Você está na página 1de 88

SSD 2: Stability

M. Ahmer Wadee
8 November 2002

Introduction
Demonstrations of instabilities in structures
Single degree of freedom (SDOF) systems [1, 2, 3, 4]
Buckling, snap-through, bifurcations and imperfection sensitivity
Multiple degrees-of-freedom (MDOF) systems [1, 2, 3]
Diagonalized systems
Introduction to Mode Interaction
Instabilities of Compression Members [5, 6]
Ideal & Real Columns
Approximate methods
Instabilities in Frames
Instabilities in Beams [5, 6]
Lateraltorsional buckling (LTB)
Instabilities in Plated Structures [5, 2]

Bibliography
[1] J. M. T. Thompson and G. W. Hunt. A general theory of elastic stability. Wiley, London,
1973.
[2] J. M. T. Thompson and G. W. Hunt. Elastic instability phenomena. Wiley, Chichester,
1984.
[3] Z. P. Bazant and L. Cedolin. Stability of Structures: Elastic, Inelastic, Fracture and
Damage theories. Oxford University Press, Oxford, 1991.
[4] J. G. A. Croll and A. C. Walker. Elements of structural stability. Macmillan, London,
1972.
[5] S. P. Timoshenko and J. M. Gere. Theory of elastic stability. McGraw-Hill, New York,
USA, second edition, 1961.
[6] G. W. Owens and P. R. Knowles, editors. Steel Designers Manual. Blackwell, London, fifth edition, 1994.

1.1 Structural Failure


Material instability
Plasticity and plastic hinges
Combinations of fracture and fatigue (covered qualitatively in the Materials
course)
Geometric instability
Buckling (structure usually remains elastic)
Small deflection (linear) theory is not usually valid when structural failure is considered. The behaviour is primarily nonlinear: whether material nonlinearities (e.g. plasticity or large strains) or geometric nonlinearities (e.g. buckling) govern depends very
much on the structure in question.

1.2 Preliminaries
In this course we address structures under conservative static loadings.
Conservative means roughly that loads do not change magnitude or direction as the
structure deformsor more exactly the loads have a potential in the mathematical
sense of the word, i.e. the work done when a structure moves from position (1) to
position (2) is independent of the path taken from (1) to (2).
Most of the loads we use are conservative: typical nonconservative loads include hydrodynamic and aerodynamic loads, e.g. jet forces or wind loads.
We can then define the stability of a conservative static systeminitiallyin terms
of the equilibrium path, which is another name for the loaddeflection diagram. The
name is used as any point on the path is associated with equilibrium values of load
and deflection. At other points, off the path, no equilibrium exists (and the structure
is accelerating i.e. it is about to be dynamic). In general, there will be more than one
displacement component: if there are n components then the equilibrium path will be
a curve (or curves) in an (n + 1) dimensional space the so-called configuration space.
P
Equilibrium path

n=1

u (or u1 )
2

As a definition, we now say that any point on the equilibrium path is STABLE if, at
that point, the total potential energy (V ) is locally minimum. If V is not a minimum
then the system at that point is probably UNSTABLE or possibly (but highly unlikely)
in a condition of NEUTRAL STABILITY.
As a further definition, we define buckling as the dynamic process which occurs when
an equilibrium path loses stability or becomes unstable (usually because the load P has
just been increased). After buckling occurs, the structure can be said to have buckled.
We shall see that there are different kinds of buckling of greater and lesser severity.

1.3 Example: Bucket in a well


Let us consider an example of a system that exhibits stability phenomena. Fig. 1 a rotating disc of radius R has a rope tied to its circumference which lifts the mass m when
the load P is applied. As the load P moves the disc rotates by an angle measured
from the vertical is the degree of freedom.






Figure 1: Bucket in a well.
If we take rotational equilibrium about the centre of the disc O, we obtain
mgR P R sin = 0

and so simple manipulation gives

P =

mg
sin

(1)

If, initially, = 0 and three initial cases are considered as shown in Fig. 2, we can plot
the response of how the load P varies with the angle (Fig. 3)

Notes on equilibrium path


1. Cases (a) and (b) follow the path smoothly from their initial points up to infinity
as the load P increases monotonically.
3








(a) 0 = 3/4 (135 )

(b) 0 = /2 (90 )

(c) 0 = /4 (45 )

Figure 2: Three key cases for the bucket in a well.


2
1.8
1.6

C0

JUMP

P/mg

1.4
1.2
1
0.8

(c)

0.6

(b)

(a)

0.4
0.2
0

0.5

1.5

2.5

Figure 3: Loadrotation response or equilibrium path of bucket in a well.


2. Case (c) at & 0 (i.e. say = 45.01 ) we have too much weight for equilibrium.
The system therefore accelerates, a DYNAMIC process, similar to the Von Mises
truss or the buckling demonstrations.
At constant P we get a dynamic (non-equilibrium) shift from C to C 0 . It
arrives at C 0 and settles down to follow the curve for (a) after vibrating for a
short while.

Clues for pinpointing loss of stability


1. Multiple values of the displacement parameter (in this example its ) for given
values of load P . This is sometimes called the adjacent equilibrium criterion.
2. Falling curve at C implies negative stiffness. Negative stiffnesses should always
set alarm bells ringing for instability.
4

1.4 Total Potential Energy


Now let us think about this problem by using total potential energy as opposed to
direct equilibrium. Defining a datum of zero potential energy at the level of the centre
of the disc O (Fig. 1), the total potential energy V of the system is defined thus:
V =U +
where U is the gain in potential energy and is the work done by the load. The work
done is equal to the load P multiplied by the distance the load moves in the direction
of loadthis quantity is negative as the structure moves in the same direction as the
load. Therefore, it is more common to write down the expression for V thus
V = U P .

(2)

In the above example, the gained potential energy U is given by the distance the mass
m rises and is given by the vertical distance the load P moves:
U = mgR,
P = P R(1 cos ),
and so the total potential energy for this system is
V = mgR P R(1 cos )

(3)

Before we can use this function, two axioms need to be introduced which define equilibrium and stability for a structural system in terms of total potential energy:
Axiom 1 A stationary value of the total potential energy with respect to the degrees of freedom
is necessary and sufficient for the equilibrium of the system.
Axiom 2 A complete relative minimum of the total potential energy with respect to the degrees
of freedom is necessary and sufficient for the stability of an equilibrium state.

Notes on the axioms


1. Axiom 1 says:

V
= 0 for Equilibrium

2. Axiom 2 says: Vmin Stability.


3. Axiom 1 can be used to derive Newtons laws of motion, but there is no completely general proof of Axiom 2.
Returning to the example and applying the axioms, equation (3) can be differentiated
to obtain equilibrium:
V
= mgR P R sin = 0,

which, as expected, gives an identical result to equation (1):


P =

mg
.
sin

(4)

What about stability? For this, Axiom 2 states that the profile of the potential energy
V needs to be a minimum. To determine the characteristic of the stationary point of
V the equilibrium solutionwe look at higher derivatives of V , so:
2V
= P R cos ,
2

(5)

recall that for V to be minimumthe stability conditionthe second derivative of V


needs to be positive. If the second derivative is negative then V is a maximum and
therefore the system is unstable.
For the example, the valid range for is
= [0, ].
Equation (4) shows that P is always positive in this range. Therefore the second derivative depends entirely on cos which behaves thus:
(
> 0 for = [0, /2],
cos
(6)
6 0 for = [/2, ].
and so equation (5) gives:
2V
2

60
>0

for = [0, /2] UNSTABLE,


for = [/2, ] STABLE.

(7)

What about the case where = /2 radians? This is the point where the system
changes from being stable to unstable, in general it is called a singular point and we
shall examine it in detail later.

1.4.1

Rolling ball analogy

An intuitive way of visualizing why energy minima implies stable equilibrium is to


think about a ball on a physical surface in static equilibrium as shown in three examples in Fig. 4. Case (1) shows a ball on a local minimumperturbing the ball away
from its position by a small distance causes the ball to roll around about its original
position, implying that the original position was stable. Case (2) shows a ball on a
local maximumperturbing the ball away from the top causes the ball permanently
to leave its original position, implying that the original position was unstable. Case
(3) shows a ball where one side rises and the other side falls awayi.e. a point of
inflectionalthough this is called a position of neutral stability, it is actually unstable
as small perturbations would make it behave identically to case (2). For systems of
6



Equilibrium points

(3)

(2)

(3)

(2)

(1)

(1)

(a) Potential energy profile

(b) Ball on a profiled surface

Figure 4: The rolling ball analogy for stability for single degree of freedom systems.
Case (1) is a minimum and is therefore stable. Case (2) is a maximum and is therefore
unstable. Case (3) is a point of inflection and is termed neutrally stable but is in fact
unstable.
n degrees of freedom the potential energy profile becomes an n + 1dimensional surface. Systems with two degrees of freedom have the stability possibilities as outlined
in Fig. 5. Cases (1) and (2) are essentially the same as for the single degree of freedom
case. For systems with two degrees of freedom and more, however, we may have a
surface in which one direction is a minimum but one of the other directions is a maximum; this is a saddle point, it is intrinsically unstable and is shown as case (3) for a two
degree of freedom system.
(1) Local minimum:

(2) Local maximum:

(3) Saddle point:

Figure 5: The rolling ball analogy for stability for more than one degree of freedom.
Case (1) is stable, but (2) and (3) are both unstable.

Question Why use potential energy as opposed to direct equilibrium?


Answer The equilibrium approach by Newtons laws becomes very cumbersome for
multiple degrees of freedom whereas potential energy gives a simpler systematic
approach to check stability.

Elastic Systems

For the use of potential energy for determining stability of structural systems, there
must be a way of measuring the energy stored in the deformed structure. The stored
energy in a structure is called strain energy U and is defined locally in terms of a structural domain S with the stress and strain tensors ij and ij respectively:
1
dU = ij ij dS.
2
This is a rather complicated view of this issue. By assuming a small elementsuch as a
springcontains the energy stored in the structure, we can now simplify matters considerably. Consider a linearly elastic and isotropic bar of length L, cross-sectional area
A and Youngs modulus E as shown in Fig. 6(a). If this bar lengthens or compresses by
P
x

(a) Real strut: cross-sectional area A, Youngs Modulus E.

P
x

(b) Model of real strut: spring stiffness k = EA/L

Figure 6: Real and modelled struts. The model strut, made up of rigid links and a
spring simplifies the deformed structure to one degree of freedom which in this case is
x.
an amount x whilst under load P , then the strain energy stored can be determined by
the use of the following assumptions:
1. The bar may be assumed to be one dimensional and material volume is conserved.
2. Strains are small and so || = |x/L|.
1
U = kx2
(8)
2
where k is EA/L and is called the stiffness of the bar. With such a simple result we
can change the bar into rigid link elements where the elastic deformation is stored in a
separate linear spring of stiffness k as shown in Fig. 6(b).
A different way of finding the strain energy stored is by examining the equilibrium
path (Fig. 7); the energy stored is equal to the force multiplied by the displacement in
8



 



(a)

 !

" #$&')% (*+!,-.+

(b)

Figure 7: Strain energy stored in (a) linear and (b) nonlinear elastic springs.
the direction of the force, i.e. in this case it is equal to the area under the graph. This
principle can be extended to nonlinear springs, if P = P () is a nonlinear function in
which is the total deflection, the strain energy stored in the structure when = x is
Z x
P () d.
(9)
U=
0

Strain energy can also be found in beams in bending. Similar in sense to the axial
spring where P = k, bending moment M relates to curvature thus
(10)

M = EI

and the flexural rigidity EI is analogous to the axial stiffness k and the curvature
is analogous to the axial displacement . The work done in a small beam element of
length x is given:
1
U = EI2 x
(11)
2
and the exact expression for curvature for a beam displaced by y(x) is:
"
 2 #1/2
dy
d2 y
1
=
2
dx
dx

(12)

and so the total strain energy stored is an integral over the beam length L:
1
U = EI
2

L
0

d2 y
dx2

2 "

dy
dx

2 #1

dx

(13)

which is often linearized for small curvatures to:


1
U = EI
2

L
0

d2 y
dx2

2

dx .

(14)

Defining a general degree of freedom Q enables the energy function V to be stated:


V (P, Q) = U (Q) P (Q).

