A cohesive model for fatigue failure of polymers Spandan Maiti, Philippe H. Geubelle * Department of Aerospace Engineering and Beckman Institute, University of Illinois at Urbana-Champaign, 306 Talbot Laboratory, 104 S. Wright Street, Urbana, IL 61801, USA Received 19 April 2004; received in revised form 25 June 2004; accepted 27 June 2004 Available online 17 September 2004 Abstract A cohesive failure model is proposed to simulate fatigue crack propagation in polymeric materials. The model relies on the combination of a bi-linear cohesive failure law used for fracture simulations under monotonic loading and an evolution law relating the cohesive stiffness, the rate of crack opening displacement and the number of cycles since the onset of failure. The fatigue component of the cohesive model involves two parameters that can be readily calibrated based on the classical log–log Paris failure curve between the crack advance per cycle and the range of applied stress intensity factor. The paper also summarizes a semi-implicit implementation of the cohesive model into a cohesive-vol- umetric finite element framework, allowing for the simulation of a wide range of fatigue fracture problems. Ó 2004 Elsevier Ltd. All rights reserved. Keywords: Cohesive model; Fatigue failure; Cohesive finite element; Polymeric materials 1. Introduction Polymeric materials subjected to cyclic loading are highly susceptible to fatigue failure, which often pre- cedes monotonic or creep failures at similar applied load levels [19]. Substantial effort has thus been devoted to understand the mechanisms underlying the fatigue failure of this class of materials and to obtain predic- tive capabilities for their service life. Early attempts to characterize the fatigue failure in polymers were inspired by similar works for metals and were generally empirical in nature. Fatigue life test results were generally presented in the form of SN curves. But these curves are of limited use in practical situations as, unlike metals, the crack initiation life for these materials are often short compared to the fatigue crack propagation life [15]. Subsequently, different relationships were proposed to relate the per cycle crack 0013-7944/$ - see front matter Ó 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.engfracmech.2004.06.005 * Corresponding author. Tel.: +1 217 244 7648; fax: +1 217 244 0720. E-mail address: [email protected](P.H. Geubelle). Engineering Fracture Mechanics 72 (2005) 691–708 www.elsevier.com/locate/engfracmech
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Engineering Fracture Mechanics 72 (2005) 691–708
www.elsevier.com/locate/engfracmech
A cohesive model for fatigue failure of polymers
Spandan Maiti, Philippe H. Geubelle *
Department of Aerospace Engineering and Beckman Institute, University of Illinois at Urbana-Champaign,
306 Talbot Laboratory, 104 S. Wright Street, Urbana, IL 61801, USA
Received 19 April 2004; received in revised form 25 June 2004; accepted 27 June 2004
Available online 17 September 2004
Abstract
A cohesive failure model is proposed to simulate fatigue crack propagation in polymeric materials. The model relies
on the combination of a bi-linear cohesive failure law used for fracture simulations under monotonic loading and an
evolution law relating the cohesive stiffness, the rate of crack opening displacement and the number of cycles since the
onset of failure. The fatigue component of the cohesive model involves two parameters that can be readily calibrated
based on the classical log–log Paris failure curve between the crack advance per cycle and the range of applied stress
intensity factor. The paper also summarizes a semi-implicit implementation of the cohesive model into a cohesive-vol-
umetric finite element framework, allowing for the simulation of a wide range of fatigue fracture problems.
Polymeric materials subjected to cyclic loading are highly susceptible to fatigue failure, which often pre-
cedes monotonic or creep failures at similar applied load levels [19]. Substantial effort has thus been devotedto understand the mechanisms underlying the fatigue failure of this class of materials and to obtain predic-
tive capabilities for their service life. Early attempts to characterize the fatigue failure in polymers were
inspired by similar works for metals and were generally empirical in nature. Fatigue life test results were
generally presented in the form of S�N curves. But these curves are of limited use in practical situations
as, unlike metals, the crack initiation life for these materials are often short compared to the fatigue crack
propagation life [15]. Subsequently, different relationships were proposed to relate the per cycle crack
0013-7944/$ - see front matter � 2004 Elsevier Ltd. All rights reserved.
advance with various loading and fracture parameters, such as the range of energy release rate, the range of
the stress intensity factor (DK), or the mean stress intensity factor [21]. It has since been realized that the
Paris power relation between the crack advance rate per cycle (da/dN) and DK, which is extensively used to
represent the fatigue failure of metals, can also correlate the experimental data well for different polymers
for various loading and material parameters. But the slope of the Paris log–log curve, i.e., the exponent mof the power relationship between da/dN and DK in da/dN = CDKm, is generally much higher than that for
metals, typically exceeding a value of 6, indicating a much higher sensitivity to the applied loading.
