Académique Documents
Professionnel Documents
Culture Documents
P. D. Soden et al.
Warton, UK
McCartney, L. N. National Physical Laboratory, London, UK Physically based `Damage Mechanics' McCartney
Puck, A. and Schurmann, H. Technische Hochchule, Darmstadt, Germany Physically based 3-D phenomenological models Puck
Wolfe, W. E. and Butalia, T. S. Department of Civil Engineering, Ohio State University, Maximum strain energy method, due to Sandhu Wolfe
Ohio, USA
Sun, C. T. and Tao, J. X. Purdue University School of Aeronautics & Astronautics, Linear and non-linear analysis (non-linear is Sun
West Lafayette, Indiana, USA. FE based)
Zinoviev, P., Grigoriev, S. V., Institute of Composite Technologies, Orevo, Development of Maximum stress theory Zinoviev
Labedeva, O. V. and Tairova, L. R. Moskovkaya, Russia.
Tsai, S. W. and Liu, K.-S. Aeronautics and Astronautics Department, Interactive progressive quadratic failure criterion Tsai
Stanford University, California, USA.
Rotem, A. Faculty of Mechanical Engineering, Technion-Israel Interactive matrix and bre failure theory Rotem
Institute of Technology, Haifa, Israel.
Current failure theories for composite laminates 1227
originated from the work of Sandhu and colleagues at were postulated in the theories, see Table 3. The
Wright Patterson Air Force Base. modes of failure basically range from bre failure
Some of the participants (Hart-Smith, Sun and Chamis) (tension, compression or shear) to matrix failure
presented more than one method of solving the test prob- (transverse tension, transverse compression, shear
lems. Hart-Smith provided two contributions to the cur- or a combination of any of these three modes). In
rent exercise. One is a generalized Tresca model,18 which most cases the properties used to predict the
considers shearing of the bres as one of the dominant modes of failure are lamina properties rather than
failure modes and is referred to here as the Hart-Smith(1) constituent (bre or matrix) properties.
theory. The second paper,17 contains an interpretation of 6. Micromechanics: A few theories relied on micro-
the widely used maximum-strain and truncated-max- mechanics in their formulation. Although the prop-
imum-strain failure theories. The maximum-strain theory erties of all the laminae provided in the exercise are
is referred to here as the Hart-Smith(2) theory. The trun- inuenced by the properties of the constituents, i.e.
cated-maximum-strain theory is only slightly dierent the bre and the matrix, certain theories (Chamis,
from the original maximum-strain theory. Hart-Smith(1), Puck, Rotem and Tsai) explicitly
Sun and Tao21 presented results for linear and non- required certain properties of the constituents.
linear analyses. Their detailed non-linear analysis, which
is based on nite-element analysis and allows for elastic/ . Unlike all the other contributors, both of Cha-
plastic material properties and progressively increasing mis's programs are based on micromechanics
matrix crack density, was used only to predict stress/ relationships and predict the lamina properties
strain curves. Their linear analysis was applied to gen- from the properties of the constituents. The rele-
erate the failure envelopes and stress/strain curves. vant equations for calculating ply properties and
Chamis13 and his group at NASA employed two stress limits can be found in the ICAN code.13
computer programs (a) `Integrated Composite Analy- Using ICAN, Chamis adjusted the constituent
ser' (ICAN) and (b) `Composite Durability Structural properties to give lamina properties close to those
Analyser' (CODSTRAN), which is integrated with a provided in the exercise and then used the derived
nite-element package that allows analysis of complex lamina properties to predict laminate behaviour.
structures. Results are presented for each program. . Hart-Smith(1) theory uses micromechanics to
ICAN results are referred to as Chamis(1) and COD- obtain the failure-strain limits needed for
STRAN results as Chamis(2). establishing the failure envelopes in the strain
plane, which can then be converted into an
2.2 Characteristics of the theories appropriate envelope in the stress plane.
Each of the theories can be characterised by a number Among the properties needed are bre volume
of key features. Table 2 lists some of the features, fraction, bre moduli, bre Poisson ratios and
including the following: the modulus and Poisson ratio of the matrix.
. Tsai's theory uses micromechanics to carry out
1. Identication of theory: This is shown in column 1. progressive post-failure analysis.
2. Method of analysis: Whether classical thin-lami- . Rotem's theory requires the knowledge of the
nate theory or a nite-element (FE) program was matrix modulus and the tensile and compressive
used. All the contributors relied on classical thin- strengths for use in his matrix failure criterion.
laminate theory. Sun and Chamis have used nite- . Puck's theory requires the properties of the
element codes besides. bres (strengths, moduli and Poisson ratios)
3. Type of analysis: Whether linear or non-linear and the modulus of the matrix. They are used
analysis was performed. Chamis, Eckold, Hart- to compute, for instance, a stress magnication
Smith, McCartney, Tsai, Sun (linear) and Zinoviev factor that is needed for establishing biaxial
used linear material properties whereas Edge, failure envelopes.
Rotem, Puck, Wolfe and Sun (non-linear) used
non-linear analysis. 7. Post-initial failure degradation models: The entry in
4. Thermal stresses: Whether or not thermal residual the seventh column of Table 2 species whether or
stresses were included. Half of the participants not the theory used any degradation model to
(Eckold, Hart-Smith, Rotem, Wolfe and Zinoviev) account for post initial failure. Eckold and Hart-
did not include the thermal stresses while the oth- Smith did not employ any post-failure model
ers attempted to do so but with some other con- whereas the rest did.
