ARTICLE IN PRESS

Engineering Analysis with Boundary Elements 31 (2007) 783–792

www.elsevier.com/locate/enganabound

Geometrically nonlinear analysis of cylindrical shells using

the element-free kp-Ritz method
X. Zhaoa, Y. Yangb, K.M. Liewc,
a
School of Mechanical and Aerospace Engineering, Nanyang Technological University, Nanyang Avenue, Singapore 639798, Singapore
b
Department of Engineering Mechanics, Kunming University of Science and Technology, Kunming, Yunnan, China
c
Department of Building and Construction, City University of Hong Kong, Tat Chee Street, Kowloon, Hong Kong, China
Received 12 December 2005; accepted 10 January 2007
Available online 23 March 2007

Abstract

In this paper, the geometrically nonlinear analysis of cylindrical shells is carried out using the element-free kp-Ritz method. The first-
order shear deformation shell theory, which can cater for both thin and relatively thick shells, is utilized in the present study. Meshfree
kernel particle functions are employed to approximate the two-dimensional displacement field. The nonlinear equilibrium equations are
formulated by applying the Ritz procedure to the energy functional of shells. The Newton–Raphson method and the arc length technique
are used to determine the load–displacement path. To validate the accuracy and stability of this method, convergence studies based on
the support size and number of nodes were performed. Comparisons were also made with the existing results available in the open
literature, and good agreement is obtained.
r 2007 Elsevier Ltd. All rights reserved.

Keywords: Geometrically nonlinear analysis; Cylindrical shells; kp-Ritz method

1. Introduction Hinton [6], and Hughes and Liu [7,8], and the SHEBA
family of finite elements by Argyris et al. [9,10]. Recently,
Extensive research has been performed to investigate the Choi and Paik [11] developed a four-node degenerated shell
response of shell structures to external loads. For practical element for the analysis of shell structures that are
purposes, the geometrical nonlinear analysis of shells, in undergoing a large deformation. A hybrid stress nine-node
which large deformations are involved, is usually carried degenerated shell element was presented by Sze and Zheng
out. The finite element method (FEM) is so far the most [12] for geometrically nonlinear analysis, and an eight-node
effective way to study the shell problems. Among the hybrid stress solid-shell element was developed by Sze et al.
earliest researchers, Galleagher [1] developed a 24 degree of [13]. Other new elements that have been presented by
freedom, doubly curved, thin shell finite element, which researchers include the curved quadratic triangular degen-
was used to analyze shell buckling problems [2]. Later, a erated and solid-shell elements of Kim et al. [14], the four-
summary on progress of doubly curved thin shell finite node and nine-node finite elements of Sansour and
element analysis and a formulation for a triangular thin Kollmann [15], and the triangular finite shell element of
shell element were presented by Gallagher and Ashwell [3]. Campello et al. [16].
Horrigmoe and Bergan [4] conducted a nonlinear analysis As for other methods of shell analysis, Dennis [17]
of free-form shells using flat shell elements. The earlier shell provided a Galerkin solution to laminated shallow shell
elements used in nonlinear analysis include the degenerated equations that included a parabolic transverse shear
shell element by Hughes and Cornoy [5], Huang and deformation, an assumption that the distribution of the
transverse shear strains through the thickness is parabolic.
Corresponding author. Tel.: +852 3442 6581; fax: +852 2788 7612. Krysl and Belytschko [18] applied the element-free
E-mail address: [email protected] (K.M. Liew). Galerkin (EFG) method to the analysis of arbitrary

0955-7997/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.enganabound.2007.01.003
 ARTICLE IN PRESS
784 X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792

Kirchoff shells, and the EFG was extended by Noguchi

et al. [19] to solve three-dimensional (3-D) general shell L
and spatial structures. Li et al. [20] carried out a numeri-
cal simulation of the large deformation of thin struc-
tures using meshfree methods. Boundary element method z
x
(BEM) was employed by Wen et al. [21] to study the h
stiffened shear-deformable shells. Haas and Kuhn [22] θ
R
presented a mixed-dimensional coupling FEM and
θ0
BEM shell element. Liu [23] applied the BEM for 3-D
elastostatic problems to the analysis of shell or shell-like
structures.
Ritz methods are widely used in engineering problems
such as vibration and buckling analysis. Some noteworthy w
studies in this area include those of Kitipornchai et al. [24],
Liew et al. [25,26], Cheung and Zhou [27], and Reddy and
Savoia [28]. v
Recently, Liew et al. [29] and Zhao et al. [30] presented a
new kp-Ritz method for the vibration analysis of cylind-
rical and conical shell panels. In this paper, the kp- h2
Ritz method is extended to investigate the nonlinear h1

