COVER SHEET

NOTE: This coversheet is intended for you to list your article title and author(s)
name only
—this page will not appear on the CD-ROM.

Paper Number: 1102 (replace with your paper number)

Title: Modeling Fatigue Damage Onset and Progression in Composites Using

an Element-Based Virtual Crack Closure Technique Combined with the Floating
Node Method

Authors: Nelson V. De Carvalho

Ronald Krueger
 (FIRST PAGE OF ARTICLE – align this to the top of page – leave space blank
above ABSTRACT)

ABSTRACT

A new methodology is proposed to model the onset and propagation of matrix

cracks and delaminations in carbon-epoxy composites subject to fatigue loading. An
extended interface element, based on the Floating Node Method, is developed to
represent delaminations and matrix cracks explicitly in a mesh independent fashion.
Crack propagation is determined using an element-based Virtual Crack Closure
Technique approach to determine mixed-mode energy release rates, and the Paris-Law
relationship to obtain crack growth rate. Crack onset is determined using a stress-
based onset criterion coupled with a stress vs. cycle curve and Palmgren-Miner rule to
account for fatigue damage accumulation. The approach is implemented in
Abaqus/Standard® via the user subroutine functionality. Verification exercises are
performed to assess the accuracy and correct implementation of the approach. Finally,
it was demonstrated that this approach captured the differences in failure morphology
in fatigue for two laminates of identical stiffness, but with layups containing θ° plies
that were either stacked in a single group, or distributed through the laminate
thickness.

INTRODUCTION

Fatigue damage in composite materials generally results in a combination of

multiple delaminations and matrix cracks. These damage mechanisms are not
isolated and often interact, forming complex damage patterns. Few approaches
attempt to model these damage modes and their interaction explicitly in fatigue [1,

Nelson V. De Carvalho, R. Krueger, National Institute of Aerospace, 100 Exploration Way,

Hampton, VA, 23666. This work was performed at the Durability, Damage Tolerance and
Reliability Branch, MS 188E, NASA Langley Research Center, Hampton, VA, 23681, U.S.A.
2]. In [1], both delamination and matrix crack onset are modeled using stress-based
criteria and ad-hoc stiffness degradation. Such an approach is mesh-subjective,
limiting its applicability as a predictive tool. In [2], both matrix cracks and
delamination are modeled with a fatigue cohesive element approach, rendering an
approach that can be mesh-objective. Matrix cracks are represented using the
regularized Extended Finite Element Method (rXFEM), and delaminations are
represented through interface elements. Both damage onset and propagation for
matrix cracks and delaminations are modeled through a fatigue cohesive approach
[2].
The Floating Node Method (FNM) has been proposed in [3] to represent
multiple discontinuities in solids in a mesh independent fashion. The method
provides a single approach to model both damage mechanisms (matrix cracks and
delamination) and their interaction. Furthermore, one of its main advantages is the
simplicity with which multiple cracks and their interaction can be accommodated
explicitly within an element.
In the present work, a three-dimensional (3D) extended interface element, based
on the FNM, is proposed and combined with an element-based implementation of
the Virtual Crack Closure Technique (VCCT) [4] and an onset criterion. The
resultant approach is capable of simulating both fatigue damage onset and
propagation. In the first section, an overview of the 3D interface element is
provided. Subsequently, a description of the element-based VCCT technique and
how it is used to propagate both matrix cracks and delaminations is given.
Afterwards, the procedure used to determine matrix cracks and delamination onset
in fatigue using a stress-based approach while assuming Linear Elastic Fracture
Mechanics (LEFM) is described. Two sets of verification exercises for matrix
cracks and delaminations are performed to assess accuracy and implementation of
the approach being proposed. Finally, the approach is demonstrated by simulating
crack onset and accumulation in fatigue in cross-ply laminates with different
stacking sequences.

3D CRACK REPRESENTATION IN COMPOSITES USING THE FLOATING

NODE METHOD: EXTENDED INTERFACE ELEMENT

In the present work, the FNM [3] was used to develop a 3D extended interface
element as shown in Figure 1. This element is used to model an interface and the plies
immediately above and below that interface via sub-elements 1 and 2. The plies
modeled with sub-elements 1 and 2 can have the same or different orientations and
thicknesses. Formulated in this fashion, the extended interface element provides
access to the stress state at the interface and within the plies immediately above/below
(Figure 1a). Additionally, it enables resolution of the kinematics of multiple crack
interaction (delamination/matrix crack) within a single element. Each sub-element can
be divided along any vertical plane to represent matrix cracks of any orientation
(Figure 1b). In addition to 16 real nodes, the element has a total of 32 floating nodes,
which are used as required to represent discontinuities, or are otherwise condensed out
of the system of equations. After the initial partitioning, the sub-elements can be
further sub-divided as needed for integration purposes. This is illustrated in Figure 1b,
sub-element 2, where the green lines represent further subdivisions needed for
integration. No limits to the number of cracks are imposed. Crack spacing is only
limited by the assumption that each sub-element can only represent one matrix crack.