siderations, see Table 2. Multi-directional laminates subjected to uni-
5. Modes of failure: Whether the theory is able to axial or biaxial stresses may still be capable of
identify the modes of failure encountered during carrying load beyond rst-ply failure or initial
loading. Almost all of the theories were able to failure occurrence. The modelling of post-failure
discriminate between two or more modes of fail- behaviour of a laminate requires that assump-
ure. Various modes of failure and failure criteria tions be made regarding the properties of the
1228
Table 2. A summary of key features of the theories used by various contributors
Contributor Method Type of Thermal Failure Micromechanics Degradation Failure criterion Computer program used
analysis stresses modes model
a
Chamis CLT+FE Linear Yesb Yes Yes Micromechanics based ICAN and CODSTRAN
Eckold CLT Linear No No No No BS4994 None
Edge CLT Nonlinear Yes Yes No Yes GrantSanders Modied ESDU package
Hart-Smith CLT Linearc No Yes Yes No Maximum strain theories and None
generalised Tresca criteria
McCartney CLT Linear Yes Yes No Yes Fracture mechanics Program developed at NPL, UK
P. D. Soden et al.
Puck CLT Non linear Yesd Yes Yes Yes Puck's theory FRACUAN developed in Kessel,
Germany
Rotem CLT Nonlinear Yes Yes Yes Yes Rotem theory In-house program
Sun CLTe Linear Yes Yes No Yes RotemHashin theory In-house program
f
Sun CLT+FE Non-linear Yes No No Yes Plasticity model based on Hill's yield ABAQUS program
Tsai CLT Linear Yes g Yes h Yes Yes Tsai-Wu quadratic theory Mic-Mac
Wolfe CLT Nonlinear No Yes No Yes Sandhu's strain energy model In-house program
Zinoviev CLT Linear No Yes No Yes Maximum stress theory STRAN software
a
Not in all cases.
b
The theory identies failure modes but Chamis chose not to present them for some of the cases he analysed.
c
Secant properties rather than initial properties are occasionally used in the analysis.
d
Only 50% of the thermal residual stresses are considered.
e
Used to generate the failure envelopes and stress/strain curves.
f
The nite element (FE) analysis was used only to generate the stress/strain curves.
g
Tsai introduced a certain amount of moisture to compensate for the thermal stresses.
h
Additional criteria identify modes of failure.
Current failure theories for composite laminates 1229
Table 3. Summary of the post initial failure degradation models used in the theories
Name Failure mode Properties degraded
Eckold No post failure
Rotem After nal matrix failure E2 00, G12 00, E1 E01 expk"1 , k is a large constant
Puck Cracking under tension Mode (A) E2 E02 , G12 G012 , 12 012 , where is a parameter which varies
with stress
Cracking under compression Modes G12 1 G012 , 12 012 ; where 1 is smaller than
(B) and (C)
Chamis Matrix failure Em is replaced by a negligible value and E2 , G12 , 12 and E1 are
computed from micromechanics
Edge Matrix failure E2 1 E02 , G12 2 G012 , 12 3 012 , where 1 , 2 and 3 are
empirical parameters that decrease with increasing strain
Sun (Nonlinear) Matrix shear failure E2 E02 expE l) and G12 G012 expG l), where E and G are
constants, l is normalised crack density
Transverse matrix failure E2 E02 expE l
Tsai Matrix failure "2 > 0 Em 015E0m , 12 015012 ; where E2 and G12 are computed from
micromechanics
degraded lamina. Table 3 summarises the dier- represented by Puck, Edge, Rotem, Zinoviev,
ent post-initial-failure models adopted. All these Rotem, Chamis and Sun (non-linear).
models shared the following common features.
Full details of all the models are given in the con-
. All rely on ply-by-ply analysis. tributors papers.1324
. All assume that bre failure, be it in tension or
in compression, constitutes nal failure. 8. Failure criterion: This is to identify the origin, the
. Almost all models distinguish between failures nature or the name of the failure theory used. The
under transverse tension and that under trans- theory used by Sun in his linear analysis is
verse compression. Here the term `transverse' is attributed to Hashin and Rotem but was slightly
used to refer to the direction perpendicular to dierent from the theory Rotem applies in his
the bres in a lamina. contribution to the exercise. Zinoviev's theory uses
a maximum-stress criterion with further develop-
However, the post-failure methods employed do dif- ment to cater for various post-failure scenarios.
fer and, for the sake of simplicity, they can be classied Equations governing the failure criteria used are
into two main groups: described by the participants in their own papers,
see also Table 4.
(a) Models employing sudden reduction in the 9. Title of computer program used: The entry in the
properties of the failed lamina. These are last column of Table 2 lists the name, if any, of the
presented by Tsai, Wolfe and Sun (linear). computer program used by the participants. The
(b) Models employing a gradual drop in the participants have either used commercial codes
properties of the failed lamina. These are (Chamis, Edge, Sun (non-linear), Tsai and Zinoviev),
Table 4. Modes of failure in the theories
1230
Mode of failure Failure criterion Theory
1 f12
Fibre failure in tension "1T "1 Ef1 mf 2 1 Puck
Longitudinal tensile failure "1 "1T (and Eckold's "1 0004) Hart-Smith(2) and Eckold
2 2 2
1
Fibre tension/ compression and matrix tension XT XC YTY2 C X1T X1C 1 Y1T Y1C 2 p
2F12 1 2
X X Y Y
12
S12 1 Tsai
T C T C
and compression and shear
2 3m1 2 3mi
1 d"1 1 d"i
6"^1 7 P 6"^i 7
Fibre failure (in tension and compression) 6 7 = 6 7 01 Wolfe
4 1 d"1
5 4 i d"i
5
i1;2;6
"u "u
1 i
Fibre failure in compressive 1
"1C j "1 Ef12
f1
m f 2 j 1 10
21 2 Puck
Longitudinal compressive failure 1 Xc (and Eckold 1 XT ) Zinoviev, Edge, Rotem, Sun, Hart-Smith(1)
and Eckold
P. D. Soden et al.
Longitudinal compressive failure "1 "1C (and Eckold "1 0004) Hart-Smith(2) and Eckold
Shear of bres Tresca type criterion (see eqn (13) in Ref. 18) Hart-Smith(1)
Transverse tensile failure "2 "2T (and Eckold "2 0001) Hart-Smith(2) and Eckold
Transverse compressive failure 2 YC (and Eckold 2 YT ) Zinoviev, Edge, Eckold and Sun
Transverse compressive failure "2 "2C (and Eckold "2 0001) Hart-Smith(2) and Eckold
r 2
21 YT 2 2 2
Inter-bre failure Mode A (for transverse tension) S21 1 p ?k S21 YT p?k S212 1 1D1 Puck
r !
2
Inter-bre failure Mode B (for moderate 1
S21
2
21 p?k 2 p?k 2 1 1D1 Puck
transverse compression)
" 2 #
2
21 YC
Inter-bre failure Mode C (for large transverse Y2C 1
2 1 1D Puck
21p?? S21
compression)
(continued)
Table 4contd
Mode of failure Failure criterion Theory
1231
1232 P. D. Soden et al.
or in-house programs (McCartney, Rotem, Sun . Chamis did not present a nal failure envelope for
(linear) and Wolfe) or no code (Eckold, Hart-Smith). the quasi-isotropic laminate (Case 4).
