Composites Science and Technology 59 (1999) 2025±2035

Statistical model for multiaxial fatigue behavior of unidirectional

plies
Xiaoxue Diao a, Larry B. Lessard a,*, Mahmood M. Shokrieh b
a
Department of Mechanical Engineering, McGill University, 817 Sherbrooke Street West, Montreal, Quebec, Canada, H3A 3K6
b
Mechanical Engineering Department, Iran University of Science and Technology, Narmak, Tehran, Iran

Received 10 February 1998; received in revised form 2 March 1999; accepted 12 April 1999

Abstract
A statistical model is developed for the prediction of the fatigue life of unidirectional composite laminae subjected to multiaxial
fatigue loading. This model is based on the experimental data for the fatigue behavior of those laminae subjected to uniaxial fatigue
loading. The method is based on the extension of the generalized residual material property degradation model previously developed
to describe fatigue behavior of composite materials under a wide range of stress ratios, combined with the statistical nature of static
and fatigue failure of ®ber-reinforced composites. The distribution function of fatigue life is determined in terms of the distribution
function of static strength of the composite laminae in di€erent loading modes, which is considered to be stochastic in nature. The
fatigue life and fatigue strength of unidirectional composite laminae under any multiaxial fatigue loading are evaluated statistically
by using the fatigue behavior of the laminae under uniaxial loading in longitudinal, transverse and in-plane shear directions, which
is determined experimentally through material characterization. The application of the statistical model to evaluation of fatigue life
and residual strength of unidirectional composite laminae subjected to biaxial fatigue loading shows good agreement with the cor-
responding experimental data. # 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Composites; Strength degradation; Statistical model; Stress ratio; B. Fatigue; C. Failure criteria

1. Introduction basic set of material fatigue data for the prediction of

fatigue behavior in all laminates and structures. In
Fatigue failure of composite materials under multiaxial addition, the statistical nature of the basic data leads to
cyclic loading has always been an important concern in design guidelines for structure reliability.
the application of composite structures [1±7]. Most ®ber- Although there is an extensive amount of research on
composite structures are made from laminates which biaxial/multiaxial fatigue of metals [8], research in the
consist of uniaxial by reinforced plies where each ply same ®eld on composite materials is much less complete.
may have a di€erent ply angle. Fatigue testing of all An early work in the study of composite failure was
laminate con®gurations and loading conditions is done by Tsai and Hill [9] and a failure criterion was
impossible, both cost- and time-wise. Thus, an analy- proposed for static strength of composites under multi-
tical method for the optimum design of composite axial stresses. To study fatigue failure, Sims and Brog-
laminates and structures against fatigue failure is don [10] extended the Tsai±Hill criterion to fatigue
urgently needed. A practical approach to this problem is failure by replacing the static strength by fatigue
to establish a fatigue failure theory so that the fatigue strength of the composite material. With the assumption
behavior and life of a composite laminate of complex that the ®ber-failure and matrix-failure are independent
con®guration under multiaxial loading can be predicted events, Hashin and Rotem [11±16] developed a fatigue
in terms of fatigue properties of the individual plies failure criterion which distinguishes the ®ber-failure
under uniaxial loading. This approach would eliminate mode and matrix-failure mode. Following their
a large amount of experimental work by providing a research, Aboudi [17] developed a micromechanical
model for the fatigue failure of unidirectional composite
laminates using the concept of a representative unit cell,
* Corresponding author. associated with the failure properties of ®bers and the
0266-3538/99/$ - see front matter # 1999 Elsevier Science Ltd. All rights reserved.
PII: S0266-3538(99)00062-7
2026 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035

bulk matrix. In a similar way, Reifsnider and co-work- [10] proposed a fatigue failure criterion based on the
ers [18±20] further proposed a micromechanical model Tsai±Hill theory [9],
which incorporates the fatigue behavior of the con-  2  
stituents with interactions between them, such as ®ber- 11 11 22
ÿ ‡
matrix debonding. However, these models have only R11 …n; 11 ; † …R11 …n; 11 ; ††2
been applied to a speci®c stress ratio, and not to general  2  2 …1†
22 12
cases. For example, the stress states are not the same at ‡ ˆ1
di€erent points in a pin/bolt fatigue loaded composite R22 …n; 22 ; † R12 …n; 12 ; †
laminate, thus, an analysis using a constant stress ratio
may lead to incorrect results. To eliminate this impor- where R11 …n; 11 ; †; R22 …n; 22 ; † and R12 …n; 12 ; † are
tant restriction, Shokrieh and Lessard [21,22] developed the residual strengths in the longitudinal, transverse and
a generalized residual material property degradation in-plane shear directions, respectively. An ecient
model, coupling a normalized residual strength model method for incorporating strength parameters that
and a normalized fatigue life model. The model can be degrade with fatigue cycles has been developed by
used to predict the fatigue failure for unidirectional [21,22] and is not presented here. From the generalized
laminates under multiaxial loading with arbitrary stress residual material property degradation model [21,22], the
ratio. However, all these models [10±22] are determinis- residual strengths for each state of stress can be expressed
tic in approach, treating the ®ber-reinforced composite in following forms:
as a homogeneous anisotropic continuum without con-
" !a11 #b1
sideration of the statistical nature of fatigue failure. The log…n† ÿ log…0:25† 11