h4
analysis of shells under external loads. The displace- h

h5
ment field is approximated by two-dimensional (2-D)
kernel particle estimates, and the geometrically nonli-
near equilibrium equations are obtained by applying Fig. 1. (a) Coordinate system of the cylindrical panel. (b) Cross-sectional
the kp-Ritz method to the energy functional of shells. view of the laminated cylindrical panel.
The equilibrium path is determined by using the
modified Newton–Raphson method combined with the
arc length technique, which was originally proposed According to the first-order shear deformation theory
by Risk [31] and Wempner [32] and alternated by Cris- (FSDT), the displacement field is expressed as
field [33], to handle the snap-through and snap-back
uðx; y; zÞ ¼ u0 ðx; yÞ þ zbx ðx; yÞ,
phenomena.
The shape functions in the present method are con- vðx; y; zÞ ¼ v0 ðx; yÞ þ zby ðx; yÞ,
structed entirely in terms of discrete nodes, mesh genera- wðx; y; zÞ ¼ w0 ðx; yÞ, ð1Þ
tion and mesh distorsion are avoided. Therefore, it is more
where u0, v0 and w0 denote the displacements of the mid-
flexible than FEM. In conventional Ritz method, it is
surface of the shell in the x, y, and z directions. bx and by
difficult to choose the appropriate trial functions for
are the rotations of the transverse normal about the y and x
problems with complicated boundary conditions, the
axes, respectively.
present approach overcomes this shortcoming by using
The nonlinear strain–displacement equations of Sanders
only one shape function type to describe the interior
nonlinear theory [34] were modified by Reddy [35] as
domain. The boundary conditions can be imposed easily. 8 9
>  ( )
Numerical examples are provided to verify the present < xx > = gyz
method, and good agreement is attained with the results yy ¼ e0 þ zj; ¼ c0 , (2)
available in the open literature. :g >
> ; gxz
xy

where
e0 ¼ e0L þ e0NL ,
2. Nonlinear formulation of shells 8 qu 9 8  2 9
> 0
> >
> 1 qw0 >
>
>
> qx >
> >
> 2 qx >
>
2.1. The energy formulation of shells < = <   =
1 qv0
e0L ¼ R qy R ; þ w0
e0NL ¼ 1 1 qw0 2 ,
>
> >
> >
>
2 R qy >
>
A part of the cylindrical shell is shown in Fig. 1a, where : 1 qu0 þ qv0 >
> ; >
> >
>
R qy qx
:1 qw0 qw0 ;
R qx qy
a coordinate system (x, y, z) is fixed on the middle surface
8 qbx 9
of the shell panel. This panel is considered to be thin and of >
> qx >
> 8 9
a length L, radius R, and thickness h, and is bounded along >
< >
= < by þ R1 qw0
 vR0 =
1 qby qy
its edges by the lines x ¼ 0, x ¼ L, y ¼ 0, and y ¼ y0. The j ¼ R qy ; c0 ¼ . ð3Þ
>
> >
> : bx þ qw0 ;
displacements of the panel in the x, y, and z directions are >
: 1 qbx þ qby ;> qx
denoted by u, v, and w, respectively. R qy qx
 ARTICLE IN PRESS
X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792 785

k
The strain energy of the shell is expressed by total number of layers in the laminated panel, and Q̄ij are
Z Z the transformed reduced stiffness coeffieients for the kth
1 L y0 T
U ¼ e S e R dy dx, (4) layer and defined as
2 0 0
½Q̄ ¼ ½T1 ½Q½TT , (10)
where e is given by
8 9 8 9 where ½T T is the transpose of matrix ½T 1 , [T] is the
e
< 0L >
> = >< e0NL >
= transformation matrix for the principle material coordi-
e ¼ eL þ eNL ¼ j þ 0 (5) nates and the panel’s coordinates, and is defined as
:c >
> ; >: >
; 2 3
0 0 cos2 a sin2 a 2 sin a cos a 0 0
6 7
in which eL and eNL represent the linear and nonlinear 6 sin2 a cos2 a 2 sin a cos a 0 0 7
6 7
6
½T  ¼ 6  sin a cos a sin a cos a cos2 a  sin2 a 0 0 7,
parts of the strain at the middle surface of the shell, 6
7
7
respectively. 4 0 0 0 cos a  sin a 5
The stiffness matrix S is defined as 0 0 0 sin a cos a
2 3 (11)
A11 A12 A16 B11 B12 B16 0 0
6 7 where a is the angular orientation of the fibers and [Q] is
6 A12 A22 A26 B12 B22 B26 0 0 7
6 7 the reduced stiffness matrix, which, in turn, is defined as
6 A16 A26 A66 B16 B26 B66 0 0 7
6 7 2 3
6 7 Q11 Q12 0 0 0
6 B11 B12 B16 D11 D12 D16 0 0 7
S¼6 6 B12 B22 B26 D12 D22 D26 0
7, (6) 6Q
6 12 Q22 0 0 0 7
6 0 77 6
7
7
6 7 6
½Q ¼ 6 0 0 Q 0 0 7. (12)
6 B16 B26 B66 D16 D26 D66 0 0 7 66 7
6 7 6 0 0 0 Q44 0 7
6 0 0 0 0 0 0 S44 S 45 7 4 5
4 5
0 0 0 0 0 0 S45 S 55 0 0 0 0 Q55