Extensive experiments performed on different polymers established further differences between the mech-
anisms responsible for the crack growth in metals and glassy polymers. While it is mostly plasticity and
distributed damage ahead of the crack tip in the case of metals, the primary mechanism for crack growth
in polymers is the crazing of the material in a thin strip ahead of the crack tip. Though brittle to ductile
transition is widely reported for this class of materials [22], most polymers tend to fail in a brittle manner
under tension, and almost all of them fail in a macroscopically brittle way under cyclic loading provided thestress level is above some minimum value [18]. It has also been noted that, in the case of cyclic loading, in
addition to continuous per cycle crack growth as in metals, an additional crack propagation mechanism is
at work. In many polymers, the craze zone may grow over a number of cycles before the fibrils fail and the
crack jumps through two thirds of the craze length, giving rise to discontinuous crack growth [20,8].
One of the most successful techniques used to model crack propagation in polymers relies on the cohe-
sive modeling of fracture, which assumes the fracture process to take place in a vanishingly thin region
ahead of the crack tip. This approach seems indeed particularly appropriate to capture the failure of pol-
ymeric systems where, as mentioned earlier, the appearance of a thin crazing zone is a key component [7].Models based on cohesive technique have appeared in the literature for quasi-static [14] and dynamic crack
propagation [23] in polymers. Though there is a wealth of experimental results on fatigue crack propagation
in polymers, very few attempts have been undertaken to model fatigue crack propagation by cohesive tech-
nique in this class of materials.
Cohesive modeling technique has been attempted for fatigue crack growth in metals [3,13,5,4], along
interfaces [17] and in quasi-brittle materials [24]. In one of the early models proposed in [3], no distinction
was made in the model between the loading and unloading paths, but a damage parameter was assumed.
The evolution of this parameter with the number of load cycles was prescribed explicitly in the model. Thepresence of plasticity in the bulk material around the crack tip influenced the crack closure and hence the
failure of the material. But it was found that the crack ceases to grow after a few cycles due to plastic shake-
down [13]. It has since been identified that a distinction needs to be made between the loading and unload-
ing paths allowing for hysteresis so that subcritical crack growth becomes possible. Nguyen and co-workers
has worked out a one-parameter cohesive model for metals [13], which is able to capture experimental Paris
curves quite well and, in particular, the slope m of the curves (equal to approximately 3 for most metals).
However, this one parameter model cannot yield steeper Paris curves, thus precluding the glassy polymers
where the slope is at least 6. Another study deals with near threshold crack growth along a metal-rigid sub-strate interface and in a single crystal [5]. In this work, the unloading path is assumed to be parallel to the
previous loading path, thus leaving certain amount of residual separation in the cohesive zone after each
cycle. It is argued that the environment-assisted oxidation of the crack faces can give rise to this kind of
behavior. But this type of crack closure effect is not very common in polymers. The fatigue model for inter-
face cracks presented by Roe and Siegmund [17] is based on damage mechanics, where a history-dependent
damage parameter gives rise to irreversibility. In this model also, the unloading path is not toward the ori-
gin. Values of the slope m of the Paris curves up to 3.1 have been reported in that study. Finally, fatigue
crack growth in quasi-brittle materials has also been studied by cohesive techniques in [24], where the irre-versibility of the loading and unloading paths is taken into account. A polynomial expression for the cyclic
behavior is postulated in that study. A special case of Paris law, where the multiplicative constant is func-
tionally dependent on the maximum loading, has been reported by these authors.
In the present study, we formulate a new cohesive model specialized for polymers with higher fatigue
crack growth sensitivity on the range of cyclic loading. Special emphasis is placed on the mode I fatigue
failure of quasi-brittle glassy polymers such as PMMA, PC, epoxy etc. Our approach is inspired by the
work presented in [13,24,17], where all the nonlinear effects associated with the failure process are
accounted for in the cohesive zone model. We also describe the implementation of the cohesive model ina finite element framework (referred to as the Cohesive Volumetric Finite Element or CVFE scheme) allow-
ing for the solution of a large set of structural problems. For simplicity, the examples presented in the pre-
sent study involve its incorporation in a 1-D Euler–Bernoulli beam bending model of a double cantilever
beam (DCB) specimen.