(15)

2.1 General Stability Analysis


For any system we can examine the stability of an equilibrium position by examining
the terms of a Taylor series expansion for the potential energy V at the equilibrium
position. For a one degree of freedom system:



dV
1 dn V n
1 d2 V 2
V (Q + ) = V (Q) +
+ ... +
+ ...
(16)
+
dQ Q
2! dQ2 Q
n! dQn Q
Axiom 1 states that

and so the change in V is


dV
= 0,
dQ Q





1 dn V n
1 d2 V 2 1 d3 V 3 1 d2 V 4
+
+
+. . .+
+. . . . (17)
V (Q+)V (Q) =
2! dQ2 Q
3! dQ3 Q
4! dQ4 Q
n! dQn Q

For V (Q) to be stable (or minimum) this series in equation (17) has to be positive for
any small perturbation . Thus the following observations can be made:

For a stable system the first nonzero term must be positive for any , i.e. it must
involve an even power of with a positive differential coefficient.
If the series is zero, all derivatives must be zerothis is the definition of neutral
stability.
All other combinations are unstable.
For example if:


d3 V
d2 V
= 0, but
>0
dQ2 Q
dQ3 Q

V is neither maximum nor minimum but for negative the dominant term in the series
also becomes negative which implies instability.
The analysis process is very systematic and a flowchart can be constructed as shown
in Figure 8. Also note:
if both

dV
d2 V
=
=0
dQ
dQ2

(18)

at a certain value of Q then the energy surface is locally flat; as we saw in the
bucket-in-a-well example, this defines a change in the systems stability or a singular point.
Linearizing deflections (e.g. saying sin ) would only allow the determination
of singular points and would not allow analysis of their stability.

10

Total Potential
Energy V
#?
dV
=0
dQ
"

Equilibrium:

Unstable
"

?
@ For any equilibrium position
@
#
2 @ >0
<
0
d
V
@

Calc.
Stable
@
dQ2
! @
"
!
@
@ =0
?
@
@

Unstable
"

 6= 0

Calc.
@

@
d3 V @
dQ3

@
@

=0

?
@
@

Unstable
"

@
d4 V @> 0

Calc.
@
dQ4
! @
@
@ =0
<0

Stable
"

?
@
@

Unstable
"

@
d5 V @
Calc.
@
dQ5
! @
@
@ =0
 6= 0

?
and so on, if and only if all derivatives=0
#?

Neutral
Stability
"

Ball on a perfectly flat surface.

Figure 8: Flowchart for checking stability of an equilibrium configuration.

11

2.2 Worked Example 1: Axially-loaded Rigid Cantilever


Figure 9 shows the system to be analysed for its stability.


Figure 9: Rigid cantilever supported by a rotational spring of stiffness c.

12

2.3 Examples of Stability Phenomena


This section defines different classes of singular points associated with single degreeof-freedom structural systems. The first singular point encountered in a system is
known as the critical point C.

2.3.1

Limit Point

Figure 10 shows a springlink model that exhibits a limit point. Initially = .




(a) Model

 

0.2

0.1

0.2

0.4

0.6
x

0.8

0.1

 1.2

(b) Equilibrium Path

Figure 10: Model exhibiting a limit point.


V = 2kL2 (cos cos )2 P L(sin sin )

(19)

Also known as a snap-through instability. In this there is a smooth transition from a


stable response to an unstable response.
Practical examples: tied arch and Von Mises truss.
13

2.3.2

Stable-Symmetric Bifurcation

Figure 10 shows a springlink model that exhibits a stable-symmetric point of bifurcation.




(a) Model

 

2.4
2.2
2O

1.8
1.6
1.4
1.2
1
0.8
0.6
0.4

0.2
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

(b) Equilibrium Path

Figure 11: Model exhibiting a stable-symmetric bifurcation.


V = 2c2 2P L (1 cos )

(20)

Also known as the Supercritical bifurcation. In such a system that is under controlled
displacement (rigid loading), the system can take increasing load after the bifurcation
C.1
Practical examples: columns and plates in axial compression.
1

Controlled loading may also be called dead loading

14

2.3.3

Unstable-Symmetric Bifurcation

Figure 12 shows a springlink model that exhibits an unstable-symmetric point of bifurcation.





(a) Model



0.8
0.7
0.6
0.5 O

0.4
0.3
0.2
0.1


1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

(b) Equilibrium Path

Figure 12: Model exhibiting an unstable-symmetric bifurcation.




p
1
V = kQ2 L2 2P L 1 1 Q2
2

(21)

Also known as the Subcritical bifurcation. In such a system that is under rigid loading, the system cannot take increasing load after the bifurcation C. The load carrying
capacity of the structure decreases with increasing displacementthe definition of an
unstable structure. This class of behaviour implies high sensitivity to initial imperfections.
Practical example: cylindrical shells in axial compression.

15

2.3.4

Asymmetric Bifurcation

Figure 13 shows a springlink model that exhibits a asymmetric point of bifurcation.









(a) Model



0.8
0.7
0.6
0.5 O

0.4
0.3
0.2
0.1

0.6

0.4

0.2

0.2

0.4

0.6

(b) Equilibrium Path

Figure 13: Model exhibiting an asymmetric bifurcation.



2


p
p
V = kL2 1 1 + Q P L 1 1 Q2

(22)

The main feature of this system class is the asymmetry of response: in the example
shown in Fig. 13, if the displacement after the bifurcation is positive the response is
unstable, if the displacement is negative the response is stable.
Practical example: asymmetric frames.
16

Imperfections

In reality structures are not perfect, there are imperfection that arise from the manufacturing processes such as members being initially not perfectly straight, residual
stresses, material defects and so on. We can extend our stability analysis further to
cover this class of problem.

3.1 Worked Example 2: Imperfect Axially-loaded Rigid Cantilever


Figure 14 shows the imperfect system to be analysed for its stability. It is a direct
extension of example 1.


Figure 14: Imperfect Rigid cantilever supported by a rotational spring of stiffness c.


Initially, = .
The following notes can be made.
The imperfect paths are asymptotic to the perfect case.
Stability of the imperfect paths can be checked by the second derivative of V for
both positive and negative .
The complementary paths are part of the overall picture, but have little practical
significance (see Fig. 15).

3.2 Imperfection Sensitivity


For appropriate classes of bifurcations we can construct extensions to the perfect cases.
17





QSR=TOUIVBR WXR$Y"Z%W"[
\5]
ZI^
_a``bdcfe g9hjiOk2iBbl
mon7pq5rqsut"qpv
KMLKON


 "!$#"%"
&'
0213547698%02: ;=<
>8%?3:A@B:=CD6%E2;$0F8%0<G8IH
( )*+-,/.*


(a) Step 1

(b) Step 2

(c) Step 3

Figure 15: Example loading history leading to complementary paths.


3.2.1

Unstable-Symmetric Bifurcations

Figure 16 shows the equilibrium diagram for the case shown in Section 2.3.3 if we

wyxz|{
0.8

DBO=OI
%FuG%2

0.7

0.6
0.5 O
0.4

=o$=j

2=XI%7F

O
0.3


1

0.8

0.6

0.2

} ~

0.1

0.4

0.2

0.2

0.4

0.6

0.8

Figure 16: Equilibrium diagram for an imperfect unstable-symmetric system.


assume that the spring is unstressed when Q = . This changes the strain energy to:
1
U = kL2 (Q )2 ,
2

(23)

with the work done remaining unchanged. For this case we can construct a graph
which plots the level of the limit point loads versus the size of the initial imperfection
 (Fig. 17). This is called an Imperfection Sensitivity diagram and it is the locus of limit
18


1.4

 
 

1.2
1O
0.8
O

O
0.6

0.4
0.2

0.3

0.2

0.1

0.1

0.2

0.3

Figure 17: Typical imperfection sensitivity plot for systems with unstable-symmetric
bifurcations. The value of a changes for different system geometries, for the example
of Fig. 16, a = 1.35.
points P/P C against initial . It shows that with the load increasing with an initial
imperfection, deflections grow nonlinearly until at some load (where P  P C ) the
structure becomes unstable. This type of structure is imperfection sensitive 2 .
A first-order approximation for the imperfection sensitivity curve is
P
1 a2/3
C
P

(24)

this has vertical tangents at  = 0, which implies a rapid loss of load carrying capacity for trivial imperfections. A well-known example of structural component that is
highly imperfection sensitive is the axially-loaded cylindrical shell where imperfections a fraction of the wall thickness of can remove 80% of the theoretical value of
P C.

3.2.2

Asymmetric Bifurcations

Figure 18 shows the equilibrium diagram for the case shown is Section 2.3.4 if we
assume that the spring is unstressed when Q = . This changes the strain energy to:
U = kL2

1+

i2
p
1+Q ,

(25)

with the work done remaining unchanged. In this case  > 0 gives the only limit loads
with negative values being stable, the system is therefore unreliable in that it is difficult
to predict the sign of the imperfections in advance.
2
Elastic systems undergoing a stable-symmetric bifurcation can be thought of as imperfectioninsensitive.

19

9;:=<?>A@B<?BCEDFEGIH#>JFKDLM


1.4
1.2


 

O
1
O
O

*,+-/.0+214351678+

0.8
0.6
O C
0.4

$
% &
0.8

0.6

'
( )
O

O!#"

0.2

0.4

0.2

0.2

0.4

0.6

0.8

Figure 18: Equilibrium diagram for an imperfect asymmetric system.


For this case we can also plot an imperfection sensitivity curve (Fig. 19). The first-order
approximation this time is
P
(26)
1 a1/2
PC
which tends to reduce the limit point load more than the symmetric case.

NOEN?P
O
2
O
O

1O

O
O
O
0

0.1

0.2

0.3

Figure 19: Typical imperfection sensitivity plot for systems with asymmetric bifurcations. The value of a changes for different system geometries, for the example of Fig. 20,
a = 1.70 for this first order approximation.

3.2.3

Limit Points

Figure 20 shows the equilibrium diagram for the case shown is Section 2.3.1 if we
20

 

O
0.2

O
O
O

0.1

O
O
O
0

0.1

0.2

0.3



0.4

0.5

0.1

Figure 20: Equilibrium diagram for an imperfect limit point system.


assume that the spring is unstressed when = + . This changes the strain energy to:
U = 2kL2 [cos( + ) cos )]2 ,

(27)

with the work done remaining unchanged. For this case we can also plot an imperfection sensitivity curve (Fig. 21).

  
O

1.4
O
1.2
O
1O



O
0.8
O
0.6

0.4
0.2

0.2

0.1

0.1

0.2

Figure 21: Typical imperfection sensitivity plot for systems with limit points. The value
of a changes for different system geometries, for the example of Fig. 20, a = 2.80.
There is no real marked effect on behaviour in this case, just the limit point loads vary
linearly with .

21

3.3 Summary
Actual physical structures do not exhibit bifurcations in realityimperfections always
distort the behaviour as discussed which is invariably asymptotic to the perfect case.
Nevertheless, the perfect cases are often easier to model/analyse and a very useful
guide to real (imperfect) behaviour.

22

Strut & Column Buckling

Consider a simply-supported strut with length L and flexural rigidity EI loaded axially with a force P (Fig. 22). At buckling it assumes a profile (mode) y(x), where at x





Figure 22: Buckling strut.
we have a bending moment M introduced by the mode shape eccentricity,
M = P y.

(28)

Bending theory states: M = EI curvature, which from linear bending theory can be
written:
d2 y
(29)
M = EI 2 .
dx
Combining eqs. (28) and (29) we obtain the differential equation:
EI

d2 y
+ Py = 0
dx2

(30)

Rewriting this in the form:


P
d2 y
+
y=0
dx2 EI
gives the form of the equation for a free undamped oscillator, or the equation for simple
harmonic motion. It is useful to differentiate this equation twice with respect to x so that
two boundary conditions can be applied at each support.
d4 y
P d2 y
+
=0
dx4 EI dx2

(31)

The solution of this equation has the form:


y(x) = A sin x + B cos x + Cx + D

(32)

where A, B, C and D are constants dependent on the boundary conditions of the strut
and is defined:
r
P
=
(33)
EI
In the case shown in Fig. 22
y(0) = y 00 (0) = 0 B = D = 0
y(L) = y 00 (L) = 0 C = 0, A sin L = 0
23

(34)

where primes denote differentiation with respect to x, a nontrivial solution (A 6= 0), is


given by:
sin L = 0 L = 0, , 2, . . . , n where: n Z.
(35)
Therefore the condition for a nontrivial y(x) is
P =

n2 2 EI
.
L2

(36)

Unless the strut is restrained somewhere the lowest value of P (n = 1) is the practical
value for bucklingthe Euler load (PE )
PE =

2 EI
.
L2

(37)

In reality the equilibrium paths are not flat, they exhibit a stable-symmetric profile.
However, this stable nature is not significant until deflections get very large, 3 so they
are assumed to be flat and the critical load is a fair estimate of the failure load when
struts buckle.
However it is important to note that linearized deflections, which we have assumed in
the struts, can only give information on critical loads and say nothing for post-buckling
responses. This means that analysing systems which are asymmetric or unstable with
linearized deflections would always overestimate the real structural strength because of
their inherent sensitivity to imperfections. For systems that have a significant postbuckling stiffness (e.g. rectangular plates supported on all edges) assuming linearized
deflections can seriously underestimate the real structural strength.

