Professional Documents
Culture Documents
www.elsevier.com/locate/compstruct
Abstract
Due to their high strength to weight ratios, laminated composite materials have found extensive applications in the construction of
mechanical, aerospace, marine and automotive structures. Predictions of the failure of composite structures and the maximum loads
that the structures can withstand have become an important topic of research for reliability assurance. In particular, the first-ply fail-
ure analysis of laminated composite plates subjected to transverse loads has drawn close attention in recent years. Accurate prediction
of failure of composite structures has become more challenging to designers in the presence of inherent scatter in the material prop-
erties. The objective of the present study is to improve the efficacy of composite material design by predicting the statistics of first-ply
failure of orthotropic plates with random material properties under random loading. In the present study stochastic finite element
solutions of probabilistic failure load of composite laminated plates are obtained with layer-wise plate theory and analytical solutions
are found using Kirchhoff–Love plate theory. Tsai-Wu and Hoffman criteria have been adopted to predict the first-ply failure load
and first-order perturbation technique has been used to evaluate the mean and variance of failure statistics. The capability of stochas-
tic finite element in predicting the first-ply failure load statistics has been studied by comparing the results with analytical solutions. It
has been found that the stochastic finite element has high accuracy for the cases considered. The present solutions are also compared
with results available in the literature. Numerical results have been presented to illustrate the application of the developed procedure.
2005 Elsevier Ltd. All rights reserved.
Keywords: Composite laminated plates; Random material properties; Random loads; Stochastic finite element method; Probabilistic failure
0263-8223/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruct.2005.06.006
80 A.K. Onkar et al. / Composite Structures 77 (2007) 79–91
Let the plate be subjected to a uniformly distributed 3. Mean and variance of response: a perturbation
transverse loading Q(x, y). The work done by the exter- based approach
nal forces is:
Z Z Mean centered first-order perturbation approach has
a b
been adopted for obtaining the first-ply failure statistics
WD ¼ Qðx; yÞwðx; yÞ dx dy. ð3Þ
0 0 of flat laminated composite plate with randomness in
material properties and loading. It is assumed that all
The boundary conditions for a cross-ply plate having the material properties and loading components are
all sides simply supported with edges free to move in uncorrelated to each other. It can be shown that in case
their respective in-plane normal directions are: of correlated random variables, it is possible to express
them in terms of an uncorrelated set by a transformation
Along x ¼ 0 and x ¼ a for all y; using eigen vectors of the covariance matrix [12]. It is
also assumed that the dispersion of each random quan-
v ¼ 0; w ¼ 0 : N x ¼ 0; M x ¼ 0
ð4Þ tity about its mean value is small, which is true in most
Along y ¼ 0 and y ¼ b for all x; sensitive engineering application.
Using a Taylor series expansion any system param-
u ¼ 0; w ¼ 0 : N y ¼ 0; M y ¼ 0.
eter, which represents structural uncertainty in the
plate, can be expanded in terms of the basic random
The following sets of admissible functions are as-
variables. The primary variables bl are assumed to be
sumed that satisfy the above boundary conditions:
the basic material properties and the elements of the
mpx npy applied load. As assumed, the random part in the pri-
u0 ¼ U mn cos sin ; mary variables is small in magnitude compared to their
a b
mpx npy mean values, the second and higher order terms are ne-
v0 ¼ V mn sin cos ; ðm; n ¼ 1; 2; . . . ; 1Þ ð5Þ glected and the expression for the system operator may
a b
mpx npy be put as:
w0 ¼ W mn sin sin
a b T i ¼ T 0i þ T ;ri ðbr b0r Þ ðr ¼ 1; 2; . . . ; RÞ; ð8Þ
where Umn, Vmn and Wmn are the maximum displace- where br, r = 1, 2, . . . , s are the material variables and br,
ments in x and y directions for a particular value of r = s + 1, s + 2, . . . , s + s1 are the external loading vari-
(m, n). ables. Thus the total number of the independent basic
The strain energy and work done by external forces random variables is s + s1 = R (say). Here the symbol
can be obtained by substituting Eq. (5) into Eqs. (2) (Æ)0 represents the value of function taken at the mean
and (3) and performing the required integrations. Based value of the input random variables while (Æ)r stands
on Rayleigh–Ritz principle, the governing modal equa- for the first partial derivative with respect to the random
tions of motion for uniformly distributed transverse variables br evaluated at the mean of the input random
loading q0 are: variables.
