Sie sind auf Seite 1von 14

Finite Elements in Analysis and Design 7071 (2013) 114

Contents lists available at SciVerse ScienceDirect

Finite Elements in Analysis and Design


journal homepage: www.elsevier.com/locate/finel

On the modeling and design of composite multilayered structures


using solid-shell nite element model
H. Naceur a,n, S. Shiri b, D. Coutellier a, J.L. Batoz c
a

Lab. LAMIH, UMR8201 CNRS, Universit Lille Nord de France, 59313 Valenciennes, France
Lab. I2M, UMR 5295 CNRS, Arts & Mtiers ParisTech, Universit Bordeaux I, 33607 Pessac, France
c
Lab. Roberval, UMR 6253 CNRS, Universit de Technologie de Compigne, 60205 Compig`ne, France
b

art ic l e i nf o

a b s t r a c t

Article history:
Received 19 July 2012
Received in revised form
31 December 2012
Accepted 20 February 2013
Available online 1 April 2013

In this investigation a coupling between a 3D solid-shell element for the analysis of multilayered
composite shell structures and a specic response surface method is proposed. The rst part of the paper
is dedicated to the nite element formulation of a developed composite 8-node solid-shell element
called SCH87, based only on translational degrees of freedom. The basis of the present nite element
formulation is the standard 8-node brick element with tri-linear shape functions. A particular attention is
given to alleviate shear, trapezoidal and thickness locking, without resorting to the classical plane-stress
assumption. Assumed natural strain method and enhanced assumed strain method are used to improve
the relative poor element behavior of a standard hexahedral displacement element. The anisotropic
material behavior of layered shells is modeled using a fully three dimensional elastic orthotropic material
law in each layer, including the thickness stress component. The second part of the paper will focus on an
adaptive response surface method for the structural optimization problem. The response surfaces are
built using moving least squares approximations and design of experiments by means of a specic
method called Diffuse Approximation.
Several numerical applications to composite multilayered shell structures are studied to show the
applicability and effectiveness of the proposed procedure. Good results of analysis and optimization
using the developed SCH87 solid-shell element have been obtained in comparison with reference
analytical solutions and with those obtained using the SC8R solid-shell nite element available in

ABAQUS code.
& 2013 Elsevier B.V. All rights reserved.

Keywords:
Composite structures
Solid-shell
Optimization
Design of experiment

1. Introduction
In the modeling of shell structures various problems may
appear especially when shell nite elements are used in combination with solid elements. Therefore special connection elements
are necessary to link shell elements with solids having different
degrees of freedom. It turns obvious to develop general-purpose
brick elements, which are able to deal with any type of structures
(solid, shell, and/or their combination).
3D-continuum elements are a variety of Finite Element (FE)
models halfway between solid elements and thin shells. They have
the same freedom conguration of solid elements but account for
shell-like behavior in the thickness direction. They are very
attractive for modeling shell-like regions of a 3D structure without
the need of special elements to connect solid elements to shell
nodes. 3D-continuum elements also called solid-shell elements

Corresponding author. Tel.: 33 3 2751 1412; fax: 33 0 3 2751 1316.


E-mail address: hakim.naceur@univ-valenciennes.fr (H. Naceur).

0168-874X/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.nel.2013.02.004

have many benets compared to the degenerated shell elements,


mainly because of the simplicity of their kinematics, their ability
in modeling complex structures generally composed of massive
and thin-walled regions and also special rotations treatment in
geometric nonlinear analysis may be avoided.
The 3D-continuum FE concept has attracted many researchers
since the late 1990s. Domissy et al. [15] and Sze et al. [37] are
probably the rst authors to investigate a solid-shell approach
for the analysis of plates and shells. In 1998, Hauptmann and
Schweizerhof [20] propose an extension to the original solid-shell
theory called double-node models where the position of an arbitrary point in the element is assumed to depend either linearly or
quadratically, on the thickness coordinate. In 1998, Cho et al. [12]
proposed a solid-shell element model based on the assumed
strain formulation for buckling and postbuckling analysis of shell
structures.
During the last decade, solid-shell FE models for thin shell
structures have attracted considerable attention. Numerous models
have been proposed in the literature and can be found in Sze and
Yao [38], Vu-Quoc and Tan [40], Areias et al. [3], Harnau and

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Schweizerhof [19], Alves de Sousa et al. [2], Hannachi et al. [18], Quy
and Matzenmiller [32], Nguyen et al. [29], Shiri et al. [35] and more
recently the work of Schwarze and Reese [34], Moreira et al. [25].
Solid-shell element properties make them appropriate also for the
modeling of laminated structures as can be seen in some representative references [3,21,25,33,40], to name just a few. In all of
these works it has been shown that it is possible to use a solid-shell
nite element models which possess no rotational degrees of
freedom and obtain good results for the resolution of various
structural problems. However, development of these elements is
not straightforward, transverse shear, trapezoidal and thickness
locking phenomena must be alleviated and therefore special treatments have to be included to suppress the numerical locking effects
become crucial.
Transverse shear locking is characterized by an overestimation
of stiffness associated to transverse shear strain energy [2]. The
most common methods that have been used to solve the shear
locking effect are the Selective Reduced Integration (SRI) scheme
[2] and also the Assumed Natural Strain (ANS) which has been
applied rstly on shells by Dvorkin and Bathe [17] and as can be
seen in the works of [3,11,15,35] for a fully integrated solid-shell
element.
Trapezoidal locking is only found in structures where the
directors of the element edges are not perpendicular to the midplane. One method to resolve this pathology is by using the ANS
interpolation of the transverse strain in thickness direction as
performed in [8,9,37,38], where it was proposed to avoid articial
thickness straining.
Thickness locking, caused by Poisson's ratio coupling to the
in-plane and transverse normal stress and normal strain responses
[14,15] can be overcome by plane-stress assumption, or more
generally by the Enhanced Assumed thickness Strain (EAS), where
a 7th parameter is added [10,21] and suppressed by condensation
technique. This locking pathology usually appears when a solidshell element with only translational degrees of freedom is to be
used in problems involving bending.
The 7th parameter as proposed by Bchter and Ramm [10] is an
extension of classical shear deformation theory. The interest of this
model is especially useful when full three-dimensional constitutive law is used which allows solving problems involving large
strains. In their work, Bchter and Ramm [10] describe the 7th
parameter model along with a FE formulation and they introduce
it on the element level by means of the hybrid-mixed formulation.
In the present solid-shell model, the 7th parameter is resulting
independently from the FE formulation, i.e. it can be understood as
semi-discretization of the solid-shell element through the thickness. Based on this technique, the obtained 7th parameter model
is considered simply as a two-dimensional continuous concept
with 7 degrees of freedom per node of the reference surface. The
7th parameter is then eliminated using the static condensation
procedure on the element level.
Another alternative, has been proposed recently by Professor
Brunet and his collaborators [4], concerning a solid-shell element
with 9 nodes: 8 are located at the element vertices and the 9th is
located at the centroid. The authors used classically a reduced
integration with one in-plane quadrature point with an assumed
shear-strain eld to avoid locking phenomena. The centroid node
is used as an extra parameter to enhance the displacement in the
thickness direction and thus obtain a linear normal strain, allowing the use of full three-dimensional constitutive strainstress
behavior.
During last few years, Response Surface Method (RSM) [24] has
gained more and more importance in the optimization of general
shell structures [30]. RSM has the advantage of replacing a
complex response model by an approximate one based on results
calculated at various points in the design space. The optimization

is then performed at a lower cost over such response surfaces.


