@Seismicisolation

@Seismicisolation
@Seismicisolation
@Seismicisolation
The Practice of Engineering Dynamics

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
The Practice of Engineering Dynamics

Ronald J. Anderson
Queen’s University
Kingston, Canada

@Seismicisolation
@Seismicisolation
This edition first published 2020
© 2020 John Wiley & Sons Ltd

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form
or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how
to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions.

The right of Ronald J. Anderson to be identified as the author of this work has been asserted in accordance with law.

Registered Offices
John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

Editorial Office
The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www
.wiley.com.

Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that appears in
standard print versions of this book may not be available in other formats.

Limit of Liability/Disclaimer of Warranty

In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of
information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and
evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or
device for, among other things, any changes in the instructions or indication of usage and for added warnings and
precautions. While the publisher and authors have used their best efforts in preparing this work, they make no
representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically
disclaim all warranties, including without limitation any implied warranties of merchantability or fitness for a particular
purpose. No warranty may be created or extended by sales representatives, written sales materials or promotional
statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or
potential source of further information does not mean that the publisher and authors endorse the information or services
the organization, website, or product may provide or recommendations it may make. This work is sold with the
understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained
herein may not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers
should be aware that websites listed in this work may have changed or disappeared between when this work was written
and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial
damages, including but not limited to special, incidental, consequential, or other damages.

Library of Congress Cataloging-in-Publication data

Names: Anderson, Ron J. (Ron James), 1950- author.

Title: The practice of engineering dynamics / Ronald J Anderson, Queen's
University, Kingston, Canada.
Description: First edition. | Hoboken, NJ, USA : John Wiley & Sons, Inc.,
[2020] | Includes bibliographical references and index.
Identifiers: LCCN 2020004354 (print) | LCCN 2020004355 (ebook) | ISBN
9781119053705 (hardback) | ISBN 9781119053682 (adobe pdf) | ISBN
9781119053699 (epub)
Subjects: LCSH: Machinery, Dynamics of. | Mechanics, Applied.
Classification: LCC TJ170 .A53 2020 (print) | LCC TJ170 (ebook) | DDC
620.1/04–dc23
LC record available at https://lccn.loc.gov/2020004354
LC ebook record available at https://lccn.loc.gov/2020004355

Cover Design: Wiley

Cover Image: © kovop58/Shutterstock

Set in 9.5/12.5pt STIXTwoText by SPi Global, Chennai, India

Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY

10 9 8 7 6 5 4 3 2 1

@Seismicisolation
@Seismicisolation
To June, Stacey, and Kate

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 vii

Contents

Preface xi
About the Companion Website xv

Part I Modeling: Deriving Equations of Motion 1

1 Kinematics 3
1.1 Derivatives of Vectors 3
1.2 Performing Kinematic Analysis 5
1.3 Two Dimensional Motion with Constant Length 6
1.4 Two Dimensional Motion with Variable Length 8
1.5 Three Dimensional Kinematics 10
1.6 Absolute Angular Velocity and Acceleration 13
1.7 The General Acceleration Expression 14
Exercises 16

2 Newton’s Equations of Motion 19

2.1 The Study of Motion 19
2.2 Newton’s Laws 19
2.3 Newton’s Second Law for a Particle 20
2.4 Deriving Equations of Motion for Particles 21
2.5 Working with Rigid Bodies 25
2.6 Using F⃗ = m⃗a in the Rigid Body Force Balance 26
⃗
2.7 Using F⃗ = ddtG in the Rigid Body Force Balance 28
2.8 Moment Balance for a Rigid Body 30
2.9 The Angular Momentum Vector – H ⃗ O 33
2.10 A Physical Interpretation of Moments and Products of Inertia 36
2.11 Euler’s Moment Equations 40
2.12 Throwing a Spiral 41
2.13 A Two Body System 42
2.14 Gyroscopic Motion 48
Exercises 52

@Seismicisolation
@Seismicisolation
viii Contents

3 Lagrange’s Equations of Motion 55

3.1 An Example to Start 55
3.2 Lagrange’s Equation for a Single Particle 58
3.3 Generalized Forces 62
3.4 Generalized Forces as Derivatives of Potential Energy 64
3.5 Dampers – Rayleigh’s Dissipation Function 65
3.6 Kinetic Energy of a Free Rigid Body 67
3.7 A Two Dimensional Example using Lagrange’s Equation 70
3.7.1 The Kinetic Energy 70
3.7.2 The Potential Energy 71
3.7.3 The 𝜃 Equation 72
3.7.4 The 𝜙 Equation 73
3.8 Standard Form of the Equations of Motion 73
Exercises 74

Part II Simulation: Using the Equations of Motion 77

4 Equilibrium Solutions 79
4.1 The Simple Pendulum 79
4.2 Equilibrium with Two Degrees of Freedom 80
4.3 Equilibrium with Steady Motion 81
4.4 The General Equilibrium Solution 84
Exercises 85

5 Stability 87
5.1 Analytical Stability 87
5.2 Linearization of Functions 92
5.3 Example: A System with Two Degrees of Freedom 95
5.4 Routh Stability Criterion 99
5.5 Standard Procedure for Stability Analysis 103
Exercises 105

6 Mode Shapes 107

6.1 Eigenvectors 107
6.2 Comparing Translational and Rotational Degrees of Freedom 111
6.3 Nodal Points in Mode Shapes 115
6.4 Mode Shapes with Damping 116
6.5 Modal Damping 118
Exercises 122

7 Frequency Domain Analysis 125

7.1 Modeling Frequency Response 125
7.2 Seismic Disturbances 132
7.3 Power Spectral Density 133

@Seismicisolation
@Seismicisolation
 Contents ix

7.3.1 Units of the PSD 138

7.3.2 Simulation using the PSD 139
Exercises 143

8 Time Domain Solutions 145

8.1 Getting the Equations of Motion Ready for Time Domain Simulation 146
8.2 A Time Domain Example 147
8.3 Numerical Schemes for Solving the Equations of Motion 149
8.4 Euler Integration 149
8.5 An Example Using the Euler Integrator 151
8.6 The Central Difference Method: An 𝒪(h2 ) Method 153
8.7 Variable Time Step Methods 155
8.8 Methods with Higher Order Truncation Error 157
8.9 The Structure of a Simulation Program 159
Exercises 163

Part III Working with Experimental Data 165

9 Experimental Data – Frequency Domain Analysis 167

9.1 Typical Test Data 167
9.2 Transforming to the Frequency Domain – The CFT 169
9.3 Transforming to the Frequency Domain – The DFT 172
9.4 Transforming to the Frequency Domain – A Faster DFT 174
9.5 Transforming to the Frequency Domain – The FFT 175
9.6 Transforming to the Frequency Domain – An Example 176
9.7 Sampling and Aliasing 179
9.8 Leakage and Windowing 184
9.9 Decimating Data 187
9.10 Averaging DFTs 189
Exercises 189

A Representative Dynamic Systems 193

A.1 System 1 193
A.2 System 2 193
A.3 System 3 194
A.4 System 4 194
A.5 System 5 195
A.6 System 6 195
A.7 System 7 196
A.8 System 8 197
A.9 System 9 197
A.10 System 10 198
A.11 System 11 198
A.12 System 12 199

@Seismicisolation
@Seismicisolation
x Contents

A.13 System 13 200

A.14 System 14 200
A.15 System 15 201
A.16 System 16 201
A.17 System 17 202
A.18 System 18 202
A.19 System 19 203
A.20 System 20 203
A.21 System 21 204
A.22 System 22 204
A.23 System 23 205

B Moments and Products of Inertia 207

B.1 Moments of Inertia 207
B.2 Parallel Axis Theorem for Moments of Inertia 208
B.3 Parallel Axis Theorem for Products of Inertia 210
B.4 Moments of Inertia for Commonly Encountered Bodies 210

C Dimensions and Units 213

D Least Squares Curve Fitting 215

Index 219

@Seismicisolation
@Seismicisolation
 xi

Preface

The design of a mechanical system very often includes a requirement for dynamic analysis.
During the early concept design stages it is useful to create a mathematical model of the
system by deriving the governing equations of motion. Then, simulations of the behavior
of the system can be produced by solving the equations of motion. These simulations give
guidance to the design engineers in choosing parameter values in their attempt to create a
system that satisfies all of the performance criteria they have laid out for it.
There is a logical progression of analyses that are required during the design. The design
engineer needs to determine, from the nonlinear differential equations of motion:
● The equilibrium states of the system – these are places where, once put there and not

disturbed, the system will stay. The time varying terms are removed from the differential
equations of motion, leaving a set of nonlinear algebraic equations. The solutions to these
equations provide knowledge of all of the equilibrium states.
● The stability of the equilibrium states – the question here is: if the system is disturbed

slightly from an equilibrium state, will it try to get back to that state or will it move
farther away from it? It is usually not good practice to design systems around unstable
equilibrium states since the system will always tend to move towards a stable equilib-
rium condition. Answering the stability question involves a linearization of the equations
of motion for small perturbations away from the equilibrium states.
● How the system behaves around a stable equilibrium state – the study of small motions

of a mechanical system around a stable equilibrium state lies in the realm of vibrations
and leads to predictions of natural frequencies, mode shapes, and damping ratios, each
of which is very useful during the design process. The linearized differential equations of
motion are used.
● The response to harmonically applied external forces – systems, in stable equilibrium, are

often subjected to harmonic external disturbances at known forcing frequencies and their
response to these forces provides critical design information. The linearized equations of
motion are used.
● The response of the system in the time domain – the fully nonlinear equations of

motion are solved numerically to simulate the response of the system to known external
forces. Large scale motions and the nonlinear characteristics of system elements are
included. This is the numerical equivalent to conducting performance experiments with

@Seismicisolation
@Seismicisolation
xii Preface

a prototype of the system. The design information gleaned from these simulations is,
perhaps surprisingly to some, not very useful in the early stages of the design. Accurate
nonlinear simulations require precise knowledge of system parameters that simply
isn’t available in the early design phases of a project. The time domain simulations are
best left to the prototype testing stage when they can be validated through comparison
of predicted and measured system response. Validated time domain simulations are
valuable tools to use when considering design changes aimed at improving the measured
performance of the system.

The presentation of material in this book divides the practice of engineering dynamics
into three parts.

Part 1. Modeling: Deriving Equations of Motion

Dynamic analysis is based on the use of accurate nonlinear equations of motion for a sys-
tem. Deriving these complicated equations is a task that is prone to error. Because of this, it
is important to derive the governing equations twice, using two different methods of anal-
ysis, and then prove to yourself that the two sets of equations are the same. This is a time
consuming activity but is vital because predictions made from the equations of motion are
critical in the design process. Predictions made using equations with errors are not of any
use. The first part of the book discusses the generation of nonlinear equations of motion
using, firstly, Newton’s laws and, secondly, Lagrange’s equation. Only when the two meth-
ods give the same equations can the analyst proceed to part 2 with confidence.

Part 2. Simulation: Using the Equations of Motion

The second part presents a logical progression of analysis techniques and methods applied
to the governing equations of motion for systems. The progression is from equilibrium solu-
tions that find in what states the system would like to be, to analyzing the stability of these
equilibrium states (stability is usually considered only in textbooks on control systems but
it is vitally important to dynamic systems), to considering small motions about the stable
equilibrium states (this topic is covered in textbooks on vibrations but is, again, vital to engi-
neers doing dynamic analysis), to frequency domain analysis (vibrations again), and finally
to time domain solutions (these are rarely covered in textbooks).

Part 3. Working with Experimental Data

While not usually considered a part of the design process, analysis of experimental data
measured on dynamic systems is critical to creating a successful product. To assist engineers
in developing capabilities in this area, part 3 covers the practical use of discrete fourier
transforms in analyzing experimental data.
In order to emphasize the idea that any dynamic mechanical system can be analyzed
using the sequence of steps presented here, all the exercises at the ends of the chapters are
based on 23 mechanical systems defined in an appendix. Any one of these systems could be
used as an example of all of the types of dynamic analysis.

@Seismicisolation
@Seismicisolation
 Preface xiii

This book is based on course notes that I have developed while teaching a one-semester
graduate course on dynamics over more than two decades. It could just as well be used in
a senior undergraduate dynamics course.

November, 2019 Ronald J. Anderson

Kingston, Canada

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 xv

About the Companion Website

This book is accompanied by a companion website:

www.wiley.com/go/anderson/engineeringdynamics

The website includes:

● Animations
● Fully worked examples
● Software
Scan this QR code to visit the companion website.

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 1

Part I

Modeling: Deriving Equations of Motion

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 3

Kinematics

Kinematics is defined as the study of motion without reference to the forces that cause the
motion. A proper kinematic analysis is an essential first step in any dynamics problem.
This is where the analyst defines the degrees of freedom and develops expressions for the
absolute velocities and accelerations of the bodies in the system that satisfy all of the phys-
ical constraints. The ability to differentiate vectors with respect to time is a critical skill in
kinematic analysis.

1.1 Derivatives of Vectors

Vectors have two distinct properties – magnitude and direction. Either or both of these prop-
erties may change with time and the time derivative of a vector must account for both.
The rate of change of a vector ⃗r with respect to time is therefore formed from,
( )
1. The rate of change of magnitude d⃗ r
.
( ) dt m

2. The rate of change of direction d⃗ r

dt
.
d

Figure 1.1 shows the vector ⃗r (t) that changes after a time increment, Δt, to ⃗r (t + Δt).
The difference between ⃗r (t) and ⃗r (t + Δt) can be defined as the vector q⃗ (t) shown in
Figure 1.1 and, by the rules of vector addition,
⃗r (t) + q⃗ (t) = ⃗r (t + Δt) (1.1)
or,
q⃗ (t) = ⃗r (t + Δt) − ⃗r (t). (1.2)
Then, using the definition of the time derivative,
d⃗r ⃗r (t + Δt) − ⃗r (t) q⃗ (t)
= lim = lim . (1.3)
dt Δt→0 Δt Δt→0 Δt

Imagine now that Figure 1.1 is compressed to show only an infinitesimally small time
interval, Δt. The components of q⃗ (t) for the interval Δt are shown in Figure 1.1. They are,
1. A component d⃗qm aligned with the vector ⃗r . This is a component that is strictly due to
̇ where ṙ is the rate of
the rate of change of magnitude of ⃗r . The magnitude of d⃗qm is rΔt
The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.
© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
4 1 Kinematics

Figure 1.1 A vector changing with time.

⇀
r(t+Δt) ⇀
q(t) dq⇀d

ϕ
⇀
dq⇀m
r(t)

change of length (or magnitude) of the vector ⃗r . The direction of d⃗qm is the same as the
direction of ⃗r . Let d⃗qm be designated1 as ⃗rΔt.
̇
2. A component d⃗qd that is perpendicular to the vector ⃗r . That is, a component due to the
rate of change of direction of the vector. Terms of this type arise only when there is an
angular velocity. The rate of change of direction term arises from the time rate of change
of the angle 𝜙 in Figure 1.1 and 𝜙̇ is the magnitude of the angular velocity of the vector.
The rate of change of direction therefore arises from the angular velocity of the vector. The
magnitude of d⃗qd is 𝜙rΔt ̇ where r is the length of ⃗r . By definition the rate of change of
̇
the angle 𝜙 (i.e. 𝜙) has the same positive sense as the angle itself. It is clear that 𝜙ṙ is the
“tip speed” one would expect from an object of length r rotating with angular speed 𝜙.̇
The angular velocity is itself a vector quantity since it must specify both the angular speed
(i.e. magnitude) and the axis of rotation (i.e. direction). In Figure 1.1, the speed of rotation
is 𝜙̇ and the axis of rotation is perpendicular to the page. This results in an angular velocity
vector,
⃗ = 𝜙̇ k⃗
𝜔 (1.4)
⃗ is defined in Figure 1.2. Note that it is
where the right handed set of unit vectors, (⃗𝚤, ⃗𝚥, k),
essential that right handed coordinate systems be used for dynamic analysis because of the
extensive use of the cross product and the directions of vectors arising from it. If there is a
⃗ then the cross
right handed coordinate system (x, y, z), with respective unit vectors (⃗𝚤, ⃗𝚥, k),
products are such that,
⃗𝚤 × ⃗𝚥 = k⃗

⃗𝚥 × k⃗ = ⃗𝚤

k⃗ × ⃗𝚤 = ⃗𝚥.
Using this definition of the angular velocity, the motion of the tip of vector ⃗r , resulting
from the angular change in time Δt, can be determined from the cross product
⃗ × ⃗r )Δt
(𝜔
which, by the rules of the vector cross product, has magnitude,
|𝜔||⃗ ̇
⃗ r |Δt = 𝜙rΔt

1 The convention used here is that a vector with an overdot such as ⃗ṙ is used to represent the rate of change
of magnitude of the vector and the overdot is not to be interpreted as a shorthand method of signifying the
total derivative of the vector. For a scalar function there is only a magnitude and the overdot will represent
its rate of change.

@Seismicisolation
@Seismicisolation
 1.2 Performing Kinematic Analysis 5

Figure 1.2 Even 2D problems are 3D. · ⇀

⇀ 𝜔
⇀
=𝜙k
k
⇀
j

⇀ ϕ ⇀
r(t+Δt)
i
⇀
r(t) ⇀
q(t)

dq⇀m dq⇀d

and a direction that, according to the right hand rule2 used for cross products, is perpendic-
⃗ and ⃗r and, in fact, lies in the direction of d⃗qd .
ular to both 𝜔
Combining these two terms to get q⃗ (t) and substituting into Equation 1.3 results in,
d⃗r ⃗rΔt
̇ + (𝜔 ⃗ × ⃗r )Δt ⃗
= lim = ṙ + 𝜔 ⃗ × ⃗r . (1.5)
dt Δt→0 Δt
The time derivative of any vector, ⃗r , can therefore be written as,
d⃗r ⃗ṙ
= + ⃗ × ⃗r
𝜔 (1.6)
dt ⏟⏟⏟ ⏟⏟⏟
rate of change rate of change
of magnitude of direction .
of the vector of the vector
It is important to understand that the angular velocity vector, 𝜔,
⃗ is the angular velocity
of the coordinate system in which the vector, ⃗r , is expressed. There is a danger that the
rate of change of direction terms will be included twice if the angular velocity of the vector
with respect to the coordinate system in which it measured is used instead. The example
presented in Section 1.3 shows a number of different ways to arrive at the derivative of a
vector which rotates in a plane.

1.2 Performing Kinematic Analysis

Before proceeding with examples of kinematic analyses we state here the steps that are
necessary in achieving a successful result. This first step in any dynamic analysis is vitally
important. The goal is to derive expressions for the absolute velocities and accelerations of the
centers of mass of the bodies making up the system being analyzed. In addition, expressions
for the absolute angular velocities and angular accelerations of the bodies will be required.

⃗ and B
2 Let the cross product of two vectors, A ⃗
⃗ , be the vector, C

⃗ ×B
A ⃗
⃗ = C.
⃗ is the direction aligned with the thumb of your right
According to the right hand rule, the direction of C
⃗ and curl your fingers towards B
hand if you point that hand in the direction of A ⃗.

@Seismicisolation
@Seismicisolation
6 1 Kinematics

It is at this first step of the analysis that degrees of freedom are defined and constraints on
relative motion between bodies are satisfied.
For this general description of kinematic analysis, we assume that we are analyzing a
system that has multiple bodies connected to each other by joints and that we are attempting
to derive an expression for the acceleration of the center of mass of a body that is not the
first in the assembly.
The procedure is as follows.
1. Find a fixed point (i.e. one having no velocity or acceleration) in the system from which
you can begin to write relative position vectors that will lead to the centers of mass of
bodies in the system.
2. Define a position vector that goes from the fixed point, through the first body, to the next
joint in the system. This is the position of the joint relative to the fixed point.
3. Determine how many degrees of freedom, both translational and rotational, are required
to define the motion of the relative position vector just defined. The degrees of freedom
must be chosen to satisfy the constraints imposed by the joint that connects this body to
ground.
4. Define a coordinate system in which the relative position vector will be written and deter-
mine the angular velocity of the coordinate system.
5. Repeat the previous three steps as you go from joint to joint in the system, always being
careful to satisfy the joint constraints by defining appropriate degrees of freedom.
6. When the desired body is reached, define a final relative position vector from the joint
to the center of mass.
7. The sum of all the relative position vectors will be the absolute position of the center
of mass and the derivatives of the sum of vectors will yield the absolute velocity and
acceleration of the center of mass.

1.3 Two Dimensional Motion with Constant Length

Figure 1.3 shows a rigid rod of length, 𝓁, rotating about a fixed point, O, in a plane. An
expression for the velocity of the free end of the rod, P, relative to point O is desired.
By definition, the velocity of P relative to O is the time derivative of the position of P
relative to O. This position vector is designated p⃗ P∕O and is shown in the figure.
In order to differentiate the position vector, we must have an expression for it and this
means we must first choose a coordinate system3 in which to work. For a start, we can
choose a right handed coordinate system fixed in the ground. The set of unit vectors
(⃗𝚤0 , ⃗𝚥0 , k⃗0 ) is such a system. The angular velocity of this coordinate system is zero (i.e.
𝜔⃗ 0 = 0⃗) since it is fixed in the ground.
An expression for the position of P relative to O in this system is,
p⃗ P∕O = 𝓁 cos 𝜃⃗𝚤0 + 𝓁 sin 𝜃⃗𝚥0 . (1.7)

3 Three dimensional sets of unit vectors shown in two dimensional figures such as Figure 1.3 will be
shown with the positive sense of the vector out of the plane represented by a curved arrow using the right
hand rule (e.g. k⃗0 in Figure 1.3).

@Seismicisolation
@Seismicisolation
 1.3 Two Dimensional Motion with Constant Length 7

⇀
Figure 1.3 A rigid rod rotating about a fixed ⇀ i1
point. j1

⇀
k1
P

⇀
PP/O

⇀
j0
θ ⇀
k0
O ⇀
i0

We apply Equation 1.6 to p⃗ P∕O to get,

d
𝑣⃗P∕O = (⃗p ) = p⃗̇ P∕O + 𝜔⃗ 0 × p⃗ P∕O
dt P∕O
d
= (𝓁 cos 𝜃⃗𝚤0 + 𝓁 sin 𝜃⃗𝚥0 ) + 0⃗ × (𝓁 cos 𝜃⃗𝚤0 + 𝓁 sin 𝜃⃗𝚥0 ). (1.8)
dt
In this coordinate system, it is clear that there is a rate of change of magnitude of the vector
only and the velocity of point P relative to O after performing the simple differentiation is,
𝑣⃗P∕O = −𝓁 𝜃̇ sin 𝜃⃗𝚤0 + 𝓁 𝜃̇ cos 𝜃⃗𝚥0 = 𝓁 𝜃(−
̇ sin 𝜃⃗𝚤0 + cos 𝜃0 ⃗𝚥0 ). (1.9)
Another derivation of the velocity of P relative to O might use the system of unit vectors
(⃗𝚤1 , ⃗𝚥1 , k⃗1 ) that are fixed in the rod. The advantage of using this system is that the position
vector is easily expressed as,
p⃗ P∕O = 𝓁⃗𝚤1 . (1.10)
Note that the length of this vector is a constant so that the total derivative must come
from its rate of change of direction. The angular velocity of the coordinate system is equal
to the angular velocity of the rod since the coordinate system is fixed in the rod. That is,
⃗ 1 = 𝜃̇ k⃗1
𝜔 (1.11)
and the velocity of P relative to O is therefore4 ,
d ̇ 𝚤 ) + (𝜃̇ k⃗ ) × (𝓁⃗𝚤 ).
𝑣⃗P∕O = (⃗p ) = p⃗̇ P∕O + 𝜔
⃗ 1 × p⃗ P∕O = (𝓁⃗ (1.12)
dt P∕O 1 1 1

Since 𝓁 is constant, 𝓁̇ = 0, and the final result is,

𝑣⃗P∕O = (𝜃̇ k⃗1 ) × (𝓁⃗𝚤1 ) = 𝓁 𝜃⃗
̇ 𝚥1 . (1.13)

4 Readers are encouraged to review the rules for cross multiplication of vectors.

@Seismicisolation
@Seismicisolation
8 1 Kinematics

We now have two expressions for 𝑣⃗P∕O (Equations 1.9 and 1.13). Since there can only be
one value of this relative velocity, the two expressions must be equal to each other. However,
the use of two different coordinate systems makes them look different. In order to compare
them, we must be able to transform results from one coordinate system to the other.
Keep in mind that sequential sets of unit vectors are related to each other by simple plane
rotations. Also note that the unit vectors are not related to any point in the system – they
simply express directions. Given these two facts, we can relate the two sets of unit vectors
we have been using by noting that k⃗1 = k⃗0 (i.e. the plane rotation relating the two sets is a
rotation about the k⃗0 or k⃗1 axis). The relationships between the two sets of unit vectors can
be expressed as follows.
⎧ ⃗𝚤 ⎫ ⎡ cos 𝜃 sin 𝜃 0 ⎤ ⎧ ⃗𝚤 ⎫
⎪ 1⎪ ⎢ ⎪ 0⎪
⎨ ⃗𝚥1 ⎬ = ⎢ − sin 𝜃 cos 𝜃 0 ⎥⎥ ⎨ ⃗𝚥0 ⎬ (1.14)
⎪ k⃗ ⎪ ⎣ 0 0 1⎦⎪ ⃗ ⎪
⎩ 1⎭ ⎩ k0 ⎭
or
⎧ ⃗𝚤 ⎫ ⎡ cos 𝜃 − sin 𝜃 0 ⎤ ⎧ ⃗𝚤 ⎫
⎪ 0⎪ ⎢ ⎪ 1⎪
⎨ ⃗𝚥0 ⎬ = ⎢ sin 𝜃 cos 𝜃 0 ⎥⎥ ⎨ ⃗𝚥1 ⎬ . (1.15)
⎪ k⃗ ⎪ ⎣ 0 0 1⎦⎪ ⃗ ⎪
⎩ 0⎭ ⎩ k1 ⎭
The first transformation (Equation 1.14) can be used with Equation 1.13 to show that,
̇ 𝚥1 = 𝓁 𝜃(−
𝑣⃗P∕O = 𝓁 𝜃⃗ ̇ sin 𝜃⃗𝚤0 + cos 𝜃⃗𝚥0 ). (1.16)
This is the same result as that shown in Equation 1.9.
Similarly, the second transformation (Equation 1.15) can be used with Equation 1.9
to show that,
𝑣⃗P∕O = −𝓁 𝜃̇ sin 𝜃⃗𝚤0 + 𝓁 𝜃̇ cos 𝜃⃗𝚥0
= −𝓁 𝜃̇ sin 𝜃(cos 𝜃⃗𝚤1 − sin 𝜃⃗𝚥1 ) + 𝓁 𝜃̇ cos 𝜃(sin 𝜃⃗𝚤1 + cos 𝜃⃗𝚥1 ). (1.17)
Expanding this yields,
̇ sin 𝜃 cos 𝜃⃗𝚤1 + sin2 𝜃⃗𝚥1 + cos 𝜃 sin 𝜃⃗𝚤1 + cos2 𝜃⃗𝚥1 ) = 𝓁 𝜃⃗
𝑣⃗P∕O = 𝓁 𝜃(− ̇ 𝚥1 . (1.18)
This is the same result as that shown in Equation 1.13.
The transformations described in this example are typical of those used in dynamic
analysis. Dynamicists are prone to using whatever coordinate system is appropriate at the
time and, sometimes, there are many intermediate coordinate systems used in deriving
the final system. Nevertheless, each coordinate system in the sequence must be right
handed and must be generated by a simple plane rotation from the preceding system.

1.4 Two Dimensional Motion with Variable Length

Figure 1.4 shows a rigid body rotating in a plane about a fixed point O. The body has a slot
cut in it and a small object A slides in this slot. Expressions for the velocity and acceleration
of A are desired.

@Seismicisolation
@Seismicisolation
 1.4 Two Dimensional Motion with Variable Length 9

⇀
Figure 1.4 A slider in a slot. i
⇀
j

⇀
k

A
X

Since the distance from O to A changes with time, we start by defining a variable distance
⃗ fixed in the rotating body, as shown, is
x from O to A. A set of rotating unit vectors (⃗𝚤, ⃗𝚥, k),
appropriate for this analysis since the position vector p⃗ A∕O is aligned with ⃗𝚤, thereby making
it easy to write.
The angular velocity of the body is not specified in magnitude but the fact that the body
⃗ We assume that the angu-
rotates in a plane fixes the direction of the angular velocity to be k.
lar velocity is,

⃗ = 𝜔k⃗
𝜔

where 𝜔 is not constant so that 𝜔̇ (i.e. the rate of change of magnitude of the angular velocity
vector) exists.
The position of A with respect to O is then,

p⃗ A∕O = x⃗𝚤

and, differentiating this, we find the velocity of A with respect to O to be,

d d ⃗ × (x⃗𝚤) = x⃗
𝑣⃗A∕O = ̇ 𝚤 + (𝜔k)
(⃗p ) = (x⃗𝚤) = x⃗ ̇ 𝚤 + 𝜔x⃗𝚥. (1.19)
dt A∕O dt
The acceleration of the slider relative to point O is defined to be,
d
a⃗ A∕O = (𝑣⃗ )
dt A∕O

@Seismicisolation
@Seismicisolation
10 1 Kinematics

or
d ⃗ × (x⃗
a⃗ A∕O = ̇ 𝚤 + 𝜔x⃗𝚥) = (̈x⃗𝚤 + 𝜔x⃗
(x⃗ ̇ 𝚥) + (𝜔k)
̇ 𝚥 + 𝜔x⃗ ̇ 𝚤 + 𝜔x⃗𝚥)
dt
= (̈x − 𝜔 x)⃗𝚤 + (𝜔x
2
̇ 𝚥.
̇ + 2𝜔x)⃗ (1.20)
Since both the velocity 𝑣⃗A∕O and the acceleration a⃗ A∕O are relative to the fixed or inertial
point O, they are in fact the absolute velocity and acceleration of point A. We commonly
write absolute velocities and accelerations without subscripts yielding,
̇ 𝚤 + 𝜔x⃗𝚥
𝑣⃗A = x⃗ (1.21)
and
a⃗ A = (̈x − 𝜔2 x)⃗𝚤 + (𝜔x ̇ 𝚥.
̇ + 2𝜔x)⃗ (1.22)

1.5 Three Dimensional Kinematics

Figure 1.5 shows a three degree of freedom robot. The horizontal arm AB is of fixed length
𝓁 and is free to rotate about a vertical axis through point A with angular speed 𝜔0 . Arm BC
has a variable length x and is free to rotate about an axis passing through points A and B
with angular speed 𝜔1 . The end effector is located at point C. Of interest for the kinematic
analysis are expressions for the absolute velocity and acceleration of the end effector.
As a first step we define the right handed coordinate system (⃗𝚤0 , ⃗𝚥0 , k⃗0 ) fixed in the arm
AB. This is a rotating coordinate system with angular velocity 𝜔 ⃗ 0 = 𝜔0 k⃗0 .
The process of finding the absolute velocity and acceleration of point C is just as outlined
in Section 1.2. The first step is to find a fixed point. In this system, point A serves the purpose
as it has no velocity or acceleration.
The next step is to define a position vector that goes from A to C. This is the vector P⃗ C∕A
shown in Figure 1.6. Notice that it goes through space from A to C and its rate of change
cannot be described directly using the motions of the physical components of the system.
We must, in fact, work with relative position vectors that go from the fixed point to the
point of interest by passing from joint to joint. In this case, we define first a vector that goes
from A to B (P⃗ B∕A ) and then add to it a vector that goes from B to C (P⃗ C∕B ). That is,
p⃗ C∕A = p⃗ B∕A + p⃗ C∕B . (1.23)

C Figure 1.5 A three dimensional robot.

⇀
ω0 k0
x ⇀

A θ j0
⇀
i0
B ω1

@Seismicisolation
@Seismicisolation
 1.5 Three Dimensional Kinematics 11

Figure 1.6 Relative position vectors. C

⇀
pC/A ⇀
k0
x ⇀

p⇀ C/B j0
A θ
⇀
p⇀B/A i0
𝓁
B

By definition, the absolute velocity of C is the time rate of change of its position with
respect to a fixed point. That is,
d
𝑣⃗C = (⃗p ) (1.24)
dt C∕A
which, upon substitution of Equation 1.23, becomes,
d d
𝑣⃗C = (⃗p ) + (⃗pC∕B ) (1.25)
dt B∕A dt
which in turn becomes,
𝑣⃗C = 𝑣⃗B∕A + 𝑣⃗C∕B (1.26)
where we see that the absolute velocity of C can be expressed as the sum of the velocities of
points in the vector chain relative to previous points in the chain so long as the first point
is stationary.
Simply differentiating Equation 1.26 with respect to time gives the corresponding expres-
sion for accelerations.
a⃗ C = a⃗ B∕A + a⃗ C∕B . (1.27)
With respect to the particular system being considered here, we can write,
p⃗ B∕A = 𝓁 ⃗𝚤0 (1.28)
and,
p⃗ C∕B = −x sin 𝜃 ⃗𝚥0 + x cos 𝜃 k⃗0 . (1.29)
Considering first the position of B with respect to A we see that the length 𝓁 is constant so
there will be no rate of change of magnitude of the vector but there will be a rate of change
of direction since the coordinate system is rotating. We find,
( )
𝑣⃗B∕A = 𝜔0 k⃗0 × (𝓁 ⃗𝚤0 ) = 𝜔0 𝓁 ⃗𝚥0 . (1.30)

We differentiate again, noting that 𝜔0 is not constant so that there will be a rate of change
of magnitude this time, and find,
( )
a⃗ B∕A = (𝜔̇ 0 𝓁 ⃗𝚥0 ) + 𝜔0 k⃗0 × (𝜔0 𝓁 ⃗𝚥0 )
= 𝜔̇ 0 𝓁 ⃗𝚥0 − 𝜔20 𝓁 ⃗𝚤0 . (1.31)
The rate of change of p⃗ C∕B is a little more complicated for two reasons. First, the vector
is not in the body to which the coordinate system being used is fixed so there will be a rate

@Seismicisolation
@Seismicisolation
12 1 Kinematics

of change of magnitude arising from the time derivatives of the trigonometric functions.
Second, the vector itself is not of fixed length so terms involving the rate of change of x will
appear. Differentiating yields,
( )
𝑣⃗C∕B = −ẋ sin 𝜃 − x𝜃̇ cos 𝜃 ⃗𝚥0
( )
+ ẋ cos 𝜃 − x𝜃̇ sin 𝜃 k⃗0
( )
+ 𝜔0 k⃗0 × (−x sin 𝜃 ⃗𝚥0 + x cos 𝜃 k⃗0 ) (1.32)

which, noting that 𝜃̇ = 𝜔1 (see Figure 1.5), expands to,

𝑣⃗C∕B = (x𝜔0 sin 𝜃) ⃗𝚤0 + (−ẋ sin 𝜃 − x𝜔1 cos 𝜃) ⃗𝚥0
+ (ẋ cos 𝜃 − x𝜔1 sin 𝜃) k⃗0 . (1.33)
Finally, we differentiate 𝑣⃗C∕B to get a⃗ C∕B as follows.
( )
̇ 0 sin 𝜃 + x𝜔0 𝜃̇ cos 𝜃 ⃗𝚤0
a⃗ C∕B = x𝜔̇ 0 sin 𝜃 + x𝜔
( )
+ −̈x sin 𝜃 − ẋ 𝜃̇ cos 𝜃 − x𝜔̇ 1 cos 𝜃 − x𝜔̇ 1 cos 𝜃 + x𝜔1 𝜃̇ sin 𝜃 ⃗𝚥0
( )
̇ 1 sin 𝜃 − x𝜔̇ 1 sin 𝜃 − x𝜔1 𝜃̇ cos 𝜃 k⃗0
+ ẍ cos 𝜃 − ẋ 𝜃̇ sin 𝜃 − x𝜔
( )
+ 𝜔0 k⃗0 × 𝑣⃗C∕B (1.34)

which, after considerable effort and again noting that 𝜃̇ = 𝜔1 , expands to,
( )
a⃗ C∕B = x 𝜔̇ 0 sin 𝜃 + 2 ẋ 𝜔0 sin 𝜃 + 2 x 𝜔0 𝜔1 cos 𝜃 ⃗𝚤0
( )
+ −̈x sin 𝜃 − 2 ẋ 𝜔1 cos 𝜃 − x 𝜔̇ 1 cos 𝜃 + x(𝜔20 + 𝜔21 ) sin 𝜃 ⃗𝚥0
( )
+ ẍ cos 𝜃 − 2 ẋ 𝜔1 sin 𝜃 − x 𝜔̇ 1 sin 𝜃 − x 𝜔21 cos 𝜃 k⃗0 . (1.35)
If you have worked through the derivation of Equation 1.35 you will be aware that the
probability of making a mistake when deriving equations such as this is high. A quick,
approximate check on the accuracy of your work can be made by verifying that every term
in the acceleration expression has dimensions of acceleration or, more simply, contains
two derivatives of displacement variables. That is, terms like ẍ are obviously accelerations
whereas terms like x 𝜔̇ 0 might require a little thought before realizing that 𝜔̇ 0 is the sec-
ond derivative of an angle and must be scaled by a length, in this case x, in order to be
a translational acceleration. Products of angular velocities such as x 𝜔0 𝜔1 and x 𝜔21 have
two derivatives of angles multiplied together and are again scaled by a length, x, to become
translational accelerations. In addition, it is somewhat comforting to see several terms that
have the Coriolis factor of 2 associated with them. Suspicion should be raised when factors
other than 1 or 2 are seen in acceleration expressions.
We now simply add the relative acceleration vectors to arrive at the absolute acceleration
of point C. That is,
a⃗ C = a⃗ A + a⃗ B∕A + a⃗ C∕B
( )
= −𝜔20 𝓁 + x 𝜔̇ 0 sin 𝜃 + 2 ẋ 𝜔0 sin 𝜃 + 2 x 𝜔0 𝜔1 cos 𝜃 ⃗𝚤0
( ( ) )
+ 𝜔̇ 0 𝓁 − ẍ sin 𝜃 − 2 ẋ 𝜔1 cos 𝜃 − x 𝜔̇ 1 cos 𝜃 + x 𝜔20 + 𝜔21 sin 𝜃 ⃗𝚥0
( )
+ ẍ cos 𝜃 − 2 ẋ 𝜔1 sin 𝜃 − x 𝜔̇ 1 sin 𝜃 − x 𝜔21 cos 𝜃 k⃗0 . (1.36)

@Seismicisolation
@Seismicisolation
 1.6 Absolute Angular Velocity and Acceleration 13

1.6 Absolute Angular Velocity and Acceleration

When working with three dimensional dynamic systems it is important to have an expres-
sion for the absolute angular velocity vector for a rigid body in order to be able to write
an expression for its angular momentum vector. The angular momentum is required for
moment balances.
Relative angular velocity vectors can be added together in the same way that relative
velocity vectors were in Section 1.5. That is, having established the angular velocity of one
body in a chain of bodies with respect to a stationary body (i.e. the absolute angular veloc-
ity of the body), we simply go through the chain adding the relative angular velocity of
neighboring bodies as we pass through the joints connecting them.
For example, the absolute angular velocity of body BC in Figure 1.5 can be determined as
follows,

𝜔
⃗ BC = 𝜔
⃗ AB + 𝜔
⃗ BC∕AB (1.37)

where the joint at A constrains AB to rotate about a vertical axis relative to the ground, so
that,

⃗ AB = 𝜔0 k⃗0
𝜔 (1.38)

is the absolute angular velocity of AB. The joint at B constrains BC to rotate about the axis
of AB with an angular velocity that is relative to AB giving,

⃗ BC∕AB = 𝜔1 ⃗𝚤0 .
𝜔 (1.39)

Substituting Equations 1.38 and 1.39 into Equation 1.37 gives the absolute angular veloc-
ity of BC

⃗ BC = 𝜔1 ⃗𝚤0 + 𝜔0 k⃗0 .
𝜔 (1.40)

The absolute angular acceleration of BC is, by definition, the time rate of change of the
absolute angular velocity vector of BC. In this example we note that the angular velocity
vector is expressed in a rotating coordinate system so that there will be both a rate of change
of magnitude and a rate of change of direction. The coordinate system has angular velocity
⃗ 0 = 𝜔0 k⃗0 . Using the symbol 𝛼 for angular acceleration we can write,
𝜔

d
𝛼⃗BC = (𝜔 ⃗𝚤 + 𝜔0 k⃗0 ) (1.41)
dt 1 0
which becomes, upon differentiation,

𝛼⃗BC = (𝜔̇ 1 ⃗𝚤0 + 𝜔̇ 0 k⃗0 ) + (𝜔0 k⃗0 ) × (𝜔1 ⃗𝚤0 + 𝜔0 k⃗0 ). (1.42)

The final result, after performing the cross-multiplication in Equation 1.42, is that the
absolute angular acceleration of BC is,

𝛼⃗BC = 𝜔̇ 1 ⃗𝚤0 + 𝜔0 𝜔1 ⃗𝚥0 + 𝜔̇ 0 k⃗0 . (1.43)

@Seismicisolation
@Seismicisolation
14 1 Kinematics

1.7 The General Acceleration Expression

In Section 1.4 we derived an acceleration expression for a very specific example. The final
result (shown in Equation 1.20) has an interesting and, perhaps, unexpected form. In par-
ticular, the origin of the term that has twice the product of an angular velocity and a trans-
̇ is not immediately obvious. The origin of all of the acceleration
lational velocity (i.e. 2𝜔x)
terms in a general expression like that in Equation 1.20 is described below. The description
is offered twice – first in a mathematical form then in a graphical form.
In general, the derivation of an expression for the acceleration of a point (say P) relative
to another point (say O) starts with the position vector of P with respect to O and then
differentiates it twice. Each differentiation must take account of the angular velocity of the
coordinate system being used to express the vectors.
Let the position vector be
P⃗ P∕O . (1.44)
Then, applying Equation 1.6, the velocity is,

𝑣⃗P∕O = P⃗̇ P∕O + 𝜔

⃗ × P⃗ P∕O (1.45)
⏟⏟⏟ ⏟⏞⏟⏞⏟
radial tangential

where the directions of the two components are defined. The rate of change of magnitude
term is aligned with the position vector and is thus termed radial and the rate of change of
direction component is perpendicular to the position vector and is therefore tangential.
Differentiating the velocity expression of Equation 1.45 yields,
d ⃗̇ d
a⃗ P∕O = (P ) + (𝜔 ⃗ × P⃗ P∕O )
dt P∕O dt
= (P⃗̇ P∕O ) + (𝜔)
d d d
⃗ × P⃗ P∕O + 𝜔
⃗ × (P⃗ P∕O ) (1.46)
dt dt dt
and, applying Equation 1.6 to Equation 1.46, we can write,
d𝜔
⃗ ⃗
a⃗ P∕O = (P⃗̈ P∕O + 𝜔
⃗ × P⃗̇ P∕O ) + ⃗ × (P⃗̇ P∕O + 𝜔
× PP∕O + 𝜔 ⃗ × P⃗ P∕O ). (1.47)
dt
d𝜔
⃗
After collecting terms and substituting 𝛼⃗ (the angular acceleration vector) for , we get
dt
the well-known result,
a⃗ P∕O = P⃗̈ P∕O + 𝛼⃗ × P⃗ P∕O + 2𝜔
⃗ × P⃗̇ P∕O + 𝜔 ⃗ × P⃗ P∕O ).
⃗ × (𝜔 (1.48)
⏟⏟⏟ ⏟⏞⏟⏞⏟ ⏟⏞⏞⏟⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏟
radial tangential Coriolis centripetal

Each of the terms in Equation 1.48 has a name and a physical meaning, as follows.

1. P⃗̈ P∕O is the radial acceleration. This is nothing more than the second derivative of the
distance between O and P and it is aligned with P⃗ P∕O .
2. 𝛼⃗ × P⃗ P∕O is the tangential acceleration. It is called the tangential acceleration because it
is aligned with the direction in which the point P would move if it were a fixed distance
from O and were rotating about O (i.e. in a direction perpendicular to a line passing
through O and P). Notice that 𝛼⃗ is the total derivative of the angular velocity including

@Seismicisolation
@Seismicisolation
 1.7 The General Acceleration Expression 15

its rate of change of magnitude and its rate of change of direction. As a result, 𝛼⃗ may not
be aligned with 𝜔. ⃗
3. 2𝜔⃗ × P⃗̇ P∕O is the Coriolis acceleration5 . The vectorial approach to finding the Coriolis
acceleration is in many ways preferable to the scalar approach put forward in many
books on dynamics. The magnitude of the radial velocity P⃗̇ P∕O is often referred to in
reference books as 𝑣rel and the Coriolis acceleration is seen written as 2𝜔𝑣rel where the
reader is left to determine its direction from a complicated set of rules. Consideration of
Equations 1.45–1.47 shows that there are two very different types of terms that combine
to form the Coriolis acceleration with its remarkable 2. The two terms are equal in mag-
nitude and direction (i.e. each is 𝜔 ⃗ × P⃗̇ P∕O ). One of these arises from part of the rate of
change of magnitude of the tangential velocity of P. The second arises from the rate of
change of direction of the radial velocity of P.
4. 𝜔⃗ × (𝜔⃗ × P⃗ P∕O ) is the centripetal acceleration. In 2D circular motion. this is commonly
written as 𝜔2 r and points toward the center of the circle. For the general points O and P
used here, the centripetal acceleration points from P to O.
It is possible to visualize the acceleration components using a simple graphical construc-
tion. As an example, we can use the slider in a slot system shown in Figure 1.4 for which we
have already derived both the velocity (Equation 1.19) and the acceleration (Equation 1.20)
in body fixed coordinates.
Remember that rates of change of magnitude are aligned with the vector that is changing
and rates of change of direction are perpendicular to the original vector and are pointed in
the direction that the tip of the vector would move if it had the prescribed angular velocity
and were simply rotating about its tail.
Figure 1.7 shows the two components of the velocity of the slider in the inner circle. A
component is labeled with a Δm to indicate that it results from a rate of change of magnitude
or a Δd to show that it results from a rate of change of direction. The two terms here are
the radial velocity ẋ aligned with the original position vector (a rate of change of magnitude
term) and the tangential velocity 𝜔x (a rate of change of direction term) perpendicular to
the position vector, pointing in the direction that the angular velocity would cause point A
to move if it were simply rotating about O, and with magnitude equal to the vector length
multiplied by the angular speed. These are the same two terms that appear in Equation 1.19.
Between the inner and outer circles on Figure 1.7 are the components of the acceleration.
To get these terms, we treat the velocity components as separate vectors that can change in
magnitude and direction and apply to each of them the same procedure we used on the
position vector in the preceding paragraph.
̇ Its rate of change of magnitude will be aligned with
Consider first the radial velocity (x).
it and will be equal to the rate of change of its length (i.e. the derivative of ẋ with respect to
time or ẍ ). As we watch it rotate about its tail with angular speed 𝜔, the tip of this vector will
move at the rate 𝜔ẋ in the direction ⃗𝚥 shown on the figure. This rate of change of direction
of the radial velocity is one half of the Coriolis acceleration.

5 Gaspard Gustave de Coriolis (1792–1843), an engineer and mathematician, introduced the terms “work”
and “kinetic energy” to engineering analysis but is best remembered for showing that the laws of motion
could be used in a rotating reference frame if an extra term called the Coriolis acceleration is added to the
equations of motion.

@Seismicisolation
@Seismicisolation
16 1 Kinematics

..
x
.
ωx
Δd Δm
. .
(ωx + 𝜔x)
ωx
Δm Δm
Δd A
velocity

acceleration
Δd ⇀
⇀
j i
ω 2x
⇀
k

Figure 1.7 The velocity and acceleration components of the slider.

Next consider the tangential velocity (𝜔x). Its rate of change of magnitude is aligned with
it and consists of the time derivative of 𝜔x which has two terms by the chain rule of dif-
ferentiation (i.e. 𝜔x ̇ Notice that the second of these terms is the other half of the
̇ and 𝜔x).
Coriolis acceleration. The rate of change of direction of the tangential velocity is found by
rotating it about its tail with angular speed 𝜔 and finding that it has magnitude 𝜔(𝜔x) or
𝜔2 x and points from A to O. The rate of change of direction of the tangential velocity is, in
fact, the centripetal acceleration.
The vector sum of these components yields the same acceleration as Equation 1.20.

Exercises
Descriptions of the systems referred to in the exercises are contained in Appendix A.

1.1 Show that the absolute velocity of point B of system 1 is,

𝑣⃗B = [−d2 (𝜃̇ 1 + 𝜃̇ 2 ) sin 𝜃2 ]⃗𝚤 + [d1 𝜃̇ 1 + d2 (𝜃̇ 1 + 𝜃̇ 2 ) cos 𝜃2 ]⃗𝚥.

1.2 Show that the absolute acceleration of point B of system 1 is,

a⃗ B = [−d1 𝜃̇ 12 − d2 (𝜃̈1 + 𝜃̈2 ) sin 𝜃2 − d2 (𝜃̇ 1 + 𝜃̇ 2 )2 cos 𝜃2 ]⃗𝚤
+ [d1 𝜃̈1 + d2 (𝜃̈1 + 𝜃̈2 ) cos 𝜃2 − d2 (𝜃̇ 1 + 𝜃̇ 2 )2 sin 𝜃2 ]⃗𝚥. (1.49)

1.3 Let the coordinate system shown in system 2 be fixed in AB and show that the accel-
eration of C with respect to B is,
a⃗ C∕B = [−d(𝜃̈ − 𝜙)
̈ cos 𝜙 − d(𝜃̇ − 𝜙)̇ 2 sin 𝜙]⃗𝚤
+ [d(𝜃̈ − 𝜙)
̈ sin 𝜙 − d(𝜃̇ − 𝜙)
̇ 2 cos 𝜙]⃗𝚥. (1.50)

@Seismicisolation
@Seismicisolation
 Exercises 17

1.4 Let the coordinate system shown in system 2 be fixed in BC and repeat the previous
problem. Note that the angle 𝜙 is constant in this coordinate system so that the vector
p⃗ C∕B has no rate of change of magnitude.

1.5 Show that the “rolls without slipping condition” in system 5 requires that the drum
have an angular velocity 𝜔 = 𝜃̇ + x∕r
̇ in the counterclockwise direction.

1.6 Show that the absolute acceleration of the center of the drum in system 5 has a com-
ponent,

ẍ + r 𝜃̈ + (d − x)𝜃̇ 2

down the plane and a component perpendicular to the plane that can be written as,

−(d − x)𝜃̈ + r 𝜃̇ 2 + 2ẋ 𝜃.

̇

1.7 Define a right handed coordinate system in system 6 where ⃗𝚤 is aligned with OA and
k⃗ points upward and show that,

a⃗ m = [r 𝜃̈ sin 𝜃 + r 𝜃̇ 2 cos 𝜃 − 𝜔2 (d − r cos 𝜃)]⃗𝚤

+ [𝜔(d ̇ − r cos 𝜃) + 2𝜔r 𝜃̇ sin 𝜃] ⃗𝚥
⃗
+ [r 𝜃̈ cos 𝜃 − r 𝜃̇ 2 sin 𝜃]k.

1.8 Show that the “rolls without slipping condition” in system 14 requires that the disk
̇ + r)∕r in the clockwise direction.
have an angular velocity 𝜔 = 𝜃(R

1.9 Show that the acceleration of the center of mass of the uniform rod in system 16 has
a component,

r 𝜃̈ sin 𝜙 − r 𝜃̇ 2 cos 𝜙 − 𝓁(𝜃̇ + 𝜙)

̇ 2

aligned with the rod and a component perpendicular to the rod that can be written
as,

r 𝜃̈ cos 𝜙 + r 𝜃̇ 2 sin 𝜙 + 𝓁(𝜃̈ + 𝜙).

̈

1.10 Using the coordinate system of Exercise 1.7, show that the absolute angular acceler-
ation of the massless rigid rod in system 6 is,
⃗
𝛼⃗Am = −𝜔 𝜃̇ ⃗𝚤 + 𝜃̈ ⃗𝚥 + 𝜔̇ k.

1.11 Using a coordinate system fixed in the ground with ⃗𝚤 positive to the right and ⃗𝚥 posi-
tive up, show that the acceleration of the center of mass of the the rod OA in system
12 is

a⃗ G = (d𝜃̈ sin 𝜃 + d𝜃̇ 2 cos 𝜃)⃗𝚤 + (−d𝜃̈ cos 𝜃 + d𝜃̇ 2 sin 𝜃)⃗𝚥.

@Seismicisolation
@Seismicisolation
18 1 Kinematics

1.12 Show that the absolute acceleration of the center of mass of the rod in system 16 is,
a⃗ G = [−r 𝜃̇ 2 − 𝓁(𝜃̈ + 𝜙)
̈ sin 𝜙 − 𝓁(𝜃̇ + 𝜙)
̇ 2 cos 𝜙]⃗𝚤
+ [r 𝜃̈ + 𝓁(𝜃̈ + 𝜙)
̈ cos 𝜙 − 𝓁(𝜃̇ + 𝜙)
̇ 2 sin 𝜙]⃗𝚥

where ⃗𝚤 points from O to A.

1.13 Consider system 23. Using the rotating coordinate system shown, show that the
absolute velocity of the mass is,
⃗
𝑣⃗m = (𝓁̇ sin 𝜙 + 𝓁 𝜙̇ cos 𝜙)⃗𝚤 + (𝓁 𝜃̇ sin 𝜙)⃗𝚥 + (−𝓁̇ cos 𝜙 + 𝓁 𝜙̇ sin 𝜙)k.

1.14 Consider system 23 again. Define a ground fixed coordinate system (⃗𝚤0 ,⃗𝚥0 ,k⃗0 )
⃗ system through an angle 𝜃 in the negative
obtained by a plane rotation of the (⃗𝚤,⃗𝚥,k)
⃗k direction. That is, ⃗𝚤 points from B to A and k⃗ is aligned with k.
⃗ Show that the
0 0
position of the mass with respect to the point O in this system is,
p⃗ m∕O = (𝓁 sin 𝜙 cos 𝜃)⃗𝚤0 + (𝓁 sin 𝜙 sin 𝜃)⃗𝚥0 + (−𝓁 cos 𝜙)k⃗0 .
Differentiate this position vector to get the absolute velocity of the mass and show
that you could get the same result by transforming the result of Exercise 1.13 using
a plane rotation.

1.15 Finally, for system 23. Define a body fixed coordinate system (⃗𝚤1 ,⃗𝚥1 ,k⃗1 ) obtained by
⃗ system through an angle 𝜙 in the negative ⃗𝚥 direction.
a plane rotation of the (⃗𝚤,⃗𝚥,k)
That is, k⃗1 points from m to O and ⃗𝚥1 is aligned with ⃗𝚥. The position of the mass with
respect to the point O in this system is,
p⃗ m∕O = −𝓁 k⃗1 .
Differentiate this position vector to get the absolute velocity of the mass and show
that you could get the same result by transforming the result of Exercise 1.13 using a
plane rotation. Be sure to get the correct angular velocity for the coordinate system
before differentiating.

@Seismicisolation
@Seismicisolation
 19

Newton’s Equations of Motion

2.1 The Study of Motion

The study of motion has a long history. Much of the early work was motivated by a desire
to study the motions of celestial bodies and the early scientists who worked in this area
have contributed to the body of knowledge we rely upon when we analyze the motion of
mechanical systems.
Galileo Galilei (1564–1642) was the first person to describe the concept of inertia. He
stated that inertia was the property of matter that causes it to resist changing its state of
motion unless compelled to by applied forces. This idea is fundamental to the field of
dynamics.
Issac Newton (1642–1727) developed the three laws of motion that form the basis for our
study of dynamics. His first and third laws are basically a restatement of Galileo’s concept of
inertia and forces of interaction. His second law, however, formally introduces the fact that
the effect of a force on an object is to cause a rate of change of its momentum that equals the
force both in magnitude and direction. This is the fundamental building block upon which
all of the material presented here is based.

2.2 Newton’s Laws

Newton developed three laws of motion that are, for the most part, common sense. He did
however introduce the concept of “mass”, which is something that remains with us today,
although his original definition has been somewhat changed by the work of Einstein et al.
Nevertheless, the majority of problems in mechanical engineering can still be handled by
reference to Newton’s laws. They are,
1. Every body perseveres in its state of rest or of uniform motion in a straight line, except
in-so-far-as it is compelled to change that state by impressed forces.
2. The rate of change of momentum of a body is equal to the impressed force and takes
place in the direction of the straight line along which the force acts.
This is a very interesting statement. It relates two vector quantities (momentum and
force) and says that the momentum vector has a rate of change (derivative) that is

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
20 2 Newton’s Equations of Motion

equal to the applied (impressed) force. This is Newton’s famous second law that is often
stated as F = ma. If you performed experiments and plotted the applied force versus
the resulting acceleration you would see that the curves were straight lines and would
conclude that the acceleration was proportional to the applied force. The constant of
proportionality is the “mass” and, to this day, we are unable to measure mass directly.
We can measure force because of the way in which it causes materials to deflect but
mass can only be inferred from a force measurement . There are four basic quantities
related to the study of dynamics – force, mass, length, and time. Of these, we can only
measure three – force, length, and time. This fact becomes important when we choose
a system of units for analysis.
3. Reaction is always equal and opposite to action, that is to say, the actions of two bodies
upon each other are always equal and directly opposite.
Newton’s three laws, being simple and intuitive, are the foundation upon which most of
the field of dynamics is built.

