of degradations. An exem-
plary characteristic stressstrain curve along with a corresponding
linear material model is illustrated in Fig. 9.
The titanium sheets used as reinforcement layers are modeled
as an isotropic material with bilinear kinematic hardening option.
This is a plasticity model which is dened by a yield stress and a
secondary modulus for the plastic domain. An exemplary run using
a basic brick-shaped test specimen under uniaxial, arbitrary load-
ing exhibited the characteristic curve shown in Fig. 10. In [4] the
titanium alloy Ti 15-3-3-3 is applied, for which an elastic modulus
of 116 GPa and yield stress of 1534 MPa are specied. The second-
ary modulus is estimated to 1500 MPa on the basis of a stress
strain curve given in [19]. Rupture of titanium is not included in
the model, which is legitimated by the fact that rupture strains
of the employed alloy is greater than 10% as opposed to about 1%
and 3.5% for the composite material in longitudinal and transverse
direction respectively. For the bolts and washers, the same tita-
nium material denition as for the reinforcement plies has been
used, though without plasticity feature.
Table 2
Initial elastic moduli of the CFRP ply taken from [4].
E|| (MPa) E\ (MPa) G||\ (MPa) G\\ (MPa)
211,424 6287 3895 2095.7
Table 3
Strength data of the material used as identication base taken from [4]. The shear
strength is estimated to the value stated here by the author.
S||,T (MPa) S||,C (MPa) S\,T (MPa) S\,C (MPa) S||\ (MPa)
2132 994 47 217 67
Rp 2Rp
Fig. 10. Stressstrain curve of the plasticity model considered for Ti layers under
uniaxial load. Rp is the yield stress of the specied material.
5. Implementation
The requirements involved in the PDM procedure include the
necessity to enable material denition changes and altering
boundary conditions after previous load steps have been solved.
For correct modeling of hysteretic effects, such as plasticity, a solu-
tion always has to be obtained with regard to the previous load
step. Beyond that, a certain amount of solution control should be
provided in order to overcome convergence problems. These crite-
ria can all be met by restarting the static analysis in the ANSYS
environment. In case of nonlinear static analyses, the user can
choose the reference for the next load step from any solution yet
gained. Within this model, the default, being the last step, is used.
Restarting the analysis, while changing material properties and
parameters, necessitates some work-around to avoid unintentional
side-effects. For instance, all parameters dened and all material
properties included in the model will be set back to the state they
had during the rst restart Therefore, with each restart, all param-
eters have to be saved before and restored afterwards in order to
be accessible at any time. The same applies for the reassignment
m||\ m\\ of material properties for each element. Every degradation-active
0.3 0.5
element has its own set of material properties and additionally,
the stiffness moduli are archived in an array with the dimension
(n  9), with n being the total number of degradation-active ele-
ments. After the moduli have been calculated, according to the
degradation routine and stored into the array mentioned above,
the material properties of those elements, that once faced degrada-
tion, have to be deleted, redened and reassigned at every load
S\\ (MPa) step. Element numbers of failed elements are recorded to an array
in order to be able to limit the extent of the reassignment loop to
30
its necessary minimum.
58 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163

B1 B6 Newton, LevenbergMarquart, etc.) or derivative-free heuristic

0 0 algorithms, such as the NelderMead simplex-reection method,
45 45
0 0 in combination with penalty function terms for constraint imple-

