Buckling - Equations

Computers & Structures Vol 18.No 3. pp.
471-536, 1984
Printed in GreatBrntain
0045-7949/84 $3.00+.00
Pergamon Press Ltd.
COMPUTERIZED ANALYSIS OF
SHELLS-GOVERNING EQUATIONSt
DAVID BUSHNELL
Applied Mechanics Laboratory, Lockheed Palo Alto Research Laboratory, 3251 Hanover Street,
Palo Alto, CA 94304, U.S.A.
(Received 20 April 1982; received for publication 19 July 1982)
Abstract-This paper opens with a general discussion of terms in an energy functional which might be the basis
from which equations governing stress, stability, and vibration analyses are derived. The energy expression
includes strain energy of the shell and discrete stiffeners, kinetic energy of the shell and stiffeners, constraint
conditions with Lagrange multipliers, and other terms arising from the change in direction of applied loads during
deformation. Brief discussions are included of the coupling effect between bending and extensional energy needed
for the analysis of layered composite shells or elastic-plastic shells, nonlinear terms, and the form that the energy
expression takes upon discretization of the structure.
A section follows in which the energy formulation for stress, stability, and vibration analyses of an elastic curved
beam is given, including thermal effects, moderately large rotations, boundary conditions, and distributed and
concentrated loads. The matrix notation and type of discretization are introduced here which will later be used for
the analysis of shells of revolution. Terms in the local element stiffness, mass, and load-geometric matrices are
derived in terms of nodal point displacements, and it is shown how these local matrices are assembled into global
matrices. The purpose of the section is to demonstrate the procedure for derivation of the analogous equations and
quantities for shells of revolution or more complex structures.
The next section is on elastic shells of revolution. It opens with a summary of what computer programs exist for
stress, buckling, and stability analyses of such structures. The assumptions on which these programs are based are
listed and the various components of the energy functional, such as strain energy of the shell and discrete rings, are
identified and derived in terms of nodal point displacements. Included are a derivation of the constitutive law for
anisotropic shell walls and a formulation of nonlinear constraint conditions, which are required for the treatment of
segmented or branched shells with meridional discontinuities between segments or branches. Derivations of terms
in the global stiffness and load-geometric matrices and the force vector are given, with tables tracing the origin of
each term. The computational strategy for calculation of critical bifurcation buckling loads in the presence of
prebuckling nonlinearities is given, with an example of buckling under axial compression of a very thin cylinder.
This is a simple problem to formulate but a difficult one to solve numerically, owing to the existence of closely
spaced eigenvalues corresponding to nonsymmetric buckling at loads close to the load corresponding to nonlinear
axisymmetric collapse. A description of various pitfalls encountered in the search for the lowest bifurcation
buckling load is given, including estimates of the critical number of circumferential waves in the buckling mode.
Computerized formulations and run times are compared for various discretization methods, including finite
difference energy models and standard finite element models, with an example showing comparisons of rate of
convergence with increasing nodal point density and computer times required to form stiffness matrices.
Hybrid bodies of revolution are discussed next. By "hybrid" is meant a body of revolution with both
one-dimensionally and two-dimensionally discretized regions. The formulation is particularly useful for the stress,
buckling, and vibration analyses of branched shells or ring-stiffened shells in which one is particularly interested in
local effects within a distance equal to a shell wall thickness of a branch or ring. An appropriate strategy for the
solution of nonlinear problems with simultaneous geometric nonlinearity and path-dependent material properties is
described, including the development of the incremental constitutive law for the tangent stiffness method of
treatment of elastic-plastic structures. The two-dimensionally discretized regions are modeled with use of 8-node
isoparametric quadrilaterals of revolution. Details are presented on the formulation of constraint conditions for
compatibility at junctions between rotationally symmetric shell segments (one-dimensionally discretized regions)
and solid segments (two-dimensionally discretized regions).
The paper closes with a summary of linear equations for general shells. Surface coordinates, the first and second
fundamental forms, and the definition of a shell are introduced, and the assumptions corresponding to Love's first
approximation are identified. The differences in commonly used or referenced formulations are listed, including
differences with regard to kinematic relations, expressions for total strain anywhere in the thickness of the shell
wall, and expressions for stress and moment resultants. Comments are offered on which theory is the most suitable
for engineering estimates.
tThis paper describes work performed by Lockheed Palo

Alto Research Laboratory, Palo Alto, CA 94304. The work was
sponsored by Air Force Office of Scientific Research, Bolling
AFB, Washington, D.C. under grant F4%20-77-C-0122 and by
the Flight Dynamics Laboratory, Air Force Wright Aeronautical
Laboratories, Wright-Patterson AFB, Ohio under contract
F33615-76-C-3105.
The work was completed under Task 2307N I, "Basic Research

in Behavior of Metallic and Composite Ccmponents of Airframe
Structures", administered by Lt Col J. D. Morgan (AFOSR) and Dr
N. S. Khot (AFWAL/FIBRA).
The contract work was performed between October 1977 and
December 1980. The technical report was released by the Author
in December 1981.
471
D. BUSHNELL
472
SECTION 1
GOVERNING EQUATIONS-AN INTRODUCTORY SUMMARY
The majority of computerized analyses of thin shells are

based on an energy formulation, important exceptions
being the programs for shells of revolution by Cohen[l],
Kalnins [2], and Svalbonas [31 based on forward integration.
Energy expressions can be used to demonstrate the kinds
of terms that should be included in a reasonably comprehensive computer program intended to be widely used for
the analysis of stress, buckling, and vibration of practical
engineering shell structures.
STRAIN ENERGY
If the displacement method is used, the strain energy

of the shell is expressed in terms of the strains and
changes in curvature of the reference surface, which is
not necessarily the middle surface or the neutral surface
Ush]ei =
(e
+2Ne) dS
+Ce
(1)
where dS is the elemental area of the reference surface.

The six element vector e represents the reference surface strains e, e,, e,2, and changes in curvature K,, K2,
KI
and C is a 6 x 6 symmetric matrix of coefficients
which depends on the location of the wall material
relative to the reference surface, on the details of the
wall construction, on the temperature, and, if plasticity is
present and the tangent stiffness method is used[4], on
the stress-strain curve and flow law of the material. The
quantity N is a vector containing thermal expansion
effects, creep strains, and plastic strains. If plasticity is
present or if the material properties depend on the temperature, the elements of C and N at a point on the
reference surface must in general be determined by
numerical integration through the thickness. Stricklin et
al.[5] and Bushnell[4] point out that Simpson's rule
should be used for the integration. Jones[6], Ashton et
al. [7], and Ashton and Whitney[8] derive C for
laminated wall construction.
Coupling between bending and extensional behavior
If the middle surface is the reference surface, and if
the properties of the wall are symmetric with respect to
this surface, then all those elements of C are zero
through which stress resultants N, N2, N12 cause
changes in curvature K,, K2, K12 , and through which
moment resultants MI, M2 , M12 cause normal strains of
the reference surface. Generally, however, there exists
coupling of bending and extensional behavior which
cannot be eliminated by a shifting of the reference
surface. Three common examples are shells reinforced in
one direction by stiffeners that are eccentrically located
with respect to the shell's neutral surface, shells stressed
into the plastic region by a combination of stretching and
bending, and nonuniformly heated shells constructed of
temperature-dependent material. In the first example the
neutral surface with regard to bending and stretching in
one direction is in a different plane from the neutral
surface with regard to bending and stretching in an
orthogonal direction. In the second and third examples
the location of the neutral surface changes with strain
and temperature distribution. In all three of the examples
it is not possible to choose a priori a reference surface
location in order to eliminate coupling between extensional and bending terms in the energy expression. In
addition, it is often advantageous not to have to choose

the middle surface as a reference surface, since practical
monocoque shell structures often have variable wall
thicknesses and doublers which make the middle surface
difficult to describe mathematically and which cause its
position to change abruptly in space. Coupling between
bending and stretching energy is also present in shells
with composite walls such as layered orthotropic, fiberwound, or semisandwich corrugated construction.
One of the first requirements of a computer program
for shell analysis, therefore, is to permit arbitrary location of the reference surface with respect to the wall
material and to include in the mathematical model the
energy coupling between changes in curvature and normal strains of this surface. If the engineer is interested in
performing analyses of many different kinds of shell
structures, he is advised to obtain a computer program or
programs which include coupled membrane and bending
behavior. The traditional finite element model of a shell
in which membrane and bending behavior are introduced
through separate elements is not generally adequate.
Kinematic relations and nonlinear terms
In eqn (I) the strain vector e can be expressed in terms
of displacement and rotation components and derivatives
of these quantities. Kinematic relations have been given
by many authors. The nonlinear strain-displacement
equations of Love[9], Novoshilov[10], or Sanders[ll]
are acceptable as a basis for the displacement method. In
general, non-linearities need not be retained in the
change-in-curvature-displacement relations as long as the
largest reference surface rotations are less than about
200, which is usually the case. The normal strains e,, e2
and in-plane shear strain e12 can always be written so
that the highest order nonlinearities are quadratic.
There is a good physical explanation for the need to
retain nonlinear terms in the strain-displacement but not
in the curvature-displacement relations. If a thin shell
deflects a large amount, let us say an amplitude many
times the thickness, the strains are usually small even
though the deflections are rather large. Hence, the linear
terms in the strain-displacement relations will tend to
cancel each other, and the nonlinear terms will become
significant for much smaller displacements than they
would have if the linear terms had not tended to cancel.
The linear terms in the expressions for the change in
curvature, however, do not tend to cancel, and the wall
rotations must be large indeed before nonlinear terms
have to be included in these expressions.
Discretization-a brief summary
Finite element method. By far the majority of computer programs for shell analysis are based on the finite
element method. Gallagher[12-14] gives surveys of the
use of finite elements for linear and nonlinear analysis of
general shells. Brombolich and Gould[15] present such a
survey for shells of revolution. A detailed description of
the various elements with evaluation will therefore not
be presented here. Gallagher encapsulates the state-ofthe-art as of 1972: "Three alternative forms of finite
element representation of this curved shells are popular:
(1) in 'faceted' form via the use of flat elements, (2) by
means of isoparametric solid elements which have been
specialized to represent curved thin shells, and (3) via the
theories formulated directly for shallow or deep curved
shells" [13].
Until about 1970 finite element experts using the dis-
Computerized analysis of shells-governing equations
placement method were insistent that the displacements

of adjacent elements be fully compatible at the common
boundary. Maintenance of slope and displacement compatibility does have the advantage of guaranteeing that
convergence of displacements is monotonic from below
and that eigenvalues for bifurcation buckling and modal
vibrations converge monotonically from above (assuming
that in the case of vibrations a consistent mass matrix is
used). However, the enforcement of full interelement
compatibility results in an overestimation of the stiffness
of the structure, which tends to decrease the rate of
convergence as the nodal point density is increased.
Wilson[16] introduces incompatible displacement functions in order to improve the convergence properties.
A major drawback of incompatible elements used in
bifurcation buckling analysis is the tendency of the discretized model to yield spuriously low buckling eigenvalues. For example, an engineer may wish to set up a
discretized model in which the nodal points are locally
concentrated in order that local buckling near some
stress concentration be accurately predicted. However,
because of the unfortunate property of convergence
from below, this model may yield a physically unreasonable prediction of buckling in some areas where
the compressive stresses are lower but the nodal point
density is sparse. This spurious mode and others would
likely prevent calculation of the local mode for which the
locally dense mesh was originally established. The problem of spurious buckling modes becomes especially
severe in cases for which an intuitive grasp of the
expected behavior is weak.
Finite difference energy method. A few shell analyses
have been performed and computer programs written
based on the finite difference energy method, in which
the displacement derivatives appearing in e (eqn 1) are
replaced by finite difference expressions. Johnson[17]
was the first to perform such an analysis with use of an
arbitrary quadrilateral finite difference mesh. A widely
used computer program based on this approach is
BOSOR4, which treats stress, buckling, and vibration of
axisymmetric shells[18].
A good test case. Bushnell[19] presents a comparison
of the finite element method and the finite difference
energy method, showing that in certain cases the finite
difference energy method is actually a rapidly convergent
kind of finite element method in which the element
0
#.
20
I
40
473
displacements and rotations are incompatible at interelement boundaries. Figures I and 2, taken from [19],
show the results of a convergence study involving a free
hemisphere pinched by a cos20 pressure distribution.
This rather ill-conditioned problem is a very good test of
various methods of discretization. The problem is illconditioned because small forces cause large displacements. Thus, the predicted reference surface strains are
very small differences of relatively large numbers. The
dotted line in Fig. 2 is obtained with use of a half-station
finite difference energy method, which is equivalent to a
finite element method based on linear functions for u and
v and a quadratic function for w.Detailed descriptions of
the finite elements are given in [19,201. Users and developers of computer programs for shell analysis and
for general structures are urged to employ this case in
order to determine the adequacy of the shell elements in
the finite element libraries of their programs.
Discretized kinematic relations. With use of a discretization method, analytical kinematic relations e=
L(d), where L is a nonlinear differential operator and d
is the displacement vector, can be expressed in the
algebraic form
e, = BLdi + dTBNLdi.
The vector e, represents reference surface strains and

changes in curvature at some point i; di is the local nodal
point displacement vector associated with i; and BL and
BNL are 6 x m matrices dependent on the local reference
surface geometry and mesh spacing at i. The number of

columns m of BL and BNL depends on how many nodal
degrees of freedom are used in the discrete model. If eqn
(1) is expressed in discrete form, and if the r.h.s. of eqn
(2) is substituted into it, the strain energy expression for
the thin shell becomes a quartic algebraic form in the d, if
C and N are independent of di.
Stiffener strain energy
Many practical shell structures are reinforced by
stiffeners. Depending on the configuration these might be
"smeared out" or treated as discrete elastic structures.
"Smeared" stiffeners. If there exists a regular pattern
of reasonably closely spaced stiffeners, their contribution
to the wall stiffness of the shell or plate might be
modeled by an averaging of their extensional and bend-
ARC LENGTH, i
60
80
I
(inches)
100
120
-_
I -_
140
-l
-2
w
E * 107psi
9
u
6
pcos 2
EOBES FREE
4
2
u.
,I
(2)
l
.
Fig. 1. Meridional and normal displacements at 0 = 0 of a hemisphere with a free edge subjected to pressure
p = cos 20 (from Bushnell[19]).
D. BUSHNELL
474
8
0
- - -
S
etion; Three and Five Gauss Points
Finite Element
Cubic u, v, w;
p~cos 28 Adz/
ti
Difference (BOSOR4)
- - @- - - - - - -Z:
-
0 -Finite
-0 9- - JI o ---
Lu
CL,
Vf)
a
4
No Static Reduction;
Two Gauss Points;
Khojosteh-Bakht
Type Element;
Linear u, v
Cubic w;
Finite Element
-J
20
40
60
NUMBER OF MESH POINTS
80
100
Fig. 2. Comparison of convergence of finite element method with finite difference energy method (from
Bushnell [19]).
ing rigidities over arc lengths equal to the local spacings
between them. Thus, the actual wall is treated as if it
were orthotropic. This "smearing" process accounts for
the fact that the neutral axes of the stiffeners do not in
general lie in the plane of the reference surface of the
shell wall. Predictions of buckling loads and vibration
frequencies of stiffened cylinders have been found to be
very sensitive to this eccentricity effect. A general rule
of thumb for deciding whether to smear out the stiffeners
or to treat them as discrete is that for smearing there
should be about 2-3 stiffeners per half-wavelength of the
deformation pattern. It may be appropriate to smear out
stiffeners in a buckling or vibration analysis but, because
of local stress concentrations caused by the stiffeners,
not in a stress analysis. The stiffeners can be smeared as
an analytical device to suppress local buckling and
vibration modes. In order to handle problems involving
smeared stiffeners, a computerized analysis must include
coupling between bending and extensional energy as
described earlier. The paper by Baruch and Singer[21] is
a classic in the field of stiffened shell analysis.
Discrete stiffeners. If the stiffeners are so far apart that
significant variations of displacement and stress occur
between them, then they cannot be averaged over the
entire shell surface but must be treated as discrete onedimensional bodies. The standard approach is to assume
that the cross section of the stiffener does not deform
but that it translates and rotates in a fashion compatible
with the shell to which it is attached. If plane sections of
the stiffener remain planar and normal to the reference
axis, the strain energy can be written ina form analogous to
that for the shell:
Ustiffer
~=
2f
(e,-Ge, + 2N,e,) dL
represents the reference axis normal strain e,,, changes

in curvature KI, K,2 in two orthogonal planes, and twist
K, 12 . G is in general a full 4x4 symmetric matrix of
coefficients which depends on the location of the
stiffener material relative to the reference axis, on details
of the stiffener construction, on the temperature, and if
plasticity is present and the tangent stiffness method is
used, on the stress-strain curve and the flow law of the
stiffener material. The vector N, is analogous to N in
eqn (1). If plasticity is present or if the material properties of the stiffener are temperature dependent, the elements of G and N, at a point on the reference axis must
in general be determined by numerical integration over
the stiffener area. The reference axis strain e,,, changes
in curvature K,,, K,2, and twist K, 12 can be expressed in
terms of the displacement and rotation components and
derivatives of these quantities referred to the stiffener
reference axis. With appropriate discretization, the
stiffener strain vector e, can be expressed in algebraic
form as
e=i = BLdi + d iBNLdi
(4)
in which all quantities are analogous to those in eqn (2).

Since the reference axis of the stiffener does not in
general lie in the plane of the reference surface of the
shell, the local displacement vector dj must be expressed in terms of the local shell reference surface displacement di
d-i = Edi
(5)
so that eqn (4) in terms of the dependent variables di

becomes
(3)
e-i = BLEdi + djTETBNLEdi.
where dL is the incremental length along the reference

axis of the stiffener. The four-element vector e,
(6)
If eqn (3)is expressed in discrete form and if the r.h.s. of

eqn (6) is substituted into it, the strain energy expression
for the stiffener becomes a quartic algebraic form which
is added to the shell strain energy.
LOADING
Two aspects of loading are of particular interest when

thin shells are involved: (I) "live" load or following loads
vs "dead" or constant-directional loads; and (2) loading
by means of enforced displacement vs loading by prescribed external forces.
Live loads
A "live" or following load is a load the direction of
which changes as the shell surface rotates. The expressions for work done by the external forces distributed
over the shell surface and along the discrete stiffeners
are, respectively
WA.11
Wstiff-ner
(pd + dT Pd) dS
(7)
L (qd, + dTQd,) dL.
(8)
The second terms in each integrand represent the live

load effect. This effect should be included if the
deflections or rotations are moderately large or, in modal
vibration of bifurcation buckling problems, if the halfwavelength of the mode is the same order of magnitude
as the smallest principal radius of curvature. Two examples in which the live load effect is significant are the
bifurcation buckling or nonlinear collapse of a very long
cylinder under external pressure and that of a ring under
external radial compression. Inclusion of the live load
effect lowers the predicted failure loads by about 30% in
these cases.
Displacement vs force loading
Loading may be applied by means of a controlled
displacement distribution (such as uniform end shortening of a cylinder) or by means of a controlled force
distribution (such as uniformly applied axial force). A
given thin shell structure may behave very differently
under these two loading conditions. If a boundary displacement is imposed, a significant amount of stress
redistribution can occur. Flexible or "soft" parts of the

structure deform considerably with more load subsequently being taken up by the stiff or "hard" parts.
Figures 3 and 4 show shells for which this type of
behavior occurs. At an axial load of about 200 lb the flat
portions of the pear-shaped cylinder shown in Fig. 3
begin to bend. The load initially carried by these portions is transferred to the curved parts, which absorb an
increasing percentage of the total load until these also
buckle, resulting in a decreased load-carrying capacity. A
similar phenomenon occurs in the case of the axially
compressed cylinder with an elliptical cross section
shown in Fig. 4. The perfect cylinder (( = 0) buckles at
point A in a mode Aw shown in insert (a). However,
significant post-buckling load-carrying capacity is exhibited at B and finally at C because the load initially
carried by the flatter portions of the cross section (S =
2.2) has been transferred to the more highly curved
portions (S = 0). Less stress redistribution can take place
if the boundary forces ar, imposed, leading in general to
earlier failure than for cases in which boundary displacements are imposed.
KINETIC ENERGY
In thin shell analysis it is not necessary to include

rotatory inertia of the shell wall. There is a stronger case
for including rotatory inertia of the discrete stiffeners,
however. The kinetic energy of the shell and the stiffeners
has the analytical form
K.E. =
f1
dT d dS+ f(drTmrd +6.i7Ird,)dL
(9)
in which () indicates differentiation with respect to time,
m is the mass/area of the shell reference surface, m, is
the mass/length of discrete stiffener reference axis, wr is
the rotation vector of the stiffener reference axis, and I,
is a matrix of rotatory inertia components of the stiffener
referred to its reference axis. As before, various transformations are used in order to express all quantities in
terms of the shell wall displacements. Whether or not the
mass matrix is diagonal depends, of course, on the
discretization model and the choice of nodal degrees of
freedom.
25001
2000 -
C1500
00
0.
)'
wo
C
0
4
NORMAL DISPLACEMENT AT l8W AND MIDLENGTH-IN
Fig. 3. Noncircular cylinder subjected to uniform end shortening (from Almroth and Brogan [221).
CAS Vol. 18, No. 3-G
475
D. BUSHNELL
476
Arc length, S
,rc length, S
20
2 5
30
3 5
40
45
End shortening/shell thickness
Fig. 4. Load-deflection curves for axially compressed perfect and imperfect elliptic cylinders
(from Bushnell et al.[23]).
homogeneous quartic functional
BOUNDARY AND OTHER CONSTRAINT CONDITIONS
The energy minimization problem (displacement

method) is subject to constraint conditions corresponding to behavior at the boundaries of the shell or other
locations within the domain where certain relationships
between nodal point displacements are postulated to
hold. These conditions may be linear or nonlinear. Two
types of nonlinearity may exist: the first may result from
continually changing geometry as loads are varied; the
second may result from a sudden change in behavior as
one part of a structure contacts another.
The constraint conditions might be introduced into the
analytical model by means of Lagrange multipliers or by
appropriate elimination of rows and columns of stiffness
matrices. If the Lagrange multiplier method is used,
a general nonhomogeneous, nonlinear constraint condition might assume the form
U, =A[da - TLd, - dITNI dh d,
H=ff(q,
q;
t)df
(I I)
tI
can be obtained in which the coefficients as well as the

displacements may be time dependent. (Note that damping as well as fluid or soil structure interaction effects
have been omitted in the above development.) Numerical
solutions for problems involving linear and nonlinear
static stress, bifurcation (eigenvalues) buckling, modal
vibration with prestress, and linear and nonlinear
dynamic response can be based on this functional.
Through minimization with respect to the nodal point
variables q, a set of simultaneous algebraic equations is
generated. The nature of these equations and the best
numerical methods for their solution depend on the type
of problem that is being solved.
(10)
SECTION 2
in which U, denotes an energy-like term pertaining to

constraint conditions; A is a vector of Lagrange multipliers; da and db are displacement vectors at different
points, a and b, in the structure; and do is an applied
displacement.
A total energy expression H can be constructed from the
r.h.s. of eqn (1)-(10). For a branched, segmented stiffened
shell of revolution, the expression H might include:
* strain energy of shell segments, including smeared
stiffeners (1)
* strain energy of discrete rings (3)
* potential energy of applied loads (7), (8)
* kinetic energy of shell segments (9)
* kinetic energy of discrete rings (9)
* constraint conditions for boundaries (10)
* constraint conditions for junctions between shell segments (10)
With appropriate substitutions of discretized displacement components q for continuous variables and numerical integration over the shell reference surface
and over the lengths of discrete stiffeners, a non-
ANALYSIS OF A CURVED BEAM
The foregoing discussion will be illustrated by a onedimensional example-a curved beam shown in Fig. 5.
The total energy expression H is given by
H = U-W+ U -TT
(12)
in which U, = strain energy; W = work done by external

loads; U, = constraint conditions (Lagrange multiplier
formulation); and T = kinetic energy.
STRAIN ENERGY
The strain energy in the beam is given by

U
.E dV
I ;
4-
(13)
vo
in which or is the stress and e. is the strain that produces

0,:
e=E- aT.
(14)

REFERENCE SURFACE
-
---
s= L
s=0
Using eqns (14), (15) and (18), and assuming that the
beam is of unit depth normal to the plane of the paper,
one can write eqn (13) in the form
Us =2-
Fig. 5. Curved beam of developed length, L.
477
(Le,
K]
-Ezz] {K }
-2EaT [1, -zJ{e}+
The quantity e is the total strain and T is the temperature rise above the zero-stress state.
It is assumed that plane sections remain plane, normal
sections remain normal, and the beam deforms only in

the plane of the paper. Displacements and coordinates
are shown in Fig. 6.
The total strain e(s, z) can be expressed in terms of the
strain and change in curvature of the reference surface:
e(s, z) = e(s) - ZK(S)
(15)
(19)
el+2NT,MTJf e)ds
(20)
e = duldx + w/R +If 2
Cl=|
(16)
C 2,
l U
= J, ([e,KJ [C
in which
= dflds
dzds
in which [j denotes a row vector, [I a matrix, and {} a

column vector.
The quantity [e, KJ denotes a row vector with two
elements, e and K. Equation (19) is derived in Fig. 8.
Integration with respect to z, and neglect of the term
Ea2 T2 , which does not contain any dependent variables,
leads to
and the reference surface quantities e and K can be

expressed in terms of the displacements:
Ea2T2)
Edz;
C1 2
-fEz dz; C 2 2 = Ez2dz
NoT=-f EaTdz;
MT- fEaTzdz.
(21)
where
3= dwlds - uIR
(17)
The strain energy can be expressed in terms of the

displacements and their derivatives with use of eqns (16)
and (17).
R(s) is the local radius of curvature of the reference
surface of the beam, and /3 is the rotation as shown in

Fig. 7.
The stress o-is given by
(a = E(e - aT)
EXTERNAL LOADING
Suppose that the beam is submitted to loading as

shown in Fig. 9(a). The reference surface is considered
to be loaded by constant-directional distributed pressure
(18)
p. and traction p,, and constant-directional end loads, V,
where E is the modulus of elasticity and a is the

coefficient of thermal expansion.
H and M.
The work done by the applied loads shown in Fig. 9(a)
is
W=
REFERENCE
SURFACE
(pu +pw) ds + HuL + VWL*+ MI3L
(22)
provided the loads act in a constant direction as the

beam deforms.
In matrix notation, eqn (22) becomes
W=
zw
JS 1,POj
ds +[HV
(23)
Fig. 6. Coordinates (sz) and displacements (uw).

CONSTRAINT CONDITIONS
Suppose the beam is supported as shown in Fig. 10.

The terms in U, (eqn 12) are
UC = Ao u*(O) + Ao2w*(0) + A03/,(O)
R(s)
+ ALIO +
(24)
AL 2w*(L) + AL3 0
or in matrix notation
U~= [A01, A02, A03

Fig. 7. Local radius of curvature R(s) and rotation A.
{W
0
AI
[ALI()0
LA
s*j
IL
(25)
D, BUSHNELL
478
13)
i5
'
= - aT
sJ
e - ZK
E(Ll,
II
E J
14)
IC
- AT)
( 16)
T) 23
-zJ {J-
IT)
daids
Us
= fs
Ef
_!-z 4,OY - 24aT ti, -Z
AT
:,.
-44
LfS
13$
{I
Il*Z}
{
B
B
{I,} - EWTI dzds

LET
I , zJ I}
EdT
\__1__
XE
-E2
t Ez
EzI
CI
I
J 2EaTLi,
12
dzds
2LIT, MTj
-zJ dz
C22]
[LI 2
Fig. 8. Derivation of eqns (19) and (20).
Equation (25) can be written in a more general form
f*
3x3
A~.
U' = [Ao] tKA] 3W"'
I/
3X3
+ [XLJ [Ks]
UK
*
w*
o0
where
iM
n 0
[Ml C-I0 m 0 .
0 0 L.
(26)
L
where, in the case of eqn (25)

3x3
[-
tKA]
l ;
DISCRETIZATION
3x3F
1 B]
(27)
KINETIC ENERGY
The kinetic energy of the beam is

T
(31)
[m(a 2 + 2)+ IJ421ds
(28)
in which () denotes differentiation with respect to time,

m is the mass/length of reference surface arc, and I, is
the mass moment of inertia of the beam cross section.
If one is concerned with modal vibrations
Now assume that the beam is modeled as a series of

discrete elements, with nodal point displacements distributed as shown in Fig. It. The continuous dependent
variables (u, w) must be expressed in terms of nodal
point quantities.
In the special case with the degrees of freedom distributed as shown in Fig. lI the finite elements are most
appropriately chosen to extend between adjacent upoints. The energy in the beam is then given approximately by the sum over the number of elements of
the energy density at the midlength of each element
times the arc length hi of that element.
If the nodal point spacing is constant, the energy in the
ith finite element in Fig. 11, e.g. is evaluated at w;. At this
point the quantities appearing in eqns (16) and (17) can
be written in terms of nodal point displacements as
follows:
u =(ui I + ui)12
(a, w)= i- (ZI, w)
(29)
(32)
w-w
/ , (win,- wi- )/(2hi) -a(Used + ui)I(2Ri)
in which A is the modal angular frequency.

In matrix form eqn (28) becomes
e.= (ui - ui,)!h, + wRi +2, 2

T= -_fl, J
[wp
wMds
(30)
K;=(iwi-2w,+ wi
-
)1hi
[(ui + u, 1)12](/1R,)'
(33)
(ui - ui_,)1(h,RJ
479
S =L
,p
.H
(a)
Pt
(b)
(u*,w*)
(horizontal,
vertical)
(u'W)
(tangential,
normal)
Fig. 9. (a) Loading on the beam; (b) Displacement notation.
HINGE ON ROLLERS
AT s = L
CLAMPED
0;
AT s = 0
Fig. 10. Boundary conditions.
in which ()' denotes differentiation with respect to

the reference surface arc length, s.
In matrix form eqns (32) can be written as
w }[D]{q}
J
(34)
and
5x5
5s1
Ix5
[BNLI={R} [RJ
(39)
IRj is given by the third row of [D]:
in which the vector
[P] = [-l/(2hj),-l/(2Rj), 0, -1/(2Rj), +1/(2hi)J.