Two important issues when applying RSM to a particular problem
concern the Design of Experiments (DOE) and construction of
accurate approximation functions [13,22] so that rapid convergence may be achieved.
In the present work, we exploit the RSM based on Diffuse
Approximation (DA) [7,26] and particularly the notion of pseudoderivative to design a specic optimization technique, custom
built for this regression model. This new method is an extension of
pattern search in two aspects: (1) accommodation of arbitrary
regular and irregular patterns; (2) design points eligible for
inclusion in any pattern instance belong to a predened set in
the design where experiments are authorized.
The outline of the paper is as follows. In Section 2 we present
the kinematics, the material law, and the variational equations of a
3D-shell formulation with convective coordinates with the associated FE description. A special technique, which is necessary for
integration through the layers in the 3D-case, is also given. In
Section 3 the optimization procedure dedicated for the design of
material parameters is given, a special attention is set to the built
of response surface model, we use to approximate the implicit
objective function. In the fourth section we discuss the applicability of the proposed procedure which consists in coupling
between RSM, DOE and solid-shell solution for the modeling of
several standard benchmarks which have been given by NAFEMS
[28]. The obtained results are compared with those obtain using

ABAQUS commercial software via its solid-shell element called


SC8R. Finally some conclusions are drawn.

2. FE formulation of the multilayered solid-shell element


2.1. Kinematics
To overcome the difculties of rotational degrees of freedom
(dof) in conventional shell elements, the shell kinematics of
deformation is described using the position vectors of a pair of
material points at the top and at the bottom of the shell midsurface (Fig. 1).
In this kinematic description, a straight transverse ber before
deformation remains straight after deformation. Such transverse
ber does not need necessarily to be normal to the shell midsurface before deformation, as well as after deformation. With
respect to nodal designation (Fig. 2), the coordinate vector X and
displacement vector u of the element, are [5,6,16]


4
1 1
X,, X 0 , X n , N i ,
Xi
Xi
1
2
2
i1
4

u,, u0 , un , N i ,
i1

1 1
u
ui
2 i
2

Fig. 1. An 8-node solid-shell element.


2

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Using Eqs. (2) and (3), the displacement gradient vector in the
curvilinear coordinate system can be expressed as
du L ds

12

with
L F T L

13

and du L d is the displacement gradient tensor in the parametric coordinate system. In order to calculate the strain tensor
directly in the curvilinear coordinate system, we have to dene a
tensor which is given by [6,15,41]
C F 1 Q

14

where
2 1
a  t 1
6 2
a t
C 6
4 1
0

Fig. 2. Position vectors and covariant basis.

a1  t 2
a2  t 2
0

a1  n

7
a2  n 7
5
a3  n

15

where Ni are the two-dimensional 4-node Lagrangian interpolation functions, X i , ui and X i , ui are respectively, the coordinate
and displacement vectors of the ith node on the bottom and top
shell surfaces respectively (Fig. 1).
Using Eq. (1), the gradient vector dX can be expressed in terms
of d fd, d, dg

E c 12L LT

dX F d

where components of E c are as follows:

and
F 1 a1 ; a2 ; a3 T

16

The covariant strain tensor can be calculated using


17

with F the covariant basis, relative to the point q, it is given by

E aT1  u, ,

2E aT1  u, aT2  u,

F a1 ; a2 ; a3 

E aT2
E aT3

 u, ,

2E aT1  u, aT3  u,

 u, ,

2E aT2  u, aT3  u,

where a1 and a2 (Fig. 2) are the covariant or natural basis vectors


dened by [6,15]

Therefore, we obtain a simple relationship between the curvilinear and covariant strain tensors

a1 X p, 12 V ,
a2 X p, 12 V ,
a3 V
1
2

The element volume dV of the hexahedron can be obtained


using
dV J d d d

with
J detF ja1  a2 a3 j

An orthonormal basis can be constructed for each point q and


dened by
Q t 1 ; t 2 ; n 

where t 1 and t 2 are unit vectors in the plane A for Cst. The
normal unit vector is dened by
n

a1  a2
ja1  a2 j

Several techniques may be used to dene the tangent unit


vectors t 1 , t 2 , following Batoz and Dhatt [6], one can dene
2
3
1 2
1
6 1 ny 1 nx ny nx 7
6
7
6
7
1
1 2
10
Q 6
7
nx ny
nx ny 7
6
4 1
5
1
nx
ny
nz
with n  k (see Fig. 2).
Finally the relationship between the curvilinear basis and the
global cartesian framework can be expressed by
dX Q ds

11

E C T Ec C

18

Now, by using the engineering notation, relation of Eq. (18)


may be rewritten in a simple form
E C~ E c
with
2

C 211
6
6 C 212
6
6 2C C
6
~
C 6 112 12
6 C 13
6
6 2C C
4 11 13
2C 12 C 13

19

C 221

C 11 C 21

C 222
2C 21 C 22

C 12 C 22
C 12 C 21 C 11 C 22

0
0

0
0

C 223

C 13 C 23

C 233

C 13 C 33

2C 21 C 23

C 23 C 11 C 21 C 13

C 11 C 33

2C 22 C 23

C 23 C 12 C 22 C 13

C 12 C 33

7
7
7
7
7
7
C 23 C 33 7
7
C 21 C 33 7
5
C 22 C 33
0
0

and E c fE E 2E E 2E 2E g , E fExx Eyy 2Exy Ezz 2Exz 2Eyz gT .


2.2. Material law in convected basis
The constitutive relation of laminated composites [39] can be
described using an orthotropic material law. For that purpose, we
express the components of the tangent elastic moduli tensor
relative to the ber reference axis fm1 ; m2 ; m3 g of lamina [23]
2
3
H 1111 H 1112
0
H 1113
0
0
6 1122
7
6H
H 2222
0
H 2223
0
0 7
6
7
6 0
1212
0
0
0 7
0
H
6
7
H 6 1133
20
7
2233
3333
6H
7
H
0
H
0
0
6
7
6 0
7
1313
0 5
0
0
0
H
4
0
0
0
0
0
H 2323
Each layer is constituted of an orthotropic elastic material
directions fm1 ; m2 ; m3 g (see Fig. 3). The non-zero coefcients of

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Fig. 4. Orthonormal basis for composite shell.

