Composite Structures 203 (2018) 153–163

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

A novel univariate method for mixed reliability evaluation of composite T

laminate with random and interval parameters
⁎
Xiao Li1, Zheng Lv1, Zhiping Qiu
Institute of Solid Mechanics, Beihang University, Beijing 100191, China

A R T I C LE I N FO A B S T R A C T

Keywords: Requiring probability density distribution functions of uncertain variables, it is difficult for conventional
Composite laminates structural reliability analysis methods to calculate the reliability of composite laminates with both random and
Mixed reliability interval variables. To evaluate the safety of composite laminated structures under this circumstance, this paper is
Random variable aimed at developing a precise and efficient univariate method for the reliability assessment of composite la-
Interval variable
minates taking use of Legendre orthogonal polynomials and Monte Carlo Simulation. In this paper, based on the
Univariate method
Legendre orthogonal polynomial
Tsai-Wu failure criterion and first-ply failure assumption, the failure criterion of composite laminate is con-
structed firstly. Then the performance function of each lamina is approximately expressed as the sum of uni-
variate contributions of all uncertain variables on the basis of the univariate method. Taking use of Legendre
orthogonal polynomials, the univariate functions of random and interval variables are constructed by the least
square fitting method and the univariate contribution bounds of interval variables are derived afterwards.
Furthermore, the interval of failure probability can be calculated by substituting contribution bounds of interval
variables and sample points of random variables into univariate function. At last, two examples and an en-
gineering application are provided to demonstrate the efficiency and accuracy of the proposed method.

1. Introduction [10] have firstly studied the statistical behaviors of ultimate strengths
of composites with random strength parameters. The uncertainties of
Having advantageous characteristics of high strength, high stiffness material properties were not considered. Yang [11] has introduced the
and low weight, composite laminates have been widely applied in many HL reliability index constructed by Hasofer and Lind into the safety
engineering fields, such as aerospace industries [1,2], automotive en- evaluation of composite laminates. Engelstad and Reddy [12] have
gineering [3,4] and naval equipment [5,6], etc. However, it has been developed a probabilistic finite element analysis procedure for com-
found that the properties of composite laminates are usually un- posite laminated shells considering material properties, ply thickness
avoidably subjected to a certain degree of uncertainties due to com- and ply angles as random variables. Lin and Kam [13] have evaluated
plicated manufacturing processes and inherent dispersions of con- the failure probability of composite laminates under transverse loads
stituents [7]. Traditional approaches of safety factors may result in a utilizing stochastic finite-element analysis. Employing statistical data
costly and unnecessary conservatism, which may block the applications from experiments, they determined the probability distributions of la-
of composite laminates [8,9]. As a consequence, the safety analysis and mina strengths, material properties, and laminate thickness and ex-
optimization design of composite laminates with uncertain variables amined the probabilistic properties of different failure criterion. Phi-
are quite important in the applications of composite laminates. lippidis and Lekou [14] have established two efficient and accurate
Probabilistic reliability-based analysis methods are powerful tools approaches for the definition of cumulative distribution function of the
to evaluate the safety of composite laminates with uncertain parameters failure condition or unidirectional FRP laminae under complex in-plane
and many researches have been done based on the probability theory. loading. The statistical aspects of the basic strengths of the material are
Probabilistic uncertain analysis methods, such as moment-based taken into account. Kogiso [15] et al. have studied the reliability of
methods, probabilistic/stochastic finite element methods (FEM), sur- composite laminates with initial imperfection regarding the initial im-
rogate model based methods and their combinations are generally ap- perfection, strength and elastic parameters as random variables. In the
plied in the uncertain analysis of composite laminates. Sun and Yamada literature [16], Haeri and Fadaee have introduced an advanced Kriging

⁎
Corresponding author.
E-mail address: [email protected] (Z. Qiu).
1
These authors contributed equally to this work.

https://doi.org/10.1016/j.compstruct.2018.06.097
Received 27 March 2018; Received in revised form 8 June 2018; Accepted 25 June 2018
Available online 28 June 2018
0263-8223/ © 2018 Published by Elsevier Ltd.
X. Li et al. Composite Structures 203 (2018) 153–163

