You are on page 1of 10

Composites Science and Technology 62 (2002) 16731682

www.elsevier.com/locate/compscitech

The prediction of failure envelopes and stress/strain behavior of


composite laminates: comparison with experimental results
C.T. Suna,*, J. Taoa, A.S. Kaddourb
a
School of Aeronautics and Astronautics, Purdue University, West Lafayette, IN 47907-1282, USA
b
DERA Farnborough, Farnborough, Hampshire, GU14 0LX, UK

Accepted 1 November 2001

Abstract
Two theoretical models were used by Sun and Tao [Comp. Sci. Technol. 58 (1998) 1125] in Part A of the exercise to predict the
strength and stress/strain curves for a number of test problems. The strength model was based on linear elasticity in conjunction
with a ply-discount method using parallel spring stiness reduction, and the stress/strain curves were predicted by the linear model
and a non-linear model incorporating simplied shear-lag analysis and an ABAQUS nite-element package. This paper provides a
description of the correlation between experimental results provided by the organizers of the failure exercise and the theoretical
predictions, published in Part A for (a) biaxial failure envelopes of [0 ] unidirectional and [0 /  45 /90 ]s, [ 30 /90 ]s and [  55 ]s
multi-layered composite laminates, and (b) stress/strain curves for [0 / 45 /90 ]s, [ 55 ]s, [0 /90 ]s and [ 45 ]s under uniaxial and
biaxial loadings. # 2002 Published by Elsevier Science Ltd.
Keywords: C. Failure criterion; C. Laminates; B. Strength; B. Stress/strain curves; Combined loading

1. Introduction matrix cracking and the corresponding stiness reduc-


tion were predicted by use of a simplied shear-lag
In Part A of the failure exercise, Sun and Tao [1] pre- analysis and nite-element analysis.
sented theoretical predictions of biaxial failure envelopes The theoretical results based on the models used by
of unidirectional and multi-directional composite lami- Sun and Tao [1] are compared in Ref. [3] with those of
nates and also for seven stress/strain curves for a number other models used by ten dierent groups who partici-
of laminates under uniaxial and biaxial stresses. Details pated in the failure analysis exercise. All these groups
of laminates analysed, the unidirectional ply material used the same input data to analyze 14 test problems.
properties and the types of loading applied are given in Soden et al. [4] provided details of test results for all
Ref. [2]. The strength was predicted by using the linear the cases analyzed in Ref. [1]. Almost all the tests were
laminate theory in conjunction with ply-by-ply discount carried out on tubular specimens. In some of the test
method using a parallel spring stiness reduction model. results, initial failure in the form of matrix cracks or
The stress/strain curves were predicted by using the lin- leakage was described and, in other test cases, only nal
ear model and a non-linear model based on simplied failure stresses were provided.
shear-lag analysis and nite-element analysis with The theoretical and test results were superimposed and
ABAQUS. The linear analysis was used to generate the the gures were supplied to the authors. These gures are
biaxial failure envelopes and also the stress/strain curves. described in the following sections.
The non-linear analysis, which was used only to predict the
stress/strain curves, took into account both material non-
linearity and progressive matrix cracking. Material non- 2. Comparison between theory and experiments
linearity was considered by using a one-parameter plasti-
city model for ber-reinforced composites. Progressive 2.1. Biaxial failure envelopes

* Corresponding author. Tel.: +1-765-494-5130; fax: +1-765-494-


Figs. 13 show the failure envelopes of unidirectional
0307. laminae under various types of loading. Fig. 1 presents
E-mail address: sun@ecn.purdue.edu (C.T. Sun). a comparison between test results and theoretical
0266-3538/02/$ - see front matter # 2002 Published by Elsevier Science Ltd.
PII: S0266-3538(01)00211-1
1674 C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682

Fig. 1. Biaxial failure stresses for 0 lamina made of GRP material E-glass/LY556/HT907/DY063.