4.1 Potential Energy Formulation of Euler Buckling


The following potential energy functional contains large displacement (small-strain)
expressions for strain energy and work done by load P of the elasticathe inextensible
strut:

s
 2 #1
 2 2 "
 2
Z L
dy
dy
1
1 1 dy dx .
1

P
V =
EI
(38)

dx2
dx
dx
0 2
This can be expanded as a power series (with primes denoting d/dx):


Z L


1
1 02 1 04
00 2
02
04
EIy
y + y + ...
V =
1 + y + y + ... P
dx
2
2
8
0

(39)

Small deflection assumptions allow us to truncate the series to first order (linearization):

Z L
1 02
1
00 2
dx
V =
EIy P y
(40)
2
2
0
3

In this case, eq. (12) becomes valid for curvature and the problem becomes that of the elastica (see
Section 1.3 of Thompson & Hunt 1973 or Section 2.7 of Timoshenko & Gere 1961).

24

To obtain the equilibrium equation of the structure, the stationary point of the potential
energy needs to be found. As this is in a form of an integral, the condition for an
integral to be stationary is that the first variation of the integral must be zero. If V is
expressed thus:
Z L
L(y 00 , y 0 ) dx
(41)
V =
0

where L is the integrand and is like a Lagrangian function from dynamics theory, we
can express the first variation of V (denoted as V ) thus:
V =
where

L
0



L
y 00

00

y +

L
y 0

dx.

1
1
2
2
L = EIy 00 P y 0 ,
2
2

and

(42)

(43)

L
= P y 0 .
y 0

(44)

d 0
d
y and y 0 =
y
dx
dx
and then integrate by parts twice we get the following:

(45)

L
= EIy 00 ,
y 00

If we also say that:


y 00 =

00

000

V = [EIy y (EIy + P y

L
)y]0

(EIy 0000 + P y 00 ) y dx.

(46)

So if V = 0 at equilibrium, the terms in the square brackets must vanish when appropriate boundary conditions are applied:
Simple supports (zero deflection and bending moment): y = y 00 = 0
Fixed supports (zero deflection and slope): y = y 0 = 0
Free (zero bending moment and shear force): y 00 = 0, EIy 000 + P y 0 = 0.
More importantly the integral vanishes when the integrand is equal to zero, namely
when:
EIy 0000 + P y 00 = 0 ,
(47)
which gives the equation of the Euler column. This analysis has used a technique called
the calculus of variations which provides the link between Newtons laws and potential
energy.

4.2 Effective Length Concept


With the solution of the governing ODE we could consider any boundary condition
combination by obtaining a solution of y(x). However, there is a more intuitive way

25

of analysing different cases using the concept of Effective length (L e ). Common cases
follow, the adjusted Euler loads take account of the effective lengths thus:
PE =

2 EI
.
L2e

(48)

1. Simple supports both ends:


2. Fixed both ends:

 
 
3. Cantilever (free fixed):



4. Sway column (fixed both ends but can also sway across):



5. Simple support fixed:

!#"%$'&(

There are many cases where end fixity lies between the above cases. This is especially
prevalent in frames. Fig. 23 shows two examples. When in doubt, take the largest effective length Le that gives the lowest P C , and always choose the correct mode carefully
to avoid the example pitfall in Fig. 24.

26




(a) Sway

(b) Non-sway

Figure 23: Sway and non-sway frames. For the effective lengths of the vertical members: (a) L < Le < 2L (between cases 1 & 3); (b) 0.5L < Le < 0.7L (between cases 2 &
5).




(a) System


  

(b) Correct mode





!

(c) INCORRECT

Figure 24: Choosing buckling effective lengths carefully. The central restraint is a pin.

27

4.3 Strut and Column Analysis


Sometimes the combination of boundary conditions and non-uniform cross-section
properties of columns are such that exact evaluation of critical loads is not possible
using the differential equation (31). However, good approximations can be made if
the system is formulated by total potential energy. The strain energy U is given by the
expression in eq. (14), but the work done by the load is thus:
Z L  2
1 dy
P = P
dx,
(49)
dx
0 2
which makes the expression for V:
 2 2
 2 #
Z L"
1
dy
P dy
EI
V =

dx
2
dx2
2 dx
0

For the standard column, substituting:


X
nx
y=
Qn sin
L
n=1

(50)

(51)

gives correct results since the chosen function in this case reproduces the correct buckling mode. Approximate results for P C can be obtained for approximate expressions
for y(x).

4.3.1

Rayleighs Method

1. Assume a one DOF form for y(x).


2. Substitute y into V given in eq. (50).
3. For critical equilibrium, set:
d2 V
=0
dQ2
Example
For a simply-supported strut of length L, assume:
y = Qx(L x)

Substituting the assumed y into V gives:


 2 2
Z L
Z L
1
1
dy
U=
EI
EI(2Q)2 dx = 2EILQ2
dx
=
2
2
dx
2
0
0
Z L  2
Z L
P L 3 Q2
P dy
P
(QL 2Qx)2 dx =
P =
dx =
2 dx
2
6
0
0
3 2
PL Q
V = 2EILQ2
6
28

(52)

(53)
(54)
(55)

2
Applying d V2 = 0 gives:
dQ

12EI
(56)
L2
which is around 22% too high, this is a pretty poor estimate as the assumed form for y
does not satisfy all the boundary conditions.
PC =

4.3.2

Timoshenkos Method

Note from eqs. (28) and (29) that when a strut buckles assuming small deflections:
d2 y
Py
=
2
dx
EI

(57)

hence for U we could write


1
U = EI
2

L
0

d2 y
dx2

2

P2
dx =
2EI

y 2 dx.

(58)

Example
Assuming the same form as in eq. (52) this time gives:
Z L 2
P2 2
P
P 2 L5 Q 2
U=
w dx =
Q2 x2 (L x)2 dx =
60EI
0 2EI
0 2EI
Z L
Z L  2
P
P dy
P L 3 Q2
dx =
(QL 2Qx)2 dx =
P =
2 dx
2
6
0
0
3 2
2 5 2
P LQ
PL Q
V =

60EI
6
Z

(59)
(60)
(61)

2
Applying d V2 = 0 gives:
dQ

10EI
(62)
L2
which is only 1.3% too high, this is a very good estimate even though the assumed
form for y does not satisfy all the boundary conditions.
PC =

4.3.3

Notes on the approximate methods

The approximation for y should always satisfy the geometric boundary conditions
to be useful. In the example, the geometric conditions at the boundaries are the
zero displacements. The zero bending moments at the boundaries are not satisfied by ythis is an example of a static boundary condition.
The origin of integration should be a position of zero bending moment.

29

Timoshenkos method relies on the expression for relating curvature to deflection. Timoshenkos method is a special case of Rayleighs method only really
useful for columns.
As y is generally modelled more precisely than its 2nd derivative, Timoshenkos
method turns out to be more accurate.
Rayleighs method can be used in a wide variety of applications including plates,
shells and even structural dynamics.
Extending the Rayleigh technique to more than one DOF makes the approximation better. Generally, this is called the RayleighRitz method.
It can be shown that an approximate energy method always arrives at upper
bound for P C .

30

Real Columns

For columns that are designed and used in practice, the Euler buckling load is only
part of the story. We also have to consider the effect of material failure, i.e. plasticity
or yielding. A idealized view is given in the graph in Fig. 25 where is the nondimen-

  

!#"%$
&!'("*),+-."*/0&!1,2
  
 
OP

3465#7%8
9;:=<0>-:@?A4!96B,C
DEFG H I*J KLDNM


Figure 25: Idealized graph of real column failure.
sional slenderness ratio of the section defined from the second moment of area I, the
cross-sectional area A and the sectional radius of gyration r. Thus:
I = Ar2 ,
= Le /r,
2 EI
2E
E =
=
.
AL2e
2

(63)
(64)
(65)

The expression in eq. (65) is the mean axial stress in the section during Euler buckling.
The slenderness is defined to incorporate the section property and the effective length
into one parameter: high values imply buckling, low values imply yielding failure.
The codes, however, also take account of initial imperfections from experimental data.

5.1 Column Formulae


5.1.1

Some preliminaries

We can plot E against to obtain a 2nd order hyperbola, the so-called Euler hyperbola.
The Euler curve is clearly an upper bound on column strength. We can normalize both
31

stress and slenderness with respect to the material yield stress y thus:
Stress : /y
where r =

Slenderness : /r

(66)

E/y .

For a stocky column, we do not expect buckling but rather squashing (yielding) of the
column. For low values of , this occurs when the mean axial stress reaches y . This
second upper bound on column strength is thus defined when /y = 1.0 for all /r .

5.1.2

The Real Column: PerryRobertson Formula

Object: to determine the value of P that causes the largest stress in the column to reach
y .
At any section we have a combination of (i) a mean axial stress ( = P/A) and (ii)
bending stresses corresponding to a moment (M = P y). The latter is compressive inside
the curve, and tensile outside. Since P/A is compressive, the most stressed location is at
midspan on the inside of the column (i.e. T in the Fig. 26), and this is where first yield

yi

yi + y

No Load

Load P

Figure 26: Imperfect column under axial load P .


will occur as P is increased. At T the stress T is given by
T =

Md P
+
I
A

(67)

where d is the distance from extreme fibre to the neutral axis. As M = P (


y + yi ) and

32

letting = P/A, we get


P (
y + yi )d
+
I

PA

=
d yi
+ yi +
AI
E



E
d
yi
+ .
=
r2
E

T =

(68)

This is after defining the magnification factor / (E ) which relates the deflection
owing to the load, y, to the imperfection yi . Putting = d
y i /r2 finally gives


E
T =
+1 .
(69)
E

If we put T = y and solve for the value of which causes yield at T we obtain:
s
2
y + (1 + )E
y + (1 + )E
y E
=
(70)

2
2

This is the PerryRobertson Formula for the mean axial stress to cause failure in a
column. This leads to a graph (Fig. 27) that rounds the corners of the curve in Fig. 25.
BS 449 used the PerryRobertson curve and factored it downwards using an allowable
stress a criterion:
a = 0.6failure .
(71)

  


!"#

i8jkCl:mn(opl:q>r
sStfkCuMnq>rFv
$&%(')*
+-,/.!0,213%+(45

687:99<;>=?A@>BC7:9<DFE@>GIHKJ9<LM7
NPORQSQMTUWVYX[Z\^]`_ba
c ZSZdfehg

Figure 27: PerryRobertson curve in comparison with the ideal column failure curve.
BS 5950 improved on 449 by combining experimental data with the PerryRobertson
curve with factors accounting for:
33

section shape, whether UC, UB etc;


residual stresses in rolled and welded sections;
the stocky column effect where failure & y A;
real support conditions.
It has been plotted for specific values of which reflect the influence of geometric
imperfections and residual stresses.
Values of from BS5950:
= 0.001d0 ( 0 ) 0
p
where = L/r and the limiting slenderness 0 = 0.2 E/y .
Values of d0 :

Curve a d0 = 2.0
Curve b d0 = 3.5
Curve c d0 = 5.5
Curve a represents rolled hollow sections with relative strengths decreasing as we go
from curves a to c.

5.1.3

Designing columns to BS 5950

Section 4.7 in BS 5950 Part 1 (2000 revision) is the relevant section:


Table 22 defines the effective length;
Table 23 defines which of the curves a, b or c to usedepending on section shape,
manufacture and so on;
Table 24 has the curves a, b and c tabulated against slenderness to obtain the
design strength.

34

Multiple Degrees of Freedom Systems

Up to now, we have mainly dealt with SDOF systems with only a hint of multiple
degrees of freedom (MDOF) systems when the multiple Euler buckling loads and the
RayleighRitz technique were mentioned. The study of nonlinear MDOF systems as a
whole is an immense topic and is still a topic of current research. In this module, we
shall study the following key aspects of MDOF systems:
General theory approach.
Diagonalized systems.
Elimination of passive coordinates.
Non-trivial fundamental paths.
Once these concepts are introduced, we shall examine briefly the phenomenon of mode
interaction mainly through a model of practical structural component.

