You are on page 1of 11

Composites: Part A 35 (2004) 977987

www.elsevier.com/locate/compositesa

Identification by modal analysis of composite structures modelled with


FSDT and HSDT laminated shell finite elements
J. Cugnoni, Th. Gmur*, A. Schorderet
School of Engineering (STI), Swiss Federal Institute of Technology (EPFL), 1015 Lausanne, Switzerland

Abstract
This paper presents an improvement and an extension to modal analysis of an existing multilayered composite shell finite element.
Generalising the formulation to a set of elements, the proposed models are based upon the first- and higher-order shear deformation theories
and are well suited for evaluating the global dynamic response of thin and thick laminated shells respectively. Characterized by a throughthe-thickness displacement approximation of a freely chosen order, they display excellent convergence properties when the polynomial order
is increased and present a higher computational effectiveness in comparison to the classical layerwise models. The models considered are
compared to closed-form solutions based on the layerwise plate theory and the so-called zig zag formulation. Experimental and numerical
modal test cases on thin and thick plates are next investigated in order to validate the proposed shell models. Good agreement is found with
the analytical, experimental and numerical references.
q 2004 Elsevier Ltd. All rights reserved.
Keywords: A. Layered structures; B. Vibration; C. Finite element analysis (FEA); D. Mechanical testing

1. Introduction
The fundamental natural frequencies and associated
mode shapes of fibre-reinforced composite structures
considered as thin to moderately thick shells can often
be accurately determined using the equivalent single-layer
(ESL) laminate theory. In this approach, which has led to
more or less sophisticated finite elements, the heterogeneous laminated shell component is treated as a
statically ESL, reducing the 3D-continuum problem to a
2D-one. In the classical laminated shell theory, which
forms the simplest ESL approach, problems are based on
the Kirchhoff hypothesis, which amounts to neglecting
both transverse shear deformation and transverse normal
effects. Since the ratio of the shear modulus to the inplane elastic moduli for composite materials is much
lower than that of isotropic materials, the shear deformation effect must be included in the analysis of
moderately thick laminated structures. The next approach
in the hierarchy of ESL shell finite element models is so
the first-order shear deformation theory (FSDT) [1 4],
which includes a rudimentary form of shear deformation
in the kinematic assumptions. In order to further increase
* Corresponding author. Tel.: 41-21-693-2924; fax: 41-21-693-3509.
E-mail address: thomas.gmuer@epfl.ch (T. Gmur).
1359-835X/$ - see front matter q 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compositesa.2004.01.014

the accuracy of the finite element models, refined


transverse shear deformation models, based on second
and higher-order shear deformation theories (HSDT) have
been developed [5].
In order to assess the validity range of these finite
element models for computing the global behaviour of
laminated shells, experimental studies should be conducted on specimens of different thicknesses. Due to the
complexity and the cost of experimental tests, the
number of experiments undertaken on composites is
unfortunately only a small fraction of the number of
analytical and numerical investigations and such tests are
seldom reported in the literature. Though some purely
experimental or mixed experimental numerical studies
have been recently conducted on composite structures
[6 11], no rigorous experimental research, which
enhances the applicability limits of the ESL shell models,
is available.
Two families of C 0 -compatible doubly curved shell finite
elements for the modal analysis of multilayered composite
structures are presented in this paper. Closely related to the
formulation of the classical Ahmad-type 3D-degenerated
shell element, the proposed models are based on the firstand HSDT, respectively. The first element set constitutes
a generalisation of a standard quadratic serendipian FSDT
laminated shell model [1], whereas the second one forms an

978

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

improvement of an HSDT multilayered shell formulation


[12] extended to dynamic analysis. Suitable for triangular or
quadrangular meshes, both element sets are based on linear,
quadratic or cubic in-plane shape functions. They are
accurate in thin structures if only the overall global response
of the structure is required. For large thickness-to-span
ratios, only the HSDT model, however, yields to precise
results. A validation of these sets of composite shell
elements is next addressed. In order to illustrate the
accuracy and effectiveness of models based on first- and
HSDT when used for discretizing thin to relatively thick
shells, three modal test cases are considered, where the
values obtained with the investigated formulations are
compared to closed-form solutions, experimental results
and numerical predictions extracted from a commercial
finite element software.