surrogate model in combination with Monte Carlo Simulation (MCS) direction of the applied loads, the failure of composite lamina may be
into the reliability analysis of composite laminates regarding both encountered on the direction of fiber or matrix [16]. There are mailny
elastic and strength variables as random variables. two kinds of failure criteria for the failure judgment of fiber reinforced
It can be seen that great achievement has been gained in the lamina, namely non-interactive and interactive failure criteria [7]. Non-
probability theory on the basis of uncertain analysis methods regarding interactive failure criteria such as Maximum stress or Maximum strain
material properties, strength properties et al. as random variables. But criteria are non-conservative for specific stress combinations [40].
it should be noticed that the probability density functions needed in the Considering independence of each stress direction, more limit state
application of probabilistic reliability methods usually demand suffi- functions are needed for the reliability evaluation of lamina. These two
cient information of uncertain parameters for the consideration of ac- reasons limit the application of non-interactive failure criteria. Taking
curate uncertain analysis [17–19]. While, it is usually hard to get suf- interaction between different stress components into consideration, the
ficient information of uncertain parameters such as destructive constructed limit state functions are much less than that of non-inter-
mechanical property experiments for some developed composites [20]. active failure criteria and have been widely applied in reliability eva-
According to the statistical methods described in Composite Materials luation of composite laminated structures [7]. Using all the strength
Handbook [21], tests for goodness-of-fit should be performed for data to define a failure surface in stress space, Tsai-Wu failure criterion
normal, Weibull and log-normal distribution characterization of elastic is a typical interactive failure criterion and has been widely applied in
and strength variables. If the observed significance level (OSL) is less the reliability analysis of composite laminated structures. In Ref [13],
than 0.05, that means that the accuracy of the quantification for the Lin and Kam evaluated reliability of composite laminates utilizing dif-
distribution is unsatisfactory which may cause errors in reliability ferent failure criteria. The results demonstrated that the Tsai-Wu failure
evaluation. The reliability evaluation of composites with the probabil- criterion obtained the most accurate result in comparation with ex-
istic distributions may be not applicable. Having the characteristic of periments. Considered accuracy and convenient, Tsai-Wu failure cri-
low dependence of uncertain information, interval methods exhibit teria will be employed in this paper to construct performance function.
high advantages in dealing with insufficient uncertain information Based on Tsai-Wu failure criteria, the failure index can be expressed
[22–25], and have been applied in many fields such as buckling ana- as
lysis [26], vibration analysis [27] and structural reliability analysis
F . I . =F1 σ1 + F2 σ2 + F11 σ12 + F22 σ22 + F66 τ12
2
+ 2F12 σ1 σ2 (1)
[28] et al. Although it has achieved many successes in probability and
interval theory, there usually exist both random and interval variables where F.I. denotes the failure index, σ1, σ2 and τ12 are the 1 direction
in some certain cases [20,29]. As it may lead to wrong results or unsafe stress, 2 direction stress and shear stress in the 1–2 plane, respectively.
design considering only random or interval uncertainties, the safety Fi and Fij are strength coefficients related to strength parameters. They
evaluation of composite structures with both these two aspects is an can be expressed as
important issue which hasn’t been explored in depth. Though some
1 1 1 1 1 1 1
inspiring progresses have been made in many works like Refs [30–35], F1 = − , F2 = − , F11 = , F22 = , F12 = − F11 F22 , F66
Xt Xc Yt Yc Xt Xc Yt Yc 2
the low efficiency still seems one of the most important difficulties
1
existing in current mixed reliability analysis [36]. = 2
S (2)
Univariate method is an effective means to evaluate structural re-
liability. A class of methods based on univariate method were proposed where Xt and Yt are the tension strengths in the 1 and 2 directions,
by Rahman [37–39] and demonstrated a good accuracy and efficiency respectively. Xc and Yc are the compression strengths in the 1 and 2
in structural reliability analysis. The application of univariate methods directions, respectively. S is the shear strength in the 1–2 plane. When
in the reliability analysis of composite structures with both random and the failure index is less than one, namely F . I . < 1, one can think that
interval parameters can obtain better results. the composite laminate is safe. On the contrary, laminate is in failure
In this paper, we aimed at proposing a new univariate method based when F . I . > 1.
on Legendre orthogonal polynomials, which can be applied to evaluate Based on the Tsai-Wu criterion, the performance function of l-th
the reliability of composite laminates with both random and interval lamina can be constructed and expressed as
variables. The remainder of this paper is organized in details as follows.
Gl = 1−F . I .l = 1−(F1 σ1, l + F2 σ2, l + F11 σ1,2 l + F22 σ2,2 l + F66 τ12,
2
Firstly, in section 2, we construct the performance of fiber-reinforced l

lamina utilizing Tsai-Wu failure criterion and state the failure criteria of + 2F12 σ1, l σ2, l ) (3)
composite laminates taking use of first-ply failure assumption. In sec-
Furthermore, the safety state of lamina shown in Fig. 2 then can be
tion 3, the structural performance function with random and interval
predicted by
variables is defined in the first place, and the univariate contribution of
random and interval variables are calculated by taking use of univariate
⎧Gl > 0, safe state
method and Legendre orthogonal polynomials. Combining univariate Gl = 0, limit state
contribution with MCS, the failure probability can be calculated. In ⎨
⎩Gl < 0, failure state (4)
section 4, the reliability analysis of two simple composite laminates
with different random and interval variables and a composite wingtip
structure are given to demonstrate the accuracy, efficiency, and ap- 2.2. The failure of composite laminate
plicability of the proposed method. Finally, some conclusions are drawn
in the last section. As composite laminates are composed of laminas, the laminates can
be regarded as systems in the procedure of reliability analysis. As a
2. Problem statement consequence, the failure of composite laminate can be evaluated by
first-ply failure (FPF) assumption and last-ply failure (LPF) assumption.
2.1. The performance function of lamina Some researches have been done based on the LPF such as mesoscale
approach [41] and probabilistic progressive failure model [42], but the
A composite laminate is a stack of layers of fiber-reinforced laminae, reliability evaluation using LPF is still a complex and time consuming
and the fiber-reinforced laminae are made of fibers and matrix, that are process because of the step-by-step failure process and degeneration of
of two different materials [13]. A typical sketch of composite laminate mechanical properties and may produce non-conservative results. Also,
and lamina is given in Fig. 1. the correlations of failure modes and failure of different plies, which are
Since the strength of the orthotropic material is dependent to the important and necessary, are hard to get in reliability evaluation

154
X. Li et al. Composite Structures 203 (2018) 153–163

Fig. 1. Sketch map of a typical composite laminate and fiber-reinforced lamina.

obeying normal distributions.

As normal random variables, the probability density functions of
random variables can be easily obtained with uncertainty information
and can be expressed as

X P ∼ N (μ X P , σ 2 P ) and X P = [x1P , x 2P , …, x mP ]T (5)

X

where N(•) denotes the normal random distribution,

μ X P = (μ x P , μ x P , ⋯, μ x P )T and σ X P = (σ x P , σ x P, ⋯, σ x P )T are the vectors
1 2 m 1 m m
of mean values and standard deviations of uncertain parameters, re-
spectively, m is the number of uncertain normal random variables ex-
isting in composites. On the contrary, when available information is
limited or insufficient so that probability density functions of uncertain
elastic parameters cannot be precisely obtained, uncertainties existing
in composites can be treated as independent interval variables, which
can be represented as

Y I = [ Y , Y ] = [y c −y r , y c + y r ], Y I = [y1I , y2I , …, ykI ] (6)

