Notes on Vibrations: 14:650:443 Fall 2015

c Andrew Norris

November 20, 2015

Contents

1 Single Degree of Freedom System 3

1.1 Single Degree of Freedom System with No Damping . . . . . . . . . . . . . . . . . . 3
1.1.1 The equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1.2 Solutions of the SDOF equation of motion: natural frequency n . . . . . . . 4
1.1.3 Amplitude and phase, period and frequency . . . . . . . . . . . . . . . . . . . 5
1.1.4 Equation of motion from conservation of energy . . . . . . . . . . . . . . . . . 6
1.2 Damped Single Degree of Freedom Systems . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Equation of motion of mkc system . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Underdamped SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 Overdamped SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.2.5 Critically damped SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.6 Energy in a damped SDOF system . . . . . . . . . . . . . . . . . . . . . . . . 17
1.2.7 Springs and dashpots in parallel and in series . . . . . . . . . . . . . . . . . . 18

2 Forced Vibration of a Single Degree of Freed System 20

2.1 The forced SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.1 Special case: F is constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2 Solution for a time harmonic force acting on the mass . . . . . . . . . . . . . . . . . 21
2.2.1 Initial value problem: harmonic forcing starting from rest . . . . . . . . . . . 27
2.2.2 Base excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.3 Rotating unbalance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3 Periodic forcing of a SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3.1 Example: A sawtooth wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Impulse response function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.1 Motivation: delta function and convolution integral . . . . . . . . . . . . . . . 37
2.4.2 Deriving the impulse response function . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Measurement of vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5.1 Decibels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.5.2 Measuring instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3 Multi-degree of freedom systems 45

3.1 Matrices M and K and the equation M u + Ku = 0 . . . . . . . . . . . . . . . . . . 45
3.1.1 Stiffness and Flexibility matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.1 A 3-DOF system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.2 n-DOF chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

1
 2

3.2.3 A simple model of a car . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.2.4 Coupled pendula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.5 The double pendulum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3 Modes of n-DOF Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.3.1 Calculating modes of n-DOF Systems . . . . . . . . . . . . . . . . . . . . . . 50
3.3.2 Example: 3-dof system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.3 Matlabs solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.4 Formal solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3.5 Example: Rigid body modes . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3.6 Damping, proportional damping . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.7 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3.8 Initial value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.4 Forced multi-DOF Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.1 Forced 2-DOF Systems: Vibration Absorber . . . . . . . . . . . . . . . . . . . 57
3.4.2 Time harmonic forcing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.3 Solution in modal coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Impedance (SKIP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5.1 Attached masses and impedance . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.5.2 Impedance and power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.5.3 Impedance of a chain of masses . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.4 Impedance of a bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.5 General structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4 Continuous systems: -DOF 68

4.1 The String . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2 Modes of a string . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 -DOF chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4 Longitudinal modes of a bar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.5 Back to discrete: The finite mk chain. . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.6 Transverse vibration of beams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.6.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.7 Wave motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.7.1 The 1D wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.7.2 Waves and modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

Appendices 80
A Basic equations review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
B Linear algebra 101 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
B.1 Basic matrix algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
B.2 Matlab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
C Lagranges Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
C.1 SDOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
C.2 n-DOF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
C.3 General coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
C.4 Origin of Lagranges equations (extra) . . . . . . . . . . . . . . . . . . . . . . 86
Chapter 1

Single Degree of Freedom System

1.1 Single Degree of Freedom System with No Damping

1.1.1 The equation of motion
The SDOF is a mass and spring, with single coordinate x. The system is in static equilibrium in
Fig. 1.1, i.e. x = 0. Now consider the mass moved a distance x from the equilibrium position

m
x
Figure 1.1: Spring and mass at rest, x = 0

x = 0, Fig. 1.2. As shown, there is no external force acting on the mass, the only force on the mass
is the restoring force of the spring. Therefore the mass cannot be in static equilibrium, it must be
moving.

x
F F
m

x
Figure 1.2: The mass is in dynamic equilibrium under the force F = kx.

Another way of saying it is that the mass is in dynamic equilibrium. The general equation for
dynamic equilibrium of a particle (the mass) is

F~ = m~a (1.1)

3
 4

where in the case considered we can just think of the components of each side in the positive
xdirection. Then1
F = ma where a = x m x F = 0. (1.2)
At the same time, the force is proportional to the spring displacement,

F = kx. (1.3)

Combining these two equations into one gives

m
x + kx = 0. (1.4)

This is the generic form of the equation of motion for a SDOF, x(t). Note that the minus sign in
(1.3) is crucial. it guarantees that the spring provides a restoring force that always pulls or pushes
the mass back towards the static equilibrium position x = 0.

1.1.2 Solutions of the SDOF equation of motion: natural frequency n

Using complex variables
First we simplify the equation of motion so that it only contains one parameter. Divide by m to
get
+ n2 x = 0
x (1.5)
where r
k
n = . (1.6)
m
n is called the natural frequency, for reasons we will see. Try x(t) = Cet where C and are
constants. Then x = Cet and x = 2 Cet , i.e. x
= 2 x, so eq. (1.5) becomes

(2 + n2 ) x = 0. (1.7)

This implies that either x = 0 at all times, which is not of interest (it is just the static equilibrium
solution) or
2 = n2 = in (1.8)
where i2 = 1. The means that can be equal to in or in . Since eq. (1.5) is a linear
ordinary differential equation, we can add solutions to get new solutions. The fact that there are
two possible values of , and two possible values of C, say C1 and C2 . In other words, the most
general solution of eq. (1.5) is
x(t) = C1 ein t + C2 ein t . (1.9)

Using trigonometric functions

We can rewrite the general solution (1.9) in a different way, starting from the identity

ei = cos + i sin (1.10)

use = n t, = n t along with cos() = cos , sin() = sin to write (1.9) in the alternative
form
x(t) = C1 (cos n t + i sin n t) + C2 (cos n t i sin n t). (1.11)
1 dx
The dot notation is used to shorten equations, i.e. x = dt
.
 5

Combining the terms we get

x(t) = A1 cos n t + A2 sin n t (1.12)
where
A1 = C1 + C2 , C1 = 21 (A1 iA2 ),
(1.13)
A2 = i(C1 C2 ), C2 = 21 (A1 + iA2 ).
In summary, eqs. (1.9) and (1.12) give two equally valid and equivalent ways to write the general
solution.

Meaning of the constants C1 , C2 (A1 , A2 )

Putting t = 0 in (1.12) and using cos 0 = 1, sin 0 = 0 gives

x(0) = A1 . (1.14)

Equation (1.14) tells us that the constant A1 is equal to the value of the displacement at t = 0.
This is called an initial condition. Another initial condition is the velocity of the mass at t = 0, i.e.
x(0).
Differentiating (1.14) gives

x(t)
= n A1 sin n t + n A2 cos n t. (1.15)

Putting t = 0 gives
x(0)
= n A2 . (1.16)
In summary, if the initial conditions are

x(0) = x0 , v(0) = v0 , (1.17)

then
A1 = x0 , C1 = 12 (x0 i v0n ),
(1.18)
A2 = v0n , C2 = 21 (x0 + i v0n ).
The solution for x becomes
v0
x(t) = x0 cos n t + sin n t (1.19)
n
which clearly satisfies the two initial conditions at t = 0. Equation (1.19) then gives x(t) for all
values of t 0.

1.1.3 Amplitude and phase, period and frequency

Equation (1.12) can be written in another very useful form,

x(t) = cos n tA1 + sin n tA2

q  
A1 A2
= A21 + A22 cos n t p 2 + sin n t p
A1 + A22 A21 + A22
q

= A21 + A22 cos n t cos + sin n t sin (1.20)

where the angle is defined by

A1 A2
cos = p , sin = p 2 . (1.21)
A21 + A22 A1 + A22
 6

Using the trigonometric identity cos cos + sin sin = cos( ) we get
q
A2
x(t) = A cos(n t ) where A = A21 + A22 , tan = . (1.22)
A1

A is the amplitude, and is called the phase angle. As an example, for the solution (1.19)
s
v2 v0
A = x20 + 02 , tan = . (1.23)
n x 0 n

Some things to note about the eq. (1.22): The maximum and minimum values of x(t) are A and
A. The maximum values are obtained when n t = 2n where n = 0, 1, 2, . . . because cos 2n =
1. The amount of time, T , between successive maxima is called the period and corresponds to
n T = 2, that is, T = 2/n . The inverse of the period, called the frequency (as compared to the
natural frequency n ) is f = 1/T . In summary

2 1 n
T = , f= = period and frequency - undamped SDOF. (1.24)
n T 2

Using units of seconds, the system oscillates 1 cycle in T seconds; in one second it oscillates f
cycles.

1.1.4 Equation of motion from conservation of energy

Consider the SDOF of Fig. 1.1 again. As the mass moves it has kinetic energy for a particle,
T = 12 mv 2 where v = x is the velocity. At the same time, there is potential energy V associated
with the spring, equal to Z x
V = F dx. (1.25)
0
Here F is the force acting on the spring, F = kx, see Fig. 1.2. Hence
Z x
1
V = kxdx = kx2 . (1.26)
0 2
The total energy E is the sum of the kinetic and potential,
1 1
E = T + V = mx 2 + kx2 . (1.27)
2 2
Since the spring force is conservative, the total energy is conserved, which means that it does not
change with t. Hence,
1 1
E = E0 mx 2 + kx2 = E0 (1.28)
2 2
where E0 is a constant which can be identified with the energy at t = 0, for instance, i.e.
1 1
E0 = mv02 + kx20 . (1.29)
2 2
The conservation of energy equation implies dE/dt = 0, or from (1.28)

1 dx 2 1 dx2
m + k = mx
x + kxx
2 dt 2 dt
= (mx + kx) x = 0. (1.30)
 7

Dividing by v gives us the equation of motion (1.4). The important point is that we have arrived
at the equation of motion without using F = ma.
Now consider some generic SDOF system with single coordinate y(t). The kinetic and potential
energy of the system are

1 1 1 1
T = meq y 2 , V = keq y 2 meq y 2 + keq y 2 = E0 (1.31)
2 2 2 2
where meq , keq and E0 are positive constants. The total energy T + V is constant, therefore
proceeding as above we find that

keq
meq y + keq y = 0 y + n2 y = 0 where n2 = . (1.32)
meq

In other words, the equation and motion and the natural frequency can be found directly once we
know the form of T and V , i.e. the equivalent mass and stiffness meq and keq . This approach allows
us to quickly reduce complicated looking systems to standard SDOF form.

Examples
Consider Case (a) in Fig. 1.3. Let keq be the equivalent stiffness, then the potential energy is
V = 12 keq x2 . then it follows that the equation of motion is

keq
x + keq x = 0 n2 =
m . (1.33)
m

Figure 1.3: Four different SDOF systems. All springs have stiffness k, the wheels have moment of
1
inertia I0 = 21 mR2 , the beam I0 = 12 mL2 . Ignore the effects of gravity. (In (b) there should be a
fixed point of rotation in the beam center).

Case (b) involves rotation about the fixed point at the beam center. Let be the angle of
rotation from the equilibrium position (assumed it to be the vertical position for simplicity, it is not
important). The kinetic energy is T = I0 2 . The PE is 2( 21 kx2 ) where the springs are both either
 8

L
compressed or stretched by an amount (distance) x = 2 (using the small angle approximation).
The total energy is therefore

1 1 k
mL2 2 + kL2 2 = E0 n2 = 6 . (1.34)
24 4 m
1
Note in this case meq = 12 mL2 and keq = 21 kL2 . These do not have the same units as the mass
and stiffness for a mass-spring system since they both contain length2 . But the ratio keq /meq has
the correct units of 1/(time)2 , as it should.
In Case (c) let be the angle of rotation from the equilibrium position. The kinetic energy is
T = I0 2 . The PE is 2( 21 kx2 ) where one spring is compressed and the other stretched by x = r
(using the small angle approximation). The total energy is therefore

1 kr2
mR2 2 + kr2 2 = E0 n2 = 4 . (1.35)
4 mR2

Case (d) is just like Case (c) except that x = R, so

1 k
mR2 2 + kR2 2 = E0 n2 = 4 . (1.36)
4 m

1.2 Damped Single Degree of Freedom Systems

1.2.1 Equation of motion of mkc system
The governing equation for a spring-mass-damper with no forcing is

m
x = kx cx m
x + cx + kx = 0, (1.37)

or dividing by m, and using k/m = n2 (eq. (1.6)) and the definition of the parameter by

c
= 2n , (1.38)
m

gives
+ 2n x + n2 x = 0.
x (1.39)

Try x(t) = Aet , x = x, x

= 2 x, so that
 
2 + 2n + n2 x = 0. (1.40)

This can be zero in general only if the bracketed term is zero, and so there are two possible values
for :
p
= n n 2 1 (1.41)

There are three cases that need to be looked at separately:

1. < 1, underdamped
2. > 1, overdamped
3. = 1, critically damped
 9

1.2.2 Underdamped SDOF

p p
Since 2 1p< 0 we have 2 1 = i 1 2 , and hence the two s are complex valued =
n in 1 2 or p
= n id where d = n 1 2 (1.42)
The general solution is a linear combination of the two types of solutions:


x(t) = C1 eid t + C2 eid t en t (1.43)

where C1 and C2 are constants. Equation (1.43) can be written

x(t) = C1 (cos d t + i sin d t) + C2 (cos d t i sin d t) en t

(1.44)
= (C1 + C2 ) cos d t + i(C1 C2 ) sin d t) en t ,

or, by defining the new constants A1 = C1 + C2 and A2 = i(C1 C2 ),

x(t) = A1 cos d t + A2 sin d t) en t . (1.45)

Let us consider the Initial Value Problem (IVP):

x(0) = x0 , v(0) = v0 , (1.46)

where x(t) is given by (1.45) the velocity of the mass v = x follows from (1.45) as


v(t) = (d A2 n A1 ) cos d t (d A1 + n A2 ) sin d t) en t . (1.47)

Put t = 0 in (1.45) and (1.47) and use (1.46):

x0 = A1 , v0 = d A2 n A1 . (1.48)

These are now easily solved for A1 and A2 , and then put back into eqs. (1.45) and (1.47) to give
 (v0 + n x0 ) 
x(t) = x0 cos d t + sin d t en t ,
d (1.49)
 n

v(t) = v0 cos d t v0 + n x0 sin d t en t .
d

We can rewrite x(t) in terms of an amplitude and phase, as we did before with the undamped
system, except now there is the extra en t term,

x(t) = C cos(d t ) en t (1.50)

where, from (1.49),

s
(v0 + n x0 )2 v0 + n x0
C= x20 + and tan = (1.51)
d2 d x 0

We could have found (1.50) and (1.51) by first starting with (1.50) as the assumed form of the
general solution. The actual values of C and for the IVP (1.46) follow by writing the velocity,
from (1.50),  
v(t) = C d sin(d t ) n cos(d t ) en t (1.52)
 10

and then plugging in t = 0 into (1.50) and (1.52):

x0 = C cos (1.53)
v0 = C(d sin n cos ) = Cd sin n x0 (1.54)

and hence
v0 + n x0
C cos = x0 , C sin = (1.55)
d

from which C and follow as in (1.51). Another way of writing x follows from (1.50) and (1.55),

x0 v0 + n x0
x(t) = cos(d t ) en t , where = tan1 (1.56)
cos d x 0

Zero initial velocity

If v0 = 0 then from (1.49)

x(t) = x0 cos d t + p sin d t en t ,
1 2
n (1.57)
v(t) = x0 p sin d t en t .
1 2

After starting at t = 0 with zero velocity, the next time the velocity is zero is after a half period,

i.e. when d t = , at which time x = x0 e 1 2 . This shows that the amplitude is never again

as large as it is at t = 0. The value x0 e 1 2 is called the overshoot.

Zero initial displacement

If x0 = 0 then from (1.49)

v0
x(t) = sin d t en t ,
d

(1.58)
v(t) = v0 cos d t p sin d t en t .
1 2

The first maximum of |x| is the largest value it can be, and it occurs when v = 0, which happens
1 2
when tan d t = . Using relations like cos = 1/ 1 + tan2 and sin = tan cos gives
p
sin d t = 1 2 . Therefore (1.58) gives
p
v0  1 2 
|x|max = exp p tan1 . (1.59)
n 1 2

For a given v0 and n the

maximum displacement depends on damping through the function
 
1 1 2
F () = exp 2 tan , see Fig. 1.4.
1
 11

0.9

0.8

0.7
F
0.6

0.5

0.4

0.3

0.2
0 0.5 1
1.5 2

n
Figure 1.4: The function F () = v0 |x|max for the SDOF with zero initial displacement, see eq.
(1.59)

Period and logarithmic decrement

If we plot x using, e.g. (1.56), it is clear that the period is
2
T =
d
which follows from the fact that the zero crossings are spaced apart by /d . Actually, we need to
be a little bit careful about defining the period for an exponentially decreasing sinusoidal response.
For instance, is the period T also equal to twice the time between subsequent maxima? The answer
is YES, as we can see by using eq. (1.50) to write the velocity:
 
v(t) = C d sin(d t ) + n cos(d t ) en t (1.60)

This in turn can be rewritten using the trig identity sin(a + b) = sin a cos b + cos a sin b as

v(t) = n C sin(d t + ) en t (1.61)

where the additional phase angle is

n
tan = =p (1.62)
d 1 2
which depends only on the damping factor. Equation (1.61) implies in particular that the zeros
of v(t) are also spaced apart by T /2. But the zeros of v correspond to the maxima and minima
of x, and therefore the period T can also be defined by the time between subsequent maxima or
subsequent minima of x(t). In summary,

2
T = period of damped SDOF. (1.63)
d
p
Since d = n 1 2 it follows from (1.24) and (1.63) that the ratio of the damped to un-
damped periods is (assuming < 1)
Tdamped n 1
= =p 1 (1.64)
Tundamped d 1 2
 12

with equality when there is no damping. The effect of adding a damper to a spring-mass system
therefore always increases the period. If the damping is small the increase will be very small, e.g.
if = 0.05 then Tdamped /Tundamped = 1.001.
Note that, for any value of t, we have

x(t + T ) = x(t) en T (1.65)

This clearly displays how the amplitude is decreasing because of the damping (when = 0 we have
x(t + T ) = x(t)). More generally, after n periods, where n = 1, 2, 3, . . .,

x(t + nT ) = x(t) enn T (1.66)

Note that
n
n T = 2 = 2 p . (1.67)
d 1 2
This only involves , which means that the damping factor can be found from the ratio of two
consecutive amplitudes x(t + T )/x(t), or more generally from the ratio of two amplitudes separated
by n periods in time: x(t + nT )/x(t). At the same time, eq. (1.65) and (1.66) imply

x(t + T ) 1 x(t + nT )
n T = log = log . (1.68)
x(t) n x(t)

The logarithmic decrement is defined as

x(t + T ) 1 x(t + nT )
= log = log (1.69)
x(t) n x(t)

and hence, = n T , or equivalently, using (1.67) shows that the logarithmic decrement depends
only on the damping factor:

= 2 p . (1.70)
1 2

The damping factor can be expressed in terms of , from (1.70),

=p (1.71)
(2)2 + 2

This suggests a way to measure the damping factor by measuring the logarithmic decrement. In
general the latter is given by
1 x(0)
= log (1.72)
n x(nT )
which can be measured by observation of the damped oscillation over several n cycles. If the
system is weakly damped then typically a large value for n will be required (n > 10) in order to
see significant change from x(0) to x(nT ). Also, for small damping will be small ( 1), and
therefore from (1.70) is also small. Hence, by (1.71)

= for small damping. (1.73)
2
 13

1.2.3 Overdamped SDOF

When > 1, the two roots given by (1.41) are real. Thus, let
p p
1 = n n 2 1, 2 = n + n 2 1, (1.74)
p p
then it is clear that 1 < 0. Also, 2 = n ( 2 1), and > 2 1, so 2 < 0 also. Thus,
they are both negative
1 < 2 < 0
It can easily be shown that in fact
1 < n < 2 < 0 (1.75)
The general solution
x(t) = A1 e1 t + A2 e2 t (1.76)
is a sum of two exponentially decreasing terms, and therefore the solution must decay to zero. It
could initially be increasing or decreasing in magnitude, depending on whether or not A1 and A2
are of the same or opposite signs.
The IVP (1.46) can be solved by first writing the velocity, from (1.76),

v(t) = A1 1 e1 t + A2 2 e2 t (1.77)

and hence the initial values are

x0 = A1 + A2 , v0 = A1 1 + A2 2 (1.78)

or,
        1  
1 1 A1 x0 A1 1 1 x0
= = (1.79)
1 2 A2 v0 A2 1 2 v0

Using the general relation

 1  
a b 1 d b
= (1.80)
c d ad bc c a

we get
    
A1 1 2 1 x0
= (1.81)
A2 2 1 1 1 v0

Hence,
2 x 0 v 0 1 x0 + v0
A1 = p , A2 = p , (1.82)
2n 2 1 2n 2 1
or
x0 (n x0 + v0 ) x0 (n x0 + v0 )
A1 = p , A2 = + p , (1.83)
2 2n 2 1 2 2n 2 1
Equations (1.74), (1.76) and (1.83) imply that the solution to the IVP is

 x0 n 2 1t
2
(n x0 + v0 ) n 2 1t
2

x(t) = e + en 1t + p e en 1t en t (1.84)
2 2n 2 1
 14

Using
1 1
cosh x = (ex + ex ), sinh x = (ex ex ) (1.85)
2 2
gives
 p (n x0 + v0 ) p 
x(t) = x0 cosh n 2 1t + p sinh n 2 1t en t . (1.86)
n 2 1

We could have derived (1.86) by starting with the assumed form

 p p 
x(t) = A1 cosh n 2 1t + A2 sinh n 2 1t en t (1.87)
p p
since cosh n 2 1t en t and sinh n 2 1t en t are an alternative pair of linearly inde-
pendent solutions of the general equation (1.39).
Note that (1.86) is formally the same as the result for the underdamped system, eq. (1.49),
which can be seen by writing the latter as
 p (n x0 + v0 ) p 
x(t) = x0 cos n 1 2 t + p sin n 1 2 t en t solution for < 1. (1.88)
n 1 2

The connection between (1.86) and (1.88) is then evident from the identities

cos(iz) = cosh(z), sin(iz) = i sinh(z) (1.89)

for any real or complex number z.