large scatter in residual strength and fatigue life mea- R11 …n; 11 ; † ˆ 1 ÿ ÿ

log Nf11 ÿ log…0:25†
surements of composite materials is a well-known fact,
which should be included in the design of composite …Rs11 ÿ 11 † ‡ 11
structures. Yamada and Sun [23,24] introduced a statis- …2†
tical function of the static strength in the longitudinal
and transverse directions in their study on the o€-axis " !a22 #b1
strength of unidirectional ®ber reinforced composites. log…n† ÿ log…0:25† 22

However, that work was limited to static strength. R22 …n; 22 ; † ˆ 1 ÿ ÿ

log Nf22 ÿ log…0:25†
In this paper, a statistical approach is employed to
study the fatigue failure of ®ber-reinforced composite …Rs22 ÿ 22 † ‡ 22
laminae subjected to multiaxial loading from the static …3†
and fatigue properties of unidirectional laminae under
unidirectional fatigue loading. The statistical distribu- " !a12 #b1
tion functions of material strength in three principal log…n† ÿ log…0:25† 12

R12 …n; 12 ; † ˆ 1 ÿ ÿ

failure modes (longitudinal and transverse tension as log Nf12 ÿ log…0:25†
well as in-plane shear) are obtained by ®tting the
experimental data of material strength to a two-para- …Rs12 ÿ 12 † ‡ 12
meter Weibull distribution. The generalized residual …4†
material property degradation model is used to gen-
eralize the statistical model for composite laminae under where Rsi …i ˆ 11; 22; 12† are the static strengths, ai and
fatigue loading with arbitrary stress ratio. The statistical bi …i ˆ 11; 22; 12† are curve ®tting parameters for the
model is then applied to simulate the S±N curve and the degradation behavior of residual strength of the uni-
residual strength degradation of unidirectional laminae directional laminae under uniaxial fatigue loading.
under biaxial fatigue loading and compares these results Nfi …i ˆ 11; 22; 12† are fatigue cycles to failure of the
with corresponding experimental data. For the sake of unidirectional laminae under uniaxial loading 11 ; 22
experimental simplicity, the o€-axis specimen test and 12 , respectively. The dependence of residual
method is used to realize the biaxial fatigue loading strength on the stress ratio  is re¯ected by a relation-
case. ship between number of cycles to failure and the corre-
sponding state of stress. Thus, additional information is
required to input the correct fatigue life into Eqs. (2±4).
2. Multiaxial failure criteria To include the e€ects of stress ratio on the fatigue life, the
normalized fatigue life model is developed in [21,22] and
Consider unidirectional laminae under a biaxial state is used to express Nfi …i ˆ 11; 22; 12† in a generalized form
of fatigue stress in longitudinal (11 ), transverse (22 ) by the alternating stress a ˆ …max ÿ min †=2, tensile
and shear (12 ) directions. To study the fatigue failure strength t and compressive strength c . For longitudinal
behavior of composite laminates, Sims and Brogdon loading, the equation is:
 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035 2027

ln…a=f† 4) the number of cycles to failure can also be determined

uˆ ˆ A ‡ B log Nf …5†
ln‰…1 ÿ q†…c ‡ q†Š as a function of relevant stress state and static strength
denoted by functions g1 and g2 for ®ber failure mode
and matrix failure mode, respectively.
and for transverse loading,
n ˆ g1 …Rs11 ; Rs12 ; 11 ; 12 † for fiber failure mode
 
ln…a=f† …10†
u ˆ log ˆ A ‡ B log Nf …6†
ln‰…1 ÿ q†…1 ‡ q†Š
n ˆ g2 …Rs22 ; Rs12 ; 22 ; 12 † for matrix failure mode

