You are on page 1of 31

2.

5 Static Bifurcations of Conservative Structures 111

q0>0 q0<0

q0<0 q0>0

q
0

Fig. 2.30 Bifurcation diagram for the perfect rigid rod and for rods with initial imperfections
qo D ˙0:1. The locus of fold bifurcations L is indicated by the dashed-dotted line

from which the equilibrium path is obtained in nondimensional form as

 D f.q; / such that .sin q  sin qo / cot q D g:

The parametric representation of the locus of fold bifurcations, denoted by L in


Fig. 2.30, is obtained by setting k D 0. The equation governing the equilibrium
path  is solved for qo and substituted into k D 0: The obtained equation is L D
f.q; / such that cos3 q   D 0g.
This curve has a special meaning from an engineering design point of view
because the limit points described by the curve represent a limit state beyond which
the structure undergoes a likely catastrophic failure since there are no adjacent stable
equilibrium states beyond it.
The dependence of the limit load on the initial imperfection is obtained as (see
Problem 2.6)
L D cos3 Œarcsin.sin qo /1=3 : (2.95)
The derivative of L with respect to qo is
 q
dL 2=3
D  cos qo 1  sin qo = sin qo1=3 :
dqo
The derivative becomes unbounded at qo D 0 thus implying that small imperfections
can cause a large degradation of the limit load. For this reason, these structures are
said to be strongly sensitive to initial imperfections.
112 2 Stability and Bifurcation of Structures

Fig. 2.31 A rigid rod P


restrained by an elastic spring
at an angle and subject to a A
compressive force P

q
l

α e2
b2
B O e1
b1

2.5.3 Example of Transcritical Bifurcation

The upward infinitely rigid rod shown in Fig. 2.31 is supported by the inclined
elastic spring attached to its free point A where a downward compressive force
is applied. The motion is constrained to take place in the plane fe 1 ; e 2 g. The actual
configuration of the rod is described by the angle q taken as positive in the clockwise
direction. For conservative systems, the equilibrium paths and their bifurcations
can be determined in a straightforward manner through the energy method. For
consistency with the previous treatments, the derivation of the equation of motion is
shown also for this system.
The position vector of a material point of the rod is r.s; t/ D sb2 .t/ where
b2 .t/ D sin q.t/e 1 C cos q.t/e 2 . The velocity and acceleration are given by
rP D s qb
P 1 and rR D s qb
R 1  s qP 2 b2 . The equation of motion consists of the balance of
angular momentum with respect to O written as
Z l
r.l; t/  f C r.l; t/  nO D r  rds
R (2.96)
0

where f D P e 2 is the force,  is the mass per unit length of the rod, and nO D
kL.r.l; t/  r B /=jr.l; t/  r B j is the restoring elastic force of the spring. The
vector r B D l cot ˛e 1 describes the fixed position of the grounded constraint point
B of the spring. The elongation of the spring is
q 
L.q/ D l 1 C cot2 ˛ C 2 sin q cot ˛  1= sin ˛ : (2.97)

The moment of the applied force is r.l; t/  f D P l sin qe 3 while the moment of
the spring tension is
r.l; t/  r B klL.q/ cot ˛ cos q
r.l; t/  nO D kL Dp e3:
jr.l; t/  r B j 1 C cot2 ˛ C 2 sin q cot ˛
2.5 Static Bifurcations of Conservative Structures 113

The equation of motion thus reads


klL.q/ cot ˛ cos q
%Jo qR C p  P l sin q D 0 (2.98)
1 C cot2 ˛ C 2 sin q cot ˛
where %Jo is the mass moment of inertia of the rod with respect to O.
The equation of motion (2.98) can also be obtained via Euler–Lagrange’s
equation with the kinetic energy given by the same expression as in the previous
example, while the potential energy is expressed as
1
V D kL2 .q/  P l.1  cos q/: (2.99)
2

The nondimensionalization
p of (2.98) based on the characteristic time 1=!o with
!o WD kl 2 =%Jo yields
L.q/= l cot ˛ cos q
qR C p   sin q D 0 (2.100)
1 C cot2 ˛ C 2 sin q cot ˛
where  WD P =.kl/ is the load multiplier and L is given by (2.97). There are two
equilibrium paths, namely, the trivial path o D fq D 0; 8g and the nontrivial
bifurcated path given by

L.q/= l cot ˛ cot q


 D f.q; / such that p D g:
1 C cot2 ˛ C 2 sin q cot ˛

The stability of the trivial equilibrium along o is studied through the linearized
variational equation
ı qR C .cos2 ˛  /ıq D 0 (2.101)
from which the vanishing of the tangent stiffness at the divergence bifurcation yields
the critical load multiplier o WD cos2 ˛. The eigenvalues obtained from (2.101) are
( p
˙i j  o j;  < o ;
D p (2.102)
˙   o ;  > o :

Therefore, the equilibrium is marginally stable (center) for  < o (two purely
imaginary eigenvalues); it becomes unstable (saddle) for  > o (one real positive
eigenvalue).
The bifurcated path  is expressed, up to third-order terms, as

 D o C 1
48
cot ˛.7 C 9 cos 4˛  16/q  1
16
cos ˛ 2 .5 cos 4˛ C 3/q 2
CŒcot ˛.540 cos 4˛  1575 cos 8˛ C 1035/=46080q 3 C O.q 4 /:
(2.103)
114 2 Stability and Bifurcation of Structures

q0>0
L

0 q0<0
q0<0

q0>0
q
0

Fig. 2.32 Bifurcation diagram for the perfect rigid rod of Fig. 2.31 and for rods with initial
imperfections qo D ˙0:1 together with the locus of fold bifurcations L

2
For example, for ˛ D =4; the path becomes  D o  3q q
8 C 16  128 q with
3 3

o D 2 : The bifurcated path, which is tangential to the straight line  D o  3q


1
8 ;
is highlighted in Fig. 2.32 by the thicker line. The calculation of the Jacobian along
the nontrivial path with q > 0 shows that k < 0 (unstable) while the part of the path
with q < 0 has k > 0 (stable). Therefore, the divergence bifurcation is manifested
in the form of a supercritical transcritical bifurcation.
Sensitivity to initial imperfections. When the rod is initially rotated by a small
angle qo ; by letting q denote the total angle from the vertical line, according to the
energy method, the potential energy is calculated as

1
V D kL2  P l.cos qo  cos q/ (2.104)
2
where
q q 
L.qI qo / D l 1C cot2 ˛ C 2 sin q cot ˛  1C cot2 ˛ C 2 sin qo cot ˛ :

(2.105)

The equilibrium states are obtained as the stationary points of V according to

Vq D kL.L/q  P l sin q D 0: (2.106)


2.5 Static Bifurcations of Conservative Structures 115

A m

q
k k
e2 P

b q0
e1
a a

Fig. 2.33 The von Mises structure exhibiting limit points at fold bifurcations at which the snap-
through phenomenon takes place

The parametric representation of the locus of fold bifurcations, denoted by L in


Fig. 2.32, is obtained by setting kD0 and substituting into it the equation governing
the equilibrium path  solved for qo so as to eliminate the explicit appearance of qo .

2.5.4 Example of Fold Bifurcation and the Snap-Through


Phenomenon

The simplest example of a fold bifurcation occurs in the von Mises structure,
discussed in Chap. 1. The structure is composed of two identical elastic (massless)
truss bars, mutually hinged at A and at an angle qo with the horizontal (see
Fig. 2.33). This system is a paradigm for initially curved structures such as imperfect
rods, arches, imperfect plates, and shells. The actual configuration is described by
the rotation angle q (taken to be positive in the clockwise direction as in Fig. 2.33).
Let the von Mises truss structure be subject to a downward load P at the hinge
A where two trusses are joined and m be the point mass at A. The tension in
the left truss is nO D kL b where k is the truss equivalent spring constant
and b is the unit vector collinear with the current orientation of the left truss,
b D cos.qo  q/e 1 C sin.qo  q/e 2 (see Fig. 2.33). The elongation in the trusses
is L D aŒsec.qo  q/  sec qo : By virtue of the symmetry of the trusses,
the mass at A can only undergo vertical motion described by the displacement
vector u WD r  r o D aŒtan.qo  q/  tan qo e 2 from which the acceleration is
obtained as uR D a sec2 .qo  q/Œ2 tan.qo  q/qP 2  qe
R 2 : The equation of motion is
thus 2kLb  e 2  P D muR  e 2 which gives

ma sec2 .qo q/Œq2


R tan.qo q/qP 2 C2kaŒsin.qo q/ sec qo  tan.qo  q/  P D 0:
(2.107)
The linearized variational equation about q D 0 is