1.2.4 Examples
Example 1
Find the damping factor necessary for an overshoot of 15%.

Solution: The overshoot x0 e 1 2 is 15% the value of x0 if e 1 2 = 0.15. Taking the
natural logarithm 2 = log 0.15. We can solve 2 = a by squaring 2 = (1 2 )a2 / 2
1 1
which gives
1
=q
.
2
1+ a

In this case = 0.5169.

Example 2
If m = 1 kg, c = 2 kg/s, and k = 10 N/m, calculate the values of and n . Is the system under-
or over-damped? p
Solution: n = 10/1 = 3.163 rad/s, = c/(2mn ) = 2/(2 3.163) = 0.3163, so the SDOF is
underdamped.

Example 3
Find the solution of x
+ 4x + x = 0 with x(0) = 1, v(0) = 0.
Solution: n = 1 and = 2, so it is overdamped. The general solution is

x(t) = A1 e2t 3t
+ A2 e2t+ 3t
v(t) = A1 (2 3)e2t 3t
+ A2 ((2 + 3))e2t+ 3t
 15

and with the given initial conditions:

A1 + A2 = 1, A1 (2 3) + A2 ((2 + 3)) = 0

Solving for A1 and A2 :

32 3+2
A1 = , A2 =
2 3 2 3
and so
x(t) = 0.077e3.73t + 1.077e0.27t
The same result follows from the general solution (1.84).

Example 4
It is known that the mass of a device must be between 2 and 3 kg. The support system and
materials are such that the stiffness of the
p system, modeledp as a SDOF, is 200 N/m. The natural
frequency is therefore restricted to n 200/2 and n 200/3, that is 8.16 n 10 rad/s.
This system is being designed to be subject to zero initial displacement and initial velocity always
less than 16 cm/s in magnitude. Choose a damping system such that the amplitude of vibration
is always less than 1 cm, that is find the damping c. Solution: From eq. (1.59), the maximum
amplitude is
v0
|x|max = F () (1.90)
n
where F () is plotted in Fig. 1.4. For v0 = 16 cm/s and n = 8.16 we have v0n = 1.96 cm. If |x|max
is to be less than 1 cm then we need F () < 1/1.96 = 0.51. From Fig. 1.4, this is satisfied if
is larger than about 0.6. We could get a precise estimate by solving F () = 0.51, but that is not
necessary. The dimensional damping coefficient c is related to the damping factor by c = 2 km,
and taking the maximum mass m = 3 kg, implies that c 29.39 kg/s ensures that the amplitude
of vibration is always less than 1 cm.

Pendulum with dashpot

Consider a bar of length L, mass m, suspended from O. The other, free end is connected to a
horizontally positioned dashpot. The equation for the moment balance about the support O is, for
small angular displacement ,
L
I0 = mg L(cL)
(1.91)
2
2
where I0 = m L3 . The first term on the right is moment of the weight mg with moment arm
L L
2 sin 2 . The second term is the moment of the dashpot force cv with moment arm L, where
Both moments are in the sense opposite to the motion, hence the minus
the velocity is v = L.
signs.
Dividing by I0 ,
cL2 mgL
+ + =0 (1.92)
I0 2I0
Therefore,
mgL 3g
n2 = = (1.93)
2I0 2L
cL2 3c
2n = = (1.94)
I0 m
 16

1.2.5 Critically damped SDOF

When = 1 the two roots of (1.41) become identical: 1 = n , 2 = n . It is useful to draw
a picture of how the roots behave for different values of the damping factor - see Figure.
Both 1 and 2 are complex-valued for underdamping < 1, and it follows that the magnitude
of both is p
|| = 2 n2 + n2 (1 2 ) = n (1.95)
and so the underdamped rots lie on a circle in the complex plane of radius . The real parts are
both negative, and so they lie to the left of the imaginary axis. For overdamped systems, we saw
in (1.75) that they are both negative, and positioned as shown in the Figure. As 1 the pair 1
and 2 both coalesce on n , and this is true whether we think of approaching = 1 from above
or below.
The general solution of the IVP can also be considered as a limit of the underdamped solution,
eqs. (1.49), (1.50) or (1.56), or as a limit of the underdamped solution of eqs. (1.84) or (1.86).
One way to find the IVP solution for the critically damped SDOF is to take any one of these and
take the appropriate limit. We should get the same result no matter how we do it, so we may as
well take one that looks relatively easy. For instance, suppose we consider (1.86) in the limit as
1. There are three terms there that can be considered separately:
p (n x0 + v0 ) p
x0 cosh n 2 1t, p sinh n 2 1t, en t . (1.96)
n 2 1
p
In the limit as 1 the last one becomes simply en t and the first is just x0 , because 2 1 0
and cosh 0 = 1. We need to be more careful about the other term, because the numerator tends to
zero, as does the denominator. But it is easy to show that
p
sinh n 2 1t
lim p =t (1.97)
1 n 2 1

as can be seen using the Taylor series for sinh x = x + 16 x3 + . . .. Hence, the IVP solution is
 
x(t) = x0 + (n x0 + v0 )t en t (1.98)

It can be easily checked that this indeed satisfies the ICs of eq. (1.46).
The surprising feature of (1.98) is the term proportional to ten t . One way of seeing how this
arises is to go back to the general equation (1.113) for = 1:

+ 2n x + n2 x = 0.
x (1.99)

This can be written

d
(x + n x) + n (x + n x) = 0 (1.100)
dt
or
dy
+ n y = 0 (1.101)
dt
where y = x + n x. Equation (1.101) is a first order ordinary differential equation (ODE) that has
the general solution y = Ben t , or

x + n x = Ben t (1.102)
 17

Multiply both sides by en t and combine the terms on the left:

d
(xen t ) = B (1.103)
dt
with solution
xen t = A + Bt x = Aen t + Bten t (1.104)
In other words, the two linearly independent solutions are en t and ten t . This is a common
phenomenon, and occurs when the solutions of a second order (or higher) ODE are no longer
linearly independent. In summary, the general form of the solution for critical damping is

x(t) = (A + Bt)en t (1.105)

and the IVP solution is (1.98).

1.2.6 Energy in a damped SDOF system

We saw in Section 1.1.4 that the mechanical energy E of eq. (1.27) is constant in the undamped
SDOF system. When damping is present the mechanical energy must decrease with time. In order
to see this we can reverse the steps that led to the result (1.30), as follows. First, starting with the
equation of motion for the damped system, eq. (1.37), we multiply it by x to get

x + cx 2 + kxx = 0.
mx (1.106)

Noting that dx2 /dt = 2xx and dx 2 /dt = 2x

x, we can rewrite (1.106) as

d 1 1

mx + kx2 + cx 2 = 0. (1.107)
dt 2 2
Alternatively, this is
dE
= cx 2 (1.108)
dt
and the right hand side term must be negative since c > 0 when there is damping. This means
that the mechanical energy must decrease.
Integrating the identity (1.108) over one period, from t = 0 to t = T = 2/d gives
Z T
E(T ) E(0) = c x 2 dt. (1.109)
0

This provides another way of interpreting the damping: it measures the amount of energy lost per
cycle. Let us consider the case in which the damping is small, 1. Then over one cycle the
exponential decay is small, and we can reasonably approximate x using the undamped solution
x = A cos(n t ) where A and are constants, see eq. (1.22). Then, using the fact that the
integral of sin2 u over one period is ,
Z 2/n
E(T ) E(0) = cA2 n2 sin2 (n t )dt
0
Z 2
= cA2 n sin2 (u )du (u = n t)
0
= cA2 n . (1.110)
 18

At the same time, the energy of the undamped SDOF system is E = 21 mA2 n2 sin2 (n t ) +
1 2 2 2 2 2 2
2 kA cos (n t ), or using k = mn , E = mA n = kA . Define the loss of energy over one
cycle: E = E(0) E(T ), then (1.110) gives
cE
E = = 2E (1.111)
mn
where we have used eq. (1.38). Therefore, since we are considering the case of small damping, we
can use (1.73) to get
E
= 2 . (1.112)
E
This provides another way of interpreting the logarithmic decrement: it is the relative loss of
energy over one cycle in the small damping case. One could generalize this result to larger values
of damping but the algebra is more complex. The concept of logarithmic decrement is most useful
when there is small damping; hence the identification of as the relative decrease in energy per
cycle is always a good way to think of it.

1.2.7 Springs and dashpots in parallel and in series

Let us briefly consider some other examples of damped SDOF systems.

Springs in parallel and in series

If two springs, k1 and k2 , are placed in parallel, the effective spring has stiffness k = k1 + k2 . This
follows from the fact that the two have equal displacement x and the force due to each one is k1 x
and k2 x, so the total force is (k1 + k2 )x kx.
If the same two springs are placed in series, in this case the force on each is the same, say F ,
and the displacements are x1 = F/k1 , x2 = F/k2 . The total displacement is x = x1 + x2 = F/k.
The effective stiffness is
1 1 1 k1 k2
= + k=
k k1 k2 k1 + k2

Dashpots in parallel and in series

The same type of relations hold for dashpots in parallel and in series. Thus two dashpots c1 and
c2 in parallel have effective damping coefficient c = c1 + c2 . The same two dashpots in series gives
effective damping coefficient c = c1 + c2 . The same two dashpots in series gives effective damping
1 1 1
c = c1 + c2 .

Dashpots and springs in parallel and in series

What happens when a spring and a dashpot are combined? If the spring and dashpot are in parallel,
then the forces add. One force is kx and the other is cx.
The total force is F = kx + cx,
and we
get the equation of motion mx = (kx + cx),
or the usual

+ 2n x + n2 x = 0
x (1.113)
q
k
with n = m and
c c
= =
2mn 2 km
This is the standard example as we have considered it. It is generally more realistic, since it is
more practical and easier to place springs and dashpots in parallel, rather than in series (next).
 19

The situation is a little different than all of the previous cases if the spring and dashpot are
in series. Then the forces on each is the same, say F , the spring displacement is x1 = F/k,
and the dashpot velocity is v2 = F/c. The total velocity is v = v1 + v2 where v1 = x1 . Hence,
v = F /k F/c. The equation of motion is

mv = F

which can be written in terms of the force alone as

d F F
m m
m =F F + F +F =0 (1.114)
dt k c k c
Dividing by m and multiplying by k, gives
k k
F + F + F = 0 F + 2 n F + n2 F = 0 (1.115)
c m
q
k
where n = m as usual but the damping factor is now

k km
= =
2cn 2c
Equation (1.115) has the same general form as the damped SDOF equation but it is for the force
not the displacement. The effective damping factor has a quite different form. Note that it is
inversely proportional to the damping coefficient c. Also, in order to calculate the displacement we
need to do a little bit more calculation, which we leave for now.
Summary: when we have sets of springs alone in series or in parallel, or in more complex
configurations, we can always reduce it to a single effective spring constant. The same goes for
dashpots. When we mix the two then it can be complicated. For a spring and dashpot in parallel
we get the classical equation for a damped SDOF. When they are in series we get a similar looking
equation but for the force, not the displacement.
Chapter 2

Forced Vibration of a Single Degree of

Freed System

2.1 The forced SDOF

So far we have considered SDOF systems, damped and undamped, which oscillate on the basis
of Initial Conditions (ICs). Any spring-mass-damper or equivalent system exhibits free vibration
when given an initial displacement and/or velocity. In the presence of damping all free vibration
is transient, i.e it lasts for a finite time, depending on the amount of damping.
Vibration problems in Mechanical and Aerospace Engineering and all other Engineering disci-
plines do not usually arise from ICs but result from forcing. Typically, the forcing is associated
with some energy source, e.g. a motor, or some other regular source of energy input. The vibration
can be desirable or unwanted, and depending on the nature and design goal, the job of the vibration
engineer can be different. An example of desirable vibration energy is music, where the source is
either analog or synthetic, but the goal is to produce or reproduce the desired vibration - in this
case ultimately the vibration of the tympani in the inner ear - with the best precision possible.
Undesirable vibration is unfortunately more common, and all engineering design needs to take into
account the possibility of vibration, and if necessary its suppression. An example of this that is
well known is the Tacoma Narrows bridge where wind could excite a mode of oscillation of the
bridge to the extent that it caused the structure to fail. More commonly, we experience unwanted
vibration in everyday circumstances. Think of examples.
A very common example of forced vibration is where the forcing is periodic - or harmonic -
as occurs in many if not most examples. The forcing typically comes from some external source,
such as the periodic forcing from a motor, which is very typical with electrical motors. Automobile
engines are another very important example, where the engine acts as a periodic source of many
different frequencies, depending on the RPM of the engine. Cars and similar engineering device
must deal with the possibility of a wide range of harmonic forcing - over a wide frequency range.
Other systems, such as electric motors, have more well defined frequencies. In general, any energetic
system exhibits forcing at narrow or broad frequency bands. Here we consider the generic problem
of the mass-damper-spring system - the m-c-k system - subject to forcing at a single frequency.
We imagine a mass, with spring k and dashpot c in parallel, connected to ground, i.e. a
rigid block. The mass is subject to the force of the spring and dashpot, Fspring = kx and
Fdashpot = cv, respectively. In addition there is an external force F acting, so that the eq. of
motion is
mx = kx cx + F. (2.1)
The force F can be imagined as acting in the positive xdirection, but F can be positive or

20
 21

negative. We rewrite (2.1):

m
x + cx + kx = F. (2.2)

2.1.1 Special case: F is constant

The simplest situation is when the force F = is constant. Let us briefly consider this case. The
main idea is that the constant force shifts the static equilibrium position, and the resulting equation
is the same as that for free vibration. Thus, when there is no motion (x = x = 0) we find x = F/k.
Redefine y = x F/k, then y = 0 in the static situation. But x = y and x = y (because x and y
only differ by a constant) so (2.2) becomes

F
m
y + cy + ky = 0 for y = x , F = constant (2.3)
k
This is the equation for free vibration, already discussed.
As a practical and common example consider the spring-mass system of Figure 2.1 subject to
the constant force of gravity. Here x = 0 describes the equilibrium position without gravity (no
spring extension). In reality Fk = mg, indicating the equilibrium position with gravity is at
x = mg k , which is the same as y = 0.
The system as shown has no damping, but if we imagine damping present then it is clear that
any initial conditions of the system will end up with the mass coming to rest at y = 0. As a
thought experiment imagine attaching the mass to the spring and then letting it go from x = 0:
what happens? We can think of this as an IVP with initial position y0 = mg k and initial velocity
y 0 = 0. The solution for the subsequent motion of the mass follows from eq. (1.57).

x
Fk
y

mg
Figure 2.1: A spring-mass system subject to the constant force of gravity.

We now consider the case where F is not constant but periodic.

2.2 Solution for a time harmonic force acting on the mass

We assume that the forcing is periodic:

F = F0 cos t (2.4)
 22

Here is the frequency of the applied forcing, or just the forcing or drive frequency. It is considered
an an independent quantity, that is, it is not related to the frequency of the oscillator, whether n or
d . We will think of as free tuning parameter, that could take on any value, and sometimes does.
We will be interested in how the response depends not only on F0 but on the variable frequency .
As we will see, things change considerably for different drive frequencies .
Equations (2.2) and (2.4) give

m
x + cx + kx = F0 cos t. (2.5)

There are several ways to solve this. Here we use complex numbers, based on the fact that

F0 cos t = Re F0 eit

so
x + cx + kx = Re F0 eit .
m (2.6)
The idea is simple but maybe subtle: let

x = Re z (2.7)

so that (2.6) becomes

z + cz + kz} = Re F0 eit .
Re {m (2.8)
This is true if we drop the Re, and require z to satisfy

z + cz + kz = F0 eit .
m (2.9)

We now have an equation where the RHS has a simple exponential form. In such cases it makes
sense to look for a particular solution that is of the same exponential form, so we try

z(t) = Aeit , (2.10)

so that z has the same frequency as the drive or forcing frequency. This is reasonable. Consequen-
tially, (2.9) becomes

2 m + ic + k Aeit = F0 eit . (2.11)
We can divide both sides by eit to get the algebraic equation:


k 2 m + ic A = F0 (2.12)

or
F0
A= 2
. (2.13)
k m + ic
To summarize so far: using (2.7), (2.10) and (2.13),

 F0 eit
x(t) = Re . (2.14)
k 2 m + ic
What does this mean?
One way to rewrite (2.14) is

F0
x(t) = Re A eit , A= . (2.15)
k 2 m + ic
 23

A is a complex number, which we can write as1

A = |A|ei (2.16)

where
F0 c
|A| = p , = tan1 . (2.17)
(k m)2 + (c)2
2 k 2 m

Combine (2.16) and (2.15):

x(t) = Re |A| eiti , (2.18)

or
x(t) = |A| cos(t ). (2.19)

This is in a form that we can now interpret. The amplitude is |A| and the response has the same
frequency as the drive but with a phase shift .
Remember,(2.19) is only a particular solution of (2.5), one that gives us the correct RHS. In
addition, we always can add to this the general solution of the homogeneous equation (F = 0), i.e.
x(t) = C cos(d t ) en t , so that the general solution of the inhomogeneous equation (2.5) is
the sum of the particular and the homogeneous solutions:

x(t) = |A| cos(t ) + C cos(d t ) en t (2.20)

Interpretation of the particular solution

It is important to note that both the amplitude |A| and phase depend on the drive frequency .
We now examine this dependence. First, let us write them in terms of the parameters and n ,
or d . Thus,
k c


k 2 m + ic = m 2 + i = m n2 2 + 2in
m m
k c
using m = n2 and m = 2n . Hence, the A in eq. (2.15) becomes

F0 1
A= 2 2
(2.21)
m n + i2in

The magnitude is
F0 1
|A| = p (2.22)
m (n )2 + (2n )2
2 2

and the phase (A = |A|ei ) is

2n
= tan1 (2.23)
n2 2
Let us consider the amplitude and how it varies with drive frequency. Rewrite it as

F0 1 F0 1
|A| = q = q (2.24)
mn2 (1

2
) 2 + 2

2 k (1

2 2 2


n n n ) + 2 n

1
Let z = a + ib, then z = a2 + b2 ei where tan = b/a. Hence 1/z = 1/ a2 + b2 ei . Eq. (2.15) follows with
z = k 2 m + ic.
 24

Apart from the term F/k, the amplitude only depends on the ratio /n and the damping factor
. Both of these are non-dimensional, and we can get a good idea of all possibilities by looking at
the plot of
k 1
B |A| =q
2 , r (2.25)
F0 n
(1 r2 )2 + 2r

for 0 < r < and for different values of , see Fig. 2.2.