Fig. 3. Fiber-reinforced lamina orientation axes.

the behavior matrix are given by the following expressions:


H 1111 E1 123 32 =K,
H

2222

H 1122 E1 21 23 =K
H 1113 E1 31 21 32 =K

E2 113 31 =K,

H 2223 E2 32 12 31 =K,
H 1212 G12 ,
H

2323

G23 ,

H 3333 E3 112 21 =K

for i,j 1,2,3 and ij

H 1313 G13
ij Ej ji Ei

with K 112 21 13 31 23 32 12 23 31 13 21 32 , E1, E2 and


E3 are the elastic moduli in the principal material directions
fm1 ; m2 ; m3 g, ij and Gij are respectively the Poisson's ratio and
the shear modulus.
Since matrix is associated with the principal material directions, we need to transform it from the lamina coordinate axes
fm1 ; m2 ; m3 g to the global cartesian coordinate axes fX; Y; Zg. With
being the ber orientation angle relative to the global cartesian
system (Fig. 3), the relationship between the lamina coordinate
system and the global Cartesian system is given by
m1 cos X sin Y
m2 sin X cos Y
m3 Z

21

The nal constitutive tensor H~ can be expressed in the


convective coordinates as
H~ RT HR
with
2

r 1 2
6 1 2
6 r 1
6 2
6 1 1
6 2r 1 r 2
R6
6 r 1 2
6 3
6 1 1
6 2r r
4 2 3
2r 11 r 13

22

r 21 2

r 11 r 21

r 31 2

r 21 r 31

r 22 2
2r 21 r 22

r 12 r 22
1 2
r 1 r 2 r 21 r 12

r 32 2
2r 31 r 32

r 22 r 32
2 3
r 1 r 2 r 22 r 31

r 23 2

r 13 r 23
1 2
r 2 r 3 r 13 r 22
r 11 r 23 r 13 r 21

r 33 2

r 23 r 33
2 3
r 2 r 3 r 23 r 32
r 21 r 33 r 23 r 31

2r 22 r 23
2r 21 r 23

2r 32 r 33
2r 31 r 33

r 11 r 31

7
7
r 22 r 32
7
1 3
1 37
r 1 r 2 r 2 r 1 7
7
7
r 13 r 33
7
7
1 3
1 37
r 2 r 3 r 3 r 2 5
r 11 r 33 r 13 r 31
23

r ji

Q j mi .

and
The stresses can be evaluated in the curvilinear coordinate
system (Fig. 4) as
S H~ Z
where the stress tensor S fSxx Syy Sxy Szz Sxz Syz g

24
T

2.3. Variational formulation using solid-shell element


The principal of virtual work can be stated as
Wu,u W int W ext 0

25

u0 and u 0, u u on Su.
With Su is the shell contour where displacements are imposed.
The Principal of Virtual Work can be expressed in the curvilinear

Fig. 5. Composites modeling using solid-shell elements.

coordinate system as
Z
Z
Z
W W int W ext E T S dV uT f v dV uT f s dS
V

26

Sf

where f V and f s are volume and surface traction forces respectively. In order to deal with the several locking phenomena
separately, we need to split the expression of virtual internal work
by separating the membrane/bending, thickness and transverse
shear effects [6,15].
W int W mb W tr W sh

27

with

Z
W mb E Ts H~ 1 E s dV
Z V
W tr E Tz H~ 2 E z dV
ZV
W sh GTs H~ 3 Gs dV

28

and E s fExx Eyy 2Exy gT , E z fExx Eyy Ezz gT , Gs f2Exz 2Eyz gT .


H~ 1 , H~ 2 and H~ 3 are sub-matrices extracted from the global
material matrix H~ according to the different strain components.
In the context of modeling multilayered composite structures
using solid-shell elements, there exist two possibilities regarding
the numerical implementation:
 Case of one element per layer: This is the easiest way for
modeling the entire thickness of the structure using several
elements (1 element per layer) as indicated in Fig. 5. In this case
the numerical procedure of integration is straightforward and
does not require any efforts in the implementation compared to
the case of isotropic material modeling. The user has to provide
the following basic properties:
1. Declaration of n groups of different elements in the FE mesh,
these groups correspond to the n different material layers
constituting the laminate.
2. Provide the physical characteristics of each layer.
 Case of several layers per element: This second technique
consists in stacking the different material layers within the
same element (see Fig. 5). Each FE is a stack of several layers,

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

therefore stress calculation and numerical integration of the


stiffness matrix is carried out using the single-layer approach
identical to the one commonly used for the integration of
plasticity, by using either one or several Lobatto integration
points through the thickness for each layer.
In the present investigation, this second method has been
implemented because it is more general and more convenient
since it requires only one element in the thickness of the
structure (gain of CPU time in the case of industrial applications). On the other side the drawback the chosen element
kinematic will not be able to capture delamination effects.
In this case, the numerical implementation of the FE model
requires some modications of the stresses calculation and the
integration of the stiffness matrix. For each element the user
has to provide the following properties:
1. Declaration of n groups of materials, these n groups correspond
to the different layers constituting the multilayered structure.
2. Provide the physical characteristics for each material.
i
3. Provide for each layer i, the thickness hi (with ni 1 h h total
thickness of the structure).
For instance, the numerical implementation for the in-plane
membrane/exion effect is done in the following way:
!
Z Z
Z Z
W mb

z h=2

z h=2

E Ts S s

z hi =2

nl

dz dA

i1

Ai

z hi =2

E Ts S s

dz dA
29

where nl is the number of layers, hi represents the layer thickness


and h is the total thickness of the structure.
Eq. (29) implies that for each layer of material, all operators are
reported to the mid-plane of every layer i then the numerical
integration is performed using Lobatto integration scheme with
N points through the thickness direction. In the present model, 2
Lobatto integration points are used in the thickness direction for
each layer, since all applications involve only elastic material
behavior, for the case of material nonlinearities, at least 5 integration points will be necessary to achieve a correct stress integration.
A solid-shell element formulated using Eqs. (26)(28) with
standard integration based on 2  2  N Gauss schema will fail
because of numerous locking phenomena.
2.4. Remedies for shear locking
An effective method of resolving shear locking is the Assumed
natural Strain method in which the natural transverse shear
strains are sampled and then interpolated at some discrete
element points. The transverse shear strains 2E and 2E are
calculated according to the average surface plan ( 0), assuming
that they vary linearly (Fig. 6), and are function of and at the
mid-side points:
2EANS

1 A1 1 A2

2
2

30

2EANS

1 B1 1 B2

2
2

31

Fig. 6. Assumed natural transverse shear strains.