Fig. 2. The safety state of lamina. where Y and Y are the lower and upper bounds of the interval vector
for uncertain parameters, respectively. y c and y r are the corresponding
process. These factors depressed the application of LPF in the reliability nominal value and interval radius of uncertain vector, respectively. k is
evaluation of composite laminates. the number of uncertain interval parameters.
The FPF holds the opinion that the composite laminate fails when
anyone of plies failed in composite laminate. Based on FPF, many re- 3.2. The univariate method for performance function approximation
searches on reliability evaluation have been done. Lin and Kam [13]
developed a method for the reliability evaluation of composite laminate Assuming the contribution of the variable x iP or yjI to performance
in combination of FPF and phenomenological failure criterion. Haeri are independent, the performance function can be extended in terms of
and Fadaee [16] evaluated the reliability of composite laminates using input variables as
FPF. Many other researchers also studied the reliability evaluation of
m k
composite laminates using FPF such as Refs [43–48]. These applications ∼
have demonstrated the applicability and feasibility of FPF. The appli-
G ≈ G = g [(x , y c )] + ∑ gi (xiP ) + ∑ gj (yjI )
i=1 j=1 (7)
cation of FPF in composite laminates reliability evaluation is relatively
simple and can obtain conservative results. To a certain extent, the FPF where g [(x , y c )]
is the value of performance at mean value of random
can reflect the safety state of composite laminates. and central value of interval variables. gi (x iP ) is the term expressing the
Accordingly, the first-ply failure assumption will be utilized in this contribution of random variable x iP to performance function. gj (yjI ) is
paper to predict the failure of composites. the contribution of interval variable yjI to performance function. In
consideration of cooperative effects, the order of these contribution
terms should be high enough. The contribution of each random para-
3. The reliability evaluation utilizing univariate method meter to the total response gi (x iP ) can be expressed as

3.1. The performance function with mixed uncertainties gi (x iP ) = gi [(x (i) , y c )]−g [(x , y c )], i = 1, 2, …, m (8)

As the manufacturing process of lamina is a quite complex process, gj (yjI ) = gj [(x , y(cj) )]−g [(x , y c )], j = 1, 2, …, k (9)
the elastic properties, and strength parameters of each lamina in-
evitably have disturbances and dispersions. That is to say, the elastic where gi [(x (i) , y c )]
is the univariate function of random parameters x i ,
mechanical parameters (E11, E22, G12, v12) and the lamina strengths gj [(x , y(cj) )] is the univariate function of interval parameters yj ,
(Xt , Xc , Yt , Yc , S ) can be treated as uncertain variables. Taking good use g [(x , y c )] is the response of performance function at mean value and
of some quantification methods, uncertain variables with sufficient central value of uncertain variables. The vector (x (i) , y c ) and (x , y(cj) ) can
uncertain information can be quantified as random variables while be written as
those with insufficient uncertain information can be described as in- (x (i) , y c ) = [x1, x2 , …, x i , …, x n , y1c , y2c …,ymc ] (10)
terval variables.
Base on the existing researches and conclusions, the elastic variables (x , y(cj) ) = [x1, x2 , …, x n , y1c , y2c …,yj , …ykc ] (11)
and strength variables can be quantified as normal distribution, Weibull
distribution, or lognormal distribution based on goodness-of-fit test Then the total contribution of random parameters and interval
result. For convenience, random variables are treated as variables parameters to performance function are

155
X. Li et al. Composite Structures 203 (2018) 153–163

x (i) = [μ x1 , μ x2 , …, ∼
m m
x iI , …, μ xm ], y = [y1c , y2c …,ykc ] (26)
GP = ∑ gi (x iP ) = ∑ gi [(x (i) , y c )]−m ∗g [(x , y c )]
i=1 i=1 (12)
and
k k
x = [μ x1 , μ x2 , …, μ xm ], y(j) = [y1c , y2c , …, yjI , …, ykc ] (27)
GI = ∑ gj (yjI ) = ∑ gj [(x , y(cj) )]−k ∗g [(x , y c )]
j=1 j=1 (13)
As stated in Eq. (22), in order to utilize the Legendre orthogonal
Consequently, the response of performance function G can be ap- polynomial series, the parameters needed to be transformed into stan-
proximately expressed as dard interval. Then the detonation as follows will be used to express
m k univariate standard interval.
∼
G ≈ G = g (x , y c ) + GP + GI = ∑ gi [(x (i), y c )] + ∑ gj [(x , y(cj) )] e (i) = [0, 0,
 … ,  …
ei,0 , 0…,0 ] ≡ [0,
 0,
… ,…
z,0 , 0…,0 ]
(28)
⏟ ⏟
i=1 j=1 m k m k