(40)
with
[DI =
0
- 1/(2hi)
0
1/2
0
- 1/(2Ri) 0
0
0
1/(2hi)
1/2
0
- 1/(2Ri)
(35)
and
[qj = {q}T =
In the derivation of the total energy functional H of the

ith finite element, matrix formulas for Lu*, w*, 03j,
which appears in eqns (23) and (26) are also needed. If
the angle between the tangent to the reference surface
and the horizontal is c, as shown in Fig. 9, then the
vector [u*, w*, P3 is related to the vector of nodal point
degrees of freedom by
U*
(36)
[w ,u,
1 u,-,, wi, ui, wi+ij.
The reference surface strain and change in curvature are

given in matrix form by
3x5 33
w=
3
[T][D]{q}
in which
cosq
[T]=I sino
{}
q4+
= [B]{q}+
2 [q] [BNL]{q}
(41)
-sino
cosO
0 .
(42)
TOTAL ENERGY OF AN ELEMENT
(37)
With use of eqns (20) and (37) for U.; eqns (23), (34)
and (41) for W; eqns (26) and (41) for U,, and eqns (30)
and (34) for T, one can form the expression for the total
with
2.s
-1/hi
hlRi
1/hi
I2
L2/i
(h 2
I2ih HR2
) i)hIR
(38)
D. BUSHNELL
480
-THE iTH FINITE
ELEMENT
Fig. 11. Discretized niodel

iodel of the curved beam.
beam.
'energy" Hi of the ith finite element
The 5x5 matrix [K]k=hk[BIT[CJ[B] is called the local
Hi = Ui - Wi + U1
1 - T1
(43)
in terms of the nodal point degrees of freedom. The

strain energy of the ith finite element is:
element stiffness matrix of the unloaded, undeformed

finite element. The (i, j)th member of this matrix K'' is
generated by differentiating Uk with respect to q, and q,, or
calculating the second variation of Uk:
2U,,
U.hi
rt.5 5-21
lI
[ qI [B] ]
|2
+24NTMTJ
i"
([B]qi+|
12
(in0
C] [B] iq}+
(44}
I}]
The work done by external loads is:

W-= [p., p, 0] [D]{q~h, + [H, V, M] [T][D]{q}&j.(45)
The constraint condition "energy" is:
uci
= [AoJ [KAI[T][D]{q18oi+[AL] [KR][T][D]{q}BL'
(46)
and the kinetic energy is:
Ti= (212) Lq [D] T [M][D]h,{q}.
(47)
Note that in eqns (44), (45) and (47) the reference axis
arc length increment dsi has been replaced by the nodal
point spacing hi. The Kronecket deltas appearing in eqns
(45) and (46) are defined by
6&,=0 if ii,;
81'=1 if i=iL
(48)
6,' =
(50)
a qi dq;
2l~5,1 [1l32
if i70io; &'i=l if i=io
in which iLmeans "i at s=L"; iomeans "i at s=0". At

the ends of the beam the energy density is multiplied by
one-half the spacing between adajcent u-nodes.
The global stiffness matrix [K] for the entire beam is

calculated by assembling or accumulating local element
stiffness matrices into a 'master" or global array, as
shown in Fig. 12. The positions of the filled members of
the global stiffness matrix depend on the numbering
scheme used for the nodal point degrees of freedom. For
the simple example of the beam, the nodal point degrees
of freedom are logically numbered in increasing order
from left to right, resulting in compact storage of each
5x5 local stiffness matrix within the NxN global array,
where N is the number of degrees of freedom of the
entire discretized model. As seen from Fig. 12, this
numbering scheme results in a global stiffness matrix
which is narrowly banded about the main diagonal. Solving such one-dimensional equilibrium, vibration, or
stability problems on the computer requires much less
computer time and storage than do problems of higher
dimensionality.
Figure 13 shows the lower triangular part of a stiffness
matrix for a more complex "branched" one-dimensionally discretized structure. Included in the matrix are
terms of the type K j,, indicated by x's, boundary condition terms of the type
(51)
(Q]=[KAIIT [D]
from eqn (46), and juncture condition terms, not yet

described, corresponding to stations at which Segment
O) is fastened to Segment (D. Although the bandwidth
of this stiffness matrix is locally large at structural
branch points, the average bandwidth becomes relatively
smaller as the nodal point density in the two structural
segments is increased, resulting in inexpensive computer
runs.
The local mass matrices

LOCAL AND GLOBAL STIFFNESS AND MASS MATRICES
If the nonlinear term (1/2)j 2 in eqn (33a) is neglected,

the strain energy Uk of the kth finite element is
simplified:
U,
[Mk"h,[DDI T [MI[D]
are assembled into a global mass matrix, and the local

force vectors (eqn 45)
IMrqIhJ[B] [C][B][q}+2h, N ,MT][B]{q}1

5x5
WK]".
(49)
(52)
{F,,j"= hRP , p,, 011[D]
(distributed>
V loads
I
(53)
481
U0
w2
U2
w3
U3
w4
U4
w5
u5
Fig. 12. Local and Assembled matrix architecture.
boundary or
{Fb}5 = [H, V, MJ[ T][D]fIL"
concentrated
loads
are assembled into a global force or "right-hand-side"

vector in a manner completely analogous to that just
described in connection with the stiffness matrix.
If the discretization scheme shown in Fig. 11 and
specified by eqn (32) is used for derivation of the mass
matrix, this matrix will have the same form as the
stiffness matrix, part of which is shown in Fig. 12. In
order to obtain a diagonal mass matrix, one must assign a
lumped mass to each displacement degree of freedom.
For example, in the interior of the beam shown in Fig.
II, half the mass of element i might be assigned to nodal
degree of freedom ui-,, half to ui and the full mass to wi.
The boundary and segment junction conditions
represented by the matrices [Q] and [D] in the global
stiffness matrix shown in Fig. 13 would be filled with
zeroes in the corresponding global mass matrix.
EQUILIBRIUM, BUCKLING AND VIBRATION
According to the principle of stationary energy, or

minimum potential energy, a structure is in equilibrium if
aH
- 0
axi
i-=l, 2,.,..N
(55)
where xi represents a nodal degree of freedom or a

Lagrange multiplier, and N is the total number of
degrees of freedom in the system including the Lagrange
multipliers. The terms in the global matrices governing
equilibrium, buckling, or modal vibration can be derived
by application of this principle to each finite element of
the structural system.
Equilibrium
Because of the appearance of (i2 in eqn (44), eqn (55)
represents a system of simultaneous nonlinear algebraic
equations. These nonlinear algebraic equations are
solved with use of the Newton-Raphson method. The
first variation aHi axi is expanded in a Taylor series about
a known solution, {xo), with retention up to linear terms
in {Ax} only:
aH/dax =aH(x0 +Ax)/8xi=aH(x0 )/axi+ -d(AAxj =0

axiax,
(56)
i=1,2,....N.
Equation (56) is solved for Axj,j=1,2... N; a new trial
solution {xo+Ax} is then available; and the solution of
eqn (56) with {x0 } replaced by {xo+Ax} is carried out for a
new {Ax}. Iterations continue until J{Ax/xlI is smaller than
some prespecified number, (In the discussion of the
482
D. BUSHNELL
Fig. 13. Stiffness matrix configuration for 2-segment, one-dimensionally discretized structure with intermittent
fasteners (from Bushnell[18]).
Newton-Raphson method the kinetic energy is assumed

to be omitted from eqn 12.)
Since {xo} is an equilibrium state, the first term of eqn

(59) is zero, so that the bifurcation buckling equations
become
Bifurcation buckling
Figure 14 shows a load-deflection curve with a bifurcation point at (p_,lxolJ), in which IlxoII
is a generalized
displacement conjugate to the load Pcr
Since the bifurcation point is on the equilibrium path
O-A, it is known that
x,=0 i=1,2 ... N.
x
x.
Ox
1
b =0
(60)
i=1,2 .... N
These equations are linear and homogeneous in x, j=

1,2,...N. A non-trivial solution {x"} exists only for certain discrete values, the eigenvalues, of the matrix
(57)
H(xo)1axi"axj", iJ=l,2,...N.
a2
In order to determine if {xo} is a bifurcation point, one
must check to see if

H(XOlXb) =0 i=1,2,...N
(58)
where {xb} is a non-trivial infinitesimal buckling mode.

Expansion of H in a Taylor Series about {xo}, as before,
yields
aH(xo)+
H(xo)Xib+h o t.=0 i=1,2,
=I dxj axj x+o
N.
(59)
(61)
General equations
In order to solve equilibrium and buckling (or vibra-
tion)
problems,
one
must
obtain
the
vector
aH(xo)lax, i = 1,2,.. .N and the matrix a 2 H(xo)l

axiaxj,i,j = 1,2,.. N. To derive these quantities one
starts from eqns (44) to (47), assuming that the
displacement state {q} is given by {qo+q 5b },

where {q } and gb are considered to be small compared
to {qo} and go, and the superscript b may signify either a
correction to the trial solution in the Newton-Raphson
iterations or a buckling or vibration modal quantity. One
pr
483
If xi is a Lagrange multiplier one obtains the same

expression as eqn (65) except the fourth line in eqn (65)
11041is replaced by
Bifurcation Point,
Post-Buck ing Curve
[qo] {[[KA] [T] [D]]T }I 6o"+ [qo] {[[KB] [T] [D]]T}i}
PreBuckling
curve
Pllx
DEFLECTION-
Fig. 14. Bifurcation buckling at pc, from prebuckling equilibrium
The elements in the generalized stiffness matrix

a2 H(xo)Iaxiaxj are derived in a similar way. The contribution of the first line of eqn (44) to the second
variation of H is obtained by differentiating the r.h.s. of
eqn (64) with respect to q4b and allowing {qb} to vanish:
state determined from nonlinear equations.
a2H(q0)
aq
1 aql
inserts
{q}={qo+qb}; 0=3
in which [R] is given by eqn (40).

Dealing with just the first line of eqn (44), assuming
that {q} = + qb}, and making use of eqn (63) for 1,
one finds that this first line of eqn (44) contributes the
following terms to the first variation of H for the kth
finite element:
{qo
[B] +
[B]T
+I?,RJ, OJ
(C]
[B] {qo}+ j2
})
})]
(67)
The r.h.s. of eqn (67) can be simplified by combining

lines I and 4 and lines 2 and 3. Doing so, adding the
kinetic energy term derived from eqn (47), and including
the remaining terms from eqns (44) and (46), one obtains
for the second variation of H associated with the kth
finite element:
2
H(qo) a2H(x)
++2p
'0)
go2
(63)
0=[Rl{q}
Fqo+qi
oJ)[c] oI}
F([q.o
cl
L
+([Bj]+[PoR,OJ)[C]({B};+0
Note that the rotation 3is given by
aH(qO+-q)=hk
hk
+([Bj j + [okoR 1,0J)[CI({B}{

+{oRi})
into eqns (44)-(47), then differentiates with respect to xi,

which represents one of the components of the vector
{q}.
Lq-q5
(62)
+0b
6L.
(66)
[(oqo)[B]
[-2,oJ)[C]{
I[C]
+(QBi]+
+[o ROD)[C] (JBj}+
X({Bi}+IjORi+0bRil)
(O}
(64)
+[N
, MTJ Rit ]hk
+([BiI+ [g3oR+0bRiOi)[C]
+[[KA]J[T][D]
x [B ] qo+q }
t(
If one now allows {qb}I0, identifies qib as one of the

global displacement degrees of freedom xi, and includes
terms from eqns (45) and (46) in the first variation of H,
one obtains
aH(qo)=aH(xo)
aq
ax,
[([qo
+[N
[B T +
['
32,0J)
})
[C] ({B }+{fP
M T ] JBi}+{oA'})]h,
-([pLe,POJ {Di}h,+ [H, V,MJ{[TI[D]i

+ IAkOj
{[KA] [T] [D]}
50 +
+fl 2 [[Df [MI [D]], hk.
)})
(65)
S
[AL] {[KB] [T] [D]}i 8L.
iJOk +[[KB][T][D]]LJO"
(68)
In eqns (65) and (68) subscript and superscript k denotes

finite element number and subscripts i and j refer to
degree of freedom numbers the range of which includes
all degrees of freedom associated with the kth finite
element.
The governing equations for linear and nonlinear stress
analysis, buckling analysis, and modal vibration analysis
can be obtained by insertion of the r.h.s. of eqns (65) and
(68) into eqns (56) and (60).
Linear stress analysis
From eqn (56) it is seen that the global equations

governing equilibrium are
a2 H(xo)_x_=-__H_
axi
ax,
j= 1,
N
i=1,2,...N_
(69)
For linear systems the initial deformations {xo} or {Jq}

are zero and {Ax} is replaced by {x}, since the A denotes
"change from a previous known solution {x0 }".
D. BUSHNELL
484
The (i,j)th term of the local element stiffness matrix

for the kth finite element is therefore given from eqn (68)
by
Kij=([BiJ [C]{Bj}+NRjR1j)hk+
QAjA5k
QUBLk
(70)
in which the range of i and j covers the nodal degrees of

freedom associated with the kth finite element. (See Fig.
II and eqns (32), (33), (36), and (37), for example.) In eqn
(70) [BiJ denotes the ith row of [B]T , {Bj} the jth column
of [B], QiA the (i,j)th element of the matrix [KA][T][D],
and QijB the (i,j)th element of the matrix [Ka][T][DI.
The ith component of the local force vector corresponding to the kth finite element is given from eqns
(65) and (69) by
Unlike the case for linear stress analysis, described in

connection with eqn (70), the case of bifurcation buckling or modal vibration in the presence of prestress
involves initial deformations. The expression for the
stiffness matrix is therefore somewhat more complicated.
The (i,j)th term of the local element stiffness matrix
[K1]k for the kth finite element is given from eqn (68) by
[B;J+ [poRiOJ])[C]({Bj}+ {IORu)
Kij
Fik = (-N T,MTJ {Bi+
[p,, p_ 0]J{Di}) hk
+ [H, V,MJ {[T][D1}6btk
(71)
+N Tij]
hk +QijA5ok+Qii SLk
(74)
in which the range of i and j covers the nodal degrees of

freedom associated with the kth finite element and the
other terms are defined as before in connection with eqn
(70). Equation (74) should be compared to the simpler
expression (70) for the linear stress analysis.
in which the range of i covers the nodal degrees of
Bifurcation buckling. The (ij)th term of the local
freedom associated with the kth finite element. The

linear stress analysis problem is formulated by assembly
of the terms K ik of the local finite element stiffness
matrices into the global stiffness matrix [K] and assembly of the components F k of the local force vector into a
global force vector {F}, yielding the global linear equation system
element "load-geometric" matrix [K2]k for the kth finite

element arises from the first line of eqn (68):
[K] {x}={F.
(72)
Bifurcation buckling and modal vibration analyses
Bifurcation buckling and modal vibration are governed

by eqn (60). The matrix of coefficients, a 2H(x0 )/axibaxjb,
i,j=1,2.... N can be derived from eqn (68).
The terms in eqn (68) have the following physical
significance: The first line represents the work done by
the prebuckling (or pre-vibration) stress and moment

resultants, shown in Fig. 15, during infinitesimal buckling
or vibration modal rotation, pb These stress and
moment resultants are given by
K2ij= N 0,MoJ 1 0 j 1= NNRRj
where [No,MoJ are given by eqn (73). In cases for which

the temperature rise is regarded as an eigenvalue, that is
one wishes to find buckling temperatures, the term
NTRkRi which in this presentation contributes to the
stiffness matrix [K,]k would instead appear in the "load
geometric" matrix (K2 ]".
The global bifurcation buckling problem is formulated
by assembly of the terms K ,Iii of the local finite element
stiffness matrices into the global stiffness matrix [K,]
and assembly of the terms K2Ui of the local finite element
load-geometric matrix into the global load-geometric
matrix [K2 ], yielding the global bifurcation buckling
problem
[K,]{x"}+A[K 2 ]{x"1=0.
[NoMoJ=([qoj [B]'+L go'2,0J)[C].
(73)
Note that the prebuckling moment resultant Mo does not

enter the buckling equations because the second term in
the vector [RiRj,0J is zero. This follows directly from
the linearity of the assumed kinematic relationship (16b)
between the change in curvature K and the displacement
components u and w. The second line in eqn (68)
represents a contribution to the stiffness matrix of the
structure as deformed by the loads. The amount of
deformation is given by 1o. The third line represents a
contribution to the work done by the thermal stress
resultants during buckling or vibration modal rotation g'.
The fourth line represents the constraint conditions. The
last line represents a contribution to the mass matrix and
the associated modal vibration eigenvalue Q2
Mo
Fig. 15. Prestress resultants at an interior point along the beam
reference surface.
(74a)
(75)
In eqn (75) A is a load factor to be multiplied by

whatever the prestress distribution in [K2] is. This distribution may, of course, vary along the reference surface.
Note that the effect of prebuckling rotation fo, which
usually varies with the loading, has been assembled into
the stiffness matrix [K,] rather than into the loadgeometric matrix [K2 ]. This is not a rigorously correct
procedure. If f8o varies with the buckling load, as is
usually the case, the eigenvalue problem assumes the

general form
[K,]{x}+A [K2 1{x}+A 2[K3I{x}=0.
(76)
However, experience with difficulties associated with the

extraction of eigenvalues of such quadratic systems has
led to implementations in computer programs in which
the loading is divided into two parts, a fixed part and an
"eigenvalue" part. The prebuckling rotations go, associated with the fixed part are retained by inclusion of the
go in the stiffness matrix as in eqn (74). The prebuckling
rotations associated with the part of the loading to be
multiplied by the eigenvalue A are neglected, leading to a

load-geometric matrix generated only from prebuckling
stress resultants, as in eqn (75). Bifurcation buckling

loads for systems in which prebuckling rotations go are
important are calculated through a converging sequence
of eigenvalue problems in which the fixed part of the
load becomes very large compared to the "eigenvalue"
part of the load. An example of this type will be discussed later.
Modal vibrations. The stiffness matrix for the modal
vibration of a pre-stressed structure is given by
[Kvib]=[K]l+[K2]
(77)
with K, and K2 derived as just described. The mass

matrix is derived in a straightforward manner from the
last line of eqn (68). The global modal vibration eigenvalue problem takes the form
[Kvib]{x1+f1
(78)
[Mvib]{x}=O
with [Kvib] given by eqn (77).

SECTION 3
ANALYSIS OF SHELLS OF REVOLUTION
The importance of this class of structures is attested to

by the numerous computer programs that have been
written for analysis of stress, buckling and vibration of
axisymmetric shells.
COMPUTER PROGRAMS
In Fig. 16 the names of computer programs or their

authors are located in a space with coordinates that
measure complexity of geometry versus generality of
phenomenon. Each name indicates the capability of the
corresponding computer program to perform the analysis
indicated by the intersection of these coordinates. In this
coordinate system increasingly general-purpose computer codes lie increasing distances from both axes.
485
Other codes, existing just outside of the region depicted,

apply to structures that are "almost" shells of revolution,
such as shells with cutouts, shells with material properties that vary around the circumference, or panels of
shells of revolution.
The region shown in Fig. 16 is divided by a heavy line
into two fields: Programs lying within the heavy line are
based on numerical methods that are essentially onedimensional, that is, the dependent variables are separable and only one spatial variable need be discretized;
programs lying outside the heavy line are based on
numerical methods in which two or more spatial variables are discretized. It is generally true that analysis
methods and programs lying outside the heavy line
require perhaps an order of magnitude more computer
time for a given case with given nodal point density than
do those lying inside the line. This distinction arises
because the band-widths and ranks of equation systems
in two-dimensional numerical analyses are greater than
those in one-dimensional numerical analyses. Certain of
the areas in Fig. 16 are blank. Those near the origin
correspond in general to cases for which closed-form
solutions exist and for which slightly more general programs are clearly applicable. The blank areas lying near
the outer boundaries of the chart are for the most part
covered by more general programs such as NASTRAN,
SPAR, STAGS, STRUDL, ASKA, MARC, ANSYS and
other general-purpose programs described in Ref. [24].
As of 1980 the most commonly used computer programs for complex shells of revolution are those by
Cohen[l], Kalnins [2], Svalbonas[3], and Bushnell[18,25]. A typical summary of the capabilities of
such programs is listed inTable 1. In general the shell-ofrevolution codes represent implementation of three distinct analyses:
(I) A nonlinear stress analysis for axisymmetric
behavior of axisymmetric shell systems (large
deflections, elastic or elastic-plastic).
(2) A linear stress analysis for axisymmetric and non-
Table 1. State-of-the-art for computer programs for the stress, buckling, and vibration analysis of complex
axisymmetric shells
Type of analysis
Shell geometry
Nonlinear axisymnMultiple-segment
metric stress
shells, each segment
Linear symmetric or
with its own wall connonsymmetric stress
struction. geometry,
Stability with linear
and loading
symmetric or nonsym-Cylinder, cone,
metric prestress or
spherical, ogival,
with nonlinear symtoroidal, ellipsoidal,
metric prestress
ctc.
N ibration with nonCieneral meridional
linear prestress
shape: point-byanalysis
point input
Variable inesh point
Axial and radial disspacing within each
continuities in shell
segment
meridian
Arbitrary choice of
reference surface
General edge
conditions
Branched shells
Prismatic shells and
composite built-up
panels
a From Bushnell
18]
Wall construction
Loading
Monocoque, variable
Axisymmetric or nonor constant thickness
symmetric thermal and/
Skew-stiffened shells
or mechanical line loads
Fiber-wound shells
and moments
Layered orthotropic
Axisymmetric or nonshells
symmetric thermal and/
Corrugated, with or
or mechanical diswithout skin
tributed loads
Layered orthotropic
Proportional loading
with temperature
Non-proportional
dependent material
loading
properties
Any of above wall
types reinforced by
stringers and 'or
uisgs treated as
'smeared out"
Any of abose wall
types further reinforced by rings treated
is discrete
Wall properties variable along meridian
D.
486
BUSHNELL
C,
.'
0 0
04
>4
0z
-1rz
S00
v
0
Ev
ZO
EE
.-
--
-.o-6
01L
V)
4
>0
O0
0t
E0
E
o
-UZ
4,
M2
.2
-I.-
.,;-I
2 I. I
I .I II
E
VFO
O
-
Z
E
i=
symmetric behavior of axisymmetric shell systems submitted to axisymmetric and nonsymmetric loads.
(3) An eigenvalue analysis in which the eigenvalues
represent buckling loads or vibration frequencies of
axisymmetric shell systems submitted to axisymmetric
loads. (Eigenvectors may correspond to axisymmetric or
nonsymmetric modes.)
Some of the codes [1, 18] have an additional branch
corresponding to buckling of nonsymmetrically loaded
shells of revolution. In the BOSOR4 program [18] this
branch is really a combination of the second and third
analyses just listed.
Advantage of axisymmetric geometry: separation of
variables
The great advantage of the computer programs cited
above is their efficiency. This efficiency derives
from the fact that for the three types of analysis
just listed the independent variables can be separated
and an analytically two-dimensional problem thus
reduced to a numerically one-dimensional model. Such a
model leads to compact, narrowly banded stiffness, loadgeometric, and mass matrices, as we have seen from the
beam analysis of the previous section. The reduction of
these matrices for solving equilibrium and eigenvalue
problems is performed speedily on the computer.
For example, the independent variables of the
BOSOR4 analysis[18] are the arc length s measured
along the shell reference surface and the circumferential
coordinate 0. The dependent variables are the displacement components u, v and w of the shell wall reference
surface. For the three types of analyses listed above it is
possible to eliminate the circumferential coordinate 0 by
separation of variables: in the nonlinear stress analysis 0
is not present; in the linear stress analysis the nonsymmetric load system is expressed as a sum or harmonically
varying quantities, the shell response to each harmonic
being calculated separately; and in the eigenvalue analysis the eigenvectors vary harmonically around the circumference. Buckling under non-symmetric loads is
handled by calculation of the nonsymmetric prestress
distribution from the linear theory and establishment of
an eigenvalue problem in which the prestress distribution
along a given meridian, presumably the meridian with the
most destabilizing prestress, is assumed to be axisymmetric. Thus, the 0-dependence, where applicable, is
eliminated by the assumption that displacements u(s,0),
v(s, 0), w(s, 0) are given by u,(s)sin nO, v,(s)cos nO,
w,(s)sin no, or by u,(s)cos nO, v,(s)sin no, w,(s)cos no.
The advantages of being able to eliminate one of the
independent variables cannot be overemphasized. The
number of calculations performed by the computer for a
given nodal point spacing along the arc length s is greatly
reduced, leading to significant reductions in computer
time. Because the numerical analysis is "one-dimensional" a rather elaborate composite shell structure can
be analyzed in a single "pass" through the computer. The
disadvantage is, of course, the restriction to axisymmetric structures.
ENERGY FORMULATION-A SUMMARY
The following analysis of a segmented, ring-stiffened

shell of revolution is similar to that for the beam given in
the previous section. It is based on energy minimization
with constraint conditions. The total energy of the system involves (1) strain energy of the shell segments U,,
(2) strain energy of the discrete rings U,, (3) potential
487
energy of the applied line loads and pressures Up, (4)

kinetic energy of the shell segments T,, and (5) kinetic
energy of the discrete rings T,. In addition the total
energy functional includes constraint conditions U. arising from (1) displacement conditions at the ends of the
composite shell, and (2)compatibility conditions between
adjacent segments of the composite shell.
These components of energy and the constraint conditions are initially integrodifferential forms. They are
then written in terms of the shell reference surface nodal
point displacement components ui, vi and w, and
Lagrange multipliers A,. The integration along the
reference surface meridian is performed numerically.
Now an algebraic form, the energy is minimized with
respect to the discrete dependent variables, ui, vi, w, and
Ai.
In the nonlinear stress analysis the energy expression

has terms linear, quadratic, cubic, and quartic in the
dependent variables. The cubic and quartic terms arise
from the "rotation-squared" terms which appear in the
constraint conditions and in the kinematic expressions
for reference surface strains e,, e2, and en2. Nonlinear
material properties (plasticity) are not included here. For
details on plastic buckling the reader should consult Ref.
[26]. Energy minimization leads to a set of nonlinear
algebraic equations which are solved by the NewtonRaphson method. Stress and moment resultants are calculated in a straightforward manner from the mesh point
displacement components through the constitutive equations (stress-strain law) and kinematic (strain-displacement) relations.
The results from the nonlinear axisymmetric stress
analysis are used in the eigenvalue analyses for buckling
and vibration. The "prebuckling" or "prestress" meridional and circumferential stress resultants N1 o and N20
and the meridional rotation g3oappear as known variable
coefficients in the energy expression which governs
bifurcation buckling and modal vibration. This bifurcation buckling or modal vibration energy expression is a
homogenous quadratic form. The values of a parameter
(load or frequency) which render the quadratic form
stationary with respect to infinitesimal variations of the
dependent variables represent buckling loads or natural
frequencies. These "eigenvalues" are calculated from a
set of linear, homogeneous equations.
Similar linear equations, with a "right-hand-side" vector added, are used for the linear stress analysis of
axisymmetrically and nonsymmetrically loaded shells.
The "right-hand-side" vector represents load terms and
terms due to thermal stress. The variable coefficients
N, 0 , N20 and PO mentioned above are zero, of course,
since there is no nonlinear "prestress" phase in the linear
nonsymmetric equilibrium analysis.
BASIC ASSUMPTIONS
The assumptions upon which the following analysis is

based are:
(I) The wall material is elastic and behaves linearly.
(2) Thin-shell theory holds; i.e. normals to the undeformed surface remain normal and undeformed.
Transverse shear deformation is neglected.
(3) The structure is axisymmetric, and in vibration
analysis and nonlinear stress analysis the loads and
prebuckling or prestress deformations are axisymmetric.
(4) The axisymmetric prebuckling deflections in the
nonlinear theory, while considered finite, are moderate;
488
D. BUSHNELL
i.e. the square of the meridional rotation can be neglected compared with unity.
(5) In the calculation of displacement and stresses in
nonsymmetrically loaded shells, linear theory is used.
This analysis is based on standard small-deflection
analysis.
(6) A typical cross-section dimension of a discrete ring
stiffener is small compared with the radius of the ring.
(7) The cross-sections of the discrete rings remain
undeformed as the structure deforms, and the rotation
about the ring centroid is equal to the rotation of the
shell meridian at the attachment point of the ring.
(8) The discrete ring centroids coincide with their
shear centers.
(9) If meridional stiffeners are present, they are
numerous enough to include in the analysis by an
averaging or 'smearing' of their properties over any
parallel circle of the shell structure.
(10) The shell is thin enough to neglect terms of order
tIR compared to unity, where t is a typical thickness and
In eqns (79) and (80) subscript I refers to the meridional

coordinate direction and subscript 2 refers to the circumferential coordinate direction. As with the beam
analysis, T is the temperature rise above the zero stress
state. The coordinate z along the normal to the reference
surface is measured outward from an arbitrary reference.
not necessarily from the middle surface.
If "normals remain normal" and undeformed (a classical thin shell theory approximation), the strains as
functions of the thickness e,(z), E2(z), and e12(z) can be
expressed in terms of reference surface strains e,, e2,
and e12 and changes in curvature K,, K2, and Kl, thus
E =eI-zK,
NI.fjudz
(11) Prebuckling in-plane shear resultants are neglected in the stability analysis.
(12) The integrated constitutive law is restricted to the
form given in eqn (84). For example, any coupling between normal stress resultants and shearing and twisting
motions is neglected.
C[,
N2
C12
N 12
MI
M2
C12
C22
0
C33
C,5
C
C24
C25
I- O
C14
M12
C36
Shell strain energy
A shell element is shown in Fig. 17. The strain energy

in the shell wall is
,(e,-aT)+
(e
E1
jO2
FF2
IT121
2
E22
O
E2
-a2T
E12
Cl4
C15
C24
0
C25
0
el
e2
C36
e1 2
C44
C4 5
C4 5
C55
0
0
K1
M I
K2
C66]
ci=JfEdz
E12 = v12E22.
N2T
0T
(84)
T
2
0
C,2 =fEl2 dz
C 5 =-fE2z dz
C22=fE2 2 dz
C25 =
C 33 =|
E2,zdz
C(66= GZ2 dz
E2 2 z
C 14=--fEzdz
G dz
C44 =f E, z2 dz
C24=-fEi 2 z
C 36 =f
dz
Gz dz
C 4 5=fEi2 z2 dz
dz
(80)
N 2 T=fJ(E1
MIT =
E 22 =E2 1(-vP2 v 21 )
2K
NI
(85)
M1 T. M2T are given
NIT =-(EIa IT+E 1 2a 2 T) dz
in which
EI 1=E,
1 (l-v1 2 v 21 );
M]2 fT12 z dz.
These stress and moment resultants are shown in Fig. 17.