Fig. 7. ANS interpolation for transverse normal strain.

element mid-surface (Fig. 7), namely


4 1
ANS
E~ z 1 i 1 i E i ,i
i14

32

Poisson's ratio coupling requires the thickness strain to be a


linear function of . Because our solid-shell element has only two
layers, as consequence the thickness strain does not vary with
thus the element fail in reproducing the plane-stress condition. In
order to obtain a linear distribution of the normal strain in
thickness direction, we enhance the thickness strain eld by
adding an internal degree of freedom as
E EAS
E~ z
z

ANS

33

where is the 7th independent internal parameter. Eq. (33) is


known as the Enhanced Assumed Strains (EAS), in which the
enrichment variable will be eliminated by static condensation
technique within the element level. By substituting Eqs. (3031)
and (33) into Eq. (28), the virtual internal work takes the nal
expression
Z Z Z i 1
nl
~ EAS
W int
ETs H~ 1 E s E EAS
z T H 2 Ez

i1

GANS
T H~ 3 GANS
J d d d
s
s

34

where nl is the number of element layers and i is the transverse


coordinate position of the ith layer along the thickness. After
applying the FE discretization with the use of Eqs. (30), (31) and
(33) into the previous equation, one can obtain
T

ANS
T
W int uTn K mb K EAS
tr K sh un k k u  un un ku

35

2.5. Remedies for thickness and trapezoidal locking


Similar to shear locking, trapezoidal locking occurs when lower
order elements such as 8-node hexahedral elements are used to
model curved shells so that their cross-sections assume the
trapezoidal shape these excessive number of sampled thickness
strains can be reduced by using a bilinear interpolation of the
transverse normal strains sampled at the four corners of the

The expressions of ku and k for the composite multilayered


element are given by
Z Z Z i 1
nl
33
ku
H~ 2 BTz e J d d d
36

i1

nl

i1

Z Z Z

i 1
i

33
2 H~ 2 e J d d d

37

Finally we carry out a static condensation, in order to eliminate


the 7th unknown parameter in the previous expression which
T
leads to 1=k ku  un . Then the nal stiffness matrix takes

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

given by (Fig. 9)

the following expression:


1
T
ANS
ku ku
K K mb K EAS
tr K sh
k

38

The resulting 8-node hexahedral solid-shell FE model with 24


degrees of freedom is called SCH87.

3. Optimization using RSM based on Diffuse Approximation


The optimization problem can be stated as [30]
Minimize f x,

xRn

39

subject to a set of m 2n constraints:


g i x0, j 1,,m
Li xi U i , i 1,,n

40

where f is the objective function (OF), xi are the design variables,


gj is the jth constraint. The region of interest is dened by Li and Ui
which are respectively the lower and upper bounds on the design
variables. The RSM approach consists in solving a problem where
the OF is replaced by its approximation f~ .
This new problem may be written as
Minimize f~ x, xRn
Subject to g~ i x0, j 1,,m
Li xi U i , i 1,,n

wxi ,x wr

exp2 r 2 exp2
1exp2

42

where r xi x=di , with di being the radius of a spherical


support. As we can observe in Fig. 9, the parameter affects the
shape of the weight function. Generally, small values of the
parameter lead to a smooth and diffuse approximation but with
the cost of a slow convergence process, and for larger values of
the optimization process converges more rapidly but increases the
risk of divergence. Through the present investigation, it has been
found by the authors, that 2 (Fig. 9) is a good compromise since
it insures a good convergence rate while maintaining a certain
amount of diffusive character on the approximation.
The approximation is local, which means that only the points
closest to the current optimum are taken into account (Fig. 10). The
approximation coefcients are continuous when panning and/or
zooming of the region of interest is performed.
Given the function values for a set of experimental points xi
distributed according to a chosen Design of Experiment, the
function f~ can be dened in terms of basis functions p and some
adjusting coefcients a as
f~ x pT x  ax

43

41

A common choice for the basis functions p are linear and


quadratic monomials

The approximation Eq. (41) is based on a set of numerical


experiments with the function f. Generally, the approximate
functions encountered in RSM rely on second-order models, over
a given region of interest, with the constant regression coefcients
tted by means of least squares. The idea in this work is to apply
minimization algorithm progression by building new response
surfaces centered each successive solution (Fig. 8). During the
progression of the process, the region of interest moves and new
numerical experiments are performed at each iteration, which is
known as the Moving Least Square (MLS) approximation.
In the present investigation, we explore the application of DA
regression for building the response surface during the iterative
process. In the DA Method used in this work [26,27], the approximating functions are polynomials tted to the nodal values of
each local domain by a weighted least squares approximation.
Belytschko et al. [7] developed an Element Free Galerkin (EFG)
method which is an alternative implementation using Moving
Least Square (MLS) approximation [24]. In this paper we adopt the
moving least squares (MLS) interpolation. In the DAM the idea is to
replace the OF f(x) computed using the FE method, for a local
moving square tting. The resulting function f~ is more regular that
the function of the FE method, since the discontinuous coefcients
are placed, by continuous functions of weight, which gives a
continuity C m m1. The approximate function becomes smooth
by using continuous weighting functions. Different weight functions have been proposed in the literature, they differ in both the
shape of the domain of inuence, and in functional form. The
truncated Gaussian spherical weight function used in this work is


T
x2
x2
px 1 x1 x2 x3 xn x1 x2 x1 x3 xi xi 1 1 n
2
2

Fig. 8. Surface response optimization using the MLS approximation.

Fig. 10. Local moving response surface.

The size N of the vector p depends on the number of variables n


and on the polynomial degree of the approximation, for instance if

Fig. 9. Gaussian weight function: inuence of parameter .

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

quadratic approximation is used, then N n 1n 2=2. The


generalized coefcients a are determined by performing a
weighted least squares t for the local approximation, which is
obtained by minimizing the error Ja between the experimental
and approximated values of the OF

where M is the number of performed experiments and xi are the


experimental designs. The weights wi insure the continuity and
the locality of the approximation and are dened wi > 0, decreasing within a xed region around the point i called domain of
inuence of xi and vanish outside. The weight functions play a
crucial role by inuencing the way that the coefcients a depend
on the location of the design point x.
The minimization of Ja leads to the coefcients a and
consequently the function to the expression of the function f~ .
Thus, to obtain nonlinear parameters vector a, we use

The structure is a two-layer, composite, orthotropic, square


plate that is simply supported on its bottom edges. The layers are
oriented at 7 451 with respect to the plate edges. Fig. 11 shows the
plate dimensions and the loading pressure which is applied on the
top surface of the plate. Each layer has the material properties
given in Table 1.
Table 2 summarizes the results of deection by comparing the
normal displacement at the center of the plate and the in-plane
displacement at (x 0,y b=2) to the analytical solution and to

those obtained using ABAQUS by means of the SC8R solid-shell FE


model. We can observe from Table 2; that the present model gave
more precise results than those obtained with the solid-shell

model of ABAQUS with the same number of elements. For


instance, with only 10 elements, we obtain a maximum deection
of 22.461 mm which represents 3.4% of error when compared to
the analytical solution 23.250 mm, while the solution obtained

using the SC8R solid-shell element of ABAQUS is 28.543 mm which


presents and error of 22.7%.

J
AxaBxZ 0
a

4.2. Laminated strip under 3-point bending

Ja wxi ,xpT xi  axf xi 2

44

i1

45

with
ANN x P T WxP

46

BNM x P T Wx

47

and

6
6
W MM 6
6
4

wx1

0
wx2

wxM

0
2
6
P MN 4

pT x1

pT xM

3
7
5

and

One-quarter of the laminated strip is modeled. The same


problem is analyzed using different meshes of the developed 3D
solid-shell FE model. Various options are used to model the

3
7
7
7
7
5

8
9
>
< f x1 >
=

Z MN
>
: f x >
;
M

48

49

Once the objective function is estimated, then an optimization


algorithm can be used to minimize the resulting OF. In this work
we have adopted an algorithm based on a Sequential Quadratic
Programming technique [31] and a second algorithm based on the
Simplex method. Both algorithms are robust enough and suitable
to deal with constrained nonlinear optimization problems. But any
Minimization algorithm may be used at this stage.
We have to notice, that as the minimization procedure is an
iterative process, therefore at least few iterations (generally less than
10) are often needed for the SQP algorithm in order to nd the optimal
solution of the metamodel obtained using the RSM based on DA.

