The Relationship Between Tensile Strength and

Flexure Strength in Fiber.reinforced Composites,

Flexure- and tensile-coupon
are compared

data on unidirectional

to a Weibull two-parameter

graphite-epoxy

statistical-strength

composites

model

by J.M. Whitney and M. Knight

ABSTRACT--Tensile
data o n unidirectional composites
generated from a flexure test usually yield a higher strength
than observed from a standard tensile coupon. According to
a statistical-strength theory based on a Weibull distribution,
the presence of a stress gradient in the flexureqest results in
an apparent increase in tensile strength as compared to the
tensile test under uniform stress. In the present paper, this
concept is explored by utilizing data from unidirectional
graphite-epoxy composites to compare with theoretical
results generated from a two-parameter Weibull distribution. A
larger variation in tensile strength is observed from tensilecoupon data than from flexure data. Such differences are not
in accordance with strength theories based on a uniform
flaw distribution and raise questions concerning variability of
the test methods, as well as sources of material variability.

List of Symbols
B ( S ) = risk of rupture at stress S
b = width of test specimen, mm (in.)
L = length of tensile specimen or span of beam
h

S=
Sb =
So =So
Sb =
S t .=

X=
P~o =

Xo=

specimen, mm (in.)
thickness of test specimen, mm (in.)
maximum stress, MPa (psi)
characteristic bending strength, MPa (psi)
scale parameter in two-parameter Weibull
distribution
maximum likelihood estimate of So
adjusted maximum likelihood estimate for scale
parameter under bending load, MPa (psi)
characteristic tensile strength, MPa (psi)
adjusted maximum-likelihood estimate for scale
parameter under tensile load, MPa (psi)
normalized strength
scale parameter for normalized strength
maximum likelihood estimate of normalizedscale parameter
shape parameter in Weibull distribution

J.M. Whitney and M. Knight are Materials Research Engineers, Nonmetallic" Materials Division, Air Force Materials Laboratory, WrightPatterson Air Force Base, OH 45433.
Paper was presented at 1979 SESA Spring Meeting hem in San Francisco,
CA on May 20-25.
Original rnanuscripl submitted: October 12, 1979. Final version received:
January 29, 1980.

= maximum likelihood estimate of shape parameter

& = unbiased estimate of pooled-shape parameter for
tensile loading
~b = unbiased estimate of pooled-shape parameter
for bending

Introduction
Tensile data on unidirectional composites are often
used as one of the key factors in materials selection, and
also provide basic ply properties which are used in
laminate design. Such data generated from a unidirectionalflexure test usually yield higher strength than data obtained from a standard tensile coupon. It is primarily for
this reason that flexure data are not considered appropriate
for design purposes. This difference in apparent tensile
strengths can be accounted for, however, if one considers
the brittle nature of most polymeric-matrix composites.
In particular, a statistical-strength theory based on a
Weibull distribution' can be used to explain the difference
between unidirectional-tensile data generated from a
flexure test and a standard tensile coupon. Tbe presence
of a stress gradient in the flexure-test results in an apparent
increase in tensile strength compared to the tensile test
under uniform stress. Establishment of a viable relationship between the flexure test and standard-tensile-coupon
test would provide a potential basis for use of the flexure
test in the generation of unidirectional-design data. Since
flexure tests are easy to run and relatively inexpensive, a
large statistical-data base obtained with this method rather
than tensile coupons is far more economical.
A two-parameter Weibull model was used by Bullock 2
in correlating 4-point flexure and tensile-coupon data for
unidirectional graphite-epoxy composites.
Excellent
agreement was obtained between theory and experiment.
The Weibull model has been previously applied to ceramic
materials '.4 and more recently to randomly oriented
short-fiber composites.'
In the present work, unidirectional graphite-epoxy
tensile data are obtained on both 3-point and 4-point
flexure tests, as well as on straight-sided tensile coupons.
The influence of specimen thickness on tensile strength is
investigated in addition to the effect of stress gradient.
Thus, a much broader data base for comparison to
Weibull statistical theory is available in the current work
than presented by Bullock. 2 Unlike the experimental

Experimental Mechanics 9 211

results discussed in Ref. 2, however, a significantly larger

variation in tensile strength vs. flexure strength is obtained
with the current data. This trend is observed in two
entirely different graphite-epoxy material systems. Such
differences are not in accordance with statistical-strength
theories based on a uniform flaw distribution. Possible
sources of this departure from classical brittle-failure
theory are discussed in detail.

TABLE 1--TEST MATRIX

Material

Tension

4-PT Flex.