. Eckold did not analyse the carbon-bre composites
because his simple design method was developed
2.3 Breadth of cases analysed by each participant specically for glass-bre-reinforced pressure ves-
The participants were set 14 test Cases for analysis and sels. He also chose not to analyse shear loading of
these are summarised in Table 5. The composite sys- laminates (Case 5) because that type of loading is
tems, laminate congurations and load combinations not covered by the pressure-vessel design code.
have been described in detail.12 . Hart-Smith did not provide solutions for the seven
Table 6 shows the test cases attempted by each con- test cases involving stress/strain curves. Neither
tributor. The majority of the participants were able to did he predict initial failures in either of the two
analyse all the cases. However, there were some notable theories oered (Hart-Smith(1) and Hart-Smith
shortfalls: (2)), although that would be possible for some
Current failure theories for composite laminates 1233
laminates by using other interpretations of the dierent theories for (i) E-glass/LY556 lamina under
maximum-strain theory.26 combined shear and direct loading perpendicular to the
. The analytical method used by McCartney is in its bres (y and xy ), (ii) T300/BSL914C lamina under
infancy and at the time of writing he was able to combined shear and direct stresses parallel to the bres
apply it to only two cases, (Case 12) a 0 /90 (x and xy ) and (iii) E-glass/MY750 lamina subjected to
laminate under uniaxial tension (SR y : biaxial stresses applied in directions parallel and per-
x 0:1) and (Case 13) a 45 laminate under pendicular to the bres (x and y ). The bar charts in
SR 1:1 which is equivalent to a 0 /90 laminate Figs 1b, 2b and 3c show the failure strengths for each
under biaxial tension. He did not attempt to pre- theory at the selected stress ratios indicated in Figs 1a,
dict nal failure loads for those cases. 2a, 3a and b.
Table 7. Ratios of the highest:lowest predicted unidirectional lamina strengths for selected stress ratios
No. Lamina studied Stress ratio Final failure prediction Highest:lowest
ratio of predictions
Highest Lowest
1 E-glass/MY750 x : y 883 : 1 Edge, Zinoviev and Sun Eckold 3.6
(Loading Case 3)
2 x : y 32 : 1 Eckold Wolfe 4.77
3 x : y 78 : 1 Hart-Smith(2) Rotem 4.4
4 x : y 375 : 1 Hart-Smith(2) Eckold 5.7
5 x : y 14 : 1 Tsai Eckold 2.72
6 E-glass/LY556 y : xy 158 : 1 Edge, Hart-Smith(2), Zinoviev Eckold 3.22
(Loading Case 1)
7 y : xy 1 : 206 Hart-Smith(2), Zinoviev, Eckold Tsai 1.54
8 T300/914C x : xy 125 : 1 Zinoviev, Sun, Hart-Smith(1), Edge 1.64
(Loading Case 2) Hart-Smith(2)
9 x : xy 1875 : 1 Edge, Zinoviev, Hart-Smith(1), Chamis 1.414
Hart-Smith(2), Sun
1234
Table 8. Summary of theoretical results showing the range of initial and nal failure predictions for multidirectional laminates
No. Laminate studied Stress ratio Final failure prediction Initial failure prediction Largest final/initial prediction
Highest Lowest Ratio Highest Lowest Ratio Name Ratio
1 (30 /90 ) GRP (loading Case 5) x : xy 2:35 : 1 Puck Chamis(2) 3.8 Sun Chamis(2) 4.16 Puck 1.99
2 x : xy 1:1 Hart-Smith(1) Wolfe 5.21 Tsai Chamis(2) 5.74 Edge 12.8
3 x : xy 0:1 Puck Chamis(2) 4.71 Zinoviev Chamis(2) 2.91 Puck 6
4 x : xy 1 : 0 Eckold Tsai 3.58 Sun Eckold 4.61 Eckold 12
5 x : xy 1:0 Eckold Chamis(2) 3.13 Tsai Chamis(2) 6.5 Eckold 24
6 (30 /90 ) GRP (Loading Case 4) y : x 1:3 Hart-Smith(1) Rotem 6.2 Eckold Chamis(2) 12.8 Edge 34
7 y : x 1 : 1 Hart-Smith(1) Chamis(2) 8.42 Zinoviev Chamis(2) 2.83 Eckold 5.7
8 y : x 1 : 3 Eckold Zinoviev 3.1 Sun Eckold 1.47 Eckold 3.74
P. D. Soden et al.
9 y : x 1:1 Hart-Smith(2) Wolfe 7 Tsai Chamis(2) 5.03 Chamis(2) 19
10 y : x 1 : 1 Edge Tsai 1.66 Chamis(1) Eckold 4 Eckold 3.66
11 y : x 1:0 Hart-Smith(1) Wolfe 2.42 Tsai Chamis(2) 2.94 Edge 7.5
12 y : x 4:26 : 1 Zinoviev Chamis(2) 2.11 Zinoviev Chamis(2) 7.46 Eckold 5
13 (0 /45 /90 ) CFRP (Loading Case 6) y : x 0 : 1 Rotem Wolfe 1.33 Sun Chamis(2) 9.23 Zinoviev 1.24
14 y : x 1:5 : 1 Zinoviev Wolfe 2.71 Zinoviev Chamis(2) 19.6 Edge 15.7
15 y : x 1 : 1 Tsai Wolfe 1.72 Tsai Wolfe 1.72 All 1
16 y : x 2:1 Sun Rotem 2.63 Chamis(1) Chamis(2) 18 Chamis(2) 51
17 y : x 1:0 Zinoviev Chamis(2) 2 Zinoviev Chamis(2) 16 Chamis(2) 24
18 y : x 1:1 Zinoviev Rotem 3.28 Chamis(1) Chamis(2) 17 Edge 28
19 (55 ) GRP (Loading Case 9) y : x 0 : 1 Zinoviev Eckold 2.53 Edge Eckold 3.64 Eckold 1.47
20 y : x 1 : 0 Eckold Chamis(2) 3.61 Sun Eckold 4.95 Eckold 9.6
21 y : x 2 : 1 Eckold Chamis(2) 4.13 Wolfe Eckold 2.82 Eckold 5
22 y : x 2:1 Hart-Smith(1) Wolfe 8.7 Eckold Chamis(2) 3.66 Chamis(2) 19
23 y : x 1:0 Eckold Chamis(2) 4.47 Puck Eckold 4.88 Eckold 9.55
24 (45 ) GRP (Loading Cases 13 and 14) y : x 1 : 1 Chamis 2 Wolfe 9.7 Eckold Puck 3.71 Puck 13
25 y : x 1 : 1 Edge Chamis(2) 5.8 Edge Chamis(2) 1.48 Edge 4
26 (0 /90 ) GRP Loading Case 12 y : x 0 : 1 Puck Wolfe 2.28 Eckold Edge 2.83 Puck 12
Current failure theories for composite laminates 1235
Fig. 1. (a) Biaxial failure stress envelope for 0 unidirectional E-glass/LY556 epoxy lamina under transverse and shear loading (y
vs xy ); (b) bar charts showing the biaxial failure stresses for an E-glass/LY556 epoxy lamina under y : xy 158 : 1 and 1:2.08.