Fig. 11. Geometry and loading of the anisotropic layered plate.

Table 1
Material properties of the anisotropic layered plate.
Material property

Value

Units

E11
E22 E33
G12 G13 G23
13 23
12

27,6000
6900
2000
0
0.25

MPa
MPa
MPa

4. Numerical applications
In order to
model dened
with the aim
phenomena, as

evaluate the effectiveness of the solid-shell FE


previously, several benchmarks are carried out
of studying its behavior regarding the locking
well as the global rate of convergence.

4.1. Analysis of an anisotropic layered plate


The problem considered is the analysis of a at plate made
from two layers oriented at 7451, subjected to a uniform pressure
loading. The example veries simple laminated composite plate
analysis. Our numerical results are compared with the analytical
solution given in Spilker et al. [36] and a second verication is

done by a comparison to the results obtained using ABAQUS SC8R


solid-shell FE model. In this benchmark, the cross-section is not
balanced, so the response includes membrane-bending coupling.

Table 2
Plate deection convergence.
No. of elements

uA (mm)
Present model

uA (mm)

uC (mm)
Present model

uC (mm)

ABAQUS

2
4
6
8
10
20
30
40
50
100

0.335
0.360
0.365
0.368
0.370
0.371
0.371
0.371
0.371
0.371

0.423
0.414
0.416
0.417
0.417
0.418
0.418
0.419
0.419
0.419

17.960
20.806
21.766
22.204
22.461
22.956
23.114
23.191
23.236
23.241

29.435
28.649
28.584
28.555
28.543
28.460
28.475
28.480
28.485
28.495

Analytical

0.376

0.376

23.250

23.250

ABAQUS

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

laminated strip through the thickness. The models using the


developed SCH87 solid-shell employ a single element in the
thickness direction using a composite section based on 7 singlelayers through the thickness.
The structure is a 7-layer, composite, orthotropic, rectangular
strip that is simply supported on its supports on A and B and
subjected to a line load of 10 N/mm at C onto the top surface
(Fig. 12). The layers are oriented at 01/901 consecutively with
respect to the strip edges. Each layer has the material properties
given by Table 3 below.
Our numerical results are compared with the reference solution given by NAFEMS [28] and a second verication is done using
a comparison with the results obtained using the SC8R solid-shell

element of ABAQUS software.


Table 4 summarizes the results by comparing the normal
displacement at the center of the strip (point E) to the reference

solution and to those obtained using ABAQUS by means of the


SC8R solid-shell FE model. The displacement eld obtained using
the present SCH87 element is of a very good accuracy. We can see
from Table 4, that even with only 1 element per edge AE, the
obtained deection (0.807 mm) presents an error less than 23.9%
compared to NAFEMS reference solution, while the obtained

solution using ABAQUS solid-shell element, gave an error of


67.5%. The error vanishes as the number of element increases
(only through the edge, while keeping only 1 element in the

thickness), for only 9 elements per edge AE we obtained a


deection (1.059 mm) with an error less than 0.1%.
4.3. Multilayered composite plate with ply drop-offs
This example demonstrates the applicability of the present
solid-shell formulation to analyze a composite structure with ply
drop-offs; it consists on a composite multilayered plate (with
6 layers) with ply drop-offs (Fig. 13).
In this example, each layer is made of unidirectional berreinforced material [40], with the ber directions aligned at
45=45= 45=45= 45=451 with respect to the length direction. The structure, with length L 12 m, width b 6 m and
thickness h 0.1 m is clamped on its thicker side and the free
thinner end is subjected to a transverse normal load distribution
uniformly along the free edge equivalent to a concentrated force of
F600 N. The location of the ply drop-offs are at x 4 m and
x 8 m with the top two layers removed after each drop-off. The
layer material properties are given in Table 5.
Table 6 shows the computed solution obtained using the
developed SCH87 solid-shell model with 12  6 elements (12 along
the length and 6 elements along the width). One can observe that
the present model can predict accurately the global deection even
for a very thin structure (L=h 12=0:004 3000). Comparison of
the maximal deection obtained with the present model with the

one obtained using ABAQUS SC8R solid-shell element with the


same mesh (Fig. 13), gave a small error of only 2.4%, which shows
that the present model is free of shear locking.
4.4. Optimization of a warped thick cylinder under pressure
In this application a warped thick cylinder submitted to an
internal pressure is studied [28]. The continuum shell model
employs one single element with a composite section based on
two single-layers through the thickness. The structure is

Fig. 12. Geometry and loading of the laminated strip.

Table 3
Material properties of the laminated strip.
Material property

Value

Units

E11
E22 E33
G12
G13 G23
13 23
12

100,000
5000
3000
2000
0.3
0.4

MPa
MPa
MPa
MPa

Fig. 13. Multilayered composite plate with ply dropoffs: undeformed mesh.

Table 5
Material properties of the multilayered composite plate.

Table 4
Convergence of the laminated strip deection.
No. of elements
(edge AE)

uE (mm)
Present model

ABAQUS

1
3
6
9
12
15
18
30
60

0.807
1.034
1.055
1.059
1.061
1.061
1.062
1.062
1.062

1.776
1.062
1.052
1.056
1.057
1.057
1.058
1.058
1.058

NAFEMS

1.060

1.060

uE (mm)

Material property

Value

Units

E11
E22 E33
G12 G13 G23
12 13 23

25,000
1000
1000
0.2

MPa
MPa
MPa

Table 6
Deection of the multilayered composite plate.
Thickness

Deection (m)
ABAQUS

Deection (m)
Present model

0.0039
2.6933
40.9174

0.0029
2.5685
39.9480

0.1
0.01
0.004

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Fig. 14. Geometry and FE mesh of a warped thick cylinder.

composed of an inner isotropic cylinder E 210 GPa, 0:3 and an


outer orthotropic circumferentially wound cylinder having the
following material parameters, expressed in a cylindrical coordinate system: a circumferential modulus E11 130 GPa, a longitudinal and a transverse moduli E22 E33 5 GPa (as indicated in
Fig. 14), 12 13 0:25, 23 0, G12 G13 10 GPa, and
G23 5 GPa. The boundary conditions correspond to a displacement dz 0 at z0 and the cylinder is subjected to an internal
pressure of 200 MPa (Fig. 14).
At rst and prior to the optimization of the material parameters, the cylinder is modeled using the initial material data and
a mesh of 40 elements through its circumference, 10 elements
through the length and only 1 element in thickness direction. The
circumferential stress at z 0 is measured for 2 radius values
(R 23 mm, R27 mm) and compared to the NAFEMS reference

solution [28] and to the one obtained using the ABAQUS SC8R
solid-shell FE model.
For the inner ange we obtain S11 1534:8 MPa, the SC8R

ABAQUS gave S11 1477 MPa, while the reference solution was
1565 MPa. For the outer ange we obtain S11 892:8 MPa, the

ABAQUS solution was S11 900 MPa, while the reference solution
was 875 MPa. These results conrm that our solid-shell model and the
used mesh are good enough to carry out the optimization process.
The optimization problem consists in nding optimal ber
orientation angle of the outer orthotropic circumferentially
wound layer and the Young's modulus E of the inner isotropic
cylinder while keeping constant the cylinder expansion for
z 0 mm at a value of uR 0:5 mm. The objective function is based
on the general Hill criterion [42].
nel

J,E J J e

Fig. 15. Design of experiments: (a) central composite design and (b) Box-Behnken
design.

