Beruflich Dokumente
Kultur Dokumente
This book deals with the elastic stability of solids and structures, for
which Warner Koiter was the world’s leading expert of his time. It
begins with fundamental aspects of stability, relating the basic notions
of dynamic stability to more traditional quasi-static approaches. The
book is concerned not only with buckling, or linear instability, but
most importantly with nonlinear postbuckling behavior and imperfec-
tion sensitivity. After laying out the general theory, Koiter applies the
theory to a number of applications, with a chapter devoted to each.
These include a variety of beam, plate, and shell structural problems
and some basic continuum elasticity problems. Koiter’s classic results
on the nonlinear buckling and imperfection sensitivity of cylindrical
and spherical shells are included. The treatments of both the funda-
mental aspects and the applications are completely self-contained. This
book was recorded as a detailed set of notes by Arnold van der Heij-
den from W. T. Koiter’s last set of lectures on stability theory at TU
Delft.
Edited by
www.cambridge.org
Information on this title: www.cambridge.org/9780521515283
© Arnold van der Heijden 2009
1. Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Discrete systems 1
3. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1 The incompressible bar (the problem of the elastica) 55
3.2 Bar with variable cross section and variable load distribution 59
3.3 The elastically supported beam 61
3.4 Simple two-bar frame 67
3.5 Simple two-bar frame loaded symmetrically 72
3.6 Bending and torsion of thin-walled cross sections
under compression 78
3.7 Infinite plate between flat smooth stamps 84
3.8 Helical spring with a small pitch 101
3.9 Torsion of a shaft 110
3.10 Torsion of a shaft with a Cardan (Hooke’s) joint 119
3.11 Lateral buckling of a beam loaded in bending 126
3.12 Buckling of plates loaded in their plane 137
3.13 Post-buckling behavior of plates loaded in their plane 158
3.14 The “von Kármán-Föppl Equations” 166
v
vi Contents
These lecture notes were made after Professor Koiter’s last official course at Delft’s
University of Technology, in the academic year 1978–79. Although these notes
were prepared in close collaboration with Professor Koiter, they are written in the
author’s style. The author is therefore fully responsible for possible errors.
This course covers the entire field of elastic stability, although recent develop-
ments in the field of stiffened plates and shells are not included. Hopefully, these
lecture notes reflect some of the atmosphere of Dr. Koiter’s unique lectures.
vii
1
Stability
Consider a system with a finite number of degrees of freedom. The position of this
system is represented by a position vector q(q1 , q2 . . . qn ), where qi (i ∈ 1 . . . n) are
n independent coordinates. It is assumed that the system is holonomic, i.e., no rela-
tions exist between the derivatives of the coordinates, and scleronomic, i.e., the fac-
tor time is not explicitly needed in the description of the system.† Let q̇i be the gen-
eralized velocities. The kinetic energy T is then a homogeneous quadratic function
of the generalized velocities, and hence T can be written as
1
T= aij (q)q̇i q̇ j . (1.1.1)
2
When the system is non-sclerononic, terms linear in the velocities and a term inde-
pendent of q̇i must be added. The coefficient aij (q) is called the inertia matrix. The
forces acting upon the system can be expressed by a generalized force vector Q
defined by
where the right-hand side stands for the virtual work of all the forces acting upon
the system. In general, this expression is not a total differential. However, for an
important class of problems, it is. Systems for which 1.1.2) is a total differential are
called conservative systems. In that case we have
where δP (q) is a total differential and P (q) is called the potential energy. In the fol-
lowing, we mainly restrict our attention to conservative systems because for elastic
systems, conservative forces play an important role.
Introducing a kinetic potential L defined by
L = T − P, (1.1.4)
1
2 Stability
P,i = 0. (1.1.9)
By stability we mean that a small disturbance from the state of equilibrium does
not cause large deviations from this state of equilibrium. A disturbance from the
state of equilibrium implies that the velocities are nonzero or that the position dif-
fers from the equilibrium position. We can always choose our coordinate system
such that the equilibrium position is given by q = 0. Furthermore, we can always
choose the potential energy in such a way that it vanishes in the equilibrium posi-
tion. Doing so, we may write
1
P= P,ij (0) qi q j + · · · . (1.1.10)
2
To be able to give a more exact definition of stability, we need a measure to denote
the deviation from the state of equilibrium. Remembering that in equilibrium we
have q = q̇ = 0, a number ρ (q, q̇) is introduced with the following properties:
ρ = max qi + max q̇i .
T + P = E, (1.1.12)
where E = T(t = 0) + P(t = 0). This equation enables us to make the following
statement about stability.
n
q̇2 = (q̇i )2 .
i=1
4 Stability
Let d(c) denote the minimum of P(q) on the hypersphere q = c . P(q) is positive-
definite when d(c) is a monotonically increasing function of c on the sphere δ ≤ c <
R.
This means that T (t = 0) < d (c1 ). Because T + P < d (c1 ) and P is positive-definite,
it follows that T < d (c1 ) for all t. On the other hand, because T is positive or
zero, it follows that P < d (c1 ) for all t. A similar argument holds for a disturbance
q̇(0) < c 2 . Hence, we may choose an arbitrary (small) disturbance and the dis-
placements and velocities will always remain within definable bounds.
The converse of this theorem has not yet been proven in all generality. To see
some of the difficulties that are encountered, we consider the following example
(one degree of freedom):
−2
P(q) = e−q cos q−2 .
For q = 0, all the derivatives vanish. However, in the immediate vicinity of the ori-
gin there are always negative values of P. In spite of this, the system is stable for
sufficiently small disturbances.
Actual physical systems are never exactly conservative, i.e., there is always
some dissipation. The approximation by a conservative system is often a very good
approximation. In the presence of damping forces, we need the Lagrangian equa-
tions with an additional term for the non-conservative forces. Multiplying by q̇ i ,
d ∂L ∂L
q̇i − q̇i i = Q∗i q̇i , (i ∈ 1, . . . , n)
dt ∂q̇i ∂q
from which follows
d
(T + P) = Q∗i q̇i ≡ −D (q, q̇) . (1.1.13)
dt
Damping implies that the dissipation function D > 0 for q̇ = 0. We now make the
following assumptions:
1) The damping forces have the property that Q∗i → 0 for q̇ → 0.
2) D(q, q̇ i ) > 0 for q̇ = 0.
3) P (q) does not possess stationary values for q̇ < c except at q = 0.
The great advantage of this stability theorem is that it does not involve the
kinetic energy, and hence the inertia matrix aij (q). For a conservative or pseudo-
conservative system, the stability criterion only depends on the potential energy
(a quasi-static criterion). In general, the stability problem is a dynamic problem,
and the kinetic energy plays an essential role. An example of such a problem is the
behavior of the wings of an airplane in an airflow. In this case, the forces do not
depend on only the geometry but also on the velocities.
For static loads, it is often sufficient to restrict oneself to conservative loads (e.g.,
deadweight loads). A more severe restriction for continuous systems is that we must
restrict ourselves to elastic systems, i.e., to systems where there is a potential for the
internal energy. Such a potential does not exist when plasticity occurs.
Let us now have a closer look at the stability problem. As mentioned previ-
ously, the stability criterion is fully determined by the potential energy P(q). In the
equilibrium position, we have chosen P (0) = 0 and q = 0 so that we may write
1 1
P (q) = P,ij (0) qi qj + · · · ≡ cij qi q j + · · · , (1.1.14)
2 2
where cij denotes the stiffness matrix in the equilibrium position. It follows that
when the stiffness matrix is positive-definite, P (q) is positive-definite and the system
is stable. If cij is indefinite (or negative-definite) then the system is unstable. If cij
is semi-definite-positive (i.e., non-negative and zero for at least a deflection in one
direction), then we must consider higher-order terms in the expansion for P. This
case is called a critical case of neutral equilibrium. We shall consider this case in
more detail.
It is convenient to transform the quadratic form (1.1.14) to a sum of quadratic
terms. If the form is positive-definite, then the coefficients in the transformed form
are all positive. Applying this transform to (1.1.14) and denoting the transformed
coordinates again by qi , we may write
1
n
P (q) = ci (qi )2 + · · · + () (q1 )3 + · · · . (1.1.15)
2
i=1
c1 ≤ c2 ≤ c3 ≤ · · · ≤ cn .
We now consider the case c1 = 0, c2 > 0. Taking all qi = 0 (i > 1), the dominant
term will be (q1 )3 . This term can attain negative values, and hence the system will be
unstable. A necessary condition for stability is that the coefficient of (q1 )3 is equal
6 Stability
x = y2
f ( x, y) < 0
2
x = 2y
x
Figure 1.1.1
to zero. A further necessary condition for stability is that the coefficient of (q1 )4 is
positive. However, this condition is insufficient, as will be shown in the following
example. Consider the function
The graphs of the functions x − y2 = 0 and x − 2y2 = 0 are given in Fig. 1.1.1.
The function f (x, y) in an arbitrary small neighborhood of the origin takes on
both positive and negative values. In this case, the quadratic form in y vanishes at
the origin, and there is no cubic term, but the coefficient in the quartic term is posi-
tive. Hence, the necessary conditions for stability are satisfied. However, this system
is unstable because in an arbitrarily small neighborhood of the origin, P takes on
negative values.
The reason that the conditions mentioned here are not sufficient is that we have
restricted our investigation to straight lines through the origin (see Fig. 1.1.1). Fol-
lowing these straight lines, we always find only positive values in a sufficiently small
neighborhood of the origin. However, if we follow curved lines through the origin
(see the dashed lines), we easily find negative values. Once we have recognized the
reason why the conditions imposed are insufficient, it is easy to find a remedy. To
this end, we consider a line y = constant in the neighborhood of the origin, and we
minimize f (x, y) with respect to x, i.e.,
This yields 2x − 3y21 = 0, and hence x = 3/2 y21 . Substitution of this value into f (x, y)
yields min f (x, y) = −1/4 y41 , which means that the function is indefinite.
In general, the function P is minimized with respect to qi (i > 1) for fixed q1 .
When the coefficient of (q1 )4 is positive-definite, the system is stable.
2
Consider a body that is in a state of equilibrium under conservative loads. Our aim
is to investigate this equilibrium state.
For an elastic body, the internal energy per unit mass may be represented by
U(s, γ), where s denotes the specific entropy and γ is the deformation tensor. Let
xi (i = 1, 2, 3) be the components of the position vector x, which describe the posi-
tion of a material point in the “fundamental state” I, which is to be investigated. Let
u(x) be the displacement vector from the fundamental state (u is a small but finite
displacement). The corresponding position in the “adjacent state” II is then x + u.
The (additional) deformation tensor is now defined by
1 1
γij = (ui,j + uj ,i ) + uh,i uh,j . (2.1.1)
2 2
The fact that the body has undergone deformations to arrive in the fundamental
state is unimportant because the state I is kept fixed.
The temperature T is now defined by
∂U
T= (2.1.2)
∂s
(γ is kept constant).
From (2.1.2) we obtain
∂T ∂2 U
= 2.
∂s ∂s
The specific heat of the material is now defined by
∂s
T = Cγ , (2.1.3)
∂T
where Cγ > 0 for a thermodynamically stable material. As ∂2 U/∂s2 is positive (non-
zero) we may solve (2.1.2) for s, which yields s = s(T, γ).
We now introduce the function F (T, γ), defined by
7
8 Continuous Elastic Systems
The heat flux will cause an entropy flux. The entropy flux vector h is given by
1
h= q (per unit time and per unit area). (2.1.8)
T
For an arbitrary part of the body, the entropy balance is given by
ρṡ dV = h · n dA, (2.1.9)
V A
where ρ is the specific mass. This equation only holds in the absence of irreversible
processes in the body. When the state of the body also depends on the deformation
rates, irreversible processes will occur, which implies entropy production. In that
case, the entropy balance reads
ρṡ dV = h · n dA + ρσ dV, σ ≥ 0, (2.1.10)
V A V
where σ denotes the entropy production per unit time and mass. This is the more
general formulation of the second law of thermodynamics (Clausius-Duhem). The
first law of thermodynamics states that the total amount of heat that flows into a
body is transformed into internal energy.
Let PL [u (x (t))] be the potential energy of the external loads and let
1
K [u̇ (x (t))] = ρ u̇ · u̇ dV
2 V
where we have a negative sign on the right-hand side of this equation because the
heat flux is regarded as positive in the outward direction.
To draw conclusions from the first and the second laws, we subtract (2.1.11)
from (2.1.10) multiplied by TI . This yields
d
ρ [U (s, γ) − TI s] dV + K [u̇ (x (t))] + PL [u (x (t))]
dt
V
(2.1.12)
TI
= − 1 q · n dA − TI ρσ dV ≤ 0
T
V V
(Duhem, 1911).
Here we have made use of the relation
d
ρṡ dV = ρs dV (2.1.13)
dt
V V
The first term on the right-hand side of (2.1.12) is negative because the heat flux is
in the outward direction when T > TI , and the second term is negative because the
entropy production is always positive. The integral on the left-hand side of (2.1.12)
may be expressed in terms of the free energy. Using the relation
∂F
U (s, γ) − TI s = U (s, γ) − Ts + (T − TI ) s = F (T, γ) + (T − TI ) , (2.1.14)
∂T
we obtain
d ∂F
ρ F (T, γ) + (TI − T) dV + PL + K ≤ 0. (2.1.15)
dt ∂T
V
Duhem (1911) already discussed the stability of a system on the basis of this equa-
tion and came to the conclusion that a system is stable when the form between the
braces is positive-definite. In this form, K is a positive-definite function. However,
the terms between the square brackets depend on the deformation tensor and the
temperature, whereas PL depends on the displacement field. The problem is to sep-
arate the influence of the temperature and the displacement field. A straightforward
expansion
∂F 1 ∂2 F
F (T, γ) = F (TI , γ) + (T − TI ) + (T − TI )2 + · · ·
∂T TI 2 ∂T 2 TI
10 Continuous Elastic Systems
does not solve the problem. Following Ericksen (1965), we may write the Taylor
expansion of the free energy at constant deformation γ in the form
∂F 1 ∂2 F
F (TI , γ) = F (T, γ) + (TI − T) + (TI − T)2 , (2.1.16)
∂T T 2 ∂T 2 ∗
where the first derivative is evaluated at the deformation γ and temperature T, and
the second (starred) derivative at the deformation γ and an intermediate tempera-
ture T∗ = T + θ (TI − T), where 0 < θ < 1. Using (2.1.16) we may rewrite the term
between the square brackets in (2.1.15) as follows:
∂F 1 ∂2 F
F (T, γ) + (TI − T) = F (TI , γ) − (TI − T)2
∂T 2 ∂T2 ∗
(2.1.17)
1 cγ 2
= F (TI , γ) + (TI − T) ,
2 T ∗
where we have used the relation
∂s ∂2 F
cγ ≡ T = T 2.
∂T ∂T
The first term on the right-hand side of (2.1.17) depends only on the displacement
field. The second term is positive-definite. The energy balance may now be written
in the form
d 1 cγ
ρF (TI , γ) dV + PL [u (x (t))] + K [u̇ (x (t))] + ρ (TI − T)2 dV ≤0.
dt 2 T ∗
V V
(2.1.18)
The last two terms in the left-hand member are positive-definite, and the remaining
terms depend only on the displacement field. Our energy balance is not affected
when we subtract from the expression between the braces a time-independent
quantity,
ρF (TI , 0) dV.
V
where W (γ) is the (additional) stored elastic energy in the isothermal (additional)
deformation γ at constant temperature TI , from the fundamental state I to the cur-
rent state. The potential energy functional P is now defined by
P [u (x (t))] = W (γ) dV + PL [u (x (t))] . (2.1.20)
V
In words: The potential energy is equal to the sum of the increase of the elastic
energy for isothermal deformations and the potential energy of the external loads.
2.2 Theorems on stability and instability 11
Hence, stability for the class of problems discussed depends on isothermal con-
stants.† The question of stabilty when TI is not constant is still unsolved (which is
important, for example, in problems with thermal stresses).
In our discussion on the stability of discrete systems, we have seen that we need
measures to be able to specify expressions like “small disturbance” and “not large
deviations.” In our discussion of the stability of continuous systems, which is gov-
erned by the character of the potential energy, we need a measure for the displace-
ments. A suitable measure is the L2 -norm of the displacement field, defined by
u (x (t))2 = 1 ρ u (x (t)) · u (x (t)) dV, (2.2.1)
M
V
where M is the total mass of the elastic body. We shall employ the same measure
for the initial disturbance. We shall assume that the potential energy functional is
regular in the following sense. On every ball u = c in the function space of kine-
matically admissible displacement fields u (x (t)) where the radius c is sufficiently
small, the energy functional P [u (x (t))] has a proper minimum that is a continuous
function d(c) in the range of c under consideration. The potential energy functional
is called positive-definite if the function d(c) is a (positive) increasing function in a
range 0 ≤ c < c1 . The functional is called indefinite if the function d(c) is a (negative)
decreasing function for 0 ≤ c < c1 .
We are now in a position to formulate the stability criterion: The equilibrium
in the fundamental state is stable if the potential energy functional P [u (x (t))] is
positive-definite.
To show this, we introduce the notation
† In the literature, one frequently encounters vague and loose statements to the effect that buckling
is “rapid” and that it is therefore “reasonable” to assume that the motion is adiabatic. This would
imply that elastic stability would be governed by the adiabatic elastic constants rather than by the
isothermal elastic properties. This reasoning is erroneous as follows from the foregoing analysis.
A simple example is a strut with pinned ends under a compressive load N. For sufficiently small
values of the compressive load the straight configuration is stable, and this stability is manifested
by a non-vanishing fundamental frequency. This frequency decreases when the critical Euler load
N1 is approached, and it vanishes for N1 . The motion at the critical load is thus infinitely slow, and
hence isothermal.
‡ Duhem called it the “ballistic energy.”
12 Continuous Elastic Systems
In words: For a given, sufficiently small initial disturbance and a given total energy,
the displacements at t > 0 are bounded by the value of u (x(0)).
For an instability criterion, we need the following assumptions:
where the elastic potential W (γ) is expanded about its value in the fundamental
state
2
∂W 1 ∂W
W (γ) = γij + γij γk + · · · (2.2.6)
∂γij I 2 ∂γij ∂γk I
2.2 Theorems on stability and instability 13
This expansion contains linear and quadratic terms in the strains, which in their
turn contain linear and quadratic terms in the derivatives of the displacements.
Consider now the displacement fields u = αu1 (x), where u1 (x) is a given (fixed)
displacement field and α is a positive number. Stability is now determined by α
for α → 0. Now consider all kinematically admissible displacement fields u1 (x). It
has been demonstrated that it is impossible to show that W is positive-definite on
this basis. We shall also need restrictions on the derivatives of the displacements.
Strictly speaking, this condition implies that we cannot use the stability criterion.
However, it can be shown that for an indefinite elastic energy functional, the system
is unstable.
Let us now consider the first term in (2.2.6). By virtue of
so that the discussion of stability is focused on the integral in (2.2.12). We now define
a tensor of elastic moduli
∂2 W
EIijk ≡ , (2.2.14)
∂γ(ij ) ∂γ(k ) I
where W is written symmetrically with respect to γij and γk . Notice that this is not
the tensor of elastic moduli that is usually used in the theory of elasticity because
that tensor is defined by
∂2 W
Eijk ≡
0
, (2.2.15)
∂γ(ij ) ∂γ(k ) 0
where the index 0 indicates that the second derivatives of W must be evaluated in
the undeformed state. Then the tensor of elastic moduli for a homogeneous isotropic
material is given by
2ν
E0ijk = Gγ δikδj + δi δjk + δij δk . (2.2.16)
1 − 2ν
This tensor gives a complete description of the elastic material when the elastic
potential is given as
1 0 0 0
W0 = E γ γ , (2.2.17)
2 ijk ij k
i.e., when W 0 is a homogeneous quadratic form in the strain components. Notice
that here we have used Cartesian coordinates in the undeformed state. The defor-
mation tensor is
1
γij = (ui,j + uj ,i + uh,i uh,j ) , (2.2.18)
2
where u now denotes the displacements with respect to the undeformed configura-
tion and (),i = ∂ () /∂xi , where xi are Cartesian coordinates. The description of W 0
by (2.2.17) is in principle only valid for infinitesimally small strains, and even then
only the linear terms in the strain tensor are important. The expression for the elas-
tic potential may now be generalized by assuming that for finite strains (where the
quadratic terms in γij may become important), the elastic potential can still be rep-
resented by a quadratic function. The fact that quadratic terms in the strain tensor
may become important even when the linearized strain tensor θij ,
1
θij ≡ (ui,j + uj ,i ) , (2.2.19)
2
is small is immediately clear from the fact that θij 1 does not imply that the
linearized rotation tensor
1
ωij ≡ (ui,j − uj ,i ) (2.2.20)
2
is small.
2.2 Theorems on stability and instability 15
αβλµ 0 1 αβλµ 0 0
δW = E0 0
γαβ δγλµ + E0 δγαβ δγλµ , (2.2.22)
2
where the contravariant components of the tensor of elastic moduli are given by
βµ αµ βλ 2ν αβ λµ
Eαβλµ = G gαλ g + g g + g g . (2.2.23a)
0 0 0 0
1 − 2ν 0 0
The difference of the metric tensors in the undeformed configuration and in the
fundamental state is
ij
g0 − δij = O (ε) ,
where ε is largest principal extension in the fundamental state. Hence it follows that
Remarks
ϕ
x
N w N
x
u*
Figure 2.2.1
Notice that now the potential energy of the external loads is a nonlinear function
of the displacement, in contradistinction to what we have used in our theory.
This is due to the fact that we have used an auxiliary condition; namely we have
assumed that the strut is inextensible, which means
1 + u2 + w2 = 1 or u = 1 − w2 − 1.
In this case, we may say that buckling occurs when the energy supplied by the
loads is equal to the strain energy in bending. This may not be generalized.
The potential energy functional may now be written as (only for dead-weight loads)
1 1
P [u (x)] = Sij uh,i uh,j + Eijk θij θk dV
2 2
V
1 1
+ Eijk θij um,kum, + Eijk θk uh,i uh,j dV + Eijk uh,i uh,j um,kum, dV,
2 8
V V
(2.2.28)
where we have arranged the terms so that the integrants contain only terms of sec-
ond, third, and fourth degree, respectively, in the displacements. Writing
so that
Sij uh,i uh,j = Sij θhi θhj − 2Sij θhi ωhj + Sij ωhi ωhj . (2.2.32)
The components of the stress tensor Sij are small compared to those of the tensor
of elastic moduli Eijk , which are of O(G) where G is the shear modulus. In the
elastic range (for engineering materials) the strains must be small so that only terms
involving the rotations might compete with terms with Eijk . This is only possible
for large rotations, so the last term in (2.2.32) is the principle term. To show that the
18 Continuous Elastic Systems
ω = ω0 + O (ε)
ω = O(ε),
i.e., the rotations and the strains are of the same order of magnitude.
For conservative dead-weight loads, and under the assumption that the material
follows the generalized Hooke’s Law, the potential energy is given by
1 1
P [u (x)] = Sij uh,i uh,j + Eijk γij γk dV. (2.3.1)
2 2
As discussed in Section 2.2, stability is primarily determined by the character of P2 ,
given by
1 1
P2 [u (x)] = Sij uh,i uh,j + Eijk θij θk dV. (2.3.2)
2 2
If the second variation is positive-definite, the equilibrium is stable, and if the
second variation is indefinite (or negative-definite), the equilibrium is unstable. The
second variation is unable to give a valid decision on the stability of instability in the
critical case that it is semi-definite positive. Let u (x) be a minimizing displacement
field. Then
for all kinematically admissible displacement fields ζ(x) and for sufficiently small
values of ε ∈ R. Here and in the following it will be assumed that ζ is continuously
2.3 The stability limit 19
This equation is the variational equation for neutral equilibrium. From the func-
tional (2.3.2), we now obtain for the bilinear term
1
P11 [u, ζ] = Sij uh,i ζh,j + Eijk θij (ζk, + ζ ,k) dV, (2.3.7)
2
where ζk are the Cartesian components of ζ.
Due to the symmetry of Eijk in the indices k , we may write
1
Eijk θij · (ζk, + ζ ,k) = Eijk θij ζk, .
2
Further, we introduce the notation
Here σk are the stresses corresponding to the linearized strain tensor in the absence
of prestresses. Thus (2.3.7) may now be rewritten as
(Sij uh,j + σhj ) ζh,j dV = 0, (2.3.9)
According to the principal theorem in the calculus of variations, we then must have
‡
Suppose (2.3.11) does not hold in a point x∗ , say, Sij uh,i + σhj j > 0. Then choose
2
xi − x∗i xi − x∗i − R2 for xi − x∗i xi − x∗i ≤ R2
ζ= 0 ∗
for xi − xi xi − xi ≥ R2∗
and thus, the surface integral in (2.3.10) vanishes. However, the volume integral is positive, so
(2.3.10) is violated, and hence (2.3.11) must hold.
20 Continuous Elastic Systems
Sy S
yx
Sy σy
) dy
Sxy
Sx σx
y
(1 + ε
dy ψ/ 2
) dx rxy
dx
(1 + ε x ψ/ 2
ψ/ 2 ω
I II III
Fundamental State No Rotation Final State
Figure 2.3.1
Notice that in the absence of prestresses Sij , these equations reduce to the equations
from the classical theory of elasticity. Performing differentiation by parts, we obtain
from (2.3.11)
The equilibrium equations and boundary conditions in the fundamental state I are
given by
Sij ,j + Xi = 0 in V
, (2.3.14)
Sij nj = Pi on Ap
where Xi are the mass forces and Pi are prescribed tractions on Ap . Using these
expressions, we can rewrite (2.3.12) and (2.3.13) as
These equations and boundary conditions were derived for the first time by Trefftz
(1930, 1933).
Different but equivalent equations were derived earlier by Biezeno and
Hencky (cf. C.B. Biezeno and R. Grammel, Engineering Dynamics, Vol. I). We
shall reproduce here their derivation for the two-dimensional case (to simplify the
analysis).
Consider a rectangular material element with dimensions dx and dy in the fun-
damental state, loaded by stresses Sx , Sy , and Sxy (see Figure 2.3.1).
The final state is reached in two steps: First, a deformation without a rotation
of the deformed element (state II), and then a rotation of the deformed element
(state III).
In state II, the element will not be in moment equilibrium under the forces
Sx , Sy , and Sxy acting on the deformed element. To restore equilibrium (to a first
approximation), we add additional (small) forces σx , σy , and τxy (τxy = τyx ). These
additional forces do not enable us to satisfy the equilibrium of moments exactly.
2.3 The stability limit 21
To reach this goal, skew-symmetric shear forces rxy (ryx = −rxy ) must be added. It is
obvious that the final rotation does not disturb the equilibrium of moments.
The equilibrium of moments requires
1
Sx dy dx ψ+ Syx dx (1 + εy ) dy − Sxy dy (1 + εx ) dx
2
1
− Sy dx ψdy − rxy dy dx − rxy dx dy = 0,
2
from which follows
1
2rxy = ψ(Sx + Sy ) + Sxy (εy − εx ) . (2.3.17)
2
The equilibrium of forces yields
Sx,x + Syx,y + X = 0
(2.3.19)
Sy,y + Sxy,x + Y = 0.
Substitution of these equations into (2.3.18) and using the relation (2.3.17) yields
1 1
σx,x + τyx,y − (Sxy ω) ,x − (Sy ω) ,y − [ψ(Sx − Sy )] ,y − [Sxy (εy − εx )] ,y = 0
4 2 (2.3.20)
1 1
τxy,x + σy,y + (Syx ω) ,y + (Sx ω) ,x + [ψ(Sx − Sy )] ,x + [Sxy (εy − εx )] ,x = 0.
4 2
With ω = 12 (v,x −u,y ) and ψ = v,x +u,y , for the first of the equations we finally
obtain (2.3.20)
1 1
σx,x + τyx,y − Sxy (v,x −u,y ) − Sy (v,x −u,y )
2 ,x 2 ,y
(2.3.21)
1 1
− [(v,x + u,y ) (Sx − Sy )],y − [Sxy (v,y − u,x )],y = 0.
4 2
The general result from Biezeno and Hencky may be written in the form
1 1
σij + Shi θhj − Shj θhi + Sih ωhj =0 (2.3.22)
2 2 ,i
or rewritten as
1 1
σij + Shi (θhj + ωhj ) − Shi θhj − Shj θhi = 0
2 2 ,i
or
1 1
σij + Shi uj ,h − Shi θhj − Shj θhi = 0. (2.3.23)
2 2 ,i
22 Continuous Elastic Systems
The additional terms in (2.3.23) can easily be derived from the variational approach
by adding to the energy density the term 12 Shi θhj θij . This is small, of order O(ε) com-
pared to 12 Eijk θij θk , in which we had already admitted a relative error of O(ε).
Hence it follows that the equations for neutral equilibrium derived by Biezeno
and Hencky and those derived by Trefftz are equivalent within the scope of our
theory.
We now continue with our general discussion of neutral equilibrium, and we
consider the case that
where u1 (x) is the buckling mode. The question now arises whether u1 is the only
kinematically admissible displacement field for which P2 [u (x)] vanishes. To investi-
gate this condition we might look for additional solutions of the equations of neutral
equilibrium, but it proves to be more useful to proceed as follows. Consider the set
of orthogonal (kinematically admissible) displacement fields. To define orthogonal-
ity, we introduce the positive-definite auxiliary functional
T2 [u (x)] ≥ 0. (2.3.25)
This functional defines the measure in the energy space. Applying similar arguments
as in the discussion of P2 , we find that the bilinear term must vanish,
where ρ is the mass density. This choice is motivated by Rayleigh’s principle for the
determination of the lowest eigenfrequency of small vibrations about an equilibrium
configuration, which states
P2 [u (x)]
ω2 = Min 1
. (2.3.30)
2
ρui ui dV
V
In the critical case of neutral equilibrium, ω vanishes, which means that the motion
is infinitely slow. It will turn out that the results are independent of the particular
choice for T2 .†
An arbitrary displacement field can always be written in the form
u (x) = a u1 (x) + u (x) , T11 [u1 , u] = 0. (2.3.31)
Namely,
T11 [u1 , u] = a T11 [u1 , u1 ] + T11 [u1 , u] ,
where the last term vanishes by virtue of (2.3.31). It follows that
T11 [u1 , u] T11 [u1 , u]
a= = , (2.3.32a)
T11 [u1 , u1 ] 2T2 [u1 ]
where we have used the relation
T2 [u1 + u1 ] = T2 [2u1 ] = 4T2 [u1 ] = T2 [u1 ] + T11 [u1 , u1 ] + T2 [u1 ] .
The second variation for an arbitrary displacement field can now be written as
P2 [u] = P2 [au1 + u] = a2 P2 [u1 ] + aP11 [u1 , u] + P2 [u] , (2.3.32b)
where the first term on the right-hand side vanishes because u1 is the buckling mode,
and the second term vanishes because P11 vanishes for all displacement fields, and
hence also for displacement fields orthogonal to u1 . In other words, it follows that
when P2 [u] = 0 for u = u1 , P2 [u] also vanishes for a displacement field orthogonal
to u1 . We may thus restrict ourselves to displacement fields that are orthogonal to
u1 . As P2 [αu] → 0 for α → 0, it is more suitable to consider the following minimum
problem:
P2 [u (x)]
Min = λ2 , (2.3.33)
T11 [u1 ,u]=0 T2 [u (x)]
where λ2 is the minimum of the left-hand member. If λ2 = 0, then we have a second
buckling mode, and when λ2 > 0, u1 is then the unique buckling mode. Let u2 be the
second buckling mode; then
P2 [u2 + εη] P2 [u2 ]
≥ = λ2 (2.3.34)
T2 [u2 + εη] T2 [u2 ]
and
T11 [u2 , u1 ] = T11 [η, u1 ] = 0. (2.3.35)
Because this equation must hold for arbitrary small values of ε, the term linear in ε
must vanish, i.e.,
for all displacement fields with T11 [u1 , η] = 0. In this condition, the displacement
field η is orthogonal to u1 . In our original condition of P11 [u, ζ] = 0, ζ was not sub-
mitted to this requirement. We shall now show that (2.3.37) also holds for arbitrary
displacement fields ζ. To show this, we write ζ = tu1 + η, whereη satisfies the orthog-
onality condition. Replacing η in (2.3.37) by ζ, we find
To investigate the stability of the critical case of neutral equilibrium, we write for an
arbitrary displacement field u (x)
T11 [uk, u] = T11 [uk, ah , uh + u] = ah T11 [uk, uh ] + T11 [uk, u] = ak. (2.3.41)
When the material follows the generalized Hooke’s Law, the expansion terminates
after P4 [u]; in other cases, higher-order terms follow. Because uh are buckling
modes, the first term on the right-hand side vanishes, P11 [ah uh , ζ] = 0, for all dis-
placement fields ζ, and hence also for u. Because we have already found m buckling
modes, P2 [u] must satisfy the relation
Let us now first consider the case that u = ≡ 0. We then only have to deal with the
terms P3 [ah uh ] and P4 [ah uh ]. A necessary condition for stability is that P3 [ah uh ]
vanishes and that P4 [ah uh ] ≥ 0. Hence it follows that from the mathematical point
of view, systems will generally be unstable because functions for which P2 = P3 =
0 and P4 ≥ 0 are exceptions. However, in applications this often happens due to
the symmetry of the structure. The conditions mentioned are necessary conditions;
to obtain sufficient conditions, we consider small values of u. In fact, the stability
conditions must be satisfied for small deviations from the fundamental state.