where a ˆ a =t ; q ˆ m =t ; c ˆ c =t , and f; u; v are …11†

curve ®tting constants. The A and B are model para-
meters, which are determined by ®tting experimental If the relationship between the normalized fatigue life,
data of di€erent stress ratios [21,22]. stress state and stress ratio is extracted from experi-
The curve ®tting parameters for di€erent loading mental data, the above discussion can be generalized for
modes are Ai and Bi …i ˆ 11; 22; 12†. By substituting arbitrary stress ratios. However, because of the inho-
Eqs. (2±4) into Eq. (1), the number of cycles to failure mogeneous microstructures of composite materials, the
can be numerically solved as an explicit function of sta- fatigue strength of composite laminae can not be deter-
tic strength and stress state. It is also an implicit func- ministic, but rather, statistical in nature. The large scat-
tion of stress ratio by means of the Goodman-type ter in experimental data of mechanical properties and
diagram. fatigue life of composite materials requires an appro-
priate interpretation. Therefore, a deterministic
n ˆ g…Rs11 ; Rs22 ; Rs12 ; 11 ; 22 ; 12 † …7† approach is not appropriate for the prediction of fatigue
life and a statistical approach should be employed
[25,26].
In a deterministic approach, fatigue life of the compo-
site laminae under biaxial loading is calculated in terms
of Eq. (7). However, the failure criterion Eq. (1) has the 3. Development of statistical model
disadvantage of giving no information as to the mode by
which failure occurs, namely longitudinally, transversely 3.1. Arbitray multiaxial loading
or by shear.
With the assumption that the ®ber failure and matrix In the statistical description, the fatigue strength of
failure are independent events, Hashin [14] developed a laminae in the longitudinal, transverse and in-plane
set of fatigue failure criteria for a unidirectional laminae shear directions is no longer deterministic, but random
under a biaxial fatigue loading, which describe ®ber in nature. Because the fatigue life of laminae subjected
failure mode and matrix failure mode separately. to biaxial loading is dependent on the residual strength
According to their studies, failure of unidirectional in the three material principal axes, it is consequently a
composite laminae occurs when one of the following random variable. In this discussion, the randomness in
conditions is satis®ed: fatigue strength is considered to be attributed only to
the random static strength of the lamina in its three
 2  2 principal material axes for the simplicity of discussion.
11 12
‡ The ¯uctuation in degradation behavior of residual
R11 …n; 11 ; † R12 …n; 12 ; †
strength, which involves the interaction between the
ˆ 1 for fiber failure mode …8† inhomogenous microstructure and damage development
[27], is not considered. Therefore, the Rs11 ; Rs22 ; Rs12 in
Eqs. (7, 10 and 11) are now random variables and n is
 2  2 consequently a random variable. Fatigue life of the
22 12 laminae subjected to biaxial loading must be determined
‡
R22 …n; 22 ; † R12 …n; 12 ; † statistically in terms of distribution functions of the
variables Rs11 ; Rs22 ; Rs12 . Fig. 1 shows the statistical
ˆ 1 for matrix failure mode …9†
distribution of static strength, residual strength, fatigue
life and their relations [25,26].
Whether the laminate fails in ®ber mode or in matrix It is assumed here that the static strengths in the long-
mode is determined by which one of Eqs. (8) and (9) is itudinal, transverse and in-plane shear modes are inde-
satis®ed ®rst. From a combination of Eqs. (8, 9 and 2± pendent of each other and their distribution functions
2028 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035

where  is the Dirac Delta function. The cumulative

distribution function of fatigue life can be determined as
…n
F…n; 11 ; 22 ; 12 † ˆ f …n; 11 ; 22 ; 12 †dn
… 0:25
……
ˆ dRs11 dRs22 dRs12 f11 …11 †
g4n
f22 …22 † f12 …12 †
…17†

With the density distribution function of fatigue life Eq.

Fig. 1. Schematics of statistical distribution of static strength, residual (16), the average fatigue life of the composite lamina sub-
strength and fatigue life. jected to biaxial loading and its mean square deviation
can be calculated.
are of the form of Weibull functions [28], thus the den- …1
sity distribution functions of static strength in three NF …11 ; 22 ; 12 † ˆ nf …n; 11 ; 22 ; 12 †dn
loading modes are given by:
… 0:25
1 …1 …1
    i  ˆ dRs11 dRs22 dRs12
i Rsi iÿ1 Rsi
fi …Rsi † ˆ exp ÿ ; 0 0 0
i i i …12† g…Rs11 ; Rs22 ; Rs12 ; 11 ; 22 ; 12 † f11 …Rs11 †
i ˆ 11; 22; 12
f22 …Rs22 † f12 …Rs12 †
where ai ; bi (i=11, 22, 12) are the shape and scale para- …18†
meters of the Weibull function, which are determined …1
from the experimental data of static strength in three load-
N2F ˆ …n ÿ NF †2 f …n; 11 ; 22 ; 12 †dn
ing modes, respectively. The corresponding cumulative … 0:25
1 …1 …1
distribution functions of static strength are ˆ dRs11 dRs22 dRs12
… Rsi   i  0 0 0
Rsi g2 …Rs11 ; Rs22 ; Rs12 ; 11 ; 22 ; 12 † f11 …Rs11 † f22
Fi …Rsi † ˆ f …Rsi †dRsi ˆ 1 ÿ exp ÿ ;
i …13†
0
…Rs22 † f12 …Rs12 † ÿ N 2F
i ˆ 11; 22; 12
…19†