6.1 General Theory Approach


To generalize the results for the SDOF systems and to provide a systematic procedure
to analyse the post-buckling characteristics of MDOF systems, we now present the
General Theory approach that was developed by Thompson et al at University College London in the 1960s.
Consider a potential energy function V :
V (Qi , P ) = U (Qi ) P (Qi )

(72)

assuming for the moment that:


1. system is of 1 DOF (i = 1),
2. fundamental equilibrium path is trivial (ie a solution Q1 = 0 exists for all P ).
The complete solution of a structural system is rarely possible in closed form. So assume V is expanded as a Taylor power series initially expanded about the unloaded
state O where P = 0 and Q1 = 0. So we now write:
1
1 O 3
V = V11O Q21 + V111
Q1 + . . .
2 
6

1 0O 2 1 0O 3
+P
V Q + V Q + ...
2 11 1 6 111 1
where
V11O

O
2 V
=
Q21
35

(73)

subscript 1 means a partial derivative with respect to Q1 and superscript O means


that the term is evaluate at the unloaded state. Also:
O
3 V
0O
V11 =
Q21 P

where the prime represents a partial derivative with respect to P . It is worth noting
the following points:

The constant and linear terms of the series are absent because the constant term
would vanish on differentiation for equilibrium. The linear terms because we
have assumed that the fundamental path is trivial. If these terms existed and we
differentiated wrt Q1 :
h 0
i
 O

V
0O
O
O
= V1 + V11 Q1 + . . . + P V1 + V11 Q1 + . . .
Q1

then under these circumstances we cannot get a solution Q 1 = 0 for all P :


V
0
= V1O + P V1 O = 0
Q1
when Q1 = 0.
Linearity in P : which means that we can expand V about a general point F on
the fundamental path where Q = 0 and P = P F without any approximation
involved, ie the potential energy may be written:
1
V = V11F Q21 +
2

1 F 3
V Q + ...
6 111 1

1
1
0F
0F
2
3
+ (P P F )
V Q + V Q + ...
2 11 1 6 111 1

(74)

which is the same as equation (73).


Rewriting equation (74 in the form of (73) and equate the coefficients:

 


1 F
1
F
2
3
F 0F
F 0F
V P V11 Q1 +
V P V111 Q1 + . . .
V =
2 11
6 111


1 0F 2 1 0F 3
V Q + V Q + ...
+P
2 11 1 6 111 1
we get:

V11F = V11O ,
and:
hence also:

(75)

O
F
V111
= V111
and so on
0

V11F P F V11F = V11O V11F = V11O + P F V11O


F
O
O
V111
= V111
+ P F V111
and so on
0

The new energy function expanded about the general state F describes the form of V
at any load P F .
36

6.1.1

Linear Eigenvalue Analysis (to find P C )

The critical equilibrium state is defined by:


(76)

V11C = 0,

which only requires the quadratic terms of the potential energy (small deflection assumption):
0
V11C = V11O + P C V11O = 0
PC =

V11O
0
V11O

(77)

Expanding V about the critical state where:


P = P C,

Q1 = 0,

V11C = 0,

we get:
1 C
1 C 3
Q1 + V1111
Q41 + . . .
V = V111
6
24


1 0C 2 1 0C 3
C
+ (P P )
V Q + V Q + ...
2 11 1 6 111 1
where

(78)

C
O
O
V111
= V111
+ P C V111

and:

6.1.2

V11C = V11O and so on


0

Post-Buckling Analysis

For information on the post-buckling response, we use a perturbation method. Express


the post-buckling path in parametric form. Assume:
P = P (Q1 ).
as shown in Fig. 28. However, the post-buckling path is also defined by the equilibrium
equation:
V
V1 =
=0
(79)
Q1
substituting P (Q1 ) into the equilibrium equation:
V1 [Q1 , P (Q1 )] 0

(80)

Plotting V1 against Q1 (see Fig. 29) we see that V1 is always zero for all Q1 hence
all derivatives of V1 wrt Q1 are zero also. This would not necessarily be the case if
we are dealing with solutions to an equation V1 = 0. Differentiating the identity (80)
sequentially wrt Q1 :
dV1
0 dP
(81)
= V11 + V1
=0
dQ1
dQ1
37

 


Figure 28: Post-buckling path: P (Q1 )



!#"$%'&)(*+"

Figure 29: Plot of V1 versus Q1 : zero for all Q1 .

38

Evaluating (81) at C we get no information since:


V11C = 0 : critical equilibrium
V1 C = 0 : no linear termstrivial fundamental path.
0

Differentiating (81) again wrt Q1 before evaluation at C:


dP
d2 V 1
0
00
= V111 + 2V11
+ V1
2
dQ1
dQ1

dP
dQ1

2

+ V1

d2 P
=0
dQ21

(82)

Evaluating (82) at C:
V1 C = 0 : no linear termstrivial fundamental path
0

V1 C = 0 : linear in P.
00

we obtain the expression for the slope of the post-buckling path at an asymmetric point of
bifurcation:
C
C
dP
V111
(83)
= 0
dQ1
2V
11

C
If V111
= 0, which it often is because of fundamental symmetries in the structure, the
slope is zero and so (82) must be differentiated again wrt Q 1 to give an expression for
curvature of the post-buckling path at a symmetric point of bifurcation:

C
C
V1111
d2 P
= 0 .
dQ21
3V11

(84)

The post-buckling path can thus itself be expanded as a Taylor series:


C
C
dP
1 d2 P 2
P P =
Q + ...
Q1 +
dQ1
2 dQ21 1
C

the series should be truncated after the first non-zero term.

39

(85)

6.2 General Theory Forms of V for SDOF Phenomena


The expressions of the Taylor series for V measured from the critical load P C for the
four distinct SDOF system types have specific features. Let q 1 be an incremental coordinate relative to the fundamental path,  be a generalized imperfection and dots
represent differentiation wrt .

6.2.1

Limit Point
h 0
i
h
i
1 C 3
V = V111
q1 + . . . + (P P C ) V1 C q1 + . . . +  V 1C q1 + . . .
6

6.2.2

Asymmetric point of bifurcation




i
h
1 0C 2
1 C 3
C
V = V111 q1 + . . . + (P P ) V11 q1 + . . . +  V 1C q1 + . . .
6
2

6.2.3

(86)

(87)

Symmetric points of bifurcation




h
i
1 C 4
1
0C 2
C
C

V = V1111 q1 + . . . + (P P ) V11 q1 + . . . +  V1 q1 + . . .
24
2

C
Stable system: V1111
>0
C
Unstable system: V1111
<0

40

(88)

6.3 Diagonalized Systems


Consider the example in Fig. 30, this is a 3 link model that is restrained by 2 rotational










 




 




Figure 30: Example system with 2 degrees of freedom: Q1 and Q2 .


springs of stiffness k at points B and C. To formulate the total potential energy V we
need the rotations of the pins at B and C, to facilitate this let us assume that the springs
are unstressed when Q1 = Q2 = 0. When the structure buckles (as shown in Fig. 30),
for the points A and B we have:
sin A = Q2 ,

sin D = Q1 .

and the angle made by BC to the horizontal () is


sin = Q1 Q2 .
The total rotations at B and C are thus:
B = A = arcsin(Q2 ) arcsin(Q1 Q2 ),
C = D + = arcsin(Q1 ) + arcsin(Q1 Q2 ),

(89)
(90)

Expanding arcsin x as a Taylor series we obtain


arcsin x = x +

x3
+ ...
6

and so the expression for the rotations to first order become:


B = 2Q2 Q1 + . . . ,
C = 2Q1 Q2 + . . . .

(91)
(92)

Note that the rotations are only to first order as this analysis is going to be confined to
critical equilibrium points only. The strain energy stored (U ) in the two springs are
1 2
1
U = kB
+ kC2
2
2

1  2
= k 5Q1 8Q1 Q2 + 5Q22 + . . . higher order terms
2
41

(93)

The work done by the load P is given by the distance the load has moved in the direction of the load
P = P (3L h)


q
q
q
2
2
2
= PL 3
1 Q1 + 1 Q2 + 1 (Q1 Q2 )


= P L Q21 Q1 Q2 + Q22 + . . . higher order terms .

(94)

Therefore the truncated total potential energy V is given by


1 
V = k 5Q21 8Q1 Q2 + 5Q22 + . . . higher order terms
2


P L Q21 Q1 Q2 + Q22 + . . . higher order terms .

(95)

Because of the Q1 Q2 terms being present, this system is called non-diagonalized. This is
because for critical equilibrium, which for SDOF systems required that:
2V
= 0,
Q2
for more than one DOF the condition becomes that the matrix V ij of the second derivatives (including cross derivatives) becomes singular. The matrix for second derivatives
is also called the Hessian matrix:

 F
V11 V12F
F
(96)
Vij =
V21F V22F

where the F superscript means on the fundamental path and the matrix elements are
as follows:
2V
2V
2V
V11 =
.
, V22 =
, V12 = V21 =
Q21
Q22
Q1 Q2
So for the critical point:
C

V11 V12C
C
=0
det(Vij ) = C
(97)
V21 V22C
where the C superscript means at the critical point. So for our example in Fig. 30 with
its potential energy given in Eq. (95), the matrix elements are:
V11 = 5k 2P L
V22 = 5k 2P L
V12 = V21 = 4k + P L

and so at the critical point, the matrix becomes singular thus:




C
C


(5k

2P
L)
(4k
+
P
L)
C
=0
det(Vij
) =
C
C
(4k + P L) (5k 2P L)

(98)
(99)
(100)

(101)

giving the following after a bit of manipulation:

(P C L 3k)(P C L k) = 0

(102)

P1C = k/L,

(103)

and so there are two separate critical loads:

P2C = 3k/L .

Because the DOFs were not distinct (V is not diagonalized), i.e. there was a cross-term
present, we cannot say much about the buckling modes.
42

6.3.1

Change of variables

Now let us approach this problem with a new set of degrees of freedom:
Q1 = u 1 + u 2 ,

(104)

Q 2 = u1 u2

or
u1 = (Q1 + Q2 )/2,

u2 = (Q1 Q2 )/2

Substituting these into the potential energy V (Eq. (95)) gives the following:

1 
V = k 2u21 + 18u22 + . . . higher order terms
2


P L u21 + 3u22 + . . . higher order terms

(105)

This time we have no u1 u2 cross term present and so there will be no terms other
than on the leading diagonal in the matrix Vij . The system has been diagonalized and
Eq. (104) is an example of a diagonalizing transformation. Completing the analysis we
can again analyse when the matrix Vij becomes singular, now the elements are:
(106)
(107)
(108)

V11 = 2k 2P L
V22 = 18k 6P L
V12 = V21 = 0
and so at the critical point:
C
det(Vij
)


(2k 2P C L)
0
=
0
(18k 6P C L)

giving the following directly:

(2k 2P C L)(18k 6P C L) = 0



=0

(109)

(110)

and as before, there are the two separate critical loads:


P2C = 3k/L .

P1C = k/L,

6.3.2

(111)

Notes

1. Diagonalization transformation has picked out special directions u 1 and u2 in the


(V, Q1 , Q2 ) space, for example P = P1C = k/L is purely in the u1 direction.
2. The new degrees of freedom u1 and u2 measure buckling mode amplitudes as
shown:
u1 = 0 Q1 = Q2


 
43

u2 = 0 Q 1 = Q 2



 

3. For a two degree of freedom non-diagonalized system, V is replaced by W (or


W, see later) and the Hessian matrix is thus


W11 W12
Wij =
W21 W22
and this is singular at the critical point
C
C
C
C
C
det(Wij
) = W11
W22
W12
W21
=0.

4. For a two degree of freedom diagonalized system, V is replaced by A (or A, see


later) and the Hessian matrix is thus


A11 0
Aij =
0 A22
and this is singular at the critical point
C C
det(AC
ij ) = A11 A22 = 0
C
and so either AC
11 = 0 or A22 = 0.

5. Finite element formulations give non-diagonalized energy forms with large numbers of DOFs.
6. Diagonalizing transformations always exist but may be difficult to find. General
theory of elastic stability works just as well for non-diagonalized systems but
needs more computational power.
7. If separate mode critical loads are identical (e.g. if P1C = P2C ) then the critical
point is called a compound bifurcation point and there are likely to be nonlinear
mode interaction effects.

44

6.4 Elimination of Passive Coordinates


Consider a system with n-degrees of freedom, a diagonalized potential energy function
A = A(ui , P )
with a trivial fundamental path
uFi = 0
for all P . This system has an associated Hessian matrix whose determinant:
P

n bifurcation points
Fundamental path: uFi

uj
ui
Figure 31: Equilibrium diagram with an MDOF system of n-degrees and n bifurcation
points.
det(Aij ) = AF11 AF22 AF33 . . . AFnn

(112)

is zero at the bifurcation points (critical states). Therefore we can define n critical states,
suppose the first critical point encountered on the fundamental path from the load
increasing from zero is m-fold, i.e. m values: AC
ii = 0 at point C in Fig. 31. In other
words, let us consider the case where there are m coincident bifurcation points at the
lowest instability point C.
Suppose, further, that generalized coordinates (DOFs) are numbered such that:
C
C
AC
11 = A22 = . . . = Amm

AC
m+1,m+1 6= 0

AC
m+2,m+2 6= 0

AC
m+2,m+2 6= 0

(113)

...

