1652

Santa FeParana, Argentina, October 2002
S. R. Idelsohn, V. E. Sonzogni and A. Cardona (Eds.)
Mecanica Computacional Vol. XXI, pp. 16521667





COMPUTATIONAL BUCKLING ANALYSIS OF SHELLS:
THEORIES AND PRACTICE

Luis A. Godoy
*
and Eduardo M. Sosa
**


*
Professor, Universidad Nacional de Córdoba, FCEFyN, Departamento de Estructuras,
and researcher, CONICET
Casilla de Correo 916, Correo Central, Córdoba 5000, lgodoy@com.uncor.edu

**
Graduate student, University of Puerto Rico, Mayaguez, PR 00681-9041, Puerto Rico


Key words: ABAQUS, Buckling, Finite elements, Shells, Structural stability, Tanks.

Abstract. Shell buckling problems belong to the class of geometrically nonlinear behavior,
and may be coupled with material nonlinearity of the shell. There are many general-purpose
finite element programs that perform geometric and material nonlinear analysis of shells;
however, this does not mean that a user can feed data and collect reliable results without a
full understanding of the physics of the problem. This paper discusses the theories involved in
the explanation and classification of phenomena, and in the prediction of results. Next, those
approaches are considered in the practical analysis of one shell form, namely thin-walled
steel tanks used to store oil. Results have been obtained with the general-purpose package
ABAQUS, and they tend to show that their interpretation requires the use of Koiter’s theory in
order to make sense of what is obtained. Some thoughts on possible ways to implement Croll’s
lower bound reduced energy approach in practice are given.




1653
L. A. Godoy and E. M. Sosa
*****************************************************************
1. INTRODUCTION

In this paper we consider the phenomenon of buckling as a process of change in the shape
of a structure. Engineers design structures with a well-defined shape thoughtfully chosen to
fulfill some purpose. But under certain critical conditions the structure cannot withstand
further load with the same shape and changes its shape in a slow or sometimes in a violent
way. The view of buckling as the development of a new form is relatively recent and has
emerged as a consequence of the work of W. T. Koiter
24
. This introduction attempts to
establish what needs to be computed in a buckling analysis.
Buckling belongs to a class of problems in mechanics for which some form of singularity
occurs in an equilibrium situation. Other problems of the same class may be the phase
changes in matter, the propagation of cracks, and many others. What such problems have in
common is that they are described in terms of control parameters (such as the external loads)
and response parameters (such as the displacements), and that at some configuration of the
above parameters the system reaches a singularity and enters into a new set of possibilities,
not available at the beginning of the process. The singularity is usually called “critical state”
and the states beyond it are identified as “post-critical states”.
At present, the usual representation of buckling problems is made in terms of equilibrium
paths, i.e. a graph showing the evolution of the structural system in a space defined by the
control and the response variables. Essentially, one needs to identify a state that separates two
regimes, and to compute equilibrium states of both regimes. First, a pre-critical (or
fundamental, or primary) equilibrium path is required, often a linear equilibrium problem.
Second, the critical state needs to be identified. Third, a post-critical (or secondary)
equilibrium path is computed: this is a nonlinear path and often displays an evolution, i.e.
further changes in the shape of the structure occur along its way (mode jumping).
In this paper we argue that there are several levels of theories involved in the
computational buckling analysis of shells
a
. There are (a) predictive theories at the level of
computations, in which results are obtained; (b) explanatory theories, in which results make
sense (or not); and (c) classification theories, in which possible solutions are classified.
Computational mechanics tends to be associated only to predictive theories, but it will be
shown that explanations and classifications are as essential as predictions in any adequate
analysis. Three theories are examined: Koiter’s initial post-critical behavior
24
; Croll’s
reduced energy approach
6
; and non-linear computational mechanics. The complementarity of
these approaches is illustrated with reference to the buckling analysis of thin-walled steel
tanks employed to store liquids.

2. BRIEF REVIEW OF RELEVANT THEORIES

2.1 Koiter’s initial post-critical theory

Koiter
24
developed both a theoretical framework for the understanding and classification
of the mechanics of buckling, and also a tool to carry out the computations (for a recent
review of the developments of the theory, see [12,13]). But while it has almost no competitors
at present to understand the process of buckling, only in a few cases a tool has been
implemented to perform the computations
7, 8, 9, 22
. In this paper we consider three roles of the

a
In looking at the different theories used for the solution of a problem, Lakatos
23
stressed the need to use a
pluralist model: “In this pluralist model, the conflict does not arise between ‘theory and facts’, but between two
high level theories: an interpretative theory that provides the facts and an explanatory theory that explains them.
The interpretative theory may be of a level as high as that of the explanatory theory… The problem is about
deciding which theory will be considered as an interpretative theory and which one as an explanatory theory that
tentatively explains them”.
1654
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************
theory of Koiter and will attempt to show how other approaches used to perform the
computations are influenced at present from Koiter’s analysis tools.
The predictive side of Koiter’s approach relies on perturbation analysis. Once a critical
state has been detected (meaning that the displacement degrees-of-freedom Q
i
c
, the critical
load Ȝ
c
, and the mode shape x
i
have been evaluated) then the post-critical equilibrium path
[Ȝ(s), Q
i
(s)] is computed in terms of a perturbation parameter s as:

Q
i
(s) = Q
i
c
+ q
i
(1)
s + ½ q
i
(2)
s
2
+ … (1)

Ȝ(s) = Ȝ
c
+ Ȝ
(1)
s + ½ Ȝ
(2)
s
2
+ … (2)

where the notation Ȝ
(j)
refers to the j-th derivative of the variable Ȝ with respect to the
perturbation parameter. The perturbation equations of the equilibrium conditions are used to
evaluate the unknowns q
i
(j)
and Ȝ
(j)
. The accuracy of the solution may be improved by
including more terms in the perturbation expansions (1-2); however this accuracy is limited by
the order of the energy V in terms of the Q
i
.
The main limitation of Koiter’s analysis is the quality of the quantitative predictions that
can be obtained from perturbation analysis: for shells, the results are only accurate up to
displacement amplitudes less than the wall-thickness. Thus, a perturbation analysis can only
carry out computations close to the critical state, and as such gives information limited to the
initial post-critical behavior of the shell.
As a theory of classification, Koiter predicts well all possible solutions that emerge from a
critical state. To achieve this, Koiter uses the total potential energy of the system, V. For
equilibrium, the condition of stationary energy with respect to the displacement degrees-of-
freedom Q
i
of a discrete system is stated as
b
:

V
i
= ˜V/˜Q
i
= 0 (3)

Given a small perturbation q
j
, equilibrium states for which V
ij
q
j
> 0 are said to be stable,
while a critical state is given by V
ij
q
j
= 0 in which case q
j
= x
j
is identified as a critical mode.
Thus, a critical state is characterized by

V
ij
x
j
= 0 (4)

The nature of the critical state is given by the scalar quantity V’
i
x
i
, where ( )’ = ˜V/˜Ȝ.

For V’
i
x
i
= 0 Æ Bifurcation state
V’
i
x
i
 0 Æ Limit state

Bifurcation states can be classified into asymmetric and symmetric behavior by means of
the coefficient C = V
ijk
x
i
x
j
x
k
as

For C = 0 Æ Symmetric Bifurcation
C  0 Æ Asymmetric bifurcation

This is not the end of the road, because symmetric bifurcations can be further classified
using the stability coefficient V
4
= V
ijkl
x
i
x
j
x
k
x
l
as

For V
4
> 0 Æ Stable symmetric bifurcation
V
4
< 0 Æ Unstable symmetric bifurcation


b
This is the extension of Koiter’s theory to discrete systems, carried out at University College London, since
Koiter only worked with continuous systems. However, for computational purposes, the discrete version seems
more suitable because the degrees-of-freedom may be those arising from a finite element analysis. See Ref. [12].
1655
L. A. Godoy and E. M. Sosa
*****************************************************************
The importance of this is enormous: one can classify all possible solutions and be confident
that only one of them will occur. The theory also classifies cases in which buckling modes
couple at the critical state leading to compound critical states and mode interaction, and this
task was completed by Chilver, Thompson, and Supple at University College London and
elsewhere in England.
A classification scheme identical to Koiter’s but based on dynamic (rather than static)
analysis has been formulated by Huseyin
19
. Finally, the only alternative for classification is
catastrophe theory, which is more qualitative in nature and allows for multiple load parameter
systems, but its success in elastic stability analysis has been very limited.
Koiter’s success has been highest regarding its power to explain the phenomena of
buckling of shells. Here the main point was the discovery that small imperfections (either in
the geometry or in the loads) destroy a bifurcation behavior leading to a nonlinear
fundamental path with a different critical state. For unstable post-critical behavior this means
that there is a reduction in the critical load, and this reduction may be significant for shell
problems.