.ma sec2 qo /ı qR C .2ka tan2 qo /ıq  ıP D 0: (2.108)


116 2 Stability and Bifurcation of Structures

L1
q0=
+
L
E
q
0
-
F L
L2

Fig. 2.34 Equilibrium paths of the von Mises structure with qo D Œ=8; =6; =4. The dashed-
dotted lines L1 and L2 indicate the loci of the limit points

The characteristic time for nondimensionalizing the equation


p is taken as the inverse
of the frequency of small oscillations, !o WD sin qo 2k=m: The nondimensional
linearized variational equation is

ı qR C ıq  ı= tan2 qo D 0 (2.109)

where ı is the nondimensional form of ıP according to the following definition


for the load multiplier:  WD P =.2ka/.
The equilibrium path is obtained as

 D f.q; / such that sin.qo  q/ sec qo  tan.qo  q/   D 0g: (2.110)

The nondimensional tangent stiffness along the equilibrium path is

k D sec2 .qo  q/  sec qo cos.qo  q/: (2.111)

The loci of static bifurcation points (folds) are found by solving k D 0 in


(2.111) and using the equation for the equilibrium path (2.110). A convenient
parametrization of the curve of bifurcation points is found by solving (2.110)
for qo and substituting the result6 in k D 0. Figure 2.34 shows two loci of
bifurcation points at which the bifurcations are represented by limit points. Locally,

6
An explicit parametrization can be obtained by considering the total angle q that the left truss
makes with the horizontal line as Lagrangian coordinate. The relationship between the limit
points and the initial angle qo is cos3 qL D cos qo while the locus of limit loads is expressed
as L D tan3 qL D tan3 Œarcos.cos qo /1=3 . The unstable branch is arcos.cos qo /1=3 < q <
arcos.cos qo /1=3 .
2.5 Static Bifurcations of Conservative Structures 117

the bifurcation at the limit point is a fold: at the first limit point denoted by LC ,
for increasing q the stable state merges with the unstable state and they disappear
through a blue-sky catastrophe; at the second limit point denoted by L , the unstable
state coalesces with the stable state. For better readability, consider in Fig. 2.34 the
equilibrium path with qo D =4. The structure suffers a snap-through instability at
LC that causes the sudden jump of the trusses, indicated by the arrow, to a far-away
equilibrium state E. During the snapping-through phase, the trusses go through the
horizontal unstable equilibrium q D q0 : If the downward load is decreased from E;
the structure encounters the second limit point L where the trusses suffer a reverse
snapping to an upward configuration F with q < 0: The equilibrium path between
the two limit points LC and L is the set of unstable equilibrium states (indicated
by the dashed lines) where the trusses are compressed and inclined to a level such
that the negative geometric stiffness overcomes the elastic stiffness.
The mechanical asymmetry of initially curved structures. A further insight into
this problem can be gained if the Lagrangian coordinate is chosen as the total angle
q that the left truss makes with the horizontal line. This angle is taken as positive in
the counterclockwise direction. The equilibrium path, in this case, turns out to be

 D f.q; / such that sin q.sec qo  sec q/   D 0g:

This equilibrium equation has the virtue of exhibiting the symmetry of the solutions.
If .q; ) is a solution, then .q; / is also a solution. By dropping the inertia forces
and taking into account the nonlinear variational equation with terms up to cubic
order, the following equation is obtained:
 
  1
cos q sec qo  sec3 q ıq  sin q sec qo C sec3 q ı 2 q
2
1    
C cos q 2.cos 2q  2/ sec5 q  sec qo ı 3 q C O ı 4 q  ı D 0: (2.112)
6
This incremental form of the equilibrium shows that, as expected, there are two
antagonistic effects in the restoring force. The projection of the tension of the trusses
nO D ka.sec q o  sec q/b along the vertical direction gives the restoring force
ka sin q.sec q o  sec q/. The restoring force in the truss is always positive (i.e.,
compression) although only the vertical component of this force contributes to the
equilibrium. Therefore, in the incremental form of the equilibrium, the linearized
truss restoring force exhibits two terms: one is always positive (cos q sec qo ıq)
due to the elastic restoring effect, the other ( sec2 qıq) is negative due to the
decrement of the truss angle. The quadratic part of the incremental force given
by  sin q 1=2 sec qo C sec3 q ı 2 q is always negative as far as q > 0; hence
it contributes a softening effect. Therefore, the stiffness suffers a continuous
degradation until vanishing at the limit point. On the other hand, if the force acts
upward, the stiffness increases with the angle q, a situation that signals a hardening
behavior.
118 2 Stability and Bifurcation of Structures

A distinguishing feature of this kind of structures is that they exhibit nontrivial


precritical equilibrium paths before reaching the bifurcation, in the sense that the
precritical equilibrium states are possible because of the nonlinearities. This is
different from the other examined perfect structures where the precritical state is
always the trivial state. Another significant feature is that the limit loads decrease
significantly with the decrease in the initial angle which can be considered as a
measure of shallowness of the structure. Thus, shallow curved structures (e.g.,
shallow arches, shells or imperfect rods) become increasingly more prone to the
snap-through instability as they become shallower.

2.6 The Buckling Problem

Consider a conservative n-dof structure described by the Lagrangian coordinates


q and subject to forces that induce negative geometric stiffness effects. For
conservative systems, there is no need to study the variational equation associated
with the equations of motion. The eigenvalue problem that governs the loss of
stability at the divergence bifurcation (called buckling) is obtained directly from
the equilibrium equations in the adjacent configuration.
Let the forces be parametrized by a load multiplier  and assume that the
structure admits the trivial equilibrium state q D o. Let the linearization of the
equilibrium equations (i.e., obtained as the equilibrium equations of the adjacent
configuration or derived from the stationarity of the potential energy based on a
second-order kinematic formulation) be

K./  q D o: (2.113)

Let the Taylor expansion of K./ about  D 0 be K D KE  KG where KE


(the elastic stiffness) and KG (the geometric stiffness per unit load multiplier) are
positive-definite, symmetric matrices. The critical condition is reached at those load
multipliers, denoted by j ; for which

.KE  j KG /  qj D o: (2.114)

Equation (2.114) is the statement of the buckling eigenvalue problem. A necessary


and sufficient condition for the existence of nontrivial equilibrium states qj is that

det.KE  j KG / D 0 (2.115)

which is the characteristic equation for the buckling problem. Given the positive-
definite, symmetric nature of the matrices KE and KG ; there are n real and positive
eigenvalues, 1 ; 2 ; : : : ; n (usually ordered in increasing order), together with the
associated eigenvectors denoted by u1 ; u2 ; : : : ; un . The lowest eigenvalue, denoted
2.6 The Buckling Problem 119

by o WD 1 ; is called the critical load multiplier and the corresponding eigenvector


uo WD u1 is the critical buckling mode shape.
It is straightforward to show that these eigenvectors satisfy orthogonality condi-
tions with respect to the elastic and geometric stiffness according to

ui  KE  uj D 0; ui  KG  uj D 0; for i ¤ j: (2.116)

Since the eigenvectors are determined within an arbitrary constant, a normalization


condition with respect to the geometric stiffness KG is employed so that

uj  KG  uj D 1; uj  KE  uj D j ; j D 1; : : :; n: (2.117)

It is convenient to introduce the modal matrix U D Œu1 ; u2 : : : ; un  such that

U|  KG  U D I; U|  K E  U D  (2.118)

where I is the identity matrix and  is a diagonal matrix with entries j .


Rayleigh quotient. The eigenvalue equation (2.114), solved for j after putting
qj D uj , yields
uj  KE  uj
j D : (2.119)
uj  KG  uj
The ratio between the numerator and denominator in (2.119) can be recognized to
be V E .uj /=V G .uj /, where V E .uj / is the modal elastic energy and V G .uj / is the
(geometric) potential energy (per unit load multiplier), respectively, given by

1 1
V E .uj / D uj  KE  uj ; V G .uj / D uj  KG  uj : (2.120)
2 2
When the argument u of the energies is not exactly the eigenvector corresponding
to one of the buckling mode shapes, the ratio is called the Rayleigh quotient and is
expressed as
u  KE  u
R.u/ WD : (2.121)
u  KG  u
According to the Rayleigh Theorem, R.u/ is stationary at the eigenvectors and
attains values corresponding to the buckling load multipliers. Moreover, the critical
load multiplier is expressed as

min u  KE  u
o D : (2.122)
u 2 IRn u  KG  u

The Rayleigh quotient is employed to obtain approximations of the buckling loads


and buckling mode shapes. For a given choice of basis vectors fv1 ; v2 ; : : : ; vn g;
the buckling mode shapes are expressed as linear combinations of vj according
120 2 Stability and Bifurcation of Structures

P
to u D j bj vj : This trial vector u is substituted into the Rayleigh quotient and its
stationarity is sought with respect to bj thus obtaining

@R.bi vi /
D 0; j D 1; 2; : : : ; n (2.123)
@bj

where the summation convention on the repeated index is used.