AC
nn 6= 0

The first step in a systematic post-buckling analysis is reduce the problem to m-degrees
of freedom which are active; modelled by a new m-DOF potential energy function:
A = A(ui , P )
45

(114)

where the m elements are associated with the vanishing stability coefficients: A C
11 = 0
and so on. Therefore the m-DOFs are termed the active coordinates u i . The remaining
n m of the DOFs are associated with non-vanishing stability coefficients which are
termed the passive coordinates which are henceforth written with Greek letter subscripts,
e.g. u . Hence:
A = A(ui , u , P )
(115)
and the associated equilibrium equations are:
Ai = 0 : m of these
A = 0 : n m of these
6.4.1

(116)

Perturbation scheme

Suppose that a new function is defined:


(117)

u = u (ui , P )

with m+1 independent variables and this is substituted into nm passive equilibrium
equations:
A [ui , u (ui , P ), P ] 0.
(118)
After differentiating each of the n m equations m + 1 times, this results in an ordered
series of solvable equations for the derivatives:
C
u
=0
ui
C
C
2 u
Aij
=
ui uj
A

(119)
(120)

and so on along with derivatives wrt P . This can be followed by u expanded as a


Taylor series in ui and P , the result sought for.
Hence substitute into A(ui , u , P ) to give equation (114) which can be achieved by the
single statement
A(ui , P ) A[ui , u (ui , P ), P ].
(121)

Differentiation of both sides and evaluation at C (or more generally anywhere on F


where AF 6= 0) leads to derivatives of A in terms of derivatives of A. For m = 1,
evaluating at F:
AF11 = AF11

AF111 = AF111
0

(122)

A11F = A11F

AF1111 = AF1111 3

n
X
=2

(A11 )
A

C
1 2 u
u1
u =
2 u21
46

2 F

(123)

Of course, when the above are evaluated at C, the only change is that:
C
AC
11 = A11 = 0.

The slope and the curvature of the post-buckling path are given by the following:
C
C
A111
dP
= 0
du1
2A11

C
C
d2 P
A1111
= 0 .
du21
3A11

(124)

(125)

For m = 2 with a more general and complete treatment, see Thompson and Hunt 1984.

47

6.5 Non-Trivial Fundamental Paths


Consider Fig. 32, suppose we have:
V = V (Qi , P )

(126)

where the fundamental path is non-trivial:


QFi 6= 0.
P

Qi = QFi (P )

PC
QFi (P )

qi

Qi
Figure 32: Non-trivial fundamental path.

The fundamental path (F) is often represented by a simple solution such as an initial
pre-buckling compression. For the purposes of post-buckling analysis, we assume it to
be given by:
Qi = QFi (P )
(127)
Define an incremental coordinate by:
Qi = QFi (P ) + qi

(128)

for all states on and off the fundamental path. Substituting this into V gives the nondiagonalized total potential energy W:


W (qi , P ) V QFi (P ) + qi , P .
(129)
W has the same form as the earlier defined V for a trivial fundamental path and can
be handled in the same way.

6.5.1

Example

Consider the system in Fig. 33, this is a 2 DOF system. Assume in this case that the
spring stiffnesses are equal:
k1 = k 2 = k
48





 



 



Figure 33: Example 2-DOF system (Q1 and Q2 ) to illustrate a non-trivial fundamental
path and the elimination of passive coordinates
The strain energy stored in the outer in-plane springs is simple, they both compress by
Q2 L. However, the compression in the central in-plane spring is not quite so obvious
it comprises the Q2 L component but the Q1 L deflection releases some of the compressive displacement (xc ) thus:



q
2
xc = L Q2 2 1 1 Q 1


(130)
1 4
2
= L Q2 Q 1 Q1
4
Formulating the total potential energy gives

2
1 4
1 2 2
1 2
2
2 2
V = kL Q1 + kL Q2 + kL Q2 Q1 Q1 P LQ2
2
2
4
1
3
1
1
= kL2 Q21 + kL2 Q22 kL2 Q21 Q2 + kL2 Q41 kL2 Q2 Q41 + higher order terms
2
2
2
4
P LQ2
(131)
where the first two terms in V term come from the vertical spring and the two in-plane
edge springs, the third term comes from kx2c /2, and the work done by the load is purely
the final term.
Noticing that the work done by load term is linear in Q2 , we can say that the system
has a non-trivial fundamental path in Q2 and a trivial fundamental path in Q1 .

49

Step (a): V W transformation


Differentiating with respect to Q2 and assuming the triviality of the fundamental path
in Q1 give the following:
QF1 = 0,
P
.
QF2 =
3kL

(132)

Introducing incremental coordinates q1 and q2 thus:


Q1 = q 1 ,
P
+ q2 .
Q2 =
3kL

(133)

Substituting these incremental coordinates into V gives the following expression for
W (the general MDOF potential energy function) after removing irrelevant terms (constant in q1 and q2 ):



1 4
1 2 2
1 2
2
2
4
q + q + ...
(134)
W = kL q1 + 3q2 2q1 q2 + q1 + . . . P L
2
3 1 12 1
A couple of things to note here:

1. V W has removed P from acting on Q2 and obliged P to act on q12 , thus we


have the possibility of a symmetric point of bifurcation, rather than a limit point.
2. As there is no term in q1 q2 in W we can say that the potential energy is diagonalized. Henceforth we write W as A and qi as ui , this in recognition of this
diagonalized form of the potential energy:



1 2 2
1 2
1 4
2
2
4
A = kL u1 + 3u2 2u1 u2 + u1 + . . . P L
u + u + ...
(135)
2
3 1 12 1
Step (b): Eliminate u2 as a passive coordinate
By expressing the potential energy A in the incremental coordinates u 1 and u2 and
expressing the potential energy in the general form:


1
1 0
1
1
1
1 0 2
4
2
2
2
4
A = A11 u1 + A22 u2 + A112 u1 u2 + A1111 u1 P
A u + A u + . . . (136)
2
2
2
24
2 11 1 24 1111 1

we can say:

2
AF11 = kL2 P F L
3
2
F
A22 = 3kL
AF1111
0

A11F

= 12kL 2P L
2
= L
3
50

(137)
(138)
(139)
(140)

As AF22 is greater than zero for all P , u2 is a passive coordinate. Thus we can introduce
the function A:
A(u1 , P ) = A [u1 , u2 (u1 , P ), P ]
(141)
and we can say:

2
AF11 = AF11 = kL2 P F L
3
2
0F
0F
A11 = A11 = L
3
F
3A2112
F
F
= 8kL2 2P F L
A1111 = A1111
A22

(142)
(143)
(144)

Step (c): Linear eigenvalue analysis


The quadratic terms of u1 in A enable the determination of the critical load P C :
3
C
AC
11 = 0 P = kL .
2

(145)

Step (d): Post-buckling analysis


2
2
C
AC
1111 = 8kL 2P L = 5kL

C
C
d2 P
A1111
5
= 0 = kL > 0 Stable
2
du1
3A11
2
1
P P =
2
C

 C
d2 P 2
u
du21 1

(146)

(147)

(148)

This finally gives us the first order approximation to the post-buckling path:
3
5
P = kL + kLu21 .
2
4

(149)

Step (e): Effective stiffness after bifurcation


The process of eliminating the passive coordinate u2 gives an expression for u2 in terms
of the active coordinate u1 :
d 2 u2
A112
2
=
= ,
2
du1
A22
3
 2 
1 d u2
1
u2 (u1 ) =
u21 = u21 ,
2
2 du1
3
51

(150)
(151)

P
q2
PC

Q2
Figure 34: Post-buckling stiffness
The effective stiffness after buckling (see Fig. 34) can be determined by returning the
post-buckling solution (Eq. (149)) to the original Q2 DOF:
Substituting (151) into (152) gives:

(152)

u2 = Q2 QF2 .

P
.
kL
Rewriting (149) in terms of Q2 gives the equation of the post-buckling path:
u21 = 3Q2

5
2
P = kL + kLQ2 ,
3
3

(153)

(154)

and the equilibrium paths are shown in Fig. 35. The effective stiffness is 5/9 times the


3
2.8
2.6
2.4
2.2
2
1.8
C
O

1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

0.2

0.4

0.6

0.8

 1

Figure 35: Equilibrium path of the example: P versus Q2 (load against end-shortening).
original stiffness. By inspection the system is stable: the load carrying capacity of the
structure continues to increase after the initial instability. This can be thought of as a
model of plate buckling.
52

6.6 Introduction to Mode Interaction


The subject of mode interaction or interactive buckling is an advanced topic of nonlinear
buckling theory. It is a subject of current research in steel structures and encompasses
all types of structural components: columns, beams, plates, shells and frames.
As opposed to the previous section where there were two DOFs where one was passive
and the other by implication was active, in 2-DOF structures undergoing interactive
buckling both DOFs are active. The following are common examples where buckling
mode interaction becomes an issue in design:
Thin-walled construction
Design of box girders
Design of plate girders
Built-up or reticulated columns
Ordinary or stiffened plates and shells
Lightweight sandwich construction
The detailed coverage of thin-walled construction is covered in the Plated Structures
elective module.
The interaction usually manifests itself when two critical modes are triggered at the
same load or near enough to each other such that the modes couple together to give a
thoroughly different response than predicted by individual modal analysis. A common
interaction is between overall (Euler) buckling and local (plate-type or short wavelength periodic) buckling. In this section we shall study an example of an idealized
model of a compression sandwich panel, which reproduces most of the major features
exhibited by the actual structure. This model was presented by Hunt in 1986 [1] and is
called the pilot model. Direct advancements on this model were made by Hunt and
da Silva between 198890 [2, 3, 4] and by Hunt and (M. A.) Wadee from 1998
Simoes
to this date [5, 6, 7, 8]

Bibliography
[1] G. W. Hunt. Hidden (a)symmetries of elastic and plastic bifurcation. Appl. Mech.
Rev., 39(8):11651186, 1986.
[2] G. W. Hunt, L. S. Da Silva, and G. M. E. Manzocchi. Interactive buckling in sandwich structures. Proc. R. Soc. Lond., A 417(1852):155177, 1988.
[3] L. S. Da Silva and G. W. Hunt. Interactive buckling in sandwich structures with
core orthotropy. Mech. Struct. & Mach., 18(3):353372, 1990.

53

[4] G. W. Hunt and L. S. Da Silva. Interactive bending behaviour of sandwich beams.


Trans. ASME J. Appl. Mech., 57(1):189196, 1990.
[5] G. W. Hunt and M. Ahmer Wadee. Localization and mode interaction in sandwich
structures. Proc. R. Soc. Lond., A 454(1972):11971216, 1998.
[6] M. Ahmer Wadee. Effects of periodic and localized imperfections on struts on
nonlinear foundations and compression sandwich panels. Int. J. Solids Struct.,
37(8):11911209, 2000.
[7] M. Ahmer Wadee and A. Blackmore. Delamination from localized instabilities in
compression sandwich panels. J. Mech. Phys. Solids, 49(6):12811299, 2001.
[8] M. Ahmer Wadee. Localized buckling in sandwich struts with pre-existing delaminations and geometrical imperfections. J. Mech. Phys. Solids, 50(8):17671787, 2002.

6.6.1

Pilot model of compression sandwich panel

Consider the model in Fig. 36 consisting of two columns of thickness t, length L and
Youngs modulus E connected to each other by a series of longitudinal springs which
represent the core material. The two columns represent the face plates of the sandwich
structure. Real core materials have shearing resistance, but this model has this idealization that reduces the complexity significantly. The longitudinal springs resistance
model is sometimes termed the Winkler approximation. The integrity of the overall
mode is included by insisting the ends of the panel tilt. From the figure it can be seen
that four degrees of freedom have been defined:
1. q1 : local buckling amplitude for the right-hand face plate
2. q2 : local buckling amplitude for the left-hand face plate
3. u3 : overall buckling amplitude for both face plates
4. U4 : end-compression in direction of the load

Potential energy function


To accumulate the potential energy, we need to identify the components of the strain
energy and the work done by load. Using Rayleighs technique to model each of the
local deflection components of the face plates from their original positions w n , the local
buckling of the face plates are modelled by:
L
ix
sin
i
L
L
ix
w2 = q2 sin
i
L

(155)

x
.
L

(156)

w1 = q 1

and the overall buckling by


w3 = u3 L sin
54





 

  

Figure 36: Pilot model of a compression sandwich panel.
We can immediately get a first-order expression for the total compression in the righthand face plate r
2 
2 #
Z L "
1
dw1
1
dw3
1
r = U4 L + bu3
+
dx = U4 L + bu3 2 Lq12 2 Lu23
dx
dx
4
4
0 2
(157)
Similarly, we get an expression for the compression in the left-hand face plate l
2 
2 #
Z L "
dw3
1
dw2
1
1
l = U4 L bu3
+
dx = U4 L bu3 2 Lq22 2 Lu23
dx
dx
4
4
0 2
(158)
The final two terms in r and l need some explanationthey represent the release
in compression arising from the face plates becoming curved rather than straight. The
strain release from this is:
 2
1 w
x =
,
(159)
2 x
and is one of the terms that accounts for large deflections in plated structures (see
later). This term is averaged over the length by the integration process and gives the
nonlinear terms when the strain energy is computed. The strain energy stored in these
55

face plates from direct compression is therefore (assuming the panel is of unit breadth) 4
1
Um = Et
2

1
= ELt
2

2x dx

"

1
b
U4 + u3 2 (q12 + u23 )
L
4

2

b
1
+ U4 u3 2 (q22 + u23 )
L
4

2 # (160)

The strain energy in bending can be written thus if we ignore the small contribution
from the overall mode
Z L  2 2
1
1
dw
dx
=
Ub = EI
KL(q12 + q22 )
(161)
2
2
dx
2
0
where I = t3 /12 and K = i2 4 I/2L2 , if we assume the core has a Youngs modulus
Ec , the strain energy stored in the core springs is
Uc =

L
0

b
0

1 (w1 w2 )2
1
E
dy dx = kL(q1 q2 )2
2
2
b
4

(162)

given that y = (w1 w2 )/b and k = Ec L2 /2i2 b. The work done by the load P is simply
P = P LU4 .