2.2 Croll’s reduced energy theory

Perhaps the structural application for which buckling problems are most crucial are shell
problems, for which high imperfection-sensitivity has been observed. Croll’s approach
emerges from the theory of Koiter and uses it as a theoretical framework to obtain
approximate values of critical loads and mode shapes for shells. As in the case of Koiter, this
is both an explanatory theory and a predictive theory, but it makes no attempt to produce a
new classification.
The explanatory part of Croll’s theory is based on two postulates
6
: (1) Systems that may
have changes in their membrane resistance are likely to show significant geometric nonlinear
response. On the other hand, systems with changes mainly in their bending resistance do not
show highly geometric nonlinear response. (2) Practically significant post-buckling loss of
stiffness can only occur when the fundamental equilibrium path contains a significant
contribution from membrane energy.
Croll’s theory led to the reduced energy approach as a predictive tool to carry out the
computations. According to this, lower bounds to buckling into a particular mode are given by
an analysis in which the membrane energy of the shell is eliminated.
The total potential energy is written in the form

( )
1
2
ij ij ij ij
V N M dA ε χ ψ = + −
³
(5)

for i, j = 1,2; where
ij ij
N ε is the membrane energy and
ij ij
M χ is the bending energy, and ȥ is
the load potential. Next, each field is represented in terms of the fundamental path (-)
F
in the
form

;
;
F F
ij ij ij ij ij ij ij ij
F F
ij ij ij ij ij ij
N N N N
M M M
λ ε λε ε ε
λ χ λχ χ
′ ′′ ′ ′′ = + + = + +
′ ′ = + = +
(6)

where (-)’ represents the linear contribution to the variable, while (-)’’ is the nonlinear
contribution. Substitution of each term in the energy V, and considering stationary energy
leads to the condition

1656
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************
( ) ( )
2
1
[ ]
2
F F
ij ij ij ij ij ij ij ij
V N M N N dA δ ε χ λ ε ε ′′ ′′ = + + +
³
(7)

The first parenthesis leads to the stiffness matrix, while the second one leads to the load-
geometry matrix.
Croll and co-workers (see Refs. [2, 27], and others cited in [6]) investigated the
contribution of each individual term into the energy, and found those that contributed to
stabilize the shell and those that were de-stabilizing. The erosion of a critical state due to the
coupling of modes and imperfections could only occur because some of the stabilizing
contributions were lost, so that the new approach eliminated those terms from the eigenvalue
problem in order to obtain a lower bound to the buckling loads.
The consequences of this are that we may now compute bifurcation buckling of a shell but
with a modified version of the total potential energy and obtain good estimates (both
quantitative and qualitative) of the true buckling resistance. The modifications consist in the
elimination of the membrane energy contribution to the critical state. This does not mean that
the membrane energy is eliminated from the fundamental path.

2.3 Nonlinear computational mechanics

There is a whole branch of mechanics dealing with the computation of nonlinear problems.
This is present in most general-purpose packages, such as ABAQUS, ADINA, ALGOR,
DIANA … The main purpose is to develop efficient algorithms to carry out the computation
of equilibrium paths. Many engineers view this approach as a heavenly solution to all
problems, since all you need is brute-force computer capabilities.
Nonlinear algorithms are highly specialized and efficient for the computation of an
equilibrium path. A good text in this field is Belytschko et al.
3
and the computational
buckling approach is well presented by Bushnell
5
.
There are two main tools incorporated into this class of predictive approaches: nonlinear
analysis and eigenvalue analysis. The eigenvalue buckling problem starts with the
computation of the pre-critical states u
F
from the condition of equilibrium

F
K u + = 0 p (8)

Next, the stress field (ı
F
) is evaluated. The results of this first stage are carried into a second
stage to solve the eigenvalue problem

F
G
[K + K ( )] = 0 λ σ φ (9)

where K is the stiffness matrix, K
G
the load-geometry matrix, ĭ is the eigenvector (buckling
mode) and Ȝ
c
is the eigenvalue (critical load). Basically, equation (9) can compute bifurcation
from the linear fundamental path of equation (8).
Notice that this is a tool of analysis and not a phenomenological theory, so that no
classification or explanation is attempted. It tells you how to compute, but it does not help in
the understanding of the computed results.

3. PRACTICE: THE COMPUTATIONAL BUCKLING ANALYSIS OF THIN-
WALLED STEEL TANKS

Steel tanks used to store oil and petrochemical products are very thin shells, simple in their
overall geometry, and very costly. Perhaps the most significant behavior of a tank is its
buckling strength: a tank may easily loose its geometry. There are many reasons to be
1657
L. A. Godoy and E. M. Sosa
*****************************************************************
concerned about such uninvited changes in the geometry: a floating roof may not slide along
the side walls, a damaged geometry triggers imperfection-sensitivity, and finally there is a
huge loss for each day a tank does not operate. The stakeholders are the owners of the tanks,
environmental protection agencies, and the insurance companies.
Buckling of tanks is a field in which there is no chance of finding appropriate analytical
closed-form solutions so that computational mechanics is a must. Tanks are designed by
engineering firms with standard finite element computer packages and medium to low level
engineering knowledge about the mechanics of the problem. The basic tools of analysis are
linear/nonlinear static analysis, bifurcation-buckling analysis, and nonlinear dynamic analysis.
The tools are available, thus we face here a problem of efficient use of the available software.
In this section we review some examples of studies carried out with different theories in order
to illustrate what phenomena may be identified with the models.

4. ELEMENTARY BUCKLING ANALYSIS FOR PRESSURE-INDUCED
INSTABILITY

Vacuum pressure is a static load produced during the process of emptying a tank with
insufficient venting. This form of buckling is usually associated to operational problems of a
plant and is frequently found in practice. Koiter’s predictive theory has been applied by
Jorgensen
22
to evaluate isolated buckling loads and modes, and imperfection sensitivity under
pressure. What features of real tanks are included in the analysis? A steel cylinder with
R=28m, L=24m, and a wall thickness tapered between h=0.032 at the bottom and h=0.012m
at the top; the shallow cap itself is not part of the model (the finite difference discretization
does not allow for a roof) and has been substituted by simply supported boundary conditions.

0
2
4
6
8
0 0.5 1
Liquid level
N
o
r
m
a
l
i
z
e
d

c
r
i
t
i
c
a
l

p
r
e
s
s
u
r
e
Wind pressure
Vacuum


Figure 1. Influence of liquid level on the critical pressures of tanks. The results for wind pressure are for an open-
top tank
10
, while the results for vacuum
22
correspond to a simply-supported condition at the top.

In an independent study, Flores and Godoy
10
considered wind pressures on open-top tanks
to address the influence of a liquid on the critical wind pressures, and the results are also
plotted in Figure 1.
Results of the influence of the liquid on the critical pressure (normalized with respect to an
empty tank) are shown in Figure 1 for the two loading conditions, and they show the same
trend: if the tank is filled with liquid up to 50% of its height, the increase in critical pressure is
of only 3 to 5%. This is of great importance both from the handling of tanks in practice and
also for what studies are worth doing. The information in Figure 1 does not tell anything about
1658
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************
other changes that may have occurred due to the liquid in the postbuckling behavior, or in the
imperfections sensitivity. Such studies are reported by Jorgensen using Koiter’s perturbation
analysis
22
.

5. STATIC OR DYNAMIC BUCKLING?

Wind acting on a tank produces a pressure variable both in time and space. Three-second
wind gusts are considered as the basic information for design. The main questions regarding
the computational modeling of wind-load buckling may be posed as: Is this a static or
dynamic stability problem? How important is the influence of the roof? What is the influence
of previous damage on the buckling strength of a tank?
The question about the importance of dynamic effects cannot be solved solely within the
limitations of static theories, and we need to employ the criterion due to Budianski and Roth
4,
18
for dynamic buckling. Flores and Godoy
10, 11
, have performed such studies previously for
open-top tanks and silos.
For three-second gusts, an example of the nonlinear dynamic response of a tank under
increasing values of step load is shown in Figures 2 to 6. The tank itself has a cylindrical part,
a fixed roof and rafters supporting the roof, as indicated in Figure 2, and the finite element
discretization with ABAQUS
17
employs about 12,600 shells elements. The space variation of
pressures in the cylindrical part of the tank is constant in elevation and variable around the
circumference, as in most works in this field
10
, but for the pressures acting on the roof there is
no clear indication as to what should be assumed. Here we consider the wind-tunnel pressures
evaluated by MacDonald et al.
25
, shown in Figure 3. The values of the pressures are scaled
using the load parameter Ȝ, which is the scalar that multiplies the pressure distribution shown
in Figure 3. The maximum pressure in the reference case is 1KPa acting on the windward
meridian on the cylindrical part of the shell. This pressure distribution is applied as a
rectangular impulse (constant amplitude in time) with 3 sec duration. Notice that this problem
is far more complex than the earlier investigations on simple cantilever cylinders.




