SCHOOL OF MECHANICAL ENGINEERING

DEPARTMENT OF AERONAUTICAL ENGINEERING

UNIT 1-THEORY OF VIBRATIONS-SME1306

UNIT – I- AIRCRAFT
STRUCTURES – SA
 Chapter 1: Basics of Vibrations for Simple Mechanical Systems

Introduction:

The fundamentals of Sound and Vibrations are part of the broader field
of mechanics, with strong connections to classical mechanics, solid
mechanics and fluid dynamics. Dynamics is the branch of physics
concerned with the motion of bodies under the action of forces.
Vibrations or oscillations can be regarded as a subset of dynamics in
which a system subjected to restoring forces swings back and forth
about an equilibrium position, where a system is defined as an
assemblage of parts acting together as a whole. The restoring forces
are due to elasticity, or due to gravity.

The subject of Sound and Vibrations encompasses the generation of

sound and vibrations, the distribution and damping of vibrations, how
sound propagates in a free field, and how it interacts with a closed space, as
well as its effect on man and measurement equipment. Technical
applications span an even wider field, from applied mathematics and
mechanics, to electrical instrumentation and analog and digital signal
processing theory, to machinery and building design. Most human
activities involve vibration in one form or other. For example, we hear
because our eardrums vibrate and see because light waves undergo
vibration. Breathing is associated with the vibration of lungs and walking
involves (periodic) oscillatory motion of legs and hands. Human speak due
to the oscillatory motion of larynges (tongue).
In most of the engineering applications, vibration is signifying to and
fro motion, which is undesirable. Galileo discovered the relationship
between the length of a pendulum and its frequency and observed the
resonance of two bodies that were connected by some energy transfer
medium and tuned to the same natural frequency. Vibration may results in
the failure of machines or their critical components. The effect of vibration
depends on the magnitude in terms of displacement, velocity or
accelerations, exciting frequency and the total duration of the vibration. In
this chapter, the vibration of a single-degree-of-freedom (SDOF), Two
degree of freedom system with and without damping and introductory
multi-degree of freedom system will be discussed in this section.

1. LINEAR SYSTEMS
Often in Vibrations and Acoustics, the calculation of the effect of a
certain physical quantity termed as the input signal on another physical
quantity, called the output signal; (Figure 1-1). An example is that of
calculating vibration velocity v(t), which is obtained in a structure
when it is excited by a given force F(t). That problem can be solved by
making use of the theory of linear time- invariant systems.

Linear
F(t),v(t) F’(t),v’(t)
time-invariant
p(t),u(t) p’(t),u’(t)
system
Input Signal Output Signal

Fig. 0-1 A linear time-invariant system describes the relationship

between an input signal and an output signal. For example, the input
signal could be a velocity v(t), and the output signal a force F(t), or
the input signal an acoustic pressure p(t) and the output signal an
acoustic particle velocity u’(t). [Sound and vibration book by KTH[1]]

From a purely mathematical standpoint, a linear system is defined as

one in which the relationship between the input and output signals can
be described by a linear differential equation. If the coefficients are,
moreover, independent of time, i.e., constant, then the system is also
time invariant. A linear system has several important features.

Example 0-1 [1]

The figure below, from the introduction, shows an example in which

the forces that excite an automobile are inputs to a number of linear
systems, the outputs from which are vibration velocities at various
points in the structure. The vibration velocities are then, in turn, inputs
to a number of linear systems, the outputs from which are sound
pressures at various points in the passenger compartment. By adding
up the contributions from all of the significant excitation forces, the
total sound pressures at points of interest in the passenger
compartment can be found. The engine is fixed to the chassis via
vibration isolators. If the force F1 that influences the chassis can be cut
in half, then, for a linear system, all vibration velocities v1 – vN caused
by the force F1 are also halved. In turn, the sound pressures p1 – pN,
which are brought about by the velocities v1 – vN, are halved as well. In
this chapter, linear oscillations in mechanical systems are considered,
i.e., oscillations in systems for which there is a linear relation between
an exciting force and the resulting motion, as described by
displacements, velocities, and accelerations. Linearity is normally
applicable whenever the kinematic quantities can be regarded as small
variations about an average value, implying that the relation between
the input signal and the output signal can be described by linear
differential equations with constant coefficients.
 Yik Zji

Body Passenger compartment

Vibration velocities
Forces Sound Pressure

(Picture: Volvo Technology Report, nr 1 1988) [1]

1.1 SINGLE DEGREE OF FREEDOM SYSTEMS

In basic mechanics, one studies single degree-of-freedom systems
thoroughly. One might wonder why so much attention should be given
to such a simple problem. The single degree-of-freedom system is so
interesting to study because it gives us information on how a system’s
characteristics are influenced by different quantities. Moreover, one
can model more complex systems, provided that they have isolated
resonances, as sums of simple single degree-of-freedom systems.

1.2 Spring Mass System

Most of the system exhibit simple harmonic motion or oscillation.

These systems are said to have elastic restoring forces. Such systems can be
modeled, in some situations, by a spring-mass schematic, as illustrated in
Figure 1.2. This constitutes the most basic vibration model of a machine
structure and can be used successfully to describe a surprising number of
devices, machines, and structures. This system provides a simple
mathematical model that seems to be more sophisticated than the problem
requires. This system is very useful to conceptualize the vibration problem
in different machine components. The single degree of freedom system is
indicating as :

k
̈ + ( + )

m m
m
mg mg
X(t
)

(a) (b) (c)

Fig.1.2 (a) Spring-mass schematic (b) free body diagram, (c) free body
diagram in static condition
If x = x(t) denotes the displacement (m) of the mass m (kg) from its
equilibrium position as a function of time t (s), the equation of motion for
this system becomes,

mẍ + k(x + x ) − mg = 0 (1.1)

where k =the stiffness of the spring (N/m),
x = static deflection
m = the spring under gravity load,
g = the acceleration due to gravity (m/s2),
ẍ = acceleration of the system
Applying static condition as shown in Fig. 1.2 (c) the equation of
motion of the system yields
mẍ + kx = 0 (1.2)
This equation of motion of a single-degree-of-freedom system and is a
linear, second-order, ordinary differential equation with constant
coefficients. A simple experiment for determining the spring stiffness by
adding known amounts of mass to a spring and measuring the resulting
static deflection x is shown in Fig. 1.3. The results of this static experiment
can be plotted as force (mass times acceleration) v/s x , the slope yielding
the value of spring stiffness k for the linear portion of the plot as illustrated
in Figure 1.4.

Fig. 1.3 Measurement of the spring stiffness

Force (N)

Displacement (mm)

Fig. 1.4 Determination of the spring stiffness

 Once m and k are determined from static experiments, Equation (1.2)
can be solved to yield the time history of the position of the mass m, given
the initial position and velocity of the mass. The form of the solution of
previous equation is found from substitution of an assumed periodic motion
as,

x(t) = A sin(ω t + ϕ) (1.3)

Where, ω = k/m is the natural frequency (rad/s).

Here, A= the amplitude
Φ= phase shift,
A and Φ are constants of integration determined by the initial
conditions.
If x0 is the specified initial displacement from equilibrium of mass m,
and v0 is its specified initial velocity, simple substitution allows the
constants A and Φ to be obtained. The unique displacement may be
expressed as,

ω x +υ ω x
x(t) = sin[ω t + tan ] (1.4)
ω υ
Or,

( )= sin + cos

Equation 1.2 can also be solved using a pure mathematical approach as

described follows.
Substituting x(t) = C e
mλ e + ke =0 (1.5)
Here C ≠ 0 and eλ ≠ 0 ,
 Hence mλ + k = 0
Or
⁄
=± =±

where, j is an imaginary number = √−1

Hence the generalized solution yields as,

x(t) = C eω +C e ω
(1.6)

where C and C are arbitrary complex conjugate constants of

integration.

The value of the constants C and C can be determined by applying

the initial conditions of the system. Note that the equation 1.2 is valid only
as long as spring is linear.

1.3 Spring Mass Damper system

Most systems will not oscillate indefinitely when disturbed, as

indicated by the solution in Equation (1.4). Typically, the periodic motion
damped out after some time. The easiest way to model this mathematically
is to introduce a new term, named as damping force term, into Equation
(1.2).
Incorporating the damping term in equation (1.2) yield as,

̇ kx = 0
mẍ + cx + (1.7)

Physically, the addition of a dashpot or damper results in the dissipation

of energy, as illustrated in Figure 1.5 the mass, damper and spring
arrangement is as:

̈ + ( ̇ + ̇ )+ ( + )
+ ( ̇ )
k

m m
m
mg mg
X(t
)

(a) (b)
(c)

Fig. 1.5 (a) Schematic of the spring–mass–damper system, (b) free

body diagram of the system in part (a), (c) free body diagram due to static
condition

If the dashpot exerts a dissipative force proportional to velocity on the

mass m, the equation (1.7) describes the equation of the motion.
Unfortunately, the constant of proportionality, c, cannot be measured by
static methods as m and k are measured in spring mass system.
The constant of proportionality c is known as damping coefficient and
its unit in MKS is Ns/m. A general mathematical approach can be used to
solve the equation 1.7 as described below.
Substituting, x(t) = a eλ in equation 1.7, get,

a(mλ eλ + cλeλ + keλ ) = 0 (1.8)

here a ≠ 0 and eλ ≠ 0
hence, mλ + cλ + k = 0
 c k
λ + λ + = 0 (1.9)
m m
The solution of equation 1.8 yields as follows

c 1 c k
λ , =− ± −4
2m 2 m m

The quantity under the radical is called the discriminant. The value of
the discriminant decides that whether the roots are real or complex.
Damping ratio: It is relatively convenient to define a non-dimensional
quantity named as damping ratio. The damping ratio is generally given by
symbol Zeeta () and mathematically defined as;

c
=
2√km

Substituting the value of k ,m and c in terms of  and ω , the equation

(1.7) yields as,

ẍ + 2 ω ẋ + ω x = 0 (1.10)
And equation (1.9) yields as

λ , = − ω ± ω  − 1 = −ξ ω ± ω j (1.11)

where, ω is the damped natural frequency for (0<  <1) the damped

natural frequency is defined as ω = ω 1−

Clearly, the value of the damping ratio,(), determines the nature of the
solution of Equation (1.6).

“Fundamentals of Sound and Vibrations” by KTH Sweden [1], this book is used
under IITR-KTH MOU for course development.
 SCHOOL OF MECHANICAL ENGINEERING
DEPARTMENT OF AERONAUTICAL ENGINEERING

UNIT 2-THEORY OF VIBRATIONS-SME1306

UNIT – I- AIRCRAFT
STRUCTURES – SA
Two degree of freedom systems
•Equations of motion for forced vibration
•Free vibration analysis of an undamped system
 Introduction
• Systems that
h require two independent
d d coordinates
d to describe
d b their
h
motion are called two degree of freedom systems.

Number
N b off
degrees of freedom  Number of masses  number of possible types
of the system
y in the system
y of motion of each mass
 Introduction
• There
h are two equations for
f a two degree
d off freedom
f d system, one for
f each
h
mass (precisely one for each degree of freedom).

• They are generally in the form of coupled differential equations‐that is,

each equation involves all the coordinates.

• If a harmonic solution is assumed for each coordinate,the equations of

motion lead to a frequency
q y equation
q that ggives two natural frequencies
q of
the system.
 Introduction
• If we give
i suitable
it bl iinitial
iti l excitation,
it ti th
the system
t vibrates
ib t att one off th
these
natural frequencies. During free vibration at one of the natural
frequencies, the amplitudes of the two degrees of freedom (coordinates)
are related in a specified manner and the configuration is called a normal
mode, principle mode, or natural mode of vibration.