where the extensional Aij, coupling Bij, bending Dij, and The external work due to the body force and surface
transverse shear Sij stiffnesses are defined as traction is given by
Z h=2 Z h=2 Z L Z y0 Z
ðAij ; Bij ; Dij Þ ¼ Qij ð1; z; z2 Þ dz; S ij ¼ K Qij dz. We ¼ u f̄R dy dx þ uT t̄ dG,
T
(13)
0 0 G
h=2 h=2
(7) where f̄ and t̄ represent the external load and prescribed
traction on the natural boundary, u is the displacement
The stiffness Aij, Bij, and Dij are defined for i, j ¼ 1, 2, 6 vector.
whereas Sij is defined for i, j ¼ 4, 5. K denotes the The total potential energy of the shell panel is expressed
transverse shear correction coefficient, which is computed as
such that the strain energy due to the transverse shear
stresses equals the strain energy due to the true transverse P ¼ U   W e. (14)
stresses predicted by the 3-D elasticity theory [35]. Qij are
the engineering constants, which are defined as
E 11 n12 E 22 E 22 2.2. Two-dimensional kernel particle shape functions
Q11 ¼ ; Q12 ¼ ; Q22 ¼ ,
1  n12 n21 1  n12 n21 1  n12 n21
For a shell panel discretized by a set of nodes xI,
Q66 ¼ G 12 ; Q44 ¼ G 23 ; Q55 ¼ G 13 , ð8Þ
I ¼ 1; . . . ; NP, the discrete displacement approximations
where E11 and E22 are the elastic moduli in the principle take the form
material coordinates; G12, G13, and G23 are the shear 0 h1 0 1
u0 uI
moduli; and n12 and n21 are Poisson’s ratio. B hC
B v0 C B v C
For a panel that is composed of different layers of B C X NP B I C X NP
B h C B C
orthotropic materials, the stiffnesses Aij, Bij, and Dij, can be h w
u0 ¼ B 0 C ¼ cI ðxÞB wI C ¼ cI ðxÞuI , (15)
B h C I¼1 B C
defined as B bx C B b C I¼1
@ A @ xI A
X
Nl
k 1X
N l
k bhy byI
Aij ¼ Q̄ij ðhk  hkþ1 Þ; Bij ¼ Q̄ ðh2  h2kþ1 Þ,
k¼1
2 k¼1 ij k
where x  ½x; y, cI ðxÞ and uI are the shape function and
1X Nl
k 5X Nl
k nodal parameter associated with node I.
Dij ¼ Q̄ij ðh3k  h3kþ1 Þ; Sij ¼ Q̄ij ðhk  hk1 Þ, ð9Þ The 2-D shape function is expressed as (see Chen et al. [36])
3 k¼1 6 k¼1
where hk and hk+1 denote the distances from the panel cI ðxÞ ¼ Cðx; x  xI ÞFa ðx  xI Þ, (16)
reference surface to the outer and inner surfaces, respec- where Cðx; x  xI Þ is the correction function and Fa ðx  xI Þ
tively, of the kth layer, as shown in Fig. 1b. Nl denotes the is the kernel function. The correction function Cðx; x  xI Þ is
 ARTICLE IN PRESS
786 X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792

described as follows: The first derivatives of the shape function need to be

determined from these equations. Eq. (20) can be rewritten
Cðx; x  xI Þ ¼ H T ðx  xI ÞbðxÞ,
as
bðxÞ ¼ ½b0 ðx; yÞ; b1 ðx; yÞ; b2 ðx; yÞ; b3 ðx; yÞ; b4 ðx; yÞ; b5 ðx; yÞT ,
MðxÞbðxÞ ¼ Hð0Þ. (28)
Hðx  xI Þ ¼ ½1; x  xI ; y  yI ; ðx  xI Þðy  yI Þ; ðx  xI Þ2 ,
The vector b(x) can be determined by the LU decom-
ðy  yI Þ2 T , ð17Þ
position of the matrix M(x) followed by backward
where H is a vector of the quadratic basis and bi ðx; yÞ are the substitution, and the derivatives of b(x) can be similarly
functions of x and y that are to be determined. Thus, the obtained. For instance, taking the derivative of Eq. (28) to
shape function can be written as x, we have
cI ðxÞ ¼ bT ðxÞHðx  xI ÞFa ðx  xI Þ. (18) M ;x ðxÞbðxÞ þ MðxÞb;x ðxÞ ¼ 0, (29)
Eq. (18) can be rewritten as which can be rearranged as
T
cI ðxÞ ¼ b ðxÞBI ðx  xI Þ, (19)
MðxÞb;x ðxÞ ¼ M ;x ðxÞbðxÞ. (30)
where
It can be seen that the first derivative of b(x) can be
bðxÞ ¼ M 1 ðxÞHð0Þ, (20) formulated using the LU decomposition procedure again.
The first derivative of the shape function can be obtained
B I ðx  xI Þ ¼ Hðx  xI ÞFa ðx  xI Þ, (21) by taking the derivative of Eq. (19)
the moment matrix M is a function of x, and H(0) is a
constant vector. The expressions for M and H(0) are given by cI;x ðxÞ ¼ bT;x ðxÞBI ðx  xI Þ þ bT ðxÞB I;x ðx  xI Þ. (31)