3
6 10

5
2
10
4
B B
1
3 10

2
0
10
1

1
0 10
0 1 2
/n 3 4 0 0.5 1
/n 1.5 2

(a) Top to bottom: =.1, .2, .3, .4, .5, 1, 2, 4 (b) =.0001, .001, .01, .1

Figure 2.2: The magnification factor B of eq. (2.25). Note the log scale for (b).

It is clear that there is a frequency at which the amplitude is maximum. We can find this by
differentiating B of (2.25) with respect to r and setting it to zero. Alternatively, the maximum of

2
B will be where the minimum of (1 r2 )2 + 2r occurs. Setting the derivative to zero:
d
2
[(1 r2 )2 + 2r ] = 2r[2(1 r2 ) + 4 2 ] = 0
dr
which occurs if r = 0 (which we can ignore as it is = 0) or
p
r= = 1 2 2 (2.26)
n
Plugging in this value:

2
(1 r2 )2 + 2r = 4 2 4 4 = 4 2 (1 2 )

and so the maximum amplitude is

1 p
Bmax = p at = n 1 2 2 (2.27)
2 1 2
This is a very interesting result. It says that the maximum possible amplitude depends only on
the damping factor . What happens when there is no damping? We can understand this limit by
taking small damping, in which case
1
Bmax for small damping ( 1) (2.28)
2
 25

That is - it becomes unbounded.

Equation (2.26) implies that the maximum moves towards zero, and equals r = 0 for =
1/ 2 = 0.707. For damping larger than this value there is no maximum (or the max is at = 0).
These features are best seen using a plot.
What about the phase?
c 2n 2r
A = |A|ei = tan1 2
= tan1 2 2
= tan1 . (2.29)
k m n 1 r2

The response to time harmonic forcing can be best understood by considering the three cases in
which r = /n is much less than one, one, and much greater than one. Thus:

F0
Low frequency r1 B1 |A| 0,
k

1 F0
Resonance r=1 B= |A| = , (2.30)
2 2k 2

1 F0
High frequency r1 B |A| .
r2 m 2

At low frequency the motion is in phase with the driving force. At high frequency it is
completely out of phase with the driving force2 And at resonance, the motion is 90 out of phase.
Note- there is no maximum if
1
= 0.707 . . .
2
Summary: The particular solution for the forced SDOF system (2.5) is

F0 cos(t )
x(t) = q . (2.31)
k
2
(1 )2 + 2 2


n n

As the particular solution to the equation of motion this does not take into account the homogeneous
solution, that is required to satisfy the initial conditions, see eq. (2.20). However, it should be
realized that the additional homogeneous solution in (2.20) has the en t factor, which means
that whatever effect the initial condition have, it eventually dies away, and we are left with the
steady state3 solution of (2.31). In other words, no matter what the initial conditions are, the
motion eventually reaches the steady state motion.

Example
A machine, m = 25 kg, rests on an elastic foundation. An oscillating force of magnitude 25 N is
applied to the machine at different drive frequencies. The maximum steady state response is found
to be 1.3 mm and occurs when the forcing has period of 0.22 s.
Find the equivalent stiffness and damping ratio of the foundation.
Given: m = 25 kg, Amax = 1.3 mm, F0 = 25 N, = 2/0.22 = 28.6 rad/s.
Find: k,
2
A phase shift = means that the forcing F = F0 cos t always has the opposite sign of the displacement x
given by (2.19).
3
Steady state is here synonymous with time harmonic.
 26

p
(a) Max occurs at r = /n = 1 2 2 , so

28.6
n = p
1 2 2

(b)Max amp is
F 25
p0 = 1.3 103 m 1.3 103 = p
k2 1 2 k2 1 2 2

Use k = mn2 , (a) and (b) imply

25(1 2 2 ) 1 2 2
1.3 103 = p p = 1.066
25(28.6)2 2 1 2 2 1 2

Simplifying
4 2 (1 2 )(1.066)2 = 1 4 2 + 4 4

4 4(1 + (1.066)2 ) 2 4(1 + (1.066)2 ) + 1 = 0

4 2 + 0.117 = 0
r
2 1 1
= 0.117
2 2
= 0.368, or 0.930

but = 0.930 > 1 does not give a maximum, so

2

= 0.368

Go back:
28.6
n = p = 33.5 rad/s
1 2 2
and so
k = mn2 = 2.81 104 N/m

Matlab implementation

The simplest way to compute the solution for time harmonic forcing is to use complex numbers.
The expression (2.14) can be written very simply in Matlab. More generally, suppose

m
x + cx + kx = F, F = F1 cos t + F2 sin t. (2.32)

We can write the forcing term as the real part of a complex forcing using the fact that cos = Reeit ,
sin = Re ieit . Therefore, F = Re(F1 iF2 )eit and the particular solution for the forcing
becomes
 (F1 iF2 ) eit
x(t) = Re
. (2.33)
m n2 2 + i2n
 27

2.2.1 Initial value problem: harmonic forcing starting from rest

Suppose the harmonic forcing (2.4) is turned on at time t = 0. Then the most general form of
the response is, see (2.20),
x(t) = |A| cos(t ) + C cos(d t ) en t for t 0 (2.34)
where |A| and are given by (2.17) in terms of the applied force magnitude F0 and the drive
frequency . The remaining two parameters C and in (2.34) are determined by the initial
conditions x0 and v0 of the system at t = 0.
Here we consider the important special case in which the system starts from rest: x0 = 0,
v0 = 0. It is straightforward to use (2.34) to show that these ICs translate into the following two
conditions for C and :
C cos = |A| cos ,
|A| cos + C cos = 0,
n (2.35)
|A| sin + d C sin n C cos = 0, C sin = |A| sin |A| cos .
d d
The exact solution for t 0 is therefore
cos  
n
x(t) = |A| cos(t ) |A| cos(d t ) en t where = tan1 + tan . (2.36)
cos d d

Example: undamped system

For the case of a forced undamped system ( = 0, d = n ) starting from rest, the solution (2.36)
simplifies since = 0 and hence = 0. Also, using (2.21) with = 0, gives
F0 cos t cos n t
x(t) = for t 0. (2.37)
m n2 2
It is easy to check that x(0) = 0 and v(0) = 0, and that the displacement (2.37) satisfies eq. (2.5)
with zero damping (c = 0). This solution can be rewritten
F0 cos rn t cos n t
x(t) = for t 0, where r = . (2.38)
k 1 r2 n

15 10

10
5
0

k x(t)F0

5
k x(t)F

0 0

5
5
10

15 10
0 2 4 6 8 10 0 2 4 6 8 10
t t
(a) r=.2 (blue), .6 (green), .9 (red) (b) r=1.1 (blue), 2 (green), 5 (red)

Figure 2.3: The response of an undamped system under harmonic forcing eq. (2.38) for different
values of the drive frequency, with n = 2.
 28

The dependence of the solution on r is shown in Fig. 2.3. It is clear from the curves for r = 0.9
and r = 1.1 that something special happens when the drive frequency is close to the system
natural frequency n .

40

30

20

10

0
k x(t)F
0

10

20

30

40
0 2 4 6 8 10
t

Figure 2.4: The response of an undamped system under harmonic forcing eq. (2.38) for = 1.001n
with n = 2.

The limiting solutions that occurs when n is clear from Fig. 2.4. Starting at x = 0 at
t = 0, the amplitude of x grows linearly with time, while it oscillates with the natural frequency.
The exact form of the solution when = n can be obtained from(2.38) by taking the limit as
r 1 by LHopitals rule:

d
F0 dr cos rn t cos n t)
= n : x(t) = d

at r = 1. (2.39)
k dr 1 r
2

This gives

F0
= n : x(t) = t sin n t for t 0. (2.40)
2mn

Check that this satisfies eq. (2.5) with zero damping (c = 0).
Another way to look at the solution (2.37) is to use the trigonometric identity

1 1
cos cos = 2 sin ( ) sin ( + )
2 2

to express it as

2F0 sin 12 (n )t sin 12 (n + )t

x(t) = for t 0. (2.41)
m n n +
 29

10

0
k x(t)F
0

10
0 10 20 t 30 40 50

Figure 2.5: The response of an undamped system under harmonic forcing eq. (2.38) or eq. (2.41)
for = 1.1n with n = 2.

Figure 2.5 plots the response for r = 1.1 (see Fig. 2.3b) for 50 cycles of the oscillator (which has
unit period). The two frequencies 12 (n + ) = 1.05n and 21 (n ) = 0.05n are evident in fast
and the slow oscillation, respectively. This effect, a slow modulation of a much faster oscillation
is common, e.g. the phenomenon of beats where the sound from two tuning forks have slightly
different frequencies. The two sounds interfere constructively and destructively to give rise to a
single acoustic sound of the form shown in Fig. 2.5.
Note that the maximum amplitude of (2.41) occurs when both the sin functions reach the
maximum (unity) at approximately the same time. It follows that
F0
|xmax | . (2.42)
|r 1|k

This agrees with the maximum amplitude of 10 Fk0 in Fig. 2.5. Note that it takes about |r1|
1
cycles
to reach the maximum amplitude.
In summary, when the drive frequency equals the natural frequency and there is no damping,
then the system amplitude increases with time. In principle the amplitude grows without bound,
but in practice some physical effect limits the amplitude. The solution (2.40), shown in Fig. 2.4,
explains how the particular solution for harmonic forcing can become infinite, see Fig. 2.2b. In
reality, it never reaches the steady state, or put another way, it takes an infinite time to become
infinite in amplitude.

2.2.2 Base excitation

General solution
In many engineering situations the external forcing on a SDOF system arises from the motion of
the base, rather than a force applied directly to the mass. In order to model this situation, we
consider a mck SDOF system, where the position of the base is allowed to vary with time. Let x
still represent the position of the mass, but now with respect to a fixed point (e.g. the position of
the base if it is not moving). The base itself is at position y(t), so that y = 0 would be the case when
the base is fixed. The restoring forces on the mass therefore depend on x y rather than x alone.
That is, the spring exerts a force Fspring = k(x y) and the dashpot force is Fdashpot = c(x y).
The equation of motion of the mass is therefore

m
x = k(x y) c(x y)
+F (2.43)
 30

where F is the applied force acting directly on the mass. For the present we assume F = 0, so the
equation of motion can be written

m
x + cx + kx = cy + ky (2.44)

Consider harmonic motion of the base:

y = Y sin b t (2.45)

where Y (constant) is the maximum amplitude, and b is the frequency of the base motion. Eq.
(2.44) becomes

m
x + cx + kx = cb Y cos b t + kY sin b t
= F0 cos(b t 0 ) (2.46)

where
p k
F0 = Y k 2 + (cb )2 , 0 = tan1 (2.47)
cb
Thus, the motion of the mass, described by eq. (2.46), is the same as if the base were fixed
and the mass subject to a harmonic forcing of amplitude F0 at frequency b , with a phase shift 0 .
Using k = mn2 and c = 2mn , allows us to rewrite the amplitude and phase as
p n
F0 = Y mn n2 + (2b )2 , 0 = tan1 (2.48)
2b

We can write down the particular solution for the mass motion using (2.19), and taking into account
the phase shift and the special form of the forcing:

F0 cos(b t 0 )
x(t) = q
k
2
(1 nb )2 + 2 nb

2
s
n2 + (2b )2
= Y n
2 cos(b t 0 ) (2.49)
(n2 b2 )2 + 2b n
or
x(t) = X cos(b t 0 ) (2.50)
where s
1 + (2r)2 b
X=Y
2 , r= . (2.51)
(1 r2 )2 + 2r n

The ratio X/Y is called the transmissibility ratio. It clearly shows a resonance behavior, partic-
ularly for small damping. For small values of , we can easily see that the maximum transmissibility
ratio will be 1/(2). Hence, the motion of a lightly damped system can be very sensitive to the
motion of its base. This is the fundamental reason for vibration isolation of lightly damped systems.

Relative motion
The displacement of the mass relative to the base, z = x y, is sometimes of interest, e.g. in
vibration measuring instruments, see Section 2.5.2. The equation is, from (2.44),

m
z + cz + kz = m
y for z = x y. (2.52)
 31

For the base motion (2.45) this becomes

z + cz + kz = mb2 Y sin b t.
m (2.53)

This is just like eq. (2.4) withsin instead of cos, but that is not a major complication. The amplitude
of z, say Z follows directly from (2.24) with F0 = mb2 Y , b as

Y r2 b
Z=p , r= . (2.54)
(1 r2 )2 + (2r)2 n

Transmissibility: Forces on the mass and on the base

The base motion induces motion of the mass in the mck SDOF system. We saw that the trans-
missibility ratio for the displacement X/Y can be resonant, meaning large motion of the mass for
small base motion. There are other measures of transmissibility, depending on the situation and
depending on the quantity of interest. For example, the force on the mass due to the base motion
is
F = m x = b2 mx FT cos(b t 0 ) (2.55)
where
FT = mb2 X (2.56)
Thus,
FT X b2 X
= 2
= r2 (2.57)
kY Y n Y

Force on base from forced SDOF

Lets go back to the first case of the fixed base and forced mass, and consider the force on the base
due to the harmonic forcing. The base is subject to the force of the spring and the dashpot (in
parallel):
F = kx + cx = m(n2 x + 2n x)
(2.58)
Using
x = A cos(t ) (2.59)
gives
 
F = m n2 A cos(t ) 2n A sin(t )
 
= mn2 A cos(t ) 2r sin(t )
p
= kA 1 + (2r)2 cos(t + 2 )
Fb cos(t + 2 ) (2.60)

using cos(a + b) = cos a cos b sin a sin b, and where where r = /n and

2 = tan1 2r (2.61)

Also, from the solution for the forced mass:

F0 1
A= q (2.62)
k
2
(1 r2 )2 + 2r
 32

and hence the ratio of base force to applied force is

p
Fb 1 + (2r)2
=q
2 , r= (2.63)
F0 n
(1 r2 )2 + 2r

Note that this force transmissibility FF0b is identical to the displacement transmissibility X
Y of (2.51)
for the system subject to base motion.
The base force is often a quantity that one wants to minimize. Hence, it is important to examine
the behavior of the ratio in (2.63). It clearly exhibits resonance, and if the base excitation is to
be minimized, then one should avoid forcing at the resonant frequency. What about low and high
frequencies? These can be understood by considering the limiting behavior of the ratio in (2.63)
for small and large values of r, respectively. Thus,

Fb 1 , r 0

(2.64)
F0

0 , r

At low frequency the applied force is completely transmitted to the base, with no amplification, but
with no reduction. However, at high frequencies the transmitted base force tends to zero. This is
a fundamental result of great significance in designing systems, as it guarantees vibration isolation
if the drive frequency is far greater than the resonance frequency of the support system.

Example: Car on a bumpy road

The response of a car moving along a rolling road can, as a first approximation, be considered
as a SDOF system subject to base motion. Thus, suppose a car is traveling along a road with
periodic peaks spaced 10 m apart. Calculate the the best choice for the damping factor among
= 0.1, 0.4, 0.6 so that the motion of the car is as small as possible for the case r = 2.
The period of the base motion experienced by the car depends upon the car speed, and is
10 3
T = 10 (3600) s
v
where v is the speed in km/hr. The base motion frequency is therefore

2 2v
b = = rad/s (2.65)
T 36
The problem is to find the value of that minimizes the transmissibility X/Y at r = 2. Thus, from
(2.51), at r = 2
s
X 1 + (2r)2
=
2
Y (1 r2 )2 + 2r
s
1 + 16 2
=
9 + 16 2
r
8
= 1 (2.66)
9 + 16 2

This shows that smaller smaller transmissibility. So = 0.1 is optimal.

 33

Lo/Hi frequency: stiffness and mass regimes

We have seen several types of transmissibility, e.g. kX/F0 for the displacement of the SDOF mass
subject to the force of magnitude F0 . Similarly, FT /(kY ) of (2.57) is the force transmitted to the
mass for base motion of amplitude Y . We also saw the displacement and force ratios X/Y of (2.51)
and Fb /F0 of (2.63), respectively. They all exhibit resonance, and have limiting behavior at low
and high frequencies that can be understood simply.
For instance, consider FT /(kY ) of (2.57). At low frequency (r 1), it is approximately unity:
FT
1 FT = kY, b n (2.67)
kY
while at high frequency (r 1) it is approximately 1/r2 , or

FT 2
n2 FT = mb2 Y, b n (2.68)
kY b

where we used k = mn2 again. These two equations for low and high frequency show that in each
regime only one of the three physical parameters of the m c k system comes into play. Thus, at
low frequency, the stiffness dominates the motion, while at high frequency the mass is the dominant
parameter. These separate frequency regimes are sometimes referred to as the stiffness and mass
regimes.
What happened to the dashpot parameter c? It does not have much effect in the stiffness or
mass-dominated regions, but it becomes important in the crossover at resonance. We have seen
1
that the limiting amplitude at resonance is always of the form 2 = km
c which involves all three
parameters.
Another way of looking at the low frequency, or stiffness regime, is that it is governed pri-
marily by the static response, because inertia and acceleration are small (they are proportional to
(frequency)2 ). Therefore, it is sometime appropriate to talk about a quasi-static response, where
the motion is governed by the stiffness primarily. However, near and above resonance, inertial
effects are all-important, and cannot be ignored.

2.2.3 Rotating unbalance

We have seen how resonance and vibration occurs as a result of (a) direct forcing on the mass and
(b) indirect forcing due to base motion. Another important source of excitation that is common in
engineering practice is the unbalanced rotor. The generic situation is the main mass as a box inside
of which a second mass rotates about a center, producing an oscillating force - an effective force on
the SDOF system.
Let the total mass of the box and the rotating object be m, and the rotating mass is m0 . Let
x represent the position of the main mass and the rotor is at

x0 = x + e cos r t

where e is the radius of the rotor. We consider the system of box and rotor, which is subject to the
external forces kx and cx, while the momentum in the xdirection is made up of two parts:
(m m0 )v and m0 v0 where v = x and v0 = x 0 . Hence, the equation of motion for the system is

(m m0 )
x + m0 x
0 = kx cx (2.69)

or

x + m0 x
(m m0 ) r2 e cos r t = kx cx (2.70)
 34

Rewrite this as a forced SDOF equation:

m
x + cx + kx = F0 cos r t, where F0 = em0 r2 (2.71)
Hence, the rotating mass induces a force of magnitude em0 r2 . In real systems, the distance e can
arise from a motor with an eccentric shaft, or in general an unbalanced motor. It is important to
note that the magnitude of the force increases as the square of the operating frequency.
Based on previous results, we have
x(t) = X cos(r t ) (2.72)
where
r2 1 2r r
X = em0 q
2 , = tan1 , r= (2.73)
k 1 r2 n
(1 r2 )2 + 2r
Using k = mn2 , gives the nondimensional displacement ratio
mX r2
=q
2 , (2.74)
m0 e
(1 r2 )2 + 2r
As usual, this displays a resonance. The low and high frequency limits for this nondimensional
ratio are r2 and 1, respectively, or
( 2
e m0kr , r n ,
X m0
(2.75)
em, r n .
The latter result implies that at frequencies far above resonance, we have
m0
x(t) e cos(r t), r n (2.76)
m
which is completely out of phase (180 ) with the motion of the rotor.
Is the center of mass of the system fixed? the simple answer is no. The center of mass is at x
,
where
m
x = (m m0 )x + m0 x0 = mx + m0 e cos r t
= mX cos(r t ) + m0 e cos r t
= (mX cos + m0 e) cos r t + mX sin sin r t (2.77)
But
1 r2 2r
cos = q
2 , sin = q
2 , (2.78)
(1 r2 )2 + 2r (1 r2 )2 + 2r
so, substituting into (2.69) and simplifying,
m0 e  2 2 3

x
=
2 (1 (1 4 )r ) cos r t + 2r sin r t
m[(1 r2 )2 + 2r ]
= X
cos(r t ) (2.79)
where p
= e m0 (1 (1 4 2 )r2 )2 + (2r3 )2 2r3
X
2 , = tan1 , (2.80)
m (1 r2 )2 + 2r 1 (1 4 2 )r2
At low frequency the center of mass moves with the rotor, and is at the fractional radius em0 /m.
At high frequencies, above resonance, the center of mass does not move and becomes increasingly
stationary. Thus, while the center of mass does move (why?), at high frequency it becomes fixed
as the main mass and the rotating mass move out of phase with each other (one moves up as the
other moves down, and vice versa).
 35

Example
A motor, m = 150 kg, has a rotating unbalance m0 = 0.5 kg, e = 0.2 m. The mount is a cantilever
beam, length L = 1 m, E = 2.1 1011 N/m2 . The operating range is 500 1200 rpm and = 0.1.
Find the cross-sectional moment of inertia I which ensures that the amplitude of vibration is less
than 1 mm.
Solution: Given,
mX 150 103
= = 1.5, (2.81)
m0 e 0.5 0.2
this maximum amplitude will be achieved at nondimensional frequency r satisfying

r2
1.5 = q
2 , (2.82)
2 2
(1 r ) + 2r

or
0.556 r4 1.96 r2 + 1 = 0 r = 0.787, 1.706
Based on the curve of (2.74), we know that amplitude would be even greater if

0.787 < r < 1.706

and therefore we want to make sure that the operating frequency range does not lie in this region.
This gives us two options: the operating range is such that (a) r 0.787, or (b) r 1.706.
(a) r 0.787 implies that the upper frequency, r = 2(1200/60) corresponding to 1200 rpm
is r 0.787. That is,
1200(2)
0.787 if n 159.7 rad/s
60n
But r
3EI (159.7)2 (1500 rad2 m3 kg
n = I
mL3 3(2.1)1011 s2 N/m2
or
(a) I 6.07 106 m4
(b) r 1.706 must hold for the lowest operating frequency, implying

500(2)
1.706
60n

which in turn implies that

(b) I 2.2 107 m4
Hence, we get two design options which satisfy the criteria specified. We would need to consider
other constraints in order to choose between the two.