Sub-element 2

Interface

Sub-element 1

Floating node
Real node
a) extended interface element comprised of two b) representation of matrix cracks with different
sub-elements orientations
Figure 1. Extended interface element with interface and matrix crack representation.

DELAMINATION AND MATRIX CRACK PROPAGATION

Matrix cracks and delaminations are represented with the FNM and the extended
interface element summarized above. To determine crack propagation, an element-
based VCCT technique is proposed and will be detailed in the next section.

Element-based Virtual Crack Closure Technique (VCCT)

In the element-based VCCT, the shape functions of interface elements are used to
obtain tractions and displacements at integration points. The tractions and
displacements are then used to compute energy release rates for the elements at the
crack front. The same methodology is applied for both matrix cracks and
delaminations, as illustrated in Figures 2a and 2b. Figure 2c shows a planar view of a
crack front (matrix crack or delamination). For each pair of elements 𝑒! , 𝑒! , where
𝑒! designates elements adjacent to 𝑒! , energy release rates can be computed in opening
mode and shear mode using VCCT as:
! ! ,! ! ,!
! !! 𝜎!! 𝛿!! 𝐴!
𝐺!! ! = (1)
2𝐴!!
!!!

! ! ,! ! ,!
! !! 𝜏! ! 𝛿! ! 𝐴!
𝐺! ! ! = (2)
2𝐴!!
!!!

! ! ,! ! ,!
! !! 𝜏! ! 𝛿! ! 𝐴!
𝐺! ! ! = (3)
2𝐴!!
!!!
where 𝑠 and 𝑡 designate two in-plane orthogonal directions and n is the normal to that
plane, as shown in Figure 2c. In equations 1 to 3, 𝐴! corresponds to the area of each
integration point i, 𝐴!! to the area of the element 𝑒! , and I is the total number of
! ,! ! ,! ! ,!
integration points. The normal traction 𝜎!! , and shear tractions   𝜏! ! , 𝜏! ! are
! ,! ! ,! ! ,!
computed at the integration points of element 𝑒! ; and the openings, 𝛿!! , 𝛿! ! , 𝛿! ! ,
are obtained at equivalent positions of elements 𝑒! (Figure 2c). The contributions from
each integration point pair are summed to obtain the energy releases for each pair of
elements. The maximum energy release rate for a pair of elements 𝑒! , 𝑒! is
determined as:
! !! ! !!! ! !!! ! !!!
!
𝐺!"# !
= 𝐺!! + 𝐺! ! + 𝐺! ! (4)

!
The maximum energy release for the element 𝑒! , 𝐺!"#!
, is assumed to be the
!! !!!
maximum 𝐺!"# obtained with the four pairs 𝑒! , 𝑒! :
! ! !! ! !!
𝐺!"#
!
= max 𝐺!"#
! !
, … , 𝐺!"#
! ! (5)

Once the maximum is determined, the adjacent element 𝑒!∗ , associated with the
maximum value, is identified. It is then possible to determine the mode-mixity, 𝛽 !! ,
as:

!! !!!∗ !! !!!∗ (6)

𝐺! + 𝐺!
𝛽 !! = !
𝐺!"#
!

and the characteristic length, 𝑙 !! , corresponding to the length of the common edge
between 𝑒! and 𝑒!∗ (see Figure 2 where 𝑒!∗ = 𝑒! is assumed). The characteristic length,
𝑙 !! , is used in the calculation of the cycles needed for a given element to open, as
detailed in the following section.

a) delamination b) matrix crack

4 3
e3
1 2
4 3 4 3 4 3
e4 e0 e2
1 2 1 2 1 2
l e0
t 4 3
n s e1
integration point
1 2
c) crack plane view. The solid circles represent integration points in each element.
Figure 2. Element-based VCCT applied to both delaminations and matrix cracks.
Propagation delamination and matrix cracks