The paper is organized as follows: To describe the cohesive model for fatigue failure, we start in Section 2
with a brief review of its foundation, i.e., a bi-linear cohesive model used to simulate quasi-static crack
propagation under monotonic loading, and its implementation in the cohesive finite element scheme. We
then discuss in Section 3 the formulation and implementation of the proposed cohesive model for fatiguecrack propagation in polymeric materials. Finally, we present in Section 4 a detailed discussion of the re-
sults associated with the proposed model, including a comparison with some experimental results obtained
on epoxy.
2. Cohesive model for monotonic loading
2.1. Cohesive failure law
The cohesive model that serves as the foundation for the proposed cohesive modeling of fatigue crack
propagation is characterized by a bi-linear, rate-independent, damage-dependent failure law between the
cohesive traction vector T and the displacement jump vector D acting across the cohesive surfaces Cc. This
particular cohesive model has been successfully used to simulate various fracture events in brittle media,
namely, the impact-induced delamination of composite materials [6], quasi-static fiber pushout in a model
composite [10], the instability of dynamic crack path in ceramic materials [11], and the dynamic fragmen-
tation of ceramics [12]. In the tensile (mode I) case, which is the focus of the work presented hereafter, thiscohesive model takes the simple form
T n ¼S
1�S
Dn
Dnc
rmax
Sinit
; ð1Þ
where Tn and Dn respectively denote the normal component of the cohesive traction and crack opening dis-
placement vectors, rmax is the tensile cohesive failure strength, and Dnc the critical opening displacement
jump (Fig. 1). The evolution of the damage process is quantified by the monotonically decreasing damageparameter S defined as
S ¼ min Sp; h1� Dn=Dnci� �
; ð2Þ
where hai = a if a > 0 and =0 otherwise, and Sp denotes the previously achieved S value.
As the material starts failing along its cohesive interface, the value of S gradually decreases from an
initial value Sinit (chosen close to unity) to zero, point at which complete failure is achieved. The monoto-nicity of S is ensured by (2): the accumulated damage is preserved and no healing of the cohesive zone
occurs due to unloading. As schematically indicated in Fig. 1, upon reloading, the cohesive stiffness main-
tains its most recent value until further failure is achieved (i.e., when the cohesive failure envelope is once
again reached).
A contact algorithm based on the penalty method has been incorporated in the numerical model to pre-
vent overlapping of the crack faces and thus simulate the crack closure effects. To that effect, the damage
Ic
loading
unloading
∆
T n
σmax
nc
∆ n
G
Fig. 1. Cohesive traction–separation law for tensile failure described by (1). After an unloading phase, reloading occurs along the
unloading path rather than the initial loading path, thereby preserving the previously achieved damage level.
T ndDn dCc|fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}cohesive
¼ 0; ð3Þ
where u is the displacement vector, S and E denote the internal stress and strain tensors, respectively, Tex is
the externally applied traction, and X, Cc and Cex respectively denote the volume, cohesive boundary andexterior boundary of the deformable body. The last term in (3) corresponds to the virtual work done by
cohesive traction Tn for a virtual separation dDn and is associated with the contribution of the cohesive ele-
ments. The expression of the internal component of the virtual work (first term in (3)) depends on the type
of volumetric element used in the analysis, which, in turn, affects the expression of the stress S and virtual
strain dE. As mentioned in the introductory section, we use in the present study classical 2-node Euler–
Bernoulli beam elements as volumetric elements since the primary focus of this work is the simulation of
fatigue delamination failure in a double cantilever beam (DCB) specimen. The corresponding finite element
formulation can be found in Appendix A.To ensure the spatial convergence of the CVFE scheme, it is critical to capture the fracture process accu-
rately, i.e., to have a sufficient number of cohesive elements in the active cohesive zone. A rough estimate of
where E is the Young�s modulus, GIc the mode I fracture toughness (=rmaxDnc/2 for the bi-linear cohesive
model described earlier), and rave the average stress in the cohesive zone (taken as rmax/2 for the bi-linear
cohesive law). Our experience with cohesive element modeling suggests that at least 4 cohesive elements
must be present in the active cohesive zone.