Figure 2. Model of the tank with fixed roof and rafter supports on the roof.

The nonlinear dynamic response has been computed using ABAQUS
17
, and the radial
displacement is shown in Figure 4 for a specific location in the cylindrical part (node 8938).
According to the criterion for dynamic buckling, the response is computed for several values
of the load parameter Ȝ. For Ȝ = 2.50 and Ȝ = 2.51, the response shows small amplitude
vibrations, of the order of the shell thickness. However, for Ȝ = 2.515 we get a qualitative
1659
L. A. Godoy and E. M. Sosa
*****************************************************************
change in the response, with large amplitude vibrations starting at about t = 2.4 sec. This is an
indication of a non-proportional change in the response due to a small increase in the load
parameter. As the load parameter is increased, the shell exhibits instability, but the process
starts earlier: for example, for Ȝ = 2.52, large vibrations start at t = 1 sec, while for Ȝ = 2.60
the time is reduced to approximately t = 0.5 sec. As the load parameter increases, the number
of small-amplitude oscillations is reduced before the shell buckles.

Figure 3. Pressure distribution assumed on the roof of the tank (after MacDonald et al.
25
).

Notice that Figure 4 has been plotted for a specific point of the shell, however, a similar
behavior is obtained if other degrees-of-freedom (DOF) are considered, provided the
nonlinear vibrations have a nonzero component in this DOF.

Figure 4. Nonlinear dynamic response for 3 sec wind gust assumed constant.
Each curve is for different pressure amplitude. Instability occurs for Ȝ = 2.515.

The vibration pattern due to the nonlinear dynamic analysis, for Ȝ = 2.515, at time t = 2.55
sec, is shown in Figure 5, while Figure 6 shows the vibrations for Ȝ = 2.55, at time t = 3 sec.
Figure 5 represents the onset of dynamic buckling, as this is the first stage for the lowest load
for which instability occurs. Figure 6, on the other hand, is an advanced vibration pattern, and
would only occur if the shell is suddenly slammed by the pressure Ȝ = 2.55. The first figure
shows a symmetric pattern of vibration, with blue deflections towards the inside and red
towards the outside of the shell. The second figure illustrates a break in the symmetry, due to
coupling of modes in an advanced postbuckling state.

Node: 8938
-0.18
-0.16
-0.14
-0.12
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Time [sec]
D
i
s
p
l
a
c
e
m
e
n
t

[
m
]
2.500
2.510
2.515
2.520
2.550
2.600
1660
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************


Figure 5. Nonlinear dynamic analysis, deflected pattern for Ȝ = 2.515, at time t=2.55 sec.



Figure 6. Nonlinear dynamic analysis, deflected pattern for Ȝ = 2.55, at time t=3 sec.

Static models. The above procedure is time consuming and computationally expensive, since
several cases of dynamic nonlinear response have to be solved. Thus, a simpler approach has
been attempted, using a static, geometrically nonlinear analysis of the shell with the same
pressure distribution but monotonically increasing in time.

1661
L. A. Godoy and E. M. Sosa
*****************************************************************
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
-0.005 -0.004 -0.003 -0.002 -0.001 0.000
Displacement [m]
F
r
a
c
t
i
o
n

o
f

a
p
p
l
i
e
d

l
o
a
d
N: 8938

Figure 7. Static nonlinear analysis. The maximum load is Ȝ = 2.55.

The equilibrium path for load parameter versus a DOF of the structure is shown in Figure
7. Here the same DOF considered in Figure 4 has been chosen. The path is nonlinear and
reaches a maximum at Ȝ = 2.55, that is a value slightly higher than the dynamic buckling
pressure identified in the nonlinear dynamic study (Ȝ = 2.515), the difference being of only
1.4%. According to Koiter’s classification theory, this is an unstable symmetric bifurcation. A
symmetric shape is found at the onset of instability, as illustrated in Figure 8. Figures 5 and 8
show similar patterns of behavior of the shell, so that not only the maximum load but also the
buckling shape is well predicted by a nonlinear static analysis of this problem.


Figure 8. Static nonlinear analysis, deflected shape for Ȝ = 2.55.

The reasons for the close agreement between nonlinear static and dynamic solutions are to
be found in the natural frequencies of the shell
26
. For the present configuration, the period of
vibration of the shell is 0.36 sec, thus far from the excitation considered of 3 sec.
Bifurcation buckling analysis has also been investigated for this problem using ABAQUS
17
, and the results lead to a lowest critical load Ȝ = 2.6107 (see Figure 9). A second mode
(Figure 10) is identified at almost the same value Ȝ = 2.6133. There is a small difference
between static nonlinear and bifurcation buckling (2.3 %).

1662
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************


Figure 9. Bifurcation buckling, first mode, Ȝ = 2.6107.

A lesson learned is that this is mainly a static problem. It is only now that one can
confidently employ static predictions for this kind of environmental action. In the previous
studies, Koiter’s predictive approach has not been used, but a nonlinear computational
mechanics approach has been adopted. However, Koiter’s explanatory theory was present
throughout the study: all explanations are possible thanks to the framework provided by the
stability theory due to Koiter.



Figure 10. Bifurcation buckling, Second mode, Ȝ = 2.6133.

6. INITIAL VERSUS ADVANCED POSTCRITICAL BEHAVIOR

The static response in the previous section was limited to the close vicinity of the critical
state. This is reasonable, especially for the case of a tank with a roof, but in more flexible
structures one may want to consider the behavior along the postcritical equilibrium path.
1663
L. A. Godoy and E. M. Sosa
*****************************************************************
Let us consider an open-top cylindrical tank, again under wind load (Figure 11).
Figure 11. Nonlinear equilibrium paths for an open-top tank with initial load imperfections
20
.
R=19m, Height=7.6m, h=0.01m.

The equilibrium paths for several load imperfections computed with ABAQUS are shown
in Figure 11, and an unstable symmetric bifurcation behavior is identified, according to
Koiter’s classification theory. In the close vicinity of the bifurcation, a symmetric behavior is
computed similar to what was observed in Figure 7; however, as the deflections increase in
the postcritical path, a more elaborate behavior occurs: The left branch (negative
displacements in the figure) tends to decrease, showing an unstable postcritical behavior. The
right branch (positive displacements) decreases until a certain value, at which a snap occurs
and the structure changes from positive to negative displacements. This is a feature that could
not be predicted by Koiter’s initial poscritical analysis, since it occurs far from the critical
state and is due to advanced nonlinearity in the response.
An advanced postcritical behavior is also required in other buckling problems of steel
tanks. Often, tanks are constructed near a shore with poor soil conditions and may face
support settlements. Traditionally, this problem has been viewed in the literature as a linear
static behavior, so that the geometric nonlinearity is not accounted for. But recent studies
15,16

have shown that the nonlinear behavior of the shell plays an important role in the response,
with out-of-plane deflections of the order of five or more times the settlement in the vertical
direction. A lesson learned is that bifurcation is highly relevant in this case and may provide
good estimates of the buckling capacity.

7. IMPERFECTION SENSITIVITY

One of the crucial aspects of Koiter’s theory was the inclusion of imperfection sensitivity
into the field of buckling. A key feature in Koiter’s explanatory theory is the sensitivity of
buckling loads to the influence of small imperfections in some shells. In the previous case of a
tank with a stiff roof, the shell under wind pressure is almost insensitive to imperfections. But
for open-top tanks, the shell may display high sensitivity, as reported in a recent paper
14
.
How this can permeate into practice for other problems can be illustrated with respect to
the results of Figure 1. Let us plot again the results but now with a critical load normalized
with respect to the tank full of liquid in the vertical axis, and the liquid level in the horizontal
axis. What we now see is a diagram of sensitivity of the critical load to changes in the level of
liquid inside the tank. The shape of the diagram is not surprising: it is a typical diagram with a
singularity at the origin (1 in this case) and which is highly sensitive to small changes and
0.00
0.50
1.00
1.50
2.00
2.50
-0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10
U2
L
o
a
d

F
a
c
t
o
r
1664
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************
almost insensitive to large changes, for which there is a plateau. It is possible to look for this
form of diagrams only within the explanatory theory of Koiter.
Figure 12. Sensitivity of critical load to a decrease in the liquid level filling the tank. Data as in Figure 1.