3-PT Flex.

T300/5208,8 Ply
T300/5208,16 Ply
AS/3501-5,16 Ply

25
20
36

--

28
25
28

21
30

Statistical.strength Model
According to the Weibull statistical-strength theory for
brittle materials,' the probability of survival, P , at a
maximum stress level S for a uniaxial-stress field in a
homogeneous material governed by a volumetric flaw
distribution is given by
>_ S ) = R ( S )

P(S/

= exp [ - B ( S ) ]

B(S)

(5)

SA = SoA -'/~

(6)

where

(1)

where S/is the value of the maximum stress at failure, R

is the reliability, and B is the risk of rupture. A nonuniform stress field, o, can always be written in terms of
the maximum stress in the following manner

and the reliability function, eq (1), can be written as a

two-parameter Weibull distribution

(2)

Thus, tensile tests from specimens containing different

stress fields can be represented by a two-parameter Weibull distribution with the same shape parameter, but with
a scale parameter which will shift according to eq (6).
For the case of a simple tensile test under uniform
stress, eq (6) takes the form

o(x,y,z)

= Sf(x,y,z)

For a two-parameter Weibull model, the risk of rupture is

of the form
B(S)

= A

(So, c~> 0)

(3)

o0

e(s) = exp [ - (~)~

(4)

A = Jv [ f ( x , Y , Z ) l ' ~ d V

and So is the scale parameter, sometimes referred to as the

characteristic strength, and c~ is the shape parameter
which characterizes the flaw distribution in the material.
Both of these parameters are considered to be material
properties independent of size. Thus, the risk to break will
be a function of the stress distribution in the test specimen.
Equation (3) can also be written in the form

where the subscript t denotes simple tension. Thus, the

scale parameter for uniform tension is a function of
specimen volume.
For flexural loading, the integration in eq (4) can be
performed in closed form and results in the following
relationships between the scale parameters for tension
and flexure
__

Sb
S,

~L/2

"q J L

P/2
~_29cm(9in )

P/2

t
P/2

P/2

L/h=52,
b=l.27cm(O.5in.)
h = 0 0127 cm/PLY ( 0 0 0 5 in / PLY )

b = 1.27 cm (0.5 in.)

Fig: 1 - - G e o m e t r y of test s p e c i m e n s

212

June 1980

]]l/,~

)]

O-point)
~,~

(4-point)

(9)
(10)

where the subscript b denotes bending. The results for

4-point flexure correspond to loading at quarter points.
In order to illustrate the effect of nonuniform stress
distribution, consider the case V, = Vb. For values of
c~ = 15 and 25, eqs (9) and (10) yield

--

Sb
=
St

1.52, a = 15
1.33, c~ = 25

(3-point)

(11)

Sb
=
St

1.31, a = 15
1.20, c~ = 25

(4-point)

(12)

--

P/2
J-

~.~-b -,,.{

Vt

[2 (c~ + 1)2 . - - ~ -

S~ _ [4(c~+ 112
S,
( a + 2)

L=15.2 cm ( 6 i n )

(8)

SA = S, = S o ( V , ) -''~

where

if- b.~

(7)

These values of c~ are typical of currently utilized composites such as glass-epoxy and graphite-epoxy. Thus, the
flexure test can, in theory, produce significantly higher
tensile strengths than the tensile test, with the 3-point
loading producing the highest strength. This is due to the
fact that the maximum stress is produced at the outer
surface in the center of the beam, while the 4-point
loading produces the maximum stress at the outer surface
throughout the center section. In particular, the smaller

TABLE 2--WEIBULL PARAMETERS

Test

..T300/5208,
Si, MPa (kSl)

8 Ply,
A
oq

Tension
Tension *
Tensiont
4-Pt Flex
4-Pt Flex*
4-Pt Flext
3-Pt Flex
3-Pt Flex*

1790 (259)
1776 (257)
.
--.
2377 (344)
2377 (344)

17.7
20.4
.

.
---

16 Ply,

^T300/5208,
Sol, MPa (kSl)

.
41.4
42.6

1665
.
.
1734
1741
.
1790
1797

(~i

(241)

^AS/3501-5,
Sol, MPa (kSI)

16 Ply,
^i
ot

1506 (218)

13.3

1506 (218)
1624 (235)
-1624 (235)
1617 (234)
--

13.2
29.2
-32.7
22.9
--

18.5
.