−(k + m−1) ∗g [(x , y c )] (14) e (j) = [0,

0,
… , 0 , 0, …, ej…,0 ] ≡ [0,
 m  k 0,
… ,0 , 
 m
0,
…
, z
…
,0 ]
k (29)
Once the univariate functions gi [(x (i) , y c )] and gj [(x , y(cj) )] are ob-
tained, the value of performance function can be easily calculated. As where −1 ⩽ z ⩽ 1. And the intervals in Eq. (26) and (27) can be ex-
the high order univariate function is needed for univariate contribution pressed as:
stated upwards, the Legendre orthogonal polynomial series are adopted ∼
x iI = μ xi + fσxi·z
to approximate the univariate function of random variables and interval
variables. Assuming the mean value and standard deviation of random yjI = yjc + yjR ·z (30)
variable x iP are μ xi and σxi , the sample points can be transformed into
interval variable The zero roots of Legendre polynomial of order r can be calculated
for calculating coefficients of the approximate polynomial in the se-
∼
x iI = [μ xi −fσxi, μ xi + fσxi] (15) quence as follows:
When f is large enough, the approximated univariate function of x iP z r = [z1, z2, …, z r ] (31)
in the interval ∼ x iI can be utilized in structural reliability analysis. Then
the problem turned into the univariate contribution function of interval The corresponding input matrix can be calculated by
∼
x iP and yjI , which can be denoted as r
⎧ x (i) = [x (1i) , x (2i) , …, x (ri) ]
∼
x iI = [∼
x iL , ∼
x iU ] = [μ xi −fσxi, μ xi + fσxi], i = 1, 2, …, m (16) ⎨ y(rj) = [y(1j) , y(2j) , …, y(rj) ]
⎩ (32)
and
where
yjI = [yjL , yU
j ], j = 1, 2, …, k (17) l c c l c T
⎧ x (i) = [x1 , x 2 , …, x i , …, x m] , l = 1, 2, …, r
where the superscripts L and U represent the lower and upper bounds ⎪ l
y = [y1c , y2c , …, yjl , …, ykc ]T , l = 1, 2, …, r
of the interval variable, respectively. The lower bounds and upper ⎨ (j)
⎪ x il = μ xi + fσxi·z l , yjl = yjc + yjR ·z l
bounds can be expressed in the following vector form as ⎩ (33)
∼
x L = [∼
x 1L , ∼
x 2L , …, ∼
x mL], y L = [y1L , y2L , …, ykL ] Based on Eq. (33), we can get the equations as
(18)
and P l c
⎧ gz (l, i) = g [(x (i) , y )], t ⩽ m
∼ , i = 1, 2, …, m , j = 1, 2, …, k , l
x U = [∼
x 1U , ∼
x 2U , …, ∼
x mU ], yU = [y1U , y2U , …, ykU ] (19)
I l
⎨ gz (l, j ) = g [(x , y(j) )], t > m
⎩
The intervals of random and interval variables can be expressed by = 1, 3, …, r (34)
radius, median value and standard interval e = [−1, 1] as
where g [(x (li) , y c )] and g [(x , y(lj) )] are actual response with input vector
x iI = x ic + x iR ei = μ xi + fσxi ei , i = 1, 2, …, m (20) given in Eq. (33).
Assuming C [−1, 1] is a set of all continuous functions within the
yjI = yjc + yjR ej , j = 1, 2, …, k (21) interval [−1, 1] and the set LN = Span{L1 (z ), L2 (z ), …, LN (z )} is the
subspace included in it, the approximated functions of i-th random
The Legendre orthogonal polynomial series for approximation of variable ∼
gi (z ) and j-th interval variable ∼ gj (z ) in LN can be expressed as
univariate contributions can be expressed as:
N
1 dn ∼
gi (z ) = ∑ cni Ln (z ), i = 1, 2, …, m
Ln (z ) = [(z 2−1)n],
2 k! dz n
n (22) n=0 (35)

where n is a non-negative integer and {Ln (z )} is a series of orthogonal N

∼
gj (z ) = ∑ cnj Ln (z ), j = 1, 2, …, k
polynomials within the interval [−1, 1]. The recurrence relation of
n=0 (36)
Legendre orthogonal polynomials can be written as
L1 (z ) = 1 (23) where the coefficients cni and cnj can be calculated by Gauss quadrature
formula and expressed as
L2 (z ) = z (24) r
(u, Ln ) 2n + 1 1 2n + 1
⋮
cni =
(Ln , Ln )
=
2
∫−1 Ln (z ) g (z ) dz = 2
∑ Al Ln (z l ) gzP (l, i)
l=1
2n + 1 n (37)
Ln + 1 (z ) = zLn (z )− Ln − 1 (z ), n = 1, 2, …
n+1 n+1 (25) r
(u, Ln ) 2n + 1 1 2n + 1
In order to calculate the univariate response of i-th random variable cnj =
(Ln , Ln )
=
2
∫−1 Ln (z ) g (z ) dz = 2
∑ Al Ln (z l ) gzI (l, j )
l=1
or j-th interval variable, the uncertain variables except i-th random or j-
th interval should be fixed at their mean value or median value and (38)
expressed as where i = 1, 2, …, m , j = 1, 2, …, k and

156
X. Li et al. Composite Structures 203 (2018) 153–163

Fig. 3. Flow chart of mixed reliability analysis based on the proposed method.

z jroot = [z1j , z2j , …, zNj − 1], j = 1, 2, …, k

∫−1 (z−Lz lr)(Lz′)(z l) dz = 2
1
Al = , l = 1, 2, …, r (44)
l (1−z l2)[Ll′ (z l )]2 (39)
It is known that the roots should be a real number within interval
It can be known that the truncation error for calculation of ap- [−1, 1]. Then, the roots with imaginary part in Eq. (44) can be rewritten
proximate polynomial can be expressed as as
F (2r ) (η) 1 F (2r ) (η) 22r (r!) 4 2 znj = 2, Im(znj ) ≠ 0, n = 1, 2, …, N −1 (45)
R=
ar2 (2r )!
∫−1 Lr2 (x ) dx = (2r )! [(2r ) !]2 2r + 1
, η ∈ (−1, 1)
(40)
where Im(•) is the imaginary part of the variable. Furthermore, the
Then the approximate function of random or interval variables can imaginary roots and roots out of standard interval can be rewritten as
be rewritten as j j
⎧− 1, Im(z n ) = 0and|Re(zn)|gt;1
znj = , n = 1, 2, …, N −1
∼
N
⎡ (2n + 1) gzP (l, i) Ln (z l ) ⎤
r
⎨ z nj , Im(z nj ) = 0and|Re(znj)| ⩽ 1
gi (z ) = ∑ ∑ Ln (z ), x ∈ [−1, 1] ⎩ (46)
⎢ (1−z l2)[L′r (z l )]2 ⎥
n=0 ⎣ l=1 ⎦ (41)
where Re(•)is the real part of the variable. Then we can obtain the
minimum and maximum points of the j-th approximated polynomial by
∼
N
⎡ (2n + 1) gzI (l, j ) Ln (z l ) ⎤
r
gj (z ) = ∑ ∑ ⎢ Ln (z ), x ∈ [−1, 1] ∼
gj (z jmin ) = min{∼
gj (z )}∼
gj (z jmax ) = max{∼
n=0 ⎣ l=1
(1−z l2)[L′r (z l )]2 ⎥ (42)
gj (z )}, j = 1, 2, …, k
(47)
⎦ z ∈ zj z ∈ zj