Delaminations and matrix cracks are assumed to propagate following the Paris
Law [5] given by:

d𝑎 !
= 𝑐 𝐺!"# (7)
d𝑁

where 𝐺!"# corresponds to the maximum energy release rate obtained at peak load, 𝑎
designates the crack length and 𝑁 the cycles. The coefficient 𝑐 and exponent 𝑛 are
assumed to be a piece-wise linear function of the mode-mixity. At a given step 𝑠, the
energy release rate, mode-mixity, and the growth rate are determined for each element
𝑒! at the crack front. A binary failed/not failed approach is implemented. The un-
!
cracked area 𝐴!"! is used as an internal state variable that tracks crack accumulation for
the elements at the crack front that do not fail in a given step. An element is
!
considered to fail if its un-cracked area 𝐴!"! is reduced below a fraction 𝑓   of the
original area:
!
𝐴!"! !!!
< 𝑓𝐴!! (8)

Otherwise the element is considered to be pristine. In the present study, 𝑓 = 0.1,

which assumes an element fails if the un-cracked area is less than 10% of the original
area. Hence, before damage:
!
𝐴!"! = 𝐴!! (9)

where 𝐴!! corresponds to the area of the element. The cycles needed to fail each
element at the crack front, at step 𝑠, can be obtained as:
!
𝐴!"! !
!!
∆𝑁 = !! (10)
d𝑎
𝑙 !!
d𝑁

in which length 𝑙 !! is the characteristic length associated with element 𝑒! . The values
of all ∆𝑁 !! are afterwards used to determine the cycle increment in the step, ∆𝑁!"# ,
which is assumed to be the minimum number of cycles needed to fail an element or
initiate a new crack. Knowing the cycle increment and the growth rate, the crack
increment,  ∆𝐴!! , can be calculated:
!!
d𝑎
 ∆𝐴 !!
=𝑙 !!
∆𝑁!"# (11)
d𝑁

and an updated un-cracked area obtained as:

! ! !!
𝐴!"! !!!
= 𝐴!"! !
− ∆𝐴 (12)
DELAMINATION AND MATRIX CRACK ONSET

In the present work, a stress-based criterion is used to determine fatigue damage

onset. The definition of a propagation zone, where the stress criterion is not active,
enables the combination of a stress-based criterion for fatigue crack onset with the
procedure detailed above for modeling crack growth. The definition of the propagation
zone, the onset criterion, and their implementation are discussed in the subsequent
sections.

Damage onset

The onset of delaminations and matrix cracks is determined by comparing a

maximum principal stress criterion to an assumed stress vs. cycle (S-N) curve and
using the Palmgren-Miner rule [6] to account for fatigue damage accumulation not
leading to onset.

MAXIMUM PRINCINPAL STRESS CRITERION

For delamination, the maximum (positive) principal stress, 𝜎!!"# , is computed in

the delamination plane:

(13)
𝜎! 𝜎! !
𝜎!!"# = + + 𝜏!! + 𝜏!!
2 2

in which 𝜎! , 𝜏! and 𝜏! are, respectively, the opening and the two orthogonal shear
tractions acting in the delamination plane. Once onset is determined in a given
element, the elements immediately adjacent are also considered to have failed,
enabling the calculation of energy release rates via VCCT, following the procedure
outlined above.
For matrix cracks, the maximum principal stress criterion, is written as:

𝜎!! 𝜎!! (14)

𝜎!! = + !
+ 𝜎!" !
+ 𝜎!"
2 2

where direction ‘1’ corresponds to the fiber direction, ‘2’ is orthogonal to ‘1’ in-plane
and ‘3’ is orthogonal to ‘1’, corresponding to the through-thickness direction. The
stress components, 𝜎!! and 𝜎!" , are assumed to not contribute to the onset of matrix
cracks. Their effect is accounted for in the delamination onset criterion. The stress in
the fiber direction, 𝜎!! , is also assumed not to contribute to the onset of matrix cracks.
Once matrix crack onset is detected in a given element, the crack is assumed to
propagate through the thickness and to the two adjacent elements along the crack
direction. This enables the determination of energy release rates via VCCT, and the
application of the propagation methodology outlined previously. Additionally, local
delaminations are assumed directly above the failed elements, enabling the initiation
of local delaminations triggered by matrix cracks.
STRESS VS. CYCLES (S-N) CURVE AND WEIBULL SCALING

Fatigue crack onset in composites is assumed to be reasonably well approximated

by an onset curve with the form [7]:

𝑆 = 𝑆! (1 − 𝛼log 𝑁 ) (15)

where 𝑆! is the static strength, and 𝛼 a parameter determined experimentally. The

onset of cracks in fatigue is an intrinsically probabilistic event. To capture its
probabilistic nature, the static strength in Equation 15 is assumed to follow a Weibull
distribution such that the probability of failure at a given stress 𝑃 𝜎 is given by [8]:

! !
! (16)
𝑃 𝜎 =1−𝑒 !!

