ARTICLE IN PRESS

Thin-Walled Structures 46 (2008) 1338 1350

Contents lists available at ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

Imperfection sensitivity and postbuckling analysis of elastic shells of

revolution
T. Hong a, J.G. Teng b,
a
College of Mechanical Engineering, Zhejiang University of Technology, Hangzhou, Zhejiang, PR China
b
Department of Civil and Structural Engineering, The Hong Kong Polytechnic University, Hong Kong, PR China

a r t i c l e in f o a b s t r a c t

Article history: The imperfection sensitivity of thin shells of revolution has been a topic of great interest to researchers
Received 28 July 2007 and designers. With the gradual emergence of the new approach of structural design based on advanced
Received in revised form numerical simulation, the efcient and accurate prediction of the behaviour of imperfect shells has
3 April 2008
recently assumed new signicance. This paper rst presents an efcient semi-analytical nite element
Accepted 3 April 2008
formulation for the nonlinear analysis of imperfect shells of revolution subject to general nonsymmetric
Available online 9 June 2008
loads. Both the applied loads and the initial geometric imperfections may take any form and are
Keywords: approximated by Fourier series expansions. Application of the analysis to study the effects of geometric
Shells
imperfections on the behaviour of shells of revolution is then presented to demonstrate the accuracy
Imperfection sensitivity
and capability of the present method and the imperfection sensitivity of shell structures. As a special
Postbuckling analysis
Finite element analysis case of the present formulation, two different approaches are also presented for the postbuckling
analysis of perfect thin shells of revolution under axisymmetric loads. The accuracy and capability of
both methods are demonstrated using a number of numerical examples, which also allows some new
insight to be gained into the postbuckling behaviour of perfect shells of revolution.
& 2008 Elsevier Ltd. All rights reserved.

1. Introduction afne imperfection in Ref. [2]. In the present study, an eigenmode-

Small geometric imperfections (or simply referred to as
imperfections for brevity) are widely known to have a strong
detrimental effect on the buckling strengths of shells, and to be
the reason for the large discrepancy between experimental
buckling loads and theoretical predictions based on the perfect
geometry [1]. Since the geometric imperfection in a shell to be
constructed is not known at the design stage, both the form and
the amplitude of the expected imperfection have to be specied
based on certain assumptions or prior knowledge in the stability
assessment at the design stage. For example, the European
standard for steel shell structures [2] requires that when a
geometrically nonlinear elastic or elasticplastic analysis with
explicit representation of imperfections is used in design, a range
of potentially damaging imperfection forms should be explored if
the most unfavourable imperfection form cannot be readily
identied. This code also recommends that the imperfection
should be specied in the form of the critical buckling mode from
a linear elastic bifurcation analysis, with its amplitude linked to
fabrication quality, unless a different unfavourable pattern is
justied. Such an imperfection is referred to as an eigenmode-
 Corresponding author. Tel.: +852 2766 6012; fax: +852 2334 6389.
E-mail address: cejgteng@polyu.edu.hk (J.G. Teng).
afne imperfection is more loosely dened as an imperfection in
the form of an eigenmode/eigenvector from either a linear or
nonlinear bifurcation analysis, which may or may not be the
critical buckling mode. Only eigenmodes from a nonlinear
bifurcation analysis are considered in the present paper.
The need to investigate the effect of a range of potentially
damaging geometric imperfection forms in the structure to be
constructed at the design stage compromises the attractiveness of
the direct use of nonlinear analysis in stability design of shell
structures. With todays computers, the problem often does not lie
in the amount of computational time, but in the time required to
prepare the nite element model and to interpret the numerical
results. As the direct use of nonlinear analysis becomes widely
accepted in practical design, an efcient nonlinear analysis
computer program for use in shell stability design, minimizing
the total user effort, is highly desirable.
Given the above considerations and noting that most civil
engineering metal shell structures are axisymmetric, it is clear
that the semi-analytical method for shells of revolution [3,4]
using an axisymmetric shell element and Fourier series approx-
imations of circumferential variables provides an attractive
alternative to a general nite element method employing general
shell elements. In the semi-analytical method, as discretization of
the shell is required only in the meridional direction, the
preparation of input data and the interpretation of numerical

0263-8231/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.tws.2008.04.001
 ARTICLE IN PRESS
T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350
results are both more efcient and simpler. The basic features of
such an analysis are also easier for the user to master than a
general shell element analysis. In addition, it is more accurate in
general, as the circumferential description of displacements in
many buckling and postbuckling problems is exact. In particular,
for shells of revolution subjected to axisymmetric loads and with
an eigenmode-afne imperfection described by a single harmonic
term, the semi-analytical method can be much more efcient than
the general nite element method, as the circumferential
variations of geometric imperfections and other variables can be
described using a small number of Fourier terms.
Against the above background, this paper rst presents a new
efcient semi-analytical nite element formulation for the non-
linear analysis of imperfect shells of revolution subject to general
nonsymmetric loads. The present formulation represents an
extension of the formulation for nonlinear analysis for perfect
thin shells of revolution under arbitrary loads presented by Hong
and Teng [4]. The advantages of Hong and Tengs semi-analytical
method for shells of revolution over previous formulations are
detailed in Ref. [4]. Both the applied loads and the initial
geometric imperfections may take any form and are approximated
by Fourier series expansions. Application of the analysis to study
the effects of eigenmode-afne geometric imperfections and
multi-mode imperfections on the behaviour of shells of revolution
subjected to axisymmetric loads is then presented. Finally, a
cylindrical shell with a multi-mode imperfection under wind
pressure is studied to demonstrate the greater capability of the
present formulation than is illustrated by the simpler numerical
examples. As a special case of the present formulation, two
different approaches are presented for the postbuckling analysis
Z Z
w s
w
u
u
R R
w
u 2
Fig. 1. Doubly curved axisymmetric shell element: (a) local and global displacements, (b) geometry of an element, (c) displacements within the element, (d) nodal
displacements.
1339
of perfect thin shells of revolution under axisymmetric loads. The
accuracy and capability of both methods are demonstrated using a
number of numerical examples, which also allow some new
insight to be gained into the postbuckling behaviour of perfect
shells of revolution.
2. Nonlinear analysis of imperfect shells
2.1. Displacements of imperfect shells
The isoparametric doubly curved shell element used in the
present nite element formulation is shown in Fig. 1. The accuracy
of the element has been demonstrated in many successful
applications [46]. The element geometry is described in
cylindrical coordinates and is uniquely dened by the radius R,
the axial coordinate z and the element meridional curvature dj/ds
at the nodal points. The intermediate values of R, z, and dj/ds of
the shell element are interpolated in terms of the nodal values
using cubic Hermitian functions. The nodal displacements in
global coordinates are taken as ui, (du/ds)i, vi, (dv/ds)i, wi and (dw/
ds)i at the two nodes of the element (Fig. 1d). The displacements
at any point, dened in the global coordinate system, u, v and w
(Fig. 1c) are interpolated between the nodal points in terms of the
nodal values also using cubic Hermitian functions. The set of
global displacements u, v and w at any point is related by a
transformation matrix [T] to the local displacements u; v and w (in
curvilinear coordinates) [5].
For a perfect shell of revolution subject to arbitrary loads, the
displacements in general include both symmetric and anti-
1 L
1
L
2
2
1
w
w2 ,
s 2
v
v2 ,
s 2
u
u2 ,
s 2
 ARTICLE IN PRESS