Suppose now that the necessary conditions are satisfied; thus the most important
terms in (2.3.42) are
namely, for u = O(a(h) , u(h) 2 ) these terms are of the same order of magnitude,
whereas all other terms have integrands of order O(a(h) , u(h) n ), n ≥ 5.
We now first consider the minimum problem,
P2 [v] + εP11 [v, η] + ε2 P2 [η] + P21 [ah uh , v] + εP21 [ah uh , η] ≥ P2 [v] + P21 [ah uh , v]
or
Because this inequality must hold for all sufficiently small values of ε, it follows that
P11 [v − v∗ , η] = 0. (2.3.47)
26 Continuous Elastic Systems
P2 [v − v∗ ] = 0. (2.3.49)
However, this result contradicts our assumption (2.3.43) that λm+1 > 0. This com-
pletes our proof that v is unique.
Equation (2.3.46) is quadratic in the buckling modes but linear in v, so the solu-
tion can be written as
Min P11 [u] = P4 [ah uh ] + P2 [ah akvhk] + P21 [ah uh , aka vk ] + · · · . (2.3.52)
ak=const.
we find
† The equation with ζ is preferable because then we do not have the side condition of orthogonality.
2.4 Equilibrium states for loads in the neighborhood of the buckling load 27
We shall now show that the results obtained are independent of the particu-
lar choice of the auxiliary functional T2 . Let T2∗ be a positive-definite functional
(T2∗ = T2 ), so then we may construct a minimizing displacement field
This field satisfies the same equations as the field v = ah akvhk(x) but is subjected to
a different orthogonality condition. The difference of these fields must satisfy the
homogeneous variational equation
which means that vhk − v∗hk is a linear combination of buckling modes, say
which does not satisfy one of the orthogonality conditions implied by T2 and T2∗ . The
minimum of the potential energy functional is now given by
where P2 ah akv∗hk may be written as
The second term on the right-hand side of (2.3.60) vanishes because P11 [ah uh , ζ] = 0
for all fields ζ, and hence also for ζ = v, and the last term vanishes because the argu-
ment is a linear combination of buckling modes. Hence it follows that the minimum
value of P [u] is independent of the particular choice of the auxiliary functional.
2.4 Equilibrium states for loads in the neighborhood of the buckling load
0 I II
x x +U x +U + u
Notice that this expression starts with quadratic terms because the fundamental
state is an equilibrium configuration, and also that this expression depends on U,
i.e., it refers to the fundamental state.
2.4 Equilibrium states for loads in the neighborhood of the buckling load 29
For sufficiently small loads, the potential energy will be positive-definite and the
equilibrium configuration will be stable and unique (Kirchhoff’s uniqueness theo-
rem). Now let λ ≥ 0 be monotonically increasing; then for a certain value of λ, say
λ = λ1 , the potential energy will become semi-definite-positive. (The case λ < 0 is a
different stability problem that may be treated similarly.)
We shall now expand the potential energy with respect to the load parameter λ
in the vicinity of λ = λ1 (assuming that such an expansion is possible), e.g.,
Notice that this equation is satisfied for u(x) = 0, the fundamental state, as it should
be. At the critical load, the infinitesimal displacement field is a linear combination
of buckling modes, i.e., u = ah uh . Assuming that for small but finite displacements
the field can also (approximately) be represented by this expression, we write
By known arguments, the first two terms on the right-hand side of (2.4.10) van-
ish. We shall now first consider the terms depending on u. First of all, we have
the positive-definite term P2 [u; λ1 ], thus the term (λ − λ1 ) P11 [ah uh , u; λ1 ] is impor-
tant because it contains u linearly, and is also linear in ah uh and λ − λ1 . The term
(λ − λ1 ) P2 [u; λ1 ] may be neglected because it is quadratic in u and is multiplied by
λ − λ1 . The next important term depending on u is P21 [ah uh , u; λ1 ], which is linear in
u and quadratic in ah uh . Other terms depending on u may be neglected because they
are either of higher order in u or they contain higher-order terms in ah uh or (λ − λ1 ).
We now determine the minimum with respect to u of the three terms mentioned,
Min P2 [u; λ1 ] + (λ − λ1 ) P11 [ah uh , u; λ1 ] + P21 [ah uh , u; λ1 ] . (2.4.11)
w.r.t. u
30 Continuous Elastic Systems
The solution to this minimum problem is unique. Suppose that u∗ is a second solu-
tion. Then the following equation must holds,
The solutions of this equation are linear combinations of the buckling modes, but
u and u∗ are orthogonal with respect to these buckling modes, so we must have
u − u∗ = 0. Guided by the structure of (2.4.11), we try a solution of the form
Because uMin (x) is the solution to (2.4.11), we have for a variation of this field, say
uMin + εη,
P2 [uMin + εη; λ1 ] + (λ − λ1 ) P11 [ah uh , uMin + εη; λ1 ] + P21 [ah uh , uMin + εη; λ1 ]
≥ P2 [uMin ; λ1 ] + (λ − λ1 ) P11 [ah uh , uMin ; λ1 ] , for ∀ε. (2.4.14)
This inequality must be satisfied for all sufficiently small values of ε ∈ R, which
implies
P11 [uMin , η; λ1 ] + (λ − λ1 ) P11 [ah uh , η; λ1 ] + P21 [ah uh , η; λ1 ] = 0. (2.4.15)
and
1
P11 [vhk, η; λ1 ] + P111 [uh , uk; λ1 ] = 0, (2.4.18)
2
where for (2.4.18) we have used the relation
1
P21 [ah uh , η; λ1 ] = ah akP111 [uh , uk, η; λ1 ] .
2
From these linear equations, vh and vhk can be determined.
In the following, we shall need the following property:
With
1 1 1 1
F2 [u] = F2 u + u = F2 [u] + F11 [u, u]
2 2 2 4
or
F11 [u, u] = 2F2 [u]
we find
F1 [u∗ ] = −2F2 [u∗ ] ,
so that
F2 [u∗ ] + F1 [u∗ ] = −F2 [u∗ ] .
Using this property, we find that the minimum of (2.4.11) is
Min P2 [u; λ1 ] + (λ − λ1 ) P11 [ah uh , u; λ1 ] + P21 [ah uh , u; λ1 ]
w.r.t .u
= −P2 [(λ − λ1 ) ah vh (x) + ah akvhk (x)]
= − (λ − λ1 )2 P2 [ah vh (x) ; λ1 ] − (λ − λ1 ) P11 [ah vh (x) , ah akvhk (x) ; λ1 ]
−P2 [ah akvhk (x) ; λ1 ] . (2.4.21)
Let us now first consider the case that P3 [ah uh ; λ1 ] ≡ 0. In this case, the term
P4 [ah uh ; λ1 ] is small compared to the cubic term, and may thus be neglected in
(2.4.10). The minimum of the remaining terms with u is given in (2.4.21), and
here the term P2 [ah akvhk (x) ; λ1 ] may be neglected because it is of fourth degree
in the amplitudes of the buckling modes. Further, the term (λ − λ1 ) P11 ah vh (x),
ah ak vh (x) ; λ1 ] may be neglected because it is of third degree in the amplitudes of
the buckling modes and multiplied by (λ − λ1 ). Finally, we may also neglect the first
term on the right-hand side of (2.4.21) because it is of order O[(λ − λ1 )2 a2 ], where
a represents the order of magnitude of the amplitudes of the buckling mode. The
only remaining term in (2.4.10) that still is to be discussed is (λ − λ1 )P2 [ah uh ; λ1 ].
This term is of order O[(λ − λ1 )a2 ] and may thus be important. Our final result is
now given by
In other words, to a first approximation we might have neglected the terms contain-
ing u to obtain the same result.
Let us now consider the case P3 [ah uh ; λ1 ] ≡ 0. Now we must take into account
the term P4 [ah uh ; λ1 ], and hence also the last term in (2.4.21). Further, the term
(λ − λ1 ) P2 [ah uh ; λ1 ] is important, whereas the first two terms on the right-hand
side of (2.4.21) may be neglected compared to this term (under the assumption that
P2 [ah uh ; λ1 ] ≡ 0).
In this case, our result is
+ O[a5 , (λ − λ1 ) a3 , (λ − λ1 )2 a2 ]. (2.4.23)
32 Continuous Elastic Systems
λ
λ
λ1
λ1
u u
Notice that the last two terms in this expression are the terms that decide on the
stability of neutral equilibrium (cf. 2.3.55).
In our discussion, we have restricted ourselves to cases where P2 [ah uh ; λ1 ] ≡ 0,
i.e., we have assumed that the derivative of P2 [ah uh ; λ] with respect to the load
parameter λ exists for λ − λ1 (the fundamental state). This means that we have
assumed that there are equilibrium states in the vicinity of the fundamental state.
This is not always the case (see Figure 2.4.2).
In the following, we shall see that the present theory only enables us to treat
branch points. Finally, we notice that when P2 [ah uh ; λ1 ] ≡ 0, this term must be neg-
ative because for λ − λ1 < 0, the equilibrium in the fundamental state is stable, i.e.,
Let us now consider the simple case that m = 1, i.e., there is only one buckling
mode, say u (x) = au1 (x). Let us further introduce the notation
First, consider the case A3 ≡ 0, A2 < 0. Then we obtain from (2.4.22)
λ
A3 > 0
A3 < 0
unstable
λ1
stable
a
Figure 2.4.3
Figure 2.4.4
34 Continuous Elastic Systems
λ λ
λ1
λ1
a a
A3 = 0 A4 > 0 A3 = 0 A4 < 0
(I) (II)
Figure 2.4.5
where
A4 ≡ P4 [uh , λ1 ] − P2 [vhk; λ1 ] .
The branch point is stable when A4 > 0 and unstable when A4 < 0.
The equilibrium condition reads
from which
a1 = 0 (fundamental state)
1 A (2.4.33)
a22 = − (λ − λ1 ) 2 .
2 A4
Because for real values of a2 the left-hand side is positive, the right-hand side must
be positive, i.e.,
λ ≷ λ1 for A4 ≷ 0. (2.4.34)
so the fundamental state is stable for (λ < λ1 ) and unstable for λ > λ1 .
Substitution of a22 into this condition yields
− 4 (λ − λ1 ) A2 ≥ 0, (2.4.36)
from which follows that the branched equilibrium state is stable for λ > λ1 and A4 >
0, and unstable for λ < λ1 and A4 < 0. These results are plotted in Figure 2.4.5 and
Figure 2.4.6.
2.4 Equilibrium states for loads in the neighborhood of the buckling load 35
rigid rod
Figure 2.4.6
Examples of structures with the behavior of (I) are the Euler column and plates
loaded in their plane. An example of a structure with the behavior of (II) is given in
Figure 2.4.6.
From Figure 2.4.3 and Figure 2.4.5, we see that branched equilibrium states for
loads below the critical load are unstable. Later, we shall prove that this is always
the case.† The converse that branched equilibrium states would always be stable for
loads exceeding the critical load is not true (except for the single mode case).
Let us now proceed with the discussion of Minah =const P[u, λ], which is given by
(2.4.32) when P3 [ah uh ; λ1 ] ≡ 0 and by (2.4.23) when this term is identically equal to
zero. The term P2 [ah uh ; λ1 ] is quadratic in the buckling modes and may be trans-
formed to a form containing only quadratic terms, which will be normalized so that
the coefficients are all equal to −1 (remember P2 [ah u; λ1 ] ≤ 0), so that we may write
We may normalize the load factor λ so that at the critical load, λ = λ1 = 1. Denoting
Min P [u; λ] by F [ai ; λ], our discussion of stability is reduced to the discussion of the
quadratic form
Aijkai aj ak
F (ai ; λ) = (1 − λ) ai ai + (2.4.39)
Aijk ai aj aka
where the term on top is important when Aijkai aj ak0, else the quartic term must be
used.
The equilibrium equations are obtained from δF = 0, for all variations δai , so that
3Aijkaj ak
2 (1 − λ) ai + = 0. (2.4.43)
4Aijk aj aka
Let δah = bh be a unit vector, i.e., b = bi bi = 1, and let ah = aeh , where e is a unit
vector and a is the amplitude of a. The equilibrium equations now read
3aAijkej ek
2 (1 − λ) ei + = 0, (2.4.45)
4a2 Aijk ej eke
We are now in a position to prove the statement that the branched equilibrium states
are unstable for loads below the critical load, i.e., for λ < 1.
To show this, we first notice that (2.4.46) must hold for all unit vectors b, and
hence also for b = e, so that
3aAijkei ej ej
1−λ+ ≥ 0. (2.4.47)
6a2 Aijk ei ej eke
from which
3aAijkei ej ek = −2 (1 − λ) , (2.4.49)
2.4 Equilibrium states for loads in the neighborhood of the buckling load 37
− (1 − λ) ≥ 0 resp. − 2 (1 − λ) ≥ 0,
and hence λ ≥ 1. It follows that for λ < 1 the equilibrium state is unstable.
Let us now consider stationary values of the multilinear forms
on the unit sphere t = 1. Introducing a Lagrangean multiplier µ, we look for sta-
tionary values of
i.e., the first variation of these forms must vanish. This yields
These equations are of the same form as our equilibrium equations – in other
words, the equilibrium equations yield stationary directions on the unit sphere
e = 1. Among these stationary directions, there are minimizing directions. Let
t = t∗ be a minimizing direction, and let
Aijkti∗ tj∗ tk∗ = A3 (t∗ ) = A∗3 ≤ 0 and Aijk ti∗ tj∗ tk∗ t ∗ = A4 (t∗ ) = A∗4 . (2.4.53)
For A∗3 ≡ 0, i.e., A∗3 < 0, the equilibrium state is unstable. For A∗3 ≡ 0 and A∗4 > 0,
the equilibrium state is stable. To show this, we need the following property† :
From the equilibrium equation (2.4.45) for e = t∗ and multiplied by t∗i , we obtain
− 2 (1 − λ) ≥ 0. (2.4.57)
Hence the branched equilibrium state is stable for (λ > 1). This property has been
proved for a minimizing direction t∗ . For other solutions of the equilibrium equa-
tions, the question of stability cannot be established in general.
† W. T. Koiter, Some properties of (completely) symmetric multi-linear forms . . . . Rep. Lab. For
Appl. Mech. Delft, 587 (1975).
38 Continuous Elastic Systems
λ other solutions
stable
1 unstable
1 1
A*3 = 0 A*3 = 0
A*4 < 0 (steepest descent)
0 0
0 0
a a
Figure 2.4.7
21−λ
a=− (A∗3 < 0)
3 A∗3
(2.4.58)
11−λ
a2 = − (A∗3 = 0) .
2 A∗4
εl
N
Figure 2.4.8
where ueq indicates that u is the displacement from the equilibrium state B to the
equilibrium state C. The first variation of P [ueq ; λ] is given by
or
∂P [ueq ; N]
ε = − . (2.4.61)
∂N
N I
II
B
NB δN
C
N1
A A1
}
O D ∆εl E
εl
Figure 2.4.9
40 Continuous Elastic Systems
I
λ − A3*
4 (1 − λ )
2
1 ∆ε = a 2 =
( )
9 A3* 2
unstable other solutions
A*3 < 0
A3*
0
0
ε
I I
λ A4* λ
1 1
unstable
other solutions
A*4
0 0 other solutions
0 0
ε ε
A*3 = 0, A*4 > 0 A*3 = 0, A*4 < 0
Figure 2.4.10
i. Our results are only valid in the immediate vicinity of the branch point. It is
not possible to assess the accuracy of the results for finite deflections from the
fundamental state.
ii. A structure will (unless prohibited) follow the path of steepest descent or weak-
est ascent. This suggests that for a structure with geometrical imperfections,
2.5 The influence of imperfections 41
the imperfections coinciding with minimizing directions are the most danger-
ous imperfections. It is therefore meaningful to consider these imperfections in
particular.
iii. The fact that the slope of the branched equilibrium path is always smaller than
that of the fundamental path implies that the stiffness of the structure after
buckling is always decreased.
In the foregoing discussions, it was assumed that the geometry of the structure and
the distribution of the loads were known exactly. Further, it was assumed that the
material was elastic and followed the generalized Hooke’s Law. However, in prac-
tice the quantities are not known exactly, e.g., it is impossible to manufacture a
perfectly straight column of constant cross section, or a perfect cylindrical shell. The
deviations from the “idealized structure” in the absence of loads are called initial
imperfections. Some structures are extremely sensitive to initial imperfection, e.g.,
cylindrical shells under axial compressive loads.
The primary effect of initial imperfection is that the fundamental state of the
idealized perfect structure, described by the displacement vector U from the unde-
formed state, does not represent a configuration of equilibrium of the actual imper-
fect structure. The bifurcation point of the perfect structure can also not be a bifur-
cation point of the structure with imperfections.
In our previous discussions, we have seen that for structures under a dead–
weight load in an equilibrium configuration, the expansion of the potential energy
functional starts with terms quadratic in the displacements. However, in the pres-
ence of imperfections the fundamental state is not an equilibrium configuration, and
hence the expansion of the potential energy functional will start with a term linear
in the displacements. Furthermore, if we restrict ourselves to small initial imper-
fections, this term will also be linear in these imperfections. This implies that our
energy functional F (ai ; λ) is replaced by the functional F ∗ (ai ; λ), defined by
∗ Aijkai aj ak
F (ai ; λ) = (1 − λ) ai ai + + µBi ai , (2.5.1)
Aijk ai aj aka
The stability condition is not affected by the linear term µBi ai , and is still given by
(2.4.44). Writing ai = aei , |e| = 1, the equilibrium equations become
3a2 Aijk ej ek
2 (1 − λ) aei + + µBi = 0, (2.5.4)
4a3 Aijk ej ek e
Multiplying this equation by ti , summing over i, and using the notation Aijkti tj tk =
A3 (t), we obtain
This equation is to be solved under the side condition a ∈ R+ . The stability condition
reads
Now consider the particular case that the stationary directions are minimizing direc-
tions, i.e., t = t∗ . Using the property
Min Aijk ai bj ck = Aijk ti∗ tj∗ tk∗ = A3 (t∗ ) = A∗3 < 0, (2.5.9† )
a=b=c=1.
† W. T. Koiter, Some properties of (completely) symmetric multi-linear forms . . . Rep. Lab. For
Appl. Mech. Delft, 587 (1975).
2.5 The influence of imperfections 43
λ λ −A3*
1 µ<0 1 µ>0
λ*
A3* < 0
0 0
0 * 0
a a a
Figure 2.5.1
These amplitudes are real for (1 − λ)2 > 3µA∗3 , and they are both positive when
µ < 0. When µ > 0, only the negative sign can be used because a ∈ R+ . These results
are shown in Figure 2.5.1.
One easily shows that for µ < 0, the path is stable for a < a∗ and unstable for
a > a∗ , and that for µ > 0 the path is stable. Notice that due to the presence of
imperfections, the branch point is not reached. The stability limit is now a limit
point and not a branch point, as was the case for the perfect structure.
In the foregoing discussion, we have made it plausible that imperfections in the
minimizing directions are the most harmful ones. To show this, we return to (2.5.4).
Multiplying this equation by ei and summing over i, we find
2 (1 − λ) a + 3a2 A3 (e) + µBi ei = 0. (2.5.16)
The stability condition (2.5.5) reads
1 − λ + 3aAijkbi bj ek ≥ 0 for ∀ b| b = 1. (2.5.17)
44 Continuous Elastic Systems
Let us now assume that e is known, and let the minimum of the trilinear form in
(2.5.17) for fixed e be C3 (e). We then have the following inequalities;
hence,
1−λ
a=− . (2.5.20)
3C3 (e)
Because we are interested in the reduction of the critical load due to imperfections
(1 < λ), it suffices to consider only negative values of C3 (e). Substitution of (2.5.20)
into (2.5.16) yields
A3 (e)
(1 − λ)2 2 − = 3µBi ei C3 (e) . (2.5.21)
C3 (e)
For the right-hand side of (2.5.21), we have the estimate
3µBi ei C3 (e) ≤ 3µC3 (e) ≤ 3µA∗ . (2.5.22)
3
The term between the square brackets in (2.5.21) is always larger than 1, for A3 (e) >
0 it is larger than 2, and for A3 (e) < 0 we have A3 (e) /C3 (e) ≤ 1 as then A3 (e) ∈
[C3 (e) , 0) ⊂ R− . The critical load occurs for µ < 0, so we may write
which implies λ∗ < λ, i.e., λ∗ is a lower bound for the critical load.
Let us now consider the case P3 [ah uh ; λ] ≡ 0 and Aijk ai aj aka indefinite. The
equilibrium equation now reads
2 (1 − λ) a + 4A∗4 a3 + µ = 0. (2.5.28)
2.5 The influence of imperfections 45
The stability condition (2.5.25) must hold for all b with b = 1, and hence also for
b = t∗ , so
1 − λ + 6A∗4 a2 ≥ 0. (2.5.29)
1 − λ∗
a∗1,2 = ± − (λ∗ < 1, because A∗4 < 0) . (2.5.31)
6A∗4
Because we are interested in the reduction of the critical load due to imperfec-
tions, it suffices to restrict ourselves to values of λ for which λ < 1. From (2.5.36), it
then follows that C4 (e) < 0. With the stability limit, we have
λ λ
1 1
other imperfections
other imperfections
λ*
*
λ
A4* < 0
A3* < 0
a A3* = 0 a
Figure 2.5.2
so that
1−λ
a1,2 = ± − . (2.5.39)
6C4 (e)
Substitution into (2.5.34) yields
A4 (e)
2 (1 − λ)3/2 1 − = ±µBi ei −6C4 (e), (2.5.40)
3C4 (e)
which determines the reduction of the critical load at the stability limit in the pres-
ence of imperfections in arbitrary directions. Let us start the discussion of this
expression with the right-hand member. Because B and e are unit vectors, we have
∗ −
Bi ei ≤ 1. Because C (e) ∈ A , 0 ⊂ R we have, by virtue of (2.5.37), −6C4 (e) ≤
∗
4 4 ∗
−6A4 , so that Bi ei −6C4 (e) ≤ −6A4 . On the left-hand side of (2.5.40), the term
between the square brackets is positive and greater than 1 when A4 (e) is positive,
and greater than or equal to 2/3 for A4 (e) < 0, so that we can write
3
(1 − λ)3/2 ≥ ∓ µ −6A∗4 . (2.5.41)
4
Because the left-hand member is positive, we must choose the negative sign in
(2.5.39) when µ < 0 and the positive sign when µ > 0. Hence, we may write
3
(1 − λ)3/2 ≥ |µ| −6A∗4 . (2.5.42)
4
Comparing this expression with (2.5.33), we find that λ∗ yields the largest possible
reduction of the critical load, and is thus a lower bound for the critical load.
Summarizing our results, we conclude that imperfections in the direction of the
minimizing directions always cause the largest reduction of the critical load.† The
results obtained so far are shown in Figure 2.5.2.
Notice that for imperfections in the direction of the minimizing direction, the
critical load is obtained as a limit point, whereas for other imperfections the critical
load λ > λ∗ may be obtained as either a bifurcation point or a limit point.
† This theorem was first proved by D. Ho (Int. J. Sol. Struct. 10, 1315–1330, 1974) for cubic systems.
2.6 On the determination of the energy functional for an elastic body 47
λ* λ*
0 0
0 µ 0 µ
Figure 2.5.3
In the foregoing section we have given the general theory while assuming that the
elastic energy functional is known. In this section, we shall determine the energy
functional for a new class of problems.
Our line of thought is depicted in Figure 2.6.1:
O U I u II
u0
U u
O* I* II*
Figure 2.6.1
48 Continuous Elastic Systems
Let O be the undeformed (stress-free) state for the idealized (perfect) structure.
The position of a material point in this configuration is given by the Cartesian com-
ponents of the position vector x. The position of this point in the fundamental state
I (equilibrium configuration) is given by x + U, and the position in an adjacent state
II (not necessarily an equilibrium configuration) by x + U + u.
Let O∗ be the undeformed (stress-free) state for the actual (imperfect) struc-
ture, obtained by superposition of a displacement field u0 on the state O, without
introducing stresses. The position of a material point is now given by the Cartesian
components of the position vector x + u0 , where u0 = u0 (x). In the state I∗ (not an
equilibrium configuration), the position of the material point is x + u0 + U. In the
adjacent state II∗ , the position is x + u0 + U + u, and we shall try to determine u so
that the state II∗ is an equilibrium state.
In our discussion, we shall need the metric tensors in the various states,
Further, we introduce the following notation for the linearized strain tensors and
the linearized rotation tensors:
1 1
θij = (ui,j + uj ,i ) ij = (Ui,j + Uj ,i )
2 2 (2.6.8)
1 1
ωij = (uj ,i − ui,j ) ij = (Uj ,i − Ui,j ) .
2 2
We shall now derive an approximate expression for the energy functional based on
the following assumptions:
Because we are dealing with dead-weight loads, the terms linear in u are balanced
by the energy of the external loads, so the potential energy is given by
!
1 1 1
P [u] = Eijk k uh,i uh,j + Eijk γij + (Uh,i uh,j + Uh,j uh,i )
2 2 2
V
" (2.6.10)
1
× γk + (Um,kum, + Um, um,k) dV.
2
This expression is fully exact for dead-weight loads and a material that follows the
generalized Hooke’s Law. To simplify this functional, we need estimates for the
various terms. Using the relations
1
γij = O(θ, ω2 ) (Uh,i uh,j + Uh,j uh,i ) = O (ω, θ) = O (ω) (2.6.12)
2
50 Continuous Elastic Systems
where
θ = θij max , ω = ωij max , and so on. (2.6.13)
Notice that for the estimate of the second term in (2.6.12), we have made use of the
fact that the fundamental state is linear. We now have the following estimates:
1
γij γk = O(θ2 , θω2 , ω4 )γij (Uh,i uh,j + Uh,j uh,i )
2
3 1
= O(θω, ω ) (Uh,i uh,j + Uh,j uh,i )(Um,kum, + Um, um,k) (2.6.14)
4
1
= O(2 ω2 ) kl uh,i uh,j = O(ω2 ).
2
Because buckling in the elastic range only occurs in slender structures, we have
ω θ.
Now consider the linear terms in (2.6.9),
Because in the energy functional these terms are balanced by the energy of the
external loads, we find for the equilibrium equations
is proportional to the external loads. Here Sij is the symmetric stress tensor of
Kirchhoff-Trefftz.† It follows that we may write
(1)
Sij = λSij , (2.6.17)
(1)
where Sij is the stress tensor for a unit load. Further, we shall normalize λ so that
at the critical load, λ = 1.
The first term in (2.6.10) is now proportional to ω2 , so that of the other terms in
(2.6.14), the term of O(θ2 ) may be comparable. Competition of the terms of O(ω2 )
√
and O(θ2 ) is only possible when θ = O(ω ). It follows that the term O(θ ω) is of
order O(3/2 ω2 ) and may thus be neglected compared to O(ω2 ). Terms of O(ω3 )
and O(2 ω2 ) in (2.6.14) may also be neglected. This means that the energy func-
tional (2.6.10) can be simplified to yield
1 (1) 1
P [u] = λS uh,i uh,j + Eijk γij γk dV (2.6.18)
2 ij 2
V
√
with a relative error of O( ). Notice that this expression is linear in the load factor
λ.
† Notice that this is not the same stress tensor as the one introduced in (2.2.7).
2.6 On the determination of the energy functional for an elastic body 51
Let us now consider the actual (imperfect) structure. The increment in the elas-
tic energy density going from state I∗ to state II∗ is
1 ∗ 1
WII∗ ∗ − WI∗∗ = E (∗ + γij∗ )(∗k + γk ∗
) − E∗ijk ∗ij ∗k
2 ijk ij 2
1 1
= E∗ijk ∗k γij∗ + E∗ijk γij∗ γk ∗
= E∗ijk k + Um,ku0m, + Um,ku0m,
2 2
1 1
× γij + uh,i u0h.j + uh,j u0h.i + (Uh,i uh,j + Uh,j uh,i ) (2.6.19)
2 2
1 1 1
+ E∗ijk γij + uh,i u0h.j + uh,j u0h.i + (Uh,i uh,j + Uh,j uh,i )
2 2 2
1 1
× γk + um,ku0m, + um,l u0m,k + (Um,kum, + Um, um,k) ,
2 2
so we may replace E∗ijk by Eijk . For our discussion of the terms with the imperfec-
tions, we shall need the following expressions,
1 0 1 0
θ0 = u + u0j ,i max and ω0 = u − u0i,j max . (2.6.21)
2 i,j 2 j ,i
We now have the following estimates,
Further, (2.6.19) contains terms that are quadratic in the imperfections, which can
be neglected. At the critical load, we have ω θ. We know that is small but finite,
and we can always choose deviations from the fundamental state as small as we need
so that θ . The leading term in (2.6.22) is of order ωω0 . The only other term
that may be important is the term O (ωθ0 ), and all other terms are small compared
to these terms. Using (2.6.20) and (2.6.17) and the fact that we have dead-weight
loads, we can simplify the energy functional corresponding to (2.6.19) to yield
∗ (1)
P [u] = P [u] + λSij u0h,i uh,j dV, (2.6.23)
V
where P [u] is the functional for the perfect structure defined in (2.6.18). Notice that
the term with the imperfections depends linearly on the load factor λ.
Let us now consider the effect of eccentric loads. In this case, only the functional
of the external loads must be modified. Let x denote the position of a material point,
and let ρX be the force per unit volume at this point. The potential energy of the
dead-weight load is then −ρX · u (x), where u denotes the displacement of the point
52 Continuous Elastic Systems
λ λ
1 1
λ*
*
λ A*3 = 0, A*4 < 0
A3* < 0
0 0
0 0
a a
Figure 2.6.2
in question. When the load is applied eccentrically, say, in a point y close to x, the
potential energy becomes
so that it explicitly shows the linear dependence on the load factor in the last
term. This expression is valid for λ ≥ 0, e.g., the equilibrium equations, when
P3 (ah uh , λ) ≡ 0, now read
u0 = a0h uh + u0 (2.6.28)
u = ah uh + u. (2.6.29)
2.6 On the determination of the energy functional for an elastic body 53
Expanding (2.6.31) with respect to λ about λ = 1, and using the fact that P2 [u, 1] =
0, we obtain
(1) 1
P2 [u, λ] = (1 − λ) Sij uh,i uh,j dV + · · · ≡ (1 − λ) P2 [u, 1] + · · · (2.6.32)
2
V
The last term in (2.6.33) may be neglected because it is small compared to the first
term. The buckling modes are now normalized so that
1 (1) (1)
S akal ukh,i u h,i dV = akal Sij ukh,i u h,j dV = akal δk = akak. (2.6.35)
2 ij
V V
Using (2.6.28) and (2.6.29), we find for the term with the imperfections in (2.6.30),
(1)
λSij a0kal ukh,i u h,j + a0kukh,i uh,j + a u h,j u0h,i + u0h,i u0h,j dV. (2.6.36)
V
The second term vanishes by virtue of (2.6.34), and the third term vanishes by virtue
of the orthogonality condition,
(1)
Sij a u h u0h,i dV = 0. (2.6.37)
V
The last term in (2.6.36) may be neglected because it is small compared to the first
term, so we have
(1)
λakal Sij ukh,i u h,j dV = 2λa0kal δk = 2λa0kak,
0
(2.6.38)
V
Applications
Here we have made use of the fact that the bar is incompressible, i.e., inextensional.