The displacements are also influenced by the struc-
T 1 U mn þ T 2 V mn ¼ 0; tural uncertainty and thus similar expression for dis-
T 2 U mn þ T 3 V mn ¼ 0; ðm; n ¼ 1; 2; . . . ; 1Þ ð6Þ placements can be written as:
and T 4 W mn ¼ S mn q0 ; U mn ¼ U 0mn þ U ;rmn ðbr b0r Þ;
where T1, T2, T3, T4 and Smn are expressed as: V mn ¼ V 0mn þ V ;rmn ðbr b0r Þ;
ð9Þ
mp2 np2 and W mn ¼ W 0mn þ W ;rmn ðbr b0r Þ
T 1 ¼ A11 þ A66 ;
a b ðm; n ¼ 1; 2; . . . ; 1; r ¼ 1; 2; . . . ; RÞ.
mpnp
T 2 ¼ ðA12 þ A66 Þ ; Substituting Eqs. (8) and (9) into Eq. (6), we arrive at
a b
np2 mp2 the following two equations by equating the zeroth and
T 3 ¼ A22 þ A66 ; first powers of small order terms:
b a
mp4 mp2 np2 np4
T 4 ¼ D11 þ 2ðD12 þ D66 Þ þ D22 ; • Zeroth-order:
a a b b
m n T 01 U 0mn þ T 02 V 0mn ¼ 0;
ab½ð1Þ 1½ð1Þ 1
S mn ¼ ðm; n ¼ 1; 2; . . . ; 1Þ.
mnp2 T 02 U 0mn þ T 03 V 0mn ¼ 0; ðm; n ¼ 1; 2; . . . ; 1Þ ð10Þ
ð7Þ and T 04 W 0mn ¼ S mn Q00 .
82 A.K. Onkar et al. / Composite Structures 77 (2007) 79–91
• First-order: • Zeroth-order:
Z Z
T ;r1 U 0mn þ T 01 U ;rmn þ T ;r2 V 0mn þ T 02 V ;rmn ¼ 0; C 0ijkl e0ij de0kl dX ¼ t0i du0i dC1 ði; j; k; l ¼ 1; . . . ; 3Þ.
T ;r2 U 0mn þ T 02 U ;rmn þ T ;r3 V 0mn þ T 03 V ;rmn ¼ 0; X C1
ð11Þ ð14Þ
T ;r4 W 0mn þ T 04 W ;rmn ¼ S mn ðQ0 Q00 Þ
ðm; n ¼ 1; 2; . . . ; 1; r ¼ 1; 2; . . . ; RÞ. • First-order:
Z Z Z
Having solved Eqs. (10) and (11) for the mean value of C 0ijkl e;rij de0kl dX þ C ;rijkl e0ij de0kl dX ¼ t;ri du0i dC1
displacement amplitudes and their first-order deriva- X X C1
Z Z Z
tives, one can also obtain the mean values of stresses or C 0ijkl e;rij de0kl dX ¼ t;ri du0i dC1 C ;rijkl e0ij de0kl dX
and their derivatives with respect to the input random X C1 X
variables by using the linear stress–strain relationship. ði; j; k; l ¼ 1; . . . ; 3; r ¼ 1; 2; . . . ; RÞ.
Subsequently the expression for the variance of the
ð15Þ
response can be expressed as follows:
2
VarðU mn Þ ¼ E½fU ;rmn ðbr b0r Þg ; where R is the number of basic random variables chosen
for the analysis. The zeroth-order equations are used to
VarðV mn Þ ¼ E½fV ;rmn ðbr b0r Þg2 ; ðr ¼ 1; 2; . . . ; RÞ obtain the mean response and the first-order equations
and VarðW mn Þ ¼ E½fW ;r
mn ðbr b0r Þg2 . give the first-order response derivative with respect to
ð12Þ the basic random variables.
