Professional Documents
Culture Documents
net/publication/241653305
CITATIONS READS
19 2,252
6 authors, including:
Some of the authors of this publication are also working on these related projects:
Analysis and Optimization of Omega Composite Stiffener Under Impact View project
All content following this page was uploaded by Xiushan Sun on 17 March 2015.
A multi-axial fatigue model for fiber-reinforced composite laminates based on Puck's criterion
X.S. Sun, A. Haris, V.B.C. Tan, T.E. Tay, S. Narasimalu and C.N. Della
Journal of Composite Materials 2012 46: 449
DOI: 10.1177/0021998311418701
Published by:
http://www.sagepublications.com
On behalf of:
Additional services and information for Journal of Composite Materials can be found at:
Subscriptions: http://jcm.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://jcm.sagepub.com/content/46/4/449.refs.html
What is This?
Abstract
A new multi-axial fatigue model for fiber-reinforced composite laminates based on Puck’s criterion is proposed in this
article. In the fatigue model, fatigue master curves from the ATM are used to determine the uniaxial ply fatigue strengths
and the multi-axial fatigue failure is then determined by Puck’s criterion with the fatigue strengths at the ply level. The
fatigue master curves from ATM are generated with limited uniaxial fatigue tests and can be applied to fatigue loading
conditions with various frequencies and stress ratios. Both uniaxial and multi-axial S-N curves can be derived from
the fatigue model. Fatigue failure envelopes are also generated from the model to better interpret the multi-axial fatigue
failure in multi-axial stress spaces. The proposed multi-axial fatigue model is based on ply-level predictions, but it can
be extended to laminate-level predictions with the CLT or numerical methods such as the FEM. Multi-axial fatigue failures
caused by either local or global multi-axiality can be predicted by the model. Both uniaxial and biaxial fatigue experi-
ments were carried out to provide test data for establishing and validating the proposed fatigue model. The application
of the proposed multi-axial fatigue model is demonstrated with predictions of S-N curves and fatigue failure envelopes of
unidirectional laminates and multi-directional laminates with typical lay-up configurations. The predictions from the pro-
posed fatigue model are also compared with various experimental results and reasonably good agreement is observed.
Keywords
Fiber-reinforced composite, multi-axial fatigue, Puck’s criterion, fatigue master curve, fatigue failure envelope, biaxial
fatigue test
However, most of these models are formulated for and Philippidis10 proposed a statistical fatigue model
uniaxial loading and not extendable to multi-axial load- based on a strength degradation scheme named
ing conditions. In the fatigue failure models based on S- Strength-Life Equal Rank Assumption (SLERA),
N curves, the fatigue strengths are determined from S- where several degradation models are studied and
N curves which must be determined by experiments large experimental data sets are required for implemen-
with defined load frequency and stress ratio. If struc- tation of the complicated model.
ture or load information changes, S-N curves usually An example of a progressive damage approach is
need to be redetermined by more experiments for this given by Shokrieh and Lessard11–14 who proposed a
kind of fatigue models. A typical example is a semi-log generalized residual material property degradation
S-N curve2,3 with linear relationship between the model for fiber-reinforced composite laminates. They
applied cyclic stress and the number of cycles to failure. used experimental data from an off-axis UD laminate
The multi-axiality is indicated by two axial ratios in the under uniaxial fatigue in order to predict the behavior
formula for the curve. A similar S-N curve was also of a ply in multi-axial fatigue environment. This
adopted by Philippidis and Vassilopoulos4 to fit fatigue model is complex and integrates stress analysis, failure
strength data with a Tsai-Wu-based fatigue failure analysis, and material property degradation rules.
criterion under multi-axial stress. Huang5 used the max- Diao et al.15 incorporated statistics into the above
imum normal stress criterion with S-N curves for a plain- model to study the fatigue failure of fiber-reinforced
woven fabric reinforced composite subjected to biaxial composite laminates subjected to multi-axial loading,
cyclic loads. Recently, Gude et al.6 modified Puck’s cri- where a Weibull distribution function of material
terion with a group of semi-log S-N curves obtained from strengths was fitted by experimental data. In general,
linear regression analysis of various test data for predic- such a model, however, requires more distribution
tion of multi-axial fatigue failure of carbon/epoxy functions corresponding to the stochastic variables con-
tubular composites under tension/compression-torsion sidered. Liu and Mahadevan16 proposed another S-N
fatigue loads. curve-based fatigue progressive damage model for
The phenomenological residual stiffness/strength multi-directional (MD) composite laminates.
models indirectly qualify the damage accumulation However, only constant amplitude fatigue loading
during fatigue. These fatigue models are usually more was considered in this model, and the model was not
complicated, ranging from mainly empirical curve fits to validated for different stress ratios. Quaresimin et al.17
more sophisticated damage evolution functions linking recently assessed several typical multi-axial fatigue
residual stiffness to damage variables and loading cycles. models and proposed an approach based on damage
An example is due to Kawakami et al.7 who proposed a mechanics to develop life predictions for multi-axial
fatigue damage accumulation model for plain woven fatigue of composites, where failure mechanisms
glass fabric polymer composite based on continuum including fiber failure (FF), rectangular cracking and
damage mechanics, where the modulus decay ratios in matrix damage from shear deformation are considered.
tension and shear were used as indicators for damage Passipoularidis et al.18 also developed a progressive
variables. Under proportional combined tension- damage model for fatigue life prediction of composites
torsion loadings, this model is reported to be able to under block and spectrum loading, where Puck’s crite-
predict fatigue lives for different biaxial stress ratios. rion was used to determine fatigue failure and uniaxial
Van Paepegem and Degrieck8 presented a phenomeno- and/or multi-axial load time series can be simulated.
logical residual stiffness model that provides the stiffness Fatigue models based on progressive damage analysis
degradation and permanent strain under generalized in- are generally more computationally expensive and
plane fatigue loading. This model takes into account the complicated.