2. FSDT and HSDT finite element description


2.1. Displacement approximation
A general doubly curved composite shell model with a
variable number of nodal points on the finite element
midsurface can be derived in a similar way to the
classical C0 shell formulation by degenerating a multilayered 3D element. Although the original solid element
is usually characterized with linear shape functions
across the degeneration line (Fig. 1) in order to satisfy
the main assumptions in the Reissner Mindlin theory for
thin to moderately thick plates or shells, a higher-order
displacement field in the thickness direction (Fig. 2) can
also be adopted [12]. When degenerating such a solid
element, the displacement approximation in the thickness
direction of the resulting shell element can be of an
arbitrary polynomial order p; which improves the element
behaviour in comparison to the linear formulation when
the shell thickness increases. Shell elements based upon
this approximation type have moreover a hierarchical
structure, the element shape functions associated to an
order p 2 1 being a subset of those corresponding to
order p; which guarantees some computational
advantages.
The approximate generalized displacement vector e uh for
a p-type HSDT shell element e V (Fig. 2) may be expressed
in terms of the following polynomial series up to order p
(
e
p
e i
X
t3 e 1
a
e h
u j; t
hi j1 ; j2 e p0
t

j
p t
3
i
2 i
i1
!2
!p
)
j23 e t3i e 2
jp3 e t3i e p

pi t
pi t
2! 2
p! 2
1
where hi j1 ; j2 i 1; 2; ; p represents the i-th polynomial shape function of the parent or master element a V
a

Fig. 1. Transformation of a multilayered solid element into a laminated


FSDT shell element.

(Fig. 3), e p denotes the total number of nodal points on the


element mid-surface, j j1 ; j2 ; j3 is the natural co-ordinate
vector, e t3i i 1; 2; ; e p is the element thickness at nodal
pointi;and t denotes time.In Eq.(1), the quantities stored in the
e
vectors e pj
i t i 1; 2; ; p; j 0; 1; ; p; which constitute the nodal unknowns, are the j-th derivatives of the
displacement approximation with respect to the shell thickness global co-ordinate z3 2e t3i =2; e t3i =2 at node i (Fig. 3)

j e uh 
e j
pi
j 0; 1; ; p
2

zj3 j1 ji1 ; j2 ji2
where jik k 1; 2 are the local in-plane co-ordinates for
nodal point i: Taking the relationship z3 2j3 =e t3i between the
local j3 and global z3 co-ordinates into account, the unknown
vectors Eq. (2) may be rewritten as
!j
!j

2 j e uh 
2 e j
e j
pi e i
ei
qi j 0;1;;p

t3
t3
jj3 j ji ; j ji
1

3
e j
qi t

where the nodal degrees of freedom


i 1;2;; e p;
j 0;1;;p are the j-th derivatives of the displacement
approximation with respect to the shell thickness local

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

979

Fig. 3. HSDT shell element with 3p 1 degrees of freedom per nodal point.

and the last three ones are the normalized p-th derivatives of
the displacements at nodal point iall the components
being measured in the global reference system, Eq. (4)
may be expressed in the following compact form
Fig. 2. Transformation of a multilayered solid element into a laminated porder HSDT shell element.

co-ordinate j3 21;1 at node i: By replacing the nodal


displacements by the new unknowns, the displacement
approximation simplifies in the following expression, which
is computationally more stable than interpolation (1) if the
shell element is relatively thin,
(
e
p
X
j23 e 2
e 1
a
e h
u j;t
hi j1 ; j2 e q0
q t
i t j3 qi t
2! i
i1

jp
t
4
3 e qp
p! i
The numerical conditioning of the proposed HSDT model is
thus improved in comparison to the formulation given in Ref.
[12] which is unstable for large values of the order p and for
thin plates.
When collecting the 3p 1 degrees of freedom at nodal
point i into the vector e qi
e

e 0 e 0 e 1 e 1 e 1
e p e p e p T
qi {e q0
1;i ; q2;i ; q3;i ; q1;i ; q2;i ; q3;i ;; q1;i ; q2;i ; q3;i }

5
in which the first three components are the displacements at
node i; the next three terms are the normalized rotations or
first derivatives of the displacements at the same point

u j; t

e h

p
X
a

Hi je qi t a Hje qt

i1

with
a

H a H1 ; a H2 ; ; a Hi ;; a Hep 

7a

q {e q1 ; e q2 ;; e qi ; ; e qep }T

7b

where a H denotes the interpolation matrix and e q is the


generalized nodal displacement vector. The 3 3p 1
submatrices a Hi i 1; 2; ; e p located in a H are written as
3
2
j23 a
a
a
0 j 3 hi 0
0
h
0 7
6 hi 0
7
6
2! i
7
6
p
7
6
j3 a
7
6
7
6
h
0
0
0

i
7
6
p!
7
6
7
6
6
2
j3 a 7
7
6
a
a
60
hi 0
0 j 3 hi 0
0
hi 7
a
7
6
2!
Hi 6
7
7
6
p
7
6
j
3a
7
6
0

0
h
0
i
7
6
p!
7
6
7
6
a
a
60
0
hi 0
0 j 3 hi 0
0 7
7
6
7
6
7
6
p
2
5
4
j3 a
j3 a
h 0
0
h
2! i
p! i
8

980

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

a hi I; j3 I; j23 I;; jp3 I

where I is the identity matrix of order 3.