2.3. Verification and validation
To verify the CVFE scheme for crack propagation in a DCB specimen subjected to monotonic loading,
we compare our numerical results to the theoretical predictions obtained in the limiting case of a vanish-
ingly small cohesive zone size. For reference, the analytical expression of the evolution of the crack length
a and load P for a displacement-controlled DCB specimen (of initial uncracked length b0 and initial crack
length a0) can be found in Appendix B. In the example shown hereafter, b0 = 100mm, the Young�s modulus
E = 3.4GPa, the fracture toughness GIc = 88.97J/m2, and the cohesive strength rmax = 35MPa. In Fig. 2,the reaction force developed in the beam at the loading point is plotted against the applied displacement
D for three values of the initial crack length a0, namely 50, 100 and 200mm. Excellent agreement is achieved
between numerical and theoretical curves, especially for the larger value of a0 for which the assumptions
underlying the theoretical solution are increasingly applicable. This agreement between theoretical and
numerical predictions is also observed in the evolution of the crack length presented in Fig. 3. As predicted
by the theory, the crack remains stationary until sufficient energy is stored in the deformed beam elements
to allow for crack propagation. Once again, the longer the initial crack, the better this agreement.
Despite its simplicity, the CVFE model of DCB delamination can be used very efficiently to model theprogressive failure of actual DCB specimens. To validate the CVFE scheme, we choose to model the mode I
delamination experiments performed by Borg and co-workers [1] on HTA/6376C composite DCB speci-
mens. The corresponding constitutive and failure properties are E = 146GPa and GIc = 259J/m2 as reported
in their paper. For the simulation of this problem, the number of cohesive elements ahead of the pre-crack
is chosen as 220. The cohesive strength rmax is taken to be 35MPa, and, as shown in Fig. 4, the cohesive
finite element model is able to capture quite well the evolution of both the reaction force and the crack
length (a0 = 35mm).
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.040
100
200
300
400
500
600
700
800
900
1000
theoretical
a0 = 200
a0 = 100
a0 = 50
∆ (mm)
P(N
)
mm
mm
mm
Fig. 2. Comparison between theoretical (dashed curve) and numerical (solid curves) evolution of the reaction force P at the tip of a
DCB specimen under prescribed displacement D, for 3 values of the initial crack length a0.
0 5 10 15 20 25 30 350
50
100
150
200
250
∆ (mm)
a (m
m)
theoretical
a0 = 200 mm
a0 = 50 mm
a0 = 100 mm
Fig. 3. Crack length a versus prescribed displacement D: comparison between numerical (solid curves) and theoretical (dashed curve)
results obtained for 3 values of the initial crack length a0.
0 1 2 3 4 5 60
10
20
30
40
50
60
∆ (mm)
Cra
ck E
xten
sion
(m
m)
P (N
)
60
50
40
30
20
10
0
experimental reaction
computed reaction
computed crack extension
experimental crack extension
Fig. 4. Delamination of a composite DCB specimen: experimental measurements [1] and numerical predictions of the reaction and
After describing, verifying and validating the cohesive model for mode I crack propagation under mono-
tonic loading, we now turn our attention to the main topic of the present paper: the formulation and imple-mentation of a cohesive model for fatigue crack propagation in polymers. As mentioned earlier, although it
prevents healing of the fracture surfaces through the enforcement of monotonic decay of the damage
parameter S, the cohesive model discussed in the previous section leads to similar unloading and reloading
paths in the traction–separation curve. This characteristic prevents crack growth under subcritical cyclic
loading due to the progressive degradation of the cohesive properties in the cohesive failure zone. This lim-
∆nc ∆n
σ
Tn
max
Fig. 5. Schematic evolution of the cohesive stiffness during cyclic loading, assuming unloading towards the origin and degradation of
itation suggests the need for an evolution law to describe the changes incurred by the cohesive strength
under fatigue [24,17,13,5,4].
As mentioned earlier, complex physical phenomena take place in the craze (or cohesive) zone ahead of
the advancing crack with the appearance of irreversible processes associated with the creation, extension
and failure of fibrils. A phenomenological model of such processes involves the progressive degradationof the cohesive zone strength during reloading events as illustrated schematically in Fig. 5.