X
NP
MðxÞ ¼ Hðx  xI ÞH T ðx  xÞFa ðx  xI Þ,
I¼1

2.3. Nonlinear equilibrium equation formulation

Hð0Þ ¼ ½1; 0; 0; 0; 0; 0T . (22)
For this 2-D problem, the kernel function Fa ðx  xI Þ is It is necessary to correctly enforce the different boundary
expressed as conditions to accurately solve the problem. There are
Fa ðx  xI Þ ¼ Fa ðxÞFa ðyÞ (23) several approaches to enforcing essential boundary condi-
tions, such as the Lagrange multiplier approach and
in which modified variational principles. In this study, the penalty
1  x  xI  method is utilized to implement the essential boundary
Fa ðx  xI Þ ¼ f , (24) conditions. The penalty formulation is developed as
a a
follows.
where fðxÞ is the weight function. The cubic spline function is
For the domain that is bounded by lu, the displacement
chosen as the weight function, and is given by
82 9 boundary condition is
> 2
< 3  4zI þ 4zI
3
for 0pjzI jp 12 >
= u ¼ ū on l u , (32)
fz ðzI Þ ¼ 43  4zI þ 4z2I  43 z3I for 1
2 pjzI jp1 ,
>
: >
;
0 otherwise where ū is the prescribed displacement of the displacement
boundary lu. The variational form is given by
Z
ðx  xI Þ a
zI ¼ , (25) Pū ¼ ðu  ūÞT ðu  ūÞ dl, (33)
d 2 lu

where the dilatation parameter d is the size of the support. where a is the penalty parameter that is taken to be 103E11,
The size of the domain of influence at a node is calculated by with E11 being the elastic modulus of the shell in the
principal coordinate direction.
d I ¼ d max aI , (26) Therefore, the total energy functional for this problem is
where dmax is a scaling factor that ranges from 2.0 to 4.0. The Pt ¼ P þ Pū . (34)
distance aI is determined by searching for enough nodes to
avoid the singularity of the matrix M. For instance, the The strain energy of the shell in Eq. (4) can be expressed
number of nodes should not be less than 6 for the quadratic as two parts: the linear strain energy UL and the nonlinear
basis. Otherwise, the matrix M cannot be invertible. strain energy UNL. Substituting Eq. (5) into Eq. (4) yields
The shape function can therefore be expressed as Z Z
1 L y0
U ¼ ðeL þ eNL ÞT SðeL þ eNL ÞR dy dx, (35)
cI ðxÞ ¼ H T ð0ÞM 1 ðxÞHðx  xI ÞFa ðx  xI Þ. (27) 2 0 0
 ARTICLE IN PRESS
X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792 787

2 qcI 3
the energy expression can be expanded to 0 0 0 0
qx
Z Z 6 1 qcI cI 7
1 L y0 T BLe 6 0 0 07
U ¼ e S eL R dy dx I ¼4 R qy R 5, (46)
2 0 0 L 1 qcI qcI
0 0 0
Z Z R qy qx
1 L y0  T 
þ eL S eNL þ eTNL S eL þ eTNL S eNL R dy dx 2 3
2 0 0 qcI
0 0 0 qx 0
¼ U L þ U NL . ð36Þ 6 7
60 0 0 1 qcI 7
BLm
I ¼ 6 0 R qy 7,
By substituting the displacement functions of Eq. (15) 4 5
1 qcI qcI
into the total energy functional Eq. (34) and applying the 0 0 0 R qy qx
Ritz procedure to the potential energy expression, we 2 3
cI 0 0 0 0
obtain 6 7
n o 6 0 cI 0 0 0 7
qP 6 7
¼ 0; uI ¼ uI vI wI bxI byI ; I ¼ 1; 2; . . . ; NP. 6 7
quI B1I ¼ 6
B
6
0 0 cI 0 0 7,
7 ð47Þ
6 7
(37) 6 0 0 0 cI 0 7
4 5
Because the shape functions do not possess Delta 0 0 0 0 cI
function properties, the uI are not the real nodal values,
therefore a full transformation method [36] is used in the 2 qcI
3
present study. The final nonlinear equilibrium equation can 0 0 qx cI 0
BLs 4 5,
be expressed in matrix form as I ¼
0  cRI 1 qcI
R qy 0 cI
^ s ð^uÞ^u ¼ F,
K ^ (38) 2 3
cI 0 0 0 0
6 7
where 6 0 cI 0 0 0 7
6 7
6 7
^ s ¼ K1 Ks KT ; F^ ¼ K1 F; u^ ¼ Ku,
K (39) WTI ¼ 6
6
0 0 cI 0 0 7,
7 ð48Þ
6 7
6 0 0 0 cI 0 7
Ks ðuÞ ¼ KL þ KNL ðuÞ þ KB (40) 4 5
0 0 0 0 cI
in which KL represents the linear stiffness matrix, KNL
denotes the nonlinear stiffness matrix, which is a function 2 3T
qw 1 qw
of the displacements, KB is the stiffness due to the qx 0 R qy 0 0 0 0 0
boundary conditions, and K is the transformation matrix. ^ ¼4
H 5 ,
1 qw qw
0 R qy qx 0 0 0 0 0
The matrices K, KL, KNL, KB, and F are given as follows:
2 qcI
3
KIJ ¼ cI ðxJ ÞI; I is the identity matrix; (41) 0 0 qx 0 0
G¼4 5. ð49Þ
1 qcI
Z 0 0 0 0
LZ Z R qy
y0
T T
KLIJ ¼ R BLI S BLJ dy dx; KBIJ1 ¼ a B1BI B1BJ dl, The nonlinear equilibrium equations can be solved by
0 0 lu
using either the conventional or the modified Newton–
(42)
Raphson method. Due to the necessity of calculating the
Z L Z y0 
tangent stiffness during each iteration, the conventional
1 LT NL T Newton–Raphson technique suffers from a high computa-
KNL
IJ ¼ R B S BNL
J þ BI S BLJ
0 0 2 I tion cost. Although this can be reduced through the
1 T adoption of the modified Newton–Raphson method, in
þ BNL S BNL dy dx, ð43Þ
2 I J which the tangent stiffness is only computed at the
beginning of each load step and is then held constant
Z L Z y0 Z Z throughout the iterative cycles, it is difficult to obtain the
FI ¼ WTI f̄R dy dx þ WTI t̄ dG þ a B1BI ū dl, complete load–displacement path using this method,
0 0 G lu because the load limit points cannot be passed. In this
(44) paper, the arc length control technique is introduced and
where combined with the Newton–Raphson method to solve the
nonlinear equilibrium equation. In the modified New-
2 3
BLe
I ton–Raphson method, the tangent stiffness must be
6 Lm 7 ^ G, determined at the beginning of each load step, the
BLI ¼ 4 BI 5; BNL
I ¼H (45)
procedure to compute the tangent stiffness is briefly
BLs
I introduced here, The details is given by Crisfield [33].
 ARTICLE IN PRESS
788 X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792