2.3 Periodic forcing of a SDOF

A periodic forcing with period T which satisfies F (t + T ) = F (t) can be written as a Fourier series

X

F (t) = a0 + ap cos p t + bp sin p t (2.83)
p=1
 36

where
2
p = p, (2.84)
T
and the Fourier coefficients can be expressed as integrals:
Z
1 T
a0 = F (t)dt, (2.85)
T 0
Z
2 T
ap = F (t) cos p tdt, p 1, (2.86)
T 0
Z
2 T
bp = F (t) sin p tdt, p 1. (2.87)
T 0

By splitting the forcing into a sum of harmonic forcing terms we can use the linearity of the system
to write the solutions as a sum of solutions for each of the harmonics terms. Note that the particular
solution for F = a0 is x = a0 /k. The particular solutions for the cos and sin terms follow from eqs.
(2.32) and (2.33). Putting it all together gives

a0 X (ap ibp ) eip t
x(t) = + Re
. (2.88)
k
p=1
m n2 2p + i2n p

Even and odd forcing functions

The forcing is even if F (t) = F (t) and odd if F (t) = F (t). Since cos p t and sin p t are even
and odd, respectively, it follows that the Fourier series of an even function only involves the cos p t
terms, and conversely, the Fourier series of an odd function only involves sin p t terms. That is,

bp = 0, p 1 for an even function; ap = 0, p 0 for an odd function. (2.89)

2.3.1 Example: A sawtooth wave

Consider a sawtooth wave forcing defined by

t T T
F (t) = F0 for < t < ; F (t + T ) = F (t), T = 2, (2.90)
2 2
so that p = p, see (2.84). This is an odd function, so the ap coefficients are all zero. Simple
integration gives

2 2 X (1)p+1
bp = (1)p+1 F (t) = F0 sin pt. (2.91)
p p
p=1

The particular solution follows from (2.83) as

F0 X (1)p+1 eipt
x(t) = Re 2
. (2.92)
m
p=1
p n2 p2 + i2n p
 37

2.4 Impulse response function

2.4.1 Motivation: delta function and convolution integral
Consider the forced SDOF system (2.1) with forcing

F (t) = I (t) (2.93)

where I is a constant and (t) is the delta function. The latter can be considered as a spike at
t = 0 which is zero at all other times, and which has area of one underneath the spike. It can also
be considered as the limit of a step function of width 1/N and height N as N :

1 1
(t) = lim N H(t )H(t + ). (2.94)
N 2N 2N

The delta function is an even function, (t) = (t), and has the property4
Z b Z b (
f (t0 ) if a < t0 , b > t0 ,
(t0 t)f (t)dt = (t t0 )f (t)dt = (2.95)
a a 0 otherwise.

It can select out the value of a function at a given time. The same property means that we could
express a function as an integral of delta functions. Assuming f (t) = 0 for t < 0 then it can be
expressed for t > 0 as a sum/integral of all the values of f ,
Z
f (t) = (t t )f (t )dt . (2.96)
0

This integral only picks out the value at t = t so it seems like we are making things more
complicated than necessary. However, it provides the basis for expressing the solution of the
forced SDOF for any forcing function f (t). To see this, suppose the solution for forcing (2.93) is
x(t) = I g(t). Then the solution for any force F (t) that is zero for t < 0 will be
Z
x(t) = g(t t )F (t )dt . (2.97)
0

We can simplify this without known anything about g(t) except for the obvious physical property
that it must be zero for t < 0, which is just a statement of cause and effect (or causality), that the
system cannot respond before the force is applied. This means that g(t t ) = 0 for t > t and
hence the integral reduces to
Z t
x(t) = g(t t )F (t )dt . (2.98)
0

This type of integral is known as a convolution integral and the function g(t) is the impulse response
function.

2.4.2 Deriving the impulse response function

Combining eqs. (2.1) and (2.93) gives

m
x + cx + kx = I (t). (2.99)
4
The function f (t) has to be smooth at the value t0 , i.e. it is continuous at t0 .
 38

In addition, we have the initial conditions that the velocity and displacement are zero before the
force is applied, i.e. for t < 0. Consider two instants in time, once before t = 0 and the other after
t = 0, say t1 < 0 and t2 > 0. If we integrate both sides of (2.99) from t1 to t2 ,
Z t2 Z t2 Z t2 Z t2
m x
dt + c xdt
+k xdt = I (t)dt (2.100)
t1 t1 t1 t1

then using the property of the delta function that it picks out t = 0, and directly integrating the
derivatives x
and x,
we have
Z t2



m v(t2 ) v(t1 ) + c x(t2 ) x(t1 ) + k xdt = I. (2.101)
t1

where v(t) = x is the velocity. But since x and v are zero for negative times, this simplifies to
Z t2
mv(t2 ) + cx(t2 ) + k xdt = I. (2.102)
0

This is valid for any t2 > 0. Now let t2 0. How should we interpret (2.102)? Since x(t) and v(t)
are zero for t < 0, this equation means that one or both of them will be non-zero just after t = 0.
The correct way to interpret (2.102) is that x remains zero just after t = 0 but v takes a jump.
The integral in (2.102) is then zero and the value of v after the jump follows from

mv(t2 ) = I for t2 0. (2.103)

This makes sense: it says that the change in momentum of the system caused by the delta force at
t = 0 is equal to I, which must have the units of an impulse (i.e. force time).
With this interpretation we can exactly replace (2.99) with an initial value problem for the
system with no forcing:
I
m
x + cx + kx = 0 for t > 0; x(0) = 0, x(0)
= . (2.104)
m
The solution is now relatively straightforward: the sum of the two linearly independent solutions of
the homogeneous equation that satisfies the initial conditions. This is, using eq. (1.49) with x0 = 0,
v0 = I/m,
I
x(t) = sin d t en t . (2.105)
md
Since x(t) = I g(t) by definition, it follows that the impulse response function is
(
1
md sin d t en t , damped,
g(t) = 1
(2.106)
mn sin n t, undamped.

Note that g(t) = 0 for t < 0.

2.5 Measurement of vibration

2.5.1 Decibels
Vibrational and acoustic energy is measured in dB, or decibels. The decibel level is defined as
Power P
decibel level 10 log10 = 10 log10 (2.107)
Power reference level P0
 39

Power and energy are proportional to x2 , and since we are dealing with ratios, we can just as well
define the dB level in terms of a ratio of the displacement magnitude:
 
x2 x
decibel level 10 log10 2 = 20 log10   (2.108)
x0 x0

Often, x0 and P0 are chosen as the maximum values, so that the dB level is then negative. For
instance, if  
1 x 1
P = P0   =
2 x0 2
then the dB level is 10 log10 2 3 (numerically). This particular dB level is called the 3 dB
point, and is commonly used in defining resonances and in practical measurements of vibrating
systems. The 3 dB points on either side of a resonance maximum define the frequency range over
which the resonance is reduced to one half (power) and the associated frequency width is called
the 6 dB width. It gives a measure of the sharpness of the resonance.
What are the 3 dB points? We can find them from the standard formula, e.g. for the forced
SDOF system, for which
F0
A= q
2 , (2.109)
2 2
k (1 r ) + 2r
F0
The maximum value, assuming small damping ( small) is A0 k2 , and hence
q
A 2 1
2
=q
2 = 2 if (1 r2 )2 + 2r = 2 2 (2.110)
A0
(1 r2 )2 + 2r

Squaring and regrouping terms,

r4 2r2 (1 2 2 ) + 1 8 2 = 0 (2.111)

and hence
p
r2 = 1 2 2 2 1 + 2
1 2 (2.112)

since is small by assumption. Therefore, the 3 dB points are at

r 1 = n n (2.113)

Hence we have the important result: the 6 dB width is 2n (for lightly damped SDOF sys-
tems). This provides another way to measure the damping, and it is a very practical and useful
method. Note that the width of the resonance is inversely proportional to its strength - which is a
characteristic feature of the resonance of a lightly damped system.
Other terminology used: The quality factor Q is the resonant amplification, i.e.
1 n
Q= = (2.114)
2 6 dB width
A high Q system is one that exhibits a sharp and strong resonance. In short:

High Q small damping narrow resonance.

 40

2.5.2 Measuring instruments

How do we measure vibration? We have seen enough of SDOF systems to understand the principles
used to measure vibration. All vibrational measurement instruments have mass and stiffness, and
therefore they have their own natural frequency, n . The characteristic feature of instruments is
that they operate in either the low or the high frequency regimes, relative to n , but not at or near
the natural frequency itself.
The general feature of instruments is that they measure the relative motion of the interior, or
active mass, and the instrument body. The motion of the mass is therefore the motion of the
mass in a SDOF system subject to base excitation, at the operating frequency b . The quantity
of interest and the quantity that is measured in practice is the relative motion z = x y, which
satisfies eq. (2.52) or equivalently,

z + 2n z + n2 z = b2 Y cos b t (2.115)

for harmonic (single frequency) base motion

y = Y cos b t.

Then

z = Z cos(b t )

where (see (2.54))

Z r2 2r b
=q
2 , = tan1 , r= . (2.116)
Y 2 2 1 r2 n
(1 r ) + 2r

We consider two common examples; the seismometer and the accelerometer.

Seismometer: This has a heavy internal mass that is spring-mounted. The mass is surrounded
by coils so that its motion induces an electric current, which is the actual signal that comes from
the instrument. It operates above the natural frequency, i.e. b n , or r 1. In this case the
mass remains effectively stationary as the body of the seismometer moves up and down about it.
In terms of the model above, the seismometer operates in the range r 1 and therefore

Z
1, seismometer
Y

The mass is usually surrounded by a coil so that it induces an electric voltage as it moves. The
measured voltage is therefore directly proportional to the velocity of the base.
 41

Figure 2.6: The heavy mass of the seismograph remains stationary as the surface of the earth moves.
This seismograph measures vertical motion. Variations on the same idea are used for measuring
horizontal motion. Based on this wikimedia file.

Accelerometer: A stiff piezoelectric crystal (the spring) is placed in series with a small mass
m. The accelerometer is designed so that it operates at frequencies far below its own natural
frequency, b n . Therefore, the mass moves, and its relative displacement, from (2.116), is
approximately

b2
z y
n2

The voltage that is measured is proportional to the relative displacement (through the squeezing
of the piezoelectric crystal) and therefore the voltage is directly proportional to the acceleration of
the base. That is, the accelerometer measures acceleration, hence its name.

Accelerometer design considerations

The accelerometer is the workhorse of all vibrational measurements, and it is worth considering
some of the details that go into its design. When measuring vibration, we first must choose an
accelerometer with a frequency range of operation that includes the frequencies we are interested
in, e.g. 0 10 kHz. The natural frequency of the accelerometer must greatly exceed the upper
frequency, this case 10 kHz or 10, 000 2 63, 000 rad/s. This can be achieved, e.g., by selecting
a device with a small mass (the smaller the mass the higher the natural frequency), and as a result,
as the operating frequency increases, the accelerometer size decreases.
 42

Figure 2.7: Schematic of how an accelerometer in a cellphone works. The mass, called the seismic
mass (not to be confused with the seismometer) translates relative to the body, causing an electric
signal from the varying capacitance.

Flatness of the response: the model is based on the low frequency approximation to the
response, which gives z proportional to the acceleration of the base. How well does this approxi-
mation hold, or how flat is the response? The answer depends on the error, which is the difference
between the approximation and the exact response. Specifically, we define the error as the difference
between the measured and predicted values of Z/(r2 Y ),
1
error 1 q
2
(1 r2 )2 + 2r

Now, from (2.116)

1
1 q
2 = 1 [1 + (4 2 2)r2 + r4 ]1/2
(1 r2 )2 + 2r
1
= 1 [1 (4 2 2)r2 + . . .]
2
= (2 2 1)r2 + O(r4 ) (2.117)

where we have used the fact that r is small. The error can be minimized over a wide rangeof
frequency if the damping is such that (2.117) is zero, to leading order,
which occurs for = 1/ 2.
Therefore, the optimal damping for an accelerometer is = 1/ 2 which ensures a flat response
over a wide range of frequency. Recall that this value of is the one separating the occurrence of
a definite peak in the response (for > 1/sqrt2), from the case when there is no maximum.
Phase distortion: Practical measurements are not usually for a simple harmonic motion, but
comprise many frequencies simultaneously. Consider

y = Y1 cos 1 t + Y2 cos 2 t (2.118)

for which, using the linearity of the system and the equations

12 12
z= Y 1 cos( 1 t 1 ) + Y2 cos(2 t 2 ) (2.119)
n2 n2
But z is not proportional to y:

y = 12 Y1 cos 1 t 22 Y2 cos 2 t
 43

The problem comes from the phase terms 1 and 2 . Any phase means that the measured response
is not exactly the same as the excitation, but is delayed in time. The problem can therefore be
solved if both signals have the same time delay, in which case the response will be proportional to
the acceleration, but delayed slightly. We therefore need

1 2
T (2.120)
1 2

where T is a time constant. If this is the case, then (2.119) implies

1
z(t) y(t T ) (2.121)
n2

Can we find T ? From (2.116),

2b /n
= tan1 (2.122)
1 (b /n )2

Let = T b , then
2b /n sin T b T b + . . .
= = (2.123)
1 (b /n )2 cos T b 1 21 (T b )2 + . . .
We can therefore get a very good approximation if both of the following hold:

b2 1 b
1 2
1 (T b )2 and 2 T b (2.124)
n 2 n

Solving the first gives T , and plugging into the second gives :

2 1
T = and = (2.125)
n 2

We found = 12 before as the optimal damping (for giving a flat response). In addition, we see
that the same choice of damping is the only one that ensures a phase delay proportional to the
excitation frequency, and therefore it is the unique value of damping that minimizes the response
distortion. In summary, the optimal damping gives

1 2
z(t) 2 y(t ) (2.126)
n n

The actual time delay n2 is not significant; it is more important to ensure the fidelity of the
instrument under general excitation.

2.6 Problems
1. Another way to solve eq. (2.5) is to assume that

x(t) = C1 cos t + C2 sin t (2.127)

where C1 and C2 are found by plugging x into eq. (2.5) and then comparing terms on both
sides that have cos t and sin t. This gives two separate algebraic equations (instead of the
single equation (2.12)). Show that the solution is the same as that of eqs. (2.17) and (2.19).
 44

2. (a) An undamped spring-mass system with m = 4 103 kg, k = 1 105 N/m is subject to
harmonic forcing at twice the natural frequency n . The steady state amplitude |A| of the
mass is observed to be 2 mm. Find the magnitude of the forcing, F .
Hint: First write out the equation relating |A|, F , m, n and .
(b) A machine of mass m = 1 103 kg is constrained by a spring of stiffness k. Find the value
of k such that the maximum amplitude of the machine displacement is |A|=10 mm when
subject to harmonic forcing of magnitude F = 3 103 N at radial frequency = 10 s1 .
Hint: No damping, use the same equation as in (a).
Circle the correct answer for (a) and (b), and show your work below.
1 1
(a) 3 2 1 2 3 4 6 8 20 30 60 ( 102 N)
(b) 0.2 0.4 0.6 0.8 2 4 6 8 ( 105 N/m)

Solution:
F0
(a) (10). |A| = 2 2 +2i |
m|n n
= 0 |A| = m|F2 02 |
n
F0 F0
= 2n 2 = 4n2 |A| = m3n2 = 3k F0 = 3k|A|

F0 = 3k|A| = 3 105 2 103 = 6 102

F0 F0
(b) (10). |A| = |km 2 |
|k m 2 | = |A|
F0
|A|= 3 103 102
= 3 105
m 2 = 1 103 102 = 1 105
F0
k m 2 = |A|
F0
k = m 2 + = 4 105
|A|
Chapter 3

Multi-degree of freedom systems

3.1 Matrices M and K and the equation M

u + Ku = 0
We first consider multi-degree of freedom systems made of springs and masses in a linear chain.
The notation is as follows:

1. |km indicates a SDOF spring and mass. The | means that the spring is attached at this end.

2. |kmk| is a SDOF mass with springs on either side. E.G. |k1 mk2 | 2 = (k1 + k2 )/m.

3. |kmkm is a 2-DOF system with the mass on the right attached only to one spring.

4. |kmkmk| is a 2-DOF system with the mass on the right attached to two springs.

5. mkm is a 2-DOF system with the masses connected by one spring, otherwise free.

6. |kmkmkmk| is a 3-DOF system with all masses attached to two springs.

7. mkmkm is a 3-DOF system with the masses connected by two springs, otherwise free.

8. |k mk kmkmk} | is a n-DOF system with all masses attached to two springs.

| . . .{z
n times

3.1.1 Stiffness and Flexibility matrix

The Stiffness matrix K for an n-DOF system is an n n matrix with element kij , i, j = 1, 2, . . . , n.
These elements can be defined via


f1 k11 .. .. k1n x1


f2 k 21 .. .. k2n
x2
f = Kx = (3.1)

.. .. .. .. .. ..

fn kn1 .. .. knn xn

Thus, if mass j is moved by the displacement xj and all the others are kept fixed (zero), then kij
is the force on mass i required to maintain this displacement for mass j.
The Flexibility matrix A is defined as the inverse of K:

A = K1 x = Af (3.2)

The element aij , i, j = 1, 2, . . . , n can be thought of as follows: if mass j is subject to the force fj
and all the others are not forced (zero forces), then aij is the displacement on mass i .

45
 46

3.2 Examples
3.2.1 A 3-DOF system
Three masses and four springs in series |k1 m1 k2 m2 k3 m3 k4 | gives three eqs:
m1 x
1 + (k1 + k2 )x1 k2 x2 = 0 (3.3)
m2 x
2 k2 x1 + (k2 + k3 )x2 k3 x3 = 0 (3.4)
m3 x
3 k3 x2 + (k3 + k4 )x3 = 0 (3.5)
The 3 eqs are the same as a single vector eq:

m1 x1 + (k1 + k2 )x1 k2 x2 0
m2 x 2 k2 x1 + (k2 + k3 )x2 k3 x3 = 0 (3.6)
m3 x3 k3 x2 + (k3 + k4 )x3 0
The vector is 3 1. But that is not much use. How about the sum of two vectors equals zero?

m1 x
1 (k1 + k2 )x1 k2 x2 0
m2 x2 + k2 x1 + (k2 + k3 )x2 k3 x3 = 0 (3.7)
m3 x
3 k3 x2 + (k3 + k4 )x3 0
Better but still not much different from (3.3). The real use of linalg is to use vectors together with
matrices.
The idea is to multiply vectors by matrices. If A is a 3 3 matrix and v is a 3 1 (rows
columns) vector, then Av is 3 1, i.e. a column vector.

a11 a12 a13 v1 a11 v1 + a12 v2 + a13 v3
Av a21 a22 a23 v2 = a21 v1 + a22 v2 + a23 v3 (3.8)
a31 a32 a33 v3 a31 v1 + a32 v2 + a33 v3
So we can write (3.7) as

m1 0 0 x
1 k1 + k2 k2 0 x1 0
0 m2 0 x 2 +
k2 k2 + k3 k3 x2 = 0
(3.9)
0 0 m3 x
3 0 k3 k3 + k4 x3 0
Its simpler to write it as
M
u + Ku = 0, (3.10)
where

m1 0 0 k1 + k2 k2 0 x1
M = 0 m2 0 , K = k2 k2 + k3 k3 , u = x 2 . (3.11)
0 0 m3 0 k3 k3 + k4 x3
Thus, M and K are 3 3 and u is 3 1.