The relations between ai ; bi (i=11, 22, 12) and the If Hashin's criteria are used the density distribution
average value and square deviation of static strength Rsi function of fatigue life is calculated in a similar way,
(i=11, 22, 12) are expressed as separately for ®ber failure mode,
  …1 …1
1
Rsi ˆ i ÿ 1 ‡ …14† f …n; 11 ; 12 † ˆ dRs11 dRs12 
i
0 0
     …n ÿ g1 …Rs11 ; Rs12 ; 11 ; 12 †† f11 …11 † f12 …12 †
2 1

R2si ˆ 2i ÿ 1 ‡ ÿ ÿ2 1 ‡ …15† …20†
i i
and matrix failure mode,
where ÿ is the Gamma function. From probability the-
ory [29], the density distribution function of the random …1 …1
variable n can be calculated based on Eq. (7) if the fail- f …n; 22 ; 12 † ˆ dRs22 dRs12 
0 0
ure criterion Eq. (1) is utilized.
…n ÿ g2 …Rs22 ; Rs12 ; 22 ; 12 †† f22 …11 † f12 …12 †
…1 …1 …1
…21†
f …n; 11 ; 22 ; 12 † ˆ dRs11 dRs22 dRs12 
0 0 0
…n ÿ g…Rs11 ; Rs22 ; Rs12 ; 11 ; 22 ; 12 †† With these density distribution functions of fatigue
life, the average fatigue life of the composite lamina
f11 …11 † f22 …22 † f12 …12 †
subjected to biaxial loading and its mean square devia-
…16† tion can be calculated for ®ber failure mode
 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035 2029
…1
NF1 …11 ; 12 † ˆ nf …n; 11 ; 12 †dn
… 0:25
1 …1
ˆ dRs11 dRs12 g1 …Rs11 ; Rs12 ; 11 ; 12 †
0 0
f11 …Rs11 † f12 …Rs12 †
…22†
…1

N2F1 ˆ …n ÿ NF1 †2 f …n; 11 ; 12 †dn
… 0:25
1 …1
ˆ dRs11 dRs12 g21 …Rs11 ; Rs12 ; 11 ; 12 † f11
0 0
…Rs11 † f12 …Rs12 † ÿ N 2f 1
…23†

and for matrix failure mode

…1 …1
NF2 …22 ; 12 † ˆ nf …n; 22 ; 12 †dn ˆ dRs22
… 0:25
1
0
…24†
dRs12 g2 …Rs22 ; Rs12 ; 22 ; 12 † f22
0
…Rs22 † f12 …Rs12 †
Fig. 2. O€-axis specimen under fatigue loading.
…1

N2F2 ˆ …n ÿ NF2 †2 f …n; 22 ; 12 †dn
… 0:25
1 …1 in which k11 ; k22 ; k12 are the o€-axis angle -dependent
ˆ dRs22 dRs12 g22 …Rs22 ; Rs12 ; 22 ; 12 † f22 parameters:
0 0
8
…Rs22 † f12 …Rs12 † ÿ N2F2 < k11 ˆ cos2 
…25† k ˆ sin2  …28†
: 22
k12 ˆ ÿ cos  sin 
If NF1 < NF2 , ®ber failure mode dominates. Conse-
quently, if NF1 > NF2 , then failure is in matrix mode. The average fatigue life, and its mean square devia-
From the above discussion, it is clear that the fatigue tion, of the composite laminae subjected to an o€-axis
life of a unidirectional lamina can be calculated in terms unidirectional loading can be calculated using Eqs. (18±
of the statistical model if the degradation behavior of 19), or (23±26), in which the i …i ˆ 11; 22; 12† are
residual strength and the statistical distribution of static determined from Eqs. (27 and 28). The failure of the
strength of the lamina under uniaxial loading in long- laminae depends on the o€-axis angle of the applied
itudinal, transverse and in-plane shear directions, stress. If the Hashin's criteria are used, whether the
respectively, are known. laminate fails in ®ber mode or in matrix mode is deter-
mined by which one of Eqs. (10 and 11) is satis®ed ®rst.
3.2. Unidirectional laminae under o€-axis unidirectional The transition from ®ber failure mode to matrix failure
tension can be de®ned by the transition angle T at which both
criteria Eqs. (10 and 11) are satis®ed simultaneously.
For o€-axis unidirectional laminae under an uniaxial The transition angle can thus be determined as:
fatigue stress with maximum applied stress XX as
shown in Fig. 2, the stresses in longitudinal, transverse s
R22 …n; Rs22 ; †
and in-plane shear directions in Eqs. (1, or 10 and 11) T ˆ arctan …29†
can be expressed as R11 …n; Rs22 ; †

8 The transition angle T is also a random variable. The

< 11 ˆ k11 XX average transition angle is calculated statistically in
 ˆ k22 XX …27† terms of distribution functions of static strength in
: 22
12 ˆ k12 XX longitudinal and transverse directions.
2030 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035