7. Results and discussions of the SFEM, the present approach is first validated
with closed form solutions as developed in the previous
In the present failure study of laminated composite section and also with results available in the literature.
plates, it is assumed that the ply is composed of laminae The effects of material property and load dispersion
which possess the same material properties throughout along with variations in thickness ratio, aspect ratio
the thickness. The laminates considered for generating and lay-up sequences on failure statistics are also
the results are made of T300/5208 graphite/epoxy mate- explored.
rial with properties as listed below:
7.1. Validation
E11 ¼ 132.5 GPa; E22 ¼ E33 ¼ 10.8 GPa;
G23 ¼ 3.4 GPa; G12 ¼ G13 ¼ 5.7 GPa; A thin square laminated plate of size 50 · 50 mm con-
m12 ¼ m13 ¼ 0.24; m23 ¼ 0.49. sisting of four layers [0/90/90/0] of equal thickness with
b/h = 100 having all edges simply supported is consid-
The ultimate strengths for the above material which are
ered for the present validation. The plane profile of
used to calculate the strength parameters are defined as
the uniform mesh of 20 · 20 elements in each layer is
X T ¼ 1515 MPa; X C ¼ 1697 MPa; shown in Fig. 3. Each element is six-noded prismatic.
Y T ¼ Z T ¼ 43.8 MPa; Y C ¼ Z C ¼ 43.8 MPa; A uniformly distributed load is applied on the top sur-
face. Both Tsai-Wu and Hoffman failure criteria are
R ¼ 67.6 MPa; S ¼ T ¼ 86.9 MPa.
used to compare the mean failure load and the statistics
In the present analysis the elastic moduli of failure index.
(E11, E22, m12, m23, G12, G23) of the material are treated as In order to validate the layer-wise model implementa-
independent random variables. The laminated plate is tion, the failure loads obtained using the layer-
subjected to a uniform distributed random load. The wise model are compared with that obtained using a
geometry and boundary conditions are shown in Figs. closed form solution and Kirchhoff–Love (KL) model.
1 and 2 respectively. The finite element analysis is per- From the results given in Table 1, it can be concluded
formed using pxy = 3 and pz = 1. To verify the accuracy that
Fig. 2. Boundary conditions of a laminated composite plate. (a) SSSS, (b) SCSC and (c) SFSF.
A.K. Onkar et al. / Composite Structures 77 (2007) 79–91 85
Table 2
Comparison of COV of failure index for [0/90/90/0] square laminate
with b/h = 100 having SSSS boundary condition
COV of COV of failure index f(r)
all BRVs
Tsai-Wu Hoffman
SFEM Closed form SFEM Closed form
0.00 0.0000 0.0000 0.0000 0.0000
0.04 0.1299 0.1298 0.1171 0.1171
0.08 0.2599 0.2597 0.2343 0.2343
0.12 0.3898 0.3896 0.3514 0.3514
016 0.5197 0.5195 0.4685 0.4685
0.20 0.6497 0.6494 0.5856 0.5856
Fig. 3. Finite element discretization in x–y plane for full plate analysis. This example validates the SFEM implementation of
the present study. It should be noted that the two failure
models, i.e., Tsai-Wu and Hoffman, give different values
Table 1 of the mean failure load (Table 1). It is also noteworthy
Comparison of the mean failure load for [0/90/90/0] square laminate that effect of material defects on the failure index, is
with b/h = 100 having SSSS boundary condition significant.
