You are on page 1of 14

Thin-Walled Structures 76 (2014) 14–27

Contents lists available at ScienceDirect

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

A generalized Ritz-based method for nonlinear buckling


of thin cylindrical shells
Probal Som, Arghya Deb n
a
Department of Civil Engineering, Indian Institute of Technology Kharagpur, Kharagpur 721302, India

art ic l e i nf o a b s t r a c t

Article history: A generalized Ritz based approach that accounts for pre-buckling bending deformations as well as geometric
Received 7 January 2013 non-linearity to find the buckling load for thin cylindrical shells under uniform axial compression is
Received in revised form proposed. The approach accommodates within a unified framework, based on a shell and spring restraint
23 September 2013
system, the entire range of boundary conditions: ranging from free boundaries to compliant boundaries. The
Accepted 24 September 2013
Available online 27 November 2013
classical kinematic boundary conditions are recovered as limiting cases. The method involves choice of a
single trial function that is invariant with respect to boundary conditions. The proposed approach is
Keywords: validated using existing analytical results, where available, and numerical solutions, otherwise.
Buckling & 2013 Elsevier Ltd. All rights reserved.
Cylindrical shell
Elastic boundary condition
Bending prebuckling
Ritz method

1. Introduction buckling load. The resultant stability equations were solved using
the finite difference method.
Experimental results [7] indicate that the critical buckling Many real world applications involve boundary conditions that
stress for thin cylinders in axial compression under certain simply are ‘weak’ and not kinematically exact. The limiting case is the free
supported edge conditions is less than 50% of the classical buckling boundary condition without any restraints. Nachbar and Hoff [6]
stress obtained by Lorenz [1] and Timoshenko [2]. Koiter [16] studied the case of buckling of an infinitely long shell under free-
suggested that this over-prediction was because the classical free edge conditions and found a 37% reduction in the buckling
buckling theory did not account for the effect of initial imperfections load from the classical buckling stress. Later, Hoff and Soong [7]
in the shell geometry on the buckling stress. This paper focuses on solved the problem for a finite shell using Donnell's [8] stability
an additional cause for the discrepancy between experimentally equations and the membrane prebuckling assumption, and found
observed buckling stresses and those predicted by the classical a very low critical load for the free-free boundary condition.
theories, apart from the role of imperfections. This factor was first In this paper an analytical framework, that accounts for von-
discussed by Flugge [15], who explained that formulations that only Kármán type nonlinearity as well as prebuckling deformations
assume membrane deformations in the pre-buckled configuration that allow for bending, is developed. It allows for calculation of the
may not accurately predict the buckling load since, even in the pre- buckling load of a cylindrical shell for the entire range of possible
buckled state, radial deformations from the perfect cylindrical shape boundary conditions. It yields idealized boundary conditions such
can be observed in cylindrical shells for certain boundary conditions. as simply supported, fixed or free as limiting cases, while being
Subsequently, several authors proposed revised estimates of able to handle a wide variety of intermediate boundary conditions
the buckling load using Donnell's [3] nonlinear stability equations that are often encountered in practice. The generalized analytical
and the membrane pre-buckling assumption. von Kármán and approach, unlike the standard Ritz method, does not require
Tsien [4] modified Donnell's shell theory by including intermedi- choice of separate trial functions for each and every boundary
ate nonlinear kinematics, which allowed for finite strains but condition. A single trial function, corresponding to the free-free
assumed small rotations. Almroth [5] used the modified Donnell case is adopted for the entire range of boundary conditions.
theory but incorporated the effect of bending pre-buckling on the Translation and rotational springs connect the shell edges to fully
fixed restraints for all degrees of freedom. The stiffness of each
spring can be varied from zero (corresponding to the free case) to
infinite (corresponding to the fully restrained case) thus allowing
n
Corresponding author. Tel.: þ 91 3222283412. the entire range of boundary conditions to be modeled. The fully
E-mail address: arghya@civil.iitkgp.ernet.in (A. Deb). fixed restraints coincide with the shell edges in the undeformed

0263-8231/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.tws.2013.09.024
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 15

Nomenclature Nx1 ; N xy1 ; N y1 corresponding incremental stress resultants per


unit width
A1 ; A2 integration constants n circumferential wave number
a radius of the cylinder pa axial compressive load at shell edge per unit width
C extensional stiffness of the cylindrical shell ¼Eh= of shell
ð1  ν2 Þ Qx shear force resultant in the x direction
C ui ; C vi ; C w
i Ritz coefficients
r gl
0 ; r loc
0 offset vector between the spring restraint and the
C ui ; C vi ; C w modified Ritz coefficients shell edge
i
D bending stiffness of the cylindrical shell Eh =½12
3 Ry transformation matrix from global to local axes of
ð1  ν Þ
2 the shell
E Young's modulus of the shell u; v; w displacement fields in axial, circumferential and radial
F gl ; F loc
resultant force vector at shell ends in global and local directions
sh sh
axis of the shell u0 ; v0 ; w0 corresponding pre-buckling displacements
F gl loc u1 ; v1 ; w1 corresponding incremental displacements
sp ; F sp translational spring force vector in global and local
shell axis usp translation of the spring restraint due to preload
h thickness of the shell δ2 V second variation of total potential energy
k ð2γ Þ  1=2 x; y; z coordinates in the axial, circumferential and radial
ku ; kv ; kw translational spring stiffness values in u; v; w directions directions
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n n n
ku ; kv ; kw non-dimensional spring stiffness: ki ¼ ki ðLa=Eh Þ
n 2 Z L2 =ah ð1  ν2 Þ (Batdorf parameter)
where i¼u, v and w a1 ξ1 L=2a
kr ; kt rotational springs about tangential and twist direction a2 ξ2 L=2a
n
kr ; kt
n
non-dimensional rotational springs: ki ¼ ki ð2π La=Eh Þ
n 2 β shell rotation
where i¼r and t εx ; εy ; γ xy axial, circumferential and transverse shear strain
L length of the shell εx0 ; εy0 ; γ xy0 corresponding pre-buckling strain fields
M x ; M xy ; M y axial, twisting and circumferential moment resul- εx1 ; εy1 ; γ xy1 corresponding incremental strain fields
tants of shell γ ðh=aÞ=½12ð1  ν2 Þ1=2
M x0 ; M xy0 ; M y0 corresponding pre-buckling moment resultants ξ1 kð1 þ NÞ1=2
M x1 ; M xy1 ; M y1 corresponding incremental moment resultants ξ2 kð1  NÞ1=2
M gl ; M loc κ x ; κ y ; κ xy bending curvatures due to bending in x; y and cross
sh sh resultant moment vectors at shell edges in global and
local axes directions
M gl ; M loc ν Poisson's ratio
sp sp rotational spring moment vectors in global and local
shell ρ ratio of buckling stress and classical buckling stress:
 axes scr =scl
N pa = 2γ Eh
Nx ; N xy ; Ny axial, shear and circumferential stress resultants scr buckling stress
per unit width scl classical buckling stress as obtained by Timoshenko
∂4 ∂4 ∂4
Nx0 ; N xy0 ; N y0 corresponding pre-buckling stress resultants per ∇4 ¼ ∂x 4 þ 2 ∂x2 ∂y2 þ ∂y4