The displacement approximation for an FSDT shell finite
element with six degrees of freedom per node is then easily
derived from Eq. (4) by retaining the first two terms in the
Taylor series p 1 for a linear distribution of the throughthe-thickness displacement)
e

p
X
a

u j; t

e h

e 1
hi j1 ; j2 {e q0
i t j3 qi t}

i1
e

p
X
a

Hi je qi t

10

i1

where the 3 6 submatrices a Hi i 1; 2; ; e p and the


subvectors e qi i 1; 2; ; e p read
2a
3
hi 0
0 j 3 a hi
0
0
6
7
a
a
Hi 6
hi 0
0
j 3 a hi
0 7
11a
4 0
5
0
e

hi

j 3 a hi

e 0 e 0 e 1 e 1 e 1 T
qi {e q0
1;i ; q2;i ; q3;i ; q1;i ; q2;i ; q3;i }

11b

the first and last three components in e qi being the global


translational and rotational degrees of freedom at nodal
point i; respectively.
In order to enforce the fibre inextensibility in the
thickness direction, an alternative FSDT shell formulation
can be developed [13], where the displacement field is
written in terms of five degrees of freedom per node,
(
)
e
p
e i
X
t3 e i e i e 1
a
e h
e 0
u j; t
hi j1 ; j2 qi t j3 2 v1 ; v2  qi t
2
i1
e

p
X
a

Hi je qi t

12

i1
e
where e q0
i t i 1; 2; ; p stores the three global
displacements at node i; whereas the two components
e
located in e q1
i t i 1; 2; ; p are the rotations of the unit
e i
vector v3 ; normal to the e-th element mid-surface at node i;
about two mutually orthogonal unit vectors e vi1 and e vi2
which form a plan normal to e vi3 at nodal point i (Fig. 4). The
interpolation submatrices a Hi i 1; 2; ; e p and the subvectors e qi i 1; 2; ; e p are now expressed as
2a
3
hi 0 0 2j3 e gi1;2 j3 e gi1;1
6
7
6
7
a
13a
Hi 6 0 a hi 0 2j3 e gi2;2 j3 e gi2;1 7
4
5
a
e i
e i
0 0 hi 2j3 g3;2 j3 g3;1
e

e 0 e 0 e 1 e 1 T
qi {e q0
1;i ; q2;i ; q3;i ; q1;i ; q2;i }

13b

where the scalars e vij;k j 1; 2; 3 are the components of the


unit vector e vik k 1; 2:
In comparison to the FSDT element with six degrees of
freedom per nodal point derived from the HSDT p-order
formulation with p 1; the FSDT shell model based upon
five unknowns per node is less time and storage
consuming and satisfies automatically the zero-rotation
condition about the normal direction. However, the FSDT
formulation with five or six degrees of freedom per node
contains only a rudimentary form of shear deformation
since the displacement is simply assumed to be linear in
the thickness direction. The use of an HSDT formulation
improves the modal response of the structurein large
thickness-to-span situations , but enlarges the number of
degrees of freedom per node by six for each odd
order increase, so that the total number of
unknowns may become prohibitive. Fortunately, as
shown farther on, the order p can generally be chosen
less than or equal to 5.
2.2. Strain and stress approximations
Assuming a linearized strain tensor, the strain vector for
the e-th element may be written as
1 j; t

e h

with
e i
gj;k

Fig. 4. FSDT shell element with 5 degrees of freedom per nodal point.

e i
t
hi 3 e vij;k

ep
X
i1

j 1; 2; 3; k 1; 2

14

7a Hi je qi t

ep
X
e

Bi je qi t

i1

7a Hje qt e Bje qt

15

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

where f symbolizes the global differential operator and e B


is the strain-displacement matrix
e

B e B1 ; e B2 ; ; e Bi ; ; e Bep 

16

p1 (six degrees of freedom per node) are then obtained by


simply keeping the first two 6 3 blocks in matrix (17).
For the FSDT shell element with five degrees of freedom
per node, the 6 5 submatrices e Bi i 1; 2; ; e p stored in
the strain-displacement matrix e B are calculated from the
corresponding submatrices a Hi i 1; 2; ; e p shown in
Eq. (13a)

For the p-order HSDT model, the 6 3p 1 submatrices e Bi i 1; 2; ; e p stored in matrix e B are derived
from the corresponding submatrices a Hi i 1; 2; ; e p
2 a
j3 e gi1;2
j3 e gi1;1
hi
0
0
2
6
6 x 1
x 1
x 1
6
6
e i
a
6
j3 g2;2
j3 e gi2;1
hi
6 0
0
2
6
6
x 2
x 2
x 2
6
6
6
j3 e gi3;2
j3 e gi3;1
a hi
6 0
0
2
6
x 3
x 3
x 3
6
e
Bi 6
6
6
j3 e gi2;2
j3 e gi3;2 j3 e gi2;1
j3 e gi3;1
a hi a hi
6 0
2
2