• Thus a two degree of freedom system has two normal modes of vibration
corresponding to two natural frequencies.

• If we give an arbitrary initial excitation to the system, the resulting free

vibration will be a superposition of the two normal modes of vibration.
However,, if the system
y vibrates under the action of an external harmonic
force, the resulting forced harmonic vibration takes place at the frequency
of the applied force.
 Introduction
• As is evident
d ffrom the
h systems shown
h in the
h ffigures, the
h configuration
f off a
system can be specified by a set of independent coordinates such as
length, angle or some other physical parameters. Any such set of
coordinates is called generalized coordinates.

• Although the equations of motion of a two degree of freedom system are

generally coupled so that each equation involves all coordinates, it is
always possible to find a particular set of coordinates such that each
equation
i off motion
i containsi onlyl one coordinate.
di The
Th equations
i off motion
i
are then uncoupled and can be solved independently of each other. Such
a set of coordinates, which leads to an uncoupled system of equations, is
called principle copordinates.
 Equations of motion for forced
vibration
• Consider
d a viscouslyl ddamped
d two d
degree off ffreedom
d spring‐mass system
shown in the figure.

• The motion of the system is completely described by the coordinates x1(t)

and x2(t), which define the positions of the masses m1 and m2 at any time t
from the respective equilibrium positions
positions.
 Equations of motion for forced
vibration
• The
h externall forces
f F1 and
d F2 act on the
h masses m1 and
d m2, respectively.
l
The free body diagrams of the masses are shown in the figure.
• The application
pp of Newton’s second law of motion to each of the masses
gives the equation of motion:
 Equations of motion for forced
vibration
• It can be
b seen that
h theh first
f equation contains terms involving
l x2, whereas
h
the second equation contains terms involving x1. Hence, they represent a
system of two coupled second‐order differential equations. We can
therefore expect that the motion of the m1 will influence the motion of
m2, and vica versa.
 Equations of motion for forced
vibration
• The
h equations can be
b written in matrix form
f as:

where [m],
[m] [c] and [k] are mass,
mass damping and stiffness matrices,
matrices
respectively and x(t) and F(t) are called the displacement and force
vectors, respectively.which are given by:
 Equations of motion for forced
vibration
• It can be
b seen that
h theh matrices [m],
[ ] [c]
[ ] and
d [k] are allll 2x2 matrices whose
h
elements are the known masses, damping coefficienst, and stiffness of the
system, respectively.
• Further, these matrices can be seen to be symmetric, so that:

Free vibration analysis of an undamped system

• For the free vibration analysis of the system shown in the figure,
figure we set
F1(t)=F2(t)=0. Further, if the damping is disregarded, c1=c2=c3=0, and the
equations of motion reduce to:
 Free vibration analysis of an
undamped system
• We are interestedd in kknowing whether
h h m1 and d m2 can oscillate
ll
harmonically with the same frequency and phase angle but with different
amplitudes. Assuming that it is possible to have harmonic motion of m1
and m2 at the same frequency  and the same phase angle , we take the
solutions to the equations

as:

where
h X1 and
d X2 are constants that
h denote
d the
h maximum
i amplitudes
li d off
x1(t) and x2(t) and  is the phase angle.Substituting the above two
solutions into the first two equations, we have:
 Free vibration analysis of an
undamped system

• Since the above equations must be satisfied for all values of time t, the
terms between brackets must be zero. This yields,

which represents two simultaneous homogeneous algebraic equations in

the unknowns X1 and X2. It can be seen that the above equation can be
satisfied by the trivial soution X1=X2=0, which implies that there is no
vibration.
ib i For
F a nontrivial
i i l solution
l i off X1 and
d X2, the
h determinant
d i off
coefficients of X1 and X2 must be zero.
 Free vibration analysis of an
undamped system

(m1m2 ) 4  {k1  k 2 m2  (k 2  k3 )m1} 2  {(k1  k 2 )(k 2  k3 )  k 22 }  0

• The above equation is called the frequency or characteristic equation

because solution of this equation yields the frequencies of the
characteristic values of the system. The roots of the above equation are
ggiven by:
y
 Free vibration analysis of an
undamped system
• This shows
Thi h that
th t it is
i possible
ibl for
f the
th system
t to
t have
h a nontrivial
t i i l harmonic
h i
solution of the form

when =1 and =2 given by:

We shall denote the values of X1 and X2 corresponding to 1 as

and those corresponding to 2 as .
 Free vibration analysis of an
undamped system
• F th since
Further, i

the above equation is homogeneous, only the ratios and

r2= can be found. For , the equations

give:

• Notice that the two ratios are identical.

 Free vibration analysis of an
undamped system
• The normall modes
Th d off vibration
ib ti corresponding
di tot can be
b
expressed, respectively, as:

• The vectors , which denote the normal modes of vibration are

y
known as the modal vectors of the system. The free vibration solution or
the motion in time can be expressed using

as:

where the constants are determined by the initial

conditions.
 Free vibration analysis of an
undamped system
IInitial
iti l conditions:
diti
Each of the two equations of motion ,

involves second order time derivatives; hence we need to specify two

initial conditions for each mass.
mass
The system can be made to vibrate in its ith normal mode (i=1,2) by
subjecting it to the specific initial conditions.

However, for any other general initial conditions, both modes will be
excited. The resulting motion, which is given by the general solution of the
equations

can be obtained by a linear superposition of two normal modes.

 Free vibration analysis of an
undamped system
I iti l conditions:
Initial diti   
x (t )  c1 x1 (t )  c2 x2 (t )
where c1 and c 2 are constants.

Since and already involve the unknown constants and

we can choose c1=c2=1 with no loss of generality. Thus, the components of the

vector x (t ) can be expressed as:

where the unknown can be determined from the initial

conditions
Free vibration analysis of an
undamped system
Free vibration analysis of an
undamped system
Free vibration analysis of an
undamped system
Frequencies of a mass
mass‐spring
spring system
Example:
l Findd the
h naturall ffrequencies and
d
mode shapes of a spring mass system , which
is constrained to move in the vertical
direction.
Solution: The equations of motion are given
by:

By assuming harmonic solution as:

the frequency equation can be obtained by:

 Frequencies of a mass
mass‐spring
spring system

• The solution to the above equation gives the natural frequencies:

 Frequencies of a mass
mass‐spring
spring system
• From

the amplitude ratios are given by:

 Frequencies of a mass
mass‐spring
spring system
• From

• The natural modes are given by

 Frequencies of a mass
mass‐spring
spring system
• The
h naturall modes
d are
given by:
 Frequencies of a mass
mass‐spring
spring system
• It can be
b seen that
h whenh theh system vibrates
b in its ffirst mode,
d the
h
amplitudes of the two masses remain the same. This implies that the
length of the middle spring remains constant. Thus the motions of the
mass 1 and mass 2 are in phase.
 Frequencies of a mass
mass‐spring
spring system
• When
h the h system vibrates
b in its second
d mode,
d the
h equations b
below
l showh
that the displacements of the two masses have the same magnitude with
opposite signs. Thus the motions of the mass 1 and mass 2 are out of
phase. In this case, the midpoint of the middle spring remains stationary
for all time. Such a point is called a node.
 Frequencies of a mass
mass‐spring
spring system
• Using equations

the motion (general solution) of the system can be expressed as:

 Forced vibration analysis
• The equation
Th ti off motion
ti off a generall two
t degree
d off freedom
f d system
t under
d
external forces can be written as:

• We shall consider the external forces to be harmonic:

where  is the forcing frequency. We can write the steady state solution
as:

where X1 and X2 are, in general, complex quantities that depend on  and

the system parameters. Substituting the above two equations into the first
one:
 Forced vibration analysis
• We obtain:
b

• If we define a term called ‘mechanical impedance’ Zrs(i) as:

and
d write
i the
h first
fi equation
i as:
where
 Forced vibration analysis
• The
h equation
can be solved to obtain:
• Where the inverse of the impedance matrix is given by:

• Therefore, the solutions are:

• By substituting these into the below equation

equation, the solutions can be
obtained.
Multi‐degree of freedom systems

•Modeling of continuous systems as multidegree of freedom systems

•Eigenvalue problem
 Multidegree of freedom systems
• A
As stated
t t d before,
b f mostt engineering
i i systems
t are continuous
contin o s and
d
have an infinite number of degrees of freedom. The vibration
analysis of continuous systems requires the solution of partial
differential equations
equations, which is quite difficult
difficult.

• In fact, analytical solutions do not exist for many partial differential

equations.
ti The
Th analysis
l i off a multidegree
ltid off ffreedom
d system
t on th
the
other hand, requires the solution of a set of ordinary differential
equations, which is relatively simple. Hence, for simplicity of
analysis continuous systems are often approximated as
analysis,
multidegree of freedom systems.

• For a system having n degrees off freedom,

f there are n associated
natural frequencies, each associated with its own mode shape.
 Multidegree of freedom systems
• Different methods can be used to approximate a continuous system as a
multidegree of freedom system. A simple method involves replacing the
distributed mass or inertia of the system by a finite number of lumped masses or
rigid bodies.

• The lumped masses are assumed to be connected by massless elastic and damping
members.

• Linear coordinates are used to describe the motion of the lumped masses. Such
models are called lumped parameter of lumped mass or discrete mass systems.

• The minimum number of coordinates necessary to describe the motion of the

lumped masses and rigid bodies defines the number of degrees of freedom of the
system. Naturally, the larger the number of lumped masses used in the model, the
higher the accuracy of the resulting analysis
analysis.
 Multidegree of freedom systems
• Some problems automatically
indicate the type of lumped
parameter model to be used.

• For example, the three storey

building shown in the figure
automaticallyy suggests
gg usingg a
three lumped mass model as
indicated in the figure.

• In this model, the inertia of the

system is assumed to be
concentrated as three point
masses located at the floor
levels, and the elasticities of the
columns are replaced by the
p g
springs.
 Multidegree of freedom systems
• Another
h popular l method
h d off approximating a continuous system as a
multidegree of freedom system involves replacing the geometry of the
system by a large number of small elements.
• By assuming a simple solution within each element, the principles of
compatibility and equilibrium are used to find an approximate solution to
the original system.
system This method is known as the finite element method.
method
 Using Newton’s second law to derive
equations of motion
The following procedure can be adopted to derive the equations of motion of
a multidegree of freedom system using Newton’s second law of motion.

1
1. Set up suitable coordinates to describe the positions of the various point
masses and rigid bodies in the system. Assume suitable positive directions
for the displacements, velocities and accelerations of the masses and rigid
bodies.

2. Determine the static equilibrium configuration of the system and measure

the displacements of the masses and rigid bodies from their respective static
equilibrium positions
positions.