1340 T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350

symmetric components. The displacements u, v and w in the 2.3. Vector of nodal variables
global coordinate system can be expanded into truncated Fourier
series [4]: A nonlinear analysis of imperfect shells of revolution under
arbitrary loads is presented here as an extension of the nonlinear
P
M P
N
analysis of perfect shells of revolution presented by Hong and
u fusm cos ksm yg fuan sin kan yg
m0 n0 Teng [4]. The vector of nodal variables for each perfect element is
P
M P
N denoted by {dp}e below:
v fvsm sin ksm yg fvan cos kan yg (1)
m0 n0
fdp ge fds0 ; . . . ; dsm ; . . . ; dsM ; da0 ; . . . ; dan ; . . . ; daN gT (5)
P
M P
N
w fwsm cos ksm yg fwan sin kan yg where
m0 n0
     
qu sm sm qv sm sm qw sm
where y is the circumferential angular coordinate, M and N are the fdsm g usm
1 ; ; v1 ; ; w1 ; ;
qs 1 qs 1 qs 1
number of the symmetric and anti-symmetric Fourier terms about  sm  sm  sm T
y 0, respectively, and usm, vsm and wsm, and uan, van and wan are qu qv qw
usm
2 ; ; vsm
2 ; ; wsm
2 ; (6a)
the coefcients of the mth symmetric term and the nth anti- qs 2 qs 2 qs 2
symmetric term, respectively. For simplicity, these Fourier   an  an  an
coefcients are referred to as displacements directly hereafter. qu qv qw
fdan g uan
1 ; ; van
1 ; ; wan
1 ; ;
ksm and kan are the harmonic numbers of the mth and nth Fourier qs 1 qs 1 qs 1
 an  an  an T
terms which are referred to as harmonic modes ksm and kan, qu qv qw
respectively. uan
2 ; ; van ; ; wan
2 ; (6b)
qs 2 2 qs 2 qs 2
where {dsm} and {dan} are the vectors of nodal variables for the
mth symmetric Fourier term and the nth anti-symmetric Fourier
2.2. Geometric imperfections of shells
term, respectively. These vectors each contain 12 unknowns, with
six for each of the two nodes denoted by the subscripts 1 and 2,
Initial geometric imperfections in shells of revolution can be
respectively.
similarly described using truncated Fourier series:
For perfect shells, the required number of symmetric and anti-
P
I J
P symmetric Fourier harmonic terms M and N in the analysis
u~ fu~ si cos lsi yg fu~ aj sin laj yg depend on the Fourier series approximation of the external load
i0 j0
and nonlinear coupling [4]. For imperfect shells as considered
P
I J
P
v~ fv~ si sin lsi yg fv~ aj cos laj yg here, the required number of symmetric and anti-symmetric
(2)
i0 j0 Fourier harmonic terms K and L depend also on the Fourier series
P
I J
P approximation of the geometric imperfection (Eq. (2)). The Fourier
~
w ~ si cos lsi yg
fw ~ aj sin laj yg
fw terms required to describe the load and nonlinear coupling may
i0 j0
not include those required to describe the geometric imperfection,
where u, ~ v~ and w
~ are the geometric deviations from the perfect consequently the vector of nodal variables for each element for a
geometry in the radial, circumferential and axial directions, perfect shell (Eq. (5)) may not be used directly in the nonlinear
respectively, of any point on the shell reference surface in analysis of an imperfect shell. For imperfect shells, the vector of
global cylindrical coordinates; I and J are the numbers of the nodal variables for each element therefore needs to include all
symmetric and anti-symmetric Fourier terms, respectively, symmetric and anti-symmetric harmonics contained in Eqs (3)
required to describe these deviations, u~ si , v~ si and w
~ si are the and (5) and is denoted by
Fourier coefcients of the ith symmetric term and u~ aj , v~ aj fdge fds0 ; . . . ; dsk ; . . . ; dsK ; da0 ; . . . ; dal ; . . . ; daL gT (7)
and w ~ aj are the Fourier coefcients of the jth anti-symmetric
In general, some of the Fourier harmonics in {dp}e are the same
term; and lsi and laj are the harmonic numbers of the ith and jth
as those in fdg~ e , so the total numbers of symmetric and anti-
Fourier terms which are referred to as harmonic modes lsi and laj,
symmetric Fourier terms in {dp} are fewer than M+I and N+J,
respectively.
respectively, that is, KpM+I and LpN+J.
The vector of nodal imperfection components for each element
~ e is dened as follows:
fdg
2.4. Effect of geometric imperfections
~ e fd~ s0 ; . . . ; d~ si ; . . . ; d~ sI ; d~ a0 ; . . . ; d~ aj ; . . . ; d~ aJ gT
fdg (3)
The nonlinear straindisplacement relationship of Rotter and
Here, fd~ si g and fd~ aj g are the vectors of nodal imperfection Jumikis for shells of revolution [7], which is a special case of the
components for the symmetric harmonic mode lsi and the anti- nonlinear straindisplacement relationship for general shells
symmetric mode laj, respectively, and are given by given by Teng and Hong [8], is adopted for dening the linear
(  si  si  si and nonlinear straindisplacement matrices in the present nite
qu~ qv~ ~
qw
fd~ si g u~ si
1 ; ; ~
v si
; ; ~
w si
; ; u~ si ; element formulation. The linear strains {el}are not affected by
qs 1 1 qs 1 1 qs 1 2
geometric imperfections and are described by the following
 si  si  si )T equation:
qu~ qv~ qw~
; v~ si ; ; ~
w si
; (4a)
qs 2 2 qs 2 2 qs 2 fl g B0 fdge (8)
in which {d}e is the vector of nodal variables for each element
(  aj  aj  aj  aj  aj  aj )T
qu~ qv~ ~
qw qu~ qv~ ~
qw (Eq. (7)), and [Bo] is the linear straindisplacement matrix. [Bo] is
fd~ aj g u~ aj
1; ; v~ aj
1;
~ aj
;w 1; ; u~ aj
2; ; v~ aj
2;
~ aj
;w 2;
qs 1 qs 1 qs 1 qs 2 qs 2 qs 2 independent of displacements and is detailed in Ref. [4].
The effect of geometric imperfections is to modify the
(4b)
nonlinear straindisplacement matrix [BL] so that the nonlinear
where subscripts 1 and 2 denote the two nodes of an element. strain increments {denl} are related to the displacement
 ARTICLE IN PRESS