An explicit condition for incompressibility reads
−∆ l
x
N w(x) N
Figure 3.1.1
55
56 Applications
1 2 1 2
P2 [w; N] = EIw − Nw dx (3.1.7)
2 2
0
P3 [w; N] = 0 (3.1.8)
1 2 2 1 4
P4 [w; N] = EIw w − Nw dx. (3.1.9)
2 8
0
The variational equation for neutral equilibrium now follows from (3.1.7),
P11 [w, ζ; N] = [EIw ζ − Nw ζ ] dx = 0. (3.1.10)
0
Because this equation must hold for all kinematically admissible displacement fields,
we have
EIw + Nw = 0 (3.1.12)
w( ) = w(0) = 0. (3.1.14)
sin α = 0, (3.1.17)
where w1 (x) is the buckling mode and w(x) is the orthogonal remainder. We must
now consider the minimum problem, as per (2.3.44),
but the last term vanishes because P3 [w; N] ≡ 0. Conseqently, the quantity
that governs stability in the critical state is P4 [w1 ; N1 ] = 0 (see 2.3.54). Taking
w1 (x) = 1 · sin πx/ , we find
1 π6 πx πx 1 π4 πx
P4 [w1 ; N1 ] = EI 6 cos2 sin2 − N1 4 cos4 dx
2 8
0 (3.1.21)
2
1 π4 1 2πx 1 1 2πx 1 π4
= N1 sin2 − · 1 + cos2 dx = N1 ,
2 4 4 4 4 64 4
0
π 2i 2
N1 Et = E
2l2
E
1
0
0
∆l
−
l
Figure 3.1.2
It should be noted that these results are only valid for small deflections. Com-
parison with the exact solution shows that the results are valid up to deflections of
the middle of the bar of about 20% of its length.†
3.2 Bar with variable cross section and variable load distribution
Consider a bar with variable cross section loaded in compression by a normal force
N(x) and a load distribution P(x), clamped at one end (see Figure 3.2.1). The bar is
considered to be incompressible.
Consider an infinitesimal part of the bar in the deflected configuration (see Fig-
ure 3.2.2).
Due to the deflection of the bar, the load P (x) dx will move over a distance
x
0 dx [1 − cos ϕ (x)], and hence the contribution to the potential energy is given by
x
−P (x) dx [1 − cos ϕ (ξ)] dξ,
0 0
where B (x) denotes the bending stiffness. The normal force in the bar is given by
N (x) = N + P (ξ) dξ, (3.2.2)
x
P ( x) N
A
+x
l
Figure 3.2.1
[ ]
Figure 3.2.2
and hence we have N (x) = −P (x). Using this relation to rewrite the second term
in (3.2.1), we find
x
−N (x) [cos ϕ (ξ) − 1] dξ dx
0 0
x
= −N (x)
[cos ϕ (ξ) − 1] dξ + N (x) [cos ϕ (x) − 1] dx
0 0 0
= N ( ) [cos ϕ (x) − 1] dx + N (x) [cos ϕ (x) − 1] dx,
0 0
1 1
P2 [ϕ (x) ; λ] = B (x) ϕ2 (x) − λN (x) ϕ2 (x) dx (3.2.4)
2 2
0
P3 [ϕ (x) ; λ] ≡ 0 (3.2.5)
1
P4 [ϕ (x) ; λ] = λN (x) ϕ4 (x) dx. (3.2.6)
24
0
Because P3 [ϕ (x) ; λ] ≡ 0 at the critical load factor, and where P2 [ϕ (x) ; λ1 ] is semi-
definite-positive, stability is governed by the sign of P4 [ϕ (x) ; λ1 ]. The condition
P11 [ϕ (x) , ψ(x) ; λ1 ] = 0 yields
P11 [ϕ (x) , ψ(x) ; λ] = [B (x) ϕ (x) ψ (x) − λN (x) ϕ (x) ψ(x)] dx = 0, (3.2.7)
0
The first term vanishes when the ends of the bar are hinged or clamped. From the
second term, we obtain
[B (x) ϕ (x)] + λN (x) ϕ (x) = 0. (3.2.9)
3.3 The elastically supported beam 61
From the equation for neutral equilibrium (3.2.9) we obtain at the buckling load
λ1 N (x) ϕ1 (x) = − [B (x) ϕ1 (x]) , (3.2.11)
so that
1
P4 [ϕ1 (x) ; λ1 ] = − [B (x) ϕ1 (x)] ϕ31 (x) dx
24
0
1 1
= − B (x) ϕ1 (x) ϕ31 (x)0 + B (x) ϕ1 (x) · 3ϕ21 (x) ϕ1 (x) dx.
24 24
0
The first term vanishes when the ends are either clamped or hinged, so
1
P4 [ϕ1 (x) ; λ1 ] = B (x)ϕ2 2
1 (x) ϕ1 (x) dx > 0, (3.2.12)
8
0
and hence the equilibrium at the buckling mode is stable. Notice that N (x) may
also be a locally tensile force. In the case that A is an elastic hinge, we must add an
additional term to the potential energy given by 12 Cϕ2 (0). The boundary condition
for x = 0 then becomes
and in the expression for P4 [ϕ1 (x) ; λ1 ], from the stock term we now have a contri-
bution at x = 0,
1
B (0) ϕ1 (0) ϕ31 (0) > 0, (3.2.14)
24
so that for a elastically hinged bar the equilibrium at buckling is also stable.
† Cf., J. G. Lekkerkerker, On the stability of an elastically supported beam . . . (Proc. Kon. Ned. Ak.
Wet. B65, no. 2, 190–197 (1962).
62 Applications
w (x) N
Figure 3.3.1
The reaction of the springs per unit length of the beam is cw, and the corresponding
energy is 12 cw2 (x). The potential energy of this system is now given by
1 w2 1 2 2 − 1
P [w (x) ; N] = B + cw + N 1 − w dx, (3.3.1)
2 1 − w2 2
0
and hence
1 1 1
P2 [w; N] = Bw2 + cw2 − Nw2 dx (3.3.2)
2 2 2
0
P3 [w; N] ≡ 0 (3.3.3)
1 1
P4 [w; N] = Bw2 w2 − Nw4 dx. (3.3.4)
2 8
0
P11 [w, ζ; N] = (Bw2 ζ + cwζ − Nw ζ ) dx = 0. (3.3.5)
0
Bw ζ 0 (−Bw ζ + cwζ + Nw ζ) dx − Nw ζ 0
+
0 (3.3.6)
= Bw ζ 0 − Bw ζ 0 − Nw ζ 0 +
(Bw
+ Nw + cw) ζdx = 0,
0
w(0) = w( ) = 0. (3.3.9)
Notice that each term in this series satisfies the kinematic boundary conditions. To
determine the stability limit, we might substitute this series into (3.3.7); however,
then we need the fourth derivative of this series, and it is unknown beforehand
whether the series obtained is convergent. It is therefore more suitable to return to
the second variation as this only requires the second derivative of this series, which is
expected to be convergent in view of the fact that the dynamic boundary conditions
also contain a second derivative.
Substitution of (3.3.10) into (3.3.2) yields
%∞ &2 %∞ &2
k2 π2
1 kπx 1 kπx
P2 [w; N] = B − 2 ak sin + c ak sin
2 2
0 k=1 k=1
%∞ &2 ∞
1 kπ kπx k4 π4 k2 π2 2
− N ak cos dx = B 4 + c − N 2 ak.
2 4
k=1 k=1
(3.3.11)
This expression is positive when all its coefficients are positive. The coefficient of ak
becomes zero for
k2 π2 2
N1 = B + C. (3.3.12)
2 k2 π2
This is the stability limit. To obtain the lowest (critical) value of N, we must minimize
(3.3.12) with respect to k,
dN1 kπ 2 2
= 2 2 B − 3 2 C = 0, (3.3.13)
dk kπ
from which
1/4
kπ C
= . (3.3.14)
B
However, here we have assumed that k ∈ R+ , but k ∈ N+ , so k cannot generally take
on the value in (3.3.14). For k ∈ R+ , we have from (3.3.14) and (3.3.12)
√
NMin = 2 BC. (3.3.15)
√
It is now convenient to plot N/2 BC versus (C/B)1/4 /π for various values of
k ∈ N+ (see Figure 3.3.2).
64 Applications
N
k =1
2 BC k=2 k=3
STABLE
0
0 1 2 3
l
(C B )1 4
π
Figure 3.3.2
Suppose now that we have /π (C/B)1/4 = 3.5. We then investigate what the
value of N is for k = 3 and k = 4. From (3.3.12), we obtain for k = 3
9 √ 1 √ √
N= 2
BC + (3.5)2 BC = 2.096 BC,
(3.5) 9
√
whereas the actual minimizing value is N = 2 BC. Similarly, we obtain for k = 4,
√
N = 2.072 BC.
In general, for a sufficiently high wave number k we may assume that k ∈ R+ ,
and then NMin is given by (3.3.15). As indicated in Figure 3.3.2, equilibrium is stable
in the shaded region but not necessarily at the stability limit, i.e., on the curves
bounding the shaded region. Because P3 [w; N] ≡ 0, stability is governed by the sign
of P4 [w1 ; N1 ].
With N1 given by (3.3.12) and
kπx
w1 (x) = ak sin , k ∈ N+ , (3.3.16)
we obtain from (3.3.4)
2 2 2
1 kπ kπx kπ kπx 2 1 kπ kπx 2
P4 [w1 ; N1] = B − 2 ak sin ak cos − N1 ak cos dx
2 8
0
1 k6 π6 4 2 2kπx 1 k4 π4 4 4 kπx
= B a sin − N1 4 ak cos dx (3.3.17)
8 6 k 8
0
1 k6 π6 4 3 k6 π6 4 1 k2 π2 4 k4 π4
= B a − N1 3 ak = a B − 3C .
16 5 k 64 64 k 4
The sign of this expression depends on C, and the critical value is
k4 π4
C= B. (3.3.18)
3 4
3.3 The elastically supported beam 65
N
2 BC k =1
k=2 stable
k=3
1 unstable
0
0 1 2 3
l
(C B )1 4
π
Figure 3.3.3
1 k2 π2 kπx
P2 [w1 ; N] = − (N − N1 ) a2k 2 cos2 dx
2
0
1 k2 π2
= (N1 − N) a2k (3.3.20)
4
1 k2 π2
= (1 − λ) a2kN1 ,
4
and for N1 = Nmin and kπ/ given by (3.3.14), we obtain
1
P2 [w1 ; N] = (1 − λ) a2kc . (3.3.21)
2
The function F (ak; λ) is now given by
)
1 1 C 1 1 k2 π2 4
F (ak; λ) = c (1 − λ) ak −2
a = c (1 − λ) ak −
4 2
a , (3.3.22)
2 16 B k 2 16 2 k
Let us now consider the influence of imperfections on the behavior of this struc-
ture. In (2.6.39), we saw that for imperfections of the shape of the buckling mode,
we must add to (3.3.22) a term −2λak0 ak, hence
∗ 1 1 k2 π2 4
F (ak; λ) = c (1 − λ) ak −
2
a − 2λakak .
0
(3.3.23)
2 16 2 k
66 Applications
N1
I C
Et = −3 E
A E
−∆ l l
Figure 3.3.4
w ( ) = v· ( ) = − sin θ (3.4.1)
w ( ) = 1 − v·2 − 1 dy (3.4.2)
0
v ( ) = 1 − 1 − w2 dx. (3.4.3)
0
N l
v ( y) C
−θ y
A
– w (x)
x
B
Figure 3.4.1
68 Applications
Figure 3.4.2
1 w2 2
1 v··2
P [v (y) , w (x) ; N] = B + N 1 − w − 1 dx + B dy,
2 1 − w2 2 1 − v·2
0 0
(3.4.4)
where B is the bending stiffness of the bars. However, it is inconvenient to formulate
the problem this way due to the nonlinear kinematic side conditions (3.4.2) and
(3.4.3). It is more convenient to replace the rigid connection in A by the elastic
element of Figure 3.4.2, where both springs have a stiffness c. The corresponding
additional energy (penalty function) is given by
2 2
1 1
c w ( ) − 1 − v·2 − 1 dy + c w ( ) − 1 − 1 − w2 dx . (3.4.5)
2 2
0 0
Using the appropriate expansions, we find the following expressions for the second
and third variation:
1 2 1 2 1 ··2 1
P2 [v, w; N, c] = Bw − Nw dx + Bv dy + cw2 ( ) + cv2 ( ) (3.4.6)
2 2 2 2
0 0
1 ·2 1
P3 [v, w; N, c] = cw ( ) v dy − cv ( ) w2 dy. (3.4.7)
2 2
0 0
1 1
P11 [v, η, w, ζ; N, c] = Bw ζ − Nw ζ dx
2 2
0
1 ·· ··
+ Bv η dy+ cw ( ) ζ( ) + cv ( ) η( ) = 0. (3.4.8)
2
0
3.4 Simple two-bar frame 69
The second of these conditions can be simplified because the first two terms cancel,
so that
N
A1 sin α + A2 1 − = 0. (3.4.18)
c
For c → ∞, we obtain from this equation and the fourth equation of (3.4.17),
sin θ − sin α sin θ
A1 = , A2 = . (3.4.19)
sin α − α cos α sin α − α cos α
From the third equation of (3.4.17) and the fourth equation of (3.4.17), for c → ∞
we obtain
1 1
A3 = sin θ, A4 2 = − sin θ. (3.4.20)
2 2
From the first of the equations of (3.4.17), we now obtain the bifurcation condition
α2 2 sin α
+3=0 (3.4.21)
sin α − α cos α
or
α
tan α = . (3.4.22)
1
1 + (α )2
3
This bifurcation equation is valid for c → ∞, i.e., for our actual structure.
We now determine the value of P3 at the critical load,
1 1
P3 [v1 , w1 ; N1 , c → ∞] = lim cw1 ( ) v·2 1 dy − cv1 ( ) w2
1 dx . (3.4.23)
c→∞ 2 2
0 0
3 sin θ 3 sin θ
A1 = , A2 = , (3.4.27)
−(α )2 sin α (α )2
which follow from (3.4.19) and (34.22), after some algebra we find
3N1 sin3 θ 7 9
P3 [v1 , w1 ; N1 , c → ∞] = + . (3.4.28)
(α1 )2 20 2 (α1 )2
α1 = 1.1861π, (3.4.29)
Wmax
N
N
stable unstable
Figure 3.4.3
72 Applications
We consider again the simple two-bar frame, but now loaded symmetrically by two
compressive forces N (see Figure 3.5.1).
The stiff joint is again replaced by an elastic element, and the potential energy
is now given by
1 w2 2
P [v (y) , w (x) ; N] = B +N 1 − w − 1 dx
2 1 − w2
0
1 v··2 ·2 − 1
+ B + N 1 − v dy
2 1 − v·2
0
2 (3.5.1)
1
+ c w ( ) − 1 − v·2 − 1 dy
2
0
2
1
+ c v ( ) − 1 − 1 − w2 dx .
2
0
N l
N
Figure 3.5.1
3.5 Simple two-bar frame loaded symmetrically 73
The third variation is still given by (3.4.7) but vanishes due to the symmetry of the
structure and its loading. We will also need the fourth variation,
1 2 2 1 4 1 ··2 ·2 1 ·4
P4 [v, w; N, c] = Bw w − Nw dx + Bv v − v
2 8 2 8
0 0
2 2 (3.5.3)
1 1 ·2 1 1 2
+ c v dy + c w dx .
2 2 2 2
0 0
The kinematic condition (3.4.12) is now automatically satisfied because the bending
moments vanish at x = y = .
The dynamic boundary conditions are
w (0) = v (0) = 0
w ( ) = v· ( ) = − sin θ. (3.5.6)
πx πy N1 π2
w1 = a sin , v1 = a sin , = 2, (3.5.7)
B
and we have
a= sin θ. (3.5.9)
π
To investigate the stability of this neutral equilibrium state, for small displacements
from this configuration we write
πx πy
w (x) = a sin + w (x) , v (x) = a sin + v (y) , (3.5.10)
where w and v are orthogonal to w1 and v1 , respectively.
We now consider (see 2.4.11)
Min P2 [v, w; N1 , c → ∞] + (N − N1 ) P11 [v1 , w1 , v, w; N1 , c → ∞]
w.r.t. v, w
+ P21 [v1 , w1 , v, w; N1 , c → ∞] . (3.5.11)
74 Applications
Now
·
P11 [v1 , w1, v, w ; N1 , c → ∞] = − w1 w dx − v·1 v· dy = 0 (3.5.12)
0 0
1 1 1 ··2 1 ·2
P2 [v, w· ; N1 , c] = Bw2 − Nw2 dx + Bv − Nv dy
2 2 2 2
0 0
(3.5.13)
1 2 1 2
+ cw ( ) + cv ( ) .
2 2
From (3.4.7), we obtain
1 ·2
P3 [v1 + v, w1 + w; N1 , c] = c [w1 ( ) + w ( )] v1 + 2v·1 v· + v2 dy
2
0
1 2
− c [v1 ( ) + v ( )] w1 + 2w1 w + w2 dx
2
0
1 1
= cw1 ( ) v·2
1 dy − cv1 ( ) w2
1 dx
2 2
0 0
(3.5.14)
+ cw1 ( ) v·1 v· dy − cv1 ( ) w1 w dx + · · ·
0 0
1 1
= cw ( ) v·2
1 dy − cv ( ) w2
1 dx + · · ·
2 2
0 0
= P21 [v1 , w1 , v, w· ; N1 , c] + · · · .
Let v, w be the minimizing solution. The minimizing solution now must satisfy
the condition of 3.4.18
i.e.,
(Bw ζ − N1 w ζ ) dx + (Bv·· η·· − N1 v· η· ) dy
0 0
+ cw ( ) ζ( ) + cv ( ) η( ) (3.5.16)
1
+ cζ( ) v·2 1 dy − cη( ) w2
1 dx = 0.
2
0 0
3.5 Simple two-bar frame loaded symmetrically 75
which yields
2 1
(A∗1 + A∗2 ) + (A∗∗ + A∗∗
2 ) = 0. (3.5.25)
π2 2 1
The boundary condition (3.5.20), the first two conditions of (3.5.21), and the first two
conditions of (3.5.22) are automatically satisfied by (3.5.23). From the third equation
of (3.5.21), we obtain
π2 π 1 2 π2 c
−B 2 + N1 A∗∗ ∗
1 + (−N1 + c ) A1 = − a ,
4
where the first term vanishes due to (3.5.7), so that
−a2 π2
A∗1 = . (3.5.26)
4 2 (1 − N1 /c )
Similarly, we obtain from the fourth equation of (3.5.21),
a2 π2
A∗2 = . (3.5.27)
4 2 (1 − N1 /c )
Because A∗1 + A∗2 = 0, we obtain from the orthogonality condition (3.5.25),
A∗∗ ∗∗
1 + A2 = 0. (3.5.28)
Wmax
Figure 3.5.2
so that
1 π6 Ba4
lim P4 [v1 , w1 ; N1 , c] + P21 [v1 , w1 , v.w; N1 , c] = − < 0. (3.5.34)
c→∞ 2 16 6
The equilibrium state is thus unstable. The behavior is shown in Figure 3.5.2.
At first sight, this result may seem surprising as both bars behave like the Euler
bar. However, the result may be explained as shown in Figure 3.5.3.
Replace the forces N by two statically equivalent forces F (F > N). Let δ be the
horizontal and vertical components of the corner point where the loads are applied;
then
δ
F ≈ N + N. (3.5.35)
Let the deflection of the horizontal bar be
πy
v (y) = a sin ;
then δ is given by
1 ·2 1 π2 2 πx π2 a2
δ≈ v dy = a cos2 dx = , (3.5.36)
2 2 2 4 2
0 0
δ N
δ F
N
Figure 3.5.3
78 Applications
so that
π2 a2
F =N 1+ . (3.5.37)
4 2
For the post-buckling behavior of the Euler bar, we had
π2 a2
F = F1 1 + , (3.5.38)
8 2
where F1 is the critical load. From (3.5.37) and (3.5.38), we find
π2 a2
N = F1 1 − . (3.5.39)
8 2
This means that the equilibrium is already unstable for values of N smaller than the
critical load F1 .
We consider a beam with a thin-walled cross section, loaded by forces that cause
bending and torsion, as shown in Figure 3.6.1. Let (y0 , z0 ) be the center of shear.
The displacements at a point (x, y, z) are given by
Here α (x) is the specific angle of torsion, ψ0 (y, z) and ϕ (y, z) are the real and the
imaginary parts of an analytic function F (y + iz) = ψ0 (y, z) + iϕ (y, z), and 0 (y, z)
and 0 (y, z) are defined by
∂0 ∂0 ∂0 ∂0
= = ψ0 (y, z) , − = = ϕ (y, z) . (3.6.2)
∂y ∂z ∂z ∂y
*( y0 , z0 )
x y
Figure 3.6.1
3.6 Bending and torsion of thin-walled cross sections under compression 79
The approximations (3.6.1) are valid for t2 /b2 1 and b2 /L2 1, where t is the wall
thickness of the cross section, L is a characteristic wavelength, and b is a character-
istic dimension of the cross section. The center of shear is determined by
zψ0 (y, z) dA = yψ0 (y, z) dA = 0. (3.6.3)
A A
Further, the resultant of the normal stresses in the x-direction must vanish, which
yields
ψ0 (y, z) dA = 0. (3.6.4)
A
The expressions (3.6.1) further guarantee the vanishing of the stresses σy , σz,
and σyz.
In our calculations, we shall make use of the estimates
where b is the width of the cross section. To simplify the calculations, we shall use
the principal axes of inertia, i.e.,
yz dA = 0. (3.6.6)
A
where the dots represent terms containing third-order derivatives with respect to x.
With the torsional stiffness St defined by
2 2 .
∂ψ0 ∂ψ0
St = G − (z − z0 ) + + (y − y0 ) dA (3.6.8)
∂y ∂z
A
and using the relations (3.6.3), (3.6.4), and (3.6.6) we may rewrite the potential
energy expression to yield
1 2 1 2 1 2 1 2
P [u] = EIzv0 + EIy w0 + Eα dx + St α dx. (3.6.10)
2 2 2 2
0 0
with
(y − y0 )2 + (z − z0 )2 dA = Iz0 + Iy0 = i 2z0 + i2y0 A = i20 A. (3.6.13)
A
Finally, we obtain for the total potential energy
1 1 1 1 2 1 1 2 2
P [u, N] = EIzv2 2 2 2
0 + EI y v0 + Eα − Nv0 − Nw0 − Ni 0 α
2 2 2 2 2 2
0
1 2 (3.6.14)
− Nz0 α v0 + Ny0 α w0 + St α dx = P2 [u; N] .
2
Expression (3.6.14) shows that interaction between bending and torsion only occurs
in the terms with Nz0 α v0 and Ny0 α w0 . This condition implies that when the shear
center coincides with the center of gravity, there is no interaction. This is the case
when there are two axes of symmetry. When there is only one axis of symmetry, one
of the coupling terms vanishes, i.e., we then have one axis of buckling mode without
interaction and one with interaction.
We shall now restrict our discussion further to the case in which the values of
Iy , Iz, , St , and N are constant. The equations of neutral equilibrium follow from
P11 [v0 , w0 , α, η, ζ, χ; N] = [EIzv0 η + EIzw0 0 ζ + Eα χ
0
− Nv0 η − Nw0 ζ − Ni 20 α χ − Nz0 (α η + χ v0 ) (3.6.15)
+ Ny0 (α ζ + χ w0 ) + St α χ ] dx = 0.
3.6 Bending and torsion of thin-walled cross sections under compression 81
η = ζ = χ = 0 at x = 0 and x =
(3.6.18)
η = ζ= χ = 0 at x = 0 and x = ,
η = ζ = χ = 0 at x = 0 and x = . (3.6.19)
v
0 + Nv0 + Nz0 α = 0
w
0 + Nw0 − Ny0 α = 0 (3.6.20)
α + Ni 20 α + Nz0 v0 − Ny0 w0
− St α = 0.
For the simply supported beam, we have the dynamic boundary conditions
i.e., the vanishing of the bending moments Mz and My and of the bimoment.†
†
The bimoment is not a real moment; its dimension is FL2 .
82 Applications
We shall now consider the case of a simply supported beam in more detail.
Because the differential equations and the boundary conditions contain only even
derivatives, we try a solution of the form
πx πx πx †
v0 = B sin , w0 = C sin , α = A sin . (3.6.23)
These expressions satisfy the boundary conditions (3.6.21) and (3.6.22). Substitution
of these expressions into (3.6.20) yields
(Ny − N) B −Nz0 A =0
(Nz − N) C +Ny0 A =0 (3.6.24)
−Nz0 B+ Ny0 C +i 20 (Nt − N) A = 0,
where Nt is the lowest buckling load due to torsion. The condition for a non-trivial
solution is the vanishing of the determinant, i.e.,
When y0 and z0 are zero, the roots of this equation are N = (Ny , Nz, Nt ). Now let
at least y0 or z0 be nonzero. Then for N = min (Ny , Nz, Nt ), the left-hand mem-
ber of (3.6.26) is negative. The left-hand member can only become positive for
N < min (Ny , Nz, Nt ). This condition means that the smallest positive root of this
equation is smaller than Ny , Nz , and Nt , i.e., the actual buckling load is smaller than
the buckling load in bending and smaller than the buckling load in torsion.
Let us now consider the special case that the y-axis is an axis of symmetry, i.e.,
z0 = 0. Then we have
where for the first case there is no interaction, but there is interaction in the second
case. Let us now investigate this second case more closely. Equation (3.6.27) may be
rewritten in the form
% &
y20
N 1 − 2 − (Nz + Nt ) N + NzNt = 0.
2
(3.6.28)
i0
This equation is symmetric in Nt and Nz; hence without loss of generality we may
assume
† We are only interested in the lowest buckling mode k = 1. In general, one writes v0 =
1∞ kπx
k=1 Bk sin , and so on.
3.6 Bending and torsion of thin-walled cross sections under compression 83
η→ 0 and η → 1. (3.6.32)
1 1
λ = lim 1 + η − 1 − η + 2ηy20 /i 20 = 1, (3.6.33)
η→0 2 1 − y2 /i 2 η
0 0
1
λ= < 1. (3.6.34)
1 + y0 /i 0
1
< λ < 1. (3.6.35)
2
This result shows that under combined torsion and bending, the buckling load may
be considerably smaller than Nz or Nt , especially when y0 /i 0 → 1. An example of
such a structure is the thin-walled circular tube with an open cross section (see Fig-
ure 3.6.2).
In this case, we have
2 2 3 2
St = πGRt , = π
3
π − 4 R5 t, y0 /i 0 = √ . (3.6.36)
3 3 5
For η = 1, we obtain λ = 0.528, i.e., a reduction of 47% of the buckling load. We
emphasize here once again that the theory is only valid for b2 /L2 1.
*
2R
Figure 3.6.2
84 Applications
λ1 = 1 + ε1 , λ2 = 1 + ε2 , λ3 = 1 + ε3 , (3.7.2)
where ε1 (i = 1, 2, 3) are the principal strains. In (3.7.1), the elastic constant is taken
to be unity. We shall assume that the material is incompressible, i.e.,
λ1 λ2 λ3 = 1, (3.7.3)
λ1 λ2 = 1. (3.7.4)
We now consider the following states (see Figure 3.7.1), where σ1 h is the
force per unit width of the plate. Let (x, y) be the coordinates of a material point
in the undeformed state. The coordinates of this point in the fundamental state are
denoted by (x, y), where
y = λ2 y.
x = λ1 x, (3.7.5)
In the adjacent state, the coordinates are x, y , where
x = λ1 x + u, y = λ2 y + v. (3.7.6)
The square of the length of an infinitely small material element in the undeformed
state is given by
σ1 h σ1 h
h
λ 2h
l
λ1l II
I
x
σ1 h σ1 h
The components of the metric tensor in the adjacent state are thus
where λ1 and λ2 are the extension ratios in those directions. The first invariant of
the metric tensor is
h/2
1 2
P [u] = λ1 u,x + λ2 v,y + u + u,y + v,x + v,y dx dy
2 2 2
2 ,x
0 −h/2
(3.7.19)
h/2
+ σ1 u,x dx dy,
0 −h/2
where for the last term we have rewritten the term σ1 h u,x dx as a surface inte-
gral. The stability condition is now that P [u] > 0 under the nonlinear side condition
(3.7.18). Using this side condition to eliminate v,y , in the linear terms in (3.7.19), we
find
h/2
λ22 1 2
P [u] = λ1 − + σ1 u,x + u,x + u2,y + v2,x + v2,y
λ1 2
0 −h/2
(3.7.20)
λ2
+ (u,y v,x − u,x v,y ) dx dy.
λ1
The stability condition is still P [u] > 0 under the side condition (3.7.18).
A necessary condition that P [u] > 0 is the vanishing of the linear term in
P [u], i.e.,
λ22 1
σ1 = −λ1 + = −λ1 + 3 , (3.7.21)
λ1 λ1
which is the equilibrium condition in the fundamental state. The discussion of the
equilibrium in the fundamental state is now reduced to the discussion of
h/2
1 2 λ2
P [u] = u + u,y + v,x + v,y + (u,y v,x − u,x v,y ) dx dy (3.7.22)
2 2 2
2 ,x λ1
0 −h/2
under the nonlinear side condition (3.7.18). Notice that P [u] is a homogeneous
quadratic functional, i.e., P [u] = P2 [u]. A necessary condition for stability is that
P2 [u] ≥ 0 for very small displacements from the fundamental state, which means
that we may linearize the side condition to yield
A sufficient condition for stability is that P2 [u] > 0 under the linear side condition
(3.7.23).
Introducing a Lagrangian multiplier p (x, y) to take into account the side condi-
tion, we now consider the functional
h/2
P [u] + p (x, y) (λ2 u,x + λ1 v,y ) dx dy. (3.7.24)
0 −h/2
h/2
[u,x η,x + u,y η,y + v,x ζ,x + v,y ζ,y
0 −h/2
λ2
+ (u,y ζ,x + v,x η,y − u,x ζ,y − v,y η,x ) (3.7.25)
λ1
+ (λ2 u,x + λ1 v,y ) δp + p (x, y) (λ2 η,x + λ1 ζ,y )] dx dy = 0,
h/2 h/2
h/2
+ (v,x ζ)0 dy + (v,y ζ)−h/2 dx − (v,xx + v,yy ) ζdx dy
−h/2 0 0 −h/2
h/2 h/2
λ2 h/2 h/2
+ (u,y ζ)0 dy − (u,x ζ)−h/2 dx + (v,x η)−h/2 dx − (v,y η)0 dy
λ1
−h/2 0 0 −h/2
h/2
+ (λ2 u,x + λ1 v,y )δp dx dy = 0. (3.7.26)
0 −h/2
u,xx + u,yy + λ2 p ,x = 0
v,xx + v,yy + λ1 p ,y = 0 (3.7.27)
λ2 u,x + λ1 v,y = 0.
and at x = 0,
u (0) = 0, (3.7.29)
which implies
Taking into account these conditions, we find for the dynamic boundary conditions
h/2
λ2
u,x − v,y + λ2 p dy = 0 at x = (3.7.31)
λ1
−h/2
λ2
v,x + u,y = 0 at x = 0 and x = (3.7.32)
λ1
λ2
u,y + v,x = 0 at y = ±h/2 (3.7.33)
λ1
λ2
v,y − u,x + λ1 p = 0 at y = ±h/2. (3.7.34)
λ1
The condition (3.7.32) may be simplified to
due to (3.7.28). The system (3.7.27) may easily be shown to be of fourth order, and
we have two boundary conditions on each part of the boundary.