Fig. 2. Biaxial failure stresses for 0 lamina made of CFRP material T300/914C.

predictions for an E-glass/LY556 epoxy lamina under shear strength is enhanced by the presence of a com-
combined transverse and shear loading ( y  xy). The pressive transverse stress.
correlation is good for tensile  y stress. However, for Fig. 2 presents the theoretical and experimental
compressive  y stress, the correlation is rather poor as the results for T300/914C carbon/epoxy lamina under com-
theory predicts a quadratic envelope with no enhance- bined longitudinal and shear loading ( x  xy). A wide
ment in the shear stress due to application of compressive scatter is observed in the test results, especially in the
stress while the test results showed a degree of enhance- values of uniaxial shear strength. In the test results, the
ment, especially for moderate values of  y stress. It should data suggest (a) an enhancement in the shear stress
be noted that the experimental transverse compressive above the uniaxial shear strength when the longitudinal
strength [4] is greater than that provided for use in the tensile stress is around 9001000 MPa and (b) a reduc-
analysis of Ref. [1]. If a greater transverse compressive tion in the shear strength when a compressive stress is
strength were used, the predicted strength in the second applied. The interaction between the shear and tensile
quadrant would have agreed better with the experimental longitudinal stresses as displayed in the experimental
data. Nevertheless, the experimental data indicates that data cannot be easily explained.
C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682 1675

Fig. 3. Biaxial failure envelope of 0 GRP lamina under combined  x and  y stresses. Material: E-glass/MY750 epoxy.

Fig. 3 shows a comparison between test results and


theoretical predictions for E-glass/MY750 epoxy lamina
under combined longitudinal and transverse loading
( x  y). The theoretical curve showed no interaction
between the two stresses in the four quadrants while the
test results in the tension-compression quadrant suggest
a degree of interaction between the stresses. Therefore,
as far as this quadrant is concerned, the theory over-
predicted the test results and thus the correlation is
rather poor. It is, however, dicult to make a coherent
comparison between the theory and experiment as no
sucient test results are available in the other quadrants.

2.2. Biaxial failure envelopes for multidirectional


composites

Figs. 4 and 5 show the experimental and theoretical


failure envelopes of the [90/  30 ]s laminate made of E-
glass/LY556/HT907/DY063 epoxy subjected to com-
bined direct stress ( x  y) and combined axial stress Fig. 4. Biaxial failure envelope for (90 / 30 ) laminate under com-
and shear loading ( x  xy), respectively. The test results bined  x and  y stresses. Material: E-glass/LY556 epoxy.
were obtained from tests on tubes and the test data
available described only the nal failure stresses, i.e. no biaxial failure stresses in the tension-tension and com-
initial failure stresses were supplied for the laminates. pression-compression quadrants. In the latter case,
Hence, it is not possible to compare the theoretical pre- buckling of the tubes may be considered as one possible
diction of initial failure with test data. cause for such dierences.
The theoretical failure stresses under  x  y loading When the laminate is subjected to combined axial
(Fig. 4), show a trend similar to the test results. There stress and shear loading ( x  xy) (Fig. 5) the theory
are, however, dierences between the theoretical and gave a good prediction of the uniaxial shear strength
experimental data. The theory slightly overpredicted the but it slightly underestimated the axial tensile strength.
1676 C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682

Fig. 5. Biaxial failure stresses for (90 / 30 ) laminate under  xy and  x stresses.

The theory also underpredicted the laminate strength 1. This may be partly attributed to the high bre
when compressive  x stress was combined with  xy shear volume fraction of the test specimens used in this
stress. quadrant [4]. The bre volume fraction was around
Fig. 6 shows the experimental and theoretical results 67% while the theoretical prediction was for a bre
of the biaxial failure envelopes of [  55]s laminates volume fraction of 60%. It is noted that in the tension-
made of E-glass/MY750 epoxy under combined  x  y tension quadrant, the theoretical predictions were lower
biaxial loading. Good correlation between theoretical than the experimental results. For instance, at SR=
nal failure prediction and test results can be observed 10:1, the hoop strength  y was almost three times higher
in a number of areas. However, the theory under- than that predicted. On the other hand, the theoretical
predicted the experimental strength in the compression- nal strength overpredicted the experimental strength at
compression quadrant for stress ratio SR=  y:  x= 2: stress ratios between SR=0:1 and 2:1. The correlation

Fig. 6. Biaxial failure stresses for ( 55 ) E-glass/MY750 laminates.