The external load is assumed to be proportional to the where Nx and Ny are the stresses in the x and y directions,
fixed load F0 as respectively, and Nxy is the shear stress.
After deriving tangent stiffness Kt, we apply the arc
F ¼ lF0 , (50) length load strategy to solve Eq. (38), the full response path
where l is the load level parameter. The nonlinear of the load displacement can be obtained.
equilibrium Eq. (38) can be rewritten as
^ s u^  lF^ 0
gð^u; lÞ ¼ K (51) 3. Numerical examples

for a new equilibrium state, and thus the following In this section, examples for plate and shell problems are
equation should exist presented. The displacement shape functions are formu-
lated using kernel particle estimates. Gauss integration is
gð^u þ D^u; l þ DlÞ ¼ 0, (52)
used to evaluate the stiffness matrix. For some cases, both
where D^u and Dl are the displacement increment and load classical thin shell theory (CST) and FSDT are employed
increment factors, respectively. By introducing the Taylor to determine the load–displacement path.
series expansion to Eq. (52), the following incremental
form of the equation can be attained
3.1. Pinched cylinder with end diaphragms
Kt D^u ¼ DlF 0  gð^u; lÞ, (53)
where Kt is the tangent stiffness matrix and is given by The pinched cylinder, shown in Fig. 2, is a well-known
benchmark problem. The cylinder with rigid end dia-
Kt ¼ KL þ KN þ KG , (54) phragms is subjected to a point load P ¼ 1 at the center on
opposite sides of the cylinder. The cylinder has a length
where KL is the linear stiffness matrix that is given in
L ¼ 600 in, a radius R ¼ 300 in, and a thickness h ¼ 3 in.
Eq. (42) and KN is the nonlinear stiffness matrix that
The Young’s modulus and Poisson’s ratio are E ¼ 3 
relates to the displacements and is expressed as
106 psi and n ¼ 0:3, respectively. The exact solution of the
Z L Z y0 n o central deflection is 1.8248  105 in [18,37]. Due to its
T NL T T
KNIJ ¼ R BLI SBNL
J þ BI SBLJ þ BNL
I SBNL
J dy dx. symmetry, only half of the cylinder is analyzed. The nodes
0 0
are regularly distributed and a background mesh based on
(55)
the nodes is constructed for numerical integration. A 4  4
The geometric stiffness matrix KG is given as Gaussian quadrature rule is used. A convergence study that
Z L Z y0 is based on the node numbers and a scaling factor of 3.0 is
G
KIJ ¼ R GTI N̄ GJ dy dx, (56) carried out, and the solution is shown in Table 1. It can be
0 0 seen that good agreement is attained.
where In order to investigate the performance of present
2 3 " # technique in dealing with membrane and shear locking,
qcI
0 0 qx 0 0 Nx N xy the present solutions are compared with the FEM solutions
G¼4 1 qcI
5; N̄ ¼ , (57) obtained using 9-node elements with selective reduced
0 0 R qy 0 0 N xy Ny
integration (SRI) and uniform reduced integration (URI),

z, w

, v x, u

h P
L

Fig. 2. Pinched cylinder with end diaphragm.