actual stress state, but complicated damage growth rate All the above three categories of fatigue models
equations, which generally vary with different materials, are based on traditional fatigue tests, which generally
are used to interpret damage evolutions. Extensive require very costly and substantial amounts of testing
experiments and numerical simulations are required to at numerous load ratios and frequencies to generate
validate the model. Kawai9 developed a phenomenolog- huge experimental fatigue failure databases for the
ical fatigue model for the off-axis fatigue behavior of determination of the required empirical parameters.
unidirectional (UD) polymer matrix composites, where This article proposes a multi-axial fatigue model based
a new non-dimensional effective stress that considers the on fatigue master curves and Puck’s criterion. Only lim-
effects of the stress ratio and multi-axial state of stress is ited uniaxial fatigue tests of UD laminates are required to
defined for the evolution of fatigue damage. However, generate the master curves and the multi-axial fatigue
an empirical fatigue strength ratio was used in this failure is determined by Puck’s criterion with uniaxial
model, which requires fatigue experimental data to ply fatigue strengths obtained from the fatigue master
determine the appropriate quantity. Passipoularidis curves. The traditional S-N curves can be derived and
generated from the proposed fatigue model. In the certain values of stress ratio R ¼ min =max and a gener-
following sections, the proposed multi-axial fatigue alized fatigue master curve for different stress ratios can
model is introduced in detail, and then uniaxial and be obtained by introducing a coefficient gf such that:
biaxial fatigue experiments are carried out to provide
necessary test data for the proposed model. Finally,
ð2ftÞnc
predictions of S-N curves and fatigue failure envelopes log f ¼ log s,0 log 1 þ gf gf logð2ftÞnf
ð2ft1 Þnr
under multi-axial fatigue loadings are demonstrated and
compared with test data for both UD and MD laminates ð2aÞ
with typical lay-up configurations. or the normalized fatigue master curve:
Figure 1. Normalized fatigue master curves: (a) for different frequencies and (b) for different stress ratios.
Equivalent fatigue stress and multi-axial S-N curves Equation (18), the multi-axial S-N curve which
The S-N curve defined in Equation (8) cannot be accounts for the effect of all fatigue stress components,
directly used in multi-axial fatigue loadings where eq, f ðNf Þ ¼ eq, f ðrf ðNf ÞÞ, can be obtained through the
multiple fatigue stress components must be considered. equivalent fatigue stress defined by Equation (15).
Therefore, a method to determine the multi-axial S-N However, this multi-axial S-N curve cannot be explic-
curve is further developed with the above mentioned itly formulated but is usually numerically obtained
fatigue model. Towards this end, a scalar equivalent from Equations (15) and (18). It can be verified that
fatigue stress eq is defined with a general multi-axial the multi-axial S-N curve determined from Equations
stress r or multi-axiality ratios i as the following: (15) and (18) can be reduced to the uniaxial one explic-
itly defined in Equation (8) under uniaxial loading.
pffiffiffiffiffiffiffiffi rffiffiffiffiffiffiffiffiffiffiffiffi
X ffi
eq ¼ rT r ¼ jN j 2i ð15Þ In the current multi-axial fatigue model, the multi-
i axial fatigue stress rf ðR,f,Nf Þ corresponds to
the stress state causing fatigue failure and therefore
where it is the multi-axial fatigue strength Rf , that is,
8
Rf ¼ rf ðR,f,Nf Þ and eq, f ¼ eq, f ðR,f,Nf Þ. The multi-
T
>
> 1 2 3 12 23 13 , axial fatigue strength Rf is generally used in the context
>
>
< of a given group of multi-axiality ratios.
if UD laminates or plies
r¼
T ð16Þ
>
> x y z xy yz xz ,
>
>
: Implementation of the multi-axial fatigue model
if MD laminates
An Excel-based user-friendly interface was developed
i ¼ i =jN j ð17Þ for implementing the multi-axial fatigue model and fati-
gue analysis of UD or MD laminates. With the input
Here, the superscript ‘T’ denotes transpose of an ply material properties (Young’s moduli and static
array; i ¼ 1, 2, 3, 12, 23, 13 (with 12 ¼ 12 , 23 ¼ 23 , strengths, etc.) and fatigue load information (stress
and 13 ¼ 13 ) for UD laminates or plies (the coordi- ratio, frequency, multi-axiality ratios, number of
nates system is defined in local material directions), and cycles or time to failure, etc.), the multi-axial fatigue
i ¼ x,y,z,xy,yz,xz (with xy ¼ xy , yz ¼ yz , and strength Rf for a given group of multi-axiality ratios
xz ¼ xz ) for MD laminates (the coordinates system and a number of cycles or time to failure can be com-
is defined in global reference directions); N is the first puted from the fatigue model. Fatigue failure envelopes
non-zero stress component of r, for example, N ¼ 1 if for a given number of cycles or time to failure and S-N
1 6¼ 0 for UD laminates. It should be noted that the curves for a given group of multi-axiality ratios can also
definition of multi-axiality ratios i in Equation (17) be generated and plotted. The flowchart for implement-
implies a positive or negative value of the correspond- ing the multi-axial fatigue model is illustrated in
ing stress component i , which corresponds to tension Figure 3. The fatigue master curves can be generated
or compression in some loading cases. A group of from limited uniaxial fatigue tests and the correspond-
multi-axiality ratios ½ comprises the multi-axiality ing parameters can be stored for analysis of the same
ratios i covering all stress components in a given composite materials. For simplicity but without loss of
stress space. generality, the fatigue failure envelope for a number of
Under multi-axial fatigue loading r ¼ rf ðR, f, Nf Þ, cycles or time to failure is plotted with the same load
the equivalent fatigue stress can be determined from amplitude ratio r for all loading axes but possible
Equation (15) as eq ¼ eq, f ðR, f, Nf Þ. The multi-axial different stress ratio R, because the multi-axial loads
fatigue stress rf ðR, f, Nf Þ can be obtained with the may alternate between tension and compression for
above mentioned fatigue model and given multi-axiality varying multi-axiality ratios. Nevertheless, this is not
ratios, that is: a prerequisite for the multi-axial fatigue model and
different load amplitude ratios r for loading axes may
FP i, f , j, f , R, f, Nf ¼ 0 be considered in the model.