(251)
(252)

29.3
28.7

(259)
(260)

36.7
36.2

*Based on a reduced sample size of 20,

tBased on a reduced sample size of 28.

the volume under maximum stress, the higher the local

strength.
It should be noted that eqs (9) and (10) are based on
the assumption that failure in the flexure test is a direct
function of normal stress on the tension side of the beam.
The effect of interlaminar stresses is neglected.
Specimen-thickness effects, as well as stress-gradient
effects, are also of interest. For pure tension, eq (8)
becomes
(13)

S, = S o ( L b h ) . . . .

where L , b and h are gage length, width and thickness,

respectively, of the tensile coupon. For specimens of
thickness h, and h~, eq (13) yields
S,,
S,~

= (h,},,,,
.~,.
,

h~ > h,

(14)

Thus, the thin specimens will have a higher characteristic

strength compared to the thick specimens. In the case of
flexural loading, the beam span, L, must also be adjusted
for any thickness change in order to assure a constant
span-to-depth ratio in the flexure test. For 3-point loading,
eq (6) becomes
Sb = So [ 2(oe + 1)~_ ( ? ) ] , / , *
bh 2
L

(3-point)

percent for T300/5208 and 65 percent for AS/3501-5A.

Both 8-ply and 16-ply panels were processed for T300/
5208, while only 16-ply panels were fabricated for A S /
3501-5A. Test specimens were cut from the large panels
with a diamond wheel.
Specimen geometry and dimensions for both tension
and flexure are shown in Fig. 1. For tension, a straightsided coupon was utilized in accordance with ASTM
Standard D-30397 The flexure tests were run in accordance with ASTM Standard D-7907 with the following
deviations. Loads were applied at a distance of L / 4 from
the supports, rather than at a distance of L / 3 as required
by the ASTM standard. In addition, the specimens were
13 mm (0.5 in.) wide rather than 25 mm (1.0 in.). These
deviations have become accepted practice for graphiteepoxy composites. A test matrix is shown in Table 1.
The Weibull parameters for each data set were estimated
from the maximum likelihood estimator (MLE). 8 A twosample test" was utilized for testing the equality of shape
parameters in a two-parameter Weibull distribution with
unknown scale parameters.
Weibull parameters are shown in TabJe 2 for each data
1.0

(15)

TENSION

0 8 L

" v ~

[U]['"

For specimens of thickness h, and hz, with L / h constant,

eq (15) yields

&~

U~"

h, > h,

(16)

Because of the requirement for a constant L / h ratio, any

thickness change will have greater effect on the flexure
test than on the tensile-coupon test. Again, thin specimens
should yield a higher characteristic strength than thick
specimens. It is also obvious that eq (16) holds for 4point bending as well as 3-point bending.

06

2~
0 4

=~184

X.

0.2

Experimental Procedure and Data Reduction

Two graphite-epoxy material systems were chosen for
this investigation, T300/5208 (Narmco) and A S / 3 5 0 1 - 5 A
(Hercules). Unidirectional panels were fabricated in an
autoclave according to each manufacturer's recommended
cure cycle. The average fiber-volume content was 70

I
08

1
09

I
1.0

Fig. 2 - - P o o l e d Weibull distribution for tension loading,

T300/5208 graphite-epoxy c o m p o s i t e s

Experimental

Mechanics

9 213

S(MPa)
1400

I0

1600

1800

2000
I

TENSION
T500/5208

tol

0.8
St = !665 MPa

Fig. 3--Weibulldistributionsfor tension

loading,T300/5208graphite-epoxy
composites

'~

X ~

06
"~

~ / - S ~ :I790MPa
(259 KSI)

u3
0.4
5~ =18 4
8 PLY

0.2

2;o

0171

:500

S(KSI)

set. In order to use the two-sample test results tabulated

in Ref. 9, it was necessary to reduce the number of
replicates in some data sets so that equal sample sizes
could be obtained within each material system. This was
accomplished by numbering the failures in each data set
to be reduced from lowest strength to highest strength
and using a table of random numbers to discard the
appropriate number of specimens. WeibuU parameters
were then determined from the reduced sets by use of
MLE. The resulting shape parameters, ~i, represented
estimates for equal sample sizes within each material
system and the tabulated data from Ref. 9 could then be
applied. Weibull parameters associated with equal-samplesize sets are also shown in Table 2.
For cases in which the shape parameters satisfied the
two-sample test criterion for a confidence level of 98
percent, a data-pooling technique was used to estimate a
shape parameter for the pooled population. The approach
utilized in the present paper involves normalizing each
data set included in the pooling procedure by its estimated
characteristic strength ~~ and the resulting normalized
strength data was fit to a two-parameter Weibull distribution, again utilizirig MLE. For a perfect fit to the
data-pooling scheme, the scale parameter of the pooled
distribution should be exactly unity. Each scale parameter
of the individual distributions were adjusted, however, to
produce an exact value of unity for the pooled-scale
parameter. In particular
A