C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682 1677

between the initial failure prediction and the strength of Fig. 8 shows the theoretical and experimental stress
unlined tubes was also poor. The dierence between strain curves for [0/90]s laminate under uniaxial loading
them was as high as 5 fold in some areas. ( y/ x =0 / 1). The test results were obtained from
Biaxial failure envelopes for [0/  45/90]s laminate testing coupons under uniaxial tension [4]. The thick-
under  x  y biaxial loading are plotted in Fig. 7. The ness of the coupons was almost twice that of the lami-
correlation between the theory and test results is good nate analyzed, and the ber volume fraction, 62%, was
as both indicated (a) an increase in the stress  y due to slightly higher than that used in the analysis, 60%.
biaxial loads in the tensiontension quadrant and (b) a In the linear analysis, the theoretical curves are linear
decrease in stress  y in the tensioncompression quad- up to around 80 MPa stress where matrix cracking took
rant. An apparent poor correlation between test results place in the layer whose ber direction was perpendicular
and theoretical prediction can be seen in the compres- to the loading direction. A reduction in the modulus took
sioncompression and, to a lesser extent, in the tension place at this point. Thereafter, the slope changed and the
compression quadrants. This could be due to problems modulus decreased slightly but remained almost parallel
(mainly buckling) associated with the experiments. to that observed in the test.
In the nonlinear analysis, the initial modulus, up to 80
2.3. Stress/strain curves MPa stress, was identical to that obtained from the lin-
ear analysis. When compared with the test results, the
The stress/strain curves were predicted using (a) linear theoretical modulus, after initial failure, was slightly
analysis and (b) nonlinear analysis. In the linear analy- lower than that observed in the test.
sis, a ply-by-ply discount method was used [1]. As a In the linear analysis, the correlation between the
result, a kink was seen when matrix failure in a limina predicted and measured Poissons strains was good up
occured. In the latter case, material non-linearity, aris- to around 268 MPa stress at which matrix cracking
ing from matrix nonlinear behaviour, and eects of occurred in the 0 lamina. Thereafter, the predicted
matrix cracking were taken into account by using a strain was lower than the measured one.
progressive matrix cracking model [1]. Since the eect of The Poissons strain in the nonlinear analysis was
matrix cracking was modeled in a progressive manner, larger than that observed in the tests. This is in contrast
no jumps were seen in the resulting stress/strain curves. to results of the linear analysis where the predicted
It should be noted that, in the present exercise, the strain was smaller than that measured in the tests. It
nonlinear analysis was used only to produce stress/ must be mentioned here that the lamina modulus, E2, in
strain curves but not for strength predictions. To use the the nonlinear analysis was assumed to decrease expo-
result of the nonlinear analysis for strength predictions, nentially with increasing the crack density whereas in
it is necessary that new appropriate failure criteria be the linear analysis the same modulus was dropped to
developed. Thus, the ending point of each stress/strain zero suddenly after the occurrence of transverse cracks.
curve shown in Figs. 814 does not necessarily corre- Again, it is noted that the ending point in the stress
spond to the predicted ultimate strength of the laminate. strain curve obtained according to the nonlinear analysis
does not indicate failure of the laminate.
Fig. 9 shows the theoretical and experimental results
for the [ 45]s laminate under equal-biaxial tensile
loading ( y/ x=1/1). The experimental results were
obtained from tubes tested with and without the use of
plastic liner. The main source of nonlinearity in the pre-
dicted curves is matrix cracking rather than material non-
linearity. The predicted initial failure stress was 68 MPa
which is similar to that observed in the test, 70 MPa. The
post initial failure behaviour in the linear analysis was
similar to that of the nonlinear analysis with the secant
modulus in the nonlinear analysis slightly greater than
that of the linear analysis, as was the case of the [0/90]s
laminate.
Both models used in the theory predicted the strains
"x and "y to be identical. However, the test results
showed the hoop strain "y to be larger than the axial
strain "x. Furthermore, at any given stress the measured
strains were larger than those predicted. This could be
due to the short length of the test specimens, which
Fig. 7. Biaxial failure stresses for (0 / 45 /90 )AS4/3501-6 laminates. could cause signicant deviation from the boundary
1678 C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682

Fig. 8. Stress/strain curves for 0 /90 E-glass/MY750 laminate under uniaxial tension ( y=0).