Different failure criteria Mean failure load (MPa) The comparison of the mean failure load (FL) ob-
SFEM Closed form tained using the present formulation is also made with
Tsai-Wu 0.07292 0.07306 results reported in literature. A thin laminated plate of
Hoffman 0.06246 0.06250 size 229 · 127 mm and h = 0.508 mm with all edges
clamped is considered. The material properties used in
this analysis is same as defined above. Table 3 shows
the non-dimensionalised mean failure load, for four-lay-
(i) The mean failure load obtained using the layer- ered laminates with different lay-ups, using Tsai-Wu fail-
wise model is close to the closed form results. ure criterion. The non-dimensionalised mean failure
(ii) The mean load obtained using layer-wise model is load obtained using layer-wise plate model is compared
lower than that for the closed form solution. with those reported in [5], using a first-order shear
deformable plate theory (FSDT). Further, the value of
The layer-wise model gives lower failure loads be- the failure load obtained using an equivalent layer (by
cause this model is less stiff as compared to KL model. assuming all four material layers as a single solution
Further, for thin plates the behaviour is accurately pre- layer) is also presented in Table 3. It is observed that:
dicted by the KL model, i.e., shear effects are negligible.
This is the reason why the layer-wise model and KL (i) The conventional 2D plate models overpredict
model give very close values of the failure load. mean failure loads as compared to those obtained
In order to validate the SFEM implementation the ef- with the layer-wise plate model.
fect of randomness of the material data and loading, on
the calculated failure index, is calculated. In Table 2 the
coefficient of variation (COV) of the failure load is re-
ported, with respect to change in the random input vari- Table 3
ables, for the thin symmetric cross-ply laminate Comparison of the mean non-dimensionalized failure load for different
considered above. From the results it can be noted that: laminates with CCCC boundary condition
Lay-ups Non-dimensionalised mean failure load
(i) The COV of failure index obtained using SFEM is (FLa4/E22h4)
close to that obtained using the closed form Layer-wise Equivalent Reddy
solution. (present) (present) [5] result
(ii) The failure index is very sensitive to change in the [45/45/45/45] 31,901.47 34,083.44 39,354.8
input data. For example, for a COV of 4% in input [0/90/90/0] 15,440.58 16,568.27 19,050.9
[90/0/90/0] 24,521.95 25,601.95 30,412.3
material data, the COV of failure index is 13%.
86 A.K. Onkar et al. / Composite Structures 77 (2007) 79–91
(ii) The mean failure load obtained using equivalent dimensionalised failure index (FI) with dispersion in
single layer lies between the values obtained using all the basic random variables (BRVs) changing simulta-
layer-wise and Reddys 2D plate model. neously for ply schemes 1 and 2 having SSSS, SCSC and
SFSF boundary conditions are presented in Figs. 4–6
The failure loads, obtained using the layer-wise model, respectively. It is found that
are approximately 20% less than those reported in [5].
The failure load, obtained using the equivalent layer (i) Angle ply is more affected by dispersion in the
model, is closer to those reported in [5]. This is because input variables compared to cross-ply laminate.
the FSDT and equivalent models are stiffer as compared (ii) Boundary conditions also play an important role
to the layer-wise model. in the stochastic analysis.
(iii) The variation in FI is most sensitive for SCSC
7.2. Second-order failure statistics of anti-symmetric boundary condition. This can be observed from
laminated plates Fig. 5(a) and (b) that the COV of failure index,
for a COV of 10% in all BRVs, is 25% for [0/
Two different laminated plates of size 50 · 50 mm 90] laminates whereas for [45/45] laminates it is
with the following lay-ups schemes are used to illustrate 55%.
the present method. These are (iv) For a COV of 10% in all BRVs, the COV of failure
index for SFSF laminated plates with [0/90] lay-
Ply scheme 1: [0/90] ups is 10% whereas for [45/45] lay-ups it is
Ply scheme 2: [45/45] 30%. As the constraint violation is less in the
case of SFSF boundary condition compared to
Three different types of boundary conditions SSSS, SSSS and SCSC, the effect of BRV on COV of
SCSC and SFSF (illustrated in Fig. 2) are used to study FI are found to be less compared to the other
the probabilistic failure of the laminates. Plate thickness two boundary conditions.
ratio b/h = 100 has been used in this study. The finite
element mesh shown in Fig. 3 is used to generate all
the results.
