Fatigue-life distributions and failure probability for glass-ber reinforced

polymeric composites
Raman Bedi
*
, Rakesh Chandra
Department of Mechanical Engineering, National Institute of Technology, Jalandhar, Punjab, India
a r t i c l e i n f o
Article history:
Received 27 November 2007
Received in revised form 26 June 2008
Accepted 7 September 2008
Available online 19 September 2008
Keywords:
B. Fatigue
Weibull distribution
a b s t r a c t
There has been an upsurge in the application of composite materials in the last few decades, due to high
demands on material performance placed by advanced technologies. Most of these applications include
the situations where degradation of strength and life by fatigue process is most likely. In this investiga-
tion, rotating bending fatigue tests have been conducted on two types of unidirectional glass-ber rein-
forced polymeric composites, having vinylester and Epoxy as matrix materials. It has been observed that
the probabilistic distribution of fatigue-life of these glass-ber reinforced composites, at a particular
stress level, can be modeled by two-parameter Weibull distribution, with statistical co-relation coef-
cient values exceeding 0.90. Various methods have been used to obtain the parameters of Weibull distri-
bution. KolmogorovSmirnov goodness-of-t test has also been used to reinforce the above ndings. The
two-parameter Weibull distribution has also been employed to incorporate failure probability into SN
relationships.
2008 Elsevier Ltd. All rights reserved.
1. Introduction
With the increasing use of composite materials in advanced
applications like aerospace and other tactical applications, accurate
prediction of performance becomes increasingly important. In
majority of these applications fatigue loads are usually unavoid-
able. The effective and efcient use of composites thus requires a
thorough understanding of the fatigue-life, strength and mecha-
nisms. In general fatigue of ber reinforced composite materials
is a quite complex phenomenon and a large research effort is being
spent on it today. Composite materials are inhomogeneous and
anisotropic and their behavior in fatigue is more complicated than
that of homogeneous and isotropic materials like metals. The main
reasons for this are the different types of damage that can occur,
i.e., ber fracture, matrix cracking, ber buckling and their interac-
tions [1]. Fatigue failure in composites is usually accompanied with
extensive damages, which are multiplied through specimen vol-
ume instead of a predominant single crack, which is often observed
in most isotropic brittle materials [2].
Leaving aside the differences in fatigue of composites and met-
als regarding damage mechanisms, it has been observed that fati-
gue process is stochastic in nature in both these cases. The
determination of fatigue strength/life of composites has been pri-
marily based on experimental investigations. Large variability usu-
ally exists in the resulting fatigue strength/life results even at a
given stress level under carefully controlled test procedures. This
variability was quite unimportant in past because of the use of
large safety factor. Advance design however requires accurate
characterization of this variability. Thus the dispersion of fatigue-
life has been a subject of statistical analysis by various researchers
[3,6,7].
Several mathematical models have been employed to study the
statistical dispersion of fatigue-life. One of the popular model has
been the logarithmic-normal (lognormal) distribution function. It
was suggested that the fatigue-life N or some function of N be as-
sumed to be normally distributed [4]. The lognormal distribution
was thus extensively used for this purpose. However, it was
pointed out afterwards that the hazard function or risk function
of lognormal distribution decreases with increasing life or time
[5]. This violates the basic physical phenomenon of progressive
deterioration of engineering materials resulting from fatigue
process.
In assessing the reliability of composite structures, Weibull dis-
tribution function has proved to be useful and versatile means of
describing composite material properties. This is because the prob-
ability density function of the Weibull distribution has a wide vari-
ety of shapes. For example when the shape parameter of the
distribution is equal to 1, it becomes the two-parameter exponen-
tial distribution. For shape parameter nearing 3, the function is
capable of approximating a normal distribution. Thus because of
physically valid assumptions, sound experimental verication, rel-
ative ease in its use and better developed statistics, the Weibull
0266-3538/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2008.09.016
* Corresponding author. Tel.: +91 9815981054.
E-mail address: bedir@nitj.ac.in (R. Bedi).
Composites Science and Technology 69 (2009) 13811387
Contents lists available at ScienceDirect
Composites Science and Technology
j our nal homepage: www. el sevi er. com/ l ocat e/ compsci t ech
distribution has proven to be a useful and versatile means of
describing composite material properties.
Most of the previous studies on fatigue strength/life of polymer
composites has been mainly concerned with unidirectional or an-
gle ply laminates [3,6,7]. Pultruded composites are being increas-
ingly used for various applications like aerial cables, etc. In these
applications large volume fractions (6070%) of continuous glass-
bers are necessary to provide high stiffness and strength. Pultru-
sion is the only method to fabricate the composites with such a
high ber content, which is relatively unusual for other glass-ber
reinforced plastics. A few studies are reported in past on fatigue
strength/life of pultruded GRP rods [8]. Most of these studies have
used the pultruded rods made at laboratories with ber volume
fraction ranging from 15% to 45%, which is quite less as compared
to the ber content in commercially made pultruded rods. It was
thus proposed to carry out fatigue studies on different types of
commercially available glass-ber reinforced pultruded rods.
2. Experimental approach
2.1. Materials
The materials used in this investigation, i.e., standard glass-ber
reinforced composites in the form of pultruded rods have been
manufactured using ECR glass roving as reinforcement and matrix
materials as (a) vinylester and (b) epoxy. These two types of com-
posite rods are referred in this paper as glass-ber reinforced viny-