Taking use of Eq. (41), it is easy to obtain the upper bound and where
lower bound of contribution of interval variables by seeking the roots of
z j = [z jroot , −1, 1] (48)
derived approximate function of interval variables
∼ (z )
dg The minimum and maximum univariate response of performance
j
= 0, j = 1, 2, …, k function with j-th interval variable can be calculated by
dz (43)
g jmin [(x , y(cj) )] ≈ ∼
gj (z jmin ) (49)
And the roots of Eq. (43) can be written as

157
X. Li et al. Composite Structures 203 (2018) 153–163

g jmax [(x , y(cj) )] ≈ ∼

gj (z jmax ) (50) be applied with the normal distribution variable by Box-Cox transfor-
mation.
Then the lower bound and upper bound of total contribution to In the reported univariate methods, the reliability evaluation is
performance function can be calculated by performed by the following steps: (1) The responses of each variables
k are obtained by Lagrange interpolation using univariate actual response
GImin ≈ ∑ ∼
gj (z jmin )−k ∗g (x , y c ) of sample points calculated by fixing other variables at a certain value.
j=1 (51) (2) The univariate contribution of random variables can be calculated
k by subtracting response of the certain value from univariate response.
GImax ≈ ∑ ∼
gj (z jmax )−k ∗g (x , y c ) (3) The approximated value of performance function then can be cal-
j=1 (52) culated by summing up univariate contributions of all random variables
and response of the certain value. (4) The failure probability can be
obtained by MCS. It can be known that the reported univariate method
3.3. Failure probability estimation may not applicable to the reliability evaluation with random and in-
terval variables.
Based on the interval of interval variable contribution and the Different from reported univariate methods, the proposed uni-
univariate performance function of random variables, the lower and variate method in the paper can give the worst or best combination of
upper bounds of every realization of l-th X = [x1l , x 2l , …, x ml] can be ex-
interval variables with explicit approximate function in standard in-
pressed as
terval, which means that the upper bound and lower bound of failure
m
∼L ∼ probability can be calculated, respectively.
Gl = ∑ gP {[x (li) −μ xi ]/(fμ xi ), 0} + GImin−m ∗g (x , y c )
i=1 (53)
4. Numerical examples
m
∼U ∼
Gl = ∑ gP {[x (li) −μ xi ]/(fμ xi ), 0} + GImax−m ∗g (x , y c ) 4.1. A T300/QY8911 composite plate with random and interval elastic
i=1 (54)
parameters
Therefore, it is easy for us to obtain the interval of failure prob-
ability by applying MCS method to the univariate function of random Consider a T3000/QY891 composite laminated plate subjected to
variables, which can be given by in-plane loads Nx = 160kN and Ny = 75kN . Its stacking sequence is
Ns [0∘, +45∘ , −45∘ , 90∘] as shown in Fig. 4. The strength parameters of
1 ∼U composite laminates are assumed to be deterministic and
PfL = ∑ Neg[Gl < 0]
Ns l=1 (55) Xt = 1500MPa , Xc = 1200MPa , Yt = 50MPa , Yc = 250MPa and
S = 70MPa . The elastic properties are given in ref [20] based on ex-
and
perimental data and listed in Table 1.
Ns
1 ∼L Based on the first-ply failure assumption, the proposed method (PM)
PUf = ∑ Neg[Gl < 0]
Ns (56) will be utilized to evaluate the lower bound (LB) and upper bound (UB)
l=1
of failure probability of the composite laminate with both random and
where Neg is a deciding function of the safe or fail state such that interval variables. In this example, the order of Legendre orthogonal
∼L ∼U
Neg=1, if Gl < 0 or Gl < 0 otherwise zero. Fig. 3 gives a brief proce- polynomials is 4, whose polynomials can be obtained by Eqs. (23)–(25).
dure of the composite structure mixed reliability with random and in- Base on the Tsai-Wu failure criterion, the performance function can be
terval variables. constructed as
The distributions of random variables mentioned previously are
G = 1−F . I . ≈g (x , y c ) + GI (X P , y c ) + GP (x , Y I ) (60)
assumed as normal distributions for convenience. For other distribu-
tions such as log-normal distribution and Weibull distribution, the where XP I
= [E11, E22, G12] and Y = [v12]. The interval of failure prob-
random variables can be transformed to normal distributions by Box- ability based on the proposed method is given in Table 2. Inspired by
Cox power transformation [49]. Ref [34], the reference failure probabilities are also given in Table 2
Box-Cox power transformation is a family of power transformations which were obtained by
on the necessarily positive response variable X given by L sub
⎧ Pf = min(P f )
(X λ −1)/ λ, λ≠0
X (λ) = ⎧ ⎨ PUf = max(P fsub)
⎩ (61)
⎨
⎩ lnX , λ = 0 (57)
where P fsub is the failure probability obtained by MCS with 50
where transformation parameter λ can be determined by the method of
maximum likelihood. As X (λ) ∼ N (μ ,̂ σ 2̂ ) , for a value of λ , the max-
imum log likelihood is, except for a constant,
m
1 1
L max = − lnσ 2̂ + lnJ (λ, X ) = − lnσ 2̂ + (λ−1) ∑ lnXi
2 2 i=1 (58)
where
m m
1 1
σ 2̂ = ∑ (Xi(λ) −μ 2̂ ), μ ̂ = ∑ Xi(λ)
m i=1
m i=1 (59)
Substituting Eq. (59) into Eq. (58), the L max can be obtained. With
several calculations of L max with different λ , L max against λ can be
plotted. And the optimal λ∗ maximizes L max then obtained. With λ∗, X
can be transformed to a normal distribution variable. Other transfor-
mation methods such as Johnson transformation [50] can also be ap-
plied to the normal transform. Consequently, the proposed method can Fig. 4. The composite plate subjected to in-plane tensile load.