The modal equal

To find modes, we assume u(t) = Ueit where U and are constant, and unknown at this stage.

Hence u(t) = iUeit or more simply, u = iu. Hence, u = (i)2 u = 2 u, so we
= i u or u
get, after dividing by the scalar factor eit ,
(K 2 M)U = 0. (3.12)
We started with a system of coupled ODEs, and we are left with an algebraic equation. Algebraic
equations are (nearly) always easier to deal with than ODEs.
 47

Note on using complex numbers

We are using the same complex-number notation as before when we considered the SDOF system.
The real physical displacement is found by taking the real part of u, in other words to be strictly
correct we should write u(t) = Re Ueit but we will leave out the Re bit as being obvious and
automatically understood. If U is a real vector, say U = U1 then we have in fact that u(t) =
U1 cos t. Alternatively, if U = iU2 where U2 is real, then u(t) = U2 sin t. More generally, we
can always write U = U1 + iU2 , so that the real displacement is u(t) = U1 cos t U2 sin t.

3.2.2 n-DOF chain

Consider a system |k1 m1 k2 m2 k3 . . . ki mi ki+1 mi+1 . . . mn kn+1 . . . mN kN +1 |. What is kii ? To find it,
imagine moving mi by xi with the others fixed. To do this, the force fi must counter the two springs
on either side of mi , that is kii = ki + ki+1 . What about ki i+1 ? This must be equal and opposite
to the force on the other end of the spring ki+1 , so ki i+1 = ki+1 . Similarly ki i1 = ki1 . The
elements are zero except for the nearest neighbors, because only the nearest neighbors need to be
forced to maintain the displacement xj with all others zero.
This is a slightly different way of looking at the issue than we discussed in class. There we looked
at the force on each mass, say mj , and argued that it only depends on the relative displacements
of this mass and its neighbors. We get the same result for K whatever way we look at it:

k1 + k2 k2
k2 k2 + k3 k3

k 3 k 3 + k4
K=
..
(3.13)
.. .. .. .. ..

.. kN
kN kN + kN +1

The equation for mass mi in the chain is

mi x
i = ki (xi xi1 ) ki+1 (xi xi+1 ) (3.14)

The ends are taken care of by defining x0 = 0 and xN +1 = 0. The ends are fixed. We could consider
the case of free ends by letting k1 = 0 (free at left end) and/or kN +1 = 0 (free at the right end).
For example, if n = 2, we have |k1 m1 k2 m2 k3 | and

 
k + k2 k2
K= 1 (3.15)
k2 k2 + k3

The flexibility matrix can be determined from eqs. (3.2) and (3.15) as

 
1 k2 + k3 k2
A= (3.16)
k1 k2 + k2 k3 + k3 k1 k2 k1 + k2
 48

3.2.3 A simple model of a car

L1
L2
G
x

k1
k2

Figure 3.1: The 2DoF car model.

The car is modeled as a 2DoF system with x the vertical displacement of the center of mass G and
the rotation angle, both measured from static equilibrium (x = 0, = 0). The vertical displacement
of the left end of the rigid bar of mass m, centroidal moment of inertia IG is x + L1 , and of the
right end, x L2 , see Fig. 3.1. Taking the sum of the forces acting gives the equation of motion
of the center of mass
mx = k1 (x + L1 ) k2 (x L2 ), (3.17)
and taking the moments about G gives

IG = L1 k1 (x + L1 ) + L2 k2 (x L2 ). (3.18)

Hence, the matrix form of the equations is

       
m 0 x
k1 + k2 k1 L1 k2 L2 x 0
+ = . (3.19)
0 IG k1 L1 k2 L2 k1 L21 + k2 L22 0

3.2.4 Coupled pendula

1 a
2
a
L2
L1 k

m2
m1
Figure 3.2: The pendula are coupled by the spring.
 49

Two light but rigid thin rods support masses with a spring connecting them as shown in Fig. 3.2.
The extension of the spring is a(2 1 ). The equations of motion can be found by taking moments
about the two support points, left and right respectively give
m1 L21 1 = m1 gL1 1 + ka2 (2 1 ),
m2 L22 2 = m2 gL2 2 ka2 (2 1 ). (3.20)
The matrix form of the two equations is
       
m1 L21 0 1 m1 gL1 + ka2 ka2 1 0
+ = . (3.21)
0 m2 L22 2 ka2 m2 gL2 + ka2 2 0

3.2.5 The double pendulum

1
L1
m1

2 L2

m2
Figure 3.3: The double pendulum.

The first pendulum, length l1 , mass m1 , hangs from a fixed point. The second pendulum, length
l2 , mass m1 , hangs from m1 . Let j , j = 1, 2 be the angle pendulum j makes with respect to the
vertical. Assume small angles of motion. The horizontal positions of the two masses are
x1 = l1 sin 1 l1 1 ,
x2 = x1 + l2 sin 2 l1 1 + l2 2 . (3.22)
The horizontal forces acting on the masses are due to the tensions. taking the components in the
direction of xj ,
on m1 : T1 sin 1 + T2 sin 2 T1 1 + T2 2 ,
on m2 : T2 sin 2 T2 2 . (3.23)
The equations of motion follow from (3.22) and (3.23)
m1 l1 1 = T1 1 + T2 2 ,
m2 (l1 1 + l2 2 ) = T2 2 . (3.24)
 50

In the small angle approximation the tensions in the two strings are the same as in static equilibrium,
that is
T1 = (m1 + m2 )g, T2 = m2 g (3.25)
Hence,
m1 l1 1 = (m1 + m2 )g1 + m2 g2 ,
m2 (l1 1 + l2 2 ) = m2 g2 , (3.26)
Note that the mass and stiffness matrices for this pair of equations are not symmetric. If we however
replace the first equation by the sum of the two equations,
(m1 + m2 )l1 1 + m2 l2 2 = (m1 + m2 )g1 ,
m2 (l1 1 + l2 2 ) = m2 g2 , (3.27)
then multiply the first by l1 and the second by l2 , we get
(m1 + m2 )l12 1 + m2 l1 l2 2 = (m1 + m2 )gl1 1 ,
m2 (l1 l2 1 + l2 2 ) = m2 gl2 2 ,
2 (3.28)
or in matrix form with symmetric matrices,
     
(m1 + m2 )l12 m2 l1 l2 1 (m1 + m2 )gl1 0 1
+ = 0. (3.29)
m2 l1 l2 m2 l22 2 0 m2 gl2 2
The double pendulum equations are derived in Appendix C using Lagrange equations. Note
that this method yields symmetric matrices automatically. p
As an example, suppose m1 = m2 = m, l1 = l2 = l, and define 0 = g/l, the frequency of a
single pendulum of length l, then equations (3.26) become
1 = 202 1 + 02 2
1 + 2 = 02 2 (3.30)
Note that since these are two coupled linear equations, they can be rewritten in different ways, e.g.
1 = 202 1 + 02 2 (3.31)
2 = 202 (2 1 ) (3.32)

3.3 Modes of n-DOF Systems

3.3.1 Calculating modes of n-DOF Systems
Assuming a time harmonic solution x = cos t for the unforced, undamped, equations of motion
M{
x} + Kx = 0, (3.33)
implies that the constant n vector satisfies


K 2 M = 0. (3.34)


This has several interpretations. First, is a null vector of the symmetric matrix K 2 M .
A more useful way to look at it is in terms of eigenvalues, that is, is a generalized eigenvector
of the stiffness matrix K and mass matrix M. The concept of eigenvector is usually restricted
to the systems Av = v, which is a particular case of the generalized system with M = I,
the identity. However, numerical linear algebra packages, Matlab included, usually handle the
generalized system, see B.2 for details.
 51

3.3.2 Example: 3-dof system

For example, take the 3-dof system in (3.9) with m1 = m2 = m3 = m and k1 = k2 = k3 = k4 = k,
then

2 1 0 2 1 0 2 1 0
K 2 M = k 1 2 1 2 mI = k 1 2 1 kI = k 1 2 1
0 1 2 0 1 2 0 1 2
(3.35)
where = 2 m/k. Now, if det A = 0 then det cA = 0 where c is any scalar factor. That is,
multiplying every element by the same constant does not change the equation det A = 0. So, the
modal frequency is obtained by

2 1 0
det 1 2 1 = 0 (2 )[(2 )2 1] (2 ) = 0. (3.36)
0 1 2

satisfy (2 )2 2 =
(2 ) is a factor, so = 2 is a root, and the other two 0. Putting the three
in ascending order 0 < 1 < 2 < 3 , they are 1 = 2 2, 2 = 2, 3 = 2 + 2.
Multiply (3.12) from the left by M1 , and rewrite as

M1 KU = 2 U. (3.37)

In this form U is an eigenvector of M1 K.

3.3.3 Matlabs solution

The way they normally do this is to convert to the simpler eigenvalue problem using Cholesky
factorization. The idea is to express the mass matrix as

M = [L][L]T , (3.38)

where [L] is lower diagonal, i.e., with zeroes in all elements above the diagonal, Lij = 0 for all
i > j. The actual Cholesky factorization is a relatively fast numerical process, based on a simple
recursive algorithm (not discussed here). The reason for using this type of decomposition of M is
that lower diagonal matrices are easily inverted. Normally the inverse [L]1 is not stored, since
the system [L]{x} = {y} can be solved for {x} quickly. Quickly, fast and other such terms
indicate a numerical method that is an order of magnitude more efficient that a full inverse of a
dense matrix.
Returning to (3.34), we can rewrite it formally, by premultiplication by [L]1 , as

[L]1 K[L]T {v0 } = 2 {v0 }, (3.39)

where {x0 } = [L]T {v0 }. This yields a simple eigenvalue equation for the symmetric matrix

= [L]1 K[L]T .
K (3.40)

3.3.4 Formal solution

The linear algebra solution (3.38) is just a statement of the identity

M = M1/2 M1/2 . (3.41)

 52

If there are questions about (3.41), note that the square root A1/2 of a matrix is always well defined
and unique as long as the matrix A is positive definite, i.e. ut Au 0, u. The idea of square root
matrix is follows from the idea of eigenvalue/eigenvector diagonalization, which is at the heart of
modal decomposition.
If we plug (3.41) into (3.34), we can write the equation as


KM1/2 2 M1/2 M1/2 = 0. (3.42)

Multiplying in the left by M1/2 leads to (see (3.40))

K 2 I = 0 (3.43)

where I is the n n identity matrix and

= M1/2 KM1/2 , = M1/2 .
K (3.44)

Equation (3.43) is now in standard form. Note that the modified stiffness matrix is symmetric,
T = K
K if K is symmetric.

3.3.5 Example: Rigid body modes

Consider an airplane modeled as a system of three masses, the two wings each of mass m and the
fuselage between them of mass 4m. The wings are modeled as beams under flexural displacement,
each of stiffness k = 3EI/L3 , where L is the distance of the center of mass of the wing from the
center of the air plane. Let x1 and x3 represented the upward motion of the wings, and x2 is the
center of the plane upward displacement. Thus, the airplane acts like a 3-DOF system of the form
mk4mkm, or

1 0 0 3 3 0
EI
M = m 0 4 0 , K = 3 3 6 3 , (3.45)
L
0 0 1 0 3 3

Thus,

1 0 0
1 = M1/2 KM1/2
M1/2 = 0 12 0 K
m
0 0 1

3 1.5 0
EI EI
= 1.5 1.5 1.5 K (3.46)
mL3 mL3
0 1.5 1.5

and

I = EI K I , EI


K 3
where = (3.47)
mL mL3
The equation for the modal frequencies is
 
3 1.5 0 




det K I =  1.5 1.5
 1.5  (3.48)
 0 1.5 1.5 
or
3 7.5 2 + 13.5 = 0 (3.49)
 53

The three roots are = 0, 3, 4.5, or

1 = 0, 2 = 1.732, 3 = 2.121, (3.50)

where = . The eigenvectors 1 , 2 , 3 can be found by substituting these values of into
(3.48), and we get

0.4082 0.7071 0.5774
= [ 1 , 2 , 3 ] = 0.8165 0 0.5774 . (3.51)
0.4082 0.7071 0.5774

Note that this matrix is normalized, i.e. the vectors are all of length one, t1 1 = 1, etc., and
they are orthogonal, t1 2 = 0, etc. This means that the matrix is orthogonal

1 1 0 0
T = [ 1 , 2 , 3 ] 2 = 0 1 0 . (3.52)
3 0 0 1
The mode shape matrix is

0.4082 0.7071 0.5774
= [1 , 2 , 3 ] = M1/2 = 0.4082 0 0.2887 (3.53)
0.4082 0.7071 0.5774
is the mode shape matrix in physical coordinates. Note that is not orthogonal, but since
is, it means that T M = I, see section 3.3.7.
The three columns in correspond to the three modal frequencies in (3.50). Note how the
first one for 1 = 0 has identical values in each row. This corresponds to a mode where each xj ,
j = 1, 2, 3 moves the same amount and is known as a rigid body mode. The reason for this mode
appearing now where it did not before is that we have allowed the masses to be unconstrained. This
extra degree of freedom arises from the fact that we chose to use the three DOF, one for each mass.
But in practice the interesting and physically significant modes are the second and third ones. In
the second mode the middle mass does not move and each wing moves in opposite directions. In
mode 3, the two wings move in the same direction and the center in the opposite direction, such
that the center of mass does not move.
In summary, the rigid body mode occurs if the entire structure can translate or rotate.

3.3.6 Damping, proportional damping

If the springs have dashpots in parallel then the equations become
u + Cu + Ku = 0
M (3.54)
where C is the damping matrix. In practice it is often difficult or impossible to characterize this
matrix, so instead a simple and very useful alternative is used. Assume the damping matrix is
proportional in the following sense
C = K + M (3.55)
so that the entire damping is defined by only the constants 0 and 0. Then
u = Ueit KU + iCU 2 MU = 0

(1 + i)KU 2 (1 i )MU = 0

KU 2 MU = 0 (3.56)
 54

where
 1 i 
2 =

2 . (3.57)
1 + i
Note that the damped equation (3.56) is exactly the same in form to the undamped equation
except that the frequency is
rather than . It means that the modal vectors are the same as the
undamped modal vectors. The mode frequencies j can be obtained from (3.57) where now j are
the undamped frequencies (which are real valued). The damped mode frequencies j are complex
valued, just like the SDOF damped system. In fact we can solve for the damped frequencies in a
way that is just like the SDOF case by rewriting (3.57) as


2 i

+ 2 = 0,
(3.58)


or equivalently
1

2 2i 2 = 0 where =

+ . (3.59)
2

In other words, the damped mode is like a SDOF with natural (undamped) frequency j and modal
damping factor
1

j = j + . (3.60)
2
j

3.3.7 Orthogonality
Orthogonality of modes
Consider two distinct modes i and j , satisfying



K i2 M i = 0, K j2 M j = 0, i 6= j. (3.61)

Multiply the first on the left by Tj and the second on the left by Ti , to get

Tj Ki = i2 Tj Mi , Ti Kj = j2 Ti Mj . (3.62)

This is where we use the fact that the matrices M and K are symmetric, which implies

Tj Ki = Ti Kj , Tj Mi = Ti Mj . (3.63)

Hence, (3.62) implies

(i2 j2 )Ti Mj = 0 (3.64)
But by definition, i 6= j, and i 6= j , theretofore

Ti Mj = 0, and Ti Kj = 0 (3.65)

Two vectors v1 , v2 satisfying v1T Bv2 = 0 are said to be orthogonal with respect to the matrix
B. Thus, the modes are orthogonal with respect to both the mass and the stiffness matrices.
It is traditional to normalize the eigenvectors (modes) with respect to the mass matrix. Define

i = q i , (3.66)
T
i Mi

then it is clear that these vectors satisfy the simpler orthogonality condition

Ti Mj = ij , where ij = 1, i = j; ij = 0, i 6= j, (3.67)
 55

that is
T M = I. (3.68)
The normalization condition (3.68) implies that the inverse of the mode matrix can be expressed
simply as
1 = T M. (3.69)
left multiplying this by gives another useful identity

T M = I. (3.70)

The reason for normalizing with the mass rather than the stiffness is that the former is often
diagonal. For instance, if it has the simplest form M = mI, then M = mI and (3.68) becomes
Ti j = m
1
ij .

Modal coordinates
Define the N N matrix of modal vectors
 
= 1 , 2 , . . . , N (3.71)

Then


T1
T
2  
T M = . M 1 , 2 , . . . , N
..
TN

1
1
= = I. (3.72)
...
1

At the same time, we can see that

 2 
T K = nat (3.73)

where
12 1
 2
  2 22   2
nat = nat = , nat = . (3.74)
... ...
2
N N
These are the fundamental relations that we will use below

3.3.8 Initial value problems

Suppose the n-DOF system is set in motion by initial conditions for the displacements and velocities,
i.e. we are given the values of the two n-vectors

{x(0)} = {a},
(3.75)
{x(0)}
= {b},

with components aj and bj .

 56

Undamped solution
Assume the solution of the form
n
X
{x(t)} = j Aj cos j t + Bj sin j t). (3.76)
j=1

This is just like the SDOF solution, but with a contribution from each mode that is proportional
to the mode shape. It should be clear that (3.76) is a solution to the homogeneous equations of
motion (3.33). Then it follows that
n
X
{x(0)} = j Aj = {a},
j=1
n (3.77)
X
{x(0)}
= j Bj j = {b}.
j=1

We can write these in matrix form as

{A} = {a},
  (3.78)
nat {B} = {b}.

These can be solved using (3.69), to give

{A} = T M{a},
 1  T (3.79)
{B} = M{b}
nat
 1   1
where nat = nat is the diagonal matrix with values 1/1 , . . . 1/n . Substituting back into
(3.76) allows us to write the solution in matrix form as
   sin nat t  T
{x(t)} = cos nat t T M{a} + M{b} (3.80)
nat
 
where cos nat t is the diagonal matrix with entries cos 1 t, . . . cos n t, etc. It is clear from this
form of the solution when t = 0, {x(0)} = IT M{a} = {a} using the identity (3.70). Similarly
{x(0)}
= T M{b} = {a}, showing that all of the initial conditions are satisfied.

Damped solution
Assume the damping is proportion, as described in section 3.3.6. Then the modes are the same as
for the undamped case and the modal frequencies become complex valued as in the SDOF system
according to equation (3.60). Based on the undamped solution (3.80), and the initial value problem
for the damped SDOF case, it can be shown that
 sin damp t
nat t  T
{x(t)} = cos damp t + nat e M{a} (3.81)
damp
 sin damp t nat t  T
+ e M{b}
damp
q
where nat j , j and damp 1 j2
j , see section 3.3.6. It follows that this solution
satisfies the initial conditions (3.75).
 57

3.4 Forced multi-DOF Systems

A n-DOF system subject to forcing can be represented as
M{
x} + C{x}
+ Kx = {F (t)} (3.82)
where C is the damping matrix. We first consider a very important 2-DOF case.