lester composite (GFRVC) and glass-ber reinforced epoxy
composite (GFREC). Both these pultruded composite rods had a -
ber volume fraction of 62%.
2.2. Specimen details
The glass-ber reinforced pultruded rods have been cut to re-
quired sample size on a water cooled abrasive disc cutter. For car-
rying out the fatigue tests, the pultruded rods have been machined
to the required specimen dimension as shown in Fig. 1 on a CNC
lathe machine. A few of the other researchers have also used the
similar specimen for their research work [8].
2.3. Fatigue tests
All the fatigue tests were conducted on a Rotating bending type
fatigue testing machine, loading the specimen as a cantilever. The
schematics of the experimental setup is shown in Fig. 2. The ma-
chine had a collet chuck mounted on the main spindle to hold
the specimen at one end, while the other end of the specimen
was supported in a oating bearing. Load was applied on this
end of the specimen with the help of dead weights as shown in
Fig. 3. The main spindle of the machine is driven through an elec-
tric motor running at 1440 rpm ($25 Hz). The estimation of static
ultimate tensile strength (r
ult
) of the test specimen is a desirable
prerequisite for the selection of stress levels for fatigue testing.
The Ultimate tensile strength (r
ult
) values for GFREC, and GFRVC
along with the Fatigue testing stress levels are listed in Table 1.
The fatigue tests have been conducted at four different stress levels
of r = 227 MPa, r = 284 MPa, r = 341 MPa and r = 395 MPa, which
fall in the range of 3575% of the ultimate tensile strength (r
ult
) for
the two materials under investigation.
3. Mode of failure
The mode of failure of GFRC pultruded samples during rotating
bending tests is discussed below for two separate cases, i.e., (i) Fa-
tigue at stress levels of 3550% of Ultimate Tensile Strength (r
ult
)
and (ii) Fatigue at stress levels of 6070% of the Ultimate Tensile
Strength (r
ult
);
(i) It is observed that longitudinal cracks are produced parallel
to the ber direction at the specimen waist close to the
machine grip. These cracks are produced due to the machin-
ing of the specimens which produces discontinuous bers at
the waist. The cracks thus produced at the interface of dis-
continuous and continuous bers lead to de-bonding at the
interfacial surface, resulting in an increase in parallel portion
of the specimen, which extended from section AA to BB
under the specimen waist as shown in Fig. 4. Fig. 5a shows
the SEM image of the failed GFREC specimen tested at
227 MPa. De-bonding and matrix cracking is clearly visible.
Similar damage pattern has been reported earlier in other
studies on pultruded composites [8]. This de-bonding along
Nomenclature
CV coefcient of variation of the data sample under consid-
eration
E expectation
f(n) probability distribution function
F(n) cumulative distribution function
GFRC glass-ber reinforced composite
GFRVC glass-ber reinforced vinylester composite
GFREC glass-ber reinforced epoxy composite
i order number of the data point in a sample under con-
sideration
k total number of data points in a sample under consider-
ation or sample size
L
R
survivorship function/survival probability/reliability
function
n number of cycles
N number of cycles to failure or fatigue-life.
P
f
failure probability
r statistical co-relation coefcient
u characteristic life or scale parameter of Weibull distri-
bution
V
f
volume fraction of glass-bers in composite
a shape parameter of the Weibull distribution or Weibull
slope
r applied fatigue stress level
r
s
standard deviation of the data sample under consider-
ation
r
ult
ultimate tensile strength
C() gamma function
l mean value of the data sample under consideration
18
21 18
130
12
7
Fig. 1. Dimensions of the specimen used for fatigue testing (mm).
1382 R. Bedi, R. Chandra / Composites Science and Technology 69 (2009) 13811387
with cracking of the matrix weakens the material and the
full load is now carried by bers only. It is observed that
most of the specimens then fail catastrophically due to
shearing of the bers at a plain normal to the ber direction
at section BB in Fig. 4 which is having the highest stress.
SEM image of cross-section of failed GFREC specimen tested
at 227 MPa shown in Fig. 5b clearly shows failure of bers
normal to ber direction. Fig. 5c shows SEM image of the
cross-section of the failed GFRVC specimen tested at
227 MPa. Separation between the bers due to de-bonding
at the ber matrix interface is clearly visible. Fiber pullout
is also visible at certain sites, which is observed very clearly
in Fig. 5d. The result is in contradiction to that reported in
other studies [8] where no ber pull out has been observed.
(ii) A very different damage mechanism has been observed for
this case. Longitudinal cracks are observed in the specimen
which rapidly travel along the length of the specimen. As
the longitudinal cracks in the matrix propagate rapidly, no
increase in length of the parallel portion of the specimen is
observed as compared to that observed at low stress levels
as shown in Fig. 6a. It is observed that the specimen then
fails catastrophically due to failure of bers. Fig. 6b shows
SEM image of the fractured GFRVC specimen tested at
395 MPa. Matrix splitting and longitudinal cracks are clearly
visible. No ber pullout is observed at high stress levels.
The above two modes are observed in both the materials under
investigation, i.e., GFREC and GFRVC specimens.
4. Analysis and discussion of fatigue test results
The complete fatigue-life data thus obtained at various stress
levels for two different test materials, i.e., GFRVC and GFREC each
having ber volume fraction of 62% is listed in ascending order
in Table 2 and Table 3, respectively.
4.1. Rejection of outliers
Some data points in Tables 2 and 3 may deserve consideration
for rejection as outliers. Chauvenets criterion [9] was applied to
the data points at all the stress levels tested in this investigation,
and data points meeting this criterion for rejection were identied
and excluded from further analysis. A few other researchers have
also used the same criterion for rejection of outliers in their work
on fatigue [10].
4.2. Statistical analysis of fatigue-life data
The statistical analysis of fatigue-life results was carried out
using a two-parameter Weibull distribution function which is
characterized by a probability density function (PDF), f(n); and
the cumulative distribution function (CDF), F(n) as follows:
f n
a
u
n
u