158
X. Li et al. Composite Structures 203 (2018) 153–163

Table 1 Table 4
Uncertainty information of elastic parameters for T300/QY8911. Uncertainty information of elastic and strength variables.
Type Normal Interval Type Normal Interval

Variables E11 (GPa) E22 (GPa) G12 (GPa) v12 Variable E11 (GPa) E22 (GPa) G12 (GPa) v12 Xt (MPa) Xc (MPa)

Mean/Central value 131.30 9.22 5.13 0.3285 Mean/Central 131.30 9.22 5.13 0.368 1400 1300
Std./Radius 2.67 0.25 0.14 0.0395 value
C.V/Radius 0.05 0.05 0.05 0.05 140 130

Table 2
Failure probability interval by the proposed method for composite plate. Table 5
Failure probability interval by the developed method for composite plate.
PM MCS Errors
PM MCS Error
Sample points 36 50 × 10 6
–
LB of Pf 48.46% 47.98% 0.99% Sample point 54 50 × 50 × 10 6
–
UB of Pf 61.22% 61.00% 0.37% LB of Pf 68.83% 68.3% 0.776%
UB of Pf 70.04% 69.79% 0.358%

Table 3
Failure probability with different Fx utilizing proposed method and MCS. Table 6
Failure probability with different Fx utilizing proposed method and MCS.
Fx (kN) LB of Pf (%) UB of Pf (%)
Fx (kN) LB of Pf (%) UB of Pf (%)
PM MCS Error PM MCS Error
PM MCS Error PM MCS Error
152.5 4.10 4.16 1.32 8.58 8.72 1.64
155 12.36 12.34 0.16 21.20 21.18 0.08 230 15.69 15.53 1.03 17.54 17.17 2.15
157.5 27.78 27.58 0.73 40.07 39.81 0.64 240 41.70 40.59 2.73 42.23 41.27 2.33
160 48.46 47.98 0.99 61.21 60.99 0.37 250 68.83 68.3 0.776 70.04 69.79 0.358
162.5 68.99 68.76 0.34 78.93 78.69 0.30 260 88.85 89.06 0.236 87.09 87.51 0.480
270 95.86 96.82 0.992 96.93 98.20 1.29

subintervals of interval variable.

Comparing the results obtained from the proposed method and is carried out and results are presented in Table 3 together with results
subinterval based MCS in Table 2, it can be seen that the proposed obtained from MCS. It can be seen that proposed method has a high
method in the paper can evaluate the reliability of composite laminates accuracy comparing with MCS. In Fig. 5, the failure probability is il-
accurately and efficiently with mixed uncertainties. The relative errors lustrated versus the x-axis load Fx . The tendency curves of the proposed
of lower bound and upper bound of failure probability are only 0.99% method are compared well with that of MCS.
and 0.37% respectively, which means that the proposed method can
lead to results with high accuracy in the reliability evaluation. Besides, 4.2. A composite plate with random elastic parameters and interval strength
the composite laminate design will be more conservative as both the variables
lower bound and upper bound are larger than those from subinterval
MCS. The times of strength analysis is only 36, which is far less than the In example 1, the proposed method was utilized to evaluate the
number of sub-MCS method 50 × 106. In other words, the proposed failure probability of composite laminate with random and interval
method can be an important means for mixed reliability evaluation of elastic properties. In this example, we will analyze the failure prob-
composite laminates with random and interval variables in engineering ability of composite laminate as shown in Fig. 4 with random elastic
application. variables and interval x-axial strength given in Table 4. The loads acting
Also, the reliability of the composite plate with different x-axis load on the composite laminate are Nx = 250kN and Ny = 70kN . Other
strength variables are Yt = 50MPa , Yc = 250MPa and S = 70MPa .

Fig. 5. LB and UB of failure probability of different x-axis load.

159
X. Li et al. Composite Structures 203 (2018) 153–163

Fig. 6. LB and UB of failure probability of different x-axis load.

Fig. 7. Composite wingtip of aerospace plane.

Table 7 G = 1−F . I . ≈g (x , y c ) + GI (X P , y c ) + GP (x , Y I ) (62)

The stacking sequence of all components.
where X P = [E11, E22, G12, v12] and Y I = [Xt , Xc ]. By virtue of the pro-
Components Stacking
posed univariate method, the univariate performance of uncertain
Wing beam [0°, ± 45°, 90°, ± 45°]s variables is constructed and the interval of failure probability is cal-
Leading edge [0°, 0°, ± 45°, 90°, 90°, ± 45°]s culated, which is listed in Table 5. To calculate reference intervals of
Wing rib [0°, ± 45°, 90°]s
failure probability using MCS, the intervals of Xt and Xc are divided
Upper/Lower skin [0°, ± 45°, 90°, 0°, ± 45°, 90°]s
Trailing edge [0°, ± 45°, 90°, 0°, 90°, ± 45°, 0°] into 50 subintervals. Failure probability of each combinations of sub-
intervals are simulated with MCS. The lower bound and upper bound of
failure probability then can be obtained by Eq. (61) and given in
Similarly, Tsai-Wu failure criterion and first-ply failure assumption Table 5.
are adopted to construct the performance function, which could be Comparing with interval of reliability obtained from MCS, the
written as proposed univariate method can precisely and accurately predict the
lower bound and upper bound of composite laminates with random

Fig. 8. The finite element model of wingtip structure and Tsai-Wu failure index.

160
X. Li et al. Composite Structures 203 (2018) 153–163

Fig. 9. The mixed reliability calculation process utilizing proposed method.

Fig. 10. The uncertain properties of contribution of random variables.