Substitution of eqns (80) with eqns (83) yields
and the thermal resultants NT, N2T,

by
E-aT
M 2 =--fo 2 z dz
(79)
which is analogous to eqns (13) for the beam.

For an orthotropic wall material
Ml= Joiz dz
C55 =
a2T) +T, 12 E 2} rdz dO ds
N 12 -fri 2 dz
in which the Cj are given by
The various components of the energy (Us, Ur, Up, TL,

T,) and the constraint conditions U, are derived in this
section. The energy is derived for a typical discrete ring.
The total energy is obtained by summation ove- all shell
segments and all discrete rings.
Us=-f |f|
N 2z+fa,2dz
(83)
ENERGY COMPONENTS AND CONSTRAINT CONDITIONS
(82)
E 12e2+-2zK,2.
It is convenient to perform the z integration in eqn (79)

at this point. The following definitions of stress and
moment resultants are required:
R a typical radius of curvature.
N.
E2 =e 2 -zK2 ,
(81)
M2T
alT+E2 2 a 2 T) dz
f (Ela,T + EI2Q2T) dz
= f (E12 a, T + E22a 2 T) dz.
(86)
The thickness coordinate z is measured from the arbitrary reference surface outward. The integrations
through the thickness in eqns (85) and (86) can be performed explicitly for layered shells with material properties constant through the thickness of each layer, or
numerical integration can be performed (e.g. Simpson's
rule) for layered shells with temperature-dependent
material properties. Equations appropriate for laminated
composite wall construction are given later.
Equations (80) and (82) are analogous to eqns (18) and
(15), respectively, for the beam analysis. The Cj and
NT, N2 ', MIT, M2T in eqns (84)-(86) are analogous to
similar quantities in eqns (20) and (21).
Using eqns (82), (83) and (86), one can write the shell
strain energy in eqn (79) in the form
T
UsJf f [IS] {e}+[N j{e}+ C(T)]rdOds (87)

in which
LSJ
[N,,
N2 , N1 2 , MI, M2 , MI 2 J
{e}=e =[e,, e 2 , e 12 ,
N
= [N,,
, 0O M,
, MT,
01
and C(T) is a function of the shell parameters and

temperature rise only. Therefore it can be dropped. Since
[Si ={S}T =[[C]{e}+{N T }] T
=LeJ [C]T + N T
Fig. 17. Shell element with displacements, rotations, forces (from

Bushnell [27]).
quantities R. and R2 are the meridional and normal

circumferential radii of curvature. Equations (90) and
(91) are of the Novoshilov-Sanders type [10, 11]; they
are analogous to eqns (16) and (17) for the curved beam.
The same equations form the theoretical basis for
Cohen's computer program[l].
Discrete ring strain energy
K2 , 2K 2J
KI,
489
Figure 18 shows a ring cross section with displacements u,, v,, w,, p of the centroid and applied loads V,
S, H, M. The ring cross-sectional area A is greatly
exaggerated relative to its centroidal radius r,. The centroid and the shear center are assumed to coincide and
plane normal sections are assumed to remain plane and
normal during deformations. In the absence of warping,
the ring strain energy is given by
(88)
U,=(rc/2)fJ Ac(e,-ajT) dA dO
= [eJ [C]+ [N ]
+j(GJarc)
substitution of eqn (88) for [S] and the dropping of

C(T) in eqn (87) leads to
U-fjj
([eJ [C] {e}+2[N T J{e})r ddOds
(89)
(p+tc/rc)2dO
in which A is the ring cross-section area and GJ is the

torsional rigidity.
The ring hoop stress is given by
o,=E,(e,-a,T).
which is analogous to eqn (20) for the beam.

Strain-displacement and curvature-displacement relations valid for moderately large relations are
Er=e,-xK,+yKy.
a/r+r(v/r)'+f0
O'
Olkr+r'l/r
2(-3/,r+r'lr+0'/R2 )
(93)
The hoop strain e, can be expressed as a function of

strain of the centroidal axis plus terms due to in-plane
and out-of-plane curvature changes K. and K,, respectively:
u'+w/R,+2(O2+y2)
v/r+ur'/r+w/R 2 + (2
(92)
(94)
(90)
Substitution of eqns (94) and (93) into eqn (92), integration over the ring area, and the dropping of the term
which contains only ring parameters and temperature,
leads to the following expression for the ring energy:
U, =(r0/2)Jf (e, 2 E,A+ KXEIy + KyErIx -2KyKyE,1x
in which
+(GJlr,2)(I3+tjr) 2+2[e,N, +KMy +KYMT])
f3=w'-u/R,,
O=vw/r-v/R 2
(95)
(91)
1Y= I61r-v'-r'v/r).
Dots indicate differentiation with respect to the circumferential coordinate 0; primes indicate differentiation
with respect to the meridional coordinate s. Positive
values of u, v, w, p, 0 and y are shown in Fig. 17. The
do
in which
N7T=-fEraT dA
(96)
MyT =+f E,a,Tx dA MT = -J Era,Ty dA.
D. BUSHNELL
490
SHELL REFERENCE SURFACE
X,
f
e2
U /
I
yj
I3,M
t
04
RING
,/I
x ,c ,H
RING
CENTROIDAL
AXIS
<
,1/1
-,>2IF
ATTACHMENT
POINT
Fig. 18. Discrete ring with centroidal displacements, forces (from Bushnell [27]).
The ring kinematic relations are

e,=/r,
Kinetic energy of discrete ring

The kinetic energy of a discrete ring stiffener is given
by
+wdlr,+2(+
Tr=p,(r,/2) f [A (UC
K, = dIr,
,2+vV,2+ W,,t2)+1,p
2+1 +
O2+
(101)
+2I,,nc,I,y,1*do
Ky=-
+,/r
C.=(Vwf -0
(97)
Irc
7, = ucl r'.
Equations (92)-(97) are analogous to eqns (79)-(91) for

the shell.
Potential energy of mechanical loads
Two types of loads are permitted in the analysis: line
loads and moments V, S, H, and M, which act at ring
centroids and at shell segment boundaries, and surface
tractions Pl, P2 and pressure p3 . These loads are shown
in Figs. 17 and 18. The potential energy associated with
constant-directional line loads at a given ring station is
in which I, L, 4, Isn are cross-sectional area moments

of inertia relative to axes normal and tangential to the
shell meridian at the ring attachment point. In the case of
harmonic oscillations, the differentiations with respect to
time are replaced by a factor Ql, which is a frequency
parameter. Equations (100) and (101) are analogous to
eqn (28) for the beam.
Constraint conditions
Figure 19 shows a meridional discontinuity (+, -)
between two adjacent shell segment reference surfaces
and discontinuities at shell edge support points "A" and
"B". The compatibility conditions for the junction are
u*+ u* +Au*
(-Vu, +Sv, +Hw,+Mg)r, do.
(98)
The potential energy associated with constant directional

surface tractions p, and P2 and pressure p, is
v*+=v* +Av*
(102)
W*+= W*-+Aw*
U,, = -|
B+=8i
in which
Au*=-(d,1+d 2 p2 /2)
Av*=-(di+Aw*)(si*--v*-)/r--(d 2+Au*)*/*
rU,2=-f f (PIU+P2V+P3w)r dO ds.
(99)
These equations are analogous to eqn (22) for the beam.

Additional terms required to account for following or
"live" loads are-introduced later.
Kinetic energy of shell segment
The kinetic energy of the shell segment is given by
T4=If
f m
(ut2+v,2+wr2)rdO
ds.
Inc2
(100)
in which ( ),, denotes differentiation with respect to time.

The shell rotatory inertia in neglected.
(103)
Aw*=d20-d,d2/2.
The constraint conditions (102) arise from the requirement that the motion of the point (+) relative to the point
(-) involves no deformation of a line joining the meridional gap (+, - ) and compatibility of meridional rotation
1 across this ( +, - ) gap is enforced.
At a support point the terms u *, v*+, and w*' in eqns
(102) are constrained to be zero if the appropriate boundary condition integers KAI, KA2, etc. and KBI, K8 2 , etc
(see below), are equal to unity. The constraint conditions
(102) are incorporated into the total system energy by the
introduction of four Lagrange multipliers A,, A2 , A3, and
A4 for each edge support and each segment junction.
Thus, the "energy of constraint" corresponding to each
491
dependence can be eliminated from the analysis by the

assumption that
junction has the form

u*+-u*- Au*
Ur =tA , A2 ,A3 ,A4]
*+-W*--Aw*
(104)
nia,
nma,
u(s, 0) = uo(s) + I
un,(s) sin nO + I
-= min
At the shell ends the constraint conditions have the

following forms: at end point "A"
-U*_ -Au*
UC= [KA,*A,,KA2*A2 ,KA3*A3,KA4*A4]
-V **
-Av*
-AW*
-,B
(105a)
at end point "B"

Uu*- -Au*
UC
-Av*
= 1Kjg,*Ai,Kl32*X2,KB33*X3,KI34
Q~
-vW**
.-dw*
These equations are analogous to eqns (26) for the beam.

Variable transformations
The components of energy of the system are
represented by the shell strain energy U. (eqn 89), the
strain energy of a discrete ring Ur (eqn 95), the potential
energy of line loads Up, and surface tractions Up2 (eqns
98 and 99), the shell kinetic energy T. (eqn 100), and the
discrete ring kinetic energy T, (eqn 101). The constraint
conditions U. are given by eqns (104) and (105).
It is desired to express all energy components in terms
of the shell reference surface displacements u, v, and w.
The displacements uc, v,, and w, of the ring centroid
(Fig. 18), which appear in eqns (95), (97), 98), and (101),
are given by
uC=u*+Au*
v,.=v*+Av*
w =w*+Aw*.
v*=v w*=ur'+wrIR2
w(s, 0)= wo(s)+2 wv,,(s) sin no+> wn 2(s) cos nO.

The temperature distribution, surface tractions and
pressures, and thermal and mechanical line loads have
similar expansions, which are given explicitly later.
If the expressions (108) were inserted into the energy
components just derived, all the harmonics would couple
in the analysis, since the kinematic relations (90), (97)
and (103) are nonlinear.
In the analysis, large deflections are permitted in the
components,
but
the
nonsymmetric
harmonics are considered to be small. The various

harmonics do not couple, and a solution for each un(s),
vM(s), and wh(s) can be obtained with the circumferential
wave number n appearing as a parameter in the analysis.
The 0 integration indicated in eqns (89), (95) and (98)(101) is replaced by a factor of wrfor n#0 and 27r for
n=0. In a linear stress analysis for nonsymmetrically
loaded shells, the static response of a shell to arbitrary
varying loads is obtained by superposition. (In this case,
even the axisymmetric components are assumed to be
small.) In buckling and vibration analyses the "small"
deflections
unv,,V,
WJ,, Un 2 ,
vn 2 ,
W. 2
are considered to
be kinematically admissible variations from the "prebuckled" or "prestressed" axisymmetric state represented
by the large deflections uo(s) and w0(s) in eqns (108). The
uo(s) and wo(s) are determined in the nonlinear stress
phase of the analysis by Newton-Raphson iterations, as
described in the discussion associated with eqn (56).
I'
S
Separation of variables
The dependent variables u, v, and w are functions of
arc length s and circumferential coordinate 0. The 0
Pt. 2.
u**, V**, w*
End of Shelle:
Shee
Ref.
Sur face
Begirnmng
~d,
SECTION AA
Support
A4.
The total energy in the system is obtained by summing

over all shell segments, discrete ring stiffeners, junctions
and boundaries.
(a)
(107)
which is analogous to eqn (42) for the beam.

Equations (106) and (107) can be used to eliminate us,
v., wr and O*, v*, and w* from the energy components
and constraint conditions. The dependent variables are
then u, v, w and the Lagrange multipliers AI, A2, A3, and
CAS Vol. 18, No. 3--H
(108)
(106)
The quantities Au*, Av*, and Aw* are given by eqns

(103) with di and d2 replaced by el and e2, the ring
eccentricity components (Fig. 18), and u*-, v*-, w*-, rreplaced by the displacement components and radius uO,
v*, w*, r which correspond to the ring attachment joint.
The axial, circumferential, and radial displacement
components uf, v*, and wO, which appear in eqns (102)(106), are given by
u*=urIR 2 -wr'
v(sO0)=n
vn(s) cos nO+> Vn2(S)sin nIJ0
axisymmetric
-13-
u, 2 (s) cos nO
lmin
of
Support
Pt. A:
Sohel1:7
d,'
( b)
Fig. 19. Geometry for constraint conditions: (a) shell reference

surface discontinuity; (b) support points (from Bushnell[27]).
D. BUSHNELL
492
In the linear analysis for nonsymmetric behavior and

in the buckling and vibration analyses, the second summations in eqns (108) can be represented by zero or
negative values of the circumferential wave number, n.
CIonped Edge
DISCRETIZATION
The 0 dependence has been replaced with a parameter

n, so that only one independent variable remains-the
arc length s. Figure 20(a) shows a shell meridian of two
segments, and Fig. 20(b) shows a discretized model. The
continuous variables u(s), v(s), and w(s) are replaced by
discrete variables ui, vi, and wi. The ui and vi occur at
stations midway between the wi in a manner analogous
to the discretized beam model shown in Fig. 11. With
constant nodal point spacing within each shell segment,
as shown in Fig. 20(b), the energy is evaluated at the
mesh points where the w, are located. The displacements
and the s derivatives required in the energy are
- _Td,
V-1 V,-Iw'-, 0',w.
'Seg-
u=(ui+ui -)12,
v=(vj+vi_,)12
t 'I
F-., Edge With Ring
u'=(ui-ui ,)Ih, v'=(vj-vv)h

(109)
w=wi,
Ring Attmw1 Poxt
e, (N.goiN.)
Fig. 20(a). Two-segment shell meridian with discrete rings, discontinuity, and various quantities identified (from Bushnell[271).
w'=(w+,1-wi-,)/2h,
w`=(wj,+-2wj+wi_,)/h 2
labeled E (center of the length 1) involves the seven

points wi-, through w+,,. The energy per unit cirin which h is the mesh point spacing. These expressions cumferential length is simply the energy per unit area
are analogous to eqns (32) for the beam.
multiplied by the length I of the finite-difference element,
If the nodal point spacing varies within a shell seg- which is the arc length of the reference surface between
ment, the expressions for u, v, u', and v' remain as given two adjacent u or v points. This formulation yields a
ineqns (109), but the three-point formulas for w,w', and w" 7x7) local element stiffness matrix corresponding to a
become
constant-strain, constant-curvature-change finite element
that is incompatible in normal displacement and rotation
w
all a12 a13 ]w,
,1
at its boundaries but that in general gives very rapidly
W
a
a 2 2 a23
Wj
(110)
convergent results with increasing density of nodal
I IL a3, a32 a33 W,+J
points. Note that two of the w-points lie outside of the
element. If the mesh spacing is constant, the algebraic
in which
equations obtained by minimization of the energy with
respect to nodal degrees-of-freedom can be shown to be
a, =(h-k)(3k+h)/[16(h 2+hk)]
equivalent to the Euler equations of the variational
problem in finite form[19]. Further description and
proofs are given later and in Ref. [19].
al 2=(h+3k)(3h+k)/(16hk)
Figures 2 and 22 show rates of convergence with
increasing nodal point density for a poorly conditioned
a, 3 =(k-h)(3h+k)/[16(k2+-hk)]
problem-a stress analysis of a thin, non-symmetrically
a 21 =-1/2h a22 =(112h-112k) a23 =112k
(111) loaded hemisphere with a free edge. The u and w displacement components at 0=0 are plotted in Fig. 1. The
finite-element results indicated in Fig. 2 were obtained by
a31 = 2/[h(h + k)]
programming various kinds of finite elements into the
BOSOR4 program[18]. The computer times for coma32= -21(hk)
putation of the stiffness matrices K, are shown in Fig.
22. A much smaller time for computation of the finitea33=21[k(h+k)].
difference K, is required than for the finite element K,
because there are fewer calculations for each "GausThe quantities h and k are defined in Fig. 21.
sian" integration point and because there is only one
"Gaussian" point per finite-difference element. Other
Finite-difference energy method vs finite element method
The discretization technique just described has been comparisons of rate of convergence with the two
called the "finite difference energy method". This methods used in BOSOR4 are shown for buckling and
method is described in detail and compared to the finite vibration problems in Ref. [19].
element method in Ref. [19]. Figure 21 shows a typical
shell segment meridian with variable nodal point spacing. Energy functional converted to algebraicform
As in the case of constant nodal point spacing shown in
With the substitution of eqns (108) in the various
Fig. 20, the 'u' and 'v' points are located halfway be- energy components and constraint conditions, the
tween adjacent 'w' points. The energy contains up to first replacement of s derivatives by eqns (109) or (110), the
derivatives in u and v and up to second derivatives in w. replacement of time derivatives by a frequency
Hence, the shell energy density evaluated at the point parameter ifQ, and the numerical integration over s and
021
""
493
13
10
Wj
Fig.
20(b).
Shell meridian
with
exact integration over 0, the system energy and constraint conditions are now represented by an algebraic
form which contains as dependent variables u*, vi, and wi
and the Lagrange multipliers Al, A2, A3, and A4 (for each
junction and boundary). The algebraic form also contains
as parameters the shell and ring properties, the loads and
temperature, and the frequency parameter fQ.
discretized model (from Bushnell[27]).
tion with double precision on the UNIVAC 1108 computer. The number of iterations required depends on how
nonlinear the problem is. Generally, less than about five
iterations are needed for convergence at a given load
level.
Fictitious
.
ShWlI Enegy -E, l
STRESS, BUCKLING, AND VIBRATION ANALYSIS-A SUMMARY
[C]
Nonlinear stress analysis
In the nonlinear stress analysis only the axisymmetric

components of the load are considered and only uo(s)
and wo(s) in eqns (108) are nonzero. Terms linear
through quartic appear in the algebraic form for the total
energy, as with eqn (44) in the beam analysis. The
simultaneous nonlinear algebraic equations obtained by
energy minimization with respect to the nodal point
displacement components uo, and woi and Lagrange
multipliers Ai are solved by the Newton-Raphson
method, as described in the discussion associated with
eqn (56). The coefficient matrix for each iteration is
symmetric and is strongly banded about the main
diagonal. Such narrowly banded systems can be solved
in a matter of seconds of computer time. For example, a
BOSOR4 case with about 200 degrees of freedom
requires somewhat less than about 2 seconds per itera-
[B
constitltie
--
Point
[][]B
L-w
Kinatic Law
Shell
Segment
Finite
Difference /'
Element,,
.
-Wi
Fig. 21. Nodal displacement degrees of freedom for variable
nodal point spacing.
494
D. BUSHNELL
7
-
w, Normal
Edge Displacement
p- cos 26
Edges Free
Linear equilibrium for nonaxisymmetric loading
Cr
4o
za
Finite Element Analysis

With Cubic u, v, w
-X---
Finite Difference (BOSOR4)
a)
z
2
Conmputer Tme
to. Compute
K,
U 0
20
40
60
80
NUMBER OF MESH POINTS
The linear stress analysis is based on the same equations as the stability and vibration analysis, except that
the "prestress" terms which appear in the stability and
vibration quadratic form are not present, and the gradient of the energy functional is not homogeneous, since
a "right-hand-side" vector is nonzero. This vector arises
from the thermal terms in eqns (89) and (95) and the load
terms in eqns (98) and (99).
Corresponding to the nonsymmetric portions of u(s, 0),
v(s,O), and w(s,O) given in eqns (108), the temperature
rise distribution T, surface tractions P., P2, and pressure
pi, mechanical line loads V, S. H, M and thermal line
loads N , , Mv , MiT have the following expansions:
0
0
based is a homogeneous quadratic form. The form is

stationary with respect to the dependent variables u,, v.,
w, and the Lagrange multipliers A for certain discrete
values of a parameter-the so-called eigenvalues. The
eigenvalue parameter can be a load or load ratio, a
temperature, or a frequency.
100
T(s,O)=X TLi(s) sin n0+X T52(s) cos no
Fig. 22. Computer times to form stiffness matrix KI and rates of
convergence of normal edge displacement for free hemisphere

with nonuniform pressure p(sO) = piios 20 (from Bushnell[18]).
p,(s,o)=Z pt,,(s) sin n0+2,p, 2 (s) cos nO

n
P2 (s,0)=
A reasonable convergence criterion for the nonlinear

prebuckling solution at each load step is that successive
values of all u0e, wio greater than a tenth of the largest
displacement be different by less than 0.1% of their
absolute values. The starting vector for the first iteration
at the first load value is zero, which means that the first
solution represents the linear theory solution. The starting vectors for the first iterations at subsequent load
values are the converged solutions obtained at the loads
immediately preceding the current load. Once the displacements uio and wio have been calculated, the
reference surface strains and stress resultants are
obtained in a straightforward manner by means of eqns
(84) and (90).
P2nI(s)cos nO+2p2n2(s)
P3 (s, 0)=X p1N.(s)
sin no
sin n6+Z P3.2(s) cos nO
V(O)=E V., sin nO+,
V.2 cos
no
S(O)=X
S., Cos nO+X
S. 2
sin nO
(112)
H(o)=j H., sin no+j H. 2 cos no

n
M(0)=Z M., sin nO+X Mn2 cos no

n
Bifurcation buckling and modal vibration
In the buckling and vibration analyses the symmetric

and nonsymmetric displacement components contained
in the summations indicated in eqn (108) are considered
to be infinitesimal, kinematically admissible variations of
the displacements from the "prebuckled" or "prestressed" state obtained in the nonlinear stress analysis described above. Since the "buckling" displacements un, v.,
and w. are infinitesimal, one need only retain linear
terms in un, v., and w. in the kinematic relations and
constraint conditions. However, it must be remembered
that the displacements and relations in eqns (90)-(106)
represent the total displacements from the undeformed
state. Hence, cross-product terms such as wows obtained
by insertion of eqns (108) (with the use of eqns 91 and
109) into eqns (90), for example, must be retained. The
energy minimization in the buckling and vibration
analyses is performed with respect to un, v., w. and the
Lagrange multipliers.
The buckling and vibration analysis is described in
detail below. Here it is sufficient to point out that the
energy expression on which the numerical analysis is
NT (O)=X
=
NL sin nO+X NT52 cos nO
M
My (
n()yM
no+
M.,
Y)=sin n
MTy+
5n2 cos nOJ
Mu(O)=2MxLi sin nO+Y,M 2 cos no

n
in which Nr, MYT, and MiT are given by eqns (96). As

with eqns (108), the first summations on the left-handside correspond to positive or zero n and the second
summations to negative or zero n.
ENERGY IN TERMS OF NODAL DEGREES OF FREEDOM
Strain energy of shell segment
For constant nodal point spacing the shell reference

surface displacement components u, v, w can be written
3X
{U
and for n =0
in terms of the nodal displacements in the form
U(S,) =U(S); V(S,) n
DJ{q}
v=
495
(113)
v(s); W(sO) =w(s).
(119)
The rotation-displacement relations (91) can be written in

the form
in which
3+7
-[0
[DI
He
1/2 0 0 1/2 0 0
0 1/2 0 0 1/2
Q]
0
0
00
=-1/2h
3x7
[R]=
0
nl4r
(114)
tqI= {q} = tw,-,, ui-l, si,, wi, ub v, w,, 1J. (115)
The reference surface strain, change in curvature, and

twist are given, from eqns (90) and (109) by
+ 2)
2(2+ 7,2)
21
6x7
{E}l
= [B1JfqJ +
El,
go
0
0
0
KI
K2
2K12
RIq
- 1/2R,
0
nl4r
0
1/2h]
- 112R 2
0
(-112h - r'/4r) 0 -
(121)
Equations (113)-(117) are analogous to eqns (34)-(38) for

the curved beam. Equations (120) and (121) are analogous to eqns (63) and (40), respectively. Insertion of
eqn (116) into eqn (89) yields an expression for the strain
energy of the shell which is analogous to eqn (44) for the
beam. The only differences are the dimensions of the
vectors and matrices and the fact that the shell energy
must be integrated over 6 as well as over s.
The nonlinear axisymmetric "prebuckling" (or in a
modal vibration calculation, "prestress") analysis is carried out after specializing eqns (116) and (117) to
axisymmetric displacements, that is, after setting e12, K,
qy,and n equal to zero.
and
02
(120)
{O} =3[R7{q}
where
0
0
- 112R 2
nir
(112h - r'/4r) 0
-112RI
14=
Other components of the energy
A similar procedure is followed for the strain energy

of the discrete rings. (Remember the smeared rings and
stringers are included by appropriate modification of the
in which
0
0
FI
- r'/2rh
n
- h
0
- n2r
I/h
r'12r
n/2r
f(RR,)'l
1
hR2 J
- r'I2rR,
n
rR,
Vh
1/Rl
1IR 2
2r)
hr
- nZlr2
n!2rR2
(_
r' _2
rR2 hR_2)
1/h
r'/2r
(117)
0
I
Equation (117) is based on the assumption that for n > 0

u(s,6)=u,(s)sin nO, v(s,0)=-v(s)cosnO;
= w,(s) sin no
(I 18)
hR,
2 ]
- r'/2rR1
nr'
rR1
0
nI2rR2
+ 2
rR2 hR)
0
r'02rh
n
rh
6 x 6 integrated constitutive law [Cl!) The potential

energy of the applied loads, the kinetic energies of the
shell and of the discrete rings, and the junction and
boundary conditions are handled in an analogous way.
D.
496
BUSHNELL
BIFURCATION BUCKLING ANALYSIS
Two sets of loads

The bifurcation buckling problem represents perhaps
the most difficult of the three types of analyses performed by shell-of-revolution computer programs.
Therefore, details of the formulation are given here.
It is practical to consider bifurcation buckling of
complex, ring-stiffened shell structures under various
systems of loads, some of which are considered to be
known and constant, or "fixed" during a computer run
and others of which are considered to be unknown
eigenvalue parameters, or "variable".
The notion of "fixed" and "variable" systems of loads
not only permits the analysis of structures submitted to
nonproportionally varying loads, but also helps in the
formulation of a sequence of simple or "classical"
eigenvalue problems of the form of eqn (75) for the
solution of problems governed by "nonclassical" eigenvalue problems of the form of eqn (76). An example is a
shallow spherical cap under external pressure, such as
shown in Fig. 23. Very shallow caps fail by nonlinear
collapse, or snap-through buckling, not by bifurcation
buckling. Deep spherical caps fail by bifurcation buckling in which nonlinearities and edge effects in the prebuckling phase are not particularly important. There is a
range of cap geometries for which bifurcation buckling is
the mode of failure and for which the critical pressures
are somewhat sensitive to predictable nonlinearities and
nonuniform deformations in prebuckling behavior. The
analysis of this intermediate class of spherical caps is
simplified by the concept of "fixed" and "variable"
pressure.
Figure 23 shows the load-deflection curve of a shallow
cap in this intermediate range. Nonlinear axisymmetric
collapse (p,,), linear bifurcation (pb), and nonlinear
bifurcation (Pb) loads are shown. The purpose of the
analysis in this section is to determine the pressure Pnb
It is useful to consider the pressure Pnb as composed of
two parts
(122)
Pnb =p+AkAp
in which pf denotes a known or "fixed" quantity, A is an

eigenvalue, and Ap is a known load increment. The fixed
Pnl
-
Prebuckling
portion pf is an initial guess or represents the results of a

previous iteration. It is clear from Fig. 23 that if pf is
fairly close to pnb the behavior in the range p=pfAX~p
is reasonably linear. Thus, the bifurcation point p,,, can
be calculated by means of a sequence of eigenvalue
problems of the form
[K,(p(.)')] {x}+A [K2(Ap(-))] {x}=0
(123)
through which, for increasing iteration index (m), ever

and ever smaller values of XAp(,) are determined and
added to the known results p(-), from the previous
iterations. The initial guess p(o)f need not be close to the
solution pb.
The matrix K, is the stiffness matrix including the
effects of the "fixed" loads p(,,)f, The matrix K2 is the
"load-geometric" matrix and is proportional to the stress
resultant increments due to the known load increment
Ap. The derivation of these matrices follows.
As in the case of the beam analysis, it is known that an
axisymmetric equilibrium state {xJ 1}={(uJ, v(f, w0 f)}
exists corresponding to the "fixed" load p=pf. The
object of the bifurcation buckling analysis is to determine whether {xof + x5}, where {Jx} is an infinitesimal,
nontrivial, kinematically admissible buckling mode, also
represents an equilibrium state at p = pf.
First a prebuckling solution {x(f + Axo} is obtained at a
neighboring load p = pf + Qp. (Note that {lxA} is a finite
quantity, not infinitesimal as is the buckling mode {x"}.)
The total displacement {x} = {u}, {D}, {W} is given by
{U}-{u- +Au +ub}; {v}-{vo+AVQD,+tv}
(124)
{w} ={wof+Aw,+wb}.
Shell strain energy

When the r.h.s. of eqns (124) are inserted into eqns
(90), the total reference surface strain and curvaturechange vector can be expressed in the form
{e}J={e
}+{e}+{e }
(125)
in which {e')} represents the contribution of the displacements {xo0 +Ax0 }; {e)} represents the contribution
of the terms that are linear in the infinitesimal modal
displacements {xb}; and {e' 2 } represents the contribution
of the terms that are quadratic in {x"}. The strain energy
U. in eqn (89) can be arranged such that terms of similar
power in {x") are collected
U. = U''+U(')+U' 2'+...
(126)
Pnb
- Postbuckling
P_
The zero-th order terms in eqn (126) can be dropped

because they are independent of the dependent variables
{xb}. The first order terms, when combined with the
other components of the total energy functional H.
cancel because {xJ-+Axo} is an equilibrium state. Terms
of higher order than second in {xb} are vanishingly small
compared to U(2). Hence, the expression for U(2)
governs bifurcation buckling. From eqn (89) one can
write
VOLUME DISPLACED
Fig. 23. Load-deflection curves for shallow spherical cap, showing bifurcation points from linear prebuckling curve (pmh) and
nonlinear prebuckling curve (p,,) (from Bushnell[18]).
+2(oe)] [C]+ [N T
I){e(2)})r
dO ds.
(127)
Analogous expressions can be written for the discrete
ring strain energy (eqn 95) and the junction and boundary
conditions, eqns (104) and (105).
If the prebuckling state {xof+Axo} is axisymmetric and
torsionless, the first order strain and curvature-change
vector {e<')} calculated from eqns (90) is given by
0
'+W IR/+(Pof+Afo)ib
v Ir+uu r/r+W /R2
bIr+r(br)'+(P3J +AIo)0b
{e)
and {qb} is given by eqn (115) with superscript "b"

added. The rotation vector [owj is expressed in terms of
the nodal point degrees of freedom in eqn (120) and
(121).
Using eqns (120), (131) and (134); integrating over 0
(multiply by 2ir if n = O.7 if n 0); and integrating over
s (multiply by h), one can transform the shell strain
energy expression (127) into the form
(128)
(
hr LqbJ[[B + P.B 2 ] [C][B, + l fB]
2(2)
1or)
|J/r+ r' pb3r

2(
Jr + r'q//r + v 5 '/R 2)
497
+7X3
33
3x7
7xl
+[RI [NJI[RI {q`J
(136)
The second order terms ic"I are given by

3x3
7x3
I(gb)2+1(yb)'2
+ [qb [-]
3.7
[ANo][R]{qb})
b 2
-2(
b52
(129)
+(-2Y
0
0
From eqn (84), the quantity (Le(J [C]+[NT J) ir i eqn
in which rk is the radius r (Fig. 20a) evaluated at the

midlength of the kth finite element. For the kth finite
element the local stiffness mass matrix [Kl("p(,))IThell of
the shell as loaded by the "fixed" loads p(.) is given by
hrk times the first set of terms in eqn (136) that are
premultiplied by [qbJ and postmultiplied by {qb}. The
local load-geometric matrix [K 2 (AP(m))] heu is given by hrk
times the second set, that is
(127) can be written as LSo], where

T
[K 2 (Ap(,,))] skheI= hr5 [RI [AN.] [R].
LS-] = [S0fJ +LASOJ

= LNofN
(130)
,0, MfM 2 ', 0]
+[AN,,,AN2 o0,, AM,., AM2o, 0].