Other theories predicted envelopes in which the sile failure strength, y , predicted by Hart-Smith(2) is
strength in one direction is aected by the stress in the more than 4.4 fold higher than that predicted by almost
other direction. Tsai's theory predicted an elliptical all the rest of the theories.
failure envelope with a large enhancement of the long- The most common form of interaction assumed that
itudinal compressive strength in the y x space combined stresses reduced the strength of a lamina to a
(Fig. 3). It is a feature of his equation that the predicted value lower than its strength under uniaxial loading.
biaxial compressive strength increases when the value of Figures 13 show many examples of this such as Rotem,
the uniaxial transverse tensile strength is reduced. Sun and Wolfe. In Fig. 3, Wolfe's theory predicts
Hart-Smith(2) also predicts an enhancement of the strengths under combined transverse tension and long-
strength under biaxial loads in y x space. Indeed, itudinal compression that are about one third of the
the largest dierence between the predicted failure uniaxial strengths for that lamina. Note also that Wolfe
stresses in the biaxial compression quadrant occurs at produces a discontinuous envelope as the stresses
the stress ratio x : y 375 : 1. At that stress ratio, approach uniaxial tension and compression along the
the transverse compressive failure strength, y , predicted bre direction (i.e. at x : y 1 : 0 and 1:0).
by Hart-Smith(2) theory is 5.70 times higher than that It can be seen from Figs 13 that although there were
predicted by Eckold. This is partly due to the low some similarities, no two theories gave identically
transverse compressive strength assumed by Eckold. shaped envelopes for the three tests cases.
Hart-Smith(2) also predicts much higher strengths in
biaxial tension than the other theories. In the tension 3.2 Failure envelopes for laminates (test cases 4, 5, 6
tension quadrant, the largest dierence between the and 9)
strength values predicted occurs at stress ratio of In order to examine the performance of the failure the-
x : y 78 : 1. At that stress ratio, the transverse ten- ories at the laminate level, four test cases were chosen,
1236 P. D. Soden et al.
Fig. 2. (a) Biaxial failure stress envelope for T300/BSL914C lamina under longitudinal and shear loading (x vs xy ); (b) bar charts
showing the biaxial failure stresses of T300/914C lamina under x : xy 125 : 1 and 18.75:1.
covering a wide range of materials, layups and biaxial With the exception of Eckold, Chamis(2) and Wolfe,
loading conditions. the initial failure envelopes shown in Fig. 4a are fairly
regular and approximately diamond-shaped, but the
3.2.1 Failure envelopes for the (90 /30 )s E-glass/ magnitudes of the predicted initial failure stresses vary
LY556 laminate under biaxial loads, (sy vs sx ) (Case 4) considerably from theory to theory. The results were
As described in Soden et al.,12 82.8% of the thickness of inuenced by thermal stresses and the highest ratio of
this (90 /30 )s E-glass/LY556 laminate is made of maximum:minimum predicted initial failure stresses
30 plies and only 17.2% of the thickness made of 90 shown in Table 8 was 12.8:1 (Eckold: Chamis) at a ratio
plies. Thus, the laminate is not quasi-isotropic. Conse- of applied biaxial tension stresses y : x 1 : 3. Further
quently, the failure envelopes are expected to lack sym- details of the eect of thermal stresses are discussed in
metry, in relation to the material principal loading Section 4.3.
directions, and also to contain a number of failure Dierences in predicted modes of failure and their
modes depending upon the loading direction. Initial eects on the nal strength values were apparent at
failure envelopes for all the theories are shown in Fig. 4a SR=1:1 in Fig. 4b. Hart-Smith(2) predicted bre ten-
and the nal failure envelopes are shown in Fig. 4b and c. sion failure in the 90 plies at a load of 471 MPa
Bar charts in Fig. 5 compare the initial and nal whereas Wolfe predicted transverse tension failure in
failure stresses predicted by the theories, at specic 30 plies at a load of 67 MPa.
ratios of applied biaxial loads y : x , and Table 8 The envelopes supplied by Chamis(2) and Eckold
shows the range of predicted values at selected load were rather coarse, data points being provided at a lim-
ratios, including those in the bar charts. ited number of biaxial stress ratios and hence the shapes
Current failure theories for composite laminates 1237
Fig. 3. (a) and (b) Biaxial failure stress envelope for 0 unidirectional lamina made of E-glass/MY750 epoxy under longitudinal and
transverse loading (y vs x ); (c) bar charts showing the biaxial failure stresses of E-glass/MY750 lamina under y : x 78 : 1,
32:1, 14:1, 3.75:1 and 8.83:1.
1238 P. D. Soden et al.
Fig. 4. (a) Initial biaxial failure stress envelope for (90 /30 /90 ) laminate made of E-Glass/LY556 epoxy under combined load-
ing (y vs x ); (b) and (c) nal biaxial failure stress envelope for (90 /30 /90 ) laminate made of E-Glass/LY556 epoxy under
combined loading (y vs x ).
of these envelopes might change if more points were larger longitudinal compressive strength than the other
computed. participants.