unit width

configuration, but following shell deformation and depending on assumption to find the buckling load of a cylindrical shell (Lim
the stiffness of the connecting springs, separations between shell et al. [11]), this work differs from the previous studies in that it (a)
edges and restraints occur. Constraint equations, general in form accommodates all types of boundary conditions within a single
and valid for the entire range of boundary conditions, are solved to analytical frame work, (b) uses a single trial function for all types
enforce shear and bending moment continuity at the shell edges. of boundary conditions, and (c) accounts for geometrical non-
Imposition of the constraint equations modify the free-free trial linearity and allows for an exact prebuckling solution (i.e. does
functions to ensure that they satisfy fixed boundary conditions at not assume membrane prebuckling). The paper by Almroth [5]
the restraints. Thus, starting with free-free trial functions for the
shell, by imposing the constraint equations, modified trial func-
tions for the entire shell and spring restraint system are arrived at.
A schematic line diagram illustrating this approach is shown in
Fig. 1. Fig. 2 shows the actual three dimensional cylindrical shell
with three translational springs (ku, kv, kw) and two rotational
springs attached to its edges. The translational springs ku, kv and kw
act in the axial, circumferential and radial directions while the
rotational springs kr and kt act on bending and twisting rotations
respectively. The results from the generalized Ritz based approach
are verified with published analytical solutions wherever they exist,
and with numerical solutions using the general purpose finite
element software ABAQUS where they do not. Results are found
to match well.
While other authors, such as Almroth [9] and Pinna and
Ronalds [10] have proposed solutions for elastic supports or
have used a variational approach with membrane pre-buckling Fig. 1. Schematic diagram illustrating the generalized Ritz method.
16 P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27

Fig. 2. Geometry and arrangement of springs and loading of the cylindrical shell.

incorporates (c), however, to the authors' knowledge, no existing and Nxyi are the axial, tangential and transverse stress resultants;
publications allow for (a) and (b) as well. M xi ; M yi are shell bending moments about the tangential and axial
directions and M xyi is the shell twisting moment; while Q xi the
shear force in the x direction.
2. Theoretical background

2.1. Nonlinear kinematics of the thin cylindrical shell 2.2. Rigorous prebuckling state

The strain-displacement relationships including von Kármán's Almroth [5] gives the general solution for axisymmetric pre-
intermediate nonlinear kinematics are expressed as (Almroth and buckling, assuming that there is bending in addition to membrane
Brush [12]) deformations in the prebuckling state as

1 w 1 w0 ¼ 2aνγ N 1 þ A1 sin ðξ1 x=aÞ sin hðξ2 x=aÞ
εx0 ¼ u0;x þ w20;x ; εy0 ¼ v0;y þ 0 þ w20;y ; 
2 a 2
þ A2 cos ðξ1 x=aÞ cos hðξ2 x=aÞ ð6Þ
γ xy0 ¼ ðv0;x þ u0;y Þ þ w0;x w0;y ð1Þ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
w where ξ1 ¼ k ð1 þ NÞ; ξ2 ¼ k ð1  NÞ; γ ¼ h=ða 12ð1  ν2 ÞÞ; N ¼
εx1 ¼ u1;x þ w0;x w1;x ; εy1 ¼ v1;y þ 1 þ w0;y w1;y ; pffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

a pa
; k ¼ 1= 2γ scl ¼ Eh= a 3ð1  ν2 Þ is the classical buckling
γ xy1 ¼ ðv1;x þ u1;y Þ þ w0;x w1;y þ w0;y w1;x ð2Þ s h
cl

stress (Timoshenko [2] and Lorenz [1]) and pa is the axial compression
κ xi ¼  wi;xx ; κ yi ¼  wi;yy ; κ xyi ¼  wi;xy ð3Þ per unit width of the shell. A1, A2 are integration constants whose
values were determined by Almroth [5] by imposing appropriate
where, x is the axial coordinate; y, the circumferential coordinate; kinematic boundary conditions. In the generalized formulation, these
εxi ; εyi and γ xyi are nonlinear strains defined at the middle surface constants are determined by enforcing constraint equations which
of the shell; κ xi ; κ yi and κ xyi are the curvatures; ui, vi and wi are arise naturally from enforcing force and moment equilibrium equa-
axial, circumferential and radial displacement fields; i¼ 0, 1 in tions at the shell edge for the shell and spring restraint system as
which ‘0’ and ‘1’ denote the prebuckling state and equilibrium described below.
state adjacent to the stable prebuckling state. Total displacements,
obtained by superposing the perturbation displacements on the
prebuckling displacements, can then be written as: u ¼ u0 þ u1 ; 2.3. Boundary conditions
v ¼ v0 þ v1 ; w ¼ w0 þ w1 .
The constitutive relations are As discussed earlier, for elastically restrained cylindrical shells,
  springs are attached at the edges of the shell. One end of the
1ν
N xi ¼ Cðεxi þ νεyi Þ; N yi ¼ Cðεyi þ νεxi Þ; N xyi ¼ C γ xyi ð4Þ spring is connected to the shell while the other end is connected to
2
the restraints. Essential boundary conditions are applied at the
M xi ¼ Dðκ xi þ νκ yi Þ; M yi ¼ Dðκ yi þ νκ xi Þ; M xyi ¼ Dð1  νÞκ xyi ; restraints as shown in Fig. 3.
By varying the stiffness of the translational and rotational springs
Q xi ¼ M xi;x þ M xyi;y ð5Þ
it is possible to account for the entire range of boundary condition.
where C ¼ Eh=ð1  ν Þ is the axial stiffness of the shell and
2
For instance, by setting the springs to very large (approximating
D ¼ Eh =½12ð1  ν2 Þ is the bending stiffness of the shell. N xi , Nyi
3
infinite) or very small (approximating zero) values, it is possible to
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 17

Fig. 3. (a) Undeformed and (b) deformed geometry of the shell.