It can easily be verified that the second product in eqn
(127) can be transformed as follows
2LSoJ {e
2
1}=
[LsbJ [N1+ANo
0]{1W}
(131)
(137)
The KI and K2" just described are analogous to the

local stiffness and load-geometric matrices for the curved
beam, given in eqns (74) and (74a), respectively. These
local matrices are assembled into the global matrices of
eqn (123) in the same way as described in connection
with the discussion associated with Figs. 12 and 13 for
the curved beam. Figure 24 shows the format of the
global stiffness matrix [K,] corresponding to the twosegment discretized model in Fig. 20(b).
in which
[bJ
Strain energy of a discrete ring stiffener

The strain energy Ur of a discrete ring, given by eqn
YbJ
=L
[gbib
N[of
0
0
N2 0 '
[N.' + AN(,] =
-
The Ago should be neglected in eqn (128) because these

terms contribute to the stiffness matrix [KI(p(,)], which
is to be evaluated at p(.) , not at p()f + Ap(,). The
vector {e(')} can be written in terms of the nodal point
degrees of freedom [qbJ:
(
[[1,] +-
{}=[[B.]
.B " 61x ]]qb
(95), can be expanded in the same manner as that of the

shell,
U,
10
[B2 ] =I
10
0
-
112R,
0
0
0
0
-1/2R2
0
0
0
0
0
0
+ UWI)+ U,(2)+
(138)
...
(Le,''1 J [G]{eI}
+ 2(Le,')j [G] + [NITJ){e,(2
in which [B,] is given by eqn (117), [B2] is given by

- 112h
Ur
Again, it is U,' ' that governs bifurcation buckling
(134)
o[B21{b
0
0
nlr
0
0
0
0
1I2h
-1/2R,
0
01
0
0
- l/2R2 0
01
0
0
01
0
0
01
0
0
J)r,
dO
(135)
(139)
D.
498
BUSHNELL
in which, from eqns (95) and (97)

e,
{e,
moment [SoJ in the ring. [SOJ can be written in the

form
j=
lbIr,
+ w,
-'j'/'r, + I
K}
LSoj = [Sro] + [ASr.]

= INr0 ,O,MrO,O] + LANro,O,AMo,O].
(140)
01lr, + u6b
|Kxy
2II
The second product in eqn (139) can be expressed in the

form
+ 22
2 [Sr.] {r(21}= [.b]

JEr(2)1 =
0
E,ly
[G],=
LwcJ 'Pcyc'
0
0
-E,lj
J'
LN T] = LNTMy T M.Tj
The quantities N,0J and AN,o represent the prebuckling

ring hoop force due to the "fixed" load p(. f and change
in load Ap(,), respectively. The prebuckling ring moment
does not appear in eqn (147) because the ring curvaturechange expressions are linear.
It is necessary to express the discrete ring strain
energy U,' in terms of the shell reference surface nodal
point degrees of freedom [qbj. First eqn (140) can be
(143)
with NT, MTMiT given by eqns (96). As with the shell

strain energy, the term ([e, 0 )] [G]+ [N T ]) in eqn (139)
represents an axisymmetric prebuckling force and
hv
Vj V,
kU*V
x x x
x x x
\'13
U.
V.
1XX X
U,
V.
X X XXX
X XX xxx
x xx
x x x
- XA
. 1
AA
--- I
I (Orl,
XX
x
u,
v,
U,
v:
[KOT]T rxiXSTA. 2
x x XI
I
x
i Z
10)
W,
xM .
STA. 3
'-A
lx x XxxlX [ofT
-'.XxXSTA.
xFxl xA
IXXKI
v,
- STA,
l
U'
v,
W,
3-STA.
X x Xlxxxxxxx
V,
W,
x X xlx)
U,
-Ix x~x
l
x
Xx X X
xxx
v,
XXX x
V,
W:
lXXXXXxXXX X
U.
x X X X X x X
v,
lX X X X X X Xl
X X X X X X X-
----
--
[r2
--V"
.[T]
lo]
A,.
j.
I
X
Wu:
V,
[T)T
U,,
X XK X X X
XX
x x
STA.
K
x
x xxK
x x
KKXSTA.
XXXX
XXX
W"
Uw,,
XXXX
XXXKXXX
W,,
x~xl
U,,
V,,
~~
x~~~
xXxxL
XX
X X|X
X|XKI
IXX
X
x~~~
U.
X XX
XK X
K&
X
xXl
xx
xlXxx
X X
AXXXST.1
12
XSA.t
X X
XXXX
XA
X Xl X X
w.,
x X XlX. X Xl X( X
xXXXX
XXX
XXXxX X
XI X
U,,
V.'
I"M.
xl
JxQ
[KOT]3
Ito)
'Al.
- -
(147)
GJ]
and
w.
(146)
[ N, f]
[0AN
01
[Nrc! + aN,0 ] =I[0 N,o']+[L
'AN,0
(142)
01
E0l
0
(145)
in which
- Er0.y
+ ARNo]{wb}
[&I
(141)
O
0
F EA
(144)
Fig. 24. Form of global stiffness matrix [K,] including constraint conditions. This matrix corresponds to the
discretized model in Fig. 20(b) (from Bushnell [30]).
499
in which
written in the form
[BJ]
I()
rb
IC()=4X4
(148)
Vb
[T]=
r/2R 2
0
1/2
0
0
r'/2
- 1/2h - 1/2R, 0
0
-r' r/2R2 0
0
1/2
0
0
r/R2 r'/2
0 - 1/2R, 0
01
0j
1/2h .
(156)
where
- n/r
0
4x4
[BJ] =
n/ r
0
2 2
l1/re
2
n /r.
Using eqns (148), (150), (152) and (155) and integrating

over 0, one can express the ring strain energy expression
(139) in the form
0
2
n /r.
l1r,
- n/r.2
nlr_
(149)
(rro 27 r.x
7x4 [l4x4
(j or 2 r) rC ( LbB [T]T[ 1
4x2
2x2
+[B. ([Nf9
Equations (97d) and (97e) for the rotation compone:nts Id

and -y. were used in the derivation of eqn (149).
Similarly,
}b|
UcI
{(Jb
b}{I:}.
[B.] =
4]
(150)
in which
r-n/r.nr
E2T]{qbl).
[
+ [ANRo])[BJJ [El+
Constraintconditions
The junction conditions (104) and boundary conditions
(105a,b) contribute only to the stiffness matrix
[K,(p(.),)] of the structure as loaded by the "fixed"
loads p(,)f.
The mth constraint condition U"' can be written in
the form
= [Al',A2",A 3 ', A4 J
fu*+]
I I];
4x7
All-of the terms in eqn (157) except those involving

[AN,0] contribute to the stiffness matrix [Kl(p(.))). The
terms involving [AN~o] contribute to the load-geometric
matrix [KAAp(,))]. The contributions of each discrete
ring to the total strain energy of the axisymmetric structure are added to the local stiffness and load-geometric
matrices of the shell finite element which contains the
attachment point of that ring.
(152)
US
04
where 00f is the prebuckling meridional rotation associated with the "fixed" loads. Consistent with eqns (106)
and (103) and Fig. 18, it can be shown that
4E4
4x4
(157)
U
[El +
2x4 -
(151)
Next, it is necessary to express the vector

Lub, VbW,
wb, b in terms of displacement components
[u*b,v*bw* lt J of the shell reference surface at the
line of attachment of the ring to the shell. (The displacement components u* and w* are shown in Fig. 18.
The circumferential component v*= v.) This transformation has the form
Cb
2x2
4x4 4x4
[0[B
r4x4
-e,l
- e2 n/r (I + elr) - elnlr 0
'1
(153)
'
*+ |+ [Q.
(158)
+o+O,Q21]
in which the superscript "b" has been dropped for

convenience, subscript I refers to the meridional station
corresponding to the mth junction between segments,
and [Ql m] and [Q2'] are analogous to the negatives of
[El] and [F2]:
0
-I
e2I
di
[m]nd2lr -(l+d,lr) nd'/r

0
e2/r
[E
02]0
L 0o
0
0
-e2n/r
0
0
-e
0o
-et
Finally, the vector Lu*b, V*b, w*1b, 3bJ

I must be expressed,
with the help of eqns (107), in terms of the nodal point
degrees of freedom
U *b
fib
[T){qb}
(159)
-I
(154)
o-.
=- nddr - d2/r
v*b
(155)
d2
nd 2lr 0
0 di
0
0.
(160)
In eqns (159) and (160) superscript m has been omitted

for convenience from the arrays. The A, A2 ",A 3", and
lA4 ' are the mth set of Lagrange multipliers associated
with the Ith station at which constraints are imposed on
the quantities u*, v*, w* and P. For example, the con-
D.BUSHNELL
500
straint conditions between Segments I and 2 in Fig. 2()(b)

(m = 2,1 = 7) arise from the requirement that the mothion
during buckling or vibration of point D relative to p()int
C involves no deformation, only rigid body translat ion
and rotation of the ring cross-section. The quantity All.
corresponds to compatibility of axial displacements IU*_
and U*+; A2' corresponds to compatibility of circi IMferential displacements L* - and
; k,"' to compatibi litY
of radial displacements w*- and w*+; and A4m to ccompatibility of meridional rotations 3 - and J h
Displacement boundary conditions applied at the A
and B ends of the meridian (see Fig. 20) take the fiorm
The boundary conditions (161) take a similar form:

[0]
U."'- l~,AqI[KQT]
q
[KQT]T
[0]
[0]
[0]
[0]
[0]j
[0]
qJ
(168)
The contributions of the junction conditions (164) and

boundary conditions (168) to the global stiffness matrix
of the structure shown in Fig. 20 appear in Fig. 24. The
boundary conditions at A contribute the elements
[KQTJ, and [KQTL T,; the compatibility conditions for
conformity of displacements and rotation at the junction
between Segment I and Segment 2 contribute the ele-
IA*1
U," -
[K]
,A 2 " ,A"" A4 ']
ments [QT]2 , [QT2 T, [T], and [T]T; and the boundary

*
conditions at B contribute the elements [KQT] 3 and
+3oQ21"'
V* |[KQT]3T.
[Q.'
(I161)
Corresponding to the end A of the meridian (m =I):

0
KA['
[K ... =
KA
-o
01
0
0
KA3
[62)
0
K,
Live load effects

The expressions (98) and (99) for the potential energy
of the applied loads are valid for constant-directional
loads, Cohen[29] gives the conditions of conservativeness for a load that rotates with structural deformations,
such as a pressure acting normal to a continuously
deforming shell. The potential energy expression, including pressure-rotation or "live" load effects, is given
by Cohen[29] for a shell of revolution as follows:
4 .
Corresponding to the end B of the meridian (m = 3):

Up 2 =
IKB 0
0
0
0
01
Ka 2
0
0
K4 3
0
KB4
J|(plu+p
(163)
Equations (162) and (163) are analogous to eqns (27a,b)

in the curved beam analysis. The quantities KAI, KA21
etc. and Kai, KB2 , etc. are assigned values, either unity
if the corresponding displacement component is zero or
zero if the corresponding force component is zero. The
displacement conditions correspond to a shell which is
supported at distances di' and d2' from the reference
surface. For the shell in Fig. 20(a) the, KA, KA2, etc.
would all be zero and the KBI, KB2 , etc. would all be
unity. The constraint conditions (158) and (161) can be
written in terms of the vectors [q'] and [q ] by use of
eqn (155). The compatibility condition (158) can be
written as a symmetric quadratic form in the following
way:
+ IP3(
[q ,Aq'
[F]JA
(17 or
~hr (LqhJ
(r
7x7
[0]
r
- 4X7
[
tQT]
7x7
7X7
7x7
[QTIT
[01
4x4
4x7
[0]
7x4
L[f0] [TIT
[T]
(166)
7x7
+u
[Qi" +
3otQ2 "].
(Rg
K)+
jp3]
r d ds.
[713
31,3
3II7
D]T[PfI[(D]{Jq"
or27)
X3 3X7
(169)
(170)
(171)
in which rk is the radius r evaluated at the midlength of

the kth finite element, [D] is given by eqn (114), and [Pt]
is
0
-
0fIR
- P30R2
t (172)
pilofR1 , + IIR 2 )
The matrix [API is given by eqn (172) with pi' replaced

by AP3.
There is an analogous contribution from the horizontal
line load H (Fig. 18) acting normal to the deformed
centroidal axis of the discrete ring. The contribution of
the line load-rotation effect to [K,(p(-)f)] arises from
+41tf4
4L4f
('7ror2nr (bj[T~r[E
2
r,(t\l CTfJ-[P 13E 21T[HfI
[0]
4x4
(7 r27 hrk([q b[ D] T[AP][D]{qb})
L 3
(165)
"m
+P3 W)
and the contribution to tK2(Ap(1,,)l arises from
[pf]3
with
2)
Oor27 hr [
(164)
wqit
The quadratic terms in eqn (169) contribute to the

stiffness matrix [K,(p(,,)')] and to the load-geometric
matrix [K 2 (Ap(-)]. The contribution to [K1(p(f.))] arises
from
fq1
U,.
(167)
4x7
7
x [E + olfE2][T]lq b})
(173)
501
Hence, the bifurcation buckling eigenvalue problem

might be better posed as
in which
[0
00]
- H'Irc 0
4..4
[H ]=
-O
The contribution to [K 2(Ap(,))] arises from the same
expressions with Hf replaced by AH and Hf by AH.

Summary of bifurcation buckling matrices
The eigenvalue problem for bifurcation buckling is
expressed in eqn (123). The contributions of the three
energy components US, U,. Up, to the kth finite element
local stiffness matrix [K,(p(-),/)]k and load-geometric
matrix [K2 (Ap ,,,)]k can be combined from eqns (136),
(157), (170), (171) and (173). These local finite element
matrices, divided by a common factor, ir or 27r (depending on n), are given by
[KI(p(o)' = 10,n = 8)]{xb}+ A8 [K 2 (Ap(0o = I,n = 8)]{xb}
= 0.
0D
T
[P, + p
0f92]T
(178)
The point on the dashed curve labeled "I" indicates the

first computed bifurcation buckling load estimate, which
is calculated from
[N.'] [RI + [D] [P] [Dilk
8jkrcj [[TIT
(177)
Figure 25 shows a sequence of critical load estimates

p_,(n) that might result, for example, from an analysis of
a spherical cap under uniform external pressure, such as
is depicted in Fig. 23. The computer program user provides a range of n, flmin - n -- n-,, which is to be
explored during a search for the minimum pc,(n). The
user also provides as input initial values for n,p(o
10 , and
Ap(O). Usually, he will choose n in the middle of the
range nmin to nma,,, p(o)f=0, and Apo=unity. Such a
choice yields eigenvalues A, which are numerically equal
to the bifurcation load pC,(n). Figure 25 indicates an
initial choice of n = 8 and p(o)f = 10.0. Suppose that
initially Ap(o)= 1.0. The first eigenvalue problem to be
posed and solved is
T
0 'B2 ]
2] [C][B, + 3
[K,(p(.~) ]k = hr, [(B. + 0fB
+ [B]
[K,(p(m,,n)]{x} + A[K2(Ap(m),n)]{xb}=0.
(174)
H'/r. 0
0 O .
([B,]T [G] [B.]
firD 1
iTr,
+ [H'])[
L-Ir .
J "-"J
~70.
p_,(8) = p(o)f + A8*Ap(O)
(179)
irn
+ [Hlf])[E, +
(175)
PlF2] [T]jj
09
From the scale shown in Fig. 25 and the initial conditions

NWf = 10.0, Ap(o) = 1.0, it is seen that the first eigenvalue
A8 calculated by the program would have been about 60.
With p(o)f fixed at 10.0 and Ap(1 ) fixed at 1.0, an
eigenvalne is next calculated for n = 9. The result.
pc,(9 )= 10.0+A 9 *1.0=60
T
T
[K2(ip(.)j)]k-= hr, [[R] [AN.] [R] + [D] [AP] [D]]k
+ Sjkri [[TIT [E, + 00f3E 2 ] ([B.]
+ [AH])[E, + 10 1E2 ] [T]]j
[AN,0 ] [B.]
(176)
in which k indicates "kth finite element", Sjk are

Kronecker deltas, and j indicates a station to which is
attached a discrete ring.
Figure 24 shows how the local stiffness matrices [K 1 ]k
are assembled or accumulated into the global stiffness
matrix [K,] for the two-segment structure shown in Fig.
20(a), discretized as indicated in Fig. 20(b). Each of the
thirteen 7x7 subarrays [K,]" is centered on the main
diagonal of [K,] and overlaps its neighbors as shown. The
constraint condition arrays given by eqns (166) and (168)
are assembled into [K,] right after assembly of the local
stiffness matrix corresponding to the station at which the
constraint is applied. Assemblage of [K2]k into [K2] is
similar, the only difference being that there are no contributions from the constraint conditions.
Table 2 summarizes where the various terms in eqns
(175) and (176) come from and their physical meanings.
is labeled "2" on the dashed curve and corresponds to a

smaller eigenvalue than that for n = 8. Hence, n is further increased by the computer program until a minimum
p_(n) is perceived. The loads p(o) and Ap(o) are held
fixed during this phase of the calculations, so that the
1i.'
MINIMUM
FROM
VAPPROXIMATE
THEORY
'
_-<_
50
LUJ
~0
FINAL
4--4
LOAD-
U0
30
0
I-
-MINIMUM
"EXACT'
20 -
I-)
10
Computational strategy for calculation of critical bifurcation load

The stiffness matrix [K,], load-geometric matrix [K2 ],
and eigenvalue A in eqns (123) depend on the number of
circumferential waves n in the buckling mode {xb}.
(180)
FROM
THEORY
0
1 INITIAL
n)
u
ESTIMATE
(INPUT)
10
11
CIRCUMFERENTIAL
WAVE
OF LOAD
I
12
NUMBER,
13
n
Fig. 25. Strategy for calculation of critical buckling load pb.
502
D.
BUSHNELL
Table 2.Physical explanations of terms in local stiffness and load-geometric matrices eqns (175) and (176)
Term
Number
Derived
From
Equation
Number
Physical
136
Stiffness
p
(i)
136
170
Explanation of the Term
matrix for
(Smeared
shell
stiffeners
Modification
of shell
prestress
shell
in
as deformed by the
'fixed'
load
included here.)
stiffness
due to
'fixed'
membrane
wall
Pressure-rotation
(live
load)
effect
from
'fixed'
load
f
P (m)
157
Contribution
by
157
'fixed'
Modification
buckling
to
173
Line
stiffness
of discrete
ring
as deformed
load.
of discrete
hoop force
load-rotation
ring
from
stiffness
'fixed'
(live
due
to
pre-
load
load)
effect
shell
wall
from
'fixed'
load pf
9,
136
Work done by prebuckling
171
increments
due to
load
modal shell
wall
rotations.
Pressure
crement
157
rotation
crement due to
x
X
modal
@
(live
load)
173
Line
effect
resultant
during buckling
from load in-
Ap( ).
Work done by prebuckling
stress
increment Ap (M)
load
discrete
increment
ring hoop
AP(m)
force in-
during buckling
ring rotations.
load-rotation
(live
load)
effect
from load
increment Ap
no
nonlinear prebuckling analysis is not repeated for n = 9,

10, 11, or 12.
The program "perceives" that the estimated critical
load p,,(ll) is a minimum in the range of n provided. At
this point new values of the "fixed" load and load
increment are established
p(m) = p(O) + A(ll)*Ap(p),
(181)
Apo) = P(o)llOO.
(182)
The load increment Ap() is set very small compared to

"fixed" component p(,, to minimize the difficulty of
finding a nonlinear prebuckling solution at the load
P() + Ap(l). Also, the small increment added to a relatively large fixed load yields an accurate approximation
of the rate of change of prebuckling stress resultants in
the neighborhood of p(Vf. New prebuckling solutions are
obtained for loads P(,) and p(lf + AP(,). A new eigenvalue problem
[K,(pj)fn = II)I{x6 } + A,,[K 2 (Ap(I), n = Il1)]{xbl
6
From Fig. 25 it is seen that p,' ,)(l1)
is considerably less
than p_(4)(11), so that Al,(6) must have been negative.
For the next two critical load estimates, per() and
pr(8 ), the number of circumferential waves n is held
constant at 11 and new values of p(,)f and AP(m
established until IA*Ap(,)l is smaller than lp(.)fllOOOI.
First, new nonlinear prebuckling solutions are obtained
corresponding to
P(2,f= P",(.)(l 1);
AP)
=P(2)flO000
(185)
and a new eigenvalue problem

7
[K,(p( 2 )f,n = lI)]{x }+ A,( '[K2(Ap( 2 ),n
ll)]{x
=0
(186)
is solved, leading to the result labeled "7" in Fig. 25:
7
PC,( )(11) = p(.)- = P(2)f + AII() *Ap( 2 ) - 36.
(187)
This process is repeated once more before convergence

at
(183)
P-zr
is set up and solved, leading to the result labeled "6" in

Fig. 25:
P-
(6)(1 1) = P(J)f + A 11(6)* AP")
29.
(184)
(l 1)= p( 4 )f = p(3 )f + AII
81
*Ap(3).
(188)
At this point the circumferential wave number is again

varied with P(4)= p_,(8)(ll) and AP(4) = p( 4 )f1000 held
constant. A new minimum critical load is perceived at
503
n = 10. Once more the procedure described in connection with eqns (181)-(188) is followed. The final critical load estimate is p(,r2) and the corresponding critical
circumferential wave number is n = 10. This is the load
denoted Pnb in Fig. 23.
Pitfalls
The strategy just described works well if the collapse

load p,, corresponding to axisymmetric snap-through is
higher than any of the estimates p(.)f and if the eigenvalues An(J) always correspond to the lowest bifurcation
point. The strategy must be modified if the situations
depicted in Figs. 26 or 27 exist.
LostUt Ilfirc.
Point,
If the program user sets p(O/f = 0 and Ap(O) equal to a
very small fraction of p.1, the dashed curve in Fig. 25 will

correspond closely to a linear bifurcation buckling
analysis. If the minimum pj,,4)(n = 11) corresponds to Plb
in Fig. 26, that is, if it is higher than the axisymmetric
collapse load p.1, the program will be unable to determine nonlinear prebuckling solutions for p(I) or p(Jf +
VOLUME DISPLACED
Fig. 27. Situation in which the strategy corresponding to eqns

(178}(188) fails because AII in eqn (184) is not the smallest
eigenvalue inabsolute value.
Ap(o).
Note that the result labeled "6" in Fig. 25 implies the

calculation of a negative eigenvalue A11(6), as discussed
immediately following eqn (184). If the situation shown
in Fig. 27 exists, conventional subroutines for the
extraction of eigenvalues will yield the second bifurcation point, A,1 (2), rather than the first, A,,('), for n = 11
circumferential waves.
creased until one or more eigenvalues are detected

between two sequential load steps (e.g. stability determinant changes sign in Fig. 28(a)) or until the maximum
allowable user specified load has been reached.
At this point in the calculations a series of eigenvalue
problems of the form
Modifications of strategy to avoid pitfalls
The pitfalls illustrated in Figs. 26 and 27 can almost

always be avoided by the following approach. The program user first selects an initial number of circumferential waves no which he feels corresponds to the
minimum bifurcation load. For this wave number no the
stability determinant I[Ki(p(,,) no)]I is calculated for a
sequence of load increments Ap(o) as shown in Fig. 28(a).
The initial load p(o)f and load increment are chosen to be
fairly small compared to the expected critical load, say
about p-/10.
For each load increment the nonlinear prebuckling
equilibrium state is determined and the stability determinant is calculated for n = no. The load p(,)' is in-
[K1 (p ('-.l),n)]{x }+ An[K 2(Ap(o),n)]{x}
0= (189)
is set up and solved, where [K 1 (p(,, I),n)] = the stiffness
matrix corresponding to n circumferential waves, of the

structure as loaded by L, = pf,, ; [K 2(Ap(o), n)] = the
load-geometric matrix corresponding to the prestress increment

resulting
from
the
load
increment L2 - L,= Ap(O); p( ) = Li = the load state just
before the sign change of the stability determinant, or the
second-to-last load; prf, )+Ap(O=L2 =the load state
just after the sign change of the stability determinant, or
the last load; An = the eigenvalue; xb = the eigenvector;
and n = the number of circumferential waves; n lies in a

range nin -< n - n-,, with noin and n,-,, provided by the
program user. Note that the initial guess no also lies in the
range ni.n < no
tn.m,,.
Also note that the increment in n
need not be unity, but may be a suitable fairly small

percentage of the average of n.,. and n.,, say An = 0.05
(nmin + n,,,.). From this point on the strategy is the same as
already described in connection with eqns (178)-(188). If
Fig. 28(b) is used as a reference, the program next
calculates eigenvalues corresponding to A(o), Ano)+ Ano(),
no-2, and A
3.
New prebuckling solutions are then obtained for the

loads
P+
=P1)
p(.-1)+ A(nrit* AP(o)
(190)
and p[f +)+).+Ap(I), where Ap(1)=pf+1)/1000. A series of

new eigenvalue problems of the type in eqn (189) would
then be set up with n held constant at n = ncri, until
VOLUME DISPLACED
Fig. 26. Situation in which the strategy corresponding to eqns

(178)-(188) fails because the axisymmetric collapse load p", is
lower than the critical load estimate p~',?(n = Il).
|AAp(,)l becomes smaller than Jp(,f/1l000J, as described

earlier in connection with eqns (187) and (188). If some
of the eigenvalues An') corresponding to the problem
given in eqn (189) take on negative values, such as shown
in Fig. 28(c), the pitfall illustrated in Fig. 27 might still
prevent a finding of the true lowest eigenvalue. The
D.
504
BUSHNELL
IS
I-
LI
Er
(n
MAX
LOAD
LOAD, L
LOADL
(a )
&l
,
OD
(I)
Buckling Mode
ncr 18
Prebuckling _;
Deformations
I. b
CIRCUMFERENTIAL WAVE NUMBER,n
(c)
(b)
10274 Ibs/in
Fig. 28. (a) Stability determinant at n = no as function of load L,

(b), (c) eigenvalues as function of number of circumferential
waves, n.
Fig. 29.
following example illustrates such a case and presents a

remedy.
In this case, it is discovered from a count of the

negative terms on the main diagonal of the factored
stiffness matrix [K 1(p( 1)f = 12008, n = 12 )](factored) that
for n = 12 circumferential waves, three eigenvalues exist
below the "fixed" load p(,f= L2= 12,008 lb/in. Hence
the load is automatically reduced by a factor of 0.7, to
L3= 8414 lb/in. After the eigenvalues corresponding to
points 5-9 in Fig. 30 have been determined, the new base
load p(2 =L3=10,819 lb/in. is established corresponding to n = 18 waves. It is also determined that at n = 18
one eigenvalue exists below this new "fixed" or base
Example-buckling of a very thin cylinder under axial

compression
This example is included here because it is a difficult

case from a numerical point of view, since eigenvalues
are close together and close to the axisymmetric collapse

load, and the case demonstrates some of the internal
checks and automatic internal control in one of the
computer programs for buckling of shells of
revolution[18]. Because of these properties and because
the geometry is simple, this is a good test case for
computer programs that perform buckling analysis including nonlinear prebuckling behavior and nonuniform
prebuckling stress and displacement.
Figure 29 shows the model of a cylinder with radius
R = 500 in., thickness t = I in., length L = 2000 in.,
Young's modulus E = 107 psi and Poisson's ratio v = 0.3.
The cylinder is treated as being symmetric about the
midlength, and the 1000-inch half-cylinder thus analyzed
is divided into two segments: a 200-inch-long edge zone
segment with 83 nodal points, and an 800-inch-long interior segment with 99 nodal points. The axisymmetric
prestress model contains 379 degrees-of-freedom, and
ted critical load of 10,274 lb/in. and the critical buckling
mode corresponding to n = 18 circumferential waves.