2.3 Newton’s Second Law for a Particle

The second law can be stated for a single particle as,

⃗
dG
F⃗ = (2.1)
dt
where F⃗ is the total externally applied force acting on the particle. G ⃗ = m𝑣⃗ is the linear
momentum of the particle. m is the mass of the particle. 𝑣⃗ is the absolute velocity 1 .
Immediately upon considering applications of Equation 2.1, it becomes apparent that
an ability to work with derivatives of vectors is required. The second law makes specific
⃗
⃗ = m𝑣⃗
reference to the time rate of change of the linear momentum vector ddtG . Substituting G
into Equation 2.1 yields,
d dm d𝑣⃗
F⃗ = (m𝑣)
⃗ = 𝑣⃗ + m . (2.2)
dt dt dt
The first term in the expansion (i.e. dmdt
𝑣)
⃗ is non-zero only when systems with variable
mass are considered. Rockets, for instance, have a fairly large rate of change of mass as
fuel is consumed during takeoff and dm dt
must be considered. The majority of mechanical
systems on earth are composed of rigid or flexible bodies that do not suffer a mass change
during their motions. For this reason, Equation 2.2 is most often used in the form,
d𝑣⃗
F⃗ = m = m⃗a (2.3)
dt
where a⃗ is defined to be the acceleration of the particle.
d𝑣⃗
a⃗ = . (2.4)
dt

1 Absolute velocity is defined as the velocity of an object with respect to an inertial reference frame and
absolute acceleration is defined in the same way. Newton’s laws apply only to absolute velocities and
accelerations. See Chapter 1.

@Seismicisolation
@Seismicisolation
 2.4 Deriving Equations of Motion for Particles 21

Both the velocity and the acceleration of the particle relate its motion to an inertial refer-
ence frame and they are termed the absolute velocity and absolute acceleration respectively.

2.4 Deriving Equations of Motion for Particles

Newton’s laws provide a very convenient method for deriving the equations of motion of
simple systems. Equations of motion are the differential equations that, when solved, can
be used to predict the response of the system to a set of applied forces.
The procedure for deriving the equations has only four steps and, if they are followed, the
derivations are very straightforward. The steps are,

1. Kinematics – choose the coordinates (also known as degrees of freedom) to be used to

describe the motion and derive expressions for the absolute velocities and accelerations
of the masses under consideration. Use the methods of Chapter 1.
2. Free body diagrams (FBDs) – sketch the masses under consideration as if they are in
space with no forces acting on them. Then add to the sketches all of the externally
applied forces acting on the masses. Also add the internal forces of interaction between
the masses, being sure that they act in equal and opposite pairs as stipulated by Newton’s
third law. The FBDs should show the positive sense of the accelerations derived in step 1.
3. Force balance equations (Newton’s second law) – using the FBDs, equate the vector sum
of forces on each body to its mass multiplied by its vector acceleration.
4. Manipulate and solve the equations – the first three steps will lead to a set of equations
with a set of unknowns. All that is left is to manipulate and combine the equations in
order to extract the desired solution.

As an example, we can apply this procedure to the slider in the slot first shown in
Figure 1.4 and repeated here for convenience as Figure 2.1. The step by step procedure
goes as follows.

Figure 2.1 A slider in a slot. ⇀

⇀ i
j

⇀
k

x A

@Seismicisolation
@Seismicisolation
22 2 Newton’s Equations of Motion

⇀ ⇀ ⇀
⇀ i ⇀ i ⇀ i
j j j

⇀ ⇀ ⇀
k k k

Ff

A
A A N
x

O O O

Original System Slider Removed from System FBD of the Slider

Figure 2.2 Creating the Free Body Diagram of the slider.

1. Kinematics – the absolute velocity and acceleration of the slider have been previously
derived in Equations 1.21 and 1.22 and shown to be,

𝑣⃗A = ẋ ⃗i + 𝜔x⃗j

and,

a⃗ A = (̈x − 𝜔2 x)⃗i + (𝜔x ̇ ⃗j.

̇ + 2𝜔x)
⃗ so that the force balances
Notice that these expressions use the body fixed axes (⃗i, ⃗j, k)
must use the same directions. This is, in fact, a very convenient set of directions for this
system since the forces acting on the slider are either aligned with or perpendicular to
the slot.
2. Free body diagrams (FBDs) – Figure 2.2 shows the FBD of the slider being developed.
At first, the slider is simply shown in space, detached from all other bodies and without
having any forces acting on it. The forces acting on the slider are then shown along with
the positive directions of the accelerations from step 1 which are, in fact, the directions
of the unit vectors . Consideration must be given at this stage to the forces that are acting
on the slider.
The figure shows two forces.
The first is a normal force N that is acting on the slider due to its interaction with the wall
of the slot in the rotating body. The magnitude of this force is not known and its direction
is only known to the extent that it must be perpendicular to the line of contact between
the slider and the slot. In the figure it is assumed that the force acts in the positive ⃗j
direction on the slider (and, by Newton’s third law, in the negative ⃗j direction on the wall
of the slot) although this is not a certainty. The final solution may reveal that N has a
negative value and is actually acting in the opposite direction.

@Seismicisolation
@Seismicisolation
 2.4 Deriving Equations of Motion for Particles 23

The second, Ff , is a frictional force that opposes the motion of the slider along the length
of the slot. Frictional forces always oppose the relative motion between two bodies so
the force shown in the figure is in a direction that assumes that the slider is moving out-
ward (i.e. the assumption is that ẋ is positive). It is not necessary to make this particular
assumption but some assumption concerning the direction of action of the force must
be made and, once made, must be carried through to the final solution.
We can make the further assumption that the frictional force is due to “dry” or
“Coulomb”2 friction. In its simplest form, this means that the force due to friction
can be modeled as being limited to a constant coefficient of friction (usually called 𝜇)
multiplied by the normal force between the two bodies in contact. In other words, if the
two bodies are assumed not to be moving relative to each other and the force required
to prevent them from moving is less than 𝜇N, then they will not move and the force
will be equal to the required value. If, on the other hand, the two bodies are assumed
not to be moving relative to each other and the force required to prevent the motion is
greater than 𝜇N, then there will be motion and the frictional force will be equal to its
maximum value (i.e. Ff = 𝜇N) and will be in a direction opposing the relative motion.
3. Force balance equations (Newton’s second law) – for this simple system of constant mass
(assume the mass of the slider is m), the force balance equations are derived simply by
summing the applied forces in the ⃗i and ⃗j directions respectively, and equating the total
force in each direction to the mass multiplied by the acceleration (derived in step 1) in
that direction. The resulting equations are,
∑
F⃗i ∶ m(̈x − 𝜔2 x) = −Ff (2.5)
∑
F⃗j ∶ m(𝜔x ̇ =N
̇ + 2𝜔x) (2.6)

where it should be noted that the positive directions assumed for the forces when draw-
ing the FBD have been maintained.
4. Manipulate and solve the equations – given the simplicity of this problem, it is quite
remarkable how much information we can get from the equations of motion. There are
three cases to consider.
(a) Consider first the case where the slider has been moving in the slot but has now come
to rest so that x is a constant. Then, ẋ = ẍ = 0 and Equations 2.5 and 2.6 simplify to,

Ff = m𝜔2 x (2.7)

N = m𝜔x.
̇ (2.8)

The action of specifying that x be constant has introduced a constraint on the motion
that can only exist if the forces Ff and N are exactly as specified in Equations 2.7 and
2.8. The forces are then known as constraint forces.

2 Charles-Augustin de Coulomb (1736–1806), a military engineer, won the Grand Prix from the Académie
des Sciences of France in 1781 for his work on friction. He also performed groundbreaking work on
electricity and magnetism, where he developed the laws of attraction and repulsion, the theory of electric
point charges, and the distribution of electricity on the surface of charged bodies.

@Seismicisolation
@Seismicisolation
24 2 Newton’s Equations of Motion

In fact, the normal force N, in this or any other system, is always a constraint force
because it takes on any value required to stop the two bodies from intruding on each
other.
Equation 2.8 indicates that the normal force can take on negative values if 𝜔̇ < 0.
While it is not physically possible for the normal force to hold the two bodies together
(only apart), the normal force can become negative by having the slider press against
the wall opposite to that assumed in the FBD (Figure 2.2).
Another consideration at this point is whether or not there is sufficient frictional
force available for x to remain constant. That is, the maximum available frictional
force is Ffmax = ±𝜇N, which, after substituting N from Equation 2.8, is given by
Ffmax = ±𝜇m𝜔x. ̇ If the angular acceleration 𝜔̇ should go to zero, for instance, there
is no available friction force (i.e. the slider is touching neither of the two walls of the
slot) and therefore there is nothing to maintain a constant value of x. If the frictional
force required to maintain the constraint is larger in magnitude than the available
frictional force, the slider will begin to accelerate along the slot and Equations 2.7
and 2.8 are no longer valid. In that case we must revert to Equations 2.5 and 2.6,
which are the governing equations of motion for this system and are always valid.
(b) Consider now the case where the slider has developed a motion outward within the
slot. That is, ẋ > 0 and the frictional force naturally opposes the motion (i.e. acts
inward as was assumed when the FBD in Figure 2.2 was constructed). This means
that the frictional force is positive and has magnitude,
Ff = 𝜇|N|.
Then, substituting for N from Equation 2.6, we find that,
Ff = 𝜇|m(𝜔x ̇
̇ + 2𝜔x).|
This can be substituted into Equation 2.5 to yield the following differential equation
governing motion of the slider outward in the slot.
ẍ − 𝜔2 x + 𝜇|(𝜔x ̇ = 0.
̇ + 2𝜔x)|
Except for the absolute value in the equation, it is not a particularly difficult equation
to solve. Consider, for example, the case where 𝜔 has a positive, known, constant
value (say 𝜔 = Ω and 𝜔̇ = 0). The differential equation simplifies to,
ẍ + 2𝜇Ωẋ − Ω2 x = 0.
This is a linear, second order, differential equation with constant coefficients. If we
make a term-by-term comparison with the standard equation for a damped oscilla-
tor, written in the form,
m̈x + cẋ + kx = 0
we can see that the slider has a negative stiffness term that will cause x to grow
without bound.
(c) For the final case where the slider has a motion directing it inward within the slot
(ẋ < 0), the frictional force again opposes the motion (i.e. acts outward) and therefore
has a negative magnitude in the sense of the force assumed on the FBD. That is,
Ff = −𝜇|N|.

@Seismicisolation
@Seismicisolation
 2.5 Working with Rigid Bodies 25

Following a procedure similar to that in the previous case gives the result,
ẍ − 𝜔2 x − 𝜇|(𝜔x ̇ = 0.
̇ + 2𝜔x)|
Considering again the case where 𝜔 has a positive, known, constant value (say 𝜔 = Ω
and 𝜔̇ = 0). The differential equation simplifies to,
ẍ − 2𝜇Ωẋ − Ω2 x = 0
and the solution again grows without bound, having now both negative stiffness and
negative damping.
Part 2 of this text discusses methods of solving the equations of motion.

2.5 Working with Rigid Bodies

To this point, we have considered only the translational motion of particles. We now move
on to consider bodies of finite size that comprise groups of particles. These particles are
subject to internal constraint forces that cause them to maintain a prescribed geometrical
relationship to each other. The simplest constraint causes them to remain a fixed distance
from each other and the group of particles becomes a rigid body. Another type of constraint
will allow the particles to move relative to each other so long as they obey a specified con-
stitutive law. These groups of particles are flexible bodies.
Rigid body analysis must take into account the fact that particles within the body have
different accelerations since, in the general case, rigid bodies will translate and rotate simul-
taneously. It is the rotational motion that causes the individual particles to experience dif-
ferent velocities and accelerations.
The concept of angular momentum is introduced in Section 2.8 to account for rotational
motion. Angular momentum does not arise from a “fourth law of Newton” but is, in fact,
derived from what we already know about linear momentum and the relationship of its rate
of change to applied forces.
We start the derivation of rigid body equations of motion with a single particle. Figure 2.3
shows a three dimensional rigid body. The body has a general reference point O with known

Figure 2.3 A single particle in a rigid body. ⇀

j

⇀
⇀ ω
i
⇀
k
⇀
mi α

⇀
pi/O

a⇀O
⇀
vO

@Seismicisolation
@Seismicisolation
26 2 Newton’s Equations of Motion

absolute velocity 𝑣⃗O and acceleration a⃗ O . Vectors are expressed in a body-fixed reference
⃗ The body and reference frame have angular velocity 𝜔
frame having unit vectors (⃗i, ⃗j, k). ⃗
and angular acceleration 𝛼⃗ . The particle to be considered (particle i) is located with respect
to the reference point by the position vector p⃗ i∕O and has mass mi .

2.6 Using F⃗ = ma⃗ in the Rigid Body Force Balance

The force balance for the single particle in the rigid body is relatively straightforward. The
absolute acceleration of the particle is required and is found to be,

a⃗ i = a⃗ O + a⃗ i∕O (2.9)

which, using Equation 1.48 with the understanding that p⃗ i∕O cannot change in length
because both points O and i are in the same rigid body3 , becomes,

a⃗ i = a⃗ O + 𝛼⃗ × p⃗ i∕O + 𝜔 ⃗ × p⃗ i∕O ).
⃗ × (𝜔 (2.10)

The FBD of the particle is presented in Figure 2.4 where two forces are shown being
applied to the particle. The first of these, F⃗ i , represents the vector sum of all externally
applied forces acting on the particle. This is a force arising from the interaction of the rigid
body with its external environment. The second applied force, f⃗i , is the vector sum of inter-
nal forces of constraint acting on particle i. This is the force of interaction between particle
i and its neighboring particles that enforces the rigid body constraint. Forces of this type
must occur in equal and opposite pairs on neighboring particles in accord with Newton’s
third law.
According to Newton’s second law, and assuming that mass does not change with time,
the vector sum of the two forces acting on the particle must be equal to its mass multiplied
by its acceleration, as follows,

f⃗i + F⃗ i = mi a⃗ i (2.11)

where a⃗ i is known from Equation 2.10 and can be substituted into Equation 2.11 to yield,

f⃗i + F⃗ i = mi [⃗aO + 𝛼⃗ × p⃗ i∕O + 𝜔 ⃗ × p⃗ i∕O )]

⃗ × (𝜔 (2.12)

or,

f⃗i + F⃗ i = mi a⃗ O + mi (⃗
𝛼 × p⃗ i∕O ) + mi [𝜔 ⃗ × p⃗ i∕O )].
⃗ × (𝜔 (2.13)

Consider the case where Equation 2.13 has been written for every particle in the rigid
body. We can then add together all such equations to get,
∑ ∑ ∑ ∑ ∑
f⃗i + F⃗ i = mi a⃗ O + 𝛼 × p⃗ i∕O ) +
mi (⃗ mi [𝜔 ⃗ × p⃗ i∕O )]
⃗ × (𝜔 (2.14)
i i i i i

3 The rigid body constraint dictates that any two points in the same rigid body must maintain a constant
separation distance. This is, in fact, the definition of a rigid body. Points in elastic bodies are not subject to
this constraint. In reality every body has some degree of elasticity but the assumption of rigid bodies can
often be made without loss of accuracy.

@Seismicisolation
@Seismicisolation
 2.6 Using F⃗ = m⃗a in the Rigid Body Force Balance 27

Figure 2.4 Free body diagram of a single particle

in a rigid body. mi
⇀ ⇀
fi Fi
⇀
pi/O

∑
where indicates that the summation is over all particles i in the body. Noting that mi is
i
a scalar, it is possible to move it around within Equation 2.14 without affecting the result4 .
Terms that do not depend on the index of summation i may also be moved outside the
summation process. As a result, Equation 2.14 can be rewritten as,
( ) ( ) [ ( )]
∑ ∑ ∑ ∑ ∑
f⃗ +
i F⃗ =
i m a⃗ + 𝛼⃗ ×
i O m p⃗ +𝜔 ⃗× 𝜔
i i∕O ⃗× m p⃗ . i i∕O
i i i i i
(2.15)
There are several terms of interest in Equation 2.15. These are,
∑
1. f⃗i – this is the sum, over all the particles in the rigid body of the internal forces of
i
interaction. By Newton’s third law, all of these forces appear in equal and opposite pairs
and the vector sum of them over the body must be zero since each pair adds to zero. The
result is,
∑
f⃗i = 0.
i
∑
2. F⃗ i – this is the vector sum of all externally applied forces acting on the body. It includes
i
forces that act on all particles (e.g. gravity forces or magnetic forces) and those that act
only on one or a few particles (e.g. a force applied at a point). We write the total externally
applied force as F⃗ and make the substitution,
∑
F⃗ i = F⃗ .
i
∑
3. mi – this is clearly the sum of the masses of all particles making up the rigid body
i
and, as a result, is the total mass of the body. Let the total mass of the body be called M,
so that,
∑
mi = M.
i
∑
4. mi p⃗ i∕O – this term is a little more difficult to explain than the others and appears twice.
i
Each component of the summation is the product of a mass (i.e. mi ) and a position (both
distance and direction) relative to the reference point O (i.e. p⃗ i∕O ). The sum of all such
terms will therefore be the product of a mass and a position vector.

4 It is often not possible to change the order of vector operations without affecting the result so care should
be taken if tempted to do so.

@Seismicisolation
@Seismicisolation
28 2 Newton’s Equations of Motion

In fact, the sum of all of the individual terms of this type will be equal to the total mass
of the body multiplied by the position of the center of mass of the body relative to the
reference point O5 . The center of mass is commonly designated as the point G and the
position of G relative to O can be written as p⃗ G∕O , yielding,
∑
mi p⃗ i∕O = M⃗pG∕O .
i

Substituting these definitions into Equation 2.15 and moving the scalar mass M outside
the vector operations simplifies it to,
F⃗ = M⃗aO + M(⃗
𝛼 × p⃗ G∕O ) + M[𝜔 ⃗ × p⃗ G∕O )]
⃗ × (𝜔 (2.16)
or, even more simply,
F⃗ = M[⃗aO + 𝛼⃗ × p⃗ G∕O + 𝜔 ⃗ × p⃗ G∕O )]
⃗ × (𝜔 (2.17)
where the expression,
a⃗ O + 𝛼⃗ × p⃗ G∕O + 𝜔 ⃗ × p⃗ G∕O )
⃗ × (𝜔
is easily recognized as the absolute acceleration of the center of mass G. That is, applying
Equation 2.10 to find the acceleration of G rather than i would give,
a⃗ G = a⃗ O + 𝛼⃗ × p⃗ G∕O + 𝜔 ⃗ × p⃗ G∕O )
⃗ × (𝜔 (2.18)
which is exactly the same result.
The force balance equation for the rigid body is therefore completely analogous to that
used for a particle. That is,
F⃗ = M⃗aG .

⃗
2.7 Using F⃗ = dG
dt
in the Rigid Body Force Balance

Section 2.6 presented a derivation of the force balance relationship for a rigid body using
Newton’s second law in the form F⃗ = m⃗a. Here, we treat the case again using the more
⃗
general form of the second law, F⃗ = ddtG .
Using Equation 2.1, the force balance equation for the particle i shown in Figure 2.4 can
be written as,
dG⃗i
f⃗i + F⃗ i = . (2.19)
dt
Equation 2.19 can be written for every particle in the body and all of the resulting
equations can be added together to give,
( )
∑ ∑ ∑ dG⃗i d ∑
f⃗i + F⃗ i = = ⃗i .
G (2.20)
i i i
dt dt i

5 Readers are encouraged to review the process used in determining the location of the center of mass of a
body. Most reference books will discuss this process in one dimension only but three successive
applications of their procedures in three orthogonal directions can readily be visualized as the vector
approach used here.

@Seismicisolation
@Seismicisolation
 dG⃗
2.7 Using F⃗ = dt
in the Rigid Body Force Balance 29

The linear momentum of particle i in Figure 2.3 is simply the mass mi multiplied by the
absolute velocity 𝑣⃗i , where,

⃗ × p⃗ i∕O .
𝑣⃗i = 𝑣⃗O + 𝜔

That is,
⃗ i = mi (𝑣⃗O + 𝜔
G ⃗ × (mi p⃗ i∕O )
⃗ × p⃗ i∕O ) = (mi )𝑣⃗O + 𝜔 (2.21)

and, the total linear momentum of the body is given by,

( ) ( )
∑ ∑ ∑
⃗
G= ⃗
Gi = mi 𝑣⃗O + 𝜔
⃗× mi p⃗ i∕O . (2.22)
i i i
∑
As explained in Section 2.6, the expression mi is equal to the total mass of the body
∑ i
M and the expression mi p⃗ i∕O is equal to the total mass of the body multiplied by the
i
position of the center of mass relative to the reference point O (i.e. M⃗pG∕O ).
As a result, the total linear momentum of the body can be written as,
⃗ = M(𝑣⃗O + 𝜔
G ⃗ × p⃗ G∕O ). (2.23)

Clearly, 𝑣⃗O + 𝜔⃗ × p⃗ G∕O is the expression for the velocity of the center of mass of the body
(i.e. 𝑣⃗G ) and the total linear momentum is, in fact,
⃗ = M 𝑣⃗G .
G (2.24)

Then, combining Equations 2.22 and 2.24, shows that,

∑
⃗ i = M 𝑣⃗G .
G (2.25)
i

This can be substituted into Equation 2.20 to give,

∑ ∑ d ⃗
dG
f⃗i + F⃗ i = (M 𝑣⃗G ) = . (2.26)
i i
dt dt

As in Section 2.6,
∑
f⃗i = 0 (2.27)
i

and
∑
F⃗ i = F (2.28)
i

where F is the total applied force. As a result, the force balance equation for the body can
be written as,
⃗
dG
F⃗ = . (2.29)
dt
A force balance for a rigid body therefore can be handled in one of two ways, both of
which require that the location of the center of mass, G, be known,

@Seismicisolation
@Seismicisolation
30 2 Newton’s Equations of Motion

1. Using the absolute acceleration of the center of mass,

F⃗ = M⃗aG .
2. Using the linear momentum of the rigid body,
⃗
dG d
F⃗ = = (M 𝑣⃗G ).
dt dt

2.8 Moment Balance for a Rigid Body

Performing a moment balance on a rigid body requires an understanding of the process of
“taking moments” about a point. We will use the reference point O defined in Section 2.6
and, particularly, in Figure 2.3.
If a force acts on a point in the body, then the moment or torque produced by that force
around point O is defined to be,
⃗ O = p⃗ F∕O × F⃗
M (2.30)
⃗ O is the moment about O, F⃗ is the applied force, and p⃗ F∕O is the position vector
where M
from O to the point of application of the applied force6 .
Consider the total moment about point O resulting from the two forces applied on the
particle i. Remember that the forces are an externally applied force, F⃗ i , and an internal
constraint force, ⃗fi . The moment about O due to these forces is therefore,

⃗ O = p⃗ i∕O × ⃗fi + p⃗ i∕O × F⃗ i = p⃗ i∕O × (f⃗i + F⃗ i ).

M (2.31)

Using Equation 2.19 with Equation 2.31, the following equality can be written.

dG⃗i
p⃗ i∕O × (f⃗i + F⃗ i ) = p⃗ i∕O × . (2.32)
dt
As was done for the force balance, we consider the case where Equation 2.32 has been
written for every particle in the body and the resulting equations have all been added
together to give,
∑ ∑ ∑ dG⃗i
p⃗ i∕O × ⃗fi + p⃗ i∕O × F⃗ i = p⃗ i∕O × . (2.33)
i i i
dt

Two of the terms in Equation 2.33 can be dealt with immediately. These are,
∑
1. p⃗ i∕O × f⃗i : this is the total moment about the reference point O caused by internal
i
forces. Since these forces occur in equal and opposite
∑ pairs, the moment caused by any
given force is canceled out by its counterpart and, p⃗ i∕O × f⃗i = 0.
i

6 Most engineers are more familiar with 2D moments where you simply multiply a force by a distance and
visualize the direction of the moment. This vector approach includes both the magnitude and the direction
of the moment and is indispensable in 3D problems.

@Seismicisolation
@Seismicisolation
 2.8 Moment Balance for a Rigid Body 31

∑
2. p⃗ i∕O × F⃗ i : this is the total moment about the reference point O caused by externally
i
⃗ O.
applied forces. Let this be denoted by M

Equation 2.33 can then be rewritten as,

∑ dG⃗i
⃗O =
M p⃗ i∕O × . (2.34)
i
dt

∑
dG⃗i
The third term (i.e. p⃗ i∕O ×
) is more difficult to deal with.
i
dt
We start by defining the angular momentum of the rigid body about point O. In the same
way that we defined moments caused by forces as a vector cross product, we define the angu-
lar momentum about point O due to the particle i (i.e. H ⃗ i∕O ) to be the moment of the linear
momentum of i about O. That is,

H ⃗ i.
⃗ i∕O = p⃗ i∕O × G (2.35)

It is then possible to express the total angular momentum of the rigid body about the
reference point as,
∑
H⃗O = ⃗ i.
p⃗ i∕O × G (2.36)
i

⃗ O with respect to time and find the third term of Equation 2.33,
We differentiate H
⃗ O ∑ d⃗pi∕O
dH ∑ dG⃗i
= ⃗i +
×G p⃗ i∕O × . (2.37)
dt i
dt i
dt
⏟⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏟
the third term!

We reorder this to write the third term as,

∑ dG⃗i ⃗ O ∑ d⃗pi∕O
dH
p⃗ i∕O × = − ⃗ i.
×G (2.38)
i
dt dt i
dt

⃗ i gives,
Substituting mi 𝑣⃗i for G
∑ dG⃗i ⃗ O ∑ d⃗pi∕O
dH
p⃗ i∕O × = − × (mi 𝑣⃗i ). (2.39)
i
dt dt i
dt
At this point, we can substitute Equation 2.39 into Equation 2.34 to get,
⃗ O ∑ d⃗pi∕O
dH
⃗O =
M − × (mi 𝑣⃗i ). (2.40)
dt i
dt

d⃗pi∕O d⃗pi∕O
Notice that is simply an expression for the velocity of i with respect to O (i.e. =
dt dt
𝑣⃗i∕O ) and we can write,
⃗O ∑
dH ⃗O ∑
dH
⃗O =
M − 𝑣⃗i∕O × (mi 𝑣⃗i ) = − mi (𝑣⃗i∕O × 𝑣⃗i ). (2.41)
dt i
dt i

@Seismicisolation
@Seismicisolation
32 2 Newton’s Equations of Motion

The absolute velocity of i, 𝑣⃗i , can be written as 𝑣⃗i = 𝑣⃗O + 𝑣⃗i∕O . Substituting this expression
in Equation 2.41 and expanding yields,
⃗O ∑
dH ∑
⃗O =
M − (mi 𝑣⃗i∕O ) × 𝑣⃗O − mi (𝑣⃗i∕O × 𝑣⃗i∕O ). (2.42)
dt i i

Since 𝑣⃗i∕O × 𝑣⃗i∕O = 0 by definition, and

( )
∑ ∑ d⃗pi∕O d ∑
mi 𝑣⃗i∕O = mi = mi p⃗ i∕O .
i i
dt dt i
Equation 2.42 can be rewritten as,
[ ( )]
⃗O
dH d ∑
⃗O =
M − mi p⃗ i∕O × 𝑣⃗O . (2.43)
dt dt i

The term including the mass of particle i was previously discussed in Section 2.6 (see the
material following Equation 2.15) and found to be equal to the total mass of the body M
multiplied by the position of the center of mass with respect to the reference point O, p⃗ G∕O .
Making this substitution, and noting that the mass of a rigid body is constant, gives,
⃗O [ d
dH ] dH⃗O
M⃗O = − (M⃗pG∕O ) × 𝑣⃗O = − M(𝑣⃗G∕O × 𝑣⃗O ). (2.44)
dt dt dt
We now note that the velocity of G with respect to O can be rewritten as 𝑣⃗G∕O = 𝑣⃗G − 𝑣⃗O
simply by rearranging the relative velocity expression 𝑣⃗G = 𝑣⃗O + 𝑣⃗G∕O . Substituting this into
Equation 2.44 and noting that 𝑣⃗O × 𝑣⃗O = 0 gives,
dH⃗O
⃗O =
M − M(𝑣⃗G × 𝑣⃗O ). (2.45)
dt
Equation 2.45 is the general form of the moment balance for a rigid body where the
moments are taken about an arbitrary, moving reference point O. There are two special
reference points that are more often used for the moment balance. These are,
1. The case where the reference point is the center of mass. That is, O and G are coinci-
dent. In this case, the term M(𝑣⃗G × 𝑣⃗O ) is equal to zero because 𝑣⃗O = 𝑣⃗G and the moment
balance equation becomes,
dH⃗G
⃗G =
M .
dt
2. The case where the reference point has zero velocity. That is, point O is a “fixed point” by
virtue of being connected to ground through a joint. In this case, 𝑣⃗O = 0 and the moment
balance is,
dH⃗O
M⃗O = .
dt

A moment balance for a rigid body therefore can be handled in three possible ways,
1. Using a moving reference point O on the rigid body,
⃗O
dH
⃗O =
M − M(𝑣⃗G × 𝑣⃗O ).
dt

@Seismicisolation
@Seismicisolation
 ⃗
2.9 The Angular Momentum Vector – H 33
O

2. Using the center of mass of the rigid body as the reference point,
dH⃗G
⃗G =
M .
dt
3. Using fixed point O on the rigid body as the reference point,
⃗O
dH
⃗O =
M .
dt

2.9 ⃗O
The Angular Momentum Vector – H

Before being able to put the moment balance equations from Section 2.8 to use, we must
first consider generating the angular momentum vector for the rigid body.
The angular momentum vector, H, ⃗ cannot be defined without reference to some point on
7
the rigid body in question . This is because angular momentum is defined as the moment of
linear momentum. Equation 2.35 defines angular momentum about the reference point O
due to the mass particle i as,
H ⃗i
⃗ i∕O = p⃗ i∕O × G (2.46)
⃗ i = mi 𝑣⃗i , resulting in,
where the linear momentum is defined to be G
⃗ i∕O = mi (⃗pi∕O × 𝑣⃗i ).
H (2.47)
We assume that the absolute velocity of the reference point, 𝑣⃗O , and the angular velocity
of the rigid body, 𝜔,
⃗ are known. We can then write an expression for the velocity of particle
i (noting that the position vector p⃗ i∕O has no rate of change of magnitude because both i and
O are in the same rigid body and cannot move apart) as,
𝑣⃗i = 𝑣⃗O + 𝑣⃗i∕O
d
= 𝑣⃗O + (⃗pi∕O )
dt
= 𝑣⃗O + 𝜔⃗ × p⃗ i∕O . (2.48)
Referring to Figure 2.3, we can write expressions for the position of i with respect to O
⃗
and the absolute velocity of point O in the reference frame with unit vectors (⃗i, ⃗j, k).
⃗
Let the distance in the i direction from point O to point i be xi . Similarly, we define yi and
zi to be the distances in the ⃗j and k⃗ directions respectively. The position vector is then,
⃗
p⃗ i∕O = xi⃗i + yi ⃗j + zi k.
Let the absolute velocity of O have scalar components 𝑣Ox , 𝑣Oy , and 𝑣Oz . The velocity of O
is then,
⃗
𝑣⃗O = 𝑣Ox⃗i + 𝑣Oy ⃗j + 𝑣Oz k.

7 There are instances during the analysis of dynamic systems where it may be advantageous to define the
angular momentum about a point that is not on the body being analyzed. Often this is done for bodies
which can be approximated as particles. For rigid bodies, the reference point is always located on the body.

@Seismicisolation
@Seismicisolation
34 2 Newton’s Equations of Motion

⃗ be,
Further, let the angular velocity of the coordinate system (⃗i, ⃗j, k)
⃗ = 𝜔x⃗i + 𝜔y ⃗j + 𝜔z k⃗
𝜔
⃗ × p⃗ i∕O can then be written as,
The cross product 𝜔

⃗ × p⃗ i∕O = [𝜔y zi − 𝜔z yi ]⃗i + [𝜔z xi − 𝜔x zi ] ⃗j + [𝜔x yi − 𝜔y xi ]k⃗

𝜔
and the velocity of particle i is,
⃗
𝑣⃗i = [𝑣Ox + 𝜔y zi − 𝜔z yi ]⃗i + [𝑣Oy + 𝜔z xi − 𝜔x zi ]⃗j + [𝑣Oz + 𝜔x yi − 𝜔y xi ]k. (2.49)
Equation 2.49 can be substituted into Equation 2.47 to give, after performing another
cross product and gathering some terms, an expression for the angular momentum about
O due to particle i as follows:
⃗ i∕O = mi [yi 𝑣Oz − zi 𝑣Oy + (y2 + z2 )𝜔x − xi yi 𝜔y − xi zi 𝜔z ]⃗i
H i i

+ mi [zi 𝑣Ox − xi 𝑣Oz − yi xi 𝜔x + (xi2 + zi2 )𝜔y − yi zi 𝜔z ]⃗j

⃗
+ mi [xi 𝑣Oy − yi 𝑣Ox − zi xi 𝜔x − zi yi 𝜔y + (xi2 + y2i )𝜔z ]k. (2.50)
To get the total angular momentum vector about point O, we write Equation 2.50 for every
particle and add the resulting equations together to get,
∑
H⃗O = H⃗ i∕O (2.51)
i
or,
⃗ O = [My𝑣Oz − Mz𝑣Oy + Ixx 𝜔x − Ixy 𝜔y − Ixz 𝜔z ]⃗i
H O O O

+ [Mz𝑣Ox − Mx𝑣Oz − IyxO 𝜔x + IyyO 𝜔y − IyzO 𝜔z ]⃗j

+ [Mx𝑣Oy − My𝑣Ox − IzxO 𝜔x − IzyO 𝜔y + IzzO 𝜔z ]k⃗
(2.52)
where the terms used in Equation 2.52 are defined as follows.
∑
1. M = mi . This is the total mass of the rigid body.
∑
i
2. Mx = (mi xi ). This summation is one component of that used to locate the center of
i
mass of the body with respect to the reference point. x is the x-component of p⃗ G∕O .
∑
3. My = (mi yi ). y is the y-component of p⃗ G∕O .
∑
i
4. Mz = (mi zi ). z is the z-component of p⃗ G∕O .
∑
i
5. IxxO = mi (y2i + zi2 ). This term is a function of the spatial distribution of mass particles
i
around the x-axis (i.e. ⃗i direction) passing through point O. The term is always positive
because of the sum of squares term. It is called the x moment of inertia of the body about
point O.
∑
6. IxyO = mi xi yi . This term is also a function of the distribution of the mass in the body
i
but it has the potential to be negative or positive depending upon the signs of xi and yi .
It is called the x–y product of inertia about point O.

@Seismicisolation
@Seismicisolation
 ⃗
2.9 The Angular Momentum Vector – H 35
O

∑
7. IxzO = mi xi zi . This is the x–z product of inertia about point O.
∑
i
8. IyxO = mi yi xi . This is the y–x product of inertia about point O. Note that IyxO = IxyO
i
since the
∑ order of the x and y terms in the summations do not change the result.
9. IyyO = mi (xi2 + zi2 ). This is the y moment of inertia of the body about point O.
∑i
10. IyzO = mi yi zi . The y–z product of inertia about O.
∑i
11. IzxO = mi zi xi . The z–x product of inertia about O. Note that this is equal to the x–z
i
product∑of inertia.
12. IzyO = mi zi yi . The z–y product of inertia about O, which is equal to the y–z product
i
of inertia
∑ about O.
13. IzzO = mi (xi2 + y2i ). The z moment of inertia about point O.
i

In fact, the particle of mass we have been considering, mi , can be considered to be

infinitesimally small. In this case the summations over all particles in the rigid body can
be replaced by integrations. The mass particle becomes dm and the moments and products
of inertia are,

IxxO = (y2 + z2 ) dm (2.53)

∫body

IyyO = (x2 + z2 ) dm (2.54)

∫body

IzzO = (x2 + y2 ) dm (2.55)

∫body

IxyO = IyxO = xydm (2.56)

∫body

IxzO = IzxO = xzdm (2.57)

∫body

IyzO = IzyO = yzdm. (2.58)

∫body

Referring to Equation 2.52 and remembering that we wish to restrict the moment balance
on the body to be about a reference point, which is either the center of mass (x = y = z = 0)
or a “fixed point” (𝑣Ox = 𝑣Oy = 𝑣Oz = 0), we can now write the angular momentum vector
for the body as,
⃗ O = [Ixx 𝜔x − Ixy 𝜔y − Ixz 𝜔z ]⃗i
H O O O

+ [−IyxO 𝜔x + IyyO 𝜔y − IyzO 𝜔z ]⃗j

+ [−IzxO 𝜔x − IzyO 𝜔y + IzzO 𝜔z ]k⃗ (2.59)

or, if we write the vector as,

⃗ O = Hx⃗i + Hy⃗j + Hz k⃗
H (2.60)

@Seismicisolation
@Seismicisolation
36 2 Newton’s Equations of Motion

⃗ O can be found from,

the components of H
⎧ Hx ⎫ ⎡ IxxO −IxyO −IxzO ⎤ ⎧ 𝜔x ⎫
⎪ ⎪ ⎢ ⎥⎪ ⎪
⎨ Hy ⎬ = ⎢ −IyxO IyyO −IyzO ⎥ ⎨ 𝜔y ⎬ . (2.61)
⎪ ⎪ ⎢ ⎥⎪ ⎪
⎩ Hz ⎭O ⎣ −IzxO −IzyO IzzO ⎦ ⎩ 𝜔z ⎭
We can write Equation 2.61 in general as,
{H}O = [IO ]{𝜔} (2.62)
where [IO ] is known as the inertia tensor.
For the moment balance we have been considering, we need the angular momentum
vector only to write its derivative with respect to time. As with all vectors, the angular
momentum vector will have a rate of change of magnitude and a rate of change of direc-
tion. Of these, the rate of change of direction is by far the most interesting because it leads
to gyroscopic effects, which will be discussed more fully later. The derivative of the angular
momentum vector is,
⃗O
dH
= Ḣ x ⃗𝚤 + Ḣ y ⃗𝚥 + Ḣ z k⃗ + 𝜔 ⃗O
⃗ ×H . (2.63)
dt ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏟⏟
rate of change rate of change
of magnitude of direction

2.10 A Physical Interpretation of Moments and Products

of Inertia
Consider the simple system shown in Figure 2.5. There is a single mass particle m that is
constrained by a massless rigid link to remain a fixed distance from the point O. We consider
the case where the mass moves in a circular horizontal path in three dimensional space. A
⃗ is defined so that k⃗ is aligned with the axis of rotation of the mass,
coordinate system (⃗i, ⃗j, k)
and ⃗i and ⃗j rotate about k⃗ so that the mass remains in the plane formed by ⃗i and k. ⃗ The
position of the mass with respect to point O is therefore,
p⃗ m∕O = x⃗i + zk⃗ (2.64)

Figure 2.5 A single mass system to help interpret

ω inertia properties.
⇀
k

x
m
⇀
j

z
O ⇀
i

@Seismicisolation
@Seismicisolation
 2.10 A Physical Interpretation of Moments and Products of Inertia 37

where the distances x and z are constant with time.

Let the angular velocity of the reference frame be,

⃗ = 𝜔k⃗
𝜔 (2.65)

where the magnitude 𝜔 is not constant. The mass therefore has an angular acceleration of,
⃗
𝛼⃗ = 𝜔̇ k. (2.66)

The reason for choosing such a simple system is that the expression for the angu-
lar momentum is relatively simple but still contains all of the terms needed to see how
moments and products of inertia enter the equations of motion. Considering the expression
for the angular momentum vector given in Equation 2.59 and setting 𝜔z = 𝜔 as well as
𝜔x = 𝜔y = 0 results in,

H ⃗
⃗ O = (−Ixz 𝜔)⃗i + (−Iyz 𝜔)⃗j + (Izz 𝜔)k. (2.67)
O O O

The integrations over the body required to find the moment of inertia (IzzO ) and the prod-
ucts of inertia (IxzO and IyzO ) are made simple by the fact that there is only a single mass
particle and that its ⃗j coordinate is zero (i.e. y = 0). The resulting expressions are,

IxzO = xzdm = m × x × z = mxz

∫body

IyzO = yzdm = m × 0 × z = 0
∫body

IzzO = (x2 + y2 ) dm = m × (x2 + 02 ) = mx2 . (2.68)

∫body

Making these substitutions in Equation 2.67 results in a simplified expression for the
angular momentum vector, as follows.

H ⃗
⃗ O = (−mxz𝜔)⃗i + (mx2 𝜔)k. (2.69)

Applying the expression for the moment balance about a fixed point
⃗O
dH
⃗O =
M
dt
to Equation 2.69 (noting that, since x and z are constant, ẋ and ż are both zero) yields,

M ̇ k⃗ + 𝜔k⃗ × H
̇ ⃗i + (mx2 𝜔)
⃗ O = (−mxz𝜔) ⃗O (2.70)

or, after performing the cross product,

M ̇ ⃗i + (−mxz𝜔2 )⃗j + (mx2 𝜔)

⃗ O = (−mxz𝜔) ⃗
̇ k. (2.71)
⏟⏞⏟⏞⏟ ⏟⏞⏞⏟⏞⏞⏟ ⏟⏟⏟
1 2 3

Before considering each of the terms in Equation 2.71 it is necessary to have an expression
for the acceleration of the mass. As usual, the derivation of an expression for the accelera-
tion starts with the position vector and differentiates twice, as follows, again noting that ẋ

@Seismicisolation
@Seismicisolation
38 2 Newton’s Equations of Motion

and ẏ are both zero,

p⃗ m∕O = x⃗i + zk⃗
d⃗pm∕O
𝑣⃗m = ⃗ + (𝜔k)
= (ẋ ⃗i + ż k) ⃗ × (x⃗i + zk)
⃗ = 𝜔x⃗j
dt
d𝑣⃗ ⃗ × (𝜔x⃗j) = −𝜔2 x⃗i + 𝜔x
a⃗ m = m = (𝜔x ̇ + 𝜔x) ̇ ⃗j + (𝜔k) ̇ ⃗j. (2.72)
dt
The terms in Equation 2.71 can be explained as follows.
1. The first term is a product of inertia term. This can be determined from the form of the
term where two different coordinates are multiplied together rather than having coordi-
nates squared as would be expected in a moment of inertia term.
The term is, in fact, an expression of the moment which must be applied about the x-axis
in order to have the specified motion of the mass. Note that the problem being consid-
ered is not one where forces are applied and the equations of motion are used to predict
the resulting motion. Instead, the motion has been predetermined (i.e. circular motion
in a horizontal plane) and the forces that cause the motion can be found from Newton’s
second law. In other words, knowing the acceleration of the mass (see Equation 2.72),
the force applied to the mass is inferred to be,
F⃗ m = −m𝜔2 x⃗i + m𝜔x
̇ ⃗j. (2.73)
Figure 2.6 shows the mass as viewed from a point on the positive x-axis. From this point,
we can see the ⃗j component of F⃗ m and its moment arm z. The resulting component
of the moment about O is MO1 = −mxz𝜔. ̇ The negative sign indicates that the moment
is in a direction opposite to a rotation about the positive ⃗i direction. This is exactly equal
to “term 1” from Equation 2.71 and indicates that this product of inertia term accounts
for the moments required to maintain the tangential acceleration x𝜔. ̇ While we have
considered only a single particle here, this statement would be true for every particle
in a general rigid body and the summation of all such terms leads to the integral over
the body which ultimately leads to the total x–z product of inertia of the body.
The part of the figure shown in dashed lines indicates what would happen if there were
another mass symmetrically placed in the body. This mass produces a moment that

⇀ Figure 2.6 Term 1 – a product of inertia term.

k

•
F = mxω

•
z M0 = − mxzω
⇀
0 j

z •
M0 = + mxzω

•
F = mxω

@Seismicisolation
@Seismicisolation
 2.10 A Physical Interpretation of Moments and Products of Inertia 39

Figure 2.7 Term 2 – a product of inertia term. ⇀

k

F = mxω2

z M0 = − mxzω2
⇀
i 0

z M0 = + mxzω2

F = mxω2

is equal and opposite to that produced by the mass we have been considering. If both
masses were present, there would be a net moment of zero about point O in this plane.
This illustrates that products of inertia for objects that are symmetric in a plane are zero.
2. The second term is also a product of inertia term. In this case, it is the moment which
must be applied about the y-axis in order for the mass to maintain its rotation in the x–y
plane.
The centripetal acceleration component, x𝜔2 , must exist during the circular motion
and there must therefore be a force equal to mx𝜔2 acting on the mass as shown
in Figure 2.7 (a view from a point on the positive y-axis). This force exerts a negative
moment, relative to the y-axis, about point O. The magnitude of the moment depends
upon the magnitude of the force and the length of the moment arm, in this case z. The
moment is therefore MO = −mxz𝜔2 . This is exactly “term 2”. Once again, dashed lines
in Figure 2.7 indicate what would happen if another, symmetrically placed, mass was
present. Since this is a product of inertia term, the two moments cancel each other.
3. The third term is different from the others because it displays the squared coordinate (in
this case x2 ) characteristic of a moment of inertia. Figure 2.8 shows a view from a point

Figure 2.8 Term 3 – a moment of inertia ⇀

j
term.
•
M0 = mx2 ω

•
F = mxω
x

F = mxω2 0 ⇀
i
F= mxω2

x
•
F = mxω

•
M0 = mx2 ω

@Seismicisolation
@Seismicisolation
40 2 Newton’s Equations of Motion

on the positive z-axis and we can see the same force that we saw in Figure 2.6. That is,
the force necessary to produce the tangential acceleration x𝜔.
̇ The difference in this view
is that the moment arm is now equal to the radius in the acceleration term. The result is
that the moment is MO = mx2 𝜔. ̇
The force causing the centripetal acceleration (F = mx𝜔2 ) can also be seen on Figure 2.8
but, since the force acts through the reference point O, it does not contribute to the
moment.
The dashed lines indicating the effect of a symmetrically placed mass in Figure 2.8 show
a moment that adds to that already there. Moments of inertia are always additive. Extra
mass simply makes the moment of inertia larger. As a result, moments of inertia are
never negative and are only zero when considering a particle or a line of particles on the
axis of rotation. If mass is distributed away from the axis of rotation, there will be a
non-zero moment of inertia.

2.11 Euler’s Moment Equations

Leonhard Euler8 developed a form of the moment balance equations for a rigid body.
Given that the angular momentum vector H ⃗ O can be expressed as three scalar compo-
⃗ fixed in the rigid body,
nents in a reference frame (⃗i, ⃗j, k)
⃗ O = Hx⃗i + Hy⃗j + Hz k⃗
H

and that the rigid body has angular velocity,

⃗ = 𝜔x⃗i + 𝜔y⃗j + 𝜔z k⃗
𝜔

then,
⃗O
dH
=H⃗̇ + 𝜔⃗ ×H⃗ O. (2.74)
O
dt
Expanding Equation 2.74 yields,
⃗O
dH
= (Ḣ x − Hy 𝜔z + Hz 𝜔y )⃗i
dt
+ (Ḣ y − Hz 𝜔x + Hx 𝜔z )⃗j
⃗
+ (Ḣ z − Hx 𝜔y + Hy 𝜔z )k. (2.75)
⃗ are the principal axes of the body (i.e. all of the products
If it is now assumed that (⃗i, ⃗j, k)
of inertia about these axes are zero) we can write,
⃗ O = (Ixx 𝜔x )⃗i + (Iyy 𝜔y )⃗j + (Izz 𝜔z )k⃗
H (2.76)
O O O

8 Leonhard Euler (1707–1783). His book Mechanica (1736), presented Newtonian dynamics in the form of
mathematical analysis for the first time. He went on to extend Newton’s laws of motion to include the
dynamics of rigid bodies.
√ Euler introduced the notation f (x) for a function, e for the base of natural
∑
logarithms, i for the −1, for summation, the notation for finite differences Δy and Δ2 y and many
others.

@Seismicisolation
@Seismicisolation
 2.12 Throwing a Spiral 41

and then,
Ḣ x = IxxO 𝜔̇ x (2.77)

Ḣ y = IyyO 𝜔̇ y
Ḣ z = IzzO 𝜔̇ z (2.78)
so that, from Equation 2.75,
⃗O
dH
= (IxxO 𝜔̇ x − IyyO 𝜔y 𝜔z + IzzO 𝜔y 𝜔z )
dt
+ (IyyO 𝜔̇ y − IzzO 𝜔x 𝜔z + IxxO 𝜔x 𝜔z )
+ (IzzO 𝜔̇ z − IxxO 𝜔x 𝜔y + IyyO 𝜔x 𝜔y ). (2.79)
⃗ O = Mx ⃗i + My ⃗j +
Finally, if we equate the total externally applied moment about O (M O O
⃗O
⃗ to
M k)
d H
from Equation 2.79 we get,
zO dt

MxO = IxxO 𝜔̇ x − (IyyO − IzzO )𝜔y 𝜔z

MyO = IyyO 𝜔̇ y − (IzzO − IxxO )𝜔x 𝜔z
MzO = IzzO 𝜔̇ z − (IxxO − IyyO )𝜔x 𝜔y . (2.80)
The moment balance equations (Equation 2.80) are known as Euler’s equations. The
assumptions built into them are,
1. They are only applicable if x–y–z are body-fixed principal axes of inertia.
2. The reference point O must be either the center of mass G or a fixed point.

2.12 Throwing a Spiral

If we consider axisymmetric bodies, the products of inertia are zero by virtue of the sym-
metry and Euler’s equations are readily applicable. In the case of bodies with no external
moments applied, Euler’s equations become very useful and interesting. Most modern text-
books take up the subject of symmetric satellites at this point but, to keep this description
more down to earth, we look at the problem of throwing a perfect spiral with an American
style football.
Figure 2.9 shows the situation to be considered. The quarterback has thrown the football
⃗ shown
and it is in flight. We describe its motion using the body-fixed reference frame (⃗i, ⃗j, k)
in the figure. We take the reference point to be the center of mass of the football, which
is, by symmetry, located at its geometric center. The moments of inertia are unknown but,
again by symmetry, we know that IyyG = IzzG and that IxxG is different. Let IxxG = I1 and IyyG =
IzzG = I2 .
Since there is no externally applied moment, MGx = MGy = MGz = 0, and we can write
Euler’s equations (Equation 2.80) for this case as,
0 = I1 𝜔̇ x − (I2 − I2 )𝜔y 𝜔z
0 = I2 𝜔̇ y − (I2 − I1 )𝜔x 𝜔z
0 = I2 𝜔̇ z − (I1 − I2 )𝜔x 𝜔y . (2.81)

@Seismicisolation
@Seismicisolation
42 2 Newton’s Equations of Motion

⇀ Figure 2.9 An American football in flight.

j

⇀
ωy i

ωx

G
⇀
k
ωz

The first relationship in Equation 2.81 can be simplified to,

𝜔̇ x = 0.

This simply says that the angular velocity about the x-axis will not change with time. In
other words, when the football leaves the hands of the quarterback, the rate of spin about
the long axis through the ball is fixed. Since the quarterback intends to throw a “spiral”,
this angular velocity is usually relatively large. We can define the constant angular velocity
about the x-axis to be 𝜔x = Ω.
The second and third relationships in Equation 2.81 speak to the development of angular
velocities about the short axes of the football. A true spiral will have an angular velocity
only about the long axis (i.e. Ω ≠ 0, 𝜔y = 𝜔z = 0).
The second and third relationships can be simplified to,
( )
I − I1
𝜔̇ y = 2 Ω 𝜔z
I2
( )
I − I2
𝜔̇ z = 1 Ω 𝜔y .
I2
If the initial values of the angular velocities about the short axes are zero (i.e. when
the ball leaves the hand of the quarterback 𝜔y = 𝜔z = 0), then these two equations show
that the rates of change of 𝜔y and 𝜔y will be zero and angular velocities about the short axes
will never develop. In other words, the football will spin only about its long axis and the
pass is a perfect spiral. If, on the other hand, the quarterback introduces a slight “wobble”
by having an initial non-zero value of either 𝜔y or 𝜔z , then the rates of change of both 𝜔y
and 𝜔z will be non-zero as time progresses and the pass will not be a spiral.

2.13 A Two Body System

Figure 2.10 shows a wedge of mass M resting on a frictionless horizontal surface. A cylinder
(radius = R, mass = m) is resting on the inclined surface of the wedge. A spring of stiffness
k and rest length 𝓁0 connects the center of the cylinder to a point on the wedge. The spring
is parallel to the inclined surface. There is Coulomb friction between the cylinder and the
wedge. The coefficient of friction is 𝜇.
The equations of motion for the system are desired.

@Seismicisolation
@Seismicisolation
 2.13 A Two Body System 43

Figure 2.10 A cylinder on a wedge.

x2
g ⇀
θ A j
⇀ k
j1 ⇀
G2 i ⇀
⇀ k
⇀
i1 k1
G1 x1
β

Section 2.4 gave a step by step procedure for deriving equations of motion. We will employ
that procedure here.
1. Kinematics – choose the coordinates (also known as degrees of freedom) to be used
to describe the motion and derive expressions for the absolute velocities and accelerations
of the masses under consideration.
We begin by choosing the degrees of freedom of the system. Degrees of freedom are coor-
dinates that, when given values, will specify the translational and rotational position
of each body in the system. It is often helpful to consider making a drawing of the sys-
tem as it moves and asking yourself how many coordinates (degrees of freedom) you
would need to make the drawing. In this case, there are three degrees of freedom.
The wedge can only translate horizontally on the frictionless surface – use x1 to describe
the horizontal motion of the wedge away from its initial position. Clearly, if we know
where the wedge was initially and we know how far it has translated horizontally from
that position, we can draw the wedge. x1 is the first degree of freedom.
The center of the cylinder can move up and down the inclined surface of the wedge. As
it does, the length of the spring changes. If we use the variable x2 to denote the length
of the spring, we can determine both the force in the spring and the position of the center
of the cylinder relative to a point on the wedge. x2 is the second degree of freedom.
Since the cylinder is perfectly round and appears to have no distinguishing marks on its
surface, the third degree of freedom is not quite as obvious. Imagine though that the end
of the cylinder had a mark that we could follow as the cylinder rotated. At some times
the mark would be at the top of the cylinder. At other times it would be at the bottom.
We could not draw the system unless we knew how far the cylinder had rotated from its
initial position. The third degree of freedom is the rotation of the cylinder. Let this be 𝜃.
x1 , x2 , 𝜃 are shown in their assumed positive senses on Figure 2.10.
Next, we need to derive kinematic expressions for the absolute acceleration of the var-
ious bodies. In the end we will write differential equations of motion related to each
degree of freedom so we will, at a minimum, need the absolute accelerations in the direc-
tions of the degrees of freedom. That is, we will need the horizontal acceleration of the
wedge since it has already been decided that it can only move horizontally. We will need
the translational acceleration of the center of mass of the cylinder in a direction aligned
with the inclined plane on the wedge because of the definition of x2 . Finally, we will need
an expression for the angular acceleration of the cylinder in order to write an equation
of motion for the angular coordinate 𝜃.

@Seismicisolation
@Seismicisolation
44 2 Newton’s Equations of Motion

The accelerations can be expressed as follows.