6
x 3
x 2
x 3
x 2
x 3
x 2
6
6
6 a
i
i
i
e
e
e
6 hi
j3 g1;2
j3 g3;2 j3 g1;1
j3 e gi3;1
a hi
6
0
2
2

6 x
x 1
x 3
x 1
x 3
x 1
6 3
6
6 a
i
i
i
e
e
e
a
4 hi hi
j3 g1;2
j3 g2;2 j3 g1;1
j3 e gi2;1
0
2
2

x 2
x 1
x 2
x 1
x 2
x 1
given by Eq. (8) as
2

a hi
6
6 x 1
6
6
6
6 0
6
6
6
6
6
6 0
6
6
6
6
6 0
6
6
6
6 a
6 hi
6
6
6 x 3
6
6 a
4 hi
x 2

a hi
x 2

a hi
x 3

a hi a hi
x 3 x 2
a hi
x 1

0
a

hi
x 1

3
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
5

19

with


jp3 a
hi
x1 p!

3
0

7
7
7
 p 
7
7
j3 a
7

0
hi
0
7
x2 p!
7
 p 7
j3 a 7
7

0
0
h 7
x3 p! i 7
7
 p 
 p 7
j3 a
j3 a 7

0
hi
h 7
7
x3 p!
x2 p! i 7
7



7
jp3 a
jp3 a 7

h
h 7
0
7
x3 p! i
x1 p! i 7
7




7
5
jp3 a
jp3 a

h
h
0
x2 p! i x1 p! i
17

with

xj

981

a hi
a hi e 21 a hi
e Jj221
Jj1
x j
j 1
j2
j3 e gik; e t3i e i

v
x j
2 k;

 


jk3 a
1 k e 21 a hi e 21 a hi
a
hi
j3 Jj1
Jj2
h
ke Jj321 jk21
i
3
k!
k!
j 1
j 2
j1;2;3; k 0;1;;p

18

where e Jj21
j; 1;2;3 denotes the j th component in the
inverse of the Jacobian matrix associated to the transformation mapping from the master element a V to the
geometrically deformed element e V: The 6 6 straindisplacement submatrices e Bi i1;2;; e p for an FSDT
element derived from the p-order HSDT formulation with

20a



hi e 21 a hi
Jj2
j3 e Jj321 a hi
j 1
j 2

e 21
Jj1

j 1;2; 3; k 1; 2; 3; 1;2

20b

Assuming a linear constitutive law in each layer, the stress


vector for the k-th lamina in the e-th FSDT or HSDT shell
finite element with e q plies is given by
e

shk j;t e Ck e eh j; t e Ck e Bje qt

21

k 1; 2; ; e q
where e Ck denotes the global 6 6 elasticity matrix for the
k-th ply. In order to align the global axes with the material
axes, e Ck must be expressed in the shell co-ordinate system
e



j 1; 2; 3

 keQ
Ck e QTe C

22

 k is the local elasticity matrix for the k-th lamina


where e C
and e Q denotes the 6 6 matrix which transforms the
stress strain law from the global to the local co-ordinate
 k can further be written
system. The shell-aligned matrix e C
in terms of the layer elastic properties as
e

 k e Tk Te Dk e Tk
C

23

where e Tk represents the 6 6 transformation matrix from


the shell to the k-th layer co-ordinate system through the socalled lamination angle of the ply, whereas e Dk is

982

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

the elasticity matrix written in the laminate or material coordinate system.


2.3. Structural matrices
The mass e M and stiffness e K matrices for the e-th FSDT
or HSDT shell finite element e V is obtained classically as

e
e a Ta e a
M
r H H j dV
24a
aV

e
aV

BTe Ce Be j a dV

24b

where e r denotes the mass density, e C constitutes the global


elasticity matrix, a V represents the volume of the
parent domain, and e j is the jacobian of the coordinate transformation from the parent domain to the e-th
element (Fig. 3-4). In these expressions, a H is the interpolation matrix (7a), whose components are given by Eq. (8),
(11a) or (13a) for a p-order HSDT element, a first-order
HSDT element or an FSDT element respectively. Similarly,
e
B is the strain-displacement matrix (16), the components of
which are obtained from Eq. (17) or (19) according as the
model is based on the p-order (p 1 included) HSDT
approach or the FSDT formulation respectively. It should be
observed that the mass and stiffness matrices derived from
the p-order HSDT model are hierarchical, since the
interpolation and strain-displacement matrices for this
formulation type are hierarchical in nature.
For a layered FSDT or HSDT shell, the components of the
structural matrices (24) are calculated per lamina, and the
contributions over the e q plies are then simply assembled as
e

e
q
X

k1
e

(25) become
e
aV
k

e
q
X

k1
q
X

25a
e

e
aV
k

rk aHTa He j a dVk

k1

aV
k

BTe Ck e Be j a dVk
e

e Tk e Qe BTe Dk e Tk e Qe Be j a dVk

i e i
uk e t3k
= t3

i 1; 2; ; e p; k 1; 2; ; e q

25b

26
a

is constant per layer, and mapping the k-th parent layer Vk


into the normalized domain Vk (Fig. 5) with a bi-unit
thickness of variable z21 # z # 1; structural matrices