ð18Þ
i, f : j, f ¼ i : j For UD laminates, the multi-axial fatigue strength
Rf is directly determined by Puck’s criterion with uni-
where FP is the function of Puck’s criterion under axial fatigue strengths obtained from the fatigue master
fatigue loadings; i, j ¼ 1, 2, 3, 12, 23, 13ði 6¼ j Þ with curves. For MD laminates, the multi-axiality ratios
fatigue stress components 12, f ¼ 12, f , 23, f ¼ 23, f , are based on global reference directions and should
and 13, f ¼ 13, f . be transformed to the local multi-axiality ratios in
Once the multi-axial fatigue stress for a given stress the ply material directions, and the stress ratio R is
ratio R and load frequency f, rf ðNf Þ, is determined with same for local ply directions and global reference
directions. Fatigue failure predictions are based on the the simultaneous degradation scheme is used.
last-ply-failure (LPF) theory, as shown in Figure 4. Nevertheless, the proposed model is straightforward
In determining the multi-axial fatigue strength Rf of to laminates with non-uniform stress distribution
MD laminates by Puck’s criterion based on LPF, (such as laminates with an open hole) with progressive
usually a progressive degradation based on ply-by-ply damage analysis based on numerical methods such as
discount is required to calculate the degraded laminate FEM. In the simultaneous degradation scheme, three
stiffness after first-ply-failure (FPF). For laminates with potential failure patterns are considered, i.e. the FPF
uniform ply stress distribution, however, a simulta- pattern where the laminate stiffness D is based on intact
neous degradation scheme can be used to simplify the materials, the LPF-matrix pattern where the laminate
failure analysis.23 In this article, only laminates with stiffness D m is based on materials with degraded matrix,
uniform ply stress distribution are considered for dem- and the LPF-fiber pattern where the laminate stiffness
onstrating the multi-axial fatigue model, and therefore D f is based on materials with degraded fibers. A coef-
ficient dm ¼ df ¼ 0:01 is used in the degradation of
matrix or fiber material to calculate the degraded lam-
inate stiffness (D m or D f ). The degradation of matrix or
fibers is simultaneous for all plies, and therefore no
iteration is required in the computation. The stress dis-
tribution based on the laminate stiffness D or the stress
redistribution based on the degraded laminate stiffness
(D m or D f ) in each ply is calculated by Puck’s criterion
with CLT for the three failure patterns, respectively.
The laminate failure stress for each pattern is the min-
imum value of failure stresses in all plies, and the final
laminate failure stress (the multi-axial fatigue strength)
is the maximum value of failure stresses from the three
patterns.
The current fatigue model is focused on predicting
the ultimate laminate strength under fatigue loading,
and only constant amplitude fatigue loading is consid-
ered in the article. Although the variable amplitude
fatigue may arise due to stress redistributions caused
by ply failures and stiffness degradation for MD
laminates, the amplitude can be assumed to be constant
for each failure pattern with the simultaneous degrada-
tion scheme (Figure 4), and the final failure stress or
strength is determined by the maximum value corre-
Figure 3. Flowchart for implementing the multi-axial fatigue sponding to the stress state in plies.
model.
Fatigue experiments
Both uniaxial and biaxial fatigue tests were carried
out under constant stress amplitude using a universal
testing machine INSTRON 8501. The uniaxial fatigue
tests are used to generate fatigue master curves for the
proposed multi-axial fatigue model and the biaxial fati-
gue tests are used to demonstrate the applicability of
the model.
material properties and static strengths for the current The limited uniaxial fatigue tests of glass/epoxy UD
fatigue tests which are used to establish and demon- laminate [0]8 were carried out for three values of stress
strate the multi-axial fatigue model. The laminates were ratio R ¼ 0:1, 10, 1, respectively. The geometry and
manufactured in infused glass fibers with fiber volume dimensions of the fatigue test coupons are similar to
fraction Vf ¼ 65%. The geometry and dimensions (in those for static tests, as shown in Figure 5, with
mm) of the UD test coupons with end tabs are shown L ¼ 40 mm for R ¼ 0.1 and L ¼ 20 mm for R ¼ 10, 1;
in Figure 5. The coupons were cut from UD laminate W ¼ 11 mm for all uniaxial fatigue test coupons. To get
panels using water jet cutting method. For longitudinal good bonding surfaces, peel plies of the coupons and
compressive tests, L ¼ 20 mm and W ¼ 25 mm; for end tabs were removed just before the bonding, and a
longitudinal tensile tests, L ¼ 40 mm and W ¼ 11 mm, proper adhesive with high durability (3M-DP420
where the width of the test coupons was reduced from black) was used to bond the coupon and tabs in the
the standard 25 mm to the current 11 mm due to the fatigue tests. All the coupons were tested using
load cell capacity (100 kN) of INSTRON 8501. INSTRON 8501 under sinusoidal loads with frequency
Nevertheless, a comparison test has been performed f ¼ 4Hz. The uniaxial test data obtained for R ¼ 0:1 and
and the results show that both Young’s modulus and R ¼ 10 are used to generate the fatigue master curves
strength of the coupons with width 11 mm are consis- (as shown in Figure 9). The uniaxial fatigue test
tently similar to those coupons with width 25 mm. data obtained for R ¼ 1 are plotted in Figure 11 for
The static tensile tests of glass/epoxy UD laminates comparison with predictions from the proposed fatigue
were performed according to EN ISO 527-424 using model.