3. Draw the free body diagram of each mass or rigid body in the system.
Indicate the spring
spring, damping and external forces acting on each mass or rigid
body when positive displacement or velocity are given to that mass or rigid
body.
 Using Newton’s second law to derive
equations of motion
4. Apply
l Newton’s’ second
d law
l off motion to each
h mass or rigid
d body
b d shown
h by
b
the free body diagram as:

Example: Derive the equations of motion of the spring‐mass‐damper system

shown
h iin the
h figure.
fi
 Using Newton’s second law to derive
equations of motion
• Draw free‐body
f b d ddiagrams off masses andd apply
l Newton’s ’ second
d law
l off
motion. The coordinates describing the positions of the masses, xi(t), are
measured from their respective static equilibrium positions, as indicated
in the figure. The application of the Newton’s second law of motion to
mass mi gives:

• or

• The equations of motion of the masses m1 and m2 can be derived from the
above equations by setting i=1 along with xo=0 and i=n along with xn+1=0,
respectively.
respectively
Equations of motion in matrix form
• The
h equations off motion in matrix fform in the
h above
b example
l can b
be
expressed as:
Equations of motion in matrix form
Equations of motion in matrix form
• F an undamped
For d d system,
t th
the equations
ti off motion
ti reduce
d to:
t

• The differential equations of the spring‐mass system considered in the

example, can be seen to be coupled. Each equation involves more than
one coordinate. This means that the equations can not be solved
individually one at a time; they can only be solved simultaneously.

• In addition, the system can be seen to be statically coupled since

stiffnesses are coupled‐
coupled that is the stiffness matrix has at least one
nonzero off‐diagonal term. On the other hand, if the mass matrix has at
least one off‐diagonal term nonzero, the system is said to be dynamically
p
coupled. Further,, if both the stiffness and the mass matrices have nonzero
off‐diagonal terms, the system is said to be coupled both statically and
dynamically.
 Undamped free vibrations
• The
h equations off motion for
f a freely
f l vibrating
b undamped
d d system can be
b
obtained by omitting the damping matrix and applied load vector from:
mx  cx  kx  0
in which 0 is a zero vector. The problem of vibration analysis consists of
determining the conditions under which the equilibrium condition expressed
by the above equation will be satisfied.
• By analogy with the behavour of SDOF systems, it will be assumed that the
free‐vibration motion is simple harmonic (the first equation below), which
may be expressed for a multi degree of freedom system as:
x(t )  xˆ sin(t   )
x   2 xˆ sin(t   )   2 x
• In the above expressions, x̂ represents the shape of the system (which does
not change with time; only the amplitude varies) and  is a phase angle. The
q
third equation above represents
p the accelerations in the free vibration.
 Undamped free vibrations
• S b tit ti
Substituting
x(t )  xˆ sin(t   )
x   2 xˆ sin(t   )   2 x

in the equation
mx  cx  kx  0

we obtain:   2mxˆ sin(

i (t   )  kxˆ sin(
i (t   )  0
which (since the sine term is arbitrary and may be omitted) may be written:
k   mxˆ  0
2

• The above equation is one way of expressing what is called an eigenvalue

or characteristic value problem. The quantities  2 are the eigenvalues or
characteristic values indicating the square of the free‐vibration
frequencies, while the corresponding displacement vectors x̂ express the
corresponding shapes of the vibrating system‐ known as the eigenvectors
or mode shapes.
 Undamped free vibrations
• It can b
be shown
h b
by C
Cramer’s
’ rule
l that
th t the
th solution
l ti off thi
this sett off
simultaneous equations is of the form:
0
x̂ 
x
k   2m
• Hence a nontrivial solution is possible only when the denominator
determinant vanishes. In other words,, finite amplitude
p free vibrations are
possible only when
k   2m  0

• The above equation is called the frequency equation of the system system.
Expanding the determinant will give an algebraic equation of the Nth
degree in the frequency parameter  2 for a system having N degrees of
freedom
freedom.
• The N roots of this equation 12 , 22 , 32 ,....,  N2  represent the frequencies of
the N modes of vibration which are possible in the system.
 Undamped free vibrations
• The mode
Th d having
h i the th lowest
l t frequency
f is
i called
ll d the
th fi
firstt mode,
d ththe nextt
higher frequency is the second mode, etc.
• The vector made up of the entire set of modal frequencies, arranged in
sequence, willll be
b called
ll d the
h frequency
f vector .
1 
 
 2 
  3 
 
 
Normalization:  N 

It was noted
t d earlier
li ththatt th
the vibration
ib ti mode d amplitudes
lit d obtained
bt i d ffrom th
the
eigenproblem solution are arbitrary; any amplitude will satisfy the basic
frequency equation
k   2m  0
and only the resulting shapes are uniquely defined.
 Normalization of modes
• In the
h analysis
l process ddescribed
b d above,
b the
h amplitude
l d off one degree
d off
freedom (the first actually) has been set to unity, and the other
displacements have been determined relative to this reference value. This
is called normalizing the mode shapes with respect to the specified
reference coordinate.

• Other normalizing procedures also are frequently used; e.g., in many

computer programs, the shapes are normalized relative to the maximum
di l
displacement value
l in
i eachh moded rather
h than
h with
i h respect to any
particular coordinate. Thus, the maximum value in each modal vector is
unity, which provides convenient numbers for use in subsequent
calculations.
 Normalization of modes
• The normalizing
Th li i procedure
d mostt often
ft used d in
i computert programs ffor
structural vibration analysis, however, involves adjusting each modal
amplitude to the amplitude ˆ,n which satisfies the condition
ˆnT mˆn  1
• This can be accomplished by computing the scalar factor
vˆ nTmvˆ m  M̂ n
where v̂ n represents an arbitrarily determined modal amplitude, and then
computing the normalized mode shapes as follows:
ˆn  vˆ n Mˆ n1/ 2
By simple substitution, it is easy to show that this givs the desired result. A
consequence of this type of normalizing together with the modal
orthogonality relationships relative to the mass matrix is that
ˆn I
ˆ nT mφ
φ
where  is the complete set of N normalized mode shapes and I is an NxN
identity matrix. The mode shapes normalized in this fashion are said to be
orthonormal relative to the mass matrix.
Mode shapes of a four storey 2D frame
• A model of a four‐story three‐bay frame can be evaluated to
determine the mode shapes. This 2 D model is from a typical
building from the Marmara region in Turkey.
Turkey

• Generally
Generally, the first mode of vibration is the one of primary
interest. The first mode usually has the largest contribution to
the structure's motion. The p period of this mode is the longest
g
and the natural frequency is the lowest.

• Please click on the movie to start!

Mode shapes of a four storey 2D frame
• First
st mode
ode shape
s ape
Mode shapes of a four storey 2D frame
• Seco
Second
d mode
ode shape
s ape
Mode shapes of a four storey 2D frame
• Third
d mode
ode sshape
ape
 Undamped free vibrations
Example:
l

Determine the eigenvalues and eigenvectors of a vibrating system for which

Example
 Example
Solution:
l
When the characteristic equation possesses repeated roots, the
corresponding
p g mode shapes
p are not unique.
q
 Example
Solution:
l
 Rigid body motion
• An unrestrainedd system is one that
h has
h no restraints or supports and d that
h
can move as a rigid body. It is not uncommon to see in practice systems
that are not attached to any stationary frame.
• Such systems are capable of moving as rigid bodies, which can be
considered as modes of oscillation with zero frequency.
• A semidefinite system such as this,
this has a singular stiffness matrix
matrix. In
systems that are not properly restrained, rigid‐body displacements can
take place without the application of any force. Thus, denoting a possible
rigid‐body
i id b d displacement
di l by
b ur, we have
h
fr  Ku r  0
• For a nonzero ur, the above equation can be satisfied provided only that K
is singular. In this case, the below equation can only be satisfied when
=0.
K   2M ur  0
 Rigid body motion
• The
h rigidd body
b d displacements
d l are those
h displacement
d l modesd that
h the
h
element must be able to undergo as a rigid body without stresses being
developed in it.
• Rigid body displacement shapes are also referred to as rigid body modes.
• A system can, of course, have more than one rigid body mode. In the most
general case,
case up to six rigid body modes are possible.
possible For example,
example a
spacecraft or an aeroplane in flight has all six possible rigid‐body modes,
three translations and three rotations, one along each of the three axis.

Rigid
g bodyy modes of a p
plane stress element
 Orthogonality of modes
• The
h naturall modes
d correspondingd to different
d ff naturall frequencies
f can be
b
shown to satisfy the following orthogonality conditions. When    :
n r

nT kr  0 nT mr  0

• Proof: The nth natural frequency and mode satisfy

kn  n2mn

Premultiplying the above equation by r

T

rT kn  n2rT mn

Similarlyy the rth natural frequency
q y and mode shape
p satisfyy
kr  r2mr
 Orthogonality of modes
b n gives:
kr  r2mr by
T
Premultiplying
l l

nT kr  r2nT mr

The transpose of the matrix on the left side of r kn  n r mn will
T 2 T

equal the transpose of the matrix on the right side of the equation:
nT kr  n2nT mr
Subtracting the first equation from the second equation:

 2
n 
 r2 nT mr  0
The equation
q nT mr  0 is true when    which for systems
n r y with
positive natural frequencies implies that    n r
 Modal equations for undamped
systems
• The
h equations off motion for
f a linear
l MDOF system without
h d
damping is:

m
x
k
x
p
t
  ︵ ︶
• The simultaneous solution of these coupled equations of motion that we
have illustrated before for a 2 dof system subjected to harmonic
excitation is not efficient for systems with more DOF, nor is it feasible for
systems excited by other types of forces
forces. Consequently,
Consequently it is advantegous
to transform these equations to modal coordinates.
• The displacement vector x of a MDOF system can be expanded in terms
of modal contributions. Thus, the dynamic response of a system can be
expressed as:
N
x(t )   r qr (t )  φq(t )
r 1
 Modal equations for undamped
systems
N
• h equation x(t )   r qr (t )  φq(t ) , the
Using the h coupled
l d equations in xj(t)
()
r 1
given below

m
x
k
x
p
t
  ︵ ︶
can be transformed to a set of uncoupled equations with modal
coordinates qn(t) as the unknowns. Substituting the first equation into the
second:
m

p
t
N N


r 1
r qr(t )   r qr(t ) ︵ ︶
r 1

Premultiplying each term in this equation by nT gives :

m

p
t
N N


r 1
T
n r qr(t )   nT r qr(t )  nT︵ ︶
r 1
 Modal equations for undamped
systems
• Because off th
B the orthogonality
th lit relations
l ti nT kr  0 nT mr  0 , allll
terms in each of the summations vanish except the r=n term, reducing the
equation to:

p
t
nT n qn(t )  nT n qn(t )  nT︵ ︶
or
M n qn (t )  K n qn (t )  Pn (t )
where
M n  nT mn K n  nT kn Pn (t )  nT p(t )

• The above
Th b equation
ti may b be interpreted
i t t d as the
th equation
ti governing i the
th
response qn(t) of the SDOF system with mass Mn, stiffness Kn, and exciting
force Pn(t).
• Th f
Therefore Mn is
i called
ll d the
h generalized
li d mass ffor the
h nthh naturall mode,
d Kn
the generalized stiffness for the nth mode, and Pn(t) the generalized force
for the nth mode. These parameters only depend on the nth mode.
Modal equations for damped systems
• When
h damping
d is included,
l d d the
h equations off motion ffor a MDOF system
are:

m
x
c
x
k
x
p
t
    ︵ ︶
• Using the transformation N
x(t )   r qr (t )  φq(t )
r 1
where r are the natural modes of the system without damping, these
equations can be written in terms of the modal coordinates. Unlike the
p systems,
case of undamped y , these modal equations
q mayy be coupled
p
through the damping terms. However, for certain forms of damping that
are reasonable idealizations for many structures, the equations become
uncoupled just as for undamped systems
uncoupled, systems. Substituting the second
equation into the first, we obtain:
m

p
t
N N N


r 1
r qr((t )   r q r((t )   r qr((t ) ︵ ︶
r 1 r 1
Modal equations for damped systems
b n gives:
T
• Premultiplying
l l each
h term in this
h equation by

p
t
N N N


r 1
T
n r qr((t )    r q r((t )   nT r qr((t )  nT︵ ︶
r 1
T
n
r 1

which can be rewritten as:

N
M n qn(t )   Cnr q r(t )  K n qn(t )  Pn(t )
r 1
where
h
Cnr  nT cr

The above N equations can be written in matrix form as:

  Cq  Kq  P(t)
Mq
Here C is a nondiagonal matrix of coefficients Cnr.
Modal equations for damped systems
• The modal equations will be uncoupled if the system has classical
damping. For such systems Cnr=0 if nr and Cn can be expressed as:
Cn  2 n M nn

• For such systems:

M n qn  Cn q n  K n qn  Pn (t )
• Dividing by Mn:
Pn (t )
qn  2 nn q n  n2 qn 
Mn
where  n is the damping ratio for the nth mode.
 SCHOOL OF MECHANICAL ENGINEERING
DEPARTMENT OF AERONAUTICAL ENGINEERING

UNIT 3-THEORY OF VIBRATIONS-SME1306

UNIT – I- AIRCRAFT
STRUCTURES – SA
 CHAPTER 9
MULTI-DEGREE-OF-FREEDOM SYSTEMS
Equations of Motion, Problem Statement,
and Solution Methods

Two-story shear building

A shear building is the building whose floor systems are rigid
in flexure and several factors are neglected, for example, axial
deformation of beams and columns.