Second-order effects due to prestresses. The structures subject to prestresses
that induce negative geometric stiffness terms exhibit an overall reduced stiffness.
As a consequence, if the prestresses are below their critical values and some
incremental forces are applied on the prestressed structure at some instant, the
ensuing displacements and stresses turn out to be larger than those calculated by
neglecting the prestresses and accounting only for the elastic stiffness of the natural
state.
Consider the example of the rigid rod subject to the compressive force of
Fig. 2.21. If a horizontal force H is applied at the free end of the rod, the equation
governing the adjacent equilibrium becomes .kT  P l/q D H l: Therefore, if
P < Po DW kT = l; then
H l=kT 1
qD D q (1) (2.124)
1  P =Po 1  P =Po
where q (1) WD H l=kT is the elastic rotation calculated according to the first-
order theory by which the second-order (detrimental) effects of the compressive
force P are neglected. The nondimensional number 1=.1  P =Po / is known as the
amplification factor of the elastic solution. Also the elastic moment of the torsional
spring is amplified by the same factor since M D kT q whence
1
M D M (1) ; M (1) WD H l:
1  P =Po
Let the stated problem be generalized by recasting the incremental problem as

.KE  KG /  q D f; (2.125)

where f is the vector of incremental forces. The following transformation is


introduced:
Xn
qD i ui D U  ; (2.126)
i D1
|
where  D Π1 ; : : : ; n  . Substituting (2.126) into (2.125) yields

.j  / j D pj ; pj WD uj  f; (2.127)

where pj is the j th modal component of the incremental force vector. The


coordinates j in (2.127) are the so-called normal coordinates as they allow a full
2.8 Flutter of Wings: Reduced-Order Models 121

uncoupling of the governing equations. The solution is j D pj =.j  / and from


this, the solution is
X n
p
1 j
qD uj : (2.128)
j D1
1  =j j

By setting  D 0; the solution of the first-order theory is obtained as


X
n
pj
q(1) D uj : (2.129)
j D1
j

Equation (2.128) can be rewritten as

X n
p
1 1 1 j 1
qD q(1)   uj q(1) : (2.130)
1  =1 j D1
1  = 1 1  = j  j 1  = 1

Equation (2.130) shows that an estimate (from above) of the actual solution of the
incremental problem can be obtained by multiplying the first-order solution q(1)
by the amplification factor 1=.1  =o / associated with the critical buckling load
o WD 1 .

2.7 Dynamic Bifurcations: Flutter of Lifting Airfoils

Flutter is a dynamic instability that takes place through a Hopf bifurcation in a


variety of structures subject to a flow field or other nonconservative force fields.
In this section, flutter of airfoils subject to uniform flows is discussed. This
phenomenology is typical of lifting surfaces such as aircraft wings, bridges, and
suspended structures.

2.8 Flutter of Wings: Reduced-Order Models

The flutter condition of aircraft wings is often determined by employing a two-


or three-dof model of a wing treated as a thin airfoil (see Fig. 2.35). Here the
mechanical formulation is shown in its major aspects for the three-dof model
although the two-dof linearized model with plunge and pitch is considered for
computations. Let fe 1 ; e 2 g be the fixed frame whose origin is taken to coincide
with the elastic center C E in the stress-free configuration B while fbo1 ; bo2 g are body-
fixed unit vectors collinear with the principal axes of inertia at an angle denoted
by ˛ o . The center of mass and the aerodynamic center are denoted by C and C A ,
respectively (see Fig. 2.35 bottom).
122 2 Stability and Bifurcation of Structures

b2o e2

kh
kp ka b1o V V
.W a
C E
e1 -r r
Vr

ao+a

e2 r E

E A
C W
C C C
O e1

d
W
e d
ab
b b

Fig. 2.35 Lifting surface, the fixed frame fe 1 ; e 2 g, and the wing directors fb1 ; b2 g

The position vector of the elastic center in the actual configuration is r E .t/ D
p.t/e 1 C h.t/e 2 so that p.t/ and h.t/ represent the lagging (or sway) and the
flapping (or plunge) degrees of freedom (see Fig. 2.35). Let bok D R o  e k ;
bk D R.˛/  bok by which bk D .R  R o /  e k . In component form, b1 D
cos.˛ o C ˛/e 1 C sin.˛ o C ˛/e 2 and b2 D  sin.˛ o C ˛/e 1 C cos.˛ o C ˛/e 2 . The
counterclockwise angle ˛ by which the airfoil is rotated denotes the pitching degree
of freedom.
The velocity and acceleration of the material points of the cross section are,
respectively, given by rDP rP C C!  xM and rD
R rR C C ! P  xC!
M  .!  x/ M in which
rP Dpe
C P 2  e !  eM 1 and rR Dpe
P 1 Che C R 2  e !  .!  eM 1 /  e !
R 1 Che P  eM 1 where
eM 1 D R  e 1 , r C is the current position vector of the center of mass C whose
eccentricity with respect to the elastic center C E is denoted by e. On the other hand,
xM is the current position vector of a material point with respect to C . Therefore, the
linear momentum and angular momentum of the airfoil are
Z Z
lD P
rdA D %ArP ; h D C
r  rdA
P D %Ar C  rP C C %J C  !
B B

where %A is the mass of the airfoil, %J C is the mass polar moment of inertia about
the center of mass C , and %J C  ! D %J C ˛e
P 3.
Let the nonlinearly viscoelastic restoring force and couple be denoted by nO
O
and m.
2.8 Flutter of Wings: Reduced-Order Models 123

Third-order expansions of the (horizontal and vertical) spring forces and of the
torsional spring moment are: NO (p) .p; p/ P D kh h C
P NO (h) .h; h/
P D kp p C k3(p) p 3 C cp p,
k3 h C ch h,
(h) 3 P and MO .˛; ˛/
P D k˛ ˛ C k3 ˛ C c˛ ˛.
(˛) 3
P On the other hand, .f A ; c A /
denote the aerodynamic resultant force (i.e., lift and drag resultants) and moment
reduced to the aerodynamic center C A and .f; c/ are the external resultant force and
moment reduced to the center of mass. Therefore, the balance of linear and angular
momentum leads to the following equations of motion for the plane problem:

nO C f A C f D lP (2.131)
O C r E  nO C r A  f A C c A C r C  f C c/  e 3 D hP  e 3 :
.m (2.132)

For the two-dof reduced model, the lagging degree of freedom together with
the drag force are neglected. To obtain the lift and moment induced by a uniform
airstream of velocity V e 1 (with zero initial angle of attack), according to the
theory of thin airfoils of Glauert [179], the effective angle of attack is first
expressed as

˛P 1 hP
˛e D ˛ C a b (2.133)
V 2 V
where a regulates the offset of the elastic center with respect to the mid-chord
position (see Fig. 2.35). To obtain ˛ e D ˛ C ˛r , consider the (linearized) downwash
P e 2 . The airstream velocity relative to the airfoil and
velocity as rP w D .d w ˛P C h)
its angle with respect to the horizontal line (see Fig. 2.35) become, respectively,
V e 1  rP w and ˛r D.˛d P
P w  h/=V . Considering the point C W , where the downwash
velocity is calculated, at three-quarter of the chord from the leading edge gives
d w D .1=2  a/b (see Fig. 2.35).
The lift force and aerodynamic moment, reduced to the elastic center, are given
by7

L WD f A  e 2 D bsV 2 CL ; CL D CLo ˛ e ; CLo WD @˛ CL .˛ e D 0/; (2.134)


 
c E D b 2 sV 2 CM ; CM D CL 12 C a C 2CM˛ (2.135)
where s is the wing span,  is the air density, the aerodynamic center is assumed
at one-quarter of the chord, and CM˛ D 0 for symmetric profiles. The linearized
aerodynamic force and moment thus become
 1  
f L  e 2 D bsV 2 CLo ˛ C ˛=V
P P
2  a b  h=V ;
 1   1 
c E  e 3 D b 2 sV 2 CLo ˛ C ˛=V
P P
2  a b  h=V 2 Ca : (2.136)

7
Note that the lift and aerodynamic moment have slightly different forms (also the reference frame
is chosen differently) when considering the Theory of Theodorsen [427–429].
124 2 Stability and Bifurcation of Structures