exp
n
u

1
Fn 1 exp
n
u

a

2
in which n = specic value of the random variable N; a = shape
parameter or Weibull slope at stress level r and u = scale parameter
or characteristic life at stress level r. Various methods like graphical
method, method of moments, maximum likelihood method of
obtaining the parameters a and u are available. In this investigation,
rst the graphical method is employed to show that the statistical
distribution of fatigue-life of glass-ber reinforced pultruded com-
posites at a certain stress level r follows the two-parameter Wei-
bull distribution and then the parameters of the Weibull
distribution are obtained by the graphical method as well as by
method of moments.
4.3. Analysis of fatigue-life data by graphical method
The probability of survival or survivorship function or reliability
function, L
R
(n), may be dened as L
R
(n) = 1 F(n), and substituting
this value of F(n) in Eq. (2) it is modied to:
GRIP
LOAD
Specimen
Fig. 2. Schematic diagram of fatigue testing setup.
Fig. 3. A view of the fatigue testing setup.
Table 1
Ultimate tensile strength and fatigue test levels for GFREC and GFRVC
Material GFREC GFRVC
Ultimate tensile Strength (r
ult
) 571 MPa 654 MPa
Fatigue stress parameters: 227 MPa (40%) 227 MPa (35%)
Fatigue stress MPa (% (r
ult
)) 284 MPa (50%) 284 MPa (43%)
341 MPa (60%) 341 MPa (52%)
395 MPa (70%) 395 MPa (60%)
Fig. 4. Failed GFRVC specimens tested at 227 MPa.
R. Bedi, R. Chandra / Composites Science and Technology 69 (2009) 13811387 1383
L
R
n exp
n
u