 ARTICLE IN PRESS
X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792 789

Table 1
Central deflection of a pinched shell

Refs. [18,37] (  105) Present results (  105)

dmax ¼ 3.0

1.8248 20  20 24  24 28  28 30  30 32  32 34  34 38  38 40  40 44  44
1.7245 1.7453 1.7951 1.8098 1.8197 1.8268 1.8264 1.8265 1.8265

1.2 x, u
Rigid diaphragm
1.0 z, w q
Normalized deflection

L
0.8
, v
free
Present
0.6 free
9-node SRI
4-node RSDS
0.4 Heterosis SRI h
9-node γ
9-node URI
0.2 Heterosis URI R
EFG θ0
0.0
0 4 8 12 16 20
Number of nodes per side
Fig. 4. Cylindrical shell roof.
Fig. 3. Comparison of deflection under point load in the pinched shell.

is g ¼ 0.09 k/ft2. Table 2 shows the convergence study that

9-node g method, 4-node resultant-stress degenerated-shell is based on the number of nodes and scaling factors. The
element (RSDS), Heterosis elements with SRI and URI nodes are regularly distributed, and a 4  4 Gaussian
[15], and element free Galerkin method (EFG). Due to quadrature rules is carried out on the background mesh.
symmetry of the structure, an octant of the shell is The vertical deflection of the center of the free edge in the
modeled. The vertical displacements at the load point are literature [37–39] is 0.3024 ft. It is observed that the present
normalized with the exact solution. The comparisons are approach gives satisfactory results. The present solutions
depicted in Fig. 3. It is observed that the solution of present are also compared with the FEM solutions obtained using
approach is superior to those obtained from finite element 9-node elements with SRI and URI, 9-node g method,
solutions with SRI when the number of nodes per side 4-node RSDS element, heterosis elements with SRI and
exceeds 16. The elements with URI shows better conver- URI [15] and EFG method. Due to symmetry of the
gence performances than present method when the number structure, a quarter of the shell is modeled. The vertical
of nodes per side is less than 16, however, the URI elements displacements at midpoint of the free edge are normalized
exhibit spurious modes for shells with different boundary with the solutions in Refs. [37–39]. The comparisons are
conditions, such as pinched shell with free ends [37]. Due to illustrated in Fig. 5. It is seen that the present approach
the adoption of a higher order weight function, the quartic performs very well compared to other well-known finite
spline function, and a quadrature at 6  6 integration, the element solutions. The EFG method also shows a better
EFG method in [18] shows a better performance than convergence rate in this case because of its higher-order
present technique. shape function and quardrature scheme.

3.2. Scordelis-Lo barrel vault 3.3. Simply supported square plate under a uniform load

The Scordelis-Lo Barrel loaded by gravity force is A simply supported square plate subjected to a uniform
supported on a rigid diaphragm on the curved edges. The transverse pressure q is investigated. The length of the
geometry of the shell is given in Fig. 4. The length of the square plate is a ¼ b ¼ 1 m and the thickness t ¼ 0.02 m.
roof L ¼ 50 ft, the radius R ¼ 25 ft, the thickness t ¼ 3 in, The material properties are E ¼ 3  109 N/m2 and u ¼ 0.3.
and the angle y0 ¼ 80 . The material properties are Geometrically nonlinear analysis is performed using 8  8
E ¼ 3.0  106 psi and u ¼ 0. The weight of the roof itself nodes and a scaling factor of 3.0. The nondimensional
 ARTICLE IN PRESS
790 X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792

Table 2
Vertical deflection at the midpoint of the free edge of the Barrel vault roof

Refs. [37–39] Present results

dmax ¼ 2.5 dmax ¼ 3.0 dmax ¼ 3.5

14  14 16  16 18  18 14  14 16  16 18  18 14  14 16  16 18  18
0.3024 0.2559 0.2641 0.2687 0.2581 0.2709 0.2761 0.2814 0.2976 0.3014

2.4
Present
9-node SRI hinged L
2.0 free
4-node RSDS
Heterosis SRI P
9-node γ
Normalized deflection

1.6
9-node URI w
h u
Heterosis URI
v
1.2 EFG free
R
hinged

0.8 θ
θ0
0.4

0.0 Fig. 7. Hinged cylindrical shell under a concentrated load.

0 4 8 12 16 20
Number of nodes per side

Fig. 5. Comparison of deflection at midpoint of free edge in the Scordelis- results for the central deflection with those that are given
Lo roof. by Chua [40].

3.4. Hinged cylindrical shell under a concentrated load

2.5
A hinged cylindrical shell subjected to a concentrated
load is shown in Fig. 7. This case has been used to test
2.0 geometric nonlinearities by many researchers [4,12,13,33].
Present solution
An isotropic shell and a composite shell with a lamination
Chia [40]
scheme [901/01/901] are investigated in present study. Both
1.5
shells have length L ¼ 508 mm, the radius R ¼ 2540 mm,
w
_