recover each of the limiting cases below, for which analytical After transformation to the local shell system the spring forces and
solutions exist. moments are given by
Simply supported at both edges 8 9 8 9
> p þ w0 w0;x kw > > 0 >
< a = < =
SS1 : u ¼ v ¼ w ¼ 0 Floc Mloc  kr w0;x
0;sp ¼ 0;sp ¼ ð11Þ
0
SS2 : v ¼ w ¼ 0 >
:  p w þw k > ; >
: >
;
a 0;x 0 w 0
SS3 : u ¼ w ¼ 0
SS4 : w ¼ 0 ð7Þ Resultant shear forces and bending moments at the shell ends
can be written in vector form as:
Clamped at both edges 8 9 8 9
< N x0 >
> = < 0 >
> =
CC1 : u ¼ v ¼ w ¼ w;x ¼ 0
Floc
0;sh ¼
0 Mloc
0;sh ¼
M x0 ð12Þ
CC2 : v ¼ w ¼ w;x ¼ 0 >
:Q ; > >
:Q > ;
x0 x0
CC3 : u ¼ w ¼ w;x ¼ 0
CC4 : w ¼ w;x ¼ 0 ð8Þ In the pre-buckling step an axial preload is applied to the shell and
spring restraint system. In order to apply the pre-load, the transla-
Free–free (F–F) at both edges: for which it is assumed that the tional restraint in the axial direction is released during the applica-
shell edges are so weakly supported that it can be assumed that tion of the pre-load, while all other degrees of freedom remain fully
there are no supports at all. fixed. Following the deformation of the shell due to the pre-load,
the shell edge and spring restraints no longer coincide, as shown in
2.4. Determination of A1 and A2 for the shell and spring Fig. 3. The resultant forces at the shell edge therefore give rise to
restraint system additional moments at the restraints. The position vector from the
restraint to the shell edge determines the “offset” for this additional
The elastic springs are assumed to act in the global system, moment contribution. The offset vector in the global system and
while the shear forces and bending moments at the shell edges act following transformation to the shell local system is given by
in the local shell system, which in the undeformed configuration 8 9 8 p 9
> þw0 w0;x >
< ðu0  usp Þ >
a
> = >
< ku >
=
coincides with the global system. Shell rotations are assumed to be
small. For small rotations β , cos β  1; sin β  β. The transfor- rgl ¼ 0 r loc
¼ 0 ð13Þ
0 >
: >
;
0
>
> >
mation matrix from the global system to the local shell system is w0 :  pka w0;x þw0 >
;
u

then given by
2 3 2 3 where usp is the translation of the spring restraint due to the
cos β 0  sin β 1 0 w0;x preload and the relationship u0  usp ¼ pa =ku has been used.
6 0 7 6 0 7
R y0 ¼ 4 1 0 5¼4 0 1 5 ð9Þ The extra coupling moment due to the offset is then given by
sin β 0 cos β  w0;x 0 1
Mloc loc loc
0;ex ¼ r0  Fsh ð14Þ
Translational and rotational spring forces in global directions at
x ¼ L=2 and x ¼  L=2 can be written in vector notation as For force and moment equilibrium at the shell edges therefore,
8 9 8 9
Floc loc
0;sh þ F0;sp ¼ 0 ð15Þ
< ðu0  usp Þku >
> = >
< 0 >
=
Fgl ¼ 0 Mgl
¼  k r w 0;x ð10Þ
0;sp >
: w k >
;
0;sp >
: >
;
0 w 0 Mloc loc loc
0;sh þ M0;ex þ M0;sp ¼ 0 ð16Þ
18 P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27

Using Eqns. (11)1 and (12)1 in Eq. (15) separable and can hence be written as (Lim et al. [11])
8 9 8 9
pa þ w0 w0;x > u1 ðx; yÞ ¼ u1 ðxÞ cos ðny=aÞ ð22Þ
< Nx0 >
> = >< =
0 þ 0 ¼0 ð17Þ 
>
:Q > ; >: p w þw k > ; v1 ðx; yÞ ¼ v1 ðxÞ sin ny=a ð23Þ
x0 a 0;x 0 w
 
Using Eqns. (11)2, (12)2 and (14) in Eq. (16) w1 ðx; yÞ ¼ w1 ðxÞ cos ny=a ð24Þ
8 9 8 pa 9 8 9 where
> > ku þw0 w0;x > pa þ w0 w0;x >
< 0 > = >< >
= > < =  Γ u1  Γ u2
M x0  0  0 m x
i  1 2x 2x
>
: >
; >> > > > u1 ðxÞ ¼ f u ðxÞ ∑ C ui ; f u ðxÞ ¼ 1 þ1
0 :  pka w0;x þw0 >; :  pa w0;x þw0 kw ; i¼1 L L L
u
8 9
>
< 0 >
= x
i  1  Γ v1  Γ v2
2m 2x 2x
þ  kr w0;x ¼ 0 ð18Þ v1 ðxÞ ¼ f v ðxÞ ∑ C vi ; f v ðxÞ ¼ 1 þ1
>
: >
; i ¼ mþ1 L L L
0
3m x
i  1  Γ w1  Γ w2
The equilibrium equations in shear (Eq. (17)3) and in bending 2x 2x
w1 ðxÞ ¼ f w ðxÞ ∑ Cw ; f w ðxÞ ¼ 1 þ1
moment (Eq. (18)2) are used to find the values of the integration i ¼ 2m þ 1
i
L L L
constants. Thus the constraint equations are the following at
n¼ 1, 2, … is the number of circumferential waves, Γ 1 and Γ 2 ¼0
i i
x ¼ L=2:
for the free–free shell and m is the number of terms in the
Q x0  pa w0;x þ w0 kw ¼ 0 ð19Þ polynomial. C iu ; C iv and C iw are the Ritz coefficients.
Using Eqs. (2), (3), (22) and (24) in Eq. (21), and after integrat-
w0 kw pa ing with respect to ‘y’, the second variation of potential energy can
M x0 þ  pw0  kr w0;x ¼ 0 ð20Þ
ku be written as
Constants A1 and A2 can be found by substitution of w0 from Eq. (6) Z L=2 " (

1 2 aπ C  2 n w1 2
in Eqs. (19) and (20). The full solution is included in Appendix A. δ V¼ u1;x þ w0;x w1;x þ v1;x þ
2  L=2 2 a a


n w1 n
w0;x w1;x w1
2.5. Energy functional þ2ν u1;x v1 þ u1;x þ v1 w0;x w1;x þ
a a a a
n

o
The second variation of the total potential energy is (Almroth ð1  νÞ n n 2
and Brush [12]) þ v1;x  u1  w0;x w1
2 a a
Z Z  