(163)

Therefore the total potential energy function is given by (after dividing through by L):
V = U b + Uc + Um P
"
2 
2 #
b
1 2 2
1
1
b
= Et U4 + u3 (q1 + u23 ) + U4 u3 2 (q22 + u23 )
2
L
4
L
4

(164)

1
1
+ K(q12 + q22 ) + k(q1 q2 )2 P U4
2
2

We can eliminate U4 as a passive coordinate as per the previous section:




1
1 2 2
V
2
2
= Et 4U4 (q1 + q2 + 2u3 ) P = 0
U4
2
2
giving
P
1
(165)
+ 2 (q12 + q22 + 2u23 ),
2Et 8
the first term being the non-trivial fundamental path. Substituting this expression back
into V has reduced the system effectively to three DOFs.
 2

1
b
2 2
2 2b
2
2
2
2 2
V = Et
u
u3 (q1 q2 ) + (q1 q2 )
2
L2 3 2L
32
1
P 2 2
1
(q1 + q22 + 2u23 )
+ K(q12 + q22 ) + k(q1 q2 )2
2
2
8
U4 =

The subscript m stands for membrane

56

Note that we have ignored a term in P 2 as this would vanish on differentiation in any
case. We can further simplify V by diagonalizing it. Similar to the example studied
earlier we can transform V by the following:
u1 = q 1 + q 2 ,
u2 = q 1 q 2 ,

(166)

 2

1
b
2 2 2
2
2 2b
V = Et
u
u1 u2 u3 + u1 u2
2
L2 3 2L
32
1
1
P 2 2
+ K(u21 + u22 ) + ku22
(u1 + u22 + 4u23 )
4
2
16

(167)

on substitution V becomes:

Equilibrium equations and critical loads


The new coordinates define the modes shown in Fig. 37: u1 is the snake mode which,



(a) Snake



(b) Hourglass



(c) Overall

Figure 37: Diagonalizing DOFs showing modes of buckling.


in the absence shearing resistance, does no work in the core; u 2 is the hourglass mode
which accounts for the only strain energy stored in the core springs; u 3 , as stated before,
is the overall mode which turns out to be the snake mode with a fixed value of i = 1.
As the system is diagonalized we can evaluate the critical loads of each mode:
1
P C2
4K
V11C = K 1
= 0 P1C = 2 ,
2
8

C 2
P
4 (K + 2k)
1
= 0 P2C =
,
V22C = K + k 2
2
8
2
2Etb2 2 P3C 2
4Etb2
C
V33C =

=
0

P
=
.
3
L2
2
L2
57

(168)
(169)
(170)

It can be deduced from this that the hourglass mode (u2 ) is never critical in this model,
but either u1 or u3 could be. If we measure the potential energy from the critical point
of the snake mode (u1 ), by introducing a new load p = P P1C , we can rewrite V :



Etb2 2
Etb 3
Et 4 2 2 p 2 2
1 2
2

K
u

u
u
u
+
u1 u2
u1 + u22 + 4u23 (171)
V = ku2 +
1 2 3
3
2
2
L
4L
64
16
The three equilibrium equations are therefore:

Et 4
Etb 3
V
=
u1 u22
u2 u3
u1
32
4L
V
Etb 3
Et 4 2
= ku2
u1 u3 +
u u2
u2
4L
32 1


V
2Etb2 2
Etb 3

2K
u

=
u1 u2
3
u3
L2
4L

p 2
u1 = 0
8
p 2
u2 = 0
8
p 2
u3 = 0
2

(172)
(173)
(174)

By multiplying eq. (172) by u1 and eq. (173) by u2 and taking the difference we get:
p 2 2
u =
8 1


p 2
k u22 .
8

(175)

Conversely by multiplying eq. (173) by u1 and eq. (172) by u2 and taking the difference
we get and substituting for u3 by rearranging eq. (174), we get:


(Etb 3 /4L)2
2Etb2 2 /L2 2K p 2 /2

Et 4

32

u21

u22

k u1 u2 = 0 .

(176)

Within the equilibrium equations uncoupled solutions can be found that have neutral
post-buckling paths at the critical loads found in eqs. (168)(170). Coupled solutions
that contain the modal interactions can be found by setting the {. . .} term in eq. (176)
to zero. In combination with eq. (175), at a fixed value of p we can determine u 1 and u2 ,
with u3 being determined from eq. (174). The equilibrium paths shown in Figs.3840
are computed by determining u1 , u2 , u3 and U4 for different p values. The equations
are:

1
p 2 Et 4
(Etb 3 /4L)2
2
u2 =

,
8
32
2Etb2 2 /L2 2K p 2 /2


8k
2
u1 = 1 2 u22 ,
p
(177)
(Etb 3 /4L)u1 u2
u3 =
,
2Etb2 2 /L2 2K p 2 /2

P
2 2
U4 =
+
u1 + u22 + 4u23 .
2Et 16
6.6.2

Secondary Bifurcation

In this case instability initially triggers the overall mode u 3 . However, as the amplitude
of this grows, a secondary bifurcation point is encountered; an interactive mode is
58


10700

10680

10000

10660
8000

10640
10620

6000

10600
10580

4000

10560
10540

2000

10520
0

 5 

10500 3.5

(a) Full graph

3.52

3.54

3.56

3.58



(b) Zoom into P3C

Figure 38: Equilibrium path: Load P versus end-shortening U4 .


triggered at S and the load-carrying capacity is suddenly reduced from the overall
modal critical load.
The location of the secondary bifurcation can be determined analytically. For the
example presented we know that the overall mode is the critical uncoupled mode
(u1 = u2 = 0, u3 6= 0 and P = P3C ). Like at all bifurcation points the Hessian matrix Vij
becomes singular at the secondary bifurcation and gives the following relationship:


S
V11S
V123
uS3 0

S
S
uS3
V22S
0 = 0
Vij
= V123

0
0
V33S

where V123 u3 is the coefficient remaining once the cross-derivative with respect to u 1
and u2 has been evaluated, and thus:
s

V11 V22
S
u3 =
.
2

V123
S

The different plots show the variations in the modal amplitudes as the bifurcations are
encountered. The key to the interactive behaviour is in the cubic cross-term (in this
case u1 u2 u3 ) in the energy; these break the symmetry that is normally in V when it
only contains even power terms of energy in the DOFs. Forms of V which only contain
quadratic and quartic terms are familiar from the symmetric responses discussed in
the SDOF section.
It has been demonstrated that an initial stable (flat) post-buckling response can be subsequently destabilized by modes which have critical loads above the lowest value.

59

3.6

24000
22000
20000
18000

 
 

16000
14000
12000



10000
8000
0.02

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Figure 39: Equilibrium path: Load versus overall mode amplitude u3 .





0.3

0.6

0.2

0.06

0.4

0.1

0.04

0.08

0.12

0.16

0.2

0.24



0.2

0.28

0.1
0

0.1

0.2

0.3

0.4

0.5

0.6



0.2
0.2

0.3

(a) Overall mode vs Snake mode

(b) Local modes u1 vs u2

Figure 40: Modal interactions: Variations of relative amplitudes after the bifurcations.

60

LateralTorsional Buckling

Lateraltorsional buckling (LTB) is the type of instability most likely in beams under
bending. For example, Fig. 41 shows a cantilever undergoing LTB. The beam initially




 !  #"$%&' ( )

Figure 41: Cantilever undergoing lateraltorsional buckling. Dashed lines indicate the
original position of the beam. Note the sideways displacement and the twisting of the
cross-section.
bends about the minor axis which in turn encourages the compression flange to bend
more in sympathy.
The theory mainly deals a simply-supported beam with a uniform bending moment M
61

as shown in Fig. 42 which deflects as shown in Fig.43. The notation for the following
derivation is shown in Fig. 44.

(a) Elevation

 
 




(b) Plan

Figure 42: Simply-supported beam with a uniform applied moment M . The distance
between lateral restraints is L and so the beam is simply-supported in both planes.

 !#"$

%'&)( (*&,+.- /0#12


Figure 43: Deflection of simply-supported beam undergoing LTB. Dashed lines indicate the bottom flange.

62

 





Figure 44: Notation for the LTB critical moment derivation where M is for bending
moments and T is for the twisting moment.

7.1 LateralTorsional Buckling for Beams in Pure Bending


Object: to determine the critical value of M = Mcrit that causes lateraltorsional buckling in a simply-supported beam of length L and doubly-symmetric cross-section under a uniform bending moment M .
The cross-section is assumed to be doubly-symmetric with second moments of area: I x
(major axis), Iy (minor axis); St Venant torsion constant:
X
J=
[breadth (thickness)3 ]/3;
(178)
Youngs modulus: E; Poissons ratio: ; and shear modulus: G = E/[2(1 + )]. Fig. 45
shows the displacement of the cross-section with its combination of twisting and lateral
movement.
u

Figure 45: Section twisting and moving laterally.

With the angle of twist , the lateral displacement u and the vertical displacement v,
we can first say that the bending moment about the major axis is M x = M and about
the minor axis My = M . Then the following differential equations can be written
63

from bending theory:


d2 v
= M,
dz 2
d2 u
EIy 2 = M.
dz

(179)

EIx

(180)

As the section is not circular, the torsion arising from the angle of twist is not uniform
because of section warping. The equation for non-uniform torsion is given by:
GJ

d
d3
EIw 3 = T
dz
dz

(181)

where the second term in (181) is the correction for warping with T being the induced
torque from the external bending moment M and the change in lateral displacement u
along the beam (in z direction):
du
T = M
,
(182)
dz
and Iw is called the warping stiffness (mm6 ) which is equal to Iy h2 /4 for an I-section,
where h is the section height. Substituting (180) into the derivative with respect to z of
(181) gives the following governing differential equation for in terms of z
EIw

d2 M 2
d4

GJ

=0.
dz 4
dz 2
EIy

(183)

This equation has two solutions:


1. = 0: the undeformed state, and
2. = A sin z
: where A is the amplitude of the twist.
L
Substituting the second solution into (183) and solving the resulting quadratic equation
in M gives the expression of the critical moment:
Mcrit
Note the following:

p
=
EIy GJ
L

1+

2 EIw
.
L2 GJ

(184)

Ix has no influence on the critical moment.


The post-buckling behaviour is stable-symmetric with a very flat response and
so Mcrit is a fair estimate of the failure load (cf. Euler buckling: Failure load for
slender columns PE ).
q
2
w
Examining term 1 + L 2 EI
GJ
For large L, the term goes to unity.