the thickness h ¼ 12.7 mm, and the angle y ¼ 0.2 rad. For
1.0 the isotropic shell, Young’s modulus and Poisson ratio are
E ¼ 3:10275 kN=mm2 and u ¼ 0.3, respectively. The mate-
rial properties of the laminated shell are E11 ¼ 3.3 GPa,
0.5 E22 ¼ 1.1 GPa, G12 ¼ 0.66, and u12 ¼ 0.25. The number of
nodes that is used for the whole shell is 10  10, and the
scaling factor is 3.0. Due to the snap-through phenomenon,
0.0 the arc length method combined with the modified New-
0 20 40 60 80 100
ton–Raphson approach is adopted to obtain the full path
q
of the load displacement. Figs. 8–10 describe the variations
Fig. 6. Simply supported plate under a uniform load. of the vertical displacements of center point with the loads.
The results in Fig. 8 are obtained by using the CST. Fig. 9
shows the solution based on the FSDT. Fig. 10 shows the
result for the laminated shell. The present solutions are
deflection and load are defined as w̄ ¼ w=h and compared with those given by Sze et al. [41]. It can be seen
q̄ ¼ qa4 =Eh4 . Only a quarter of the plate is analyzed due that the present results are in good agreement with those in
to its symmetry. Fig. 6 shows the comparison of the present the literature.
 ARTICLE IN PRESS
X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792 791

4.0 3.5

3.5 3.0
Present solution
3.0 Sze et al. [41]
Present solution 2.5
Sze et al. [41]
2.5

Load (kN)
2.0
Load (kN)

2.0
1.5
1.5
1.0
1.0
0.5
0.5
0.0
0.0 0 5 10 15 20 25 30 35
0 5 10 15 20 25 30 35
Central point displacement
Central point displacement
Fig. 10. Load deflection curve for the hinged laminated cylindrical shell
Fig. 8. Load deflection curve for the hinged isotropic cylindrical shell [901/01/901].
(CST).

nodes on areas where more refinement is needed. The

present approach also performs very well in eliminating the
membrane and shear locking in shell problems compared
3.5
to other well-known finite element solutions with SRI. The
3.0
present method shows a slower convergence rate than some
Sze et al. [41]
elements with URI; however, the performance of URI
Present
2.5 elements is not stable. Furthermore, in the conventional
Ritz method, the trial functions must satisfy at least the
Load (kN)

2.0 essential boundary conditions, and the difficulty of finding

appropriate trial functions for certain boundary condition
1.5 types has been well documented in open literature. The
present kp-Ritz method elegantly overcomes this by using a
1.0 common shape function type to describe the interior, while
the boundary conditions are separately dealt with via
0.5 penalty enforcement. Thus, the kp-Ritz is relatively much
more robust. Numerical examples are performed to
0.0 validate the present method and formulation, and satisfac-
0 5 10 15 20 25 30 35
tory results are obtained in comparison with those given in
Central point displacement
the literature.
Fig. 9. Load deflection curve for the hinged isotropic cylindrical shell
(FSDT).
References

[1] Gallagher RH. The development and evaluation of matrix methods

for thin shell structural analysis. Report No. 8500-902011, Bell
4. Conclusion Aerosystems, Buffalo, NY, 1966.
[2] Gallagher RH, Yang HTY. Elastic instability predictions for doubly-
In this paper, the kp-Ritz method is extended to the curved shells. In: Proceedings of second conference on matrix
geometrically nonlinear analysis of shells. The formulation methods in structural mechanics, Dayton, OH, 1968.
for nonlinear shells is based on the first-order shear [3] Gallagher RH, Ashwell D. Finite elements for thin shells and curved
members. New York: Wiley; 1976.
deformation shell theory. The kp-Ritz method combined [4] Horrigmoe G, Bergan PG. Nonlinear analysis of free-form shells by
with the Newton–Raphson and arc length approaches are flat shell elements. Comput Methods Appl Mech Eng
utilized to solve the nonlinear equilibrium equations. In 1978;16(19):11–35.
present study, the meshfree kernel particle estimate is [5] Hughes TJR, Cornoy E. Nonlinear finite element shell formulation
employed to approximate the 2-D transverse displacement accounting for large membrane strains. Comput Methods Appl Mech
Eng 1983;39:69–82.
field. No mesh generation is required, therefore, mesh [6] Huang HC, Hinton E. Elastic–plastic and geometrically nonlinear
distorsion in FEM is avoided. It is very convenient to add analysis of plates and shells using a new nine node elements. Finite
 ARTICLE IN PRESS
792 X. Zhao et al. / Engineering Analysis with Boundary Elements 31 (2007) 783–792