The evolution law of the instantaneous cohesive stiffness kc, i.e., the ratio of the cohesive traction Tn to
the displacement jump Dn, during reloading can be expressed in the general form
kc ¼dT n
dDn
¼ FðN f ; T nÞ; ð4Þ
where Nf denotes the number of loading cycle experienced by the material point since the onset of failure,
i.e., at the time the cohesive traction Tn first exceeded the failure strength rmax. Simplifying (4), we adopt
the separable form
dT n
dDn
¼ �cðN fÞT n; ð5Þ
leading to an exponential decay of the cohesive strength, with the rate of decay controlled by the parameter
c. In the present study, we use the following two-parameter power-law relation
c ¼ 1
aN�b
f ; ð6Þ
where a and b are material parameters describing the degradation of the cohesive failure properties. The
physical significance of these two parameters is discussed in Section 4 in terms of their effect on the resulting
Paris� fatigue curve. Note that the parameter a has the dimension of length and b denotes the history
dependence of the failure process. When it vanishes, we retrieve the one-parameter model of Nguyen
The proposed evolution law for the cohesive model can also be expressed in terms of the rate of change
of the cohesive stiffness _kc as follows:
_kc ¼ � 1
aN�b
f kc _Dn if _Dn P 0;
¼ 0 if _Dn 6 0;
ð7Þ
where _Dn is the rate of change of the normal separation. The second equation simply states that the cohesivestiffness is assumed to remain constant during unloading.
3.2. Implementation issues
During the reloading phase, (7) can be described in its discretized form as
kjþ1c � kjc ¼ � 1
aðNj
fÞ�bkjcðDjþ1
n � DjnÞ;
where the superscripts j and j + 1 stand for loading steps j and j + 1, respectively. Recalling the expression
for kc obtained from (1), the updated value of the cohesive stiffness kjþ1c can be written as
kjþ1c ¼ Sj
1�Sj
1
Dnc
rmax
Sinit
1� 1
aðNj
fÞ�bðDjþ1
n � DjnÞ
� �; Djþ1
n P Djn: ð8Þ
A quasi-implicit load stepping scheme is used in this investigation to form the equations of equilibrium. In
this simple DCB implementation, the stiffness matrix arising from the volumetric (beam) elements is con-
stant throughout the simulation. The only nonlinear component of this problem is associated with the cohe-
sive component of the stiffness matrix, which is updated at each load step based on the displacement jump
values achieved at the previous loading step.
As shown in relation (8), during the reloading phase, the cohesive stiffness at each material point along
the cohesive zone gradually reduces proportionally to the increment in crack opening displacement. This
proportionality factor c evolves with the cycles to failure Nf in accordance with (6) and thus gives a measureof the total accumulated damage in the failure process. Note that the second term in the square-bracketed
quantity in (8) can be viewed as a damage parameter which depends on the increment of deformation and
starts accumulating only after the parameter S goes below its initial valueSinit. Therefore, the material can
cycle infinitely without failure when the cohesive traction is below its critical value rmax. The basic tenets of
damage evolution laws are thus also satisfied [9].
To further elaborate on the difference between subcritical and critical failure, we consider in Fig. 6 the
evolution of the failure process taking place in the cohesive zone ahead of a delamination crack propagating
in a DCB specimen subjected to a prescribed sinusoidal displacement loading. The DCB specimen is150mm long and has an initial crack length a0 equal to 50.1mm. The cyclic displacement D applied at
the end of the beam has a maximum amplitude of 1.5mm and the ratio of minimum to maximum amplitude
of loading R = 0. The specimen is made of epoxy with a Young�s modulus E = 3.9GPa, and a mode I frac-
cohesive strength rmax is taken as 35MPa and the values of the two fatigue parameters a and b are chosen
as 4lm and 0.3 respectively. Since the estimated cohesive zone size is approximately 0.45mm, the cohesive
element size is chosen as 0.1mm. Fig. 6 represents the evolution of the failure process at a point located
ahead of the pre-existing crack, more precisely, at the location of the first Gauss point of the tenth cohesiveelement. During the initial phase of the failure process, the stress concentration present at the crack tip
leads to a critical fatigue failure, as the material degradation curve follows the envelope of the bi-linear fail-
ure law. The failure process then becomes subcritical as the degradation leads to peak cohesive traction val-
ues well below the critical value (denoted by the dashed curve). In the event of an overload, the fatigue
0 0.2 0.4 0.6 0.8 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
∆ n /∆ nc
Tn/σ
max
15 20 25–1
0
1
2
x 105
Cycles
∆(m
)
Fig. 6. Subcritical versus critical failure: evolution of the traction–separation curve for a point located in the path of a fatigue crack.