INSTRON 8501 (load cell capacity: 100 kN), and the
compressive tests were performed according to ASTM
Biaxial fatigue tests
D3410/D3410M25 using Shimazu universal testing
machine (load cell capacity: 250 kN). Strain was The biaxial fatigue tests were performed using a biax-
recorded using two single strain gauges attached on ial fixture mounted on INSTRON 8501 (Figure 6).
the central area of the coupons, one on each face. The
uniaxial static test results are listed in Tables 1 and 2,
respectively, where the fiber material properties and the Table 2. Uniaxial static strengths of the glass/epoxy UD
ply shear and transverse compressive strengths are not laminate
tested but referred to the values of typical glass/epoxy
Ply strengths Values
materials from the ‘World-Wide Failure Exercise
(WWFE)’.26 Longitudinal tensile strength XT (MPa) 1170
Longitudinal compressive strength XC (MPa) 977
Transverse tensile strength YT (MPa) 30.5
Transverse compressive strength YC (MPa) 114a
Shear strength S12 (MPa) 72.0a
a
Values referred to the material in Word-Wide Failure Exercise.26
A cruciform type specimen was set up on the biaxial thicker than the thickness of the biaxial zone in order to
fixture. The center of the cruciform specimen (biaxial ensure higher stresses in the biaxial zone. The ratio of
zone) was milled to reduce the thickness from 7.14 mm the thicknesses of the arm to the biaxial zone is 4:1,
(32 plies in arms) to 1.78 mm (8 plies in the central biax- which is much higher than the specimen of Smits
ial zone), as shown in Figures 7(a) and 8(a). The area of et al.27 which is 1.8:1.
the biaxial zone is about 18 18 mm2. The cruciform The biaxial loadings are achieved through the biaxial
geometry is selected from the optimum type designed fixture under a vertical loading by the INSTRON actu-
by Smits et al.27 This geometry has a reduced thickness ator. For example, if the actuator moves upwards, the
in the central region of the specimen, in combina- cruciform specimen will experience compression in the
tion with a fillet corner between two arms inside the vertical direction and tension in the horizontal direc-
material. These features cause failure to occur in the tion, and vice versa, as indicated by the white arrows
biaxially loaded test zone, rather than in the uniaxially in Figure 6. The horizontal and vertical directions cor-
loaded arms. Additionally, the uniformity of the strain respond to the x and y directions in a global reference
field in the central zone was also checked using finite coordinates system, respectively. The biaxiality ratio
element analysis and digital image correlation tech- between the horizontal and vertical loads was deter-
nique. Lamkanfi et al.28–30 found that cruciform speci- mined by the angle of link bars in the fixture. Strains
men type of Smits et al.27 may still lead to unwanted in both horizontal and vertical directions were recorded
stress concentrations at the inner radius of the arms using three strain gauges (strain rosettes) attached to
under certain conditions. They suggested the use of a the central area (biaxial zone) of the specimens as
continuous spline for the arms to replace the inner shown in Figures 7(a) and 8(a). The biaxial fatigue
radius. However, as an alternative, in this article, the tests were carried out under sinusoidal loads with fre-
thickness of the cruciform arms was instead made much quency f ¼ 4 Hz, and stress ratio R ¼ 10 for vertical
Figure 7. Photographs of off-axis specimen ([45/45]4 laminate; x ¼ 71:7 MPa, y ¼ 8:53 MPa): (a) Nf ¼ 0 (before loading),
(b) Nf ¼ 4944 (visible failure), (c) Nf ¼ 8902 (final failure), and (d) after final failure.
Figure 8. Photographs of cross-ply specimen ([0/90]4 laminate; x ¼ 269:9 MPa, y ¼ 240:6 MPa): (a) Nf ¼ 0 (before loading),
(b) Nf ¼ 14364 (visible failure), (c) Nf ¼ 14453 (final failure) and (d) after final failure.
loading and R ¼ 0:1 for horizontal loading. Two kinds the load for many cycles after the visible matrix crack-
of glass/epoxy MD laminates including off-axis lami- ing occurs as shown in Figure 7(b)–(d), whereas the
nate (lay-up configuration: [45/45]4) and cross-ply cross-ply specimens are fractured soon after the visible
laminate (lay-up configuration: [0/90]4) were tested failure, as shown in Figure 8(b)–(d).
under biaxial loads. Figures 7(b)–(d) and 8(b)–(d)
show the photograph of the biaxial fatigue specimens Demonstration and application of the
for MD [45/45]4 and [0/90]4 laminates before and
after final failure (fracture), respectively.
fatigue model
Five and ten specimens were tested under different The fatigue model may be applied to multi-axial fatigue
stress levels and biaxiality ratios for off-axis and cross- analysis of both UD and MD laminates. In this section,
ply laminates, respectively. The biaxial test data are glass/epoxy UD and MD laminates are used to demon-
plotted in Figures 18 and 23 for comparison with pre- strate the applicability of the proposed multi-axial
dictions from the proposed multi-axial fatigue model. fatigue model. Master curves of uniaxial fatigue
It was observed that damage started in the biaxial zone strengths are first obtained from the above mentioned
of each specimen and no delamination was observed in limited uniaxial fatigue tests of glass/epoxy UD lami-
the arms of the cruciform specimens. For the off-axis nates before fatigue analysis is performed based on the
specimens ([45/45]4 laminate), extensive matrix crack- master curves and Puck’s criterion. For the current
ing occurred first in the biaxial zone and then followed fatigue tests, the ply material properties and static
by fiber failure. For the cross-ply specimens ([0/90]4 strengths used in the demonstration are listed in
laminate), damage clearly started in the biaxial zone Tables 1 and 2, respectively, as mentioned in the previ-
of each specimen, and the damage was dominated by ous section. Besides the current fatigue tests, various
fiber failure. The off-axis specimens can still withstand fatigue tests on glass/epoxy laminates in literature,
Figure 9. Fitting fatigue master curves of glass/epoxy UD laminates from limited uniaxial test data: (a) longitudinal tension and
(b) longitudinal compression.