element methods for nonlinear problems. Euor-US Symposium, [23] Liu YJ. Analysis of shell-like structures by the boundary element
Trondheim, Norway, 1985. method based on 3-D elasticity: formulation and verification. Int J
[7] Hughes TJR, Liu WK. Nonlinear finite element analysis of shells: Num Methods Eng 1998;41(3):541–58.
Part I. Three-dimensional shells. Comput Methods Appl Mech Eng [24] Kitipornchai S, Xiang Y, Wang CM, Liew KM. Buckling of thick
1981;26:331–62. skew plates. Int J Num Methods Eng 1993;36:1299–310.
[8] Hughes TJR, Liu WK. Nonlinear finite element analysis of shells: [25] Liew KM, Xiang Y, Wang CM, Kitipornchai S. Flexural vibration of
Part II. Two-dimensional shells. Comput Methods Appl Mech Eng shear deformable circular and annular plates on ring supports.
1981;27:167–81. Comput Methods Appl Mech Eng 1993;110:301–15.
[9] Argyris JH, Scharpf DW. The SHEBA family of shell elements for [26] Liew KM, Xiang Y, Kitipornchai S, Meek JL. Formulation of
the matrix displacement method. Part I. Natural definition of Mindlin–Engesser model for stiffened plate vibration. Comput
geometry and strains. J R Aeronaut Soc 1968;72:873–8. Methods Appl Mech Eng 1995;120:339–53.
[10] Argyris JH, Scharpf DW. The SHEBA family of shell elements for [27] Cheung YK, Zhou D. Three-dimensional vibration analysis of
the matrix displacement method. Part II. Interpolation scheme and cantilevered and completely free isosceles triangular plates. Int J
stiffness matrix. J R Aeronaut Soc 1968;71:878–83. Solids Struct 2002;39:673–87.
[11] Choi CK, Paik JG. An effective four node degenerated shell element [28] Reddy JN, Savoia M. Layer-wise shell theory for postbuckling
for geometrically nonlinear analysis. Thin-Walled Struct of laminated circular cylindrical shells. AIAA J 1992;30(8):
1996;24:261–83. 2148–54.
[12] Sze KY, Zheng SJ. A hybrid-stress nine-node degenerated shell [29] Liew KM, Ng TY, Zhao X. Free vibration analysis of conical shells
element for geometrical nonlinear analysis. Comput Mech via the element-free kp-Ritz method. J Sound Vibr 2005;281:627–45.
1999;23:448–56. [30] Zhao X, Ng TY, Liew KM. Free vibration of laminated two-side
[13] Sze KY, Chan WK, Pian THH. An eight-node hybrid-stress solid- simply supported cylindrical panels via the kp-Ritz method. Int J
shell element for geometric nonlinear analysis of elastic shells. Int J Mech Sci 2004;46:123–42.
Num Methods Eng 2002;55:853–78. [31] Risk E. The application of Newton’s method to the problem of elastic
[14] Kim CH, Sze KY, Kim YH. Curved quadratic triangular degener- stability. J Appl Mech 1972;39:1060–6.
ated- and solid-shell elements for geometric nonlinear analysis. Int J [32] Wempner GA. Discrete approximation related to nonlinear theories
Num Methods Eng 2003;57:2077–97. of solids. Int J Solids Struct 1971;7:1581–99.
[15] Sansour C, Kollmann FG. Families of 4-node and 9-node finite [33] Crisfield MA. A fast incremental/iterative solution procedure that
elements for a finite deformation shell theory. An assessment of handles ‘‘snap-through’’. Comput Struct 1990;13:55–62.
hybrid stress, hybrid strain and enhanced strain elements. Comput [34] Sanders Jr JL. Nonlinear theories for thin shells. Q Appl Math
Mech 2000;24:435–47. 1963;21:21–36.
[16] Campello EM, Pimenta PM, Wriggers P. A triangular finite shell [35] Reddy JN. Mechanics of laminated composite plates and shells:
element based on a fully nonlinear shell formulation. Comput Mech theory and analysis. 2nd ed. Boca Raton, FL: CRC Press; 2004.
2003;31:505–18. [36] Chen JS, Pan C, Wu T, Liu WK. Reproducing kernel particle
[17] Dennis ST. A Galerkin solution to geometrically nonlinear laminated methods for large deformation analysis of non-linear structures.
shallow shell equations. Comput Struct 1997;63(5):859–74. Comput Methods Appl Mech Eng 1996;139:195–227.
[18] Krysl P, Belytschko T. Analysis of thin shells by the element-free [37] Liu WK, Law ES, Lam D, Belytschko T. Resultant-stress degener-
Galerkin method. Int J Solids Struct 1996;33:3057–80. ated-shell element. Comput Methods Appl Mech Eng
[19] Noguchi H, Kawashima T, Miyamura T. Element free analyses of 1986;55:259–300.
shell and spatial structures. Int J Num Methods Eng [38] Belytschko T, Stolarski H, Liu WK, Carpenter N, Ong JS. Stress
2000;47:1215–40. projection for membrane and shear locking in shell finite elements.
[20] Li S, Hao W, Liu WK. Numerical simulations of large deformation Comput Methods Appl Mech Eng 1985;51:221–58.
of thin shell structures using meshfree methods. Comput Mech [39] Simo J, Fox DD, Rifai MS. On a stress resultant geometrically exact
2000;25:102–16. shell mode, Part II: The linear theory. Comput Methods Appl Mech
[21] Wen PH, Aliabadi MH, Young A. Boundary element analysis of Eng 1989;73:53–92.
reinforced shear deformable shells. Int J Num Methods Eng [40] Chua CY. Nonlinear analysis of plates. New York: McGraw-Hill;
2002;54(6):789–808. 1980.
[22] Haas M, Kuhn G. Mixed-dimensional, symmetric coupling of FEM [41] Sze KY, Liu XH, Lo SH. Popular benchmark problems for geometric
and BEM. Eng Anal Bound Elem 2003;27(6):575–82. nonlinear analysis of shells. Finite Elements Anal Des 2004;40:1551–69.