In practice this means that little contribution can be expected from the liquid to stabilize
the tank, so that no real advantage can be gained by filling the tank with say half its contents.

8. POSSIBILITIES OF IMPLEMENTATION OF CROLL’S APPROACH

Croll’s approach has not been mentioned in the sections about applications mainly because
it is more a predictive tool than an explanatory theory. The use of Croll’s theory in
engineering practice is conditioned to the availability of a suitable predictive tool to carry out
the analysis. In most cases, the reduced energy lower bound approach has been used with an
analytical solution of the buckling modes, so that simple loading and boundary conditions can
be taken into account. Furthermore, in cases of more complex structures, such as those shown
in Figure 2 with compound shells and rafters on the roof, there are no analytical solutions
available that could be used for the computation of the total potential energy. Thus, although
this approach is physically sound and convenient, it faces the shortcoming of not being easy to
use in practice. Only if the engineer is ready to develop his/her own computer program for the
lower bound analysis, then it may be employed with advantage over a full nonlinear analysis.
An alternative approach would be to consider a homogeneous steel shell as if it was made
with a laminated material. In a laminated material, modeled by the classical lamination theory
developed for composite materials
21, 1
, the stiffness of an individual lamina is computed as in
plane stress behavior. The lamina is rotated to the global directions of the laminate. Individual
contributions from laminae are assembled into a laminate and define a constitutive relation of
the form

N A B
M B D
ε
χ
­ ½ ­ ½ ª º
=
® ¾ ® ¾
« »
¯ ¿ ¯ ¿ ¬ ¼
(10)

where {N, M} represent the membrane and bending stress resultants; {İ, Ȥ} are the membrane
strains and changes in curvature of the shell; and [A, B, D] contain the coefficients of the
laminate. For a symmetric laminate, the bending-extension coupling vanishes (B = 0), and we
are left with the uncoupled membrane and bending contributions. One of the advantages of
0
0.2
0.4
0.6
0.8
1
1.2
0.8 0.85 0.9 0.95 1
Liquid level
N
o
r
m
a
l
i
z
e
d

c
r
i
t
i
c
a
l

l
o
a
d
wind
vacuum
1665
L. A. Godoy and E. M. Sosa
*****************************************************************
this indirect definition of the material is that now one can perform separate evaluations of the
individual terms of the membrane and bending contributions by adequately controlling the
individual coefficients of the constitutive relations (10). Rather than deleting terms in the
energy, one could delete coefficients in the constitutive relation.
Consider again the eigenvalue analysis (equations 8 and 9). The lower bound approach
requires the computation of the fundamental equilibrium path (usually a linear path in
equation 8) using the full energy contributions to the stiffness of the shell. But for the
eigenvalue problem (equation 9), we need separate computation of K and K
G
. Using a
general-purpose finite element program, one needs to work in two steps, with different shell
properties in each. ABAQUS, for example, allows the use of steps in an eigenvalue analysis,
but does not easily allow for changes in the properties from one step to the next.
We have not developed these ideas in detail at present; however, it is our feeling that if the
reduced energy approach is to be adopted in practice, ways to implement it into standard finite
element codes should be found. This section outlines a possibility of using laminate elements
to account for the energy contributions of basically homogeneous shell structures.

9. CONCLUSIONS

This paper stresses the importance of explanatory theories for the computational buckling
analysis of shells. It is shown that the algorithms of nonlinear computational mechanics that
are implemented in most general-purpose finite element programs are necessary but not
sufficient to perform adequate instability studies, and that it is the power of explanation in
Koiter’s theory that allows to make sense of results.
As a predictive tool, several authors have implemented Koiter’s perturbation analysis in
conjunction with finite elements for the analysis of shells; however, its use has been limited to
specific-purpose programs not available for commercial use.
Croll’s reduced energy approach has suffered the same limitations in practice. However,
some possibilities are indicated in this paper that may help to turn this into a practical tool:
The use of classical lamination theory for the solution of shells made with homogeneous
materials. This indirect approach may deserve closer attention to model the reduced-energy
contributions in which some of the energy contributions are dropped.
The framework of explanatory theories and predictive theories is considered in practice for
the analysis of thin-walled, steel tanks to store oil and petrochemical fluids. Simple parametric
studies of bifurcation buckling may be enhanced by considering them with respect to the
imperfection-sensitivity of shells. Thus, a case of influence of liquid inside a tank as a
stabilizing factor can be seen as the imperfection-sensitivity of tanks full of liquid with respect
to a decrease in the liquid level. This shows a high sensitivity, so that for practical purpose it
may not be convenient to speculate with the liquid level as a positive factor in the event of a
hurricane.
To understand the behavior of shells under wind gusts it is necessary to enlarge the
theoretical frame and consider dynamic buckling. However, the results show that the problem
behaves essentially as a static one, and inertia forces de not play an active role because the
frequencies of excitation are far from the natural frequencies of the shells. In this case,
Koiter’s theory becomes a good theoretical framework for the analysis.

Acknowledgements. This work was supported by Agencia Córdoba Ciencia, CONICET
(Consejo Nacional de Investigaciones Científicas y Técnicas), and the National University of
Córdoba. The first author acknowledges the valuable contribution of F. G. Flores, R. Jaca and
J. C. Virella.

1666
MECOM 2002 First South-American Congress on Computational Mechanics
*****************************************************************
REFERENCES

[1] E.J. Barbero, Introduction to Composite Materials Design, Taylor and Francis,
Philadelphia, PA (1999).
[2] R.C. Batista and J.G.A. Croll, “A design approach for unstiffened cylindrical shells under
external pressure”, Proc. Int. Conf. Thin-Walled Structures, Ed. J. Rhodes, Elsevier (1979).
[3] T. Belytschko, K.W. Liu and B. Moran, Nonlinear Finite Elements for Continua and
Structures, Wiley, New York (2000).
[4] B. Budianski, “Dynamic buckling of elastic structures: Criteria and estimates”, in Dynamic
Stability of Structures, Ed. G. Herrmann, Pergamon Press, pp. 83-106 (1967).
[5] D. Bushnell, Computerized Buckling Analysis of Shells, Martinus Nijhoff, Dordrecht
(1985).
[6] J.G.A. Croll, “Shell buckling: A return to basic mechanics”, in Applied Mechanics in the
Americas, vol.1, American Academy of Mechanics, pp. 410-417 (1995).
[7] F.G. Flores and L.A. Godoy, “Instability of shells of revolution using ALREF: Studies for
wind loaded shells”, In: Buckling of Shells in Land, in the Sea and in the Air, Elsevier Applied
Science, Oxford, pp. 213-222 (1991).
[8] F.G. Flores and L.A. Godoy, “Elastic postbuckling analysis via finite element and
perturbation techniques, Part I: Formulation”, Int. Journal Numerical Methods in
Engineering, vol. 33, pp. 1775-1794 (1992).
[9] F.G. Flores and L.A. Godoy, “Elastic postbuckling analysis via finite element and
perturbation techniques, Part II: Application to shells of revolution”, Int. Journal Numerical
Methods in Engineering, vol. 36, pp. 331-354 (1992).
[10] F.G. Flores and L.A. Godoy, “Buckling of short tanks due to hurricanes”, Journal
Engineering Structures, 20(8), pp. 752-760 (1998).
[11] F. G. Flores and L. A. Godoy, “Forced vibrations of silos leading to buckling”, Journal
of Sound and Vibration, 224 (3), pp. 431-454 (1999).
[12] L.A. Godoy, Theory of Elastic Stability: Analysis and Sensitivity, Taylor and Francis,
Philadelphia, PA (2000).
[13] L.A. Godoy, “Sobre el sentido del progreso en la teoría de estabilidad elástica”,
Mechanics, vol. 31 (3-4), pp. 30-52 (2002).
[14] L.A. Godoy and F.G. Flores, “Imperfection sensitivity of wind loaded tanks”, Int. J.
Structural Engineering and Mechanics, vol. 13(5), pp. 533-542 (2002).
[15] L.A. Godoy and E.M. Sosa, “Deflections of thin-walled storage tanks with roof due to
localized support settlements”, In: Advances in Structural Engineering and Mechanics, Tecno
Press, Seoul (2002).
[16] L.A. Godoy and E.M. Sosa, “Localized support settlement of thin-walled storage tanks”,
submitted for publication (2002).
[17] H.D. Hibbit, B.I. Karlson and Sorensen, ABAQUS/ User’s Manual, Hibbitt, Karlsson and
Sorensen, Inc. 1997.
[18] N.J. Hoff, “Dynamic stability of structures”, in Dynamic Stability of Structures, Ed. G.
Herrmann, Pergamon Press, pp. 7-41 (1967).
[19] K. Huseyin, Vibrations and Stability of Multiple Parameter Systems, Ivaordhoff, Alphen
(1978).
[20] R. Jaca, L.A. Godoy and F.G. Flores, “Analisis postcritico de tanques de pared delgada
bajo cargas de viento”, I South American Congress on Computational Mechanics, Paraná,
Argentina (2002).
[21] R.M. Jones, Mechanics of Composite Materials, Hemisphere Publishing, New York
(1975).
1667
L. A. Godoy and E. M. Sosa
*****************************************************************
[22] F. Jorgensen, “Buckling behavior of a liquid storage tank”, Thin-Walled Structures, vol.
1(4), pp. 309-323 (1983).
[23] I. Lakatos, The Methodology of Scientific Research Programmes, vol. I, Cambridge
University Press, Cambridge, UK (1978).
[24] W.T. Koiter, On the Stability of Elastic Equilibrium, Ph. D. Thesis, Delft Institute of
Technology, Delft, Holland (in Dutch) (1945). There is an English translation by NASA,
1967.
[25] P.A. MacDonald, K.C. Kwok, J.D. Holmes, “Wind loads on circular storage bins, silos
and tanks: I. Point pressure measurements on isolated structures”, Journal of Wind
Engineering and Industrial Aerodynamics, vol. 31, pp. 165-188 (1988).
[26] J.C. Virella, L.A. Godoy and L.E. Suárez, “Influence of the roof on the natural
frequencies of steel tanks”, Journal of Sound and Vibration, submitted for publication (2002).
[27] S. Yamada and J.G.A. Croll, “Buckling and postbuckling characteristics of pressure-
loaded cylinders”, ASME Journal of Applied Mechanics, vol. 60, pp. 290-299 (1993).