This plot illustrates the initial critical failure followed by subcritical degradation of the cohesive properties, with a transient critical
cycle associated with an overload (inset) in applied displacement.
failure may become critical again and follow the failure envelope until the next unloading phase, as illus-
trated in Fig. 6. In all cases, the unloading portion of the cohesive failure curve points to the origin.
4. Results and discussion
4.1. Characteristics of fatigue crack propagation
In this section, we comment on some basic characteristics of the fatigue crack propagation results ob-
tained from the numerical model developed in Section 3. As a basis for this discussion, we consider the
fatigue failure of the DCB specimen whose geometry and constitutive and failure properties have been de-scribed at the end of Section 3.2. The cyclic displacement applied at the tip of the beam has the same ampli-
tude (1.5mm) but its minimum value is chosen as 0.15mm, resulting in an amplitude ratio R = 0.1. In this
initial simulation, the parameters a and b are taken as 4lm and 0.3 respectively.
Fig. 7 presents the evolution of the crack tip and cohesive zone tip locations as a function of the number
of loading cycles. As expected, in this prescribed displacement problem, the crack driving force decreases
with crack advance, and the crack extension rate decreases with the number of loading cycles giving rise to
a stable crack growth. The cohesive zone tip follows a similar evolution, but, as the crack extension rate
decreases, the cohesive zone length increases significantly, from an initial size of 1.1mm to a final size of2.4mm.
Using the analytical solution of the DCB fracture problem summarized in Appendix B, the crack length
can be related to the applied stress intensity factor K, in terms of its maximum applied value (Kmax) or its
range (DK). As mentioned earlier, the proposed model leads to subcritical material degradation character-
ized with energy levels well inferior to the (monotonic) fracture toughness GIc. This fact is illustrated in Fig.
8, which presents the variation of the expended energy G (i.e., the area under the fatigue cohesive failure
curve) with the applied Kmax value (normalized by KIc). As apparent in that figure, the fracture toughness
0 1000 2000 3000 4000 500050
55
60
65
70
75
80
Cycles
Cra
ck a
nd c
ohes
ive
zone
tip
loca
tion
(
mm
)
crack tip
cohesive zone tip
Fig. 7. Evolution of crack and cohesive zone tip locations with number of loading cycles, a = 4lm and b = 0.3.
0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
G/G
IC
Kmax /KIc
0.4 0.5 0.6 0.7 0.8 0.9 10
1000
2000
3000
4000
5000
6000
7000
Cyc
les
to fa
ilure
Fig. 8. Variation of the fatigue fracture toughness G and the number of cycles to failure with the applied Kmax, for a = 4lm and
b = 0.3. G and Kmax both are normalized by their monotonic counterparts, GIc and KIc, respectively.
under fatigue loading can represent a very small fraction of its monotonic counterpart, with values inferior
to 5% of GIc for Kmax < 0.4KIc. For higher values of the applied cyclic load, the required energy for failure
increases rapidly, especially as the applied Kmax approaches 0.9KIc. The figure also illustrates the number of
cycles to failure as a function of Kmax/KIc for the same fatigue crack propagation problem, showing a rap-
idly decaying trend down to zero for Kmax = KIc.