We shall now restrict ourselves to the lowest buckling load, and taking into
account the boundary conditions (3.7.28) and (3.7.29), we can try a solution for
u (x, y) of the form
πx
u (x, y) = U (y) sin . (3.7.36)
From the first of the equations of (3.7.27), it then follows that p (x, y) is of the form
πx
p (x, y) = P (y) cos , (3.7.37)
and from the second or third of the equations of (3.7.27), we find
πx
v (x, y) = V (y) cos . (3.7.38)
Substitution of these expressions into (3.7.27) yields
π2 π
Ü − 2
U − λ2 P = 0
π2
V̈ − 2 V + λ1 Ṗ = 0 (3.7.39)
π
λ1 V̇ + λ2 U = 0,
3.7 Infinite plate between flat smooth stamps 89
where ()· = d ( ) /dy. This is a set of three homogeneous linear ordinary differential
equations, which can be solved by the substitutions
which yield
π2 π
µ − 2 A
2
− λ2 C = 0
π2
µ − 2 B
2
+ λ1 µC = 0 (3.7.41)
π
λ2 A + λ1 µB = 0.
The condition for a nontrivial solution is the vanishing of the determinant of the
coefficients, which yields
2
π2 2π
µ − 2
2
−λ1 µ + λ2 2 = 0.
2 2
(3.7.42)
The eigenvalues are thus
π π
µ1,2 = ± , µ3,4 = ±λ , (3.7.43)
where λ = λ2 /λ1 .
Because we have two pairs of roots with opposite signs, we can also express the
solutions as hyperbolic functions instead of exponential functions. Hence, we can
write
2
2
U (y) = Ai1 sinh µi y + Ai2 cosh µi y
i=1 i=1
2
2
V (y) = Bi1 cosh µi y + Bi2 sinh µi y (3.7.44)
i=1 i=1
2
2
P (y) = Ci1 sinh µi y + Ci2 cosh µi y,
i=1 i=1
C1 = 0, B1 = −λA1 , (3.7.46)
λ2 − 1 π
B2 = −A2 , C2 = A2 . (3.7.47)
λ2
The boundary conditions (3.7.33) and (3.7.34), with the aid of (3.7.36) to (3.7.38),
can be rewritten to yield
π π
U̇ − λ V = 0, V̇ − λ U + λ1 P = 0, (3.7.48)
for y = ±h/2. Substitution of (3.7.45) into (3.7.48) and using (3.7.46) and (3.7.47)
yields
πh πh
l 1 + λ2 cosh A1 + 2λ cosh λ A2 = 0
2 2
(3.7.49)
πh 1 πh
−2λ sinh A1 − λ + sinh λ A2 = 0.
2 λ 2
Notice that λ = λ2 /λ1 is the single significant load parameter. Its value follows from
the condition for a non-trivial solution of (3.7.49), which leads to
2 2
λ + 1 tanh λθ = 4λ3 tanh θ, (3.7.50)
πh
θ= . (3.7.51)
2
Apart from the solution λ = 1, which corresponds to the fundamental state
only, (3.7.50) has one solution, λ > 1. Let us consider two limiting cases θ → 0
and θ → ∞.
3.7 Infinite plate between flat smooth stamps 91
For θ 1, we can use series expansions for the hyperbolic functions. Then
(3.7.50) can be rewritten to yield
1 2 2 2
(λ2 − 1)2 = λ θ [(λ + 1)2 − 4] + O(θ3 ). (3.7.52)
3
As for θ = 0, λ2 = 1 is a root of this equation; we try a solution of the form
λ2 = 1 + ε, 0 < ε 1, (3.7.53)
which yields
4 2
ε= θ + O(εθ2 ), (3.7.54)
3
so that
λ2 2
= 1 + θ2 . (3.7.55)
λ1 3
Using the incompressibility condition λ1 λ2 = 1, we find
1 π2 h2
λ2 = 1 + θ 2 = 1 +
3 12 2
(3.7.56)
1 π2 h2
λ1 = 1 − θ 2 = 1 − .
3 12 2
This result for the thin plate (h/ 1) is in complete agreement with the result for
the Euler column, and as it is derived from a two-dimensional theory, it justifies the
latter result.
For θ → ∞, (3.7.50) reduces to
from which follows λ = 3.383 (and of course, λ = 1). Hence, it follows that
1
λ1 = √ 0.55, (3.7.58)
3.383
which means a reduction in length of 45%. In this case, the deformations are concen-
trated near y = ±h/2. In the case of symmetric deformation, the bifurcation equa-
tion becomes
A2 (1 + λ2 ) cosh θ
=− . (3.7.60)
A1 2λ cosh λθ
92 Applications
−2λ−1 cosh λθ
A1 = . (3.7.62)
2 cosh λθ − (λ−2 + 1) cosh θ
where u and v are orthogonal (in some suitable sense) to the buckling mode. Stabil-
ity is now governed by the behavior of the functional (3.7.22) under the nonlinear
side condition (3.7.18). To make full use of the properties of the functional in the
critical case of neutral equilibrium, we consider the functional (3.7.22). Substituting
(3.7.65) into this functional, we find
h/2
1 2 2
PII − PI = a u1,x +u21,y +v21,x +v21,y +a (u1,x u,x +u1,y u,y + v1,x v,x +v1,y v,y )
2
0 −h/2
1 2
+ u,x + u2,y + v2,x + v2,y + λa2 (u1,y v1,x − u1,x v1,y ) (3.7.66)
2
+ λa (u1,y v,x + u,y v1,x − u1,x v,y − u,x v1,y ) +λ (u,y v,x − u,x v,y ) dx dy,
3.7 Infinite plate between flat smooth stamps 93
where the terms involving a2 vanish because u1 and v1 satisfy the equations for the
critical case of neutral equilibrium. Using (3.7.25) with η = u, ζ = v, and u = u1 ,v =
v1 , p = p 1 , we can rewrite the bilinear terms in (3.7.66) to yield
h/2
− ap 1 (λ2 u,x + λ1 v,y ) dx dy, (3.7.67)
0 −h/2
We now use this nonlinear condition to rewrite the term involving p1 in (3.7.68),
which yields
h/2
− ap 1 (λ2 u,x + λ1 v,y ) dx dy
0 −h/2
h/2
3
= a p 1 (u1,x v1,y − u1,y v1,x ) + a2 p 1 (u1,x v,y + u,x v1,y − u1,y v,x − u,y v1,x )
0 −h/2
+ a2 p 1 (u1,x v,y + u,x v1,y − u1,y v,x − u,y v1,x ) + ap 1 (u,x v,y − u,y v,x )
+ p[λ2 u,x + λ1 v,y + a2 (u1,x v1,y − u1,y v1,x ) (3.7.71)
+ a (u1,x v,y + u,x v1,y − u1,y v,x − u,y v1,x ) + u,x v,y − u,y v,x ]
"
+ µ (u1 u + v1 v) dx dy,
94 Applications
where we have taken into account the nonlinear side condition with the Lagrangian
multiplier p (x, y) and the orthogonality condition
h/2
(u1 u + v1 v) dx dy = 0 (3.7.72)
0 −h/2
with the multiplier µ.† We now minimize this functional with respect to u, v, p, and
µ for a fixed value of the amplitude a. Variation with respect to u yields
h/2
{u,x δu,x + u,y δu,y + λ (v,x δu,y − v,y δu,x )
0 −h/2
h/2
− [u,xx + u,yy + a2 (p 1 v1,y ),x + a2 (p 1 v1,y ),x − a2 (p 1 v1,x ),y
0 −h/2
u,xx + u,yy + a2 (p 1,x v1,y − p 1,y v1,x ) + a (p 1,x v,y − p 1,y v,y )
(3.7.75)
+ λ2 p ,x + a(p ,x v1,y − p ,y v1,y ) + p ,x v,y − p ,y v,x + µu1 = 0,
h/2
[v,x δv,x + v,y δv,y + λ (u,y δv,x − u,x δv,y )
0 −h/2
h/2
− [v,xx + v,yy + a2 (p 1 u1,x ),y − a2 (p 1 u1,y ),x
0 −h/2
+ a (p 1 u,x ),y − a (p 1 u,y ),x + λ1 p ,y
+ a (p u1,x ),y − a (pu,y ),x + (p u,x ),y − (p u,y ),x + µv1 ] δv dx dy (3.7.78)
h/2
+ (v,y − λu,x + a2 p 1 u1,x + ap 1 u,x + λ1 p + apu1,x + pu,x ) δv |−h/2 dx
0
h/2
+ (v,x + λu,y − a2 p 1 u1,y − ap 1 u,y − apu1,y − p u,y ) δv | 0 dy = 0,
−h/2
v,xx + v,yy + a2 (p 1,y u1,x − p 1,x u1,y ) + a(p 1,y u,x − p 1,x u,y )
(3.7.79)
+ λ1 p ,y + a(p ,y u,x − p ,x u1,y ) + p ,y u,x − p ,x u,y + µv1 = 0,
u = u 0 + au1 + a2 u2 + · · ·
v = v 0 + av1 + a2 v2 + · · ·
(3.7.82)
p = p 0 + ap 1 + a2 p 2 + · · ·
µ = µ0 + aµ1 + a2 µ2 + · · · .
96 Applications
Introducing these expansions into the differential equations and the corresponding
boundary conditions, we find to zeroth order
0
u,xx + u,yy
0
+ λ2 p ,x0 + p ,x0 v,y0 − p ,y0 v,x0 + µ0 u1 = 0
0
v,xx + v,yy
0
+ λ1 p ,y0 + p ,y0 u,x0 − p ,x0 u,y0 + µ0 v1 = 0
(3.7.83)
λ2 u,x0 + λ1 v,y0 + u,x0 v,y0 − u,y0 v,x0 = 0
h/2
u1 u 0 + v1 v 0 dx dy = 0,
0 −h/2
h/2
0
u,x − λv0,y + p 0 v0,y + λ2 p 0 dy = 0 for x= (3.7.85)
−h/2
Because the equations and the boundary conditions are homogeneous, the unique
solution is
u0 = v0 = p 0 = µ0 = 0. (3.7.88)
h/2
1
u,x − λv1,y + λ2 p 1 dy = 0 for x= (3.7.91)
−h/2
3.7 Infinite plate between flat smooth stamps 97
u1 = v1 = p 1 = µ1 = 0. (3.7.94)
h/2 h/2
2
u,x − λv2,y + λ2 p 2 dy = − p 1 v1,y dy for x= (3.7.97)
−h/2 −h/2
The present equations and boundary conditions are the conditions for the sta-
tionary value of the functional
h/2
∗ 1 22 2 2 2
(PII − PI ) = a 4
u,x + u2,y + v2,x + v2,y
2
0 −h/2
+ λ u2,y v2,x − u2,x u2,y (3.7.100)
+ p 1 u1,x v2,y + u2,x v1,y − u1,y v2,x − u2,y v1,x dx dy
h/2
u1 u2 + v1 v2 dx dy = 0. (3.7.102)
0 −h/2
This is the variational problem of a quadratic functional with the linear side condi-
tion, and has a second variation that is positive-definite under the requirement of
orthogonality of the field u2 , v2 with respect to the buckling mode u1 . The solution
of the variational problem thus represents a minimum of the functional (PII − PI )∗ .
Writing
Because πy/ is of order θ, we can write the buckling mode (3.7.64) in the form
πy πx
u1 = + O θ3 sin
π2 y2 4 πx
v1 = 1 − + O θ cos (3.7.105)
2 2
π 2πy πx
p1 = + O θ3 cos ,
2
where we must note that the derivative with respect to y of an error term θn is of
order θn−1 −1 .
We can now evaluate the expressions
π3 2 2πx
p 1,x v1,y − p 1,y v1,x = 1 + O θ sin
3
2π4 y 2 2πx
p 1,y u1,x − p 1,x u1,y = 4 1 + O θ sin
π2 2 2πx
u1,x v1,y − u1,y v1,x = 2 1 + O θ 1 − cos
2
π3 y 2 2πx
−p 1 v1,x = 3 1 + O θ sin (3.7.106)
π4 y2 2 2πx
p 1 u1,x = 4 1 + O θ 1 + cos .
3.7 Infinite plate between flat smooth stamps 99
π4 y2 2 2πx
p 1 v1,y = − 1 + O θ 1 + cos
4
3
π y 2πx
−p 1 u1,y = − 3 1 + O θ2 sin .
The equations (3.7.95) now become
πy πx π3 2πx
u2,xx + u2,yy + λ2 p 2,x = −µ2 sin − 3 sin
2 2
π y πx 2π4 y
v2,xx + v2,yy + λ1 p 2,y = −µ2 1 − cos −
2 2 4
2 (3.7.107)
π 2πx
λ2 u2,x + λ1 v2,y = − 2 1 − cos 2
2
h/2
πy πx 2 π2 y2 πx 2
sin u + 1− cos v dx dy = 0,
2 2
0 −h/2
h/2
2 π4 h3
u,x − λv2,y + λ2 p 2 dy = 4 for x= (3.7.109)
6
−h/2
The value µ2 in (3.7.107) is still unknown. The terms in (3.7.107) involving µ2 lead
to terms in u2 and v2 proportional to sin πx/ and cos πx/ , respectively, which do
not satisfy the orthogonality condition, and hence we require
µ2 = 0. (3.7.112)
where all numbers C1 − C6 are of order of magnitude unity. We have made use of
the fact that λ1 and λ2 are of order unity with an error of order π2 h2 /4 2 , but in some
places we had to allow for the more accurate relation λ2 /λ1 = λ = 1 + π2 h2 /6 2 .
It is easily verified that the expressions (3.7.113) satisfy the equations (3.7.107)
and the first boundary conditions (3.7.109) with relative errors of O h2 /L2 . For the
left-hand member of the second of the boundary conditions (3.7.108), we obtain
2
π 3 π4 h2 2πx π2 π4 h2 2πx π4 h2
+ cos − + C 3 cos + C 4
4 2 8 4 4 2 4 4
2
π π4 h2 2πx π2 2π4 h2 2πx π4 h2
−λ − cos − + C 1 cos + C2
4 2 8 4 4 2 4 4
4 2
π h 2πx π4 h2 π4 h2 2πx π4 h2
+ λ1 − 4 cos − + C 5 cos + C 6
4 4 4 4 4
2 4 2 2 4 2 4 2 4 2
π 3π h π π h π h π h 2πx
= + − 2+ − − + · · · cos
4 2 8 4 4 8 4 24 4 4 4
2 2 4 2 4 2 4 2 4 2 4 2
π π π h π h π h π h π h
− 2+ 2+ 4
− 4
+ C4 4 − C2 4 + C6 4 + · · · (3.7.114)
4 4 24 4
π4 h2 π4 h2 π4 h2 2πx
+ C3 4 − 2C1 4 + C5 4 cos
4 2 4 2
5 π h 2πx 5 π h
= + C3 + C5 − 2C1 4
cos + − + C4 − C2 + C6
24 24 4
4 2 4 2
π h 2πx 1 π h
=− 4
1 + cos + + C4 − C2 + C6
4 24 4
4 2
11 π h 2πx
+ + C3 + C5 − 2C1 cos .
24 4
It follows that
1
− C2 + C4 + C6 + =0 (3.7.115)
24
11
− 2C1 + C3 + C5 + = 0. (3.7.116)
24
For the left-hand member of (3.7.109), we obtain
!
πh π3 h3 2π π2 h π3 h3 2π π4 h3
2 − 3
− 2 + C1 3 + C2 4
16 96 8 2 2
2 4 3 2 4 3 4 3
π h π h π h π h π h
−λ + − 2 + C3 4 + C4 3
8 2 16 4 8 2 2
4 3 4 3 4 3 4 3 "
π h π h π h π h
+ λ2 − 4
− 4
+ C5 4 + C6 4 (3.7.117)
24 24 2 2
π4 h3 π4 h3
= − 4 + 4 (2C1 + C2 − C3 − C4 + C5 + C6 )
3
π4 h3 1 π4 h3
= + 2C1 + C2 − C3 − C4 + C5 + C6 − .
6 4 2 4
3.8 Helical spring with a small pitch 101
π6 h2 a4
Min (PII − PI )∗ = h 1 + O h2 / 2 . (3.7.120)
192 6
which shows that the critical state of neutral equilibrium is stable.
A comparison with Euler column theory requires the evaluation of
h/2
1 2
P2 [u1 , λ = 1] = a 2
u1,x + u21,y + v21,x + v21,y + (u1,y v1,x − u1,x v1,y ) dx dy.
2
0 −h/2
(3.7.121)
Using the expressions (3.7.105), we find
2 5
1π h 1 π4 h3 h 1 π2 h 1 π4 h3
P2 [u1 , λ = 1] = a2 + + O − +
2 48 3 5 2 16 3
(3.7.122)
1 π4 h2 a2 h2
= 4
h 1 + O 2 .
12
The ratio of the expressions (3.7.121) and (3.7.122) is π2 a2 /16 2 with a relative error
of order h2 / 2 . This is indeed in full agreement with the result of the Euler column
in the initial post-buckling range, cf. (3.1.23).
The elastic stability of helical springs was first dealt with by Hurlbrink (1910) and
Grammel (1924). Both ignored the shear elasticity of the spring. Shortly thereafter,
Biezeno and Koch took into account the effect of shear but they overestimated the
effect, implying as a consequence that any spring, however short, would be liable to
buckling, which is not confirmed by experiment. Later, Haringx (1942) took into
account the shear effect correctly. In the first part of this section we shall follow his
approach.†
† Cf. J. A. Haringx, On highly compressible helical springs and rubber rods, and their application for
vibration-free mountings (Philips Research Laboratories, 1950).
102 Applications
Figure 3.8.1
If the helical spring has a sufficiently small pitch when compressed, the deforma-
tion of one single coil under a certain load will differ little from that of an unclosed
circular ring lying in a flat plane. Consequently, we may imagine the compressed
helical spring as being replaced by a number of similar rings a connected by per-
fectly rigid elements b, as represented in Figure 3.8.1.
The influence of the axial force having been discounted as the successive coils
are imagined as being flat after compression of the spring, we now must see what
distortions occur as a result of the bending moments and transverse forces to be
transmitted by these flat coils (see Figure 3.8.2).
In the following, we shall assume that the principal axes of inertia of the cross
section of the ring are in the plane perpendicular to the plane of the ring. Let Sb1
and Sb2 be the bending stiffness of the cross section of the ring in the plane of the
ring and perpendicular to it, respectively, and let St be the torsional stiffness of the
cross section. Let u, v be the displacements due to N and D, respectively, and let ϕ
be the rotation due to M. Elementary calculations show
St Sb1 Sb2 · St
N= u, D= v, M= ϕ. (3.8.1)
2πR3 πR3 πR (Sb2 + St )
Haringx now smears the stiffness, i.e., he replaces the spring with n coils by a con-
tinuous rod. Let be the length of the spring; then the elongation of the spring is
2πR3 N
= Nn ≡ , (3.8.2)
St EA
D D
N
M
M
Figure 3.8.2
3.8 Helical spring with a small pitch 103
Z d x + du
N
ψ
X
M ψ
ψ
D dw
II
ds dv
dx dx
x = (1 − λ) x (3.8.12)
ds
= (1 − λ + u ) cos ψ + w sin ψ
dx
(3.8.13)
dv
= − (1 − λ + u ) sin ψ + w cos ψ.
dx
The elongation per unit length is thus
dv
γ= = − (1 − λ + u ) sin ψ + w cos ψ. (3.8.15)
dx
Further, the curvature of the rod is given by
dψ
κ= = ψ , (3.8.16)
dx
so we can write
X = N = −λEA. (3.8.18)
for the equilibrium of moments. Solving N and D from (3.8.19), our results are
Substituting these expressions into (3.8.20) and using the relations (3.8.13) to
(3.8.17), we find
1 − λ + u = (1 + ε) cos ψ − γ sin ψ
(3.8.24)
1
= (1 − λ cos ψ + ζsin ψ) cos ψ − (λ sin ψ + ζcos ψ) sin ψ
β
w = γ cos ψ + (1 + ε) sin ψ
(3.8.25)
1
= (λ sin ψ + ζcos ψ) cos ψ + (1 − λ cos ψ + ζsin ψ) sin ψ.
β
When there are no transverse shear forces in the spring (e.g., for a spring hinged
on both ends), ζ = 0 and, in other cases, ζ 1 in the case of small deflections ψ,
ζ = O (ψ).
For the determination of the buckling mode, (3.8.26) can be linearized, which
yields
1 1
αR2 ψ + λψ − λ2 1 − ψ+ ζ 1 − 1 − λ = 0, (3.8.27)
β β
where we have neglected the term in ζ2 . As mentioned previously, when the spring
is hinged at both ends, ζ = 0. Nonzero values of ζ occur, e.g., in the cases shown in
Figure 3.8.4.
106 Applications
Figure 3.8.4
or, linearized,
! "
1 ζ
ψ 1−λ 1− + dx = 0. (3.8.30)
β β
0
Let us now first consider the case that the spring is hinged at both ends. The
lowest buckling mode is then given by
πx
ψ = cos (3.8.31)
and the bifurcation condition becomes
1 π2 αR2
1− λ2 − λ + = 0, (3.8.32)
β 2
from which we obtain
1/2
β β2 π2 αβ R2
λ1 = − − . (3.8.33)
2 (β − 1) 4 (β − 1)2 β − 1 2
It follows that for sufficiently large values of R/ (short spring), there is no critical
load, i.e., no buckling can occur. We shall discuss this result in more detail later in
this section.
When the spring is clamped at both ends, ψ = 0 for x = 0 and x = , and for
a symmetric buckling mode, there are no transverse shear forces, i.e., ζ = 0. Then
(3.8.30) is satisfied by
2πx
ψ = sin (3.8.34)
3.8 Helical spring with a small pitch 107
+ β [− (1 − λ + u ) sin ψ + w cos ψ] } dx
2
(3.8.36)
1
− EA λ dx + λEA (−λ + u ) dx.
2
2
0 0
The equations for the equilibrium of forces are obtained by variations of this func-
tional with respect to u and w . Variation with respect to u yields
[(1 − λ + u ) cos ψ + w sin ψ − 1] cos ψ
0 (3.8.37)
− β [− (1 − λ + u ) sin ψ + w cos ψ] sin ψ + λ δu dx = 0,
where we have introduced the Lagrangian multiplier c because δw is not arbitrary,
as it must satisfy the kinematic condition 0 δw dx = 0.
The second equilibrium equation now reads
Multiplying (3.8.38) by cos ψand (3.8.40) by sin ψand adding the resulting equations,
we obtain
Multiplying (3.8.38) by sin ψ and (3.8.40) by cos ψ and subtracting the resulting equa-
tions, we find
Using (3.8.41) and (3.8.42), we can rewrite the functional (3.8.36) to yield
2
1
PII − PI = EA αR2 ψ2 + (λ cos ψ + c sin ψ)
2
0
(3.8.43)
1
+ (λ sin ψ − c cos ψ)2 + 2λu dx,
β
where we have omitted the constant terms in (3.8.36) because they are unimpor-
tant for our further discussion. We now still need an expression for u . Multiplying
(3.8.41) by β cos ψ and (3.8.42) by sin ψ and adding the resulting equations, we find
from which
1 1
u = −1 + cos ψ + λ 1 − sin2 ψ − c 1 − sin ψcos ψ. (3.8.44)
β β
The functional (3.8.43) can now be rewritten to yield
!
1
PII − PI = EA αR2 ψ2 − 2λ (1 − cos ψ)
2
0
" (3.8.45)
1 1
+λ 1+ 1−
2
sin ψ + c sin ψ + cos ψ dx.
2 2 2 2
β β
The second variation is now given by
1 1 c2
P2 [ψ; λ] = EA αR2 ψ2 − λψ2 + λ2 1 − ψ2 + dx, (3.8.46)
2 β β
0
where we have made use of the fact that c 1. When there are no kinematic con-
ditions, c = 0. In this case, the equation for neutral equilibrium is given by
2 1
αR ψ + λ 1 − λ 1 − ψ = 0, (3.8.47)
β
in full agreement with (3.8.27) with ζ = 0. There are no cubic terms, so P21 [au, u] =
0. The stability in the critical case of neutral equilibrium is thus governed by the
fourth degree terms, i.e., by
1 1 1 2 1 1
P4 [ψ ; λ] = EA λψ − λ 1 −
4
ψ +c 1−
4 2
ψ2 dx. (3.8.48)
2 12 3 β β
0
3.8 Helical spring with a small pitch 109
For a hinged-hinged spring, we have already discussed the buckling mode and the
critical load, cf. (3.8.31) to (3.8.33). The critical load λ1 given in (3.8.33) is real for
π2 R2 β
≤ . (3.8.50)
2 4α (β − 1)
The sign of P4 [ψ; λ] is determined by the coefficient of ψ4 in (3.8.49), i.e., by the sign
of
1
f (λ1 ) = λ1 − 4λ1 1 −
2
. (3.8.51)
β
Using (3.8.32), we obtain
π2 αR2 αR2
f (λ1 ) = λ1 − 4 λ1 − = 4π2 − 3λ1 . (3.8.52)
2 2
Let us now first consider the critical case where the equality sign holds in
(3.8.50), i.e.,
β
λ1 = . (3.8.53)
2 (β − 1)
In this case,
β 3β 3β
f (λ1 ) = − = < 0, (3.8.54)
β − 1 2 (β − 1) 2 (1 − β)
so the equilibrium is unstable in that point.
Let us now investigate what happens in the points where f (λ1 ) = 0, i.e., for
4π2 R2
λ1 = α. (3.8.55)
3 2
Substitution into the bifurcation condition (3.8.32) yields
2
β−1 4π2 R2 4π2 R2 2 αR
2
α − α + π = 0,
β 3 2 3 2 2
or
π2 R2 3β β
α= < , (3.8.56)
2 16 (β − 1) 4 (β − 1)
i.e., the equilibrium is unstable for
β 3β
≤ λ1 ≤ . (3.8.57)
4 (β − 1) 4 (β − 1)
We can now represent our results in the graph in Figure 3.8.5.
110 Applications
λ1
πx
ψ = sin
β / 2 ( β − 1) UNSTABLE l
UNSTABLE
β / 4 ( β − 1)
STABLE
β −1 l/R
2π α
β
Figure 3.8.5
For a spring with a circular cross section, no buckling will occur for
β−1 1 + 2ν
< 2π α = 2π . (3.8.58)
R β 2+ν
It is still an open question whether the unstable domain on the boundary for values
of λ1 such that β/4 (β − 1) < λ1 < β/2 (β − 1) does actually exist, or that it stems
from the approximation that the spring is built up from flat circular rings.
z1 z
y1
kx
0 y
Figure 3.9.1
3.9 Torsion of a shaft 111
In the fundamental state, we have pure torsion, and the only non-vanishing com-
ponents of the stress tensor are
∂F ∂F
τxy = , τxz = − , (3.9.1)
∂z ∂y
where F is the stress function of torsion. In the adjacent state, bending will occur, and
the point 0 will have displacements v0 (x) , w0 (x) along the y and z axes, respectively.
Assuming that Bernoulli’s hypothesis holds, the displacement u of a point (x, y, z)
is given by
It is now convenient to employ the principle axes of inertia. Using the relations
y = y1 cos kx − z1 sin kx
(3.9.7)
z = y1 sin kx + z1 cos kx,
Because we are dealing with buckling, we cannot neglect the stresses in the funda-
mental state. Recalling our general expression for the second variation in the case
of dead-weight loads,
1 1
P2 [u] = Sij uh,i uh,j + Eijkl θij θk dV, (3.9.9)
2 2
where we have already evaluated the second term (3.9.8), we are still in need of an
explicit form of the first term. Because for torsion we have only two nonzero stress
components, we find
1
Sij uh,i uh,j = τxy (u,x u,y + v,x v,y + w,x w,y ) + τxz (u,x u,z + v,x v,z + w,x w,z)
2 !
1 2
= τxy (−yv0 − zw0 ) (−v0 ) + v0 + νv0 (y − z ) + νw0 yz 2
2
"
1 2
× (νv0 y + νw0 z) + w0 + νv0 yz + νw0 (z − y ) (νv0 z − νw0 y)
2
(3.9.10)
2
!
1
+ τxz (−yv0 − zw0 ) (−w0 ) + v0 + νv (y2 − z2 ) + νw 0 yz
2 0
"
1 2
× (−νv0 z + νw0 y) + w0 + νv0 yz + νw0 (z − y ) (νv0 y + νw0 z) .
2
2
We shall now argue that the terms with third derivatives may be neglected. Consider
the term
% &
2 5
1 2 τv b
τxy × νv0 (y − z2 ) × νv
0 y dx dy = O
0
. (3.9.11)
2 5
A
In our discussion of ψxy , we have already neglected terms of O Ev20 b6 / 6 , and have
2 4 4
taken into account terms of O Ev0 b / . Hence, it follows that we can neglect the
contribution of (3.9.11) in the energy and, similarly, the other terms involving third
derivatives with respect to x.
The expression (3.9.10) can now be rewritten to yield
1
Sij uh,i uh,j = v0 v0 [yτxy + νyτxy − νzτxz] + w0 w0 [−νyτxy + zτxz + νzτxz]
2 (3.9.12)
+ v0 w0 [zτxy + νzτxy + νyτxz] + v0 w0 [νzτxy + yτxz + νyτxz] .
Integrating this expression over the cross-sectional area, we must note that
4 4
∂F
yτxy dy dz = y dy dz = yFnz ds = Fedge ynz ds = 0, (3.9.13)
∂z
A edge
This implies that after integration, the first and the second term between brackets
in (3.9.12) vanish and in the third and the fourth term between brackets the terms
3.9 Torsion of a shaft 113
Bv
0 + Ww0 = 0, Bw
0 − Wv0 = 0. (3.9.22)
Let us now consider the case that the shaft is supported at its ends. It is now
convenient to choose the origin of our coordinate system in the middle of the shaft,
114 Applications
v0 = w0 = 0 for x = ± . (3.9.23)
µ4 BC − iµ3 WD = 0
(3.9.26)
µ4 BD + iµ3 WC = 0.
v0 = C1 cos µx + C2 sin µx + C3 x2 + C4 x + C5
(3.9.29)
w0 = C1 sin µx − C2 cos µx + C6 x2 + C7 x + C8
v0 (x = ) = C1 cos µ + C3 2 + C5 = 0
(3.9.30)
w0 (x = ) = C1 sin µ + C7 2 = 0
Because C5 only occurs in the first of the equations of (3.9.31), the condition for a
non-trivial solution is
sin µ 0
1 2 1
− µ cos µ 2 µ
2
2 = 0, (3.9.32)
1 2
− µ sin µ −µ 0
2
from which follows
1
tan µ = − µ . (3.9.33)
3
The smallest non-negative root of this transcendental equation is
π
µ = 1.566 , (3.9.34)
2
which yields
B
W = 1.566π , (3.9.35)
2
which is the critical value of the torque W.
Let us now examine the boundary conditions more closely. We consider the end
cross section at x = (see Figure 3.9.2).
The axes y, z are fixed in space, and y1 , z1 are co-rotating axes. The displacement
u( , y1 , z1 ) is given by
which implies that the shear stresses τxy cause a bending moment
τxy u dy dx = Mz (3.9.37)
A
This result means that the load on the end faces is not directed along the unde-
formed axis. Bending moments My and Mz must be added. Evaluating the integrals
z, z1
τ xz
τ xy
y, y1
Figure 3.9.2
116 Applications
These moments are called by Ziegler† semi-tangential moments, i.e., the moments
are obtained by multiplying the torque by half the angle of rotation.
In the problem treated previously, the loading was conservative. Greenhill‡
treated a similar problem, but in his case the bending moments were absent. The
loading is then non-conservative. Greenhill’s boundary conditions are obtained
from our previous problem by superimposing loads
1 1
Mz = − Ww0 , My = − Wv0 (3.9.40)
2 2
for x = ± . The boundary conditions now become
Bw0 − Wv0 = 0, Bv0 + Ww0 = 0, x = ± . (3.9.41)
The differential equations are still given by (3.9.22), and the general solution is
v0 = C1 cos µx + C2 sin µx + C3 x2 + C4 x + C5
(3.9.42)
w0 = C1 sin µx − C2 cos µx + C6 x2 + C7 x + C8 ,
where µ = W/B. By arguments similar to those used in the previous problem, we
may restrict ourselves to the underlined terms. The kinematic boundary conditions
v0 = w0 = 0 at x = require
C1 cos µ + C3 2 + C5 = 0
(3.9.43)
C1 sin µ + C7 = 0,
and the dynamic boundary conditions become
−2C3 W = 0
(3.9.44)
2BC3 + WC7 = 0,
which yields
C3 = C7 = 0, C5 = −C1 cos µ , C1 sin µ = 0, (3.9.45)
and hence, µ = π, so that the critical torque is given by
B
W = 2π
. (3.9.46)
2
The correctness of this result, obtained from the equations for neutral equilibrium
for a conservative system, is in this case verified by the solution of the dynamic
problem.
x W
l
Figure 3.9.3
To show that this equivalent is not always justified, we consider the problem
shown in Figure 3.9.3.