Figure 10. Fatigue master curves of glass/epoxy UD laminates Figure 11. Comparison of predictions of uniaxial S-N curves
for the reference stress ratios: transverse tension/compression with the current test data for R ¼ 1.
and shear.
including the well-known OPTIMAT database,31–38 s, 0 can be obtained by the corresponding static
SNL/MSU/DOE database,39 and some published fatigue strength 0 through the relationship in Equation (6),
test data,40 are also used to validate the model, where the and the other five parameters are determined by fitting
basic material properties are provided by corresponding data with uniaxial fatigue tests of UD laminates.
static tests. Figure 9 shows the fitted fatigue master curves of
longitudinal tension and compression with the current
Generation of fatigue master curves with limited uniaxial test data of glass/epoxy UD laminates for
f ¼ 4Hz, where the values of reference stress ratio
uniaxial fatigue tests R0 ¼ 0:1 and R0 ¼ 10 are used in the fitting data of
For fiber-reinforced UD laminates, five uniaxial ply tension and compression, respectively. The obtained fit-
fatigue strengths (longitudinal tension and compres- ting parameters are listed in Table 3.
sion, transverse tension and compression, and in- The fatigue master curves of transverse and shear
plane shear) may be determined from the fatigue strengths can also be fitted with corresponding test
master curves. Six parameters from Equation (2a) data. However, in order to simplify the analysis and
need to be determined from tests. The CSR strength without loss of generality for application of the fatigue
Figure 12. Comparison of predictions of uniaxial S-N curves with test data31,32 for various frequencies: (a) R ¼ 0.1 and (b) R ¼ 1.
Figure 13. Comparison of predictions of uniaxial S-N curves with test data33–35 for various stress ratios: (a) longitudinal direction
and (b) transverse direction.
model, it can be assumed that the fitting parameters in parameters for shear are same as those for tension.
the fatigue master curves for transverse tension and Figure 10 shows the fatigue master curves of uniaxial
compression are same as those for longitudinal tension transverse tension/compression and shear with the fit-
and compression, respectively, and that the fitting ting parameters and corresponding static strengths,
based on the reference stress ratios R0 ¼ 0:1 and
R0 ¼ 10. With these fatigue master curves, multi-axial
Table 3. Fitting parameters for fatigue master curves of the fatigue analysis can be implemented and S-N curves,
glass/epoxy UD laminate fatigue strengths, or fatigue failure envelopes can be
obtained for given load frequencies, stress ratios, and
Parameters Tension Compression multi-axiality ratios.
t1 ðminÞ 5:8 108 5:8 108
nf 0.05 0.03 Uniaxial and multi-axial S-N curves
nr 0.20 0.10
As mentioned in Section 2, both uniaxial and multi-
nc 0.15 0.11
axial S-N curves can be derived from the proposed
Figure 14. Comparison of predictions of uniaxial S-N curves with test data39 for stress ratio R ¼ 0.1: (a) longitudinal tension and
(b) transverse tension.
Figure 16. Comparison of predictions of multi-axial S-N curves for UD laminates with test data40 for different stress ratios:
(a) R ¼ 0.5, (b) R ¼ 0, and (c) R ¼ 1.
Figures 15 and 16 plot multi-axial S-N curves of UD For comparison, the uniaxial S-N curve (with a
laminates for different local multi-axiality ratios special group of local multi-axiality ratios
(1 : 2 : 12 ) and stress ratios (R), and the predictions 1 : 2 : 12 ¼ 1 : 0 : 0 or 1 : 2 : 12 ¼ 0 : 1 : 0) of lon-
are also compared with test data which are obtained gitudinal or transverse tension is also plotted. It can be
through transforming fatigue test data of UD off-axis seen that the multi-axial S-N curves vary with different
coupons from the loading direction to ply material direc- multi-axiality ratios, and therefore a logarithmic scale
tions.37,40 The relationship between the local multi-axi- in the stress ordinate is used to better show the com-
ality ratios in the material directions and the off-axis parison for a wide range of stress values. The compar-
angle is 1 : 2 : 12 ¼ cos2 : sin2 : cos sin . ison shows that the predicted multi-axial S-N curves
It should be noted that the multi-axial S-N curves in based on the equivalent stress generally agree with the
Figures 15 and 16 are generated from Equations (15) test data. In Figure 16, the predicted S-N curves in the
and (18) for given multi-axiality ratios, and the corre- longitudinal direction (1 : 2 : 12 ¼ 1 : 0 : 0) are a
sponding test data are also transformed to the equivalent little more conservative than test data, which may be
stress defined in Equation (15). The relationship between caused by the lower static longitudinal tensile strength
the equivalent stress eq and the off-axis stress x is: (XT ¼ 800 MPa). For other multi-axiality ratios in
Figure 16, the predictions generally agree well with
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
eq ¼ x 1 cos2 sin2 ð19Þ test data, except for cases with higher transverse
stress components (2 ) for stress ratio R ¼ 1 as
Figure 17. Comparison of predictions of multi-axial S-N curves for MD laminates with test data37-39 under global uniaxial loading
(special multi-axiality ratios: x : y : xy ¼ 1 : 0 : 0): (a) different lay-ups and (b) different stress ratios.
Figure 18. Comparison of predictions of biaxial S-N curves for MD laminates with the current test data for different global biaxiality
ratios (x : y ): (a) [45/45]4 laminate and (b) [0/90]4 laminate.
shown in Figure 16(c). It should be noted that the are applied with a global uniaxial load, the failure in
compressive S-N curves for R ¼ 1 are not plotted in the plies is actually caused by local multi-axial stress
Figure 16(c) because the compressive strengths are not states. For R ¼ 1 in Figure 17(a), failure was under
available in the test data.40 Nevertheless, the failure for tension in all test cases.37 However, the failure mode of
the off-axis tests for R ¼ 1 can be determined by the tests for R ¼ 1 by DOE MSU39 was not reported.
tensile S-N curves because it was dominated by the The comparison in Figure 17 shows reasonably good
matrix.40 agreement of predictions with test data, which indicates
The multi-axial S-N curves for MD laminates are plot- that the ply-level-based multi-axial fatigue model is able
ted in Figures 17 and 18, and the predictions are also to predict fatigue failure of MD laminates. Figure 18
compared with test data. In Figure 17, special global shows biaxial S-N curves of MD laminates, including
multi-axiality ratios, that is, x : y : xy ¼ 1 : 0 : 0, lay-up configurations of [45/45/]4 and [0/90]4, for
are investigated for different MD laminates and different different global biaxiality ratios (x : y ), f ¼ 4 Hz
stress ratios, respectively. Although these MD laminates and r ¼ 0:1. For the experiment, the biaxial stresses of
Figure 20. Comparison of predictions of fatigue failure envelopes for UD laminates with test data (f ¼ 1.0 5.2 Hz, R ¼ 0.1)36,37in
various stress spaces: (a) stress space 1 2 with 12 6¼ 0, (b) stress space 1 12 with 2 6¼ 0, and (c) stress space 2 12 with
1 6¼ 0.