4.2. Effect of the parameters on fatigue crack propagation
We now turn our attention on the effect of the parameters a and b on the predicted fatigue failure. Fig. 9
presents the evolution of the total crack extension Da with the number of loading cycles for three values of
a, with the parameter b kept constant at 0.3. As apparent there, the crack growth rate decreases with
increasing a: for a = 4lm, the crack extends by 26mm after 4000 loading cycles, whereas it has extended
by only 20mm and 17mm for a = 6lm and 8lm, respectively. Differentiating these curves, we can charac-terize the effect of a on the fatigue response in terms of the evolution of the crack advance per cycle (da/dN)
0 1000 2000 3000 4000 5000 6000 7000 800050
55
60
65
70
75
80
Cycles
Cra
ck e
xten
sion
(m
m)
∆f = 4 µ m
∆f = 6 µ m
∆f = 8 µ m
Fig. 9. Evolution of the crack extension with the number of cycles for three values of a, with b = 0.3.
versus the applied range of stress intensity factor, DK (normalized with KIc), better known as the Paris fa-
tigue failure law (Fig. 10). As apparent in that figure, all the curves have the same slope but different inter-cepts, suggesting that a unit change in DK produces an equal amount of change in the crack growth rate,
though the actual value of da/dN can be different.
10–0.4
10–0.3
10–0.2
10–0.1
10–3
10–2
∆ K/KIC
da/d
N
(m
m/c
ycle
)
α = 4 µm
α = 6 µm
α = 8 µm
Fig. 10. Effect of the parameter a on the fatigue curve, showing that a affects the intercept but not the slope of the curves (b = 0.3).
0 2000 4000 6000 8000 10000 12000 14000 1600050
55
60
65
70
75
80
85
Cycles
Cra
ck e
xten
sion
(m
m)
β = 0.0
β = 0.3
β = 0.5
Fig. 11. Effect of b on the evolution of the crack extension (a = 8lm).
The effect of the parameter b on the fatigue crack propagation is more pronounced, as illustrated byFig. 11, which presents the crack extension versus loading cycles for b = 0.0, 0.3 and 0.5, with a kept con-
stant at 8lm. As apparent in that figure, b affects the crack growth rate, leading to Paris failure curves
showing very different slopes (Fig. 12). This parameter plays therefore a key role in capturing the fatigue
response of polymeric materials, where the rate of crack growth per cycles may vary substantially with
the crack driving force from material to material. The effect of the parameter b on the slope m of the Paris
fatigue failure curve is summarized in Fig. 13: for lower values (b < 0.4), the predicted value for the expo-
nent m is about 3, which is characteristic of the fatigue failure of metals. As b increases, increasing values of
the exponent m of the Paris fatigue law are obtained.
4.3. Fatigue failure of epoxy—comparison with experiments
To conclude this result discussion, we simulate in this last section the fatigue crack growth in epoxy and
compare it with experimental results [2]. As many polymeric materials, epoxy has a very high sensitivity of
the crack growth rate with the change in stress intensity factor, hence a large slope of the Paris curve. The
slope m of the Paris fatigue curve for epoxy is about 10 compared to about 3 for most metals. So the change
in fatigue crack growth rate is strongly correlated with the crack driving force. The experimental values ofthe crack growth rate da/dN were obtained by Brown [2] with a tapered cantilever beam specimen made of
neat EPON 828 epoxy resin with 12 pph Ancamine DETA (diethylenetriamine) curing agent subjected to a
range of cyclic loading conditions. Owing to the tapered configuration, the stress intensity factor remained
constant during the experiments. Therefore, every test led to a single data point on the da/dN versus DKplot. In our numerical simulations, we use a cantilever beam with constant cross section so that the stress
intensity factor varies continuously with the propagating crack, allowing us to obtain in one simulation the
entire fatigue failure curve. The maximum amplitude of the applied displacement Dapplied is taken to be
1.8mm resulting in an initial value of the maximum stress intensity factor 0.6MPaffiffiffiffim
p, so that the crack
quickly establishes itself. Once the crack starts growing, the stress intensity factor goes down and fatigue
crack propagation is established. The load ratio R is taken to be 0.1.
Fig. 14 presents the comparison between the experimental points and numerical simulations. The oscil-
lations apparent in the da/dN versus DK curve for low values of DK are associated with the difficulty of
differentiating the crack evolution curve for very small crack advances. The two parameters of the cohesive
fatigue model are taken to be a = 8lm and b = 0.65. The value of b has been chosen based on the results
10–0.4
10–0.3
10–0.2
10–4
10–3
10–2
10–1
∆ K/KIC
da/d
N
(m
m/c
ycle
)β = 0.0
β = 0.3
β = 0.5
Fig. 12. Effect of b on the Paris fatigue curve, with a = 8lm.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.72
4
6
8
10
12
14
16
18
20
22
Parameter β
Slo
pem
Fig. 13. Variation of slope of the Paris fatigue curve m with the parameter b.
presented in Fig. 13. The proposed two-parameter model captures the entire fatigue failure response curve
quite well, including in the higher crack growth rate regime, where the numerical results show the rapidacceleration of the fatigue crack for DK/KIc > 0.8.