T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350 1341

increments d{d}e through denition of [K]e is also available in Ref. [4]. The inuence of
imperfections on [K]e is accounted for through the nonlinear
fdnl g B~ L ddfdge (9a) straindisplacement matrix B~ L  which is different from that for
perfect shells.
where the nonlinear straindisplacement matrix The element tangent stiffness matrices [KT]e (Eq. 14) are
condensed to reduce the inter-element continuity in a way
B~ L d OdG
~ (9b)
similar to that described in Ref. [5]. The nodal displacement
The rotationdisplacement matrix [G] is not affected by unknowns after condensation consist of the three translations u, v
imperfections and is detailed in Ref. [4]. The matrix of pseudo- and w, and the meridional rotation b. The tangent stiffness
~
rotations for shells of revolution Od is dened by Eq. (30) in Ref. matrix for the structure [KT] may then be found by assembling the
[4], where the pseudo rotations bi d; d ~ (i 16) are those condensed element tangent stiffness matrices. After the
including the effect of imperfections. These pseudo rotations are condensed displacement vector is determined, the complete
given by set of nodal displacements can be recovered. This condensation
and recovery procedure is necessary for the analysis to be
~ b d b~ di
bi d; d i i
~ 1  6 (10) applicable to segmented or branched shells featuring slope
discontinuities [5].
in which bi(d) is due to the displacements and b~ i d
~ is due to the
initial imperfections. The pseudorotations bi d; d~ can also be
described using Fourier series as follows: 2.6. Solution method

X
K X
L To determine the pre- and post-critical responses of the
bi fbi sk cos mk yg fbi al sin nl yg i 1; 3; 5 (11a)
k0 l0
X
K X
L loaddeection path. The arc-length method [1214] appears to
bi fbi sk sin mk yg fbi al cos nl yg i 2; 4; 6 (11b)
k0 l0
The above approach of including the effect of geometric
imperfections has been employed by previous researchers
[3,911]. This approach is valid provided the summation of the
initial imperfection and the subsequent displacement is within
the range of validity of the total Lagrangian approach for
deformations of moderately large rotations.
2.5. Governing nonlinear equations
The equations governing the nonlinear deformations of perfect
shells of revolution under nonsymmetric loads can be adapted for
the nonsymmetric deformations of imperfect shells with the
nodal variables dened by Eq. (7). The total Lagrangian approach
is adopted here in which all the quantities are referred to the
undeformed conguration. The application of the principle of
virtual displacements leads to a set of nonlinear equations for the
nite element model of a given structure which may be
represented by
X Z
fFdg fFg  BT fSg dV 0 (12)
element
in which {d} is the vector of nodal displacement variables, {F} is
the vector of equivalent nodal forces due to body forces and
surface tractions, [B] is the total incremental straindisplacement
matrix and [B] [B0]+B~ L ( Eqs. (8) and (9b)), {S} is the vector of
generalized stress resultants and {F(d)} is the vector of nodal
residual forces. For each iteration, the nodal displacement
increments for the structure {Dd} are obtained by solving the
following linearized system of equations:
fFdg K T fDdg
where [KT] is termed the global tangent stiffness matrix. The
tangent stiffness matrix for each element is given by
K T e Ke K s e
In Eq. (14), the matrix [Ks]e which accounts for the effect of
internal stresses is independent of imperfections, so [Ks]e is the
same as that given by Hong and Teng [4]. [K]e is the stiffness
matrix including the effect of changes in geometry and the
imperfect shell, the nonlinear loaddeection path needs to be
traced. Research on nonlinear solution procedures has led to many
incremental-iterative solution methods which can trace the entire
be most reliable and powerful. Besides being able to trace the
postcritical response, the method can also accelerate the iterative
procedure in the pre-critical range. The arc-length method was
adopted in the present study for the nonlinear analysis of
imperfect elastic shells. It should be noted that even with the
use of the arc-length method, difculty in convergence might still
be encountered in the vicinity of a bifurcation point around which
the loaddisplacement curve features a sharp turn. Reductions in
the load step size are generally required in such situations. In the
present study, user intervention was used when necessary,
although an automated procedure could have been developed.
This aspect is beyond the scope of the present study and is not
further discussed.
3. Imperfection sensitivity of shells
3.1. Specication of imperfections
Imperfections in real shells are not known at the stage of
design and may take a complex form. In order to assess the
buckling strength of a shell at the design stage, some assumptions
have to be made about both the form and amplitude of the
expected imperfection. A common approach for specifying
imperfections is to assume that the imperfection is in the form
of the critical bifurcation buckling mode of a suitable magnitude.
For shells of revolution under axisymmetric loading, the bifurca-
tion buckling mode is described using a single harmonic mode.
For other loading cases, the bifurcation buckling mode needs to be
described as the superposition of a number of harmonic modes.
The nite element formulation for the nonlinear analysis of
imperfect shells of revolution under arbitrary loads presented
(13) above was coded into the CHASH program [4] assuming that both
the geometric imperfections and the deformations of the shell are
y-symmetric as is the case for most real problems. Furthermore, in
the present paper, only imperfections in the form of a single
(14)
eigenmode or the combination of a number of eigenmodes found
from nonlinear bifurcation analysis of a shell of revolution under
axisymmetric loads are considered, although the CHASH program
is capable of dealing with y-symmetric imperfections of any form.
A y-symmetric initial geometric imperfection of the type outlined
 ARTICLE IN PRESS