t
90 90 mentation, are the most common potentially appropriate methods.
-45 -45
symm. 0 symm. 0 More information about numerical optimization algorithms can be
found in [20]. These algorithms, however, cannot be conveniently
B7 B8 applied to the present numerical model due to the fact that with
0 0
45 45 any method mentioned above, a great number of iterations, i.e. to-
0 0 tal simulation runs on the level of identication, are required. It is
90 90
-45 -45 therefore necessary to deviate from standard optimization algo-
symm. 0 symm. 0 rithms and make use of engineers experience and knowledge
about the system, in order to minimize the iterations involved. In
Fig. 11. Schematic description of the laminate layouts studied herein (t = 0.25 mm).
fact, the procedure adopted here for identication is a heuristic
manual approach, based on few assumptions and simplications.
6. Parameter identication
It is performed by iterating the sequence simulation runevalua-
tiondrawing conclusionsaltering parameters. As an additional
Due to the implementation as a gray-box model, an obligatory
instrument of evaluation, damage monitoring has been also imple-
step within the creation of the model is the determination of iden-
mented. This means, that in each simulation sub-step, for each fail-
tication parameters. Parameter identication is the process of
ure criterion the numbers of failed elements (with damage variable
matching numerical simulation results with data gained from mea-
di > 1) have been recorded to an array. Furthermore, for each lam-
surements by adjusting the identication parameters accordingly.
inate layer and each failure criterion, images that visualize the
failed elements can be generated at user-dened times during
6.1. Model specications of identication base
the analysis.
During the process of parameter identication, the operator can
For the model presented in this work, the identication base
account for the following principles:
comprises measurement plots of tensile tests conducted with four
different specimen layouts of a Ti-reinforced single-bolted CFRP-
 If the simulated bearing stress in general exceeds the mea-
joint (see Fig. 11), with the geometrical parameters d = 6 mm,
sured data in the nonlinear domain, the stiffness residuals
w = 42 mm and e = 24 mm referring to the denitions illustrated
rij have to be reduced in total, and vice versa.
in Fig. 6. These measurement plots are given in [4] in terms of
 Reducing rij of one failure criterion will impede further
nominal bearing stress against absolute cross-head displacement.
stress accumulation after the highest fail-rate for this crite-
Strength data of the used material M40-J/CYCOM977-2, specied
rion occurred. Here, the damage monitoring mentioned
ibid., are summarized in Table 3 and implemented into the set of
above is a helpful tool for controlling enhancement.
failure criteria involved. However, the exact layout of the pulling
 Criteria that do not incorporate ber failure should not
strap and clamp mechanism are not sufciently known. Therefore,
entail degradation of stiffness terms other than those
in the model the pulling strap is kept short and during numerical
involving the normal of the fracture plane.
evaluation of the simulation data, a serial compliance element is
 Material softening due to secondary failure may inhibit
included in order to t the linear slope in the low-displacement do-
ber failure.
main of the measurement curve. Note that in [4] the transitions be-
 Coupling between two laminate layers is decreased by low-
tween the titanium replacement plies and CFRP plies are arranged
ering those shear stiffness terms involving the laminate
in a staggered form, in order to avoid notching effects. As opposed
normal. This affects the reinforcement capabilities by the
to that, in the model applied herein, all transitions are located in
titanium plies.
the same longitudinal coordinate. This is justiable by the fact that
 Stress distribution around the bolt hole from the bearing
nonlinear effects in the transition region are not the focus of this
zone to the tensile loaded lateral zone is alleviated by low-
work and, as described above, the global compliance is manipu-
ering the in-plane shear stiffness.
lated in post-processing.