Fig. 9. Stress/strain curves for 45 E-glass/MY750 laminate under  y/ x= 1/1.
C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682 1679

Fig. 10. Stress/strain curves for 45 E-glass/MY750 laminate under  y/ x= 1/ 1.

Fig. 11. Stress/strain curves for 55 GRP laminate under uniaxial tension ( y/ x= 1/0).

condition assumed in the theoretical predictions. Nei- In contrast, the nonlinear analysis gave nonlinear
ther the linear analysis nor the nonlinear analysis was curves. But the experimental nal failure strains of
able to predict leakage in this laminate. about 10% obtained from testing tubes under combined
Fig. 10 shows the predicted and measured stress/ internal pressure and axial compression were not pre-
strain curves for [  45]s laminate subjected to equal dicted by the linear analysis.
biaxial tension-compression loading with  y/ x=1/1. Fig. 11 shows the predicted and measured stress/
This is the only case among all the problems solved in strain curves for [ 55]s laminate subjected to uniaxial
this exercise, in which the nonlinearity is predicted to tensile loading with  y/ x=1/0. The measured curves
come primarily from material nonlinearity as the beha- were obtained from internal pressure tests on thin tubes
viour is dominated almost fully by the unidirectional tested with and without the use of plastic liner. The
shear stress. The linear analysis produced straight lines correlation between the measured and predicted failure
indicating that the rst failure was the ultimate failure. stresses (and thus the failure strains) with the linear
1680 C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682

measured curves were obtained from internal pressure


tests on thin tubes tested with and without the use of
plastic liner. It is evident that the theoretical (linear and
nonlinear) results agree with the experimental results in
the early stages of deformation, i.e. up to around 120
MPa. In both the linear analysis and the measured
results, the "x strain curve crossed the "y curve as the
stress increased, see also comments on corrected results
in Section 2.4. In contrast, in the nonlinear analysis, the
"x strain (always smaller than the measured curve) did
not cross the measured "y curve. Rather, it exhibited a
stiening behaviour and reached an almost constant
value beyond 500 MPa stress. Since the nonlinear ana-
lysis was not used for strength predictions, the strain at
the end point of the curve should not be taken as the
failure strain of the laminate.
Fig. 12. Stress/strain curves for a 55 laminate made of E-glass/
Figs. 13 and 14 show the theoretical and experimental
MY750 epoxy under  y/ x= 2/1.
stress/strain curves for [0/  45/90]s AS4/3501-6 laminate
under uniaxial ( y/ x=1/0) and biaxial stresses ( y/
 x=2/1). It was suggested in the Part A paper by Sun
analysis is poor. However, the stress/strain curves and Tao [1], that material nonlinearity and matrix
according to the nonlinear analysis seem to agree with cracking contribute nearly the same amount to the
the experimental curves quite well. nonlinearity of the stress-strain curve under  y/ x=1/0).
It is known that angle ply composite laminates such Also, when the laminate is subjected to  y/ x=2/1 biax-
as the ones considered in Figs. 10 and 11 tend to yield ial loading, matrix cracking has a greater eect on the
large ow strains when subjected to uniaxial loads. laminate stress/strain nonlinearity. That is because the
Large strains can signicantly alter the ber orientation contribution to failure by the shear stress component
with the consequence of load-shifting to bers and, gets smaller as the loading moves from uniaxial tension
thus, yielding higher laminate strengths. The experi- to equal biaxial tension.
mental results seem to conrm this conjecture. The correlations between experimental and the theo-
The behaviour of [  55]s laminate subjected to biaxial retical results are in general good. The worst case was
tensile loading with  y/ x=2/1 is depicted in Fig. 12. The that of the predicted "x strain under  y/ x=2/1 loading