where 𝜎! is the material characteristic strength and 𝑚 is the shape parameter. Both 𝜎!
and 𝑚 can be determined from experimental data. When perfoming the simulations,
the distribution is sampled for each element. This procedure inherently leads to a mesh
dependent solution: the finer the mesh, more sampling will be performed, and hence
the greater the likelihood of obtaining values at the extremes of the assumed
distribution. To balance this effect, Weibull scaling is used [8]. Weibull scaling
provides a relationship between the strength of two volumes of the same material,
based on the notion that the larger the volume, the more likely it is to have a weak link
(weakest link assumption). The S-N data used in the present work were obtained using
a 3-Point-Bending (3PB) configuration [9]. Using Weibull scaling, the characteristic
strength, 𝜎!!!!" , obtained in [9], can be related to the strength of each element, 𝜎!! , in
a given discretization. Assuming predominantly tensile loading conditions, this
relation is given by [9]:
𝜎!!!!"
𝜎!! = ! !
! 𝑉! (17)
2 𝑚+1 𝑉!!"

where 𝑉 ! designates the volume of a given element and 𝑉!!" is the volume of the 3PB
specimen [9]. However in [9], Weibull scaling alone was not sufficient to account for
the differences between the strength obtained using a 3PB and a 4-Point-Bending
(4PB) test setup, despite predicting the trend correctly. Thus, further investigation may
be needed to assess the accuracy of the procedure outlined above.

PALMGREN-MINER RULE

Fatigue damage accumulation prior to the onset of discrete cracks is obtained by

the Palmgren-Miner rule [6]:
!
∆𝑁!"# !
C ! = (18)
𝑁! !
!!!
where  𝑁! ! and ∆𝑁!"# ! are, respectively, the number of cycles to failure and the
number of cycles accumulated at the stress corresponding to the 𝑠 !! step. Equation 18
can also be used to determine the cycles needed to fail an element at a given stress
level and step 𝑘:

∆𝑁!"#$% ! = 𝑁! !
1.0 − 𝐶 !!! (19)

In the current implementation, ∆𝑁!"#$% ! is computed for all elements at each step. At
the end of each step, the minimum number of cycles to onset and/or to propagate a
crack is determined and assumed to equal the cycle increment ∆𝑁!"# ! . Finally, all
elements for which:

∆𝑁!"#$% !
<1+𝑓 (20)
∆𝑁!"#

where 𝑓 = 0.1, are considered to fail in step 𝑘.

Propagation Zone

Previously, VCCT was used to simulate crack growth from an assumed damage
state [10]. This methodology assumes LEFM and hence, that stresses tend to infinity
at the crack front. Consequently, these assumptions prevent use of an approach that
predicts damage onset using a stress-based failure criterion, since refinement in the
vicinity of the crack tip would lead to continuously increasing stress, rendering a
mesh-subjective approach. For sufficient refinement (high stress at the crack tip),
propagation of a crack would no longer be controlled by a fracture mechanics
criterion, but always governed by the stress-based onset criterion. Therefore, a zone of
length 𝑙!! must be defined, within which the onset criterion is not activated.
For further clarification, let us first assume predominantly Mode I loading. The
stresses ahead of a crack tip can be estimated by:

𝐾! (21)
𝜎 𝑥 = + 𝑂(𝑥)
2𝜋𝑥  

where 𝐾! is the Mode I stress intensity factor, and 𝑥 the distance from the crack tip
along the crack path (Figure 3). Assuming plane strain conditions and neglecting
higher order terms, one can write:

𝐺! 𝐸 (22)
𝜎 𝑥 ≈
2𝜋𝑥  
in which 𝐺! is the Mode I energy release rate and 𝐸 the Young’s modulus.
In fatigue, it is generally possible to assume a threshold stress level, 𝜎!! , below which
fatigue onset is assumed not to occur within the number of cycles of interest. Using
Equation 22, it is possible to determine a distance away from the crack tip, 𝑙!!!! ,
associated with Mode I loading, where the stresses caused by the presence of a crack
decay below 𝜎!! :
 𝐺! 𝐸 (23)
𝑙!!!!   ≈ !
2𝜋𝜎!!!!  

The same approximation can be made for Mode II loading. Additionally, if the critical
energy release rates for Mode I and Mode II are assumed, 𝐺!" and 𝐺!!" , a further
conservative estimate can be obtained:
! ! ! ! (24)
𝑙!!!!"   ≈ !!!!"! , 𝑙!!!!!"   ≈ !!!!!"!
!!!!   !!!!  

where the subscripts 𝑛 and 𝑠 designate shear and normal tractions. Finally, the
maximum distance 𝑙!! , can be approximated by:

𝑙!! = max 𝑙!!!!" , 𝑙!!!!!" (25)

The distance 𝑙!! can then be used to define an area within which the onset criterion is
not activated (Figure 3).

th lth

lth x

Propagation
zone

Figure 3. Definition of the propagation zone, within which the onset criterion is not activated.