and many others. But under certain critical conditions the structure cannot withstand further load with the same shape and changes its shape in a slow or sometimes in a violent way. Computational mechanics tends to be associated only to predictive theories. not available at the beginning of the process. a post-critical (or secondary) equilibrium path is computed: this is a nonlinear path and often displays an evolution. Essentially. and (c) classification theories.  . Three theories are examined: Koiter’s initial post-critical behavior 24. In this paper we argue that there are several levels of theories involved in the computational buckling analysis of shells a. and that at some configuration of the above parameters the system reaches a singularity and enters into a new set of possibilities. First. one needs to identify a state that separates two regimes. or primary) equilibrium path is required. in which results make sense (or not). This introduction attempts to establish what needs to be computed in a buckling analysis. in which possible solutions are classified. 9. Second. The interpretative theory may be of a level as high as that of the explanatory theory… The problem is about deciding which theory will be considered as an interpretative theory and which one as an explanatory theory that tentatively explains them”. There are (a) predictive theories at the level of computations. and non-linear computational mechanics. Other problems of the same class may be the phase changes in matter. the usual representation of buckling problems is made in terms of equilibrium paths. i. BRIEF REVIEW OF RELEVANT THEORIES 2. the propagation of cracks. 22./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** 1. Buckling belongs to a class of problems in mechanics for which some form of singularity occurs in an equilibrium situation. 8. In this paper we consider three roles of the a In looking at the different theories used for the solution of a problem. Koiter 24. (b) explanatory theories. At present.e. i. only in a few cases a tool has been implemented to perform the computations 7. a pre-critical (or fundamental. but between two high level theories: an interpretative theory that provides the facts and an explanatory theory that explains them. The singularity is usually called “critical state” and the states beyond it are identified as “post-critical states”.13]). see [12. Croll’s reduced energy approach 6. and also a tool to carry out the computations (for a recent review of the developments of the theory. Lakatos 23 stressed the need to use a pluralist model: “In this pluralist model. but it will be shown that explanations and classifications are as essential as predictions in any adequate analysis. Engineers design structures with a well-defined shape thoughtfully chosen to fulfill some purpose. 2. a graph showing the evolution of the structural system in a space defined by the control and the response variables. T. and to compute equilibrium states of both regimes. The view of buckling as the development of a new form is relatively recent and has emerged as a consequence of the work of W. in which results are obtained. the conflict does not arise between ‘theory and facts’.1 Koiter’s initial post-critical theory Koiter 24 developed both a theoretical framework for the understanding and classification of the mechanics of buckling. the critical state needs to be identified. But while it has almost no competitors at present to understand the process of buckling. further changes in the shape of the structure occur along its way (mode jumping). Third. INTRODUCTION In this paper we consider the phenomenon of buckling as a process of change in the shape of a structure. often a linear equilibrium problem. What such problems have in common is that they are described in terms of control parameters (such as the external loads) and response parameters (such as the displacements).e. The complementarity of these approaches is illustrated with reference to the buckling analysis of thin-walled steel tanks employed to store liquids.

the results are only accurate up to displacement amplitudes less than the wall-thickness. since Koiter only worked with continuous systems. As a theory of classification.  . Koiter uses the total potential energy of the system. a critical state is characterized by Vij xj = 0 The nature of the critical state is given by the scalar quantity V’i xi . and the mode shape xi have been evaluated) then the post-critical equilibrium path [ (s). The perturbation equations of the equilibrium conditions are used to evaluate the unknowns qi(j) and (j). and as such gives information limited to the initial post-critical behavior of the shell. The predictive side of Koiter’s approach relies on perturbation analysis. the discrete version seems more suitable because the degrees-of-freedom may be those arising from a finite element analysis. See Ref. To achieve this. For equilibrium. the critical load c. Once a critical state has been detected (meaning that the displacement degrees-of-freedom Qic. the condition of stationary energy with respect to the displacement degrees-offreedom Qi of a discrete system is stated as b: Vi = V/ Qi = 0 (3) Given a small perturbation qj. however this accuracy is limited by the order of the energy V in terms of the Qi. a perturbation analysis can only carry out computations close to the critical state. However. Koiter predicts well all possible solutions that emerge from a critical state. because symmetric bifurcations can be further classified using the stability coefficient V4 = Vijkl xi xj xk xl as For V4 > 0 V4 < 0 Stable symmetric bifurcation Unstable symmetric bifurcation b This is the extension of Koiter’s theory to discrete systems. [12]. V. while a critical state is given by Vij qj = 0 in which case qj = xj is identified as a critical mode.0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** theory of Koiter and will attempt to show how other approaches used to perform the computations are influenced at present from Koiter’s analysis tools. Thus. Qi(s)] is computed in terms of a perturbation parameter s as: Qi(s) = Qic + qi(1) s + ½ qi(2) s2 + … (s) = (j) c (1) (2) + (1) s+½ (2) 2 s +… where the notation refers to the j-th derivative of the variable with respect to the perturbation parameter. Thus. carried out at University College London. The accuracy of the solution may be improved by including more terms in the perturbation expansions (1-2). (4) Bifurcation states can be classified into asymmetric and symmetric behavior by means of the coefficient C = Vijk xi xj xk as For C = 0 C 0 Symmetric Bifurcation Asymmetric bifurcation This is not the end of the road. where ( )’ = For V’i xi = 0 V’i xi 0 Bifurcation state Limit state V/ . The main limitation of Koiter’s analysis is the quality of the quantitative predictions that can be obtained from perturbation analysis: for shells. for computational purposes. equilibrium states for which Vij qj > 0 are said to be stable.