The boundary conditions in this case are
v0 = w0 = 0, v0 = w0 = 0, at x = 0
. (3.9.47)
Bv0 + Ww0 = 0, Bw0 − Wv0 =0
at x = .
Bv
0 + Ww0
= 0, Bw
0 − Wv0
=0
Integrating the differential equations (3.9.22) twice and making use of the boundary
conditions at x = , we find
v0 = C1 cos µx + C2 sin µx + C3
(3.9.49)
w0 = C1 sin µx − C2 cos µx + C4 .
C1 + C3 = 0, −C2 + C4 = 0
(3.9.50)
C2 µ = 0, C1 µ = 0.
These results mean that for µ = 0, there is no neutral equilibrium (Ziegler, 1950).
The dynamic problem yields increasing amplitudes for non-vanishing W (flutter).
We shall analyze this phenomenon in the following (simpler) problem, in which
we consider a shaft without mass and a concentrated mass m at its end x = (see
Figure 3.9.4).
Because the shaft has no mass, the differential equations are unaltered. The
boundary conditions in this case are
v0 = w0 = v0 = w0 = 0 at x = 0
. (3.9.51)
Bv0 + Ww0 = 0, Bw0 − Wv0 = 0
at x = ,
Bw
0 + Ww0 = mv̈0 , Bw0 − Wv0 = mẅ0
W
m
W
x
l
Figure 3.9.4
118 Applications
where () = ∂ () /∂x and ()· = ∂ () /∂t. Introducing the complex function
V = V = 0 at x = 0, (3.9.54)
U = D1 eiµx + D2 x2 + D3 x + D4 . (3.9.56)
D1 + D4 = 0, iµD1 + D3 = 0. (3.9.57)
−µ2 D1 + 2 (1 − iµ ) D2 = 0
(3.9.59)
mω2 iµ mω2 2
− e − 1 − iµ D1 + − + 2iµ D2 = 0.
B B
The condition for a non-trivial solution is
−µ2 2 (1 − iµ )
mω2 mω2 2 = 0, (3.9.60)
− eiµ − 1 − iµ − + 2iµ
B B
which yields
mω2 2 2
[µ + 2(1 − iµ )(eiµ − 1 − iµ )] − 2iµ3 = 0, (3.9.61)
B
so that the square of the frequency is given by
2iµ3 B/m
ω2 = . (3.9.62)
µ2 2 + 2 (1 − iµ ) (eiµ − 1 − iµ )
It follows that ω2 is complex unless the denominator is purely imaginary, which is
not the case, as we shall show.
The fact that ω2 is complex implies that there is one root with a negative imag-
inary part, which implies that V, and hence v0 and w0 , increase exponentially with
time. This fact means that for µ = 0, i.e., when there is a torque, the system is always
3.10 Torsion of a shaft with a Cardan (Hooke’s) joint 119
so that
3B 3
ω =
2
1 − iµ + O(µ ) .
2 2
(3.9.63)
m 3 8
Although at first sight this result seems alarming, it is not important for actual
constructions because it is virtually impossible to apply a torque the way it is
assumed in this problem. To apply a torque, one usually makes use of the universal
(Cardan) joint, and then the system is conservative. We shall treat this problem in
the next section.
We consider a shaft loaded in torsion with a Cardan (Hooke’s) joint (see Fig-
ure 3.10.1).
Because large rotations may occur, we shall first derive an expression for the
rotation vector for finite rotations.
Let n be a unit vector along the axis of rotation, and let α be the angle of rota-
tion. The position vector of a point is denoted by r (see Figure 3.10.2).
The position of the point after rotation is denoted by r , and is given by
r = (r · n) n + {r − (r · n) n} cos α + (n × r) sin α
(3.10.2)
= r cos α + (r · n) n (1 − cos α) + (n × r) sin α.
z
y
W
x
x = −l x=l
Cardan’s joint
Figure 3.10.1
120 Applications
e2
r
α
r′ α
θ
e1
n
e3
Figure 3.10.2
with components ωi (i ∈ {1, 2, 3}) . Notice that two subsequent rotations do not yield
a rotation vector that is the sum of the two corresponding rotation vectors. We now
consider the rotation of the triad (e1 , e2 , e3 ) to obtain the rotation matrix. Let us
denote the rotated triad by e1 , e2 , e3 . The triad (e1 , e2 , e3 ) is fixed to the body
under consideration (see Figure 3.10.3).
With r = e1 and n = ω/(2 sin α/2), we obtain from (3.10.2),
ω1 ω ω × e1
e1 = e1 cos α + (1 − cos α) + sin α (3.10.4)
2 sin α/2 2 sin α/2 2 sin α/2
with
e2
e′2
e1′
e1
e3 e′3
Figure 3.10.3
3.10 Torsion of a shaft with a Cardan (Hooke’s) joint 121
and
) )
1 1
cos α/2 = (1 + cos α) = 1 − ω2 .
2 4
We can then rewrite this expression to yield
1 2 1 1 2 1/2
e1 = 1 − ω + ω3 e1 +
2
ω1 ω2 + ω3 1 − ω e2
2 2 2 4
(3.10.5)
1 1 2 1/2
+ ω1 ω3 − ω2 1 − ω e3 .
2 4
1 1 2 1/2 1 1 2 1/2
e3 = ω3 ω1 + ω2 1 − ω e1 + ω3 ω2 − ω1 1 − ω e2
2 4 2 4
(3.10.7)
1 2
+ 1− ω + ω2 e3 .
2
2 1
We can now write our results in the form
For our purpose, this expression is insufficient because we need quadratic terms for
the second variation of the elastic energy. Taking into account quadratic terms, we
obtain
1 2 1
Rij = δij 1 − ω + ωi ωj + εijkωk + O(ω3 ). (3.10.11)
2 2
Before we begin with the discussion of the buckling problem, let us make a
few remarks. The Cardan Joint was invented by the physician and mathematician
G. Cardan in Italy in the 16th century, but in most English-speaking countries it is
referred to as Hooke’s joint. The construction is sketched in Figure 3.10.4.
122 Applications
e3
B′
A′ m
θ
A
e2
e1
B
x = −l
x=l
W
Figure 3.10.4
The bars AA and BB are rigid and are rigidly connected perpendicular to each
other. The semi-rings are also rigid. The bearings in A, A , B, B are frictionless.
Along the bars AA and BB , we introduce the unit vectors m and , respec-
tively. These vectors are expressed in terms of the triad (e1 , e2 , e3 ), which is fixed in
space,
= − sin θ e2 + cos θ e3
(3.10.12)
m = cos θ e2 + sin θ e3 .
Let e1 , e2 , e3 be connected to the deformable shaft; thus
which yields
1 2 1 2 1/2 1
− sin θ cos (θ + ϕ) 1 − ω + ω1 − sin θ sin (θ + ϕ) ω1 1 − ω
2
+ ω2 ω3
2 3 4 2
1/2
1 1
+ cos θ cos (θ + ϕ) −ω1 1 − ω2 + ω3 ω2 (3.10.16)
4 2
3.10 Torsion of a shaft with a Cardan (Hooke’s) joint 123
1 2
+ cos θ sin (θ + ϕ) 1 − ω1 + ω22 = 0,
2
or
1/2
1 1
− cos ϕ ω1 1 − ω2 + ω2 ω3 cos (2θ + ϕ)
4 2
(3.10.17)
1 1 1
+ sin ϕ 1 − ω21 + ω23 sin θ cos (θ − ϕ) − ω22 cos θ sin (θ + ϕ) = 0.
2 2 2
For known ω, the angle ϕ can be determined from this (exact) equation. Because
we only need an approximate solution that is correct up to and including quadratic
terms, we can simplify this equation to yield
1 1 1
− ω1 + ω 2 ω 3 cos 2θ + ϕ + ω23 sin θ cos θ − ω22 cos θ sin θ = 0, (3.10.18)
2 2 2
from which follows
1 1 2
ϕ = ω1 − ω2 ω3 cos 2θ − ω3 − ω22 sin 2θ. (3.10.19)
2 4
The second variation is now obtained by adding to the functional (3.9.20) the
terms due to the rotations at x = ± ,
1 2 2
1
P2 [u] = B v0 + w0 − W (v0 w0 − v0 w0 ) dx
− 2 2
1 1 2
+ W ω2 ω3 cos 2θ + ω − ω2 sin 2θ
2
(3.10.20)
2 2 3 x=
1 1 2
− W ω2 ω3 cos 2θ − ω3 − ω22 sin 2θ .
2 2 x=−
With
1
+ W [−w0 cos 2θ + v0 sin 2θ] η x= (3.10.23)
2
1
+ W [−v0 cos 2θ − w0 sin 2θ] ς x=
2
1
+ W [w0 cos 2θ + v0 sin 2θ] η x=−
2
1
+ W [v0 cos 2θ − w0 sin 2θ] ς x=− = 0,
2
where η and ζ are arbitrary kinematically admissible displacement fields. Assuming
that x = ± , we have the kinematic conditions
v0 (± ) = w0 (± ) = 0. (3.10.24)
and at x = − , we have
1 1
−Bv0 + Wv0 sin 2θ − Ww0 (1 − cos 2θ) =0
2 2 x=−
1 1
−Bw0 + Wv0 (1 + cos 2θ) − Ww0 sin 2θ = 0.†
2 2 x=−
Bv
0 + Ww0 = 0, Bw
0 − Wv0 = 0. (3.10.26)
As already discussed in our previous examples, the general solution to these equa-
tions is
v0 = C1 cos µx + C2 sin µx + C3 x2 + C4 x + C5
(3.10.27)
w0 = C1 sin µx − C2 cos µx + C6 x2 + C7 x + C8
where µ = W/B.
We shall now first discuss the case that v0 is an even function in x. Then w0 is an
odd function in x, as follows from the differential equation. This means that we must
only consider the underlined terms in (3.10.27). Introduction of these expressions
into the kinematic boundary conditions yields
C1 cos µ + C3 2 + C5 = 0
(3.10.28)
C1 sin µ + C7 = 0.
† These boundary conditions can also be derived directly by requiring that the moments along and
m vanish.
3.10 Torsion of a shaft with a Cardan (Hooke’s) joint 125
where λ = µ .
First, notice that there are no solutions for λ 1. For θ = 0, we have the solu-
tion λ = π/2, which is half the value of Greenhill’s result. For θ = π/2, we have
λ = π (Greenhill’s value). Evaluating θ for given values of λ, we obtain the graph
shown in Figure 3.10.5.
2
λ
π
stable
2
v0 symmetric
w0 symmetric
1
arctan
0 1+υ
0 2 1 2
θ0
π 2
θ
π
Figure 3.10.5
126 Applications
which has roots that are shifted over a distance π/2 compared to the roots of
(3.10.31). The corresponding curve is drawn as a dashed curve in Figure 3.10.5.
The torsion angle follows from
W
θ − θ0 = , (3.10.33)
St
which yields
W B
θ0 = θ − =θ−λ . (3.10.34)
St St
This result is a straight line in the λ − θ graph. For a circular shaft, we have
θ0 = θ − λ (1 + υ) , (3.10.35)
and its tangent in the λ − θ plane is arctan (1/1 + υ) < π/4, which implies that this
line may be tangent to the curve. The part of the curve above this line then cannot be
reached, which implies that the critical torque is a discontinuous function of θ. From
the practical point of view, these results are not as important because the critical
value of the torque is very large.
( y0, z0 )
*
M
Figure 3.11.1
3.11 Lateral buckling of a beam loaded in bending 127
where the starred functions are added to obtain a uni-axial state of stress. However,
the values of these functions are small compared to those of the other terms, and
may for our purpose be neglected. In (3.11.1), ψ0 (y, z) is the warping function with
respect to the shear center and α (x) is the torsion angle per unit length. We know
from our general theory that for dead-weight loads, the second variation is given by
1 1
P2 [u] = Sij uh,i uh,j + Eijkl θij θkl dV. (3.11.2)
V 2 2
The second term in (3.11.2) is readily obtained in the form
1 1 1 2 1
Eijkl θij θkl dV = EIzv2
0 + St α + Eα 2
dx, (3.11.3)
2 0 2 2 2
V
where is the warping constant.
For the evaluation of the energy in the fundamental state, we notice that
Mz
S11 = σx = (3.11.4)
Iy
is the only non-vanishing stress component. The contribution to the second variation
is
1 2 1 2
S11 u1,1 + u2,1 + u3,1 dV =
2 2
σx u,x + v2,x + w2,x dV
2 2
V V
1 ∂v∗ 2
σx (−yv0 + ψ0 α ) + v0 − α (z − z0 ) +
2
= (3.11.5)
2 ∂x
.
∗ 2
∂w
+ α (y − y0 ) + dV.
∂x
First, notice that the first term between the brackets can be neglected because in it is
σx u2,x , which is always small compared to the term Eu2,x , which is taken into account
in (3.11.3). Second, the terms with an asterisk can be neglected as already discussed,
so that we obtain
1
Sij uh,i uh,j dV
2
V
Mz 2
= dx v0 − α (z − z0 ) + α2 (y − y0 )2 dy dz (3.11.6)
2Iy
Mz
(v0 + α z0 ) + α2 y20 − 2α (v0 + α z0 ) z − 2α2 yy0 + α2 z2 + α2 y2 dy dz.
2
= dx
2Iy
The first two terms between the brackets vanish after integration because the origin
of our (y, z) axes is the center of gravity, and the fourth term vanishes after integra-
tion because our axes are principle axes of inertia. Carrying out the integration, we
obtain
1
Sij uh,i uh,j dV
2
V
(3.11.7)
1 Mα2
= dx −Mα (v0 + α z0 ) + z(y + z ) dy dz .
2 2
2 Iy
128 Applications
and that
2
αα (y − y0 ) = O bα . (3.11.12)
so that by only taking into account the term αα (y − y0 ) in the first line of (3.11.10),
we make a relative error of order b2 / 2 , which is admissible for b2 / 2 1. A similar
† In the literature, the term with c is often missing. However, this is not always admissible.
3.11 Lateral buckling of a beam loaded in bending 129
argument holds for terms in the second line of (3.11.10), and our final result
becomes
1
Sij uh,i uh,j dV
2
= dx [(yτxy + zτxz) αα − αα τxy y0 + τxzα (v0 + α z0 )] dy dz (3.11.14)
= dx αα (yτxy + zτxz) dy dz + M α (v0 + α z0 ) dx.
where D (x) is the transverse shear force in a cross section. The first term in (3.11.14)
can now be rewritten by noticing that
(yτxy + zτxz)dy dz
! "
1 ∂ 1 ∂
= [(y2 + z2 )τxy ] + [(y2 + z2 )τxz) dy dz
2 ∂y 2 ∂z
(3.11.16)
1 ∂τxy ∂τxz
− (y2 + z2 ) + dy dz
2 ∂y ∂z
4
1 1 ∂σx
= (y2 + z2 ) (τxy ny + τxznz) ds + (y2 + z2 ) dy dz.
2 2 ∂x
edge
The line integral vanishes because τxy and τxz vanish at the edge, and using (3.11.4)
our final result is
1 M
(yτxy + zτxz) dy dz = z y2 + z2 dy dz
2 Iy
(3.11.17)
= M (z0 − c) .
The second variation for a beam loaded by a transverse shear force (dead-
weight load), applied in the center of shear-by-shear stresses according to Saint-
Venant’s theory, is now given by
1 1 2 1
P2 [u] = EIzv2
0 + St α + Eα − Mα v0
2
0 2 2 2
− cMα2 − M α (v0 + z0 α ) + M αα (z0 − c) dx (3.11.18)
1 2 1 2 1 2
= EIzv0 + St α + Eα − (Mα) v0 − c (Mα) α dx.
0 2 2 2
This expression is also (approximately) valid for distributed loads, provided that
they are distributed conforming to the shear stress distribution of Saint-Venant’s
theory. When the transverse shear force is not applied in the center of shear, we
must add the contribution of this force to the energy (see Figure 3.11.2).
130 Applications
z F
d
* ( y 0, z 0 )
y
Figure 3.11.2
F
F
destabilizing
F
F
stabilizing
Figure 3.11.3
3.11 Lateral buckling of a beam loaded in bending 131
EIzv
0 + (Mα) = 0
(3.11.22)
Eα − St α + Mv0 + c (Mα) + cMα = 0.
v0 = 0, α = 0. (3.11.23)
v0 = 0, α = 0, v0 = 0, α = 0. (3.11.25)
iii) Free end. In this case, we have only dynamic boundary conditions
EIzv0 = 0, EIzv
0 + (Mα) = 0,
(3.11.26)
Eα = 0, St α − Eα − Mv0 − cMα − c (Mα) = 0.
For narrow cross sections, |ψ0 | = O (ht) (see Figure 3.11.4), and the warping con-
stant is of order || = O(h3 t3 ), which means the warping is not important, so that
the elementary theory ( = 0) may be used. This is also true for a T profile (see
Figure 3.11.5).
However, for other profiles we have different conditions (see Figure 3.11.6).
t
h
Figure 3.11.4
center of shear
*
Figure 3.11.5
132 Applications
* * e.g.
shear *
center
( )
Γ = O b 5t , and St = O ( )
b 3t .
Figure 3.11.6
( )
Γ = O b6 , St = O ( )
b4 .
Figure 3.11.7
EIzv
0 + Mα = 0, −St α + Mv0 + 2Mcα = 0. (3.11.27)
By choosing
kπx kπx
α = A sin v0 = B sin , (3.11.28)
the boundary conditions (3.11.23) and (3.11.24) are satisfied identically. Substitution
into the differential equations yield
k4 π4 k2 π2
EIz B − MA = 0
4 2 (3.11.29)
k2 π2 k2 π2
− 2 MB + 2 (St − 2Mc) A = 0.
3.11 Lateral buckling of a beam loaded in bending 133
z 1 1 3 1 3
St = Gt3h, Iy = th , Iy = th
3 12 12
π Et3 h
h y Mcr =
6 2 (1 + v)
π Et2 5π t2 E
σcr = ≅ (v = 0.28)
t h 2 (1 + v) 8 h
Figure 3.11.8. Example of slender rectangular section.
When c = 0, we have
kπ
M=± EIzSt . (3.11.31)
In general, we have 2Mc/St 1, and then
kπ kπc EIz
M≈± EIzSt 1 ∓ . (3.11.32)
St
For example, if t/h = 0.1, h/ = 0.1, then σcr ≈ 0.002E, which for most construc-
tion materials is still within the elastic range.
)
π Eht3 πh 1
Mcr = 1− (1 + υ)
24 2 (1 + υ) 5 2
)
t2 h E πh 1
σcr = π 2 1− (1 + υ)
h 2 (1 + υ) 5 2
5π t2 h
≈ E for υ = 0.28 and h/ 1.
8 h2
Notice that this is (approximately) the same result as for the rectangular cross
section.
z
2 1
t z0 = h, c = h
15 5
1 1
EIz0 = Eht 3 , EIy = Eth3
y 48 36
h 3
Eht EIz0 1
St = , = (1 + v)
24 (1 + v) St 2
x
F
Figure 3.11.10
Let us now consider a beam clamped at one end and loaded by a trans-
verse shear force (applied through the shear center) at the other end (see Fig-
ure 3.11.10).
We further assume that the cross section is such that = 0, c = 0. In this case,
the differential equations are
EIzv
0 + [F ( − x) α] = 0,
(3.11.33)
F ( − x) v0 − St α = 0.
EIzv0 = 0, EIzv
0 + [F ( − x) α] = 0,
(3.11.35)
St α = 0 at x =
Integrating the first equation and taking into account the second condition of
(3.11.35), we obtain
EIzv
0 + [F ( − x) α] = 0. (3.11.36)
Integrating once again and taking into account the first and the third conditions of
(3.11.35), we obtain
EIzv0 + F ( − x) α = 0. (3.11.37)
Eliminating v0 from (3.11.37) and the second of the equations (3.11.33), we obtain
an equation in α,
F 2 ( − x)2
α + α=0 (3.11.38)
EIzSt
to be solved under the conditions
α=0 at x = 0, α = 0 at x = . (3.11.39)
The coefficient of ηk is
Ak+2 (k + 2) (k + 1) + λAk−2 = 0, (3.11.47)
so that
λ
Ak+2 = − Ak−2 . (3.11.48)
(k + 1) (k + 2)
The solution then becomes
λ 4 λ λ 8 λ λ λ 12
α = A0 1 − η + η − η + ··· . (3.11.49)
3.4 3.4 7.8 3.4 7.8 11.12
This solution satisfies the boundary condition at η = 0. The value of λ follows from
α = 0 for η = 1, i.e., the series between the brackets must vanish for η = 1, which
yields
λ = 16.104, (3.11.50)
so the critical load is
√
EIzSt
Fcr = 4.013 . (3.11.51)
2
For a beam with a rectangular cross section (t, h) the critical stress becomes
E t2 h
σcr = 4.013 . (3.11.52)
2 (1 + υ) h2
As we have seen, the problem in this case can be reduced to a single second-order
differential equation. This reduction is always possible when we have a beam with
136 Applications
at least one free edge. In this case, the first of the equations of (3.11.33) can be
integrated twice to yield
α=0 at x = 0, α = 0 at x = , (3.11.58)
which yields
F 2 4 105
= = 17.5, (3.11.61)
EIzSt 6
so that
√
EIzSt
Fcr = 4.18 , (3.11.62)
2
which gives an error of about 4% compared to the exact value.
3.12 Buckling of plates loaded in their plane 137
We consider a homogeneous, isotropic elastic flat plate, loaded in its plane. Without
going into details, we note that a plate theory can be derived under the following
assumptions:
Further, we note that the stiffness of the plate in its plane is considerably larger than
the bending stiffness, which means that after buckling, the displacements perpendic-
ular to the middle plane are considerably larger than the displacement in the middle
plane.
We now recall that stability in the fundamental state is governed by the func-
tional
1 1
P[u] = Sij uh,i uh,j + Eijkl γij γk dV, (3.12.1)
2 2
V
where γ is the nonlinear strain tensor. For plates, the non-vanishing components of
the strain tensor are
1
γ11 (z) = u,x + w2,x − zw,xx
2
1
γ22 (z) = v,y + w2,y − zw,yy (3.12.2)
2
2γ12 (z) = u,y + v,x + w,x w,y − 2zw,xy ,
where u(x, y), v(x, y) are the displacements in the mid-plane, w(x, y) is the dis-
placement perpendicular to the mid-plane, and z is the distance to the mid-plane
(−h/2 ≤ z ≤ h/2). The contribution of the second term in (3.12.1) to the energy
density can now be written as
1 E 2
Eijkl γij γk = γ (z) + γ22
2
(z) + 2υγ11 (z)γ22 (z) + 2(1 − υ)γ12
2
(z) .
2 2 (1 − υ2 ) 11
(3.12.3)
Integrating (3.12.3) over the thickness of the plate, we obtain the energy per unit
area,
Eh 1 2 2 1 2 2
V= u,x + w,x + v,y + w,y
2 (1 − υ2 ) 2 2
1 1 1
+ 2υ u,x + w2,x v,y + w2,y + (1 − υ) (u,y + v,x + w,x w,y )2 (3.12.4)
2 2 2
.
h2 2
+ w + w,yy + 2υw,xx w,yy + 2(1 − υ)w,xy ,
2 2
12 ,xx
Assuming that in the fundamental state the only non-vanishing stress compo-
nents are
S11 = Sx , S22 = Sy , S12 = Sxy ,
the corresponding energy per unit area of the middle plane is
1 1
Sx w,x + Sy w,y + Sxy w,x w,y h.
2 2
(3.12.5)
2 2
The energy functional for a plate loaded in its plane now becomes
Eh 1 2 2 1 2 2
P[u] = u,x + w,x + v,y + w,y
12 (1 − υ2 ) 2 2
1 1 1
+ 2υ u,x + w2,x v,y + w2,y + (1 − υ) (u,y + v,x + w,x w,y )2
2 2 2
. (3.12.6)
h2 2
+ w + w,yy + 2υw,xx w,yy + 2(1 − υ)w,xy dx dy
2 2
12 ,xx
1
+ h Sx w2,x + Sy w2,y + 2Sxy w,x w,y dx dy.
2
We now assume that the fundamental state is linear, i.e., that the stresses are pro-
portional to the loads. This assumption is valid when the rotations are of the same
order of magnitude as the strains, which are small. This is the case for a property-
supported plate (no rotations, as in a rigid body). Under these assumptions, we can
consider the x-, y-coordinates are the same as the coordinates in the undeformed
state.
Stability is primarily determined by the second variation, which is given by
!
Eh
P2 [u] = u2·x + v2,y + 2υu,x v,y + 2(1 − υ) (u,y + v,x )2
2 (1 − υ2 )
h2 2
+ w,xx + w2,yy + 2υw,xx w,yy + 2(1 − υ)w2,xy (3.12.7)
12
"
1 − υ2
+ Sx w2,x + Sy w2,y + 2Sxy w,x w,y dx dy.
E
We may now draw some important conclusions from the structure of this second
variation:
† To show this we recall that a2 + b2 + 2υab + 2(1 − υ)c2 = (a + υb)2 + (1 − υ)b2 + 2(1 − υ)c2 > 0
for ∀a, b, c, |υ| < 1, and a, b, c not vanishing simultaneously.
3.12 Buckling of plates loaded in their plane 139
ii) Because in (3.12.6) the third and fourth line together are semi-positive-definite
and the first two lines together are positive-definite, it follows that in the critical
state of neutral equilibrium P[u] > 0, so that for flat plates loaded in their plane,
the critical state of neutral equilibrium is stable. This conclusion does not hold
for curved plates and shells.
Let us now derive the conditions for neutral equilibrium, which follow from
!
Eh2
P11 [u, ζ] = 0 = w,xx ζ,xx + w,yy ζ,yy
12 (1 − υ2 )
+ υw,xx ζ,yy + υw,yy ζ,xx + 2 (1 − υ) w,xy ζ,xy (3.12.9)
"
12(1 − υ ) 2
+ [Sx w,x ζ,x + Sy w,y ζ,y + Sxy (w,x ζ,y + w,y ζ,x )] dx dy.
Eh2
Using the divergence theorem once, with υ as the unit normal vector to the edge,
we obtain positive in the outward direction
4
Eh3
[w,xx ζ,x υx + w,yy ζ,y υy + υw,xx ζ,y υy
12 (1 − υ2 )
edge
+ υw,yy ζ,x υx + (1 − υ)w,xy ζ,x υy + (1 − υ)w,xy ζ,y υx ] ds
Eh3
+ − (w,xx + υw,yy ),x ζ,x − (w,yy + υw,xx ),y ζ,y (3.12.10)
12 (1 − υ2 )
−(1 − υ)w,xyy ζ,x − (1 − υ)w,xyx ζ,y
.
12 1 − υ2
+ [Sx w,x ζ,x + Sy w,y ζ,y + Sxy (w,x ζ,y + w,y ζ,x )] dx dy = 0.
Eh2
where is the Laplacian operator. The differential equation for neutral equilibrium
is now given by
12 1 − υ2
w − (Sx w,xx + Sy w,yy + 2Sxy w,xy − Xw,x − Yw,y ) = 0, (3.12.12)
Eh2
140 Applications
where X and Y are the mass forces per unit area. Here we have used the equilibrium
equations
Sx,x + Sxy,y + X = 0
(3.12.13)
Sxy,y + Sy,y + Y = 0.
where t is the unit tangent vector to the edge, we can rewrite the line integral in
(3.12.11) to yield
4
(w,xx + υw,yy ) υ2x + 2(1 − υ)w,xy υx υy + (w,yy + υw,xx ) υ2y ζ,υ
edge
Let us now consider a part of the edge curve (see Figure 3.12.1).
It then follows that
ty = υx , tx = −υy . (3.12.16)
which holds when F (s) is a differentiable function. This implies that the edge curve
has no corners in s1 < s < s2 . In the following, we shall assume that the edge curve
is a smooth curve. From (3.12.15), we now obtain
(w,xx + υw,yy ) υ2x + 2(1 − υ)w,xy υx υy + (w,yy + υw,xx ) υ2y ζ,v edge = 0 (3.12.18)
y
t
x
Figure 3.12.1
3.12 Buckling of plates loaded in their plane 141
!
(1 − υ) (w,yy − w,xx ) υx υy + υ2x − υ2y w,xy ,s
s2
(1 − υ) (w,yy − w,xx ) υx υy + υ2x − υ2y w,xy ζ s = 0. (3.12.20)
1
First, we notice that the left-hand side of (3.12.20) vanishes identically along a closed
curve. Further, it vanishes along parts of the edge that are supported such that w =
0, and it also vanishes along a free edge provided that the edge in the parts of the
edge curve adjacent to the free part is properly supported.
In the following discussion, we shall assume that (3.12.20) is satisfied. Let us
now consider the various boundary conditions.
i) The free edge. Because there are no restrictions imposed on ζ and ζ,υ , their coef-
ficients must vanish, i.e.,
which means that the bending moment along the edge vanishes, and
(1 − υ) (w,yy − w,xx ) υx υy + υ2x − υ2y w,xy ,s
+ (w,xx − w,yy ),x υx + (w,xx + w,yy ),y υy (3.12.22)
12(1 − υ )
2
− [(Sx w,x + Sxy w,y ) υx + (Sy w,y + Sxy w,x ) υy ] = 0,
Eh
which means that the reduced transverse shear force vanishes at the edge.
ii) The simply supported edge. Here ζ ≡ 0 along the edge, so (3.12.19) is satisfied,
so that we have
∂w
w= = 0 at the edge (3.12.24)
∂υ
so that ζ = ζ,υ = 0 along the edge, which implies that the conditions (3.12.18)
and (3.12.19) are satisfied automatically.
Let us now apply our result to a square plate, simply supported at its edges,
loaded in compression by forces σ per unit length at its edges at x = 0 and x = a
(see Figure 3.12.2).
142 Applications
y=b
σ
σ
x
x=a
Figure 3.12.2
In this case, the only non-vanishing stress component in the fundamental state
is Sx = −σ. Further, there are no mass forces, so the differential equation for neutral
equilibrium is given by
Eh3
w + hσw,xx = 0. (3.12.25)
12(1 − υ2 )
In this case, we have
υx =1 υy =0 at x=a
υx = −1 υy =0 at x=0
υy =1 υx =0 at y=b
υy = −1 υx =0 at y=0
so that the boundary conditions x = a are
w,xx = 0 w = 0 at x = a, x = 0. (3.12.27)
Similarly,
w,yy = 0 w = 0 at y = 0, y = b. (3.12.28)
Because both the differential equation and the boundary conditions contain
only even derivatives, we can attempt a solution of the form
mπx mπy
w(x, y) = Cmn sin sin , (3.12.29)
a b
which satisfies all the boundary conditions.† Substitution into the differential equa-
tion yields
2 2 2
Eh3 mπ n2 π2 m2 π2
+ − hσ 2 = 0. (3.12.30)
12(1 − υ )
2 a2 b 2 a
The critical load is obtained for the minimum value of the right-hand term, which
for fixed m has a minimum value for n = 1, so that we must minimize the expression
π2 Eh2 2b
2
a2
σ= m + 2 + (3.12.32)
12(1 − υ2 )b2 a2 b2 m2
π2 Eh2
σcr ≥ (3.12.34)
3(1 − υ 2 )b2
where the equality sign holds when a/ b is an integer. For sufficiently large values of
a/b, (3.12.34) is a good approximation,† which means that for a sufficiently long plate
we can always approximate a/b by an integer. The critical load for other boundary
conditions is often expressed as
π2 Eh2
σcr = k, (3.12.35)
3(1 − υ2 )
i.e., it is expressed as a multiple of the critical load for a simply supported load.
When a/b < 1, the smallest wave number is m = 1, so we can write the critical
load in the form
π2 Eh2 a2 a4
σcr = 1+2 2 + 4 . (3.12.36)
12(1 − υ2 )a2 b b
π2 Eh2
σcr = , (3.12.37)
12(1 − υ2 )a2
which also follows from the result for the Euler bar when we take into account that
there is no anti-elastic bending, which accounts for the factor (1 − υ2 )−1 .