Examination of Fig. 4b and c shows that Hart- Table 8 shows that the highest ratio of maximum:
Smith(1) and Hart-Smith(2) theories gave similar nal minimum nal failure stresses was 8.4:1 (Hart-Smith(1):
failure envelopes. Rotem's nal failure prediction is very Chamis(2)) at stress ratio y : x 1 : 1.
conservative in the biaxial tension quadrant. The bar charts in Fig. 5 show that at some stress
Eckold predicts the highest (x ) compressive strengths ratios (e.g. y : x 1 : 1) most of the theories pre-
in the biaxial compression quadrant. This is as a result dicted nal stresses of similar magnitude to the initial
of his starting assumption to use a value for the long- failure stresses, whilst at other stress ratios (e.g.
itudinal lamina compressive strength equal to the long- y : x 1 : 0) the nal stresses are predicted to be much
itudinal lamina tensile strength and hence a much greater than the initial failure stresses. In Table 8 the
Current failure theories for composite laminates 1239
Fig. 6. (a) Initial biaxial failure stress envelope for (90 /30 /90 ) E-glass/LY556 laminate under combined loading (x vs xy );
(b) nal biaxial failure stress envelope for (90 /30 /90 ) E-glass/LY556 laminate under combined loading (x vs xy ).
The largest dierence between the theoretical 3.2.4 Biaxial envelope for (55)s E-glass/MY750
predictions of nal failure strength, Zinoviev: epoxy laminate under combined sy and sx (Case 9)
Rotem=3.31:1, occurred under biaxial tensile loading This type of laminate is commonly employed in pipes
(y : x 1 : 1). The bar charts in Fig. 9 show the varia- and pressure vessels. The initial failure envelopes are
tion between predictions for other stress ratios. shown in Fig. 12 and the nal envelopes in Fig. 13. The
Almost all of the theories predicted the initial failure bar charts (Fig. 14) compare initial and nal failure
load to be the same as the nal failure load over the stresses at selected loading ratios.
whole of the biaxial compression quadrant of the failure The various theories gave a range of results for the
envelope. All of the theories predicted nal failures that initial failure loads. Edge, Rotem, Chamis(2) and Puck
were dierent from the initial failures when tensile loads all predict very low strengths in biaxial tension. A rela-
were applied (except for Hart-Smith who did not predict tively small number of data points appear to have been
initial failure). The greatest ratio of nal: initial failure calculated by Chamis(2) and Eckold. Eckold's design
loads shown in the whole of Table 8 was 51:1, predicted approach is seen to be conservative in comparison with
by Chamis(2) for this quasi-isotropic laminate loaded at the other theories over a wide range of stress ratios, but
y : x 2 : 1. That result was taken from the stress/ predicts higher initial strengths than the others for a small
strain data (test Case 8 presented below) as Chamis did range of biaxial tensile loads (e.g. compare with Puck).
not provide data for the nal failure envelope for this The biggest dierence between predicted initial failure
laminate. envelopes was at stress ratios of 1:0 and 1:0 where the
Current failure theories for composite laminates 1241
Fig. 8. (a) Initial biaxial failure stress envelope for (0 /45 /90 ) AS4/3501-6 laminate under combined loading (y vs x ); (b) and
(c) nal biaxial failure stress envelopes for (0 /45 /90 ) AS4/3501-6 laminate under combined loading (y vs x ).
tension in the 90 plies. Failure in the second and third sile stresses in direction perpendicular to the bres. The
stages occurred in the 45 plies and 90 plies respec- transverse stress component changes from being tensile
tively with the same mode of failure as that in the 0 to being compressive when non-linear analysis is carried
plies. out.
The initial Young's modulus is identical in all predic-
3.3.2 Stress/strain curves for (55)s E-glass/MY750 tions but some of the theories predicted large changes of
epoxy laminate under uniaxial tension sy:sx=1:0 and laminate stiness after initial failure (Fig. 15). Tsai, Sun,
biaxial tension sy:sx=2:1 (Cases 10 and 11) Edge, Wolfe and Eckold showed only one failure point,
Stress/strain curves for Case 10 (SR=1:0) are shown in coincident with initial failure. Edge and Chamis
Fig. 15, the failure stresses in Fig. 14 and the failure predicted very dierent curves from the rest, with very large
strains in Fig. 17. In this case the laminae are, according strains, up to 12%, which was the specied limit of the
to linear analysis, subjected to high shear and low ten- graph provided by the organisers. At the other extreme,
Current failure theories for composite laminates 1243
Fig. 10. (a) and (b) Stress/strain curves for (0 /45 /90 )
AS4/3501-6 laminate under uniaxial tensile loading in y direc-
tion (y : x 1 : 0).
Fig. 9. Bar charts showing the biaxial failure stresses for (0 / linear (Edge, Sun (non-linear)). In Sun's linear analysis,
45 /90 ) AS4/3501-6 laminate under y : x 1 : 0, 1.5:1,
2:1,1:1 and 1:0. the "x strain curve intersected the "y strain curve at
around "y 1% while Zinoviev predicted an intersec-
tion point near nal failure, i.e. at "y 28%.
Eckold's curve stopped at only 0.5% strain. Thus, the ratio The range of initial failure occurrence (45156 MPa)
of Chamis' to Eckold's maximum strains is 24:1. is marked in Fig. 16a and b and that was predicted by
The nal failure stresses, shown in the bar chart in most of the participants to be due to tension
Fig. 14, dier by a maximum factor of 4.5 between perpendicular to the bres. Also marked on these g-
Eckold (640 MPa) and Chamis(2) (140 MPa). Note that ures is the range of nal failure occurrence (112
Eckold's results were taken from the envelope shown in 977 MPa). This nal failure was due to tension along the
Fig. 10 of Eckold.14 bre direction according to Zinoviev, Edge, Chamis,
Sun, Tsai, Rotem, Wolfe and Edge predicted that Puck and Sun and due to matrix failure in the case of
initial and nal failures were coincident events and the Wolfe. The latter gave the lowest failure strength.
failure was dominated by in-plane shear (Sun, Wolfe There was a large variation in the strains predicted in
and Edge). It was only Zinoviev who predicted separate the x direction. All predicted failure strains in that
failure loads with the initial failure due to shear and direction were tensile, except that of Edge. In Edge's
nal failure due to compression perpendicular to the analysis, the strain "x started as a positive value and
bres. became negative at high strains, see Fig. 16, while the
For test Case 11 (at y : x 2 : 1), a very wide range strain "y was comparable to the other predictions.
of stress/strain curves was predicted as shown in Fig. 16a
and b. In addition, the predicted initial and nal failure 3.3.3 Stress/strain curves for (0 /90 )s E-glass/MY750
strains in the x and y directions are shown as bar charts epoxy laminate under uniaxial tension sy:sx=0:1 (Case
in Fig. 17. The shape of the stress/strain curves varied 12)
from linear (Wolfe, Rotem and Eckold), and bi-linear This type of laminate has been researched fairly heavily
(Tsai, Zinoviev, Sun (linear) and Puck) to highly non- over the last 20 years, see for instance Bailey et al.,28 by
1244 P. D. Soden et al.
Fig. 12. Initial biaxial failure stress envelope for angle ply 55 E-glass/MY750 epoxy laminate under combined loading (y vs
x ).