The horizontal acceleration of the wedge is found by successive differentiations starting
from the position vector, as follows.
p⃗ G1 = x1⃗i (2.82)

𝑣⃗G1 = ẋ 1⃗i (2.83)

a⃗ G1 = ẍ 1⃗i. (2.84)
The absolute acceleration of the center of mass of the cylinder can be developed from
the position vector as shown in Equation 2.85. Two facts used in the development are,
• a⃗ A = a⃗ G1 since all points on the wedge have the same motion as it translates
horizontally.
• The unit vectors are related by the constant angle 𝛽 as follows.
⃗i = cos 𝛽⃗i1 + sin 𝛽⃗j1

⃗j = − sin 𝛽⃗i1 + cos 𝛽⃗j1

p⃗ G2 = p⃗ A + p⃗ G2 ∕A
p⃗ G2 = p⃗ A + x2⃗i1
𝑣⃗G2 = 𝑣⃗A + ẋ 2⃗i1
a⃗ G2 = a⃗ A + ẍ 2⃗i1
a⃗ G2 = ẍ 1⃗i + ẍ 2⃗i1
a⃗ G2 = ẍ 1 (cos 𝛽 ⃗i1 + sin 𝛽 ⃗j1 ) + ẍ 2⃗i1
a⃗ G2 = (̈x1 cos 𝛽 + ẍ 2 )⃗i1 + ẍ 1 sin 𝛽 ⃗j1 . (2.85)
The angular acceleration of the cylinder is found from,
⃗ cyl = −𝜃̇ k⃗
𝜔
⃗
𝛼⃗cyl = −𝜃̈ k. (2.86)

Note that the angle 𝜃 is defined to be positive in the negative k⃗ direction so that its angular
velocity and angular acceleration are also in the negative k⃗ direction.
2. Free body diagrams (FBDs) – sketch the masses under consideration as if they are in space
with no forces acting on them. Then add to the sketches all of the externally applied forces
acting on the masses. Also add the internal forces of interaction between the masses, being
sure that they act in equal and opposite pairs as stipulated by Newton’s third law. The FBDs
should show the positive sense of the accelerations derived in step 1.
Figure 2.11 shows the free body diagrams of the cylinder and the wedge. The forces
shown are,

@Seismicisolation
@Seismicisolation
 2.13 A Two Body System 45

Figure 2.11 FBDs of the cylinder and the Fs

wedge.
G
Ff1

g mg N1
⇀
A j
⇀ Fs
j1 ⇀
N1 i ⇀
⇀ Ff1 k
⇀
i1 k1

β Mg

Ff 2
N2

• N1 – the normal force acting between the cylinder and the wedge. By definition, this
force must be perpendicular to the plane of the wedge. Its magnitude is unknown.
• N2 – the normal force between the wedge and the supporting surface. Again, the direc-
tion is known but the magnitude is not. This force is not, in fact, required for the analysis
being done here since we are only interested in horizontal motions of the wedge, but is
included for completeness.
• Ff1 – the frictional force acting between the cylinder and the wedge. The direction is
known to be perpendicular to the normal force and in opposition to the relative velocity
between the two bodies at the point of contact if slip is occurring. If there is slip, the mag-
nitude of the friction force is ±𝜇N1 – if there is no slip, the magnitude of the friction force
lies between −𝜇N1 and +𝜇N1 and acts in the direction required to enforce the “no-slip”
condition.
• Ff2 – the frictional force acting between the wedge and the supporting surface is
included for completeness. It is zero in this problem because the contact was designated
as frictionless.
• Fs – this is the force in the spring connecting the cylinder to the wedge. The spring stiff-
ness was given as k, and the undeflected length as 𝓁0 . In order to be sure that the sign
of the force is correct as we develop an expression for it, we must refer to the assumed
direction shown on the FBD. Figure 2.11 shows that the spring was assumed to be in ten-
sion (i.e. the forces are acting to pull the two bodies together). If this is the case, then it
must also be assumed that the spring is longer than its undeflected length (i.e. x2 > 𝓁0 ).
The displacement of the spring is therefore 𝛿 = x2 − 𝓁0 and the force in the spring, acting
in the direction on the FBD, is Fs = k𝛿 = k(x2 − 𝓁0 ).
• mg and Mg – these are the weights of the two bodies acting vertically downward.
3. Force Balance Equations (Newton’s second law) – using the FBDs, equate the vector sum
of forces on each body to its mass multiplied by its vector acceleration and the vector sum
of moments on each body to its moment of inertia multiplied by its angular acceleration9 .
Referring to Figure 2.11, we write the force and moment balance equations as follows.

⃗G
⃗ =
9 In general we should use M
dH ⃗ = I 𝛼⃗ .
but, in two dimensional cases, this is often simplified to M
G dt G G

@Seismicisolation
@Seismicisolation
46 2 Newton’s Equations of Motion

First we recognize that we simply require a horizontal force balance on the wedge, lead-
ing to,
∑
F⃗i ∶ M ẍ1 = Ff1 cos 𝛽 − N1 sin 𝛽 + Fs cos 𝛽. (2.87)

Next, the cylinder is accelerating in both the ⃗i1 and ⃗j1 (see Equation 2.85) directions so
that two force balances are required. They are,
∑
F⃗i ∶ m(̈x1 cos 𝛽 + ẍ 2 ) = −Ff1 − Fs + mg sin 𝛽 (2.88)
1

∑
F⃗j ∶ m̈x1 sin 𝛽 = N1 − mg cos 𝛽. (2.89)
1

Finally, the angular acceleration of the cylinder requires that we write,

∑
Mk⃗ ∶ −I 𝜃̈ = −Ff1 R. (2.90)
1

4. Manipulate and solve the equations – the first three steps will lead to a set of equations
with a set of unknowns. All that is left is to manipulate and combine the equations in order
to extract the desired solution.
We first count the number of equations and the number of unknowns to see if we need
further information before attempting a solution. The full set of equations available to us
are five in total.

M ẍ1 = Ff1 cos 𝛽 − N1 sin 𝛽 + Fs cos 𝛽

m (̈x1 cos 𝛽 + ẍ 2 ) = −Ff1 − Fs + mg sin 𝛽
m̈x1 sin 𝛽 = N1 − mg cos 𝛽
− I 𝜃̈ = −Ff R 1

Fs = k(x2 − 𝓁0 ).

Counting the unknowns, we find that there are the following six.
• x1 and its derivatives10
• x2 and its derivatives
• 𝜃 and its derivatives
• Ff1
• N1
• Fs .
Given five equations with six unknowns means we need to find another relationship
between variables before we have a chance of generating a solution. In fact, we have
already discussed the relationship required – we know how the frictional force between
the cylinder and the wedge behaves.
We have two alternatives. If the cylinder rolls without slipping, Ff1 is whatever it must be
to enforce the “no-slip” condition and we can generate a kinematic relationship between
the velocity of the center of the cylinder and its angular velocity.

10 A variable and its derivatives are counted as a single unknown since they are related by differential
equations we do not derive. For example, ẋ = dx
dt
goes without saying.

@Seismicisolation
@Seismicisolation
 2.13 A Two Body System 47

If the cylinder rolls and slips, then there is relative motion at the point of contact between
the two bodies and the frictional force is known to be at its limiting value, Ff1 = 𝜇N1 ,
with a direction opposing the relative motion.
Consider first the case where the cylinder rolls without slipping. Define the point of con-
tact between the cylinder and the wedge to be point B. The definition of the “no-slip”
condition is that point B must have the same velocity whether we consider it a point
on the wedge or a point on the cylinder.
If point B is on the wedge, we know its velocity is the same as that of any other point
on the wedge because the wedge simply translates. That is,
= ẋ 1⃗i.
wedge
𝑣⃗ B (2.91)
If point B is on the cylinder, we find its velocity as follows.
cyl
p⃗ B = p⃗ A + p⃗ B∕A
p⃗ B = p⃗ A + x2⃗i1 − R⃗j2
cyl

𝑣⃗ B = 𝑣⃗A + ẋ 2⃗i1 − (−𝜃̇ k⃗2 ) × (R⃗j2 )

cyl

𝑣⃗ B = 𝑣⃗A + ẋ 2⃗i1 − R𝜃̇ ⃗i2

cyl

𝑣⃗ B = ẋ 1⃗i + ẋ 2⃗i1 − R𝜃̇ ⃗i2

cyl

𝑣⃗ B = ẋ 1⃗i + (ẋ 2 − R𝜃)

̇ ⃗i1 .
cyl
(2.92)
What was just done needs explanation. Defining the position of point B in the cylinder
requires that the relative position of two points in a rigid body (the cylinder) be defined.
The relative velocity of these two points (G2 and B) can only be due to the angular velocity
of the coordinate system in which they are expressed. Clearly, the two coordinate systems
⃗ and (⃗i , ⃗j , k⃗ ), do not rotate and are incapable of expressing
used in the analysis, (⃗i, ⃗j, k) 1 1 1
𝑣⃗B∕G2 . It was therefore necessary to introduce the coordinate system (⃗i2 , ⃗j2 , k⃗2 ), which is
fixed in the cylinder and has angular velocity −𝜃̇ k⃗2 . (⃗i2 , ⃗j2 , k⃗2 ) is aligned with (⃗i1 , ⃗j1 , k⃗1 )
at the instant being considered so the final result has been written in the (⃗i , ⃗j , k⃗ ) system.
1 1 1
Equating the two expressions for 𝑣⃗B from Equations 2.91 and 2.92 gives,
wedge cyl
𝑣⃗ B = 𝑣⃗ B
ẋ 1⃗i = ẋ 1⃗i + (ẋ 2 − R𝜃)
̇ ⃗i1
̇ ⃗i1
0 = (ẋ 2 − R𝜃)
ẋ 2 = R𝜃.̇ (2.93)
The result from Equation 2.93 becomes the sixth equation required for a solution. We
then combine the equations and eliminate the variables Ff1 , N1 , Fs , and 𝜃 and its deriva-
tives to get,
1
𝜃̇ = (ẋ 2 )
R ( )
I ̈ I ẍ 2 I
Ff1 = (𝜃) = = 2 (̈x2 )
R R R R
N1 = m̈x1 sin 𝛽 + mg cos 𝛽
Fs = k(x2 − 𝓁0 ) (2.94)

@Seismicisolation
@Seismicisolation
48 2 Newton’s Equations of Motion

and, finally,
⎡ (M + m sin2 𝛽) − I2 cos 𝛽 ⎤ { }
⎢ (R )⎥ ẍ 1
⎢ m cos 𝛽 m + RI2 ⎥ =
⎢ ⎥ ẍ 2
⎢ ⎥
⎣ ⎦
{ }
k(x2 − 𝓁0 ) cos 𝛽 − mg cos 𝛽 sin 𝛽
. (2.95)
mg sin 𝛽 − k(x2 − 𝓁0 )
The end result is a set of two, simultaneous, nonlinear, second order, differential
equations in the two variables x1 and x2 . Later, we will discuss methods for solving
the differential equations arising from derivations such as this. For the time being, we
must simply realize that there is a significant assumption built into these equations.
Namely that the cylinder rolls without slipping on the wedge. Any solution to the
equations must continually check to see if the required value of Ff1 exceeds 𝜇N1 .
If it does, these differential equations are no longer applicable and we must derive
the equations of motion corresponding to Ff1 = 𝜇N1 and use them for the solution.

2.14 Gyroscopic Motion

Gyroscopic motion occurs whenever the axis about which a body is spinning is itself rotating
about another axis. Gyroscopic effects hold a certain mystique because they never behave
in the way everyday experience would tell people they should.
In fact, gyroscopic effects can be readily understood from the equations we have already
developed and it is the purpose of this section to give the reader that understanding without
unduly complicating matters with equations. We seek a feeling for gyroscopic effects that
comes from an understanding of the rate of change of direction of a vector – in this case,
the angular momentum vector.
A moment balance is always involved in gyroscopic considerations. The terms involving
the rate of change of direction of the angular momentum vector are the “gyroscopic terms”
in the equations of motion.
∑ ⃗O
dH
⃗O =
M = H⃗̇O + 𝜔 ⃗O .
⃗ ×H (2.96)
dt ⏟⏟⏟
gyroscopic
terms
Equation 2.59 gives a general expression for the angular momentum vector, either
about a fixed point or about the center of mass of the body. If we restrict the discussion to
axi-symmetric bodies and use the center of mass G as the reference point, the products of
inertia will vanish and the angular momentum vector will simply become,
⃗ G = Ixx 𝜔x⃗i + Iyy 𝜔y⃗j + Izz 𝜔z k⃗
H (2.97)
G G G

⃗ are fixed in the rigid body.

where the axes (⃗i, ⃗j, k)

@Seismicisolation
@Seismicisolation
 2.14 Gyroscopic Motion 49

Further, if we assume that one component of the angular velocity vector is significantly
larger than the other two (say, for example, 𝜔x ≫ 𝜔y and 𝜔x ≫ 𝜔z ) then we can approximate
the angular momentum vector as,
H⃗G ≈ IxxG 𝜔x⃗i (2.98)
or,
⃗ ≈ I𝜔
H ⃗ (2.99)
where the subscripts have been dropped to show that the angular momentum of the body
is essentially aligned with its “spin axis”. That is, the axis about which the angular speed is
the greatest.
It is easy to think of many examples where there is a dominant component to the angular
velocity vector. A top spinning on a horizontal surface, for instance, has an angular velocity
vector and a resulting angular momentum vector that are directed primarily in the vertical
direction. The top may well be wobbling but the components of the angular velocity about
directions other than the vertical will be relatively small.
Another example is a bicycle. If we consider the bicycle to be made of four compo-
nents – the front forks, the frame, and two wheels, only the wheels rotate as the bicycle
moves. A bicycle moving forward will have a large angular momentum vector generated
by each wheel and the angular momentum vectors can be visualized as pointing out of the
wheel centers to the left11 . The angular momentum vectors will get longer as the bicycle
goes faster but the direction will not change.
There are numerous other examples of systems with rapidly rotating components that
generate significant angular momentum vectors. Think of the engines in cars that rotate
rapidly about fixed directions in the car. Think of jet aircraft whose engines have very high
rates of rotation and always point the same way relative to the plane.
Figure 2.12 shows a top spinning on a horizontal surface. The coordinate system is fixed
in the body of the top and therefore spins with it, having an angular velocity 𝜔 ⃗ = 𝜔k.⃗ The

Figure 2.12 A spinning top. ⇀ ⇀

H = Iω

⇀
j

⇀
i

11 Check this statement by visualizing the direction in which the wheels turn and using the right hand
rule to determine the direction of the angular momentum vector.

@Seismicisolation
@Seismicisolation
50 2 Newton’s Equations of Motion

⇀ Figure 2.13 A bicycle.

V

⇀ ⇀
H = Iω
ω

⇀ ⇀
H = Iω

moment of inertia of the top about the spin direction is I so that the angular momentum
vector is H⃗ = I 𝜔.
⃗ From the figure it is clear that the angular momentum vector is aligned
with the angular velocity vector and is oriented vertically upward as shown12 .
Figure 2.13 shows the situation for a bicycle that is being ridden in a straight line. The
angular momentum vectors of the two wheels are shown pointing horizontally to the left
of the bicycle. In fact, the moment of inertia of a bicycle wheel is relatively large and the
angular velocity grows with forward speed so that the magnitude of the angular velocity
vectors is large. The reader can imagine that the equivalent vectors for a motorcycle trav-
eling at high speeds will be very large indeed. The total angular momentum vector of the
bicycle is the sum of the two vectors shown in Figure 2.13.
We now turn our attention to the action that we would be required to take in order to
reorient the top or the bicycle.
Consider first the top. Figure 2.14 shows the top after someone has tipped it slightly so that
the point A has moved vertically downward. We can assume that the change in orientation
was caused by a force applied vertically downward somewhere on the periphery of the top.
The question is “where do you apply the force?”.
Quite clearly, if the top were not spinning a force applied vertically downward at point
A would do the trick. However, since there is a significant angular momentum vector, we
have to take account of the gyroscopic effect13 . Figure 2.14 shows the angular momentum
vector both before (H ⃗ 1 ) the tipping takes place. The magnitude of the vector
⃗ 0 ) and after (H
is not likely to change significantly as it tips. The major change in the angular momentum
vector is a directional change. The directional change is indicated by the vector ΔH ⃗ shown
in the figure.
ΔH⃗ is aligned with the unit vector ⃗i in Figure 2.14 since the point A lies on that axis. If
dH⃗
the change in the orientation of the top occurred over a period of time Δt, then can
dt

12 The angular velocity vector of the top is that which would be imparted if it were thrown by a
left-handed person. Right-handed readers would spin it in the opposite direction but that would mean that
the angular momentum vector would point downward and the figure wouldn’t be nearly so clear.
13 It is the angular momentum vector and its resistance to changing direction that causes the top to stand
upright in the first place.

@Seismicisolation
@Seismicisolation
 2.14 Gyroscopic Motion 51

Figure 2.14 The “tipped” top. ⇀ ⇀

H0 ΔH

⇀
H1

F
⇀
j
B

A
⇀
i

ΔH ⃗ ΔH ⃗
be approximated by and, since Δt is a scalar, the direction of is the same as the
Δt Δt
⃗
direction of ΔH and the moment applied to the top must also be in this direction since
⃗
M⃗ = dH .
dt
By the right hand rule, a positive moment in the ⃗i direction would require a vertical force,
F, to be applied, not at point A since there would be no moment about the ⃗i axis in that
case, but at a point like B that has a moment arm that will give rise to a moment about the ⃗i
axis. In fact, the maximum moment arm on the axisymmetric top occurs when the force is
applied at a point B located on the negative ⃗j axis. Tops always tip 90∘ away from the point of
application of a vertical force on their periphery. This is easily demonstrated experimentally
with a simple top on a tabletop.
The fact that the response to an action occurs 90∘ away from the action is a basic property
of gyroscopic effects. An analyst, using basic geometrical information about the angular
velocity vector and its directional rate of change can easily determine what to expect from
a system with significant rotating masses.
Next, we consider the bicycle and what it would take to reorient it so that it will move
in a different direction on the horizontal plane. In other words, we would like to steer the
bicycle so that it turns away from the straight path it is following in Figure 2.13. Let us
consider the case where we would like the bicycle to turn to the right. Figure 2.15 shows
the effect we must have on the angular momentum vectors in order to do this. The angular
momentum vectors are still perpendicular to the plane of the rotating wheels but they have
turned with the bicycle and, as a result, there must have been a directional rate of change
of the vectors. The original angular momentum vectors (H ⃗ 0 ) and the rotated vectors (H
⃗ 1 ) as
⃗ ⃗ ⃗
well as the change in the vectors (ΔH = H1 − H0 ) are shown on Figure 2.15. As we know, the
rate of change of the angular momentum vector (approximately equal to the total change
ΔH⃗ divided by the scalar length of time Δt over which the change occurred) must be equal
to the applied moment both in magnitude and direction. Leaving aside the magnitude, the
direction of the applied moment is easily seen to be aligned with the forward motion of the
bicycle. By the right hand rule, this means that a vertical force must have been applied to
the right or left of the centerline of the bicycle. The most easily applied force is that from

@Seismicisolation
@Seismicisolation
52 2 Newton’s Equations of Motion

⇀ Figure 2.15 The bicycle turning to the right.

ΔV ⇀
⇀ V1
V0

⇀
H1
⇀
ΔH
⇀
H0
⇀
H1
⇀
ΔH ⇀
H0

shifting the weight of the rider to the right so that the vertically downward gravity force on
the rider causes the required moment.
Any reader who has ever ridden a bicycle “no-hands” will know that shifting the rider’s
weight to the right or left is sufficient to steer a bicycle accurately along the chosen path.

Exercises

Descriptions of the systems referred to in the Exercises are contained in Appendix A.

2.1 Draw FBDs, write force and moment balance equations, and derive the equations
of motion for system 1. Show that the force acting on body OA due to its interaction
with the pin at A is
⃗ A = m2 [−g sin 𝜃1 + d1 𝜃̇ 12 + d2 (𝜃̈1 + 𝜃̈2 ) sin 𝜃2 + d2 (𝜃̇ 1 + 𝜃̇ 2 )2 cos 𝜃2 ]⃗𝚤
R
+ m2 [−g cos 𝜃1 − d1 𝜃̈1 − d2 (𝜃̈1 + 𝜃̈2 ) cos 𝜃2 + d2 (𝜃̇ 1 + 𝜃̇ 2 )2 sin 𝜃2 ] ⃗𝚥.

2.2 Draw FBDs, write force and moment balance equations, and derive the equations of
motion for system 2. Show that the tension in the string BC can be expressed as

T = mg cos(𝜃 − 𝜙) − m̈x sin 𝜙 + m𝜃̇ 2 (x − 𝓁) sin 𝜙

̇ 2 + m(𝓁 − x)𝜃̈ cos 𝜙.
− 2mẋ 𝜃̇ cos 𝜙 + md(𝜃̇ − 𝜙)

2.3 Show that the normal force that the slot exerts on the mass in system 4 has
magnitude
̇ + 2Ωx)
N = m(Ωx ̇

and indicate in which direction it acts.

@Seismicisolation
@Seismicisolation
 Exercises 53

2.4 Derive the equations of motion for system 5. Show that the friction force acting on
the oil drum can be expressed as
Ff = mg sin 𝜃 − m(d − x)𝜃̇ 2 − mr 𝜃̈ − m̈x
and indicate in which direction it acts.

2.5 Consider the particle of mass m and the rigid rod of system 6 to be a single rigid
body. Where is the center of mass of the body? Draw an FBD and explain why there
are five unknown reaction components at point A. What is the angular momentum
of the body about the center of mass?

2.6 Show that the equation of motion governing the angle 𝜃 in system 6 is
mr 𝜃̈ − m𝜔2 (d − r cos 𝜃) sin 𝜃 + mg cos 𝜃 = 0.

2.7 Draw FBDs, write force and moment balance equations, and derive the equation of
motion for system 7. Show that the component of the reaction force acting on rod
OA at point O in the direction from O to B can be expressed as
FO = F − 8mL𝜃̈ sin 𝜃 − 8mL𝜃̇ 2 cos 𝜃.

2.8 Derive the equations of motion for system 16 and show that the reaction force acting
on the rod is
RA = mg sin 𝜃 + mr 𝜃̇ 2 + m𝓁(𝜃̈ + 𝜙)
̈ sin 𝜙 + m𝓁(𝜃̇ + 𝜙)
̇ 2 cos 𝜙.

2.9 Show that the angular momentum of the rectangular mass of system 22 about its
center of mass and expressed in a right-handed set of body fixed principal axes where
⃗𝚤 is aligned with AB and k⃗ is aligned with BC is
( )
⃗ G = 1 mb2 𝜔1⃗𝚤 + 5 𝜔0 sin 𝜃⃗𝚥 + 5 𝜔0 cos 𝜃 k⃗ .
H
6 2 2

2.10 Consider system 22 in the case where 𝜔0 and 𝜔1 are both constant. Show that the
moment that must be applied about the center of mass of the rectangular body is

M⃗ G = 1 mb2 𝜔0 𝜔1 (− cos 𝜃⃗𝚥 + sin 𝜃 k)

⃗
4
where the unit vectors are the same set used in Exercise 2.9.

2.11 Use Newton’s Laws to derive the equations of motion for system 23.

2.12 Using the angle, 𝜃, as the single degree of freedom, show that the equation-of-motion
of the rod in system 12 can be written as
4 2̈
md 𝜃 + 4kd2 sin 𝜃 cos 𝜃 − mgd cos 𝜃 = 0.
3

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 55

Lagrange’s Equations of Motion

In this chapter, we consider the development of Lagrange’s1 equations of motion. We start

with Newton’s laws and work from them to derive a method of writing equations of motion
by taking derivatives of scalar expressions for kinetic and potential energy in the system.
Use of Lagrange’s equations gives the analyst two distinct advantages when deriving the
equations of motion. First, the vector kinematic analysis is shorter than it is with a direct
application of Newton’s laws since acceleration vectors need not be found. This is because
the kinetic and potential energy expressions can be derived from velocity vectors and posi-
tion vectors respectively. Secondly, there is no need to draw free body diagrams for each of
the rigid bodies in the system because the forces of constraint between the bodies do no
work and are therefore not required for the analysis.
Of course, there are also disadvantages. The method requires a great deal of
differentiation, sometimes of relatively complicated functions. Some analysts prefer
the kinematics of Newton’s method over the differentiation involved in Lagrange’s
equations. Some point to a lack of physical feeling for problems without free body diagrams
as being a disadvantage of the method. Finally, if the intent of analyzing the dynamics of
a system is to predict loads, which could be carried forward into a structural analysis for
instance, the forces of interaction between bodies are not available from a straightforward
application of Lagrange’s equations.

3.1 An Example to Start

Rather than going directly into the derivation of Lagrange’s equation we will start with a
relatively simple example so that the form of the equation and the way it is applied can
be seen.
Lagrange’s equation is
( )
d 𝜕T 𝜕T 𝜕U
− + = Qq (3.1)
dt 𝜕 q̇ 𝜕q 𝜕q

1 Joseph-Louis Lagrange (1736–1813), an Italian/French mathematician, is well known for his work on
calculus of variations, dynamics, and fluid mechanics. In 1788 Lagrange published the Mécanique
Analytique summarizing all the work done in the field of mechanics since the time of Newton, thereby
transforming mechanics into a branch of mathematical analysis.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
56 3 Lagrange’s Equations of Motion

k Figure 3.1 A mass on a wire.

i

j
O

R
g
θ m

where

● T = the total kinetic energy of the system

● U = the total potential energy of the system
● q = a generalized coordinate
● q̇ = the time derivative of q
● Qq = the generalized force corresponding to a variation of q.

The reader will certainly have some feeling for the kinetic and potential energy of a system
but probably not for the concepts of generalized coordinate and generalized force. The full
definitions of these terms will be given later. At this point we will simply apply Lagrange’s
equation to a system as if all were understood.
Figure 3.1 shows a small mass, m, which slides on a semi-circular wire that rotates about
a vertical axis. The wire has radius R. Gravity acts to pull the mass to the bottom of the
semi-circle while centripetal effects try to move it to the top. The equation of motion gov-
erning the single degree of freedom, 𝜃, is desired.
We first determine the kinetic energy of the system. This requires that we find an expres-
sion for the absolute velocity of the mass. Using the rotating coordinate system shown in
the figure, we can write
⃗
p⃗ m∕O = R sin 𝜃⃗𝚤 − R cos 𝜃 k. (3.2)

The absolute velocity is

d
𝑣⃗m = 𝑣⃗O + p⃗ . (3.3)
dt m∕O
Then, recognizing that 𝑣⃗O = 0 and that Ṙ = 0,
⃗ + 𝜔k⃗ × (R sin 𝜃⃗𝚤 − R cos 𝜃 k)
𝑣⃗m = (R𝜃̇ cos 𝜃⃗𝚤 + R𝜃̇ sin 𝜃 k) ⃗ (3.4)

@Seismicisolation
@Seismicisolation
 3.1 An Example to Start 57

which can be simplified to,

⃗
𝑣⃗m = R𝜃̇ cos 𝜃⃗𝚤 + 𝜔R sin 𝜃⃗𝚥 + R𝜃̇ sin 𝜃 k. (3.5)
The kinetic energy of the system is then
1
T= m(𝑣⃗m ⋅ 𝑣⃗m ) (3.6)
2
which becomes, after substitution of Equation 3.5 and some simplification,
1
T= mR2 (𝜃̇ 2 + 𝜔2 sin2 𝜃). (3.7)
2
The potential energy of the system is due to gravity only. If the datum for potential energy
is taken to be at point O, the potential energy, U, of the system is simply
U = −mgR cos 𝜃. (3.8)
Having expressions for T and U and a single degree of freedom, 𝜃, we can apply Lagrange’s
equation (Equation 3.1) and find
q=𝜃
q̇ = 𝜃̇
𝜕T
= mR2 𝜃̇
𝜕 𝜃̇
( )
d 𝜕T
= mR2 𝜃̈
dt 𝜕 𝜃̇
𝜕T
= mR2 𝜔2 sin 𝜃 cos 𝜃
𝜕𝜃
𝜕U
= mgR sin 𝜃
𝜕𝜃
Q𝜃 = 0. (3.9)
Substituting the expressions from Equation 3.9 into Lagrange’s equation gives the desired
equation of motion
mR2 𝜃̈ − mR2 𝜔2 sin 𝜃 cos 𝜃 + mgR sin 𝜃 = 0. (3.10)
Clearly, Equation 3.10 could be further simplified by factoring out the group mR but this
would take away the ability to look at the individual terms and give a physical explanation
for them. Whenever an equation is derived, the first test for correctness is to see if all of
the terms have the same dimensions. In this case, the first term has dimensions of ML2 ∕T 2
where M is mass, L is length, and T is time. Note that angles such as 𝜃 are dimensionless since
they are defined by an arc length divided by a radius. It follows that trigonometric functions
such as sin 𝜃 and cos 𝜃 are also dimensionless. Angular velocities therefore have dimensions
derived from angles divided by time, 1∕T, and angular accelerations are expressed as 1∕T 2 .
Using these conventions, it is easy to see that all three terms in Equation 3.10 have the same
dimensions2 .

2 Note the difference between dimensions and units. Dimensions refer to physical characteristics such as
mass, length, or time. Units refer to the system of measurement we use to substitute numbers into an
equation. Examples are kilograms for mass, feet for length, and minutes for time.

@Seismicisolation
@Seismicisolation
58 3 Lagrange’s Equations of Motion

The dimensions of force are ML∕T 2 or mass times acceleration. Taking this into account,
we can see that the three terms in Equation 3.10 all have dimensions of FL or force times
length. The terms are, in fact, all moments. The third term is the most obvious because
it contains the gravity force mg multiplied by a moment arm of R sin 𝜃. The moment arm
is simply the horizontal distance between the mass and point O. Lagrange’s equation has
produced an equation of motion based on a dynamic moment balance about the stationary
point O and it did so without requiring the derivation of acceleration expressions, the draw-
ing of free body diagrams, or the production of force and moment balance relationships.
This is the power of using Lagrange’s equation for deriving equations of motion.

3.2 Lagrange’s Equation for a Single Particle

In this section we start with Newton’s laws and derive from them Lagrange’s equation for
a particle.
Assume there is a single particle of mass m moving in three dimensional space and its
position is expressed using three Cartesian coordinates3 x, y, and z. Associated with each
Cartesian direction is a unit vector so that the position of the particle with respect to a
⃗ The Cartesian coordinate system is
fixed point in space can be written as p⃗ = x⃗𝚤 + y⃗𝚥 + zk.
non-rotating.
Further assume that there is a total external force, F⃗ = Fx ⃗𝚤 + Fy⃗𝚥 + Fz k⃗ acting on the par-
ticle. Then we can write the equations of motion for the particle as
m̈x = Fx
m̈y = Fy
mż = Fz . (3.11)
Now consider the work done by the externally applied forces during a completely arbi-
trary, infinitesimally small, displacement 𝛿⃗s in three dimensional space. We call this virtual
work.
⃗ The virtual work is then
In Cartesian coordinates, we can write 𝛿⃗s = 𝛿x⃗𝚤 + 𝛿y⃗𝚥 + 𝛿zk.
⃗
found from 𝛿W = F ⋅ 𝛿⃗s or,
𝛿W = Fx 𝛿x + Fy 𝛿y + Fz 𝛿z. (3.12)
Consider multiplying each of the expressions in Equation 3.11 by its respective compo-
nent of 𝛿⃗s and adding together the terms. The result is,
m(̈x𝛿x + ÿ 𝛿y + z𝛿z)
̇ = Fx 𝛿x + Fy 𝛿y + Fz 𝛿z. (3.13)
The right hand side of Equation 3.13 is simply the virtual work defined in Equation 3.12.
The left hand side of Equation 3.13 corresponds to the change in energy in the system that
arises from having work done on it. In this case, there will be a change in kinetic energy
only. Any possible potential energy terms due to gravity or elasticity have been included in
the total force acting on the particle.

3 Named for René Descartes (1596–1650), a French philosopher and mathematician whose major work,
La géométrie, includes his application of algebra to geometry from which we now have Cartesian geometry.

@Seismicisolation
@Seismicisolation
 3.2 Lagrange’s Equation for a Single Particle 59

There are often constraints on the motion of particles that need to be taken into account.
Let us now assume that the particle in question is not free to move arbitrarily in three
dimensional space but that there is some constraint acting on it so that its motion can be
completely described by two “generalized coordinates”, q1 and q2 . The three Cartesian coor-
dinates x, y, and z must be able to be derived from the two generalized coordinates. We
are simply saying that, given values for q1 and q2 , we can use algebraic equations arising
from the constraints to find x, y, and z. As an example, consider the case where the par-
ticle is constrained to remain on an inclined surface. We can locate the particle by using
two coordinates on the inclined surface (the generalized coordinates) and then, using the
equation of the plane, we can find the three Cartesian coordinates from the two generalized
coordinates. That is
x = x(q1 , q2 )
y = y(q1 , q2 )
z = z(q1 , q2 ). (3.14)

Given that the Cartesian coordinates are related to the generalized coordinates through
the relationships in Equation 3.14, we can write
𝜕x 𝜕x
𝛿x = 𝛿q + 𝛿q
𝜕q1 1 𝜕q2 2
𝜕y 𝜕y
𝛿y = 𝛿q + 𝛿q
𝜕q1 1 𝜕q2 2
𝜕z 𝜕z
𝛿z = 𝛿q + 𝛿q . (3.15)
𝜕q1 1 𝜕q2 2
The generalized coordinates q1 and q2 must be independently variable. That is, 𝛿q1 and
𝛿q2 must be able to be given arbitrary small values independently without violating the
constraints.
We further assume that the constraint is smooth so that the work done by the force of
constraint is zero for arbitrary 𝛿q1 and 𝛿q2 . This is not as restrictive a condition as it may
seem at first sight. Most constraints in mechanical systems act to prevent displacements of
bodies with respect to each other. Consider, for example two bodies that are pinned together
at a common point. The constraint is that the two bodies cannot move apart at that point and
the constraint forces will ensure that this is actually the case. Since there is no displacement
associated with the constraint forces, no work is done.
Under these conditions, we can substitute Equation 3.15 into Equation 3.13 to yield the
following expression for the virtual work.
( ) ( )
𝜕x 𝜕y 𝜕z 𝜕x 𝜕y 𝜕z
𝛿W = m ẍ + ÿ + ż 𝛿q1 + m ẍ + ÿ + ż 𝛿q2
𝜕q1 𝜕q1 𝜕q1 𝜕q2 𝜕q2 𝜕q2
( ) ( )
𝜕x 𝜕y 𝜕z 𝜕x 𝜕y 𝜕z
= Fx + Fy + Fz 𝛿q1 + Fx + Fy + Fz 𝛿q2 . (3.16)
𝜕q1 𝜕q1 𝜕q1 𝜕q2 𝜕q2 𝜕q2
Since q1 and q2 are independently variable, we may consider the case where 𝛿q2 = 0 and
𝛿q1 ≠ 0. In this case, Equation 3.16 reduces to the work done during an arbitrary variation
of q1 with q2 being held constant. In more general systems, we think of varying one of the
generalized coordinates while holding all others constant.

@Seismicisolation
@Seismicisolation
60 3 Lagrange’s Equations of Motion

Equation 3.16 then reduces to

( )
𝜕x 𝜕y 𝜕z
𝛿Wq1 = m ẍ + ÿ + ż 𝛿q1
𝜕q1 𝜕q1 𝜕q1
( )
𝜕x 𝜕y 𝜕z
= Fx + Fy + Fz 𝛿q1 . (3.17)
𝜕q1 𝜕q1 𝜕q1
We now consider the terms on the left hand side of Equation 3.17. A representative term
is
𝜕x
ẍ . (3.18)
𝜕q1
𝜕x
We can derive this term from differentiating ẋ with respect to time. That is
𝜕q1
( ) ( )
d 𝜕x 𝜕x d 𝜕x
ẋ = ẍ + ẋ . (3.19)
dt 𝜕q1 𝜕q1 dt 𝜕q1
⏟⏟⏟
the term

Equation 3.19 can be reorganized to give the expression from Equation 3.18 as two expres-
sions, A and B, as shown below
( ) ( )
𝜕x d 𝜕x d 𝜕x
ẍ = ẋ − ẋ . (3.20)
𝜕q1 dt 𝜕q1 dt 𝜕q1
⏟⏞⏞⏞⏞⏟⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏟⏞⏞⏞⏟
A B

Consider now these two expressions.

1. Expression A
Since x = x(q1 , q2 ), it is true that
𝜕x 𝜕x
ẋ = q̇ 1 + q̇ . (3.21)
𝜕q1 𝜕q2 2
Differentiating this expression with respect to q̇ 1 yields
𝜕 ẋ 𝜕x
= (3.22)
𝜕 q̇ 1 𝜕q1
and, therefore,
( ) ( )
d 𝜕x d 𝜕 ẋ
̇x = ̇
x . (3.23)
dt 𝜕q1 dt 𝜕 q̇ 1
2. Expression B
The order of differentiation in term B can be changed to show that
( ) ( )
d 𝜕x 𝜕 dx 𝜕 ẋ
= = . (3.24)
dt 𝜕q1 𝜕q1 dt 𝜕q1
Substituting the new expressions for A and B into Equation 3.19 gives
( )
𝜕x d 𝜕 ẋ 𝜕 ẋ
ẍ = ẋ − ẋ (3.25)
𝜕q1 dt 𝜕 q̇1 𝜕q1
where the right hand side now depends on ẋ only whereas in Equation 3.19 both x and ẋ
appeared on the right hand side.

@Seismicisolation
@Seismicisolation
 3.2 Lagrange’s Equation for a Single Particle 61

We now define a scalar function 𝜓 = ẋ 2 ∕2 and take its partial derivatives with respect to
q̇ 1 and q1 to find
𝜕𝜓 ( ̇ )( ̇ )
x 𝜕x 𝜕 ẋ
=2 = ẋ
̇
𝜕 q1 2 𝜕q̇ 𝜕 q̇ 1
( ) ( 1)
𝜕𝜓 ẋ 𝜕 ẋ 𝜕 ẋ
=2 = ẋ . (3.26)
𝜕q1 2 𝜕q1 𝜕q1
The terms derived in Equation 3.26 from the scalar function 𝜓 are exactly those that
appear on the right hand side of Equation 3.25 so that we can make a substitution and
write
[ ]
𝜕x d 𝜕(ẋ 2 ∕2) 𝜕(ẋ 2 ∕2)
ẍ = − . (3.27)
𝜕q1 dt 𝜕 q̇1 𝜕q1
The term we were actually concerned with in Equation 3.17 was that shown in
Equation 3.27 multiplied by the particle mass m. We can multiply Equation 3.27 by the
scalar m and factor it inside the derivatives since it is constant. The result is
[ ]
𝜕x d 𝜕(mẋ 2 ∕2) 𝜕(mẋ 2 ∕2)
m̈x = − . (3.28)
𝜕q1 dt 𝜕 q̇1 𝜕q1
𝜕y 𝜕z
There are two similar terms in Equation 3.17. These are m̈y and mż . We could
𝜕q1 𝜕q1
𝜕x
repeat the treatment given to m̈x on each of these and the result would be
𝜕q1
[ ]
𝜕y d 𝜕(mẏ 2 ∕2) 𝜕(mẏ 2 ∕2)
m̈y = −
𝜕q1 dt 𝜕 q̇1 𝜕q1
[ ]
𝜕z d 𝜕(mż ∕2)
2 𝜕(mż 2 ∕2)
mż = − . (3.29)
𝜕q1 dt 𝜕 q̇1 𝜕q1
Substituting Equations 3.28 and 3.29 into Equation 3.17 yields
( { [ ]} { })
d 𝜕 1 𝜕 1
𝛿Wq1 = m(ẋ 2 + ẏ 2 + ż 2 ) − m(ẋ 2 + ẏ 2 + ż 2 ) 𝛿q1
dt 𝜕 q̇ 1 2 𝜕q1 2
( )
𝜕x 𝜕y 𝜕z
= Fx + Fy + Fz 𝛿q1 . (3.30)
𝜕q1 𝜕q1 𝜕q1
We can recognize the kinetic energy of the particle appearing twice in Equation 3.30.
Since the absolute velocity of the particle is

𝑣⃗m = x⃗ ̇ 𝚥 + ż k⃗
̇ 𝚤 + y⃗ (3.31)

then the kinetic energy of the particle is

1
T= m(𝑣⃗ ⋅ 𝑣)
⃗ (3.32)
2
or,
1
T= m(ẋ 2 + ẏ 2 + ż 2 ) (3.33)
2
which is the expression seen in Equation 3.30.

@Seismicisolation
@Seismicisolation
62 3 Lagrange’s Equations of Motion

We can therefore write

( )
d 𝜕T 𝜕T 𝜕x 𝜕y 𝜕z
− = Fx + Fy + Fz . (3.34)
dt 𝜕 q̇ 1 𝜕q1 𝜕q1 𝜕q1 𝜕q1
The paragraph following Equation 3.16, said “since q1 and q2 are independently variable,
we may consider the case where 𝛿q2 = 0 and 𝛿q1 ≠ 0. In this case, Equation (3.16) reduces
to the work done during an arbitrary variation of q1 with q2 being held constant. In more
general systems, we think of varying one of the generalized coordinates while holding all others
constant.”
Therefore, we consider now the case where 𝛿q1 = 0 and 𝛿q2 ≠ 0. Clearly, this case would
follow all of the same steps as the previous one and we would find Lagrange’s equation for
generalized coordinate q2 to be
( )
d 𝜕T 𝜕T 𝜕x 𝜕y 𝜕z
− = Fx + Fy + Fz . (3.35)
dt 𝜕 q̇ 2 𝜕q2 𝜕q2 𝜕q2 𝜕q2

3.3 Generalized Forces

We define the right hand side of Lagrange’s equation for generalized coordinate qr to be the
generalized force, Qqr , where
𝜕x 𝜕y 𝜕z
Qqr = Fx + Fy + Fz . (3.36)
𝜕qr 𝜕qr 𝜕qr
Lagrange’s equation for the generalized coordinate qr can therefore be written as
( )
d 𝜕T 𝜕T
− = Qqr . (3.37)
dt 𝜕 q̇ r 𝜕qr
We go on later to show that some common forces can be included in Lagrange’s equation
through the use of potential energy but there will often be applied forces that can only enter
the equations of motion as generalized forces. These forces are the most difficult for the
analyst to handle. In this section, the two most common alternative methods for finding
the generalized forces are introduced.
1. The formal method
The most formal approach, and one that always works, starts with the vector expression
of the absolute position of the point of application of the force
p⃗ F = x⃗𝚤 + y⃗𝚥 + zk⃗
and then recognizes that the generalized force in Equation 3.36 can be written as
𝜕⃗pF
Qqr = F⃗ ⋅ . (3.38)
𝜕qr
Equation 3.38 can be written for each of N applied forces and the resulting scalar gen-
eralized forces can be added together to give the total generalized force for generalized
coordinate qr as
∑
N
𝜕⃗pFi
Qqr = F⃗ i ⋅ . (3.39)
i=1
𝜕qr

@Seismicisolation
@Seismicisolation
 3.3 Generalized Forces 63

2. The intuitive approach

Let there be n generalized coordinates specifying the position of a force acting on a
dynamic system in Cartesian coordinates. The force will be acting at the point (x, y, z)
where the coordinates x, y, and z are functions of the generalized coordinates q1 through
qn and of time, t, as follows
x = x(q1 , q2 , … , qr , … , qn , t)
y = y(q1 , q2 , … , qr , … , qn , t)
z = z(q1 , q2 , … , qr , … , qn , t). (3.40)

Variations in the position of the force as the generalized coordinates are varied while
time is held constant can be written as
𝜕x 𝜕x 𝜕x 𝜕x
𝛿x = 𝛿q + 𝛿q + … 𝛿q + … 𝛿q
𝜕q1 1 𝜕q2 2 𝜕qr r 𝜕qn n
𝜕y 𝜕y 𝜕y 𝜕y
𝛿y = 𝛿q + 𝛿q + … 𝛿q + … 𝛿q
𝜕q1 1 𝜕q2 2 𝜕qr r 𝜕qn n
𝜕z 𝜕z 𝜕z 𝜕z
𝛿z = 𝛿q + 𝛿q + … 𝛿q + … 𝛿q . (3.41)
𝜕q1 1 𝜕q2 2 𝜕qr r 𝜕qn n
If we are trying to find the generalized force corresponding to only one of the generalized
coordinates, say qr , we rewrite Equation 3.41 with 𝛿qi = 0; i = 1, n; i ≠ r and 𝛿qr ≠ 0,
giving
𝜕x
𝛿x = 𝛿q
𝜕qr r
𝜕y
𝛿y = 𝛿q
𝜕qr r
𝜕z
𝛿z = 𝛿q . (3.42)
𝜕qr r
Now consider Equation 3.36 with each side multiplied by 𝛿qr
𝜕x 𝜕y 𝜕z
Qqr 𝛿qr = Fx 𝛿q + Fy 𝛿q + Fz 𝛿q . (3.43)
𝜕qr r 𝜕qr r 𝜕qr r
Clearly, the terms from Equation 3.42 can be substituted into the right hand side of
Equation 3.43 to yield,

Qqr 𝛿qr = Fx 𝛿x + Fy 𝛿y + Fz 𝛿z. (3.44)

The right hand side of Equation 3.44 can be seen to be the work done by the applied
force as its position varies due to changes in the generalized coordinate qr while all other
generalized coordinates and time are held constant.
Using the intuitive approach to finding generalized forces, the analyst will consider, in
sequence, the variation of individual generalized coordinates and will write expressions
for the total work done during each variation. The generalized force associated with each
generalized coordinate will be the work done during the variation of that coordinate
divided by the variation in the coordinate. That is,

Qqr = 𝛿Wqr ∕𝛿qr . (3.45)

@Seismicisolation
@Seismicisolation
64 3 Lagrange’s Equations of Motion

3.4 Generalized Forces as Derivatives of Potential Energy

The concept of potential energy is well known to engineers and scientists so it is unneces-
sary to define it in detail here. We simply state that, if the forces Fx , Fy , and Fz are conser-
vative, then they may be written as partial derivatives of a scalar potential energy function,
U. That is
𝜕U
Fx = −
𝜕x
𝜕U
Fy = −
𝜕y
𝜕U
Fz = − . (3.46)
𝜕z
Note that the negative signs in Equation 3.46 indicate that we are interested in the forces
applied to the body by the field.
Figure 3.2 can be used as an example. In this figure, a mass, m, has been raised a distance
x vertically above the datum being used for changes in potential energy4 . We know that the
force of gravity on the body is mg, acting vertically downward. If we define the gravitational
𝜕U
potential energy to be U = mgx, then Fx = − = −mg.
𝜕x
Another example of a generalized force that can be included in Lagrange’s equation
through the potential energy is the force in a linear spring that obeys Hooke’s Law5 , F = kx
(Figure 3.3). The potential energy in a spring is well known to be
1 2
U= kx
2
where x is the displacement from the unstretched length of the spring. We can then write
𝜕U
Fx = − = −kx.
𝜕x

m Figure 3.2 Gravitational force acting on a mass.

x
g
mg

Datum

x Figure 3.3 Spring force acting on a

mass.
kx kx

4 We rarely ask for the “absolute” gravitational potential energy but only for the changes from a reference
height (“the datum”), which can be chosen arbitrarily.
5 Robert Hooke (1635–1703), an English scientist, discovered the linear relationship between the force
acting on, and the resulting deflection of, elastic bodies.

@Seismicisolation
@Seismicisolation
 3.5 Dampers – Rayleigh’s Dissipation Function 65

Substituting Equation 3.46 into Equation 3.36 gives

𝜕U 𝜕x 𝜕U 𝜕y 𝜕U 𝜕z 𝜕U
Qqr = − − − =− . (3.47)
𝜕x 𝜕qr 𝜕y 𝜕qr 𝜕z 𝜕qr 𝜕qr
Therefore, for conservative forces, where the force may be derived from differentiating
the potential energy, we may use the result of Equation 3.47. Note that the expression in
Equation 3.47 appears on the right hand side of Lagrange’s equation. It is usually taken to
the left hand side with a change of sign to give
( )
d 𝜕T 𝜕T 𝜕U
− + = Qqr . (3.48)
dt 𝜕 q̇ r 𝜕qr 𝜕qr
Lagrange’s equation is most often used in the form given in Equation 3.48. That is,
( )
d 𝜕T 𝜕T 𝜕U
− + = Qqr (3.49)
dt 𝜕 q̇ r 𝜕qr 𝜕qr
where the potential energy, U, contains all gravitational and elastic effects. All other applied
forces are accounted for in the generalized force, Qqr .
The equation is written once for each of the generalized coordinates chosen to describe
the system.
The only assumptions made in the derivation of Lagrange’s equation were,

1. The generalized coordinates are independent. That is, each of the generalized coordi-
nates can be arbitrarily varied while all others are held constant. Essentially, this means
that the chosen degrees of freedom (i.e. generalized coordinates) should not be related
by constraints.
2. The internal constraints are “smooth” and do no work during the variation of a general-
ized coordinate.

Some authors define a function called the Lagrangian, L,

L=T−U

so that Lagrange’s equation may be written as

( )
d 𝜕L 𝜕L
− = Qqr
dt 𝜕 q̇ r 𝜕qr
where the assumption that potential energy
( is)never a function of velocity has been used. If
d 𝜕U
U actually depended on q̇ r , a term − would appear.
dt 𝜕 q̇ r

3.5 Dampers – Rayleigh’s Dissipation Function

Devices called “dampers” are common in mechanical systems. These are elements that dis-
sipate energy and they are modeled as producing forces that are proportional to their rate
of change of length. The rate of change of length is the relative velocity across the damper.
“Proportional” implies linearity and a force proportional to speed implies laminar, viscous
flow. As a result, these elements are often referred to as “linear viscous dampers”.

@Seismicisolation
@Seismicisolation
66 3 Lagrange’s Equations of Motion

v Figure 3.4 A linear viscous damper.

m
c

v
m
c Fd

Figure 3.4 shows a system where a body is attached to ground by a damper. The body is
moving to the right with speed 𝑣 and the damping coefficient (constant of proportionality)
is c. The physical connection of the damper to both the ground and the body dictates that
the rate of change of length of the damper is equal to the speed 𝑣. The force in the damper
will therefore be Fd = c𝑣. The direction of the force will be such that it causes the damper
to increase in length as shown in the lower part of Figure 3.4. By Newton’s third law, the
force on the body must be equal and opposite to the force acting on the damper. The force Fd
therefore acts to the left on the body. In other words, the damping force opposes the velocity
of the body.
Consider now the more general case of a particle as described in Section 3.2 where the
velocity of the body is given by Equation 3.31 as
̇ 𝚤 + y⃗
𝑣⃗m = x⃗ ⃗
̇ 𝚥 + ż k. (3.50)
Given this velocity, the force that the damper applies to the particle will be
F⃗ d = −c𝑣⃗m = −cx⃗
̇ 𝚤 − cy⃗ ⃗
̇ 𝚥 − cż k. (3.51)
The components of F⃗ d can be substituted into Equation 3.36 to get the following expres-
sion for the generalized force arising from the damper
𝜕x 𝜕y 𝜕z
Qdqr = −cẋ − cẏ − cż (3.52)
𝜕qr 𝜕qr 𝜕qr
where, using the result from Equation 3.22, we can write
𝜕x 𝜕 ẋ 𝜕y 𝜕 ẏ 𝜕z 𝜕 ż
= ; = ; = (3.53)
𝜕qr 𝜕 q̇ r 𝜕qr 𝜕 q̇ r 𝜕qr 𝜕 q̇ r
which can be substituted into Equation 3.52 to yield the following expression for the gen-
eralized force
( )
𝜕 ẋ 𝜕 ẏ 𝜕 ż
Qqr = −c ẋ
d
+ ẏ + ż . (3.54)
𝜕 q̇ r 𝜕 q̇ r 𝜕 q̇ r
The generalized force, as expressed in Equation 3.54, can be derived from a scalar function
called Rayleigh’s dissipation function which is defined as
1 1
ℜ= c(𝑣⃗m ⋅ 𝑣⃗m ) = c(ẋ 2 + ẏ 2 + ż 2 ). (3.55)
2 2

@Seismicisolation
@Seismicisolation
 3.6 Kinetic Energy of a Free Rigid Body 67

A simple differentiation with respect to q̇ r yields

( )
𝜕ℜ 𝜕 ẋ 𝜕 ẏ 𝜕 ż
= c ẋ + ẏ + ż = −Qdqr . (3.56)
𝜕 q̇ r 𝜕 q̇ r 𝜕 q̇ r 𝜕 q̇ r
Lagrange’s equation can then be written as
( )
d 𝜕T 𝜕T 𝜕U 𝜕ℜ
− + =− + Qqr (3.57)
dt 𝜕 q̇ r 𝜕qr 𝜕qr 𝜕 q̇ r
where Qqr now represents the generalized force corresponding to all externally applied
forces that are neither conservative nor linear viscous in nature. Finally, we can transfer
the Rayleigh dissipation term to the left hand side and write Lagrange’s equation with dis-
sipation as
( )
d 𝜕T 𝜕T 𝜕U 𝜕ℜ
− + + = Qqr . (3.58)
dt 𝜕 q̇ r 𝜕qr 𝜕qr 𝜕 q̇ r

3.6 Kinetic Energy of a Free Rigid Body

The extension of Lagrange’s equation to rigid bodies follows logically from its application
to single particles. We again consider a rigid body to be composed of individual particles
that maintain constant inter-particle distances. The rigid body constraint means that each
rigid body will have only a limited number of generalized coordinates6 , far fewer than the
number of particles in the body.
The analysis requires that the kinetic and potential energy for the rigid body be available.
The scalar nature of Lagrange’s equations and of the energy of a body allows the total kinetic
energy of a rigid body to be found simply by adding the contributions of all the particles.
Figure 3.5 shows the same three-dimensional rigid body that was first introduced in
Figure 2.3. The body has a general reference point O with known absolute velocity 𝑣⃗O =
⃗ Vectors are expressed in a reference frame having unit vectors (⃗𝚤, ⃗𝚥, k).
𝑣Ox ⃗𝚤 + 𝑣Oy ⃗𝚥 + 𝑣Oz k. ⃗
The reference frame has angular velocity 𝜔 ⃗ The particle to be consid-
⃗ = 𝜔 ⃗𝚤 + 𝜔 ⃗𝚥 + 𝜔 k.
x y z
ered (particle i) is located with respect to the reference point by the position vector p⃗ i∕O =
x ⃗𝚤 + y ⃗𝚥 + z k⃗ and has mass m .
i i i i
The velocity of particle i can be expressed as
⃗ × p⃗ i∕O .
𝑣⃗i = 𝑣⃗O + 𝜔 (3.59)
⃗ where the components of the velocity of i are
This leads to 𝑣⃗i = 𝑣x ⃗𝚤 + 𝑣y⃗𝚥 + 𝑣z k,
𝑣x = 𝑣Ox + 𝜔y zi − 𝜔z yi
𝑣y = 𝑣Oy + 𝜔z xi − 𝜔x zi
𝑣z = 𝑣Oz + 𝜔x yi − 𝜔y xi . (3.60)

6 The maximum is six for a free rigid body that moves without constraint. Three translational coordinates
are required to locate the center of mass of the body in space and three angular coordinates are needed to
specify its orientation. Any constraint on the motion of the rigid body (e.g. a pin joint) will reduce the
number of generalized coordinates.

@Seismicisolation
@Seismicisolation
68 3 Lagrange’s Equations of Motion

j Figure 3.5 A single particle in a

rigid body.

ω
i
k
mi α

pi/O

aO

vO

The kinetic energy contribution of particle i is then

1
Ti = m (𝑣⃗ ⋅ 𝑣⃗ )
2 i i i
1
= mi (𝑣2x + 𝑣2y + 𝑣2z ) (3.61)
2
which, after substituting the expressions for the velocity components and reorganizing, can
be written as
1
Ti = mi [(𝑣2O + 𝑣2O + 𝑣2O ) + 𝜔2x (y2i + zi2 ) + 𝜔2y (xi2 + zi2 ) + 𝜔2z (xi2 + y2i )
2 x y z

− 2𝜔x 𝜔y xi yi − 2𝜔x 𝜔z xi zi − 2𝜔y 𝜔z yi zi

+ 2𝑣Ox (𝜔y zi − 𝜔z yi ) + 2𝑣Oy (𝜔z xi − 𝜔x zi ) + 2𝑣Oz (𝜔x yi − 𝜔y xi )]. (3.62)

Note that the expression (𝑣2O + 𝑣2O + 𝑣2O ) can be replaced by 𝑣2O where 𝑣O is the magni-
x y z
tude of the velocity of point O (i.e. the speed of point O).
The total kinetic energy of the rigid body is then
∑ 1 ∑
T= Ti = 𝑣2O mi
i
2 i
1 ∑ 1 ∑ 1 ∑
+ 𝜔2x mi (y2i + zi2 ) + 𝜔2y mi (xi2 + zi2 ) + 𝜔2z mi (xi2 + y2i )
2 i
2 i
2 i
∑ ∑ ∑
− 𝜔x 𝜔y mi xi yi − 𝜔x 𝜔z mi xi zi − 𝜔y 𝜔z mi yi zi
i i i
( ) ( )
∑ ∑ ∑ ∑
+ 𝑣Ox 𝜔y mi zi − 𝜔z mi yi + 𝑣Oy 𝜔z mi xi − 𝜔x m i zi
i i i i
( )
∑ ∑
+ 𝑣Oz 𝜔x mi yi − 𝜔y mi xi . (3.63)
i i

The summation terms in Equation 3.63 have already been considered in Section 2.9. They
are

@Seismicisolation
@Seismicisolation
 3.6 Kinetic Energy of a Free Rigid Body 69

∑
1. mi = M
∑
i
2. mi (y2i + zi2 ) = IxxO
∑
i
3. mi (xi2 + zi2 ) = IyyO
∑
i
4. mi (xi2 + y2i ) = IzzO
∑
i
5. mi xi yi = IxyO
∑
i
6. mi xi zi = IxzO
∑
i
7. mi yi zi = IyzO
∑
i
8. mi xi = Mx
∑
i
9. mi yi = My
∑
i
10. mi zi = Mz.
i

Making these substitutions into Equation 3.63 gives a general expression for the kinetic
energy of a rigid body, as follows
1
T= M𝑣2O
2
1
+ [IxxO 𝜔2x + IyyO 𝜔2y + IzzO 𝜔2z − 2IxyO 𝜔x 𝜔y − 2IxzO 𝜔x 𝜔z − 2IyzO 𝜔y 𝜔z ]
2
+ M[𝑣Ox (𝜔y z − 𝜔z y) + 𝑣Oy (𝜔z x − 𝜔x z) + 𝑣Oz (𝜔x y − 𝜔y x)]. (3.64)

There are cases of Equation 3.64 that deserve special attention because, since they cause
many of the terms to become zero, they are used almost exclusively in dynamic analysis.
These are:

1. The case where O is a fixed point.

In this case, 𝑣Ox = 𝑣Oy = 𝑣Oz = 0, and the kinetic energy becomes

1
T= [I 𝜔2 + IyyO 𝜔2y + IzzO 𝜔2z − 2IxyO 𝜔x 𝜔y − 2IxzO 𝜔x 𝜔z − 2IyzO 𝜔y 𝜔z ]. (3.65)
2 xxO x
2. The case where O is the center of mass, G.
In this case, x = y = z = 0 and 𝑣⃗O = 𝑣⃗G . The kinetic energy is
1 1
T= M𝑣2G + [IxxG 𝜔2x + IyyG 𝜔2y + IzzG 𝜔2z
2 2
− 2IxyG 𝜔x 𝜔y − 2IxzG 𝜔x 𝜔z − 2IyzG 𝜔y 𝜔z ]. (3.66)

While the general expression for the kinetic energy is relatively complex, it can be easily
remembered for the case where G is the reference point if we simply change mathemati-
cal notation. Rather than writing the scalar components of the translational and angular
velocities of the body with their respective unit vectors, we can write the velocities as the

@Seismicisolation
@Seismicisolation
70 3 Lagrange’s Equations of Motion

following column vectors.