VERIFICATION

The approach outlined above was implemented in Abaqus/Standard®. The user

subroutine SDVINI was used to seed the elements with the initial strength values,
obtained by randomly sampling a Weibull distribution. The extended interface
element was implemented as a 3D UEL (user element subroutine). In addition, the
user subroutine UEXTERNALDB was used to interface with an external Python
script. This script is used to manage the crack insertion and propagation. To verify the
approach, two sets of verification exercises were performed, the first dedicated to
matrix cracks, and the second to delaminations.

Matrix Cracks

To assess the accuracy of the energy release rate calculation for matrix cracks,
numerical results are compared to the analytical solutions for an orthotropic plate with
a center crack under Mode I and Mode II loading conditions. The elastic properties are
assigned such that the fiber direction is aligned with the crack, in a similar fashion to
what was performed in [11]. Two cases were considered, one corresponding to Mode I
(Figure 4a), and another to Mode II (Figure 4b). The model is a square plate of 50×50
mm in-plane and 1 mm thickness. A crack of 5 mm length is assumed to be located at
the center of the plate. The plate was loaded with applied tractions as illustrated in
Figure 4. The center region of the specimen was progressively refined to assess the
accuracy and mesh objectivity of the approach. Numerical results were compared to
analytical solutions for Mode I [12]:
!
𝜎! 𝜋𝑎 (26)
𝐺! = ! !
4𝐺!" 𝐸! 𝐸!! !

2𝐺!" 𝐸!! !
+ 𝐸! 𝐸!! !
− 2𝜈!" 𝐺!" 𝐸!! !

and Mode II [12]:

!
𝜏! 𝜋𝑎 (27)
𝐺!! = ! !
4𝐺!" 𝐸!! 𝐸!! !

2𝐺!" 𝐸!! !
+ 𝐸! 𝐸!! !
− 2𝜈!" 𝐺!" 𝐸!! !

energy release rates, where 𝜎! and 𝜏! are the normal and shear remote loadings and
𝑎 is half of the crack length (Figure 4). The variables 𝐸, 𝐺 and 𝜈 are the in-plane
Young’s modulus, shear modulus and Poisson ratio. The subscripts ‘1’ and ‘2’ refer to
fiber direction and transverse direction, respectively.
Figure 4 shows that, for the same refinement, Mode II results have a slightly larger
error than Mode I. Nevertheless, for both loading modes, the solution is demonstrated
to converge to the analytical value.
 30" σ∞

25"

20" 2a"

Error"
15" element""
(%)"
0°" width"
10"

5" σ∞

0"
0.5" 0.25" 0.125"
Element"width"(mm)"
a) Mode I
30"
τ∞

25"

20"
τ∞
Error" τ∞
15"
(%)"
0°"
10"
τ∞
5"

0"
0.5" 0.25" 0.125"
Element"width"(mm)"
b) Mode II
Figure 4. Energy release rate error between numerical and analytical solutions as a function of the
mesh refinement.

Delaminations

In [13, 14], numerical benchmarks were proposed to assess the accuracy of

fracture mechanics based approaches to simulate delamination propagation under
Mode I and Mode II fatigue loading. The numerical models used correspond to
Double Cantilever Beam (DCB) and End-Notch-Flexure (ENF) specimens. These
benchmarks were used in the present work to assess the accuracy and correct
implementation of the approach for delamination propagation. The mesh, material
properties, and load conditions were prescribed following [13, 14] and for brevity are
not repeated here. The results obtained are provided in Figure 5, where the maximum
crack length at each step is plotted against number of cycles and compared to the
benchmark results [13, 14]. Further mixed-mode cases showing similar acceptable
agreement, and a detailed comparison to the current Abaqus/Standard® low-cycle
fatigue capability, are provided in [15].
 45   50   Simula.on  
Simula.on  
40   Benchmark   40   Benchmark  

a     35   a     30  
(mm)   30   (mm)  
20  
25   10  
20   0  
1   100   10000   1000000   1   100   10000   1000000  
N  (cycles)   N  (cycles)  
a) DCB b) ENF
Figure 5. Comparison between the numerical results obtained for Mode I (DCB) and Mode II
(ENF) loadings and the benchmarks provided in [13, 14].