Figure 21. Comparison of predictions of fatigue failure envelopes for UD laminates with test data (f ¼ 3.3 Hz, R ¼ 0)40 in various
stress spaces: (a) stress space 1 2 (1 0 and 2 0) with 12 6¼ 0, (b) stress space 1 12 (1 0 and 12 0) with 2 6¼ 0,
and (c) stress space 2 12 (2 0 and 12 0) with 1 6¼ 0.
Figure 22. Biaxial fatigue failure envelopes of MD laminates: (a) [45/45]4 laminate and (b) [0/90]4 laminate.
Figure 23. Comparison of predictions of fatigue failure envelopes for MD laminates with the current test data in biaxial stress space
x y (x 0 and y 0): (a) [45/45]4 laminate and (b) [0/90]4 laminate.
The fatigue failure envelopes of UD laminates for plots the tri-axial failure envelopes in the 1 2
different stress ratios are plotted in various biaxial stress space with different non-zero values of shear
stress spaces, as shown in Figures 20 and 21, where stress 12 . In Puck’s criterion, however, the longitudi-
some of the test results36,37,40 are also included and nal stress 1 usually has little effect on failure envelopes
compared with the predicted failure envelopes. Since in stress space 2 12 if 1 is not close to the longitu-
the longitudinal and transverse compressive strengths dinal tensile or compressive strength. So the failure
are not available in the test data,40 only parts with envelopes in Figures 20(c) and 21(c) with different
1 0 and 2 0 of the failure envelopes are plotted values of 1 5 5 XT (denotes fatigue stress or
in Figure 21. The comparison demonstrates that the fatigue strength in fatigue loading) for each number
predictions from the fatigue model generally agree of cycles are nearly same as the corresponding biaxial
well with the experimental results. It should be noted 2 12 failure envelopes with 1 ¼ 0. The fatigue fail-
that Figures 20 and 21 actually plot the multi-axial (tri- ure envelopes in Figures 20 and 21 clearly show fatigue
axial) failure envelopes in different biaxial stress failures of UD laminates for various multi-axiality
spaces through a sectional view with the non-zero ratios. For example, in Figure 20(c), the fatigue failures
value of the third stress. For example, Figure 20(a) for muli-axiality ratios 1 : 2 : 12 ¼ 1 : 0:03 : 0:18
(corresponding to 10o off-axis tests)37 and [0/90]4 laminate, the predicted failure envelopes are a
1 : 2 : 12 ¼ 1 : 3 : 1:73 (corresponding to 60o off- little more optimistic than test data. Usually, the failure
axis tests36,37 are mainly dominated by shear and trans- of the [45/45]4 laminate under biaxial tension–
verse tension, respectively. compression loads is dominated by the matrix and the
failure strengths obtained from FPF analysis and LPF
MD LAMINATES. For MD laminates, the last-ply analysis are similar. For the [0/90]4 laminate under
failure (LPF) usually dominates the laminate ultimate biaxial tension–compression loads, however, the failure
strength and is more important in the case of fatigue is dominated by the fibers and failure strengths
loadings, and therefore the LPF envelopes are plotted obtained from FPF and LPF analyses are quite differ-
to show the laminate failure characteristics under multi- ent. The difference between the FPF and LPF analyses
axial fatigue loadings. Biaxial fatigue failure enve- may cause optimistic predictions. The experimental
lopes are illustrated in Figures 22 and 23, which are results are supposed to be between predictions from
generated with the fatigue model and LPF analysis the FPF and LPF analyses, and closer to predictions
based on CLT. from the LPF analysis.
Figure 22 plots the fatigue failure envelopes of MD
laminates with lay-up configurations of [45/45]4 and
[0/90]4 in global biaxial stress space x y , for various
Conclusions
number of cycles to failure and r ¼ 0:1. The fatigue A multi-axial fatigue model based on fatigue master
failure envelopes are usually different for laminates curves and Puck’s criterion is developed for fatigue
with different configurations as shown in Figure 22. analysis of fiber-reinforced composite laminates in
The biaxial fatigue strengths of the same laminate this article. The fatigue master curves from ATM are
vary for different global biaxiality ratios. In addition, generated with limited uniaxial fatigue test data of UD
the biaxial fatigue strengths also vary with different laminates, and can be applied to fatigue loading condi-
laminates for the same global biaxiality ratios, for tions with different frequencies and stress ratios. Under
example, x : y ¼ 1 : 0:2, as shown in Figure 22. The fatigue loading, the uniaxial ply fatigue strengths are
failure envelopes clearly show different load capacity obtained through the fatigue master curves, and then
of the laminates for different biaxiality ratios. For the multi-axial fatigue failure is determined with these
example, for the [45/45]4 laminate, the load capacity fatigue strengths by applying Puck’s criterion at the ply
under biaxial tension–tension or compression– level. The traditional uniaxial S-N curve can be explic-
compression is generally much better than that under itly derived from the proposed fatigue model, and the
tension–compression or compression–tension, and the multi-axial S-N curve is numerically determined with
failure is dominated by either the matrix or the fibers an equivalent fatigue stress by the fatigue model,
depending on the biaxiality ratios, as shown in Figure where all stress components in the fatigue loadings
22(a). For the [0/90]4 laminate, however, the load are considered to account for the effects of multi-
capacity is similar for different biaxiality ratios because axiality. The fatigue model is also applicable to general
the final failure is always dominated by the fibers, as MD laminates through analysis methods such as CLT
shown in Figure 22(b). For the [45/45]4 laminate, or FEM. With several basic fatigue master curves for
since the in-plane shear stress-strain relationship usu- a ply material, fatigue strength of any laminate
ally presents significant non-linearity for glass/epoxy may be predicted by the multi-axial fatigue model for
fiber-reinforced composites, the secant in-plane shear various frequencies, stress ratios, and multi-axiality
modulus at failure strain is used in the LPF analysis. ratios.