6.2. The optimization process

By formal description, the identication process is an optimiza-

tion problem described by
min Dr ij subject to 0 < r ij  1 19
r ij

where D(rij) is the deviation function that quanties the deviation of

the simulated stress characteristic from the measurement data, for
instance by applying a mean square error approach. The stiffness
residuals rij represent the ratio of initial-to-minimal moduli in prin-
cipal direction j with respect to the ber coordinate system for
failure criterion i and have been specied as the identication
parameters in Section 3.2. Under consideration of
i 2 fMTF; MCF; DT; DC; FTF; FCFg and j e {x, y, z, xy, yz, xz}, the identi-
cation emerges as an optimization with 36 degrees of freedom.
For unidentied nonlinear numerical systems, the optimization
theory provides a manifold range of optimization algorithms, in Fig. 12. Inuence of maximum number of iterations per load step. The source of
theory suited to the class of problem at hand. Algorithms with local this plot is a simulation of B1 specimen with a preliminary set of identication
gradient approximation and gradient-based iterating (e.g. quasi- parameters.
 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163
Table 4
Parameters retrieved from identication with a = 1 and nd = 2.
rij j
x y z xy yz
i MTF 1 0.012 1 0.0075 0.0075
MCF 1 0.012 1 0.0075 0.0075
DT 1 1 0.006 1 0.006
DC 1 1 0.006 1 0.006
FTF 0.008 1 1 0.008 0.008
FCF 4.8  106 1 1 4.8  106 4.8  106
Aiming at the reduction of costs, the maximum number of dam-
age iterations per load step nd is limited to 2 throughout the iden-
tication process. See Fig. 12 for a comparison between plots
gained with nd = 2 and nd = 9 respectively, using the same [0/
45/0/90/45/0]s CFRP lay-up and a preliminary set of identi-
cation parameters. The lower sub-iteration number produces the
smoother characteristic, yet the global trends prove themselves
mutually well comparable. Increasing the damage transition width
a (see Fig. 4) generally inhibits stress degradation. However, after
the initial stages of the identication performed during this work,
a = 1 was applied constantly. The results gained from the identi-
cation process using the above methodology after a given comput-
ing time available for this task are listed in Table 4.
7. Numerical results and discussion
Mesh density of the model has a signicant effect on computa-
tional costs involved. Since simulation time for the present model
is extensive, the coarsest acceptable mesh is sought. However, as
degradation alters element properties, meshing parameters can
be expected to affect the prediction of nal failure load data. In
Fig. 13, the results gained from a parametric study on mesh density
are plotted. Mesh density is here dened by the number of subdi-
visions nDiv per quarter circle in the transition zone around the bolt
hole while the meshing algorithm used ensures that the radial ele-
ment dimension matches the tangential dimension. Basis for the
parametric study is the specimen layout B7 (see Fig. 11) with the
nal identication parameters given in Table 4. It is shown that
from a reference point of nDiv = 18 reducing the element size by
50% does not entail a signicant change in characteristics, while
an increase in size of 50% will lead to more extensive alterations.
Based on these results, for further procedure nDiv = 18 is used.
A comparison between results gained by models considering
bolt plasticity (plasticity model described in Fig. 10) and neglecting
it is provided in Fig. 14. The comparison is based on the specimen
layout B7 (see Fig. 11) with the nal identication parameters
Fig. 13. Parametric study on mesh density for the B7 specimen conguration.
59
xz
1
1
0.006
0.006
0.008
4.8  106
Fig. 14. Comparison between results with and without bolt plasticity in case of a B7
specimen.
given in Table 4. Given the negligible difference in the predicted
bearing stress courses, the model simplication by omission of bolt
plasticity is legitimated.
The numerical results comprise nominal bearing stress against
displacement characteristics, damage accumulation data and visu-
alization of damage zones. The data presented herein have been
gained primarily during the identication process itself.
Fig. 15 compares the bearing stress with respect to the applied
displacement obtained from the tensile tests performed in [4] to
the course simulated through the model for a [0/45/0/90/
45/0]s laminate and three titanium-reinforced variations (see
Fig. 11 for laminate congurations). It is unknown from the source
at which point ultimate material failure was dened, thus, in the
following it is assumed to occur at the high-displacement end of
the provided data string. A notable feature of the computed plots
is the good agreement with measurements in case of the pure CFRP
specimen B1 until the maximum recorded bearing stress. At this
point, the numerical model produces a sudden drop in stress, while
the test specimen is subject to continuous softening until nal fail-
ure. So, the beginning of the rst sudden drop in stress appears to
be a decent description for the nal failure criterion regarding the
determination of strength. Dening a criterion for nal failure is vi-
tal for a model that applies nonsingular degradation and consistent
FE-entities, since zero-force will never be reached in a deected
state. The resulting force will at any point of displacement remain
higher than a linear characteristic which corresponds to a model
with all active elements at their lowest possible moduli, dened
by the stiffness residuals rij.
For the titanium-reinforced laminates, however, the onset of
the rst stress drop proves to deliver a very conservative strength
prediction, compared to the experimental data. The drop occurs
consistently near the end of the linear domain of the measurement
plots, yet ends soon and passes into a gradient that closely resem-
bles the further characteristic of the measurement, until a second
drop follows near the occurrence of nal failure in the tensile test.
Accordingly, the onset of the second drop is considered a suitable
nal failure criterion for simulated reinforced specimens, though
the accuracy of strength prediction does not match the one ob-
served in case of pure CFRP laminates. For comparison, the pre-
dicted and measured nominal bearing stresses of the B1 and B7
conguration are depicted in Fig. 16. There, the predicted course
is only shown within the limits of the aforementioned nal failure
criteria. The comparison shows an almost identical close- to-linear
behavior at the early stages of loading for both joints. At applied
displacement of 0.075 mm, the stiffness of joint B1 drops signi-
cantly due to damage accumulation. Joint B1 is predicted to fail
at around 650 MPa. On the contrary, the stiffness of joint B7 drops
60 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163