1 2 C 2π a L=2 1ν 2 aπ C n 2 2
δ V¼ ε2x1 þ ε2y1 þ 2νεx1 εy1 þ γ xy1 dxdy þ N x0 w1;x 2 þ N y0 w1
2 2 0 2 2 a
 L=2 n
4 n
2
Z Z aπ D
1 2π a L=2 þ w1;xx þ 2
w1 2 þ 2ð1  νÞ w1;x 2
þ ðN x0 w21;x þ N y0 w21;y þ N xy0 w1;x w1;y Þdxdy 2 a a
2 0  L=2 n
2
Z Z 2ν
D 2π a L=2 w1;xx w1 dx ð25Þ
þ ðw21;xx þw21;yy þ2νw1;xx w1;yy a
2 0  L=2
Constraint equations, corresponding to the equilibrium equations
þ 2ð1  νÞw21;xy Þdxdy ð21Þ
at the shell edges have to be imposed on the trial functions. In
where from equilibrium of the shell the ends, Nx0 ¼  pa ; N y0 ¼ order to do this, the equilibrium equations are written in the local
Eh=w0  νpa ; N xy0 ¼ 0. shell system at the shell edges. During the perturbation loading,
Unlike the solution by Pinna and Ronalds [10], this expression the deformed shell configuration is no longer symmetric, since
does not explicitly include separate contributions to the potential rotations about both the global x and y axes are permitted. The
energy from the elastic springs in the spring restraint system. This transformation matrix, again assuming small rotations is given by
is because after imposition of the constraint equations the spring 2 3
1 0 w;x
stiffness values form part of the definition of the modified trial 6 w;y 7
Ryx ¼ 4 0 1 5 ð26Þ
functions, and their contribution is automatically accounted for
 w;x  w;y 1
when the perturbation displacements are written in terms of the
modified trial functions. Spring forces in the local system are obtained by transforming the
global spring reactions
2.6. Modified trial solutions 2 38 9
1 0 w;x > ðu  usp Þku >
< =
6 0 1 w;y 7
F loc
sp ¼ 4 5 vkv
Trial solutions need to be constructed for the perturbation > >
 w;x w;y 1 : wkw ;
quantities of the displacement fields. In the following the trial 8 9
functions are adapted from Lim et al. [11]. Their trial solution for > ðu  usp Þku þ ww;x kw >
< =
the free–free case is adopted as the unconstrained trial function ¼ vkv þ ww;y kw ð27Þ
>
:  ðu  u Þw k þ w vk þ wk > ;
which on imposition of the appropriate force and moment con- sp ;x u ;y v w
tinuity constraints becomes the trial function for the shell and
spring restraint system (see Fig. 1). During the perturbation Reaction moments due to the rotational springs in the local
loading, the restraints in the shell and spring restraint system shell system are given by
are fully fixed, and the modified trial functions satisfy the kine- 2 38 9 8 9
1 0 w;x > kt w;y > > kt w;y >
< = < =
matic (fully-fixed) conditions at the restraints, thereby satisfying loc 6 0 7
1 w;y 5  kr w;x ¼  kr w;x
M sp ¼ 4
the requirement of the Ritz method that trial solutions satisfy > >
; > : k w w k w w >
 w;x w;y 1 : 0 t ;x ;y r ;x ;y
;
essential boundary conditions. The dependence of the trial solu-
tion on the axial and circumferential coordinates is assumed to be ð28Þ
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 19

The force and moment resultants at the shell edges in the Eq. (37) yields the characteristic equation which enables
perturbed configuration are calculation of the buckling load
8 9 8 9 2 38 u 9
< Nx >
> >
< M xy > a11 … a1q > > C >
= = < i >=
F loc M loc 6 ⋮ 7
sh ¼ sh ¼ ð29Þ
N xy Mx 4 ⋱ ⋮ 5 C vi ¼0
>
:Q > ; >
: >
; > >
x 0 aq1 ⋯ aqq > : Cw >
;
i

As discussed earlier, the shell edges no longer coincide with fixed ½A ð38Þ
restraints following loading. The extra moment that arises due to
this “offset”, denoted as M loc loc
 F loc where q ¼ ð3m  sÞ, in which s is the number of constraint equations,
ex is calculated by M ex ¼ r
loc
sh
loc
where r is the position vector from the restraints to the shell in the present case being equal to 10. The entire procedure described
edges in the local system. The force and moment equilibrium above was implemented in Mathematica [13]. Eq. (25) was integrated
equations at the shell edges are then numerically using Gauss quadrature (with Kronrod extension) as
described in the Mathematica [13] documentation. A graphical
F loc loc
sh þ F sp ¼ 0 ð30Þ solution of the characteristic equation was obtained by plotting a
curve of the determinant of the matrix [A] above as a function of the
axial compressive load (pa). The value of pa for which the curve
M loc loc loc
sh þM ex þ M sp ¼ 0 ð31Þ
intersected the load axis yielded the buckling load of the shell.
where
2 38 9 8 9
1 0 w;x > u  usp > > ðu  usp Þ þww;x > 3. Finite element modeling
< = < =
loc 6 0 1 7
w;y 5 v
r ¼4 ¼ vþ ww;y
> >
; :  w; ðu u Þ  vw; þ w >
>
w;x w;y 1 :w x sp y
; For validation of the results obtained using the generalized Ritz
method, a finite element model of the shell was set up in ABAQUS
ð32Þ
[14]. Nonlinear Bifurcation Load analysis (GNA, Schmidt [17]) was
and hence M loc
ex is performed in ABAQUS on an axially prestressed cylindrical shell.
8 9 The shell was modeled using four-noded S4R as well as nine-
> ðu  usp Þ þ ww;x >
< = noded S9R5 elements, both being finite strain, finite rotation
M loc
ex ¼ 
v þ ww;y elements that are part of the ABAQUS element library. No
>
: w ðu  u Þ  vw þw > ;
;x sp ;y significant variation in the buckling stress was found for the two
8 9 element types. The subsequent numerical studies were therefore
> ðu  usp Þku þ ww;x kw >
< = performed using the S4R finite element with a converged mesh.
 vkv þww;y kw ð33Þ
>  
:  u  u w k þ w vk þ wk > ;
The results of the mesh convergence study with S4R and S9R5
sp ;x u ;y v w elements for the SS1 boundary condition are shown in Fig. 4. The
converged values of buckling load obtained using the two element
Displacements u, v and w are composed of prebuckling and
perturbation quantities. Therefore using Eqs. (27)–(33) in Eqs. (30)
and (31), subtracting the prebuckling equilibrium state defined by
Eqs. (17) and (18), and collecting the terms that are of first order in
the perturbation quantities u1 ; v1 ; w1 , the following equations are
obtained:
8 9 8   9
< N x1 >
> = > < u1 ku þ w0 w;x þ w1 w0;x kw > =
N xy1 þ v1 kv þ w0 w1;y kw ¼0 ð34Þ
>
: Q ; :  u w k  pw þ w k >
> > ;
x1 1 0;x u 1;x 1 w