For large h and small L (deep and short beam), Mcrit rises significantly.
64

7.1.1

Alternative formulation using potential energy

Assuming no pre-buckling deformation and no stored strain energy in major axis


bending we can define two separate lateral displacements:
us : representing the displacement of the centroid,
uw : representing the displacement of the flanges.
The lateral displacements of the flanges are thus (see Fig. 46):
Top: us + uw
Bottom: us uw
Because we are purely concerned with minor axis bending the values of I y for the







Figure 46: Lateral displacements due to LTB


flange can be approximated to Iy /2 of the whole section by assuming the contribution
by the web is small. The strain energy due to bending (Ub ) is therefore:
2  2
2 )
Z L ( 2
d
1
d
Ub = EIflange
(us + uw ) +
(us uw )
dz
2
dz 2
dz 2
0
(185)
Z L  2 2
Z L  2 2
d us
1
d
1
dz + EIw
dz,
= EIy
2
dz 2
2
dz 2
0
0
as uw = h/2 and Iw was defined in the earlier formulation. The two terms represent
strain energy stored in bending and warping respectively. The strain energy stored
65

from uniform torsion (UT ) is:


Z
1 L
UT =
T d
2 0
Z L  2
1
d
= GJ
dz
2
dz
0

(186)

The work done by the load in this case equal to the applied moment multiplied by the
resulting rotation. Fig. 47 shows a sequence of diagrams observing a beam element of













%'&)(*(+&-,.&/ 0 1
2$3
465798;:=<?>@8+A"7CBEDGFH7I

  "!$#

WXY
WZ

JKL
JM

\^]_

N RP Q S
PVU
O PT
Figure 47: Plan view of differential beam element
length dz in plan. A constant sideways movement us over the element, or a constant
rotation over the element, gives no end rotations. Only by introducing du s and d,
can we see any end rotations in the plane of M . The top and bottom parts of the web

66

move:

h
2
h
Bottom moves: +
2
Top moves:

dus
d
dz
dus
d,
dz

the resulting rotation is


Rotation:

dus
dus d
d =
dz (by the chain rule).
dz
dz dz

Using the generic notation, the work done by the load is


P =

M
0

dus d
dz
dz dz

(187)

and so the total potential energy is thus:


V = U b + UT P
Z L  2
Z L  2 2
Z L  2 2
d
d us
d
1
1
1
dz
= EIy
dz + EIw
dz + GJ
2
2
2
dz
2
dz
2
dz
0
0
0
Z L
dus d

M
dz
dz dz
0

(188)

We can use a two DOF RayleighRitz formulation, where:


us = Q1 sin

z
,
L

= Q2 sin

z
,
L

(189)

to obtain the critical load. The formulation is in small deflections (linear) and so only a
critical equilibrium analysis is possible. Performing the integrations and then forming
the Hessian matrix Vij gives a non-diagonalized form, which at the critical point C:


EIy (/L)4

Mcrit (/L)2
C


Vij =
(190)
2
2
4 = 0,
Mcrit (/L) GJ(/L) + EIw (/L)
gives the expression for the critical moment:
Mcrit

p
=
EIy GJ
L

67

1+

2 EIw
.
L2 GJ

(191)

7.2 Modifications for Differing Loading Cases


7.2.1

Non-uniform bending moments

Up to now we have considered a beam with a constant bending moment. However,


the critical moment changes for different cases such as point loads, UDLs and so on.
The adjustment on Mcrit is encompassed in the ratio m where
Mcrit base case
Mcrit new loading case

m=

(192)

To recap: for the base case (Fig. 42)


r
p
2 EIw
Mcrit =
,
EIy GJ 1 + 2
L
L GJ
but for the case with a central point load W (Fig. 48)

(193)

 

Figure 48: Beam with central point load W and maximum moment M .
r
4.24 p
2 EIw
Mcrit =
EIy GJ 1 + 2
(194)
L
L GJ
and so m = /4.24 = 0.74. Similarly, for the loading cases shown in Fig. 49(a) and (b),
m = 0.89 and m = 0.40 respectively


 





(a) UDL

(b) Anti-symmetric loading

Figure 49: Beams with various loading configurations.

7.2.2

Destabilizing and stabilizing loads

Another modification is for the level at which the load is applied on the section. Fig. 50
shows the two different cases. Case (a) shows a destabilizing load where, as the buckling
occurs, W adds an extra torque that exacerbating the angle of twist . Conversely, case
(b) shows a stabilizing load where W acts to reduce to make the web vertical. For
strength concerns, case (a) is clearly worse than case (b).
68







(a) Destabilizing load

(b) Stabilizing load

Figure 50: Destabilizing and stabilizing load levels.


7.2.3

Effective lengths

The effective buckling lengths for LTB (Le ) owe a great deal to column theory. Fig.51
shows plan views of beams and the type of lateral restraints. Cantilevers have an
effective length of at least 2L and this depends on the position of the load as discussed
in the Section 7.2.2.


(a) Le = L


(b) Le = L/2


 

(c) Le = L/2

Figure 51: Effective buckling lengths for LTB. Cases (a) and (b) compare with a simplysupported column and a fixed-ended column respectively. Case (c) shows a beam that
is fixed to a secondary cross-beam at midspan.

69

7.3 Designing beams susceptible to LTB to BS 5950


The design philosophy of BS 5950 Part 1 is quite similar to that of column design. The
relevant section is 4.3.
Two bounds of design strength:
Plasticity (Mp )
Stability (Mcrit )
The code works to find a maximum bending capacity pb which is defined by a
slenderness parameter LT
where
LT = uv
and
Le
=
,
ry

ry =

w
r

Iy
A

Le comes from either Tables 13 or 14


u is the buckling parameter and is defined by the manufacturing process of
the section (whether rolled or welded)usually 0.9 or 1.0
v is the slenderness factor and depends on section geometryTable 19
w is the slenderness correction factor which depends on the section classification, for most rolled sections (i.e. class 1 and 2 sections): w = 1.0
pb is obtained either from Table 16 or 17 depending on whether the section was
rolled or welded and m is calculated in Table 18.

70

Instabilities in Framed Structures

8.1 Introduction
Frames, comprising beams and columns and in general loaded axially and laterally,
represent perhaps the most common stability problem for engineers. Usually, although
there are notable exceptions (see later), only the critical load is required.
Buckling of frames involves buckling of their individual members, which can be generally represented by elastically restrained beamcolumns. In particular, buckling of a
beamcolumn with arbitrary spring supports at both ends exhibit some, but not all, of
the essential features of the buckling of frame members.
However, in a frame, the individual members interact and buckle simultaneously, the
actual axial load in any one member influencing the critical loads of both the particular
member considered and any adjacent members. In other words, the appropriate values
for the stiffness of the elastic supports at the two ends of a member (which represent the
elastic reaction of the rest of the frame) are not constant but depend on the unknown
critical load of the frame. Neglecting the effect of the rest of the frame on the spring
stiffness can lead to unsafe results.

8.1.1

Classification of frames

Frames can be classified into those with or without sway (Fig. 52). It is also important

 

(a) No sway

(b) Sway

Figure 52: Classification of frame types


to distinguish between general loading and frames under axial loading alone. General
loading cases are given in Fig. 52. If members are loaded only axially, the frame will
bifurcate at some critical value of the loading factor = C (See Fig. 53). For general
71





 

Figure 53: Bifurcation of frame during buckling ( = C ).


loading, the deflection approaches infinity at some critical load. The response is like
that of an imperfect systemrounding off the corner of a bifurcation (Fig. 54) For any







Figure 54: Behaviour of a buckling frame under general loading.
symmetrical form of loading, the same effective C applies. The axially loaded frame,
with columns taking the same load as in the imperfect generally loaded frame, thus
represents its corresponding perfect (bifurcating) system.

8.1.2

Effects of structural connections

The actual behaviour of a frame carrying both lateral and axial loads is also influenced
by the characteristics of the members joints, which are usually considered to be either
simple, semi-rigid or rigid, according to their ability to transmit moment.
1. Simple connections may be defined as those that will not develop restraint moments, and which will therefore allow the members meeting at this point to be
72

pin-jointed. Clearly, the joint should have sufficient rotation capacity to allow for
member end-rotations. If there are sufficient pin-joints to make the frame statically determinate, each member will act independently of the others, hence a
frame stability analysis is not required.
2. Semi-rigid connections are those which have some moment capacity and which
partially restrain the relative rotations of the members at the joints. The methods
of analysis for semi-rigid frames tend to be approximate, based on either of the
two limiting cases (i.e. frames with simple or rigid joints).
3. Rigid connections may be defined as a joint with sufficient rigidity to prevent
virtually relative rotation between the members connected. There are important
interactions between the members of frames with rigid joints. A frame stability
analysis must be carried out in this case.

8.2 Elastic Critical Loads of No-sway Frames


Analytical methods abound. Perhaps nowadays numerical methods such as finite elements provide the most effective tool. Here we adopt a modified stiffness method
employing stability functions for axially loaded members.

8.2.1

Stability functions

Consider a clamped beam under axial loading (Fig. 55), first with P = 0, we can relate
moments and rotations at the ends of the member by a stiffness matrix K:
 



4EI/L 2EI/L
MA
A
=
(195)
MB
2EI/L 4EI/L
B
For P 6= 0 the elements of the stiffness matrix K change, and these equations can be







Figure 55: Axially loaded clamped beam showing moments and rotations
re-written (k = EI/L):


MA
MB

=k

s sc
sc s
73



A
B

(196)

where s and c are so called stiffness and carry-over factors respectively; so named
because of their role in the moment distribution method. For P = 0, s = 4 and c = 1/2.
The general solution of the differential equation for an axially loaded beam can be used
to find functions for s and c as follows:
L [1 L cot(L)]
2 tan(L/2) L
L sin(L)
c=
sin(L) L cos(L)

s=

where:
=

P
=
EI
L

(197)
(198)

P
PE

Buckling occurs when the total stiffness matrix for the frame becomes singular, that is
to say
det(K) = 0.
(199)

8.2.2

Worked example

Fig. 56 is the frame for the worked example.




Figure 56: No-sway frame example. All elements (except bracing) have flexural rigidity EI and length L.

74

8.3 Sway Frames


For a member which includes sway (Fig. 57), we must replace the 22 element stiffness
matrix by a new 3 3 matrix involving a shear force FB and sway displacement sway .
The matrix changes thus:









Figure 57: Beam now including sway displacement

MA
s
sc
s(1 + c)
A
MB = k
B
sc
s
s(1 + c)
FB
s(1 + c) s(1 + c) 2s(1 + c) P L/k
sway

(200)

8.4 Limitations of Analysis Using Stability Functions


1. Inapplicable for asymmetric loading (Fig. 58). General asymmetric loading cannot be simply replaced by an equivalent axially loaded system to find the critical
load.

Figure 58: Example of asymmetric loading in frames


2. Nonlinear geometry: e.g. Roorda frames (Fig. 59). Frame asymmetry generates
an asymmetric point of bifurcation which implies imperfection sensitivity and
the critical load analysis becomes non-conservative. If the frame is inextensional,
no imperfection gives a perfect response with a critical point P C . However,
75

extensibility introduces an effective (negative) imperfection. The system buckles


at the limit point where P = P M < P C .






Figure 59: Roorda frame with an asymmetric point of bifurcation.
3. Plasticity may cause collapse before P C is attained.
4. Sizing of frame: local instability of a single column can be approximated by summing incoming stiffnesses from adjacent members. In this case, stability of the
whole frame is not checked. Code methods, such as that given in BS 5950, are
often based on approximate techniques and it is important to understand their
range of application.

76

Buckling of Plates

Plated structures are common in structural engineering. Any structure that requires
cross-section properties over and above those provided by rolled sections are built up
using plates and thereby become plated structures. The most common example of a
type of structure where plates are prevalent is in bridgeswhere plate and box girders
are common and their elements (flanges and webs) behave like separate plates. In this
section we will discuss the behaviour of simply-supported rectangular plates in axial
compression, in-plane shear stresses and under uniformly-distributed lateral loads.
The following key aspects are discussed:
determination of critical buckling loads,
introduction to post-buckling analysis.
Plated structures are renowned for their stiff post-buckling response and to take advantage of this the origin of this stability needs to be discussed and will come later.

9.1 Critical Buckling of Plates


In this section we will derive the critical buckling loads by the use of potential energy.
Consider a uniaxially loaded rectangular plate which is simply-supported on all four
edges and is of length a, width b and thickness t (Fig. 60).




Figure 60: Simply-supported rectangular plate under uniaxial loading.

9.1.1

Strain energy

If a rectangular differential element is considered under cylindrical bending about the


y axis (Fig. 61) we can write down the strain energy of bending in the x direction:
 2

w
1
dU = Mx dy
dx ,
(201)
2
x2
77






!












 

"


(a) Undeflected

(b) Deflected

Figure 61: Differential element for plate bending.


and immediately also get the corresponding result for the y direction:
 2

w
1
dy .
dU = My dx
2
y 2

(202)

If the same differential element is twisted about the x axis with a twisting moment M xy
(Fig. 62) then this component of strain energy is:
#$

/0
(

12
%&

*,+.'

576
598

:9; =
:9<

:>?; A
@
:9@9:9<

4
)

(a) Undeflected

(b) Deflected

Figure 62: Differential element for plate twisting.