Figure 30 shows the sequence of wave numbers and
explored by the computer
loads automatically
program[18]
to
obtain
the
final
result
L 6 = Nc,=
10,274 lb/in. With an initial base or "fixed" load p(o)f = 0

and a load increment Ap(o) = 1.0 lb/in., eigenvalues
labeled (1-(4) are calculated. The base or "fixed" load is
then set equal to the local minimum or 12,008 lb/in. and
the load increment is set equal to 12,008/1000 lb/in.
of axially compressed
Bushnell[181).
cylinder
(from
-]
12500
L) d
(4)
C
(3 -' - -`\ Ba~se Load
120006
0 lbs/in
i6s)
Li
Cal
.0
0
I1500~
Bose Loud - B414 lbs/in
(D
= L3
'\z
-jZ
is)
.
s(71
the stability model 566 degrees-of-freedom. Simple sup-
port conditions are applied at the edge, and symmetry

conditions at the midlength. Also shown in Fig. 29 are
the prebuckling displacement distribution at the predic-
Buckling
11000
jC,
(i)
0U
C.)j
0_
10500
10000
(lt
L4
Final Predicted Load = L6 -,2)=
21
lB
12
15
9
NUMBER OF CIRCUMFERENTIAL WAVES
Fig. 30. Sequence of axial load and circumferential wave number

estimates during calculation of buckling of cylinder with nonlinear prebuckling effects included (from Bushnell[18]).
load. However, the new base load need not be reduced

by some factor because initial inverse power iterations
for the eigenvalue nearest to L 3 = 10,819 indicate that
subsequent critical load estimates will further reduce the
base loads L4, L5, etc. to the lowest eigenvalue rather
than increase them toward the second eigenvalue. Had
the opposite trend developed, the program would have
caused the base load to be reduced to 0.9* 10,819. Figure
30 shows the final three load estimates, L4, L5, and L6.
Figure 31 gives the prebuckling load deflection curve
for this cylinder. The abscissa represents the difference
between the actual end shortening and the linear end
shortening that would exist if there were no prebuckling
rotation. Eigenvalues computed with Ap = 1.0 and
505
results, each run corresponding to a different "fixed"

load pf = No. The open circles in the load-end shortening
curve correspond to the various loads, No = 5000,
10000,... 11900. The large dots represent the "fixed"
loads used in the sequence shown in Fig. 30. Two to four
eigenvalues are calculated corresponding to each opencircle fixed load. These eigenvalues are indicated by
crosses on the same vertical lines as the open circles.
The eigenvectors are shown in Fig. 32. Notice that for
"fixed" load L, = p(o)f =0, the lowest four eigenvalues
are very close and are all approximately equal to the

"classical" load 0.605 Et2/R. The lowest eigenvalues are
also close for no = 15, the starting circumferential wave
number in the search for the minimum critical load
P(=)f =No=0, 5000, 10000, 11000, 11500, and 119001b/in.
p(,,)(n = ncrit). Therefore, several inverse power itera-
and n = 18 circumferential waves are indicated as

crosses. Several runs were made in order to obtain these
tions and spectral shifts are required to obtain the lowest

eigenvalue at that wave number. For No = 5000 lb/in. the
IIC
0.02
0.04
0o06
NONLINEAR PORTION OF END SHORTENING
008
ALTOT - L N, /Et
Fig. 31. Eigenvalue "separation" for axially compressed cylinder (from Bushnell[18]).
D.
906
BUSHNELL
I
3
ZI
In
NI
(-I-
15-~
x,
-t
I
'1
iS5
I-
L.
xo
*.0
WV
80
e t
I"
I I=
! iE
Iz
AI~
!s
S2

lowest eigenvalue "separates" from the others, and the
localized nature of the corresponding eigenvector is
strongly developed (Fig. 32). Because of this separation of
the lowest eigenvalue, fewer inverse power iterations
and spectral shifts are required for convergence. Thus,
the user may save computer time by choosing a base or
"fixed" load to be some reasonable percentage of the
estimated final buckling load. This is particularly true if
many values of the wave number n are to be explored
and if the predicted n corresponding to the minimum
p_,(n) is likely to depend strongly on the fixed portion of
the load, as is the case for axially compressed very thin
cylinders.
507
nP
U,
I-1
(
(CIRCUMFERENTIAL WAVES)
Fig. 33. Ring-stiffened conical shell with simultaneous local and

general instability modes (from Bushnell[34]).
Another pitfall-failure to find the global minimum

p.,(n)
In all of the examples shown in Figs. 25, 28 and 30 the

curve of critical load or eigenvalue vs circumferential
wave number n has a single minimum and corresponding
n-rit, given p(-)f. However, the curve of pNr(n) vs n for
optimally designed stiffened shell structures often has

several minima, all of them at approximately the same
load, as shown in Fig. 33. This is because the minimum
weight design often yields a configuration in which
general and local instability occur at almost the same
load. In Fig. 33 the general instability mode and local
modes corresponding to primary failure in the first,
second, and third bays of the conical shell are shown as
insets in a plot of pcr(n) vs n. In addition the rings might
cripple at still higher values of n. It is generally up to the

computer program user to provide a wide enough range
of n or to make several runs with different ranges of n in
order to cover all possible failure modes.
Physical intuition is invaluable as a guide for finding
the absolute minimum load. One may idealize each bay
of a ring stiffened shell by assuming that the bay is
simply supported, calculate corresponding "panel"
buckling loads with certain appropriate ranges of n, and
then use the critical loads and values of n as starting
points in an investigation of the assembled structure.
It is not necessary always to increase the circumferential wave number n by one. In the search for the
minimum buckling load, for example, one may only be
certain that the n corresponding to the minimum buckling load, ncri,, lies in the range 2 - n - 100. One
might, therefore, choose An = 10 and "zero in" on a more
accurate value in a subsequent run.
Experimental evidence is of course very useful in
determining a good choice of initial number of circumferential waves no and range limits nmin and na,. If none
is available the user is advised to try the following

formulas:
(I) "Square" buckles for short cylindrical or conical
(4) For axially compressed conical shells and frustrums:
Use formula 2 where the circumferential radius of curvature, R, is the average or the radii at the ends.
(5) Spherical segments of any depth under axial tension
n = 1.8 *(R/t)" 2 sin [a, + 4.2(tlR)" 2 ]
where a, and a2 are the meridional angles measured

from the axis of revolution to the segment beginning and
end, respectively (a, < a2).
The above list of formulas is by no means complete.
However, notice that (Rlt)112 is a significant parameter. If
n is known for a shell of a given geometry loaded in a
certain way, a new value can be predicted for a new Rit

through the knowledge that n often seems to vary as
(R/t)' 2. (R is the normal circumferential radius of cur-
vature.)
MODAL VIBRATION ANALYSIS FOR PRESTRESSED SHELLS
For modal vibration analysis the stiffness matrix

[K,(pf,n)] of the shell as loaded by pf is the same as that
appearing in the eigenvalue problem for bifurcation
buckling. The eigenvalue problem for modal vibration
analysis is
2
[KI(p',n)]lxb}+ Q1
[M(n)]{xh} = 0
(191)
in which [M(n)] is the global mass matrix corresponding

to n circumferential waves.
Starting from eqn (100) for the shell and eqn (101) for
the discrete ring, and assuming that the displacement as a
function of time is
(192)
{x} = {xb} eift
shells or panel buckling
n = rr/L, where L is the shell meridional arc length

corresponding to the half-wavelength.
one can derive the local mass matrix for the kth finite
element:
7x3
(2) Formonocoque deep shells, axial compression
[M]
= hrkmk[[D
3x33x7
[I][D]]k
n proportional to [(Nominal circumferential radius of

curvature)/t] 12
(3) For shallow spherical caps supported rigidly at

their edges; external pressure:
+ [RJ T [Ta][R]Ij
n = 1.8 * a2 * (R/t)"2 - 5 (a2 = angle in radians from the
axis of revolution to the edge).

CAS Vol. 18, No. 3-1
jkr
[A[T]E +
f[TA][fI +
pf- T
3x3
(193)
in which mk is the shell wall mass/area, [I] is the identity

matrix, pj is the jth ring material mass density, A is the
D. BUSHNELL
Uring(linear in q)= (r or2?r)r, LNrT ] [B,][EiJ[TI
discrete ring cross section area, and

I
TA
15
Is
TB =
(194)
1 0
T 4 0
Ik
/Ak
1N
n
in which the transformations involving [B.], [D], [B,],

[E1], and IT] appear in eqns (134), (113), (148), (152) and
(155), respectively, and
LPI
(195)
components of the moment of inertia of the discrete ring

with respect to the (s,n) axis system, which is shown in
Fig. 20(a).
(201)
Corresponding to the kth finite element, the local force
vector, divided by the common factor 7r or 27r (depending on n), is given by the negatives of the r.h.s. of eqns
(199) and (200) without the {q}:
I()
LINEAR STRESS ANALYSIS
surface and L(0), acting along discrete ring centroidal

axes, are decomposed into Fourier harmonics, as listed in
eqns (112). For each Fourier harmonic, the stiffness
matrix [K,(pf = 0,n)] and "right-hand-side" vector {(fn)}
are formed. A solution {x(n)} to the linear system
(196)
is obtained and added to solutions obtained for previous

values of n. The stiffness matrix [K,] is the same as that
used in the buckling and vibration analyses with the
"fixed" load P(i)f now set equal to zero. The right-handside vector F(n) is formed from assembly or accumulation of the local right-hand-side vectors for each finite
element into the global vector. The local vectors are
generated from the first variation of the total energy
evaluated for the undeformed structure. Contributions to
these local vectors arise from the second term in the
integrand in eqn (89) (thermal loads in the shell), the last
three terms in eqn (95) (thermal loads in the discrete
rings), and all of the terms in eqn (98) (line loads on the
discrete ring centroidal axes) and eqn (99) (distributed
loads on the shell reference surface). The total shell and
ring energy components corresponding to these linear
terms in q are:
U,,,,h(inear in q) = Jf | 2LNTJ {J}r do ds

|(pIU +
p V2 + paw)r do ds
Ur,ng(linear in q) =f
-
(197)
in which [N T J is defined in eqns (84) and (86);
is
defined in eqn (128) (Remember ,8of = 0 in this analysis);

[NRTJ is defined in eqns (143) and (96); and {6e2"} is
defined in eqn (140). Using principles and equations
introduced in the discussions and derivations leading to
eqns (136) and (157), one can express eqns (197) and
(198) in the forms
U.he,,(linear in q) = (r or 27r) hri I.LN T] [B.I
- Lp [DI]{q}
[B,][EI[TI
LU [EI[T]1 j. (202)
The local "right-hand-side" or force vectors {F}k are

assembled into the global force vector {F} as described
in the discussion following eqns (175) and (176) regarding
the local stiffness and load-geometric matrices. Table 3
lists the equations that give rise to the various terms in
eqn (202) and identifies the physical significance of these
terms.
VARIOUS DISCRETIZATION METHODS
In the preceding development the discretizations

shown in Figs. 11 and 21 have been given the appellation
"Finite-Difference Energy method". However, as
emphasized in Ref. [191, the categorization of discretization methods into "Finite Element" and "FiniteDifference Energy" is somewhat artificial. In both the
finite-element and finite-difference energy methods the
unknowns of the problem are certain generalized displacement components located at discrete nodes in the
domain. Between these nodes the variations of the
generalized displacements are expressed as power series
in s. Integration can then be performed analytically or
numerically. The differing choice of generalized displacement components and locations of the nodes are the
only characteristics of the two solution techniques which
justify giving them different names. Once the nodes and
the appropriate generalized displacement components
have been selected, the solution procedure is identical
for both methods.
Figure 34 shows five types of discretization. The nodes
are denoted by large dots or crosses. The "element" is
crosses. Nodal point variables ui, v,, wi, xi, etc. are shown
(198)
{e('J}
O
&"kri LLN,'
8i-
defined as the solid horizontal line bounded by dots or
dLFrT]{EP)}rO
Vu, + SvD + Hw, + MP3)r dO
(2)
{F}J= -hrkL N'] [B,] - LPI [D]] k
Arbitrary loads p(s,0), acting on the shell reference
[K,(p' =0,n)]{x(n)}={F(n)J
Lp=,P2,P3]
[L] -- L- V,S,H,MJ.
1.
In
The matrices [D], [T], [Pi + P 0 'E2], and [RI are the same
as those used in the derivation of [K"]k and [K2]k,
appearing in eqns (175) and (176). The I, In, 1 are
-|
(200)
-[L] [R,][T]] {q}
(199)
next to the nodes with which they are associated. The

first three models fall into the category "finite element
method", the last two into the category "finite difference
method".
The three models 0, (0) and (3) represent standard

finite elements such as described in Kotanchik et al. [301,
Mebane and Stricklin[31], and Adelman et al.[32]. A
curved element (2) with extra internal degrees of

freedom (dof) ii,,DI, a2, and i32 permits rigid body
motion without excessive storage of energy. The internal
degrees of freedom represent corrections to the linear
function. Elements of this type are described in [311. An
alternate way of obtaining higher-order displacement
509
total element energy is obtained by multiplication of the

energy per meridional arc length by the element length 1.
The finite-difference formulas for variable mesh spacing,
given in eqns (110) and (111), are obtained by Taylor
series expansions of the displacements about the centroid of each element. Since first and second derivatives
of w and only first derivatives of u and v occur in the
integrand of eqn (12), the appropriate polynomials for the
lowest-order difference formulas in each case are shown
in Fig. 34(d-e). As in the case of the finite-element
method, the a, can easily be expressed in terms of the
nodal point variables. The finite-difference energy discretization model ( has been used in computer programs by Stein[33], Bushnell[18], and Brogan and
Almroth[22].
Figure 36 shows the finite-difference discretization of
the same five-element structure depicted in Fig. 35. In
Fig. 36 the element boundaries are at u and v nodal
points (crosses). The quadratic w expansions pass
through three adjacent w nodes, spanning a longer arc
functions is to define more degrees of freedom at the

nodes [321. Element (D is of this type. The displacements within each of these elements are given by the
polynomials shown in the figure. Integration of the
energy functional can be performed analytically or
numerically. Gaussian quadrature seems to be the most
accurate and economical method of integration.
Figure 35 shows schematically a structure consisting
of five elements. The displacement function w and its
first derivative are continuous throughout the domain.
The displacement function for u and v corresponds to
model (D. Of course, the elements need not be flat.
However, if the element is curved, higher-order displacement functions than linear in u and v are required
for representation of rigid body motions.
Model ()represents the discretization method described here in previous sections. As has already been
mentioned, in this finite-difference energy discretization,
the integrand of the energy functional H (eqn 12) is
evaluated at only one point within each element, and the
Standard lowest order finite element
0D
al
0
V '
C1+3+
+ 025
(a)
04S
+ oss
87s
0,5+ 065 +
Energy evaluated at Gaussian integration points.
Finite element with extra internal d.o.f.
2
U -
v = a,5 +
a1 + a25 +035
+ a7s
06s
+,
2
4S
+ a8s
w = a9 + alos + al-s
(b)
+_ . 1 2 s
Energy evaluated at Gaussian integration points.

Static reduction used to get local IKI.
Finite element with extra nodal d. o. f.
U
alI +
V -
a5
09 +
a2s + ce3s2
2 +
F 06 + a7s
010S +
a4s3
3
(c)
+ o8S
a II"
.S2s
Energy evaluated at Gaussian integration points
Finite difference with (u.ivi) on half-stations
01 + a2s
(d)
[.
wII
wit
i+1
22
= a 35 + a4s
w
as + a6s + a7S
Energy evaluated at s
i/2.
Finite difference with (w ) on half-stations

u
'Pi
KvCi
al
+ 02S
(e)
V '
Wii+1
w=a
Q3 + o45
5+
a6s + a7s
Energy evaluated at s - e/2.
Fig. 34. Various discrete models for energy methods (from Bushnell[19]).
D. BUSHNELL
510
Table 3. Physical explanations of terms in local force vector and where they came from
Term
Number
Derived
From
Equations
G
0
Physical
Explanation
of Term
89),
134)
Thermal
99)
201)
113),
Surface tractions
shell
reference
and pressure
surface.
95),
152),
148),
155)
Thermal loading
on discrete
98),
(
155), (
152),
201)
Line loading along discrete

centroidal
axis.
length than the element. However, the integration, or

"lumping" corresponds only to the lengths between adjacent crosses. The in-plane displacements u and v are
continuous everywhere. Notice that at element boundaries the normal displacements and derivatives are discontinuous. It can be shown that the displacement discontinuity Awsis of maximum order
Loading on shell
wall.
acting on
ring.
ring
CONIC SECTIONS
L\W2
zw
ES
1I
U,V
IWij = hih;w (hi-, +2h, + 2hi+, + hi,2 )
(203)
and the slope discontinuity is of maximum order

tAj341h,
- h
(h, + 2hi + 2hi+ + hi+,2 ). (204)
The finite-difference discretizations and @T

are
similar to replacement of the actual structure by a structure consisting of elements linked as shown in Fig. 37.
The normal displacement w is continuous at the pinned
joints and u and v are continuous at the stations where
the projections stick into the rounded holes.
At first glance it would seem that this structure is far
too flexible to represent the behavior of a continuous
shell. However, notice in Fig. 36 that u and v must be
continuous at the stations where w is discontinuous.
Since the circumferential strain, for example, involves at
least both w and v, the Aw at the element boundaries
must remain small enough to keep this membrane strain
component, and hence the energy, at a reasonable level.
In other words, the minimum energy state will involve
small discontinuities in w at the element boundaries.
In Fig. 22 are given the computer times required to
form the global stiffness matrix [K] corresponding to
the problem shown in Figs. I and 2. The curve labeled
"Finite Element Analysis" corresponds to Model ( in
Fig. 34. As seen from Fig. 2, a high order element of this
type (cubic in u and v) is required for convergence to a
reasonably accurate answer with a reasonable number of
elements.
Figure 38 illustrates why the computer time for the
ELEMENT NUMERS
Fig. 36. Displacement functions for finite difference model

(from Bushnell[19]).
TJ
formation of [K,] is much higher for Model ( than for

Models ( or ():
(I) There is an extra loop over the number of Gaussian
integration points per element.
(2) There is more algebra required for formation of the
kinematic matrix B and more products required in the
formation of BTCB (B has more columns).
w, X
u,v
Big. 35. Displacement functions for finite difference model T

(from Bushnell[19].
Fig. 37. Structural equivalent of finite difference model (ior (D

(courtesy Carlos Felippa) (from Bushnell[19]).
(3) An extra step is required outside of the Gaussian

integration loop in order to condense out the internal
degrees of freedom in each element.
In spite of the higher order of the finite element Q, the
convergence of edge displacement w with increasing
nodal point density is far more rapid with the finite
difference element 3). For an accuracy of 2.0% in w
about 20 finite difference elements would be required, with
0
FINITE ELEMENT
METHOD
STIFFNESS MATRIXCALCULATION
I 5nuber ca
exIlement
about 0.25 sec. needed to form the global stiffness matrix

[K1]. For the same accuracy, about 60 finite elements of
the type 0 would be required, with about 3.5 sec.
needed to form [KM]. The difference in the rate of convergence for the two discretization methods probably
results from the following:
(I) The lower order (stiffer) approximation for u, v,
and w in the finite difference element is compensated by
FINITE ELEMENT METHOD

m
STIFFNESS MATRIX CALCULAI ON
WITH STATIC REDUCTION
511
FINITE DIFFERENCE METHODS

S
AND
S
STIFFNESS MATRIX CALCULATION
> number o
erent
nmber to\
e/h point
I>
YES |YES
FSTIFFNESS MATRIX FTIFNESS
HES
MATRIX FINISHED
TIFFNESS MATRIX FINISHED
Fig. 38. Flow charts for the calculation of stiffness matrix K1 with finite element models (Dand(
models or ) (see Fig. 34).
and finite difference
512
D. BUSHNELL
discontinuities in w and : at finite difference element

boundaries.
(2) The fact that the curved finite difference element
energy is evaluated at only one station per element
results in a more energy-free representation of rigid body
motion than of the curved finite element with cubic u, v,
and w, in which the energy is evaluated at more than one
Gaussian integration point.
CONSTITUTIVE LAW [C] FOR COMPOSITE SHELL WALLS
Here the phrase "composite shell wall" is taken in a

broad sense. It may mean laminated as described by
Jones et al. in Ref. [6-8]. It may also mean a shell wall
modified by stiffeners that are to be smeared out in the
computerized model.
Equations (80) and (84) are based on an assumption
that the principal axes of orthotropy of the shell wall
material are aligned with the orthogonal coordinate lines
on the shell reference surface. Thus, they are valid for a
stringer and ring stiffened shell wall such as shown in
Fig. 39, in which the stringers follow meridians (s-coordinate) and the rings follow parallel circles (0-coordinate). If the analyst wishes to set up a model in which
both sets of stiffeners are smeared or averaged over the
entire surface, he can treat the shell wall as if it consisted
of four orthotropic layers with properties G, E,, E2 , v, 2,
m, a,, and a2 assigned as listed in Fig. 39. If the
stiffeners do not follow coordinate lines, the more
general model described in Ref. [6] is required.
On p. 154 of Ref. [6] Jones writes the equivalent of
N.
N2
N, 2
A,,
A 12
A,6
A 12
A2 2
A26
A, 6
A2 6
A66
B,1
M,
B.,
B, 2
B16
B,
B 22
B 26
B, 6
B 26
B66
M2
M, 2
B 16
B12
B 22
B2 6
B,6
B2 6
B66
D,,
D,2
D16
D,2
D2 2
D26
D, 6
D2 6
D66
1B12
e,
e2
e, 2
K,
K2
2K 1,
2
(205)
in which the stress and moment resultants are defined as
in Fig. 17 and the reference surface strains and changes
in curvature are defined as in eqns (90). The Ail, Bj, and
Djj in eqn (205) are, of course, equivalent to the corresponding Cj in eqn (84). For the laminate Jones gives
N
A,j =
Zk
-I
(Q)k (Zk
6
k=l
i2 2 , 4 + )kE
(206)
Z k+
2k=l
IN
D(j
j- z
(0ix(Zl,
k=l
c E+(E,
1
2G
1)2
22
c ,s E 22
in which the Zk are measured from an arbitrary reference

surface, as shown in Fig. 40. (Jones [6] uses the middle
surface, but this is not a necessary restriction.) The Qij
for each lamina are given by Jones on p. 51 of Ref. [6]:
oil = E,,'c4 + 2(E,2 + 2G) S2CI + E21 S4

Q,2 =
(El,, +
E22
- 4G)s c+
Gs+c)
Q22 = E,,s4 + 2(E,2 + 2G) S2C2 + E22C4

Q,16
= (E,,I - E,12 - 2G)sc3 + (E,2 -
E22
+ 2G)s'c
Q26
(E,,
Q66 = (vt,
F, 2 -2G)s 3 c +(E,2 - E2 2 + 2G)sc'

2 2
+ E 22 - 2E,2 - 2G)s c
+ G(s4
c4)
(207)
in which E,,, E, 2 , E22 are defined in eqns (81),
s - sin y
c - cos 'y
(208)
and y is the angle from the meridional direction to the

"I" axis of the lamina (direction in which E, is
measured).
Note that use of the full matrix in eqn (205) in the
analysis of shells of revolution would prevent the
separation of variables according to eqns (108). Hence, in
most one-dimensionally discretized analyses of stress,
buckling, and vibration of composite, laminated shells of
revolution, the Cj, in eqn (84) are derived as in eqns
(205)-(208) and the terms A, 6 , A2 6 , B,6, B2 6, D,6 . D26 are
subsequently ignored.
SECTION 4
HYBRID BODIES OF REVOLUTION
INTRODUCTION
By "hybrid" is meant a body of revolution with both

one-dimensionally and two-dimensionally discretized
regions, such as shown in Fig. 41. The stresses and
strains for distances equal approximately to one wall
thickness from the junction shown in Fig. 41 cannot
adequately be predicted with thin shell theory. Therefore, a small region is defined in which the domain is
discretized in two dimensions. Figure 42 shows other
examples in which such a hybrid model might be used
for accurate prediction of local stresses and strains near
structural junctions.
The existence of a special-purpose hybrid body-ofrevolution computer program for stress, buckling and
vibration analysis is justified because there exist many
practical problems in which the geometry is axisymmetric or essentially so.
The user of such a computer program can obtain
solutions involving fairly complex configurations with
reasonable computer times. A hybrid program of this
type should be used primarily for cases in which most of
the structure can be modeled with use of thin shell
elements, only very localized areas being modeled with
use of two-dimensional solids of revolution. If the twodimensional finite elements are introduced sparingly, the
bandwidths of the stiffness, mass amd load-geometric
matrices are generally narrow. Such matrices can be
stored very compactly with a minimum amount of indexing. Thus, the computer time to decompose the
stiffness matrix is small, and input from and output to
auxiliary mass storage devices is kept to a minimum.
Since most of the structure is modeled with very simple
thin shell elements, the program user is able to simulate
the behavior of an entire complex structure in one model
while retaining reasonable nodal point density and twodimensional discretization in critical areas.
There are in existence several general purpose computer programs[24] which can yield solutions to problems of the type just described. In fact, the idea of
analyzing hybrid bodies of revolution, consisting of thin
and thick segments is not new. Strickland et al. [36]
created a computer program, WASP, for the axisymmetric linear static analysis of such bodies. Zudans and
513
is
SECTION AA
ACTUAL
MATERIAL
PROPERTIES
Elastic
Moduli
Es
Rings
Er
Thermal Exp.
Coefficients
4
F
ShellWall EG
Stringers
Mass
Density
as
mr
ar
EQUIVALENT
LAYERED
ORTHOTROPIC
SHELLWALL
Shell
Wall
Layer
Thickness
El
F2
V12
2
02
hs
Ests/dI
msts/dI
a.
L1
Ert1/d 2
t2
ErL 2 /d 2
Orthotropic Material Properties
Ml
mrtl/d2
ar
mrL 2 /d 2
ar
Fig. 39. How to model a shell wall with smeared stringers and rings. Stiffeners and parts of stiffeners are treated as
if they were orthotropic layers or lamina (from Bushnell[34]).
Chow[37] wrote a hybrid program, BOXSHL, which

permits nonsymmetric linear treatment by means of
Fourier Series expansion in the circumferential direction.
Many authors have written computer codes which can
handle solids of revolution by means of axisymmetric
finite elements. Displacement functions and integration
schemes in these elements can be modified such that
they are adequate for the analysis of thin shells. Thus,
Wilson[16] introduces incompatible modes in order to
create correct bending behavior, Ergatoudis et al.[38]
derive higher order isoparametric solid elements with
one or two midside nodes, and Ahmad et al.[39] and
Pawsey[40] specialize these isoparametric elements by
taking advantage of certain characteristics of the
behavior of thin shells. Isoparametric solid elements are
also used for shell analysis by Larsen and Popov[41], by
Bathe et al.[42] in NONSAP, by Dunham and
Becker[43] in TEXGAP and by Sharifi[44] in NEPSAP.
Ferguson [45] extended the capabilities of FARSS [46] to
handle stress, buckling and vibration of thin and thick
shells and bodies.
These and other investigators favor using solid isoparametric two-dimensional or three-dimensional finite
elements for large deflection analysis of shells because the
kinematic relations are well known and the rigid body
behavior is represented exactly. Other work on thick shells
or solids in conjunction with thin shells is presented
in [47-50].
CHOICE OF FINITE ELEMENT FOR THE TWO-DIMENSIONAL
REGIONS
The guiding principle in the selection of an appropriate

element for analysis of the two-dimensionally discretized
"solid" or "thick shell" regions is that the element
should be capable of reproducing thin shell behavior as
SHELLS
TWO-DIMENSIONAL
REGION
.ayer
Nu, ber
Fig. 40. Geometry of an N-layered laminate (adapted from

Jones [6]).
Fig. 41. Hybrid body of revolution: discretization in one and two

dimensions (from Bushnell[35]).
D.
514
BUSHNELL
Hz
I+ led
ui
Fig
Pobemsfo
4.
wicahyrdpormiusfl(omBhnl[5)
well as general three-dimensional behavior. If this were

not the case, large spurious stresses and artificial constraints would be introduced at every junction between
"thick" (two-dimensionally discretized) and "thin" (onedimensionally discretized) regions.
The pressurized flat circular plate (Fig. 43) is a good
example to use in a discussion of what properties a
"solid" element must have in order to lead to accurate
predictions of "thin shell" behavior. Suppose that the
plate shown in Fig. 43 is loaded well into the plastic
range. Then at the axis of revolution, for example, the
topmost fibers are compressed and the bottom-most
fibers are stretched plastically. Since the plastic flow is
assumed to be associated with zero volume change, the
axial strain (normal to the surface) must be extensional
at the topmost fiber and compressive at the bottom-most.
The simplest assumption is that the axial strain varies
linearly through the plate thickness. It follows that the
axial displacement must vary quadratically through the
plate thickness. Thus, at least three nodes are required
through the plate thickness if the nodal point unknowns
are the displacement components. If only two nodes are
used through the plate thickness, the axial strain can only
be constant. Such a gross simplification leads to the
prediction of very large stresses normal to the surface of
the plate at its extreme fibers, a result that naturally

generates serious errors, especially in analyses in which
plasticity or creep are included.
It is easy to see that at least three nodes are required
in the radial direction in each element in order that
bending be possible. Wilson's[16] incompatible element
permits bending but still suffers from the insufficiency
just discussed relative to normal stress and strains.
The requirement of at least three modes in each direction immediately suggests use of an isoparametric 8-node
element. Such elements were first introduced by Irons in
1966[51]
and
subsequently
popularized
by
Zienkiewicz[52]. These elements can, as seen from the
results listed in Table 4, reproduce thin shell behavior. If
the programming is done in double precision, the "aspect
ratio" is essentially unlimited. In another analysis of a
flat plate, use of elements with a radial length-to-thickness ratio of 1000 still led to good prediction of the
deformed state.
BASIC EQUATIONS
In this section the possibility of plastic flow and creep

will be allowed. Therefore, the equations governing
equilibrium and bifurcation buckling will be derived from
the principle of virtual work, rather than from minimiza-
Table 4. Comparison of solutions and computer times for uniformly pressurized (P = 3.0 psi) flat circular plate with
10 elements (see Fig. 43)
Type of model
8-Node, 2-Dimensional Thin shell theory:
finite difference
isoparametric
finite elements
energy method
Item
(10 elements)
(10 elements)
(I) Maximum displacement (in.)