M x1  w0 u1 ku þ w0 u1 kw  w1 pa þ w1 kw pa =ku  kr w1;x ¼ 0 ð35Þ Fig. 4. Mesh Convergence with S4R and S9R5 elements .

v1 kw w0;x pa
M xy1 v1 w0;x pa  w0 v1 kw  þ v1 w0 kv  kt w1;y ¼ 0 ð36Þ
ku
Thus a set of five constraint equations are obtained at each edge,
yielding ten equations in all. The trial solutions have to satisfy these
equations. On substitution of the trial solutions in place of the
perturbation quantities u1 ; v1 ; w1 in the constraint equations, it is
possible to eliminate ten Ritz coefficients (equal to the number of
constraint equations) by expressing them in terms of the remaining
Ritz coefficients. After imposition of these constraints, the trial
solutions with 3m-10 Ritz coefficients (where 3m, say, was the
number of Ritz coefficients in the unrestrained trial solution) are
the modified trial solutions appropriate for the shell and spring
restraint system. The modified trial solutions are substituted in
Eq. (25) and the potential energy minimized as

∂ðδ Π Þ ∂ðδ Π Þ ∂ðδ Π Þ


2 2 2
¼ 0; ¼ 0; ¼0 ð37Þ
∂C ui ∂C vi ∂C w
i

where, C iu ; C iv ; C iw are the modified Ritz coefficients. Fig. 5. The converged finite element mesh.
20 P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27

restraint was varied over a wide range of values. The stiffness


values in the remaining degrees of freedom were held constant at
a very large value to simulate full fixity in those degrees of
freedom. In order to enforce fixity in translational degrees of
freedom a spring constant 102 times the axial stiffness (C) of the
shell was prescribed, while to enforce fixity in rotational degrees
of freedom, the rotational spring stiffness was made equal to 104
times bending stiffness (D) of the shell. To model unconstrained
degrees of freedom, spring stiffness values 10–5 times softer than
the bending rigidity of the shell were used. These spring constants
were chosen because they were found to be sufficiently large or
sufficiently small i.e. any further increase or decrease in their
values did not change the results significantly. Stiffness values
identical to those used in the generalized Ritz solutions were used
for the finite element solutions using ABAQUS as well.
In Fig. 7, the effect of varying the rotational spring about the
bending i.e. the y axis on the buckling load is shown. The
generalized Ritz solution and the finite element solution are found
Fig. 6. Comparison of finite element results with the benchmark solution for the
to be in good agreement. It is found that bending rotational
SS1 boundary condition (Almroth [5]).
springs stiffer than 100–1000 times the bending rigidity of the
shell give results that are near identical to those obtained for fully
types are seen to be nearly the same. Mesh sizes smaller than clamped boundaries, while rotational springs 10–5 times softer
2.5 mm make no significant difference in the results; hence a than the bending rigidity of the shell give results identical to those
mesh size of 2.5 mm was used for the numerical simulations. Fig. 5 obtained when the rotation about the y axis at the boundary is
shows a typical converged mesh. The converged numerical model free. To investigate the effect of varying the twisting spring on the
was validated using the published solution of Almroth [5] for the buckling load, the rotational spring corresponding to twisting
SS1 boundary condition. The comparison is included in Fig. 6. rotations is varied while the other springs are made extremely
The material was assumed to be elastic, with the material stiff, with values as described earlier. The results are shown in
properties of steel as described below. The springs in the spring Fig. 8. It is found that changing the magnitude of the twisting
restraint system were modeled using connector elements. Three spring has almost no effect on the buckling load.
‘cartesian connector’ elements along with two ‘rotation connector’ The effect of varying translational spring stiffness in the axial
elements about the tangential and axial axes were connected to direction on buckling load is shown in Fig. 9. It is found that ku
the top and bottom edges of the cylinder. Each connector element with magnitude close to the axial stiffness of shell (C) can be
comprised two nodes, one of which was a node on the shell edge considered equivalent to an infinitely stiff constraint while pre-
while the other node, modeling the restraint, coincided with the scribing ku 10–5 times softer than the axial stiffness is equivalent to
shell edge node in the initial configuration. During the pre-load allowing the shell to freely move in the axial direction. It is
step, the axial degree of freedom at the ‘restraint’ nodes were observed in Fig. 9 that with increasing spring stiffness, the curves
released to allow application of the axial pre-load, while during are transitioning from case CC2 to case CC1, as described in Eq. (8)
the perturbation calculations, these nodes had all degrees of 1 and Eq. (8)2 above. Analytical and finite element buckling loads
freedom fully constrained. In order to model the elastic stiffness increase by nearly the same amount as the transition is made from
of the spring restraint system, the connectors were prescribed case CC2 to case CC1. The differences between the buckling load
stiffness values using the connector behavior option. magnitudes are also small. However, the threshold stiffness for
transition from CC2 type to CC1 type behavior appears to be
somewhat higher for the analytical as compared to the finite
4. Results and discussion

As in any Ritz type analysis, a convergence analysis to deter-


mine the number of terms in the trial solution had to be
performed. This was done by gradually increasing the number
terms in the polynomial of Eqs. (22)–(24). It was found that 14
terms in the trial solution i.e. m ¼14 resulted in convergent
buckling loads up to 3 decimal places.
For all the results the following geometric and material para-
meters were used: a=h ¼ 100, where a was the diameter of the
shell and h its thickness, Young's modulus E equal to 2.E þ5
N/mm2, and Poisson's ratio ν ¼0.3. Shells with a ¼100 mm and
h¼1 mm were used for the examples considered in this section.
The value of the Batdorf parameter, Z, was 244.2. The buckling load
as well as the translational and rotational spring stiffness were
expressed in terms non-dimensional ratios:ρ ¼ pa =hscl , ktran ¼
n

ktran ðLa=Eh Þ and kmom ¼ kmom ð2π La=Eh Þ.