1342 T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350

above is described by types of loading conditions (axisymmetric loading and nonax-

isymmetric loading) to demonstrate the accuracy and capability of
u~ Sxni tui cos ni y
i the present analysis and the imperfection sensitivity of thin shells.
b~ Sxni tbi cos ni y The loaddeection curve determined considering an imperfec-
i
(15) tion containing a number of harmonic modes is referred to as the
v~ Sxni tvi sin ni y
i loaddeection curve for an imperfection of harmonic modes
~ Sxni twi cos ni y
w n1+n2+? or simply for n n1+n2+?. The plus sign is used to
i
mean that these harmonic modes are included in the geometric
in which ui*, vi*, wi* and bI* are the radial, circumferential and imperfection. Except for the inclusion of harmonic mode 1 in
axial displacements of the eigenvector normalized by the some of the analyses, additional harmonic modes due to nonlinear
maximum buckling displacement and the meridional rotation, coupling were ignored. Harmonic mode 1 due to nonlinear
respectively, of harmonic mode ni, xni is the imperfection factor for coupling needs to be retained in the analysis for some problems,
the corresponding normalized eigenvector, and t is the wall in order to achieve more rapid convergence to equilibrium states.
thickness of the shell. If the imperfection is in a single harmonic No imperfection needs to be included in harmonic mode 1 as
mode n, then the imperfection becomes nonzero displacements in this mode arise naturally from non-
linear coupling [4].
u~ xn tun cos ny
b~ xn tb cos ny n
(16) 3.2. Axisymmetric loading
v~ xn tvn sin ny
~ xn twn cos ny
w
3.2.1. Conecylinder intersection
In the remainder of the section, numerical results from the The rst example is the real conecylinder intersection of an
CHASH program are presented for imperfect shells under two internally pressurized oil storage tank which failed by elastic

0.2

No imperfection
Present
ABAQUS
Internal pressure (MPa)

Cone n = 29
300
0.1
1500
Internal Pressure
340 t = 3.2 Cylinder

0.0
0 10 20 30 40
Vertical displacement at cone apex (mm)

0.2 0.2

No imperfection
Present
ABAQUS
Internal pressure (MPa)

Internal pressure (MPa)

n = 29 n = 29

0.1 0.1
Bifurcation load =
0.06984 MPa No imperfection
29 = 1E-4
29 = 0.01
29 = 0.1
29 = 1

0.0 0.0
0 10 20 30 40 50 0 10 20 30 40 50
Vertical displacement at cone apex (mm) Vertical displacement at cone apex (mm)

Fig. 2. Conecylinder intersection under internal pressure: (a) geometry and loading, (b) comparison with results from ABAQUS: x29 0.01, (c) comparison with results
from ABAQUS: x29 1, (d) effect of imperfection amplitude.
 ARTICLE IN PRESS
T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350
buckling during service due to mis-operation [15,16]. The
geometry of the conecylinder intersection is shown in Fig. 2a.
The material properties are: elastic modulus E 193 GPa and
Poisson ratio n 0.28. This conecylinder intersection under
internal pressure was predicted by CHASH through a nonlinear
bifurcation analysis to buckle at p 0.0698 MPa with 29 circum-
ferential waves. In Fig. 2b and c, the effects of eigenmode-afne
imperfections of two different amplitudes are examined (im-
perfection factor x29 0.01 and 1 for eigenmode n 29). The
results obtained using CHASH are in close agreement with those
from ABAQUS [17]. The effect of imperfection amplitude on the
load-deection curve is shown in Fig. 2d. The postbuckling path of
the imperfect structure with x29 0.01 is almost identical to that
for x29 104. The curve for x29 1 is far away from the
axisymmetric path well before the bifurcation load. The curve
for x29 104 only starts to deviate from the axisymmetric path
when the load exceeds the nonlinear bifurcation load.
3.2.2. Spherical cap under uniform external pressure
An imperfect spherical cap under uniform external pressure
was investigated using CHASH and ABAQUS. The shallow
spherical cap is clamped, and has a rise parameter l 12
pp
(l 2 4 31  n2 H=t ), where H is the height of the spherical
cap, and a Poissons ratio n 0.3. The classical buckling load of a
corresponding perfect complete spherical shell pcr is dened by
2Et2
pcr p (17)
R 2
31  n2
where R is the radius of the spherical cap. The loaddeection
curves for an imperfection of harmonic mode 7 obtained by the
present analysis are compared with those from ABAQUS in Fig. 3a
and b. Close agreement is seen between the present results and
those from ABAQUS, although some small differences are found.
The degrading effect of an initial imperfection on the buckling
load of this spherical cap is shown in Fig. 3c. As the amplitude of
the imperfection increases, the postbuckling response changes
from an unstable one to a stable one. For an imperfection
amplitude of about half the thickness of the cap, the buckling
load is just about 38% of the classical buckling load of the
perfect shell.
The effects of small imperfections in the forms of three
different harmonic modes (n 6, 7 or 8) are shown in Fig. 3d.
The imperfection in the form of the critical buckling mode (n 7)
is seen to have the greatest effect on reducing the buckling
strength.
Fig. 3e and f illustrates the effects of multi-mode imperfections
consisting of two harmonic modes. It can be seen that the
postbuckling path of the multi-mode imperfection of n 7+8
follows closely that for harmonic mode n 7, except for a smaller
buckling load of n 7+8 than that of n 7 (Fig. 3e). The reason
for the lower buckling load is that the superposition of imperfec-
tions of harmonic modes 7 and 8 leads to an imperfection of
greater amplitude. These results mean that harmonic mode n 8
has little inuence on the postbuckling path of the multi-mode
imperfection of n 7+8. In Fig. 3f, the results for a multi-mode
imperfection of n 6+7 are compared with those for the single
mode imperfections of n 6 and 7. The postbuckling loadde-
ection curves demonstrate some initial interaction between
these two harmonic modes, but in the deep postbuckling range,
the curve for n 6+7 follows that for n 6 closely. The results in
Fig. 3e and f therefore show that the shell tends to deform into a
harmonic mode of a lower number rather than one of a higher
number in the postbuckling process.
1343
3.2.3. Cylindrical shell under axial compression
Axially compressed cylindrical shells are known to be the most
imperfection sensitive among known cases, matched only by
complete spherical shells under external pressure. The present
method was also applied to a circular cylindrical shell under axial
compression which is one of the many cylindrical shells studied
by Yamaki [18]. The geometric and material properties of the shell
are as follows: radius R 405 mm, thickness t 1 mm, length
L 291.18 mm, elastic modulus E 5.56 GPa and Poissons
pratio
n 0.3. The Batdorf parameter of the cylinder Z L2 1  n2 =Rt)
is 200. In the present study, both ends of the cylinder were
clamped with the top end allowed to move axially so that
axial compression could be applied. The shell was found by a
nonlinear bifurcation analysis using CHASH to buckle into 16
circumferential waves at an axial stress sxcr 0.911s cl, in
p
which the classical buckling stress given byscl Et=R 31  n2 .
The imperfection sensitivity of this cylinder is now considered.
A series of load-shortening curves of the shell with an eigenmode-
afne imperfection of different amplitudes is shown in Fig. 4a.
As the imperfection amplitude increases, the maximum
load-carrying capacity reduces, and the postbuckling drop
in the load also reduces before the load rises again.
The postbuckling path is almost stable for a shell with
the amplitude of imperfection approaching half the wall
thickness. For this imperfection amplitude, the buckling load
is only about 37% of the classical buckling load of the
perfect cylinder.
The load-shortening curves of the cylindrical shell with a
multi-mode imperfection are shown in Fig. 4b and c. Fig. 4b
compares the postbuckling path for an imperfection of harmonic
modes 15+16 (x15 x16 0.1) (with harmonic mode 1 included in
the analysis) to those for imperfections of a single harmonic mode
of 15 or 16. The load-shortening curve for the multi-mode
imperfection in the advanced postbuckling range is close to the
postbuckling path for harmonic mode 15. Since the amplitude of
the multi-mode imperfection of n 15+16 is greater than that of
the single mode imperfections, the buckling load decreases from
0.67scl for n 15 and 0.66 scl for n 160.57 scl for n 15+16.
Fig. 4c shows that the load-shortening curve of the imperfect
cylinder with an imperfection of harmonic modes
n 16+17(x16 x17 0.1) is predominated by the imperfection
component of harmonic mode 16. The buckling load does decrease
from 0.66 scl for n 160.60 scl for n 16+17. These results again
show that the shell tends to deform into a harmonic mode of a
lower number rather than one of a higher number in the
postbuckling process.
3.3. Nonaxisymmetric loading
Here, a cylindrical shell with a multi-mode imperfection
subjected to nonaxisymmetric wind loading is discussed to
demonstrate the greater capability of the present analysis than
is illustrated by the simpler problems presented above. Bulenda
[19] investigated a cylindrical shell restrained against both radial
and meridional displacements at the two ends subject to uniform
external pressure. It is further restrained against axial displace-
ments at the mid-height so that rigid body translations in the
axial direction are excluded. The material and geometric proper-
ties of the shell are: elastic modulus E 3  104 KN/m2, radius
R 100 m, height H 140 m, and thickness t 2 m. For this shell
subjected to uniform external pressure, CHASH predicted a
nonlinear bifurcation buckling load pcr 1.1829 KN/m2 and a
corresponding buckling mode ncr 6 [20]. The imperfection is
assumed to consists of the eigenvector of the critical buckling
mode n 6 and those of the two lower modes of n 4 and 5 for
 ARTICLE IN PRESS