and is the load potential. A classification scheme identical to Koiter’s but based on dynamic (rather than static) analysis has been formulated by Huseyin 19. and this task was completed by Chilver. Next. and considering stationary energy leads to the condition  . The explanatory part of Croll’s theory is based on two postulates 6: (1) Systems that may have changes in their membrane resistance are likely to show significant geometric nonlinear response. On the other hand. the only alternative for classification is catastrophe theory. The theory also classifies cases in which buckling modes couple at the critical state leading to compound critical states and mode interaction.2.2 Croll’s reduced energy theory Perhaps the structural application for which buckling problems are most crucial are shell problems. (2) Practically significant post-buckling loss of stiffness can only occur when the fundamental equilibrium path contains a significant contribution from membrane energy. Thompson. but it makes no attempt to produce a new classification. Here the main point was the discovery that small imperfections (either in the geometry or in the loads) destroy a bifurcation behavior leading to a nonlinear fundamental path with a different critical state. lower bounds to buckling into a particular mode are given by an analysis in which the membrane energy of the shell is eliminated. and this reduction may be significant for shell problems. but its success in elastic stability analysis has been very limited. where N ijε ij is the membrane energy and M ij χ ij is the bending energy. and Supple at University College London and elsewhere in England. Koiter’s success has been highest regarding its power to explain the phenomena of buckling of shells. j = 1. 2. systems with changes mainly in their bending resistance do not show highly geometric nonlinear response./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** The importance of this is enormous: one can classify all possible solutions and be confident that only one of them will occur. According to this. Substitution of each term in the energy V. For unstable post-critical behavior this means that there is a reduction in the critical load. which is more qualitative in nature and allows for multiple load parameter systems. this is both an explanatory theory and a predictive theory. Finally. for which high imperfection-sensitivity has been observed. χ ij = λχ F ij + χ ij (6) where (-)’ represents the linear contribution to the variable. As in the case of Koiter. Croll’s theory led to the reduced energy approach as a predictive tool to carry out the computations. The total potential energy is written in the form V= 1 2 (N ε ij ij + M ij χ ij )dA − ψ (5) for i. each field is represented in terms of the fundamental path (-)F in the form ′ ′′ ′ ′′ N ij = λ N F ij + N ij + N ij . while (-)’’ is the nonlinear contribution. Croll’s approach emerges from the theory of Koiter and uses it as a theoretical framework to obtain approximate values of critical loads and mode shapes for shells. ε ij = λε F ij + ε ij + ε ij ′ ′ M ij = λ M F ij + M ij .

since all you need is brute-force computer capabilities. There are many reasons to be  . while the second one leads to the loadgeometry matrix. simple in their overall geometry. The modifications consist in the elimination of the membrane energy contribution to the critical state. The eigenvalue buckling problem starts with the computation of the pre-critical states uF from the condition of equilibrium K uF + p = 0 F (8) Next. 3 and the computational buckling approach is well presented by Bushnell 5. Nonlinear algorithms are highly specialized and efficient for the computation of an equilibrium path. This is present in most general-purpose packages. such as ABAQUS. is the eigenvector (buckling mode) and c is the eigenvalue (critical load).0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** δ 2V = 1 ′′ ′′ [( N ijε ij + M ij χ ij ) + λ ( N F ijε ij + N ijε F ij )]dA 2 (7) The first parenthesis leads to the stiffness matrix. ALGOR. so that the new approach eliminated those terms from the eigenvalue problem in order to obtain a lower bound to the buckling loads. Notice that this is a tool of analysis and not a phenomenological theory. A good text in this field is Belytschko et al. equation (9) can compute bifurcation from the linear fundamental path of equation (8). and others cited in [6]) investigated the contribution of each individual term into the energy. but it does not help in the understanding of the computed results. the stress field ( ) is evaluated. and found those that contributed to stabilize the shell and those that were de-stabilizing. It tells you how to compute. 3. Many engineers view this approach as a heavenly solution to all problems.3 Nonlinear computational mechanics There is a whole branch of mechanics dealing with the computation of nonlinear problems. Basically. KG the load-geometry matrix. This does not mean that the membrane energy is eliminated from the fundamental path. Perhaps the most significant behavior of a tank is its buckling strength: a tank may easily loose its geometry. 27]. The consequences of this are that we may now compute bifurcation buckling of a shell but with a modified version of the total potential energy and obtain good estimates (both quantitative and qualitative) of the true buckling resistance. The erosion of a critical state due to the coupling of modes and imperfections could only occur because some of the stabilizing contributions were lost. 2. so that no classification or explanation is attempted. [2. and very costly. PRACTICE: THE COMPUTATIONAL BUCKLING ANALYSIS OF THINWALLED STEEL TANKS Steel tanks used to store oil and petrochemical products are very thin shells. Croll and co-workers (see Refs. The results of this first stage are carried into a second stage to solve the eigenvalue problem [K + λ K G (σ F )] φ = 0 (9) where K is the stiffness matrix. DIANA … The main purpose is to develop efficient algorithms to carry out the computation of equilibrium paths. ADINA. There are two main tools incorporated into this class of predictive approaches: nonlinear analysis and eigenvalue analysis.

The stakeholders are the owners of the tanks. thus we face here a problem of efficient use of the available software. and they show the same trend: if the tank is filled with liquid up to 50% of its height. Koiter’s predictive theory has been applied by Jorgensen 22 to evaluate isolated buckling loads and modes. 4. This form of buckling is usually associated to operational problems of a plant and is frequently found in practice. bifurcation-buckling analysis. the increase in critical pressure is of only 3 to 5%.012m at the top. L=24m. and finally there is a huge loss for each day a tank does not operate. while the results for vacuum 22 correspond to a simply-supported condition at the top. a damaged geometry triggers imperfection-sensitivity. 8 Normalized critical pressure 6 Wind pressure 4 2 0 0 Vacuum 0. the shallow cap itself is not part of the model (the finite difference discretization does not allow for a roof) and has been substituted by simply supported boundary conditions. Flores and Godoy 10 considered wind pressures on open-top tanks to address the influence of a liquid on the critical wind pressures. This is of great importance both from the handling of tanks in practice and also for what studies are worth doing. and nonlinear dynamic analysis. The tools are available. environmental protection agencies.5 Liquid level 1 Figure 1. In an independent study. Tanks are designed by engineering firms with standard finite element computer packages and medium to low level engineering knowledge about the mechanics of the problem. and a wall thickness tapered between h=0. In this section we review some examples of studies carried out with different theories in order to illustrate what phenomena may be identified with the models. Influence of liquid level on the critical pressures of tanks. and the insurance companies. The results for wind pressure are for an opentop tank 10. The information in Figure 1 does not tell anything about  . Buckling of tanks is a field in which there is no chance of finding appropriate analytical closed-form solutions so that computational mechanics is a must. What features of real tanks are included in the analysis? A steel cylinder with R=28m. and imperfection sensitivity under pressure./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** concerned about such uninvited changes in the geometry: a floating roof may not slide along the side walls. ELEMENTARY BUCKLING ANALYSIS FOR PRESSURE-INDUCED INSTABILITY Vacuum pressure is a static load produced during the process of emptying a tank with insufficient venting. The basic tools of analysis are linear/nonlinear static analysis. Results of the influence of the liquid on the critical pressure (normalized with respect to an empty tank) are shown in Figure 1 for the two loading conditions. and the results are also plotted in Figure 1.032 at the bottom and h=0.

the response is computed for several values of the load parameter . and we need to employ the criterion due to Budianski and Roth 4. Figure 2. However.50 and = 2. Such studies are reported by Jorgensen using Koiter’s perturbation analysis 22.0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** other changes that may have occurred due to the liquid in the postbuckling behavior. as in most works in this field 10. Model of the tank with fixed roof and rafter supports on the roof. for = 2. 5. 18 for dynamic buckling. Here we consider the wind-tunnel pressures evaluated by MacDonald et al. This pressure distribution is applied as a rectangular impulse (constant amplitude in time) with 3 sec duration. or in the imperfections sensitivity. The tank itself has a cylindrical part. which is the scalar that multiplies the pressure distribution shown in Figure 3. The space variation of pressures in the cylindrical part of the tank is constant in elevation and variable around the circumference.515 we get a qualitative  . and the finite element discretization with ABAQUS 17 employs about 12.51. 11.600 shells elements. but for the pressures acting on the roof there is no clear indication as to what should be assumed. STATIC OR DYNAMIC BUCKLING? Wind acting on a tank produces a pressure variable both in time and space. For = 2. shown in Figure 3. For three-second gusts. an example of the nonlinear dynamic response of a tank under increasing values of step load is shown in Figures 2 to 6. 25. have performed such studies previously for open-top tanks and silos. The main questions regarding the computational modeling of wind-load buckling may be posed as: Is this a static or dynamic stability problem? How important is the influence of the roof? What is the influence of previous damage on the buckling strength of a tank? The question about the importance of dynamic effects cannot be solved solely within the limitations of static theories. a fixed roof and rafters supporting the roof. The maximum pressure in the reference case is 1KPa acting on the windward meridian on the cylindrical part of the shell. as indicated in Figure 2. Notice that this problem is far more complex than the earlier investigations on simple cantilever cylinders. According to the criterion for dynamic buckling. The values of the pressures are scaled using the load parameter . the response shows small amplitude vibrations. of the order of the shell thickness. and the radial displacement is shown in Figure 4 for a specific location in the cylindrical part (node 8938). The nonlinear dynamic response has been computed using ABAQUS 17. Flores and Godoy 10. Three-second wind gusts are considered as the basic information for design.