Fig. 13. (a) and (b) Final biaxial failure stress envelope for Fig. 14. Bar charts showing the biaxial failure stresses for
angle ply 55 E-glass/MY750 epoxy laminate under com- 55 E-glass/MY750 laminate under y : x 1 : 0, 2:1, 2:1
bined loading (y vs x ). and 1:0.
1246 P. D. Soden et al.
Fig. 15. (a) and (b) Stress/strain curves for 55 E-glass/MY750 under uniaxial tensile loading with y : x 1 : 0.
The predicted curves were either linear (Tsai, Sun stresses but Rotem, Zinoviev, Puck and Edge predicted
(linear), Eckold and Chamis)) or non-linear (Puck, Sun nal failure stresses that were higher than the initial
(non-linear), Rotem and Wolfe) reecting the type of failure stresses (see the bar chart in Fig. 19b).
analysis employed in these theories.
Zinoviev showed unusual behaviour after initial fail-
ure, predicting failure strains up to 8%, Fig. 21. The 4 DISCUSSION
initial failure was by shear and the nal failure was by
transverse compression. Edge's curves are very dierent Theoretical results, obtained by the participants in the
from the others. His predicted failure strains reached failure exercise, have been presented in the form of
15%, far higher than any other strain predicted by the superimposed failure envelopes, bar charts, and super-
rest of the participants. imposed stress/strain curves for 0 , (90 /30 )s, (0 /
Wolfe, Tsai, Sun, Eckold and Chamis predicted nal 45 /90 )s, (55 )s, (45 )s and (0 /90 )s laminates.
failure stresses to be the same as the initial failure The dierences between the predictions of all the
Current failure theories for composite laminates 1247
Fig. 16. (a) and (b) Stress/strain curves for 55 E-glass/MY750 laminate under biaxial tensile loading with y : x 2 : 1.
theories for each of the cases analysed have been unidirectional bre-reinforced laminae, are reected in
described. Among the factors that have aected the the failure envelopes for the multidirectional laminates.
predicted failure envelopes and stress/strain curves for The high failure strength of the quasi-isotropic (0 /
the laminates are: 45 /90 )s AS4/3501-6 carbon/epoxy laminate under
equal biaxial compressive load predicted by Tsai in
. Dierences in the failure criteria applied at the Fig. 8 results from the enhanced strength that he predicts
lamina (or constituent) level (or otherwise). for unidirectional laminae under biaxial compression
. Allowances made in some theories for the dier- (e.g. Fig. 3).
ences in the lamina strength when tested in isola- The enhancement of the lamina biaxial transverse
tion and when embedded in a laminate. strength, predicted by Hart-Smith(2), is not shown in
. Inclusion of residual thermal stresses as a result of the results for multidirectional laminates presented here.
the curing process. That is primarily due to the fact the Hart-Smith chose
. Type of analysis (linear or non-linear). to omit the constant transverse strain failure lines from
. Dierences between post-failure models and his laminate failure envelopes for reasons stated in his
mechanisms. papers.17,18
In Eckold's predictions, the eects of equating long-
These factors are discussed in more detail below. itudinal compressive and tensile strengths of a lamina
are reected in the failure envelopes for the multi-direc-
4.1 Dierences between lamina failure criteria tional laminates. For example, the failure strength of
Some of the characteristics of the dierent failure cri- (90 /30 ) and 55 (Figs 4b and 13b) and laminates
teria, which were observed in the failure envelopes for was higher than that predicted by the other participants
1248 P. D. Soden et al.
Fig. 17. Bar charts showing failure strains for (a) 55 E- Fig. 18. Stress/strain curves for (0 /90 ) E-glass/MY750 lami-
glass/MY750 under y : x 2 : 1 (strain in y direction); (b) nate under uniaxial tensile loading with y : x 0 : 1.
55 E-glass/MY750 under y : x 2 : 1 (strain in x direc-
tion) and (c) 45 E-glass/MY750 under y : x 1 : 1
(strain in y direction).
4.2 Properties of embedded lamina
Some of the contributors chose to dierentiate between
in the compressioncompression quadrant. BS4994,27 the behaviour of an isolated lamina and that of a lamina
which uses Eckold's method, does allow for dierent embedded within a laminate. They assumed that the
tensile and compressive strengths to be employed if strength (occasionally the stiness) of a lamina embed-
measured values are available. Use of the compressive ded in a laminate is higher than that of an isolated
strength values provided in the exercise would remove lamina. For example, in his analysis of multidirectional
some of the anomalies from Eckold's results for laminae laminates, Sun, after studying work by Bailey et al.28
and laminates. and Flaggs and Kural,29 assumed new values for the
A number of theories showed that the biaxial lamina shear and transverse tensile strengths of the laminae,
strength under compressive load in one direction com- which were 50% higher than those provided by the
bined with tension in the other direction is smaller that organisers and employed by other participants.