Vk

e
q
X

k1

where a Vk constitutes the image in the parent domain a V of


the k-th lamina from the e-th element e V (Fig. 5) and e rk is the
mass density of the k-th layer, whereas e Ck and e Dk again
represent the global and local elasticity matrices for the k-th
ply, e Tk and e Q being the aforementioned transformation
matrices from the k-th layer co-ordinate system to the shellaligned axes and finally to the global reference.
Assuming for the sake of simplicity that the ratio e uk
i
between the thickness e t3k
of the k-th ply at node i and the
e i
element thickness t3 at the same nodal point
e

e
q
X

k1

Fig. 5. Normalization of the domain Vk into the k-th layer a Vk of an HSDT


shell element.

Vk

e e
e
e STe
k Dk Sk jlj3 j3 z uk dVk

27a

e e
e
e STe
k Dk Sk jlj3 j3 z uk dVk

27b

where e Sk e Tk e Qe B is the local fibre-oriented straindisplacement matrix for the k-th ply and where the natural coordinate j3 is related to variable z by the following expression

j3 21 2

kX
21

uj 1 ze uk

k 1; 2; ;e q

28

j1

which holds for every nodal point since the ratio between the
k-th lamina depth and the total shell element thickness is
assumed to be identical for each node within a given element.
2.4. Numerical integration
As each layer Vk k 1; 2; ; e q of the composite FSDT
or HSDT shell element is mapped into a normalized parent
domain, a Gauss Legendre numerical integration can be
efficiently used for evaluating the components of

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

the corresponding contributions to the stiffness matrix with


a selective quadrature scheme.

Table 1
Single-layer material properties
Property

E1 =E2
()

E3 =E2
( )

G23 =E2
()

G31 =E2 G12 =E2


()

n12 n13 n23


()

Value

40

0.6

0.5

0.25

3. Closed-form, experimental and numerical model


validation

the structural matrices. For an hexahedral element, the


integrals (27) are calculated as
q
X
e

k1

Vk

lj3 j3 z e uk dVk

q X
r X
s X
t
X
k1 1 m1 n1

983

n ;
v1 vm2 vne
3 uk lj1 j1 ; j2 jm
2 ; j3 j3 z

29

The case studies considered to validate the described sets


of composite shell elements are modal analyses of thin to
thick multilayered square or rectangular freely or simply
supported plates. The aim of these examples is to highlight
the ability of the FSDT and HSDT finite element models to
accurately predict the vibration behaviour of conventional
laminated plates and shells with different thicknesses.
3.1. Validation with closed-form solutions

j1 1;2; ;r; jm2 m 1; 2; ; s

where
and j3 z n
1; 2;; t are the abscissae of the r;s and t Gauss points per
lamina in the j1 -; j2 - and j3 -directions, respectively, whereas
v1 1; 2;; r; vm2 m 1; 2; ; s and vn3 n 1; 2;; t
represent the corresponding Gaussian weights, the total
number of quadrature points per element being thus equal to
r s t e q: For a prismatic element, the numerical integration in the j1 - and j2 -directions is as usually performed
simultaneously in order to obtain an optimal quadrature
strategy, and the integration through the thickness (j3 direction) is carried out as for the hexahedral element.
It should be pointed out that the number of Gauss points
needed per layer for an exact integration in the j3 -direction
depends on the order of the polynomial approximation
chosen for the through-the-thickness displacement. For the
FSDT shell model, two integration points are thus sufficient,
but up to p 1 points are necessary with a p-order HSDT
formulation. It can be shown that despite the increased
number of integration pointsespecially with a high-order
HSDT modelin comparison to explicit through-thethickness integration schemes, an exact quadrature per ply
leads to results which are more accurate than with explicit
methods. Furthermore, with this standard integration
strategy, the shear locking phenomenon inherent in thin
shells can be circumvented as in the classical shell
formulation by splitting the strain-displacement matrix
into normal and shear components and by integrating