APPLICATION

To demonstrate the adequacy of the approach to capture the onset and

accumulation of multiple cracks, the experiment proposed in [16] was simulated. In
[16], two different laminates with the same stiffness were considered. In laminate A,
[0! , 𝟗𝟎, 0! , 𝟗𝟎, 0! ], the two 90° plies are isolated, while in laminate B, [0! , 𝟗𝟎𝟐 , 0! ],
they are stacked in the center of the specimen. Specimens with the same dimensions
were manufactured from the two laminates. The specimens were loaded under
uniaxial tension-tension fatigue at the same load level, 𝜎!"# , corresponding to 60% of
the ultimate strength with a 0.1 load ratio and a frequency of 10 Hz. Experimental
results showed that the damage accumulation in fatigue was markedly different for the
two layups [16]. Specimens from laminate A exhibited a significantly higher crack
saturation density than from laminate B. Contrarily, matrix cracks in laminate B grew
faster and to a greater length than in laminate A. Additionally, the authors highlighted
the importance of capturing the 3D effects caused by the interaction and shielding of
cracks growing through the width. Indeed, two-dimensional (2D) analysis would
assume cracks would traverse the specimens completely. However, this was seldom
observed experimentally, with cracks arresting before traversing the specimen, due to
the interaction with other growing cracks. In [16], the material used in the experiments
was carbon/epoxy T300/914. In the present work, due to a lack of a complete set of
fatigue material data for T300/914, properties for the carbon-epoxy IM7/8552, which
has similar elastic properties (~15% difference in 𝐸!! ), were assumed instead.

Material properties and input data

In this section, the material properties and input data used in the application
example are summarized. The elastic properties assumed for IM7/8552 are provided
in Table I.

TABLE I. ELASTIC PROPERTIES, IM7-8552 [17]

𝐸!! 𝐸!! = 𝐸!! 𝜈!" = 𝜈!" 𝜈!" 𝐺!" = 𝐺!" 𝐺!"
161.0   GPa 11.38   GPa 0.32 0.44 5.17 GPa 3.98 GPa
 In [10], S-N data for the transverse strength of IM7/8552 using 3PB and 4PB tests
were produced. In the present work, the 3PB results were used, since they required the
least amount of scaling. A least squares fit of an S-N curve, as given in equation 19, to
the average of the failure strength obtained for each load level, yielded 𝛼 = 0.071
with a correlation coefficient of 𝑅! = 0.95. The Weibull characteristic strength, 𝜎! ,
and shape parameter, 𝑚, were also obtained from [10]. The material properties used to
determine crack onset, together with the assumed threshold value, 𝜎!! , are given in
Table II. The same S-N curve is used to model both matrix crack (intra-laminar) and
delamination (inter-laminar) onset. This assumption is a first approximation and can
be updated as reliable experimental data are obtained. Nevertheless, the equivalence
between intra-laminar and inter-laminar Mode I critical energy release rate has been
experimentally demonstrated for IM7-8552 [18].

TABLE II. MATERIAL PROPERTIES USED TO PREDICT CRACK ONSET IN THE

PRESENT STUDY
𝛼 𝜎! MPa [17] 𝑚 [17] 𝜎!! MPa
0.071 128.95 18.83 60.0

To model crack growth, the growth rate obtained in [19-21] for pure Mode I,
Mode II and mixed-mode I/II is linearly interpolated to obtain the Paris Law
coefficient and exponent, 𝑐 and 𝑛, in equation 7. The values used are summarized in
Table III.

TABLE III. PARIS LAW COEFFICIENT 𝑐 AND EXPONENT 𝑛 AS A FUNCTION OF THE

MODE-MIXITY 𝛽. NORMALIZED DATA ARE CONSIDERED FOR ALL 𝛽 EXCEPT 𝛽 = 1.0
𝛽=0 𝛽 = 0.2 𝛽 = 0.5 𝛽 = 0.8 𝛽 = 1.0
𝑐 173.83 2291.31 6.90 4.29 0.231
𝑛 6.77 8.38 5.41 5.07 5.45

Results

The models used are illustrated in Figure 6. For laminate A, symmetry conditions
were applied, and only half of the thickness was considered (Figure 6a). For laminate
B, no symmetry conditions were applied, due to the difficulty in adding boundary
conditions as damage evolved (note that no pre-defined damage location is assumed)
(Figure 6b). The two elements near the boundaries where traction is applied were not
allowed to fail - ‘no failure’ zone. This prevented numerical issues and ensured regular
load introduction throughout the analysis. In both cases, the lengthwise dimension was
assumed to be 2.0 + 0.18 mm, where 0.18 is the length of the ‘no failure’ zone. The
use of a representative region, rather than the full specimen, enables a significant
reduction in the computational time. For laminate A, FNM elements were used in the
center of the specimen. Each element modeled half of the ply above and below the
90/0 interfaces. For laminate B, one additional FNM element was used in the center
region, modeling half of the ply above and below the mid-plane 90/90 interface.
Elsewhere in the models, native Abaqus/Standard elements C3D8 (bi-linear, full
integration) were used. Tensile fatigue loading, 𝜎!"# , was applied corresponding to
approximately 60% of the ultimate strength, as in [16]. Assuming the 0° plies will
carry the final failure load, 𝜎!"# was estimated to be:
 𝜎!"# = 0.6𝑛!° 𝑋! (28)