Fig. 15. Comparison between simulation results and experimental data [4] of the parameter identication base.

at the applied displacement of 0.2 mm but in smaller degree than

the stiffness of joint B1. Joint B1 is predicted to fail at around
1200 MPa.
The measured and predicted bearing strengths and displace-
ments at fracture under consideration of the nal failure deni-
tions explained above are summarized in Table 5. The most
evident deviations between numerically and experimentally
gained results are observed at high titanium content regarding
the joint strength. These particularly arise from the rst accumula-
tion of failures that entail a drop in tensile force, which is never
compensated in the later course. Predicted displacements at ulti-
mate failure slightly are below their measured equivalence
throughout the tested range, i.e. material brittleness is overesti-
mated. This may partly be due to the lack of modeled plasticity
of the resin component of CFRP.
As a background for discussion on the effect of insertion of tita-
Fig. 16. B1 and B7 congurations compared in measurement and simulation under
consideration of nal failure criteria. (For interpretation of the references to color in
nium foils in this model, Fig. 17 provides data on the number of
this gure legend, the reader is referred to the web version of this article.) failed elements during the analysis, for both pure CFRP-layout B1
and the reinforced layout B7. The sub-plots (.1) show the progress
of element failures, normalized to the respective values at dis-
placement Dx0 = 0.5 in favor of comparability. Most noticeably,
Table 5 the total number of failures decreases signicantly from 9550 to
Predicted strengths and fracture displacements of the identication base in compar-
3229 due to the titanium layers. In this regard, note that the num-
ison with measurement data.
ber of potentially degradable elements is reduced to 2/3 of the ini-
B1 B6 B7 B8 tial number through the replacement of CFRP layers by titanium.
Nominal bearing strength N
mm2
Measured 680 1040 1390 1715 Still, the decrease in failures to 1/3 clearly overcompensates that.
Predicted 691 934 1273 1529 Moreover, a general shift of the normalized failure curves to the
Displacement at fracture mm Measured 0.39 0.53 0.54 0.59
right indicates that failure onset occurs later in total, except for -
Predicted 0.32 0.47 0.50 0.53
ber tensile failure (FTF). In relation to the total number of failures,
 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163 61

0
Fig. 17. Course of normalized numbers of element failures (.1) and total numbers of element failures at Dx = 0.5 mm (.2) for B1 (a) and B7 (b) specimen respectively. (For
interpretation of the references to color in this gure legend, the reader is referred to the web version of this article.)