2 n 2

4.1. Effect of varying spring restraint in a single degree of freedom

As a first attempt to validate the generalized Ritz approach, the


elastic stiffness in a single degree of freedom in the spring Fig. 7. Variation of buckling load with bending spring stiffness (kr).
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 21

prescribing very stiff rotational springs the rotational springs are


made extremely soft, a 50% reduction in buckling load is seen as
the circumferential spring transitions from extremely stiff to
extremely soft values. Similar behavior was reported earlier for
the kinematically prescribed boundary conditions described by Eq.
(7)3 and 4. The influence of varying the stiffness of the translational
spring in the radial direction is shown in Fig. 11. It is observed that
with decrease in spring stiffness, buckling load increases. During
buckling, membrane strain energy transforms to bending strain
energy. If the radial springs at the shell edges are compliant, the
shell can displace in the radial direction even in the pre-buckling
state. This results in energy building up in the bending mode,
accompanied by a relative reduction of energy in the membrane
mode. Thus the critical value of the membrane strain energy,
coinciding with the onset of buckling, is reached for comparatively
higher axial loads. This results in increase in the buckling load as
spring stiffness in the radial direction decreases.

Fig. 8. Variation of buckling load with twisting spring stiffness (kt). 4.2. Effect of varying spring restraint in two translational
degrees of freedom

Next the elastic stiffness in two translational degrees of free-


dom was varied over a wide range of values. The stiffness values in
the remaining degrees of freedom were held constant at a very
large value (as described in the previous section) to simulate full
fixity in those degrees of freedom.

Fig. 9. Variation of buckling load with axial spring stiffness (ku).

Fig. 11. Variation of buckling load with radial spring stiffness (kw).

Fig. 10. Variation of buckling load with tangential spring stiffness (kv).

element results. In Fig. 10 the effect of increasing spring stiffness in


the circumferential direction on buckling load is shown. Similar to
the behavior noticed on varying the twisting spring, in this case
too change in circumferential spring stiffness seems to have no Fig. 12. Variation of buckling load with change in axial and tangential spring
appreciable effect on the buckling load. However if instead of stiffness.
22 P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27

The combined effect of varying the translational springs in the kv springs increase in stiffness, the buckling load increases. But the
axial and circumferential directions on the buckling load is shown range of variation in the buckling load is not too high. When both
in Fig. 12. The translational spring stiffness values in the axial and the springs are very soft, the buckling load is about 9% smaller
circumferential directions are set equal and their magnitudes than the classical buckling load. The maximum difference between
varied from low to high values. Fig. 12 shows that as the ku and the buckling loads predicted using the two approaches is only

Fig. 16. Variation of buckling load with translational spring stiffness. Both rota-
Fig. 13. Variation of buckling load with change in axial and radial spring stiffness. tional springs are very stiff.

Fig. 14. Variation of buckling load with change in tangential and radial spring Fig. 17. Variation of buckling load with translational spring stiffness. Bending
stiffness. spring (kr) very stiff, twisting spring (kt) very soft.

Fig. 15. Variation of buckling load with translational spring stiffness. Both rota- Fig. 18. Variation of buckling load with translational spring stiffness. Twisting
tional springs are very soft. spring (kt) very stiff, bending spring (kr) very soft.
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 23

about 1%. However, as in Fig. 9, the threshold stiffness for the restraint in the radial direction is loosened while the restraints
transition from low to high buckling load appears to be somewhat in all other degrees of freedom are held fixed, Fig. 13 shows that
higher for the analytical case as compared to the finite element the buckling stress increases with reduction in kw. On the other
solution. The combined effect of varying translational springs ku hand, the results in Fig. 9 show that reducing the value ku only,
and kw is shown in Fig. 13. As for the case shown in Fig. 11, when while ensuring fixity in other degrees of freedom, results in a
reduction in the buckling load. Thus the dominant influence on the
buckling load for the present case appears to be the value of kw
rather than that of ku. It is also seen that for very soft springs kw
and ku, the buckling load is about 9% less than the classical
buckling load. When kw and ku are made very stiff, the solution
for fully fixed boundary, CC1, is recovered. The analytical and finite
element solutions in Fig. 13 match well for very stiff and very soft
springs. However for intermediate values there are some discre-
pancies, though never in excess of 2%. Similar to the behavior
observed in Figs. 9 and 12, the analytical solution transitions
comparatively slowly to higher buckling load values with increase
in spring stiffness, as compared to the finite element solution.
Fig. 14 includes plots showing the variation of buckling load with
spring stiffness in the circumferential and radial directions. As in
the previous case just discussed, the reduction of spring stiffness
in the radial direction appears to have a more dominant effect
than the influence of the spring kv, and results in an overall
increase in the buckling load. Again as both kw and kv are made
very stiff, the solution for the CC1 (fully fixed) case is recovered.
Fig. 19. Comparison of buckling loads for the free–free boundary condition.

Fig. 20. (a) Comparison of buckling loads for limiting case SS1. (b)First buckled Fig. 21. (a) Comparison of buckling loads for limiting case SS2. (b)First buckled
mode shape obtained from the finite element solution for SS1. mode shape obtained from the finite element solution for SS2.
24 P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27

4.3. Effect of varying spring restraint isotropically in three edge rotations are not small the ABAQUS finite element solution,
translational degrees of freedom which accounts for finite rotations in the shell formulation, may be
yielding somewhat more accurate results. However for extremely
In real life structures, cylindrical shells may be embedded soft translational springs, i.e. the free-free case, the generalized
inside an isotropic elastic media. The stiffness of the restraint Ritz and finite element results match well. Finite rotations are not
acting in each global direction will hence be equal and may cover a a factor for the free-free case. This is because the extremely small
wide range of values: extremely soft, extremely stiff or some springs in the ku and kv directions act to reduce the buckling load
intermediate value. Hence it is desirable to examine the influence to very small values. At such loads, all deformations, including
on buckling load of the translational springs, ku, kv and kw varying edge rotations, are extremely small.
isotropically, while the rotational springs are very soft. When the The small rotation assumption works well and finite element
translational springs become very soft the results should match and generalized Ritz solutions match for all values of translational
that of the free-free case, while extremely stiff translational spring stiffness when both rotational springs (kr and kt) are made
springs should replicate the result for the SS1 case described in infinitely stiff. This fixes the edge rotations, and ensures compli-
Eq.(7)1. Results are plotted in Fig. 15 for spring stiffness values ance with the small rotation assumption elsewhere in the shell.
varying between 10–6 and 10 times the axial stiffness of the shell, The results are shown in Fig. 16. The buckling stress is no longer
C. It is found that at the two limiting ends, the generalized Ritz found to decrease with decreasing translational spring stiffness.
results and the finite element results match, but for intermediate The very stiff rotational spring about the circumferential axis,
stiffness values there are some differences. The differences, which coupled with the reduction in stiffness of the radial spring kw,
are never in excess of 15%, may be due to large rotations about the ensures significant prebuckling bending deformations and conse-
edges. In the analytical model, consistent with the von- Kármán quent increase in bending strain energy. Because of this, the axial
assumptions, it is assumed that rotations are small. But as the load necessary for the membrane strain energy to reach a critical
buckling load increases with reduction in stiffness in the w value and hence cause buckling also increases.
direction, the small rotation assumption may no longer hold. It is When one of the rotational springs is made very soft and the
conjectured, subject to experimental verification, that when the other made very stiff, the generalized Ritz and finite element
solutions show a somewhat larger mismatch for some intermediate