00 0. while Figure 6 shows the vibrations for = 2. Instability occurs for = 2. 25). with blue deflections towards the inside and red towards the outside of the shell. at time t = 3 sec. for = 2. Each curve is for different pressure amplitude.55.0 0. As the load parameter is increased. The first figure shows a symmetric pattern of vibration. Nonlinear dynamic response for 3 sec wind gust assumed constant.18 -0. Pressure distribution assumed on the roof of the tank (after MacDonald et al.510 2.5 Node: 8938 2. This is an indication of a non-proportional change in the response due to a small increase in the load parameter. and would only occur if the shell is suddenly slammed by the pressure = 2. however. but the process starts earlier: for example. The second figure illustrates a break in the symmetry.12 -0. The vibration pattern due to the nonlinear dynamic analysis.5 Time [sec] 2. provided the nonlinear vibrations have a nonzero component in this DOF. the shell exhibits instability. on the other hand.4 sec./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** change in the response. Figure 3.16 -0.  .520 2. due to coupling of modes in an advanced postbuckling state.515.515 2. Figure 6.550 2.02 0. with large amplitude vibrations starting at about t = 2. Figure 5 represents the onset of dynamic buckling. a similar behavior is obtained if other degrees-of-freedom (DOF) are considered.14 Displacement [m] -0.55. the number of small-amplitude oscillations is reduced before the shell buckles. while for = 2.0 3. as this is the first stage for the lowest load for which instability occurs.0 2. large vibrations start at t = 1 sec. is shown in Figure 5.04 -0.10 -0. for = 2. Notice that Figure 4 has been plotted for a specific point of the shell.06 -0.0 1.5 sec. -0.5 3. As the load parameter increases.515.600 Figure 4.08 -0. at time t = 2.55 sec.500 2. is an advanced vibration pattern.60 the time is reduced to approximately t = 0.52.5 1.

Nonlinear dynamic analysis. at time t=2. geometrically nonlinear analysis of the shell with the same pressure distribution but monotonically increasing in time. Figure 6. a simpler approach has been attempted.55 sec.0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** Figure 5. deflected pattern for = 2. Nonlinear dynamic analysis. using a static.  .515. since several cases of dynamic nonlinear response have to be solved. Thus. Static models. at time t=3 sec. The above procedure is time consuming and computationally expensive.55. deflected pattern for = 2.

Static nonlinear analysis.4%. The reasons for the close agreement between nonlinear static and dynamic solutions are to be found in the natural frequencies of the shell 26. the period of vibration of the shell is 0. as illustrated in Figure 8. The path is nonlinear and reaches a maximum at = 2.20 0.3 %).6133.36 sec. For the present configuration. Bifurcation buckling analysis has also been investigated for this problem using ABAQUS 17 .55. so that not only the maximum load but also the buckling shape is well predicted by a nonlinear static analysis of this problem. Here the same DOF considered in Figure 4 has been chosen.002 -0./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** 1.55.000 -0.40 0. the difference being of only 1. A second mode (Figure 10) is identified at almost the same value = 2. Figure 8.60 0.30 0.90 0.10 0. Figures 5 and 8 show similar patterns of behavior of the shell. Static nonlinear analysis.50 0.6107 (see Figure 9). that is a value slightly higher than the dynamic buckling pressure identified in the nonlinear dynamic study ( = 2.005 Displacement [m] N: 8938 Figure 7.55. According to Koiter’s classification theory.80 Fraction of applied load 0.001 -0. thus far from the excitation considered of 3 sec.  . deflected shape for = 2.515).00 0. and the results lead to a lowest critical load = 2. The equilibrium path for load parameter versus a DOF of the structure is shown in Figure 7. A symmetric shape is found at the onset of instability. The maximum load is = 2. this is an unstable symmetric bifurcation.00 0. There is a small difference between static nonlinear and bifurcation buckling (2.70 0.004 -0.003 -0.

Koiter’s predictive approach has not been used. first mode. Koiter’s explanatory theory was present throughout the study: all explanations are possible thanks to the framework provided by the stability theory due to Koiter. Bifurcation buckling. This is reasonable. = 2. However.  . Second mode. especially for the case of a tank with a roof. = 2. Bifurcation buckling. In the previous studies. It is only now that one can confidently employ static predictions for this kind of environmental action.0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** Figure 9.6133. INITIAL VERSUS ADVANCED POSTCRITICAL BEHAVIOR The static response in the previous section was limited to the close vicinity of the critical state. but in more flexible structures one may want to consider the behavior along the postcritical equilibrium path. A lesson learned is that this is mainly a static problem. Figure 10.6107. but a nonlinear computational mechanics approach has been adopted. 6.

IMPERFECTION SENSITIVITY One of the crucial aspects of Koiter’s theory was the inclusion of imperfection sensitivity into the field of buckling. as reported in a recent paper 14. so that the geometric nonlinearity is not accounted for. R=19m. The equilibrium paths for several load imperfections computed with ABAQUS are shown in Figure 11. at which a snap occurs and the structure changes from positive to negative displacements.05 0. In the previous case of a tank with a stiff roof. But for open-top tanks.00 -0. But recent studies 15. h=0.50 2. Let us plot again the results but now with a critical load normalized with respect to the tank full of liquid in the vertical axis. however. with out-of-plane deflections of the order of five or more times the settlement in the vertical direction./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** Let us consider an open-top cylindrical tank. Height=7. The right branch (positive displacements) decreases until a certain value.50 1. Traditionally.20 -0.01m. 7.05 0.10 -0. tanks are constructed near a shore with poor soil conditions and may face support settlements.6m.16 have shown that the nonlinear behavior of the shell plays an important role in the response. A lesson learned is that bifurcation is highly relevant in this case and may provide good estimates of the buckling capacity. The shape of the diagram is not surprising: it is a typical diagram with a singularity at the origin (1 in this case) and which is highly sensitive to small changes and  . the shell may display high sensitivity. This is a feature that could not be predicted by Koiter’s initial poscritical analysis. this problem has been viewed in the literature as a linear static behavior. and the liquid level in the horizontal axis. again under wind load (Figure 11). Often. a more elaborate behavior occurs: The left branch (negative displacements in the figure) tends to decrease. a symmetric behavior is computed similar to what was observed in Figure 7.50 0. according to Koiter’s classification theory. as the deflections increase in the postcritical path. and an unstable symmetric bifurcation behavior is identified. 2.25 -0. the shell under wind pressure is almost insensitive to imperfections. since it occurs far from the critical state and is due to advanced nonlinearity in the response. showing an unstable postcritical behavior. How this can permeate into practice for other problems can be illustrated with respect to the results of Figure 1.00 0. What we now see is a diagram of sensitivity of the critical load to changes in the level of liquid inside the tank. Nonlinear equilibrium paths for an open-top tank with initial load imperfections 20. An advanced postcritical behavior is also required in other buckling problems of steel tanks. In the close vicinity of the bifurcation.00 Load Factor 1.00 0.15 -0. A key feature in Koiter’s explanatory theory is the sensitivity of buckling loads to the influence of small imperfections in some shells.10 U2 Figure 11.

for which there is a plateau.2 1 0. the reduced energy lower bound approach has been used with an analytical solution of the buckling modes. modeled by the classical lamination theory developed for composite materials 21. although this approach is physically sound and convenient. 1. so that no real advantage can be gained by filling the tank with say half its contents. It is possible to look for this form of diagrams only within the explanatory theory of Koiter.6 0.9 0. For a symmetric laminate.8 0. In most cases.95 Liquid level 1 wind vacuum Figure 12.0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** almost insensitive to large changes. and [A. Furthermore. } are the membrane strains and changes in curvature of the shell. Data as in Figure 1.4 0.8 0. In practice this means that little contribution can be expected from the liquid to stabilize the tank. in cases of more complex structures. and we are left with the uncoupled membrane and bending contributions.2 0 0. B. An alternative approach would be to consider a homogeneous steel shell as if it was made with a laminated material. Only if the engineer is ready to develop his/her own computer program for the lower bound analysis. Normalized critical load 1. Sensitivity of critical load to a decrease in the liquid level filling the tank. D] contain the coefficients of the laminate. such as those shown in Figure 2 with compound shells and rafters on the roof. One of the advantages of  . then it may be employed with advantage over a full nonlinear analysis. it faces the shortcoming of not being easy to use in practice. The use of Croll’s theory in engineering practice is conditioned to the availability of a suitable predictive tool to carry out the analysis. In a laminated material. { . there are no analytical solutions available that could be used for the computation of the total potential energy. so that simple loading and boundary conditions can be taken into account. 8. The lamina is rotated to the global directions of the laminate. Individual contributions from laminae are assembled into a laminate and define a constitutive relation of the form N A B = M B D ε χ (10) where {N. the bending-extension coupling vanishes (B = 0).85 0. Thus. the stiffness of an individual lamina is computed as in plane stress behavior. M} represent the membrane and bending stress resultants. POSSIBILITIES OF IMPLEMENTATION OF CROLL’S APPROACH Croll’s approach has not been mentioned in the sections about applications mainly because it is more a predictive tool than an explanatory theory.