1344 T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350

1.2 1.2

Present
Dimensionless pressure p/pcr

Dimensionless pressure p/pcr

ABAQUS n=7
0.8 n=7 0.8

0.4 0.4
Present
ABAQUS

0.0 0.0
-0.8 -0.4 0.0 0.4 -0.4 0.0 0.4 0.8
Dimensionless vertical displacement at cap apex w/t Dimensionless vertical displacement at cap apex w/t

1.2 1.0

n=7
0.8
Dimensionless pressure p/pcr

Dimensionless pressure p/pcr

0.8
Imperfection
factor 0.6
7 = 1E-5
0.01
0.05 0.4
0.1 Imperfection factor
0.4 0.2
0.3 n=6 n = 0.1, (n = 6, 7, 8)
0.4 0.2 n=7
0.5 n=8
1.0
2.0
0.0 0.0
-0.8 -0.4 0.0 0.4 0.8 -0.6 -0.3 0.0 0.3 0.6
Dimensionless vertical displacement at cap apex w/t Dimensionless vertical displacement at cap apex w/t

1.0 0.8

n=8 n=7
n=7+8 n=6+7
0.8
Dimensionless pressure p/pcr

Dimensionless pressure p/pcr

n=7 n=6
0.6
0.4

0.4

Imperfection factor
Imperfection factor 7= 8= 0.1
0.2 7= 8= 0.1

0.0 0.0
-0.6 -0.3 0.0 0.3 0.6 -0.4 0.0 0.4 0.8
Dimensionless vertical displacement at cap apex w/t Dimensionless vertical displacement at cap apex w/t

Fig. 3. Spherical cap under uniform external pressure: (a) comparison with results from ABAQUS: x7 0.01, (b) comparison with results from ABAQUS: x7 0.5, (c) effect of
inperfection amplitude, (d) effect of imperfection form, (e) effect of a multi-mode imperfection (n 7+8), (f) effect of a multi-mode imperfection (6+7).

the uniform external pressure loading condition with imperfec- harmonics around the circumference as follows [21]:
tion factors x4 x5 x6 0.2. This imperfection is illustrated in p pr a0 a1 cos y a2 cos 2y a3 cos 3y a4 cos 4y
Fig. 5a and b.
a5 cos 5y a6 cos 6y a7 cos 7y
The wind pressure is assumed to be constant along the
meridian and approximated by a combination of eight Fourier The pressure coefcients for these harmonics are given in Table 1.
 ARTICLE IN PRESS

T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350 1345

1.0

Imperfection factor 16 = 0.0001

0.01
0.05
0.1
0.2
0.8 0.3
0.4
0.5
1.0
Dimensionless axial compression x / cl

2.0

0.6

n = 16

0.4

0.2

0.0
0.0 0.1 0.2 0.3 0.4 0.5
End shortening l / t at = 0

0.8 0.8
Imperfections factor
Dimensionless axial compression x / cl

Dimensionless axial compression x / cl

Imperfections factor
15= 16 = 0.1
16= 17 = 0.1

0.4 0.4

n = 15 n = 16
n = 16 n = 17
n = 1 + 15 + 16 n = 16 + 17

0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
End-shortening l / t at = 0 End-shortening l / t at = 0

Fig. 4. Cylindrical shell under axial compression: (a) effect of imperfection amplitude, (b) effect of a multi-mode imperfection (n 15+16), (c) effect of a multi-mode
imperfection (n 16+17).