3.4.1 Forced 2-DOF Systems: Vibration Absorber

Consider a SDOF system m, k, without damping, where the mass is forced by the steady state
forcing of frequency ,
F = F0 sin t
p
The system resonates if the drive frequency equals the system natural frequency k/m. The
vibration absorber is a passive method to remove this resonance by adding an extra DOF with
chosen properties.
Consider an added mass ma attached to the main mass by a spring of stiffness ka . The main
mass is attached to a rigid base via the spring k but the added mass is only connected by ka , that
is the system is like | k m ka ma where the forcing acts on m only. The coupled equations
of motion are therefore
       
m 0 x
k + ka ka x F0 sin t
+ = (3.83)
0 ma x a ka ka xa 0
or
M{x} + Kx = {F0 } sin t (3.84)
where
 
F0
{F0 } = (3.85)
0
Assume x has the form (3.96), where the constant vector is
 
X
{X} = (3.86)
Xa
Equation (3.99) becomes
    
k + ka m 2 ka X F0
= (3.87)
ka ka ma 2 Xa 0
These simultaneous equations for X and Xa can be readily solved, giving
(ka ma 2 )F0
X= (3.88)
(k + ka m 2 )(ka ma 2 ) ka2

ka F0
Xa = (3.89)
(k + ka m 2 )(ka ma 2 ) ka2
The displacement of the main mass, given by (3.88), will be zero if the added mass and its
spring are such that ka ma 2 = 0 or
ma 2 = ka (3.90)
Thus, it is possible to completely eliminate the resonance by tuning the added mass to the drive
frequency! The added mass has non-zero displacement, equal to
F0
Xa = (3.91)
ka
 58

Example
A rotating saw operating at 180 cycles per minute is positioned on a table. The saw is unbalanced
so that it exerts a force of magnitude 13 N on the table. Assume that the legs of the table have
effective stiffness of 2600 N/m. Find the parameters for an added mass vibration absorber which
eliminates the vibration of the saw, and is such that no part of the whole system vibrate with
amplitude more than 2 mm.
Here the saw acts as the main mass (unknown but assumed large relative to ma ) and the added
mass should make the amplitude of the saw vibration zero. Hence, according to (3.90),

ma (2180/60)2 = ka (3.92)

Since the saw does not vibrate (because we are choosing the added mass according to (3.90)), the
only vibration comes from the added mass itself, and is given by (3.91). hence we require that

13
2 103 m (3.93)
ka

Solving (3.92) and (3.93) implies that the vibration absorbed should be such that

ka 6, 500 N/m (3.94)

If ka = 6, 500 N/m then the mass should be 18.29 kg. This choice of vibration absorber will give
the desired response and is a valid design option.

3.4.2 Time harmonic forcing

Time harmonic forcing
F (t) = F0 cos(t 0 ), (3.95)
implies that the forcing has no beginning or end - it is a steady state situation. In this case we can
assume that the response of the system has the same steady state behavior, at the same frequency.
That is
{x(t)} = Re {X}ei(t0 ) (3.96)
where {X} is a constant n 1 complex-valued vector. Note, we have kept the arbitrary phase 0 in
both the forcing and the response. Normally, we consider either F (t) = F0 cos t or F (t) = F0 sin t,
corresponding to 0 = 0 and 0 = 2 , respectively. In the latter case, we could replace (3.96) by
{x(t)} = Im {X}eit . In general, we will write

{x(t)} = {X}eit (3.97)

where it is understood that 0 = 0, and that the real part of the solution is implicit (complex-valued
physical quantities have no real meaning - pun intended). Plugging into (3.82) gives

2 M{X}eit iC{X}eit + K{X}eit = {F0 }eit (3.98)

or, since this is true for all t,

K iC 2 M {X} = {F0 } (3.99)
This can be formally solved
1
{X} = K iC 2 M {F0 } (3.100)
We have already seen the very important example for n = 2: the vibration absorber in 3.4.1.
 59

3.4.3 Solution in modal coordinates

Assume that the displacements can be represented as

{x(t)} = {q(t)} . (3.101)

Substituting into (3.82) gives

M{
q } + K{q} = {F (t)} (3.102)
followed by left multiplication with T ,

T M{
q } + T K{q} = T {F (t)} (3.103)

This simplifies using the modal relation (3.72) and (3.73),

2
{
q } + [nat ]{q} = T {F (t)} . (3.104)

The modal coordinates now satisfy uncoupled, or independent, equations of motion,

qj (t) + j2 qj = Tj {F (t)} , j = 1, 2, . . . , N. (3.105)

Each equation is similar to that for a SDOF system, and can be solved accordingly. As with the
SDOF system we need to distinguish different types of forcing:

Transient, or forcing for t 0 only

Time harmonic {F (t)} = {F0 } sin t

In the former case, the general solution can be written formally using our previous results for the
SDOF forced system in terms of a convolution integral with the impulse-response function:
Z t
sin j
qj (t) = d Tj {F (t )} , j = 1, 2, . . . , N. (3.106)
0 j

Time harmonic forcing: Modal coordinates

An alternate representation for the solution under time harmonic forcing can be found in the form
(3.101), specifically,

{q(t)} = {Q} cos t , qj (t) = Qj cos t , j = 1, 2, . . . , N. (3.107)

Then (3.105) implies

(j2 2 )Qj = Tj {F0 } , j = 1, 2, . . . , N, (3.108)
which determines Qj . Remember, j is the normalized modal vector j, see eq. (3.71).
Define the matrix
 1  1 1 1

2 2 = diag , , . . . , 2 2 , (3.109)
nat 12 2 22 2 N

then combining (3.101), (3.107) and (3.108), gives

 1  T
{x(t)} = 2 2
{F0 } cos t. (3.110)
nat
 60

It is interesting to compare (3.110) with the direct solution found previously, (3.100). Comparing
this with the modal approach gives two alternate representations for the amplitude vector:

1
{X} = K 2 M {F0 }, (3.111)
 1 
{X} = 2 T {F0 }. (3.112)
nat 2
Since the forcing vector is arbitrary, this implies the identity

1  1  T
K 2 M = 2 2
. (3.113)
nat


The proof of this identity can be found directly by multiplication with K 2 M , followed by use
of the identities (3.72) and (3.73), and relations of the type
 2 
K = M nat , MT = I. (3.114)
The first of (3.114) is matrix version of the modal identities (3.61), the second one follows from
(3.68) by right multiplication with T and then left multiplication with T .

3.5 Impedance (SKIP)

3.5.1 Attached masses and impedance
We can find explicit solutions for a very useful type of N-DOF system, comprising N distinct
masses attached to a common mass-less base. As a preliminary, consider a single spring-mass-
dashpot system (m, k, c), attached to a base, where the base is driven by the forcing F . Let x(t)
and y(t) denote the motion of the mass and base, respectively. The equilibrium equations for the
mass and the base are then,
m
x(t) = k(x y) c(x y),
(3.115)
0 = F + k(x y) + c(x y),
(3.116)
respectively (recall, the mass of the base is zero). Assuming time harmonic forcing: F = F0 eit ,
we let x = Xeit and y = Y eit , and solving for Y we find
F0 02 2 2i0

Y = . (3.117)
m 2 02 2i0
Define the drive impedance as the ratio of the force to the velocity at the drive point:
F0
,
Z= (3.118)
V
where V = iY is the velocity of the base. Thus,


im02 1 2i 0
Z= (3.119)
2 02 + 2i0
Now consider the same base with N attached spring-mass-dashpot systems (mj , kj , cj ), j =
1, 2, . . . , N , and the base is again subject to the forcing F . Let xj (t) represent the motion of mass
j, and y is the base motion, then we obtain N + 1 equations for the masses and the base:
mj x
j (t) = kj (xj y) cj (x j y),
j = 1, 2, . . . , N (3.120)
N
X  
0=F+ kj (xj y) + cj (x j y)
. (3.121)
j=1
 61

Assuming time harmonic forcing and motion, F = F0 eit , xj = Xj eit and y = Y eit , the
equation for mass j yields
 
j2 2ij j
Xj = Y. (3.122)
j2 2 2ij j
Substituting into the equation for the base, gives
N
X mj 2 (j2 2ij j )
F = Y. (3.123)
j=1
2 j2 + 2ij j

Hence, the drive point impedance Z = F0 /(iY ), is now

N im 2 1 2i
X j j j j
Z= (3.124)
j=1
2 j2 + 2ij j

or
N
X
Z= Zj , (3.125)
j=1

where the individual impedances are

imj j2 1 2i j j
Zj = (3.126)
2 j2 + 2ij j

It is perhaps not surprising that the impedances add, since the masses are attached in parallel.

3.5.2 Impedance and power

The power dissipated by the force is, at any given instant,

P (t) = F v (3.127)

With V = iY , this can be written

P (t) = Re(F0 eit ) Re(V eit )

1 1 1 1
= F0 V ei2t + F0 V ei2t + F0 V + F0 V (3.128)
4 4 4 4
where signifies the complex conjugate. Taking the average over one cycle, 0 t 2/, gives
1
P = Re(F0 V ) (3.129)
2
or
1
P = |V |2 Re Z . (3.130)
2
since the impedance is defined as Z = F0 /V .
In any physically realistic system, which does not generate its own power, a force that is applied
to the system should use external energy. That is, the work done - the integral of the power over
time - should be non-negative. The limit of zero is achieved for a system of zero damping. There,
eq. (3.130) implies that

Re Z 0, with equality only for a system of zero damping. (3.131)

 62

We may check this on the above example. Thus, from (3.126), we have

2mj j j 4
Re(Zj ) = , (3.132)
( 2 j2 )2 + (2j j )2

which is positive as long as cj = 2mj j j > 0. That is, power is dissipated if and only if the internal
damping is positive.

3.5.3 Impedance of a chain of masses

The drive point impedance of a linear chain can be derived, formally but not explicitly, using the
system defined in (3.13). Thus, assume the force is applied to mass m1 , and in order to be consistent
with the previous example, let k1 = 0, m1 = 0. We must then consider a system of size N 2 to
model a chain of N 1 masses. The condition at the right end of the chain can be free (kN +1 = 0)
or fixed. The force, applied only to mass 1, is {F } = (1, 0, 0, . . . , 0)T F0 eit . The impedance is
Z = F0 /(iX1 ), where x1 = X1 eit , or
1
Z= , k1 = 0, m1 = 0. (3.133)
i(K 2 M)1
11

Alternatively, based on (3.113)

1
Z=   . (3.134)
i( 1
2 2
nat
T )11

Impedance of a chain of masses: N

Consider the case of a chain that is free at the right end. Let the masses and springs be equal, then
in the limit as N the modes that need to be considered, according to the previous discussion
are the free-free modes of a bar. The non-normalized eigenvectors are
x
j (x) = cos(2j 1) , j = 1, 2, 3, . . . . (3.135)
2L
How do we normalize these? The inner product of (3.67) translates to the integral identity

ZL
i (x)j (x)1 (x)dx = ij , (3.136)
0

where 1 is the density per unit length, which is uniform in this case and equal to M/L, M being
the total mass (M = AL). Therefore, the eigenvalues and eigenvectors are, from (4.14), (3.135)
and (3.136)
r
c 2 x
j = (2j 1) , j (x) = cos(2j 1) , j = 1, 2, 3, . . . , (3.137)
2L M 2L
p
where c = E/.
The general result (3.134) becomes, in the -DOF limit,
1
Z= 2 (0) . (3.138)
P j
i j2 2
j=1
 63

In the particular case of the uniform bar, we get

M
Z= . (3.139)
P 1
i2 c

2
j=1 (2j1) 2L 2

or

1 X 1
= , (3.140)
Z Zj
j=1

with
M c
2
Zj = [ (2j 1) 2 ]. (3.141)
i2 2L

3.5.4 Impedance of a bar

Consider a uniform elastic bar of length L, cross-sectional area A, Youngs modulus E and density
. The end x = 0 is subject to the driving force F = F0 cos t, and the end x = L is free (zero
traction). The displacement is assumed to be of the form

u(x, t) = U (x) eit , (3.142)

so that equation (4.31) becomes

U (x) + k 2 U = 0, 0 < x < L, (3.143)

p
where k = /c, and c = E/. and the end conditions are

EAU (0) = F0 , EAU (L) = 0. (3.144)

These arise from the fact that the axial stress, xx , in the xdirection is EU , and the force is
Axx . The negative sign in the first of (3.144) arises from the fact that the normal at that end is in
the negative direction, and therefore the force acting in the positive xdirection is Axx nx = Axx
since nx = 1.
The general solution of (3.143) is U = a1 eikx + a2 eikx , where a1 , a2 are constants. These are
determined from the end conditions (3.144) as a2 = a1 ei2kL , and a1 = F0 /[EAik(1 ei2kL )]. Thus,

F0 eik(xL) + eik(xL)
ikL

U (x) = e (3.145)
EAik 1 eik2L
or, more simply,
F0 cos k(x L)
U (x) = (3.146)
EAk sin kL
F0
The impedance of the bar - more precisely the impedance of the drive point, is Z = iU (0) ,

Z = icA tan kL . (3.147)

This can be written in resonance form, i.e. a form that more clearly displays the resonances, by
noting that tan y is a meromorphic function (it only has poles) and it is bounded at infinity in the
complex plane. It can be shown that this implies it must be of the form
X dj
tan y = (3.148)
y yj
j
 64

where yn are the poles. These are clearly (n 12 ), n = 1, 2, . . .. We can obtain the values of
dj by equating the residues of (3.148) with those of the function at each pole. Thus, we find that
dj = 1 for all j, and hence

X 1 X 1
tan y = +
n=1
y (n 2 ) n=1 y + (n 12 )
1

X 2y
= (3.149)
n=1
y2 (n 21 )2 2

Therefore, the impedance (3.147) can be rewritten

X 1
Z = i2m (3.150)
n=1
(kL)2 (n 12 )2 2

where we have used k = /c, and m = AL is the total mass of the base. Alternatively, we can
write Z as

X imn n2
Z= (3.151)
2 n2
n=1

where n = (n 21 )c/L are the resonance frequencies of the free-free bar, and the masses are
defined as
2m
mn = (3.152)
(n 12 )2 2
This form of the impedance is identical in form to that for the finite system of uncoupled masses
in (3.125) and (3.126). w Damping can be included in the usual manner. For a SDOF system we
replace k by k ic. This translates to the replacement n2 n2 2in n in the multi-DOF
system, with k mn n2 and c 2mn n n .

3.5.5 General structures

The drive point impedance Z for any structure can be expressed in terms of its discrete modes, as

N
X
Z= Zj , (3.153)
j=1

where N for a continuous system - one with an infinite number of DOFs. The impedance
associated with each mode can be written, in general, as

imn n2 (1 21 n n )
Zn = . (3.154)
2 n2 + 2in n

This form is motivated by the previous results for the uncoupled masses, and for the bar, each of
which can be cast in this form. It is not difficult to recognize that (3.154) should be true in general,
based on the fact that the impedance must possess resonances, i.e.poles in the plane, and that
these poles must be in the lower half plane, see Appendix A.
Note that the real part of this expression disappears in the limit as n 0, except at resonance
frequency. We need to take care in taking this limit. Let y = 2 n2 , then dy = 2d. Let be a
 65

real positive number consider the integral

Z Z
id 1 idy (y i)
Re = Re
n2 + i
2 2 y 2 + 2

Z
1 dy
=
2 y2 + 2

Z
1 ds
=
2 s2 +1


= , (3.155)
2
independent of . Therefore in the limit of zero damping, we should interpret the real part of the
impedance as

Re Zn = mn n2 ( n ), n 0 (3.156)
2
where is the Dirac delta function.
The impedance, being complex valued, can be expressed
Z = R + iX (3.157)

3.6 Problems
1. Consider the system shown, with two masses, each m, and three springs each of
stiffness k.

(a) Write out the two equations for the system without forcing. Let uj (t) be
the displacement of mass j.

(b) Write the equations as a single matrix-vector equation and define the
matrices.

(c) Using (b), assume solutions with cos t and find the equation for the
modal frequency .

(d) p
Find the two modal frequencies 1 < 2 in terms of 0 , where 0 =
k/m.
!
(j) 1 (j)
(e) The modal vectors can be expressed as U = (j) . Find U2 , j = 1, 2.
U2

(f) Describe the two distinct ways one would release the system from rest so as to
generate each of these modes. That is, describe the initial displacements
necessary to give each separate mode, starting from rest.

(g) The system is released from rest with u1 (0) = 1 and u2 (0) = 0. Find
u1 (t) and u2 (t) for t 0. Hint: assume a sum of the two modes u =
A1 U(1) cos 1 t + A2 U(2) cos 2 t
 66

2. The |kmkmk| system shown has twomasses  and three springs. Let x1 (t) and x2 (t) be the
x1
displacements to the right, with u = . The equations of motion in the absence of forcing
x2
 
4 1
are mu + ku = 0. Consider the case m1 = m2 = 1 (in arbitrary units) and k =
1 4
(a) What are the values of the three spring stiffnesses kj , j = 1, 2, 3?
(b) Find the modal frequencies 1 and 2 (> 1 ).
 
U1
(c) Find the modal vector U = for mode 2.
U2
(d) This mode vector has a pretty simple form. Give (in words) a simple physical explanation
for this.

k1 k2 k3
m1 m2

Solution:

   
1 0 k1 + k2 k2
m= , k= , k1 = k3 = 3, k2 = 1
0 1 k2 k2 + k3

 
2 4 2 1
k m=
1 4 2
 
k 2 m = [( 2 4)2 1] = ( 2 3)( 2 5) = 0

so
12 = 3, 22 = 5

     
4 5 1 1 1 0
(k 22 m)U = U= U=
1 4 5 1 1 0
U1 + U2 = 0
U1 + U2 = 0

so  
1
U2 = U1 U(2) =
1
The masses are 180 out of phase. The middle spring extends/contracts symmetrically - its
center is fixed. Therefore each mass is like a SDOF with stiffness k from k1 in parallel with
2k2 , i.e. k = 3 + 2 1 = 5.

3. Systems (a), (b) and (c) have 2 masses m1 and m2 . (a) has 3 springs k1 , k2 , k3 , with k1 and
k3 attached to rigid supports. (b) is like (a) except k3 is removed. (c) has k1 and k3 removed.
Let u1 (t) and u2 (t) be the displacements of the masses to the right. No external forces act.
 67

(a) Write the equations of motion for system (a).

(b) Write the stiffness matrices for systems (a), (b) and (c).
(c) Consider (b) with m1 = m2 = 1kg, k1 = 2 N N 2 2
m , k2 = 1 m . Find 1 and 2 .
(d) One of the natural frequencies of system (c) is zero. Why?

(a)
k1 k2 k3
m1 m2
(b)

(c)

Solution:

m1 u
1 = k1 u1 k2 (u1 u2 ),
m2 u
2 = k3 u2 k2 (u2 u1 ),

so
   
m1 0 k1 + k2 k2
m= , k= , for (a),
0 m2 k2 k2 + k3

System (b) is the same with k3 = 0, and for (c) k1 = k3 = 0, so

   
k1 + k2 k2 k2 k2
k= , for (b), and , k= , for (c).
k2 k2 k2 k2

System (b) with m1 = m2 = 1 kg, k1 = 2 N/m, k2 = 1 N/m, implies

 
2 3 2 1  
k 2 m = ( 2 3)( 2 1) 1 = 4 4 2 + 2
k m= 2
1 1
so
4 1p 2
2 = 4 (4)(2) = 2 2 rad/s
2 2
System (c) can translate as a rigid body, which corresponds to a zero natural frequency.
Chapter 4

Continuous systems: -DOF

Realistic engineering systems are -DOF systems, also called continuous systems. Examples
abound: a string, a metal bar, a beam, a membrane, a volume of air (organ pipe), etc. Let
us look at strings.

4.1 The String

We consider a string tied/clamped at both ends and in tension, so that the string is characterized
by a tension T and a mass per unit length 1 . Note, 1 = A where is the mass per unit volume
and A the cross-sectional area (the subscript 1 indicates a 1D density). Each element of length dx
moves up and down. Let ((x) be the angle from the horizontal, then the net vertical or transverse
force acting is

T sin((x + dx)) T sin((x)) T (x + dx) T (x)

T (x) + dxT (x) T (x)
= dx T (x) (4.1)

where (x) = d/dx. Let y(x, t) be the transverse displacement then Newtons 2nd law for the
element of length dx, with mass 1 dx and acceleration y, is

1 dx y = dx T (x) (4.2)

or
1 y = T (x). (4.3)
In the same small angle approximation
= y (x, t) (4.4)
where y = y/x, so that (4.3) becomes

1 y = T y . (4.5)

Equivalently,
2y 2y
1 = T (4.6)
t2 x2
The main difference with the n-DOF system is that now instead of n distinct masses we have an
infinite number of them, at every value of x. We will see below in section 4.3 how to make the
connection clearer. But first let us look at the modes of the string.