The comparisons of failure envelopes from the Limited uniaxial fatigue tests on glass/epoxy UD
fatigue model (r ¼ 0:1) with the current test data laminates were carried out to provide test data for
(R ¼ 0:1) for various number of cycles to failure are establishing the fatigue master curves, and MD lami-
illustrated in Figure 23(a) and (b) for the [45/45]4 nates were also tested under biaxial fatigue loadings.
and [0/90]4 laminates, respectively, where x and y cor- The application of the proposed multi-axial fatigue
respond to the stresses along the horizontal and vertical model is demonstrated with predictions of S-N curves
directions in the tests. The predictions from the model and fatigue failure envelopes of UD and MD laminates.
show that the failures of the [45/45]4 and [0/90]4 lam- By using multi-axial S-N curves and fatigue failure
inates under biaxial tension-compression loads are envelopes, multi-axial fatigue failure can be illustrated
dominated by the matrix and the fibers, respectively, with any number of cycles to failure and any group
which agree with the experimental observations. It of multi-axiality ratios in a defined stress space. The
can be seen that the theoretical predictions for the multi-axial fatigue model can account for both local
[45/45]4 laminate are in reasonably good agreement and global multi-axial failures. Two features of the
with the experimental results. For the multi-axial fatigue model are demonstrated:
1. Ply-level prediction of multi-axial fatigue failure, under in-plane fatigue loading. Compos Sci Technol
which means the model is easy to be used for lami- 2003; 63: 677–694.
nates with various lay-up configurations. 9. Kawai M. A phenomenological model for off-axis fatigue
2. Generation of fatigue failure envelope in stress behavior of unidirectional polymer matrix composites
under different stress ratios. Compos Part A-Appl S
spaces, which can clearly show the multi-axial failure
2004; 35: 955–963.
and different failure modes, such as tension, com- 10. Passipoularidis VA and Philippidis TP. Strength degra-
pression or shear. dation due to fatigue in fiber dominated glass/epoxy
composites: a statistical approach. J Compos Mater
The fatigue of glass/epoxy UD and MD laminates 2009; 43: 997–1013.
with various lay-up configurations is investigated 11. Shokrieh MM and Lessard LB. Multiaxial fatigue behav-
with the proposed fatigue model. The predictions of iour of unidirectional plies based on uniaxial fatigue
uniaxial and multi-axial S-N curves with different experiments – I. Modelling. Int J Fatigue 1997; 19: 201–207.
stress ratios and multi-axiality ratios, for both local 12. Shokrieh MM and Lessard LB. Multiaxial fatigue behav-
and global multi-axialities, from the proposed fatigue iour of unidirectional plies based on uniaxial fatigue
model are compared and validated with various test experiments – II. Experimental evaluation. Int J Fatigue
1997; 19: 209–217.
data. Fatigue failure envelopes are also generated for
13. Shokrieh MM and Lessard LB. Progressive fatigue
multi-axial loadings and compared with limited test
damage modeling of composite materials, Part I:
data. Reasonably good agreement is observed between Modeling. J Compos Mater 2000; 34: 1056–1080.
the results of prediction and experiment. Although the 14. Shokrieh MM and Lessard LB. Progressive fatigue
multi-axial fatigue model is demonstrated and validated damage modeling of composite materials, Part II:
with glass/epoxy laminates with uniform stress distribu- Material characterization and model verification.
tion, it can be extended to non-uniform stress distribu- J Compos Mater 2000; 34: 1081–1116.
tion in structures, where the model can be coded into 15. Diao X, Lessard LB and Shokrieh MM. Statistical model
numerical analysis software. for multiaxial fatigue behavior of unidirectional plies.
Compos Sci Technol 1999; 59: 2025–2035.
16. Liu Y and Mahadevan S. Probabilistic fatigue life predic-
Acknowledgements tion of multidirectional composite laminates. Compos
Struct 2005; 69: 11–19.
Financial support from Vestas and NUS Incentive Grant is
17. Quaresimin M, Susmel L and Talreja R. Fatigue behav-
gratefully acknowledged.
iour and life assessment of composite laminates under
multiaxial loadings. Int J Fatigue 2010; 32: 2–16.
18. Passipoularidis VA, Philippidis TP and Brondsted P.
References
Fatigue life prediction in composites using progressive
1. Degrieck J and Van Paepegem W. Fatigue damage model- damage modeling under block and spectrum loading.
ing of fiber-reinforced composite materials: Review. Appl Int J Fatigue 2011; 33: 132–144.
Mech Rev 2001; 54: 279–299. 19. Miyano Y, Nakada M and Cai H. Formulation of long-
2. Fawaz Z and Ellyin F. Fatigue failure model for fiber- term creep and fatigue strengths of polymer composites
reinforced materials under general loading conditions. based on accelerated testing methodology. J Compos
J Compos Mater 1994; 28: 1432–1451. Mater 2008; 42: 1897–1919.
3. Fawaz Z and Ellyin F. A new methodology for the predic- 20. Puck A and Schürmann H. Failure analysis of FRP lam-
tion of fatigue failure in multidirectional fiber-reinforced inates by means of physically based phenomenological
laminates. Compos Sci Technol 1995; 53: 47–55. models. Compos Sci Technol 1998; 58: 1045–1067.
4. Philippidis TP and Vassilopoulos AP. Fatigue strength 21. Puck A and Schürmann H. Failure analysis of FRP lam-
prediction under multiaxial stress. J Compos Mater 1999; inates by means of physically based phenomenological
33: 1578–1599. models. Compos Sci Technol 2002; 62: 1633–1662.