…1 …1 s …1 …1
R22 …n; Rs22 ; XX †
T ˆ dRs11 dRs22 arctan …30†
R2 ˆ dRs22 dRs12
0 0 R11 …n; Rs22 XX † 0 0
f22 …Rs22 † f12 …Rs12 †
" 2  2 #
k22 k12
‡
For the graphite/epoxy AS4/3501-6 laminate the transi- R22 …n; XX ; Rs22 † R12 …n; XX ; Rs12 †
tion angle was calculated to be 9.3 [21,22]. For the case ÿ
2
that the o€-axis angle is larger than the average transi- ÿ R
tion angle, i.e.  > T , failure is controlled by the matrix …34†
mode so that the fatigue life of composite lamina is
determined by Eq. (17) rather than Eq. (16). From the above discussion, it is clear that fatigue life
and residual strength of a unidirectional lamina under
3.3. Residual strength degradation o€-axis loading can be calculated in terms of the statis-
tical model if the degradation behavior of residual
The statistical model can be used to simulate the resi- strength and the statistical distribution of static strength
dual strength degradation of a o€-axis lamina. Hashin's of the lamina under uniaxial loading in transverse and
fatigue theory [14] is combined with Eq. (11) to give a in-plane shear directions, respectively, are known
relation between the residual strength of unidirectional through experiments of material characterization.
laminae under o€-axis fatigue loading and the residual
strengths of the laminae under longitudinal and transverse
fatigue loading. 4. Statistical evaluation

 2  2   The application of a statistical model for simulation

22 12 XX 2 of fatigue life and residual strength requires two types of
‡ ˆ …31†
R22 …n; 22 ; Rs22 † R12 …n; 12 ; Rs12 † R tests. The ®rst is the full material characterization of
unidirectional laminae in both the transverse direction
and the in-plane shear condition under uniaxial static
where R is the residual strength of the laminae under and fatigue loading. This includes a series of tests for
o€-axis fatigue loading. It is a function of the o€-axis static tensile and compressive strength, residual strength
fatigue stress and the fatigue cycles. From Eqs. (31, 27 and fatigue life at di€erent applied stress levels and
and 28), the residual strength can be obtained as: stress ratios. These results will be used to determine the
material parameters, which then will be used as input
1 for the statistical model. The second series of tests is the
R ˆ "
 2  2 #12 static and fatigue tests on a laminate under o€-axis
k22 k12 loading, which will be compared with the simulated
‡
R22 …n; XX ; Rs22 † R12 …n; XX ; Rs12 † results by the statistical model. The material selected is
the graphite/epoxy AS4/3501-6 unidirectional laminae,
…32†
and the loading tests are conducted on 30 o€-axis
laminates. The test cases were chosen so that  > T .
Because Eq. (10) depends on the static transverse and This way Eq. (9) rather than Eq. (8) could be used and
in-plane shear strengths of unidirectional laminae, unidirectional results for matrix shear were necessary to
which are considered random in nature, the residual verify the test. This was done so that a good series of
strength of an o€-axis lamina is also a random variable tests could be performed over a wide range of test con-
and must be determined statistically. From the distribu- ditions. If this had been done for  < T , then many of
tion function of static strength Eq. (12) the average resi- the low level tests would have taken too long to perform
dual strength and its square deviation can be determined (in the matrix dominated tests, even low level tests yiel-
as [25,26]: ded failures within a reasonable numbers of cycles). The
scenario for the statistical model is shown schematically
…1 …1 in Fig. 3.
R ˆ dRs22 dRs12
0 0
4.1. Material characterization
1
" 2  2 #12 …33†
k22 k12 A series of tests were conducted on graphite/epoxy
‡ AS4/3501-6 unidirectional lamina under transverse
R22 …n; XX ; Rs22 † R12 …n; XX ; Rs12 †
and in-plane shear loading conditions for material
f22 …Rs22 †f12 …Rs12 † characterization.
 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035 2031

4.1.1. Transverse direction tests Fig. 6 with the normalized axes based on Eq. (13). The
The results of static and compressive strength [21,22] ®tting parameters a22 and b22 in the degradation formula
of unidirectional lamina under uniaxial loading and of normalized residual strength against the normalized
their deviations are shown in Table 1. From these fatigue life are thus determined, shown in Table 3.
results the Weibull scale and shape parameters 22 and
22 are determined in terms of Eqs. (17 and 18). 4.1.2. In-plane shear tests
The results of fatigue life of unidirectional 90 lamina To increase the stability of the specimen during the
under tension-tension (R ˆ 0:1) and compression-com- test, a [0=90]s con®guration is selected instead of 0 ply
pression (R ˆ 10) fatigue loading with various stress for the specimen with the assumption that theoretically
levels [21,22] are shown in Fig. 4. The results are con- a [0=90]s specimen behaves the same as a unidirectional
verted to a relation between the life parameter u22 and 0 ply under in-plane shear loading conditions, even
logarithmic fatigue life based on Eq. (9) in order to
extract a master curve shown in Fig. 5. The ®tting
parameters A22 and B22 for the master curve are deter-
mined as 1.0117 and 0.0937, respectively, as is listed in
Table 2. The parameter f is taken to be a constant of
1.06 for better ®tting to the experimental results [21,22].
The results of residual strength of a unidirectional 90
lamina under tension±tension fatigue loading (R=0.2,
at frequencies from 1 to 10 Hz) with maximum stress of
40% and 60% of static strength [21,22] are presented in