As a first example, the FSDT and HSDT formulations are


compared to several closed-form solutions available from
the literature [14]. In this test case, the natural frequency
analysis of a simply supported four-layered square plate is
considered.
The structure analysed is a (08/908)s laminate with mass
density r; side a (aspect ratio of 1) and thickness h; the spanto-thickness factor varying from 4 to 100. The single-ply
material properties are known in terms of ratios and given in
Table 1, where E1 ; E2 and E3 are the Youngs moduli in the
principal in-plane and normal directions, G23 ; G31 and G12
are the shear moduli, and n23 ; n31 and n12 are the associated
Poissons ratios. The FSDT and HSDT finite element meshes
are characterized by 15 15 9-node quadratic composite
shell elements based on a selective integration scheme in the
in-plane directions. A transverse shear correction factor of
0.833 is adopted for the FSDT shell model.
Expressed in a non-dimensional form, the predicted
fundamental bending natural frequency v of the plate is
compared in Table 2 to four closed-form solutions based on
the first- and third-order layerwise plate theory (linear
and cubic displacement approximation per layer) and on
the first- and third-order ESL formulations including the socalled zig zag terms. The relative errors on the numerical
FSDT and HSDT simulations in comparison to the thirdorder layerwise theory considered as reference are shown

Table 2
p
Nondimensional natural frequency v a4 r=E2 h2 of the square plate
a=h

4
10
20
100
a
b

Numerical results ( )

Closed-form solutions ( ) [14]

FSDT

HSDT p 1

HSDT p 3

HSDT p 5

HSDT p 7

LW1a

LW3a

ESL1b

ESL3b

9.388
15.13
17.65
18.82

9.960
15.57
17.83
18.83

9.497
15.26
17.66
18.75

9.394
15.20
17.64
18.75

9.388
15.19
17.64
18.75

9.473
15.34
17.70
18.76

9.224
15.15
17.63
18.75

9.595
15.57
17.83
18.83

9.394
15.27
17.72
18.83

First- and third order layerwise theory.


First- and third order ESL theory with zig-zag term.

984

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987


Table 3
Glass/polypropylene single-layer material properties

Fig. 6. Relative errors on the fundamental natural frequency.

graphically in Fig. 6. It can be seen that in the full range of


the span-to-thickness ratio the predictions derived from the
HSDT models of the third or higher order are in good
agreement with the reference values and more accurate than
the other closed-form solutions when p . 3: However, the
precision of the results obtained with the FSDT and firstorder HSDT formulations is smaller, especially when the
plate is relatively thick. It should also be mentioned that if
an equivalent finite element model is derived from the firstor third-order layerwise plate theory, the total number of
degrees of freedom needed for this model is in general much
higher than for the HSDT formulations, since the layerwise
models require at least two degrees of freedom per layer and
per nodal point.
3.2. Validation with experimental results
The second test case investigated to validate the FSDT
and HSDT elements is devoted to the experimental and
numerical modal analysis of a thin free-free multilayered
rectangular plate.
The specimen adopted for the experimental tests is an inhouse manufactured glass/polypropylene laminate produced
using a compression-moulding technique. Supplied by
Vetrotex, France, the Twintex prepegs used for fabrication
are based on a 4 1 fibre weave (ratio of the number of fibres
in the weft and warp directions) and are 0.625 mm thick.
The plate is a 290 234 5 mm3 8-ply (908/08)2s laminate,
thus characterized by an aspect ratio of 1.24 and a span-tothickness factor of about 60, which corresponds to a rather
thin plate. Usual quality checks are carried out on the
specimen to verify that the porosity level and the thickness
uniformity are acceptable. Single-layer material constants
are determined for the glass/polypropylene laminae by
means of standard static tensile and bending tests, as well as

Property

E1
(GPa)

E2
(GPa)

G23
(GPa)

G31 G12
(GPa)

n23
()

n31
()

n12
( )

Value

23.5

9.4

1.8

2.5

0.38

0.14

0.08

a modal numerical/experimental identification [15]. The


estimated single-layer constitutive values are listed in
Table 3, where it should be observed that the uncertainty
on the given shear moduli G23 ; G31 and G12 is larger than for
the Youngs moduli E1 and E2 n12 denotes Poissons ratio
for the transverse strain). The mass density for the laminate
is taken as 1463 kg/m3.
Experimental modal analysis is performed on the specimen by means of a contact-free measurement system. The
equipment is constituted by a controlled pseudo-random and
periodic chirp signal generator, an acoustic system with
filters, power amplifiers and one-way 150 mm loudspeaker
to generate a contact-free excitation of the specimen (the
reference signal for the excitation is measured with a B &
K4134 microphone), and a Polytec PSV200 laser scanning
vibrometer to reconstruct the FFT functions (grid of 15 13
measurement points) and to visualize the mode shapes
(Fig. 7). Quasi free boundary conditions are considered (the
plate is suspended at two points with a thin nylon yarn).
The numerical simulations are carried out with the
described sets of FSDT and HSDT (third and fifth order)
shell elements implemented in an in-house finite element
software developed for structural identification [16]. The
plate is modelled with 8 8 8-node quadratic finite
elements based on a selective quadrature scheme for
the integration of the structural components in the transverse
directions. The single-layer material constants given in
Table 3 are assigned to each lamina, as well as the
appropriate ply orientation and thickness.

Fig. 7. Experimental set-up.