rSpi = X o S o i

(17)

where Spi, Xo, and So i are adjusted values of the scale

parameter for the ith data set, the MLE estimate of the
pooled-location parameter, and the MLE estimate of the
location parameter for the ith data set, respectively. It
should be noted that MLE is asymptotically unbiased,
i.e., it is a biased estimator for small sample sizesP Unbiasing factors are tabulated in Ref. 11. Values of pooled
parameters were adjusted for bias and denoted by &v-

I0 ~

BENDING ~

0.8 ~

[0]

06

T3_OOJ52o8

o ss

090

Application of the two-sample test to all of the data

within each material system failed to indicate a constant
value of a i . Application of the two-sample test to flexure

214 9 June 1980

095

"~,q~

data and tensile-coupon data separately, however, indicated

that pooling procedures would be appropriate for each of
these test methods. Since tensile-coupon data on AS/
3501-5A composites were obtained for only one specimen
geometry, data pooling could only be accomplished on
flexure strength for this material.
Comparison between strength data and Weibull distributions obtained from the data-reduction procedures
are shown in Figs. 2-9. Data points are converted to
probabilities of survival from the "median rank" (MR)
defined as
MR

Discussion of Results

too
1.05
x
Fig.4--PooledWeibulldistributionfor flexuralloading,
T300/5208graphite-epoxycomposites
00%

0.3
n+0.4

j -

(18)

where j is the survival order number (data listed in decreasing order of strength) and n is the total number of
samples tested. Pooled-shape parameters are denoted by

1600

I 014OO

O
\O

~_\

06

2200

~_~ g

O 3PT, 16 PLY

r~c~

~ 3PT, 8PLY

gb =1734MP(]

2400

BENDING
T300/5208

\o~

08

04

S(MPo)
2000

1800

~] OI

Sb=2377 MISo

~[3 O~r

Fig. 5--Weibull distributions for flexural

loading, T300/5208 graphite-epoxy
composites

"~

= 1790 MPo

02

0
200

225

,,L,

250

\q

275
StKSI)

300

325

I ~
350

S(MPa)
I100

1.0

1300

1500

1700

TENSION
0.8
AS/3501-5A
0.6

~)~

?,--128

g,

: 1499 MPa

Fig. 6--Weibull distribution for tension

loading, AS/3501-5A graphite-epoxy
composites

~3

(217 KSI)
04

0.2

~0

150

175

~ '- ~ 0 ~

O608 ~

m
N

I-~

200
S(KSI)

225

ab :251

04

02

O
O 80

I
O 85

J
090

I,
095
X

I
IOO

105

Fig. 7--Pooled Weibull distribution for flexural loading,

AS/3501-5A graphite-epoxy composites

JO

250

~t and ~b where the subscripts l and b denote tension and

bending, respectively.
Note that, for both graphite-epoxy material systems,
the ratio of bending-shape parameters to tension-shape
parameters, ~b/~,, is approximately 2, which is a departure
from classical Weibull theory. The characteristic flexure
strengths are, however, consistently higher than the
characteristic tensile strengths as predicted by the Weibull
failure model.
The difference obtained in shape parameters between
tensile coupons and flexure tests suggests that their
failures are governed by two different flaw distributions.
Typical failure modes for tension and flexure demonstrate
the same brooming type of failure mode for both loading
methods. Tension tends to produce a more catastrophic
failure due to the uniform stress field, while flexure
loading produces a more localized failure due to stress
gradients. Similar modes could lead one to believe that the
failures are governed by the same flaw distribution. This
can, however, be misleading as further discussion will
show.

Experimental Mechanics 9 215

S(MPe)
1500

1300
I0,

I ~

(3 I 0

1900

~Q.~
~

0.8

1700

BENDING
(4PT)

AS/3501-5A~ _
E

Sb=1624M P o ~
(235 KSI)
~

04
02
i

017!

200

225
S(KSI)

.250

275

Fig. 8--Weibull distribution for 4-point flexural loading,

AS/3501-5A

graphite-epoxy

composites

S(MP~)
1300
p

tO

1500
I

08
O6
0.4

BENDING
(SPT)

~
AS/SSOL-~

1900

17OO
I

Sb=1617MPo
(234KS~1

The authors wish to acknowledge Ran Cornwall, R a n

Esterline, and Chuck Fowler of the University of Dayton
Research Institute for the fabrication and testing of
composite specimens.