1.00.4
10–5
10–4
10–3
∆ K/KIC
da/d
N
(m
m/c
ycle
)
Fig. 14. Comparison between experimental (symbols) and computed (solid curve) fatigue response curves for epoxy. The experimental
For the geometry and material parameters used in the DCB problem described in Section 3, with R = 0.1
and Sinit ¼ 0:98, the computed value of DK th=KIc ¼ 0:09MPaffiffiffiffim
p. Due to the high fatigue sensitivity of
epoxy (m � 10), this threshold value corresponds to an extremely small value of the crack extension per
cycle (da/dN < 10�6mm/cycle), making its numerical capture excessively expensive. However, if we consider
a material system with a = 4lm and b = 0.3 (i.e., m � 3), the cohesive modeling of the fatigue crack prop-
agation yields the results shown in Fig. 15, which clearly illustrates the near-threshold behavior. As DK/KIc
decreases and approaches the threshold value (DKth = 0.09KIc) indicated by a vertical arrow, the crack ad-
vance slows down to a halt. The difference between theoretical and numerical values of DKth is due to
numerical error associated with the finite element approximation of the beam on (nonlinear) elastic
foundation.
5. Conclusion
We have presented a cohesive failure model specially developed for the simulation of fatigue crack prop-
agation in polymeric structures subjected to mode I cyclic loading. The phenomenological model is based
on a two-parameter evolution relation between the cohesive stiffness, the crack opening displacement rate
and the number of cycles since the onset of failure. The fatigue evolution law has been combined with a
bi-linear cohesive model for crack propagation under monotonic loading. The two parameters (a and bin (7)) entering the evolution law define the irreversible degradation process during each reloading cycle
and can be readily calibrated based on the slope and intercept of the Paris fatigue failure curve character-
izing the material.
The semi-implicit implementation of the cohesive model into a cohesive-volumetric finite element
(CVFE) framework has also been described. Although the simulations presented in the present paper have
used a simple 1-D CVFE scheme combining cohesive elements with conventional Euler–Bernouilli beamelements, the formulation presented here applied to other volumetric finite elements and therefore allows
for the simulation of a wide range of fatigue failure problems.
The cohesive model has been shown to be quite successful in capturing the fatigue response of an epoxy,
not only within the range of loading amplitudes where the Paris fatigue law is applicable, but also with re-
gards to the low and high amplitude regimes.
Acknowledgments
The authors gratefully acknowledge funding for this project provided by AFOSR Aerospace and Mate-
rials Science Directorate (Grant # F49620-02-1-0080). They also wish to thank Dr. Eric Brown and Prof.
Nancy Sottos for helpful discussions.
Appendix A. Cohesive element formulation
The expression of the local stiffness matrix kcoh of the cohesive element for this DCB problem is derived
from the cohesive component of the principle of virtual work described by (3). Let d denote the vector
containing the nodal deflection (w1 and w2) and rotation (h1 and h2) degrees of freedom associated with
the 2-node Euler–Bernouilli (C1 continuous) element. Only the upper half of the DCB specimen has been
considered for computations due to symmetry. The normal displacement jump or crack opening displace-
ment along that cohesive element (of length l) is approximated by
DnðsÞ ¼ 2d �N ; ðA:1Þ
where the vector N contains the four cubic shape functions of the beam element. The introduction of the
cohesive failure law described by (1) leads to
kcoh ¼ 2
Z l
0
NTkcN dCc; ðA:2Þ
where kc denotes the cohesive stiffness and, for the bi-linear cohesive relation (1), is given by
kc ¼S
1�S
1
Dnc
rmax
Sinit
: ðA:3Þ
A 3-point Gauss quadrature scheme is used to evaluate the integral appearing in (A.2).
Appendix B. LEFM solution for a DCB
We summarize here the classical linearly elastic fracture mechanics (LEFM) solution for the delamina-
tion of a double cantilever beam (DCB) specimen with an initial crack length a0 and subjected to an end
displacement 2D. This solution is used in Section 2 to verify the CVFE scheme in the limit of a very small