Fig. 4. Experimental data of static strength and fatigue life of unidirec-

tional laminae under transverse loading with di€erent stress ratios.

Fig. 5. Normalized fatigue life curve for unidirectional laminae under

Fig. 3. Schematics of statistical model. transverse loading.

Table 1
Static tensile strength of unidirectional lamina and Weibull parameters Table 2
Constant life model parameters
Loading mode (i) Average Standard i i
(MPa) deviation (MPa) Loading mode (i) Ai Bi

Transverse tension (22) 52.56 4.10 16.13 54.31 Transverse tension (22) 1.0117 0.0937
In-plane shear (12) 136.33 7.41 18.63 139.42 In-plane shear (12) 0.1245 0.1785
2032 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035

thought there may be some constraining e€ects between 8. Here 1.06 is also used as the value for the constant f in
plies in an actual test [30]. The average strength and its all the calculations of residual strength and fatigue life.
deviation of the test results are listed in Table 1, from The results of residual strength of a unidirectional
which the Weibull parameters 12 and 12 are deter- lamina under tension±tension (R=0.1, at frequencies
mined. equal to 1 and 2 Hz) fatigue loading with 40% and 59%
The results of fatigue life of unidirectional lamina under of static strength [21,22] are presented in Fig. 9. The
in-plane shear fatigue loading with the maximum stress of ®tting parameters a12 and b12 in the formula of normal-
40% and 80% of static strength and for two di€erent ized strength degradation expressed by Eq. (14) are
stress ratios 0.1 and 0.0 [21,22] are shown in Fig. 7. These obtained as listed in Table 3.
data are also converted to a relation between the life
parameter u12 and logarithmic fatigue life based on Eq.
(10), from which a master curve is extracted with ®tting 5. Results
A12 and B12 determined consequently, and shown in Fig.
With all the model parameters obtained from materi-
als characterization as the input data, the statistical
model can be applied to calculate the S±N curve of any
o€-axis lamina. First the statistical model [Eqs. (33 and
34)] is applied to the static strength of unidirectional
lamina with di€erent o€-axis angles. The results are
shown in Table 4. It can be seen that the static strength

Fig. 6. Normalized residual fatigue strength curve for unidirectional

laminae under transverse tensile loading conditions.

Table 3
Degradation parameters for normalized residual strength

Loading mode (i) ai bi

Transverse tension (22) 9.6287 0.1255

In-plane shear (12) 0.1600 9.1100 Fig. 8. Normalized fatigue life curve for [0=90]s laminae under in-
plane shear loading conditions.

Fig. 7. Experimental data of static strength and fatigue life of [0=90]s Fig. 9. Normalized residual fatigue strength curve for [0=90]s laminae
lamina under in-plane shear loading with di€erent stress ratios. under in-plane shear loading conditions.
 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035 2033

decreases with the increase of o€-axis angle, while in Figs. 11±13, respectively. It can be seen from these
deviation increases simultaneously. The comparison ®gures that the calculated results for a speci®ed o€-axis
between the calculated static strength and experimental angle under di€erent applied stress levels are very similar
data for the laminate with 30 o€-axis angle shows that in pro®le, so that a master curve can be extracted. If the
the calculated result is 8.9% higher than the experi- normalized residual strength of unidirectional laminates
mental data. The reason for this di€erence could be that with o€-axis angle (is of the same form as Eqs. (2±4):
the Eqs. (33 and 34) only include the matrix failure
mode without the consideration of the ®ber failure
mode, which, though not dominant, also contributes to
the degradation of strength. Because no experimental
data of static strength for 45 and 60 o€-axis laminates
are available for comparison, the calculated results can
only be considered as predictions.
The statistical model [Eq. (33)] is then used to calcu-
late the residual strength of unidirectional laminates
with di€erent o€-axis angles of 30 , 45 and 60 . The
calculated results of residual strength of 30 o€-axis
laminates subjected to fatigue loading with maximum of
70% static strength are shown in Fig. 10. From Fig. 10
it is clearly seen that residual strength degrades with the
increase of fatigue cycles. The dashed lines show the
95% con®dence interval, which means that the expected Fig. 11. Normalized residual strength of [30]16 o€-axis laminate.
value of residual strength lies in the region between two
dashed lines with a probability of 95%. Similar calcula-
tions can be made for other applied stress levels and
di€erent o€-axis angles. These results are converted to
the normalized residual strength de®ned in Eqs. (2±4)
against the normalized fatigue cycles for di€erent per-
centages of applied stress over the static strength, shown

Table 4
Calculated static strength of di€erent o€-axis unidirectional laminates

O€-axis Strength Coecient of Experimental

angle (MPa) variance (%) data (MPa) [7]

30 173.51 5.45 158.00

45 97.34 6.39 Not available
60 67.92 6.94 Not availabe
Fig. 12. Normalized residual strength of [45]16 o€-axis laminate.