ber failures gain weight over the secondary failures. This particu-
larly applies to ber tensile failure, whose total number increases
in the reinforced specimen layout. In spite of comparably low total
failure count for this failure type, the higher step-like accumula-
tions of ber tensile failures apparently coincide with drops in
bearing strength (compare Figs. 15 and 16). Since softening
through secondary failures is diminished by the titanium rein-
forcement, stresses are increasingly remote around the bolthole
leading to early ber failure in the lateral zone. However, the tita-
nium sheets increase the tolerance of ber failure thus, allowing
for a further incline in bearing strength, which is in agreement
with reality.
Evaluation of the area of failure propagation, that emerged in
the pure CFRP specimen after nal failure was detected, provides
information on the global failure mechanisms involved. In case of
the secondary failure modes delamination and matrix failure, dam-
age accumulates in the bearing pressure zone and propagates
towards the specimen longitudinal free edge in shape of a longitu-
dinal projection of the bolt hole. In this band, matrix compressive
failure in the layers with diagonal ber direction particularly oc-
curs on the side where the plain shear stress caused by the central
bolt force contributes to matrix compression under consideration
of polar stress transformation. Areas of elements, exhibiting matrix
compressive failure at Dx0 = 0.64 mm, are depicted in Fig. 18. The
opposite effect can be found in the damage visualization for matrix
tensile failure.
The secondary failures show correlation with the occurrence of
ber compressive failure in the adjacent layers. It is concluded that
secondary failure is additionally triggered by interlaminar shear
and stress concentration that evolves from the stress gradients
caused by ber failure. The appearance of ber compressive failure
(see Fig. 19) is limited to areas in the projection of the bearing zone
along the ber direction of each layer. Fiber tensile failure exclu-
sively occurs in low quantities in the lateral areas directly along-
side the bearing hole.
Similar damage plots depicting matrix compressive failure and
ber compressive failure are given in Fig. 20 for a titanium-rein-
forced B7 specimen layout at a comparable displacement. The
reduction of damaged area is an evident feature. particularly with
regard to non-ber failure. These are limited to the longitudinal
projections of the lateral hole edges where in-plane shear stress
prevails and lack the asymmetric characteristic induced by the
diagonal ber layers.
8. Conclusions and outlook
A progressive damage model for the numerical representation
of bearing effects in hybrid CFRP/Ti bolted joints has been
62 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163
0
Fig. 18. Visualization of failure after simulation of B1 specimen at Dx = 0.64 mm:
matrix compressive failure (MCF). (For interpretation of the references to color in
this gure legend, the reader is referred to the web version of this article.)
0 quires a tensile test of a bolted joint with an arbitrary laminate
Fig. 19. Visualization of failure after simulation of B1 specimen at Dx = 0.64 mm:
ber compressive failure (FCF). (For interpretation of the references to color in this
gure legend, the reader is referred to the web version of this article.)
0
Fig. 20. Visualization of failure after simulation of B7 specimen at Dx = 0.58 mm:
matrix compressive failure (MCF) and ber compressive failure (FCF). (For
interpretation of the references to color in this gure legend, the reader is referred
to the web version of this article.)
developed in this work. Its main foundation is a new set of stress-
based failure criteria and a continuous stress-based degradation
strategy.
Identifying the free parameters that characterize the degrada-
tion process by matching results with measurement data pro-
vided by an external source proves to be a demanding and
expensive process. However, the results show that the model
developed herein is capable of reproducing the bearing behavior
of both pure CFRP plane laminates and Ti-sheet-reinforced lam-
inate types. Strength prediction of the identication base is in
agreement with measurement data within accuracy fewer than
2% for a pure CFRP-specimen and within 11% for three different
layouts with titanium fractions up to 33%. A clear correlation be-
tween titanium addition to the model and strength enhancement
has been observed.
In order to use this model, the set of degradation parameters
that is associated with the materials failure characteristic, has to
be identied for every type of bermatrix combination. This re-
layout to be performed as a base of the identication process, as
well as the identication itself. Given the similarity of
 G. Kolks, K.I. Tserpes / Composites: Part A 56 (2014) 5163
mechanical properties within the same class of resin and bers,
the identication process can be comparably quickly done if