Table 7
DOE plan using central composite design.
Run

Factor x1

Factor x1

Response J (MPa2)

Disp. uR (mm)

1
2
3
4
5
6
7
8
9

1
1
1
1
1
1
0
0
0

1
1
1
1
0
0
1
1
0

268316.8
294069.1
282546.0
293785.5
278756.9
294282.9
280609.1
286957.9
285223.1

0.355
0.833
0.150
0.196
0.210
0.316
0.584
0.134
0.211

50

e1

where e is the element index, nel is the total number of elements


on the whole structure, the element OF is given by
J e FS22 S33 2 GS33 S11 2 HS11 S22 2 2NS212 2LS223 2MS231
51
The design variables (x1 ,x2 E) are constrained between
01901 and 60 GPaE300 GPa. Before performing the optimization procedure, a DOE based on composite design (Fig. 15a) built
of 9 functions evaluations for the full cylinder, is carried out. In this
case we used 3 groups of design points (4 two-level factorial
design points, 4 axial points and a 1 center point). Then design
variables are coded as (1, 1) in order to facilitate the data
treatment. The DOE plan with the OF values is given in Table 7.
Fig. 16 shows the global quadratic response surface model
based on DA which has been calculated and given explicitly by
J~ ,E 291900 4107:30 1842:52E7520:69E
10230:382 261:44E2
The response approximation function for the constraint
(uR 0.5) on the radial displacement of the cylinder at z 0 is also

Fig. 16. Approximated response surface.

10

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

carried out using (DA) and given explicitly by (see Fig. 17).
~
u,E
0:49 0:026 0:071E 0:13E0:132 0:098E2

~
The minimization of J~ ,E under constraint u,E
0:5 (see
Fig. 18) has been done using the SQP algorithm based on the work
of Powell [31], the optimal solution was obtained in 5 iterations
leading to the optimal solution in coded form (0.830, 0.961) which
corresponds to n ,En 82:341,295:40 GPa.
Fig. 19 shows the hoop stress distribution on the cylinder
before and after optimization. We can observe just a few amount
of stress reduction is obtained after optimization, this is due to the
~
presence of the constraint u,E
0:5 which cannot lead to the
minimum of the unconstrained objective function J~ ,E.

4.5. Optimization of a multilayered strip panel

Fig. 17. Response surface of constraint function.

Fig. 18. Optimal solution in contour plot.

The structure is a thick composite multilayered shell (Fig. 20),


composed by two layers having the same thickness. The lower layer
is made of an aluminum alloy 6066-T651 (E67 GPa, 0:34) and
the upper layer is an orthotropic lightweight based aramid bers
with strong characteristics (E11 130 GPa, E22 E33 5 GPa, 12
13 0:25, 23 0, G23 5 GPa and G12 G13 10 GPa).
The structure is submitted to a transverse load in point A of
4.17 N/mm, and clamped at its end BC. The structure was meshed
using 1596 3D-shell elements corresponding to 3588 nodes
(Fig. 20).
Prior to performing the material optimization, a FE model
validation is carried out using initial material parameters. A
numerical comparison of our results to those obtained with

ABAQUS software by means of the SC8R solid-shell model is


performed. Fig. 21 shows the deection of the strip panel under
the maximal load. Comparison of the deection value obtained
using the present SCH87 solid-shell model gave 13.747 mm which

is in good agreement with 13.676 mm obtained using ABAQUS .


The optimization problem consists in nding optimal ber
orientation angle of the upper orthotropic lightweight based
aramid layer and the Young's modulus E of the lower aluminum
alloy layer, while the maximal deection under load is maintained
equal to 15 mm. The objective function is based on the general Hill
criterion [1] and dened by Eq. (51).

Fig. 20. Initial undeformed mesh and loading conditions of a multilayered


strip panel.

Fig. 19. Hoop stress distribution.

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

11

Fig. 21. Comparison of the global deection of a multilayered strip panel (SC8R vs. SCH87). (a) ABQUS (SC8R). (b) Present model SCH87.

Fig. 22. Approximated response surface.

The optimization problem is stated us


nel

Minimize

J,E J e ,

Fig. 23. Optimal solution in contour plot.

,ERn
obtained for the optimal structure is about un 15:11 mm which
respects the imposed constraint u15 (Fig. 23).

e1

with

u,E15 mm0,

22:51 22:51,
60 GPaE79 GPa:

52

The response surface model based on DA is built up using the


same DOE based on central composite design using 9 experiments.
Fig. 22 shows the approximated RSM calculated which is given by
the quadratic metamodel
J~ ,E 939800 58864:99 29763:57E 11549:15E
1465002 1431:34E2

The response approximation for the constraint (ux15 0)


has been also carried out using DA by means of a quadratic
metamodel, it is given explicitly by
~
u,E
13:08 31:22E0:17E 4:582 0:14E2
The optimal solution obtained after un-normalizing the DV
corresponds to n ,En 3:401,60:12 GPa, the displacement

4.6. Optimization of a composite leaf spring for automobile


suspension
In this application, a composite leaf spring for automobile
suspension is studied. The structure shown in Fig. 24 with a
10 mm constant thickness is made of a composite orthotropic
material whose mechanical characteristics are given in Table 8.
The optimization problem here, consists in nding the best
orientation of the material bers which allows reducing the
maximal stresses over the structure under ultimate loading
conditions.
The FE mesh using 1220 solid-shell elements and the boundary
conditions and loading are shown in Fig. 24. The structure
supports a concentrated force at its free end (applied on the
bottom shell surface) of 500 N, the other edge being clamped.
Prior to the optimization process, the FE model has been
validated using initial material parameters with an orientation
angle 01 with reference to the global x-axis (Fig. 24). A rst

12

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Fig. 24. Geometry and boundary conditions of a composite leaf spring suspension.