than under uniaxial load, see Fig. 3. This is reected for Following the ndings of Rotem and Hashin,25
example in the low initial failure stresses predicted by Rotem assumed that, for all the laminae in the exercise,
Tsai compared with those of Zinoviev in the following the shear and transverse strengths, as well as the corre-
areas (i) (0 /45 /90 )s AS4/3501-6 carbon/epoxy sponding stiness of the embedded laminae, increased
laminate under equal biaxial compressive-tension load- by 20% above their values measured on isolated lami-
ing of y : x 1 : 1, see Fig. 8b, (ii) (30 /90 )s glass/ nae. He took the measured isolated lamina strengths as
epoxy laminate under y : x 425 : 1, Fig. 4b and c being the cause of initial cracking in the constrained
and (iii) (30 /90 )s glass/epoxy laminate under lamina and the modied strengths as controlling the
x : xy 235 : 1, Fig. 6b. In the last two cases, onset of nal failure. Thus, Rotem predicted, for
with the same input data, Tsai's theory predicted instance, that 45 glass/epoxy laminate, whose beha-
initial failure due to combined longitudinal com- viour is shear dominated under SR=1:1, nally failed
pression and transverse tension of the laminae whereas at an applied stress of 87.6 MPa, which is 20% higher
Zinoviev's theory predicted failure due to in-plane than the initial failure stress of 73 MPa. He also pre-
lamina shear. dicted that the 45 glass/epoxy laminate, subjected to
Current failure theories for composite laminates 1249
Fig. 20. Stress/strain curves for 45 E-glass/MY750 laminate under biaxial tensile loading with y : x 1 : 1.
was by matrix failure rather than the bre tension (e.g. those of Sun and Edge) predicted non-linear beha-
predicted by others. viour in both the "x and the "y curves, although, the
After initial failure in transverse tension, in addition degree of nonlinearity was more pronounced in the "x
to reducing E2 and G12, Rotem also reduced the mod- curve as this is matrix dominated.
ulus (E1) parallel to the bres in the failed lamina. In the In the case of Edge's prediction, which was based on
quasi-isotropic laminate loaded at SR=1:1, where all reducing gradually and simultaneously the transverse,
layers experience the same stress after initial failure, the shear moduli and Poisson's ratio, once transverse ten-
value of E1 will be reduced in all the laminae simulta- sion failure occurs, the slope of the "x curve changed
neously and the laminate cannot carry further loads. On gradually and eventually the strain switched from being
the other hand, in the 0 /90 laminate loaded in uniaxial tensile to being compressive.
tension (SR=0:1), only the 90 layer fails initially and The simultaneous drop to zero of both the transverse
the 0 layer can still carry load. and shear moduli in the case of Wolfe and Rotem, as
The eects of dierent post-failure models can also be described above, truncated the stress/strain curve.
observed in the stress/strain curves. One case which The two models proposed by Sun (non-linear) and
showed striking dierences between the theories was the McCartney, which relied on a crack density formulation
behaviour of the 55 laminate under y : x 2 : 1 for predicting the stiness degradation after initial fail-
(Fig. 16). Theories which employ a gradual non-linear ure, gave very similar shaped stress/strain curves to
reduction in lamina properties after the initial failure nearly all the other theories for the 0 /90 laminate
Current failure theories for composite laminates 1251
Fig. 21. Stress/strain curves for 45 E-glass/MY750 laminate under biaxial tensile loading with y : x 1 : 1:
under uniaxial tension, Fig. 18. There was one notice- 4.5 Non-linear behaviour
able dierence between the models of McCartney and The main source of non-linearity in the behaviour of the
Sun and that was the transverse strain predicted by Sun, laminates preceding initial failure was the decrease in in-
which was larger than that predicted by McCartney. plane lamina shear stiness with increasing shear
However, when the same two models (Sun and strain.12 The 45 laminate loaded at SR=1:1 (load
McCartney) were used to predict the stress/strain curve Case 14) was equivalent to a 0 /90 cross-ply laminate
for the 45 laminate under equal biaxial tension, they loaded in pure shear so the stress/strain curves in Fig. 21
gave dierent results. Sun's results were remarkably should reect the given non-linear shear curve for the
similar to most of the others in terms of the slope of the unidirectional lamina supplied by Soden et al.12 For that
stress/strain curves and also the nal failure strength case the initial failure strains in the non-linear analyses
which was around 640 MPa, representing bre tension (Puck, Edge, Rotem, Wolfe and Sun (non-linear)) were
failure. McCartney's results were stier than the others around 2% which is , as expected, half the shear strain
and he did not predict nal failure. to failure of an isolated lamina. That predicted strain
1252 P. D. Soden et al.
was up to 3.2 fold higher than the initial failure strains the applied load would lead to imminent failure at the
in the linear analyses (Zinoviev, Chamis, Sun and Tsai) onset of that large deformation.
which produce truncated stress/strain curves. There were other cases where large deformations led
In the laminates whose behaviour is not aected by to open (discontinuous) failure envelopes (i.e. failure
in-plane lamina shear, linear analysis should give iden- envelopes with indeterminate regions). Puck reported
tical answers to non-linear analysis for initial failure that both the initial and nal biaxial failure envelopes
prediction. The behaviour of a 45 laminate under for the 55 laminate were open where large shear
SR=1:1 and the 0 /90 laminate under SR=0:1 are strains, combined with transverse tension stress, were
clear examples of stress/strain curves where shear non- induced in the plies. The initial failure envelope was
linearity is absent. open in the range 7 : 1 > y : x > 25 : 1. The nal failure
Zinoviev introduced an additional geometric non-lin- envelope was also open at the same range of stress
earity by allowing for the change of ply angle as defor- ratios and in another area between 0 : 1 < y : x < 2 : 1.
mation progressed. It is this geometric non-linearity Zinoviev also reported the occurrence of an open
which results in the slight increase in stiness at large envelope for the nal biaxial failure stresses between
strains in the stress/strain curves for the cross ply lami- stress ratios 0 : 1 < y : x < 2 : 1, which is the same as the
nate loaded in pure shear, Fig. 21. second range reported by Puck.
Certain theories (Edge, Sun(non-linear) and Puck)
introduced additional material non-linearity by assum-
ing that the transverse and shear moduli decreased 5 CONCLUSIONS
gradually after initial failure. It is the combined non-
linearities which result in the non-linear divergence of (a) Leading theories have been employed, by their orig-
the strains "x in the stress/strain curves predicted by Sun inators, to predict the strength and deformation
(non-linear) and Edge for the 55 laminate loaded at response in 14 test cases, involving six dierent
SR=2:1 (Fig. 16). FRP laminates (0 , (90 /30 )s, (0 /45 /90 )s,
The inclusion of any of these forms of non-linear (55 )s, (45 )s and (0 /90 )s laminates)
behaviour into the laminate analysis usually requires the subjected to a range of in-plane biaxial loading
introduction of iterative numerical methods of solution. conditions.