Fig. 10. Residual strength of [30]16 o€-axis laminate subjected to an

applied stress with maximum of 70% static strength. Fig. 13. Normalized residual strength of [60]16 o€-axis laminate.
2034 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035

" !a #b1

R …n;  ; † ÿ  log…n† ÿ log…0:25† 

ˆ 1ÿ ÿ

Rs ÿ  log Nf ÿ log…0:25†
…35†

where Rs ; R …n;  ; † and  are the o€-axis static

strength, residual strength and applied stress, the value
of parameters a and b … ˆ 30 ; 45 ; 60 † are deter-
mined, as shown in Table 5. Therefore, if the fatigue life
of an o€-axis laminate under di€erent applied stress
levels is known, the normalized residual strength of the
laminate can be determined based on Eq. (35).
The fatigue life of unidirectional laminate with di€er-
ent o€-axis angles of 30 , 45 and 60 is calculated from
Eqs. (24 and 25). The results are shown in Figs. 14±16,
Fig. 15. Calculated S±N curve and deviations for [45]16 o€-axis laminate.
respectively. The dashed lines in the ®gures present the
95% con®dence interval. The comparison of the calcu-
lated results for the 30 o€-axis laminate in Fig. 14
shows a good agreement with the corresponding
experimental data [21,22]. Most fatigue data are in the
region between the two dashed lines. However, the
model overestimates the static strength due to the rea-
son mentioned above. Because there are no experi-
mental data for 45 and 60 o€-axis laminates available,
the results in Figs. 15 and 16 can only be considered as
predictions, which need to be veri®ed experimentally.

Table 5
Fitting parameters for normalized residual strength of o€-axis lami-
nates in Figs. 13±15
Fig. 16. Calculated S±N curve and deviations for [60]16 o€-axis laminate.
O€-axis angle a b

30 3.272 1.3460

45 7.067 0.5948
60 9.628 0.3148 6. Summary

Based on the experimental results of material char-

acterization for unidirectional laminae subjected to uni-
axial static and fatigue loading, the statistical model is
applied to simulate the fatigue behavior of unidirec-
tional laminae subjected to multiaxial fatigue loading.
With the statistical model, the static strength, residual
strength and fatigue life are calculated for the 30 , 45
and 60 o€-axis unidirectional laminate, respectively.
The results are summarized as follows.

1. The residual strength and fatigue life of an o€-axis

laminate can be determined from the statistical
model if the fatigue behavior of unidirectional
lamina under transverse and in-plane shear load-
ing are known through material characterization.
2. The calculated static strength for the 30 o€-axis
laminate is 8.9% higher than the corresponding
Fig. 14. Calculated S±N curve and deviations for [30]16 o€-axis lami- experimental data because the ®ber failure mode
nate in comparison with experimental results. was not considered.
 X. Diao et al. / Composites Science and Technology 59 (1999) 2025±2035 2035