Table 8 other words, the proposed method can be a credible implement for
Failure probability interval by the developed method and MCS. safety evaluation of composite laminates with random elastic variables
PM MCS Error
and interval strength variables.
Similarly, the reliability of composite plate with different x-axis
Sample point 36 50 × 105 – load Fx are calculated by the proposed method and MCS and the results
LB of Pf 4.79% 4.71% 1.06% are given in Table 6 and Fig. 6. It can be seen that the proposed method
UB of Pf 11.26% 11.19% 0.63%
has a high accuracy comparing with that of MCS with random elastic
parameters and interval strength parameters. Fig. 6 demonstrates that
elastic variables and interval strength variables. With only 54 times the results obtained from the proposed method can predict the tendency
strength analysis for univariate performance construction, the proposed of failure probability precisely.
univariate method has high precision of lower bound and upper bound,
whose relative errors are only 0.818% and 0.645%, respectively. In

161
X. Li et al. Composite Structures 203 (2018) 153–163

4.3. Engineering application proposed method can be a viable alternative to evaluate safety of
composite laminates with both random and interval variables.
To demonstrate the feasibility and efficiency of the proposed
methods in the engineering problems, the mixed reliability of a com- Acknowledgement
posite wingtip as shown in Fig. 7 with a concentrated load 65600N
acting on the tip will be analyzed utilizing the proposed method. The work is supported by the Defence Industrial Technology
The composite wingtip structure can be divided into six components Development Program (No. JCKY2016601B001) for the financial sup-
as shown in Fig. 7 by different stacking sequences of laminas given in port.
Table 7. Composite laminates applied to manufacture the wingtip
structure is T300/QY8911, whose uncertain properties of elastic para- References
meters and deterministic strength parameters are given in Tables 1 and
2, respectively. [1] Richards J, Suleman A. 3.10 composite aerostructures for unmanned aircraft. In:
The response calculation of the composite wingtip structure is car- Beaumont PWR, Zweben CH, editors. Comprehensive composite materials II.
Oxford: Elsevier; 2018. p. 261–87.
ried out with MATLAB and MSC. Patran/Nastran (see Fig. 8). A cal- [2] Choi J. Architecture of a knowledge based engineering system for weight and cost
culation process is constructed as shown in Fig. 9, the calculation estimation for a composite airplane structures. Expert Syst Appl
process also can be applied to other structure using corresponding finite 2009;36(8):10,828–36.
[3] Brooks R, Shanmuga Ramanan SM, Arun S. Composites in automotive applications:
element model and PCL codes. design. Reference module in materials science and materials engineering Elsevier;
Evaluating the mixed reliability of the composite wingtip structure 2017.
utilizing the proposed method, the probabilistic and interval properties [4] Araújo HAM, Machado JJM, Marques EAS, da Silva LFM. Dynamic behaviour of
composite adhesive joints for the automotive industry. Compos Struct
of contribution of random variables E11, E22 and G12 are given in Fig. 10.
2017;171(Supplement C):549–61.
It can be seen from Fig. 10 that the contribution bounds of random [5] Mathijsen D. Now is the time to make the change from metal to composites in naval
variables are both approximately normal distribution, the minimum shipbuilding. Reinf Plast 2016;60(5):289–93.
[6] Hassoon OH, Tarfaoui M, El Moumen A. Progressive damage modeling in laminate
contribution and mean value of minimum contribution are of E11 are
composites under slamming impact water for naval applications. Compos Struct
both the smallest among random variables. As small contribution may 2017;167(Supplement C):178–90.
produce small value of performance function, Fig. 10 also indicates that [7] Chiachio M, Chiachio J, Rus G. Reliability in composites–a selective review and
E11 has the largest impact on the upper bound of failure probability of survey of current development. Compos B Eng 2012;43(3):902–13.
[8] Di Sciuva M, Lomario D. A comparison between Monte Carlo and FORMs in cal-
the composite wingtip structure, followed by E22 and G12 . While E22 has culating the reliability of a composite structure. Compos Struct 2003;59(1):155–62.
the largest impact on the lower bound of failure probability, followed [9] Sriramula S, Chryssanthopoulos MK. Quantification of uncertainty modelling in
by E11 and G12 . stochastic analysis of FRP composites. Compos A Appl Sci Manuf
2009;40(11):1673–84.
The interval of failure probability of composite wingtip structure [10] Sun C, Yamada S. Strength distribution of a unidirectional fiber composite. J
obtained by the proposed method and MCS are given in Table 8. Compos Mater 1978;12(2):169–76.
Comparing with reference result obtained from MCS, it can be found [11] Yang L. Reliability of composite laminates. Mech Struct Mach 1988;16(4):523–36.
[12] Engelstad S, Reddy J. Probabilistic nonlinear finite element analysis of composite
that the proposed method has a good accuracy. Avoiding of numerous structures. AIAA J 1993;31(2).
complex computations, the total number of finite element method [13] Lin S, Kam T. Probabilistic failure analysis of transversely loaded laminated com-
calculation is only 36, which is acceptable for the application in solu- posite plates using first-order second moment method. J Eng Mech
2000;126(8):812–20.
tion of practical engineering problems. Comparing with traditional [14] Philippidis TP, Lekou DJ. Probabilistic failure prediction for FRP composites.
MCS, which need a huge number of finite element method calculations, Compos Sci Technol 1998;58(12):1973–82.
the proposed method shows a high efficiency and accuracy in reliability [15] Kogiso N, Murotsu Y. Relialility analysis of laminated composite plate with initial
imperfection. Jpn Soc Mech Eng, Trans A 2000;66(648):43–50.
evaluation of implicit performance function.
[16] Haeri A, Fadaee MJ. Efficient reliability analysis of laminated composites using
advanced Kriging surrogate model. Compos Struct 2016;149(Supplement C):26–32.
5. Conclusions [17] Li J, Tian X, Han Z, Narita Y. Stochastic thermal buckling analysis of laminated
plates using perturbation technique. Compos Struct 2016;139(Supplement C):1–12.
[18] Wang L, Wang X, Xia Y. Hybrid reliability analysis of structures with multi-source
Owing to complicated manufacturing processes and inherent dis- uncertainties. Acta Mech 2014;225(2):413–30.
persions of constituents, mechanical properties of composite laminates [19] Wang L, Xiong C, Wang X, Li Y, Xu M. Hybrid time-variant reliability estimation for
are always uncertain in nature. According to the amount of available active control structures under aleatory and epistemic uncertainties. J Sound Vib
2018;419:469–92.
uncertainty information, uncertainty variables can be quantified as [20] Chen X, Qiu Z. A novel uncertainty analysis method for composite structures with
random or interval variables. To evaluate the safety of composite la- mixed uncertainties including random and interval variables. Compos Struct
minates with both random and interval variables, this paper is aimed at 2018;184(Supplement C):400–10.
[21] Polymer Matrix Composites Guidelines for Characterisation of Structural Materials
developing a novel reliability assessment method based on the uni- MIL-HDBK-17-1F, Composite Materials Handbook MIL-HDBK-17, vol. 1. West
variate thought and interval analysis technique. Constructing the uni- Conshohocken, PA, ASTM, 2002.
variate performance of all variables at the same time, the interval of [22] Wang X, Wang L. Uncertainty quantification and propagation analysis of structures
based on measurement data. Math Comput Modell 2011;54(11–12):2725–35.
failure probability can be evaluated based on the contribution bounds [23] ZhiPing Q, Lei W. The need for introduction of non-probabilistic interval concep-
of interval variables and MCS. The numerical examples are carried out tions into structural analysis and design. Sci China Phys Mech Astron
to demonstrate the feasibility and validity of the proposed univariate 2016;59(11):114,632.
[24] Lv Z, Liu H. Nonlinear bending response of functionally graded nanobeams with
method. The results show that the proposed method can be efficiently
material uncertainties. Int J Mech Sci 2017;134:123–35.
and accurately applied to evaluate the reliability of composite lami- [25] Qiu Z, Lv Z. The vertex solution theorem and its coupled framework for static
nates with both random and interval variables. It can also be applied in analysis of structures with interval parameters. Int J Numer Meth Eng
the uncertainty optimization of composite laminates in face of the de- 2017;112(7):711–36.
[26] Lv Z, Liu H. Uncertainty modeling for vibration and buckling behaviors of func-
mands of lightweight. tionally graded nanobeams in thermal environment. Compos Struct
Last but not least, the major contributions and highlights may lie in 2018;184:1165–76.
the construction of a novel univariate method for reliability evaluation [27] Lv Z, Qiu Z, Zhu J, Zhu B, Yang W. Nonlinear free vibration analysis of defective FG
nanobeams embedded in elastic medium. Compos Struct 2018.
of composite laminates with random and interval variables. The per- [28] Wang L, Wang X, Su H, Lin G. Reliability estimation of fatigue crack growth pre-
formance function is constructed utilizing univariate method and diction via limited measured data. Int J Mech Sci 2017;121:44–57.
Legendre orthogonal polynomials. And the interval of failure prob- [29] Du X, Sudjianto A, Huang B. Reliability-based design with the mixture of random
and interval variables. J Mech Des 2005;127(6):1068–76.
ability then can be calculated by MCS. Results of examples demon- [30] Hui L, Yu D. Stability analysis and improvement of uncertain disk brake systems
strated that the proposed method is accurate and efficient. The