⎧ 𝑣Gx ⎫
⎪ ⎪
{𝑣G } = ⎨ 𝑣Gy ⎬ (3.67)
⎪ ⎪
⎩ 𝑣Gz ⎭
⎧ 𝜔x ⎫
⎪ ⎪
{𝜔} = ⎨ 𝜔y ⎬ . (3.68)
⎪ ⎪
⎩ 𝜔z ⎭
Then, the kinetic energy becomes simply
1 1
T= M{𝑣G }T {𝑣G } + {𝜔}T [IG ]{𝜔} (3.69)
2 2
where [IG ] is the inertia tensor defined in Equations 2.61 and 2.62.
In the case where O is a fixed point, the kinetic energy is simply
1
T= {𝜔}T [IO ]{𝜔}. (3.70)
2

3.7 A Two Dimensional Example using Lagrange’s Equation

Figure 3.6 shows a rigid body moving in a plane. Point B on the body is attached to a fixed
point A by a massless rigid rod7 of constant length, r. The body has mass M and moment
⃗ are fixed in the body. The
of inertia about point B of IzzB = I. The unit vectors shown, (⃗𝚤, ⃗𝚥, k),
center of mass, G, is located relative to B by the position vector p⃗ G∕B = x0⃗𝚤 + y0⃗𝚥.
The problem is to write the equations of motion of the system in terms of the generalized
angular coordinates 𝜃 and 𝜙, each of which measures an angle relative to a fixed direction
in space8 .

3.7.1 The Kinetic Energy

We note that this is a two-dimensional problem and that the only possible angular velocities
are due to rotations about the k⃗ direction. The angular velocity of the coordinate system is
therefore

𝜔 ⃗
⃗ = 𝜙̇ k. (3.71)

Since the body does not have a fixed point (A is not on the body and B is moving), we
can either find the velocity of the center of mass (𝑣⃗G ) and the moment of inertia about G
(IzzG ) and use Equation 3.69 or we can work with the moving reference point B and use
Equation 3.64. We choose to work with B in this case.

7 This example will be used later and we will consider cases where the rod is inverted and supports
the mass from below.
8 𝜃 and 𝜙 are in fact generalized coordinates because it is possible to vary 𝜃 without changing 𝜙 and vice
versa.

@Seismicisolation
@Seismicisolation
 3.7 A Two Dimensional Example using Lagrange’s Equation 71

Figure 3.6 2D rigid body example.

A

θ
j
r
k x0
G
B
y0
ϕ
i
Mg

We must first determine the velocity of point B. We start by writing the position of B with
respect to A (i.e p⃗ B∕A ). Since A has no velocity, the absolute velocity of B can be found simply
by differentiating p⃗ B∕A
p⃗ B∕A = r cos(𝜙 − 𝜃)⃗𝚤 − r sin(𝜙 − 𝜃)⃗𝚥. (3.72)
The velocity of B is then
d
𝑣⃗B = p⃗ = p⃗̇ B∕A + 𝜔
⃗ × p⃗ B∕A . (3.73)
dt B∕A
Consideration of Equation 3.72 shows that there is a rate of change of magnitude of p⃗ B∕A
and, since the unit vectors are rotating, there will also be a rate of change of direction. The
result, after differentiating and simplifying, is
𝑣⃗B = r 𝜃̇ sin(𝜙 − 𝜃)⃗𝚤 + r 𝜃̇ cos(𝜙 − 𝜃)⃗𝚥. (3.74)
For this 2D example, Equation 3.64 simplifies to
1 1
T= M𝑣2B + IzzB 𝜙̇ 2 − M𝑣Bx 𝜙y
̇ 0 + M𝑣B 𝜙x ̇ 0. (3.75)
2 2 y

Substituting for the velocity terms, setting IzzB = I, and simplifying yields
1 2 ̇2 1 ̇2
T= Mr 𝜃 + I 𝜙 + Mr 𝜃̇ 𝜙[x
̇ 0 cos(𝜙 − 𝜃) − y0 sin(𝜙 − 𝜃)]. (3.76)
2 2

3.7.2 The Potential Energy

The next step is to find the potential energy of the system. In this example, there are no
springs so all potential energy arises from gravitational effects. Since we are only interested
in changes in potential energy from some arbitrary datum, we are free to choose the datum
to be at any convenient height. Figure 3.7 shows a suitable datum position corresponding
to 𝜃 = 𝜙 = 0. As the system moves and 𝜃 and 𝜙 assume non-zero values, the height of the
center of mass changes as shown in the figure. The vertical distance from A to G in the two
configurations can be used to calculate h, the increase in height of G from the datum.

@Seismicisolation
@Seismicisolation
72 3 Lagrange’s Equations of Motion

Datum
j
A A
θ
r k x0
r
G
B
B y0
h ϕ
x0
i
Mg
G
y0

Figure 3.7 2D rigid body potential energy.

The vertical distance from A to G in the datum configuration is (r + x0 ). In the displaced

configuration, it is (r cos 𝜃 + x0 cos 𝜙 − y0 sin 𝜙). As a result,

h = (r + x0 ) − (r cos 𝜃 + x0 cos 𝜙 − y0 sin 𝜙)

or, after grouping terms,

h = r(1 − cos 𝜃) + x0 (1 − cos 𝜙) + y0 sin 𝜙

and, the change in potential energy from the datum to the displaced configuration is

U = Mgh = Mg[r(1 − cos 𝜃) + x0 (1 − cos 𝜙) + y0 sin 𝜙]. (3.77)

3.7.3 The 𝜽 Equation

We proceed to write each of the terms required in Lagrange’s equation for the generalized
coordinate 𝜃.
𝜕T
= Mr 2 𝜃̇ − Mr 𝜙ẏ 0 sin(𝜙 − 𝜃) + Mr 𝜙ẋ 0 cos(𝜙 − 𝜃) (3.78)
𝜕 𝜃̇
( )
d 𝜕T
= Mr 2 𝜃̈ − Mr 𝜙y ̇ 𝜙̇ − 𝜃)y
̈ 0 sin(𝜙 − 𝜃) − Mr 𝜙( ̇ 0 cos(𝜙 − 𝜃)
dt 𝜕 𝜃̇
− Mr 𝜙( ̇ 𝜙̇ − 𝜃)x
̇ 0 sin(𝜙 − 𝜃) + Mr 𝜙x
̈ 0 cos(𝜙 − 𝜃) (3.79)

𝜕T ̇ 0 cos(𝜙 − 𝜃) + Mr 𝜃̇ 𝜙x
= Mr 𝜃̇ 𝜙y ̇ 0 sin(𝜙 − 𝜃) (3.80)
𝜕𝜃
𝜕U
= Mgr sin 𝜃. (3.81)
𝜕𝜃

@Seismicisolation
@Seismicisolation
 3.8 Standard Form of the Equations of Motion 73

These are then substituted into Equation 3.49 to arrive at the equation of motion for the
generalized coordinate 𝜃, which simplifies to
Mr 2 𝜃̈ + Mr 𝜙[x
̈ 0 cos(𝜙 − 𝜃) − y0 sin(𝜙 − 𝜃)]
− Mr 𝜙̇ 2 [y0 cos(𝜙 − 𝜃) + x0 sin(𝜙 − 𝜃)] + Mgr sin 𝜃 = 0. (3.82)

3.7.4 The 𝝓 Equation

The required terms for the generalized coordinate 𝜙 are
𝜕T
= I 𝜙̇ − Mr 𝜃ẏ 0 sin(𝜙 − 𝜃) + Mr 𝜃x
̇ 0 cos(𝜙 − 𝜃) (3.83)
𝜕 𝜙̇
( )
d 𝜕T
= I 𝜙̈ − Mr 𝜃y
̈ 0 sin(𝜙 − 𝜃) + Mr 𝜃x
̈ 0 cos(𝜙 − 𝜃)
dt 𝜕 𝜙̇
− Mr 𝜃(̇ 𝜙̇ − 𝜃)y
̇ 0 cos(𝜙 − 𝜃) − Mr 𝜃(̇ 𝜙̇ − 𝜃)x
̇ 0 sin(𝜙 − 𝜃) (3.84)

𝜕T
= −Mr 𝜃̇ 𝜙y
̇ 0 cos(𝜙 − 𝜃) − Mr 𝜃̇ 𝜙x
̇ 0 sin(𝜙 − 𝜃) (3.85)
𝜕𝜙
𝜕U
= Mgx0 sin 𝜙 + Mgy0 cos 𝜙. (3.86)
𝜕𝜙
The equation of motion for the generalized coordinate 𝜙 becomes
I 𝜙̈ − Mr 𝜃[y
̈ 0 sin(𝜙 − 𝜃) − x0 cos(𝜙 − 𝜃)]
+ Mr 𝜃̇ 2 [y0 cos(𝜙 − 𝜃) + x0 sin(𝜙 − 𝜃)] + Mg[y0 cos 𝜙 + x0 sin 𝜙] = 0. (3.87)

3.8 Standard Form of the Equations of Motion

Equations of motion for systems with more than one degree of freedom are often written in
a standard form as
̇ t)]{q}
[M(q, q, ̈ = {f (q, q,
̇ t)} (3.88)
where
{q} is the vector of degrees of freedom.
̇ and {q}
{q} ̈ are the vectors of first and second derivatives of the degrees of freedom.
[M(q, q,̇ t)] is the mass matrix with elements that may depend upon the degrees of free-
dom, their first derivatives, and time.
̇ t)} is a forcing vector containing all of the remaining terms in the equations of
{f (q, q,
motion.
The point to note is that accelerations (i.e. second derivatives of the degrees of freedom)
always appear linearly in the equations of motion so that it is always possible to write them
in the standard form of Equation 3.88.
Applying this to the equations of motion derived in Section 3.7 yields
[ ]{ } { }
Mr 2 MrP 𝜃̈ Mr 𝜙̇ 2 Q − Mgr sin 𝜃
= (3.89)
MrP I 𝜙̈ −Mr 𝜃̇ 2 Q − Mg[y0 cos 𝜙 + x0 sin 𝜙]

@Seismicisolation
@Seismicisolation
74 3 Lagrange’s Equations of Motion

where
P = x0 cos(𝜙 − 𝜃) − y0 sin(𝜙 − 𝜃)
and
Q = y0 cos(𝜙 − 𝜃) + x0 sin(𝜙 − 𝜃).
Once the equations of motion have been formulated, the analyst must decide what to do
with them. That is, the equations have been derived in order to extract some information
about the system being studied. Methods of extracting the required information will be
considered in the following chapters. Discussions of equilibrium solutions, stability and
mode shapes, frequency response, and time response are included.

Exercises
Descriptions of the systems referred to in the exercises are contained in Appendix A.

3.1 Show that the kinetic energy and potential energy expressions for system 1 are
1 1
T = (m1 + m2 )d21 𝜃̇ 12 + m2 d22 (𝜃̇ 1 + 𝜃̇ 2 )2 + m2 d1 d2 𝜃̇ 1 (𝜃̇ 1 + 𝜃̇ 2 ) cos 𝜃2
2 2
and
U = (m1 + m2 )gd1 sin 𝜃1 + m2 gd2 sin(𝜃1 + 𝜃2 ).

3.2 Use Lagrange’s equation to show that the equations of motion for system 1 are

⎡ (m1 + m2 )d21 + m2 d2 (d2 + 2d1 cos 𝜃2 ) m2 d2 (d2 + d1 cos 𝜃2 ) ⎤ ⎧ 𝜃̈1 ⎫

⎢ ⎥⎪⎨
⎪
⎬
⎢ ⎥
⎣ m2 d2 (d2 + d1 cos 𝜃2 ) m2 d22 ⎦⎪ 𝜃̈2 ⎪
⎩ ⎭
{ }
m2 d1 d2 𝜃̇ 2 (2𝜃̇ 1 + 𝜃̇ 2 ) sin 𝜃2 − (m1 + m2 )gd1 cos 𝜃1 − m2 gd2 cos(𝜃1 + 𝜃2 )
= .
−m2 d1 d2 𝜃̇ 12 sin 𝜃2 − m2 gd2 cos(𝜃1 + 𝜃2 )
Show that these equations are equivalent to those found in Exercise 2.1.

3.3 Use Lagrange’s equation to show that the equations of motion for system 5 are
[ 3 3 ]{ }
2
m 2
mr ẍ
3
mr 9
mr 2 + m(d − x)2 𝜃̈
2 2
{ }
mg sin 𝜃 − m(d − x)𝜃̇ 2
= .
Fd cos 𝜃 + mgr sin 𝜃 − mg(d − x) cos 𝜃 + 2m(d − x)ẋ 𝜃̇
Show that these equations are equivalent to those found in Exercise 2.4.

3.4 Use Lagrange’s equation to show that the equation of motion for system 6 is
mr 2 𝜃̈ − m𝜔2 r(d − r cos 𝜃) sin 𝜃 + mgr cos 𝜃 = 0.

@Seismicisolation
@Seismicisolation
 Exercises 75

3.5 Use Lagrange’s equation to show that the equation of motion for system 7 is
( )
4
2mL2 + 12sin2 𝜃 𝜃̈ + 24mL2 𝜃̇ 2 sin 𝜃 cos 𝜃 + 4kL2 sin 𝜃 cos 𝜃 = 4FL sin 𝜃.
3

3.6 Use Lagrange’s equation to show that the equation of motion for system 9 is
1
mR2 𝜃̈ + kh2 sin 𝜃 cos 𝜃 = FR sin 𝜃.
3

3.7 Use Lagrange’s equation to show that the equations of motion for system 10 are
[ ]{ } { }
m 0 ẍ −2kx + kd sin 𝜃
= .
0 md2 𝜃̈ dF(t) + kxd cos 𝜃 − kd2 sin 𝜃 cos 𝜃

3.8 Use Lagrange’s equation to show that the equation of motion for system 14 is
11 3
m(R + r)2 𝜃̈ − mg(R + r) sin 𝜃 = −F(R + r) sin 𝜃.
6 2

3.9 Use Lagrange’s equation to show that the equations of motion for system 16 are

⎡
4
m𝓁 2 + mr 2 + 2mr𝓁 cos 𝜙 4
m𝓁 2 + mr𝓁 cos 𝜙 ⎤ ⎧ 𝜃̈ ⎫
⎥⎪ ⎪
3 3
⎢
⎢ ⎥⎨ ⎬
⎣
4
m𝓁 2 + mr𝓁 cos 𝜙 4
m𝓁 2
⎦⎪ 𝜙̈ ⎪
3 3 ⎩ ⎭

⎧ 2mr𝓁 𝜃̇ 𝜙̇ sin 𝜙 + mr𝓁 𝜙̇ 2 sin 𝜙 + mgr cos 𝜃 + mg𝓁 cos(𝜃 + 𝜙) ⎫

⎪ ⎪
=⎨ ⎬.
⎪ −mr𝓁 𝜃̇ 2 sin 𝜙 + mg𝓁 cos(𝜃 + 𝜙) ⎪
⎩ ⎭

3.10 Use Lagrange’s equation to show that the equation of motion for system 17 is
mr 2 𝜃̈ − m𝜔2 r 2 sin 𝜃 cos 𝜃 = 0.
Verify this result using Newton’s laws.

3.11 Use Lagrange’s equation to derive the equations of motion for system 23 and show
that they are equivalent to the equations of motion found in Exercise 2.11.

3.12 Use Lagrange’s equation to confirm (see Exercise 2.12) that the equation-of-motion
of the rod in system 12, using the angle 𝜃 as the single degree-of-freedom, is
4 2̈
md 𝜃 + 4kd2 sin 𝜃 cos 𝜃 − mgd cos 𝜃 = 0.
3

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 77

Part II

Simulation: Using the Equations of Motion

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 79

Equilibrium Solutions

Equilibrium solutions of the equations of motion are those where the degrees of freedom
assume values that cause their first and second derivatives to go to zero. Under these condi-
tions, there will be no tendency for the values of the degrees of freedom to change and the
system will be in an equilibrium state.
We will make extensive use of the equilibrium solutions when we consider frequency
response and stability of motions.

4.1 The Simple Pendulum

Consider the simple pendulum1 shown in Figure 4.1. Using the angle 𝜃 as the single degree
of freedom, the equation of motion is,
m𝓁 2 𝜃̈ + mg𝓁 sin 𝜃 = 0. (4.1)
Once started in motion the pendulum will swing about the point of connection to the
ground. In the case of the simple pendulum there is no mechanism for removing energy
from the system as it swings (i.e. no friction or other forces that do work) so the motion,
once started, will persist.
The motion will depend on the way in which it is started. That is, if the pendulum is
rotated to some arbitrary starting angle, 𝜃0 , and released from rest, it will swing through
the position where 𝜃 = 0 and will eventually return to where it started before reversing and
starting the cyclic motion over again. If the pendulum is stopped and returned to 𝜃0 and then
released, not from rest but with an initial velocity, the resulting motion will be different and
the pendulum will pass through 𝜃0 when it returns. The motion will still, however, be cyclic.
Solutions of the equation of motion (Equation 4.1) will predict the results of releasing the
pendulum from any initial angle 𝜃0 with any initial angular velocity. These solutions we
leave for later. The question we ask now is are there initial values of 𝜃 where the pendulum
can be released from rest and remain stationary? These are the equilibrium states.
Consider Equation 4.1 under the conditions that there is an initial angle 𝜃0 and there is no
angular velocity (i.e. 𝜃̇ = 0 so that 𝜃0 does not change with time) and that there is no angular

1 The definition of the, often quoted, simple pendulum is that it has a massless rigid rod supporting a point
mass. The rod is free to swing in a plane about the frictionless point where it is connected to the ground.
The only external force is that due to gravity.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
80 4 Equilibrium Solutions

Figure 4.1 A simple pendulum.

g
θ

acceleration (i.e. 𝜃̈ = 0 so that 𝜃̇ does not change with time and thus there will never be a
change in 𝜃0 ). This is an equilibrium position and Equation 4.1 becomes,
mg𝓁 sin 𝜃0 = 0. (4.2)
Since m and g are never zero, this can only be satisfied by,
sin 𝜃0 = 0
and the total range of 𝜃 is 0 ≤ 𝜃 ≤ 2𝜋. In this range, only 𝜃 = 0 (the pendulum hangs verti-
cally downward) and 𝜃 = 𝜋 (the pendulum stands upright) satisfy the requirements. These
are the two equilibrium states for the pendulum. We will see later that one equilibrium state
is stable and the other is unstable.

4.2 Equilibrium with Two Degrees of Freedom

At the end of Chapter 2, the equations of motion for a two dimensional rigid body with two
degrees of freedom were derived. These equations, taken from Equation 3.89, are repeated
here as Equation 4.3. The system being analyzed is shown in Figure 3.6.
[ ]{ } { }
Mr 2 MrP 𝜃̈ Mr 𝜙̇ 2 Q − Mgr sin 𝜃
= (4.3)
MrP I 𝜙̈ −Mr 𝜃̇ 2 Q − Mg[y0 cos 𝜙 + x0 sin 𝜙]
where,
P = x0 cos(𝜙 − 𝜃) − y0 sin(𝜙 − 𝜃)
and
Q = y0 cos(𝜙 − 𝜃) + x0 sin(𝜙 − 𝜃).
Finding the equilibrium solutions for this problem is completely analogous to finding
the solutions for the simple pendulum. We assume that 𝜃 and 𝜙 have taken up equilibrium
values 𝜃0 and 𝜙0 respectively and that their first and second derivatives are zero. Under
these conditions, Equation 4.3 reduces to,
{ } { }
0 −Mgr sin 𝜃0
= . (4.4)
0 −Mg [y0 cos 𝜙0 + x0 sin 𝜙0 ]
There are four solutions to the set of equations presented as Equation 4.4. First note that
the two equations in the set are not coupled because the first equation contains only 𝜃0 and
the second contains only 𝜙0 . As a result, the two equations can be solved independently2 .

2 Equations are coupled if more than one of the variables describing the motion appears in the same
equation.

@Seismicisolation
@Seismicisolation
 4.3 Equilibrium with Steady Motion 81

G i
⇀
j
B
B
⇀
j
G
⇀
i A A
⇀
A i A

⇀ G
j
B
B
⇀
j
G
⇀
i

Figure 4.2 The equilibrium solutions for the 2D example.

The first equation is satisfied when sin 𝜃0 = 0. The possible range of motion is 0 ≤ 𝜃0 ≤ 2𝜋
and, in this range the only solutions are 𝜃0 = 0 and 𝜃0 = 𝜋.
The second equation is satisfied when
y
tan 𝜙0 = − 0 . (4.5)
x0
There are two angles where this is the case, one in the second quadrant and one in the
fourth.
Figure 4.2 shows the four equilibrium solutions for this problem3 . Notice that all of the
equilibrium solutions correspond to a position where the potential energy has an extreme
value – either a maximum or a minimum. Notice also that some of the equilibrium solutions
are clearly unstable in the sense that the system, if disturbed, will move away from these
positions.

4.3 Equilibrium with Steady Motion

Classical static equilibrium is not the only type of equilibrium solution that exists. There are
cases where several of the degrees of freedom of a multi-degree of freedom system can have
motion that affects the ultimate equilibrium values of the degrees of freedom of interest.
Consider, for example, a system consisting of a large vehicle in which are suspended sev-
eral objects making up the payload. An analyst may write the equations of motion for the
entire system and then decide that the equilibrium values for the degrees of freedom of the
payload are of interest in the case where the vehicle is driving in circles at a constant speed.

3 Notice the importance of having a massless rigid rod rather than a string in the third and fourth
equilibrium positions. The mass cannot stand on a string.

@Seismicisolation
@Seismicisolation
82 4 Equilibrium Solutions

Figure 4.3 An eccentric rotating body.

X m
k

g θ
O

d T

Another example, perhaps more realistic but again dealing with vehicles, is the case
where the steady-state value of the roll angle of a car negotiating a long curve at high speed
is required as part of the suspension design process. The system is certainly not static but
equilibrium values can be found so long as the motion is steady.
Consider, as an example, the system shown in Figure 4.3, which has an eccentric rotating
rigid body (mass = M, moment of inertia about the fixed point O = I, center of mass at G) in
which a slot has been milled to support a small slider (mass= m) that is attached by a spring
(stiffness = k, undeflected length = 𝓁0 ) to the fixed point O. The system has two degrees of
freedom, x and 𝜃. There is an applied torque T acting on the rigid body.
The equations of motion for this system are,
[ ]{ } { }
I + mx2 0 𝜃̈ T − 2mxẋ 𝜃̇ + Mgd cos 𝜃 − mgx cos 𝜃
= . (4.6)
0 m ẍ mx𝜃̇ 2 − mg sin 𝜃 − k(x − 𝓁0 )
We consider two scenarios for establishing equilibrium. The first is in the absence of the
applied torque T and the second is in the case where T is specified through a control scheme
that maintains a constant angular velocity 𝜃̇ = 𝜃̇ 0 .
For T = 0, we let x = x0 , ẋ = ẍ = 0, 𝜃 = 𝜃0 , and 𝜃̇ = 𝜃̈ = 0 to get the two equilibrium
equations,

Mgd cos 𝜃0 − mgx0 cos 𝜃0 = 0 (4.7)

and

−mg sin 𝜃0 − k(x0 − 𝓁0 ) = 0. (4.8)

Equation 4.7 can be written as,

(Mgd − mgx0 ) cos 𝜃0 = 0 (4.9)

which yields three possible equilibrium states.

@Seismicisolation
@Seismicisolation
 4.3 Equilibrium with Steady Motion 83

First, (Mgd − mgx0 ) = 0 satisfies Equation 4.9. This requires that x0 = Md∕m and then,
from Equation 4.8, 𝜃0 must take the value,
[ ( )]
k Md
𝜃0 = sin−1 𝓁0 − . (4.10)
mg m
What this means physically is that there is a state where the center of mass of the system
is at the fixed point O and the system is therefore balanced in an equilibrium position.
If, on the other hand, (Mgd − mgx0 ) ≠ 0, then Equation 4.9 requires that cos 𝜃0 = 0 so that
𝜃0 = 𝜋∕2 and 3𝜋∕2 become the equilibrium states. Equation 4.8 then yields x0 = 𝓁0 − mg∕k
and x0 = 𝓁0 + mg∕k for 𝜃0 = 𝜋∕2 and 3𝜋∕2 respectively. These are the equilibrium states
where the slot is vertically oriented and the small mass and the center of mass of the rigid
body are located on the same vertical line.
We can now consider the case where the torque T is not zero but is, in fact, controlled so
that 𝜃̇ = 𝜃̇ 0 , a constant angular velocity. In this case, the two differential equations making
up Equation 4.6 are treated differently. One is used to specify the torque required and the
other is used to calculate the equilibrium value of x.
As before, we specify that x = x0 , ẋ = ẍ = 0. The specification of 𝜃 is different this time.
We define the constant angular velocity 𝜃̇ = 𝜃̇ 0 from which we get 𝜃̈ = 0 and 𝜃 = 𝜃̇ 0 t. Notice
that the angle 𝜃 now grows linearly with time. The result of substituting these values into
Equation 4.6 is a specification for the required torque,
( )
T = (mx0 − Md)g cos 𝜃̇ 0 t (4.11)
and the value of x0 ,
( )
k𝓁0 − mg sin 𝜃̇ 0 t
x0 = . (4.12)
k − m𝜃̇ 2 0
There is clearly a problem here since Equation 4.12 says that the equilibrium value of x
is a time varying function. This shows that not every system can have equilibrium values
with steady motion. The problem here is the gravitational force that causes the small mass
to move harmonically in the slot as the rigid body rotates and the direction of gravity relative
to the slot changes.
The same system operating in a horizontal plane yields a much better result. Simply
removing the gravitational effects from Equation 4.6 gives the mathematical model of this
system in a horizontal plane. The equilibrium result is that T = 0, which is to be expected
since the system does not have an angular acceleration and,
k𝓁0
x0 = (4.13)
k − m𝜃̇ 02
indicating that the slider takes up a position where the force in the spring balances the effect
of the centripetal acceleration.
It is also possible to have equilibrium states where some degrees of freedom have constant
acceleration. Consider for example the system shown in Figure 4.4 where the large body is
subjected to the applied force F and the smaller body rolls on top of the large body and is
restrained by the spring of stiffness k. We can use the degrees of freedom x and y where
x is an absolute coordinate locating the large body relative to ground and y is a relative
coordinate locating the small body relative to the large body.

@Seismicisolation
@Seismicisolation
84 4 Equilibrium Solutions

y Figure 4.4 The constant acceleration case.

k
m

x F
M

Assuming an undeflected length of 𝓁0 for the spring, the equations of motion are,
[ ]{ } { }
(M + m) −m ẍ F
= . (4.14)
−m m ÿ −k(y − 𝓁0 )
It is reasonable to seek the equilibrium value of y given that the force F is controlled
so that the large mass has a constant acceleration. That is, we can set y = y0 = a constant
so that ẏ = ÿ = 0 and let ẍ = ẍ 0 = a constant. Note that this means that x and ẋ will both
be functions of time. In fact, since ẍ is a constant, we can easily write that ẋ = ẍ 0 t and
x = ẍ 0 t2 ∕2. These time varying values do not prevent us from finding the equilibrium value
of y because neither x nor ẋ are present in Equation 4.14.
We therefore use the first equation in 4.14 to find the force required for a constant accel-
eration,
F = (M + m)̈x0 (4.15)
and the second equation to find,
m̈x0
y0 = 𝓁0 + . (4.16)
k

4.4 The General Equilibrium Solution

The standard form of the equations of motion was previously presented as Equation 3.88,
which is repeated here for easy reference.
̇ t)]{q}
[M(q, q, ̈ = {f (q, q,
̇ t)}. (4.17)
The equilibrium solution is simply a set of constant values of the elements of vector {q},
say {q0 }, that will satisfy Equation 4.17. Since we are seeking equilibrium values that do
not vary with time, the explicit dependence on t disappears from the equations of motion.
Under these conditions, the equations of motion become a set of algebraic equations as
follows.
{f (q0 )} = {0}. (4.18)
Equation 4.18 will normally contain nonlinear algebraic equations, often requiring
numerical solutions.

@Seismicisolation
@Seismicisolation
 Exercises 85

In some cases, equilibrium states exist under conditions of steady motion. In these cases,
a subset of degrees of freedom in the {q} vector can be specified as having constant first or
second derivatives so long as no time varying elements appear in the equations of motion as
a result. If they appear, there will not be an equilibrium solution. The equations of motion
corresponding to these degrees of freedom can be used to calculate control inputs or may
simply be removed from the analysis. We may therefore say that for the case of steady
motion, {q̇ 0 } and {q̈ 0 } may have some specified, constant, non-zero values and the equi-
librium problem can be specified as,
{f (q0 , q̇ 0 , q̈ 0 )} = {0}. (4.19)

Exercises

Descriptions of the systems referred to in the exercises are contained in Appendix A.

4.1 Find four equilibrium positions for system 1.

4.2 Show that the equilibrium distance from O to the mass in system 4 is
k 𝓁0
x0 = .
k − m Ω2

4.3 Consider system 6. For the case where the angular velocity is constant at 60 RPM and
the disk has a radius of 30 cm, what rod length will result in an equilibrium value of
𝜃 equal to 60∘ ? What is the other equilibrium value of 𝜃 under these conditions? Note
that finding the second equilibrium value will require the use of numerical methods.

4.4 For the case where F = kL∕2 in system 7, find the two equilibrium values of 𝜃.

4.5 Given that system 8 has reached a steady-state where 𝜔 is constant, find the equilib-
rium positions of the mass attached to the pin A relative to the rotating body. Draw
sketches of the system with the mass in its equilibrium positions.

4.6 For system 9, using 𝜃 as the single degree-of-freedom, find two equilibrium positions
of the rod in the range 0 ≤ 𝜃 ≤ 𝜋∕2.

4.7 Derive the equation governing equilibrium in system 11 and show that setting
mg∕kL = 0.07735 will make the equilibrium value of 𝜃 equal to 30∘ .

4.8 Find the four equilibrium values of 𝜃 in system 17.

4.9 For the case where mg = 2kd, find four equilibrium positions for the rod in system
12. Sketch the rod in each equilibrium position.

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 87

Stability

The question of stability of dynamic systems is conceptually simple. In fact, the question is
not about the stability of the system as such but about the stability of equilibrium solutions
to the equations of motion. The question is: if the system has reached an equilibrium operat-
ing point1 and it is disturbed so that it moves slightly away from the equilibrium state, will it
return or not?
We say the operating point is stable if the system returns to it and unstable if it tends to
move away from it after the disturbance. The information about which way the system will
move resides in the equations of motion.
Sometimes stability characteristics become obvious simply by looking at the system as it
takes up each of its equilibrium states. A good example of this is the two degree of freedom
system presented in Chapter 4 for which the four equilibrium states are shown in Figure 4.2.
Clearly, only one equilibrium operating point is stable for this system. The second solution
shown has the center of mass at its lowest possible point. If the system moves away from
this position, it has no choice but to come back. This is the stable equilibrium operating
point for this system.
The three other equilibrium operating points have something in common. In each of
these, the center of mass is located at a point where, with a small disturbance, it can be
pulled further down by the gravitational force. These three equilibrium operating points
are all unstable. We will say more about this later.

5.1 Analytical Stability

Stability analysis is complicated by the nonlinear form of the equations of motion. Every-
thing is simpler if it is linear.
There are two ways of analyzing the stability of a dynamic system about an equilibrium
operating point.
1. We may work with the fully nonlinear equations of motion. This leads to analytical tech-
niques that, after a great deal of (interesting) mathematics, give regions in state-space
near the equilibrium state where perturbations away from equilibrium will eventually
be reduced to zero and the system will return to the equilibrium state if it is stable.

1 An operating point is a condition where we know the values of all the degrees of freedom.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
88 5 Stability

This type of analysis is not commonly used for practical problems in mechanics. It is
therefore not presented in this work. Readers who are interested in learning about sta-
bility of nonlinear equations of motion should start by looking up Lyapunov functions.
2. We may consider a linearized version of the equations of motion for small perturbations
away from the equilibrium operating point. In this case, the equations of motion will
reduce to a set of homogeneous, linear, differential equations for which solutions may
be readily written. The solutions tell us if the perturbations away from equilibrium grow
or decrease with time, thus establishing the stability of the system about the operating
point.
Consider the process of linearizing the equations of motion about an equilibrium operat-
ing point. The equations of motion have been derived and written in the form,
̇ t)]{q}
[M(q, q, ̈ = {f (q, q,
̇ t)}. (5.1)
We know from Chapter 4 that the operating point corresponds to an equilibrium solu-
̇ = {q̇ 0 }, {q}
tion of Equation 5.1 where {q} = {q0 }, {q} ̈ = 0, and the variation with time, t,
disappears, giving rise to the equation,
{f (q0 , q̇ 0 )} = {0}. (5.2)
We now consider an infinitesimally small variation, {x}, away from the equilibrium solu-
tion, {q0 }. The degrees of freedom and their derivatives become,
{q} → {q0 } + {x}
̇ → {q̇ 0 } + {x}
{q} ̇
̈ → {̈x}.
{q} (5.3)
Making these substitutions into Equation 5.1 yields,
[M(q0 , q̇0 )]{̈x} = {f (q0 , q̇0 )} + {f (x, x)}.
̇ (5.4)
The expression on the left hand side of Equation 5.4 includes a constant matrix, [M(q0 , q̇0 )],
which we term the mass matrix and represent simply as [M].
The right hand side of Equation 5.4 has two terms, one of which, {f (q0 , q̇0 )}, is zero
by virtue of Equation 5.2. The second term is a vector of nonlinear functions of the
infinitesimally small variables, x and x,̇ which were introduced as perturbations away from
equilibrium.
Given that the magnitudes of the elements of {x} and {x} ̇ are very small, then products
of elements of {x} and {x} ̇ (the nonlinear terms) are extremely small and can be ignored
relative to linear terms. The result is that terms involving for instance x1 x2 , x10 ẋ 4 , ẋ 35 and so
on are eliminated from the equations of motion in favor of linear terms.
As a result, the following approximation can be made.
̇ ≈ −[C]{x}
{f (x, x)} ̇ − [K]{x} (5.5)
where [K] and [C] are constant coefficient matrices, which are called the stiffness and damp-
ing matrices respectively. They are defined with the negative signs so that, when the approx-
imation is substituted into Equation 5.4, it can be written as,
̇ − [K]{x}
[M]{̈x} = −[C]{x} (5.6)

@Seismicisolation
@Seismicisolation
 5.1 Analytical Stability 89

and, immediately, in the standard form for linear equations of motion,

̇ + [K]{x} = {0}.
[M]{̈x} + [C]{x} (5.7)
The coefficient matrices are square and of dimension N × N where N is the number of
degrees of freedom in the system. The vector of degrees of freedom, {x}, and its derivatives
are each of length N.
Equation 5.7 is a very interesting and useful result. It says that very small perturbations
of the system degrees of freedom away from an equilibrium operating point are governed
by a set of second order, linear, differential equations with constant coefficients.
Solutions to differential equations of this type are easy to generate and will tell us whether
the perturbations will grow (the unstable case) or decay (the stable case) with time. The solu-
tions are always exponential functions of time. Think about a function of time which, when
multiplied by a constant and added to other constants multiplied by its first two derivatives,
gives a result of zero. This is what the differential equation requires and the only function
which has the same form as its first two derivatives is the exponential function.
We therefore assume that the solution to Equation 5.7 looks like,
{x(t)} = {X}e𝜆t (5.8)
where {X} is a vector of complex2 but constant amplitudes of the degrees of freedom. 𝜆 is
also a constant complex number but is a scalar rather than a vector.
Equation 5.8 can be differentiated to give,
̇
{x(t)} = 𝜆{X}e𝜆t
{̈x(t)} = 𝜆2 {X}e𝜆t . (5.9)
Substituting Equations 5.8 and 5.9 into Equation 5.7 yields,
(𝜆2 [M] + 𝜆[C] + [K]){X}e𝜆t = {0}. (5.10)
Equation 5.10 can only be satisfied if the left hand side is identically equal to zero. Since
e𝜆t can never be zero for any value of 𝜆, it can be taken out of the equation. For a system with
N degrees of freedom, the result is a set of N algebraic equations with N + 1 unknowns. The
unknowns are the N amplitudes contained in {X} and the scalar 𝜆, which appears in the
coefficient matrix. A general form of the system of equations is,
[A(𝜆)]{X} = {0} (5.11)
where [A(𝜆)] = (𝜆2 [M] + 𝜆[C] + [K]). This is the standard form of an eigenvalue problem
where there are characteristic values of the eigenvalue, 𝜆, which make the determinant of
[A(𝜆)] equal to zero.
Equation 5.11 is a set of N simultaneous algebraic equations with a zero right hand side.
If the determinant of [A(𝜆)] is non-zero, then there will be a well-defined, unique solution
to the equations and that solution will be {X} = {0}. In other words, the amplitude of every
degree of freedom will be zero and the system will not move.
We therefore seek to find values of 𝜆 that make the determinant zero. These are the eigen-
values. For each eigenvalue, we determine the value of the amplitude of each degree of

2 Complex here does not mean complicated but rather means that the amplitudes are in the form of
complex numbers.

@Seismicisolation
@Seismicisolation
90 5 Stability

freedom and fill in the {X} vector. These will be the eigenvectors, each corresponding to a
different eigenvalue.
Note that we can never solve for all of the elements of {X} uniquely. There are only N
equations to work with and, as was mentioned earlier, there are N + 1 unknowns. We must
solve for 𝜆, after which we can only extract N − 1 further pieces of information from the set
of equations. We, in fact, choose to solve for the ratio of each of the elements of {X} to a
reference element. That is, for a chosen reference degree of freedom xp , we find the N − 1
complex values of (xj ∕xp , j = 1, N, j ≠ p) where, of course, (xp ∕xp = 1 + 0i)3 by definition.
This means that the eigenvector shows only the ratio of the amplitudes of the motions in
the system and not their actual values.
If we were to write a polynomial to represent the determinant of [A(𝜆)], it would be as
follows
det[A(𝜆)] = a2N 𝜆2N + a2N−1 𝜆2N−1 + · · · + a1 𝜆 + a0 = 0.
There are 2N roots to this polynomial and therefore 2N eigenvalues. The reason for the
polynomial being of order 2N while [A(𝜆)] is only N × N is that 𝜆2 appears in the coefficient
matrix.
We therefore have a set of 2N eigenvalues and 2N eigenvectors. The entire set can be
denoted by (𝜆j , {X}j , j = 1, 2N).
In general, the eigenvalues are complex numbers of the form 𝜆j = 𝜎j + i𝜔j . The time func-
tions assumed (see Equation 5.8) as solutions to the linearized differential equations of
motion (i.e. e𝜆t ) now take on the form,
e𝜆j t = e(𝜎j +i𝜔j )t
= e𝜎j t ei𝜔j t
= e𝜎j t (cos 𝜔j t + i sin 𝜔j t) (5.12)
where the last line of Equation 5.12 uses Euler’s formula, ei𝜃 = cos 𝜃 + i sin 𝜃.
Since we are considering the solution of a set of linear differential equations
(Equation 5.7), we can apply superposition to say that the total response of the sys-
tem to an arbitrary set of initial conditions will be a linear combination of all of the natural
solutions. The natural solutions are those that arise from the eigenvalue analysis just
performed. Therefore, the total response is,
∑
2N
{x(t)} = cj {X}j e𝜆j t
j=1

= c1 {X}1 e𝜆1 t + c2 {X}2 e𝜆2 t + · · · + c2N {X}2N e𝜆2N t . (5.13)

The constant coefficients in Equation 5.13 (i.e. the cj ) are derived from the initial condi-
tions of the motion. That is, the initial perturbation from the steady-state solution of the
nonlinear equations of motion will determine how much of each natural mode enters the
response. Since perturbations tend to be random, it is likely that all of the natural modes
will be excited from time to time. Therefore, from the point of view of stability, we must ensure
that all of the natural modes decay with time.
√
3 The notation used here for complex numbers has i = −1

@Seismicisolation
@Seismicisolation
 5.1 Analytical Stability 91

Whether or not a mode decays, depends upon the real part of the eigenvalue. This can be
seen from Equation 5.12, where the real part of the eigenvalue (i.e. 𝜎j ) appears in an expo-
nential function that can decay or grow with time depending upon the sign of 𝜎j . If 𝜎j > 0
the exponential term grows inexorably with time. The magnitude of 𝜎j is related only to the
speed at which the function grows and has no bearing on whether it grows or not. Small
positive values of 𝜎j imply slow growth and large positive values of 𝜎j imply rapid growth
but any positive value of 𝜎j implies that the system will move away from the steady-state
position and that the particular operating point is unstable. Conversely, if 𝜎j < 0, perturba-
tions away from the steady-state solutions will decay with time and the system will return
to the operating point. Only if all 2N eigenvalues have negative real parts is the system stable.
The imaginary parts of the eigenvalues appear in the harmonic terms (i.e. sin and cos) of
Equation 5.12 and have no effect on whether or not the functions grows or decays. In fact,
the magnitude of the expression (cos 𝜔j t + i sin 𝜔j t) is always 1.0. This term shows whether
the response to the perturbation is oscillatory or not. If 𝜔j = 0, then the response is purely
exponential and no oscillations occur. If, on the other hand, 𝜔j ≠ 0, then the response is
oscillatory and the frequency of oscillation of the jth mode is given by 𝜔j .
Figure 5.1 shows schematically the response types related to different values of 𝜎j and 𝜔j .
It contains a number of names that are commonly used in describing the stability charac-
teristics of a system. The two types of instability, divergent and flutter, are descriptive of the
modes of instability and have their origins in aircraft stability analysis performed early in
the 20th century. Neutrally stable modes have 𝜎j = 0 so that no definite statement can be
made about whether they are stable or not. In practice neutrally stable modes are consid-
ered undesirable and treated as instabilities. Undamped oscillations are simply harmonic
functions that go on forever without amplitude modulation. rigid body modes correspond
to eigenvalues that are identically zero and indicate that the system, having no resistance
to motion in a particular mode, will simply behave as an unrestrained rigid body which, as
Newton said, “perseveres in its state of rest or of uniform motion”. Rigid body modes in a

σj < 0 σj = 0 σj > 0

ωj = 0 Divergent
Overdamped Rigid Body Mode
Instability

Underdamped Undamped Flutter

Instability
ωj ≠ 0

STABLE NEUTRALLY UNSTABLE

STABLE

Figure 5.1 Stability response types.

@Seismicisolation
@Seismicisolation
92 5 Stability

mechanical system often indicate that supporting structures have not been included in the
model.

5.2 Linearization of Functions

The linearization of functions about specified operating points is an important topic,

deserving of explanation. The trigonometric functions being considered here give us a good
opportunity to look closely at the process. We therefore leave the discussion of stability
briefly to have a look at linearization.
When asked to give the small angle approximations for the sine and cosine, most engi-
neers will reply that, for small 𝛼,

sin 𝛼 ≈ 𝛼

cos 𝛼 ≈ 1.

However, they are unlikely to have given much thought to why this is so or, more impor-
tantly, to what assumptions are implicit in these statements.
The small angle approximations are a linearization of the sine and cosine about the oper-
ating point 𝛼 = 0. They are, in fact, the first terms arising from a Taylor series4 expansion
of these trigonometric functions about 𝛼 = 0. The expression for a one-dimensional Taylor
series expansion for a function f (x) about the point x = x0 is,
df || 1 d2 f || 1 d3 f ||
f (x0 + Δx) = f (x0 ) + | Δx + | Δx2 + | Δx3 + · · · . (5.14)
dx |x=x0 2! dx ||x=x
2 3! dx3 ||x=x
0 0

If we take x to be 𝛼 and f (x) = f (𝛼) = cos 𝛼, then the derivatives required for the series
are,
d(cos 𝛼)
= − sin 𝛼
d𝛼
d2 (cos 𝛼) d(− sin 𝛼)
= = − cos 𝛼
d𝛼 2 d𝛼
d3 (cos 𝛼) d(− cos 𝛼)
= = sin 𝛼
d𝛼 3 d𝛼
and the expansion gives,
1 1
cos(𝛼 + Δ𝛼) = cos 𝛼 − Δ𝛼 sin 𝛼 − Δ𝛼 2 cos 𝛼 + Δ𝛼 3 sin 𝛼 + · · · . (5.15)
2 6
To this point the analysis has been general and the relationship just derived can be used
to produce an expansion about any angle 𝛼. We now consider the expansion about 𝛼 = 0
where cos 𝛼 = 1 and sin 𝛼 = 0. The expansion about 𝛼 = 0 therefore yields,
1
cos Δ𝛼 ≈ 1 − Δ𝛼 2 + · · ·
2

4 Brook Taylor (1685–1731), an English mathematician, developed the calculus of finite differences and
invented integration by parts. He is best known for introducing the well-known Taylor series.

@Seismicisolation
@Seismicisolation
 5.2 Linearization of Functions 93

which, with nonlinear terms ignored because Δ𝛼 is so small that Δ𝛼 2 is negligible5 , gives
the common reply by an engineer,
cos Δ𝛼 ≈ 1.
Doing this again for the sine function gives,
d(sin 𝛼)
= cos 𝛼
d𝛼
d2 (sin 𝛼) d(cos 𝛼)
= = − sin 𝛼
d𝛼 2 d𝛼
d3 (sin 𝛼) d(− sin 𝛼)
= = − cos 𝛼
d𝛼 3 d𝛼
from which,
1 1
sin(𝛼 + Δ𝛼) = sin 𝛼 + Δ𝛼 cos 𝛼 − Δ𝛼 2 sin 𝛼 − Δ𝛼 3 cos 𝛼 + · · · . (5.16)
2 6
Now let 𝛼 = 0 so that cos 𝛼 = 1, and sin 𝛼 = 0. The expansion yields,
1
sin Δ𝛼 ≈ Δ𝛼 − Δ𝛼 2 + · · ·
6
which, with nonlinear terms ignored, once again yields the common reply,
sin Δ𝛼 ≈ Δ𝛼.
Now consider what happens if the expansion is about some angle other than zero. Say
that we wish to linearize about the angle 𝛼 = 𝛼0 where 𝛼0 ≠ 0.
Equation 5.15 becomes,
1 1
cos(𝛼0 + Δ𝛼) = cos 𝛼0 − Δ𝛼 sin 𝛼0 − Δ𝛼 2 cos 𝛼0 + Δ𝛼 3 sin 𝛼0 + · · ·
2 6
which can be linearized to,
cos(𝛼0 + Δ𝛼) ≈ cos 𝛼0 − Δ𝛼 sin 𝛼0 .
Similarly, Equation 5.16 becomes,
1 1
sin(𝛼0 + Δ𝛼) = sin 𝛼0 + Δ𝛼 cos 𝛼0 − Δ𝛼 2 sin 𝛼0 − Δ𝛼 3 cos 𝛼0 + · · · .
2 6
In linear form, this is,
sin(𝛼0 + Δ𝛼) ≈ sin 𝛼0 + Δ𝛼 cos 𝛼0 .
As an aside, these same relationships can be derived from the well known “sum of angles
formulae”6
cos(𝛼 + 𝛽) = cos 𝛼 cos 𝛽 − sin 𝛼 sin 𝛽

5 The term linearization expresses the fact that the perturbation in the variable being considered is so
small that squaring it, cubing it, or taking it to any higher power produces terms whose magnitudes are
negligibly small compared to the linear term and can therefore be ignored so that retaining the linear term
only provides a good approximation to the value of the function very close to the operating point.
6 While these may be “well known”, they are not easily remembered. However, they can be quickly
derived using Euler’s formula: ei𝜃 = cos 𝜃 + i sin 𝜃. We write the equation twice, once for the angle 𝛼 and
again for the angle 𝛽, giving,
ei𝛼 = cos 𝛼 + i sin 𝛼

@Seismicisolation
@Seismicisolation
94 5 Stability

sin(𝛼 + 𝛽) = cos 𝛼 sin 𝛽 + sin 𝛼 cos 𝛽.

We substitute 𝛼0 for 𝛼 and Δ𝛼 for 𝛽 and get,
cos(𝛼0 + Δ𝛼) = cos 𝛼0 cos Δ𝛼 − sin 𝛼0 sin Δ𝛼

sin(𝛼0 + Δ𝛼) = cos 𝛼0 sin Δ𝛼 + sin 𝛼0 cos Δ𝛼

and then we use the “usual” small angle approximation to say that,
cos Δ𝛼 ≈ 1
and
sin Δ𝛼 ≈ Δ𝛼.
Substituting these above yields,
cos(𝛼0 + Δ𝛼) ≈ cos 𝛼0 − Δ𝛼 sin 𝛼0

sin(𝛼0 + Δ𝛼) ≈ sin 𝛼0 + Δ𝛼 cos 𝛼0 .

These are the same results obtained with the Taylor series expansion.
A trigonometric function of two angles (e.g. sin(𝛼 + 𝛽) or cos(𝛼 + 𝛽)) is a function of two
variables so that the Taylor series expansion of Equation 5.14 is insufficient for the purpose.
Instead we use a two variable form of the Taylor series expansion of the function f (x, y) about
the operating point (x0 , y0 ).
𝜕f || 𝜕f ||
f (x0 + Δx, y0 + Δy) = f (x0 , y0 ) + Δx + Δy
𝜕x ||x0 ,y0 𝜕y ||x0 ,y0
( )
1 𝜕 2 f || 𝜕 2 |
f | 𝜕 2 |
f |
+ | Δx2 + | ΔxΔy + 2 | Δy2 + · · · .
2! 𝜕x2 ||x ,y 𝜕x𝜕y ||x ,y 𝜕y ||x ,y
0 0 0 0 0 0

(5.17)
The point to note is that, whether we are using the single variable expansion
(Equation 5.14) or the two variable expansion (Equation 5.17), the linear terms involve
only the first derivative(s) of the function7 . Second and higher derivatives always involve

ei𝛽 = cos 𝛽 + i sin 𝛽

We then multiply these two equations together giving, on the left hand side,
ei𝛼 ei𝛽 = ei(𝛼+𝛽) = cos(𝛼 + 𝛽) + i sin(𝛼 + 𝛽)
and, on the right hand side,
(cos 𝛼 cos 𝛽 − sin 𝛼 sin 𝛽) + i(cos 𝛼 sin 𝛽 + sin 𝛼 cos 𝛽)
Equating the real and imaginary parts of these two results gives the “sum of angles” formulae we couldn’t
remember,
cos(𝛼 + 𝛽) = cos 𝛼 cos 𝛽 − sin 𝛼 sin 𝛽

sin(𝛼 + 𝛽) = cos 𝛼 sin 𝛽 + sin 𝛼 cos 𝛽.

7 The linear terms in multi-variable Taylor series expansions also involve only the first derivatives of the
function.

@Seismicisolation
@Seismicisolation
 5.3 Example: A System with Two Degrees of Freedom 95

Figure 5.2 Linearization.

nonlinear function

force - f(x)
operating point
x0
f(x0)
linear approximation
deflection - x

nonlinear terms that will eventually be ignored. Linearization of functions about operating
points is therefore an exercise in finding the local slope of the function and assuming that
small deviations away from the operating point can be approximated by points lying on
this straight line. This is shown graphically in Figure 5.2 for a function of one variable and
can be generalized to functions of many variables.
The linearized version of a function of many variables, f (x, y, z, …), about the operating
point, (x0 , y0 , z0 , …), is therefore,
f (x0 + Δx, y0 + Δy, z0 + Δz, …) = f (x0 , y0 , z0 , …)
𝜕f || 𝜕f || 𝜕f ||
+ Δx + Δy + Δz + · · · . (5.18)
𝜕x ||x0 ,y0 ,z0 ,… 𝜕y ||x0 ,y0 ,z0 ,… 𝜕z ||x0 ,y0 ,z0 ,…
If we apply Equation 5.18 to cos(𝛼 + 𝛽), for instance, we first need to find,
𝜕 cos(𝛼 + 𝛽)
= − sin(𝛼 + 𝛽)
𝜕𝛼
and
𝜕 cos(𝛼 + 𝛽)
= − sin(𝛼 + 𝛽)
𝜕𝛽
then, substituting into Equation 5.18, we find,
cos[(𝛼 + Δ𝛼) + (𝛽 + Δ𝛽)] ≈ cos(𝛼 + 𝛽) − Δ𝛼 sin(𝛼 + 𝛽) − Δ𝛽 sin(𝛼 + 𝛽).
Similarly, the linearized version of sin(𝛼 + 𝛽) is,
sin[(𝛼 + Δ𝛼) + (𝛽 + Δ𝛽)] ≈ sin(𝛼 + 𝛽) + Δ𝛼 cos(𝛼 + 𝛽) + Δ𝛽 cos(𝛼 + 𝛽).
Equation 5.18 can be used whenever a function needs to be linearized about a known
operating point.

5.3 Example: A System with Two Degrees of Freedom

At the end of Chapter 3, the equations of motion for a two dimensional rigid body (see
Figure 3.6) with two degrees of freedom were derived. The equilibrium positions of this
system were found in Section 4.2 and are demonstrated graphically in Figure 4.2. Here, we
consider the stability of each of the four equilibrium positions.