Let us now consider the solution of our buckling problem in more detail. We
have attempted a solution in the form (3.12.29), which means that we have silently
† E.g., a/b = 3.5 yields for the factor between the brackets in (3.12.32), 4.072 for m = 4 and 4.095
when m = 3, which means an error of 2.5% when (3.12.34) is used.
144 Applications
a b
mπx nπy
w,xx w,yy sinsin dx dy
0 0 a a
m2 n2 π4 a b mπx nπy
= 2 2
w sin sin dx dy.
ab 0 0 a b
Our final result for the Fourier transform of the differential equation can now be
written as
2 2 2
Eh3 mπ n2 π2 m2 π2
+ 2 − hσ 2 Wmn = 0 (3.12.43)
12(1 − υ2 ) a2 b a
3.12 Buckling of plates loaded in their plane 145
where
a b
mπx mπy
Wmn = w(x, y) sin sin dx dy (3.12.44)
0 0 a b
is the Fourier transform of w(x, y). The equation (3.12.43) is valid for all integer
values of m and n, and because Wmn ≡ 0 the expression between the brackets must
vanish, which leads to our earlier result (3.12.31).
Introducing the series (3.12.38) into (3.12.44), we find the following relation
between Cmn and the Fourier transform of w(x, y):
1
Wmn = Cmn ab. (3.12.45)
4
To obtain this result, we have interchanged integration and summation, which is
admissible because the series (3.12.38) is a uniformly convergent series. We see that
only one term of the series is left, which justifies our earlier approach.
Let us now consider the influence of a free edge at y = 0 for the case that a/b
1, when the other edges are simply supported (see Figure 3.12.3).
In this case, we shall only apply a Fourier transform in the x direction because w =
w = 0 at x = 0 and x = a, but w = 0 at y = 0. With
a
mπ
Wm(y) = w(x, y) sin dx, (3.12.46)
0 a
we find
Eh3 d4 Wm m2 π2 d2 Wm m4 π4 m2 π2
− 2 + Wm − hσ Wm = 0, (3.12.47)
12(1 − υ2 ) dy4 a2 dy 2 a4 a2
where we have used the relation
a n
d w mπx dn Wm
n
sin dx = . (3.12.48)
0 dy a dyn
As discussed previously for a short plate (a/b 1), the wave number is m = 1.
Introducing the notation
a
y = η, d () /dη = ()· , W1 = W, (3.12.49)
π
y
s⋅s
s⋅s s⋅s
σ σ
x
free edge
Figure 3.12.3
146 Applications
from which
√
2υ2
ε= 1 + O(ε2 ) , (3.12.66)
(2 − υ) 2
so that
4υ4
λcr ≈ 1 − . (3.12.67)
(2 − υ)4
The largest reduction is obtained for υ = 1/2, where ε ≈ 0.157 and
77
λcr = = 0.9506. (3.12.68)
81
f ( λ,υ )
υ=0
υ>0
1 λ
Figure 3.12.4
148 Applications
b/2
σ
x=a x
σ
−b/2
Figure 3.12.5
which means a reduction of the critical load of about 5% compared to the plate that
is simply supported at y = 0.
Let us now consider a rectangular plate with clamped edges at y = ±b/2, and
simply supported edges at x = 0 and x = a, loaded in compression by forces σh per
unit length at x = 0 and x = a (see Figure 3.12.5).
However, the clamped edges are allowed to slide in the x-direction. We con-
sider the case that the plate is sufficiently long, and that 2 is the wavelength of
the deformation pattern in the buckled state. The differential equation for neutral
equilibrium is
Eh3
w + σhw,xx = 0, (3.12.71)
12(1 − υ2 )
and the boundary conditions are
w = 0, w,xx = 0 at x = 0, x = m (≤ a)
(3.12.72)
w = 0, w,y = 0 at y = ± b/2.
Because the lowest bifurcation load will occur when m = 1, we try a solution of the
form
πx
w(x, y) = W(y) sin . (3.12.73)
Substitution into the differential equation yields
4
Eh3 d W 2π2 d2 W π4 π2
− + W − σh W = 0. (3.12.74)
12(1 − υ2 ) dy 4 2 dy 2 4 2
3.12 Buckling of plates loaded in their plane 149
To avoid the rather tedious calculations involved with the solution of the tran-
scendental equation (3.12.83), we now try an approximate solution to our problem.
Introducing a dimensionless coordinate η defined by
1 1
y = ηb, − ≤ η≤ , (3.12.86)
2 2
we assume the following symmetric polynomial for W,
Because u and v are zero at bifurcation, it follows from (3.12.7) that we can write
Eh3 12(1 − υ2 ) σ 2
P2 [u1 ] = (w)2 − 2 (1 − υ) w,xx w,yy − w2,xy − w dx dy.
24(1 − υ2 ) h2 E ,x
(3.12.88)
The second variation is semi-positive definite for the exact solution. For the exact
buckling load and an assumed (approximate) displacement, the field P2 [u] > 0, so
that the application of Rayleigh’s method (also called Rayleigh-Ritz method) where
we put P2 [u] = 0 for an assumed displacement field yields an upper bound for the
critical load.
For the evaluation of (3.12.88), it will be convenient to make use of the following
property:
w,xx w,yy − w2,xy dx dy = 0 (3.12.89)
S
for a domain bounded by straight lines (polygonal edge curve) with w = const. at
the boundary ∂S, and ∂w/∂υ = 0 in the corner points at the boundary where υ is the
unit normal to ∂S, positive in the outward direction. To show this property, we apply
†
√
This result is valid for /b ≥ 1/2 λcr = 0.37878.
3.12 Buckling of plates loaded in their plane 151
So far, our result is fully general. When w is constant along ∂S, we have w,s = 0
along ∂S, and along a straight line υx and υy are constant, so that for a polygonal
edge curve where w = const., we can write
si+1
1
n
2
(w,xx w,yy − w2,xy )dx dy =− υx υy w,υ (3.12.91)
2
S i=1 si
for a polygonal edge curve with n corner points, where si denotes the ith corner point
and sn+1 ≡ s0 . When w,υ = 0, in the corner points we obtain the result (3.12.89).
Notice that this result is valid for ∀w w ∈ C3 , w = 0 on ∂S, ∂w/∂υ = 0 in the corner
points on ∂S, where ∂S is a polygonal curve.
For simply supported and clamped straight edges, we have w = 0, ∂w/∂υ = 0 in
the corner points, and thus the result (3.12.89) applies,† so that for these plates we
can write
Eh3 12 1 − υ2
σ
P2 [u1 ] = (w)2 − w2 dx dy. (3.12.92)
24 (1 − υ2 ) h2 E ,x
† Notice that the limiting case that n → ∞ for a simply supported plate bounded by a regular poly-
gonal curve with n corner points is not a simply supported but a clamped circular plate, since now
∂w/∂υ = 0 along the edge.
152 Applications
where
π2 b2
C= , (3.12.94)
16 2
and
1 1
2 b 2
π2 πx 2
w2,x dx dy = b cos2 1 − 8η2 + 16η4 dx dη
2
0−1b 0 1
2 2
64 16C
= , (3.12.95)
315 b
we obtain from the condition P2 [u1 ] = 0,
4 64 128 2
128 + C + C
Eh2 5 105 315
σ= . (3.12.96)
12(1 − υ )b
2 2 64
· 16C
315
Minimizing with respect to C, we obtain
)
63
Cmin = = 1.403 (3.12.97)
32
and from (3.12.94),
y
τ
y=b
τ τ
x
τ x=a
Figure 3.12.6
3.12 Buckling of plates loaded in their plane 153
Figure 3.12.7
Notice that the differential equation now contains even and odd derivatives with
respect to x and y. In the following, we shall assume that the plate is infinitely long,
so that we do not take into account the boundary conditions on the vertical ends.
The boundary conditions at y = 0, y = b read
w = 0, w,yy = 0 at y = 0, y = b. (3.12.101)
The exact solution to this problem was given by Southwell,† and his result was
π2 Eh2
τcr = 5.35 . (3.12.102)
12(1 − υ2 )b2
However, the numerical factor is not very accurate. Later, we shall give an upper
bound that is slightly smaller.
The exact solution of this problem requires rather tedious calculations, and
therefore we shall try to construct an approximate solution. We try a solution of
the form
π πy
w(x, y) = f sin (x − my) sin , (3.12.103)
b
x
x=l x = 2l
Figure 3.12.8
† Cf. R.V. Southwell, Proc. Roy. Soc., London, Series A, 105, 582.
154 Applications
w=0
x
A
w=0
Figure 3.12.9
3.12 Buckling of plates loaded in their plane 155
Because w = 0 along the edge and along the nodal line, we must have w,x =
w,y = 0 in A but not w,yy = 0. To satisfy this condition, we try a form for w(x, y) that
is slightly more general than (3.12.102):
π
w(x, y) = sin (x − ϕ(y))W(y), W(0) = W(b) = 0. (3.12.111)
We then obtain
π π π
w,y = − W(y)ϕ (y) cos [x − ϕ(y)] + W (y) sin [x − ϕ(y)] (3.12.112)
π2 2
w,yy = W(y) − 2 W(y)ϕ (y) sin [x − ϕ(y)]
π (3.12.113)
− [2W (y)ϕ (y) + W(y)ϕ (y)] cos [x − ϕ(y)]
π π
w,yy (x, 0) = W (0) sin[x − ϕ(0)] − 2 W (0)ϕ (0) cos [x − ϕ(0)]. (3.12.114)
Because w,y (x, 0) is free, W (0) = 0, so that the conditions for w,yy (x, 0) = 0 are
which means that the nodal line x = ϕ(y) is perpendicular to the edge. These condi-
tions are satisfied by choosing
π πy
w(x, y) = f sin (x − ϕ(y)) sin , (3.12.116)
b
where
mb πy
ϕ(y) = 1 − cos . (3.12.117)
π b
Figure 3.12.10
156 Applications
∞ ∞
Eh2 π2 kπy kπy
(µ + k ) ak sin
2 2 2
− 2τµ kbk cos =0
12(1 − υ2 )b2 b b
k=1 k=1
∞ ∞
(3.12.123)
Eh2 π2 kπy kπy
(µ + k ) bk sin
2 2 2
+ 2τµ kak cos = 0.
12(1 − υ2 )b2 b b
k=1 k=1
π2 Eh2
τ=λ , (3.12.124)
3(1 − υ2 )b2
Assuming that the left-hand members can be written as a sine series, we multiply
both sides of these equations by sin j πy/b and integrate (term-wise) from y = 0 to
y = b. With
b
j πy kπy 1
sin sin dy = bδjk (3.12.126)
b b 2
0
3.12 Buckling of plates loaded in their plane 157
b
j πy kπy [1 − (−1)j +k]
sin cos dy = j b (= 0 for j = k) , (3.12.127)
b b (j 2 − k2 ) π
0
we obtain
∞
32 kj
(µ2 + j 2 )2 aj − λµ 1 − (−1)k+j bk = 0
π 2 ( j 2 − k2 )
k=1
∞
(3.12.128)
32 kj
(µ + j )bj + λµ
2 2
1 − (−1)k+j ak = 0.
π 2 (j − k )
2 2
k=1
There are two infinite sets of linear algebraic equations for the unknowns ak, bk
(k = 1, 2, . . .). The condition for a non-trivial solution is that the determinant of the
matrix of coefficients vanishes, i.e., with
32
λµ = ρ (3.12.129)
π
a1 b2 a3 b4 ···
2 2ρ 4ρ
(µ + 1)2 0 · · ·
3 15
2ρ 6ρ
(µ + 1)2
2
· · ·
0
3 5
6ρ 12ρ
0 − (µ2 + 9)2 · · · . (3.12.130)
5 7
4ρ
12ρ
(µ + 1)2
2
· · ·
0
15 7
.. .. .. ..
. . . .
4ρ2
(µ2 + 1)2 (µ2 + 4)2 − = 0, (3.12.131)
9
which yields
2ρ 64µλ
(µ2 + 1)(µ2 + 4) = = ,
3 3π
so that
3π 4
λ= µ3 + 5µ + . (3.12.132)
64 µ
Minimizing with respect to µ yields
5 4
µ4 + µ2 = = 0,
3 3
so that
√
−5 + 73
µ =
2
≈ 0.59 (3.12.133)
b
158 Applications
and
which shows that this process converges rapidly with the exact result.
For a square plate loaded by compressive forces σ per unit length in the x-direction,
the energy functional reads
!
Eh3
P [u] = (w)2 − 2(1 − υ) w,xx w,yy − w2,xy
12(1 − υ )2
Eh 1 2 2 1 2 2
+ u,x + w,x + v,y + w,y
2(1 − υ2 ) 2 2
(3.13.1)
1 2 1 2
+ 2υ u,x + w,x v,y + w,y
2 2
"
1 1
+ (1 − υ)(u,y + v,x + w,x w,y ) − σhw,x dx dy.
2 2
2 2
The second variation is given by
!
Eh3
P[u] = [(w)2 − 2(1 − υ) w,xx w,yy − w2,xy X]
12(1 − υ )
2
"(3.13.2)
Eh 1 1
+ u2
+ v2
+ 2υu,x v,y + (1 − υ)(u,y + v ,x ) 2
− σhw 2
dx dy,
2(1 − υ2 ) ,x ,y
2 2 ,x
and
Eh 2 2
P4 [u] = w,x + w2·y dx dy, (3.13.4)
8(1 − υ2 )
respectively.
The displacement field for small but finite deflections from the fundamental
state is written as
u = au1 + u, (3.13.5)
3.13 Post-buckling behavior of plates loaded in their plane 159
where u is orthogonal to the buckling mode. The equations for u are obtained by
minimizing the functional
P2 [u] + a(λ − 1)P11 [u1 , u] + a2 P21 [u1 , u], (3.13.6)
Because P21 [u1 , u] does not contain w, we must minimize P2 [u] with respect to w.
P2 [u] is positive-definite, and its minimum with respect to w is obtained when the
first line in (3.13.2), which is positive-definite, vanishes, which yields
w ≡ 0. (3.13.10)
b/2 !
1 1
− u,xx + (1 − υ) u,yy + (1 + υ) v,xy
2 2
0 −b/2
a2 2 " (3.13.13)
+ w1,x + w21,y ,x + (1 − υ) (w1,x w1,yy − w1,y w1,xy ) ξ
2
!
1 1
+ v,yy + (1 − υ) v,xx + (1 + υ) u,xy
2 2
"
a2 2
+ w1,x + w21,y ,y + (1 − υ) w1,y w1,xx − w1,x w1,xy η dx dy = 0,
2
from which we obtain
1 1
u,xx + (1 − υ) u,yy + (1 + υ)v,xy
2 2
a2 2
=− w1,x + w21,y ,x + (1 − υ) (w1,x w1,yy − w1,y w1,xy )
2 (3.13.14)
1 1
v,yy + (1 − υ)v,xx + (1 + υ)u,xy
2 2
a2 2
=− w1,x + w21,y ,y + (1 − υ) (w1,y w1,xx − w1,x w1,xy ) .
2
We shall now discuss various boundary conditions. First, notice that for the
determination of the buckling load we must only deal with w, whereas for the post-
buckling only in-plane quantities (u, v) must be dealt with. We shall now adapt
this theory to the problem of a plate with many fields, loaded in compression
by forces σh per unit length. The plate is simply supported at x = ± /2, and at
y = ±((2k + 1)/2)b (k = 0, 1, 2, . . .) in such a way that in-plane displacements are
free (see Figure 3.13.1). We shall now consider the field {x, y ||x| ≤ /2, |y| ≤ b/2}.
The edges y = ±b/2 remain straight (due to symmetry), so that
σ b σ
x
Figure 3.13.1
3.13 Post-buckling behavior of plates loaded in their plane 161
b σ x
Figure 3.13.2
Further, we shall assume that at x = ± /2 the supports are such that u = const.
For the rectangular plate simply supported at its edges, the buckling mode is
(see (3.12.29)
πx πy
w1 = cos cos ≡ cos λx cos µy. (3.13.16)
The equations for u and v now become
1 1
u,xx + (1 − υ)u,yy + (1 + υ)v,xy
2 2
a2
= − λ[(λ − υµ ) sin 2λx + (λ2 + µ2 ) sin 2λx cos 2µy],
2 2
4 (3.13.17)
1 1
v,yy + (1 − υ)v,xx + (1 + υ)u,xy
2 2
a2
= − µ[(µ − υλ ) sin 2µy + (λ2 + µ2 ) sin 2µy cos 2λx].
2 2
4
By inspection, we notice that particular solutions of (3.13.17) are of the form
upart = A sin 2λx + B sin 2λx cos 2µy
(3.13.18)
vpart = C sin 2µy + D sin 2µy cos 2λx.
from which
a2 λ2 − υµ2 a2 µ2 − υλ2
A= , C= ,
16 λ 16 µ
(3.13.20)
a2 a2
B= λ, D= µ.
16 16
162 Applications
a2
upart = [(λ2 − υµ2 ) sin 2λx + λ2 sin 2λx cos 2µy]
16λ (3.13.21)
a2
vpart = [(µ2 − υλ2 ) sin 2µy + µ2 cos 2λx sin 2µy].
16µ
To satisfy the boundary conditions, we must add solutions of the homogeneous
equations. With the condition (3.13.15), the boundary conditions become
.
u = const.
at x = ± /2
u,y + v,x = 0
. (3.13.22)
v = const.
at y = ± /2,
u,y + v,x = 0
which means that the shear strains vanish along the edge of the plate. Writing the full
solution as the sum of the solution of the homogeneous equations and the particular
solution
.
uh = const.
at x = ± /2
uh,y + vh,x = 0
. (3.13.25)
vh = const.
at y = ± /2.
uh,y + vh,x = 0
satisfy both the equations and the boundary conditions. The unknown strains −a2 ε1 ,
−a2 ε2 follow from the conditions
b/2
a2 2
u,x + υv,y + w1,x dy = 0 at x = ± /2
2
−b/2
(3.13.27)
/2
a2 2
v,y + υu,x + w1,y dx = 0 at y = ± b/2,
2
− /2
3.13 Post-buckling behavior of plates loaded in their plane 163
which yields
1 2
ε1 + υε2 = (λ + υµ2 )
8 (3.13.28)
1
υε1 + ε2 = (υλ2 + µ2 ),
8
from which we obtain the expressions for the strains ε1 , ε2 ,
1 2 1 2
ε1 = λ, ε2 = µ . (3.13.29)
8 8
The energy in the plate is given by (3.13.7), where u = au1 + u, so that with u1 =
v1 ≡ 0 and w ≡ 0, we obtain
!
Eh3 a2 2
P [au1 + u] = (w 1 ) − 2(1 − υ) w 1,xx w 1,yy − w 2
12(1 − υ2 ) 1,xy
2 2
Eh 1 1
+ u,x + a2 w21,x + v,y + a2 w21,y
2(1 − υ2 ) 2 2
(3.13.30)
a2 2 a2 2
+ 2υ u,x + w1,x v,y + w1,y
2 2
"
1 2 1
+ (1 − υ) u,x + v,y + a2 w1,x w1,y − σha2 w21,x dx dy.
2 2
Making use of the fact that
Eh3
(w1 )2 − 2(1 − υ) w1,xx w1,yy − w21,xy dx dy
12(1 − υ )
2
1
= σcr h w21,x dx dy
2
and the fact that u and v are the solutions of the minimum problem, we have
Eha2
u,x w21,x + v,y w21,y + υ u,x w21,y + v,y w21,x
12(1 − υ )
2
+ (1 − υ) (u,x + v,y ) w1,x w1,y dx dy (3.13.31)
Eh 1
=− u2.x + v2,y + 2υu,x v,y + (1 − υ) (u,x + v,y )2 dx dy.
1 − υ2 2
Then, we may write
/2 b/2
Eh 2 2
P [au1 + u] = w1,x + w21,y dx dy
8 (1 − υ2 )
− /2 −b/2
/2 b/2
Eh 1 2
− u2,x + v2,y + 2υu,x v,y + (1 − υ) (u,y + v,x ) dx dy
2(1 − υ2 ) 2
− /2 −b/2
/2 b/2
1
+ (σcr − σ)h w21,x dx dy = F (a). (3.13.32)
2
− /2 −b/2
164 Applications
σ
Et
asymptotic result
σ cr actual behavior
ε tot
Figure 3.13.3
3.13 Post-buckling behavior of plates loaded in their plane 165
σ σ b
l
Figure 3.13.4
However, it should be noted that our solution is only valid in a small neighbor-
hood of the critical load. Comparison of our result with exact numerical calculations
show that for sufficiently long plates, the range of validity is moderately large. How-
ever, for short plates the range of validity is extremely small. Let us consider this
case in some detail.
From (3.12.37), the critical load is then given by
π2 Eh2
σcr = , (3.13.41)
12(1 − υ2 ) 2
and for /b → 0, we have µ/λ → 0 so that from (3.13.35) we obtain
σ − σcr σ − σcr 2
a2 = 16 = 16 . (3.13.42)
λ2 E E h2
The total strain εtot is now given by
σ a2 h2 σ 2σcr
εtot = + = − , (3.13.43)
E 8 2 1 E
E
3
so the tangent modulus now is given by Et = 1/3E. This result has an extremely
small range of validity because almost immediately after buckling, the plate behaves
like a buckled bar.
This result is valid for the lowest bifurcation load, with only one wave.
asymptotic result
Et
σ cr
actual behavior
ε tot
Figure 3.13.5
166 Applications
From (3.12.36), the exact formula for the critical load for a plate is
Eh2 λ2 λ4
σcr = λ 2
1 + 2 + . (3.13.44)
12(1 − υ2 ) µ2 µ4
For fixed λ and, say, λ/µ = 10−2 for one wave, we have
Eh2
(1 + 2.10−4 ).
(1)
σcr ≈ λ2 (3.13.45)
12(1 − υ2 )
When there are two waves, we have λ/µ = 2.10−2 and
Eh2
(1 + 8.10−4 ),
(2)
σcr ≈ λ2 (3.13.46)
12(1 − υ2 )
which is very close to the lowest bifurcation load. This result shows that the lowest
bifurcation point is a cluster point.
Let us finally make some remarks regarding the results obtained. We have
assumed that the edges at x = ± remain straight. This assumption holds exactly
for sufficiently long plates because the nodal lines are straight lines. However, for
short plates this must be accomplished with an edge beam, without constraining the
strain long the edge. This condition can be done between two smooth stamps in a
testing machine.
The post-buckling behavior of plates can also be analyzed using the so-called “von
Kármán-Föppl equations.” For the derivation of these equations, we consider the
deviation from the undeformed state. The strains in the mid-plane are
1 1
γxx = u,x + w2,x , γyy = v,y + w2,y ,
2 2 (3.14.1)
2γxy = u,x + v,y + w,x w,y .
The elastic energy with respect to the undeformed state is now given by
Eh 1 2 2 1 2 2 1 2 1 2
u,x + w + v,y + w + 2υ u,x + w v ,y + w
2(1 − υ2 ) 2 ,x 2 ,y 2 ,x 2 ,y
1
+ (1 − υ) (u,y + v,x + w,x w,y )2 (3.14.4)
2 .
h2 2
+ w,xx + w,yy + 2υw,xx w,yy + 2(1 − υ)w,xy dx dy.
2 2
12
3.14 The “von Kármán-Föppl Equations” 167
This expression is in full agreement with (3.12.4) because the fundamental state is
linear. Carrying out the variations with respect to u and v, we obtain a stationary
value when
!
Eh 1 2 1 2
u,x + w,x + υ v,y + w,y δu,x
1 − υ2 2 2
Eh 1 2 1 2
+ υ u,x + w,x + v,y + w,y δv,y (3.14.5)
1 − υ2 2 2
"
Eh
+ (u,y + v,x + w,x w,y ) (δu,y + δv,x ) dx dy = 0.
2(1 + υ)
and the in-plane boundary conditions, when either u, v are zero or free,
(Nx υx + Nxy υy ) δu∂S = 0
(3.14.10)
(Ny υy + Nxy υx ) δv∂S = 0.
∗ ∗
Nx,x = F,xyy , and from the second Ny,y = F,xyx . Integrating these expressions with
respect to x and y, respectively, we obtain
∗ ∗
Nx = F,yy + f (y), Ny = F,xx + g(x),
or defining
∗
f (y) = f ,yy , g(x) = g∗,xx ,
∗ ∗ ∗
Nx = F,yy + f ,yy , Ny = F,xx + g∗,xx .
Eh3
w − Nx w,xx − Ny w,yy − 2Nxy w,xy = 0. (3.14.13)
12(1 − υ2 )
Using (3.14.11), we can rewrite (3.14.13) in the form
Eh3
w − F,yy w,xx + 2F,xy w,xy − F,xx w,yy = 0. (3.14.14)
12(1 − υ2 )
The equation for the stress function is obtained from the condition that the displace-
ment field is compatible. To this end, we write the inverse of (3.14.6), making use of
(3.14.11), which yields the following expressions:
1 1
u,x + w2,y = (F,yy − υF,xx )
2 Eh
1 1
v,y + w2,y = (F,xx − υF,yy ) (3.14.15)
2 Eh
2(1 − υ)
u,y + v,x + w,x w,y = − F,xy .
Eh
3.15 Buckling and post-buckling behavior of shells using shallow shell theory 169
1
F = w2,xy − w,xx w,yy . (3.14.16)
Eh
This equation and (3.14.14), without the bending term, were first derived by Föppl.
Later, von Kármán derived the full equations.
Notice that in this approach there are only two unknowns, Airy’s stress function
and the displacement w, whereas in the general theory we use the three displace-
ments as unknowns.
3.15 Buckling and post-buckling behavior of shells using shallow shell theory
In cases where the wavelength of the buckling mode is small compared to the small-
est radius of curvature, shallow shell theory may be used. Shallow shell theory can
be derived as follows.
Consider a surface Z(x, y), where x and y are Cartesian coordinates in a tangent
plane to the surface (see Figure 3.15.1), chosen so that the projection of the lines
with principal radius of curvature on the tangent plane coincide with the x and y axis,
respectively. The z axis is perpendicular to the tangent plane (see Figure 3.15.1). In
the undeformed state, the square of a line element on the surface Z(x, y) is given by
For the description of the surface in the deformed state, we shall make use of the fact
that the in-plane displacements u, v are considerably smaller than the displacement
z y
w v
u
x
Figure 3.15.1
170 Applications
and that
(x + u − Z,x w, y + v − Z,y w, z + w)
in the deformed state. The square of the length of a line element in the deformed
state is now
Neglecting higher order terms, except for w2,x , w2,y and w,x , w,y , we obtain
1
(ds̄)2 − (ds)2 = 2 u,x − Z,xx w + w2,x (dx)2
2
+ 2(u,y + v,x + w,x w,y ) dx dy
(3.15.6)
1
+ 2 u,y − Z,yy w + w2,y (dy)2
2
≡ 2γxx (dx)2 + 4γxy dx dy + 2γyy (dx)2 .
where
are the principle radii of curvature. The changes of curvature are given by
Notice that all our expressions are only valid in a sufficiently small neighbor-
hood of the tangent point. Under the assumption that the loads are dead-weight
loads, which cause stresses σx , σy , and τxy in the fundamental state, the energy is
given by
!
Eh w 1 2 2 w 1 2 2
P[u] = u,x − + w,x + v,y − + w,y
2(1 − υ 2 ) R1 2 R2 2
w 1 w 1 1
+ 2υ u,x − + w2,x v,y − + w2,y + (1−υ)(u,y +v,x +w,x w,y )2
R1 2 R2 2 2
(3.15.10)
h2 2
+ w + w,yy + 2υw,xx w,yy + 2(1 − υ)w,xy
2 2
12 ,xx
"
1 − υ2
+ σx w,x + σy w,y + 2τxy w,x w,y dx dy,
2 2
E
where the radii of the curvature can be assumed to be constants. Later we shall show
that this expression may also be used for a shallow shell under uniform pressure
(which is not a dead-weight load).
The second and the third variations are given by
!
Eh w 2 w 2
P2 [u] = u,x − + v,y −
2(1 − υ 2 ) R1 R2
w w 1
+ 2υ u,x − v,y − + (1 − υ)(u,y + v,x )2
R1 R2 2
(3.15.11)
2
h
+ w + w,yy + 2υw,xx w,yy + 2(1 − υ)w,xy
2 2 2
12 ,xx
"
1 − υ2
+ σx w,x + σy w,y + 2τxy w,x w,y dx dy
2 2
E
!
Eh w w
P3 [u] = u,x − + υ v ,y − w2,x
2(1 − υ 2 ) R1 R2
" (3.15.12)
w w
+ v,y − + υ u,x − w2,y + (1 − υ)(u,y + v,x )w,x w,y dx dy.
R2 R1
The variational equation for neutral equilibrium is P11 [u, ζ] = 0, where ζ = (ξ, η, ζ) is
a kinematically admissible displacement field,
Eh w ζ w ζ
P11 [u, ζ] = u,x − ξ,x − + v,y − η,y −
1 − υ2 R1 R1 R2 R2
w ζ w ζ
+ υ u,x − η,y − + υ v,y − ξ,x −
R1 R2 R2 R1
172 Applications
1
+ (1 − υ)(u,y + v,x )(ξ,y + η,x ) (3.15.13)
2
h2
+ w,xx ζ,xx + w,yy ζ,yy + υw,xx ζ,yy + υw,yy ζ,xx + 2(1 − υ)w,xy ζ,xy
12
1 − υ2
+ (σx w,x ζ,x + σy w,y ζ,y + τxy w,x ζ,y + τxy w,y ζ,x ) dx dy.
E
where υ is the unit normal vector at the edge. This yields the differential equations
1 1 1 υ
u,xx + (1 − υ)u,yy + (1 + υ)v,xy − + w,x = 0
2 2 R1 R2
(3.15.15)
1 1 1 υ
v,yy + (1 − υ)v,xx + (1 + υ)u,xy − + w,y = 0
2 2 R2 R1
% &
h2 1 υ 1 υ 1 2υ 1
w − + u,x − + v,y + 2
+ + 2 w
12 R1 R2 R2 R1 R1 R1 R2 R2
(3.15.16)
1 − υ2
− (σx w,xx + σy w,yy + 2τxy w,xy ) = 0,
E
where we have used the fact that σx , σy , and τxy (approximately) satisfy the equa-
tions
For the reduction of the remaining terms, we make use of the relations
ζ,x = −υy ζ,s + υx ζ,v , ζ,y = υx ζ,s + υy ζ,v , (3.15.19)
and (3.12.17) to obtain
(w,xx + υw,yy )υ2x + 2(1 − υ)w,xy υx υy + (w,yy + υw,xx )υ2y ζ,v ∂S = 0 (3.15.20)
!
(1 − υ) (w,yy − w,xx )υx υy + w,xy υ2x − υ2y
,s
Notice that only the in-plane boundary conditions are affected by the curvature of
the surface.
We shall now apply this theory to a spherical shell under a uniform external
pressure p per unit area of the middle surface.† The stresses in the fundamental
state are given by
pR
σx = σy = −σ = − , τxy = 0. (3.15.22)
2h
The equations now read
1 1 1+υ
u,xx + (1 − υ)u,yy + (1 + υ)v,xy − w,x = 0
2 2 R (3.15.23)
1 1 1+υ
v,yy + (1 − υ)v,xx + (1 + υ)u,xy − w,y = 0
2 2 R
Eh2 E w
w − u ,x + v,y − 2 + σ(w,xx + w,yy ) = 0. (3.15.24)
12(1 − υ 2 ) (1 − υ)R R
We try a solution of the form
u = A sin px/R cos qy/R
v = B cos px/R sin qy/R (3.15.25)
w = C cos px/R cos qy/R, (p, q ∈ R).
Substitution into (3.15.23) yields
1 1
− p + (1 − υ)q A − (1 + υ)pqB + (1 + υ)pC = 0
2 2
2 2
(3.15.26)
1 1
− (1 + υ)pqA − (1 − υ)p + q B + (1 + υ)qC = 0.
2 2
2 2
We can now solve A/C and B/C from these equations, which yield
A p B q
= (1 + υ) 2 , = (1 + υ) 2 . (3.15.27)
C p + q2 C p + q2
In our derivation of the shallow shell equations, we have use the assumption |w|
|u|, |v|, which means A/C, B/C 1, so that for p or q 1, our assumptions are
satisfied.