Thus the linearized Equations of motion become


" #
˛P 1  hP
%AhR  %Ae ˛R C ch hP C kh h D bsV CL ˛ C
2 o
a b ; (2.137)
V 2 V
" #
  P
˛   P
h
%J E ˛R  %Ae hR C c˛ ˛P C k˛ ˛ D b 2 sV 2 CLo 12 C a ˛ C 1
a b
V 2 V
(2.138)

where %J E WD %J C C %Ae 2 is the polar mass moment of inertia with respect to the
elastic center C E .
By dividing the vertical coordinate
p h by b (i.e., hN WD h=b) and introducing the
characteristic time 1=!˛ (!˛ WD k˛ =%J E is the frequency of the pitch mode), the
following nondimensional form of the equations of motion is obtained:
" #
˛P  1  hP
R P
h  "˛R C 2 h ! h C ! h D kU ˛ C
2 2
a  ; (2.139)
U 2 U
" #
1  ˛P 1  hP
r ˛R  "hR C 2r ˛ ˛P C r ˛ D kU 2 C a ˛ C
2 2 2 2
a  (2.140)
U 2 U

where the bar on h is dropped and the dot indicates differentiation with respect
to nondimensional timeptN WD t!˛ . The following nondimensional parameters
are introduced: r WD %J E =.%Ab 2 /,p" WD e=b, h WD ch =.2%A!h /, ˛ WD
c˛ =.2J E !˛ /, ! WD !h =!˛ ; !h D kh =%A. The nondimensional velocity is
U WD V =.b!˛ / and k WD sb 2 CLo =%A is the aerodynamic constant.
Flutter condition. The two equations of motion (2.139) and (2.140) can be written
in compact form by letting q D Œh; ˛| thus obtaining

M  qR C .C  kU CA /  qP C .K  kU 2 HA /  q D o (2.141)

where
2 3
  1 1
a  
1 " 2 0 1
M WD ; C WD 4 
A
 5; H WD
A
:
" r 2  12 C a 1
 a2 0 1
2
Ca
4
(2.142)
The stiffness and damping matrices are diagonal with entries f! 2 ; r 2 g and
f2 h !; 2r 2 ˛ g, respectively. The matrices CA and HA are nonsymmetric. In
particular, HA can be decomposed into its symmetric and skew-symmetric parts as
 1
  
0 0 12
Hsym WD 1 1
A 2 ; Hskw WD
A
:
2 2 Ca  12 0
2.8 Flutter of Wings: Reduced-Order Models 125

The symmetric part HAsym contributes to the stiffness of the airfoil. On the other hand,
kU CA  qP and kU 2 HAskw  q embody the nonconservative terms.
The system is cast in first-order form as

xP D A  x (2.143)
P ˛
where x D Œh; ˛; h; P | and
 
O I
AD : (2.144)
M1  .K  kU 2 HA / M1  .C  kU CA /

The eigenvalues are the roots of the characteristic equation

detŒ2 M C .C  kU CA / C .K  kU 2 HA / D 0 (2.145)

which is written explicitly as

a0 4 C a1 3 C a2 2 C a3  C a4 D 0: (2.146)

According to the Routh–Hurwitz criterion, a necessary and sufficient condition


for all the eigenvalues to have negative real parts is that all minors i .i D 1; : : : ; 4/,
along the principal diagonal of the Hurwitz matrix be positive.
If one or more minors vanish, the system is in a critical condition. Hence, to
find the flutter condition, each individual i is forced to vanish. The coefficients ai
.i D 0; : : : ; 4/ in (2.146) generate the Hurwitz matrix
2 3
a1 a0 0 0
6 a3 a2 a1 a0 7
HD6
40
7 (2.147)
a4 a3 a2 5
0 0 0 a4

from which the minors are 1 D a1 and


2 3
  a1 a0 0
a1 a0
2 D det ; 3 D det 4 a3 a2 a1 5 ; 4 D det H D a4 3 : (2.148)
a3 a2
0 a4 a3

The lowest positive value U D Uo corresponding to which one of the minors


vanishes is the critical flutter velocity.
Example 2.3 (Computation of the flutter speed of a wing). Calculate the critical
flutter speed of the wing model employed in [222], whose parameters are:  D
1:225 kg/m3 , a D 0:4, b D 0:135, " D Œ0:0873  b.1 C a/=b, s D 1 m, kh D
2844:4 N/m, k˛ D 99:4 Nm, ch D 27:43 Ns/m, c˛ D 0:036 Nsm, m D 12:387 kg,
%J E D 0:065 kgm2, CLo D 6:28.
126 2 Stability and Bifurcation of Structures

a
0.8

0.4

0
-0.06 -0.04 -0.02 0.02
-0.4

-0.8

1.2
0.01
b c
0 0.8

-0.01 0.4

-0.02 0

-0.03 -0.4

-0.04 -0.8

-0.05 -1.2
2 2.4 2.8 3.2 2 2.4 2.8 3.2

Fig. 2.36 (a) Loci of the eigenvalues of the two-dof wing model in the complex plane, (b) real
parts of the eigenvalues, and (c) imaginary parts of the eigenvalues vs. the nondimensional flow
speed U

The corresponding nondimensional parameters are: ! D 0:3875, " D 0:0467,


r D 0:5366, h D 0:0731, ˛ D 0:0071, and k D 0:0113. By employing the
governing equations (2.139) and (2.140), the Routh–Hurwitz criterion or the direct
p gives Uo D 2:47. The dimensional flutter speed is
calculation of the eigenvalues
V D U b!˛ . Since !˛ D k˛ =%J E D 39:101=s, the dimensional flutter speed
is V D 13:04 m/s. The loci of the eigenvalues in the complex plane and the
variations of the real and imaginary parts with the air speed are shown in Fig. 2.36.
In particular, Fig. 2.36a shows the transversal crossing of the imaginary axis which
signals the bifurcation. The substantial insensitivity of the frequencies of the plunge
and pitch modes is for this example demonstrated in Fig. 2.36c.
Flutter of a two-dof system using the flutter derivatives. Two-dof systems
undergoing heave and pitch motions due to uniform airstreams can well describe
sectional models of lifting surfaces such as bridges or aircraft wings (see Fig. 2.37).
Consider a two-dof lifting surface of width b, subject to a steady uniform flow
of velocity V . The body can experience heave (vertical motion) described by h
and pitch (torsional motion) described by the rotation angle ˛ (see Fig. 2.37). The
balance of linear and angular momentum of the airfoil gives the following equations
of motion:
1  hP ˛P h
mhR C ch hP C kh h D bV 2 KH1 C KbH2 C K 2 H3 ˛ C K 2 H4 ;
2 V V b
(2.149)
2.8 Flutter of Wings: Reduced-Order Models 127

Fig. 2.37 Two-dof model of e2


a bluff body subject to a
uniform airstream.
b2
kh
b1 V
kα α e1
b

1 2 2 hP ˛P h
%J ˛R C c˛ ˛P C k˛ ˛ D b V KA1 C KbA2 C K 2 A3 ˛ C K 2 A4
2 V V b
(2.150)

where the overdot indicates differentiation with respect to the dimensional time
t; m is the mass, %J is the mass moment of inertia with respect to the elastic
center C E (here it is taken to coincide with the mass center); .kh ; k˛ / are the spring
constants of the vertical and torsional springs, respectively; .ch ; c˛ / are the damping
coefficients for the heave and pitch motions;  is the air density; K WD b! o =V is the
reduced oscillation (circular) frequency (K is related to the so-called reduced flow
velocity Ur :DV =.b fo /=2 =K where !o =2fo is the airfoil oscillation frequency);
and .Hj ; Aj / .j D 1; : : : ; 4/ are the flutter derivatives (also called aeroelastic
derivatives).
The right-hand sides of (2.149) and (2.150) are the lift force and aerodynamic
moment expressed in terms of the flutter derivatives .Hj ; Aj / .j D 1; : : : ; 4/. The
flutter derivatives are identified through wind tunnel tests as functions of K or Ur .
Notice that the velocity-dependent terms in the lift and moment give rise to the
aerodynamic damping matrix CA , while the displacement-dependent terms give rise
to the nonsymmetric matrix HA . Thus the equations of motion can be written as
   
M  qR C C  12 bV 2 KCA  qP C K  12 bV 2 K 2 HA  q D o: (2.151)