@Seismicisolation
@Seismicisolation
96 5 Stability

Before proceeding with the analysis we should inspect the equilibrium positions and see
how we would expect the system to behave if it were moved very slightly away from equilib-
rium. Referring to Figure 4.2, it should be clear that the first, third, and fourth equilibrium
positions are unstable. Each is comparable to an inverted pendulum that, given a slight dis-
turbance, will fall. Further, we expect that these instabilities will be divergent. That is, there
will be no oscillations that cause the system to pass through the equilibrium position many
times as the amplitude of motion increases. They will simply fall. The second equilibrium
is different because the center of mass of the rigid body is at its lowest possible position and
cannot fall. If we disturb this equilibrium we should see a neutrally stable harmonic motion
about the equilibrium. The motion is not stable unless the amplitudes decrease with time
and, since there is no element in this system which removes energy, we expect undamped
oscillations.
The equations of motion for the system, taken from Equation 3.89, are repeated here as
Equation 5.19.
[ ]{ } { }
Mr 2 MrP 𝜃̈ Mr 𝜙̇ 2 Q − Mgr sin 𝜃
= (5.19)
MrP I 𝜙̈ −Mr 𝜃̇ 2 Q − Mg[y0 cos 𝜙 + x0 sin 𝜙]
where,
P = x0 cos(𝜙 − 𝜃) − y0 sin(𝜙 − 𝜃)
and
Q = y0 cos(𝜙 − 𝜃) + x0 sin(𝜙 − 𝜃).
This example provides quite a challenge in terms of linearization. Some terms in the
equations of motion are obviously nonlinear and can be discarded immediately. These are
the terms on the right hand side containing 𝜙̇ 2 and 𝜃̇ 2 . What makes the linearization dif-
ficult are the trigonometric functions of more than one variable such as cos(𝜙 − 𝜃) and
sin(𝜙 − 𝜃). Linearization was discussed in Section 5.2 and the techniques described there
will be used here.
In order to derive the linearized version of Equation 5.19, we first recognize that the
terms in the mass matrix must be constant. That is, they cannot be linear functions of any
of the degrees of freedom since each of these terms is already multiplied by the second
derivatives of the degrees of freedom that reside in the column vector to the right of the
mass matrix. The expansion for the [x0 cos(𝜙 − 𝜃) − y0 sin(𝜙 − 𝜃)] term is therefore simply
[x0 cos(𝜙0 − 𝜃0 ) − y0 sin(𝜙0 − 𝜃0 )], keeping only the constant terms from the expansions of
cos(𝜙 − 𝜃) and sin(𝜙 − 𝜃) as discussed in Section 5.2. During the linearizing process, the
equilibrium expressions of Equation 4.4 are also found and are set equal to zero because of
the requirement that they be zero at an equilibrium point.
On the right hand side, we immediately drop the terms containing squares of angular
velocities as being nonlinear. The remaining expressions involve quite simple linearizations
of sine and cosine, which, when done, give the linearized equations of motion as,
[ ]{ }
Mr 2 MrP Δ𝜃̈
MrP I Δ𝜙̈
[ ]{ } { }
Mgr cos 𝜃0 0 Δ𝜃 0
+ = (5.20)
0 Mg[−y0 sin 𝜙0 + x0 cos 𝜙0 ] Δ𝜙 0

@Seismicisolation
@Seismicisolation
 5.3 Example: A System with Two Degrees of Freedom 97

where,
P = [x0 cos(𝜙0 − 𝜃0 ) − y0 sin(𝜙0 − 𝜃0 )].
We now assume an exponential solution to the linear differential equation (Equation 5.20)
and let,
{ } { }
Δ𝜃 X1
= e𝜆t
Δ𝜙 X2
which gives, after two differentiations,
{ } { }
Δ𝜃̈ X1
=𝜆 2
e𝜆t .
Δ𝜙̈ X2
Substituting into Equation 5.20 gives,
[ ]{ }
Mr 2 𝜆2 + Mgr cos 𝜃0 MrP𝜆2 X1
e𝜆t
MrP𝜆2 I𝜆2 + Mg[−y0 sin 𝜙0 + x0 cos 𝜙0 ] X2
{ }
0
= . (5.21)
0
To find the eigenvalues, we write the determinant of the coefficient matrix in
Equation 5.21 as a polynomial in 𝜆 and set it equal to zero. The polynomial is,
𝜆4 [Mr 2 (I − MP2 )] + 𝜆2 {Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )]}
+ M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ) = 0. (5.22)
It is clear that, even with this relatively simple example, making general comments about
the stability of the system by working algebraically with the polynomial in Equation 5.22
will be difficult. In fact, we will be able to say something about the stability later when we
consider a different method. For the time being, we will assign values to the parameters in
the system and calculate the eigenvalues numerically to see if they agree with the intuitive
results presented at the beginning of this section.
Consider the case where the system has parameter8 values as specified in Table 5.1. The
resulting eigenvalues and eigenvectors of the system are presented in Table 5.2.
The data presented in Table 5.2 describe the natural motions of the system when it is
moved slightly away from an equilibrium state. Reference to Figure 4.2 will be helpful in
understanding the descriptions that follow.
● Case 1: There is a single unstable9 mode (i.e. an eigenvalue with a positive real part) in
this equilibrium position. The mode shape (i.e. the eigenvector) indicates that 𝜙 is much
greater than 𝜃 for this mode10 . Referring to Figure 3.6 for the definitions of the degrees
of freedom, 𝜃 and 𝜙, and to Figure 4.2 for a view of the system when it is the equilibrium
position of case 1, it is clear that the mass will tend to fall and that changes in the angle
𝜙 will be the dominant motion as is does. This is a divergent instability.

8 Equation 4.5 expresses the equilibrium value of 𝜙 as a function of x0 and y0 . The parameter values
chosen here result in an angle that is approximately −26.5∘ (−0.46365 rad) or 𝜋 plus this value, depending
on the case. Similarly, the equilibrium values of 𝜃 were found to be 0 or 𝜋.
9 The eigenvalues related to unstable modes are shown in boldface in Table 5.2.
10 Remember that the eigenvector is defined with 𝜃 as the first element and 𝜙 as the second.

@Seismicisolation
@Seismicisolation
98 5 Stability

Table 5.1 Parameter values for the two degree of freedom example.

Parameter Symbol Value Units

Mass M 10.0 kg
Moment of inertia about B I 5.5 kg m2
Length of supporting rod r 0.5 m
Distance from B to G (⃗i direction) x0 0.2 m
Distance from B to G (⃗j direction) y0 0.1 m
Acceleration due to gravity g 9.807 m s−2
Supporting rod angle (case 1) 𝜃0 0 rad
Mass angle (case 1) 𝜙0 𝜋-0.46365 rad
Supporting rod angle (case 2) 𝜃0 0 rad
Mass angle (case 2) 𝜙0 −0.46365 rad
Supporting rod angle (case 3) 𝜃0 𝜋 rad
Mass angle (case 3) 𝜙0 −0.46365 rad
Supporting rod angle (case 4) 𝜃0 𝜋 rad
Mass angle (case 4) 𝜙0 𝜋-0.46365 rad

Table 5.2 Eigenvalues and eigenvectors for the two degree of freedom example.

Case Eigenvalue/eigenvector

1 𝜆1 = −4.6087i 𝜆2 = 4.6087i 𝝀𝟑 = 𝟐.𝟎𝟏𝟐𝟓 𝜆4 = −2.0125

{ } { } { } { }
1.0 1.0 0.07654 0.07654
{X}1 = {X}2 = {X}3 = {X}4 =
0.1712 0.1712 1.0 1.0
2 𝜆1 = −4.6970i 𝜆2 = 4.6970i 𝜆3 = −1.9746i 𝜆4 = 1.9746i
{ } { } { } { }
1.0 1.0 0.1110 0.1110
{X}1 = {X}2 = {X}3 = {X}4 =
−0.2481 −0.2481 1.0 1.0
3 𝝀𝟏 = 𝟒.𝟔𝟎𝟖𝟕 𝜆2 = −4.6087 𝜆3 = −2.0125i 𝜆4 = 2.0125i
{ } { } { } { }
1.0 1.0 0.07654 0.07654
{X}1 = {X}2 = {X}3 = {X}4 =
0.1712 0.1712 1.0 1.0
4 𝝀𝟏 = 𝟒.𝟔𝟗𝟕𝟎 𝜆2 = −4.6970 𝝀𝟑 = 𝟏.𝟗𝟕𝟒𝟔 𝜆4 = −1.9746
{ } { } { } { }
1.0 1.0 0.1110 0.1110
{X}1 = {X}2 = {X}3 = {X}4 =
−0.2481 −0.2481 1.0 1.0

@Seismicisolation
@Seismicisolation
 5.4 Routh Stability Criterion 99

● Case 2: Case 2 is the only stable equilibrium position for this system. Here, all of the eigen-
values are purely imaginary, indicating undamped oscillatory motion. That this state is
only neutrally stable is no surprise since there are no forces that remove energy from the
system. To be truly stable some “damping” terms would need to be added to the system.
There are two distinct modes in case 2. There is an oscillation at 4.6970 rad s−1 (approxi-
mately 0.75 Hz) where changes in the angle 𝜃 are dominant. There is a lower frequency
oscillation (1.9746 rad s−1 or 0.31 Hz), which is dominated by changes in the angle 𝜙.
● Case 3: There is a single divergent instability here which will demonstrate itself as a
change in the angle 𝜃. Again, this makes sense when we look at case 3 in Figure 4.2 and
see that, in order to “fall”, the system will see large changes in 𝜃.
● Case 4: There are two divergent instabilities in this equilibrium state. Again referring to
the figure, this makes sense. Both 𝜃 and 𝜙 will see large changes as the system leaves this
precarious equilibrium state. The mode shapes of the two instabilities indicate that each
is dominated by one of the two angles.
In reality, systems never stay in unstable equilibrium positions. There will always be a
slight disturbance that will start the system moving and it will always move from the unsta-
ble equilibrium to a stable equilibrium position. In this example, only case 2 was a stable
equilibrium and the system would eventually find itself there. Whether an equilibrium state
is stable or not is not as obvious for systems more complex than this example.

5.4 Routh Stability Criterion

In the previous section, we derived the polynomial form of the determinant11 of the coef-
ficient matrix for the linearized equations of motion of the system for small motions away
from an equilibrium point (Equation 5.22). We decided it was too difficult to pursue an alge-
braic solution to the polynomial and proceeded to handle it numerically. We now return to
the polynomial and consider its stability using the Routh12 stability criterion.
The Routh stability criterion is a method for determining system stability from the nth
order characteristic equation of the form,
an 𝜆n + an−1 𝜆n−1 + · · · + a1 𝜆 + a0 . = 0 (5.23)
From the coefficients of the characteristic equation, we form a Routh table as shown
in Table 5.3. The first two rows of the table contain the coefficients of the characteristic
equation and are padded with zeros at the end. We find the coefficients bi and ci from 2 × 2
determinants involving the coefficients in the two rows just preceding them. That is,
a a − an an−3 a a − an an−5
b1 = n−1 n−2 b2 = n−1 n−4 ···
an−1 an−1
(5.24)
b1 an−3 − an−1 b2 b1 an−5 − an−1 b3
c1 = c2 = ··· .
b1 b1

11 This polynomial is called the characteristic polynomial for the system

12 Edward John Routh (1831–1907), born in Canada, studied mathematics in England and was well
known for the teaching skills he demonstrated while employed at Cambridge University. He is especially
remembered for his contributions to the field of dynamic stability.

@Seismicisolation
@Seismicisolation
100 5 Stability

Table 5.3 The Routh table.

𝜆n an an−2 an−4 ··· 0

𝜆 n−1
an−1 an−3 an−5 ··· 0
𝜆n−2 b1 b2 b3 ···
𝜆n−3 c1 c2 c3 ···
⋮ ⋮ ⋮ ⋮ ⋮

Each new element generated for the table depends on the leading elements of the preced-
ing two rows and on the two elements just to the right on the same two rows. For instance,
the calculation of c2 , the element in row 4 and column 2, uses an−1 and b1 , the two leading
elements in rows 2 and 3, and an−5 and b3 , the two elements in column 3 of rows 2 and
3. The pattern is easy to follow and the table is continued horizontally and vertically until
only zeros are obtained.
A statement on the stability of the equilibrium state being analyzed for the system can be
made from consideration of the first column of the Routh table as follows: all of the roots
of the characteristic equation have negative real parts if and only if the elements of the first
column of the Routh table have the same sign. Otherwise, the number of roots with positive
real parts is equal to the number of sign changes in the first column.
Remember that the roots of the characteristic equation are the eigenvalues of the lin-
earized equations of motion of the system for small perturbations about an equilibrium
state. Our previous work said that having all eigenvalues with negative real parts ensured
stability. Therefore, having the same sign, whether positive or negative, on all of the ele-
ments in the first column of the Routh table ensures stability. If the column starts out being
positive, for example, and then changes to a negative value at some element, staying nega-
tive thereafter, there is a single sign change and, as a result, one unstable mode. If, on the
other hand, there is a change of sign as just described but it is followed later by a return to
the original sign, there will be two unstable modes and so on.
We now return to the two degree of freedom system for which we have been consider-
ing the stability. In particular, we reconsider getting an analytical statement on stability by
reconsidering the characteristic equation first presented as Equation 5.22 and reproduced
below as Equation 5.25.
𝜆4 [Mr 2 (I − MP2 )] + 𝜆2 {Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )]}
+ M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ) = 0 (5.25)
where
P = [x0 cos(𝜙0 − 𝜃0 ) − y0 sin(𝜙0 − 𝜃0 )]. (5.26)
The characteristic equation is of the form,
a 4 𝜆4 + a 3 𝜆3 + a 2 𝜆2 + a 1 𝜆 + a 0 = 0 (5.27)
where
a4 = Mr 2 (I − MP2 )

@Seismicisolation
@Seismicisolation
 5.4 Routh Stability Criterion 101

Table 5.4 The initial Routh table for the two

degrees of freedom example.

𝜆4 a4 a2 a0 0
𝜆3 𝜖 0 0 0
𝜆2
b1 b2 0
𝜆1 c1 c2
𝜆0 d1

a3 = 0
a2 = Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )]
a1 = 0
a0 = M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ). (5.28)
The Routh table for this case starts out as shown in Table 5.4 where a small but non-zero
term 𝜖 has been introduced as the first element in the second row. We work with 𝜖 to avoid
division by zero while we fill in the remainder of the table and then we let 𝜖 → 0 to get the
final form of the table.
The elements in the table are calculated as follows.
𝜖 ⋅ a2 − a4 ⋅ 0
b1 = = a2
𝜖
𝜖 ⋅ a0 − a4 ⋅ 0
b2 = = a0
𝜖
b ⋅ 0 − 𝜖 ⋅ b2 𝜖 ⋅ b2
c1 = 1 =−
b1 b1
b1 ⋅ 0 − 𝜖 ⋅ 0
c2 = =0
b1
c ⋅ b − b1 ⋅ c2
d1 = 1 2 = a0 .
c1
Making these substitutions into Table 5.4 and letting 𝜖 → 0 yields the final Routh table
presented in Table 5.5.
The first column of Table 5.5 contains all of the coefficients in the characteristic equation
plus three zeros. The zeros are representative of the same neutrally stable behavior we noted

Table 5.5 The final Routh table for the two

degrees of freedom example.

𝜆4 a4 a2 a0 0
𝜆3
0 0 0 0
𝜆2 a2 a0 0
𝜆1 0 0
𝜆0 a0

@Seismicisolation
@Seismicisolation
102 5 Stability

earlier. They cannot be taken as changes of sign but rather of a “near” change in sign. What
is of more interest to us is that all of the coefficients in the characteristic equation must have
the same sign or the system will be unstable.
The coefficients are given in Equation 5.28. It is immediately clear that it will be difficult
to make comments about the signs of any of these coefficients if we do not have relatively
easy values of 𝜃0 and 𝜙0 to work with. For that reason, we will consider the case where
y0 = 0. This means that the center of gravity G is located a distance x0 from point B and that
the equilibrium values of 𝜙 (found from Equation 4.5) are,
{
y0 0
𝜙0 = arctan(− ) = arctan(0) = .
x0 𝜋
The four equilibrium cases, expressed as their values of the angles 𝜃0 and 𝜙0 and the
resulting value of p from Equation 5.26 are now,
● Case 1: 𝜃0 = 0, 𝜙0 = 𝜋 ⇒ P = −x0 .
● Case 2: 𝜃0 = 0, 𝜙0 = 0 ⇒ P = x0 .
● Case 3: 𝜃0 = 𝜋, 𝜙0 = 0 ⇒ P = −x0 .
● Case 4: 𝜃0 = 𝜋, 𝜙0 = 𝜋 ⇒ P = x0 .
We first consider the sign of a4 = Mr 2 (I − MP2 ). In all of the equilibrium cases being
considered, the value of a4 is a4 = Mr 2 (I − Mx02 ). This is, using the parallel axis theorem13 ,
a4 = Mr 2 IG > 0 since IG is greater than zero by definition.
Having a4 > 0 sets the stage for the Routh stability calculation. The system will be stable
in the chosen equilibrium if and only if a2 > 0 and a0 > 0.
Consider case 1. We have 𝜃0 = 0 and 𝜙0 = 𝜋. As a result, with y0 = 0,
a2 = Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )] = Mgr(I − Mrx0 ).
The sign of a2 depends on the relative magnitudes of I and MRx0 . Given the parameter
values we are using for this analysis, I = 5.5 and Mrx0 = 1.0 so that a2 > 0.
We now consider a0 = M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ). Given the values of 𝜃0 , 𝜙0 , and
y0 , this becomes a0 = −M 2 g2 rx0 . a0 is definitely negative.
The final result for case 1 is that the first column of the Routh table (see Table 5.5) starts
with a4 > 0, sees no sign change as we move down the first column to a2 > 0 and sees one
sign change as we move down to a0 < 0. As a result, there is one eigenvalue with a positive
real part (one sign change) and, therefore, one unstable mode. This is the same result
we obtained from the numerical study. The difference is that applying the Routh stability
criterion does not give us information related to “how unstable the mode is” (i.e. where the
eigenvalue with the positive real value lies on the complex plane) or on the mode shape
that is unstable.
We now consider case 2 where 𝜃0 = 0 and 𝜙0 = 0. This time, again with y0 = 0, we find,
a2 = Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )] = Mgr(I + Mrx0 ).

13 The reader is encouraged to review the definitions of moments and products of inertia in Appendix B.
The parallel axis theorem states that the moment of inertia about some point B on a rigid body is equal to
the moment of inertia about the center of mass, IG , plus the mass of the rigid body, M, multiplied by the
square of the distance between G and B. That is, in this case, I = IG + Mx02 or IG = I − Mx02 .

@Seismicisolation
@Seismicisolation
 5.5 Standard Procedure for Stability Analysis 103

In this case, a2 > 0 for all parameter values.

In case 2, a0 = M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ) = M 2 g2 rx0 , which is greater than zero
for all parameter values. Therefore, case 2 sees positive values for all elements in the first
column of the Routh table and the equilibrium position is stable as we predicted numeri-
cally.
Consideration of cases 3 and 4 follow directly from what has just been presented.
In case 3, a4 > 0, a2 = Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )] = Mgr(−I + Mrx0 ) < 0,
and a0 = M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ) = −M 2 g2 rx0 < 0 so that there is one sign
change and one unstable mode.
Similarly, in case 4, a4 > 0, a2 = Mgr[I cos 𝜃0 + Mr(−y0 sin 𝜙0 + x0 cos 𝜙0 )] = Mgr(−I −
Mrx0 ) < 0, and a0 = M 2 g2 r cos 𝜃0 (−y0 sin 𝜙0 + x0 cos 𝜙0 ) = M 2 g2 rx0 > 0 so that there are
two sign changes and two unstable modes.
Notice that case 2 is stable and case 4 has two sign changes, and therefore two unstable
modes, for any set of parameter values. Cases 1 and 3 have the option of changing the sign of
a2 depending on whether (I − Mrx0 ) > 0 or not. In fact, if (I − Mrx0 ) changes sign because
different parameter values are used, the result is simply a shift of where the sign change
occurs in the first column of the Routh array and not a change in the number of sign rever-
sals. As a result, cases 1 and 3 have a single sign change and therefore one unstable mode
regardless of the sign of a2 .

5.5 Standard Procedure for Stability Analysis

We start by assuming that the equations of motion for a system have been derived, an equi-
librium solution has been found, and the equations of motion have been linearized for small
motions about this equilibrium position to give N second order linear differential equations
for the N degrees of freedom, as follows.
̇ + [K]{x} = {0}.
[M]{̈x} + [C]{x} (5.29)
Whether or not the system is stable for small perturbations away from the equilibrium
operating point requires that the eigenvalues and eigenvectors (if we are interested in mode
shapes) of these differential equations be found.
You will find that computer programs for finding eigenvalues work with the standard
eigenvalue problem, which starts with a set of first-order differential equations,
̇ = [A]{x}
{x} (5.30)
and then assumes exponential solutions {x(t)} = {X}e𝜆t to arrive at, after exchanging sides
in the equation, the standard eigenvalue problem,
[A]{X} = 𝜆{X}. (5.31)
The problem is then one of transforming Equation 5.29 to a form equivalent to that in
Equation 5.30. This can be accomplished using the following steps.
● Define a new set of variables for the displacements
{y1 } = {x} ⇒ {ẏ 1 } = {x}.
̇ (5.32)

@Seismicisolation
@Seismicisolation
104 5 Stability

● Define a second set of new variables for the rates of change of the displacements
̇ ⇒ {ẏ 2 } = {̈x}.
{y2 } = {x} (5.33)
● From the preceding two definitions, we can see that we have established two expressions
for the rates of change of the displacements. These two expressions must agree, resulting
in a set of N first-order differential equations,
{ẏ 1 } = {y2 }. (5.34)
● We still have our initial differential equations (Equation 5.29)
̇ + [K]{x} = {0}
[M]{̈x} + [C]{x}
which, after noting that {̈x} = {ẏ 2 }, can be written as,
[M]{ẏ 2 } + [C]{y2 } + [K]{y1 } = {0}
or
{ẏ 2 } = −[M]−1 [K]{y1 } − [M]−1 [C]{y2 }. (5.35)
● Equations 5.34 and 5.35 are then written together as a set of 2N differential equations.
{ } [ ]{ }
ẏ 1 0 I y1
= . (5.36)
ẏ 2 −M −1 K −M −1 C y2
● The 2N × 2N constant coefficient matrix,
[ ]
0 I
−M −1 K −M −1 C
is defined to be [A] so that the differential equations can be written as,
̇ = [A]{y}
{y}
which is in the desired form of Equation 5.30.
Standard eigenvalue routines for finding eigenvalues and eigenvectors simply require
that the matrix [A] be provided. The user must be careful to determine that the algorithm
used is for non-symmetric matrices when sending a typical [A] matrix from a dynamic sys-
tem since in our case [A] is most often non-symmetric. Computer routines for finding the
eigenvalues and eigenvectors of symmetric matrices are much more efficient than those
for non-symmetric matrices and are very common but they cannot process non-symmetric
matrices. It is up to the user to determine which algorithm to use in any given software
package.
If used properly, the computer package will return a set of 2N complex eigenvalues of the
form,
𝜆j = 𝜎j + i𝜔j ; j = 1, 2N.
We have previously noted that the stability of the system is dependent upon the signs of
the 𝜎j and the frequencies of oscillation on the 𝜔j .
For each eigenvalue there will be a corresponding eigenvector. This means that 2N eigen-
vectors will be returned, each consisting of a column of 2N complex values. They contain

@Seismicisolation
@Seismicisolation
 Exercises 105

valuable information about the mode shapes inherent in the system and deserve a little extra
explanation which will be presented in the next chapter.

Exercises
Descriptions of the systems referred to in the exercises are contained in Appendix A.

5.1 Let an equilibrium position for system 1 be 𝜃10 and 𝜃20 . Show that the mass and stiff-
ness matrices for small motions about the equilibrium position are,
[ ]
(m1 + m2 )d12 + m2 d22 + 2m2 d1 d2 cos 𝜃20 m2 d22 + m2 d1 d2 cos 𝜃20
[M] =
m2 d22 + m2 d1 d2 cos 𝜃20 m2 d22
[ ]
−(m1 + m2 )gd1 sin 𝜃10 − m2 gd2 sin(𝜃10 + 𝜃20 ) −m2 gd2 sin(𝜃10 + 𝜃20 )
[K] = .
−m2 gd2 sin(𝜃10 + 𝜃20 ) −m2 gd2 sin(𝜃10 + 𝜃20 )

5.2 Consider system 1 for the case where m1 = m2 = m and d1 = d2 = d. Write the char-
acteristic equation for small motions about each of the four equilibrium states found
in Exercise 4.1. Determine the stability of each equilibrium state using the Routh sta-
bility criterion and confirm the results by finding the eigenvalues.

5.3 Linearize the equation of motion for system 4 about the equilibrium position
determined in Exercise 4.2 and discuss the stability for: (1) (k − m Ω2 ) < 0; (2)
(k − m Ω2 ) = 0; and (3) (k − m Ω2 ) > 0.

5.4 Linearize the equation of motion for system 6 around the two equilibrium positions
found in Exercise 4.3 and determine whether or not they are stable.

5.5 Two equilibrium states were found for system 8 in Exercise 4.5. Comment on the
stability of the two states. If there is an instability, specify the type.

5.6 Are the two equilibrium states found for system 9 in Exercise 4.6 stable?

5.7 Comment on the stability of the four equilibrium states found for system 17 in
Exercise 4.8.

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 107

Mode Shapes

This is a separate chapter on mode shapes, included because of the importance of mode
shapes in understanding the dynamic behavior of systems. Assume that the analyst has cre-
ated a dynamic model of a system, found the equilibrium states, linearized the equations of
motion about one of them, and tested that state for stability. In the case where the eigenval-
ues indicate an instability, the question becomes – what design changes can be made to make
the system stable? The answer may be as simple as adding stiffness or damping somewhere
in the system – but where? The mode shapes answer this question.
Equation 5.13 in Section 5.1 shows that the total response of the linear system is a com-
bination of all of the natural modes. The job of the analyst is to identify which mode is
unstable, visualize the body as it moves in that mode and then choose locations where
there are relatively large deflections between bodies as the locations to attempt to stabi-
lize the system by adding stiffness or damping. Adding stiffness or damping between two
points that have no relative motion in this mode shape will be fruitless.
Visualization of mode shapes is much enhanced by computer animation but the effort
involved in doing that is often not justified. Quick, hand-drawn sketches can be just as
effective in achieving the design goals.

6.1 Eigenvectors

The eigenvalues and eigenvectors are calculated for a general dynamic system that has been
cast into a first-order form as shown in Equation 5.36 and our interpretation of mode shapes
begins with that general system. An eigenvalue 𝜆j = 𝜎j + i𝜔j will be accompanied by an
eigenvector,
{ }
Y1
Y2 j

which contains, in its first N elements, the complex amplitudes of {y1 (t)} and, in its last N
elements, the complex amplitudes of {y2 (t)}. In Equations 5.32 and 5.33 we defined {y1 (t)}
as the original set of displacement degrees of freedom, {x(t)}, and {y2 (t)} as the derivatives
̇
of {x(t)}. That is, {y1 (t)} = {x(t)} and {y2 (t)} = {x(t)}. We therefore expect the first N ele-
ments of the eigenvector to represent the displacements in the system and to be very useful
The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.
© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
108 6 Mode Shapes

in describing how the system degrees of freedom move in relation to each other. The last
half of the eigenvector contains derivatives of the first half that, given that we assumed expo-
nential solutions of the form {x(t)} = {X}e𝜆t , can be regenerated simply by multiplying the
̇
first N elements of the eigenvector by 𝜆j since {x(t)} = 𝜆{X}e𝜆t = 𝜆{x(t)}.
In other words, the eigenvector returned by whatever computer package is being used is
actually,
{ } { } { }
Y1 X X
= = . (6.1)
Y2 j Ẋ j
𝜆X j
It is convenient therefore to discard the last N elements of the eigenvector and work with
the displacements contained in the first N elements.
Consider now an eigenvector of length N, the elements of which are all complex numbers
representing the displacement degrees of freedom in a system. Such a vector would look
like,
⎧ a1 + ib1 ⎫
⎪ ⎪
⎪ a2 + ib2 ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎪ a + ibp ⎪
{X}j = ⎨ p ⎬ (6.2)
⎪ ⋮ ⎪
⎪ a + ib ⎪
⎪ n n
⎪
⎪ ⋮ ⎪
⎪ ⎪
⎩ aN + ibN ⎭j
where the coefficients ap and bp , for all p, are real valued constants.
Remembering that the elements in the eigenvector are of arbitrary magnitude since they
represent only the relationship of one degree of freedom to another, there is no loss of
information1 if we normalize the elements of the eigenvector by dividing by the element
of largest absolute value. Assume that the element an + ibn has the largest magnitude and
divide all of the others by it. Note that this √
involves dividing by a complex number, an + ibn ,
and not by the magnitude of the number, a2n + b2n . The eigenvector is transformed to,
⎧ c1 + id1 ⎫
⎪ ⎪
⎪ c2 + id2 ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎪ c + idp ⎪
{X}j = ⎨ p ⎬ (6.3)
⎪ ⋮ ⎪
⎪ 1 + i0 ⎪
⎪ ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎩ N idN
c + ⎭j

1 This was shown in Section 5.1 where we noted that we could never solve for all of the elements of {X}
uniquely since there were only N equations to work with and there were N + 1 unknowns – the eigenvalue
and the N elements of the eigenvector. This meant that we could find only the eigenvalue and N − 1 other
pieces of information that are components of the eigenvector relative to each other. As a result, we can
scale all of the elements of an eigenvector by the same factor without losing any information.

@Seismicisolation
@Seismicisolation
 6.1 Eigenvectors 109

where cp and dp are real valued constants and the nth degree of freedom, which is now rep-
resented by 1 + i0, becomes the reference degree of freedom as we move forward from here.
We now transform
√ the eigenvector one more time by changing each element to an ampli-
tude, |Xp | = c2p + d2p , and a phase angle with respect to the degree of freedom with maxi-
mum amplitude, 𝜙p = arctan(dp ∕cp ). The final form of the eigenvector is then,

⎧ |X | ∠ 𝜙 ⎫
⎪ 1 1
⎪
⎪ |X | ∠ 𝜙 ⎪
⎪ 2 2 ⎪
⎪ ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎪ |Xp | ∠ 𝜙p ⎪
{X}j = ⎨ ⎬ (6.4)
⎪ ⋮ ⎪
⎪ ⎪
⎪ 1∠0 ⎪
⎪ ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎪ |X | ∠ 𝜙 ⎪
⎩ N N ⎭j

where the expression ∠ 𝜙p means “at an angle 𝜙p ” with respect to the phase angle of the
reference degree of freedom.
The reason for producing the eigenvectors is so that we can observe the natural move-
ments of the system as functions of time. To this point, we have generated only an amplitude
and a phase angle with respect to the degree of freedom having the largest amplitude. Since
we cannot have an eigenvector without having first found the corresponding eigenvalue, we
also have some idea of the time dependence of the modes of motion. That is, we know if a
mode oscillates or not and we know how much damping each mode has. In general, we are
mostly interested in the mode shapes of those modes that are unstable or nearly unstable.
Any corrective action we can take to improve the stability of these modes depends on visu-
alizing the motions and applying force-producing elements to areas in the system where
they will have some effect. For example, if an unstable mode involves motions where two
adjacent bodies have no relative motion with respect to each other, there is little to be gained
from placing a spring or damper between these two bodies. Such an element will produce
no force and, as a result, make no contribution to stabilizing the system.
We know that the motion resulting from mode-j of the system can be written, as a function
of time, as {x(t)} = {X}j e𝜆j t or, given that 𝜆j = 𝜎j + i𝜔j and substituting e𝜆j t = e𝜎j t (cos 𝜔j t +
i sin 𝜔j t), we can write {x(t)} = {X}j e𝜎j t (cos 𝜔j t + i sin 𝜔j t).
In order to visualize the motion, we normally ignore the damping term, e𝜎j t , for the simple
reason that it causes the amplitudes to grow or decay with time and therefore clouds our
image of the motion. We already know if the mode is unstable or not, we simply want to
see what type of motion is involved.
The apparent conflict between the “real” and “imaginary” parts of the response has been
partly resolved by transforming the eigenvector into an amplitude and a phase angle. Given
that we are ignoring the decay or growth for the purpose of visualization, we can write the

@Seismicisolation
@Seismicisolation
110 6 Mode Shapes

time variation of the system in the jth mode as,

⎧ |X1 |(cos(𝜔j t + 𝜙1 ) + i sin(𝜔j t + 𝜙1 )) ⎫
⎪ ⎪
⎪ |X2 |(cos(𝜔j t + 𝜙2 ) + i sin(𝜔j t + 𝜙2 )) ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎪ |Xp |(cos(𝜔j t + 𝜙p ) + i sin(𝜔j t + 𝜙p )) ⎪
{x(t)} = ⎨ ⎬. (6.5)
⎪ ⋮ ⎪
⎪ 1(cos(𝜔j t) + i sin(𝜔j t)) ⎪
⎪ ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎩ |XN |(cos(𝜔j t + 𝜙N ) + i sin(𝜔j t + 𝜙N )) ⎭
The real and imaginary parts of Equation 6.5 differ only in that each is the projection of
the complex function on a different axis of the complex plane. We can pick either the real
or the imaginary component to describe the motion since each has the same amplitude and
frequency. Without loss of generality, we can choose to watch the motion as it is projected
on the imaginary axis and write,

⎧ |X1 | sin(𝜔j t + 𝜙1 ) ⎫
⎪ ⎪
⎪ |X2 | sin(𝜔j t + 𝜙2 ) ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎪ |Xp | sin(𝜔j t + 𝜙p ) ⎪
{x(t)} = ⎨ ⎬. (6.6)
⎪ ⋮ ⎪
⎪ 1 sin(𝜔j t) ⎪
⎪ ⎪
⎪ ⋮ ⎪
⎪ ⎪
⎩ |XN | sin(𝜔j t + 𝜙N ) ⎭
Table 5.2 lists the eigenvalues and eigenvectors for the two degrees of freedom system we
have considered so many times. Here, we are interested in case 2 – the equilibrium state
with two stable eigenvalues. From the table, there is a mode with 𝜆 = ±4.6970i. This mode2
has a frequency of 4.6970 rad s−1 or 0.746 Hz and is described by a motion where the angle3
𝜙 is roughly 1/4 (actually 0.2481) of the magnitude of the angle 𝜃 and is out-of-phase (the
negative sign on the second element of the eigenvector indicates a 180∘ phase angle) with
𝜃. Figure 6.1 shows the system in equal increments of time as it moves in this mode. The

Figure 6.1 The higher frequency mode for the two degrees of freedom system (0.746 Hz).

2 Table 5.2 actually shows two modes with similar eigenvalues, one with 𝜆 = +4.6970i and another with
𝜆 = −4.6970i. Since eigenvalues with non-zero imaginary parts appear as complex conjugate pairs, these
are actually the same mode.
3 The degrees of freedom, 𝜃 and 𝜙 are defined in Figure 3.6.

@Seismicisolation
@Seismicisolation
 6.2 Comparing Translational and Rotational Degrees of Freedom 111

Figure 6.2 The lower frequency mode for the two degrees of freedom system (0.314 Hz).

motion consists mainly of rotation of the supporting rod with the rigid body maintaining
nearly the same orientation at all times as it swings back and forth.
Figure 6.2 shows the motion for the second mode associated with the equilibrium of case
2. From Table 5.2, we see that this is a lower frequency mode than the first (1.9746 rad s−1
or 0.314 Hz) and is dominated by the angle 𝜙. 𝜃 has roughly 10% (actually 0.1110) of the
amplitude of 𝜙 and the angles are in-phase since both elements have the same sign. We
expect a motion where the supporting rod moves very little and the rigid body essentially
swings about the supporting point B. This motion is shown in Figure 6.2 and confirms the
point just made.

6.2 Comparing Translational and Rotational Degrees

of Freedom
The example just presented did not contain an important element in the interpretation of
mode shapes. In looking at the example we were comparing two rotational degrees of free-
dom, 𝜃 and 𝜙, and were able to make, and understand, statements like “𝜃 has roughly 10%
(actually 0.1110) of the amplitude of 𝜙” and understand what they mean. The equations of
motion for many systems have both rotational and translational degrees of freedom. What
does it mean to say that “x is 10% of 𝛼” when x is a displacement and 𝛼 is an angle?
To answer this question, we use another example. Figure 6.3 shows a thin, uniform, rigid
rod of length 2a that is suspended at its ends by two springs of stiffness k1 and k2 . The first
view in the figure shows the system in its equilibrium position. The rod has mass m and a
moment of inertia about its center of mass of ma2 ∕3. A linear damper with coefficient c is
included at a distance b from the center of mass as shown.
We let the rod have two degrees of freedom. Its center of mass is free to move up and down
and we designate this motion by the variable x, which is positive upwards and is a measure
of how far the center of mass moves away from its equilibrium position. In addition, the rod
can rotate in the plane of its motion and we use the angle 𝛼 to designate the rotation away
from the equilibrium orientation.
The kinematic analysis of this system is very simple. The acceleration of the center of
mass is ẍ vertically upward and the rod has an angular acceleration of 𝛼̈ in the direction
shown for positive 𝛼 in Figure 6.3.
Figure 6.3 also shows the free-body diagram for the system. Notice that the weight of the
rod is missing since it will be offset by preloads in the springs4 . As a result, we also don’t

4 Finding eigenvalues and eigenvectors requires that the linear equations of motion about an equilibrium
state be derived. If this is all that is required, there is no need to derive the fully nonlinear equations of

@Seismicisolation
@Seismicisolation
112 6 Mode Shapes

a a
b

k1 c k2

x α

f2
fc

f1

Figure 6.3 A rigid rod supported on springs.

include the preloads when we establish expressions for the spring forces f1 and f2 . They
are only the changes in the spring forces that result from displacements away from their
equilibrium displacements.
Using the symbol 𝛿 to designate the displacements of the elements from their equilibrium
lengths, we can write,

𝛿1 = x − a𝛼
𝛿2 = x + a𝛼
𝛿̇c = ẋ + b𝛼̇ (6.7)

motion, establish the equilibrium positions, and then linearize the equations of motion about that
equilibrium. We simply assume that the system is in an easily determined equilibrium (e.g. resting on the
springs in this case) and then derive the linear equations of motion for motions away from equilibrium
directly. This is the method most often used in the study of vibrations as opposed to dynamics. Vibrations is
primarily concerned with the linear, frequency domain, response of systems. One of the subtleties in
deriving these linear equations of motion is that the forces that acted to put the system in the equilibrium
position are canceled out by constant forces in elements such as springs and both are ignored in
determining the equations of motion. In the case considered here, gravity acted on the rod to deflect the
springs downward until an equilibrium was established. As we consider motions away from that
equilibrium, we consider neither gravity nor the preloads in the springs.

@Seismicisolation
@Seismicisolation
 6.2 Comparing Translational and Rotational Degrees of Freedom 113

where the assumption of small motions has been used implicitly. That is, no trigonometric
functions of the angle 𝛼 appear because the small angle assumptions sin 𝛼 ≈ 𝛼 and cos 𝛼 ≈ 1
have been used.
A short discussion of the directions assumed for the forces in the elements is in order
at this point because it is very easy to lose a sign. Construction of the free-body diagram in
Figure 6.3 was not possible without deciding on directions for the forces acting in the springs
and the damper. The spring forces, f1 and f2 , are shown as if the springs are in tension. That
is, the assumption is that the motions of the system have caused both springs to become
longer than they were in the equilibrium position and the springs are both trying to pull the
rod down in order to return to equilibrium. Whether or not this is actually the case depends
on what the actual motion of the system is and we cannot predict that without first deriving
the equations of motion. Assuming tension in both springs is as good an assumption as any
so long as we maintain that assumption for all the following steps in the derivation. The
drawing in the center of Figure 6.3 seems to imply that if the spring on the right hand end
is in tension than the spring on the left must be in compression. Again, we have no way
of saying in advance whether this is so or not. The center drawing is simply there to show
the positive directions assumed for the degrees of freedom, x and 𝛼. No prior knowledge of
the motion was available when it was drawn. The point is that the degrees of freedom have
been assigned positive directions and assumptions were made about the displacements of
the springs. The analyst must respect these choices when deriving the equations of motion.
With respect to the force in the damper, fc , the same arguments as were made for the
springs apply. Linear dampers resist getting longer or shorter by virtue of a constitutive
relationship that says that fc = c𝛿̇ where c is a constant damping coefficient. The free-body
diagram shows that fc has been assumed to be acting downward on the rod. This can only
happen if the damper is getting longer so the expression for 𝛿̇ c in Equation 6.7 has been
derived so that it is positive if the damper is increasing in length.
A final note about Equation 6.7 is that the expressions presented respect the assumed
directions of the forces and use the positive senses of the degrees of freedom to calculate
the corresponding spring displacements and damper velocity.
The forces acting can then be written as,
f1 = k1 𝛿1 = k1 (x − a𝛼)
f2 = k2 𝛿2 = k2 (x + a𝛼)
fc = c𝛿̇c = c(ẋ + b𝛼).
̇ (6.8)

We now sum forces vertically upward. Remember that x was assumed to be positive
upward and that positive acceleration is in the same direction as x. According to Newton’s
second law, there can only be an acceleration in that direction if there is a net force in
that direction. The implication is that some or all of the force directions we assumed were
incorrect but consistency in applying our assumptions will lead to the correct equations of
motion. The force balance results in,
m̈x = −f1 − f2 − fc
= −k1 (x − a𝛼) − k2 (x + a𝛼) − c(ẋ + b𝛼)
̇
= −cẋ − cb𝛼̇ − (k1 + k2 )x + a(k1 − k2 )𝛼. (6.9)

@Seismicisolation
@Seismicisolation
114 6 Mode Shapes

Summing moments about the center of mass in the direction of positive 𝛼̈ (i.e. the same
counter-clockwise direction as positive 𝛼) yields,
ma2
𝛼̈ = af1 − af2 − bfc
3
= ak1 (x − a𝛼) − ak2 (x + a𝛼) − bc(ẋ + b𝛼)
̇
= −bcẋ − cb2 𝛼̇ + a(k1 − k2 )x − a2 (k1 + k2 )𝛼. (6.10)
Combining Equations 6.9 and 6.10 into the standard matrix form gives,
[ ]{ } [ ]{ }
m 0 ẍ c cb ẋ
1
+
0 3 ma 2
𝛼̈ cb cb 2 𝛼̇
[ ]{ } { }
k1 + k 2 −a(k1 − k2 ) x 0
+ = . (6.11)
2
−a(k1 − k2 ) a (k1 + k2 ) 𝛼 0
We consider two cases. The first will be a case with no damping (i.e. c = 0) in order to
demonstrate a mode shape where we compare a translation (x) with a rotation (𝛼). The sec-
ond will include a significant amount of damping so that the phase angles between degrees
of freedom becomes obvious.
Let the parameter values be as shown in Table 6.1.
The eigenvalues and corresponding eigenvectors for the first case are,
{ }
1.0000
𝜆1 = ±11.652i ; {X}1 =
−0.1056
and
{ }
0.1407
𝜆2 = ±20.597i ; {X}2 = .
1.0000
The first mode is therefore an undamped oscillation at a frequency of 1.85 Hz
(11.652 rad s−1 ) with the amplitude of x dominating. 𝛼 is roughly 10% of x (actually
10.56%) and has the opposite sign. The second mode is another undamped oscillation (no
surprise since c = 0 for this case), this time at 3.28 Hz (20.597 rad s−1 ) with x being only
14% of 𝛼. The question is, what do these mode shapes look like? How do we compare
translation to rotation?

Table 6.1 Parameters for the rigid rod example.

Parameter Case 1 Case 2

m 1000 kg 1000 kg
a 2m 2m
b 1m 1m
k1 60000 Nm−1 60000 Nm−1
k2 80000 Nm−1 80000 Nm−1
c 0 N ms−1 5000 N ms−1

@Seismicisolation
@Seismicisolation
 6.3 Nodal Points in Mode Shapes 115

Visualization of the mode shapes is akin to drawing the system as it moves in time. To
draw the rigid rod, we first need to know where the center of mass is (i.e we need to know
x). After making this translation, we need to know where the ends of the rod are located and
we do this using the same linear equations we used to find spring deflections (Equation 6.8).
It is clear that the positions of the ends of the rod relative to the position of the center of
mass depend on the angle 𝛼 and the distance from the center of mass to the ends of the rod.
The absolute vertical positions of the rod ends (written as pvleft and pvright ) can be expressed
as,
pvleft = x − a𝛼
pvright = x + a𝛼. (6.12)
Given the positions of the ends of the rod as a function of time or, more precisely, as a
function of the what proportion of a cycle of motion has passed, we can connect the ends
with a straight line and visualize the motion. The expressions for the motions of the ends of
the rod (substituting the eigenvectors for the two mode shapes and a = 2 m from Table 6.1)
are, for the first mode,
pvleft = [1.000 − 2.0(−0.1056)] sin(11.652t)
= 1.2112 sin(11.652t)
pvright = [1.000 + 2.0(−0.1056)] sin(11.652t)
= 0.7888 sin(11.652t) (6.13)
and, for the second,
pvleft = [0.1407 − 2.0(1.0000)] sin(20.597t)
= −1.8593 sin(20.597t)
pvright = [0.1407 + 2.0(1.0000)] sin(20.597t)
= 2.1407 sin(20.597t). (6.14)
Having generated the expressions for the motions, presented as Equations 6.13 and 6.14,
one can draw the rod for various values of 𝜔t in the range 0 < 𝜔t < 2𝜋, thereby showing a
complete cycle of the harmonic motion. Figure 6.4 shows the two modes. Remember that
the amplitudes given in Equations 6.13 and 6.14 are not absolute since the expressions are
derived from the eigenvectors that contain only relative amplitude information. A sketch of
a mode shape can use any convenient value for the vertical displacement of one end of the
rod. The motion of the other end must satisfy the relative amplitudes given in the equations.

6.3 Nodal Points in Mode Shapes

In Section 6.2, the vertical motions of the left and right ends of the rigid rod were given in
Equation 6.12. If you compare these to the expressions for the spring deflections at the ends
of the rod (see Equation 6.7), you will see that the expressions are identical. Basically, the
deflection of a spring connected from the moving body to the ground is a measure of the
motion of the point on the body relative to ground (i.e. absolute motion) in the direction

@Seismicisolation
@Seismicisolation
116 6 Mode Shapes

mode 1 mode 2
ωt = 0

ωt = π/2

ωt = π

ωt = 3π/2

ωt = 2π

Figure 6.4 Undamped mode shapes for the rigid rod supported on springs.

aligned with the spring. We can easily write an expression for the deflection of a spring
connecting the rod to the ground at some distance, 𝓁, to the right of the center of mass as,

𝛿 = x + 𝓁𝛼 (6.15)

and, if we are looking for a point that has no motion at all, we simply set the deflection of
the spring to zero and solve for the location of that point. For 𝛿 = 0, we get
−x
𝓁= (6.16)
𝛼
which says that the point with no motion in the first mode is located at 𝓁 = 9.4697 m and,
in the second mode, at 𝓁 = −0.1407 m.
Figure 6.5 shows the nodal points for the rigid rod. Nodal points are useful aids in visu-
alizing mode shapes. The motion of the system is simply a rotation about the nodal points.

6.4 Mode Shapes with Damping

In Table 6.1, we introduced a second case for the rigid rod supported on springs that we
have been considering. The second case included significant damping and the calculated
eigenvalues and eigenvectors for that case are,
{ }
−0.0144357 + 0.0815538i
𝜆1 = −2.1045 ± 11.889i; {X}1 =
−0.0122252 − 0.0086485i

@Seismicisolation
@Seismicisolation
 6.4 Mode Shapes with Damping 117

Figure 6.5 Nodal points. mode 1 nodal point

mode 2 nodal point

and,
{ }
0.0188754 + 0.0071370i
𝜆2 = −2.2705 ± 19.747i; {X}2 = .
−0.0057466 + 0.0499797i
These are in the raw form of Equation 6.2.
The element having the largest amplitude in each eigenvector is selected and divided into
the other element(s) using complex division. In the first case, the first element of {X}1 is
the largest so we divide by −0.0144357 + 0.0815538i to get,
{ }
1.0000000 + 0.0000000i
𝜆1 = −2.1045 ± 11.889i; {X}1 = .
−0.0770966 + 0.1635498i
Similarly, we divide {X}2 throughout by the first element, (i.e. 0.0188754 + 0.0071370i,
to get
{ }
0.0980779 − 0.3889390i
𝜆2 = −2.2705 ± 19.747i; {X}2 = .
1.0000000 + 0.0000000i
These are now in the form of Equation 6.3.
The final step is to change the elements of the eigenvectors to magnitudes and phase
angles as in Equation 6.4. This results in,
{ }
1.0000 ∠ 0.00∘
𝜆1 = −2.1045 ± 11.889i; {X}1 =
0.1808 ∠ − 115.24∘
and,
{ }
0.4011 ∠ 75.85∘
𝜆2 = −2.2705 ± 19.747i; {X}2 = .
1.0000 ∠ 0.00∘
Using Equation 6.6, we can write the time variation of the mode shapes as,
{ }
1.0000 sin(𝜔1 t)
{x(t)}1 =
0.1808 sin(𝜔 t − 115.24∘ )
1

and,
{ }
0.4011 sin(𝜔2 t + 75.85∘ )
{x(t)}2 =
1.0000 sin(𝜔2 t)

@Seismicisolation
@Seismicisolation
118 6 Mode Shapes

where, from the eigenvalues, 𝜔1 = 11.889 rad s−1 (1.89 Hz) and 𝜔2 = 19.747 rad s−1 (3.14 Hz).
In fact, the actual values of the natural frequencies, while useful, are not important to
visualizing the mode shapes. We simply want to plot the system’s position for fractions of
a period of motion, 0 ≤ 𝜔t ≤ 2𝜋, so we use a set of values of 𝜔1 t and 𝜔2 t between 0 and 2𝜋.
We are basically creating frames of an animation when we do this. If you are actually doing
computer animation, you use as many frames as necessary to get a smooth animation. When
using paper, as we are here, it is generally sufficient to use the set, 𝜔t = 0, 𝜋∕2, 𝜋, 3𝜋∕2, 2𝜋.
To make the plotting easier, we can use the sine of the sum of angles identity to write

sin(𝜔t + 𝜙) = sin 𝜔t cos 𝜙 + cos 𝜔t sin 𝜙

so that the time variation of the mode shapes becomes

{ }
1.0000 sin 𝜔1 t
{x(t)}1 =
0.1808 sin 𝜔1 t cos(−115.24∘ ) + 0.1808 cos 𝜔1 t sin(−115.24∘ )

and,
{ }
0.4011 sin 𝜔2 t cos 75.85∘ + 0.4011 cos 𝜔2 t sin 75.85∘
{x(t)}2 = .
1.0000 sin 𝜔2 t

Substituting numerical values for the sines and cosines of the phase angles gives
{ }
1.0000 sin 𝜔1 t
{x(t)}1 =
−0.0771 sin 𝜔1 t − 0.1635 cos 𝜔1 t

and,
{ }
0.0981 sin 𝜔2 t + 0.3889 cos 𝜔2 t
{x(t)}2 =
1.0000 sin 𝜔2 t

which are readily evaluated for the values of 𝜔t we have chosen to use.
Figure 6.6 shows the damped mode shapes. The concept of nodal points (Section 6.3)
does not exist for heavily damped systems. Using these two damped mode shapes in
Equation 6.16 in order to find the nodal point leaves you with tangent or cotangent
functions of 𝜔t that go off to infinity more than once in the range 0 ≤ 𝜔t ≤ 2𝜋 so that the
nodal points are undefined.

6.5 Modal Damping

The last concept to discuss with respect to the natural modes of a system is how damping
can be quantified in an understandable way. Saying that the first mode in the example just
considered has a real part of −2.2705 doesn’t give much information about how long we
would expect the mode to sustain motion once it starts. That is, if we disturb the system so

@Seismicisolation
@Seismicisolation
 6.5 Modal Damping 119

Figure 6.6 Damped mode shapes mode 1 mode 2

for the rigid rod supported on
springs. ωt = 0

ωt = π/2

ωt = π

ωt = 3π/2

ωt = 2π

that the response is entirely composed of the first mode, does the amplitude decay5 within
three oscillations or does it take one hundred oscillations?
We start with the equation of motion for a single degree of freedom system for small
perturbations about an equilibrium state, which can be written as,
m̈x + cẋ + kx = 0. (6.17)
Assuming the usual exponential solution to Equation 6.17,
x(t) = Xe𝜆t (6.18)
this can be rewritten as,
(m𝜆2 + c𝜆 + k)Xe𝜆t = 0. (6.19)
The characteristic equation is therefore,
m𝜆2 + c𝜆 + k = 0 (6.20)
which, after division throughout by m, becomes,
c k
𝜆2 + 𝜆+ = 0. (6.21)
m m

5 The exponential decay of oscillations predicted from linear models means that the amplitude never goes
to zero but only tends toward zero as time approaches infinity. In fact, there is always some Coulomb
friction in systems and the oscillations eventually stop completely because of the work done by friction
forces. Decay in this section uses the somewhat arbitrary measure of the number of oscillations that
reduces the initial amplitude by 90%.

@Seismicisolation
@Seismicisolation
120 6 Mode Shapes

The eigenvalues (i.e. characteristic roots of Equation 6.21) are,

√ ( )⎤
⎡ ( )2
1⎢ c c k ⎥
𝜆= − ± −4 . (6.22)
2⎢ m m m ⎥
⎣ ⎦
Whether or not the motion is oscillatory at all depends on the sign of the term within the
square root of Equation 6.22. This term can be negative, zero, or positive.
If it is negative, the square root will generate an imaginary number and the system will
oscillate at the frequency indicated by the result (see the discussion following Equations
5.12 and 5.13). In this case, the system is said to be underdamped.
If the term inside the square root is zero, the solution will no longer oscillate. We call
the system critically damped in this case and define a critical damping coefficient, ccr , as the
amount of damping which just makes the system stop oscillating. The value,
√
c = ccr = 4km
makes the term inside the square root equal to zero.
If the term inside the square root is greater than zero, the system will have a purely expo-
nential behavior and will decay exponentially with time. If the value of c is increased beyond
the critical damping value, the system is said to be overdamped and the decay will become
less rapid with larger values of c.
Consider now introducing two new variables√ into Equation 6.21. The first is the
undamped natural frequency of the system6 , 𝜔n = k∕m. The second is the damping ratio,
𝜁 = c∕ccr , which is defined to be the ratio of the actual damping in the system to the critical
damping.
Making these substitutions into Equation 6.21 leads to the following sequence of
equations,
c k
𝜆2 + 𝜆+ =0
m m
c ccr
𝜆2 + 𝜆 + 𝜔2n = 0
ccr m
√
4km
𝜆 +𝜁
2
𝜆 + 𝜔2n = 0
m
√
k
𝜆 + 2𝜁
2
𝜆 + 𝜔2n = 0
m
𝜆2 + 2𝜁𝜔n 𝜆 + 𝜔2n = 0. (6.23)
The eigenvalues of Equation 6.23, plotted in Figure 6.7, are,
√
𝜆 = −𝜁𝜔n ± 𝜔n 𝜁 2 − 1. (6.24)

6 The undamped natural frequency is found by determining the eigenvalues of the one degree of freedom
system without damping,
m̈x + kx = 0
The eigenvalues are the complex conjugate pair,
√
𝜆 = ± k∕m i
√
which give undamped oscillatory motion at a frequency 𝜔n = k∕m.

@Seismicisolation
@Seismicisolation
 6.5 Modal Damping 121

Figure 6.7 Calculating the damping Imag

ratio from the location of an
eigenvalue. ζωn

ωn√1 – ζ2
θ Real

Oscillations will occur when the eigenvalues of Equation 6.24 are a complex conjugate
pair. This requires that 𝜁 2 < 1. That is, the actual damping should be less than critical.
If this
√ is the case, then the real part of the eigenvalue is −𝜁𝜔n and the imaginary part is
±𝜔n 1 − 𝜁 2 . Notice that the term (𝜁 2 − 1) has been rewritten as (1 − 𝜁 2 ) because it is now
the magnitude of the imaginary part of the eigenvalue
√ and must therefore be positive so that
the square root yields a real value. That is, i = −1 has been factored out. The magnitude
of the eigenvalue (i.e. the distance from the eigenvalue to the origin in Figure 6.7) is,
√
|𝜆| = 𝜁 2 𝜔2n + 𝜔2n (1 − 𝜁 2 ) = 𝜔n (6.25)

and the damping ratio (refer to Figure 6.7) can be found from the trigonometric expression,
𝜁𝜔n
cos 𝜃 = = 𝜁. (6.26)
𝜔n
By percent damping of a mode, we mean the percentage of critical damping that is present.
Applying the trigonometric analysis just presented to systems with more than one degree
of freedom means that any oscillatory eigenvalue can be described by its percent damping.
For example, for the rigid rod on springs just described, the first mode has 11.4% damping
and the second has 17.4% damping7 .
The question now becomes – how many cycles of motion does it take for modes with
11.4% and 17.4% damping to disappear?

7 These values are calculated as follows. For the first mode,

2.2705
𝜁 = cos 𝜃 = √ = 0.11423 ≈ 11.4%
(2.2705)2 + (19.747)2

and, for the second mode,

2.1045
𝜁 = cos 𝜃 = √ = 0.17430 ≈ 17.4%.
(2.1045)2 + (11.889)2

@Seismicisolation
@Seismicisolation
122 6 Mode Shapes

ζ = 1% ζ = 5% ζ = 10%

ζ = 20% ζ = 30% ζ = 40%

Figure 6.8 Waveforms for different percent damping.

Figure 6.8 shows typical oscillatory behavior of a single degree of freedom system for
various levels of percent damping. Also shown on the figure are two horizontal lines repre-
senting ±10% of the initial amplitude. It is clear from the figure that oscillations in systems
with 30% to 40% damping disappear very rapidly, losing 90% of their amplitude in 1 to 1.5
cycles. With 20% critical damping the oscillation is essentially gone in 2 cycles. A system
with 10% critical damping will sustain oscillations for approximately 4 cycles using this
measure and those with 1% or 5% will go on for a very long time.
Designers of vehicle suspension systems typically look for modal damping in the 10%
to 20% range for the primary modes of a vehicle. Lower levels of damping than this mean
that oscillations will be sustained over too long a period for good ride quality. Higher levels
of damping will lead to larger shock loads in the system as rapidly changing forces (e.g.
from hitting a bump on the road) are transmitted directly to the vehicle by the dampers
rather than being attenuated by the springs. Even garage mechanics will check the state of
shock absorbers on a car by pushing down a corner of the vehicle and then counting the
number of oscillations before the motion stops. If it oscillates more than three times, the
shock absorbers need to be replaced.