Substitution of the displacement field (3.15.25) into the third equilibrium equa-
tion and using (3.15.27), we obtain
σ 1 h2
= 2 + (p 2
+ q2
), c = 3(1 − υ 2 ). (3.15.28)
E p + q2 4c2 R2
We still must minimize this expression with respect to p and q, but because the
expression only contains the combination p 2 + q2 , we can minimize with respect to
this parameter, which yields
2cR
p 2 + q2 = . (3.15.29)
h
This means that our assumptions are satisfied as R/h 1. The critical load is now
given by
Eh
σcr = . (3.15.30)
cR
Notice that the critical load is proportional to h/R, whereas for plates the critical
load was proportional to h2 / b2 .
Because p and q must only satisfy the condition (3.15.29), there is an infinite
number of buckling
modes. In the p − q plane, (3.15.29) represents a circle with
radius p 0 = 2cR/h (see Figure 3.15.2).
Because we have a continuous spectrum, the displacement field is given by inte-
grals instead of a series. However, for our purpose we may consider some discrete
values, p i , qi . Because for the evaluation of P3 [u1 ] we must deal with cubic terms,
( 2c p
R
/h
) 1/ 2
Figure 3.15.2
3.15 Buckling and post-buckling behavior of shells using shallow shell theory 175
(p 1 , q1 ), (p 2 , q2 ), (p 3 , q3 ), (3.15.31)
p 1 ± p 2 ± p 3 = 0, (3.15.34)
q1 ± q2 ± q3 = 0. (3.15.35)
This means that the points (p 1 , q1 ), (p 2 , q2 ), (p 3 , q3 ) divide the circle in three equal
parts. We shall now consider the special case of (see Figure 3.15.2)
1 1 √
(p 1 , q1 ) = (p 0 , 0), (p 2 , q2 ) = − p 0 , p 0 3 ,
2 2
(3.15.36)
1 1 √
(p 3 , q3 ) = − p 0 , − p 0 3 .
2 2
Let us introduce the notation
1 1√
p0, q =
p= 3 p0; (3.15.37)
2 2
then our displacement field is given by
h2 x x y
u = (1 + υ) 2pa0 sin 2p + pa1 sin p cos q
2cR R R R
h2 x y
v = (1 + υ) qa1 cos p sin q (3.15.38)
2cR R R
x x y
w = h a0 cos 2p + a1 cos p cos q ,
R R R
where we have used (3.15.27).
For the evaluation of P3 [u1 ], we shall need the following expressions,
w h x 1 x y
u,x − = υa0 cos 2p − (3 − υ)a1 cos p cos q
R R R 4 R R
w h x 1 x y
v,y − = −a0 cos 2p − (1 − 3υ)a1 cos p cos q
R R R 4 R R
†
Let L x R, e.g., x = O R3/4 h1/4 , so that our shallow shell assumptions still hold approxi-
mately.
176 Applications
w w h 3 x y
u,x − + υ v,y − = − a1 (1 − υ 2 ) cos p cos q
R R R 4 R R
w w h x 1 x y
v,y − + υ u,x − = −(1 − υ )a0 cos 2p − (1 − υ )a1 cos p cos q
2 2
R R R R 4 R R
h 1√ x y
u,y + v,x = (1 + υ) − 3 a1 sin p sin q (3.15.39)
R 2 R R
ph x 1 x y
w,x = 2 −a0 sin 2p − a1 sin p cos q
R R 2 R R
ph 1√ x y
w,y = 2 − 3 a1 cos p sin q .
R 2 R R
R R
1
( )= ( )dx dy, (3.15.40)
4R2
−R −R
9 Ech3
P3 [u1 ] = − a0 a21 . (3.15.42)
32 R2
so that
Eh3 1 2 1 2
P2 [u1 ; λ] = (1 − λ) a0 + a1 . (3.15.44)
R2 2 4
In a sufficiently small neighborhood of the critical load, the average energy P[u, λ]
is given by
Eh3 1 1 2 9c
F (ai , λ) = P[u, λ] = 2 (1 − λ) a0 + a1 − a0 a1 .
2 2
(3.15.45)
R 2 2 32
3.15 Buckling and post-buckling behavior of shells using shallow shell theory 177
The work of the uniform pressure p is −pV, where V is the volume increment
passing from the fundamental path to the branched path. It follows that
∂F
4πR2 = −V, (3.15.52)
∂p
or with
Eh2
p = 2λ (3.15.53)
cR2
2πR4 c ∂F
= −V. (3.15.54)
Eh2 ∂λ
Using (3.15.45), we obtain
64π 2
V = − hR (1 − λ)2 . (3.15.55)
27c
178 Applications
p
λ=
pcr
1
∆vtot
∆vcr
Figure 3.15.3
σcr 4π
Vcr = −4πR2 wcr = − (1 − υ)R · 4πR = − (1 − υ)hR2 , (3.15.56)
E c
so that
Vtot 16
= (1 − λ)2 + λ. (3.15.57)
Vcr 27(1 − υ)
This relation is illustrated in Figure 3.15.3. Notice that there is no stable equilibrium
state in the vicinity of the bifurcation point, and that the load drops very rapidly after
bifurcation, which means that the shell is very imperfection-sensitive. We will come
back to this later.
Let us now first have a closer look at the buckling mode,
x x √ y
w = ha0 cos 2p + 2 cos p cos p 3 . (3.15.58)
R R R
Introducing the notation
x y
2p = α, 2p = β, (2.15.59)
R R
we can rewrite this expression as
√
w = ha0 f (α, β), f (α, β) = cos α + 2 cos α/2 cos 3β/2. (3.15.60)
2j π
α = 2kπ, β= √ . (3.15.67)
3
We now obtain the following values for f (α, β),
2j + 1
f (2k + 1)π, √ π =1
3
1
+ j even
2π 2j π 2
f ± + 4kπ, √ =
3
3 −3 j odd
2 (3.15.68)
4π 2j π 3
f ± + 4kπ, √ =−
3 3 2
2j π −1 j + k odd
f 2kπ, √ =
3 3 j + k even.
3/2 3/2 3
3
3/2 3 3/ 2
−3π −2π −π 0 π 2π 3π α
3 3
3/2 3/2
Figure 3.15.4
smaller than that of the shell inside the shell.† The experimental results are in agree-
ment with our buckling pattern.
As we have seen previously, the load drops rapidly after bifurcation, which
implies that the shell is very imperfection-sensitive. To get quantitative informa-
tion about the imperfection sensitivity, we assume imperfections corresponding to
the buckling mode, which according to the general theory are the worst type of
imperfections.
Let the imperfections be given by
x x √ y
w0 = h ā0 cos 2p + ā1 cos p cos 3 p , (3.15.69)
R R R
where ā1 = 2ā0 . In the presence of imperfections, the function F (ai ; λ) (see 3.15.45)
must be replaced by (see 2.6.39)
Eh3 1 1 9c 1
F ∗ (ai ; λ) = 2 (1 − λ) a20 + a21 − a0 a21 − λ a0 ā0 + a0 ā1 . (3.15.70)
R 2 2 32 2
With ā1 = 2ā0 and a0 and a1 given by (3.15.47) and (3.15.48), we can rewrite this
expression in the form
Eh3 9c 3
F ∗ (ai ; λ) = 3(1 − λ)a2
0 − a − 6λ ā0 a0 . (3.15.71)
2R2 4 1
3.16 Buckling behavior of a spherical shell under uniform external pressure using
the general theory of shells
In this section, we shall again consider the behavior of a spherical shell under uni-
form external pressure, but now we shall employ the general theory of shells. For
the derivation of the equations of the theory of shells, we refer to the literature.†
However, we shall give a short survey of the most important quantities.
Let r(xα ) (α ∈ 1, 2) be the position vector from a fixed origin in space to a
generic point on the middle surface of the undeformed shell. The tangential base
β
vectors are aα = r,α = ∂r/∂xα . The reciprocal base is defined by aα · aβ = δα . The
covariant and contravariant metric tensors are given by aαβ = aα · aβ and aαβ =
aα · aβ , respectively. The unit normal to the middle surface is n = 12 εαβ aα × aβ where
εαβ is the contravariant alternating tensor. The second fundamental tensor is speci-
fied by bαβ = n · r,αβ .
A point in shell space is identified by its distance z to the middle surface and by
the surface coordinates of its projection on to the middle surface. The coordinate z is
orthonormal to the middle surface. The shell faces z = ± 12 h, where h is the constant
shell thickness, are surfaces parallel to the mid-surface.
The covariant components of the spatial metric tensor gij are specified by
where cαβ = bκα bκβ is the third fundamental tensor. On the mid-surface, we have
gαβ = aαβ . Let g be the determinant of gij and a be the determinant of ααβ ; then
)
g
= 1 − 2z H + z2 K, (3.16.2)
a
where H = 12 bαα is the mean curvature in a point (x1 , x2 ) of the middle surface and
K = b11 b22 − b12 b21 is the Gaussian curvature.
The edge of the shell is assumed to be a ruled surface formed by normals to the
middle surface along an edge curve on this surface. Let υ be the unit vector in the
tangent plane, normal to the edge curve and positive outward. The positive sense
on the edge curve is defined by the tangential unit vector t = n × υ.
A deformation of the middle surface is described by the two-dimensional dis-
placement field,
† Cf. W. T. Koiter and J. G. Simmonds, Foundations of shell theory. Proc. 13th IUTAM Congress,
(Springer Verlag, 1972), 150–176 (also WTHD Rep. No. 40).
3.16 Buckling behavior of a spherical shell under uniform external pressure 183
1
ωαβ = (uβ|α − uα|β ) (3.16.6)
2
is a rotation in the tangent plane to the shell, and
where the integration is to be carried out over the mid-surface of the shell. Here
Eαβλµ is the tensor of elastic moduli corresponding to a plate state of stress
2υ αβ λµ
E αβλµ
=G a a +a a +
αλ βµ αµ βλ
a a , (3.16.14)
1−υ
184 Applications
where we have used the divergence theorem on a closed surface so that there are
no stock terms. The reduction of the term quadratic in u is considerably more
difficult.
1 1
εijk εpqr δip uj ,quk,r dV0 = (δj qδkr − δj r δkq) uj ,quk,r dV0
2 2
V0 V0
1
= (uj ,j uk,k − uj ,kuk,j ) dV0
2
V0
1 1
= uj ,j uknk dA − uj ,jkuk dV0 (3.16.22)
2 2
A V0
1 1
− uj ,kuknj dA + uj ,kj uk d0
2 2
A 0
1
= (uj ,j nk − uj ,knj ) uk dA.
2
A
n1 = n2 = 0, n3 = −1, (3.16.24)
Now
1
33α = g3kk3α = g33 33α = (g33,α + g3α,3 − g3α,3 ) = 0. (3.16.30)
2
Here, ijk are Christoffel symbols of the first kind.
Using the relation
uα α = uα,α + αβ α uβ , (3.16.31)
Because p = O(N/R), where N denotes the maximum of the membrane forces and
R the minimum radius of curvature, the second term is of order O (Nwθαα /R) =
O(Nw2 /R2 ), whereas the first term is of order O(Nw/R). The potential energy of
the membrane stresses in the fundamental state is O(Nw2 /L2 ), so that for L/R 1
the contribution to the energy of the second term in (3.16.33) may be neglected.
The third term, which is even smaller, may then certainly be neglected. It follows
that for L/R 1 (shallow shell theory), only the first (linear) term must be taken
into account, which means that to a first approximation, uniform external pressure
may be considered as a dead-weight load. This implies that the energy functional is
given by (3.16.17).
We shall now apply these results to a spherical shell under uniform external
pressure. However, we shall also take into account the second term in (3.16.33) and
show that this term may indeed be neglected. The basic property of a spherical shell
is that the first and second fundamental tensors are proportional to each other,
1 1 α 1 αβ
bαβ = aαβ , bαβ = a , bαβ = a , (3.16.34)
R R β R
where R is the radius of the middle surface. The sign convention in (3.16.34) implies
that the positive direction of n points inward. The linearized strain tensor (3.16.5)
3.16 Buckling behavior of a spherical shell under uniform external pressure 187
where the term εαλ ψ|λα vanishes because ψ|λα is symmetric, and is the Laplacian.
This is for a known displacement field for an equation with ϕ, which has a unique
solution‡ for sufficiently smooth displacement fields. To get an equation for ψ, we
multiply (3.16.38) by εαµ to obtain
uα εαµ = εαµ ϕ,α + εαµ εαλ ψλ = εαµ ϕ,α + ψ|µ ,
so that
is symmetric in µ and υ, so that the left-hand term must also be symmetric, which
means
This is exactly the equation for ϕ, and this completes our proof.
As we shall show, the advantage of expressing uα in terms of ϕ and ψ is that we
obtain a separate equation for ψ in which ϕ and w do not appear. Using (3.16.38),
we can rewrite (3.16.35) to (3.16.37) to yield
1 1 1
θαβ = ϕ αβ + εαλ ψ.λβ + εβλ ψ.λα − aαβ w (3.16.43)
2 2 R
† Cf. A. van der Neut, De elastische stabiliteit vaan den dunwandigen bol (Thesis, Delft; H. J. Paris,
Amsterdam, 1932).
‡ Apart from an arbitrary constant, which is unimportant as only derivatives are needed.
188 Applications
1 1
ϕα = w,α + ϕ,α + εαλ ψλ (3.16.44)
R R
1 1 1
ραβ = w αβ + ϕ αβ + εαλ ψ.λβ + ε·βλ ψ.λα (3.16.45)
R 2R 2R
We now return to our energy functional (3.16.17). The fundamental state of a
spherical shell under uniform external pressure p is a membrane state of stress, and
the contravariant tensor of stress resultants is
1
Nαβ = σhaαβ = pRaαβ . (3.16.46)
2
It will be convenient to introduce a dimensionless load parameter λ, defined by
Eh
σ= λ, c = 3(1 − υ 2 ). (3.16.47)
cR
Also taking into account the second term of (3.16.33), the second variation of our
energy functional can now be written as
!
Eh h2
P2 [u; λ] = (1 − υ)θ θαβ + υ(θ κ ) +
αβ κ 2
(1 − υ)ραβ ρ + υ(θ κ )
αβ κ 2
2(1 − υ 2 ) 12
A
"
Eh2 Eh2 α
− λ (θαβ − ωαβ )(θ − ω ) + ϕα ϕ +
αβ αβ α
λ wθα − u ϕα dA.
α
2cR cR2
(3.16.48)
The first of these integrals vanishes after the application of the divergence theorem
on the closed surface. Changing the order of covariant differentiation in the remain-
ing integral, we can write
− ϕ αβλ εαλ ψβ dA = − ϕ αλβ εαλ ψβ dA − Rκ·αβλ ϕ,κ εαλ ψβ dA.
A A A
3.16 Buckling behavior of a spherical shell under uniform external pressure 189
Here the first integral on the right-hand side vanishes because ϕ αλβ is symmetric in
α and λ ; Rκ·αβλ is the Riemann-Christoffel tensor of the surface, which is related to
the Gaussian curvature by
The first of these integrals vanishes after the application of the divergence theorem
on a closed surface, and the second integral vanishes because ϕ |κα is symmetric in κ
and α.
We may now rewrite (3.16.49) in the form
1 1
θαβ θαβ dA = ϕ αβ − aαβ w ϕ αβ − aαβ w
R R
A A
1 λ λ α κβ
+
εαλ ψ ·β + εβλ ψ ·α ε·κ ψ + ε·κ ψ| β κα
dA
4
(3.16.52)
αβ 2 2 2
=
ϕ αβ ϕ − wϕ + 2 w dA
R R
A
1 λ ·β
+ ψ ·β ψλ + εαλ εβ·κ ψλ·β ψ|κα dA.
2
A
Applying the divergence theorem on the first term, this term vanishes for a closed
surface. Changing the order of differentiation, in the second term we obtain
1 κα β
ϕ αβ ϕ αβ dA = ϕ αβ ·α ϕ β dA − ε ε·α ϕ,κ ϕ,β dA
R2
A A
A
β 1
= − (ϕ) ϕ β dA − 2 ϕ |κ ϕ,κ dA
R
A A
β 1
=−
(ϕ) ϕ β dA + ϕϕ dA − 2 (ϕ |κ ϕ )κ dA
R
1
+ 2 ϕϕ dA.
R
190 Applications
Applying the divergence theorem to the first and the third terms, these terms vanish
because we are dealing with a closed surface. Our final result can now be written as
1
ϕ αβ ϕ αβ dA = ϕϕ + 2 ϕϕ dA, (3.16.53)
R
A A
Similarly,
·β λ β
ψλ·β ψλ dA = ψ·β ψ λ dA − ψλ·βλ ψβ dA
A A A
1 κλ β
=− ψλ·λβ ψβ dA −
ε εβλ ψ,κ ψ dA
R2
A A
β 1
= − (ψ)β ψ dA − 2 ψλ ψλ dA
R
A
A (3.16.55)
β
=− (ψ) ψ β dA + (ψ)2 dA
A A
1 λ 1
− 2 ψ ψ λ dA + 2 ψψdA
R R
A A
2 1 1
= (ψ) + 2 ψψ dA ≡ ψψ + 2 ψψ dA.
R R
A A
Finally,
εαλ εβ·κ ψλ·β ψ| κα
dA = εαλ εβκ ψλ·β ψακ dA
A A
α
= εαλ εβκ ψλ·β ψ κ dA − εαλ εβκ ψλβκ ψ|α dA
A A
1
= − εαλ εβκ ψλ·βκ − ψλ·κβ ψ|α dA.
2
Interchanging the order of covariant differentiation in the second term, we obtain
1
εαλ εβ·κ ψλ·β ψ|κα dA = εαλ εβκ εµλ εκβ ψ,µ ψ|α dA
2R2
A A
1 µ α
1
= − 2 2δα ψ ψ,µ dA = − 2 ψ|µ ψ|µ dA (3.16.56)
2R R
A A
1 1 1
= − 2 (ψ|µ ψ)µ dA + 2 ψψdA = 2 ψψdA.
R R R
A A A
3.16 Buckling behavior of a spherical shell under uniform external pressure 191
Applying the divergence theorem to the first integral, this term vanishes on a closed
surface. Changing the order of covariant differentiation in the second term, we
obtain
·β 1
w αβ εα·κ ψκβ dA = − εακ ψβ·κ w,α dA − 2 εακ ελβ εκβ ψ,λ w,α dA
R
A A 1
=− ακ
ε ψw,α κ dA + εακ ψw |ακ dA − 2 εακ ψ,κ w,α dA.
R
A A A
192 Applications
Applying the divergence theorem to the first of these integrals, this term vanishes
because we are dealing with a closed surface. The second integral vanishes because
w |ακ is symmetric in α and κ. Hence, we are left only with the third term, which can
be rewritten as
1 1
− 2 (εακ ψw,α )κ dA + 2 εακ ψw |ακ dA = 0,
R R
A A
where the first term vanishes on a closed surface and the second term vanishes due
to symmetry in α and κ of w |ακ . Our result is thus
w αβ εα·κ ψ κβ dA = 0. (3.16.61)
A
where the terms involving ψ have been evaluated in (3.16.55) and (3.16.56). The
term containing only ϕ is given in (3.16.53) or (3.16.54), which are obviously also
valid when ϕ is replaced by w.
Completely analogously, we obtain
w ϕ
w αβ w αβ dA = ϕ w + 2 dA = w ϕ + 2 dA, (3.16.63)
R R
A A A
Next, we must determine the contribution P2 [u, λ] of the stresses in the funda-
mental state. From (3.16.48), consider the integral
(θαβ − ωαβ ) θαβ − ωαβ + ϕα ϕα dA. (3.16.67)
A
where we have made use of (3.16.54). For the second term in (3.16.71), we obtain
2 1
w |α w |α + w |α ϕ |α + 2 ϕ |α ϕ |α dA
R R
A
2 2 1 1
= (w |α w)α − ww + (ϕw |α )α − wϕ + 2 (ϕ |α ϕ)α − 2 ϕϕ dA
R R R R
A
194 Applications
2 1
=− ww + wϕ + 2 ϕϕ dA.
R R
A
= (ψ)2 dA,
A
where we have made use of (3.16.55). The contribution of the stresses in the funda-
mental state to the energy functional is thus given by
(s) Eh2 1 2w 2
P2 [u; λ] = λ (ϕ)2 + 2 ϕϕ − ϕ + 2 w2
2cR R R R
A
2 1
− ww − wϕ − 2 ϕϕ + (ψ)2 dA (3.16.72)
R R
Eh2 2 4w 2
=− λ (ϕ)2 − ϕ − ww + 2 w2 + (ψ)2 dA.
2cR R R
A
To be in the elastic range, we must have σ/E <<< 1 σ/E = O 10−3 for most
engineering materials, which means that the underlined terms in (3.16.74) may be
neglected, implying a relative error of O (σ/E) compared to unity. Further, we may
neglect the term with ψ in the third line compared to that in the second line, which
introduces a relative error of O h2 /R2 , which is smaller than the error in shell the-
ory. Similarly, we may neglect the term with (ϕ)2 in the third line compared to
the corresponding term in the first line, again with a relative error of O h2 /R2 . To
show that the term 2R−1 h2 wϕ in the third line may also be neglected, we employ
the inequality
/ 0
2h h
hwϕ dA ≤ (hw)2 + (ϕ)2 dA, (3.16.75)
R R
A A
which shows that this term may be neglected in comparison with the sum of the
terms h2 (w)2 in the third line and (ϕ)2 in the first line, introducing a relative
error of O (h/R), which is consistent with the error inherent to the shell equations.
However, we shall accept this error. To discuss the term h2 ww/R2 in the fourth
line, we consider the inequality
! "
h2 h w2 2
2 2
ww dA ≤ + h 2
(w) dA, (3.16.76)
R R R2
A A
which shows that this term may be neglected compared to the sum of the terms with
w2 /R2 in the second line and h2 (w)2 in the third line, again with a relative error of
O (h/R). The terms h2 /R2 wϕ/R and h2 /R2 ϕϕ/R2 in the fourth line may be
neglected compared to the terms wϕ/R and ϕϕ/R2 in the first line with a relative
error of O h2 /R2 . Using the relation
ϕw dA = wϕ dA, ∀ϕ, w|w, ϕ ∈ C2 , (3.16.77)
A A
it follows that the term h2 ϕw/R3 in the fourth line may also be neglected with a
relative error of O h2 /R2 . This means that the second variation (3.16.74) can be
simplified to yield
Eh 1−υ 2 (1 + υ)
P2 [u; λ] = (ϕ)2 +
ϕϕ − wϕ
2 (1 − υ 2 ) R2 R
A
2 (1 + υ) 2 1 2
+ w + (1 − υ) ψ ψ + ψ
R2 2 R2
h2 σ h
+ (w)2 + 1 − υ 2 ww dA 1 + O . (3.16.78)
12 E R
θ
R
y
β
x
Figure 3.16.1
3.16 Buckling behavior of a spherical shell under uniform external pressure 197
This expression vanishes for n = 0 and n = 1 and is positive for n > 1. Because n = 1
implies a rigid body motion, we are only interested in values of n > 1. This means
that the terms with ψ are positive-definite, so a minimum of P2 [u; λ] can only be
obtained when the terms with ψ vanish, i.e.,
2
ψ = 0 or ψ + ψ = 0. (3.16.85)
R2
Later, we shall see that only the first of these equations holds. From (3.16.6), the
rotation in the tangent plane is
1 αβ 1 1 1
= ε ωαβ = εαβ ϕ βα + εβλ ψλα − ϕ αβ − εαλ ψλβ = − ψλλ = − ψ,
2 2 2 2
(3.16.86)
so that the first of the equations of (3.16.86) means that the rotation around the
normal vanishes identically in all buckling modes.
We may now restrict our discussion to the functional
Eh 1−υ 2 (1 + υ)
P2 [u, ω; λ] = (ϕ)2 + ϕϕ − wϕ
2 (1 − υ 2 ) R2 R
A
2 (1 + υ) 2 h2 2
2 σ
+ w + (w) + 1 − υ ww dA. (3.16.87)
R2 12 E
A necessary condition for a minimum value of this functional is that its variation
with respect to ϕ and w vanishes. For the reduction of the various terms, we need
the following results:
α
ϕδϕ dA = ϕδϕ α dA = − ϕ |α δϕ |α dA
A A A
=− ϕ |α δϕ |α dA = ϕ αα δϕ dA = ϕδϕ dA (3.16.88)
A A A
delta wϕ dA = ϕδw dA + wδϕ αα dA = ϕδw dA + wδϕ dA.
A A A A A
Assuming that these series can be differentiated term-wise four times and using the
relation (3.16.81), we obtain from (3.16.90) the following equations:
/ 0
[n (n + 1)]2 − (1 − υ) n (n + 1) Dn + (1 − υ) n (n + 1) Cn = 0
! 2 " (3.16.92)
h 2
2 σ
(1 − υ) n (n + 1) Dn + [n (n + 1)] − 1 − υ n (n + 1) Cn = 0.
12R2 E
The condition for a non-trivial solution is
1 − υ 2 [n (n + 1) − 2] + [n (n + 1) − (1 − υ)]
2 (3.16.93)
h 2
2 σ
× n2
(n + 1) − 1 − υ n (n + 1) = 0,
12R2 E
which yields
σ n (n + 1) − 2 h2 n (n + 1)
= + . (3.16.94)
E n (n + 1) [n (n + 1) − (1 − υ)] 12R2 1 − υ 2
For values of n of O(1), the first term on the right-hand side is of order unity, which
means σ/E = O (1). Because we must stay in the elastic range, we require σ/E 1,
which is only possible for n 1. This means that (3.16.94) can be simplified to yield
σ 1 h2 n (n + 1)
= + . (3.16.95)
E n (n + 1) 12R2 1 − υ 2
It follows that because n 1, the wavelength L of the deformation pattern is small,
i.e., L/R 1, which means that we can restrict ourselves to shallow shell theory.
The functional (3.16.78) can now further be simplified to yield
Eh w 1
P2 [u; λ] = (ϕ)2 − 2 (1 + υ) ϕ + (1 − υ) (ψ) 2
2 (1 − υ 2 ) R 2
A
(3.16.96)
h 2
2
2 σ h L2
+ (w) + 1 − υ ww dA 1 + O + .
12 E R R2
Omitting the error term, we rewrite this functional in the form
!
Eh w 2 1
P2 [u; λ] = ϕ − (1 + υ) + (1 − υ) (ψ)2
2 (1 − υ 2 ) R 2
A
" (3.16.97)
2
2 w h2 2
2 σ
− 1+υ + (w) + 1 − υ ww dA.
R2 12 E
Notice that for the present simplified functional, we did not need any a priori
assumptions with respect to the wavelength of the deformation pattern. The con-
dition that L/R 1 follows from the condition that σ/E 1 to stay in the elastic
range.
3.16 Buckling behavior of a spherical shell under uniform external pressure 199
In our present functional, we only have quadratic terms with positive constants,
except for the load term. For the functional to become semi-positive-definite, the
term with ψ must vanish, which leads to the first of the equations of (3.16.85), and
the term with ϕ must also vanish, which yields the equation
w
ϕ = (1 + υ) . (3.16.98)
R
This admits of a single-valued and non-singular solution for the potential ϕ if the
integral of the normal deflection w over the entire surface of the shell vanishes for
all buckling modes. To show this, apply the operator A () dA to both sides of this
equation. The left-hand member then vanishes (divergence theorem), so the right-
1
hand side must also vanish. This condition is satisfied because w = ∞ α
n Cn Sn (x )
and A Sn dA = 0 (n ≥ 1). Apart from irrelevant arbitrary constants, the solution to
(3.16.98) is
∞
1+υ
ϕ (xα ) = − RCn Sn (xα ) . (3.16.99)
n
n (n + 1)
Let us now return to the functional (3.16.97). The remaining terms contain only
w, so that a stationary value of P2 [u; λ] is obtained by putting the variation with
respect to w equal to zero. Using (3.16.91), the condition for a non-trivial solution
yields again the expression for σ/E given in (3.16.96). To obtain the critical value
of σ/E, we minimize this expression with respect to n (n + 1), which yields (for a
continuous variable n∗ )
2cR
n∗ (n∗ + 1) = , (3.16.100)
h
so that
Eh
σcr = . (3.16.101)
cR
Notice that this is in complete agreement with our earlier result of (3.15.30).
The positive root n∗ of (3.16.100) is in general non-integral. However, the near-
est integers do not differ more from n∗ than O n∗−1 , which means a relative error
of O (h/R), and which is negligible within the framework of shell theory. The crit-
ical load factor λ1 is therefore always equal to unity within the limits of accu-
racy of shell theory. The associated buckling mode is any spherical surface har-
monic of integral degree [n∗ ] or [n∗ ] + 1 associated with the integer that yields the
smallest value of the right-hand member of (3.16.95). The corresponding buckling
mode is
wn = ah Sn (xα ) , (3.16.102)
where A is the amplitude factor and Sn (xα ) is given in (3.16.82). This representation
shows that at the same value of the load factor λ1 , there are 2n + 1 independent
buckling modes.
200 Applications
In the special case that R/h is such that (3.16.100) yields two integer values for
n∗ , the right-hand side of (3.16.95) is the same for both values. In that case, the
corresponding buckling mode is given by
ψ = 0, ∀n. (3.16.104)
A necessary condition for stability at the critical bifurcation load is that the
third variation vanishes identically for any linear combination of buckling modes.
Returning to (3.16.17) and remembering that γαβ = θαβ + 12 w,α w,β , we see that the
only contribution to P3 [u] originates from the membrane strains,
Eh α β
P3 [u] = (1 − υ)θ αβ w w + υθα w,κ wκ dA
2(1 − υ 2 ) α
A
!
Eh 1 w α β
= (1 − υ) (uα|β + uβ|α ) − aαβ w w (3.16.105)
2(1 − υ )
2 2 R
A
α
w κ 0
+ υ(u α
− 2 )w,κ w dA.
R
For the evaluation of this functional, we recall that
(1 − υ)R (1 − υ)R
ϕ=− w, ψ = 0, uα = ϕ,α = − w,α , (3.16.106)
n(n + 1) n(n + 1)
n (n + 1)
w β w β w dA = w3 dA. (3.16.109)
2R2
A A
where we have made use of (3.16.100). For the evaluation of this term and a further
discussion of the problem we refer to Koiter’s paper.†
The result is that P3 [ah uh ] vanishes if the degree of the buckling mode is odd,
but it does not vanish when the degree of the buckling mode is even. This means
that in the latter case, the critical bifurcation point is unstable. In the case of an even
degree of the buckling mode, stability is governed by the sign of A4 ; cf. (3.4.38) and
(3.4.53). It turns out that A4 is negative for both even and odd degrees of buckling
modes, so the critical bifurcation point is unstable.
In the following discussion, we shall assume that the fundamental state is a mem-
brane state of stress. (This assumption is usually violated in experiments.) The
second variation of the energy functional is then given by
! 0
Eh κ 2 h2 / κ 2
P2 [u] = (1 − υ) θαβ θ αβ
+ υ (θ ) + (1 − υ) ραβ ραβ
+ υ (ρ )
2 (1 − υ 2 ) κ
12 κ
A
1
+ Nαβ [(θ·α
κ
− ωκ·α ) (θκβ − ωκβ ) + ϕα ϕβ ] (3.17.1)
2
"
1
+ P (wθα − u ϕα ) dA.
α α
2
† W.T. Koiter, The nonlinear buckling problem of a complete spherical shell under uniform external
pressure. Proc. Kon Ned. Ak Wit., Series B, 72, 40–123 (1969).
202 Applications
that
∂ 1 ∂ 1
( ) ≡ ( ) , ( ) ≡ ( ). (3.17.2)
∂x R ∂y R
In the following, we shall use the Koiter-Sanders strain-displacement relations for
cylindrical shells.†
1 1 · 1
θxx = u θyy = (v + w) θxy = (u· + v )
R R 2R
1 1 1
ωxy = (v − u· ) ϕx = w ϕy = (w· − v) . (3.17.3)
2 R R
1 1 1 3 1
ρxx = 2 w ρyy = 2 (w· − v)· ρxy = 2 w· − v + u·
R R R 4 4
Because we are dealing with Cartesian coordinates, the functional (3.17.1) can read-
ily be rewritten to yield
!