(a) Linear elastic
(b) Nonlinear elastic
(c) Nonlinear elastic-plastic
(2) Number of unknowns
(3) Maximum matrix bandwidth
(4) Total Newton iterations req'd
for entire case
(5) Computer time spent in the
Newton-Loop (sec)
(6) Number of "trials" (times that
material properties must be
updated)
(7) Computer time spent in updating
material properties (sec)
(8) Total run time (sec)
1.9356
0.38357
0.58089
109
19
37
1.9491
0.38043
0.57221
31
8
34
66.439
2.343
10
10
21.874
188.313
4.589
l6.932
515
2.0
10N.
UNIFORiM PRESSU'E,
PSI
= 330
0 i.1
0.0
N
10x 106PSI
i=
-2.0
-4.0
= 0-3
EACH ELEMENT
I-
HAS 8 NODES
-6.
I
.
2.0
.
0. 0
IU.
.
4.0
,.
I I
.0
6.0
R
i I
10.0
12.0
Fig. 43. Flat plate under uniform pressure modeled with 10

isoparametric 8-node finite elements (from Bushnell[35]).
AXIS OF REVOLUTION
tion of the total potential energy as was done in the

previous sections on the analysis of elastic curved beams
and thin shells.
Principle of virtual work
The first variation of a total energy functional H
(analogous to H for the beam in eqn 12) is
U3
[r
L] {be} dV (strain energy)
= fH
y
(h!
dV
mu
Vum
Fb 8 dV
- fu
Fig. 44. Variables used in the analysis of (a) thin shell segments,
(b) solids of revolution (from Bushnell[351).
(kinetic energy)
(body forces)
The variation AH is therefore given by
Volume
(surface tractions)
AH = dH
-q
- f
Fe
FJ 8i[f(11)] dA ("following" pressure)
Jf
.dA
+ EA,
(E
alkUk
+ ao) +
(constrInAl
m'
(209)
Strains are considered to be small but rotations may be
moderately large. The material is elastic-plastic and
primary or secondary creep are included in the analytical
model. The strain components [EJ are nonlinear functions of the displacements a. The displacements a anywhere in each finite element can be expressed in terms of
nodal point values 4.
hi(x, r)4.
(210)
In eqn (210) fi represents a displacement vector and 4i

represents the value of a at the ith node of a finite
element. For example, in Fig. 44(b) a has the components U ,
U2 , U3 = qi,, qi2 , qi3 .
a[f(q)i)dA
dO
-JFL'drL
(equilibrium condition).
-=
aL,-aadV
' aq )dV
alk8Uk
(constraint conditions)
=0
u
at-F
1 dqj1 -
- A(.aq
k=l
(211)
Since 5q is arbitrary, the equilibrium condition is
fv ( '
(line loads)
F, -8urc dO
-f
= lfaq =
8A
(' n
+ X aq',(X alkUk
+
+
xi(
ak
A1 ( ''
l/
u,, ) =
ak--)
-0
(212)
i = 1, 2, 3,. .. N.
Because the strains [eJ are nonlinear functions of the

nodal point displacements q, eqn (212) represents a
simultaneous set of nonlinear equations. These equations
must be solved incrementally because the material is
elastic-plastic creeping and hence its state is pathdependent. In addition to providing for an incremental
procedure, one must provide for some kind of iterative
or self-correcting technique at each load or time increment in order to prevent drifting from the correct solution path.
516
D.
BUSHNELL
Appropriate use of the Newton-Raphson method with

path-dependent material properties
Suppose that for some known starting vector qo of
dimension N, we have qj(qo) 6 0. We wish to determine
Aq such that i(qo + Aq) = 0. A system of equations for
the unknowns Aq can be generated by expansion of pi
about qo in a Taylor series. If only the linear terms in Aq
are retained in this expansion, one obtains the following
linear simultaneous equations:
N
Aqj
i = 1,2,3,...N.
(213)
These equations are solved for Aqj and a new estimate of

the solution q, - qo + Aq becomes available. Iterations
continue until Aq is smaller than a certain prescribed
percentage of q. This is called the Newton-Raphson
method. The short description above is a reiteration of
the discussion associated with eqn (56) for the nonlinear
analysis of the elastic curved beam.
A certain refinement in the solution strategy is
required if large deflection effects and elastic-plastic
material behavior are simultaneously present. The Newton-Raphson method can be used only if i/ can be
expanded in a Taylor series about the known origins qO
or q, or q2 , etc. The Taylor series expansion exists if and
only if the gradient afldq exists. In problems involving
material which loads plastically but unloads elastically
along a different path in stress-strain space, the existence of a unique aolaq depends on the use of a proper
strategy for taking into account both geometric and
material nonlinearity. One strategy is to establish a double iteration loop for each load or time increment: the
tangent stiffness coefficients and plastic and creep strains
are updated in the outer loop and the geometric nonlinearities are handled by the Newton-Raphson method
in the inner loop. In the Newton-Raphson loop the
tangent stiffness is not recomputed with each new estimate of the displacement vector q. While calculations are
proceeding in this inner loop, it is as if the material were
elastic, loading and unloading along the same path in
stress-strain space. Therefore, a unique value of olpaq
exists and the Newton-Raphson procedure is valid. It is
important to be assured of this validity because the
Newton method has certain favorable convergence properties which are well understood. If one is attempting to
predict loads at which a structure collapses, it is important to know that failure of convergence of the iterations indicates failure of the structure and not simply
failure of the algorithm to predict the behavior of the
structure.
Details on various solution methods for nonlinear
structural analysis, including methods for reliable prediction of collapse loads are given in Ref. [26]. More
information is also given there on the double iteration
loop and on the method used to update elastic-plastic
material properties in the outer loop of the computational
process.
Calculationof ali/aqj
The stiffness matrix for each Newton-Raphson iteration is ail/q, and the (i,j)th element of this matrix is
oil/aq;. From eqn (212), we can write:
do__2e
d =J
dJ
Lb
I+ [da,
(q
a
IdS
d- d\
md
dJd
-J
Ft .
[f(i)]
a qiaqj
',
ak
+uk
.'
,=, aqj
,
allk
aq
kE=a dq
A
qj
(214)
Note that we have assumed by use of eqn (210) that the

displacements a anywhere in the structure are linear
functions of the nodal point displacements, q. This may
be a questionable model for discrete rings attached to
areas on the shell which undergo moderately large rotations. The displacements anywhere in the cross section
of a discrete ring are expressed in terms of nodal points
on the shell reference surface in the neighborhood of the
attachment point of the ring to the shell. Since the cross
section of the discrete ring is assumed to remain undeformed in "classical" ring theory, the displacements
anywhere in that cross section involve sines and cosines of
angles of rotation of the shell wall. If these angles are
relatively large, retention of only the linear terms in the
series expansions for their sines and cosines may not be
very accurate. Hence, for the discrete ring analysis, the
displacements a are considered for certain analysis branches to be nonlinear functions of q. (For example, see eqns
(106) and related discussion.)
Equilibrium
Equilibrium is determined by iterative solution of eqns
(213), with eqn (214) used on the l.h.s. and eqn (212) on
the r.h.s. For statics problems ii= 0 and thus the last
term on the first line of eqn (214) drops out.
Bifurcation buckling and eigenvibrations
The following discussion is a review of that associated
with eqn (60) for the buckling of the beam. The calculation of bifurcation buckling loads and of modal
vibration frequencies can be posed as a problem of
determination of the values of a parameter for which the
equilibrium solution is not unique. In bifurcation buckling problems the parameter is a load parameter, and in
modal vibration problems it is a frequency.
If qo represents a solution of the equilibrium equations
(215)
Oi(q,,) = O.
Then we need to know if

(216)
oi(qo+ qh) = O
in which q is a non-trivial bifurcation buckling mode or

vibration mode of infinitesimal amplitude. Following the
Newton-Raphson iteration strategy, we can write
Ma
l= I qi' Iq-q,,-qi~
= - i(q.)
=0
(217)
i = 1,2,3 .... M.
Equations (217) are linear and homogeneous. A nontrivial solution q" exists only for certain discrete values
of some parameter contained in the matrix a8 laqqb. These
eigenvalues are the bifurcation buckling loads or modal
vibration frequencies.
Variation of displacements in the circumferential direction
In nonlinear problems the loading and the structural

response are assumed to be axisymmetric. In cases involving bifurcation buckling or modal vibrations, the
eigenvectors vary harmonically around the circumference with only one harmonic participating in each
mode. Thus, the modal displacements are denoted lb,
u
u=
U
U
u'
{U2
sin nO
b(n)
OS nl
sin no
U3
where u1b, u2 b, U3 b are the axial, circumferential and

radial displacement components, respectively. In linear
nonsymmetric stress problems, the displacement field u,,
u2, U3 can be written as infinite series
u, = 2 u, sin no + f u,(-n cos no
l
n=O
u, = E u2," cos no +
n=O
>
u2' 'lsin
In the discussion about the Newton-Raphson method, it

was emphasized that within the inner iterative loop the
material tangent stiffness is held constant. Since the
solution corresponds to an instant in time, the creep
strains [e'] and the thermal strains [eT] are also held
constant. The plastic strains [eP] can be expressed in the
form
(218)
csnon
mnbo
(219)
[epj
n-I
u,3
sin no +
(223)
= [eopj + [e _ eoJ [CO ]
in which [COT] is the transpose of a matrix which will be

derived later. The subscript ( )Oin eqn (223) denotes
"value obtained when the material properties were last
updated". These values are held constant throughout the
Newton-Raphson iterations. The total strain Le] changes,
of course, with each Newton-Raphson iteration. With the
use of eqn (223) in eqn (222), and recognition that LeOP j,
Le'] and [ET] are independent of q in the NewtonRaphson loop, we can write eqn (222) in the form:
fJ
da~dU
u=
517
cos no.
([eJ[DTo.It
q}+
[DTO]{
"-nl
+ eO. COT -
ln=O
These equations are analogous to eqns (108) for the thin

shell analysis.
Displacement components used for shells, rings, solids
In the analysis of the thin shell segments, u,, u2, U3 are

written in terms of u, v, w the meridional, circumferential and normal displacement components of the
shell reference surface. These components are shown in
Fig. 44(a).
In the analysis of a discrete ring the displacements
anywhere in the ring cross section are ultimately
expressed in terms of the displacements u, v, w of the
point on the shell reference surface to which the ring is
considered to be attached (see eqns 106 and 107). Further
details on the elastic-plastic discrete ring analysis are
given in Ref. [26].
In the analysis of solids of revolution u,, u2, U3 are the
dependent variables. These are shown in Fig. 44(b).
Eop
- e' - ETj [D
d d ) dV
(224)
where [DTo] is a matrix which can be shown to be

symmetric,
T
[Dm] =[I CO
] [D].
(225)
The expression (223) is also used in eqn (221) and eqn

(220) so that,
f
dU
e DO
OT _
tj}d e]+
aqiJ+LOCo-e
[DI I'
1) dV
D_0
dqi
_ET]
cEl
-p
aqj
(226)
Derivation of CoT
This derivation follows Stricklin et al. [53]. The change

in stress at an instant of time for a given temperature is
STRAIN ENERGY
{da} = [D]{de - de0 }.
General equations
The strain energy gradient is given by the first term in

the integrand on the r.h.s. of eqn (212):
The stress and strain components are

[do-] = [do-,, doa2, do-3 , do,13 , do,12 , do2 3 ]
6whr
dU f, [0" -
de V
6.6
e-e-ecC
(228)
(220)
where
Lo-]=
(227)
(221)
J. [D]
in which e', e', eT are the plastic, creep, and thermal

strains, respectively, and [DI is the constitutive matrix
given later.
The first two terms in the integrand on the r.h.s. of eqn
(214) can thus be written
[de] = Lde,, de2 , de 3 , de 13 , de 12 , de2 3 ] .
The changes in plastic strain components {deP}, are

related to the change in effective plastic strain by the
flow law associated with the von Mises yield criterion:
{dEP} = diP{}
(229)
where the effective stress a is given by

-
2 )2 + (o-2 - a3)2 + (0-3 -
{-[(,Ri -
cr1)2
l/2
+ 6(o122
d2
=d
+da
e-e
E - E -el -e
[D]
[ 7
|' [DI~
).
}).
3 + 0.23 )]}1
(230)
(2
and
(222)
da/ao-J - La/l ao-1, aalaO-2, . .etc.].
(231)
518
D. BUSHNELL
A change in effective plastic strain di' is related to a

change in effective stress da by
da = EET
E-ET
For thin shells the stress and strain vectors given in

eqns (228) reduce to
[do-] - [do, dor 2, d- 12 ]
(232)
dip = Ht dip
[del - [de,, de 2, dcE

2]
(239)
and for small di, we can write

da = Lda&laoi {do}.
(233)
If we premultiply eqn (227) by [a6loa] and make appropriate use of eqns (229) and (232) and (233), we
obtain:
in which "I" denotes meridional direction, "2" denotes

circumferential direction, and "12" denotes in-plane
shear. For isotropic materials the 6 x 6 elasticity matrix
[D] in eqn (238) reduces to
[a&faoj [D] ({deI -
})'
dip {
235
(25)
d=Hp [ailauj [D4fe

] [D]e ai
(240)
(234)
We can solve eqn (234) for di',
a wtu
(229
f
Equation (240) can be derived from eqn (238) by

remembering that for shell theory the stress component
0`3 normal to the shell wall is zero. From the third row of
the [D] matrix in eqn (238) one can derive
E3
and with the use of eqn (229) we obtain

6)<6
(236)
{de'l = [C.] {de}
where
{' }[a&la,
[CO]
[D]l
a[
(237)
H + [ al ao] [D] {ail

The matrix CoT, used in eqn (223), is the transpose of
[CO]. In eqns (221)-(237) the quantity D or [D] is given
by
-(I
V)(E
e2).
(241)
With use of eqn (241) and recalling that the transverse

shear strains e 13
= E23
= 0, one can condense the matrix
in eqn (238) to that in eqn (240).

The 6 x 6 [CO] matrix given in eqn (237) for general
three-dimensional plasticity analysis becomes a 3 x 3
matrix in shell analysis since the stress vector in eqn
(239a) contains only three components. The effective
stress in eqn (230) simplifies to
a=
(ul2 + (22 - go,2 + 3o-122)1/2.
(242)
0
0
0
0
0
0
0
0
2
2v
1-2,
Discretization in the meridional direction is by the finite

difference method, as described previously. Figure 21
shows the arrangement of nodal points used for the
discretized shell analysis. As already stated, the energy
density is evaluated midway between adjacent "u" or
"v" nodes. The u, v, w, displacement components in the
neighborhood "E" 1/2 are expanded in a Taylor series
about the point "E" , where the independent variable is
the meridional arc length s. Thus for portions of the
structure being treated as thin shell segments, eqn (210)
might be written in the form
1 -2vj
u = E hi(s)qi
[D] (
(It + )(- 2v
VI
Laila] {da} = di = H'dE'
(l - V)
V
V
(I- V)
v
V
(I -v)
(238)
Equations for thin shells, discrete rings, solids of
revolution
The integration over the volume of the structure indicated in eqns (224) and (226) includes portions of the
structure modeled as thin shell segments, discrete rings
and solids of revolution. Each of these analytical models
of the actual structure has its own kinematic law.
Thin shells
Each shell segment may contain a number of layers,
each layer with its own orthotropic properties G, El, E2 ,
v12 and each with its own stress-strain curve. A
reference surface is selected; the strains anywhere in the
wall are expressed in terms of the strains and changes in
curvature of this reference surface as in eqns (82), and
numerical integration is carried out through the thickness
of each layer by Simpson's rule, with five integration
points being used through the thickness of each layer.
(243)
i-l
in which the domain over which eqn (243) holds is the

neighborhood "E" 1/2 and the qi are the 7 nodal displacement components wi-,, ui, v,, wi, ui. 1, vi,,, Wi+
1
indicated in Fig. 21 and eqn (115).
The hi(s) could be derived from consideration of the
linear variation of u and v and the quadratic variation of
w within each shell finite difference element as shown in
Fig. 36. However, the hi(s) are not needed, since there is
only one Gaussian integration point per element. Therefore, expressions such as given in eqns (110), (Ill), (113)
and (116) are sufficient.
Discrete rings
If warping is neglected, the strain energy associated
with a discrete ring arises from circumferential strain and
torsion. The integrals over the volume indicated in eqns
(224) and (226) have the form
f(x', y')(r, + x') dy' dx' dO
(244)

x.UI
in which the (x',y') coordinate system has its origin at

the ring shear center r. and the limits of x' and y' define
the shape of the ring cross section. The function f(x',y')
represents the integrand of eqns (224) or (226). The strain
vector contains two elements
leJ =
-lEhOOP,
EtOmtJ
519
(245)
and the elastic matrix [DI is given by
Il 0[DI = E,
2(1 +v)_
(246)
The 6 x 6 Co matrix given in eqn (237) for general threedimensional plasticity analysis becomes a 2x 2 matrix in
discrete ring analysis since we are only concerned with
the two stress components Whops at-i~t
Each discrete ring cross section can be modeled as if it
consists of an assemblage of K straight segments of
thickness Tk, length Lk, and orientation angle Ok, k =
1,2,...,K. Figure 45 shows an example. The material of
each ring segment may have a different stress-strain
curve and different creep properties. The temperature
may vary along the length Lk of the ring segments but
must be constant through the thickness Tk. The integrated ring properties are determined by Simpson's
rule integration within each segment for all K segments.
Further details are given in Refs. [26, 54].
Solids of revolution
Isoparametric finite elements [52] are used to represent

those portions of the hybrid body of revolution for which
shell theory might be inadequate. The element geometry
is given in Fig. 46. The independent variables are x, 6
and r and the corresponding displacement components
are UI, U2 , U3.
This element was chosen because it can adequately
predict shell behavior as long as all eight nodes are
present. Hence, it does not create spurious discontinuity
stresses in the neighborhood of a junction with a finite
difference shell element.
Within the element the displacement field Ul, u2, u3
and rx coordinates can be expressed in terms of the
nodal point values and interpolation formulas written in
(S.
Fig. 46. Isoparametric 8-node solid element of revolution used in

hybrid computer program (from Bushnell[35]).
terms of the local coordinates s, t which vary from - I
to 1:
(x, r) =
hi(s,t)(x', r')
=1
(247)
(u], U2, U2) = fI
hi(s, t)(u,', U2i, U3i)
in which the interpolation formulas hi(s,t) are given

along with the kinematic expressions and other details in
the following section. Integration over the x-r plane is
performed by Gaussian quadrature in which the numbers
of Gaussian points in the "s" and "t" directions are
input variables. The circumferential variations of displacement components are given by eqns (219).
DETAILS OF THE ANALYSIS OF A TWO-DIMENSIONALLY
DISCRETIZED (SOLID) REGION
In this section expressions will be derived for most of

the components of the energy functional which appear in
eqn (209).
Strain energy-general
The strain displacement relations valid for small

strains and moderately large rotations are given by
Novoshilov [55]:
e, I(U3+
=
n22
!
= -(
Fig. 45. Discrete ring as modeled in the hybrid program (from

Bushnell [251).
032)
ax
2
2
u 3 -nu2 )+2-(w
+w3 2 )
du=
3
NOTE: Discrete ring ottochment point is considered to be

located on the shell reference surface.
a3
r 2(l+ 0
au,
au3 at. +au,
613
6,2
=nuI/r+
WI(
623 = d2+
ax-X--''"'
ar rI n3-
(248)
U2 ) -
f02(03
in which the rotation components Aot, o., wo

3, shown in
520
D. BUSHNELL
In terms of s and t the infinitesimal volume element dV

in eqn (209) is
Fig. 46, are given by

2(j,= I(nU3-
U2) -
dr2
dV = JI rdsdtdd.
a
au
au 3
ar
(249)
ax
2(o3 =ad2 _ nu, /r.
ax
Strain energy: nonlinear axisymmetric prebuckling

analysis
The kinematic relations for axisymmetric deformations
including moderate rotations are obtained by simplification of eqns (248) and (249):
In writing eqns (248) and (249) it has been assumed that
au,+ I
ax
EI
u, = u,(x, r) sin nd
U2 = U2 (x,
(250)
r)cos nd
-2
U3 = u3 (x, r)sin nO.
and r, x coordinates can be expressed in terms of the

nodal point values and interpolation formulas written in
terms of the local coordinates s, t which vary from -I
to I as in eqns (247). In Ref. [42] the interpolation
formulas h,(s,t) are given by:
h2 = RS14
- h5/2
(256)
W2
au3
a
du,
ar
ax
in which the rotation component w0is given by
au, au
ar
(257)
ax
If we let
Delete if node I is not included

- h1/2
uilr
=-
Within the element the displacement field u,, U2 , U3
h, = RS14
102
au.,+ I
e3
The kinematic relations (248) and (249) are consistent

with those used for the shell analysis, eqns (90) and (91).
1=5
(255)
1=6
1=8
1=7
ar
as
dx
d;
ds
a II
- h8 /2
a 12 =;
a2,=d-;
ax
at'
a-2 =d
ar
at2 (258)
- h6 12
h 3= RS/4
- h6/2
then, using eqn (253), we obtain
- h,/2
- h7/2
h 4 = RS14
I/aud,
- h812
d,
el = ' (a2 "as-a
h 5 =R*S/2
h 6= RS*12
t2J)+
22
e 2 = u 3 1r
(259)
h, = R*S/2
I/
(251)
h8 = RS*/2
3=
in which
E13 =
R=l+t
R=l-st
au3\
du
aJi-d
12
at J-2'2
I aidu, au,
du
J (- a2' d + a,,adt+ a22 all3 -a
F7
R*=It2
ati7
aul\
0'
I' dt,)
at
(252)
S =I+
= I-
S* = I-
and
and the numbered subscript refers to node point number

in the 8-node isoparametric quadrilateral of revolution.
The nodes are numbered as shown in Fig. 46.
The kinematic relations (248) and (249) contain derivatives with respect to x and r. These can be expressed in
terms of derivatives with respect to s and t by means of
at
- ar
as
- ax
at
ax
as
ar
ar
a
as
a
at
au,
2W2=
22lJi
au
au \
3
3
,-a0p
Idu.
(260)
-- a2ds
+ a'dtJ(6)
With use of eqn (247b), we can write eqns (259) in the

form
{e = L
e3
= [B]{q}l
+i {
(261)
au ]
(262)
(253)
E 13
in which
where
Iqi
ija =ax ar - ar ax
as at ds at'
du,
I(_aIau~aau
ads
+adt
(254)
Iu',u,
u 2 uu 3
The superscripts in eqn (262) denote nodal point numbers
521
as shown in Fig. 46. The terms in the 4 x 16 matrix [B]
Equation (268) is analogous to eqn (68) in the curved
are
beam analysis.
B(l, 21- 1) =I
a22 As
_ a12 ;hr )
B(1,2I) =0
B(2,21) = hlr
B(2,21- 1)= 0
(263)
l(-
(3i2
B(3,21 - 1) = 0
B(4,21 - 1) = 1(- a2' dA + al t

VI\_
ds
at
in which I is the local node point number, I = 1,2,

The rotation (02 may be written in the form
B(4,21) = I
... 8.
2w2 = [RJ{q}
(264)
as
(a22
a0
112
as
Jj\
at
a1 )
at.
Equation (226) can be written in the form

au
aU~
e
2 7r7,
Ask I f Atk [ (JAE] [DTro]4 [oo]MfBi)
in which
R(21 - 1) = B(4,21 - 1) R(21)
The hi in eqns (263) are given by eqns (251).

Using eqns (255), (261) and (264), we can writ e eqn
(224) in terms of the nodal point displacement components. The first and third terms in eqn (2244) are
expressed in the combined form
([A.EJ [Drol +
[au]) t aq dq
}28
(266)
where
LAeJ
Le-o
loo = [[E. --
E -C
eT j
[D]
(267)
in order to save computer storage space and to i-educe

the number of calculations required to obtai n the
stiffness matrix. Equation (224) thus becomes
j) IJ|r j
+ [Doi)
+(IADI
(265)
B(4,21)
(269)
In the axisymmetric prebuckling analysis the terms in

eqn (268) contribute to the l.h.s. of eqn (213) and those in
eqn (269) contribute to the r.h.s. of eqn (213). Equation
(268) represents a contribution to the (i,j)th term of the
local stiffness matrix for a single solid element of revolution. The entire local stiffness matrix is obtained by
variation of i and j to cover all 16 nodal degrees of
freedom associated with that element. The stiffness
matrix of the entire solid axisymmetric structural segment is obtained by assembly of other local matrices
derived for other solid elements. Equation (269)
represents the negative of an analogous contribution to
the "loading" vector.
Strain energy: nonsymmetric analysis
a U
Ks
l =2 2KSI
AS,
q,3qd~
k,=}
.I4
K,
4x4
Ix
(Le[DT..]
LJA
+
xiEA
4)
~l(I)
Fzijj
R2R1
+
LBj1 [DToltBI}
Linear nonsymmetric stress analysis

The linear solution can be thought of as the solution
obtained after one Newton-Raphson iteration with zero
used for the starting vector. Thus, eqn (213) holds and Aq
now represents the entire linear solution, not just a
correction vector. The strain displacement relations (248)
should be linearized. For each circumferential harmonic
these relations have the form
OL
LBj} [DToi
(268)
+ [w)2 R 1,0,w 2 Rj,OJ [DToI{Bi}
[w2Ri
+ L[ 2 R,08,wR-0IOJ [DTO] X .
1
6X1
6X24
{e} = [B]{q}
\
lJtrI
kt. k,
in which Ask, Atk are Gaussian integration weights; K,

and K, are the numbers of Gaussian integration points in
the s and t directions; RPis the ith member of the vector
JRi given in eqn (265); IBij is the transpose of the jth
column of [B]; and {B;} is the ith column of [B]. All
terms in eqn (268) are evaluated at s and t corresponding
to the Gaussian integration points (Sk,, tk,) in the element.
sin no
cos no
sin no
cos nO
sin no or cos no
sin no
cos no
cos no
sin no
cos nO
sin no
(270)
with [eJ given by eqn (228b) without the d's, and

Lq] =
u 2 U',ue,u
,U2 ,u
2
3
, ... ul
,u
8
2
,1u3 J .
(271)
Equation (224) becomes

a
= 7,
iiqi~q
K6,
s
Xt
Ask
k,-
Y
kil
Atk[[BjJ
[DI{Bi}J1Jr]k,,k,
272
(27)
D.
BUSHNULL
in which the subscripts []k,,k, signify "evaluated at the

Gaussian integration station (sk,,tk,)`.
In the bifurcation buckling and vibration analysis the

term AE in eqn (266) is zero because the stiffness matrix
is evaluated at q = qo. Therefore, f =,o in the first term
in eqn (266). The six-element veetor Laoj is given by
Equation (226) becomes

au
T2j7
Ks
7
KE
IT ,~
Kr
As, J At,
tk
k,
JJ-ft[{1,}1J
T]Ak,,,,
(273)
in which [D] is given by eqn (238). If n = 0 the factor ir

should be replaced by 2ir, The 6 x 24 matrix [B} is given
by
13(1,31-2)= 1 a22 411 -
(277)
[qO! - t~llO,0f2(10s3(i,0130,tfllO~ff230l]
in which the first four components are calculated from

eqn (267) and the last two components are zero if the
axisymmetric prebuckled state is torsionless.
dt i
B(1,31-1) = 0
B(1,31) = 0
B (2,31-2) = 0
B(2,31-I)= -nh 1/r
B(2.31)= hj~r
B(3.31-2) =0
I /
B(3,31-1)=0
Oh,
B(3,31)= I
dhr
a2 , dh-+ a" hJ
BO4X31 2)-= 1-a
2 1
j + 1aaI)
(274)
B(4,31 - 1)=0
B(4, 311)
(ah
22
B(5,31 -2)- nhlr
B(5, 31- 1 = 1allJ
B(6,31-2) = 0
R{6, 3, - 1) = 1J
a1 2, a
j aS12
_
dtj
atl IhL + all ANhi - hefr
B(6,31) = nhlr
As in the shell analysis (eqn 125) the strain vector can
in which I = 1,23,3.. .8, refers to the local node point
number- hT is given by eqn (251); and all, G2 , a2l, 4a2 be divided into three parts:
are given by eqn (258) with use of eqn (247a).
The deformations due to a general nonsymmetric load
(278)
[El = If]J + [cJl + [fC
are calculated by superposition of deformations of the
where superscripts('), i = 0,1,2 indicate zeroth, first and
form (250) or, if n is zero or negative, of the form
second order in the infinitesimal buckling or vibration
a1 =u,(xricosfO; U2=u2 (.Xr)sinlnte:
mode qb = IbU2b u3". The zeroth order represents the
U3 = U3(X, r) cos no
(275) axisymmetric prebuckled state. The first and second
corresponding to each harmonic of the Fourier series order terms, derived from eqns (248), are given by
representation of the load.
Equation (272) represents the contribution in the nth {f11} -1,
circumferential harmonic to the (i,j)th term of the local
stiffness matrix for a single solid element of revolution.
11o)
( du + 6o20(
sin no
The entire local stiffness matrix is obtained by variation
of i and j to cover all 24 nodal degrees of freedom
associated with that element. The stiffness matrix for the
E(1)
(u2
I2
el
3 br - nu2 kir) sin nO
entire axisymmetric solid region is obtained by assembly
of similar local matrices derived for other solid elements
(-,+4070w2b
l,@
. sinno
)SQ
3 +
in the two-dimensionally discretized structural segment.
Or
/
2
(L
f(1)
Nonsvrmetric vibrations and bifurcation buckling

The starting point for this derivation is eqn (224) with
eqn (266). The (i,j)th element of the stability or vibration
stiffness matrix is given by
(u
f(I)
E12
If 23
+ tBElaq'hj [DTO] {
1 r ds dt do) ,= 2..
j;j})
Or
a,axF)I sin no
(nulbIr +d-
O2
bb)cosnO
,)20,1
Cos
no
,
(279I)

1(2)
O2
h2 +
I(oj2 sin nd+w
hi'
Uo
Cos2 no)
(M
(2)
2(I+0)
(2
523
B2 (5,31- 2) = 0
B2(5,3I- 1)- hlfr + 9 (- 21h, + a,
t)
)cos' no
B2(5,3 n)= - nhtrr

(2)
(6 (2)) =
t(
E3
-to
1(2)
sin' no)
cos2 n H+ 192
I'
(UWI
COS
(280)
no
B2(6,31-2)= nhd/r
12adt
82(6,31- 1)=- 1Ua 2 -ll -lJI\ Os
at
t)2 sinn0cos no
Wj
C23
(284)
82(6,31) = 0.
sin nO cos n O.
to (03
Equations (279) and (280) are based on the assumption

that
U,
= u, 5 (x,r) sin nO
2b
U2b(x, r) cos nO
3=N3bix, r) sin nO.
After integration over d in closed form and numerical

integation with respect to s and t, we obtain eqn (276) in
the form
a( r-1T
0)
odq
j
KIN
'sVk9 1
(281)
Ask
I,
Xl
KI
At,
=l
{aW
b od'i. bI
+ 189~
WU Ow
93
Two of the prebuckling rotation components iD and wjo

are zero because the prebuckled state is axisymmetric
and torsionless. The prebuckling shear terms ai0o2
and
W, b aW bx
+
O3
100),S 00)35,
00)3
0 aqi
3W3oa5q
q
-a~o
I\20-w--m
O + q,
aLo are also zero. Hence, the circumferential coordinate
i can be separated from the other independent variables

s, t in the usual way, thereby leading to reduction of a
three-dimensional phenomenon to a numerically twodimensional analysis.f
It is clear that only Iel2>J will contribute terms to the
vector 1a2 E/Oqib q)j' and only [L(1)] will contribute
terms to the vectors [deldqtl} and {0e10q97j. As in the
case of the thin shell theory, we can write eqn (279) in
the form
6)524
sin nO
sin nO
sin no
sin nd
eos no
6x24
{.E< = ([Btl + i&20[B2J)jq I
(282)
in which [B1] is given by eqns (274) and [qbJ is