where 𝑋! = 2560 MPa is the fiber tensile strength of IM7/8552 [22], and 𝑛!° is the
number of 0° plies in the laminates. Figure 6 also shows the damage state obtained
after 5×10! cycles. Qualitatively, it is possible to see the marked differences in crack
density observed at the edges in the two laminates. Laminate A shows a significantly
higher crack density at the edges of the two specimens compared to B. This is in
agreement with the experimental observations. Furthermore, in Figure 7, it is also
clear that not only the crack density differs, but also the average crack length:
Laminate B shows fewer longer cracks. Once again, this is in agreement with what
was reported in [16]. Additionally, the crack shielding, also reported in [16], can be
observed in both laminates. Figure 8 compares the crack density accumulation
observed experimentally and obtained numerically. The numerical results show
similar trends to the experimental observations. In a first stage, Laminate B shows
higher crack density than A. The cross-over point coincides approximately with the
saturation density for Laminate B. After this stage, laminate A continues to
accumulate cycles until saturation is reached. Quantitatively, the simulations show an
overall higher crack density than that obtained experimentally. Moreover, more cycles
are needed to obtain saturation. Indeed, saturation was only obtained for laminate A
after 5×10! cycles (squares with dark outline). However, it is interesting to observe
that the difference in saturation density between laminates A and B, ∆!"# , is
approximately the same in both experiments and simulations, despite the differences
in absolute values of crack density. It is worth recalling that the material simulated is
not the same material used in the experiments, which may contribute to the
quantitative differences observed. Nevertheless, the qualitative agreement
demonstrates that the approach is able to simulate both onset and growth of multiple
cracks and has the potential to capture the complex fatigue damage development,
being able to simulate the development of distinct damage patterns in two seemingly
equivalent laminates.

sym$

2.18$mm$

1$mm$
15$mm$

2$mm$
a) laminate A b) laminate B
Figure 6. Numerical model dimensions and loading. Edge crack density after 𝑁 ≈ 5×10! cycles.
Laminate A shows a higher number of cracks at the edges (elements in red) compared to Laminate B.
 a) laminate A

b) laminate B
Figure 7. Planar view of the 90 ply matrix cracks after 𝑁 ≈ 5×10! cycles. Fewer longer cracks are
observed in B compared to A. Crack shielding is evident in both specimens.

6" 6"
Laminate"A":"Experiments" Laminate"A":"SimulaBon"
T300/914" IM7/8552"
5" Laminate"B":"Experiments" 5" Laminate"B":"SimulaBon"
T300/914" IM7/8552" Δcds"
4" 4"
crack"density" crack"density"
Δcds"
3" 3"
(mm:1)" (mm:1)"
2" 2"

1" 1"
5.E+06" 5.E+06"
0" 0"
1.E+00" 1.E+02" 1.E+04" 1.E+06" 1.E+08" 1.E+00" 1.E+02" 1.E+04" 1.E+06" 1.E+08"
N"(Cycles)" N"(cycles)"
a) Experiments [14] b) Simulations
Figure 8. Comparison between the observed and predicted crack density.

CONCLUSIONS

A new approach is proposed to model damage onset and accumulation in fatigue.

It uses an extended interface element, based on the FNM method, to represent
delaminations and matrix cracks. Crack propagation is determined using an element-
based Virtual Crack Closure Technique and a Paris-Law relationship. Crack onset is
determined using a stress-based onset criterion and Miner’s rule to account for fatigue
damage accumulation. The approach is implemented in Abaqus/Standard® via the user
subroutine functionality. Preliminary verification shows good agreement between the
numerical results obtained and analytical solution/benchmarks. Finally, the approach
is demonstrated by capturing the experimentally observed differences in damage
development and morphology between two seemingly equivalent cross-ply laminates,
involving multiple crack onset and propagation.