162
X. Li et al. Composite Structures 203 (2018) 153–163

with random and interval parameters for squeal reduction. J Vib Acoust [40] Orifici AC, Herszberg I, Thomson RS. Review of methodologies for composite ma-
2015;137(5):101–8. terial modelling incorporating failure. Compos Struct 2008;86(1):194–210.
[31] Jiang C, Han X, Liu W, Liu J, Zhang Z. A hybrid reliability approach based on [41] Gurvich MR, Pipes RB. Probabilistic analysis of multi-step failure process of a la-
probability and interval for uncertain structures. J Mech Des 2012;134(3):031,001. minated composite in bending. Compos Sci Technol 1995;55(4):413–21.
[32] Jiang C, Li WX, Han X, Liu LX, Le PH. Structural reliability analysis based on [42] Mahadevan S, Liu X, Xiao Q. A probabilistic progressive failure model of composite
random distributions with interval parameters. Comput Struct laminates. J Reinf Plast Compos 1997;16(11):1020–38.
2011;89(23):2292–302. [43] Kam TY, Chang ES. Reliability formulation for composite laminates subjected to
[33] Guo J, Du X. Reliability sensitivity analysis with random and interval variables. Int first-ply failure. Compos Struct 1997;38(1):447–52.
J Numer Meth Eng 2010;78(13):1585–617. [44] Lin SC, Kam TY, Chu KH. Evaluation of buckling and first-ply failure probabilities of
[34] Du X, Interval reliability analysis. In: ASME 2007 International design engineering composite laminates. Int J Solids Struct 1998;35(13):1395–410.
technical conferences and computers and information in engineering conference, [45] Gurvich MR, Pipes RB. Reliability of composites in a random stress state. Compos
2007, 1103–1109. Sci Technol 1998;58(6):871–81.
[35] Shuxiang G, Zhenzhou L. Hybrid probabilistic and non-probabilistic model of [46] Jeong HK, Shenoi RA. Probabilistic strength analysis of rectangular FRP plates using
structural reliability. J Mech Strength 2002;24(4):476–524. Monte Carlo simulation. Comput Struct 2000;76(1):219–35.
[36] Han X, Jiang C, Liu L, Liu J, Long X. Response-surface-based structural reliability [47] Lopes PAM, Gomes HM, Awruch AM. Reliability analysis of laminated composite
analysis with random and interval mixed uncertainties. Sci China Technol Sci structures using finite elements and neural networks. Compos Struct
2014;57(7):1322–34. 2010;92(7):1603–13.
[37] Rahman S, Wei D. A univariate approximation at most probable point for higher- [48] Chen N-Z, Sun H-H, Guedes Soares C. Reliability analysis of a ship hull in composite
order reliability analysis. Int J Solids Struct 2006;43(9):2820–39. material. Compos Struct 2003;62(1):59–66.
[38] Wei D, Rahman S. A multi-point univariate decomposition method for structural [49] Box GEP, Cox DR. An analysis of transformations. J Roy Stat Soc
reliability analysis. Int J Press Vessels Pip 2010;87(5):220–9. 1964;26(2):211–52.
[39] Wei D, Rahman S. Structural reliability analysis by univariate decomposition and [50] Johnson NL. Systems of frequency curves generated by methods of translation.
numerical integration. Probab Eng Mech 2007;22(1):27–38. Biometrika 1949;36(Pt. 1–2):149–76.

163