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987


Table 4
Nonzero bending frequencies of the thin multilayered plate
Mode

1
2
3
4
5
6
7

Numerical results (Hz)

Table 5
Carbon/epoxy single-layer material properties
Experimental
results (Hz)

FSDT

HSDT p 3

HSDT p 5

101.5
187.3
277.0
338.3
393.1
515.6
562.4

101.5
189.4
278.3
340.9
395.3
520.9
563.8

101.5
188.8
277.9
340.2
394.7
519.5
563.5

985

101.7
190.1
273.9
340.9
389.7
524.5
558.2

The lowest seven nonzero natural frequencies (below


600 Hz) are then extracted numerically from the finite
element models and experimentally from the measurements
(the rigid body modes introduced by the free boundary
conditions are not considered in the analysis). The
numerical and experimental frequencies are reported in
Table 4 and the relative errors between the different finite
element simulations and the test results represented in Fig. 8.
It can be seen that the predictions derived from all the
numerical models agree very well with the experimental
values and that the increase in accuracy when using HSDT
elements in thin shell situations is not really convincing.
Since the error distribution seems to be independent of
the element type, the uncertainties on some single-layer
constitutive properties probably account for the discrepancy
between the numerical and experimental results.
3.3. Validation with other numerical simulations
In a last validation study, the accuracy of the FSDT and
HSDT formulations is compared to the precision of an
equivalent 3D model through the purely numerical modal
analysis of a thick free-free multilayered rectangular plate.
The aim of this example is to highlight the superiority of

Property

E1
(GPa)

E2 E3
(GPa)

G23
(GPa)

G31 G12
(GPa)

n23
()

n31 n12
()

Value

114

8.0

2.9

3.1

0.44

0.29

the p-order approach over the first-order model when the


thickness-to-span ratio of the shell becomes large.
The plate considered here is a rectangular carbon/epoxy
150 100 40 mm3 8-ply (08/908)2s laminate with a fibre
volume fraction of 60%. The span-to-depth and span-tothickness ratios of the plate are thus equal to 1.5 and 3.75,
respectively, which corresponds to a rather thick laminated
structure. The single-layer material properties chosen are
given in Table 5, where E1 ; E2 and E3 are the Youngs
moduli in the principal in-plane and normal directions,
G23 ; G31 and G12 are the shear moduli, and n23 ; n31 and n12
are the associated Poissons ratios. The mass density for the
laminate is taken as 1480 kg/m3 and a transverse shear
correction factor of 0.833 is adopted for the FSDT shell
formulation (this factor is unused in the 3D and p-order
models). The FSDT and HSDT finite element meshes are
characterized by 15 10 8-node quadratic composite shell
elements based on a selective integration scheme in the
transverse directions, whereas the 3D discretization used as
reference is formed by 20 13 16 20-node quadratic
Abaqusw C3D20R-type solid elements based on a reduced
quadrature strategy. The number of nodal points is thus
equal to 501 for the shell models and 19103 for the 3D
formulation, so that the total number of degrees of freedom
amounts to 2505 for the FSDT mesh (5 unknowns per node),
3006 for a first-order HSDT model (6 unknowns per node),
12024 for a 7th-order HSDT formulation (24 unknowns per
node) or 57309 for the 3D discretization (3 degrees of
freedom per nodal point).
The first nine bending frequencies (below 15000 Hz) of
the plate under free boundary conditions are next extracted
by means of the aforementioned in-house finite element
software (FSDT and HSDT shell models) and the Abaqusw
computer code (3D solid model). The results are shown in
Table 6
Nonzero bending frequencies of the thick multilayered plate
Mode

1
2
3
4
5
6
7
8
9
Fig. 8. Relative errors on the computed natural frequencies for the thin
plate.

Numerical results (Hz)


FSDT

HSDT
p 1

HSDT
p 3

HSDT
p 5

HSDT
p 7

3D

3045
7104
7819
10726
11173a
11026a
11804
13247
14944

3045
7113
7824
10780
11219a
11030a
11807
13288
14961

3053
7333
8016
10259
10772
11560
12261
13078
14532

3052
7314
7997
10190
10705
11496
12198
13002
14430

3051
7306
7992
10140
10664
11482
12187
12968
14358

3051
7304
7990
10122
10648
11482
12184
12961
14343

Permutations in the mode shape numbering.

986

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

Fig. 11. Seventh mode shape of the thick plate (3D model).

Fig. 9. Relative errors on the computed natural frequencies for the thick
plate.