Exercises

Descriptions of the systems referred to in the exercises are contained in Appendix A.

6.1 Find the two mode shapes for system 10 given that the applied force, F(t), is zero.

@Seismicisolation
@Seismicisolation
 Exercises 123

6.2 Consider system 13 for the special case where k=0 and d=𝓁. Show that the linearized
equations of motion about the equilibrium position, 𝜃0 = 180∘ and 𝜙0 = 0∘ , can be
written as
⎡ 17 m𝓁 2 1 m𝓁 2 ⎤ { } [ ]{ } { }
⎢ 3 3 ⎥ 𝛼̈ 3mg𝓁 0 𝛼 0
⎢ 1 1 ⎥ ̈
+ =
⎢ m𝓁 2 m𝓁 2 ⎥ 𝛽 0 0 𝛽 0
⎣ 3 3 ⎦
where 𝛼 and 𝛽 represent small variations in 𝜃 and 𝜙 respectively.
Find the eigenvalues and eigenvectors of this system and describe in words the two
natural modes.

6.3 Derive the equations of motion for system 15 using the horizontal and vertical posi-
tions of the mass with respect to point A as degrees of freedom. Find the equilibrium
position, linearize about it, and use the mode shapes to show that small horizontal
and vertical motions of the mass about equilibrium are independent of each other.

6.4 Find the three mode shapes for system 18 and sketch them.

6.5 A transit system operator is complaining that streetcars (see system 19) manufactured
by your company are excessively noisy. The operator has made measurements and
cites a peak noise level at 68 Hz as being the main problem. You are told that the
cause of the problem is a resonance in the suspension system of the streetcars. Your
manager feels that a doubling of the vertical secondary suspension stiffness (ks ) will
shift the offending frequency enough that it will no longer be a problem. Your task is
to provide analytical evidence to support this solution. Perform the following analysis
to do this.
(a) Derive the linearized differential equations of motion governing bounce (x) and
pitch (𝜃) of the truck frame. Use m for the truck frame mass, I for the truck frame
pitch moment of inertia, ks for the secondary stiffness where the truck frame is
connected to the car body, and kp for the primary stiffnesses that connect the
wheels to the truck frame. The car body is massive enough that it can be assumed
to have no motion and the wheels remain stationary on the rails. Assume that
damping is small enough to be neglected. The stable equilibrium state has both
bounce and pitch equal to zero.
The result should be,
[ ]{ } [ ]{ } { }
m 0 ẍ 2kp + ks kp (b − a) − ks c x 0
+ = .
0 I 𝜃̈ 2 2
kp (b − a) − ks c kp (a + b ) + ks c 2 𝜃 0
(b) Find the natural frequencies and mode shapes of the system using, as parameters:
m = 500 kg, I = 125 kg m2 , a = 1.6 m, b = 1.4 m, c = 0.1 m, kp = 5 × 106 N m−1 ,
and ks = 1 × 107 N m−1 .
(c) Which mode is causing the problem? Comment on the effectiveness of curing the
problem by doubling the secondary suspension stiffness as your manager sug-
gested rather than changing the primary stiffness values. Do not recalculate the
natural frequencies. Simply base your comments on the mode shapes.

@Seismicisolation
@Seismicisolation
124 6 Mode Shapes

(d) Add primary and secondary dampers in parallel with the primary and secondary
stiffnesses and select damping coefficients so that the two modes each have
approximately 15% damping.

6.6 Find the two mode shapes for system 20 and sketch them.

6.7 Find the stable equilibrium position for system 23, linearize the equations of motion
around the equilibrium, and find the three natural frequencies and mode shapes.

6.8 In Exercise 4.9, four equilibrium positions were found for the rod in system 12. Derive
an expression for the natural frequency of the rod for small motions around the most
physically realistic of the equilibrium positions.

6.9 A system with three degrees of freedom has an eigenvalue and corresponding eigen-
vector as follows:
⎧ −4.4153 × 10−4 − 9.2574 × 10−3 i⎫
⎪ ⎪
𝜆1 = −5.1405 + 107.78 i ; {X}1 = ⎨ −9.1116 × 10−4 + 3.0405 × 10−3 i⎬.
⎪ ⎪
⎩ +7.0807 × 10−4 − 2.6808 × 10−4 i⎭
Interpret this so as to give the damped natural frequency in Hz, the damping ratio in
percent, and a description of the mode shape including amplitudes and phase angles.

@Seismicisolation
@Seismicisolation
 125

Frequency Domain Analysis

Determining the response of a mechanical system to a disturbance that can be described

primarily by its frequency is often useful. For example, a sensitive piece of equipment may
be required to be installed in a power generation plant where the floor is subject to vibrations
at the frequency of the rotating machinery that generates the electricity. We must design
the new equipment for an environment where it will be disturbed by vibrations having this
known frequency but, often, an unknown amplitude. Another use of frequency domain
analysis is the case where we have measured the input to a system and have found it to
be apparently random. There are well known techniques for decomposing the input into
a series of harmonic functions at different frequencies. The task of the analyst is then to
predict how the system will respond to the various harmonic excitations and to construct
an overall predicted response.

7.1 Modeling Frequency Response

Frequency response calculations use the equations of motion in the linearized form for
small motions around a stable equilibrium state. We used the equilibrium states for stability
and mode shape analysis in Chapter 5. The difference here is that stability considers the
natural motions of the system with no externally applied forces whereas frequency response
analysis considers externally applied harmonic forces1 .
Consider the standard set of linearized equations of motion that were presented as
Equation 5.29. They are,
̇ + [K]{x} = {0}
[M]{̈x} + [C]{x} (7.1)
and they apply to motions of the system away from the equilibrium state without any
applied forces other than those which put the system into the equilibrium state. We modify
the equations by saying that the system is in the equilibrium state but is disturbed by a set
of harmonically varying external forces which can be written as,
{F(t)} = {F}ei 𝜔t = {F}(cos 𝜔t + i sin 𝜔t) (7.2)

1 As we will see in Chapter 8, only time domain simulations permit the use of the fully nonlinear
equations of motion.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
126 7 Frequency Domain Analysis

where {F} is a vector of constant force amplitudes and 𝜔 is a known forcing frequency. It
is important that we represent these forces as being harmonic at the forcing frequency and
that we use the complex exponential form of the harmonic function. Using only sin 𝜔t or
cos 𝜔t works if there is no damping in the system but the complex exponential form must
be used in the general case where damping and, as a result, phase angles are present.
The applied forces replace the zero vector on the right hand side of Equation 7.1 to give,
̇ + [K]{x} = {F}ei 𝜔t .
[M]{̈x} + [C]{x} (7.3)

As is for any set of linear differential equations, there are two components of the solution
to Equation 7.3. The first is the complementary solution to the homogeneous differential
equations (the same equations but with a zero vector on the right hand side). This solution
involves the motion of the system at its natural frequencies in response to specified initial
conditions. The second is the particular solution that is an oscillation at the forcing fre-
quency. The frequency domain analysis always assumes that the complementary solution
will decay with time because of damping in the system and that the particular solution will
give the steady state motions in response to the externally applied forces with which we are
concerned.
The particular solution is a time-varying vector {x(t)} which, along with its first two
derivatives, must vary with time as ei 𝜔t . The result can only be that {x(t)} = {X}ei 𝜔t where
{X} is a vector of complex constants representing the amplitudes and phases of the degrees
of freedom in response to the forcing function. Notice that the response will be at the forc-
ing frequency and not at any of the natural frequencies of the system, as determined from
the eigenvalue analysis, unless the forcing frequency is equal to a natural frequency.
We substitute
{x(t)} = {X}ei 𝜔t
̇
{x(t)} = i 𝜔{X}ei 𝜔t
{̈x(t)} = (i 𝜔)2 {X}ei 𝜔t = −𝜔2 {X}ei 𝜔t (7.4)

into Equation 7.3 to get,

(−𝜔2 [M] + i 𝜔[C] + [K]){X}ei 𝜔t = {F}ei 𝜔t . (7.5)

The ei 𝜔t terms in Equation 7.5 cancel out and a single, complex, coefficient matrix

[A(𝜔)] = (−𝜔2 [M] + i 𝜔[C] + [K]) (7.6)

can be defined to represent the group of matrices on the left hand side.
We are left with a simple set of linear algebraic equations

[A(𝜔)]{X} = {F}. (7.7)

Equation 7.7 is remarkably like the general form of the system of eigenvalue equations
presented as Equation 5.11 in Section 5.1 as,

[A(𝜆)]{X} = {0} (7.8)

where [A(𝜆)] = (𝜆2 [M] + 𝜆[C] + [K]) and characteristic values of 𝜆 = 𝜎 + i 𝜔, the eigenval-
ues, made the determinant of [A(𝜆)] equal to zero.

@Seismicisolation
@Seismicisolation
 7.1 Modeling Frequency Response 127

The difference is that, in the frequency response equations (Equation 7.7), the coeffi-
cient matrix, [A(𝜔)], is a function only of an imaginary variable, i 𝜔. As a consequence, the
determinant of [A(𝜔)] will never be zero as 𝜔 is varied since i 𝜔 ≠ 𝜎 + i 𝜔 in the general
case2 .
Given a forcing frequency 𝜔 and a forcing vector {F}, all of the terms in Equation 7.7 will
be known except for {X}. The problem then becomes one of solving a relatively simple set
of linear algebraic equations3 to find the response amplitudes.
Conceptually, although not computationally because of its inefficiency, it is often use-
ful to look at the solution of sets of linear algebraic equations that would result from an
application of Cramer’s Rule4 .
Cramer’s rule says that, given the set of N linear algebraic equations [C]{x} = {r}, xj , the
jth element of the vector of unknowns {x}, can be found from the relationship,
det[{C}1 {C}2 · · · {C}j−1 {r}{C}j+1 · · · {C}N ]
xj = (7.9)
det[C]
where {C}i is the ith column of [C]. That is, the right hand side of the set of equations is
substituted for the jth column of [C] and xj is found from the ratio of two determinants.
Applying this to Equation 7.7 in an attempt to find Xj yields,

det[{A(𝜔)}1 {A(𝜔)}2 · · · {A(𝜔)}j−1 {F}{A(𝜔)}j+1 · · · {A(𝜔)}N ]

Xj = . (7.10)
det[A(𝜔)]
The point of this is that, whatever value the numerator of Equation 7.10 might have,
the amplitude of any Xj depends strongly on the denominator of the equation, which will
become very nearly zero when the forcing frequency, 𝜔, is near a natural frequency of the
system. In fact, det[A(𝜔)] is equal to zero when the system is undamped and the forcing
frequency is equal to a natural frequency. This situation results in the prediction of infinite
response amplitudes when an undamped system is forced at one of its natural frequencies.
Consider, as an example, the system shown in Figure 7.1. The equations of motion are,
[ ]{ } [ ]{ }
m1 0 ẍ 1 c1 + c2 −c2 ẋ 1
+
0 m2 ẍ 2 −c2 c2 ẋ 2
[ ]{ } { }
k1 + k2 −k2 x1 F1 (t)
+ = . (7.11)
−k2 k2 x2 F2 (t)

If we let,
{ } { }
F1 (t) F1 𝜔t
= ei (7.12)
F2 (t) F2

2 There are, of course, cases without damping where 𝜎 = 0 and i 𝜔 may indeed be an eigenvalue of the
system causing the determinant to be zero. Such cases will be discussed later.
3 The equations are a set of linear algebraic equations with complex coefficients. The standard methods for
solving linear algebraic equations still apply but must be used with complex arithmetic.
4 Gabriel Cramer (1704–1752), a Swiss mathematician, published, in 1750, a treatise entitled Introduction
á l’analyse des lignes courbes algébraique where, in a chapter on the classification of curves, he presents the
now famous “Cramer’s rule”.

@Seismicisolation
@Seismicisolation
128 7 Frequency Domain Analysis

F2 Figure 7.1 A linear system with two masses.

x2
m2

c2 F1
k2

x1
m1

c1 k1

we get, after substitution and canceling of the ei 𝜔t terms,

[ ]{ } { }
−𝜔2 m1 + i 𝜔(c1 + c2 ) + (k1 + k2 ) −i 𝜔c2 − k2 X1 F1
= .
−i 𝜔c2 − k2 −𝜔2 m2 + i 𝜔c2 + k2 X2 F2
(7.13)
After rearranging the elements of the coefficient matrix, this becomes,
[ ]{ } { }
(−𝜔2 m1 + k1 + k2 ) + i 𝜔(c1 + c2 ) −k2 − i 𝜔c2 X1 F1
=
−k2 − i 𝜔c2 (−𝜔2 m2 + k2 ) + i 𝜔c2 X2 F2
(7.14)
which is in the standard form [A(𝜔)]{X} = {F} and the complex coefficients are obvious.
The determinant of the coefficient matrix is,
det[A(𝜔)] = [(−𝜔2 m1 + k1 + k2 ) + i 𝜔 (c1 + c2 )][(−𝜔2 m2 + k2 ) + i 𝜔c2 ]
− [−k2 − i 𝜔c2 ]2 (7.15)
and we can use Cramer’s rule to solve for X1 and X2 . The solution is,

F1 [(−𝜔2 m2 + k2 ) + i 𝜔c2 ] − F2 [−k2 − i 𝜔c2 ]

X1 =
[(−𝜔2 m1 + k1 + k2 ) + i 𝜔(c1 + c2 )][(−𝜔2 m2 + k2 ) + i 𝜔c2 ] − [−k2 − i 𝜔c2 ]2
(7.16)
and,

F2 [(−𝜔2 m1 + k1 + k2 ) + i 𝜔(c1 + c2 )] − F1 [−k2 − i 𝜔c2 ]

X2 = .
[(−𝜔2 m1 + k1 + k2 ) + i 𝜔(c1 + c2 )][(−𝜔2 m2 + k2 ) + i 𝜔c2 ] − [−k2 − i 𝜔c2 ]2
(7.17)

@Seismicisolation
@Seismicisolation
 7.1 Modeling Frequency Response 129

There is no elegant form of these relationships. The point in showing them is to indicate
that solutions can be found and that they are complex. The solutions, once determined, can
be written as,

X1 = a1 + ib1 ; X2 = a2 + ib2 (7.18)

or,
√ √
X1 = a21 + b21 ∠tan−1 (b1 ∕a1 ); X2 = a22 + b22 ∠tan−1 (b2 ∕a2 ) (7.19)

where a1 , b1 , a2 and b2 represent the numerical results of substituting parameter values into
the solutions.
Consider the case where the system has the parameter values given in Table 7.1. With
these values, an eigenvalue analysis gives two modes as follows.

● Mode 1: 0.797 Hz, 9.33% damping,

{ } { }
X1 0.6695 ∠4.9∘
= .
X2 1.0000 ∠0∘

● Mode 2: 2.713 Hz, 13.97% damping,

{ } { }
X1 1.0000 ∠0∘
= .
X2 0.3321 ∠ − 163.0∘

We then expect that if the system is forced at varying frequencies, we should see rela-
tively large response at forcing frequencies near the two natural frequencies: 0.797 Hz and
2.713 Hz. Figures 7.2–7.5 show the forced responses, both amplitudes and phases, as the
forces shown in Table 7.1 are applied at varying frequencies.
Mass 1 (the lower mass in Figure 7.1) shows increased response at both of the natu-
ral frequencies and has larger amplitude at the first natural frequency. Mass two is quite
responsive at the first natural frequency but shows a fairly weak response at the second

Table 7.1 Parameters for the linear system with two masses.

Parameter Value Units

m1 10 kg
m2 20 kg
c1 50 N s m−1
c2 5 N s m−1
k1 1000 N m−1
k2 1500 N m−1
F1 100 N
F2 10 N

@Seismicisolation
@Seismicisolation
130 7 Frequency Domain Analysis

0.7 Figure 7.2 Amplitude response of mass 1.

X1 Amplitude (m)

0.6
0.5
0.4
0.3
0.2
0.1
0.0
0 1 2 3 4 5 6 7 8
Frequency (Hz)

180 Figure 7.3 Phase response of mass 1.

120
ϕ1 (degrees)

60
0
–60
–120
–180
0 1 2 3 4 5 6 7 8
Frequency (Hz)

0.7 Figure 7.4 Amplitude response of mass 2.

X2 Amplitude (m)

0.6
0.5
0.4
0.3
0.2
0.1
0.0
0 1 2 3 4 5 6 7 8
Frequency (Hz)

180 Figure 7.5 Phase response of mass 2.

120
ϕ2 (degrees)

60
0
–60
–120
–180
0 1 2 3 4 5 6 7 8
Frequency (Hz)

@Seismicisolation
@Seismicisolation
 7.1 Modeling Frequency Response 131

frequency. Notice on Figure 7.2 that there is a frequency between 1 and 2 Hz where the
amplitude of the lower mass is reduced to zero5 .
The phase angle plots show that the two masses are approximately in-phase with each
other up to the point between 1 and 2 Hz where the lower mass stops moving. They both
have negative phase angles with respect to a positively applied force up to that point. At
higher frequencies, the upper mass moves to a positive phase angle and the lower mass
maintains a negative phase so that the two masses are approximately out-of-phase with
each other.
The observed phase relationship is very much in line with the predicted mode
shapes – in-phase at the lower natural frequency and out-of-phase at the higher.
The relative amplitudes at the two natural frequencies are also in line with the mode
shapes. Both masses have relatively large amplitudes in the mode shape corresponding to
the lower natural frequency with X2 being larger than X1 . Figures 7.2 and 7.4 show this
behavior clearly. The eigenvector at the higher natural frequency indicates that X1 should
be significantly larger than X2 and the figures indicate that this is so.
In fact, a close look at the predicted response at the natural frequencies shows very close
agreement with the mode shapes. The comparison is:

● Mode 1: 0.797 Hz, 9.33% damping,

{ } { }
X1 0.6695 ∠4.9∘
= .
X2 1.0000 ∠0∘

● Response at 0.80 Hz,

{ } { }
X1 0.67 ∠ = 0.37∘
= .
X2 1.0 ∠0∘

● Mode 2: 2.713 Hz, 13.97% damping,

{ } { }
X1 1.0000 ∠0∘
= .
X2 0.3321 ∠ − 163.0∘ .

● Response at 2.71 Hz,

{ } { }
X1 1.0 ∠0∘
= .
X2 0.36 ∠ − 188∘

It can be expected that any system with an externally applied harmonic force that is near
a natural frequency, will respond by moving approximately in the mode shape associated
with that frequency regardless of where the force is applied to the system.

5 The fact that the forced response of a body to which a harmonic force is applied can be reduced to zero
simply by having another mass elastically attached to it lies behind the theory of the well known and
widely used vibration absorber. For undamped systems, the frequency at which the motion of m1 is reduced
√
to zero is precisely the natural frequency of the system attached to m1 . That is, at the frequency k2 ∕m2 ,
the upper mass and spring system will have all of the response and the lower mass will √stop moving. In the
lightly damped case presented here, the motion of m1 is minimized at 1.45 Hz where k2 ∕m2 = 1.38 Hz.

@Seismicisolation
@Seismicisolation
132 7 Frequency Domain Analysis

7.2 Seismic Disturbances

Seismic disturbances are harmonic motions of the ground or, more particularly, the sup-
porting base of a system. We consider again the system we considered in Section 7.1 but
with the applied forces removed and replaced by a harmonic ground disturbance y (t) given
by

y (t) = Y ei 𝜔t (7.20)

where Y is the amplitude of the base motion and 𝜔 is the forcing frequency (see Figure 7.6).
The equations of motion are slightly modified from those of Equation 7.9 to become
[ ]{ } [ ]{ }
m1 0 ẍ 1 c1 + c2 −c2 ẋ 1
+
0 m2 ẍ 2 −c2 c2 ẋ 2
[ ]{ } { }
k1 + k2 −k2 x1 c1 ẏ + k1 y
+ = (7.21)
−k2 k2 x2 0

where we notice that the effect of the ground disturbance is to apply forces only to the lower
mass and that these forces pass through the elements connecting that mass to the base.
Since the disturbance is harmonic and at a known frequency, the response will also be
harmonic at the same frequency as pointed out in Equation 7.4. That is,
{ }
X1
{X(t)} = ei 𝜔t . (7.22)
X2

After substitution and simplification, the resulting algebraic equations are

[ ]{ }
−𝜔2 m1 + i 𝜔(c1 + c2 ) + (k1 + k2 ) −i 𝜔c2 − k2 X1
−i 𝜔c2 − k2 −𝜔2 m2 + i 𝜔c2 + k2 X2
{ }
i 𝜔c1 + k1
= Y. (7.23)
0

x2 Figure 7.6 Seismic disturbance of a system.

m2

c2 k2
x1
m1

c1 k1
y

@Seismicisolation
@Seismicisolation
 7.3 Power Spectral Density 133

Cramer’s rule yields

[ ]
i 𝜔c1 + k1 −i 𝜔c2 − k2
det
0 −𝜔2 m2 + i 𝜔c2 + k2
X1 = [ ] Y (7.24)
−𝜔2 m1 + i 𝜔(c1 + c2 ) + (k1 + k2 ) −i 𝜔c2 − k2
det
−i 𝜔c2 − k2 −𝜔2 m2 + i 𝜔c2 + k2
and
[ ]
−𝜔2 m1 + i 𝜔(c1 + c2 ) + (k1 + k2 ) i 𝜔c1 + k1
det
−i 𝜔c2 − k2 0
X2 = [ ] Y. (7.25)
−𝜔2 m 1 + i 𝜔(c1 + c2 ) + (k1 + k2 ) −i 𝜔c2 − k2
det
−i 𝜔c2 − k2 −𝜔2 m2 + i 𝜔c2 + k2

Equations 7.24 and 7.25 give the output response, X1 and X2 , as a function of the magni-
tude of the input, Y , and the forcing frequency, 𝜔. These equations can be written as,

X1 = T1 (𝜔)Y (7.26)

and

X2 = T2 (𝜔)Y (7.27)

where T1 (𝜔) and T2 (𝜔) are the complex transfer functions relating output, X1 and X2 , to
input magnitude, Y . The complex numbers carry both amplitude and phase information.
Often we simply want to know the amplitude relationship between output and input. In
that case we take the absolute values and write,

|X1 | = |T1 (𝜔)||Y | (7.28)

and

|X2 | = |T2 (𝜔)||Y |. (7.29)

7.3 Power Spectral Density

Consider the apparently random variable with zero mean value, x (t), shown in Figure 7.7.
Assume that we have sampled it and used the discrete Fourier transform6 to write it in the
form of Equation 9.3. That is,
∑
N
∑
N
x(t) = a0 + an cos (n𝜔0 t) + bn sin (n𝜔0 t). (7.30)
n=1 n=1

The term a0 represents the mean value of x (t). We can drop a0 in the following since we
are considering a variable with a mean value of zero.

6 Discrete Fourier transforms are covered in Chapter 9 Analysis of Experimental Data.

@Seismicisolation
@Seismicisolation
134 7 Frequency Domain Analysis

20

10

x(t) 0

–10

–20
0 2 4 6 8 10 12 14 16
time (s)

Figure 7.7 A measured variable x(t) plotted versus time.

Equation 7.30 can be rewritten in terms of amplitudes and phase angles 7 as,
∑
N
x (t) = An cos (n𝜔0 t + 𝜙n ). (7.31)
n=1
We now recognize that the random time signal has been decomposed into a finite number
of terms, each of which is a harmonic function with known amplitude and frequency of the
form,
f (t ) = A cos (𝜔t + 𝜙). (7.32)
The mean-square value of f (t) is,
T
2 1
f = x2 (t) dt (7.33)
T ∫0
where T = 2𝜋∕𝜔 is the period of the harmonic function.
Since f (t) repeats cyclically, the mean-square value of any cycle will be equal to the total
mean-square value of the function. The total mean-square value can therefore be calculated
from a single cycle of the wave and, since the phase angle 𝜙 does not affect the value, it can
be assumed to be zero (note that this is the same as shifting the integration limits to coincide
with the beginning and end of a single cycle of the wave).
A cycle is therefore represented by the range 0 ≤ 𝜔t ≤ 2𝜋 resulting in limits on the inte-
gration of 0 ≤ t ≤ (2𝜋∕𝜔). As a result, using Equation 7.32, we can write,
t=2𝜋∕𝜔
2 1
f = A2 cos2 𝜔t dt. (7.34)
(2𝜋∕𝜔) ∫t=0
We define a dummy variable u = 𝜔t and transform the integration by replacing t with
u∕𝜔 and dt with du∕𝜔 to get,
u=2𝜋
2A2
f = cos2 u du. (7.35)
2𝜋 ∫u=0
The integral can be looked up in standard tables of integrals or worked out using integra-
tion by parts to yield,
u=2𝜋 [ ]u=2𝜋
1 1
cos2 u du = u + sin 2u = 𝜋. (7.36)
∫u=0 2 4 u=0

7 See Equation 9.53

@Seismicisolation
@Seismicisolation
 7.3 Power Spectral Density 135

Substituting Equation 7.36 into Equation 7.35 gives the total mean-square value of a har-
monic function as,
2 A2
f = . (7.37)
2
Applying Equation 7.37 to each of the terms in Equation 7.31 and adding them together
gives the following expression for the total mean-square value of the signal.

2
∑
N
A2n
xT = . (7.38)
n=1
2

Figure 7.8 shows an example of a Fourier analysis of a signal that has generated five mag-
nitude components. The frequency increment referred to in the figure is f0 8 whereas it is
more convenient to write our equations in terms of 𝜔0 in rad s−1 . The five components of
the mean-square value are (A2n ∕2 for 𝜔 = n𝜔0 where n = 1, 5) and are shown in Figure 7.9.
Figure 7.10 shows how the components of the mean-square value can be added together
to arrive at the cumulative mean-square curve. At each increment of 𝜔0 , the new component
of the mean-square value is added to the value that has been accumulated from previous

Figure 7.8 DFT component amplitudes.

DFT amplitude

0 f0 2f0 3f0 4f0 5f0 6f0

Frequency

Figure 7.9 Component mean-square values.

Component mean-square

A22
2 A25
A21 2
value

2
A23
2 A24
2

0 f0 2f0 3f0 4f0 5f0 6f0

Frequency

8 You will see in Chapter 9 that, if we have a continuous time domain function x(t) and we wish to
approximate it in the range 0 ≤ t ≤ T by using a series of harmonic functions, we choose a base frequency
f0 that allows one complete cycle in the range and then add higher frequency components that are
multiples of f0 . Clearly, f0 in cycles per second will be such that there is one cycle in T seconds so that
f0 = 1∕T cycles per second or Hz. The base frequency in radians per second will then be 𝜔0 = 2𝜋∕T.

@Seismicisolation
@Seismicisolation
136 7 Frequency Domain Analysis

Figure 7.10 The cumulative mean-square

Cumulative mean-square

X2T
value.
A25
2
A24
value

A23
2
2
A22
2
A21
2
0 f0 2f0 3f0 4f0 5f0 6f0
Frequency

components. Eventually the cumulative mean-square curve arrives at the total mean-square
2
value for the signal xT as specified in Equation 7.38.
We know the total mean-square value of the signal from the time domain analysis (see
Equation 9.2)and the value derived from Equation 7.38 must agree with that value. This
fact shows that the terms in the summation (i.e. the values of A2n ∕2) must be strong func-
tions of the frequency increment 𝜔0 that was used to generate the discrete Fourier transform
in Equation 7.31. If we use a large value of 𝜔0 we would expect a small value of N would
cover the frequency range of interest and a few relatively large values of An would com-
2
bine to give xT . Conversely, a small value of 𝜔0 would require a large value of N to cover
the same frequency range and we would see many small values of An , again combining to
2
yield xT .
We consider the contribution that any term makes to the summation in Equation 7.38
2
as an incremental contribution to the total mean-square value and denote it by Δxn . That
is,
A2n 2
. Δxn = (7.39)
2
The incremental frequency between terms has been specified to be 𝜔0 . If we consider
the cumulative mean-square curve in Figure 7.10, it is clear that, at any given frequency
𝜔 = n𝜔0 , the slope of the cumulative mean-square curve can be approximated by the ratio
2
Δxn ∕𝜔0 .
We designate this ratio to be W (𝜔) and define it to be the power spectral density or PSD.
As the frequency increment approaches zero, the cumulative mean-square curve
approaches a smooth function and the PSD can be defined to be the derivative of the
function9 .
2 2
Δx dx
W(𝜔) = lim = . (7.40)
𝜔0 →0 𝜔 d𝜔
0

We can once again refer to the component mean-square values shown in Figure 7.9 and
to the cumulative mean-square curve of Figure 7.10 to see how they can be interpreted.

9 The precise definition of the derivative is,

2 2
Δx dx
lim =
Δ𝜔→0 Δ𝜔 d𝜔
but, since it is always the case in this analysis that Δ𝜔 = 𝜔0 , the notation of Equation 7.40 is acceptable.

@Seismicisolation
@Seismicisolation
 7.3 Power Spectral Density 137

Figure 7.11 Cumulative mean-square 50

Cumulative mean
curve for the example system.

square value
40
30
20
10
0
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)

First of all, the component mean-square values are representative of the level of the signal
that occurs at a known discrete frequency. That is, if there is a large amount of activity at
a frequency (e.g. 2𝜔0 in Figure 7.9) then the bar will be higher than it is at a frequency
without much content (e.g. at 4𝜔0 ). The corresponding frequencies in Figure 7.10 show
a steep slope or high PSD value approaching 2𝜔0 and a gentler slope or lower PSD value
approaching 4𝜔0 . The result is that the PSD curve will show peaks at frequencies where the
content is high (e.g. the natural frequencies of a system).
Figure 7.11 shows the cumulative mean-square curve for the apparently random signal
presented in Figure 7.7. Notice that the curve demonstrates a relatively linear growth with
frequency over the entire range except for two rapid changes below 10 Hz.
Consider first the linear behavior. If we imagine forming a PSD by taking the derivative
or slope of this function, we will end up with a small positive value which is nearly con-
stant. This is indicative of nearly the same amplitude at all frequencies. This is exactly the
characteristic of a random signal and the slope indicates that there is a randomness to the
signal in Figure 7.7.
The frequencies where the cumulative mean-square value changes rapidly are frequen-
cies where the slope (i.e. the PSD) will increase sharply and then decrease just as rapidly to
the previous value. These are peaks in the PSD curve and indicate frequencies where the
signal has a great deal of content. These could be natural frequencies of the system or they
could be forcing frequencies. Whichever is the case, they are invisible in Figure 9.1 but are
of great interest to the analyst.
Figure 7.12 shows the PSD for the apparently random signal being considered. It is shown
on a frequency range compressed from that of Figure 7.11 because nothing of interest hap-
pens after 10 Hz. The waviness seen near the frequency axis is representative of the less than
perfectly random background signal. The two peaks are clearly visible at 1 Hz and 5 Hz.

Figure 7.12 PSD for the example system. 80

70
60
50
PSD

40
30
20
10
0
0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz)

@Seismicisolation
@Seismicisolation
138 7 Frequency Domain Analysis

To this point we have demonstrated that it is possible to use the PSD to extract frequency
information about a signal. Of course we are also interested in amplitude information. If,
for example, the function x(t) being considered in this example is an acceleration measured
on a body then we know at which frequencies the accelerations are large. How large are
they?
To answer this question we turn to the commonly used RMS or root-mean-square value
of the signal. The name means exactly what it says, the root-mean-square is the square
root of the mean-square value of the signal. That is, for a single frequency component
such as that presented in Equation 7.32 (i.e. f (t) = A sin(𝜔t + 𝜙)), the mean-square value
2 2
(Equation 7.37) is f = A2 and the RMS value is,
√ √
2 A2 A
fRMS = f = = √ . (7.41)
2 2
Equation 7.41, while being commonly quoted, is actually not very useful if you have more
than a single frequency. In our case, we have performed a Fourier analysis and have a very
large number of frequency components. The RMS value we use will depend on the band-
width in which we are interested. That is, we need to decide on a frequency interval in which
to specify the RMS value of the signal. So-called wideband RMS covers the entire spectrum
of the signal and is a request for the square root of the total mean-square value of the signal.
Clearly this gives no frequency discrimination to the amplitude determination. In a case
such as the example we are considering we are likely to be more interested in a narrow
band RMS analysis that gives the RMS value as a function of frequency on a bandwidth of,
say, 1 Hz.
The RMS is calculated by integrating under the PSD curve between two specified fre-
quencies, 𝜔L and 𝜔R , and then taking the square root of the result. To see this, consider
Equation 7.40. It can be rearranged to give,
2
dx = W(𝜔)d𝜔. (7.42)
This can be integrated to yield,
2
xR 𝜔R
2
dx = W(𝜔)d𝜔 (7.43)
∫x2 ∫𝜔L
L

or,
𝜔R
2
Δx𝜔L →𝜔R = W(𝜔)d𝜔 (7.44)
∫𝜔L
2
and the RMS value between 𝜔L and 𝜔R is the square root of Δx𝜔L →𝜔R from Equation 7.44.
That is,
√
𝜔R
RMS𝜔L →𝜔R = W(𝜔)d𝜔. (7.45)
∫𝜔L

7.3.1 Units of the PSD

The units of PSD’s often seem arcane and random but they are in fact specified in a consis-
tent manner. Notice that the integral being considered in Equations 7.44 and 7.45 is the area

@Seismicisolation
@Seismicisolation
 7.3 Power Spectral Density 139

under the PSD curve between the frequency limits 𝜔L and 𝜔R . This integral must therefore
have units derived from those on the vertical axis multiplied by those on the horizontal
or frequency axis. In addition, the units of the area under the PSD curve must satisfy the
requirement that the square root of the units give an amplitude of the measured signal. That
is, the square root must result in an RMS amplitude.
Consider the case where the signal we have measured is taken from an accelerometer
and has been calibrated to specify the acceleration in g’s where a g is the acceleration
due to gravity. The measured acceleration has been transformed to the frequency domain
using an FFT with the frequency specified in Hertz10 (Hertz are abbreviated as Hz where 1
Hz = 1 cycle s−1 ).
In this case, the units of the PSD must be g2 ∕Hz. The area under the curve between √ two
frequencies then has units of (g2 /Hz) × (Hz) or g2 . The square root of the area is then g2
or g which is the desired amplitude unit.
There are many units for the PSD, all of them derived from the units used to measure
the quantities characterizing the system in question. For example, the surface roughness
of a road is best characterized by measuring the vertical irregularities of the surface and
then transforming them, not to the frequency domain, but as a function of wavelength.
Road irregularities are used as input disturbances to dynamic models of vehicles and the
frequency of the disturbance depends on the wavelength of the irregularity and the speed
of the vehicle. In the case of the US customary units11 , the amplitudes of the irregularities
will typically be in inches (abbreviated in) and the wavelengths will be in feet/cycle (abbre-
viated ft/cycle). The units of the PSD will then be in2 ⋅cycle/ft. If we repeated the analysis in
the SI system, the irregularities would be measured in meters (m) and the wavelengths in
meters/cycle (m/cycle) resulting in a PSD in m2 ⋅cycle/m. Note the unfortunate coincidence
that this can be simplified to m⋅cycle and the feeling for what dimensions a PSD must have
is lost.

7.3.2 Simulation using the PSD

There are instances where the input to a dynamic system cannot be expressed as a known
function of time. In these cases, the F (t) input that we use in our models cannot be exactly
specified even though we know it must exist. Consider for example, a vehicle driving over
uneven ground. The tires follow the profile of the ground and the chassis responds to the
up-and-down motion of the tires through the action of the suspension. The forcing input in
this case will come from the time varying normal forces acting between the ground and the
tires. These forces will be functions of the speed, V, of the vehicle coupled with the profile
over which the vehicle is being driven. If, for example, the ground has a purely sinusoidal
profile with amplitude, Y , and wavelength, 𝜆, each tire will experience a seismic distur-
bance12 exactly like that described in Section 7.2.

10 Heinrich Hertz (1857–1894), a German physicist, proved the existence of the electromagnetic waves
predicted theoretically by Maxwell. In recognition this, the unit of frequency, one cycle per second, is
named the Hertz.
11 US customary units are still widely used in the United States and cannot be dismissed as being
irrelevant. (See Appendix C)
12 The amplitude of the seismic input will be Y and the forcing frequency, 𝜔, can be calculated from the
period of the input. That is, the vehicle will travel over one wavelength of the disturbance in the period, T,

@Seismicisolation
@Seismicisolation
140 7 Frequency Domain Analysis

It is very unlikely that the profile of the ground will ever be a simple sinusoid but, using
the Fourier series again, we can fit the random looking profile with a series of sinusoidal
functions that, when superimposed, model it very closely. Then, given a vehicle speed, we
can calculate the response to each of the frequencies that the vehicle encounters while trav-
eling over the profile. The superposition of all of these responses will then give the total
vehicle response.
A typical problem of this type is addressed by firstly obtaining a PSD of the input distur-
bance. These inputs are usually expressed as PSD amplitude as a function of wavelength (or,
sometimes, as a function of one over the wavelength). You can develop them by measuring
the profile of your specific input and producing a PSD or you can easily find published PSDs
for various systems such as rails, roads, sea states, etc.
The simulation procedure here is as follows: 1. Derive the linearized equations of motion
about a stable equilibrium state using degrees of freedom including those that will act as the
seismic inputs. 2. Assume harmonic variation of the seismic inputs with known amplitude
and frequency. 3. Assume that all DOFs move with the same frequency as the input but
have different amplitudes and derive the transfer functions for the degrees of freedom that
are not seismic inputs. 4. Calculate the PSDs of the output variables of interest. 5. Integrate
the output PSDs over frequency bandwidths of interest and calculate the RMS values for
the bandwidths.
The transfer functions we use are of the type we derived in Equations 7.26 and 7.27 in
Section 7.2. That is, the output, Xout , is related to the input, Yin , by,

Xout = T (𝜔) Yin . (7.46)

Since the PSD represents amplitude only, we take the absolute values of the terms in
Equation 7.46, and then square both sides of the equation, divide each side by 2𝜔0 and take
the limit as 𝜔0 goes to zero and find,
( ) ( )
|Xout |2 |Yin |2
lim = |T (𝜔)| 2
lim . (7.47)
𝜔0 →0 2𝜔 𝜔0 →0 2𝜔
0 0

Comparing this result to Equation 7.40 shows that the PSD of the output appears on the
left hand side and the PSD of the input appears on the right hand side so we can write,

PSDout = |T (𝜔)|2 PSDin . (7.48)

As an example, consider the case where we are looking at the response of a vehicle as it
travels over a terrain for which we know the PSD as a function of wavelength, 𝜆. The first
step is to derive the equations of motion, linearized about an equilibrium state where the
vehicle is sitting on level ground. The wheels of the vehicle, Nw in number, are given vertical
degrees of freedom, {xw }. The remainder of the vehicle is given Nv degrees of freedom.
These are a set of translations and rotations that the analyst decides are able to describe
fully the motion of the vehicle relative to the wheels. They are contained in the vector {xv }.

giving VT = 𝜆 and the frequency will be such that 𝜔T = 2𝜋. Equating T from each of these expressions
gives a forcing frequency 𝜔 = 2𝜋V∕𝜆.

@Seismicisolation
@Seismicisolation
 7.3 Power Spectral Density 141

The equations of motion, in partitioned form13 , are,

Mvv Mvw ẍ v Cvv Cvw ẋ v

+
Mwv Mww ẍ w Cwv Cww ẋ w
(7.49)
Kvv Kvw xv
+ = .
Kwv Kww xw Fw

The partitioning is shown by the dashed lines in Equation 7.49. The two sub-matrices on
the diagonal are square matrices. The two off-diagonal matrices are rectangular. Consider
the stiffness matrix as an example. The overall stiffness matrix is square and is dimensioned
(Nv + Nw ) × (Nv + Nw ). The sub-matrix Kvv is Nv × Nv . Kww is Nw × Nw . The upper right
sub-matrix, Kvw , is Nv × Nw and, on the lower left, Kwv is Nw × Nv .
The right hand side of the equations of motion contains the externally applied forces
acting on the system. Since we are considering small motions about equilibrium, we see,
in this vector, only those forces that change as we move away from equilibrium. This is the
reason for the first Nv elements being zero. The only external forces acting on the vehicle
degrees of freedom are gravitational forces and, since they are constant, they are canceled
out by constant suspension forces. The wheel degrees of freedom see external forces, {Fw }.
These are forces of interaction with the ground that vary as the vehicle traverses the profile.
The total external force acting on each wheel will be the static normal load in equilibrium
plus the dynamic force found from the solution of the equations of motion.
The partitioned equations of motion can be written as two separate sets of differential
equations. The first is,

[Mvv ]{ẍ v } + [Cvv ]{ẋ v } + [Kvv ]{xv }

= −[Mvw ]{ẍ w } − [Cvw ]{ẋ w } − [Kvw ]{xw } (7.50)

where the wheel degrees of freedom have been taken to the right hand side because they
represent the seismic forcing input.
The second set is,

[Mwv ]{ẍ v } + [Mww ]{ẍ w }

+ [Cwv ]{ẋ v } + [Cww ]{ẋ w }
+ [Kwv ]{xv } + [Kww ]{xw } = {Fw }. (7.51)

Equation 7.51 is useful for finding the dynamic loads that the wheels experience as the
profile is traversed. This is excellent structural design information but is not always the focus
of these simulations. At the early stages of design, engineers are often more interested in
the vehicle ride quality, a quantity that can be assessed from the accelerations of the vehicle
found by solving Equation 7.50.

13 In this example, the wheel degrees of freedom appear as the final Nw elements of the vector containing
the degrees of freedom. They don’t necessarily have to be ordered like this but the analyst has to be able to
keep track of which degrees of freedom they are because, as you will see, they are treated differently.
Keeping track of the wheel DOFs is an important bookkeeping exercise.

@Seismicisolation
@Seismicisolation
142 7 Frequency Domain Analysis

The total solution will come from the superposition of solutions to harmonic inputs. We
have the input PSD as a function of wavelength and we generate solutions for a large num-
ber of wavelengths covering the input data. Given a constant forward speed, V, for the
vehicle and choosing a wavelength, 𝜆, we can calculate the forcing frequency, 𝜔 = 2𝜋V∕𝜆
(see footnote 12), and write the wheel vertical displacement as,
{xw } = A{ei(𝜔t+𝜙w ) } = A{ei𝜙w }ei𝜔t (7.52)
where A is the amplitude of the wave and 𝜙w is a phase angle that accounts for the distance
between wheels. That is, if two wheels are spaced exactly a multiple of one wavelength
apart, their input will be exactly in phase and the wheels will go up and down together. If
we take the leading wheel to have zero phase, then a wheel a distance, 𝓁w , behind it will
have a phase angle, 𝜙w = −(𝓁w ∕𝜆) (2𝜋). The vector, {ei𝜙w }, is a constant vector of length, Nw .
Differentiating gives the wheel velocities,
{ẋ w } = i𝜔A{ei𝜙w }ei𝜔t (7.53)
and, differentiating again, gives the wheel accelerations,
{ẍ w } = −𝜔2 A{ei𝜙w }ei𝜔t . (7.54)
The steady-state response of the vehicle degrees of freedom will be harmonic at the forc-
ing frequency, giving,
{xv } = {Xv }ei𝜔t . (7.55)
Differentiating Equation 7.55 twice and substituting into the equations of motion yields,
[−𝜔2 Mvv + i𝜔Cvv + Kvv ]{Xv }ei𝜔t =
− [−𝜔2 Mvw + i𝜔Cvw + Kvw ]A{ei𝜙w }ei𝜔t . (7.56)
Finally, dividing throughout by the input amplitude, A, and solving gives the transfer
functions for all the degrees of freedom of the vehicle (see Equations 7.26 and 7.27). They
appear in a vector, {T (𝜔)}, which is defined by,
( )
1
{T (𝜔)} = {Xv }
A
= −[−𝜔2 Mvv + i𝜔Cvv + Kvv ]−1 [−𝜔2 Mvw + i𝜔Cvw + Kvw ]{ei𝜙w }. (7.57)
We are often interested in predicting accelerations rather than displacements. In this
case, the same procedure applies except that the transfer functions used will be those
associated with acceleration rather than displacement. Given our assumption of har-
monic motion, where acceleration is always −𝜔2 multiplied by displacement, we can use
acceleration transfer functions,
{Tacc (𝜔)} = −𝜔2 {T (𝜔)}. (7.58)
Furthermore, given our linearity assumptions, we can create transfer functions for points
away from where degrees of freedom are defined simply by producing an expression for the
displacement of a point using the defined degrees of freedom and the dimensions that locate
the point of interest with respect to the points where degrees of freedom are defined. This
is completely analogous to defining linear spring displacements as we have done so many
times before.

@Seismicisolation
@Seismicisolation
 Exercises 143

Once we have the desired transfer function, we simply use Equation 7.48 to calculate the
output PSD for a range of frequencies of our choice. The RMS values of the output can be
found by integrating the output PSD over specified bandwidths.

Exercises

Descriptions of the systems referred to in the exercises are contained in Appendix A.

7.1 Consider system 3 for the case where the parameters for the vehicle are: mass
= 30.0 kg; pitch moment of inertia about G =20.0 kg m2 ; a =0.5 m; b =1.0 m;
k =3000 N m−1 .
Find the transfer functions for vertical motion, x, and pitch rotation, 𝜃, as the vehicle
travels at V = 10.0 m s−1 over sinusoidal terrain having a wavelength 𝜆 = 1.0 m.

7.2 The power spectral density curve for a measured signal, covering the 90 to 110 Hz
frequency band, can be approximated by the quadratic,

PSD = −f 2 + 200f − 9800

where the PSD has units of (m s−2 )2 Hz−1 and f is the frequency in Hz.
Calculate the RMS value of this signal for the 90 to 110 Hz frequency band. Using the
units of the calculated RMS value, explain what type of signal was originally mea-
sured.

7.3 Consider system 10 (see Exercise 3.7 for the equations of motion) for the case where
there is, initially, no applied force and the system is in equilibrium with both x and
𝜃 equal to zero. A harmonic force, F (t) = F cos 𝜔t, is then applied to the rod in the
direction shown in the figure.
a) Show that the steady state response of the system will be,
{ }
X
cos 𝜔t
Θ

where,

kd2 F
X=
(−m𝜔2 + 2k)(−md2 𝜔2 + kd2 ) − (−kd)2

and,

Fd(−m𝜔2 + 2k)
Θ=
(−m𝜔2 + 2k)(−md2 𝜔2 + kd2 ) − (−kd)2

b) At what forcing frequency, 𝜔, does the amplitude of 𝜃 go to zero? This is an example

of the vibration absorber described in footnote 5.

@Seismicisolation
@Seismicisolation
144 7 Frequency Domain Analysis

7.4 The transit vehicle of system 19 runs on a track with a surface displacement PSD that
can be described by
( )6
(2 × 10−10 ) 𝜆1
PSD = ( )3 ( )6
1 + (1 × 10−2 ) 𝜆1 + (1 × 10−7 ) 𝜆1

where 𝜆 is the wavelength in m and the PSD is in m2 /(cycles/m). Use the parameter
values from Exercise 6.5(b) and 6.5(d), augmented with a wheel mass of mw = 1000 kg,
to perform the following analyses.
a) Extend the model of Exercise 6.5 by adding vertical degrees of freedom to the
wheels to get a four degree of freedom model.
b) Use the methods of Subsection 7.3.2 to derive, numerically, the transfer function
for vertical acceleration of the center of mass. Use a vehicle speed of V = 50 km h−1
and a rail surface wavelength of 𝜆 = 0.50 m.
c) Assessment of ride quality in vehicles is often based on RMS accelerations plotted
versus 1/3 octave frequency bands14 . To predict ride quality for the vehicle being
considered here would require a dynamic model of the full vehicle and we don’t
have that. We can, however, use the simple model we have to predict the RMS
accelerations of the center of mass of the truck frame. Make that prediction for 1/3
octave bands with center frequencies spanning the range from 1 Hz to 80 Hz. Use
a vehicle speed of 50 km h−1 .
Note: This will require that you integrate the PSDs over the bandwidths in order to
get the RMS accelerations. This will have to be done numerically. At a minimum,
you can calculate the PSDs at the bandwidth boundaries and use the trapazoidal
rule. To get better accuracy, you will need to calculate the PSDs at more frequencies
and use a better numerical integration method.

14 An octave is defined as a doubling of frequency. That is, starting from a center frequency, fo , the octaves
are 2 × fo = 2fo , 2 × 2 × fo = 4fo , and so on. For any given center frequency, fc , the resulting bandwidth
[ ( ) ( ) ]
−1 1
spans the frequencies from 2 2 × fc to 2 2 × fc . In dynamics and vibrations, it is common to work
( ) ( ) ( )
1 1 1
with 1/3 octave bands for which the center frequencies are defined as fo , 2 3 × fo , 2 3 ×2 3 × fo , etc.
[ ( ) ( ) ]
−1 1
The resulting bandwidth around center frequency, fc , is 2 6 × fc to 2 6 × fc .

@Seismicisolation
@Seismicisolation
 145

Time Domain Solutions

The only solution technique that can include all of the nonlinearities in the equations of
motion is one which solves the governing differential equations in the time domain. That
is, powerful numerical integration techniques work directly with a first-order differential
equation representation of the equations of motion, including all of their nonlinear terms,
and solve them with respect to time.
The solutions generated in this way are known as simulations of the system. They are
approximations to what would come from experimental measurements on the real system.
There are a few issues to keep in mind when performing simulations.
● The approximation to reality is no better than the data used in the simulation. Detailed
dynamic models with very accurate parameter values provide the best simulations.
● The data generated from a time domain simulation have as much value in design as an
experiment does. You can see how a system behaves but no natural understanding of
the underlying mechanics arises from the simulation. Compare this to a linear stability
analysis where the mode shapes give the analyst a “feeling” for what is happening and
therefore some idea of how to solve the problem. If you don’t like the dynamic behavior
of a system you are simulating, you will likely have to resort to a trial and error variation
of parameters in order to improve the response.
● One advantage of simulations in software is that it is much cheaper than building and
testing prototypes. A good simulation will lead to early design iterations that avoid much
of what typically goes wrong with early prototypes. Any system that goes to market will
need prototypes built and tested at some stage but the total number required can be min-
imized by using a good simulation as a starting point.
● Test results can be used to validate simulations so that the designers know they can rely
on the simulations to make accurate predictions of how a system’s dynamic response will
change when system parameters are varied away from their nominal values. In the final
design stages, simulations become powerful design tools.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
146 8 Time Domain Solutions

8.1 Getting the Equations of Motion Ready for Time Domain

Simulation

The set of N simultaneous nonlinear differential equations of motion, Equation 3.88, from
the end of Chapter 3 is the starting point for time domain simulations. It is,
̇ t)]{q}
[M(q, q, ̈ = {f (q, q,
̇ t)}. (8.1)

It is only in the case of time domain simulations that we are able to keep all of the non-
linear terms and use the equations of motion without any simplifying assumptions such as
the small motions that allow a linearization of the equations. The solutions are generated
by solving the set of ordinary differential equations, Equation 8.1, with respect to time. The
solution requires a specified set of initial conditions for both displacement and velocity of
all the degrees of freedom.
Standard digital computer solution methods (commonly, although erroneously, referred
to as “integrators”) require that the differential equations first be put in first-order form.
This is completely analogous to the standard eigenvalue form discussed in Section 5.1.
Equation 8.1 can be transformed into an equivalent first-order set as follows.

● Define a new set of N variables for the displacements

{x1 } = {q}. (8.2)

● Define a second set of new variables for the rates of change of the displacements
̇
{x2 } = {q}. (8.3)

● From the preceding two definitions, a new set of N first-order differential equations has
been created. These are,

{ẋ 1 } = {x2 }. (8.4)

● With straightforward substitutions, the nonlinear differential equations given in

Equation 8.1 can be written in terms of the new variables as,

[M(x1 , x2 , t)]{ẋ 2 } = {f (x1 , x2 , t)} (8.5)

or, solving for {ẋ 2 },

{ẋ 2 } = [M(x1 , x2 , t)]−1 {f (x1 , x2 , t)}. (8.6)

● Equations 8.4 and 8.6 can then be written together as a set of 2N first-order differential
equations,
{ } { }
ẋ 1 x2
= (8.7)
ẋ 2 [M(x , x , t)]−1 {f (x , x , t)}
1 2 1 2

with user specified initial conditions,

{ } { } { }
x1 (0) q(0) displacements at t = 0
= = .
x2 (0) ̇
q(0) velocities at t = 0

@Seismicisolation
@Seismicisolation
 8.2 A Time Domain Example 147

Considering Equation 8.7, it is clear that the derivatives (i.e. rates of change) of the dis-
placements ({ẋ 1 }) and of the velocities ({ẋ 2 }) can be calculated at any time t if the numerical
values of the displacements ({x1 }) and the velocities {x2 }) are known at that time. That is,
only the displacements and velocities appear on the right hand side of Equation 8.7 so that
the rates of change on the left hand side of the equations can be easily calculated. Notice, in
particular, that specifying the initial conditions allows the calculation of the rates of change
at time t = 0.

8.2 A Time Domain Example

We first look at an example of the type of solution of the equations of motion that we are
seeking here and later discuss the numerical methods used to generate the solution.
Consider the simple pendulum shown in Figure 8.1. It is stationary (i.e. has zero veloc-
ity) and is hanging vertically downward at time t = 0. A horizontal force, F(t), is suddenly
applied to the pendulum bob at time t = 0 and is maintained at a constant magnitude and
orientation thereafter. We seek to explore how the motion of the pendulum (measured by
the angle 𝜃) varies with time for t ≥ 0.
The equation of motion for this system is easily generated using Lagrange’s Equation as
follows.
The kinetic and potential (datum at the connection to ground) energy expressions are,
1 ̇ 2 and U = −mg𝓁 cos 𝜃.
T = m(𝓁 𝜃)
2
Taking the necessary partial derivatives and recognizing that the generalized force asso-
ciated with the angle 𝜃 is,
Q𝜃 = F(t)𝓁 cos 𝜃
leads to the single differential equation of motion,
m𝓁 2 𝜃̈ + mg𝓁 sin 𝜃 = F(t)𝓁 cos 𝜃.
Following the procedure established previously in this section for reorganizing the
equations of motion for time domain solutions, we define new variables,
x1 = 𝜃 and x2 = 𝜃̇

Figure 8.1 The pendulum.

ℓ
θ

m F(t)

mg

@Seismicisolation
@Seismicisolation
148 8 Time Domain Solutions

and differentiate them to get,

ẋ1 = 𝜃̇ ̈
and ẋ2 = 𝜃.
Then we recognize that ẋ1 is identically equal to x2 as stated in Equation 8.4. This gives a
simple first-order differential equation,
ẋ1 = x2 .
We then substitute the new variables into the differential equation of motion to find,
m𝓁 2 ẋ2 + mg𝓁 sin x1 = F(t)𝓁 cos x1
from which we can solve for ẋ2 as,
g F(t)
ẋ2 = −
sin x1 + cos x1 .
𝓁 m𝓁
The two first-order differential equations can then be written together as,

{ } ⎧ x2 ⎫
ẋ 1 ⎪ ⎪
=⎨ g F(t) ⎬
ẋ 2 ⎪ − 𝓁 sin x1 + m𝓁 cos x1 ⎪
⎩ ⎭
̇
with initial conditions where 𝜃(0) = 0 (i.e. hanging straight down) and 𝜃(0) = 0 (i.e. motion-
less) at t = 0. In terms of the new variables, the initial conditions become,
{ } { }
x1 (0) 0
= .
x2 (0) 0
After setting up the equations of motion in this form, we proceed to a numerical solution.
For the purposes of this example we will let the time varying force, F(t), start at a value
equal to the weight of the pendulum bob (F(0) = mg at t = 0), then have it ramp down (i.e.
decrease linearly) to F(1) = 0 at t = 1 s, and hold it steady at F(t) = 0 thereafter. This can
be stated as,
F(t) = mg (1 − t) for 0 ≤ t ≤ 1 and F(t) = 0 for t > 1.
The predicted results are shown in Figure 8.2. They were generated using the variable
time-step Runge–Kutta–Fehlberg solution routine. The numerical solution routines for
ordinary differential equations, such as these equations of motion, are discussed next and
then we will return to a description of the procedure for solving the equations.