The loaddisplacement curve of the perfect shell is compared buckling load. The buckling load is just about 58% of the buckling
to that of the imperfect shell in Fig. 5c, where the negative values load of the perfect shell (Fig. 5c). The deformed circumference at
of displacements indicate that they are radially inwards. It is seen mid-height is shown in Fig. 5d. This is a simple amplication of
that the imperfection has a strong detrimental effect on the the imperfection form.
 ARTICLE IN PRESS

1346 T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350

1.2
0.6
0
-0.6
-1.2

300
1.2
0.6 240
0
-0.6 180
-1.2
120 120
100
80
60 60
40
20
0 0

1.2

Imperfect cylinder
Dimensionless external pressure p/pr

Perfect cylinder

0.8

0.4
A
Deformed shape
Initial shape
Perfect
0.0
-16 -12 -8 -4 0 Scale of displacements = 1
Radial displacement at = 0 of mid-height u(m)
Fig. 5. Imperfect cylindrical shell under wind loading: (a) imperfection surface of the cylinder, (b) imperfect circumference at mid-height, (c) effect of imperfection (d)
deformed circumference at mid-height at point A in Fig. 5c.

Table 1 closely. When the applied load reaches the bifurcation load within
Pressure coefcients of the Fourier terms
a small margin, the solution path continues to the postbuckling
Coefcient a0 a1 a2 a3 path. This method is called the imperfection-disturbance method,
similar to the load-disturbance method which involves the
Pressure 0.111390 0.269745 0.605218 0.490069 addition of a small load disturbance, as presented in Ref. [20].
Coefcient a4 a5 a6 a7 An imperfection disturbance, with the value of imperfection factor
Pressure 0.090330 0.097418 0.017047 0.048120
xn being usually set to be less than or equal to 104, is preferred in
many cases to load a disturbance as the smallness of imperfection
disturbance can be more directly judged by comparison with the
shell wall thickness. This method has been used by several
4. Postbuckling analysis of perfect shells
researchers (e.g. [10,22,23]), but its application in complex
postbuckling behaviour considering mode switching and interac-
4.1. The imperfection-disturbance method tion does not appear to have been reported before.

As a special case of the nonlinear analysis of imperfect shells

presented above, postbuckling behaviour of perfect shells of 4.2. The direct method
revolution can be closely predicted by including a very small
geometric imperfection in the analysis. The small imperfection Alternatively, based on the present formulation, a direct
acts only as a trigger for the modal deections, and the solution method for postbuckling analysis is possible as explained
provides a close approximation of the behaviour of the perfect below. A complete postbuckling analysis of perfect shells of
shell. Before the applied load is sufciently close to the bifurcation revolution by the direct approach consists of three steps:
load, the shell follows the axisymmetric pre-buckling path very (a) pre-buckling nonlinear axisymmetric analysis of the perfect
 ARTICLE IN PRESS

T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350 1347

shell, (b) determination of the bifurcation buckling load and where [K]ce is the stiffness matrix including the effect of changes in
mode, and (c) tracing of the postbifurcation buckling path. The geometry, and [Ks]ce accounts for the effect of internal stresses.
procedures for the rst two steps are described in Ref. [6]. Once These two matrices are given in Ref. [4]. In the above equations,
the critical buckling mode and load are determined, a number of only axisymmetric terms are retained. [KT]ce is thus a 12  12
adjacent buckling modes (eigenvectors) can also be determined. matrix.
Without any special measure, the solution will advance along the In the postbuckling analysis, the displacements comprise two
axisymmetric path even after the bifurcation point as the loading components: the axisymmetric component and the nonsym-
is still axisymmetric. A special technique is thus generally metric component in harmonic mode n. Therefore, the vector of
required to guide the solution path from the pre-buckling nodal variables for each element in a postbuckled state at the step
equilibrium path into the postbuckling equilibrium path. The following the critical step (i.e. the (c+1)th step) is
direct method of postbuckling analysis which is capable of (      
effecting this switching is described below. c1 qu0 c1 c1 qv0 c1 c1 qw0 c1
fdc1 ge u01 ; ; v01 ; ; w01 ; ;
The direct postbuckling analysis method consists of adding a qs 1 qs 1 qs 1
small displacement increment to the deformed state at the      
qu0 c1 c1 qv0 c1 c1 qw0 c1
bifurcation point, and continuing with the incremental-iterative uc1
02 ; ; v 02 ; ; w 02 ; ,
qs 2 qs 2 qs 2
analysis from the new position (e.g. [2426]). During this  c1  c1  c1
displacement increment, no increment is applied to the load. qun qvn qwn
uc1
n1 ;
c1
; vn1 ; c1
; wn1 ; ,
Different possibilities exist for the choice of this displacement qs 1 qs 1 qs 1
      )T
increment. It appears that the preferred approach [27,28] is to c1 qun c1 c1 qvn c1 c1 qwn c1
take an increment which is a vector parallel to a bifurcation un2 ; ; vn2 ; ; wn2 ;
qs 2 qs 2 qs 2
buckling mode (not necessarily the critical bifurcation buckling
mode), obtained by multiplying the normalized buckling mode by (22)
a suitable scaling factor. This proposal is adopted here, so the in which the subscripts 0 and n denote the axisymmetric and
displacement increment is obtained as nonsymmetric terms of harmonic n, respectively. The number of
* * * nodal variables for each element is now 24. [KT]n(c+1)
e is now a
u u 0 lt j (18)
24  24 matrix.
*
where u denotes the perturbed deformed conguration which is Therefore, as the analysis switches from the pre-buckling path
used as the starting displacement vector for tracing the post- to the postbuckling path, the number of nodal variables, and the
*
buckling equilibrium path, u 0 is the displacement vector of the dimensions of the element tangent stiffness matrix and other
last converged solution (the pre-buckling displacements at the associated vectors and matrices are enlarged accordingly. Once
*
bifurcation point), j is the normalized eigenvector of choice this switching is effected, the solution then continues like a
(usually the critical buckling mode), t is the shell wall thickness, conventional nonlinear analysis for the postbuckled shell gov-
and l denotes a scaling factor. erned by the equations presented above. This method, which
Here, for convenience, the load step during which the bifurcation switches the solution path from the axisymmetric pre-buckling
load is reached is called the critical step. The displacements at the path to the nonsymmetric postbuckling path without any
next step can be expressed in the following form: intermediate transition, is therefore called the direct method.
This contrasts with the imperfection-disturbance method where
u u0 ltun cos ny
some additional small initial geometric imperfections have to be
b b0 ltbn cos ny
(19) applied to induce the shell to deform along the postbuckling path.
v v0 ltvn sin ny An important advantage of the direct method is that the
w w0 ltwn cos ny bifurcation point and mode together with the postbuckling path
can be determined precisely without any inuence from an
where u0, v0, w0 and b0 are the converged solution for the
articial disturbance such as a small load or small displacement.
displacements of the critical step, un*, vn*, wn* and bn* are the
As no a priori knowledge is required of the buckling mode, the
radial, circumferential and axial displacements and meridional
pre-buckling path, the bifurcation buckling load and mode, and
rotations of the normalized buckling mode, n is the harmonic
the postbuckling path are all determined in a single analysis in
number of the buckling mode, and l is a scaling factor whose
which the switching from the pre-buckling path to the postbuck-
value has to be chosen carefully to successfully guide the
ling path can be automated.
solution into the postbuckling equilibrium path. If the value of l is
too small, the solution path will fall back to the pre-buckling path.
On the other hand, convergence cannot be achieved if the value is 4.3. Numerical results
too large [26].
In this approach, the number of nodal variables for each The direct method and the imperfection-disturbance method
element is enlarged when crossing the bifurcation point. In the were coded into the CHASH program. The computer program was
pre-buckling analysis, the deformation is axisymmetric. The then applied to study the postbuckling behaviour of several
vector of nodal variables for each element at the critical step example problems. The examples are used to clarify: (a) the effect
contains 12 terms and is given by of imperfection disturbance on the predicted postbuckling
  c  c  c response; and (b) the ability and accuracy of the imperfection-
qu qv qw
fdc ge uc1 ; ; vc1 ; ; wc1 ; ; disturbance method in tracing postbuckling paths involving mode
qs 1 qs 1 qs 1
 c  c  c T switching and interaction. Results obtained from ABAQUS are also
qu qv qw included for comparison with the present results. The scaling
uc2 ; ; v2 ; ; wc ; (20)
qs 2 qs 2 2 qs 2 factor l for the direct method and the imperfection factor x for the
The corresponding element tangent stiffness matrix [KT]ce is imperfection-disturbance method were set to be 0.01 and 104
then obtained as respectively in the following numerical examples unless other-
wise specied, and to zero for harmonic modes which arose due to
K T ce Kce K s ce (21) nonlinear coupling.
 ARTICLE IN PRESS