a

3
taking the logarithm twice of both sides of Eq. (3), it can be rewrit-
ten as
ln ln
1
L
R

alnn alnu 4
Eq. (4) represents a linear relationship between ln[ln(1/L
R
)] and
ln(n). In order to obtain a graph from Eq. (4), the fatigue-life data
corresponding to a particular stress level are rst arranged in
ascending order of cycles to failure and the empirical survivorship
function L
R
for each fatigue-life data at a given stress level is ob-
tained from the following relation [9]:
L
R
1
i 0:3
k 0:4
5
where i denotes the failure order number and k represents the num-
ber of data points in a data sample under consideration at a partic-
ular stress level r. The empirical survivorship function in the form
of ln[ln(1/L
R
)] for each fatigue-life data is then plotted on a graph
with the corresponding fatigue lives ln(N). If a linear trend is estab-
lished for the data points, the best t line is drawn using method of
least squares. It can then be assumed that fatigue-life data for that
particular stress level follows the two-parameter Weibull distribu-
tion. The slope of the line provides an estimate of shape parameter a
and the characteristic life u can be obtained as that value of n which
corresponds to L
R
= 0.368.
One such typical graph for fatigue-life data at stress level
r = 227 MPa for GFRVC is shown in Fig. 7. The approximate straight
line plot in Fig. 7 with statistical correlation coefcient r of 0.94,
indicates that the two-parameter Weibull distribution is a reason-
able assumption for the statistical distribution of fatigue-life at this
stress level. Similar trends are observed for GFRVC pultruded rods
at r = 284 MPa, r = 341 MPa and r = 395 MPa with correlation
coefcient exceeding 0.9. The parameters obtained by this method
are listed in Table 4. The fatigue-life data for GFREC pultruded
composites at different stress levels was also analysed and it has
been shown to follow two-parameter Weibull distribution with
statistical correlation coefcient values exceeding 0.9. The esti-
mated parameters are listed in Table 5.
4.4. Parameter estimation by method of moments
Estimating parameters by method of moments requires esti-
mating the appropriate sample moments, such as sample mean
and sample variance. The moments of Weibull distribution may
be written in following forms
En uC
1
a
1

6
and En
2
u
2
C
2
a
1

7
Fig. 5. (a) SEM image of failed GFREC specimen tested at 227 MPa. (b) Cross-sectional SEM images of failed GFRVC specimen tested at 227 MPa. (c) Cross-sectional SEM image
of failed GFRVC specimen tested at 227 MPa showing de-bonding. (d) SEM image of failed GFRVC specimen tested at 227 MPa showing ber pullout.
1384 R. Bedi, R. Chandra / Composites Science and Technology 69 (2009) 13811387
where U() is the gamma function and E denotes expectation. Noting
that the mean of the data sample under consideration at a given
stress level r, l = E(n) and the variance of the data sample under
consideration at a given stress level r, r
2
s
En
2
l
2
, Eq. (6) and
Eq. (7) give
r
s
l