Fig. 22. (a) Comparison of buckling loads for limiting case SS3. (b) First buckled Fig. 23. (a) Comparison of buckling loads for limiting case SS4. (b) First buckled
mode shape obtained from the finite element solution for SS3. mode shape obtained from the finite element solution for SS4.
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 25

cases. This is seen in Fig. 17 where the translational springs are stiffness value close to 10–5 times the rigidity of the shell can be
reduced isotropically while the twisting spring is very soft and the assumed to be so soft as to approximate a free boundary. Thus all
bending spring is very stiff, as well as in Fig. 18 where the bending the spring were assigned values 10–5 times the rigidity of the shell.
spring is very soft and the twisting spring is very stiff. Again, a soft The extremely small springs in the axial direction resulted in huge
rotational spring about any one axis, appears to engender some axial translations, orders of magnitude higher than the length of
differences between the generalized Ritz and finite element results. the shell, during the application of the preload. This resulted in
The maximum difference in buckling stress for soft twisting spring the ABAQUS solution failing to converge in the preload step for
and very stiff bending spring is about 25%, although this appears to L/a o0.7 due to the accumulation of numerical errors. Hence for
be limited to one particular value of the translational spring L/a ratios less than 0.7, the pre-load was applied directly to the
stiffness. The differences for soft bending spring and very stiff shell edge instead of applying it through the springs in the
twisting spring (which is probably the case more likely to be ABAQUS model. This prevented the numerical singularity from
encountered in practice), are seen to be comparatively less. Again occurring and enabled the pre-load step to converge. However it
these differences are most likely due to the differences in the way in made no material difference to the results, as was verified by
which intermediate rotation values are handled: the von-Kármán comparing the ABAQUS solutions for L/a 40.7 with and without
assumption treating the rotations as small while the ABAQUS finite this modification. Rotational and twisting springs were assigned
elements use the full finite rotation formalisms. values 10–6 times the rigidity of the shell to ensure that they had
no effect on buckling load. The same spring values were used in
4.4. The free-free limiting case the analytical model. Finite element and analytical results are
shown to match in Fig. 19. The results are also found to be close to
If the restraint at the shell edges is extremely small, the Hoff's solution that assumes pre-buckling membrane deforma-
cylindrical shell can be thought of as a freely supported cylinder. tions only. The buckling load values for the free-free case are very
The buckling load for the free-free boundary condition was first small compared to the other boundary conditions. Thus there is
derived by Hoff, but with the membrane prebuckling assumption. very little bending deformation in the prebuckling state. This
In Fig. 19, the buckling load for the free-free case is plotted for results in the close match between the rigorous pre-buckling
increasing values of shell length. As mentioned earlier, a spring solution and Hoff's membrane prebuckling solution.

Fig. 24. (a) Comparison of buckling loads for limiting case CC1. (b) First buckled Fig. 25. (a) Comparison of buckling loads for limiting case CC2. (b) First buckled
mode shape obtained from the finite element solution for CC1. mode shape obtained from the finite element solution for CC2.
26 P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27

4.5. Recovery of solutions for kinematic boundary conditions slightly higher due to the finite strain, finite rotation kinematics of
the S4R shell element. Plots of the first buckled mode shape for
It is possible to recover all the eight variants of the classical cases SS1 through SS4, obtained from the ABAQUS analyses, are
simply supported boundary condition described by Almroth included in Figs. 20(b), 21(b), 22(b) and 23(b) respectively.
(Eq. (7)) using the elastically restrained model and the modified
Ritz method. Kinematic constraints were modeled using very stiff
springs in the spring restraint system, while free degrees of
freedom were assigned extremely small values of spring stiffness. 4.5.2. Clamped
The magnitudes of the very stiff and extremely soft springs were Results are plotted in Fig. 24(a) for both the finite element and
chosen as described earlier. In all cases the solution was also found the analytical model for case CC1. Both sets of results compare
using ABAQUS, in order to compare the results. well. It is found from the finite element solution that for extremely
small lengths, due to the preponderance of edge effects, the
4.5.1. Simply supported buckling load is in excess of the classical buckling load. The
For case SS1 the results are shown in Fig. 20(a). Finite element analytical model in its present form does not include these effects.
and generalized Ritz solutions are compared. They match well and The limitation of the analytical model to ρ r 1.0 thus restricts the
are close to the published results of Almroth. Fig. 21(a) shows the generalized Ritz results to L/a ratios greater than or equal to 0.7.
results for case SS2. Again, analytical results are found to be close Results are plotted in Fig. 25(a) for the CC2 case. Finite element
to the finite element results. A strong length effect is visible for L/a and analytical solutions compare well. Again, the limitation of the
ratios less than 0.5. The results for case SS3 are shown in Fig. 22(a), analytical model to ρ ¼1.0, restricts the results to L/a ratios greater
which also shows good agreement between the analytical and than or equal to 0.7. For L/a ¼ 0.7, the buckling stress is found to be
numerical solutions. As in the published analytical results, a 50% about 14% smaller than the classical buckling stress. But for other
reduction of the buckling load is obtained for this case as L/a ratios the reduction varies between 9% and 7%. Fig. 26(a) shows
compared to the classical simply supported boundary conditions. the results for the CC3 case. Reduction of the buckling load is
SS4 results are shown in Fig. 23(a). Again, the analytical and finite approximately 9% from the classical solution. As shown in Fig. 27(a),
element solutions match closely. Both solutions show a nearly 50% for case CC4, for L/a¼ 0.7, the reduction in the buckling load is
reduction in the buckling stress. The ABAQUS buckling load is significantly higher (14%) than for higher shell lengths. Plots of the
first buckled mode shape for cases CC1 through CC4, obtained from