parameters for a similar material have already been determined.
Nevertheless, an implementation of a physical degradation mod-
el based on fracture toughness in a future work would improve
the usability and potentially the accuracy of the model. Applying
damage transition widths a differentially to each failure criterion
would be a sub-step towards a more accurate model, yet would
further increase the number of degrees of freedom of the
identication.
An eminent restriction of the model is the limitation to three
orthogonal fracture planes. This decision was made upon the prin-
ciple guideline to create a clear and computationally lean model. In
this case, with the application to non-crimp CFRP laminates, the
restrictions are justiable given the results. The effect of a variety
of parameters, such as mesh density, bolt clearance, bolt pre-stress,
and friction in the contact zone, yet have to be investigated in fu-
ture work, in order to gain better verication and deeper under-
standing in the behavior of the model. Another interesting aspect
to do research on is damage distribution when applying the model
to serial multiple-row bolt joint design. For future use of the model
in parametric studies or applications in design optimization, the -
nal failure criterion that species the event of total specimen fail-
ure has to be dened quantitatively in order to gain exact
information on predicted total strength. The qualitative a posteri-
ori criteria fall short of requirements needed for scientic studies,
such as objectivity.
Acknowledgements
The research leading to these results has received funding from
the European Union Seventh Framework Programme (FP7/2007-
2013) under grant agreement no. 266026 (Project CERFAC: Cost
Effective Reinforcement of Fastener Areas in Composites). The rst
author would also like to thank Professor Sp. Pantelakis, director of
LTSM (University of Patras) and Dr. A. Dafnis, engineer-in-chief at
ILB (Institute of Lightweight Design at RWTH Aachen University,
Germany).
63
References
[1] Camanho PP, Matthews FL. Stress analysis and strength prediction of
mechanically fastened joints in FRP: a review. Compos A Appl Sci Manuf
1997;28:52947.
[2] Prof. Reimerdes HG. Faserverbundwerkstoffe I. Lecture notes; 2005.
[3] Schrmann H. Konstruieren mit Faser-Kunststoff-Verbunden. Heidelberg: Springer-
Verlag Berlin Heidelberg; 2007.
[4] Camanho PP, Fink A, Obst A, Pimenta S. Hybrid titanium CFRP laminates for
high-performance bolted joints. Compos A Appl Sci Manuf 2009;40:182637.
[5] Camanho PP, Tavares CML, de Oliveira R, Marques AT, Ferreira AJM. Increasing
the efciency of composite single-shear lap joints using bonded inserts.
Compos B Eng 2005;36:37283.
[6] Kolesnikov B, Herbeck L, Fink A. CFRP/titanium hybrid material for improving
composite bolted joints. Compos Struct 2008;83:36880.
[7] Tserpes KI, Labeas G, Pantelakis S, Kermanidis T. Strength prediction of bolted
joints in graphite/epoxy composite laminates. Compos B Eng 2002;33:5219.
[8] Hashin Z. Failure criteria for unidirectional ber composites. J Appl Mech
1980;47:32935.
[9] Tserpes KI, Labeas G, Pantelakis S. Multi-scale modeling of the mechanical
response of plain weave composites and cellular solids. Theoret Appl Fract
Mech 2010;54:1729.
[10] Hhne C, Zerbst AK, Kuhlmann G, Steenbock C, Rolfes R. Progressive damage
analysis of composite bolted joints with liquid shim layers using constant and
continuous degradation models. Compos Struct 2010;92:189200.
[11] Maim P, Camanho PP, Mayugo JA, Dvila CG. A continuum damage model for
composite laminates: Part I constitutive model. Mech Mater
2007;39:897908.
[12] Maim P, Camanho PP, Mayugo JA, Dvila CG. A continuum damage model for
composite laminates: Part II computational implementation and validation.
Mech Mater 2007;39:90919.
[13] Davila CG, Camanho PP, Rose CA. Failure criteria for FRP laminates. J Compos
Mater 2005;39:32345.
[14] Pinho ST, Davila CG, Camanho PP, Iannucci L, Robinson P. Failure models and
criteria for FRP under in-plane or three-dimensional stress states including
shear non-linearity. NASA Langley Research Center: Technical memorandum;
2005.
[15] Tserpes KI, Papanikos P, Kermanidis T. A three-dimensional progressive
damage model for bolted joints in composite laminates subjected to tensile
loading. Fatigue Fract Eng Mater Struct 2001;24:66375.
[16] Michaeli W, Mannigel M, Preller F. On the effect of shear stresses on the bre
failure behaviour in CFRP. Compos Sci Technol 2009;29:13547.
[17] ANSYS, Inc. Release 11.0 Documentation for ANSYS, 2006.
[18] McCarthy MA, McCarthy CT, Lawlor VP, Stanley WF. Three-dimensional nite
element analysis of single-bolt, single-lap composite bolted joints: Part I
model development and validation. Compos Struct 2005;71:14058.
[19] Fukushima A, Fujiwara C, Kagawa Y, Masuda C. Effect of interfacial properties
on tensile strength in SiC/Ti-15-3 composites. Mater Sci Eng A
2005;276:2439.
[20] Nocedal J, Wright SJ. Numerical optimization. second ed. New York: Springer
Science+Business Media; 2006.