We will formulate the equations of motion of a simple 2-story

shear building whose mass are lumped at the floor.

The equations of motion are formulated by considering

equilibrium of forces acting on each mass. Any of the two
approaches can be used
(1) Newton’s second law of motion
(2) D’Alembert’s principle of dynamic equilibrium

11 - 1
Newton’s second law of motion
∑ F = mu&&
For each floor mass (j=1 and 2)
p j − f Sj − f Dj = m j u&&j or m j u&&j + f Sj + f Dj = p j ( t )

Two equations can be written in matrix form

⎡ m1 0 ⎤ ⎧ u&&1 ⎫ ⎧ f D1 ⎫ ⎧ f S 1 ⎫ ⎧ p1 ( t ) ⎫
⎢0 ⎨ ⎬+⎨ ⎬+⎨ ⎬= ⎨ ⎬
⎣ m2 ⎥⎦ ⎩u&&2 ⎭ ⎩ f D 2 ⎭ ⎩ f S 2 ⎭ ⎩ p2 ( t ) ⎭

&& + f D + f S = p ( t )
mu
where

⎧u ⎫ ⎡m 0⎤ ⎧ f D1 ⎫ ⎧f ⎫ ⎧p ⎫
u = ⎨ 1⎬ m = ⎢ 1 f = ⎨ ⎬ fS = ⎨ S1 ⎬ p = ⎨ 1⎬
⎩u2 ⎭ ⎣0 m2 ⎥⎦ D ⎩ f D 2 ⎭ ⎩ fS 2 ⎭ ⎩ p2 ⎭

Because all beams are assumed rigid, the story shear force can
be directly related to the relative displacement between
stories.
Vj = k jΔ j
12 EI c
where Δ j = u j +1 − u j and kj = ∑ h3
columns

The elastic force acting on the first story mass comes from
columns below ( f Sb1 ) and above ( f Sa1 ) the floor.
f S1 = f Sb1 + f Sa1

f S1 = k1u1 + k2 ( u1 − u2 )

f S 2 = k2 ( u2 − u1 )

11 - 2
 ⎧ f S 1 ⎫ ⎡ k1 + k2 − k2 ⎤ ⎧ u1 ⎫
⎨ ⎬=⎢ ⎨ ⎬ or f S = ku
⎩ f S 2 ⎭ ⎣ − k2 k2 ⎥⎦ ⎩u2 ⎭

The elastic resisting force vector f S is related to displacement

vector u through the stiffness matrix k .

The damping forces f D1 and f D 2 are related to floor velocities

u&1 and u&2 . The jth story damping coefficient c j relates the story
shear V j due to the damping effects to the velocity Δ& j
associated with the story deformation by
V j = c j Δ& j

We can derive
f D1 = c1u&1 + c2 ( u&1 − u&2 )

f D 2 = c2 ( u&2 − u&1 )

⎧ f D1 ⎫ ⎡ c1 + c2 −c2 ⎤ ⎧ u&1 ⎫
⎨ ⎬=⎢ ⎨ ⎬ or f D = cu&
⎩ f D 2 ⎭ ⎣ −c2 c2 ⎥⎦ ⎩u&2 ⎭

The damping force vector f D and velocity vector u& are related
through the damping matrix c .

Therefore, the equations of motion are

&& + cu& + ku = p ( t )
mu

This matrix equation represents two ordinary differential

equations governing the displacements u1 ( t ) and u2 ( t ) of the
two-story frame subjected external dynamic forces p1 ( t ) and
p2 ( t ) .
Each equation contains both unknowns u1 and u2 , so two
equations are coupled and must be solved simultaneously.

11 - 3
Dynamic Equilibrium (D’Alembert’s principle)

For each of the mass in the system, the external force must be
in balance with
(1) inertia force (resisting acceleration) acting in the
opposite direction to acceleration
(2) damping force (resisting velocity) acting in the
opposite direction to velocity and
(3) elastic force resisting deformation

11 - 4
11 - 5
11 - 6
11 - 7
General Approach for Linear Systems
Discretization
A frame structure can be idealized by an assemblage of
elements—beams, columns, walls—interconnected at nodal
points or nodes. Displacements of nodes are degrees of
freedom. A node in a planar two-dimension frame has 3
DOFs—two translations and one rotation.

If axial deformations are neglected, the number of DOFs can

be reduced because some translational DOF are equal.

The external forces are applied at the nodes which correspond

to the DOFs.

11 - 8
Elastic forces
The elastic forces are related to displacement through stiffness
matrix. The stiffness matrix can be obtained from stiffness
influence coefficient kij , which is the force required along
DOF i due to a unit displacement at DOF j and zero
displacement at all other DOFs.
For example, the force ki1 ( i = 1, 2,...,8 ) are required to
maintain the deflected shape associated with u1 = 1 and all
other u j = 0 .

The force f Si at DOF i associated with displacement u j

( j = 1 to N ) is obtained by superposition:
f Si = ki1u1 + ki 2u2 + ... + kiN u N

11 - 9
Such equation applies to each of f Si where i = 1 to N , so
⎧ f S 1 ⎫ ⎡ k11 k12 .. k1N ⎤ ⎧ u1 ⎫
⎪ f ⎪ ⎢k .. k2 N ⎥ ⎪⎪ u2 ⎪⎪
⎪ S 2 ⎪ ⎢ 21 k22
⎥⎨ ⎬
⎨ ⎬= or f S = ku
⎪ : ⎪ ⎢ : : .. : ⎥ ⎪ : ⎪
⎪⎩ f SN ⎪⎭ ⎣⎢ k N 1 kN 2
⎥
.. k NN ⎦ ⎩⎪u N ⎭⎪

where k is the stiffness matrix of the structure.

This approach can be cumbersome for complex
structures in order to visualize a deflected shape with a unit
displacement at DOF j and zero displacement at all other
DOFs.
The direct stiffness method must be used instead. It
involves assembling of stiffness matrices of structural
members into the stiffness matrix of the whole system. The
appropriate method should be used for a given problem.

Damping forces
Damping forces are related to velocities of nodes through
damping matrix. The method of damping influence coefficient
cij can be used to derive the damping matrix in a similar
manner as stiffness matrix relating elastic forces to
displacements.
However, it is impractical to compute the coefficient cij
of damping matrix directly from the size of the structural
elements. Instead, damping of a MDF system is usually
specified in term of damping ratio and the corresponding
damping matrix can be constructed accordingly.

11 - 10
Inertia forces
Inertia forces are forces related to acceleration of the
mass. An approach to consider inertia forces acting at nodes is
to lump the mass of structural components to nodes.

Inertial forces are related to acceleration at nodes through

the mass matrix m . Mass matrix can be derived using mass
influence coefficient mij which is the external force in DOF i
due to unit acceleration along DOF j . For example, the force
mi1 ( i = 1, 2,...8 ) are required in various DOF to equilibrate the
inertia forces associated with u&&1 = 1 and all other u&&j = 0 .
The force at DOF i due to acceleration at various nodes
can be obtained by superposition

f Ii = mi1u&&1 + mi 2u&&2 + ... + miN u&&N

Such inertia forces at all DOFs are written together in the

inertia force vector f I , which is equal to

⎧ f I 1 ⎫ ⎡ m11 m12 .. m1N ⎤ ⎧ u&&1 ⎫

⎪ f ⎪ ⎢m .. m2 N ⎥ ⎪⎪ u&&2 ⎪⎪
⎪ I 2 ⎪ ⎢ 21 m22
⎥⎨ ⎬
⎨ ⎬= or f I = mu
&&
⎪ : ⎪ ⎢ : : .. : ⎥⎪ : ⎪
⎪⎩ f IN ⎪⎭ ⎣⎢ mN 1 mN 2
⎥
.. mNN ⎦ ⎩⎪u&&N ⎭⎪

When lumped-mass model is used, the mass matrix will be

diagonal. Rotational inertia forces at the nodes are neglected,
so the mass associated with rotational DOFs are zero.

mij = 0 if i ≠ j m jj = m j or 0

11 - 11
Equations of motion

fI + fD + fS = p ( t )

&& + cu& + ku = p ( t )
mu

The off-diagonal terms in the coefficient matrices m, c, and k

are known as coupling terms. The coupling in a system also
depends on the choice of DOFs.

11 - 12
11 - 13
11 - 14
11 - 15
11 - 16
11 - 17
11 - 18
11 - 19
11 - 20
Static Condensation

Static condensation is a method to exclude the DOFs with no

force from dynamic analysis. Typically the formulation of
stiffness matrix in static analysis considers all unrestrained
DOFs at joints between structural members. Some of DOFs
may not be associated with any mass in dynamic analysis, for
example, rotation DOFs in a lumped-mass model, so they
should be excluded to simplify the dynamic analysis.

The equations of motion for a building shown above is

⎡m tt 0⎤ ⎧u &&t ⎫ ⎡ k tt k to ⎤ ⎧ u t ⎫ ⎧pt ( t ) ⎫
⎨ ⎬+ ⎢ ⎨ ⎬=⎨ ⎬
⎢ 0
⎣ 0 ⎥⎦ ⎩u
&&o ⎭ ⎣k ot k oo ⎥⎦ ⎩u o ⎭ ⎩ 0 ⎭

It is partitioned into translation ( ut ) and rotation ( u o ) DOFs.

Each part involves vectors and sub-matrices.

Each group of partitioned equations are

&&t + k tt u t + k tou o = p t ( t )
m tt u and k ot ut + k oou o = 0

11 - 21
Because no inertia terms and external forces are associated
with the rotations, u o can be solved.

u o = −k oo −1k ot ut

Then, we can substitute u o into the equation for translational

DOFs and obtain equations of motion which are simpler as
they involve only translation DOFs.
&&t + kˆ tt ut = pt ( t )
m tt u

where the condensed stiffness matrix is

−1
kˆ tt = k tt − k toT k oo k ot

Note that k to = k Tot because k is a symmetric matrix.