1348 T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350

0.25

Axisymmetric path
ABAQUS
0.20
Direct method
Imperfection-disturbance method
Internal pressure (MPa)
0.15

0.10

n = 29
0.05

0.00
0 5 10 15 20 25 30 35
Vertical displacement at cone apex (mm)

Fig. 6. Conecylinder intersection under internal pressure.

1.0 1.0
n = 16
Dimensionless axial compression x / cl

Dimensionless axial compression x / cl

0.8 0.8

0.6 0.6
n = 16 n = 15

0.4 n = 1 + 15 + 16
0.4

0.2 Imperfection-disturbance method 0.2

Direct method

0.0 0.0
0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6
End shortening l / t at = 0 End shortening l / t at = 0

1.2
n = 16 + 17
Dimensionless axial compression x / cl

n = 16
n = 17
0.8

0.4

0.0
0.0 0.1 0.2 0.3 0.4 0.5
End shortening l / t at = 0

Fig. 7. Cylindrical shell under axial compression: (a) postbuckling path of harmonic mode 16, (b) effect of mode switching and interaction, (c) effect of a higher mode.
 ARTICLE IN PRESS
T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350
4.3.1. Conecylinder intersection
The conecylinder intersection of an internally pressurized oil
storage tank is discussed in Section 3.1.1 of this paper. The
postbuckling paths predicted by CHASH and ABAQUS for the
critical bifurcation mode (ncr 29) are compared in Fig. 6.
In this gure, it is shown that when a very small imperfection
in the critical buckling mode (ncr 29) with x29 104 was
included in an ABAQUS analysis, the imperfection was too small to
guide the shell into the postbuckling path. By contrast, the present
analysis was able to predict the postbuckling path properly,
demonstrating the superiority of the present method to ABAQUS
in such situations. Fig. 6 shows that the predictions of the direct
method and the imperfection-disturbance method are in close
agreement, indicating that the inclusion of a small imperfection
has a negligible effect on the predicted postbuckling path.
4.3.2. Cylindrical shell under axial compression
The imperfection sensitivity of a cylindrical shell with Z 200
under axial compression is examined in Section 3.1.3. The
postbuckling paths of the same cylindrical shell of perfect
geometry predicted using both the direct method and the
imperfection-disturbance method are discussed here. Fig. 7a
shows that the postbuckling paths of harmonic mode 16 obtained
by the two different methods, with l 0.01 and x16 104, are in
very close agreement.
The imperfection-disturbance method was applied to study
the postbuckling behaviour considering mode switching and
interaction. The multi-mode postbuckling path of this cylindrical
shell for n 15+16 (with harmonic mode 1 included in the
analysis), obtained with x15 x16 104, is shown in Fig. 7b. The
effect of mode interaction on the postbuckling response is clearly
seen, but in the advanced stage of postbuckling deformation, the
postbuckling path for n 1+15+16 follows closely that of n 15.
The postbuckling path for n 16+17 is shown in Fig. 7c. This curve
is the same as the postbuckling path of harmonic mode 16, so no
interaction between harmonic modes 16 and 17 is seen in the
postbuckling process. The results of Fig. 7b and c support the
observation that re-bifurcation along the postbuckling path is
associated with reductions in the number of circumferential
waves [20]. A detailed investigation of the postbuckling behaviour
of shells of revolution considering mode switching and interaction
is given in Ref. [20].
5. Conclusions
An efcient and accurate nite element formulation for shells
of revolution with arbitrary imperfections subjected to arbitrary
loads has been presented. Both the applied loads and the initial
geometric imperfections are expressed using truncated Fourier
series, with the total number of Fourier terms required being
dependent on the complexity of both the loading and the
imperfection. Imperfections are treated as initial displacements
whose inuence can then be included into the stiffness matrix of
the structure. The formulation was coded into the CHASH
program.
The CHASH program was applied to study the imperfection
sensitivity of several typical problems. The present results are in
close agreement with results from ABAQUS, demonstrating the
accuracy of the present an analysis. The effects of imperfection
form and amplitude were investigated for a number of cases,
conrming the well-known sensitivity to imperfections of axially
compressed cylinders and externally pressurized spherical shells.
The nal example of a cylinder with a multi-mode imperfection
subjected to wind loading demonstrates the greater capability of
the present analysis than is seen in the other simpler examples.
1349
As a special case of the more general formulation presented in
this paper, postbuckling analysis of perfect shells of revolution can
be carried out using two different methods. The direct method can
determine the bifurcation point and mode together with the
postbuckling path precisely without the inuence of any
articial disturbance. No a priori knowledge is required of the
buckling mode, and the pre-buckling path, the bifurcation
buckling load and mode, and the postbuckling path can all be