The mass, stiffness, and damping matrices are diagonal, M D fm; %J g, K D


fkh ; k˛ g, and C D fch ; c˛ g. On the other hand,
   
1 H1 bH2 H4 =b H3
CA WD ; HA WD :
V bA1 b 2 A2 A4 bA3
128 2 Stability and Bifurcation of Structures

The decomposition of HA into its symmetric and skew-symmetric parts leads to


" #
H4 1
.H3 C A4 /
Hsym WD 1 
A b 2 ;
2
.H3 C A4 / bA3
 
0 1
.H3  A4 /
Hskw WD
A 2 :
 12 .H3  A4 / 0

While the symmetric part HAsym contributes to the stiffness of the airfoil, the terms
proportional to CA  qP and HAskw  q represent the nonconservative forces.
To make the equations nondimensional, the p heave motion is scaled by b and
time by the characteristic time 1=!h (!h WD kh =m is the heave frequency). The
following nondimensional variables and parameters are introduced:
r
%J !˛ b 2
rWD ; N
! ˛ WD ; N
 WD ;
mb 2 !h m
b!o c˛ ch
KWD ; ˛ WD 2
; h WD ;
V 2!˛ b m 2!h m

where N is the relative density of air/fluid with respect to the structure; the
nondimensional oscillation frequency is !N o WD !o =!h .
The nondimensional form of the equations of motion becomes

1
hRN C 2 h hPN C hN D N!N o2 .H1 =!N o hPN C H2 =!N o ˛P C H3 ˛ C H4 h/; N (2.152)
2
1
r 2 ˛R C 2 ˛ !N ˛ ˛P C r 2 !N ˛2 ˛ D N!N o2 .A1 =!N o hPN C A2 =!N o ˛P C A3 ˛ C A4 h/:
N (2.153)
2
p
The critical flutter mode is sought as .h; N ˛/ D .u1 ; u2 /ei!N o t (where i WD 1 is
the imaginary unit) whose substitution into (2.152) and (2.153) yields

1 2  
.!N o2 C 2i!N o h C 1/u1 D N!N o iH1 u1 C iH2 u2 C H3 u2 C H4 u1 ;
2
1  
.r 2 !N o2 C 2i ˛ !N ˛ !N o C r 2 !N ˛2 /u2 D N!N o2 iA1 u1 C iA2 u2 C A3 u2 C A4 u1 :
2
The set of equations governing the eigenvalue problem is rewritten in compact
form as A.!N o /  u D o with u D Œu1 ; u2 | and

1  
A11 D .!N o2 C 2i!N o h C 1/  N!N o2 iH1 C H4 ; (2.154)
2
1   1  
A12 D  N!N o2 iH2 C H3 ; A21 D  N!N o2 iA1 C A4 ; (2.155)
2 2
2.8 Flutter of Wings: Reduced-Order Models 129

1  
A22 D .r 2 !N o2 C 2i ˛ !N ˛ !N o C r 2 !N ˛2 /  N!N o2 iA2 C A3 : (2.156)
2

A real-valued form of the eigenvalue problem can be obtained by letting u1 .t/ D


y1 cos !N o t C y2 sin !N o t and u2 .t/ D y3 cos !N o t C y4 sin !N o t: Substituting this
transformation into the equations of motion, setting the coefficients of cos !N o t and
sin !N o t to zero, and letting y D Œy1 ; y2 ; y3 ; y4 | yield the following system of real-
valued equations: A.!N o /  y D o where

1 1 1
A11 D 1  !N o2  H N 4 !N o2 ; A12 D 2 !N o  HN  !N 2 ; A13 D 2 !N o  H N  !N 2 ;
2 2 1 o 2 3 o
1 1 1
A14 D  H N 2 !N o2 A21 D 2 h !N o C H1 N!N o2 ; A22 D 1  !N o2  H N  !N 2 ;
2 2 2 4 o
1 1 1  2 1  2
A23 D H N 2 !N o2 ; A24 D  H N  !N 2 ; A31 D  A N !N ; A32 D  A N !N ;
2 2 3 o 2 4 o 2 1 o
1  2 1  2 1  2
A33 D r 2 !N ˛2  r 2 !N o2  A N !N ; A34 D 2 ˛ !N ˛ !N o  A N !N ; A41 D A N !N ;
2 3 o 2 2 o 2 1 o
1  2 1  2 1  2
A42 D  A N !N ; A43 D 2 ˛ !N ˛ !N o C A N !N ; A44 D r 2 !N ˛2  r 2 !N o2  A
N !N :
2 4 o 2 2 o 2 3 o
The flutter speed is calculated as the lowest real root of the characteristic equation
detŒA.!N o / D 0: To this end, an iterative procedure is employed as follows [406]:
1. A tentative value (initial guess) of K is chosen.
2. The values of the experimentally obtained coefficients Hi and Ai are extracted
for the guessed value of K.
3. The characteristic equation is solved and the complex-valued roots are
determined. In general, the imaginary part of !N o is different from zero for
all roots.
4. The procedure is iterated spanning a suitable range of K until the condition
Im.!N o / D 0 is satisfied.
The procedure described seeks only the critical flutter condition. It can be
modified to give information about the actual behavior of the eigenvalues of (2.152)
and (2.153) in the vicinity of flutter if they are set in the general form k D ˛k Ciˇk
with ˛k WD Re.k / and ˇk WD Im.k /. Within the spectrum of frequencies of the
modes of the unforced structure, !o is set, in its initial guess, to the frequency of the
mode that is expected to undergo the Hopf bifurcation, say the mth mode (often this
mode is the torsional mode). This is the initialization. Then the procedure unfolds
as follows:
1. An initial value of the flow velocity V is chosen so that the corresponding K D
b!o =V or reduced speed Ur D V =.bfo / are calculated.
2. The values of the experimentally obtained coefficients Hi and Ai are extracted
for the given value of K or Ur .
130 2 Stability and Bifurcation of Structures

3. The characteristic equation is solved and the complex-valued eigenvalues k D


˛k C iˇk are determined among which the eigenvalue of interest is m D
˛m C iˇm . In general, ˇm ¤ !o . Hence, !o is corrected and the steps 1–3 of
the procedure are iterated until achieving convergence, i.e., until the condition
jˇm  !o j < " is met with " denoting the tolerance. If the velocity V of step
1 does not correspond to the critical value, the real part ˛m of the eigenvalue is
different from zero.
4. The velocity is updated and the procedure is iterated assuming the imaginary part
ˇm of the critical mode found in the previous step as the oscillation frequency !o .
5. The critical velocity is determined when ˛m < "N (i.e., the real part of the
eigenvalue becomes sufficiently small), where "N is the numerical tolerance within
which the critical condition is determined.
The behavior of the loci of the real and imaginary parts of the eigenvalues in
the range of velocities signals the bifurcation when a pair of complex-conjugate
eigenvalues intersects transversally the imaginary axis (whereby ˛m D 0). The
advantage of this procedure, notwithstanding its higher computational costs, is that
it carries information both on the (damped) oscillation frequency and damping of
the critical mode of the airfoil subject to self-excited aeroelastic forces below and
above flutter.

2.9 Dynamic Instabilities Due to Parametric Resonances

Nonautonomous systems subject to time-varying excitations are said to be para-


metrically excited when the excitation input (force, displacement, or other driving
inputs) is such that it gets multiplied by the motion of the system by various physical
mechanisms. It is also said to be a multiplicative excitation as opposed to a direct
excitation. The latter can be, for example, a force directly applied to the system
mass. The work done by a parametric excitation depends on the motion. This is
the case for a pendulum whose pivot point is excited in the vertical direction by
a prescribed displacement denoted by y.t/ (see Fig. 2.38a). The apparent (inertia-
induced) force myR (m is the point-wise mass of the pendulum whose arm is assumed
to be massless) induces the couple myl R sin  given that l sin  is the lever arm of
the force if  denotes the angle that the pendulum arm makes with a vertical line.
This demonstrates the multiplicative nature of the parametric excitation. Besides
the parametric excitation couple, the pendulum is subject to the restoring couple
induced by gravity given by mgl sin .
The trivial downward vertical configuration is a stable equilibrium state of
the parametrically excited system provided that the ratio between the excitation
frequency of the pivot and the pendulum natural frequency is away from certain
critical values. When the frequency of the prescribed motion is in a suitable ratio
with the natural frequency of the pendulum, the trivial equilibrium is no longer
stable and the pendulum may transition into other kinds of motion. The instability
2.9 Dynamic Instabilities Due to Parametric Resonances 131

a b c
y(t)
y(t)
m1 y(t) O1
l1 l1C
l
l m2
1

B l2C
O2
c l2
mg
2

Fig. 2.38 (a) The parametrically excited pendulum, (b) the autoparametric vibration absorber, (c)
the double pendulum

often appears as an abrupt and violent burst of sustained large-amplitude forced


pendular oscillations. An important feature of the phenomenology is that the ensuing
parametrically excited motion is orthogonal, in a broad sense, to the direction of the
excitation.
For small-amplitude oscillations (i.e., sin  ), the equation of motion reduces
to Hill’s equation [401],