11 - 22
11 - 23
Equation of motion:
Planar systems subjected to translational ground motion
At each instant of time, displacement of each mass is
u tj ( t ) = u j ( t ) + u g ( t )

For N masses, the displacements can be written in compact

form as a vector.
ut ( t ) = u ( t ) + ug ( t ) 1

where 1 is a vector of order N with each element equal to

unity.

The equations of motion previously derived for a MDF

system subjected to external for p ( t ) is still valid except that
the external force for this case (ground excitation) is zero.
fI + fD + fS = 0

Only relative displacements u between masses and the base

produce deformation and elastic and damping forces.

11 - 24
The inertia forces f I are related to the total acceleration u&&t .
f I = mu
&&t

Substituting f I = mu&&t in the equilibrium equation, we get

&& + cu& + ku = −m1u&&g ( t )
mu

The right hand side is the effective earthquake forces due to

ground motion excitation.
p eff ( t ) = −m1u&&g ( t )

This is valid when a unit ground displacement results in a unit

total displacement of all DOFs. In general, this is not always
the case. We introduce the influence vector ι to represent the
influence of ground displacement on total displacement at
DOFs.
u t ( t ) = ιu g ( t ) + u ( t )

The equations of motion are

&& + cu& + ku = −mιu&&g ( t )
mu

11 - 25
For example, vertical DOF u3 is not displaced when the
ground moves horizontally. The influence vector is
⎧1 ⎫
⎪ ⎪
ι = ⎨1 ⎬
⎪ ⎪
⎩0 ⎭

The effective earthquake force is

⎡ m1 ⎤ ⎧1 ⎫ ⎧ m1 ⎫
p eff ( t ) = −mιu&&g ( t ) = −u&&g ( t ) ⎢ ⎥ ⎪1 ⎪ = −u&& t ⎪m + m ⎪
⎢
m2 + m3
⎥⎨ ⎬ g ( )⎨ 2 3⎬

⎣⎢ m3 ⎥⎦ ⎩⎪0 ⎭⎪ ⎪
⎩ 0 ⎭
⎪

Note that the mass corresponding to u&&2 = 1 is m2 + m3 because

both masses will undergo the same acceleration since the
connecting beam is axially rigid.

The effective earthquake force is zero in the vertical DOF

because the ground motion is horizontal.

11 - 26
Inelastic systems
For inelastic systems, the force resisting deformation is no
longer linear relationship and is described by a nonlinear
function
f s = f s ( u, u& )

The equation of motion becomes

&& + cu& + f s ( u, u& ) = −mιu&&g ( t )
mu

Such equation has to be solved by numerical methods as

presented in Chapter 5.

Problem statement
Given a system with known, mass matrix m, damping matrix
c, stiffness matrix k, and excitation p(t) or u&&g ( t ) , we want to
determine the response of the system.
Response can be any response quantity such as displacement
u ( t ) , velocity, acceleration of masses or internal forces, which
is closely related to the relative displacement.
By the concept of equivalent static force, internal forces can
be obtained by static analysis of structure subjected to a set of
equivalent static forces
f S ( t ) = ku ( t )

11 - 27
 SCHOOL OF MECHANICAL ENGINEERING
DEPARTMENT OF AERONAUTICAL ENGINEERING

UNIT 4-THEORY OF VIBRATIONS-SME1306

UNIT – I- AIRCRAFT
STRUCTURES – SA
 TORSIONAL VIBRATION

Single Rotor System

If a rigid body oscillates about a specific reference axis, the resulting motion is called
torsional vibration. In this case, the displacement of the body is measured in terms of an
angular coordinate. In a torsional vibration problem, the restoring moment may be due to the
torsion of an elastic member or to the unbalanced moment of a force or couple. Figure 1
shows a disc, which has a polar mass moment of inertia J0 mounted at one end of a solid
circular shaft, the other end of which is fixed. Let the angular rotation of the disc about the
axis of the shaft be  ,  also represents the shaft’s angle of twist.

Figure 1 Torsional vibration of a disc

Let
 = angular twist of the disc from its equilibrium position
GJ
T = torque required to produce the twist = 
l

J is the polar moment inertia of the rod =  d

4

32
d = rod dia.
l = rod length
Then the torsional spring constant can be defined as,

T GJ
kt  
 l
Applying D’Alembert’s principle the equation of motion may be written as
I  kt  0
kt
  0
I
So the natural frequency n may written as

kt
n 
I

n 1 kt
And f n   Hz
2 2 I
Double Rotor System
Consider a torsional system consisting of two discs mounted on a shaft, as shown in Fig. 2.
The three segments of the shaft have rotational spring constants kt1 , kt 2 , kt 3 and as indicated

in the figure. Also shown are the discs of mass moments of inertia J1 and J2 and the applied
torques Mt1 and Mt2and and the rotational degrees of freedom 1 and  2 and The differential

equations of rotational motion J1 and J2for the discs and can be derived as:

Figure 2 Torsional vibration of a two rotor system

Example
Transverse vibration of beam with various boundary conditions

Figure 3 Transverse vibration of beam

Commonly used boundary conditions for the transverse vibration of beam are as shown in
Figure 4

Figure 4 Commonly used boundary conditions for transverse vibration of beam

References

1. Mechanical Vibration by Morse and Hinkle

2. Mechanical Vibration with application by W.T. Thomas
3. Mechanical Vibrations by V.P. Singh
4. Mechanical Vibrations by S.S. RAo
5. Mechanical Vibrations by G.K. Grover
 SCHOOL OF MECHANICAL ENGINEERING
DEPARTMENT OF AERONAUTICAL ENGINEERING

UNIT 5-THEORY OF VIBRATIONS-SME1306

UNIT – I- AIRCRAFT
STRUCTURES – SA
1 General comments
Vibration phenomena that might be modelled well using linear vibration theory include small am-
plitude vibrations of long, slender objects like long bridges, aeroplane wings, and helicopter blades;
small rocking motions of ships in calm waters; the simplest whirling motions of flexible shafts, and so
on. However, interactions between bridges and foundations, between wings/blades and air, between
ships and waves, between shafts and bearings, and so on, are all nonlinear.
Nonlinear systems can display behaviours that linear systems cannot. These include:

(a) multiple steady state solutions, some stable and some unstable, in response to the same inputs,

(b) jump phenomena, involving discontinuous and significant changes in the response of the system
as some forcing parameter is slowly varied,

(c) response at frequencies other than the forcing frequency,

(d) internal resonances, involving different parts of the system vibrating at different frequencies,
all with steady amplitudes (the frequencies are usually in rational ratios, such as 1:2, 1:3, 3:5,
etc.),

(e) self sustained oscillations in the absence of explicit external periodic forcing, and

(f) complex, irregular motions that are extremely sensitive to initial conditions (chaos).

Analytical intractability and limitations in computational resources make it difficult to system-

atically study the abovementioned phenomena in large systems (though harmonic balance is a useful
technique; see below). For the most part, detailed studies of nonlinear vibrations are conducted
using small systems (with perhaps just one or two degrees of freedom). A good qualitative under-
standing of the phenomena observed for the small system is invaluable when the same phenomena
are subsequently encountered in larger systems.
The utility of precise numerical solutions remains high where appropriate. However, in nonlinear
dynamics it is difficult to extract the qualitative essence from simulations alone. Therefore, an
essential complement to all-numerical studies of large nonlinear systems is the analytical/theoretical
study of simplified systems.

1
2 Analysis techniques
Three broad categories of techniques for analyzing nonlinear systems are:

(a) heuristic techniques like Galerkin methods, including harmonic balance

(b) asymptotic techniques, including the methods of averaging and multiple scales, and
(c) rigorous mathematical results about dynamical systems.

This introduction will concentrate on the first two categories.

2.1 Convergent, asymptotic, and heuristic

To make the later discussion more meaningful, let us distinguish between the terms convergent,
asymptotic, and heuristic.
A convergent series dependent on a parameter (say, ǫ) is one where if we fix ǫ and take more
and more terms, the sum converges to the correct answer. An asymptotic series dependent on a
parameter (say, ǫ “small”) is one where if we take a fixed number of terms and take ǫ smaller and
smaller, the sum gets more and more accurate. Convergent series need not be asymptotic, and vice
versa1 .
In harmonic balance, there is a periodic solution we wish to approximate. That periodic solution
has a convergent Fourier series representation. However, in the application of harmonic balance with
many terms, we obtain equally many coupled, usually nonlinear, equations in terms of the coefficients
(see below). In practice, harmonic balance is often used with only a few harmonics, usually with
excellent results but never any formal advance guarantees of how accurate the solution will be with
a given number of terms included. In this sense, harmonic balance is a heuristic method.
We now discuss these methods in more detail.

2.2 Galerkin methods, and harmonic balance

The basic Galerkin method is now described using a simple boundary value problem,

ẍ + x − 3t = 0 , with x(0) = x(π/2) = 0 .

N
3π X
The exact solution is x = 3t − sin t . As an approximation we assume, say, x ≈ ak sin 2kt .
2 k=1
Substituting into the governing equation, we obtain a nonzero quantity r(t) called the residual. We
make r(t) orthogonal to the assumed basis functions, i.e., set
Z π/2
r(t) sin 2kt dt = 0 , for k = 1, 2, · · · , N .
0

The above process, called a Galerkin projection, yields N equations for the N unknown ak ’s, which
upon solution give the approximate solution. The approximation to 3 terms is
1 1
x ≈ − sin 2t + sin 4t − sin 6t ,
10 35
which has an error ≤ 0.024. More terms yield more accuracy.
1
See, e.g., E. J. Hinch, Perturbation Methods, Cambridge University Press, 1991.

2
 Note that for this linear ODE, the equations for the unknown ak ’s are linear and algebraic,
while for general nonlinear ODE’s these will be nonlinear algebraic equations (see below). For
partial differential equations in time and space, the approximation will typically be of the form
N
X
ak (t)φk (x), where the φk are functions of space chosen to suit the problem (e.g., satisfy boundary
k=1
conditions).

The technique of Harmonic Balance is a specialized application of the Galerkin method to find
periodic solutions in vibration problems. There are several slightly different versions of the method.
Here, we consider unforced, undamped, conservative problems, e.g.,
ẍ + x3 = 0 . (1)
We start with, say, x ≈ A sin ωt + B sin 3ωt . Note that the unknown ω appears in the functions
sin ωt and sin 3ωt, and so there are actually three unknowns in the two term approximation. Sub-
stituting into the differential equation, multiplying in turn by sin ωt and sin 3ωt, and integrating in
each case from 0 to 2π/ω and then equating to zero (the Galerkin projection), we obtain:
−Aω 2 + 3A3 /4 − 3A2 B/4 + 3AB 2 /2 = 0 ,
−9Bω 2 − A3 /4 + 3A2 B/2 + 3B 3 /4 = 0 .
Treating the indeterminate A as a parameter, we obtain ω = 0.8869A and B = −0.04482A.
Variations of the above method are used as the problem changes.
Harmonic balance with a few terms usually gives good approximations to periodic solutions. For
example, some numerical results for the above nonlinear oscillations of Eq. 1, as compared with the
two term harmonic balance calculation given above, are shown in Fig. 1. Oscillations at four different
amplitudes are shown, and the figure appears to have four different curves. Each of these curves is
in fact two superimposed and nearly indistinguishable curves (one solid, one dash-dot). The small
difference between the solid (numerical) and dash-dot (harmonic balance) is visible towards the right
side of the figure (for larger t).
The results show that the two term harmonic balance solution is very accurate. The strong
dependence of frequency on amplitude is also clearly seen.