3. The normalized residual strength of an o€-axis [10] Sims DF, Brogdon VH. Fatigue behavior of composites under
laminate against normalized fatigue cycles for dif- di€erent loading modes. In: Fatigue of ®lamentary composite
materials. ASTM STP-636. Philadelphia: ASTM, 1977. p. 185±
ferent applied stress levels converges to a master
205.
curve, which can be used to determined the residual [11] Hashin Z, Rotem A. A fatigue failure criterion for ®bre rein-
strength of the laminate under any applied stress forced materials. Journal of Composite Materials 1973;7:448±
levels if the corresponding fatigue life is known. 64.
4. The fatigue lives calculated for the unidirectional [12] Rotem A, Hashin Z. Failure modes of angle ply laminates. Jour-
laminates with o€-axis angle 30 show a good nal of Composite Materials 1975;9:191±206.
[13] Rotem A, Hashin Z. Fatigue failure of angle ply laminates.
agreement with corresponding experimental data, AIAA Journal 1976;14:868±72.
indicating the usability of the statistical model. [14] Hashin Z. Fatigue Failure criteria for unidirectional ®ber com-
posites. Journal of Applied Mechanics 1981;14:846±52.
For further development of the statistical model the [15] Hashin Z. Fatigue failure criteria for combined cyclic stress.
stochastic nature of other factors such as fatigue life, International Journal of Fatigue 1981;17:101±9.
[16] Hashin Z, Rotem A. A Fatigue criterion for ®bre reinforced
residual strength degradation of unidirectional lamina materials. Journal of Composites 1985;7:246±56.
under uniaxial loading must be considered. This [17] Aboudi J. Micromechanics prediction of fatigue failure of com-
requires the introduction of more distribution functions posite materials. Journal of Reinforced Plastics and Composites
corresponding to the stochastic variables considered. 1989;8:150±66.
The statistical model could be further developed to [18] Reifsnider KL. The critical element model: A modelling philoso-
phy. Engineering Fracture Mechanics 1986;25:739±49.
include the e€ects of out-of-plane stresses 33 ; 13 and [19] Reifsnider KL, Stinchcomb WW. A critical-element model of the
23 on fatigue life of unidirectional lamina by introdu- residual strength and life of fatigue-loaded composite coupons.
cing a failure criterion including these stresses and the In: Hahn HT, editor. Composite materials: fatigue and fracture,
density distribution functions of static strength of the ASTM STP 907. Philadelphia: American Society Testing and
lamina under uniaxial fatigue loading 33 ; 13 and 23 , Materials, 1986. p. 298±313.
[20] Reifsnider KL, Gao Z. A micromechanics model for composite
respectively. The statistical model remains to be eval- under fatigue loading. International Journal of Fatigue
uated by numerical simulation and comparison with 1991;13:149±56.
experimental data. [21] Shokrieh MM, Lessard LB. Multiaxial fatigue behaviour of uni-
directional plies based on uniaxial fatigue experimental: Part I.
Modeling. International Journal of Fatigue 1997;19(3):201±7.
References [22] Shokrieh MM, Lessard LB. Multiaxial fatigue behaviour of
unidirectional plies based on uniaxial fatigue experimental: Part
[1] Pipes RB, Cole BW. On the o€-axis strength test for anisotropic II. Experimental evaluation. International Journal of Fatigue
materials. Journal of Composite Materials 1973;7:245±56. 1997;19(3)209±17.
[2] Francis PH, Walrath DE, Sims DF, Week DN. Biaxial fatigue [23] Sun CT, Yamada SE. Strength distribution of a unidirectional
loading of notched composites. Journal of Composite Materials ®bre composite. Journal of Composite Materials 1978;12:169±
1977;11:488±591. 76.
[3] Owen MJ, Rice DJ. Biaxial strength behavior of glass fabric [24] Yamada SE, Sun CT. Analysis of laminate strength and its dis-
reinforced polyester resins. Composite 1981:13±25. tribution. Journal of Composite Materials 1978;12:275±84.
[4] Garud YS. Multiaxial Fatigue: A survey of the state of the art. [25] Diao X, Ye L, Mai Y-W. A statistical model of residual strength
Journal of Testing and Evaluation, JTEVA 1981;9:165±78. and fatigue life of composite laminates. Composites Science and
[5] Found MS. A review of the multiaxial fatigue testing of ®ber Technology 1995;54:329±36.
reinforced plastics. In: Miller KJ, Brown MW, editors. Multiaxial [26] Diao X, Ye L, Mai Y-W. Statistical prediction of fatigue failure
fatigue, ASTM STP 853. Philadelphia: American Society for of ®bre reinforced composite materials. Applied Composite
Testing and Materials, 1985. p. 381±95. Materials 1995;2:153±73.
[6] Krempl A, Elzey DM, Hong BZ, Ayar T, Loewy RG. Uniaxial [27] Diao X. A statistical equation of damage evolution. Engineering
and biaxial fatigue properties of thin-walled composite tubes. Fracture Mechanics 1995;52:33±42.
Journal of American Helicopter Society 1988:3±10. [28] Weibull W. A statistical distribution function of wide applic-
[7] Makinde A, Neale KW. A general criterion for low-cycle multi- ability. Journal of Applied Mechanics 1951;18:293±7.
axial fatigue failure. Journal of Engineering Materials and Tech- [29] Melsa JL, Sage AP. An introduction to probability and stochastic
nology 1989;111:263±9. processes. Englewood Cli€s, New Jersey: Prentice-Hall, Inc.,
[8] Garud YS. Multiaxial Fatigue: A survey of the state of the art. 1973. p. 110±88.
Journal of Testing and Evaluation, JTEVA 1981;9:165±78. [30] Lessard LB, Shokrieh MM. Fatigue behaviour of compolsite
[9] Tsai SW. In: Schwartz RT, Schwatz HS, editors. Fundamental pined/bolted joints, Final Report, NRE #32171 and IAR-CR-29,
aspects of ®bre reinforced plastic composites. New York: Wiley Structure and Materials Laboratory, Institute for Aerospace
Interscience, 1968. p. 3±11. Research, National Research Council, Canada, 1995.