(~
~

References
I

200

225
S(KSI)

250

275

Fig. 9--Weibull distribution for 3-point flexural loading,

graphite-epoxy

composites

A possible source of apparent scatter in tensile-coupon

data is specimen misalignment which induces bending
and/or a nonuniform stress field. The straight-sided
geometry associated with composite tensile coupons makes
them particularly sensitive to misalignment, with unidirectional composites being the most sensitive due to the
high ratio of axial to transverse strength and stiffness. It
can easily be seen from eqs (3) and (4) that a constant
nonuniform stress field will change the characteristic
strength but not the shape parameter. Misalignment,
however, is likely to induce a nonuniform stress field
which varies from specimen to specimen depending on the
degree and nature of the misalignment. Such variations
can reduce the estimated value of the shape parameter,
~g, by producing artificially large scatter in measured
tensile strengths.

It is obvious from the data presented that the experimental results do not correlate with a two-parameter Weibull statistical model. This lack of correlation may be a
result of test methodology or may simply be an indication
that the Weibull distribution is an inadequate model for
describing failure in the fiber direction for a unidirectional
fiber-reinforced composite. It is important from a design
standpoint to determine the precise source of the discrepancy. This can only be accomplished by establishing
failure mechanisms. Until this is done, it appears that any
attempt to predict tensile-coupon data from flexural data
for design purposes is premature.

Acknowledgments
~t~= 2 5 I

175

216

Conclusions

b \

02

AS/3501-5A

Another anomaly associated with the experimental data

is the extremely high tensile-strength values obtained
from 3-point flexural loading of 8-ply T300/5208 unidirectional composites. It is possible that the ratio of
load-nose radius-to-specimen thickness is too large,
producing a distributed load rather than a concentrated
load.
One final observation concerning the data, is the
obvious possibility that the Weibull distribution is not a
valid model for describing failure in the fiber direction for
a unidirectional fiber-reinforced composite.
It should be noted that Figs. 2-9 do not indicate a
minimum strength at the origin, but are simply for convenience plotted with an initial point corresponding to a
very high reliability. For example, the origin of Fig. 2
(x = 0.75) corresponds to a reliability of 0.9950 and not
to a reliability of 1. For any value o f x > 0, R ( x ) < 1.

June 1980

1. Weibull, W., "'A Statistical Theory o f the Strength of Materials,"

lng. Vetenskaps Akad. Hand/ (Roy Swedish Inst. Eng. Research Proc.),
NR 151 (1939).
2. Bullock, R.E., "'Strength Ratios o f Composite Materials in Flexure
and in Tension, "' J. o f Camp. Marls, 8, 200 (1974).
3. Daniel, I.M. and 14"eil, N.A., "The Influence o f Stress Gradient
Upon Fracture o f Brittle Materials," A S M E Paper ~,o. 63-H'A-228,
presented at Winter Annual Meeting, Amer, Sac. o f Mech. Eng., Philadelphia, PA, Nov. 17-22, 1963.
4. Weil, N.A. and Daniel, I.M., "Analysis o f Fracture Probabilities
in Uniformly Stressed Brittle Materials, " J. o f the Amer. Car. Sac., 47,
268 (1964).
5. Knight, M. and Hahn, H.T., "'Strength and Elastic Modulus o f a
Randomly-Distributed Short Fiber Composites," J. o f Camp. Marls, 9,
77 (1975),
6. A S T M Standard D-3"039, "Standard Test Method for Tensile
Properties o f Oriented Fiber Composites, "" Book o f A S T M Standards,
Part 36, 721 (1978).
7, A S T M Standard D-790, "'Standard Test Method for Flexural
Properties o f Plastics and Electrical Insulating Materials, " Book o f A S T M
Standards, Part 35, 321 (1979).
8. Mann, N.R., Schafer, R.E. and Singpurwalla, N.D., Method for
Statistical Analysis o f Reliability and Life Data, John H'i/ey and Sons,
New York (1974).
9. Thoman, Darrel R. and Bain, Lee J., "Two Sample Tests in the
Weibull Distribution," Technometrics, I 1, 805 (1969).
10. Wolf, R.V. and Lemon, G.H., Reliability Prediction for Composite Joints--Bonded and Bolted Air Force Technical Report AFML-TR74-197 (March 1976).
11. Thoman, Darrel R., Bain, Lee J. and Antle, Charles E., "'Inferences on the Parameters o f the 14eibull Distribution, " Technonletrics,
11 (3), 445-460 (Aug. 1969),