R C Œ! 2 C y.t/=
R l D 0 (2.157)
p
where ! WD g= l is the natural frequency of the pendulum. If yR is harmonic (i.e.,
y D Y cos t), this equation is referred to as Mathieu’s equation [70, 217, 332]. If
the dissipative effects are neglected, the trivial equilibrium undergoes instabilities
whenever the excitation frequency is twice the natural frequency divided by an
integer, namely,  D m2 !, m D 1; 2; : : :
To prove this result, let time be rescaled by the characteristic time 2= by which
Mathieu’s equation is rewritten as

R C .ı C 2 cos 2t  / D 0 (2.158)

where ı WD .2!=/2 and  WD 2Y = l: The nondimensional excitation amplitude 


is a small number (i.e.,   1/. The natural frequency of the rescaled system is
p
!  WD ı D 2!=: (2.159)
P
A straightforward perturbation based on  D N k
kD1  k shows that the governing
R 
equation at mth order is m C ım D 2 cos 2t m1 . If the generating solution
 
is 1 / ei! t (i is the imaginary unit), the forcing term of the .m C 1/th order
 
system is / ei .2m! /t : This forcing function can cause an unbounded growth of
the motion if the forcing frequency .2m  !  / is equal to the natural frequency of
132 2 Stability and Bifurcation of Structures

the pendulum !  : This occurs if and only if !  D m, hence ı D m2 . The definition


of !  according to (2.159) implies that the forcing frequency must be  D m2 !,
m D 1; 2; : : :
If the excitation frequency is not exactly twice the natural frequency divided by
an integer, the instability can be triggered if the excitation amplitude is sufficiently
high. This means that in the plane .ı; /, there are instability regions which emanate
from the critical frequencies ı D m2 and become wider with increasing . In
the physical .; Y /-plane, the corresponding critical dimensional frequencies that
emanate from the instability regions are  D m2 !, m D 1; 2; : : : The first instability
region which emanates from ı D 1 for m D 1 (i.e.,  D 2!) is the principal
parametric resonance. The second instability region corresponds to ı D 4 for m D
2 ( D !) and is referred to as parametric resonance. The other regions correspond
to higher-order resonances, such as ı D 9; 16; 25; : : : (i.e.,  D 23 !; 12 !; 25 !; : : :/
and thus accumulate on the origin of the frequency axis at  D 0 as m goes to 1.
This accumulation of the critical frequencies on the origin of the physical frequency
axis shows that the time rescaling is very useful to study these instabilities since it
shifts the accumulation toward 1. The collection of instability regions is called a
Strutt–Ince diagram while the curves along which the instability occurs are called
transition curves. Each instability region is called a Mathieu tongue.
At amplitude levels beyond the instability, further bifurcations can lead to
more complex quasi-periodic or chaotic motions of the pendulum as observed in
experimental or numerical studies [205, 395].
Parametric instabilities arise in virtually all physical systems governed by
ordinary differential equations having coefficients (almost) periodic in time. They
also arise in distributed-parameter systems governed by partial differential equa-
tions, but there is no simple mathematical characterization of those equations that
admit parametric instabilities (cf., e.g., [260]). The essential physical phenomenon
of a parametric instability is that a small parametric excitation can produce a
large response when the driving frequency is close to twice one of the natural
frequencies of the system divided by an integer. However, the distinguishing feature
of parametrically excited linear systems is that the amplitude of the nontrivial
solution grows exponentially unbounded even when there is viscous dissipation,
whereas in “directly excited” linear systems, the resonance may be bounded as a
consequence of the dissipation. Geometric and material nonlinearities can act to
limit the motions caused by parametric instabilities because the frequency of the
motion, varying with the amplitude, can be shifted out of resonance. The large
growth of parametrically excited motions is a powerful natural amplifier, which
can be exploited in fields such as microengineering, structural health monitoring,
vibration suppression, or quenching of self-excited vibrations [396, 442].
In multi-body mechanical systems or in distributed-parameter systems, one mode
of vibration can effectively act as the parametric excitation of another mode through
the presence of multiplicative nonlinearities. Thus autoparametric resonance occurs
when the frequency of one of the modes is a multiple of half of the other. Other
effects include combination resonances, where the excitation is a sum or difference
2.9 Dynamic Instabilities Due to Parametric Resonances 133

of two modal frequencies. Studies include the motion of water waves in vertically
forced containers [153], of longitudinally forced strings [314, 411, 418, 419], of
transversally forced membranes [315], of longitudinally forced columns and plates
[62, 103, 205, 219, 412, 473, 481], of base-excited cantilever rods [16, 470], and of
plates and shells [24, 144, 340, 359, 449]. Studies in other areas of physics feature
the propagation of electromagnetic waves in media with a periodic structure and the
motion of electrons in a crystal lattice. The parametric resonance phenomenon also
arises in the study of the stability of (almost) periodic motions [292, 311, 332, 476].
Furthermore, as mentioned, the parametric resonance instability can also arise in the
presence of modal couplings of various kinds such as combination resonances of the
sum or difference type, two-to-one, three-to-one, and one-to-one internal resonances
[142, 332, 468].
Besides Faraday waves in vertically forced containers, the more general problem
of linear and nonlinear interaction of liquid sloshing dynamics with elastic con-
tainers and supported structures, which is of great concern to aerospace, civil, and
nuclear engineers, is treated in [202]. The problem of stability of parametrically
excited motions becomes much wider when the excitations exhibit stochasticity.
A good reference for an account of stochastic stability theorems and analytical
techniques for determining the random response of nonlinear systems is [203].
A parallel field of engineering interest is that of stabilization of the undesirable
parametric resonances in discrete and continuous systems. To mention but a few,
control of the principal parametric resonance of beams was sought by means of
various active/passive control strategies [247,472,475] or by finding optimal shapes
of the rods through a suitable optimization problem [293]. An active parametric
resonance cancellation method for magnetically levitated bodies was proposed in
theory and experiments in [474].
Methods to calculate the onset of parametric instabilities. A variety of methods
have been proposed to construct the instability regions of systems governed by
linear differential equations with time-periodic coefficients. The methods can be
grouped according to two fundamental approaches. The first approach involves
the determination of the characteristic exponents and generally leads to very
complicated mathematical analyses. The second method finds directly the boundary
surfaces or curves between the regions of stability and instability determining the
critical conditions under which periodic solutions may exist. In the analysis of
second-order equations, the method of variation of the parameters was combined
with a series expansion typical of a perturbation method in [420]. The method
has the advantage of providing information about the behavior of the nontrivial
solutions in both stable and unstable cases. This method was generalized to systems
of second-order differential equations in [198]. In [415] this approach was further
exploited in the presence of eigenvalues with negative real parts. A number of
variants of the second approach have been devised to treat parametrically excited
systems according to different perturbational procedures [332, 376, 389].
The expansion of the monodromy matrix in terms of the system parameters was
proposed as a method for computing the boundaries between the stable and unstable
134 2 Stability and Bifurcation of Structures

regions of Hill’s equation with damping [401]. The eigenvalues of the matrix (i.e.,
the Floquet multipliers) are forced to take the values that they should have on
the associated transition curves, thus obtaining a parametric representation of the
transition curves together with the instability regions.

2.10 Parametric Resonances of Conservative Systems


with Linear Damping

The case of parametrically excited mechanical systems is treated first. In nondimen-


sional form, the equations of motion, expanded in Taylor series up to third order,
can be cast as
M  qR C C  qP C .K C P .t/B/  q D n2 .q; q/ C i(1) P q/
2 .q; P C i(2) R
2 .q; q/

Ci(1) P q;
3 .q; P q/ C i(2) R C n3 .q; q; q/
3 .q; q; q/

(2.160)
where the overdot indicates differentiation with respect to the nondimensional time
tI q.t/ is the vector of generalized coordinates; M is the positive-definite, symmetric
inertia matrix; K is the positive-definite, symmetric stiffness matrix; C is the linear
damping operator; n2 and n3 are quadratic and cubic elastic and geometric stiffness
2 ; i2 / and .i3 ; i3 / are quadratic and cubic inertia operators; P .t/B  q
operators; .i(1) (2) (1) (2)

collects the parametric excitation terms. The adopted form of inertia nonlinearities
is typical of various mechanical systems.
In general, the nonlinear operators do not commute, i.e. n2 .u; v/ ¤ n2 .v; u/:
Because the linear, unforced, undamped, conservative problem is symmetric and
positive-definite, the eigenvectors um obtained from the eigenvalue problem

.K  ! 2 M/  u D o (2.161)

are mutually orthogonal and are normalized according to

um  M  un D ımn ; um  K  un D !n2 ımn (2.162)

where ımn is the Kronecker delta.