2.3 A first look at asymptotic techniques

Asymptotic techniques depend on some parameter in the problem being very small (or very large,
which is the same thing on taking reciprocals). In the limit as the small parameter becomes zero, the
problem should be analytically tractable. The basic ideas can be demonstrated using the following
root-finding example:
ǫ x6 + x − 1 = 0 , (2)
where 0 < ǫ ≪ 1. If ǫ = 0, x = 1 is the only root. For nonzero ǫ, that root is perturbed to
x = 1 − ǫ + 6ǫ2 − 51ǫ3 + O(ǫ4 ) .
The O(ǫ4 ) above represents a quantity that is no bigger than some finite constant times ǫ4 , as ǫ goes
to zero.
For ǫ 6= 0, Eq. 2 has five other “large” roots, obtainable via a singular perturbation scheme. One
of them is
1 3 14 2/5
x = −ǫ−1/5 − + ǫ1/5 − ǫ + O(ǫ3/5 ) .
5 25 125
The two “asymptotic” approximations above are useful for sufficiently small ǫ.

3
 3

x(t) 1

−1

−2

−3
0 2 4 6 8 10 12 14 16 18 20
time
Figure 1: Solutions for Eq. 1. Solid line: numerical. Dashdot: harmonic balance (can be viewed as
slightly distinct from solid line, for larger times).

2.4 Averaging and multiple scales

The method of averaging is a specialized asymptotic technique for systems of the form

ẋ = ǫf (x, t) , ǫ ≪ 1. (3)

Here, we assume f (x, t) = f (x, t + T ) for all x, t. An approximation to the solution is found by
solving the simpler equation

1
Z T
ẋ = ǫf0 (x) , where f0 (x) = f (x, t) dt .
T 0

Nonlinear oscillators, e.g.,

ẍ + x = ǫẋ(1 − x2 ) , (4)
are not directly amenable to averaging; but they can be put in that form via a change of variables to
x = A(t) sin(t + φ(t)) , along with the added constraint equation ẋ = A(t) cos(t + φ(t)) . In this form,
the asymptotic method of averaging has been widely used to study a variety of weakly nonlinear
oscillators that are slightly perturbed versions of the harmonic oscillator (ẍ + x = 0).
For illustration, Eq. 4 yields the two equations
³ ´
Ȧ = ǫ A/2 − A3 /8 + A cos(2t + 2φ)/2 + A3 cos(4t + 4φ)/8 ,
³ ´
φ̇ = ǫ − sin(2t + 2φ)/2 + A2 sin(2t + 2φ)/4 − A2 sin(4t + 4φ)/8 .

4
Finally, by first order averaging (higher order averaging is possible, but not done here), we get
³ ´
Ȧ = ǫ A/2 − A3 /8 , and φ̇ = 0 .

The above two equations show that A = 0 is an unstable equilibrium; all other solutions slowly but
eventually approach A = 2 (assuming A > 0); and the phase of the oscillation remains steady, at
least at first order.

The method of multiple scales, also applicable to Eq. 3, involves an additional issue, namely the
identification and removal of secular terms, as illustrated below for Eq. 4 using two time scales.
Let t be the actual time; and τ = ǫ t be a slow time. Assume x = x(t, τ ). Now

∂x ∂x ∂2x ∂2x
ẋ = +ǫ , and ẍ = 2 + 2ǫ + O(ǫ2 ) .
∂t ∂τ ∂t ∂τ ∂t
Using subscripts t and τ to denote partial derivatives with respect to these quantities, we have
n ³ ´o
xtt + x = ǫ −2xτ t + xt 1 − x2 + O(ǫ2 ) .

Assuming a solution of the form x = x0 + ǫx1 + · · ·, we obtain

n ³ ´o
x0,tt + x0 = ǫ −x1,tt − x1 − 2x0,τ t + x0,t 1 − x20 + O(ǫ2 ) .

Collecting terms, at leading order we obtain

x0,tt + x0 = 0 ,

which has the general solution x0 = A(τ ) sin(t + φ(τ )). Substituting this at the next order we obtain
(dropping the explicit dependence of A and φ on τ , and using primes to denote a τ -derivative)
′ ′
x1,tt + x1 = A3 cos(3t + 3φ)/4 + (−2A + A − A3 /4) cos(t + φ) + 2Aφ sin(t + φ) .

In the above equation, the solution for x1 can contain t sin(t + φ) and t cos(t + φ) (effectively the
same as t sin t and t cos t). These secular terms make the approximation break down by the time
t = O(1/ǫ). The validity of the expansion can be extended by removing the secular terms, which
can be done here by requiring that the coefficients of the sine and cosine in the forcing be zero, i.e.,
′ ′ ′
−2A + A − A3 /4 = 0 and 2Aφ = 0. Noting that Ȧ = ǫA , etc., we find the evolution of A and φ
are governed, at this order of approximation, by the same equations as obtained by averaging:

Ȧ = ǫ (A/2 − A3 /8) , and φ̇ = 0 . (5)

3 The phase plane

Our study of entrainment in section 9 will involve the use of a popular and powerful idea from
nonlinear dynamics: the idea of the phase space. The essential idea is described below.
Consider a system of two equations

ẋ = f (x, y) , and ẏ = g(x, y) .

Sometimes, instead of plotting x and y individually versus t, we just plot x versus y. If, say, x rises
monotonically from 0 to 1 as t increases, while y rises from −1 to 3 during the same time, then on

5
 (a) x
(c)
k,c
m F(t)
k1
x1
m1
(b)
k2
h
m2
k,c k3 x2
h
x
m

Figure 2: (a) A linear, damped, forced system. (b) A nonlinear system. The spring has a free length
L0 > h. (c) A nonlinear two degree of freedom system. Mass m1 is constrained to move frictionlessly
in the vertical direction, while mass m2 moves in the horizontal direction. Gravity is neglected, for
simplicity.

the x versus y plane we have a single curve that goes from the point (0, −1) to (1, 3). The (x, y)
plane is called the phase plane. In a more general case, with n dependent variables, we would have
an n-dimensional phase space.
Looking at solutions in the phase space has the disadvantage of losing detailed information about
the exact way in which x and y vary with time. However, it has the obvious advantage of reducing
the dimensionality of the system by one: the solution goes from a curve in the three-dimensional
(x, y, t) space to the two-dimensional (x, y) plane. In addition, there are other advantages involving
geometrical ideas about various types of solutions and how they behave. For example, if x and y
approach constant values, then the graphs of x and y versus t are horizontal lines; but in the phase
plane, the graph of x versus y approaches a point. Similarly, if x and y are periodic functions with
some period T , and with some phase difference between them, then in the phase plane we see a
closed curve. Interested readers will find many excellent books available on nonlinear dynamics, and
topics touched upon in these notes are discussed properly in such books. A representative sample of
references is provided at the end.

4 Multiple solutions
A damped linear system, such as sketched in Fig. 2(a), governed by the linear differential equation
mẍ + cẋ + kx = f (t) ,
has a uniquely defined long term behaviour (after transients die out). For example, consider
ẍ + 0.3 ẋ + x = sin 3.2 t . (6)

6
 0.5

0.4

0.3

0.2

0.1
x(t)

−0.1

−0.2

−0.3

−0.4

−0.5
0 5 10 15 20 25 30
time
Figure 3: Solutions for Eq. 6 converge to the same long-time behaviour regardless of initial conditions.

Two different solutions, for two different initial conditions, are shown to converge to the same “long-
time” solution in Fig. 3.
In contrast, consider the system shown in Fig. 2(b), with the spring’s free length L0 greater than
h. Now it is clear that this nonlinear system will have three equilibrium positions: one at x = 0,
which will be unstable, while one stable position at some nonzero positive x, and another (reflected)
one for negative x. This simple example shows that it is possible for general deterministic nonlinear
systems to have more than one steady state solution in response to the same inputs (but, of course,
with different initial conditions).
This system is not analyzed here in detail; other examples of multiple solutions will soon be
analyzed.
In practical engineering, examples of multiple solutions are encountered in a variety of situations.
A few examples are provided below.

• Buckling. Beyond a certain load, the structure has more than one equilibrium; the nominal
equilibrium loses stability, and new stable equilibrium positions appear. This is related to the
system in Fig. 2(b).

• Whirling of shafts at, near and possibly beyond critical speed. A non-whirling solution still
exists, but is now unstable.

• Resonances in nonlinear systems. When the forcing frequency is near the linear natural fre-
quency, there can be more than one possible stable steady state solution. This example will be
covered again under “jumps”.

7
 • Machine tool chatter. Under certain operating conditions, the cutting tool might chatter a lot
(poorer surface finish) or very little: there is more than one stable steady state solution.

• Systems with dry friction. Some systems with dry friction, for small forcing near resonance,
can have two solutions: one with large amplitude, and one without vibrations.

5 Forced vibrations (via harmonic balance)

Consider the damped nonlinear forced system given by

ẍ + cẋ + x + ax3 − F sin ωt = 0 . (7)

We will study this system using single term harmonic balance. Let us assume x ≈ A sin ωt+B cos ωt.
The assumption is that the solution is dominated by a response at the same frequency, though not
at the same phase, as the forcing. The assumption is exactly true for the linear system (with a = 0),
and approximately true for reasonable values of a and most values of ω. This single harmonic
approximation is sufficient for the purposes of this section.
Substituting into the equation of motion and using some trigonometric identities such as sin3 x =
(3 sin x − sin 3x)/4, we obtain

−ω 2 A sin ωt − ω 2 B cos ωt + cωA cos ωt − cωB sin ωt + A sin ωt + B cos ωt − 41 aA3 sin 3ωt
· · · + 34 aA3 sin ωt + 43 aA2 B cos ωt − 34 aA2 B cos 3ωt + 43 aAB 2 sin 3ωt + 34 aA sin ωtB 2
· · · + 41 aB 3 cos 3ωt + 34 aB 3 cos ωt − F sin ωt = negligible terms.

Multiplying by sin ωt or cos ωt, integrating w.r.t. t from 0 to 2π/ω, and then setting them equal
to zero, is equivalent to simply picking out the coefficients of sin ωt or cos ωt, respectively, and setting
them equal to zero. This gives:
3 3
−Aω 2 − cBω + A + aA3 + aAB 2 − F = 0 ,
4 4
3 3
−Bω 2 + cAω + B + aA2 B + aB 3 = 0 .
4 4
The solutions to the two simultaneous equations above provide a fairly accurate picture of the
dynamics of the system in Eq. 7.

5.1 Unforced, undamped case

If we put c = 0 and F = 0, then we obtain an approximate solution to the unforced, undamped
system, for which
1p
B = 0 , and ω = 4 + 3aA2 .
2
The above (approximate) result tells us that for undamped, unforced periodic oscillations the fre-
quency of oscillations depends on the amplitude. The graph of A versus ω (i.e., with amplitude
along the vertical axis) is usually called a “backbone curve” because of its shape. In this system,
the strength of the nonlinearity is measured by the single quantity a, and so it is not surprising
that the amplitude dependence of the frequency (which happens only for nonlinear systems) involves
a-dependence as well. It is usual to call the case of a > 0 a stiffening nonlinearity, and the case
a < 0 a softening nonlinearity. In the presence of a stiffening nonlinearity, frequency increases with
amplitude; in the case of a softening nonlinearity, frequency decreases with amplitude.