(b) The exercise revealed that some theories lack
4.6 Large and unbounded deformations exibility and other require further development.
Several of the participants encountered problems in One contributor who adopted a `damage
deciding when to terminate numerical solutions that mechanics' approach solved only two of the 14
predicted ever-increasing deformations. For example, test cases.
Edge produced stress/strain curves for 55 laminates (c) These were many similarities between the predic-
under SR y : x 1 : 0 and SR=2:1 and for the 45 tions of various theories, but some striking dif-
laminate under SR=1:1 in which the predicted failure ferences.
strains were very large compared with those obtained by (d) A number of the test cases (Nos 13) were chosen
others. to compare the theories at the most simplistic
In Chamis's results, large failure strains were shown level, i.e. that of a unidirectional bre-reinforced
for the (0 /45 /90 )s AS4/3501-6 quasi-isotropic lami- lamina. No two theories gave the same biaxial
nate under SR=1:0, in the 45 laminate under failure envelopes for all these cases, and dier-
SR=1:1 and in the 55 laminates under SR=1:0. ences as great as 570% were observed in the
Chamis13 pointed out that the large increase in the strength predictions.
strains with very little increase in the stress would hap- (e) At the next level of complexity, certain test cases
pen only in monotonic loading and any uctuation in were chosen by the organisers because they
Current failure theories for composite laminates 1253
3. Owen, M. J. and Rice, D. I., Biaxial strength behaviour 17. Hart-Smith, L. J., Predictions of the original and trun-
of glass-reinforced polyester resins. In Composite cated maximum-strain failure models for certain brous
Materials: Testing and Design, 6th Conf., ASTM STP composite laminates. Compos. Sci. Technol., 1998, 58(7),
787, ed. I. M. Daniel. 1982, pp. 124144. 1151.
4. Soni, S. R., A comparative study of failure envelopes in 18. Hart-Smith, L. J., Predictions of a generalized maximum-
composite laminates. J. Reinf. Plast. Compos., 1983, 2, shear-stress failure criterion for certain brous com-
3442. posite laminates. Compos. Sci. Technol., 1998, 58(7),
5. Tsai, S. W., A survey of macroscopic failure criteria for 1179.
composites materials. J. Reinf. Plast. Compos., 1984, 3, 19. Puck, A. and Schurmann, H., Failure analysis of FRP
4062. laminates by means of physically based phenom-
6. Tsai, S. W., Composite Design, 4th edn. Think Com- enological models. Compos. Sci. Technol., 1998, 58(7),
posites, Dayton, Ohio, 1988. 1045.
7. Rowlands, R. E., Strength (failure) theories and their 20. Rotem, A., Prediction of laminate failure with the Rotem
experimental correlation. In Handbook of Composites, failure criterion. Compos. Sci. Technol., 1998, 58(7), 1083.
Vol 3, Failure Mechanics of Composites, ed. G. C. Sih 21. Sun, C. T. and Tao, J. X., Prediction of failure envelopes
and A. M. Skudra. Elsevier Science Publishers B.V., 1985, and stress/strain behaviour of composite laminates.
pp. 71125. Compos. Sci. Technol., 1998, 58(7), 1125.
8. Nahas, M. N., Survey of failure and post failure theories 22. Liu, K.-S. and Tsai, S. W., A progressive quadratic
of laminated bre reinforced composites. J. Compos. failure criterion for a laminate. Compos. Sci. Technol.,
Technol. Res., 1986, 8, 138153. 1998, 58(7), 1023.
9. Chen, A. S. and Matthews, F. L., A review of multiaxial/ 23. Wolfe, W. E. and Butalia, T. S., A strain-energy based
biaxial loading tests for composite materials. Composites, failure criterion for nonlinear analysis of composite lami-
1993, 24, 395406. nates subjected to biaxial loading. Compos. Sci. Technol.,
10. Daniel, I. M. and Ishai, O., Engineering Mechanics of 1998, 58(7), 1107.
Composite Materials. Oxford University Press, Oxford, 24. Zinoviev, P., Grigoriev, S. V., Labedeva, O. V. and
UK, 1994. Tairova, L. R., Strength of multilayered composites
11. Hinton, M. J. and Soden, P. D., Predicting failure in under plane stress state. Compos. Sci. Technol., 1998,
composite laminates: the background to the exercise 58(7), 1209.
Compos. Sci. Technol., 1998, 58(7), 1001. 25. Rotem, A. and Hashin, Z., Failure modes of angle ply
12. Soden, P. D., Hinton, M. J. and Kaddour, A. S., Lamina laminates. J. Compos. Mater., 1975, 9, 191206.
properties, lay-up conguration and loading conditions 26. Hinton, M. J., Soden, P. D. and Kaddour, A. S., Strength
for a range of bre reinforced composite laminates, of composite laminates under biaxial loads. Appl. Com-
Compos. Sci. Technol., 1998, 58(7), 1011. pos. Mater., 1996, 3, 151162.
13. Gotsis, P. K., Chamis, C. C. and Minnetyan, L., Predic- 27. BS4994: Specication For Design And Construction Of
tion of composite laminate fracture: micromechanics and Vessels And Tanks In Reinforced Plastics, BSI, London,
progressive fracture. Compos. Sci. Technol., 1998, 58(7), UK, 1987.
1137. 28. Bailey, J. E., Curtis, P. T. and Parvisi, A., On the trans-
14. Eckold, G. C., Failure criteria for use in the design verse cracking and longitudinal splitting of glass and car-
environment. Compos. Sci. Technol., 1998, 58(7), 1095. bon bre reinforced epoxy cross ply laminates and the
15. Edge, E. C., Stress based Grant-Sanders method for pre- eect of Poisson and thermally generated strain. Proc. R.
dicting failure of composite laminates. Compos. Sci. Soc. Lond., 1979, A366, 599623.
Technol., 1998, 58(7), 1033. 29. Flaggs, D. L. and Kural, M. H., Experimental deter-
16. McCartney, L. N., Predicting transverse crack formation mination of the in-situ transverse laminate strength in
in cross-ply laminate. Compos. Sci. Technol., 1998, 58(7), graphite epoxy laminates. J. Compos. Mater., 1982, 16,
1069. 103116.