68
 69

4.2 Modes of a string

Consider a string of length L stretched from x = 0 to x = L, and attached at both ends so that the
motion at the two ends is zero. We need to find homogeneous solutions of eq. (4.6) that satisfies
these boundary conditions. Start with the assumed separation of variables form

y(x, t) = Y (x)f (t) (4.7)

then (4.6) implies

1 Y f = T Y f, (4.8)
where the prime indicates differentiation with respect to x and the dot with respect to t. Thus,

f(t) T Y (x)
= , (4.9)
f (t) 1 Y (x)

and since each side depends on only one of either x or t they must be independent of both. That
is,
f(t) T Y (x)
= 2 , = 2 , (4.10)
f (t) 1 Y (x)
where is a constant. The constant has been chosen so that the equations look familiar, e.g. the
f equation becomes
f(t) + 2 f (t) = 0. (4.11)
The solution of this homogeneous equation for f is clearly sinusoidal, with frequency which
is as yet unknown. Thus, as with finite DOF systems, we could just as well start by assuming
time-harmonic time dependence:
y(x, t) = Y (x) cos t (4.12)
The values of at which solutions like this exist are the modal frequencies.
To find the modal frequencies, either use the second equation in (4.10) or substitute from (4.12)
into (4.6), to get
d2 Y
T 2 = 1 2 Y (4.13)
dx
or
d2 Y 2
+ Y =0 (4.14)
dx2 c2
where s
T
c= . (4.15)
1
The general solution of (4.14) is



Y (x) = A cos x + B sin x (4.16)
c c
The boundary conditions on the ends of the string at x = 0, x = L are assumed fixed, so that

Y (0) = 0, and Y (L) = 0, (4.17)

Which gives us two conditions to find the two unknowns A and B. The first implies A = 0, while
the second gives


B sin L = 0. (4.18)
c
 70

So, either B = 0, which is not very interesting, or sin cL = 0, which occurs if

L = , 2, 3, . . . (4.19)
c
The modal frequencies are therefore
c
j = j , j = 1, 2, 3, . . . (4.20)
L
There are several things to note. First, there is an infinite number of modes - which is to be
expected. Second, the modal frequencies are all an integer multiple of the first one: j = j1 . If
we plot the mode shapes - which are sine curves,
jx

Yj (x) = sin (4.21)
L
we see that the mode 2 has a node or a point in the middle where the displacement is always zero.
Similarly, mode 3 has 2 nodes, and in general mode j has j 1 nodes equally spaced along the
string.
What is perhaps surprising is that we can find all of the modal frequencies; remember - we had
to solve algebraic equations for the finite DOF systems!. Let us return to the n-DOF system and
see how it can help us understand the DOF system.
Equation (4.6) can be written so that it clearly depends only on the speed c:
2y 1 2y
= 0. (4.22)
x2 c2 t2
This is call the wave equation; see Section 4.7 for a discussion of wave motion, i.e. solutions of this
equation and the connection with modes.

4.3 -DOF chain

Think of the springs in the n-DOF chain of 3.2.2 becoming infinitely short, so that the spacing
between masses is h (constant). Let z be the coordinate along the chain, so that mass i is at
zi = ih. Then let
xi x(zi ), (4.23)
actually, x(zi , t), but we will leave the time dependence implicit for now. Thus,
h2
xi1 = x(zi1 ) = x(zi h) = x(zi ) hx (zi ) + x (zi ) + . . . (4.24)
2
Note that we are thinking of the displacements at neighboring positions as being almost the same,
and that we are also assuming that the displacement as s function of the position is a smoothly
varying function. Both of these are true for the string in the limit when we take very closely spaced
points on the string. Similarly, assume the stiffnesses and masses depend on the position in the
same smooth way, so that we can replace them by functions of the position:

ki k(zi ), mi m(zi ), (4.25)

so we now have continuous functions x, k and m instead of discrete values. Equation (3.14) becomes,
using (4.25) and similar ones for the stiffness and mass,
1 1 1 1
x+[2k+hk + h2 k +. . .]xk[xhx + h2 x +. . .][k+hk + h2 k +. . .][x+hx + h2 x +. . .] = 0.
m
2 2 2 2
(4.26)
 71

Here it is understood that the position is evaluated at z(= zi ) in all quantities, and the unwritten
terms are of order h3 . Simplifying we obtain

x (x k + x k)h2 + . . . = 0.
m (4.27)

Let the mass be m = 1 h, where 1 (z) is the mass per unit length in the chain, and let
kh = E1 (z), then the leading order limit of (4.27) becomes independent of h:

(x E1 + x E1 ) = 0.
1 x (4.28)

This can be rewritten

2x x

1 = E 1 . (4.29)
t2 z z
Consider a bar of cross-sectional area A, length L, under a force F . Its extension is L/L =
F/EA where E is the Youngs modulus of the material. Thus, for a section of length h the
stiffness is k = EA/h, and hence E1 = EA. Similarly, 1 = A is the mass density per unit volume
(the usual definition of density). Then we get

2x x

A = EA (4.30)
t2 z z
We have derived the equation of motion for longitudinal motion of a bar or rod. Remember, we
started with discrete springs and masses, and let them become vanishingly small, so that we end
up with an infinite DOF system, or a continuous system. The N discrete functions xi (t) becomes
the single function of two variables x(z, t).

4.4 Longitudinal modes of a bar

Consider a bar of length L between x = 0 and x = L. Let the displacement in the axial direction,
or longitudunal motion, be denoted by u(x, t), then the equation which follows from (4.30),

2u 2u
= E (4.31)
t2 x2
is similar to the equation for the string, but the boundary conditions can be different. If the bar is
clamped at both ends, then the situation is exactly the same as for the string, the modes are sine
curves and the modal frequencies are again given by (4.20) where now
s
E
c= (4.32)

Suppose however that the bar is free at both ends, which is described as free/free boundary
conditions. Then instead of the displacement being zero, the stress or force must vanish at the
ends. The stress is = Eu/x and the resultant force is A where A =cross-sectional area. Let

u(x, t) = U (x) cos t

then as before we find that

U (x) = A cos x + B sin x (4.33)
c c
The boundary conditions are now
dU dU
E (0) = 0 and E (L) = 0 (4.34)
dx dx
 72

Thus, B = 0 and the condition at x = L implies

A sin L =0 (4.35)
c
So again the modal frequencies are given by (4.20) but the mode shapes now are cosine curves.
A third possibility for the bar is that one end is clamped and the other free, call clamped/free.
In this case the conditions become
dU
U (0) = 0 and E (L) = 0 (4.36)
dx
Using (4.33), the first of (4.36) implies A = 0, and the second gives


B cos L =0 (4.37)
c
Hence,
3 5
L= , , , ... (4.38)
c 2 2 2
The modal frequencies are therefore
c
j = (2j 1) , j = 1, 2, 3, . . . (4.39)
2L

4.5 Back to discrete: The finite mk chain.

Consider the bar of length L, fixed at both end z = 0 and z = L, with uniform EA (independent
of z), then A drops out and we have the well known equation

2x 2x
E = 0. (4.40)
t2 z 2
p
The natural nondimensional variables are Z = z/L and = tL E/. Let x(z, t) = X(Z, , then
eq. (4.30) and the end conditions become

2X 2X
E = 0, Z(0, ) = 0, Z(1, ) = 0. (4.41)
2 Z 2
Let us reconsider the discrete N DOF system for the chain of springs and masses, with diagonal
mass matrix and stiffness matrix (3.13). For simplicity, let us assume that all masses are identical,
and all stiffnesses are identical. Then, using the relation which we determined above: mi = Ah,
and ki = EA/h, where h is the spacing, we find [M ] = Ah[I]N N and [K] = (EA/h)[J]N N
where the matrix J is defined as

2 1
1 2 1


1 2 1

[J] = ..
.. .. .. .. .. .. ..
(4.42)
.. .. .. .. .. .. .. ..

.. 1 2 1
1 2

So, the discrete equations [M ]{

x} + [K]{x} = 0 are
EA
Ah{
x} + [J]{x} = 0 (4.43)
h
 73

Introduce the nondimensional time (the natural thing to do) by letting {

x}(t) = {X}(t), then we
get
h2 2
{X} + [J]{X} = 0 (4.44)
L2 2
Note that L = (N + 1)h, so that the equation can be rewritten
2
{X} + (N + 1)2 [J]N N {X} = 0 (4.45)
2
What does this mean? Consider the lowest mode of the system - the lowest frequency. This
2 2
corresponds to 2 {X} = {X} where = , for example, if the end conditions are clamped.
This means that the lowest eigenvalue of [J]N N must scale as 1/(N + 1)2 . Numerical experimen-
tation shows that 1 , the lowest eigenvalue of [J] satisfies (N + 1)2 1 2 as N gets larger and
larger.
What about the case when the ends are free? This can be handled by setting k1 and kN +1 to
zero, resulting in the alternate form for [J]

1 1
1 2 1


1 2 1

[J] = ..
.. .. .. .. .. .. ..
, free-free chain. (4.46)
.. .. .. .. .. .. .. ..

.. 1 2 1
1 1

In this case we find that the lowest eigenvalue of [J] satisfies (N + 1)2 1 0 for N 2. The reason
for this can be seen directly by evaluating det[J], which can be seen to be identically zero by using
column or row addition and subtraction. Physically, this corresponds to the possibility of a rigid
body mode, in which the entire chain is translated with no relative motion between the masses. The
second eigenvalue of [J], 2 , satisfies (N + 1)2 2 2 as N becomes large.
If one end is free and the other fixed, this corresponds to, for instance,

1 1
1 2 1


1 2 1

[J] = ..
.. .. .. .. .. .. ..
, free-fixed chain. (4.47)
.. .. .. .. .. .. .. ..

.. 1 2 1
1 2
This corresponds to the left end free and the right end fixed, and we find that the lowest eigenvalue
of [J] satisfies (N + 1)2 1 (/2)2 as N .

4.6 Transverse vibration of beams

Beams vibrate in quite a different manner than the string or the bar. The reason is that the
equation of motion for the transverse motion of a beam is the equation of flexural motion. Let
y(x, t) be the transverse displacement, and let V be the shear force across the beam, M the bending
moment, A the cross-sectional area, and I the cross-sectional moment of inertia. The balance of
forces in the transverse direction is
dV
1 y = (4.48)
dx
 74

where 1 = A and is the mass per unit volume. The balance of moments is
dM
V = (4.49)
dx
so that
d2 M
A
y= (4.50)
dx2
The moment is related to the curvature or second derivative of y by the Euler-Bernoulli relation

d2 y
M = EI . (4.51)
dx2
The equation of flexural motion is therefore

d4 y
A
y = EI (4.52)
dx4
As before, let
y(x, t) = Y (x) cos t (4.53)
then Y satisfies
d4 Y A 2
4Y = 0 where 4 = (4.54)
dx4 EI
The general solution of (4.54) can be found by trying Y = ex , implying 4 = 4 . Therefore,
= , , i, i. Thus

Y (x) = A0 ex + B0 ex + C0 eix + D0 eix (4.55)

Using eix = cos x i sin x, and ex = cosh x sinh x (where cosh z = (ez + ez )/2 and
sinh z = (ez ez )/2) we can write

Y (x) = A cos x + B sin x + C cosh x + D sinh x (4.56)

It remains to apply boundary conditions in order to determine .

4.6.1 Examples
Bar hinged at both ends
This example is a bit academic but it shows some of the general properties of flexural modes.
Consider a beam of length L that is hinged at both ends, also called simply supported. This
means that the displacement and the moment are both zero at x = 0 and x = L, or

d2 Y d2 Y
Y (0) = 0, EI (0) = 0 and Y (L) = 0, EI (L) = 0 (4.57)
dx2 dx2
Using (4.56) and

dY

(x) = A sin x + B cos x + C sinh x + D cosh x (4.58)
dx
the conditions at x = 0 are

A + C = 0 and 2A + 2C = 0 (4.59)
 75

implying A = C = 0 and Y (x) = B sin x + D sinh x. The conditions at x = L become

B sin L + D sinh L = 0 (4.60)

2 2
B sin L + D sinh L = 0 (4.61)

Therefore
B sin L = 0 and D sinh L = 0 (4.62)
But sinh x = 0 only if x = 0, and so D = 0. We cannot have B = 0 so we must have sin L = 0 or

=n , n = 1, 2, 3, . . . (4.63)
L
q
EI
Note from (4.54) that = 2 A , and hence (4.63) gives us
s
2 EI
n = n 2 2 , n = 1, 2, 3, . . . (4.64)
L A

Note that
n = n 2 1
unlike the string and longitudinal motion in a bar (n = n1 ).
q As an example, consider a beam and a bar each of length L = 1 m and each of steel: c =
E
= 5500 m/s. The fundamental frequency, f1 = 1 /(2), for the bar in longitudinal motion is
f1 = c/(2L) or f1 = 2750 Hz. Assume the beam is of thickness
h m, so that I = Ah2 /12, then the
fundamental beam flexural frequency is f1 = 5500h/(2 12). E.g. if h = 2 cm then f1 = 15.88 Hz.
Note that the fundamental flexural frequency is much lower than the longitudinal frequency, which
is common in real structures. This means in practice that flexural modes and flexural vibration are
the more important source of vibration in engineering structures.

Bar free at both ends

At a free end both the shear force and the bending moment are zero. The four boundary conditions
now become
d2 Y d3 Y d2 Y d3 Y
EI (0) = 0, EI (0) = 0 and EI (L) = 0, EI (L) = 0 (4.65)
dx2 dx3 dx2 dx3
Assuming the general solution (4.56), then

d2 Y
(x) = 2 A cos x B sin x + C cosh x + D sinh x),
dx2 (4.66)
d3 Y
(x) = 3 A sin x B cos x + C sinh x + D cosh x).
dx3
Plugging these into the four conditions (4.65) gives four equation, which can be written in matrix
form (after dividing by common factors EI 2 and EI 3 )

1 0 1 0 A 0
0 1 0 1 B 0
cos L sin L cosh L sinh L C = 0 (4.67)

sin L cos L sinh L cosh L D 0

 76

Setting the determinant of the matrix to zero gives us the desired equation for . Use the fact that
the determinant is unchanged by adding rows and columns: add the first column to the third, and
the second to the fourth, to get

1 0 1 0
0 1 0 1
det
cos L sin L cosh L sinh L

sin L cos L sinh L cosh L

1 0 0 0
0 1 0 0
= det cos L sin L cosh L cos L sinh L sin L
(4.68)
sin L cos L sinh L + sin L cosh L cos L

cosh L cos L sinh L sin L
= det
sinh L + sin L cosh L cos L
= 2 2 cosh L cos L = 0
where we used cosh2 L sinh2 L = 1, cos2 L + sin2 L = 1.

Cantilever beam
One end is fixed (x = 0) and the other is free:
dY d2 Y d3 Y
Y (0) = 0, (0) = 0 and EI (L) = 0, EI (L) = 0 (4.69)
dx dx2 dx3
These become, as before,

1 0 1 0 A 0
0 1 0 1 B 0

cos L sin L cosh L sinh L C = 0 (4.70)

sin L cos L sinh L cosh L D 0

This time we subtract columns, to get

1 0 1 0

0 1 0 1
cos L sin L cosh L sinh L
sin L cos L sinh L cosh L (4.71)

cosh L + cos L sinh L + sin L
= det = 2 + 2 cosh L cos L = 0
sinh L sin L cosh L + cos L

Fixed-fixed
Both ends are fixed:
dY dY
Y (0) = 0, (0) = 0 and Y (L) = 0, (L) = 0 (4.72)
dx dx
implying
1 0 1 0 A 0
0 1 0 1 B 0
cos L sin L cosh L sinh L C = 0 (4.73)

sin L cos L sinh L cosh L D 0

 77

Hence,
cosh L cos L sinh L sin L
det = 2 2 cosh L cos L = 0 (4.74)
sinh L + sin L cosh L cos L

Summary of beam modes

The mode frequency is given by
(
1 free-free and fixed-fixed,
cosh L cos L = (4.75)
1 fixed-free.

The mode shapes (not normalized) can be expressed

(cos L cosh L)(sin x + sinh x) (sin L sinh L)(cos x + cosh x),
free-free,
Y (x) = (cos L cosh L)(sin x sinh x) (sin L sinh L)(cos x cosh x), fixed-fixed,


(sin L sinh L)(sin x sinh x) + (cos L + cosh L)(cos x cosh x), fixed-free.
(4.76)
Check that these satisfy the conditions (4.65), (4.69) and (4.72). The roots of (4.75) can be found
using this matlab script for s = 1.

s=-1;for n=1:8; x0=(n-s/2)*pi;x=fzero(@(x) cos(x)*cosh(x)+s, x0); [x,x-x0], end

This generates the first eight roots n L, and it also shows that n L (n 12 ) are very good
approximations to the higher order roots.

4.7 Wave motion

4.7.1 The 1D wave equation
The equations for transverse motion of the string and for longitudinal motion of the bar are each
of the form
2u 2
2 u
c = 0. (4.77)
t2 x2
The parameter c has units of length/time, i.e. velocity or speed, and it is known as the wave speed.
Thus, (p
T /1 in a string,
c= p (4.78)
E/ in a bar.
For instance, c = 5, 500 m/s for steel, and 6, 400 m/s for aluminum. The value of c is much less
in strings and wire. For instance a typical piano wire, 0.04 thick and 234 feet per lb has linear
density of 1 =6.4 102 kg/m. At a high tension value of T = 200 lb (= 890 N) the wave speed is
c =117 m/s. Most strings in musical instruments and elsewhere will have a lower value of c.
The reason c is called the wave speed is simple. Consider u(x, t) = f (x ct) where f (z) can
be any function that has a second derivative. Then u u
x = f (x ct) and t = cf (x ct) where
2 2
f (z) is just the derivative. Similarly, xu2 = f (x ct) and t2u = c2 f (x ct), so eq. (4.77) is
automatically satisfied. Consider u(x, t) = f (x ct) at two different times, t1 and t2 . and at two
different positions x1 , x2 . The value of x ct will be the same if x1 ct1 = x2 ct2 , that is, if
x2 x1 = c(t2 t1 ). In other words, for a given time difference, the difference in position required
 78

to get the same shape for u is defined by c, hence the shape or pulse propagates with velocity c,
the wave speed.
Note that we can also get solutions like u(x, t) = g(x + ct) where g(z) can be any function that
has a second derivative. In general,

u(x, t) = f (x ct) + g(x + ct), (4.79)

where f and g represent waves in opposite directions: f (x ct) is a wave traveling in the direction
of increasing x, to the right, say, and g(x + ct) is a wave traveling to the left.

4.7.2 Waves and modes

Consider (4.79) with f (z) = g(z) = B cos c z, then using well known formulas,
x x
u(x, t) = B cos( t) + B cos( t)
c c
x x
x x

= B cos cos t + sin sin t + B cos cos t sin sin t
c c c c
x
= 2B cos cos t. (4.80)
c
More generally, setting f (z) = g(z) = 21 B1 cos c z + 21 B2 sin c z gives
x x

u(x, t) = B1 cos + B2 sin cos t. (4.81)
c c
Comparing (4.81) with eqs. (4.12) and (4.16), it is clear that the modes of the string correspond
to the linear superposition of two waves traveling in opposite directions. This shows the fundamental
connection between waves and modes. Modes correspond to linear combinations of waves which
together satisfy the boundary conditions.

4.8 Problems
1. A string of length L under tension (force) T and mass per unit length m (= A where is
density, A cross-sectional area) is attached to grips at its two ends as shown.

x
(a) Write the equation of motion and boundary conditions for the transverse displace-
ment w.
(b) Assume w(x, t) = (A cos x + B sin x) cos t. Find A, B, and , and use this to
determine the modal frequencies n for free vibration.