8
5.2 Forced, damped case
In the general case, if we select a certain forcing amplitude F and angular frequency ω, then we
can in principle solve for A and B (and hence the response) in terms of a and c. In practice, it is
convenient to solve the equations numerically. Some specific results are shown in the three plots of
Fig. 4,√where frequency ω is plotted along the horizontal axes and amplitude of response (taken to
mean A2 + B 2 from the harmonic balance equations) is plotted along the vertical axes.
Fig. 4(a) shows the effect of nonlinearity. For a = 0, we have the familiar linear resonance curve.
For increasing a while holding all other things constant, the resonance curve leans over to the right
(for a stiffening nonlinearity; if we took a < 0 it would lean over to the left).
Fig. 4(b) shows the effect of varying damping c, while holding all other things fixed. Since a is
fixed, the backbone curve is fixed. It is seen that the hump in the amplitude versus frequency curve
follows the backbone curve in each case – hence the importance of the backbone curve. All other
things held fixed, decreasing c raises the hump, i.e., raises the maximum response amplitude possible
with a given amplitude of harmonic forcing. If we allow both c and F to become very small, the
amplitude-frequency curve follows the backbone curve even more closely (not surprising, because the
backbone curve is obtained by setting c = 0, F = 0).
Finally, Fig. 4(c) shows the effect of increasing forcing amplitude while holding other things
constant. It is seen that the amplitude versus frequency curve has a hump that leans over to the
right; it would lean to the left if the nonlinearity was of the opposite sense, i.e., a was negative. It is
seen that for relatively small damping, the hump in the amplitude versus frequency plot follows the
backbone curve. Larger F leads to a higher hump.
A few further remarks may be made about the response of this simple nonlinear system. For
very high frequencies of forcing, inertia dominates and the amplitude of motion is very small; in such
cases, the x3 nonlinearity is insignificant because |x3 | ≪ |x|, and the system behaves essentially like
a linear system. The most visible qualitative difference between the linear and nonlinear system is in
the leaning over of the hump near resonance; this is not surprising because large amplitude motions
are (in this system, though not for all systems) the reason for nonlinear terms to become important2 .
It is also clear that for F = 2 and ω = 3, say, there are three different possible amplitudes of
response (thus, multiple solutions in response to the same input forcing). Of these, it is possible to
show that the smallest and largest amplitude solutions are stable, while the intermediate amplitude
solution is unstable (this stability issue will not be discussed fully in these lectures, but a limited
study will be presented below).

6 Jumps
Figure 4(a) also shows clearly the existence of multiple solutions for this problem. There are three
vertical lines in the figure, marked 1, 2 and 3. For the solid line (marked 2), we see that there are
three amplitudes possible (shown in the figure with heavy dots marked P, Q and R). Of these, the
point Q is unstable, while P and R are stable. Though an analysis of the stabilities of these points is
not conducted here, I mention that such stability analyses can be conducted using several techniques.
These include direct numerical simulation; Floquet theory (a topic not covered in these lectures); and,
under more limited circumstances (weak nonlinearity, light damping and small forcing), asymptotic
techniques like the method of averaging or the method of multiple scales.
2
Consider a “simply supported” slender rod supported on pins at its two ends. A small clearance will exist in the pin
holes. For small amplitude vibrations, when the amplitude of vibrations is comparable to that clearance, the vibrations
will in fact be strongly nonlinear. For somewhat larger amplitudes, nonlinearities will be unimportant. Finally, for
very large amplitudes, nonlinearities will be important again.

9
 (a) Changing a, for c=0.2, F=0.5
3
1 2 3 a=0
2.5 a=0.1
amplitude

a=0.4
2 a=2
1.5
R
1
Q
0.5
P
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
ω

(b) Changing c, for a=2, F=0.5

3
backbone curve
2.5 c=0.05
amplitude

c=0.1
2 c=0.2
c=0.4
1.5

0.5

0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
ω

(c) Changing F, for a=2, c=0.2

3
backbone curve
2.5 F=0.5
amplitude

F=1.0
2 F=1.5
F=2
1.5

0.5

0
0 1 2 3 4 5 6
ω

Figure 4: Forced vibrations of a nonlinear system, via harmonic balance. See text for details.

10
 The figure also shows the phenomenon of jumps, which are discontinuous changes in the steady
state response of a system as a parameter (here, forcing frequency) is slowly varied. Imagine that
we start by forcing the system at a low frequency; there is a unique steady state periodic solution,
on which the system response settles. As we raise the frequency quasistatically (very slowly; so
slowly that there are no transients and we get a sequence of steady states), we eventually reach the
first vertical dashed line (marked 1). Beyond this frequency, there are three possible solutions, but
the system stays on the uppermost branch. Passing through point R, the system does not show
any awareness of the alternative solutions at P and Q. Eventually reaching the vertical dashed line
marked 3, the system response jumps to the lower branch, as indicated by the downward arrow. On
further increasing the forcing frequency, the system has a unique, stable solution. Finally, if we now
start decreasing the forcing frequency from some initially large value, then the system response stays
on the lower branch as we cross line 3, and jumps up, as shown by the arrow, when we reach line
1. For frequencies between line 1 and line 3, if we start the system from arbitrary initial conditions,
then which response the system chooses (upper or lower; not, for generic initial conditions, the
intermediate unstable one) depends on initial conditions.

7 Harmonics and subharmonics

It is possible for the response of a nonlinear system to contain frequencies other than that of the
forcing frequency. In fact, it is quite common for the response to have frequencies that are multiples
of the forcing frequency. To see this through simple examples, consider the following system:

ẍ + 0.05 ẋ3 + x3 = sin ωt . (8)

Two values of ω were chosen, based on a preliminary study using harmonic balance (details not
given here), for detailed numerical study: these values are ω = 0.4 and ω = 1.66. Numerical results
obtained are summarized in Fig. 5. In the figure, the numerically obtained power spectral density
of the forcing is plotted for each case, and shows a single peak in each case (see Figs. 5 (a) and
(d)). For both values of ω, the time series (direct numerical solution) settles down to qualitatively
similar periodic solutions (see Figs. 5 (c) and (e)). For both cases, the power spectral density of
the system response shows multiple peaks, at frequencies in the proportion 1 : 3 : 5 : · · ·. In other
words, in the response in each case has content or “energy” at frequencies other than the forcing
frequency. This feature is common in nonlinear systems. Note that in the harmonic ratios above,
even numbers are missing. That is, no frequency component at twice the forcing frequency appears
in the response. This is because the system chosen here has only odd order nonlinearities (cubic
terms). In the presence of some even order nonlinear terms, even order harmonics would also be
expected.
Finally, note that for ω = 0.4 the fundamental frequency of the response equals the forcing
frequency (compare the peaks in Figs. 5 (a) and (c)). This situation, with higher harmonics, is very
common in nonlinear vibrations. In contrast, for ω = 1.66, the fundamental frequency of the response
is one third of the forcing frequency (compare the peaks in Figs. 5 (d) and (f)). The 1/3 frequency
response is called a subharmonic, and occurs somewhat less frequently than higher harmonics.
The above example provides a simple picture of a relatively simple nonlinear system. For more
general, higher dimensional systems under more complex forcing, many different modes as well as
harmonics can interact in complex ways to produce responses that are difficult to characterize and
understand. There is no simple yet general theory for such cases, and problems that arise need to
be tackled on a case by case basis.

11
 (a) ω = 0.4; PSD of forcing (d) ω = 1.66; PSD of forcing

0 0
10 10
squared amplitude

squared amplitude
−5 −5
10 10

−10 −10
10 10

−15 −15
10 10
0 0.2 0.4 0.6 0 0.2 0.4 0.6 0.8 1
frequency frequency

(b) ω = 0.4; time series of response (e) ω = 1.66; time series of response
1.5
0.8
1 0.6
0.4
0.5
0.2
x(t)

x(t)

0 0
−0.2
−0.5
−0.4
−1 −0.6
−0.8
−1.5
0 10 20 30 40 50 0 10 20 30 40 50
time, t time, t

(c) ω = 0.4; PSD of response (f) ω = 1.66; PSD of response

0 0
10 10
squared amplitude

squared amplitude

−5 −5
10 10

−10 −10
10 10

−15 −15
10 10
0 0.2 0.4 0.6 0 0.2 0.4 0.6 0.8 1
frequency frequency

Figure 5: Numerical simulation results for Eq. 8. See text for details.

12
 (a) Response at forcing frequency (b) Subharmonic response

response 1.5 response

4 forcing forcing
1

2
0.5
x(t)

x(t)
0 0

- 0.5
-2

-1
-4
0 1 2 3 0 1 2 3 4 5
time, t time, t
Figure 6: Numerical simulation results for Eq. 9. See text for details.

Before concluding this section, it is worth looking at another system with a clearer and more
convincing subharmonic response. The equation

ẍ + x + 4x3 = sin 6t

has the exact solution

x = − sin 2t ,
a subharmonic resonance. In this solution, the response has no component at the forcing frequency.
The same system also has a solution that is dominated by the forcing frequency. By first order
harmonic balance, that solution is approximately A sin 6t, with A satisfying

3A3 − 35A − 1 = 0, or A ≈ 3.43, −0.03, −3.40 .

From our previous experience with forced vibrations (section 5), we can arrange the three solutions
by magnitude, to get
|A| = 0.03, 3.40, 3.43 ;
and we expect that the intermediate solution (3.40) will turn out to be unstable.
To see that these solutions are meaningful, we add a little damping, and numerically integrate
the equation
ẍ + 0.03ẋ + x + 4x3 = sin 6t (9)
for different initial conditons. Partial numerical results are shown in Fig. 6. For initial conditions
that are sufficiently close to the steady state motion of interest, the numerical solution converges in
each case to a steady state solution approximately equal to that expected from the above calculations
(due to the introduction of small damping, the solutions are slightly different). In particular, Fig.
6(a) shows a solution at the forcing frequency, with an amplitude of roughly 3.4 (consistent with our
undamped estimate of 3.43); while Fig. 6(b) shows a solution at 1/3 of the forcing frequency, with
an amplitude of approximately 1.

13
8 Limit cycles
Some systems have self-sustained vibrations. These include squealing door hinges, electric wires
whistling in the wind, and whirling shafts. These self-sustained vibrations are periodic motions that
are locally unique, and which occur in the absence of external periodic forcing.
To study limit cycles, we will again use the van der Pol equation (Eq. 4), reproduced here as
ẍ + x − ǫẋ(1 − x2 ) = 0 . (10)
Let us start with one-term harmonic balance,
x ≈ A sin ωt . (11)
Note that the cosine is not explicitly included here, because time t does not appear explicitly in the
equation, and so we can choose t = 0 in such a way as to make the coefficient of the cosine equal to
zero; however, for the same reasons, the cosine is implicitly included, i.e., the same solution form, on
shifting time, automatically includes the cosine.
Substituting Eq. 11 into Eq. 10, we obtain
à !
³
2
´ A3 ǫωA3
−ω A + A sin ωt + ǫω − A cos ωt − cos 3ωt = negligibly small terms.
4 4
From the coefficient of sin ωt, we find that either A = 0 or ω = 1. From the coefficient of cos ωt, we
find that A3 /4 − A = 0, which means either A = 0 or A = 2 (we ignore the negative roots, which
provide no new physical information).
Thus, with one term harmonic balance, we have partly verified the information obtained from
first order averaging, or from the method of multiple scales, for the case 0 < ǫ ≪ 1, in section
2.4, Eq. 5. However, Eq. 5 had in fact provided more information, which harmonic balance has not
provided. Harmonic balance can find periodic solutions, but it cannot say whether they are stable
or not. Equation 5, on the other hand, shows that A = 0 is unstable and A = 2 is stable, by the
following simple analysis:
1. For the A = 0 case, we linearize for small A, and obtain
A
Ȧ = ǫ ,
2
which shows that solutions grow exponentially; thus, the A = 0 solution is unstable.
2. For the A = 2 case, we let A = 2 + B, to obtain
à !
3B 2 B 3
Ḃ = ǫ −B − − ,
4 8
which we linearize for small B to obtain
Ḃ = −ǫB ,
which in turn shows that provided B (or the deviation from A = 2) is sufficiently small to start
with, it decays exponentially to zero. Thus, the A = 2 solution is stable. (In fact, it is easy to
show that all initial conditions other than A = 0 eventually settle on A = 2, but we skip that
demonstration here.)
Note that the previous stability conclusions are exactly reversed if ǫ is negative instead of positive;
while the harmonic balance results, blind to stability issues, remain unaffected.