1
dU = Mxy dy
2


2w
dx .
xy

(203)

the component from Myx relies on the general plate theory result that Mxy = Myx and
gives the same result as above. By using the standard expressions relating moment
(bending and twisting) to curvature:

 2
2w
w
(204)
Mx = D
+ 2
x2
y
 2

w
2w
My = D
+ 2
(205)
y 2
x
2w
(206)
Mxy = D(1 )
xy
78

where D = Et3 /[12(1 2 )] and is called the flexural rigidity of a plate. Integrating over
the volume of the plate (assuming that the thickness is constant), the total strain energy
in bending is:
"
 2 2 #)
2
Z Z ( 2
w 2w
w
D b a
2w 2w
dx dy (207)
+

2(1

)
U=
2 0 0
x2
y 2
x2 y 2
xy
This expression for U has two distinct parts: the first part relates to engineers bending
theory where M = EI (curvature), the second part with the 2(1 ) coefficient
is known as Gaussian curvature and arises from asymmetries in the plate deflections.
Both parts account for double curvature in the plate which implies that the plate has to
stretch to buckle and that its buckled configuration is non-developable. Buckling occurs
when the work done by the load supplies just enough energy to allow for the in-plane
stretching to occur.

9.1.2

Work done by load

For a general plate problem, the work done by load depends very much on the problem. For the uniaxially-loaded rectangular plate shown in Fig. 60 the plate can be considered as inextensional and the expression is very similar to the column case (except
for the double integration required):
Z Z  2
x t b a w
dx dy.
(208)
P =
2 0 0
x

However under general loading where lateral loads and shear are possible (Fig. 63),
we can extend this simple expression to:
"  
 2
   #)
Z bZ a(
2
t
w
w
w
w
P =
wq +
x
+ y
+ 2xy
dx dy .
2
x
y
x
y
0
0

(209)








 


Figure 63: Plates under biaxial loads, shearing stresses and lateral loads.

9.1.3

Critical load evaluation for an axially loaded plate

Returning to the uniaxial case, we now have enough information to calculate critical
loads. For post-buckling analysis we require the consideration of in-plane stretching
79

of the plate surface. We can initiate Rayleighs method for a critical load analysis by
assuming:
y
mx
sin
(210)
w(x, y) = Q sin
a
b
i.e. one half sine wave in the y direction and m half waves in the x direction. This
assumes that the plate is long and thin, i.e.
a  b.
Substituting the expression for w into the energy double integrals give:
w
x
2w
x2
2w
y 2
2w
xy

m
mx
y
cos
sin
a
a
b
 m 2
mx
y
sin
= Q
sin
a
a
b
 2
mx
y
sin
= Q
sin
b
a
b
2
m
mx
y
=Q
cos
cos .
ab
a
b
=Q

(211)
(212)
(213)
(214)

The strain energy expression is thus:


 2 #2
mx 2 y
1
sin2
sin
dx dy
+
a
b
a
b
0
0
Z Z
i
D(1 )m2 4 2 b a h 2 mx 2 y
2 mx
2 y
Q
sin

cos
cos

sin
dx dy (215)
a2 b 2
a
b
a
b
0
0
"
 2 #2
D 4 ab 2  m 2
1
Q
+
=
ZERO.
8
a
b

D 4 2
Q
U=
2

Z bZ

"

 m 2

The work done by the load is thus:


Z Z
x t  m 2 2 b a
mx 2 y
sin
dx dy
Q
P =
cos2
2
a
a
b
0
0
x tm2 2 b 2
Q
=
8a

(216)

Accumulating the total potential energy V :


V = U P

"
 2 # 2
D 4 ab 2  m 2
1
x tm2 2 b 2
=
Q
Q
+

8
a
b
8a

(217)

At critical equilibrium:
2V
= 0,
Q2

(218)

k 2 E
12(1 2 )(b/t)2

(219)

which gives the following expression:


xC =

80

where:
k=

2

(220)

= a/b

Consider for the moment that m is a constant and the plate aspect ratio is variable:



m
dk
1

m
=2
+

=0
(221)
d

m
m 2

hence:

(222)

2 = m 2

for practical values when = m we get a minimum value of the buckling load coefficient k and for the rectangular simply-supported plate, this is equal to 4. Fig. 64 shows
different curves relating to different buckling mode wave numbers m and how these
affect their respective k values against .

   

10







Figure 64: Curves of buckling load coefficient k plotted against plate aspect ratio .
Some points to note:
1. The minimum value of k for any aspect ratio (Fig. 64) gives the valid mode m.
Other values of m also give valid modes but these are associated with higher k
values (i.e. higher critical loads). A buckled plate divides itself approximately
into squares (Fig. 65)
2. k only reaches kmin = 4 for integer values of and only rises significantly above
this for small .
3. Transitional cases ( = trans when km = km+1 )
2 
2


m+1
m
+
=
+

m+1
leads to:

trans =

81

m(m + 1)

(223)



Figure 65: Plate buckling into square waves.

2 and k = 4.50.

2nd transition: trans = 6 and k = 4.17.

3rd transition: trans = 12 and k = 4.08.

1st transition: trans =

Hence as and m increase, the value of k at transition reduces. For a long plate,
kmin 4. The transitional cases may seem the best case to design for (e.g. k = 4.5
is better than kmin = 4 by approximately 12%). However, this is only a result
of small deflection critical buckling analysis which tells us nothing about postbuckling. At transitional points post-buckling is likely to be eroded by nonlinear
mode interaction effects.
4. For wide plates ( 0) the critical stress reduces to:
xC =

2E
12(1 2 )(a/t)2

(224)

Comparing this expression with the critical stress of a buckling strut of square
cross-section of side t and length a:
I=

t4
2 Et4 C
2E
, PC =
,

=
.
12
12a2
12(a/t)2

(225)

We find that the only difference is a factor of 1/(1 2 ), which represents a stiffening against anticlastic bending which is allowable in a sequence of struts but
not a plate. The buckling of a wide plate is governed by the length a and not the
width b and can be an important factor in design.
Failure of Koblenz box-girder bridge (Germany) in the late 1960s during
construction (6 people killed)wide plate introduced by a mistake in the
weld detailing (shown in Fig. 66). The shaded region is the wide plate
which emerged as the stiffeners were not welded to the flange plate in the
region of the splice. The design considered long plates of width b but inadvertently missed the failure of the wide plate of length a.

9.1.4

Notes on obtaining the critical load

The critical load can be obtained in two separate ways

82

JKLGMN=OPRQTSL L KUUWVIXZY)VIUWV
"!#%$'&)(*"+-,.//10325476)81#
 





9;:=<)>?@AB>CDEGF@H>I<
Figure 66: Koblenz bridge collapse from wide plate buckling.

83

1. Inextensibility assumed, as shown above, with:


Z b Z a  2
1 w
=
dx dy
x
0 2
0
which is fine for just obtaining xC but does not suffice for any post-buckling considerations.
2. The same result can be obtained by assuming the load does no work at all, so
the ends do not move, but stretching of the middle surface takes place (similar to
Fig. 65). From the buckling deformation w: the plate is stretched out longer than
a by an amount a , thus:
Z  2
1 a w
dx.
a =
2 0
x
See the discussion of Von Karman equations in the plated structures course.

9.2 Shear Buckling


Plates buckling in shear are, again, common in plate and box girder webs. The buckling
of a plate in shear is represented in Fig. 9.2. We can adapt our earlier energy analysis
by assuming w(x, y) has a double Fourier sine series form:
w(x, y) =

Cmn sin

m=1 n=1

ny
mx
sin
a
b

(226)

This analysis can be found in Theory of Elastic Stability by Timoshenko & Gere, 1961.



Figure 67: Buckling of a simply-supported plate in shear.


Again, we shall only consider plates that are simply-supported. For a longitudinally
compressed plate of thickness t and length a and width b, one sine wave model suffices. Now, the buckle is made up of a number of sine waves. Substituting w into the
expression for strain energy of bending gives Ub [Eq. (207)]

2

D 4 ab X X 2  m 2  n 2
C
+
.
(227)
Ub =
8 m=1 n=1 mn
a
b
For the remaining energy assume no work done by the load as point (2) in 9.1.4 and assume that at buckling strain energy stored in bending equals the release in membrane
(in-plane stretching) energy, with applied Nxy only, where
Nxy = xy t.
84

The release in stretching Us is thus:


Z bZ
Us = Nxy
0

a
0

w
x



w
y

dx dy

(228)

where the integrand is the shear angle produced by distortion due to w. Substituting
the Fourier approximation for w(x, y) gives:
Us = 4Nxy

X
X
X
X

Cmn Cpq

m=1 n=1 p=1 q=1

(m2

mnpq
p2 )(n2 q 2 )

(229)

where m p and n q are odd numbers. As we have stated above buckling occurs
when Ub = Us , so:
 
2

X
X
m 2  n 2
2
Cmn
+
4
a
b
D
ab
n=1
m=1
C
Nxy
=
(230)
X
X
X

32 X
mnpq
Cmn Cpq 2
(m p2 )(n2 q 2 )
m=1 n=1 p=1 q=1
Post-buckling responses of plates in shear is quite complicated and is related to the
tension field that is created diagonally across the buckled rectangular panel Fig. 9.2.
Timoshenkos approach does not deal with degrees of freedom, rather it minimizes


 
Figure 68: Tension field action of a buckled plate in shear
Nxy with respect to the Cmn terms directly. The accuracy of the result depends on the
number of input sine waves (number of Cmn terms), e.g. for a short plate = a/b < 2:
1st, with just C11 , C22

9 2 (1 + 2 ) 2 D
32
3
b2
error is approximately 15% for a square plate.
C
=
Nxy

With more degrees of freedom:

2D
b2
where k is a function of and k = 9.4 with five terms in the Fourier series. The
exact answer is k = 9.35.
C
Nxy
=k

The general expression for k in terms of the plate aspect ratio is


k = 5.35 +

4
for 1
2

5.35
k = 2 + 4 for 1

85

(231)

9.3 Introduction to Plate Post-Buckling


Significant post-buckling strength is exhibited by axially-loaded plates and this section discusses a simplified formulation that gives a lower bound on the post-buckling
stiffness. In this introductory section, the following assumptions are made:
Plate is made up of a set of struts (strips) of width dy.
Strips are constrained to act together during buckling.
Ignore transverse and shear in-plane (membrane) effects
The Rayleigh method can be applied to this problem, a general form for w(x, y) can be
thus:
ix
L
(232)
w(x, y) = u1 g(y) sin
i
L
means that each strip deforms as shown in Fig. 69, where g(y) is the y variation of w

!"#%$&')(*

+-,/.0132546.879;:<3=?>!@BA;CD09FE!.HGGI98J=

 



Figure 69: Axial deformation of strip of width dy.


(g(y) = sin y/b for simply-supported edges), L/i is a nondimensionalization. If each
strip was inextensional, it would shorten by
1
=
2

L
0

w
x

2

dx

(233)

to first order. However, they are not inextensional and shorten by U 3 L, hence compressive strain x :

L
1 2 2 2
= U 3 g u1 ,
4

x = U 3

(234)

if the strain is assumed to be constant along length (and also through the plate thickness). In-plane membrane energy is the principal source of post-buckling stiffness and
this is obtained by integrating the strain energy:
Z
1
(x x + y y + xy xy ) dS
Us =
2
Z
(235)
1 2
=
E dS
2 x
86

where S represents the volume of plate. Only one term remains in this expression as we
have ignored shear and transverse strains (y = xy = 0). Performing the integration
gives:
2
Z b
1
1 2 2 2
Us = ELt
g u1 U 3
dy
(236)
2
4
0
In the strain energy of bending (Ub ), nonlinear effects are negligible and the Timoshenko view of this suffices, i.e. bending energy is equivalent to the released energy
in stretching at P C .
Ub =

b
0

1L C 2 2
P C
dy =
P u1
b
4b

(237)

g 2 dy.
0

The work done by the load and hence the total energy are given by:
P = P LU3 ,

(238)

V = U s + Ub P
2
Z
Z b
1L C 2 2 b 2
1
1 2 2 2
g dy P LU3
dy +
= ELt
g u1 U 3
P u1
2
4
4b
0
0

(239)

End-shortening U3 is made up of a non-trivial fundamental path component U 3F , and


an extra bit u3 which is entirely due to buckling:
U3 = U3F + u3 =

P
+ u3
Ebt

(240)

The post-buckling analysis of this is completed by eliminating u 3 as a passive coordinate and evaluating the post-buckling stiffness as earlier methods have shown. Table 1
shows the effective stiffness values for some common edge and membrane conditions.
Long edge support conditions
Lower bound stiffness
2 SS, straight edges
0.333
2 SS, edges free to pull in
0.333
2 CL, straight edges
0.449
1 SS + 1 FR, edges free to pull in
0.444

Exact stiffness
0.500
0.408
0.497
0.444

Table 1: Stiffness of post-buckling path as a proportion fundamental stiffness of long


plates under axial compression. SS: Simple support, CL: Clamped, FR: Free edge.

87

Você também pode gostar