Consider again the eigenvalue analysis (equations 8 and 9). ABAQUS. CONICET (Consejo Nacional de Investigaciones Científicas y Técnicas). The lower bound approach requires the computation of the fundamental equilibrium path (usually a linear path in equation 8) using the full energy contributions to the stiffness of the shell. R. C. so that for practical purpose it may not be convenient to speculate with the liquid level as a positive factor in the event of a hurricane. Jaca and J. However. Thus. one needs to work in two steps. But for the eigenvalue problem (equation 9). steel tanks to store oil and petrochemical fluids. allows the use of steps in an eigenvalue analysis. the results show that the problem behaves essentially as a static one. This indirect approach may deserve closer attention to model the reduced-energy contributions in which some of the energy contributions are dropped. Virella. Simple parametric studies of bifurcation buckling may be enhanced by considering them with respect to the imperfection-sensitivity of shells. G. however. some possibilities are indicated in this paper that may help to turn this into a practical tool: The use of classical lamination theory for the solution of shells made with homogeneous materials. To understand the behavior of shells under wind gusts it is necessary to enlarge the theoretical frame and consider dynamic buckling./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** this indirect definition of the material is that now one can perform separate evaluations of the individual terms of the membrane and bending contributions by adequately controlling the individual coefficients of the constitutive relations (10). We have not developed these ideas in detail at present. The framework of explanatory theories and predictive theories is considered in practice for the analysis of thin-walled. its use has been limited to specific-purpose programs not available for commercial use. ways to implement it into standard finite element codes should be found. but does not easily allow for changes in the properties from one step to the next. and that it is the power of explanation in Koiter’s theory that allows to make sense of results. CONCLUSIONS This paper stresses the importance of explanatory theories for the computational buckling analysis of shells. however. and the National University of Córdoba. and inertia forces de not play an active role because the frequencies of excitation are far from the natural frequencies of the shells. In this case. The first author acknowledges the valuable contribution of F. However. we need separate computation of K and KG. Rather than deleting terms in the energy.  . This shows a high sensitivity. it is our feeling that if the reduced energy approach is to be adopted in practice. This work was supported by Agencia Córdoba Ciencia. Acknowledgements. It is shown that the algorithms of nonlinear computational mechanics that are implemented in most general-purpose finite element programs are necessary but not sufficient to perform adequate instability studies. Koiter’s theory becomes a good theoretical framework for the analysis. one could delete coefficients in the constitutive relation. As a predictive tool. with different shell properties in each. a case of influence of liquid inside a tank as a stabilizing factor can be seen as the imperfection-sensitivity of tanks full of liquid with respect to a decrease in the liquid level. Flores. 9. Croll’s reduced energy approach has suffered the same limitations in practice. for example. several authors have implemented Koiter’s perturbation analysis in conjunction with finite elements for the analysis of shells. This section outlines a possibility of using laminate elements to account for the energy contributions of basically homogeneous shell structures. Using a general-purpose finite element program.

W. Godoy.  . Sosa. Hibbitt.G.G.G. Int. Journal Engineering Structures. [4] B. 13(5).A.0(&20  ± )LUVW 6RXWK$PHULFDQ &RQJUHVV RQ &RPSXWDWLRQDO 0HFKDQLFV ***************************************************************** REFERENCES [1] E. Barbero. pp. vol. Godoy.A.D.C. Int. “Buckling of short tanks due to hurricanes”. J. Elsevier Applied Science. in Dynamic Stability of Structures. [8] F.G. Belytschko. Theory of Elastic Stability: Analysis and Sensitivity. 20(8). 410-417 (1995). Dordrecht (1985). Godoy. Journal Numerical Methods in Engineering. Croll. American Academy of Mechanics. [21] R. “Deflections of thin-walled storage tanks with roof due to localized support settlements”. Introduction to Composite Materials Design. vol. Taylor and Francis. 31 (3-4). “Localized support settlement of thin-walled storage tanks”. ABAQUS/ User’s Manual. “Sobre el sentido del progreso en la teoría de estabilidad elástica”. 213-222 (1991). pp.G. Int.A. [13] L. Mechanics. vol.G. B. Flores. Philadelphia. New York (2000). pp. Ivaordhoff. [20] R. 33. [19] K. Journal Numerical Methods in Engineering. Nonlinear Finite Elements for Continua and Structures.A. “Instability of shells of revolution using ALREF: Studies for wind loaded shells”. Paraná.M. Sosa. “Forced vibrations of silos leading to buckling”. pp. New York (1975).A. Godoy. Flores and L. A. Ed.A. [14] L. Karlson and Sorensen. Hemisphere Publishing. “Elastic postbuckling analysis via finite element and perturbation techniques. Karlsson and Sorensen. Pergamon Press. K. [7] F. Godoy and E. “Dynamic buckling of elastic structures: Criteria and estimates”. Vibrations and Stability of Multiple Parameter Systems. pp. [17] H. 1997. Seoul (2002). Philadelphia.J. pp. 224 (3). [5] D. pp.G. PA (1999). Flores. [15] L. [11] F. “Dynamic stability of structures”. Thin-Walled Structures. 7-41 (1967). Herrmann. Rhodes. Part II: Application to shells of revolution”. In: Advances in Structural Engineering and Mechanics. submitted for publication (2002). [18] N. Wiley. 83-106 (1967).A.G. Pergamon Press. 331-354 (1992). Int. Elsevier (1979). Conf. pp. Martinus Nijhoff. “Elastic postbuckling analysis via finite element and perturbation techniques. 36. vol.J.A. Jones. “Shell buckling: A return to basic mechanics”.A. [12] L.I. Budianski. Huseyin. Flores and L. G. Argentina (2002). Inc. Part I: Formulation”. G. Godoy and F. Alphen (1978).1. Godoy. I South American Congress on Computational Mechanics. Herrmann. Ed. L. [2] R. Croll. 533-542 (2002). [16] L. “A design approach for unstiffened cylindrical shells under external pressure”. [6] J. Computerized Buckling Analysis of Shells.A. Ed. 431-454 (1999). “Analisis postcritico de tanques de pared delgada bajo cargas de viento”. [9] F. 30-52 (2002). Bushnell. J. Jaca. Taylor and Francis. Flores and L. Godoy and E. 752-760 (1998). Mechanics of Composite Materials. Hoff. Proc. pp.A. In: Buckling of Shells in Land.A. Liu and B. in Dynamic Stability of Structures. pp. [10] F. PA (2000). Flores and L. Journal of Sound and Vibration. Batista and J. Godoy. Godoy. in Applied Mechanics in the Americas. [3] T. Oxford.M. Godoy and F. Structural Engineering and Mechanics. vol. Hibbit. 1775-1794 (1992). Moran. Tecno Press.M. G. “Imperfection sensitivity of wind loaded tanks”. Flores and L. in the Sea and in the Air.

submitted for publication (2002). [25] P. Kwok. Cambridge University Press. Thin-Walled Structures. Delft. Point pressure measurements on isolated structures”. L. Holland (in Dutch) (1945). 165-188 (1988). 1(4). 60. Suárez. MacDonald.C.D. [24] W. [27] S. K.C. “Buckling and postbuckling characteristics of pressureloaded cylinders”. Journal of Wind Engineering and Industrial Aerodynamics. Yamada and J. Ph. “Influence of the roof on the natural frequencies of steel tanks”. pp. pp. Koiter.A. Jorgensen. 1967. The Methodology of Scientific Research Programmes. UK (1978). Thesis. ASME Journal of Applied Mechanics. D. vol.T.A. Cambridge. “Buckling behavior of a liquid storage tank”. J. vol. Godoy and L. Delft Institute of Technology. [26] J./ $ *RGR\ DQG ( 0 6RVD ***************************************************************** [22] F. There is an English translation by NASA. Journal of Sound and Vibration.  . Lakatos.A. 309-323 (1983). Croll. silos and tanks: I. I. Holmes. On the Stability of Elastic Equilibrium. pp. Virella. vol. [23] I. 290-299 (1993).G. 31. vol. “Wind loads on circular storage bins.E.

Sign up to vote on this title
UsefulNot useful