The mean failure load predicted by Tsai-Wu and
Hoffman criteria by keeping failure index equal to 0.8
are listed in Table 4 for square plate with different lay-
ups and boundary conditions. These failure loads ensure
that if a factor of safety of 1.2 were chosen for the de-
sign, the plates would still be assumed to be safe in the
deterministic environment. It can also be observed that
Hoffman criterion predicts higher mean failure load
compared to Tsai-Wu criterion for this set of loading,
material and boundary condition.
Table 4
Mean failure load for square plate with different lay-ups and boundary
conditions with b/h = 100
Ply schemes Failure Mean failure load for boundary
criteria conditions
SSSS SCSC SFSF
[0/90] Tsai-Wu 0.02469 0.01869 0.01113
Hoffman 0.02827 0.02397 0.01274
Fig. 4. Influence of SD of all basic random inputs changing simulta-
[45/45] Tsai-Wu 0.04323 0.03059 0.01412
neously on COV of failure index for different lay-ups with SSSS
Hoffman 0.05526 0.04098 0.01589
boundary condition and b/h = 100. (a) [0/90] and (b) [45/45].
A.K. Onkar et al. / Composite Structures 77 (2007) 79–91 87
Fig. 5. Influence of SD of all basic random inputs changing simulta- Fig. 6. Influence of SD of all basic random inputs changing simulta-
neously on COV of failure index for different lay-ups with SCSC neously on COV of failure index for different lay-ups with SFSF
boundary condition and b/h = 100. (a) [0/90] and (b) [45/45]. boundary condition and b/h = 100. (a) [0/90] and (b) [45/45].
Assuming a normal distribution for the FI, which (iii) The dispersion depends strongly on the failure cri-
roughly implies that two-thirds (68.26%) of the occur- terion used.
rences fall within 1r limit on either side of the average
and practically all (99.73%) fall within 3r limits, it can Fig. 8(a)–(g) shows the effect of individual random
be observed that both antisymmetric cross-ply and an- variable on failure index for the same laminate with
gle-ply fail when the effect of dispersion in the basic in- SCSC boundary condition. Trends similar to SSSS plate
put variables are incorporated using a probabilistic have been observed with higher dispersion in failure
model. index.
Fig. 7. Influence of SD of individual basic random input on COV of failure index of [0/90] square laminate with SSSS boundary condition and b/
h = 100. Variation in (a) E11, (b) E22, (c) m12, (d) m23, (e) G12, (f) G23 and (g) q0.
for a COV of 10% in the BRVs. The variation in FI for to 50 and decreases with further decrease in thickness ra-
ply scheme 2 is sensitive to change in thickness ratio. It tio. The dispersion in the FI is generally observed to be
initially increases for thickness ratio changing from 100 more than that in the BRVs.
A.K. Onkar et al. / Composite Structures 77 (2007) 79–91 89
Fig. 8. Influence of SD of individual basic random input on COV of failure index of [0/90] square laminate with SCSC boundary condition and b/
h = 100. Variation in (a) E11, (b) E22, (c) m12, (d) m23, (e) G12, (f) G23 and (g) q0.
Fig. 9. Influence of SD of all basic random inputs changing simulta- Fig. 10. Influence of SD of all basic random inputs changing
neously on COV of failure index of square laminates with SSSS simultaneously on COV of failure index of square laminates with
boundary condition for different aspect ratios. (a) [0/90] and (b) [45/ SSSS boundary condition for different thickness ratios. (a) [0/90] and
45]. (b) [45/45].
[10] Frangopol DN, Recek S. Reliability of fiber-reinforced composite [13] Harkovitch CT. Mechanics of fibrous composites. New York: -
laminate plates. Probab Eng Mech 2003;18:119–37. John Wiley and Sons; 1998.
[11] Ahmad NU, Basu PK. Higher-order finite element modeling of [14] Jones RM. Mechanics of composite materials. New York: Mc-
laminated composite plates. Int J Numer Method Eng Graw-Hill; 1975.
1994;37:123–39. [15] Kleiber M, Hein TD. The stochastic finite element method. New
[12] Nigam NC. Introduction to random vibration. Cambridge, York: John Wiley and Sons; 1992.
MA: MIT Press; 1983.