(c) The modal frequency (f = 2 ) of the mode with shape indicated by the dashed line
above is 400 Hz. What is the modal frequency of the mode with shape indicated by the
solid curve?
(d) Draw the mode shape of the mode with frequency 800 Hz.
(e) Staying with the same string, suppose it is now pinned at A which is 23 of the way along
the string. What are the first two modal frequencies f1 , f2 of the section on the left
(between x = 0 and A)?
(f) The shorter section (to the right of A) is plucked so that its fundamental (first mode)
oscillates. Which, if any, mode on the left will be dominant, and why?
 79

A
x

Solution:

(a)
2w 2w
T m =0
t2 x2
w = 0 at x = 0, and w = 0 at x = L
p
(b) A = 0, B 6= 0, = n /c with c = T /m and n = nc/L
(c) f2 = 600 so f1 = 300 Hz

(d)
(e) n = nc/L where L = 23 L so n = 3nc/(2L)or fn = 23 fn . f1 = 300 so f1 = 450,
f2 = 900 and f3 = 1350 Hz
(f) The second mode because it has the same frequency

2. A bar of uniform cross-sectional area A, length L, density per unit volume, and Youngs
modulus is clamped at x = 0 and free at the other end. Let u(x, t) be the longitudinal
displacement at 0 x L at time t.

(a) Write the equation of motion for u.

(b) Write the boundary conditions for the problem.
(c) Find the modal frequencies n for free vibration.
(d) If the lowest frequency of resonance of an
q aluminum bar with L = 1 m is measured as
f1 = 3, 000 Hz, where = 2f , find c = E in m/s for aluminum.

Hint: The normal force is N = EA u

x and mass per unit length is A. For (2), assume
u(x, t) = U (x) cos t.

Solution:

(a)
2u 2u
A EA =0
t2 x2
u
(b) w = 0 at x = 0, and x = 0 at x = L
(c) u(x, t) = U (x) cos t implies, with the BCs, U (x) = sin x L
c where cos c = 0. So
L 1
c = (n 2 ), or
1 c
n = (n )
2 L
c c
(d) 1 = 2L so f1 = 4L , or c = 4Lf1 = 4(1)3, 000 = 12, 000 m/s
 80

Appendix

A Basic equations review

Complex numbers
i is defined as the square root of 1,

i = 1 = (1)1/2 ii = 1, i29 = i(1)14 = i, etc.
Any complex number z can be written z = x + iy where x and y are real (they can be positive,
p
negative or zero). The conjugate of z is z = x iy. The magnitude of z is |z| = x2 + y 2 .
If z1 = x1 + iy1 and z2 = x2 + iy2 , then z1 z2 = x1 x2 y1 y2 + i(x1 y2 + x2 y1 ). For instance,

z z = (x + iy)(x iy) = x2 + y 2 + i0 |z| = z z .
Any complex number can be reduced to the simple form z = x + iy. For example, z = zz21 =
x1 +iy1
x2 +iy2 . Multiply above and below by z 2 z = zz12 zz22 . But z2 z2 = |z2 |2 and using the product
formula for z1 z2
z1 x 1 x 2 + y1 y2 y1 x 2 y 2 x 1
= x + iy, where x = 2 2 , y= .
z2 x 2 + y2 x22 + y22
y
Any complex number can be expressed as z = |z|ei where tan = . We can see this by
x
using the basic identity
ei = cos + i sin .
x y x y
First write z as z = |z|( |z| +i |z| ). Both |z| and |z| are real numbers such that the sum of their squares
x y
= 1. Therefore, we can define such that cos = |z| and sin = |z| . Then z = |z|(cos + i sin ) =
sin y
|z|ei and cos = tan = x .
Multiplication and division by complex numbers is much simpler in the form z = |z|ei . For ex-
ample, z1 = |z1 |ei1 and z2 = |z2 |ei2 z1 z2 = |z1 ||z2 |ei(1 +2 ) , z11 = |z11 | ei1 , z16.2 = |z1 |6.2 ei6.21 ,
z1 |z1 | i(1 2 )
z2 = |z2 | e , etc.

Trigonometry formulas
1
Start with ei = cos + i sin , ei = cos i sin , then add them cos = (ei + ei ) ,
2
1 i
subtract them sin = (e ei ) . Multiply them ei(1 +2 ) = cos(1 + 2 ) + i sin(1 +
2i
2 ). But e i( 1 + 2 ) = e e 2 = (cos 1 + i sin 1 )(cos 2 + i sin 2 ) and multiply out ei(1 +2 ) =
i 1 i

cos 1 cos 2 sin 1 sin 2 + i(sin 1 cos 2 + sin 2 cos 1 ). Comparing the real and imaginary parts

cos(1 + 2 ) = cos 1 cos 2 sin 1 sin 2 , sin(1 + 2 ) = sin 1 cos 2 + sin 2 cos 1 .

How can you express cos 3 in terms of cos and sin ? A quick solution is to use ei3 =
(cos + i sin )3 = (cos )3 + 3(cos )2 i sin + 3 cos (i sin )2 + (i sin )3 . Taking the real part gives
cos 3 = (cos )3 3 cos (sin )2 . Other basic identities can be found in the same way.
Interesting math fact: five of the most important numbers (1, 0, , e, i) are related by the formula
1 + ei = 0.
 81

B Linear algebra 101

B.1 Basic matrix algebra
Algebra involves multiplication (division is a special case of multiplication). We can multiply
matrices: if A and B are m n and n p then AB is m p. Note the matrix on the left has to
have the same number of columns as the number of rows of the matrix on the right. If A and B
are 3 2 and 2 4 then AB is 3 4 but BA does not make sense.
A matrix is square if the number of rows and columns is equal. If A and B are both square
n n matrices, then AB and BA are square n n matrices but in general

AB 6= BA (B.1)

The order of multiplication is important!

The identity n n matrix I has the property that AI = A and IA = A for any n n matrix
A.
The inverse of a square n n matrix A is a square n n matrix A1 such that AA1 = I and
A1 A = I. Not every square n n matrix has an inverse, e.g.
 
1 0
(B.2)
0 0

The condition that the inverse exists is that det A 6= 0. Another way of saying this is that the
inverse does not exist if det A = 0.
If A is an n n matrix and v is an n 1 vector and

Av = 0 (B.3)

then there are two possibilities: (i) the inverse of exists A, or (ii) it does not. If (1) then multiply
by A1 , so
A1 Av = 0 Iv = 0 (B.4)

but Iv = v so v = 0. In conclusion, the only way to get a nonzero v is if (ii) holds, i.e. there is no
inverse matrix. A necessary condition for the matrix to have no inverse is that

det A = 0. (B.5)

Another common matrix-vector equation is

Bv = v, (B.6)

where B is an n n matrix and v is an n 1 vector. This is in the form (B.3) with A = B I.

So the condition that (B.6) has a non-zero v is that det(B I) = 0. Generally, a matrix times a
vector is not proportional (parallel) to the vector. So (B.6) is a special relation, for a given matrix
B, it only works for certain vectors, called the eigenvectors, and the associated values of are the
eigenvalues. An n n matrix can have n distinct eigenvalues, each with its own eigenvector.
In the n-dof vibration problem, A = K 2 M and the condition for a mode is (see eq. (3.12),
(3.34))
det(K 2 M) = 0. (B.7)
 82

B.2 Matlab
In matlab, if A is an n n matrix, then eig(A) returns the n eigenvalues. Do help eig to see
how you can get the eigenvectors using eig. So in principle, we just need to use the matrix M1 K.
However, the function eig is even better. If k and m are the matrices in matlab, then a=eig(k,m)
returns a vector of the eigenvalues, i.e. a(1) = 12 , . . . a(n) = n2 .
Try running the following:

m=eye(3)
k = [2 -1 0; -1 2 -1; 0 -1 2]
[V,D]=eig(k,m)

The output is

m =
1 0 0
0 1 0
0 0 1

k =
2 -1 0
-1 2 -1
0 -1 2

V =
0.5000 -0.7071 -0.5000
0.7071 0.0000 0.7071
0.5000 0.7071 -0.5000

D =
0.5858 0 0
0 2.0000 0
0 0 3.4142

The eigenvalues are on the diagonal of the diagonal matrix D, with D(1, 1) = 12 , . . . D(n, n) = n2 .
The columns of the matrix V are the eigenvalues

C Lagranges Equations
Deriving the equations of motion is often a non-trivial task, especially for multi-DOF systems. We
start by analyzing the forces, and then equate the force with the mass times acceleration for each
DOF. Lagranges equations provide an alternate procedure to get the equations of motion, and is
generally preferable when there are many DOFs. For simplicity, here we will discuss the procedure
for conservative systems, that is, with no damping.

C.1 SDOF
Consider a mass/spring system, and define the kinetic and potential energies

1 1
T = mx 2 , V = kx2 ,
2 2
 83

respectively. The Lagrangian is defined as

L = T V.

Note that it is not T + V which is the total energy of the system. L can be considered as a function
that depends on x, on x and on t. Lagranges equation for the mass/spring system is
 
d L L
=0 (C.1)
dt x x

Let us see what this equation yields. First, the only place that x occurs in L is in V and so

L V
= = kx (C.2)
x x
Similarly, x only appears in T , and so

L T
= = mx (C.3)
x x
Then,  
d L d
= mx = m
x (C.4)
dt x dt
Combining eqs. (C.1)-(C.4) gives
m
x + kx = 0 (C.5)
Summary: Once we have an expression for the Lagrangian L(x, x) then we plug it into the
Euler-Lagrange equation (C.1). The only real work that we have to do is (1) construct L, and (2)
do some differentiation. Constructing L means writing out expressions for the KE and PE, which
is always easier than dealing with free body diagrams, writing forces, etc. The real power of the
method is for n-DOF. In these problems dealing with forces can be extremely difficult, particularly
if there are moments and forces, linear and angular momentum. In an n-DOF system, each mass
can be acted on by many forces, most of which depend upon the other n 1 coordinates.

C.2 n-DOF
The power of Lagranges equations is the applicability to arbitrary systems, of one, two or an
infinite number of DOFs. Let us do some examples.

Two masses and springs

For instance, consider a two DOF system of two mass spring systems (m1 , k1 ) and (m2 , k2 ) with
a spring kc connecting the masses. We have seen this system before in class, and know that the
equations are

m1 x
1 = k1 x1 kc (x1 x2 ) (C.6)
m2 x
2 = k2 x2 kc (x2 x1 ) (C.7)

The same equations can be derived by first writing out the Lagrangian, again defined as L = T V ,
where now the KE and the PE depend on the two coordinates and their time derivatives:
1 1 1 1 1
T = m1 x 21 + m2 x 22 , V = k1 x21 + k2 x22 + kc (x1 x2 )2 (C.8)
2 2 2 2 2
 84

The final term in V comes from the fact that the extension/compression of the coupling spring is
(x1 x2 ).
Lagranges equations for a multi-DOF Lagrangian L(x1 , x 1 , . . . , xn , x n , t) are
 
d L L
= 0, i = 1, 2, . . . , n (C.9)
dt x i xi
Taking i = 1, we have
L T
= = m1 x 1 (C.10)
x 1 x 1
and
L V
= = k1 x1 kc (x1 x2 ) (C.11)
x1 x1
Therefore,  
d L L
= m1 x
1 + k1 x1 + kc (x1 x2 ) = 0 (C.12)
dt x 1 x1
which is exactly equation (C.6). The other equation follows taking i = 2 in (C.9).

The double pendulum

As a second example, consider the double pendulum, i.e. one pendulum of mass m1 , length l1 , with
another pendulum, m2 , l2 , hanging from m1 . Let 1 and 2 be the angles from the vertical for each
pendulum. The elevation of the first mass from its equilibrium position is l1 (1 cos 1 ) and its PE
is therefore m1 gl1 (1 cos 1 ). The elevation of the second mass from its equilibrium position is
l1 (1 cos 1 ) + l2 (1 cos 2 ) because it is raised up by the first pendulum also. The total PE is
therefore
V = (m1 + m2 )gl1 (1 cos 1 ) + m2 gl2 (1 cos 2 ) (C.13)
The KE is
1 1
T = m1 |v1 |2 + m2 |v2 |2 (C.14)
2 2
where vi is the vector velocity of mass i. The string holding the first mass is attached to a fixed
point, and v1 = l1 1 e1 so that |v1 |2 = (l1 1 )2 . The velocity of the second mass is v2 = v1 + l2 2 e2
and therefore
|v2 |2 = |l1 1 e1 + l2 2 e2 |2 = (l1 1 )2 + (l2 2 )2 + 2l1 l2 1 2 cos(1 2 )
Thus,
1 1
T = (m1 + m2 )(l1 1 )2 + m2 [(l2 2 )2 + 2l1 l2 1 2 cos(1 2 )] (C.15)
2 2
Note that we have not made any small angle approximations yet. The theory is exact no matter
how big or small the angles.
The Lagranges equations for this system are
 
d L L
= 0, i = 1, 2, (C.16)
dt i i
where, as usual, L = T V , and L = L(1 , 1 , 2 , 2 ). This just means that our independent
coordinates are 1 and 2 (n=2). Note that in this case, the KE depends on the coordinates 1 and
2 , in addition to the angular velocities. This makes the differentiation a little bit more complicated
but we can be confident that the final equations are exact: thus
(m1 + m2 )l1 1 + m2 l2 2 cos(1 2 ) + m2 l2 22 sin(1 2 ) + (m1 + m2 )g sin 1 = 0 (C.17)
m2 l2 2 + m2 l1 1 cos(1 2 ) + m1 l1 12 sin(2 1 ) + m2 g sin 2 = 0 (C.18)
 85

Small angle approximation

In the above example we have not assumed the usual small angle approximation. This amounts to
assuming that all angle and angular velocities are small, so that
sin i i , cos i 1
Applying this to (C.17) and (C.18), with cos(1 2 ) = 1 and sin(2 1 ) = 2 1 , we get
(m1 + m2 )l1 1 + m2 l2 2 + m2 l2 22 (1 2 ) + (m1 + m2 )g1 = 0 (C.19)
m2 l2 2 + m2 l1 1 + m1 l1 12 (2 1 ) + m2 g2 = 0 (C.20)
We can now consistently neglect the nonlinear terms 22 (1 2 ) and 12 (2 1 ), so that the linearized
equations are
(m1 + m2 )l1 1 + m2 l2 2 + (m1 + m2 )g1 = 0 (C.21)
m2 l2 2 + m2 l1 1 + m2 g2 = 0 (C.22)
We can also obtain the linearized equations (small angle approx) by first approximating the PE
and KE in eqs. (C.13) and (C.15). In this case we need to retain the terms that quadratic in the
small quantities, but ignore those of higher order. Thus, we need to use cos i 1 12 i2 so that
we get
1 1
T = (m1 + m2 )(l1 1 )2 + m2 [(l2 2 )2 + 2l1 l2 1 2 ] (C.23)
2 2
1 1
V = (m1 + m2 )gl1 12 + m2 gl2 22 (C.24)
2 2
Applying the Lagrange equations (C.16) directly now gives the linearized equations (C.21) and
(C.22).
It is worth noting that the [M ] and [K] matrices of (C.21) and (C.22) correspond to

(m1 + m2 )l1 m2 l2 (m1 + m2 )g 0
[M ] = , [K] = (C.25)
m2 l1 m2 l2 0 m2 g
This might look unusual, because the mass matrix is not diagonal - it contains off-diagonal terms.
However, we can rearrange (C.21) and (C.22), by combining them, to get
m1 l1 1 + (m1 + m2 )g1 m2 g2 = 0 (C.26)
m1 l2 2 (m1 + m2 )g1 + (m1 + m2 )g2 = 0 (C.27)
Divide the first by m2 and the second by (m1 + m2 ),
m1 m1
l 1 1 + ( + 1)g1 g2 = 0 (C.28)
m2 m2
m1 l2
2 g1 + g2 = 0 (C.29)
m1 + m2
or
m1 m1
m2 l 1 0 ( m2 + 1)g g
[M ] = , [K] = (C.30)
m 1 l2
0 m1 +m2 g g
Now the mass matrix is diagonal and the stiffness is not. Comparison of (C.25) and (C.30) shows
that the mass and stiffness matrices are not unique, which is not surprising because we can always
rearrange systems of linear equations. However, no matter how we rearrange them the modes and
modal frequencies are not changed.
 86

C.3 General coordinates

The real potential of Lagranges equations is for systems with coordinates that are more general.
It is common to call the generalized coordinates for an n-DOF system qi , i = 1, 2, 3, . . . , n. Then
we express the PE and KE in terms of qi and qi , and the n Lagranges equations are
 
d L L
= 0, i = 1, 2, . . . , n (C.31)
dt qi qi

For instance, consider a pendulum of length r mass m2 that hangs from the mass m1 of a
mass/spring system of stiffness k. The mass m1 moves horizontally with coordinate x and the
mass m2 swings from the moving mass m1 with angle from the vertical. The two generalized
coordinates are q1 = x and q2 = .
The KE and PE of the system are
1 1 1
T = m1 x 2 + m2 (x 22 + y 22 ), V = kx2 + m2 gy2
2 2 2
where (x2 , y2 ) are the rectangular coordinates of m2 relative to its equilibrium position: x2 =
x + r sin , y2 = r(1 cos ). Hence and hence

1 1 1
T = m1 x 2 + m2 [(x + r cos )2 + (r sin )2 ], V = kx2 + m2 gr(1 cos ) (C.32)
2 2 2
The 2 equations of motion then follow from the lagrange equations
   
d L L d L L
= 0, =0 (C.33)
dt x x dt

In the small angle approximation, we can simplify T and V of (C.32) to become

1 1 2, 1 1
T = m1 x 2 + m2 (x + r) V = kx2 + m2 gr2 (C.34)
2 2 2 2
It is easier to find the equations of motion in the small angle approximation. Applying (C.33) gives

x + m2 r + kx = 0
(m1 + m2 ) (C.35)
m2 ( + m2 g = 0
x + r) (C.36)

C.4 Origin of Lagranges equations (extra)

What is the basis for Lagranges equations? One way is to derive them from Newtons 2nd Law.
A more general and more profound basis is the so-called Principle of Least Action. The idea is
to define the action as the integral Z
I Ldt

Then the actual solution of a dynamical system is the one which minimizes the action. That is,
we consider all possible qi (t) and minimize the integral with respect to these. Since there are n-
variables, we get n minimization conditions, and each one is the Lagrange equation for that variable.
The actual minimization proceeds as follows:
Z
I L L qi

= + dt
qi qi qi qi
 87

The second term is simplified by integrating by parts (ignoring the end contributions to the integral):
Z Z Z
L qi d L
qi d L

dt = dt = dt
qi qi dt qi qi dt qi
and so Z
I  d L
L 
= dt
qi dt qi qi
The integrand has to be zero at each instant, hence we get the Lagrange equations. This powerful
method of deriving equations is also known as Variational Calculus, or the Calculus of Variations,
and is at the heart of Finite Element Methods.

Problems
1
1. The system shown consists of a bar of mass m, length L, centroidal MoI IG = 12 mL2 sup-
ported by springs at its ends, plus a mass supported from the center by a spring. Using the
three coordinates shown, write out in full in terms of m, k, L, xj , x j and x
j , j = 1, 2, 3:
(a) The expression for the kinetic energy.
(b) The expression for the potential energy.
(c) The three equations of motion.
Rigid body: T = 21 m|x G |2 + 12 IG 2
Ignore gravity + horizontal motion. Assume small
d L L
angles. Lagrange eqs. dt qj qj = 0, L = T V

Solution
(a)
1 1 1
T = mx 2c + IG 2 + mx 23
2 2 2
1 x2 x1
where xc = 2 (x1 + x2 ) and is the angle of rotation of the bar: = L . So

1 1 1
T = m(x 1 + x 2 )2 + m(x 1 x 2 )2 + mx 23
8 24 2

(b)
1 1 1
V = 4kx21 + 4kx22 + k(xc x3 )2
2 2 2
Hence
1
V = 2k(x21 + x22 ) + k(x1 + x2 2x3 )2
8

(c) Since T depends only on the velocities and V only on the coordinates, the Lagrange eqs.
d T V
become dt x j + xj = 0, j = 1, 2, 3. Hence:

1 1 1
j=1: m(
x1 + x
2 ) + m(
x1 x2 ) + 4kx1 + k(x1 + x2 2x3 ) = 0,
4 12 4
1 1 1
j=2: m(
x1 + x
2 ) + m(
x2 x1 ) + 4kx2 + k(x1 + x2 2x3 ) = 0,
4 12 4
1 1
j=2: m
x3 k(x1 + x2 2x3 ) = 0
2 2
 88

Can simplify to
4 2 k
j=1: x
1 + x 2 + (17x1 + x2 2x3 ) = 0,
3 3 m
2 4 k
j=2: x
1 + x 2 + (x1 + 17x2 2x3 ) = 0,
3 3 m
k
j=2: x
3 + (2x3 x1 x2 ) = 0
m