80 10 Figure 8.2 Predicted results for

θ the sample time domain
8 simulation.
40
6
F(t) (N)
θ (deg)

0 4
F(t) 2
–40
0
–80 –2
0 1 2 3 4 5 6 7 8 9 10
time (s)

@Seismicisolation
@Seismicisolation
 8.4 Euler Integration 149

8.3 Numerical Schemes for Solving the Equations of Motion

Numerical solution schemes for ordinary differential equations use an extrapolation

approach that says – “I know where I am now and I know how quickly things are changing
with respect to time so I can estimate where I will be in the future”. This statement is not
mathematical at all but, if you consider it, you will see quickly that the potential for making
errors in estimating “where I will be” is large. The statement assumes that derivatives
remain constant as you extrapolate into the future. If this is true, one can make arbitrarily
large steps into the future with zero error.
We can easily show that this is not the case for dynamic systems where there are accel-
erations. Consider the case where the derivative of the displacement (i.e. the velocity) is
constant. Say that q(t) is the displacement and that q(t)̇ = c where c is a known constant.
Then we can start from an initial displacement q(0) and make predictions about the value of
the displacement at any time in the future from the simple expression, q(t) = q(0) + c t. This
is mathematically correct but not applicable to dynamic systems since a constant velocity
requires zero acceleration. If we differentiate the original expression, q̇ = c, we find that,
q̈ = 0. It immediately becomes clear that we are not working with a system of interest in
dynamics since the acceleration is zero.
The fact is, systems with non-zero accelerations do not have solutions that can be extrap-
olated along straight lines indefinitely. We still rely on extrapolation methods but must be
very concerned about the length of time over which we extrapolate because we know that
we are trying to follow curved lines even in the simplest case and that we can only extrap-
olate over small time steps without introducing significant errors.

8.4 Euler Integration

The simplest integration scheme is one developed by Leonhard Euler. Extrapolation is often
based on the Taylor series expansion of a function and Euler’s method is so defined.
We use the following expression for a one-dimensional Taylor series in time
( ) ( )
dx 1 d2 x
x(t0 + h) = x(t0 ) + h+ h2
dt t0 2! dt2 t0
( ) ( )
1 d3 x 1 dn x
+ 3
h3 + · · · + n
hn + · · · (8.8)
3! dt t0 n! dt t0
where we see that the value of the function x(t) at time t = t0 + h is equal to the value at
t = t0 plus extrapolation terms depending on the derivatives of x(t), all evaluated at time
t = t0 , and the time step h. The Taylor series expansion of a function is an infinite series
where you can see the high order terms going to zero as n → ∞. In most cases, successive
terms quickly become negligible in size compared to the sum of the preceding terms before
n gets anywhere near ∞. Lagrange developed a theorem that shows that the Taylor series for
a function can be truncated at any value of n without loss of accuracy if the nth derivative
term is evaluated at a time t = 𝜏 where t0 ≤ 𝜏 ≤ t0 + h. This allows the Taylor series to be
written as an exact series with a finite number of terms even though the actual value of 𝜏

@Seismicisolation
@Seismicisolation
150 8 Time Domain Solutions

cannot be determined,
)
( ( )
dx 1 d2 x
x(t0 + h) = x(t0 ) + h+ h2
dt t0 2! dt2 t0
( ) ( )
1 d3 x 1 dn x
+ h3
+ · · · + n
hn . (8.9)
3! dt3 t0 n! dt 𝜏

In developing integration schemes based on the Taylor series, we will be truncating the
series and losing all of the terms after the last one we wish to save. The terms discarded dur-
ing truncation will introduce an error since they are not zero. This is called the truncation
error and we wish to keep it as small as possible. Clearly the error will be a function of the
time step raised to the power n (i.e. hn ) as can be seen by considering retaining the terms up
to and including the hn−1 term in Equation 8.9 and discarding the last term. The truncation
error introduced by doing this is 𝜖T where,
( )
1 dn x
𝜖T = n
hn . (8.10)
n! dt 𝜏
From Equation 8.10, it can be seen that 𝜖T decreases as n increases. The magnitude of 𝜖T
is also sensitive to the magnitude n
( n of
) the time step h in that, if h is much less than one, h is
d x
very small. The magnitude of dtn is finite but indeterminate. Overall, we can control the
𝜏
magnitude of the truncation error for the integration procedure by choosing the values of n
and h. The goal is to make the truncation error as small as possible so that the integration
scheme approximates the solution to the differential equation very closely.
We now get to the Euler numerical integration scheme. First we define a standard differ-
ential equation that we are trying to solve as,
dx
= f (x, t) with initial condition x(0) = x0 . (8.11)
dt
Euler suggested truncating the Taylor series immediately after the term that is linear in
h. That is, we approximate the function using,
( ) ( )
dx 1 d2 x
x(t0 + h) = x(t0 ) + h+ h2 (8.12)
dt t0 2! dt2 𝜏
which it has become common to write as,
( )
dx
x(t0 + h) = x(t0 ) + h + 𝒪(h2 ) (8.13)
dt t0
where 𝒪(h2 ) shows that the magnitude of the truncation error is on the order of h2 (i.e. the
magnitude varies according to the square of the time step).
To use the approximation given in Equation 8.13 to solve the differential equation
(Equation 8.11) we need to have an approximation for the first derivative. We can use
Equation 8.13 to write,
( ) x(t0 + h) − x(t0 )
dx
= + 𝒪(h) (8.14)
dt t0 h
where the truncation error on the approximation to the derivative is 𝒪(h) because of the
division by h that took place during the reorganization of Equation 8.13 to get Equation 8.14.

@Seismicisolation
@Seismicisolation
 8.5 An Example Using the Euler Integrator 151

x(t+h)

dx
dt t
x
x(t)

x(t) t t+h

Figure 8.3 Graphical representation of Euler’s method.

We now truncate Equation 8.14 and substitute the resulting approximate derivative into
the differential equation from Equation 8.11 for t = t0 to get,
x(t0 + h) − x(t0 )
≈ f (x(t0 ), t0 ). (8.15)
h
Rearranging this gives the following extrapolation formula for the Euler integrator
x(t0 + h) ≈ x(t0 ) + hf (x(t0 ), t0 ) (8.16)
where we recognize that the truncation error introduced during the extrapolation using this
formula is 𝒪(h2 ) per time step. If we use the Euler method to propagate a solution from an
initial time (t = 0) to a final time (t = tf ), we will need to take N time steps where tf = Nh.
From this we can see that N = tf ∕h and that the overall truncation error at tf will be,
tf
N × 𝒪(h2 ) = × 𝒪(h2 ) = 𝒪(h).
h
We therefore categorize the Euler integrator as an order-h method. Note that the order
of the total truncation error is the same as that of the local truncation error on the approx-
imation to the derivative (see Equation 8.14 for the Euler method). We will later discuss
integrators with higher order truncation error.
Figure 8.3 gives a graphical explanation for Euler’s method. The figure shows that the
solution to the differential equation has been propagated from the initial condition out to
time t. The derivative, f (x, t), is the local slope, dx
dt
, at time t. The extrapolation is enlarged in
the circle and shows that the method simply “slides up the slope” over the time step.

8.5 An Example Using the Euler Integrator

We consider here a simple differential equation where f (x, t) in Equation 8.11 is simply
equal to the function x(t) itself and the initial condition is that x(0) = 1. This can be written

@Seismicisolation
@Seismicisolation
152 8 Time Domain Solutions

as,

dx
= x with initial condition x(0) = 1. (8.17)
dt

Equation 8.17 is a well-known differential equation and the exact solution is the expo-
nential function. That is,

x(t) = et . (8.18)

We now use the Euler integrator to generate an approximate solution to the differential
equation for a time step of h. We will later compare the approximate solution to the exact
solution so we can see the effect of the truncation error.
This differential equation is simple enough to allow the approximate solution to be gen-
erated in a tabular format (see Table 8.1). We construct the table by first filling in a col-
umn with the time in steps of h. Beside this we write the value of x(t) at each time noting
that the initial value in this column is given by the initial condition in Equation 8.17 but
that subsequent entries in this column arise from extrapolation using the Euler integrator
(Equation 8.16). The third column contains the derivative f (x, t) calculated from the differ-
ential equation using the values of x and t in the same row of the table. The fourth column
contains the approximate value of x(t + h) calculated using the Euler integrator. This value
is then carried down to become the next value of x(t) in column two. The process is repeated
from there as the approximate solution to the differential equation is propagated forward
in time. After a few rows are filled in, the pattern becomes obvious and we can write the
approximate solution at the final simulated time ts = Nh to complete the table.
The exact solution at t = Nh is x(Nh) = eNh and the approximate solution is
x(Nh) ≈ (1 + h)N . To see how the two solutions differ we compare their infinite series
representations. The exact solution can be written as,

∑ (Nh)i
i=∞
(Nh)2 (Nh)3
x(Nh) = eNh = = 1 + Nh + + +··· . (8.19)
i=0
i! 2 6

Table 8.1 Euler integration example.

t x(t) f (x, t) x(t + h)

0 1 1 (1 + h)
h (1 + h) (1 + h) (1 + h)2
2h (1 + h)2 (1 + h)2 (1 + h)3
3h (1 + h)3 (1 + h)3 (1 + h)4
4 4
4h (1 + h) (1 + h) (1 + h)5
⋮ ⋮ ⋮ ⋮
N
ts = Nh (1 + h) − −

@Seismicisolation
@Seismicisolation
 8.6 The Central Difference Method: An 𝒪(h2 ) Method 153

The approximate solution ((1 + h)N ) can be expanded using the binomial theorem,
∑
k=n
n!
(a + b)n = an−k bk = an + nan−1 b
k=0
(n − k)!k!
n(n − 1) n−2 2 n(n − 1)(n − 2) n−3 3
+ a b + a b +··· (8.20)
2 6
where a = 1, b = h, and n = N in this case. The result is,
N(N − 1) 2 N(N − 1)(N − 2) 3
x(Nh) ≈ (1 + h)N = 1 + Nh + h + h +··· . (8.21)
2 6
The difference between the exact solution and the approximate solution can be found by
subtracting Equation 8.21 from Equation 8.19. This is the truncation error 𝜖T and can be
written as,
N 2 N
𝜖T = h + (3N − 2)h3 + · · · . (8.22)
2 6
Notice that the largest term in the truncation error is 𝒪(h2 ). For typical integration time
steps of h = 0.001 or smaller, the h2 term is on the order of 1000 times larger than the h3
term so the truncation error can be approximated as,
N 2 (Nh)h ts h
𝜖T ≈ h = = (8.23)
2 2 2
where ts is the total length of time simulated (see Table 8.1 where ts = Nh). As a result, the
truncation error at any simulated time (ts ) is 𝒪(h) and varies linearly with the time-step.
Notice that it also varies linearly with the total simulation time as may be expected.

8.6 The Central Difference Method: An 𝓞(h2 ) Method

We can develop an integrator with higher order truncation error using the truncated Taylor
series in Equation 8.9 where we now keep terms up to 𝒪(h3 ),
( ) ( ) ( )
dx 1 d2 x 2 1 d3 x
x(t0 + h) = x(t0 ) + h+ h + h3 (8.24)
dt t0 2! dt2 t0 3! dt3 𝜏
and a second truncated Taylor series that goes back one time step from t0 . That is,
( ) ( ) ( )
dx 1 d2 x 1 d3 x
x(t0 − h) = x(t0 ) − h+ 2
h − h3 . (8.25)
dt t0 2! dt2 t0 3! dt3 𝜏
We subtract Equation 8.25 from Equation 8.24 to get,
( ) ( )
dx 2h3 d3 x
x(t0 + h) − x(t0 − h) = 2h + (8.26)
dt t0 3! dt3 𝜏
from which we can write the derivative,
( ) ( )
dx x(t0 + h) − x(t0 − h) h2 d3 x
= − (8.27)
dt t0 2h 3! dt3 𝜏

@Seismicisolation
@Seismicisolation
154 8 Time Domain Solutions

or,
( ) x(t0 + h) − x(t0 − h)
dx
= + 𝒪(h2 ). (8.28)
dt t0 2h
The approximation to the derivative that we use is then,
( ) x(t0 + h) − x(t0 − h)
dx
≈ (8.29)
dt t0 2h
where it is recognized that the truncation error is 𝒪(h2 ).
This method is called the central difference method and it has a higher order truncation
error than the Euler integrator and should therefore produce more accurate results for the
same time step.
The propagation formula (derived from Equation 8.29 and the fact that f (x(t0 ), t0 ) is the
actual value of the derivative at time t0 ) is,

x(t0 + h) = x(t0 − h) + 2hf (x(t0 ), t0 ). (8.30)

Notice that this method cannot start with only the initial conditions at time t = 0. It
requires information from one time step previous to the current time and that information
is not available at t = 0. Methods like this are classified as being non-self-starting methods.
Another method must be used to propagate the solution out to t = h before the central dif-
ference method can be used.
Figure 8.4 shows the magnitude of the errors generated by the Euler integration
method and the central difference method as a function of the time step used. The
example is the same one used above in Equation 8.17 with the exponential solution as
stated in Equation 8.18. The percent errors quoted are those that exist after propagating
the solution to t = 10 s. According to Equation 8.18, the actual solution at that time is
x(10) = e10 = 22026 and this is used as a basis for calculating the percentage error. The
exact solution at t = h and the initial condition at t = 0 were used to get the central
difference method started. The Euler method started with the initial condition only.
The difference between the two methods is clear with the Euler method beginning
to lose accuracy at much smaller time steps than the central difference method. For

–20
Euler method
–40
% error

Central difference method

–60

–80

–100
1E–05 1E–04 1E–03 1E–02 1E–01 1E+00
time step “h” (s)

Figure 8.4 Solution errors as a function of time step.

@Seismicisolation
@Seismicisolation
 8.7 Variable Time Step Methods 155

example, at h ≈ 0.02 the Euler error is nearly −10% and the central difference error is
only −0.07%.

8.7 Variable Time Step Methods

To see how variable time step methods work, the midpoint method, a self-starting method
with 𝒪(h2 ) truncation error is introduced. This is a method that uses information from the
midpoint of the time step being considered. It will later be combined with the Euler method
in order to form a method with variable time steps.
Let there be an ordinary differential equation
dx
= f [x(t)] with the initial condition x(0) = x0 (8.31)
dt
with a solution, x(t), that has been propagated to time ts . The solution at ts + h is desired. The
midpoint method makes the extrapolation in two steps. First the value of x(t) at the midpoint
of the step (i.e. x(ts + h∕2)) is estimated by an Euler step from ts using the derivative f [x(ts )].
This midpoint value is designated as xmp .
h
xmp = x(ts ) + f [x(ts )]. (8.32)
2
The derivative at the midpoint, f [xmp ], is then calculated and used to propagate across the
entire time step. The estimated solution at time ts + h is then,

x(ts + h) = x(ts ) + hf [xmp ]. (8.33)

Figure 8.5 shows the steps graphically.

Notice that this method requires two evaluations of the derivative per time step and uses
a point within the time step. This is a feature of many higher order methods that will be
discussed later.
The truncation error of this method is determined as follows. First Equation 8.32 is sub-
stituted into Equation 8.33 to give,
h
x(ts + h) = x(ts ) + hf [x(ts ) + f [x(ts )]]. (8.34)
2

Figure 8.5 The midpoint method.

x(t+ h)

x(t) xmp

t t + h/2 t+h

@Seismicisolation
@Seismicisolation
156 8 Time Domain Solutions

The derivative at the midpoint, f [x(ts ) + h2 f [x(ts )]], appears in Equation 8.34 and can be
expanded using a Taylor series as follows1 ,
h h df [x(ts )]
f [x(ts ) +f [x(ts )]] = f [x(ts )] + f [x(ts )] + 𝒪(h2 ). (8.35)
2 2 dx
This is substituted into Equation 8.34 to yield,
( )
h df [x(ts )]
x(ts + h) = x(ts ) + hf [x(ts )] + h f [x(ts )] + h(𝒪(h2 )). (8.36)
2 dx
This can be simplified somewhat by noting that (see equation 8.31),
dx df [x(t)] dx df [x(t)] df [x(t)] d2 x
f [x(t)] = so that f [x(t)] = = = 2
dt dx dt dx dt dt
Substituting this into Equation 8.36 and noting that h(𝒪(h2 )) = 𝒪(h3 ) gives,
( ) ( )
dx h2 d2 x
x(ts + h) = x(ts ) + h + + 𝒪(h3 ) (8.37)
dt ts 2 dt2 ts
which is the Taylor series expansion for x(t) across the time step with 𝒪(h3 ) local trunca-
tion error and thus 𝒪(h2 ) global truncation error. Based on the global truncation error, the
midpoint method is classed as a second-order, self-starting integration technique.
Now consider the case where two different methods, Euler and midpoint, are used simul-
taneously to find a solution to the differential equation. They differ in truncation error by
one order so, given that they use the same time step, it can be expected that they will give
slightly different solutions.
The difference between the solutions can be determined from the Taylor series approxi-
mations.
( ) ( )
dx h2 d2 x
xMP (ts + h) = x(ts ) + h + + 𝒪(h3 ) (8.38)
dt ts 2 dt2 ts
and,
( )
dx
xEU (ts + h) = x(ts ) + h + 𝒪(h2 ). (8.39)
dt ts
Subtracting gives the difference, 𝜖, as,
( )
h2 d2 x
𝜖 = xMP (ts + h) − xEU (ts + h) = + 𝒪(h3 ) − 𝒪(h2 ) (8.40)
2 dt2 ts

which is numerically indeterminate but, assuming a small time step h, is 𝒪(h2 ) so that 𝜖
can be expressed as some constant, A, multiplied by h2 .
𝜖 = Ah2 . (8.41)

1 This step may not be obvious. It can be helpful to first write the Taylor series as,
( )
df
f (x + Δx) = f (x) + Δx + 𝒪(h2 )
dx x
and then note that, in this case,
h
Δx = f [x(ts )].
2

@Seismicisolation
@Seismicisolation
 8.8 Methods with Higher Order Truncation Error 157

The goal is to choose a value of h that keeps the difference between the solutions small.
To that end, the user specifies a tolerance on the difference, 𝜖tol , that must be satisfied in
order to maintain an accurate solution. If the difference between the solutions is larger than
the acceptable tolerance, the step-size must be changed in order to bring it under control.
Equation 8.41 allows the constant, A, to be determined as,
𝜖
A = 2. (8.42)
h
There will be a smaller step size, h∗ , that makes the difference between the solutions
satisfy the tolerance specification. That step size can be found from,
𝜖
𝜖tol = Ah2∗ which yields h2∗ = tol . (8.43)
A
Substituting the value of A from Equation 8.42 into Equation 8.43 gives the step size that
will satisfy the tolerance as,
√
𝜖tol
h∗ = h. (8.44)
𝜖
Using the variable step size method therefore requires the user to specify a tolerance and
an initial step size. The numerical routine can then adjust the step size throughout the
solution so that the tolerance is maintained. If the error is too large the step size is reduced
and, if the difference between the solutions is much smaller than the tolerance, the step
size can be increased to speed up the solution.
This has been a very rudimentary discussion of a complex field of numerical analysis
and is only meant to give an impression of the logic behind solution routines that employ
variable step size. There are many powerful routines available that employ this method and
they are the preferred methods of solution for simulations.

8.8 Methods with Higher Order Truncation Error

Introducing a few simple numerical methods for solving ordinary differential equations
has demonstrated the concept of truncation error and how it varies with the time step used
for the solution. The development of methods with higher order truncation error has been
of interest to numerical analysis experts for decades and there are many methods with very
high order truncation error (𝒪(h10 ) and higher methods exist). The truncation errors intro-
duced by these methods will clearly be smaller than those of lower order methods but one
must also consider the computational efficiency of the methods. The total error in a numer-
ical solution is not strictly due to truncation. There is also an error introduced by round-off
during any calculation. The numbers being used have a finite accuracy and something is
lost during each calculation as the numbers are rounded off to match the limited number of
significant figures in the computer. As a result, the round-off error grows with the number
of computations and the number of computations grows as the time-step gets smaller. There
is thus a trade-off between truncation error and round-off error. A very large step-size will
result in extreme truncation error and almost no round-off error. A step-size very close to
zero will yield almost no truncation error but extreme round-off errors due to the number of

@Seismicisolation
@Seismicisolation
158 8 Time Domain Solutions

calculations required to propagate the solution for any length of time. The result is an “op-
timum”, not too small and not too large, time step that minimizes the total solution error.
Unfortunately, it is impossible to know the value of that time step so numerical solutions
make use of “reasonable” time steps. Simulations with fixed time steps are often executed
twice using different time steps in order to ensure that the solution is not overly sensitive
to the time step used.
The most often used higher order method is the classical fourth-order Runge–Kutta2 for-
mula that requires four derivative (f (x, t)) evaluations per step. For the case where ts is the
time at the beginning of a step and xs is the known value of x(ts ), the derivatives are,
k1 = f (xs , ts ) (8.45)
( )
k h
k2 = f xs + h 1 , ts +
2 2
( )
k2 h
k3 = f xs + h , ts +
2 2
k4 = f (xs + hk3 , ts + h)
and the extrapolation formula is,
( )
k1 k 2 k3 k4
xs+h = xs + h + + + . (8.46)
6 3 3 6
Notice that the derivative is required at the beginning of the step, twice at the midpoint,
and finally at the end of the step. Since no information is required previous to the beginning
of the step, this is a self-starting method.
Variable time step versions of the Runge–Kutta method are very common and robust.
One of the best known is the Runge–Kutta–Fehlberg Method. It simultaneously generates
fourth and fifth order solutions using six derivative evaluations as follows.
k1 = f (xs , ts ) (8.47)
h
x1∕4 = xs + k1
4
( )
1
k2 = f x1∕4 , ts + h
4
3 9
x3∕8 = xs + hk1 + hk2
32 32
( )
3
k3 = f x3∕8 , ts + h
8
1932 7200 7296
x12∕13 = xs + hk − hk + hk
2197 1 2197 2 2197 3
( )
12
k4 = f x12∕13 , ts + h
13
439 3680 845
x1 = xs + hk − 8hk2 + hk − hk
216 1 513 3 4104 4

2 Carl David Tolme Runge (1856–1927) was a German mathematician, physicist, and spectroscopist.
Martin Wilhelm Kutta (1867–1944) was a German mathematician. In 1895 Carl Runge published the first
Runge–Kutta method. In 1905 Martin Kutta described the popular fourth-order Runge–Kutta method.

@Seismicisolation
@Seismicisolation
 8.9 The Structure of a Simulation Program 159

k5 = f (x1 , ts + h)
8 3544 1859 11
x1∕2 = xs − hk1 + 2hk2 − hk + hk − hk
27 2565 3 4104 4 40 5
( )
1
k6 = f x1∕2 , ts + h .
2
The fourth order Runge–Kutta approximation is then,
25 1408 2197 1
(xs+h )4 = xs + hk + hk + hk − hk (8.48)
216 1 2565 3 4104 4 5 5
and the fifth order approximation is,
16 6656 28561 9 2
(xs+h )5 = xs + hk + hk + hk − hk + hk . (8.49)
135 1 12825 3 56430 4 50 5 55 6
Given a tolerance, 𝜖tol , the scaling factor for step size, s, such that the correct step size is
h∗ = sh is,
( )1∕4
𝜖tol h
s= . (8.50)
2|(xs+h )5 − (xs+h )4 |

8.9 The Structure of a Simulation Program

This chapter ends with a description of the way in which a time domain simulation program
is structured.
There are three parts to any software designed for simulation of dynamic motions. They
will be called here, the main program, the solution routine, and the derivative routine. They
are shown graphically in the flowchart of Figure 8.6 and described below.

1. The main program. This is used primarily for input and output functions. Parameter
values necessary for defining the system (e.g. masses, moments of inertia, spring con-
stants – anything that is required to numerically form the equations of motion) are
accepted as input. These parameter values are stored in common memory so that they
can be used by the derivative routine. The main program sets the variables to their initial
conditions and then calls the solution routine, sending the current values of the variables
and time as well as a time when the updated solution is to be returned. The main program
receives the values of the degrees of freedom from the solution routine at the specified
time, writes them to an output file, checks to see if the total simulation times has been
exceeded, and then returns control to the solution routine or stops.
2. The solution routine. This is an implementation of one of the numerical solution meth-
ods described earlier in this chapter. The main program supplies it with numerical values
of the variables, the starting time for the step, the time for the next output and the
required tolerance (for variable time step methods). The solution routine needs to be
able to call the derivative routine to get the derivatives whenever required. Recall that
this operation happens four times per step for the classical fourth-order Runge–Kutta
method and six times per step for the Runge–Kutta–Fehlberg method. The solution rou-
tine retains control and propagates the solution of the differential equations forward in

@Seismicisolation
@Seismicisolation
160 8 Time Domain Solutions

Figure 8.6 Structure of a simulation program.

time until it reaches the time at which the main program is set to write output. It then
returns control to the main program.
3. The derivative routine. This is where the equations of motion are coded. This routine is
designed simply to provide the derivatives to the solution routine whenever requested.
The solution routine calls the derivative routine, providing the current time and the val-
ues of the variables, and expects, in return, the numerical values of the derivatives it
needs in order to propagate the solution. The derivative routine needs access to the sys-
tem parameters that the main program stored in common memory in order to set up the
equations of motion and solve them for the derivatives.
Consider, as an example, the pendulum from Section 8.2 (see Figure 8.1). The equations
of motion are repeated here for easy reference. They are,
{ } { x2
}
ẋ 1
=
ẋ 2 g
− 𝓁 sin x1 + F(t) cos x1
m𝓁

@Seismicisolation
@Seismicisolation
 8.9 The Structure of a Simulation Program 161

with,
F(t) = mg (1 − t) for 0 ≤ t ≤ 1 and F(t) = 0 for t > 1.
Table 8.2 shows the pseudocode for the Main Program and the Derivative Routine for this
system. Any other system would have exactly the same programming structure.
Systems with more than one degree of freedom have derivative routines that are more
complicated than the example used in Table 8.2. The general case was presented in
Equation 8.7, repeated here as Equation 8.51.

Table 8.2 Pendulum simulation pseudo-code.

MAIN PROGRAM

START
define common memory for m, g, and 𝓁
open an input file
read system parameters (m, g, and 𝓁)
read initial conditions (x1 (0) and x2 (0))
read time between output values (tout )
read total simulation time(tf )
read integration control parameters (𝜖, hmint , hmax , hstart )
WHILE t is less than tf
set tret = t + tout
call the solution routine with the current values of {x},
t, and tret
receive updated values of {x} at t = tret
write values of t and {x} to the output file
ENDWHILE
STOP

DERIVATIVE ROUTINE

START
receive m, g, and 𝓁 from common memory
receive {x} and t from the solution routine
IF t is greater than 1
set F(t) = 0
ELSE
set F(t) = mg (1 − t)
ENDIF
set ẋ 1 = x2
set ẋ 2 = − 𝓁g sin x1 + F(t)
m𝓁
cos x1
RETURN

@Seismicisolation
@Seismicisolation
162 8 Time Domain Solutions

{ } { }
ẋ 1 x2
= (8.51)
ẋ 2 [M(x1 , x2 , t)]−1 {f (x1 , x2 , t)}
with user specified initial conditions
{ } { } { }
x1 (0) q(0) displacements at t = 0
= = .
x2 (0) ̇
q(0) velocities at t = 0
It is clear that, in the general case, a set of linear algebraic equations has to be solved
whenever the derivative routine responds to a request for the values of the derivatives.

Table 8.3 General simulation pseudo-code.

MAIN PROGRAM

START
define common memory for system parameters
open an input file
read system parameters
read initial conditions
read time between output values (tout )
read total simulation time(tf )
read integration control parameters (𝜖, hmin , hmax , hstart )
WHILE t is less than tf
set tret = t + tout
call the solution routine with the current values of {x},
receive updated values of {x} at t = tret
write values of t and {x} to the output file
ENDWHILE
STOP

DERIVATIVE ROUTINE

START
receive system parameters from common memory
receive {x} and t from the solution routine
set {ẋ 1 } to {x2 }
calculate the RHS of the equations of motion ({f (x1 , x2 , t)})
calculate the mass matrix ([M(x1 , x2 , t)])
send [M(x1 , x2 , t)] and {f (x1 , x2 , t)} to a linear algebra solver
receive {ẋ 2 }
RETURN

@Seismicisolation
@Seismicisolation
 Exercises 163

Equation 8.51 specifies 2N derivatives. The first N are obtained by simply equating elements
of the vector {ẋ 1 } to {x2 } N times. The second N require that Equation 8.5 (repeated here
as Equation 8.52 be solved

[M(x1 , x2 , t)]{ẋ 2 } = {f (x1 , x2 , t)}. (8.52)

In the text the solution has been written as if the inverse of the mass matrix had been
calculated and then multiplied by the right hand side of the equations of motion. While
convenient for use in the text, this is actually a very inefficient way to solve linear algebraic
equations. It is much better to use a routine that is designed for efficient solutions of linear
algebraic equations such as Gaussian elimination. The pseudocode for the general case is
shown in Table 8.3.

Exercises

Descriptions of the systems referred to in the exercises are contained in Appendix A.

8.1 You previously found, in Exercise 3.2, that the equations of motion for system 1 are
[ ]{ }
(m1 + m2 )d12 + m2 d2 (d2 + 2d1 cos 𝜃2 ) m2 d2 (d2 + d1 cos 𝜃2 ) 𝜃̈1
m2 d2 (d2 + d1 cos 𝜃2 ) m2 d22 𝜃̈2
{ }
m2 d1 d2 𝜃̇ 2 (2𝜃̇ 1 + 𝜃̇ 2 ) sin 𝜃2 − (m1 + m2 )gd1 cos 𝜃1 − m2 gd2 cos(𝜃1 + 𝜃2 )
= .
−m2 d1 d2 𝜃̇ 12 sin 𝜃2 − m2 gd2 cos(𝜃1 + 𝜃2 )

(a) Set up these equations in the form you would need for a time domain solution.
(b) For the case where d = 0.05 m and m = 1.0 kg, implement an Euler integration
scheme and solve the equations of motion for 20 s of simulated time. Experiment
with different time steps to see how the solution is affected.

8.2 The nonlinear equation of motion for system 6 was found twice, in Exercises 2.6 and
3.4. In Exercise 4.3, one equilibrium value of 𝜃 was specified to be 60∘ and the second
was found numerically. Exercise 5.4 asked for an analysis of the stability of the two
equilibria.
(a) Create a time domain simulation of system 6 using the parameters determined in
Exercise 4.3.
(b) Run your simulation for arbitrary initial conditions and note that the system is
always trying to get to the stable equilibrium state but cannot settle out there
because energy needs to be removed from the system in order for it to stop moving.
Undamped, oscillatory solutions about the stable equilibrium value of 𝜃 should
be clear, especially if you start with no initial velocity and an initial value of 𝜃
close to the stable equilibrium value.
(c) Add viscous damping to the equation of motion (a moment on the arm opposing,
and proportional to, 𝜃̇ will do the trick) and see if your time domain solution
converges to the same stable equilibrium value you found in Exercise 4.3.

@Seismicisolation
@Seismicisolation
164 8 Time Domain Solutions

8.3 Create a time domain simulation of system 12 using the equation of motion from
Exercise 3.12. Use parameter values: m = 1000 kg, k = 30 000 N m−1 , and d = 1.2 m.
(a) Implement the classical, fourth-order, Runge–Kutta method described in Section
8.8 and use it for the simulation.
(b) Calculate the equilibrium value of 𝜃 from the equation of motion using the tech-
niques from Chapter 4. Start your simulation from 𝜃 = 0 with zero initial velocity
and plot the results versus time. You should see an oscillation about the equilib-
rium value you found.
(c) Linearize the equation of motion about equilibrium and find the natural fre-
quency analytically using the techniques of Chapter 5. How well does this match
the frequency you get by counting cycles on the plotted results of the nonlinear
simulation?

8.4 Use parameter values of your choice for system 16 and solve the nonlinear equations
of motion (see Exercise 3.9) with initial conditions: 𝜃(0) = 90∘ and 𝜙(0) = 10∘ . Using
your predicted values of 𝜃 and 𝜙, sketch the system at several times over one cycle of
motion and visualize the animated motion of rod AB.

@Seismicisolation
@Seismicisolation
 165

Part III

Working with Experimental Data

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 167

Experimental Data – Frequency Domain Analysis

Up to now, we have concentrated on the theoretical simulation of systems in order to obtain

information useful in their design. The theoretical study of dynamic systems and the ability
to understand the physical behavior of the systems that the theory imparts are immensely
important to practicing engineers. The theory is what leads to the initial system design
through simulation and the theory is what gives the ability to understand and correct the
behavior of the prototype systems when they are tested.
Equally as important to engineers is the analysis of experimentally derived response data
for systems. The validation of the models used in the design phase relies on being able to
record and analyze test data. Such analysis is most often done in the frequency domain and
that is what we will concentrate on here.

9.1 Typical Test Data

Dynamic measurements are made by sensors of various kinds and are most often recorded
in digital form1 . Figure 9.1 shows a typical measured dynamic variable x(t) plotted as a
function of time. Measured variables very often appear to be random because, even after
looking at the signal over the entire plot, there is no way to predict where the next point
will be.
Characterization of the time domain signal is unrewarding. The mean value of the signal
can be found but is not representative of anything but the “average” value of x(t). It contains
no information with respect to peak values for instance. Given that x(t) is known in the
range 0 ≤ t ≤ T, the mean value is defined as
T
1
x= x(t) dt. (9.1)
T ∫0

1 The analog measurements that used to be recorded on analog devices such as multi-channel tape
recorders have been almost completely superseded by in-field A/D converters and digital recording devices.
The sampling rate of the A/D converters must be chosen very carefully to satisfy the requirements of the
data analysis that will be done later. With analog devices it was possible to re-play the experiments if you
got the sampling rate wrong. Now you are required to get it right or repeat the, sometimes very costly,
experiments.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
168 9 Experimental Data – Frequency Domain Analysis

20

10

x(t) 0

–10

–20
0 2 4 6 8 10 12 14 16
time (s)

Figure 9.1 A measured variable x(t) plotted versus time.

A characteristic of the signal that can be calculated and that has meaning is the total
2
mean-square value of the signal, xT . The total mean-square value of the signal is defined
as the mean value of another signal generated by squaring the original signal and then
finding its mean value over the period 0 ≤ t ≤ T using Equation 9.2. Figure 9.2 shows the
2
squared signal and the mean-square value xT = 43. Characterizing the signal by this single
scalar value seems less than satisfying but that is all that can be done in the time domain.
However, the total mean-square value will be useful to us when we discuss scaling of results

250

200
x(t) squared

150

mean square
100 value = 43

50

0
0 2 4 6 8 10 12 14 16 18
time (s)

Figure 9.2 The square of x(t) plotted versus time.

@Seismicisolation
@Seismicisolation
 9.2 Transforming to the Frequency Domain – The CFT 169

in the frequency domain in Section 9.8

T
2 1
xT = x2 (t) dt (9.2)
T ∫0

9.2 Transforming to the Frequency Domain – The CFT

The most common approach to analyzing experimental signals begins with a transforma-
tion that takes the measured data into the frequency domain. Readers may be familiar with
the terms FFT and fast Fourier transform that describe a technique that is widely used
to accomplish the transformation but may not familiar with the details of the transfor-
mation. The mathematical development of the Fourier transform equations is most easily
done using complex exponentials and that is the approach used in modern reference books.
Unfortunately, the complex exponential approach, while being mathematically elegant,
leaves the reader without any physical feeling for what lies behind the equations. The goal
here is to present the techniques in a manner that leaves the reader with an understanding
of the techniques and a physical feeling for what they represent.
It is best to think of the transformation as a curve fitting exercise. That is, we have a
continuous time domain function x(t) and we wish to approximate it in the range 0 ≤ t ≤ T
by using a series of harmonic functions. We choose a base frequency f0 that allows one
complete cycle in the range and then add higher frequency components that are multiples
of f0 . Clearly, f0 in cycles per second will be such that there is one cycle in T seconds so that
f0 = 1∕T cycles per second or Hz. The base frequency in radians per second will then be
𝜔0 = 2𝜋∕T.
Let the approximating function xA (t) be expressed in terms of (2N + 1) undetermined
coefficients a0 , an , and bn as follows
∑
N
∑
N
xA (t) = a0 + an cos(n𝜔0 t) + bn sin(n𝜔0 t). (9.3)
n=1 n=1

While there are several techniques that could be used to find the “best” values of the coef-
ficients, we will use the traditional least squares2 approach because it is commonly used in
fitting curves to experimental data. We first define an objective function J that is a measure
of the difference between the experimental data and the approximating function. Since the
difference [x(t) − xA (t)] can be either positive or negative, we can’t use it to establish the
goodness of fit so we work with the square of the difference instead and define,
T
1
J= [x(t) − xA (t)]2 dt. (9.4)
2 ∫0
We choose the coefficients such that they minimize the objective function, thereby mini-
mizing the difference between the experimental curve x(t) and the approximating function

2 Appendix D has a description of least squares curve fitting for those who would like to review the
method before proceeding with the remainder of this chapter.

@Seismicisolation
@Seismicisolation
170 9 Experimental Data – Frequency Domain Analysis

xA (t). The conditions for minimizing J are that the partial derivatives of J with respect to all
of the undetermined coefficients must simultaneously be equal to zero. J can be expanded
into3 .
[ ]2
1
T ∑N
∑N
J= x(t) − a0 − an cos(n𝜔0 t) − bn sin(n𝜔0 t) dt (9.5)
2 ∫0 n=1 n=1

The partial derivative with respect to a0 is

[ ]
𝜕J
T ∑N
∑N
= −x(t) + a0 + an cos(n𝜔0 t) + bn sin(n𝜔0 t) dt = 0. (9.6)
𝜕a0 ∫0 n=1 n=1

For the remaining coefficients,

[ ]
𝜕J
T ∑N
∑N
= −x(t) + a0 + an cos(n𝜔0 t) + bn sin(n𝜔0 t) cos(p𝜔0 t)dt = 0,
𝜕ap ∫0 n=1 n=1

p = 1, N (9.7)
and
[ ]
𝜕J
T ∑
N
∑
N
= −x(t) + a0 + an cos(n𝜔0 t) + bn sin(n𝜔0 t) sin(p𝜔0 t)dt = 0,
𝜕bp ∫0 n=1 n=1

p = 1, N. (9.8)
Consider first satisfying the condition on a0 from Equation 9.6. This can be written as
[ T ] [ T ]
T T ∑N
∑N
− x(t)dt + a0 dt + an cos(n𝜔0 t)dt + bn sin(n𝜔0 t)dt = 0.
∫0 ∫0 n=1
∫0 n=1
∫0
(9.9)
Three of the four integrals in Equation 9.9 can be directly evaluated, yielding
T T T
dt = T; cos(n𝜔0 t)dt = 0 ; sin(n𝜔0 t)dt = 0. (9.10)
∫0 ∫0 ∫0
The integrals of the trigonometric functions give the area under them as time goes from zero
to T. Since they have been defined as multiples of a single wavelength spanning 0 ≤ t ≤ T,
they will all complete an integer number of cycles and the area under them will be zero. A
mathematical proof is left as an exercise.
The result is that a0 is
T
1
a0 = x(t)dt. (9.11)
T ∫0
Comparing this to Equation 9.1, we see that a0 is the mean value of x(t). Examination of
the approximating function in Equation 9.3 makes it clear that this must be the case. The

3 You may wonder why a0 is included and b0 is not. In fact, a more elegant way of writing the series is

∑
N
∑
N
xA (t) = an cos(n𝜔0 t) + bn sin(n𝜔0 t)
n=0 n=0

but it is clear that sin(n𝜔0 t) = 0 when n = 0 so the form shown in Equation 9.3 is universal.

@Seismicisolation
@Seismicisolation
 9.2 Transforming to the Frequency Domain – The CFT 171

function has one constant term (a0 ) and superimposes harmonics on top of it. The constant
value can only be the mean value.
We now turn to the conditions on the other coefficients as specified in Equations 9.7 and
9.8. These involve integrals of the form
T
cos(n𝜔0 t) sin(p𝜔0 t)dt (9.12)
∫0
T
sin(n𝜔0 t) sin(p𝜔0 t)dt (9.13)
∫0
T
cos(n𝜔0 t) cos(p𝜔0 t)dt (9.14)
∫0
and
T
sin(n𝜔0 t) cos(p𝜔0 t)dt (9.15)
∫0
all of which can be shown to be zero so long as n ≠ p (see Exercise 9.2). This is a result of the
functions 1, cos pt, and sin pt being orthogonal over the interval [0 ≤ pt ≤ 2𝜋]. This concept
of orthogonality will reappear when we consider discrete Fourier transforms in Section 9.4.
In the case where p = n Equations 9.7 and 9.8 become
T T T
− x(t) cos(n𝜔0 t)dt + a0 cos(n𝜔0 t)dt + an cos2 (n𝜔0 t)dt
∫0 ∫0 ∫0
T
+bn sin(n𝜔0 t) cos(p𝜔0 t)dt = 0 (9.16)
∫0
and
T T T
− x(t) sin(p𝜔0 t)dt + a0 sin(p𝜔0 t)dt + an cos(n𝜔0 t) sin(p𝜔0 t)dt
∫0 ∫0 ∫0
T
+bn sin2 (n𝜔0 t)dt = 0. (9.17)
∫0
Several of the integrals in Equations 9.16 and 9.17 have already been shown to be zero.
In addition, it can easily be shown that
T T
T
cos2 (n𝜔0 t)dt = sin2 (n𝜔0 t)dt = (9.18)
∫0 ∫0 2
and, as a result, we find that the least squares curve fit to the experimental data is
∑
N
∑
N
xA (t) = a0 + an cos(n𝜔0 t) + bn sin(n𝜔0 t) (9.19)
n=1 n=1

where
T
1
a0 = x(t)dt (9.20)
T ∫0
T
2
an = x(t) cos(n𝜔0 t)dt, n = 1, N (9.21)
T ∫0
T
2
bn = x(t) sin(n𝜔0 t)dt, n = 1, N. (9.22)
T ∫0

@Seismicisolation
@Seismicisolation
172 9 Experimental Data – Frequency Domain Analysis

If we now let N increase so that we minimize the objective function at more and more
points we see that as N approaches infinity the approximation becomes equal to the func-
tion x(t) so that we can write
∑
∞
∑
∞
x(t) = a0 + an cos(n𝜔0 t) + bn sin(n𝜔0 t) (9.23)
n=1 n=1

where
T
1
a0 = x(t)dt (9.24)
T ∫0
T
2
an = x(t) cos(n𝜔0 t)dt, n = 1, ∞ (9.25)
T ∫0
T
2
bn = x(t) sin(n𝜔0 t)dt, n = 1, ∞. (9.26)
T ∫0
The transformation given by Equations 9.23 through 9.26 is the well-known continuous
Fourier rransform (CFT)4 .
Notice that we have considered a continuous function x(t) that was defined over the inter-
val 0 ≤ t ≤ T. The Fourier series representation of this function is based on this interval even
though it can be evaluated for any time t. There is an implicit assumption that the function
being analyzed is periodic with a period equal to T. This fact will become important to us
as we proceed.

9.3 Transforming to the Frequency Domain – The DFT

We have shown that a periodic, continuous function can be exactly represented by a curve
fit using an infinite series of harmonic functions, with a base frequency derived from the
period of the function and multiples of that frequency. Clearly, practicing engineers will
never have to deal with functions like this. Measurements we take are not continuous but
are sampled and stored at some sampling rate that gives an equal time between samples,
Δt. The measured values are not periodic but we have to finish sampling at some time so
we get data over the range 0 ≤ t ≤ T where T is the time at which we stop sampling. As a
result, we get a finite number of points to analyze. Consider the curve fitting process again
but without a continuous function to fit this time.
Let there be 2N measured data points stored in a vector x(t). Let the sampling rate be
fs = 1∕Δt where Δt is the constant time between samples. Define the base frequency f0 in
Hz as
1
f0 = (9.27)
2NΔt
giving a base frequency 𝜔0 in rad s−1
2𝜋 𝜋
𝜔0 = = . (9.28)
2NΔt NΔt

4 Jean-Baptiste Joseph Fourier (1768–1830), a French mathematician, is best remembered for his work on
the propagation of heat in solid bodies and for his expansions of functions as trigonometric series that we
now call Fourier series.

@Seismicisolation
@Seismicisolation
 9.3 Transforming to the Frequency Domain – The DFT 173

Use a series of harmonics to fit the measured data. The series is

∑
N
x(t) = [an cos(n𝜔0 t) + bn sin(n𝜔0 t)]. (9.29)
n=0

Since we have 2N data points, we can use them to find exactly 2N coefficients, an and bn ,
in Equation 9.29. As it stands, Equation 9.29 has (2N + 2) undetermined coefficients. We
can rewrite the series with its first and last terms extracted to see which two coefficients
are unnecessary. We do this for some time t = mΔt where 0 ≤ m ≤ (2N − 1). Substituting
Equation 9.28 into Equation 9.29 and extracting the first and last terms gives 2N simulta-
neous equations of the form
xm = x(mΔt) = a0 cos(0) + b0 sin(0)
N−1 [ ( ) ( )]
∑ n𝜋 n𝜋
+ an cos m + bn sin m
n=1
N N
+ aN cos(𝜋m) + bN sin(𝜋m) , 0 ≤ m ≤ (2N − 1) (9.30)
where we note that cos(0) = 1 and sin(0) = sin(𝜋m) = 0 so that the coefficients b0 and bN
do not contribute to the curve fit and can be removed5 . This leaves us with exactly 2N unde-
termined coefficients and 2N data points so we can write 2N simultaneous equations of the
form
N−1 [ ( ) ( )]
∑ n𝜋 n𝜋
xm = a0 + an cos m + bn sin m
n=1
N N
+aN cos(𝜋m) , 0 ≤ m ≤ (2N − 1). (9.31)
The equations can be assembled into the standard matrix form for linear, algebraic
equations with a known coefficient matrix, [C], multiplied by a vector of unknowns,
{a0 … aN b1 … bN−1 }T , set equal to a vector of known values {x0 … x2N−1 }T
⎧ a ⎫
⎪ 0 ⎪
⎪ ⋮ ⎪
⎪ ⎪ ⎧ x0 ⎫
[ ] ⎪ aN ⎪ ⎪ ⎪
C ⎨ ⎬=⎨ ⋮ ⎬ (9.32)
b
⎪ 1 ⎪ ⎪ ⎪
⎪ ⋮ ⎪ ⎩ x2N−1 ⎭
⎪ ⎪
⎪b ⎪
⎩ N−1 ⎭
where the 2N by 2N coefficient matrix, [C], has terms cij , for the an (left side of the matrix)
[ ]
(i − 1)(j − 1)𝜋
cij = cos ; i = 1, 2N, j = 1, N + 1 (9.33)
N
and, for the bn (right side of the matrix)
[ ]
(i − 1)(j − N − 1)𝜋
cij = sin ; i = 1, 2N, j = N + 2, 2N. (9.34)
N

5 We are treating the case where we have an even number of samples. The case for an odd number of
samples, (2N + 1), can be handled in a similar way with the small difference that an extra coefficient, bN , is
retained.

@Seismicisolation
@Seismicisolation
174 9 Experimental Data – Frequency Domain Analysis

Solving for the unknown coefficients using Equation 9.32 is a simple matter of using a
standard elimination technique such as Gaussian elimination. The computational effort
required to do so is approximately proportional to N 3 where “computational effort” is mea-
sured in terms of the number of multiplications and divisions required to implement a
solution on a digital computer.

9.4 Transforming to the Frequency Domain – A Faster DFT

Here, we return to the concept of least squares curve fitting that we used in Section 9.2 but
now it is applied to the finite number of data points we discussed in Section 9.3. We start
with the series from Equation 9.31 rewritten so that a0 and aN are taken into the summation

∑
N ( ) N−1
∑ ( )
n𝜋 n𝜋
xm = an cos m + bn sin m , 0 ≤ m ≤ (2N − 1) (9.35)
n=0
N n=1
N

and construct an objective function, J, as

[ ]
( ) N−1 ( ) 2
1 ∑ ∑ ∑
2N−1 N
n𝜋 n𝜋
J= x − a cos m − bn sin m . (9.36)
2 m=0 m n=0 n N n=1
N

As before, we set the partial derivatives of J with respect to each of the coefficients to zero.
That is,
[N
∑
2N−1 ( p𝜋 ) 2N−1 ∑ ∑ ( ) ( p𝜋 )
𝜕J n𝜋
=− xm cos m + an cos m cos m
𝜕ap m=0
N m=0 n=0
N N
]
∑
N−1 ( ) ( p𝜋 )
n𝜋
+ bn sin m cos m = 0 , p = 0, N (9.37)
n=1
N N

and
[N
∑
2N−1 ( p𝜋 ) 2N−1 ∑ ∑ ( ) ( p𝜋 )
𝜕J n𝜋
=− xm sin m + an cos m sin m
𝜕bp m=0
N m=0 n=0
N N
]
∑
N−1 ( ) ( p𝜋 )
n𝜋
+ bn sin m sin m = 0 , p = 1, N − 1. (9.38)
n=1
N N

We used orthogonality in Section 9.2 to show that the vast majority of the integrals involv-
ing products of harmonic functions were equal to zero. Orthogonality also holds in the case
of summations. The rules are as follows (for 0 ≤ p, n ≤ N),
⎧0 if p = n = 0, N
∑
2N−1 ( ) ( p𝜋 ) ⎪
n𝜋
sin m sin m = ⎨0 if p≠n (9.39)
m=0
N N ⎪
⎩N if p=n≠0
∑
2N−1 ( ) ( p𝜋 )
n𝜋
sin m cos m =0 (9.40)
m=0
N N

@Seismicisolation
@Seismicisolation
 9.5 Transforming to the Frequency Domain – The FFT 175

⎧
∑
2N−1 ( ) ( p𝜋 ) ⎪0 if p ≠ n
n𝜋
cos m cos m = ⎨N if p = n ≠ 0, N . (9.41)
m=0
N N ⎪2N if p = n = 0, N
⎩
Applying the orthogonality rules to the conditions on the partial derivatives specified in
Equations 9.37 and 9.38 yields the following expressions for the coefficients.

1 ∑
2N−1
a0 = x (9.42)
2N m=0 m
( )
1 ∑
2N−1
n𝜋
an = xm cos m ,1 ≤ n ≤ N − 1 (9.43)
N m=0 N

1 ∑
2N−1
aN = x cos(𝜋m) (9.44)
2N m=0 m
( )
1 ∑
2N−1
n𝜋
bn = xm sin m , 1 ≤ n ≤ N − 1. (9.45)
N m=0 N
The computational effort required to calculate the coefficients using the series given in
Equations (9.42–9.45) is approximately proportional to N 2 . Comparing this to the N 3 result
given in Section 9.3, where a set of linear, algebraic equations was solved, shows the inher-
ent advantage to using the series solution, especially when N is large, as it usually is. For
N = 1000, using the series method is approximately 1000 times faster than using the linear,
algebraic equations.

9.5 Transforming to the Frequency Domain – The FFT

Up to this point, we have been attempting to maintain a geometric feeling for the trans-
formations by describing the process as a curve fitting exercise using sines and cosines to
represent the experimentally measured data. From a purely mathematical view, it is much
better to do the analysis in the complex domain where the Fourier transform is given by
T √
X(f , T) = x(t)e−i2𝜋ft dt where i = −1. (9.46)
∫0
Euler’s formula
ei𝜃 = cos 𝜃 + i sin 𝜃 (9.47)
shows how this is equivalent to the sine and cosine series we have been using.
The fast Fourier transform or FFT was developed in the mid-1960s. It is beyond the scope
of this text to develop the actual algorithm. Suffice it to say that the computational effort is
significantly less than other methods. The FFT algorithm requires that the number of data
points be a power of two (i.e. 2N = 2p in our cases where we were working with 2N points).
The computational effort for the FFT is proportional to Nlog2 N which is significantly less
than for the series method presented in Section 9.4. For example, if N = 210 = 1024, the
computational effort of the three methods we have discussed is summarized in Table 9.1.

@Seismicisolation
@Seismicisolation
176 9 Experimental Data – Frequency Domain Analysis

Table 9.1 Fourier transforms computational effort.

Method for N = 2p e.g. p = 10 Computer time

Linear algebraic 23p 1.06 × 109 104 858

equations
Series 22p 1.05 × 106 102
p 4
FFT p2 1.02 × 10 1

The column labeled computer time shows the relative amount of computer processing time
that the three methods would use on the same computer and is not in any particular unit
of time. The table makes it abundantly clear why the FFT is the method of choice for all
analysts. Implementations of the algorithm are commonplace and can even be found in
spreadsheets.

9.6 Transforming to the Frequency Domain – An Example

As an example of the techniques for transforming to the frequency domain, consider the
function
x(t) = 2 + 5 sin(2𝜋t) + 10 sin(4𝜋t) + 15 cos(6𝜋t) (9.48)
which is shown in Figure 9.3. This function has a mean value of 2 and three harmonics at
frequencies of 1, 2, and 3 Hz. The harmonics have amplitudes of 5, 10, and 15 respectively.
Consider sampling the function at a rate of 10 samples per second (Δt = 0.10 s) and col-
lecting 20 samples (2N = 20). The data are shown in Table 9.2. The base frequency for the

30

20

10

x(t) 0

–10

–20

–30
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
time (s)

Figure 9.3 The example function, x(t), plotted versus time.

@Seismicisolation
@Seismicisolation
 9.6 Transforming to the Frequency Domain – An Example 177

Table 9.2 Sampled data.

n Time (s) x(t)

1 0.0 17.000
2 0.1 9.814
3 0.2 0.498
4 0.3 13.013
5 0.4 0.064
6 0.5 −13.000
7 0.6 13.207
8 0.7 15.258
9 0.8 −20.768
10 0.9 −15.085
11 1.0 17.000
12 1.1 9.814
13 1.2 0.498
14 1.3 13.013
15 1.4 0.064
16 1.5 −13.000
17 1.6 13.207
18 1.7 15.258
19 1.8 −20.768
20 1.9 −15.085

transformation is given by Equation 9.27. That is,

1 1
f0 = = = 0.50 Hz (9.49)
2NΔt 20 × 0.10
and we expect 20 coefficients, a0 … a10 and b1 … b9 . Solving the set of linear, algebraic
equations from Section 9.3 or using the series from Section 9.4 both result in the coefficients
shown in Table 9.3. Notice that the FFT can’t be used on this data set since the number of
points is not a power of two.
The original function can then be reconstructed using the coefficients in Table 9.3 and
Equation 9.3. That is, retaining only the non-zero coefficients,
xA (t) = a0 + a6 cos(6𝜔0 t) + b2 sin(2𝜔0 t) + b4 sin(4𝜔0 t) (9.50)
where
𝜔0 = 2𝜋f0 = 2𝜋 × 0.50 = 𝜋 (9.51)
yielding
xA (t) = 2.000 + 15.000 cos(6𝜋t) + 5.000 sin(2𝜋t) + 10.000 sin(4𝜋t) (9.52)
which is exactly the same as the original function defined in Equation 9.48.

@Seismicisolation
@Seismicisolation
178 9 Experimental Data – Frequency Domain Analysis

Table 9.3 DFT coefficients.

Coefficient Value Frequency (Hz)

a0 2.000 0.00
a1 0.000 0.50
a2 0.000 1.00
a3 0.000 1.50
a4 0.000 2.00
a5 0.000 2.50
a6 15.000 3.00
a7 0.000 3.50
a8 0.000 4.00
a9 0.000 4.50
a10 0.000 5.00
b1 0.000 0.50
b2 5.000 1.00
b3 0.000 1.50
b4 10.000 2.00
b5 0.000 2.50
b6 0.000 3.00
b7 0.000 3.50
b8 0.000 4.00
b9 0.000 4.50

So it is clear that the transformation to the frequency domain can be undone and we can
transform back to the time domain. This is called the inverse Fourier transform.
There is another way to write the DFT. We can use amplitudes and phase angles instead
of the coefficients, an and bn . We did this in Section 6.1 when we were considering eigen-
vectors. We can write
an cos(n𝜔0 t) + bn sin(n𝜔0 t) = An cos(n𝜔0 t + 𝜙n ) (9.53)
where An is the amplitude
√
An = a2n + b2n (9.54)
and 𝜙n is the phase angle
𝜙n = arctan(bn ∕an ). (9.55)
In fact, most data analysis concentrates on amplitudes only and phase angles are ignored.
The function x(t) that we have been considering (Equation 9.48 or 9.53), would appear on a
plot of amplitude versus frequency as shown in Figure 9.4. We will be working strictly with
amplitudes in the following.

@Seismicisolation
@Seismicisolation
 9.7 Sampling and Aliasing 179

16

14

12

10
amplitude

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
frequency (Hz)

Figure 9.4 The DFT amplitudes of the example function, x(t), plotted versus frequency.

9.7 Sampling and Aliasing

Consider the case where there is a single frequency harmonic signal at f Hz. That is

x(t) = A cos(2𝜋ft) + B sin(2𝜋ft). (9.56)

If we used a DFT to detect this signal we would need to sample fast enough to capture
the information required to calculate the coefficients an and bn using Equations 9.43 and
9.45, for example. The question is, how fast must we sample to gather the necessary informa-
tion about the signal? The answer is fairly straightforward. You can visualize the process as
follows.
Since there is only a single frequency present, all but one of the terms in the summation
required for the DFT of Equation 9.29 will be zero and the DFT will reduce to

x(t) = ap cos(p𝜔0 t) + bp sin(p𝜔0 t) (9.57)

where p𝜔0 = 2𝜋f . That is, only the term at the frequency of the signal will remain.
Equation 9.57 has two unknowns, ap and bp . Returning to our curve fitting analogy, we
have two unknowns so we will need two points per cycle in order to be able to calculate the
amplitudes required. We therefore need to sample at a frequency at least twice as high as
the frequency to be detected.
Let Δt be the time between samples. The sampling rate is then
1
fs = Hz (9.58)
Δt
and the highest frequency that can be detected, called the Nyquist frequency, is
fs 1
fc = = Hz (9.59)
2 2Δt

@Seismicisolation
@Seismicisolation
180 9 Experimental Data – Frequency Domain Analysis

1.5
sampled points actual signal
1.0

0.5

x(t) 0.0

–0.5

–1.0
DFT approximation
–1.5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
time (s)

Figure 9.5 Aliasing.