Eh
P2 [u] = θ2 + θyy
2
+ 2υθxx θyy + 2 (1 − υ) θxy
2
2 (1 − υ 2 ) xx
A
"
h2 2
+ ρ + ρyy + 2υρxx ρyy + 2 (1 − υ) ρxy dA
2 2
12 xx
! (3.17.4)
1 1
2 2
+ Nxx θxx 2
+ (θyx − ωyx ) + ϕ2x + Nyy θyy 2
+ (θxy − ωxy ) + ϕ2y
2 2
A
"
1
+ P [w (θxx + θyy ) − uϕx − vϕy ] dA.
2
2
The term with Nxy is absent because there are no shear forces. The terms Nxx θxx and
2
Nyy θyy can be neglected compared to the terms Eh θxx + θyy + 2υθxx θyy .
2 2
Koiter‡ has shown that within the context of the classical theory of shells, the
changes of curvature can be altered by adding terms of the order of the strain in
the middle surface divided by the minimum radius of curvature without altering the
accuracy of the theory. We now consider the following modified energy functional,
∗ Eh3 2 (1 − υ) 2
P2 [u] = P2 [u] + − θxx ρyy + θyy (ρxx + ρyy )
24 (1 − υ 2 ) R R
A
(3.17.5)
1 2 2 (1 − υ) 3 (1 − υ) 2
+ 2 θyy + θxy ρxy − θxy dA.
R R 2R2
Koiter has shown that an upper bound for the change in strain energy is
∗
P [u] − P2 [u] < 0.467 h P2 [u] 1 + O h . (3.17.6)
2
R R
† Cf. W. T. Koiter, Summary of equations for modified simplistic possible accurate linear theory of
thin circular cylindrical shells. Rep. Lab. For Appl. Mech. Delft, 442.
‡ W. T. Koiter, A consistent first approximation in the general theory of thin elastic shells. Proc.
IUTAM Symp. On Theory of Thin Elastic Shells (Delft, Aug. 1959; North-Holland Publ. Co., 1960),
pp. 12–33.
3.17 Buckling of circular cylindrical shells 203
Substitution of the relations (3.17.3) into this modified energy functional yields
!
∗ Eh 1
u2 + (v· + w)2 + 2υu (v· + w) + (1 − υ) (u· + v )
2
P2 [u] =
2 (1 − υ ) R
2 2 2
A
h2 2
+ w + w··2 + 2υw w·· + 2 (1 − υ) w·2 + 2w (w + w·· ) + w2
12R2
"
h2 · · · · ·· · ·
+ 2(1 − υ) [u v − u v + u w − u w + v w − v w ] dA (3.17.7)
12R2
!
Nxx 2 Nyy / ·2
+ 2
v + w2 + 2
u + (w· − v)2
2R 2R
A
"
P · ·
+ u w − uw + v w − vw + v + w 2 2
dA
2R
This modified functional has the advantage that the first term with h2 /R2 contains
only the displacement w, whereas the second term with h2 /R2 contains only expres-
sions of the divergence type, which do not alter the differential equation. Consider,
e.g.,
R−1 (u v· − u· v ) dA = (u vvy − u· vvx ) ds
A ∂S
· ·
− (u − u )v dA = − u· vvx ds,
A ∂S
∂S
· ·
− (v − v ) u dA = uv· vx ds,
A ∂S
v = w = 0 at x = 0, x = . (3.17.14)
u = 0 at x = 0, x = . (3.17.16)
3.17 Buckling of circular cylindrical shells 205
The second line integral in (3.17.12) vanishes identically because all the functions
must be single-valued at θ = 0, and θ = 2π.
Let us now consider the terms with η,
2π !
· (1 − υ) h2 1
R 2
v + w + υu + (u + w ) − λy w η·
12 R2 2
0 0
1 · (1 − υ) h2
× + (1 − υ) (u + v ) + (−u − w ) + λx v η
· ·
(3.17.17)
2 12 R2
"
1 dx
− λy w· η dθ = 0.
2 R
2π
(1 − υ) h2 · 1
−R 2
v·· + w· + υu· + (u + w· ) − λy w·
12 R2 2
0 0
1 (1 − υ) h2 · 1 dx
+ (1 − υ) (u· + v ) − (u + w ·
) + λ x v
+ λ y w ·
ηdθ
2 12 R2 2 R
(3.17.18)
2π
(1 − υ) h 2
1 dx
+ R2 v· + w + υu + (u + w ) − λy w η
12 R2 2 R
0 0
2π
1 (1 − υ) h2 ·
+ R2 (1 − υ) (u· + v ) − 2
(u + w ·
) + λ x v
η dθ = 0,
2 12 R 0
0
1 1
v·· + (1 − υ) v + (1 + υ) u· + w· + λx v = 0, (3.17.19)
2 2
and in the case of a simply supported edge, both line integrals vanish.
Finally, we consider the terms with ζ,
2π !
h2 1
v· + w + υu + (w
+ w ··
+ w) + λ y (−u
− v·
− 2w) ζ
12R2 2
0 0
1 1
+ λx w + λ y u ζ + λ y w · − v ζ ·
2 2
2
h
+ [w + υw·· + w + (1 − υ) v· ] ζ (3.17.20)
12R2
h2
+ [w·· + υw − (1 − υ) u + w] ζ ··
12R2
"
h2 · · · dx
+ [2 (1 − υ) w + (1 − υ) u − (1 − υ) v ] ζ dθ = 0.
12R2 R
206 Applications
2π !
h2
v· + w + υu + (w
+ w ··
+ w) − λ x w
− λ y (u
+ w ··
+ w) ζ
12R2
0 0
h2
− [w + υw·· + w + (1 − υ) v· ] ζ
12R2
h2
− [w··· + υw· − (1 − υ) u· + w· + 2 (1 − υ) w· + (1 − υ) u·
12R2
"
dx
− (1 − υ) v ] ζ · dθ
R
(3.17.21)
2π 2π
1 1 dx
+ λx w + λy u ζ dθ + λy w· − v ζ
2 2 R
0 0 0 0
2π
h2
+ [w + υw·· + w + (1 − υ) v· ] ζ
12R2
0
+ [2 (1 − υ) w· + (1 − υ) u· − (1 − υ) v ] ζ | 0 dθ
2π
2
h ··
dx
·
+ 2
(w + υw − (1 − υ) u + w) ζ R = 0,
12R
0 0
where the second and the fourth line integrals vanish because the displacements and
their derivatives are periodic in the y-direction. For simply supported edges, the first
line integral and the second term in the third line integral vanish. Integrating by
parts once again, we obtain
2π !
h2
v· + w + υu + [w + w·· + w + w + υw·· + w + (1 − υ) v·
12R2
0 0
+ w···· + υw·· − (1 − υ) u·· + w·· + 2 (1 − υ) w·· + (1 − υ) u·· − (1 − υ) v· ]
"
dx
− λx w − λy (u + w·· + w) ζ dθ
R
2π
h2
+ 2
[w + υw·· + w + (1 − υ) v· ] ζ 0 dθ (3.17.22)
12R
0
2π
h2
− [w + υw·· + w + (1 − υ) v· ] ζ 0 dθ
12R2
0
h2
− [w··· + υw − (1 − υ) u· + w· + 2 (1 − υ) w·
12R2
0
2π dx
+ (1 − υ) u· − (1 − υ) v ] ζ 0 = 0.
R
3.17 Buckling of circular cylindrical shells 207
Here, the last line integral vanishes because the displacements and their derivatives
are periodic in the y-direction. For simply supported edges, the second of the line
integrals vanishes. We now obtain the third equilibrium equation,
h2
v· + w + υu + [w + 2w·· + w···· + 2 (w + w·· ) + w]
12R2
(3.17.23)
− λx w − λy (u + w·· + w) = 0,
w = 0 at x = 0, x = . (3.17.25)
The equations (3.17.13), (3.17.15), and (3.17.26) are known as the Koiter-Morley
Equations.
We now try to find a solution of these equations of the form
! 1
1 2
+ (1 − υ) q2 p 2 + q2 − 3p 2 − q2 − (1 − υ) (1 − 2υ) p 4 (3.17.29)
2 2
.
h2 2 2 1
+ p +q −12
(1 − υ) p q + q
2 2 4
λy
12R2 2
1
+ q2 q2 − 1 − p 2 (1 − υ) p 2 + q2 λ2y + · · · = 0.
2
12R 2 (3.17.30)
+ λy q2 p 2 + q2 − 3p 2 + q2 = 0.
i) q = 0. In this case, the term with λy vanishes, i.e., there is no influence of the
external pressure P. The resulting equation now reads
h2 4 2 2
1 − υ2 p 4 + p p − 1 + λx p 6 = 0, (3.17.31)
12R2
from which (for p = 0 )
2
1 − υ2 h2 p 2 − 1
λx = − − . (3.17.32)
p2 12R2 p2
To stay in the elastic range, we must have large values of p, which means that
we can neglect unity with respect to p 2 , so that the expression for λx can be
simplified to yield
1 − υ2 h2 2
λx = − 2
− p . (3.17.33)
p 12R2
Notice that λx is negative, i.e., the cylinder is loaded in compression. The crit-
ical value for λx follows from (3.17.33) by minimizing the right-hand side with
respect to p 2 , which yields
2 2cR h
p min = , (λx )min = − 1 − υ2 . (3.17.34)
h cR
Notice the p 2 min is indeed a large number, as R/h 1.
ii) q = 1. In this case, the equation becomes
h2 4 2 2 2
1 − υ2 p 4 + p p + 1 + λ x p 2
p 2
+ 1 + 1
12R2 (3.17.35)
+ λy (p − p 2 ) = 0.
4
3.17 Buckling of circular cylindrical shells 209
π2 E 3
F = −2πRNxx = πR h, (3.17.38)
2
where πR3 h is the moment of inertia of the cylinder. When λx = 0, λy is pos-
itive, namely,
π2 R2
λy = 1 − υ2 . (3.17.39)
2
This is a case of buckling under internal uniform pressure P.
Using (3.17.8), we can write
π2 E 3
F = −PπR2 = πR h, (3.17.40)
2
which is again the Euler load.
c) p = O(1) : In this case, λx is of order unity, which is not possible in the elastic
range. Hence, values of p = O (1) must be excluded.
Before proceeding with the general discussion of (3.17.30), we notice that this
is the equation of a straight line in the λx , λy plane. Let p i , qi (i ∈ N+ ) denote
minimizing values of p and q, respectively. The corresponding values (λx )cr , (λy )cr
then lie on a straight line i , which is the boundary between a stable and an unstable
region. Let Dis be the stable region. Let Ds = ∩i=1...n Dis ; then Ds is convex (i.e., if
x, y ∈ Ds , then µx + (1 − µ) y ∈ Ds , 0 ≤ µ ≤ 1 ), if the fundamental state is linear.
(The origin (λx , λy ) = (0, 0) ∈ Ds .)
(k)
To prove this property we proceed as follows: Let Sij be the stress due to a
unit load in the loading case k. The total stress under n loads can then be written
Pπ R 2 −P Pπ R 2
Figure 3.17.1
210 Applications
1 (k)
as Sij = nk=1 λkSij (linearity). Substitution of this expression into the functional
P2 [u] yields
n
(k) 1
P2 [u] = λkSij uh,i uh,j + Eijkl θij θk dV. (3.17.41)
2
V k=1
and hence
n
.
(1) (2) (k) 1
µλk + (1 − µ) λk Sij + Eijkl θij θk dV ≥ 0 (3.17.44)
2
V k=1
(1) (2)
so that µλk + (1 − µ) λk ∈ Ds + ∂Ds , q.e.d.
Let us now return to the buckling equation (3.17.30) and let us first consider the
case that there is no axial loading, λx = 0. In this case, λy is given by
h2 2 2 2
1 − υ2 p 4 + 2
p + q2 p 2 + q2 − 1
λy = − 12R . (3.17.45)
2
q2 (p 2 + q2 ) − (3p 2 + q2 )
h2 2
λy = − 2
q −1 . (3.17.46)
12R
The critical value is attained for the smallest possible value of q. However, q = 0
cannot be taken into account because then the limit process is not valid. The value
q = 1, the Euler buckling, also can not be taken, so that the minimum value for q is
q = 2,
h2
λycr = − . (3.17.47)
4R2
This is a long cylinder (or a ring) under uniform external pressure (see Fig-
ure 3.17.2).
3.17 Buckling of circular cylindrical shells 211
E h3
P P=−
(
4 1 − υ R3
2
)
Figure 3.17.2
Let us now consider the case that λy = 0, i.e., we have only axial loading. We
then have
2 2
1 − υ2 p 2 h2 p 2 + q2 p 2 + q2 − 1
− λx = 2
+ . (3.17.48)
(p 2 + q2 ) + q2 12R2 p 2 (p 2 + q2 )2 + q2
Let us now consider the case q = O (1) but q = 1. Then, to stay in the elastic range,
we must have either p 2 1 or p 2 1.
Let us first consider the case p 2 1. In this case, (3.17.48) can be simplified to
yield
2
1 − υ2 p 2 h2 q2 q2 − 1
− λx = + . (3.17.49)
q2 (q2 + 1) 12R2 p 2 (q2 + 1)
from which
h 2 2
p2 = q q −1 . (3.17.50)
2cR
Because p = kπR/ 1, we must have
)
2cR 1
= kπR 1, (3.17.51)
h q q2 − 1
h q2 − 1
(−λx )cr = 1 − υ2 . (3.17.52)
cR q2 + 1
For q = 2, we have
h
(−λx )cr = 0.6 1 − υ2 . (3.17.53)
cR
212 Applications
p0 = 2cR / h
pq1 pq 2 p p
0
Figure 3.17.3
3.17 Buckling of circular cylindrical shells 213
For the evaluation of this integral, one must deal with integrands of the form
% & % & % & % &
cos cos cos cos
(q1 y/R) (q2 y/R) (q3 y/R) (p 1 x/R)
sin sin sin sin
% & % & (3.17.65)
cos cos
× (p 2 x/R) (p 3 y/R) ,
sin sin
214 Applications
For a sufficiently long cylinder, the only non-vanishing terms (after integration) are
the cosine terms with an argument of zero; i.e., we must only consider the combina-
tions
p 1 ± p 2 ± p 3 = 0, q1 ± q2 ± q3 = 0, (3.17.67)
under the side condition that (p i , qi ) (i = 1, 2, 3) lies on the semicircle (3.17.59). The
only combination that satisfies these conditions is
1 1 1
p 1,2 = p 0 ± p 20 − 4q2 , 0/q ≤ p 0
2 2 2 (3.17.68)
p 3 = p 0 , q3 = 0.
The result of a fairly lengthy but elementary calculation is
% &
3π Eh4 2 1
P3 [u] = q b0 cq1 cq2 + mb0 cm = 0,
2
(3.17.69)
4 R2 q
2
so that
1 Eh2
P2 [ah uh ; λ] = (1 − λ) 3 w2 dA, (3.17.72)
2 cR
A
q q
2
The equations of equilibrium are obtained by taking the variations with respect to
b0 , cq1 , cq2 , and cm, respectively, which yields
% &
1
4 (1 − λ) p 20 b0 + 3c q2 cq1 cq2 + m2 c2m = 0
q
2
3.17 Buckling of circular cylindrical shells 215
The second and third of these equations contain only cq1 and cq2 , and have a non-
trivial solution when
Bearing in mind that p q1 and p q2 lie on the semicircle and satisfy the relation
p q1 p q2 = q2 (see Figure 3.17.4), we can simplify (3.17.75) to yield
from which
3cb0 = ±2 (1 − λ) . (3.17.77)
3cb0 = −2 (1 − λ) ∨ cm = 0. (3.17.78)
When the first of these solutions is chosen, compatible with (3.17.77), cm remains
undetermined. This result is due to the fact that we have not taken into account the
higher order terms P4 [ah uh ]. With this choice for b0 , we find from the second and
third of the equations of (3.17.75),
cq2 2 (1 − λ) p 2q1 pq
=− = 1 (317.79)
cq1 3cq2 [−2 (1 − λ) /3c] p q2
1 8 (1 − λ)2 2
q2 cq1 cq2 + m2 c2m = p0, (3.17.80)
q
2 9c2
pq pq p
1 2
Figure 3.17.4
216 Applications
32 (1 − λ)2 R 1
c2m = , (3.17.83)
9c h m2
which with m = 12 p 0 = cR/2h can be rewritten in the form
8 (1 − λ)
cm = ± , (3.17.84)
3c
i.e., |cm| = 4b0 , twice the shell thickness, much larger than the amplitude of the
axisymmetric buckling mode.
Although the amplitudes of the buckling modes are indeterminate, the shorten-
ing of the shell is determined. From (2.4.61), the general formula for the additional
shortening is given by
∂P [ueq ; N]
ε = − . (3.17.85)
∂N
The compression force N is given by
2π 2
N = 2πRhσ = Eh λ, (3.17.86)
c
and the elastic energy is given by (3.17.73). Using this relation, we find
∂P ∂P c
ε = − =−
∂N ∂λ 2πEh2 . (3.17.87)
1 h2
= 2p 20 b20 + p 2q1 c2q1 + p 2q2 c2q2 + m2 c2m ,
8 R2 q
σ
λ=
σ cr
unstable (3.17.90)
2/3 1
ε / ε cr
Figure 3.17.5
Because
∂2 P π Eh4 2
= p (1 − λ) < 0 for λ > 1, (3.17.90)
∂b02 c R2 0
it follows that for λ > 1 the equilibrium path is unstable. Because there are no sta-
ble equilibrium configurations in the vicinity of the bifurcation point, the shell will
collapse explosively.
Because the post-buckling behavior of the perfect cylinder is unstable, the cylin-
der is an imperfection-sensitive structure. To get quantitative results for the imper-
fection sensitivity, we consider first the case of an axisymmetric imperfection in the
direction of the buckling mode. Let this imperfection be given by
w0 = κh cos p 0 x/R. (3.17.91)
According to the general theory, the term 12 σhw2,x in the energy functional must be
replaced by
1 2
σh w,x + w0x w,x (3.17.92)
2
so that in the presence of the axisymmetric imperfection, the energy functional is
given by
P2 [u, u0 ; λ] + P3 [u] = P2 [u; λ] + P3 [u]
(3.17.93)
π Eh4 2
− p b0 κλ.
c R2 0
The extra term contains only the amplitude b0 , so that only the first of the equilib-
rium equations (3.17.76) is affected. This equation must be replaced by
% &
1
4 (1 − λ) p 20 b0 + 3c q2 cq1 cq2 + m2 c2m − 4λp 20 κ = 0. (3.17.94)
q
2
Because prior to buckling cq1 , cq2 , and cm are zero, we readily obtain
λ
b0 = κ. (3.17.95)
1−λ
218 Applications
0
0 .2 .4 .6 κ
Figure 3.17.6
The values of b0 for a non-trivial solution for cq1 , cq2 of the other equilibrium
equations follow from (3.17.76), so we must have
4 (1 − λ)2 λ2
b20 = = κ 2, (3.17.96)
9c (1 − λ)2
from which follows
3 5 5
(1 − λ)2 = cλ |κ| ≈ λ |κ| for c = . (3.17.97)
2 2 2
In this case, cq1 and cq2 are nonzero, which means that branching from the axisym-
metric mode to a non-axisymmetric buckling mode will occur. It follows that small
imperfections reduce the buckling load considerably, e.g., a reduction of the buck-
ling load by 50% is obtained when κ = 1/5, say, for an amplitude of the imperfection
of 20% of the shell thickness. This effect is shown in Figure 3.17.6.
Let us now consider a more general imperfection again in the direction of the
buckling mode,
where we have used the fact that cm = 4b0 ; i.e., the amplitude of the non-symmetric
mode is four times that of the symmetric mode. In this case, the energy functional is
given by
q
2
For λ = 0, these equations yield the null solution. For λ < λcr , cq1 and cq2 are equal
to zero, and the first and last of these equations are simplified to read
3
4 (1 − λ) b0 + cc2m − 4λκ = 0
8 (3.17.101)
2 (1 − λ) cm + 3cb0 cm − 8λκ = 0
and have a solution for cm = 4b0 . The resulting equation for b0 then reads as
2 2
b20 + (1 − λ) b0 − λκ = 0, (3.17.102)
3c 3c
which yields
1−λ (1 − λ)2 2
b0 = − ± 2
+ λκ. (3.17.103)
3c 9c 3c
We must choose the positive solution because the result of the negative solution
does not refer to the undeformed state.
The expression under the square-root sign is always positive when κ > 0, and
then the imperfection has a stabilizing effect. However, for κ < 0 the expression
may become negative, and the stability limit is reached when
Notice that now (1 − λ)2 is four times larger than in the case of an axisymmetric
imperfection. In this case, we do not have a branch point but a limit point. The
behavior is shown in Figure 3.17.7. Notice that an imperfection of 5% of the shell
thickness (κ = 0.05) yields a reduction of the critical load by 50%.
2
b0 = − (1 − λ )
3c
1
stable
unstable
b0
Figure 3.17.7
220 Applications
2π −5 −5
−5 0 3
2 3 1
1 1 1
2
π
3 3 −5
0 0
−5
3 −5 3 −5
−3π −2π −π 0 π 2π 3π β
0
3
−5 −π −5 3
3 2
0 0
23 2
0
−5 −5 3 −5
−2π
Figure 3.17.8
For
. .
α = (2j + 1) π α = 2j π
and , (3.17.106)
β = (2k + 1) π β = 2kπ
w0
5
4 µ 2 / m2 = 0.1
3
2
1
0 2 4 6 8 10 mx / R
Figure 3.18.1
So far, we have only considered overall imperfections. In practice, one also often
encounters localized imperfections (dimples), which are usually inward deflections.
To investigate the influence of these local imperfections, we consider an imperfec-
tion of the form
υ 1−υ
sin(mx/R) cos(my/R) e− 2 µ (x +y ) /R
1 2 2 2 2
u = hb0 − sin( p 0 x/R) + (3.18.2)
2m m
3+υ
cos(mx/R) sin(my/R) e− 2 µ (x +y ) /R .
1 2 2 2 2
v = −hb0
m
† Cf. Koiter’s paper under the same title, Rep. Lab. For Appl. Mech. Delft 534, I. N. Vekua Anniver-
sary volume (Moscow, 1978), pp. 242–244.
222 Applications
R
Let us now first consider integration in the x-direction. First, we notice that the first
term in the integrand leads to integrals in the x-direction of the form
/2
cos (nx/R) e−µ
2 2
x /R2
dx. (3.18.5)
− /2
Because the integrants are rapidly decaying functions, we can replace the bound-
aries of the integrals by −∞ to ∞, so that we must deal with the following integrals,
∞ ∞
−µ2 x2 /R2 x
sin (nx/R) e−µ x /R dx,
2 2 2
cos(nx/R) e dx,
R
−∞ −∞
(3.18.8)
∞ 2
x
cos (nx/R) e−µ x /R dx.
2 2 2
R 2
−∞
3.18 The influence of more-or-less localized short-wave imperfections 223
iy
a C
A′ i
2 B′
−R R
A 0 B x
Figure 3.18.2
Let us consider the first of these integrals. It is convenient to consider the inte-
gral
4
e−z dz
2
(3.18.9)
C
and this expression tends uniformly to zero as R tends to infinity. Thus, the portions
of the whole integral that arise from the vertical sides tend to zero, and if we carry
out the passage to the limit R → ∞ and note that dz = d(x + 12 ia) = dx, on A B we
can express the result of Cauchy’s theorem as follows,
∞ 2
∞
−(x+ 12 ia)
e−x dx = 0.
2
e dx −
−∞ −∞
−∞ −∞ −∞ −∞ 0
so that we have
∞
√
a2 /4
e−x (cos ax − i sin ax) dx =
2
e π,
−∞
and because sin ax is an odd function, its contribution in the integral vanishes so that
∞
√ −a2 /4
cos ax e−x dx =
2
πe . (3.18.10)
−∞
224 Applications
Noting that n is either zero or a multiple of m (m 1), it follows that only the case
n = 0 must be taken into account.
Let us now consider the other integrals in (3.18.8). Differentiating (3.18.11) with
respect to n, we obtain
∞ √
x −µ2 x2 /R2 R π − 4µn22
sin(nx/R) e dx = ne , (3.18.12)
R 2µ3
−∞
which is zero for n = 0 and is exponentially small for large values of n, and may thus
be neglected. Differentiating (3.18.12) with respect to n, we obtain
∞ √
x2 −µ2 x2 /R2 R π n − n
2
2
cos(nx/R) e dx = 3
1 − 2
e 4µ2 . (3.18.13)
R 2µ 2µ
−∞
For large values of n this integral vanishes, and for n = 0 the contribution is
√
R π/2µ3 . Summarizing our results, we find that the contribution of the first term in
√
(3.18.4) is of order m2 R π/µ, whereas the contribution of the third term in (3.18.4)
√
is of order (1/4)µ2 R π/µ. For small values of µ2 /m2 , we may neglect the contri-
bution of the third term compared with the first term, making a relative error in the
energy of order µ2 /m2 .
Let us now evaluate the first term using the fact that the integrand is also expo-
nentially decaying in the y-direction, so that the boundaries of integration 0 to 2π
can be replaced by −∞ to ∞,
∞ ∞
[2m sin(2mx/R) + 4m sin(mx/R) cos(my/R)]2 e−µ (x +y ) /R dx dy
2 2 2 2
−∞ −∞
∞ ∞
= 4m 2
[sin2 (2mx/R) + 4 sin(2mx/R) sin(mx/R) cos(my/R)
−∞ −∞
∞ ∞ !
3 1
= 4m 2
− cos(2mx/R) − cos(4mx/R) − cos(2my/R) (3.18.14)
2 2
−∞ −∞
+ 2 [cos(mx/R) − cos(3mx/R)] cos(my/R)
"
+ cos(2mx/R) cos(2my/R) e−µ (x +y ) /R dx dy
2 2 2 2
3 R2
= 4m 2
π + exponentially small terms .
2 µ2
3.18 The influence of more-or-less localized short-wave imperfections 225
The corresponding term for an imperfection of the form (3.17.98) is 4m2 π R, which
means that we can use the results from the previous section if we replace by R/2µ2 .
By similar arguments, we find that for the cubic terms must be replaced by R/3µ2 ,
so the energy expression becomes
∗∗ 3π Eh3 2 3
F = (1 − λ) b0 + cb0 − 2λκb0 .
2
(3.18.15)
2 µ2 3
The equilibrium equation now reads
2 (1 − λ) b0 + 2cb20 − 2λκ = 0, (3.18.16)
from which
1−λ (1 − λ)2 λκ
b0 = − ± + . (3.18.17)
2c 4c2 c
For positive values of κ, the term under the square-root sign is always positive, which
means that an imperfection in the outward direction is stabilizing. However, for
negative values of κ a limit point will occur, i.e., when
(1 − λ∗ )2 = −4λ∗ cκ, κ < 0. (3.18.18)
When we compare this expression with (3.17.104), we see that the factor 6 is
replaced by a factor 4; i.e., a local imperfection is equally harmful as an imperfection
of the corresponding periodic type, with an amplitude reduced by a factor 2/3.
To investigate the accuracy of the present first-order approximation,
Gristchak,† who worked as a research fellow in our laboratory, calculated a second
approximation with an error of order µ4 /m4 . His improved approximation yields
the following equation for the critical load:
2
µ2 µ2
1 + (d − ε) 2 − λ∗ = −4λ∗ cκ 1 + (e − ε) 2 , (3.18.19)
m m
where
3 7 + 5υ − 6υ2
ε= , e= ,
8 12 (1 − υ2 )
(3.18.20)
1
d= {(3 − υ) [υ2 + 2 (1 − υ)2 + 2 (3 + υ)2 ] + 80 1 − υ2 },
9b (1 − υ )
2
† V. Z. Gristchak, Asymptotic formula for the buckling stress of axially compressed circular cylin-
drical shells with more-or-less localized shortwave imperfections. W.T.H.D., Rep. 88.
226 Applications
λ*
1.2
1.116
1.0
0.8
0.6
µ 2/m 2 = 0.1
0.4
µ 2/ m 2 = 0
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 cκ
Figure 3.18.3
These results show that the first approximation is indeed a good approximation.
In our analysis, we have employed the Rayleigh-Ritz method, which in itself yields
an upper bound for the critical load. However, we have also neglected terms in
powers of µ/m, so it is not certain that the present results yield an upper bound.
Let us finally remark that actual shells usually have local imperfections due
to damage, and these imperfections are in the inward direction, and are therefore
strongly destabilizing. It is further interesting to note that even the carefully manu-
factured cylindrical shells used for the Lockheed tests† had local imperfections.
Over de stabiliteit van het elastisch evenwicht (On the stability of elastic equilibrium). Proef-
schrift T. H. Delft (1945).
On the stability of elastic equilibrium (Washington, 1967), 6 + 202 pp. (NASA Technical
Translation F-10, 833, Clearinghouse US Dept. of Commerce/Nat. Bur. of Standards N67-
25033).
The stability of elastic equilibrium (Palo Alto, Cal., 1970), 13 + 306 pp. (Stanford Univ., Dept.
of Aeronaut. and Astronaut. Report AD 704124/Air Force Flight Dynamics Lab., Report
TR-70-25).
“Buckling and post buckling behaviour of a cylindrical panel under axial compression,”
Reports and Transactions National Aeronautical Research Institute, 20 (1956) pp. 71–84
(Nat. Aero. Res. Inst. Report No. S476).
“A consistent first approximation in the general theory of thin elastic shells.” Proc.
I. U. T. A. M. Symposium on the Theory of Thin Elastic Shells (Deflt, August 1959). North-
Holland Publishing Cy., Amsterdam, 12 (1960).
“A sysematic simplification of the general equations in the linear theory of thin shells.” Proc.
Kon. Ned. Ak. Wet., B64, 612–619 (1961).
Introduction to the post-buckling behaviour of flat plates. Actes du Colloque sur le Comporte-
ment Postcritique des Plaques Utilisées en Construction Métallique, Liège (1963).
“Elastic stability and post-buckling behaviour.” Proc. Symp. Nonlinear Problems, University
of Wisconsin Press, Madison, 257–275 (1963).
“The concept of stability of equilibrium for continuous bodies.” Proc. Kon. Ned. Ak. Wet.,
B66, 173–177 (1963).
“The effect of axisymmetric imperfections on the buckling of cylindrical shells under axial
compression. Proc. Kon. Ned. Ak. Wet., B66, 265–279 (1963).
“Knik van schalen (buckling of shells).” De Ingenieur, 76, 033–041 (1964).
“On the instability of equilibrium in the absence of a minimum of the potential energy.” Proc.
Kon. Ned. Wet., B68, 107–113 (1965).
“The energy criterion of stability for continuous elastic bodies.” Proc. Kon. Ned. Ak. Wet.,
B68, 178–202 (1965).
“Post buckling analysis of a simple two-bar frame.” Recent Progress in Applied Mechanics
(The Folke Odqvist Volume), Almqvist & Wiksell, Stockholm (1967) 337–354.
“On the nonlinear theory of thin elastic shells” Proc. Kon. Ned. Ak. Wet., B69, 1–54.
“General equations of elastic stablity for thin shells.”Proc. Symposium on the Theory of
Shells, April 4–6, 1066, Houston, Texas, U. S. A. (Houston, Texas U. S. A., 1967) pp. 187–
227.
“A sufficient condition for the stability of shallow shells.” Proc. Kon. Ned. Ak. Wet., B70,
367–375 (1967).
227
228 Selected Publications of W. T. Koiter on Elastic Stability Theory
“The nonlinear buckling problem of a complete spherical shell under uniform external pres-
sure.” Proc. Kon. Ned. Ak. Wet., B72, 40–123 (1969).
“Postbuckling theory. Jointly with J. W. Hutchinson.” App. Mech. Reviews, 23, 1353–1366
(1970).
“Thermodynamics of elastic stablity.” Proc. 3rd Canadian Congress of Applied Mechanics,
Calgary, May 1971, 29–37.
“An alternative approach to the interaction of local and overall buckling in stiffened panels.
Jointly with M. Pignataro. Proc. IUTAM Symposium on Buckling of Structures, Harvard
University, June 1974, Springer-Verlag (1976), 133–148.
“A basic open problem in the theory of elastic stability.” Joint IUTAM/IMU Symposium on
Applications of Methods of Functional Analysis to Problems of Mechanics. Résumés des
conférences, September 1 to 6, 1075, Marseille. Part XXIX (Marseille, 1975).
“Buckling of a flexible shaft under torque loads transmitted by cardan joints.” Ingenieur-
Archiv, 49, 369–373 (1980).
Index
229
230 Index