5. Huang ZM. Fatigue life prediction of a woven fabric com- 22. Puck A, Kopp J and Knops M. Guidelines for the deter-
posite subjected to biaxial cyclic loads. Compos Part A- mination of the parameters in Puck’s action plane
Appl S 2002; 33: 253–266. strength criterion. Compos Sci Technol 2002; 62: 371–378.
6. Gude M, Hufenbach W, Koch I and Protz R. Fatigue 23. Tsai SW. Theory of composites design. In: Tsai SW (ed.)
failure criteria and degradation rules for composites Strength & life of composites. CA, USA: Stanford
under multiaxial loadings. Mech Compos Mater 2006; 42: University, 2008, pp. A9-8–A9-17.
443–450. 24. EN ISO 527-4. Plastics – Determination of tensile proper-
7. Kawakami H, Fujii TJ and Morita Y. Fatigue degradation ties – Part 4: test conditions for isotropic and orthotropic
and life prediction of glass fabric polymer composite under fiber-reinforced plastic composites. Geneve, Switzerland:
tension/torsion biaxial loading. J Reinf Plast Comp 1996; International Organization for Standardization, 1997.
15: 183–195. 25. ASTM D3410/D3410M. Standard test method for com-
8. Van Paepegem W and Degrieck J. Modelling damage pressive properties of polymer matrix composite materials
and permanent strain in fiber-reinforced composites with unsupported gage section by shear loading.
Philadelphia, PA, USA: American Society for Testing 34. Philippidis TP, Passipoularidis VA, Assimakopoulou TT
and Materials, 2003. and Antoniou AE. ‘Fatigue Tests in the Fiber Direction
26. Soden PD, Hinton MJ and Kaddour AS. Lamina prop- of UD OB Standard Specimen: Main Test Phase I’,
erties, lay-up configurations and loading conditions for a OPTIMAT BLADES, ENK6-CT2001-00552, OB_TG1_
range of fiber-reinforced composite laminates. Compos R013, http://www.wmc.eu/public_docs/10219_001.pdf
Sci Technol 1998; 58: 1011–1022. (2006, accessed 16 July 2011).
27. Smits A, Van Hemelrijck D, Philippidis TP and Cardon 35. Philippidis TP, Assimakopoulou TT, Antoniou AE and
A. Design of a cruciform specimen for biaxial testing of Passipoularidis VA. ‘Fatigue Tests on OB Standard
fiber reinforced composite laminates. Compos Sci Technol Coupons at 90o: Main Test Phase I’, OPTIMAT
2006; 66: 964–975. BLADES, ENK6-CT2001-00552, OB_TG2_R021, http://
28. Lamkanfi E, Van Paepegem W, Degrieck J, Ramault C, www.wmc.eu/public_docs/10253_000.pdf (2005, accessed
Makris A and Van Hemelrijck D. Strain distribution in 16 July 2011).
cruciform specimens subjected to biaxial loading condi- 36. Philippidis TP, Antoniou AE, Assimakopoulou TT and
tions. Part 1: Two-dimensional vs. three-dimensional Passipoularidis VA. ‘Static Tests on the Standard OB UD
finite element model. Polym Test 2010; 29: 7–13. and MD Off-axis Coupons: Main Test Phase I’,
29. Lamkanfi E, Van Paepegem W, Degrieck J, Ramault C, OPTIMAT BLADES, ENK6-CT2001-00552, OB_TG2_
Makris A and Van Hemelrijck D. Strain distribution in R022, http://www.wmc.eu/public_docs/10254_000.pdf
cruciform specimens subjected to biaxial loading condi- (2005, accessed 15 July 2011).
tions. Part 2: Influence of geometrical discontinuities. 37. Philippidis TP, Antoniou AE, Assimakopoulou TT and
Polym Test 2010; 29: 132–138. Passipoularidis VA. ‘Fatigue Tests on OB Unidirectional
30. Makris A, Vandenbergh T, Ramault C, Van Hemelrijck & Multidirectional Off-axis Coupons: Main Test Phase
D, Lamkanfi E and Van Paepegem W. Shape optimisa- I’, OPTIMAT BLADES, ENK6-CT2001-00552, OB_
tion of a biaxially loaded cruciform specimen. Polym Test TG2_R030, http://www.wmc.eu/public_docs/10350_000
2010; 29: 216–223. .pdf (2006, accessed 15 July 2011).
31. Lekou DJ. ‘Test Results of Benchmark Static Tests on 38. Philippidis TP, Assimakopoulou TT, Passipoularidis VA
UD and MD Coupons’, OPTIMAT BLADES, ENK6- and Antoniou AE. ‘Static and Fatigue Tests on ISO
CT2001-00552, OB_TG1_R007, http://www.wmc.eu/pub- Standard 45o Coupons: Main Test Phase I’, OPTIMAT
lic_docs/10154_000.pdf (2003, accessed 16 July 2011). BLADES, ENK6-CT2001-00552, OB_TG2_R020, http://
32. Lekou DJ. ‘Test Results of CA Fatigue Tests on UD www.wmc.eu/public_docs/10207_000.pdf (2004, accessed
Coupons’, OPTIMAT BLADES, ENK6-CT2001-00552, 15 July 2011).
OB_TG1_R011, http://www.wmc.eu/public_docs/10197_ 39. MSU Composite Group. ‘SNL/MSU/DOE Composite
001.pdf (2004, accessed 16 July 2011). Material Fatigue Database, Version 20.0’, http://www.
33. Philippidis TP, Vassilopoulos AP, Assimakopoulou TT coe.montana.edu/composites/ (2011, accessed xx xx xxxx).
and Passipoularidis V. ‘Static and Fatigue Tests of 40. El Kadi H and Ellyin F. Effect of stress ratio on the
OPTIMAT UD Coupons: Benchmark Tests’, OPTIMAT fatigue of unidirectional glass fiber/epoxy composite lam-
BLADES, ENK6-CT2001-00552, OB_TG2_R012, http:// inae. Composites 1994; 25: 917–924.
www.wmc.eu/public_docs/10120_000.pdf (2003, accessed
16 July 2011).