determined in a single analysis in which the switching from the
pre-buckling path to the postbuckling path is automated.
Using the imperfection-disturbance method, complex postbuck-
ling behaviour considering mode switching and interaction
can be explored. Numerical results presented in the paper
show that the imperfection-disturbance method can lead to
accurate predictions of the postbuckling paths of perfect shells,
as the geometric imperfection disturbance can be made so small
that it practically has no inuence on the predicted postbuckling
response.
Acknowledgements
Both authors are grateful to The Hong Kong Polytechnic
University for its nancial support. The authors would also like
to thank Dr Y. Zhao and Dr C.Y. Song for all the ABAQUS results
presented in this paper.
References
[1] Teng JG, Rotter JM. Buckling of thin metal shells. UK: Spon Press; 2004.
[2] ENV 1993-1-6. Eurocode 3: Design of steel structures, Part 1.6: General rules-
supplementary rules for the strength and stability of shell structures, CEN,
Brussels, 1999.
[3] Wunderlich W, Obrecht H, Springer H, Lu Z. A semi-analytical approach to the
nonlinear analysis of shells of revolution. In: Noor AK, Belytschko T, Simo JC,
editors. Analytical and computational models for shells. New York: ASME;
1989. p. 50536.
[4] Hong T, Teng JG. Nonlinear analysis of shells of revolution under arbitrary
loads. Comput Struct 2002;80:154768.
[5] Teng JG, Rotter JM. Elasticplastic large deection analysis of axisymmetric
shells. Comput Struct 1989;31:21133.
[6] Teng JG, Rotter JM. Non-symmetric bifurcation of geometrically nonlinear
elasticplastic axisymmetric shells under combined loads including torsion.
Comput Struct 1989;32:45375.
[7] Rotter JM, Jumikis PT. Nonlinear straindisplacement relations for thin shells
of revolution. Research report R563, School of Civil and Mining Engineering,
University of Sydney, Australia, 1988.
[8] Teng JG, Hong T. Nonlinear thin shell theories for numerical buckling
predictions. Thin-Walled Struct 1998;31:89115.
[9] Wunderlich W, Cramer H, Obrecht H. Application of ring elements in the
nonlinear analysis of shells of revolution under nonaxisymmetric loading.
Comput Meth Appl Mech Eng 1985;51:25975.
[10] Combescure A. Static and dynamic buckling of large thin shells. Nuclear Eng
Des 1986;92:33954.
[11] Combescure A, Pernette E, Reynouard JM. Linear and nonlinear buckling of a
discretely supported cooling tower using special axisymmetric shell ele-
ments. In: Proceedings of international colloquium on stability of plate and
shell structures. Ghent, Belgium 1987. p. 28391.
[12] Riks E. An incremental approach solution of snapping and buckling problems.
Int J Solids Struct 1979;15:52451.
[13] Criseld MA. A fast increment/iterative solution procedure that handles snap-
through. Comput Struct 1981;13:5562.
[14] Ramm E. Strategies for tracing nonlinear response near limit points. In:
Wunderlich W, Stein E, Bathe KJ, editors. Nonlinear nite element analysis in
structural mechanics. New York: Springer; 1981. p. 6389.
[15] Jones DRH. Buckling failures of pressurised vessels-two case studies. Eng
Failure Anal 1994;1:15567.
[16] Teng JG, Zhao Y. On the buckling failure of a pressure vessel with a conical
end. Eng Failure Anal 2000;7:26180.
[17] ABAQUS theory and users manual, version 5.8. Pawtucket, Rhode Island:
Hibbit, Karlsson & Sorensen Inc.; 1998.
[18] Yamaki N. Elastic stability of circular cylindrical shells. North-Holland, 1984.
[19] Bulenda T. Harmonic-by-harmonic method for the postcritical computation
of axisymmetric shells. In: Valliappan S, Pulmano VA, Tin-Loi F, editors.
Computational mechanics. Balkema: Rotterdam; 1993. p. 438.
[20] Teng JG, Hong T. Postbuckling analysis of shells of revolution considering
mode switching and interaction. Int J Solids Struct 2006;43:55168.
 ARTICLE IN PRESS
1350 T. Hong, J.G. Teng / Thin-Walled Structures 46 (2008) 13381350
[21] Chan ASL, Trbojevic VM. Thin shell nite element by the mixed method
formulationParts 2 and 3. Comput Meth Appl Mech Eng 1977;10:75103.
[22] Koiter WT. Over de Stabiliteit van het Elastisck Evenwich (on the stability of
elastic equilibrium). PhD thesis, Delft University, 1945. English translation
issued as NASA TT F-10, 1967.
[23] Budiansky B, Amazigo JC. Initial post-buckling behavior of cylindrical shells
under external pressure. J Mat Phys 1968;47:22335.
[24] Maewal A, Nachbar W. Stable postbuckling equilibria of axially compressed,
elastic circular cylindrical shells: a nite-element analysis and comparison
with experiments. J Appl Mech, ASME 1977;44:47581.
[25] Soric J. Stability analysis of a torispherical shell subjected to internal pressure.
Comput Struct 1990;36:14756.
[26] Flores FG, Godoy LA. Elastic postbuckling analysis via nite element and
perturbation techniquesPart II: application to shells of revolution. Int J
Numer Meth Eng 1993;36:33154.
[27] Kiciman OK, Popov EP. Post-buckling analysis of cylindrical shells. J Eng Mech
Div, ASCE 1978;104:75162.
[28] Wagner W, Wriggers P. A simple method for the calculation of postcritical
branches. Eng Comput 1988;5:1039.