Table 8
Material properties of a composite leaf spring suspension.
Material property

Value

Units

E11
E22 E33
G12 G13 G23
12 13 23

39,000
10,000
5000
0.3

MPa
MPa
MPa

analysis is carried out using ABAQUS software and its SC8R solidshell element with the same mesh. Fig. 25 shows the stress
distribution on the top ange of the structure in the 1st principal
direction. Comparison of maximal values of S11 obtained using the
present solid-shell model gave 185.79 MPa which is almost the
same maximal value 185.41 MPa obtained with the SC8R element

of ABAQUS .
Unlike the stress S11, the distribution of the transverse stress
S33 in the thickness direction, shows that a maximal value of

20 MPa is reached, while the ABAQUS solution indicates zero


stress as indicated in Fig. 26 (This results from the fact that the

SC8R solid-shell element of ABAQUS is based on plane stress


condition). Analysis of other stress components shows small
values compared to the principal stress S11 values and a compar
ison with ABAQUS results shows good agreement. This rst study
conrms that the present model is valid and therefore the
optimization process can be performed.
In order to minimize the number of design variables, a
1-dimensional parameterization is used through the curvilinear
abscissa s where 5 DVs are dened as indicated in Fig. 27. The
points represented by 1 ,2 ,,5 are the optimization poles of
control, where the bers orientation angle i is dened. Inside each
nite element, the ber orientation angle e is calculated using a
linear interpolation between two successive poles where the
element is located. For instance, if an element is located between
two poles i and j , the material orientation angle inside the
element is dened as




sj s
ssi
s
i

sj si
sj si j
Here again, the objective function is based on the Hill criterion
[1] and dened by Eq. (51). Thus the optimization problem can be
stated in terms of the DV vector f1 ; 2 ; 3 ; 4 ; 5 g as follows:
nel

Minimize

J J e ,
e1

with

601i 601,

Rn
i 1,5

53

In order to reduce the number of evaluations of the OF, a DOE


based on the Box-Behnken algorithm is used with three levels
coded between {1, 0, 1} (Fig. 15b). The Box-Behnken DOE with
5 variables, allows a screening of only 46 experiments (or
combination of variables) coded between 1 and 1. The quadratic
response surface model based on DA is built up using the above
mentioned DOE based on 46 experiments in the space of 5 coded
DV. Fig. 28 shows the approximated RSM calculated and given by
the following quadratic metamodel:
J~ 1874:041206:351 2210:642 1240:793 331:034 71:525

2146:901 2 0:421 3 1:501 4 0:301 5 1281:272 3


4:252 4 1:142 5 444:043 4 2:083 5 64:004 5
1245:7521 2475:7022 1375:8423 269:3524 31:3325
The minimization of the resulting metamodel was carried out
with two different algorithms, namely the VF02AD algorithm
using the SQP method [31] and a second optimizer based on the
simplex algorithm. Different initial points were used to minimize
the OF, and the best result obtained after minimization corresponds to n f15:771; 15:451; 17:331; 15:191; 0:001g.
After obtaining the optimal material bers orientations, we
carried out a last FE calculation using optimal solution n found, by
means of the developed SCH87 solid-shell model. Review of the
principal results is given in Table 9, where maximal stresses in all
principal directions are compared to values obtained using the
initial material data with 01. As we can observe, the optimal
bers orientation, allow a reduction of the maximal stress distribution in the structure, we found also that the obtained results
if physically good since the orientation of the material bers
follows the general orientation of the structure geometry.

5. Conclusion
In the present investigation, an efcient 8-node solid-shell
element formulation for the analysis of multilayered composite
shell is presented. While the ANS method has been used in order
to remedy to shear locking, the enhancement of transverse normal
strain is adopted, thus full 3D anisotropic constitutive model is
incorporated without resorting to the plane-stress assumption.
The present formulation can predict the through-thickness effects
with a high degree of accuracy.
In the second part of the paper we proposed a specic method
based on RSM for the optimization of laminated structures, where

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Fig. 25. Comparison of stress distribution on the top ange in the 1st principal direction. (a) ABQUS (SC8R). (b) Present model SCH87.

Fig. 26. Comparison of stress distribution on the top ange in the 3rd principal direction (thickness direction). (a) ABQUS (SC8R). (b) Present model SCH87.

Fig. 27. Parameterization model of design variables.

the design variables are the material bers orientation. To solve


the optimization problem, we proposed a technique which has the
advantage of replacing a complex response model by an approximate model evaluated in a limited number of points obtained
through an experimental design. In this study, we proposed a
specic response surface method based on Diffuse Approximation
involving pattern search optimization. The resulting response

Fig. 28. Approximated response surface in the space f1 ,2 g.

13

14

H. Naceur et al. / Finite Elements in Analysis and Design 7071 (2013) 114

Table 9
Summary of principal results obtained before and after optimization.
Maximal stress
component

Initial solution using 01


(in [MPa])

Optimal solution using n


(in [MPa])

S11
S22
S12
S33

185.79
20.41
34.53
20.43

173.90
19.12
12.66
19.20

surface algorithm involves iterative improvement of the objective


function employing locally supported nonlinear approximations.
Numerous applications have been treated, they conrmed that
the optimization method based on RSM and DA coupled to the
developed SCH87 FE solid-shell model is efcient and particularly
suited for industrial problems in structural mechanics.
References
[1] S. Abid, J.L. Batoz, C. Knopf-Lenoir, P. Lardeur, H. Naceur, Thickness Optimization of Beams and Shells with Large Displacements, Integrated Design and
Manufacturing in Mechanical Engineering, Editions Kluwer Academic
Publishers173182.
[2] R.J. Alves de Sousa, J.W. Yoon, R.P.R. Cardoso, R.A. Fontes Valente, J.J. Gracio, On
the use of a reduced enhanced solidshell (RESS) element for sheet forming
simulations, Int. J. Plasticity 23 (2007) 490515.
[3] P.M.A. Areias, J.M.A. Csar de S, C.A. Concei ao Antnio, A.A. Fernandes,
Analysis of 3D problems using a new enhanced strain hexahedral element, Int.
J. Numer. Meth. Eng. 58 (2003) 16371682.
[4] B. Bassa, F. Sabourin, M. Brunet, A new nine-node solid-shell nite element
using complete 3D constitutive laws, Int. J. Numer. Meth. Eng. 92 (7) (2012)
589636 16.
[5] K.J. Bathe, E.N. Dvorkin, A formulation of general shell elements- the use of
mixed interpolation of tensorial components, Int. J. Numer. Meth. Eng. 22
(1986) 697722.
[6] J.L. Batoz, G. Dhatt, Finite Element Modeling of Structures, 3rd Edition, vol. 3,
Herms 1992, Paris (in french).
[7] T. Belytschko, Y.Y. Lu, L. Gu, Element free Galerkin method, Int. J. Numer. Meth.
Eng. 37 (1994) 229256.
[8] P. Betsch, E. Stein, An assumed strain approach avoiding articial thickness
straining for a non-linear 4-node shell element, Commun. Numer. Meth. Eng.
11 (1995) 899909.
[9] M. Bischoff, E. Ramm, Shear deformable shell elements for large strains and
rotations, Int. J. Numer. Meth. Eng. 40 (1997) 44274449.
[10] N. Bchter, E. Ramm, Shell theory versus degeneration a comparison in large
rotation nite element analysis, Int. J. Numer. Meth. Eng. 37 (1992) 5562.
[11] R.P.R. Cardoso, J.W. Yoon, M. Mahardika, S. Choudry, R.J. Alves de Sousa, R.A.
F. Valente, Enhanced assumed strain (EAS) and assumed natural strain (ANS)
methods for one-point quadrature solid-shell elements, Int. J. Numer. Meth.
Eng. 75 (2008) 156187.
[12] C. Cho, H.C. Park, S.W. Lee, Stability analysis using a geometrically nonlinear
assumed strain solid shell element model, Finite Elem. Anal. Des. 29 (1998)
121135.
[13] D.R. Cox, N. Reid, The Theory of the Design of Experiments, Edition, Chapman
Hall/CRC, 2000.
[14] S. Doll, K. Schweizerhof, R. Hauptmann, C. Freischlager, On volumetric locking
of low-order solid and solid-shell elements for nite elastoviscoplastic
deformations and selective reduced integration, Eng. Comp. 17 (2000)
874902.
[15] E. Domissy, S. Bouabdallah, J.L. Batoz, Formulation and evaluation of a solid
nite element model for the linear and nonlinear shell analysis, in: 12eme
Congrs Franais de Mcanique Strasbourg, France, September 1995 (in french).
[16] D. Durban, D. Givoli, J.G. Simmonds, Advances in the Mechanics of Plates and
Shells, Kluwer Academic Publishers, 2002.