Fig. 26. (a) Comparison of buckling loads for limiting case CC3. (b) First buckled Fig. 27. (a) Comparison of buckling loads for limiting case CC4. (b) First buckled
mode shape obtained from the finite element solution for CC3. mode shape obtained from the finite element solution for CC4
P. Som, A. Deb / Thin-Walled Structures 76 (2014) 14–27 27

the ABAQUS analyses, are included in Figs. 24(b), 25(b), 26(b) and where
27(b) respectively.
b1 ¼ aDku kw ðξ1  ξ2 Þ
2 2

b2 ¼ 2aDku kw ξ1 ξ2
5. Conclusion
b3 ¼ ξ1 fa2 ðkr ku kw þ ðku  kw Þp2 Þ  Dðku  kw Þpðξ1  3ξ2 Þg
2 2

A Ritz based variational approach to find the buckling load of a b4 ¼ ξ2 fa2 ðkr ku kw þ ðku  kw Þp2 Þ  Dðku  kw Þpð3ξ1  ξ2 Þg
2 2

cylindrical shell that accommodates all types of boundary condi-


b5 ¼ 2aDku ξ1 ξ2 ða2 kw þ kr ðξ1 þ ξ2 ÞÞ
2 2
tions within a single analytical frame work, uses a single trial
b6 ¼ 2aDku ξ1 ξ2 ða2 kw þ kr ðξ1  ξ2 ÞÞ
2 2
function for all cases, accounts for geometrical nonlinearity, and
allows for an exact prebuckling solution is proposed. The enhanced
b7 ¼ ξ2 fa4 ðkr ku kw þ ðku  kw Þp2 Þ þ D2 ku ðξ1 þ ξ2 Þ2
2 2
Ritz based approach proposed here enables analytical solution of
þa2 Dpa ð  2ku ξ1 þ3kw ξ1 þ2ku ξ2  kw ξ2 Þg
2 2 2 2
the buckling load for non-standard boundary conditions for which
such solutions may not exist. Thus it appears to be a useful tool for
b8 ¼ ξ1 fa4 ðkr ku kw þ ðku  kw Þp2 Þ þ D2 ku ðξ1 þ ξ2 Þ2
2 2
benchmarking numerical solutions. It also yields valuable insight
þa2 Dpa ð  2ku ξ1 þkw ξ1 þ 2ku ξ2  3kw ξ2 Þg
2 2 2 2
into the buckling behavior since, unlike finite element solutions, it
has the advantage of being able to predict the buckling load for
different circumferential wave numbers. In addition, it does not
require prior calculation of the optimum magnitude of preload that References
would result in the critical buckling load. In contrast, numerical
solutions that account for the effect of geometrical nonlinearity on [1] Lorenz R. Zeitschrift des Vereines Deutscher Ingenieure. Achsensymmetrische
Verzerrungen in dü nnwandigen Hohlzylindern, 43; 1908. p. 52.
buckling load typically require evaluation of the optimum preload
[2] Timoshenko SP. Erzwungene Schwingungen prismatischer Stdbe. Z Math Phys
using a trial and error approach. This often requires several 1910;58:337–85.
simulations to be performed before the optimum value of the [3] Donnell LH. Stability of thin-walled tubes under torsion. NASA, Rep. 479. 1933.
[4] von Kármán Th, Tsien HS. The buckling of thin circular cylindrical shells under
preload is found.
axial compression. J Aerosp Sci 1941;8:303–12.
The generalized Ritz approach is validated by recovering the [5] Almroth BO. Influence of edge conditions on the stability of axially com-
classical boundary condition and Almroth's variations thereto. The pressed cylindrical shells. AIAA J 1965;4:1.
effect of restraints in each global degree of freedom on the [6] Nachbar W, Hoff NJ. The buckling of a free edge of an axially compressed
circular cylindrical shell. Quart Appl Math 1962;XX:267–77.
buckling load is investigated. Also the synergic effects of varying [7] Hoff NJ, Soong TC. Buckling of circular cylindrical shells in axial compression.
two or three springs in the global degrees of freedom on the Int J Mech Sci 1965;7:489–520.
buckling load are investigated. Overall, the solutions are found to [8] Donnell LH. A new theory for the buckling of thin cylinders shells under axial
compression and bending. ASME Trans 1934;56:795–806.
match well, though for the case of soft rotational springs some [9] Almroth BO. Influence of imperfection and edge restraint on the buckling of
discrepancies are noticed. These discrepancies are not very large axially compressed cylinders. NASA-CR432; April 1966.
and may possibly be due to differences in the rotation kinematics [10] Pinna R, Ronalds FB. Hydrostatic buckling of shells with various boundary
conditions. J Constr Steel Res 2000;56:1–16.
between the analytical approach and the finite element solution.
[11] Lim CW, Ma YF, Kitipornchai S, Wang CM, Yuen RKK. Buckling of vertical
Finally, while in this paper the focus has been on evaluating the cylindrical shells under combined end pressure and body force. J Eng Mech
buckling load under axial compression, there is nothing in the ASCE 2003;129:876–84.
[12] Almroth BO, Brush DO. Buckling of bars, plates and shells. New York: McGraw-
formulation of the generalized Ritz approach that precludes it
Hill Inc; 1975.
from being used for other loading conditions as well. [13] Mathematica, version 7.0. Champaign, IL: Wolfram research, inc.
[14] ABAQUS analysis user's manual. Providence, Rhode Island: Simulia Corp;
2008.
Appendix A [15] Flugge W. Ingenieur-Archiv 1932;3:463.
[16] Koiter WT. Over de Stabiliteit van het Elastische Evenwicht. Delft, Netherlands:
University [Ph.D. thesis. Trans. National Aeronautics and Space Agency Techni-
Solutions of A1 and A2 cal. 1945. Translation number: TT F 10833].
[17] Schmidt H. Stability of steel shell structures—general report. J Constr Steel Res.
2a2 f cos hα2 ðb1 cos α1 þ b3 sin α1 Þ þ ðb2 sin α1  b4 cos α1 Þ sin hα2 g
A1 ¼  2000;55:159–81.
ðb5 cos hα2 þ b6 cos α1 þ b7 sin α1 þ b8 sin hα2 Þ
2a2 f cos hα2 ðb2 cos α1 þb4 sin α1 Þ þ ðb3 cos α1 b1 sin α1 Þ sin hα2 g
A2 ¼ 
ðb5 cos hα2 þ b6 cos α1 þ b7 sin α1 þb8 sin hα2 Þ

You might also like