14
9 Entrainment
Recall the van der Pol equation encountered above:
³ ´
ẍ + x = ǫẋ 1 − x2 , where 0 < ǫ ≪ 1 .

As shown above, this equation has a stable limit cycle of amplitude about 2, and angular frequency
1 + O ǫ2 . Now consider a small perturbation where the “spring” is a little stiffer or a little softer,
¡ ¢

h ³ ´ i
ẍ + x = ǫ ẋ 1 − x2 + ∆x ,

where the O(1) quantity ∆ is called a detuning parameter. It may be expected that changing the
spring stiffness a little, just in itself, does nothing except change the angular frequency of the limit
cycle a little, so that it becomes 1 + O (ǫ). Indeed, on averaging the above equation, we find
³ ´ ∆
Ȧ = ǫ A/2 − A3 /8 , and φ̇ = −ǫ .
2
The above nonzero φ̇ indicates that the period of the solution is now slightly different from 2π.
Now, consider the equation
h ³ ´ i
ẍ + x = ǫ ẋ 1 − x2 + ∆x + F sin t , (12)

which represents a van der Pol oscillator periodic with forcing at a frequency slightly different from
that of the unforced limit cycle3 .
What sort of behaviour can we expect? If F is small, then we expect the original limit cycle
with period slightly different from 2π because of the detuning. If F is sufficiently large, perhaps the
forcing will overwhelm the unforced dynamics, and the oscillation will phase-lock with the forcing,
a phenomenon called entrainment. For intermediate values of F , perhaps some sort of transition
region might be observed.
By first order averaging, we obtain:
à !
A A3 F cos φ
Ȧ = ǫ − − , and (13)
2 8 2
∆ F sin φ
µ ¶
φ̇ = ǫ − + . (14)
2 2A
It will be more convenient for us to study the above “slow flow” (or averaged equations) after
transforming to polar coordinates. Recall that the approximate solution is

A sin(t + φ) = A cos φ sin t + A sin φ cos t = C sin t + D cos t ,

where
C = A cos φ , D = A sin φ .
In terms of C and D, we get
à !
C C 3 ∆D CD2 F
Ċ = ǫ − + − − , and (15)
2 8 2 8 2
à !
D D3 ∆C C 2D
Ḋ = ǫ − − − . (16)
2 8 2 8

15
 (a) ∆ = 1, F=0 (b) ∆ = 1, F=1
5 5
D

D
0 0

-5 -5
-5 0 5 -5 0 5
C C

(c) ∆ = 1, F=1.3 (d) ∆ = 1, F=1.446

5 5
D

0 0

-5 -5
-5 0 5 -5 0 5
C C

(e) ∆ = 1, F=1.53 (f) ∆ = 1, F=2

5 5
D

0 0

-5 -5
-5 0 5 -5 0 5
C C
16
Figure 7: Phase plane for Eqs. 15 and 16. See text for details.
 (a) ∆ = 1, F=1.3
4
x(t)
3 F sin t
x(t) and F sin t

2
1
0
-1
-2
0 20 40 60 80 100 120 140 160 180 200
time, t

(b) ∆ = 1, F=1.446
4
x(t)
3 F sin t
x(t) and F sin t

2
1
0
-1
-2
0 20 40 60 80 100 120 140 160 180 200
time, t

(c) ∆ = 1, F=1.53
4
x(t)
3 F sin t
x(t) and F sin t

2
1
0
-1
-2
0 20 40 60 80 100 120 140 160 180 200
time, t

(d) ∆ = 1, F=2
4
x(t)
3 F sin t
x(t) and F sin t

2
1
0
-1
-2
0 20 40 60 80 100 120 140 160 180 200
17
time, t

Figure 8: Numerical solutions of Eq. 12. See text for details.

 What do Eqs. 15 and 16 say?
We begin by looking at the case ∆ = 0, F = 0. In this case, the equations reduce to
à !
C C 2 D2
Ċ = ǫ 1− − , and (17)
2 4 4
à !
D C 2 D2
Ḋ = ǫ 1− − . (18)
2 4 4

By temporarily calling à !
C 2 D2
E= 1− − ,
4 4
we find
EC ED
, and Ḋ = ǫ
Ċ = ǫ .
2 2
Thus, if E > 0, then in the (C, D) phase plane points move radially outwards; while if E < 0, points
move radially inwards. By the definition of E, we conclude that all points move radially until they
reach the circle C 2 + D2 = 4, or (in terms of the original quantity) A = 2.
Now consider F = 0 but ∆ 6= 0. This causes the trajectories to spiral out (or in) instead of
moving purely radially. The steady state solution now goes round and round on the circle with
radius 2 and centre at the origin. The situation is shown using the (C, D) phase plane in Fig. 7(a).
Now, as we increase F , we find that the circle is shifted and deformed into a smaller closed
curve; and the unstable equilibrium point shifts away from the origin as well. The situation is
depicted in Figs. 7(b) through 7(e). Finally, for even larger values of F , the closed curve shifts to a
point and merges with the unstable equilibrium point, which now becomes stable. The situation is
shown in 7(f). At this point, the solution is periodic, and completely phase locked with the solution
(entrained).
An interesting transition occurs at F ≈ 1.446, when the origin leaves the closed curve. For F
below this critical value, the closed curve encloses the origin, and therefore as the point in the (C, D)
space slowly goes round and round the closed curve, the phase of the oscillations drifts further and
further away from the forcing (gaining or losing 2π with every encircling of the origin). For the
critical value of F ≈ 1.446, there is a point when the oscillation amplitude becomes zero (when
the (C, D) trajectory passes through the origin); and then as the oscillation grows again, there has
been a sudden change in the phase by π. Finally, for F above this critical value, although the point
in (C, D) space goes round and round on its closed curve or limit cycle, the phase angle oscillates
between limits. For such F values, there is weak phase locking and the phase does not drift away.
Finally, as mentioned above, for large enough F , the phase of the oscillation is exactly locked with
the forcing. The above situations are also seen in numerical solutions of the original Eq. 12, as shown
in Fig. 8. The figure shows the steady state solutions, after transients have died out, for the case
ǫ = 0.1 (recall that the averaging procedure is based on ǫ being small). A close look at the time
histories of the vibration response (x(t)) and the forcing (F sin t), for ∆ = 1 and various F values as
given in the figure, shows that the predictions from a study of the averaged Eq. 15 and 16 are borne
out completely. In particular, for F = 1.446, it is seen that the oscillation amplitude periodically
comes to zero, as predicted; and, as shown by arrows towards the left of the figure, the response
3
By getting a slightly slow or fast clock, as appropriate, we slightly change our definition of time so that the forcing
has a period of 2π while the van der Pol oscillator’s unperturbed limit cycle period is slightly different from 2π. We
could have done the reverse, letting the unperturbed limit cycle period be 2π and the forcing period be slightly different
from 2π. But the former is more convenient for averaging.

18
changes from being about π/2 behind the forcing to about π/2 ahead, every time the amplitude
becomes small.
Note that harmonic balance, which only finds periodic solutions and says nothing about stability,
would have only found the steady periodic solution even for small F values, and missed the stable,
modulated solutions shown in Fig. 8 (corresponding to the limit cycles or closed curves of Fig. 7).

10 Modal interactions
The term modal interaction refers to a situation in which energy is exchanged between modes in a
system. Thus, situations in which modal interactions occur are in distinction to the case for linear,
completely diagonalizable systems in which the normal modes do not exchange energy. Here we will
look at an example of modal interactions caused by nonlinear terms. Consider the system shown in
Fig. 2(c). Let the free length of spring k2 be h, as shown; let the equilibrium positions of masses m1
and m2 be at x1 = 0 and x2 = 0 respectively.
We now take m1 = m2 = 1, k1 = k2 = k3 = 1, and h = 1. The equations of motion for the above
system are strongly nonlinear; however, for small displacements, we retain linear and quadratic terms
but drop third order and higher order terms. Then the equations of motion are

x22
ẍ1 = −2x1 + , and (19)
2
ẍ2 = −x2 + x1 x2 . (20)

Note that the linearized system is decoupled, and thus

√ each degree of freedom by itself constitutes
one vibration mode. The natural frequencies are: 2 for x1 and 1 for x2 . Due to the presence of
nonlinearities, however, the modes are coupled and energy exchange can occur between them.
For simplicity, we now add small amounts of forcing and damping to these systems, writing

x2
ẍ1 = −2x1 − cẋ1 + 2 + F sin 2t , and (21)
2
c
ẍ2 = −x2 − ẋ2 + x1 x2 . (22)
3
In the above, the choice of damping is not critical, but it is significant that the forcing frequency is
twice the natural frequency of the x√ 2 mode. Note that the forcing is applied to mass m1 , and the
natural frequency of the x1 mode is 2, so the forcing frequency is not equal or close to the natural
frequency of the forced mode.
On numerically solving the above equations for c = 0.04 and F = 0.06, we obtain the steady
state responses shown in Fig. 9. It is seen that the response of x2 , i.e., the unforced mode, is at its
natural frequency, which is one half of the forcing frequency; moreover, this response has a greater
amplitude than that of the forced mode itself; and finally, the forced mode responds at the forcing
frequency.
Since the unforced mode is damped, it dissipates energy; that energy comes from the forced mode,
showing that the two modes are interacting. This system also provides an example of a motion where
different parts of a system vibrate with different frequencies.

References
[1] Hinch, E. J., 1991. Perturbation Methods, Cambridge University Press.

19
 0.5
x1(t)
0.4
x2(t)
0.3

x1(t) and x2(t) 0.2

0.1

- 0.1

- 0.2

- 0.3

- 0.4
0 5 10 15 20 25 30
time, t
Figure 9: Numerical solutions of Eqs. 21 and 22 with c = 0.04 and F = 0.06.

[2] Verhulst, F., 1990. Nonlinear Differential Equations and Dynamical Systems, Springer-Verlag.

[3] Rand, R. H., 2005. Lecture Notes on Nonlinear Vibrations, version 52. Available online at
http://audiophile.tam.cornell.edu/randdocs/nlvibe52.pdf

[4] Guckenheimer, J., and Holmes, P., 1983. Nonlinear Oscillations, Dynamical Systems, and Bifur-
cations of Vector Fields. Springer-Verlag.

[5] Strogatz, S. H., 1994. Nonlinear Dynamics and Chaos: With Applications to Physics, Biology,
Chemistry, and Engineering. Addison-Wesley.

[6] Drazin, P. G., 1992. Nonlinear Systems. Cambridge University Press.

[7] Rand, R. H. and Armbruster, D., 1987. Perturbation Methods, Bifurcation Theory and Computer
Algebra. Springer, Berlin.

20