[=bl
blU iU
3bi
.1b
Ujj,U
U,
tU
) 2 08 2]T
[DTo
X<
[Bl+w 2nBJl{$Y}r)IJ1.~k
(285)
In eqn (2853l),b
w w, 1035 are given in eqns (249) with
superscript b added; O;so, a20, a30, a,30 are the stresses
associated with the x, 0, r, xr directions, respectively;
[DTh] iS the 6 x 6 constitutive matrix including effects of
plasticity (teqn 225); and ,, is the prebtickling rotation
component analogous (but with opposite sign) to the
meridional rotation,0 of thae shell Note that eqn(285) is
valid only if the terms DTois,
,
r DTO1 dito
25 , DTO26, Dv35,
DTro6, DTO4s,
DTO46 in the constitutive coefficient matrix
[DTol are zero. If these terms were nonzero, one would
not be able to separate variables as has been done.
The terms in eqn (285) multiplied by alo, a,.2 . ar30, and
aio can be expressed in the form
eos nO
[qbJ
+ |aqd/aq'j [B1 +
,U3
boiSSU~j.
(283)
The matrix [BR is derived from eqns (279), (249), (247)

and (253):
24x3
B2(1, 31-2)
32i~s~)=
+ aI, ah)
(ds
Mr
"2tn S
-h
atrz
B2(1,31-1) = 0
3x3 /qX24P
B2(2,3I-1) = 0;
B2(2,3I) = 0
B2(4,3I-2) = 0;
Vol. 18, No. 3-}
-a,1
(a) o + (o2
30
Orrio
(02)7
(287)
B2(3,3I-I)0=
and [RI occurs in the transformation from the rotation

vector
B(3,31) = B2(1,31)
CAS
0
(a10+
B2(3,3I-2)=B2(l,31-2);
(286)
in which [Iaol is given by
[(aw+a30)
B2(2,3I - 2) = 0;
Ldq Idlqjbj [RI T[io-olRIIt qi-I
B2(4,31- 1) = 0
2(4,33)
i )'] = t01i,.
(288)
54
D.
BUSHNELL
term of the stiffness matrix K ,ijis given by
to the nodal degree of freedom vector, [qb]:
K,
(289)
{o} = [RI {qb},
K lij
K,
tkjJ(SR7
E Ask,, E
k
I=I
The matrix [RI can be derived from eqns (249) with use
x (Q(B, + (oo'B 2),I [DToI{((B
of eqn (253). The derivatives of the nodal degree of

freedom vectors [aq"/aqjb| and [aq"1aqj'J are simply
+ [Rj] [O11p{Ri})k.tg
I 24
Laqblaqbb]
to,0,
o .... o,l,o,o
(290)
K2i=E
K,
ASkS
k,=1
[O
..
,,,0
,i,o,o
...
Ibx24
24[3
Laq`1aqi I
[RIT=
3x24
241
[R]
Ix3
[fiJ]
(292)
qF = {Rk}.
Kinetic energy
The kinetic energy in eqn (209) has the general form
T=l
124
24>6
fx6
+ o2 oB 2 ' T = I(B, + W2 oB2),J
aqib] [B,
24x
6x24
aq
+ W2 oB2]
T-T
(293)
{(B. +
qqo
sin no
= a0 o+-ib(x,r) or
eint
. cos no)
/ sin no
kI= I
u = ifl"(x,r)
woB 2 ).})IJIrI.,.-,.
(294)
In modal vibration analysis eqn (294) represents a contribution to the local tangent stiffness matrix for a solid
finite element of revolution that is loaded by stress
o10,(T2 0 ,
0130,
r1 30
(300)
Since the prestressed state is static,
K,
Kr
= 'Tr E Ask E Atk[(qRjJ[uO]{R,1
+ [(B, + 1 2oB2),J [DTo]{(B, +
components
in which m is the mass density and the total displacement vector a is the sum of the axisymmetric prestressed state ao and the infinitesimal vibration mode ub:
0 2B 2 )i,}.
(aulb
)
(299)
mu dV
6XI
Using the expression (286) and eqns (292) and (293) in

eqn (285), one obtains
Qj
(298)
3,,x
Similarly,
(-e
r(sg, t)
where superscript k indicates finite element number and

Sg, ts indicate the (s, t) coordinates of the Gaussian integration point. One can readily see the analogy between
eqn (297) and (298) for the two-dimensionally discretized
solid of revolution with eqns (136) and eqn (137) for the
thin shell. The local stiffness and load-geometric matrices
for the isoparametric quadrilateral 8-node elements of
revolution are dimensioned 24x24, whereas those for
finite-difference shell element are dimensioned 7 x 7.
so that
[B.
(291)
corresponds to
d.o.f. number "i"
q,
tt|(g
kj=1
x (Rkj| [Ao]{Ri})jR
I. 24
La qblaqibj
Laqh
+ (oo'B 2 )}
and the (i,j)th term of the load-geometric matrix K2,1 is

given by
K,
corresponds to
d.o.f. number "j"
(297)
Ir(s,
k,=I
and deformed by rotation
In bifurcation buckling analysis the loading is

divided into two parts, as described earlier (see eqn 122):
or
e'nt.
cos no
(301)
For the calculation of vibration modes and frequencies

one needs the matrix [22 Tlaqibaqba]; ij= 1,2,3,... N,
where N is the number of degrees of freedom in the
vibration problem. Recalling that a" is a linear function
of the nodal degrees of freedom qb, one can write
020.
a2T
aT__
_a
(295)
= p + AAp
-Nb
and the prebuckling rotation AW20 associated with the

load increment Ap is neglected. Assuming that the
"fixed" part pf gives rise to go', C020' and the increment
du b ab
aii
aqj
sin2 no
aqq, a
co
no d V
(302)
The displacement field in the quadrilateral isoparametric

finite element is given by eqn (247b). Therefore, eqn
(302) can be written in the form
Ap gives rise to Ago, A(020, one can write eqn (294) in the
form
ada
MU
=K
ij
= 7r(K,,j +
XK 2i)
(296)
in which k denotes finite element number. The (i,j)th
Q2-
f"J fmh,(s, t)hj(s, t) IJjr ds dt.
(303)
The integrals in eqn (303) are evaluated by Gaussian

quadrature. The local mass matrix is dimensioned 24 x
24.
Body forces
The term in the energy functional (209) related to body
forces is
Since U1, U2, u3 are functions of the nodal degrees of
freedom q, eqn (306) can be written in the form

F -dV=
SuidV =
LFba
Fb da dV~q.
525
V (307)
m(g,-+g2-+g3-u
(304)
If the displacement vector a is a linear function of the

nodal point degrees of freedom q, the body force term
generates contributions only to the first variation 4,l of
the energy functional. It is assumed here that body
forces are due to rigid body accelerations of the center
of mass, angular accelerations about the center of mass,
and angular rotation about the axis of revolution. Figure
47 shows a body of revolution with components of
translational acceleration a., ay, a, and rotational acceleration i,, 9yb,89z.In order to transform eqn (304) into
a form suitable for programming, one must have the local
after cancellation of Sq.

In general the displacement field U, U2,
Because g1,
can be
g2,
g3 contain only terms independent of 6
and terms which vary as sin 0 and cos 6, only the terms
in eqns (219) with n = 0 or n = I contribute to eqn
(307). After performing the integration with respect to 6,
one obtains for the r.h.s. of eqn (307)
For n = 0:
27r f.
2
~m( - a.
+.6)
-iroq--
aq
)rdxdr.
(308)
components of acceleration g1 , g2 , g3 corresponding to

U1, U2, U3 at a point at a radius r from the axis of
revolution and a distance d from the yz plane in which

the center of mass is located. These components, in
W are
terms of a., ay, a.s, W, wX,,,6,
U3
expanded in the trigonometric series as in eqn (219).
Forn- +1:
r [r6,Ld-(a, +d6,)(LUL dq L)]rdxdr
g=- a + r(6, sin 6- 6, cos 6)
(309)
g2 = r6. + (a. - dd,) sin d - (a, + da,) cos 6 (305)
Forn = -1:
g3 = rw,2 - (ay + d&o.) sin 6 - (a, - day) cos 6

JI, fJ'
Equation (304) can be written in the form
[-
r.
dui-'>
2
+(a,-dwy)("
dV = J(mg] Su + mg 2 8u 2 + mg 3 Su 3 ) d V.
iAu
(306)
~-d
)]rdxdr (310)
Thin shell segment body forces. The displacement

components a1 , U2, u3 are assumed to be constant
through the thickness. In terms of u, v, w (Fig. 44a), uI,
U2 , U3 are
a
J
,= u(rIR 2 )-
XXX
r
wr';
U2 = v
(311)
U3 = ur' + w(rIR2 ).
Furthermore u, v, w can be written in terms of the nodal
displacement quantities
]
Lq] = Lwj 1,ujvIwjuj+1, v+vw+
1
(312)
by means of eqns (114). The double integration indicated

in Expressions (308)-(310) is replaced by single integration along the shell meridian, and the quantity m is
interpreted as the mass per area of shell reference surface.
Solid of revolution body forces. The displacement
components U1, U2, U3 are written in terms of the nodal
degrees of freedom by eqn (247b). If the r.h.s. of eqn

(247) is used in eqns (308)-(310) and the (xr) integration
is transformed into (s,t) integration according to eqn
(255), the contributions of the body forces to the loading
vector of the discretized problem are obtained.
Surface tractions on solid of revolution
The term in the energy functional (209) related to
constant-directional surface traction is
Fig. 47. Local accelerations g1, g2, g3 due to rigid body translational (a., a,, a,) and angular (6., ay, fi) accelerations and
angular velocity wAof and about the center of mass.
JfsraiU
fArea
dA
or
Jfs
A
-"dAq.
q
(313)
D. BUSHNELL
As with body forces, this term generates only contributions to Ti.
The three components of external surface load p,, p,,
pn, acting on a solid region (one element in this example)
are shown in Fig. 48. These positive values form a
right-handed system.
The work done by the surface tractions acting over an
elemental length dl (shown in Fig. 48) is
dW = (p,dlu, + pdlu2 +pdlu) rdM
(314)
in which the normal and tangential displacement components un and u, are given by
dr
dI
dx
dl
(315)
dx
dr
dl +U dl
Fig. 48. Surface tractions on isoparametric quadrilateral finite

element of revolution. pt, p,, pl form a right-handed system.
The elemental work done is therefore

dW = [p,(u, dx + uadr)+pdlu
2,+p,(- uIdr+ u3 dx)]rdo
(316)
The variables that appear in eqn (316) can be written in
terms of the local coordinates (s,t) of the element by
means of eqns (247) and the relationships
dr=
dt; dx = d ds + dt.
as ds + at'
as
at
(317)
The finite element depicted in Fig. 48 has four faces,

labeled 0, (2), (0, I. On each of the faces the work
done by the surface tractions can be shown to be given
by:
On Face,: s t 1.0, t varies
W(D =Ir I
rkAtk [
hi(+
+ p, u2,(r,,
p,(UI X,, + u3 r,,)
ftk){
(318)
+ x,,2)1 + p.(uir, - U3Xt)}k].
On Face (2): s varies, t
In eqns (318)-(321) the factor ir arises from integration

of cos 2 nO or sin 2 nO from 0 to 27r. If n = 0 this factor
should be 2ir. The quantities At, and Ask are the Gaussian integration weights and Ks, K, are the numbers of
integration points in the s and t directions, respectively.
Subscript k here denotes a quantity evaluated at the kth
Gaussian integration station. Subscripts (),, and (), t
denote differentiation with respect to the local element
coordinates s and t.
The contributions to the loading vector are obtained
from eqns (318)-(321) by differentiation of WID through
W with respect to the nodal degrees of freedom.
Of course, in a solid region consisting of many finite
elements, only those elements with faces exposed to
external surface loads yield contributions to the global
loading vector.
Following pressure for solid of revolution
The relevant term in the energy functional (209) is
1.0
'=
Are.|
FI
.8 [f(R)] dA.
(322)
= IretP
AW
8
K,
W(- = 7r E
E h,(sk, - 1){
rk Sk
k=1
+ pAU
2 (r,,
On Face 3: s
W(j = Tr
p,(uix,,
u3 r,,)
i =l
2 1 12
+ x,, )
+ p.(uI r., - U3iX,)}k].
(319)
t varies
-1.0,
K,
hi(-
rkAtk
2
WIk){p
(UIIX,4 + u3 r,,)
+ pcu 2i(r,,7 + X,, )Pn + pn( - u ir,, + U3'X,,)}k].
(320)
On Face (:
s varies, t
= r
-
+ pu
+ 1.0
=
Ks
rkASk[
2
2 i(r,,
The term arises from the fact that the direction of the
pressure may change as the structure deforms. As
derived in the discussion associated with eqns (169)(171), the so-called "live-load" terms in the energy functional are quadratic; they contribute to the stiffness and
load-geometric matrices.
The expressions for each face of the isoparametric
quadrilateral can be derived from the nonlinear terms in
eqn (169) with hR,= 0, with the meridional arc length
element ds replaced by dl (see Fig. 48), and with use of
the transformations (315), (247), and (317). For any of the
four faces it can be shown that after integration with
respect to 0, W' is given by
hi(Sk, + 1) {PI(UliX,
2
+ u3 ir,,)
ds + x,, dt) (-ul d +u~dx)
-p,(x,, ds + x,, dt)u 22
(pn ds +-p
Pe.r dt) r
+ x,, )'/ + p,( - ulir,, + U3 1Xe)}k].
(321)
x ul dx + u, dr) (- Udr+ u~dx)
(323)
527
Differentiating with respect to the nodal degrees of

freedom q, and qj, one obtains
au, dr + u~dx
a2W,
+ x,, dt)
dqiqjJfLce{PnXds
aq
Ji
f ftcl.
x (---+-d-)
dq,
a Tl
_(-a
dd
Oxqi
di
ud d/
aqi
p(x,,
' ds + x,,' dt)--aq
aqj
qj,
- (pn ds +pdr
dt)r E
[(dx
dU3 dr
U2I ,.tU)
aqiT + -dl
q dl)
au,dr+ au3 dx + {audx au3dr)

d
aqi dl aqj dl )aq;dl daqjdl
audxIl
x (audr
-
(324)
qi dl aqi dilJJ
Reference Surface
of Thin Shell
Segment
In analogy with eqns (318)-(221), the following conditions prevail on the four faces of the quadrilateral
element:
On Face (I): s = + 1.0, ds -0, t varies, dt is negative
dr/dl =
rI(r.t 2 +
X,1 )
dxldl = - x,,d(r,
Surface
Fig. 49. Junction between thin shell segment and two-dimensional

finite element region (from Bushnell [35]).
'
On Face (Z: s varies, ds is negative, t = - 1.0, dt =0

dr/dl = - r,(r, 2
xc 2)112; dx/dl = - x, 0/(r,, 2
+ X's2)/2
On Face 3: s = - 1.0, ds = 0, t varies, dt is positive

dridl
= + r t2/(r ,2 + x, 2)"12 ; dx/dl = x /(r
+ x 2)12
On Face 0: s varies, ds is positive, t = + 1.0, dt = 0

2
dr/dl = + r,,/(r,, 2 + xK 2)"2; dxldl = + x,,j(r,.2 +x,, 2 )11
.
The derivatives duldq and r,,, r,,, x,_, x,t can be obtained
from eqns (247). The integration in eqn (324) is carried
out by Gaussian quadrature as with eqns (318)-(321).
Constraint conditions for junctions between thin shells
and two-dimensionally discretized regions
The relevant terms in the energy functional (209) are
, 5A,
SU"
alkuk
+aio) + Al
alkUk
These terms contribute to At (eqn 212) and

214):
i' =
7_
alkUk
+ ao) +
-
A,
lying along the junction line AB must be constrained to

remain on a straight line normal to the shell reference
surface as the structure deforms. This is accomplished
by imposition of appropriate constraints on differences
of un for compatibility and uniformity of the meridional
rotation 3 of AB and constraints on differences of u2 for
compatibility and uniformity of the rotation 4 of AB
about an axis through si, tj parallel to the meridional
direction of the shell reference surface. The length of AB
is free to change, of course, since there should not be
any constraint preventing strain normal to the plane of
the thin shell reference surface. Thus, typical constraint
conditions at a junction line AB are:
U*= u,(Si, ti); V - u2(s1, t1); w* = U3(Si, ti)

/3
(325)
atplaqi (eqn
(U
(uE
-_un
(u2n
u.
d
axl
X1
'. a
UqaqI
uk ,daA,
d
-
vt01,
)/L
74
3 4
)/Li4 ;
t=
tb =
(u 2
+ (2
(U
7
-u 2
u2
)/L
74
)/L
34
(328)
4
2
-U2i)Li4
in which
alk
a(
(326)
u-' = u1 i sin , + u3 i cos 4)

L, = [(xi
ad,.
un )L
al 'k
d .
aqi
x )2 + (r, - r )2 ]1 /2
(329)
37)
Figure 49 shows a junction AB between a thin shell

segment and a two-dimensionally discretized region. At
the junction u*, v, w*, and 3 at the end of the shell
reference surface must be equal to u1 (sj,ti), u2(si,ti),
U3(S1 , t,) in the "solid" region and a rotation derived from
the differences of un. (Subscript "n" denotes "normal to
the junction line AB" and superscript "i" denotes "ith
nodal point on the junction line AB".) The coordinates si
and tj coincide with the end of the thin shell reference
surface on the line AB. In addition, all nodal points
/ = w'- u/R,
4 = (llr)dwlda - v/R2
Shell wall
rt(sota)
and superscript i denotes any node on the junction line

AB. For each such junction, therefore, there are 3 +
2(k - 1) constraint conditions, where k is the number of
nodal points on the junction line AB. The sign for 4 in
eqns (328) as well as the nodal indices in all of the
equations depend on which end of the shell segment and
which faces of the two-dimensional finite element are
involved in the junction.
58
D. BUSHNELL
SECTION 5
a -CONSTANT
(2 CJRvE)
LINEAR EQUATIONS FOR GENERAL SHELLS
N-
__
The purpose of this section is to summarize certain

aspects of the more widely used linear shell theories, to
explain where the differences originate, and to comment
on the significance of these differences to the engineer or
designer of shell structures. Much of the material here is
abstracted from Leissa's excellent survey given in chapter I of his monograph Vibration of Shells [56].
- - CONSTANT
(
CJRVE)
N
INTRODUCTION
Whereas in thin plate theory the differential equation

of motion is universally agreed upon, the same cannot be
said for thin shell theory. Differences arise from different
simplifying approximations and different points in a
derivation where a given approximation is introduced.
The more commonly referred to shell theories are those
by Donnell[57,58], Mushtari[59,60], Love [61,62],
Timoshenko[63], Reissner[64], Naghdi and Berry[65],
Vlasov[66,67], Sanders[68], Byrne[69], Flugge[70,71],
Goldenveizer[72], Lur'ye[73], and Novoshilov[74]. All
of these theories result from Love's "first approximation" and apply to shells of arbitrary curvature.
For comparisons of various thin shell theories the reader
is referred to the work of Leissa[56], Koiter[75],
Goldenveizer[76], Klosner and Levine[77], Naghdi and
Berry[65], Kraus[78], Naghdi[79], and Kalnins[80].
Fig. 50. Middle surface coordinates (from Leissa[561).
which is derived from the square of the length of an

infinitesimal arc
dF = T,,da + P,,3d = d-
lying in the surface. The square of dF is a scalar quantity

obtained from the dot product df df:
d dt
CONCEPTS FROM THE THEORY OF SURFACES NEEDED FOR SHELL

THEORY
The essential feature of thin shell theory is the complete characterization of stress and deformation
throughout the three-dimensional domain of the shell
wall by knowledge of the deformation of a reference
surface. The expressions for stress and deformation
throughout the domain therefore depend on parameters
of this surface, such as its original shape and the extent
to which it has been stretched and bent.
(332)
ds 2 =A 2 da2 + 2AB cos y dad13 + B2 dp2. (333)
The three coefficients of the first quadratic form (333) are

essential, for example, in the derivation of strain due to
stretching of the surface. The engineering strain components expressed in terms of surface coordinates are
found by comparing dS2 for the undeformed surface with
ds*2 for the deformed surface, with the location of the
deformed surface r* = *(a, 1) being expressed in terms
of that of the undeformed surface r plus displacement
components u, v, w in the directions of the base vectors
F., l3, I.:
Surface coordinates
In order to measure deformations of a surface, we
must attach a two-dimensional coordinate system to it.
Figure 50 shows such a coordinate system and its relationship to a three-dimensional system fixed in space.
Any point on the undeformed surface may be located by
a vector
(330)
Tr= (a,3)
in which a and ,8 are independent coordinates of the

surface. A system of unit vectors T., I,3, f. parallel and
normal to lines of constant A = go and constant a = aO is
depicted in Fig. 50. These so-called "base" vectors are
given by
T = r,cJA,
f, =3
B;
r,,
(f, X ,3)lsin y
(331)
in which (),, and (),, indicate differentiation with respect to a and 1; A=-I, I; B -_ ,,I; and y is the angle
between the surface coordinate lines.
First fundamental form
Three of the surface parameters needed for characterization of surface deformations are the three
coefficients of the first fundamental quadratic form,
F* = F + Uf
(334)
+ VE13+ WT-
It can be shown [74] that for orthogonal coordinate lines

these reference surface strains are given for linear theory
by
I au
qAda
v dA
AB da
u aB
I ay
AB aa
_B'ap
e,Bd+
Aed u \
Bda+
w
RZ
w
B+
BaI
_R,
da B
(335)
in which e, and e,3 are the strains in the i'. and 1,

directions, respectively, and ea is the in-plane shearing
strain, a measure of the change in angle between the a
and 1 coordinate lines. All shell theoreticians agree on
eqns (335).
There are differences of opinion, however, as to the
expressions for the change in curvature and twist of the
reference surface as it deforms. A derivation of these
expressions requires what is called the second fundamental form.
529
Second fundamental form
The second fundamental form has to do with the

curvature and torsion of the surface coordinate lines.
Leissa[56] presents a derivation of an expression for the
normal curvature IIR of any line element lying in the
reference surface
2
Lda +2Mdad3 + Ndf
A2da 2 + 2AB cos ydadp + B2df 2
(336)
dz
h
2
14
IGo
in which the numerator on the r.h.s. is the second fundamental form with coefficients L, M, and N given by
the dot products
L = T. zIn; M =
3 *I;
iM N = r as] tI. (337)
The normal curvatures of the a curves and A3curves are

obtained by setting either df3 or da in eqn (336) equal to
zero:
l
-=
- NIB 2 .
(338)
By "normal curvature" is meant the curvature of the line

formed by intersection of the surface with a plane normal to it at the point (a, 3).
In the comparisons of shell theories given by
Leissa[56] it is assumed that the a and 3 coordinate
lines are lines of principal curvature of the undeformed
surface, that is, they are characterized by cos y = 0 and
M =0.
A shell: a surface with finite thickness
The shell theories developed in Refs. [57-74] are formulated considering an element such as shown in Fig. 51.
An infinitesimal slice of thickness dz located a constant
distance z above the reference surface has the following
geometrical properties:
Lengths of Edges:
ds,'z' = A(l + z/R 0.)da
(339)
ds(z) = B(l + z/R,)dp.

Areas of Edge Faces:
dA(z = dsjz)dz;
dAz)= ds('dz
(340)
Volume:
dV('z) = AB(I + z/R.) (I + zlRp)dadocdz.
Fig. 51. Notation and positive directions of stress in shell coordinates (from Leissa[56]).
(341)
Inclusion or neglect of the terms zIR, and zIR, in eqns

(339)-(341) gives rise to many of the differences in the
various shell theories. Differences also arise between two
theories both of which include these zIR effects initially
but which use different simplifications involving neglect
of zIR compared to unity later in their derivations.
LOVE'S FIRST APPROXIMATION
Love[61] made the following approximations in his

classical linear theory of thin shells:
(1) The thickness of the shell is small compared with
the smallest radius of curvature of its reference surface.
(2) Strains and displacements are small. Hence secondorder terms in the strain-displacement relations may be
neglected in comparison with first-order terms.
(3) The transverse normal stress is small compared

with the other normal stress components and may be
neglected.
(4) Normals to the undeformed middle surface remain
straight and normal to the deformed middle surface and
do not change in length.
These four assumptions constitute what Love called his
"first approximation" shell theory. The approximations
are almost universally accepted in the derivation of
linear thin shell theories.
The first assumption gives the raison d'&tre for a
discipline called "shell theory"; the second justifies
linearization of the theory; the third restricts applications
of shell theory to situations in which rates of change of
phenomena and geometry with respect to surface coordinates a and p3have characteristic lengths that are large
compared to the shell thickness; the fourth permits the
reduction of a fundamentally three-dimensional problem
to one or two dimensions and is equivalent to neglect of
transverse shearing strains. The fourth assumption,
known as Kirchhoff's hypothesis, restricts the applications
of shell theory in the same way as the third.
Several authors, including Leissa[56] point out the
inconsistencies in the four assumptions with Hooke's
law. For example, if the normal literally could not change
in length at all, then for an isotropic material a considerable normal stress o, = v(o, so)
+
would be generated
by uniform biaxial stretching of a plate. This normal
stress, arising from the Poisson effect, would not be
negligible compared to the in-plane stress. Conversely, if
the normal stress is small and the in-plane stress is not,
then the normal must change length. Similarly, from
Hooke's law, the Kirchhoff postulate implies zero transverse shear stress. However, the shell element cannot in
general remain in equilibrium without transverse shear
force resultants (which are the integrals of the transverse
shear stresses over the wall thickness) acting along its
edges.
These inconsistencies do not, of course, seriously
diminish the value of shell theory as an engineering tool.
Emphasis on them represents an unfair misinterpretation
of the mathematical model. The inconsistencies can be
deemphasized by introduction of the following two approximations to replace the third and fourth above:
(3a) The work done by the maximum normal stresses
530
D. BUSHNELL
acting through a distance equal to the maximum change

in length of the normal and the work done by the
maximum transverse shear stress acting through a distance equal to the maximum transverse shear strain
times the thickness are negligible compared to the total
change in strain energy during deformation.
(4) The displacements in planes parallel to the
reference surface may be calculated as if the normal to
the undeformed reference surface remains straight and
normal and unextended during deformation of this surface.
The apparent inconsistencies can also be deemphasized (as they deserve to be-at least by
engineers) by introduction of a corollary rule stating that
the average normal stresses and transverse shear stress
resultants must be calculated from considerations of
equilibrium rather than directly from the kinematics
embodied in the Kirchhoff hypothesis.
As a result of the Kirchhoff hypothesis, the displacements U, V, W anywhere along the normal to the reference
surface can be calculated from
Differences in the kinematic relations for reference surface deformation

As stated above, all shell theoreticians agree on the
expressions (335) for the strains of the reference surface.
However, there are differences in the various theories
for change in curvature Ka, K, and twist K., These
expressions are listed in Tables 5 and 6.
Change in curvature K,,K1 3 . Except for the expressions
by Donnell and Mushtari, there is general agreement
among the various theories concerning curvature
changes K. and K,,. The expressions of Vlasov in Table 5
differ from those of Byrne, FRigge, Goldenveizer, etc.
only in terms of order eatRe, ep3 IRB
3, this small difference
arising from replacement of 1/(1 + Z/RO) and 11(1 + zIR O)
by their series expansions.
The Donnell-Mushtari expressions differ in a more
fundamental way from the others in that they are
obtained by neglect of terms containing tangential displacements u and v.A simple example will show that for
certain commonly occurring cases of great engineering
significance the Donnell-Mushtari expressions are not
sufficiently accurate. Suppose that we consider inextensional deformations of an infinitely long cylindrical shell
of radius R. For a cylinder the surface coordinate a can
be identified with the axial coordinate x and the surface
coordinate p with the circumferential coordinate 0. Then
U(a,13,z) = u(a,3)+ zO(aP3)

V(a,3,z)= v(a,13) + z%(a,f3)
(342)
W(a,f3,z)= w(a,3)
A = Ir|
1oa
.0;
R&
o;
in which Leissa[56] gives

0
= u1
W'o;
Rn
w 3.
W,
= R
R,
B = If ,j3 = Irk| = R.
The displacement components, u, v, w are the axial,

circumferential, and normal (outward) displacements of a
(343)
Table 5. Change in curvature of the middle surface

Theory
Byrne, Flugge, Goldensetzer,
Lur'ye, Novozhilov, Love,
Timoshenko, Reissner,
Naghdi, Berry, Sanders
I d6.+ 63 3A
A da
AB do
vIasv.
I 86
A da
6:aA
AB8MO
I8/
Donnell, Nfushtari
I lds.
oAAa
a.w
I d6+
6aaB
3 MO AB as
IA a'"
A de 138,
I
00s 6,8B
1383 AB1 a
)A o,,8,w
1 1d B
R.
B ad
Terms given for the V'lasov theory correspond only to the linear is
8I ,
1a33
B
W
Rdd)
Bd.8
aI
A B a. ad
I) terms of table
Table 6.Change in twist Ka3of the middle surface

A 8
a(
'+B 8 (O6 / I j 1I_
B
A) A 8A\B dRaB a8
Byrne, FlUgge, Lur'ye, Goldenvei.er,

Novozhilov, Tisonhenko, Love
~~A
8 /e \
I
Reissner, Berry, Naghdi
vlasovinaSanders
1)
A adu)
B833\A
A d/aX
B Md\AO
MushtarmDgnvenl for tar