2

C
2
a
1

C
1
a
1
1 8
where r
s
/l = CV is the coefcient of variation of the fatigue data
sample under consideration at a given stress level r and r
s
is the
Fig. 6. (a) Failed GFRVC specimens tested at 395 MPa. (b) SEM image of failed
GFRVC specimen tested at 395 MPa.
Table 2
Laboratory fatigue-life data for glass-ber reinforced vinylester (GFRVC)
Stress level r
227 MPa (35% r
ult
) 284 MPa (43% r
ult
) 341 MPa (52% r
ult
) 395 MPa (60% r
ult
)
Cycles to failure (N)
264,123 29,437 4841 227
*
408,337 31,878 6597 2895
604,208 45,731 6599 3214
750,771 48,043 6870 3645
771,391 67,333 8867 3981
901,212 67,641 8921 4053
10,41,682 67,688 8937 4408
10,84,983 76,994 13,356 4899
12,39,046 115,230 13,478 6452
19,85,623 197,448 15,625 7131
218,230 27,336 8562
55,023
*
*
Rejected as outlier by Chauvenets criteria, not included in analysis.
Table 3
Laboratory fatigue-life data for glass-ber reinforced epoxy (GFREC)
Stress level r
227 MPa 0 (40% r
ult
) 284 MPa (50% r
ult
) 341 MPa (60% r
ult
) 395 MPa (70% r
ult
)
Cycles to failure (N)
154,213 22,103 3826 3315
225,196 25,534 4707 3335
351,227 36,297 6102 3619
479,771 52,915 6964 3746
485,623 53,576 7613 4306
498,335 56,036 7923 5102
652,312 66,325 8047 5131
956,213 100,230 9635 5156
965,231 126,319 10,021 5211
11,56,231 133,392 17,556 5501
16,52,312 182,454 18,963 6215
Fig. 7. Graphical analysis of fatigue-life data for GFRVC.
R. Bedi, R. Chandra / Composites Science and Technology 69 (2009) 13811387 1385
standard deviation of the data sample under consideration at a gi-
ven stress level r. Since it is difcult to obtain the value of shape
parameter a from Eq. (8), it can be reduced to a simple expression
as follows [11]:
a CV
1:08
9
The characteristic life u can be estimated from Eq. (9) by substi-
tuting l for E(n) as follows:
u
l
C
1
a
1
10
The parameters obtained by this method for GFRVC and GFREC
at different stress levels are also listed in Table 4 and 5, respec-
tively. It is seen from Table 4 and 5 that higher shape parameters
are observed at higher stress levels, indicating less scatter in the fa-
tigue-life data. Similar results have been reported earlier by other
investigators [3].
4.5. KolmogorovSmirnov test for goodness-of-t
In the preceding sections, graphical method has been employed
to show that the statistical distribution of fatigue-life of various
types of glass-ber reinforced composites, at various stress levels,
can approximately be described by the two-parameter Weibull dis-
tribution. The KolmogorovSmirnov test [9] has been applied to
validate the results, which can be carried out by using the follow-
ing equation:
D
i
max
k
i1
jF

x
i
Fx
i
j 11
where F
*
(x
i
) = i / k = observed cumulative histogram, i = order num-
ber of the data point, k = total number of data points in the sample
under consideration at a given stress level and F(x
i
) = hypothesized
cumulative distribution given by Eq. (2). The value of D
i
thus ob-
tained is compared with D
c
values for each data set obtained from
KolmogorovSmirnov Table [9]. If D
i
< D
c
The model is acceptable
with 5% signicance level. This goodness-of-t test was applied to
fatigue-life data of GFRVC and GFREC at all the stress levels and it
was found that the model was acceptable at 5% level of signicance.
The results are compiled in Table 6.
4.6. Failure probability and SN relationships
The SN relationships reported by earlier investigators do not
incorporate failure probability P
f
, which is an important parameter
in fatigue studies. The Weibull distribution has been employed
here to incorporate the failure probability into SN relationships
for GFRCs. Substituting 1 P
f
= L
R
in Eq. (4), the following relation
is obtained
ln ln
1
1 P
f