REFERENCES

1. Fang, E., N. Simon, and J. Lua. 2014. “A Continuum Damage and Discrete Crack Module for
Fatigue Damage Prediction of Laminated Composite Structures,” presented at the 55th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, 13-
17 January, 2014.
2. Hoos, K., E. V. Iarve, M. Braginsky, E. Zhou, and D. H. Mollenhauer. 2016. “Progressive Failure
Simulation in Laminated Composites under Fatigue Loading by Using Discrete Damage
Modeling,” presented at the 57th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and
Materials Conference, 4-8 January, 2016.
3. Chen, B. Y., S. T Pinho, N. V. De Carvalho, P. M. Baiz, and T. E. Tay. 2014. “A Floating Node
Method for the Modelling of Discontinuities in Composites,” Engineering Fracture Mechanics,
27:104-134.
4. Rybicki, E. F., M. F. Kannine. 1977. “A Finite Element Calculation of Stress Intensity Factors
by a Modified Crack Closure Integral,” Engineering Fracture Mechanics, 9:931-938.
5. Paris, P. C., M. P. Gomez and W. E. Anderson. 1961. “A Rational Analytic Theory of Fatigue,”
The Trend in Engineering 13, 9-14.
6. Miner, M.A. 1945. “Cumulative Damage in Fatigue,” Journal of Applied Mechanics 67:A159-
A164.
7. May, M. and S. R. Hallett. 2011. “An Advanced Model for Initiation and Propagation of
Damage Under Fatigue Loading – part I: Model formulation,” Composites Structures, 93:2340-
2349.
8. Weibull, W. 1939. “A Statistical Theory of the Strength of Materials,” Ing. Vetenskaps Akad.
Handl. (Royal Swedish Institute Engineering Research Proceedings), NR151, 1939.
9. O’Brien, T. K., A. D. Chawan, R. Krueger, and I. Paris. 2001. “Transverse Tension Fatigue Life
Characterization Through Flexure Testing of Composite Materials,” NASA/TM-2001-211035,
2001.
10. De Carvalho, N. V., B. Y. Chen, S. T. Pinho, J. G. Ratcliffe, P.M. Baiz, and T.E. Tay. 2015.
“Modeling Delamination Migration in Cross-Ply Tape Laminates,” Composites Part A: Applied
Science and Manufacturing, 71:192-203.
11. Mabson, G. E., O. Weckner and M. Ramnath. 2012. “Finite Element Based Decohesive Failure
Simulation Sensitivity Studies,” presented at the 57th AIAA/ASCE/AHS/ASC Structures,
Structural Dynamics, and Materials Conference, 23-26 April, 2012.
12. Sih, G. C., P. C. Paris, G. R. Irwin. 1965. “On Cracks in Rectilinearly Anisotropic Bodies,”
International Journal of Fracture Mechanics, 1:189.
13. Krueger, R. 2010. "Development of a Benchmark Example for Delamination Fatigue Growth
Prediction," NASA/CR-2010-216723.
14. Krueger, R. 2011. "Development and Application of Benchmark Examples for Mode II Static
Delamination Propagation and Fatigue Growth Predictions," NASA/CR-2011-217305.
15. Krueger, R. and N. V. De Carvalho. 2016. “In Search of a Time Efficient Approach to Crack
and Delamination Growth Predictions in Composites,” to be presented at the 31st ASC
Technical Conference, 19-21 September, 2016.
16. Lafarie-Frenot, M. C. and C. Hénaff-Gardin. 1991. “Formation and Growth of 90° Ply Fatigue
Cracks in Carbon/Epoxy Laminates,” Composites Science and Technology 40:307-324.
17. O'Brien, T.K., R. Krueger. 2003. “Analysis of Flexure Tests for Transverse Tensile Strength
Characterization of Unidirectional Composites,” Journal of Composites Technology and
Research, 25:50-68.
18. Czabaj, M. W. and J. G. Ratcliffe. 2013. “Comparison of Intralaminar and Interlaminar Mode I
Fracture Toughnesses of a Unidirectional IM7/8552 carbon/epoxy composite,” Composites
Science and Technology 89:15-23.
19. Murri, G. B. 2013. “Evaluation of Delamination Onset and Growth Characterization Methods
Under Mode I Fatigue Loading,” NASA/TM-2013-0010557.
20. Ratcliffe, J. G. and W. M. Johnston. 2014. “Influence of Mixed Mode I-Mode II Loading on
Fatigue Delamination Growth Characteristics of a Graphite,” presented at the 29th ASC
Technical Conference, 8-10 September, 2014.
21. O’Brien, T. K., W. M. Johnston, and G. J. Toland. 2010. “Mode II Interlaminar Fracture
Toughness and Fatigue Characterization of a Graphite Epoxy Composite Material,” NASA/TM-
2010-216838.
22. Kaddour, A. S., M. J Hinton, P. A. Smith, and S. Li. 2013. “Mechanical Properties and Details
of Composite Laminates for the Test Cases Used in the Third World-Wide Failure Exercise,”
Journal of Composite Materials, 47(20-21):2427-2442.