Table 6 and the relative errors between the predictions


obtained with the 3D model and the values derived from the
FSDT formulation and the HSDT models up to the 7th order
are given in Fig. 9. It can be observed that the results
obtained for the HSDT models of the third or higher order
are in excellent agreement with the 3D predictions, whereas
the accuracy of the frequencies extracted from the FSDT

and first-order HSDT models is reduced by the coarse


approximation of the transverse shear deformations in thick
plates or shells. The small differences between the results
found for the FSDT formulation and the first-order HSDT
model can be attributed to the use of a shear correction factor
in the former model. In order to measure the global
convergence of the various shell formulations, the average
relative error on the first 15 natural bending frequencies of
the plate is represented in function of the HSDT order in
Fig. 10, where the corresponding values for 10 and 20 mm
thick plates are also given. It can be seen that a drastic
improvement of the accuracy is obtained when increasing the
order of the HSDT model. This is easily explained by the
highly non-linear through-the-thickness displacement field,
as shown in Fig. 11 for the 7th mode shape (3D discretization). Finally, it should be mentioned that the associated
mode shapes for the third and higher-order HSDT models
perfectly coincide with the corresponding 3D configurations,
whilst differences between the 3D mode shapes and the ones
associated to the FSDT model are clearly observable.

4. Conclusions

Fig. 10. Average relative errors on the lowest 15 frequencies in function of


the HSDT order.

Closely related to the standard C0 -compatible shell


approach, two FSDT and HSDT multilayered composite
finite element formulations have been presented in this
paper. Based on a through-the-thickness displacement
which is approximated by a first- or higher-order polynomial respectively, these elements have been validated
with closed-form solutions based on the layerwise and socalled zig zag plate theories, an experimental modal test on
a glass/polypropylene specimen and a numerical modal
simulation on a carbon/epoxy 3D model derived from a
well-known commercial software. Good agreement
between the analytical, measured or 3D-computed frequencies and the FSDT or HSDT predictions has been obtained.
The FSDT model has, however, shown an accuracy
deterioration with an increase of the shell thickness-tospan ratio, while an excellent convergence for the HSDT

J. Cugnoni et al. / Composites: Part A 35 (2004) 977987

formulation has been observed when augmenting the order


of the normal displacement approximation.

Acknowledgements
This work was partially supported by the Swiss National
Science Foundation (FNS), grant No. 2000-068103.02.

References
[1] Panda S, Natarajan R. Analysis of laminated composite shell structures
by finite element method. Comp Struct 1981;14(3-4):22530.
[2] Reddy JN. Bending of laminated anisotropic shells by a shear
deformable finite element. Fibre Sc Tech 1982;17:924.
[3] Yunus SM, Kohnke PC, Saigal S. An efficient through-thickness
integration scheme in an unlimited layer doubly curved isoparametric
composite shell element. Int J Num Meth Engng 1989;28(12):277793.
[4] Wung PM. Laminated composite structures by continuum-based shell
elements with transverse deformation. Comp Struct 1997;62(6):
107390.
[5] Reddy JN. Mechanics of laminated composite plates: theory and
analysis. Boca Raton: CRC Press; 1997.
[6] Sotiropoulos SN, GangaRao HVS, Mongi ANK. Theoretical and
experimental evaluation of FRP components and systems. ASCE J
Struct Engng 1994;120(2):464 85.
[7] Ayorinde EO. Elastic constants of thick orthotropic composite plates.
J Compos Mater 1995;29(8):1025 39.

987

[8] Anderson TJ, Nayfeh AH. Natural frequencies and mode shapes of
laminated composite plates: experiments and FEA. J Vibr Control
1996;2(4):381414.
[9] Araujo AL, Mota Soares CM, Moreira de Freitas MJ. Characterization
of material parameters of composite plate specimens using optimization and experimental vibrational data. Comp Part B 1996;27(2):
18591.
[10] De Visscher J, Sol H, De Wilde WP, Vantomme J. Identification of the
damping properties of orthotropic composite materials using a mixed
numerical experimental method. Appl Compos Mater 1997;4(1):
1333.
[11] Frederiksen PS. Experimental procedure and results for the identification of elastic constants of thick orthotropic plates. J Compos Mater
1997;31(4):36082.
[12] Surana KS, Sorem RM. p-Approximation curved shell elements for
elastostatic analysis of laminated composite plates and shells. In:
Proceedings of the ASME Pressure Vessels and Piping Conference.
Honolulu, July; 1989. pp. 185 99.
[13] Cugnoni J, Gmur T, Schorderet A. Modal validation of a set of C 0 compatible composite shell finite elements. Compos Sci and Technol
2003; submitted for publication.
[14] Carrera E. An assessment of mixed and classical theories on global
and local response of multilayered orthotropic plates. Compos Struct
2000;50(2):18398.
[15] Bosia F, Gmur T, Botsis J. Deformation characteristics of
composite laminatesPart II: an experimental/numerical study on
equivalent single-layer theories. Compos Sci Technol 2002;62:
5566.
[16] Gmur T, Schorderet A. MAFEA Code for Modal Analysis by Finite
Elements (Including Substructuring): UsersManual. Lausanne: Ecole
polytechnique federale de Lausanne; 2002.

You might also like