Fig. 13. Stress/strain curves for (0 / 45 /90 ) laminate under uniaxial tension ( y/ x= 1/0).
C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682 1681

and [0 /  45 /90 ]s carbon/epoxy laminates under uni-


axial and biaxial tensile stresses.
(d) Very poor agreement with the experimental failure
strains which are dominated by transverse and shear
matrix cracking. These include both branches of the
failure strains of [ 45 ]s under SR=1/-1, both branches
of [ 55 ]s glass/epoxy laminate under SR=1/0, where
the predicted curves are terminated prematurely, the
Poissons strain in [0 /90 ]s glass/epoxy laminate under
uniaxial tension and the Poissons strain of [ 55 ]s
glass/epoxy laminate under SR=2/1.
(e) The assumption that the in situ lamina tensile
transverse and shear strengths are one and half times
those measured from individual laminae seems to work
reasonably well in the prediction of laminate strengths.
The laminate stress/strain curves predicted by both
the linear analysis, which accounted for eects of matrix
cracking by a ply discount method, and by the non-
Fig. 14. Stress/strain curves for (0 / 45 /90 ) laminate under biaxial
stress  y/ x= 2/1. linear analysis, which took into account nonlinearities
due to material and progressive matrix cracking, gen-
erally agree well with the measured curves for ber-domi-
where the linear analysis resulted in a strain of 0.51%, nated laminates. For matrix-dominated laminates, except
almost 50% higher than that measured in the tests. for the [ 55 ] glass/epoxy laminate under SR=2/1, the
nonlinear analysis gave reasonably accurate stress/strain
2.4. Modied results curves up to moderate strain ranges.
One should be cautious when drawing conclusions
Revised stress/strain curves were made after the pub- from the comparison between theoretical and experi-
lication of Part A paper [1]. The revised gures, shown mental results for laminate strength. Discrepancies
in Figs. 8, 9, 12, 13 and 14, were corrected for minor should not be attributed automatically to deciencies of
numerical and plotting errors discovered after the pub- the theory. There are a number of factors that may
lication of Part A. The changes are apparent as steps in the result in such discrepancies.
stress/strain curves predicted by the linear analysis. The (a) The lamina strength properties in the laminate
most noticeable changes are in Figs. 8 and 12. These analysis were obtained from unidirectional composite
changes did not aect the results of failure predictions. specimens. These strength properties could be quite dif-
ferent from the in situ lamina properties due to the
absence of the constraining eects from adjacent laminae
3. Conclusions in the laminate.
(b) Most tests for biaxial loading were performed using
Employing the linear elastic laminate theory in con- short, thin-walled tubular specimens. The gripping condi-
junction with a ply discount method based on parallel tion could induce undesirable boundary eects, especially
spring stiness reduction, together with HashinRotem in tests that involve torsion. Additional failure mechan-
failure criterion gave the following features in failure isms that were not accounted for by the theory could be
prediction of composite laminates. responsible for failure in the test. The most notable
(a) Good correlation with test results for nal failure example is buckling of thin-walled tubes under com-
strengths of [0 / 45 /90 ]s AS4/3501-6 carbon/epoxy and pression or torsion.
[ 30 /90 ]s glass/epoxy laminates under biaxial stresses. (c) The eect of curvature of the tubular specimen on
(b) An acceptable correlation with test results in a the composite strength is unknown. Moreover, since tub-
number of segments of the biaxial failure envelope for ular specimens are manufactured by dierent methods,
[ 55 ]s glass/epoxy laminate. However, the model they may not be of the same quality as at laminates.
underpredicted the nal strength in some other portion
of the envelope by a factor up to 3.
(c) Good agreement with the nal strength of lami- References
nates whose failure is largely dominated by the ber
tensile strength, such as and [  45 ]s under equal biaxial [1] Sun CT, Tao J. Prediction of failure envelopes and stress-strain
tension, [0 /90 ]s glass/epoxy laminate under uniaxial behaviors of composite laminates: comparison with experimental
tension, [ 55 ]s glass/epoxy laminate under SR=2/1 results. Composites Science and Technology 1998;58:112536.
1682 C.T. Sun et al. / Composites Science and Technology 62 (2002) 16731682

[2] Soden PD, Hinton MJ, Kaddour AS. Lamina properties and lay- nates, published in. Composites Science and Technology 1998;
up congurations and loading conditions of a range bre rein- 58(7):122554.
forced composite laminates. Composites Science and Technology [4] Soden PD, Hinton MJ, Kaddour AS. Experimental failure stres-
1998;58:7 1011-22. ses and deformations for a range of composite laminates sub-
[3] Soden PD, Hinton MJ, Kaddour AS. Comparison of the pre- jected to uniaxial and biaxial loads. Composites Science and
dictive capabilities of current failure theories for composite lami- Technology [in press].

You might also like