The obvious next question is, since the Nyquist frequency is the highest that we can detect
with our sampling rate, what happens if the signal has frequency content above the Nyquist
frequency?
Figure 9.5 shows a high frequency signal (solid line) sampled at five points (squares) and
the DFT approximation that would result (dashed line). Clearly, the high frequency signal
is in the data and the sampling will detect it but at less than two points per cycle so the DFT
approximation will be a signal with a lower frequency. This is called aliasing, a fitting name
for a high frequency signal masquerading as a low frequency signal.
In order to determine what frequency we can expect to see in the DFT, consider the follow-
ing and note that the complex exponential notation for harmonics is used simply because
it shortens the argument. Let there be a high frequency signal, x(t), with amplitude A, fre-
quency 𝜔, and phase angle 𝜙 so that we can write
x(t) = Aei(𝜔t+𝜙) = Aei𝜔t ei𝜙 (9.60)
and let the signal detected by the DFT, xd (t), have an amplitude a, be at a detected frequency,
𝜔d , and have a phase angle 𝜙d
xd (t) = aei(𝜔d t+𝜙d )t = aei𝜔d t ei𝜙d (9.61)
where 𝜔d < 𝜔.
Let there be a time t0 where x(t) and xd (t) are equal. This corresponds to one of the sampled
points shown in Figure 9.5 and can be expressed as
Aei𝜔t0 ei𝜙 = aei𝜔d t0 ei𝜙d . (9.62)
The two signals will be equal again at the next sampled point where t = t0 + Δt so we can
write
Aei𝜔(t0 +Δt) ei𝜙 = aei𝜔d (t0 +Δt) ei𝜙d (9.63)

@Seismicisolation
@Seismicisolation
 9.7 Sampling and Aliasing 181

or
Aei𝜔t0 ei𝜔Δt ei𝜙 = aei𝜔d t0 ei𝜔d Δt ei𝜙d . (9.64)
Equation 9.62 can be used to cancel terms in Equation 9.64, resulting in
ei𝜔Δt = ei𝜔d Δt . (9.65)
Define 𝜔′ to be the difference between 𝜔 and 𝜔d so that
𝜔 = 𝜔d + 𝜔′ . (9.66)
Substituting Equation 9.66 into Equation 9.65 yields
′
ei(𝜔d +𝜔 )Δt = ei𝜔d Δt (9.67)
or
′
ei𝜔d Δt ei𝜔 Δt = ei𝜔d Δt (9.68)
from which
′
ei𝜔 Δt = 1. (9.69)
Equation 9.69 can be written as
cos 𝜔′ Δt + i sin 𝜔′ Δt = 1 (9.70)
and, by equating the real and imaginary parts on the left and right hand sides, we get two
simultaneous equations, as follows
cos 𝜔′ Δt = 1 and sin 𝜔′ Δt = 0. (9.71)
The dual requirements in Equation 9.71 are satisfied by
𝜔′ Δt = ±2k𝜋; k = 0, 1, 2, … (9.72)
from which
)(
1
𝜔′ = ±2k𝜋 ; k = 0, 1, 2, … . (9.73)
Δt
Substituting this into Equation 9.66 yields
( )
1
𝜔 = 𝜔d ± 2k𝜋 ; k = 0, 1, 2, … . (9.74)
Δt
We can change Equation 9.74 from frequencies in rad s−1 to frequencies in Hz by noting
that 𝜔 = 2𝜋f and 𝜔d = 2𝜋fd , resulting in
( )
1
2𝜋f = 2𝜋fd ± 2k𝜋 ; k = 0, 1, 2, … (9.75)
Δt
Canceling the 2𝜋 terms gives
( )
1
f = fd ± k ; k = 0, 1, 2, … . (9.76)
Δt
We make one last change to Equation 9.76 by introducing the Nyquist frequency fc =
1∕2Δt from Equation 9.59 to get
f = fd ± 2kfc ; k = 0, 1, 2, … . (9.77)

@Seismicisolation
@Seismicisolation
182 9 Experimental Data – Frequency Domain Analysis

Equation 9.77 specifies which high frequency components, f , will appear at the detected
frequency, fd . It is clear that, since fd ≤ fc , some of the high frequencies will be negative.
Consider only the positive frequencies, which are
f = fd + 2kfc ; k = 0, 1, 2, … (9.78)
and
f = −fd + 2kfc ; k = 0, 1, 2, … . (9.79)
Equations 9.78 and 9.79 can be combined as
f = 2kfc ± fd ; k = 0, 1, 2, … . (9.80)
The Nyquist frequency is sometimes called the folding frequency because of Equation 9.80.
For the case where k = 1, we find that the first frequency aliased with fd is
f = 2fc − fd (9.81)
and this frequency can be found by folding the frequency scale shown in Figure 9.6 around
the Nyquist frequency, fc .
Consider again the example of Section 9.6 but, in addition to the signal, x(t), that has
three frequency components, we will add another higher frequency component at 8.5 Hz.
The sampling time is Δt = 0.10 s, so that the highest frequency that will be detectable is the
Nyquist frequency fc = 1∕(2Δt) = 5 Hz. As a result, the 8.5 Hz signal is outside the detectable
range but we expect to see it as an aliased signal at (according to Equation 9.81) fd = 2 ×
5 − 8.5 = 1.5 Hz. Let the 8.5 Hz component have an amplitude of 20 so that the function in
Equation 9.48 becomes
x(t) = 2 + 5 sin(2𝜋t) + 10 sin(4𝜋t) + 15 cos(6𝜋t) + 20 cos(17𝜋t). (9.82)
Table 9.4 shows the coefficients that are calculated by the DFT for this revised case. Notice
that a3 , the coefficient related to the cosine term at 1.5 Hz, erroneously has the ampli-
tude of the 8.5 Hz component. Aliasing is a serious problem in experimental data analysis.
Figure 9.7 shows the dramatic change from Figure 9.4.
There will invariably be signal components at frequencies higher than the frequency
range of interest in the experiment. The only way to avoid aliasing is to use a low-pass analog
filter when the data are recorded. The filtering has to be done using hardware. Section 9.9
discusses digital filtering after the data have been recorded but it needs to be emphasized

fd fc - fd fc - fd fd

Frequency
f=0 fd fc 2fc - fd 2fc

Figure 9.6 The folding frequency.

@Seismicisolation
@Seismicisolation
 9.7 Sampling and Aliasing 183

Table 9.4 DFT coefficients.

Coefficient Value Frequency (Hz)

a0 2.000 0.00
a1 0.000 0.50
a2 0.000 1.00
a3 20.000 1.50
a4 0.000 2.00
a5 0.000 2.50
a6 15.000 3.00
a7 0.000 3.50
a8 0.000 4.00
a9 0.000 4.50
a10 0.000 5.00
b1 0.000 0.50
b2 5.000 1.00
b3 0.000 1.50
b4 10.000 2.00
b5 0.000 2.50
b6 0.000 3.00
b7 0.000 3.50
b8 0.000 4.00
b9 0.000 4.50

25

20

15
amplitude

10

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0
frequency (Hz)

Figure 9.7 Aliased DFT results.

@Seismicisolation
@Seismicisolation
184 9 Experimental Data – Frequency Domain Analysis

that digital filtering relies on accurate DFTs and, if the aliased results are in the data, the
digital filtering process will not remove them because they are seen to be legitimate low
frequency signals.

9.8 Leakage and Windowing

To illustrate the concept of leakage, we consider an example where we have a signal with
two frequencies. Let
x(t) = 60 cos(40𝜋t) + 90 cos(80𝜋t). (9.83)
That is, we have an amplitude of 60 at 20 Hz and an amplitude of 90 at 40 Hz. Choose Δt =
0.01 s and collect 100 samples. As a result we have a sample time T = 100 × 0.01 = 1 s, a
frequency resolution Δf = 1∕T = 1 Hz, and a Nyquist frequency fc = 1∕(2Δt) = 50 Hz. The
DFT is calculated and the results, shown in Figure 9.8, are as expected.
Now we consider a different signal with the same amplitudes but slightly different fre-
quencies (amplitude of 60 at 20.5 Hz and an amplitude of 90 at 39.5 Hz)
x(t) = 60 cos(41𝜋t) + 90 cos(79𝜋t). (9.84)
Using the same settings for the DFT as in the first signal, the result is as shown in Figure 9.9.
The immediate question is, what happened?
Before answering the what happened? question, it must be pointed out that we are looking
at what is called leakage. This is a phenomenon where the amplitudes that are present in
the signal at discrete frequency bins leak into neighboring bins.
As to the question of what happened?, we can start by considering the continuous Fourier
transform of a square wave. Figure 9.10 shows the square wave and some of the components

100
90
80
70
60
amplitude

50
40
30
20
10
0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
frequency (Hz)

Figure 9.8 The DFT for the first example (no leakage).

@Seismicisolation
@Seismicisolation
 9.8 Leakage and Windowing 185

70

60

50
amplitude

40

30

20

10

0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
frequency (Hz)

Figure 9.9 The DFT for the second example (leakage).

Figure 9.10 CFT approximation to the square wave.

that are used in approximating it. The thin lines are components in the expansion and the
thick line is the sum of the components. Theoretically, an infinite number of components
will exactly duplicate the square wave but, as can already be seen, this will require some
amplitude at every frequency in order to fill in the discontinuities. The square wave has the
same period as the first component of the Fourier series but the corners can only be filled
in by adding higher and higher frequency components.
Going back to our two examples, the difference is that the first example (20 Hz and
40 Hz) is a signal where both components have completed an integer number of cycles in
the 1 s sample that we considered. In the second example (20.5 Hz and 39.5 Hz), neither

@Seismicisolation
@Seismicisolation
186 9 Experimental Data – Frequency Domain Analysis

component is at the end of a cycle when the sample is truncated so the DFT sees a
discontinuity and tries to deal with it by using all available frequency components.
The Fourier transform is only valid for a periodic function and requires information for
complete cycles of the signal. The chances of any experimental data sample being periodic
and being sampled over complete cycles is essentially zero. Leakage can’t be completely
eliminated but it can be reduced by windowing the data. Windowing forces the data to
appear to be periodic by multiplying them by a function that is zero at both the start and
the end and that rises to a value of one in the center.
You can imagine all sorts of windows. For example, the rectangular window goes from
zero to one at the beginning of the data, stays at one, and then goes back down to zero at the
end of the data. This is the default window that we get without manipulating the data so it
has no effect on leakage. The simplest window that has an effect is the triangular window
that ramps from zero at the start of the data to one at the center of the data and then back
down to zero again at the end of the data. It is, however, more common to use windows that
have continuous derivatives and zero slopes at their ends. A very common window function
is the Hanning window shown in Figure 9.11
The Hanning window is expressed as a weighting function, 𝑤(x)
1
𝑤(x) = (1 + cos 𝜋x) ; − 1 ≤ x ≤ 1 (9.85)
2
and is shown time-scaled in Figure 9.11 in order to cover the one second of data we have
been considering in our examples.
Figures 9.12 and 9.13 show the original data and the windowed data respectively. It is
clear from the change in shape of the data that the windowing will have an effect on the
amplitudes of the DFT coefficients. We take the amplitude effect into account by realizing
that the area under the Hanning window is one-half the area under a rectangular window.
As a result, the windowed amplitudes need to be multiplied by two in order to restore their
actual values. This area scaling needs to be done whichever windowing function is used.
Figure 9.14 shows the DFT for the windowed data of Figure 9.13. While not being perfect,
the result is far superior to that shown in Figure 9.9.

1.2

1.0

0.8
amplitude

0.6

0.4

0.2

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
time (s)

Figure 9.11 The Hanning window.

@Seismicisolation
@Seismicisolation
 9.9 Decimating Data 187

200

150

100
amplitude

50

–50

–100

–150
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
time (s)

Figure 9.12 The data from Equation 9.83.

200

150

100
amplitude

50

–50

–100

–150
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
time (s)

Figure 9.13 The windowed data.

9.9 Decimating Data

There are times when we realize that we have sampled at a higher rate than is necessary
and we wish to reduce the sampling rate. For example (and this has happened), if we set the
sampling rate in our A/D converter to 100 000 samples per second and we analyze using an
FFT limited to 8192 points, we will have a total sample time T= 0.08192 s and the frequency
resolution will then be Δf = 1∕T = 12.2 Hz. If your analysis is meant to be able to distin-
guish between signals at 20 and 30 Hz, for instance, then you are out of luck because both
of these signals will be in the same frequency bin. It becomes tempting to say that, since

@Seismicisolation
@Seismicisolation
188 9 Experimental Data – Frequency Domain Analysis

80

70

60

50
amplitude

40

30

20

10

0
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
frequency (Hz)

Figure 9.14 The DFT for the second example with windowing.

this is all digital data, we can simply ignore some of the data and we will get the same result
as if we had sampled more slowly. That is, just using every second point will be equivalent
to sampling at 50 000 samples per second or, even better, using every hundredth point will
be equivalent to sampling at 1000 Hz so that 8192 points will give Δf = 0.122 Hz.
While it is true that we can leave out data points and the result will be exactly as if we had
sampled at a slower rate, we need to keep in mind the possibility of aliasing as discussed
in Section 9.7. If we did our work correctly when we sampled the data, we used a low-pass
analog filter with a cut-off frequency we desired and then sampled at a rate that was at
least twice as high as the cut-off frequency. For the example just discussed, we sampled at
100 000 samples per second so our cut-off frequency would have been around 50 000 Hz.
The digital signal we recorded therefore has frequency components up to 50 000 Hz. If we
then decide to use every hundredth point to set our sampling rate to 1000 Hz, we are going
to get serious aliasing.
The process of leaving out data points is called downsampling.
There is a difference between the aliasing stemming from downsampling and that which
comes from not using a low-pass filter when sampling data. If you omit the low-pass fil-
ter your DFT will interpret high frequency signals as low frequency signals and you won’t
be able to differentiate between actual peaks and aliased peaks. In the case of downsam-
pling, you know where the aliased peaks actually are in the sense that you see them in the
DFT before you do the downsampling. This gives you the option of using a digital filter to
remove them before you do the downsampling. The sequence of using a low-pass digital
filter followed by downsampling is called decimation.
For the example we are using, we want to get rid of anything above 25 Hz before we down-
sample. This is best accomplished in the frequency domain by performing a DFT on the
full data set, then implementing a low-pass filter with a cut-off at 25 Hz, and then perform-
ing the inverse DFT to get a revised time signal. Implementing the filter in the frequency

@Seismicisolation
@Seismicisolation
 Exercises 189

domain is simply a matter of using a rectangular window that is equal to one for frequencies
below the cut-off frequency and zero above it. This can be accomplished by setting the DFT
coefficients related to frequencies above the cut-off frequency to zero.

9.10 Averaging DFTs

The last topic to be covered in this chapter has to do with suppression of noise in experi-
mental measurements. Noise in measurements consists of random changes in signal values
that can, in some cases, reach levels that approach that of the signal you are trying to mea-
sure. It is good experimental practice to attempt to minimize noise levels before recording
data. Nevertheless, there will always be some level of noise in experimental data.
Since the signal comprises two components, one of which is regular and repeatable while
the other is random, it can be expected that two DFTs calculated over different parts of the
time signal will give different results. Both will have large amplitudes at the frequencies of
the signal but smaller and non-repeatable amplitudes at other frequencies. It can therefore
be expected that if several of the DFTs are averaged, the real signal will be prominent since
it appears in every DFT whereas the noise will give different peaks in different spectra and,
on average, will not be prominent. This is exactly what happens. The noise is part of the
signal and will never be zero but averaging the spectra results in a smoothing of the noise.
This is a simple linear averaging process where the DFT amplitudes at each frequency are
added and then divided by the number of DFTs performed.

Exercises

Descriptions of the systems referred to in the exercises are contained in Appendix A.

9.1 Prove that the trigonometric integrals in Equation 9.10 are zero. That is, for 𝜔0 =
2𝜋∕T, show that,
T T
cos(n𝜔0 t)dt = 0 and sin(n𝜔0 t)dt = 0.
∫0 ∫0

9.2 Show that, for 𝜔0 = 2𝜋∕T and p ≠ n,

T
cos(n𝜔0 t) sin(p𝜔0 t)dt = 0
∫0
T
sin(n𝜔0 t) sin(p𝜔0 t)dt = 0
∫0
T
cos(n𝜔0 t) cos(p𝜔0 t)dt = 0
∫0
T
sin(n𝜔0 t) cos(p𝜔0 t)dt = 0
∫0

@Seismicisolation
@Seismicisolation
190 9 Experimental Data – Frequency Domain Analysis

by using the identities

1 1
cos(n𝜔0 t) cos(p𝜔0 t) = cos[(n + p)𝜔0 t] + cos[(n − p)𝜔0 t]
2 2
1 1
sin(n𝜔0 t) sin(p𝜔0 t) = cos[(n − p)𝜔0 t] − cos[(n + p)𝜔0 t]
2 2
1 1
cos(n𝜔0 t) sin(p𝜔0 t) = sin[(n + p)𝜔0 t] − sin[(n − p)𝜔0 t]
2 2
1 1
sin(n𝜔0 t) cos(p𝜔0 t) = sin[(n + p)𝜔0 t] + sin[(n − p)𝜔0 t].
2 2

9.3 Use Equation 9.85 to show that the area under the Hanning window is one-half the
area under the rectangular window.

9.4 Your data acquisition system can sample at any rate you choose between 1 and 4000
samples per second and it has sufficient memory to store up to 8192 samples. You also
have a low-pass filter with an adjustable cut-off frequency that can be set anywhere
between 1 and 2000 Hz. You have been asked to come into a building experienc-
ing excessive floor vibrations and determine whether the cause is a rotary machine
operating at 1800 RPM in this building or a reciprocating compressor operating at 28
strokes per second in an adjacent building.
You have an accelerometer and plan to measure and record accelerations on the floor.
You will then perform an FFT and calculate the response spectrum to see which
machine is at fault.
Specify your sampling rate, how many samples you should take per measurement,
what your cut-off frequency should be, and how many measurements you can aver-
age. Give reasons for your choices.

9.5 Program the DFT series of Section 9.4 using software of your choice and then generate
some data similar to that used as an example in Section 9.6 (i.e. Equation 9.48) and
test your software on it. Experiment with the software – try out aliasing, leakage, and
so on.

9.6 If you install strain gauges on the spindle of a milling machine, you will be able to
pick up both the relatively constant normal force on the cutter and the cyclic forces
caused by tooth engagement as the spindle turns. If you plot the total force versus
time, you will see the cyclic forces as a relatively small oscillation superimposed on
the large, constant, normal force. The amplitude of the cyclic force variation is about
5% of that of the normal force.
a) What is the mean value of this signal?
b) What would you expect the DFT of the signal to look like?
c) If you calculated the mean value in the time domain and subtracted it from each
of the data points before doing the DFT, what would your DFT look like?
d) Using the DFT of the zero mean signal in part (c) is the preferred way of getting
the frequency content of a signal in most cases. Why do you think this is?

@Seismicisolation
@Seismicisolation
 Exercises 191

9.7 Using the simulation data you generated for system 1 in Exercise 8.1 and your DFT
series software, find the frequency spectrum. Compare the frequencies where your
DFT shows peaks to the natural frequencies determined in Exercise 5.2.

9.8 Run your simulation of system 12 from Exercise 8.3 and generate enough data to be
able to produce a DFT. Do you see the natural frequency?

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 193

Representative Dynamic Systems

A.1 System 1
Two massless rigid rods (lengths d1 and d2 ) support two particle masses (m1 and m2 ) at
points A and B as shown in Figure A.1. The joints at O and A are frictionless pinned joints
and the bodies move in two dimensions under the influence of gravity. Use the angles 𝜃1
⃗ is fixed in link OA and ⃗𝚤 is
and 𝜃2 as degrees of freedom. The coordinate system (⃗𝚤 − ⃗𝚥 − k)
aligned with OA.

⇀
j ⇀
i
B ⇀
m2 k

d2

g θ2

A
m1
d1
θ1
O

Figure A.1

A.2 System 2
A thin, uniform, rigid rod (AB) has length 2𝓁 and mass M. The center of mass of the rod is
at G. The rod is free to slide on the frictionless pivot at O as shown in Figure A.2. Attached
to one end of the rod (B) is an inextensible string of length d that supports a particle with
mass m (C). A known force F(t) is applied to the other end of the rod (A) and remains
The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.
© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
194 A Representative Dynamic Systems

perpendicular to the rod at all times. The bodies move in two dimensions under the influ-
ence of gravity. The three degrees of freedom are 𝜃, 𝜙, and x.

⇀
k B
⇀

i ⇀
j
O θ ϕ
x

G d g

A m C
F(t)

Figure A.2

A.3 System 3

A small vehicle has been designed as a Mars rover and a simple dynamic model of it is
shown in Figure A.3.
The vehicle is modeled as a single rigid body supported by two vertical springs, each
having stiffness, k. The lower ends of the springs are attached to small, rigid, massless
wheels that exactly follow the terrain as the vehicle moves forward. The center of mass of
the vehicle is at G. One spring is located a distance, a, ahead of G and the other is a distance,
b, behind G. Motions of the vehicle are described by the vertical translation of the center of
mass, x, and a pitch rotation, 𝜃, with positive directions as shown. The vehicle has a total
mass of m, a pitch moment of inertia about G of IG , and moves forward with a velocity V.

g
x

V
θ G
k k

b a

Figure A.3

A.4 System 4

The circular disk rotates in a horizontal plane about the fixed point O. It has constant angu-
@Seismicisolation
@Seismicisolation
lar speed Ω in the direction shown in Figure A.4. A slot has been milled into the disk in a
 A.6 System 6 195

radial direction. The small mass, m, is able to move along the slot without friction and is
attached to the center of the disk by a spring of stiffness, k. The spring has an undeflected
length of 𝓁0 (i.e. there is no force in the spring when it has a length of 𝓁0 ).

m
k

O
Ω

Figure A.4

A.5 System 5

Figure A.5 shows an inclined plane that is used to unload cylindrical oil drums from a
platform.
The drums are rolled onto the end of the plane when it is in a horizontal position (𝜃 = 0).
A person applies a vertical force, F, to the opposite end of the plane in order to pivot the
plane about the frictionless pin at O, thereby raising the drums so that they roll without
slipping down the plane, eventually being deposited on the ground. The force varies in mag-
nitude as the drum rolls down the plane but is always perfectly vertical. The center of mass
of the drum is at G and the center of mass of the platform is at O.
The parameters are:
Oil drum: mass = m, radius = r, moment of inertia about the center of mass = 12 mr 2
Inclined plane: mass = m, moment of inertia about the pivot point = 3mr 2 .

g
d
d x
r G

F O θ

Figure A.5

A.6 System 6

Figure A.6 shows a disk with an angular velocity 𝜔 about a vertical axis. Attached to brackets
@Seismicisolation
@Seismicisolation
at the edge of the disk is a massless rigid rod of length r that has a point mass m at its free
196 A Representative Dynamic Systems

end. The brackets welded to the disk contain bearings that allow the rigid rod to pivot freely
so that the mass can move in a vertical plane as the disk spins.

θ A

ω d
g O

Figure A.6

A.7 System 7
The system shown in Figure A.7 consists of three rigid bodies. The two slender rods are
identical, each having mass = m and length = 2L. They are connected at their ends by
frictionless pin joints at O, A, and B. The rectangular body also has mass = m and slides
without friction in its guideway. A constant force F is applied to this body in the direction
indicated. A spring with stiffness = k connects the system to ground at point A as shown.
The spring is undeflected when 𝜃 = 0. The system operates in the horizontal plane so that
gravity has no effect on it.

m
B

2L
θ
m
k
A

m
θ
2L

O
@Seismicisolation
@Seismicisolation
Figure A.7
 A.9 System 9 197

A.8 System 8

Figure A.8(left) shows a rectangular object with four point masses attached to it by massless,
slender, rigid rods connected to frictionless pin joints at the corners (such as at A). The
object is at rest and the masses rest against it. We consider the rectangular object and the
four attached bodies to constitute a system. The system is free to rotate in the horizontal
plane about the fixed point O.
Figure A.8(right) shows the system after some applied torque has started it rotating. The
rectangular object now has an angular velocity 𝜔 and an angular acceleration 𝜔̇ in the direc-
tion shown and the point masses are in motion relative to it.

A L
L m
m
m
a a
L L
m
O ω, ω• O
m
L L
m
m m
L
L

Figure A.8

A.9 System 9

Figure A.9 shows a slotted, thin, uniform rod OA that is connected to the ground by a fric-
tionless pin at O and moves in a horizontal plane. It is also connected to the ground by a

A
B
k

h R

@Seismicisolation
@Seismicisolation
Figure A.9
198 A Representative Dynamic Systems

linear spring of stiffness, k, that is connected to a slider in the slot. The slot is frictionless
so that the spring always remains perpendicular to the rod. The spring can be modeled as
having an undeflected length of zero. That is, there is no force in the spring when 𝜃 = 0.
The force F, applied at point A, is constant and remains parallel to line OB.

A.10 System 10
Figure A.10 shows a slender uniform rod of mass 3m and length d that is connected to
the ground by the frictionless pin at O. The rod is connected to a linear spring of stiffness
k which is in turn connected to a slider of mass m that moves in a frictionless slot. The
slider is also connected to the ground by another linear spring of stiffness k. The springs
are undeflected when the angle 𝜃 and the displacement x are zero. The system moves in a
horizontal plane. A force F(t) is applied to the free end of the rod as shown. This force is
always perpendicular to the rod.

F(t) k k
m

θ d

Figure A.10

A.11 System 11
Figure A.11 shows a system in two different positions. The system consists of a massless
rigid rod of length L connecting two massless rollers that are constrained to move in hor-
izontal and vertical slots as shown. Connected to the roller on the right is a mass m that
moves freely in the vertical slot. The spring is undeflected when the system is in the posi-
tion shown in the left-hand figure. When the system is in motion, it has a single degree of

@Seismicisolation
@Seismicisolation
 A.12 System 12 199

freedom 𝜃 as shown in the figure to the right. There is an equilibrium value of 𝜃 as is implied
by the figure to the right.

Undeflected State Equilibrium State

k k
θ

L
m

g
m

Figure A.11

A.12 System 12
Figure A.12 shows a slender uniform rod of mass m and length 2d that is connected to
the ground by the frictionless pin at O. The rod is supported by a linear spring of stiffness k
connected to it at A and to a slider in a frictionless, ground-fixed, slot at B. As the rod moves,
the slider aligns itself in the slot to ensure that the spring is always truly vertical. The spring
is undeflected when the angle 𝜃 is zero.

k
O

θ
G
d

d
A

Figure A.12

@Seismicisolation
@Seismicisolation
200 A Representative Dynamic Systems

A.13 System 13
Figure A.13 shows a thin uniform rod OA (mass = m, length = 2𝓁) pinned to the ground
at O and to an identical thin uniform rod BC at A. Both pins are frictionless. The system
moves in a vertical plane. A spring of stiffness k has one end attached to rod BC at B and the
other end attached to a slider D that moves freely within a frictionless slot in OA. The slider
causes the spring to remain perpendicular to rod OA at all times. The force in the spring is
zero when its length is zero.

C
2
ϕ

d
A
B
k D

g
θ
2

Figure A.13

A.14 System 14
Figure A.14 shows a uniform disk (mass = m; radius = r) that rolls without slipping on a
stationary circular body that has radius R. The center of mass of the disk is at point A and

g F

r
A

R
θ

@Seismicisolation
@Seismicisolation
Figure A.14
 A.16 System 16 201

the disk is pinned there to a uniform slender rod (mass = m) that is also pinned to the center
of the stationary body at point O. The pins are frictionless. There is a constant vertical force
F applied to the rod at point A.

A.15 System 15
Figure A.15 shows a particle of mass, m, that is supported in the vertical plane by two springs
of stiffness, k, and, 4k. The springs have undeflected lengths of zero. A constant horizontal
force, F, is applied to the particle as shown.

g
A B

k 4k

m F

Figure A.15

A.16 System 16
Figure A.16 shows a slender uniform rod AB (mass = m; length = 2𝓁) attached to a massless,
frictionless roller at point A. The roller travels in the vertical plane inside a circular slot
(radius = r) cut into a solid structure.

O θ g
r
A
2
ϕ

Figure A.16

@Seismicisolation
@Seismicisolation
202 A Representative Dynamic Systems

A.17 System 17

The small mass, m, in Figure A.17 moves freely on the rotating circular hoop. There is
friction between the mass and the hoop and there is no gravity. The hoop has a constant
angular velocity vector of magnitude 𝜔 in the direction shown.

r θ
m

Figure A.17

A.18 System 18

Three masses are connected by three identical springs as shown in Figure A.18. The masses
are constrained to have vertical motion only and the degrees of freedom, relative to the
equilibrium positions of the masses, are x1 , x2 , and x3 as shown. The vertical force, F(t) =
F sin 𝜔t, is applied as shown in the figure.

k
g

2m
x3
k k

m m

x1 x2

Fsinω t

Figure A.18
@Seismicisolation
@Seismicisolation
 A.20 System 20 203

A.19 System 19
Figure A.19 shows one truck of a transit vehicle with degrees of freedom specified as bounce
(x) and pitch (𝜃) of the truck frame. Use m for the truck frame mass, I for the truck frame
pitch moment of inertia, ks for the secondary stiffness where the truck frame is connected
to the carbody, and kp for the primary stiffnesses that connect the wheels to the truck frame.
The car body is massive enough that it can be assumed to have no motion and the wheels
remain stationary on the rails. Assume that damping is small enough to be neglected. The
dimensions a, b, and c are known.

Car body

x ks
θ Truck frame
c
b a
kp kp
Wheels

Figure A.19

A.20 System 20
The two carts (masses m and 4m) in Figure A.20 are connected together and to the ground
by springs of stiffness, k, as shown. There is a force, f (t), applied to the lower cart.

g
x1
k

x2 m
k

4m
f(t)
θ

Figure A.20

@Seismicisolation
@Seismicisolation
204 A Representative Dynamic Systems

A.21 System 21
Figure A.21 shows a cart of mass, m, that is attached to the ground at point A by a spring of
stiffness, k, and a damper with damping coefficient, c. The motion of the cart is forced by a
harmonic motion, y(t), at the end of another damper with coefficient, c. y(t) has amplitude,
Y , and frequency, 𝜔.

x(t)

y(t)
k
c
A m

Figure A.21

A.22 System 22

Figure A.22 shows a robot moving a rectangular solid block of mass, m, with dimensions,
b × b × 2b. The arm, AB, has constant length, 𝓁, and rotates in a horizontal plane with angu-
lar velocity, 𝜔0 . The arm, BC, has variable length, x, and angular velocity about the line, AB,
of magnitude, 𝜔1 .

b
2b

b
C

ω0
x
A θ

 ω1
B

Figure A.22

@Seismicisolation
@Seismicisolation
 A.23 System 23 205

A.23 System 23
Figure A.23 shows a three dimensional pendulum with a massless, axially elastic rod sup-
porting a particle of mass, m. The system is free to move in 3D except in so far as point, O,
is constrained to remain attached to the ground and the rod, Om, cannot bend. The rod has
an axial stiffness, k, and an undeflected length, 𝓁0 . The position of the mass is specified by
the length of the rod, 𝓁, the angle, 𝜃, measured in the horizontal plane from fixed line, AB,
and the angle, 𝜙, measured from a vertical line in the vertical plane of the rod.

ϕ
g

k

A
B
⇀ θ
k
m
⇀ ⇀
i j

Figure A.23

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 207

Moments and Products of Inertia

In any problem with rotational degrees of freedom, there is a need to know the moments
and products of inertia of the bodies in the system. In systems where masses are modeled as
particles, the determination of the inertia properties is simply a matter of summing particle
masses multiplied by products of their distances from a reference point. In rigid bodies the
number of particles is infinite and the summations become integrals.
Moments and products of inertia were introduced in Section 2.9 where the definitions
about a reference point O were given as,

Moment of inertia about the x-axis: IxxO = (y2 + z2 )dm

∫body

Moment of inertia about the y-axis: IyyO = (x2 + z2 )dm

∫body

Moment of inertia about the z-axis: IzzO = (x2 + y2 )dm

∫body

Product of inertia about the xy-axes: IxyO = IyxO = xydm

∫body

Product of inertia about the xz-axes: IxzO = IzxO = xzdm

∫body

Product of inertia about the yz-axes: IyzO = IzyO = yzdm.

∫body

B.1 Moments of Inertia

There are several points to note about moments of inertia. First, it is clear that moments of
inertia are always positive due to the fact that the integrals involve the sums of squares of
distances and squared numbers are always positive. Notice that, by Pythagoras’ theorem,
these sums of squares are simply the square of the distance from the particle to the reference
point in the plane perpendicular to the axis about which the rotation is being considered.
Also notice that adding mass to a body always increases the moment of inertia unless the
mass is added at the reference point.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
208 B Moments and Products of Inertia

It is relatively simple to show that, for a rigid body, the sum of the any two moments of
inertia must be greater than the third. Simply add together Ixx and Iyy to show1 ,

Ixx + Iyy = (y2 + z2 )dm + (x2 + z2 )dm

∫body ∫body

= (x2 + y2 + 2z2 )dm

∫body

= (x2 + y2 )dm + 2z2 dm

∫body ∫body

= Izz + 2z2 dm > Izz . (B.1)

∫body
Clearly this statement could be written for the sum of any two moments of inertia with the
result that that sum must be greater than the third moment of inertia. This is a useful check
on moments of inertia, especially those that are provided to the analyst by someone else. If
they fail this test, then they are incorrect.
The dimensions of moments of inertia are mass times distance squared. In SI units this
will be kg m2 . In US customary units, different forms, depending on the units being used
for the problem at hand, are appropriate. Most commonly, one will work with lb ft sec2 or
lb in sec2 . Appendix C explains the use of different systems of units.
Because the moment of inertia is the product of mass and distance squared and the units
can be somewhat complicated, it has become customary and convenient to specify the
moment of inertia of a body through the use of the mass of the body and a characteris-
tic length called the radius of gyration, kO , about the reference point. The radius of gyration
is defined as,
√
IO
kO = (B.2)
m
where m is the total mass of the rigid body and IO is the body’s moment of inertia about the
reference point for the plane of motion being considered. Equation B.2 can be manipulated
to read,
IO = mkO2 (B.3)
which can be interpreted as saying that the radius of gyration is the distance that a particle
having the same mass as the rigid body would need to be from the reference point to have
the same moment of inertia as the rigid body. The inertial properties of a rigid body are
therefore often specified to an analyst as simply a mass and a distance. The actual moment
of inertia is reconstructed using Equation B.3.

B.2 Parallel Axis Theorem for Moments of Inertia

The integration required to derive an expression for the moment of inertia is often tedious
and to be avoided to the extent possible. Undergraduate textbooks on dynamics usually

1 The reference point O is dropped from the notation for brevity.

@Seismicisolation
@Seismicisolation
 B.2 Parallel Axis Theorem for Moments of Inertia 209

YG
YO
XG
dm

XO

G
y
d
x
v
u
O

Figure B.1 Parallel axis theorem.

include extensive tables of moments of inertia for rigid bodies of various shapes. In con-
structing the tables, the authors have had to choose a reference point for the integration and
the point chosen is most often the center of mass of the body, G. The parallel axis theorem
allows the analyst to use the moment of inertia about the center of mass to calculate the
moment of inertia about a parallel axis passing through another point without performing
any integrations.
Figure B.1 shows a rigid body rotating in a plane with the center of mass, G, and another
reference point, O, shown. √The distance from O to G is u in the x-direction, any 𝑣 in the
y-direction or simply d = u2 + 𝑣2 .
Izz can be calculated in this plane since all rotations are around the z-axis. The moment
of inertia about the center of mass is,

IzzG = (x2 + y2 )dm (B.4)

∫body
and the moment of inertia about the reference point O is,

IzzO = [(u + x)2 + (𝑣 + y)2 ]dm. (B.5)

∫body
Equation B.5 can be expanded and written as2 ,

IzzO = [(u2 + 2ux + x2 ) + (𝑣2 + 2𝑣y + y2 )]dm (B.6)

∫
which can be rearranged to be,

IzzO = (u2 + 𝑣2 )dm + (x2 + y2 )dm + (2ux + 2𝑣y)dm (B.7)

∫ ∫ ∫
where we can recognize that u2 + 𝑣2 = d2 , that the second integral is the moment of inertia
about G (see Equation B.4), and that u and 𝑣 are constants so they can be factored out of
the last integral. The result is,

IzzO = d2 dm + IzzG + 2u xdm + 2𝑣 ydm. (B.8)

∫ ∫ ∫

2 The integrals are assumed to be over the entire body for the remainder of this discussion so the word
body can be omitted from the integral signs to improve readability.

@Seismicisolation
@Seismicisolation
210 B Moments and Products of Inertia

The first integral term in Equation B.8 is clearly just the total mass, m, of the rigid body.
The two other integrals are both definitions of the location of the center of mass with respect
to the origin for the integral, G, and so define the location of the center of mass with respect
to itself. Both of these integrals are therefore zero and the parallel axis theorem becomes,

IzzO = IzzG + md2 . (B.9)

B.3 Parallel Axis Theorem for Products of Inertia

There is also a parallel axis theorem for products of inertia. Figure B.1 shows the x–y plane of
the body being considered. The products of inertia Ixy = Iyx are found in this plane according
to,

IxyG = IyxG = xydm (B.10)

∫
or:

IxyO = IyxO = (u + x)(𝑣 + y)dm. (B.11)

∫
Equation B.11 can be expanded to give,

IxyO = IyxO = (u𝑣 + uy + 𝑣x + xy)dm (B.12)

∫
which, upon taking constants outside the integration, can be written as,

IxyO = IyxO = u𝑣 dm + u ydm + 𝑣 xdm + xydm (B.13)

∫ ∫ ∫ ∫
For the same reasons given in Section B.2, the second and third integrals in the sequence
are definitions of the location of the body center of mass with respect to itself and must
therefore be zero. The first integral is the total mass of the body and the last integral is IxyG ,
yielding the parallel axis theorem for products of inertia,

IxyO = IyxO = IxyG + mu𝑣. (B.14)

Note that the signs associated with u and 𝑣 in Equation B.14 are important. The product
u𝑣 can be positive or negative depending upon these signs. Referring to Figure B.1, you can
see that the constant distances u and 𝑣 are defined as being from the reference point O to
the center of mass G along the positive X and Y axes.

B.4 Moments of Inertia for Commonly Encountered Bodies

Tables of moments of inertia are easy to find in undergraduate textbooks on dynamics. Some
common moments of inertia are given here for quick reference.
Thin uniform rod: mass = m, length = 𝓁
1
IG = m𝓁 2 .
12

@Seismicisolation
@Seismicisolation
 B.4 Moments of Inertia for Commonly Encountered Bodies 211

Uniform disk: about an axis perpendicular to the plane in which it lies, mass = m,
radius = r
1
IG = mr 2 .
2
Uniform disk: about a radial axis, mass = m, radius = r
1 2
IG = mr .
4
Uniform rectangular block: mass = m, base = b, height = h
1
IG = m(b2 + h2 ).
12

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 213

Dimensions and Units

This appendix is included simply because of the difficulty that many people have with the
concepts of dimensions and units – particularly units. Dimensions are measurable quan-
tities that we use in an abstract way in that we don’t specify a scale for them. They are
simply referred to as length, mass, time, force, and so on. Units on the other hand are used
to quantify dimensions using scales based on agreed upon standards. Most of the world
uses SI (Système International d’unités) units while US customary units are still used in
the United States.
Applied mechanics uses four dimensions – length, time, force, and mass. Three of these
(length, time, and force) are measurable using common instruments such as rulers, clocks,
and scales. The fourth, mass, can only be inferred from a measurement of force in the pres-
ence of a known gravitational field. When you put your tomatoes on a scale in the vegetable
section of your food store and it tells you that you have 1 kg or 2.2 lb of tomatoes, the scale
has actually measured the weight and not the mass. The 2.2 lb is a direct measurement of
weight (i.e. force) and the 1 kg comes with the assumption that the local gravitational accel-
eration is 9.81 m s−2 , which may or may not be exactly the case since the gravitational field
varies slightly around the earth. So it becomes clear that gravitational acceleration must be
included in these discussions if we want to work with both mass and force. In US customary
units, the gravitational acceleration is generally accepted to be 32.2 ft sec−2 or 386 in sec−2 .
A difficulty arises from the fact that the four dimensions – length, time, force, and
mass – are related by a constraint equation. That is, Newton’s second law, F = ma, must be
satisfied by whatever units we choose to use. This means that only three dimensions can
be independent and the fourth must be derived to satisfy Newton’s second law.
The SI system specifies mass, length, and time as the independent dimensions and force is
the derived unit. Mass is in kilograms (kg), length is in meters (m), and time is in seconds (s).
Applying Newton’s second law then requires that force be mass times acceleration where
acceleration is length divided by time squared or kg m s−2 . This derived force quantity is
called a Newton (1 N = 1 kg m s−2 ).
US customary units are not as restrictive as SI units in that there is no specification as to
what the independent dimensions are. The field of applied mechanics commonly uses force,
length, and time as the independent dimensions and mass is the derived unit. Typically, but
not always, force is in pounds (lb), length is in feet (ft), and time is in seconds (sec). Mass
must therefore be m = F∕a which is lb sec2 /ft. This mass unit was once called a Slug but
that term is confusing and adds a restriction to the analysis that needn’t be there. It is much
The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.
© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
214 C Dimensions and Units

simpler to work with units that satisfy the constraint and not give the mass unit a name
at all. For example, if you are working on a vehicle dynamics problem, you are likely to
measure drag force in pounds, distance in miles, and time in hours since the speed of cars
in US customary units is typically measured in miles per hour (mph). The correct mass unit
for this set of units is lb hour2 /mile since you simply need to generate a force in pounds
when you multiply the mass by the acceleration which, in this quite ridiculous example,
must be measured in miles per hour per hour or mile/hour2 . The US customary units are
much more flexible than the SI units but the user needs to be careful. Rigid specifications
usually lead to fewer mistakes.
A final word needs to be said about engineers who work in thermodynamics and fluid
mechanics. They insist on using a quantity called the pound mass (lbm) as well as the stan-
dard pound which they label (lbf) to indicate that it is a force. The relationship between
these two quantities is derived from the fact that a body weighing 1 lbf on earth (g = 32.2
ft/sec2 ) has a mass of 1 lbm. That is, applying Newton’s second law,
ft
1 lbf = 1 lbm × 32.2
sec2
or, dividing both sides by 1 lbf,
lbm ft
1 = 32.2 = gc
lbf sec2
where gc is a constant equal to 1 that can be used whenever it seems appropriate. My
thermodynamics professor used to say “you can multiply or divide by 1 anytime with-
out changing anything”. This was advice that often caused more problems than it solved.
The metric system, before it became the rigidly controlled SI system, had a kilogram force
unit (kgf).

@Seismicisolation
@Seismicisolation
 215

Least Squares Curve Fitting

This appendix is a short review of least squares curve fitting in aid of enhancing the under-
standing of Fourier transforms in Chapter 9.
Consider the three data points, (xn , yn ) shown in Table D.1. We want to define a function
that will best approximate these data.
There is a class of curve fitting methods called trial functions with undetermined coeffi-
cients. To use one of these methods, we first define a set of functions, yi (x); i = 1, r, that
we think are good fits to the data we have and then generate an approximation by adding
together these functions multiplied by a set of coefficients, ai , which we later determine in
order to get the best possible fit to the data. That is, we let the approximation be
∑
r
yA (x) = ai yi (x) (D.1)
i=1

and then use one of several possible methods to find ai . Here we restrict ourselves to using
the method of least squares.
Consider first the case where we attempt to fit a straight line to the three points. That is,
we define an approximation, yA (x), as the sum of the constant function, y1 (x) = 1, and the
linear function, y2 (x) = x, multiplied by undetermined coefficients, a1 and a2 , yielding
yA (x) = a1 + a2 x. (D.2)
The goal is to find the “best” possible values of a1 and a2 . The “best” values are those that
minimize the error at the three data points where we define error as the difference between
yA (xn ) and yn . The error at the nth data point is then
en = (a1 + a2 xn ) − yn . (D.3)
We can’t minimize the errors as shown in Equation D.3 since they have no lower limit.
Negative errors, no matter how large, would be deemed to be superior to small positive or
negative errors in any minimization attempt.

The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.

© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
216 D Least Squares Curve Fitting

The solution to this problem is to define a function of the errors that is always positive
and then minimize that. The least squares method uses the squared errors as the function
to minimize. We therefore define a function, J, as

1∑ 1∑
N N
J= (en )2 = (a + a2 xn − yn )2 (D.4)
2 n=1 2 n=1 1
where N is the number of data points.
To minimize J, we simply take the partial derivatives of J with respect to the undeter-
mined coefficients, in this case a1 and a2 , and ensure that they are both zero. If we think of
a three dimensional plot with J plotted vertically against a1 and a2 axes, it would appear as
a bowl with the two partial derivatives being simultaneously zero at the bottom of the bowl
where J is a minimum.
The partial derivatives are

𝜕J ∑ 3
= (a + a2 xn − yn )(1)
𝜕a1 n=1 1
[ 3 ] [ 3 ] [ 3 ]
∑ ∑ ∑
= a1 (1) + a2 (xn ) − (yn ) (D.5)
n=1 n=1 n=1

and
𝜕J ∑ 3
= (a + a2 xn − yn )(xn )
𝜕a2 n=1 1
[ 3 ] [ 3 ] [ 3 ]
∑ ∑ ∑
= a1 (xn ) + a2 2
(xn ) − (xn yn ) . (D.6)
n=1 n=1 n=1

Substituting the values for xn and yn from Table D.1 results in the set of linear algebraic
equations
[ ]{ } { }
3 11 a1 163
= (D.7)
11 59 a2 963
with the solution yielding
{ } { }
a1 −17.42
= (D.8)
a2 19.57
so that the least squares linear approximation is
yA (x) = −17.42 + 19.57x (D.9)
which is the straight line plotted on Figure D.1 where the three dots are the points from
Table D.1.

Table D.1 Sample data points.

n xn yn

1 1 9
2 3 31
3 7 123

@Seismicisolation
@Seismicisolation
 D Least Squares Curve Fitting 217

160

140

120

100

80

y
60

40

20

−20
0 1 2 3 4 5 6 7 8 9
x

Figure D.1 Three data points and two least squares curve fits.

Also plotted on Figure D.1 is a dashed line that goes through all three data points exactly.
This is the result of using the least squares method with three undetermined coefficients.
That is, let the approximation be

yA (x) = a1 + a2 x + a3 x2 . (D.10)

Formulating J and setting the three partial derivatives to zero give the set of equations

⎡ 3 11 59 ⎤ ⎧ a1 ⎫ ⎧ 163 ⎫
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎢ 11 59 371 ⎥ ⎨ a2 ⎬ = ⎨ 963 ⎬ . (D.11)
⎢ ⎥⎪ ⎪ ⎪ ⎪
⎣ 59 371 2483 ⎦ ⎩ a3 ⎭ ⎩ 6315 ⎭

@Seismicisolation
@Seismicisolation
218 D Least Squares Curve Fitting

Solving Equation D.11 yields

yA (x) = 4 + 3x + 2x2 (D.12)
which is the function used to generate the data points in the beginning.
The fact that the method of least squares will fit a curve that goes through all the data
points if the number of undetermined coefficients is equal to the number of data points is
relied upon in Chapter 9 where the trial functions are sines and cosines and the result is the
discrete Fourier transform.

@Seismicisolation
@Seismicisolation
 219

Index

1/3 octave bands 144 Complex number 89, 90, 104, 107
coefficient matrix 126
a complex division 117
Acceleration conjugate pair 110, 120
absolute 3, 20 exponential 90, 180
centripetal 15 real and imaginary parts 110
constant 83 Constraint force 24, 25, 30, 59
Coriolis 15 Continuous Fourier Transform, see
general expression 14–16 Experimental data: the CFT
graphical construction 15–16 Coordinate system
radial 14 body fixed 9, 15
relative 9, 12 ground fixed 6
tangential 14 right handed 4, 8
Angular acceleration rotating 10, 13, 56
absolute 13–14 Coordinate transformation 8
Angular momentum, definition 33 Coupled 80
Angular velocity 4 Cramer’s rule 127, 128, 133
absolute 13–14 Critical damping 120
as a vector 4 coefficient 120
of a coordinate system 5
d
b Damper 65, 109, 113
Bandwidth 138, 140, 144 force transmitted through 122
Binomial theorem 153 generalized force 66
Body fixed coordinate system 9, 15 linear, viscous 66
Brook Taylor 92 Damping matrix 88
Decay 90, 109, 119
c Degree of freedom 6, 21, 43, 65, 73, 79, 83,
Carl David Tolme Runge 158 89, 108, 140, 159, 207
Center of mass 6, 28–30, 32, 34, 41, 44, 48, reference 90, 109
67, 69, 71, 83, 87, 114, 209 Differential equations 46, 83
Central difference method 154 constant coefficients 24, 89
Charles-Augustin de Coulomb 23 eigenvalues 103
The Practice of Engineering Dynamics, First Edition. Ronald J. Anderson.
© 2020 John Wiley & Sons Ltd. Published 2020 by John Wiley & Sons Ltd.
@Seismicisolation
@Seismicisolation
Companion Website: www.wiley.com/go/anderson/engineeringdynamics
220 Index

Differential equations (contd.) Experimental data 167–191

eigenvectors 103 averaging DFTs 189
equations of motion 21, 43, 90 decimating data 187–189
exponential solution 97 leakage and windowing 184–187
first-order 104 sampling and aliasing 179–184
homogeneous 88 the CFT 169–172
linear 24, 88, 126 the DFT 172–174
linearized 90, 103 a faster DFT 174–175
N degrees of freedom 103 the FFT 175–176
nonlinear 48 transforming to the frequency domain
second order 24, 48, 89, 103 169–189
sets of 88, 89, 103 typical test data 167–169
simultaneous 48 Extrapolation 149, 151, 158
time domain solution 145
Dimensional check of equations of f
motion 12 Fast Fourier Transform, see Experimental
Dimensions and units 213–214 data: the FFT
Discrete Fourier Transform, see Experimental FBD, see Free body diagram
data: the DFT Fixed point 6, 8, 10, 32, 35, 37, 69, 82, 83
Downsampling 188 Folding frequency 182
Force balance equations 21, 23, 28, 29, 45
e Free body diagram 21, 22, 27, 44
Edward John Routh 99 with Lagrange’s equation 55, 58
Eigenvalue 90, 97, 103, 104, 107, 114 Frequency domain analysis 125–144
Eigenvector, see also Mode shape 90, 97, Modeling frequency response 125–131
103, 104, 107, 109, 114, 117 Power spectral density 133–143
Eigenvectors 107–111 simulation with 139–143
Equations of motion, see also Lagrange’s units 138–139
equations of motion, see also Newton’s Seismic disturbances 132–133
equations of motion Friction
standard form 73–74 coefficient of 23, 42
Equilibrium Coulomb 23, 42, 119
general solution 84–85 dry 23
solutions 79–85
extreme value of potential energy g
81 Gabriel Cramer 127
states 79, 82, 83, 85, 87, 97, 99–100, 105, Galileo Galilei 19
107, 119, 125, 140 Gaspard Gustave de Coriolis 15
static 81 Gaussian elimination 163, 174
with constant acceleration 83 Ground fixed coordinate system 6
with constant velocity 83
with steady motion 81–84 h
Euler’s formula 90, 175 Hanning window 186
Euler’s moment equations 40 Harmonic function 91, 125, 134, 135, 169,
Euler’s integration 149 172, 174

@Seismicisolation
@Seismicisolation
 Index 221

Harmonic motion 83, 91, 96, 132, 142, 204 m

Heinrich Hertz 139 Martin Wilhelm Kutta 158
Mass
i concept of 19
Inertia 19 constant 23
Inertial reference frame 21 measurement of 20
Initial conditions 90, 126, 146, 148, 151, 154, variable 20
159, 162 Mass matrix 73, 88, 96, 163
Instability, see Stability Mean value 133, 167, 171
Issac Newton 19 Mean-square value 134–136, 168
cumulative curve 137
j Midpoint method 155
Jean-Baptiste Joseph Fourier 172 Mode shape, see also Eigenvector
Joint constraints 6 Mode shapes 107–122
Joseph-Louis Lagrange 55 computer animation 107
modal damping 118–122
k nodal points 115–116
Kinematics 3–18 rotational degrees of freedom
2-D, constant length example 6–8 111–115
translational degrees of freedom
2-D, variable length example 8–10
111–115
3-D 10–13
visualizing 109
constraints 6
with damping 116–118
definition 3
Moments and products of inertia
fixed points 6
207–211
goal of 5
dimensions 208
joint 6, 10, 13, 67, 193, 196, 197
for commonly encountered bodies 210
steps for performing analysis 5–8
parallel axis theorem for moments of
inertia 209
l parallel axis theorem for products of inertia
Lagrange’s equations of motion
210
55–75
physical interpretation 36–40
damping 65–67 radius of gyration 208
derivation 58–62 Momentum
for a particle 58–62 angular momentum vector 25
generalized forces 62–63 linear momentum vector 20
kinetic energy of a free rigid body
67–70 n
potential energy 64–65 Natural motion 90, 97, 107, 109, 125
Rayleigh’s dissipation function 65–67 Newton’s equations of motion 19–53
Lagrangian 65 angular momentum vector 33–36
Least squares curve fitting 215–218 equations of motion for particles 21–25
Leonhard Euler 40 Euler’s moment equations 40
Linearization of functions 92–95 general form of the rigid body force balance
Lyapunov functions 88 30

@Seismicisolation
@Seismicisolation
222 Index

Newton’s equations of motion (contd.) Stability analysis

general form of the rigid body moment Routh Stability Criterion 99–103
balance 32 Routh table 99
moment balance for a rigid body 30–33 standard procedure for 103–105
Newton’s laws 19–20 eigenvalue routines 104
second law for a particle 20–21 transforming the equations of motion
Nonlinear terms 88, 93, 95, 145 103
Nyquist frequency 179, 182, 184 Stable 87, 91, 103
Standard eigenvalue problem 103, 146
o Steady-state 90
Octave 144 Steady-state 82
Overdamped 120 Stiffess matrix 88, 141
Sum of angles formula 94
p
Percent damping 121 t
Perturbation 87, 90, 100, 103, 119 Taylor’s series 92, 94, 149, 153
Phase angle 109, 114, 117, 126, 131, 134,
two dimensional 94
142, 178, 180
Time domain solutions 145–164
position vector 6, 7, 10, 14, 18
central difference method 153–155
relative 6, 11
Euler integration 149–153
PSD, see Power spectral density
getting the equations of motion ready
146–147
q
methods with higher order truncation error
Rectangular window 186
157–159
René Descartes 58
Ride quality 144 non self-starting method 154
Right Hand Rule 5 numerical schemes 149
Right handed coordinate system 4, 8 self-starting method 155, 156, 158
Rigid body 8, 13, 25–27, 29, 30, 32, 34, 35, structure of a simulation program
38, 40, 47, 67, 68 159–161
RMS, see Root-mean-square variable time-step methods
Robert Hooke 64 155–157
Root-mean-square value 138 Time steps 149, 151, 153, 155
Rotation in a plane 8 Transfer function 133, 140, 142
Runge-Kutta-Fehlberg method 148, 158 Truncation error 150, 151, 153, 156

s u
Simulation 125, 139–141, 145, 148, 157, 159, Undamped natural frequency 120
167 Underdamped 120
Small angle approximation 92, 94 Unstable 80, 87, 89, 91, 96, 97, 99, 103, 107,
Stability 87–105 109
analytical 87–92
linear 88–92 v
nonlinear 87 Vector
of equilibrium states 79 cross product 4, 7
response types 91 derivative of 3–5
@Seismicisolation
@Seismicisolation
 Index 223

properties 3 Velocity
rate of change of direction 4, 5 absolute 3, 20
rate of change of magnitude radial 15
4, 5 relative 6, 8
unit vectors 4, 6, 8, 9, 22, 26, 33, 44, 50, tangential 16
58, 67, 71 Vibration absorber 131, 143

@Seismicisolation
@Seismicisolation
@Seismicisolation
@Seismicisolation
 WILEY END USER LICENSE AGREEMENT
Go to www.wiley.com/go/eula to access Wiley’s ebook EULA.

@Seismicisolation
@Seismicisolation