[17] E.N. Dvorkin, K.J. Bathe, A continuum mechanics based four node shell
element for general nonlinear analysis, Eng. Comput. (1984) 177.
[18] M. Hannachi, H. Naceur, J.L. Batoz, Continuum based solid-shell element
modeling for the optimization of composite multilayered structures, Int. Rev.
Mech. Eng. 1 (4) (2007) 150163.
[19] M. Harnau, K. Schweizerhof, Articial kinematics and simple stabilization of
solid-shell elements occurring in highly constrained situations and applications in composite sheet forming simulation, Finite Elem. Anal. Des. 28 (34)
(2006) 10971111.
[20] R. Hauptmann, K. Schweizerhof, A systematic development of Solid-Shell
element formulation for linear and non-linear analysis employing only
displacement degrees of freedom, Int. J. Numer. Meth. Eng. 42 (1998) 4969.
[21] S. Klinkel, F. Grutmann, W. Wagner, A continuum based d-shell element for
laminated structures, Comput. Struct. 71 (1999) 4362.
[22] R.L. Mason, R.F. Gunst, J.L. Hess, Statistical Design and Analysis of Experiments
with Applications to Engineering and Science, 2nd ed, John Wiley & Sons,
2003.
[23] S.K. Mazumdar, Composites ManufacturingMaterials, Product, and Process
Engineering, Edition, CRC Press LLC, Florida, 2002.
[24] R.H. Meyers, D.C. Montgomery, Response Surface Methodology Process and
Product Optimization Using Designed Experiments, 2nd Edition, Wiley, 2003.
[25] R.A.S. Moreira, R.J. Alves de Sousa, R.A.F. Valente, A solid-shell layerwise nite
element for non-linear geometric and material analysis, Compos. Struct. 92 (6)
(2010) 15171523.
[26] H. Naceur, A. Delameziere, J.L. Batoz, Y.Q. Guo, C. Knopf-Lenoir, Some
improvements on the optimum process design in deep drawing using the
Inverse Approach, J. Mater. Process. Technol. 146 (2) (2004) 250262.
[27] H. Naceur, Y.Q. Guo, S. Ben-Elechi, Response surface methodology for design
of sheet forming parameters to control springback effects, Comput. Struct.
84 (2006) 16511663.
[28] NAFEMS, National Agency for Finite Element Methods and Standards
(NAFEMS), Test R0031/2 from NAFEMS publication R0031, Composites Benchmarks Issue 2, February 5, UK, 2001.
[29] N.H. Nguyen, V.N. Pham, M. Hogge, J.P. Ponthot, An assumed natural strain
technique for 2D solid-shell elements, in: M. Hogge et al. (Eds.), Fourth
International Conference on Advanced Computational Methods in Engineering, University of Lige, Belgium, 2628 May 2008.
[30] J. Nocedal, S.J. Wright, in: P. Glynn, S.M. Robinson (Eds.), Numerical Optimization, Springer-Verlag, New York, 1999.
[31] M.J.D. Powell, Extensions to subroutine VF02AD, in: R.F. Drenick, F. Kozin
(Eds.), Systems Modeling and Optimization, Lecture Notes in Control and
Information Sciences, vol. 38, Springer-Verlag, Berlin, 1982, pp. 529538.
[32] N.-D. Quy, A. Matzenmiller, A solid-shell element with enhanced assumed
strains for higher order shear deformations in laminates, Tech. Mech. 28 (34)
(2008) 334355.
[33] K. Rah, W. Van Paepegem, A.M. Habraken, J. Degrieck, A partial hybrid stress
solid-shell element for the analysis of laminated composites, Comp. Meth.
Appl. Mech. Eng. 200 (4952) (2011) 35263539.
[34] M. Schwarze, S. Reese, A reduced integration solid-shell nite element based
on the EAS and the ANS conceptgeometrically linear problems, Int. J. Numer.
Meth. Eng. 80 (2009) 13221355.
[35] S. Shiri, H. Naceur, J. Roelandt, Numerical modeling of stamping and crashworthiness of steel/polymer/steel structures using solid-shell element, in:
International Conference on Computational Plasticity, COMPLAS X 2009,
Barcelona, Spain, 2009.
[36] R.L. Spilker, S. Verbiese, O. Orringer, S.E. French, E.A. Witmer, A. Harris, Use of
the hybrid-stress nite-element model for the static and dynamic analysis of
multilayer composite plates and shells, Report for the Army Materials and
Mechanics Research, Center Watertown, MA, 1976.
[37] K.Y. Sze, S. Yi, M.H. Tay, An explicit hybrid-stabilized eighteen-node solid
element for thin shell analysis, Int. J. Numer. Meth. Eng. 40 (1997) 18391856.
[38] K.Y. Sze, L.Q. Yao, A hybrid stress ANS solidshell element and its generalization for smart structure modeling, Part I-solidshell element formulation,
Int. J. Numer. Meth. Eng. 48 (2000) 545564.
[39] J.R. Vinson, R.L. Sierakowski, The Behavior of Structures Composed of Composite Materials, 2nd Edition, Kluwer Academic Publishers, 2004.
[40] L. Vu-Quoc, X.G. Tan, Optimal solid shells for non-linear analyses of multilayer
composites. I. Statics, Comp. Meth. Appl. Mech. Eng. 192 (2003) 9751016.
[41] G. Wempner, D. Talaslidis, Mechanics of Solids and ShellsTheories and
Approximations, Edition, CRC PRESS LLC, New York, 2003.
[42] X. Xiao, C. Hsiung, Z. Zhao, Analysis and modeling of exural deformation of
laminated steel, Int. J. Mech. Sci. 50 (2008) 6982.