Terms gis'en for the Vlasov theory corsodonly to the linear
A 8a\B/
B
A di\BJ
)
B a "I\
A da\B/
a (I
r LB5L( I
_ au 8A\
AB a. R# A da AB ad
B d 8\
B 83 A/
R R
(344)
RO=R
A]a__B_ 8
1 / 1 I
\_(dBvr
2AB1Rg R./\ da
a.)
I)MO
Ad/
a.
Btms
of (abl
n~ 1) terms of table
Ad/a / ao
_3)
d:
dAS a,A.
da.
point on the cylindrical reference surface. Suppose that

for the infinite cylinder we have displacements u =0
v = vn cos no, and w = w. sin no, in which v. and w. do
not depend on x. From eqns (335)
e,=e,=u'=O; e,3=e9=
=-
nv + w )sin nO
e.,, = e., =
U+ V' = 0.
Total strains
(345)
eo =(~nvn +w.) sin nO = 0
(346)
which leads to
(347)
V.= w.]In.
From Table 5 and eqns (343) the curvature change
expressions of Byrne, FlUgge, etc. yield

-W;
K13 =K, =
(348)
(-W).
With v = v. cos no, w - w. sin nO and no x-dependence,
we have
K. = 0;
v(vn + wn 2) sin nO
K0 =
which, for inextensional deformation (Vn

K-= 0; K&
-nI)wn
(349)
w,.n/n) yields
sin nO.
'0; Kg=K= -
wW=
Table 7 shows differences in the expressions for total

strain at any point z in the wall thickness. The total
strain is always represented as the sum of stretching and
bending components. The expressions of Byrne et al. are
the most general and result from application of the
Kirchhoff hypothesis to the kinematic relationships of the
three-dimensional theory of elasticity. In the Love,
Timoshenko et al. theory zIR, and zIR, are everywhere
neglected compared to unity. The theory of Vlasov
represents a sort of middle ground between the Byrne et
al. and Love et al. formulations in that series expansions
are used for 1/(1 + z/R.), 1/(1 + z/Rl).
Force and moment resultants
Since the strains are known functions of the thickness

coordinate (Table 7) and, given Hooke's law, the stresses
are known functions of the strains, the forces acting on
the edges of the shell element shown in Fig. 51 can be
derived by integration of the stresses over the thickness
coordinate z. If the reference surface is chosen as the
middle surface and if we rigorously note the dimensions
(eqns 339-341) of the slice of thickness dz shown in Fig.
51, we can derive three force resultants acting on the
face perpendicular to the a coordinate
TN =
N,
5f
_ cr9
(1
dz
(350)
(352)
and three more force resultants acting on the face perpendicular to the / coordinate
The expressions of Donnell in Table 5 yield

K = KX,= -w
with regard to rigid body motions. Kadi[811 found the

same for the theories of Love, Timoshenko, and Vlasov,
but that the Donnell-Mushtari theory gives non-zero
curvature changes and twist due to rigid body translations.
DIFFERENCES IN RELATIONS INVOLVING STRESSES AND STRAINS
THROUGH THE THICKNESS
For inextensional deformations
KQ = K=
531
sin
Wn
no.
(351)
For n = 2 the expression for
K0 is 33% in error. This

error occurs when, for example, one uses Donnell theory
to calculate buckling loads of long cylindrical shells
under external pressure, for which the buckling mode

corresponds to n = 2. The buckling modal displacements
correspond to nearly inextensional hoop strain, so that
the relationship (347) holds with good accuracy. For
shells which buckle with higher values of n, the DonnellMushtari theory is more accurate.
Twist, K,,3. Table 6 shows comparisons of the expressions for twist K.0 of the reference surface. The Vlasov
and the Donnell-Mushtari expressions differ from the
others for the same reasons given in the discussion of K.
and Kp. The Reissner, Berry, Naghdi expression differs
because of neglect of zIR., z/Ru compared to unity at an
earlier stage in the derivation than in the Byrne, Fligge et
al. formulation. Sanders' expression is derived through
correction of that of Reissner et al. by addition of the
term with the factor (I/R,3 - h/R.) to eliminate non-zero
KQp arising from rigid body rotation. Kraus[781 demonstrates that the kinematic relations of Byrne, Flfgge,
Goldenveizer, Lur'ye, and Novoshilov are consistent
N03 l =
lQ0 J
t2
aU 1(
-hI2 l ;3 J
(353)
The positive directions of the force resultants are shown

in Fig. 52. These forces act at the reference surface and
have units of force/length.
dNa
'
3a
Fig. 52. Notation and positive directions of force resultants in

shell coordinates.
532
D. BUSHNELL
Table 7. Total strains at any point in a shell

E', CC.
Theory
e0 + ZKj1
(+/
Byrne,
FlUgge, a
(1zR
Goldenveizer,
Lur'ye,
j+
e
(1+z/R
Love, Timoshenko,
Reissner, Naghdi,
Berry, Sanders
Donnell,
Mushtari
e,
+ ZKl
2R
2R
ZK5
KanZ
ea +
Generalized
(+/R+
(1 +z/R)(l+z/
Novoshilov
ea
Vlasov
KaKn
1
nn=
eS+
n1lK
nZ
Similar expressions for moment resultants can be

derived
t MQ
I = fw2 ta0
I ( I + R -)
z dZ
(354)
tM} I=fh/t
5'/2
B,
01.
R
(1+RQ)zdZ
which are shown in Fig. 53 and have dimensions

moment/length. Note that even though oa3 = ua from
the symmetry of the stress tensor, the same does not
hold for stress and moment resultants: Np $ Np, and
MP $ M#, unless R. = Ra because the areas over which
the stresses a3uaQ. act are different on the different
edges of the shell element shown in Fig. 51.
Tables 8 and 9 show expressions for the stress and
moment resultants in terms of middle surface strains and
changes in curvature for an isotropic homogeneous shell
wall. The theories of Love, Timoshenko, Reissner,
Naghdi, Berry, Sanders, Mushtari, and Donnell are
arrived at by indiscriminately neglecting zIR. and zIR&
compared to unity. Novoshilov and Goldenveizer obtain
resultants by taking variations of the strain energy functional and discarding selected terms. Byrne, Fligge, and
Lur'ye simplify the z-integration by using series expansions for the quotients 11(1 + z/R.), 1/(1 + z/R,). Vlasov
follows a similar procedure.
WHICH THEORY IS BEST?
In modern computerized structural analysis in which

energy methods are almost universally applied to
engineering shell problems it is most often advisable to
avoid the use of Donnell-Mushtari theory or any theory
in which rigid body motion generates finite reference
Fig. 53. Notation and positive directions of moment resultants in

shell coordinates.
surface strains or changes in curvature. The DonnellMushtari theory might still be applied profitably in computer-oriented optimization analyses for preliminary
design. Such analyses usually involve sequential solution
of many structural problems. The Donnell-Mushtari
theory is computationally efficient because it permits the
use of fewer unknowns in equilibrium and eigenvalue
formulations. However, the analyst should be aware of
the limitations illustrated by the above example of inextensional bending of a cylinder. The Donnell theory is
accurate enough if the wavelength of the deformation
pattern is small compared to a typical radius of curvature
of the shell.
The differences attributable to retention of zIR., zIR,3
are of little importance for most engineering problems,
and it is best to choose the simplest theories in this

regard. In fact, retention of the zIR terms can lead to
results that puzzle computer program users and cause
them to distrust the programs they use. Two good
examples have arisen in the writer's experience
which caused him to remove terms involving zIR compared to unity in the BOSOR4 [18] and BOSOR5 [25]
computer programs, which are now based on Sanders'
equations. One example is a rather thick (Rlt = 10) isotropic hemispherical shell clamped at the equator and
uniformly heated. Far away from the clamped boundary
the stresses should be essentially zero. However, in the
original versions of BOSOR4 and BOSOR5, which were
based on total strain relations of the Byrne, Flugge, type
in Table 7, the stresses did not die away but instead
approached the values
533
At the extreme fibers, z = + t12, the stress a-(z) in eqn

(355) is therefore given by the spurious values
o( t12) = + -
aAT(t/2R).
I- V
(357)
If E = 10' psi, v = 0.3, a = 10-5, AT = 300, tIR = 0.1 the

maximum stresses are about - 2000 psi at the outer fiber
and + 2000 psi at the inner fiber, values large enough to
stimulate a program user to telephone the program
developer. This error arises because actually the radial
displacement due to uniform heating is not uniform
throughout the thickness as the theory implies, but increases linearly with z as one moves radially outward
from the reference surface. This linear variation of w
with z gives rise to I + zIR terms in the numerators of
1+vAT the first two terms in eqn (355) which cancel the like
E r ea+ ZKa + e0 +ZKterms in the denominators, resulting in a correct predic(l+ zIR )
(l + zIR) vaJ.
(355) tion of zero stress far from the clamped edge of the
uniformly heated hemispherical shell. However, the
same correct result is obtained simply by neglecting
Far away from the clamped edge deformation of the
uniformly heated hemisphere consists of a sum of a the zIR terms in eqn (355) and simultaneously ignoring
the true nature of the linearly varying radial displaceuniform radial (normal to surface) expansion plus a rigid
ment, as the Kirchhoff hypothesis requires. This example
body axial displacement. The rigid body displacement
demonstrates that it is inconsistent to include zIR terms
does not give rise to any strains or changes in curvature.
From the first row of Table 5, the Byrne, Flfigge et al. compared to unity, as in the first row of Table 7, while
relations for K. and Ki yield zero curvature change neglecting the effect of extension of the normal to the
reference surface.
corresponding to uniform radial expansion. The strains
The second example in which "small" zIR terms
ea and ep for uniform radial expansion are given, from
proved troublesome in BOSOR4 involved the axisymeqns (335a,b), by
metric shell structure shown in Fig. 54(a) subjected to an
eQ = en = wIR = aAT.
(356) axial load V. The shell is a wheel rim. Half of the wheel
Table 8. Force resultants according to the various theoriesa

Theory
(1 -v')Ne/Eh
(I1-')Np/Eh
2(1 +v)No/Eh
Byrne, Flugge, Lur'ye
2(1 +0No./Eh
.. h'f I
h'I 1
'/
12IR.
'
I/Ih
+ kX-
zoo-
Re)(
Goldenveizer, Novozhilov
'I
Rp)
ie)2
Ro)
-R-)(Xi-i-)
V~
'i12R '
h05+R2R
Love, Timoshenko, Reissner,

Berry, Naghdi, Mushtari,
Donnell, Sanders
Vlasov
aNote: T =
Kg.
Table 9. Moment resultants according to the various theories'

Theory
Byrne, FlAgge, Lur'ye
Goldenveizer, Novozhilov,
Love, Timoshenko,
Iteissner, Naghdi, Berry,
Mushtari, Donnell,
Sanders
I\
na+r'
Same as Flugge,
Byrne, Lur'ye
Vlasov
aNote:
12(1 -")Me/EhV
7=
K,.5
12(1 - v)M 1 /EV

-
24(10+)M.o/EhA
24(1+0)Mo.1EhS
i\
'8+'.
Same as Byrne,
Flugge, Lur'yerR
+R-
534
D.
BUSHNELL
6.00
5.00
.0I
ccI
4.00
4Symietry
Plane
t0.
'
3. 00
2.00
175-
BOO4 l.9
MODEL
1 .00
0.00
i
Radius = 8.65
(a)
I
7.00
8.00
9.00
RADIAL COORDINATE, r
Fig. 54. Wheel rim modeled as shell with 10 segments: (a) dimensions; (b) model for analysis with BOSOR4; (c)
deformation due to axial load V.
I
0
-I
-2
E
c
2
I:
0
Io
-I
MERIDIONAL ARC LANGTH -
Fig. 55. Inner and outer fiber stresses along meridian of wheel rim predicted from theories including and neglecting
-IR compared to unity (see eqn (339).
rim is modeled as a shell with 10 segments as shown in

Fig. 54(b). Some of these segments have small ratios of
meridional curvature R. to thickness t. For example, in
SegmentS R.It = 1.29. In Segments 0,3,(!(D,, andS
Rift = 2.14, 2.10, 2.42, 3.36, and 3.36, respectively. Use of
thin shell theory for prediction of stresses in these segments is questionable. Figure 54(c) shows how the wheel
rim deforms under uniform axial load V.The extreme fiber
stresses predicted in computer runs with and without the
zIR terms are plotted in Fig. 55. The discontinuities in the
''zIR included" curves are due to large discontinuities of
zJR at segment boundaries. These discontinuities stimulated a program user to call the program developer, who
decided to eliminate the offending zIR terms from the
BOSOR4 and BOSOR5 computer programs permanently.
REFERENCES
1. G. A. Cohen, User document for computer programs for
ring-stiffened shells of revolution, NASA CR-2086, 1973;
Computer analysis of ring-stiffened shells of revolution,
NASA CR-2085, 1973; Computer program for analysis of
imperfection sensitivity of ring-stiffened shells of revolution," NASA CR-1801, 1971; National Aeronautics and
Space Administration, Washington, D.C.
2. A. Kalnins, Users Manual for KSHEL Computer Programs
(1970), available from Dr. Kalnins, Lehigh University, Pa.
Also see Free vibration, static and stability analysis of thin
elastic shells of revolution, AFFDL-TR-68-144, March 1%9,
Wright-Patterson Air Force Base, Ohio.
3. V. Svalbonas, Numerical analysis of stiffened shells of revolution, NASA CR-2273, 1973, National Aeronautics and
Space Administration. Washington, D.C.

4. David Bushnell, Large deflection elastic-plastic creep analysis of axisymmetric shells. Numerical Solution of Nonlinear
StructuralProblems, ASME AMD 6, 103-138 (1973).
5. J. A. Stricklin, W. E. Haisler and W. A. von Riesemann,
Formulation, computation, and solution procedures for
material and/or geometric nonlinear structural analysis by
the finite element method, SC-CR-72 3102, Sandia Laboratories, Albuquerque, New Mexico, July 1972.
6. R. M. Jones, Mechanics of Composite Materials. McGrawHill, New York (1975).
7. J. E. Ashton, J. C. Halpin and P. H. Petit, Primer on
Composite Materials, Technomic, Connecticut (1969).
8. J. E. Ashton and J. M. Whitney, Theory of Laminated Plates.
Progress in Materials Science Series, Vol. IV. Technomic,
Connecticut (1970).
9. A. E. H. Love, Phil. Trans., Roy. Soc. (Ser. A), 179 (1888).
10. V. V. Novoshilov, Foundations of the Nonlinear Theory of
Elasticity. Graylock Press (1953).
11. Sanders, J. Lyell, Jr., Nonlinear theories for thin shells.
Quart. Appl. Math. 21, 21-36 (1%3).
12. R. H. Gallagher, Analysis of plate and shell structures. Proc.
Symp. Appl. Finite Element Methods in Civil Engng, (Edited
by Rowan, William H., Jr. and Robert M. Hackett), Vanderbilt University and the American Society of Civil Engineers
(1969).
13. R. H. Gallagher, Applications of finite element analysis.
Advances in Computational Methods in Structural
Mechanics and Design (Proc. 2nd U.S.-Japan Seminar on
Matrix Methods of Structural Analysis and Design), (Edited
by J. T. Oden, R. W. Clough and Y. Yamamoto) University
of Alabama Press, Huntsville, Alabama, pp. 641-678 (1972).
14. R. H. Gallagher, Geometrically nonlinear finite element analysis. Specialty Conference on the Finite Element Method in Civil
Engineering. McGill University (1972).
15. J. L. Brombolich and P. L. Gould, Finite element analysis of
shells of revolution by minimization of the potential energy
functional. Proc. Symp. on Appl. of Finite Element Methods in
Civil Engng (Edited by W. H. Rowan, Jr. and R. M. Hackett)
Vanderbilt University and the ASCE, pp. 279-307 (1969).
16. E. L. Wilson, R. L. Taylor, W. P. Doherty and J. Ghaboussi,
Incompatible displacement models. Numerical and Computer
Methods in Structural Mechanics (Edited by S. J. Fenves, N.
Perrone, A. R Robinson and W. C. Schnobrich), pp. 43-57.
Academic Press, New York (1973).
17. D. E. Johnson. A difference based variational method for
shells. Int. J. Solids Structures 6, 699-724 (1970).
18. D. Bushnell, Stress, stability, and vibration of complex branched shells of revolution. Comput. Structures 4, 399-435
(1974).
19. D. Bushnell, Finite-difference energy models versus finite
element models: two variational approaches in one computer
program. Numerical and Computer Methods in Structural
Mechanics (Edited by S. J. Fenves, N. Perrone, A.
R. Robinson and W. C. Schnobrich), pp. 291-336. Academic
Press, New York (1973).
20. M. Khojasteh-Bakht, Analysis of elastic-plastic shells of
revolution under axisymmetric loading by the finite element
method. Ph.D. Dissertation, Department of Civil Engineering, University of California, Berkeley, CA 1967. (Also published as SESM Report 67-8.)
21. M. Baruch and J. Singer Effect of eccentricity of stiffeners
on the general instability of stiffened cylindrical shells under
hydrostatic pressure. J. Mech. Engng Sci. 5, 23-27 (1963).
22. B. 0. Almroth and F. A. Brogan, Bifurcation buckling as an
approximation of the collapse load for general shells. AIAA
J. 10, 463-467 (1972).
23. D. Bushnell, B. 0. Almroth and F. Brogan, Finite-difference
energy method for nonlinear shell analysis. Comput. Structures 1, 361-387 (1971).
24. W. Pilkey, K. Saczalski and H. Schaeffer (Eds), Structural
Mechanics Computer Programs: Surveys, Assessments, and
Availability. University Press of Virginia, Charlottesville
(1974).
535
25. D. Bushnell, BOSOR5-Program for buckling of elasticplastic complex shells of revolution including large
deflections and creep. Comput. Structures 6, 221-239 (1976).
26. D. Bushnell, Plastic Buckling, Pressure Vessels and Piping:
Design Technology-1982, A Decadeof Progress,(Edited by S.
Y. Zamrik and D. Dietrich) pp. 47-117. ASME, New York
(1982).
27. D. Bushnell Analysis of ring-stiffened shells of revolution
under combined thermal and mechanical loading. AIAA J. 9,
401-410 (1971).
28. D. Bushnell, Analysis of buckling and vibration of ringstiffened, segmented shells of revolution. Int. J. Solids
Structures 6, 157-181 (1970).
29. G. A. Cohen, Conservativeness of a normal pressure field
acting on a shell. AIAA J. 4, 1886-1887 (1966).
30. J. J. Kotanchik, R. P. Yeghiayan, E. A. Witmer and B. A.
Berg, The transient linear elastic response analysis of complex thin shells of revolution subjected to arbitrary external
loadings, by the finite-element program, SABOR5-DRASTIC.
SAMSO TR 70-206, ASRL TR 146-10, Aeroelastic and
Struct. Res. Lab., MIT, April 1970.
31. P. M. Mebane and J. A. Stricklin, Implicit rigid body motion
in curved finite elements. AIAA J. 9, 344 (1971).
32. H. M. Adelman, D. S. Catherines and W. C. Jr. Walton, A
method for computation of vibration modes and frequencies
of orthotropic thin shells of revolution having general meridional curvature. NASA TN D-4972, Lengley Res. Center,
Hampton, VA Jan. 1969.
33. M. Stein, The effect on the buckling of perfect cylinders of
prebuckling deformations and stresses induced by edge support. NASA TN D-1510, p. 217. Langley Res. Center,
Hampton, VA, Dec. 1962.
34. D. Bushnell, BOSOR4: program for stress, buckling, and
vibration of complex shells of revolution. Structural
Mechanics Software Series, (Edited by N. Perrone and W.
Pilkey) , Vol. 1, pp. 11-143. University Press of Virginia,
Charlottesville, VA (1977).
35. D. Bushnell, Stress, buckling, and vibration of hybrid bodies
of revolution. Comput. Structures 7, 517-537 (1977).
36. G. E. Strickland, W. A. Loden and S. K. Ferriera, WASP-a
digital computer program for the linear elastic analysis of
hybrid symmetrically loaded bodies of revolution, Report
9-87-68-2, Lockheed Missiles & Space Co., Sunnyvale, CA,
June 1968.
37. Z. Zudans, Analysis of elastically coupled shells on elastic
foundation by hybrid method. Nuclear Engng Design 20,
85-129 (1972).
38. J. Ergatoudis, B. M. Irons and 0. C. Zienkiewicz, Curved,
isoparametric, quadrilateral elements for finite element
analysis. Int. J. Solids Structures 4, 31-42 (1968).
39. S. Ahmad, B. M. Irons and 0. C. Zienkiewicz, Curved thick
shell and membrane elements with particular reference to
axisymmetric problems. Proc. 2nd Conf. on Methods in
Structural Mechanics, Wright-Patterson Air Force Base, OH
(1968).
40. S. F. Pawsey, The analysis of moderately thick to thin shells
by the finite element method. UCSESM 70-12, University of
California, Berkeley, CA, Aug. 1970.
41. P. K. Larsen and E. P. Popov, Elastic-plastic analysis of
axisymmetric solids using isoparametric finite elements.
UCSESM 71-2, University of California, Berkeley, CA, Jan.
1971.
42. K. J. Bathe, E. L. Wilson and R. H. Iding, NONSAP-a
structural analysis program for static and dynamic response
of nonlinear systems. UCSESM 74-3, University of California, Berkeley, Feb. 1974.
43. R. S. Dunham and E. B. Becker, TEXGAP-the Texas grain
analysis program. TICOM Rep. 73-1, The Texas Institute for
Computational Mech., University of Texas at Austin, Aug.
1973.
44. P. Sharifi and D. N. Yates, Nonlinear thermo-elastic-plastic
and creep analysis by the finite element method. AIAA J. 12,
1210-1215 (1974).
5 36
D. BUSHNELL
45. G. Ferguson and N. Cyr, "DIAL-Structural Analysis System,

User's Manual, Unnumbered Lockheed Missiles and Space
Company Report, Sept. 20, 1982.
46. M. B. Marlow, FARSS User's Manual, unnumbered Lockheed Missiles & Space Company, Inc. Report, 1974.
47. J. C. Speare, Discrete-element static analysis of corestiffened shells of revolution for asymmetric mechanical
loading. ASRL TR 146-5, SAMSO TR 68-400, Aeroelastic &
Structures Research Laboratory, M.I.T., Cambridge, Mass,
Sept. 1968.
48. A. P. Cappelli, G. L. Agrawal and V. E. Romero, EPAD, a
computer program for the elastic plastic analysis of domed
structures and arbitrary solids of revolution subjected to
asymmetrical loadings. Service Bureau Corporation, Inglewood, California, AFWL-TR-69-146, Air Force Weapons
Laboratory, Kirtland AFB, New Mexico, Feb. 1970.
49. S. Ghosh and E. Wilson, Dynamic stress analysis of
axisymmetric structures under arbitrary loading. Earthquake
Engineering Res. Center Rep. EERC/69-10, Sept. 1969.
50. F. C. P. Yin, Discrete element analysis of core-stiffened
shells of revolution. M.I.T. Aeroelastic & Structures
Research Lab., TR 139-8, Feb. 1967.
51. B. M. Irons, Numerical integration applied to the finite element method. Int. Symp. on the Use of Electronic Digital
Computers in Structural Engineering, Working Session No. 4,
Paper No. 19, University of Newcastle (July 1966).
52. 0. C. Zienkiewicz, Isoparametric and applied numerically
integrated elements, a review. Numerical and Computer
Methods in Structural Mechanics (Edited by S. J. Fenves, N.
Perrone, R. Robinson and W. C. Schnobrich), pp. 13-41.
Academic Press, New York (1973).
53. J. A. Stricklin, W. E. Haisler and W. A. von Riesemann,
Formulation, computation, and solution procedures for
material and/or geometric nonlinear structural analysis by
finite element method. SC-CR-72-3102. Sandia Labs., Albuquerque, NM, July 1972.
54. D. Bushnell, Stress, buckling, and vibration of hybrid bodies
of revolution, Vol. 1. Analysis, LMSC Report D501503,
Lockheed Missiles & Space Co., Inc., Palo Alto, CA (March
1976).
55. V. V. Novoshilov, Foundations of the Nonlinear Theory of
Elasticity, Chap. 1. Graylock Press, Rochester, NY (1953).
56. A. W. Leissa, Vibration of shells. NASA SP-288, Chap. 1,
1973.
57. L. H. Donnell, Stability of thin walled tubes under torsion.
NACA Rep. 479, 1933.
58. L. H. Donnell, A discussion of thin shell theory. Proc. 5th
Int. Congr. Appl. Mech. (1938).
59. Kh. M. Mushtari, On the stability of cylindrical shells subjected to torsion. Trudy Kaz. avais, in-ta, Vol. 2, in Russian
(1938).
60. Kh. M. Mushtari, Certain generalizations of the theory of
thin shells. Izv. Fiz. Mat. ob-va. Pri Kaz., un-te 11, in
Russian (1938).
61. A. E. H. Love, A Treatise on the Mathematical Theory of
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
Elasticity, Ist Edn. Cambridge University Press (1892);,

4th Edn. Dover Pub., Inc. New York (1944).
A. E. H. Love, The small free vibrations and deformations of
a thin elastic shell. Phil. Trans. Roy. Soc., London, ser. A,
179, 491-549 (1888).
S. Timoshenko, Theory of Plates and Shells. McGraw-Hill,
New York (1959).
E. Reissner, A new derivation of the equations of the
deformation of elastic shells. Am. J. Math. 63, 177-184
(1941).
P. M. Naghdi and J. G. Berry, On the equations of motion of
cylindrical shells. J. Appl. Mech. 21, 160-166 (1964).
V. Z. Vlasov, Osnovnye differentsialnye Uravnenie obshche
teorii uprugikh obolochek. Prikl. Mat. Metkh. 8 (1944).
(English translation: NACA TM 1241, Basic Differential
Equations in the General Theory of Elastic Shells. Feb.
1951.)
V. Z. Vlasov, Obshchaya teoriya obolochek; yeye prilozheniya v tekhnike. Gos. Izd. Tekh.-Teor. Lit., MoscowLeningrad (1949). (English translation: NASA TT F-99,
General Theory of Shells and Its Applications in Engineering, April 1964.)
J. L. Sanders, Jr., An improved first approximation theory
for thin shells. NASA TR-R24, 1959.
R. Byrne, Theory of small deformations of a thin elastic
shell. Seminar Reports in Math., University of California
Pub. in Math, N.S. vol. 2, No. 1, pp. 103-152 (1944).
W. Flugge, Statik und Dynamik der Schalen. Julius Springer,
Berlin (1934). Reprinted by Edwards Brothers, Inc., Ann
Arbor, Michigan (1943).
W. Fligge, Stresses in Shells. Springer-Verlag, Berlin (1962).
A. L. Goldenveizer, Theory of Thin Shells. Pergamon Press,
New York (1961).
A. 1. Lur'ye, General theory of elastic shells. Prikl. Mat.
Mekh 4, 7-34 (1940), in Russian.
V. V. Novozhilov, The Theory of Thin Elastic Shells. Noordhoff, Groningen, The Netherlands (1964).
W. T. Koiter, A consistent first approximation in the general
theory of thin elastic shells. Proc. Symp. on Theory of Thin
Elastic Shells, I.U.T.A.M., Delft, 24-28 Aug. 1959, North
Holland Pub. Co., Amsterdam, pp. 12-33 (1960).
A L. Goldenveizer, Method for justifying and refining the
theory of shells (survey of recent results). Appl. Math. Mech.
(transl. of Prikl. Mat. Mekh.), 32, 704-718 (1968).
J. M. Klosner and H. S. Levine, Further comparison of
elasticity and shell theory solutions. AIAA J. 4, 467-480
(1966).
H. Kraus, Thin Elastic Shells. Wiley, New York (1967).
P. M. Naghdi, A survey of recent progress in the theory of
elastic shells. Appl. Mech. Rev. 9, 365-367 (1956).
A. Kalnins, Dynamic problems of elastic shells. Appl. Mech.
Rev. 18. 867-872 (1965).
A. S. Kadi, A study and comparison of the equations of thin
shell theories. Dissertation, Ohio State University, June
1970.

Buckling - Equations

Uploaded by

Copyright:

Available Formats

Buckling - Equations

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Buckling - Equations

Uploaded by

Copyright:

Available Formats

Computers & Structures Vol 18.No 3. pp.

tThis paper describes work performed by Lockheed Palo

The work was completed under Task 2307N I, "Basic Research

The majority of computerized analyses of thin shells are

If the displacement method is used, the strain energy

where dS is the elemental area of the reference surface.

addition, it is often advantageous not to have to choose

Computerized analysis of shells-governing equations

placement method were insistent that the displacements

The vector e, represents reference surface strains and

surface geometry and mesh spacing at i. The number of

represents the reference axis normal strain e,,, changes

in which all quantities are analogous to those in eqn (2).

so that eqn (4) in terms of the dependent variables di

where dL is the incremental length along the reference

If eqn (3)is expressed in discrete form and if the r.h.s. of

Computerized analysis of shells-governing equations

Two aspects of loading are of particular interest when

L (qd, + dTQd,) dL.

The second terms in each integrand represent the live

redistribution can occur. Flexible or "soft" parts of the

In thin shell analysis it is not necessary to include

dT d dS+ f(drTmrd +6.i7Ird,)dL

NORMAL DISPLACEMENT AT l8W AND MIDLENGTH-IN

End shortening/shell thickness

homogeneous quartic functional

BOUNDARY AND OTHER CONSTRAINT CONDITIONS

The energy minimization problem (displacement

can be obtained in which the coefficients as well as the

in which U, denotes an energy-like term pertaining to

ANALYSIS OF A CURVED BEAM

in which U, = strain energy; W = work done by external

The strain energy in the beam is given by

in which or is the stress and e. is the strain that produces

Computerized analysis of shells-governing equations

Fig. 5. Curved beam of developed length, L.

-2EaT [1, -zJ{e}+

sections remain normal, and the beam deforms only in

e = duldx + w/R +If 2

in which [j denotes a row vector, [I a matrix, and {} a

and the reference surface quantities e and K can be

-fEz dz; C 2 2 = Ez2dz

The strain energy can be expressed in terms of the

R(s) is the local radius of curvature of the reference

surface of the beam, and /3 is the rotation as shown in

Suppose that the beam is submitted to loading as

p. and traction p,, and constant-directional end loads, V,

where E is the modulus of elasticity and a is the

(pu +pw) ds + HuL + VWL*+ MI3L

provided the loads act in a constant direction as the

Fig. 6. Coordinates (sz) and displacements (uw).

Suppose the beam is supported as shown in Fig. 10.

U~= [A01, A02, A03

_!-z 4,OY - 24aT ti, -Z

{I,} - EWTI dzds

Fig. 8. Derivation of eqns (19) and (20).

Equation (25) can be written in a more general form

U' = [Ao] tKA] 3W"'

where, in the case of eqn (25)

The kinetic energy of the beam is