The method of multiple scales is employed to unfold both the onset of the para-
metric resonance (i.e., critical condition) and the postcritical response. The method
of multiple scales is illustrated by means of an extraordinary number of examples
featuring various external and internal resonance conditions in [220,328,332]. Here,
uniform asymptotic expansions of the postcritical periodic motion are sought by
introducing the following time scales:

"2
t0 WD t; t1 WD "t; t2 WD t (2.163)

2.10 Parametric Resonances of Conservative Systems with Linear Damping 135

where " is a small dimensionless number. The time scale t0 accounts for the rapidly
varying part of the motion (occurring at the natural frequencies) and the stretched
time scales t1 and t2 account for the slowly varying part of the motion, namely, the
slowly modulated envelope and phase of the motion. Two time scales (i.e., t0 and t1 )
are sufficient if only the critical condition is to be determined. Provided that the data
are sufficiently differentiable, asymptotic expansions of the solutions are sought in
the form
X
3
"k
q.t; "/ qk .t0 ; t1 ; t2 / (2.164)

kD1

where the functions qk .t0 ; t1 ; t2 / are to be determined. The following notational


convention is adopted for simplicity:

@ @2 @ @2
@k WD ; @k @l WD ; so that @0 WD ; @0 @2 WD ; etc.
@tk @tk @tl @t0 @t0 @t2

In accord with the method of multiple scales, the following expansions are
substituted into the equation of motion (2.160):
" 3 #
X3
"k "2 X "k
q.t; "/ P "/ Œ@0 C "@1 C @2 
qk .t0 ; t1 ; t2 / ; q.t; qk .t0 ; t1 ; t2 / ;
kŠ 2Š kŠ
kD1 kD1
" 3 #
X " k
R "/ Œ@20 C 2"@0 @1 C "2 .@0 @2 C @21 /
q.t; qk .t0 ; t1 ; t2 / : (2.165)

kD1

The forcing function is assumed to be sinusoidal: P .t/DP cos t. In seeking


periodic solutions to the nonlinear problem, it is essential to allow the frequency 
to depend on the small parameter ":

."/ D 0 C " C O."2 /; 0 D 2!k (2.166)

where 0 is taken to be twice the natural frequency of the kth mode at the critical
condition. The damping matrix C and the parametric excitation function P .t/ are
ordered as " C and "P .t/, respectively.
Substituting (2.165) into (2.160), using the independence of the time scales, and
equating coefficients of like powers of " yield the following hierarchy of linear
problems.
Order ":
M  @20 q1 C K  q1 D o: (2.167)
136 2 Stability and Bifurcation of Structures

Order "2 :

M  @20 q2 C K  q2 D 2M  @0 @1 q1  C  @0 q1  P cos t0 B  q1


Cn2 .q1 ; q1 /Ci(1) (2) 2
2 .@0 q1 ; @0 q1 /Ci2 .q1 ; @0 q1 /: (2.168)

Because the kth mode is activated by the parametric instability and no internal
resonances engage this mode with other modes, the generating solution at order
" is assumed to be

 
q1 D Ak .t1 /ei!k t0 C ANk .t1 /ei!k t0 uk (2.169)

where Ak .t1 / is the complex-valued amplitude of the motion and the overbar indi-
cates the complex conjugate. Substituting (2.169) into (2.168) yields the following
inhomogeneous problem at second order:

M  @20 q2 C K  q2 D i!k ei!k t0 Œ2@1 Ak M C Ak C  uk C hC N


k Ak Ak

2 2i!k t0 1
 
Ch
k Ak e  P Ak ei.C!k /t0 CANk ei.!k /t0 B  uk Ccc
2
(2.170)

where cc stands for the complex conjugate of the preceding terms and

h
k D n2 .uk ; uk /  !k Œi2 .uk ; uk / C i2 .uk ; uk /;
2 (1) (2)

hC
k D n2 .uk ; uk / C !k Œi2 .uk ; uk /  i2 .uk ; uk /:
2 (1) (2)
(2.171)

The right-hand side of (2.170) contains the following terms that can cause an
unbounded growth in time of the solution at order  2 :
1
i!k ei!k t0 Œ2@1 Ak M C Ak C  uk  P ANk ei.!k /t0 B  uk : (2.172)
2
A solvability condition, according to the Fredholm Alternative Theorem, is imposed
by multiplying the resonant terms (2.172) in the right-hand side of (2.170) by the
solution of the linear adjoint problem, uk exp.i!k t0 /. The result is the following
complex-valued modulation equation for the amplitude Ak :

2i!k .@1 Ak C k Ak / D Pk ANk ei t1 (2.173)

where 2 k WD uk  C  uk is the definition of the modal damping coefficient and

1
k WD uk  B  uk (2.174)
2

is called the effective parametric resonance coefficient of the kth mode.


2.10 Parametric Resonances of Conservative Systems with Linear Damping 137

To transform (2.173) into real form, let

1 k  t1
Ak D ak ei 2 ei 2 (2.175)
2
and substitute it into (2.173) thus obtaining

Pk
@1 ak D  k ak  ak sin k ;
2!k
(2.176)
Pk
@1 k D   cos k :
!k

The critical solutions on the transition curves correspond to the steady-state


solution of (2.176), namely, @1 ak D 0 D @1 k . Solving for sin k and cos k ,
summing the squares, and employing the fundamental trigonometric identity yield
q
 D ˙ P 2 2k =!k2  .2 k /2 (2.177)

from which the transition curves associated with the principal parametric resonance
are obtained as
q
 2!k ˙ P 2 2k =!k2  .2 k /2 : (2.178)
The parametric resonance is activated if the excitation amplitude is above a
threshold value called the critical force amplitude expressed as

Po D 2 k !k =k (2.179)

where k is given by (2.174).


To obtain the postcritical response, the solution of the second perturbation is
substituted into the third perturbation for which a solvability condition is enforced.
This condition captures the effects of the nonlinear terms due to the elastic,
geometric, and inertial forces. By employing the method of reconstitution [332] so
as to express the time derivative of the critical amplitude AP D "@1 A C "2Š @2 A C    ;
2

the bifurcation equation reads:

2i !k .APk C k Ak / D Pk ANk ei t C  S Ak C k A2k ANk (2.180)

where k is the effective nonlinearity coefficient of the kth mode and  S is a linear
frequency shift. The effective nonlinearity coefficient governs the bending of the
backbone of the kth mode8.

8
The backbone is the curve expressing the relationship between the frequency and the oscillation
amplitude of the unforced, undamped problem. The frequency reduces to that of the kth mode as
the amplitude goes to zero.
138 2 Stability and Bifurcation of Structures

2.10.1 Multi-pendulum Systems and the Autoparametric


Transfer of Energy

The dynamics of multi-pendulum systems have been investigated in depth both from
a theoretical and from an experimental point of view. Some features of the chaotic
dynamics of the single pendulum were discussed in [390, 409], while the forced
double pendulum and the triple pendulum with impacts were addressed in [410]. For
small forcing amplitudes, there are many theoretical studies dealing with parametric
resonances in the planar double pendulum [143, 318, 407]. In particular, the method
of multiple scales was employed in [143, 392] to study the principal parametric
resonance of the in-phase and out-of-phase modes of a double pendulum.
Experimental studies of double pendulum systems have considered several
different geometric configurations as well as forcing conditions. One such forcing
condition is the high-frequency excitation that can stabilize the upright unstable
position. The effects of resonant high-frequency excitation on the linear stability
and nonlinear behavior of the pendulum were investigated in [212] by using the
method of direct partition of motion due to Blekhman [70]. It was shown that the
support excitation has a stabilizing effect for most system parameters but can also
destabilize the upright pendulum position: supercritical bifurcations may turn into
subcritical bifurcations and chaotic behaviors of the pendulum exist for a wide range
of system parameters and initial conditions. In a similar system, the normal form and
bifurcation theory were used in [479] to find closed-form solutions for equilibria,
periodic, and quasi-periodic motions.
Existence, bifurcations, and stability of high-frequency periodic motions were
studied again in a double pendulum [227]. The linear stability analysis of the four
equilibria was carried out for generic geometries, although when the two arms were
identical, the stability problem could be studied in the full nonlinear setting. For
the case of vertical base excitations at an arbitrary frequency and amplitude, a
rigorous stability analysis of the equilibria of the double pendulum was carried out
in [47]. Along the same lines, a large body of works have addressed high-frequency
parametric excitations [70, 217].
The effects of follower forces in inverted double pendula (with rotational springs
and dashpots between the arms), subject to base excitations, have also been studied
in depth.
Double pendulum. The equations of motion of a double pendulum, subject to
a vertical base motion y.t/ (see Fig. 2.38c), are given by the following two
autonomous ordinary differential equations with time-varying coefficients [392]:

J1 R1 C I1 .g C y/
R sin 1 C c1 P1
CI2o l1 R2 cos.1  2 / C I2o l1 sin.1  2 /P22 D 0; (2.181)
J2o R2 C I2o R sin 2 C c2 P2
.g C y/
CI2o l1 R1 cos.1  2 /  I2o l1 sin.1  2 /P12 D 0 (2.182)
2.10 Parametric Resonances of Conservative Systems with Linear Damping 139

where 1 and 2 are the absolute angles (taken to be positive if counterclockwise)


of the upper and lower arms, respectively, with respect to a fixed vertical line;
the overdot indicates differentiation with respect to time. The governing physical
parameters are expressed as: J1 D J1o C m2 l12 ; Iio D mi lic ; I1 D I1o C l1 m2 ;
where mi , li ; lic , and Jio .i D 1; 2/ denote the mass, the arm length, the distance
of the center of mass of the i th arm from Oi , and the moment of inertia about
Oi . .Ik ; Jk / represent first- and second-order mass moments of inertia. Equations
of motion (2.181) and (2.182) hold for a double pendulum with nonuniform mass
properties of the arms as is the case in various engineering applications.
p is nondimensionalized by rescaling it by the characteristic time 1=!c with
Time
!c WD g= l , lWDl1 C l2 ; g is the acceleration due to gravity so that !c has the
meaning of circular frequency of a simple pendulum whose length is equal to the
sum of the lengths of the two arms. The following nondimensional parameters
are introduced: ˛WDI1 l=J1 ; ˇDI2o l=J1 ; o WDJ2o =J1 ; ıWDl1 = l: The nondimensional
amplitude and frequency of the prescribed periodic base motion are Y = l and =!c ;
respectively. The equations of motion in nondimensional form thus become

R1 C ˛Œ1 C y.t/


R sin 1 C 1 P1 C ˇı R2 cos.1  2 /
Cˇı sin.1  2 /P22 D 0; (2.183)
o R2 C ˇŒ1 C y.t/
R sin 2 C 2 P2 C ˇı R1 cos.1  2 /
ˇı sin.1  2 /P12 D 0 (2.184)
p
where i WD.ci =J1 / l=g is the nondimensional damping coefficient for the i th arm.
The nondimensional parameters associated with ˝ and Y are denoted by the same
notation as the dimensional quantities. The following four independent parameters
regulate the dynamics of the undamped double pendulum: .˛; ˇ; ı; o /:
Without the forcing term, under the mere action of gravity, the double pendulum
exhibits four equilibria obtained from (2.183) and (2.184) putting Pi D 0 D Ri .
The equilibria are .1 ; 2 / D .0; 0/; .0; /; .; 0/; and .; / of which only .0; 0/
is stable (marginally or asymptotically stable depending on the absence or presence
of damping). The double pendulum possesses two vibration modes about the stable
equilibrium .0; 0/ of which the lowest represents an in-phase mode while the second
is an out-of-phase mode. The frequencies and eigenvectors associated with these
modes are respectively given by
(  1=2 ) 1=2
˛o C ˇ ˛ 2 .o /2 C 4˛ˇ 3 ı 2  2˛ˇo C ˇ 2
!1;2 D ; (2.185)
2.o  ˇ 2 ı 2 /
" #|
.˛ 2 .o /2 C 4˛ˇ 3 ı 2  2˛ˇo C ˇ 2 /1=2 ˛o ˙ ˇ
u1;2 D ;1 : (2.186)
2˛ˇı
The eigenvectors uk D Œ1 ; 2 | are normalized according to uk  M  uk D 1:
140 2 Stability and Bifurcation of Structures

The transition curves are constructed by the method of multiple scales [332]. To
this end, let  2!k ; k D 1; 2, and rescale the damping and amplitude of the base
motion as "2 k and "2 Y . The equations of motion are then expanded in Taylor series
about .1 ; P1 ; 2 ; P2 / D .0; 0; 0; 0/ up to terms of fifth polynomial order.
The following time scales are introduced t0 WD t; t2 WD "2 t, and t4 WD "4 t: Only
three time scales are introduced because the perturbation procedure is terminated
at fifth order. A one-term expansion is obtained by carrying out the analysis up to
the cubic order where the solvability condition gives the modulation equations. A
two-term expansion is obtained by carrying out the analysis up to the quintic order.
Provided that the data are sufficiently smooth and differentiable, asymptotic
expansions of the solutions are sought in the form

X
3
k .t; "/ k;j .t0 ; t2 ; t4 /"j (2.187)
j D1

where the functions k;j .t0 ; t2 ; t4 / are to be determined. A detuning from the critical
condition is introduced according to  D 2!k C "2 . To within second order, the
prescribed pivot acceleration is
1 1
yR D Œ@20 C 2"2 @0 @2  Y .ei.2!k t0 C t2 / C cc/ D 4!k Y .!k C "2 / .eit0 C cc/:
2 2
(2.188)

The principal parametric resonance of the kth mode is considered, thus the
solution of the linearized equations of motion at order O."/ is expressed as

Œ1;1 ; 2;1 | D uk ŒAk .t2 ; t4 /ei!k t0 C cc (2.189)

where Ak .t2 ; t4 / is the complex-valued amplitude and cc indicates the complex


conjugate of the preceding term. This generating solution, substituted into the
problem at third order, requires the imposition of a solvability condition by the
Fredholm Alternative Theorem. This yields the complex-valued equation as
1
@2 Ak D  k Ak C ik A2k ANk C iYk ANk ei t2 (2.190)
2
where .k ; k ; k / are given, respectively, by

k D .˛Uk;1
4
C ˇUk;2
4
/=.4!k / C ˇı!k ŒUk;2 Uk;1
3
 2Uk;2
2 2
Uk;1 C Uk;2
3
Uk;1 ; (2.191)
k D 1 Uk;1
2
C 2 Uk;2
2
; k D !k .˛Uk;1
2
C ˇUk;2
2
/: (2.192)

The eigenvectors appearing in (2.191) and (2.192) are expressed as uk D


ŒUk;1 ; Uk;2 | after being normalized. The coefficients .k ; k / are the effective
nonlinearity coefficient (which regulates the bending of the backbone of the
considered pendulum mode) and the effective parametric resonance coefficient,
respectively.
2.10 Parametric Resonances of Conservative Systems with Linear Damping 141

If the perturbation is arrested at this order, by introducing the polar form for the
complex-valued amplitude as Ak .t/ D 12 ak ei k =2 ei t2 =2 into the solvability condition
(2.190), the following coupled ordinary differential equations are obtained:
1
aP k D  k ak C k Y ak sin k ; (2.193)
2
1
Pk D  C k ak2 C 2Yk cos k : (2.194)
2
The two-periodic solutions emanating from the parametric instability are solu-
tions of aP k D0 and Pk D0: The equation relating the amplitude of the motion a, the
base excitation Y; and the bifurcation parameter  is obtained as:
1 p
D k ak2 ˙ 4Y 2 k 2  k 2 : (2.195)
2
Equation (2.195) is the bifurcation equation for the principal parametric instabil-
ity which can be expressed in terms of the frequency as
 p 
1
 2!k C k ak ˙ 4Y k  k :
2 2 2 2 (2.196)
2

According to (2.195), the parametric resonance is initiated only if the excitation


amplitude Y is above Yko , the critical amplitude for the onset of the instability in the
kth mode expressed as

k
2
1 Uk;1 C 2 Uk;2
2
Yko WD D : (2.197)
2jk j 2!k .˛Uk;12
C ˇUk;22
/

An explicit expression of the local approximation of the transition curves around


the tip of the kth Mathieu tongue can be obtained from (2.195) putting ak D 0,
solving for Y , and expanding its formula about  D 0: When the pendulum is
frictionless (i.e., k D 0), the instabilities are obtained from (2.196) in the form
 D 2!k ˙ 2Yk : Consequently, the instability region emanates directly from the
-axis at  D 2!k :
The perturbation analysis can be carried out to the next order (fifth polynomial
degree) if the validity of the formulas for the transition curves is sought to be
extended to higher excitation amplitudes and away from the exact frequency tuning
condition at resonance,  D 2!k . At fifth order, the solvability condition yields
@4 A: The solvability conditions are combined to obtain the time rate of change of
A according to AP D "2 @2 A C "4 @4 A: The outcome of the perturbation treatment is
a two-term expansion of the pendulum resonant motion featuring linear and cubic
nonlinear terms. The transition curves for three values of damping coefficients are
shown in Fig. 2.39 considering the parameters of a prototype used in experiments
[392]. As expected, the critical excitation amplitude for the onset of the instability

You might also like