alnn alnu 12
rearranging
n ln
1
ln ln
1
1P
f

alnu
a

13
Thus, using the average values of the parameters of Weibull dis-
tribution for fatigue-life at any stress level (as listed in Tables 4 and
5), Eq. (13) has been used to calculate the fatigue lives at a given
Table 4
Values of the Weibull parameters for fatigue-life of GFRVC
r = 227 MPa (35% r
ult
) r = 284 MPa (43% r
ult
) r = 341 MPa (52% r
ult
) r = 395 MPa (60% r
ult
)
a u a u a u a u
Graphical method 1.92 10,35,091 1.6 98,716 2.106 12,582 2.963 5486
Method of moments 1.96 10,21,138 1.41 96,618 1.81 12,430 2.867 5525
Average 1.94 10,28,115 1.505 97,667 1.958 12,506 2.915 5506
Table 5
Values of the Weibull parameters for fatigue-life of GFREC
r = 227 MPa (40% r
ult
) r = 284 MPa (50% r
ult
) r = 341 MPa (60% r
ult
) r = 395 MPa (70% r
ult
)
a u a u a u a u
Graphical Method 1.55 790,167 1.6 89,321 2.2 10,404 4.864 5014
Method of Moments 1.58 767,330 1.57 86,513 2.01 10,397 5.297 4983
Average 1.56 778,749 1.585 87,917 2.105 10,400 5.080 4999
Table 6
Goodness-of-t test results
Stress level D
i
D
c
Remarks
Material GFRVC
r = 227 Mpa (35% r
ult
) 0.1435 0.41 Accepted
r = 284 Mpa (43% r
ult
) 0.2380 0.39 Accepted
r = 341 Mpa (52% r
ult
) 0.2511 0.39 Accepted
r = 395 Mpa (60% r
ult
) 0.1928 0.41 Accepted
Material GFREC
r = 227 Mpa (40% r
ult
) 0.1584 0.39 Accepted
r = 284 Mpa (50% r
ult
) 0.1737 0.39 Accepted
r = 341 Mpa (60% r
ult
) 0.2164 0.39 Accepted
r = 395 Mpa (70% r
ult
) 0.1177 0.39 Accepted
Fig. 8. P
f
SN diagram for GFRVC.
1386 R. Bedi, R. Chandra / Composites Science and Technology 69 (2009) 13811387
stress level corresponding to different failure probabilities. These
calculated values of fatigue lives for GFRVC and GFREC for failure
probabilities of P
f
= 0.2, 0.3, 0.5, 0.8 and 0.95, are plotted in Figs.
8 and 9 for GFRVC and GFREC, respectively, to obtain P
f
SN
diagrams.
5. Conclusions
The following conclusions can be drawn from this investigation:
1. For both the materials investigated in this study, different
modes of failure are observed at low and high fatigue stress lev-
els, which establish different damage mechanisms.
2. The probabilistic distribution of fatigue-life for glass-ber rein-
forced vinylester (GFRVC) and glass-ber reinforced epoxy
(GFREC) pultruded composites at any stress level can be
approximately modeled by two-parameter Weibull distribution
with statistical correlation coefcient values exceeding 0.9.
3. The shapes of the Weibull distribution for fatigue-life of glass-
ber reinforced composites (GFRC) are different for different
levels of the applied fatigue stress. Both the methods of estimat-
ing the parameters of Weibull distribution yield almost similar
results.
4. Higher shape parameters are observed at higher stress levels,
indicating less scatter in the fatigue-life data indicating a more
uniform damage mechanism.
5. Using two-parameter Weibull distribution, the P
f
SN rela-
tionships have been generated for glass-ber reinforced com-
posites (GFRCs), incorporating failure probability into SN
relationships. These relationships can be used by the design
engineers to obtain the fatigue strength of GFRCs for the
desired level of failure probability.
References
[1] Joris D, Paepegam WV. Fatigue damage modeling of ber reinforced composite
materials review. Appl Mech Rev 2001;54(4):279300.
[2] Tsai SW. Composite design, 4th ed. Think Composites; 1988.
[3] Lee J, Harris B, Almond DP, Hammett F. Fibre composite fatigue life
determination. Composites A 1997;28:515.
[4] ASTM Special Publication 91-A, A guide for fatigue testing and the statistical
analysis of fatigue data. ASTM 1963; 83.
[5] Gumble EJ. Parameters in the distribution of fatigue life. J Eng Mech, ASCE
1963:4563.
[6] Philippidis TP, Vassilopoulos AP. Fatigue design allowable for GRP laminates
based on stiffness degradation measurements. Comp Sci Tech
2000;60:281928.
[7] Tai NH, Ma CCM, Wu SH. Fatigue behaviour of carbon ber/PEEK laminate
composites. Composites 1995;26:5519.
[8] Khashaba UA. Fatigue and reliability analysis of unidirectional GFRP
composites under rotating bending loads. J Comp Mater 2003;37:31730.
[9] Kennedy JB, Neville AM. Basic statistical methods for engineers and
scientists. A Dun-Donnelley Publishers; 1986. 613.
[10] Singh SP, Kaushik SK. Flexural fatigue life distributions and failure probability
of steel brous concrete. ACI Mater J 2000;97-M74:65867.
[11] Wirsching PH, Yao JTP. Statistical methods in structural fatigue. J Struct Div
Proc ASCE 1970:120119.
Fig. 9. P
f
SN diagram for GFREC.
R. Bedi, R. Chandra / Composites Science and Technology 69 (2009) 13811387 1387